[PATCH tip/core/rcu 0/8] Documentation updates for 4.5

All of lore.kernel.org
 help / color / mirror / Atom feed

* [PATCH tip/core/rcu 0/8] Documentation updates for 4.5
@ 2015-12-04 23:49 Paul E. McKenney
  2015-12-04 23:50 ` [PATCH tip/core/rcu 1/8] documentation: Record RCU requirements Paul E. McKenney
                   ` (8 more replies)
  0 siblings, 9 replies; 18+ messages in thread
From: Paul E. McKenney @ 2015-12-04 23:49 UTC (permalink / raw)
  To: linux-kernel
  Cc: mingo, jiangshanlai, dipankar, akpm, mathieu.desnoyers, josh,
	tglx, peterz, rostedt, dhowells, edumazet, dvhart, fweisbec, oleg,
	bobby.prani

Hello!

This series contains documentation updates, perhaps most notably a first
installment of RCU design documentation, which first appeared on LWN
(http://lwn.net/Articles/652156/, http://lwn.net/Articles/652677/, and
http://lwn.net/Articles/653326/).  The individual patches are as follows:

1.	Add RCU requirements design document.

2-7.	Updates to the RCU requirements.

8.	Fix memory-barrier.txt typo, courtesy of Chris Metcalf.

							Thanx, Paul

------------------------------------------------------------------------

 b/Documentation/RCU/Design/Requirements/GPpartitionReaders1.svg |  374 +
 b/Documentation/RCU/Design/Requirements/RCUApplicability.svg    |  237 
 b/Documentation/RCU/Design/Requirements/ReadersPartitionGP1.svg |  639 ++
 b/Documentation/RCU/Design/Requirements/Requirements.html       | 2967 +++++++++-
 b/Documentation/RCU/Design/Requirements/Requirements.htmlx      | 2809 +++++++++
 b/Documentation/RCU/Design/htmlqqz.sh                           |  108 
 b/Documentation/memory-barriers.txt                             |    8 
 7 files changed, 7069 insertions(+), 73 deletions(-)


^ permalink raw reply	[flat|nested] 18+ messages in thread

* [PATCH tip/core/rcu 1/8] documentation: Record RCU requirements
  2015-12-04 23:49 [PATCH tip/core/rcu 0/8] Documentation updates for 4.5 Paul E. McKenney
@ 2015-12-04 23:50 ` Paul E. McKenney
  2015-12-05  0:07   ` Josh Triplett
  2015-12-05  0:34   ` Josh Triplett
  2015-12-04 23:50 ` [PATCH tip/core/rcu 2/8] Documentation: Record bottom-bit-zero guarantee for ->next Paul E. McKenney
                   ` (7 subsequent siblings)
  8 siblings, 2 replies; 18+ messages in thread
From: Paul E. McKenney @ 2015-12-04 23:50 UTC (permalink / raw)
  To: linux-kernel
  Cc: mingo, jiangshanlai, dipankar, akpm, mathieu.desnoyers, josh,
	tglx, peterz, rostedt, dhowells, edumazet, dvhart, fweisbec, oleg,
	bobby.prani, Paul E. McKenney

This commit adds RCU requirements as published in a 2015 LWN series.
Bringing these requirements in-tree allows them to be updated as changes
are discovered.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
---
 .../RCU/Design/Requirements/2013-08-is-it-dead.png |  Bin 0 -> 100825 bytes
 .../Design/Requirements/GPpartitionReaders1.svg    |  374 +++
 .../RCU/Design/Requirements/RCUApplicability.svg   |  237 ++
 .../Design/Requirements/ReadersPartitionGP1.svg    |  639 +++++
 .../RCU/Design/Requirements/Requirements.html      | 2799 ++++++++++++++++++++
 .../RCU/Design/Requirements/Requirements.htmlx     | 2643 ++++++++++++++++++
 Documentation/RCU/Design/htmlqqz.sh                |  108 +
 7 files changed, 6800 insertions(+)
 create mode 100644 Documentation/RCU/Design/Requirements/2013-08-is-it-dead.png
 create mode 100644 Documentation/RCU/Design/Requirements/GPpartitionReaders1.svg
 create mode 100644 Documentation/RCU/Design/Requirements/RCUApplicability.svg
 create mode 100644 Documentation/RCU/Design/Requirements/ReadersPartitionGP1.svg
 create mode 100644 Documentation/RCU/Design/Requirements/Requirements.html
 create mode 100644 Documentation/RCU/Design/Requirements/Requirements.htmlx
 create mode 100755 Documentation/RCU/Design/htmlqqz.sh

diff --git a/Documentation/RCU/Design/Requirements/2013-08-is-it-dead.png b/Documentation/RCU/Design/Requirements/2013-08-is-it-dead.png
new file mode 100644
index 0000000000000000000000000000000000000000..7496a55e4e7b41becdb658a2bd34765fbe80013f
GIT binary patch
literal 100825
zcmaI8c|2C@8#TTWl}t}j$kbCx3K5Zr>M<lKr3@7^WS)o2o(2+1GBhYeGG+=PQ>lnb
z5(*hAnadoRertEW@8|b<|9Ly7Pv__v_P+1yzOHqxwXS{jsE!6ZD?cki5bOsvX~zhH
z$%G&n9aga5CvPs*`Qi@-w__UniR||R1Na{nvqKs*Vu}1Gu{<>jKUwLlY3PQxx01gZ
z_KU@N;fKrJ4<1%q{*{%ohGTi>PcaUH5FrlI_UU=G4SzlFclgiJ^3m2e#`{@TYrXrd
zCi=iA#kMd-Ho7gN&_g6+I_)pN#K@6M>l2Up)zT$SjzqR}Zcwik%8!cc`13nLA@xt)
z;Y}2aviOJ^|6<4YehF?)@3$qmsR%REXmrLzMZ$@gVkl;GX6c@+t=zSphv)bJVMfpx
zLUzr_^H+HgTNv)K+<1A6LeceHcbcX*Z@^^D-pxaFh9-Gk)jW87jggU2l3w~TXIIyY
zj~`2H(|EPhJ0*4F=FKd+y1F-S-@e`a`@JmDC(M!;MR+n6Hw;oJl<IZ5V}6zLOdk&E
zMrQ|~%{*P=`fXq!-OI-(NjF(N?dSu>d-v{j|M;=Y*Vp&t^;)`4>H&9mSu-;;E+L`o
zPpleVbae?YE-n_k53K&!)YN=%ikN0FSZTVEu5dCUBxIdAO|PJwFilGmQ>Oh;Bb;4a
zB7c2-w#}if;6iHZmX5Dq4MmguZrr{dveW8)cYkBDLAj6I;=-S<aFe3Yo7<^z4D)QJ
zQrpyMFGtmAYKK-5CgPzo$7uFhSy>McT;b44*y}&qa?0uX$B#F8s6KU)#$_Qq+fJJH
z*2NT`{~iA~UzpxVywBfH!>^z?X8MNEq{|FkTefUjkni%_h(*~?YtZoVDIM-8InFQd
zxSd$WC+M}E`DnbbA&vHG`KgMK(AZl<7W2`jx;m}BbHh6=JBnSzPGoRzJD{hUb?Pa{
z_t)2|udSC)6w|sH63#rTyzKx@Z}QHQWi&N)eKp-8M(5v+$;)>+v^^|aTAa(xJNtW-
z#(eO;qQ?%4TCU>a;_jKT?`z+@dE>;8#%j7;#z%V-HHjK3VL}T^;G+)^uiHL<uKo5Z
zET*&E_lY<ysOnvud}E%u#_3{b24bF}Y`I8C%m%923^nOu4L{!*gUT}x_A`Geu!$NU
zua_d<pt*lP<KO9#F!uGkRCS}BDc-B9RI=D8x{cNC99LPNCNQ`=be8R&9{!T7mwECU
z&jxV|EbQpw-0$iaFBs>hN1o_&r!djG7+yY6q|w++XnH&RSWWF0{!UqTRrs^+X%-PL
z<5qh<lv8>A(<eOvMfbF2bp1$a1vkaa1OIlcLI04N?#a^PVljN!ajjdD??SnFg5*8x
zV-@`W%4Qpv|7E1R5{qUhLPA100{X(zMqQ3__r~72bLUAyLc;j#^-j4|nwqAcpOUe6
zFHebc&qc<>J#@J&R@{Q=>(yL5icD%RU$wU0dLEXx!qwTiYOpzzM@^S7-M(!<P0xL_
z=s?noSI2at3tL)RL|Y!8O8wm4{sf<R^p1?>r*_VIMun9k^b;GX;`5)!)%1!dxg*CH
z=Kko{MQ^RP<>EPO5-^+c*PShMA5BkobPsl=-q6i$;#T&VXTjnpGIIH8Y8~sH$}TW1
z=OC);2_1YgEoeQ0hLv+Y;jv!MR&V>ITgEw3(_09;qx0?~Z72K(U)gd|r2Qz3Pmc4@
z4W|a1S9Ju7ii)27l5cs$VHT0#)f=Oq8@ROCSDFwI5D;?ps<-5j&cTEGnjRg_vuY5k
zSQARQ^vBoZ<fV|1p0JR3K~5s=<YS`??YWJ~8p$S=0R~StJ1Lu)6gzKdPe921R`4F*
z<TEum)iFJWSo0cs{PaWm@zm7(`I=_aucaOeF5i~<Ob@>~zhByHG~kb^Tu2CSMM(db
zbPnDH(Xy&y|G8a+h=hbh>x86X{xXEo%lG&9Jh}H10Z9;h_B`t=ium*E>j57h#b2G}
zalh-0i|qOMWv%ItPd#1R&;Q|xex8|{!KqVrzXPfN;?$~Ikf@8<rr*75yR<ZgFtxLb
z8}hblh`&})AS<xfPoduCMRhd;A*vN!+cf>|{+@MReiQH3_*)Iy1a#RVPBo7lsd;eV
ziqI*h<;$02F`A@ocJ7RpIJGu3sTli#bHjW{RnJ<C7<KM$X5_1|x3?d6cbQc<vRdip
z&6_)HKZy~leedt<mbmsRjU<OCHLn`3S-EPJZxW#uG;=K{*W8Rs4VI!NSqW5(uU6W=
zeS3FL4+FlcxU}@Aw{XI_`dcL>-q-D_Vx_46UT8K`*W_?z_evp7A*EWk(cFT9m%a73
z9<&afJsbHaqO9-hIM4d^aeagq5)gTQXb&GgBnTX`c&*Wph={<X?z894#U@W#Xrwk-
z`dJ(OSgJXv!<c5oROE-;%ws(4d?_T}0Lx1cKlvz?qO^MT>Uc4~J4W%PZ|7Iswyigb
zR|!ejM@U?ER}@T2N(wh##m?URUZ09PxP~+m?bNeP{_%eE*Wu4EE$963#Oeqe?+zj|
zGV->JC9779I5SdVwD<{DK}C1^ojZn)P9|KwyliH6_AHL~=gL48Y(`Ii{|YB3r?=P(
z%erW3qt-W3TXdf~p85Rc%Z3db7(9krf>KgM7iar~$$WTA^349DN4b#K$9_Hg{cWn%
zq@h46Jw2UB^qhM<HZgH2B7)PlHT&9{Ejm9k9L;O4tD%^bdX9v0s?2jC9~-fM(zwDQ
zXzJ(~zm>ju^JcS#c;(MeorI@H+OOp1?od!ve9_dj#;xytP+VNx8xOpKtUswo5?+6K
zV%0M+a5Y}pH_777%_Dw7_jbFnWS)HdqTKgSV$;BfVi$Aw;m^|TJ(T?i4v<IbbcG+^
z*|TTUOe*)5m6!MQ_Qv!oF)%PFc#VoknpA{iA;=nN)AT6K@X4b_cL}3{Dr9<Re8%ZA
zFN%r(6pv24($$YChmwyZ?7h9yiet|c5d;M_s)9*m?$7UA7hg9wpCD?RGxgnM7jXT|
zbm9hX_-aV)<Umtl>!+HlYqrp6FHu;CTB`*A&DbQ$mMy8;O&JmSmi1bGeq}}3I7X~Q
zI=(+7<mKzv_)p{VQdygiFMr7J%h_&PxpJj?yVChzrx`?Y%Em*d+mQVV9Wo8_%(88Z
zeE&E)_+u$oQz8wYoU?3;-KcRrBqXG~yqvhvQ+=hr*ya1nFJE@o^CPsia7-R~WSdkP
z<80ot@i}$s)Ey;n*Z7DfTzzk0tMSefiO{6l#C@0YJA~G)TNgFq@w-#GN&3Uk)cDBL
zuF%55!Z%aew|DFV0w^3n;@d<#JoVJk454ikKhTtV=<MjXFsaZau?O=jHX+%1wQry(
z=XDhMsnKXvZEgH0*VSL2op<?BOL_R{QIczK9dN=0gjL4L#~Fp4{qBm}PZx9flnXsM
z_ijG|hNyb?j{On~`-_)X*XU-Rlpy#dwr1+*w6vrY-A$iJyMNKk*LS11IIE|p=jimb
zIu3l{+Pi^^v&;SGrsdksq4=S`eaO_8a{t{ai0I6H`m{zka6U5gu9~{~3tZ5zU%ytc
zvc8DisBz=oy>R4uYq>PN^qae~<8)H15vs(YLx<kpmOk+8ezx`<8D0fvk<B{yLKHm)
zt*6eExTZAz-McV(+<o99^Pk^cHT6=J;beGMR8&}Z6ieTXijwvi)bpF^xl%INEG;U^
z9JWsCHC~i;t?8VlWf0JUDL%WuDV1}0WF*hJky^I+cSuu5r{(9@a$osU_koMGwVEho
z$=KKak11hAoVa)OXPzVBii&sSorKh{@drgma|2!^p2$4e|0#WqxPCSp<vGGm+P(j%
zgHNHTw(|15^P>mszn1c#cr7Ee@5-)R&LPmf{x(7+L}qKQgo)^(SZ=~~sO7Q1;hff?
z!oGKR<<<}~jvWfgHlNnAl^aVDn-mp|_Abr4F^Cu&8|V8>*&aN6xcO0~`@qeXq{hm?
zz~q*##J8V6AC;f8vDtbeLoc#HW$E;<juP{N&Y)H6Z0lFDvVJs9P;f~?fjCqjD{;-J
zbMRC8#&wdXYdzkv*^Q$5uOMa-INB|$w3i(nr(e8y@v^m5-|qgYLfaT5QtPJE&pZ#e
zXQLR|<p}TfGvHQ>TrH|q`}*4YoeuTLc(2~QJ51Exm9uj#<HUVN+csys3Fq6nj9y^Z
zs)KNMls!zNUH+7D;t)}rnh=mt-ia*1AcAm=+NPh)`|gSRz%d7vfQ2c23gv1<#4Q=i
zI_o(hAt6V-dX!r`3DcSg;dPYAt;Zj&Ado1oWtmiVTK{$YT6#4oC<x{3Hc~Kt>3O}l
z#jnoD#N?*zgatj!!WpBJx`s$<8n8e~%eQS7Pwr0A!UjuzW_Mzr2l{2E7e9Mum$O?@
zk^k$jTg?v9!TT(K)JA^9s*a6~MV_q=zIy4>CBLa=eW3c8ms7rTBTt#>;k?q+dK*I1
z%q+<dKgjo+b&ZnyFyP)c;(;A%8#)}fi-#CTq+68CcLmHHCNzQM+{(TJkAA$^`8m)0
zuwxYeA`ya*-Xtpe=2h6bLUTF0mMr^*uv^AH2WhmH>@%6~o;bAcikdS9(qs@(_L=0O
zMB?b^5}Fe~-yP20C@FdUDRNHZ&Q~1rA^mSuG=(}HZ{NzxMJO}Ox5msRBSoCGvAK$@
zkbEpf3yA;9n}Mvyk9&UnxZK}w-BIds$oI_G&DGa=hKGl_)~^q{eVaF`-qp?R0FAb6
z|AD1vo!25FG&D6=;61vy2B0b)z%<JWcV#CltL22|p+oigETKu=aZVK%o;};+^XJ$7
zUKeK%4>bdW+b5aIjU&iB(An9Uo10tp=y06&`}GtG1EJtMy&ZKyx5)lWjpNhg<c&l(
zqRODc&pWnB&Y_Ll(a~|m%9YOn^$LU}frlD%OjM#K{808rCnl<AC;BKnJVA+xB7|xX
zE6<1QGkc=u_WI2@BiPiR#e8`6ikaYAyY`%`;BZmGKZKR`L09UL+l2oEjY}GyJ%>`8
z&LZ>;ik;6E{kvCp_kwA`gy9$KM!TwBMI7@zq*uA+QE<*sYYw?+yn8&i)0P1r7@!V6
z65^WJLm!;-30blE@ZHP80W&>$aw}G?YO>|x<z0dLUSQkI`n&v(@%19)!969cCWb-k
zF-c-#CKXEi_wPp+mn|4|Q}oce!8^yA6ZbKV%m9yW*tBUGakS93dHb`U`hB|37WI9d
zWtFb~55(A4w|>8P<*c>O>1|P1VpL0e*<Z2gqrK=7i4~lj@(a&roRx#W?@RBguc=vn
zrqum;_w!3*ttNpz_s@4e|M@PiaG^G=hWTi<Ltsa-HP%+){4W9Ev&7aR3O|47jT>Az
zJiI_G?;EFo$ZBVN?acf-YNf$$ZDV88HvVLlgYEJygVSj>o3>1()$2$2pY#y@4_@xq
z5IntG>Qd|8U$J|tbH|FidxnRz95l4FSP?FT4l?!<y4Ct8pC+eI@cxSM2}_EKTEoI2
zumVv5P-*tNv)pTG;ZL4reXIaG@BLfE-CO)9EvkUMmkSH!E?&Ci(p`1IPq0<XVOEAs
z>rqFE+j3x6W&+LAVY`vgYu8>OtQ&LBR%=hpPc*2sxvgKnUe@;03c`8p8%iCms<~ON
z-W$a>e3AFYEfrQS3PqWH*fu^S{`TLmeAU@66E=CF!DKoR%(hQYPcIvbTy?Dap3k$p
zJt8k`Ri)>3Zf%${JHsAR75Fqe{kg+v_w$*`r8yO%>bl^b9rmAdm0BD0>h(=cMS?j5
zLqkGVh=nEvJ*p{OR5oHlO_xO=x8|7CWEnmgpDNCjFbRI@y>}4p_)PaDPJ%$={L(I`
zoQ0E<Gg?fS5#<&owvPKcS+IO2tqE213q9T4i9tCysNwu_brw}|fIal^-?w`=aj&M3
zeT{A&tI42&F=~k_&L|S`m9|1V^PU$$yz-9sbh~{{kM6K<);&;rf6sZMs{fgqExk_M
zFI~bo`0475@3wqJ+u4Qp3ZPIDbhWUxXUD$1^3F&=h!NcgAo2iSUb*Hd3jX|L(-GdC
z=fZVI-ySATBt!_S2!$q@zPP+{rfX@@DnCXV$n?-HNwuerUvrU;PY2}y&e7D=Hpt3y
zqhF~;2a;?5MTP{!Q$wvadFC}hoY#@UZT7i-dwG?hA5Yck`Fg(V6~Hb|cFKau$PSB{
z(H?>h(EI>+kDvqQn`8Zpy~e22jT`gM|28gg>Jqwo_3DnDJF9POI6%y|=G*X=T_p>u
zYf_b`Tvx`4Z44slgX^ctq@|^gW$0~X5J6vRbpG4nyRwm}=@iQ2xyeGi)^pFYuqk#E
zTSP^#hG^m)%ya;iGjWc?pYs@10ag!7_BHF9AeRF*Sm%8C@?|9f_<Q8hSgD_%-!eL&
z`5RoG+nGzn#R@K-p6~UTq@<*9C0vw7W0MkB%cA<hug{+DsywSdSRKsXWc+7z^oCUB
z{4qd}O*_xMy>Y(FUw<e-nC{;HVS`3{xyyHT1`$As=AUv?&YqrYRa7dqJ?z?x9Ev6Y
z|0Ry!S5i{Szi-u6VAD1kwdF|5YvyflKQ}8OK~bvpAC1LpU#pB>GiX{on0)4$=W1bL
zC+>{XXU_EE9=}MYYD5UN$$WZ#WjPz~4I6<xAl}fDl9B?)uWdhESvZBZpFDXITe5Pb
z5D@whD)Tj#KPN;tZw|e6>lSh2&Q7bJ=`1FJ3x&GuMo%3bxZ{!UCEmY(A0b#<>eN-y
zCc|czcHK86So3(qcyE3E72$x}-`(CeG#nu`N7@Q^O4Pr9uY)zScooXCT2SzQFRx=)
z;8OY8ismcH$;ofy<d1y(_)+HFn)U0qk??%mwoL8z3dgUC$u@zrz3Ui604gkINYI~@
zM5%D>s*oVx-O?7lZqNB&(SQi`h@Zn8iNn8tFMYO5+gA^;|IGQw;lnW~s}`j<R9t^(
z7PYVE=C+tzijI!9uQoxE!s%51)m53$$3NUze#+0^|Jwe9p1+5if9fJFPzW=CDewfF
zPkjM)*huac7NHIa+yVvI6({Dj)9<nGH9!!>#kDz=i{kB1oo*qV51N`LzVWaUm6JOG
zmQ3tak$oiEsDrk)gXVn)H8pPpE-mPz`cYf6&(yop`*3HPXgjOdk-NxqZS=%ZG$2}t
zMN5ztuZmazL&nkJ)rInGi~n1SPfRORQuZ39ZxJEiqRjH_?v^BKGVSW>&-s>z^0F_K
zmn-jb>YRT&_*hF*^FwL;7AzVXZEBqc(tp~ir;^8x9gDZVVe`KqC)pbq8JPi|pb<Em
zNnXUik3B4B`^l6-o~V#Z+t-knMkbyA>(75LB!BYXrT_nb{=Z(xvzP4r0vG)v+jFoG
z>z+Dw{bbk-$ZO~Pt7G@ti~q4Av%E@PJB<COPRz{Bt-Is@acHXl>oa-qhT&CJszMfw
z1RIhUK$aR<x3GOQnzz~W)m*wKEB+2{lePXppvm4b>qdEw7$NBnGSSD!$E%rD9yoBo
zW#FSIQ3dV^%!Fk;+cL^Nd<H6_sgu*v%lc<-eH?&~$V4#%l=bUZ0kEK&sL})&vb=+z
z#c_zRsD9cQp-HS`Q0nH}_m6p-wLs^5Th?YZfENWv&h;*U)Weh%20}nUV7wq6=>`G9
zo#u~h-x_~QO!wlJ>YaLA4a;Nh7`}c-RlG0RIx*<3NH)<pp|8*hIZyWMc04;TB_Ec{
z@Zm8%dOv5k0KJ;vP@ey@x54<|Zq{v_n$tl)WvH`e?b>PD2FuPeZ(ccD?w@b(TmqTQ
zQ#%^ddh}pP-|ycM`X>H7kBtgn6gjlBc1+)ii|bSLyRYoKcGcSLRT(ET!yWbA-Q8(%
zyBtP89+G+v0EI){R*gTeU%w8Da5iOIxlm(cV@P;7O;0a++R55lOy{1$S<k<d11tjr
z1MCD_Ajl-5YeTY!WQ^6$e__zuvwvsm22rV=AwUo=%Ccq4aQ{~ItrHY(?#m!BPge$%
zt2&zKoT@l>{P-Afvgz}S%nCk}V*mP0lq_|yaE%2v$3LYX*EnT-mRndj0v&x1YSYZz
z?@AXGh+t~Iwr;c`b)1PFR9mbT`7dz9R9@&?{0MA4>20$CeLJ^S?Dnft6*EbyBhPvQ
zr2UlljI=+k`SB*IvBa&9aUs4|(hsaGAL#8@3v+zudyp&GkfurMWV?h!*xNi1ovP>~
z8cRLwh$<v{X;1)J26;*LT3U7PYPzpOYj+U!Ugy?O7Os>Fu8thwydp4vqI2|_srSmQ
zG2$m)pb-SCV~{v~=Bt#Oo7?BlpGPgc)IoObu&m?$gl1B;a*K9Q4)DWde<KsRgZ%~u
zyue8zNEqbJ_Jc`{WHo+-#%Yzx()>>Jbjf4~xog)hPa=RRmzBP;!STO<P}ec4V5_V?
z7G1GEC0MG+herw!uGo8G^ZR?d&GYxnW*Jp1&QBb{f-WbUSlp(|_>Y6zjSAALf@@RL
z(OlnF^bprc*!#F_e!QOea6mVDC!ypWOreYw9iaNqfb5czlOylm4FzZNqN77#>(;Hg
zMMdvr9*TvUeVXz_3WJ(pLZY$Uyu1rxVXHa==0<8*O6^*+&o0i-0+tAXtcr|}zuw(#
z`tsGQ2j|}1HS|<<{`2e1Jw*>zz$J}i$5w+~6A%>a`S$H1_75}&O9|_-!Jt3M7mlSi
z^?-yx>QF+z(*5h#b<|Lc5&<B}(NX8i!NEyLfuxACeY@+QyiDNR)vFmn9z;u?Q8bD0
z0995OD`A&^;wQS*nZH9h2mvcuG}-OFh|a2yA6Juc@+LvWYm|8K-~oEM(b3Ubmu*Mw
z?YE<O36?xlns_j3Gl`6X+1ESwykI6l2*efd;r#q|jhK+`cBjf$W!@77?aUG&GeMA{
zckH>uxtER(^-XLJN+%E6H)$}EJIr6Rg$pSww$>DjW#}fe08$38T(g;Mq|&p*>WnEi
zq2Z2$B%Wa{`%Tm?F3k1X@GCec5;xFRgoS9@+GY@F*w(waxkVvy)&U)>(Jau`FsdTu
z&}bF_UgMp<zYo5-DauT5Oh!U*UHI#f?D6j2?&GMzKTzPQ)bU0Q;g!TdYfb{mm%n}Y
zu5h+KX4@8&zR1s?4ICDM;|9JwNxgmNj-^pTPc;ha->IQ3$OOAHp3@nameI*>9zCWd
zLg^#=u)HTv>I;UU_8dpEBO_ssE?T{vQx(DfvZ?7fvQO;DU-Y}ZKvE<<a5W|-W>@9?
z`}e^W>Vj_=ceih9YU%|!a1%89)v&NI;sz+cu$-J6w5s*iZIw%lr{2DO`?9ms7}tNR
zR|X8I_|Km|KLS%~tL$;^rn~>@_+l-L@JRtB25KutyXiONCZj@IQNU!h&w_hAR}kr-
zm*!{t622h}^%_V3HtFO4$p3V%>4mb{pK)z_W49hx2N)=vczZ)9H3~GRZN1;vt98Av
z)|to}9YLw;g;4aZr)LB4<HLgkRBFD(o7KA9YRENPNtQy!p)pB~QI%V*WL2VoN*%e4
zJwqF&E*1hBsQ1*5HWoRYqEPtucxI7oOpbB+L*sIv&6gsQ#M&k}_s&#VpBYs9<y5(_
znMzIY8^21Bcu}TaWD4TjK_jEPAt5B6rLN-U;Sp2nF;s`3*$OBZi{rw(F#kP5<-IOD
zWC&3X4i1vUNI#ZRXa1+YwN<{J%g)hJk|dkZpd`=YQk`T3{LxpNIXK+!U3~WJnH6i^
zZkO3{OPh}m560Ge;0us*`QGicYuU~qQdCs*t*`Hv%^jpB2tO}DxuQ@Owr-?SE&4w^
zP?z(WJZo;gQDtE=s&V}9pWjl%!}Gs8snpfHykf*doP}daw3Q4Zz#zw-c{(_J0m1n3
zCrM%bJB0p9F5fOUDpt<@+B3qr0zj(OxDpx^$sWbEK7YQ?*k?eLK?GROVze&B<@@XO
zLM0~pp9!==l<HeP`xe0D)DM4o(hH6>_Gzcjp_VMejUfACcf0+pMZOnRP%xm;DrUZ|
z;@j=|aM|+Z>(A)@8EQR(Lbd_it}Aj-`q2j$TNqQC1`0e*WEtoZwMcUXy$m;S<p)td
zN!;p+gt^<j|HjPUzqUg&1E~p{0P*b3`u5#F{{tbh`HvN<R$Gt{@PM1%C4laF<bxl`
zxatzSs;_X~B2nP6V>h8G?n84UN<1`s6?!K&^7d^p2v9)lIfc3QT@`0mT}D)^?^Dfb
zwd-M`&&<q_m+}L5=eo4$4=V2_mdCEX{u=5iNm#eseHyUfvyzQ18YdRSe-uuKorE1u
zhYVi`*#R?t1|RUa@aM?aXZpSi2-%Mn{=UhHtzc!zyV<mPa~%T60-J7z06BN=Ts}B^
z2V*(yKLbq~nO7e_eyp%(PZ+uY?S#E&(4r#OpmSqX1>O<ez1t9QvY-uo)v1;o6DhQT
zAE(E_Jv2^_p$(u?fvl+L@_C4SvsbK9g>%p;3fDfBI8{`;N~%m`X|8Lj;OM*V7`Jkt
zsbdXSgXR?gSJko(t@%G*06<qJ-0wPqq_}%!lhv=?1k$gE)Ijq2IHuqus7>yJBbSMA
zNBQFizTjJNfzTBhRedJ=$=x+9+U(K)Vc&Iu-ANKnU0o{e{x)p|=`9y;u0fL&lMuLQ
zcwJaU66Zz6|F1g|Prd9JprHbX9aC<{zkU1m5SfHgb@P$<tnz^Q85ajb2ofhMgzEWr
zo&y)zl640}`rUiyYHw^XI_e`^Txi|6e^p{%L&7a^Rd#)^!Si2xgf#R7iS$i`%A#0y
zcD7-B4teLx=p>!t`xs&(qbkJX%yMKtIR+7Ax@V2^DBt;DMGcFcAL#Phcb1jtR>=e0
ziX=PZTq#Eda(jIuvjp)2I3eF_%xt9WYk|$0H76v&*J^@LnhZ^P$dL9X!ksF9ede(#
zRp#J`>VceA{2MpyvsYPJ5O=KL_LGfN@{y14oQ{RiGd+*%O+!thhhOI)tOad8x3w`6
zNYAELR<~?+v-8Um#4;8Zeh-GF;_FJEdi&f*J0~aI!I5ti6I+3f<`VMAn*lbHN}nll
zq@;%nOCTN<oV$tl2`UkIC7O*FEiG#Sj`o8`CpDgq3jb4a#)9tpIju-Hf+)N6OP83)
z8vpGZqrSdADF~+5V_R=kzkbaEG*<7rxZyx>H!8&oRKHOi{pv4Y_y|*Qo#2niDqS3~
zp!l;PAp&8|NUYs%Kt~eimq-ObfQ~}?9<brpfqn0u?gE9h0~!G^K==IYM4En%@uLX_
z;`Hg$TwGkHvRM~=-p)-8r7i?QHU_MD4vyu?@TeI!+n~Uj60o#z-tW)X^N<YQH}3xQ
z)XBuYtzad95vaKu$SjR1ht{FIMGAA5jOHF+odPKIlQ_rN|1yK?U(;mD9e}Ra)Y3A>
z<}Uh_xADsBr>Cb^u(7>$H4N^lcP74Pz0k+C6xBmHJpjJ5pjGO0aU$~*4fN&TQ1;b(
zB)0SnJ!#l`0A=Zk;HZ4&0ttJ6zP}&W>xVP)8Z4Q#*~=9pa!^*Y&OBR7N;&A3T>v+=
zCk)|X!3~r2&RSDp=%PKKk@~wTRdAQLM%)Lb!AAtItsttfyx<_J!CQ|(lQ9J~M@p6~
z?EKL-y9aMOpoOXLP6;kRIU+T}EjssVW&G0H&l+@Tq&#{=-LZoU?Un{U!L2Of8o%5N
zoP0VMT#|XjW!JlQ;#lyW|J6b2CL}wJgEMh}%GnvX<PUl2JMy<6Lm5l=j?Df4l-*%I
zPr;m~0qtNZ%04eI1=QU;Bl&EnLfC4K_)MCeDtC?u4V5Vo@K8qcE_S9=7O0aVAP|n*
zCRq^>)bC^)P*anFN&sm3th<^QGEka}n)rgHJSw48T=$;%$v`N24zB>0^Xc)abp#iv
z(hnbwZrZd-1m`AI=e~@QG4ktZ-L@Q3+9y#q*);O;U7eZnQUZd+34I0dk_YBXJ8l>I
z_nL^n;R>=1MHPk9;5j%Xw}K_y@&%;TF{xFnR*8i&s_EuJN`%nz?nl|6wD#e{{=+_S
z1QqPC81y3&*<DO+J0wo1suCn!4+bUEe{yn?MqUQ!v$N<7z66s!#&BEVs)Gj)-j=hA
zZOlvUxU<XQ?%yl)3P|Flv1i%$*c^@JF>+-ha`W=rkeJ`;&Kx_n8wXrlQX{98sn+@q
zBb~BQy*o^*g5(vAitMlM_5Yg)IS=COW~9ShpdEl$qqxe`CHJOWA9tF~7LVO!ld4*?
zW=&!47OM5<+zli$cdbAgThAk|_hGu`=<Im!HKUh8>}=akW>q~r7`5Z<i)DIQ1{8$o
zd(ZV}aN5?by8_U*K|w)5V{Xa%wvMKzD%P3g1d2;adcY<Tgsf9%M}Lb-^=M4<Zga2g
z4GAhvdRE7K4k@=hvb8LZuvK_BvNr%g%>d2_T)|}|1B)*oOQ6jND0y+k#>OHOEmskV
zuL5LRsiC29_1ZQ6xMq>0Z?D5eMUpx$=lu9#@z_#46&P*K5KGwRHsz~iI~3fudc6BN
z6SAs>(RH5<*~Ske_~lsl?gcYB(br&(lazKMb1i9L=<g4~(pf(ij|i?BR(cbVK7=;H
zuv6U1!QmRfrbKS99#DjIz`PephZ4EgjY&1{kC(dJpcCYTvkO|nrMS3ubK${}gbY&0
zgp7dRRB(q=r_$s|`#ROve~KiHWrcR?3%>p9GB=IBSp@O{Dt6n*jfi;8;F{A5@|n&i
zImXKOl)P8>^z@XpX`DZQ{&7k^7&-+gl8<@=Ay<(Z5E%UfXi_i}qfFmc8W<QPI<yzD
z%DkS>XbAn6oOpbLp9-qeXasRG#~7#9wi{J_pO#kWS{<R*{T81Mue4=Q)~reX%3**z
zAa?wXk4$fonqoZ79*hezs0>gh)jTgRFN?bzgi(R@gRj%Kn@+uQ<*#5nWvz!+_hOP*
zv2zb&r{BcksQZ3tQ#(BDI>|GGB4A&wj@Sd;eYKYi@EhnE(fMqYm6fd@pe?&4WwOiG
z@-Kp76f2MjfM#U1cGWsb8o&YRZxd4X;bpdz^HXly7c8!zz>Cw@Y7~J>xUkOHi$PVr
zKE7u*k!4KdQubCb4K35snXye-(|K4TGWWsvLHgXz0^*R)sl~-PCqS?&sG)I1h|qcG
z+U38?Ca(XDeoQKkdLqhA*+122q-pB$oMYiL7XmnH&DHDIW2Mbs|43W)hh+QW&I6B7
zsjlD=wZArWRR)^)jsO<kiG5;Ozm`D+#jws*q=ys=Uq=W_9e<=93I9VVluLd6*6p`!
zv?7F*m(k1Sem(o5^9NYYru`WXVRCEp^9%+}tfU3l>U>~863w9i9V8o`&?H_7DKRlk
zbWF?W(NZSHq^Isn?%7NCR055W<Rk$(+YAigH9!({D^NkMJO;X-K?Lj%i6(FdEi9gU
z&;Ppj7ENP@evTyZ5JCakMrg|+up!{2l8%dOvB;K(9q!3h?|UTg)S1Q3vhZe00_x*|
zALvD5(a?xA(VLOxqhKm=FrHFanE4*jW(STNq&iBs)w2lls2QA-l$8AUy;rX`mZqrq
zPG^zdJ5;)OO}?{MwrVQ*lIQxpXj0X5xgG9ITHp?{TRbIH+R$F5iQHvX1-H>)<$0k1
zQLMf00s8VrmcRFr1Bn~45~t2Ry9d*w^~905T@p?C$w^5CtE@d#f8e!u1a|Pzk>llS
z<`6n!YuB!2R3-gY$=-Wse?)12KYS-Ccpoti-`zP@Cg>b7_%x_b*L2y@&wC()@DO)i
zfBoBg%jD(PugdjY^eas3Gd<p>BOI+6w9lx6S1n|?%q|T1_QrBfA?tl=g3IP~p)R(*
zo>O%Ff!=Vq{pk@R<jNI|*mjfjW1@F<oW0ms0Wr}|WeE#d(AqrW>#ihy9PlZJIk#ry
zd-C@?=?wNr_s4MT929r%-02O?@gdBl_`!hHbmysZACFj%z{z=XNqRvZ%;tB3gZu)y
z<9Jskbc({)%V!#O5WLK~M+h0K_xG&+t{X&k3_~-nm!WqMP<J~pj5^b(Ct7n5AJHQ+
zY^O>d9ZAS$FK$EcFpDiz%V}NDZru(N#e1}S-=kjC##-3kUV-rU9YF^ogDhwG)bW0!
ztKPHozgr|<vOiVclb4tG5#8@f4vzOO3Jd7f@*QHnJatO&TbOG7SnhkC4<yIwa-S1W
zIkY)^?qNenDC{@>X3Mvc5ecw!-0(-)d@U_}duL}~UVF^fpub=@MPV4?+Oua*UQtmj
zVB#TQoFC<XI-{6Zt<_XjrH6!E4hlL*Rv?y(yDVRZP{sj3YAgKbXf(1nV^q!1OQ(YE
z=0dgiMs`er=p+M3xEtL-(WQOzb}iZ$FJ3HYjg>NyM81dxi6`*;dx1^ULqz_a>BgEG
z4M0p0dB=`gPd<?K`EGqWU-Hba08)k+=d6QDF2MP5;UQ?ajbv{H3|Ld~^pgs=m$$d_
zn2(3zePEohBryd4x73OR00bZ6*8lb=Ql7w)bnmQG6Wg_`#Ul4jI$|gUkO3&V4hs3W
zM?1bckvWkx4WM@r1-E`PK3?+q^XDHSFf)iCZ`b`PUB=9O5L!YC@b5p_;$2*P{Nib+
zxw<!FvOH|UeSrLJW8inUkIvm*gZDD3D!TWNY8UyUk!4hk1<eAT@FogqZ^7W>o7e`w
zZ&R;pYV2x$B}eS_8Jsgr(c;nN77Z01q3PYr&Gb%s@tvQD5>!xfu-(Gy-HWnFLIDsC
z8W2t<Jle@r45v>wxM5O`hAVQVNOFzReVtT~tPgOQ`Q;qKcDbP00vKXP+`exv3%-9t
ziSv=-D(~o&l-@P>w~>yl`)S-S{ok%EWQwqoE4tNopbxc!>?oqQ=}Xr4EdL}q!R|Zt
zc3ZXa=EGCF?sfr4JL3Y5&0B9t`gR$7Esu~iadBZ`L2X1dcSl}f;VW27BY)RFMmfiB
zT>=x&Kec%4=FKGL97vmIK{_9?th{&sej_q`{C$?Qs_YKI)vVm-2E8Y<7@sH>RKv&+
z2^SMwDi&W3rSMk>$!8nsptf2o`|S-Ioyv`8?UEyxs5dOlc~(?X((zl8T3C2#rQ|QI
zXt4|E=GLy+tX&1ut21;s7(<>nH}jxXropmf_|OW>mk_d!+@w#r?w_S-H`{G3C=*AY
zoI4D`$P8s?bB)LGrS2QT5+<tZL;5g|7ab6U@J-=am%^iV(DnTJ*1m#?6;vM?tnpex
z#`;5|Rel9@MVqejyDEVd1Q+Pi5|5#ik8DUDhCu`_#5cD*aI#hr105w3$GaNPA7suQ
zr0LoJ^ZSaPrp2->U%svWmXwuTNT^8l$t={~Do6*4_c@@K)yh6rtnv;h^h>{FsPg;w
zrq%y+>wRMq!~)diA6Ts{7F~+@%z^~+9SLxVQIM!`M7*>%nws;N9lI~pU8y4vn=EOS
z6{}_^j>9>^BP^U?HU7$r>NDL_kB$&>G<If(yu603t$4uvm_}63wo0RlPqDI7CjYz&
zR8vLz)RrUhFF_|8uqRP@D?Xh}L(C@wX%KW6r1+<(>C-y5Pt1e@nj>M3H62lT-rBkj
z3JJZkQUwylVVxY$g0oIe4NUuJliZ_6PSJv*i2mh?f5FTQ>I2B4fq3`S<yuJptnQ^q
z{w%^2dFFGm3&=5QuIGr+<WOtF4AVZDKc~j}5(3T8Kq2$jPVTg_vVsW7f<ASZ&lz?X
zfOOgs(hJX;vy2erQCqn<n3+SG#5eBU8-HShI2F*erLxu7mmP8zL5CYhv`Hx<I0uYK
zbxlo;#V8{^ElsQpR3Qv54;qK;Xa6&6WPM#rIYQLP+kd9RxgsJ$I<dtWd+Kw0R}rMI
zjHn`2ClIFFt6QOvNW1;i9$6w=MdTPRC^j%Onj*f)H3rR3A_)jMXX0W>(-p2cx_pi>
zB4(z*fsIa1MuSngbLrt~p1jfKLOXpNMh*f^!%ciP(ruw7na*)}8}7<GT}Qf3h8F<J
zL&mN^Y@hmdf%JEo0glnz4R0=o76{{Y{td?NjYPG%x%m{7QW~x**RECaWN_Mz;2aHK
zfdcq|!A27VY-F;3nZnM_$oO?}3h(Xsp$LKc8qy3wpp0EW;v{BrdNVPzp~Us?swa6>
zLNx`6ILXt+bkC8t2j~q+aRKuNwJ%zeB}|A>2vgM%I$*wq^{9T*4N?O@48hm6W@2=7
z$3lJATebxMX+C;*OLnqGglz-g{{poyo6(C?bth4Sj|~F6`VP{o28%-x6nw`@;V~Wj
z?EmORE7=8~@AO_y<f21KUW;bd!s20z&k2x?_N`61XAg{cySmEImy9aV1bT`5cLvsR
zFj=!EhJ^_1aqs6s1AZ9+7z(*+mt#lTS6h^HjgwCWGoa?P5ud>ZAqNz+PWhDYQa8}y
zy}k~nM`l7yKU=`w!C?$la8#9xqRcFSX@LIr(NSGNRj2zptwR8KS@Q_f{kq(M?^X5n
zt2#S7N5{waW9?1N&8ssg%6+4j8Hk;>B8M<AnLSXc0Qk$NOG6TYL&%;Rc4o+@aU(wX
zkDWm|5YAt`dv^`n`Oi!L!Gnk*v{^>^>5kNoKj5XE!hnKpp=+<!$i;>)UkrT}cJH=n
zaBy%ie8rdzh61YfuQgJAAOgrUwTjic>w&mfPyO4sY)mX{Re&8XV8oI~nF*3zK&O!B
zKj(>}{{W&E_|@3hU~qNTOH!wbwiAf}aZ23)KeQu}Jj4;=0I{9HnV;`#>lQ|?y{xTd
zJ9kr9J)HNs&)#ufS?e20t6-|45E#_C(<_VGNSXrhrCd0<3?e{DRH_Wr9EZ3#SsSX0
zi_4v#3Cg}jyAI<*3R;^!e*9r*(@~sP%8nTq7MyD_9{N&sO>!1Pw2W8oKxTJ7ORfV*
zv!9sm3HTPoH+f@Lh<@A5xZRcr2sr0dTT|f2K#z-CPjV3;?aTgj`Wob0Tx}VbgPma>
z;YoHVB$oxm-veER1{M|6z%>l7m=AnRY0JlHUq%#ziW_Ws90t<C9BIi<gVgLPmfv>}
zC4ozeYo2Ps)qj#(?W);924ae+MNc)aZFCuXyN!5=!}&$$DfnQLH;3(MAE5~bd-@4b
z45=;&x8Lm>8p?d<26GC7NZ48lpUravy}hEuq*8fEn!;J9D>Rx~4F_>Vsn`?BCbIVi
zs+Hj}LRE|a39k|r1GDPNs=%)5IxW;VQuSLe?^yNzJ;%9oOAA`iMijG8ms|u^+pim~
zJpm-bk7NE4Qh7nEvEP{5+XNM1VgJd-onIP&aCdZ*J_X&gQ!>OXY*z7UYNV&#w!ZfK
z_N{~}07_xN!rz96YGM)+hhJY8+y<4*cenrsgjo2)>kz}3w!mr4N500;N+@aUgyOLU
zpmn_LOMCke!|5Lbq_>O~B)93Mx0jbvquLlaxaQI31kvUWcNZ5*oiJQUgS(3&QrA*=
zwv>(B2ubVbqO2_QpDRkcFZCm6pVj)Zkx`X>7z+pq{dUH6wg)1?VABrFn2<#-Ew?lV
zXyfwb%StvgTFaeQ?sR7)YWCJ|@%{7bM8YBlW&W{-JZ*?{5bbqMZ{Zw+iI<CLB{n^|
zclQL5d+r_oNvbgV!2|xK_nSrj89&57^$Y=#d*9tHB0`|gWA&=<Nq$chyLVSHTn-x3
zDcM4*x}aiM>3d9v#Q*s`bT_PJCdf&Prnev=n&R`L@64m-TnA&6xB7@FHcL}mTd(S*
z6gg`GPtItGqr%VQj=4{+*-M<@77c(H>Du>R@Z?5LD1_23-)JJqsnw!cEB^cNvyX@!
zyK^0N?kQ(lN*4K9dYaH)Z{D;?4BF#axp7l~zhb{x8BBw9lVk+Br@ILHcEkKlt>#i-
z6t%9d76rRa%MUk{VR9CrXBicSGI{p8+UKzHi1!2PFdq#{T}|O>HNlvWp8JoI-l5T+
z>X5Lo<1~vc^|wWp=?vh;5jPrUA1oCu^<~QeTPCYO*z!<jDOHD02K{e4o-h+IBAsq1
zh7osa;AE`ysXn!bzTiL*k<*Bcnoy<JwUHb1Y??L`3M{4q7=^+wwdeAL#nR$JNrglf
zG;T%_cKVw7+Z)ets~vn&RMY~V;r<_<b7w&Y{8I@zs$x>v$rH)L6G+@Z{QnqaoCVOj
z7%tS$k$<^_(-dvPIMgN#DkwFz93Y#hWDQO^uOOzhx~3oiiIC8nYeJJWwY0RdC8MD{
z<OJtP8N4{{dt5jA)IX8wFKQdlT(`&Ir>*Hmg&B)IyFEBmCE_=eSF88wG8nerzh)(P
zk?l;zyVXChD$|#kZg8lgpQNI0Zufk!O+-YMG}`Co<{rJ&8bR25`TW}#T{gl;N-F2{
znH^g;Z6a9#q2UXXDrKzQAFI1J;Oinm<sQ=3){f*jLk?;{=nloinNQr=Wpo`SLGsem
z(rlSj;i0^nxvH;n@$Vsrsk;rUf3Dg|I5Etu;uE|b0{N)|P9GS;xxbYAJ_Wm)c=XC+
zD1o)88<wAQ&yuvo)I5c_#lXg3TGK=i5!<;cmL8H;Yt2BQGpudahJK`<kRLWX|6qCU
z_nu@%`URW<;gQbrc7Nmmz5&z`px{@Bq{^3J<N{)s9VhvxqT;zdC*y*Gn3a^3C(14$
z`C~}Svt>_W>qfZ6|DEB`kBjW=?5{BA<&0>re$&J-%rGiQKf`%|m)ud^NKbA$tO-9e
zc#`)Au_lRWF%ovMqv;teiSjZP%crD-ib8v*C|q1O`eHNTC*II7gu}y8<w!&^6fkyI
z9XLdb-bsQuasFWjFKBG*Nk~4XjXfsi1%@h4Ro40$J!U1ptQ`vHPDDy)-(5pL#?k<j
zea1khBK;|9OPI^|@m<DwQBhH6O;s$-Zn_hBM3S&KEkv`bI^mzOi|3H{-wG0&HLtxE
zVfbQywV>)>tVMKRY!1L+W{tmM{XyE~@a?>1i0%EWM9z2p(ZX;TV8Zs$q)kkpt5&b&
ze@<{D>^aY&pMCmG6C;~Zg@B>Xf37p1x*j`dHkk1^Mzdg-aon2x!d#Gm6tU~b&i3~9
z%p+g6ZTPoTj@c;;4_nfYZ_$4mUue^$UOts2eaN7{m_aiKWKq21o(_l-KEqad(KK2E
z2Wn{uYo$Je!Ous{2u_sC0K%ESYGrKhhm{P@{lJx^iZb<<Wnvlm1`=3DyGhVHzPf^Q
z+#j-n*lsb)`nZW?_U{hy9l7^m+>@x+B+o;1n4OK9?k4cL8c*-RpndhplPAWy(FPVB
z`I<pZdPF`#<@8rq?C%D+aD3Ese_cB89x&?o{0C90c!w<`=3u7PQGMNHt09PR!P6>Y
zUY5QLS60{bbsIR!{jRS07sF3B$i~971W45Hz4b!)KF9l!jqK<5RVw?65#$&O`hL~x
z=Wg$Id-V5{FsGB@!~Jj86BsYK_ni+(B?yU_p^AyHVjUY>N4V7Xr|e|qW7@y!$V#es
zKR3cZ^uah$CS+XZ!;l6e@a_wN(MkI^!>jA*8Sz^>0B!K-*w_V3Oom!&F6Wxq5zeeK
zU;Sb$#-2O64foTSnWbb<zr~mm*VAJ;ne=y83%7mwLN3KOIM1ha;iv^fUmg0QQdjqR
zZ+G{N<ZdYE-Ef;7**p8NcG&myX#u!}zz<)7gLwkKZ{c?zs1sY_ZhF(!(AA>UR6<}`
zF2AZsg8<dX*z3eTnp%t?{W{^pV14OIpgn+lG^WDDQFM+9!<s|QRMVqfzH+6S975B%
ze-VAz{K5isovcLzLVzK6H6POnH)ofB<;aIIhS1jIPRR^WAXEXTKRi3{M2-l7uCxjY
z3B7}IdKN2y=`Ch}r+62JhhmD9FN~&&LZ0nM{>dcKJ5z{oeeT}llss41AW)^Z`Tw0H
z=cqAmOUSVBn&7=Bvrk`*lXqg|dwe#N2w(1eh38+$CuJEj@MehBZKAk{pD8WSj#n}S
zi6a6-nV}%H?jyRRtKccVA$uL=7FDBFEFV>|aPR!jOK6eEN<T~t!e@LHz84vI!9la;
z%Fv=WLJgh)%SighR<g67;8t6?h9Yd?Cm$XnW<pLu>uDAq<9@|i6^8444w5pOLr}4*
zt}ZlYLy`bE!b`aHU^c#_kQE%ld+l??w48_y#G++Th?q)VZ&(OxvMzDX5SG|?2`VOh
z@Lr;l&6=Ug_rn1Muhv##EiuEU%8#LI93?%01;W$-Ax>I<v|!{gpit}darXC-lFiiM
z3F7mrZU-VA^Uum#cOn>BBOegK%}8GmIoOP8lX{=B3^A8@@1GJwI5EzspyZA-;ZUdL
zT4#&Zbz=15&g|UWwi6lCJiK^jz<<vL&^xOZ(x2`#+RXrGb}vLqR{8}Z;2cJpaf(7q
z*YT*0>2trbMAFm3I3g~}XEXW=R&a(;^$)Fy{7GET6Q@P%lfaDi($X4At5d~n@N*3J
zrjz={pRPcCZnZo;ZcCoe$7LzX3;}d5-F^D2s{TOsJ_<9NJaAh4ZrC`<fkxmiXGUkf
zDrWN1FQOCH`{2)@ibjKN+df*u?@Rq*Qfw+wAtB?JyJj&pKFqA9%LwplA28A;_p0OG
zwJF-<JiOQZzSF}D^diC&BhAsLkn58!W+j9e!1NzVUf7;XEjgTq>||IcB|YLGJhB9-
zNqvJL#bGOBCO)0R7%_SC{G5>f^s-Oii4jTadykQF0P71$VnSuX|6)@77`NJ3nn4~F
z?lS{6syG)IRuSqhLOh@UU%aaY#f2#7<rv?`+ZlN(guY!_^N*{E)MX{meU6vWA3ofK
zIOz3W#Nn}N$bm|V)?gpa%OXD)j8T?x`Q0F~ZnBle3)fO3z-K+!TL@O@UxIr8f+88N
z9aMq3sWR-%esB$WP#Fk%PL34#=5SD{JoQGKw`@`Iyo&*mBsgKoVbmiPjJCj^q^8Kf
z%f8m^-$Xa1Hv#ag2RZ(8`JO-3WAH}sDc;u+Y+4Rwh=qWkW$iP?M=2=}@T%K;{e5Da
zKu&k#3=>r+V;G0gA7XT=8nhx-;@=4J{BaPNL={*VS7j%1=rIwEBSD9)kWharz_T9b
zfTZ@ML6vV$)g3zCVH<t(=0$LC)Xke89ktI%5+ex81)bDwx$Wqv0!BE6{;N2z#QX-_
z6-U0cJ%9e-bF=i8rvhtRN9C;E3&5zxhe9B(*RzTMyR_=ZE*2J+B&2lfPid4&^Mn8D
zRghY9#)%1m3Je<;jPd|XY`+vV9rj{)tWun<THl2=4OB4^pJmP)EM*Se0t3-Bm=q#!
z-h9C+Llr-^-GGrQp7AL*r>d&zbc5aJ@szz%***0g9ZEG*;*N6sd;JuUg06rymiFt7
z0&pgu39;1B>3>Dd<#8af^gh4pqpwub3avt7wQpFtU3lIrgQVBi+7!3L;ahFZwo9u=
zd0pc#*EPQ;&yT-cHhG0vjhV9V;mV7XrMYv9iw2&O+YS|8s7UDhqH*v3?wu3whlk62
z-`^w0$!jfSDsD)>R+u#l&wCW@m-6soB7VJ>lhI09jgbVi@e@anid6ie?%I=e=v`|V
z|3-g*fAE7(-`G1H*Q_8!cI>zzWg4E>H?b$?KozgAaY=Eps(8%Bkin_Z07u8{nXEU}
zZ`QT?HQ)1Fd6C$-QD`wFO6oPQd7JaQb>_pf<y8k$$T2m*X~&M@3veQgl>_FAY}>X9
zBGieyyHZk9!vgmxDq?%p=L36<?xS{*f+`~M5{-HxJ$);*5RSmWCH|+|@Fu_)a#$Xc
zSHWg?&<n3H^qU7IzkhJ>8-(jB@VYFZ3&Jrv11U16X#VY6Z2%3}c9!$XSZH2~dH}!W
z=~DNVe0+THVz50twiQ5`1*U-8ad9L81?K&RR*Km6?aK&WcnsucSc4cLi@w0A!gn(;
zF)^XAc0=;Gl$>mS?cYNTFv7>p%1lmNojG&n+MgILQF3hd<Huw6?M3Y1&Pjb1hNz3M
z(`(;X<N<GxQ8bUEi|bOQMkaz6xS_Rv(E~aIE7tz;ZwdvfSQrduBrE;-GcRm>qf=9l
z^N+}Z19^@ipBrFVU||u9(JHWRWJJ;3j|w|<TBp}Y+haTie+dH|37gTqz%8Q=NJx4z
z(0G$pV&dq@lff9JIG~}i92%<{Jhw2Dk&bT|0(g<6Hv>sM%Bu(uXQOFoe8*YDf?vtY
zlZLgU=lgdiJG&vpSF+d&&~WPTndD*6lavn7+pEL3jATa+<o6E@j4?+@BM^ou1uUu<
zGkKsfhQXdS1nbI`jkhZ>S4*Xq;sD230)I)t^R*crQh|952cs$>ws~{#$t;6?P%_t~
zo(LEi9Q=qLt<bR^)}l0=T>gpme0+@n(F534{)y?q<}l&|xZT?3=E6FkfWHH&;=NAo
zMb|nyI%I$Dl$L$~O6(X`vD~~BBq^gRh<_1k68=`aoiv$KskkN>OuM^PA&+veUcDbj
z+{oPG+&Oh*q4fNU@o`t95?9eLZEd!1o$Txmk=9~3YHi-;NCsjE%);LObg#m_J?C}I
z14>ToASy)>s@J>)1q4u-DiX)X$9-!0QAR}U?Ccc1$1P$!;Do<ZR#sN<IuQtW%BTT0
z=Ym)!dN8N34u+^a#H}!rvWyN1)&`aT6Cl;c#>Vs_dlM5A4AjD`nsV%*y1M!@`dSGi
zT5xbM|HR<XP!lFw<yz!{V;=X`S;&y<>|-pX;7kw41Yg5202$_~K7alIV|Qk>rJ31&
zq}t@vUhwZHa(op`9`TunOTT`jFA6b1gpf6@BCZWUf_sDmWK`xQZE0blchF3{OT_6*
zeh?~~HAF58E2~tO?^f2<n#dQaINL6FF<+#fbkxX53uvWDb1yuDZ>p;g-22(vs|KGA
z&#gOhc54|Dv2AMzS6A00zyK5N5DOVTa6McUJSl;K!L=jX>tDRMI6pt11`+&Q0qiA5
z(QcQUrB;*#j1;zJ9Ioi}8Iqt<@9sG->Fwk55vlbcyldt{SoOgV4=$-wKf)b$%>$|K
zDArMKbv788DA=JDNcUfKLV}cJ6CO8njdyl-wh^w;+V=K1;s#96UiR<~9zc6zW{4Sk
zZZ59G$bJ}G!b3wt2YEII1_m;!!Z+|(VFdiuiJNW^YsM!hH^KB_i^t6zj#nx%9xB1$
z+#3rS+pnc^<U2_i7)eu9R6KdGyAUn4JVPQX-!Ukqx69aslx~yb>TRVS+eSJ{q5vc}
z$EyIfeH$M%C(Vb9su0anP_6MWfX6EcIj7Eh`T6-TXDb(`Pkn4|PJzt@g_^e<BLojo
z+8zHOD`wb#N3C8-Ff%h>9b2YK#$vjj7NJRvypfWUq8?5Tc<EH|l3N{bIX*=`=0LvE
z7A^Xw6^#Dy^L)beg%R|O*D`Pqi4bJyQK=Y#Tf;7>7#$J<qtr@*T=S=00f%<)-u=xr
zr{Zsh*G4ZmnO*+EQL>WQy=PAo2DG%aQjYma$;jM<;<$CQrfrx1pZmq^c(7115~FnF
z@h;iGiFfie8i*lb3V})JWFQEe;d)pP)8XA+M#p0Z^m0B21qUCqu-N3#Q5?y}##Y<b
zb{8f-GCu(RsKV|czm=O6k)Z6m%OA&{d)>M%uM?5VU`^QN(vB^CwM<iA|JD!JlHzQA
zWA7m1D2$N(u)ckB#}NW5=5qSg<@KF?tQ`40Oxg&9uY$6iTnwzdD+to)Orwz}fl(Dx
zP9M@J$Hlna@iYghds~Rf>2{~qZR&#ycwPZ0B|OAHKP%nc!y^SsQ4lUtsQ&O$_0_Y6
zhJ60GQM3F#Za?1|)zAD6TslPbV^kqAV(;G&4gm#K^u0!X{s7vAj$iK*A5hS)hFF7D
zTt+v>1h`R!Ux|VfX5VkKDWCp6GwO-z)<lZ`6BZck<^LTBIsaQ8m5LLxI_f(_kVeRs
zqK<B83a8-QCvKFNm(P4D1;yv^wKN8StARl+<ptOk+~FnpKcQeTJ_QjXRNi~=ZUQuC
zE@FIaYy*msjJ$lS_2MMv8z_VizUkFw6Kvr|Y7&C=BStk|K1`PJC?TxjB{fST69KZ4
z`}f!EkJ}B79iw*6FM&7bm*XZaBAP9qCh%B<i>$1yVi>1~pRV@p+i*-5J}#WUc#+Z7
z)3ZfWeei&PwIedkh!Q*`4>7t+j)ct4%zQu>SMXlm=Hn)5bxcL@zf(F@V7(8mo6)Uj
zzJDb7czJK=iKn1sQmKDiD;L!WYe;A(em1;&mx3}C`tI{l-Dd_RKeO^t@glRcC2>Sc
zG243I{QJ<*A;KC%ll(R81a8peNa5y3Am<9-`(jGeWv)POA?sI780w{+-!32#W6Mrh
z0&OFn?HUwj7IIuYkBL<Z!OFq0HcFMSfT|`mR|gjMl|OREYV(gtYOBVlrc#imZ(uUz
zH7E6+9}XVRFXv0r)2An<rg)YrurrvVARpaAQ~{=d>bd|);X9OpmjE;?RGYsrB@RkZ
zh`Rdv2bq~+D1RH2l*-x|JVr)FUR6S>V<491Cphtaqc8(iH8xUkE{$%De9qfQItXzP
z*!X2v0M-z6nAoog4H75*`~(5bLXeX{t|bC!l6NDaXGm1S{|3D98sj4$uxdW9K79DV
zMbD@X8WiJZjgE~q1Azy9!?sX)lcIux^~*HWE`@-3g-xqhR?LqdB0iLIK6>;Bo?=y;
zSW=DyA&F5IR>B#I(c=ls0B{n{FdFpt_jkkOqweip(lYK~Z;#Tw6UOBA$ko-*q|c;&
zPV;@GC!CD@23VJrmBj{^)^j8`Q)}xL8#ivmC`UD(W&(6@z}eZk{Q7gW8+p(vfu}^I
zq&R(id_FWa0iJCnpVxx`{0=*RVeIjesEE7QQ;m=S@F2Z6Z-}GjsITwjO26K!?=>j2
z<ssc?B#;5%D=jVEv113R29c^83fqsqKD%q_rEXkPsx98$-eTh78TrOgiTa0zk_!sB
zF&qfZ?ggHA6Wp+OE?kNpE=``SVVnG?Pp{yCFG$<Y$dH(2pkh>;*nwv#@UCBPPVyqK
z2f{(T>W+3W+(N|C`YqZIj+z3o5p>wO5?_0yr`N1}2G6Up?}(g$f-`|Zi);_yHLF(>
zs@fI!TvghM)jJA7jhI0OAZ19HRF3wIC(y<VLB7FF*A3rDhJH90qA|orH(&^fa>~so
z5p4W@5!bK3#>u}9FQw{(qe?Bj%J=+AN=r$<Gp|a35;RRrnkzo8XI{3992>;gqu%48
znc*i5qnPTk^Z$*d4+#z3ucO1UcX7rX?);X|Q5L;FG9}Nv%k&j^TVKyaj9s~c@Ipj3
zXWp7I^ff?c!oriI<)CXyUf%{*S#gn|X536m$AR7;ElmM!?bm1??lS(0*Sz5-N*n1A
zK`GlujiiTkcX!hf87nT*Nn<*0f*cOnfkrQ{rS9w3axFu-Yu+bz=N(<zA2>Jd0Jla`
zaWOA2a)!sLW;}5ra7tf{`=}Uqvu98M{yxt$OR`t~;UPy4{VCeAp#8e$Y_{vy{C{M9
zcRbha9{0~IiHIU2qe93kD>HlVkTOCNDkBZMDIv-VDKjfOWm7guGAqeUMx-)Rp7*!=
zJpVk;>vhh3?sM+bkKgxuU7zdod9O?Fco_ChWPCi0w6yd>Q>93cxZr^Uc^)-*M{qPJ
zV-tD{T4b9^$*h|2n-gFRR{ZGL*uBWepxIeB=AAo7(c;K>PcuW&bpr=aM&F+Uf`WvE
zo<xFgr5>t(0C?LlcBE%!?t#vk2rwWVv<(evj;Bub%C6?<c37C1F_Y>b@)}#|i{1Me
zo>iFT7%0gQ9!3WVbfZMiM}GhuL;!$<L~+xDE!Rk<rl#XxzL+PT=$+U^-o!dK9*S~v
z4GWlXykcT+LTMo*VW00fKL^nr6Cu_ma(e2nQZ(AG_O@C~+(`W)qrI)9v~<UYd4%u8
zw{PKSLQY~qUPg7Csp^Pvb#~FI0VY(0=6x1^dxKH|x|Eutqn`&|bPIdBU36=5Lc7qY
zGwj;6e<-=1U!%0w=?awP|FJ?$t_QhT&bG!K4w-wKLILShvUfZL>-{q3i602G?%_iY
zXn5?fx7nJ(V$-2(-%L+`OOH<#TriHIYPNP$ep{35X9kkCIR0O(?8@Po`LNy0yL$-w
ziBYJtdoOo_V*r$iCk5f~(S!w;&}{5Mv>1|2!>%pmby!phH;7qwsQTACT~hMG>t6hF
zlJ`p>M?ra^Km$xl_IcCL+N!-LCF?BrP}_X2t3ywxlRqATeX;T9@!Tiu*50c`Kuc1>
ztfbn+kp!WBVX*O3fWv&<#MhAHFjHzhF)`t-L)kMSYW3`EQM1OU^?XH&W3<<Tp#N}@
z=Q^4MG3^ls%JK2>+xoVp#?srK9!N3i_cMUy)U&**$*LaD0uPu4aK&Xr;>_&u@_x4q
z;&{BTKG($3Mzf^kPql%yO`W>l9bRbOiLu+at+gNhJy;H6Dg2u5)pd9TK$rX)H=z_V
zFfhO%x#Rr!v{kNYxw&4o8!|Nj?FB=~62>l_|AAgST4x_X%P=(lsi2?SZ1%<IG(Nfg
zfk1Q-wnH$5pt;Sy9TZ2EgwL-IFWLxNuY#3$h=G55;piW%4dg<-J_7MPp_u@pMct_N
zXBpJpNyI?mUGLeuH~qsChsMUnff|3gySo+cV7^*7Cd$MV#nvzD^qTcSLV`Wc0+>FG
zOH23Q++67l{Ciz#W4**cQUKkOy@XM@hA1`-q%sZYIUDZ^?v4dt3TagT{=aA=sTfDl
z*fw3GqHTB*6qD^5>f(AWJAYLW(=+-ABFAR^^TwaYeOG`h-X0ybhiyQAP|5o>w5F~n
zI#E(iLPU<Ik3Mk|Ja(orRzizXN)P>D1ThcEUR$l1cB$!Uwe!Dkhxs;4d`T%HBln6V
zN<~{sOENT0cdk<_xgoZL7-!KtM>RdrCVqRFWY4o>VkxQZGx#m;bQhc`=e^rVOr2=8
zlkp1Bk?j7nT)UBhq!>>y+;N|P$`O+yDKb1<KkxU&zw1}-YwaDadqq@$#)bxrMO!d=
zMG)o<&kGlr+bu4=wBaIlck}jx($XyLCYPB0zLA?}AWfs5=Q#vt$K}Zz^k5oloWhjz
z>|{MP+D6nz87NKVI*|P3Pz_P#Ar#X*3MtnT{Bw`@_<y5fCE~%~gfLINcKrt}C_;S>
zD8zu>^jkUU>4o)qd;)T&#X#vHHVk=W2n6rW@M=6m#D)(9vkgZcK<+xjya5SLa9CJa
z!(yr1`}_D%4VW%zFc94(4iA_2#rrsw0;&u?B3Qz3k46|b2}6nZw+}H0|JejkV&@<D
zoncInUs%^gw;P9))Wc*9&DFWuNEM*tkTH^%-nX5`8y+C1&PV##CQm?Sb+Op`9U&dM
zWoop|5-0q8Xtzeh?!2yvKlMG2AK!ZTaj;jO7={KM3~JzFGTd_W8sS|0Q0|r4cLjdB
zuh2~bWZ@-L$A|8-W0tyl^{QlkT6(%#m2AKY-vV?pY&5aY<IsLWOZX07#u+-~qtrI}
zt=Q)+KqH%eEdKhH1(5Xu`X`@eQFL!QS1@}sef|uA4Ouc=yd_PZL<-mhPLk|}A^gbJ
zhKBSee=!M(R8%z+ZEYI7oV2#OcF9#J@mt=$Eo}OM#d-!_(4$S}-Cr>Bje5o~G4rN&
zXBwKK2)UlHKuZqnK-XK58vWoe$eZ86FwsC3-dj*yEU{(lR_5lA-2FTzqC^@P6p4z6
z7dZ1t5yp~=%L}P}9v_@uKQNyBb~iLM4Y-0SRPkCsAB1ZGNk>OTsge#+`+$4P_Y44w
z@tE`$o2E2B4d)_W8D@ER9Nn%xW%YDT1GGfq41qEs!{9m~u)LojefJzZ$OK030R-4G
z3V@x^zYuwLt|v4nCiQm*ETtoWsO;i5wx#*cgbWO(WLg4th1WeioDQc(5Ny05`d<mh
z49?pyPPyj>zx(09C8Tn~N+Kfk=y&Z`MXP^phcD6C)vIV4y4;h{2@pIKfEOZ#jZi0z
z{)U)P{o+L_APRYG?OlC`OwaxpZ@s)a5r<GDC^l~LNJ{QNoDK1#5%xjE)Zs|K-FW~V
z#TGmO$%`Ad?~k9azKOGA3=aXel6urQ$+r2vJa&w*>0Nv5^obG+#+eLl<>=3=zy`eQ
zfk-hRX91>`J@WD#r%s&`lawS9CjkUe8ya@{a8iIsf-0X0bI=Ww>;?ef?l|l5{WyVI
z5b1aDmb(ZOVR=LgPE=9EEWB!I*^Cv?j88!@*DY*;SHKUk9_w-b!(^9Sw2t3p<ja>^
zk&%&y&tHOws1RLZ(JlvIn1mY(l^C#>Jzf=CCpmgM7Nx%{d$3XY42~B&(rl%pBjk#C
zGx(OOaD$;aZfB7rYG4ud<)@OihVAhAwG0eA0PL^%R1jDgD#so`GXSl=1Rat90+DVc
zAj5FtXCtexO4%1QC%e-C0dWB@IE*MOSyn3o$A#HY#mR}UIg8PyMaCVQbNa5YoMTu~
zb6C`tyNv%(Di-3e___aD#D9NF3|9YOjsKdSe}7A0-v9lt1~oF2|NGmf*8l$x|2sSX
zedNE6>Hqn||NcZ#(|%OI>u4VS`=PViZEbBTIHtsruR>4lJBzs%a<Go8cK`e5s~T!6
z7(P<W1ap%nar%`7ZYW~>ZYFbx_=NTk*5FNQ8XBHrtPcNvb>9H!CpLrMdj*wYPSPaS
z6N!Y~NYCm!OEi%fkx|2gufP27`wjdgq=S^Cz5M(Za)ufj)TFwBft8zgI)G=r1;`0v
z=10g6O3b;WAeEG1JF%t&;r+ou;!T!kkwD^-q?zPb?JQ{vyMP9QAONP2@R}5lJz~Jn
zPs%Eq68hiQeA*hN<#Jx&9$Mky#@y8eSudcEykg+X6^$Qp=i>qYdj+#Mjp7mt*4O<=
zqBb4<nq2-3{4J(+lD6Ot$dvH59iA?Uh92(m+ri9S71_$-d$>o^miXHL-q6yCT-}bi
z!>jN5X2s;>*fHO0X=`6SArk2JaS;I|p_1oFXeL7-!q_7!8nNwzmk)j4#0kg$W~`?n
z52Vhr5ydL=N(4Z5$j%6%*XT0<Nogon5Xs&sFPFtA*p+F>MMT?KMP0+0k5I$iFl|}&
ziv!;w4ti8cwr;e4MTF-LrZa?PvWnYNq8nNtaG?gkk4kbCFVF$@qT6@x-inK3!GV1j
zP25Rf4VU3ey_b|kM<SSm@3jd9C@w_Y983ZT?zr{hg*2gMue##3jx(q?e;M0wGytdI
zcARr{AmMW4XG+e0RwSZ!aDJ=n>1_q$-t|PE9a($SAfGHtoWlElh$6w2Xy@>Uf{-7E
zlPnx8hqS;sbYmNTmuq1)HO`wq?dWI+y&$WC&mov8-r|Q699R^e{wq`wwhJc+cQ77P
z3Xbu@l9KC5Nu1C;-iDBV7>YbPK`$V?tcreRD}y%rm>RnP0iq6`;Rpp40Oj*0L{}j$
zO9P^~@8~U{%hYz2#YFk_;|ozF;-oG9fqN4g>}^C0goE|!`hR;c!qAm;2+f_(2Mj?P
zBmo|tl)fvP;cV)puNcq&p}Y*FJW(LzTafiD;DED|G6=v(Hcr~HYDkc7w?9cw2S$AI
zBoBs|2(02~`Dv)$#2l-6w;r%rqr^5&1bV9}0aT4yBp}t>w;{gOW}r+!Vi2-jt3R-2
z*%I*4jF7N!I=Za;IvG}z^FT^YIXT_=85|tUL~{A~`~Y%$EaGhiW8T5h+qnAk=g%qF
z@pwYiQcKq%UNb@)U2T5>As-&AAW1ydQ17SEZ}L%4Tz6Rxe4BiP?yv}Y(SqlCA!;h<
zaI~x2Ik1w*k_qs*?+Vx?9g;I9oY7|h9W5&LCE~*Zur38BvN|)c6kM!3H*xtXlbj)1
zQ&VGRW>zB|x;s_!9b<0J^4Wht{bD2MIkBbzRh$7tqzi03JMekA9yW_X1u>H!tF9a*
zns`)UBJdnT^#G7(&?5XqZAV`U5GQAaHUMeJCmkJ$h#-R1Jzu>TYuSgfm5dPhxJQ>j
zvGYGIz=aExB&2YX!&MSoQ=>>oVuFH5;K8_oW*~Cw2;`=RFU^JF#e>9>lA*_0O)Lk<
zFgzyOb`XORN<R_K3@RitE^ZUN4+JPhGP1F`3n3aU=Gp<E;H2IXdqQLk?=$rO5l9fR
z14Lr~kZ=R#pYO>sCQ^%jEiMv=5RCD!U`RmJ@d}U#kGVlvVv*y&65K4tRWcHwEx=Jo
z^<gK_Da&KW{@DzX3=S?S7==#-1Q?;a$5Cu&XBUM3NX4jISX}%H;$5UA?j-OrK!NJc
z&dy1$7Pqz5k26q`a`<8i)d$!Vp3UU{YiZyvI)V=ap}IYU5_{A6t;ZGQR1n`3qyco@
zfL4a*Yd4yJnuj98R@kPB6%&?5ZNQ&t?>F<GYq_Ai&hxH(mh=;S;dS&G=o}~k^a5)b
z#%PMoPlqQbhHMZx+!@h&Y>$KZm5hWRe8#%7z5~!E&QbtQtyi2RZK0v33_=$hYBgC}
zSjdf@6_Y3cO(bd*J=d{5o(`=10y!bpdG}xl!4%}mTz=oLY8;5q%@z8D#<%S-sF^IN
zj{osXzXt3jfLD<#qZ4_0aW~o$M4ct$@OcRCwEEA|M}Y2`;)Wg`MV5OA-o;}LGvQ_u
zXksHQ&zNFpC9;9KQYxo8H3Pl}Sor6A{kC*7!Nw`<@ib)d-2PaJr%kR5_YI7{>4Fo;
z1qwowhb;h--TSUCwEAUA(;u4Pn`0so%m-3VGcOJ~IXW7FfOLJTpYsN5orZ*z#j3t5
zDEj&&0SIOLuAIlfg5(9tAhh-c3pRM(n@Jd`j?EOKZ1R(yq11=7O@I+54l>DOL|@{t
z4HFPT3E+%-cyVd=7_b~(G^kk7q@b!Qc~D;Uh~G_P3PN%(0XQO$nUtKFxd|r+MiUMN
z-zCi7{O-_bhkL(Ckfddoa#Kb%$NuIbhG<BO2v~?@x8mOGKLkS^RfyU(ps~O@^4gv!
z#>byT<^DQUEe8Q&C`@&?s~Z%yAW4m+eC*h<(w_ig0UXwU`osyG*vj>gqeQ$V`(*(6
z9)vrl;SY$YD=`ybX^2zo!c3q3pePt?tbHCd+-PI%JD5nwj~I)y4d@cEwXk4819QEk
z#4F|p(q9*_ksRuR$OueQab=i7E2A_}-6eOmxTuJX1bSRG+l&`eCGZH~tG>dMr>|U*
zB}jcJZMc<`xc-N`mXsXi(|Zuq($XS0X`%XN6-IBu9)OJmWSAin+EsA1#4Ju=W`ty@
z`~<y%f&yd;2%i{@v9KX>A9CHI){cqcZjQW<>G`V?5)#`xMT1-csS-`YfEVsYc+nQ_
zj3yC`>0NNv)L6k}q;DMsH5-3@5pcMTAXVCJDlY6NjM306>p1uzXQbiDHT+ct6}S+1
zJNR>0vXj9!841SVNY}u8K)4x*T$q`|z>CPT5^5Wpo9jP*{GGkewxD{6jpzZu1z5;E
z#wIN;DS?<<kWUP}<=6fKF2eePn$=Tgi(Uc>=<AS`4LtpUlG>YZC=$0vN{Wbt*q$iw
zk|si6emFF(qUtL4+7hHYQLw>ZV2!BzVbJ2jbcC7X2%@wI_c&gz4`(nQv>k{cwvd^6
zZ8MyMfO<)QEGHk4;Q{i*=n(KD)+q3k(N{NV3DBqrHV@))1mz?y16ssgDYunWIOHyF
zMK_>=8$$YuaRlGQpYa2G#?}qOjEqEtcgC4s8h+J5zw3X`8#pd4N*g>EEgc;<RZ{9O
z&fF$?Ap|oL?H7#CIeUqd^#O>ue-RcqLZHpObwY$NF@X(pN1P6DdxD%i+9!uX2Bj{M
zOiQ%LMKE!a>K=$YaBn7uBOJw;1R2@W6XnGQzlk7d%lo3<-&D`G_G4o~UO3Oz4p?0i
z+96OPJaO%K2ZU|{;5s6Hc6bko;_)psRXxd{0679AhfZ4hg^RCpk9mM<Q~|gRLJhbA
zgpLR&Y-n!Yg8>hx6Csp<1&HvF0kr@cMh|s3^CpJ;j_CcSurlfs`>wNd8@?E#P>HJ{
zK=XLu@WsL%pO_%VL<m}d3L6@31NcaQ;+Uz34iO|avHIcM-%q%tK#cIn%I?C*ht`yt
zDGO399;hiWuzp`&4u1TY3mjtOSwR7TppX#lm`AL=N=9bVzm!qC2&3Wxt5A#%2$b7O
zlwwV+{I)=!IK&Pex&e?Jd96guFj^&UtTe1@btDYoB?Tk#qfF-~p?M}FVF(PuSm<)V
z7Z0C6_(37$R7u0)f>pySDrz`uM@}NR2JwZ~H*bjR2Npop^qsNoNivbY`vAw^`CjAx
z;-0<}1)c{>UcFL@j)~#n=idaEWB$*8Y$hG=<?oVbEx>x>@er!)$seCSeVRG+`1v@k
zkz^FgYUxT~hDiwiBvDxv6?YRki~x5CqDmm9&Baw*NDj_0ya%iX7@A6@W}@3CzV_y_
zAPp%u&p*!qAq^1v-h_=u1^a{m&e6W#i;B9AxGg*MG)Ld8;&J1d>aL5w>gjo=Q;z+K
zhE@jlmQ?Xiz(H~D)Z+jrQt3Q=pg8~m6N*+()zHvTPY)k&mH8wY(;_h_N{SO80v@f$
zm)E-pT@q;+VPs2!%SZ{mQ#+pfIM|rq44l0rAGzGc*|&!y^!Vzb?IK(XL<Ec)B@~P>
zCzFgM>AZl@C4S9L^jFBcVOIU8bf0ahq<8M1+taM9%}8HIPkrufKz3qk?;(viC~m=-
z5ML9id<uO|u2`WPzg6;h#3Qx#iiwL~#qmENG>kwU#9`ki8YSS2(2t%DpD~dReO6ME
zyCl0GMS*Bx)Xe5zKD9-rm)rYdDhOJRKoWiVDaH{3!ADMts{CZHr0w+C9|!}wdi6-@
z<V~LwFWU5ZDuH!CmPeEiyd!ly*?WnJq0ndS!SY4VyNQHh9g#nxHYX<8RGL^_pI(kU
z)x_GJil&?t1TupY@G6M}`}igZZj@21A>!RnFTsug@8Q5C`*m>8gmtpdvse0l?<ZEn
zhcb+$@9$4P1}F|h$9(J$fyd)ufqJzk%R8I!R1x7r#HA1`^7_`<iJa#7`mnTfq0fM}
z+^S$)mv{BEMT8F*BGDo+Fyroq6r0~ikvTnc7|0-TWe~Jyb{OCy7UfZW8JQ;AIz}+Z
z=j8n{%N<+@UaQJ#g|%xd3I{?(M}Yb5-1H*g!C{)G4_W&rAJzi?CZwsPoXJvj6x(=T
zVbV4GCL4$xY$B$`)zjw6XTkH`1fhl|h#;Qu^9TzSfwEv!cffE!kZ0U+?Tr(%3;!k~
zU(ZL!EPkq~;f>l$_k%wlpPv`Kn3R{6ri#`Dxqm_uQc`?3Rgk<&VBgCFq?YdPC}4z|
zz{oOyeu44^Z{1o~7X+7QzIU}w#(hYev_#@eAsflrss;pJ53Ux<AQU45Ol>(PO@NA=
z@e~}PWdc0=Nahn-n!)EdvV9uIuzKy6r=XN&tYqJ$jNoGo{OH#qeF@y~4?k56!|P}<
z_*1C&5jRF3J%}s;czLsn2SN9q0SWo<Q)_;UFfxKj+)lEp4e$@235$r}M~g(RjOID}
zT~f74c6#97)nJebYM5@Hfm=M>cnTU!`~HoXjz<c|ISvo!7SyC-j0&BkVeG=h*Lac=
z!`u|sgtygaX!f-Dq8LdM2M?yfq8m<}@;Z2-!E$eVd-p+xhjHXY_1z+q?7d)VeQ**^
zWyIO=q8DoG?99aYj}wGtQ!2GGVl<fJ2yBbynlh<nXy_3Uk+KG~^ecW4z$!v&0(lb6
zHSPd<LE3HGGBGtK$awJL5}&v?j}&zAus^&wZ%FzI2rCs!ZW^r{)C8fAB<EZn5iR*c
zKLs*aI-~Itz)WNidDIX8en~w44t(}k3_3k%{sh~TAdG+?R~rbS{e7|^%s5jW(x20j
zT>&qNY3U5E-`K~<=nt2i)#uM}=3J%_*-TDFP5lm-&>6IB_T=dS*qOMRjF7r00J~rd
zQMS%Le@m#=2H6>cLPBsy<4KS!qaV#GdWb>f_z6WM>Nz|usF;iM2@}H!O0JB<HQ)R9
z#?Qwah=JKn#*O(EPMkw{meU_!6g*hTw#9UlWvYjQsY?=oPszatqaV7RGtp=4?35vz
zazrelK{r}-z>xhjG7tm5MXw;xbo>vP3&->44b|1vdvFN*#ObOLVT5S&8UD=JZV<P!
zg><8s8)CW!P;QMh*te{vFnQy&J_OgNgt&Ou03T)<WU%l8S;Wwc6K1CBz_fUKPwRv7
z@>Nq#>F|HjQ$SLS%mk|dO@uS--xt91{x_v#Z>Qr$CDb5$(j^X&Ey>V1yh9dQHiic9
zuIgCsPY@>(iI%vuU0qYt6s<dcV9tjtXmpQ2xw$<mq)ik$Gd5^Pp=t?1Am8oV#Csh5
zO_!$tD7NfDI=Za4kZ(NoV5-4fgR+{8{m9mC+`ok^8M4j1-x&ruE+0f8Kq96P96EUv
z{H|UYlwlk_9@D`3Po;;8Y@9I+A|bu{aLlEZ?<etXnOsMA?>1VioUiFlke71p3j=L3
zrpYE|nQdl*W&;D=9YR5fx4$>46`uF{`s-lMP3I*Lex;2qOGq1Q5%M8ponrJB?3gpU
z|3ld6E^&}RFtf4g5S$6teH40l*n(~W^qTtmwMRC8noEiF7Tde=Y6&~V|1?9b{Y)lM
z4ogkuJWWr(fs7Q>dCaIk1jNOQ`q*0`^2(d(ea;E@R3=)HF3?o_`S~?x{D@wcEE#%s
z1rvFUnO}k4pgyNB`k(D++x;ANP=sK_#zLgGq4hfh*y({ZJ}`yo6*P#Cg1f9gL99>h
z(}iLwh%Ta&lB_ewJ{Vz9p%2OE5-zL(p_K*5XyzB3or0w{rTc;Ozz$~$k}@ey)^mI_
z8MmrxJvFB-@ioF@>UiwJBu$)1o@haCVa<9S+D*pz5~f)6c?fZ0aFlp|@beRFtu&xz
zc+Y#dl3C2q-C#VUMVSGs@Uo$S5@O;NAV5Uth-ugX41y$m*dO64OvlqF_f=5H)N9`V
z*H6r@mMPP`^2cD42-t<bX-`XQ>r3!5?hqU8PL7pi!zbaI!<>sN<cMh?fD?4%;VzZG
z!JXh3Bn2r1T~NILAM)858c4bzA|Sv}dHD1qv;vg8qpZam{09!uVa9m{GAa^MBDD6b
zY;5+#xq}?ulc!Jje|`QQNf%fRSJWJvG@%Sfn=^am$H5?J^jvR|7;~4}VK!b+Q1BWw
zzqbWAl*Ke}h@0EWw@5fV6|bbsqW*a*tmyPvtjp;mb$evk!isJUCx|v#IymueR=E%9
z#u=EF?cCg6GoB1nU{a$50z$ABxY+N+Se97RbSN>#g&$L7^nV-`wFfrmt7$J;eVs7>
zJT<7I_ZkIpB=Bz>zxezbHz73oiV=)<ZEcN%v1!1R5b&aJP$FFt?_~9*Bqny(**T`J
zFL<0Ic`t@WoyD3SRRH_VWhw}i!3nFdg_EAl;VF#_h22+3r`__<R&l&eLVNe<E;heC
zjzQ9Ql5!xaoMT6mrK(4XYLlgDmkPs;(Cr}&>wr0!iHqcj?_vx&HB1$$c%=IntG5m_
zYu*%J5-5Am<M1^5Hh*dt1dmu>T8Du!;fXCZ5LzOl<O3+UdV6TBAyj-x#{J*Ndu~!T
zIdf(z5^xg@4ty0jd1M5^Z@8c_g-U%hw6>c`u@shV#wY*5>%NTE@XgmnqFk^KfeHgg
z76cKS<}L9hE$mT@fkAMqZv)EXNirop9c>Z&2dr1O%#jaJB{yUBaso@FfX@;pn1ns`
ztgexOr%z&(TtVl+vsp!uagoCP2L6HX!<BfvOVgOm$p?e-@P!EooETzI89W_X9lUEJ
z_wv~zt_u%>fXT>)_cBtE%IdoCe9jGi<Z<YXwQ1C?9nG+q`m9G<mQnNGzSWK@DE}I_
z-TKPH%Gb1r;faaHdnd!r#J~m(r6Hm8Ftf4}+<c8eU5$%;<$78@L2wj3X=kg;CD~9j
zkm_~^oy+LtI<u;0(x^+r%~>az^~kG&G)x`zthvm8JLwx~h0LAGj+&cH*|O1RLSLXy
zl}wmon2H4G9sD9W+~>4V>MAnr@_jimIq5*$0SC4lx17;rn9v+ym}BhXB`HxDxkvv)
z7y5pck*^f;P8A1)P2@yLPzC)a{Vf>1H9{G5-zPea!ftO+1nF7!7pvr_E6Zd-bflVG
z%hL{-6m}dztt1~317K;S>TnSR0!~h11inkJ^rLOy=iS@xIEG-V8-vycenV1BbaYn1
zr8qJ-szOF(Mor*%*ogX}ly#J($0X-f=KkELm#xVP>FbV=j`hCI-DF1=w2kzj=irnn
zNyerqFfP$(yF6K)8p-}V2M`&EmAO%jcn}PVlkXOoM3^e=TwK~tOE~E3VaCCn7a0?C
z6Vu@^umK`76G8<6+|2f)+F(o2SyI5&oE>EUNFPfky^5wIi?8qfIXJY_+qALLsMq}4
zwHzqq%SZ#;j&?U|8q*I;lMa$ON~er5Z%)3{SbLQ;_^e}ZjLyk(=L}H2uYqu(yeb-{
z--zTF<Z{Moikt?iFi`5svgz5q_V#vFFs}H8xZ^B79zIN0*LbIYS60G#p&57mOH^;S
zD3|E;oBMp7Z_}?I6Us03RFwEv@O(WhlzZJW)c1T+R9C-tfm)fXr{H<+^SuRSykx&K
zLUzl|+}SIf%qw0>X}9pU_K9DT+hJrNEM#3jH#s?pjQyKHbXLoxm)Kd42ymwgXSuAE
z2>tQhmh)jXlyu*9e-o|pCeouFt>!{mRw6BIh1zX*78@<9Mg{+H(s4hIU;{naCw7a}
zXYTj@9qqc{_31cA2gQ8@2cJ=q&?t1LAr>NLZP2Fqr2n~6vcfW6o}8sMzI$3)nwr7D
zJZY{Ojp-^B+`VhyJrP5(dDC~vpuc_{j<b!8wX@>pfmcZvDV*P=kOa>O-%rV*(i?X#
zJUvVb;sGgNSXz44od<9M5)7#@1Me{E4cP1M{5W@JNoa@YrEqqhBSGAoxp^^+5%VzI
z0EBc=kMFu3UvB9tI$*UkXR=oWMMN|pBN+FTg~Dg=_7E8^#uVONm89YMI~J_ZF0GA_
zj-Gue7o?KqvN%VkMP*bM98>f}vyI2@sf~}aBy|{L@Mz$&h>_&SO0wRAM@KleFGU|C
z!4}uJXGbwqE|_g75;}E?pXBWcDj;3xk9F71FexhGB#znwHQsx041t^0a<%7+Yd5yC
zx`P@DotlbBobJBu+L{?A@9*7Kyf%HT==+^GfAv=rAqk0hSj_g?Pm<QR94&Y;;n%Xu
zbzNFNtf*qw=A)WSYaDl6TFI(xJbtzfbbHogZpnqE>E-7$YLrS!N*HM=j3jkK#|x#-
z{3XSbSbw;`%E-;7Mb`#tnKECLLR2_}+yVjuSr5Y?DKyL9)vYz#Bk85KcMcQaQNhN)
zs=JkY1;;E%8_Hf=wg15^lKTtD2k7$;<cE4@J#w{dNwG83&H-GeYHweEy!tOtETV9e
zIId0y@9aBN8Z><+zi8T*!@Jnt#bpb2+38sK-q<kIpU_rT#v^=DFYD?^1d!U_AB<OJ
zCa8{{0)r?q7zWaA*<ys_{0Ku&tL65fqokTm{kQ(B4zHO>N=JJCh~<t|t2OG+3XybV
zIc90v(P9>s!5Mg+mKAnVm{BRS5_UfR5-Fu!K(WG@Lw}Ey<q{6^9+@9Mv^usVUV{Pz
zKsF)5$jHcP&o{{qqY|Ub*?s;W-q-iFgQ_q=Nr(<tOrs2_GX`H*I5LGgU81k0<vb9g
zLy2*)4HM%YUQPPbQEA(T<w)H^{MI<4;q<y$v=G`~gnZ!rM8%1XeR3|pEOYet?C!@R
zVG76Bbrl_Yk8d76ZcxRhM;hLA)X--AA&2k%Hdj{17v`|OpdVtqEfA*qe*92yPNs5r
zLwjp04dy3N$5AR0@e|zLgMg8txN+ap(omHg8sJbo_3y-Jauzo0q$|`ktRpAmpi&ZT
ze)LeH{KU}k8!B)EnSX!E2N!GaqU^P6`$l#<=?tz9o4$98<Ffhf%@|*Z3t{iv*=^`R
zmTh(iqjf!cZHah2Y~PyeIb#CWivIuoh}3;SfnKb;XS|=V2Yp^msDBjcxp+{7&N#c=
zUO2e8m351q63I*X=tsC~$dR1Ix9VnLW3$`d{v0nPDH8FsMf1>iGXky{o}Jwqq9#F}
zBi{6S^^iTGKX@eRpH(p}IbtNg<f<P?l6%=9^M1_2%G|tf_$D}%-ay4+BPp6cWV1)q
zzlFmY2LKvgtK_t_?(vIHl9LgG%SWOl{9p)`G<35DeoRo`C^+m#80<-93sVgQT&SQc
zMD}b<OulHQLx99C%RDQt3*>(X-qmi&)wXdXDWB@(oP~F${5Aj|@UWfl%b6M(AsGEK
z_D*U>aPyvR%r-1&T8N_!IA@+3ZwU3<5JMrN<Cs0hN@FJ#JQ8B-#doCl9*skUE;$Jk
zYUUAIi<|@EqefCL)+7!RH$EMsBCe-Ae_t*tK0X8l3_^%r0^2c}0<mBY2o+*Oz?#w8
zd+#4KTGVO(_xLlL=<4|B>rGQk?~Qu~^UZpc{}yOPKwsRNAs79rv<b?inv5FL4ef}$
zzI#{yyd#NM#onF>MJFh5a3Bhav_*I65ikLkFlEm@Edy0aWl4uwnz5l~R#OG(LfZB3
z3*82Ff2p{+>OAQ;`Kr-G-1mcYH8jvufc;x_n+Qn?pAJDgMpXTCnmW3=jr2B9!lo2#
zxM_|+ZVIs~l*JVgh%@SByxXBH^4K#>L6+!IzwZd5=eTv+Q?-=2PMDs`SM3FfTSDT8
zag=EOs>{O~>bJ*|gb4<utn5&Ex)54;;1W0D<9FduS$xe@(1sL0!g^?WgdzWGWy<N<
zkWriFflv_!WtqU8kFwGAk*x=-zGO>ip}iJ42F3-%q1pV+kj-$_L68u(Y52{Eo0*1k
z?;!O2*$<Inb-|lGFhE(Y1!wWelVkwsQ!+DYPoHLakSP5|V<+SUz)S^%gesNQ*01j+
zSb!z9hc~Br&MnO{6&8_}JOwN%IH}%>Vp$)s>e=XufyDSk)n#LvUNI(hjyD=FY3k@m
zPC+^4DUd0Ej>@%8hKwIl&dBx&JfcXg8-gzs)N~lK+j7?lYugW?1Co3eU{LWd{B-;k
zh-YNGuCa~zt%rjrR#tt|e(G`-dbCTGRY+Y2OqoVs26Ev)5<r^lK~pBT1MW-v*=XZf
z;6B=$m9H)*c+O{=my9LFL+ZEl@CwrD)DTzQ0ji@3vLJICahKYXEBJ3AS}_&iJi_3X
zQkcweYT}f0f8REnJo%BYcH7upg}_=|yN!Ar_UaWc^A47@!a{~Ct5kS!ngK_6e*FA-
zFDWXtYS0fd-c;^dhMVf@+`OzUj)Vh2P4gP`RMD;NX01skH2(eyy~!=q)L7O*nU(B5
z%hR_EM5$8zj|WD_#no3@vkZKcJG9n7YqWR`=1C&#z~HE0j2zj+;qJ(yW=u^3p*md3
zhl0Lv+|kI`TBrQ?)oHr7jLHK9+6?r<N7N6TBf&&xn+>6|2Fav%n$5TMlf~;%b3NWr
zxUZ?!0>RR`bM#O1gvB@hy#Lylu}5BE_xaqecbHp#{)mi<qAjo5+SH^0+2M_%dyHGR
zI+luT%vIeS(7U^S{r_qKmOrZ&y4rslfyjhgR<>L#+bqH2MBKlFV@d0a>LP=*>qDA>
zT+-k66uZTB#ZX{D-n#nwJ0S0xv92Qme1JNWSSj<b`p5nvsgXNVWw}%nd(ttT?N(k@
z6$ge6caMt=(-WcM$+q(vm3t2OA6CxE-R9sh#&?MUh6E!^%gYaMuPy9v9T-ci=dlyz
zK2GvY&iUhV_)_dO+&HpD!5XB@TI8-4?(Vpk%1^e%>v3IEQ@(O#^{1yb8x=QW-BCsB
zV+ySMW7>vLhM=W0t!6pb_ms%7rX<}ZJ@=G(9M3><=;>F0KL1}P|DDiKB3@%ip>lLE
zyZ*cL`?BCCAC^BWMMghNZ4+4h#l2a7Qu;g;*N%>kWlUY(Syp$%nszt;;Rp_@4(g+L
z&^5qxAVe()qWY@iFI1X`l~Ik9Z4G0br)DhHb2B#qsn^OdX=wcXQt9M~jNdQAmiD&D
z$F^^KB1~n2u8{h^?{A-OAyc*opqmbsjjo~#fjV-)b||uZRH_?z{HJG04HROOX4ULC
zBUJ=rQo$(z>QyP2vvyy9`-rdTp!kb(4*f%)J%%IF${0B*aH8Oj1j+b>Lc)y=%*HI+
z%!*84ygQ0;uH=d;kr<tCl-8JJGxwRtKKRpT7RGv*VaO(r&Fn7dzmz?J_Knl?9#cP#
zMhh_u2>Jfv`#rt;C1%mvk3NivO&EpOKfi5gZLay76gF7oun9BR@%fuj=iC4pN|EDK
zut(;h;U^0tx;pw0HF5|Ypd|^q?GE*LnWWdtMcvcfF;dlkUu~fc{oa08!LoRkJ-TfI
z%8YZ{wJo6}h1AYK5O2@VM}iod7QH!?A2$_?udU@c38ZP#)z8M8zU3+W<8WZZK!vx)
zww6ZePZb?X3KiAO#Sc=PoSbD9G7ie0hd{7iE05omZo>I=HGtxrTBCfP*nW~5NsK+{
zzTru8T2Yi+bO+j4bd>01eGcfU!Hm#^!HLwn;*}_yZWvAzwAbvHJDB7V*g1|cL`Fy7
zSYMY<Z_s>|dhc%3QKF*BEzdSTy`*@f{HE;Fyt8bJOR+IABT!mF0^wiRv27PI?)tE9
z9kz2AfPj$mvQL@AP-ThOmzWW3M8DLTsP>70$_?%ss|8M=Y@zqxnFui75&Ve;8+8UT
zk@Y>&O%9`FUCv!q@qH438)E_*5s7P-XmC7h$%oAw#lyx-<9qGtf-VuMQhrPCPcPcY
zpl}1GPJ=$ogm(+Mdy?v!*$t8<J@z68CnsQ{U=Vaf(rNeFdGRx^9!xr&W}4$W`MqVK
zrR@z`57lsoI|{Bq;NT69$kBo-5=Vr&kS1wkQ{8HOz%V6Qeel-0`<*)vl^q*NtJr>p
zb7N?Q>!BjkosSDN0-~b9ckj{*A9yVuFe`M~;g#$6<)8Vc$9BbUd)v0L&++u6VqQm_
z>QUOI1N<kSQh&L|J=oXXU9y#ymg&Qbg23<};8wmPtP>y&EAJMP()^F~DY7zCkGxxI
zyhnh2p)-WGcALW<)xAv`@|WbStpuB2%l>p9j&N{yzi>m3&o%mEwwWNV5TLoK(nVKy
zOwsz^OGscPm!aSuK6W+;pJSvf$01in>*Vw#e46%XwLe`<R3h#CbL+_1M|U%q1z4F$
z7s(&gWlHBZ<&rLf=z!EZj;&{L9X9mupbyCFq)1FyAU$Rz_TK(IfZoyJpfnTYvy8`O
zE}x6@>w|OafzRh(JzQ40mX;PZkP+bE1(qBNn7!b5S@28w-tbkSC-Wr>up3wsW`dGq
z*OzKJ+R09iGiD)iYkZT&i4_kL<41H>J{P_y@lE(7Faswhex;Y^NJ870m~Mg@(RV%g
z@dFY=*nl;utKV0OQC(9j=;n#x$8DNqBq$h0hV7P5@Y|5SudRp=)Vk6lB7^-~BoVjt
zwyVp7S}eiv7}452J#0I+Ysu2d*)C|_+&`osPVsZn;`<Mld)6O26h1HZOQ5loc|7!V
zCoO4x6{+b2wzL?aXpDgroWd<2i8R*=ch|YoPceye?4S^&y++(MUL;))0i9pz<q=xO
z8#W3R_HT}x-n%$&Gxj=?iVSTmeuLF}|NVTu#O%vo`+G81_nmTcyz!^<h5d(dmb6n3
zOQQC?GT^+m+MWAJt}Rn0_a=GHOP~7X+4t9pWE1pLHdN4G*#WJ*hHmvP_hw(xx{~iQ
z=aAWZMfdf@Ua6^6^uGtvW~(y=RMn)&+Zfbix4x`)uP(b{yEwUXhq0M5!s65&TO+Ga
z#aMqZshE20oZt}<m>!<TEWl~>@wb=ti4N^IZArmVgOl=;0dw~YV=Ld^j7Wl{EhzU#
z`g=F0S?gZ@N&i!C+<0gDdG_tgFqyN3?S?I!kKyZ{4)e|0H8%ui&4+J2OzL)#gsQF3
zzx(`P!0~yP9ltc__N;o<7p$zV*+g`ljdc>y;if2~8>m9?Ah^tSuz2}rsKCUY25}|`
z9wA&OV@*UUg=r#<<!veV0-UFf7-UG5pYBGykUpH0zqT%5n!7snyFi@H3xig6^lxCp
zIq#ego!9N&k$NG0_*T}P1ai5nLF@~SyR3w6d^+CGz-UuxvS~kmvzDihpmyv2z56ng
z1-ASR2v}KN9i5*{-00tS{(QaUrG0}aQT}TYb&e7iJ#&-fK4+#!*)ES%Y;((cX=en!
zDA(s}N^t(vqh?Tf2QPoOYjl&6yGm0hy}eK3_r3M}5tq^g=*&&Tq!JtLU)oD;_xH0c
zrEObNdM@VqT8&NeT$n~$w%PA5?lLldD}~)t7jBVfzq@fWa$Hv{afs?dIU{?)$b4rF
z=ggUb<iB+KSzGSsjE}2Cn4ajyFP)zLBxU&rtd}>eczAlY&@`QU5fzi9fAJ#yvG=l|
zeehqx>qn9uem}o3{~;u?;!=ulzrl^ZE#=cSZtp9uAE}>ja(iypaQqYbuh=glIv#!x
zFftFBF*m+|u#LE`3}aVngAgHML~O2^9$yr>9(7G@VXPHHWKiyFdCaeeT>D=C0Tv+m
zsq*4*rTXKe?ZboqpAS9%HaIDwXcFJh&;Z6w(bhP6*Sj#{TbZ3-9?m&;`O_6&_JOi5
zniUtO-K}5vZcs!NdHgiLZ))<%VrXzf#X8XSCZ(-}^}Bp|Sy_#xS-F-spM2k!`*KxY
zoNxa8IWDonyVbDeaQBm=q4u3`L^c*!_ze76OUefeg2KrIBJo}JbF8qkKIxu{#&r!x
z7C}wNu52`bU1b%s3j)}bM(P}4W3{|jXj8K3>b95Oy}1w$Q&O&fI*Hti%U`OOl{ALL
z%dDs;U%&Ty_U#?P=bBaI+U^X!3ASP!K|$y1_Uj+R8dqCP-(*1E@4t2X^_O+UjgQ40
z-+t@=qyJdu3L6v>E;lL1zMakb{%!xASFNU}2g-eI|JGJNJzn~-@a<^l8%60!Z^f0%
zN3L8vB2RlLma8&V{{H0{oyB{p0z5!F&57m3cF5T%q=J{YPXNA90qE=mhfmw_M(*RO
z=JUO3K-FsTGP!M-?fUo2X3YC1BjEcsJxg(Dspib=K<Q(+A#`4x$H-$T9P1l&K4U*E
zPitJ(BO8)U^$`2wlGJxl-yyq9`P=y2;~mcqb6zT0?fhn$*XOyaKrS5Ba>sK!sftdI
zJj=GWhDfOGg=DDgL)~jA35jees8Lc(vfl{ak=QOH%yZygY@M<Djt`qRZ^j7t?vFjh
z`#dkN&JI>pSZ$gjp5u&<P`J=YwJnRN%{aJ8WBTB<j6<h+bPRGh@8rCqz5K=Jd{6Vo
z#ADt3x})Ayf-!e?=(22Q+-?4aEa|GVUAfxP=(MdTmd|-p#o`42@aW2yUC`0Q;6{rG
z{3g~HU*ZN#T#5S%!VF{Pj1--W4^&}sk+Ch5%i9=B6NZO}c|{ZSG@j^{Wq#O{ZCtN)
zYwNV;T4QL;xz8>=Sy#KK+!8ka2;C5vn&+jtn#IdDJbm|~wfFSfyVW22wSVbyONl%(
z=REh)St90?<hw?7vuAd8cXTp*l#Qb+NMHd^Ln9u>A$uQ5N%ky6un0&u3*nuiTUMAa
zQ?^98@6@Guxak$?nbiTDSl+eHrNs4w`(&lIh>CW)a6LVEy0fe1V;-pEba?~2m|RdI
zLQW-DWWI+Ua^=y{$oHc$)`3L1YGplY`|a&H7Sh)&Y<;NDpPzr1Z?E({NzKHp&}=vm
zd8Xu-NHg@i<+G5;69Z14AbD_AjW)Q*dpB`0NGJF+Ga`yCXjj>-O6NTz!sfNt8gw&E
ztUY^$U<8>m<<lCCDZu0TP<mG1+<c-k1F)-6Mwj#MXhZGW_oIYtQw5LFZLP9lr=`tP
zUw^L}GpFZx%IU?&5VyXN6<V793TNTBoLbQWci5A~^XG1;1}#y+$DtwN*usE_P9l#N
zdP(FYh2oAaA|xG4X!cHO75!`V{A)X&Gxt{~X7#!N7Z}-X`F`qxCt@k=4d(c(E<R8b
z{Cqz{N$mHk1x>DXc~itQt2|hQk0J5@f&#P3eNkgNb&}{<MO?DVyx!8n(wEY>M-%rp
zWgh)tC&NK6QBtgJ*vYN1=itV9>O16zlXfN<1REr7SY-bGEt|A6&pgy17qNhG-#?t}
zmv8y8<KVa?JL|y!K@EqHN-igoLeSW|2K9Xm^joQKWtruCWLEE@z?BO`pr7&L*RJd6
z+&m$q+8wy=L0s%8c;e<hzOkG+i%+X$t>d5h)B9&a!%AWi_!XCV*HrVOcKQ}^HvOKQ
z{$i8ZBqgHoJC%4-)3@;NyRp|c)yE}lCLnZVRVlhH36Xmt;+mUbtWg<KHu2mtX+K(M
zDQqcR{@~|^m2j*|gqL{Y0ZY#VP{2n$x_89KhmWPZ8{f3SSz=^zM%T`c?ix7-Vl26n
zHRyU=c1R_1CU)HFc(l>)gzufmi6gY*T)m@2)J5^6wZ-YTSx5GUvuV65Gp0a*xWVMz
zZ#wcK@01p+&a%o~ny)|i!uD*R{IHZ=?jcuWnk!^4Mk{!wShHpVUbgF=>B+h`jhTJ!
zCr_PR9S8Rdwq`f#?=)rkKD|gb)jHY=Q|tG>K-<xg_+{OhwcwF7{_p)GS-IvTRYNdt
z7~*DY{yV9!vB-$d;n}>ws_hcf{$^i}L}^!*o|k8LJRh?aJ=T;-Tl30JEV<to2j45Z
zw`u(|UA|XR`dRdRPrhY`&~Tp^H@8yPmc3$y)Huv1-tBwm(lO@tOV%JM<V(*~$-~3`
zvHqL>@~)i_R;ai%(l_+N-E}_k%;)FcAum|G*Y`><?$$HExYX-pO&=EfMELJMHZp~)
z$A4@{m5nE-n<SYqPlya_Nrrck{ir<m*UPQo=~H10QDRRTaaSu(-V~R4o=tPlF++yk
zyN$D#gyiHBp%F+b{rD|>+;lsZD9@H><U<r|vEJ%ddABmT0&n-*2IO09`epA!owzah
z=E;GN1J6#$g@m*|wJDE2XMNbl`+aoJAt(7?dtLj_zjAwi;!K~?RhxHyjkRW%Q`Ba6
zo^Uu$cBomB+t0h3aUkyT4cwEd1y|fB(MR%E?!UEq`B1UqxRrspn-sg`o&@9C0LN@!
zPP3i>P6eNyFBcB)bJ&@>w|37y%G3mzJGrdQd?%K-#Uva#Z}&FnV~)S_b#jU_uN8(l
z4n@kK@o2MkKY!h4Zd&g8U*wnfvP++($baA`8T~})Z==X?&f8-~83!4ht@W8iNz3}h
zsNJYN%e2U0d(h$ZL_hZAI}uzBCw^I3>Ei5=ZFa?OTvlx+7{iuz-E}h=Ps0wacq!wg
zfglOn{9;St3)626QdhT?s$A=u(Y5*B=Jwp^+Tq=jdy7-IXm5IM*1bGOF=)NAsq&?r
zd+h@ZrW$D@Zx~f|tFJ~(2DYZ2uPJzUghSDd`(kC@yKiwe1r`*X<F#cB_x)%V3<F06
zkK4FdB^k_V3T)zw+BGp<qMdv*i>-;gX1JmsFoe|{i||fbs1<thm<+DB_+$(zFP}?V
zu^S>^qpJxm=@LKn;aIL@#Jt5*$GSrCs=rN@F?_dQeEJae>(^Fh^=vb){D_DMFy@Qs
zYO$A9m?zba?WEuOJ#KR>Mag>=nqtv#HJ#gnDJ^9jFx<{H0~JEKd!kpBsmJd8`MG4}
zi~N&#{9lJ$xzAVqIjJp}vvZ+;Bu#$3!AtFsTRn@de|^J(!<Er{ZKe&PV%$O0_ZG$C
z^<t*$7ZDhnXTBWcB+}5^s$iRXSVy4!6%WtL?Js|5)d)L1ykknvqt?r^^@7L)Y?Dh7
zM`Hj`qa_8NSE(5x{PWf!Ho6FfL)Ybkw~eGUKX#Z)obO7pX0Py`(&`uCoC2sOd7WHa
zOON-j3NH)%AhnUoetx{{(&zUzJ3Ef96mbkJZt~`RC3H$GcpGO+|7Slu2g8i_{J8HR
zRp8#c9-{fXJsLgr?&xL92Wo}#_eD#nGRJaS;~Y<`tp7gsWy97y*Gv$oGg-g#cRUrD
zB!g+S8>Y23NuD3=J{@L!dZOm`LykSZ%Uq7jLll1H&ifDU_PhTpOi$&S>vkKtu-o6o
z&<)()T+p#;{`JI{#ERIN<R3duey<DgTWTC_TXZnV#kEhOXQoxOe(Y28xYkzeUhTuU
z{iUa!M0}APhehH6L*9K2wU6kus9!Ow^S}?FZcN;4NgqrhYG}x3$ZEmEzO=a5$`$&3
z=dzbUX32>r<tpQf3oa4w?VFrRjwamy$retnZhR<6WaH;fl@qrHf9}gkGg<C^>r&cN
zU}3>)N&kAMuJ&AWOQaT!@xBHYc}UK2eG7fb`><4jEJ>H9yKEkR0#>p%J}LR*U07FM
z0Qs0%{4I`YbL5kMYiZro^9;+cA3yd!Grz4?$N_csNZDFUfn`WE!_ilL*ETto#N%?z
zhXxh}qlXzsdUAynw&XXyJ@Ti>DlcYcQ-)REFk8F9W^{lI>d`(7THzY1bb(4CcCxr&
zZZ2_7XtZRzikYAxsXoZ?p~0jmRY;t9?#Z5(liQPAFSr&Y7FoWIae0jXCA#Q%EwP(;
zS}hoHw=qXrkx5kuUX8|B_yOsM(GOoe{&;cwCB+)ERtuVnjRq~R9jR?AJ057Kyu4gZ
zH}q?&SC!W*-0FqP-B0J5opR-0Se!Y;w!=tOk&`E1Z_7DU|NN_;ocIqO{9Q5U=48I}
z<MVl^=KN|7MYhKHlF72jNG+R3bn3?Lhm(48*<XH@o45OM!{gAoo@AA!dZ&`2v(4uM
zPOP%eNV)ON`N<2$Y%$zsC|thNCHSUi|Lv5m>=Hq9)^gU1Ik5)02ef~b9!<z)II2S$
zo^x<+?Z`MDXQ1vR!_KLSlXh`hgBTJAFCq0G1$6iBjUn;Dqk=n#b7OqUOT9_2^4>#(
zO!9FZCKjT!|J|SejP}cWH%F&aF~8~Gf4fM}Dce;3K*q%6otCyfaRxpAHlS~BA^{3$
ze4O>uvevrkuX4-3^VJcn!(&wP%fHg(e~#Dhn|f!#+cjyw+S$G_mT4QHn%iqH5g$4m
z+&H?HufW38?91ifv@W`H#9f9}>M}`GabvH3;vqrB)pt`yw&m$wuwAH_$x5?4a_&i1
zq`tn}{it%a>!735MmeXbq3F{q=4{d~bq5U<UuWH`-%?Q;^FC!y^MgZ+C+X*gRto6H
zv0P+-^!<);3gn;<zQ(Nm@6@_uJS}8PAAG&%S?EMDQ*eY@(XZcJ$0`GujQI_4sl}8*
zgQr)X`gV@o#pxaW-!95stV}9DU!#f1nyX6Th1-!kwRK)zHm4gqa|_?k?#tG@-QnbQ
z>H4w1W1{QpGcMG>Dq{>4{X(raSRYztcBfYSh$)y1ayyf+Vz%FUi{Z2L{nYN>Q#B4f
zm!HoMs-DY_89b5uTSoS5ACvsAfs)q&KTbLgDv{ZKE~XQyI@2yyadFC0<UAi7wX`(a
zj;BsX#l+}4IX(Yflq%4Zdpx*h%{$w+05Z6m3APr+YM<}xx?VyZn-Xo6web@7u5KUQ
zz7Rwn?XtHcI=aW?ybOPH%QaD!gY(@t0TI}h{P}dcb}x;&_JfX{I|Ke4@b_IFznem#
zQ+hW%@r8unFG@U!jGD}3?p^ENi-!%8WZP@Lu_HoBINHy8p)HPF`2G_?wHPL=%c&PF
z50K2)tL}6J&beDC_fB27%)V7-2fs!dO<GrjZb5Y`@$YA?%M0J1W~=pyek$ZjeC~F7
z@WH)|01Az?kk9>pUXGrR``P?N$vC&y`4F2~mXVRW498s5Q7Zmk;n)+;tmsarlsXS2
zD_+{VtL3&FHRDdJ+@9ZERmYevkz4MnFv1WknNL=Fli13}H+E=I=^Q<%_|4uYTdKlw
zYMcD}ikIcjrw=+}xA<S@i;90S^e4-#wg0rO@MuxyhH0ulaP`Z+?03G6W(jO@DsdGq
zVB+r-JNN2>x!>}>ob?;JncWp!EPLn+wM=)UPV?CwPKq5KP9TM`$%fIfZ#cHTo5(T~
zthhMW;$^6GM<eY^$a_)G>DR|DT}?6E>DPLimIc2q(#bT$3j=Mfq|U==oU)gO{P>G+
z_J1`uEJiusw6*a~H|(e9l5JWR3OM=u%_jtX-%s*i^mx7{f4Kbq-u^#sr_&q;id;n_
zI{$79XP30YAvGxLC$KovTmHpvX|^l8zS&?q^96^{tzSL|KBX>B4<V*whDPyNsa2iM
zvgb^`MIH-!2}@#l@F319&IY8=WOj!bWMWH}J*K$SkA_;(C5Da4V!KzKMY2xR-hfIs
zs))`<Uhj#4YGh{SdiPYiZYDkYnb^KOtGDQ?<rOv*f4Y=@VPs-&Mr}yF`?%rbN9Ed@
z>Y|U|zFw$U=d!5~+Gnez&9oy`Cui}}kBK8j!=D{~1gwo5|5cG8muGb-Q-5aX+~o2%
z%TA&fNHeSsYIVH0yd6cWe&&2tvmBS74z>u}-+*UsU17$1f;~3UeFvKpUeWJ9tfHmW
z!E5o*er)vEi<3<6Wo73?`a-l@XO-M<n`Wko$9J&)^5HOmVO(Qbd`{<=k0^@$oz&zm
zXA#+3EPL;3MkEcdvP}LMKPcK&<bS-z>N`)Km6VW&Sn$@0q(6b2<s)1<9Zsrw=Esxb
zD#ecx2Tj(}x^AVN)0fhufn1ENbh&}UDk?;ElVusx(J>R;p{O{7bq|Y^(LsU8;bCr<
zFOOfoe)EQ>*?I4tJyEKfcL`XC{;xN4)cYGSrt?LW5Bx5~W2e6Uh?vyYhZG;YAVp2h
z7diO(RkgO4penwddbV{skbT*(A}+w+*Lv~Ga(<HBi3M+qAKvfdGTKuZjz(R2Nuh3h
zXRkQgR1b$a;idG>KT&fTr$U{!vP#{;Np>f-Ebtf0UdjF0XVTr?H`P!CF8r7|zwl*{
zWtY-9vwKnUSP_|*I#V;gJr48`QCj~+Ymk}i^AzY^L9MclO>4zXmJ4|`zQ42{`MB-0
z%8MQ2H}E(#O$m5*e0-|pgqBu=qlAOlR}UGS;qO_!`rU2f4<_=3v)h=8{r*)EpZUml
z^;TdOUldKLmA{eazp9iv7=^24U27+G^UO{8c^~3{|5Y*Z#MLGM9S(7J<bTUdFTV8o
z2mzL1j<_A75P|M4AuhOn#BSy75&6F@TMv6XU+VbiA@gv~chkA4u9!2Xb7{7pZSVpn
z0t~MP8#M1IF1Y?q?Pk_}5z+w?+F&$W<79lX&M$Ff7P+?1|H2#?f9!qS0rKc5hrOtb
zuQeax4!^{Em$r&T`<)#ds+<>;*r>krxv{f0U&M2C%4&6OZFpj;)nQww-SqfE&(^id
zb>FmG_qK8l*0!vdmz$5iX-@0+HI@^8*3)~2T{27!r(LPlr{%a$n_=(1Q$J!U(sBHD
zuBrBBHOcYt+>a--zg{@}ZqzW^cOs{ad%r9dR@->6s`~rF1w)PSL_H_-=NySQvJ{7?
zcZ)u7?k$$pTsm9+eD(m(y+<$T##B^Bk5+ChR+Y=)p7{%gpA{XMZ4azOemJQ<9U$g0
zt?E~6pW*v8Y<pn*Wp?MmLb8JDU&0i>mk+cK+?+T!cw?8_=h6t^NbE}~u*>(HE48X5
z--N2+amodV(JhylQzxs3Lc33=<vQA_xZXSw*Ff7+tEuNH%##zJc9Z{2^r$5pI#mj}
zjpVa4v8=z=+TWZO`x-Cl;CXKn^&voeluB&AS~_dQ_6I#9R(QgZt9}bpz8r(q_g+%`
z{uJ3jYxf~~;elQg`_k2l3WG0af>hsC{jaY+6U5S(?Y^e>Osy#F#3$ArG&dM}b?;3x
zxRl;f_?`d9KBTsm%hqpgH(TD50m0y*A?jP(I2+JB33RPhRntBFv*`Ou#Wkhh*LtDs
zl-O6y7x4{Ae?x=VBrmT~fA;E~iA}`5dZ72j)Ao|~+(_G-HyUYOSwV6cS7<%FJv-yF
z?jFXsk9z-lfhn`#;~OjCaz{s}oxabPd=C5bB}5J<to;5R)4m2i+(_iH@6<h$pDJeR
znjUlw?8Ap^j$NLPk9%Kya$-ls)y_9p=-jPOVQch!U*_L)Fbcy-`l+8!<mDzMmxq7h
zsq9)d9PcRiy7QF(pb%}6`K9;wf8`yQob0)0trx!_P@8WPYM#4~zZo}NX8sPyITO|V
z^Q&t2r{=NG{eR&)t)rvM(#c5eKOkM?+Ia-yM*F@IB6|2h;sbBy|EmRn(!7I+wtyEo
zu6ka2|NaojK%gbDxg^8!S@(bKUbfe<+6R9R`~9l;Hk*1rcHv_J-TIIEQFgXxzI=*x
zQ4t-&3-RN*e73uH$C)t;r3)1vOUPk3Y9R8R{<~nA;O@DN83x}a&Sc%pQK9slUYF_q
z-0NF*V1x+Qs4stD8DLeOKvDTO`J$x&#ktxTMh8AmsWlE<Z`1jp@|JD!qs&MC@u+L=
zsP3YbUMQo&$BTn#X+mZPt0U?)Q?r<i6QAg_b>-<c>u!%cXDui3=OzEKQXfvX@LRJF
zIjZQW*(BZSdRmubbLvdad{wB<zaW1=yz-L)qWN%7Ctyfc@GjumWO27990Jvje_!kM
zzZC-e10GLpblP~|zHP+P+bmtM&b4&PbJTIFz|UGsYrNx<qUYn62TlwPOkF^47*Y@+
z{o>cxTl=KrWDBZZemE-q*Kzdb4YdTRbChNI*EnS_a%6V54wL=KAJCsHZjH<2H<&2?
z+;vF@Lt4xt*>_4KiaNLF-*M(FOy07+^ZLS<S=sGT_qBBlUaC$W`E+>X*}2|pKHhss
zmsgcaJ~cl=74^J4?}Rh%48yT`iDr(7+?`(q7_jXsHMUOVGpZKIe8v;>YL2`*`=La1
zW8j<W-L95=h2LcxSDQD}<hQ<4|CxQ9KUIZddTPR0q%{sFT4pyW)v17dI9&f(j9V|X
z#+1U-jCr%mr4&eROox+e_~35>yi;L*$0Qc*y>EeKyUxbGsVVLHmbJNqWiF*{lF6Qb
zm2!J^oXn-T{MV?An+=P;^q+d>+C6=1)44ZZhVC;#_r6}Rg7)p}Kn-2&;0Y%MiNuhx
zW(kLBH85-k7S`nDFIZIm*mol<q4?^ht8p*A7ia2hgJlConL83CJvw^=2P|JZ(U!X0
zB$D6)j=zWp<OXfDdTuQC%L)UZVcAD(0oAmw6H*K0-+z5?NH~4q=bOyaj)iC1FK%aw
z($>@A6W>pT62=$K_BP!^HQ~^?mimL4KfYe?z7jZedTaGh8~V8qmMK)Se-B@HdP1k;
z<F_UI_MuHwR8Su^^0rz4@yEcmvA)8-l^V)BE4;*ikEE|c-h9+urnTku-eiq%SDWWf
zM%>#v6MTACos<?B3^IS-I{J1kb86Ypt|j@S@&BReE5oYXzos{!pn!rPU4oQIhten|
zDIg#!DJaq*-5^~8B3;tbozf}NEhXLE4e#3L|2`kjb$Qr(-}f)pnwd3o-`j6zd}~<`
z%9KJ2jwz=*3o)6LOZ^y`QM|Wp_~SpD;H5Y^QjHF?iCEYyUoBRYZ7r;a_tw!odPEY}
zEeQ&2us{9)p>2?^>jN|_SP?fwX|pbrEhUf$7~Q}TVo#Tvdws!?xu8~;PrK!eXEl8m
zdUAqQ3f65XXy`~CKg`UYxLhUySQr3$GoL%Fy+hK+#a8><1dilSW4g%sKL>%BGRi^o
zN!%T{fK2uX2nfI&?1oVpX~|d@a$NB5&i3<UK&DZn(+%N%c(n<$&aHPD=jS(rlL~ME
z$cOWAcF|^$%MS|UuC9%MYn<$SjcbW!V_yR48pPUg@C?A$AA8HooUfZ@Y`0n7E^;d1
z;E+cZq|tdYV=0w=L@~yshcdtJ{EFc?(7bM+TE7H-{?gSsOn{@oH1so=2Y6e&K$ep7
z6(8?)lrWm)UdJbGO%OwNvz$j1kl+aSbcAPuA<=s<RI51}H=eYcc5yzLeUbb2SLDiy
zl}4q72W7?NpN9(F$(p0;4(n46A3myf*|kx{U8VWdyHz=B1H<^9BhFvMEfJ0cI35d;
zH7q|^qf5)n(>;9}1;!eR0Qb|VH9;@=?Yat+I{#S@Yv7u`NFihW=2LWjCnqxd!}L!p
zQzaOQl$0$kiOVfnT+CNXPEZ$%Kc{Bvwe1LCjtqF*mdS{T1b)_!H2<2|!r^sxye<~r
zcE{Wv0o^7b({QuXKr#0Oifrg_)<*nw&NHSJvt*vwM)~?;yv$7fZN(Ik{p0Z*sbRLL
zibFk0Di;0MBtdycNp!75En0dyqL_Z*8V**{y}yFC-T{UxqhS~lNY4lS^pQ`slax0s
zK^Yeq6x7AJX#rh3^!95n=W<^^270&RpJO0M1Ez$MV7eKMu5?Fo-)?|OkHfxj3Ou>D
zH5Qalr&a1|v(+vx$f`S4Xlh2b6DBpV$Xu^c%HD}hp4vyUT*s7=J-MgB6%D3vFO!7j
zVPYRloVK=nZkjxaLrs${Aw)t90nYUZe1!G`0Qv^YlznM57h|6qH#J<jIBX=Z!mrrx
z)9WpES*HB<O*z2X|EckKrH0+ciNA7<h&hUH2<7sdAxn^Hz|iy4@Yq<H?%3xw=lxSF
zwKc>XCO>j&J_lxuvKWZuUj)mmms=hDy$wH727GF6-?{~5+0MxZ(eD1?#{~*cVm@bf
ziirxAetyt*6;4oWwZpS7H+^S$@Ym9}<Ac^Cc~w8A?c)o*V$6olLC=#D@$pi#MkujA
zHa5`#YPCr9P%wu7aBr$?eJHE>VTj-#G#o+q*QvAh1$BxW9GSUOcL%vRS*w~rKnQuf
zYagRlK@|ozS)V|x9aLwBH_i4|G5_}8ZR4l7uW~lT!hrmERj;M#Xl=Kt4%I%gqUPk}
z%9<{HSm)|WjExRU@-`+WQ*rLFQR}S5-_!O7<D!1D(iLJ#nnLI|)@p0)kMl-Z1(e?y
zu(-ad4t=^@x8mBP)3?9y5a+vk)U%SYCN8!b$?dC$21BI;+coFH#v{4uF6Vcd-<A4-
zm-o(;T_-WG1@5)eygYYe-2n;vmt&E$+$K-!g%wN3L$aRhe~B+@(d8ZHvc|xmvo^SK
zrW@YF+j>{D%X6HWtjtgIDoz|{{|uqoyrsWCT!5La!o*FAY;uHa&6Zk{6E?>81gIN%
zR2x{3<@w&LM=FsTv)lP2fd1U45GaK+{m0-c(n|m)Igyg=z}7E=tF-+5vET;41yin&
zO!RxeoUr8cDUFufX_fBNrR4nwy|2p`a+G~-;R?!@Qo;nfTE8dWVfZ5kHgQ8Uj!4t*
zqHyRUv{_LEtxt#@o(KSL_3>Ru^Xvl_mWbY^T9*!E6dogXt8rQwR`w4HN(1mbsw+G!
z45bj;LaKyt?$5#>s0)n{AvwcFX9smLPx90?Y&_gaub}hCMi&~)oNV7<pkl2^a99#4
zz_Swehp{zeo<|BJhzf#W*uHAH<IkTt5FIUn^$aqpR4$g=ADJdhO8RQU{>dWsl_+ZM
zInFXnK<mSv1022X>dZJpF6I@ZCHFoIxkDT}urs>$$n7lbLOP$>rQ15-tv2N&nY!5T
ztH1NdEz$Nmx(XV#!|U4P-32gE6iLg%R7%fsuP-6K$xW)ZN1(yw7tvX-f8HZR&TH?d
z*RHU8q89=ah|-cU_is#jZ??>E@+S&Fa34;eVyz4%PZ2!T?=G2JRXPfApLRZcgys{x
zco4&7?{p6Kbwn@-`vOL9<zu;%!C(y_3*C}2DL5^usXf5c6Vi-mupQapQtbPP;aK;Y
zs_PtETYH)_uX)w_Td0KxwDbL`PEoqc4hPAr&L_LeQ5w5UZ-nXSTDPqE>qF^T#5?Xf
zi>f?jLt#5RJu7xx9O!f%yI-z1s`c-Bg-c&9;aQT_)2NBBNER&a{O;n(F8&r<8O*;Y
zfE$CRZ43kq^n<fI(hA^ezsZXHZ&k9!(c)!@uwDy==j43<B-bhmDYsD%Ot7ydW`ds^
zx!@aTmLRYPj|9yQG;W4}3sG|1&W^JE^XwFbkWdiX+0>e=86frtWX@{(FTZj;Lsr4D
ziK4^12aZCSUpH#fekwRP9M+rbm5e!j?AhQfkX5=r$3|<hX3d@1D{ggWQhIJ_GU)(t
z1Qa)oJjQsWq@tre6w$n{U8?vk#$Zk-wuznxpL`FLKq~7<RGp8{|M{i4-t5J301KD4
z&c(vpiz-v1Mrvw_L=mT>+;0g9wIi$6cfZrK09O*hrbH2ZLrvt8HDNr;zyqWuOBrSf
zoh~nIH^&q}UlkZHe{nJ&Qc&5&kmTHiG>Cbb35F>ZlShkj+pqgdr60n50~oRE)uNrQ
zuCQf&9VA<H(rY1*vChcQInZa_7`nx%G<ydq=Y9jGJq_Sh72^Y_aPMk_E&&{<y6|Id
zeO~juKaqRC0SzN#p~gu1tE`{+_g*IWS)3g`8kZfPwFpfJSI$nB%5RMdjDplqyVjP+
zVCjtHtur4J-y-pC_UJcgF?pib)$zV(Uty=^q1yHnhu*&er+WJ%G`*TN_F>@f&z(8w
z5QUUnNN%0t4a>9lVgAtbEl(F#f1~C2Z!!Uls*vHYa`b9=1KGn77%FDN*#y9a2_F4T
z4XXcjQMYY0j7iT&H+<PV2Mo&|c(zt!h_dGYk`AMKSyu;_KXL=+j%WbJY;m|Cn?tuw
zMRz`I#4|gaZ1f|YTa7xqw9z{}=E^(YUCoO;^<I)_djDg9|FU*kux2)oMH;8C_a=a@
zfPkR6_0;z0aEXJl5^{cPKbT*~y}#1y$whB<)ul}*Q-gTNZT_X-fO$_tRP@W#s`XcN
zxq*0ZMv|&My6iZ0lnCI>n|-huZG8@cuu;+S6+$Y2-ri|t*xu{f$mF#-V{UBZ0UX$M
z_S?DyY{R|IKM`Fq`&5+C{9%vqw;xrw+`gT^`mzpDcd_AOM#UKm#ZgpSFlm3T0~aWA
zp20X|Q76PUc!3vgA;dJ5o1H9X>!o|F{~7g7a_(t4ZzEjY`X*v4r-_8I4Vy73LF~qE
zKVwQz&5_y<t~Lr0w<djp-E@l<@bEsLGWmkC2=rO+FS8f7mN#sFt+#7mJ^KUfiUN^y
z_qX&;|KPw()40f*s{yN0Q>)dg0>FR%?wNqF0jMI7^74{@RnaNo{`en&{rL=GZ@=Pq
z3+M5kC}_b3shqz><hW*nrsJQ}I*-|==B~uC4`6oGo@VW<)_C|%i2zW&FCrfMwLQS<
z3PLX3r6DInk?lJ`nnbR<g0PI9?{s@12YS&L$JhR2CcuAik&9SjwkAUDn?7u0&`<jf
zdd$7uUj-5N)6k|wvk?9}H>f+`$Vp+07a+n4rSFE4pp(qt75n?LCDoJN5d&_>T+*cb
za=iMbfSox1LW`M@E8&Xc;KlO+)t!V<w-tl?60^fZN$Jl&2#w@U{5JL6`ml%-$hK-!
z@G1*zT3e0EblHg^uJ$whbhWhqgiW|=c-TKgRnt0?2UkA~2`K{^C@(W<`NZtspTE0<
ze~Vq|piin|707s<1I=LwP^ZL#g;W?g6tnwmQHA<blA!sD&I)f=TY1n4ROZ{m^J>||
zr^E-P<`c0;rs`Z~T?sWiUHHytr)th6);Aj)pUcJA<AgH=rRqQ0O5f>ou$agbaya%p
z734G~1=#Nr183HPA0;Kd3RCIhlxVU#6+BQp4X&S}Powp?@f4e{wW)Z0n;WY|p5T*=
z%>?Yj>>AfuzCeJBrqX@-$?FFR{N?kP*G^NKb8aI+9l%KU!0+I-)mf)g!&Dv|FzEY1
z4IM~ozX1lJtnz;qVtxx&!-pp&F70NL!PbY4r>CRS9bH{Z6-jjxIm4VLzY$vUGWsK_
zy@i(DAx}TNn5Z!AhCuL2vk5Z(6I)8Bvzwrk`8n%$O2x>c&b`=dchxCiUi0e6X-Q?q
zf)_UCmYFZWzzwU`@7x~_Xn6lD_PH;pm@|w``M03yDX|M{Bs|osLk|Hm3eT50zvgzI
zLahLece=~ONNe5ImT%HG5SlDbyIOWcM~Gn@f0`Yg-kbk39c{5qe7p}W@tie36ygxK
zkn$>Pg!<U=wq-X`y=m6txt=@oN7Z;r!F6}|so+V&cN$GjAZ6RbQkBGiRX3<Qz(u?|
zStKNKz31e=#1^7U2h&O*bzJ(}Xu7rLM-mhBck{?zLoXkbI-B0k<_?uPnvmJrt7Nj0
zF%#O%wAThat?R%K=c~va>AoW+Eeoi)I)7DE@Pb05!)@9*NkY5p@m_*jG@xj|jy4__
zcke3eL#?vG|MJrZ*o5t`4BL>ra?5Zon|)~P(J_lecPzIp&emJrDq@>u_Bze-l<^y%
zx-^^L9yRxlnmRc+Ji$gsgOYBl7c4qqPiTHe5<^#;vs+@D7QCx{`SRXoI~sE2HtXUV
zpCgv1r3#Ll^h=1W1iK${P@~`^M1qD@tNgJ&^dkJW2C^S6J4d<KppW9pl($td2XSZg
z-~FLn5-6r_s^W~w{QuR_rrFx0+hHSedPK&CT+UHlf<_DMt{0~jM)>hcc6ap#BTY5r
zdMkwi_wcEBuDwuIw*EAkQ6`96l{!1O&HY8A*VOC}rRLw5Q!}ev{RDNv>*@N6xUkm2
zgbg}=$V7YyCQerU{c@Fd+~>BrnXjdGK1OZ-O=daAA*Xr5dR4qHM=k<vNbIl7QS-?!
zTIPrc;IdWrjsh+1|CVrDf-019pFanw#q$-r3gO{}XUkiFl5!A0$P>p7yU~wMPvY`^
z<O(7HpC{*c#!j;?!c~3ys|g>ZOOXbON5#B0OU%KfD!~h#I(=3K+w^DCBY``GxIor-
z=2+=l(Hr4dX**nHFCH|D8gra?_*Q+I3)1!Fe$<maygu2!1|)jxsx|9(eP$$j@m-wo
z^cZiow&riKO*_G-tBX!LY*GvLGR(e*B^kT@4B9i+s^qyl3pqPKGLxK-=$OjJZ|8<0
z-=-=p7AE%36BPdn(`osNe2N@7#KR`z5{s*I6Rl&eX|~h&4*WLFU87MEq21qTFhn5#
z>hmAf<xAIAnYnp<5LrG~x(~HAML>!(L-(8i1l6}0a53_Wo3&Zq+K!Qo=1;eP++{g<
zbXg?_ac5(i2x@rOOH)m6^G>sh`T2F3I%tC&))O2+{+aii7>MDIl->+Af<lk2_26z{
zQMc>M9PP$Cb%y>8c+rBthqy?YqhU$n18!#Q2pLt20yR|5`+i7@(DIHell{*5_?@&*
z_ubqEm$g}E52KQDc8WeR%}4DaAH`KoZ@$>xVnciTu*6cmd7ZXJ*(QZy3S55@G8?cP
zRae@eqm<knPafWS<I3Y$dljI$9sW(ZGJRVS70t76UGsQJ3uf&s`4gyQ54@0i5#!md
zfec=Pe~zEDinuoamPM&MTIWQf#KLiLmqNHLEVcbH`)+aStE@gJhi|dAVIjhresoWd
zaX~ey1hoduyo$Y}>PAF?$P?3u9Ifj@omt(U)|P=XD_>o0g<sJxLxeq*vOt_(olOr(
zZUFbrD&Cin>H3cc9Q`0*LEP{yuJc8b>vLP%mtiEp?WSqUG0+I+zC-`5)5m3b@n)&J
zpwcNjZ$uET;Y7;Bnu{}aUW&W!4CxM-O10nE5?-n&lc^V5^F8#r^X(^E%-gp?aqOPQ
z%PwuP+L5!%%T67Fa57C!e!BJK*Pu{7(K8d1+!V$?0t<FiME~hjq#reC9D6CxdWI0)
zk7?}e3<pqm5A;C8Po(g-D4L5GMfBi4{e9L=h{d09x9nXvkb;3cY~~Dw&to%jRkgy5
zgg^Bp<`mx9{O5!Sq>K6|<fL-`!wq=KJpKb(@Zxisc-oh1I4>L0sgEUeez*BU2OTYK
z%cb+{*MyR(9GXhs`39W-qZkI>+>J+PkLj*C+!BNbA)!w!Y>iuRy**___5(UQ#d<9Y
zS4~Zhd4slu5RiHmf139A!ZrPQSk`ZC)_T^zgdp14rF!OZ#!E`Jdl;;=v>d>6oXl)i
z*=uIxET43gNcH|sfeun4F(f$HhBu+7$X%7@(SQ5T2meI7(Mu*>_mV0uuxsaHqn#Ju
z<ttm=S;jcJVbci*^Odc`O*(BN#rxjcY6SrON8i{8)qf;jW|6Yv)nU?Rz^(eMIPCWy
z?NN4kWB$B5LwY5ZdMoaoU$c%HYi^p1m3zKRPL8Sp4eM}25}pLIhDu>3oM(!Wh3cu(
zjrrX&QYGcRKfSy=W|eJv+zh1QhvqgGSPX2(HtlT*FH%PeZ7=EF=?2KXe?AFN*#k|Y
z<=#Qep28D=>0<^Nl(-lbX<P|99%Tt$4`?(s8Ku6~J{x3rW7qc;m-%mFW0CFXH4=3e
z)aS9Be6d`In=HwBhn`T@tPEQbG<UI2O`ERJeE+yiK~FS4n=F<&jOxtNt>kC)@Qy`N
zKM0th9;?=PT4;053Nn*!VAYxWSv@{!*++n*u^^ksQmoAX<P9l;m27r=1m2ezZ1S9K
z!ByeF?Ws;6Zen0qa@1ZFI4AH^z{~*c#PH(IPi_h7><GP^ICAt>Rd<cU62lt5lZ5Q`
z#1-hUuh0#sS+4AGASDbxe#ExcC4z^ZWNHV9n~)^>;(M;pJ32)Rd-J{Z(&AUtjO?L-
zM4i`8Pgz+Lb2pgpOPsz;VHPi&uiJ)h?{)rJt;XxvsHpbe9O0FEzY$xc94aZ!1aL$-
zG4!VCck6JI$`|cN%h9_};j#SpdIbNv^B|tY1)E3Zh5B7^onm6n=sKP<X#00`7z~lR
zOW=K^&Q73dCDYZN_p1+8T#9&^aLW7K1Eq%VSi;}pHWCoa$|9OVdql(leu}_08PmS|
zZ^JpfeY^93k1qIcrM<rtHB$2S%f3isG$zit)ucT1Q%Q`6=YM}&weqDa3Q*%Gr=!K1
zmyVlm+ZGoshLO&v9UmW`sWN1~Ols9H(L6fb-zBn#iscq7cj9Pz`<ez^k}#3_;lIz~
zdna}k-108i2dF9Pdyi;oE9s@BA1zk*fXt8>+Yv-qg#LxN!>{P`&8FRrEv6-=owe5@
zST3(#)Sax6xn5e{cC3vOCn6TqB^Nrwem|Oc*;*(3@hla58MVrP`%IZJ(9^eNG5!g3
z2Q8X7U%$*PuLl%rZbxxJ^!S}_8XZAd__R!?Ks?FHZJd7ATn5_M;x7eapRP$%MO_5t
z5|bVdSxLY-0O!#dlqolxZk<=?oeh~*Q+!POCMCAOJSow7eIU#dKO-Ev@TeY&mGP>F
zpc8_(@_>sg&hc<vyEs0D5zNv;zkNe3lves~M;9ylovMnJ&8u_2!59Ip@rTU#R!GuJ
z`I58#N4^##%y#y}=_MmE?>7J;QyBy97RrA2{uyb6wtKT)0&u+O<9}p9R8QApUvqv~
z5SkQnym1^<D@*gvZ_K1or?2GW`dy%bzl0P}IXJvlY-k@P0=oj)NVrmvK~doy){U#y
z|4=rfq%f80(k7B|S#z)`p;tQ58yu0<|K0Vc`U834+7GQM{<KdYbh%^yd~|`IqxtJN
zl~E>^R!daH`1bwml&Ek$q=03#^vlKu`N(zA%?KISo(tvYF_)VH$1~}p_@KfyFd$BT
z|Gt%8D<B1FZ`%5SGtZVk;gD|kKS34Kfc7(*%C>kCm+edE+7BcXb`U0?G~octXrR!6
zsP3JHLrw|<U1s6jvTE+z3Rr;atE<o78039kcdXoQpth#ucP~gAyREc&sbW9Zn<pUa
zoU#n^X6sEp1Q<&KeNee2a5baXyVHY_qysm^8p3?Ch5!erH>W2G9Q(Ai{yone<Rj+e
zphq6DD(v1D-}YZ-;)C?4IYgvA^xix%ha;+a?fE^bI)Nx4p<e$<m<EbHwtN+=@!gu%
z%E<n>T-vOASqWc>R^=K``N)GL9bqnAf}b*eUbTZft(z`!v}}^~lo(bWCfF&R4ERej
z6lQC29v>a`gPQo&tHA)l<KMZ4CR62s=-mRlO9E5?<tRTwtFfEbws&&Uwxk5I4DunT
zW-X+q<^OAJs~sqKRoG!LT?|Mx5F)^5lJk?R8g>$0ADuhWPYU>K#<|XWRxC%JpKaP=
z03M_<ayZDA(*f2z=&#Z`;CUr9soGK^WX)JXAh<MtipBP;;q(LxyGD<ByY7%_Fk`YE
zKSZ!$HdP7kxf;xg4jhrIc)NesX}_&NLNt@PU9Ufty+|I&|Gb$n0V&aY%v+_x_r>Xq
znKXDQpxs<T3H+9AnF97uEndc^{Z}{D#@{q+;z^Xo43&4i?HJeHR!rrZT`J&e6_S5Z
zY;@<gzcfMdRqlofgiQRYp#+Wa&~7+lBCYB+^`1Op4wD`>tAfxu=Q$1~lkJivzW&Y(
znsOyTmJO;u_13BCi>=2(1g{R_F{_^1EpNHNpjXx3(Rbwre@2tsTCNhvc!}^|Ot99D
zmFwB1-`e=9>9o6ByH_MYO^uQ>>?o!5{b~9mOyr3@8pn6-`;4W5%07R_8-#xF8{a;%
zJL{Ph0};5Q<G&}llbuayK>5sXFRLvbnp{~4II^*Yp4WQey2)ZlDjW2Dgs1B18v@e=
zL@}0^RO-KfcP(VK>k~K#!@(9edfDYse|NB2S>ft@<b2VVS>w8$@QC6h5P$<M%erbC
znm2ok?KQh+<Xhwd{t@;0RkI#b7t2SwfnkgR|HY<!r2tD~9suyMn}#p7ZYec>h-ahD
z-@ux=gNW<y0pU{<QBs}SBfEpmr&?{zVlTo+rdIm$d0fs9zzsPL4s;Nk;FA%B)F-iN
z3fQF?6OG!lg&nR(GN1nGX`8O!=onSs?s@oF;6*aAO-k$kX#o^%q^Z3cR1bF^&D4Hb
zKsMK)eXMdDsNIUuaz4DHWxuqm`^)|O^t9M;Qvu1o!{PU1o#oQu%pU_OqllU^0L>?i
zZ2-9}4ScnP4>2FWPn~WT-VRBu3(%Rf>=NrjeVC#(bqgJb3Y?O0TpXvIo4=ebA8AT#
z9fKncov?5s#CCyTzGK#+2Bn6nTfuMrMH^@p|5#V%`xv4Clq${fVt0Q(uT&r&uIaY!
ziNCO2LADBNw`4fZK!NNF!lqxxlsdh2Xd7crp~$QrK2yBoe?osWBTteBSTpcnW%6YN
z7Ofm=rT;4!{FGuik^YoWc8l+a2R93~I7c~(5GH*rk7?I#eL0#1EB)6Vp|h~hSG-1t
z!vb3^ZLNb<FGP@tFR=Yfu2;Z1F~2=Sn|3}%gI*W#z-RCjIEt2mLQ1yXRQwjM+`nF1
zd75F?<+8}PHd1wD6i}c0uKBUmwEpzh!>6>NZ{Lo2jhKVSG>c-E3YgYfLP!{NWoBsq
z()wy`MdA0uOqAuvhgjDjfK)~5!9xUj^9Hsobe<vXf{FiVKJ+0xKb!HmM2y<2gDRx`
z1>kZU%i1471}MmGN+VDDl$rCoeV>TaMCzK8P748YdnB5&z?_3Sw<pXR@7&tXU~wXK
z-}GI@gGzl|I2zPaTCm`5xK(#`@vv=DAOa|_Zy+wc@D>Lpn_UQHW5MIPWSg9&ty2OC
z>P=mKm_)sSJ^~UBgP4E6`x;?MoT~k;ba-5GRJT}-@fq%G7>m9CvPKRVaKf1%++5UV
z<bYqcL|op*2G#iA&HVQYO~$^KfOtlf@4YvS=XHp6*e5DJnOL;mob-S>IWCwsG?}Uk
zyqC&QY&*=!Z)c*tcBm~Sl_s5I1{mDK8Q)3u2VRVc47fN$H~I{?p(~e*sdoDG@neZ?
zGd-YDuPR>h0@PJ7vFo(-P+>4z!b<;TlJyiC-*UwD?(nzm-Q$AiDk|RrU@b@cDvghi
zWFeVKu%2x;bwm{u;5?3Wl?L%*_GBCnsXX)9lG9SRw$#h7VIVOf>~Tv%>cR#c&8av8
z;4*(-A@C287^Y9>Rr#gDodJs;c-&&;R9WZN9L_59Kg?%`o82dtsrl#SP6s4%GVjpA
z$oOS=2e-P?_v`SC|AYi8`v2VTJ)oia0xap#)>I$}d1ROGnw-z8m5ygmSlP`l<$;v&
z`)UADW!qu3`5to+eK^4_RJnmtFI=@gQeF*qn2GA_VaYM<uUW0%3eLT-xdY5i+S|dl
zQJY_I|4q-JLqN4V@E~LNC3ew1ryJh3oR`jWpk^&lYV?GBjr~3`9F+WX)UV{hDbq7b
zAdh7_Uozixk&VHJSs;@ww;2vNuNzFHYrJsjEOK{>nzDGu+;(>A_kt%mtQ%|;Y7oF}
zi*7G^_Zx`ky*7_%kwupOCr>Th^uRqf5}5=+zQM#}@<jb_?I)3^ha}>X`WfPDoEahq
zAD(}4M7P(bq48owVc1b?^y)XzUCyt?i!d%+iTk*6=m;b36nA~~>cqwW`<tV*+UTAh
z5NHfCoDML2l!N5L8PXdEEggf_E4E}X#6kD&x^QolC-=;h0w#4Z*~Q^+SE6HJDws@E
zsnfg?1%cj@Tiva=sjsc73o9@D1CSTmZh>^2H9rMVSST|Yab~)ZNdoSFXi#O`bEEb+
znmYTuH$>w9rmi$_RWAMvm%fB##&jl4T;i6}_`>?ZD&V>SX^|~8r+}OAca*%_-P@Bx
z*}x;9X2N+?N!_w&islGE1!+P2=1ugucV#87S^cH$%S7wN`yj8NqB0-3E{|tT;8!_s
zxI<!zL-xe9(WpDHk>^M8g%O^pM&Of%<S)0W%*`QZQDKliIKQ$F?)R>1=X(%u2V-=e
z?(U>FhISZW9PZfa-#;HFvQkgquNGEG)Nl6p09t)aVO}lI5!03D&S>~bROK$`?_%7j
zQDP0%R7@}vfMDQ{kj#NhIEO$C<PYfSNmjTHyC~gwY*wFVS9WK-qSN)m{s;`wI=q~D
zf?NjaxHsQe2A_M&<mp~FimA!hfqYd<E3u`W;WN;~0MNj+9xVvnf`XXgKZ?dS^1YZv
zIC9h(zC^Ob4irmpf&5tpG?UkVWXi|J_^>I3@0PSSrJZkO7=}v<9yDBL;|OaN_L-Dh
zhtt!EFFsI%x%)(_(3^@muUWt&Dj)%}vN*uM6R5{!*Ovvyy0aHJ!~#b4-mV>5pw|)=
z#rxklPXnm30CWRKETl>WHaAkGV(+9}U`D}jV<M|_b!kLE$MGm8BkxzZ9ye;ujT&Zo
z>03%NPVPMjz>j<>E0d0=U2D4@m%ZEH3onuz;NtZ651xNBB%Sgyo_6iz5s-R)Ki|``
z#PX!O{2IZosbzzs_2)q&sdf95$z)ALwavr7^?GRVu>e8-c$sP&(5jn-5g$eWzrFvO
z#Um8~mY>UDycLzuuV2FPzUf9JtL8gelyjp--=MnvuXv_iW))U*3nU=R`(A&SVzHq4
z#mRb~Qm%u;jhpDW_lczx6%jl}CnZLQ{Ge@ygU|mKPUp3H9RIyCScjl!7601r2OJ#O
z0)Q$E5KeMIYp)*-2vE+x$o#qq^ezj<cttW&{&F)<;d<6&x{4^S+ZNA&zFX>+5_%Hq
zAQt_2g>kn>f}6^Tq?J`(yr7NGPke^4lH&*Ud2d5nzjeZmL<}R8XGg34jXw%hJRE^9
zxL97io>Ph67%$F{D&}49A7j2aoA_8&HBvN?_8$)RCmXAf><93Q;u4xwziLt#1<zJe
zop%~Byz$tDrdeWp3#rjqW3FeCADT=TwI`S)2)g(Hpl3LmL0~#jnx+{yQffSb33E)W
z8JWmY&(c!smr4HmjZuNH-k2{PuXqxIRquPlr9Wmj!*)c)+jZsX%WJ=K3lQpk{NnN;
zGAjm{CXgX0yop7$e==yr9xG@20M9Q<pVo`f)6o5wT=l<BEM-S#rs3-nyOH-w`liJ7
zzQoZqv;?@~ij5C$JVzn>g=Tzb#8zI#q`z$BrHLj%>+6w`GR3i0R6Wi6O3xHgBFv&4
z>O^pbmApuzpKJ7PS1*13@XzYsexhTq%bWv~+sdBc99>U*y%*Q=Rm;?QvW4OJ4Nd-X
zvE{E@>|8vj88M}fmxNvFmDo@@{H^-P;)6q`vT<CPIcs5U8kG<h)`2mB-lN05+p43Y
z^bWNW9RZ<XZD3NKhRF>{cOd2X8I2IIT(Pq@7$rr7#((>=3L1ui#r2hcWoRa3Jnnh&
zXhRj*7(6`rL;eC5E5!5)ae=|cQOR!8iHojMG@>W_M3q!$xW`)}Ezx0OO6|qvUAyA-
z0TM1wYT%xP+Mt3kenSA96+sKivoKFK;UwP|c4$fylN)jLOgubyqVhG`W2~YquI32d
zyo;V68Rm*e{2CvxflaZC)f0b?4>%z)=WjG8J2R%;cb(`j@9)wKP&5#{&LhUAuOkl1
zmSRABLiqvpQl;x<^$^OZu4oUCrpyeYG_TH|hkcL#BCWjqLoKA|O|`{cFVW}Tc)TA$
zQ0PSIVTFQGZzy^1UVNNg!)4zK+`v{G#NVQ|W*zngcT#t2XYRWh$<$5xdFy?dpGKL&
zpFytA@1JTBthc7yTAORFv}=NbL|?ytGG6+*IZN-nzMA6m;Tg-DLi^U>gM%Q#+;<$r
z(}YXQfsKqYH0mv)$%+~7V1C~YijKgmq?j1*M9OCN_D_$#o^3bz49h-duAvlm{BnsA
zqC1hHIxUzwVmj{Q*4#u*7-)r3S(1(XR^ouKdS7?^{@nuNkL-~LGSXQC-lNkM+sSpc
zWQpD{V78nrye*{3TiI7u$oEO1Hrv8-peAAnWybqU0>wEBHu{gwY*Nq079!8xyhpEI
z(1b}XHGm_1`%3@K<rW7)n;<Fy@{ld97s?F2m<$_hhjA-?c~s?Q?{4DjrJBb|_clFb
zH*Yitz!KrUi}?4yFZKJfiA}eN_~~`^WwU<EIgURuzpmi8^@ev^+6VBPn?2qnD>Pf2
z<CLFQd&=!DpL+I((Zar-^rrec2FhPwRNs-a+-wfw2(<nER-NzI^)z!>2=IdZk)wPk
z^Xqn0;0i>-G-uny6@5IXj2A2>>X9dP8^p#7Y+V_i#(z6DhS+)PPLVs#PK?cmk2<s5
zbbYZ;yW!i3D0&mrJz=X?-unuf<-))~G<@fd#oli%UcPZzirz}4#FM8RB%E9&4{&cG
zjqJcm6zs*L;R|2Kg@wJxeh{b1hb~rB7{P`<FDs5Xv#iqinVl`FE;;=M3Wl8na`bC7
z*IT#iXozNQxEP6Td-5#Lw_GkL?6=~u*bpZ+t46Q`Jj7`_BpXTi&^;R!y-t|u#jhOp
z4tyO>$SSUUKB9f<imkt~x=I4REfP-0uPcv7lQ?g)A$}@NA;ghE^$_9cjzcB<01SX3
zP|O!T_G@+ZndR##V(ixD*Rb;E!tW)L-~GfB(muE`UiNqmy!Nw03VNlbq@ERM2`M!z
z;#B0v=NGa?<M>NQ3mc>4Yt{ST$%%?ID6b2R3hVa^aP{u$Vuq<bnEAxvA1+lIY#~0(
zTmQq8DoQ2X2G1Os5s64de0zKNo}CyIPx}#q8|8lG81H9PS;C$x1(AcY2KAk{rKr@&
zA)!iry-hWdk4Ey}sH{2tp#AnOd{)Tm9;;TJ51ce9kcQUkN_6W+rS&QvL<{kwvfsdM
z&!+H9=8Y~BLr-;ILfB$@i^_sK8|Vk|;e3*W0ANsx`GRA{_bOCE3W{JT=Nham+`8fx
zRo`oKJ6+s;KT^{QG&Y;_lAQ0e2DZbqGO~)ft}2nn<^6``-G;?%n%OBQ!7B#rZp}E|
z`SFi^gTq-ClIa;4$;o0?EiH8@_cU+xd8ksMdnP|SN^W4k0fC!8(RFDkFP14e+1PMi
ze7#0Vl6Z_ub)CP%%yQyPssA$0Z&kTXOjA=ngZG$95Ph4kK1I9!3K?sig`S3vLm33R
z<Bgu~Zi=34^8dD3GNTd74#BM(6sOL@NHaPdoDf|iY!c2J$f#V1gj0LyaD64Dp{dDk
zIC(3Gf+tt>c0iy)!*5G1`+fSmPPQiE&FHTQ0@%2?u<BVgz2AGDr)p_-Vz$zSN%d8W
zu}#)lQ8}ER;@-NU041;+3hwW0aAQ%#)kKMBx(IpJ(7s7Vv{;vNod;^y)}_sQz2SG-
z;L=Or&kEq)KcQ-!!#(exhPuqk@w2_!IvXt{QoUVGBY6`#pVZ1RLVa<lnp#`wxE?+f
zYk=RA%=;Id#6tpZH5jo*oyYes+Z`MowdJT4)ZuC&fJ(JTW)9-{r?vP&x>sM76~n>y
zU~N;=`ij4Dp_Nf^o9&ZImwi!hZz4r@<Fko52LeL)-x`exx4Z@2-+(=rA}&bTl<OT2
z*VMRqc$!g4KMe;o>M**?GX8bE^+U(oU^($}`ch0nqFtxU%kby>f4G5>d6LjvL%sDQ
zqpfx4z2{EvB^mEpy)6A~1k`_g!m|C7wE~XEo4)necOzx&QQRSG*h@5--(52I4+pWF
zGY9cr?*2nw@|W02y~k@~C7GW=%Z$u{v9USCLt9y2e+cQNkWz3xR{<augc6Y*Jvl7W
z**f^l!1RcZ#M39(yElb^*F{%^uq)w8b+Azz-y=Z3-ArU-!VMw>Vj!E{J3Jg88EF<B
zP5cpJtYnpy2TMpvN&69w?s?f<#Mq}{o(iw)K3nY<nl1}_iMZ`*3U_V(qFx!&L*$xj
z7CIG(z##=%_B#ni<Iw_xSpBof$U8naFHL)_v#YCVZ<y@;SRaWw-nX5ZMloc{m5faB
zoqc$U>1in`dt5o6v3=%r&}iF2#IX<p!Xl%pq`z!8n;@YRM7Ceye7TRWBkl#(S`WB{
zZoW9$4n3t)lln09n3vla3oNRcqt}BiQ-uPxGVHdzXB=8&;VXJ9vuVvah|_9GN<IgZ
z<E1h=rC*UG%toVM(!KdeCnkO<p&7$r`A29T1SlRcRzweBUd71B$R5(zh(2NX-GbD~
zcK8CVkUn-BY<;xAqA|2EyIqH!NlFNTgJR)<y4)7LNjlq;Cp#@M=YPvBzRLZyp0&O6
zHzNpMJ^T=eNW!IP&iB|pS3?T=59NVk{ae?GBl|j63{=R42K~Bk-bT2!w;&UtV10T)
zKqK1g#s8HHd#|kXWgp<`K#nm4TAF8}%+1ZQ>$cnis9pzz<3Ck%2*KUd65_$$ydyN3
zhAhlSteU(bSebX9%Ll9Hd(qW6;TWr{!FX-5<l&mZoB>YKnAYDSCy}Eqk)UcDX2puH
zH^B!FOSr82z0xnj7F`Y+Ey+1}aN;$7ErC%=w{E+5`9*~pyqU(l8*KXi?F6|EvvH~!
z1y(fiC}R|FS8s27^zA{*69WV+>%KY{CuG9s>E5Ejc<HKIa{`Dt;V0s~c=3WEGc(f(
zjTifYr>X?HrxFSwO0@b3m4`ddv`zi{6@|+jlP_QiS+w1t#QdfP6>01zd+qjgWxVUQ
zGDC=1MbGBPtt5z0I-(Nvxf!U|RQ!^H`EBw2yLSofjK2XACoeBA4OV#^?iI+bf{iXl
zD2^DhsG4;4c=inPRMNozgDJ4t7biS9nVLnTG8BSB0<M5^B}XQ+-n$2K+yGD$JUVI>
zOUd>6?D7g<>zjYky{anSa9L#r)MunL_YD2K7B+bjQ<}v-eC6WWC77yq^cl(1kkgaA
zn?Axa^}FS--wjmZ!;KODyQij^K{)-dS|d)L#K-ape0$cwfPh>YG042&HX34tctIlR
zX0Yq~JvF7-HNX2A(9&Ks1GMs~m`0uNI5voTh@U$T@i149y0I3wd<hPv{oS5-5_Z#>
zvdH6M-jrBXef^p~<lj_hg1)6BIfcl9z+3g<)C~EARrt%a=HMhG4Iw1PZ*26S%<w6_
zpSUhFyMH<a>Ew>EPGZ>NK?6d{u7h~1u09CnZS^NYkTC>8IYuEN@Rr&CkzU{P=s1vX
z?d`3%!#QH9J~+wgfkfY=7}MVA^BA5l@qCXtk=a7?Sg5=OO>`89HC_;w8TbGCo0+a9
zU_PP(UlDXcl8E2ncn`w{etrgEz0Ym4rV^vIL?!I}1FlOz<5cUOJ@`Gde;S8)Q<D8h
zhy6p^R3_9?yQhNV<%X&exSHmvSJVa4deRSO@RNEUN4ELb2c`YaJnkge7CAN_07|Um
z?{7cwibUF{rZIvsr5LKm0}OYj95ZYvU>E}l5ls>__h2|Nf-yyA4s5JGn3%|cSeeDf
zCRZUz7|JXO!m{p10(|8fZ2ex*?AfMflIv<L$%JOzKh;85tvBa7WXQ-b_}$Q}xJ~yY
z&O~Y)B!L-WBIe$?5;2(htQzP=bgv&mTYry?L=v*R6Nj)l&icR2O--=&4b#9-u^=n!
z^c^%gSniZL)vzRGmQrV!4V6UbaCZ!M8tT4><DJWktDVg2kaGU^v|n432!D-`Qp+S|
zi55usSMU0omAXQ|wcx|22|uF|?qmqNP=>4{Rdfh_htV!;e^nlIv8zHZ?VxMSCB&1C
z&-qvgi0I)k9p~1lw!RIiI?rJyK(_V@CSV}Xlk8}7(hHQc9r3>b|0*?_MoSQ=Zq7EN
z;I=FcA9D*7ToJI~-7q0b@vsrnsIx^Qx~}p0xk3yT=0gYvL_EN|&0#a}CSK6x9uB1-
zGv@hRD;Co08}db3L0SoMD?u=m^FzDd^`};C^^hT?#6f?<4>n(-kP;U0Gb`&dJU+gL
zvOyZrkt@SR9Q11CZUfb`uww_x5DYXI&A($$z9i<~Jvg*Z&da;aA}?=XXecVHjCSVe
zSol*@R{s;!5I|hVJ+CNoywumIzXvX-&hQS=5OT((dTVKx#KI<_a+2I|``}ufkzZVd
z%pOghgUu0!4>m~vydL-FtQzVG@cNovSb)Ugp$^1I$y(cgZRUL%)IzX@eyO6Ws+!25
zhn}I3bfsFdoAep{jIk;Bkx5Q%u(Rq5|Eb{th}N9tK40lL>(zBTLW*`}Y8s!Pf4I=9
zbKE@EFsk6+vuTf;BVHi-3ydis@`rV0djq5tSl40<J148MCKKhgD|SV1ux)H?z_1yC
z$k(pF3jsC21UEpKFwC)tgOWKJ(lcxvp)9dDJ21R}yl!hSruBzvIvoBV8#zA?czlI7
zEbh8X_@j~E#mLUiR#a8>E-uCcg`1Qpu%lnfBk7SBoO~*}!B7LK&H9j}jToArxTv3(
zsGoQfTaZpmE5n0130h<im&$O}kl>U&^SMguJPpTmsE_oY)_i|~LP$Y1*6DX>AuF9k
z?%Qn(H6)xT+wbgfK#&o;{SNNIN<W$)<#yc1E#$ogRdWy=pwu}=$la%i%Koi+Htp68
z)V1k)ky@AY*V&u*VJmXO;)I`-0s}`<wRiNIeeS3XwnS_BOg$bg|9*R3Qqk8o?-AQ(
z_fwIm(5NGy8_uppR`{aH?oi6UF>&frKoc|on<niAZyz6YI%ty4)nC56d-KK(#d_CE
zkhBFN1Lhz;F$Jc{V#<F38PN}Rzgb{9hX)msv#IH(FG3K-pvb6Khh13A9W{!O?_`5H
zx|B*d<V9>OX&N<Nn0qe+WJ16g6qtJ(8u&?RYlpyM@#|%)+!{0;^G<@(VMU{%5$}BB
zMD2PYk%G+dz%O~I)|lb5(cH5P9(gHE%><)i&WAW5b>Q&Ngo%xv29G6M#=yuZ042KE
zs7>cLPWfKz38mp}onKP4#igaCNVh{Bb8|M;28gc)#F7}Wvv%0D$*HLiRV!ggZGI)J
zxzhejv8wzw^h6{WcPZP0J2M~L$IKR|KO))Tf|rOIB0SaKlt-cw`@{p93)|qTh=(r|
zgdy#f^#1uTfB)Oyv#W@_iRWJ`D6~_#U7*0CySO}_hSL)t{Vw?}J3G6Wi_rAU%z1F-
zlX$EPZ&n%%QDhKsa}{Sa!1E&Mtwdu-(Yf`E%AM7H91-u1It}qIMiS~doi_M)F=)bW
z!KI_9q~t*wh4qo(qlKz!<KpM!jJBImoj0R$0)~dTk!ub}Z?OIK%*i3kRx7v*j`Cmt
z+5}M)s2)FLJ{dtSt_1`I4I*4oaGSG9kUhzKdlR{Pfg<BZ!}2~jH3SN~oJkR|!=FJ0
zRYT|~pryW7m+AXh;HS?Hd2U`llg(q{7|@W*FNr7dTcPMNIL`mmc1x*FzPNb3w7mQf
zhS4F|H0dHPsHVo5m~8l&JpRLIVlt2R#|~Ial=Whj7OPBfg|qD*=Zu=`4ji1?Uz{Yd
zxf=KVK!Ie718>I1{!oX`z}doUzw@A}sR<%s&9RBu5#YYhned+RzAhlo;Wsca0;@E(
zPGR+^BFu_28g<y~nm!<mQJz7#034wR-{$A`_Pj8fy1OyPCFpuV3^r5H^MQix36xTy
zK(RC9B`YH6=sqIPS$LCQz_R!T>hpeIzfd>{fo6o(2pNYgnp`OLA)>ukS9PW@S&W#+
z=sw)KD|*`NI5WiVIJrjubq-g*ZXuY@4m6@*T6$<GO@RFpJ{8&7fpcp!SPd$sO5Vw8
zT3lL!q%IFB=`(kHUstx#fZ>vzTa{(20U}>dle3ghDRh>Wmgd$D)p;G($XT`Th&TN@
z{@v5#0mn^gS(&1|I@WNup_sZX78BDSKbK4N0$r_#^hZ2is3=CU#6*4^_^Bz|G|9N(
z8QsrM=4aS$D(m7+3LLk#NGr2L4i}rNIhde>w#*v>-a8J+A&mpPR@pBuFL$@51P->R
zsUUYM)N!#Gra<k%{<U#*RP6%N;ts$(e;Pt2O%Yl}OJF)D{2s?W*?aCWHFa`JARMq?
z`AR;`jeF0-2?wAVgtdsnI3f5fvOAn<43(F*rAP$2{_F~uqD3aRLn9BTC>&B?;=G5V
zal+Vgg_!V$4M2ST$Pvw+4fobhs2*M=t<%szWB|ljA!B$`;fnZE+AGAv6Y;LXOyctF
zh`C~Q?)wRmN<lu|=HHr($yppI3SC_sL(3xx%S^UJk(I`$kgE*%H?P|9uh+Dz$*ldq
zQk^A?<ZFm4(+4F(yW4j)U4a{tsl}W>IW$^o+0Dtx&HVsmG6hQtkejy(B@F?;EG`7o
z?JxH-fSYl)dIc*0dhW2MK7NImYcI(VN_m<vMs0O<0qh1HXI+%wMxBek$@X@uw-Bz8
zencl5#hffX|03#d$kE0>Z&Xy&l^3+^?4htVUxK55I*f4PCG|4gxf9&9Z$^SRLR-1<
z%v<#-9PlTuMF<PA2e_dzF(i-z`RAY*@`jPJz%Yn#ueIM5%_zKm`}R;}k<cZlKCeI5
z+#kAl+L&E9C^1L{jH0{Vw~|<vnl0)MWM#-}@|UG*s*gsj-LIlaqQ84&efbI@%{jli
zb48&tQA1c}K8+mpl<j^9qFH{hV4T$2vfKWb{zy4PF_@GG8DCuLcxV(5BNspOe_DXx
z;D|?GA@+>lVUHGa1SOG0_D$4ALiX$T*li(xqzEU(L&O46Kmmtd+xKx_zgAtNf0Mzi
ziNbpl+I$AjwD*pWGpPt5%BQ%n@CO9DfsH~27=0?k-GV&e#c7n1ab*#OBsV{8M|<ic
zpT|o@j6!i$RUU;0xP9jJmy*!qan{n>ZvT^abaX7$ZN&r#kTbBH--g+6wuXcNj|Xen
zAv`XSYcZoP^3fzzwBN#0h008)Y9?P3d^`;z;rreeyhrQXX1lPm@(kdp2&N>-AmZoH
z0!E!<!3@SAT)Kt}zr&!;6oYA?H1I0z2LR+lz%p2zayuRv%<Z36p?WkB-Ne`?!ESTI
zXF&*|oS@p@dVrJk_MwWJS|;4skz;n^l9I1#t8Ex}P3?>7J}>lo8p4$E!Nb7{^JzBQ
zFfcY7+?x6YeT%=nAm#40xx~}k?$hm`-wWJ>9xF9qfTTek8Y~E?OaZ_kMhhES<RJ*i
z16%J@=$bn(bp0WAm5WPxyv9|+`$qTLB31{WZmsoaYHRF#kBYe6j<^R#2h-v8tHySm
zQpAmhdel);@O}1zgs340m_U^PtG`z(fx-+7zAqE_C7@k_d_k7Afpo0X^YhLR0jF?T
zasz2Lw`*VJEbPXG;2P;74%(*VLF6HrQ7p*ZPppbSi6TXd7bZnZ16Ml)(L$u9P}W&6
zG+=!r!H)wEBLIX>fxmOzH)lN4;8AKknqP1QWiuHh#|n*2OoUfg^WXMvYTElP?$O&u
zjVOJ4_>rAB$Xr1lL39HZWf;gxR;?Pwec4%v;rw7=AZ4}Gl?<ydrg?d3=^F$;u|%(H
zQgd)YZ&X~KEnCD!DMLXI9b}o=sw8spS14VpWY-s*yJ^!K8$XWOH*WlL6R3_08K+ff
zng?MQ{)DtVsAiBqKNDIQrI#<is?3`ItC1F@41ub3AJ(=bV)#vo9(x@bFJc2Bhhech
zuXBY}l%AaYT!S?V#!AJJ$)cVZIIp0G06W3rva<B9F1f6;K){P&w$v0J19UejA3Gy2
z3!B!V>PUmxCy=(k(gYa9X}s3yX_-O)m!aWd5ut~X7&Dxha-(EL+1+q1Mqm*&kP{F@
zXb2}Mz)6$|iNG;hw?GsC4E=*_<;=SvS&cex<un?SS{uwEgRR}^(q??|cXmO?T)j)$
z@@-OX8nMwtHQ$|`6^YYT^T$(TpD%LUq!kp@mM-4aIrBjRj$8Qi23Xz(P<{`=;3Fiq
zBtzFR0GmK|N_TB-O$Ihb=E{lzDdz{jh54AmGK6?838(OpL<XcOlp^913NUY>V55Of
z-0Axa`CG>O-=N-$iH#kAl+zX9De3ZqGSkz;f`fzQF6CT|AODecQXW!c-HVY1U(n0<
zt!hx27(n5&h#40dDGA>sgH^K{H>CX`;0|DzaO2sRFA=iaA>oDpRlg7NP{grrpc;)8
z)73hkzFN9_30AYn_7h^LLmeecXh9#@F_J<3(b<nR%DqF@Nz4=DC_S=UlaXgOU9Krd
zb0TsG>Th$J8pg*SS9fY9=-b*jm&YK>&JbG5Y3FCS`1oS5F=!yhGg~PYT|S1>8+@nL
zH)3IAJS;2Aki(LYghUz)M#FF2_>valY1$Uz&523A-GjQrP^Td+jWUDw%=4pax*m?s
zV|SJB7G?l3g+h7ZU~eB<Uth2Cm;)}D@VC5UW8;eYb2+*b9kn$zu2Qo$M;mXm!DG5>
zxj>c%OjD8UKlCiLVN&9;v9SQMXfU>zCF@AYCGnuasTfK%j1s@Ff0~w=sm3~33kG!D
zwi|D<-<Hs8);U*+vV8%!7S3#?1i)1Ou;`SVX-Fft1FVLu=7r4q^sk!}iql$C0p0np
zkCsDH&wo1Fh{f!i0-B&|_Fxa{m%ZKHWI&}up<U`u5=nxYgy*0Lxm>eNhPS3vW;DF#
zB|7{1KC#pL3%h=FcG8B`kh!(ai2EMyrQAGH%3_}4H<wlC8uS+)qfxsnA-0mehFM?{
z!&#aRCh>!?m_KfjK|t`c>#Oqux%cE0q(pIY6E&_kgDC`-eDSqlo=+w7B}Mh;#em8A
z9>{eDa%wP|L+>Y~CZ)m7gM!!#6)PeNVUODPq1zKmz&|2`8|k^Z@vZZ_NR;4vSeO!P
z(TDf%pTmr9Iy85f>LVpZhT=9XEKIaIaA9&{F`5_b4&|G6nlsgGO;$^xpE=Ur$6SH0
z$GJEa%VUg8(N?u>uz({imfwbn0!A>cZEPM<Q_tp%LiW-tuyvpnP2OQ~Kg-R!%A84Q
z7$3il;7h{UICZAnwO>n9mX8xWv(Bs)c-kMa-~Nd91QW^?A8VWJuxAYlAGvlh#RwXP
zdRq$v0x)+^oxy+m-s$P7((Bi+tIofioSbl8Iz@;pf6CM)#HlcSCy^={d@<WH>UA3*
zqJPC^kKxQkwyt{z2goG?II_1oz7^sIVg*B=Ck!4(S_X;xvS@N>d_ux1czG<wy2L^M
zO+rdK&=JnCI51Z5w#0{=Y(j}^nLdKa5va!$fS^{J<34@%SNKMv05{0@Bi3)ifXtZF
zPz~lRC#Rn=IuV_J4+=Kwi^0b*`2E8bS)s-3LAuUc)w!K2ahgYE!c#R4;gD3$s;c4Y
z>e{lrU!tfZ5xnc7=ldnk*gpx2_0FApw{N1ZO*ST!pXX=zG1$pWk!}o>-iGI8{x<SY
z!d*n-T!|_xTzs(F(is#If2W81t+afY_ohg6$pW@fEAz+}JVm&$uf%S%nVXw`wzbWG
zMnI<49Maelz=s3U6J(*qvhwZ0Kv-Y$7D3fY9Ms!)bEbEGS4v$y4vfLy<bUOb05Z}i
zCQ=|_VYay`MhN-1$c7ZwfjB?^_-B04fs@^V!Qte-3AIj&_Rd15Qs)lm&(?(x0mzs`
zv@Cv>W<+l5C#p}mABdp|;rZVf%22feD+;CrEZ#RJgcVj8cCkDzcuKEP6RziR`2++y
zzPpL}U{i~T)H4;)Nz&rE*brDlm(FcCK$|s>g^dlb@q12gu3$9J^fsO`qFvF#HO)Yg
zZDwCuQ$84s$fZ$U9Ca+Xv2AM74(Aw4Lx`{M+oL}>u@IR%V{hTX-hoB0?ESh*rm3St
zTEBu${J-$^jnOKJL!ITwGH&OSUXh@kSOGg$9>}|H5|Cq(fdLl^5!ZNr+l`Ydv1L01
zA)-%D9+(z(1RX*TV#Wp>1Sw%ZmTPc*f~8sdp#u*J^6)==8O%9!P#OII3XzbnS7U)o
zcDCv9qb<I-@1nz}x4JofgI#t*UDX9cfC32h@SqH)<oFI1;Hx#hlFAhZYv5k@68SwQ
zHRAnymz8VzBZs=i!T?iY8sCcsXaWM#8JU=5)&xxkGr$7YHcti{Z*ppCHALi!arY@$
zGL4sf3X$7|z_BpM3S$jvk86(Iv*Y<O9ing{1Y-<yyF(r0*{r7j3g@sgtBRdz5+|vJ
zLl~Yn29Ty+9JDh`h(J86U+?lDloXF1J(AvXX@(R*!;_u8yw_Egm5SZ5^rf!DcqzFz
z#%oQ5931bS_Xg%e&4GlodElniUp@+;hj!%R;^HUBQILQvs)sw`1KMAdh5#NmbkE(?
z0pbP`*n7xK``KpS#V3g~3k%{9DV_}FW=9zs*a9=i%Lmhx?Dbw($)BB7Td#I0dEYL$
zIH(x<S4-gF6dc93VLLti;YYy>d^Zwl54S6&HQS~E;i{~HRkJ?2*e4F;-3feUf6eD5
zujM+UuIX(k-@SSBy;C%u<_+M2YHAh8j|PB-vz>&9=mnfs4uE7O=1Qv>g3wIdWBW%+
z#C^(;7v1p`NF@BTadV)HewZSopYB^NGaommuim26z%na={7m%B8!mr1os_ERuXOeF
zI{N#=Qc_Z&HH~z%UKdw?lAo63N_vn)s09%VaLkGW-v7#vIFbRF>da}^^I$-~F)Qd~
z5b<bzK0M&(v77qircNf@Y2spHo?s^>0Ysrj&MU~Se*mG|<CT_rRdb=}EG+UeUjrFQ
z-`t$8eCC6GW_@<z>g~r?;IEfu^};U3OL-&EcL=<hVN0e$vll7L|M`^`zMAS`V10o`
zwR>)EnpJsxT%4@4a1!LW<+RB1^72N*)cVdai~pi!wPq!cs<JW{(KFeRnjCT&#1cU2
zpU}BCOvwfS|M_4<w5^*o(8kmG(=>;0f$xKk|E!g=c6ePPE6Pe!v9><_Y<S)ec<E4V
z5R5QDq5mx}k43ZIwN_&mYCZ)4+YQFG7vLl!4dpEiG@mEk*MyE6*{eXGnFc7mtD0eN
zXXiseil8B+^cPPNHTDldV50~S0d9KLj13-+kDvhcxkD!D@9)33CazuUC;<xy<O>Wy
zLd1v)n0)%ga3-+hT+}%1IWhV(!~fpl@rv~9XFP%L5zp$)r=K0Jy?Z5dftA%v4lKs%
z+FEFD{Oad2zvzhjVj8R!e;pz@QU#^8wUg`x6ICYU0BZqs*->?AF)k@d0p5=G$@%R&
zcj%m)&M{jl?%x*&1_0}s&`4Jfb{it!jj0+4nc<3jS+_(XdS!Lq8aE7^)uFGYrQEBT
z9xf|TQ5b-B<%5xt9DZByMat%7oeP@3c+-v7H#j*eDxWP40=7UfLIpyL(6TZfxHU>9
zC$@5PbL$G*oKE*XLWgzD_rH<|9uFukip$G05(MlbJMk=4XF!8Gl&>iO{|L@uEKP|n
zRK;K(F<4pJ$tdpU@LG@tKXzM_kL)t5KJ`)}L4?*oBDFsO0h!C}u%})8(TXEF36Uog
zOZ@Wx<LSHOv2Ne6Z%TtwnWZ8^wo=H<sAQ$=WTv4MG9qrPl8h)LNk&GpNwQZcLPjKe
z@9e$b<9>ec`~LNOdW!r0Uf=6Fuk$>P<2+6SlMEXiKcV`TnCTy->YieDHU9USP6my?
zNl=SaQ+=siDk@M9Mj0SzV4K~T$`65c#a{xcw#(DKLURJgj=5pFymp*no<W{1xKnES
zwRV(s)y`f08bDVRkL@QCj@?k*4D0t*<2#u6=@ZcV*yB4NDRu66k{TWt>A8pf!2^D4
z%l3f9iP^X>_OB~aNtBd<M0N~AYAu(*s$#}O+4qaWy9lQ%*^Wv{GaAl*)?O2KSsgvm
zhI#I}Wh|)ZIMpKb`z$eFg6{C)waEPjuIn>9#m=661&*5(-*Es&0X;V5+T2OhPGM4g
zds5bd;RJ6Ow(4&TRX`|YjG(F}Pao|s--}F&n7vG>9G-Fbf{FJitp}xDn`0FA-2+Ak
zY*<@wxPZ7~{Z4vtd^qhJ%T5hd+89cSlYdZYMw#D<qo!RRW<hzQytGx6{AvPuHcQ|;
zrVt1{Muszr>)p@&s$aQYYP+d@@@2%?P?yb>I|;KEV@(NKW4H3*y1^v&IcHzQZd%UD
z2UN9gWZT%u-HmseFLnpD5oo5pD9S!wrk-CNJf6VEFxmAr0=LVRqty^yC8&k=BBT3a
z)wKtAHfy(Cwa?c*EMn<_=h;kzc>RP9kiRHBF+7dBB6hlnq@DgZF>x2!oh?`hh>8Na
ziOBei^5&x9@@EShw<&EaMRpp}uy+9K=*T%5F8TO#dCCSN+!^qiq%#Z<IS^J?g98IZ
z`?|2VZ!e*Y<hSnGgGLPbKW6|7EM1ZyN}dEhzi4gm3pci7^78Vti;ExOK%z7eq=teU
z3Av!%^ey8kg@;$S-Y5?&PjxCG2eq9_JBM|oA$owZ8EvZwq&#34P~*Tft8wmLSEAkP
zp4Iux=DJ8<-<Q=j`M6zl%x+&hAn7pecCRJX;Gbn`YAWfma?7^>>^0WcS8g*dd$%Pj
zyHn<@v{<|F>~QIYi#F05&!l!mul}(kWifYEO@%}fW%cx!nzHRhSLgm^(3AxYGDqiL
z*SL4zQr_OefL-yn<h4H{CkVyqH@JhZEtdzaqqzHd{|<IgNHsfhJ!w*Q4B3l{-B{BH
z?El|bpHa9~#XWbL9oU_lpYOV`gfsYid8WTu;y^%XsN&}4ChnBu_d0TBbW+ORe4&`b
z{K%#}JM+etmJ`V99Pu9Pm*9_@J1{(~z`($OY*K?@mN`@fv^Tr8h6{dk)(;<Y>nOZM
zGZ(7wkKmE2A6V%}JlLI8YaT2<l6bZ370r!rHyOkvMAAMmHSZmrDLkz1sp#hu^^WtR
zZQnF=?(P&)KC(4cZR@Z_y?U$0O6YWoBB!wG$aJd@hj-ds_7lbKk|>oEe@Ux~;)q8y
zCMhXN8DknRY_$OEMZC-HM5cNaCCLFYx3WBUrD0;Pj3F-fj`lC>i8D-ftlLJiE9SU(
zxt|*r6VIPIu#aoQS<|DvsmP#6v-3(M=S!9gC-zOTzP|MKr+1ttGvk}kOY8wZ$BcnW
zkDw~Go??N|>Urfie5{8LA0C4T7*2FSdGLE|=K8ea&DCZH!y=+0CPNCPeRQTG9d(R=
zKZ7A$jeq`p1pU#7<iGl8cjlcsl6e~nm;`J|+u4CC9#dOICV;Ud3K@5l0!sad1cziV
zAz(Zns>^06N({wUJl0o+BX8cgA!qvC_rv`ByyfZ*z(Ga}fx+*l1J0ah3*B%Kwtt{x
zH?CWGKiq0sX5YJ$77n>PpFeC+8v)38#$vusv%Ae3jE`2*M^#{ou>i#D@mvXBfwcU3
zS&K>E#@ADf$LizEL$5F30}=;vJ&_bNfRgok!6hmAo{yh@P+%JAOa(@0$6OP=l3><D
zt1%0ygvNKz5p{i@f)DS5bUl)pPyD!WrEW6(Q(PETXy_kyLYeMN(513(v+CN~AFsw-
z3~ns~QB&F2sBXewK?MYrRZvh6ov1t_Uh@N0v;b{5q@)VfYr91sv;`p_)VX!xyOHb0
z1z?jWz@=cTv65`4bk=XOVr~yQFO(51IF27bu9t2s8+>l3{p^6Ekf>-pE>Scv_*3rn
zmE0Ni?Hy15^%5Rtd4>*xFYT`F2hX`kGD|wR0cc|7<lJG|T|}bW@Yqp+vS1|Fl&D4J
z=jR8Xk;<ByXW$cn8>ib0U$z~o?+ufFfzjS)&YZt=S<(GN-X>%5otw9x2)=r7Lq%DG
zHnZ?VS)nQw7xLMolMW?^Itp35UlpfSt{?fa?N`a22Fik$KAw;g>V3WM@!FI#%Azx`
zXU03;budxulR^CDgqhZ}XZt!<=I{-PUd6@EYYZr9aHD9YTE7bm>mRHRA-I7<``cR8
z!Xf~_7Tm$Xs2g(NX6wB6qm)B_=8PawXZLnu&3lmp#LbrWkBNzWhUS2R&bmz7*KFB|
zeLHR_bF9a*wz9ILdh8qH7ru42{ku-$$%M9ny4{QVmfSPuug5J32?@+EH0dk%!i=O6
zXk<?T>q0$jJoq#C@0Yu@5z@}ykdz!gTML0qSE18~<YZr@2GKZ%TAZ0%8%xMs*7|!&
zpF#lvN4jGe3~X}yHs_L4LQ;}x<ttXwEO%yU;tTK!(eNL`bT^9czK2MeD0ji0I1aoP
zNcYZT66sf<<020z_1!u#bRNTG835Ta;()b(9{OqA=W|u->dtg!MMbZe*w|;lGx||i
zpA;5`>4|Vdx`{l1nSNl34x8&@_y#l6z4GjwoY5fcF?d@Bv*MV~x$Jo`kyzZR7ynMA
zt5|LOfvS17yHj0CAJ+;VANGW|O5gm<M$b%B9QD%3Oz~SuXXXHP^$eFXPLQ98nx#vC
z4hMz@W%J+R<+VwOxKE7*<Ags=OdMk@#yF+@U%!6U`dov5>rq*Gi6BF==4?sjUvkRH
z%aha5(Fux(=wc{|Ouqh!&inmGC3c!G4?5SwIG@8R`*od4Q+BH7>;8|QV3cjlW__bV
zb5GJY!O+l9e0i>gzSvmy_u_Vx5LBgp*^XSX4+Ao-Y;DhjbR$Fo;(^!=R;@s(%YOXW
z31F@Se-C2`GIyD2RgVf}Nj+0j{`e&%W{-@F%<ONvNnBhhNP1|jN)a3$_LPv-yceVG
z|NgQJTww3$=x&@%%K*{y&j7)jb`?CvIDu~v;o;ru&*w)(Fym+&o0|US1f_S@%^<l-
z=<V_i%?mCR##Q|@u*&0EC=p>}Bgbc<=G%!X=HbGa);#X1pXZVj6M2|~)4zWWcRF=@
z;G{?A9Y_ifj*1<mUToa`j-;2(H$HSZyzqqwm1_k$i1?a$+Ht<G<X`R-O8Z^k^dtie
z@YAQgLqm%dQy?E-t$7C&U~;PhmdgE;^V|-*pSpv*f)JT$08{!CjVpt*h{yAfhY^4>
zlN<WN+6HS&6J&FG!VMqgr#MFs9g-$6e2!jKCj!%RujcP;nJ98zt5484F)}u$eUA--
zrc<|T4x($W4@+~Q@7+kXM-3p_804P7$99l!+je08em9ueY29<ArKNrN@9(K2XRQdB
zi$3x@exHa%$1X<35X-zaU|pv`X81BoF_o5<lI9yC)HWbT^gVR?W4XOo^b23#^6B2P
z*MWgg4~tq5GoGv@EO%wz*@tVs0I_m|i}2`q;MKca8GOc<WTgm(lFa+?!UFGJDE;=0
z9M3Bc9Vft+T`Vjtwn*{JoOHFmTV@Mtr*!VAYoL8SJz3b%f6lRwj?M7A$z%<;x&g}u
z4}<D;+b1&eJHZ2%y*O|@wySG-q~ZA6KSCj6dST!Fa$siYRjg$<qoteHRpjgg@SSTw
z8nB8iW?5__9RDNCbwrBpJqtM);2(<*5;gmd{CkjbW0W}UFm{?u^G?M=-12H$&yoiE
z%$_&I6xUw;In)26pR|zV8E0k&|NYxTIC#cvKZx>}=`VxB!~KgB?YbE^C=lvkxJ89k
z|JL4q81z6X5}e+#F}+(~SIFElGf(3up5mOL$}|RbgkgcwSm_s#8+vqxyhukTjS;aQ
zFi|q+@!aiDyEK5wX6;{7c>I5u(%vOCFKGFB5Q2b9oP@2WwL)X^wbRd@KmS%+yBF8N
zS-4jQyO^0L<vBPkiSJq1TVMY{jDi39Q~RlNl^m0?12<LPiSLxiPUg4KK1hg=cJdo~
zkZ;mvYniN<MalO1{-%q(oV*EVoY_~K`%mT=1%j5_A=gVvUS73r@5L-~I?kf=YzKgw
zB_&5k3q{!N^=7H=jj>bQ`+660)FeH@D;ykV=6A^mj5WLMS=$JEC_PlJU8lw>>?W9X
zt+(GlF=(BuzJOg+YGrNB!}JAse`#6SHn`?lD=S-A2;g+4+m5P#vFv6R7Z*P$<#@)h
z*ySZy1~x?7ozzEUpgj2yt)3++aM#@26p6XR>O40@($BD9r8PBsj$iaG0jD<Ln1)t3
z?4{iben(0X2~Itpy_s-=6Q$m6Vc{%J!j8`$tN)A4=@j)Lb;G9Eb@b9{afylE*D`PS
z&JI=|6tiZ>m6|wTMzx*Dcm2%HrL8>Qzb$E3ySwNYRohFP7Ueot$1i)Gj%_^o;5+&J
zSi>PeB-I5XLSyuLb|W6$Z_-LoG5xeF22b(fjYphj%OA_XyL#&)Sh=|q{&t^+dU18Z
zya{)U(wpF~Uqt|-y?Xoh)0G(t0y8u7eoBw0x=T@3e3ZqPE|E)wJqOacA?5hZrjo$n
z?&cay&*-M6T?pO_`&x6X0@J&n#mC<XRwOF&8Ud+tm>b$llJ?K*i&}KJrKSo1k$M6g
zt_K90diQY%)zu;-OObtK7?+aIPqZ`s`SYj#&!6WQ8i8@2q*k%>3JU!|g>*r_K^DBA
zTK87#8JOi>9COl=*VNWF#H+E8HZf`mq#(&xI^xwLFXC8ZuX~r2$RI&c{y2@7_!mM;
zFaVbcKa*F4c(y$n^0~_3L$D%I{}3YOUbu%b(-4P7Au6G(K^7vSL|aZMLQ1}>CS}|z
zQ|h34>^8@kcB6yclHKLfH??hReJv|p>a`K%C1@qcAx`Yr%yrGIF+Y7evp%!?WX;=C
zv`{E;>9kecIAwVLeALH;P}}e)2;FUMy|^(D+SD4lsb1<<-l)oGtgiZd5aY96`1r^#
zjCr?H4E6Or0V5aV?`$SL&PjK%!LSvs6-{Rn-bqM_CL4O69u_!s>*h^rvh=B)f|=2U
zhF}*kk(_6tZ*6JHw5KGoaKvi(7qZ!o9Nq38gVq^1#PEhD184p12?Z4tdYE5Lh4_q%
zl@U-p24};wmj^=p*KgmxzzX&ue<I}uNa~@da&~ZVh{CMZPe=oUjT#bn?bxvcy_6yV
zjJ&XB#*p7kLr-7gz73K7E!#uhy|@ZU|0kGN2d>rx1%)5h@wj{qbqgJZr>2UxA{5S3
z>1J17Idgx|=PJ`LLLSs%H4#$Rhf*wVk#Z4I5>L7@u`PKvdcptJ$jK=x?p8hWulMvF
zCg#pUziLLm-qyKclGYpwmpG@_-BcBC_VOO136}B^BmTR2=5t2HKviJ%#`|IAr%+uV
zBPgDi)x>s_wTof8!d>dv+hW%Amu$F>vD@+BfdgbVHgh}%CSpgZ)z#IZruy*l<Hz(h
z=ng+ze93f#u$UZnh3kUTEjznJt!uq#8ulGj9+H=5*Q@YO&WS5){M8BBfaK#@Qg4J-
z;=UE);5g#&h3nMkNQSF_aR(vzQ$a&N|F4x6oBWEpI*)ZNl6(@Hz8zVYCrWc{`I-4q
z4U$_u7L`wLWmS+Wt*$!;jQTg!L<g9dC_#7U<mbl##Kkx`h0B*OC-nLsmDpi!ZjR3f
zHl=yVf&R1gpc0tAL&ix)zDjN>#8XP9yRW2c!O+BuGS?m86)Ex~kL<;{!ubxKH3oGB
zJKLwJm$N7<G}8Py{8pyE*O_9#`t-wHXVn*Dmmsi>^>*^_W__k`o|S7PjWVvGyPM@@
z*x72j@863~{oylQ?(`BwkJ-{xH?M8A2`f^nLG<+`F&^kNG4SZC2noGb%M*SDI^|xm
zd(qSc1HzPRj}R}75puCOi(smK`xH>r4N7O|TvehmLaZ0mxN=rPg7MIx@g)W1rx|7q
zY{;mO9Xm!qgeAS$Q}_R#L?)2;);$dqWPMrAG2{GTQ9M!Zb~HBNI%-NYGTK00l<8%5
zfcMaIxf9298BP|-nG;ed$jOLAVv*HfUNyk<4Ikups)P!~H8Kl`i!*V4zi1t*sH3fY
z(e(OZ&EEl$Tc1hSRj4#03Q`R4aLnxoku}|3%#zYpJ=D~e{6-8Zy(iXwN{Q%&nRRqV
z^^J{VKl3O(iXv+$w=}vF{k~aVIuD0iexn<~LB?iLf@H?)PeZQ1UgoLpECUV@yLkBE
zp+jEaj<_UeIId9_*bGzCYEfRiaDf7oN=tDP575zVhfeg)ojdyZkM_#AOOOLqMOICT
z$qv@^h8{%yjdy#$u2$eGnR)qE`>ewj&JVpyCaISYbq^-usl-X@-@M6wFEsQghX7ub
zfz!UI>mk(E%H7G2m3#Weo<lif0!|y6s6p>xJW&tTcJ;7OK_)jY?$W9XdH}>SkXP>6
zyB8=2mB(3uHzz!Jjl3nNv|b>WK#gIZnJV|N-<Xi-vF*m=td_JH9!>^}2YEM^ik{NH
z^xoPQ651C3Bp06&XUOOCdl98*FDf4IG~37r6288^&wxAQmtXTqW=B$&0Sym^P?sQ^
zz+8*BV1RHTpTsQVPe0Q=3>i8)I<A0fwIdwY<vvNun?-ID+Rz0T11V5FYQE*YXj%Gd
z(TOR&Y1UMGTPs4m^!vo^dxe&X(x3piLY6JM_m;L&g*=?@kp<zu6NLT~^tzzXJ!Yft
z&n)%imO+wrT#U|^yKd&~y{Nfgy?<Y$T_ha?EgDLee)4)_gXcHAX=RwMq_Zuso>Ke}
zU3e>>d6{=vz@m#%K<9`nMR$c~dF)Ms>sHCFj~ZuFuV`pI_3*fmz!(pgOg-VzuP%{i
zACe<QNS)uwGf}`OpYaLB9tuW$plSkj;9ynqwalo51hJ4o>8D<T7jK;=cMHJo^VlO*
zsRZorN};12EOtMsmxqKV5X)OU%h{m-?9)<7*6*8(nB#rJzqS`DOhmD>uA7pu$R_3E
zFzBhi`ECgb<jGT~d3D59%tOm6V>yA@?_iU=af1!bBT2Ru5Her7bm<e_45Tu~78bu`
z-6eKBzi~-mJ$jeIu*bhP8^TnVvs>Ofrg!3EK}lV;Z|9QOd3l}42&?YZ%JI`aiHY|@
zzbRe2b}g~$E9Qw3KGCr}jZ7nYJbM;2n&kW8AOoK+?2gWL6g;NAJxKJgQp&dIguc#q
zy6UmU4m|7g=&NKWj}Wjaes+kkS>mLogU5qthr%Z<<P%5ywsmL6M;Ai;CX9!hm;&SL
zN|XM!4El|y_2fIuiEL73&O(Gx4iMz^t5;WUa;P&+b`(<CRL@2R9}>BJ1So<B6oPz4
zMfxV~QUlyY*y=jmedB?pQ4-@KxM;0gFiPv})_Y%H6^FVNdQ>-81{1$E$S#s?&bLy{
zps;-nW>LardNy$}A)QB=2&_MWGw|~Jv+e*=oC9|{A}vWyNx7z+d=tUP6Ody6>cV*N
zpGd|(RfxI|LH)i-S0`!BKk?;lJ4v@|yY<322S0AQ=U|KdSk9#y&u|o<HX<iufh?>)
z@>STC{=~Ab1mQ_fX+9l&Wu4?7^P^o?v^Qr9((nVMV2)^O^6ziPNJ|>aWOwmaa}M^t
zx92|QC_k;0oz%_d2SLK@1mJd1)IvWvtpFXbK$C*4ho}f>nP8=(G`N=i!C^)#Jw|Pl
z?^@B<%F5@!ZT>yDWi#+^*p;sR(WI)fs<Judv8A1YX>3{87Ev)V8e<hOlPEQE4SrCY
zB~;Ne%xi>#E%<Zi%NbWvp#z+^ANfFiOH11P8ty0tfMHUldo2GuuQV~R8D!}GIWq{9
zKLe|xwBKk}*ubE(o=_c9QwH)nJ^KN>&9Z|3)`gL>9|y&4%@SI&9<(#~vz&X5c!SKJ
z3Wxx1%H+R`;h#Qnt<P2qo)k&AW}zH>{u%$cb_{&k5pgddZ@~it7y=R!ETYjCCA-Zk
zS6_rb^{h9Yj<t1S$8j|krEiYiN0B;`N7g-E3NpH>*IDQ+y4{phrLL~B^8SvVqr9nG
zK7tv#`a?f;Am%-A@ZgK}^|Uo^SG;Ik*|=&ll^Bh1hmx{gpcfTNUuo8+ak-k}omkn}
zta08&Y9^PWm!(N&dz0fEg7gQodt_$Sy;d-nZ&BaK<N}q_eq2)W2p|VauXFtTd(kLT
z-|}1Q(mE4UG)*miNQGsRsPXIox4PKQp&5s_LboVE|DQw~&Hav?1CZhB=2~;$yTF`s
zCz=UtM?2#?>RyMw_FKz-`fz>uA->{{$;n9+gh`F`x^PLvu7o*FGnND1P^0Oo@+NdC
z#JG=fRZ_E=!-DdP2~jE5C8aqHjn^MJZw&E^+f51n!oZVlJuKRq1ZlCk;E)iVp`Rhx
zf6%P&KYDbd*&I9955R9wl3EH^UF10hTqepLUL{xdAcEsoc`LEh-c~R@KR>X`Lrzee
zjZ{K<e`F;10Zv?9n%b(R`4AmV4pk>ge;L*p#)Ivj^r)|!-Z1K~<8?oH`nK==y}F9K
zA$~Y-Y<vtA(KL{alJfHHNC!Lu0-zkW<F<bneLSY5e<lj$6VL=%b{Z1|R4qf|?(O90
z{!+?~bay91z8^$5DbYm9V1%=6?cY(iwX5;>>8RAy=U2--W#r|>?-Wm+V~<8Szoe|@
zy0_;P??*Q8H+k09)=_bBTDRTs!0Ks9|K<!MQ;AZ}t3eOMl1V1I)9seUBn`5F^jTgx
zv~8gG-JtbsC`9Y>4YoVopo4>ya59W@AU@LN8;MKgEE*^rOpi|P{<1A1;+#A?O|S$t
zki>tT`AhvW9^148vgI->vQv#pV-*7lbh-lh)`OI!!;ZP!eaHs2K9dg*56f$4WU5c0
z53!1@{+S&MCF{y>_t1h|=M%9bYhSn<!|xIgGw$=eJ^8|go=Qur`JKL2^GSk~?p1n5
zBcSKEQjz&0-;GK!`22Jg#|V?zKS1=?JzxTM)%ez}3Z<-ozXgv^qscevnL90tMI<T-
z1hZ15=cEU_Izn!?gMh1?B691&#xrp-F<I;_oP2Bl{6w!bsQmx6y6&PqaXCnt`26{E
zz`&`~HLN?iICm^8EF7Src>;-43_fpqW+ow^FG*;<(b*e)4umha4PMAyTHTU4TKV(m
zo9pE`zlT_KenU1Mb8+D_qhO3tMfAhWuOv~qek#w!y)5KT`4{y+T(LUnb_$2pQX=R!
z8@HI)A-2~eNzTncIU7>0i+=j_NwaMmv1qoUTxSrYAly-#+yF4A#i@!MY-oCAojci<
z)K^P);SS-a+=u-reG+MQkNwR!4up!Hy+|cd<-29~geNG9Z+~^AO+}$#85sOebuZRB
zdpH@M=6~DT`p_Ionm~+sFniJIctp&4I|O5vKsrGdr~QiNrG!q;%*^bJslkgI6^tR(
z7p#@MJJ>EOs(Lj32tE|UbSv_}9@kBBKY_^?{;n{t5I|p{Zx$~@Z`$~9PeZ;$$dCS`
z#uW=&!Gd?c9^vHVeB&ta4<uE0c{%aJ@{Kpty#6S#Ntx4LrZb;}?#@n66FkdIQqTYQ
zXcz6Hj0dBjY1Lwr6%Y{EMM)V@yzIE5ZgG2Pu=LMSr&EtVuXDCMlVjZ_bNP`-Kw!W?
zgdu1^)Yr#Op4<=PN8KDNRtVZCkcz%X0s}?aD(A}EhzM%zE<{0mrW@>QD8}XJ<Ql%p
z)6?H#KT)klylZ~<gR+7>seg<oZ(iJK&0^Pk7+1h>0j+Zn=Su{9r`HqA)ce3Ym5_B1
z4zBkMYJC~jdG}};n<-82yswW>8E*96NyA*z?r$k==bwSci@{yKWja;RACuT9y%}!U
zNTqm*5O^i<irfM6O9Zm3?<g;6?B|wi3U<EQZD-d>Z`Yk7az80YK;YP?GY9GkA3VBt
zKf5ePCH$<ZO66mr>z+`G(%@&@_P2j6&GnWA)fC2eI8CmE4~x|upOo8o(rNMjN3{@b
zq7TpqH1~ZVo1~Xw-`}U)bMH&ZHqsg35Yc9u^IV(E_s1V{4T*}C6VubLO?8vCLpU?3
zckd?4@0@Uol#`Ql7z(|kn{9CvOs(X~Kp=5-G#)rp1c@F4KQ6u8vn~5X&7)t9%E~ZR
zXYQU0hR75eu7lKSs_y0wQ<*CVt$+SBI6O1mAL<H!c<ZtGYlh3UQ<hSX?{ce2NxfWS
zAqF*mwzk#dhsph|lBdp=tNA+`u1ZQf=BOlS7SQx;(ErmkvA1G@aF6E55n}rkgP>sz
zxNEqVF*6!r3@|#f>Z4?zI>wLf?+^o^@Wbp(BZS;l!4qC5B3|9iYBTaNq}esYVI}PB
zcFaSr^cAZ-yT&KXR)g1Fwfrx}2=NP_&+{2oyXw}KxIfi*tZ6G`IF*s1Uu4@bVfVtz
zixs`4=)r|BytT+#0-rVzs3zSTtYb(*yqND+lUO6Bmp>|WAWvpxWfeMhZ4ZaQaO2Cs
zzUmNzFeP3Ak>{2}zvR+>&-|)!tuI`XSsWg;S*|=Ky)j&@c)#^EMYq6r<4@l$vT2VT
zDNl<Gle^ZMBKG6E-(%mXsHjt9T}GEV>`qvCA_;mgVL$n`+j*{5)c4@2Z|3x3kN&+)
zcYOQ;gHas$(?$eOp}Sg>r+I&aGPzDPo$9e3$EcUksfOqo>&vE$e@F_@2bXy$Mcd{F
zYg}ioH%rTOScs9`=&3<#F{QWwZT{<B!ONfZ9xVK{{EdqAov@8V?_(hy{kwO$d4k9M
zIGs~JB2oE_18ckOvUhKJ-1{`=tEP#)`VABzk;Zs+Vvnp2mno<jSAPwUwI6+{mvS*i
zyPcPZx7y9httaZ(?YdPrS$URWJw<L%vggyb{Iz18T7Laa{(32Dw-u+%S~YjkWckEN
z%tFK}8)xZRGMPi2UI~2OVT84AI432G7%t`X#%VHFp~Kw*6VXq>RR>oc`?VsZBr(57
z=RsQEl_?rI<*qz*H+uJW8S%4=PIe>pvg#BzS>q<MvSd|?UICxNsx-?|(PG*9JL~m*
zULKwo(2L<E^N+VQubn;Nck}~Pand#iLf*WI*5r}0SRq%iy_~DxrHuC)|86|hv-#rj
zJ9YHAgX}YV`$;1o!dg;{*Z*GS3Rb%5S9`8k^fHak?KIX5MN^n^)XcJIcYYo@WhQej
zSB2AwM`lfF$61~0E*63<FJVFdPUOK)5bKt!L`=om!*r&yU%*q;da^@|hK8p6=*Vr6
z`+sHh_20*)kPI(^Fy-jL!j)s?AtiBNcK+C5s&nnF{)MH<P6M7IN-8SF%K_i-4Pnk8
z;E{)qbzKV${Z%+)kZ$Yqr$&ZptE%p$_~HLMPMI+>_l2N-PLJ?RFh+ENM{zn8l;bg%
z)&F}Y(8EINYPPS_O3R(o++wc2uC?zJBfF&P?`GlQ5??l#e?JptWaRnXF!NDAe=*Bx
zYwY+^+HGGV=d*0Q^Ey0TH^z-ZB_9i+1;@^1NHwHPuUhBo)z{T*^^`6V*rrOH-)8G(
zFzN;aB`$h*(d1C}KH_>o!9*UgC!rnUgkO>8>X2n_CLOK@X6p6gMHDVv+ljVdaJcPx
z1>hn5zVDt1_|!g>B9Svy#GD#QHoEY4;t{D!ptoxEacQWz3Rf@&TfMB)uO#&mr?og$
zOq@J*>VSt_gYKRa7PLlp2>=WiDt8Mr3CpuESJr#(?3xX)y?vZJV|VdmyYKwn3x&le
zI0L!K6e<@22;(Q>t3N2k<@G<C0Y2I@k=>0JTj_t@#Vo-}6=-i5=`JqTsc6BYJzKUW
zV-(Og2G*Y4sb+D8=~|vGw`JkNo7nMstT`ZoC(x9NljmPZqnu<gS)uuo=G<dD&RM8=
z!8(^DH%P~$Ms9?mIT$^^<-?J#zouNboE(=XRngc5xX5<0Bh+YPOos>#Rv+g$rs0J)
z?_SxATkQg99-W-+|1<Onod*gD>Q^6}++OZcfSEA}aXNfPkhs0A9lCPreU8V^HpU8_
z!n8XB)?(q}U;6vvc;EA#wfk+am-5{ASNP46D8j!J@yqI5boP8j1d==sK@Bn1gNjiI
zG;x~XIvqXz?reERlHu^?=2CI@Vm=U`-x;O~Q2TPNII2MW^zrwoA*~b->ADe`p{Ju`
zzs8tp)V3sy!#X@ZJ#7W)eKm&J)q-I3^6@dP3g94Fjh(UVdY|3x66Q+W<nZ^eyRrD5
z;{DlcBXqnGj7Ab2Y(<wL=>f@#wL&Z7E7xOn=FYGguE8&wSe!$Z08}T#=~(&Zgjr+k
z?&juZo>L$5D9k6Q!fJWE$c2?&4CivwKHiF{vskgHsr>M{{?)79a>Qo-PY`p<q_xc9
z%{mJI{E$w}b;;;M^sWgLRht#?%p&=GgsLpQI@nK<DwA`R^8K1-TXx9Mn_zTr&W*-L
zZbmPwCpAkhG-;H7|K3>ax)s);jj<bOqC##Q0um+DdF@e=LBXCZJp@F~Gwg5w+)Wv2
zjydc+S;x*F$8j&K?SY;6qjh$GW91bUx$A#nbO#cXuw(C$O5CucO&n$B`;s9l$=#3y
zAkKE`)a%(<>$IGs80d=onF~rz!Y^-=3aP&8ds&JfuUymk9@sVLVz@Rzb>@ZNYMtA=
zt{u5@T{O$g%XGTT$2RD!W84z_mw5yQhXH9xPM2&mOo6jCu$SlcS4?Uig)BqLdIR_-
zab++>`l>(NfC65<iV}BOU*1;YAAN_y-p=R+=PrjvG8N81_9deG@fjOg+0qBas5Zmq
zBycRCU9woScPxxRFEeTl67)`LJITv?cU0ohGzL1&j?|mz=jwg;-Es;`mD;)ANzJ@3
zu>Do5{Vu_+N!33O{@S*8NH5$f%~pO4|GCcN|NM^13)pwGLHgQvX_KtD_ywRd)%eTd
zNYUoJ2dR#+vdZrGLND*VRvr>h^Qe3GOPapXrVmHkzfR?i&b{5S_heY@)wBBhinoz-
z$R=q#Xq{LT_emQYyQiJ3OG*bi@}?J*D{l&snjWe@_?(3>?d9R_yrUu*Fo_oENSYjt
zk1DUx3slVGsD<d;W5M#Kw&b<tg@mAwtY^f;Lcxx7|B;%;%rfgD3$X3s-<o;oD8Wx0
zI(>W3RMDCMsmCWe`U#4QlMuLl{v2^3X<x6AP0^6M?D3rVTM9Rr;v0*qHzYUyQSs@|
zN?jNpkFc*aJ2TS|Z#3OuQ*5p51n#o_Yl{4~4yUQY_hQcP%sF(%McT94UR4o?<pcx;
z(W>@4sW0j7?~lJK_P9HhhiJ+5QAI_6rU}(wq@AIES#?vrS=zMu-a>v!@wT4Fhizgm
z6$MHfZ5^V0X?BJwE&oUtR}miGoFoBR>a_3Xj~pHts0UjWnOF_pJ6)I0fGlg|n)`mE
zC##Lr3Kk5Co^&na3+K<DzoMx*PwQ?x7-m^SCKLaA2E#$~2Z+!jp&wV4H1!iSsW;?%
zSM~L4KHsWd?lezp5aM|Gej)6Tlm#XC=+c4UdT)jYLJt3Cyw*+s-1j>owP~a?axQd1
zur1qiN?X}<qFg~BgqDd$>6ibxrqLye6MnA)0>(}=&SLzkL!Z|P3<Wi;-WwDU@I+rg
zjlRsgIw92=Q$snl(nB_f#W+<W9tCRY$N!FgB|)<TlVnf9L22G`L-!b?!r4cz9A6hb
zs%|C*DrX@OAVFQ8HUVHWzTh47<_%?a2vvm+Al7v`qVWB{ufUmdxg8*#fv11x=>L96
zqZd{8wRJe2mFg*)dJ)pAH+W95ZvR84zdsn>M@4n~^yxqlV-c>Kk4rKH(Lo-GbDUxN
zgBl|8Z{94MZkT%TkBpK!TC4i@&90VUblEn3wPR|3^sg^%ig?~)lG8t{^KXV)y2<a`
z&;1|w`MCy0hA5z3Cyw1nHIQ0jKzrkM3W}`cPoMT5_c0gKSc@)*CdY_Y+oS~%|2-}*
zT$Yy3r?qs>KQ=SlHAo+fM&wAfa~nbU3PM&rB^JTK!8;#_T-WIdU|9^*c+lKDQEw&1
zW{$SGs|lKhDd$rLl@l$Ni@u|-!N#gpjWj593`S{#35>lq3#7@3p!T<}8yZUM%%?^b
z?WzvhZ-hjlYDfEW!gAj0!v@sXv+fPf1S<tAU2t?1$4?Gp)Z%Tp|DXpG+J(w<TPtec
zrNknbs0!9f@7!E;e*0YU!NKXugKWoRmPcazA%h=++So!c-qbF`;=s|p!So2Tt4$U9
z8XEmtGr^$$l;UWr2UCZeUM_Wb_%t^1%gXx24v9uL-`ll!@1T!h5#h*z>-Lbr-@ku9
zKq*d~ayiDMiMwcwX&wqHN}r0ptPaC-aRmkXx<&e2w_P1wlkpZ&7u6*>t&LTqD;<75
ziWKM1Fvu1jcq7PK1Yeq8QgTIXMZOo0`!EU?Nw4g_0UkVn+gx3p2AzX&-Dp~AMR#9o
zwj~SFpj%LC@WHib-?{buNXT#nh={zwU3Thb<*P84VM;Vz*EW=S7T!38?;96eemXZ4
z)&vEsC)HF`$}w5p$E3o!PZ>G;YK<68UtAdDh~8c9L3%B|>tFwU-0(y<)d@jCe^?7V
zgK&{(1ha@^#=+;j(_o*_H0rtIDH-M!TU&o-FPFP@ivt4NzF#(ESq;>d#Kq<4Y7>5-
zpioj(XFq5cIr#R3qwbZoHMujgCDB8xW9DFB#KV8Zm$!fb<`%U7mzl1YvWG=lzj4qX
z9n(mI3YPnP_zd!bz%{8G8vcU*X1Skb>k=}cR+y4ORbY$}$pTXL$X|L*O!T?Lhi4oS
zxx>MTxa~J00kXy?&C2Whq-SnrwsJ|@={0(UuO@BXLT&*srmk57TiY|F$cc|iL@mVP
zerMw<v02#TVoLjm>_zjgR7tzt2icZy&Db3;Lb`&UaegXlY8SovRGAlovIch_8MtAp
zHuuUmoZBEYk&1VBk66FO-ler$<`xh4LnsXa!1YeBX!!}!E~MNYLSzj)DNKYyB@KU|
z%-GrU#+ZAu<fhPLE#loQJy@X2Y&%tu6U26GUsQB-KpF$zoBOiEzl<s~Xg$_P)m%5z
zTB;D@-a^i-lGft7i`Zk?H~#2D#nxez!~qA4ghG8Ghl(VAy!5&)wmA#)BAx@?cpWjB
z?`{G$f^Wr*s{J-*q!^pm_x{*L1%<U|%ZXV;;Fp=`Z6M&f2bk*JZL6fE6?}a*0zJ+)
zn;l|lP27BZzEECfNB(GcDkAjt_upC`c44-hyD;ICHEzS5`$*NAqUM}!5Tomw_}q~5
zf{zaYUxhSp5E<9uLc^^iez7S<M~d|Jo_8y21gWkY&w}~dX%~4m?sjWxYg5h)CMpi8
z@fiUEi^m(x{<m$81IP)c!@1Lp@Hy(@Qj9X}O9D0?Ab1~djWf^-4~k0Nx4~?C$&HMY
zH!JrSkJ7<jWq2Ugel<OZI1{Zih8{wa&PR%^sG!h&Z8aj8`i8?pu{k#;LayAejx+|P
zM5~ZAy3-DSX-Pw)V?TD~E=HhJ5XX*_i|bRd3TOLuUSs0^JGRW2kbNh+3@SlWkuZ+H
z>q*hkaxVl$z0x3BY(QBNnbv}nJcufmyP=GnwP?Ft<LN9xtdnz0NkhYXbhV*tc$**n
zPe7e5d5$ZR|BbvGVVG%-eQ27>U^Me?8AY0o%!@QAj&2HFGvA*2@!)b-hRoGkiPPsw
z?`0`P8()m^@?<&uBep7VJZtHT>(=J|*3XazR%*tnKc>61Si1Qzkv~h%;Lc2z>y``F
zuD~L9Q!SFIv(NolIU^kdLqJf_z`j1J@XMkq^Xa7h8(#4uL6iCaX#ql%;%0YOlcqQE
zA(`R?MPik6<np^SRhHthu+K@+9)I)V84`W{+<~LjA=*iBU;I8XJhK=fX!Xwy&<uaY
zbRWa!_P+#gALZqq{<=cLWBHxGG&CaOFI3YRJ|1j#WQj+q2w$StK|xJ2F%1LTI*G}a
zH(bQETRh++uw&1`V_x2vkAqD}Y~#N?)~8H)m4E6DU@emjT{gn&KpD~|upq^%7BRkb
z(TW!AuVMX}udVk<Nv6whR`kSM_=vZDzNP#tQfgdS*4i|k&Ysmn$_RSzPnwAbak4x*
zOFRi$JGP}2ZMwcMwrWVCaNI~KzG~Y`nR&ajaKEp=KNOx)@VYfyaC8=!&e^x{a*p*z
zhVBcmgm!*3BC_|F8ViMvy^mM18tzMqiYgK0GXk^$+N|2WH#&quj<|b_y(ma2PHlU~
zCwe3FWZO`Wpi59bPJKny6KH$R19Fl$%==e(A@^8N{0Q7W(N7$cm)ADYSKa<54ZT_e
z){oyFH0=03Wp(}>rlFn|GtX<G6tXOp8*(muTj3Lu^Dhrk2ye*q!10Tf3yvH5J*q~u
zMi3U-xqL*+yW&M+b5ql<nL*AO^x~6J*w7T};&SMoa07(%IbI{J<F9?nMhA>A`TN^9
zxnXuhL@$?S(gNbP&mH{hSjZe!`@Hr3qC+8K5~4b6v$*kNwYRC{dcF!%_UKU0+%j>V
zWN@kcHRHf){bz9YTXk9{afLF(Ka$ep38_pcWo9D_tKDS&D~)rwX}W^+S1-SMH>kY5
z-3sLaAKyTP)8x$IkYAq6j=N$S{!;E=IZ9f;$@>wzs9U6r9!f}}89|kH0!vbLKhM4@
z`E&G>o*v`uqecmujwyx6ep`!Oi`!4As#{5X(dFxt{b%^5XHSud@%ryN4TA^sf2%Te
z(oAcX6ZrJLyai5~J0U4pye<Sit4xCCBaE@pigc4jEA|5_H94Z90+BRd_SyEWRyKe<
zq?ru^GL{X$NrxGw>R3NpA3yj8o4{A-i9e0n1Kc;sk-W=kO6pIlf%3zL8ETI%J${_4
zTprSV@4|;G9pVz+ULOV$?b|Tzq;!3{Y@dacrzW2fXP_eMbH?nsBUdP{T)SFjetL7@
zbzthUiLlz`P5SCL&T4>LI$esuh3oM#RO=bh7%e<(4Gd`i&|o$nN}*`f^Zj9Nbo4j=
z24Z0qIxcRBhg*1$+$Du{2?nFvnXM<*+w0zoJ*~QFT?AncjxMR4AFv0m`#lsC6a_Qi
z55qL)350m7^P^={f(0d@{4rc+?oxSSKD|-uE=!nLsc338@kfgKnpb8e=@hDqhF>18
zsU528Q%-9+ZnNC9t3_UP-|mtrvOUes#M##S)GT5{e`p0ZHg|GLoem$Y)*%5B**f*N
zGn)m@K3H_$%;6P%#GQL|wed-;;FW9FM)frT|IbUzEMQq;u5dB$m*Ph?KkZ{<BD2r0
zP@|~FwKG<cR>ZPP1w1?sQGa&vM#`K7z*UK<f-<nA*Pwvo;NXC$#)r|+!5zs+I&?nJ
zfy+a~j6S<wOxhk6bX}o3w?4h?!i5U~38oh=cql)c2AE)Zc<dV951rl5K<xI1Xjdy{
zgpq$fO{VD5oU|2PiJ8#Mthr_G#CfU^V%_QAE*K=fehJl}q<Sv2NkEpo-OY>Gk>cGh
z(k3caOyD^cKk^m%Rop1G$nK^)SbbPFvO_UU<N>Bt=pDN-pMz*ydQ^&QJnvLn7}861
zEHJ173mQ6z=JPnlOz@wBFdZcAP9x_9dfe3LObHUR8LBUKbxqaJ{gyvfq>L^VFQi4d
z9aX|aUb=41MUo75pO*H_tm6KrJ#gS*%a4S_#Jq`tQr4n^%wcnEyZhZmohh=SvFXE9
z?=5t^ZoGXcq1_{xMLX0iHqOl@IWZeZTv%ZHShy^K^8q`SydON-CS^O4wa5L77F81J
zBAD^k+U|>F$F1)=6Qsn!^peDd6QHK81n;w+8Ue28P4(dHiSzzc&z5P{Mc*gQ7JbP;
zb2gn%g`AcaJ#6;2A=Y<1Y?99+eSkQmi2Z<_Bp}-8gY&}?Y2sTQBj;e`LY`>R#oTi_
z5KU%qhwztLJZM?8!rAS-bQG<^76w_o+}yb-`B6zP%nvFkX4rRNR3JaN2U1X-IC4Al
z8co8^)O<b;FO%=bYx*5W9{W^$3v+e+ack$-_KBM1{_d5DsV#qDb8EBCyetLYuh=iZ
z%l@p$u$Z|UTW#0*<i8w%V^(`RQ03aXJ0dzT;v)pGI~AbDa^PcN)I5{%gZVf+`_1H$
z2mc*78l$2i!vlv8S1)zD*1_)Z{S&_9$NN+L4#9&0HAtwmvp7=M+qhAJkYPc#K?1z&
z?1bSqM(QMZ=HUTzID0@C%Rnq!zuXP9(NY`hOS&pY-lCN1m@?Hdv1pdv&e1F9BvpE!
z_Ee3-Go~H`&t>TIvz5>4*lwoo(=#+IJ$^ib{Z1Lfqsfj7O5S&*yaBKL{d|iG$;cKV
zt9-5G?n0y48N18re8Mxgvh+Nbty$}&*Z8Oy_y{tbfejGCKukSkPyXBN3@3q2_I;1v
z6>rX6m!4R>-@itKuGgB#bKNUZk!S7(v_F%BDiZaIFO%@kmB)bXfkNRz&#>&4-rSf%
z?HH-1sY!=hzHrnCZ+^U~2r^Wgq{E-o6Uy(M!~^j$P$|W2?GtOEeZl#{<h$wmHih}J
z^FGZ#|L(5)Ea0SAxbM;Y-`&iIX4sdOMytOqbuD=dFH{Q_v!l)hKF$7K%IQ&l#f?$I
zCA&;t4Pt_3qT|kvS3l*Vn(x_#wfBM>{tF5PMXTtBvXT-TH+Pj_=TQxHbuKP0t0GIZ
zqvL4Z#JO<MPzb-QfI0#L3UZ&i<=(i&00JlH=w+XH{mJX!>qzkg?To`~5C=KUulM%$
zqFzAP+vqxJI0HI`;<>J;8<V9K%u)i9J;uQ?$Z&PhCXiO=f<(|~E^aQaV`5^h{69jx
z2YIEXnRE4-r&4-2$zaa5TNeprYp#=^P0{L`4AZfQu;n%SRu#wF;;yT|BSTM`JkjAX
z&2C@mdAe^Iu@Uti*deLHoE!&_P79V}b~79y)#LOZ4V0XU_!xZJf6ii*nH9bd(0P1n
z4>0Xy-3%bX<iFmF$=|#gtKqw$`5x0Smv374LHcGh(I$LHhSc`<C(UTe{%Jjsmz&5x
z!t}0QWi1aO#i<7m-<c9?XK$Th3NJkgD~QXeFAEJhLsc$f*P6kE3<}P|yK<x!4=h4~
z!zq&fJN|dWN7ci)kw?J~<Ne2h$HA6gGdS3Fc(gV11e(L6&@AQRY$I$|PYA!vsH3ID
zagnL+^|l&vP93d74vQ_tA6$#O%FpDS=Jf@8a(HXWesiNaoA3H!p@Ll1))B-DxdhEd
z;6(%o^DjZ7#tejUYmbB8&%SEhGs;8??{*V2GcF*a@c==xiFZ))@$f`PMY&-)us+uc
z9kb?~;c~pxVN<92<ljYi&gMIw`T{37I6NS+{Ed!eJirP-fprT46B>^Ix=s0-KD?e>
zu$vUy3YKSuMbXUli_DKm(sKm}n4>`g8u{#hiV~o>angs>Tvr2s$5ok~xslnr-ZtBe
zOs`mfW}W+g2`Jb0h@lHTtWmEU%=!}IBkxoMq`Ax}9JbCrIhY$_783Cb0M9r*PiHWC
zX99tTh;?V*Yxag#0(g-`E*gXQrax*{=)hMeydtF2;>S%P#qWc^?u$5G*!6SJnn~0`
z1<@7DJ^X;;UP?vsVmHFgdpXtk-<^NLPXR^525?%q){O^6Ceu%6L{j13&RlsvT<aPG
zDI(BeZmg5Uw#}>rRU~^Q!u<sr9xJZ~$AOT%mwZfPmGj*9h*;x^!^Mh$oNsGy*W8{{
z_3{K~r~<`vMe{Ya*uF;x;qR`(NkuQ!_weUK-RIixYH$Cs?5K&3t3I??8P&e|uqw;E
zV7^^cH{XPo#bx;b)M@jM6j00jf;{XRG85IvJ2!8h1e6b@@xt>`<Fh;0l@t4tQV-Zd
z%Gz)*D<HwQ_5PcIF5W^zHMG1!?2)~(iJL^!D-&1+DQ4pta$!ynfG{sb!W7uV^=(pQ
zlCzb^3s#20NI*K`?g6BDCMF-)C3W<xaA*c0!ruIi3>x_--gU<lwY;M)ph0z0{3Nxc
z9Otdpwv~mGU`9wyextsCMe^Tb*K;IvZ0F9ki3{=S2^KVAn@XrmMZ#2{DbG=#BV@YI
z?2s3xjo*A{7~1LOT_EP#RX{=Ycr;iP4HSEzu~6pxTQ;6n{40oC!-DrBx^ilO7yL*=
z?(E96=lPt>*QWB2%kfEbr!OdhC+MqYj5+$fqT&GjAP`k8S5g6}Lx3)pauaDCQi0zq
z(uk7acyd$jqobyGp~dJ9Y9SaI!qZg~?8U}qq7-T5%3|xCeQ17t@tRasmz*naK>`?l
z3x#iDok&sbZIVunL;)>OR<A4nC~}yIuJ#|NoGe^2G<05Q2Cjmu_GQWS^?lQX{YC~8
z`;C>0S7-tgo~IQVmgln&y|v@-tjyA*FHly<!w{`EkR%Oe;>Kw$CJ}dtxq*cni|xF)
zrNqhC;@FE)PaI6kTn<qG8Hhx+m`XSelkL&d$FKJ#XJTp6kS?(5=6dfwS7Ri1q-R0w
z0zXp@RDC@$Jl}ToN%OtEKMqWMy|ky|Nu>P13gxuRQm~W}<ij_yD=)kER-4>|C#A1p
zn!S$<LdMS6P|19x_b8OEY7Ru)2?*+-et9v5+&#&QB7*!mVTmjyP0e~<X3zA=={v@U
zT5JyeX?lCD@V1niio=^&?eMp23-$#|n>(Rq(@_5g!xvvjkY-!&L~cay4@LhLt|?8F
zJr86|xxLa>eI?YrZFgO|w-p*c{j7Z$nWPF}GGI=yYI_MtXe&&&YH)tMMU7hWIjO$w
zIkEhH&>T9PU!|&X>d5(tUuPf)CC!lhQJ18hl%dOKnBs&S7h7O%Jd9Ke?@-?@-=oCi
zrsg)~rr$%bqE`5{qFkchBPz2`hCi_C2-*FcLTB>!qtAsTSw^N(JixQqs4Un3E=j^~
zySVuuJ`Cv2(n}FhBX0@SAU2=S^=j&%-e{QRptvb?$X{}H7s=ZLhAD{rE4QQ+UNS*Q
zxO}sr|H;;KD+#*3>Jyau<JO;kkHShh|7P-$1rtcrvlpt<D$LGUBfAFmIIxb6mmCSD
z#l)8Jt4YlWCbJB#tI#I<dc;poD{|XUrrB_Syt#?u2d+eIX_o2uOJu$@R6$<7-)|js
z+j~}@5-0;0ZG0!7CrQoTN^!6TdiCm+4Dzed<|IZ^E*i^`^9J^x+Nt_gkQ`1ei-OLB
znO!5wscHN@^T|+^TC%|G3LS9XYCdV$aRb)5b#cZs^jHIA&$I0>JPAnh<MZ<JIHG3y
zq_|O+9J@6$NM+%Pw>AuE(dC&1V9cw@)8r&7!^mZY18M4zDtXdkWb1u^uYYpg#RUZX
zP@8wWCQ)~A7r7x4X34wDx4GF~^LzQgM{J$4LAxm!Fr#^EsOm(d?Mg-*d*yL9Hl&$K
zLRKn!L<KnVwTYp%x3$v@!L0NF9102!25v^eKAkzORFPT(ZydrGNu6`yetA(IuNF6-
zC2^=GbV55RYg<H}U@ZB!Qj=D*j+}>uZ8AJgMetu!bJDsLe)zaI8iTFJnn-%hjYK~Z
zMEmjMPb_`=PF-$C18l=6hDb``bBLw!>2Hyfs8OW>Mg+g20;+-})`5DHGzbJYh8RUi
ze!wlr+0>n-WETV*@(jeOZoBr*3PC(4wiLH=8do2LseFoigaz(sO;90fKYlE>Pip>^
znl?azS!i7v(T?f3lNoYMQ{);zi?ewJ1Pt$Hx3kR8#~H54rD(4ew0|fy-j%aq*cC0(
z_o3KJ&<)KZRS;Od1HwVNoR<`Hjzvu$st#$rKZKb2NI%(H3~a#C=yT-Z5|FlF<8(QW
z^8<1a!wiRC&P$yN$WS{u3rb9c&xB6aP_doi3xbo^!}zAa@xc1;(H#1}WP7+|&~*Sy
z;0!hGZx-37(XPhB%NvR|P=ljL>>^+U15CG(Wx1nF>#kMIa-#W@TM1r*34=%G?yL)s
z>^Lhoj^b$Tf*`3lT8>>Nd1dD09JFFTaamDu2Hy=Szx}vzHf|7pj_fj6x4Jo;YRd5_
zZeFy<3x(^%bS|Yzq;xn|=rdX;o5u3%l<E_;c%c<)7+;`uKj$D%tQxeAj)@s35)yKc
z0!I~_8BCKEMcH!U%9U3Pmk#__Sz9R=J4vi|3SRy1oD#*eb@-7`-6(M*BXjD7ozz>L
zk35dIo<UHmC?sf#(8<4adQOo#Tz@L87Oh)l@swNN4_l@`n$QEI*{klOcJbm46cFIG
zi>XznzbumzyJk7EDt$IpfSec4B6F>#r>t=JAAFK^ww7h`a;k$UsR+8?DpZnj?goCT
zbms_$U*k~D{xx+EBaa1_ft(`pi*^&qu*g>2fREh%%C-VBzls=y>kRW2fk>`pl>Vr*
zs38_xP`b3POd4Xh)aA4f2(wsyHv~~PGL}6s64@w%m7l-%&`dqcA!*`Sh^9=>tC8M8
z$JxgZ3qB>hzR!xZeT2>vKWkYiVx#PZ(DtEa9GKxl3Zc+{eudP{O#BQ!t^9rZDcYv?
znAL(?A|@v0Di_`BT^$vC#x@hr9>e83AiJ*^zhiyE($y4<_5x*<OT0>iWGJH%@hSry
z-4}ARDef7h(&NHk6}hWErxPFhA+Fsimv@at{r6VoqeH5~Hu_!oW^XNt#!A6$9foMa
zl$@(ME!{nWi<5)9%86_mdo0I#rSa%S3Wh~q^Lc}Id({s91v<d}eCNMBYfs;%aqq`x
zfYWn;NmK(UQoyg{%b|lTZ0}(a2lmocv;?DdGFC3F#ggougB+1N_@|@CUD1Pk0@~&{
z%0q^y@3fHWbKEooC1ug^BpA+u#z-A<C<tQkI|$VFZxomxmk%G_2npcL_kR2KZF?#w
zCp)`U*7DDzEb(`J7*Gf1y41g}o;o`gE$Abd)Xei}GCzdala>`YB-%|31kR{PKMlkR
zGLu8$mw!OsAjPEZ1oJsGfoUZD9-X!79?RDt!cbwgpyz;vXGBHhQwsu`n%<#_CE<Tx
zs$^#e^>A&NXYFly&Onl{K(WIyi>`uDw4&y1NFp86FL0Q7Q+L$D(|&UDL`ZdKpZ^vT
zdhu-uIW#yPf0|&@mlKxlMw?~YN~<Qx8v2AZ@B$1fL^r?i(t2msuj0nwP=50YSHs5!
zGit(QCG$Tzqt~w&<~|4OmFzWAte+6<tHxK`bL8y)%+`+AhuKu9w#yMI1Ol>&&3Q+5
zP*9K<>NVoikf5L|`uhJ4REIPnncs`ZJvT0nHm0i@8pjz;liN>`Ym^=cx&t!wS4P=&
zC`=3;dMRAy!n$E_T?^lYLT!fs1$Kn5mqAK{)S~+SmuIxP3zOBUgav~?Z}(oFy;F^w
z><A>5TI6iQr25VM3%?NxI+A~an<<NIq_i`!s1%O1mw_?8s;=N?+{H!b-L@fU5-r?u
znR?p4&1?O1ouoW8R63NOalXavCu>mjw!+jD_sNoDKayok7f=Qde;qQF7@vJCkNeMO
z7Cli=vhe&iJ1B;lNz8iS#dY$NiXJE;s!)8o@H<{7rIvSBgu-eAc=acKW?_~#Zao)P
zU27W~MOD>>1R+m_i((hj1+VC2E7{Z1(2yUqc7FVvF?Z5M*TA6G)L&|-!JO0*1&aYU
zQi3&>l=d01Bl}rEnjpFtAoNE=%PW@>Mghq$ze9xB*uD){gy4L=!TCZnz`95uYe~9E
z8DW(m)X)zQ_t`Re?46a*t@nuw%{p#nV}q(8M|D1<#)rfEh{mm=;w^c=#FhMPRY#qw
zfZxY{d+F)nLB{Ug`~A2knJHOP#SIva_#^n_ZNvglRC#~)HCjMVS@p<}+PzL(4<I#J
zz=wsFj&7c=C$Xj}FOpBBsj#1&;a=Bor#bSYuH8$)aJo@Z@nr&|g$2?v?$^{;uYMUA
z7+41V6G5S=2qqRQMAP>BBGgCDmY^<iPVCT2*^$(|z?OiL>xh(N)8G%S1kHfY1BUx#
z>bJYc|J|{k`$%pJtCKe<z73-89{b|8qa-&`s0lz$&HT%(i&jry@03BI%BP!7K%{gs
zo;6^;O!jFK;AK@Cd-PF9W>;!aqiQ4*0>4LXf1kHRsHD8lZoZOeC)RW?OC@}a<$&JY
z+#L2HE8NTFqZO6y?6PjQ$3VvT-**xlisdXy-FfT<W7$iEP_lms;g|M-rjDahL32<6
zx=~1oGg!6HV5V>jr#MpWvpk8@+vU3D4537PiCg~Jpx@_XHF?G#oC<jNUjVP@#_5+_
z{A<b;{IfJVn-r2}C@xjc;=<Pj>d=$UpvZlHzO!d^;3u~KAkIrR{R#CEOC%WB>d>6~
zD*GpC@^Ekz!7hTVZ@+`%{lipmABY^j^Xq-@Z?$uUN!wtG`kKU1qMkTuTtc9uq<~Mt
zz+BzA6VhFxSqngu97f|pZpzBaR=ZK#^A_PB=snnIY=b_HwP?L_wi)uNc|RXXRT9FR
zR&pW0!V~h0IIv|_x(ZMX4luqf#14bM*ignC8on@z!mfIl1X*fn-^kJ{?Be7J{yDf-
z6w4-He)n%WE5f<<9#Q7u{Vg#hH8N?>tAl$PR$K-|S|Ptr>O9~#z%4!D_FcUg?Xo`O
z?;4BdSLyCca1s*#zm*4nQ=QZz=Zu}`8^PX32XnT|mCL&!KK|6=X+9nO1M7%L<@?CN
z)B^MXU-8}hY8*P76=4PfCnVXP%zkc^pgnzW@1y)$egR*kHy=T94fZK3wb-DvMTu~6
zkEmhq-l!u-Qkx+?{NuV|3i==RQAWfW3KC+AqgfL!5rRN)Ra3JZ5mlDtvKm<4u2u6+
z*e62D>7K24Cs|Z#FXfaTvGu-4&25_pn3DV=ZB=i2`@-@4Y|Ms?Mkt@I{`XCH&kVl#
zWV6r<`w@~Af|YK6EYqpvR5azIV_0tZSK}_**k-eYQ=P5S=ScAKkITvWxs8K^Ik=Ns
zv&=(`eA3|Qpvj{Zd(fgl?$mUA+wKKV78eJN8&kk>GLTV`LN3Y07D!HKo?SosMC~sw
z6Nkg?QKNq`%Kw1m>J8sDz@GO$U5pXi)pUVE?C?d-z~-dz*rB02LZQ!=XUW~ur;?B!
z6MA3l%K0zS1b8pQRESJFi5{pKI_E*VE%-d|f_zQwgpQw=`N(YHwH_<shj#4Z+}?s7
zatFLit{3clPvPm|3T+oui=OD=)%#*rs-Y%z3z)#t<tcP1k&403Xx7U@N1;!rMSMZQ
zdl;bDpr5JOlKuZ*6f;V3);dRrDD-q*2y*AHgsN~>>GqO)$-Hpj3aoM%c&&hXhZr#O
z5csvbm(6c~y2UL5lB%6<VAg=X_lfQeN3_>_H#M2@(z4?%Nfx-CL+SBtuqC32+=5ig
zWVc`Vu03V4-7r_6=pVozzH695_8nd4Es04<5^q96%rW?x?f*3O-tk!X{rk94RwXOS
zrbs192_+<>NTRaJC?nZqXH_IbJ4GmxBqOqCAxR~~sbpto@2v0f>b^d|$M3rTxvwjo
z&i8md$MHOl<2kl*%&}wF7)wU0L+(i<Bcf_NJHo$16-pW?(L{uW<Fk*6)|8aT<yTCI
zBtHM3?Bf-z#85yz1|*!FoBQwme_xSWqU5>R{C=0;422$}?j`<U>n*(?CY-~UsvkH$
zxA&$87}(EliOd|1C;eJ4<x#a)q?eMTV7)`GI`zRTxlVxzrLC2^%<>F>FeoN=4xneu
zF6a;vLp8q0mMzaE$1II2xK2opizFTircz7L$>Z}r>N8*1B826O#h~^%2Q;q{|A&r_
z>hf40es*FPHdP@<`0veZ)PQk`_NZyt&4{kSMz%OsQ{7A-q=$3+q8~66h42}yzu(j;
zlUPPuR>HHo8KziEC2wHibQGH`yo>#K$9iKzB@0_6elf3}CYGeYAs2ir4uj)JyIx+f
z>JIZ^<Wa%0j3cT(;1ou)?a+?;tvK=xjVzX&1%5nI`{69asdGO<;^S7^-g`S23PLmN
zOqu7LQx<lLC@FRT81Dgk=Q>ayg!*v7w{nk^)Ld`0<F}#dQrg%Oc0!B^Y<c#y?R%ui
z{n7v90;tK$`~H2~PH{ud#$N0TAtqd1nk=42K1^U^fM{O;4=1PS<Cg(xW*E(ZBXW7M
zh=hulFZ*-aSFff&X}=4@XKF-CEV~vwwpLEz=<aXyvUYC_(B3k^X0a1+ouBDA*v(Rq
zVj}@41{@kexq0aTM;&vq`qMuPWHpXj!fr;EcaKmT<J0(TXzf8Ggi+!j{qR4C`5wy0
zu5HxNOzc$ax8Y(()8(|Kh_0*Rc6ZiUx)J&aLvc<p`;VYaiIg;*SqRlex{fsQK1zym
zj{e`w<08_zhZE|=#7#|zBah&W@4&P-Sk4mkDqlguT6JZ6Nzs70z0~8mnHI;G(W-Es
zKK<|_%Now3Tl7oryzCeIRP$*_(E7W9xK(6W6lTHsi-$<I+2BOW+jLg&C@%U-<q*xE
z=apDhJwHhp=3QOqOVj5F;D<>(2cABAZtqr?s2i{KY>UO!aPhB`lUciJG*^LeYPL8x
zjvKGdRV;z$pK-e)HXY*i6Gns0^zHC?PtWVm4}*b(u94ApIT?-ovuId+)0(W096N4>
zGE-uwI!Zt}I3tUQq-DK7S~ub5TWV^)`}dr@s%*SyEUGdnJtDxY$e;R)?Fqg!o1i58
z-tStlq*O7y4Bf*s-Q+=$lIQ110Y_8^Qg2jG-jJM~KrMY<>TwipU3_2A9NQ}9<28}T
zL??~X(v2d^Aq0u_>fKao(*f~rJ*)xK4}K&n0I3}(CiS*K2UeBXb5e|haFD<IYKC^D
zK$J+u=(o@B<rOPg8r@^G3iSJEf;18{ftheib8Cg`rAUJlbxkj2C&eaP#3*$p9w8Xh
zSLT$)!P74Ji?g!7esC!0j%HC-H$O(X4vm=>K#Z6*NBj;Py<KCNX+c&JKI0DW6<m4a
z`NPpC;`;?5ZA?x;L-jxr+Jk1qPhdR5{p%joBKgp`HKkLj1|DK|B2$((MyX;Pu`w$M
z8Dz9v_NGe~>X%X*Ge=SVh+DUko<Cm&a^KyIC0*CP$qnNhczD*76!&RT{|ehY_4k@m
zs9FLekV~noA2*es3BtP;AszO$N=T!aCczsBaTLups{XzenIZ?bf%-&0V3v!%7*iXk
z>Iuj^KGy!zE@#7{53|HJC#K*vUDsg_5r{6zd#az1ni$tjw=Yd%APY1C+@s(*K;M5d
z4aDuQM1R-w75#!Z6k^8HCIX*qP~yqG0fyL4k%9Rg3RSJ0(5S0Um4G%t#ctY`=|e<6
z&|*_zYstX4)1MK%($W@|UR$H=P_R~2zGX4ZO#=JsqebRMfiqvUiX?LAXngN$Sl&J>
zq5kXy3;~V-P#|I^MD~4A-NqpYuDvS*T7Hs=iOJ6&^b-A=nwnvo?7vE25C)gAbwpMC
z?ECFTozrF^B`ps^G|b8De!dMs8Y(R={Vz*k^fip=g@n>F!4tQ6`i;+~wRBkFBC@$8
zu&1W*yXYC_RpX#SN+QL}e&hVeUMz{WW93b!)22IKnm&!aGfa$(<55SPqcgLO<Xro2
zBmd4+zdM$E|H)ivM1)#GT?Kcz+R-6BUcJ=iGPB3WnpDv4XB&~t9Eip*w4$2|P89H>
zq}0%0f`JcJC^+4?pr50N<W0CbMHJ(gk@haov{OG|yO?Dz`~F+H40rf$_(LTkO`%QE
z(XCrJtL5V(k4Yn1-+T|uOh1OVH(t2%48mMMK8QBQ#~;MS)k1T&Sds?b97_Fac*)FT
z`r)ba?-`YqF^LTop7)Xgk1G@AHAyEn$vQnXmHvYm`sC>5#Oo$Wm<G-FtWHI(-^Y(y
zKg1|C;_l12P}&z%hDxv15)LKm)?x62s^SpJ<D&D28V6jlHsj92hoX4ZCQv)N!W*nM
z65%5&<<u)JwS@7*W+Eqsdu9&!7zd>8;SR4uE&t1ZW$E>q+%%JO6yG31+>eRTLW{Y7
zQ(E8JBA%Ma`!hZ$jz>W``GlKC^P&WVbPGU|v`%>*XDT<4rf}s=W2k_3`b?uyF*G7b
zc(a9#9}qF_-rw@2ziS=0%1ING-rO$D&^qP~OhMoG(OOcK(x^e@m#F;gDc(Q~V(Vd)
zepK}`2G$S}M-byb2qHjQ8_5Q!!jI8lCh!euT!FdSh7;Y1!u+fL(%6QpnOneN9*Z9~
zZ{1pzo+^f$<nuqJ+f^fKz4n8qx{T6bbF~p<9vtSqW`iyNL~6D#U@^<X<g+6!pzM&!
zfwgB(buh60**Ts_8G4MNc)Ro;@KOo%9jU5Ei%ri?HGRN5Pd$vJ3>auU__S5H;Hd2P
zB94bM(q*z{TQg3l>fHha{t5gEta_}(J^&0=A3ra*8=Ceeyb74PYyb>;a(V>&;hUSA
zMGz5{QyL{`uTu+|i2d~Ay(6HBN!X7<56Sfkb5Ad6r<8(ftwg+sOwwn=tb<WW#bMyX
zETF*4!n)qpYb3u9(CR{Qzc1dj<y&;OsU<?L>vOsPX?_QqHdXNDI)I^l9^r+^ZffBZ
z4ikUz=v+V!I-@`taD8Mm*z5kNEqPB^q&RC=N-TlcJR2gnZS<s7e>5eDIGu#>0!xP~
zP@z9!bmS}5taaoF1JXQvHZ|ulYfZj9Jtz$juG2P1*%{Nwo*IixQ(CQq2G0-5&U}6V
z9w_-P_^Tow2MeM?*tgz|j;?{vpX%Gg?~G4E*@^m;`eYvfbx`kg=O6QW)2yiE@zII?
z<k|W<_IcA_qFxG_oVSa-6_=2Bfr$+0&qs4^y4jC^Z?c#5{(WPLeh(wvFRWpDk723p
z6F%l5`Yiy20aa|2_slRA*+b8;e*M<Xo3BAo0<L}+o<TCCFx~|2_ovqY_mRRG)?;h7
zW@O#DQd&=jkNY;ReFv|@bFiZ%qcVQe-xFKlpxc2tSAxO<cVbSTra?)Ow*9wo5L>ni
z4x(UE1J2WfiPYsbhum5geI+wNaa7=6dB_dTfZtcy9gErIq}YkV>(lNombo{i^Bvc+
z>#og1yLbq8v_<?;vkulwi&Q=-C&bvk+ejm0|JE=4`=FgMJube^8j?IqI9g{$n)(^;
z#{GN=!;lLALR=7|AE+f5uDeP7%W|ZJW-Zk#92UeP!Xvj0`E&57?`xu`JzpEy!8r~o
z<FkNQm-O^Fp%$tQX<N@5*t26-U7M`oIN4EnHs57+fT&i~H;h0_ZJR;Z0lR!>+Ac}K
zQVBNuebJNI6ncY1HYkzmtk<so#W$crTjaA8r<baC@6n@=(9kaKryi@7;QnZVU9UGV
z;H}GbG_}M$0Nkdb%iv`!V3cF4pL{Av^xnXw!q$>i>iwRP>uAf&%v8Zhc7TZjgsE2@
zOPBPdD2b<79D=XLv01aOr^bLh?5xBs8eJz9oegsfKB{{UnT1JmZ8=(O?y&Z4r(=Gx
z4IJdwfm+0-|0wcyS4sL%h=+NrPZYhAO50`RTie6R%EVj_Ho1)0<#Isz@ZkZh|HB9&
zvK075C<2ePZA;E@+*>325~EZJJu!|^YGdc{2QGYLyUgH0NmjnVxsCpmAT_v{?}txt
ze=p#pV*?iwXj;u%qf{~})9k2{ZDl?<oskTn2AnQh-jQHH`!j`V4|H8D^UwOV^qU){
za%TPL%H_)8_$T-cD0Erhe<eM@4$0|Qs$K(F!Z@C4h!Cuci*EHrUYGcNr*wWe`c3lm
zKo=E{RRmHy_wIE~&`5lxn|#WYmNx&~vse7D>8{gA<1rI<2%2+7YIj;vZa<V9#nkl^
z-~pNu1w=D8lvU?;OQPEEDE0g`>;%o|Dg6U3@$Tp!u*Is>_B$F-Bw*XEEhcbUy$-me
zb;1jD*_gv$7T9dEja-CEu=2p}&-KF>s22k2xWnm+O=^M7z)L#Ul-|=xI(sB!z5Eyz
zxE1<1rKc&VVp%~lYTw2Y$`r1aMZHmTX(nVTVDnfA2oIqc-AOrjsnqpT?f?}6ycY*}
z55)n^_O5fO{dV=wLzr07_5d1MH1-*A-GA$e6L>EHn^oSH77%Zn(^Ayy0G`A1XK`no
zI*?bx&eAT;qy{<Fl2b63qjhrYPa7z!Qo02208K105`hkqVI%_P4`>Fy0vSzsq~Kcs
z+-mtTxX)P_YFLJ=UE26aPkr~&!Cg!l;I2<V*(U|2YJlSf3?(6edO1|cVp?)Ykgk==
z6SH<p93_7ieGJY44va*_#|wjdvtITrGw2WJO6|U*emf%S_@g!-z0>M*d-WEEgHjs?
zAOpF(y}|9!h{pIdeTTi1L0UsY!=_zk0nIHvbO>BHsMjvVHE+HpU%y>idfGVK$QO0<
zNZGl@0`}>Qjh?(2K#-&y+vK4Qnh$U|vrWW+t|}IYkQm?i2J3|qD&m=2ZDF&%;zkOA
zjRn5y9Nbn?@^^5D-98@FAef$N4qh43cz*lr{PCAqT&Iz|1ppDE&XE!a?_v{|hv^j+
zDz5uuQf7RL*A*?F>=YUMd{qr~5@Z>-mqu5X)8ROOXm+2?ck2?3#O<e38sn(!{W@9N
zsk*7oU<SVOms09EUdct9b8sEqHH(?cS2BR`_4o>J-n>cL=KisyhMgPqulR3f*Al$Q
zQ8wg~UzBc?>TWoLu)xo>ca{2<9Mn#^dR@T5P`3(3)}Uyc)g%lz5TTv|6}u}|i_(xf
zcuiyY#&9)sgDThZ3i6X47#CjGOGs`l{hc`^tf0`A=%5Zn5N^(Dx;vV|`~}mw+Ag<c
z%}B2-8bjxO(L4XBbxwafygo#yz6L|sH&DHz<gm>BVaJ)$Xe1n2Q)4Ohc=&=Jn^kU0
zciEeEaro2s_xA_b%4Gfc4BhS#4UM}hUZG)OcWQbh__|k3m;Y}pPS%k)x#43tI5^C5
zTWV?C+`BQJHf<iBDGg0I0)qb1ZQ85{8a>T<d3gaII;;#dkMVJLBO~c$uZDtOafM#M
z=fvKdfl#NBvJk2-FkiqJdxlequ6*Hiz1tNf!PedVpp>K(zkNHUg`vz$1mn2KOZ8Hp
zv{T!+VhdRSz*Ax|Sht<#5Y@!{HIiePM^iC}=_CIU?XmSn#*@azTBw|e+Bx~uL}B|<
z1c@N5qAU!*KEX`{qxVP_Gn~P}x~(fG#5lC!^N|IZ00&Pcg<%!-N7LmAF8CmJa+qME
zz{M#kknr;54Llkk<rTODB{j9{)*M%qI`pWUvMy*}HZ*+v@8d>pug7uAf(a0;yy%|S
zW+@2<5N+AYeB26kQJ;-9xb54vb)rA`x&oXAYj4V|vd(22QL8(|q9j4?!y2h=xF2F1
zOm!hLPW|76Am#4|8_j>d#<U_nN=cA~<``c6Ri641kt_S9r0xR6xclgln30MggA%IN
za=cq_T88=AO+bt%A`bWvIxR1Hg@%SgsYC|cE2?U0Y@iZ3Iqh=E=M7T60$dLD2c$A-
zCdHpn%H)BFgia~}vpk3O{oR`A_lPtPz}@40_S}-iQ+HXCv5K&j<_{PRe+#TxQKRKf
z=;xnZ)&<Ff(B?2YidS7lBUIGi{3hYdtS8(NZUWdpm)k<R6d|E2I0j@968<TUFg$&l
z*fSGi0)}A(Nkl(Ymro)2W?{5+EBx4TG8lI}QWA7Zq1v%?r*FAc79+JNh$nQ*00>PO
zF@#5n47^_X?F}WCfW?LO6&Q5Hf+yys;z<qu<wKa5Ap;ou+2Qp@Dr&~02m*(D#|)ge
z;WhpNe;i<SR+#Gvn?{pcEUv_|%-q)JhUt+J3PT{JgdK=$x=LjM9-92ZpDg|jCdA*^
zp(gCb2{D|Pcue<+rVYEiz0Wi(t^<sEPZU+<kq$@y1s=d_uzo(1C&aXhCX#K_rfRTi
z#6ibAq<ybCu+^bm(5SMeyVRN{ec}}i5q6;WhT{J&8l8X7wav48-ST3eBJES%Wb!gV
ziy8-JFX~OWpglpd4pYfdd<HEsWVOtgQ^JD-HUTW*Fgrg>o~TXvx7v&!c6(I97{)i;
zSd<)##4X0v9)>(wH4N{&8hj8y4<xkNZABzGV?wr+Spr1H9P~UGOaWY}SUtv$mywv!
zp{x=<5rXHDM0L5<Dyvw?NaaSlQkQ>#2co<|w#^1|Xm$#-tlF5Em<amaT~u9YoS@vm
zuHq-4^ewm*b4i*ULu~`Cq`J~<Y32EI*pdkC7ai^!rh`N$hvJi0-t#t6&J|z>VLYmW
zEDY4W7e1=8dp_y22_g!-pfNTEn%e|mm>yJx=`Un0xUyKkzZXlnOsK#cQp6QPsIuQ|
zVFfp(2tGcDl_BUMCcdB4*1?;HI-T%80lrI;AICs{gG?B%Y#UAhd8;yE2v6y8nIBaR
z^W=Y%Maeowtx!KSok;}#5De4_(;N!G$1y_+RE7o{5>XPMaL?J`$;vVf6bM{zvIIHc
zSn|2O+;EH029Ak43K8B>96=1KEkO66B4``x1jL*}Xa@88VlP*OZ^q+OgWwR%xI0(`
z`!urJA~Q^oH&)f4M80ZO2Ya!J<QZsO&=mfAGyj<e%ynGj;NnWqOF4S{xD_A_%q0SB
zf-x<g8Ex0bBQ7(hRa~Cite5ft<ssntCvjjoJ;Hg!5x-hzy#>Tlp~KgAK_vpdV8s{m
z6+$xtkmN>h(psTh`GM`aixQ!al?3N3OXj-P#Cmoy-c(E2bAJ&C;fd`Js41KfnxIU2
z<2KY3@M;#(|06c(0h7mc@7w%5sw2dquit4vBcdaL0z|S~_>E*Wx-&a$zC4&RDC9lu
zk-SN_oY*k%yWMAU9a1>QzxPSNcZ#+4;4H6uAs^(kK+Xn$x=I;#e2}+k$=Qkc56TO@
ziZ4cH5W^?~KR~fw#_DQbInhHngyKedn830ijkz?$=7E++x>sNr2+FK75NkQ*6LZoS
zadC0G>m;Dh)83-K<hDP?XL)fRkx8Az7c-y5@v5S!dd)I_7tavYjcC!(>*|6;&N7X8
z%ns;>32_s>8Wp*xt=1L`i}UwQ3qSmI8zP4ng$gDkEB==~`1RdlMx7Vk^kjp3GDe?@
zUlqJj_~Ksv?Wmo45f5duUSIs4({jo@tM=DphcCyz>^!3uU4LwiDxX%vS?Bg|_bXPP
zIHaVbbX%*l%+POk{q(U_BX(VV^PW>TvX;J%&K_(R2+nP9pPh6o%0VdzchfNx1g&*J
z8i6c36OeB0+KLQm9tEbiv~4b<x%Td9tfIdHse?FKyO@Eh@cT`tm{J3aN63z>iSQSM
zM3*2gbKS5;v;+@n?(bhVKE8DU@nZMkU6FN5L1WaR^~jlwfScFoF?D@cSs4VQHM)eW
zL`tT^L2%XqS!D)BGq#e7Tk;IzQXAJs!2#|G0)S>1gL5$Ay;=S9qXwyZr?zi^RuZw_
zl)IO{z<pQ)OJ0K&zk;Jp?F4Cs+(->LN%605vR<R3Hq+-g6*(dR>a$hWW3cF|CQonQ
zKu$7X)Ym4(M*DBEDB~(6ovVGzN>Q}Un6CWgB2=!oH&{HL;$J8zWvIB+i$#w;a#H7{
z<VC@8nnSe{ae;4UBV6zZVHf<CbxqD$MyXSwVVi}wtO{W9_vc({UwLzB!5uX>vyzdK
z(KUMjzgQ9YPE)d21<kLkppZ%!-L7!8$paTz8BK><i_4^NmlrTA+62?Xf4?3`(^#Xv
zsqvyniXMq3Fv@&>E(-kYYT$GS{)|Hu#P~p?(d_#gL7A<=ffB~tzh*mb3t`<a82Lb7
z*I?ZNVg0?kgvw!Ta<V%$x9#rN&4Jc+RJ|i3%NNMK{+acxj0uXHz3^V;_g{3r$SO2)
z!hdC<9bhpP#uM9eE^UBKIHHv6NAr~ba;}sV8`gwXErrJ42Hs<<eXwzCjbW-)McEbb
z;%*{ZKw*a@5S|l#|9%Bva;V<#GH(gA9;bStm*T%;r*LOVaXya$=H5^6z>GzYKM~t0
zT){U8kRn1y5+@g?C2ehO^}9UG=6+_Jqs2`5jUUAlqUH}Z7Vp1QCf3>V^D8{LS|pDP
zL*~N7{%zeQc-RfJ<Vu3~T=MVk((k%gJ;n9!BJW)(<Hi~(&xPS!Fm^M^-Oc#@+oI}a
zq)B=bSp4;36b4gNM0_K8;B6nrC=1RtWi>S^@63M~8ooUJN|fTI$-OH8Pvj|O2oqr;
z85v4loSztq6@+^8HwMqy8HJ4Y1{fRBA0wy6{h2^U6WJn;v^+T(w{jLK@W3wwM_$Dp
zp~@;MfqOqmzt>minSYg*W(7AF><5wYs=in^&`|XE`$?_qlBSZ91yvP*u8%?qP6eDY
zJQlNA_~;=nud5HiA>EFpkgR2}?OC;p(bWEhOh%j8t~}x|(-*{L^6!{>!2DKyz!mos
z)LNaC9?G9aY}(M(*QfmM|2q7#q=eWD#s5Q^l<!feG}&1htSsyQ`jGBFvuEZ-B3Zn!
zmwO$6jKVGf6%=3O*QRoo!BJTxpj9~evd<-NDjr1l1;jqDo(KBRRRdX2-^ltk4NZdq
zm8CVCl#_$j|0W#c3mp#>wvNpC!%qrmF&uF)PSoL8P5#?@tXuEs?7VX8)_QE{gnkW)
zuw`l^cBC1Cmx(n;pr37u6zz2-o1#w!1$}??MgP~={&&}$y=RRv#)i)Z*teP0eU+7!
zyFvZ6v9;a3ajgptlaP@}z&gJDQ<6Xln(cB{9_ZZ}Yvg7NIF(S>KOSbfcooM8u|)_U
zfYL1E(A*$s<Iy7x@W%;HPnSfUJ-p8ewaR*^eqpeQu_@SmPhw)mCnkgi1p|opyp1#l
zg5hf28;`zDOCvZPSa&*L`+)ocrCeTqJ`6k6QLrHLV>U+xq2N&r@KE+(wX%-BesH`P
zI-w32*;iY<w~*@W=?TJy9q;1hf(V#plGYT&DZ-4!$6A8gsLwktf9daD#f1TF#;NJz
zu6>ARw3nxzzAoT~TPhkd9gf60F>J@-!STVVxQ~WbjxmUZt!#^ro9vr1j0gIsyTT+V
zTG~IoOkwGYPcQlV>5a()lk^K1(`%WyG#Q~0RE3@kFp;DF@F^XgLm+5CtcIY{{pSzS
zE~0yZQFq7iux~?{4s5h_Pn=LfZ3Ui*#r#AM`qI~WJK+9^_}?`&6joSx0EP+xBk-A|
z<=F2Qin-Qjt37Zr<ti1-{z<ojvH`CdXIRkRzur|V*8l|I6a%yXmvF}jAs{PC-cpob
z+I#NG75#$`=U$3gnMBC|58MrT1;wKA7%W9mjBN2?6#BV6U>n2=oc#C48imgXe8J`R
z^(n5%5g`(JIjNfM_AP^9BI_{ZTCi`yyxAc)3lK5hG%tAc@oiSLPtD1BhyxA&(qU9z
zU~-CLM~jEYtkvc9_o`Ktm9Hk82m&7$T@-n4qoe+ZCy4?KB30*s-kudoV`Jx^KQJ>~
z1MfxL0TT}%Si*Gl^nN%<(V?E?x+gkb8??sR%|ks=7u@4t7IQ8h9x$WOKCUdJ61^$|
zl}?eNv2iNi*4)I?TYRLGUTVIpal~q@qUndLe1t0e#SJq9k;gi^UQ(bJ`1$<X2b^I8
zlzI4t#N)hLPTk(JGIRshUhaZt3$U~L?V|&Da*)4j73(E6JZl($Wvy+_(l&|h=YQQ`
z?f2W8sc&QwQG2VmC;8NNywNa$?04M{V;|s2wr{nt-4roVV(T-OcrBV{l-j3JaL%`5
zY_ui7SYTVZn|){GReiD*$qOVCa+}wIFGCs6YIWGEpipwU-BgDIREkI7xZ-`bcpu0(
zyxr3<RZm^18I2V>pLP7{I6E+rEuXztf0ym=@=Hgbm?({p8RzVlMh;Z@a@pAO#ii6)
zyPoC$!GUk1KE|bdBZt8ZYE)gIzDwz93G>-+KW0~dq%;OmpSh`KJ_fSg2NVdBJB0(S
zVWp69p?2Kim(~Aq0r;!zuf6;B>OminwQ>+SuN_Y^$1T<}iE3)Xn}{F;@B!n3i;y{L
zVY-RBrlPeq1~3=g8II*P8wkh8qpTae-MH{~df?hm>n*u?8x`+`DEcq%p43m<@<x6A
z^@EdZ9ZMFa1|C0t-If!6xF&MXb5ljmRKsiWV)|=M;p-2vMs<i-p!FU=g^_&$)eOiQ
zXV_e}c?Kz<JoK5WQ?D0RzM+P9Gr4%LE;S1$XVdo4Gp44su%D<wQM76I`BjkdSm)Y5
z*G=9iZNmXFga^kf_^A?unW@3XHDQ$yl8Ah?h%YI6;_tV(n|gWK)7$&}#F(IhQ0J>4
zaR)AlP0lU6F}Eu%a)?aYs;w~1L3iy6hThA?mTNAAhVG(J?qzyQJC;nbX*Q*ztv256
zV=Q4L9Ef=0u%7$kRngTZOeU<D*$86bZWyh%lU@0(DY2T6mfnyjtNjw072!O)|1v+y
z0=%afv_Fz{76y7aToQ9klpXKG^R%=G&<LucjG=u-Yr>AP!tq1Yx3-A9Su9^au!;S!
zg_)q6w6ssQ=Wcaoffq$Zyg^|bmiNCOwJ5&F4pBd{o01Y$-R|8Wc93;9de2{Ub$7li
z;5&duLM-$>hA=$#=4b@~bnV6-5V=i3EcTzOs)FhHJ(aS}-d#P%8J2il=l*d0_VA<e
z)}L2&$3lVDaiD~?y%#d6e50itm}$2`drft9$Zqowdkr@}lf&nN0wL!%$bnk%L`v;W
zTM5`h6EY$ok2uNA513N*!roPfNW|>Fvp8GI?t3w2ud2?G+RqkCX~yCvxe@iBH0WA=
zWiVCaOH+g>!gW6%F}^Xk+1?p*;L_mKMdY<P=ug}G7sN2c0cRN#q#aumSmKT0?_&k1
zR0Bh2M@*4T6IHqIo~pYo$we6tCDd^6)`C_^3RRpoHenxh39wFX--K8TcKS6O*%!xS
zyj$qfA;80^^8?rd1R=-z371vkP&YTuPg?LE)VQTC&~r<`|7Ly~GwyDBo@|@vjA4o9
z(N0tUIojZv(F<QM#G$E;po@Dq{Po9=O<-!F8Q&Cpzv7zthmNi;mr5^Fko-W%V2=JG
z`tM8U`7p*uN4J~-b3wmecwP0A^p!C>T?{aP9i07j7`Q>9p@2F=ou8rjL%Q5E!aj`@
z%4D6oKxcP%IRcAgO}iFKARho5B!Li=0Bsr-ia%JnS)xCNrE~z&3L-IQM{BL(g{JpG
zi+(@%jtIy<8q-YQwA)E=Gaa)<>(?0hUH06#=?Km4FZ~h$m!Lm%a^9H^gK5mXdA|5!
zDh$I8C}$j7lq3TC^Qi4`uN{q}Xhk=pTIL>Gg*n;vK+LFNM<V=Alfl{8&vW2OyqH*M
z{hv{?!~)q5n24=MX*3=oyi6b@1q~n42E<hS7cogmLUV&o3XdETh2D4c0<fEsX}*r~
z2Z6={6E9c_2STtzw8vWZJEgO;i;%O%QfAw_CuQ0#eto>d)zWI!eNQ#)$c<a0PVa`Q
zii(Rt{3<?Q7Y>X7ww_x%H5uwgXqo|~oF+gP)eP<3U1hG@Jf{f@2@|`+NexF+7=+Sl
zuquZ8bryP@-y0h*GUFX4rwE8q_vmQ*NQ7xp!yD|00>L>sJv}KS<2LHw6FNFqOwy~`
z+v8yTvd6Xm1U8MObMrUaNMn5_DEimn68qL@Cx-3`FLF2`SkFC^n>k%L`ZCwRU~D+|
zbxU!y^R0tTos#R<GcX9^-G-DE>w`no5+DoagX=J6d9Wgclh_0l!=D(p1zPK)@yP==
za~fbJuY{`X79Z+7>Y=rChua}(heOAN+KTXwKo<Qh2r7VMtUt>;@8af`2w3h*&Jrpd
zCVVO;bBzuy@HoDWJG%!17;sJwp_)Nq29eTL)N*iGa`22+OSpp|0=On1E=~Z7iVnMr
z+vbzBv-~I$CAFX8GhZ=DI*z(;(uhWh!0znqOzawP6ad#$qq--S2Z%}6-YH>f0Plm$
zv)Mv{)}z#)UyFT;IFNsE<hKo5S{W7*^(|xeg2@$G^C5RQkVWQq)rMKrZ_%*84+}y$
z<6hd4fp35^=ym>EL8nX5yH`|I5xEYAm4R`kid^7tzE1mKlci%~;-@~f2UQReG$L&q
zl=#TuBC8r3H&7v6VkbB-w+BydE8lQW#yuefjvlOr%TY`4I{cmzQ^oMw)!m(+LqMSR
z)#((AI~qbNS%K-u*j6d>`X^2V98c;-x4W+QVYWJ`4M`YbtzN_V)?oS_=s!ez^x>Ak
zEfk#I#U+)LQ?}Z;uUGU^dQc&SH!Th}rSbqQc*w5H;9R&WV}Pw>gqN;^XY1Ca%b+kr
zzU)@GqHigrval?5aZK}cYA;gPm&{@KAa$JYDFj56J>-u4UGS@mEh{g-2K$H;Cr*45
z9!=0qMsZRO5A2jiZ}lL_aJ6eVKDb%oy`oi>m7svAf&TyGKBBi!z|!>bb!Hw$b1pO`
z?k+CeoP2y=AU?yhyR+vIgL4C!i-PLV?QU#LVzB)~`@(_(bVbl5l0i5^%Tr(fXD=Ix
zYy<qRafnbSL8=8yb==Bd09;`m9aiM?jo}{-yef?Wqz+2(z5>&93O3!9V|Kl5YvMnQ
zA>IAZdhy1g>CG4i!Rw>A+uKVhFmHpl{J6@s7R<@MH#P0koXcz(F00HgA!<K-$urNN
z_+BC#RwYa3gaMLS0&?`3sNRYQ@%+ck0`3-CbSV2>)u*#?(>_~enV%aJ&;+Us>cZ`R
zcYSq+sm9!(r)$~yViU#vK+8d-KnIuDw?-(c5_MA3(@#%~QhUL-$}KX+{vz0)cISZN
zhgq&U>LX;$>`_Us>?m~gNFJY1bT-4N3<9#<?fxspn@_Z@Rn;N-Iiu8v5D&EKEWO6=
z$$)?WI^V8fqP!po9~QIU!y9mBRR(_f1LT4Bk>^zITfaYVq}hgMr!ymC7mQJE1_uXE
zdE)KEqd5n|zKH1(4DNuJVQ<*%L)x48mUycog!y5viee`V%zK;|MkXd3%A_6|Aq52m
zQubiCAMn(2<Tuz|&&|)bj&U0U`Rju60`MhvEP3N*5x0J$ok&vpc0;fRMdDDLuh#xu
z!J+*2GAdC#YolzZ&i7lwDs>aYJ{@N?&0QJy1>E=nj+Uflz`m8|Nw|YCdMTDrBuBKM
z<H7lCciA;DJ4^I0sI*)#SLx{LzxBTX3Z9<x%pXHU%NE~8J)BDtfY4(=*Z`>_1Pxei
zRGK!O{KuV`JpdjA&R*AY-nj?o2M+#ZITl8FXs_NH4QtsCA$6laO1XX<cs|(p#N!OF
z`#<>1=mGbmq=o^}RjXEIz@v_!M!{=<!2?tQq0aTbA=n5=M#!EH!;w>oOdja)&4C3&
z>2VVKWW@r9;lc_sId&3#AEAiveS8eeTo!)*uNa_$7`+n`x3aRbl<OJr`GnC2YR^-J
zX)Z6tn84#VTaKUqQ~jfyb(%Hvk0Gzz<v{zI-0`KlQ8`n$*o=-09;VwU*?~HzT$l9l
z@PJiw8O9XjQ&acZPcjHmAA7H#a)SC2E^!M0CbAJRE-r4$4p13n@|c-TEX}qbSf-MM
z2?D5S_?R<4bI>rumkw=u7PLM{9{3hcub1590*UW6GOR9oZY3q9LIK80ND9Djv2pIw
z?~knbbTwC3*Lw^0*+xLa)zT?@0I%=(df7$;;|{!F=w6l!)S4ZFARwQtjJxM2!FTD(
zrpvGRh8$B3;xf+1*r|FQekb1ydFln(j9x0%XRH)sHyIfjE*Nnb7&Jd9T|Bj&Y>0tU
z2)(<@c_twT3YGh_1iFj}1E<5Q7c~>=<cEibYM>h%pPh~5-x+A_tbK6$R00I-uMHc-
zTSkBfqdkCu;1eVzq7jFZquuO2+#%#X)%(wN1|u>9z5je>yHg!E9`r`YQt_hoKn#+Y
zk&#h%3B^D#h9}tn7*Qvzssj_18tEOHe!#(i`0h30Ny*<9`u=XCrdX)t`-#M!u9%hC
zgsm3?+RO|uKI_un;In__WL3{|3cFtV1-^99B_2u`Z{PZyr7MS>hv#xWLsdH@7`A-l
z8xM4Sq$o;rf8oHIY;t3roHC{?gfJ{+WF(vXFwl@&-iYnSNkEo?inLD5l)rxuhEgBm
zlo_B(q6n2~^St)+rx83Z`xy)))!##xD(AN-i^WtRfYV{qle_Vg%@}<J1W=VMPG}Ae
z4dJe-r&AKd2$2ybIhX@o8>b<NBFF~@a##@f^S9z#ODiiEAYmwmC|#E-+=p9L3+_f(
z=CA%rV@hRMcIs(z&qwbo+f9QSSd(@<ytN{P)kS3Zb?n&Z(-fIYE-vRW>sg$LEWJ-+
zSWCV<&BMEQT^zB?#>*@+Fnz}Mx>2fM9Tr&wfNphukP*mYpROIJ0I6c(JCQ4ZPJoUj
z8Rm=*4r)MZjHaY$y6FrW%!y#d#VgxRzBr+;UxkW`Z}O`IJxBDH571*Z!v_=OQS9gJ
z93F0!Sfh6~Ht8)H(vSSA#&|k8Elm{M9OM}w?8BaEFv#4#9rkMnvS4(C(iIcWORGMk
z9>yZ3Z`&v15JlHSN<MdR=37OjS{50wT9)pn<6k$8DCLe2CE@bT>GS-c=&0(TT78Q1
zq3;t5JO~{VSzeGC`(O6roW$>04oXVeiXTZtc&NV=#FF|IJ2f#h1@n)07Bzc!Pfw%W
zC2a&_?6n01cn@{RWR>Fb9rVs7e)E>gb5)9LV1L192j)|MYX#;b)yj#J7XT!pMF(gI
zms5Fpc?WCe)0(h^vqByK=mxqX{Jemm5rAJ>?rMK;uNnsAc}|@)Scom{<jmm$%mcc<
zI*@NT+xq~Q-5I_cZvVGZgj$TImqt3$&o3c_E0umpa{4B()I@>0+OYcpWQ78&M~d!$
zHWHv`%NCX68bTHJd}W*xXFVN#eSL!&&G;<8Qd(d;Bz>>C0+V69v61{c(6mFG<bBNs
zPE*rR@qF6N-E<bXU_R!nL|!E(AwfDc;?Rw+L@H<;LqaBOwhjr$v=P|`zxsS9niQh4
zZG3fF>ua{n8e~umi93J)4vZHAQ1#@=lQg5w6RCQH%1sI_l;OlgN7id%3$D-pfp43?
zo#J#YWu=c<yJJFY&sgUzliN&x$Gu~D7q(ova)q*_fGWIjghnEga-AKe&}kSU0;W5a
zUsza*3d-ccSfj$CVWRHRC$Ql`fY>3Sunh<=VALR>B3bWzQHsKVB=_S2T0C$r`Ii?;
zS9p-RV!a2qp=^3AcS#gJV!Mz(y2&{Pb#R?p9*63)v@~YC6xc!VNXC($csavh2?c%5
zMINAw*@vFp9x_g-YiPKKCK5gZ=q)awOIasTZnsS}DXb&7a`ge~d(>P{ISqXCHwU;E
zaF%z|i7M<Lzb<i0(R)e^hMEMer1+%8i{)1Cw^n9wrltPV8o^TLFk8dypRSv{j8yRf
z=Y0SR3a3s8sT6+0dKk>C`4W!)Z}|laQrr2kV(bQ>p`{(bRxN;<Bxrz9HxB!ur(#TS
z*{z+&)6qnVwzRYaG9fc(j$yWrsj2q4b3!C*CnX{Bpy-e(-ZhrWTLVcaj0tHaIC<dE
zUYJIa!0a7^PU4&lZ98#yGchsQnbXt_Y`FPx4*Np8->K6ReouLifY7}9ssAzU=8#>n
z{3}z^v+6n=sN-dQ=63DSdo2h@Z796v3M5i+>o1@|({<Hgs1rgD>w@|;1#hctJZhGx
zmN_<+M*Bs%MU6z=;$egvAh#G*)b9ym873~BoA$ekp&@gQL+1jQ75m23PFSf&9EnjH
z5p=|J%nZ#1S(#0g^jHIkTk^J=GC1vxq6B&9mm9@6b$_rm7>$6#=~Q*{VEUVGTLB;K
zdl*!cZB^<Q=g}h)b;07&Qo5|;A3Z`c`LQiaF3yv`#%|4`la1+ymls~uef`$e-CKUJ
zcoa0H-q4o)a2q1dTYY^e(Q?#TKSzMU%zOHJ+Mh>w%kan}Qw{~P@L|>`ig)ECtX53)
zIeJ#M_+-aY0z8b3jmZs0PgT!@tF1>`^cv1y5kHMOetc}q?C0ByJ4;4Rog^*!*RP`J
zI>FW>#6!F!>bPD+6+X>};++^kg_nw-+K%2Fcq+MGY)Ic{VE$SJ7knLaS!gvt=IBZ*
z94Y91vxLEp1=@gcPT?G*RN`HV+)J!L|ES6$=0<zii4H4q&#YvchiAsu)E|;kQCGd>
z(O$^Nz+gR*4ZI8c{chkWgFSHrDkpNugWm^Uq--de{CWTYZpX+7fnCXfGX2bi)UI+I
zOz=K(G&bxYorB`SkJ=;0kGubB8gK=3XS~_`R`!4&qVT=QNOg>`(9L7b1H`~b5X&KG
zCOj4UT{a#+dg?Non+?$8emftT^99;QFd#d+ES~_kC>)%mq@D?Gb-3{TrH;N2JRKN;
z(7GZDtIp&Dx40I(Uf1wUNbtLBu?M~PZF-7X%lt^;sH|&03qU4R!0ZA7{*?jln}P5P
z_>3Kju5Cf&(PJ=6-xfL%8<_rZ&$_uJz~gwq;!R8a>K6i8hE444cjB26lb&@npV5e<
z%SAOr!N^V)*&mz)40qU2)Fa@?p>Mx7e=ah)(FWB!jvm<$YAAVBeEatEsQDwt`UVF-
zL5_r*a5hc!$Ir<IhwsRjWUB;9eex(563IfCEtuZms8?5S4eZZ|;FWui62-BQ%Kbo4
zdS8GxCR&3qX1}QKpTQthk&SnfL|~-XzC7)u2C|?Wc*T#A847&U;5b5zo`A<<K3ztk
zM1sjlcbcJ7F>(tXJ^dbt>TnJm2Z~|M1ojoUMo6=~=cZxhxD~3BYTwl4WGtek!;B0A
zpw`x6<5XjJLwE`T+a`t*5GU?in89BLwGnUNi%PZGK)eqIRqIW)X)oc6+MxOP1+sg6
zOCU>b47w})V1r;0%6oK$#8K$ybO|Oyq~4X1l0tj}@`L$Qt5_h$(dW;f_w2Yr@8s@Y
z(m9))lCoJur1A}?;nAZZ-Mi5V!fNXS32Nfv&(ZXwoIu5e!5Uf+<htb7uVLiM&cdPu
zu|z-Gd~U54uG!So)SheW6*}M@WdG(<7-83-8w8O;u9JYqTYOsYhm>(ak63I`eb;@3
zK9oiI?3b-^KJT#xU~psIm9T{UM6>!VnBOKcL0W6|8z2t%(9Giv!RxgyLOrE1zF~l9
zK5s-sl#C2>a&g_mLJLGx)KypoNQ^iinbRS>lPCt03{NkwbGtU;3LU;X;1h_mAGizH
zrlE<RD*DzcSpJeWjodZi7D;h&3-}a)`}^zL-*d|U`VordE$V8zYlmX)-c{Dsy*nj=
z3c(Qr+r>W@6cb8H+MMH#CvAq6G$Xs{+XXpI#rq)oqZ)!cUztO$dDahZ-ji@C#p_)-
z_|pnqKG2stAS5BVgDF+iy*u0J3u%C6XRBIT7GAs`Ahja2;Anx-tQx!VgvQINH6lqN
ziaE)zOvHbe*%MDchWm@fMK{zQY>eK);txQeX!-9DYG~u*QbH9myk#b*IeL5<O=TR!
zZ=c__aW5K5(5!U5AF#xiH-m$M;T~|DPtI|ADU~fP(Gcf!0<}HErwwZu!gK?o3Z6yP
zkw<zds)@SzGi6)@Fb?Z%BlW~e!1bVDpf=z?=#Aa*;#lt2{!|No$Pc0=^8F0aJRy|6
zv)((=5V!oct{W<rUL+W#WHh-al20Kfaj3p)ZRG=OjGx6rl3$+dmdRdE$U7_3rYntq
z!t5|tBnDJ<>86TsBI1$DC$G<|SANgLKHZ$jyA|dB)7?Bk70=}W_`nE23IkfoH<%NV
z)(A|4+F!2s_>W<CL?J*rl@?O1E}fxGzyFQFI0)pobt^rX&0nm_9dZvuqyX6jAPdxu
z)K+9zJnRmL#0$0xq{IQe176y}nAlQtNmfIIP*Gmq{cr}cfEAb-9yV}>aE$`Y{s_~H
zV9;TeI&x@Ob<dp9#1-PkVZ}@5^(L7eVq#`Eu%sw-Y*&CDrQ!8Z0sxM2EJa~sVJX9K
z4LTYZ7e=8Gga1&72Y2sEW#a$SSbMOWwV2Np2+ZZaGyC=hMkp{DKH_~BkJRtzO}+yU
zarp6GuQ^-v_KQ^(*$Ihy5i~#R@yeUg_mENHvWiL?bp`j?>vM;(F$Fn*c!Ypew^6|-
z&%AB`MQs!i0xB-a18d>NyldCk!&xm5^x>xoGzGl7v)5|qwf>E5;B%y_r*}PG3}zN(
ztR;Azfk1q6+iB#3Jxst34d9Z7(2c}FmUqHV!r9&ZLv8Ko!&|C494G=S?}}*lLydsu
zE^nYlD2OFK!8f-J&IQB!0<v>!nAK@^z}smVTf;6+)UyZ$_FJPH^je<pGt99wOvnK5
zC?A|W9F|&potxdkq%i`$&I>>xru1N>f`EeI*i&4y{#q&6N{J%0vDS|i3`u~4Q)jo3
zjcy9hJI!<cX}lOlaj>t)<V+a!HiYs9n2V%1<Ra+do?<pZc3lYxJ$py<{l6ZC5V&&R
zu3(HWZc_9Jc^k%K{tqQhBi@<aW<Sax<VRd9aelK4HCb93K2n@IctEkiNKb$y>7(ha
z%D-3<LqYTr2NjAq+UhuUO^6x5bpUN2l$31~zml3bczHj=J`wc>*_N>Qd#hq6>JDLf
z`NdZWUN}`&2(J+>+1c6P$ff_7*@2{n7oU}nuhru6t5e(Y3v@Vw@Lq%43w?SI-ZoST
z(%Wz7Vu%RH!4QuKS#fbHNl^r_;HOU-&{usXe}EP40BKOLgE{PZGT{_zW&*Fc4(MV=
z;sx$3_BE&7^kfuLVF$SUvxmXp^8AJGmjfE%JUs^Y=-n6Vw+!r7zHvYoq<F`oaXK<f
z<6uC_#!(^|92=X<8D4E?=iTVr0cHL!koZAMP;2;DJN*KAgfyOCj1)kjT9yHih*%;~
z)6(r_I)z%qXK9*(IP~A1C6qKGkbfc)VMi(CwLPfd>{<&R0M7^bcnr`i6sXS&3b<pA
z2g6YHv%s~mgvU^;lHVZnwiGn^b3h-!Zo(4#GVC?@gh<sB$f|t9*4lBi*|3n$I+`X3
zG`sOlJYq5991y`}wV`VC>I~unbJ3CGJ!(aqPjXrO{B0w*ERjKo*(LDPg@RZ|;<Fy|
z?B&ak&Yy9!<;N@7zvO7+A*0i!$Cv;lbc~}QAjV%l4k{TW8;I+yC{<|8VD^cHLl}RN
zG7$qRL>BkL*73(+)6><^pA9*llz<ZI7u*!|ePc0f2P6Q|Aq3QjYv!nU7i(7**CW>+
zXG66(PBjx6FDCG-y3>{_13M)02nqqhfU0*ea<EJ1`_;tuTVZ+590!dDgNomQ)fVS5
zi~@lmu~o0&4EE3B@sLWIvJ{0Gbad%rkK{U<8t~a%TN5r&i1H7V+ue_kA$6e6Cg(6!
z5A-B+=EfdP#i7AL-qnXB^j(U$K?Z@TDF{|ip=~4vmg|}%VbTJS5xmlmYLRA5O--4I
zk2usJjW&s(DZy@!BpPyj<#q?MbnRm{v)@$mmnIo6+%ODE*$L}>iV33@V$e)v7{qCo
z?=-7F?|Ukc_WNO_2g{RTn-y#|35wgcUDT0^ilO?b%Ar$n0&xJ;-AZ4;of^MJR^zMf
zUT$cwYQR^GvWz1nB&<+!3qNIC(^Ef<(+ZK<eXI_-{peBiQ-%-dlgkmLepsIyu8BKi
zW+sTUg4W^^HsVzMynqT7w>XGma#^d_9mj>9!OVU`i8oC&Dz^cd()o25PQ99<!{fOQ
z;7eg;@e+9mkUa1l^vmBLpOv)-_@(IX-Qy1_RW{713CSj_(&g!NnW+L)X8Rs3U~d=^
z<nW}{x$Wuw8lws82J_whLkMRONNxrv6B9Ca*lU8tfJ93<9pZvggydTYO!of$1-1hU
zj3(6~825o}DNj0p)Q8+G$~R_4P>7Rv%bv6Br?8J3{XU*>wOZ&NB8lC4{Te(4y6<Q;
zP<<77O@cA+aj&+{dM|Kmpf@+d!$tP*kHY?ueffKxL5{_%;@^oaRwd6UeBf-O8I{Sa
zMK=SZ^UjC8@-Yg&QJ~|s`AljdV<TT8QQ>1C9g#h7z<k%nUEJZo(^2_>#Ay4_Wdsm!
z#}|x3;ZH^m!!6XZsNl*0j!H&yF$t**t;$%@<LD`m;4LX2gph>U9qC2yNBw3G%ufZe
zbaA2<$%H-;UV)^exL2DmFZ@D`31e_!xE#W9$SPXVO&Tz_|6;cQIObz7I5n2PIQo@?
zQqnH?6bKcekYniT=y@5~0Ks5A+Ps1hJ_^1l<f8w*jkQWY?7%((v+ENCmprf6hz7F6
zcdyQ9T5D>2B^7@%ixE;^t^abC&!93x-A%%}a&eJD;dC)msR|DG%&I!5T164lF&qJH
zo&i1Mj9W#?(##nUCu$-kO9#egfCfQ2-MixBZ2U^X6U1jMpNbnQ5eLwE3pKG6Xxe}P
z1^Lt%gti1^dsEq_jy0}sw%W_R6bC%1&2Pbe{w-2(5E3TI$%v`IRS1BdH~32wlE>po
zAbo%Zj@!N&qI+=X$zJA(iGnmaEyaiMc!KK%YRh*G4U;c!5tRrEpiCTUH6b<$iRS}D
zLz-BHhf;PO(@FdasLY9pH<2dcHY2reYm~rU(V{W14NSkG`EEx)%Aj8C(R!e5$RK2-
z1CS87kzI4f75r?~Uqk&RmlMfg$N#T@1mFq|05)r2#qIm{_La~pn0}E04Dd5vhCNX|
z6D7dfh#L-`bNbwqngv^_#~|h5uLDkm@g@p7(##Pod7GIKzKISM9JWc0GsVl3&?4vK
zR>D!FOepLa3<$-gu_(WNgH{=h6p+#mw5#Y8;8a~b<eh4g4o25AF(#xIxZ#7`ABy#G
zSX2P}MN~sC(H3EB=s1&Kj0)lczMdB6>pV69&3?0m?YlA5MVxe{!gs)FGm`QN0<9(n
zNu~;~R?o=|7~H-pD*DA*ArYL3!4<6RrOP*Vyo~qsqVBp7>vbIjoQ@X~`bGRe?sygb
zp5QyeWb+H6G0*E~Jt%rn&l9>3HnLru$+tsev!A%!yTFDg$C}DEd2E9ZXh(`z=s;H0
z?wnaIj+<eL+jHmf<7!YHWSu&wz{eNev18~}I%cH@FetG;C0vqt`qZi0adEd`U6-nd
zfxr%O_yVmxs9%e)=}>W{aJze<y_{Q|tQ8@}0=+XRRah%jfh(I|UKj;7<@|Xy{8o4{
zK!2bKC_woLm2Ki!PgSK`Q9%S$v1Ed-#W}@+Z;U^@F~eCFwZ~_kDjK{Ic=}&|gEKr&
z(I$U}A6S)P#cXO)CJJiT65!Hv5PG6cn6LQXpR(loQXPP3Qq#R~se6W{L_XjYe$W<A
z5!;x)zZnC{{qL8!22$x68!yGji>=tzjP)?vLA3}}XSgPG!wz1{;GqBgSo~(Z0cJ?l
zxx8f#B0l*vLoqQi^q4*pBZ{MjrvLx1B=K+FUx;ZiDXhMJ{W=ly?>CV%DQ!L{MU{f;
z`TB>dDl)%9tny*|?{W`p3n+dn_UZG(wWeRoWDpgP5;lNoDL)q%*ZivgE{m-!YPXLZ
zwdoKHrcg;>l-SkVE4owie}8<Ns_AB@COsKDuLqWp2@M;W{Z{^W8Ej=~djgEL=jv&~
zroETS!7{P}IRyNdix>ZQ1r>awc+E-gD2IE8@?+^0^)cUY`-=e<F~?_}kj9ae?ska$
z_hlkcw-Cp@2Pr8-RW{k)KO19H3e?7;(#bRugHuc+5j7Lt{`>V^>VCUuW{O%J;u^nS
zO4~b;NT&cn3_D^B;zSrYO#l1XysH8?1C~yv>S-6ZjV013G}>fq|6N!hV^|+w-$d#E
z`!r6PM+G4r(kC%h>jrNKEenF?ZN2|{d(NyiQ;eOQz609BJ2btR)?gO=bMQcl*ryGS
zrUQM^zOO}c>h47JxVL=x@^&ev{zd8=-;0iN>wn9WhrYZSs3Bz90B~Ly$I{M}k4^KB
zF$G{@Z>nB<J=a2Z?B$)?AG|s(VEe`X*XPNfe~Y8L-G-fchR2IlT9yyB@Q$`dwdDPn
zwHX<1|LFI&_j2xxIX+f}efyPnN*{Vk#a8H!MY^;8_w21am^QwmS*BSWVba!38=L&1
u@ZuL;OJw)`|L=cAkL}58q<?8*g}UkjqrR0^z8e+(IjVYGB~97v>i+|+b-vyJ

literal 0
HcmV?d00001

diff --git a/Documentation/RCU/Design/Requirements/GPpartitionReaders1.svg b/Documentation/RCU/Design/Requirements/GPpartitionReaders1.svg
new file mode 100644
index 000000000000..4b4014fda770
--- /dev/null
+++ b/Documentation/RCU/Design/Requirements/GPpartitionReaders1.svg
@@ -0,0 +1,374 @@
+<?xml version="1.0" encoding="UTF-8" standalone="no"?>
+<!-- Created with Inkscape (http://www.inkscape.org/) -->
+
+<svg
+   xmlns:dc="http://purl.org/dc/elements/1.1/"
+   xmlns:cc="http://creativecommons.org/ns#"
+   xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#"
+   xmlns:svg="http://www.w3.org/2000/svg"
+   xmlns="http://www.w3.org/2000/svg"
+   xmlns:sodipodi="http://sodipodi.sourceforge.net/DTD/sodipodi-0.dtd"
+   xmlns:inkscape="http://www.inkscape.org/namespaces/inkscape"
+   width="447.99197"
+   height="428.19299"
+   id="svg2"
+   version="1.1"
+   inkscape:version="0.48.3.1 r9886"
+   sodipodi:docname="GPpartitionReaders1.svg">
+  <defs
+     id="defs4">
+    <marker
+       inkscape:stockid="Arrow2Lend"
+       orient="auto"
+       refY="0"
+       refX="0"
+       id="Arrow2Lend"
+       style="overflow:visible">
+      <path
+         id="path3792"
+         style="fill-rule:evenodd;stroke-width:0.625;stroke-linejoin:round"
+         d="M 8.7185878,4.0337352 -2.2072895,0.01601326 8.7185884,-4.0017078 c -1.7454984,2.3720609 -1.7354408,5.6174519 -6e-7,8.035443 z"
+         transform="matrix(-1.1,0,0,-1.1,-1.1,0)"
+         inkscape:connector-curvature="0" />
+    </marker>
+    <marker
+       inkscape:stockid="Arrow2Lstart"
+       orient="auto"
+       refY="0"
+       refX="0"
+       id="Arrow2Lstart"
+       style="overflow:visible">
+      <path
+         id="path3789"
+         style="fill-rule:evenodd;stroke-width:0.625;stroke-linejoin:round"
+         d="M 8.7185878,4.0337352 -2.2072895,0.01601326 8.7185884,-4.0017078 c -1.7454984,2.3720609 -1.7354408,5.6174519 -6e-7,8.035443 z"
+         transform="matrix(1.1,0,0,1.1,1.1,0)"
+         inkscape:connector-curvature="0" />
+    </marker>
+  </defs>
+  <sodipodi:namedview
+     id="base"
+     pagecolor="#ffffff"
+     bordercolor="#666666"
+     borderopacity="1.0"
+     inkscape:pageopacity="0.0"
+     inkscape:pageshadow="2"
+     inkscape:zoom="1.6184291"
+     inkscape:cx="223.99599"
+     inkscape:cy="214.0965"
+     inkscape:document-units="px"
+     inkscape:current-layer="layer1"
+     showgrid="false"
+     inkscape:window-width="979"
+     inkscape:window-height="836"
+     inkscape:window-x="571"
+     inkscape:window-y="335"
+     inkscape:window-maximized="0"
+     fit-margin-top="5"
+     fit-margin-left="5"
+     fit-margin-right="5"
+     fit-margin-bottom="5" />
+  <metadata
+     id="metadata7">
+    <rdf:RDF>
+      <cc:Work
+         rdf:about="">
+        <dc:format>image/svg+xml</dc:format>
+        <dc:type
+           rdf:resource="http://purl.org/dc/dcmitype/StillImage" />
+        <dc:title></dc:title>
+      </cc:Work>
+    </rdf:RDF>
+  </metadata>
+  <g
+     inkscape:label="Layer 1"
+     inkscape:groupmode="layer"
+     id="layer1"
+     transform="translate(-28.441125,-185.60612)">
+    <flowRoot
+       xml:space="preserve"
+       id="flowRoot2985"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"><flowRegion
+         id="flowRegion2987"><rect
+           id="rect2989"
+           width="82.85714"
+           height="11.428572"
+           x="240"
+           y="492.36218" /></flowRegion><flowPara
+         id="flowPara2991"></flowPara></flowRoot>    <g
+       id="g4433"
+       transform="translate(2,0)">
+      <text
+         sodipodi:linespacing="125%"
+         id="text2993"
+         y="-261.66608"
+         x="412.12299"
+         style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+         xml:space="preserve"
+         transform="matrix(0,1,-1,0,0,0)"><tspan
+           y="-261.66608"
+           x="412.12299"
+           id="tspan2995"
+           sodipodi:role="line">synchronize_rcu()</tspan></text>
+      <g
+         id="g4417"
+         transform="matrix(0,1,-1,0,730.90257,222.4928)">
+        <path
+           style="fill:none;stroke:#000000;stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;stroke-opacity:1;marker-start:url(#Arrow2Lstart);marker-end:url(#Arrow2Lend)"
+           d="m 97.580736,477.4048 183.140664,0"
+           id="path2997"
+           inkscape:connector-curvature="0"
+           sodipodi:nodetypes="cc" />
+        <path
+           style="fill:none;stroke:#000000;stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;stroke-opacity:1"
+           d="m 96.752718,465.38398 0,22.62742"
+           id="path4397"
+           inkscape:connector-curvature="0"
+           sodipodi:nodetypes="cc" />
+        <path
+           style="fill:none;stroke:#000000;stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;stroke-opacity:1"
+           d="m 281.54942,465.38397 0,22.62742"
+           id="path4397-5"
+           inkscape:connector-curvature="0"
+           sodipodi:nodetypes="cc" />
+      </g>
+    </g>
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="112.04738"
+       y="268.18076"
+       id="text4429"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4431"
+         x="112.04738"
+         y="268.18076">WRITE_ONCE(a, 1);</tspan></text>
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="112.04738"
+       y="439.13766"
+       id="text4441"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4443"
+         x="112.04738"
+         y="439.13766">WRITE_ONCE(b, 1);</tspan></text>
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="255.60869"
+       y="309.29346"
+       id="text4445"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4447"
+         x="255.60869"
+         y="309.29346">r1 = READ_ONCE(a);</tspan></text>
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="255.14423"
+       y="520.61786"
+       id="text4449"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4451"
+         x="255.14423"
+         y="520.61786">WRITE_ONCE(c, 1);</tspan></text>
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="396.10254"
+       y="384.71124"
+       id="text4453"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4455"
+         x="396.10254"
+         y="384.71124">r2 = READ_ONCE(b);</tspan></text>
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="396.10254"
+       y="582.13617"
+       id="text4457"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4459"
+         x="396.10254"
+         y="582.13617">r3 = READ_ONCE(c);</tspan></text>
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="112.08231"
+       y="213.91006"
+       id="text4461"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4463"
+         x="112.08231"
+         y="213.91006">thread0()</tspan></text>
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="252.34512"
+       y="213.91006"
+       id="text4461-6"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4463-0"
+         x="252.34512"
+         y="213.91006">thread1()</tspan></text>
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="396.42557"
+       y="213.91006"
+       id="text4461-2"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4463-2"
+         x="396.42557"
+         y="213.91006">thread2()</tspan></text>
+    <rect
+       style="fill:none;stroke:#000000;stroke-width:1;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0"
+       id="rect4495"
+       width="436.28488"
+       height="416.4859"
+       x="34.648232"
+       y="191.10612" />
+    <path
+       style="fill:none;stroke:#000000;stroke-width:1;stroke-linecap:butt;stroke-linejoin:miter;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0"
+       d="m 183.14066,191.10612 0,417.193 -0.70711,0"
+       id="path4497"
+       inkscape:connector-curvature="0" />
+    <path
+       style="fill:none;stroke:#000000;stroke-width:1;stroke-linecap:butt;stroke-linejoin:miter;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0"
+       d="m 325.13867,191.10612 0,417.193 -0.70711,0"
+       id="path4497-5"
+       inkscape:connector-curvature="0" />
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="111.75929"
+       y="251.53981"
+       id="text4429-8"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4431-9"
+         x="111.75929"
+         y="251.53981">rcu_read_lock();</tspan></text>
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="396.10254"
+       y="367.91556"
+       id="text4429-8-9"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4431-9-4"
+         x="396.10254"
+         y="367.91556">rcu_read_lock();</tspan></text>
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="396.10254"
+       y="597.40289"
+       id="text4429-8-9-3"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4431-9-4-4"
+         x="396.10254"
+         y="597.40289">rcu_read_unlock();</tspan></text>
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="111.75929"
+       y="453.15311"
+       id="text4429-8-9-3-1"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4431-9-4-4-6"
+         x="111.75929"
+         y="453.15311">rcu_read_unlock();</tspan></text>
+    <path
+       style="fill:none;stroke:#000000;stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;stroke-opacity:1"
+       d="m 33.941125,227.87568 436.284885,0 0,0.7071"
+       id="path4608"
+       inkscape:connector-curvature="0" />
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="394.94427"
+       y="345.66351"
+       id="text4648"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4650"
+         x="394.94427"
+         y="345.66351">QS</tspan></text>
+    <path
+       sodipodi:type="arc"
+       style="fill:none;stroke:#000000;stroke-width:1;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0"
+       id="path4652"
+       sodipodi:cx="358.85669"
+       sodipodi:cy="142.87541"
+       sodipodi:rx="10.960155"
+       sodipodi:ry="10.253048"
+       d="m 358.86939,132.62237 a 10.960155,10.253048 0 1 1 -0.0228,0"
+       transform="translate(36.441125,199.60612)"
+       sodipodi:start="4.7135481"
+       sodipodi:end="10.994651"
+       sodipodi:open="true" />
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="112.11968"
+       y="475.77856"
+       id="text4648-4"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4650-4"
+         x="112.11968"
+         y="475.77856">QS</tspan></text>
+    <path
+       sodipodi:type="arc"
+       style="fill:none;stroke:#000000;stroke-width:1;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0"
+       id="path4652-7"
+       sodipodi:cx="358.85669"
+       sodipodi:cy="142.87541"
+       sodipodi:rx="10.960155"
+       sodipodi:ry="10.253048"
+       d="m 358.86939,132.62237 a 10.960155,10.253048 0 1 1 -0.0228,0"
+       transform="translate(-246.38346,329.72117)"
+       sodipodi:start="4.7135481"
+       sodipodi:end="10.994651"
+       sodipodi:open="true" />
+    <path
+       sodipodi:type="arc"
+       style="fill:#ffffff;fill-opacity:1;stroke:#000000;stroke-width:1;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0"
+       id="path4652-7-7"
+       sodipodi:cx="358.85669"
+       sodipodi:cy="142.87541"
+       sodipodi:rx="10.960155"
+       sodipodi:ry="10.253048"
+       d="m 358.86939,132.62237 a 10.960155,10.253048 0 1 1 -0.0228,0"
+       transform="translate(-103.65246,202.90878)"
+       sodipodi:start="4.7135481"
+       sodipodi:end="10.994651"
+       sodipodi:open="true" />
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="254.85066"
+       y="348.96619"
+       id="text4648-4-3"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4650-4-5"
+         x="254.85066"
+         y="348.96619">QS</tspan></text>
+  </g>
+</svg>
diff --git a/Documentation/RCU/Design/Requirements/RCUApplicability.svg b/Documentation/RCU/Design/Requirements/RCUApplicability.svg
new file mode 100644
index 000000000000..ebcbeee391ed
--- /dev/null
+++ b/Documentation/RCU/Design/Requirements/RCUApplicability.svg
@@ -0,0 +1,237 @@
+<?xml version="1.0" encoding="UTF-8" standalone="no"?>
+<!-- Creator: fig2dev Version 3.2 Patchlevel 5d -->
+
+<!-- CreationDate: Tue Mar  4 18:34:25 2014 -->
+
+<!-- Magnification: 3.000 -->
+
+<svg
+   xmlns:dc="http://purl.org/dc/elements/1.1/"
+   xmlns:cc="http://creativecommons.org/ns#"
+   xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#"
+   xmlns:svg="http://www.w3.org/2000/svg"
+   xmlns="http://www.w3.org/2000/svg"
+   xmlns:sodipodi="http://sodipodi.sourceforge.net/DTD/sodipodi-0.dtd"
+   xmlns:inkscape="http://www.inkscape.org/namespaces/inkscape"
+   width="1089.1382"
+   height="668.21368"
+   viewBox="-2121 -36 14554.634 8876.4061"
+   id="svg2"
+   version="1.1"
+   inkscape:version="0.48.3.1 r9886"
+   sodipodi:docname="RCUApplicability.svg">
+  <metadata
+     id="metadata40">
+    <rdf:RDF>
+      <cc:Work
+         rdf:about="">
+        <dc:format>image/svg+xml</dc:format>
+        <dc:type
+           rdf:resource="http://purl.org/dc/dcmitype/StillImage" />
+        <dc:title />
+      </cc:Work>
+    </rdf:RDF>
+  </metadata>
+  <defs
+     id="defs38" />
+  <sodipodi:namedview
+     pagecolor="#ffffff"
+     bordercolor="#666666"
+     borderopacity="1"
+     objecttolerance="10"
+     gridtolerance="10"
+     guidetolerance="10"
+     inkscape:pageopacity="0"
+     inkscape:pageshadow="2"
+     inkscape:window-width="849"
+     inkscape:window-height="639"
+     id="namedview36"
+     showgrid="false"
+     inkscape:zoom="0.51326165"
+     inkscape:cx="544.56912"
+     inkscape:cy="334.10686"
+     inkscape:window-x="149"
+     inkscape:window-y="448"
+     inkscape:window-maximized="0"
+     inkscape:current-layer="g4"
+     fit-margin-top="5"
+     fit-margin-left="5"
+     fit-margin-right="5"
+     fit-margin-bottom="5" />
+  <g
+     style="fill:none;stroke-width:0.025in"
+     id="g4"
+     transform="translate(-2043.6828,14.791398)">
+    <!-- Line: box -->
+    <rect
+       x="0"
+       y="0"
+       width="14400"
+       height="8775"
+       rx="0"
+       style="fill:#ffa1a1;stroke:#000000;stroke-width:21;stroke-linecap:butt;stroke-linejoin:miter"
+       id="rect6" />
+    <!-- Line: box -->
+    <rect
+       x="1350"
+       y="0"
+       width="11700"
+       height="6075"
+       rx="0"
+       style="fill:#ffff00;stroke:#000000;stroke-width:21;stroke-linecap:butt;stroke-linejoin:miter"
+       id="rect8" />
+    <!-- Line: box -->
+    <rect
+       x="2700"
+       y="0"
+       width="9000"
+       height="4275"
+       rx="0"
+       style="fill:#00ff00;stroke:#000000;stroke-width:21;stroke-linecap:butt;stroke-linejoin:miter"
+       id="rect10" />
+    <!-- Line: box -->
+    <rect
+       x="4050"
+       y="0"
+       width="6300"
+       height="2475"
+       rx="0"
+       style="fill:#87cfff;stroke:#000000;stroke-width:21;stroke-linecap:butt;stroke-linejoin:miter"
+       id="rect12" />
+    <!-- Text -->
+    <text
+       xml:space="preserve"
+       x="7200"
+       y="900"
+       font-style="normal"
+       font-weight="normal"
+       font-size="324"
+       id="text14"
+       sodipodi:linespacing="125%"
+       style="font-size:427.63009644px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;writing-mode:lr-tb;text-anchor:middle;fill:#000000;font-family:Nimbus Sans L;-inkscape-font-specification:Nimbus Sans L"><tspan
+         style="font-size:427.63009644px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;writing-mode:lr-tb;text-anchor:middle;font-family:Nimbus Sans L;-inkscape-font-specification:Nimbus Sans L"
+         id="tspan3017">Read-Mostly, Stale &amp;</tspan></text>
+    <!-- Text -->
+    <text
+       xml:space="preserve"
+       x="7200"
+       y="1350"
+       font-style="normal"
+       font-weight="normal"
+       font-size="324"
+       id="text16"
+       sodipodi:linespacing="125%"
+       style="font-size:427.63009644px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;writing-mode:lr-tb;text-anchor:middle;fill:#000000;font-family:Nimbus Sans L;-inkscape-font-specification:Nimbus Sans L"><tspan
+         style="font-size:427.63009644px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;writing-mode:lr-tb;text-anchor:middle;font-family:Nimbus Sans L;-inkscape-font-specification:Nimbus Sans L"
+         id="tspan3019">Inconsistent Data OK</tspan></text>
+    <!-- Text -->
+    <text
+       xml:space="preserve"
+       x="7200"
+       y="1800"
+       font-style="normal"
+       font-weight="normal"
+       font-size="324"
+       id="text18"
+       sodipodi:linespacing="125%"
+       style="font-size:427.63009644px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;writing-mode:lr-tb;text-anchor:middle;fill:#000000;font-family:Nimbus Sans L;-inkscape-font-specification:Nimbus Sans L"><tspan
+         style="font-size:427.63009644px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;writing-mode:lr-tb;text-anchor:middle;font-family:Nimbus Sans L;-inkscape-font-specification:Nimbus Sans L"
+         id="tspan3021">(RCU Works Great!!!)</tspan></text>
+    <!-- Text -->
+    <text
+       xml:space="preserve"
+       x="7200"
+       y="3825"
+       font-style="normal"
+       font-weight="normal"
+       font-size="324"
+       id="text20"
+       sodipodi:linespacing="125%"
+       style="font-size:427.63009644px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;writing-mode:lr-tb;text-anchor:middle;fill:#000000;font-family:Nimbus Sans L;-inkscape-font-specification:Nimbus Sans L"><tspan
+         style="font-size:427.63009644px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;writing-mode:lr-tb;text-anchor:middle;font-family:Nimbus Sans L;-inkscape-font-specification:Nimbus Sans L"
+         id="tspan3023">(RCU Works Well)</tspan></text>
+    <!-- Text -->
+    <text
+       xml:space="preserve"
+       x="7200"
+       y="3375"
+       font-style="normal"
+       font-weight="normal"
+       font-size="324"
+       id="text22"
+       sodipodi:linespacing="125%"
+       style="font-size:427.63009644px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;writing-mode:lr-tb;text-anchor:middle;fill:#000000;font-family:Nimbus Sans L;-inkscape-font-specification:Nimbus Sans L"><tspan
+         style="font-size:427.63009644px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;writing-mode:lr-tb;text-anchor:middle;font-family:Nimbus Sans L;-inkscape-font-specification:Nimbus Sans L"
+         id="tspan3025">Read-Mostly, Need Consistent Data</tspan></text>
+    <!-- Text -->
+    <text
+       xml:space="preserve"
+       x="7200"
+       y="5175"
+       font-style="normal"
+       font-weight="normal"
+       font-size="324"
+       id="text24"
+       sodipodi:linespacing="125%"
+       style="font-size:427.63009644px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;writing-mode:lr-tb;text-anchor:middle;fill:#000000;font-family:Nimbus Sans L;-inkscape-font-specification:Nimbus Sans L"><tspan
+         style="font-size:427.63009644px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;writing-mode:lr-tb;text-anchor:middle;font-family:Nimbus Sans L;-inkscape-font-specification:Nimbus Sans L"
+         id="tspan3027">Read-Write, Need Consistent Data</tspan></text>
+    <!-- Text -->
+    <text
+       xml:space="preserve"
+       x="7200"
+       y="6975"
+       font-style="normal"
+       font-weight="normal"
+       font-size="324"
+       id="text26"
+       style="font-size:427.63009644px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;writing-mode:lr-tb;text-anchor:middle;fill:#000000;font-family:Nimbus Sans L;-inkscape-font-specification:Nimbus Sans L"
+       sodipodi:linespacing="125%">Update-Mostly, Need Consistent Data</text>
+    <!-- Text -->
+    <text
+       xml:space="preserve"
+       x="7200"
+       y="5625"
+       font-style="normal"
+       font-weight="normal"
+       font-size="324"
+       id="text28"
+       sodipodi:linespacing="125%"
+       style="font-size:427.63009644px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;writing-mode:lr-tb;text-anchor:middle;fill:#000000;font-family:Nimbus Sans L;-inkscape-font-specification:Nimbus Sans L"><tspan
+         style="font-size:427.63009644px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;writing-mode:lr-tb;text-anchor:middle;font-family:Nimbus Sans L;-inkscape-font-specification:Nimbus Sans L"
+         id="tspan3029">(RCU Might Be OK...)</tspan></text>
+    <!-- Text -->
+    <text
+       xml:space="preserve"
+       x="7200"
+       y="7875"
+       font-style="normal"
+       font-weight="normal"
+       font-size="324"
+       id="text30"
+       style="font-size:427.63009644px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;writing-mode:lr-tb;text-anchor:middle;fill:#000000;font-family:Nimbus Sans L;-inkscape-font-specification:Nimbus Sans L"
+       sodipodi:linespacing="125%">(1) Provide Existence Guarantees For Update-Friendly Mechanisms</text>
+    <!-- Text -->
+    <text
+       xml:space="preserve"
+       x="7200"
+       y="8325"
+       font-style="normal"
+       font-weight="normal"
+       font-size="324"
+       id="text32"
+       style="font-size:427.63009644px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;writing-mode:lr-tb;text-anchor:middle;fill:#000000;font-family:Nimbus Sans L;-inkscape-font-specification:Nimbus Sans L"
+       sodipodi:linespacing="125%">(2) Provide Wait-Free Read-Side Primitives for Real-Time Use)</text>
+    <!-- Text -->
+    <text
+       xml:space="preserve"
+       x="7200"
+       y="7425"
+       font-style="normal"
+       font-weight="normal"
+       font-size="324"
+       id="text34"
+       style="font-size:427.63009644px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;writing-mode:lr-tb;text-anchor:middle;fill:#000000;font-family:Nimbus Sans L;-inkscape-font-specification:Nimbus Sans L"
+       sodipodi:linespacing="125%">(RCU is Very Unlikely to be the Right Tool For The Job, But it Can:</text>
+  </g>
+</svg>
diff --git a/Documentation/RCU/Design/Requirements/ReadersPartitionGP1.svg b/Documentation/RCU/Design/Requirements/ReadersPartitionGP1.svg
new file mode 100644
index 000000000000..48cd1623d4d4
--- /dev/null
+++ b/Documentation/RCU/Design/Requirements/ReadersPartitionGP1.svg
@@ -0,0 +1,639 @@
+<?xml version="1.0" encoding="UTF-8" standalone="no"?>
+<!-- Created with Inkscape (http://www.inkscape.org/) -->
+
+<svg
+   xmlns:dc="http://purl.org/dc/elements/1.1/"
+   xmlns:cc="http://creativecommons.org/ns#"
+   xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#"
+   xmlns:svg="http://www.w3.org/2000/svg"
+   xmlns="http://www.w3.org/2000/svg"
+   xmlns:sodipodi="http://sodipodi.sourceforge.net/DTD/sodipodi-0.dtd"
+   xmlns:inkscape="http://www.inkscape.org/namespaces/inkscape"
+   width="735.25"
+   height="516.21875"
+   id="svg2"
+   version="1.1"
+   inkscape:version="0.48.3.1 r9886"
+   sodipodi:docname="ReadersPartitionGP1.svg">
+  <defs
+     id="defs4">
+    <marker
+       inkscape:stockid="Arrow2Lend"
+       orient="auto"
+       refY="0"
+       refX="0"
+       id="Arrow2Lend"
+       style="overflow:visible">
+      <path
+         id="path3792"
+         style="fill-rule:evenodd;stroke-width:0.625;stroke-linejoin:round"
+         d="M 8.7185878,4.0337352 -2.2072895,0.01601326 8.7185884,-4.0017078 c -1.7454984,2.3720609 -1.7354408,5.6174519 -6e-7,8.035443 z"
+         transform="matrix(-1.1,0,0,-1.1,-1.1,0)"
+         inkscape:connector-curvature="0" />
+    </marker>
+    <marker
+       inkscape:stockid="Arrow2Lstart"
+       orient="auto"
+       refY="0"
+       refX="0"
+       id="Arrow2Lstart"
+       style="overflow:visible">
+      <path
+         id="path3789"
+         style="fill-rule:evenodd;stroke-width:0.625;stroke-linejoin:round"
+         d="M 8.7185878,4.0337352 -2.2072895,0.01601326 8.7185884,-4.0017078 c -1.7454984,2.3720609 -1.7354408,5.6174519 -6e-7,8.035443 z"
+         transform="matrix(1.1,0,0,1.1,1.1,0)"
+         inkscape:connector-curvature="0" />
+    </marker>
+    <marker
+       inkscape:stockid="Arrow2Lstart"
+       orient="auto"
+       refY="0"
+       refX="0"
+       id="Arrow2Lstart-4"
+       style="overflow:visible">
+      <path
+         id="path3789-9"
+         style="fill-rule:evenodd;stroke-width:0.625;stroke-linejoin:round"
+         d="M 8.7185878,4.0337352 -2.2072895,0.01601326 8.7185884,-4.0017078 c -1.7454984,2.3720609 -1.7354408,5.6174519 -6e-7,8.035443 z"
+         transform="matrix(1.1,0,0,1.1,1.1,0)"
+         inkscape:connector-curvature="0" />
+    </marker>
+    <marker
+       inkscape:stockid="Arrow2Lend"
+       orient="auto"
+       refY="0"
+       refX="0"
+       id="Arrow2Lend-4"
+       style="overflow:visible">
+      <path
+         id="path3792-4"
+         style="fill-rule:evenodd;stroke-width:0.625;stroke-linejoin:round"
+         d="M 8.7185878,4.0337352 -2.2072895,0.01601326 8.7185884,-4.0017078 c -1.7454984,2.3720609 -1.7354408,5.6174519 -6e-7,8.035443 z"
+         transform="matrix(-1.1,0,0,-1.1,-1.1,0)"
+         inkscape:connector-curvature="0" />
+    </marker>
+  </defs>
+  <sodipodi:namedview
+     id="base"
+     pagecolor="#ffffff"
+     bordercolor="#666666"
+     borderopacity="1.0"
+     inkscape:pageopacity="0.0"
+     inkscape:pageshadow="2"
+     inkscape:zoom="1.3670394"
+     inkscape:cx="367.26465"
+     inkscape:cy="258.46182"
+     inkscape:document-units="px"
+     inkscape:current-layer="g4433-6"
+     showgrid="false"
+     inkscape:window-width="1351"
+     inkscape:window-height="836"
+     inkscape:window-x="438"
+     inkscape:window-y="335"
+     inkscape:window-maximized="0"
+     fit-margin-top="5"
+     fit-margin-left="5"
+     fit-margin-right="5"
+     fit-margin-bottom="5" />
+  <metadata
+     id="metadata7">
+    <rdf:RDF>
+      <cc:Work
+         rdf:about="">
+        <dc:format>image/svg+xml</dc:format>
+        <dc:type
+           rdf:resource="http://purl.org/dc/dcmitype/StillImage" />
+        <dc:title />
+      </cc:Work>
+    </rdf:RDF>
+  </metadata>
+  <g
+     inkscape:label="Layer 1"
+     inkscape:groupmode="layer"
+     id="layer1"
+     transform="translate(-29.15625,-185.59375)">
+    <flowRoot
+       xml:space="preserve"
+       id="flowRoot2985"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"><flowRegion
+         id="flowRegion2987"><rect
+           id="rect2989"
+           width="82.85714"
+           height="11.428572"
+           x="240"
+           y="492.36218" /></flowRegion><flowPara
+         id="flowPara2991" /></flowRoot>    <g
+       id="g4433"
+       transform="translate(2,-12)">
+      <text
+         sodipodi:linespacing="125%"
+         id="text2993"
+         y="-261.66608"
+         x="436.12299"
+         style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+         xml:space="preserve"
+         transform="matrix(0,1,-1,0,0,0)"><tspan
+           y="-261.66608"
+           x="436.12299"
+           id="tspan2995"
+           sodipodi:role="line">synchronize_rcu()</tspan></text>
+      <g
+         id="g4417"
+         transform="matrix(0,1,-1,0,730.90257,222.4928)">
+        <path
+           style="fill:none;stroke:#000000;stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;stroke-opacity:1;marker-start:url(#Arrow2Lstart);marker-end:url(#Arrow2Lend)"
+           d="M 97.580736,477.4048 327.57913,476.09759"
+           id="path2997"
+           inkscape:connector-curvature="0"
+           sodipodi:nodetypes="cc" />
+        <path
+           style="fill:none;stroke:#000000;stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;stroke-opacity:1"
+           d="m 96.752718,465.38398 0,22.62742"
+           id="path4397"
+           inkscape:connector-curvature="0"
+           sodipodi:nodetypes="cc" />
+        <path
+           style="fill:none;stroke:#000000;stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;stroke-opacity:1"
+           d="m 328.40703,465.38397 0,22.62742"
+           id="path4397-5"
+           inkscape:connector-curvature="0"
+           sodipodi:nodetypes="cc" />
+      </g>
+    </g>
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="112.04738"
+       y="268.18076"
+       id="text4429"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4431"
+         x="112.04738"
+         y="268.18076">WRITE_ONCE(a, 1);</tspan></text>
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="112.04738"
+       y="487.13766"
+       id="text4441"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4443"
+         x="112.04738"
+         y="487.13766">WRITE_ONCE(b, 1);</tspan></text>
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="255.60869"
+       y="297.29346"
+       id="text4445"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4447"
+         x="255.60869"
+         y="297.29346">r1 = READ_ONCE(a);</tspan></text>
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="255.14423"
+       y="554.61786"
+       id="text4449"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4451"
+         x="255.14423"
+         y="554.61786">WRITE_ONCE(c, 1);</tspan></text>
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="396.10254"
+       y="370.71124"
+       id="text4453"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4455"
+         x="396.10254"
+         y="370.71124">WRITE_ONCE(d, 1);</tspan></text>
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="396.10254"
+       y="572.13617"
+       id="text4457"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4459"
+         x="396.10254"
+         y="572.13617">r2 = READ_ONCE(c);</tspan></text>
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="112.08231"
+       y="213.91006"
+       id="text4461"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4463"
+         x="112.08231"
+         y="213.91006">thread0()</tspan></text>
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="252.34512"
+       y="213.91006"
+       id="text4461-6"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4463-0"
+         x="252.34512"
+         y="213.91006">thread1()</tspan></text>
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="396.42557"
+       y="213.91006"
+       id="text4461-2"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4463-2"
+         x="396.42557"
+         y="213.91006">thread2()</tspan></text>
+    <rect
+       style="fill:none;stroke:#000000;stroke-width:1;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0"
+       id="rect4495"
+       width="724.25244"
+       height="505.21201"
+       x="34.648232"
+       y="191.10612" />
+    <path
+       style="fill:none;stroke:#000000;stroke-width:1;stroke-linecap:butt;stroke-linejoin:miter;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0"
+       d="m 183.14066,191.10612 0,504.24243"
+       id="path4497"
+       inkscape:connector-curvature="0"
+       sodipodi:nodetypes="cc" />
+    <path
+       style="fill:none;stroke:#000000;stroke-width:1;stroke-linecap:butt;stroke-linejoin:miter;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0"
+       d="m 325.13867,191.10612 0,504.24243"
+       id="path4497-5"
+       inkscape:connector-curvature="0"
+       sodipodi:nodetypes="cc" />
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="111.75929"
+       y="251.53981"
+       id="text4429-8"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4431-9"
+         x="111.75929"
+         y="251.53981">rcu_read_lock();</tspan></text>
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="396.10254"
+       y="353.91556"
+       id="text4429-8-9"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4431-9-4"
+         x="396.10254"
+         y="353.91556">rcu_read_lock();</tspan></text>
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="396.10254"
+       y="587.40289"
+       id="text4429-8-9-3"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4431-9-4-4"
+         x="396.10254"
+         y="587.40289">rcu_read_unlock();</tspan></text>
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="111.75929"
+       y="501.15311"
+       id="text4429-8-9-3-1"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4431-9-4-4-6"
+         x="111.75929"
+         y="501.15311">rcu_read_unlock();</tspan></text>
+    <path
+       style="fill:none;stroke:#000000;stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;stroke-opacity:1"
+       d="m 33.941125,227.87568 724.941765,0"
+       id="path4608"
+       inkscape:connector-curvature="0"
+       sodipodi:nodetypes="cc" />
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="394.94427"
+       y="331.66351"
+       id="text4648"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4650"
+         x="394.94427"
+         y="331.66351">QS</tspan></text>
+    <path
+       sodipodi:type="arc"
+       style="fill:none;stroke:#000000;stroke-width:1;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0"
+       id="path4652"
+       sodipodi:cx="358.85669"
+       sodipodi:cy="142.87541"
+       sodipodi:rx="10.960155"
+       sodipodi:ry="10.253048"
+       d="m 358.86939,132.62237 a 10.960155,10.253048 0 1 1 -0.0228,0"
+       transform="translate(36.441125,185.60612)"
+       sodipodi:start="4.7135481"
+       sodipodi:end="10.994651"
+       sodipodi:open="true" />
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="112.11968"
+       y="523.77856"
+       id="text4648-4"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4650-4"
+         x="112.11968"
+         y="523.77856">QS</tspan></text>
+    <path
+       sodipodi:type="arc"
+       style="fill:none;stroke:#000000;stroke-width:1;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0"
+       id="path4652-7"
+       sodipodi:cx="358.85669"
+       sodipodi:cy="142.87541"
+       sodipodi:rx="10.960155"
+       sodipodi:ry="10.253048"
+       d="m 358.86939,132.62237 a 10.960155,10.253048 0 1 1 -0.0228,0"
+       transform="translate(-246.38346,377.72117)"
+       sodipodi:start="4.7135481"
+       sodipodi:end="10.994651"
+       sodipodi:open="true" />
+    <path
+       sodipodi:type="arc"
+       style="fill:#ffffff;fill-opacity:1;stroke:#000000;stroke-width:1;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0"
+       id="path4652-7-7"
+       sodipodi:cx="358.85669"
+       sodipodi:cy="142.87541"
+       sodipodi:rx="10.960155"
+       sodipodi:ry="10.253048"
+       d="m 358.86939,132.62237 a 10.960155,10.253048 0 1 1 -0.0228,0"
+       transform="translate(-103.65246,190.90878)"
+       sodipodi:start="4.7135481"
+       sodipodi:end="10.994651"
+       sodipodi:open="true" />
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="254.85066"
+       y="336.96619"
+       id="text4648-4-3"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4650-4-5"
+         x="254.85066"
+         y="336.96619">QS</tspan></text>
+    <path
+       style="fill:none;stroke:#000000;stroke-width:1;stroke-linecap:butt;stroke-linejoin:miter;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0"
+       d="m 470.93311,190.39903 0,504.24243"
+       id="path4497-5-6"
+       inkscape:connector-curvature="0"
+       sodipodi:nodetypes="cc" />
+    <path
+       style="fill:none;stroke:#000000;stroke-width:1;stroke-linecap:butt;stroke-linejoin:miter;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0"
+       d="m 616.22755,190.38323 0,504.24243"
+       id="path4497-5-2"
+       inkscape:connector-curvature="0"
+       sodipodi:nodetypes="cc" />
+    <g
+       id="g4433-6"
+       transform="translate(288.0964,78.32827)">
+      <text
+         sodipodi:linespacing="125%"
+         id="text2993-7"
+         y="-261.66608"
+         x="440.12299"
+         style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+         xml:space="preserve"
+         transform="matrix(0,1,-1,0,0,0)"><tspan
+           y="-261.66608"
+           x="440.12299"
+           id="tspan2995-1"
+           sodipodi:role="line">synchronize_rcu()</tspan></text>
+      <g
+         id="g4417-1"
+         transform="matrix(0,1,-1,0,730.90257,222.4928)">
+        <path
+           style="fill:none;stroke:#000000;stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;stroke-opacity:1;marker-start:url(#Arrow2Lstart);marker-end:url(#Arrow2Lend)"
+           d="M 97.580736,477.4048 328.5624,477.07246"
+           id="path2997-2"
+           inkscape:connector-curvature="0"
+           sodipodi:nodetypes="cc" />
+        <path
+           style="fill:none;stroke:#000000;stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;stroke-opacity:1"
+           d="m 96.752718,465.38398 0,22.62742"
+           id="path4397-3"
+           inkscape:connector-curvature="0"
+           sodipodi:nodetypes="cc" />
+        <path
+           style="fill:none;stroke:#000000;stroke-width:1px;stroke-linecap:butt;stroke-linejoin:miter;stroke-opacity:1"
+           d="m 329.39039,465.38397 0,22.62742"
+           id="path4397-5-4"
+           inkscape:connector-curvature="0"
+           sodipodi:nodetypes="cc" />
+      </g>
+    </g>
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="541.70508"
+       y="387.6217"
+       id="text4445-0"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4447-5"
+         x="541.70508"
+         y="387.6217">r3 = READ_ONCE(d);</tspan></text>
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="541.2406"
+       y="646.94611"
+       id="text4449-6"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4451-6"
+         x="541.2406"
+         y="646.94611">WRITE_ONCE(e, 1);</tspan></text>
+    <path
+       sodipodi:type="arc"
+       style="fill:#ffffff;fill-opacity:1;stroke:#000000;stroke-width:1;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0"
+       id="path4652-7-7-5"
+       sodipodi:cx="358.85669"
+       sodipodi:cy="142.87541"
+       sodipodi:rx="10.960155"
+       sodipodi:ry="10.253048"
+       d="m 358.86939,132.62237 a 10.960155,10.253048 0 1 1 -0.0228,0"
+       transform="translate(182.44393,281.23704)"
+       sodipodi:start="4.7135481"
+       sodipodi:end="10.994651"
+       sodipodi:open="true" />
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="540.94702"
+       y="427.29443"
+       id="text4648-4-3-1"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4650-4-5-7"
+         x="540.94702"
+         y="427.29443">QS</tspan></text>
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="686.27747"
+       y="461.83929"
+       id="text4453-7"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4455-1"
+         x="686.27747"
+         y="461.83929">r4 = READ_ONCE(b);</tspan></text>
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="686.27747"
+       y="669.26422"
+       id="text4457-9"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4459-2"
+         x="686.27747"
+         y="669.26422">r5 = READ_ONCE(e);</tspan></text>
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="686.27747"
+       y="445.04358"
+       id="text4429-8-9-33"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4431-9-4-2"
+         x="686.27747"
+         y="445.04358">rcu_read_lock();</tspan></text>
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="686.27747"
+       y="684.53094"
+       id="text4429-8-9-3-8"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4431-9-4-4-5"
+         x="686.27747"
+         y="684.53094">rcu_read_unlock();</tspan></text>
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="685.11914"
+       y="422.79153"
+       id="text4648-9"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4650-7"
+         x="685.11914"
+         y="422.79153">QS</tspan></text>
+    <path
+       sodipodi:type="arc"
+       style="fill:none;stroke:#000000;stroke-width:1;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0"
+       id="path4652-8"
+       sodipodi:cx="358.85669"
+       sodipodi:cy="142.87541"
+       sodipodi:rx="10.960155"
+       sodipodi:ry="10.253048"
+       d="m 358.86939,132.62237 a 10.960155,10.253048 0 1 1 -0.0228,0"
+       transform="translate(326.61602,276.73415)"
+       sodipodi:start="4.7135481"
+       sodipodi:end="10.994651"
+       sodipodi:open="true" />
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="397.85934"
+       y="609.59003"
+       id="text4648-5"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4650-77"
+         x="397.85934"
+         y="609.59003">QS</tspan></text>
+    <path
+       sodipodi:type="arc"
+       style="fill:none;stroke:#000000;stroke-width:1;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0"
+       id="path4652-80"
+       sodipodi:cx="358.85669"
+       sodipodi:cy="142.87541"
+       sodipodi:rx="10.960155"
+       sodipodi:ry="10.253048"
+       d="m 358.86939,132.62237 a 10.960155,10.253048 0 1 1 -0.0228,0"
+       transform="translate(39.356201,463.53264)"
+       sodipodi:start="4.7135481"
+       sodipodi:end="10.994651"
+       sodipodi:open="true" />
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="256.75986"
+       y="586.99133"
+       id="text4648-5-2"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4650-77-7"
+         x="256.75986"
+         y="586.99133">QS</tspan></text>
+    <path
+       sodipodi:type="arc"
+       style="fill:none;stroke:#000000;stroke-width:1;stroke-miterlimit:4;stroke-opacity:1;stroke-dasharray:none;stroke-dashoffset:0"
+       id="path4652-80-5"
+       sodipodi:cx="358.85669"
+       sodipodi:cy="142.87541"
+       sodipodi:rx="10.960155"
+       sodipodi:ry="10.253048"
+       d="m 358.86939,132.62237 a 10.960155,10.253048 0 1 1 -0.0228,0"
+       transform="translate(-101.74328,440.93395)"
+       sodipodi:start="4.7135481"
+       sodipodi:end="10.994651"
+       sodipodi:open="true" />
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="546.22791"
+       y="213.91006"
+       id="text4461-2-5"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4463-2-6"
+         x="546.22791"
+         y="213.91006">thread3()</tspan></text>
+    <text
+       xml:space="preserve"
+       style="font-size:10px;font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;text-align:center;line-height:125%;letter-spacing:0px;word-spacing:0px;writing-mode:lr-tb;text-anchor:middle;fill:#000000;fill-opacity:1;stroke:none;font-family:Symbol;-inkscape-font-specification:Symbol"
+       x="684.00067"
+       y="213.91006"
+       id="text4461-2-1"
+       sodipodi:linespacing="125%"><tspan
+         sodipodi:role="line"
+         id="tspan4463-2-0"
+         x="684.00067"
+         y="213.91006">thread4()</tspan></text>
+  </g>
+</svg>
diff --git a/Documentation/RCU/Design/Requirements/Requirements.html b/Documentation/RCU/Design/Requirements/Requirements.html
new file mode 100644
index 000000000000..95ef7b5de951
--- /dev/null
+++ b/Documentation/RCU/Design/Requirements/Requirements.html
@@ -0,0 +1,2799 @@
+<!-- DO NOT HAND EDIT. -->
+<!-- Instead, edit Requirements.htmlx and run 'sh htmlqqz.sh Requirements' -->
+<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN"
+        "http://www.w3.org/TR/html4/loose.dtd">
+        <html>
+        <head><title>A Tour Through RCU's Requirements [LWN.net]</title>
+        <meta HTTP-EQUIV="Content-Type" CONTENT="text/html; charset=iso-8859-1">
+
+<h1>A Tour Through RCU's Requirements</h1>
+
+<p>Copyright IBM Corporation, 2015</p>
+<p>Author: Paul E.&nbsp;McKenney</p>
+<p><i>The initial version of this document appeared in the
+<a href="http://lwn.net/">LWN</a> articles
+<a href="http://lwn.net/Articles/652156/">here</a>,
+<a href="http://lwn.net/Articles/652677/">here</a>, and
+<a href="http://lwn.net/Articles/653326/">here</a>.</i></p>
+
+<h2>Introduction</h2>
+
+<p>
+Read-copy update (RCU) is a synchronization mechanism that is often
+used as a replacement for reader-writer locking.
+RCU is unusual in that updaters do not block readers,
+which means that RCU's read-side primitives can be exceedingly fast
+and scalable.
+In addition, updaters can make useful forward progress concurrently
+with readers.
+However, all this concurrency between RCU readers and updaters does raise
+the question of exactly what RCU readers are doing, which in turn
+raises the question of exactly what RCU's requirements are.
+
+<p>
+This document therefore summarizes RCU's requirements, and can be thought
+of as an informal, high-level specification for RCU.
+It is important to understand that RCU's specification is primarily
+empirical in nature;
+in fact, I learned about many of these requirements the hard way.
+This situation might cause some consternation, however, not only
+has this learning process been a lot of fun, but it has also been
+a great privilege to work with so many people willing to apply
+technologies in interesting new ways.
+
+<p>
+All that aside, here are the categories of currently known RCU requirements:
+</p>
+
+<ol>
+<li>	<a href="#Fundamental Requirements">
+	Fundamental Requirements</a>
+<li>	<a href="#Fundamental Non-Requirements">Fundamental Non-Requirements</a>
+<li>	<a href="#Parallelism Facts of Life">
+	Parallelism Facts of Life</a>
+<li>	<a href="#Quality-of-Implementation Requirements">
+	Quality-of-Implementation Requirements</a>
+<li>	<a href="#Linux Kernel Complications">
+	Linux Kernel Complications</a>
+<li>	<a href="#Software-Engineering Requirements">
+	Software-Engineering Requirements</a>
+<li>	<a href="#Other RCU Flavors">
+	Other RCU Flavors</a>
+<li>	<a href="#Possible Future Changes">
+	Possible Future Changes</a>
+</ol>
+
+<p>
+This is followed by a <a href="#Summary">summary</a>,
+which is in turn followed by the inevitable
+<a href="#Answers to Quick Quizzes">answers to the quick quizzes</a>.
+
+<h2><a name="Fundamental Requirements">Fundamental Requirements</a></h2>
+
+<p>
+RCU's fundamental requirements are the closest thing RCU has to hard
+mathematical requirements.
+These are:
+
+<ol>
+<li>	<a href="#Grace-Period Guarantee">
+	Grace-Period Guarantee</a>
+<li>	<a href="#Publish-Subscribe Guarantee">
+	Publish-Subscribe Guarantee</a>
+<li>	<a href="#RCU Primitives Guaranteed to Execute Unconditionally">
+	RCU Primitives Guaranteed to Execute Unconditionally</a>
+<li>	<a href="#Guaranteed Read-to-Write Upgrade">
+	Guaranteed Read-to-Write Upgrade</a>
+</ol>
+
+<h3><a name="Grace-Period Guarantee">Grace-Period Guarantee</a></h3>
+
+<p>
+RCU's grace-period guarantee is unusual in being premeditated:
+Jack Slingwine and I had this guarantee firmly in mind when we started
+work on RCU (then called &ldquo;rclock&rdquo;) in the early 1990s.
+That said, the past two decades of experience with RCU have produced
+a much more detailed understanding of this guarantee.
+
+<p>
+RCU's grace-period guarantee allows updaters to wait for the completion
+of all pre-existing RCU read-side critical sections.
+An RCU read-side critical section
+begins with the marker <tt>rcu_read_lock()</tt> and ends with
+the marker <tt>rcu_read_unlock()</tt>.
+These markers may be nested, and RCU treats a nested set as one
+big RCU read-side critical section.
+Production-quality implementations of <tt>rcu_read_lock()</tt> and
+<tt>rcu_read_unlock()</tt> are extremely lightweight, and in
+fact have exactly zero overhead in Linux kernels built for production
+use with <tt>CONFIG_PREEMPT=n</tt>.
+
+<p>
+This guarantee allows ordering to be enforced with extremely low
+overhead to readers, for example:
+
+<blockquote>
+<pre>
+ 1 int x, y;
+ 2
+ 3 void thread0(void)
+ 4 {
+ 5   rcu_read_lock();
+ 6   r1 = READ_ONCE(x);
+ 7   r2 = READ_ONCE(y);
+ 8   rcu_read_unlock();
+ 9 }
+10
+11 void thread1(void)
+12 {
+13   WRITE_ONCE(x, 1);
+14   synchronize_rcu();
+15   WRITE_ONCE(y, 1);
+16 }
+</pre>
+</blockquote>
+
+<p>
+Because the <tt>synchronize_rcu()</tt> on line&nbsp;14 waits for
+all pre-existing readers, any instance of <tt>thread0()</tt> that
+loads a value of zero from <tt>x</tt> must complete before
+<tt>thread1()</tt> stores to <tt>y</tt>, so that instance must
+also load a value of zero from <tt>y</tt>.
+Similarly, any instance of <tt>thread0()</tt> that loads a value of
+one from <tt>y</tt> must have started after the
+<tt>synchronize_rcu()</tt> started, and must therefore also load
+a value of one from <tt>x</tt>.
+Therefore, the outcome:
+<blockquote>
+<pre>
+(r1 == 0 &amp;&amp; r2 == 1)
+</pre>
+</blockquote>
+cannot happen.
+
+<p><a name="Quick Quiz 1"><b>Quick Quiz 1</b>:</a>
+Wait a minute!
+You said that updaters can make useful forward progress concurrently
+with readers, but pre-existing readers will block
+<tt>synchronize_rcu()</tt>!!!
+Just who are you trying to fool???
+<br><a href="#qq1answer">Answer</a>
+
+<p>
+This scenario resembles one of the first uses of RCU in
+<a href="http://en.wikipedia.org/wiki/DYNIX">DYNIX/ptx</a>,
+which managed a distributed lock manager's transition into
+a state suitable for handling recovery from node failure,
+more or less as follows:
+
+<blockquote>
+<pre>
+ 1 #define STATE_NORMAL        0
+ 2 #define STATE_WANT_RECOVERY 1
+ 3 #define STATE_RECOVERING    2
+ 4 #define STATE_WANT_NORMAL   3
+ 5
+ 6 int state = STATE_NORMAL;
+ 7
+ 8 void do_something_dlm(void)
+ 9 {
+10   int state_snap;
+11
+12   rcu_read_lock();
+13   state_snap = READ_ONCE(state);
+14   if (state_snap == STATE_NORMAL)
+15     do_something();
+16   else
+17     do_something_carefully();
+18   rcu_read_unlock();
+19 }
+20
+21 void start_recovery(void)
+22 {
+23   WRITE_ONCE(state, STATE_WANT_RECOVERY);
+24   synchronize_rcu();
+25   WRITE_ONCE(state, STATE_RECOVERING);
+26   recovery();
+27   WRITE_ONCE(state, STATE_WANT_NORMAL);
+28   synchronize_rcu();
+29   WRITE_ONCE(state, STATE_NORMAL);
+30 }
+</pre>
+</blockquote>
+
+<p>
+The RCU read-side critical section in <tt>do_something_dlm()</tt>
+works with the <tt>synchronize_rcu()</tt> in <tt>start_recovery()</tt>
+to guarantee that <tt>do_something()</tt> never runs concurrently
+with <tt>recovery()</tt>, but with little or no synchronization
+overhead in <tt>do_something_dlm()</tt>.
+
+<p><a name="Quick Quiz 2"><b>Quick Quiz 2</b>:</a>
+Why is the <tt>synchronize_rcu()</tt> on line&nbsp;28 needed?
+<br><a href="#qq2answer">Answer</a>
+
+<p>
+In order to avoid fatal problems such as deadlocks,
+an RCU read-side critical section must not contain calls to
+<tt>synchronize_rcu()</tt>.
+Similarly, an RCU read-side critical section must not
+contain anything that waits, directly or indirectly, on completion of
+an invocation of <tt>synchronize_rcu()</tt>.
+
+<p>
+Although RCU's grace-period guarantee is useful in and of itself, with
+<a href="https://lwn.net/Articles/573497/">quite a few use cases</a>,
+it would be good to be able to use RCU to coordinate read-side
+access to linked data structures.
+For this, the grace-period guarantee is not sufficient, as can
+be seen in function <tt>add_gp_buggy()</tt> below.
+We will look at the reader's code later, but in the meantime, just think of
+the reader as locklessly picking up the <tt>gp</tt> pointer,
+and, if the value loaded is non-<tt>NULL</tt>, locklessly accessing the
+<tt>-&gt;a</tt> and <tt>-&gt;b</tt> fields.
+
+<blockquote>
+<pre>
+ 1 bool add_gp_buggy(int a, int b)
+ 2 {
+ 3   p = kmalloc(sizeof(*p), GFP_KERNEL);
+ 4   if (!p)
+ 5     return -ENOMEM;
+ 6   spin_lock(&amp;gp_lock);
+ 7   if (rcu_access_pointer(gp)) {
+ 8     spin_unlock(&amp;gp_lock);
+ 9     return false;
+10   }
+11   p-&gt;a = a;
+12   p-&gt;b = a;
+13   gp = p; /* ORDERING BUG */
+14   spin_unlock(&amp;gp_lock);
+15   return true;
+16 }
+</pre>
+</blockquote>
+
+<p>
+The problem is that both the compiler and weakly ordered CPUs are within
+their rights to reorder this code as follows:
+
+<blockquote>
+<pre>
+ 1 bool add_gp_buggy_optimized(int a, int b)
+ 2 {
+ 3   p = kmalloc(sizeof(*p), GFP_KERNEL);
+ 4   if (!p)
+ 5     return -ENOMEM;
+ 6   spin_lock(&amp;gp_lock);
+ 7   if (rcu_access_pointer(gp)) {
+ 8     spin_unlock(&amp;gp_lock);
+ 9     return false;
+10   }
+<b>11   gp = p; /* ORDERING BUG */
+12   p-&gt;a = a;
+13   p-&gt;b = a;</b>
+14   spin_unlock(&amp;gp_lock);
+15   return true;
+16 }
+</pre>
+</blockquote>
+
+<p>
+If an RCU reader fetches <tt>gp</tt> just after
+<tt>add_gp_buggy_optimized</tt> executes line&nbsp;11,
+it will see garbage in the <tt>-&gt;a</tt> and <tt>-&gt;b</tt>
+fields.
+And this is but one of many ways in which compiler and hardware optimizations
+could cause trouble.
+Therefore, we clearly need some way to prevent the compiler and the CPU from
+reordering in this manner, which brings us to the publish-subscribe
+guarantee discussed in the next section.
+
+<h3><a name="Publish-Subscribe Guarantee">Publish/Subscribe Guarantee</a></h3>
+
+<p>
+RCU's publish-subscribe guarantee allows data to be inserted
+into a linked data structure without disrupting RCU readers.
+The updater uses <tt>rcu_assign_pointer()</tt> to insert the
+new data, and readers use <tt>rcu_dereference()</tt> to
+access data, whether new or old.
+The following shows an example of insertion:
+
+<blockquote>
+<pre>
+ 1 bool add_gp(int a, int b)
+ 2 {
+ 3   p = kmalloc(sizeof(*p), GFP_KERNEL);
+ 4   if (!p)
+ 5     return -ENOMEM;
+ 6   spin_lock(&amp;gp_lock);
+ 7   if (rcu_access_pointer(gp)) {
+ 8     spin_unlock(&amp;gp_lock);
+ 9     return false;
+10   }
+11   p-&gt;a = a;
+12   p-&gt;b = a;
+13   rcu_assign_pointer(gp, p);
+14   spin_unlock(&amp;gp_lock);
+15   return true;
+16 }
+</pre>
+</blockquote>
+
+<p>
+The <tt>rcu_assign_pointer()</tt> on line&nbsp;13 is conceptually
+equivalent to a simple assignment statement, but also guarantees
+that its assignment will
+happen after the two assignments in lines&nbsp;11 and&nbsp;12,
+similar to the C11 <tt>memory_order_release</tt> store operation.
+It also prevents any number of &ldquo;interesting&rdquo; compiler
+optimizations, for example, the use of <tt>gp</tt> as a scratch
+location immediately preceding the assignment.
+
+<p><a name="Quick Quiz 3"><b>Quick Quiz 3</b>:</a>
+But <tt>rcu_assign_pointer()</tt> does nothing to prevent the
+two assignments to <tt>p-&gt;a</tt> and <tt>p-&gt;b</tt>
+from being reordered.
+Can't that also cause problems?
+<br><a href="#qq3answer">Answer</a>
+
+<p>
+It is tempting to assume that the reader need not do anything special
+to control its accesses to the RCU-protected data,
+as shown in <tt>do_something_gp_buggy()</tt> below:
+
+<blockquote>
+<pre>
+ 1 bool do_something_gp_buggy(void)
+ 2 {
+ 3   rcu_read_lock();
+ 4   p = gp;  /* OPTIMIZATIONS GALORE!!! */
+ 5   if (p) {
+ 6     do_something(p-&gt;a, p-&gt;b);
+ 7     rcu_read_unlock();
+ 8     return true;
+ 9   }
+10   rcu_read_unlock();
+11   return false;
+12 }
+</pre>
+</blockquote>
+
+<p>
+However, this temptation must be resisted because there are a
+surprisingly large number of ways that the compiler
+(to say nothing of
+<a href="http://www.openvms.compaq.com/wizard/wiz_2637.html">DEC Alpha CPUs</a>)
+can trip this code up.
+For but one example, if the compiler were short of registers, it
+might choose to refetch from <tt>gp</tt> rather than keeping
+a separate copy in <tt>p</tt> as follows:
+
+<blockquote>
+<pre>
+ 1 bool do_something_gp_buggy_optimized(void)
+ 2 {
+ 3   rcu_read_lock();
+ 4   if (gp) { /* OPTIMIZATIONS GALORE!!! */
+<b> 5     do_something(gp-&gt;a, gp-&gt;b);</b>
+ 6     rcu_read_unlock();
+ 7     return true;
+ 8   }
+ 9   rcu_read_unlock();
+10   return false;
+11 }
+</pre>
+</blockquote>
+
+<p>
+If this function ran concurrently with a series of updates that
+replaced the current structure with a new one,
+the fetches of <tt>gp-&gt;a</tt>
+and <tt>gp-&gt;b</tt> might well come from two different structures,
+which could cause serious confusion.
+To prevent this (and much else besides), <tt>do_something_gp()</tt> uses
+<tt>rcu_dereference()</tt> to fetch from <tt>gp</tt>:
+
+<blockquote>
+<pre>
+ 1 bool do_something_gp(void)
+ 2 {
+ 3   rcu_read_lock();
+ 4   p = rcu_dereference(gp);
+ 5   if (p) {
+ 6     do_something(p-&gt;a, p-&gt;b);
+ 7     rcu_read_unlock();
+ 8     return true;
+ 9   }
+10   rcu_read_unlock();
+11   return false;
+12 }
+</pre>
+</blockquote>
+
+<p>
+The <tt>rcu_dereference()</tt> uses volatile casts and (for DEC Alpha)
+memory barriers in the Linux kernel.
+Should a
+<a href="http://www.rdrop.com/users/paulmck/RCU/consume.2015.07.13a.pdf">high-quality implementation of C11 <tt>memory_order_consume</tt> [PDF]</a>
+ever appear, then <tt>rcu_dereference()</tt> could be implemented
+as a <tt>memory_order_consume</tt> load.
+Regardless of the exact implementation, a pointer fetched by
+<tt>rcu_dereference()</tt> may not be used outside of the
+outermost RCU read-side critical section containing that
+<tt>rcu_dereference()</tt>, unless protection of
+the corresponding data element has been passed from RCU to some
+other synchronization mechanism, most commonly locking or
+<a href="https://www.kernel.org/doc/Documentation/RCU/rcuref.txt">reference counting</a>.
+
+<p>
+In short, updaters use <tt>rcu_assign_pointer()</tt> and readers
+use <tt>rcu_dereference()</tt>, and these two RCU API elements
+work together to ensure that readers have a consistent view of
+newly added data elements.
+
+<p>
+Of course, it is also necessary to remove elements from RCU-protected
+data structures, for example, using the following process:
+
+<ol>
+<li>	Remove the data element from the enclosing structure.
+<li>	Wait for all pre-existing RCU read-side critical sections
+	to complete (because only pre-existing readers can possibly have
+	a reference to the newly removed data element).
+<li>	At this point, only the updater has a reference to the
+	newly removed data element, so it can safely reclaim
+	the data element, for example, by passing it to <tt>kfree()</tt>.
+</ol>
+
+This process is implemented by <tt>remove_gp_synchronous()</tt>:
+
+<blockquote>
+<pre>
+ 1 bool remove_gp_synchronous(void)
+ 2 {
+ 3   struct foo *p;
+ 4
+ 5   spin_lock(&amp;gp_lock);
+ 6   p = rcu_access_pointer(gp);
+ 7   if (!p) {
+ 8     spin_unlock(&amp;gp_lock);
+ 9     return false;
+10   }
+11   rcu_assign_pointer(gp, NULL);
+12   spin_unlock(&amp;gp_lock);
+13   synchronize_rcu();
+14   kfree(p);
+15   return true;
+16 }
+</pre>
+</blockquote>
+
+<p>
+This function is straightforward, with line&nbsp;13 waiting for a grace
+period before line&nbsp;14 frees the old data element.
+This waiting ensures that readers will reach line&nbsp;7 of
+<tt>do_something_gp()</tt> before the data element referenced by
+<tt>p</tt> is freed.
+The <tt>rcu_access_pointer()</tt> on line&nbsp;6 is similar to
+<tt>rcu_dereference()</tt>, except that:
+
+<ol>
+<li>	The value returned by <tt>rcu_access_pointer()</tt>
+	cannot be dereferenced.
+	If you want to access the value pointed to as well as
+	the pointer itself, use <tt>rcu_dereference()</tt>
+	instead of <tt>rcu_access_pointer()</tt>.
+<li>	The call to <tt>rcu_access_pointer()</tt> need not be
+	protected.
+	In contrast, <tt>rcu_dereference()</tt> must either be
+	within an RCU read-side critical section or in a code
+	segment where the pointer cannot change, for example, in
+	code protected by the corresponding update-side lock.
+</ol>
+
+<p><a name="Quick Quiz 4"><b>Quick Quiz 4</b>:</a>
+Without the <tt>rcu_dereference()</tt> or the
+<tt>rcu_access_pointer()</tt>, what destructive optimizations
+might the compiler make use of?
+<br><a href="#qq4answer">Answer</a>
+
+<p>
+This simple linked-data-structure scenario clearly demonstrates the need
+for RCU's stringent memory-ordering guarantees on systems with more than
+one CPU:
+
+<ol>
+<li>	Each CPU that has an RCU read-side critical section that
+	begins before <tt>synchronize_rcu()</tt> starts is
+	guaranteed to execute a full memory barrier between the time
+	that the RCU read-side critical section ends and the time that
+	<tt>synchronize_rcu()</tt> returns.
+	Without this guarantee, a pre-existing RCU read-side critical section
+	might hold a reference to the newly removed <tt>struct foo</tt>
+	after the <tt>kfree()</tt> on line&nbsp;14 of
+	<tt>remove_gp_synchronous()</tt>.
+<li>	Each CPU that has an RCU read-side critical section that ends
+	after <tt>synchronize_rcu()</tt> returns is guaranteed
+	to execute a full memory barrier between the time that
+	<tt>synchronize_rcu()</tt> begins and the time that the RCU
+	read-side critical section begins.
+	Without this guarantee, a later RCU read-side critical section
+	running after the <tt>kfree()</tt> on line&nbsp;14 of
+	<tt>remove_gp_synchronous()</tt> might
+	later run <tt>do_something_gp()</tt> and find the
+	newly deleted <tt>struct foo</tt>.
+<li>	If the task invoking <tt>synchronize_rcu()</tt> remains
+	on a given CPU, then that CPU is guaranteed to execute a full
+	memory barrier sometime during the execution of
+	<tt>synchronize_rcu()</tt>.
+	This guarantee ensures that the <tt>kfree()</tt> on
+	line&nbsp;14 of <tt>remove_gp_synchronous()</tt> really does
+	execute after the removal on line&nbsp;11.
+<li>	If the task invoking <tt>synchronize_rcu()</tt> migrates
+	among a group of CPUs during that invocation, then each of the
+	CPUs in that group is guaranteed to execute a full memory barrier
+	sometime during the execution of <tt>synchronize_rcu()</tt>.
+	This guarantee also ensures that the <tt>kfree()</tt> on
+	line&nbsp;14 of <tt>remove_gp_synchronous()</tt> really does
+	execute after the removal on
+	line&nbsp;11, but also in the case where the thread executing the
+	<tt>synchronize_rcu()</tt> migrates in the meantime.
+</ol>
+
+<p><a name="Quick Quiz 5"><b>Quick Quiz 5</b>:</a>
+Given that multiple CPUs can start RCU read-side critical sections
+at any time without any ordering whatsoever, how can RCU possibly tell whether
+or not a given RCU read-side critical section starts before a
+given instance of <tt>synchronize_rcu()</tt>?
+<br><a href="#qq5answer">Answer</a>
+
+<p><a name="Quick Quiz 6"><b>Quick Quiz 6</b>:</a>
+The first and second guarantees require unbelievably strict ordering!
+Are all these memory barriers <i> really</i> required?
+<br><a href="#qq6answer">Answer</a>
+
+<p>
+In short, RCU's publish-subscribe guarantee is provided by the combination
+of <tt>rcu_assign_pointer()</tt> and <tt>rcu_dereference()</tt>.
+This guarantee allows data elements to be safely added to RCU-protected
+linked data structures without disrupting RCU readers.
+This guarantee can be used in combination with the grace-period
+guarantee to also allow data elements to be removed from RCU-protected
+linked data structures, again without disrupting RCU readers.
+
+<p>
+This guarantee was only partially premeditated.
+DYNIX/ptx used an explicit memory barrier for publication, but had nothing
+resembling <tt>rcu_dereference()</tt> for subscription, nor did it
+have anything resembling the <tt>smp_read_barrier_depends()</tt>
+that was later subsumed into <tt>rcu_dereference()</tt>.
+The need for these operations made itself known quite suddenly at a
+late-1990s meeting with the DEC Alpha architects, back in the days when
+DEC was still a free-standing company.
+It took the Alpha architects a good hour to convince me that any sort
+of barrier would ever be needed, and it then took me a good <i>two</i> hours
+to convince them that their documentation did not make this point clear.
+More recent work with the C and C++ standards committees have provided
+much education on tricks and traps from the compiler.
+In short, compilers were much less tricky in the early 1990s, but in
+2015, don't even think about omitting <tt>rcu_dereference()</tt>!
+
+<h3><a name="RCU Primitives Guaranteed to Execute Unconditionally">RCU Primitives Guaranteed to Execute Unconditionally</a></h3>
+
+<p>
+The common-case RCU primitives are unconditional.
+They are invoked, they do their job, and they return, with no possibility
+of error, and no need to retry.
+This is a key RCU design philosophy.
+
+<p>
+However, this philosophy is pragmatic rather than pigheaded.
+If someone comes up with a good justification for a particular conditional
+RCU primitive, it might well be implemented and added.
+After all, this guarantee was reverse-engineered, not premeditated.
+The unconditional nature of the RCU primitives was initially an
+accident of implementation, and later experience with synchronization
+primitives with conditional primitives caused me to elevate this
+accident to a guarantee.
+Therefore, the justification for adding a conditional primitive to
+RCU would need to be based on detailed and compelling use cases.
+
+<h3><a name="Guaranteed Read-to-Write Upgrade">Guaranteed Read-to-Write Upgrade</a></h3>
+
+<p>
+As far as RCU is concerned, it is always possible to carry out an
+update within an RCU read-side critical section.
+For example, that RCU read-side critical section might search for
+a given data element, and then might acquire the update-side
+spinlock in order to update that element, all while remaining
+in that RCU read-side critical section.
+Of course, it is necessary to exit the RCU read-side critical section
+before invoking <tt>synchronize_rcu()</tt>, however, this
+inconvenience can be avoided through use of the
+<tt>call_rcu()</tt> and <tt>kfree_rcu()</tt> API members
+described later in this document.
+
+<p><a name="Quick Quiz 7"><b>Quick Quiz 7</b>:</a>
+But how does the upgrade-to-write operation exclude other readers?
+<br><a href="#qq7answer">Answer</a>
+
+<p>
+This guarantee allows lookup code to be shared between read-side
+and update-side code, and was premeditated, appearing in the earliest
+DYNIX/ptx RCU documentation.
+
+<h2><a name="Fundamental Non-Requirements">Fundamental Non-Requirements</a></h2>
+
+<p>
+RCU provides extremely lightweight readers, and its read-side guarantees,
+though quite useful, are correspondingly lightweight.
+It is therefore all too easy to assume that RCU is guaranteeing more
+than it really is.
+Of course, the list of things that RCU does not guarantee is infinitely
+long, however, the following sections list a few non-guarantees that
+have caused confusion.
+Except where otherwise noted, these non-guarantees were premeditated.
+
+<ol>
+<li>	<a href="#Readers Impose Minimal Ordering">
+	Readers Impose Minimal Ordering</a>
+<li>	<a href="#Readers Do Not Exclude Updaters">
+	Readers Do Not Exclude Updaters</a>
+<li>	<a href="#Updaters Only Wait For Old Readers">
+	Updaters Only Wait For Old Readers</a>
+<li>	<a href="#Grace Periods Don't Partition Read-Side Critical Sections">
+	Grace Periods Don't Partition Read-Side Critical Sections</a>
+<li>	<a href="#Read-Side Critical Sections Don't Partition Grace Periods">
+	Read-Side Critical Sections Don't Partition Grace Periods</a>
+<li>	<a href="#Disabling Preemption Does Not Block Grace Periods">
+	Disabling Preemption Does Not Block Grace Periods</a>
+</ol>
+
+<h3><a name="Readers Impose Minimal Ordering">Readers Impose Minimal Ordering</a></h3>
+
+<p>
+Reader-side markers such as <tt>rcu_read_lock()</tt> and
+<tt>rcu_read_unlock()</tt> provide absolutely no ordering guarantees
+except through their interaction with the grace-period APIs such as
+<tt>synchronize_rcu()</tt>.
+To see this, consider the following pair of threads:
+
+<blockquote>
+<pre>
+ 1 void thread0(void)
+ 2 {
+ 3   rcu_read_lock();
+ 4   WRITE_ONCE(x, 1);
+ 5   rcu_read_unlock();
+ 6   rcu_read_lock();
+ 7   WRITE_ONCE(y, 1);
+ 8   rcu_read_unlock();
+ 9 }
+10
+11 void thread1(void)
+12 {
+13   rcu_read_lock();
+14   r1 = READ_ONCE(y);
+15   rcu_read_unlock();
+16   rcu_read_lock();
+17   r2 = READ_ONCE(x);
+18   rcu_read_unlock();
+19 }
+</pre>
+</blockquote>
+
+<p>
+After <tt>thread0()</tt> and <tt>thread1()</tt> execute
+concurrently, it is quite possible to have
+
+<blockquote>
+<pre>
+(r1 == 1 &amp;&amp; r2 == 0)
+</pre>
+</blockquote>
+
+(that is, <tt>y</tt> appears to have been assigned before <tt>x</tt>),
+which would not be possible if <tt>rcu_read_lock()</tt> and
+<tt>rcu_read_unlock()</tt> had much in the way of ordering
+properties.
+But they do not, so the CPU is within its rights
+to do significant reordering.
+This is by design:  Any significant ordering constraints would slow down
+these fast-path APIs.
+
+<p><a name="Quick Quiz 8"><b>Quick Quiz 8</b>:</a>
+Can't the compiler also reorder this code?
+<br><a href="#qq8answer">Answer</a>
+
+<h3><a name="Readers Do Not Exclude Updaters">Readers Do Not Exclude Updaters</a></h3>
+
+<p>
+Neither <tt>rcu_read_lock()</tt> nor <tt>rcu_read_unlock()</tt>
+exclude updates.
+All they do is to prevent grace periods from ending.
+The following example illustrates this:
+
+<blockquote>
+<pre>
+ 1 void thread0(void)
+ 2 {
+ 3   rcu_read_lock();
+ 4   r1 = READ_ONCE(y);
+ 5   if (r1) {
+ 6     do_something_with_nonzero_x();
+ 7     r2 = READ_ONCE(x);
+ 8     WARN_ON(!r2); /* BUG!!! */
+ 9   }
+10   rcu_read_unlock();
+11 }
+12
+13 void thread1(void)
+14 {
+15   spin_lock(&amp;my_lock);
+16   WRITE_ONCE(x, 1);
+17   WRITE_ONCE(y, 1);
+18   spin_unlock(&amp;my_lock);
+19 }
+</pre>
+</blockquote>
+
+<p>
+If the <tt>thread0()</tt> function's <tt>rcu_read_lock()</tt>
+excluded the <tt>thread1()</tt> function's update,
+the <tt>WARN_ON()</tt> could never fire.
+But the fact is that <tt>rcu_read_lock()</tt> does not exclude
+much of anything aside from subsequent grace periods, of which
+<tt>thread1()</tt> has none, so the
+<tt>WARN_ON()</tt> can and does fire.
+
+<h3><a name="Updaters Only Wait For Old Readers">Updaters Only Wait For Old Readers</a></h3>
+
+<p>
+It might be tempting to assume that after <tt>synchronize_rcu()</tt>
+completes, there are no readers executing.
+This temptation must be avoided because
+new readers can start immediately after <tt>synchronize_rcu()</tt>
+starts, and <tt>synchronize_rcu()</tt> is under no
+obligation to wait for these new readers.
+
+<p><a name="Quick Quiz 9"><b>Quick Quiz 9</b>:</a>
+Suppose that synchronize_rcu() did wait until all readers had completed.
+Would the updater be able to rely on this?
+<br><a href="#qq9answer">Answer</a>
+
+<h3><a name="Grace Periods Don't Partition Read-Side Critical Sections">
+Grace Periods Don't Partition Read-Side Critical Sections</a></h3>
+
+<p>
+It is tempting to assume that if any part of one RCU read-side critical
+section precedes a given grace period, and if any part of another RCU
+read-side critical section follows that same grace period, then all of
+the first RCU read-side critical section must precede all of the second.
+However, this just isn't the case: A single grace period does not
+partition the set of RCU read-side critical sections.
+An example of this situation can be illustrated as follows, where
+<tt>x</tt>, <tt>y</tt>, and <tt>z</tt> are initially all zero:
+
+<blockquote>
+<pre>
+ 1 void thread0(void)
+ 2 {
+ 3   rcu_read_lock();
+ 4   WRITE_ONCE(a, 1);
+ 5   WRITE_ONCE(b, 1);
+ 6   rcu_read_unlock();
+ 7 }
+ 8
+ 9 void thread1(void)
+10 {
+11   r1 = READ_ONCE(a);
+12   synchronize_rcu();
+13   WRITE_ONCE(c, 1);
+14 }
+15
+16 void thread2(void)
+17 {
+18   rcu_read_lock();
+19   r2 = READ_ONCE(b);
+20   r3 = READ_ONCE(c);
+21   rcu_read_unlock();
+22 }
+</pre>
+</blockquote>
+
+<p>
+It turns out that the outcome:
+
+<blockquote>
+<pre>
+(r1 == 1 &amp;&amp; r2 == 0 &amp;&amp; r3 == 1)
+</pre>
+</blockquote>
+
+is entirely possible.
+The following figure show how this can happen, with each circled
+<tt>QS</tt> indicating the point at which RCU recorded a
+<i>quiescent state</i> for each thread, that is, a state in which
+RCU knows that the thread cannot be in the midst of an RCU read-side
+critical section that started before the current grace period:
+
+<p><img src="GPpartitionReaders1.svg" alt="GPpartitionReaders1.svg" width="60%"></p>
+
+<p>
+If it is necessary to partition RCU read-side critical sections in this
+manner, it is necessary to use two grace periods, where the first
+grace period is known to end before the second grace period starts:
+
+<blockquote>
+<pre>
+ 1 void thread0(void)
+ 2 {
+ 3   rcu_read_lock();
+ 4   WRITE_ONCE(a, 1);
+ 5   WRITE_ONCE(b, 1);
+ 6   rcu_read_unlock();
+ 7 }
+ 8
+ 9 void thread1(void)
+10 {
+11   r1 = READ_ONCE(a);
+12   synchronize_rcu();
+13   WRITE_ONCE(c, 1);
+14 }
+15
+16 void thread2(void)
+17 {
+18   r2 = READ_ONCE(c);
+19   synchronize_rcu();
+20   WRITE_ONCE(d, 1);
+21 }
+22
+23 void thread3(void)
+24 {
+25   rcu_read_lock();
+26   r3 = READ_ONCE(b);
+27   r4 = READ_ONCE(d);
+28   rcu_read_unlock();
+29 }
+</pre>
+</blockquote>
+
+<p>
+Here, if <tt>(r1 == 1)</tt>, then
+<tt>thread0()</tt>'s write to <tt>b</tt> must happen
+before the end of <tt>thread1()</tt>'s grace period.
+If in addition <tt>(r4 == 1)</tt>, then
+<tt>thread3()</tt>'s read from <tt>b</tt> must happen
+after the beginning of <tt>thread2()</tt>'s grace period.
+If it is also the case that <tt>(r2 == 1)</tt>, then the
+end of <tt>thread1()</tt>'s grace period must precede the
+beginning of <tt>thread2()</tt>'s grace period.
+This mean that the two RCU read-side critical sections cannot overlap,
+guaranteeing that <tt>(r3 == 1)</tt>.
+As a result, the outcome:
+
+<blockquote>
+<pre>
+(r1 == 1 &amp;&amp; r2 == 1 &amp;&amp; r3 == 0 &amp;&amp; r4 == 1)
+</pre>
+</blockquote>
+
+cannot happen.
+
+<p>
+This non-requirement was also non-premeditated, but became apparent
+when studying RCU's interaction with memory ordering.
+
+<h3><a name="Read-Side Critical Sections Don't Partition Grace Periods">
+Read-Side Critical Sections Don't Partition Grace Periods</a></h3>
+
+<p>
+It is also tempting to assume that if an RCU read-side critical section
+happens between a pair of grace periods, then those grace periods cannot
+overlap.
+However, this temptation leads nowhere good, as can be illustrated by
+the following, with all variables initially zero:
+
+<blockquote>
+<pre>
+ 1 void thread0(void)
+ 2 {
+ 3   rcu_read_lock();
+ 4   WRITE_ONCE(a, 1);
+ 5   WRITE_ONCE(b, 1);
+ 6   rcu_read_unlock();
+ 7 }
+ 8
+ 9 void thread1(void)
+10 {
+11   r1 = READ_ONCE(a);
+12   synchronize_rcu();
+13   WRITE_ONCE(c, 1);
+14 }
+15
+16 void thread2(void)
+17 {
+18   rcu_read_lock();
+19   WRITE_ONCE(d, 1);
+20   r2 = READ_ONCE(c);
+21   rcu_read_unlock();
+22 }
+23
+24 void thread3(void)
+25 {
+26   r3 = READ_ONCE(d);
+27   synchronize_rcu();
+28   WRITE_ONCE(e, 1);
+29 }
+30
+31 void thread4(void)
+32 {
+33   rcu_read_lock();
+34   r4 = READ_ONCE(b);
+35   r5 = READ_ONCE(e);
+36   rcu_read_unlock();
+37 }
+</pre>
+</blockquote>
+
+<p>
+In this case, the outcome:
+
+<blockquote>
+<pre>
+(r1 == 1 &amp;&amp; r2 == 1 &amp;&amp; r3 == 1 &amp;&amp; r4 == 0 &amp&amp; r5 == 1)
+</pre>
+</blockquote>
+
+is entirely possible, as illustrated below:
+
+<p><img src="ReadersPartitionGP1.svg" alt="ReadersPartitionGP1.svg" width="100%"></p>
+
+<p>
+Again, an RCU read-side critical section can overlap almost all of a
+given grace period, just so long as it does not overlap the entire
+grace period.
+As a result, an RCU read-side critical section cannot partition a pair
+of RCU grace periods.
+
+<p><a name="Quick Quiz 10"><b>Quick Quiz 10</b>:</a>
+How long a sequence of grace periods, each separated by an RCU read-side
+critical section, would be required to partition the RCU read-side
+critical sections at the beginning and end of the chain?
+<br><a href="#qq10answer">Answer</a>
+
+<h3><a name="Disabling Preemption Does Not Block Grace Periods">
+Disabling Preemption Does Not Block Grace Periods</a></h3>
+
+<p>
+There was a time when disabling preemption on any given CPU would block
+subsequent grace periods.
+However, this was an accident of implementation and is not a requirement.
+And in the current Linux-kernel implementation, disabling preemption
+on a given CPU in fact does not block grace periods, as Oleg Nesterov
+<a href="http://marc.info/?l=linux-kernel&amp;m=143431078202914&amp;w=2">demonstrated</a>.
+
+<p>
+If you need a preempt-disable region to block grace periods, you need to add
+<tt>rcu_read_lock()</tt> and <tt>rcu_read_unlock()</tt>, for example
+as follows:
+
+<blockquote>
+<pre>
+ 1 preempt_disable();
+ 2 rcu_read_lock();
+ 3 do_something();
+ 4 rcu_read_unlock();
+ 5 preempt_enable();
+ 6
+ 7 /* Spinlocks implicitly disable preemption. */
+ 8 spin_lock(&amp;mylock);
+ 9 rcu_read_lock();
+10 do_something();
+11 rcu_read_unlock();
+12 spin_unlock(&amp;mylock);
+</pre>
+</blockquote>
+
+<p>
+In theory, you could enter the RCU read-side critical section first,
+but it is more efficient to keep the entire RCU read-side critical
+section contained in the preempt-disable region as shown above.
+Of course, RCU read-side critical sections that extend outside of
+preempt-disable regions will work correctly, but such critical sections
+can be preempted, which forces <tt>rcu_read_unlock()</tt> to do
+more work.
+And no, this is <i>not</i> an invitation to enclose all of your RCU
+read-side critical sections within preempt-disable regions, because
+doing so would degrade real-time response.
+
+<p>
+This non-requirement appeared with preemptible RCU.
+If you need a grace period that waits on non-preemptible code regions, use
+<a href="#Sched Flavor">RCU-sched</a>.
+
+<h2><a name="Parallelism Facts of Life">Parallelism Facts of Life</a></h2>
+
+<p>
+These parallelism facts of life are by no means specific to RCU, but
+the RCU implementation must abide by them.
+They therefore bear repeating:
+
+<ol>
+<li>	Any CPU or task may be delayed at any time,
+	and any attempts to avoid these delays by disabling
+	preemption, interrupts, or whatever are completely futile.
+	This is most obvious in preemptible user-level
+	environments and in virtualized environments (where
+	a given guest OS's VCPUs can be preempted at any time by
+	the underlying hypervisor), but can also happen in bare-metal
+	environments due to ECC errors, NMIs, and other hardware
+	events.
+	Although a delay of more than about 20 seconds can result
+	in splats, the RCU implementation is obligated to use
+	algorithms that can tolerate extremely long delays, but where
+	&ldquo;extremely long&rdquo; is not long enough to allow
+	wrap-around when incrementing a 64-bit counter.
+<li>	Both the compiler and the CPU can reorder memory accesses.
+	Where it matters, RCU must use compiler directives and
+	memory-barrier instructions to preserve ordering.
+<li>	Conflicting writes to memory locations in any given cache line
+	will result in expensive cache misses.
+	Greater numbers of concurrent writes and more-frequent
+	concurrent writes will result in more dramatic slowdowns.
+	RCU is therefore obligated to use algorithms that have
+	sufficient locality to avoid significant performance and
+	scalability problems.
+<li>	As a rough rule of thumb, only one CPU's worth of processing
+	may be carried out under the protection of any given exclusive
+	lock.
+	RCU must therefore use scalable locking designs.
+<li>	Counters are finite, especially on 32-bit systems.
+	RCU's use of counters must therefore tolerate counter wrap,
+	or be designed such that counter wrap would take way more
+	time than a single system is likely to run.
+	An uptime of ten years is quite possible, a runtime
+	of a century much less so.
+	As an example of the latter, RCU's dyntick-idle nesting counter
+	allows 54 bits for interrupt nesting level (this counter
+	is 64 bits even on a 32-bit system).
+	Overflowing this counter requires 2<sup>54</sup>
+	half-interrupts on a given CPU without that CPU ever going idle.
+	If a half-interrupt happened every microsecond, it would take
+	570 years of runtime to overflow this counter, which is currently
+	believed to be an acceptably long time.
+<li>	Linux systems can have thousands of CPUs running a single
+	Linux kernel in a single shared-memory environment.
+	RCU must therefore pay close attention to high-end scalability.
+</ol>
+
+<p>
+This last parallelism fact of life means that RCU must pay special
+attention to the preceding facts of life.
+The idea that Linux might scale to systems with thousands of CPUs would
+have been met with some skepticism in the 1990s, but these requirements
+would have otherwise have been unsurprising, even in the early 1990s.
+
+<h2><a name="Quality-of-Implementation Requirements">Quality-of-Implementation Requirements</a></h2>
+
+<p>
+These sections list quality-of-implementation requirements.
+Although an RCU implementation that ignores these requirements could
+still be used, it would likely be subject to limitations that would
+make it inappropriate for industrial-strength production use.
+Classes of quality-of-implementation requirements are as follows:
+
+<ol>
+<li>	<a href="#Specialization">Specialization</a>
+<li>	<a href="#Performance and Scalability">Performance and Scalability</a>
+<li>	<a href="#Composability">Composability</a>
+<li>	<a href="#Corner Cases">Corner Cases</a>
+</ol>
+
+<p>
+These classes is covered in the following sections.
+
+<h3><a name="Specialization">Specialization</a></h3>
+
+<p>
+RCU is and always has been intended primarily for read-mostly situations, as
+illustrated by the following figure.
+This means that RCU's read-side primitives are optimized, often at the
+expense of its update-side primitives.
+
+<p><img src="RCUApplicability.svg" alt="RCUApplicability.svg" width="70%"></p>
+
+<p>
+This focus on read-mostly situations means that RCU must interoperate
+with other synchronization primitives.
+For example, the <tt>add_gp()</tt> and <tt>remove_gp_synchronous()</tt>
+examples discussed earlier use RCU to protect readers and locking to
+coordinate updaters.
+However, the need extends much farther, requiring that a variety of
+synchronization primitives be legal within RCU read-side critical sections,
+including spinlocks, sequence locks, atomic operations, reference
+counters, and memory barriers.
+
+<p><a name="Quick Quiz 11"><b>Quick Quiz 11</b>:</a>
+What about sleeping locks?
+<br><a href="#qq11answer">Answer</a>
+
+<p>
+It often comes as a surprise that many algorithms do not require a
+consistent view of data, but many can function in that mode,
+with network routing being the poster child.
+Internet routing algorithms take significant time to propagate
+updates, so that by the time an update arrives at a given system,
+that system has been sending network traffic the wrong way for
+a considerable length of time.
+Having a few threads continue to send traffic the wrong way for a
+few more milliseconds is clearly not a problem:  In the worst case,
+TCP retransmissions will eventually get the data where it needs to go.
+In general, when tracking the state of the universe outside of the
+computer, some level of inconsistency must be tolerated due to
+speed-of-light delays if nothing else.
+
+<p>
+Furthermore, uncertainty about external state is inherent in many cases.
+For example, a pair of veternarians might use heartbeat to determine
+whether or not a given cat was alive.
+But how long should they wait after the last heartbeat to decide that
+the cat is in fact dead?
+Waiting less than 400 milliseconds makes no sense because this would
+mean that a relaxed cat would be considered to cycle between death
+and life more than 100 times per minute.
+Moreover, just as with human beings, a cat's heart might stop for
+some period of time, so the exact wait period is a judgment call.
+One of our pair of veternarians might wait 30 seconds before pronouncing
+the cat dead, while the other might insist on waiting a full minute.
+The two veternarians would then disagree on the state of the cat during
+the final 30 seconds of the minute following the last heartbeat, as
+fancifully illustrated below:
+
+<p><img src="2013-08-is-it-dead.png" alt="2013-08-is-it-dead.png" width="431"></p>
+
+<p>
+Interestingly enough, this same situation applies to hardware.
+When push comes to shove, how do we tell whether or not some
+external server has failed?
+We send messages to it periodically, and declare it failed if we
+don't receive a response within a given period of time.
+Policy decisions can usually tolerate short
+periods of inconsistency.
+The policy was decided some time ago, and is only now being put into
+effect, so a few milliseconds of delay is normally inconsequential.
+
+<p>
+However, there are algorithms that absolutely must see consistent data.
+For example, the translation between a user-level SystemV semaphore
+ID to the corresponding in-kernel data structure is protected by RCU,
+but it is absolutely forbidden to update a semaphore that has just been
+removed.
+In the Linux kernel, this need for consistency is accommodated by acquiring
+spinlocks located in the in-kernel data structure from within
+the RCU read-side critical section, and this is indicated by the
+green box in the figure above.
+Many other techniques may be used, and are in fact used within the
+Linux kernel.
+
+<p>
+In short, RCU is not required to maintain consistency, and other
+mechanisms may be used in concert with RCU when consistency is required.
+RCU's specialization allows it to do its job extremely well, and its
+ability to interoperate with other synchronization mechanisms allows
+the right mix of synchronization tools to be used for a given job.
+
+<h3><a name="Performance and Scalability">Performance and Scalability</a></h3>
+
+<p>
+Energy efficiency is a critical component of performance today,
+and Linux-kernel RCU implementations must therefore avoid unnecessarily
+awakening idle CPUs.
+I cannot claim that this requirement was premeditated.
+In fact, I learned of it during a telephone conversation in which I
+was given &ldquo;frank and open&rdquo; feedback on the importance
+of energy efficiency in battery-powered systems and on specific
+energy-efficiency shortcomings of the Linux-kernel RCU implementation.
+In my experience, the battery-powered embedded community will consider
+any unnecessary wakeups to be extremely unfriendly acts.
+So much so that mere Linux-kernel-mailing-list posts are
+insufficient to vent their ire.
+
+<p>
+Memory consumption is not particularly important for in most
+situations, and has become decreasingly
+so as memory sizes have expanded and memory
+costs have plummeted.
+However, as I learned from Matt Mackall's
+<a href="http://elinux.org/Linux_Tiny-FAQ">bloatwatch</a>
+efforts, memory footprint is critically important on single-CPU systems with
+non-preemptible (<tt>CONFIG_PREEMPT=n</tt>) kernels, and thus
+<a href="http://lkml.org/lkml/2009/1/14/449">tiny RCU</a>
+was born.
+Josh Triplett has since taken over the small-memory banner with his
+<a href="https://tiny.wiki.kernel.org/">Linux kernel tinification</a>
+project, which resulted in
+<a href="#Sleepable RCU">SRCU</a>
+becoming optional for those kernels not needing it.
+
+<p>
+The remaining performance requirements are, for the most part,
+unsurprising.
+For example, in keeping with RCU's read-side specialization,
+<tt>rcu_dereference()</tt> should have negligible overhead (for
+example, suppression of a few minor compiler optimizations).
+Similarly, in non-preemptible environments, <tt>rcu_read_lock()</tt> and
+<tt>rcu_read_unlock()</tt> should have exactly zero overhead.
+
+<p>
+In preemptible environments, in the case where the RCU read-side
+critical section was not preempted (as will be the case for the
+highest-priority real-time process), <tt>rcu_read_lock()</tt> and
+<tt>rcu_read_unlock()</tt> should have minimal overhead.
+In particular, they should not contain atomic read-modify-write
+operations, memory-barrier instructions, preemption disabling,
+interrupt disabling, or backwards branches.
+However, in the case where the RCU read-side critical section was preempted,
+<tt>rcu_read_unlock()</tt> may acquire spinlocks and disable interrupts.
+This is why it is better to nest an RCU read-side critical section
+within a preempt-disable region than vice versa, at least in cases
+where that critical section is short enough to avoid unduly degrading
+real-time latencies.
+
+<p>
+The <tt>synchronize_rcu()</tt> grace-period-wait primitive is
+optimized for throughput.
+It may therefore incur several milliseconds of latency in addition to
+the duration of the longest RCU read-side critical section.
+On the other hand, multiple concurrent invocations of
+<tt>synchronize_rcu()</tt> are required to use batching optimizations
+so that they can be satisfied by a single underlying grace-period-wait
+operation.
+For example, in the Linux kernel, it is not unusual for a single
+grace-period-wait operation to serve more than
+<a href="https://www.usenix.org/conference/2004-usenix-annual-technical-conference/making-rcu-safe-deep-sub-millisecond-response">1,000 separate invocations</a>
+of <tt>synchronize_rcu()</tt>, thus amortizing the per-invocation
+overhead down to nearly zero.
+However, the grace-period optimization is also required to avoid
+measurable degradation of real-time scheduling and interrupt latencies.
+
+<p>
+In some cases, the multi-millisecond <tt>synchronize_rcu()</tt>
+latencies are unacceptable.
+In these cases, <tt>synchronize_rcu_expedited()</tt> may be used
+instead, reducing the grace-period latency down to a few tens of
+microseconds on small systems, at least in cases where the RCU read-side
+critical sections are short.
+There are currently no special latency requirements for
+<tt>synchronize_rcu_expedited()</tt> on large systems, but,
+consistent with the empirical nature of the RCU specification,
+that is subject to change.
+However, there most definitely are scalability requirements:
+A storm of <tt>synchronize_rcu_expedited()</tt> invocations on 4096
+CPUs should at least make reasonable forward progress.
+In return for its shorter latencies, <tt>synchronize_rcu_expedited()</tt>
+is permitted to impose modest degradation of real-time latency
+on non-idle online CPUs.
+That said, it will likely be necessary to take further steps to reduce this
+degradation, hopefully to roughly that of a scheduling-clock interrupt.
+
+<p>
+There are a number of situations where even
+<tt>synchronize_rcu_expedited()</tt>'s reduced grace-period
+latency is unacceptable.
+In these situations, the asynchronous <tt>call_rcu()</tt> can be
+used in place of <tt>synchronize_rcu()</tt> as follows:
+
+<blockquote>
+<pre>
+ 1 struct foo {
+ 2   int a;
+ 3   int b;
+ 4   struct rcu_head rh;
+ 5 };
+ 6
+ 7 static void remove_gp_cb(struct rcu_head *rhp)
+ 8 {
+ 9   struct foo *p = container_of(rhp, struct foo, rh);
+10
+11   kfree(p);
+12 }
+13
+14 bool remove_gp_asynchronous(void)
+15 {
+16   struct foo *p;
+17
+18   spin_lock(&amp;gp_lock);
+19   p = rcu_dereference(gp);
+20   if (!p) {
+21     spin_unlock(&amp;gp_lock);
+22     return false;
+23   }
+24   rcu_assign_pointer(gp, NULL);
+25   call_rcu(&amp;p-&gt;rh, remove_gp_cb);
+26   spin_unlock(&amp;gp_lock);
+27   return true;
+28 }
+</pre>
+</blockquote>
+
+<p>
+A definition of <tt>struct foo</tt> is finally needed, and appears
+on lines&nbsp;1-5.
+The function <tt>remove_gp_cb()</tt> is passed to <tt>call_rcu()</tt>
+on line&nbsp;25, and will be invoked after the end of a subsequent
+grace period.
+This gets the same effect as <tt>remove_gp_synchronous()</tt>,
+but without forcing the updater to wait for a grace period to elapse.
+The <tt>call_rcu()</tt> function may be used in a number of
+situations where neither <tt>synchronize_rcu()</tt> nor
+<tt>synchronize_rcu_expedited()</tt> would be legal,
+including within preempt-disable code, <tt>local_bh_disable()</tt> code,
+interrupt-disable code, and interrupt handlers.
+However, even <tt>call_rcu()</tt> is illegal within NMI handlers.
+The callback function (<tt>remove_gp_cb()</tt> in this case) will be
+executed within softirq (software interrupt) environment within the
+Linux kernel,
+either within a real softirq handler or under the protection
+of <tt>local_bh_disable()</tt>.
+In both the Linux kernel and in userspace, it is bad practice to
+write an RCU callback function that takes too long.
+Long-running operations should be relegated to separate threads or
+(in the Linux kernel) workqueues.
+
+<p><a name="Quick Quiz 12"><b>Quick Quiz 12</b>:</a>
+Why does line&nbsp;19 use <tt>rcu_access_pointer()</tt>?
+After all, <tt>call_rcu()</tt> on line&nbsp;25 stores into the
+structure, which would interact badly with concurrent insertions.
+Doesn't this mean that <tt>rcu_dereference()</tt> is required?
+<br><a href="#qq12answer">Answer</a>
+
+<p>
+However, all that <tt>remove_gp_cb()</tt> is doing is
+invoking <tt>kfree()</tt> on the data element.
+This is a common idiom, and is supported by <tt>kfree_rcu()</tt>,
+which allows &ldquo;fire and forget&rdquo; operation as shown below:
+
+<blockquote>
+<pre>
+ 1 struct foo {
+ 2   int a;
+ 3   int b;
+ 4   struct rcu_head rh;
+ 5 };
+ 6
+ 7 bool remove_gp_faf(void)
+ 8 {
+ 9   struct foo *p;
+10
+11   spin_lock(&amp;gp_lock);
+12   p = rcu_dereference(gp);
+13   if (!p) {
+14     spin_unlock(&amp;gp_lock);
+15     return false;
+16   }
+17   rcu_assign_pointer(gp, NULL);
+18   kfree_rcu(p, rh);
+19   spin_unlock(&amp;gp_lock);
+20   return true;
+21 }
+</pre>
+</blockquote>
+
+<p>
+Note that <tt>remove_gp_faf()</tt> simply invokes
+<tt>kfree_rcu()</tt> and proceeds, without any need to pay any
+further attention to the subsequent grace period and <tt>kfree()</tt>.
+It is permissible to invoke <tt>kfree_rcu()</tt> from the same
+environments as for <tt>call_rcu()</tt>.
+Interestingly enough, DYNIX/ptx had the equivalents of
+<tt>call_rcu()</tt> and <tt>kfree_rcu()</tt>, but not
+<tt>synchronize_rcu()</tt>.
+This was due to the fact that RCU was not heavily used within DYNIX/ptx,
+so the very few places that needed something like
+<tt>synchronize_rcu()</tt> simply open-coded it.
+
+<p><a name="Quick Quiz 13"><b>Quick Quiz 13</b>:</a>
+Earlier it was claimed that <tt>call_rcu()</tt> and
+<tt>kfree_rcu()</tt> allowed updaters to avoid being blocked
+by readers.
+But how can that be correct, given that the invocation of the callback
+and the freeing of the memory (respectively) must still wait for
+a grace period to elapse?
+<br><a href="#qq13answer">Answer</a>
+
+<p>
+But what if the updater must wait for the completion of code to be
+executed after the end of the grace period, but has other tasks
+that can be carried out in the meantime?
+The polling-style <tt>get_state_synchronize_rcu()</tt> and
+<tt>cond_synchronize_rcu()</tt> functions may be used for this
+purpose, as shown below:
+
+<blockquote>
+<pre>
+ 1 bool remove_gp_poll(void)
+ 2 {
+ 3   struct foo *p;
+ 4   unsigned long s;
+ 5
+ 6   spin_lock(&amp;gp_lock);
+ 7   p = rcu_access_pointer(gp);
+ 8   if (!p) {
+ 9     spin_unlock(&amp;gp_lock);
+10     return false;
+11   }
+12   rcu_assign_pointer(gp, NULL);
+13   spin_unlock(&amp;gp_lock);
+14   s = get_state_synchronize_rcu();
+15   do_something_while_waiting();
+16   cond_synchronize_rcu(s);
+17   kfree(p);
+18   return true;
+19 }
+</pre>
+</blockquote>
+
+<p>
+On line&nbsp;14, <tt>get_state_synchronize_rcu()</tt> obtains a
+&ldquo;cookie&rdquo; from RCU,
+then line&nbsp;15 carries out other tasks,
+and finally, line&nbsp;16 returns immediately if a grace period has
+elapsed in the meantime, but otherwise waits as required.
+The need for <tt>get_state_synchronize_rcu</tt> and
+<tt>cond_synchronize_rcu()</tt> has appeared quite recently,
+so it is too early to tell whether they will stand the test of time.
+
+<p>
+RCU thus provides a range of tools to allow updaters to strike the
+required tradeoff between latency, flexibility and CPU overhead.
+
+<h3><a name="Composability">Composability</a></h3>
+
+<p>
+Composability has received much attention in recent years, perhaps in part
+due to the collision of multicore hardware with object-oriented techniques
+designed in single-threaded environments for single-threaded use.
+And in theory, RCU read-side critical sections may be composed, and in
+fact may be nested arbitrarily deeply.
+In practice, as with all real-world implementations of composable
+constructs, there are limitations.
+
+<p>
+Implementations of RCU for which <tt>rcu_read_lock()</tt>
+and <tt>rcu_read_unlock()</tt> generate no code, such as
+Linux-kernel RCU when <tt>CONFIG_PREEMPT=n</tt>, can be
+nested arbitrarily deeply.
+After all, there is no overhead.
+Except that if all these instances of <tt>rcu_read_lock()</tt>
+and <tt>rcu_read_unlock()</tt> are visible to the compiler,
+compilation will eventually fail due to exhausting memory,
+mass storage, or user patience, whichever comes first.
+If the nesting is not visible to the compiler, as is the case with
+mutually recursive functions each in its own translation unit,
+stack overflow will result.
+If the nesting takes the form of loops, either the control variable
+will overflow or (in the Linux kernel) you will get an RCU CPU stall warning.
+Nevertheless, this class of RCU implementations is one
+of the most composable constructs in existence.
+
+<p>
+RCU implementations that explicitly track nesting depth
+are limited by the nesting-depth counter.
+For example, the Linux kernel's preemptible RCU limits nesting to
+<tt>INT_MAX</tt>.
+This should suffice for almost all practical purposes.
+That said, a consecutive pair of RCU read-side critical sections
+between which there is an operation that waits for a grace period
+cannot be enclosed in another RCU read-side critical section.
+This is because it is not legal to wait for a grace period within
+an RCU read-side critical section:  To do so would result either
+in deadlock or
+in RCU implicitly splitting the enclosing RCU read-side critical
+section, neither of which is conducive to a long-lived and prosperous
+kernel.
+
+<p>
+In short, although RCU read-side critical sections are highly composable,
+care is required in some situations, just as is the case for any other
+composable synchronization mechanism.
+
+<h3><a name="Corner Cases">Corner Cases</a></h3>
+
+<p>
+A given RCU workload might have an endless and intense stream of
+RCU read-side critical sections, perhaps even so intense that there
+was never a point in time during which there was not at least one
+RCU read-side critical section in flight.
+RCU cannot allow this situation to block grace periods:  As long as
+all the RCU read-side critical sections are finite, grace periods
+must also be finite.
+
+<p>
+That said, preemptible RCU implementations could potentially result
+in RCU read-side critical sections being preempted for long durations,
+which has the effect of creating a long-duration RCU read-side
+critical section.
+This situation can arise only in heavily loaded systems, but systems using
+real-time priorities are of course more vulnerable.
+Therefore, RCU priority boosting is provided to help deal with this
+case.
+That said, the exact requirements on RCU priority boosting will likely
+evolve as more experience accumulates.
+
+<p>
+Other workloads might have very high update rates.
+Although one can argue that such workloads should instead use
+something other than RCU, the fact remains that RCU must
+handle such workloads gracefully.
+This requirement is another factor driving batching of grace periods,
+but it is also the driving force behind the checks for large numbers
+of queued RCU callbacks in the <tt>call_rcu()</tt> code path.
+Finally, high update rates should not delay RCU read-side critical
+sections, although some read-side delays can occur when using
+<tt>synchronize_rcu_expedited()</tt>, courtesy of this function's use
+of <tt>try_stop_cpus()</tt>.
+(In the future, <tt>synchronize_rcu_expedited()</tt> will be
+converted to use lighter-weight inter-processor interrupts (IPIs),
+but this will still disturb readers, though to a much smaller degree.)
+
+<p>
+Although all three of these corner cases were understood in the early
+1990s, a simple user-level test consisting of <tt>close(open(path))</tt>
+in a tight loop
+in the early 2000s suddenly provided a much deeper appreciation of the
+high-update-rate corner case.
+This test also motivated addition of some RCU code to react to high update
+rates, for example, if a given CPU finds itself with more than 10,000
+RCU callbacks queued, it will cause RCU to take evasive action by
+more aggressively starting grace periods and more aggressively forcing
+completion of grace-period processing.
+This evasive action causes the grace period to complete more quickly,
+but at the cost of restricting RCU's batching optimizations, thus
+increasing the CPU overhead incurred by that grace period.
+
+<h2><a name="Software-Engineering Requirements">
+Software-Engineering Requirements</a></h2>
+
+<p>
+Between Murphy's Law and &ldquo;To err is human&rdquo;, it is necessary to
+guard against mishaps and misuse:
+
+<ol>
+<li>	It is all too easy to forget to use <tt>rcu_read_lock()</tt>
+	everywhere that it is needed, so kernels built with
+	<tt>CONFIG_PROVE_RCU=y</tt> will spat if
+	<tt>rcu_dereference()</tt> is used outside of an
+	RCU read-side critical section.
+	Update-side code can use <tt>rcu_dereference_protected()</tt>,
+	which takes a
+	<a href="https://lwn.net/Articles/371986/">lockdep expression</a>
+	to indicate what is providing the protection.
+	If the indicated protection is not provided, a lockdep splat
+	is emitted.
+
+	<p>
+	Code shared between readers and updaters can use
+	<tt>rcu_dereference_check()</tt>, which also takes a
+	lockdep expression, and emits a lockdep splat if neither
+	<tt>rcu_read_lock()</tt> nor the indicated protection
+	is in place.
+	In addition, <tt>rcu_dereference_raw()</tt> is used in those
+	(hopefully rare) cases where the required protection cannot
+	be easily described.
+	Finally, <tt>rcu_read_lock_held()</tt> is provided to
+	allow a function to verify that it has been invoked within
+	an RCU read-side critical section.
+	I was made aware of this set of requirements shortly after Thomas
+	Gleixner audited a number of RCU uses.
+<li>	A given function might wish to check for RCU-related preconditions
+	upon entry, before using any other RCU API.
+	The <tt>rcu_lockdep_assert()</tt> does this job,
+	asserting the expression in kernels having lockdep enabled
+	and doing nothing otherwise.
+<li>	It is also easy to forget to use <tt>rcu_assign_pointer()</tt>
+	and <tt>rcu_dereference()</tt>, perhaps (incorrectly)
+	substituting a simple assignment.
+	To catch this sort of error, a given RCU-protected pointer may be
+	tagged with <tt>__rcu</tt>, after which running sparse
+	with <tt>CONFIG_SPARSE_RCU_POINTER=y</tt> will complain
+	about simple-assignment accesses to that pointer.
+	Arnd Bergmann made me aware of this requirement, and also
+	supplied the needed
+	<a href="http://lwn.net/Articles/376011/">patch series</a>.
+<li>	Kernels built with <tt>CONFIG_DEBUG_OBJECTS_RCU_HEAD=y</tt>
+	will splat if a data element is passed to <tt>call_rcu()</tt>
+	twice in a row, without a grace period in between.
+	(This error is similar to a double free.)
+	The corresponding <tt>rcu_head</tt> structures that are
+	dynamically allocated are automatically tracked, but
+	<tt>rcu_head</tt> structures allocated on the stack
+	must be initialized with <tt>init_rcu_head_on_stack()</tt>
+	and cleaned up with <tt>destroy_rcu_head_on_stack()</tt>.
+	Similarly, statically allocated non-stack <tt>rcu_head</tt>
+	structures must be initialized with <tt>init_rcu_head()</tt>
+	and cleaned up with <tt>destroy_rcu_head()</tt>.
+	Mathieu Desnoyers made me aware of this requirement, and also
+	supplied the needed
+	<a href="https://lkml.org/lkml/2010/3/18/417">patch</a>.
+<li>	An infinite loop in an RCU read-side critical section will
+	eventually trigger an RCU CPU stall warning splat.
+	However, RCU is not obligated to produce this splat
+	unless there is a grace period waiting on that particular
+	RCU read-side critical section.
+	This requirement made itself known in the early 1990s, pretty
+	much the first time that it was necessary to debug a CPU stall.
+<li>	Although it would be very good to detect pointers leaking out
+	of RCU read-side critical sections, there is currently no
+	good way of doing this.
+	One complication is the need to distinguish between pointers
+	leaking and pointers that have been handed off from RCU to
+	some other synchronization mechanism, for example, reference
+	counting.
+<li>	In kernels built with <tt>CONFIG_RCU_TRACE=y</tt>, RCU-related
+	information is provided via both debugfs and event tracing.
+<li>	Open-coded use of <tt>rcu_assign_pointer()</tt> and
+	<tt>rcu_dereference()</tt> to create typical linked
+	data structures can be surprisingly error-prone.
+	Therefore, RCU-protected
+	<a href="http://lwn.net/Articles/609973/#RCU List APIs">linked lists</a>
+	and, more recently, RCU-protected
+	<a href="https://lwn.net/Articles/612100/">hash tables</a>
+	are available.
+	Many other special-purpose RCU-protected data structures are
+	available in the Linux kernel and the userspace RCU library.
+<li>	Some linked structures are created at compile time, but still
+	require <tt>__rcu</tt> checking.
+	The <tt>RCU_POINTER_INITIALIZER()</tt> macro serves this
+	purpose.
+<li>	It is not necessary to use <tt>rcu_assign_pointer()</tt>
+	when creating linked structures that are to be published via
+	a single external pointer.
+	The <tt>RCU_INIT_POINTER()</tt> macro is provided for
+	this task and also for assigning <tt>NULL</tt> pointers
+	at runtime.
+</ol>
+
+<p>
+This not a hard-and-fast list:  RCU's diagnostic capabilities will
+continue to be guided by the number and type of usage bugs found
+in real-world RCU usage.
+
+<h2><a name="Linux Kernel Complications">Linux Kernel Complications</a></h2>
+
+<p>
+The Linux kernel provides an interesting environment for all kinds of
+software, including RCU.
+Some of the relevant points of interest are as follows:
+
+<ol>
+<li>	<a href="#Configuration">Configuration</a>.
+<li>	<a href="#Firmware Interface">Firmware Interface</a>.
+<li>	<a href="#Early Boot">Early Boot</a>.
+<li>	<a href="#Interrupts and NMIs">
+	Interrupts and non-maskable interrupts (NMIs)</a>.
+<li>	<a href="#Loadable Modules">Loadable Modules</a>.
+<li>	<a href="#Hotplug CPU">Hotplug CPU</a>.
+<li>	<a href="#Scheduler and RCU">Scheduler and RCU</a>.
+<li>	<a href="#Tracing and RCU">Tracing and RCU</a>.
+<li>	<a href="#Energy Efficiency">Energy Efficiency</a>.
+<li>	<a href="#Performance, Scalability, Response Time, and Reliability">
+	Performance, Scalability, Response Time, and Reliability</a>.
+</ol>
+
+<p>
+This list is probably incomplete, but it does give a feel for the
+most notable Linux-kernel complications.
+Each of the following sections covers one of the above topics.
+
+<h3><a name="Configuration">Configuration</a></h3>
+
+<p>
+RCU's goal is automatic configuration, so that almost nobody
+needs to worry about RCU's <tt>Kconfig</tt> options.
+And for almost all users, RCU does in fact work well
+&ldquo;out of the box.&rdquo;
+
+<p>
+However, there are specialized use cases that are handled by
+kernel boot parameters and <tt>Kconfig</tt> options.
+Unfortunately, the <tt>Kconfig</tt> system will explicitly ask users
+about new <tt>Kconfig</tt> options, which requires almost all of them
+be hidden behind a <tt>CONFIG_RCU_EXPERT</tt> <tt>Kconfig</tt> option.
+
+<p>
+This all should be quite obvious, but the fact remains that
+Linus Torvalds recently had to
+<a href="http://marc.info/?l=linux-kernel&amp;m=142905739823385">remind</a>
+me of this requirement.
+
+<h3><a name="Firmware Interface">Firmware Interface</a></h3>
+
+<p>
+In many cases, kernel obtains information about the system from the
+firmware, and sometimes things are lost in translation.
+Or the translation is accurate, but the original message is bogus.
+
+<p>
+For example, some systems' firmware overreports the number of CPUs,
+sometimes by a large factor.
+If RCU naively believed the firmware, as it used to do,
+it would create too many per-CPU kthreads.
+Although the resulting system will still run correctly, the extra
+kthreads needlessly consume memory and can cause confusion
+when they show up in <tt>ps</tt> listings.
+
+<p>
+RCU must therefore wait for a given CPU to actually come online before
+it can allow itself to believe that the CPU actually exists.
+The resulting &ldquo;ghost CPUs&rdquo; (which are never going to
+come online) cause a number of
+<a href="http://paulmck.livejournal.com/37494.html">interesting complications</a>.
+
+<h3><a name="Early Boot">Early Boot</a></h3>
+
+<p>
+The Linux kernel's boot sequence is an interesting process,
+and RCU is used early, even before <tt>rcu_init()</tt>
+is invoked.
+In fact, a number of RCU's primitives can be used as soon as the
+initial task's <tt>task_struct</tt> is available and the
+boot CPU's per-CPU variables are set up.
+The read-side primitives (<tt>rcu_read_lock()</tt>,
+<tt>rcu_read_unlock()</tt>, <tt>rcu_dereference()</tt>,
+and <tt>rcu_access_pointer()</tt>) will operate normally very early on,
+as will <tt>rcu_assign_pointer()</tt>.
+
+<p>
+Although <tt>call_rcu()</tt> may be invoked at any
+time during boot, callbacks are not guaranteed to be invoked until after
+the scheduler is fully up and running.
+This delay in callback invocation is due to the fact that RCU does not
+invoke callbacks until it is fully initialized, and this full initialization
+cannot occur until after the scheduler has initialized itself to the
+point where RCU can spawn and run its kthreads.
+In theory, it would be possible to invoke callbacks earlier,
+however, this is not a panacea because there would be severe restrictions
+on what operations those callbacks could invoke.
+
+<p>
+Perhaps surprisingly, <tt>synchronize_rcu()</tt>,
+<a href="#Bottom-Half Flavor"><tt>synchronize_rcu_bh()</tt></a>
+(<a href="#Bottom-Half Flavor">discussed below</a>),
+and
+<a href="#Sched Flavor"><tt>synchronize_sched()</tt></a>
+will all operate normally
+during very early boot, the reason being that there is only one CPU
+and preemption is disabled.
+This means that the call <tt>synchronize_rcu()</tt> (or friends)
+itself is a quiescent
+state and thus a grace period, so the early-boot implementation can
+be a no-op.
+
+<p>
+Both <tt>synchronize_rcu_bh()</tt> and <tt>synchronize_sched()</tt>
+continue to operate normally through the remainder of boot, courtesy
+of the fact that preemption is disabled across their RCU read-side
+critical sections and also courtesy of the fact that there is still
+only one CPU.
+However, once the scheduler starts initializing, preemption is enabled.
+There is still only a single CPU, but the fact that preemption is enabled
+means that the no-op implementation of <tt>synchronize_rcu()</tt> no
+longer works in <tt>CONFIG_PREEMPT=y</tt> kernels.
+Therefore, as soon as the scheduler starts initializing, the early-boot
+fastpath is disabled.
+This means that <tt>synchronize_rcu()</tt> switches to its runtime
+mode of operation where it posts callbacks, which in turn means that
+any call to <tt>synchronize_rcu()</tt> will block until the corresponding
+callback is invoked.
+Unfortunately, the callback cannot be invoked until RCU's runtime
+grace-period machinery is up and running, which cannot happen until
+the scheduler has initialized itself sufficiently to allow RCU's
+kthreads to be spawned.
+Therefore, invoking <tt>synchronize_rcu()</tt> during scheduler
+initialization can result in deadlock.
+
+<p><a name="Quick Quiz 14"><b>Quick Quiz 14</b>:</a>
+So what happens with <tt>synchronize_rcu()</tt> during
+scheduler initialization for <tt>CONFIG_PREEMPT=n</tt>
+kernels?
+<br><a href="#qq14answer">Answer</a>
+
+<p>
+I learned of these boot-time requirements as a result of a series of
+system hangs.
+
+<h3><a name="Interrupts and NMIs">Interrupts and NMIs</a></h3>
+
+<p>
+The Linux kernel has interrupts, and RCU read-side critical sections are
+legal within interrupt handlers and within interrupt-disabled regions
+of code, as are invocations of <tt>call_rcu()</tt>.
+
+<p>
+Some Linux-kernel architectures can enter an interrupt handler from
+non-idle process context, and then just never leave it, instead stealthily
+transitioning back to process context.
+This trick is sometimes used to invoke system calls from inside the kernel.
+These &ldquo;half-interrupts&rdquo; mean that RCU has to be very careful
+about how it counts interrupt nesting levels.
+I learned of this requirement the hard way during a rewrite
+of RCU's dyntick-idle code.
+
+<p>
+The Linux kernel has non-maskable interrupts (NMIs), and
+RCU read-side critical sections are legal within NMI handlers.
+Thankfully, RCU update-side primitives, including
+<tt>call_rcu()</tt>, are prohibited within NMI handlers.
+
+<p>
+The name notwithstanding, some Linux-kernel architectures
+can have nested NMIs, which RCU must handle correctly.
+Andy Lutomirski
+<a href="https://lkml.org/lkml/2014/11/21/642">surprised me</a>
+with this requirement;
+he also kindly surprised me with
+<a href="https://lkml.org/lkml/2014/11/22/1">an algorithm</a>
+that meets this requirement.
+
+<h3><a name="Loadable Modules">Loadable Modules</a></h3>
+
+<p>
+The Linux kernel has loadable modules, and these modules can
+also be unloaded.
+After a given module has been unloaded, any attempt to call
+one of its functions results in a segmentation fault.
+The module-unload functions must therefore cancel any
+delayed calls to loadable-module functions, for example,
+any outstanding <tt>mod_timer()</tt> must be dealt with
+via <tt>del_timer_sync()</tt> or similar.
+
+<p>
+Unfortunately, there is no way to cancel an RCU callback;
+once you invoke <tt>call_rcu()</tt>, the callback function is
+going to eventually be invoked, unless the system goes down first.
+Because it is normally considered socially irresponsible to crash the system
+in response to a module unload request, we need some other way
+to deal with in-flight RCU callbacks.
+
+<p>
+RCU therefore provides
+<tt><a href="http://lwn.net/Articles/217484/">rcu_barrier()</a></tt>,
+which waits until all in-flight RCU callbacks have been invoked.
+If a module uses <tt>call_rcu()</tt>, its exit function should therefore
+prevent any future invocation of <tt>call_rcu()</tt>, then invoke
+<tt>rcu_barrier()</tt>.
+In theory, the underlying module-unload code could invoke
+<tt>rcu_barrier()</tt> unconditionally, but in practice this would
+incur unacceptable latencies.
+
+<p>
+Nikita Danilov noted this requirement for an analogous filesystem-unmount
+situation, and Dipankar Sarma incorporated <tt>rcu_barrier()</tt> into RCU.
+The need for <tt>rcu_barrier()</tt> for module unloading became
+apparent later.
+
+<h3><a name="Hotplug CPU">Hotplug CPU</a></h3>
+
+<p>
+The Linux kernel supports CPU hotplug, which means that CPUs
+can come and go.
+It is of course illegal to use any RCU API member from an offline CPU.
+This requirement was present from day one in DYNIX/ptx, but
+on the other hand, the Linux kernel's CPU-hotplug implementation
+is &ldquo;interesting.&rdquo;
+
+<p>
+The Linux-kernel CPU-hotplug implementation has notifiers that
+are used to allow the various kernel subsystems (including RCU)
+to respond appropriately to a given CPU-hotplug operation.
+Most RCU operations may be invoked from CPU-hotplug notifiers,
+including even normal synchronous grace-period operations
+such as <tt>synchronize_rcu()</tt>.
+However, expedited grace-period operations such as
+<tt>synchronize_rcu_expedited()</tt> are not supported,
+due to the fact that current implementations block CPU-hotplug
+operations, which could result in deadlock.
+
+<p>
+In addition, all-callback-wait operations such as
+<tt>rcu_barrier()</tt> are also not supported, due to the
+fact that there are phases of CPU-hotplug operations where
+the outgoing CPU's callbacks will not be invoked until after
+the CPU-hotplug operation ends, which could also result in deadlock.
+
+<h3><a name="Scheduler and RCU">Scheduler and RCU</a></h3>
+
+<p>
+RCU depends on the scheduler, and the scheduler uses RCU to
+protect some of its data structures.
+This means the scheduler is forbidden from acquiring
+the runqueue locks and the priority-inheritance locks
+in the middle of an outermost RCU read-side critical section unless
+it also releases them before exiting that same
+RCU read-side critical section.
+This same prohibition also applies to any lock that is acquired
+while holding any lock to which this prohibition applies.
+Violating this rule results in deadlock.
+
+<p>
+For RCU's part, the preemptible-RCU <tt>rcu_read_unlock()</tt>
+implementation must be written carefully to avoid similar deadlocks.
+In particular, <tt>rcu_read_unlock()</tt> must tolerate an
+interrupt where the interrupt handler invokes both
+<tt>rcu_read_lock()</tt> and <tt>rcu_read_unlock()</tt>.
+This possibility requires <tt>rcu_read_unlock()</tt> to use
+negative nesting levels to avoid destructive recursion via
+interrupt handler's use of RCU.
+
+<p>
+This pair of mutual scheduler-RCU requirements came as a
+<a href="http://lwn.net/Articles/453002/">complete surprise</a>.
+
+<p>
+As noted above, RCU makes use of kthreads, and it is necessary to
+avoid excessive CPU-time accumulation by these kthreads.
+This requirement was no surprise, but RCU's violation of it
+when running context-switch-heavy workloads when built with
+<tt>CONFIG_NO_HZ_FULL=y</tt>
+<a href="http://www.rdrop.com/users/paulmck/scalability/paper/BareMetal.2015.01.15b.pdf">did come as a surprise [PDF]</a>.
+RCU has made good progress towards meeting this requirement, even
+for context-switch-have <tt>CONFIG_NO_HZ_FULL=y</tt> workloads,
+but there is room for further improvement.
+
+<h3><a name="Tracing and RCU">Tracing and RCU</a></h3>
+
+<p>
+It is possible to use tracing on RCU code, but tracing itself
+uses RCU.
+For this reason, <tt>rcu_dereference_raw_notrace()</tt>
+is provided for use by tracing, which avoids the destructive
+recursion that could otherwise ensue.
+This API is also used by virtualization in some architectures,
+where RCU readers execute in environments in which tracing
+cannot be used.
+The tracing folks both located the requirement and provided the
+needed fix, so this surprise requirement was relatively painless.
+
+<h3><a name="Energy Efficiency">Energy Efficiency</a></h3>
+
+<p>
+Interrupting idle CPUs is considered socially unacceptable,
+especially by people with battery-powered embedded systems.
+RCU therefore conserves energy by detecting which CPUs are
+idle, including tracking CPUs that have been interrupted from idle.
+This is a large part of the energy-efficiency requirement,
+so I learned of this via an irate phone call.
+
+<p>
+Because RCU avoids interrupting idle CPUs, it is illegal to
+execute an RCU read-side critical section on an idle CPU.
+(Kernels built with <tt>CONFIG_PROVE_RCU=y</tt> will splat
+if you try it.)
+The <tt>RCU_NONIDLE()</tt> macro and <tt>_rcuidle</tt>
+event tracing is provided to work around this restriction.
+In addition, <tt>rcu_is_watching()</tt> may be used to
+test whether or not it is currently legal to run RCU read-side
+critical sections on this CPU.
+I learned of the need for diagnostics on the one hand
+and <tt>RCU_NONIDLE()</tt> on the other while inspecting
+idle-loop code.
+Steven Rostedt supplied <tt>_rcuidle</tt> event tracing,
+which is used quite heavily in the idle loop.
+
+<p>
+It is similarly socially unacceptable to interrupt an
+<tt>nohz_full</tt> CPU running in userspace.
+RCU must therefore track <tt>nohz_full</tt> userspace
+execution.
+And in
+<a href="https://lwn.net/Articles/558284/"><tt>CONFIG_NO_HZ_FULL_SYSIDLE=y</tt></a>
+kernels, RCU must separately track idle CPUs on the one hand and
+CPUs that are either idle or executing in userspace on the other.
+In both cases, RCU must be able to sample state at two points in
+time, and be able to determine whether or not some other CPU spent
+any time idle and/or executing in userspace.
+
+<p>
+These energy-efficiency requirements have proven quite difficult to
+understand and to meet, for example, there have been more than five
+clean-sheet rewrites of RCU's energy-efficiency code, the last of
+which was finally able to demonstrate
+<a href="http://www.rdrop.com/users/paulmck/realtime/paper/AMPenergy.2013.04.19a.pdf">real energy savings running on real hardware [PDF]</a>.
+As noted earlier,
+I learned of many of these requirements via angry phone calls:
+Flaming me on the Linux-kernel mailing list was apparently not
+sufficient to fully vent their ire at RCU's energy-efficiency bugs!
+
+<h3><a name="Performance, Scalability, Response Time, and Reliability">
+Performance, Scalability, Response Time, and Reliability</a></h3>
+
+<p>
+Expanding on the
+<a href="#Performance and Scalability">earlier discussion</a>,
+RCU is used heavily by hot code paths in performance-critical
+portions of the Linux kernel's networking, security, virtualization,
+and scheduling code paths.
+RCU must therefore use efficient implementations, especially in its
+read-side primitives.
+To that end, it would be good if preemptible RCU's implementation
+of <tt>rcu_read_lock()</tt> could be inlined, however, doing
+this requires resolving <tt>#include</tt> issues with the
+<tt>task_struct</tt> structure.
+
+<p>
+The Linux kernel supports hardware configurations with up to
+4096 CPUs, which means that RCU must be extremely scalable.
+Algorithms that involve frequent acquisitions of global locks or
+frequent atomic operations on global variables simply cannot be
+tolerated within the RCU implementation.
+RCU therefore makes heavy use of a combining tree based on the
+<tt>rcu_node</tt> structure.
+RCU is required to tolerate all CPUs continuously invoking any
+combination of RCU's runtime primitives with minimal per-operation
+overhead.
+In fact, in many cases, increasing load must <i>decrease</i> the
+per-operation overhead, witness the batching optimizations for
+<tt>synchronize_rcu()</tt>, <tt>call_rcu()</tt>,
+<tt>synchronize_rcu_expedited()</tt>, and <tt>rcu_barrier()</tt>.
+As a general rule, RCU must cheerfully accept whatever the
+rest of the Linux kernel decides to throw at it.
+
+<p>
+The Linux kernel is used for real-time workloads, especially
+in conjunction with the
+<a href="https://rt.wiki.kernel.org/index.php/Main_Page">-rt patchset</a>.
+The real-time-latency response requirements are such that the
+traditional approach of disabling preemption across RCU
+read-side critical sections is inappropriate.
+Kernels built with <tt>CONFIG_PREEMPT=y</tt> therefore
+use an RCU implementation that allows RCU read-side critical
+sections to be preempted.
+This requirement made its presence known after users made it
+clear that an earlier
+<a href="http://lwn.net/Articles/107930/">real-time patch</a>
+did not meet their needs, in conjunction with some
+<a href="https://lkml.org/lkml/2005/3/17/199">RCU issues</a>
+encountered by a very early version of the -rt patchset.
+
+<p>
+In addition, RCU must make do with a sub-100-microsecond real-time latency
+budget.
+In fact, on smaller systems with the -rt patchset, the Linux kernel
+provides sub-20-microsecond real-time latencies for the whole kernel,
+including RCU.
+RCU's scalability and latency must therefore be sufficient for
+these sorts of configurations.
+To my surprise, the sub-100-microsecond real-time latency budget
+<a href="http://www.rdrop.com/users/paulmck/realtime/paper/bigrt.2013.01.31a.LCA.pdf">
+applies to even the largest systems [PDF]</a>,
+up to and including systems with 4096 CPUs.
+This real-time requirement motivated the grace-period kthread, which
+also simplified handling of a number of race conditions.
+
+<p>
+Finally, RCU's status as a synchronization primitive means that
+any RCU failure can result in arbitrary memory corruption that can be
+extremely difficult to debug.
+This means that RCU must be extremely reliable, which in
+practice also means that RCU must have an aggressive stress-test
+suite.
+This stress-test suite is called <tt>rcutorture</tt>.
+
+<p>
+Although the need for <tt>rcutorture</tt> was no surprise,
+the current immense popularity of the Linux kernel is posing
+interesting&mdash;and perhaps unprecedented&mdash;validation
+challenges.
+To see this, keep in mind that there are well over one billion
+instances of the Linux kernel running today, given Android
+smartphones, Linux-powered televisions, and servers.
+This number can be expected to increase sharply with the advent of
+the celebrated Internet of Things.
+
+<p>
+Suppose that RCU contains a race condition that manifests on average
+once per million years of runtime.
+This bug will be occurring about three times per <i>day</i> across
+the installed base.
+RCU could simply hide behind hardware error rates, given that no one
+should really expect their smartphone to last for a million years.
+However, anyone taking too much comfort from this thought should
+consider the fact that in most jurisdictions, a successful multi-year
+test of a given mechanism, which might include a Linux kernel,
+suffices for a number of types of safety-critical certifications.
+In fact, rumor has it that the Linux kernel is already being used
+in production for safety-critical applications.
+I don't know about you, but I would feel quite bad if a bug in RCU
+killed someone.
+Which might explain my recent focus on validation and verification.
+
+<h2><a name="Other RCU Flavors">Other RCU Flavors</a></h2>
+
+<p>
+One of the more surprising things about RCU is that there are now
+no fewer than five <i>flavors</i>, or API families.
+In addition, the primary flavor that has been the sole focus up to
+this point has two different implementations, non-preemptible and
+preemptible.
+The other four flavors are listed below, with requirements for each
+described in a separate section.
+
+<ol>
+<li>	<a href="#Bottom-Half Flavor">Bottom-Half Flavor</a>
+<li>	<a href="#Sched Flavor">Sched Flavor</a>
+<li>	<a href="#Sleepable RCU">Sleepable RCU</a>
+<li>	<a href="#Tasks RCU">Tasks RCU</a>
+</ol>
+
+<h3><a name="Bottom-Half Flavor">Bottom-Half Flavor</a></h3>
+
+<p>
+The softirq-disable (AKA &ldquo;bottom-half&rdquo;,
+hence the &ldquo;_bh&rdquo; abbreviations)
+flavor of RCU, or <i>RCU-bh</i>, was developed by
+Dipankar Sarma to provide a flavor of RCU that could withstand the
+network-based denial-of-service attacks researched by Robert
+Olsson.
+These attacks placed so much networking load on the system
+that some of the CPUs never exited softirq execution,
+which in turn prevented those CPUs from ever executing a context switch,
+which, in the RCU implementation of that time, prevented grace periods
+from ever ending.
+The result was an out-of-memory condition and a system hang.
+
+<p>
+The solution was the creation of RCU-bh, which does
+<tt>local_bh_disable()</tt>
+across its read-side critical sections, and which uses the transition
+from one type of softirq processing to another as a quiescent state
+in addition to context switch, idle, user mode, and offline.
+This means that RCU-bh grace periods can complete even when some of
+the CPUs execute in softirq indefinitely, thus allowing algorithms
+based on RCU-bh to withstand network-based denial-of-service attacks.
+
+<p>
+Because
+<tt>rcu_read_lock_bh()</tt> and <tt>rcu_read_unlock_bh()</tt>
+disable and re-enable softirq handlers, any attempt to start a softirq
+handlers during the
+RCU-bh read-side critical section will be deferred.
+In this case, <tt>rcu_read_unlock_bh()</tt>
+will invoke softirq processing, which can take considerable time.
+One can of course argue that this softirq overhead should be associated
+with the code following the RCU-bh read-side critical section rather
+than <tt>rcu_read_unlock_bh()</tt>, but the fact
+is that most profiling tools cannot be expected to make this sort
+of fine distinction.
+For example, suppose that a three-millisecond-long RCU-bh read-side
+critical section executes during a time of heavy networking load.
+There will very likely be an attempt to invoke at least one softirq
+handler during that three milliseconds, but any such invocation will
+be delayed until the time of the <tt>rcu_read_unlock_bh()</tt>.
+This can of course make it appear at first glance as if
+<tt>rcu_read_unlock_bh()</tt> was executing very slowly.
+
+<p>
+The
+<a href="http://lwn.net/Articles/609973/#RCU Per-Flavor API Table">RCU-bh API</a>
+includes
+<tt>rcu_read_lock_bh()</tt>,
+<tt>rcu_read_unlock_bh()</tt>,
+<tt>rcu_dereference_bh()</tt>,
+<tt>rcu_dereference_bh_check()</tt>,
+<tt>synchronize_rcu_bh()</tt>,
+<tt>synchronize_rcu_bh_expedited()</tt>,
+<tt>call_rcu_bh()</tt>,
+<tt>rcu_barrier_bh()</tt>, and
+<tt>rcu_read_lock_bh_held()</tt>.
+
+<h3><a name="Sched Flavor">Sched Flavor</a></h3>
+
+<p>
+Before preemptible RCU, waiting for an RCU grace period had the
+side effect of also waiting for all pre-existing interrupt
+and NMI handlers.
+However, there are legitimate preemptible-RCU implementations that
+do not have this property, given that any point in the code outside
+of an RCU read-side critical section can be a quiescent state.
+Therefore, <i>RCU-sched</i> was created, which follows &ldquo;classic&rdquo;
+RCU in that an RCU-sched grace period waits for for pre-existing
+interrupt and NMI handlers.
+In kernels built with <tt>CONFIG_PREEMPT=n</tt>, the RCU and RCU-sched
+APIs have identical implementations, while kernels built with
+<tt>CONFIG_PREEMPT=y</tt> provide a separate implementation for each.
+
+<p>
+Note well that in <tt>CONFIG_PREEMPT=y</tt> kernels,
+<tt>rcu_read_lock_sched()</tt> and <tt>rcu_read_unlock_sched()</tt>
+disable and re-enable preemption, respectively.
+This means that if there was a preemption attempt during the
+RCU-sched read-side critical section, <tt>rcu_read_unlock_sched()</tt>
+will enter the scheduler, with all the latency and overhead entailed.
+Just as with <tt>rcu_read_unlock_bh()</tt>, this can make it look
+as if <tt>rcu_read_unlock_sched()</tt> was executing very slowly.
+However, the highest-priority task won't be preempted, so that task
+will enjoy low-overhead <tt>rcu_read_unlock_sched()</tt> invocations.
+
+<p>
+The
+<a href="http://lwn.net/Articles/609973/#RCU Per-Flavor API Table">RCU-sched API</a>
+includes
+<tt>rcu_read_lock_sched()</tt>,
+<tt>rcu_read_unlock_sched()</tt>,
+<tt>rcu_read_lock_sched_notrace()</tt>,
+<tt>rcu_read_unlock_sched_notrace()</tt>,
+<tt>rcu_dereference_sched()</tt>,
+<tt>rcu_dereference_sched_check()</tt>,
+<tt>synchronize_sched()</tt>,
+<tt>synchronize_rcu_sched_expedited()</tt>,
+<tt>call_rcu_sched()</tt>,
+<tt>rcu_barrier_sched()</tt>, and
+<tt>rcu_read_lock_sched_held()</tt>.
+However, anything that disables preemption also marks an RCU-sched
+read-side critical section, including
+<tt>preempt_disable()</tt> and <tt>preempt_enable()</tt>,
+<tt>local_irq_save()</tt> and <tt>local_irq_restore()</tt>,
+and so on.
+
+<h3><a name="Sleepable RCU">Sleepable RCU</a></h3>
+
+<p>
+For well over a decade, someone saying &ldquo;I need to block within
+an RCU read-side critical section&rdquo; was a reliable indication
+that this someone did not understand RCU.
+After all, if you are always blocking in an RCU read-side critical
+section, you can probably afford to use a higher-overhead synchronization
+mechanism.
+However, that changed with the advent of the Linux kernel's notifiers,
+whose RCU read-side critical
+sections almost never sleep, but sometimes need to.
+This resulted in the introduction of
+<a href="https://lwn.net/Articles/202847/">sleepable RCU</a>,
+or <i>SRCU</i>.
+
+<p>
+SRCU allows different domains to be defined, with each such domain
+defined by an instance of an <tt>srcu_struct</tt> structure.
+A pointer to this structure must be passed in to each SRCU function,
+for example, <tt>synchronize_srcu(&amp;ss)</tt>, where
+<tt>ss</tt> is the <tt>srcu_struct</tt> structure.
+The key benefit of these domains is that a slow SRCU reader in one
+domain does not delay an SRCU grace period in some other domain.
+That said, one consequence of these domains is that read-side code
+must pass a &ldquo;cookie&rdquo; from <tt>srcu_read_lock()</tt>
+to <tt>srcu_read_unlock()</tt>, for example, as follows:
+
+<blockquote>
+<pre>
+ 1 int idx;
+ 2
+ 3 idx = srcu_read_lock(&amp;ss);
+ 4 do_something();
+ 5 srcu_read_unlock(&amp;ss, idx);
+</pre>
+</blockquote>
+
+<p>
+As noted above, it is legal to block within SRCU read-side critical sections,
+however, with great power comes great responsibility.
+If you block forever in one of a given domain's SRCU read-side critical
+sections, then that domain's grace periods will also be blocked forever.
+Of course, one good way to block forever is to deadlock, which can
+happen if any operation in a given domain's SRCU read-side critical
+section can block waiting, either directly or indirectly, for that domain's
+grace period to elapse.
+For example, this results in a self-deadlock:
+
+<blockquote>
+<pre>
+ 1 int idx;
+ 2
+ 3 idx = srcu_read_lock(&amp;ss);
+ 4 do_something();
+ 5 synchronize_srcu(&amp;ss);
+ 6 srcu_read_unlock(&amp;ss, idx);
+</pre>
+</blockquote>
+
+<p>
+However, if line&nbsp;5 acquired a mutex that was held across
+a <tt>synchronize_srcu()</tt> for domain <tt>ss</tt>,
+deadlock would still be possible.
+Furthermore, if line&nbsp;5 acquired a mutex that was held across
+a <tt>synchronize_srcu()</tt> for some other domain <tt>ss1</tt>,
+and if an <tt>ss1</tt>-domain SRCU read-side critical section
+acquired another mutex that was held across as <tt>ss</tt>-domain
+<tt>synchronize_srcu()</tt>,
+deadlock would again be possible.
+Such a deadlock cycle could extend across an arbitrarily large number
+of different SRCU domains.
+Again, with great power comes great responsibility.
+
+<p>
+Unlike the other RCU flavors, SRCU read-side critical sections can
+run on idle and even offline CPUs.
+This ability requires that <tt>srcu_read_lock()</tt> and
+<tt>srcu_read_unlock()</tt> contain memory barriers, which means
+that SRCU readers will run a bit slower than would RCU readers.
+It also motivates the <tt>smp_mb__after_srcu_read_unlock()</tt>
+API, which, in combination with <tt>srcu_read_unlock()</tt>,
+guarantees a full memory barrier.
+
+<p>
+The
+<a href="http://lwn.net/Articles/609973/#RCU Per-Flavor API Table">SRCU API</a>
+includes
+<tt>srcu_read_lock()</tt>,
+<tt>srcu_read_unlock()</tt>,
+<tt>srcu_dereference()</tt>,
+<tt>srcu_dereference_check()</tt>,
+<tt>synchronize_srcu()</tt>,
+<tt>synchronize_srcu_expedited()</tt>,
+<tt>call_srcu()</tt>,
+<tt>srcu_barrier()</tt>, and
+<tt>srcu_read_lock_held()</tt>.
+It also includes
+<tt>DEFINE_SRCU()</tt>,
+<tt>DEFINE_STATIC_SRCU()</tt>, and
+<tt>init_srcu_struct()</tt>
+APIs for defining and initializing <tt>srcu_struct</tt> structures.
+
+<h3><a name="Tasks RCU">Tasks RCU</a></h3>
+
+<p>
+Some forms of tracing use &ldquo;tramopolines&rdquo; to handle the
+binary rewriting required to install different types of probes.
+It would be good to be able to free old trampolines, which sounds
+like a job for some form of RCU.
+However, because it is necessary to be able to install a trace
+anywhere in the code, it is not possible to use read-side markers
+such as <tt>rcu_read_lock()</tt> and <tt>rcu_read_unlock()</tt>.
+In addition, it does not work to have these markers in the trampoline
+itself, because there would need to be instructions following
+<tt>rcu_read_unlock()</tt>.
+Although <tt>synchronize_rcu()</tt> would guarantee that execution
+reached the <tt>rcu_read_unlock()</tt>, it would not be able to
+guarantee that execution had completely left the trampoline.
+
+<p>
+The solution, in the form of
+<a href="http://lwn.net/Articles/607117/"><i>Tasks RCU</i></a>,
+is to have implicit
+read-side critical sections that are delimited by voluntary context
+switches, that is, calls to <tt>schedule()</tt>,
+<tt>cond_resched_rcu_qs()</tt>, and
+<tt>synchronize_rcu_tasks()</tt>.
+In addition, transitions to and from userspace execution also delimit
+tasks-RCU read-side critical sections.
+
+<p>
+The tasks-RCU API is quite compact, consisting only of
+<tt>call_rcu_tasks()</tt>,
+<tt>synchronize_rcu_tasks()</tt>, and
+<tt>rcu_barrier_tasks()</tt>.
+
+<h2><a name="Possible Future Changes">Possible Future Changes</a></h2>
+
+<p>
+One of the tricks that RCU uses to attain update-side scalability is
+to increase grace-period latency with increasing numbers of CPUs.
+If this becomes a serious problem, it will be necessary to rework the
+grace-period state machine so as to avoid the need for the additional
+latency.
+
+<p>
+Expedited grace periods scan the CPUs, so their latency and overhead
+increases with increasing numbers of CPUs.
+If this becomes a serious problem on large systems, it will be necessary
+to do some redesign to avoid this scalability problem.
+
+<p>
+RCU disables CPU hotplug in a few places, perhaps most notably in the
+expedited grace-period and <tt>rcu_barrier()</tt> operations.
+If there is a strong reason to use expedited grace periods in CPU-hotplug
+notifiers, it will be necessary to avoid disabling CPU hotplug.
+This would introduce some complexity, so there had better be a <i>very</i>
+good reason.
+
+<p>
+The tradeoff between grace-period latency on the one hand and interruptions
+of other CPUs on the other hand may need to be re-examined.
+The desire is of course for zero grace-period latency as well as zero
+interprocessor interrupts undertaken during an expedited grace period
+operation.
+While this ideal is unlikely to be achievable, it is quite possible that
+further improvements can be made.
+
+<p>
+The multiprocessor implementations of RCU use a combining tree that
+groups CPUs so as to reduce lock contention and increase cache locality.
+However, this combining tree does not spread its memory across NUMA
+nodes nor does it align the CPU groups with hardware features such
+as sockets or cores.
+Such spreading and alignment is currently believed to be unnecessary
+because the hotpath read-side primitives do not access the combining
+tree, nor does <tt>call_rcu()</tt> in the common case.
+If you believe that your architecture needs such spreading and alignment,
+then your architecture should also benefit from the
+<tt>rcutree.rcu_fanout_leaf</tt> boot parameter, which can be set
+to the number of CPUs in a socket, NUMA node, or whatever.
+If the number of CPUs is too large, use a fraction of the number of
+CPUs.
+If the number of CPUs is a large prime number, well, that certainly
+is an &ldquo;interesting&rdquo; architectural choice!
+More flexible arrangements might be considered, but only if
+<tt>rcutree.rcu_fanout_leaf</tt> has proven inadequate, and only
+if the inadequacy has been demonstrated by a carefully run and
+realistic system-level workload.
+
+<p>
+Please note that arrangements that require RCU to remap CPU numbers will
+require extremely good demonstration of need and full exploration of
+alternatives.
+
+<p>
+There is an embarrassingly large number of flavors of RCU, and this
+number has been increasing over time.
+Perhaps it will be possible to combine some at some future date.
+
+<p>
+RCU's various kthreads are reasonably recent additions.
+It is quite likely that adjustments will be required to more gracefully
+handle extreme loads.
+It might also be necessary to be able to relate CPU utilization by
+RCU's kthreads and softirq handlers to the code that instigated this
+CPU utilization.
+For example, RCU callback overhead might be charged back to the
+originating <tt>call_rcu()</tt> instance, though probably not
+in production kernels.
+
+<h2><a name="Summary">Summary</a></h2>
+
+<p>
+This document has presented more than two decade's worth of RCU
+requirements.
+Given that the requirements keep changing, this will not be the last
+word on this subject, but at least it serves to get an important
+subset of the requirements set forth.
+
+<h2><a name="Acknowledgments">Acknowledgments</a></h2>
+
+I am grateful to Steven Rostedt, Lai Jiangshan, Ingo Molnar,
+Oleg Nesterov, Borislav Petkov, Peter Zijlstra, Boqun Feng, and
+Andy Lutomirski for their help in rendering
+this article human readable, and to Michelle Rankin for her support
+of this effort.
+Other contributions are acknowledged in the Linux kernel's git archive.
+The cartoon is copyright (c) 2013 by Melissa Broussard,
+and is provided
+under the terms of the Creative Commons Attribution-Share Alike 3.0
+United States license.
+
+<h3><a name="Answers to Quick Quizzes">
+Answers to Quick Quizzes</a></h3>
+
+<a name="qq1answer"></a>
+<p><b>Quick Quiz 1</b>:
+Wait a minute!
+You said that updaters can make useful forward progress concurrently
+with readers, but pre-existing readers will block
+<tt>synchronize_rcu()</tt>!!!
+Just who are you trying to fool???
+
+
+</p><p><b>Answer</b>:
+First, if updaters do not wish to be blocked by readers, they can use
+<tt>call_rcu()</tt> or <tt>kfree_rcu()</tt>, which will
+be discussed later.
+Second, even when using <tt>synchronize_rcu()</tt>, the other
+update-side code does run concurrently with readers, whether pre-existing
+or not.
+
+
+</p><p><a href="#Quick%20Quiz%201"><b>Back to Quick Quiz 1</b>.</a>
+
+<a name="qq2answer"></a>
+<p><b>Quick Quiz 2</b>:
+Why is the <tt>synchronize_rcu()</tt> on line&nbsp;28 needed?
+
+
+</p><p><b>Answer</b>:
+Without that extra grace period, memory reordering could result in
+<tt>do_something_dlm()</tt> executing <tt>do_something()</tt>
+concurrently with the last bits of <tt>recovery()</tt>.
+
+
+</p><p><a href="#Quick%20Quiz%202"><b>Back to Quick Quiz 2</b>.</a>
+
+<a name="qq3answer"></a>
+<p><b>Quick Quiz 3</b>:
+But <tt>rcu_assign_pointer()</tt> does nothing to prevent the
+two assignments to <tt>p-&gt;a</tt> and <tt>p-&gt;b</tt>
+from being reordered.
+Can't that also cause problems?
+
+
+</p><p><b>Answer</b>:
+No, it cannot.
+The readers cannot see either of these two fields until
+the assignment to <tt>gp</tt>, by which time both fields are
+fully initialized.
+So reordering the assignments
+to <tt>p-&gt;a</tt> and <tt>p-&gt;b</tt> cannot possibly
+cause any problems.
+
+
+</p><p><a href="#Quick%20Quiz%203"><b>Back to Quick Quiz 3</b>.</a>
+
+<a name="qq4answer"></a>
+<p><b>Quick Quiz 4</b>:
+Without the <tt>rcu_dereference()</tt> or the
+<tt>rcu_access_pointer()</tt>, what destructive optimizations
+might the compiler make use of?
+
+
+</p><p><b>Answer</b>:
+Let's start with what happens to <tt>do_something_gp()</tt>
+if it fails to use <tt>rcu_dereference()</tt>.
+It could reuse a value formerly fetched from this same pointer.
+It could also fetch the pointer from <tt>gp</tt> in a byte-at-a-time
+manner, resulting in <i>load tearing</i>, in turn resulting a bytewise
+mash-up of two distince pointer values.
+It might even use value-speculation optimizations, where it makes a wrong
+guess, but by the time it gets around to checking the value, an update
+has changed the pointer to match the wrong guess.
+Too bad about any dereferences that returned pre-initialization garbage
+in the meantime!
+
+<p>
+For <tt>remove_gp_synchronous()</tt>, as long as all modifications
+to <tt>gp</tt> are carried out while holding <tt>gp_lock</tt>,
+the above optimizations are harmless.
+However,
+with <tt>CONFIG_SPARSE_RCU_POINTER=y</tt>,
+<tt>sparse</tt> will complain if you
+define <tt>gp</tt> with <tt>__rcu</tt> and then
+access it without using
+either <tt>rcu_access_pointer()</tt> or <tt>rcu_dereference()</tt>.
+
+
+</p><p><a href="#Quick%20Quiz%204"><b>Back to Quick Quiz 4</b>.</a>
+
+<a name="qq5answer"></a>
+<p><b>Quick Quiz 5</b>:
+Given that multiple CPUs can start RCU read-side critical sections
+at any time without any ordering whatsoever, how can RCU possibly tell whether
+or not a given RCU read-side critical section starts before a
+given instance of <tt>synchronize_rcu()</tt>?
+
+
+</p><p><b>Answer</b>:
+If RCU cannot tell whether or not a given
+RCU read-side critical section starts before a
+given instance of <tt>synchronize_rcu()</tt>,
+then it must assume that the RCU read-side critical section
+started first.
+In other words, a given instance of <tt>synchronize_rcu()</tt>
+can avoid waiting on a given RCU read-side critical section only
+if it can prove that <tt>synchronize_rcu()</tt> started first.
+
+
+</p><p><a href="#Quick%20Quiz%205"><b>Back to Quick Quiz 5</b>.</a>
+
+<a name="qq6answer"></a>
+<p><b>Quick Quiz 6</b>:
+The first and second guarantees require unbelievably strict ordering!
+Are all these memory barriers <i> really</i> required?
+
+
+</p><p><b>Answer</b>:
+Yes, they really are required.
+To see why the first guarantee is required, consider the following
+sequence of events:
+
+<ol>
+<li>	CPU 1: <tt>rcu_read_lock()</tt>
+<li>	CPU 1: <tt>q = rcu_dereference(gp);
+	/* Very likely to return p. */</tt>
+<li>	CPU 0: <tt>list_del_rcu(p);</tt>
+<li>	CPU 0: <tt>synchronize_rcu()</tt> starts.
+<li>	CPU 1: <tt>do_something_with(q-&gt;a);
+	/* No smp_mb(), so might happen after kfree(). */</tt>
+<li>	CPU 1: <tt>rcu_read_unlock()</tt>
+<li>	CPU 0: <tt>synchronize_rcu()</tt> returns.
+<li>	CPU 0: <tt>kfree(p);</tt>
+</ol>
+
+<p>
+Therefore, there absolutely must be a full memory barrier between the
+end of the RCU read-side critical section and the end of the
+grace period.
+
+<p>
+The sequence of events demonstrating the necessity of the second rule
+is roughly similar:
+
+<ol>
+<li>	CPU 0: <tt>list_del_rcu(p);</tt>
+<li>	CPU 0: <tt>synchronize_rcu()</tt> starts.
+<li>	CPU 1: <tt>rcu_read_lock()</tt>
+<li>	CPU 1: <tt>q = rcu_dereference(gp);
+	/* Might return p if no memory barrier. */</tt>
+<li>	CPU 0: <tt>synchronize_rcu()</tt> returns.
+<li>	CPU 0: <tt>kfree(p);</tt>
+<li>	CPU 1: <tt>do_something_with(q-&gt;a); /* Boom!!! */</tt>
+<li>	CPU 1: <tt>rcu_read_unlock()</tt>
+</ol>
+
+<p>
+And similarly, without a memory barrier between the beginning of the
+grace period and the beginning of the RCU read-side critical section,
+CPU&nbsp;1 might end up accessing the freelist.
+
+<p>
+The &ldquo;as if&rdquo; rule of course applies, so that any implementation
+that acts as if the appropriate memory barriers were in place is a
+correct implementation.
+That said, it is much easier to fool yourself into believing that you have
+adhered to the as-if rule than it is to actually adhere to it!
+
+
+</p><p><a href="#Quick%20Quiz%206"><b>Back to Quick Quiz 6</b>.</a>
+
+<a name="qq7answer"></a>
+<p><b>Quick Quiz 7</b>:
+But how does the upgrade-to-write operation exclude other readers?
+
+
+</p><p><b>Answer</b>:
+It doesn't, just like normal RCU updates, which also do not exclude
+RCU readers.
+
+
+</p><p><a href="#Quick%20Quiz%207"><b>Back to Quick Quiz 7</b>.</a>
+
+<a name="qq8answer"></a>
+<p><b>Quick Quiz 8</b>:
+Can't the compiler also reorder this code?
+
+
+</p><p><b>Answer</b>:
+No, the volatile casts in <tt>READ_ONCE()</tt> and
+<tt>WRITE_ONCE()</tt> prevent the compiler from reordering in
+this particular case.
+
+
+</p><p><a href="#Quick%20Quiz%208"><b>Back to Quick Quiz 8</b>.</a>
+
+<a name="qq9answer"></a>
+<p><b>Quick Quiz 9</b>:
+Suppose that synchronize_rcu() did wait until all readers had completed.
+Would the updater be able to rely on this?
+
+
+</p><p><b>Answer</b>:
+No.
+Even if <tt>synchronize_rcu()</tt> were to wait until
+all readers had completed, a new reader might start immediately after
+<tt>synchronize_rcu()</tt> completed.
+Therefore, the code following
+<tt>synchronize_rcu()</tt> cannot rely on there being no readers
+in any case.
+
+
+</p><p><a href="#Quick%20Quiz%209"><b>Back to Quick Quiz 9</b>.</a>
+
+<a name="qq10answer"></a>
+<p><b>Quick Quiz 10</b>:
+How long a sequence of grace periods, each separated by an RCU read-side
+critical section, would be required to partition the RCU read-side
+critical sections at the beginning and end of the chain?
+
+
+</p><p><b>Answer</b>:
+In theory, an infinite number.
+In practice, an unknown number that is sensitive to both implementation
+details and timing considerations.
+Therefore, even in practice, RCU users must abide by the theoretical rather
+than the practical answer.
+
+
+</p><p><a href="#Quick%20Quiz%2010"><b>Back to Quick Quiz 10</b>.</a>
+
+<a name="qq11answer"></a>
+<p><b>Quick Quiz 11</b>:
+What about sleeping locks?
+
+
+</p><p><b>Answer</b>:
+These are forbidden within Linux-kernel RCU read-side critical sections
+because it is not legal to place a quiescent state (in this case,
+voluntary context switch) within an RCU read-side critical section.
+However, sleeping locks may be used within userspace RCU read-side critical
+sections, and also within Linux-kernel sleepable RCU
+<a href="#Sleepable RCU">(SRCU)</a>
+read-side critical sections.
+In addition, the -rt patchset turns spinlocks into a sleeping locks so
+that the corresponding critical sections can be preempted, which
+also means that these sleeplockified spinlocks (but not other sleeping locks!)
+may be acquire within -rt-Linux-kernel RCU read-side critical sections.
+
+<p>
+Note that it <i>is</i> legal for a normal RCU read-side critical section
+to conditionally acquire a sleeping locks (as in <tt>mutex_trylock()</tt>),
+but only as long as it does not loop indefinitely attempting to
+conditionally acquire that sleeping locks.
+The key point is that things like <tt>mutex_trylock()</tt>
+either return with the mutex held, or return an error indication if
+the mutex was not immediately available.
+Either way, <tt>mutex_trylock()</tt> returns immediately without sleeping.
+
+
+</p><p><a href="#Quick%20Quiz%2011"><b>Back to Quick Quiz 11</b>.</a>
+
+<a name="qq12answer"></a>
+<p><b>Quick Quiz 12</b>:
+Why does line&nbsp;19 use <tt>rcu_access_pointer()</tt>?
+After all, <tt>call_rcu()</tt> on line&nbsp;25 stores into the
+structure, which would interact badly with concurrent insertions.
+Doesn't this mean that <tt>rcu_dereference()</tt> is required?
+
+
+</p><p><b>Answer</b>:
+Presumably the <tt>-&gt;gp_lock</tt> acquired on line&nbsp;18 excludes
+any changes, including any insertions that <tt>rcu_dereference()</tt>
+would protect against.
+Therefore, any insertions will be delayed until after <tt>-&gt;gp_lock</tt>
+is released on line&nbsp;25, which in turn means that
+<tt>rcu_access_pointer()</tt> suffices.
+
+
+</p><p><a href="#Quick%20Quiz%2012"><b>Back to Quick Quiz 12</b>.</a>
+
+<a name="qq13answer"></a>
+<p><b>Quick Quiz 13</b>:
+Earlier it was claimed that <tt>call_rcu()</tt> and
+<tt>kfree_rcu()</tt> allowed updaters to avoid being blocked
+by readers.
+But how can that be correct, given that the invocation of the callback
+and the freeing of the memory (respectively) must still wait for
+a grace period to elapse?
+
+
+</p><p><b>Answer</b>:
+We could define things this way, but keep in mind that this sort of
+definition would say that updates in garbage-collected languages
+cannot complete until the next time the garbage collector runs,
+which does not seem at all reasonable.
+The key point is that in most cases, an updater using either
+<tt>call_rcu()</tt> or <tt>kfree_rcu()</tt> can proceed to the
+next update as soon as it has invoked <tt>call_rcu()</tt> or
+<tt>kfree_rcu()</tt>, without having to wait for a subsequent
+grace period.
+
+
+</p><p><a href="#Quick%20Quiz%2013"><b>Back to Quick Quiz 13</b>.</a>
+
+<a name="qq14answer"></a>
+<p><b>Quick Quiz 14</b>:
+So what happens with <tt>synchronize_rcu()</tt> during
+scheduler initialization for <tt>CONFIG_PREEMPT=n</tt>
+kernels?
+
+
+</p><p><b>Answer</b>:
+In <tt>CONFIG_PREEMPT=n</tt> kernel, <tt>synchronize_rcu()</tt>
+maps directly to <tt>synchronize_sched()</tt>.
+Therefore, <tt>synchronize_rcu()</tt> works normally throughout
+boot in <tt>CONFIG_PREEMPT=n</tt> kernels.
+However, your code must also work in <tt>CONFIG_PREEMPT=y</tt> kernels,
+so it is still necessary to avoid invoking <tt>synchronize_rcu()</tt>
+during scheduler initialization.
+
+
+</p><p><a href="#Quick%20Quiz%2014"><b>Back to Quick Quiz 14</b>.</a>
+
+
+</body></html>
diff --git a/Documentation/RCU/Design/Requirements/Requirements.htmlx b/Documentation/RCU/Design/Requirements/Requirements.htmlx
new file mode 100644
index 000000000000..c5bb5bddd7ee
--- /dev/null
+++ b/Documentation/RCU/Design/Requirements/Requirements.htmlx
@@ -0,0 +1,2643 @@
+<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN"
+        "http://www.w3.org/TR/html4/loose.dtd">
+        <html>
+        <head><title>A Tour Through RCU's Requirements [LWN.net]</title>
+        <meta HTTP-EQUIV="Content-Type" CONTENT="text/html; charset=iso-8859-1">
+
+<h1>A Tour Through RCU's Requirements</h1>
+
+<p>Copyright IBM Corporation, 2015</p>
+<p>Author: Paul E.&nbsp;McKenney</p>
+<p><i>The initial version of this document appeared in the
+<a href="http://lwn.net/">LWN</a> articles
+<a href="http://lwn.net/Articles/652156/">here</a>,
+<a href="http://lwn.net/Articles/652677/">here</a>, and
+<a href="http://lwn.net/Articles/653326/">here</a>.</i></p>
+
+<h2>Introduction</h2>
+
+<p>
+Read-copy update (RCU) is a synchronization mechanism that is often
+used as a replacement for reader-writer locking.
+RCU is unusual in that updaters do not block readers,
+which means that RCU's read-side primitives can be exceedingly fast
+and scalable.
+In addition, updaters can make useful forward progress concurrently
+with readers.
+However, all this concurrency between RCU readers and updaters does raise
+the question of exactly what RCU readers are doing, which in turn
+raises the question of exactly what RCU's requirements are.
+
+<p>
+This document therefore summarizes RCU's requirements, and can be thought
+of as an informal, high-level specification for RCU.
+It is important to understand that RCU's specification is primarily
+empirical in nature;
+in fact, I learned about many of these requirements the hard way.
+This situation might cause some consternation, however, not only
+has this learning process been a lot of fun, but it has also been
+a great privilege to work with so many people willing to apply
+technologies in interesting new ways.
+
+<p>
+All that aside, here are the categories of currently known RCU requirements:
+</p>
+
+<ol>
+<li>	<a href="#Fundamental Requirements">
+	Fundamental Requirements</a>
+<li>	<a href="#Fundamental Non-Requirements">Fundamental Non-Requirements</a>
+<li>	<a href="#Parallelism Facts of Life">
+	Parallelism Facts of Life</a>
+<li>	<a href="#Quality-of-Implementation Requirements">
+	Quality-of-Implementation Requirements</a>
+<li>	<a href="#Linux Kernel Complications">
+	Linux Kernel Complications</a>
+<li>	<a href="#Software-Engineering Requirements">
+	Software-Engineering Requirements</a>
+<li>	<a href="#Other RCU Flavors">
+	Other RCU Flavors</a>
+<li>	<a href="#Possible Future Changes">
+	Possible Future Changes</a>
+</ol>
+
+<p>
+This is followed by a <a href="#Summary">summary</a>,
+which is in turn followed by the inevitable
+<a href="#Answers to Quick Quizzes">answers to the quick quizzes</a>.
+
+<h2><a name="Fundamental Requirements">Fundamental Requirements</a></h2>
+
+<p>
+RCU's fundamental requirements are the closest thing RCU has to hard
+mathematical requirements.
+These are:
+
+<ol>
+<li>	<a href="#Grace-Period Guarantee">
+	Grace-Period Guarantee</a>
+<li>	<a href="#Publish-Subscribe Guarantee">
+	Publish-Subscribe Guarantee</a>
+<li>	<a href="#RCU Primitives Guaranteed to Execute Unconditionally">
+	RCU Primitives Guaranteed to Execute Unconditionally</a>
+<li>	<a href="#Guaranteed Read-to-Write Upgrade">
+	Guaranteed Read-to-Write Upgrade</a>
+</ol>
+
+<h3><a name="Grace-Period Guarantee">Grace-Period Guarantee</a></h3>
+
+<p>
+RCU's grace-period guarantee is unusual in being premeditated:
+Jack Slingwine and I had this guarantee firmly in mind when we started
+work on RCU (then called &ldquo;rclock&rdquo;) in the early 1990s.
+That said, the past two decades of experience with RCU have produced
+a much more detailed understanding of this guarantee.
+
+<p>
+RCU's grace-period guarantee allows updaters to wait for the completion
+of all pre-existing RCU read-side critical sections.
+An RCU read-side critical section
+begins with the marker <tt>rcu_read_lock()</tt> and ends with
+the marker <tt>rcu_read_unlock()</tt>.
+These markers may be nested, and RCU treats a nested set as one
+big RCU read-side critical section.
+Production-quality implementations of <tt>rcu_read_lock()</tt> and
+<tt>rcu_read_unlock()</tt> are extremely lightweight, and in
+fact have exactly zero overhead in Linux kernels built for production
+use with <tt>CONFIG_PREEMPT=n</tt>.
+
+<p>
+This guarantee allows ordering to be enforced with extremely low
+overhead to readers, for example:
+
+<blockquote>
+<pre>
+ 1 int x, y;
+ 2
+ 3 void thread0(void)
+ 4 {
+ 5   rcu_read_lock();
+ 6   r1 = READ_ONCE(x);
+ 7   r2 = READ_ONCE(y);
+ 8   rcu_read_unlock();
+ 9 }
+10
+11 void thread1(void)
+12 {
+13   WRITE_ONCE(x, 1);
+14   synchronize_rcu();
+15   WRITE_ONCE(y, 1);
+16 }
+</pre>
+</blockquote>
+
+<p>
+Because the <tt>synchronize_rcu()</tt> on line&nbsp;14 waits for
+all pre-existing readers, any instance of <tt>thread0()</tt> that
+loads a value of zero from <tt>x</tt> must complete before
+<tt>thread1()</tt> stores to <tt>y</tt>, so that instance must
+also load a value of zero from <tt>y</tt>.
+Similarly, any instance of <tt>thread0()</tt> that loads a value of
+one from <tt>y</tt> must have started after the
+<tt>synchronize_rcu()</tt> started, and must therefore also load
+a value of one from <tt>x</tt>.
+Therefore, the outcome:
+<blockquote>
+<pre>
+(r1 == 0 &amp;&amp; r2 == 1)
+</pre>
+</blockquote>
+cannot happen.
+
+<p>@@QQ@@
+Wait a minute!
+You said that updaters can make useful forward progress concurrently
+with readers, but pre-existing readers will block
+<tt>synchronize_rcu()</tt>!!!
+Just who are you trying to fool???
+<p>@@QQA@@
+First, if updaters do not wish to be blocked by readers, they can use
+<tt>call_rcu()</tt> or <tt>kfree_rcu()</tt>, which will
+be discussed later.
+Second, even when using <tt>synchronize_rcu()</tt>, the other
+update-side code does run concurrently with readers, whether pre-existing
+or not.
+<p>@@QQE@@
+
+<p>
+This scenario resembles one of the first uses of RCU in
+<a href="http://en.wikipedia.org/wiki/DYNIX">DYNIX/ptx</a>,
+which managed a distributed lock manager's transition into
+a state suitable for handling recovery from node failure,
+more or less as follows:
+
+<blockquote>
+<pre>
+ 1 #define STATE_NORMAL        0
+ 2 #define STATE_WANT_RECOVERY 1
+ 3 #define STATE_RECOVERING    2
+ 4 #define STATE_WANT_NORMAL   3
+ 5
+ 6 int state = STATE_NORMAL;
+ 7
+ 8 void do_something_dlm(void)
+ 9 {
+10   int state_snap;
+11
+12   rcu_read_lock();
+13   state_snap = READ_ONCE(state);
+14   if (state_snap == STATE_NORMAL)
+15     do_something();
+16   else
+17     do_something_carefully();
+18   rcu_read_unlock();
+19 }
+20
+21 void start_recovery(void)
+22 {
+23   WRITE_ONCE(state, STATE_WANT_RECOVERY);
+24   synchronize_rcu();
+25   WRITE_ONCE(state, STATE_RECOVERING);
+26   recovery();
+27   WRITE_ONCE(state, STATE_WANT_NORMAL);
+28   synchronize_rcu();
+29   WRITE_ONCE(state, STATE_NORMAL);
+30 }
+</pre>
+</blockquote>
+
+<p>
+The RCU read-side critical section in <tt>do_something_dlm()</tt>
+works with the <tt>synchronize_rcu()</tt> in <tt>start_recovery()</tt>
+to guarantee that <tt>do_something()</tt> never runs concurrently
+with <tt>recovery()</tt>, but with little or no synchronization
+overhead in <tt>do_something_dlm()</tt>.
+
+<p>@@QQ@@
+Why is the <tt>synchronize_rcu()</tt> on line&nbsp;28 needed?
+<p>@@QQA@@
+Without that extra grace period, memory reordering could result in
+<tt>do_something_dlm()</tt> executing <tt>do_something()</tt>
+concurrently with the last bits of <tt>recovery()</tt>.
+<p>@@QQE@@
+
+<p>
+In order to avoid fatal problems such as deadlocks,
+an RCU read-side critical section must not contain calls to
+<tt>synchronize_rcu()</tt>.
+Similarly, an RCU read-side critical section must not
+contain anything that waits, directly or indirectly, on completion of
+an invocation of <tt>synchronize_rcu()</tt>.
+
+<p>
+Although RCU's grace-period guarantee is useful in and of itself, with
+<a href="https://lwn.net/Articles/573497/">quite a few use cases</a>,
+it would be good to be able to use RCU to coordinate read-side
+access to linked data structures.
+For this, the grace-period guarantee is not sufficient, as can
+be seen in function <tt>add_gp_buggy()</tt> below.
+We will look at the reader's code later, but in the meantime, just think of
+the reader as locklessly picking up the <tt>gp</tt> pointer,
+and, if the value loaded is non-<tt>NULL</tt>, locklessly accessing the
+<tt>-&gt;a</tt> and <tt>-&gt;b</tt> fields.
+
+<blockquote>
+<pre>
+ 1 bool add_gp_buggy(int a, int b)
+ 2 {
+ 3   p = kmalloc(sizeof(*p), GFP_KERNEL);
+ 4   if (!p)
+ 5     return -ENOMEM;
+ 6   spin_lock(&amp;gp_lock);
+ 7   if (rcu_access_pointer(gp)) {
+ 8     spin_unlock(&amp;gp_lock);
+ 9     return false;
+10   }
+11   p-&gt;a = a;
+12   p-&gt;b = a;
+13   gp = p; /* ORDERING BUG */
+14   spin_unlock(&amp;gp_lock);
+15   return true;
+16 }
+</pre>
+</blockquote>
+
+<p>
+The problem is that both the compiler and weakly ordered CPUs are within
+their rights to reorder this code as follows:
+
+<blockquote>
+<pre>
+ 1 bool add_gp_buggy_optimized(int a, int b)
+ 2 {
+ 3   p = kmalloc(sizeof(*p), GFP_KERNEL);
+ 4   if (!p)
+ 5     return -ENOMEM;
+ 6   spin_lock(&amp;gp_lock);
+ 7   if (rcu_access_pointer(gp)) {
+ 8     spin_unlock(&amp;gp_lock);
+ 9     return false;
+10   }
+<b>11   gp = p; /* ORDERING BUG */
+12   p-&gt;a = a;
+13   p-&gt;b = a;</b>
+14   spin_unlock(&amp;gp_lock);
+15   return true;
+16 }
+</pre>
+</blockquote>
+
+<p>
+If an RCU reader fetches <tt>gp</tt> just after
+<tt>add_gp_buggy_optimized</tt> executes line&nbsp;11,
+it will see garbage in the <tt>-&gt;a</tt> and <tt>-&gt;b</tt>
+fields.
+And this is but one of many ways in which compiler and hardware optimizations
+could cause trouble.
+Therefore, we clearly need some way to prevent the compiler and the CPU from
+reordering in this manner, which brings us to the publish-subscribe
+guarantee discussed in the next section.
+
+<h3><a name="Publish-Subscribe Guarantee">Publish/Subscribe Guarantee</a></h3>
+
+<p>
+RCU's publish-subscribe guarantee allows data to be inserted
+into a linked data structure without disrupting RCU readers.
+The updater uses <tt>rcu_assign_pointer()</tt> to insert the
+new data, and readers use <tt>rcu_dereference()</tt> to
+access data, whether new or old.
+The following shows an example of insertion:
+
+<blockquote>
+<pre>
+ 1 bool add_gp(int a, int b)
+ 2 {
+ 3   p = kmalloc(sizeof(*p), GFP_KERNEL);
+ 4   if (!p)
+ 5     return -ENOMEM;
+ 6   spin_lock(&amp;gp_lock);
+ 7   if (rcu_access_pointer(gp)) {
+ 8     spin_unlock(&amp;gp_lock);
+ 9     return false;
+10   }
+11   p-&gt;a = a;
+12   p-&gt;b = a;
+13   rcu_assign_pointer(gp, p);
+14   spin_unlock(&amp;gp_lock);
+15   return true;
+16 }
+</pre>
+</blockquote>
+
+<p>
+The <tt>rcu_assign_pointer()</tt> on line&nbsp;13 is conceptually
+equivalent to a simple assignment statement, but also guarantees
+that its assignment will
+happen after the two assignments in lines&nbsp;11 and&nbsp;12,
+similar to the C11 <tt>memory_order_release</tt> store operation.
+It also prevents any number of &ldquo;interesting&rdquo; compiler
+optimizations, for example, the use of <tt>gp</tt> as a scratch
+location immediately preceding the assignment.
+
+<p>@@QQ@@
+But <tt>rcu_assign_pointer()</tt> does nothing to prevent the
+two assignments to <tt>p-&gt;a</tt> and <tt>p-&gt;b</tt>
+from being reordered.
+Can't that also cause problems?
+<p>@@QQA@@
+No, it cannot.
+The readers cannot see either of these two fields until
+the assignment to <tt>gp</tt>, by which time both fields are
+fully initialized.
+So reordering the assignments
+to <tt>p-&gt;a</tt> and <tt>p-&gt;b</tt> cannot possibly
+cause any problems.
+<p>@@QQE@@
+
+<p>
+It is tempting to assume that the reader need not do anything special
+to control its accesses to the RCU-protected data,
+as shown in <tt>do_something_gp_buggy()</tt> below:
+
+<blockquote>
+<pre>
+ 1 bool do_something_gp_buggy(void)
+ 2 {
+ 3   rcu_read_lock();
+ 4   p = gp;  /* OPTIMIZATIONS GALORE!!! */
+ 5   if (p) {
+ 6     do_something(p-&gt;a, p-&gt;b);
+ 7     rcu_read_unlock();
+ 8     return true;
+ 9   }
+10   rcu_read_unlock();
+11   return false;
+12 }
+</pre>
+</blockquote>
+
+<p>
+However, this temptation must be resisted because there are a
+surprisingly large number of ways that the compiler
+(to say nothing of
+<a href="http://www.openvms.compaq.com/wizard/wiz_2637.html">DEC Alpha CPUs</a>)
+can trip this code up.
+For but one example, if the compiler were short of registers, it
+might choose to refetch from <tt>gp</tt> rather than keeping
+a separate copy in <tt>p</tt> as follows:
+
+<blockquote>
+<pre>
+ 1 bool do_something_gp_buggy_optimized(void)
+ 2 {
+ 3   rcu_read_lock();
+ 4   if (gp) { /* OPTIMIZATIONS GALORE!!! */
+<b> 5     do_something(gp-&gt;a, gp-&gt;b);</b>
+ 6     rcu_read_unlock();
+ 7     return true;
+ 8   }
+ 9   rcu_read_unlock();
+10   return false;
+11 }
+</pre>
+</blockquote>
+
+<p>
+If this function ran concurrently with a series of updates that
+replaced the current structure with a new one,
+the fetches of <tt>gp-&gt;a</tt>
+and <tt>gp-&gt;b</tt> might well come from two different structures,
+which could cause serious confusion.
+To prevent this (and much else besides), <tt>do_something_gp()</tt> uses
+<tt>rcu_dereference()</tt> to fetch from <tt>gp</tt>:
+
+<blockquote>
+<pre>
+ 1 bool do_something_gp(void)
+ 2 {
+ 3   rcu_read_lock();
+ 4   p = rcu_dereference(gp);
+ 5   if (p) {
+ 6     do_something(p-&gt;a, p-&gt;b);
+ 7     rcu_read_unlock();
+ 8     return true;
+ 9   }
+10   rcu_read_unlock();
+11   return false;
+12 }
+</pre>
+</blockquote>
+
+<p>
+The <tt>rcu_dereference()</tt> uses volatile casts and (for DEC Alpha)
+memory barriers in the Linux kernel.
+Should a
+<a href="http://www.rdrop.com/users/paulmck/RCU/consume.2015.07.13a.pdf">high-quality implementation of C11 <tt>memory_order_consume</tt> [PDF]</a>
+ever appear, then <tt>rcu_dereference()</tt> could be implemented
+as a <tt>memory_order_consume</tt> load.
+Regardless of the exact implementation, a pointer fetched by
+<tt>rcu_dereference()</tt> may not be used outside of the
+outermost RCU read-side critical section containing that
+<tt>rcu_dereference()</tt>, unless protection of
+the corresponding data element has been passed from RCU to some
+other synchronization mechanism, most commonly locking or
+<a href="https://www.kernel.org/doc/Documentation/RCU/rcuref.txt">reference counting</a>.
+
+<p>
+In short, updaters use <tt>rcu_assign_pointer()</tt> and readers
+use <tt>rcu_dereference()</tt>, and these two RCU API elements
+work together to ensure that readers have a consistent view of
+newly added data elements.
+
+<p>
+Of course, it is also necessary to remove elements from RCU-protected
+data structures, for example, using the following process:
+
+<ol>
+<li>	Remove the data element from the enclosing structure.
+<li>	Wait for all pre-existing RCU read-side critical sections
+	to complete (because only pre-existing readers can possibly have
+	a reference to the newly removed data element).
+<li>	At this point, only the updater has a reference to the
+	newly removed data element, so it can safely reclaim
+	the data element, for example, by passing it to <tt>kfree()</tt>.
+</ol>
+
+This process is implemented by <tt>remove_gp_synchronous()</tt>:
+
+<blockquote>
+<pre>
+ 1 bool remove_gp_synchronous(void)
+ 2 {
+ 3   struct foo *p;
+ 4
+ 5   spin_lock(&amp;gp_lock);
+ 6   p = rcu_access_pointer(gp);
+ 7   if (!p) {
+ 8     spin_unlock(&amp;gp_lock);
+ 9     return false;
+10   }
+11   rcu_assign_pointer(gp, NULL);
+12   spin_unlock(&amp;gp_lock);
+13   synchronize_rcu();
+14   kfree(p);
+15   return true;
+16 }
+</pre>
+</blockquote>
+
+<p>
+This function is straightforward, with line&nbsp;13 waiting for a grace
+period before line&nbsp;14 frees the old data element.
+This waiting ensures that readers will reach line&nbsp;7 of
+<tt>do_something_gp()</tt> before the data element referenced by
+<tt>p</tt> is freed.
+The <tt>rcu_access_pointer()</tt> on line&nbsp;6 is similar to
+<tt>rcu_dereference()</tt>, except that:
+
+<ol>
+<li>	The value returned by <tt>rcu_access_pointer()</tt>
+	cannot be dereferenced.
+	If you want to access the value pointed to as well as
+	the pointer itself, use <tt>rcu_dereference()</tt>
+	instead of <tt>rcu_access_pointer()</tt>.
+<li>	The call to <tt>rcu_access_pointer()</tt> need not be
+	protected.
+	In contrast, <tt>rcu_dereference()</tt> must either be
+	within an RCU read-side critical section or in a code
+	segment where the pointer cannot change, for example, in
+	code protected by the corresponding update-side lock.
+</ol>
+
+<p>@@QQ@@
+Without the <tt>rcu_dereference()</tt> or the
+<tt>rcu_access_pointer()</tt>, what destructive optimizations
+might the compiler make use of?
+<p>@@QQA@@
+Let's start with what happens to <tt>do_something_gp()</tt>
+if it fails to use <tt>rcu_dereference()</tt>.
+It could reuse a value formerly fetched from this same pointer.
+It could also fetch the pointer from <tt>gp</tt> in a byte-at-a-time
+manner, resulting in <i>load tearing</i>, in turn resulting a bytewise
+mash-up of two distince pointer values.
+It might even use value-speculation optimizations, where it makes a wrong
+guess, but by the time it gets around to checking the value, an update
+has changed the pointer to match the wrong guess.
+Too bad about any dereferences that returned pre-initialization garbage
+in the meantime!
+
+<p>
+For <tt>remove_gp_synchronous()</tt>, as long as all modifications
+to <tt>gp</tt> are carried out while holding <tt>gp_lock</tt>,
+the above optimizations are harmless.
+However,
+with <tt>CONFIG_SPARSE_RCU_POINTER=y</tt>,
+<tt>sparse</tt> will complain if you
+define <tt>gp</tt> with <tt>__rcu</tt> and then
+access it without using
+either <tt>rcu_access_pointer()</tt> or <tt>rcu_dereference()</tt>.
+<p>@@QQE@@
+
+<p>
+This simple linked-data-structure scenario clearly demonstrates the need
+for RCU's stringent memory-ordering guarantees on systems with more than
+one CPU:
+
+<ol>
+<li>	Each CPU that has an RCU read-side critical section that
+	begins before <tt>synchronize_rcu()</tt> starts is
+	guaranteed to execute a full memory barrier between the time
+	that the RCU read-side critical section ends and the time that
+	<tt>synchronize_rcu()</tt> returns.
+	Without this guarantee, a pre-existing RCU read-side critical section
+	might hold a reference to the newly removed <tt>struct foo</tt>
+	after the <tt>kfree()</tt> on line&nbsp;14 of
+	<tt>remove_gp_synchronous()</tt>.
+<li>	Each CPU that has an RCU read-side critical section that ends
+	after <tt>synchronize_rcu()</tt> returns is guaranteed
+	to execute a full memory barrier between the time that
+	<tt>synchronize_rcu()</tt> begins and the time that the RCU
+	read-side critical section begins.
+	Without this guarantee, a later RCU read-side critical section
+	running after the <tt>kfree()</tt> on line&nbsp;14 of
+	<tt>remove_gp_synchronous()</tt> might
+	later run <tt>do_something_gp()</tt> and find the
+	newly deleted <tt>struct foo</tt>.
+<li>	If the task invoking <tt>synchronize_rcu()</tt> remains
+	on a given CPU, then that CPU is guaranteed to execute a full
+	memory barrier sometime during the execution of
+	<tt>synchronize_rcu()</tt>.
+	This guarantee ensures that the <tt>kfree()</tt> on
+	line&nbsp;14 of <tt>remove_gp_synchronous()</tt> really does
+	execute after the removal on line&nbsp;11.
+<li>	If the task invoking <tt>synchronize_rcu()</tt> migrates
+	among a group of CPUs during that invocation, then each of the
+	CPUs in that group is guaranteed to execute a full memory barrier
+	sometime during the execution of <tt>synchronize_rcu()</tt>.
+	This guarantee also ensures that the <tt>kfree()</tt> on
+	line&nbsp;14 of <tt>remove_gp_synchronous()</tt> really does
+	execute after the removal on
+	line&nbsp;11, but also in the case where the thread executing the
+	<tt>synchronize_rcu()</tt> migrates in the meantime.
+</ol>
+
+<p>@@QQ@@
+Given that multiple CPUs can start RCU read-side critical sections
+at any time without any ordering whatsoever, how can RCU possibly tell whether
+or not a given RCU read-side critical section starts before a
+given instance of <tt>synchronize_rcu()</tt>?
+<p>@@QQA@@
+If RCU cannot tell whether or not a given
+RCU read-side critical section starts before a
+given instance of <tt>synchronize_rcu()</tt>,
+then it must assume that the RCU read-side critical section
+started first.
+In other words, a given instance of <tt>synchronize_rcu()</tt>
+can avoid waiting on a given RCU read-side critical section only
+if it can prove that <tt>synchronize_rcu()</tt> started first.
+<p>@@QQE@@
+
+<p>@@QQ@@
+The first and second guarantees require unbelievably strict ordering!
+Are all these memory barriers <i> really</i> required?
+<p>@@QQA@@
+Yes, they really are required.
+To see why the first guarantee is required, consider the following
+sequence of events:
+
+<ol>
+<li>	CPU 1: <tt>rcu_read_lock()</tt>
+<li>	CPU 1: <tt>q = rcu_dereference(gp);
+	/* Very likely to return p. */</tt>
+<li>	CPU 0: <tt>list_del_rcu(p);</tt>
+<li>	CPU 0: <tt>synchronize_rcu()</tt> starts.
+<li>	CPU 1: <tt>do_something_with(q-&gt;a);
+	/* No smp_mb(), so might happen after kfree(). */</tt>
+<li>	CPU 1: <tt>rcu_read_unlock()</tt>
+<li>	CPU 0: <tt>synchronize_rcu()</tt> returns.
+<li>	CPU 0: <tt>kfree(p);</tt>
+</ol>
+
+<p>
+Therefore, there absolutely must be a full memory barrier between the
+end of the RCU read-side critical section and the end of the
+grace period.
+
+<p>
+The sequence of events demonstrating the necessity of the second rule
+is roughly similar:
+
+<ol>
+<li>	CPU 0: <tt>list_del_rcu(p);</tt>
+<li>	CPU 0: <tt>synchronize_rcu()</tt> starts.
+<li>	CPU 1: <tt>rcu_read_lock()</tt>
+<li>	CPU 1: <tt>q = rcu_dereference(gp);
+	/* Might return p if no memory barrier. */</tt>
+<li>	CPU 0: <tt>synchronize_rcu()</tt> returns.
+<li>	CPU 0: <tt>kfree(p);</tt>
+<li>	CPU 1: <tt>do_something_with(q-&gt;a); /* Boom!!! */</tt>
+<li>	CPU 1: <tt>rcu_read_unlock()</tt>
+</ol>
+
+<p>
+And similarly, without a memory barrier between the beginning of the
+grace period and the beginning of the RCU read-side critical section,
+CPU&nbsp;1 might end up accessing the freelist.
+
+<p>
+The &ldquo;as if&rdquo; rule of course applies, so that any implementation
+that acts as if the appropriate memory barriers were in place is a
+correct implementation.
+That said, it is much easier to fool yourself into believing that you have
+adhered to the as-if rule than it is to actually adhere to it!
+<p>@@QQE@@
+
+<p>
+In short, RCU's publish-subscribe guarantee is provided by the combination
+of <tt>rcu_assign_pointer()</tt> and <tt>rcu_dereference()</tt>.
+This guarantee allows data elements to be safely added to RCU-protected
+linked data structures without disrupting RCU readers.
+This guarantee can be used in combination with the grace-period
+guarantee to also allow data elements to be removed from RCU-protected
+linked data structures, again without disrupting RCU readers.
+
+<p>
+This guarantee was only partially premeditated.
+DYNIX/ptx used an explicit memory barrier for publication, but had nothing
+resembling <tt>rcu_dereference()</tt> for subscription, nor did it
+have anything resembling the <tt>smp_read_barrier_depends()</tt>
+that was later subsumed into <tt>rcu_dereference()</tt>.
+The need for these operations made itself known quite suddenly at a
+late-1990s meeting with the DEC Alpha architects, back in the days when
+DEC was still a free-standing company.
+It took the Alpha architects a good hour to convince me that any sort
+of barrier would ever be needed, and it then took me a good <i>two</i> hours
+to convince them that their documentation did not make this point clear.
+More recent work with the C and C++ standards committees have provided
+much education on tricks and traps from the compiler.
+In short, compilers were much less tricky in the early 1990s, but in
+2015, don't even think about omitting <tt>rcu_dereference()</tt>!
+
+<h3><a name="RCU Primitives Guaranteed to Execute Unconditionally">RCU Primitives Guaranteed to Execute Unconditionally</a></h3>
+
+<p>
+The common-case RCU primitives are unconditional.
+They are invoked, they do their job, and they return, with no possibility
+of error, and no need to retry.
+This is a key RCU design philosophy.
+
+<p>
+However, this philosophy is pragmatic rather than pigheaded.
+If someone comes up with a good justification for a particular conditional
+RCU primitive, it might well be implemented and added.
+After all, this guarantee was reverse-engineered, not premeditated.
+The unconditional nature of the RCU primitives was initially an
+accident of implementation, and later experience with synchronization
+primitives with conditional primitives caused me to elevate this
+accident to a guarantee.
+Therefore, the justification for adding a conditional primitive to
+RCU would need to be based on detailed and compelling use cases.
+
+<h3><a name="Guaranteed Read-to-Write Upgrade">Guaranteed Read-to-Write Upgrade</a></h3>
+
+<p>
+As far as RCU is concerned, it is always possible to carry out an
+update within an RCU read-side critical section.
+For example, that RCU read-side critical section might search for
+a given data element, and then might acquire the update-side
+spinlock in order to update that element, all while remaining
+in that RCU read-side critical section.
+Of course, it is necessary to exit the RCU read-side critical section
+before invoking <tt>synchronize_rcu()</tt>, however, this
+inconvenience can be avoided through use of the
+<tt>call_rcu()</tt> and <tt>kfree_rcu()</tt> API members
+described later in this document.
+
+<p>@@QQ@@
+But how does the upgrade-to-write operation exclude other readers?
+<p>@@QQA@@
+It doesn't, just like normal RCU updates, which also do not exclude
+RCU readers.
+<p>@@QQE@@
+
+<p>
+This guarantee allows lookup code to be shared between read-side
+and update-side code, and was premeditated, appearing in the earliest
+DYNIX/ptx RCU documentation.
+
+<h2><a name="Fundamental Non-Requirements">Fundamental Non-Requirements</a></h2>
+
+<p>
+RCU provides extremely lightweight readers, and its read-side guarantees,
+though quite useful, are correspondingly lightweight.
+It is therefore all too easy to assume that RCU is guaranteeing more
+than it really is.
+Of course, the list of things that RCU does not guarantee is infinitely
+long, however, the following sections list a few non-guarantees that
+have caused confusion.
+Except where otherwise noted, these non-guarantees were premeditated.
+
+<ol>
+<li>	<a href="#Readers Impose Minimal Ordering">
+	Readers Impose Minimal Ordering</a>
+<li>	<a href="#Readers Do Not Exclude Updaters">
+	Readers Do Not Exclude Updaters</a>
+<li>	<a href="#Updaters Only Wait For Old Readers">
+	Updaters Only Wait For Old Readers</a>
+<li>	<a href="#Grace Periods Don't Partition Read-Side Critical Sections">
+	Grace Periods Don't Partition Read-Side Critical Sections</a>
+<li>	<a href="#Read-Side Critical Sections Don't Partition Grace Periods">
+	Read-Side Critical Sections Don't Partition Grace Periods</a>
+<li>	<a href="#Disabling Preemption Does Not Block Grace Periods">
+	Disabling Preemption Does Not Block Grace Periods</a>
+</ol>
+
+<h3><a name="Readers Impose Minimal Ordering">Readers Impose Minimal Ordering</a></h3>
+
+<p>
+Reader-side markers such as <tt>rcu_read_lock()</tt> and
+<tt>rcu_read_unlock()</tt> provide absolutely no ordering guarantees
+except through their interaction with the grace-period APIs such as
+<tt>synchronize_rcu()</tt>.
+To see this, consider the following pair of threads:
+
+<blockquote>
+<pre>
+ 1 void thread0(void)
+ 2 {
+ 3   rcu_read_lock();
+ 4   WRITE_ONCE(x, 1);
+ 5   rcu_read_unlock();
+ 6   rcu_read_lock();
+ 7   WRITE_ONCE(y, 1);
+ 8   rcu_read_unlock();
+ 9 }
+10
+11 void thread1(void)
+12 {
+13   rcu_read_lock();
+14   r1 = READ_ONCE(y);
+15   rcu_read_unlock();
+16   rcu_read_lock();
+17   r2 = READ_ONCE(x);
+18   rcu_read_unlock();
+19 }
+</pre>
+</blockquote>
+
+<p>
+After <tt>thread0()</tt> and <tt>thread1()</tt> execute
+concurrently, it is quite possible to have
+
+<blockquote>
+<pre>
+(r1 == 1 &amp;&amp; r2 == 0)
+</pre>
+</blockquote>
+
+(that is, <tt>y</tt> appears to have been assigned before <tt>x</tt>),
+which would not be possible if <tt>rcu_read_lock()</tt> and
+<tt>rcu_read_unlock()</tt> had much in the way of ordering
+properties.
+But they do not, so the CPU is within its rights
+to do significant reordering.
+This is by design:  Any significant ordering constraints would slow down
+these fast-path APIs.
+
+<p>@@QQ@@
+Can't the compiler also reorder this code?
+<p>@@QQA@@
+No, the volatile casts in <tt>READ_ONCE()</tt> and
+<tt>WRITE_ONCE()</tt> prevent the compiler from reordering in
+this particular case.
+<p>@@QQE@@
+
+<h3><a name="Readers Do Not Exclude Updaters">Readers Do Not Exclude Updaters</a></h3>
+
+<p>
+Neither <tt>rcu_read_lock()</tt> nor <tt>rcu_read_unlock()</tt>
+exclude updates.
+All they do is to prevent grace periods from ending.
+The following example illustrates this:
+
+<blockquote>
+<pre>
+ 1 void thread0(void)
+ 2 {
+ 3   rcu_read_lock();
+ 4   r1 = READ_ONCE(y);
+ 5   if (r1) {
+ 6     do_something_with_nonzero_x();
+ 7     r2 = READ_ONCE(x);
+ 8     WARN_ON(!r2); /* BUG!!! */
+ 9   }
+10   rcu_read_unlock();
+11 }
+12
+13 void thread1(void)
+14 {
+15   spin_lock(&amp;my_lock);
+16   WRITE_ONCE(x, 1);
+17   WRITE_ONCE(y, 1);
+18   spin_unlock(&amp;my_lock);
+19 }
+</pre>
+</blockquote>
+
+<p>
+If the <tt>thread0()</tt> function's <tt>rcu_read_lock()</tt>
+excluded the <tt>thread1()</tt> function's update,
+the <tt>WARN_ON()</tt> could never fire.
+But the fact is that <tt>rcu_read_lock()</tt> does not exclude
+much of anything aside from subsequent grace periods, of which
+<tt>thread1()</tt> has none, so the
+<tt>WARN_ON()</tt> can and does fire.
+
+<h3><a name="Updaters Only Wait For Old Readers">Updaters Only Wait For Old Readers</a></h3>
+
+<p>
+It might be tempting to assume that after <tt>synchronize_rcu()</tt>
+completes, there are no readers executing.
+This temptation must be avoided because
+new readers can start immediately after <tt>synchronize_rcu()</tt>
+starts, and <tt>synchronize_rcu()</tt> is under no
+obligation to wait for these new readers.
+
+<p>@@QQ@@
+Suppose that synchronize_rcu() did wait until all readers had completed.
+Would the updater be able to rely on this?
+<p>@@QQA@@
+No.
+Even if <tt>synchronize_rcu()</tt> were to wait until
+all readers had completed, a new reader might start immediately after
+<tt>synchronize_rcu()</tt> completed.
+Therefore, the code following
+<tt>synchronize_rcu()</tt> cannot rely on there being no readers
+in any case.
+<p>@@QQE@@
+
+<h3><a name="Grace Periods Don't Partition Read-Side Critical Sections">
+Grace Periods Don't Partition Read-Side Critical Sections</a></h3>
+
+<p>
+It is tempting to assume that if any part of one RCU read-side critical
+section precedes a given grace period, and if any part of another RCU
+read-side critical section follows that same grace period, then all of
+the first RCU read-side critical section must precede all of the second.
+However, this just isn't the case: A single grace period does not
+partition the set of RCU read-side critical sections.
+An example of this situation can be illustrated as follows, where
+<tt>x</tt>, <tt>y</tt>, and <tt>z</tt> are initially all zero:
+
+<blockquote>
+<pre>
+ 1 void thread0(void)
+ 2 {
+ 3   rcu_read_lock();
+ 4   WRITE_ONCE(a, 1);
+ 5   WRITE_ONCE(b, 1);
+ 6   rcu_read_unlock();
+ 7 }
+ 8
+ 9 void thread1(void)
+10 {
+11   r1 = READ_ONCE(a);
+12   synchronize_rcu();
+13   WRITE_ONCE(c, 1);
+14 }
+15
+16 void thread2(void)
+17 {
+18   rcu_read_lock();
+19   r2 = READ_ONCE(b);
+20   r3 = READ_ONCE(c);
+21   rcu_read_unlock();
+22 }
+</pre>
+</blockquote>
+
+<p>
+It turns out that the outcome:
+
+<blockquote>
+<pre>
+(r1 == 1 &amp;&amp; r2 == 0 &amp;&amp; r3 == 1)
+</pre>
+</blockquote>
+
+is entirely possible.
+The following figure show how this can happen, with each circled
+<tt>QS</tt> indicating the point at which RCU recorded a
+<i>quiescent state</i> for each thread, that is, a state in which
+RCU knows that the thread cannot be in the midst of an RCU read-side
+critical section that started before the current grace period:
+
+<p><img src="GPpartitionReaders1.svg" alt="GPpartitionReaders1.svg" width="60%"></p>
+
+<p>
+If it is necessary to partition RCU read-side critical sections in this
+manner, it is necessary to use two grace periods, where the first
+grace period is known to end before the second grace period starts:
+
+<blockquote>
+<pre>
+ 1 void thread0(void)
+ 2 {
+ 3   rcu_read_lock();
+ 4   WRITE_ONCE(a, 1);
+ 5   WRITE_ONCE(b, 1);
+ 6   rcu_read_unlock();
+ 7 }
+ 8
+ 9 void thread1(void)
+10 {
+11   r1 = READ_ONCE(a);
+12   synchronize_rcu();
+13   WRITE_ONCE(c, 1);
+14 }
+15
+16 void thread2(void)
+17 {
+18   r2 = READ_ONCE(c);
+19   synchronize_rcu();
+20   WRITE_ONCE(d, 1);
+21 }
+22
+23 void thread3(void)
+24 {
+25   rcu_read_lock();
+26   r3 = READ_ONCE(b);
+27   r4 = READ_ONCE(d);
+28   rcu_read_unlock();
+29 }
+</pre>
+</blockquote>
+
+<p>
+Here, if <tt>(r1 == 1)</tt>, then
+<tt>thread0()</tt>'s write to <tt>b</tt> must happen
+before the end of <tt>thread1()</tt>'s grace period.
+If in addition <tt>(r4 == 1)</tt>, then
+<tt>thread3()</tt>'s read from <tt>b</tt> must happen
+after the beginning of <tt>thread2()</tt>'s grace period.
+If it is also the case that <tt>(r2 == 1)</tt>, then the
+end of <tt>thread1()</tt>'s grace period must precede the
+beginning of <tt>thread2()</tt>'s grace period.
+This mean that the two RCU read-side critical sections cannot overlap,
+guaranteeing that <tt>(r3 == 1)</tt>.
+As a result, the outcome:
+
+<blockquote>
+<pre>
+(r1 == 1 &amp;&amp; r2 == 1 &amp;&amp; r3 == 0 &amp;&amp; r4 == 1)
+</pre>
+</blockquote>
+
+cannot happen.
+
+<p>
+This non-requirement was also non-premeditated, but became apparent
+when studying RCU's interaction with memory ordering.
+
+<h3><a name="Read-Side Critical Sections Don't Partition Grace Periods">
+Read-Side Critical Sections Don't Partition Grace Periods</a></h3>
+
+<p>
+It is also tempting to assume that if an RCU read-side critical section
+happens between a pair of grace periods, then those grace periods cannot
+overlap.
+However, this temptation leads nowhere good, as can be illustrated by
+the following, with all variables initially zero:
+
+<blockquote>
+<pre>
+ 1 void thread0(void)
+ 2 {
+ 3   rcu_read_lock();
+ 4   WRITE_ONCE(a, 1);
+ 5   WRITE_ONCE(b, 1);
+ 6   rcu_read_unlock();
+ 7 }
+ 8
+ 9 void thread1(void)
+10 {
+11   r1 = READ_ONCE(a);
+12   synchronize_rcu();
+13   WRITE_ONCE(c, 1);
+14 }
+15
+16 void thread2(void)
+17 {
+18   rcu_read_lock();
+19   WRITE_ONCE(d, 1);
+20   r2 = READ_ONCE(c);
+21   rcu_read_unlock();
+22 }
+23
+24 void thread3(void)
+25 {
+26   r3 = READ_ONCE(d);
+27   synchronize_rcu();
+28   WRITE_ONCE(e, 1);
+29 }
+30
+31 void thread4(void)
+32 {
+33   rcu_read_lock();
+34   r4 = READ_ONCE(b);
+35   r5 = READ_ONCE(e);
+36   rcu_read_unlock();
+37 }
+</pre>
+</blockquote>
+
+<p>
+In this case, the outcome:
+
+<blockquote>
+<pre>
+(r1 == 1 &amp;&amp; r2 == 1 &amp;&amp; r3 == 1 &amp;&amp; r4 == 0 &amp&amp; r5 == 1)
+</pre>
+</blockquote>
+
+is entirely possible, as illustrated below:
+
+<p><img src="ReadersPartitionGP1.svg" alt="ReadersPartitionGP1.svg" width="100%"></p>
+
+<p>
+Again, an RCU read-side critical section can overlap almost all of a
+given grace period, just so long as it does not overlap the entire
+grace period.
+As a result, an RCU read-side critical section cannot partition a pair
+of RCU grace periods.
+
+<p>@@QQ@@
+How long a sequence of grace periods, each separated by an RCU read-side
+critical section, would be required to partition the RCU read-side
+critical sections at the beginning and end of the chain?
+<p>@@QQA@@
+In theory, an infinite number.
+In practice, an unknown number that is sensitive to both implementation
+details and timing considerations.
+Therefore, even in practice, RCU users must abide by the theoretical rather
+than the practical answer.
+<p>@@QQE@@
+
+<h3><a name="Disabling Preemption Does Not Block Grace Periods">
+Disabling Preemption Does Not Block Grace Periods</a></h3>
+
+<p>
+There was a time when disabling preemption on any given CPU would block
+subsequent grace periods.
+However, this was an accident of implementation and is not a requirement.
+And in the current Linux-kernel implementation, disabling preemption
+on a given CPU in fact does not block grace periods, as Oleg Nesterov
+<a href="http://marc.info/?l=linux-kernel&amp;m=143431078202914&amp;w=2">demonstrated</a>.
+
+<p>
+If you need a preempt-disable region to block grace periods, you need to add
+<tt>rcu_read_lock()</tt> and <tt>rcu_read_unlock()</tt>, for example
+as follows:
+
+<blockquote>
+<pre>
+ 1 preempt_disable();
+ 2 rcu_read_lock();
+ 3 do_something();
+ 4 rcu_read_unlock();
+ 5 preempt_enable();
+ 6
+ 7 /* Spinlocks implicitly disable preemption. */
+ 8 spin_lock(&amp;mylock);
+ 9 rcu_read_lock();
+10 do_something();
+11 rcu_read_unlock();
+12 spin_unlock(&amp;mylock);
+</pre>
+</blockquote>
+
+<p>
+In theory, you could enter the RCU read-side critical section first,
+but it is more efficient to keep the entire RCU read-side critical
+section contained in the preempt-disable region as shown above.
+Of course, RCU read-side critical sections that extend outside of
+preempt-disable regions will work correctly, but such critical sections
+can be preempted, which forces <tt>rcu_read_unlock()</tt> to do
+more work.
+And no, this is <i>not</i> an invitation to enclose all of your RCU
+read-side critical sections within preempt-disable regions, because
+doing so would degrade real-time response.
+
+<p>
+This non-requirement appeared with preemptible RCU.
+If you need a grace period that waits on non-preemptible code regions, use
+<a href="#Sched Flavor">RCU-sched</a>.
+
+<h2><a name="Parallelism Facts of Life">Parallelism Facts of Life</a></h2>
+
+<p>
+These parallelism facts of life are by no means specific to RCU, but
+the RCU implementation must abide by them.
+They therefore bear repeating:
+
+<ol>
+<li>	Any CPU or task may be delayed at any time,
+	and any attempts to avoid these delays by disabling
+	preemption, interrupts, or whatever are completely futile.
+	This is most obvious in preemptible user-level
+	environments and in virtualized environments (where
+	a given guest OS's VCPUs can be preempted at any time by
+	the underlying hypervisor), but can also happen in bare-metal
+	environments due to ECC errors, NMIs, and other hardware
+	events.
+	Although a delay of more than about 20 seconds can result
+	in splats, the RCU implementation is obligated to use
+	algorithms that can tolerate extremely long delays, but where
+	&ldquo;extremely long&rdquo; is not long enough to allow
+	wrap-around when incrementing a 64-bit counter.
+<li>	Both the compiler and the CPU can reorder memory accesses.
+	Where it matters, RCU must use compiler directives and
+	memory-barrier instructions to preserve ordering.
+<li>	Conflicting writes to memory locations in any given cache line
+	will result in expensive cache misses.
+	Greater numbers of concurrent writes and more-frequent
+	concurrent writes will result in more dramatic slowdowns.
+	RCU is therefore obligated to use algorithms that have
+	sufficient locality to avoid significant performance and
+	scalability problems.
+<li>	As a rough rule of thumb, only one CPU's worth of processing
+	may be carried out under the protection of any given exclusive
+	lock.
+	RCU must therefore use scalable locking designs.
+<li>	Counters are finite, especially on 32-bit systems.
+	RCU's use of counters must therefore tolerate counter wrap,
+	or be designed such that counter wrap would take way more
+	time than a single system is likely to run.
+	An uptime of ten years is quite possible, a runtime
+	of a century much less so.
+	As an example of the latter, RCU's dyntick-idle nesting counter
+	allows 54 bits for interrupt nesting level (this counter
+	is 64 bits even on a 32-bit system).
+	Overflowing this counter requires 2<sup>54</sup>
+	half-interrupts on a given CPU without that CPU ever going idle.
+	If a half-interrupt happened every microsecond, it would take
+	570 years of runtime to overflow this counter, which is currently
+	believed to be an acceptably long time.
+<li>	Linux systems can have thousands of CPUs running a single
+	Linux kernel in a single shared-memory environment.
+	RCU must therefore pay close attention to high-end scalability.
+</ol>
+
+<p>
+This last parallelism fact of life means that RCU must pay special
+attention to the preceding facts of life.
+The idea that Linux might scale to systems with thousands of CPUs would
+have been met with some skepticism in the 1990s, but these requirements
+would have otherwise have been unsurprising, even in the early 1990s.
+
+<h2><a name="Quality-of-Implementation Requirements">Quality-of-Implementation Requirements</a></h2>
+
+<p>
+These sections list quality-of-implementation requirements.
+Although an RCU implementation that ignores these requirements could
+still be used, it would likely be subject to limitations that would
+make it inappropriate for industrial-strength production use.
+Classes of quality-of-implementation requirements are as follows:
+
+<ol>
+<li>	<a href="#Specialization">Specialization</a>
+<li>	<a href="#Performance and Scalability">Performance and Scalability</a>
+<li>	<a href="#Composability">Composability</a>
+<li>	<a href="#Corner Cases">Corner Cases</a>
+</ol>
+
+<p>
+These classes is covered in the following sections.
+
+<h3><a name="Specialization">Specialization</a></h3>
+
+<p>
+RCU is and always has been intended primarily for read-mostly situations, as
+illustrated by the following figure.
+This means that RCU's read-side primitives are optimized, often at the
+expense of its update-side primitives.
+
+<p><img src="RCUApplicability.svg" alt="RCUApplicability.svg" width="70%"></p>
+
+<p>
+This focus on read-mostly situations means that RCU must interoperate
+with other synchronization primitives.
+For example, the <tt>add_gp()</tt> and <tt>remove_gp_synchronous()</tt>
+examples discussed earlier use RCU to protect readers and locking to
+coordinate updaters.
+However, the need extends much farther, requiring that a variety of
+synchronization primitives be legal within RCU read-side critical sections,
+including spinlocks, sequence locks, atomic operations, reference
+counters, and memory barriers.
+
+<p>@@QQ@@
+What about sleeping locks?
+<p>@@QQA@@
+These are forbidden within Linux-kernel RCU read-side critical sections
+because it is not legal to place a quiescent state (in this case,
+voluntary context switch) within an RCU read-side critical section.
+However, sleeping locks may be used within userspace RCU read-side critical
+sections, and also within Linux-kernel sleepable RCU
+<a href="#Sleepable RCU">(SRCU)</a>
+read-side critical sections.
+In addition, the -rt patchset turns spinlocks into a sleeping locks so
+that the corresponding critical sections can be preempted, which
+also means that these sleeplockified spinlocks (but not other sleeping locks!)
+may be acquire within -rt-Linux-kernel RCU read-side critical sections.
+
+<p>
+Note that it <i>is</i> legal for a normal RCU read-side critical section
+to conditionally acquire a sleeping locks (as in <tt>mutex_trylock()</tt>),
+but only as long as it does not loop indefinitely attempting to
+conditionally acquire that sleeping locks.
+The key point is that things like <tt>mutex_trylock()</tt>
+either return with the mutex held, or return an error indication if
+the mutex was not immediately available.
+Either way, <tt>mutex_trylock()</tt> returns immediately without sleeping.
+<p>@@QQE@@
+
+<p>
+It often comes as a surprise that many algorithms do not require a
+consistent view of data, but many can function in that mode,
+with network routing being the poster child.
+Internet routing algorithms take significant time to propagate
+updates, so that by the time an update arrives at a given system,
+that system has been sending network traffic the wrong way for
+a considerable length of time.
+Having a few threads continue to send traffic the wrong way for a
+few more milliseconds is clearly not a problem:  In the worst case,
+TCP retransmissions will eventually get the data where it needs to go.
+In general, when tracking the state of the universe outside of the
+computer, some level of inconsistency must be tolerated due to
+speed-of-light delays if nothing else.
+
+<p>
+Furthermore, uncertainty about external state is inherent in many cases.
+For example, a pair of veternarians might use heartbeat to determine
+whether or not a given cat was alive.
+But how long should they wait after the last heartbeat to decide that
+the cat is in fact dead?
+Waiting less than 400 milliseconds makes no sense because this would
+mean that a relaxed cat would be considered to cycle between death
+and life more than 100 times per minute.
+Moreover, just as with human beings, a cat's heart might stop for
+some period of time, so the exact wait period is a judgment call.
+One of our pair of veternarians might wait 30 seconds before pronouncing
+the cat dead, while the other might insist on waiting a full minute.
+The two veternarians would then disagree on the state of the cat during
+the final 30 seconds of the minute following the last heartbeat, as
+fancifully illustrated below:
+
+<p><img src="2013-08-is-it-dead.png" alt="2013-08-is-it-dead.png" width="431"></p>
+
+<p>
+Interestingly enough, this same situation applies to hardware.
+When push comes to shove, how do we tell whether or not some
+external server has failed?
+We send messages to it periodically, and declare it failed if we
+don't receive a response within a given period of time.
+Policy decisions can usually tolerate short
+periods of inconsistency.
+The policy was decided some time ago, and is only now being put into
+effect, so a few milliseconds of delay is normally inconsequential.
+
+<p>
+However, there are algorithms that absolutely must see consistent data.
+For example, the translation between a user-level SystemV semaphore
+ID to the corresponding in-kernel data structure is protected by RCU,
+but it is absolutely forbidden to update a semaphore that has just been
+removed.
+In the Linux kernel, this need for consistency is accommodated by acquiring
+spinlocks located in the in-kernel data structure from within
+the RCU read-side critical section, and this is indicated by the
+green box in the figure above.
+Many other techniques may be used, and are in fact used within the
+Linux kernel.
+
+<p>
+In short, RCU is not required to maintain consistency, and other
+mechanisms may be used in concert with RCU when consistency is required.
+RCU's specialization allows it to do its job extremely well, and its
+ability to interoperate with other synchronization mechanisms allows
+the right mix of synchronization tools to be used for a given job.
+
+<h3><a name="Performance and Scalability">Performance and Scalability</a></h3>
+
+<p>
+Energy efficiency is a critical component of performance today,
+and Linux-kernel RCU implementations must therefore avoid unnecessarily
+awakening idle CPUs.
+I cannot claim that this requirement was premeditated.
+In fact, I learned of it during a telephone conversation in which I
+was given &ldquo;frank and open&rdquo; feedback on the importance
+of energy efficiency in battery-powered systems and on specific
+energy-efficiency shortcomings of the Linux-kernel RCU implementation.
+In my experience, the battery-powered embedded community will consider
+any unnecessary wakeups to be extremely unfriendly acts.
+So much so that mere Linux-kernel-mailing-list posts are
+insufficient to vent their ire.
+
+<p>
+Memory consumption is not particularly important for in most
+situations, and has become decreasingly
+so as memory sizes have expanded and memory
+costs have plummeted.
+However, as I learned from Matt Mackall's
+<a href="http://elinux.org/Linux_Tiny-FAQ">bloatwatch</a>
+efforts, memory footprint is critically important on single-CPU systems with
+non-preemptible (<tt>CONFIG_PREEMPT=n</tt>) kernels, and thus
+<a href="http://lkml.org/lkml/2009/1/14/449">tiny RCU</a>
+was born.
+Josh Triplett has since taken over the small-memory banner with his
+<a href="https://tiny.wiki.kernel.org/">Linux kernel tinification</a>
+project, which resulted in
+<a href="#Sleepable RCU">SRCU</a>
+becoming optional for those kernels not needing it.
+
+<p>
+The remaining performance requirements are, for the most part,
+unsurprising.
+For example, in keeping with RCU's read-side specialization,
+<tt>rcu_dereference()</tt> should have negligible overhead (for
+example, suppression of a few minor compiler optimizations).
+Similarly, in non-preemptible environments, <tt>rcu_read_lock()</tt> and
+<tt>rcu_read_unlock()</tt> should have exactly zero overhead.
+
+<p>
+In preemptible environments, in the case where the RCU read-side
+critical section was not preempted (as will be the case for the
+highest-priority real-time process), <tt>rcu_read_lock()</tt> and
+<tt>rcu_read_unlock()</tt> should have minimal overhead.
+In particular, they should not contain atomic read-modify-write
+operations, memory-barrier instructions, preemption disabling,
+interrupt disabling, or backwards branches.
+However, in the case where the RCU read-side critical section was preempted,
+<tt>rcu_read_unlock()</tt> may acquire spinlocks and disable interrupts.
+This is why it is better to nest an RCU read-side critical section
+within a preempt-disable region than vice versa, at least in cases
+where that critical section is short enough to avoid unduly degrading
+real-time latencies.
+
+<p>
+The <tt>synchronize_rcu()</tt> grace-period-wait primitive is
+optimized for throughput.
+It may therefore incur several milliseconds of latency in addition to
+the duration of the longest RCU read-side critical section.
+On the other hand, multiple concurrent invocations of
+<tt>synchronize_rcu()</tt> are required to use batching optimizations
+so that they can be satisfied by a single underlying grace-period-wait
+operation.
+For example, in the Linux kernel, it is not unusual for a single
+grace-period-wait operation to serve more than
+<a href="https://www.usenix.org/conference/2004-usenix-annual-technical-conference/making-rcu-safe-deep-sub-millisecond-response">1,000 separate invocations</a>
+of <tt>synchronize_rcu()</tt>, thus amortizing the per-invocation
+overhead down to nearly zero.
+However, the grace-period optimization is also required to avoid
+measurable degradation of real-time scheduling and interrupt latencies.
+
+<p>
+In some cases, the multi-millisecond <tt>synchronize_rcu()</tt>
+latencies are unacceptable.
+In these cases, <tt>synchronize_rcu_expedited()</tt> may be used
+instead, reducing the grace-period latency down to a few tens of
+microseconds on small systems, at least in cases where the RCU read-side
+critical sections are short.
+There are currently no special latency requirements for
+<tt>synchronize_rcu_expedited()</tt> on large systems, but,
+consistent with the empirical nature of the RCU specification,
+that is subject to change.
+However, there most definitely are scalability requirements:
+A storm of <tt>synchronize_rcu_expedited()</tt> invocations on 4096
+CPUs should at least make reasonable forward progress.
+In return for its shorter latencies, <tt>synchronize_rcu_expedited()</tt>
+is permitted to impose modest degradation of real-time latency
+on non-idle online CPUs.
+That said, it will likely be necessary to take further steps to reduce this
+degradation, hopefully to roughly that of a scheduling-clock interrupt.
+
+<p>
+There are a number of situations where even
+<tt>synchronize_rcu_expedited()</tt>'s reduced grace-period
+latency is unacceptable.
+In these situations, the asynchronous <tt>call_rcu()</tt> can be
+used in place of <tt>synchronize_rcu()</tt> as follows:
+
+<blockquote>
+<pre>
+ 1 struct foo {
+ 2   int a;
+ 3   int b;
+ 4   struct rcu_head rh;
+ 5 };
+ 6
+ 7 static void remove_gp_cb(struct rcu_head *rhp)
+ 8 {
+ 9   struct foo *p = container_of(rhp, struct foo, rh);
+10
+11   kfree(p);
+12 }
+13
+14 bool remove_gp_asynchronous(void)
+15 {
+16   struct foo *p;
+17
+18   spin_lock(&amp;gp_lock);
+19   p = rcu_dereference(gp);
+20   if (!p) {
+21     spin_unlock(&amp;gp_lock);
+22     return false;
+23   }
+24   rcu_assign_pointer(gp, NULL);
+25   call_rcu(&amp;p-&gt;rh, remove_gp_cb);
+26   spin_unlock(&amp;gp_lock);
+27   return true;
+28 }
+</pre>
+</blockquote>
+
+<p>
+A definition of <tt>struct foo</tt> is finally needed, and appears
+on lines&nbsp;1-5.
+The function <tt>remove_gp_cb()</tt> is passed to <tt>call_rcu()</tt>
+on line&nbsp;25, and will be invoked after the end of a subsequent
+grace period.
+This gets the same effect as <tt>remove_gp_synchronous()</tt>,
+but without forcing the updater to wait for a grace period to elapse.
+The <tt>call_rcu()</tt> function may be used in a number of
+situations where neither <tt>synchronize_rcu()</tt> nor
+<tt>synchronize_rcu_expedited()</tt> would be legal,
+including within preempt-disable code, <tt>local_bh_disable()</tt> code,
+interrupt-disable code, and interrupt handlers.
+However, even <tt>call_rcu()</tt> is illegal within NMI handlers.
+The callback function (<tt>remove_gp_cb()</tt> in this case) will be
+executed within softirq (software interrupt) environment within the
+Linux kernel,
+either within a real softirq handler or under the protection
+of <tt>local_bh_disable()</tt>.
+In both the Linux kernel and in userspace, it is bad practice to
+write an RCU callback function that takes too long.
+Long-running operations should be relegated to separate threads or
+(in the Linux kernel) workqueues.
+
+<p>@@QQ@@
+Why does line&nbsp;19 use <tt>rcu_access_pointer()</tt>?
+After all, <tt>call_rcu()</tt> on line&nbsp;25 stores into the
+structure, which would interact badly with concurrent insertions.
+Doesn't this mean that <tt>rcu_dereference()</tt> is required?
+<p>@@QQA@@
+Presumably the <tt>-&gt;gp_lock</tt> acquired on line&nbsp;18 excludes
+any changes, including any insertions that <tt>rcu_dereference()</tt>
+would protect against.
+Therefore, any insertions will be delayed until after <tt>-&gt;gp_lock</tt>
+is released on line&nbsp;25, which in turn means that
+<tt>rcu_access_pointer()</tt> suffices.
+<p>@@QQE@@
+
+<p>
+However, all that <tt>remove_gp_cb()</tt> is doing is
+invoking <tt>kfree()</tt> on the data element.
+This is a common idiom, and is supported by <tt>kfree_rcu()</tt>,
+which allows &ldquo;fire and forget&rdquo; operation as shown below:
+
+<blockquote>
+<pre>
+ 1 struct foo {
+ 2   int a;
+ 3   int b;
+ 4   struct rcu_head rh;
+ 5 };
+ 6
+ 7 bool remove_gp_faf(void)
+ 8 {
+ 9   struct foo *p;
+10
+11   spin_lock(&amp;gp_lock);
+12   p = rcu_dereference(gp);
+13   if (!p) {
+14     spin_unlock(&amp;gp_lock);
+15     return false;
+16   }
+17   rcu_assign_pointer(gp, NULL);
+18   kfree_rcu(p, rh);
+19   spin_unlock(&amp;gp_lock);
+20   return true;
+21 }
+</pre>
+</blockquote>
+
+<p>
+Note that <tt>remove_gp_faf()</tt> simply invokes
+<tt>kfree_rcu()</tt> and proceeds, without any need to pay any
+further attention to the subsequent grace period and <tt>kfree()</tt>.
+It is permissible to invoke <tt>kfree_rcu()</tt> from the same
+environments as for <tt>call_rcu()</tt>.
+Interestingly enough, DYNIX/ptx had the equivalents of
+<tt>call_rcu()</tt> and <tt>kfree_rcu()</tt>, but not
+<tt>synchronize_rcu()</tt>.
+This was due to the fact that RCU was not heavily used within DYNIX/ptx,
+so the very few places that needed something like
+<tt>synchronize_rcu()</tt> simply open-coded it.
+
+<p>@@QQ@@
+Earlier it was claimed that <tt>call_rcu()</tt> and
+<tt>kfree_rcu()</tt> allowed updaters to avoid being blocked
+by readers.
+But how can that be correct, given that the invocation of the callback
+and the freeing of the memory (respectively) must still wait for
+a grace period to elapse?
+<p>@@QQA@@
+We could define things this way, but keep in mind that this sort of
+definition would say that updates in garbage-collected languages
+cannot complete until the next time the garbage collector runs,
+which does not seem at all reasonable.
+The key point is that in most cases, an updater using either
+<tt>call_rcu()</tt> or <tt>kfree_rcu()</tt> can proceed to the
+next update as soon as it has invoked <tt>call_rcu()</tt> or
+<tt>kfree_rcu()</tt>, without having to wait for a subsequent
+grace period.
+<p>@@QQE@@
+
+<p>
+But what if the updater must wait for the completion of code to be
+executed after the end of the grace period, but has other tasks
+that can be carried out in the meantime?
+The polling-style <tt>get_state_synchronize_rcu()</tt> and
+<tt>cond_synchronize_rcu()</tt> functions may be used for this
+purpose, as shown below:
+
+<blockquote>
+<pre>
+ 1 bool remove_gp_poll(void)
+ 2 {
+ 3   struct foo *p;
+ 4   unsigned long s;
+ 5
+ 6   spin_lock(&amp;gp_lock);
+ 7   p = rcu_access_pointer(gp);
+ 8   if (!p) {
+ 9     spin_unlock(&amp;gp_lock);
+10     return false;
+11   }
+12   rcu_assign_pointer(gp, NULL);
+13   spin_unlock(&amp;gp_lock);
+14   s = get_state_synchronize_rcu();
+15   do_something_while_waiting();
+16   cond_synchronize_rcu(s);
+17   kfree(p);
+18   return true;
+19 }
+</pre>
+</blockquote>
+
+<p>
+On line&nbsp;14, <tt>get_state_synchronize_rcu()</tt> obtains a
+&ldquo;cookie&rdquo; from RCU,
+then line&nbsp;15 carries out other tasks,
+and finally, line&nbsp;16 returns immediately if a grace period has
+elapsed in the meantime, but otherwise waits as required.
+The need for <tt>get_state_synchronize_rcu</tt> and
+<tt>cond_synchronize_rcu()</tt> has appeared quite recently,
+so it is too early to tell whether they will stand the test of time.
+
+<p>
+RCU thus provides a range of tools to allow updaters to strike the
+required tradeoff between latency, flexibility and CPU overhead.
+
+<h3><a name="Composability">Composability</a></h3>
+
+<p>
+Composability has received much attention in recent years, perhaps in part
+due to the collision of multicore hardware with object-oriented techniques
+designed in single-threaded environments for single-threaded use.
+And in theory, RCU read-side critical sections may be composed, and in
+fact may be nested arbitrarily deeply.
+In practice, as with all real-world implementations of composable
+constructs, there are limitations.
+
+<p>
+Implementations of RCU for which <tt>rcu_read_lock()</tt>
+and <tt>rcu_read_unlock()</tt> generate no code, such as
+Linux-kernel RCU when <tt>CONFIG_PREEMPT=n</tt>, can be
+nested arbitrarily deeply.
+After all, there is no overhead.
+Except that if all these instances of <tt>rcu_read_lock()</tt>
+and <tt>rcu_read_unlock()</tt> are visible to the compiler,
+compilation will eventually fail due to exhausting memory,
+mass storage, or user patience, whichever comes first.
+If the nesting is not visible to the compiler, as is the case with
+mutually recursive functions each in its own translation unit,
+stack overflow will result.
+If the nesting takes the form of loops, either the control variable
+will overflow or (in the Linux kernel) you will get an RCU CPU stall warning.
+Nevertheless, this class of RCU implementations is one
+of the most composable constructs in existence.
+
+<p>
+RCU implementations that explicitly track nesting depth
+are limited by the nesting-depth counter.
+For example, the Linux kernel's preemptible RCU limits nesting to
+<tt>INT_MAX</tt>.
+This should suffice for almost all practical purposes.
+That said, a consecutive pair of RCU read-side critical sections
+between which there is an operation that waits for a grace period
+cannot be enclosed in another RCU read-side critical section.
+This is because it is not legal to wait for a grace period within
+an RCU read-side critical section:  To do so would result either
+in deadlock or
+in RCU implicitly splitting the enclosing RCU read-side critical
+section, neither of which is conducive to a long-lived and prosperous
+kernel.
+
+<p>
+In short, although RCU read-side critical sections are highly composable,
+care is required in some situations, just as is the case for any other
+composable synchronization mechanism.
+
+<h3><a name="Corner Cases">Corner Cases</a></h3>
+
+<p>
+A given RCU workload might have an endless and intense stream of
+RCU read-side critical sections, perhaps even so intense that there
+was never a point in time during which there was not at least one
+RCU read-side critical section in flight.
+RCU cannot allow this situation to block grace periods:  As long as
+all the RCU read-side critical sections are finite, grace periods
+must also be finite.
+
+<p>
+That said, preemptible RCU implementations could potentially result
+in RCU read-side critical sections being preempted for long durations,
+which has the effect of creating a long-duration RCU read-side
+critical section.
+This situation can arise only in heavily loaded systems, but systems using
+real-time priorities are of course more vulnerable.
+Therefore, RCU priority boosting is provided to help deal with this
+case.
+That said, the exact requirements on RCU priority boosting will likely
+evolve as more experience accumulates.
+
+<p>
+Other workloads might have very high update rates.
+Although one can argue that such workloads should instead use
+something other than RCU, the fact remains that RCU must
+handle such workloads gracefully.
+This requirement is another factor driving batching of grace periods,
+but it is also the driving force behind the checks for large numbers
+of queued RCU callbacks in the <tt>call_rcu()</tt> code path.
+Finally, high update rates should not delay RCU read-side critical
+sections, although some read-side delays can occur when using
+<tt>synchronize_rcu_expedited()</tt>, courtesy of this function's use
+of <tt>try_stop_cpus()</tt>.
+(In the future, <tt>synchronize_rcu_expedited()</tt> will be
+converted to use lighter-weight inter-processor interrupts (IPIs),
+but this will still disturb readers, though to a much smaller degree.)
+
+<p>
+Although all three of these corner cases were understood in the early
+1990s, a simple user-level test consisting of <tt>close(open(path))</tt>
+in a tight loop
+in the early 2000s suddenly provided a much deeper appreciation of the
+high-update-rate corner case.
+This test also motivated addition of some RCU code to react to high update
+rates, for example, if a given CPU finds itself with more than 10,000
+RCU callbacks queued, it will cause RCU to take evasive action by
+more aggressively starting grace periods and more aggressively forcing
+completion of grace-period processing.
+This evasive action causes the grace period to complete more quickly,
+but at the cost of restricting RCU's batching optimizations, thus
+increasing the CPU overhead incurred by that grace period.
+
+<h2><a name="Software-Engineering Requirements">
+Software-Engineering Requirements</a></h2>
+
+<p>
+Between Murphy's Law and &ldquo;To err is human&rdquo;, it is necessary to
+guard against mishaps and misuse:
+
+<ol>
+<li>	It is all too easy to forget to use <tt>rcu_read_lock()</tt>
+	everywhere that it is needed, so kernels built with
+	<tt>CONFIG_PROVE_RCU=y</tt> will spat if
+	<tt>rcu_dereference()</tt> is used outside of an
+	RCU read-side critical section.
+	Update-side code can use <tt>rcu_dereference_protected()</tt>,
+	which takes a
+	<a href="https://lwn.net/Articles/371986/">lockdep expression</a>
+	to indicate what is providing the protection.
+	If the indicated protection is not provided, a lockdep splat
+	is emitted.
+
+	<p>
+	Code shared between readers and updaters can use
+	<tt>rcu_dereference_check()</tt>, which also takes a
+	lockdep expression, and emits a lockdep splat if neither
+	<tt>rcu_read_lock()</tt> nor the indicated protection
+	is in place.
+	In addition, <tt>rcu_dereference_raw()</tt> is used in those
+	(hopefully rare) cases where the required protection cannot
+	be easily described.
+	Finally, <tt>rcu_read_lock_held()</tt> is provided to
+	allow a function to verify that it has been invoked within
+	an RCU read-side critical section.
+	I was made aware of this set of requirements shortly after Thomas
+	Gleixner audited a number of RCU uses.
+<li>	A given function might wish to check for RCU-related preconditions
+	upon entry, before using any other RCU API.
+	The <tt>rcu_lockdep_assert()</tt> does this job,
+	asserting the expression in kernels having lockdep enabled
+	and doing nothing otherwise.
+<li>	It is also easy to forget to use <tt>rcu_assign_pointer()</tt>
+	and <tt>rcu_dereference()</tt>, perhaps (incorrectly)
+	substituting a simple assignment.
+	To catch this sort of error, a given RCU-protected pointer may be
+	tagged with <tt>__rcu</tt>, after which running sparse
+	with <tt>CONFIG_SPARSE_RCU_POINTER=y</tt> will complain
+	about simple-assignment accesses to that pointer.
+	Arnd Bergmann made me aware of this requirement, and also
+	supplied the needed
+	<a href="http://lwn.net/Articles/376011/">patch series</a>.
+<li>	Kernels built with <tt>CONFIG_DEBUG_OBJECTS_RCU_HEAD=y</tt>
+	will splat if a data element is passed to <tt>call_rcu()</tt>
+	twice in a row, without a grace period in between.
+	(This error is similar to a double free.)
+	The corresponding <tt>rcu_head</tt> structures that are
+	dynamically allocated are automatically tracked, but
+	<tt>rcu_head</tt> structures allocated on the stack
+	must be initialized with <tt>init_rcu_head_on_stack()</tt>
+	and cleaned up with <tt>destroy_rcu_head_on_stack()</tt>.
+	Similarly, statically allocated non-stack <tt>rcu_head</tt>
+	structures must be initialized with <tt>init_rcu_head()</tt>
+	and cleaned up with <tt>destroy_rcu_head()</tt>.
+	Mathieu Desnoyers made me aware of this requirement, and also
+	supplied the needed
+	<a href="https://lkml.org/lkml/2010/3/18/417">patch</a>.
+<li>	An infinite loop in an RCU read-side critical section will
+	eventually trigger an RCU CPU stall warning splat.
+	However, RCU is not obligated to produce this splat
+	unless there is a grace period waiting on that particular
+	RCU read-side critical section.
+	This requirement made itself known in the early 1990s, pretty
+	much the first time that it was necessary to debug a CPU stall.
+<li>	Although it would be very good to detect pointers leaking out
+	of RCU read-side critical sections, there is currently no
+	good way of doing this.
+	One complication is the need to distinguish between pointers
+	leaking and pointers that have been handed off from RCU to
+	some other synchronization mechanism, for example, reference
+	counting.
+<li>	In kernels built with <tt>CONFIG_RCU_TRACE=y</tt>, RCU-related
+	information is provided via both debugfs and event tracing.
+<li>	Open-coded use of <tt>rcu_assign_pointer()</tt> and
+	<tt>rcu_dereference()</tt> to create typical linked
+	data structures can be surprisingly error-prone.
+	Therefore, RCU-protected
+	<a href="http://lwn.net/Articles/609973/#RCU List APIs">linked lists</a>
+	and, more recently, RCU-protected
+	<a href="https://lwn.net/Articles/612100/">hash tables</a>
+	are available.
+	Many other special-purpose RCU-protected data structures are
+	available in the Linux kernel and the userspace RCU library.
+<li>	Some linked structures are created at compile time, but still
+	require <tt>__rcu</tt> checking.
+	The <tt>RCU_POINTER_INITIALIZER()</tt> macro serves this
+	purpose.
+<li>	It is not necessary to use <tt>rcu_assign_pointer()</tt>
+	when creating linked structures that are to be published via
+	a single external pointer.
+	The <tt>RCU_INIT_POINTER()</tt> macro is provided for
+	this task and also for assigning <tt>NULL</tt> pointers
+	at runtime.
+</ol>
+
+<p>
+This not a hard-and-fast list:  RCU's diagnostic capabilities will
+continue to be guided by the number and type of usage bugs found
+in real-world RCU usage.
+
+<h2><a name="Linux Kernel Complications">Linux Kernel Complications</a></h2>
+
+<p>
+The Linux kernel provides an interesting environment for all kinds of
+software, including RCU.
+Some of the relevant points of interest are as follows:
+
+<ol>
+<li>	<a href="#Configuration">Configuration</a>.
+<li>	<a href="#Firmware Interface">Firmware Interface</a>.
+<li>	<a href="#Early Boot">Early Boot</a>.
+<li>	<a href="#Interrupts and NMIs">
+	Interrupts and non-maskable interrupts (NMIs)</a>.
+<li>	<a href="#Loadable Modules">Loadable Modules</a>.
+<li>	<a href="#Hotplug CPU">Hotplug CPU</a>.
+<li>	<a href="#Scheduler and RCU">Scheduler and RCU</a>.
+<li>	<a href="#Tracing and RCU">Tracing and RCU</a>.
+<li>	<a href="#Energy Efficiency">Energy Efficiency</a>.
+<li>	<a href="#Performance, Scalability, Response Time, and Reliability">
+	Performance, Scalability, Response Time, and Reliability</a>.
+</ol>
+
+<p>
+This list is probably incomplete, but it does give a feel for the
+most notable Linux-kernel complications.
+Each of the following sections covers one of the above topics.
+
+<h3><a name="Configuration">Configuration</a></h3>
+
+<p>
+RCU's goal is automatic configuration, so that almost nobody
+needs to worry about RCU's <tt>Kconfig</tt> options.
+And for almost all users, RCU does in fact work well
+&ldquo;out of the box.&rdquo;
+
+<p>
+However, there are specialized use cases that are handled by
+kernel boot parameters and <tt>Kconfig</tt> options.
+Unfortunately, the <tt>Kconfig</tt> system will explicitly ask users
+about new <tt>Kconfig</tt> options, which requires almost all of them
+be hidden behind a <tt>CONFIG_RCU_EXPERT</tt> <tt>Kconfig</tt> option.
+
+<p>
+This all should be quite obvious, but the fact remains that
+Linus Torvalds recently had to
+<a href="http://marc.info/?l=linux-kernel&amp;m=142905739823385">remind</a>
+me of this requirement.
+
+<h3><a name="Firmware Interface">Firmware Interface</a></h3>
+
+<p>
+In many cases, kernel obtains information about the system from the
+firmware, and sometimes things are lost in translation.
+Or the translation is accurate, but the original message is bogus.
+
+<p>
+For example, some systems' firmware overreports the number of CPUs,
+sometimes by a large factor.
+If RCU naively believed the firmware, as it used to do,
+it would create too many per-CPU kthreads.
+Although the resulting system will still run correctly, the extra
+kthreads needlessly consume memory and can cause confusion
+when they show up in <tt>ps</tt> listings.
+
+<p>
+RCU must therefore wait for a given CPU to actually come online before
+it can allow itself to believe that the CPU actually exists.
+The resulting &ldquo;ghost CPUs&rdquo; (which are never going to
+come online) cause a number of
+<a href="http://paulmck.livejournal.com/37494.html">interesting complications</a>.
+
+<h3><a name="Early Boot">Early Boot</a></h3>
+
+<p>
+The Linux kernel's boot sequence is an interesting process,
+and RCU is used early, even before <tt>rcu_init()</tt>
+is invoked.
+In fact, a number of RCU's primitives can be used as soon as the
+initial task's <tt>task_struct</tt> is available and the
+boot CPU's per-CPU variables are set up.
+The read-side primitives (<tt>rcu_read_lock()</tt>,
+<tt>rcu_read_unlock()</tt>, <tt>rcu_dereference()</tt>,
+and <tt>rcu_access_pointer()</tt>) will operate normally very early on,
+as will <tt>rcu_assign_pointer()</tt>.
+
+<p>
+Although <tt>call_rcu()</tt> may be invoked at any
+time during boot, callbacks are not guaranteed to be invoked until after
+the scheduler is fully up and running.
+This delay in callback invocation is due to the fact that RCU does not
+invoke callbacks until it is fully initialized, and this full initialization
+cannot occur until after the scheduler has initialized itself to the
+point where RCU can spawn and run its kthreads.
+In theory, it would be possible to invoke callbacks earlier,
+however, this is not a panacea because there would be severe restrictions
+on what operations those callbacks could invoke.
+
+<p>
+Perhaps surprisingly, <tt>synchronize_rcu()</tt>,
+<a href="#Bottom-Half Flavor"><tt>synchronize_rcu_bh()</tt></a>
+(<a href="#Bottom-Half Flavor">discussed below</a>),
+and
+<a href="#Sched Flavor"><tt>synchronize_sched()</tt></a>
+will all operate normally
+during very early boot, the reason being that there is only one CPU
+and preemption is disabled.
+This means that the call <tt>synchronize_rcu()</tt> (or friends)
+itself is a quiescent
+state and thus a grace period, so the early-boot implementation can
+be a no-op.
+
+<p>
+Both <tt>synchronize_rcu_bh()</tt> and <tt>synchronize_sched()</tt>
+continue to operate normally through the remainder of boot, courtesy
+of the fact that preemption is disabled across their RCU read-side
+critical sections and also courtesy of the fact that there is still
+only one CPU.
+However, once the scheduler starts initializing, preemption is enabled.
+There is still only a single CPU, but the fact that preemption is enabled
+means that the no-op implementation of <tt>synchronize_rcu()</tt> no
+longer works in <tt>CONFIG_PREEMPT=y</tt> kernels.
+Therefore, as soon as the scheduler starts initializing, the early-boot
+fastpath is disabled.
+This means that <tt>synchronize_rcu()</tt> switches to its runtime
+mode of operation where it posts callbacks, which in turn means that
+any call to <tt>synchronize_rcu()</tt> will block until the corresponding
+callback is invoked.
+Unfortunately, the callback cannot be invoked until RCU's runtime
+grace-period machinery is up and running, which cannot happen until
+the scheduler has initialized itself sufficiently to allow RCU's
+kthreads to be spawned.
+Therefore, invoking <tt>synchronize_rcu()</tt> during scheduler
+initialization can result in deadlock.
+
+<p>@@QQ@@
+So what happens with <tt>synchronize_rcu()</tt> during
+scheduler initialization for <tt>CONFIG_PREEMPT=n</tt>
+kernels?
+<p>@@QQA@@
+In <tt>CONFIG_PREEMPT=n</tt> kernel, <tt>synchronize_rcu()</tt>
+maps directly to <tt>synchronize_sched()</tt>.
+Therefore, <tt>synchronize_rcu()</tt> works normally throughout
+boot in <tt>CONFIG_PREEMPT=n</tt> kernels.
+However, your code must also work in <tt>CONFIG_PREEMPT=y</tt> kernels,
+so it is still necessary to avoid invoking <tt>synchronize_rcu()</tt>
+during scheduler initialization.
+<p>@@QQE@@
+
+<p>
+I learned of these boot-time requirements as a result of a series of
+system hangs.
+
+<h3><a name="Interrupts and NMIs">Interrupts and NMIs</a></h3>
+
+<p>
+The Linux kernel has interrupts, and RCU read-side critical sections are
+legal within interrupt handlers and within interrupt-disabled regions
+of code, as are invocations of <tt>call_rcu()</tt>.
+
+<p>
+Some Linux-kernel architectures can enter an interrupt handler from
+non-idle process context, and then just never leave it, instead stealthily
+transitioning back to process context.
+This trick is sometimes used to invoke system calls from inside the kernel.
+These &ldquo;half-interrupts&rdquo; mean that RCU has to be very careful
+about how it counts interrupt nesting levels.
+I learned of this requirement the hard way during a rewrite
+of RCU's dyntick-idle code.
+
+<p>
+The Linux kernel has non-maskable interrupts (NMIs), and
+RCU read-side critical sections are legal within NMI handlers.
+Thankfully, RCU update-side primitives, including
+<tt>call_rcu()</tt>, are prohibited within NMI handlers.
+
+<p>
+The name notwithstanding, some Linux-kernel architectures
+can have nested NMIs, which RCU must handle correctly.
+Andy Lutomirski
+<a href="https://lkml.org/lkml/2014/11/21/642">surprised me</a>
+with this requirement;
+he also kindly surprised me with
+<a href="https://lkml.org/lkml/2014/11/22/1">an algorithm</a>
+that meets this requirement.
+
+<h3><a name="Loadable Modules">Loadable Modules</a></h3>
+
+<p>
+The Linux kernel has loadable modules, and these modules can
+also be unloaded.
+After a given module has been unloaded, any attempt to call
+one of its functions results in a segmentation fault.
+The module-unload functions must therefore cancel any
+delayed calls to loadable-module functions, for example,
+any outstanding <tt>mod_timer()</tt> must be dealt with
+via <tt>del_timer_sync()</tt> or similar.
+
+<p>
+Unfortunately, there is no way to cancel an RCU callback;
+once you invoke <tt>call_rcu()</tt>, the callback function is
+going to eventually be invoked, unless the system goes down first.
+Because it is normally considered socially irresponsible to crash the system
+in response to a module unload request, we need some other way
+to deal with in-flight RCU callbacks.
+
+<p>
+RCU therefore provides
+<tt><a href="http://lwn.net/Articles/217484/">rcu_barrier()</a></tt>,
+which waits until all in-flight RCU callbacks have been invoked.
+If a module uses <tt>call_rcu()</tt>, its exit function should therefore
+prevent any future invocation of <tt>call_rcu()</tt>, then invoke
+<tt>rcu_barrier()</tt>.
+In theory, the underlying module-unload code could invoke
+<tt>rcu_barrier()</tt> unconditionally, but in practice this would
+incur unacceptable latencies.
+
+<p>
+Nikita Danilov noted this requirement for an analogous filesystem-unmount
+situation, and Dipankar Sarma incorporated <tt>rcu_barrier()</tt> into RCU.
+The need for <tt>rcu_barrier()</tt> for module unloading became
+apparent later.
+
+<h3><a name="Hotplug CPU">Hotplug CPU</a></h3>
+
+<p>
+The Linux kernel supports CPU hotplug, which means that CPUs
+can come and go.
+It is of course illegal to use any RCU API member from an offline CPU.
+This requirement was present from day one in DYNIX/ptx, but
+on the other hand, the Linux kernel's CPU-hotplug implementation
+is &ldquo;interesting.&rdquo;
+
+<p>
+The Linux-kernel CPU-hotplug implementation has notifiers that
+are used to allow the various kernel subsystems (including RCU)
+to respond appropriately to a given CPU-hotplug operation.
+Most RCU operations may be invoked from CPU-hotplug notifiers,
+including even normal synchronous grace-period operations
+such as <tt>synchronize_rcu()</tt>.
+However, expedited grace-period operations such as
+<tt>synchronize_rcu_expedited()</tt> are not supported,
+due to the fact that current implementations block CPU-hotplug
+operations, which could result in deadlock.
+
+<p>
+In addition, all-callback-wait operations such as
+<tt>rcu_barrier()</tt> are also not supported, due to the
+fact that there are phases of CPU-hotplug operations where
+the outgoing CPU's callbacks will not be invoked until after
+the CPU-hotplug operation ends, which could also result in deadlock.
+
+<h3><a name="Scheduler and RCU">Scheduler and RCU</a></h3>
+
+<p>
+RCU depends on the scheduler, and the scheduler uses RCU to
+protect some of its data structures.
+This means the scheduler is forbidden from acquiring
+the runqueue locks and the priority-inheritance locks
+in the middle of an outermost RCU read-side critical section unless
+it also releases them before exiting that same
+RCU read-side critical section.
+This same prohibition also applies to any lock that is acquired
+while holding any lock to which this prohibition applies.
+Violating this rule results in deadlock.
+
+<p>
+For RCU's part, the preemptible-RCU <tt>rcu_read_unlock()</tt>
+implementation must be written carefully to avoid similar deadlocks.
+In particular, <tt>rcu_read_unlock()</tt> must tolerate an
+interrupt where the interrupt handler invokes both
+<tt>rcu_read_lock()</tt> and <tt>rcu_read_unlock()</tt>.
+This possibility requires <tt>rcu_read_unlock()</tt> to use
+negative nesting levels to avoid destructive recursion via
+interrupt handler's use of RCU.
+
+<p>
+This pair of mutual scheduler-RCU requirements came as a
+<a href="http://lwn.net/Articles/453002/">complete surprise</a>.
+
+<p>
+As noted above, RCU makes use of kthreads, and it is necessary to
+avoid excessive CPU-time accumulation by these kthreads.
+This requirement was no surprise, but RCU's violation of it
+when running context-switch-heavy workloads when built with
+<tt>CONFIG_NO_HZ_FULL=y</tt>
+<a href="http://www.rdrop.com/users/paulmck/scalability/paper/BareMetal.2015.01.15b.pdf">did come as a surprise [PDF]</a>.
+RCU has made good progress towards meeting this requirement, even
+for context-switch-have <tt>CONFIG_NO_HZ_FULL=y</tt> workloads,
+but there is room for further improvement.
+
+<h3><a name="Tracing and RCU">Tracing and RCU</a></h3>
+
+<p>
+It is possible to use tracing on RCU code, but tracing itself
+uses RCU.
+For this reason, <tt>rcu_dereference_raw_notrace()</tt>
+is provided for use by tracing, which avoids the destructive
+recursion that could otherwise ensue.
+This API is also used by virtualization in some architectures,
+where RCU readers execute in environments in which tracing
+cannot be used.
+The tracing folks both located the requirement and provided the
+needed fix, so this surprise requirement was relatively painless.
+
+<h3><a name="Energy Efficiency">Energy Efficiency</a></h3>
+
+<p>
+Interrupting idle CPUs is considered socially unacceptable,
+especially by people with battery-powered embedded systems.
+RCU therefore conserves energy by detecting which CPUs are
+idle, including tracking CPUs that have been interrupted from idle.
+This is a large part of the energy-efficiency requirement,
+so I learned of this via an irate phone call.
+
+<p>
+Because RCU avoids interrupting idle CPUs, it is illegal to
+execute an RCU read-side critical section on an idle CPU.
+(Kernels built with <tt>CONFIG_PROVE_RCU=y</tt> will splat
+if you try it.)
+The <tt>RCU_NONIDLE()</tt> macro and <tt>_rcuidle</tt>
+event tracing is provided to work around this restriction.
+In addition, <tt>rcu_is_watching()</tt> may be used to
+test whether or not it is currently legal to run RCU read-side
+critical sections on this CPU.
+I learned of the need for diagnostics on the one hand
+and <tt>RCU_NONIDLE()</tt> on the other while inspecting
+idle-loop code.
+Steven Rostedt supplied <tt>_rcuidle</tt> event tracing,
+which is used quite heavily in the idle loop.
+
+<p>
+It is similarly socially unacceptable to interrupt an
+<tt>nohz_full</tt> CPU running in userspace.
+RCU must therefore track <tt>nohz_full</tt> userspace
+execution.
+And in
+<a href="https://lwn.net/Articles/558284/"><tt>CONFIG_NO_HZ_FULL_SYSIDLE=y</tt></a>
+kernels, RCU must separately track idle CPUs on the one hand and
+CPUs that are either idle or executing in userspace on the other.
+In both cases, RCU must be able to sample state at two points in
+time, and be able to determine whether or not some other CPU spent
+any time idle and/or executing in userspace.
+
+<p>
+These energy-efficiency requirements have proven quite difficult to
+understand and to meet, for example, there have been more than five
+clean-sheet rewrites of RCU's energy-efficiency code, the last of
+which was finally able to demonstrate
+<a href="http://www.rdrop.com/users/paulmck/realtime/paper/AMPenergy.2013.04.19a.pdf">real energy savings running on real hardware [PDF]</a>.
+As noted earlier,
+I learned of many of these requirements via angry phone calls:
+Flaming me on the Linux-kernel mailing list was apparently not
+sufficient to fully vent their ire at RCU's energy-efficiency bugs!
+
+<h3><a name="Performance, Scalability, Response Time, and Reliability">
+Performance, Scalability, Response Time, and Reliability</a></h3>
+
+<p>
+Expanding on the
+<a href="#Performance and Scalability">earlier discussion</a>,
+RCU is used heavily by hot code paths in performance-critical
+portions of the Linux kernel's networking, security, virtualization,
+and scheduling code paths.
+RCU must therefore use efficient implementations, especially in its
+read-side primitives.
+To that end, it would be good if preemptible RCU's implementation
+of <tt>rcu_read_lock()</tt> could be inlined, however, doing
+this requires resolving <tt>#include</tt> issues with the
+<tt>task_struct</tt> structure.
+
+<p>
+The Linux kernel supports hardware configurations with up to
+4096 CPUs, which means that RCU must be extremely scalable.
+Algorithms that involve frequent acquisitions of global locks or
+frequent atomic operations on global variables simply cannot be
+tolerated within the RCU implementation.
+RCU therefore makes heavy use of a combining tree based on the
+<tt>rcu_node</tt> structure.
+RCU is required to tolerate all CPUs continuously invoking any
+combination of RCU's runtime primitives with minimal per-operation
+overhead.
+In fact, in many cases, increasing load must <i>decrease</i> the
+per-operation overhead, witness the batching optimizations for
+<tt>synchronize_rcu()</tt>, <tt>call_rcu()</tt>,
+<tt>synchronize_rcu_expedited()</tt>, and <tt>rcu_barrier()</tt>.
+As a general rule, RCU must cheerfully accept whatever the
+rest of the Linux kernel decides to throw at it.
+
+<p>
+The Linux kernel is used for real-time workloads, especially
+in conjunction with the
+<a href="https://rt.wiki.kernel.org/index.php/Main_Page">-rt patchset</a>.
+The real-time-latency response requirements are such that the
+traditional approach of disabling preemption across RCU
+read-side critical sections is inappropriate.
+Kernels built with <tt>CONFIG_PREEMPT=y</tt> therefore
+use an RCU implementation that allows RCU read-side critical
+sections to be preempted.
+This requirement made its presence known after users made it
+clear that an earlier
+<a href="http://lwn.net/Articles/107930/">real-time patch</a>
+did not meet their needs, in conjunction with some
+<a href="https://lkml.org/lkml/2005/3/17/199">RCU issues</a>
+encountered by a very early version of the -rt patchset.
+
+<p>
+In addition, RCU must make do with a sub-100-microsecond real-time latency
+budget.
+In fact, on smaller systems with the -rt patchset, the Linux kernel
+provides sub-20-microsecond real-time latencies for the whole kernel,
+including RCU.
+RCU's scalability and latency must therefore be sufficient for
+these sorts of configurations.
+To my surprise, the sub-100-microsecond real-time latency budget
+<a href="http://www.rdrop.com/users/paulmck/realtime/paper/bigrt.2013.01.31a.LCA.pdf">
+applies to even the largest systems [PDF]</a>,
+up to and including systems with 4096 CPUs.
+This real-time requirement motivated the grace-period kthread, which
+also simplified handling of a number of race conditions.
+
+<p>
+Finally, RCU's status as a synchronization primitive means that
+any RCU failure can result in arbitrary memory corruption that can be
+extremely difficult to debug.
+This means that RCU must be extremely reliable, which in
+practice also means that RCU must have an aggressive stress-test
+suite.
+This stress-test suite is called <tt>rcutorture</tt>.
+
+<p>
+Although the need for <tt>rcutorture</tt> was no surprise,
+the current immense popularity of the Linux kernel is posing
+interesting&mdash;and perhaps unprecedented&mdash;validation
+challenges.
+To see this, keep in mind that there are well over one billion
+instances of the Linux kernel running today, given Android
+smartphones, Linux-powered televisions, and servers.
+This number can be expected to increase sharply with the advent of
+the celebrated Internet of Things.
+
+<p>
+Suppose that RCU contains a race condition that manifests on average
+once per million years of runtime.
+This bug will be occurring about three times per <i>day</i> across
+the installed base.
+RCU could simply hide behind hardware error rates, given that no one
+should really expect their smartphone to last for a million years.
+However, anyone taking too much comfort from this thought should
+consider the fact that in most jurisdictions, a successful multi-year
+test of a given mechanism, which might include a Linux kernel,
+suffices for a number of types of safety-critical certifications.
+In fact, rumor has it that the Linux kernel is already being used
+in production for safety-critical applications.
+I don't know about you, but I would feel quite bad if a bug in RCU
+killed someone.
+Which might explain my recent focus on validation and verification.
+
+<h2><a name="Other RCU Flavors">Other RCU Flavors</a></h2>
+
+<p>
+One of the more surprising things about RCU is that there are now
+no fewer than five <i>flavors</i>, or API families.
+In addition, the primary flavor that has been the sole focus up to
+this point has two different implementations, non-preemptible and
+preemptible.
+The other four flavors are listed below, with requirements for each
+described in a separate section.
+
+<ol>
+<li>	<a href="#Bottom-Half Flavor">Bottom-Half Flavor</a>
+<li>	<a href="#Sched Flavor">Sched Flavor</a>
+<li>	<a href="#Sleepable RCU">Sleepable RCU</a>
+<li>	<a href="#Tasks RCU">Tasks RCU</a>
+</ol>
+
+<h3><a name="Bottom-Half Flavor">Bottom-Half Flavor</a></h3>
+
+<p>
+The softirq-disable (AKA &ldquo;bottom-half&rdquo;,
+hence the &ldquo;_bh&rdquo; abbreviations)
+flavor of RCU, or <i>RCU-bh</i>, was developed by
+Dipankar Sarma to provide a flavor of RCU that could withstand the
+network-based denial-of-service attacks researched by Robert
+Olsson.
+These attacks placed so much networking load on the system
+that some of the CPUs never exited softirq execution,
+which in turn prevented those CPUs from ever executing a context switch,
+which, in the RCU implementation of that time, prevented grace periods
+from ever ending.
+The result was an out-of-memory condition and a system hang.
+
+<p>
+The solution was the creation of RCU-bh, which does
+<tt>local_bh_disable()</tt>
+across its read-side critical sections, and which uses the transition
+from one type of softirq processing to another as a quiescent state
+in addition to context switch, idle, user mode, and offline.
+This means that RCU-bh grace periods can complete even when some of
+the CPUs execute in softirq indefinitely, thus allowing algorithms
+based on RCU-bh to withstand network-based denial-of-service attacks.
+
+<p>
+Because
+<tt>rcu_read_lock_bh()</tt> and <tt>rcu_read_unlock_bh()</tt>
+disable and re-enable softirq handlers, any attempt to start a softirq
+handlers during the
+RCU-bh read-side critical section will be deferred.
+In this case, <tt>rcu_read_unlock_bh()</tt>
+will invoke softirq processing, which can take considerable time.
+One can of course argue that this softirq overhead should be associated
+with the code following the RCU-bh read-side critical section rather
+than <tt>rcu_read_unlock_bh()</tt>, but the fact
+is that most profiling tools cannot be expected to make this sort
+of fine distinction.
+For example, suppose that a three-millisecond-long RCU-bh read-side
+critical section executes during a time of heavy networking load.
+There will very likely be an attempt to invoke at least one softirq
+handler during that three milliseconds, but any such invocation will
+be delayed until the time of the <tt>rcu_read_unlock_bh()</tt>.
+This can of course make it appear at first glance as if
+<tt>rcu_read_unlock_bh()</tt> was executing very slowly.
+
+<p>
+The
+<a href="http://lwn.net/Articles/609973/#RCU Per-Flavor API Table">RCU-bh API</a>
+includes
+<tt>rcu_read_lock_bh()</tt>,
+<tt>rcu_read_unlock_bh()</tt>,
+<tt>rcu_dereference_bh()</tt>,
+<tt>rcu_dereference_bh_check()</tt>,
+<tt>synchronize_rcu_bh()</tt>,
+<tt>synchronize_rcu_bh_expedited()</tt>,
+<tt>call_rcu_bh()</tt>,
+<tt>rcu_barrier_bh()</tt>, and
+<tt>rcu_read_lock_bh_held()</tt>.
+
+<h3><a name="Sched Flavor">Sched Flavor</a></h3>
+
+<p>
+Before preemptible RCU, waiting for an RCU grace period had the
+side effect of also waiting for all pre-existing interrupt
+and NMI handlers.
+However, there are legitimate preemptible-RCU implementations that
+do not have this property, given that any point in the code outside
+of an RCU read-side critical section can be a quiescent state.
+Therefore, <i>RCU-sched</i> was created, which follows &ldquo;classic&rdquo;
+RCU in that an RCU-sched grace period waits for for pre-existing
+interrupt and NMI handlers.
+In kernels built with <tt>CONFIG_PREEMPT=n</tt>, the RCU and RCU-sched
+APIs have identical implementations, while kernels built with
+<tt>CONFIG_PREEMPT=y</tt> provide a separate implementation for each.
+
+<p>
+Note well that in <tt>CONFIG_PREEMPT=y</tt> kernels,
+<tt>rcu_read_lock_sched()</tt> and <tt>rcu_read_unlock_sched()</tt>
+disable and re-enable preemption, respectively.
+This means that if there was a preemption attempt during the
+RCU-sched read-side critical section, <tt>rcu_read_unlock_sched()</tt>
+will enter the scheduler, with all the latency and overhead entailed.
+Just as with <tt>rcu_read_unlock_bh()</tt>, this can make it look
+as if <tt>rcu_read_unlock_sched()</tt> was executing very slowly.
+However, the highest-priority task won't be preempted, so that task
+will enjoy low-overhead <tt>rcu_read_unlock_sched()</tt> invocations.
+
+<p>
+The
+<a href="http://lwn.net/Articles/609973/#RCU Per-Flavor API Table">RCU-sched API</a>
+includes
+<tt>rcu_read_lock_sched()</tt>,
+<tt>rcu_read_unlock_sched()</tt>,
+<tt>rcu_read_lock_sched_notrace()</tt>,
+<tt>rcu_read_unlock_sched_notrace()</tt>,
+<tt>rcu_dereference_sched()</tt>,
+<tt>rcu_dereference_sched_check()</tt>,
+<tt>synchronize_sched()</tt>,
+<tt>synchronize_rcu_sched_expedited()</tt>,
+<tt>call_rcu_sched()</tt>,
+<tt>rcu_barrier_sched()</tt>, and
+<tt>rcu_read_lock_sched_held()</tt>.
+However, anything that disables preemption also marks an RCU-sched
+read-side critical section, including
+<tt>preempt_disable()</tt> and <tt>preempt_enable()</tt>,
+<tt>local_irq_save()</tt> and <tt>local_irq_restore()</tt>,
+and so on.
+
+<h3><a name="Sleepable RCU">Sleepable RCU</a></h3>
+
+<p>
+For well over a decade, someone saying &ldquo;I need to block within
+an RCU read-side critical section&rdquo; was a reliable indication
+that this someone did not understand RCU.
+After all, if you are always blocking in an RCU read-side critical
+section, you can probably afford to use a higher-overhead synchronization
+mechanism.
+However, that changed with the advent of the Linux kernel's notifiers,
+whose RCU read-side critical
+sections almost never sleep, but sometimes need to.
+This resulted in the introduction of
+<a href="https://lwn.net/Articles/202847/">sleepable RCU</a>,
+or <i>SRCU</i>.
+
+<p>
+SRCU allows different domains to be defined, with each such domain
+defined by an instance of an <tt>srcu_struct</tt> structure.
+A pointer to this structure must be passed in to each SRCU function,
+for example, <tt>synchronize_srcu(&amp;ss)</tt>, where
+<tt>ss</tt> is the <tt>srcu_struct</tt> structure.
+The key benefit of these domains is that a slow SRCU reader in one
+domain does not delay an SRCU grace period in some other domain.
+That said, one consequence of these domains is that read-side code
+must pass a &ldquo;cookie&rdquo; from <tt>srcu_read_lock()</tt>
+to <tt>srcu_read_unlock()</tt>, for example, as follows:
+
+<blockquote>
+<pre>
+ 1 int idx;
+ 2
+ 3 idx = srcu_read_lock(&amp;ss);
+ 4 do_something();
+ 5 srcu_read_unlock(&amp;ss, idx);
+</pre>
+</blockquote>
+
+<p>
+As noted above, it is legal to block within SRCU read-side critical sections,
+however, with great power comes great responsibility.
+If you block forever in one of a given domain's SRCU read-side critical
+sections, then that domain's grace periods will also be blocked forever.
+Of course, one good way to block forever is to deadlock, which can
+happen if any operation in a given domain's SRCU read-side critical
+section can block waiting, either directly or indirectly, for that domain's
+grace period to elapse.
+For example, this results in a self-deadlock:
+
+<blockquote>
+<pre>
+ 1 int idx;
+ 2
+ 3 idx = srcu_read_lock(&amp;ss);
+ 4 do_something();
+ 5 synchronize_srcu(&amp;ss);
+ 6 srcu_read_unlock(&amp;ss, idx);
+</pre>
+</blockquote>
+
+<p>
+However, if line&nbsp;5 acquired a mutex that was held across
+a <tt>synchronize_srcu()</tt> for domain <tt>ss</tt>,
+deadlock would still be possible.
+Furthermore, if line&nbsp;5 acquired a mutex that was held across
+a <tt>synchronize_srcu()</tt> for some other domain <tt>ss1</tt>,
+and if an <tt>ss1</tt>-domain SRCU read-side critical section
+acquired another mutex that was held across as <tt>ss</tt>-domain
+<tt>synchronize_srcu()</tt>,
+deadlock would again be possible.
+Such a deadlock cycle could extend across an arbitrarily large number
+of different SRCU domains.
+Again, with great power comes great responsibility.
+
+<p>
+Unlike the other RCU flavors, SRCU read-side critical sections can
+run on idle and even offline CPUs.
+This ability requires that <tt>srcu_read_lock()</tt> and
+<tt>srcu_read_unlock()</tt> contain memory barriers, which means
+that SRCU readers will run a bit slower than would RCU readers.
+It also motivates the <tt>smp_mb__after_srcu_read_unlock()</tt>
+API, which, in combination with <tt>srcu_read_unlock()</tt>,
+guarantees a full memory barrier.
+
+<p>
+The
+<a href="http://lwn.net/Articles/609973/#RCU Per-Flavor API Table">SRCU API</a>
+includes
+<tt>srcu_read_lock()</tt>,
+<tt>srcu_read_unlock()</tt>,
+<tt>srcu_dereference()</tt>,
+<tt>srcu_dereference_check()</tt>,
+<tt>synchronize_srcu()</tt>,
+<tt>synchronize_srcu_expedited()</tt>,
+<tt>call_srcu()</tt>,
+<tt>srcu_barrier()</tt>, and
+<tt>srcu_read_lock_held()</tt>.
+It also includes
+<tt>DEFINE_SRCU()</tt>,
+<tt>DEFINE_STATIC_SRCU()</tt>, and
+<tt>init_srcu_struct()</tt>
+APIs for defining and initializing <tt>srcu_struct</tt> structures.
+
+<h3><a name="Tasks RCU">Tasks RCU</a></h3>
+
+<p>
+Some forms of tracing use &ldquo;tramopolines&rdquo; to handle the
+binary rewriting required to install different types of probes.
+It would be good to be able to free old trampolines, which sounds
+like a job for some form of RCU.
+However, because it is necessary to be able to install a trace
+anywhere in the code, it is not possible to use read-side markers
+such as <tt>rcu_read_lock()</tt> and <tt>rcu_read_unlock()</tt>.
+In addition, it does not work to have these markers in the trampoline
+itself, because there would need to be instructions following
+<tt>rcu_read_unlock()</tt>.
+Although <tt>synchronize_rcu()</tt> would guarantee that execution
+reached the <tt>rcu_read_unlock()</tt>, it would not be able to
+guarantee that execution had completely left the trampoline.
+
+<p>
+The solution, in the form of
+<a href="http://lwn.net/Articles/607117/"><i>Tasks RCU</i></a>,
+is to have implicit
+read-side critical sections that are delimited by voluntary context
+switches, that is, calls to <tt>schedule()</tt>,
+<tt>cond_resched_rcu_qs()</tt>, and
+<tt>synchronize_rcu_tasks()</tt>.
+In addition, transitions to and from userspace execution also delimit
+tasks-RCU read-side critical sections.
+
+<p>
+The tasks-RCU API is quite compact, consisting only of
+<tt>call_rcu_tasks()</tt>,
+<tt>synchronize_rcu_tasks()</tt>, and
+<tt>rcu_barrier_tasks()</tt>.
+
+<h2><a name="Possible Future Changes">Possible Future Changes</a></h2>
+
+<p>
+One of the tricks that RCU uses to attain update-side scalability is
+to increase grace-period latency with increasing numbers of CPUs.
+If this becomes a serious problem, it will be necessary to rework the
+grace-period state machine so as to avoid the need for the additional
+latency.
+
+<p>
+Expedited grace periods scan the CPUs, so their latency and overhead
+increases with increasing numbers of CPUs.
+If this becomes a serious problem on large systems, it will be necessary
+to do some redesign to avoid this scalability problem.
+
+<p>
+RCU disables CPU hotplug in a few places, perhaps most notably in the
+expedited grace-period and <tt>rcu_barrier()</tt> operations.
+If there is a strong reason to use expedited grace periods in CPU-hotplug
+notifiers, it will be necessary to avoid disabling CPU hotplug.
+This would introduce some complexity, so there had better be a <i>very</i>
+good reason.
+
+<p>
+The tradeoff between grace-period latency on the one hand and interruptions
+of other CPUs on the other hand may need to be re-examined.
+The desire is of course for zero grace-period latency as well as zero
+interprocessor interrupts undertaken during an expedited grace period
+operation.
+While this ideal is unlikely to be achievable, it is quite possible that
+further improvements can be made.
+
+<p>
+The multiprocessor implementations of RCU use a combining tree that
+groups CPUs so as to reduce lock contention and increase cache locality.
+However, this combining tree does not spread its memory across NUMA
+nodes nor does it align the CPU groups with hardware features such
+as sockets or cores.
+Such spreading and alignment is currently believed to be unnecessary
+because the hotpath read-side primitives do not access the combining
+tree, nor does <tt>call_rcu()</tt> in the common case.
+If you believe that your architecture needs such spreading and alignment,
+then your architecture should also benefit from the
+<tt>rcutree.rcu_fanout_leaf</tt> boot parameter, which can be set
+to the number of CPUs in a socket, NUMA node, or whatever.
+If the number of CPUs is too large, use a fraction of the number of
+CPUs.
+If the number of CPUs is a large prime number, well, that certainly
+is an &ldquo;interesting&rdquo; architectural choice!
+More flexible arrangements might be considered, but only if
+<tt>rcutree.rcu_fanout_leaf</tt> has proven inadequate, and only
+if the inadequacy has been demonstrated by a carefully run and
+realistic system-level workload.
+
+<p>
+Please note that arrangements that require RCU to remap CPU numbers will
+require extremely good demonstration of need and full exploration of
+alternatives.
+
+<p>
+There is an embarrassingly large number of flavors of RCU, and this
+number has been increasing over time.
+Perhaps it will be possible to combine some at some future date.
+
+<p>
+RCU's various kthreads are reasonably recent additions.
+It is quite likely that adjustments will be required to more gracefully
+handle extreme loads.
+It might also be necessary to be able to relate CPU utilization by
+RCU's kthreads and softirq handlers to the code that instigated this
+CPU utilization.
+For example, RCU callback overhead might be charged back to the
+originating <tt>call_rcu()</tt> instance, though probably not
+in production kernels.
+
+<h2><a name="Summary">Summary</a></h2>
+
+<p>
+This document has presented more than two decade's worth of RCU
+requirements.
+Given that the requirements keep changing, this will not be the last
+word on this subject, but at least it serves to get an important
+subset of the requirements set forth.
+
+<h2><a name="Acknowledgments">Acknowledgments</a></h2>
+
+I am grateful to Steven Rostedt, Lai Jiangshan, Ingo Molnar,
+Oleg Nesterov, Borislav Petkov, Peter Zijlstra, Boqun Feng, and
+Andy Lutomirski for their help in rendering
+this article human readable, and to Michelle Rankin for her support
+of this effort.
+Other contributions are acknowledged in the Linux kernel's git archive.
+The cartoon is copyright (c) 2013 by Melissa Broussard,
+and is provided
+under the terms of the Creative Commons Attribution-Share Alike 3.0
+United States license.
+
+<p>@@QQAL@@
+
+</body></html>
diff --git a/Documentation/RCU/Design/htmlqqz.sh b/Documentation/RCU/Design/htmlqqz.sh
new file mode 100755
index 000000000000..d354f069559b
--- /dev/null
+++ b/Documentation/RCU/Design/htmlqqz.sh
@@ -0,0 +1,108 @@
+#!/bin/sh
+#
+# Usage: sh htmlqqz.sh file
+#
+# Extracts and converts quick quizzes in a proto-HTML document file.htmlx.
+# Commands, all of which must be on a line by themselves:
+#
+#	"<p>@@QQ@@": Start of a quick quiz.
+#	"<p>@@QQA@@": Start of a quick-quiz answer.
+#	"<p>@@QQE@@": End of a quick-quiz answer, and thus of the quick quiz.
+#	"<p>@@QQAL@@": Place to put quick-quiz answer list.
+#
+# Places the result in file.html.
+#
+# This program is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 2 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program; if not, you can access it online at
+# http://www.gnu.org/licenses/gpl-2.0.html.
+#
+# Copyright (c) 2013 Paul E. McKenney, IBM Corporation.
+
+fn=$1
+if test ! -r $fn.htmlx
+then
+	echo "Error: $fn.htmlx unreadable."
+	exit 1
+fi
+
+echo "<!-- DO NOT HAND EDIT. -->" > $fn.html
+echo "<!-- Instead, edit $fn.htmlx and run 'sh htmlqqz.sh $fn' -->" >> $fn.html
+awk < $fn.htmlx >> $fn.html '
+
+state == "" && $1 != "<p>@@QQ@@" && $1 != "<p>@@QQAL@@" {
+	print $0;
+	if ($0 ~ /^<p>@@QQ/)
+		print "Bad Quick Quiz command: " NR " (expected <p>@@QQ@@ or <p>@@QQAL@@)." > "/dev/stderr"
+	next;
+}
+
+state == "" && $1 == "<p>@@QQ@@" {
+	qqn++;
+	qqlineno = NR;
+	haveqq = 1;
+	state = "qq";
+	print "<p><a name=\"Quick Quiz " qqn "\"><b>Quick Quiz " qqn "</b>:</a>"
+	next;
+}
+
+state == "qq" && $1 != "<p>@@QQA@@" {
+	qq[qqn] = qq[qqn] $0 "\n";
+	print $0
+	if ($0 ~ /^<p>@@QQ/)
+		print "Bad Quick Quiz command: " NR ". (expected <p>@@QQA@@)" > "/dev/stderr"
+	next;
+}
+
+state == "qq" && $1 == "<p>@@QQA@@" {
+	state = "qqa";
+	print "<br><a href=\"#qq" qqn "answer\">Answer</a>"
+	next;
+}
+
+state == "qqa" && $1 != "<p>@@QQE@@" {
+	qqa[qqn] = qqa[qqn] $0 "\n";
+	if ($0 ~ /^<p>@@QQ/)
+		print "Bad Quick Quiz command: " NR " (expected <p>@@QQE@@)." > "/dev/stderr"
+	next;
+}
+
+state == "qqa" && $1 == "<p>@@QQE@@" {
+	state = "";
+	next;
+}
+
+state == "" && $1 == "<p>@@QQAL@@" {
+	haveqq = "";
+	print "<h3><a name=\"Answers to Quick Quizzes\">"
+	print "Answers to Quick Quizzes</a></h3>"
+	print "";
+	for (i = 1; i <= qqn; i++) {
+		print "<a name=\"qq" i "answer\"></a>"
+		print "<p><b>Quick Quiz " i "</b>:"
+		print qq[i];
+		print "";
+		print "</p><p><b>Answer</b>:"
+		print qqa[i];
+		print "";
+		print "</p><p><a href=\"#Quick%20Quiz%20" i "\"><b>Back to Quick Quiz " i "</b>.</a>"
+		print "";
+	}
+	next;
+}
+
+END {
+	if (state != "")
+		print "Unterminated Quick Quiz: " qqlineno "." > "/dev/stderr"
+	else if (haveqq)
+		print "Missing \"<p>@@QQAL@@\", no Quick Quiz." > "/dev/stderr"
+}'
-- 
2.5.2


^ permalink raw reply related	[flat|nested] 18+ messages in thread

* Re: [PATCH tip/core/rcu 1/8] documentation: Record RCU requirements
  2015-12-04 23:50 ` [PATCH tip/core/rcu 1/8] documentation: Record RCU requirements Paul E. McKenney
@ 2015-12-05  0:07   ` Josh Triplett
  2015-12-05  0:33     ` Paul E. McKenney
  2015-12-05  0:34   ` Josh Triplett
  1 sibling, 1 reply; 18+ messages in thread
From: Josh Triplett @ 2015-12-05  0:07 UTC (permalink / raw)
  To: Paul E. McKenney
  Cc: linux-kernel, mingo, jiangshanlai, dipankar, akpm,
	mathieu.desnoyers, tglx, peterz, rostedt, dhowells, edumazet,
	dvhart, fweisbec, oleg, bobby.prani

On Fri, Dec 04, 2015 at 03:50:19PM -0800, Paul E. McKenney wrote:
> This commit adds RCU requirements as published in a 2015 LWN series.
> Bringing these requirements in-tree allows them to be updated as changes
> are discovered.
> 
> Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
> ---
>  .../RCU/Design/Requirements/2013-08-is-it-dead.png |  Bin 0 -> 100825 bytes
>  .../Design/Requirements/GPpartitionReaders1.svg    |  374 +++
>  .../RCU/Design/Requirements/RCUApplicability.svg   |  237 ++
>  .../Design/Requirements/ReadersPartitionGP1.svg    |  639 +++++
>  .../RCU/Design/Requirements/Requirements.html      | 2799 ++++++++++++++++++++
>  .../RCU/Design/Requirements/Requirements.htmlx     | 2643 ++++++++++++++++++
>  Documentation/RCU/Design/htmlqqz.sh                |  108 +
>  7 files changed, 6800 insertions(+)
>  create mode 100644 Documentation/RCU/Design/Requirements/2013-08-is-it-dead.png
>  create mode 100644 Documentation/RCU/Design/Requirements/GPpartitionReaders1.svg
>  create mode 100644 Documentation/RCU/Design/Requirements/RCUApplicability.svg
>  create mode 100644 Documentation/RCU/Design/Requirements/ReadersPartitionGP1.svg
>  create mode 100644 Documentation/RCU/Design/Requirements/Requirements.html
>  create mode 100644 Documentation/RCU/Design/Requirements/Requirements.htmlx

If Requirements.html is machine-generated and shouldn't be hand-edited,
and it can be generated without any special tools, then I don't think it
should be committed in the tree; I'd suggest putting it in .gitignore
and generating it from one of the various "make docs" invocations.

Alternatively, if you want to make sure a usable version is in-tree, you
could make the script reversible (easy enough to do if the compiled
version includes some marker comments or similar), and then tell people
to run it in reverse mode, edit, and run it in forward mode.  Then you
don't need the .htmlx file at all. :)

^ permalink raw reply	[flat|nested] 18+ messages in thread

* Re: [PATCH tip/core/rcu 1/8] documentation: Record RCU requirements
  2015-12-05  0:07   ` Josh Triplett
@ 2015-12-05  0:33     ` Paul E. McKenney
  2015-12-05  0:38       ` Josh Triplett
  0 siblings, 1 reply; 18+ messages in thread
From: Paul E. McKenney @ 2015-12-05  0:33 UTC (permalink / raw)
  To: Josh Triplett
  Cc: linux-kernel, mingo, jiangshanlai, dipankar, akpm,
	mathieu.desnoyers, tglx, peterz, rostedt, dhowells, edumazet,
	dvhart, fweisbec, oleg, bobby.prani

On Fri, Dec 04, 2015 at 04:07:19PM -0800, Josh Triplett wrote:
> On Fri, Dec 04, 2015 at 03:50:19PM -0800, Paul E. McKenney wrote:
> > This commit adds RCU requirements as published in a 2015 LWN series.
> > Bringing these requirements in-tree allows them to be updated as changes
> > are discovered.
> > 
> > Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
> > ---
> >  .../RCU/Design/Requirements/2013-08-is-it-dead.png |  Bin 0 -> 100825 bytes
> >  .../Design/Requirements/GPpartitionReaders1.svg    |  374 +++
> >  .../RCU/Design/Requirements/RCUApplicability.svg   |  237 ++
> >  .../Design/Requirements/ReadersPartitionGP1.svg    |  639 +++++
> >  .../RCU/Design/Requirements/Requirements.html      | 2799 ++++++++++++++++++++
> >  .../RCU/Design/Requirements/Requirements.htmlx     | 2643 ++++++++++++++++++
> >  Documentation/RCU/Design/htmlqqz.sh                |  108 +
> >  7 files changed, 6800 insertions(+)
> >  create mode 100644 Documentation/RCU/Design/Requirements/2013-08-is-it-dead.png
> >  create mode 100644 Documentation/RCU/Design/Requirements/GPpartitionReaders1.svg
> >  create mode 100644 Documentation/RCU/Design/Requirements/RCUApplicability.svg
> >  create mode 100644 Documentation/RCU/Design/Requirements/ReadersPartitionGP1.svg
> >  create mode 100644 Documentation/RCU/Design/Requirements/Requirements.html
> >  create mode 100644 Documentation/RCU/Design/Requirements/Requirements.htmlx
> 
> If Requirements.html is machine-generated and shouldn't be hand-edited,
> and it can be generated without any special tools, then I don't think it
> should be committed in the tree; I'd suggest putting it in .gitignore
> and generating it from one of the various "make docs" invocations.

I considered doing that, but then decided that it is nice for people to
be able train their browser directly on the file without having to know
what scripts to run.

Hmmm...  I suppose I could construct a Makefile that dealt with that
though...  I will give this some thought, and if it looks good, I will
add the Makefiles and "git rm" the .htmlx files.

> Alternatively, if you want to make sure a usable version is in-tree, you
> could make the script reversible (easy enough to do if the compiled
> version includes some marker comments or similar), and then tell people
> to run it in reverse mode, edit, and run it in forward mode.  Then you
> don't need the .htmlx file at all. :)

Decades ago, back when I (against all evidence) believed I could
consistently avoid making stupid mistakes, you might have been able to
convince me that this was a good idea.  ;-)

							Thanx, Paul


^ permalink raw reply	[flat|nested] 18+ messages in thread

* Re: [PATCH tip/core/rcu 1/8] documentation: Record RCU requirements
  2015-12-05  0:33     ` Paul E. McKenney
@ 2015-12-05  0:38       ` Josh Triplett
  2015-12-05  1:56         ` Paul E. McKenney
  0 siblings, 1 reply; 18+ messages in thread
From: Josh Triplett @ 2015-12-05  0:38 UTC (permalink / raw)
  To: Paul E. McKenney
  Cc: linux-kernel, mingo, jiangshanlai, dipankar, akpm,
	mathieu.desnoyers, tglx, peterz, rostedt, dhowells, edumazet,
	dvhart, fweisbec, oleg, bobby.prani

On Fri, Dec 04, 2015 at 04:33:32PM -0800, Paul E. McKenney wrote:
> On Fri, Dec 04, 2015 at 04:07:19PM -0800, Josh Triplett wrote:
> > On Fri, Dec 04, 2015 at 03:50:19PM -0800, Paul E. McKenney wrote:
> > > This commit adds RCU requirements as published in a 2015 LWN series.
> > > Bringing these requirements in-tree allows them to be updated as changes
> > > are discovered.
> > > 
> > > Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
> > > ---
> > >  .../RCU/Design/Requirements/2013-08-is-it-dead.png |  Bin 0 -> 100825 bytes
> > >  .../Design/Requirements/GPpartitionReaders1.svg    |  374 +++
> > >  .../RCU/Design/Requirements/RCUApplicability.svg   |  237 ++
> > >  .../Design/Requirements/ReadersPartitionGP1.svg    |  639 +++++
> > >  .../RCU/Design/Requirements/Requirements.html      | 2799 ++++++++++++++++++++
> > >  .../RCU/Design/Requirements/Requirements.htmlx     | 2643 ++++++++++++++++++
> > >  Documentation/RCU/Design/htmlqqz.sh                |  108 +
> > >  7 files changed, 6800 insertions(+)
> > >  create mode 100644 Documentation/RCU/Design/Requirements/2013-08-is-it-dead.png
> > >  create mode 100644 Documentation/RCU/Design/Requirements/GPpartitionReaders1.svg
> > >  create mode 100644 Documentation/RCU/Design/Requirements/RCUApplicability.svg
> > >  create mode 100644 Documentation/RCU/Design/Requirements/ReadersPartitionGP1.svg
> > >  create mode 100644 Documentation/RCU/Design/Requirements/Requirements.html
> > >  create mode 100644 Documentation/RCU/Design/Requirements/Requirements.htmlx
> > 
> > If Requirements.html is machine-generated and shouldn't be hand-edited,
> > and it can be generated without any special tools, then I don't think it
> > should be committed in the tree; I'd suggest putting it in .gitignore
> > and generating it from one of the various "make docs" invocations.
> 
> I considered doing that, but then decided that it is nice for people to
> be able train their browser directly on the file without having to know
> what scripts to run.
> 
> Hmmm...  I suppose I could construct a Makefile that dealt with that
> though...  I will give this some thought, and if it looks good, I will
> add the Makefiles and "git rm" the .htmlx files.
> 
> > Alternatively, if you want to make sure a usable version is in-tree, you
> > could make the script reversible (easy enough to do if the compiled
> > version includes some marker comments or similar), and then tell people
> > to run it in reverse mode, edit, and run it in forward mode.  Then you
> > don't need the .htmlx file at all. :)
> 
> Decades ago, back when I (against all evidence) believed I could
> consistently avoid making stupid mistakes, you might have been able to
> convince me that this was a good idea.  ;-)

Because you don't want to complicate the script, or because you don't
want to accidentally edit the wrong version?  (Note that a carefully
written script would mean it doesn't matter which version you edit.)

A third alternative would be to include the answers inline right after
the questions, and optionally add a tiny bit of JavaScript that hides
them by default and lets you click to show the answer. :)

^ permalink raw reply	[flat|nested] 18+ messages in thread

* Re: [PATCH tip/core/rcu 1/8] documentation: Record RCU requirements
  2015-12-05  0:38       ` Josh Triplett
@ 2015-12-05  1:56         ` Paul E. McKenney
  2015-12-05 21:19           ` Paul E. McKenney
  0 siblings, 1 reply; 18+ messages in thread
From: Paul E. McKenney @ 2015-12-05  1:56 UTC (permalink / raw)
  To: Josh Triplett
  Cc: linux-kernel, mingo, jiangshanlai, dipankar, akpm,
	mathieu.desnoyers, tglx, peterz, rostedt, dhowells, edumazet,
	dvhart, fweisbec, oleg, bobby.prani

On Fri, Dec 04, 2015 at 04:38:42PM -0800, Josh Triplett wrote:
> On Fri, Dec 04, 2015 at 04:33:32PM -0800, Paul E. McKenney wrote:
> > On Fri, Dec 04, 2015 at 04:07:19PM -0800, Josh Triplett wrote:
> > > On Fri, Dec 04, 2015 at 03:50:19PM -0800, Paul E. McKenney wrote:
> > > > This commit adds RCU requirements as published in a 2015 LWN series.
> > > > Bringing these requirements in-tree allows them to be updated as changes
> > > > are discovered.
> > > > 
> > > > Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
> > > > ---
> > > >  .../RCU/Design/Requirements/2013-08-is-it-dead.png |  Bin 0 -> 100825 bytes
> > > >  .../Design/Requirements/GPpartitionReaders1.svg    |  374 +++
> > > >  .../RCU/Design/Requirements/RCUApplicability.svg   |  237 ++
> > > >  .../Design/Requirements/ReadersPartitionGP1.svg    |  639 +++++
> > > >  .../RCU/Design/Requirements/Requirements.html      | 2799 ++++++++++++++++++++
> > > >  .../RCU/Design/Requirements/Requirements.htmlx     | 2643 ++++++++++++++++++
> > > >  Documentation/RCU/Design/htmlqqz.sh                |  108 +
> > > >  7 files changed, 6800 insertions(+)
> > > >  create mode 100644 Documentation/RCU/Design/Requirements/2013-08-is-it-dead.png
> > > >  create mode 100644 Documentation/RCU/Design/Requirements/GPpartitionReaders1.svg
> > > >  create mode 100644 Documentation/RCU/Design/Requirements/RCUApplicability.svg
> > > >  create mode 100644 Documentation/RCU/Design/Requirements/ReadersPartitionGP1.svg
> > > >  create mode 100644 Documentation/RCU/Design/Requirements/Requirements.html
> > > >  create mode 100644 Documentation/RCU/Design/Requirements/Requirements.htmlx
> > > 
> > > If Requirements.html is machine-generated and shouldn't be hand-edited,
> > > and it can be generated without any special tools, then I don't think it
> > > should be committed in the tree; I'd suggest putting it in .gitignore
> > > and generating it from one of the various "make docs" invocations.
> > 
> > I considered doing that, but then decided that it is nice for people to
> > be able train their browser directly on the file without having to know
> > what scripts to run.
> > 
> > Hmmm...  I suppose I could construct a Makefile that dealt with that
> > though...  I will give this some thought, and if it looks good, I will
> > add the Makefiles and "git rm" the .htmlx files.
> > 
> > > Alternatively, if you want to make sure a usable version is in-tree, you
> > > could make the script reversible (easy enough to do if the compiled
> > > version includes some marker comments or similar), and then tell people
> > > to run it in reverse mode, edit, and run it in forward mode.  Then you
> > > don't need the .htmlx file at all. :)
> > 
> > Decades ago, back when I (against all evidence) believed I could
> > consistently avoid making stupid mistakes, you might have been able to
> > convince me that this was a good idea.  ;-)
> 
> Because you don't want to complicate the script, or because you don't
> want to accidentally edit the wrong version?  (Note that a carefully
> written script would mean it doesn't matter which version you edit.)

I prefer the simpler script.

> A third alternative would be to include the answers inline right after
> the questions, and optionally add a tiny bit of JavaScript that hides
> them by default and lets you click to show the answer. :)

That does have some advantages.  I would welcome a patch that added
the JavaScript.

							Thanx, Paul


^ permalink raw reply	[flat|nested] 18+ messages in thread

* Re: [PATCH tip/core/rcu 1/8] documentation: Record RCU requirements
  2015-12-05  1:56         ` Paul E. McKenney
@ 2015-12-05 21:19           ` Paul E. McKenney
  0 siblings, 0 replies; 18+ messages in thread
From: Paul E. McKenney @ 2015-12-05 21:19 UTC (permalink / raw)
  To: Josh Triplett
  Cc: linux-kernel, mingo, jiangshanlai, dipankar, akpm,
	mathieu.desnoyers, tglx, peterz, rostedt, dhowells, edumazet,
	dvhart, fweisbec, oleg, bobby.prani

On Fri, Dec 04, 2015 at 05:56:01PM -0800, Paul E. McKenney wrote:
> On Fri, Dec 04, 2015 at 04:38:42PM -0800, Josh Triplett wrote:

[ . . . ]

> > Because you don't want to complicate the script, or because you don't
> > want to accidentally edit the wrong version?  (Note that a carefully
> > written script would mean it doesn't matter which version you edit.)
> 
> I prefer the simpler script.
> 
> > A third alternative would be to include the answers inline right after
> > the questions, and optionally add a tiny bit of JavaScript that hides
> > them by default and lets you click to show the answer. :)
> 
> That does have some advantages.  I would welcome a patch that added
> the JavaScript.

And I did try a CSS approach with this CSS definition:

	<style media="screen" type="text/css">
	.quick-quiz-wrap {
		position: relative;
	}
	.quick-quiz-wrap .quick-quiz-content {
		display: none;
		position: absolute;
		bottom: -10px;
		left: 10px;
		right: 10px;
		background-color: lightgrey;
		padding: 20px;
	}
	.quick-quiz-wrap:hover .quick-quiz-content {
		display: block;
	}
	</style>

I linked to this as follows:

	<link rel="stylesheet" type="text/css" href="../quick-quiz.css" media="screen">

And used it like this:

	<p><b>Quick Quiz:</b>
	Wait a minute!
	You said that updaters can make useful forward progress concurrently
	with readers, but pre-existing readers will block
	<tt>synchronize_rcu()</tt>!!!
	Just who are you trying to fool???
	</p>

	<div class="quick-quiz-wrap">
		<b>Quick Quiz Answer</b>
		<div class="quick-quiz-content">
			<p>First, if updaters do not wish to be blocked
			by readers, they can use <tt>call_rcu()</tt> or
			<tt>kfree_rcu()</tt>, which will be discussed later.
			Second, even when using <tt>synchronize_rcu()</tt>,
			the other update-side code does run concurrently with
			readers, whether pre-existing or not.
		</div> 
	</div>

This proved fragile.  It worked in trivial cases, but with non-trivial
content in a non-trivial document, it refused to show anything on mouse
hover.  Even in trivial cases, it sometimes made strange choices on
where to display the text.  At this point, I drop back to the old-school
scripted approach.  It has the advantage of working consistently and
providing good user experience, with the HTML unconditionally present. ;-)

							Thanx, Paul


^ permalink raw reply	[flat|nested] 18+ messages in thread

* Re: [PATCH tip/core/rcu 1/8] documentation: Record RCU requirements
  2015-12-04 23:50 ` [PATCH tip/core/rcu 1/8] documentation: Record RCU requirements Paul E. McKenney
  2015-12-05  0:07   ` Josh Triplett
@ 2015-12-05  0:34   ` Josh Triplett
  2015-12-05  1:54     ` Paul E. McKenney
  1 sibling, 1 reply; 18+ messages in thread
From: Josh Triplett @ 2015-12-05  0:34 UTC (permalink / raw)
  To: Paul E. McKenney
  Cc: linux-kernel, mingo, jiangshanlai, dipankar, akpm,
	mathieu.desnoyers, tglx, peterz, rostedt, dhowells, edumazet,
	dvhart, fweisbec, oleg, bobby.prani

The content of the document seems fine; a few comments below on
meta-issues.

On Fri, Dec 04, 2015 at 03:50:19PM -0800, Paul E. McKenney wrote:
> --- /dev/null
> +++ b/Documentation/RCU/Design/Requirements/Requirements.html
> @@ -0,0 +1,2799 @@
> +<!-- DO NOT HAND EDIT. -->
> +<!-- Instead, edit Requirements.htmlx and run 'sh htmlqqz.sh Requirements' -->
> +<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN"
> +        "http://www.w3.org/TR/html4/loose.dtd">

Nit: these days, this should just be:
<!doctype html>

> +        <html>
> +        <head><title>A Tour Through RCU's Requirements [LWN.net]</title>
> +        <meta HTTP-EQUIV="Content-Type" CONTENT="text/html; charset=iso-8859-1">

Is there a good reason to not use charset=utf-8 here?

> +
> +<h1>A Tour Through RCU's Requirements</h1>
> +
> +<p>Copyright IBM Corporation, 2015</p>

If you're aiming for a properly formatted copyright notice, the year
typically comes first, followed by the copyright holder.  That said,
your corporate guidelines presumably have a specific format; is this
that format?

> +<p>Author: Paul E.&nbsp;McKenney</p>
> +<p><i>The initial version of this document appeared in the
> +<a href="http://lwn.net/">LWN</a> articles
> +<a href="http://lwn.net/Articles/652156/">here</a>,
> +<a href="http://lwn.net/Articles/652677/">here</a>, and
> +<a href="http://lwn.net/Articles/653326/">here</a>.</i></p>

s/http/https/g

> +<p>
> +All that aside, here are the categories of currently known RCU requirements:
> +</p>
> +
> +<ol>
> +<li>	<a href="#Fundamental Requirements">

Anchors don't typically have spaces in them.  This may work in some
browsers, but it doesn't validate.  You should either use %20 or
(better) use a '-'.

> +<p>
> +This is followed by a <a href="#Summary">summary</a>,
> +which is in turn followed by the inevitable
> +<a href="#Answers to Quick Quizzes">answers to the quick quizzes</a>.

(Note: when editing anchors, don't forget to handle the target of this
in the generation script.)

> +<p>
> +This scenario resembles one of the first uses of RCU in
> +<a href="http://en.wikipedia.org/wiki/DYNIX">DYNIX/ptx</a>,

s/http/https/

> +<p>
> +However, this temptation must be resisted because there are a
> +surprisingly large number of ways that the compiler
> +(to say nothing of
> +<a href="http://www.openvms.compaq.com/wizard/wiz_2637.html">DEC Alpha CPUs</a>)

This link sadly doesn't seem to work anymore; it redirects to HP's
general page on OpenVMS, not a copy of that specific article.o
Use this instead, assuming no current live version exists:
https://web.archive.org/web/20120720095054/http://www.openvms.compaq.com/wizard/wiz_2637.html

> +<li>	It is also easy to forget to use <tt>rcu_assign_pointer()</tt>
> +	and <tt>rcu_dereference()</tt>, perhaps (incorrectly)
> +	substituting a simple assignment.
> +	To catch this sort of error, a given RCU-protected pointer may be
> +	tagged with <tt>__rcu</tt>, after which running sparse
> +	with <tt>CONFIG_SPARSE_RCU_POINTER=y</tt> will complain
> +	about simple-assignment accesses to that pointer.
> +	Arnd Bergmann made me aware of this requirement, and also
> +	supplied the needed
> +	<a href="http://lwn.net/Articles/376011/">patch series</a>.

s/http/https/

> +<li>	Open-coded use of <tt>rcu_assign_pointer()</tt> and
> +	<tt>rcu_dereference()</tt> to create typical linked
> +	data structures can be surprisingly error-prone.
> +	Therefore, RCU-protected
> +	<a href="http://lwn.net/Articles/609973/#RCU List APIs">linked lists</a>

s/http/https/

> +<p>
> +This all should be quite obvious, but the fact remains that
> +Linus Torvalds recently had to
> +<a href="http://marc.info/?l=linux-kernel&amp;m=142905739823385">remind</a>
> +me of this requirement.

I'd suggest using the lkml.kernel.org redirector for this link, along
with a Message-Id.

> +<p>
> +The name notwithstanding, some Linux-kernel architectures
> +can have nested NMIs, which RCU must handle correctly.
> +Andy Lutomirski
> +<a href="https://lkml.org/lkml/2014/11/21/642">surprised me</a>
> +with this requirement;
> +he also kindly surprised me with
> +<a href="https://lkml.org/lkml/2014/11/22/1">an algorithm</a>
> +that meets this requirement.

These links should both use lkml.kernel.org as well.  Doubly important
because lkml.org is often down or has broken messages in its archive.

> +<p>
> +RCU therefore provides
> +<tt><a href="http://lwn.net/Articles/217484/">rcu_barrier()</a></tt>,

s/http/https/

> +<p>
> +This pair of mutual scheduler-RCU requirements came as a
> +<a href="http://lwn.net/Articles/453002/">complete surprise</a>.

s/http/https/

> +This requirement made its presence known after users made it
> +clear that an earlier
> +<a href="http://lwn.net/Articles/107930/">real-time patch</a>

s/http/https/

> +did not meet their needs, in conjunction with some
> +<a href="https://lkml.org/lkml/2005/3/17/199">RCU issues</a>

lkml.kernel.org

> +<p>
> +The
> +<a href="http://lwn.net/Articles/609973/#RCU Per-Flavor API Table">RCU-bh API</a>

s/http/https/ (and the same for all later lwn links in the document)

- Josh Triplett

^ permalink raw reply	[flat|nested] 18+ messages in thread

* Re: [PATCH tip/core/rcu 1/8] documentation: Record RCU requirements
  2015-12-05  0:34   ` Josh Triplett
@ 2015-12-05  1:54     ` Paul E. McKenney
  2015-12-05 20:19       ` Paul E. McKenney
  0 siblings, 1 reply; 18+ messages in thread
From: Paul E. McKenney @ 2015-12-05  1:54 UTC (permalink / raw)
  To: Josh Triplett
  Cc: linux-kernel, mingo, jiangshanlai, dipankar, akpm,
	mathieu.desnoyers, tglx, peterz, rostedt, dhowells, edumazet,
	dvhart, fweisbec, oleg, bobby.prani

On Fri, Dec 04, 2015 at 04:34:43PM -0800, Josh Triplett wrote:
> The content of the document seems fine; a few comments below on
> meta-issues.
> 
> On Fri, Dec 04, 2015 at 03:50:19PM -0800, Paul E. McKenney wrote:
> > --- /dev/null
> > +++ b/Documentation/RCU/Design/Requirements/Requirements.html
> > @@ -0,0 +1,2799 @@
> > +<!-- DO NOT HAND EDIT. -->
> > +<!-- Instead, edit Requirements.htmlx and run 'sh htmlqqz.sh Requirements' -->
> > +<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN"
> > +        "http://www.w3.org/TR/html4/loose.dtd">
> 
> Nit: these days, this should just be:
> <!doctype html>

Will making this change mean that https://validator.w3.org/ will
then require me to make a huge quantity of other changes?

> > +        <html>
> > +        <head><title>A Tour Through RCU's Requirements [LWN.net]</title>
> > +        <meta HTTP-EQUIV="Content-Type" CONTENT="text/html; charset=iso-8859-1">
> 
> Is there a good reason to not use charset=utf-8 here?

Beats me.  Cargo-culted that one.  ;-)

> > +
> > +<h1>A Tour Through RCU's Requirements</h1>
> > +
> > +<p>Copyright IBM Corporation, 2015</p>
> 
> If you're aiming for a properly formatted copyright notice, the year
> typically comes first, followed by the copyright holder.  That said,
> your corporate guidelines presumably have a specific format; is this
> that format?

Indeed it is.  Between you and IBM Legal, I unfortunately must follow
IBM Legal's advice.  ;-)

> > +<p>Author: Paul E.&nbsp;McKenney</p>
> > +<p><i>The initial version of this document appeared in the
> > +<a href="http://lwn.net/">LWN</a> articles
> > +<a href="http://lwn.net/Articles/652156/">here</a>,
> > +<a href="http://lwn.net/Articles/652677/">here</a>, and
> > +<a href="http://lwn.net/Articles/653326/">here</a>.</i></p>
> 
> s/http/https/g

Will change.

> > +<p>
> > +All that aside, here are the categories of currently known RCU requirements:
> > +</p>
> > +
> > +<ol>
> > +<li>	<a href="#Fundamental Requirements">
> 
> Anchors don't typically have spaces in them.  This may work in some
> browsers, but it doesn't validate.  You should either use %20 or
> (better) use a '-'.

It did when I validated it via https://validator.w3.org/.

Which is why I questioned your changes to the doctype directive.

Alternatively, what are you using to validate this?

> > +<p>
> > +This is followed by a <a href="#Summary">summary</a>,
> > +which is in turn followed by the inevitable
> > +<a href="#Answers to Quick Quizzes">answers to the quick quizzes</a>.
> 
> (Note: when editing anchors, don't forget to handle the target of this
> in the generation script.)

Good point!

> > +<p>
> > +This scenario resembles one of the first uses of RCU in
> > +<a href="http://en.wikipedia.org/wiki/DYNIX">DYNIX/ptx</a>,
> 
> s/http/https/

Will change globally.

> > +<p>
> > +However, this temptation must be resisted because there are a
> > +surprisingly large number of ways that the compiler
> > +(to say nothing of
> > +<a href="http://www.openvms.compaq.com/wizard/wiz_2637.html">DEC Alpha CPUs</a>)
> 
> This link sadly doesn't seem to work anymore; it redirects to HP's
> general page on OpenVMS, not a copy of that specific article.o
> Use this instead, assuming no current live version exists:
> https://web.archive.org/web/20120720095054/http://www.openvms.compaq.com/wizard/wiz_2637.html

Good catch!  Its new home is http://h71000.www7.hp.com/wizard/wiz_2637.html.

> > +<li>	It is also easy to forget to use <tt>rcu_assign_pointer()</tt>
> > +	and <tt>rcu_dereference()</tt>, perhaps (incorrectly)
> > +	substituting a simple assignment.
> > +	To catch this sort of error, a given RCU-protected pointer may be
> > +	tagged with <tt>__rcu</tt>, after which running sparse
> > +	with <tt>CONFIG_SPARSE_RCU_POINTER=y</tt> will complain
> > +	about simple-assignment accesses to that pointer.
> > +	Arnd Bergmann made me aware of this requirement, and also
> > +	supplied the needed
> > +	<a href="http://lwn.net/Articles/376011/">patch series</a>.
> 
> s/http/https/
> 
> > +<li>	Open-coded use of <tt>rcu_assign_pointer()</tt> and
> > +	<tt>rcu_dereference()</tt> to create typical linked
> > +	data structures can be surprisingly error-prone.
> > +	Therefore, RCU-protected
> > +	<a href="http://lwn.net/Articles/609973/#RCU List APIs">linked lists</a>
> 
> s/http/https/

Will fix these.

> > +<p>
> > +This all should be quite obvious, but the fact remains that
> > +Linus Torvalds recently had to
> > +<a href="http://marc.info/?l=linux-kernel&amp;m=142905739823385">remind</a>
> > +me of this requirement.
> 
> I'd suggest using the lkml.kernel.org redirector for this link, along
> with a Message-Id.
> 
> > +<p>
> > +The name notwithstanding, some Linux-kernel architectures
> > +can have nested NMIs, which RCU must handle correctly.
> > +Andy Lutomirski
> > +<a href="https://lkml.org/lkml/2014/11/21/642">surprised me</a>
> > +with this requirement;
> > +he also kindly surprised me with
> > +<a href="https://lkml.org/lkml/2014/11/22/1">an algorithm</a>
> > +that meets this requirement.
> 
> These links should both use lkml.kernel.org as well.  Doubly important
> because lkml.org is often down or has broken messages in its archive.

Good point, will look into finding the Message-IDs...

> > +<p>
> > +RCU therefore provides
> > +<tt><a href="http://lwn.net/Articles/217484/">rcu_barrier()</a></tt>,
> 
> s/http/https/
> 
> > +<p>
> > +This pair of mutual scheduler-RCU requirements came as a
> > +<a href="http://lwn.net/Articles/453002/">complete surprise</a>.
> 
> s/http/https/
> 
> > +This requirement made its presence known after users made it
> > +clear that an earlier
> > +<a href="http://lwn.net/Articles/107930/">real-time patch</a>
> 
> s/http/https/

Will fix these.

> > +did not meet their needs, in conjunction with some
> > +<a href="https://lkml.org/lkml/2005/3/17/199">RCU issues</a>
> 
> lkml.kernel.org

As above.

> > +<p>
> > +The
> > +<a href="http://lwn.net/Articles/609973/#RCU Per-Flavor API Table">RCU-bh API</a>
> 
> s/http/https/ (and the same for all later lwn links in the document)

Will fix.

							Thanx, Paul


^ permalink raw reply	[flat|nested] 18+ messages in thread

* Re: [PATCH tip/core/rcu 1/8] documentation: Record RCU requirements
  2015-12-05  1:54     ` Paul E. McKenney
@ 2015-12-05 20:19       ` Paul E. McKenney
  0 siblings, 0 replies; 18+ messages in thread
From: Paul E. McKenney @ 2015-12-05 20:19 UTC (permalink / raw)
  To: Josh Triplett
  Cc: linux-kernel, mingo, jiangshanlai, dipankar, akpm,
	mathieu.desnoyers, tglx, peterz, rostedt, dhowells, edumazet,
	dvhart, fweisbec, oleg, bobby.prani

On Fri, Dec 04, 2015 at 05:54:31PM -0800, Paul E. McKenney wrote:
> On Fri, Dec 04, 2015 at 04:34:43PM -0800, Josh Triplett wrote:
> > The content of the document seems fine; a few comments below on
> > meta-issues.
> > 
> > On Fri, Dec 04, 2015 at 03:50:19PM -0800, Paul E. McKenney wrote:
> > > --- /dev/null
> > > +++ b/Documentation/RCU/Design/Requirements/Requirements.html
> > > @@ -0,0 +1,2799 @@
> > > +<!-- DO NOT HAND EDIT. -->
> > > +<!-- Instead, edit Requirements.htmlx and run 'sh htmlqqz.sh Requirements' -->
> > > +<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN"
> > > +        "http://www.w3.org/TR/html4/loose.dtd">
> > 
> > Nit: these days, this should just be:
> > <!doctype html>
> 
> Will making this change mean that https://validator.w3.org/ will
> then require me to make a huge quantity of other changes?

I checked a prominent web site, and they are using the same doctype as
I am, so I am keeping this one.  However...

> > > +        <html>
> > > +        <head><title>A Tour Through RCU's Requirements [LWN.net]</title>
> > > +        <meta HTTP-EQUIV="Content-Type" CONTENT="text/html; charset=iso-8859-1">
> > 
> > Is there a good reason to not use charset=utf-8 here?
> 
> Beats me.  Cargo-culted that one.  ;-)

... they use utf-8, so I changed mine accordingly.

> > > +
> > > +<h1>A Tour Through RCU's Requirements</h1>
> > > +
> > > +<p>Copyright IBM Corporation, 2015</p>
> > 
> > If you're aiming for a properly formatted copyright notice, the year
> > typically comes first, followed by the copyright holder.  That said,
> > your corporate guidelines presumably have a specific format; is this
> > that format?
> 
> Indeed it is.  Between you and IBM Legal, I unfortunately must follow
> IBM Legal's advice.  ;-)
> 
> > > +<p>Author: Paul E.&nbsp;McKenney</p>
> > > +<p><i>The initial version of this document appeared in the
> > > +<a href="http://lwn.net/">LWN</a> articles
> > > +<a href="http://lwn.net/Articles/652156/">here</a>,
> > > +<a href="http://lwn.net/Articles/652677/">here</a>, and
> > > +<a href="http://lwn.net/Articles/653326/">here</a>.</i></p>
> > 
> > s/http/https/g
> 
> Will change.

And done, globally, at least for those whose webservers were willing to
put up with the change.  Which was the vast majority.

[ . . . ]

> > > +<p>
> > > +This all should be quite obvious, but the fact remains that
> > > +Linus Torvalds recently had to
> > > +<a href="http://marc.info/?l=linux-kernel&amp;m=142905739823385">remind</a>
> > > +me of this requirement.
> > 
> > I'd suggest using the lkml.kernel.org redirector for this link, along
> > with a Message-Id.
> > 
> > > +<p>
> > > +The name notwithstanding, some Linux-kernel architectures
> > > +can have nested NMIs, which RCU must handle correctly.
> > > +Andy Lutomirski
> > > +<a href="https://lkml.org/lkml/2014/11/21/642">surprised me</a>
> > > +with this requirement;
> > > +he also kindly surprised me with
> > > +<a href="https://lkml.org/lkml/2014/11/22/1">an algorithm</a>
> > > +that meets this requirement.
> > 
> > These links should both use lkml.kernel.org as well.  Doubly important
> > because lkml.org is often down or has broken messages in its archive.
> 
> Good point, will look into finding the Message-IDs...

This involves a second click for https://lkml.kernel.org/r/<Message-ID>
Trying https://lkml.kernel.org/g/<Message-ID>.  Which works nicely, so sold!

I will let you bug kernel.org about the fact that it translates the URL
to the http: form rather than the https: form...  But if it did that for
the /g/ form, it couldn't find the web page.  Never mind!

So all the ones that work have been converted.

							Thanx, Paul


^ permalink raw reply	[flat|nested] 18+ messages in thread

* [PATCH tip/core/rcu 2/8] Documentation: Record bottom-bit-zero guarantee for ->next
  2015-12-04 23:49 [PATCH tip/core/rcu 0/8] Documentation updates for 4.5 Paul E. McKenney
  2015-12-04 23:50 ` [PATCH tip/core/rcu 1/8] documentation: Record RCU requirements Paul E. McKenney
@ 2015-12-04 23:50 ` Paul E. McKenney
  2015-12-04 23:50 ` [PATCH tip/core/rcu 3/8] documentation: Cover requirements controlling stall warnings Paul E. McKenney
                   ` (6 subsequent siblings)
  8 siblings, 0 replies; 18+ messages in thread
From: Paul E. McKenney @ 2015-12-04 23:50 UTC (permalink / raw)
  To: linux-kernel
  Cc: mingo, jiangshanlai, dipankar, akpm, mathieu.desnoyers, josh,
	tglx, peterz, rostedt, dhowells, edumazet, dvhart, fweisbec, oleg,
	bobby.prani, Paul E. McKenney

This commit records RCU's guarantee that the bottom bit of the rcu_head
structure's ->next field will remain zero for callbacks posted via
call_rcu(), but not necessarily for <tt>kfree_rcu()</tt> or some
possible future call_rcu_lazy() variant that might one day be created
for energy-efficiency purposese.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
---
 .../RCU/Design/Requirements/Requirements.html      | 43 ++++++++++++++++++++++
 .../RCU/Design/Requirements/Requirements.htmlx     | 43 ++++++++++++++++++++++
 2 files changed, 86 insertions(+)

diff --git a/Documentation/RCU/Design/Requirements/Requirements.html b/Documentation/RCU/Design/Requirements/Requirements.html
index 95ef7b5de951..cda58d46f751 100644
--- a/Documentation/RCU/Design/Requirements/Requirements.html
+++ b/Documentation/RCU/Design/Requirements/Requirements.html
@@ -1678,6 +1678,7 @@ Some of the relevant points of interest are as follows:
 <li>	<a href="#Scheduler and RCU">Scheduler and RCU</a>.
 <li>	<a href="#Tracing and RCU">Tracing and RCU</a>.
 <li>	<a href="#Energy Efficiency">Energy Efficiency</a>.
+<li>	<a href="#Memory Efficiency">Memory Efficiency</a>.
 <li>	<a href="#Performance, Scalability, Response Time, and Reliability">
 	Performance, Scalability, Response Time, and Reliability</a>.
 </ol>
@@ -2006,6 +2007,48 @@ I learned of many of these requirements via angry phone calls:
 Flaming me on the Linux-kernel mailing list was apparently not
 sufficient to fully vent their ire at RCU's energy-efficiency bugs!
 
+<h3><a name="Memory Efficiency">Memory Efficiency</a></h3>
+
+<p>
+Although small-memory non-realtime systems can simply use Tiny RCU,
+code size is only one aspect of memory efficiency.
+Another aspect is the size of the <tt>rcu_head</tt> structure
+used by <tt>call_rcu()</tt> and <tt>kfree_rcu()</tt>.
+Although this structure contains nothing more than a pair of pointers,
+it does appear in many RCU-protected data structures, including
+some that are size critical.
+The <tt>page</tt> structure is a case in point, as evidenced by
+the many occurrences of the <tt>union</tt> keyword within that structure.
+
+<p>
+This need for memory efficiency is one reason that RCU uses hand-crafted
+singly linked lists to track the <tt>rcu_head</tt> structures that
+are waiting for a grace period to elapse.
+It is also the reason why <tt>rcu_head</tt> structures do not contain
+debug information, such as fields tracking the file and line of the
+<tt>call_rcu()</tt> or <tt>kfree_rcu()</tt> that posted them.
+Although this information might appear in debug-only kernel builds at some
+point, in the meantime, the <tt>-&gt;func</tt> field will often provide
+the needed debug information.
+
+<p>
+However, in some cases, the need for memory efficiency leads to even
+more extreme measures.
+Returning to the <tt>page</tt> structure, the <tt>rcu_head</tt> field
+shares storage with a great many other structures that are used at
+various points in the corresponding page's lifetime.
+In order to correctly resolve certain
+<a href="https://lkml.org/lkml/2015/8/19/137">race conditions</a>,
+the Linux kernel's memory-management subsystem needs a particular bit
+to remain zero during all phases of grace-period processing,
+and that bit happens to map to the bottom bit of the
+<tt>rcu_head</tt> structure's <tt>-&gt;next</tt> field.
+RCU makes this guarantee as long as <tt>call_rcu()</tt>
+is used to post the callback, as opposed to <tt>kfree_rcu()</tt>
+or some future &ldquo;lazy&rdquo;
+variant of <tt>call_rcu()</tt> that might one day be created for
+energy-efficiency purposes.
+
 <h3><a name="Performance, Scalability, Response Time, and Reliability">
 Performance, Scalability, Response Time, and Reliability</a></h3>
 
diff --git a/Documentation/RCU/Design/Requirements/Requirements.htmlx b/Documentation/RCU/Design/Requirements/Requirements.htmlx
index c5bb5bddd7ee..859a39534286 100644
--- a/Documentation/RCU/Design/Requirements/Requirements.htmlx
+++ b/Documentation/RCU/Design/Requirements/Requirements.htmlx
@@ -1837,6 +1837,7 @@ Some of the relevant points of interest are as follows:
 <li>	<a href="#Scheduler and RCU">Scheduler and RCU</a>.
 <li>	<a href="#Tracing and RCU">Tracing and RCU</a>.
 <li>	<a href="#Energy Efficiency">Energy Efficiency</a>.
+<li>	<a href="#Memory Efficiency">Memory Efficiency</a>.
 <li>	<a href="#Performance, Scalability, Response Time, and Reliability">
 	Performance, Scalability, Response Time, and Reliability</a>.
 </ol>
@@ -2173,6 +2174,48 @@ I learned of many of these requirements via angry phone calls:
 Flaming me on the Linux-kernel mailing list was apparently not
 sufficient to fully vent their ire at RCU's energy-efficiency bugs!
 
+<h3><a name="Memory Efficiency">Memory Efficiency</a></h3>
+
+<p>
+Although small-memory non-realtime systems can simply use Tiny RCU,
+code size is only one aspect of memory efficiency.
+Another aspect is the size of the <tt>rcu_head</tt> structure
+used by <tt>call_rcu()</tt> and <tt>kfree_rcu()</tt>.
+Although this structure contains nothing more than a pair of pointers,
+it does appear in many RCU-protected data structures, including
+some that are size critical.
+The <tt>page</tt> structure is a case in point, as evidenced by
+the many occurrences of the <tt>union</tt> keyword within that structure.
+
+<p>
+This need for memory efficiency is one reason that RCU uses hand-crafted
+singly linked lists to track the <tt>rcu_head</tt> structures that
+are waiting for a grace period to elapse.
+It is also the reason why <tt>rcu_head</tt> structures do not contain
+debug information, such as fields tracking the file and line of the
+<tt>call_rcu()</tt> or <tt>kfree_rcu()</tt> that posted them.
+Although this information might appear in debug-only kernel builds at some
+point, in the meantime, the <tt>-&gt;func</tt> field will often provide
+the needed debug information.
+
+<p>
+However, in some cases, the need for memory efficiency leads to even
+more extreme measures.
+Returning to the <tt>page</tt> structure, the <tt>rcu_head</tt> field
+shares storage with a great many other structures that are used at
+various points in the corresponding page's lifetime.
+In order to correctly resolve certain
+<a href="https://lkml.org/lkml/2015/8/19/137">race conditions</a>,
+the Linux kernel's memory-management subsystem needs a particular bit
+to remain zero during all phases of grace-period processing,
+and that bit happens to map to the bottom bit of the
+<tt>rcu_head</tt> structure's <tt>-&gt;next</tt> field.
+RCU makes this guarantee as long as <tt>call_rcu()</tt>
+is used to post the callback, as opposed to <tt>kfree_rcu()</tt>
+or some future &ldquo;lazy&rdquo;
+variant of <tt>call_rcu()</tt> that might one day be created for
+energy-efficiency purposes.
+
 <h3><a name="Performance, Scalability, Response Time, and Reliability">
 Performance, Scalability, Response Time, and Reliability</a></h3>
 
-- 
2.5.2


^ permalink raw reply related	[flat|nested] 18+ messages in thread

* [PATCH tip/core/rcu 3/8] documentation: Cover requirements controlling stall warnings
  2015-12-04 23:49 [PATCH tip/core/rcu 0/8] Documentation updates for 4.5 Paul E. McKenney
  2015-12-04 23:50 ` [PATCH tip/core/rcu 1/8] documentation: Record RCU requirements Paul E. McKenney
  2015-12-04 23:50 ` [PATCH tip/core/rcu 2/8] Documentation: Record bottom-bit-zero guarantee for ->next Paul E. McKenney
@ 2015-12-04 23:50 ` Paul E. McKenney
  2015-12-04 23:50 ` [PATCH tip/core/rcu 4/8] documentation: Composability analogies Paul E. McKenney
                   ` (5 subsequent siblings)
  8 siblings, 0 replies; 18+ messages in thread
From: Paul E. McKenney @ 2015-12-04 23:50 UTC (permalink / raw)
  To: linux-kernel
  Cc: mingo, jiangshanlai, dipankar, akpm, mathieu.desnoyers, josh,
	tglx, peterz, rostedt, dhowells, edumazet, dvhart, fweisbec, oleg,
	bobby.prani, Paul E. McKenney

This commit adds verbiage on boot and sysfs parameters that can be
used to control RCU CPU stall warnings, both to change the timeout
and to suppress these warnings entirely.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
---
 .../RCU/Design/Requirements/Requirements.html      | 25 +++++++++++++++++++++-
 .../RCU/Design/Requirements/Requirements.htmlx     | 25 +++++++++++++++++++++-
 2 files changed, 48 insertions(+), 2 deletions(-)

diff --git a/Documentation/RCU/Design/Requirements/Requirements.html b/Documentation/RCU/Design/Requirements/Requirements.html
index cda58d46f751..12f9d1555db3 100644
--- a/Documentation/RCU/Design/Requirements/Requirements.html
+++ b/Documentation/RCU/Design/Requirements/Requirements.html
@@ -1618,12 +1618,35 @@ guard against mishaps and misuse:
 	supplied the needed
 	<a href="https://lkml.org/lkml/2010/3/18/417">patch</a>.
 <li>	An infinite loop in an RCU read-side critical section will
-	eventually trigger an RCU CPU stall warning splat.
+	eventually trigger an RCU CPU stall warning splat, with
+	the duration of &ldquo;eventually&rdquo; being controlled by the
+	<tt>RCU_CPU_STALL_TIMEOUT</tt> <tt>Kconfig</tt> option, or,
+	alternatively, by the
+	<tt>rcupdate.rcu_cpu_stall_timeout</tt> boot/sysfs
+	parameter.
 	However, RCU is not obligated to produce this splat
 	unless there is a grace period waiting on that particular
 	RCU read-side critical section.
+	<p>
+	Some extreme workloads might intentionally delay
+	RCU grace periods, and systems running those workloads can
+	be booted with <tt>rcupdate.rcu_cpu_stall_suppress</tt>
+	to suppress the splats.
+	This kernel parameter may also be set via <tt>sysfs</tt>.
+	Furthermore, RCU CPU stall warnings are counter-productive
+	during sysrq dumps and during panics.
+	RCU therefore supplies the <tt>rcu_sysrq_start()</tt> and
+	<tt>rcu_sysrq_end()</tt> API members to be called before
+	and after long sysrq dumps.
+	RCU also supplies the <tt>rcu_panic()</tt> notifier that is
+	automatically invoked at the beginning of a panic to suppress
+	further RCU CPU stall warnings.
+
+	<p>
 	This requirement made itself known in the early 1990s, pretty
 	much the first time that it was necessary to debug a CPU stall.
+	That said, the initial implementation in DYNIX/ptx was quite
+	generic in comparison with that of Linux.
 <li>	Although it would be very good to detect pointers leaking out
 	of RCU read-side critical sections, there is currently no
 	good way of doing this.
diff --git a/Documentation/RCU/Design/Requirements/Requirements.htmlx b/Documentation/RCU/Design/Requirements/Requirements.htmlx
index 859a39534286..20bc84557f05 100644
--- a/Documentation/RCU/Design/Requirements/Requirements.htmlx
+++ b/Documentation/RCU/Design/Requirements/Requirements.htmlx
@@ -1777,12 +1777,35 @@ guard against mishaps and misuse:
 	supplied the needed
 	<a href="https://lkml.org/lkml/2010/3/18/417">patch</a>.
 <li>	An infinite loop in an RCU read-side critical section will
-	eventually trigger an RCU CPU stall warning splat.
+	eventually trigger an RCU CPU stall warning splat, with
+	the duration of &ldquo;eventually&rdquo; being controlled by the
+	<tt>RCU_CPU_STALL_TIMEOUT</tt> <tt>Kconfig</tt> option, or,
+	alternatively, by the
+	<tt>rcupdate.rcu_cpu_stall_timeout</tt> boot/sysfs
+	parameter.
 	However, RCU is not obligated to produce this splat
 	unless there is a grace period waiting on that particular
 	RCU read-side critical section.
+	<p>
+	Some extreme workloads might intentionally delay
+	RCU grace periods, and systems running those workloads can
+	be booted with <tt>rcupdate.rcu_cpu_stall_suppress</tt>
+	to suppress the splats.
+	This kernel parameter may also be set via <tt>sysfs</tt>.
+	Furthermore, RCU CPU stall warnings are counter-productive
+	during sysrq dumps and during panics.
+	RCU therefore supplies the <tt>rcu_sysrq_start()</tt> and
+	<tt>rcu_sysrq_end()</tt> API members to be called before
+	and after long sysrq dumps.
+	RCU also supplies the <tt>rcu_panic()</tt> notifier that is
+	automatically invoked at the beginning of a panic to suppress
+	further RCU CPU stall warnings.
+
+	<p>
 	This requirement made itself known in the early 1990s, pretty
 	much the first time that it was necessary to debug a CPU stall.
+	That said, the initial implementation in DYNIX/ptx was quite
+	generic in comparison with that of Linux.
 <li>	Although it would be very good to detect pointers leaking out
 	of RCU read-side critical sections, there is currently no
 	good way of doing this.
-- 
2.5.2


^ permalink raw reply related	[flat|nested] 18+ messages in thread

* [PATCH tip/core/rcu 4/8] documentation: Composability analogies
  2015-12-04 23:49 [PATCH tip/core/rcu 0/8] Documentation updates for 4.5 Paul E. McKenney
                   ` (2 preceding siblings ...)
  2015-12-04 23:50 ` [PATCH tip/core/rcu 3/8] documentation: Cover requirements controlling stall warnings Paul E. McKenney
@ 2015-12-04 23:50 ` Paul E. McKenney
  2015-12-04 23:50 ` [PATCH tip/core/rcu 5/8] documentation: Expand on scheduler/RCU deadlock requirements Paul E. McKenney
                   ` (4 subsequent siblings)
  8 siblings, 0 replies; 18+ messages in thread
From: Paul E. McKenney @ 2015-12-04 23:50 UTC (permalink / raw)
  To: linux-kernel
  Cc: mingo, jiangshanlai, dipankar, akpm, mathieu.desnoyers, josh,
	tglx, peterz, rostedt, dhowells, edumazet, dvhart, fweisbec, oleg,
	bobby.prani, Paul E. McKenney

This commit expands on RCU's composability by comparing it to that of
transactional memory and of locking.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
---
 Documentation/RCU/Design/Requirements/Requirements.html  | 8 ++++++++
 Documentation/RCU/Design/Requirements/Requirements.htmlx | 8 ++++++++
 2 files changed, 16 insertions(+)

diff --git a/Documentation/RCU/Design/Requirements/Requirements.html b/Documentation/RCU/Design/Requirements/Requirements.html
index 12f9d1555db3..c95679dff143 100644
--- a/Documentation/RCU/Design/Requirements/Requirements.html
+++ b/Documentation/RCU/Design/Requirements/Requirements.html
@@ -1495,6 +1495,14 @@ section, neither of which is conducive to a long-lived and prosperous
 kernel.
 
 <p>
+It is worth noting that RCU is not alone in limiting composability.
+For example, many transactional-memory implementations prohibit
+composing a pair of transactions separated by an irrevocable
+operation (for example, a network receive operation).
+For another example, lock-based critical sections can be composed
+surprisingly freely, but only if deadlock is avoided.
+
+<p>
 In short, although RCU read-side critical sections are highly composable,
 care is required in some situations, just as is the case for any other
 composable synchronization mechanism.
diff --git a/Documentation/RCU/Design/Requirements/Requirements.htmlx b/Documentation/RCU/Design/Requirements/Requirements.htmlx
index 20bc84557f05..b0b88f1f765c 100644
--- a/Documentation/RCU/Design/Requirements/Requirements.htmlx
+++ b/Documentation/RCU/Design/Requirements/Requirements.htmlx
@@ -1654,6 +1654,14 @@ section, neither of which is conducive to a long-lived and prosperous
 kernel.
 
 <p>
+It is worth noting that RCU is not alone in limiting composability.
+For example, many transactional-memory implementations prohibit
+composing a pair of transactions separated by an irrevocable
+operation (for example, a network receive operation).
+For another example, lock-based critical sections can be composed
+surprisingly freely, but only if deadlock is avoided.
+
+<p>
 In short, although RCU read-side critical sections are highly composable,
 care is required in some situations, just as is the case for any other
 composable synchronization mechanism.
-- 
2.5.2


^ permalink raw reply related	[flat|nested] 18+ messages in thread

* [PATCH tip/core/rcu 5/8] documentation: Expand on scheduler/RCU deadlock requirements
  2015-12-04 23:49 [PATCH tip/core/rcu 0/8] Documentation updates for 4.5 Paul E. McKenney
                   ` (3 preceding siblings ...)
  2015-12-04 23:50 ` [PATCH tip/core/rcu 4/8] documentation: Composability analogies Paul E. McKenney
@ 2015-12-04 23:50 ` Paul E. McKenney
  2015-12-04 23:50 ` [PATCH tip/core/rcu 6/8] documentation: Clarify RCU memory barriers and requirements Paul E. McKenney
                   ` (3 subsequent siblings)
  8 siblings, 0 replies; 18+ messages in thread
From: Paul E. McKenney @ 2015-12-04 23:50 UTC (permalink / raw)
  To: linux-kernel
  Cc: mingo, jiangshanlai, dipankar, akpm, mathieu.desnoyers, josh,
	tglx, peterz, rostedt, dhowells, edumazet, dvhart, fweisbec, oleg,
	bobby.prani, Paul E. McKenney

This commit adds a second option for avoiding scheduler/RCU deadlocks,
namely that preemption be disabled across the entire RCU read-side
critical section in question.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
---
 Documentation/RCU/Design/Requirements/Requirements.html  | 14 +++++++++-----
 Documentation/RCU/Design/Requirements/Requirements.htmlx | 14 +++++++++-----
 2 files changed, 18 insertions(+), 10 deletions(-)

diff --git a/Documentation/RCU/Design/Requirements/Requirements.html b/Documentation/RCU/Design/Requirements/Requirements.html
index c95679dff143..e566e78b5eb1 100644
--- a/Documentation/RCU/Design/Requirements/Requirements.html
+++ b/Documentation/RCU/Design/Requirements/Requirements.html
@@ -1942,12 +1942,16 @@ RCU depends on the scheduler, and the scheduler uses RCU to
 protect some of its data structures.
 This means the scheduler is forbidden from acquiring
 the runqueue locks and the priority-inheritance locks
-in the middle of an outermost RCU read-side critical section unless
-it also releases them before exiting that same
-RCU read-side critical section.
-This same prohibition also applies to any lock that is acquired
+in the middle of an outermost RCU read-side critical section unless either
+(1)&nbsp;it releases them before exiting that same
+RCU read-side critical section, or
+(2)&nbsp;preemption is disabled across
+that entire RCU read-side critical section.
+This same prohibition also applies (recursively!) to any lock that is acquired
 while holding any lock to which this prohibition applies.
-Violating this rule results in deadlock.
+Adhering to this rule prevents preemptible RCU from invoking
+<tt>rcu_read_unlock_special()</tt> while either runqueue or
+priority-inheritance locks are held, thus avoiding deadlock.
 
 <p>
 For RCU's part, the preemptible-RCU <tt>rcu_read_unlock()</tt>
diff --git a/Documentation/RCU/Design/Requirements/Requirements.htmlx b/Documentation/RCU/Design/Requirements/Requirements.htmlx
index b0b88f1f765c..09d9539eb61a 100644
--- a/Documentation/RCU/Design/Requirements/Requirements.htmlx
+++ b/Documentation/RCU/Design/Requirements/Requirements.htmlx
@@ -2109,12 +2109,16 @@ RCU depends on the scheduler, and the scheduler uses RCU to
 protect some of its data structures.
 This means the scheduler is forbidden from acquiring
 the runqueue locks and the priority-inheritance locks
-in the middle of an outermost RCU read-side critical section unless
-it also releases them before exiting that same
-RCU read-side critical section.
-This same prohibition also applies to any lock that is acquired
+in the middle of an outermost RCU read-side critical section unless either
+(1)&nbsp;it releases them before exiting that same
+RCU read-side critical section, or
+(2)&nbsp;preemption is disabled across
+that entire RCU read-side critical section.
+This same prohibition also applies (recursively!) to any lock that is acquired
 while holding any lock to which this prohibition applies.
-Violating this rule results in deadlock.
+Adhering to this rule prevents preemptible RCU from invoking
+<tt>rcu_read_unlock_special()</tt> while either runqueue or
+priority-inheritance locks are held, thus avoiding deadlock.
 
 <p>
 For RCU's part, the preemptible-RCU <tt>rcu_read_unlock()</tt>
-- 
2.5.2


^ permalink raw reply related	[flat|nested] 18+ messages in thread

* [PATCH tip/core/rcu 6/8] documentation: Clarify RCU memory barriers and requirements
  2015-12-04 23:49 [PATCH tip/core/rcu 0/8] Documentation updates for 4.5 Paul E. McKenney
                   ` (4 preceding siblings ...)
  2015-12-04 23:50 ` [PATCH tip/core/rcu 5/8] documentation: Expand on scheduler/RCU deadlock requirements Paul E. McKenney
@ 2015-12-04 23:50 ` Paul E. McKenney
  2015-12-04 23:50 ` [PATCH tip/core/rcu 7/8] documentation: Update RCU requirements based on expedited changes Paul E. McKenney
                   ` (2 subsequent siblings)
  8 siblings, 0 replies; 18+ messages in thread
From: Paul E. McKenney @ 2015-12-04 23:50 UTC (permalink / raw)
  To: linux-kernel
  Cc: mingo, jiangshanlai, dipankar, akpm, mathieu.desnoyers, josh,
	tglx, peterz, rostedt, dhowells, edumazet, dvhart, fweisbec, oleg,
	bobby.prani, Paul E. McKenney

The RCU requirements do not make it absolutely clear that the
memory-barrier requirements are not intended to replace the fundamental
requirement that all pre-existing RCU readers complete before a grace
period completes.  This commit therefore pulls the memory-barrier
requirements into a separate section and explicitly calls out the
relationship between the memory-barrier requirements and the fundamental
requirement.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
---
 .../RCU/Design/Requirements/Requirements.html      | 66 +++++++++++++---------
 .../RCU/Design/Requirements/Requirements.htmlx     | 66 +++++++++++++---------
 2 files changed, 78 insertions(+), 54 deletions(-)

diff --git a/Documentation/RCU/Design/Requirements/Requirements.html b/Documentation/RCU/Design/Requirements/Requirements.html
index e566e78b5eb1..f5bf27228192 100644
--- a/Documentation/RCU/Design/Requirements/Requirements.html
+++ b/Documentation/RCU/Design/Requirements/Requirements.html
@@ -80,6 +80,8 @@ These are:
 	Grace-Period Guarantee</a>
 <li>	<a href="#Publish-Subscribe Guarantee">
 	Publish-Subscribe Guarantee</a>
+<li>	<a href="#Memory-Barrier Guarantees">
+	Memory-Barrier Guarantees</a>
 <li>	<a href="#RCU Primitives Guaranteed to Execute Unconditionally">
 	RCU Primitives Guaranteed to Execute Unconditionally</a>
 <li>	<a href="#Guaranteed Read-to-Write Upgrade">
@@ -499,9 +501,37 @@ might the compiler make use of?
 <br><a href="#qq4answer">Answer</a>
 
 <p>
-This simple linked-data-structure scenario clearly demonstrates the need
-for RCU's stringent memory-ordering guarantees on systems with more than
-one CPU:
+In short, RCU's publish-subscribe guarantee is provided by the combination
+of <tt>rcu_assign_pointer()</tt> and <tt>rcu_dereference()</tt>.
+This guarantee allows data elements to be safely added to RCU-protected
+linked data structures without disrupting RCU readers.
+This guarantee can be used in combination with the grace-period
+guarantee to also allow data elements to be removed from RCU-protected
+linked data structures, again without disrupting RCU readers.
+
+<p>
+This guarantee was only partially premeditated.
+DYNIX/ptx used an explicit memory barrier for publication, but had nothing
+resembling <tt>rcu_dereference()</tt> for subscription, nor did it
+have anything resembling the <tt>smp_read_barrier_depends()</tt>
+that was later subsumed into <tt>rcu_dereference()</tt>.
+The need for these operations made itself known quite suddenly at a
+late-1990s meeting with the DEC Alpha architects, back in the days when
+DEC was still a free-standing company.
+It took the Alpha architects a good hour to convince me that any sort
+of barrier would ever be needed, and it then took me a good <i>two</i> hours
+to convince them that their documentation did not make this point clear.
+More recent work with the C and C++ standards committees have provided
+much education on tricks and traps from the compiler.
+In short, compilers were much less tricky in the early 1990s, but in
+2015, don't even think about omitting <tt>rcu_dereference()</tt>!
+
+<h3><a name="Memory-Barrier Guarantees">Memory-Barrier Guarantees</a></h3>
+
+<p>
+The previous section's simple linked-data-structure scenario clearly
+demonstrates the need for RCU's stringent memory-ordering guarantees on
+systems with more than one CPU:
 
 <ol>
 <li>	Each CPU that has an RCU read-side critical section that
@@ -554,30 +584,12 @@ Are all these memory barriers <i> really</i> required?
 <br><a href="#qq6answer">Answer</a>
 
 <p>
-In short, RCU's publish-subscribe guarantee is provided by the combination
-of <tt>rcu_assign_pointer()</tt> and <tt>rcu_dereference()</tt>.
-This guarantee allows data elements to be safely added to RCU-protected
-linked data structures without disrupting RCU readers.
-This guarantee can be used in combination with the grace-period
-guarantee to also allow data elements to be removed from RCU-protected
-linked data structures, again without disrupting RCU readers.
-
-<p>
-This guarantee was only partially premeditated.
-DYNIX/ptx used an explicit memory barrier for publication, but had nothing
-resembling <tt>rcu_dereference()</tt> for subscription, nor did it
-have anything resembling the <tt>smp_read_barrier_depends()</tt>
-that was later subsumed into <tt>rcu_dereference()</tt>.
-The need for these operations made itself known quite suddenly at a
-late-1990s meeting with the DEC Alpha architects, back in the days when
-DEC was still a free-standing company.
-It took the Alpha architects a good hour to convince me that any sort
-of barrier would ever be needed, and it then took me a good <i>two</i> hours
-to convince them that their documentation did not make this point clear.
-More recent work with the C and C++ standards committees have provided
-much education on tricks and traps from the compiler.
-In short, compilers were much less tricky in the early 1990s, but in
-2015, don't even think about omitting <tt>rcu_dereference()</tt>!
+Note that these memory-barrier requirements do not replace the fundamental
+RCU requirement that a grace period wait for all pre-existing readers.
+On the contrary, the memory barriers called out in this section must operate in
+such a way as to <i>enforce</i> this fundamental requirement.
+Of course, different implementations enforce this requirement in different
+ways, but enforce it they must.
 
 <h3><a name="RCU Primitives Guaranteed to Execute Unconditionally">RCU Primitives Guaranteed to Execute Unconditionally</a></h3>
 
diff --git a/Documentation/RCU/Design/Requirements/Requirements.htmlx b/Documentation/RCU/Design/Requirements/Requirements.htmlx
index 09d9539eb61a..4082e588baa7 100644
--- a/Documentation/RCU/Design/Requirements/Requirements.htmlx
+++ b/Documentation/RCU/Design/Requirements/Requirements.htmlx
@@ -78,6 +78,8 @@ These are:
 	Grace-Period Guarantee</a>
 <li>	<a href="#Publish-Subscribe Guarantee">
 	Publish-Subscribe Guarantee</a>
+<li>	<a href="#Memory-Barrier Guarantees">
+	Memory-Barrier Guarantees</a>
 <li>	<a href="#RCU Primitives Guaranteed to Execute Unconditionally">
 	RCU Primitives Guaranteed to Execute Unconditionally</a>
 <li>	<a href="#Guaranteed Read-to-Write Upgrade">
@@ -539,9 +541,37 @@ either <tt>rcu_access_pointer()</tt> or <tt>rcu_dereference()</tt>.
 <p>@@QQE@@
 
 <p>
-This simple linked-data-structure scenario clearly demonstrates the need
-for RCU's stringent memory-ordering guarantees on systems with more than
-one CPU:
+In short, RCU's publish-subscribe guarantee is provided by the combination
+of <tt>rcu_assign_pointer()</tt> and <tt>rcu_dereference()</tt>.
+This guarantee allows data elements to be safely added to RCU-protected
+linked data structures without disrupting RCU readers.
+This guarantee can be used in combination with the grace-period
+guarantee to also allow data elements to be removed from RCU-protected
+linked data structures, again without disrupting RCU readers.
+
+<p>
+This guarantee was only partially premeditated.
+DYNIX/ptx used an explicit memory barrier for publication, but had nothing
+resembling <tt>rcu_dereference()</tt> for subscription, nor did it
+have anything resembling the <tt>smp_read_barrier_depends()</tt>
+that was later subsumed into <tt>rcu_dereference()</tt>.
+The need for these operations made itself known quite suddenly at a
+late-1990s meeting with the DEC Alpha architects, back in the days when
+DEC was still a free-standing company.
+It took the Alpha architects a good hour to convince me that any sort
+of barrier would ever be needed, and it then took me a good <i>two</i> hours
+to convince them that their documentation did not make this point clear.
+More recent work with the C and C++ standards committees have provided
+much education on tricks and traps from the compiler.
+In short, compilers were much less tricky in the early 1990s, but in
+2015, don't even think about omitting <tt>rcu_dereference()</tt>!
+
+<h3><a name="Memory-Barrier Guarantees">Memory-Barrier Guarantees</a></h3>
+
+<p>
+The previous section's simple linked-data-structure scenario clearly
+demonstrates the need for RCU's stringent memory-ordering guarantees on
+systems with more than one CPU:
 
 <ol>
 <li>	Each CPU that has an RCU read-side critical section that
@@ -653,30 +683,12 @@ adhered to the as-if rule than it is to actually adhere to it!
 <p>@@QQE@@
 
 <p>
-In short, RCU's publish-subscribe guarantee is provided by the combination
-of <tt>rcu_assign_pointer()</tt> and <tt>rcu_dereference()</tt>.
-This guarantee allows data elements to be safely added to RCU-protected
-linked data structures without disrupting RCU readers.
-This guarantee can be used in combination with the grace-period
-guarantee to also allow data elements to be removed from RCU-protected
-linked data structures, again without disrupting RCU readers.
-
-<p>
-This guarantee was only partially premeditated.
-DYNIX/ptx used an explicit memory barrier for publication, but had nothing
-resembling <tt>rcu_dereference()</tt> for subscription, nor did it
-have anything resembling the <tt>smp_read_barrier_depends()</tt>
-that was later subsumed into <tt>rcu_dereference()</tt>.
-The need for these operations made itself known quite suddenly at a
-late-1990s meeting with the DEC Alpha architects, back in the days when
-DEC was still a free-standing company.
-It took the Alpha architects a good hour to convince me that any sort
-of barrier would ever be needed, and it then took me a good <i>two</i> hours
-to convince them that their documentation did not make this point clear.
-More recent work with the C and C++ standards committees have provided
-much education on tricks and traps from the compiler.
-In short, compilers were much less tricky in the early 1990s, but in
-2015, don't even think about omitting <tt>rcu_dereference()</tt>!
+Note that these memory-barrier requirements do not replace the fundamental
+RCU requirement that a grace period wait for all pre-existing readers.
+On the contrary, the memory barriers called out in this section must operate in
+such a way as to <i>enforce</i> this fundamental requirement.
+Of course, different implementations enforce this requirement in different
+ways, but enforce it they must.
 
 <h3><a name="RCU Primitives Guaranteed to Execute Unconditionally">RCU Primitives Guaranteed to Execute Unconditionally</a></h3>
 
-- 
2.5.2


^ permalink raw reply related	[flat|nested] 18+ messages in thread

* [PATCH tip/core/rcu 7/8] documentation: Update RCU requirements based on expedited changes
  2015-12-04 23:49 [PATCH tip/core/rcu 0/8] Documentation updates for 4.5 Paul E. McKenney
                   ` (5 preceding siblings ...)
  2015-12-04 23:50 ` [PATCH tip/core/rcu 6/8] documentation: Clarify RCU memory barriers and requirements Paul E. McKenney
@ 2015-12-04 23:50 ` Paul E. McKenney
  2015-12-04 23:50 ` [PATCH tip/core/rcu 8/8] Documentation/memory-barriers.txt: Fix ACCESS_ONCE thinko Paul E. McKenney
  2015-12-05  0:42 ` [PATCH tip/core/rcu 0/8] Documentation updates for 4.5 Josh Triplett
  8 siblings, 0 replies; 18+ messages in thread
From: Paul E. McKenney @ 2015-12-04 23:50 UTC (permalink / raw)
  To: linux-kernel
  Cc: mingo, jiangshanlai, dipankar, akpm, mathieu.desnoyers, josh,
	tglx, peterz, rostedt, dhowells, edumazet, dvhart, fweisbec, oleg,
	bobby.prani, Paul E. McKenney

Because RCU-sched expedited grace periods now use IPIs and interact
with rcu_read_unlock(), it is no longer sufficient to disable preemption
across RCU read-side critical sections that acquire and hold scheduler
locks.  It is now necessary to instead disable interrupts.  This commit
documents this change.

Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
---
 Documentation/RCU/Design/Requirements/Requirements.html  | 12 ++++++++++--
 Documentation/RCU/Design/Requirements/Requirements.htmlx | 10 +++++++++-
 2 files changed, 19 insertions(+), 3 deletions(-)

diff --git a/Documentation/RCU/Design/Requirements/Requirements.html b/Documentation/RCU/Design/Requirements/Requirements.html
index f5bf27228192..9aa8f2c402f5 100644
--- a/Documentation/RCU/Design/Requirements/Requirements.html
+++ b/Documentation/RCU/Design/Requirements/Requirements.html
@@ -1,5 +1,5 @@
 <!-- DO NOT HAND EDIT. -->
-<!-- Instead, edit Requirements.htmlx and run 'sh htmlqqz.sh Requirements' -->
+<!-- Instead, edit Documentation/RCU/Design/Requirements/Requirements.htmlx and run 'sh htmlqqz.sh Documentation/RCU/Design/Requirements/Requirements' -->
 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN"
         "http://www.w3.org/TR/html4/loose.dtd">
         <html>
@@ -1957,7 +1957,7 @@ the runqueue locks and the priority-inheritance locks
 in the middle of an outermost RCU read-side critical section unless either
 (1)&nbsp;it releases them before exiting that same
 RCU read-side critical section, or
-(2)&nbsp;preemption is disabled across
+(2)&nbsp;interrupts are disabled across
 that entire RCU read-side critical section.
 This same prohibition also applies (recursively!) to any lock that is acquired
 while holding any lock to which this prohibition applies.
@@ -1966,6 +1966,14 @@ Adhering to this rule prevents preemptible RCU from invoking
 priority-inheritance locks are held, thus avoiding deadlock.
 
 <p>
+Prior to v4.4, it was only necessary to disable preemption across
+RCU read-side critical sections that acquired scheduler locks.
+In v4.4, expedited grace periods started using IPIs, and these
+IPIs could force a <tt>rcu_read_unlock()</tt> to take the slowpath.
+Therefore, this expedited-grace-period change required disabling of
+interrupts, not just preemption.
+
+<p>
 For RCU's part, the preemptible-RCU <tt>rcu_read_unlock()</tt>
 implementation must be written carefully to avoid similar deadlocks.
 In particular, <tt>rcu_read_unlock()</tt> must tolerate an
diff --git a/Documentation/RCU/Design/Requirements/Requirements.htmlx b/Documentation/RCU/Design/Requirements/Requirements.htmlx
index 4082e588baa7..de86311e638e 100644
--- a/Documentation/RCU/Design/Requirements/Requirements.htmlx
+++ b/Documentation/RCU/Design/Requirements/Requirements.htmlx
@@ -2124,7 +2124,7 @@ the runqueue locks and the priority-inheritance locks
 in the middle of an outermost RCU read-side critical section unless either
 (1)&nbsp;it releases them before exiting that same
 RCU read-side critical section, or
-(2)&nbsp;preemption is disabled across
+(2)&nbsp;interrupts are disabled across
 that entire RCU read-side critical section.
 This same prohibition also applies (recursively!) to any lock that is acquired
 while holding any lock to which this prohibition applies.
@@ -2133,6 +2133,14 @@ Adhering to this rule prevents preemptible RCU from invoking
 priority-inheritance locks are held, thus avoiding deadlock.
 
 <p>
+Prior to v4.4, it was only necessary to disable preemption across
+RCU read-side critical sections that acquired scheduler locks.
+In v4.4, expedited grace periods started using IPIs, and these
+IPIs could force a <tt>rcu_read_unlock()</tt> to take the slowpath.
+Therefore, this expedited-grace-period change required disabling of
+interrupts, not just preemption.
+
+<p>
 For RCU's part, the preemptible-RCU <tt>rcu_read_unlock()</tt>
 implementation must be written carefully to avoid similar deadlocks.
 In particular, <tt>rcu_read_unlock()</tt> must tolerate an
-- 
2.5.2


^ permalink raw reply related	[flat|nested] 18+ messages in thread

* [PATCH tip/core/rcu 8/8] Documentation/memory-barriers.txt: Fix ACCESS_ONCE thinko
  2015-12-04 23:49 [PATCH tip/core/rcu 0/8] Documentation updates for 4.5 Paul E. McKenney
                   ` (6 preceding siblings ...)
  2015-12-04 23:50 ` [PATCH tip/core/rcu 7/8] documentation: Update RCU requirements based on expedited changes Paul E. McKenney
@ 2015-12-04 23:50 ` Paul E. McKenney
  2015-12-05  0:42 ` [PATCH tip/core/rcu 0/8] Documentation updates for 4.5 Josh Triplett
  8 siblings, 0 replies; 18+ messages in thread
From: Paul E. McKenney @ 2015-12-04 23:50 UTC (permalink / raw)
  To: linux-kernel
  Cc: mingo, jiangshanlai, dipankar, akpm, mathieu.desnoyers, josh,
	tglx, peterz, rostedt, dhowells, edumazet, dvhart, fweisbec, oleg,
	bobby.prani, Chris Metcalf, Paul E. McKenney

From: Chris Metcalf <cmetcalf@ezchip.com>

In commit 2ecf810121c7 ("Documentation/memory-barriers.txt: Add
needed ACCESS_ONCE() calls to memory-barriers.txt") the statement
"Q = P" was converted to "ACCESS_ONCE(Q) = P".  This should have
been "Q = ACCESS_ONCE(P)".  It later became "WRITE_ONCE(Q, P)".
This doesn't match the following text, which is "Q = LOAD P".
Change the statement to be "Q = READ_ONCE(P)".

Signed-off-by: Chris Metcalf <cmetcalf@ezchip.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
---
 Documentation/memory-barriers.txt | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/Documentation/memory-barriers.txt b/Documentation/memory-barriers.txt
index aef9487303d0..85304ebd187c 100644
--- a/Documentation/memory-barriers.txt
+++ b/Documentation/memory-barriers.txt
@@ -194,7 +194,7 @@ There are some minimal guarantees that may be expected of a CPU:
  (*) On any given CPU, dependent memory accesses will be issued in order, with
      respect to itself.  This means that for:
 
-	WRITE_ONCE(Q, P); smp_read_barrier_depends(); D = READ_ONCE(*Q);
+	Q = READ_ONCE(P); smp_read_barrier_depends(); D = READ_ONCE(*Q);
 
      the CPU will issue the following memory operations:
 
@@ -202,9 +202,9 @@ There are some minimal guarantees that may be expected of a CPU:
 
      and always in that order.  On most systems, smp_read_barrier_depends()
      does nothing, but it is required for DEC Alpha.  The READ_ONCE()
-     and WRITE_ONCE() are required to prevent compiler mischief.  Please
-     note that you should normally use something like rcu_dereference()
-     instead of open-coding smp_read_barrier_depends().
+     is required to prevent compiler mischief.  Please note that you
+     should normally use something like rcu_dereference() instead of
+     open-coding smp_read_barrier_depends().
 
  (*) Overlapping loads and stores within a particular CPU will appear to be
      ordered within that CPU.  This means that for:
-- 
2.5.2


^ permalink raw reply related	[flat|nested] 18+ messages in thread

* Re: [PATCH tip/core/rcu 0/8] Documentation updates for 4.5
  2015-12-04 23:49 [PATCH tip/core/rcu 0/8] Documentation updates for 4.5 Paul E. McKenney
                   ` (7 preceding siblings ...)
  2015-12-04 23:50 ` [PATCH tip/core/rcu 8/8] Documentation/memory-barriers.txt: Fix ACCESS_ONCE thinko Paul E. McKenney
@ 2015-12-05  0:42 ` Josh Triplett
  8 siblings, 0 replies; 18+ messages in thread
From: Josh Triplett @ 2015-12-05  0:42 UTC (permalink / raw)
  To: Paul E. McKenney
  Cc: linux-kernel, mingo, jiangshanlai, dipankar, akpm,
	mathieu.desnoyers, tglx, peterz, rostedt, dhowells, edumazet,
	dvhart, fweisbec, oleg, bobby.prani

On Fri, Dec 04, 2015 at 03:49:52PM -0800, Paul E. McKenney wrote:
> Hello!
> 
> This series contains documentation updates, perhaps most notably a first
> installment of RCU design documentation, which first appeared on LWN
> (http://lwn.net/Articles/652156/, http://lwn.net/Articles/652677/, and
> http://lwn.net/Articles/653326/).  The individual patches are as follows:
> 
> 1.	Add RCU requirements design document.
> 
> 2-7.	Updates to the RCU requirements.
> 
> 8.	Fix memory-barrier.txt typo, courtesy of Chris Metcalf.

I sent two mails worth of feedback to patch 1 that effectively apply to
the whole series; one covers the checkin of a generated file, and the
other covers various mechanical issues in the document.  With those two
fixed:

Reviewed-by: Josh Triplett <josh@joshtriplett.org>
for the whole series

^ permalink raw reply	[flat|nested] 18+ messages in thread

end of thread, other threads:[~2015-12-05 21:18 UTC | newest]

Thread overview: 18+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2015-12-04 23:49 [PATCH tip/core/rcu 0/8] Documentation updates for 4.5 Paul E. McKenney
2015-12-04 23:50 ` [PATCH tip/core/rcu 1/8] documentation: Record RCU requirements Paul E. McKenney
2015-12-05  0:07   ` Josh Triplett
2015-12-05  0:33     ` Paul E. McKenney
2015-12-05  0:38       ` Josh Triplett
2015-12-05  1:56         ` Paul E. McKenney
2015-12-05 21:19           ` Paul E. McKenney
2015-12-05  0:34   ` Josh Triplett
2015-12-05  1:54     ` Paul E. McKenney
2015-12-05 20:19       ` Paul E. McKenney
2015-12-04 23:50 ` [PATCH tip/core/rcu 2/8] Documentation: Record bottom-bit-zero guarantee for ->next Paul E. McKenney
2015-12-04 23:50 ` [PATCH tip/core/rcu 3/8] documentation: Cover requirements controlling stall warnings Paul E. McKenney
2015-12-04 23:50 ` [PATCH tip/core/rcu 4/8] documentation: Composability analogies Paul E. McKenney
2015-12-04 23:50 ` [PATCH tip/core/rcu 5/8] documentation: Expand on scheduler/RCU deadlock requirements Paul E. McKenney
2015-12-04 23:50 ` [PATCH tip/core/rcu 6/8] documentation: Clarify RCU memory barriers and requirements Paul E. McKenney
2015-12-04 23:50 ` [PATCH tip/core/rcu 7/8] documentation: Update RCU requirements based on expedited changes Paul E. McKenney
2015-12-04 23:50 ` [PATCH tip/core/rcu 8/8] Documentation/memory-barriers.txt: Fix ACCESS_ONCE thinko Paul E. McKenney
2015-12-05  0:42 ` [PATCH tip/core/rcu 0/8] Documentation updates for 4.5 Josh Triplett

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.