vhpt_miss handler question?

vhpt_miss handler question?

[Mario Smarduch]

Some work I'm doing has taken me to the vhpt_miss
handler. I've been looking at  the following snippet just
wondering what happens in the case where the present
bit is 0 and p10,p11 predicates are not set anotherwords
they retain their application values. Would this not
cause an errant insertion if either p10 or p11 were
true in the application prior to the VHPT handler?

- Mario.


(p7)    tbit.z p6,p7=r18,_PAGE_P_BIT            // page present bit
cleared?
        mov r22=cr.iha                          // get the VHPT address
that cau
sed the TLB miss
        ;;                                      // avoid RAW on p7
(p7)    tbit.nz.unc p10,p11=r19,32              // is it an instruction
TLB miss
?
        dep r23=0,r20,0,PAGE_SHIFT              // clear low bits to get
page ad
dress
        ;;
(p10)   itc.i r18                               // insert the
instruction TLB en
try
(p11)   itc.d r18                               // insert the data TLB
entry

Re: vhpt_miss handler question?

[Keith Owens]

On Fri, 26 Sep 2003 09:43:55 -0500, 
Mario Smarduch <cms063@email.mot.com> wrote:
>Some work I'm doing has taken me to the vhpt_miss
>handler. I've been looking at  the following snippet just
>wondering what happens in the case where the present
>bit is 0 and p10,p11 predicates are not set anotherwords
>they retain their application values. Would this not
>cause an errant insertion if either p10 or p11 were
>true in the application prior to the VHPT handler?

>(p7)    tbit.nz.unc p10,p11=r19,32              // is it an instruction

tbit.xx.unc sets both predicates to 0 if the controlling predicate is
false.  See the 'else' clause on page 3:229 of Document Number:
245319-004 (ia64 Volume 3: Instruction Set Reference).

Re: vhpt_miss handler question?

[Mario Smarduch]

Keith Owens wrote:

> On Fri, 26 Sep 2003 09:43:55 -0500,
> Mario Smarduch <cms063@email.mot.com> wrote:
> >Some work I'm doing has taken me to the vhpt_miss
> >handler. I've been looking at  the following snippet just
> >wondering what happens in the case where the present
> >bit is 0 and p10,p11 predicates are not set anotherwords
> >they retain their application values. Would this not
> >cause an errant insertion if either p10 or p11 were
> >true in the application prior to the VHPT handler?
>
> >(p7)    tbit.nz.unc p10,p11=r19,32              // is it an instruction
>
> tbit.xx.unc sets both predicates to 0 if the controlling predicate is
> false.  See the 'else' clause on page 3:229 of Document Number:
> 245319-004 (ia64 Volume 3: Instruction Set Reference).

I understand the unc sets bot predicates to 0. The question
was regarding p7úlse, but it appears in that case a TLB
entry with present bit set to 0 may be inserted which appears
to be harmless (other then taking up a TC entry).

Sorry for the noise :)

- mario.

Re: vhpt_miss handler question?

[David Mosberger]

>>>>> On Fri, 26 Sep 2003 09:43:55 -0500, Mario Smarduch <cms063@email.mot.com> said:

  Mario> Some work I'm doing has taken me to the vhpt_miss
  Mario> handler. I've been looking at the following snippet just
  Mario> wondering what happens in the case where the present bit is 0
  Mario> and p10,p11 predicates are not set anotherwords they retain
  Mario> their application values. Would this not cause an errant
  Mario> insertion if either p10 or p11 were true in the application
  Mario> prior to the VHPT handler?

  Mario> (p7) tbit.nz.unc p10,p11=r19,32

Did you miss the .unc qualifier on that instruction?  It causes p10 and
p11 to be set to FALSE when p7 is FALSE.

	--david

Re: vhpt_miss handler question?

[Mario Smarduch]

David Mosberger wrote:

> >>>>> On Fri, 26 Sep 2003 09:43:55 -0500, Mario Smarduch <cms063@email.mot.com> said:
>
>   Mario> Some work I'm doing has taken me to the vhpt_miss
>   Mario> handler. I've been looking at the following snippet just
>   Mario> wondering what happens in the case where the present bit is 0
>   Mario> and p10,p11 predicates are not set anotherwords they retain
>   Mario> their application values. Would this not cause an errant
>   Mario> insertion if either p10 or p11 were true in the application
>   Mario> prior to the VHPT handler?
>
>   Mario> (p7) tbit.nz.unc p10,p11=r19,32
>
> Did you miss the .unc qualifier on that instruction?  It causes p10 and
> p11 to be set to FALSE when p7 is FALSE.
>
>         --david

I see! I guess I didn't fully understand what Keith wrote.

- Mario.