* OProfile callgraph support not working correctly on PPC processors
@ 2007-12-21 17:15 Bob Nelson
2008-01-07 0:11 ` Anton Blanchard
0 siblings, 1 reply; 2+ messages in thread
From: Bob Nelson @ 2007-12-21 17:15 UTC (permalink / raw)
To: linuxppc
I have been investigating why I have not been able to get callgraph code fo=
r=20
OProfile on Cell to work correctly and I am pretty sure that I have run int=
o=20
a problem that is common across all the Power platforms. (At least the oth=
er=20
ones I have looked at.) I have a simple test program that is attached=20
below. =A0It has a main, that calls function1, which calls function2. =A0Ea=
ch of=20
the functions has some type of loop in it so that I can catch it spending=20
some CPU time with OProfile. =A0I have also attached the objdump -d output =
for=20
the program cut down to the three pertinent functions that shows what is=20
happening. =A0In a nutshell when a terminal function (calls no other functi=
on)=20
is called the compiler is making an optimization that seems to break the AB=
I=20
convention as far as I can tell. =A0It does not store the Link Register on =
the=20
stack like any other function. It just leaves the return address in LR,=20
knowing that nothing should change it. =A0(You can see at the top of both m=
ain=20
and function1 the first thing it does is "mflr =A0r0" to copy the link regi=
ster=20
to R0 to be saved. It does not do that in function2.) =A0 When OProfile tak=
es=20
an interrupt and needs to gather the callgraph information it does so by=20
grabbing the process' stack pointer (R1) and follows the chain back up the=
=20
stack to gather all the caller's addresses. =A0This works for most function=
s,=20
except for terminal functions for the reason noted above.
Looking at the assembly listing I drew myself a diagram of the stack while=
=20
function2 is active to convince myself of what was wrong and here is what I=
=20
see it as... =A0When the interrupt is handled OProfile grabs a copy of R1, =
it=20
ignores the first frame on the stack because there should be no address=20
stored. =A0In the second frame it expects to find function2's caller but si=
nce=20
function2 doesn't store it, it grabs some random data and proceeds. The sta=
ck=20
chain is all ok so it doesn't go off into neverland trying to follow a bad=
=20
chain, but it grabs an invalid address for the caller. =A0And that is why=20
OProfile thinks terminal functions have no callers on PPC...
Any suggestions on how this can be fixed? =A0I am guessing that changing th=
e=20
compiler and recompiling every program is probably not the answer. I assum=
e=20
the link register has to be saved in the interrupt routine when it runs, or=
=20
else it couldn't call anything else without crashing the program that was=20
interrupted. Is there a safe place to find it?
Thanks, Bob Nelson
top of stack =A0 ------------------------------
=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0| =A0 =A0 =A0 . =A0 =A0 =A0 =A0 =A0 =A0 =A0 =
=A0 =A0 =A0|
=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0| =A0 =A0 =A0 . =A0 =A0 =A0 =A0 =A0 =A0 =A0 =
=A0 =A0 =A0| <------------------------------
=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0| =A0 =A0 =A0 . =A0 =A0 =A0 =A0 =A0 =A0 =A0 =
=A0 =A0 =A0| |
=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0|----------------------------| =A0 =A0 =A0 =
=A0 =A0 |
=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0| =A0 =A0R0 (link register) =A0 =A0 =A0| =A0=
=2D-> main's caller =A0 =A0 |
=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0|----------------------------| =A0 =A0 =A0 =
=A0 =A0 |
=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0| =A0 =A0flags (unused) =A0 =A0 =A0 =A0 =A0|=
=A0 =A0 =A0 =A0 =A0 =A0 |
=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0|----------------------------|=A0 =A0 =A0 =
=A0 =A0 |
=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0| =A0 =A0R1 (previous frame) |>---------=
=2D---------------------
R1 main =A0 =A0 -> |----------------------------| 0 (Offset from R1 <----=
=2D-----
=A0 =A0(entry) =A0 =A0=A0| =A0 =A0R31 save =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0|=
=A0 at entry to main) =A0 |
=A0 =A0 =A0 =A0 =A0 =A0 =A0 |----------------------------| -8 =A0 =A0 =A0=
=A0 =A0 =A0 =A0 =A0 =A0 =A0 |
=A0 =A0 =A0 =A0 =A0 =A0 =A0 | =A0 =A0 =A0 . =A0 =A0 =A0 =A0 =A0 =A0 =A0 =
=A0 =A0 =A0| =A0 =A0 =A0 =A0 =A0 =A0 |
=A0 =A0 =A0 =A0 =A0 =A0 =A0 | =A0 =A0 =A0 . =A0 =A0 =A0 =A0 =A0 =A0 =A0 =
=A0 =A0 =A0|=A0 =A0 =A0 =A0 =A0 =A0 |=20
=A0 =A0 =A0 =A0 =A0 =A0 =A0 | =A0 =A0 =A0 . =A0 =A0 =A0 =A0 =A0 =A0 =A0 =
=A0 =A0 =A0| =A0 =A0 =A0 =A0 =A0 =A0|
=A0 =A0 =A0 =A0 =A0 =A0 =A0 |----------------------------| =A0 =A0 =A0 =A0=
=A0 =A0 |
=A0 =A0 =A0 =A0 =A0 =A0 =A0 | =A0 =A0R0 (link register) =A0 =A0 =A0| =A0--=
>function1's caller (main) |
=A0 =A0 =A0 =A0 =A0 =A0 =A0 |----------------------------| =A0 =A0 =
=A0 |
=A0 =A0 =A0 =A0 =A0 =A0 =A0 | =A0 =A0flags (not stored)=A0 =A0 =A0 |=A0 =
=A0 =A0 =A0 =A0 =A0 |
=A0 =A0 =A0 =A0 =A0 =A0 =A0 |----------------------------| =A0 =A0 =A0 =A0=
=A0 =A0 |
=A0 =A0 =A0 =A0 =A0 =A0 =A0 | =A0 =A0R1 (previous frame) |>-----------=
=2D-------------------
R1 function1-->|----------------------------| -144 <-----------------------=
=2D-
=A0 =A0(entry) =A0 =A0 | =A0 =A0R31 save =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0| =
=A0 =A0 =A0 =A0 =A0 =A0 |
=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0|----------------------------|=A0 =A0 =A0 =
=A0 =A0 =A0 |
=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0| =A0 =A0 =A0 . =A0 =A0 =A0 =A0 =A0 =A0 =A0 =
=A0 =A0 =A0| =A0 =A0 =A0 =A0 =A0 =A0 |
=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0| =A0 =A0 =A0 . =A0 =A0 =A0 =A0 =A0 =A0 =A0 =
=A0 =A0 =A0| =A0 =A0 =A0 =A0 =A0 =A0 |
=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0| =A0 =A0 =A0 . =A0 =A0 =A0 =A0 =A0 =A0 =A0 =
=A0 =A0 =A0| =A0 =A0 =A0 =A0 =A0 =A0 |
=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0|----------------------------| =A0 =A0 =A0 =
=A0 =A0 =A0 |
=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0| =A0 =A0nothing stored =A0 =A0 =A0 =A0 =A0|=
(should be function2's caller |
=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0|----------------------------| =A0function1)=
=A0 =A0 =A0 =A0 =A0 |
=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0| =A0 =A0flags (not stored) =A0 =A0 =A0| =A0=
=A0 =A0 =A0 =A0 =A0 |
=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0|----------------------------| =A0 =A0 =A0 =
=A0 =A0 =A0 |
=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0| =A0 =A0R1 (previous frame) |>---------=
=2D---------------------
R1 function2-->|----------------------------| -288 <-----------------------=
=2D-
=A0 =A0(entry) =A0 =A0 | =A0 =A0R31 save =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0| =
=A0 =A0 =A0 =A0 =A0 =A0 |
=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0|----------------------------| =A0 =A0 =A0 =
=A0 =A0 =A0 |
=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0| =A0 =A0 =A0 . =A0 =A0 =A0 =A0 =A0 =A0 =A0 =
=A0 =A0 =A0| =A0 =A0 =A0 =A0 =A0 =A0 |
=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0| =A0 =A0 =A0 . =A0 =A0 =A0 =A0 =A0 =A0 =A0 =
=A0 =A0 =A0| =A0 =A0 =A0 =A0 =A0 =A0 |
=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0| =A0 =A0 =A0 . =A0 =A0 =A0 =A0 =A0 =A0 =A0 =
=A0 =A0 =A0| =A0 =A0 =A0 =A0 =A0 =A0 |
=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0|----------------------------| =A0 =A0 =A0 =
=A0 =A0 =A0 |
=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0| =A0 =A0nothing stored =A0 =A0 =A0 =A0 =A0|=
=A0would be used if function2 =A0|
=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0|----------------------------| =A0called any=
thing |
=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0| =A0 =A0flags (not stored) =A0 =A0 =A0| =A0=
=A0 =A0 =A0 =A0 =A0 |
=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0|----------------------------| =A0 =A0 =A0 =
=A0 =A0 =A0 |
=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0| =A0 =A0R1 (previous frame) |>---------=
=2D---------------------
R1 function2-->|----------------------------| -368=A0 =A0(running)
| =A0 =A0 =A0 . =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0|
=A0 =A0 =A0 =A0 =A0 =A0 =A0 | =A0 =A0 =A0 . =A0 =A0 =A0 =A0 =A0 =A0 =A0 =
=A0 =A0 =A0|
=A0 =A0 =A0 =A0 =A0 =A0 =A0 | =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =
=A0 =A0 =A0|
/* loop.c - nonsense code for testing OProfile */
#include <stdio.h>
int function2( int count )
{
int i, j, k;
for ( i=3D0; i<count; i++ )
{
k =3D k + j * i;
}
return k;
}
int function1( int count )
{
int i, j;
i =3D function2( count );
for ( j=3D0; j<1000; j++ ) i++;
return i;
}
int main( int argc, char *argv[] )
{
int count, i, j, k;
if ( argc > 0 )
count =3D atoi( argv[1] );
else
count =3D 10000;
for ( i=3D0; i<count; i++ )
{
j =3D function1( 10000 );
for( j=3D0; j<10000; j++ ) k =3D k + j;
}
return 0;
}
loop.64: file format elf64-powerpc
=2E.. deleted ...
00000000100005b0 <.function2>:
100005b0: fb e1 ff f8 std r31,-8(r1)
100005b4: f8 21 ff b1 stdu r1,-80(r1)
100005b8: 7c 3f 0b 78 mr r31,r1
100005bc: 7c 60 1b 78 mr r0,r3
100005c0: 90 1f 00 80 stw r0,128(r31)
100005c4: 38 00 00 00 li r0,0
100005c8: 90 1f 00 38 stw r0,56(r31)
100005cc: 48 00 00 2c b 100005f8 <.function2+0x48>
100005d0: 81 3f 00 34 lwz r9,52(r31)
100005d4: 80 1f 00 38 lwz r0,56(r31)
100005d8: 7c 09 01 d6 mullw r0,r9,r0
100005dc: 7c 09 07 b4 extsw r9,r0
100005e0: 80 1f 00 30 lwz r0,48(r31)
100005e4: 7c 00 4a 14 add r0,r0,r9
100005e8: 90 1f 00 30 stw r0,48(r31)
100005ec: 81 3f 00 38 lwz r9,56(r31)
100005f0: 38 09 00 01 addi r0,r9,1
100005f4: 90 1f 00 38 stw r0,56(r31)
100005f8: 80 1f 00 38 lwz r0,56(r31)
100005fc: 81 3f 00 80 lwz r9,128(r31)
10000600: 7f 80 48 00 cmpw cr7,r0,r9
10000604: 41 9c ff cc blt+ cr7,100005d0 <.function2+0x20>
10000608: 80 1f 00 30 lwz r0,48(r31)
1000060c: 7c 00 07 b4 extsw r0,r0
10000610: 7c 03 03 78 mr r3,r0
10000614: e8 21 00 00 ld r1,0(r1)
10000618: eb e1 ff f8 ld r31,-8(r1)
1000061c: 4e 80 00 20 blr
...
10000628: 80 01 00 01 lwz r0,1(r1)
000000001000062c <.function1>:
1000062c: 7c 08 02 a6 mflr r0
10000630: fb e1 ff f8 std r31,-8(r1)
10000634: f8 01 00 10 std r0,16(r1)
10000638: f8 21 ff 71 stdu r1,-144(r1)
1000063c: 7c 3f 0b 78 mr r31,r1
10000640: 7c 60 1b 78 mr r0,r3
10000644: 90 1f 00 c0 stw r0,192(r31)
10000648: 80 1f 00 c0 lwz r0,192(r31)
1000064c: 7c 00 07 b4 extsw r0,r0
10000650: 7c 03 03 78 mr r3,r0
10000654: 4b ff ff 5d bl 100005b0 <.function2>
10000658: 7c 60 1b 78 mr r0,r3
1000065c: 90 1f 00 74 stw r0,116(r31)
10000660: 38 00 00 00 li r0,0
10000664: 90 1f 00 70 stw r0,112(r31)
10000668: 48 00 00 1c b 10000684 <.function1+0x58>
1000066c: 81 3f 00 74 lwz r9,116(r31)
10000670: 38 09 00 01 addi r0,r9,1
10000674: 90 1f 00 74 stw r0,116(r31)
10000678: 81 3f 00 70 lwz r9,112(r31)
1000067c: 38 09 00 01 addi r0,r9,1
10000680: 90 1f 00 70 stw r0,112(r31)
10000684: 80 1f 00 70 lwz r0,112(r31)
10000688: 2f 80 03 e7 cmpwi cr7,r0,999
1000068c: 40 9d ff e0 ble+ cr7,1000066c <.function1+0x40>
10000690: 80 1f 00 74 lwz r0,116(r31)
10000694: 7c 00 07 b4 extsw r0,r0
10000698: 7c 03 03 78 mr r3,r0
1000069c: e8 21 00 00 ld r1,0(r1)
100006a0: e8 01 00 10 ld r0,16(r1)
100006a4: 7c 08 03 a6 mtlr r0
100006a8: eb e1 ff f8 ld r31,-8(r1)
100006ac: 4e 80 00 20 blr
100006b0: 00 00 00 00 .long 0x0
100006b4: 00 00 00 01 .long 0x1
100006b8: 80 01 00 01 lwz r0,1(r1)
00000000100006bc <.main>:
100006bc: 7c 08 02 a6 mflr r0
100006c0: fb e1 ff f8 std r31,-8(r1)
100006c4: f8 01 00 10 std r0,16(r1)
100006c8: f8 21 ff 71 stdu r1,-144(r1)
100006cc: 7c 3f 0b 78 mr r31,r1
100006d0: 7c 60 1b 78 mr r0,r3
100006d4: f8 9f 00 c8 std r4,200(r31)
100006d8: 90 1f 00 c0 stw r0,192(r31)
100006dc: 80 1f 00 c0 lwz r0,192(r31)
100006e0: 2f 80 00 00 cmpwi cr7,r0,0
100006e4: 40 9d 00 28 ble- cr7,1000070c <.main+0x50>
100006e8: e9 3f 00 c8 ld r9,200(r31)
100006ec: 39 29 00 08 addi r9,r9,8
100006f0: e8 09 00 00 ld r0,0(r9)
100006f4: 7c 03 03 78 mr r3,r0
100006f8: 4b ff fc f9 bl 100003f0 <._init+0x38>
100006fc: e8 41 00 28 ld r2,40(r1)
10000700: 7c 60 1b 78 mr r0,r3
10000704: 90 1f 00 7c stw r0,124(r31)
10000708: 48 00 00 0c b 10000714 <.main+0x58>
1000070c: 38 00 27 10 li r0,10000
10000710: 90 1f 00 7c stw r0,124(r31)
10000714: 38 00 00 00 li r0,0
10000718: 90 1f 00 78 stw r0,120(r31)
1000071c: 48 00 00 54 b 10000770 <.main+0xb4>
10000720: 38 60 27 10 li r3,10000
10000724: 4b ff ff 09 bl 1000062c <.function1>
10000728: 7c 60 1b 78 mr r0,r3
1000072c: 90 1f 00 74 stw r0,116(r31)
10000730: 38 00 00 00 li r0,0
10000734: 90 1f 00 74 stw r0,116(r31)
10000738: 48 00 00 20 b 10000758 <.main+0x9c>
1000073c: 81 3f 00 70 lwz r9,112(r31)
10000740: 80 1f 00 74 lwz r0,116(r31)
10000744: 7c 09 02 14 add r0,r9,r0
10000748: 90 1f 00 70 stw r0,112(r31)
1000074c: 81 3f 00 74 lwz r9,116(r31)
10000750: 38 09 00 01 addi r0,r9,1
10000754: 90 1f 00 74 stw r0,116(r31)
10000758: 80 1f 00 74 lwz r0,116(r31)
1000075c: 2f 80 27 0f cmpwi cr7,r0,9999
10000760: 40 9d ff dc ble+ cr7,1000073c <.main+0x80>
10000764: 81 3f 00 78 lwz r9,120(r31)
10000768: 38 09 00 01 addi r0,r9,1
1000076c: 90 1f 00 78 stw r0,120(r31)
10000770: 80 1f 00 78 lwz r0,120(r31)
10000774: 81 3f 00 7c lwz r9,124(r31)
10000778: 7f 80 48 00 cmpw cr7,r0,r9
1000077c: 41 9c ff a4 blt+ cr7,10000720 <.main+0x64>
10000780: 38 00 00 00 li r0,0
10000784: 7c 03 03 78 mr r3,r0
10000788: e8 21 00 00 ld r1,0(r1)
1000078c: e8 01 00 10 ld r0,16(r1)
10000790: 7c 08 03 a6 mtlr r0
10000794: eb e1 ff f8 ld r31,-8(r1)
10000798: 4e 80 00 20 blr
1000079c: 00 00 00 00 .long 0x0
100007a0: 00 00 00 01 .long 0x1
100007a4: 80 01 00 01 lwz r0,1(r1)
100007a8: 60 00 00 00 nop
100007ac: 60 00 00 00 nop
=2E.. deleted ...
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: OProfile callgraph support not working correctly on PPC processors
2007-12-21 17:15 OProfile callgraph support not working correctly on PPC processors Bob Nelson
@ 2008-01-07 0:11 ` Anton Blanchard
0 siblings, 0 replies; 2+ messages in thread
From: Anton Blanchard @ 2008-01-07 0:11 UTC (permalink / raw)
To: Bob Nelson; +Cc: linuxppc
Hi,
> I have been investigating why I have not been able to get callgraph code for
> OProfile on Cell to work correctly and I am pretty sure that I have run into
> a problem that is common across all the Power platforms. (At least the other
> ones I have looked at.) I have a simple test program that is attached
> below. It has a main, that calls function1, which calls function2. Each of
> the functions has some type of loop in it so that I can catch it spending
> some CPU time with OProfile. I have also attached the objdump -d output for
> the program cut down to the three pertinent functions that shows what is
> happening. In a nutshell when a terminal function (calls no other function)
> is called the compiler is making an optimization that seems to break the ABI
> convention as far as I can tell. It does not store the Link Register on the
> stack like any other function. It just leaves the return address in LR,
> knowing that nothing should change it. (You can see at the top of both main
> and function1 the first thing it does is "mflr r0" to copy the link register
> to R0 to be saved. It does not do that in function2.) When OProfile takes
> an interrupt and needs to gather the callgraph information it does so by
> grabbing the process' stack pointer (R1) and follows the chain back up the
> stack to gather all the caller's addresses. This works for most functions,
> except for terminal functions for the reason noted above.
Sounds like a standard tail call optimisation. You would have a similar
problem on a normal function if you took a PMU exception in the function
prolog before the LR had been saved to the stack.
> Looking at the assembly listing I drew myself a diagram of the stack while
> function2 is active to convince myself of what was wrong and here is what I
> see it as... When the interrupt is handled OProfile grabs a copy of R1, it
> ignores the first frame on the stack because there should be no address
> stored. In the second frame it expects to find function2's caller but since
> function2 doesn't store it, it grabs some random data and proceeds. The stack
> chain is all ok so it doesn't go off into neverland trying to follow a bad
> chain, but it grabs an invalid address for the caller. And that is why
> OProfile thinks terminal functions have no callers on PPC...
>
> Any suggestions on how this can be fixed? I am guessing that changing the
> compiler and recompiling every program is probably not the answer. I assume
> the link register has to be saved in the interrupt routine when it runs, or
> else it couldn't call anything else without crashing the program that was
> interrupted. Is there a safe place to find it?
You could make use of the LR itself, but you would need to work out if
the LR was in the same function as the stack saved LR, or worse still if
someone uses the LR for an indirect branch it could be completely wrong.
To do a better job you would probably have to parse the dwarf debug
info. Maybe we could always dump the LR and stack LR and decide what one
to use when oprofile userspace parses it.
Anton
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2008-01-07 0:11 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2007-12-21 17:15 OProfile callgraph support not working correctly on PPC processors Bob Nelson
2008-01-07 0:11 ` Anton Blanchard
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).