Re: [pcre-dev] Fwd: Bug#840354: src:pcre3: FTBFS on powerpc …

Startseite
Nachricht löschen
Autor: Zoltán Herczeg
Datum:  
To: Christoph Biedl
CC: pcre-dev
Betreff: Re: [pcre-dev] Fwd: Bug#840354: src:pcre3: FTBFS on powerpc (G4 CPU)
Another idea just came to my mind. This issue could be cache flush issue, since the CPU executes instructions from the instruction cache, while gdb prints instructions from the data cache.

In this patch cache flush changed by Tavian Barnes' request:

https://lists.exim.org/lurker/message/20160406.071510.a067ef1f.en.html

Might be the cause of the issue.

Regards,
Zoltan

"Zoltán Herczeg" <hzmester@???> írta:
>Hi Christoph,
>
>>Very likely not. I saw SIGILL from other code, gdb pointed right to
>>the place. So, just as another example:
>
>Just asking :) Yes, SIGILL should be precise on all cpus.
>
>>| (gdb) disassemble 0xb7fe40a8,0xb7fe40c8
>>| Dump of assembler code from 0xb7fe40a8 to 0xb7fe40c8:
>>|    0xb7fe40a8:  blt     cr1,0xb7fe40b8
>>|    0xb7fe40ac:  addis   r8,r3,8188
>>|    0xb7fe40b0:  lbz     r3,13308(r8)
>>|    0xb7fe40b4:  add     r30,r30,r3
>>| => 0xb7fe40b8:  cmplw   cr1,r30,r29

>
>Perhaps it does not like the cr1 form? Unlikely though.
>
>>|    0xb7fe40bc:  bge     cr1,0xb7fe40d8
>>|    0xb7fe40c0:  lbz     r3,0(r30)
>>|    0xb7fe40c4:  cmpwi   r3,65

>
>>Can you enlighten me about the ways to debug JIT? For example, is
>>there a way to disassemble or at least dump the generated code for
>>further inspection?
>
>You can do that with gdb. There are multiple (I think three) entry points in pcre_jit_compile.c, but they all uses the call_executable_func(arguments) or call_executable_func(&arguments) form. So please search all call_executable_func calls and put a breakpoint to all of them. You can follow the executed machine instructions with the si (step instruction) command, and dump the jit code as well. The "display/i $pc" is a great help with si command. Since you are running the a full test, perhaps not the first call_executable_func crashes. With the "ignore" command you can find the offending call. E.g.
>
>b line
>-> breakpoint 1 created
>ignore 1 10000000
>-> ignores the first 10000000 hits of the breakpoints
>c
>-> crash
>info breakpoints
>-> check how many times the breakpoint 1 ignored
>-> rerun the program and set the parameter of ignore to that number-1
>ignore 1 number-1
>-> now your program stops just before the crash
>display/i $pc
>si
>-> repeat "si" az many times as you want
>
>When the location of the code is ??? (we are in jit code) you can disassemble the whole function.
>
>Regards,
>Zoltan
>
>
>--
>## List details at https://lists.exim.org/mailman/listinfo/pcre-dev