Re: [PATCH] perf test: Remove atomics from test_loop to avoid test failures

[Nick Forrington]

On 25/11/2023 03:05, Leo Yan wrote:
> Hi all,
>
> On Fri, Nov 24, 2023 at 08:57:52PM +0100, Michael Petlan wrote:
> > On Thu, 2 Nov 2023, Nick Forrington wrote:
> > > The current use of atomics can lead to test failures, as tests (such as
> > > tests/shell/record.sh) search for samples with "test_loop" as the
> > > top-most stack frame, but find frames related to the atomic operation
> > > (e.g. __aarch64_ldadd4_relax).
> I am confused by above description.  As I went through the script
> record.sh, which is the only test invoking the program 'test_loop',
> but I don't find any test is related with stack frame.
>
> Do I miss anything?  I went through record.sh but no clue why the
> failure is caused by stack frame.  All the testings use command:
>
>    if ! perf report -i "${perfdata}" -q | grep -q "${testsym}"
>      ...
>    fi
>
> @Nick, could you narrow down which specific test case causing the
> failure.
>
> [...]


All checks for ${testsym} in record.sh (including the example you 
provide) can fail, as the expected symbol (test_loop) is not the 
top-most function on the stack (and therefore not the symbol associated 
with the sample).


Example perf report output:

# Overhead  Command  Shared Object          Symbol
# ........  .......  ..................... .............................
#
    99.53%  perf     perf                   [.] __aarch64_ldadd4_relax

...


You can see the issue when recording/reporting with call stacks:

# Children      Self  Command  Shared Object          Symbol
# ........  ........  .......  ..................... 
..........................................................
#
    99.52%    99.52%  perf     perf                   [.] 
__aarch64_ldadd4_relax
            |
            |--49.77%--0xffffb905a5dc
            |          0xffffb8ff0aec
            |          thfunc
            |          test_loop
            |          __aarch64_ldadd4_relax

...

> > I believe that it was there to prevent the compiler to optimize the loop
> > out or some reason like that. Hopefully, it will work even without that
> > on all architectures with all compilers that are used for building perf...
> Agreed.
>
> As said above, I'd like to step back a bit for making clear what's the
> exactly failure caused by the program.


I don't think this loop could be sensibly optimised away, as it depends 
on "done", which is defined at file scope (and assigned by a signal 
handler).


Cheers,
Nick

> Thanks,
> Leo