Re: [PATCH 3/5 v4] Fix argument to compiled_cond, and add cases for compiled-condition.

[Pedro Alves <palves at redhat dot com>]

On 06/27/2015 05:21 PM, Wei-cheng Wang wrote:
> Hi,
> 
> Ulrich Weigand wrote:
>> > Ah, when I said to add new test cases in a separate patch, what I meant was:
>> > - use a separate patch (applied *first*) that adds the *new tests* (to be
>> >   run on existing platforms), i.e. test_ftrace_condition
>> > - as part of the patch that actually adds powerpc support, add all the small
>> >   test case snippets that specifically enable the test cases for powerpc
>> > This is again so that each set in a series is meaningful in itself (and
>> > does not introduce testsuite regressions when applied alone).
> ...
>> > Wei-cheng Wang wrote:
>>> > >if (tpoint->compiled_cond)
>>> > >   err = ((condfn) (uintptr_t) (tpoint->compiled_cond)) (ctx, &value);
>>> > >
>>> > >I think probably either we could pass ctx->regs to compiled_cond instead,
>>> > >or move the declarations of fast_tracepoint_ctx (and others) to tracepoint.h,
>>> > >so we can use "offsetof (fast_tracepoint_ctx, regs)" instead.
>>> > >Any suggestion?
>> > FWIW, passing the regs buffer directly to the compiled routine seems
>> > more straightforward to me ...
> Some of the new cases are used to testing emit-reg, and emit-reg for x86
> doesn't work due to the incorrect argument to compiled_cond - "regs" buffer
> is expected, but tracepoint context is passed
> 
> This case also includes the fix for complied_cond in order for x86 to
> pass testing for emit-reg op.
> 
> Testing result on my x86_64 Ubuntu 14.04.2 TLS
> 
>                before   w/ new cases only  w/ the fix
>   PASS         2625     2759               2765
>   FAIL           33       39                 33
>   XFAIL          16       16                 16
>   KFAIL           2        2                  2
>   UNTESTED        1        1                  1
>   UNSUPPORTED     3        3                  3
> 
> Thanks,
> Wei-cheng
> 
> ---
> 
> * Fix generating emit-reg op by passing register buffer to compiled_cond.
> * Add a new function for testing compiled-cond by checking whether
>   based on a given CONDEXP, a list of expected values should be collected.
> 
> ---

FWIW, looks good to me too.  Thanks for the new tests!

Thanks,
Pedro Alves