Re: [PATCH] perf test: Skip Sigtrap test for arm+aarch64

[Leo Yan]

On Thu, Feb 17, 2022 at 06:40:45PM +0100, Marco Elver wrote:
> On Thu, 17 Feb 2022 at 18:34, John Garry <john.garry@xxxxxxxxxx> wrote:
> [...]
> > >> -#if defined(__powerpc__) || defined(__s390x__)
> > >> +#if defined(__powerpc__) || defined(__s390x__) || \
> > >> +       defined(__arm__) || defined(__aarch64__)
> > >>   #define BP_ACCOUNT_IS_SUPPORTED 0
> > >>   #else
> > >>   #define BP_ACCOUNT_IS_SUPPORTED 1
> > >
> > > This is now equivalent to BP_SIGNAL_IS_SUPPORTED
> > > tools/perf/tests/tests.h -- and different from the original
> > > BP_ACCOUNT_IS_SUPPORTED (and makes me wonder why
> > > BP_SIGNAL_IS_SUPPORTED wasn't just used from the beginning). Perhaps
> > > just use BP_SIGNAL_IS_SUPPORTED.
> > >
> >
> > We currently have BP_ACCOUNT_IS_SUPPORTED defined now in 2x locations:
> >
> > tests/sigtrap.c
> > tests/bp_account.c
> >
> > bp_account works for arm64, and we don't want to skip that test. So, as
> > long as the macro meaning is appropriate, we can reuse
> > BP_SIGNAL_IS_SUPPORTED for sigtrap.c
> 
> BP_ACCOUNT seems to say something about the "breakpoint accounting /
> measuring" test. BP_SIGNAL is about the tests that want to use
> breakpoints to generate signals.

More general speaking, I think "BP_ACCOUNT_IS_SUPPORTED = 1" means an
architecture can support breakpoint with perf_event.

"BP_SIGNAL_IS_SUPPORTED = 1" means an architecture can support
breakpoint to generate signals with using perf_event.  So
"BP_SIGNAL_IS_SUPPORTED = 1" is subset of "BP_ACCOUNT_IS_SUPPORTED = 1".

> So it's very much appropriate to use BP_SIGNAL here if, as we have
> discovered regardless how they're generated in response to
> breakpoints, are broken on arm/arm64. On the day arm/arm64 decides to
> fix signals, I'm assuming all tests being skipped with
> BP_SIGNAL_IS_SUPPORTED can be re-enabled (or so we hope).

Yeah, I agree that BP_SIGNAL_IS_SUPPORTED is better choice for sigtrap.c.

Thanks,
Leo