[PATCH 0/1] Return EINVAL when precise_ip perf events are requested on Arm
james.clark at arm.com
Wed Jan 15 02:58:54 PST 2020
Since we're adding support for SPE in user space Perf, we've encountered an issue
where we would like some more feedback if SPE isn't available for an event.
At the moment there is a patch for perf where you can enable SPE by doing this:
perf record -r branch-misses:p ...
Perf will have a hard coded list of events that can use SPE when ":p" is specified
and open the SPE PMU instead of the specified one. But if the event isn't in that
list, then Perf will attempt to open the normal event with precise_ip = 1.
That will succeed at the moment, but we'd like the kernel to say it's not supported
so there is a chance of showing a warning to the user.
This isn't just relevant to Perf though, there may be other tools that are already
Therefore I'm looking for feedback on whether this would break backwards
compatibility with user space tools that are already setting precise_ip and
expecting it to not error out on Arm.
This change would also be beneficial for the case where if in the (distant) future
we do add some kind of precise support, there will be a chance of userspace
determining what is supported and what isn't.
Cc: Will Deacon <will at kernel.org>
Cc: Mark Rutland <mark.rutland at arm.com>
Cc: Peter Zijlstra <peterz at infradead.org>
Cc: Ingo Molnar <mingo at redhat.com>
Cc: Arnaldo Carvalho de Melo <acme at kernel.org>
Cc: Alexander Shishkin <alexander.shishkin at linux.intel.com>
Cc: Jiri Olsa <jolsa at redhat.com>
Cc: Tan Xiaojun <tanxiaojun at huawei.com>
Cc: Al Grant <al.grant at arm.com>
Cc: Namhyung Kim <namhyung at kernel.org>
Cc: linux-arm-kernel at lists.infradead.org
Cc: linux-kernel at vger.kernel.org
James Clark (1):
Return EINVAL when precise_ip perf events are requested on Arm
drivers/perf/arm_pmu.c | 3 +++
include/uapi/linux/perf_event.h | 4 ++--
2 files changed, 5 insertions(+), 2 deletions(-)
More information about the linux-arm-kernel