Re: [RFC] perf arm-spe: Track task context switch for cpu-mode events

[Namhyung Kim]

Hi Leo,

On Thu, Sep 23, 2021 at 7:23 AM Leo Yan <leo.yan@xxxxxxxxxx> wrote:
>
> Hi Namhyung,
>
> On Thu, Sep 16, 2021 at 02:01:21PM -0700, Namhyung Kim wrote:
>
> [...]
>
> > > Before we had discussion for enabling PID/TID for SPE samples; in the patch
> > > set [1], patches 07, 08 set sample's pid/tid based on the Arm SPE context
> > > packets.  To enable hardware tracing context ID, you also needs to enable
> > > kernel config CONFIG_PID_IN_CONTEXTIDR.
> >
> > Thanks for sharing this.
> >
> > Yeah I also look at the context info but having a dependency on a kconfig
> > looks limiting its functionality.  Also the kconfig says it has some overhead
> > in the critical path (even if perf is not running, right?) - but not sure how
> > much it can add.
>
> Yes, after enabled config PID_IN_CONTEXTIDR, the kernel will always
> write PID into the system register CONTEXTIDR during process context
> switching.  Please see the flow:
>
>   __switch_to() (arch/arm64/kernel/process.c)
>     `-> contextidr_thread_switch(next)

Thanks for the info.  I assume it's a light-weight operation.


> > > We also considered to use PERF_RECORD_SWITCH_CPU_WIDE event for setting
> > > pid/tid, the Intel PT implementation uses two things to set sample's
> > > pid/tid: one is PERF_RECORD_SWITCH_CPU_WIDE event and another is to detect
> > > the branch instruction is the symbol "__switch_to".  Since the trace
> > > event PERF_RECORD_SWITCH_CPU_WIDE is coarse, so it only uses the new
> > > pid/tid after the branch instruction for "__switch_to".  Arm SPE is
> > > 'statistical', thus it cannot promise the trace data must contain the
> > > branch instruction for "__switch_to", please see details [2].
> >
> > I can see the need in the Intel PT as it needs to trace all (branch)
> > instructions, but is it really needed for ARM SPE too?
> > Maybe I am missing something, but it seems enough to have a
> > coarse-grained context switch for sampling events..
>
> The issue is that the coarse-grained context switch if introduces any
> inaccuracy in the reported result.  If we can run some workloads and
> prove the coarse-grained context switch doesn't cause significant bias,
> it will be great and can give us the confidence for this approach.
>
> Even enabling PERF_RECORD_SWITCH_CPU_WIDE event, I think it's good to
> give priority for hardware PID tracing in Arm SPE trace data, if detects
> the hardware PID tracing is enabled, then we can rollback to use
> context packets from hardware trace data to set sample's PID.
>
> How about you think for this?

I think it's good as long as it has a fallback when the context info
is not available.

Thanks,
Namhyung