Re: [PATCHv3] perf kvm: add kvm-stat for arm64

From: Leo Yan <leo.yan@linaro.org>
To: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Mark Rutland <mark.rutland@arm.com>,
	Mathieu Poirier <mathieu.poirier@linaro.org>,
	Peter Zijlstra <peterz@infradead.org>,
	Marc Zyngier <maz@kernel.org>, John Garry <john.garry@huawei.com>,
	linux-kernel@vger.kernel.org,
	Arnaldo Carvalho de Melo <acme@kernel.org>,
	Suleiman Souhlal <suleiman@google.com>,
	Namhyung Kim <namhyung@kernel.org>, Will Deacon <will@kernel.org>,
	linux-arm-kernel@lists.infradead.org
Subject: Re: [PATCHv3] perf kvm: add kvm-stat for arm64
Date: Thu, 17 Sep 2020 21:08:54 +0800	[thread overview]
Message-ID: <20200917130854.GG12548@leoy-ThinkPad-X240s> (raw)
In-Reply-To: <20200917110028.GB3049@jagdpanzerIV.localdomain>

Hi Sergey,

On Thu, Sep 17, 2020 at 08:00:28PM +0900, Sergey Senozhatsky wrote:
> On (20/09/17 11:21), Marc Zyngier wrote:
> > > > On Arm64, ftrace tracepoint "kvm_entry" doesn't contain the field "id"
> > > > or field "vcpu_id", thus it always reads out the "id" is 0 and it is
> 
> Right, Leo, I saw that. I put "id", because there is ... well, there is
> nothing we can put there. The trace points on arm64, unlike on s390 or x86,
> don't contain info which can let us map trace-event to vcpu_id (even
> (void *)vcpu would have been rather helpful).
> 
> > > > Essentially, this issue is caused by different archs using different
> > > > format for ftrace event "kvm_entry", on x86 it contains feild
> > > > "vcpu_id" but arm64 only just records "vcpu_pc".
> 
> Exactly. I wish trace-points were less of pain-points.
> 
> So 'perf kvm stat' on arm64 works, but it's not as feature rich as on other
> platforms; at the same it's better than nothing, I suppose.

Agree and I don't want to introduce complexity at here.  So it's fine
for me to merge your patch v4 firstly and later use a new patch set
to support "vcpu_id".

As you might have seen the discussion with Marc for new trace events
with "vcpu_id", another way is to add new trace events with "vcpu_id"
and then enable "perf kvm stat" based on new trace events.

Either way is okay for me; except maintainers have different thinking
for this.

Thanks,
Leo

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel