Received-SPF: pass (google.com: domain of linux-kernel+bounces-40196-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.80.249 as permitted sender) client-ip=147.75.80.249;
Date: Fri, 26 Jan 2024 14:36:16 +0000
From: Mark Rutland <mark.rutland@arm.com>
To: Marc Zyngier <maz@kernel.org>
Cc: linux-kernel@vger.kernel.org, Hector Martin <marcan@marcan.st>,
	Ian Rogers <irogers@google.com>, acme@redhat.com,
	james.clark@arm.com, john.g.garry@oracle.com, leo.yan@linaro.org,
	linux-arm-kernel@lists.infradead.org,
	linux-perf-users@vger.kernel.org, mike.leach@linaro.org,
	namhyung@kernel.org, suzuki.poulose@arm.com, tmricht@linux.ibm.com,
	will@kernel.org
Subject: Re: [PATCH] perf print-events: make is_event_supported() more robust
Message-ID: <ZbPDYG2Bd2H7C_Es@FVFF77S0Q05N>
References: <20240116170348.463479-1-mark.rutland@arm.com>
 <8734uwxrca.wl-maz@kernel.org>
 <ZafEFU7kwf6W0_Qx@FVFF77S0Q05N.cambridge.arm.com>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <ZafEFU7kwf6W0_Qx@FVFF77S0Q05N.cambridge.arm.com>

On Wed, Jan 17, 2024 at 12:12:05PM +0000, Mark Rutland wrote:
> On Wed, Jan 17, 2024 at 09:05:25AM +0000, Marc Zyngier wrote:
> > However, I'm seeing some slightly odd behaviours:

I believe that this is a separate issue; info dump below.
> > $ sudo ./perf stat -e cycles:k ~/hackbench 100 process 1000
> > Running with 100*40 (== 4000) tasks.
> > Time: 3.313
> > 
> >  Performance counter stats for '/home/maz/hackbench 100 process 1000':
> > 
> >    <not supported>      apple_firestorm_pmu/cycles:k/                                         
> >    <not supported>      apple_icestorm_pmu/cycles:k/                                          
> > 
> >        3.467568841 seconds time elapsed
> > 
> >       13.080111000 seconds user
> >       53.162099000 seconds sys
> > 
> > I would have expected it to count, but it didn't. For that to work, I
> > have to add the 'H' modifier:

I gave that a spin with the aforementioned hacked-up PMUv3 driver, and I see
the same:

| # ./perf-after stat -e cycles true
| 
|  Performance counter stats for 'true':
| 
|      <not counted>      armv8_pmuv3_0/cycles/                                                   (0.00%)
|            1375271      armv8_pmuv3_1/cycles/                                                 
| 
|        0.001153070 seconds time elapsed
| 
|        0.001204000 seconds user
|        0.000000000 seconds sys
| 
| 
| # ./perf-after stat -e cycles:k true
| 
|  Performance counter stats for 'true':
| 
|    <not supported>      armv8_pmuv3_0/cycles:k/                                               
|    <not supported>      armv8_pmuv3_1/cycles:k/                                               
| 
|        0.000983130 seconds time elapsed
| 
|        0.001037000 seconds user
|        0.000000000 seconds sys
| 
| 
| # ./perf-after stat -e cycles:kH true
| 
|  Performance counter stats for 'true':
| 
|      <not counted>      armv8_pmuv3_0/cycles:kH/                                                (0.00%)
|             932067      armv8_pmuv3_1/cycles:kH/                                              
| 
|        0.001090100 seconds time elapsed
| 
|        0.001125000 seconds user
|        0.000000000 seconds sys

.. though interestingly 'cycles:u' works:

| # ./perf-after stat -e cycles:u true
| 
|  Performance counter stats for 'true':
| 
|             369753      armv8_pmuv3_0/cycles:u/                                               
|      <not counted>      armv8_pmuv3_1/cycles:u/                                                 (0.00%)
| 
|        0.001171980 seconds time elapsed
| 
|        0.001245000 seconds user
|        0.000000000 seconds sys

Looking at the output with '-vvv' the perf tool implicitly sets exclude_guest
for 'cycles', 'cycles:u', and 'cycles:kH', but does not set exclude_guest for
'cycles:k'.

It looks like that's consistent with the behaviour of opening separate events
prior to this patch:

| # ./perf-before stat -e armv8_pmuv3_0/cycles/ -e armv8_pmuv3_1/cycles/ true
| 
|  Performance counter stats for 'true':
| 
|            1407624      armv8_pmuv3_0/cycles/                                                 
|      <not counted>      armv8_pmuv3_1/cycles/                                                   (0.00%)
| 
|        0.001179205 seconds time elapsed
| 
|        0.001217000 seconds user
|        0.000000000 seconds sys
| 
| 
| # ./perf-before stat -e armv8_pmuv3_0/cycles/u -e armv8_pmuv3_1/cycles/u true
| 
|  Performance counter stats for 'true':
| 
|             329212      armv8_pmuv3_0/cycles/u                                                
|      <not counted>      armv8_pmuv3_1/cycles/u                                                  (0.00%)
| 
|        0.001050550 seconds time elapsed
| 
|        0.001081000 seconds user
|        0.000000000 seconds sys
| 
| 
| # ./perf-before stat -e armv8_pmuv3_0/cycles/k -e armv8_pmuv3_1/cycles/k true
| 
|  Performance counter stats for 'true':
| 
|    <not supported>      armv8_pmuv3_0/cycles/k                                                
|    <not supported>      armv8_pmuv3_1/cycles/k                                                
| 
|        0.000944285 seconds time elapsed
| 
|        0.000985000 seconds user
|        0.000000000 seconds sys
| 
| 
| # ./perf-before stat -e armv8_pmuv3_0/cycles/kH -e armv8_pmuv3_1/cycles/kH true
| 
|  Performance counter stats for 'true':
| 
|            1016160      armv8_pmuv3_0/cycles/kH                                               
|      <not counted>      armv8_pmuv3_1/cycles/kH                                                 (0.00%)
| 
|        0.001179220 seconds time elapsed
| 
|        0.001239000 seconds user
|        0.000000000 seconds sys

.. and per '-vvv', exclude_guest is set in the same cases.

I agree it's a bit weird that the tool sets exclude_guest for unfilted and ':u'
events, but not ':k' events, but it looks like that's separate from the way
events get expanded.

Thanks,
Mark.