Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::3:3 as permitted sender) client-ip=2620:137:e000::3:3;
Message-ID: <94a2011f-38a6-a883-99c5-5693128f803e@linux.intel.com>
Date:   Tue, 3 Oct 2023 11:58:26 -0400
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:102.0) Gecko/20100101
 Thunderbird/102.15.1
Subject: Re: [RESEND PATCH V3 1/6] perf: Add branch stack extra
Content-Language: en-US
To:     Andi Kleen <ak@linux.intel.com>,
        Peter Zijlstra <peterz@infradead.org>
Cc:     mingo@redhat.com, acme@kernel.org, linux-kernel@vger.kernel.org,
        mark.rutland@arm.com, alexander.shishkin@linux.intel.com,
        jolsa@kernel.org, namhyung@kernel.org, irogers@google.com,
        adrian.hunter@intel.com, eranian@google.com,
        alexey.v.bayduraev@linux.intel.com, tinghao.zhang@intel.com,
        Sandipan Das <sandipan.das@amd.com>,
        Ravi Bangoria <ravi.bangoria@amd.com>,
        Athira Rajeev <atrajeev@linux.vnet.ibm.com>
References: <20230911154822.2559213-1-kan.liang@linux.intel.com>
 <20231002154535.GB35785@noisy.programming.kicks-ass.net>
 <ce16c6c1-9bdc-35e2-fe85-155bd7edbaa4@linux.intel.com>
 <20231002213752.GB1539@noisy.programming.kicks-ass.net>
 <ed169d4d-76d5-c134-c685-ad3d812028be@linux.intel.com>
 <20231003102733.GC1539@noisy.programming.kicks-ass.net>
 <ZRwuE8LPkwtkjX5C@tassilo>
From:   "Liang, Kan" <kan.liang@linux.intel.com>
In-Reply-To: <ZRwuE8LPkwtkjX5C@tassilo>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit
Precedence: bulk


On 2023-10-03 11:06 a.m., Andi Kleen wrote:
>> I'm thinking we should do something like expose branch_counter_nr and
>> branch_counter_width in the sysfs node, and then rename this extra field
>> to counters.
>>
>> Then userspace can do something like:
>>
>> 	for (i = 0; i < branch_counter_nr; i++) {
>> 		counter[i] = counters & ((1 << branch_counter_width) - 1);
>> 		counters >>= branch_counter_width;
>> 	}
>>
>> to extract the actual counter values.
> 
> perf script/report won't necessarily have access to the sysfs
> values if they run on a different system
> 
> It would need extra PT style metadata written by perf record to
> perf.data and read by the user tools.
> 
> Seems complicated. It would be better if it just parsed on its own.
> 

If so, perf probably have to save the information in the
perf_branch_entry. At least, perf has to be able to support 64 counters
and 64 width. That will use at least 12 bits of the info field of the
perf_branch_entry. (We only have 31 spare bits now. Almost half.) That
seems a waste.

Perf tool already supports "branches" in the caps folder. It has been
saved in the header. We can use a similar way to support
"branch_counter_nr" and "branch_counter_width". It doesn't seem very
complex. I think I will try the sysfs method first.

Thanks,
Kan