Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::3:8 as permitted sender) client-ip=2620:137:e000::3:8;
Message-ID: <86df1293-c20b-4292-abde-852861dcedf1@linux.intel.com>
Date:   Thu, 19 Oct 2023 09:56:02 -0400
MIME-Version: 1.0
User-Agent: Mozilla Thunderbird
Subject: Re: [PATCH V4 4/7] perf/x86/intel: Support LBR event logging
To:     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, ak@linux.intel.com, eranian@google.com,
        alexey.v.bayduraev@linux.intel.com, tinghao.zhang@intel.com
References: <20231004184044.3062788-1-kan.liang@linux.intel.com>
 <20231004184044.3062788-4-kan.liang@linux.intel.com>
 <20231019092341.GE36211@noisy.programming.kicks-ass.net>
Content-Language: en-US
From:   "Liang, Kan" <kan.liang@linux.intel.com>
In-Reply-To: <20231019092341.GE36211@noisy.programming.kicks-ass.net>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit
Precedence: bulk


On 2023-10-19 5:23 a.m., Peter Zijlstra wrote:
> On Wed, Oct 04, 2023 at 11:40:41AM -0700, kan.liang@linux.intel.com wrote:
> 
>> diff --git a/arch/x86/events/intel/lbr.c b/arch/x86/events/intel/lbr.c
>> index c3b0d15a9841..1e80a551a4c2 100644
>> --- a/arch/x86/events/intel/lbr.c
>> +++ b/arch/x86/events/intel/lbr.c
>> @@ -676,6 +676,21 @@ void intel_pmu_lbr_del(struct perf_event *event)
>>  	WARN_ON_ONCE(cpuc->lbr_users < 0);
>>  	WARN_ON_ONCE(cpuc->lbr_pebs_users < 0);
>>  	perf_sched_cb_dec(event->pmu);
>> +
>> +	/*
>> +	 * The logged occurrences information is only valid for the
>> +	 * current LBR group. If another LBR group is scheduled in
>> +	 * later, the information from the stale LBRs will be wrongly
>> +	 * interpreted. Reset the LBRs here.
>> +	 * For the context switch, the LBR will be unconditionally
>> +	 * flushed when a new task is scheduled in. If both the new task
>> +	 * and the old task are monitored by a LBR event group. The
>> +	 * reset here is redundant. But the extra reset doesn't impact
>> +	 * the functionality. It's hard to distinguish the above case.
>> +	 * Keep the unconditionally reset for a LBR event group for now.
>> +	 */
> 
> I found this really hard to read, also should this not rely on
> !cpuc->lbr_users ?
>

It's possible that the last LBR user is not in the branch_counters
group, e.g., a branch_counters group + several normal LBR events.
For this case, the is_branch_counters_group(event) return false for the
last LBR user. The LBR will not be reset.

> As is, you'll reset the lbr for every event in the group.
> 
>> +	if (is_branch_counters_group(event))
>> +		intel_pmu_lbr_reset();
>>  }

Right, I forgot to change it after I modified flag. :(

Here I think we should only clear the LBRs once for a branch_counters
group, e.g., in the leader event.

+	if (is_branch_counters_group(event) && event == event->group_leader)+	
intel_pmu_lbr_reset();

The only problem is that the leader event may not be an LBR event. But I
guess it should be OK to limit that the leader event of a
branch_counters group must be an LBR event in hw_config().

Thanks,
Kan