From nobody Sun Feb 8 17:43:10 2026 Received: from mail-yb1-f201.google.com (mail-yb1-f201.google.com [209.85.219.201]) (using TLSv1.2 with cipher ECDHE-RSA-AES128-GCM-SHA256 (128/128 bits)) (No client certificate requested) by smtp.subspace.kernel.org (Postfix) with ESMTPS id 099EFBA41 for ; Sun, 28 Apr 2024 05:36:28 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; arc=none smtp.client-ip=209.85.219.201 ARC-Seal: i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1714282590; cv=none; b=gUEPs1qnNCgwzOwUNhb1WbrZeXeotWOxiYSZrs78/XTSQPCkcoTvdeMAp/f0/cESu4S/xB4K41o2d8l6fpZTyKIqs4sB7UsBWKbkrqhbUuC89uMaeesaML0Uah8RXRbGyns7aOThEmU4BdETDWzyc3O/6djElaxsmZVM/cKSylo= ARC-Message-Signature: i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1714282590; c=relaxed/simple; bh=ds3FTpAVnoEQPwjyBLlVVQYAf2s8bYy/yXDtibgv24Y=; h=Date:In-Reply-To:Message-Id:Mime-Version:References:Subject:From: To:Content-Type; b=dTKK60WqQCyVpMXEj0eoo3uCKss8Ac0+h5uK6YPgKRR0IaA5aMUxwGhkAMyS5t37R88D3TnYnqi9GsvexQVhhx+x8Uh0l5mrSQe7dD0Od4OQ1qOSMP6SM08f/pbbwOnyL49jDI95kj0sbIEfcYORVEf8UtlOHRpHzvHmLx/Csa4= ARC-Authentication-Results: i=1; smtp.subspace.kernel.org; dmarc=pass (p=reject dis=none) header.from=google.com; spf=pass smtp.mailfrom=flex--irogers.bounces.google.com; dkim=pass (2048-bit key) header.d=google.com header.i=@google.com header.b=h+QUlFSk; arc=none smtp.client-ip=209.85.219.201 Authentication-Results: smtp.subspace.kernel.org; dmarc=pass (p=reject dis=none) header.from=google.com Authentication-Results: smtp.subspace.kernel.org; spf=pass smtp.mailfrom=flex--irogers.bounces.google.com Authentication-Results: smtp.subspace.kernel.org; dkim=pass (2048-bit key) header.d=google.com header.i=@google.com header.b="h+QUlFSk" Received: by mail-yb1-f201.google.com with SMTP id 3f1490d57ef6-de45d0b7ffaso6577287276.2 for ; Sat, 27 Apr 2024 22:36:28 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=20230601; t=1714282588; x=1714887388; darn=vger.kernel.org; h=to:from:subject:references:mime-version:message-id:in-reply-to:date :from:to:cc:subject:date:message-id:reply-to; bh=JEMBdRiSYAzLnQF6W26QVCloinXfU92DZLk227BSKkk=; b=h+QUlFSkXNj/FTzkYofHlX4Mqx2McsxdsrFP+DlVlLonnd28OaQf+7vSk3VvQJNeQk ME8JaeC3T2g/rX45zfGnB0LFwZ/cPc/qzviBxoa5vqPfcm9BMUUB2A91tElv81VvI0xu tuWZ4GKw5tix4u87ogM5FZLPWnDksWpLzfvmLE5TBueSNobCIPpWRO5EVypPmQ5TSNA4 fgwILvILLX2CSU9jlIPuFZybGVBxdQpIRmvtKM/P+GiVpwtUyAXiFipMLoLHr9ihqA4Q Y/+uu8qlxOmrwOWMXfihfbEguuwk1jXxox+dF6NOlujriq6yW3FgsVxqLJo0esnW2nyX 7q7g== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20230601; t=1714282588; x=1714887388; h=to:from:subject:references:mime-version:message-id:in-reply-to:date :x-gm-message-state:from:to:cc:subject:date:message-id:reply-to; bh=JEMBdRiSYAzLnQF6W26QVCloinXfU92DZLk227BSKkk=; b=ezPT6PTLM37pHndb9KF8QuOF8l0Kgx+oRoWt6lBRnWY98VTMOi0ZC1/5F5LwSlP2NR LS5Hyw9AdMRmb1z7ejhpYgdmhgMb/Q38lSzYewoibjdu91CThhLCu0wnNCtByrsoKv5D MP8JdqbSy941njTAwJbaUgq3Vi++ZPhIZW3Sxc1GYxrO8en4flPeGJF9TGIzgK7aza5N RMKoJXvc3dXNkvSRqrc985SAlrD6IcIWAW/pwSH57wjwcZ8qWPJcFX/RMJvFNAsJT536 LR4OHTiQKIm42ewL6/VQxkNd0ox169gyc93Wug15OFZBFXJC3YlBHzOPZTlEKe44cP5T Xs1g== X-Forwarded-Encrypted: i=1; AJvYcCUU2chntkrmwhqpb5n3PrbjpNp7fzAoYjPUwInMikoa43TAQsh0B0jAO0mJMFQ/cpFY7Swx8SsnGE1lIasx1COdNT7OEfAhbDgVs+Sk X-Gm-Message-State: AOJu0YxnQYYQ86HkQpp/kmgVrcXsZE5SuqvRu+9tKP9nZlzmkGl/dcIi 9HorSgdHXbi2Xg6ZcZtImos4oegAY2+N2zOVlY7/ZaF2TG4KMnxvEMNihXGsGnydPxlSPJD5ZLR gUn0iFg== X-Google-Smtp-Source: AGHT+IHzqyKnSGf7Pk6gDt60ZrjYNxljd0CkcvR8nELq18HPFE76znx0vCTgauxHqqYDBN6fFbSAyL6qVedt X-Received: from irogers.svl.corp.google.com ([2620:15c:2a3:200:eaaa:2f79:c83c:9def]) (user=irogers job=sendgmr) by 2002:a25:aa63:0:b0:de4:654f:9ad0 with SMTP id s90-20020a25aa63000000b00de4654f9ad0mr702348ybi.6.1714282588141; Sat, 27 Apr 2024 22:36:28 -0700 (PDT) Date: Sat, 27 Apr 2024 22:36:15 -0700 In-Reply-To: <20240428053616.1125891-1-irogers@google.com> Message-Id: <20240428053616.1125891-4-irogers@google.com> Precedence: bulk X-Mailing-List: linux-kernel@vger.kernel.org List-Id: List-Subscribe: List-Unsubscribe: Mime-Version: 1.0 References: <20240428053616.1125891-1-irogers@google.com> X-Mailer: git-send-email 2.44.0.769.g3c40516874-goog Subject: [RFC PATCH v2 3/3] perf evsel: Add retirement latency event support From: Ian Rogers To: weilin.wang@intel.com, Peter Zijlstra , Ingo Molnar , Arnaldo Carvalho de Melo , Namhyung Kim , Mark Rutland , Alexander Shishkin , Jiri Olsa , Ian Rogers , Adrian Hunter , Kan Liang , Ze Gao , Leo Yan , Ravi Bangoria , Dmitrii Dolgov <9erthalion6@gmail.com>, Song Liu , James Clark , linux-perf-users@vger.kernel.org, linux-kernel@vger.kernel.org Content-Transfer-Encoding: quoted-printable Content-Type: text/plain; charset="utf-8" When a retirement latency event is processed it sets a flag on the evsel. This change makes it so that when the flag is set evsel opening, reading and exiting report values from child perf record and perf report processes. Something similar was suggested by Namhyung Kim in: https://lore.kernel.org/lkml/CAM9d7cgdQQn5GYB7t++xuoMdeqPXiEkkcop69+rD06RAn= u9-EQ@mail.gmail.com/ This is trying to add support for retirement latency directly in events rather than through metric changes, as suggested by Weilin Wang in: https://lore.kernel.org/lkml/20240402214436.1409476-1-weilin.wang@intel.com/ Signed-off-by: Ian Rogers --- tools/perf/util/evsel.c | 186 +++++++++++++++++++++++++++++++++++++++- tools/perf/util/evsel.h | 3 + 2 files changed, 188 insertions(+), 1 deletion(-) diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c index a0a8aee7d6b9..17c123cddde3 100644 --- a/tools/perf/util/evsel.c +++ b/tools/perf/util/evsel.c @@ -22,6 +22,7 @@ #include #include #include +#include #include #include #include "asm/bug.h" @@ -58,6 +59,7 @@ #include #include #include +#include =20 #include =20 @@ -493,6 +495,162 @@ struct evsel *evsel__newtp_idx(const char *sys, const= char *name, int idx) } #endif =20 +static int evsel__start_retire_latency_cpu(struct evsel *evsel, struct per= f_cpu_map *cpus, + int cpu_map_idx) +{ + char buf[16]; + int pipefd[2]; + int err, i; + struct perf_cpu cpu =3D perf_cpu_map__cpu(cpus, cpu_map_idx); + struct child_process *child_record =3D + xyarray__entry(evsel->children, cpu_map_idx, 0); + struct child_process *child_report =3D + xyarray__entry(evsel->children, cpu_map_idx, 1); + char *event =3D strdup(evsel__name(evsel)); + /* TODO: the dummy event also won't be used, but there's no option to dis= able. */ + const char *record_argv[15] =3D { + [0] =3D "perf", + [1] =3D "record", + [2] =3D "--synth=3Dno", + [3] =3D "--no-bpf-event", + [4] =3D "-W", + [5] =3D "-o", + [6] =3D "-", + [7] =3D "-e", + }; + const char *report_argv[] =3D { + [0] =3D "perf", + [1] =3D "report", + [2] =3D "-i", + [3] =3D "-", + [4] =3D "-q", + [5] =3D "-F", + [6] =3D "retire_lat", + NULL, + }; + + if (evsel->core.attr.sample_period) /* no sampling */ + return -EINVAL; + + if (!event) + return -ENOMEM; + + /* Change the R modifier to the maximum precision one. */ + for (i =3D strlen(event) - 1; i > 0; i--) { + if (event[i] =3D=3D 'R') + break; + if (event[i] =3D=3D ':' || event[i] =3D=3D '/') + i =3D 0; + } + if (i <=3D 0) { + pr_err("Expected retired latency 'R'\n"); + return -EINVAL; + } + event[i] =3D 'P'; + + i =3D 8; + record_argv[i++] =3D event; + if (verbose) { + record_argv[i++] =3D verbose > 1 ? "-vv" : "-v"; + } + if (cpu.cpu >=3D 0) { + record_argv[i++] =3D "-C"; + snprintf(buf, sizeof(buf), "%d", cpu.cpu); + record_argv[i++] =3D buf; + } else { + record_argv[i++] =3D "-a"; + } + /* TODO: use something like the control fds to control perf record behavi= or. */ + record_argv[i++] =3D "sleep"; + record_argv[i++] =3D "0.1"; + + if (pipe(pipefd) < 0) { + free(event); + return -errno; + } + + child_record->argv =3D record_argv; + child_record->pid =3D -1; + child_record->no_stdin =3D 1; + if (verbose) + child_record->err =3D fileno(stderr); + else + child_record->no_stderr =3D 1; + child_record->out =3D pipefd[1]; + err =3D start_command(child_record); + free(event); + if (err) + return err; + + child_report->argv =3D report_argv; + child_report->pid =3D -1; + if (verbose) + child_report->err =3D fileno(stderr); + else + child_report->no_stderr =3D 1; + child_report->in =3D pipefd[0]; + child_report->out =3D -1; + return start_command(child_report); +} + +static int evsel__finish_retire_latency_cpu(struct evsel *evsel, int cpu_m= ap_idx) +{ + struct child_process *child_record =3D + xyarray__entry(evsel->children, cpu_map_idx, 0); + struct child_process *child_report =3D + xyarray__entry(evsel->children, cpu_map_idx, 1); + + if (child_record->pid > 0) + finish_command(child_record); + if (child_report->pid > 0) + finish_command(child_report); + return 0; +} + +static int evsel__read_retire_latency(struct evsel *evsel, int cpu_map_idx= , int thread) +{ + struct child_process *child_record =3D xyarray__entry(evsel->children, cp= u_map_idx, 0); + struct child_process *child_report =3D xyarray__entry(evsel->children, cp= u_map_idx, 1); + struct perf_counts_values *count =3D perf_counts(evsel->counts, cpu_map_i= dx, thread); + char buf[256]; + int err; + + kill(child_record->pid, SIGTERM); + err =3D read(child_report->out, buf, sizeof(buf)); + if (err < 0 || strlen(buf) =3D=3D 0) + return -1; + + count->val =3D atoll(buf); + count->ena =3D 1; + count->run =3D 1; + count->id =3D 0; + count->lost =3D 0; + + /* + * TODO: The SIGCHLD from the children exiting will cause interval mode + * to stop, use the control fd to avoid this. + */ + err =3D evsel__finish_retire_latency_cpu(evsel, cpu_map_idx); + if (err) + return err; + + /* Restart the counter. */ + return evsel__start_retire_latency_cpu(evsel, evsel->core.cpus, cpu_map_i= dx); +} + +static int evsel__finish_retire_latency(struct evsel *evsel) +{ + int idx; + + perf_cpu_map__for_each_idx(idx, evsel->core.cpus) { + int err =3D evsel__finish_retire_latency_cpu(evsel, idx); + + if (err) + return err; + } + return 0; +} + const char *const evsel__hw_names[PERF_COUNT_HW_MAX] =3D { "cycles", "instructions", @@ -1465,6 +1623,10 @@ static void evsel__free_config_terms(struct evsel *e= vsel) =20 void evsel__exit(struct evsel *evsel) { + if (evsel->children) { + evsel__finish_retire_latency(evsel); + zfree(&evsel->children); + } assert(list_empty(&evsel->core.node)); assert(evsel->evlist =3D=3D NULL); bpf_counter__destroy(evsel); @@ -1778,6 +1940,9 @@ int evsel__read_counter(struct evsel *evsel, int cpu_= map_idx, int thread) if (evsel__is_tool(evsel)) return evsel__read_tool(evsel, cpu_map_idx, thread); =20 + if (evsel->retire_lat) + return evsel__read_retire_latency(evsel, cpu_map_idx, thread); + if (evsel->core.attr.read_format & PERF_FORMAT_GROUP) return evsel__read_group(evsel, cpu_map_idx, thread); =20 @@ -1993,10 +2158,22 @@ static int __evsel__prepare_open(struct evsel *evse= l, struct perf_cpu_map *cpus, threads =3D empty_thread_map; } =20 - if (evsel->core.fd =3D=3D NULL && + if (!evsel->retire_lat && evsel->core.fd =3D=3D NULL && perf_evsel__alloc_fd(&evsel->core, perf_cpu_map__nr(cpus), nthreads) = < 0) return -ENOMEM; =20 + if (evsel->retire_lat && evsel->children =3D=3D NULL) { + /* + * Use ylen of 2, [0] is the record and [1] is the report + * command. Currently retirement latency doesn't support + * per-thread mode. + */ + evsel->children =3D xyarray__new(perf_cpu_map__nr(cpus), /*ylen=3D*/2, + sizeof(struct child_process)); + if (!evsel->children) + return -ENOMEM; + } + evsel->open_flags =3D PERF_FLAG_FD_CLOEXEC; if (evsel->cgrp) evsel->open_flags |=3D PERF_FLAG_PID_CGROUP; @@ -2209,6 +2386,13 @@ static int evsel__open_cpu(struct evsel *evsel, stru= ct perf_cpu_map *cpus, =20 for (idx =3D start_cpu_map_idx; idx < end_cpu_map_idx; idx++) { =20 + if (evsel->retire_lat) { + err =3D evsel__start_retire_latency_cpu(evsel, cpus, idx); + if (err) + goto out_close; + continue; + } + for (thread =3D 0; thread < nthreads; thread++) { int fd, group_fd; retry_open: diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h index bd8e84954e34..3c0806436e64 100644 --- a/tools/perf/util/evsel.h +++ b/tools/perf/util/evsel.h @@ -177,6 +177,9 @@ struct evsel { __u64 start_time; /* Is the tool's fd for /proc/pid/stat or /proc/stat. */ bool pid_stat; + + /* Used for retire_lat child process. */ + struct xyarray *children; }; =20 struct perf_missing_features { --=20 2.44.0.769.g3c40516874-goog