From nobody Fri Feb 13 22:36:10 2026 Received: from foss.arm.com (foss.arm.com [217.140.110.172]) by smtp.subspace.kernel.org (Postfix) with ESMTP id 097D47EEF8; Tue, 21 May 2024 13:30:39 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; arc=none smtp.client-ip=217.140.110.172 ARC-Seal: i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1716298241; cv=none; b=ebADya0aCrQG9U2K7whx6ZElyiDUGWEyB1JONPwFjhm3oUmVkPLk4zWe28whA928nVne18ezU6bb5s7y1MVzF3FjNjLnOxgZt0ZoYxJemqgq0z4coENPC///5uGNSw7MhSaC0MprB0eNhs4xHPwLtYs4k0u5Ef1PmeUvm/OnzrE= ARC-Message-Signature: i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1716298241; c=relaxed/simple; bh=7GXDVQXA+VImFAKLoHD6wQq7/l1ZtBKXfUB9v9jbp/M=; h=From:To:Cc:Subject:Date:Message-ID:In-Reply-To:References: MIME-Version; b=kpHh4EsO2rpd84ro1TyEU2KdtAnmxTFRsxZEBNA8lY2FB3NEmGUYGRzbNIvU7FgS5I1WtAGCBOZyLNpEBenx7KXay3z4SY0JLPJjEQKXMF6hou16Xa9QLwLF2iyQHj6k6mPbo8Mb3uY+gO540BA7WJExZwILSq/PeDlp8Kk1RUg= ARC-Authentication-Results: i=1; smtp.subspace.kernel.org; dmarc=pass (p=none dis=none) header.from=arm.com; spf=pass smtp.mailfrom=arm.com; arc=none smtp.client-ip=217.140.110.172 Authentication-Results: smtp.subspace.kernel.org; dmarc=pass (p=none dis=none) header.from=arm.com Authentication-Results: smtp.subspace.kernel.org; spf=pass smtp.mailfrom=arm.com Received: from usa-sjc-imap-foss1.foss.arm.com (unknown [10.121.207.14]) by usa-sjc-mx-foss1.foss.arm.com (Postfix) with ESMTP id 1E661FEC; Tue, 21 May 2024 06:31:03 -0700 (PDT) Received: from e126817.cambridge.arm.com (e126817.cambridge.arm.com [10.2.3.5]) by usa-sjc-imap-foss1.foss.arm.com (Postfix) with ESMTPA id 602FA3F641; Tue, 21 May 2024 06:30:37 -0700 (PDT) From: Ben Gainey To: peterz@infradead.org, mingo@redhat.com, acme@kernel.org, namhyung@kernel.org Cc: james.clark@arm.com, mark.rutland@arm.com, alexander.shishkin@linux.intel.com, jolsa@kernel.org, irogers@google.com, adrian.hunter@intel.com, linux-perf-users@vger.kernel.org, linux-kernel@vger.kernel.org, Ben Gainey Subject: [PATCH v6 1/4] perf: Support PERF_SAMPLE_READ with inherit Date: Tue, 21 May 2024 14:30:26 +0100 Message-ID: <20240521133029.83654-2-ben.gainey@arm.com> X-Mailer: git-send-email 2.45.1 In-Reply-To: <20240521133029.83654-1-ben.gainey@arm.com> References: <20240521133029.83654-1-ben.gainey@arm.com> Precedence: bulk X-Mailing-List: linux-kernel@vger.kernel.org List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 Content-Transfer-Encoding: quoted-printable Content-Type: text/plain; charset="utf-8" This change allows events to use PERF_SAMPLE READ with inherit so long as PERF_SAMPLE_TID is also set. In this configuration, an event will be inherited into any child processes / threads, allowing convenient profiling of a multiprocess or multithreaded application, whilst allowing profiling tools to collect per-thread samples, in particular of groups of counters. The read_format field of both PERF_RECORD_READ and PERF_RECORD_SAMPLE are changed by this new configuration, but calls to `read()` on the same event file descriptor are unaffected and continue to return the cumulative total. Signed-off-by: Ben Gainey Acked-by: Namhyung Kim --- include/linux/perf_event.h | 1 + kernel/events/core.c | 78 ++++++++++++++++++++++++++++---------- 2 files changed, 58 insertions(+), 21 deletions(-) diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index d2a15c0c6f8a9..e7eed33c50f16 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -932,6 +932,7 @@ struct perf_event_context { =20 int nr_task_data; int nr_stat; + int nr_inherit_read; int nr_freq; int rotate_disable; =20 diff --git a/kernel/events/core.c b/kernel/events/core.c index 724e6d7e128f3..e7c847956ebff 100644 --- a/kernel/events/core.c +++ b/kernel/events/core.c @@ -1767,6 +1767,14 @@ perf_event_groups_next(struct perf_event *event, str= uct pmu *pmu) event =3D rb_entry_safe(rb_next(&event->group_node), \ typeof(*event), group_node)) =20 +/* + * Does the event attribute request inherit with PERF_SAMPLE_READ + */ +static inline bool has_inherit_and_sample_read(struct perf_event_attr *att= r) +{ + return attr->inherit && (attr->sample_type & PERF_SAMPLE_READ); +} + /* * Add an event from the lists for its context. * Must be called with ctx->mutex and ctx->lock held. @@ -1797,6 +1805,8 @@ list_add_event(struct perf_event *event, struct perf_= event_context *ctx) ctx->nr_user++; if (event->attr.inherit_stat) ctx->nr_stat++; + if (has_inherit_and_sample_read(&event->attr)) + ctx->nr_inherit_read++; =20 if (event->state > PERF_EVENT_STATE_OFF) perf_cgroup_event_enable(event, ctx); @@ -2021,6 +2031,8 @@ list_del_event(struct perf_event *event, struct perf_= event_context *ctx) ctx->nr_user--; if (event->attr.inherit_stat) ctx->nr_stat--; + if (has_inherit_and_sample_read(&event->attr)) + ctx->nr_inherit_read--; =20 list_del_rcu(&event->event_entry); =20 @@ -3529,11 +3541,18 @@ perf_event_context_sched_out(struct task_struct *ta= sk, struct task_struct *next) perf_ctx_disable(ctx, false); =20 /* PMIs are disabled; ctx->nr_pending is stable. */ - if (local_read(&ctx->nr_pending) || + if (ctx->nr_inherit_read || + next_ctx->nr_inherit_read || + local_read(&ctx->nr_pending) || local_read(&next_ctx->nr_pending)) { /* * Must not swap out ctx when there's pending * events that rely on the ctx->task relation. + * + * Likewise, when a context contains inherit + + * SAMPLE_READ events they should be switched + * out using the slow path so that they are + * treated as if they were distinct contexts. */ raw_spin_unlock(&next_ctx->lock); rcu_read_unlock(); @@ -4533,11 +4552,19 @@ static void __perf_event_read(void *info) raw_spin_unlock(&ctx->lock); } =20 -static inline u64 perf_event_count(struct perf_event *event) +static inline u64 perf_event_count_cumulative(struct perf_event *event) { return local64_read(&event->count) + atomic64_read(&event->child_count); } =20 +static inline u64 perf_event_count(struct perf_event *event, bool self_val= ue_only) +{ + if (self_value_only && has_inherit_and_sample_read(&event->attr)) + return local64_read(&event->count); + + return perf_event_count_cumulative(event); +} + static void calc_timer_values(struct perf_event *event, u64 *now, u64 *enabled, @@ -5454,7 +5481,7 @@ static u64 __perf_event_read_value(struct perf_event = *event, u64 *enabled, u64 * mutex_lock(&event->child_mutex); =20 (void)perf_event_read(event, false); - total +=3D perf_event_count(event); + total +=3D perf_event_count_cumulative(event); =20 *enabled +=3D event->total_time_enabled + atomic64_read(&event->child_total_time_enabled); @@ -5463,7 +5490,7 @@ static u64 __perf_event_read_value(struct perf_event = *event, u64 *enabled, u64 * =20 list_for_each_entry(child, &event->child_list, child_list) { (void)perf_event_read(child, false); - total +=3D perf_event_count(child); + total +=3D perf_event_count_cumulative(child); *enabled +=3D child->total_time_enabled; *running +=3D child->total_time_running; } @@ -5545,14 +5572,14 @@ static int __perf_read_group_add(struct perf_event = *leader, /* * Write {count,id} tuples for every sibling. */ - values[n++] +=3D perf_event_count(leader); + values[n++] +=3D perf_event_count_cumulative(leader); if (read_format & PERF_FORMAT_ID) values[n++] =3D primary_event_id(leader); if (read_format & PERF_FORMAT_LOST) values[n++] =3D atomic64_read(&leader->lost_samples); =20 for_each_sibling_event(sub, leader) { - values[n++] +=3D perf_event_count(sub); + values[n++] +=3D perf_event_count_cumulative(sub); if (read_format & PERF_FORMAT_ID) values[n++] =3D primary_event_id(sub); if (read_format & PERF_FORMAT_LOST) @@ -6132,7 +6159,7 @@ void perf_event_update_userpage(struct perf_event *ev= ent) ++userpg->lock; barrier(); userpg->index =3D perf_event_index(event); - userpg->offset =3D perf_event_count(event); + userpg->offset =3D perf_event_count_cumulative(event); if (userpg->index) userpg->offset -=3D local64_read(&event->hw.prev_count); =20 @@ -7194,13 +7221,14 @@ void perf_event__output_id_sample(struct perf_event= *event, =20 static void perf_output_read_one(struct perf_output_handle *handle, struct perf_event *event, - u64 enabled, u64 running) + u64 enabled, u64 running, + bool from_sample) { u64 read_format =3D event->attr.read_format; u64 values[5]; int n =3D 0; =20 - values[n++] =3D perf_event_count(event); + values[n++] =3D perf_event_count(event, from_sample); if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) { values[n++] =3D enabled + atomic64_read(&event->child_total_time_enabled); @@ -7218,8 +7246,9 @@ static void perf_output_read_one(struct perf_output_h= andle *handle, } =20 static void perf_output_read_group(struct perf_output_handle *handle, - struct perf_event *event, - u64 enabled, u64 running) + struct perf_event *event, + u64 enabled, u64 running, + bool from_sample) { struct perf_event *leader =3D event->group_leader, *sub; u64 read_format =3D event->attr.read_format; @@ -7245,7 +7274,7 @@ static void perf_output_read_group(struct perf_output= _handle *handle, (leader->state =3D=3D PERF_EVENT_STATE_ACTIVE)) leader->pmu->read(leader); =20 - values[n++] =3D perf_event_count(leader); + values[n++] =3D perf_event_count(leader, from_sample); if (read_format & PERF_FORMAT_ID) values[n++] =3D primary_event_id(leader); if (read_format & PERF_FORMAT_LOST) @@ -7260,7 +7289,7 @@ static void perf_output_read_group(struct perf_output= _handle *handle, (sub->state =3D=3D PERF_EVENT_STATE_ACTIVE)) sub->pmu->read(sub); =20 - values[n++] =3D perf_event_count(sub); + values[n++] =3D perf_event_count(sub, from_sample); if (read_format & PERF_FORMAT_ID) values[n++] =3D primary_event_id(sub); if (read_format & PERF_FORMAT_LOST) @@ -7281,9 +7310,14 @@ static void perf_output_read_group(struct perf_outpu= t_handle *handle, * The problem is that its both hard and excessively expensive to iterate = the * child list, not to mention that its impossible to IPI the children runn= ing * on another CPU, from interrupt/NMI context. + * + * Instead the combination of PERF_SAMPLE_READ and inherit will track per-= thread + * counts rather than attempting to accumulate some value across all child= ren on + * all cores. */ static void perf_output_read(struct perf_output_handle *handle, - struct perf_event *event) + struct perf_event *event, + bool from_sample) { u64 enabled =3D 0, running =3D 0, now; u64 read_format =3D event->attr.read_format; @@ -7301,9 +7335,9 @@ static void perf_output_read(struct perf_output_handl= e *handle, calc_timer_values(event, &now, &enabled, &running); =20 if (event->attr.read_format & PERF_FORMAT_GROUP) - perf_output_read_group(handle, event, enabled, running); + perf_output_read_group(handle, event, enabled, running, from_sample); else - perf_output_read_one(handle, event, enabled, running); + perf_output_read_one(handle, event, enabled, running, from_sample); } =20 void perf_output_sample(struct perf_output_handle *handle, @@ -7343,7 +7377,7 @@ void perf_output_sample(struct perf_output_handle *ha= ndle, perf_output_put(handle, data->period); =20 if (sample_type & PERF_SAMPLE_READ) - perf_output_read(handle, event); + perf_output_read(handle, event, true); =20 if (sample_type & PERF_SAMPLE_CALLCHAIN) { int size =3D 1; @@ -7944,7 +7978,7 @@ perf_event_read_event(struct perf_event *event, return; =20 perf_output_put(&handle, read_event); - perf_output_read(&handle, event); + perf_output_read(&handle, event, false); perf_event__output_id_sample(event, &handle, &sample); =20 perf_output_end(&handle); @@ -12006,10 +12040,12 @@ perf_event_alloc(struct perf_event_attr *attr, in= t cpu, local64_set(&hwc->period_left, hwc->sample_period); =20 /* - * We currently do not support PERF_SAMPLE_READ on inherited events. + * We do not support PERF_SAMPLE_READ on inherited events unless + * PERF_SAMPLE_TID is also selected, which allows inherited events to + * collect per-thread samples. * See perf_output_read(). */ - if (attr->inherit && (attr->sample_type & PERF_SAMPLE_READ)) + if (has_inherit_and_sample_read(attr) && !(attr->sample_type & PERF_SAMPL= E_TID)) goto err_ns; =20 if (!has_branch_stack(event)) @@ -13033,7 +13069,7 @@ static void sync_child_event(struct perf_event *chi= ld_event) perf_event_read_event(child_event, task); } =20 - child_val =3D perf_event_count(child_event); + child_val =3D perf_event_count_cumulative(child_event); =20 /* * Add back the child's count to the parent's count: --=20 2.45.1 From nobody Fri Feb 13 22:36:10 2026 Received: from foss.arm.com (foss.arm.com [217.140.110.172]) by smtp.subspace.kernel.org (Postfix) with ESMTP id 9F78C7F482; Tue, 21 May 2024 13:30:41 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; arc=none smtp.client-ip=217.140.110.172 ARC-Seal: i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1716298244; cv=none; b=WlIm3BkQ31fVfCiTxhd5rk88c8/qbqgpFVj3zYA6hcgvmzvO1zgeQJbWElqwUgZOrekreowISBxEgnciV61rjbgkF47GhoxTyCkZV1M4ATyDbGpPHwMzn3/9QJIgyCfz4gtrzG04Ge0UiHVHQbFryHCP2pj6/am6BBUtoP2RJs8= ARC-Message-Signature: i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1716298244; c=relaxed/simple; bh=9KsvWJ0unbJTziuYOcxjeYSAUdeTmy5OO9IsWvSeBtk=; h=From:To:Cc:Subject:Date:Message-ID:In-Reply-To:References: MIME-Version; b=FOK+KA+LDEcnejm68NypUtYSRn1EWXbxF8gYyX2CSKWAQzQYUVycF/xR1RteQmHAMUBF9tQIyGT4yvxxTa23mXLArVYuqFd84/Yw62PXoWKpBZEE7maYlnhfJNuxx7my4Qxb54K+D51a2l41LWTtdR18qN7dnvAjbDTVVGUT3JA= ARC-Authentication-Results: i=1; smtp.subspace.kernel.org; dmarc=pass (p=none dis=none) header.from=arm.com; spf=pass smtp.mailfrom=arm.com; arc=none smtp.client-ip=217.140.110.172 Authentication-Results: smtp.subspace.kernel.org; dmarc=pass (p=none dis=none) header.from=arm.com Authentication-Results: smtp.subspace.kernel.org; spf=pass smtp.mailfrom=arm.com Received: from usa-sjc-imap-foss1.foss.arm.com (unknown [10.121.207.14]) by usa-sjc-mx-foss1.foss.arm.com (Postfix) with ESMTP id 0C83B1042; Tue, 21 May 2024 06:31:05 -0700 (PDT) Received: from e126817.cambridge.arm.com (e126817.cambridge.arm.com [10.2.3.5]) by usa-sjc-imap-foss1.foss.arm.com (Postfix) with ESMTPA id 4F2003F641; Tue, 21 May 2024 06:30:39 -0700 (PDT) From: Ben Gainey To: peterz@infradead.org, mingo@redhat.com, acme@kernel.org, namhyung@kernel.org Cc: james.clark@arm.com, mark.rutland@arm.com, alexander.shishkin@linux.intel.com, jolsa@kernel.org, irogers@google.com, adrian.hunter@intel.com, linux-perf-users@vger.kernel.org, linux-kernel@vger.kernel.org, Ben Gainey Subject: [PATCH v6 2/4] tools/perf: Track where perf_sample_ids need per-thread periods Date: Tue, 21 May 2024 14:30:27 +0100 Message-ID: <20240521133029.83654-3-ben.gainey@arm.com> X-Mailer: git-send-email 2.45.1 In-Reply-To: <20240521133029.83654-1-ben.gainey@arm.com> References: <20240521133029.83654-1-ben.gainey@arm.com> Precedence: bulk X-Mailing-List: linux-kernel@vger.kernel.org List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 Content-Transfer-Encoding: quoted-printable Content-Type: text/plain; charset="utf-8" perf_sample_ids and related evlist/evsel code are modified to track which events combine inherit+PERF_SAMPLE_READ+PERF_SAMPLE_TID. Events with this combination of properties must be handled differently when calculating each sample period. They must use the combination of (ID + TID) to uniquely identify each distinct sequence of values. Signed-off-by: Ben Gainey --- tools/lib/perf/evlist.c | 1 + tools/lib/perf/evsel.c | 7 +++++++ tools/lib/perf/include/internal/evsel.h | 9 +++++++++ 3 files changed, 17 insertions(+) diff --git a/tools/lib/perf/evlist.c b/tools/lib/perf/evlist.c index c6d67fc9e57ef..d17288eeaee48 100644 --- a/tools/lib/perf/evlist.c +++ b/tools/lib/perf/evlist.c @@ -255,6 +255,7 @@ static void perf_evlist__id_hash(struct perf_evlist *ev= list, =20 sid->id =3D id; sid->evsel =3D evsel; + sid->period_per_thread =3D perf_evsel__attr_has_per_thread_sample_period(= evsel); hash =3D hash_64(sid->id, PERF_EVLIST__HLIST_BITS); hlist_add_head(&sid->node, &evlist->heads[hash]); } diff --git a/tools/lib/perf/evsel.c b/tools/lib/perf/evsel.c index c07160953224a..f7abb879f416e 100644 --- a/tools/lib/perf/evsel.c +++ b/tools/lib/perf/evsel.c @@ -537,6 +537,13 @@ void perf_evsel__free_id(struct perf_evsel *evsel) evsel->ids =3D 0; } =20 +bool perf_evsel__attr_has_per_thread_sample_period(struct perf_evsel *evse= l) +{ + return (evsel->attr.sample_type & PERF_SAMPLE_READ) + && (evsel->attr.sample_type & PERF_SAMPLE_TID) + && evsel->attr.inherit; +} + void perf_counts_values__scale(struct perf_counts_values *count, bool scale, __s8 *pscaled) { diff --git a/tools/lib/perf/include/internal/evsel.h b/tools/lib/perf/inclu= de/internal/evsel.h index 5cd220a61962e..f8de2bf89c76b 100644 --- a/tools/lib/perf/include/internal/evsel.h +++ b/tools/lib/perf/include/internal/evsel.h @@ -36,6 +36,13 @@ struct perf_sample_id { =20 /* Holds total ID period value for PERF_SAMPLE_READ processing. */ u64 period; + + /* + * When inherit is combined with PERF_SAMPLE_READ, the period value is + * per (id, thread) tuple, rather than per id, so use the stream_id to + * uniquely identify the period, rather than the id. + */ + bool period_per_thread; }; =20 struct perf_evsel { @@ -88,4 +95,6 @@ int perf_evsel__apply_filter(struct perf_evsel *evsel, co= nst char *filter); int perf_evsel__alloc_id(struct perf_evsel *evsel, int ncpus, int nthreads= ); void perf_evsel__free_id(struct perf_evsel *evsel); =20 +bool perf_evsel__attr_has_per_thread_sample_period(struct perf_evsel *evse= l); + #endif /* __LIBPERF_INTERNAL_EVSEL_H */ --=20 2.45.1 From nobody Fri Feb 13 22:36:10 2026 Received: from foss.arm.com (foss.arm.com [217.140.110.172]) by smtp.subspace.kernel.org (Postfix) with ESMTP id 48B5F7F48A; Tue, 21 May 2024 13:30:43 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; arc=none smtp.client-ip=217.140.110.172 ARC-Seal: i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1716298244; cv=none; b=Oz2JC7mdinlMOo6E/LB5YQ0BIlKp1hx1ITgemmpUP3Z8LtxwHfNXfNvLjgHKdht65evk9ut7v3AI8ckzzEqNqT8yT/Utvb0fAA7cfBWqGfh6Ptpb630xSELBTo7KL5ZCNOrB6O5NIoBBqVk10O4j3WX+tbOSuoclnya9Ya9PrbI= ARC-Message-Signature: i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1716298244; c=relaxed/simple; bh=WSvbWOCjSUrN3vc/yaZiD0nKvntc6NdAjQ0Ub1rmUUk=; h=From:To:Cc:Subject:Date:Message-ID:In-Reply-To:References: MIME-Version; b=YkEfX6YUfiYuKqAKz1KU2sju4POksYpBtxpZZ6e3MnVVaRyho21JBuyO0ue66lXynlagFLspsKcxptg77eQddbX88LxGLD+EI8cYAWXU6A0CCz1xGQdZrZ7wF5OuKvBF8rZy08cINGKZdlvFjQRBNDp3aMMVjdYuKJJjXjUBkwA= ARC-Authentication-Results: i=1; smtp.subspace.kernel.org; dmarc=pass (p=none dis=none) header.from=arm.com; spf=pass smtp.mailfrom=arm.com; arc=none smtp.client-ip=217.140.110.172 Authentication-Results: smtp.subspace.kernel.org; dmarc=pass (p=none dis=none) header.from=arm.com Authentication-Results: smtp.subspace.kernel.org; spf=pass smtp.mailfrom=arm.com Received: from usa-sjc-imap-foss1.foss.arm.com (unknown [10.121.207.14]) by usa-sjc-mx-foss1.foss.arm.com (Postfix) with ESMTP id EF50ADA7; Tue, 21 May 2024 06:31:06 -0700 (PDT) Received: from e126817.cambridge.arm.com (e126817.cambridge.arm.com [10.2.3.5]) by usa-sjc-imap-foss1.foss.arm.com (Postfix) with ESMTPA id 3D9EF3F641; Tue, 21 May 2024 06:30:41 -0700 (PDT) From: Ben Gainey To: peterz@infradead.org, mingo@redhat.com, acme@kernel.org, namhyung@kernel.org Cc: james.clark@arm.com, mark.rutland@arm.com, alexander.shishkin@linux.intel.com, jolsa@kernel.org, irogers@google.com, adrian.hunter@intel.com, linux-perf-users@vger.kernel.org, linux-kernel@vger.kernel.org, Ben Gainey Subject: [PATCH v6 3/4] tools/perf: Correctly calculate sample period for inherited SAMPLE_READ values Date: Tue, 21 May 2024 14:30:28 +0100 Message-ID: <20240521133029.83654-4-ben.gainey@arm.com> X-Mailer: git-send-email 2.45.1 In-Reply-To: <20240521133029.83654-1-ben.gainey@arm.com> References: <20240521133029.83654-1-ben.gainey@arm.com> Precedence: bulk X-Mailing-List: linux-kernel@vger.kernel.org List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 Content-Transfer-Encoding: quoted-printable Content-Type: text/plain; charset="utf-8" Sample period calculation is updated to take into account the fact that events with inherit+PERF_SAMPLE_READ+PERF_SAMPLE_TID should use the TID in combination with the ID field in the read_format data to identify which value represents the previous accumulated counter total used= to calculate the period delta since the last sample. perf_sample_id is modified to support tracking per-thread values, along with the existing global per-id values. In the per-thread case, values are stored in a hash by TID within the perf_sample_id, and are dynamically allocated as the number is not known ahead of time. deliver_sample_value is modified to correctly locate the previous sample storage based on the attribute, stream id and thread id. Signed-off-by: Ben Gainey --- tools/lib/perf/evsel.c | 41 ++++++++++++++++++++++ tools/lib/perf/include/internal/evsel.h | 45 +++++++++++++++++++++++-- tools/perf/util/session.c | 11 ++++-- 3 files changed, 92 insertions(+), 5 deletions(-) diff --git a/tools/lib/perf/evsel.c b/tools/lib/perf/evsel.c index f7abb879f416e..26f3d7ba0f267 100644 --- a/tools/lib/perf/evsel.c +++ b/tools/lib/perf/evsel.c @@ -5,6 +5,7 @@ #include #include #include +#include #include #include #include @@ -23,6 +24,7 @@ void perf_evsel__init(struct perf_evsel *evsel, struct pe= rf_event_attr *attr, int idx) { INIT_LIST_HEAD(&evsel->node); + INIT_LIST_HEAD(&evsel->per_stream_periods); evsel->attr =3D *attr; evsel->idx =3D idx; evsel->leader =3D evsel; @@ -531,10 +533,17 @@ int perf_evsel__alloc_id(struct perf_evsel *evsel, in= t ncpus, int nthreads) =20 void perf_evsel__free_id(struct perf_evsel *evsel) { + struct perf_sample_id_period *pos, *n; + xyarray__delete(evsel->sample_id); evsel->sample_id =3D NULL; zfree(&evsel->id); evsel->ids =3D 0; + + perf_evsel_for_each_per_thread_period_safe(evsel, n, pos) { + list_del_init(&pos->node); + free(pos); + } } =20 bool perf_evsel__attr_has_per_thread_sample_period(struct perf_evsel *evse= l) @@ -544,6 +553,38 @@ bool perf_evsel__attr_has_per_thread_sample_period(str= uct perf_evsel *evsel) && evsel->attr.inherit; } =20 +u64 *perf_sample_id__get_period_storage(struct perf_sample_id *sid, u32 ti= d) +{ + struct hlist_head *head; + struct perf_sample_id_period *res; + int hash; + + if (!sid->period_per_thread) + return &sid->period; + + hash =3D hash_32(tid, PERF_SAMPLE_ID__HLIST_BITS); + head =3D &sid->periods[hash]; + + hlist_for_each_entry(res, head, hnode) + if (res->tid =3D=3D tid) + return &res->period; + + if (sid->evsel =3D=3D NULL) + return NULL; + + res =3D zalloc(sizeof(struct perf_sample_id_period)); + if (res =3D=3D NULL) + return NULL; + + INIT_LIST_HEAD(&res->node); + res->tid =3D tid; + + list_add_tail(&res->node, &sid->evsel->per_stream_periods); + hlist_add_head(&res->hnode, &sid->periods[hash]); + + return &res->period; +} + void perf_counts_values__scale(struct perf_counts_values *count, bool scale, __s8 *pscaled) { diff --git a/tools/lib/perf/include/internal/evsel.h b/tools/lib/perf/inclu= de/internal/evsel.h index f8de2bf89c76b..797dc9d782544 100644 --- a/tools/lib/perf/include/internal/evsel.h +++ b/tools/lib/perf/include/internal/evsel.h @@ -11,6 +11,32 @@ struct perf_thread_map; struct xyarray; =20 +/** + * The per-thread accumulated period storage node. + */ +struct perf_sample_id_period { + struct list_head node; + struct hlist_node hnode; + /* Holds total ID period value for PERF_SAMPLE_READ processing. */ + u64 period; + /* The TID that the values belongs to */ + u32 tid; +}; + +/** + * perf_evsel_for_each_per_thread_period_safe - safely iterate thru all the + * per_stream_periods + * @evlist:perf_evsel instance to iterate + * @item: struct perf_sample_id_period iterator + * @tmp: struct perf_sample_id_period temp iterator + */ +#define perf_evsel_for_each_per_thread_period_safe(evsel, tmp, item) \ + list_for_each_entry_safe(item, tmp, &(evsel)->per_stream_periods, node) + + +#define PERF_SAMPLE_ID__HLIST_BITS 4 +#define PERF_SAMPLE_ID__HLIST_SIZE (1 << PERF_SAMPLE_ID__HLIST_BITS) + /* * Per fd, to map back from PERF_SAMPLE_ID to evsel, only used when there = are * more than one entry in the evlist. @@ -34,8 +60,18 @@ struct perf_sample_id { pid_t machine_pid; struct perf_cpu vcpu; =20 - /* Holds total ID period value for PERF_SAMPLE_READ processing. */ - u64 period; + union { + /* + * Holds total ID period value for PERF_SAMPLE_READ processing + * (when period is not per-thread). + */ + u64 period; + /* + * Holds total ID period value for PERF_SAMPLE_READ processing + * (when period is per-thread). + */ + struct hlist_head periods[PERF_SAMPLE_ID__HLIST_SIZE]; + }; =20 /* * When inherit is combined with PERF_SAMPLE_READ, the period value is @@ -65,6 +101,9 @@ struct perf_evsel { u32 ids; struct perf_evsel *leader; =20 + /* Where period_per_thread is true, stores the per-thread values */ + struct list_head per_stream_periods; + /* parse modifier helper */ int nr_members; /* @@ -97,4 +136,6 @@ void perf_evsel__free_id(struct perf_evsel *evsel); =20 bool perf_evsel__attr_has_per_thread_sample_period(struct perf_evsel *evse= l); =20 +u64 *perf_sample_id__get_period_storage(struct perf_sample_id *sid, u32 ti= d); + #endif /* __LIBPERF_INTERNAL_EVSEL_H */ diff --git a/tools/perf/util/session.c b/tools/perf/util/session.c index 06d0bd7fb4599..2c66b730780ab 100644 --- a/tools/perf/util/session.c +++ b/tools/perf/util/session.c @@ -1478,14 +1478,19 @@ static int deliver_sample_value(struct evlist *evli= st, { struct perf_sample_id *sid =3D evlist__id2sid(evlist, v->id); struct evsel *evsel; + u64 *storage =3D NULL; =20 if (sid) { + storage =3D perf_sample_id__get_period_storage(sid, sample->tid); + } + + if (storage) { sample->id =3D v->id; - sample->period =3D v->value - sid->period; - sid->period =3D v->value; + sample->period =3D v->value - *storage; + *storage =3D v->value; } =20 - if (!sid || sid->evsel =3D=3D NULL) { + if (!storage || sid->evsel =3D=3D NULL) { ++evlist->stats.nr_unknown_id; return 0; } --=20 2.45.1 From nobody Fri Feb 13 22:36:10 2026 Received: from foss.arm.com (foss.arm.com [217.140.110.172]) by smtp.subspace.kernel.org (Postfix) with ESMTP id 3291B7F7FC; Tue, 21 May 2024 13:30:45 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; arc=none smtp.client-ip=217.140.110.172 ARC-Seal: i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1716298246; cv=none; b=tek6jurWsLGq12pOb62eEDVrSRZtKSpCPa8lUfUfF88R0gHKGfhOZdihsCl8A16I3ZpMK9Y0Y/Y/wOWn0i4aT9WCBcNe0al8uZ067sMD6cAXOUxTnJj585TNOrOn8ClPGOxIsMqWqVOWst4wVX+qRoWrJVh72amZVJtxmgSwMiE= ARC-Message-Signature: i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1716298246; c=relaxed/simple; bh=v13Dt602tJnaqyKpadsMo9w6iuh3ZNfEnTiEaz+kZIw=; h=From:To:Cc:Subject:Date:Message-ID:In-Reply-To:References: MIME-Version; b=H6s6xlrL7n9F4cS+ARJ4kX5dxV0Qz0eqSexhZ3mMeSY8PvjZovHClJ25WprQgcboC9Wwvzmjza2vvrUidr+aJzmDnCYmjz/sh4rJgDwhdYIEd0BskV5uOr11GBXbUnH93FTihuV8ONx15FwQU6DPCS4OnRwnSYo63xJBGkeMSJU= ARC-Authentication-Results: i=1; smtp.subspace.kernel.org; dmarc=pass (p=none dis=none) header.from=arm.com; spf=pass smtp.mailfrom=arm.com; arc=none smtp.client-ip=217.140.110.172 Authentication-Results: smtp.subspace.kernel.org; dmarc=pass (p=none dis=none) header.from=arm.com Authentication-Results: smtp.subspace.kernel.org; spf=pass smtp.mailfrom=arm.com Received: from usa-sjc-imap-foss1.foss.arm.com (unknown [10.121.207.14]) by usa-sjc-mx-foss1.foss.arm.com (Postfix) with ESMTP id DE0D6FEC; Tue, 21 May 2024 06:31:08 -0700 (PDT) Received: from e126817.cambridge.arm.com (e126817.cambridge.arm.com [10.2.3.5]) by usa-sjc-imap-foss1.foss.arm.com (Postfix) with ESMTPA id 2C2AF3F641; Tue, 21 May 2024 06:30:43 -0700 (PDT) From: Ben Gainey To: peterz@infradead.org, mingo@redhat.com, acme@kernel.org, namhyung@kernel.org Cc: james.clark@arm.com, mark.rutland@arm.com, alexander.shishkin@linux.intel.com, jolsa@kernel.org, irogers@google.com, adrian.hunter@intel.com, linux-perf-users@vger.kernel.org, linux-kernel@vger.kernel.org, Ben Gainey Subject: [PATCH v6 4/4] tools/perf: Allow inherit + PERF_SAMPLE_READ when opening events Date: Tue, 21 May 2024 14:30:29 +0100 Message-ID: <20240521133029.83654-5-ben.gainey@arm.com> X-Mailer: git-send-email 2.45.1 In-Reply-To: <20240521133029.83654-1-ben.gainey@arm.com> References: <20240521133029.83654-1-ben.gainey@arm.com> Precedence: bulk X-Mailing-List: linux-kernel@vger.kernel.org List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 Content-Transfer-Encoding: quoted-printable Content-Type: text/plain; charset="utf-8" The tool will now default to this new mode if the user specifies a sampling group when not in system-wide mode, and when --no-inherit is not specified. This change updates evsel to allow the combination of inherit and PERF_SAMPLE_READ. A fallback is implemented for kernel versions where this feature is not supported. Signed-off-by: Ben Gainey --- tools/perf/tests/attr/README | 2 + .../tests/attr/test-record-group-sampling | 39 ------------ .../tests/attr/test-record-group-sampling1 | 50 ++++++++++++++++ .../tests/attr/test-record-group-sampling2 | 60 +++++++++++++++++++ tools/perf/tests/attr/test-record-group2 | 9 +-- tools/perf/util/evsel.c | 19 +++++- tools/perf/util/evsel.h | 1 + 7 files changed, 135 insertions(+), 45 deletions(-) delete mode 100644 tools/perf/tests/attr/test-record-group-sampling create mode 100644 tools/perf/tests/attr/test-record-group-sampling1 create mode 100644 tools/perf/tests/attr/test-record-group-sampling2 diff --git a/tools/perf/tests/attr/README b/tools/perf/tests/attr/README index 4066fec7180a8..67c4ca76b85d5 100644 --- a/tools/perf/tests/attr/README +++ b/tools/perf/tests/attr/README @@ -51,6 +51,8 @@ Following tests are defined (with perf commands): perf record --call-graph fp kill (test-record-graph-fp-aarc= h64) perf record -e '{cycles,instructions}' kill (test-record-group1) perf record -e '{cycles/period=3D1/,instructions/period=3D2/}:S' kill (t= est-record-group2) + perf record -e '{cycles,cache-misses}:S' kill (test-record-group-samplin= g1) + perf record -c 10000 -e '{cycles,cache-misses}:S' kill (test-record-grou= p-sampling2) perf record -D kill (test-record-no-delay) perf record -i kill (test-record-no-inherit) perf record -n kill (test-record-no-samples) diff --git a/tools/perf/tests/attr/test-record-group-sampling b/tools/perf/= tests/attr/test-record-group-sampling deleted file mode 100644 index 97e7e64a38f07..0000000000000 --- a/tools/perf/tests/attr/test-record-group-sampling +++ /dev/null @@ -1,39 +0,0 @@ -[config] -command =3D record -args =3D --no-bpf-event -e '{cycles,cache-misses}:S' kill >/dev/null 2>= &1 -ret =3D 1 - -[event-1:base-record] -fd=3D1 -group_fd=3D-1 -sample_type=3D343 -read_format=3D12|28 -inherit=3D0 - -[event-2:base-record] -fd=3D2 -group_fd=3D1 - -# cache-misses -type=3D0 -config=3D3 - -# default | PERF_SAMPLE_READ -sample_type=3D343 - -# PERF_FORMAT_ID | PERF_FORMAT_GROUP | PERF_FORMAT_LOST -read_format=3D12|28 -task=3D0 -mmap=3D0 -comm=3D0 -enable_on_exec=3D0 -disabled=3D0 - -# inherit is disabled for group sampling -inherit=3D0 - -# sampling disabled -sample_freq=3D0 -sample_period=3D0 -freq=3D0 -write_backward=3D0 diff --git a/tools/perf/tests/attr/test-record-group-sampling1 b/tools/perf= /tests/attr/test-record-group-sampling1 new file mode 100644 index 0000000000000..9b87306266329 --- /dev/null +++ b/tools/perf/tests/attr/test-record-group-sampling1 @@ -0,0 +1,50 @@ +[config] +command =3D record +args =3D --no-bpf-event -e '{cycles,cache-misses}:S' kill >/dev/null 2>= &1 +ret =3D 1 + +[event-1:base-record] +fd=3D1 +group_fd=3D-1 + +# cycles +type=3D0 +config=3D0 + +# default | PERF_SAMPLE_READ +sample_type=3D343 + +# PERF_FORMAT_ID | PERF_FORMAT_GROUP | PERF_FORMAT_LOST | PERF_FORMAT_TOT= AL_TIME_ENABLED | PERF_FORMAT_TOTAL_TIME_RUNNING +read_format=3D28|31 +task=3D1 +mmap=3D1 +comm=3D1 +enable_on_exec=3D1 +disabled=3D1 + +# inherit is enabled for group sampling +inherit=3D1 + +[event-2:base-record] +fd=3D2 +group_fd=3D1 + +# cache-misses +type=3D0 +config=3D3 + +# default | PERF_SAMPLE_READ +sample_type=3D343 + +# PERF_FORMAT_ID | PERF_FORMAT_GROUP | PERF_FORMAT_LOST | PERF_FORMAT_TOT= AL_TIME_ENABLED | PERF_FORMAT_TOTAL_TIME_RUNNING +read_format=3D28|31 +task=3D0 +mmap=3D0 +comm=3D0 +enable_on_exec=3D0 +disabled=3D0 +freq=3D0 + +# inherit is enabled for group sampling +inherit=3D1 + diff --git a/tools/perf/tests/attr/test-record-group-sampling2 b/tools/perf= /tests/attr/test-record-group-sampling2 new file mode 100644 index 0000000000000..8e29fc13f6668 --- /dev/null +++ b/tools/perf/tests/attr/test-record-group-sampling2 @@ -0,0 +1,60 @@ +[config] +command =3D record +args =3D --no-bpf-event -c 10000 -e '{cycles,cache-misses}:S' kill >/de= v/null 2>&1 +ret =3D 1 + +[event-1:base-record] +fd=3D1 +group_fd=3D-1 + +# cycles +type=3D0 +config=3D0 + +# default | PERF_SAMPLE_READ +sample_type=3D87 + +# PERF_FORMAT_ID | PERF_FORMAT_GROUP | PERF_FORMAT_LOST | PERF_FORMAT_TOT= AL_TIME_ENABLED | PERF_FORMAT_TOTAL_TIME_RUNNING +read_format=3D28|31 +task=3D1 +mmap=3D1 +comm=3D1 +enable_on_exec=3D1 +disabled=3D1 + +# inherit is enabled for group sampling +inherit=3D1 + +# sampling disabled +sample_freq=3D0 +sample_period=3D10000 +freq=3D0 +write_backward=3D0 + +[event-2:base-record] +fd=3D2 +group_fd=3D1 + +# cache-misses +type=3D0 +config=3D3 + +# default | PERF_SAMPLE_READ +sample_type=3D87 + +# PERF_FORMAT_ID | PERF_FORMAT_GROUP | PERF_FORMAT_LOST | PERF_FORMAT_TOT= AL_TIME_ENABLED | PERF_FORMAT_TOTAL_TIME_RUNNING +read_format=3D28|31 +task=3D0 +mmap=3D0 +comm=3D0 +enable_on_exec=3D0 +disabled=3D0 + +# inherit is enabled for group sampling +inherit=3D1 + +# sampling disabled +sample_freq=3D0 +sample_period=3D0 +freq=3D0 +write_backward=3D0 diff --git a/tools/perf/tests/attr/test-record-group2 b/tools/perf/tests/at= tr/test-record-group2 index cebdaa8e64e47..785892a54d9e1 100644 --- a/tools/perf/tests/attr/test-record-group2 +++ b/tools/perf/tests/attr/test-record-group2 @@ -9,8 +9,9 @@ group_fd=3D-1 config=3D0|1 sample_period=3D1234000 sample_type=3D87 -read_format=3D12|28 -inherit=3D0 +read_format=3D28|31 +disabled=3D1 +inherit=3D1 freq=3D0 =20 [event-2:base-record] @@ -19,9 +20,9 @@ group_fd=3D1 config=3D0|1 sample_period=3D6789000 sample_type=3D87 -read_format=3D12|28 +read_format=3D28|31 disabled=3D0 -inherit=3D0 +inherit=3D1 mmap=3D0 comm=3D0 freq=3D0 diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c index 3536404e9447b..557d409c53d6c 100644 --- a/tools/perf/util/evsel.c +++ b/tools/perf/util/evsel.c @@ -1156,7 +1156,15 @@ void evsel__config(struct evsel *evsel, struct recor= d_opts *opts, */ if (leader->core.nr_members > 1) { attr->read_format |=3D PERF_FORMAT_GROUP; - attr->inherit =3D 0; + } + + /* + * Inherit + SAMPLE_READ requires SAMPLE_TID in the read_format + */ + if (attr->inherit) { + evsel__set_sample_bit(evsel, TID); + evsel->core.attr.read_format |=3D + PERF_FORMAT_ID; } } =20 @@ -1832,6 +1840,8 @@ static int __evsel__prepare_open(struct evsel *evsel,= struct perf_cpu_map *cpus, =20 static void evsel__disable_missing_features(struct evsel *evsel) { + if (perf_missing_features.inherit_sample_read) + evsel->core.attr.inherit =3D 0; if (perf_missing_features.branch_counters) evsel->core.attr.branch_sample_type &=3D ~PERF_SAMPLE_BRANCH_COUNTERS; if (perf_missing_features.read_lost) @@ -1887,7 +1897,12 @@ bool evsel__detect_missing_features(struct evsel *ev= sel) * Must probe features in the order they were added to the * perf_event_attr interface. */ - if (!perf_missing_features.branch_counters && + if (!perf_missing_features.inherit_sample_read && + evsel->core.attr.inherit && (evsel->core.attr.sample_type & PERF_SAMP= LE_READ)) { + perf_missing_features.inherit_sample_read =3D true; + pr_debug2("Using PERF_SAMPLE_READ / :S modifier is not compatible with i= nherit, falling back to no-inherit.\n"); + return true; + } else if (!perf_missing_features.branch_counters && (evsel->core.attr.branch_sample_type & PERF_SAMPLE_BRANCH_COUNTERS)) { perf_missing_features.branch_counters =3D true; pr_debug2("switching off branch counters support\n"); diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h index 517cff431de20..21b8b7e70e75e 100644 --- a/tools/perf/util/evsel.h +++ b/tools/perf/util/evsel.h @@ -192,6 +192,7 @@ struct perf_missing_features { bool weight_struct; bool read_lost; bool branch_counters; + bool inherit_sample_read; }; =20 extern struct perf_missing_features perf_missing_features; --=20 2.45.1