linux-kernel - Re: [PATCH 1/3] perf: Optimize perf_install_in

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20191022100355.GM1817@hirez.programming.kicks-ass.net>
Date:   Tue, 22 Oct 2019 12:03:55 +0200
From:   Peter Zijlstra <peterz@...radead.org>
To:     mingo@...nel.org, linux-kernel@...r.kernel.org
Cc:     acme@...nel.org, mark.rutland@....com,
        alexander.shishkin@...ux.intel.com, jolsa@...hat.com,
        namhyung@...nel.org, andi@...stfloor.org, kan.liang@...ux.intel.com
Subject: Re: [PATCH 1/3] perf: Optimize perf_install_in_event()

On Tue, Oct 22, 2019 at 11:20:18AM +0200, Peter Zijlstra wrote:
> Andi reported that when creating a lot of events, a lot of time is
> spend in IPIs and asked if it would be possible to elide some of that.
> 
> Now when, as for example the perf-tool always does, events are created
> disabled, then these events will not need to be scheduled when added
> to the context (they're still disable) and therefore the IPI is not
> required -- except for the very first event, that will need to set
> ctx->is_active.
> 
> ( it might be possible to set ctx->is_active remotely for cpu_ctx, but
>   we really need the IPI for task_ctx, so lets not make that
>   distinction )
> 
> Also use __perf_effective_state() since group events depend on the
> state of the leader, if the leader is OFF, the whole group is OFF.
> 
> So when sibling events are created enabled (XXX check tool) then we
> only need a single IPI to create and enable the whole group (+ that
> initial IPI to initialize the context).

Arguably, we could possible do something like so as well, but I'm not
sure it makes sense as it will not help if IOC_ENABLE is called in
creation order. Because in that case we'll enable the group leader
before the siblings and we'll schedule them one at a time, instead of
the whole group at once.

---
diff --git a/kernel/events/core.c b/kernel/events/core.c
index f9a5d4356562..2a5e6d9654e1 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -2791,6 +2791,7 @@ static void __perf_event_enable(struct perf_event *event,
 static void _perf_event_enable(struct perf_event *event)
 {
 	struct perf_event_context *ctx = event->ctx;
+	struct perf_event *leader;
 
 	raw_spin_lock_irq(&ctx->lock);
 	if (event->state >= PERF_EVENT_STATE_INACTIVE ||
@@ -2808,6 +2809,17 @@ static void _perf_event_enable(struct perf_event *event)
 	 */
 	if (event->state == PERF_EVENT_STATE_ERROR)
 		event->state = PERF_EVENT_STATE_OFF;
+
+	/*
+	 * If this is a sibling event and the group leader is still OFF
+	 * then there is no point in trying to schedule this event.
+	 */
+	leader = event->group_leader;
+	if (leader != event && leader->state == PERF_EVENT_STATE_OFF) {
+		event->state = PERF_EVENT_STATE_INACTIVE;
+		raw_spin_unlock_irq(&ctx->lock);
+		return;
+	}
 	raw_spin_unlock_irq(&ctx->lock);
 
 	event_function_call(event, __perf_event_enable, NULL);