linux-kernel - [PATCH v2] sched/rt: Document why has_pushable_tasks() isn't called with a runqueue lock

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>] [day] [month] [year] [list]

Date:   Thu, 2 Mar 2017 15:50:22 -0500
From:   Steven Rostedt <rostedt@...dmis.org>
To:     Peter Zijlstra <peterz@...radead.org>
Cc:     LKML <linux-kernel@...r.kernel.org>,
        Ingo Molnar <mingo@...nel.org>,
        Andrew Morton <akpm@...ux-foundation.org>
Subject: [PATCH v2] sched/rt: Document why has_pushable_tasks() isn't called
 with a runqueue lock


From: Steven Rostedt (VMware) <rostedt@...dmis.org>

While reviewing the RT scheduling IPI logic, I was thinking that it was
a bug that has_pushable_tasks(rq) was not called under the runqueue
lock. But then I realized that there isn't a case where a race would
cause a problem, as to update has_pushable_tasks() would trigger a
push_rt_task() call from the CPU doing the update.

This subtle logic deserves a comment.

Signed-off-by: Steven Rostedt (VMware) <rostedt@...dmis.org>
---
Changes from v1:
   Removed pronouns that causes confusion, and added a statement about
   push_rt_task() being called elsewhere when has_pushable_tasks() is
   set someplace else.

Index: linux-trace.git/kernel/sched/rt.c
===================================================================
--- linux-trace.git.orig/kernel/sched/rt.c
+++ linux-trace.git/kernel/sched/rt.c
@@ -1976,6 +1976,18 @@ static void try_to_push_tasks(void *arg)
 	src_rq = rq_of_rt_rq(rt_rq);
 
 again:
+	/*
+	 * Normally, has_pushable_tasks() would be performed within the
+	 * runqueue lock being held. But if has_pushable_tasks() is false
+	 * when entering this hard interrupt handler function, then to have
+	 * it set to true would require a wake up. A wake up of an RT task
+	 * will either cause a schedule if the woken task is higher priority
+	 * than the running task, or it would try to do a push from the CPU
+	 * doing the wake up. In ether case push_rt_task() would be performed
+	 * there, and missing it here would not be an issue. Grabbing the
+	 * runqueue lock in such a case would more likely just cause
+	 * unnecessary contention.
+	 */
 	if (has_pushable_tasks(rq)) {
 		raw_spin_lock(&rq->lock);
 		push_rt_task(rq);