Re: [PATCH] sched: fix env->src_cpu for active migration

[Damien Wyart <damien.wyart@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 684 bytes --]

> Bingo, that was the problem in my setup: as the patch was applied
> through a script with others, I had missed the error message about the
> conflict (I have also another conflict which can be safely ignored so
> the new one did not catch my eye)... So the patch was only
> half-applied, and the final code is broken.

> How did you solve the conflict (I am not a scheduler expert)? I can
> retry running the patched kernel with your resolution, to check if
> everything is ok.

After looking a bit more, the conflict resolution seemed straighforward,
so I gave it a go. The attached version booted fine, so the initial
problem was purely PEBCAK... Sorry for the noise!

-- 
Damien

[-- Attachment #2: patch-sched.patch --]
[-- Type: text/x-diff, Size: 635 bytes --]

--- a/kernel/sched/fair.c	2013-02-06 20:49:17.447613049 +0100
+++ b/kernel/sched/fair.c	2013-02-13 20:52:37.409507417 +0100
@@ -5060,6 +5060,10 @@
 	ld_moved = 0;
 	lb_iterations = 1;
 	clock_updated = 0;
+
+	env.src_cpu   = busiest->cpu;
+	env.src_rq    = busiest;
+	
 	if (busiest->nr_running > 1) {
 		/*
 		 * Attempt to move tasks. If find_busiest_group has found
@@ -5068,8 +5072,6 @@
 		 * correctly treated as an imbalance.
 		 */
 		env.flags |= LBF_ALL_PINNED;
-		env.src_cpu   = busiest->cpu;
-		env.src_rq    = busiest;
 		env.loop_max  = min(sysctl_sched_nr_migrate, busiest->nr_running);
 
 		update_h_load(env.src_cpu);