sched: Fix select_idle_sibling() regression in selecting an idle SMT sibling
authorPeter Zijlstra <peterz@infradead.org>
Thu, 15 Dec 2011 19:47:00 +0000 (11:47 -0800)
committerIngo Molnar <mingo@elte.hu>
Fri, 16 Dec 2011 08:44:58 +0000 (09:44 +0100)
Mike Galbraith reported that this recent commit:

   commit 4dcfe1025b513c2c1da5bf5586adb0e80148f612
   Author: Peter Zijlstra <peterz@infradead.org>
   Date:   Thu Nov 10 13:01:10 2011 +0100

       sched: Avoid SMT siblings in select_idle_sibling() if possible

stopped selecting an idle SMT sibling when there are no idle
cores in a single socket system.

Intent of the select_idle_sibling() was to fallback to an idle
SMT sibling, if it fails to identify an idle core. But this
fallback was not happening on systems where all the scheduler
domains had `SD_SHARE_PKG_RESOURCES' flag set.

Fix it. Slightly bigger patch of cleaning all these goto's etc
is queued up for the next release.

Reported-by: Mike Galbraith <efault@gmx.de>
Reported-by: Alex Shi <alex.shi@intel.com>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Link: http://lkml.kernel.org/r/1323978421.1984.244.camel@sbsiddha-desk.sc.intel.com
Signed-off-by: Ingo Molnar <mingo@elte.hu>
kernel/sched_fair.c

index a78ed2736ba79f02a201d8256bd9e0a56d57981e..8a39fa3e3c6c7bafe368bc2e27c607801e8afb87 100644 (file)
@@ -2352,13 +2352,11 @@ again:
                if (!smt && (sd->flags & SD_SHARE_CPUPOWER))
                        continue;
 
-               if (!(sd->flags & SD_SHARE_PKG_RESOURCES)) {
-                       if (!smt) {
-                               smt = 1;
-                               goto again;
-                       }
+               if (smt && !(sd->flags & SD_SHARE_CPUPOWER))
+                       break;
+
+               if (!(sd->flags & SD_SHARE_PKG_RESOURCES))
                        break;
-               }
 
                sg = sd->groups;
                do {
@@ -2378,6 +2376,10 @@ next:
                        sg = sg->next;
                } while (sg != sd->groups);
        }
+       if (!smt) {
+               smt = 1;
+               goto again;
+       }
 done:
        rcu_read_unlock();