sched: Fix vmark regression on big machines
Mike Galbraith [Mon, 4 Jan 2010 13:44:56 +0000 (14:44 +0100)]
SD_PREFER_SIBLING is set at the CPU domain level if power saving isn't
enabled, leading to many cache misses on large machines as we traverse
looking for an idle shared cache to wake to.  Change the enabler of
select_idle_sibling() to SD_SHARE_PKG_RESOURCES, and enable same at the
sibling domain level.

Reported-by: Lin Ming <ming.m.lin@intel.com>
Signed-off-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1262612696.15495.15.camel@marge.simson.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

include/linux/topology.h
kernel/sched_fair.c

index 57e6357..5b81156 100644 (file)
@@ -99,7 +99,7 @@ int arch_update_cpu_topology(void);
                                | 1*SD_WAKE_AFFINE                      \
                                | 1*SD_SHARE_CPUPOWER                   \
                                | 0*SD_POWERSAVINGS_BALANCE             \
-                               | 0*SD_SHARE_PKG_RESOURCES              \
+                               | 1*SD_SHARE_PKG_RESOURCES              \
                                | 0*SD_SERIALIZE                        \
                                | 0*SD_PREFER_SIBLING                   \
                                ,                                       \
index 42ac3c9..8fe7ee8 100644 (file)
@@ -1508,7 +1508,7 @@ static int select_task_rq_fair(struct task_struct *p, int sd_flag, int wake_flag
                         * If there's an idle sibling in this domain, make that
                         * the wake_affine target instead of the current cpu.
                         */
-                       if (tmp->flags & SD_PREFER_SIBLING)
+                       if (tmp->flags & SD_SHARE_PKG_RESOURCES)
                                target = select_idle_sibling(p, tmp, target);
 
                        if (target >= 0) {