aboutsummaryrefslogtreecommitdiffstats
path: root/arch
diff options
context:
space:
mode:
authorGerman Monroy <german.monroy@intel.com>2012-05-14 17:14:19 -0700
committerTodd Poynor <toddpoynor@google.com>2012-07-16 13:40:02 -0700
commitb595a66a0be0e99da8854d4ff7693193e6ac1a81 (patch)
tree1db35408480f51f01acf817ded31f86f2aa9d732 /arch
parent66943ea3526d56b35880f962094ae29b1ded1d06 (diff)
downloadkernel_samsung_crespo-b595a66a0be0e99da8854d4ff7693193e6ac1a81.zip
kernel_samsung_crespo-b595a66a0be0e99da8854d4ff7693193e6ac1a81.tar.gz
kernel_samsung_crespo-b595a66a0be0e99da8854d4ff7693193e6ac1a81.tar.bz2
x86: Call idle notifiers
BZ: 35303 Google patched the idle loop for x86-64 (commit 1fd57f722c) and for ARM (commit 41fa406c26), but they forgot x86-32. This is preventing their interactive governor from upshifting CPU frequencies in intel's SOC platforms (e.g. Medfield, etc.). Fixing that. NOTE: The notifier calls are not located in the same exact place in the idle loop for x86-64 and ARM. The notifier is called inside the `while(!need_resched)' loop in x86-64 and outside of it in ARM. Since Google has likely tuned this governor for ARM, leaving it as in ARM. Change-Id: Ibefd0c8f08e4b4c24c4a5c32dcdc574f9090b2b9 Signed-off-by: German Monroy <german.monroy@intel.com>
Diffstat (limited to 'arch')
-rw-r--r--arch/x86/kernel/process_32.c2
1 files changed, 2 insertions, 0 deletions
diff --git a/arch/x86/kernel/process_32.c b/arch/x86/kernel/process_32.c
index fcdb1b3..12dae34 100644
--- a/arch/x86/kernel/process_32.c
+++ b/arch/x86/kernel/process_32.c
@@ -98,6 +98,7 @@ void cpu_idle(void)
/* endless idle loop with no priority at all */
while (1) {
tick_nohz_stop_sched_tick(1);
+ idle_notifier_call_chain(IDLE_START);
while (!need_resched()) {
check_pgt_cache();
@@ -112,6 +113,7 @@ void cpu_idle(void)
pm_idle();
start_critical_timings();
}
+ idle_notifier_call_chain(IDLE_END);
tick_nohz_restart_sched_tick();
preempt_enable_no_resched();
schedule();