[PATCH V2] X86/CPU: Avoid 100ms sleep for cpu offline during S3

[Lan Tianyu]

With some bad kernel configures, cpu offline consumes more than 100ms
during S3. This because native_cpu_die() would fall into 100ms
sleep when cpu idle loop thread marked cpu state to DEAD slower. It's
timing related issue. What native_cpu_die() does is that poll cpu
state and wait for 100ms if cpu state hasn't been marked to DEAD.
The 100ms sleep doesn't make sense. To avoid such long sleep, this
patch is to add struct completion to each cpu, wait for the completion
in the native_cpu_die() and wakeup the completion when the cpu state is
marked to DEAD.

Tested on the Intel Xeon server with 48 cores, Ivbridge and Haswell laptops.
the times of cpu offline on these machines are reduced from more than 100ms
to less than 5ms. The system suspend time reduces 2.3s on the servers.

Signed-off-by: Lan Tianyu <tianyu.lan@xxxxxxxxx>
---
Change since V1:
	Remove redundant empty line and correct some code styple issues.

 arch/x86/kernel/smpboot.c | 23 +++++++++++------------
 1 file changed, 11 insertions(+), 12 deletions(-)

diff --git a/arch/x86/kernel/smpboot.c b/arch/x86/kernel/smpboot.c
index edf80bb..bc65e42 100644
--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -102,6 +102,8 @@ DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_llc_shared_map);
 DEFINE_PER_CPU_SHARED_ALIGNED(struct cpuinfo_x86, cpu_info);
 EXPORT_PER_CPU_SYMBOL(cpu_info);
 
+DEFINE_PER_CPU(struct completion, die_complete);
+
 atomic_t init_deasserted;
 
 /*
@@ -1331,7 +1333,7 @@ int native_cpu_disable(void)
 		return ret;
 
 	clear_local_APIC();
-
+	init_completion(&per_cpu(die_complete, smp_processor_id()));
 	cpu_disable_common();
 	return 0;
 }
@@ -1339,18 +1341,14 @@ int native_cpu_disable(void)
 void native_cpu_die(unsigned int cpu)
 {
 	/* We don't do anything here: idle task is faking death itself. */
-	unsigned int i;
+	wait_for_completion_timeout(&per_cpu(die_complete, cpu), HZ);
 
-	for (i = 0; i < 10; i++) {
-		/* They ack this in play_dead by setting CPU_DEAD */
-		if (per_cpu(cpu_state, cpu) == CPU_DEAD) {
-			if (system_state == SYSTEM_RUNNING)
-				pr_info("CPU %u is now offline\n", cpu);
-			return;
-		}
-		msleep(100);
-	}
-	pr_err("CPU %u didn't die...\n", cpu);
+	/* They ack this in play_dead by setting CPU_DEAD */
+	if (per_cpu(cpu_state, cpu) == CPU_DEAD) {
+		if (system_state == SYSTEM_RUNNING)
+			pr_info("CPU %u is now offline\n", cpu);
+	} else
+		pr_err("CPU %u didn't die...\n", cpu);
 }
 
 void play_dead_common(void)
@@ -1362,6 +1360,7 @@ void play_dead_common(void)
 	mb();
 	/* Ack it */
 	__this_cpu_write(cpu_state, CPU_DEAD);
+	complete(&per_cpu(die_complete, smp_processor_id()));
 
 	/*
 	 * With physical CPU hotplug, we should halt the cpu
-- 
1.8.4.rc0.1.g8f6a3e5.dirty

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/