Skip to contentSkip to conversation

uncore doing kfree() from CPU_{STARTING,DYING}

PATCH Peter Zijlstra 1 participant 1 message Oct 4, 2012

Subsystems (2)
x86 perf events
lore
Share & tools
lore

Peter Zijlstra

author maintainer
#0 lore raw

Hi Zheng,

Can you have a look at 7fdba1ca and do a similar patch for uncore? Thomas is sad because uncore breaks -rt, the problem (as described in that patch) is that kfree() must schedule on -rt while CPU_{STARTING,DYING} really are atomic.

~ Peter

---
commit 7fdba1ca10462f42ad2246b918fe6368f5ce488e

Author: Peter Zijlstra <[email protected]>
Date:   Fri Jul 22 13:41:54 2011 +0200

perf, x86: Avoid kfree() in CPU_STARTING

On -rt kfree() can schedule, but CPU_STARTING is before the CPU is fully up and running. These are contradictory, so avoid it. Instead push the kfree() to CPU_ONLINE where we're free to schedule.

Reported-by Thomas Gleixner <[email protected]>
Signed-off-by Peter Zijlstra <[email protected]>
Link http://lkml.kernel.org/n/[email protected] open here
Signed-off-by Ingo Molnar <[email protected]>
arch/x86/kernel/cpu/perf_event.c +8 -0
index 4ee3abf..594d425 100644
@@ -129,6 +129,8 @@ struct cpu_hw_events {
 	 * AMD specific bits
 	 */
 	struct amd_nb		*amd_nb;
+
+	void			*kfree_on_online;
 };
 
 #define __EVENT_CONSTRAINT(c, n, m, w) {\

@@ -1466,10 +1468,12 @@ static int __cpuinit
 x86_pmu_notifier(struct notifier_block *self, unsigned long action, void *hcpu)
 {
 	unsigned int cpu = (long)hcpu;
+	struct cpu_hw_events *cpuc = &per_cpu(cpu_hw_events, cpu);
 	int ret = NOTIFY_OK;
 
 	switch (action & ~CPU_TASKS_FROZEN) {
 	case CPU_UP_PREPARE:
+		cpuc->kfree_on_online = NULL;
 		if (x86_pmu.cpu_prepare)
 			ret = x86_pmu.cpu_prepare(cpu);
 		break;

@@ -1479,6 +1483,10 @@ x86_pmu_notifier(struct notifier_block *self, unsigned long action, void *hcpu)
 			x86_pmu.cpu_starting(cpu);
 		break;
 
+	case CPU_ONLINE:
+		kfree(cpuc->kfree_on_online);
+		break;
+
 	case CPU_DYING:
 		if (x86_pmu.cpu_dying)
 			x86_pmu.cpu_dying(cpu);
arch/x86/kernel/cpu/perf_event_amd.c +1 -1
index 941caa2..ee9436c 100644
@@ -350,7 +350,7 @@ static void amd_pmu_cpu_starting(int cpu)
 			continue;
 
 		if (nb->nb_id == nb_id) {
-			kfree(cpuc->amd_nb);
+			cpuc->kfree_on_online = cpuc->amd_nb;
 			cpuc->amd_nb = nb;
 			break;
 		}
arch/x86/kernel/cpu/perf_event_intel.c +1 -1
index f88af2c..3751494 100644
@@ -1362,7 +1362,7 @@ static void intel_pmu_cpu_starting(int cpu)
 
 		pc = per_cpu(cpu_hw_events, i).shared_regs;
 		if (pc && pc->core_id == core_id) {
-			kfree(cpuc->shared_regs);
+			cpuc->kfree_on_online = cpuc->shared_regs;
 			cpuc->shared_regs = pc;
 			break;
 		}