Skip to contentSkip to conversation

disable preempt when calling smp_processor_id()

PATCH In review Chris Wright 2 participants 3 messages Nov 6, 2004 – Nov 8, 2004

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 disable preempt when calling smp_processor_id()

Oprofile is hitting many of the below BUG messages.

BUG: using smp_processor_id() in preemptible [00000001] code: sleep/4892
caller is task_exit_notify+0x9/0x17
Call Trace: <ffffffff802ac70f>{smp_processor_id+191}
	    <ffffffff803ff219>{task_exit_notify+9}
	    <ffffffff8014a930>{notifier_call_chain+32}
	    <ffffffff8013d181>{profile_task_exit+49}
	    <ffffffff8013eb22>{do_exit+34}
	    <ffffffff80146490>{process_timeout+0}
	    <ffffffff8013f8c8>{do_group_exit+264}
	    <ffffffff801106b6>{system_call+126}

smp_processor_id() is called w/out preempt disabled. Use get_cpu()/put_cpu() instead. Should this be put_cpu_no_resched()?

drivers/oprofile/buffer_sync.c renamed from drivers/oprofile/buffer_sync.c~orig +4 -2
@@ -62,7 +62,8 @@
 	/* To avoid latency problems, we only process the current CPU,
 	 * hoping that most samples for the task are on this CPU
 	 */
-	sync_buffer(smp_processor_id());
+	sync_buffer(get_cpu());
+	put_cpu();
   	return 0;
 }
 
@@ -86,7 +87,8 @@
 		/* To avoid latency problems, we only process the current CPU,
 		 * hoping that most samples for the task are on this CPU
 		 */
-		sync_buffer(smp_processor_id());
+		sync_buffer(get_cpu());
+		put_cpu();
 		return 0;
 	}
 
#1 lore raw
Quoted 2 lines · Chris Wright

smp_processor_id() is called w/out preempt disabled. Use get_cpu()/put_cpu() instead. Should this be put_cpu_no_resched()?

No, the patch below looks fine

regards
john

Quoted 24 lines
drivers/oprofile/buffer_sync.c renamed from drivers/oprofile/buffer_sync.c~orig +4 -2
@@ -62,7 +62,8 @@
 	/* To avoid latency problems, we only process the current CPU,
 	 * hoping that most samples for the task are on this CPU
 	 */
-	sync_buffer(smp_processor_id());
+	sync_buffer(get_cpu());
+	put_cpu();
   	return 0;
 }
 
@@ -86,7 +87,8 @@
 		/* To avoid latency problems, we only process the current CPU,
 		 * hoping that most samples for the task are on this CPU
 		 */
-		sync_buffer(smp_processor_id());
+		sync_buffer(get_cpu());
+		put_cpu();
 		return 0;
 	}
 
#2 lore raw
Quoted 5 lines · * John Levon ([email protected]) wrote:
Quoted 2 lines · Chris Wright

smp_processor_id() is called w/out preempt disabled. Use get_cpu()/put_cpu() instead. Should this be put_cpu_no_resched()?

No, the patch below looks fine

I realized over the weekend that it's still broken. And Cliff hit it in some tests he ran here. The sync_buffer() function can sleep, and while the unprotected call to smp_processor_id() is no longer a problem, scheduling with preempt disabled is.

thanks,
-chris

Signature · 1 line
Linux Security Modules     http://lsm.immunix.org     http://lsm.bkbits.net