Skip to contentSkip to conversation

fix bogus hotplug cpu warning

DISCUSSION Ingo Molnar 2 participants 3 messages Aug 28, 2007

lore
Share & tools
lore
#0 lore raw

this commit:

commit d243769d3f83b318813a04a9592bb7cfedc6c280
Author: Hugh Dickins <[email protected]>
Date: Mon Aug 27 16:06:19 2007 +0100

fix bogus hotplug cpu warning

Fix bogus DEBUG_PREEMPT warning on x86_64, when cpu brought online after bootup: current_is_keventd is right to note its use of smp_processor_id is preempt-safe, but should use raw_smp_processor_id to avoid the warning.

- int cpu = smp_processor_id(); /* preempt-safe: keventd is per-cpu */
+ int cpu = raw_smp_processor_id(); /* preempt-safe: keventd is per-cpu */

is incorrect. smp_processor_id() will _not_ warn about true per-cpu tasks:

        /*
         * Kernel threads bound to a single CPU can safely use
         * smp_processor_id():
         */
        this_mask = cpumask_of_cpu(this_cpu);

        if (cpus_equal(current->cpus_allowed, this_mask))
                goto out;

so if this check triggered for the cpu-offline code it's because keventd was _not_ per-cpu (perhaps because the cpu offline code broke its affinity?). So your patch hides a real bug.

Ingo

#1 lore raw
Quoted 28 lines · Ingo Molnar

this commit:

commit d243769d3f83b318813a04a9592bb7cfedc6c280
Author: Hugh Dickins <[email protected]>
Date: Mon Aug 27 16:06:19 2007 +0100

fix bogus hotplug cpu warning

Fix bogus DEBUG_PREEMPT warning on x86_64, when cpu brought online after bootup: current_is_keventd is right to note its use of smp_processor_id is preempt-safe, but should use raw_smp_processor_id to avoid the warning.

(quoted diff) +1 -1
-       int cpu = smp_processor_id();   /* preempt-safe: keventd is per-cpu */
+       int cpu = raw_smp_processor_id(); /* preempt-safe: keventd is per-cpu */

is incorrect. smp_processor_id() will _not_ warn about true per-cpu tasks:

        /*
         * Kernel threads bound to a single CPU can safely use
         * smp_processor_id():
         */
        this_mask = cpumask_of_cpu(this_cpu);

        if (cpus_equal(current->cpus_allowed, this_mask))
                goto out;
Quoted 3 lines

so if this check triggered for the cpu-offline code it's because keventd was _not_ per-cpu (perhaps because the cpu offline code broke its affinity?). So your patch hides a real bug.

No, keventd is properly per-cpu, but the task which is calling current_is_keventd() is a normal user task, so smp_processor_id() did warn about it. It wouldn't make much sense to have a function current_is_keventd() if you could only call it from a keventd, would it ;-? Take a look at it (kernel/workqueue.c), I believe what it does is unusual (hence that comment about being "preempt-safe") but valid.

Hugh

#2 lore raw
Quoted 10 lines · Hugh Dickins
Quoted 3 lines

so if this check triggered for the cpu-offline code it's because keventd was _not_ per-cpu (perhaps because the cpu offline code broke its affinity?). So your patch hides a real bug.

No, keventd is properly per-cpu, but the task which is calling current_is_keventd() is a normal user task, so smp_processor_id() did warn about it. It wouldn't make much sense to have a function current_is_keventd() if you could only call it from a keventd, would it ;-? Take a look at it (kernel/workqueue.c), I believe what it does is unusual (hence that comment about being "preempt-safe") but valid.

ok, you are right indeed.

Ingo