On Sat, 3 Mar 2012, Anton Vorontsov wrote:
There is not need to grab mutex in the atomic context, moreover this is wrong and causes the following bug:
BUG: sleeping function called from invalid context at kernel/mutex.c:271 in_atomic(): 1, irqs_disabled(): 1, pid: 1056, name: m no locks held by m/1056. irq event stamp: 58768 hardirqs last enabled at (58767): [<ffffffff8101ca35>] do_page_fault+0x275/0x450 hardirqs last disabled at (58768): [<ffffffff81325a2b>] apic_timer_interrupt+0x6b/0x80 softirqs last enabled at (58676): [<ffffffff81038def>] __do_softirq+0x10f/0x160 softirqs last disabled at (58671): [<ffffffff813262ac>] call_softirq+0x1c/0x26 Pid: 1056, comm: m Not tainted 3.2.0+ #3 Call Trace: <IRQ> [<ffffffff81062530>] ? print_irqtrace_events+0xd0/0xe0 [<ffffffff8102f5da>] __might_sleep+0x12a/0x1e0 [<ffffffff81321dbc>] mutex_lock_nested+0x3c/0x340 [<ffffffff81053a8c>] ? __run_hrtimer+0x4c/0x100 [<ffffffff810bd920>] ? vmevent_read+0x100/0x100 [<ffffffff810bd79e>] vmevent_sample+0x6e/0xf0 [<ffffffff810bd946>] vmevent_timer_fn+0x26/0x60 [<ffffffff81053a92>] __run_hrtimer+0x52/0x100 [<ffffffff81054463>] hrtimer_interrupt+0xf3/0x220 [<ffffffff810166d4>] smp_apic_timer_interrupt+0x64/0xa0 [<ffffffff81325a30>] apic_timer_interrupt+0x70/0x80 <EOI> [<ffffffff8101ca3a>] ? do_page_fault+0x27a/0x450 [<ffffffff8101ca35>] ? do_page_fault+0x275/0x450 [<ffffffff810036bf>] ? do_softirq+0x6f/0xc0 [<ffffffff8132485d>] ? retint_restore_args+0xe/0xe [<ffffffff8132484a>] ? retint_swapgs+0xe/0x13 [<ffffffff8116b31d>] ? trace_hardirqs_off_thunk+0x3a/0x3c [<ffffffff81324a3f>] page_fault+0x1f/0x30
Signed-off-by: Anton Vorontsov anton.vorontsov@linaro.org
The patch is for git://github.com/penberg/linux.git vmevent/core.
mm/vmevent.c | 4 ---- 1 files changed, 0 insertions(+), 4 deletions(-)
diff --git a/mm/vmevent.c b/mm/vmevent.c index 1375f9d..1dbefb5 100644 --- a/mm/vmevent.c +++ b/mm/vmevent.c @@ -71,8 +71,6 @@ static void vmevent_sample(struct vmevent_watch *watch) if (!vmevent_match(watch, &event)) return;
- mutex_lock(&watch->mutex);
- watch->pending = true;
if (watch->config.event_attrs & VMEVENT_EATTR_NR_AVAIL_PAGES) @@ -85,8 +83,6 @@ static void vmevent_sample(struct vmevent_watch *watch) watch->attr_values[n++] = event.nr_swap_pages; watch->nr_attrs = n;
- mutex_unlock(&watch->mutex);
}
Why do you think the mutex is not needed? Surely we need to synchronize vmevent_sample() against vmevent_read(), for example.
Pekka