Messages in this thread Patch in this message |  | | | Date | Mon, 09 Nov 2009 14:46:52 +0900 | | From | Tejun Heo <> | | Subject | [PATCH percpu#for-linus] percpu: fix possible deadlock via irq lock inversion |
Lockdep caught the following irq lock inversion which can lead to deadlock.
========================================================= [ INFO: possible irq lock inversion dependency detected ] 2.6.32-rc5-tip-04815-g12f0f93-dirty #745 --------------------------------------------------------- hub 1-3:1.0: state 7 ports 2 chg 0000 evt 0004 ksoftirqd/65/199 just changed the state of lock: (pcpu_lock){..-...}, at: [<ffffffff81130e04>] free_percpu+0x38/0x104 but this lock took another, SOFTIRQ-unsafe lock in the past: (vmap_area_lock){+.+...} and interrupts could create inverse lock ordering between them.
This happens because pcpu_lock is allowed to be acquired from irq context for free_percpu() path and alloc_percpu() path may call vfree() with pcpu_lock held. As vmap_area_lock isn't irq safe, if an IRQ occurs while vmap_area_lock is held and the irq handler calls free_percpu(), locking order inversion occurs.
As the nesting only occurs in the alloc path which isn't allowed to be called from irq context, A->B->A deadlock won't occur but A->B B->A deadlock is still possible.
The only place where vmap_area_lock is nested under pcpu_lock is while extending area_map to free old map. Break the locking order by temporarily releasing pcpu_lock when freeing old map. This is safe to do as allocation path is protected with outer pcpu_alloc_mutex.
Signed-off-by: Tejun Heo <tj@kernel.org> Reported-by: Yinghai Lu <yhlu.kernel@gmail.com> Cc: Ingo Molnar <mingo@elte.hu> --- If nobody objects, I'll push it to Linus tomorrow. Thanks. mm/percpu.c | 17 +++++++++++++++-- 1 file changed, 15 insertions(+), 2 deletions(-) diff --git a/mm/percpu.c b/mm/percpu.c index d907971..30cd343 100644 --- a/mm/percpu.c +++ b/mm/percpu.c @@ -372,7 +372,7 @@ static struct pcpu_chunk *pcpu_chunk_addr_search(void *addr) static int pcpu_extend_area_map(struct pcpu_chunk *chunk, unsigned long *flags) { int new_alloc; - int *new; + int *new, *old = NULL; size_t size;
/* has enough? */ @@ -407,10 +407,23 @@ static int pcpu_extend_area_map(struct pcpu_chunk *chunk, unsigned long *flags) * one of the first chunks and still using static map. */ if (chunk->map_alloc >= PCPU_DFL_MAP_ALLOC) - pcpu_mem_free(chunk->map, size); + old = chunk->map;
chunk->map_alloc = new_alloc; chunk->map = new; + + /* + * pcpu_mem_free() might end up calling vfree() which uses + * IRQ-unsafe lock and thus can't be called with pcpu_lock + * held. Release and reacquire pcpu_lock if old map needs to + * be freed. + */ + if (old) { + spin_unlock_irqrestore(&pcpu_lock, *flags); + pcpu_mem_free(old, size); + spin_lock_irqsave(&pcpu_lock, *flags); + } + return 0; }
|  |