[ofa-general] hang on close in umem_release
Roland Dreier
rdreier at cisco.com
Thu Jun 21 08:40:54 PDT 2007
hmm, I see what seems to be an "i can't believe it ever worked" type
bug -- if the context is closing but then we do manage to get the mm's
rwsem, it seems like we immediately try to lock it again, which
obviously deadlocks.
Does this patch fix your problem and look correct?
diff --git a/drivers/infiniband/core/umem.c b/drivers/infiniband/core/umem.c
index b4aec51..d40652a 100644
--- a/drivers/infiniband/core/umem.c
+++ b/drivers/infiniband/core/umem.c
@@ -225,13 +225,15 @@ void ib_umem_release(struct ib_umem *umem)
* up here and not be able to take the mmap_sem. In that case
* we defer the vm_locked accounting to the system workqueue.
*/
- if (context->closing && !down_write_trylock(&mm->mmap_sem)) {
- INIT_WORK(&umem->work, ib_umem_account);
- umem->mm = mm;
- umem->diff = diff;
-
- schedule_work(&umem->work);
- return;
+ if (context->closing) {
+ if (!down_write_trylock(&mm->mmap_sem)) {
+ INIT_WORK(&umem->work, ib_umem_account);
+ umem->mm = mm;
+ umem->diff = diff;
+
+ schedule_work(&umem->work);
+ return;
+ }
} else
down_write(&mm->mmap_sem);
More information about the general
mailing list