[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Xen-devel] rwlock and per-cpu rwlock recursive?


On Arm, we made the (wrong?) assumption that the rwlock was recursive. We have a couple of places where the read lock can be nested (mostly the memaccess code).

I found out today that it can actually deadlock in the following case:
        1) CPU A -> Ask for the read lock
                => Succeed
        2) CPU B -> Ask for the write lock
                => Already taken by CPU A so go to the slowpath
                => Take the internal lock (lock->lock)
                => Wait until the lock is released.
        3) CPU A -> Ask for the read lock recursively
                => A writer is waiting so go to the slowpath
                => Try to take the internal lock (lock->lock)
                        => Blocking because CPU B already has it

So we end up in a deadlock case. Can someone confirm whether whether rwlock was meant to be recursive?

Similarly, I was thinking to move the p2m code to the per-cpu rwlock as x86 does. From what I gathered by reading the x86 code, this lock could be taken recursively. Am I right?

Lastly, the x86 p2m code seemed to use rwlock before hand. How did the p2m code was handling recursive locking?


Julien Grall

Xen-devel mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.