[openib-general] [PATCH] IB/uverbs: Fix lockdep warning when QP is created with 2 CQs

Arjan van de Ven arjan at infradead.org
Wed Aug 16 10:11:23 PDT 2006


On Wed, 2006-08-16 at 09:43 -0700, Roland Dreier wrote:
> Arjan, here's a case that disproves your rule of thumb that rwsems
> are equivalent to mutexes as far as correctness goes: you can't have
> an AB-BA deadlock with nested rwsems when using down_read().  In other
> words, the following:
> 
> 	down_read(&lock_1);
> 					down_read(&lock_2);
> 	down_read(&lock_2);
> 					down_read(&lock_1);
> 
> is perfectly safe

it's safe as long as you never ever do a down_write nested inside or
outside a down_read of any of these locks....

-- 
if you want to mail me at work (you don't), use arjan (at) linux.intel.com





More information about the general mailing list