Messages in this thread | | | Date | Mon, 13 Jan 2014 19:35:55 +0100 | From | Oleg Nesterov <> | Subject | Re: [PATCH 0/1] lockdep: Kill held_lock->check and "int check" arg of __lock_acquire() |
| |
On 01/12, Peter Zijlstra wrote: > > On Sun, Jan 12, 2014 at 06:45:32PM +0100, Oleg Nesterov wrote: > > But could someone please explain me what should lockdep_no_validate > > actually do? 1704f47b5 "lockdep: Add novalidate class for dev->mutex > > conversion" doesn't tell which kind of warnings it tries to avoid, > > and it looks buggy (see another email from me). > > It should disable lock-order checks for the locks that set that as > class. So we can still do the simple checks like detect if we free > memory with a held lock in and make sure we do not return to userspace > with a held lock; but the class should not participate in the lock > graph.
Yes, this is what I more or less understood, but I was confused anyway. Somehow I thought that this patch tries to suppress the warnings caused by rw_semaphore -> mutex conversion, and I was completely puzzled. But 3142788b79 "drivers/base: Convert dev->sem to mutex" actually did the semaphore -> mutex change!
Everything is clear now, thanks!
Oleg.
| |