[RFC] Lockdep VS rw-semaphores strangeness

[RFC] Lockdep VS rw-semaphores strangeness

[Pavel Emelianov]

[-- Attachment #1: Type: text/plain, Size: 837 bytes --]

Hello Ingo, Arjan.

I'm playing with lockdep and have a question about rw-sems.

down_read_trylock() looks like
int down_read_trylock(struct rw_semaphore *sem)
{
        int ret = __down_read_trylock(sem);

        if (ret == 1)
                rwsem_acquire_read(&sem->dep_map, 0, 1, _RET_IP_);
        return ret;
}
i.e. it calls rwsem_acquire_read() with trylock == 1.

But down_write_trylock() -
int down_write_trylock(struct rw_semaphore *sem)
{
        int ret = __down_write_trylock(sem);

        if (ret == 1)
                rwsem_acquire(&sem->dep_map, 0, 0, _RET_IP_);
        return ret;
}
- calls lockdep with trylock set to 0. Why?

I've already caught a fake warning when trying to write-lock
an mm->mmap_sem with another mm's mmap_sem write-locked. With the
patch attached everything works, fine.

Did I miss something?

[-- Attachment #2: diff-lockdep-rwsem-trylock --]
[-- Type: text/plain, Size: 356 bytes --]

--- ./kernel/rwsem.c.pbonrem	2007-03-06 19:09:50.000000000 +0300
+++ ./kernel/rwsem.c	2007-04-06 14:02:18.000000000 +0400
@@ -60,7 +60,7 @@ int down_write_trylock(struct rw_semapho
 	int ret = __down_write_trylock(sem);
 
 	if (ret == 1)
-		rwsem_acquire(&sem->dep_map, 0, 0, _RET_IP_);
+		rwsem_acquire(&sem->dep_map, 0, 1, _RET_IP_);
 	return ret;
 }
 

Re: [RFC] Lockdep VS rw-semaphores strangeness

[Arjan van de Ven]

Pavel Emelianov wrote:
> I've already caught a fake warning when trying to write-lock
> an mm->mmap_sem with another mm's mmap_sem write-locked. With the
> patch attached everything works, fine.
> 
btw why is that a fake warning? what is your guarantee that you never 
ever do it in the opposite direction from the other thread as well?

Re: [RFC] Lockdep VS rw-semaphores strangeness

[Pavel Emelianov]

Arjan van de Ven wrote:
> Pavel Emelianov wrote:
>> I've already caught a fake warning when trying to write-lock
>> an mm->mmap_sem with another mm's mmap_sem write-locked. With the
>> patch attached everything works, fine.
>>
> btw why is that a fake warning? what is your guarantee that you never
> ever do it in the opposite direction from the other thread as well?
> 

down_write(&mm->mmap_sem);
if (down_write_trylock(&another_mm->mmap_sem)) {
      ...
      up_write(...);
}
up_write(&mm->mmap_sem);

will never deadlock as if I have another thread is doing

down_write(&another_mm->mmap_sem);
if (down_write_trylock(&mm->mmap_sem)) {
      ...
      up_write(...);
}
up_write(&another_mm->mmap_sem);

booth will fail to trylock and go on.