Messages in this thread | | | Date | Mon, 9 Jun 2008 20:03:26 +0200 | From | Cornelia Huck <> | Subject | Re: [bug, 2.6.26-rc4/rc5] sporadic bootup crashes in blk_lookup_devt()/prepare_namespace() |
| |
On Mon, 9 Jun 2008 19:15:21 +0200, Cornelia Huck <cornelia.huck@de.ibm.com> wrote:
> On Mon, 9 Jun 2008 09:15:40 -0700 (PDT), > Linus Torvalds <torvalds@linux-foundation.org> wrote: > > > And this is all still ignoring the locking issue, of course. It would be > > trivial to just remove the block_class_lock, and change > > > > mutex_[un]lock(&block_class_lock); > > > > into > > > > down|up(&block_class.sem); > > > > except for _one_ case, which is > > > > bdev_map = kobj_map_init(base_probe, &block_class_lock); > > > > which really wants a mutex, not a sempahore. > > > > So to fix that, we'd need to make the class->sem be a mutex, and pass that > > in. Which is probably a good change too, but makes the whole thing much > > bigger. > > The driver core changes in -next convert class->sem to > class->p->class_mutex, which makes it non-accessible to drivers. > Most of the locking is easily done through converting to the class > iterator functions, but there are some cases where this is not going to > work: > > - The {register,unregister}_blkdev() functions, which don't directly > involve the class. > - The iterators for /proc/partitions, which take the lock in > part_start() and give it up again in part_stop(). > > Maybe we need a possibilty for a driver to lock a class from outside?
Argh. I was just trying to hack up a patch when I realized that we had to get a reference on the dynamic private structure when we take the lock - which made the patch so ugly that I dare not post it. I'll see if I have a better idea tomorrow (or someone beats me to it :)
| |