Re: [PATCH 0/7] convert semaphore to mutex in struct class

["Dave Young" <hidave.darkstar@gmail.com>]

On Jan 8, 2008 1:20 AM, Greg KH <gregkh@suse.de> wrote:
> On Mon, Jan 07, 2008 at 06:13:37PM +0100, Stefan Richter wrote:
> > It's already in the driver core to the most part.  It remains to be seen
> > what is less complicated in the end:  Transparent mutex-protected list
> > accesses provided by driver core (requires the iterator), or all the
> > necessary locking done by the drivers themselves (requires some more
> > lock-taking but perhaps fewer lock instances overall in the drivers, and
> > respective redefinitions and documentation of the driver core API).
>
> I favor changing the driver core api and doing this kind of thing there.
> It keeps the drivers simpler and should hopefully make their lives
> easier.

What about this?

#define class_for_each_dev(pos, head, member) \
        for (mutex_lock(&(container_of(head, struct class, devices))->mutex), po
s = list_entry((head)->next, typeof(*pos), member); \
        prefetch(pos->member.next), &pos->member != (head) ? 1 : (mutex_unlock(&
(container_of(head, struct class, devices))->mutex), 0); \
        pos = list_entry(pos->member.next, typeof(*pos), member))

>
> > Semi off-topic:  What about struct device.sem?  Is there any chance to
> > rip this out of the driver core and let drivers serialize everything?  I
> > suppose not...
>
> See the previous long threads about this very topic, that is what caused
> this class.sem patches :)
>
> thanks,
>
> greg k-h
>