Re: [PATCH] use spinlock instead of binary mutex in idt77252 driver

["Satyam Sharma" <satyam.sharma@gmail.com>]

Hi,

On 4/23/07, Eddie C. Dost <ecd@atecom.com> wrote:
> as long as mutexes are not converted to nop when CONFIG_SMP is not
> defined (I don't know what current kernels do), this is of course
> correct. You need to verify the headerfiles for the above.

Yes, even on UP different threads accessing the same data could race.
Mutexes (== binary semaphores) are *required* to synchronize access to
shared data.

You might be confusing mutexes with spinlocks. Spinlocks _are_
compiled away on UP (actually !CONFIG_SMP && !CONFIG_PREEMPT) kernels,
but that is still safe because spinlocks are busy-waiting loops
(unlike mutexes and semaphores that block) and hence no thread is
allowed to sleep when holding a spinlock.

> On Mon, Apr 23, 2007 at 09:40:26AM +0200, Matthias Kaehlcke wrote:
> > El Mon, Apr 23, 2007 at 09:16:08AM +0200 Eddie C. Dost ha dit:
> > > Please note that the semaphore is used to lock the idt77252 config
> > > tables among multiple users including atmsigd even on single processor
> > > machines. Does this work with mutexes?
> >
> > afaik mutexes have the same behaviour as binary semaphores that are
> > used as mutexes (always locked and unlocked by the same
> > process/thread):

Mutexes / binary semaphores / spinlocks are used to synchronize access
to shared data by *multiple* threads ... there is no meaning in
locking access to something if we know only one thread will ever touch
it.

Cheers,
S