Re: [PATCH, RFC 1/3] Add sem_getcount() to arches that lack it

[Andrew Morton <akpm@osdl.org>]

Jody McIntyre <scjody@modernduck.com> wrote:
>
> On Thu, Mar 10, 2005 at 08:55:03PM -0800, Andrew Morton wrote:
> > Jody McIntyre <scjody@modernduck.com> wrote:
> > >
> > > parisc and frv define sem_getcount() in semaphore.h, which returns the
> > >  current semaphore value.  This is cleaner than doing
> > >  atomic_read(&semaphore.count), currently done in
> > >  drivers/ieee1394/nodemgr.c and fs/xfs/linux-2.6/xfs_buf.c, and will work
> > >  on all architectures if sem_getcount() is added.
> > 
> > That's a fairly bizarre thing to want to do.  Would it be hard to modify
> > xfs and 1394 to stop wanting to read a semaphore's up() count?
> 
> The count is the number of free transaction labels (1394 async is
> transaction-based) and is initialized to 64.  When a new transaction label
> is needed, the requestor does a down(), then locks the tlabel variables
> and allocates a new one.  When a transaction label is freed, an up()
> occurs.  The semaphore's up() count is therefore the number of free
> tlabels, and the number of outstanding transactions is (64 - count).  I
> can imagine situations in which this would be a useful statistic, but
> I'm not sure any of them actually exist.

That's a nice application of semaphores.  I can see that there's also a
need to be able to read the value back for reporting purposes.  Dang.

But I guess it's a bit hard to justify adding more infrastructure to
support a single callsite which has a simple alternative.  So if you could
please add the separate counter?

> ...
> 
> If this patch isn't accepted, we should get rid of the xfs and 1394
> hacks and delete sem_getcount (parisc, frv) and sema_count (arm) as they
> are unused.

True.