Re: [patch] IA64: only call up() in salinfo_work_to_do() if down_trylock() was successful

[Keith Owens <kaos@ocs.com.au>]

Andrew Morton (on Wed, 30 Jul 2008 02:43:29 -0700) wrote:
>On Tue, 29 Jul 2008 23:07:45 +1000 Simon Horman <horms@verge.net.au> wrote:
>
>> On Tue, Jul 29, 2008 at 10:47:09PM +1000, Keith Owens wrote:
>> > Simon Horman (on Tue, 29 Jul 2008 19:47:20 +1000) wrote:
>> > >Aesthetic issues aside is it safe to call up() if down_trylock() failed?
>> > >
>> > >arch/ia64/kernel/salinfo.c: In function `salinfo_work_to_do':
>> > >arch/ia64/kernel/salinfo.c:195: warning: ignoring return value of `down_trylock'
>> > >
>> > >Signed-off-by: Simon Horman <horms@verge.net.au>
>> > >
>> > >Index: linux-2.6/arch/ia64/kernel/salinfo.c
>> > >===================================================================
>> > >--- linux-2.6.orig/arch/ia64/kernel/salinfo.c	2008-07-29 19:06:33.000000000 +1000
>> > >+++ linux-2.6/arch/ia64/kernel/salinfo.c	2008-07-29 19:40:02.000000000 +1000
>> > >@@ -192,8 +192,8 @@ struct salinfo_platform_oemdata_parms {
>> > > static void
>> > > salinfo_work_to_do(struct salinfo_data *data)
>> > > {
>> > >-	down_trylock(&data->mutex);
>> > >-	up(&data->mutex);
>> > >+	if (down_trylock(&data->mutex) == 0)
>> > >+		up(&data->mutex);
>> > > }
>> > > 
>> > > static void
>> > 
>> > NAK.  The whole point of this function is to set the mutex to the up
>> > state, irrespective of whether it was already down or not.  Tracking
>> > the state of data->mutex in all the possible contexts is just too
>> > fragile, especially since it can be modified from NMI context.
>> > salinfo_work_to_do() ensures that the mtuex ends in the up state.
>
>boggle.  I daren't look.
>
>> > To remove the warning, just stick '(void)' in front of down_trylock().
>> 
>> Thanks, will do.
>> 
>
>For gawd's sake add a comment there too.

You mean like the comment that is already in there?

/* Kick the mutex that tells user space that there is work to do.  Instead of
 * trying to track the state of the mutex across multiple cpus, in user
 * context, interrupt context, non-maskable interrupt context and hotplug cpu,
 * it is far easier just to grab the mutex if it is free then release it.
 *
 * This routine must be called with data_saved_lock held, to make the down/up
 * operation atomic.
 */