On Sat, 1 Mar 2008, Pierre Ossman wrote:

quoted text
> On Mon, 25 Feb 2008 15:00:51 -0500 (EST)
> Alan Stern <stern@rowland.harvard.edu> wrote:
> 
> > 
> > Maybe a better approach would be to leave the workqueue unfreezable,
> > and keep cancel_delayed_work_sync() in mmc_suspend_host().  It would
> > then be necessary to add a test to verify, if there is a card attached,
> > that the card is indeed suspended.  After all, it's possible that the
> > cancel_delayed_work_sync() ended up waiting for a job already running
> > on the workqueue to register a new card!  (The same would be true even 
> > with flush_scheduled_work.)
> 
> How would that be handled? I'd prefer a patch as I'm evidently not up to date with how to fondle the pm stuff properly. :)

The easiest way to do this will be to wait until some planned updates 
are added to the PM core; then this will be a quick and simple change.  
That probably means waiting until after 2.6.25 is released, however.

quoted text
> > Also, as a bit of defensive programming, it might be a good idea to add
> > a "suspended" flag to the mmc_host structure.  If mmc_rescan() sees
> > that the flag is set then it should return immediately.  This would
> > protect against host drivers that aren't careful to disable detect
> > IRQs before calling mmc_suspend_host().
> 
> Indeed. I'll add that to my todo.

That could go in along with the previous change.  The PM update
mentioned above involves adding a "suspended" flag to every struct 
device.  With that available, this amounts to nothing more than an 
extra test added at the beginning of mmc_rescan().

Alan Stern

--
unsubscribe notice
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/