Re: Recent and not-so problems with tifm_sd driver

[Pierre Ossman <drzeus-mmc@drzeus.cx>]

Alex Dubov wrote:
> This is hard to trigger problem, so I'll spare you the rather lengthy log.
> It happens if card timeouts and mmc_remove_host is called while mmc_register_card is still in
> progress (the hint was in crash dump). If I sleep before remove, it gives the mmc_register_card
> chance to finish/mark card as dead and everything's fine.
> The remove callback of mmc_block is apparently not executed in this case (probably because device
> has not finished registering).

Let's see, you manage to call mmc_remove_host() when a detection is pending in
the workqueue? That could indeed generate new request (up until mmc_free_host()
is called), but as the card is removed the mmc layer shouldn't be able to detect
anything, hence mmc_block should never get involved.

> 
> My proposition: lets flush the workqueue first thing in the mmc_remove_host (and make sure that
> nothing gets scheduled into it after this). 
> 

This is fixed in -mm. Making sure that nothing gets scheduled is the driver's
responsibility. I've added checks to catch broken drivers though.

Rgds
-- 
     -- Pierre Ossman

  Linux kernel, MMC maintainer        http://www.kernel.org
  PulseAudio, core developer          http://pulseaudio.org
  rdesktop, core developer          http://www.rdesktop.org