Re: [PATCH v2] MMC: fix all hangs related to mmc/sd card insert/removal during suspend/resume.

[Maxim Levitsky <maximlevitsky@gmail.com>]

On Mon, 2010-06-14 at 16:01 -0700, Andrew Morton wrote: 
> On Fri, 11 Jun 2010 22:19:55 +0300
> Maxim Levitsky <maximlevitsky@gmail.com> wrote:
> 
> > If you don't use CONFIG_MMC_UNSAFE_RESUME, card will now be removed
> > in pm notified while userspace is still running.
> > Thus it will be possible to sync it propely.
> > 
> > Card detect workqueue is now freezeable, therefore a card insert/removal event will
> > wait till userspace is unfrozen.
> > 
> > Tested with and without CONFIG_MMC_UNSAFE_RESUME with suspend and hibernate.
> > 
> >
> > ...
> >
> > @@ -158,6 +162,8 @@ void mmc_remove_host(struct mmc_host *host)
> >  	device_del(&host->class_dev);
> >  
> >  	led_trigger_unregister_simple(host->led);
> > +
> > +	unregister_pm_notifier(&host->pm_notify);
> >  }
> 
> This looks a little risky.  There's a window where the pm notifier
> remains registered after we've done the device_del() and the
> led_trigger_unregister_simple().
> 
> I don't know if the code's really buggy, nor if it might become buggy
> in the future as things evolve.  But as register_pm_notifier() is the
> last thing we do before mmc_start_host(), I'd have though that
> unregister_pm_notifier() should be the first thing we do after
> mmc_stop_host()?

Absolutely right!


> 
> >
> > ...
> >
> > --- a/include/linux/mmc/host.h
> > +++ b/include/linux/mmc/host.h
> > @@ -124,6 +124,7 @@ struct mmc_host {
> >  	unsigned int		f_min;
> >  	unsigned int		f_max;
> >  	u32			ocr_avail;
> > +	struct notifier_block	pm_notify;
> >  
> >  #define MMC_VDD_165_195		0x00000080	/* VDD voltage 1.65 - 1.95 */
> >  #define MMC_VDD_20_21		0x00000100	/* VDD voltage 2.0 ~ 2.1 */
> > @@ -257,6 +258,8 @@ int mmc_card_can_sleep(struct mmc_host *host);
> >  int mmc_host_enable(struct mmc_host *host);
> >  int mmc_host_disable(struct mmc_host *host);
> >  int mmc_host_lazy_disable(struct mmc_host *host);
> > +int mmc_pm_notify(struct notifier_block *notify_block, unsigned long, void *);
> 
> It's unusual to provide names for some of the arguments and to leave
> them out for others.
Will fix that too.



Best regards,
Maxim Levitsky