Re: [linux-pm] [PATCH 2/4] PM / Loss: power loss management

["Rafael J. Wysocki" <rjw@sisk.pl>]

On Thursday, May 19, 2011, Alan Stern wrote:
> On Thu, 19 May 2011, Davide Ciminaghi wrote:
> 
> > I'm not completely sure about this. What we wanted to do was to avoid powering
> > down the mmc while it is physically writing data into its internal memory.
> > If we force a sync when the power loss warning event warning happens,
> > it is very difficult to be able to guarantee that all buffered data will be 
> > written before power actually dies. So we preferred to follow another strategy:
> > let the mmc finish any running write operation, and then stop its request 
> > queue. If power really goes down, then we hope that the file system journal 
> > will fix things on next boot (yes, some data could get lost, but the fs should
> > still be mountable). On the other hand, if power resumes, nothing bad should 
> > happen for user space processes.
> 
> You could consider a totally different approach.
> 
> Each platform will have a different set of high-power devices it wants
> to turn off when a power-loss warning occurs.  So instead of changing
> the core PM interface, you could add a new "power_loss" notifier list.  
> Only the most critical drivers would need to listen for notifications, 
> and this could be different drivers on different platforms.

Moreover, it would allow not only drivers, but also filesystems (for
one example) to get notifications.

Thanks,
Rafael