Re: PM interface to suspend/resume individual/specific device

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

On Monday 06 July 2009, HU TAO-TGHK48 wrote:
> 
> Hi, Rafael
> 
> Thank for the answer.
> 
> 2 more things
> A) The suspend/resume API assumes that  all user space process/kthread
> access to driver will be frozen first
>     This avoids race condition that processes/threads  are accessing
> driver while suspend()/resume() of the driver is called at the same
> time.
>     
>     Is my understanding correct?

Generally, it is, but while all user space processes are frozen before every
attempt to put the system into a sleep state, only some of the kernel threads
are frozen.

> B) I noticed your "Run-time PM framework" patch was submitted recently
>     Do you think it is fair to allow user space to trigger run-time
> suspend/resume for specific device?
>
>     Sine the run-time suspend/resume would be system independent per my
> understanding.

There's a problem that the user space can potentially trigger a run-time
suspend of a device at a wrong time.  It's probably better to let the driver
decide.

Best,
Rafael