lkml.org 
[lkml]   [2009]   [Jun]   [8]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
SubjectRe: [linux-pm] Run-time PM idea (was: Re: [RFC][PATCH 0/2] PM: Rearrange core suspend code)
Date
Am Montag, 8. Juni 2009 13:29:26 schrieb Rafael J. Wysocki:

> But I need to be able to call __pm_schedule_resume() (at least) from
> interrupt context and I can't use a mutex from there. Otherwise I'd have
> used a mutex. :-)

I see.

> Anyway, below is a version with synchronous resume.

You are assuming autosuspend should always be with a delay. Why?

Secondly, you are not using a counter. Therefore only one driver can
control the PM state of a device at a given time. Is that wise?

> + * __pm_schedule_suspend - Schedule run-time suspend of given device.
> + * @dev: Device to suspend.
> + * @delay: Time to wait before attempting to suspend the device.

In which unit of time? If this is to go into kerneldoc that must be specified.

> + * @autocancel: If set, the request will be cancelled during a resume from
> a + * system-wide sleep state if it happens before @delay elapses.

Why is this needed?

> + */
> +void __pm_schedule_suspend(struct device *dev, unsigned long delay,
> + bool autocancel)

[..]


> +
> +/**
> + * __pm_schedule_resume - Schedule run-time resume of given device.
> + * @dev: Device to resume.
> + * @autocancel: If set, the request will be cancelled during a resume from
> a + * system-wide sleep state if it happens before pm_autoresume() can be
> run. + */

Eeek! This is a bad idea. You never want to a resume to be cancelled.

> +void __pm_schedule_resume(struct device *dev, bool autocancel)

[..]
> +int pm_resume_sync(struct device *dev)
> +{
> + int error = 0;
> +
> + pm_lock_device(dev);
> + if (dev->power.runtime_status == RPM_IDLE) {
> + /* ->autosuspend() hasn't started yet, no need to resume. */
> + pm_cancel_suspend(dev);
> + goto out;
> + }
> +
> + if (dev->power.runtime_status == RPM_SUSPENDING) {
> + /*
> + * The ->autosuspend() callback is being executed right now,
> + * wait for it to complete.
> + */
> + pm_unlock_device(dev);
> + cancel_delayed_work_sync(&dev->power.suspend_work);

That is the most glorious abuse of an API I've seen this year :-)

Regards
Oliver



\
 
 \ /
  Last update: 2009-06-08 14:05    [W:1.952 / U:0.272 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site