On Thu, Jun 21, 2012 at 10:59 PM, Takashi Iwai tiwai@suse.de wrote:

At Thu, 21 Jun 2012 21:51:22 -0700, Dylan Reid wrote:

...

After cancel_delayed_work_sync returns, the power down work either never started (power_on == 1) or finished (power_on == 0).  In the former case there is no need to power up again.

Signed-off-by: Dylan Reid dgreid@chromium.org

There is still a small window between cancel_delayed_work_sync() and the spinlock, so the power down might happen between them, in theory. (And due to the obvious reason, cancel_delayed_work_sync() can't be in  spinlock :)

I'm probably missing something here, but I think three scenarios could happen during that window:

1) snd_hda_suspend or snd_hda_codec_reset, which will leave power_on = 0, and when power_up gets the spin lock it will turn back on. I don't think that's a problem, power_count is still > 0 so it should be on.

2) snd_hda_power_down is called and power_count goes to zero while power_on == 0. Might be a problem, this would not schedule new power down work. Is this expected? One thread calling power_up moving the count from 0 to 1 and another thread (that must not have called power_up) calling power_down? This hole could be plugged by checking if power_count is still >0 after getting the spin lock. does that sound right?

3) snd_hda_power_down is called and schedules new work, I think this is OK because when hda_power_work runs it will check power_count and do the right thing.

Thanks for the quick response,

Dylan

thanks,

Takashi

...

---

sound/pci/hda/hda_codec.c |    7 +++++++  1 files changed, 7 insertions(+), 0 deletions(-)

diff --git a/sound/pci/hda/hda_codec.c b/sound/pci/hda/hda_codec.c index 604699c..045b5e7 100644 --- a/sound/pci/hda/hda_codec.c +++ b/sound/pci/hda/hda_codec.c @@ -4444,6 +4444,13 @@ static void __snd_hda_power_up(struct hda_codec *codec, bool wait_power_down)       cancel_delayed_work_sync(&codec->power_work);

spin_lock(&codec->power_lock);

• /* If the power down delayed work was cancelled above before starting,
• * then there is no need to go through power up here.
• */
• if (codec->power_on) {
• spin_unlock(&codec->power_lock);
• return;
• }

trace_hda_power_up(codec);       snd_hda_update_power_acct(codec);       codec->power_on = 1; -- 1.7.9.rc0