diff mbox

[v3] audio: make audio poll timer deterministic

Message ID 20170214071510.6112.76764.stgit@PASHA-ISP
State New
Headers show

Commit Message

Pavel Dovgalyuk Feb. 14, 2017, 7:15 a.m. UTC
This patch changes resetting strategy of the audio polling timer.
It does not change expiration time if the timer is already set.
This patch is needed to make this timer deterministic and to use execution
record/replay for audio devices.

audio_reset_timer is used in the function audio_vm_change_state_handler.
Therefore every time VM is stopped or restarted the timer will be reset
to new timeout. Virtual clock does not proceed while VM is stopped.
Therefore there is no need in resetting the timeout when VM restarts.

v2: updated commit message
v3: now using timer_mod_anticipate function (as suggested by Yurii Zubrytskyi)

Signed-off-by: Pavel Dovgalyuk <pavel.dovgaluk@ispras.ru>
---
 audio/audio.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Comments

Paolo Bonzini Feb. 14, 2017, 11:59 a.m. UTC | #1
On 14/02/2017 08:15, Pavel Dovgalyuk wrote:
> This patch changes resetting strategy of the audio polling timer.
> It does not change expiration time if the timer is already set.
> This patch is needed to make this timer deterministic and to use execution
> record/replay for audio devices.
> 
> audio_reset_timer is used in the function audio_vm_change_state_handler.
> Therefore every time VM is stopped or restarted the timer will be reset
> to new timeout. Virtual clock does not proceed while VM is stopped.
> Therefore there is no need in resetting the timeout when VM restarts.
> 
> v2: updated commit message
> v3: now using timer_mod_anticipate function (as suggested by Yurii Zubrytskyi)

Gerd, feel free to pick either v2 or v3.

Paolo
Gerd Hoffmann Feb. 16, 2017, 3:12 p.m. UTC | #2
On Di, 2017-02-14 at 12:59 +0100, Paolo Bonzini wrote:
> 
> On 14/02/2017 08:15, Pavel Dovgalyuk wrote:
> > This patch changes resetting strategy of the audio polling timer.
> > It does not change expiration time if the timer is already set.
> > This patch is needed to make this timer deterministic and to use execution
> > record/replay for audio devices.
> > 
> > audio_reset_timer is used in the function audio_vm_change_state_handler.
> > Therefore every time VM is stopped or restarted the timer will be reset
> > to new timeout. Virtual clock does not proceed while VM is stopped.
> > Therefore there is no need in resetting the timeout when VM restarts.
> > 
> > v2: updated commit message
> > v3: now using timer_mod_anticipate function (as suggested by Yurii Zubrytskyi)
> 
> Gerd, feel free to pick either v2 or v3.

Picked up v3 now.

thanks,
  Gerd
Pavel Dovgalyuk Feb. 16, 2017, 3:16 p.m. UTC | #3
> From: Gerd Hoffmann [mailto:kraxel@redhat.com]
> On Di, 2017-02-14 at 12:59 +0100, Paolo Bonzini wrote:
> > On 14/02/2017 08:15, Pavel Dovgalyuk wrote:
> > > This patch changes resetting strategy of the audio polling timer.
> > > It does not change expiration time if the timer is already set.
> > > This patch is needed to make this timer deterministic and to use execution
> > > record/replay for audio devices.
> > >
> > > audio_reset_timer is used in the function audio_vm_change_state_handler.
> > > Therefore every time VM is stopped or restarted the timer will be reset
> > > to new timeout. Virtual clock does not proceed while VM is stopped.
> > > Therefore there is no need in resetting the timeout when VM restarts.
> > >
> > > v2: updated commit message
> > > v3: now using timer_mod_anticipate function (as suggested by Yurii Zubrytskyi)
> >
> > Gerd, feel free to pick either v2 or v3.
> 
> Picked up v3 now.

Thanks!

And what about patch for audio record/replay?
http://lists.nongnu.org/archive/html/qemu-devel/2017-02/msg00345.html

Pavel Dovgalyuk
Gerd Hoffmann Feb. 16, 2017, 3:17 p.m. UTC | #4
On Do, 2017-02-16 at 18:16 +0300, Pavel Dovgalyuk wrote:
> > From: Gerd Hoffmann [mailto:kraxel@redhat.com]
> > On Di, 2017-02-14 at 12:59 +0100, Paolo Bonzini wrote:
> > > On 14/02/2017 08:15, Pavel Dovgalyuk wrote:
> > > > This patch changes resetting strategy of the audio polling timer.
> > > > It does not change expiration time if the timer is already set.
> > > > This patch is needed to make this timer deterministic and to use execution
> > > > record/replay for audio devices.
> > > >
> > > > audio_reset_timer is used in the function audio_vm_change_state_handler.
> > > > Therefore every time VM is stopped or restarted the timer will be reset
> > > > to new timeout. Virtual clock does not proceed while VM is stopped.
> > > > Therefore there is no need in resetting the timeout when VM restarts.
> > > >
> > > > v2: updated commit message
> > > > v3: now using timer_mod_anticipate function (as suggested by Yurii Zubrytskyi)
> > >
> > > Gerd, feel free to pick either v2 or v3.
> > 
> > Picked up v3 now.
> 
> Thanks!
> 
> And what about patch for audio record/replay?
> http://lists.nongnu.org/archive/html/qemu-devel/2017-02/msg00345.html

Picked that one too (busy preparing a audio pull req atm).

cheers,
  Gerd
diff mbox

Patch

diff --git a/audio/audio.c b/audio/audio.c
index c845a44..beadf7b 100644
--- a/audio/audio.c
+++ b/audio/audio.c
@@ -1112,7 +1112,7 @@  static int audio_is_timer_needed (void)
 static void audio_reset_timer (AudioState *s)
 {
     if (audio_is_timer_needed ()) {
-        timer_mod (s->ts,
+        timer_mod_anticipate_ns(s->ts,
             qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + conf.period.ticks);
     }
     else {