2015-05-13 14:39:53

by Koro Chen

Subject: [RFC PATCH] ALSA: pcm: Modify double acknowledged interrupts check condition

Currently in snd_pcm_update_hw_ptr0 during interrupt,
we consider there were double acknowledged interrupts when:
1. HW reported pointer is smaller than expected, and
2. Time from last update time (hdelta) is over half a buffer time.

However, when HW reported pointer is only a few bytes smaller than
expected, and when hdelta is just a little larger than half a buffer time
(e.g. ping-pong buffer), it wrongly treats this IRQ as double acknowledged.

The condition #2 uses jiffies, but jiffies is not high resolution
since it is integer. We should consider jiffies inaccuracy.

Signed-off-by: Koro Chen <[email protected]>
---
sound/core/pcm_lib.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/sound/core/pcm_lib.c b/sound/core/pcm_lib.c
index ac6b33f..7d45645 100644
--- a/sound/core/pcm_lib.c
+++ b/sound/core/pcm_lib.c
@@ -339,7 +339,7 @@ static int snd_pcm_update_hw_ptr0(struct snd_pcm_substream *substream,
if (delta > new_hw_ptr) {
/* check for double acknowledged interrupts */
hdelta = curr_jiffies - runtime->hw_ptr_jiffies;
- if (hdelta > runtime->hw_ptr_buffer_jiffies/2) {
+ if (hdelta > runtime->hw_ptr_buffer_jiffies/2 + 1) {
hw_base += runtime->buffer_size;
if (hw_base >= runtime->boundary) {
hw_base = 0;
--
1.8.1.1.dirty

2015-05-19 07:34:55

by Takashi Iwai

Subject: Re: [RFC PATCH] ALSA: pcm: Modify double acknowledged interrupts check condition

At Wed, 13 May 2015 22:39:03 +0800,
Koro Chen wrote:
>
> Currently in snd_pcm_update_hw_ptr0 during interrupt,
> we consider there were double acknowledged interrupts when:
> 1. HW reported pointer is smaller than expected, and
> 2. Time from last update time (hdelta) is over half a buffer time.
>
> However, when HW reported pointer is only a few bytes smaller than
> expected, and when hdelta is just a little larger than half a buffer time
> (e.g. ping-pong buffer), it wrongly treats this IRQ as double acknowledged.
>
> The condition #2 uses jiffies, but jiffies is not high resolution
> since it is integer. We should consider jiffies inaccuracy.
>
> Signed-off-by: Koro Chen <[email protected]>

The condition looks too strict, indeed. I applied the patch as is
now.

BTW, the similar check is already present for the free-wheeling case.
I'm going to clean up the code a bit for using the common code.


thanks,

Takashi


> ---
> sound/core/pcm_lib.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/sound/core/pcm_lib.c b/sound/core/pcm_lib.c
> index ac6b33f..7d45645 100644
> --- a/sound/core/pcm_lib.c
> +++ b/sound/core/pcm_lib.c
> @@ -339,7 +339,7 @@ static int snd_pcm_update_hw_ptr0(struct snd_pcm_substream *substream,
> if (delta > new_hw_ptr) {
> /* check for double acknowledged interrupts */
> hdelta = curr_jiffies - runtime->hw_ptr_jiffies;
> - if (hdelta > runtime->hw_ptr_buffer_jiffies/2) {
> + if (hdelta > runtime->hw_ptr_buffer_jiffies/2 + 1) {
> hw_base += runtime->buffer_size;
> if (hw_base >= runtime->boundary) {
> hw_base = 0;
> --
> 1.8.1.1.dirty
>