At Fri, 26 Oct 2012 10:45:33 +0000, Wang, Xingchao wrote:
-----Original Message----- From: Takashi Iwai [mailto:tiwai@suse.de] Sent: Thursday, October 25, 2012 7:08 PM To: Wang, Xingchao Cc: alsa-devel@alsa-project.org Subject: Re: [PATCH] ALSA: HDA - Check return value to reduce useless delay
At Thu, 25 Oct 2012 10:30:35 +0000, Wang, Xingchao wrote:
-----Original Message----- From: Takashi Iwai [mailto:tiwai@suse.de] Sent: Thursday, October 25, 2012 6:01 AM To: Wang, Xingchao Cc: alsa-devel@alsa-project.org Subject: Re: [PATCH] ALSA: HDA - Check return value to reduce useless delay
At Wed, 24 Oct 2012 14:53:23 +0800, Wang Xingchao wrote:
For verb 705h, it's useless to read response, so use *write api would be better. If there's error after sending cmd, just try again without continue after wrong operation.Otherwise there's long
time delay.
Well, this is a bit sensitive part. Did you do the good test coverage for different hardware controllers and codecs, including the
non-Intel ones?
Well I only tested the patch on Haswell/Ivybridge platform.
Both are too new :)
For Haswell I meet an issue about codec suspend/resume. There's no response from codec when there's power state transition from D3
to D0, and the patch could reduced delay time.
But does it fix the issue itself?
Actually not. :)
For ivybridge the codec works well, so this patch has no active behavior. I assume the verb 705h will not work for subnodes if it failed at first for
"function group".
Is there such case when set power state for the "Audio Function Group" fail
but the sub-nodes could keep on setting power?
Not sure. But we need more testing obviously before breaking anything.
Okay, I will try to test it on other platforms. :)
Thanks.
But thinking this twice, I think the only case it matters is that when the codec returns a bogus error at the power up. As the original code, it continues to power up all sub nodes. But with your patch, the sub node power-up will be skipped completely.
So, a compromise option would be:
- Keep using snd_hda_codec_read() to sync with the power up of FG node; snd_hda_codec_write() is asynchronous, so this can be too fast for power up of sub nodes, and...
- Check the return value from the fg node, skip the rest and continue, except for the last try. (Maybe we should add some delay for the repeated try here, though.) And in the last loop count, forcibly continue to the power up of the rest.
In this way, most of time consumed for the broken power up case you faced should be avoided.
Takashi
--xingchao
thanks,
Takashi
Thanks --xingchao
Takashi
Signed-off-by: Wang Xingchao xingchao.wang@intel.com
sound/pci/hda/hda_codec.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-)
diff --git a/sound/pci/hda/hda_codec.c b/sound/pci/hda/hda_codec.c index 960800b..0946eca 100644 --- a/sound/pci/hda/hda_codec.c +++ b/sound/pci/hda/hda_codec.c @@ -3569,6 +3569,7 @@ static unsigned int hda_set_power_state(struct
hda_codec *codec,
hda_nid_t fg = codec->afg ? codec->afg : codec->mfg; int count; unsigned int state;
int err;
/* this delay seems necessary to avoid click noise at power-down */ if (power_state == AC_PWRST_D3) { @@ -3582,9 +3583,11 @@
static
unsigned int hda_set_power_state(struct
hda_codec *codec,
codec->patch_ops.set_power_state(codec, fg, power_state); else {
snd_hda_codec_read(codec, fg, 0,
err = snd_hda_codec_write(codec, fg, 0, AC_VERB_SET_POWER_STATE, power_state);if (err < 0)continue; snd_hda_codec_set_power_to_all(codec, fg, power_state, true);-- 1.7.9.5