At Thu, 26 Mar 2015 14:10:17 +0100, Takashi Iwai wrote:
At Thu, 26 Mar 2015 20:24:43 +0800, Hui Wang wrote:
On 03/21/2015 02:38 PM, Hui Wang wrote:
On 03/21/2015 12:20 AM, David Henningsson wrote:
On 2015-03-18 09:50, Takashi Iwai wrote:
Hi,
here is a patchset for supporting more aggressive PM for HD-audio. This allows to change the power state of each widget more dynamically with jack and stream states. It's activated only when the codec driver (or via sysfs or f/w patch) sets codec->power_mgmt flag.
In theory, this should work for the recent Realtek codecs, but currently I have no machine for test.
David, could you or your team check whether this works for ALC282 or such? Just add like:
--- a/sound/pci/hda/patch_realtek.c +++ b/sound/pci/hda/patch_realtek.c @@ -5415,6 +5415,7 @@ static int patch_alc269(struct hda_codec *codec)
spec = codec->spec; spec->gen.shared_mic_vref_pin = 0x18;
codec->power_mgmt = 1;
snd_hda_pick_fixup(codec, alc269_fixup_models, alc269_fixup_tbl, alc269_fixups);
The patchset is for for-next branch of sound git tree, but they might be applicable to 4.0-rc (or even older), too. The current patches are found in topic/hda-power branch.
So I hoped to be able to look at this today, but it turns out the machine I was thinking of using for testing has an ALC262 codec, which hardly counts as "new".
Hui, is this something you feel like taking on? Otherwise I'll try to talk to someone in Taipei.
OK， I will look for the machine to do the test next week.
Regards, Hui.
Sorry for late response, today is my first day in the office back from vacation, I checked all machines in the Beijing office, none of them has the ALC282 codec, I will continue to look for the machine from other office.
And I did the test on the machines with the alc283, alc255, alc292 and alc269, the testing result were same, there were no sound output from internal speaker or headphone, and the internal mic or external mic can't record any sound. The test steps as below:
- power_save_node = 0
checkout the hda-power branch, build the kernel based on this branch. Install the kernel to the above machines and boot into the desktop test internal speaker and internal mic, works very well, plug a headset, test headphone and external mic, works very well. run pm_suspend, wait 5 seconds, wakeup the system, redo the above test, everything works very well.
OK, this is expected. The patch shouldn't touch this case.
- power_save_node = 1
enable the power_save_node as below: @@ -5426,6 +5426,8 @@ static int patch_alc269(struct hda_codec *codec)
alc_auto_parse_customize_define(codec);
codec->power_save_node = 1;if (has_cdefine_beep(codec)) spec->gen.beep_nid = 0x01;rebuild the kernel, install the kernel to the above machines and boot into the desktop test internal speaker and internal mic, we can play sound to internal speaker without any errors, but I can't hear any sound from the speaker; I can use the internal mic to record without errors, but recorded file did not include any sound pcm (maybe all 0x00 or 0xff) I plug a headset into the headset jack, the detection works very well, but I can't hear sound from headphone when play a sound, and I can't use headset mic to record any sound as well.
And I attached 2 alsa-info.txt, one is the power_save_node=0, the other is the power_save_node=1
Thanks. The alsa-info.sh outputs show no difference but the power state, so the widget attributes seem kept with the power state change, as it seems.
Could you give alsa-info.sh output *during* playing with power_save_node=1?
Also, try to pull topic/hda-regmap branch in addition, and apply the patch below. This implements the partial sync for the widget path. Note that the patch is totally untested.
thanks,
Takashi
diff --git a/include/sound/hda_regmap.h b/include/sound/hda_regmap.h index 76648ccfbbf8..727a5801e298 100644 --- a/include/sound/hda_regmap.h +++ b/include/sound/hda_regmap.h @@ -202,4 +202,16 @@ snd_hdac_regmap_update_amp_stereo(struct hdac_device *codec, hda_nid_t nid, return snd_hdac_regmap_update_raw(codec, cmd, mask, val); }
+/** + * snd_hdac_regmap_sync_node - sync the widget node attributes + * @codec: HD-audio codec + * @nid: NID to sync + */ +static inline void +snd_hdac_regmap_sync_node(struct hdac_device *codec, hda_nid_t nid) +{ + regcache_mark_dirty(codec->regmap); + regcache_sync_region(codec->regmap, nid << 20, ((nid + 1) << 20) -1); +} + #endif /* __SOUND_HDA_REGMAP_H */ diff --git a/sound/pci/hda/hda_generic.c b/sound/pci/hda/hda_generic.c index f7ccef5559de..1f2ca7be1468 100644 --- a/sound/pci/hda/hda_generic.c +++ b/sound/pci/hda/hda_generic.c @@ -842,10 +842,8 @@ static hda_nid_t path_power_update(struct hda_codec *codec, snd_hda_codec_write(codec, nid, 0, AC_VERB_SET_POWER_STATE, state); changed = nid; - /* here we assume that widget attributes (e.g. amp, - * pinctl connection) don't change with local power - * state change. If not, need to sync the cache. - */ + if (state == AC_PWRST_D0) + snd_hdac_regmap_sync_node(&codec->core, nid); } } return changed;