[PATCH v3 0/3] HDA controller PM and codec PM cleanups

[PATCH v3 0/3] HDA controller PM and codec PM cleanups

[Kai-Heng Feng]

While working on the issue "ALSA: hda: fix jack detection with Realtek
codecs when in D3", I've found using pm_runtime_force_{suspend,resume}()
have surprising behavior, specifically, pm_runtime_need_not_resume()
uses pm_runtime_need_not_resume() to avoid calling resume callback, so
jackpoll was used to really power up the codec.

We can use direct-complete to do the keep the codec suspended throughout
the system PM flow, namely, keep the codec suspended all the way, unless
the codec needs to be woken up after resume.

For HDA controller, PCI core may enable direct-complete for it if
conditions are met. So make runtime and system PM distinctive to always
apply correct wake up setting.

At least point, hopefully all runtime PM issues are solved, let's enable
runtime PM by default again.

v3:
 - Drop "ALSA: hda: Stop mangling PCI IRQ"

Kai-Heng Feng (3):
  ALSA: hda: Refactor codec PM to use direct-complete optimization
  ALSA: hda: Separate runtime and system suspend
  ALSA: hda: Reinstate runtime_allow() for all hda controllers

 sound/pci/hda/hda_codec.c      | 45 +++++++++++++++---------
 sound/pci/hda/hda_controller.h |  3 +-
 sound/pci/hda/hda_intel.c      | 63 +++++++++++++++++++---------------
 3 files changed, 66 insertions(+), 45 deletions(-)

-- 
2.17.1

[PATCH v3 1/3] ALSA: hda: Refactor codec PM to use direct-complete optimization

[Kai-Heng Feng]

Upon system resume, hda_codec_pm_resume() uses hda_codec_force_resume()
to resume the codec. However, pm_runtime_force_resume() won't really
resume the codec because of pm_runtime_need_not_resume() check.

Hence, hda_codec_force_resume() schedules a jackpoll work, which is to
really power up the codec.

Instead of doing that, we can use direct-complete to make the PM flow
more straightforward, and keep codec always suspended through system PM
flow if conditions are met.

On system suspend, PM core will decide what to do based on
hda_codec_pm_prepare():
- If codec is not runtime-suspended, PM core will suspend and resume the
device as normal.
- If codec is runtime-suspended, PM core will try to keep it suspended.
If it's still suspended after system resume, we use
hda_codec_pm_complete() to resume codec if it's needed.

Signed-off-by: Kai-Heng Feng <kai.heng.feng@canonical.com>
---
v3: 
 - No change

v2:
 - Also resume when codec->jackpoll_interval is set

 sound/pci/hda/hda_codec.c | 45 +++++++++++++++++++++++++--------------
 1 file changed, 29 insertions(+), 16 deletions(-)

diff --git a/sound/pci/hda/hda_codec.c b/sound/pci/hda/hda_codec.c
index a356c21edb90..4bb58e8b08a8 100644
--- a/sound/pci/hda/hda_codec.c
+++ b/sound/pci/hda/hda_codec.c
@@ -2934,7 +2934,7 @@ static void hda_call_codec_resume(struct hda_codec *codec)
 	snd_hdac_leave_pm(&codec->core);
 }
 
-static int hda_codec_runtime_suspend(struct device *dev)
+static int hda_codec_suspend(struct device *dev)
 {
 	struct hda_codec *codec = dev_to_hda_codec(dev);
 	unsigned int state;
@@ -2953,7 +2953,7 @@ static int hda_codec_runtime_suspend(struct device *dev)
 	return 0;
 }
 
-static int hda_codec_runtime_resume(struct device *dev)
+static int hda_codec_resume(struct device *dev)
 {
 	struct hda_codec *codec = dev_to_hda_codec(dev);
 
@@ -2967,57 +2967,70 @@ static int hda_codec_runtime_resume(struct device *dev)
 	pm_runtime_mark_last_busy(dev);
 	return 0;
 }
+
+static int hda_codec_runtime_suspend(struct device *dev)
+{
+	return hda_codec_suspend(dev);
+}
+
+static int hda_codec_runtime_resume(struct device *dev)
+{
+	return hda_codec_resume(dev);
+}
+
 #endif /* CONFIG_PM */
 
 #ifdef CONFIG_PM_SLEEP
-static int hda_codec_force_resume(struct device *dev)
+static int hda_codec_pm_prepare(struct device *dev)
+{
+	return pm_runtime_suspended(dev);
+}
+
+static void hda_codec_pm_complete(struct device *dev)
 {
 	struct hda_codec *codec = dev_to_hda_codec(dev);
-	int ret;
 
-	ret = pm_runtime_force_resume(dev);
-	/* schedule jackpoll work for jack detection update */
-	if (codec->jackpoll_interval ||
-	    (pm_runtime_suspended(dev) && hda_codec_need_resume(codec)))
-		schedule_delayed_work(&codec->jackpoll_work,
-				      codec->jackpoll_interval);
-	return ret;
+	if (pm_runtime_suspended(dev) && (codec->jackpoll_interval ||
+	    hda_codec_need_resume(codec) || codec->forced_resume))
+		pm_request_resume(dev);
 }
 
 static int hda_codec_pm_suspend(struct device *dev)
 {
 	dev->power.power_state = PMSG_SUSPEND;
-	return pm_runtime_force_suspend(dev);
+	return hda_codec_suspend(dev);
 }
 
 static int hda_codec_pm_resume(struct device *dev)
 {
 	dev->power.power_state = PMSG_RESUME;
-	return hda_codec_force_resume(dev);
+	return hda_codec_resume(dev);
 }
 
 static int hda_codec_pm_freeze(struct device *dev)
 {
 	dev->power.power_state = PMSG_FREEZE;
-	return pm_runtime_force_suspend(dev);
+	return hda_codec_suspend(dev);
 }
 
 static int hda_codec_pm_thaw(struct device *dev)
 {
 	dev->power.power_state = PMSG_THAW;
-	return hda_codec_force_resume(dev);
+	return hda_codec_resume(dev);
 }
 
 static int hda_codec_pm_restore(struct device *dev)
 {
 	dev->power.power_state = PMSG_RESTORE;
-	return hda_codec_force_resume(dev);
+	return hda_codec_resume(dev);
 }
 #endif /* CONFIG_PM_SLEEP */
 
 /* referred in hda_bind.c */
 const struct dev_pm_ops hda_codec_driver_pm = {
 #ifdef CONFIG_PM_SLEEP
+	.prepare = hda_codec_pm_prepare,
+	.complete = hda_codec_pm_complete,
 	.suspend = hda_codec_pm_suspend,
 	.resume = hda_codec_pm_resume,
 	.freeze = hda_codec_pm_freeze,
-- 
2.17.1

[PATCH v3 2/3] ALSA: hda: Separate runtime and system suspend

[Kai-Heng Feng]

Both pm_runtime_force_suspend() and pm_runtime_force_resume() have
some implicit checks, so it can make code flow more straightforward if
we separate runtime and system suspend callbacks.

High Definition Audio Specification, 4.5.9.3 Codec Wake From System S3
states that codec can wake the system up from S3 if WAKEEN is toggled.
Since HDA controller has different wakeup settings for runtime and
system susend, we also need to explicitly disable direct-complete which
can be enabled automatically by PCI core. In addition to that, avoid
waking up codec if runtime resume is for system suspend, to not break
direct-complete for codecs.

While at it, also remove AZX_DCAPS_SUSPEND_SPURIOUS_WAKEUP, as the
original bug commit a6630529aecb ("ALSA: hda: Workaround for spurious
wakeups on some Intel platforms") solves doesn't happen with this
patch.

Signed-off-by: Kai-Heng Feng <kai.heng.feng@canonical.com>
---
v3:
 - Fix typos.
 - Remove from_rt from __azx_runtime_resume().
 - Change prepared to pm_prepared.
 - Concatenate azx_writew() into one line.

v2:
 - Make sure WAKEEN is disabled for system suspend.
 - Avoid waking up codec if runtime resume is for system suspend

 sound/pci/hda/hda_controller.h |  3 +-
 sound/pci/hda/hda_intel.c      | 62 +++++++++++++++++++---------------
 2 files changed, 36 insertions(+), 29 deletions(-)

diff --git a/sound/pci/hda/hda_controller.h b/sound/pci/hda/hda_controller.h
index be63ead8161f..68f9668788ea 100644
--- a/sound/pci/hda/hda_controller.h
+++ b/sound/pci/hda/hda_controller.h
@@ -41,7 +41,7 @@
 /* 24 unused */
 #define AZX_DCAPS_COUNT_LPIB_DELAY  (1 << 25)	/* Take LPIB as delay */
 #define AZX_DCAPS_PM_RUNTIME	(1 << 26)	/* runtime PM support */
-#define AZX_DCAPS_SUSPEND_SPURIOUS_WAKEUP (1 << 27) /* Workaround for spurious wakeups after suspend */
+/* 27 unused */
 #define AZX_DCAPS_CORBRP_SELF_CLEAR (1 << 28)	/* CORBRP clears itself after reset */
 #define AZX_DCAPS_NO_MSI64      (1 << 29)	/* Stick to 32-bit MSIs */
 #define AZX_DCAPS_SEPARATE_STREAM_TAG	(1 << 30) /* capture and playback use separate stream tag */
@@ -143,6 +143,7 @@ struct azx {
 	unsigned int align_buffer_size:1;
 	unsigned int region_requested:1;
 	unsigned int disabled:1; /* disabled by vga_switcheroo */
+	unsigned int pm_prepared:1;
 
 	/* GTS present */
 	unsigned int gts_present:1;
diff --git a/sound/pci/hda/hda_intel.c b/sound/pci/hda/hda_intel.c
index 749b88090970..be83f1747c5f 100644
--- a/sound/pci/hda/hda_intel.c
+++ b/sound/pci/hda/hda_intel.c
@@ -297,8 +297,7 @@ enum {
 /* PCH for HSW/BDW; with runtime PM */
 /* no i915 binding for this as HSW/BDW has another controller for HDMI */
 #define AZX_DCAPS_INTEL_PCH \
-	(AZX_DCAPS_INTEL_PCH_BASE | AZX_DCAPS_PM_RUNTIME |\
-	 AZX_DCAPS_SUSPEND_SPURIOUS_WAKEUP)
+	(AZX_DCAPS_INTEL_PCH_BASE | AZX_DCAPS_PM_RUNTIME)
 
 /* HSW HDMI */
 #define AZX_DCAPS_INTEL_HASWELL \
@@ -985,7 +984,7 @@ static void __azx_runtime_suspend(struct azx *chip)
 	display_power(chip, false);
 }
 
-static void __azx_runtime_resume(struct azx *chip, bool from_rt)
+static void __azx_runtime_resume(struct azx *chip)
 {
 	struct hda_intel *hda = container_of(chip, struct hda_intel, chip);
 	struct hdac_bus *bus = azx_bus(chip);
@@ -1002,7 +1001,8 @@ static void __azx_runtime_resume(struct azx *chip, bool from_rt)
 	azx_init_pci(chip);
 	hda_intel_init_chip(chip, true);
 
-	if (from_rt) {
+	/* Avoid codec resume if runtime resume is for system suspend */
+	if (!chip->pm_prepared) {
 		list_for_each_codec(codec, &chip->bus) {
 			if (codec->relaxed_resume)
 				continue;
@@ -1018,6 +1018,29 @@ static void __azx_runtime_resume(struct azx *chip, bool from_rt)
 }
 
 #ifdef CONFIG_PM_SLEEP
+static int azx_prepare(struct device *dev)
+{
+	struct snd_card *card = dev_get_drvdata(dev);
+	struct azx *chip;
+
+	chip = card->private_data;
+	chip->pm_prepared = 1;
+
+	/* HDA controller always requires different WAKEEN for runtime suspend
+	 * and system suspend, so don't use direct-complete here.
+	 */
+	return 0;
+}
+
+static void azx_complete(struct device *dev)
+{
+	struct snd_card *card = dev_get_drvdata(dev);
+	struct azx *chip;
+
+	chip = card->private_data;
+	chip->pm_prepared = 0;
+}
+
 static int azx_suspend(struct device *dev)
 {
 	struct snd_card *card = dev_get_drvdata(dev);
@@ -1029,15 +1052,7 @@ static int azx_suspend(struct device *dev)
 
 	chip = card->private_data;
 	bus = azx_bus(chip);
-	snd_power_change_state(card, SNDRV_CTL_POWER_D3hot);
-	/* An ugly workaround: direct call of __azx_runtime_suspend() and
-	 * __azx_runtime_resume() for old Intel platforms that suffer from
-	 * spurious wakeups after S3 suspend
-	 */
-	if (chip->driver_caps & AZX_DCAPS_SUSPEND_SPURIOUS_WAKEUP)
-		__azx_runtime_suspend(chip);
-	else
-		pm_runtime_force_suspend(dev);
+	__azx_runtime_suspend(chip);
 	if (bus->irq >= 0) {
 		free_irq(bus->irq, chip);
 		bus->irq = -1;
@@ -1066,11 +1081,7 @@ static int azx_resume(struct device *dev)
 	if (azx_acquire_irq(chip, 1) < 0)
 		return -EIO;
 
-	if (chip->driver_caps & AZX_DCAPS_SUSPEND_SPURIOUS_WAKEUP)
-		__azx_runtime_resume(chip, false);
-	else
-		pm_runtime_force_resume(dev);
-	snd_power_change_state(card, SNDRV_CTL_POWER_D0);
+	__azx_runtime_resume(chip);
 
 	trace_azx_resume(chip);
 	return 0;
@@ -1118,10 +1129,7 @@ static int azx_runtime_suspend(struct device *dev)
 	chip = card->private_data;
 
 	/* enable controller wake up event */
-	if (snd_power_get_state(card) == SNDRV_CTL_POWER_D0) {
-		azx_writew(chip, WAKEEN, azx_readw(chip, WAKEEN) |
-			   STATESTS_INT_MASK);
-	}
+	azx_writew(chip, WAKEEN, azx_readw(chip, WAKEEN) | STATESTS_INT_MASK);
 
 	__azx_runtime_suspend(chip);
 	trace_azx_runtime_suspend(chip);
@@ -1132,18 +1140,14 @@ static int azx_runtime_resume(struct device *dev)
 {
 	struct snd_card *card = dev_get_drvdata(dev);
 	struct azx *chip;
-	bool from_rt = snd_power_get_state(card) == SNDRV_CTL_POWER_D0;
 
 	if (!azx_is_pm_ready(card))
 		return 0;
 	chip = card->private_data;
-	__azx_runtime_resume(chip, from_rt);
+	__azx_runtime_resume(chip);
 
 	/* disable controller Wake Up event*/
-	if (from_rt) {
-		azx_writew(chip, WAKEEN, azx_readw(chip, WAKEEN) &
-			   ~STATESTS_INT_MASK);
-	}
+	azx_writew(chip, WAKEEN, azx_readw(chip, WAKEEN) & ~STATESTS_INT_MASK);
 
 	trace_azx_runtime_resume(chip);
 	return 0;
@@ -1177,6 +1181,8 @@ static int azx_runtime_idle(struct device *dev)
 static const struct dev_pm_ops azx_pm = {
 	SET_SYSTEM_SLEEP_PM_OPS(azx_suspend, azx_resume)
 #ifdef CONFIG_PM_SLEEP
+	.prepare = azx_prepare,
+	.complete = azx_complete,
 	.freeze_noirq = azx_freeze_noirq,
 	.thaw_noirq = azx_thaw_noirq,
 #endif
-- 
2.17.1

[PATCH v3 3/3] ALSA: hda: Reinstate runtime_allow() for all hda controllers

[Kai-Heng Feng]

The broken jack detection should be fixed by commit a6e7d0a4bdb0 ("ALSA:
hda: fix jack detection with Realtek codecs when in D3"), let's try
enabling runtime PM by default again.

Signed-off-by: Kai-Heng Feng <kai.heng.feng@canonical.com>
---
v3:
 - No change.

v2:
 - No change.

 sound/pci/hda/hda_intel.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/sound/pci/hda/hda_intel.c b/sound/pci/hda/hda_intel.c
index be83f1747c5f..d539f52009a1 100644
--- a/sound/pci/hda/hda_intel.c
+++ b/sound/pci/hda/hda_intel.c
@@ -2362,6 +2362,7 @@ static int azx_probe_continue(struct azx *chip)
 
 	if (azx_has_pm_runtime(chip)) {
 		pm_runtime_use_autosuspend(&pci->dev);
+		pm_runtime_allow(&pci->dev);
 		pm_runtime_put_autosuspend(&pci->dev);
 	}
 
-- 
2.17.1

Re: [PATCH v3 0/3] HDA controller PM and codec PM cleanups

[Takashi Iwai]

On Tue, 27 Oct 2020 14:00:35 +0100,
Kai-Heng Feng wrote:
> 
> While working on the issue "ALSA: hda: fix jack detection with Realtek
> codecs when in D3", I've found using pm_runtime_force_{suspend,resume}()
> have surprising behavior, specifically, pm_runtime_need_not_resume()
> uses pm_runtime_need_not_resume() to avoid calling resume callback, so
> jackpoll was used to really power up the codec.
> 
> We can use direct-complete to do the keep the codec suspended throughout
> the system PM flow, namely, keep the codec suspended all the way, unless
> the codec needs to be woken up after resume.
> 
> For HDA controller, PCI core may enable direct-complete for it if
> conditions are met. So make runtime and system PM distinctive to always
> apply correct wake up setting.
> 
> At least point, hopefully all runtime PM issues are solved, let's enable
> runtime PM by default again.
> 
> v3:
>  - Drop "ALSA: hda: Stop mangling PCI IRQ"
> 
> Kai-Heng Feng (3):
>   ALSA: hda: Refactor codec PM to use direct-complete optimization
>   ALSA: hda: Separate runtime and system suspend
>   ALSA: hda: Reinstate runtime_allow() for all hda controllers

Now I applied all three patches.
Since it's a nice improvement, I'm going to push to 5.10.


thanks,

Takashi