Skip to contentSkip to conversation

Support wm_adsp hibernation for runtime suspend

PATCH In review Stefan Binding 3 participants 8 messages Feb 24 – Mar 3

Subsystems (5)
lore
Share & tools
lore
#0 lore raw

v2 cover Support wm_adsp hibernation for runtime suspend

When the CS35L41 and CS35L45 drivers suspend, they are put into hibernation, and the regmap goes into cache_only, but the firmware is still running, and wm_adsp is not stopped. If userspace attempts to read a firmware control, it will perform a regmap_raw_read() and this will produce an error in the kernel log.

To prevent these spurious errors, add an apis into cs_dsp and wm_adsp to allow wm_adsp to hibernate. In this hibernation mode, reads or writes to the dsp controls would be rejected with -EPERM rather than -EBUSY, and no error will be printed to the kernel log.

Changes Since v1:
- Fix minor error in cs_dsp API documentation
- Add similar patch to use new wm_adsp api for CS35L45

Ricardo Rivera-Matos (1):
ASoC: cs35l45: Hibernate wm_adsp on runtime suspend

Stefan Binding (3):
firmware: cs_dsp: Add API to hibernate the DSP
ASoC: codecs: wm_adsp: Allow wm_adsp to hibernate without stopping DSP
ASoC: cs35l41: Hibernate wm_adsp on runtime suspend

 drivers/firmware/cirrus/cs_dsp.c       | 49 +++++++++++++++++++++++---
 include/linux/firmware/cirrus/cs_dsp.h |  3 ++
 sound/soc/codecs/cs35l41.c             |  5 +++
 sound/soc/codecs/cs35l45.c             |  3 ++
 sound/soc/codecs/wm_adsp.c             |  6 ++++
 sound/soc/codecs/wm_adsp.h             |  1 +
 6 files changed, 63 insertions(+), 4 deletions(-)
Signature · 1 line
2.43.0
#1 lore raw

PATCH v2 1/4 firmware: cs_dsp: Add API to hibernate the DSP

For some parts, the DSP is kept running when in low power mode (hibernation), leaving the firmware ALSA controls enabled, but the registers are inaccessible. Attempts to access volatile firmware controls whilst in this state would produce errors in the kernel log due to a regmap_raw_read() into DSP registers whilst the regmap is in cache_only.

To remove this error log, add a hibernating flag to indicate that the controls are inaccessible, so we no longer try to read or write to the registers whilst the regmap is in cache_only.

This would still produce an error when trying to read or write to these controls, but this would be a different error (-EPERM instead of -EBUSY), and would not produce a spurious error log in the kernel.

Upon wake from hibernation, the control caches are re-synced to the hardware, if the DSP is running.

---
 drivers/firmware/cirrus/cs_dsp.c       | 49 +++++++++++++++++++++++---
 include/linux/firmware/cirrus/cs_dsp.h |  3 ++
 2 files changed, 48 insertions(+), 4 deletions(-)
drivers/firmware/cirrus/cs_dsp.c +45 -4
index b4f1c01e3b5b..f9d8a883900d 100644
@@ -515,6 +515,7 @@ void cs_dsp_init_debugfs(struct cs_dsp *dsp, struct dentry *debugfs_root)
 
 	debugfs_create_bool("booted", 0444, root, &dsp->booted);
 	debugfs_create_bool("running", 0444, root, &dsp->running);
+	debugfs_create_bool("hibernating", 0444, root, &dsp->hibernating);
 	debugfs_create_x32("fw_id", 0444, root, &dsp->fw_id);
 	debugfs_create_x32("fw_version", 0444, root, &dsp->fw_id_version);
 
@@ -703,7 +704,7 @@ int cs_dsp_coeff_write_acked_control(struct cs_dsp_coeff_ctl *ctl, unsigned int
 
 	lockdep_assert_held(&dsp->pwr_lock);
 
-	if (!dsp->running)
+	if (!dsp->running || dsp->hibernating)
 		return -EPERM;
 
 	ret = cs_dsp_coeff_base_reg(ctl, &reg, 0);
@@ -827,7 +828,7 @@ int cs_dsp_coeff_write_ctrl(struct cs_dsp_coeff_ctl *ctl,
 	}
 
 	ctl->set = 1;
-	if (ctl->enabled && ctl->dsp->running)
+	if (ctl->enabled && ctl->dsp->running && !ctl->dsp->hibernating)
 		ret = cs_dsp_coeff_write_ctrl_raw(ctl, off, buf, len);
 
 	if (ret < 0)
@@ -920,12 +921,12 @@ int cs_dsp_coeff_read_ctrl(struct cs_dsp_coeff_ctl *ctl,
 		return -EINVAL;
 
 	if (ctl->flags & WMFW_CTL_FLAG_VOLATILE) {
-		if (ctl->enabled && ctl->dsp->running)
+		if (ctl->enabled && ctl->dsp->running && !ctl->dsp->hibernating)
 			return cs_dsp_coeff_read_ctrl_raw(ctl, off, buf, len);
 		else
 			return -EPERM;
 	} else {
-		if (!ctl->flags && ctl->enabled && ctl->dsp->running)
+		if (!ctl->flags && ctl->enabled && ctl->dsp->running && !ctl->dsp->hibernating)
 			ret = cs_dsp_coeff_read_ctrl_raw(ctl, 0, ctl->cache, ctl->len);
 
 		if (buf != ctl->cache)
@@ -1108,6 +1109,44 @@ static int cs_dsp_create_control(struct cs_dsp *dsp,
 	return ret;
 }
 
+
+/**
+ * cs_dsp_hibernate() - Disable or enable all controls for a DSP
+ * @dsp: pointer to DSP structure
+ * @hibernate: whether to set controls to cache only mode
+ *
+ * When @hibernate is true, the DSP is entering hibernation mode where the
+ * regmap is inaccessible, and all controls become cache only.
+ * When @hibernate is false, the DSP has exited hibernation mode. If the DSP
+ * is running, all controls are re-synced to the DSP.
+ *
+ */
+void cs_dsp_hibernate(struct cs_dsp *dsp, bool hibernate)
+{
+	mutex_lock(&dsp->pwr_lock);
+
+	if (!dsp->running) {
+		cs_dsp_dbg(dsp, "Cannot hibernate, DSP not running\n");
+		goto out;
+	}
+
+	if (dsp->hibernating == hibernate)
+		goto out;
+
+	cs_dsp_dbg(dsp, "Set hibernating to %d\n", hibernate);
+	dsp->hibernating = hibernate;
+
+	if (!dsp->hibernating && dsp->running) {
+		int ret = cs_dsp_coeff_sync_controls(dsp);
+
+		if (ret)
+			cs_dsp_err(dsp, "Error syncing controls: %d\n", ret);
+	}
+out:
+	mutex_unlock(&dsp->pwr_lock);
+}
+EXPORT_SYMBOL_NS_GPL(cs_dsp_hibernate, "FW_CS_DSP");
+
 struct cs_dsp_coeff_parsed_alg {
 	int id;
 	const u8 *name;
@@ -2498,6 +2537,7 @@ int cs_dsp_adsp1_power_up(struct cs_dsp *dsp,
 		goto err_ena;
 
 	dsp->booted = true;
+	dsp->hibernating = false;
 
 	/* Start the core running */
 	regmap_update_bits(dsp->regmap, dsp->base + ADSP1_CONTROL_30,
@@ -2776,6 +2816,7 @@ int cs_dsp_power_up(struct cs_dsp *dsp,
 		dsp->ops->disable_core(dsp);
 
 	dsp->booted = true;
+	dsp->hibernating = false;
 
 	mutex_unlock(&dsp->pwr_lock);
 
include/linux/firmware/cirrus/cs_dsp.h +3 -0
index 0ec1cdc5585d..4e3baa557068 100644
@@ -179,6 +179,7 @@ struct cs_dsp {
 
 	bool booted;
 	bool running;
+	bool hibernating;
 
 	struct list_head ctl_list;
 
@@ -354,4 +355,6 @@ int cs_dsp_chunk_write(struct cs_dsp_chunk *ch, int nbits, u32 val);
 int cs_dsp_chunk_flush(struct cs_dsp_chunk *ch);
 int cs_dsp_chunk_read(struct cs_dsp_chunk *ch, int nbits);
 
+void cs_dsp_hibernate(struct cs_dsp *dsp, bool hibernating);
+
 #endif
Signature · 1 line
2.43.0
#5 lore raw

PATCH v2 3/4 ASoC: cs35l41: Hibernate wm_adsp on runtime suspend

In reply to Stefan Binding

When the CS35L41 driver suspends, it is put into hibernation, and the regmap goes into cache_only, but the firmware is still running, and wm_adsp is not stopped. If userspace attempts to read a firmware control, it will perform a regmap_raw_read() and this will produce an error in the kernel log. To prevent spurious errors, put the DSP into hibernation which prevents access to the hardware for the ALSA controls.

---
 sound/soc/codecs/cs35l41.c | 5 +++++
 1 file changed, 5 insertions(+)
sound/soc/codecs/cs35l41.c +5 -0
index ee56dfceedeb..b2a076706c79 100644
@@ -1404,6 +1404,7 @@ static int cs35l41_runtime_suspend(struct device *dev)
 	if (!cs35l41->dsp.preloaded || !cs35l41->dsp.cs_dsp.running)
 		return 0;
 
+	wm_adsp_hibernate(&cs35l41->dsp, true);
 	cs35l41_enter_hibernate(dev, cs35l41->regmap, cs35l41->hw_cfg.bst_type);
 
 	regcache_cache_only(cs35l41->regmap, true);
@@ -1432,10 +1433,14 @@ static int cs35l41_runtime_resume(struct device *dev)
 	cs35l41_test_key_unlock(cs35l41->dev, cs35l41->regmap);
 	ret = regcache_sync(cs35l41->regmap);
 	cs35l41_test_key_lock(cs35l41->dev, cs35l41->regmap);
+
+	wm_adsp_hibernate(&cs35l41->dsp, false);
+
 	if (ret) {
 		dev_err(cs35l41->dev, "Failed to restore register cache: %d\n", ret);
 		return ret;
 	}
+
 	cs35l41_init_boost(cs35l41->dev, cs35l41->regmap, &cs35l41->hw_cfg);
 
 	return 0;
Signature · 1 line
2.43.0
#6 lore raw

PATCH v2 4/4 ASoC: cs35l45: Hibernate wm_adsp on runtime suspend

In reply to Stefan Binding

From: Ricardo Rivera-Matos <[email protected]>

When the CS35L45 driver suspends, it is put into hibernation, and the regmap goes into cache_only, but the firmware is still running, and wm_adsp is not stopped. If userspace attempts to read a firmware control, it will perform a regmap_raw_read() and this will produce an error in the kernel log. To prevent spurious errors, put the DSP into hibernation which prevents access to the hardware for the ALSA controls.

---
 sound/soc/codecs/cs35l45.c | 3 +++
 1 file changed, 3 insertions(+)
sound/soc/codecs/cs35l45.c +3 -0
index 7aa558d6362f..a032bb23b4ac 100644
@@ -984,6 +984,7 @@ static int cs35l45_runtime_suspend(struct device *dev)
 	if (!cs35l45->dsp.preloaded || !cs35l45->dsp.cs_dsp.running)
 		return 0;
 
+	wm_adsp_hibernate(&cs35l45->dsp, true);
 	cs35l45_enter_hibernate(cs35l45);
 
 	regcache_cache_only(cs35l45->regmap, true);
@@ -1014,6 +1015,8 @@ static int cs35l45_runtime_resume(struct device *dev)
 	if (ret != 0)
 		dev_warn(cs35l45->dev, "regcache_sync failed: %d\n", ret);
 
+	wm_adsp_hibernate(&cs35l45->dsp, false);
+
 	/* Clear global error status */
 	regmap_clear_bits(cs35l45->regmap, CS35L45_ERROR_RELEASE, CS35L45_GLOBAL_ERR_RLS_MASK);
 	regmap_set_bits(cs35l45->regmap, CS35L45_ERROR_RELEASE, CS35L45_GLOBAL_ERR_RLS_MASK);
Signature · 1 line
2.43.0
#3 lore raw

PATCH v2 2/4 ASoC: codecs: wm_adsp: Allow wm_adsp to hibernate without stopping DSP

In reply to Stefan Binding

Some parts do not stop the DSP core before runtime suspend, however, this leaves the DSP controls enabled and accessible whilst the part is suspended and the regmap is cache_only.

Add new APIs to allow for the DSP to hibernate when the part is suspended, and the DSP is not shut down.

---
 sound/soc/codecs/wm_adsp.c | 6 ++++++
 sound/soc/codecs/wm_adsp.h | 1 +
 2 files changed, 7 insertions(+)
sound/soc/codecs/wm_adsp.c +6 -0
index 2e23848e1dce..d95b54275b5e 100644
@@ -1100,6 +1100,12 @@ void wm_adsp_stop(struct wm_adsp *dsp)
 }
 EXPORT_SYMBOL_GPL(wm_adsp_stop);
 
+void wm_adsp_hibernate(struct wm_adsp *dsp, bool hibernate)
+{
+	cs_dsp_hibernate(&dsp->cs_dsp, hibernate);
+}
+EXPORT_SYMBOL_GPL(wm_adsp_hibernate);
+
 int wm_adsp_event(struct snd_soc_dapm_widget *w,
 		  struct snd_kcontrol *kcontrol, int event)
 {
sound/soc/codecs/wm_adsp.h +1 -0
index 8035fda71f8d..a9118be793d7 100644
@@ -103,6 +103,7 @@ irqreturn_t wm_halo_wdt_expire(int irq, void *data);
 
 int wm_adsp_run(struct wm_adsp *dsp);
 void wm_adsp_stop(struct wm_adsp *dsp);
+void wm_adsp_hibernate(struct wm_adsp *dsp, bool hibernate);
 int wm_adsp_event(struct snd_soc_dapm_widget *w,
 		  struct snd_kcontrol *kcontrol, int event);
 
Signature · 1 line
2.43.0
maintainer Reviewed-by on v2 1/4
#2 lore raw

Re: [PATCH v2 1/4] firmware: cs_dsp: Add API to hibernate the DSP

In reply to Stefan Binding
Quoted 15 lines · Stefan Binding

For some parts, the DSP is kept running when in low power mode (hibernation), leaving the firmware ALSA controls enabled, but the registers are inaccessible. Attempts to access volatile firmware controls whilst in this state would produce errors in the kernel log due to a regmap_raw_read() into DSP registers whilst the regmap is in cache_only.

To remove this error log, add a hibernating flag to indicate that the controls are inaccessible, so we no longer try to read or write to the registers whilst the regmap is in cache_only.

This would still produce an error when trying to read or write to these controls, but this would be a different error (-EPERM instead of -EBUSY), and would not produce a spurious error log in the kernel.

Quoted 8 lines

Upon wake from hibernation, the control caches are re-synced to the hardware, if the DSP is running.

---
  drivers/firmware/cirrus/cs_dsp.c       | 49 +++++++++++++++++++++++---
  include/linux/firmware/cirrus/cs_dsp.h |  3 ++
  2 files changed, 48 insertions(+), 4 deletions(-)
maintainer Reviewed-by on v2 2/4
#4 lore raw

Re: [PATCH v2 2/4] ASoC: codecs: wm_adsp: Allow wm_adsp to hibernate without stopping DSP

In reply to Stefan Binding
Quoted 4 lines · Stefan Binding

Some parts do not stop the DSP core before runtime suspend, however, this leaves the DSP controls enabled and accessible whilst the part is suspended and the regmap is cache_only.

Quoted 8 lines

Add new APIs to allow for the DSP to hibernate when the part is suspended, and the DSP is not shut down.

---
  sound/soc/codecs/wm_adsp.c | 6 ++++++
  sound/soc/codecs/wm_adsp.h | 1 +
  2 files changed, 7 insertions(+)
maintainer
#7 lore raw
In reply to Stefan Binding
Quoted 6 lines · Stefan Binding

When the CS35L41 and CS35L45 drivers suspend, they are put into hibernation, and the regmap goes into cache_only, but the firmware is still running, and wm_adsp is not stopped. If userspace attempts to read a firmware control, it will perform a regmap_raw_read() and this will produce an error in the kernel log.

Quoted 6 lines

To prevent these spurious errors, add an apis into cs_dsp and wm_adsp to allow wm_adsp to hibernate. In this hibernation mode, reads or writes to the dsp controls would be rejected with -EPERM rather than -EBUSY, and no error will be printed to the kernel log.

[...]

Applied to

https://git.kernel.org/pub/scm/linux/kernel/git/broonie/sound.git for-next

Thanks!

[1/4] firmware: cs_dsp: Add API to hibernate the DSP

      commit: 73942a6ea26bd7e02b7c260b8b7aa942397be894
[2/4] ASoC: codecs: wm_adsp: Allow wm_adsp to hibernate without stopping DSP
      commit: 6394a52c90c4abd10c8265983ec1f53207cd283e
[3/4] ASoC: cs35l41: Hibernate wm_adsp on runtime suspend
      commit: 4d80c0dbcda551b8b86ff14c6ae93026993970b2
[4/4] ASoC: cs35l45: Hibernate wm_adsp on runtime suspend
      commit: 17c6bf433742e0c1ff5ce175145877c0194e4a7a

All being well this means that it will be integrated into the linux-next tree (usually sometime in the next 24 hours) and sent to Linus during the next merge window (or sooner if it is a bug fix), however if problems are discovered then the patch may be dropped or reverted.

You may get further e-mails resulting from automated or manual testing and review of the tree, please engage with people reporting problems and send followup patches addressing any issues that are reported if needed.

If any updates are required or you are submitting further changes they should be sent as incremental updates against current git, existing patches will not be replaced.

Please add any relevant lists and maintainers to the CCs when replying to this mail.

Thanks,
Mark