All of lore.kernel.org
 help / color / mirror / Atom feed
From: Maxime Ripard <maxime@cerno.tech>
To: Liam Girdwood <lgirdwood@gmail.com>,
	Mark Brown <broonie@kernel.org>, Jaroslav Kysela <perex@perex.cz>,
	Takashi Iwai <tiwai@suse.com>,
	Lars-Peter Clausen <lars@metafoo.de>
Cc: alsa-devel@alsa-project.org, linux-kernel@vger.kernel.org,
	Dave Stevenson <dave.stevenson@raspberrypi.com>,
	Tim Gover <tim.gover@raspberrypi.com>,
	Phil Elwell <phil@raspberrypi.com>,
	Maxime Ripard <maxime@cerno.tech>
Subject: [PATCH] ASoC: core: Remove only the registered component in devm functions
Date: Tue,  7 Jul 2020 09:42:37 +0200	[thread overview]
Message-ID: <20200707074237.287171-1-maxime@cerno.tech> (raw)

The ASoC devm_ functions that register a component
(devm_snd_soc_register_component and devm_snd_dmaengine_pcm_register) will
clean their component by running snd_soc_unregister_component.

snd_soc_unregister_component will then remove all the components for the
device that was used to register the component in the first place.

However, some drivers register several components (such as a DAI and a
dmaengine PCM) on the same device, and if the dmaengine PCM is registered
first, then the DAI will be cleaned up first and
snd_dmaengine_pcm_unregister will be called next.

snd_dmaengine_pcm_unregister will then lookup the dmaengine PCM component
on the device, and if there's one unregister that component and release its
dmaengine channels. That doesn't happen in practice though since the first
call to snd_soc_unregister_component removed all the components, so we
never get the chance to release the dmaengine channels.

In order to fix this, instead of removing all the components for a given
device, we can simply remove the component that was registered in the first
place. We should have the same number of component registration than we
have components, so it should work just fine.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>

---

This was observed on a RaspberryPi that uses the vc4_hdmi driver
(drivers/gpu/drm/vc4/vc4_hdmi.c). This driver will register a dmaengine PCM
and two components. If the MIPI-DSI controller is enabled, it will create
an EPROBE_DEFER across the entire display pipeline, leading to multiple
bind/unbind cycles in the other display drivers including vc4_hdmi, leading
to multiple warnings since we request the same dmaengine channels on the
same device without ever releasing them.

It's not really clear to me when that bug was introduced exactly, since it
can only be seen on a rather unusual setup, and with all the drivers
built-in (otherwise we probably wouldn't get an EPROBE_DEFER for DSI), but
it still looks like something that should probably go to stable?

Changes from v1:
  - Rebased on top of next-20200706
---
 include/sound/soc.h                   |  2 ++
 sound/soc/soc-core.c                  | 27 +++++++++++++++++++++++++++
 sound/soc/soc-devres.c                |  8 +++++---
 sound/soc/soc-generic-dmaengine-pcm.c |  2 +-
 4 files changed, 35 insertions(+), 4 deletions(-)

diff --git a/include/sound/soc.h b/include/sound/soc.h
index 6791b7570a67..59235e553630 100644
--- a/include/sound/soc.h
+++ b/include/sound/soc.h
@@ -426,6 +426,8 @@ int devm_snd_soc_register_component(struct device *dev,
 			 const struct snd_soc_component_driver *component_driver,
 			 struct snd_soc_dai_driver *dai_drv, int num_dai);
 void snd_soc_unregister_component(struct device *dev);
+void snd_soc_unregister_component_by_driver(struct device *dev,
+			 const struct snd_soc_component_driver *component_driver);
 struct snd_soc_component *snd_soc_lookup_component_nolocked(struct device *dev,
 							    const char *driver_name);
 struct snd_soc_component *snd_soc_lookup_component(struct device *dev,
diff --git a/sound/soc/soc-core.c b/sound/soc/soc-core.c
index adedadcb0efb..7c58e45c1c3f 100644
--- a/sound/soc/soc-core.c
+++ b/sound/soc/soc-core.c
@@ -2513,6 +2513,33 @@ int snd_soc_register_component(struct device *dev,
 }
 EXPORT_SYMBOL_GPL(snd_soc_register_component);
 
+/**
+ * snd_soc_unregister_component_by_driver - Unregister component using a given driver
+ * from the ASoC core
+ *
+ * @dev: The device to unregister
+ * @component_driver: The component driver to unregister
+ */
+void snd_soc_unregister_component_by_driver(struct device *dev,
+					    const struct snd_soc_component_driver *component_driver)
+{
+	struct snd_soc_component *component;
+
+	if (!component_driver)
+		return;
+
+	mutex_lock(&client_mutex);
+	component = snd_soc_lookup_component_nolocked(dev, component_driver->name);
+	if (!component)
+		goto out;
+
+	snd_soc_del_component_unlocked(component);
+
+out:
+	mutex_unlock(&client_mutex);
+}
+EXPORT_SYMBOL_GPL(snd_soc_unregister_component_by_driver);
+
 /**
  * snd_soc_unregister_component - Unregister all related component
  * from the ASoC core
diff --git a/sound/soc/soc-devres.c b/sound/soc/soc-devres.c
index 11e5d7962370..4534a1c03e8e 100644
--- a/sound/soc/soc-devres.c
+++ b/sound/soc/soc-devres.c
@@ -48,7 +48,9 @@ EXPORT_SYMBOL_GPL(devm_snd_soc_register_dai);
 
 static void devm_component_release(struct device *dev, void *res)
 {
-	snd_soc_unregister_component(*(struct device **)res);
+	const struct snd_soc_component_driver **cmpnt_drv = res;
+
+	snd_soc_unregister_component_by_driver(dev, *cmpnt_drv);
 }
 
 /**
@@ -65,7 +67,7 @@ int devm_snd_soc_register_component(struct device *dev,
 			 const struct snd_soc_component_driver *cmpnt_drv,
 			 struct snd_soc_dai_driver *dai_drv, int num_dai)
 {
-	struct device **ptr;
+	const struct snd_soc_component_driver **ptr;
 	int ret;
 
 	ptr = devres_alloc(devm_component_release, sizeof(*ptr), GFP_KERNEL);
@@ -74,7 +76,7 @@ int devm_snd_soc_register_component(struct device *dev,
 
 	ret = snd_soc_register_component(dev, cmpnt_drv, dai_drv, num_dai);
 	if (ret == 0) {
-		*ptr = dev;
+		*ptr = cmpnt_drv;
 		devres_add(dev, ptr);
 	} else {
 		devres_free(ptr);
diff --git a/sound/soc/soc-generic-dmaengine-pcm.c b/sound/soc/soc-generic-dmaengine-pcm.c
index 80a4e71f2d95..61844403f181 100644
--- a/sound/soc/soc-generic-dmaengine-pcm.c
+++ b/sound/soc/soc-generic-dmaengine-pcm.c
@@ -478,7 +478,7 @@ void snd_dmaengine_pcm_unregister(struct device *dev)
 
 	pcm = soc_component_to_pcm(component);
 
-	snd_soc_unregister_component(dev);
+	snd_soc_unregister_component_by_driver(dev, component->driver);
 	dmaengine_pcm_release_chan(pcm);
 	kfree(pcm);
 }
-- 
2.26.2


WARNING: multiple messages have this Message-ID (diff)
From: Maxime Ripard <maxime@cerno.tech>
To: Liam Girdwood <lgirdwood@gmail.com>,
	Mark Brown <broonie@kernel.org>, Jaroslav Kysela <perex@perex.cz>,
	Takashi Iwai <tiwai@suse.com>,
	Lars-Peter Clausen <lars@metafoo.de>
Cc: alsa-devel@alsa-project.org,
	Tim Gover <tim.gover@raspberrypi.com>,
	Dave Stevenson <dave.stevenson@raspberrypi.com>,
	linux-kernel@vger.kernel.org, Maxime Ripard <maxime@cerno.tech>,
	Phil Elwell <phil@raspberrypi.com>
Subject: [PATCH] ASoC: core: Remove only the registered component in devm functions
Date: Tue,  7 Jul 2020 09:42:37 +0200	[thread overview]
Message-ID: <20200707074237.287171-1-maxime@cerno.tech> (raw)

The ASoC devm_ functions that register a component
(devm_snd_soc_register_component and devm_snd_dmaengine_pcm_register) will
clean their component by running snd_soc_unregister_component.

snd_soc_unregister_component will then remove all the components for the
device that was used to register the component in the first place.

However, some drivers register several components (such as a DAI and a
dmaengine PCM) on the same device, and if the dmaengine PCM is registered
first, then the DAI will be cleaned up first and
snd_dmaengine_pcm_unregister will be called next.

snd_dmaengine_pcm_unregister will then lookup the dmaengine PCM component
on the device, and if there's one unregister that component and release its
dmaengine channels. That doesn't happen in practice though since the first
call to snd_soc_unregister_component removed all the components, so we
never get the chance to release the dmaengine channels.

In order to fix this, instead of removing all the components for a given
device, we can simply remove the component that was registered in the first
place. We should have the same number of component registration than we
have components, so it should work just fine.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>

---

This was observed on a RaspberryPi that uses the vc4_hdmi driver
(drivers/gpu/drm/vc4/vc4_hdmi.c). This driver will register a dmaengine PCM
and two components. If the MIPI-DSI controller is enabled, it will create
an EPROBE_DEFER across the entire display pipeline, leading to multiple
bind/unbind cycles in the other display drivers including vc4_hdmi, leading
to multiple warnings since we request the same dmaengine channels on the
same device without ever releasing them.

It's not really clear to me when that bug was introduced exactly, since it
can only be seen on a rather unusual setup, and with all the drivers
built-in (otherwise we probably wouldn't get an EPROBE_DEFER for DSI), but
it still looks like something that should probably go to stable?

Changes from v1:
  - Rebased on top of next-20200706
---
 include/sound/soc.h                   |  2 ++
 sound/soc/soc-core.c                  | 27 +++++++++++++++++++++++++++
 sound/soc/soc-devres.c                |  8 +++++---
 sound/soc/soc-generic-dmaengine-pcm.c |  2 +-
 4 files changed, 35 insertions(+), 4 deletions(-)

diff --git a/include/sound/soc.h b/include/sound/soc.h
index 6791b7570a67..59235e553630 100644
--- a/include/sound/soc.h
+++ b/include/sound/soc.h
@@ -426,6 +426,8 @@ int devm_snd_soc_register_component(struct device *dev,
 			 const struct snd_soc_component_driver *component_driver,
 			 struct snd_soc_dai_driver *dai_drv, int num_dai);
 void snd_soc_unregister_component(struct device *dev);
+void snd_soc_unregister_component_by_driver(struct device *dev,
+			 const struct snd_soc_component_driver *component_driver);
 struct snd_soc_component *snd_soc_lookup_component_nolocked(struct device *dev,
 							    const char *driver_name);
 struct snd_soc_component *snd_soc_lookup_component(struct device *dev,
diff --git a/sound/soc/soc-core.c b/sound/soc/soc-core.c
index adedadcb0efb..7c58e45c1c3f 100644
--- a/sound/soc/soc-core.c
+++ b/sound/soc/soc-core.c
@@ -2513,6 +2513,33 @@ int snd_soc_register_component(struct device *dev,
 }
 EXPORT_SYMBOL_GPL(snd_soc_register_component);
 
+/**
+ * snd_soc_unregister_component_by_driver - Unregister component using a given driver
+ * from the ASoC core
+ *
+ * @dev: The device to unregister
+ * @component_driver: The component driver to unregister
+ */
+void snd_soc_unregister_component_by_driver(struct device *dev,
+					    const struct snd_soc_component_driver *component_driver)
+{
+	struct snd_soc_component *component;
+
+	if (!component_driver)
+		return;
+
+	mutex_lock(&client_mutex);
+	component = snd_soc_lookup_component_nolocked(dev, component_driver->name);
+	if (!component)
+		goto out;
+
+	snd_soc_del_component_unlocked(component);
+
+out:
+	mutex_unlock(&client_mutex);
+}
+EXPORT_SYMBOL_GPL(snd_soc_unregister_component_by_driver);
+
 /**
  * snd_soc_unregister_component - Unregister all related component
  * from the ASoC core
diff --git a/sound/soc/soc-devres.c b/sound/soc/soc-devres.c
index 11e5d7962370..4534a1c03e8e 100644
--- a/sound/soc/soc-devres.c
+++ b/sound/soc/soc-devres.c
@@ -48,7 +48,9 @@ EXPORT_SYMBOL_GPL(devm_snd_soc_register_dai);
 
 static void devm_component_release(struct device *dev, void *res)
 {
-	snd_soc_unregister_component(*(struct device **)res);
+	const struct snd_soc_component_driver **cmpnt_drv = res;
+
+	snd_soc_unregister_component_by_driver(dev, *cmpnt_drv);
 }
 
 /**
@@ -65,7 +67,7 @@ int devm_snd_soc_register_component(struct device *dev,
 			 const struct snd_soc_component_driver *cmpnt_drv,
 			 struct snd_soc_dai_driver *dai_drv, int num_dai)
 {
-	struct device **ptr;
+	const struct snd_soc_component_driver **ptr;
 	int ret;
 
 	ptr = devres_alloc(devm_component_release, sizeof(*ptr), GFP_KERNEL);
@@ -74,7 +76,7 @@ int devm_snd_soc_register_component(struct device *dev,
 
 	ret = snd_soc_register_component(dev, cmpnt_drv, dai_drv, num_dai);
 	if (ret == 0) {
-		*ptr = dev;
+		*ptr = cmpnt_drv;
 		devres_add(dev, ptr);
 	} else {
 		devres_free(ptr);
diff --git a/sound/soc/soc-generic-dmaengine-pcm.c b/sound/soc/soc-generic-dmaengine-pcm.c
index 80a4e71f2d95..61844403f181 100644
--- a/sound/soc/soc-generic-dmaengine-pcm.c
+++ b/sound/soc/soc-generic-dmaengine-pcm.c
@@ -478,7 +478,7 @@ void snd_dmaengine_pcm_unregister(struct device *dev)
 
 	pcm = soc_component_to_pcm(component);
 
-	snd_soc_unregister_component(dev);
+	snd_soc_unregister_component_by_driver(dev, component->driver);
 	dmaengine_pcm_release_chan(pcm);
 	kfree(pcm);
 }
-- 
2.26.2


             reply	other threads:[~2020-07-07  7:42 UTC|newest]

Thread overview: 8+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-07-07  7:42 Maxime Ripard [this message]
2020-07-07  7:42 ` [PATCH] ASoC: core: Remove only the registered component in devm functions Maxime Ripard
2020-07-07 14:17 ` Mark Brown
2020-07-07 14:17   ` Mark Brown
  -- strict thread matches above, loose matches on Subject: below --
2020-07-03  7:49 Maxime Ripard
2020-07-03  7:49 ` Maxime Ripard
2020-07-03 15:32 ` Mark Brown
2020-07-03 15:32   ` Mark Brown

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20200707074237.287171-1-maxime@cerno.tech \
    --to=maxime@cerno.tech \
    --cc=alsa-devel@alsa-project.org \
    --cc=broonie@kernel.org \
    --cc=dave.stevenson@raspberrypi.com \
    --cc=lars@metafoo.de \
    --cc=lgirdwood@gmail.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=perex@perex.cz \
    --cc=phil@raspberrypi.com \
    --cc=tim.gover@raspberrypi.com \
    --cc=tiwai@suse.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.