[alsa-devel] [PATCH 0/4 v3] ASoC: rsnd: tidyup .remove method
Hi Mark
These are v3 of unbind/bind issue fixup patches.
v2 patch-set was
Kuninori Morimoto (6): 1) ASoC: rsnd: set device data before snd_soc_register_platform/component 2) ASoC: soc-core: indicate unregister debug message once 3) ASoC: soc-core: add snd_soc_remove_card() 4) ASoC: soc-core: deactivate pins in snd_soc_instantiate_card() 5) ASoC: soc-core: call soc_cleanup_card_debugfs() from snd_soc_unregister_card() 6) ASoC: soc-core: call snd_soc_remove_card() when component del
Now, 1) / 2) are applied already. So, v3 patches is same as 3) - 6)
From: Kuninori Morimoto kuninori.morimoto.gx@renesas.com
Added snd_soc_remove_card() is termination method of snd_soc_instantiate_card()
Signed-off-by: Kuninori Morimoto kuninori.morimoto.gx@renesas.com --- sound/soc/soc-core.c | 12 +++++++++--- 1 file changed, 9 insertions(+), 3 deletions(-)
diff --git a/sound/soc/soc-core.c b/sound/soc/soc-core.c index 4bfabc7..4837b5d 100644 --- a/sound/soc/soc-core.c +++ b/sound/soc/soc-core.c @@ -1725,6 +1725,14 @@ static int soc_cleanup_card_resources(struct snd_soc_card *card)
}
+static void snd_soc_remove_card(struct snd_soc_card *card) +{ + card->instantiated = false; + snd_soc_dapm_shutdown(card); + soc_cleanup_card_resources(card); +} + + /* removes a socdev */ static int soc_remove(struct platform_device *pdev) { @@ -2393,9 +2401,7 @@ EXPORT_SYMBOL_GPL(snd_soc_register_card); int snd_soc_unregister_card(struct snd_soc_card *card) { if (card->instantiated) { - card->instantiated = false; - snd_soc_dapm_shutdown(card); - soc_cleanup_card_resources(card); + snd_soc_remove_card(card); dev_dbg(card->dev, "ASoC: Unregistered card '%s'\n", card->name); }
From: Kuninori Morimoto kuninori.morimoto.gx@renesas.com
snd_soc_instantiate_card() is the final method of snd_soc_register_card(). Deactivate pins to sleep state is related to snd_soc_instantiate_card(), not snd_soc_register_card()
Signed-off-by: Kuninori Morimoto kuninori.morimoto.gx@renesas.com --- sound/soc/soc-core.c | 33 +++++++++++++++++---------------- 1 file changed, 17 insertions(+), 16 deletions(-)
diff --git a/sound/soc/soc-core.c b/sound/soc/soc-core.c index 4837b5d..794d8ec 100644 --- a/sound/soc/soc-core.c +++ b/sound/soc/soc-core.c @@ -1652,6 +1652,23 @@ static int snd_soc_instantiate_card(struct snd_soc_card *card) snd_soc_dapm_sync(&card->dapm); mutex_unlock(&card->mutex);
+ /* deactivate pins to sleep state */ + for (i = 0; i < card->num_rtd; i++) { + struct snd_soc_pcm_runtime *rtd = &card->rtd[i]; + struct snd_soc_dai *cpu_dai = rtd->cpu_dai; + int j; + + for (j = 0; j < rtd->num_codecs; j++) { + struct snd_soc_dai *codec_dai = rtd->codec_dais[j]; + + if (!codec_dai->active) + pinctrl_pm_select_sleep_state(codec_dai->dev); + } + + if (!cpu_dai->active) + pinctrl_pm_select_sleep_state(cpu_dai->dev); + } + return 0;
probe_aux_dev_err: @@ -2372,22 +2389,6 @@ int snd_soc_register_card(struct snd_soc_card *card) if (ret != 0) soc_cleanup_card_debugfs(card);
- /* deactivate pins to sleep state */ - for (i = 0; i < card->num_rtd; i++) { - struct snd_soc_pcm_runtime *rtd = &card->rtd[i]; - struct snd_soc_dai *cpu_dai = rtd->cpu_dai; - int j; - - for (j = 0; j < rtd->num_codecs; j++) { - struct snd_soc_dai *codec_dai = rtd->codec_dais[j]; - if (!codec_dai->active) - pinctrl_pm_select_sleep_state(codec_dai->dev); - } - - if (!cpu_dai->active) - pinctrl_pm_select_sleep_state(cpu_dai->dev); - } - return ret; } EXPORT_SYMBOL_GPL(snd_soc_register_card);
From: Kuninori Morimoto kuninori.morimoto.gx@renesas.com
soc_cleanup_card_debugfs() is called from soc_cleanup_card_resources(), but, card debugfs is called from snd_soc_register_card(). cleanup function should be called paired function.
Signed-off-by: Kuninori Morimoto kuninori.morimoto.gx@renesas.com --- sound/soc/soc-core.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/sound/soc/soc-core.c b/sound/soc/soc-core.c index 794d8ec..b7ab676 100644 --- a/sound/soc/soc-core.c +++ b/sound/soc/soc-core.c @@ -1729,8 +1729,6 @@ static int soc_cleanup_card_resources(struct snd_soc_card *card) /* remove and free each DAI */ soc_remove_dai_links(card);
- soc_cleanup_card_debugfs(card); - /* remove the card */ if (card->remove) card->remove(card); @@ -2403,6 +2401,8 @@ int snd_soc_unregister_card(struct snd_soc_card *card) { if (card->instantiated) { snd_soc_remove_card(card); + soc_cleanup_card_debugfs(card); + dev_dbg(card->dev, "ASoC: Unregistered card '%s'\n", card->name); }
From: Kuninori Morimoto kuninori.morimoto.gx@renesas.com
ASoC devices are organized as CPU-CARD-CODEC. Then, CPU/CODEC are based on component structure. Now, each CARD device knows connected component. But CARD doesn't notice if connected component was removed when user called rmmod or unbind in current implementation. Thus, CARD which lost some components still exist in system. And then, ALSA sound card will have some problem if user used this CARD in such timing. This patch temporarily removes CARD from system if connected component was removed, and re-add it if some component was added.
Reported-by: Nguyen Viet Dung nv-dung@jinso.co.jp Reported-by: Bui Duc Phuc bd-phuc@jinso.co.jp Reported-by: Cao Minh Hiep cm-hiep@jinso.co.jp Signed-off-by: Kuninori Morimoto kuninori.morimoto.gx@renesas.com --- include/sound/soc.h | 1 + sound/soc/soc-core.c | 26 ++++++++++++++++++++++++++ 2 files changed, 27 insertions(+)
diff --git a/include/sound/soc.h b/include/sound/soc.h index b4fca9a..a90eff4 100644 --- a/include/sound/soc.h +++ b/include/sound/soc.h @@ -1083,6 +1083,7 @@ struct snd_soc_card { struct list_head paths; struct list_head dapm_list; struct list_head dapm_dirty; + struct list_head unbinded_list;
/* Generic DAPM context for the card */ struct snd_soc_dapm_context dapm; diff --git a/sound/soc/soc-core.c b/sound/soc/soc-core.c index b7ab676..f8d5498 100644 --- a/sound/soc/soc-core.c +++ b/sound/soc/soc-core.c @@ -56,6 +56,7 @@ static DEFINE_MUTEX(client_mutex); static LIST_HEAD(platform_list); static LIST_HEAD(codec_list); static LIST_HEAD(component_list); +static LIST_HEAD(unbinded_card_list);
/* * This is a timeout to do a DAPM powerdown after a stream is closed(). @@ -2406,6 +2407,10 @@ int snd_soc_unregister_card(struct snd_soc_card *card) dev_dbg(card->dev, "ASoC: Unregistered card '%s'\n", card->name); }
+ mutex_lock(&client_mutex); + list_del(&card->unbinded_list); + mutex_unlock(&client_mutex); + return 0; } EXPORT_SYMBOL_GPL(snd_soc_unregister_card); @@ -2669,6 +2674,9 @@ EXPORT_SYMBOL_GPL(snd_soc_component_exit_regmap);
static void snd_soc_component_add_unlocked(struct snd_soc_component *component) { + struct snd_soc_card *card, *_card; + int ret; + if (!component->write && !component->read) { if (!component->regmap) component->regmap = dev_get_regmap(component->dev, NULL); @@ -2677,6 +2685,16 @@ static void snd_soc_component_add_unlocked(struct snd_soc_component *component) }
list_add(&component->list, &component_list); + + /* re-add temporarily removed card if exist */ + list_for_each_entry_safe(card, _card, &unbinded_card_list, + unbinded_list) { + ret = snd_soc_instantiate_card(card); + if (ret < 0) + continue; + + list_del(&card->unbinded_list); + } }
static void snd_soc_component_add(struct snd_soc_component *component) @@ -2694,7 +2712,15 @@ static void snd_soc_component_cleanup(struct snd_soc_component *component)
static void snd_soc_component_del_unlocked(struct snd_soc_component *component) { + struct snd_soc_card *card = component->card; + list_del(&component->list); + + /* card is removed temporarily */ + if (card->instantiated) { + list_add(&card->unbinded_list, &unbinded_card_list); + snd_soc_remove_card(card); + } }
static void snd_soc_component_del(struct snd_soc_component *component)
At Mon, 9 Feb 2015 08:52:49 +0000, Kuninori Morimoto wrote:
From: Kuninori Morimoto kuninori.morimoto.gx@renesas.com
ASoC devices are organized as CPU-CARD-CODEC. Then, CPU/CODEC are based on component structure. Now, each CARD device knows connected component. But CARD doesn't notice if connected component was removed when user called rmmod or unbind in current implementation. Thus, CARD which lost some components still exist in system. And then, ALSA sound card will have some problem if user used this CARD in such timing. This patch temporarily removes CARD from system if connected component was removed, and re-add it if some component was added.
Reported-by: Nguyen Viet Dung nv-dung@jinso.co.jp Reported-by: Bui Duc Phuc bd-phuc@jinso.co.jp Reported-by: Cao Minh Hiep cm-hiep@jinso.co.jp Signed-off-by: Kuninori Morimoto kuninori.morimoto.gx@renesas.com
include/sound/soc.h | 1 + sound/soc/soc-core.c | 26 ++++++++++++++++++++++++++ 2 files changed, 27 insertions(+)
diff --git a/include/sound/soc.h b/include/sound/soc.h index b4fca9a..a90eff4 100644 --- a/include/sound/soc.h +++ b/include/sound/soc.h @@ -1083,6 +1083,7 @@ struct snd_soc_card { struct list_head paths; struct list_head dapm_list; struct list_head dapm_dirty;
struct list_head unbinded_list;
/* Generic DAPM context for the card */ struct snd_soc_dapm_context dapm;
diff --git a/sound/soc/soc-core.c b/sound/soc/soc-core.c index b7ab676..f8d5498 100644 --- a/sound/soc/soc-core.c +++ b/sound/soc/soc-core.c @@ -56,6 +56,7 @@ static DEFINE_MUTEX(client_mutex); static LIST_HEAD(platform_list); static LIST_HEAD(codec_list); static LIST_HEAD(component_list); +static LIST_HEAD(unbinded_card_list);
/*
- This is a timeout to do a DAPM powerdown after a stream is closed().
@@ -2406,6 +2407,10 @@ int snd_soc_unregister_card(struct snd_soc_card *card) dev_dbg(card->dev, "ASoC: Unregistered card '%s'\n", card->name); }
- mutex_lock(&client_mutex);
- list_del(&card->unbinded_list);
- mutex_unlock(&client_mutex);
- return 0;
} EXPORT_SYMBOL_GPL(snd_soc_unregister_card); @@ -2669,6 +2674,9 @@ EXPORT_SYMBOL_GPL(snd_soc_component_exit_regmap);
static void snd_soc_component_add_unlocked(struct snd_soc_component *component) {
- struct snd_soc_card *card, *_card;
- int ret;
- if (!component->write && !component->read) { if (!component->regmap) component->regmap = dev_get_regmap(component->dev, NULL);
@@ -2677,6 +2685,16 @@ static void snd_soc_component_add_unlocked(struct snd_soc_component *component) }
list_add(&component->list, &component_list);
- /* re-add temporarily removed card if exist */
- list_for_each_entry_safe(card, _card, &unbinded_card_list,
unbinded_list) {ret = snd_soc_instantiate_card(card);if (ret < 0)continue;list_del(&card->unbinded_list);- }
[Cc'ed Lars-Peter]
This would instantiate a card even if it's irrelevant with the given component? If so, it looks fragile, and possibly racy, when there are multiple cards.
thanks,
Takashi
On 02/09/2015 11:48 AM, Takashi Iwai wrote:
At Mon, 9 Feb 2015 08:52:49 +0000, Kuninori Morimoto wrote:
From: Kuninori Morimoto kuninori.morimoto.gx@renesas.com
ASoC devices are organized as CPU-CARD-CODEC. Then, CPU/CODEC are based on component structure. Now, each CARD device knows connected component. But CARD doesn't notice if connected component was removed when user called rmmod or unbind in current implementation. Thus, CARD which lost some components still exist in system. And then, ALSA sound card will have some problem if user used this CARD in such timing. This patch temporarily removes CARD from system if connected component was removed, and re-add it if some component was added.
Reported-by: Nguyen Viet Dung nv-dung@jinso.co.jp Reported-by: Bui Duc Phuc bd-phuc@jinso.co.jp Reported-by: Cao Minh Hiep cm-hiep@jinso.co.jp Signed-off-by: Kuninori Morimoto kuninori.morimoto.gx@renesas.com
include/sound/soc.h | 1 + sound/soc/soc-core.c | 26 ++++++++++++++++++++++++++ 2 files changed, 27 insertions(+)
diff --git a/include/sound/soc.h b/include/sound/soc.h index b4fca9a..a90eff4 100644 --- a/include/sound/soc.h +++ b/include/sound/soc.h @@ -1083,6 +1083,7 @@ struct snd_soc_card { struct list_head paths; struct list_head dapm_list; struct list_head dapm_dirty;
struct list_head unbinded_list;
/* Generic DAPM context for the card */ struct snd_soc_dapm_context dapm;
diff --git a/sound/soc/soc-core.c b/sound/soc/soc-core.c index b7ab676..f8d5498 100644 --- a/sound/soc/soc-core.c +++ b/sound/soc/soc-core.c @@ -56,6 +56,7 @@ static DEFINE_MUTEX(client_mutex); static LIST_HEAD(platform_list); static LIST_HEAD(codec_list); static LIST_HEAD(component_list); +static LIST_HEAD(unbinded_card_list);
/*
- This is a timeout to do a DAPM powerdown after a stream is closed().
@@ -2406,6 +2407,10 @@ int snd_soc_unregister_card(struct snd_soc_card *card) dev_dbg(card->dev, "ASoC: Unregistered card '%s'\n", card->name); }
- mutex_lock(&client_mutex);
- list_del(&card->unbinded_list);
- mutex_unlock(&client_mutex);
- return 0; } EXPORT_SYMBOL_GPL(snd_soc_unregister_card);
@@ -2669,6 +2674,9 @@ EXPORT_SYMBOL_GPL(snd_soc_component_exit_regmap);
static void snd_soc_component_add_unlocked(struct snd_soc_component *component) {
- struct snd_soc_card *card, *_card;
- int ret;
- if (!component->write && !component->read) { if (!component->regmap) component->regmap = dev_get_regmap(component->dev, NULL);
@@ -2677,6 +2685,16 @@ static void snd_soc_component_add_unlocked(struct snd_soc_component *component) }
list_add(&component->list, &component_list);
- /* re-add temporarily removed card if exist */
- list_for_each_entry_safe(card, _card, &unbinded_card_list,
unbinded_list) {ret = snd_soc_instantiate_card(card);if (ret < 0)continue;list_del(&card->unbinded_list);- }
[Cc'ed Lars-Peter]
This would instantiate a card even if it's irrelevant with the given component? If so, it looks fragile, and possibly racy, when there are multiple cards.
That shouldn't be a problem. snd_soc_instantiate_card() does the proper locking and will return an error if not all components are ready yet.
I think the main issue with this path is that snd_soc_unregister_card() will crash if the card is not on the unbinded_card_list, which in most cases it won't be.
The other issue is that it introduces subtly issues with the suspend and resume order. Suspend and resume are called in the order in which the probe functions of devices are called (and succeed). By returning -EPROBE_DEFER in the card driver we make sure that the card's probe function is always called after the probe functions of all the components of the card have run. This again causes the card's suspend function to be called before any suspend function of any of it's components. Now with this patch it is possible again for a component's probe function to be called after the card's probe function which changes the suspend and resume order and might break things.
This patch essentially is a partial revert of commit b19e6e7b763 ("ASoC: core: Use driver core probe deferral") where the card list was replaced with the -EPROBE_DEFER mechanism.
So while this patch fixes the nasty crash it introduces some other subtle issue. Maybe we should also add a big WARN() when a component of a card is removed while still in use until the other issues are also fixed.
Ideally the correct fix would somehow make sure that the card itself is unbound and put back onto the probe_defer list.
- Lars
At Mon, 09 Feb 2015 12:26:02 +0100, Lars-Peter Clausen wrote:
On 02/09/2015 11:48 AM, Takashi Iwai wrote:
At Mon, 9 Feb 2015 08:52:49 +0000, Kuninori Morimoto wrote:
From: Kuninori Morimoto kuninori.morimoto.gx@renesas.com
ASoC devices are organized as CPU-CARD-CODEC. Then, CPU/CODEC are based on component structure. Now, each CARD device knows connected component. But CARD doesn't notice if connected component was removed when user called rmmod or unbind in current implementation. Thus, CARD which lost some components still exist in system. And then, ALSA sound card will have some problem if user used this CARD in such timing. This patch temporarily removes CARD from system if connected component was removed, and re-add it if some component was added.
Reported-by: Nguyen Viet Dung nv-dung@jinso.co.jp Reported-by: Bui Duc Phuc bd-phuc@jinso.co.jp Reported-by: Cao Minh Hiep cm-hiep@jinso.co.jp Signed-off-by: Kuninori Morimoto kuninori.morimoto.gx@renesas.com
include/sound/soc.h | 1 + sound/soc/soc-core.c | 26 ++++++++++++++++++++++++++ 2 files changed, 27 insertions(+)
diff --git a/include/sound/soc.h b/include/sound/soc.h index b4fca9a..a90eff4 100644 --- a/include/sound/soc.h +++ b/include/sound/soc.h @@ -1083,6 +1083,7 @@ struct snd_soc_card { struct list_head paths; struct list_head dapm_list; struct list_head dapm_dirty;
struct list_head unbinded_list;
/* Generic DAPM context for the card */ struct snd_soc_dapm_context dapm;
diff --git a/sound/soc/soc-core.c b/sound/soc/soc-core.c index b7ab676..f8d5498 100644 --- a/sound/soc/soc-core.c +++ b/sound/soc/soc-core.c @@ -56,6 +56,7 @@ static DEFINE_MUTEX(client_mutex); static LIST_HEAD(platform_list); static LIST_HEAD(codec_list); static LIST_HEAD(component_list); +static LIST_HEAD(unbinded_card_list);
/*
- This is a timeout to do a DAPM powerdown after a stream is closed().
@@ -2406,6 +2407,10 @@ int snd_soc_unregister_card(struct snd_soc_card *card) dev_dbg(card->dev, "ASoC: Unregistered card '%s'\n", card->name); }
- mutex_lock(&client_mutex);
- list_del(&card->unbinded_list);
- mutex_unlock(&client_mutex);
- return 0; } EXPORT_SYMBOL_GPL(snd_soc_unregister_card);
@@ -2669,6 +2674,9 @@ EXPORT_SYMBOL_GPL(snd_soc_component_exit_regmap);
static void snd_soc_component_add_unlocked(struct snd_soc_component *component) {
- struct snd_soc_card *card, *_card;
- int ret;
- if (!component->write && !component->read) { if (!component->regmap) component->regmap = dev_get_regmap(component->dev, NULL);
@@ -2677,6 +2685,16 @@ static void snd_soc_component_add_unlocked(struct snd_soc_component *component) }
list_add(&component->list, &component_list);
- /* re-add temporarily removed card if exist */
- list_for_each_entry_safe(card, _card, &unbinded_card_list,
unbinded_list) {ret = snd_soc_instantiate_card(card);if (ret < 0)continue;list_del(&card->unbinded_list);- }
[Cc'ed Lars-Peter]
This would instantiate a card even if it's irrelevant with the given component? If so, it looks fragile, and possibly racy, when there are multiple cards.
That shouldn't be a problem. snd_soc_instantiate_card() does the proper locking and will return an error if not all components are ready yet.
But there is no check of card->instantiated there, so the whole path can be called again even if the card was already instantiated?
Takashi
On 02/09/2015 02:07 PM, Takashi Iwai wrote:
At Mon, 09 Feb 2015 12:26:02 +0100, Lars-Peter Clausen wrote:
On 02/09/2015 11:48 AM, Takashi Iwai wrote:
At Mon, 9 Feb 2015 08:52:49 +0000, Kuninori Morimoto wrote:
From: Kuninori Morimoto kuninori.morimoto.gx@renesas.com
ASoC devices are organized as CPU-CARD-CODEC. Then, CPU/CODEC are based on component structure. Now, each CARD device knows connected component. But CARD doesn't notice if connected component was removed when user called rmmod or unbind in current implementation. Thus, CARD which lost some components still exist in system. And then, ALSA sound card will have some problem if user used this CARD in such timing. This patch temporarily removes CARD from system if connected component was removed, and re-add it if some component was added.
Reported-by: Nguyen Viet Dung nv-dung@jinso.co.jp Reported-by: Bui Duc Phuc bd-phuc@jinso.co.jp Reported-by: Cao Minh Hiep cm-hiep@jinso.co.jp Signed-off-by: Kuninori Morimoto kuninori.morimoto.gx@renesas.com
include/sound/soc.h | 1 + sound/soc/soc-core.c | 26 ++++++++++++++++++++++++++ 2 files changed, 27 insertions(+)
diff --git a/include/sound/soc.h b/include/sound/soc.h index b4fca9a..a90eff4 100644 --- a/include/sound/soc.h +++ b/include/sound/soc.h @@ -1083,6 +1083,7 @@ struct snd_soc_card { struct list_head paths; struct list_head dapm_list; struct list_head dapm_dirty;
struct list_head unbinded_list;
/* Generic DAPM context for the card */ struct snd_soc_dapm_context dapm;
diff --git a/sound/soc/soc-core.c b/sound/soc/soc-core.c index b7ab676..f8d5498 100644 --- a/sound/soc/soc-core.c +++ b/sound/soc/soc-core.c @@ -56,6 +56,7 @@ static DEFINE_MUTEX(client_mutex); static LIST_HEAD(platform_list); static LIST_HEAD(codec_list); static LIST_HEAD(component_list); +static LIST_HEAD(unbinded_card_list);
/* * This is a timeout to do a DAPM powerdown after a stream is closed(). @@ -2406,6 +2407,10 @@ int snd_soc_unregister_card(struct snd_soc_card *card) dev_dbg(card->dev, "ASoC: Unregistered card '%s'\n", card->name); }
- mutex_lock(&client_mutex);
- list_del(&card->unbinded_list);
- mutex_unlock(&client_mutex);
- return 0; } EXPORT_SYMBOL_GPL(snd_soc_unregister_card);
@@ -2669,6 +2674,9 @@ EXPORT_SYMBOL_GPL(snd_soc_component_exit_regmap);
static void snd_soc_component_add_unlocked(struct snd_soc_component *component) {
- struct snd_soc_card *card, *_card;
- int ret;
- if (!component->write && !component->read) { if (!component->regmap) component->regmap = dev_get_regmap(component->dev, NULL);
@@ -2677,6 +2685,16 @@ static void snd_soc_component_add_unlocked(struct snd_soc_component *component) }
list_add(&component->list, &component_list);
- /* re-add temporarily removed card if exist */
- list_for_each_entry_safe(card, _card, &unbinded_card_list,
unbinded_list) {ret = snd_soc_instantiate_card(card);if (ret < 0)continue;list_del(&card->unbinded_list);- }
[Cc'ed Lars-Peter]
This would instantiate a card even if it's irrelevant with the given component? If so, it looks fragile, and possibly racy, when there are multiple cards.
That shouldn't be a problem. snd_soc_instantiate_card() does the proper locking and will return an error if not all components are ready yet.
But there is no check of card->instantiated there, so the whole path can be called again even if the card was already instantiated?
The card won't be on the list of unbound cards if it is not instantiated. And this whole section is protected by the client_mutex.
- Lars
At Mon, 09 Feb 2015 14:09:50 +0100, Lars-Peter Clausen wrote:
On 02/09/2015 02:07 PM, Takashi Iwai wrote:
At Mon, 09 Feb 2015 12:26:02 +0100, Lars-Peter Clausen wrote:
On 02/09/2015 11:48 AM, Takashi Iwai wrote:
At Mon, 9 Feb 2015 08:52:49 +0000, Kuninori Morimoto wrote:
From: Kuninori Morimoto kuninori.morimoto.gx@renesas.com
ASoC devices are organized as CPU-CARD-CODEC. Then, CPU/CODEC are based on component structure. Now, each CARD device knows connected component. But CARD doesn't notice if connected component was removed when user called rmmod or unbind in current implementation. Thus, CARD which lost some components still exist in system. And then, ALSA sound card will have some problem if user used this CARD in such timing. This patch temporarily removes CARD from system if connected component was removed, and re-add it if some component was added.
Reported-by: Nguyen Viet Dung nv-dung@jinso.co.jp Reported-by: Bui Duc Phuc bd-phuc@jinso.co.jp Reported-by: Cao Minh Hiep cm-hiep@jinso.co.jp Signed-off-by: Kuninori Morimoto kuninori.morimoto.gx@renesas.com
include/sound/soc.h | 1 + sound/soc/soc-core.c | 26 ++++++++++++++++++++++++++ 2 files changed, 27 insertions(+)
diff --git a/include/sound/soc.h b/include/sound/soc.h index b4fca9a..a90eff4 100644 --- a/include/sound/soc.h +++ b/include/sound/soc.h @@ -1083,6 +1083,7 @@ struct snd_soc_card { struct list_head paths; struct list_head dapm_list; struct list_head dapm_dirty;
struct list_head unbinded_list;
/* Generic DAPM context for the card */ struct snd_soc_dapm_context dapm;
diff --git a/sound/soc/soc-core.c b/sound/soc/soc-core.c index b7ab676..f8d5498 100644 --- a/sound/soc/soc-core.c +++ b/sound/soc/soc-core.c @@ -56,6 +56,7 @@ static DEFINE_MUTEX(client_mutex); static LIST_HEAD(platform_list); static LIST_HEAD(codec_list); static LIST_HEAD(component_list); +static LIST_HEAD(unbinded_card_list);
/* * This is a timeout to do a DAPM powerdown after a stream is closed(). @@ -2406,6 +2407,10 @@ int snd_soc_unregister_card(struct snd_soc_card *card) dev_dbg(card->dev, "ASoC: Unregistered card '%s'\n", card->name); }
- mutex_lock(&client_mutex);
- list_del(&card->unbinded_list);
- mutex_unlock(&client_mutex);
- return 0; } EXPORT_SYMBOL_GPL(snd_soc_unregister_card);
@@ -2669,6 +2674,9 @@ EXPORT_SYMBOL_GPL(snd_soc_component_exit_regmap);
static void snd_soc_component_add_unlocked(struct snd_soc_component *component) {
- struct snd_soc_card *card, *_card;
- int ret;
- if (!component->write && !component->read) { if (!component->regmap) component->regmap = dev_get_regmap(component->dev, NULL);
@@ -2677,6 +2685,16 @@ static void snd_soc_component_add_unlocked(struct snd_soc_component *component) }
list_add(&component->list, &component_list);
- /* re-add temporarily removed card if exist */
- list_for_each_entry_safe(card, _card, &unbinded_card_list,
unbinded_list) {ret = snd_soc_instantiate_card(card);if (ret < 0)continue;list_del(&card->unbinded_list);- }
[Cc'ed Lars-Peter]
This would instantiate a card even if it's irrelevant with the given component? If so, it looks fragile, and possibly racy, when there are multiple cards.
That shouldn't be a problem. snd_soc_instantiate_card() does the proper locking and will return an error if not all components are ready yet.
But there is no check of card->instantiated there, so the whole path can be called again even if the card was already instantiated?
The card won't be on the list of unbound cards if it is not instantiated. And this whole section is protected by the client_mutex.
OK, that explains. Thanks.
(BTW, client_mutex doesn't cover the whole places accessing the component_list in soc-core.c.)
Takashi
On 02/09/2015 02:35 PM, Takashi Iwai wrote: [...]
(BTW, client_mutex doesn't cover the whole places accessing the component_list in soc-core.c.)
Yea, that looks pretty broken, I'll have a look.
Hi Lars
ASoC devices are organized as CPU-CARD-CODEC. Then, CPU/CODEC are based on component structure. Now, each CARD device knows connected component. But CARD doesn't notice if connected component was removed when user called rmmod or unbind in current implementation. Thus, CARD which lost some components still exist in system. And then, ALSA sound card will have some problem if user used this CARD in such timing. This patch temporarily removes CARD from system if connected component was removed, and re-add it if some component was added.
Reported-by: Nguyen Viet Dung nv-dung@jinso.co.jp Reported-by: Bui Duc Phuc bd-phuc@jinso.co.jp Reported-by: Cao Minh Hiep cm-hiep@jinso.co.jp Signed-off-by: Kuninori Morimoto kuninori.morimoto.gx@renesas.com
(snip)
The other issue is that it introduces subtly issues with the suspend and resume order. Suspend and resume are called in the order in which the probe functions of devices are called (and succeed). By returning -EPROBE_DEFER in the card driver we make sure that the card's probe function is always called after the probe functions of all the components of the card have run. This again causes the card's suspend function to be called before any suspend function of any of it's components. Now with this patch it is possible again for a component's probe function to be called after the card's probe function which changes the suspend and resume order and might break things.
This patch essentially is a partial revert of commit b19e6e7b763 ("ASoC: core: Use driver core probe deferral") where the card list was replaced with the -EPROBE_DEFER mechanism.
So while this patch fixes the nasty crash it introduces some other subtle issue. Maybe we should also add a big WARN() when a component of a card is removed while still in use until the other issues are also fixed.
OK. I can send such patch. Oops ? does this mean my patch-set + WARN() ? or just WARN() ?
Best regards --- Kuninori Morimoto
On 02/10/2015 01:44 AM, Kuninori Morimoto wrote:
Hi Lars
ASoC devices are organized as CPU-CARD-CODEC. Then, CPU/CODEC are based on component structure. Now, each CARD device knows connected component. But CARD doesn't notice if connected component was removed when user called rmmod or unbind in current implementation. Thus, CARD which lost some components still exist in system. And then, ALSA sound card will have some problem if user used this CARD in such timing. This patch temporarily removes CARD from system if connected component was removed, and re-add it if some component was added.
Reported-by: Nguyen Viet Dung nv-dung@jinso.co.jp Reported-by: Bui Duc Phuc bd-phuc@jinso.co.jp Reported-by: Cao Minh Hiep cm-hiep@jinso.co.jp Signed-off-by: Kuninori Morimoto kuninori.morimoto.gx@renesas.com
(snip)
The other issue is that it introduces subtly issues with the suspend and resume order. Suspend and resume are called in the order in which the probe functions of devices are called (and succeed). By returning -EPROBE_DEFER in the card driver we make sure that the card's probe function is always called after the probe functions of all the components of the card have run. This again causes the card's suspend function to be called before any suspend function of any of it's components. Now with this patch it is possible again for a component's probe function to be called after the card's probe function which changes the suspend and resume order and might break things.
This patch essentially is a partial revert of commit b19e6e7b763 ("ASoC: core: Use driver core probe deferral") where the card list was replaced with the -EPROBE_DEFER mechanism.
So while this patch fixes the nasty crash it introduces some other subtle issue. Maybe we should also add a big WARN() when a component of a card is removed while still in use until the other issues are also fixed.
OK. I can send such patch. Oops ? does this mean my patch-set + WARN() ? or just WARN() ?
Both. Or an alternative would be not to allow re-binding the card and force the user to unbind/bind the card before it starts working again.
- Lars
Hi Lars
OK. I can send such patch. Oops ? does this mean my patch-set + WARN() ? or just WARN() ?
Both. Or an alternative would be not to allow re-binding the card and force the user to unbind/bind the card before it starts working again.
Thanks. My v1 patch had this idea, but Mark wanted automatically re-binding.
participants (3)
-
Kuninori Morimoto -
Lars-Peter Clausen -
Takashi Iwai