Commit 70a6404ff610 ("ASoC: rt5640: Fix sleep in atomic context") not only switched from request_irq() to request_threaded_irq(), to fix the sleep in atomic context issue, but it also added devm management of the IRQ by actually switching to devm_request_threaded_irq() (without any explanation in the commit message for this change).

This is wrong since the IRQ was already explicitly managed by the driver. On unbind the ASoC core will call rt5640_set_jack(NULL) which in turn will call rt5640_disable_jack_detect() which frees the IRQ already. So now we have a double free.

Besides the unexplained switch to devm being wrong, the actual fix for the sleep in atomic context issue also is not the best solution.

The only thing which rt5640_irq() does is cancel + (re-)queue the jack_work delayed_work. This can be done in a single non sleeping call by replacing queue_delayed_work() with mod_delayed_work(), which does not sleep. Using mod_delayed_work() is a much better fix then adding a thread which does nothing other then queuing a work-item.

This patch is a straight revert of the troublesome changes, the switch to mod_delayed_work() is done in a separate follow-up patch.

Fixes: 70a6404ff610 ("ASoC: rt5640: Fix sleep in atomic context") Cc: Sameer Pujar spujar@nvidia.com Cc: Oder Chiou oder_chiou@realtek.com Signed-off-by: Hans de Goede hdegoede@redhat.com --- sound/soc/codecs/rt5640.c | 12 +++++------- 1 file changed, 5 insertions(+), 7 deletions(-)

diff --git a/sound/soc/codecs/rt5640.c b/sound/soc/codecs/rt5640.c index 15e1a62b9e57..05ff8066171b 100644 --- a/sound/soc/codecs/rt5640.c +++ b/sound/soc/codecs/rt5640.c @@ -2565,10 +2565,9 @@ static void rt5640_enable_jack_detect(struct snd_soc_component *component, if (jack_data && jack_data->use_platform_clock) rt5640->use_platform_clock = jack_data->use_platform_clock;

- ret = devm_request_threaded_irq(component->dev, rt5640->irq, - NULL, rt5640_irq, - IRQF_TRIGGER_RISING | IRQF_TRIGGER_FALLING | IRQF_ONESHOT, - "rt5640", rt5640); + ret = request_irq(rt5640->irq, rt5640_irq, + IRQF_TRIGGER_RISING | IRQF_TRIGGER_FALLING | IRQF_ONESHOT, + "rt5640", rt5640); if (ret) { dev_warn(component->dev, "Failed to request IRQ %d: %d\n", rt5640->irq, ret); rt5640_disable_jack_detect(component); @@ -2621,9 +2620,8 @@ static void rt5640_enable_hda_jack_detect(

rt5640->jack = jack;

- ret = devm_request_threaded_irq(component->dev, rt5640->irq, - NULL, rt5640_irq, IRQF_TRIGGER_RISING | IRQF_ONESHOT, - "rt5640", rt5640); + ret = request_irq(rt5640->irq, rt5640_irq, + IRQF_TRIGGER_RISING | IRQF_ONESHOT, "rt5640", rt5640); if (ret) { dev_warn(component->dev, "Failed to request IRQ %d: %d\n", rt5640->irq, ret); rt5640->irq = -ENXIO;

When jack-detect was originally added disabling the IRQ during suspend was done by the sound/soc/intel/boards/bytcr_rt5640.c driver calling snd_soc_component_set_jack(NULL) on suspend, which calls rt5640_disable_jack_detect(), which calls free_irq() which also disables it.

Commit 5fabcc90e79b ("ASoC: rt5640: Fix Jack work after system suspend") added disable_irq() / enable_irq() calls on suspend/resume for machine drivers which do not call snd_soc_component_set_jack(NULL) on suspend.

The new disable_irq() / enable_irq() are made conditional by "if (rt5640->irq)" statements, but this is true for the machine drivers which do call snd_soc_component_set_jack(NULL) on suspend too, causing a disable_irq() call there on the already free-ed IRQ.

Change the "if (rt5640->irq)" condition to "if (rt5640->jack)" to fix this, rt5640->jack is only set if the jack-detect IRQ handler is still active when rt5640_suspend() runs.

And adjust rt5640_enable_hda_jack_detect()'s request_irq() error handling to set rt5640->jack to NULL to match (note that the old setting of irq to -ENOXIO still resulted in disable_irq(-ENOXIO) calls on suspend).

Fixes: 5fabcc90e79b ("ASoC: rt5640: Fix Jack work after system suspend") Cc: Oder Chiou oder_chiou@realtek.com Signed-off-by: Hans de Goede hdegoede@redhat.com --- sound/soc/codecs/rt5640.c | 7 +++---- 1 file changed, 3 insertions(+), 4 deletions(-)

diff --git a/sound/soc/codecs/rt5640.c b/sound/soc/codecs/rt5640.c index 5c34c045d396..1bc281d42ca8 100644 --- a/sound/soc/codecs/rt5640.c +++ b/sound/soc/codecs/rt5640.c @@ -2622,7 +2622,7 @@ static void rt5640_enable_hda_jack_detect( IRQF_TRIGGER_RISING | IRQF_ONESHOT, "rt5640", rt5640); if (ret) { dev_warn(component->dev, "Failed to request IRQ %d: %d\n", rt5640->irq, ret); - rt5640->irq = -ENXIO; + rt5640->jack = NULL; return; }

@@ -2797,7 +2797,7 @@ static int rt5640_suspend(struct snd_soc_component *component) { struct rt5640_priv *rt5640 = snd_soc_component_get_drvdata(component);

- if (rt5640->irq) { + if (rt5640->jack) { /* disable jack interrupts during system suspend */ disable_irq(rt5640->irq); } @@ -2825,10 +2825,9 @@ static int rt5640_resume(struct snd_soc_component *component) regcache_cache_only(rt5640->regmap, false); regcache_sync(rt5640->regmap);

- if (rt5640->irq) + if (rt5640->jack) { enable_irq(rt5640->irq);

- if (rt5640->jack) { if (rt5640->jd_src == RT5640_JD_SRC_HDA_HEADER) { snd_soc_component_update_bits(component, RT5640_DUMMY2, 0x1100, 0x1100);

Set "rt5640->irq_requested = true" after a successful request_irq() in rt5640_enable_hda_jack_detect(), so that rt5640_disable_jack_detect() properly frees the IRQ.

This fixes the IRQ not being freed on rmmod / driver unbind.

Fixes: 2b9c8d2b3c89 ("ASoC: rt5640: Add the HDA header support") Cc: Oder Chiou oder_chiou@realtek.com Signed-off-by: Hans de Goede hdegoede@redhat.com --- sound/soc/codecs/rt5640.c | 1 + 1 file changed, 1 insertion(+)

diff --git a/sound/soc/codecs/rt5640.c b/sound/soc/codecs/rt5640.c index 03c866c04c7a..a4a11407ab10 100644 --- a/sound/soc/codecs/rt5640.c +++ b/sound/soc/codecs/rt5640.c @@ -2625,6 +2625,7 @@ static void rt5640_enable_hda_jack_detect( rt5640->jack = NULL; return; } + rt5640->irq_requested = true;

/* sync initial jack state */ queue_delayed_work(system_long_wq, &rt5640->jack_work, 0);