| Message ID | 1548345432-11449-1-git-send-email-spujar@nvidia.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | ALSA: hda: runtime PM is always active | expand |
On Thu, 24 Jan 2019 16:57:11 +0100, Sameer Pujar wrote: > > The runtime PM count is incremented and set to active during hda codec > device init, but it is decremented and set to suspend during exit only. > Hence the runtime PM status is always active and hda device cannot be > put to runtime suspend. Keeping device usage active for entire period, > though nothing really happening on the device, seems unnecessary. > > This patch avoides incrementing runtime PM usage count of the device. > > Signed-off-by: Sameer Pujar <spujar@nvidia.com> > Reviewed-by: Ravindra Lokhande <rlokhande@nvidia.com> > Reviewed-by: Mohan Kumar D <mkumard@nvidia.com> This breaks the existing things, I'm afraid. Did you really test and verify the behavior...? The runtime PM refcount is decremented at snd_hda_codec_register(), and that's the place to let runtime suspend of the codec really effective. (It's not about the controller, though.) thanks, Takashi > --- > sound/hda/hdac_device.c | 6 ------ > 1 file changed, 6 deletions(-) > > diff --git a/sound/hda/hdac_device.c b/sound/hda/hdac_device.c > index 95b073e..72aa342 100644 > --- a/sound/hda/hdac_device.c > +++ b/sound/hda/hdac_device.c > @@ -31,9 +31,6 @@ static void default_release(struct device *dev) > * > * Returns zero for success or a negative error code. > * > - * This function increments the runtime PM counter and marks it active. > - * The caller needs to turn it off appropriately later. > - * > * The caller needs to set the device's release op properly by itself. > */ > int snd_hdac_device_init(struct hdac_device *codec, struct hdac_bus *bus, > @@ -55,8 +52,6 @@ int snd_hdac_device_init(struct hdac_device *codec, struct hdac_bus *bus, > codec->bus = bus; > codec->addr = addr; > codec->type = HDA_DEV_CORE; > - pm_runtime_set_active(&codec->dev); > - pm_runtime_get_noresume(&codec->dev); > atomic_set(&codec->in_pm, 0); > > err = snd_hdac_bus_add_device(bus, codec); > @@ -123,7 +118,6 @@ EXPORT_SYMBOL_GPL(snd_hdac_device_init); > */ > void snd_hdac_device_exit(struct hdac_device *codec) > { > - pm_runtime_put_noidle(&codec->dev); > snd_hdac_bus_remove_device(codec->bus, codec); > kfree(codec->vendor_name); > kfree(codec->chip_name); > -- > 2.7.4 > >
On 1/24/2019 9:34 PM, Takashi Iwai wrote: > On Thu, 24 Jan 2019 16:57:11 +0100, > Sameer Pujar wrote: >> The runtime PM count is incremented and set to active during hda codec >> device init, but it is decremented and set to suspend during exit only. >> Hence the runtime PM status is always active and hda device cannot be >> put to runtime suspend. Keeping device usage active for entire period, >> though nothing really happening on the device, seems unnecessary. >> >> This patch avoides incrementing runtime PM usage count of the device. >> >> Signed-off-by: Sameer Pujar <spujar@nvidia.com> >> Reviewed-by: Ravindra Lokhande <rlokhande@nvidia.com> >> Reviewed-by: Mohan Kumar D <mkumard@nvidia.com> > This breaks the existing things, I'm afraid. > Did you really test and verify the behavior...? > > The runtime PM refcount is decremented at snd_hda_codec_register(), > and that's the place to let runtime suspend of the codec really > effective. (It's not about the controller, though.) > It worked fine at my end. I could put the device to runtime suspend and resume it back multiple times. Since you are suggesting it could break things, let me rework on the patch and dig up a little more to see if there is better way to fix this. Thanks, Sameer. > thanks, > > Takashi > > >> --- >> sound/hda/hdac_device.c | 6 ------ >> 1 file changed, 6 deletions(-) >> >> diff --git a/sound/hda/hdac_device.c b/sound/hda/hdac_device.c >> index 95b073e..72aa342 100644 >> --- a/sound/hda/hdac_device.c >> +++ b/sound/hda/hdac_device.c >> @@ -31,9 +31,6 @@ static void default_release(struct device *dev) >> * >> * Returns zero for success or a negative error code. >> * >> - * This function increments the runtime PM counter and marks it active. >> - * The caller needs to turn it off appropriately later. >> - * >> * The caller needs to set the device's release op properly by itself. >> */ >> int snd_hdac_device_init(struct hdac_device *codec, struct hdac_bus *bus, >> @@ -55,8 +52,6 @@ int snd_hdac_device_init(struct hdac_device *codec, struct hdac_bus *bus, >> codec->bus = bus; >> codec->addr = addr; >> codec->type = HDA_DEV_CORE; >> - pm_runtime_set_active(&codec->dev); >> - pm_runtime_get_noresume(&codec->dev); >> atomic_set(&codec->in_pm, 0); >> >> err = snd_hdac_bus_add_device(bus, codec); >> @@ -123,7 +118,6 @@ EXPORT_SYMBOL_GPL(snd_hdac_device_init); >> */ >> void snd_hdac_device_exit(struct hdac_device *codec) >> { >> - pm_runtime_put_noidle(&codec->dev); >> snd_hdac_bus_remove_device(codec->bus, codec); >> kfree(codec->vendor_name); >> kfree(codec->chip_name); >> -- >> 2.7.4 >> >>
On Thu, 24 Jan 2019 19:15:12 +0100, Sameer Pujar wrote: > > > On 1/24/2019 9:34 PM, Takashi Iwai wrote: > > On Thu, 24 Jan 2019 16:57:11 +0100, > > Sameer Pujar wrote: > >> The runtime PM count is incremented and set to active during hda codec > >> device init, but it is decremented and set to suspend during exit only. > >> Hence the runtime PM status is always active and hda device cannot be > >> put to runtime suspend. Keeping device usage active for entire period, > >> though nothing really happening on the device, seems unnecessary. > >> > >> This patch avoides incrementing runtime PM usage count of the device. > >> > >> Signed-off-by: Sameer Pujar <spujar@nvidia.com> > >> Reviewed-by: Ravindra Lokhande <rlokhande@nvidia.com> > >> Reviewed-by: Mohan Kumar D <mkumard@nvidia.com> > > This breaks the existing things, I'm afraid. > > Did you really test and verify the behavior...? > > > > The runtime PM refcount is decremented at snd_hda_codec_register(), > > and that's the place to let runtime suspend of the codec really > > effective. (It's not about the controller, though.) > > > It worked fine at my end. I could put the device to runtime suspend > and resume it back multiple times. > Since you are suggesting it could break things, let me rework on the > patch and dig up a little more to > see if there is better way to fix this. Well, I don't see what you want to "fix". Do you see any bug? thanks, Takashi > > Thanks, > Sameer. > > > thanks, > > > > Takashi > > > > > >> --- > >> sound/hda/hdac_device.c | 6 ------ > >> 1 file changed, 6 deletions(-) > >> > >> diff --git a/sound/hda/hdac_device.c b/sound/hda/hdac_device.c > >> index 95b073e..72aa342 100644 > >> --- a/sound/hda/hdac_device.c > >> +++ b/sound/hda/hdac_device.c > >> @@ -31,9 +31,6 @@ static void default_release(struct device *dev) > >> * > >> * Returns zero for success or a negative error code. > >> * > >> - * This function increments the runtime PM counter and marks it active. > >> - * The caller needs to turn it off appropriately later. > >> - * > >> * The caller needs to set the device's release op properly by itself. > >> */ > >> int snd_hdac_device_init(struct hdac_device *codec, struct hdac_bus *bus, > >> @@ -55,8 +52,6 @@ int snd_hdac_device_init(struct hdac_device *codec, struct hdac_bus *bus, > >> codec->bus = bus; > >> codec->addr = addr; > >> codec->type = HDA_DEV_CORE; > >> - pm_runtime_set_active(&codec->dev); > >> - pm_runtime_get_noresume(&codec->dev); > >> atomic_set(&codec->in_pm, 0); > >> err = snd_hdac_bus_add_device(bus, codec); > >> @@ -123,7 +118,6 @@ EXPORT_SYMBOL_GPL(snd_hdac_device_init); > >> */ > >> void snd_hdac_device_exit(struct hdac_device *codec) > >> { > >> - pm_runtime_put_noidle(&codec->dev); > >> snd_hdac_bus_remove_device(codec->bus, codec); > >> kfree(codec->vendor_name); > >> kfree(codec->chip_name); > >> -- > >> 2.7.4 > >> > >> >
On 1/25/2019 12:13 AM, Takashi Iwai wrote: > On Thu, 24 Jan 2019 19:15:12 +0100, > Sameer Pujar wrote: >> >> On 1/24/2019 9:34 PM, Takashi Iwai wrote: >>> On Thu, 24 Jan 2019 16:57:11 +0100, >>> Sameer Pujar wrote: >>>> The runtime PM count is incremented and set to active during hda codec >>>> device init, but it is decremented and set to suspend during exit only. >>>> Hence the runtime PM status is always active and hda device cannot be >>>> put to runtime suspend. Keeping device usage active for entire period, >>>> though nothing really happening on the device, seems unnecessary. >>>> >>>> This patch avoides incrementing runtime PM usage count of the device. >>>> >>>> Signed-off-by: Sameer Pujar <spujar@nvidia.com> >>>> Reviewed-by: Ravindra Lokhande <rlokhande@nvidia.com> >>>> Reviewed-by: Mohan Kumar D <mkumard@nvidia.com> >>> This breaks the existing things, I'm afraid. >>> Did you really test and verify the behavior...? >>> >>> The runtime PM refcount is decremented at snd_hda_codec_register(), >>> and that's the place to let runtime suspend of the codec really >>> effective. (It's not about the controller, though.) >>> >> It worked fine at my end. I could put the device to runtime suspend >> and resume it back multiple times. >> Since you are suggesting it could break things, let me rework on the >> patch and dig up a little more to >> see if there is better way to fix this. > Well, I don't see what you want to "fix". Do you see any bug? I was expecting the device to be runtime suspended, once probe is finished, until I playback something on it. But it appears that rutntime PM is always active and runtime PM suspend callback never happens. > thanks, > > Takashi > >> Thanks, >> Sameer. >> >>> thanks, >>> >>> Takashi >>> >>> >>>> --- >>>> sound/hda/hdac_device.c | 6 ------ >>>> 1 file changed, 6 deletions(-) >>>> >>>> diff --git a/sound/hda/hdac_device.c b/sound/hda/hdac_device.c >>>> index 95b073e..72aa342 100644 >>>> --- a/sound/hda/hdac_device.c >>>> +++ b/sound/hda/hdac_device.c >>>> @@ -31,9 +31,6 @@ static void default_release(struct device *dev) >>>> * >>>> * Returns zero for success or a negative error code. >>>> * >>>> - * This function increments the runtime PM counter and marks it active. >>>> - * The caller needs to turn it off appropriately later. >>>> - * >>>> * The caller needs to set the device's release op properly by itself. >>>> */ >>>> int snd_hdac_device_init(struct hdac_device *codec, struct hdac_bus *bus, >>>> @@ -55,8 +52,6 @@ int snd_hdac_device_init(struct hdac_device *codec, struct hdac_bus *bus, >>>> codec->bus = bus; >>>> codec->addr = addr; >>>> codec->type = HDA_DEV_CORE; >>>> - pm_runtime_set_active(&codec->dev); >>>> - pm_runtime_get_noresume(&codec->dev); >>>> atomic_set(&codec->in_pm, 0); >>>> err = snd_hdac_bus_add_device(bus, codec); >>>> @@ -123,7 +118,6 @@ EXPORT_SYMBOL_GPL(snd_hdac_device_init); >>>> */ >>>> void snd_hdac_device_exit(struct hdac_device *codec) >>>> { >>>> - pm_runtime_put_noidle(&codec->dev); >>>> snd_hdac_bus_remove_device(codec->bus, codec); >>>> kfree(codec->vendor_name); >>>> kfree(codec->chip_name); >>>> -- >>>> 2.7.4 >>>> >>>>
On Thu, 24 Jan 2019 20:01:13 +0100, Sameer Pujar wrote: > > > On 1/25/2019 12:13 AM, Takashi Iwai wrote: > > On Thu, 24 Jan 2019 19:15:12 +0100, > > Sameer Pujar wrote: > >> > >> On 1/24/2019 9:34 PM, Takashi Iwai wrote: > >>> On Thu, 24 Jan 2019 16:57:11 +0100, > >>> Sameer Pujar wrote: > >>>> The runtime PM count is incremented and set to active during hda codec > >>>> device init, but it is decremented and set to suspend during exit only. > >>>> Hence the runtime PM status is always active and hda device cannot be > >>>> put to runtime suspend. Keeping device usage active for entire period, > >>>> though nothing really happening on the device, seems unnecessary. > >>>> > >>>> This patch avoides incrementing runtime PM usage count of the device. > >>>> > >>>> Signed-off-by: Sameer Pujar <spujar@nvidia.com> > >>>> Reviewed-by: Ravindra Lokhande <rlokhande@nvidia.com> > >>>> Reviewed-by: Mohan Kumar D <mkumard@nvidia.com> > >>> This breaks the existing things, I'm afraid. > >>> Did you really test and verify the behavior...? > >>> > >>> The runtime PM refcount is decremented at snd_hda_codec_register(), > >>> and that's the place to let runtime suspend of the codec really > >>> effective. (It's not about the controller, though.) > >>> > >> It worked fine at my end. I could put the device to runtime suspend > >> and resume it back multiple times. > >> Since you are suggesting it could break things, let me rework on the > >> patch and dig up a little more to > >> see if there is better way to fix this. > > Well, I don't see what you want to "fix". Do you see any bug? > I was expecting the device to be runtime suspended, once probe is > finished, until I playback something on it. > But it appears that rutntime PM is always active and runtime PM > suspend callback never happens. Does the behavior really appear? If so, it's a bug. The current design is to enable the runtime PM *after* the device registration. Otherwise it makes no sense -- the problem doesn't finish yet before that point. thanks, Takashi
diff --git a/sound/hda/hdac_device.c b/sound/hda/hdac_device.c index 95b073e..72aa342 100644 --- a/sound/hda/hdac_device.c +++ b/sound/hda/hdac_device.c @@ -31,9 +31,6 @@ static void default_release(struct device *dev) * * Returns zero for success or a negative error code. * - * This function increments the runtime PM counter and marks it active. - * The caller needs to turn it off appropriately later. - * * The caller needs to set the device's release op properly by itself. */ int snd_hdac_device_init(struct hdac_device *codec, struct hdac_bus *bus, @@ -55,8 +52,6 @@ int snd_hdac_device_init(struct hdac_device *codec, struct hdac_bus *bus, codec->bus = bus; codec->addr = addr; codec->type = HDA_DEV_CORE; - pm_runtime_set_active(&codec->dev); - pm_runtime_get_noresume(&codec->dev); atomic_set(&codec->in_pm, 0); err = snd_hdac_bus_add_device(bus, codec); @@ -123,7 +118,6 @@ EXPORT_SYMBOL_GPL(snd_hdac_device_init); */ void snd_hdac_device_exit(struct hdac_device *codec) { - pm_runtime_put_noidle(&codec->dev); snd_hdac_bus_remove_device(codec->bus, codec); kfree(codec->vendor_name); kfree(codec->chip_name);