On 26-11-20, 03:11, Liao, Bard wrote:
> > -----Original Message-----
> > From: Vinod Koul <vkoul@xxxxxxxxxx>
> > Sent: Wednesday, November 25, 2020 1:05 PM
> > To: Bard Liao <yung-chuan.liao@xxxxxxxxxxxxxxx>
> > Cc: alsa-devel@xxxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx;
> > gregkh@xxxxxxxxxxxxxxxxxxx; jank@xxxxxxxxxxx;
> > srinivas.kandagatla@xxxxxxxxxx; rander.wang@xxxxxxxxxxxxxxx;
> > ranjani.sridharan@xxxxxxxxxxxxxxx; hui.wang@xxxxxxxxxxxxx; pierre-
> > louis.bossart@xxxxxxxxxxxxxxx; Kale, Sanyog R <sanyog.r.kale@xxxxxxxxx>; Lin,
> > Mengdong <mengdong.lin@xxxxxxxxx>; Liao, Bard <bard.liao@xxxxxxxxx>
> > Subject: Re: [PATCH] soundwire: master: use pm_runtime_set_active() on
> > add
> >
> > On 24-11-20, 21:07, Bard Liao wrote:
> > > From: Pierre-Louis Bossart <pierre-louis.bossart@xxxxxxxxxxxxxxx>
> > >
> > > The 'master' device acts as a glue layer used during bus
> > > initialization only, and it needs to be 'transparent' for pm_runtime
> > > management. Its behavior should be that it becomes active when one of
> > > its children becomes active, and suspends when all of its children are
> > > suspended.
> > >
> > > In our tests on Intel platforms, we routinely see these sort of
> > > warnings on the initial boot:
> > >
> > > [ 21.447345] rt715 sdw:3:25d:715:0: runtime PM trying to activate
> > > child device sdw:3:25d:715:0 but parent (sdw-master-3) is not active
> > >
> > > This is root-caused to a missing setup to make the device 'active' on
> > > probe. Since we don't want the device to remain active forever after
> > > the probe, the autosuspend configuration is also enabled at the end of
> > > the probe - the device will actually autosuspend only in the case
> > > where there are no devices physically attached. In practice, the
> > > master device will suspend when all its children are no longer active.
> > >
> > > Fixes: bd84256e86ecf ('soundwire: master: enable pm runtime')
> > > Signed-off-by: Pierre-Louis Bossart
> > > <pierre-louis.bossart@xxxxxxxxxxxxxxx>
> > > Reviewed-by: Rander Wang <rander.wang@xxxxxxxxxxxxxxx>
> > > Signed-off-by: Bard Liao <yung-chuan.liao@xxxxxxxxxxxxxxx>
> > > ---
> > > drivers/soundwire/master.c | 14 ++++++++++++++
> > > 1 file changed, 14 insertions(+)
> > >
> > > diff --git a/drivers/soundwire/master.c b/drivers/soundwire/master.c
> > > index 3488bb824e84..9b05c9e25ebe 100644
> > > --- a/drivers/soundwire/master.c
> > > +++ b/drivers/soundwire/master.c
> > > @@ -8,6 +8,15 @@
> > > #include <linux/soundwire/sdw_type.h> #include "bus.h"
> > >
> > > +/*
> > > + * The 3s value for autosuspend will only be used if there are no
> > > + * devices physically attached on a bus segment. In practice enabling
> > > + * the bus operation will result in children devices become active
> > > +and
> > > + * the master device will only suspend when all its children are no
> > > + * longer active.
> > > + */
> > > +#define SDW_MASTER_SUSPEND_DELAY_MS 3000
> > > +
> > > /*
> > > * The sysfs for properties reflects the MIPI description as given
> > > * in the MIPI DisCo spec
> > > @@ -154,7 +163,12 @@ int sdw_master_device_add(struct sdw_bus *bus,
> > struct device *parent,
> > > bus->dev = &md->dev;
> > > bus->md = md;
> > >
> > > + pm_runtime_set_autosuspend_delay(&bus->md->dev,
> > SDW_MASTER_SUSPEND_DELAY_MS);
> > > + pm_runtime_use_autosuspend(&bus->md->dev);
> > > + pm_runtime_mark_last_busy(&bus->md->dev);
> > > + pm_runtime_set_active(&bus->md->dev);
> > > pm_runtime_enable(&bus->md->dev);
> > > + pm_runtime_idle(&bus->md->dev);
> >
> > I understand that this needs to be done somewhere but is the core the right
> > place. In intel case it maybe a dummy device but other controllers are real
> > devices and may not support pm.
> >
> > I think better idea would be to do this in respective driver.. that way it
> > would be an opt-in for device supporting pm.
>
> Should it be put in the same place as pm_runtime_enable?
> IMHO, pm_runtime_enable is in the core already and it seems to be harmless
> for devices which don't support pm. And pm can still be optional on md's
> parent device.
For intel case yes, but world is not only intel, there are md which do
not have a parent like sof. they are real sdw controller devices
--
~Vinod
