Hi Jisheng,
On Wed, Jan 24, 2024 at 04:46:43AM +0000, JiSheng Teoh wrote:
> Hi Andi,
> >
> > ...
> >
> > > > > +static int __maybe_unused cdns_i2c_resume(struct device *dev) {
> > > > > + struct cdns_i2c *xi2c = dev_get_drvdata(dev);
> > > > > + int err;
> > > > > +
> > > > > + err = cdns_i2c_runtime_resume(dev);
> > > > > + if (err)
> > > > > + return err;
> > > > > +
> > > > > + if (pm_runtime_status_suspended(dev)) {
> > > > > + err = cdns_i2c_runtime_suspend(dev);
> > > > > + if (err)
> > > > > + return err;
> > > >
> > > > We call the cdns_i2c_resume() functions to come up from a suspended
> > > > state. But, if we fail to resume, we call the suspend and return '0' (because this always returns '0').
> > > >
> > > > In other words, if we take this path, we call resume, but we still end up suspended and return success.
> > > >
> > > > Andi
> > > >
> > >
> > > My understanding is that during system level resume 'cdns_i2c_resume()', the i2c device itself can still be held in runtime suspend
> > regardless of the change in system level PM.
> > > Looking back at this, we invoke cdns_i2c_runtime_resume() to enable clock and init the i2c device, the runtime PM state is still
> > unchanged and kept suspended.
> > > pm_runtime_status_suspended() will be evaluated as true, and runtime suspend 'cdns_i2c_runtime_suspend()' is invoked to
> > disable the clock. This balances the clock count enabled earlier.
> >
> > If this is your issue, what if we do not enable the clock during resume? and we just mark the device as resumed?
> >
> That will work as well. The i2c device will be runtime resumed again during cdns_i2c_master_xfer() anyway, but thought that it
> would be a good idea to check if the i2c device is able runtime resume during a system level resume.
That's fine, I think it might work this way, as well, so let's
keept it at your original implementation.
If we save here, we add complication somewhere else.
Reviewed-by: Andi Shyti <andi.shyti@xxxxxxxxxx>
Before applying the patch I will read it again to make sure all
is balanced.
Thanks,
Andi
> > > The runtime PM state is only resumed during cdns_i2c_master_xfer() through pm_runtime_resume_and_get(), and subsequently
> > kept suspended through pm_runtime_put_autosuspend().
> > > Since the cdns_i2c_runtime_suspend() always return '0', I will simplify them as follow:
> > > +if (pm_runtime_status_suspended(dev))
> > > + cdns_i2c_runtime_suspend(dev);
> >
> > I'd prefer checking the error value, even though we are sure on the expected return. It's more future proof.
> >
> > Andi
> >
> Ok, I will keep the original changes.
>
> > > > > + }
> > > > > +
> > > > > + i2c_mark_adapter_resumed(&xi2c->adap);
> > > > > +
> > > > > + return 0;
> > > > > +}
