Re: [PATCH v2 2/3] cpufreq: qcom-nvmem: Enable virtual power domain devices

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Tue, 24 Oct 2023 at 14:03, Stephan Gerhold
<stephan.gerhold@xxxxxxxxxxxxxxx> wrote:
>
> On Thu, Oct 19, 2023 at 01:26:19PM +0200, Ulf Hansson wrote:
> > On Thu, 19 Oct 2023 at 12:24, Ulf Hansson <ulf.hansson@xxxxxxxxxx> wrote:
> > >
> > > On Wed, 18 Oct 2023 at 10:06, Stephan Gerhold
> > > <stephan.gerhold@xxxxxxxxxxxxxxx> wrote:
> > > >
> > > > The genpd core caches performance state votes from devices that are
> > > > runtime suspended as of commit 3c5a272202c2 ("PM: domains: Improve
> > > > runtime PM performance state handling"). They get applied once the
> > > > device becomes active again.
> > > >
> > > > To attach the power domains needed by qcom-cpufreq-nvmem the OPP core
> > > > calls genpd_dev_pm_attach_by_id(). This results in "virtual" dummy
> > > > devices that use runtime PM only to control the enable and performance
> > > > state for the attached power domain.
> > > >
> > > > However, at the moment nothing ever resumes the virtual devices created
> > > > for qcom-cpufreq-nvmem. They remain permanently runtime suspended. This
> > > > means that performance state votes made during cpufreq scaling get
> > > > always cached and never applied to the hardware.
> > > >
> > > > Fix this by enabling the devices after attaching them and use
> > > > dev_pm_syscore_device() to ensure the power domains also stay on when
> > > > going to suspend. Since it supplies the CPU we can never turn it off
> > > > from Linux. There are other mechanisms to turn it off when needed,
> > > > usually in the RPM firmware (RPMPD) or the cpuidle path (CPR genpd).
> > >
> > > I believe we discussed using dev_pm_syscore_device() for the previous
> > > version. It's not intended to be used for things like the above.
> > >
> > > Moreover, I was under the impression that it wasn't really needed. In
> > > fact, I would think that this actually breaks things for system
> > > suspend/resume, as in this case the cpr driver's genpd
> > > ->power_on|off() callbacks are no longer getting called due this,
> > > which means that the cpr state machine isn't going to be restored
> > > properly. Or did I get this wrong?
> >
> > BTW, if you really need something like the above, the proper way to do
> > it would instead be to call device_set_awake_path() for the device.
> >
>
> Unfortunately this does not work correctly. When I use
> device_set_awake_path() it does set dev->power.wakeup_path = true.
> However, this flag is cleared again in device_prepare() when entering
> suspend. To me it looks a bit like wakeup_path is not supposed to be set
> directly by drivers? Before and after your commit 8512220c5782 ("PM /
> core: Assign the wakeup_path status flag in __device_prepare()") it
> seems to be internally bound to device_may_wakeup().
>
> It works if I make device_may_wakeup() return true, with
>
>         device_set_wakeup_capable(dev, true);
>         device_wakeup_enable(dev);
>
> but that also allows *disabling* the wakeup from sysfs which doesn't
> really make sense for the CPU.
>
> Any ideas?

The device_set_awake_path() should be called from a system suspend
callback. So you need to add that callback for the cpufreq driver.

Sorry, if that wasn't clear.

Kind regards
Uffe




[Index of Archives]     [Device Tree Compilter]     [Device Tree Spec]     [Linux Driver Backports]     [Video for Linux]     [Linux USB Devel]     [Linux PCI Devel]     [Linux Audio Users]     [Linux Kernel]     [Linux SCSI]     [XFree86]     [Yosemite Backpacking]


  Powered by Linux