Re: [PATCH v2] bus: mhi: host: pci_generic: Add missing poweroff() PM callback

[Manivannan Sadhasivam]

On Tue, Apr 05, 2022 at 01:23:22PM -0700, Bhaumik Vasav Bhatt wrote:
> 
> On 4/5/2022 5:59 AM, Manivannan Sadhasivam wrote:
> > During hibernation process, once thaw() stage completes, the MHI endpoint
> > devices will be in M0 state post recovery. After that, the devices will be
> > powered down so that the system can enter the target sleep state. During
> > this stage, the PCI core will put the devices in D3hot. But this transition
> > is allowed by the MHI spec. The devices can only enter D3hot when it is in
> > M3 state.
> > 
> > So for fixing this issue, let's add the poweroff() callback that will get
> > executed before putting the system in target sleep state during
> > hibernation. This callback will power down the device properly so that it
> > could be restored during restore() or thaw() stage.
> > 
> > Cc: stable@xxxxxxxxxxxxxxx
> > Fixes: 5f0c2ee1fe8d ("bus: mhi: pci-generic: Fix hibernation")
> > Reported-by: Hemant Kumar <quic_hemantk@xxxxxxxxxxx>
> > Suggested-by: Hemant Kumar <quic_hemantk@xxxxxxxxxxx>
> > Signed-off-by: Manivannan Sadhasivam <manivannan.sadhasivam@xxxxxxxxxx>
> > ---
> > 
> > Changes in v2:
> > 
> > * Hemant suggested to use restore function for poweroff() callback as we can
> > make sure that the device gets powered down properly.
> > 
> >   drivers/bus/mhi/host/pci_generic.c | 1 +
> >   1 file changed, 1 insertion(+)
> > 
> > diff --git a/drivers/bus/mhi/host/pci_generic.c b/drivers/bus/mhi/host/pci_generic.c
> > index 9527b7d63840..ef85dbfb3216 100644
> > --- a/drivers/bus/mhi/host/pci_generic.c
> > +++ b/drivers/bus/mhi/host/pci_generic.c
> > @@ -1085,6 +1085,7 @@ static const struct dev_pm_ops mhi_pci_pm_ops = {
> >   	.resume = mhi_pci_resume,
> >   	.freeze = mhi_pci_freeze,
> >   	.thaw = mhi_pci_restore,
> > +	.poweroff = mhi_pci_freeze,
> 
> It is possible that .thaw() queues recovery work and recovery work is still
> running
> 
> while .poweroff() is called. I would suggest adding a flush_work() in freeze
> such that
> 
> we don't try to power off while we're also trying to power on MHI from
> recovery work.
> 

These two cannot run in parallel because the power down will only happen if the
MHI_PCI_DEV_STARTED bit is set in mhi_pdev->status and that's the last step of
recovery. But the real issue I found is, if the recovery didn't finish during
powerdown() stage, then the function will just return and the device might be
powered on later.

So to avoid this scenario, we should have the flush_workqueue() in freeze.

I'll submit a different patch for that.

Thanks,
Mani

> >   	.restore = mhi_pci_restore,
> >   #endif
> >   };