Re: [PATCH v2] PCI: PM: Move to D0 before calling pci_legacy_resume_early()

[Rafael J. Wysocki]

On 10/7/2019 8:57 PM, Dexuan Cui wrote:
> > -----Original Message-----
> > From: Bjorn Helgaas <helgaas@xxxxxxxxxx>
> > Sent: Monday, October 7, 2019 6:24 AM
> > To: Dexuan Cui <decui@xxxxxxxxxxxxx>
> > Cc: lorenzo.pieralisi@xxxxxxx; linux-pci@xxxxxxxxxxxxxxx; Michael Kelley
> > <mikelley@xxxxxxxxxxxxx>; linux-hyperv@xxxxxxxxxxxxxxx;
> > linux-kernel@xxxxxxxxxxxxxxx; driverdev-devel@xxxxxxxxxxxxxxxxxxxxxx; Sasha
> > Levin <Alexander.Levin@xxxxxxxxxxxxx>; Haiyang Zhang
> > <haiyangz@xxxxxxxxxxxxx>; KY Srinivasan <kys@xxxxxxxxxxxxx>;
> > olaf@xxxxxxxxx; apw@xxxxxxxxxxxxx; jasowang@xxxxxxxxxx; vkuznets
> > <vkuznets@xxxxxxxxxx>; marcelo.cerri@xxxxxxxxxxxxx; Stephen Hemminger
> > <sthemmin@xxxxxxxxxxxxx>; jackm@xxxxxxxxxxxx
> > Subject: Re: [PATCH v2] PCI: PM: Move to D0 before calling
> > pci_legacy_resume_early()
> >
> > On Wed, Aug 14, 2019 at 01:06:55AM +0000, Dexuan Cui wrote:
> > > In pci_legacy_suspend_late(), the device state is moved to PCI_UNKNOWN.
> > >
> > > In pci_pm_thaw_noirq(), the state is supposed to be moved back to PCI_D0,
> > > but the current code misses the pci_legacy_resume_early() path, so the
> > > state remains in PCI_UNKNOWN in that path. As a result, in the resume
> > > phase of hibernation, this causes an error for the Mellanox VF driver,
> > > which fails to enable MSI-X because pci_msi_supported() is false due
> > > to dev->current_state != PCI_D0:
> > >
> > > mlx4_core a6d1:00:02.0: Detected virtual function - running in slave mode
> > > mlx4_core a6d1:00:02.0: Sending reset
> > > mlx4_core a6d1:00:02.0: Sending vhcr0
> > > mlx4_core a6d1:00:02.0: HCA minimum page size:512
> > > mlx4_core a6d1:00:02.0: Timestamping is not supported in slave mode
> > > mlx4_core a6d1:00:02.0: INTx is not supported in multi-function mode,
> > aborting
> > > PM: dpm_run_callback(): pci_pm_thaw+0x0/0xd7 returns -95
> > > PM: Device a6d1:00:02.0 failed to thaw: error -95
> > >
> > > To be more accurate, the "resume" phase means the "thaw" callbacks which
> > > run before the system enters hibernation: when the user runs the command
> > > "echo disk > /sys/power/state" for hibernation, first the kernel "freezes"
> > > all the devices and creates a hibernation image, then the kernel "thaws"
> > > the devices including the disk/NIC, writes the memory to the disk, and
> > > powers down. This patch fixes the error message for the Mellanox VF driver
> > > in this phase.
> > >
> > > When the system starts again, a fresh kernel starts to run, and when the
> > > kernel detects that a hibernation image was saved, the kernel "quiesces"
> > > the devices, and then "restores" the devices from the saved image. In this
> > > path:
> > > device_resume_noirq() -> ... ->
> > >    pci_pm_restore_noirq() ->
> > >      pci_pm_default_resume_early() ->
> > >        pci_power_up() moves the device states back to PCI_D0. This path is
> > > not broken and doesn't need my patch.
> > >
> > > Signed-off-by: Dexuan Cui <decui@xxxxxxxxxxxxx>
> > This looks like a bugfix for 5839ee7389e8 ("PCI / PM: Force devices to
> > D0 in pci_pm_thaw_noirq()") so maybe it should be marked for stable as
> > 5839ee7389e8 was?
> >
> > Rafael, could you confirm?

No, it is not a bug fix for that commit. The underlying issue would be 
there without that commit too.


> > > ---
> > >
> > > changes in v2:
> > > 	Updated the changelog with more details.
> > >
> > >   drivers/pci/pci-driver.c | 7 ++++---
> > >   1 file changed, 4 insertions(+), 3 deletions(-)
> > >
> > > diff --git a/drivers/pci/pci-driver.c b/drivers/pci/pci-driver.c
> > > index 36dbe960306b..27dfc68db9e7 100644
> > > --- a/drivers/pci/pci-driver.c
> > > +++ b/drivers/pci/pci-driver.c
> > > @@ -1074,15 +1074,16 @@ static int pci_pm_thaw_noirq(struct device
> > *dev)
> > >   			return error;
> > >   	}
> > >
> > > -	if (pci_has_legacy_pm_support(pci_dev))
> > > -		return pci_legacy_resume_early(dev);
> > > -
> > >   	/*
> > >   	 * pci_restore_state() requires the device to be in D0 (because of MSI
> > >   	 * restoration among other things), so force it into D0 in case the
> > >   	 * driver's "freeze" callbacks put it into a low-power state directly.
> > >   	 */
> > >   	pci_set_power_state(pci_dev, PCI_D0);
> > > +
> > > +	if (pci_has_legacy_pm_support(pci_dev))
> > > +		return pci_legacy_resume_early(dev);
> > > +
> > >   	pci_restore_state(pci_dev);
> > >
> > >   	if (drv && drv->pm && drv->pm->thaw_noirq)
> > > --
> > > 2.19.1
The patch looks reasonable to me, but the comment above the 
pci_set_power_state() call needs to be updated too IMO.