[PATCH 1/7] PCI: Restore BARs on runtime resume despite being unbound

[Rafael J. Wysocki]

On Tue, Feb 20, 2018 at 10:29 PM, Bjorn Helgaas <helgaas at kernel.org> wrote:
> On Sun, Feb 18, 2018 at 09:38:32AM +0100, Lukas Wunner wrote:
>> PCI devices not bound to a driver are supposed to stay in D0 during
>> runtime suspend.
>
> Doesn't "runtime suspend" mean an individual device is suspended while
> the rest of the system remains active?
>
> If so, maybe it would be more direct to say "PCI devices not bound to
> a driver cannot be runtime suspended"?
>
> (It's a separate question not relevant to this patch, but I'm not
> convinced that "PCI devices without a driver cannot be suspended"
> should be accepted as a rule.  If it is a rule, we should be able to
> deduce it from the specs.)
>
>> But they may have a parent which is bound and can be
>> transitioned to D3cold at runtime.  Once the parent goes to D3cold, the
>> unbound child may go to D3cold as well.  When the child comes out of
>> D3cold, its BARs are uninitialized and thus inaccessible when a driver
>> tries to probe.
>>
>> One example are recent hybrid graphics laptops which cut power to the
>> discrete GPU when the root port above it goes to ACPI power state D3.
>> Users may provoke this by unbinding the GPU driver and allowing runtime
>> PM on the GPU via sysfs:  The PM core will then treat the GPU as
>> "suspended", which in turn allows the root port to runtime suspend,
>> causing the power resources listed in its _PR3 object to be powered off.
>> The GPU's BARs will be uninitialized when a driver later probes it.
>>
>> Another example are hybrid graphics laptops where the GPU itself (rather
>> than the root port) is capable of runtime suspending to D3cold.  If the
>> GPU's integrated HDA controller is not bound and the GPU's driver
>> decides to runtime suspend to D3cold, the HDA controller's BARs will be
>> uninitialized when a driver later probes it.
>>
>> Fix by restoring the BARs on runtime resume if the device is not bound.
>> This is sufficient to fix the above-mentioned use cases.  Other use
>> cases might require a full-blown pci_save_state() / pci_restore_state()
>> or execution of fixups.  We can add that once use cases materialize,
>> let's not inflate the code unnecessarily.
>
> Why would we not do a full-blown restore?  With this patch, I think
> some configuration done during enumeration, e.g., ASPM and MPS, will
> be lost.  "lspci -vv" of the HDA before and after the suspend may show
> different things, which seems counter-intuitive.

Right.

> I wouldn't think of a full-blown restore as "inflating the code"; I
> would think of it as "having fewer special cases", i.e., we always use
> the same restore path instead of having the main one plus a special
> one for unbound devices.

That is a fair point, but if you look at pci_pm_runtime_suspend(), it
doesn't actually save anything for devices without drivers, so we'd
just restore the original initial state for them every time.

If we are to restore the entire state on runtime resume,
pci_pm_runtime_suspend() should be changed to call pci_save_state()
before returning 0 in the !dev->driver case AFAICS.

>> Cc: Bjorn Helgaas <bhelgaas at google.com>
>> Cc: Rafael J. Wysocki <rafael.j.wysocki at intel.com>
>> Signed-off-by: Lukas Wunner <lukas at wunner.de>
>> ---
>>  drivers/pci/pci-driver.c | 8 ++++++--
>>  drivers/pci/pci.c        | 2 +-
>>  drivers/pci/pci.h        | 1 +
>>  3 files changed, 8 insertions(+), 3 deletions(-)
>>
>> diff --git a/drivers/pci/pci-driver.c b/drivers/pci/pci-driver.c
>> index 3bed6beda051..51b11cbd48f6 100644
>> --- a/drivers/pci/pci-driver.c
>> +++ b/drivers/pci/pci-driver.c
>> @@ -1277,10 +1277,14 @@ static int pci_pm_runtime_resume(struct device *dev)
>>
>>       /*
>>        * If pci_dev->driver is not set (unbound), the device should
>> -      * always remain in D0 regardless of the runtime PM status
>> +      * always remain in D0 regardless of the runtime PM status.
>> +      * But if its parent can go to D3cold, this device may have
>> +      * been in D3cold as well and require restoration of its BARs.
>>        */
>> -     if (!pci_dev->driver)
>> +     if (!pci_dev->driver) {
>> +             pci_restore_bars(pci_dev);
>
> If we do decide not to do a full-blown restore, how do we decide
> whether to use pci_restore_bars() or pci_restore_config_space()?
>
> I'm not sure why we have both.

Me neither.

> The pci_restore_bars() path looks a
> little smarter in that it is more careful when updating 64-bit BARs
> that can't be updated atomically.
>
>>               return 0;
>> +     }
>>
>>       if (!pm || !pm->runtime_resume)
>>               return -ENOSYS;

So if pci_pm_runtime_suspend() is modified to call pci_save_state()
before returning 0 in the !dev->driver case, we can just move the
pci_restore_standard_config() invocation in pci_pm_runtime_resume() up
to the very top and check dev->driver later.