Re: [linux-pm] [PATCH 2/2] PCI PM: Introduce pci_preferred_state

["Rafael J. Wysocki" <rjw@sisk.pl>]

On Wednesday, 7 of May 2008, Alan Stern wrote:
> On Wed, 7 May 2008, Rafael J. Wysocki wrote:
> 
> > On Wednesday, 7 of May 2008, Pavel Machek wrote:
> > > Hi!
> > > 
> > > > From: Rafael J. Wysocki <rjw@sisk.pl>
> > > > 
> > > > The new suspend and hibernation callbacks introduced with
> > > > 'struct pm_ops' and 'struct pm_ext_ops' do not take a
> > > > pm_message_t argument, so the drivers using them will not be able
> > > > to use pci_choose_state() in its present form.  For this reason,
> > > > introduce the new function pci_preferred_state() playing the role
> > > > of pci_choose_state(), but taking only a pointer to the device
> > > > object.
> > > > 
> > > > Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
> > > > ---
> > > >  drivers/pci/pci.c   |   33 ++++++++++++++++++++++++++++++++-
> > > >  include/linux/pci.h |    1 +
> > > >  include/linux/pm.h  |   10 ++++++++++
> > > >  3 files changed, 43 insertions(+), 1 deletion(-)
> > > > 
> > > > Index: linux-2.6/drivers/pci/pci.c
> > > > ===================================================================
> > > > --- linux-2.6.orig/drivers/pci/pci.c
> > > > +++ linux-2.6/drivers/pci/pci.c
> > > > @@ -509,7 +509,38 @@ pci_set_power_state(struct pci_dev *dev,
> > > >  }
> > > >  
> > > >  pci_power_t (*platform_pci_choose_state)(struct pci_dev *dev);
> > > > - 
> > > > +
> > > > +/**
> > > > + * pci_preferred_state - Choose the preferred power state of a PCI device
> > > > + * @dev: PCI device to be put into the low power state
> > > > + * @sp: Information aboutabout what the driver would prefer to do with
> > > > + *	the device if there were no platform-implemeted policy.
> > > > + *
> > > > + * Returns PCI power state suitable for given device and given suspend policy.
> > > > + * The policy, however, is only used if platform_pci_choose_state() fails or is
> > > > + * not present.  Otherwise, it is assumed that platform_pci_choose_state()
> > > > + * implements the right policy.
> > > > + */
> > > > +
> > > > +pci_power_t pci_preferred_state(struct pci_dev *dev, enum suspend_policy sp)
> > > > +{
> > > > +	pci_power_t ret;
> > > > +
> > > > +	if (!pci_find_capability(dev, PCI_CAP_ID_PM))
> > > > +		return PCI_D0;
> > > > +
> > > > +	ret = (sp == SP_TURN_OFF) ? PCI_D3hot : PCI_D0;
> > > > +	if (platform_pci_choose_state) {
> > > > +		pci_power_t platform_ret = platform_pci_choose_state(dev);
> > > > +
> > > > +		if (platform_ret != PCI_POWER_ERROR)
> > > > +			ret = platform_ret;
> > > > +	}
> > > > +	return ret;
> > > > +}
> > > > +
> > > > +EXPORT_SYMBOL(pci_preferred_state);
> > > 
> > > I don't get it. How is driver supposed to use this? How does the
> > > driver decide between SP_TURN_OFF and SP_TURN_ON?
> > > 
> > > ...and it seems to be clearer to just inline this in the driver... or
> > > pass  PCI_D3hot/PCI_D0 to it, instead of inventing yet another
> > > define...
> > 
> > I thought about that too.  I'd like to know what the other people think,
> > though.
> 
> The point of this isn't at all clear.
> 
> Is this routine meant to be called during a hibernation 
> transition?

Yes, it is.

> Or is it just for suspend? 
> 
> And why would the return value ever be anything other than D3_hot?  (Or 
> why would the driver ever want to put a device in a different state?)

In principle, the driver may want to put the device into a state having shorter
wake up latency than D3_hot.

> AFAICS, the only reason would be because platform_pci_choose_state() 
> suggested something else.  In which case there's no need for the 
> "policy" argument.

There is a need in two cases:
- if platform_pci_choose_state() is not defined (it only is defined for ACPI
  systems at the moment),
- if platform_pci_choose_state() returns PCI_POWER_ERROR meaning that it cannot
  handle the device.

I agree with Pavel that the driver could pass a "fallback state" as a second
argument to be used in case the platform cannot provide it with one.

Thanks,
Rafael