Re: A reference implementation of PCI suspend/resume?

[David Brownell <david-b@pacbell.net>]

[-- Attachment #1: Type: text/plain, Size: 1526 bytes --]

On Monday 16 May 2005 12:02 pm, Fabrice Gautier wrote:
> > 
> > There's PCI ... and PCI with PM capabilities.  That rule about only
> > issuing PME transactions while suspended can only apply in the latter
> > case.  That is, not all PCI devices support the PCI PM rules; drivers
> > may need to help the hardware.
> 
> You mean, you support some kind of power management on PCI devices without
> PM capabilities ? How are you even sure that the pci_set_power_state will
> work in those kind of devices.

Linux must certainly support suspend() and resume() methods for such
devices.  In terms of PCI PM terminology, they will either stay in
the PCI_D0 state during suspend, or go into PCI_D3cold ... and in
the latter case, the device driver's resume() is going to have to
know how to reinit from a reset state.  (They already have code to
init from that state, so it should be no problem.)

By definition, pci_set_power_state() will be a NOP.  There might
be firmware that knows how to switch power on/off, or maybe not;
that would be board-specific.


> In any case, anyway it works I would say this is very much device specific
> and not really part of a reference implementation. This is supposed to be a
> reference implementation for "PCI PM compliant" devices right ?

It was supposed to be a reference implementation, yes.  But if it was
only for one subset of PCI devices, I didn't notice any suggestion of
that before now.  The "works for small subset" implementation wouldn't
be anywhere near as useful.

- Dave




[-- Attachment #2: Type: text/plain, Size: 0 bytes --]