Re: [PATCH] implement pm_ops.valid for everybody

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

On Thursday 22 March 2007 12:26 pm, Tony Lindgren wrote:
> 
> In addition to offering wakeup events for individual devices,
> the device suspend states should be something like retention
> and suspend, where:

Maybe ... worth discussing.  Most PCI drivers don't make
that distinction, although they could (see below).

What they do instead is assume the lowest device power state
("suspend"), and re-initialize in resume().  If Linux starts to
support standby and STR modes properly ... then it'd make sense
to teach more PCI drivers to try using "retention" states.  But
those drivers would still need to be prepared to re-init.


> Retention is where clocks are off for a device, but power is on.
> In this case the device registers are maintained in hardware.

Analagous to PCI D1 or D2.

> Suspend is where clocks and power are off. In this state the
> device registers are maintained in software.

Analagous to PCI D3, especially D3cold ... although PCI D3
certainly allows the Vaux "power well" to power some parts
of the device, so that not all register values get reset.


> Laptops mostly have suspend, while socs allow both retention
> and suspend in many cases.

Not quite true, as noted above.  There are differences in how
things are factored, but those mechanisms exist in both x86
and SOC worlds.  One key difference from a Linux perspective
is probably that without ACPI in the way, a SOC design can
make much better use of the hardware PM capabilities.

Very few non-USB drivers address "retention" modes on laptops;
USB host controller drivers need it to handle "remote wakeup",
which one expects to work from "standby" and suspend-to-RAM.
(Plus potentialy suspend-to-disk, but that's uncommon.)

- Dave