Resume only part of device tree?

Resume only part of device tree?

[Nigel Cunningham]

Hi all.

Is there any existing code in the device model that supports resuming a
part of the device tree? For Suspend2, I'm wanting to resume storage
devices (and their parents) part way through resuming, and other drivers
later.

Regards,

Nigel

Re: Resume only part of device tree?

[Benjamin Herrenschmidt]

On Thu, 2004-03-04 at 07:03, Nigel Cunningham wrote:
> Hi all.
> 
> Is there any existing code in the device model that supports resuming a
> part of the device tree? For Suspend2, I'm wanting to resume storage
> devices (and their parents) part way through resuming, and other drivers
> later.

What is your exact goal ? Not resuming all devices when writing the
state to the swap partition ?

You really need to resume it all at this point. However, the optimisation
that can be done is for some drivers to not put the HW to sleep on a
swsusp transition, only freeze the state. I did something like that in
IDE though that doesn't always work well due our "state" paremeter passed
down to drivers not beeing consistent.

Ben.

Re: Resume only part of device tree?

[Nigel Cunningham]

Hi.

My implementation saved the image in two parts. 'Pageset 2' contains the
LRU pages (active & inactive lists). 'Pageset 1' contains all other data
to be saved. At resume time, I read pageset 1 and copy the original
kernel data back. Then I want to resume the storage devices and read
pageset 2 before resuming all devices and waking everything else up. It
would also be good to not resume all devices when writing the state to
the swap partition, but I have other means of ensuring the consistency
of the image that mean I'm not so worried then.

Regards,

Nigel

On Thu, 2004-03-04 at 11:40, Benjamin Herrenschmidt wrote:
> On Thu, 2004-03-04 at 07:03, Nigel Cunningham wrote:
> > Hi all.
> > 
> > Is there any existing code in the device model that supports resuming a
> > part of the device tree? For Suspend2, I'm wanting to resume storage
> > devices (and their parents) part way through resuming, and other drivers
> > later.
> 
> What is your exact goal ? Not resuming all devices when writing the
> state to the swap partition ?
> 
> You really need to resume it all at this point. However, the optimisation
> that can be done is for some drivers to not put the HW to sleep on a
> swsusp transition, only freeze the state. I did something like that in
> IDE though that doesn't always work well due our "state" paremeter passed
> down to drivers not beeing consistent.
> 
> Ben.
> 
-- 

Re: Resume only part of device tree?

[Nigel Cunningham]

Okay. I'll write it then. Thanks.

Nigel

On Thu, 2004-03-04 at 16:57, Benjamin Herrenschmidt wrote:
> On Thu, 2004-03-04 at 11:48, Nigel Cunningham wrote:
> > Hi.
> > 
> > My implementation saved the image in two parts. 'Pageset 2' contains the
> > LRU pages (active & inactive lists). 'Pageset 1' contains all other data
> > to be saved. At resume time, I read pageset 1 and copy the original
> > kernel data back. Then I want to resume the storage devices and read
> > pageset 2 before resuming all devices and waking everything else up. It
> > would also be good to not resume all devices when writing the state to
> > the swap partition, but I have other means of ensuring the consistency
> > of the image that mean I'm not so worried then.
> 
> I don't see any good way to do that at this point. With a tree
> structure, it would be possible to revive only the parents of
> the storage device you are concerned in, but at this point, we
> don't have this possibility
> 
> Ben.
> 
-- 

Re: Resume only part of device tree?

[Benjamin Herrenschmidt]

On Thu, 2004-03-04 at 11:48, Nigel Cunningham wrote:
> Hi.
> 
> My implementation saved the image in two parts. 'Pageset 2' contains the
> LRU pages (active & inactive lists). 'Pageset 1' contains all other data
> to be saved. At resume time, I read pageset 1 and copy the original
> kernel data back. Then I want to resume the storage devices and read
> pageset 2 before resuming all devices and waking everything else up. It
> would also be good to not resume all devices when writing the state to
> the swap partition, but I have other means of ensuring the consistency
> of the image that mean I'm not so worried then.

I don't see any good way to do that at this point. With a tree
structure, it would be possible to revive only the parents of
the storage device you are concerned in, but at this point, we
don't have this possibility

Ben.