Re: [RFC] [PATCH] Execute PCI quirks on resume from suspend-to-RAM

[Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006@gmx.net>]

Dave Jones wrote:
> On Wed, May 10, 2006 at 11:36:41PM +0200, Carl-Daniel Hailfinger wrote:
>  > Greg KH wrote:
>  > > On Wed, May 10, 2006 at 12:30:34PM +0200, Carl-Daniel Hailfinger wrote:
>  > >> Thinking about it again, if we restored the full pci config space
>  > >> on resume, this quirk handling would be completely unnecessary.
>  > >> Any reasons why we don't do that?
>  > > 
>  > > We do do that.  Look at pci_restore_state().
>  > > 
>  > > Actually, look at it in the latest -mm tree, that version works better
>  > > than mainline does right now :)
>  > 
>  > Sorry. Even the version in -mm does not restore all 256 bytes, so it
>  > will not change anything.
> 
> You can't generically look at a PCI device past the first 32 bytes.
> *anything* could be there, including registers which cause the machine
> to lock up when they get read.
> 
> This is exactly the reason that lspci by default only shows 32 bytes,
> and you need to be root to see past that limit.

You mean 64 bytes?

>  > So either we really restore the full config space (probably a good idea
>  > by itself)
> 
> No, *really* *really* bad idea :)

I had hoped the warnings in the lspci man page would be obsolete now.
Wishful thinking, it appears. Thanks for the hint.


Unfortunately, that means we either have to introduce a new PCI_FIXUP_
type or we execute PCI_FIXUP_HEADER also on resume. Which is better?


Regards,
Carl-Daniel
-- 
http://www.hailfinger.org/