Re: Fwd: [CFT] Clean up yenta_socket

Re: Fwd: [CFT] Clean up yenta_socket

[Jeff Garzik]

WIBNI?

Anyway, MSI needs more than the standard size as well.

But I would actually prefer the interface to go the other way:

	pci_save_state(pdev);
		and
	pci_restore_state(pdev);

Allocate and store the state in a pointer in struct pci_dev, or 
somesuch.  And somebody other than the low-level driver figures out the 
amount to save and restore.

	Jeff

Re: Fwd: [CFT] Clean up yenta_socket

[Russell King]

On Mon, Aug 25, 2003 at 06:41:26PM -0400, Jeff Garzik wrote:
> WIBNI?

"wouldn't it be nice if"

> Anyway, MSI needs more than the standard size as well.
> 
> But I would actually prefer the interface to go the other way:
> 
> 	pci_save_state(pdev);
> 		and
> 	pci_restore_state(pdev);
> 
> Allocate and store the state in a pointer in struct pci_dev, or 
> somesuch.  And somebody other than the low-level driver figures out the 
> amount to save and restore.

Hmm.  The reason I wanted to stear clear of that was that sometimes we
don't know what's there.  Taking the yenta as an example, we know that
the "standard" space is 0x48 bytes long.  However, some devices have
extra control registers at 0x80, and then there's the PCI PM registers
up at around 0xa0 or so.

On a different cardbus bridge, the PCI PM registers might be somewhere
else.

Do we care if we overwrite the PCI PM registers with possibly old/stale
data?

In other words, how does the PCI layer itself know how much configuration
space to save and restore over power management calls?

-- 
Russell King (rmk@arm.linux.org.uk)                The developer of ARM Linux
             http://www.arm.linux.org.uk/personal/aboutme.html