Re: [PATCH] pci-mmconfig fix for 2.6.9

[Grant Grundler <grundler@parisc-linux.org>]

On Sun, Nov 14, 2004 at 11:07:13PM -0800, Michael Chan wrote:
> Grant Grundler wrote:
>  
> > There are two tests in arch/i386/pci/mmconfig.c:pci_mmcfg_init() before
> > the raw_pci_ops is set to &pci_mmcfg. Perhaps some additional crude tests
> > could select a different set of pci_raw_ops to deal with posted writes
> > to mmconfig space. Someone more familiar with those chipsets might
> > find a more elegant solution.
> 
> Do you mean something like pci_mmcfg1 for Intel chipsets that implement
> non-posted mmconfig and pci_mmcfg2 for other chipsets that may implement
> posted mmconfig?

Yes.

> pci_mmcfg2's write method will guarantee that the write has reached
> the target before returning. If pci_mmcfg2's write method uses read
> from the target to flush the write, we are back to the original problem
> of out-of-spec read when writing the PMCSR register. If the flush does
> not require reading from the target device, then it's fine.

Yes - agreed.

I do expect chip designers are aware of this problem.
It's not a new problem.
But it's certainly possible some are not. :^(

thanks,
grant

>  
> Michael
> 
> 
>