Re: [PATCH] x86/pci: claim SR-IOV bar in pcibios_allocate_resource

[Jesse Barnes <jbarnes@virtuousgeek.org>]

On Tue, 24 Nov 2009 13:21:45 -0800
Yinghai Lu <yinghai@kernel.org> wrote:

> Jesse Barnes wrote:
> > On Sat, 14 Nov 2009 23:46:46 -0800
> > Yinghai Lu <yinghai@kernel.org> wrote:
> > 
> >> so use correct allocation from BIOS, instead of later assign
> >> another one.
> >>
> >> Signed-off-by: Yinghai Lu <yinghai@kernel.org>
> >>
> >> ---
> >>  arch/x86/pci/i386.c |    4 +++-
> >>  1 file changed, 3 insertions(+), 1 deletion(-)
> >>
> >> Index: linux-2.6/arch/x86/pci/i386.c
> >> ===================================================================
> >> --- linux-2.6.orig/arch/x86/pci/i386.c
> >> +++ linux-2.6/arch/x86/pci/i386.c
> >> @@ -155,7 +155,9 @@ static void __init pcibios_allocate_reso
> >>  
> >>  	for_each_pci_dev(dev) {
> >>  		pci_read_config_word(dev, PCI_COMMAND, &command);
> >> -		for (idx = 0; idx < PCI_ROM_RESOURCE; idx++) {
> >> +		for (idx = 0; idx < PCI_BRIDGE_RESOURCES; idx++) {
> >> +			if (idx == PCI_ROM_RESOURCE)
> >> +				continue;
> >>  			r = &dev->resource[idx];
> >>  			if (r->parent)		/* Already
> >> allocated */ continue;
> > 
> > I'm worried this might have side effects beyond just allocating
> > SR-IOV BARs, since it looks like we'll walk through all the bridge
> > resources all the time?
> 
> /*
>  *  For PCI devices, the region numbers are assigned this way:
>  */
> enum {
>         /* #0-5: standard PCI resources */
>         PCI_STD_RESOURCES,
>         PCI_STD_RESOURCE_END = 5,
> 
>         /* #6: expansion ROM resource */
>         PCI_ROM_RESOURCE,
> 
>         /* device specific resources */
> #ifdef CONFIG_PCI_IOV
>         PCI_IOV_RESOURCES,
>         PCI_IOV_RESOURCE_END = PCI_IOV_RESOURCES + PCI_SRIOV_NUM_BARS
> - 1, #endif
> 
>         /* resources assigned to buses behind the bridge */
> #define PCI_BRIDGE_RESOURCE_NUM 4
> 
>         PCI_BRIDGE_RESOURCES,
> 
> so will only add IOV BAR related.

Oh right, *up to* bridge resources...  Still we should probably add a
comment about that.  SR-IOV is still somewhat new.

-- 
Jesse Barnes, Intel Open Source Technology Center