Re: [PATCH] Detect mmconfig on nVidia MCP55

[Ingo Molnar <mingo@elte.hu>]

* Yinghai Lu <yinghai@kernel.org> wrote:

> On Wed, Feb 4, 2009 at 8:39 AM, Ed Swierk <eswierk@aristanetworks.com> wrote:
> > Detect and enable memory-mapped PCI configuration space on the nVidia
> > MCP55 southbridge even if the ACPI MCFG table is missing or wrong.
> >
> > Signed-off-by: Ed Swierk <eswierk@aristanetworks.com>
> >
> > ---
> > Index: linux-2.6.27.4/arch/x86/pci/mmconfig-shared.c
> > ===================================================================
> > --- linux-2.6.27.4.orig/arch/x86/pci/mmconfig-shared.c
> > +++ linux-2.6.27.4/arch/x86/pci/mmconfig-shared.c
> > @@ -155,6 +155,26 @@ static const char __init *pci_mmcfg_amd_
> >        return "AMD Family 10h NB";
> >  }
> >
> > +static const char __init *pci_mmcfg_nvidia_mcp55(void)
> > +{
> > +       u32 extcfg;
> > +
> > +       raw_pci_ops->read(0, 0, PCI_DEVFN(0, 0), 0x90, 4, &extcfg);
> > +
> > +       if (!(extcfg & 0x80000000))
> > +               return NULL;
> > +       pci_mmcfg_config = kzalloc(sizeof(pci_mmcfg_config[0]), GFP_KERNEL);
> > +       if (!pci_mmcfg_config)
> > +               return NULL;
> > +       pci_mmcfg_config[0].address = (extcfg & 0x00007fff) << 25;
> > +       pci_mmcfg_config[0].pci_segment = 0;
> > +       pci_mmcfg_config[0].start_bus_number = 0;
> > +       pci_mmcfg_config[0].end_bus_number = (1 << (8 - ((extcfg >> 28) & 3))) - 1;
> > +       pci_mmcfg_config_num = 1;
> > +
> > +       return "nVidia MCP55";
> > +}
> > +
> >  struct pci_mmcfg_hostbridge_probe {
> >        u32 bus;
> >        u32 devfn;
> > @@ -172,6 +192,8 @@ static struct pci_mmcfg_hostbridge_probe
> >          0x1200, pci_mmcfg_amd_fam10h },
> >        { 0xff, PCI_DEVFN(0, 0), PCI_VENDOR_ID_AMD,
> >          0x1200, pci_mmcfg_amd_fam10h },
> > +       { 0, PCI_DEVFN(0, 0), PCI_VENDOR_ID_NVIDIA,
> > +         0x0369, pci_mmcfg_nvidia_mcp55 },
> 
> may break amd family 10h + mcp55 system. because some system have setting 
> in AMD cpu and mcp55 it will find setting in CPU nb, and your code will 
> partially overwrite those setting.

That's a good point ...

We could do something like:

	if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
		boot_cpu_data.x86 >= 0x10) {
		u64 mmconf_val;

		rdmsrl(MSR_FAM10H_MMIO_CONF_BASE, mmconf_val);
		if (mmconf_val & FAM10H_MMIO_CONF_ENABLE)
			return;
	}

To detect enabled on-CPU mmconf support. We could even put this into a 
helper function as other places might want to use it too.

	Ingo