[+cc Yinghai, sorry I didn't think of it before] On Wed, Mar 5, 2014 at 11:30 PM, Suravee Suthikulpanit <suravee.suthikulpanit@xxxxxxx> wrote: > On 3/5/2014 8:13 PM, Suravee Suthikulanit wrote: >> >> On 3/5/2014 3:24 PM, Bjorn Helgaas wrote: >>> >>> [+cc linux-acpi] >>> >>> On Wed, Mar 5, 2014 at 2:06 PM, <suravee.suthikulpanit@xxxxxxx> wrote: >>>> >>>> From: Suravee Suthikulpanit <suravee.suthikulpanit@xxxxxxx> >>>> >>>> The current code only supports upto AMD hostbridge for family11h. >>>> This causes PCI numa_node information to be reported incorrectly >>>> for newer family with multi sockets. >>> >>> >>> Where is the incorrect reporting? In ACPI tables? Is this patch a >>> way to cover up firmware defects in the ACPI description? Or is this >>> for machines without ACPI (it seems unlikely that machines with new >>> AMD processors would not have ACPI)? >> >> >> This is incorrectly reported in the sysfs for each PCI device (e.g. >> /devices/pci0000:50/0000:50:00.2/numa_node). Without the patch, they >> return -1. >> >> In file arch/x86/pci/acpi.c, in function pci_acpi_scan_root(), it is >> queries the node information as following: >> >> #ifdef CONFIG_ACPI_NUMA >> pxm = acpi_get_pxm(device->handle); >> if (pxm >= 0) >> node = pxm_to_node(pxm); >> if (node != -1) >> set_mp_bus_to_node(busnum, node); >> else >> #endif >> node = get_mp_bus_to_node(busnum); >> >> In this case, I see that the acpi_get_pxm() returns -1. Therefore, it >> falls back to using the node information in mp_bus_to_node[]. So, >> without this patch, it would also returning -1. >> >> Also, the spec mentioned that the _PXM is optional, so I am not sure if >> this is a firmware bug. > > I am not quite familiar with the ACPI for this part. However, after taking > a look at the code (in driver/acpi/pci_root.c: acpi_pci_root_add()), I > believe it's trying to locate _PXM method in the DSDT table, in which I > don't see any _PXM methods. This sure looks like a firmware bug. True, _PXM is optional, but if the firmware doesn't provide it, nobody should be surprised that the OS thinks everything is in the same proximity domain. I would not endorse extending amd_bus.c for new CPUs. That just covers up firmware problems like this, and if you ever run a different OS on the box, you'll trip over them again. And I don't think a patch like this will even be a possibility for Windows. Bjorn -- To unsubscribe from this list: send the line "unsubscribe linux-pci" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html