Re: SRAT v1 support

Re: SRAT v1 support

[Yinghai Lu]

On Fri, May 15, 2009 at 7:51 AM, xb <xavier.bru@bull.net> wrote:
> Recent linux kernels suppose that the  SRAT table is in rev 2 format (ACPI
> 3.0), but some BIOSes still provide SRAT table in rev 1.
> The rev 2 of the SRAT extension mainly provides an extension of the
> "proximity_domain" item from 8 bits to 32 bits, using a "reserved" field of
> the structure.
> When the "reserved" field is not null, linux finds a wrong proximity domain,
> and numa initialization is wrong.
> Following patch tests the SRAT revision to allow a correct initialization:
>
> This patch tests the version of SRAT ACPI table to allow supporting SRAT rev
> 1 and SRAT rev 2.
>
> diff -Nru linux-2.6.29-rc7-orig/arch/x86/kernel/acpi/boot.c
> linux-2.6.29-rc7-tmp/arch/x86/kernel/acpi/boot.c
> --- linux-2.6.29-rc7-orig/arch/x86/kernel/acpi/boot.c    2009-03-12
> 14:41:38.000000000 +0100
> +++ linux-2.6.29-rc7-tmp/arch/x86/kernel/acpi/boot.c    2009-03-16
> 14:52:07.000000000 +0100
> @@ -260,7 +260,8 @@
> }
>
> static int __init
> -acpi_parse_lapic(struct acpi_subtable_header * header, const unsigned long
> end)
> +acpi_parse_lapic(const struct acpi_subtable_header * const header,
> +                const unsigned long end, const int rev)

do we need to pass that rev all over around?

looks like we could use one srat_version or pxm_mask variable to get
the same result.

YH

Re: SRAT v1 support

[xb]

yes, could be simpler :-)

    I just saw that Kurt Garloff from Suse has published 3 patches on 
kernel mailing list that fixes the SRAT v1 issue:
    [PATCH 0/3]: Discard reserved PXM bits for SRAT v1
    [PATCH 1/3]: Store SRAT revision
    [PATCH 2/3]: x86-64: Handle SRAT v1 and v2 consistently

    It looks OK for us.

Thanks.
Xavier


Yinghai Lu wrote:
> On Fri, May 15, 2009 at 7:51 AM, xb <xavier.bru@bull.net> wrote:
>   
>> Recent linux kernels suppose that the  SRAT table is in rev 2 format (ACPI
>> 3.0), but some BIOSes still provide SRAT table in rev 1.
>> The rev 2 of the SRAT extension mainly provides an extension of the
>> "proximity_domain" item from 8 bits to 32 bits, using a "reserved" field of
>> the structure.
>> When the "reserved" field is not null, linux finds a wrong proximity domain,
>> and numa initialization is wrong.
>> Following patch tests the SRAT revision to allow a correct initialization:
>>
>> This patch tests the version of SRAT ACPI table to allow supporting SRAT rev
>> 1 and SRAT rev 2.
>>
>> diff -Nru linux-2.6.29-rc7-orig/arch/x86/kernel/acpi/boot.c
>> linux-2.6.29-rc7-tmp/arch/x86/kernel/acpi/boot.c
>> --- linux-2.6.29-rc7-orig/arch/x86/kernel/acpi/boot.c    2009-03-12
>> 14:41:38.000000000 +0100
>> +++ linux-2.6.29-rc7-tmp/arch/x86/kernel/acpi/boot.c    2009-03-16
>> 14:52:07.000000000 +0100
>> @@ -260,7 +260,8 @@
>> }
>>
>> static int __init
>> -acpi_parse_lapic(struct acpi_subtable_header * header, const unsigned long
>> end)
>> +acpi_parse_lapic(const struct acpi_subtable_header * const header,
>> +                const unsigned long end, const int rev)
>>     
>
> do we need to pass that rev all over around?
>
> looks like we could use one srat_version or pxm_mask variable to get
> the same result.
>
> YH
> --
> To unsubscribe from this list: send the line "unsubscribe linux-acpi" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>
>
>