Re: [PATCH 2/2] msi: Add support for the Intel chipsets that support MSI.

[ebiederm@xmission.com (Eric W. Biederman)]

Andi Kleen <ak@suse.de> writes:

> On Friday 25 May 2007 06:26:50 Eric W. Biederman wrote:
>> 
>> This patch is the result of a quick survey of the Intel chipset
>> documents.  I took a quick look in the document to see if the chipset
>> supported MSI and if so I looked through to find the vendor and device
>> id of device 0 function 0 of the chipset and added a quirk for that
>> device id if I it was not a duplicate. 
>
> It would be better to look for any PCI bridge. Sometimes there are
> different PCI bridges around (e.g. external PCI-X bridges on HT systems) 
> which might need own quirks

Maybe but pci-pci bridges should have no problems with MSI traffic
because at that level MSI is just a normal DMA write going upstream.

The AMD 8131 hypertransport to PCI-X bridge only failed because at the
connection to hypertransport it did not implement the hypertransport
msi mapping capability and the associated logic to convert an MSI
irq into a hypertranposrt IRQ.

I currently have a quirk that looks for any hypertransport msi mapping
capability and only enables MSI on the downstream bus of the bridge.

> Also in the x86 world Microsoft defined a FADT ACPI flag that MSI doesn't
> work for Vista. It might make sense to do
>
> if (dmi year >= 2007 && FADT.msi_disable not set) assume it works

Sounds reasonable to me.  If it happens to work reliably.

>> This patch should be safe.  The anecdotal evidence is that when dealing
>> with MSI the Intel chipsets just work.  If we find some buggy ones
>> changing the list won't be hard.
>
> The FADT bit should be probably checked anyways.

Sure if we have a way to check I have no problem, although I tend to
trust the hardware more.

Eric