Re: pci_enable_msi() for everyone?

[Luben Tuikov <luben_tuikov@adaptec.com>]

On 06/07/05 01:46, Adam Belay wrote:
> On Mon, Jun 06, 2005 at 10:22:03PM -0700, Greg KH wrote:
> 
>>On Mon, Jun 06, 2005 at 05:26:32PM -0700, Roland Dreier wrote:
>>
>>>    davej> What if MSI support has been disabled in the bridge due to
>>>    davej> some quirk (like the recent AMD 8111 quirk) ?  Maybe the
>>>    davej> above function should check pci_msi_enable as well ?
>>>
>>>    Greg> Yes, you are correct.  I said it wasn't tested :)
>>>
>>>Huh?  If a host bridge doesn't support MSI, and a device below it has
>>>its MSI capability enabled, we're in big trouble.  Because that device
>>>is going to send interrupt messages whether the bridge likes it or
>>>not.
>>
>>No, that device would never get MSI enabled on it.  See the patch I
>>posted to make sure I didn't get it wrong...
>>
>>thanks,
>>
>>greg k-h
> 
> 
> How are we handling the case where a device has multiple MSI messages.
> Is any driver interaction needed for that?  Will this change affect it?
> I haven't had a chance to look through the MSI code yet.
> 
> Thanks,
> Adam

Yes, this is a very good point (PCI MSI-X).

All in all, given all the hardware quirks of both
PCI bridges and PCI devices, I'd leave PCI MSI
control to the PCI LLDD.

	Luben