Re: [PATCH] xen: use the trigger info we already have to choose the irq handler

[Tom Goetz <tom.goetz@virtualcomputer.com>]

On May 25, 2011, at 10:25 AM, Konrad Rzeszutek Wilk wrote:

> On Wed, May 25, 2011 at 12:33:23PM +0100, Stefano Stabellini wrote:
>> Do not use pirq_needs_eoi to decide which irq handler to use because Xen
>> always returns true if the guest does not support pirq_eoi_map.
>> Use the trigger information we already have from MP-tables and ACPI.
>> 
>> Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
>> Reported-by: Thomas Goetz <tcgoetz@gmail.com>
>> Tested-by: Thomas Goetz <tcgoetz@gmail.com>
> 
> Tom,
> 
> Do you prefer those email addresses or tom.goetz@virtualcomputer.com ?

It doesn't matter to me. I'm signed up to Xen devel with the gmail address because I signed up before we had a domain for the company and never bothered to change it.


> ..
>> -	if (pirq_needs_eoi(irq))
>> +	if (shareable)
> 
> Ok, so for dom0, the sharable flag is determined by the ACPI parsing.. good.
> And for MSI/MSI-X we use xen_bind_pirq_msi_to_irq which sets it to use handle_edge_irq.
> 
> The only hazard which you asked me on IRC is for PV guests with Xen PCI front.
> In those cases, we have no idea whether the "GSI" IRQ is edge or level and we
> just assume that anything under 16 is edge. Which is 99% right, except if you
> are to pass in the ACPI IRQ to the guest, or some other weird creature. The MSI/MSI-X
> part of the PCI front are OK too - they use the xen_bind_pirq_msi_to_irq call.
> 
> So, putting on the todo list an extension to the PCI PV protocol to pass information
> about the GSI's - whether they are edge or level.
> 
> Patch is on the 2.6.40-rc1 train.