Re: [PATCHv3] qemu-kvm: enable msi with irqchip

From: Avi Kivity <avi@redhat.com>
To: "Michael S. Tsirkin" <mst@redhat.com>
Cc: kvm@vger.kernel.org
Subject: Re: [PATCHv3] qemu-kvm: enable msi with irqchip
Date: Tue, 07 Jul 2009 14:21:39 +0300	[thread overview]
Message-ID: <4A532FC3.6030103@redhat.com> (raw)
In-Reply-To: <20090705182633.GA3073@redhat.com>

On 07/05/2009 09:26 PM, Michael S. Tsirkin wrote:
> On Sun, Jul 05, 2009 at 09:03:00PM +0300, Avi Kivity wrote:
>    
>> On 07/05/2009 08:35 PM, Michael S. Tsirkin wrote:
>>      
>>> Support msi-x with irqchip in kernel: allocate entries
>>> when they are used, and update when they are unmasked.
>>>
>>> @@ -340,6 +447,10 @@ void msix_notify(PCIDevice *dev, unsigned vector)
>>>            msix_set_pending(dev, vector);
>>>            return;
>>>        }
>>> +    if (kvm_enabled()&&   qemu_kvm_irqchip_in_kernel()) {
>>> +        kvm_set_irq(dev->msix_irq_entries[vector].gsi, 1, NULL);
>>>
>>>        
>> Toggle back to zero after setting to one, for consistency.
>>      
>
> You don't really want that.  It's an extra system call, on a data path
> operation.
>    

Yes I do.  I don't want 'level' to mean one thing for msi and another 
for other interrupts.

If the extra system call ends up hurting, we can add KVM_TOGGLE_IRQ or 
define a new level that means 'toggle'.

-- 
error compiling committee.c: too many arguments to function