Re: [PATCH v3 1/3] Introduce a workqueue to deliver PIT timer interrupts.

[Gleb Natapov <gleb@redhat.com>]

On Tue, Apr 17, 2012 at 01:23:52PM +0300, Avi Kivity wrote:
> On 04/17/2012 12:31 PM, Gleb Natapov wrote:
> > On Mon, Apr 16, 2012 at 06:33:26PM +0200, Jan Kiszka wrote:
> > > On 2010-06-16 23:11, Chris Lalancette wrote:
> > > > We really want to "kvm_set_irq" during the hrtimer callback,
> > > > but that is risky because that is during interrupt context.
> > > > Instead, offload the work to a workqueue, which is a bit safer
> > > > and should provide most of the same functionality.
> > > 
> > > Unfortunately, workqueues do not have fixed kthread associations (and
> > > "kvm-pit-wq" would be too unspecific when running multiple VMs). So I
> > > just realized that this subtly breaks the ability to run KVM guests with
> > > RT priority (boot managers with timeouts hang as the workqueue starves).
> > > 
> > > Before throwing some kthread_worker at this, could someone help me
> > > recalling what was "risky" here? That the PIT IRQ may have to be
> > > broadcast to a large number of VCPUs? I would offer to include this
> > > information in my changelog. ;)
> > > 
> > ioapic and pic irq_set functions uses spinlocks, but not irqsave
> > variants. Also I am always forget whether is is safe to kick vcpu from
> > irq context.
> 
> It isn't, since you need to send an IPI.
> 
That is exactly what I forget whether you can send IPI from there :)
Anyway this is another reason.

--
			Gleb.