Re: [PATCH v3 01/10] KVM: arm64: vgic: Store LPIs in an xarray

[Marc Zyngier <maz@kernel.org>]

On Tue, 20 Feb 2024 16:30:24 +0000,
Zenghui Yu <zenghui.yu@linux.dev> wrote:
> 
> On 2024/2/17 02:41, Oliver Upton wrote:
> > Using a linked-list for LPIs is less than ideal as it of course requires
> > iterative searches to find a particular entry. An xarray is a better
> > data structure for this use case, as it provides faster searches and can
> > still handle a potentially sparse range of INTID allocations.
> > 
> > Start by storing LPIs in an xarray, punting usage of the xarray to a
> > subsequent change.
> > 
> > Signed-off-by: Oliver Upton <oliver.upton@linux.dev>
> 
> [..]
> 
> > diff --git a/arch/arm64/kvm/vgic/vgic.c b/arch/arm64/kvm/vgic/vgic.c
> > index db2a95762b1b..c126014f8395 100644
> > --- a/arch/arm64/kvm/vgic/vgic.c
> > +++ b/arch/arm64/kvm/vgic/vgic.c
> > @@ -131,6 +131,7 @@ void __vgic_put_lpi_locked(struct kvm *kvm, struct vgic_irq *irq)
> >  		return;
> >   	list_del(&irq->lpi_list);
> > +	xa_erase(&dist->lpi_xa, irq->intid);
> 
> We can get here *after* grabbing the vgic_cpu->ap_list_lock (e.g.,
> vgic_flush_pending_lpis()/vgic_put_irq()).  And as according to vGIC's
> "Locking order", we should disable interrupts before taking the xa_lock
> in xa_erase() and we would otherwise see bad things like deadlock..
> 
> It's not a problem before patch #10, where we drop the lpi_list_lock and
> start taking the xa_lock with interrupts enabled.  Consider switching to
> use xa_erase_irq() instead?

But does it actually work? xa_erase_irq() uses spin_lock_irq(),
followed by spin_unlock_irq(). So if we were already in interrupt
context, we would end-up reenabling interrupts. At least, this should
be the irqsave version.

The question is whether we manipulate LPIs (in the get/put sense) on
the back of an interrupt handler (like we do for the timer). It isn't
obvious to me that it is the case, but I haven't spent much time
staring at this code recently.

Thanks,

	M.

-- 
Without deviation from the norm, progress is not possible.