Re: [PATCH 3/3] irqchip/gic-v4: Make sure a VPE is locked when VMAPP is issued

[Marc Zyngier]

On Tue, 23 Jul 2024 02:51:32 +0100,
Zhou Wang <wangzhou1@xxxxxxxxxxxxx> wrote:
> 
> On 2024/7/19 19:31, Marc Zyngier wrote:
> > On Fri, 19 Jul 2024 10:42:02 +0100,
> > Zhou Wang <wangzhou1@xxxxxxxxxxxxx> wrote:
> >>
> >> On 2024/7/5 17:31, Marc Zyngier wrote:
> >>> In order to make sure that vpe->col_idx is correctly sampled
> >>> when a VMAPP command is issued, we must hold the lock for the
> >>> VPE. This is now possible since the introduction of the per-VM
> >>> vmapp_lock, which can be taken before vpe_lock in the locking
> >>> order.
> >>>
> >>> Signed-off-by: Marc Zyngier <maz@xxxxxxxxxx>
> >>> ---
> >>>  drivers/irqchip/irq-gic-v3-its.c | 8 ++++++--
> >>>  1 file changed, 6 insertions(+), 2 deletions(-)
> >>>
> >>> diff --git a/drivers/irqchip/irq-gic-v3-its.c b/drivers/irqchip/irq-gic-v3-its.c
> >>> index b52d60097cad5..951ec140bcea2 100644
> >>> --- a/drivers/irqchip/irq-gic-v3-its.c
> >>> +++ b/drivers/irqchip/irq-gic-v3-its.c
> >>> @@ -1810,7 +1810,9 @@ static void its_map_vm(struct its_node *its, struct its_vm *vm)
> >>>  		for (i = 0; i < vm->nr_vpes; i++) {
> >>>  			struct its_vpe *vpe = vm->vpes[i];
> >>>  
> >>> -			its_send_vmapp(its, vpe, true);
> >>> +			scoped_guard(raw_spinlock, &vpe->vpe_lock)
> >>> +				its_send_vmapp(its, vpe, true);
> >>> +
> >>>  			its_send_vinvall(its, vpe);
> >>>  		}
> >>>  	}
> >>> @@ -1827,8 +1829,10 @@ static void its_unmap_vm(struct its_node *its, struct its_vm *vm)
> >>>  	if (!--vm->vlpi_count[its->list_nr]) {
> >>>  		int i;
> >>>  
> >>> -		for (i = 0; i < vm->nr_vpes; i++)
> >>> +		for (i = 0; i < vm->nr_vpes; i++) {
> >>> +			guard(raw_spinlock)(&vm->vpes[i]->vpe_lock);
> >>>  			its_send_vmapp(its, vm->vpes[i], false);
> >>> +		}
> >>>  	}
> >>>  }
> >>>  
> >>
> >> Hi Marc,
> >>
> >> It looks like there is ABBA deadlock after applying this series:
> >>
> >> In its_map_vm: vmapp_lock -> vpe_lock
> >> In its_vpe_set_affinity: vpe_to_cpuid_lock(vpe_lock) -> its_send_vmovp(vmapp_lock)
> >>
> >> Any idea about this?
> > 
> > Hmmm, well spotted. That's an annoying one.
> > 
> > Can you give the below hack a go? I've only lightly tested it, as my
> > D05 box is on its last leg (it is literally falling apart) and I don't
> > have any other GICv4.x box to test on.
> > 
> > Thanks,
> > 
> > 	M.
> > 
> > diff --git a/drivers/irqchip/irq-gic-v3-its.c b/drivers/irqchip/irq-gic-v3-its.c
> > index 951ec140bcea2..b88c6011c8771 100644
> > --- a/drivers/irqchip/irq-gic-v3-its.c
> > +++ b/drivers/irqchip/irq-gic-v3-its.c
> > @@ -1328,12 +1328,6 @@ static void its_send_vmovp(struct its_vpe *vpe)
> >  		return;
> >  	}
> >  
> > -	/*
> > -	 * Protect against concurrent updates of the mapping state on
> > -	 * individual VMs.
> > -	 */
> > -	guard(raw_spinlock_irqsave)(&vpe->its_vm->vmapp_lock);
> > -
> >  	/*
> >  	 * Yet another marvel of the architecture. If using the
> >  	 * its_list "feature", we need to make sure that all ITSs
> > @@ -3808,7 +3802,7 @@ static int its_vpe_set_affinity(struct irq_data *d,
> >  	struct its_vpe *vpe = irq_data_get_irq_chip_data(d);
> >  	unsigned int from, cpu = nr_cpu_ids;
> >  	struct cpumask *table_mask;
> > -	unsigned long flags;
> > +	unsigned long flags, vmapp_flags;
> >  
> >  	/*
> >  	 * Changing affinity is mega expensive, so let's be as lazy as
> > @@ -3822,7 +3816,14 @@ static int its_vpe_set_affinity(struct irq_data *d,
> >  	 * protect us, and that we must ensure nobody samples vpe->col_idx
> >  	 * during the update, hence the lock below which must also be
> >  	 * taken on any vLPI handling path that evaluates vpe->col_idx.
> > +	 *
> > +	 * Finally, we must protect ourselves against concurrent
> > +	 * updates of the mapping state on this VM should the ITS list
> > +	 * be in use.
> >  	 */
> > +	if (its_list_map)
> > +		raw_spin_lock_irqsave(&vpe->its_vm->vmapp_lock, vmapp_flags);
> > +
> >  	from = vpe_to_cpuid_lock(vpe, &flags);
> >  	table_mask = gic_data_rdist_cpu(from)->vpe_table_mask;
> >  
> > @@ -3852,6 +3853,9 @@ static int its_vpe_set_affinity(struct irq_data *d,
> >  	irq_data_update_effective_affinity(d, cpumask_of(cpu));
> >  	vpe_to_cpuid_unlock(vpe, flags);
> >  
> > +	if (its_list_map)
> > +		raw_spin_unlock_irqrestore(&vpe->its_vm->vmapp_lock, vmapp_flags);
> > +
> >  	return IRQ_SET_MASK_OK_DONE;
> >  }
> >
> 
> Hi Marc，
> 
> We add above code to do test again. Now it is OK.

Great, thanks for giving it a go. I have just posted an actual patch
(with the exact same change) at [1]. It would be good if you could
give it a Tested-by: tag.

Thanks,

	M.

[1] https://lore.kernel.org/r/20240723175203.3193882-1-maz@xxxxxxxxxx

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