Re: [PATCH v4 15/22] KVM: arm64: ITS: Sort the device and ITE lists

[Marc Zyngier <marc.zyngier@arm.com>]

On Mon, Mar 27 2017 at 10:31:05 AM, Eric Auger <eric.auger@redhat.com> wrote:
> Natively sort the device and ITE lists in ascending
> deviceId/eventid order. This paves the way to optimized
> DTE and ITE scan in guest RAM table where entries are chained
> together using a next ID offset.
>
> Signed-off-by: Eric Auger <eric.auger@redhat.com>
> Reviewed-by: Andre Przywara <andre.przywara@arm.com>
>
> ---
>
> v3 -> v4:
> - added Andre's R-b
> ---
>  virt/kvm/arm/vgic/vgic-its.c | 36 ++++++++++++++++++++++++++++++++++--
>  1 file changed, 34 insertions(+), 2 deletions(-)
>
> diff --git a/virt/kvm/arm/vgic/vgic-its.c b/virt/kvm/arm/vgic/vgic-its.c
> index 2a1ccbf..7364b7d 100644
> --- a/virt/kvm/arm/vgic/vgic-its.c
> +++ b/virt/kvm/arm/vgic/vgic-its.c
> @@ -726,6 +726,21 @@ static void vgic_its_free_collection(struct vgic_its *its, u32 coll_id)
>  	kfree(collection);
>  }
>  
> +static void ite_list_insert_sorted(struct list_head *h, struct its_ite *ite)
> +{
> +	struct list_head *pos = h->next;
> +	u32 id = ite->event_id;
> +
> +	while (pos != h) {
> +		struct its_ite *iter =
> +			list_entry(pos, struct its_ite, ite_list);
> +		if (id < iter->event_id)
> +			break;
> +		pos = pos->next;
> +	}
> +	list_add_tail(&ite->ite_list, pos);
> +}
> +
>  /* Must be called with its_lock mutex held */
>  static int vgic_its_alloc_ite(struct its_device *device,
>  			       struct its_ite **itep,
> @@ -742,7 +757,7 @@ static int vgic_its_alloc_ite(struct its_device *device,
>  	ite->collection = collection;
>  	ite->lpi = lpi_id;
>  
> -	list_add_tail(&ite->ite_list, &device->itt_head);
> +	ite_list_insert_sorted(&device->itt_head, ite);
>  	*itep = ite;
>  	return 0;
>  }
> @@ -835,6 +850,22 @@ static void vgic_its_unmap_device(struct kvm *kvm, struct its_device *device)
>  	kfree(device);
>  }
>  
> +static void device_list_insert_sorted(struct list_head *h,
> +				      struct its_device *dev)
> +{
> +	struct list_head *pos = h->next;
> +	u32 id = dev->device_id;
> +
> +	while (pos != h) {
> +		struct its_device *iter =
> +			list_entry(pos, struct its_device, dev_list);
> +		if (id < iter->device_id)
> +			break;
> +		pos = pos->next;
> +	}
> +	list_add_tail(&dev->dev_list, pos);
> +}
> +
>  /* Must be called with its_lock mutex held */
>  static int vgic_its_alloc_device(struct vgic_its *its,
>  				 struct its_device **devp,
> @@ -852,7 +883,8 @@ static int vgic_its_alloc_device(struct vgic_its *its,
>  	device->nb_eventid_bits = nb_eventid_bits;
>  	INIT_LIST_HEAD(&device->itt_head);
>  
> -	list_add_tail(&device->dev_list, &its->device_list);
> +	device_list_insert_sorted(&its->device_list, device);
> +
>  	*devp = device;
>  
>  	return 0;

What is the actual gain for sorting the list at runtime, vs sorting it
at save time? A save/restore operation is a very rare event compared to
the normal use of the ITS, so I'd rather put the overhead on the rarest
event if possible.

Thanks,

	M.
-- 
Jazz is not dead. It just smells funny.