On 2018年05月03日 21:54, Tiwei Bie wrote:

On Thu, May 03, 2018 at 03:25:29PM +0800, Jason Wang wrote:

quoted

On 2018年05月03日 10:09, Tiwei Bie wrote:

quoted

quoted

quoted

quoted

So how about we use the straightforward way then?

You mean we do new += vq->vring_packed.num instead
of event_idx -= vq->vring_packed.num before calling
vring_need_event()?

The problem is that, the second param (new_idx) of
vring_need_event() will be used for:

(__u16)(new_idx - event_idx - 1)
(__u16)(new_idx - old)

So if we change new, we will need to change old too.

I think that since we have a branch there anyway,
we are better off just special-casing if (wrap_counter != vq->wrap_counter).
Treat is differenty and avoid casts.

quoted

And that would be an ugly hack..

Best regards,
Tiwei Bie

I consider casts and huge numbers with two's complement
games even uglier.

The dependency on two's complement game is introduced
since the split ring.

In packed ring, old is calculated via:

old = vq->next_avail_idx - vq->num_added;

In split ring, old is calculated via:

old = vq->avail_idx_shadow - vq->num_added;

In both cases, when vq->num_added is bigger, old will
be a big number.

Best regards,
Tiwei Bie

How about just do something like vhost:

static u16 vhost_idx_diff(struct vhost_virtqueue *vq, u16 old, u16 new)
{
     if (new > old)
         return new - old;
     return  (new + vq->num - old);
}

static bool vhost_vring_packed_need_event(struct vhost_virtqueue *vq,
                       __u16 event_off, __u16 new,
                       __u16 old)
{
     return (__u16)(vhost_idx_diff(vq, new, event_off) - 1) <
            (__u16)vhost_idx_diff(vq, new, old);
}

?

It seems that there is a typo in above code. The second
param of vhost_idx_diff() is `old`, but when calling this
function in vhost_vring_packed_need_event(), `new` is
passed as the second param.

Right.

If we assume the second param of vhost_idx_diff() is new
and the third one is old, i.e.:

static u16 vhost_idx_diff(struct vhost_virtqueue *vq, u16 new, u16 old)
{
     if (new > old)
         return new - old;
     return  (new + vq->num - old);
}

I think it's still not right.

Because in virtqueue_enable_cb_delayed(), we may set an
event_off which is bigger than new and both of them have
wrapped. And in this case, although new is smaller than
event_off (i.e. the third param -- old), new shouldn't
add vq->num, and actually we are expecting a very big
idx diff.

Yes, so to calculate distance correctly between event and new, we just 
need to compare the warp counter and return false if it doesn't match 
without the need to try to add vq.num here.

Thanks

Best regards,
Tiwei Bie