On Thu, 28 Mar 2024 16:07:14 +0800, Jason Wang <jasowang@xxxxxxxxxx> wrote:
> On Thu, Mar 28, 2024 at 3:32 PM Xuan Zhuo <xuanzhuo@xxxxxxxxxxxxxxxxx> wrote:
> >
> > On Thu, 28 Mar 2024 14:56:47 +0800, Jason Wang <jasowang@xxxxxxxxxx> wrote:
> > > On Wed, Mar 27, 2024 at 7:14 PM Xuan Zhuo <xuanzhuo@xxxxxxxxxxxxxxxxx> wrote:
> > > >
> > > > In the functions vring_unmap_extra_packed and vring_unmap_desc_packed,
> > > > multiple checks are made whether unmap is performed and whether it is
> > > > INDIRECT.
> > > >
> > > > These two functions are usually called in a loop, and we should put the
> > > > check outside the loop.
> > > >
> > > > And we unmap the descs with VRING_DESC_F_INDIRECT on the same path with
> > > > other descs, that make the thing more complex. If we distinguish the
> > > > descs with VRING_DESC_F_INDIRECT before unmap, thing will be clearer.
> > > >
> > > > For desc with VRING_DESC_F_INDIRECT flag:
> > > > 1. only one desc of the desc table is used, we do not need the loop
> > > > Theoretically, indirect descriptors could be chained.
> > > > But now, that is not supported by "add", so we ignore this case.
> > > > 2. the called unmap api is difference from the other desc
> > > > 3. the vq->premapped is not needed to check
> > > > 4. the vq->indirect is not needed to check
> > > > 5. the state->indir_desc must not be null
> > >
> > > It doesn't explain the connection to the goal of this series. If it's
> > > not a must I'd suggest moving it to a separate patch.
> >
> >
> > The "no store dma ..." depends this.
> >
> > I will add this message in next version.
> >
> >
> > >
> > > >
> > > > Signed-off-by: Xuan Zhuo <xuanzhuo@xxxxxxxxxxxxxxxxx>
> > >
> > > Rethink this, it looks to me it would complicate the codes furtherly.
> > >
> > > For example, vring_map_xxx() helpers will check premappred and
> > > use_dma_api by itself. But in the case of vring_unmap() you want to
> > > move those checks to the caller. This will result in tricky codes that
> > > are hard to understand.
> > >
> > > We need to be consistent here.
> > >
> > > If we try to optimize unmap we need to optimize map as well. But
> > > generally it would complicate the logic of the caller if we want to
> > > let the caller to differ. Ideally, the caller of those function should
> > > know nothing about use_dma_api, premapped and other.
> >
> >
> > The key is that we can check "use_dma_api, premapped" to skip the loop.
> > If the vring_unmap_xxx is called, the "use_dma_api, premapped" is checked in
> > advance, so that is a waste to check thest again.
>
> Right, but we have the same logic for map.
But we can not skip the loop for map.
>
> >
> >
> > >
> > > > ---
> > > > drivers/virtio/virtio_ring.c | 78 ++++++++++++++++++------------------
> > > > 1 file changed, 40 insertions(+), 38 deletions(-)
> > > >
> > > > diff --git a/drivers/virtio/virtio_ring.c b/drivers/virtio/virtio_ring.c
> > > > index 03360073bd4a..a2838fe1cc08 100644
> > > > --- a/drivers/virtio/virtio_ring.c
> > > > +++ b/drivers/virtio/virtio_ring.c
> > > > @@ -1214,6 +1214,7 @@ static u16 packed_last_used(u16 last_used_idx)
> > > > return last_used_idx & ~(-(1 << VRING_PACKED_EVENT_F_WRAP_CTR));
> > > > }
> > > >
> > > > +/* caller must check vring_need_unmap_buffer() */
> > > > static void vring_unmap_extra_packed(const struct vring_virtqueue *vq,
> > > > const struct vring_desc_extra *extra)
> > > > {
> > > > @@ -1221,33 +1222,18 @@ static void vring_unmap_extra_packed(const struct vring_virtqueue *vq,
> > > >
> > > > flags = extra->flags;
> > > >
> > > > - if (flags & VRING_DESC_F_INDIRECT) {
> > > > - if (!vq->use_dma_api)
> > > > - return;
> > > > -
> > > > - dma_unmap_single(vring_dma_dev(vq),
> > > > - extra->addr, extra->len,
> > > > - (flags & VRING_DESC_F_WRITE) ?
> > > > - DMA_FROM_DEVICE : DMA_TO_DEVICE);
> > > > - } else {
> > > > - if (!vring_need_unmap_buffer(vq))
> > > > - return;
> > > > -
> > > > - dma_unmap_page(vring_dma_dev(vq),
> > > > - extra->addr, extra->len,
> > > > - (flags & VRING_DESC_F_WRITE) ?
> > > > - DMA_FROM_DEVICE : DMA_TO_DEVICE);
> > > > - }
> > > > + dma_unmap_page(vring_dma_dev(vq),
> > > > + extra->addr, extra->len,
> > > > + (flags & VRING_DESC_F_WRITE) ?
> > > > + DMA_FROM_DEVICE : DMA_TO_DEVICE);
> > > > }
> > > >
> > > > +/* caller must check vring_need_unmap_buffer() */
> > > > static void vring_unmap_desc_packed(const struct vring_virtqueue *vq,
> > > > const struct vring_packed_desc *desc)
> > > > {
> > > > u16 flags;
> > > >
> > > > - if (!vring_need_unmap_buffer(vq))
> > > > - return;
> > > > -
> > > > flags = le16_to_cpu(desc->flags);
> > > >
> > > > dma_unmap_page(vring_dma_dev(vq),
> > > > @@ -1323,7 +1309,7 @@ static int virtqueue_add_indirect_packed(struct vring_virtqueue *vq,
> > > > total_sg * sizeof(struct vring_packed_desc),
> > > > DMA_TO_DEVICE);
> > > > if (vring_mapping_error(vq, addr)) {
> > > > - if (vq->premapped)
> > > > + if (!vring_need_unmap_buffer(vq))
> > > > goto free_desc;
> > >
> > > I would do this to make it much more easier to be read and avoid the warn:
> > >
> > > if (vring_mapping_error(vq, addr))
> > > goto unmap_release;
> > >
> > > unmap_release:
> > > if (vring_need_unmap_buffer(vq))
> > > for (i = 0, xxx)
> > > free_desc:
> > > kfree(desc);
> > >
> > > or it could be
> > >
> > > unmap_release:
> > > if (!vring_need_unmap_buffer(vq))
> > > goto free_desc;
> > >
> > > Still tricky but better.
> >
> > I am ok.
> >
> >
> > >
> > > >
> > > > goto unmap_release;
> > > > @@ -1338,10 +1324,11 @@ static int virtqueue_add_indirect_packed(struct vring_virtqueue *vq,
> > > > vq->packed.desc_extra[id].addr = addr;
> > > > vq->packed.desc_extra[id].len = total_sg *
> > > > sizeof(struct vring_packed_desc);
> > > > - vq->packed.desc_extra[id].flags = VRING_DESC_F_INDIRECT |
> > > > - vq->packed.avail_used_flags;
> > > > }
> > > >
> > > > + vq->packed.desc_extra[id].flags = VRING_DESC_F_INDIRECT |
> > > > + vq->packed.avail_used_flags;
> > >
> > > An example of the tricky code, I think you do this because you want to
> > > differ indirect in detach_buf_packed():
> > >
> > > flags = vq->packed.desc_extra[id].flags;
> > >
> > >
> > > > +
> > > > /*
> > > > * A driver MUST NOT make the first descriptor in the list
> > > > * available before all subsequent descriptors comprising
> > > > @@ -1382,6 +1369,8 @@ static int virtqueue_add_indirect_packed(struct vring_virtqueue *vq,
> > > > unmap_release:
> > > > err_idx = i;
> > > >
> > > > + WARN_ON(!vring_need_unmap_buffer(vq));
> > > > +
> > > > for (i = 0; i < err_idx; i++)
> > > > vring_unmap_desc_packed(vq, &desc[i]);
> > > >
> > > > @@ -1475,12 +1464,13 @@ static inline int virtqueue_add_packed(struct virtqueue *_vq,
> > > > desc[i].len = cpu_to_le32(sg->length);
> > > > desc[i].id = cpu_to_le16(id);
> > > >
> > > > - if (unlikely(vq->use_dma_api)) {
> > > > + if (vring_need_unmap_buffer(vq)) {
> > > > vq->packed.desc_extra[curr].addr = addr;
> > > > vq->packed.desc_extra[curr].len = sg->length;
> > > > - vq->packed.desc_extra[curr].flags =
> > > > - le16_to_cpu(flags);
> > > > }
> > > > +
> > > > + vq->packed.desc_extra[curr].flags = le16_to_cpu(flags);
> > > > +
> > > > prev = curr;
> > > > curr = vq->packed.desc_extra[curr].next;
> > > >
> > > > @@ -1530,6 +1520,8 @@ static inline int virtqueue_add_packed(struct virtqueue *_vq,
> > > >
> > > > vq->packed.avail_used_flags = avail_used_flags;
> > > >
> > > > + WARN_ON(!vring_need_unmap_buffer(vq));
> > > > +
> > > > for (n = 0; n < total_sg; n++) {
> > > > if (i == err_idx)
> > > > break;
> > > > @@ -1599,7 +1591,9 @@ static void detach_buf_packed(struct vring_virtqueue *vq,
> > > > struct vring_desc_state_packed *state = NULL;
> > > > struct vring_packed_desc *desc;
> > > > unsigned int i, curr;
> > > > + u16 flags;
> > > >
> > > > + flags = vq->packed.desc_extra[id].flags;
> > >
> > > Can we check vq->indirect && indir_desc here? Then we don't need
> > > special care to store flags in desc_extra.
> >
> >
> > No.
> >
> > When vq->indirect is true, but the desc may has not indirect flag.
>
> But we check indir_desc as well?
>
> vq->indirect = virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC) &&
> !cfg_vq_get(cfg, vq, ctx);
I think you are right.
I will fix in next version.
Thanks.
>
> Thanks
>
