On Tue, 18 Jun 2024 08:57:52 +0800, Jason Wang <jasow...@redhat.com> wrote:
> On Mon, Jun 17, 2024 at 3:39 PM Xuan Zhuo <xuanz...@linux.alibaba.com> wrote:
> >
> > On Mon, 17 Jun 2024 13:00:13 +0800, Jason Wang <jasow...@redhat.com> wrote:
> > > On Fri, Jun 14, 2024 at 2:39 PM Xuan Zhuo <xuanz...@linux.alibaba.com>
> > > wrote:
> > > >
> > > > If the xsk is enabling, the xsk tx will share the send queue.
> > > > But the xsk requires that the send queue use the premapped mode.
> > > > So the send queue must support premapped mode when it is bound to
> > > > af-xdp.
> > > >
> > > > * virtnet_sq_set_premapped(sq, true) is used to enable premapped mode.
> > > >
> > > > In this mode, the driver will record the dma info when skb or xdp
> > > > frame is sent.
> > > >
> > > > Currently, the SQ premapped mode is operational only with af-xdp. In
> > > > this mode, af-xdp, the kernel stack, and xdp tx/redirect will share
> > > > the same SQ. Af-xdp independently manages its DMA. The kernel stack
> > > > and xdp tx/redirect utilize this DMA metadata to manage the DMA
> > > > info.
> > > >
> > > > If the indirect descriptor feature be supported, the volume of DMA
> > > > details we need to maintain becomes quite substantial. Here, we have
> > > > a cap on the amount of DMA info we manage.
> > > >
> > > > If the kernel stack and xdp tx/redirect attempt to use more
> > > > descriptors, virtnet_add_outbuf() will return an -ENOMEM error. But
> > > > the af-xdp can work continually.
> > >
> > > Rethink of this whole logic, it looks like all the complication came
> > > as we decided to go with a pre queue pre mapping flag. I wonder if
> > > things could be simplified if we do that per buffer?
> >
> > YES. That will be simply.
> >
> > Then this patch will be not needed. The virtio core must record the
> > premapped
> > imfo to the virtio ring state or extra.
> >
> >
> > http://lore.kernel.org/all/20230517022249.20790-6-xuanz...@linux.alibaba.com
>
> Yes, something like this. I think it's worthwhile to re-consider that
> approach. If my memory is correct, we haven't spotted the complicated
> issues we need to deal with like this patch.
>
> >
> > >
> > > Then we don't need complex logic like dmainfo and cap.
> >
> > So the premapped mode and the internal dma mode can coexist.
> > Then we do not need to make the sq to support the premapped mode.
>
> Probably.
>
> >
> >
> > >
> > > >
> > > > * virtnet_sq_set_premapped(sq, false) is used to disable premapped mode.
> > > >
> > > > Signed-off-by: Xuan Zhuo <xuanz...@linux.alibaba.com>
> > > > ---
> > > > drivers/net/virtio_net.c | 228 ++++++++++++++++++++++++++++++++++++++-
> > > > 1 file changed, 224 insertions(+), 4 deletions(-)
> > > >
> > > > diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> > > > index e84a4624549b..88ab9ea1646f 100644
> > > > --- a/drivers/net/virtio_net.c
> > > > +++ b/drivers/net/virtio_net.c
> > > > @@ -25,6 +25,7 @@
> > > > #include <net/net_failover.h>
> > > > #include <net/netdev_rx_queue.h>
> > > > #include <net/netdev_queues.h>
> > > > +#include <uapi/linux/virtio_ring.h>
> > >
> > > Why do we need this?
> >
> > for using VIRTIO_RING_F_INDIRECT_DESC
>
> Ok. It's probably a hint that something like layer violation happens.
> A specific driver should not know details about the ring layout ...
But the blk device did the same thing.
>
> >
> >
> > >
> > > >
> > > > static int napi_weight = NAPI_POLL_WEIGHT;
> > > > module_param(napi_weight, int, 0444);
> > > > @@ -276,6 +277,26 @@ struct virtnet_rq_dma {
> > > > u16 need_sync;
> > > > };
> > > >
> > > > +struct virtnet_sq_dma {
> > > > + union {
> > > > + struct llist_node node;
> > > > + struct llist_head head;
> > >
> > > If we want to cap the #dmas, could we simply use an array instead of
> > > the list here?
> > >
> > > > + void *data;
> > > > + };
> > > > + dma_addr_t addr;
> > > > + u32 len;
> > > > + u8 num;
> > > > +};
> > > > +
> > > > +struct virtnet_sq_dma_info {
> > > > + /* record for kfree */
> > > > + void *p;
> > > > +
> > > > + u32 free_num;
> > > > +
> > > > + struct llist_head free;
> > > > +};
> > > > +
> > > > /* Internal representation of a send virtqueue */
> > > > struct send_queue {
> > > > /* Virtqueue associated with this send _queue */
> > > > @@ -295,6 +316,11 @@ struct send_queue {
> > > >
> > > > /* Record whether sq is in reset state. */
> > > > bool reset;
> > > > +
> > > > + /* SQ is premapped mode or not. */
> > > > + bool premapped;
> > > > +
> > > > + struct virtnet_sq_dma_info dmainfo;
> > > > };
> > > >
> > > > /* Internal representation of a receive virtqueue */
> > > > @@ -492,9 +518,11 @@ static void virtnet_sq_free_unused_buf(struct
> > > > virtqueue *vq, void *buf);
> > > > enum virtnet_xmit_type {
> > > > VIRTNET_XMIT_TYPE_SKB,
> > > > VIRTNET_XMIT_TYPE_XDP,
> > > > + VIRTNET_XMIT_TYPE_DMA,
> > >
> > > I think the name is confusing, how about TYPE_PREMAPPED?
> > >
> > > > };
> > > >
> > > > -#define VIRTNET_XMIT_TYPE_MASK (VIRTNET_XMIT_TYPE_SKB |
> > > > VIRTNET_XMIT_TYPE_XDP)
> > > > +#define VIRTNET_XMIT_TYPE_MASK (VIRTNET_XMIT_TYPE_SKB |
> > > > VIRTNET_XMIT_TYPE_XDP \
> > > > + | VIRTNET_XMIT_TYPE_DMA)
> > > >
> > > > static enum virtnet_xmit_type virtnet_xmit_ptr_strip(void **ptr)
> > > > {
> > > > @@ -510,12 +538,180 @@ static void *virtnet_xmit_ptr_mix(void *ptr,
> > > > enum virtnet_xmit_type type)
> > > > return (void *)((unsigned long)ptr | type);
> > > > }
> > > >
> > > > +static void virtnet_sq_unmap(struct send_queue *sq, void **data)
> > > > +{
> > > > + struct virtnet_sq_dma *head, *tail, *p;
> > > > + int i;
> > > > +
> > > > + head = *data;
> > > > +
> > > > + p = head;
> > > > +
> > > > + for (i = 0; i < head->num; ++i) {
> > > > + virtqueue_dma_unmap_page_attrs(sq->vq, p->addr, p->len,
> > > > + DMA_TO_DEVICE, 0);
> > > > + tail = p;
> > > > + p = llist_entry(llist_next(&p->node), struct
> > > > virtnet_sq_dma, node);
> > > > + }
> > > > +
> > > > + *data = tail->data;
> > > > +
> > > > + __llist_add_batch(&head->node, &tail->node, &sq->dmainfo.free);
> > > > +
> > > > + sq->dmainfo.free_num += head->num;
> > > > +}
> > > > +
> > > > +static void *virtnet_dma_chain_update(struct send_queue *sq,
> > > > + struct virtnet_sq_dma *head,
> > > > + struct virtnet_sq_dma *tail,
> > > > + u8 num, void *data)
> > > > +{
> > > > + sq->dmainfo.free_num -= num;
> > > > + head->num = num;
> > > > +
> > > > + tail->data = data;
> > > > +
> > > > + return virtnet_xmit_ptr_mix(head, VIRTNET_XMIT_TYPE_DMA);
> > > > +}
> > > > +
> > > > +static struct virtnet_sq_dma *virtnet_sq_map_sg(struct send_queue *sq,
> > > > int num, void *data)
> > > > +{
> > > > + struct virtnet_sq_dma *head = NULL, *p = NULL;
> > > > + struct scatterlist *sg;
> > > > + dma_addr_t addr;
> > > > + int i, err;
> > > > +
> > > > + if (num > sq->dmainfo.free_num)
> > > > + return NULL;
> > > > +
> > > > + for (i = 0; i < num; ++i) {
> > > > + sg = &sq->sg[i];
> > > > +
> > > > + addr = virtqueue_dma_map_page_attrs(sq->vq, sg_page(sg),
> > > > + sg->offset,
> > > > + sg->length,
> > > > DMA_TO_DEVICE,
> > > > + 0);
> > > > + err = virtqueue_dma_mapping_error(sq->vq, addr);
> > > > + if (err)
> > > > + goto err;
> > > > +
> > > > + sg->dma_address = addr;
> > > > +
> > > > + p = llist_entry(llist_del_first(&sq->dmainfo.free),
> > > > + struct virtnet_sq_dma, node);
> > > > +
> > > > + p->addr = sg->dma_address;
> > > > + p->len = sg->length;
> > >
> > > I may miss something, but I don't see how we cap the total number of
> > > dmainfos.
> >
> > static void *virtnet_dma_chain_update(struct send_queue *sq,
> > struct virtnet_sq_dma *head,
> > struct virtnet_sq_dma *tail,
> > u8 num, void *data)
> > {
> > sq->dmainfo.free_num -= num;
> > -> head->num = num;
> >
> > tail->data = data;
> >
> > return virtnet_xmit_ptr_mix(head, VIRTNET_XMIT_TYPE_DMA);
> > }
>
> Ok, speak too fast I guess it should be more like:
>
> if (num > sq->dmainfo.free_num)
> return NULL;
static struct virtnet_sq_dma *virtnet_sq_map_sg(struct send_queue *sq, int num,
void *data)
{
struct virtnet_sq_dma *head = NULL, *p = NULL;
struct scatterlist *sg;
dma_addr_t addr;
int i, err;
if (num > sq->dmainfo.free_num)
return NULL;
Do you mean this?
Thanks.
>
> Thanks
>
> >
> >
> >
> > Thanks.
> >
> > >
> > > Thanks
> > >
> >
>
