search for: out_sg

..., > return next; > } > > -static inline int virtqueue_add_split(struct virtqueue *_vq, > - struct scatterlist *sgs[], > - unsigned int total_sg, > - unsigned int out_sgs, > - unsigned int in_sgs, > - void *data, > - void *ctx, > - gfp_t gfp) > +static int virtqueue_map_sgs(struct vring_virtqueue *vq, > +...

...struct virtio_pstore_hdr *hdr; > > + ssize_t len; > > + > > + for (;;) { > > + elem = virtqueue_pop(vq, sizeof(VirtQueueElement)); > > + if (!elem) { > > + return; > > + } > > + > > + hdr = elem->out_sg[0].iov_base; > > + if (elem->out_sg[0].iov_len != sizeof(*hdr)) { > > + error_report("invalid header size: %u", > > + (unsigned)elem->out_sg[0].iov_len); > > + exit(1); > > + } > > Pleas...

...struct virtio_pstore_hdr *hdr; > > + ssize_t len; > > + > > + for (;;) { > > + elem = virtqueue_pop(vq, sizeof(VirtQueueElement)); > > + if (!elem) { > > + return; > > + } > > + > > + hdr = elem->out_sg[0].iov_base; > > + if (elem->out_sg[0].iov_len != sizeof(*hdr)) { > > + error_report("invalid header size: %u", > > + (unsigned)elem->out_sg[0].iov_len); > > + exit(1); > > + } > > Pleas...

...e, enum dma_data_direction direction) @@ -520,6 +528,80 @@ static inline unsigned int virtqueue_add_desc_split(struct virtqueue *vq, return next; } +static void virtqueue_unmap_sgs(struct vring_virtqueue *vq, + struct scatterlist *sgs[], + unsigned int total_sg, + unsigned int out_sgs, + unsigned int in_sgs) +{ + struct scatterlist *sg; + unsigned int n; + + if (!vq->use_dma_api) + return; + + for (n = 0; n < out_sgs; n++) { + for (sg = sgs[n]; sg; sg = sg_next(sg)) { + if (!sg->dma_address) + return; + + dma_unmap_page(vring_dma_dev(vq), sg->dma_address...

...c +++ b/drivers/virtio/virtio_ring.c @@ -520,6 +520,73 @@ static inline unsigned int virtqueue_add_desc_split(struct virtqueue *vq, return next; } +static int virtqueue_map_sgs(struct vring_virtqueue *vq, + struct scatterlist *sgs[], + unsigned int total_sg, + unsigned int out_sgs, + unsigned int in_sgs) +{ + struct scatterlist *sg; + unsigned int n; + + for (n = 0; n < out_sgs; n++) { + for (sg = sgs[n]; sg; sg = sg_next(sg)) { + dma_addr_t addr = vring_map_one_sg(vq, sg, DMA_TO_DEVICE); + + if (vring_mapping_error(vq, addr)) + return -ENOMEM; + + sg-&g...

...sg->length, direction); + + if (dma_mapping_error(vring_dma_dev(vq), *addr)) + return -ENOMEM; + + return 0; } static dma_addr_t vring_map_single(const struct vring_virtqueue *vq, @@ -588,8 +593,9 @@ static inline int virtqueue_add_split(struct virtqueue *_vq, for (n = 0; n < out_sgs; n++) { for (sg = sgs[n]; sg; sg = sg_next(sg)) { - dma_addr_t addr = vring_map_one_sg(vq, sg, DMA_TO_DEVICE); - if (vring_mapping_error(vq, addr)) + dma_addr_t addr; + + if (vring_map_one_sg(vq, sg, DMA_TO_DEVICE, &addr)) goto unmap_release; prev = i; @@ -603,8 +609,9 @@...

Currently virtio-net code relys on the layout of descriptor, this patchset removed the assumptions and introduced a control command to set mac address. Last patch is a trivial renaming. V2: check guest's iov_len V3: fix of migration compatibility make mac field in config space read-only when new feature is acked V4: add fix of descriptor layout assumptions, trivial rename V5: fix

Currently virtio-net code relys on the layout of descriptor, this patchset removed the assumptions and introduced a control command to set mac address. Last patch is a trivial renaming. V2: check guest's iov_len V3: fix of migration compatibility make mac field in config space read-only when new feature is acked V4: add fix of descriptor layout assumptions, trivial rename V5: fix

XDP socket(AF_XDP) is an excellent bypass kernel network framework. The zero copy feature of xsk (XDP socket) needs to be supported by the driver. The performance of zero copy is very good. ENV: Qemu with vhost. vhost cpu | Guest APP CPU |Guest Softirq CPU | PPS -----------------------------|---------------|------------------|------------ xmit by sockperf: 90% | 100%

...gfp_t gfp) > { > struct vring_virtqueue *vq = to_vvq(_vq); > @@ -767,9 +772,11 @@ static inline int virtqueue_add_split(struct virtqueue *_vq, > goto end; > } > > - err = virtqueue_map_sgs(vq, sgs, total_sg, out_sgs, in_sgs); > - if (err) > - goto err; > + if (dma_map) { Could we simply check sg->dma_addr in this case? Then we don't need to introduce the dma_map flag. Thanks > + err = virtqueue_map_sgs(vq, sgs, total_sg, out_sgs, in_sgs); > +...

...Note that MAX_SKB_FRAGS depends on a config knob (CONFIG_MAX_SKB_FRAGS) and valid/reasonable values are up to 45. The total stack usage can be pretty large (~700 bytes). As this is under the vsk tx lock, have you considered moving such data in the virtio_vsock struct? > int ret, in_sg = 0, out_sg = 0; > struct sk_buff *skb; > bool reply; > @@ -111,12 +113,39 @@ virtio_transport_send_pkt_work(struct work_struct *work) > > virtio_transport_deliver_tap_pkt(skb); > reply = virtio_vsock_skb_reply(skb); > + sg_init_one(&bufs[out_sg], virtio_vsock_hdr(skb),...

From: Amos Kong <akong at redhat.com> Currently virtio-net code relys on the layout of descriptor, this patchset removed the assumptions and introduced a control command to set mac address. Last patch is a trivial renaming. V2: check guest's iov_len V3: fix of migration compatibility make mac field in config space read-only when new feature is acked V4: add fix of descriptor layout

From: Amos Kong <akong at redhat.com> Currently virtio-net code relys on the layout of descriptor, this patchset removed the assumptions and introduced a control command to set mac address. Last patch is a trivial renaming. V2: check guest's iov_len V3: fix of migration compatibility make mac field in config space read-only when new feature is acked V4: add fix of descriptor layout

...(struct vring_virtqueue *vq, struct vring_packed_desc *desc) { struct scatterlist *sg; - unsigned int i, n, err_idx; + unsigned int i, n; u16 head, id; dma_addr_t addr; @@ -1362,14 +1362,9 @@ static int virtqueue_add_indirect_packed(struct vring_virtqueue *vq, for (n = 0; n < out_sgs + in_sgs; n++) { for (sg = sgs[n]; sg; sg = sg_next(sg)) { - addr = vring_map_one_sg(vq, sg, n < out_sgs ? - DMA_TO_DEVICE : DMA_FROM_DEVICE); - if (vring_mapping_error(vq, addr)) - goto unmap_release; - desc[i].flags = cpu_to_le16(n < out_sgs ? 0 : VRING_DESC_F_WRIT...

XDP socket(AF_XDP) is an excellent bypass kernel network framework. The zero copy feature of xsk (XDP socket) needs to be supported by the driver. The performance of zero copy is very good. ENV: Qemu with vhost. vhost cpu | Guest APP CPU |Guest Softirq CPU | PPS -----------------------------|---------------|------------------|------------ xmit by sockperf: 90% | 100%

..._t gfp) > > { > > struct vring_virtqueue *vq = to_vvq(_vq); > > @@ -767,9 +772,11 @@ static inline int virtqueue_add_split(struct virtqueue *_vq, > > goto end; > > } > > > > - err = virtqueue_map_sgs(vq, sgs, total_sg, out_sgs, in_sgs); > > - if (err) > > - goto err; > > + if (dma_map) { > > Could we simply check sg->dma_addr in this case? Then we don't need to > introduce the dma_map flag. > > Thanks > > > > + err = virtqueue_ma...

...e, enum dma_data_direction direction) @@ -520,6 +528,80 @@ static inline unsigned int virtqueue_add_desc_split(struct virtqueue *vq, return next; } +static void virtqueue_unmap_sgs(struct vring_virtqueue *vq, + struct scatterlist *sgs[], + unsigned int total_sg, + unsigned int out_sgs, + unsigned int in_sgs) +{ + struct scatterlist *sg; + unsigned int n; + + if (!vq->use_dma_api) + return; + + for (n = 0; n < out_sgs; n++) { + for (sg = sgs[n]; sg; sg = sg_next(sg)) { + if (!sg->dma_address) + return; + + dma_unmap_page(vring_dma_dev(vq), sg->dma_address...

OK, so I've spent a few days benchmarking. Turns out 80% of virtio_add_buf cases are uni-directional (including the always-performance-sensitive networking code), and that gets no performance penalty (though tests with real networking would be appreciated!). I'm not reposting all the "convert driver to virtio_add_outbuf()" patches: just the scsi one which I didn't have

OK, so I've spent a few days benchmarking. Turns out 80% of virtio_add_buf cases are uni-directional (including the always-performance-sensitive networking code), and that gets no performance penalty (though tests with real networking would be appreciated!). I'm not reposting all the "convert driver to virtio_add_outbuf()" patches: just the scsi one which I didn't have

...e, enum dma_data_direction direction) @@ -520,6 +528,80 @@ static inline unsigned int virtqueue_add_desc_split(struct virtqueue *vq, return next; } +static void virtqueue_unmap_sgs(struct vring_virtqueue *vq, + struct scatterlist *sgs[], + unsigned int total_sg, + unsigned int out_sgs, + unsigned int in_sgs) +{ + struct scatterlist *sg; + unsigned int n; + + if (!vq->use_dma_api) + return; + + for (n = 0; n < out_sgs; n++) { + for (sg = sgs[n]; sg; sg = sg_next(sg)) { + if (!sg->dma_address) + return; + + dma_unmap_page(vring_dma_dev(vq), sg->dma_address...