Hi all: This RFC implement packed ring layout. The code were tested with pmd implement by Jens at http://dpdk.org/ml/archives/dev/2018-January/089417.html. Minor change was needed for pmd codes to kick virtqueue since it assumes a busy polling backend. Test were done between localhost and guest. Testpmd (rxonly) in guest reports 2.4Mpps. Testpmd (txonly) repots about 2.1Mpps. Notes: The event suppression /indirect descriptor support is complied test only because of lacked driver support. Changes from V1: - Refactor vhost used elem code to avoid open coding on used elem - Event suppression support (compile test only). - Indirect descriptor support (compile test only). - Zerocopy support. - vIOMMU support. - SCSI/VSOCK support (compile test only). - Fix several bugs For simplicity, I don't implement batching or other optimizations. Please review. Thanks Jason Wang (8): vhost: move get_rx_bufs to vhost.c vhost: hide used ring layout from device vhost: do not use vring_used_elem vhost_net: do not explicitly manipulate vhost_used_elem vhost: vhost_put_user() can accept metadata type virtio: introduce packed ring defines vhost: packed ring support vhost: event suppression for packed ring drivers/vhost/net.c | 138 ++----- drivers/vhost/scsi.c | 62 +-- drivers/vhost/vhost.c | 818 ++++++++++++++++++++++++++++++++++--- drivers/vhost/vhost.h | 46 ++- drivers/vhost/vsock.c | 42 +- include/uapi/linux/virtio_config.h | 9 + include/uapi/linux/virtio_ring.h | 32 ++ 7 files changed, 921 insertions(+), 226 deletions(-) -- 2.7.4
Move get_rx_bufs() to vhost.c and rename it to vhost_get_rx_bufs(). This helps to hide vring internal layout from specific device implementation. Packed ring implementation will benefit from this. Signed-off-by: Jason Wang <jasowang at redhat.com> --- drivers/vhost/net.c | 83 ++------------------------------------------------- drivers/vhost/vhost.c | 78 +++++++++++++++++++++++++++++++++++++++++++++++ drivers/vhost/vhost.h | 7 +++++ 3 files changed, 88 insertions(+), 80 deletions(-) diff --git a/drivers/vhost/net.c b/drivers/vhost/net.c index 8139bc7..57dfa63 100644 --- a/drivers/vhost/net.c +++ b/drivers/vhost/net.c @@ -658,83 +658,6 @@ static int vhost_net_rx_peek_head_len(struct vhost_net *net, struct sock *sk) return len; } -/* This is a multi-buffer version of vhost_get_desc, that works if - * vq has read descriptors only. - * @vq - the relevant virtqueue - * @datalen - data length we'll be reading - * @iovcount - returned count of io vectors we fill - * @log - vhost log - * @log_num - log offset - * @quota - headcount quota, 1 for big buffer - * returns number of buffer heads allocated, negative on error - */ -static int get_rx_bufs(struct vhost_virtqueue *vq, - struct vring_used_elem *heads, - int datalen, - unsigned *iovcount, - struct vhost_log *log, - unsigned *log_num, - unsigned int quota) -{ - unsigned int out, in; - int seg = 0; - int headcount = 0; - unsigned d; - int r, nlogs = 0; - /* len is always initialized before use since we are always called with - * datalen > 0. - */ - u32 uninitialized_var(len); - - while (datalen > 0 && headcount < quota) { - if (unlikely(seg >= UIO_MAXIOV)) { - r = -ENOBUFS; - goto err; - } - r = vhost_get_vq_desc(vq, vq->iov + seg, - ARRAY_SIZE(vq->iov) - seg, &out, - &in, log, log_num); - if (unlikely(r < 0)) - goto err; - - d = r; - if (d == vq->num) { - r = 0; - goto err; - } - if (unlikely(out || in <= 0)) { - vq_err(vq, "unexpected descriptor format for RX: " - "out %d, in %d\n", out, in); - r = -EINVAL; - goto err; - } - if (unlikely(log)) { - nlogs += *log_num; - log += *log_num; - } - heads[headcount].id = cpu_to_vhost32(vq, d); - len = iov_length(vq->iov + seg, in); - heads[headcount].len = cpu_to_vhost32(vq, len); - datalen -= len; - ++headcount; - seg += in; - } - heads[headcount - 1].len = cpu_to_vhost32(vq, len + datalen); - *iovcount = seg; - if (unlikely(log)) - *log_num = nlogs; - - /* Detect overrun */ - if (unlikely(datalen > 0)) { - r = UIO_MAXIOV + 1; - goto err; - } - return headcount; -err: - vhost_discard_vq_desc(vq, headcount); - return r; -} - /* Expects to be always run from workqueue - which acts as * read-size critical section for our kind of RCU. */ static void handle_rx(struct vhost_net *net) @@ -784,9 +707,9 @@ static void handle_rx(struct vhost_net *net) while ((sock_len = vhost_net_rx_peek_head_len(net, sock->sk))) { sock_len += sock_hlen; vhost_len = sock_len + vhost_hlen; - headcount = get_rx_bufs(vq, vq->heads + nheads, vhost_len, - &in, vq_log, &log, - likely(mergeable) ? UIO_MAXIOV : 1); + headcount = vhost_get_bufs(vq, vq->heads + nheads, vhost_len, + &in, vq_log, &log, + likely(mergeable) ? UIO_MAXIOV : 1); /* On error, stop handling until the next kick. */ if (unlikely(headcount < 0)) goto out; diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c index 1b3e8d2d..c57df71 100644 --- a/drivers/vhost/vhost.c +++ b/drivers/vhost/vhost.c @@ -2098,6 +2098,84 @@ int vhost_get_vq_desc(struct vhost_virtqueue *vq, } EXPORT_SYMBOL_GPL(vhost_get_vq_desc); +/* This is a multi-buffer version of vhost_get_desc, that works if + * vq has read descriptors only. + * @vq - the relevant virtqueue + * @datalen - data length we'll be reading + * @iovcount - returned count of io vectors we fill + * @log - vhost log + * @log_num - log offset + * @quota - headcount quota, 1 for big buffer + * returns number of buffer heads allocated, negative on error + */ +int vhost_get_bufs(struct vhost_virtqueue *vq, + struct vring_used_elem *heads, + int datalen, + unsigned *iovcount, + struct vhost_log *log, + unsigned *log_num, + unsigned int quota) +{ + unsigned int out, in; + int seg = 0; + int headcount = 0; + unsigned d; + int r, nlogs = 0; + /* len is always initialized before use since we are always called with + * datalen > 0. + */ + u32 uninitialized_var(len); + + while (datalen > 0 && headcount < quota) { + if (unlikely(seg >= UIO_MAXIOV)) { + r = -ENOBUFS; + goto err; + } + r = vhost_get_vq_desc(vq, vq->iov + seg, + ARRAY_SIZE(vq->iov) - seg, &out, + &in, log, log_num); + if (unlikely(r < 0)) + goto err; + + d = r; + if (d == vq->num) { + r = 0; + goto err; + } + if (unlikely(out || in <= 0)) { + vq_err(vq, "unexpected descriptor format for RX: " + "out %d, in %d\n", out, in); + r = -EINVAL; + goto err; + } + if (unlikely(log)) { + nlogs += *log_num; + log += *log_num; + } + heads[headcount].id = cpu_to_vhost32(vq, d); + len = iov_length(vq->iov + seg, in); + heads[headcount].len = cpu_to_vhost32(vq, len); + datalen -= len; + ++headcount; + seg += in; + } + heads[headcount - 1].len = cpu_to_vhost32(vq, len + datalen); + *iovcount = seg; + if (unlikely(log)) + *log_num = nlogs; + + /* Detect overrun */ + if (unlikely(datalen > 0)) { + r = UIO_MAXIOV + 1; + goto err; + } + return headcount; +err: + vhost_discard_vq_desc(vq, headcount); + return r; +} +EXPORT_SYMBOL_GPL(vhost_get_bufs); + /* Reverse the effect of vhost_get_vq_desc. Useful for error handling. */ void vhost_discard_vq_desc(struct vhost_virtqueue *vq, int n) { diff --git a/drivers/vhost/vhost.h b/drivers/vhost/vhost.h index ac4b605..cbc2c80 100644 --- a/drivers/vhost/vhost.h +++ b/drivers/vhost/vhost.h @@ -185,6 +185,13 @@ int vhost_get_vq_desc(struct vhost_virtqueue *, struct iovec iov[], unsigned int iov_count, unsigned int *out_num, unsigned int *in_num, struct vhost_log *log, unsigned int *log_num); +int vhost_get_bufs(struct vhost_virtqueue *vq, + struct vring_used_elem *heads, + int datalen, + unsigned *iovcount, + struct vhost_log *log, + unsigned *log_num, + unsigned int quota); void vhost_discard_vq_desc(struct vhost_virtqueue *, int n); int vhost_vq_init_access(struct vhost_virtqueue *); -- 2.7.4
Jason Wang
2018-Mar-26 03:38 UTC
[RFC PATCH V2 2/8] vhost: hide used ring layout from device
We used to return descriptor head by vhost_get_vq_desc() to device and pass it back to vhost_add_used() and its friends. This exposes the internal used ring layout to device which makes it hard to be extended for e.g packed ring layout. So this patch tries to hide the used ring layout by - letting vhost_get_vq_desc() return pointer to struct vring_used_elem - accepting pointer to struct vring_used_elem in vhost_add_used() and vhost_add_used_and_signal() This could help to hide used ring layout and make it easier to implement packed ring on top. Signed-off-by: Jason Wang <jasowang at redhat.com> --- drivers/vhost/net.c | 46 +++++++++++++++++++++----------------- drivers/vhost/scsi.c | 62 +++++++++++++++++++++++++++------------------------ drivers/vhost/vhost.c | 52 +++++++++++++++++++++--------------------- drivers/vhost/vhost.h | 9 +++++--- drivers/vhost/vsock.c | 42 +++++++++++++++++----------------- 5 files changed, 112 insertions(+), 99 deletions(-) diff --git a/drivers/vhost/net.c b/drivers/vhost/net.c index 57dfa63..f821fcd 100644 --- a/drivers/vhost/net.c +++ b/drivers/vhost/net.c @@ -422,22 +422,24 @@ static int vhost_net_enable_vq(struct vhost_net *n, static int vhost_net_tx_get_vq_desc(struct vhost_net *net, struct vhost_virtqueue *vq, + struct vring_used_elem *used_elem, struct iovec iov[], unsigned int iov_size, unsigned int *out_num, unsigned int *in_num) { unsigned long uninitialized_var(endtime); - int r = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov), + int r = vhost_get_vq_desc(vq, used_elem, vq->iov, ARRAY_SIZE(vq->iov), out_num, in_num, NULL, NULL); - if (r == vq->num && vq->busyloop_timeout) { + if (r == -ENOSPC && vq->busyloop_timeout) { preempt_disable(); endtime = busy_clock() + vq->busyloop_timeout; while (vhost_can_busy_poll(vq->dev, endtime) && vhost_vq_avail_empty(vq->dev, vq)) cpu_relax(); preempt_enable(); - r = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov), - out_num, in_num, NULL, NULL); + r = vhost_get_vq_desc(vq, used_elem, vq->iov, + ARRAY_SIZE(vq->iov), out_num, in_num, + NULL, NULL); } return r; @@ -459,7 +461,6 @@ static void handle_tx(struct vhost_net *net) struct vhost_net_virtqueue *nvq = &net->vqs[VHOST_NET_VQ_TX]; struct vhost_virtqueue *vq = &nvq->vq; unsigned out, in; - int head; struct msghdr msg = { .msg_name = NULL, .msg_namelen = 0, @@ -472,6 +473,7 @@ static void handle_tx(struct vhost_net *net) size_t hdr_size; struct socket *sock; struct vhost_net_ubuf_ref *uninitialized_var(ubufs); + struct vring_used_elem used; bool zcopy, zcopy_used; mutex_lock(&vq->mutex); @@ -494,20 +496,20 @@ static void handle_tx(struct vhost_net *net) vhost_zerocopy_signal_used(net, vq); - head = vhost_net_tx_get_vq_desc(net, vq, vq->iov, - ARRAY_SIZE(vq->iov), - &out, &in); - /* On error, stop handling until the next kick. */ - if (unlikely(head < 0)) - break; + err = vhost_net_tx_get_vq_desc(net, vq, &used, vq->iov, + ARRAY_SIZE(vq->iov), + &out, &in); /* Nothing new? Wait for eventfd to tell us they refilled. */ - if (head == vq->num) { + if (err == -ENOSPC) { if (unlikely(vhost_enable_notify(&net->dev, vq))) { vhost_disable_notify(&net->dev, vq); continue; } break; } + /* On error, stop handling until the next kick. */ + if (unlikely(err < 0)) + break; if (in) { vq_err(vq, "Unexpected descriptor format for TX: " "out %d, int %d\n", out, in); @@ -535,7 +537,8 @@ static void handle_tx(struct vhost_net *net) struct ubuf_info *ubuf; ubuf = nvq->ubuf_info + nvq->upend_idx; - vq->heads[nvq->upend_idx].id = cpu_to_vhost32(vq, head); + vq->heads[nvq->upend_idx].id + cpu_to_vhost32(vq, used.id); vq->heads[nvq->upend_idx].len = VHOST_DMA_IN_PROGRESS; ubuf->callback = vhost_zerocopy_callback; ubuf->ctx = nvq->ubufs; @@ -576,7 +579,7 @@ static void handle_tx(struct vhost_net *net) pr_debug("Truncated TX packet: " " len %d != %zd\n", err, len); if (!zcopy_used) - vhost_add_used_and_signal(&net->dev, vq, head, 0); + vhost_add_used_and_signal(&net->dev, vq, &used, 0); else vhost_zerocopy_signal_used(net, vq); vhost_net_tx_packet(net); @@ -707,14 +710,12 @@ static void handle_rx(struct vhost_net *net) while ((sock_len = vhost_net_rx_peek_head_len(net, sock->sk))) { sock_len += sock_hlen; vhost_len = sock_len + vhost_hlen; - headcount = vhost_get_bufs(vq, vq->heads + nheads, vhost_len, - &in, vq_log, &log, - likely(mergeable) ? UIO_MAXIOV : 1); - /* On error, stop handling until the next kick. */ - if (unlikely(headcount < 0)) - goto out; + err = vhost_get_bufs(vq, vq->heads + nheads, vhost_len, + &in, vq_log, &log, + likely(mergeable) ? UIO_MAXIOV : 1, + &headcount); /* OK, now we need to know about added descriptors. */ - if (!headcount) { + if (err == -ENOSPC) { if (unlikely(vhost_enable_notify(&net->dev, vq))) { /* They have slipped one in as we were * doing that: check again. */ @@ -725,6 +726,9 @@ static void handle_rx(struct vhost_net *net) * they refilled. */ goto out; } + /* On error, stop handling until the next kick. */ + if (unlikely(err < 0)) + goto out; if (nvq->rx_ring) msg.msg_control = vhost_net_buf_consume(&nvq->rxq); /* On overrun, truncate and discard */ diff --git a/drivers/vhost/scsi.c b/drivers/vhost/scsi.c index 7ad5709..654c71f 100644 --- a/drivers/vhost/scsi.c +++ b/drivers/vhost/scsi.c @@ -67,7 +67,7 @@ struct vhost_scsi_inflight { struct vhost_scsi_cmd { /* Descriptor from vhost_get_vq_desc() for virt_queue segment */ - int tvc_vq_desc; + struct vring_used_elem tvc_vq_used; /* virtio-scsi initiator task attribute */ int tvc_task_attr; /* virtio-scsi response incoming iovecs */ @@ -441,8 +441,9 @@ vhost_scsi_do_evt_work(struct vhost_scsi *vs, struct vhost_scsi_evt *evt) struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq; struct virtio_scsi_event *event = &evt->event; struct virtio_scsi_event __user *eventp; + struct vring_used_elem used; unsigned out, in; - int head, ret; + int ret; if (!vq->private_data) { vs->vs_events_missed = true; @@ -451,16 +452,16 @@ vhost_scsi_do_evt_work(struct vhost_scsi *vs, struct vhost_scsi_evt *evt) again: vhost_disable_notify(&vs->dev, vq); - head = vhost_get_vq_desc(vq, vq->iov, + ret = vhost_get_vq_desc(vq, &used, vq->iov, ARRAY_SIZE(vq->iov), &out, &in, NULL, NULL); - if (head < 0) { + if (ret == -ENOSPC) { + if (vhost_enable_notify(&vs->dev, vq)) + goto again; vs->vs_events_missed = true; return; } - if (head == vq->num) { - if (vhost_enable_notify(&vs->dev, vq)) - goto again; + if (ret < 0) { vs->vs_events_missed = true; return; } @@ -480,7 +481,7 @@ vhost_scsi_do_evt_work(struct vhost_scsi *vs, struct vhost_scsi_evt *evt) eventp = vq->iov[out].iov_base; ret = __copy_to_user(eventp, event, sizeof(*event)); if (!ret) - vhost_add_used_and_signal(&vs->dev, vq, head, 0); + vhost_add_used_and_signal(&vs->dev, vq, &used, 0); else vq_err(vq, "Faulted on vhost_scsi_send_event\n"); } @@ -541,7 +542,7 @@ static void vhost_scsi_complete_cmd_work(struct vhost_work *work) ret = copy_to_iter(&v_rsp, sizeof(v_rsp), &iov_iter); if (likely(ret == sizeof(v_rsp))) { struct vhost_scsi_virtqueue *q; - vhost_add_used(cmd->tvc_vq, cmd->tvc_vq_desc, 0); + vhost_add_used(cmd->tvc_vq, &cmd->tvc_vq_used, 0); q = container_of(cmd->tvc_vq, struct vhost_scsi_virtqueue, vq); vq = q - vs->vqs; __set_bit(vq, signal); @@ -784,7 +785,7 @@ static void vhost_scsi_submission_work(struct work_struct *work) static void vhost_scsi_send_bad_target(struct vhost_scsi *vs, struct vhost_virtqueue *vq, - int head, unsigned out) + struct vring_used_elem *used, unsigned out) { struct virtio_scsi_cmd_resp __user *resp; struct virtio_scsi_cmd_resp rsp; @@ -795,7 +796,7 @@ vhost_scsi_send_bad_target(struct vhost_scsi *vs, resp = vq->iov[out].iov_base; ret = __copy_to_user(resp, &rsp, sizeof(rsp)); if (!ret) - vhost_add_used_and_signal(&vs->dev, vq, head, 0); + vhost_add_used_and_signal(&vs->dev, vq, used, 0); else pr_err("Faulted on virtio_scsi_cmd_resp\n"); } @@ -807,11 +808,12 @@ vhost_scsi_handle_vq(struct vhost_scsi *vs, struct vhost_virtqueue *vq) struct virtio_scsi_cmd_req v_req; struct virtio_scsi_cmd_req_pi v_req_pi; struct vhost_scsi_cmd *cmd; + struct vring_used_elem used; struct iov_iter out_iter, in_iter, prot_iter, data_iter; u64 tag; u32 exp_data_len, data_direction; unsigned int out = 0, in = 0; - int head, ret, prot_bytes; + int ret, prot_bytes; size_t req_size, rsp_size = sizeof(struct virtio_scsi_cmd_resp); size_t out_size, in_size; u16 lun; @@ -831,22 +833,22 @@ vhost_scsi_handle_vq(struct vhost_scsi *vs, struct vhost_virtqueue *vq) vhost_disable_notify(&vs->dev, vq); for (;;) { - head = vhost_get_vq_desc(vq, vq->iov, - ARRAY_SIZE(vq->iov), &out, &in, - NULL, NULL); + ret = vhost_get_vq_desc(vq, &used, vq->iov, + ARRAY_SIZE(vq->iov), &out, &in, + NULL, NULL); pr_debug("vhost_get_vq_desc: head: %d, out: %u in: %u\n", - head, out, in); - /* On error, stop handling until the next kick. */ - if (unlikely(head < 0)) - break; + used.id, out, in); /* Nothing new? Wait for eventfd to tell us they refilled. */ - if (head == vq->num) { + if (ret == -ENOSPC) { if (unlikely(vhost_enable_notify(&vs->dev, vq))) { vhost_disable_notify(&vs->dev, vq); continue; } break; } + /* On error, stop handling until the next kick. */ + if (unlikely(ret < 0)) + break; /* * Check for a sane response buffer so we can report early * errors back to the guest. @@ -891,20 +893,20 @@ vhost_scsi_handle_vq(struct vhost_scsi *vs, struct vhost_virtqueue *vq) if (unlikely(!copy_from_iter_full(req, req_size, &out_iter))) { vq_err(vq, "Faulted on copy_from_iter\n"); - vhost_scsi_send_bad_target(vs, vq, head, out); + vhost_scsi_send_bad_target(vs, vq, &used, out); continue; } /* virtio-scsi spec requires byte 0 of the lun to be 1 */ if (unlikely(*lunp != 1)) { vq_err(vq, "Illegal virtio-scsi lun: %u\n", *lunp); - vhost_scsi_send_bad_target(vs, vq, head, out); + vhost_scsi_send_bad_target(vs, vq, &used, out); continue; } tpg = READ_ONCE(vs_tpg[*target]); if (unlikely(!tpg)) { /* Target does not exist, fail the request */ - vhost_scsi_send_bad_target(vs, vq, head, out); + vhost_scsi_send_bad_target(vs, vq, &used, out); continue; } /* @@ -950,7 +952,8 @@ vhost_scsi_handle_vq(struct vhost_scsi *vs, struct vhost_virtqueue *vq) if (data_direction != DMA_TO_DEVICE) { vq_err(vq, "Received non zero pi_bytesout," " but wrong data_direction\n"); - vhost_scsi_send_bad_target(vs, vq, head, out); + vhost_scsi_send_bad_target(vs, vq, + &used, out); continue; } prot_bytes = vhost32_to_cpu(vq, v_req_pi.pi_bytesout); @@ -958,7 +961,8 @@ vhost_scsi_handle_vq(struct vhost_scsi *vs, struct vhost_virtqueue *vq) if (data_direction != DMA_FROM_DEVICE) { vq_err(vq, "Received non zero pi_bytesin," " but wrong data_direction\n"); - vhost_scsi_send_bad_target(vs, vq, head, out); + vhost_scsi_send_bad_target(vs, vq, + &used, out); continue; } prot_bytes = vhost32_to_cpu(vq, v_req_pi.pi_bytesin); @@ -996,7 +1000,7 @@ vhost_scsi_handle_vq(struct vhost_scsi *vs, struct vhost_virtqueue *vq) vq_err(vq, "Received SCSI CDB with command_size: %d that" " exceeds SCSI_MAX_VARLEN_CDB_SIZE: %d\n", scsi_command_size(cdb), VHOST_SCSI_MAX_CDB_SIZE); - vhost_scsi_send_bad_target(vs, vq, head, out); + vhost_scsi_send_bad_target(vs, vq, &used, out); continue; } cmd = vhost_scsi_get_tag(vq, tpg, cdb, tag, lun, task_attr, @@ -1005,7 +1009,7 @@ vhost_scsi_handle_vq(struct vhost_scsi *vs, struct vhost_virtqueue *vq) if (IS_ERR(cmd)) { vq_err(vq, "vhost_scsi_get_tag failed %ld\n", PTR_ERR(cmd)); - vhost_scsi_send_bad_target(vs, vq, head, out); + vhost_scsi_send_bad_target(vs, vq, &used, out); continue; } cmd->tvc_vhost = vs; @@ -1025,7 +1029,7 @@ vhost_scsi_handle_vq(struct vhost_scsi *vs, struct vhost_virtqueue *vq) if (unlikely(ret)) { vq_err(vq, "Failed to map iov to sgl\n"); vhost_scsi_release_cmd(&cmd->tvc_se_cmd); - vhost_scsi_send_bad_target(vs, vq, head, out); + vhost_scsi_send_bad_target(vs, vq, &used, out); continue; } } @@ -1034,7 +1038,7 @@ vhost_scsi_handle_vq(struct vhost_scsi *vs, struct vhost_virtqueue *vq) * complete the virtio-scsi request in TCM callback context via * vhost_scsi_queue_data_in() and vhost_scsi_queue_status() */ - cmd->tvc_vq_desc = head; + cmd->tvc_vq_used = used; /* * Dispatch cmd descriptor for cmwq execution in process * context provided by vhost_scsi_workqueue. This also ensures diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c index c57df71..83cb1dc 100644 --- a/drivers/vhost/vhost.c +++ b/drivers/vhost/vhost.c @@ -1956,6 +1956,7 @@ static int get_indirect(struct vhost_virtqueue *vq, * never a valid descriptor number) if none was found. A negative code is * returned on error. */ int vhost_get_vq_desc(struct vhost_virtqueue *vq, + struct vring_used_elem *used, struct iovec iov[], unsigned int iov_size, unsigned int *out_num, unsigned int *in_num, struct vhost_log *log, unsigned int *log_num) @@ -1988,7 +1989,7 @@ int vhost_get_vq_desc(struct vhost_virtqueue *vq, * invalid. */ if (vq->avail_idx == last_avail_idx) - return vq->num; + return -ENOSPC; /* Only get avail ring entries after they have been * exposed by guest. @@ -2006,6 +2007,7 @@ int vhost_get_vq_desc(struct vhost_virtqueue *vq, return -EFAULT; } + used->id = ring_head; head = vhost16_to_cpu(vq, ring_head); /* If their number is silly, that's an error. */ @@ -2094,10 +2096,16 @@ int vhost_get_vq_desc(struct vhost_virtqueue *vq, /* Assume notifications from guest are disabled at this point, * if they aren't we would need to update avail_event index. */ BUG_ON(!(vq->used_flags & VRING_USED_F_NO_NOTIFY)); - return head; + return 0; } EXPORT_SYMBOL_GPL(vhost_get_vq_desc); +static void vhost_set_used_len(struct vhost_virtqueue *vq, + struct vring_used_elem *used, int len) +{ + used->len = cpu_to_vhost32(vq, len); +} + /* This is a multi-buffer version of vhost_get_desc, that works if * vq has read descriptors only. * @vq - the relevant virtqueue @@ -2114,13 +2122,13 @@ int vhost_get_bufs(struct vhost_virtqueue *vq, unsigned *iovcount, struct vhost_log *log, unsigned *log_num, - unsigned int quota) + unsigned int quota, + s16 *count) { unsigned int out, in; int seg = 0; int headcount = 0; - unsigned d; - int r, nlogs = 0; + int r = 0, nlogs = 0; /* len is always initialized before use since we are always called with * datalen > 0. */ @@ -2131,17 +2139,12 @@ int vhost_get_bufs(struct vhost_virtqueue *vq, r = -ENOBUFS; goto err; } - r = vhost_get_vq_desc(vq, vq->iov + seg, + r = vhost_get_vq_desc(vq, &heads[headcount], vq->iov + seg, ARRAY_SIZE(vq->iov) - seg, &out, &in, log, log_num); if (unlikely(r < 0)) goto err; - d = r; - if (d == vq->num) { - r = 0; - goto err; - } if (unlikely(out || in <= 0)) { vq_err(vq, "unexpected descriptor format for RX: " "out %d, in %d\n", out, in); @@ -2152,24 +2155,26 @@ int vhost_get_bufs(struct vhost_virtqueue *vq, nlogs += *log_num; log += *log_num; } - heads[headcount].id = cpu_to_vhost32(vq, d); + len = iov_length(vq->iov + seg, in); - heads[headcount].len = cpu_to_vhost32(vq, len); + vhost_set_used_len(vq, &heads[headcount], len); datalen -= len; ++headcount; seg += in; } - heads[headcount - 1].len = cpu_to_vhost32(vq, len + datalen); + vhost_set_used_len(vq, &heads[headcount - 1], len + datalen); *iovcount = seg; if (unlikely(log)) *log_num = nlogs; /* Detect overrun */ if (unlikely(datalen > 0)) { - r = UIO_MAXIOV + 1; + headcount = UIO_MAXIOV + 1; goto err; } - return headcount; + + *count = headcount; + return 0; err: vhost_discard_vq_desc(vq, headcount); return r; @@ -2185,14 +2190,11 @@ EXPORT_SYMBOL_GPL(vhost_discard_vq_desc); /* After we've used one of their buffers, we tell them about it. We'll then * want to notify the guest, using eventfd. */ -int vhost_add_used(struct vhost_virtqueue *vq, unsigned int head, int len) +int vhost_add_used(struct vhost_virtqueue *vq, struct vring_used_elem *used, + int len) { - struct vring_used_elem heads = { - cpu_to_vhost32(vq, head), - cpu_to_vhost32(vq, len) - }; - - return vhost_add_used_n(vq, &heads, 1); + vhost_set_used_len(vq, used, len); + return vhost_add_used_n(vq, used, 1); } EXPORT_SYMBOL_GPL(vhost_add_used); @@ -2325,9 +2327,9 @@ EXPORT_SYMBOL_GPL(vhost_signal); /* And here's the combo meal deal. Supersize me! */ void vhost_add_used_and_signal(struct vhost_dev *dev, struct vhost_virtqueue *vq, - unsigned int head, int len) + struct vring_used_elem *used, int len) { - vhost_add_used(vq, head, len); + vhost_add_used(vq, used, len); vhost_signal(dev, vq); } EXPORT_SYMBOL_GPL(vhost_add_used_and_signal); diff --git a/drivers/vhost/vhost.h b/drivers/vhost/vhost.h index cbc2c80..477bfd2 100644 --- a/drivers/vhost/vhost.h +++ b/drivers/vhost/vhost.h @@ -182,6 +182,7 @@ int vhost_vq_access_ok(struct vhost_virtqueue *vq); int vhost_log_access_ok(struct vhost_dev *); int vhost_get_vq_desc(struct vhost_virtqueue *, + struct vring_used_elem *used_elem, struct iovec iov[], unsigned int iov_count, unsigned int *out_num, unsigned int *in_num, struct vhost_log *log, unsigned int *log_num); @@ -191,15 +192,17 @@ int vhost_get_bufs(struct vhost_virtqueue *vq, unsigned *iovcount, struct vhost_log *log, unsigned *log_num, - unsigned int quota); + unsigned int quota, + s16 *count); void vhost_discard_vq_desc(struct vhost_virtqueue *, int n); int vhost_vq_init_access(struct vhost_virtqueue *); -int vhost_add_used(struct vhost_virtqueue *, unsigned int head, int len); +int vhost_add_used(struct vhost_virtqueue *vq, + struct vring_used_elem *elem, int len); int vhost_add_used_n(struct vhost_virtqueue *, struct vring_used_elem *heads, unsigned count); void vhost_add_used_and_signal(struct vhost_dev *, struct vhost_virtqueue *, - unsigned int id, int len); + struct vring_used_elem *, int len); void vhost_add_used_and_signal_n(struct vhost_dev *, struct vhost_virtqueue *, struct vring_used_elem *heads, unsigned count); void vhost_signal(struct vhost_dev *, struct vhost_virtqueue *); diff --git a/drivers/vhost/vsock.c b/drivers/vhost/vsock.c index 0d14e2f..fc20292 100644 --- a/drivers/vhost/vsock.c +++ b/drivers/vhost/vsock.c @@ -98,11 +98,12 @@ vhost_transport_do_send_pkt(struct vhost_vsock *vsock, for (;;) { struct virtio_vsock_pkt *pkt; + struct vring_used_elem used; struct iov_iter iov_iter; unsigned out, in; size_t nbytes; size_t len; - int head; + int ret; spin_lock_bh(&vsock->send_pkt_list_lock); if (list_empty(&vsock->send_pkt_list)) { @@ -116,16 +117,9 @@ vhost_transport_do_send_pkt(struct vhost_vsock *vsock, list_del_init(&pkt->list); spin_unlock_bh(&vsock->send_pkt_list_lock); - head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov), - &out, &in, NULL, NULL); - if (head < 0) { - spin_lock_bh(&vsock->send_pkt_list_lock); - list_add(&pkt->list, &vsock->send_pkt_list); - spin_unlock_bh(&vsock->send_pkt_list_lock); - break; - } - - if (head == vq->num) { + ret = vhost_get_vq_desc(vq, &used, vq->iov, ARRAY_SIZE(vq->iov), + &out, &in, NULL, NULL); + if (ret == -ENOSPC) { spin_lock_bh(&vsock->send_pkt_list_lock); list_add(&pkt->list, &vsock->send_pkt_list); spin_unlock_bh(&vsock->send_pkt_list_lock); @@ -139,6 +133,12 @@ vhost_transport_do_send_pkt(struct vhost_vsock *vsock, } break; } + if (ret < 0) { + spin_lock_bh(&vsock->send_pkt_list_lock); + list_add(&pkt->list, &vsock->send_pkt_list); + spin_unlock_bh(&vsock->send_pkt_list_lock); + break; + } if (out) { virtio_transport_free_pkt(pkt); @@ -146,7 +146,7 @@ vhost_transport_do_send_pkt(struct vhost_vsock *vsock, break; } - len = iov_length(&vq->iov[out], in); + len = vhost32_to_cpu(vq, used.len); iov_iter_init(&iov_iter, READ, &vq->iov[out], in, len); nbytes = copy_to_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter); @@ -163,7 +163,7 @@ vhost_transport_do_send_pkt(struct vhost_vsock *vsock, break; } - vhost_add_used(vq, head, sizeof(pkt->hdr) + pkt->len); + vhost_add_used(vq, &used, sizeof(pkt->hdr) + pkt->len); added = true; if (pkt->reply) { @@ -346,7 +346,8 @@ static void vhost_vsock_handle_tx_kick(struct vhost_work *work) struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock, dev); struct virtio_vsock_pkt *pkt; - int head; + struct vring_used_elem used; + int ret; unsigned int out, in; bool added = false; @@ -367,18 +368,17 @@ static void vhost_vsock_handle_tx_kick(struct vhost_work *work) goto no_more_replies; } - head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov), - &out, &in, NULL, NULL); - if (head < 0) - break; - - if (head == vq->num) { + ret = vhost_get_vq_desc(vq, &used, vq->iov, ARRAY_SIZE(vq->iov), + &out, &in, NULL, NULL); + if (ret == -ENOSPC) { if (unlikely(vhost_enable_notify(&vsock->dev, vq))) { vhost_disable_notify(&vsock->dev, vq); continue; } break; } + if (ret < 0) + break; pkt = vhost_vsock_alloc_pkt(vq, out, in); if (!pkt) { @@ -397,7 +397,7 @@ static void vhost_vsock_handle_tx_kick(struct vhost_work *work) else virtio_transport_free_pkt(pkt); - vhost_add_used(vq, head, sizeof(pkt->hdr) + len); + vhost_add_used(vq, &used, sizeof(pkt->hdr) + len); added = true; } -- 2.7.4
Instead of depending on the exported vring_used_elem, this patch switches to use a new internal structure vhost_used_elem which embed vring_used_elem in itself. This could be used to let vhost to record extra metadata for the incoming packed ring layout. Signed-off-by: Jason Wang <jasowang at redhat.com> --- drivers/vhost/net.c | 19 ++++++++++--------- drivers/vhost/scsi.c | 10 +++++----- drivers/vhost/vhost.c | 33 ++++++++++++++++----------------- drivers/vhost/vhost.h | 18 +++++++++++------- drivers/vhost/vsock.c | 6 +++--- 5 files changed, 45 insertions(+), 41 deletions(-) diff --git a/drivers/vhost/net.c b/drivers/vhost/net.c index f821fcd..7ea2aee 100644 --- a/drivers/vhost/net.c +++ b/drivers/vhost/net.c @@ -337,10 +337,10 @@ static void vhost_zerocopy_signal_used(struct vhost_net *net, int j = 0; for (i = nvq->done_idx; i != nvq->upend_idx; i = (i + 1) % UIO_MAXIOV) { - if (vq->heads[i].len == VHOST_DMA_FAILED_LEN) + if (vq->heads[i].elem.len == VHOST_DMA_FAILED_LEN) vhost_net_tx_err(net); - if (VHOST_DMA_IS_DONE(vq->heads[i].len)) { - vq->heads[i].len = VHOST_DMA_CLEAR_LEN; + if (VHOST_DMA_IS_DONE(vq->heads[i].elem.len)) { + vq->heads[i].elem.len = VHOST_DMA_CLEAR_LEN; ++j; } else break; @@ -363,7 +363,7 @@ static void vhost_zerocopy_callback(struct ubuf_info *ubuf, bool success) rcu_read_lock_bh(); /* set len to mark this desc buffers done DMA */ - vq->heads[ubuf->desc].len = success ? + vq->heads[ubuf->desc].elem.len = success ? VHOST_DMA_DONE_LEN : VHOST_DMA_FAILED_LEN; cnt = vhost_net_ubuf_put(ubufs); @@ -422,7 +422,7 @@ static int vhost_net_enable_vq(struct vhost_net *n, static int vhost_net_tx_get_vq_desc(struct vhost_net *net, struct vhost_virtqueue *vq, - struct vring_used_elem *used_elem, + struct vhost_used_elem *used_elem, struct iovec iov[], unsigned int iov_size, unsigned int *out_num, unsigned int *in_num) { @@ -473,7 +473,7 @@ static void handle_tx(struct vhost_net *net) size_t hdr_size; struct socket *sock; struct vhost_net_ubuf_ref *uninitialized_var(ubufs); - struct vring_used_elem used; + struct vhost_used_elem used; bool zcopy, zcopy_used; mutex_lock(&vq->mutex); @@ -537,9 +537,10 @@ static void handle_tx(struct vhost_net *net) struct ubuf_info *ubuf; ubuf = nvq->ubuf_info + nvq->upend_idx; - vq->heads[nvq->upend_idx].id - cpu_to_vhost32(vq, used.id); - vq->heads[nvq->upend_idx].len = VHOST_DMA_IN_PROGRESS; + vq->heads[nvq->upend_idx].elem.id + cpu_to_vhost32(vq, used.elem.id); + vq->heads[nvq->upend_idx].elem.len + VHOST_DMA_IN_PROGRESS; ubuf->callback = vhost_zerocopy_callback; ubuf->ctx = nvq->ubufs; ubuf->desc = nvq->upend_idx; diff --git a/drivers/vhost/scsi.c b/drivers/vhost/scsi.c index 654c71f..ac11412 100644 --- a/drivers/vhost/scsi.c +++ b/drivers/vhost/scsi.c @@ -67,7 +67,7 @@ struct vhost_scsi_inflight { struct vhost_scsi_cmd { /* Descriptor from vhost_get_vq_desc() for virt_queue segment */ - struct vring_used_elem tvc_vq_used; + struct vhost_used_elem tvc_vq_used; /* virtio-scsi initiator task attribute */ int tvc_task_attr; /* virtio-scsi response incoming iovecs */ @@ -441,7 +441,7 @@ vhost_scsi_do_evt_work(struct vhost_scsi *vs, struct vhost_scsi_evt *evt) struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq; struct virtio_scsi_event *event = &evt->event; struct virtio_scsi_event __user *eventp; - struct vring_used_elem used; + struct vhost_used_elem used; unsigned out, in; int ret; @@ -785,7 +785,7 @@ static void vhost_scsi_submission_work(struct work_struct *work) static void vhost_scsi_send_bad_target(struct vhost_scsi *vs, struct vhost_virtqueue *vq, - struct vring_used_elem *used, unsigned out) + struct vhost_used_elem *used, unsigned out) { struct virtio_scsi_cmd_resp __user *resp; struct virtio_scsi_cmd_resp rsp; @@ -808,7 +808,7 @@ vhost_scsi_handle_vq(struct vhost_scsi *vs, struct vhost_virtqueue *vq) struct virtio_scsi_cmd_req v_req; struct virtio_scsi_cmd_req_pi v_req_pi; struct vhost_scsi_cmd *cmd; - struct vring_used_elem used; + struct vhost_used_elem used; struct iov_iter out_iter, in_iter, prot_iter, data_iter; u64 tag; u32 exp_data_len, data_direction; @@ -837,7 +837,7 @@ vhost_scsi_handle_vq(struct vhost_scsi *vs, struct vhost_virtqueue *vq) ARRAY_SIZE(vq->iov), &out, &in, NULL, NULL); pr_debug("vhost_get_vq_desc: head: %d, out: %u in: %u\n", - used.id, out, in); + used.elem.id, out, in); /* Nothing new? Wait for eventfd to tell us they refilled. */ if (ret == -ENOSPC) { if (unlikely(vhost_enable_notify(&vs->dev, vq))) { diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c index 83cb1dc..8744dae 100644 --- a/drivers/vhost/vhost.c +++ b/drivers/vhost/vhost.c @@ -1956,7 +1956,7 @@ static int get_indirect(struct vhost_virtqueue *vq, * never a valid descriptor number) if none was found. A negative code is * returned on error. */ int vhost_get_vq_desc(struct vhost_virtqueue *vq, - struct vring_used_elem *used, + struct vhost_used_elem *used, struct iovec iov[], unsigned int iov_size, unsigned int *out_num, unsigned int *in_num, struct vhost_log *log, unsigned int *log_num) @@ -2007,7 +2007,7 @@ int vhost_get_vq_desc(struct vhost_virtqueue *vq, return -EFAULT; } - used->id = ring_head; + used->elem.id = ring_head; head = vhost16_to_cpu(vq, ring_head); /* If their number is silly, that's an error. */ @@ -2101,9 +2101,9 @@ int vhost_get_vq_desc(struct vhost_virtqueue *vq, EXPORT_SYMBOL_GPL(vhost_get_vq_desc); static void vhost_set_used_len(struct vhost_virtqueue *vq, - struct vring_used_elem *used, int len) + struct vhost_used_elem *used, int len) { - used->len = cpu_to_vhost32(vq, len); + used->elem.len = cpu_to_vhost32(vq, len); } /* This is a multi-buffer version of vhost_get_desc, that works if @@ -2117,7 +2117,7 @@ static void vhost_set_used_len(struct vhost_virtqueue *vq, * returns number of buffer heads allocated, negative on error */ int vhost_get_bufs(struct vhost_virtqueue *vq, - struct vring_used_elem *heads, + struct vhost_used_elem *heads, int datalen, unsigned *iovcount, struct vhost_log *log, @@ -2190,7 +2190,7 @@ EXPORT_SYMBOL_GPL(vhost_discard_vq_desc); /* After we've used one of their buffers, we tell them about it. We'll then * want to notify the guest, using eventfd. */ -int vhost_add_used(struct vhost_virtqueue *vq, struct vring_used_elem *used, +int vhost_add_used(struct vhost_virtqueue *vq, struct vhost_used_elem *used, int len) { vhost_set_used_len(vq, used, len); @@ -2199,27 +2199,26 @@ int vhost_add_used(struct vhost_virtqueue *vq, struct vring_used_elem *used, EXPORT_SYMBOL_GPL(vhost_add_used); static int __vhost_add_used_n(struct vhost_virtqueue *vq, - struct vring_used_elem *heads, + struct vhost_used_elem *heads, unsigned count) { struct vring_used_elem __user *used; u16 old, new; - int start; + int start, i; start = vq->last_used_idx & (vq->num - 1); used = vq->used->ring + start; - if (count == 1) { - if (vhost_put_user(vq, heads[0].id, &used->id)) { + for (i = 0; i < count; i++) { + if (unlikely(vhost_put_user(vq, heads[i].elem.id, + &used[i].id))) { vq_err(vq, "Failed to write used id"); return -EFAULT; } - if (vhost_put_user(vq, heads[0].len, &used->len)) { + if (unlikely(vhost_put_user(vq, heads[i].elem.len, + &used[i].len))) { vq_err(vq, "Failed to write used len"); return -EFAULT; } - } else if (vhost_copy_to_user(vq, used, heads, count * sizeof *used)) { - vq_err(vq, "Failed to write used"); - return -EFAULT; } if (unlikely(vq->log_used)) { /* Make sure data is seen before log. */ @@ -2243,7 +2242,7 @@ static int __vhost_add_used_n(struct vhost_virtqueue *vq, /* After we've used one of their buffers, we tell them about it. We'll then * want to notify the guest, using eventfd. */ -int vhost_add_used_n(struct vhost_virtqueue *vq, struct vring_used_elem *heads, +int vhost_add_used_n(struct vhost_virtqueue *vq, struct vhost_used_elem *heads, unsigned count) { int start, n, r; @@ -2327,7 +2326,7 @@ EXPORT_SYMBOL_GPL(vhost_signal); /* And here's the combo meal deal. Supersize me! */ void vhost_add_used_and_signal(struct vhost_dev *dev, struct vhost_virtqueue *vq, - struct vring_used_elem *used, int len) + struct vhost_used_elem *used, int len) { vhost_add_used(vq, used, len); vhost_signal(dev, vq); @@ -2337,7 +2336,7 @@ EXPORT_SYMBOL_GPL(vhost_add_used_and_signal); /* multi-buffer version of vhost_add_used_and_signal */ void vhost_add_used_and_signal_n(struct vhost_dev *dev, struct vhost_virtqueue *vq, - struct vring_used_elem *heads, unsigned count) + struct vhost_used_elem *heads, unsigned count) { vhost_add_used_n(vq, heads, count); vhost_signal(dev, vq); diff --git a/drivers/vhost/vhost.h b/drivers/vhost/vhost.h index 477bfd2..8399887 100644 --- a/drivers/vhost/vhost.h +++ b/drivers/vhost/vhost.h @@ -34,6 +34,10 @@ struct vhost_poll { struct vhost_dev *dev; }; +struct vhost_used_elem { + struct vring_used_elem elem; +}; + void vhost_work_init(struct vhost_work *work, vhost_work_fn_t fn); void vhost_work_queue(struct vhost_dev *dev, struct vhost_work *work); bool vhost_has_work(struct vhost_dev *dev); @@ -126,7 +130,7 @@ struct vhost_virtqueue { struct iovec iov[UIO_MAXIOV]; struct iovec iotlb_iov[64]; struct iovec *indirect; - struct vring_used_elem *heads; + struct vhost_used_elem *heads; /* Protected by virtqueue mutex. */ struct vhost_umem *umem; struct vhost_umem *iotlb; @@ -182,12 +186,12 @@ int vhost_vq_access_ok(struct vhost_virtqueue *vq); int vhost_log_access_ok(struct vhost_dev *); int vhost_get_vq_desc(struct vhost_virtqueue *, - struct vring_used_elem *used_elem, + struct vhost_used_elem *used_elem, struct iovec iov[], unsigned int iov_count, unsigned int *out_num, unsigned int *in_num, struct vhost_log *log, unsigned int *log_num); int vhost_get_bufs(struct vhost_virtqueue *vq, - struct vring_used_elem *heads, + struct vhost_used_elem *heads, int datalen, unsigned *iovcount, struct vhost_log *log, @@ -198,13 +202,13 @@ void vhost_discard_vq_desc(struct vhost_virtqueue *, int n); int vhost_vq_init_access(struct vhost_virtqueue *); int vhost_add_used(struct vhost_virtqueue *vq, - struct vring_used_elem *elem, int len); -int vhost_add_used_n(struct vhost_virtqueue *, struct vring_used_elem *heads, + struct vhost_used_elem *elem, int len); +int vhost_add_used_n(struct vhost_virtqueue *vq, struct vhost_used_elem *heads, unsigned count); void vhost_add_used_and_signal(struct vhost_dev *, struct vhost_virtqueue *, - struct vring_used_elem *, int len); + struct vhost_used_elem *, int len); void vhost_add_used_and_signal_n(struct vhost_dev *, struct vhost_virtqueue *, - struct vring_used_elem *heads, unsigned count); + struct vhost_used_elem *heads, unsigned count); void vhost_signal(struct vhost_dev *, struct vhost_virtqueue *); void vhost_disable_notify(struct vhost_dev *, struct vhost_virtqueue *); bool vhost_vq_avail_empty(struct vhost_dev *, struct vhost_virtqueue *); diff --git a/drivers/vhost/vsock.c b/drivers/vhost/vsock.c index fc20292..5d36208 100644 --- a/drivers/vhost/vsock.c +++ b/drivers/vhost/vsock.c @@ -98,7 +98,7 @@ vhost_transport_do_send_pkt(struct vhost_vsock *vsock, for (;;) { struct virtio_vsock_pkt *pkt; - struct vring_used_elem used; + struct vhost_used_elem used; struct iov_iter iov_iter; unsigned out, in; size_t nbytes; @@ -146,7 +146,7 @@ vhost_transport_do_send_pkt(struct vhost_vsock *vsock, break; } - len = vhost32_to_cpu(vq, used.len); + len = vhost32_to_cpu(vq, used.elem.len); iov_iter_init(&iov_iter, READ, &vq->iov[out], in, len); nbytes = copy_to_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter); @@ -346,7 +346,7 @@ static void vhost_vsock_handle_tx_kick(struct vhost_work *work) struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock, dev); struct virtio_vsock_pkt *pkt; - struct vring_used_elem used; + struct vhost_used_elem used; int ret; unsigned int out, in; bool added = false; -- 2.7.4
Jason Wang
2018-Mar-26 03:38 UTC
[RFC PATCH V2 4/8] vhost_net: do not explicitly manipulate vhost_used_elem
Two helpers of setting/getting used len were introduced to avoid explicitly manipulating vhost_used_elem in zerocopy code. This will be used to hide used_elem internals and simplify packed ring implementation. Signed-off-by: Jason Wang <jasowang at redhat.com> --- drivers/vhost/net.c | 11 +++++------ drivers/vhost/vhost.c | 12 ++++++++++-- drivers/vhost/vhost.h | 5 +++++ 3 files changed, 20 insertions(+), 8 deletions(-) diff --git a/drivers/vhost/net.c b/drivers/vhost/net.c index 7ea2aee..7be8b55 100644 --- a/drivers/vhost/net.c +++ b/drivers/vhost/net.c @@ -337,9 +337,10 @@ static void vhost_zerocopy_signal_used(struct vhost_net *net, int j = 0; for (i = nvq->done_idx; i != nvq->upend_idx; i = (i + 1) % UIO_MAXIOV) { - if (vq->heads[i].elem.len == VHOST_DMA_FAILED_LEN) + if (vhost_get_used_len(vq, &vq->heads[i]) =+ VHOST_DMA_FAILED_LEN) vhost_net_tx_err(net); - if (VHOST_DMA_IS_DONE(vq->heads[i].elem.len)) { + if (VHOST_DMA_IS_DONE(vhost_get_used_len(vq, &vq->heads[i]))) { vq->heads[i].elem.len = VHOST_DMA_CLEAR_LEN; ++j; } else @@ -537,10 +538,8 @@ static void handle_tx(struct vhost_net *net) struct ubuf_info *ubuf; ubuf = nvq->ubuf_info + nvq->upend_idx; - vq->heads[nvq->upend_idx].elem.id - cpu_to_vhost32(vq, used.elem.id); - vq->heads[nvq->upend_idx].elem.len - VHOST_DMA_IN_PROGRESS; + vhost_set_used_len(vq, &used, VHOST_DMA_IN_PROGRESS); + vq->heads[nvq->upend_idx] = used; ubuf->callback = vhost_zerocopy_callback; ubuf->ctx = nvq->ubufs; ubuf->desc = nvq->upend_idx; diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c index 8744dae..65954d6 100644 --- a/drivers/vhost/vhost.c +++ b/drivers/vhost/vhost.c @@ -2100,11 +2100,19 @@ int vhost_get_vq_desc(struct vhost_virtqueue *vq, } EXPORT_SYMBOL_GPL(vhost_get_vq_desc); -static void vhost_set_used_len(struct vhost_virtqueue *vq, - struct vhost_used_elem *used, int len) +void vhost_set_used_len(struct vhost_virtqueue *vq, + struct vhost_used_elem *used, int len) { used->elem.len = cpu_to_vhost32(vq, len); } +EXPORT_SYMBOL_GPL(vhost_set_used_len); + +int vhost_get_used_len(struct vhost_virtqueue *vq, + struct vhost_used_elem *used) +{ + return vhost32_to_cpu(vq, used->elem.len); +} +EXPORT_SYMBOL_GPL(vhost_get_used_len); /* This is a multi-buffer version of vhost_get_desc, that works if * vq has read descriptors only. diff --git a/drivers/vhost/vhost.h b/drivers/vhost/vhost.h index 8399887..d57c875 100644 --- a/drivers/vhost/vhost.h +++ b/drivers/vhost/vhost.h @@ -198,6 +198,11 @@ int vhost_get_bufs(struct vhost_virtqueue *vq, unsigned *log_num, unsigned int quota, s16 *count); +void vhost_set_used_len(struct vhost_virtqueue *vq, + struct vhost_used_elem *used, + int len); +int vhost_get_used_len(struct vhost_virtqueue *vq, + struct vhost_used_elem *used); void vhost_discard_vq_desc(struct vhost_virtqueue *, int n); int vhost_vq_init_access(struct vhost_virtqueue *); -- 2.7.4
Jason Wang
2018-Mar-26 03:38 UTC
[RFC PATCH V2 5/8] vhost: vhost_put_user() can accept metadata type
We assumes used ring update is the only user for vhost_put_user() in the past. This may not be the case for the incoming packed ring which may update the descriptor ring for used. So introduce a new type parameter. Signed-off-by: Jason Wang <jasowang at redhat.com> --- drivers/vhost/vhost.c | 14 +++++++------- 1 file changed, 7 insertions(+), 7 deletions(-) diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c index 65954d6..dcac4d4 100644 --- a/drivers/vhost/vhost.c +++ b/drivers/vhost/vhost.c @@ -847,7 +847,7 @@ static inline void __user *__vhost_get_user(struct vhost_virtqueue *vq, return __vhost_get_user_slow(vq, addr, size, type); } -#define vhost_put_user(vq, x, ptr) \ +#define vhost_put_user(vq, x, ptr, type) \ ({ \ int ret = -EFAULT; \ if (!vq->iotlb) { \ @@ -855,7 +855,7 @@ static inline void __user *__vhost_get_user(struct vhost_virtqueue *vq, } else { \ __typeof__(ptr) to = \ (__typeof__(ptr)) __vhost_get_user(vq, ptr, \ - sizeof(*ptr), VHOST_ADDR_USED); \ + sizeof(*ptr), type); \ if (to != NULL) \ ret = __put_user(x, to); \ else \ @@ -1716,7 +1716,7 @@ static int vhost_update_used_flags(struct vhost_virtqueue *vq) { void __user *used; if (vhost_put_user(vq, cpu_to_vhost16(vq, vq->used_flags), - &vq->used->flags) < 0) + &vq->used->flags, VHOST_ADDR_USED) < 0) return -EFAULT; if (unlikely(vq->log_used)) { /* Make sure the flag is seen before log. */ @@ -1735,7 +1735,7 @@ static int vhost_update_used_flags(struct vhost_virtqueue *vq) static int vhost_update_avail_event(struct vhost_virtqueue *vq, u16 avail_event) { if (vhost_put_user(vq, cpu_to_vhost16(vq, vq->avail_idx), - vhost_avail_event(vq))) + vhost_avail_event(vq), VHOST_ADDR_USED)) return -EFAULT; if (unlikely(vq->log_used)) { void __user *used; @@ -2218,12 +2218,12 @@ static int __vhost_add_used_n(struct vhost_virtqueue *vq, used = vq->used->ring + start; for (i = 0; i < count; i++) { if (unlikely(vhost_put_user(vq, heads[i].elem.id, - &used[i].id))) { + &used[i].id, VHOST_ADDR_USED))) { vq_err(vq, "Failed to write used id"); return -EFAULT; } if (unlikely(vhost_put_user(vq, heads[i].elem.len, - &used[i].len))) { + &used[i].len, VHOST_ADDR_USED))) { vq_err(vq, "Failed to write used len"); return -EFAULT; } @@ -2269,7 +2269,7 @@ int vhost_add_used_n(struct vhost_virtqueue *vq, struct vhost_used_elem *heads, /* Make sure buffer is written before we update index. */ smp_wmb(); if (vhost_put_user(vq, cpu_to_vhost16(vq, vq->last_used_idx), - &vq->used->idx)) { + &vq->used->idx, VHOST_ADDR_USED)) { vq_err(vq, "Failed to increment used idx"); return -EFAULT; } -- 2.7.4
Signed-off-by: Jason Wang <jasowang at redhat.com> --- include/uapi/linux/virtio_config.h | 9 +++++++++ include/uapi/linux/virtio_ring.h | 13 +++++++++++++ 2 files changed, 22 insertions(+) diff --git a/include/uapi/linux/virtio_config.h b/include/uapi/linux/virtio_config.h index 308e209..5903d51 100644 --- a/include/uapi/linux/virtio_config.h +++ b/include/uapi/linux/virtio_config.h @@ -71,4 +71,13 @@ * this is for compatibility with legacy systems. */ #define VIRTIO_F_IOMMU_PLATFORM 33 + +#define VIRTIO_F_RING_PACKED 34 + +/* + * This feature indicates that all buffers are used by the device in + * the same order in which they have been made available. + */ +#define VIRTIO_F_IN_ORDER 35 + #endif /* _UAPI_LINUX_VIRTIO_CONFIG_H */ diff --git a/include/uapi/linux/virtio_ring.h b/include/uapi/linux/virtio_ring.h index 6d5d5fa..e297580 100644 --- a/include/uapi/linux/virtio_ring.h +++ b/include/uapi/linux/virtio_ring.h @@ -43,6 +43,8 @@ #define VRING_DESC_F_WRITE 2 /* This means the buffer contains a list of buffer descriptors. */ #define VRING_DESC_F_INDIRECT 4 +#define VRING_DESC_F_AVAIL 7 +#define VRING_DESC_F_USED 15 /* The Host uses this in used->flags to advise the Guest: don't kick me when * you add a buffer. It's unreliable, so it's simply an optimization. Guest @@ -62,6 +64,17 @@ * at the end of the used ring. Guest should ignore the used->flags field. */ #define VIRTIO_RING_F_EVENT_IDX 29 +struct vring_desc_packed { + /* Buffer Address. */ + __virtio64 addr; + /* Buffer Length. */ + __virtio32 len; + /* Buffer ID. */ + __virtio16 id; + /* The flags depending on descriptor type. */ + __virtio16 flags; +}; + /* Virtio ring descriptors: 16 bytes. These can chain together via "next". */ struct vring_desc { /* Address (guest-physical). */ -- 2.7.4
Signed-off-by: Jason Wang <jasowang at redhat.com> --- drivers/vhost/net.c | 5 +- drivers/vhost/vhost.c | 530 ++++++++++++++++++++++++++++++++++++++++++++++---- drivers/vhost/vhost.h | 7 +- 3 files changed, 505 insertions(+), 37 deletions(-) diff --git a/drivers/vhost/net.c b/drivers/vhost/net.c index 7be8b55..84905d5 100644 --- a/drivers/vhost/net.c +++ b/drivers/vhost/net.c @@ -67,7 +67,8 @@ enum { VHOST_NET_FEATURES = VHOST_FEATURES | (1ULL << VHOST_NET_F_VIRTIO_NET_HDR) | (1ULL << VIRTIO_NET_F_MRG_RXBUF) | - (1ULL << VIRTIO_F_IOMMU_PLATFORM) + (1ULL << VIRTIO_F_IOMMU_PLATFORM) | + (1ULL << VIRTIO_F_RING_PACKED) }; enum { @@ -706,6 +707,8 @@ static void handle_rx(struct vhost_net *net) vq_log = unlikely(vhost_has_feature(vq, VHOST_F_LOG_ALL)) ? vq->log : NULL; mergeable = vhost_has_feature(vq, VIRTIO_NET_F_MRG_RXBUF); + /* FIXME: workaround for current dpdk prototype */ + mergeable = false; while ((sock_len = vhost_net_rx_peek_head_len(net, sock->sk))) { sock_len += sock_hlen; diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c index dcac4d4..6177e4d 100644 --- a/drivers/vhost/vhost.c +++ b/drivers/vhost/vhost.c @@ -324,6 +324,7 @@ static void vhost_vq_reset(struct vhost_dev *dev, vhost_reset_is_le(vq); vhost_disable_cross_endian(vq); vq->busyloop_timeout = 0; + vq->used_wrap_counter = true; vq->umem = NULL; vq->iotlb = NULL; __vhost_vq_meta_reset(vq); @@ -1136,10 +1137,22 @@ static int vhost_iotlb_miss(struct vhost_virtqueue *vq, u64 iova, int access) return 0; } -static int vq_access_ok(struct vhost_virtqueue *vq, unsigned int num, - struct vring_desc __user *desc, - struct vring_avail __user *avail, - struct vring_used __user *used) +static int vq_access_ok_packed(struct vhost_virtqueue *vq, unsigned int num, + struct vring_desc __user *desc, + struct vring_avail __user *avail, + struct vring_used __user *used) +{ + struct vring_desc_packed *packed = (struct vring_desc_packed *)desc; + + /* FIXME: check device area and driver area */ + return access_ok(VERIFY_READ, packed, num * sizeof(*packed)) && + access_ok(VERIFY_WRITE, packed, num * sizeof(*packed)); +} + +static int vq_access_ok_split(struct vhost_virtqueue *vq, unsigned int num, + struct vring_desc __user *desc, + struct vring_avail __user *avail, + struct vring_used __user *used) { size_t s = vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0; @@ -1151,6 +1164,17 @@ static int vq_access_ok(struct vhost_virtqueue *vq, unsigned int num, sizeof *used + num * sizeof *used->ring + s); } +static int vq_access_ok(struct vhost_virtqueue *vq, unsigned int num, + struct vring_desc __user *desc, + struct vring_avail __user *avail, + struct vring_used __user *used) +{ + if (vhost_has_feature(vq, VIRTIO_F_RING_PACKED)) + return vq_access_ok_packed(vq, num, desc, avail, used); + else + return vq_access_ok_split(vq, num, desc, avail, used); +} + static void vhost_vq_meta_update(struct vhost_virtqueue *vq, const struct vhost_umem_node *node, int type) @@ -1763,6 +1787,9 @@ int vhost_vq_init_access(struct vhost_virtqueue *vq) vhost_init_is_le(vq); + if (vhost_has_feature(vq, VIRTIO_F_RING_PACKED)) + return 0; + r = vhost_update_used_flags(vq); if (r) goto err; @@ -1836,7 +1863,8 @@ static int translate_desc(struct vhost_virtqueue *vq, u64 addr, u32 len, /* Each buffer in the virtqueues is actually a chain of descriptors. This * function returns the next descriptor in the chain, * or -1U if we're at the end. */ -static unsigned next_desc(struct vhost_virtqueue *vq, struct vring_desc *desc) +static unsigned next_desc_split(struct vhost_virtqueue *vq, + struct vring_desc *desc) { unsigned int next; @@ -1849,11 +1877,17 @@ static unsigned next_desc(struct vhost_virtqueue *vq, struct vring_desc *desc) return next; } -static int get_indirect(struct vhost_virtqueue *vq, - struct iovec iov[], unsigned int iov_size, - unsigned int *out_num, unsigned int *in_num, - struct vhost_log *log, unsigned int *log_num, - struct vring_desc *indirect) +static unsigned next_desc_packed(struct vhost_virtqueue *vq, + struct vring_desc_packed *desc) +{ + return desc->flags & cpu_to_vhost16(vq, VRING_DESC_F_NEXT); +} + +static int get_indirect_split(struct vhost_virtqueue *vq, + struct iovec iov[], unsigned int iov_size, + unsigned int *out_num, unsigned int *in_num, + struct vhost_log *log, unsigned int *log_num, + struct vring_desc *indirect) { struct vring_desc desc; unsigned int i = 0, count, found = 0; @@ -1943,23 +1977,274 @@ static int get_indirect(struct vhost_virtqueue *vq, } *out_num += ret; } - } while ((i = next_desc(vq, &desc)) != -1); + } while ((i = next_desc_split(vq, &desc)) != -1); return 0; } -/* This looks in the virtqueue and for the first available buffer, and converts - * it to an iovec for convenient access. Since descriptors consist of some - * number of output then some number of input descriptors, it's actually two - * iovecs, but we pack them into one and note how many of each there were. - * - * This function returns the descriptor number found, or vq->num (which is - * never a valid descriptor number) if none was found. A negative code is - * returned on error. */ -int vhost_get_vq_desc(struct vhost_virtqueue *vq, - struct vhost_used_elem *used, - struct iovec iov[], unsigned int iov_size, - unsigned int *out_num, unsigned int *in_num, - struct vhost_log *log, unsigned int *log_num) +static int get_indirect_packed(struct vhost_virtqueue *vq, + struct iovec iov[], unsigned int iov_size, + unsigned int *out_num, unsigned int *in_num, + struct vhost_log *log, unsigned int *log_num, + struct vring_desc_packed *indirect) +{ + struct vring_desc_packed desc; + unsigned int i = 0, count, found = 0; + u32 len = vhost32_to_cpu(vq, indirect->len); + struct iov_iter from; + int ret, access; + + /* Sanity check */ + if (unlikely(len % sizeof(desc))) { + vq_err(vq, "Invalid length in indirect descriptor: " + "len 0x%llx not multiple of 0x%zx\n", + (unsigned long long)len, + sizeof desc); + return -EINVAL; + } + + ret = translate_desc(vq, vhost64_to_cpu(vq, indirect->addr), + len, vq->indirect, + UIO_MAXIOV, VHOST_ACCESS_RO); + if (unlikely(ret < 0)) { + if (ret != -EAGAIN) + vq_err(vq, "Translation failure %d in indirect.\n", + ret); + return ret; + } + iov_iter_init(&from, READ, vq->indirect, ret, len); + + /* We will use the result as an address to read from, so most + * architectures only need a compiler barrier here. */ + read_barrier_depends(); + + count = len / sizeof desc; + /* Buffers are chained via a 16 bit next field, so + * we can have at most 2^16 of these. */ + if (unlikely(count > USHRT_MAX + 1)) { + vq_err(vq, "Indirect buffer length too big: %d\n", + indirect->len); + return -E2BIG; + } + + do { + unsigned iov_count = *in_num + *out_num; + if (unlikely(++found > count)) { + vq_err(vq, "Loop detected: last one at %u " + "indirect size %u\n", + i, count); + return -EINVAL; + } + if (unlikely(!copy_from_iter_full(&desc, sizeof(desc), + &from))) { + vq_err(vq, "Failed indirect descriptor: idx %d, %zx\n", + i, (size_t)vhost64_to_cpu(vq, indirect->addr) + + i * sizeof desc); + return -EINVAL; + } + if (unlikely(desc.flags & + cpu_to_vhost16(vq, VRING_DESC_F_INDIRECT))) { + vq_err(vq, "Nested indirect descriptor: idx %d, %zx\n", + i, (size_t)vhost64_to_cpu(vq, indirect->addr) + + i * sizeof desc); + return -EINVAL; + } + + if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_WRITE)) + access = VHOST_ACCESS_WO; + else + access = VHOST_ACCESS_RO; + + ret = translate_desc(vq, vhost64_to_cpu(vq, desc.addr), + vhost32_to_cpu(vq, desc.len), + iov + iov_count, + iov_size - iov_count, access); + if (unlikely(ret < 0)) { + if (ret != -EAGAIN) + vq_err(vq, "Translation failure %d " + "indirect idx %d\n", + ret, i); + return ret; + } + /* If this is an input descriptor, increment that count. */ + if (access == VHOST_ACCESS_WO) { + *in_num += ret; + if (unlikely(log)) { + log[*log_num].addr + vhost64_to_cpu(vq, desc.addr); + log[*log_num].len + vhost32_to_cpu(vq, desc.len); + ++*log_num; + } + } else { + /* If it's an output descriptor, they're all supposed + * to come before any input descriptors. */ + if (unlikely(*in_num)) { + vq_err(vq, "Indirect descriptor " + "has out after in: idx %d\n", i); + return -EINVAL; + } + *out_num += ret; + } + } while ((i = next_desc_packed(vq, &desc)) != -1); + return 0; +} + +#define DESC_AVAIL (1 << VRING_DESC_F_AVAIL) +#define DESC_USED (1 << VRING_DESC_F_USED) +static bool desc_is_avail(struct vhost_virtqueue *vq, __virtio16 flags) +{ + if (vq->used_wrap_counter) + if ((flags & cpu_to_vhost16(vq, DESC_AVAIL)) && + !(flags & cpu_to_vhost16(vq, DESC_USED))) + return true; + if (vq->used_wrap_counter == false) + if (!(flags & cpu_to_vhost16(vq, DESC_AVAIL)) && + (flags & cpu_to_vhost16(vq, DESC_USED))) + return true; + + return false; +} + +static __virtio16 get_desc_flags(struct vhost_virtqueue *vq) +{ + __virtio16 flags = 0; + + if (vq->used_wrap_counter) { + flags |= cpu_to_vhost16(vq, DESC_AVAIL); + flags |= cpu_to_vhost16(vq, DESC_USED); + } else { + flags &= ~cpu_to_vhost16(vq, DESC_AVAIL); + flags &= ~cpu_to_vhost16(vq, DESC_USED); + } + + return flags; +} + +static int vhost_get_vq_desc_packed(struct vhost_virtqueue *vq, + struct vhost_used_elem *used, + struct iovec iov[], unsigned int iov_size, + unsigned int *out_num, unsigned int *in_num, + struct vhost_log *log, + unsigned int *log_num) +{ + struct vring_desc_packed desc; + int ret, access, i; + + /* When we start there are none of either input nor output. */ + *out_num = *in_num = 0; + if (unlikely(log)) + *log_num = 0; + + used->count = 0; + + do { + struct vring_desc_packed *d = vq->desc_packed + + vq->last_avail_idx; + unsigned int iov_count = *in_num + *out_num; + + ret = vhost_get_user(vq, desc.flags, &d->flags, + VHOST_ADDR_DESC); + if (unlikely(ret)) { + vq_err(vq, "Failed to get flags: idx %d addr %p\n", + vq->last_avail_idx, &d->flags); + return -EFAULT; + } + + if (!desc_is_avail(vq, desc.flags)) { + /* If there's nothing new since last we looked, return + * invalid. + */ + if (!used->count) + return -ENOSPC; + break; + } + + /* Read desc content after we're sure it was available. */ + smp_rmb(); + + ret = vhost_copy_from_user(vq, &desc, d, sizeof(desc)); + if (unlikely(ret)) { + vq_err(vq, "Failed to get descriptor: idx %d addr %p\n", + vq->last_avail_idx, d); + return -EFAULT; + } + + if (used->count == 0) + used->elem.id = desc.id; + + if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_INDIRECT)) { + ret = get_indirect_packed(vq, iov, iov_size, + out_num, in_num, log, + log_num, &desc); + if (unlikely(ret < 0)) { + if (ret != -EAGAIN) + vq_err(vq, "Failure detected " + "in indirect descriptor " + "at idx %d\n", i); + return ret; + } + goto next; + } + + if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_WRITE)) + access = VHOST_ACCESS_WO; + else + access = VHOST_ACCESS_RO; + ret = translate_desc(vq, vhost64_to_cpu(vq, desc.addr), + vhost32_to_cpu(vq, desc.len), + iov + iov_count, iov_size - iov_count, + access); + if (unlikely(ret < 0)) { + if (ret != -EAGAIN) + vq_err(vq, "Translation failure %d idx %d\n", + ret, i); + return ret; + } + + if (access == VHOST_ACCESS_WO) { + /* If this is an input descriptor, + * increment that count. + */ + *in_num += ret; + if (unlikely(log)) { + log[*log_num].addr + vhost64_to_cpu(vq, desc.addr); + log[*log_num].len + vhost32_to_cpu(vq, desc.len); + ++*log_num; + } + } else { + /* If it's an output descriptor, they're all supposed + * to come before any input descriptors. + */ + if (unlikely(*in_num)) { + vq_err(vq, "Desc out after in: idx %d\n", + i); + return -EINVAL; + } + *out_num += ret; + } + +next: + if (unlikely(++used->count > vq->num)) { + vq_err(vq, "Loop detected: last one at %u " + "vq size %u head %u\n", + i, vq->num, used->elem.id); + return -EINVAL; + } + if (++vq->last_avail_idx >= vq->num) + vq->last_avail_idx = 0; + /* If this descriptor says it doesn't chain, we're done. */ + } while (next_desc_packed(vq, &desc)); + + return 0; +} + +static int vhost_get_vq_desc_split(struct vhost_virtqueue *vq, + struct vhost_used_elem *used, + struct iovec iov[], unsigned int iov_size, + unsigned int *out_num, unsigned int *in_num, + struct vhost_log *log, unsigned int *log_num) { struct vring_desc desc; unsigned int i, head, found = 0; @@ -2044,9 +2329,9 @@ int vhost_get_vq_desc(struct vhost_virtqueue *vq, return -EFAULT; } if (desc.flags & cpu_to_vhost16(vq, VRING_DESC_F_INDIRECT)) { - ret = get_indirect(vq, iov, iov_size, - out_num, in_num, - log, log_num, &desc); + ret = get_indirect_split(vq, iov, iov_size, + out_num, in_num, + log, log_num, &desc); if (unlikely(ret < 0)) { if (ret != -EAGAIN) vq_err(vq, "Failure detected " @@ -2088,7 +2373,7 @@ int vhost_get_vq_desc(struct vhost_virtqueue *vq, } *out_num += ret; } - } while ((i = next_desc(vq, &desc)) != -1); + } while ((i = next_desc_split(vq, &desc)) != -1); /* On success, increment avail index. */ vq->last_avail_idx++; @@ -2098,6 +2383,31 @@ int vhost_get_vq_desc(struct vhost_virtqueue *vq, BUG_ON(!(vq->used_flags & VRING_USED_F_NO_NOTIFY)); return 0; } + +/* This looks in the virtqueue and for the first available buffer, and converts + * it to an iovec for convenient access. Since descriptors consist of some + * number of output then some number of input descriptors, it's actually two + * iovecs, but we pack them into one and note how many of each there were. + * + * This function returns the descriptor number found, or vq->num (which is + * never a valid descriptor number) if none was found. A negative code is + * returned on error. + */ +int vhost_get_vq_desc(struct vhost_virtqueue *vq, + struct vhost_used_elem *used, + struct iovec iov[], unsigned int iov_size, + unsigned int *out_num, unsigned int *in_num, + struct vhost_log *log, unsigned int *log_num) +{ + if (vhost_has_feature(vq, VIRTIO_F_RING_PACKED)) + return vhost_get_vq_desc_packed(vq, used, iov, iov_size, + out_num, in_num, + log, log_num); + else + return vhost_get_vq_desc_split(vq, used, iov, iov_size, + out_num, in_num, + log, log_num); +} EXPORT_SYMBOL_GPL(vhost_get_vq_desc); void vhost_set_used_len(struct vhost_virtqueue *vq, @@ -2248,10 +2558,69 @@ static int __vhost_add_used_n(struct vhost_virtqueue *vq, return 0; } +static int vhost_add_used_n_packed(struct vhost_virtqueue *vq, + struct vhost_used_elem *heads, + unsigned int count) +{ + struct vring_desc_packed __user *desc; + int i, ret; + + for (i = 0; i < count; i++) { + desc = vq->desc_packed + vq->last_used_idx; + + ret = vhost_put_user(vq, heads[i].elem.id, &desc->id, + VHOST_ADDR_DESC); + if (unlikely(ret)) { + vq_err(vq, "Failed to update id: idx %d addr %p\n", + vq->last_used_idx, desc); + return -EFAULT; + } + ret = vhost_put_user(vq, heads[i].elem.len, &desc->len, + VHOST_ADDR_DESC); + if (unlikely(ret)) { + vq_err(vq, "Failed to update len: idx %d addr %p\n", + vq->last_used_idx, desc); + return -EFAULT; + } + + /* Update flags after descriptor id and len is wrote, + * TODO: Update head flags at last for saving barriers */ + smp_wmb(); + + ret = vhost_put_user(vq, get_desc_flags(vq), &desc->flags, + VHOST_ADDR_DESC); + if (unlikely(ret)) { + vq_err(vq, "Failed to update flags: idx %d addr %p\n", + vq->last_used_idx, desc); + return -EFAULT; + } + + if (unlikely(vq->log_used)) { + /* Make sure desc is written before update log. */ + smp_wmb(); + log_write(vq->log_base, vq->log_addr + + vq->last_used_idx * sizeof(*desc), + sizeof(*desc)); + if (vq->log_ctx) + eventfd_signal(vq->log_ctx, 1); + } + + vq->last_used_idx += heads[i].count; + if (vq->last_used_idx >= vq->num) { + vq->used_wrap_counter ^= 1; + vq->last_used_idx -= vq->num; + } + } + + return 0; +} + /* After we've used one of their buffers, we tell them about it. We'll then * want to notify the guest, using eventfd. */ -int vhost_add_used_n(struct vhost_virtqueue *vq, struct vhost_used_elem *heads, - unsigned count) +static int vhost_add_used_n_split(struct vhost_virtqueue *vq, + struct vhost_used_elem *heads, + unsigned count) + { int start, n, r; @@ -2283,6 +2652,19 @@ int vhost_add_used_n(struct vhost_virtqueue *vq, struct vhost_used_elem *heads, } return r; } + +/* After we've used one of their buffers, we tell them about it. We'll then + * want to notify the guest, using eventfd. + */ +int vhost_add_used_n(struct vhost_virtqueue *vq, + struct vhost_used_elem *heads, + unsigned int count) +{ + if (vhost_has_feature(vq, VIRTIO_F_RING_PACKED)) + return vhost_add_used_n_packed(vq, heads, count); + else + return vhost_add_used_n_split(vq, heads, count); +} EXPORT_SYMBOL_GPL(vhost_add_used_n); static bool vhost_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq) @@ -2290,6 +2672,11 @@ static bool vhost_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq) __u16 old, new; __virtio16 event; bool v; + + /* FIXME: check driver area */ + if (vhost_has_feature(vq, VIRTIO_F_RING_PACKED)) + return false; + /* Flush out used index updates. This is paired * with the barrier that the Guest executes when enabling * interrupts. */ @@ -2352,7 +2739,8 @@ void vhost_add_used_and_signal_n(struct vhost_dev *dev, EXPORT_SYMBOL_GPL(vhost_add_used_and_signal_n); /* return true if we're sure that avaiable ring is empty */ -bool vhost_vq_avail_empty(struct vhost_dev *dev, struct vhost_virtqueue *vq) +static bool vhost_vq_avail_empty_split(struct vhost_dev *dev, + struct vhost_virtqueue *vq) { __virtio16 avail_idx; int r; @@ -2367,10 +2755,58 @@ bool vhost_vq_avail_empty(struct vhost_dev *dev, struct vhost_virtqueue *vq) return vq->avail_idx == vq->last_avail_idx; } + +static bool vhost_vq_avail_empty_packed(struct vhost_dev *dev, + struct vhost_virtqueue *vq) +{ + struct vring_desc_packed *d = vq->desc_packed + vq->last_avail_idx; + __virtio16 flags; + int ret; + + ret = vhost_get_user(vq, flags, &d->flags, VHOST_ADDR_DESC); + if (unlikely(ret)) { + vq_err(vq, "Failed to get flags: idx %d addr %p\n", + vq->last_avail_idx, d); + return -EFAULT; + } + + return desc_is_avail(vq, flags); +} + +bool vhost_vq_avail_empty(struct vhost_dev *dev, struct vhost_virtqueue *vq) +{ + if (vhost_has_feature(vq, VIRTIO_F_RING_PACKED)) + return vhost_vq_avail_empty_packed(dev, vq); + else + return vhost_vq_avail_empty_split(dev, vq); +} EXPORT_SYMBOL_GPL(vhost_vq_avail_empty); -/* OK, now we need to know about added descriptors. */ -bool vhost_enable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq) +static bool vhost_enable_notify_packed(struct vhost_dev *dev, + struct vhost_virtqueue *vq) +{ + struct vring_desc_packed *d = vq->desc_packed + vq->last_avail_idx; + __virtio16 flags; + int ret; + + /* FIXME: disable notification through device area */ + + /* They could have slipped one in as we were doing that: make + * sure it's written, then check again. */ + smp_mb(); + + ret = vhost_get_user(vq, flags, &d->flags, VHOST_ADDR_DESC); + if (unlikely(ret)) { + vq_err(vq, "Failed to get descriptor: idx %d addr %p\n", + vq->last_avail_idx, &d->flags); + return -EFAULT; + } + + return desc_is_avail(vq, flags); +} + +static bool vhost_enable_notify_split(struct vhost_dev *dev, + struct vhost_virtqueue *vq) { __virtio16 avail_idx; int r; @@ -2405,10 +2841,25 @@ bool vhost_enable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq) return vhost16_to_cpu(vq, avail_idx) != vq->avail_idx; } + +/* OK, now we need to know about added descriptors. */ +bool vhost_enable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq) +{ + if (vhost_has_feature(vq, VIRTIO_F_RING_PACKED)) + return vhost_enable_notify_packed(dev, vq); + else + return vhost_enable_notify_split(dev, vq); +} EXPORT_SYMBOL_GPL(vhost_enable_notify); -/* We don't need to be notified again. */ -void vhost_disable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq) +static void vhost_disable_notify_packed(struct vhost_dev *dev, + struct vhost_virtqueue *vq) +{ + /* FIXME: disable notification through device area */ +} + +static void vhost_disable_notify_split(struct vhost_dev *dev, + struct vhost_virtqueue *vq) { int r; @@ -2422,6 +2873,15 @@ void vhost_disable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq) &vq->used->flags, r); } } + +/* We don't need to be notified again. */ +void vhost_disable_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq) +{ + if (vhost_has_feature(vq, VIRTIO_F_RING_PACKED)) + return vhost_disable_notify_packed(dev, vq); + else + return vhost_disable_notify_split(dev, vq); +} EXPORT_SYMBOL_GPL(vhost_disable_notify); /* Create a new message. */ diff --git a/drivers/vhost/vhost.h b/drivers/vhost/vhost.h index d57c875..8a9df4f 100644 --- a/drivers/vhost/vhost.h +++ b/drivers/vhost/vhost.h @@ -36,6 +36,7 @@ struct vhost_poll { struct vhost_used_elem { struct vring_used_elem elem; + int count; }; void vhost_work_init(struct vhost_work *work, vhost_work_fn_t fn); @@ -91,7 +92,10 @@ struct vhost_virtqueue { /* The actual ring of buffers. */ struct mutex mutex; unsigned int num; - struct vring_desc __user *desc; + union { + struct vring_desc __user *desc; + struct vring_desc_packed __user *desc_packed; + }; struct vring_avail __user *avail; struct vring_used __user *used; const struct vhost_umem_node *meta_iotlb[VHOST_NUM_ADDRS]; @@ -148,6 +152,7 @@ struct vhost_virtqueue { bool user_be; #endif u32 busyloop_timeout; + bool used_wrap_counter; }; struct vhost_msg_node { -- 2.7.4
Jason Wang
2018-Mar-26 03:38 UTC
[RFC PATCH V2 8/8] vhost: event suppression for packed ring
This patch introduces basic support for event suppression aka driver and device area. Compile tested only. Signed-off-by: Jason Wang <jasowang at redhat.com> --- drivers/vhost/vhost.c | 169 ++++++++++++++++++++++++++++++++++++--- drivers/vhost/vhost.h | 10 ++- include/uapi/linux/virtio_ring.h | 19 +++++ 3 files changed, 183 insertions(+), 15 deletions(-) diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c index 6177e4d..ff83a2e 100644 --- a/drivers/vhost/vhost.c +++ b/drivers/vhost/vhost.c @@ -1143,10 +1143,15 @@ static int vq_access_ok_packed(struct vhost_virtqueue *vq, unsigned int num, struct vring_used __user *used) { struct vring_desc_packed *packed = (struct vring_desc_packed *)desc; + struct vring_packed_desc_event *driver_event + (struct vring_packed_desc_event *)avail; + struct vring_packed_desc_event *device_event + (struct vring_packed_desc_event *)used; - /* FIXME: check device area and driver area */ return access_ok(VERIFY_READ, packed, num * sizeof(*packed)) && - access_ok(VERIFY_WRITE, packed, num * sizeof(*packed)); + access_ok(VERIFY_WRITE, packed, num * sizeof(*packed)) && + access_ok(VERIFY_READ, driver_event, sizeof(*driver_event)) && + access_ok(VERIFY_WRITE, device_event, sizeof(*device_event)); } static int vq_access_ok_split(struct vhost_virtqueue *vq, unsigned int num, @@ -1222,14 +1227,27 @@ static int iotlb_access_ok(struct vhost_virtqueue *vq, return true; } -int vq_iotlb_prefetch(struct vhost_virtqueue *vq) +int vq_iotlb_prefetch_packed(struct vhost_virtqueue *vq) +{ + int num = vq->num; + + return iotlb_access_ok(vq, VHOST_ACCESS_RO, (u64)(uintptr_t)vq->desc, + num * sizeof(*vq->desc), VHOST_ADDR_DESC) && + iotlb_access_ok(vq, VHOST_ACCESS_WO, (u64)(uintptr_t)vq->desc, + num * sizeof(*vq->desc), VHOST_ADDR_DESC) && + iotlb_access_ok(vq, VHOST_ACCESS_RO, + (u64)(uintptr_t)vq->driver_event, + sizeof(*vq->driver_event), VHOST_ADDR_AVAIL) && + iotlb_access_ok(vq, VHOST_ACCESS_WO, + (u64)(uintptr_t)vq->device_event, + sizeof(*vq->device_event), VHOST_ADDR_USED); +} + +int vq_iotlb_prefetch_split(struct vhost_virtqueue *vq) { size_t s = vhost_has_feature(vq, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0; unsigned int num = vq->num; - if (!vq->iotlb) - return 1; - return iotlb_access_ok(vq, VHOST_ACCESS_RO, (u64)(uintptr_t)vq->desc, num * sizeof(*vq->desc), VHOST_ADDR_DESC) && iotlb_access_ok(vq, VHOST_ACCESS_RO, (u64)(uintptr_t)vq->avail, @@ -1241,6 +1259,17 @@ int vq_iotlb_prefetch(struct vhost_virtqueue *vq) num * sizeof(*vq->used->ring) + s, VHOST_ADDR_USED); } + +int vq_iotlb_prefetch(struct vhost_virtqueue *vq) +{ + if (!vq->iotlb) + return 1; + + if (vhost_has_feature(vq, VIRTIO_F_RING_PACKED)) + return vq_iotlb_prefetch_packed(vq); + else + return vq_iotlb_prefetch_split(vq); +} EXPORT_SYMBOL_GPL(vq_iotlb_prefetch); /* Can we log writes? */ @@ -1756,6 +1785,29 @@ static int vhost_update_used_flags(struct vhost_virtqueue *vq) return 0; } +static int vhost_update_device_flags(struct vhost_virtqueue *vq, + __virtio16 device_flags) +{ + void __user *flags; + + if (vhost_put_user(vq, cpu_to_vhost16(vq, device_flags), + &vq->device_event->desc_event_flags, + VHOST_ADDR_DESC) < 0) + return -EFAULT; + if (unlikely(vq->log_used)) { + /* Make sure the flag is seen before log. */ + smp_wmb(); + /* Log used flag write. */ + flags = &vq->device_event->desc_event_flags; + log_write(vq->log_base, vq->log_addr + + (flags - (void __user *)vq->device_event), + sizeof(vq->used->flags)); + if (vq->log_ctx) + eventfd_signal(vq->log_ctx, 1); + } + return 0; +} + static int vhost_update_avail_event(struct vhost_virtqueue *vq, u16 avail_event) { if (vhost_put_user(vq, cpu_to_vhost16(vq, vq->avail_idx), @@ -2667,16 +2719,13 @@ int vhost_add_used_n(struct vhost_virtqueue *vq, } EXPORT_SYMBOL_GPL(vhost_add_used_n); -static bool vhost_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq) +static bool vhost_notify_split(struct vhost_dev *dev, + struct vhost_virtqueue *vq) { __u16 old, new; __virtio16 event; bool v; - /* FIXME: check driver area */ - if (vhost_has_feature(vq, VIRTIO_F_RING_PACKED)) - return false; - /* Flush out used index updates. This is paired * with the barrier that the Guest executes when enabling * interrupts. */ @@ -2709,6 +2758,79 @@ static bool vhost_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq) return vring_need_event(vhost16_to_cpu(vq, event), new, old); } +static bool 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); +} + +static bool vhost_notify_packed(struct vhost_dev *dev, + struct vhost_virtqueue *vq) +{ + __virtio16 event_off_wrap, event_flags; + __u16 old, new; + bool v, wrap; + int off; + + /* Flush out used descriptors updates. This is paired + * with the barrier that the Guest executes when enabling + * interrupts. + */ + smp_mb(); + + if (vhost_get_avail(vq, event_flags, + &vq->driver_event->desc_event_flags) < 0) { + vq_err(vq, "Failed to get driver desc_event_flags"); + return true; + } + + if (!(event_flags & cpu_to_vhost16(vq, RING_EVENT_FLAGS_DESC))) + return event_flags =+ cpu_to_vhost16(vq, RING_EVENT_FLAGS_ENABLE); + + /* Read desc event flags before event_off and event_wrap */ + smp_rmb(); + + if (vhost_get_avail(vq, event_off_wrap, + &vq->driver_event->desc_event_off_warp) < 0) { + vq_err(vq, "Failed to get driver desc_event_off/wrap"); + return true; + } + + off = vhost16_to_cpu(vq, event_off_wrap); + + wrap = off & 0x1; + off >>= 1; + + old = vq->signalled_used; + v = vq->signalled_used_valid; + new = vq->signalled_used = vq->last_used_idx; + vq->signalled_used_valid = true; + + if (unlikely(!v)) + return true; + + return vhost_vring_packed_need_event(vq, new, old, off) && + wrap == vq->used_wrap_counter; +} + +static bool vhost_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq) +{ + if (vhost_has_feature(vq, VIRTIO_F_RING_PACKED)) + return vhost_notify_packed(dev, vq); + else + return vhost_notify_split(dev, vq); +} + /* This actually signals the guest, using eventfd. */ void vhost_signal(struct vhost_dev *dev, struct vhost_virtqueue *vq) { @@ -2789,7 +2911,17 @@ static bool vhost_enable_notify_packed(struct vhost_dev *dev, __virtio16 flags; int ret; - /* FIXME: disable notification through device area */ + if (!(vq->used_flags & VRING_USED_F_NO_NOTIFY)) + return false; + vq->used_flags &= ~VRING_USED_F_NO_NOTIFY; + + flags = cpu_to_vhost16(vq, RING_EVENT_FLAGS_ENABLE); + ret = vhost_update_device_flags(vq, flags); + if (ret) { + vq_err(vq, "Failed to enable notification at %p: %d\n", + &vq->device_event->desc_event_flags, ret); + return false; + } /* They could have slipped one in as we were doing that: make * sure it's written, then check again. */ @@ -2855,7 +2987,18 @@ EXPORT_SYMBOL_GPL(vhost_enable_notify); static void vhost_disable_notify_packed(struct vhost_dev *dev, struct vhost_virtqueue *vq) { - /* FIXME: disable notification through device area */ + __virtio16 flags; + int r; + + if (vq->used_flags & VRING_USED_F_NO_NOTIFY) + return; + vq->used_flags |= VRING_USED_F_NO_NOTIFY; + + flags = cpu_to_vhost16(vq, RING_EVENT_FLAGS_DISABLE); + r = vhost_update_device_flags(vq, flags); + if (r) + vq_err(vq, "Failed to enable notification at %p: %d\n", + &vq->device_event->desc_event_flags, r); } static void vhost_disable_notify_split(struct vhost_dev *dev, diff --git a/drivers/vhost/vhost.h b/drivers/vhost/vhost.h index 8a9df4f..02d7a36 100644 --- a/drivers/vhost/vhost.h +++ b/drivers/vhost/vhost.h @@ -96,8 +96,14 @@ struct vhost_virtqueue { struct vring_desc __user *desc; struct vring_desc_packed __user *desc_packed; }; - struct vring_avail __user *avail; - struct vring_used __user *used; + union { + struct vring_avail __user *avail; + struct vring_packed_desc_event __user *driver_event; + }; + union { + struct vring_used __user *used; + struct vring_packed_desc_event __user *device_event; + }; const struct vhost_umem_node *meta_iotlb[VHOST_NUM_ADDRS]; struct file *kick; struct eventfd_ctx *call_ctx; diff --git a/include/uapi/linux/virtio_ring.h b/include/uapi/linux/virtio_ring.h index e297580..7cdbf06 100644 --- a/include/uapi/linux/virtio_ring.h +++ b/include/uapi/linux/virtio_ring.h @@ -75,6 +75,25 @@ struct vring_desc_packed { __virtio16 flags; }; +/* Enable events */ +#define RING_EVENT_FLAGS_ENABLE 0x0 +/* Disable events */ +#define RING_EVENT_FLAGS_DISABLE 0x1 +/* + * Enable events for a specific descriptor + * (as specified by Descriptor Ring Change Event Offset/Wrap Counter). + * Only valid if VIRTIO_F_RING_EVENT_IDX has been negotiated. + */ +#define RING_EVENT_FLAGS_DESC 0x2 +/* The value 0x3 is reserved */ + +struct vring_packed_desc_event { + /* Descriptor Ring Change Event Offset and Wrap Counter */ + __virtio16 desc_event_off_warp; + /* Descriptor Ring Change Event Flags */ + __virtio16 desc_event_flags; +}; + /* Virtio ring descriptors: 16 bytes. These can chain together via "next". */ struct vring_desc { /* Address (guest-physical). */ -- 2.7.4
cc Jens, Tiwei and Wei Thanks On 2018?03?26? 11:38, Jason Wang wrote:> Hi all: > > This RFC implement packed ring layout. The code were tested with pmd > implement by Jens at > http://dpdk.org/ml/archives/dev/2018-January/089417.html. Minor change > was needed for pmd codes to kick virtqueue since it assumes a busy > polling backend. > > Test were done between localhost and guest. Testpmd (rxonly) in guest > reports 2.4Mpps. Testpmd (txonly) repots about 2.1Mpps. > > Notes: The event suppression /indirect descriptor support is complied > test only because of lacked driver support. > > Changes from V1: > > - Refactor vhost used elem code to avoid open coding on used elem > - Event suppression support (compile test only). > - Indirect descriptor support (compile test only). > - Zerocopy support. > - vIOMMU support. > - SCSI/VSOCK support (compile test only). > - Fix several bugs > > For simplicity, I don't implement batching or other optimizations. > > Please review. > > Thanks > > Jason Wang (8): > vhost: move get_rx_bufs to vhost.c > vhost: hide used ring layout from device > vhost: do not use vring_used_elem > vhost_net: do not explicitly manipulate vhost_used_elem > vhost: vhost_put_user() can accept metadata type > virtio: introduce packed ring defines > vhost: packed ring support > vhost: event suppression for packed ring > > drivers/vhost/net.c | 138 ++----- > drivers/vhost/scsi.c | 62 +-- > drivers/vhost/vhost.c | 818 ++++++++++++++++++++++++++++++++++--- > drivers/vhost/vhost.h | 46 ++- > drivers/vhost/vsock.c | 42 +- > include/uapi/linux/virtio_config.h | 9 + > include/uapi/linux/virtio_ring.h | 32 ++ > 7 files changed, 921 insertions(+), 226 deletions(-) >
On Mon, Mar 26, 2018 at 11:38:45AM +0800, Jason Wang wrote:> Hi all: > > This RFC implement packed ring layout. The code were tested with pmd > implement by Jens at > http://dpdk.org/ml/archives/dev/2018-January/089417.html. Minor change > was needed for pmd codes to kick virtqueue since it assumes a busy > polling backend. > > Test were done between localhost and guest. Testpmd (rxonly) in guest > reports 2.4Mpps. Testpmd (txonly) repots about 2.1Mpps.And how does it compare to older ring layout?> > Notes: The event suppression /indirect descriptor support is complied > test only because of lacked driver support. > > Changes from V1: > > - Refactor vhost used elem code to avoid open coding on used elem > - Event suppression support (compile test only). > - Indirect descriptor support (compile test only). > - Zerocopy support. > - vIOMMU support. > - SCSI/VSOCK support (compile test only). > - Fix several bugs > > For simplicity, I don't implement batching or other optimizations. > > Please review. > > Thanks > > Jason Wang (8): > vhost: move get_rx_bufs to vhost.c > vhost: hide used ring layout from device > vhost: do not use vring_used_elem > vhost_net: do not explicitly manipulate vhost_used_elem > vhost: vhost_put_user() can accept metadata type > virtio: introduce packed ring defines > vhost: packed ring support > vhost: event suppression for packed ring > > drivers/vhost/net.c | 138 ++----- > drivers/vhost/scsi.c | 62 +-- > drivers/vhost/vhost.c | 818 ++++++++++++++++++++++++++++++++++--- > drivers/vhost/vhost.h | 46 ++- > drivers/vhost/vsock.c | 42 +- > include/uapi/linux/virtio_config.h | 9 + > include/uapi/linux/virtio_ring.h | 32 ++ > 7 files changed, 921 insertions(+), 226 deletions(-) > > -- > 2.7.4 >
On 2018?03?27? 03:01, Konrad Rzeszutek Wilk wrote:> On Mon, Mar 26, 2018 at 11:38:45AM +0800, Jason Wang wrote: >> Hi all: >> >> This RFC implement packed ring layout. The code were tested with pmd >> implement by Jens at >> http://dpdk.org/ml/archives/dev/2018-January/089417.html. Minor change >> was needed for pmd codes to kick virtqueue since it assumes a busy >> polling backend. >> >> Test were done between localhost and guest. Testpmd (rxonly) in guest >> reports 2.4Mpps. Testpmd (txonly) repots about 2.1Mpps. > And how does it compare to older ring layout? >No obvious difference. Vhost itself becomes a bottleneck. I plan to do more aggressive batching like dpdk on top. Thanks
Tiwei Bie
2018-Mar-30 02:05 UTC
[RFC PATCH V2 8/8] vhost: event suppression for packed ring
On Mon, Mar 26, 2018 at 11:38:53AM +0800, Jason Wang wrote:> This patch introduces basic support for event suppression aka driver > and device area. Compile tested only. > > Signed-off-by: Jason Wang <jasowang at redhat.com> > ---[...]> + > +static bool vhost_notify_packed(struct vhost_dev *dev, > + struct vhost_virtqueue *vq) > +{ > + __virtio16 event_off_wrap, event_flags; > + __u16 old, new; > + bool v, wrap; > + int off; > + > + /* Flush out used descriptors updates. This is paired > + * with the barrier that the Guest executes when enabling > + * interrupts. > + */ > + smp_mb(); > + > + if (vhost_get_avail(vq, event_flags, > + &vq->driver_event->desc_event_flags) < 0) { > + vq_err(vq, "Failed to get driver desc_event_flags"); > + return true; > + } > + > + if (!(event_flags & cpu_to_vhost16(vq, RING_EVENT_FLAGS_DESC))) > + return event_flags => + cpu_to_vhost16(vq, RING_EVENT_FLAGS_ENABLE);Maybe it would be better to not use '&' here. Because these flags are not defined as bits which can be ORed or ANDed. Instead, they are defined as values: 0x0 enable 0x1 disable 0x2 desc 0x3 reserved> + > + /* Read desc event flags before event_off and event_wrap */ > + smp_rmb(); > + > + if (vhost_get_avail(vq, event_off_wrap, > + &vq->driver_event->desc_event_off_warp) < 0) { > + vq_err(vq, "Failed to get driver desc_event_off/wrap"); > + return true; > + } > + > + off = vhost16_to_cpu(vq, event_off_wrap); > + > + wrap = off & 0x1; > + off >>= 1;Based on the below definitions in spec, wrap counter is the most significant bit. struct pvirtq_event_suppress { le16 { desc_event_off : 15; /* Descriptor Ring Change Event Offset */ desc_event_wrap : 1; /* Descriptor Ring Change Event Wrap Counter */ } desc; /* If desc_event_flags set to RING_EVENT_FLAGS_DESC */ le16 { desc_event_flags : 2, /* Descriptor Ring Change Event Flags */ reserved : 14; /* Reserved, set to 0 */ } flags; };> + > + > + old = vq->signalled_used; > + v = vq->signalled_used_valid; > + new = vq->signalled_used = vq->last_used_idx; > + vq->signalled_used_valid = true; > + > + if (unlikely(!v)) > + return true; > + > + return vhost_vring_packed_need_event(vq, new, old, off) && > + wrap == vq->used_wrap_counter; > +} > + > +static bool vhost_notify(struct vhost_dev *dev, struct vhost_virtqueue *vq) > +{ > + if (vhost_has_feature(vq, VIRTIO_F_RING_PACKED)) > + return vhost_notify_packed(dev, vq); > + else > + return vhost_notify_split(dev, vq); > +} > + > /* This actually signals the guest, using eventfd. */ > void vhost_signal(struct vhost_dev *dev, struct vhost_virtqueue *vq) > { > @@ -2789,7 +2911,17 @@ static bool vhost_enable_notify_packed(struct vhost_dev *dev, > __virtio16 flags; > int ret; > > - /* FIXME: disable notification through device area */ > + if (!(vq->used_flags & VRING_USED_F_NO_NOTIFY)) > + return false; > + vq->used_flags &= ~VRING_USED_F_NO_NOTIFY; > + > + flags = cpu_to_vhost16(vq, RING_EVENT_FLAGS_ENABLE); > + ret = vhost_update_device_flags(vq, flags); > + if (ret) { > + vq_err(vq, "Failed to enable notification at %p: %d\n", > + &vq->device_event->desc_event_flags, ret); > + return false; > + } > > /* They could have slipped one in as we were doing that: make > * sure it's written, then check again. */ > @@ -2855,7 +2987,18 @@ EXPORT_SYMBOL_GPL(vhost_enable_notify); > static void vhost_disable_notify_packed(struct vhost_dev *dev, > struct vhost_virtqueue *vq) > { > - /* FIXME: disable notification through device area */ > + __virtio16 flags; > + int r; > + > + if (vq->used_flags & VRING_USED_F_NO_NOTIFY) > + return; > + vq->used_flags |= VRING_USED_F_NO_NOTIFY; > + > + flags = cpu_to_vhost16(vq, RING_EVENT_FLAGS_DISABLE); > + r = vhost_update_device_flags(vq, flags); > + if (r) > + vq_err(vq, "Failed to enable notification at %p: %d\n", > + &vq->device_event->desc_event_flags, r); > } > > static void vhost_disable_notify_split(struct vhost_dev *dev, > diff --git a/drivers/vhost/vhost.h b/drivers/vhost/vhost.h > index 8a9df4f..02d7a36 100644 > --- a/drivers/vhost/vhost.h > +++ b/drivers/vhost/vhost.h > @@ -96,8 +96,14 @@ struct vhost_virtqueue { > struct vring_desc __user *desc; > struct vring_desc_packed __user *desc_packed;Do you think it'd be better to name the desc type as struct vring_packed_desc? And it will be consistent with other names, like: struct vring_packed; struct vring_packed_desc_event;> }; > - struct vring_avail __user *avail; > - struct vring_used __user *used; > + union { > + struct vring_avail __user *avail; > + struct vring_packed_desc_event __user *driver_event; > + }; > + union { > + struct vring_used __user *used; > + struct vring_packed_desc_event __user *device_event; > + }; > const struct vhost_umem_node *meta_iotlb[VHOST_NUM_ADDRS]; > struct file *kick; > struct eventfd_ctx *call_ctx; > diff --git a/include/uapi/linux/virtio_ring.h b/include/uapi/linux/virtio_ring.h > index e297580..7cdbf06 100644 > --- a/include/uapi/linux/virtio_ring.h > +++ b/include/uapi/linux/virtio_ring.h > @@ -75,6 +75,25 @@ struct vring_desc_packed { > __virtio16 flags; > }; > > +/* Enable events */ > +#define RING_EVENT_FLAGS_ENABLE 0x0 > +/* Disable events */ > +#define RING_EVENT_FLAGS_DISABLE 0x1 > +/* > + * Enable events for a specific descriptor > + * (as specified by Descriptor Ring Change Event Offset/Wrap Counter). > + * Only valid if VIRTIO_F_RING_EVENT_IDX has been negotiated. > + */ > +#define RING_EVENT_FLAGS_DESC 0x2 > +/* The value 0x3 is reserved */ > + > +struct vring_packed_desc_event { > + /* Descriptor Ring Change Event Offset and Wrap Counter */ > + __virtio16 desc_event_off_warp; > + /* Descriptor Ring Change Event Flags */ > + __virtio16 desc_event_flags;Do you think it'd be better to remove the prefix (desc_event_) for the fields. And it will be consistent with other definitions, e.g.: struct vring_packed_desc { /* Buffer Address. */ __virtio64 addr; /* Buffer Length. */ __virtio32 len; /* Buffer ID. */ __virtio16 id; /* The flags depending on descriptor type. */ __virtio16 flags; };> +}; > + > /* Virtio ring descriptors: 16 bytes. These can chain together via "next". */ > struct vring_desc { > /* Address (guest-physical). */ > -- > 2.7.4 >
Reasonably Related Threads
- [RFC V4 PATCH 8/8] vhost: event suppression for packed ring
- [RFC V5 PATCH 8/8] vhost: event suppression for packed ring
- [PATCH net-next 8/8] vhost: event suppression for packed ring
- [RFC V5 PATCH 8/8] vhost: event suppression for packed ring
- [RFC V5 PATCH 8/8] vhost: event suppression for packed ring