Skip to content

Commit 821c9e5

Browse files
torvaldstorvalds
committed
Merge tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost
Pull virtio updates from Michael Tsirkin: - vhost can now support legacy threading if enabled in Kconfig - vsock memory allocation strategies for large buffers have been improved, reducing pressure on kmalloc - vhost now supports the in-order feature. guest bits missed the merge window. - fixes, cleanups all over the place * tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost: (30 commits) vsock/virtio: Allocate nonlinear SKBs for handling large transmit buffers vsock/virtio: Rename virtio_vsock_skb_rx_put() vhost/vsock: Allocate nonlinear SKBs for handling large receive buffers vsock/virtio: Move SKB allocation lower-bound check to callers vsock/virtio: Rename virtio_vsock_alloc_skb() vsock/virtio: Resize receive buffers so that each SKB fits in a 4K page vsock/virtio: Move length check to callers of virtio_vsock_skb_rx_put() vsock/virtio: Validate length in packet header before skb_put() vhost/vsock: Avoid allocating arbitrarily-sized SKBs vhost_net: basic in_order support vhost: basic in order support vhost: fail early when __vhost_add_used() fails vhost: Reintroduce kthread API and add mode selection vdpa: Fix IDR memory leak in VDUSE module exit vdpa/mlx5: Fix release of uninitialized resources on error path vhost-scsi: Fix check for inline_sg_cnt exceeding preallocated limit virtio: virtio_dma_buf: fix missing parameter documentation vhost: Fix typos vhost: vringh: Remove unused functions vhost: vringh: Remove unused iotlb functions ...
2 parents 0bd0a41 + 6693731 commit 821c9e5

File tree

23 files changed

+574
-342
lines changed

23 files changed

+574
-342
lines changed

‎drivers/gpu/drm/virtio/virtgpu_drv.c‎

Lines changed: 4 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -130,10 +130,10 @@ static void virtio_gpu_remove(struct virtio_device *vdev)
130130

131131
static void virtio_gpu_shutdown(struct virtio_device *vdev)
132132
{
133-
/*
134-
* drm does its own synchronization on shutdown.
135-
* Do nothing here, opt out of device reset.
136-
*/
133+
struct drm_device *dev = vdev->priv;
134+
135+
/* stop talking to the device */
136+
drm_dev_unplug(dev);
137137
}
138138

139139
static void virtio_gpu_config_changed(struct virtio_device *vdev)

‎drivers/vdpa/mlx5/core/mr.c‎

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -908,6 +908,9 @@ void mlx5_vdpa_destroy_mr_resources(struct mlx5_vdpa_dev *mvdev)
908908
{
909909
struct mlx5_vdpa_mr_resources *mres = &mvdev->mres;
910910

911+
if (!mres->wq_gc)
912+
return;
913+
911914
atomic_set(&mres->shutdown, 1);
912915

913916
flush_delayed_work(&mres->gc_dwork_ent);

‎drivers/vdpa/mlx5/net/mlx5_vnet.c‎

Lines changed: 7 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -2491,7 +2491,7 @@ static void mlx5_vdpa_set_vq_num(struct vdpa_device *vdev, u16 idx, u32 num)
24912491
}
24922492

24932493
mvq = &ndev->vqs[idx];
2494-
ndev->needs_teardown = num != mvq->num_ent;
2494+
ndev->needs_teardown |= num != mvq->num_ent;
24952495
mvq->num_ent = num;
24962496
}
24972497

@@ -3432,15 +3432,17 @@ static void mlx5_vdpa_free(struct vdpa_device *vdev)
34323432

34333433
ndev = to_mlx5_vdpa_ndev(mvdev);
34343434

3435+
/* Functions called here should be able to work with
3436+
* uninitialized resources.
3437+
*/
34353438
free_fixed_resources(ndev);
34363439
mlx5_vdpa_clean_mrs(mvdev);
34373440
mlx5_vdpa_destroy_mr_resources(&ndev->mvdev);
3438-
mlx5_cmd_cleanup_async_ctx(&mvdev->async_ctx);
3439-
34403441
if (!is_zero_ether_addr(ndev->config.mac)) {
34413442
pfmdev = pci_get_drvdata(pci_physfn(mvdev->mdev->pdev));
34423443
mlx5_mpfs_del_mac(pfmdev, ndev->config.mac);
34433444
}
3445+
mlx5_cmd_cleanup_async_ctx(&mvdev->async_ctx);
34443446
mlx5_vdpa_free_resources(&ndev->mvdev);
34453447
free_irqs(ndev);
34463448
kfree(ndev->event_cbs);
@@ -3888,6 +3890,8 @@ static int mlx5_vdpa_dev_add(struct vdpa_mgmt_dev *v_mdev, const char *name,
38883890
mvdev->actual_features =
38893891
(device_features & BIT_ULL(VIRTIO_F_VERSION_1));
38903892

3893+
mlx5_cmd_init_async_ctx(mdev, &mvdev->async_ctx);
3894+
38913895
ndev->vqs = kcalloc(max_vqs, sizeof(*ndev->vqs), GFP_KERNEL);
38923896
ndev->event_cbs = kcalloc(max_vqs + 1, sizeof(*ndev->event_cbs), GFP_KERNEL);
38933897
if (!ndev->vqs || !ndev->event_cbs) {
@@ -3960,8 +3964,6 @@ static int mlx5_vdpa_dev_add(struct vdpa_mgmt_dev *v_mdev, const char *name,
39603964
ndev->rqt_size = 1;
39613965
}
39623966

3963-
mlx5_cmd_init_async_ctx(mdev, &mvdev->async_ctx);
3964-
39653967
ndev->mvdev.mlx_features = device_features;
39663968
mvdev->vdev.dma_dev = &mdev->pdev->dev;
39673969
err = mlx5_vdpa_alloc_resources(&ndev->mvdev);

‎drivers/vdpa/vdpa_user/vduse_dev.c‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -2216,6 +2216,7 @@ static void vduse_exit(void)
22162216
cdev_del(&vduse_ctrl_cdev);
22172217
unregister_chrdev_region(vduse_major, VDUSE_DEV_MAX);
22182218
class_unregister(&vduse_class);
2219+
idr_destroy(&vduse_idr);
22192220
}
22202221
module_exit(vduse_exit);
22212222

‎drivers/vhost/Kconfig‎

Lines changed: 18 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -95,4 +95,22 @@ config VHOST_CROSS_ENDIAN_LEGACY
9595

9696
If unsure, say "N".
9797

98+
config VHOST_ENABLE_FORK_OWNER_CONTROL
99+
bool "Enable VHOST_ENABLE_FORK_OWNER_CONTROL"
100+
default y
101+
help
102+
This option enables two IOCTLs: VHOST_SET_FORK_FROM_OWNER and
103+
VHOST_GET_FORK_FROM_OWNER. These allow userspace applications
104+
to modify the vhost worker mode for vhost devices.
105+
106+
Also expose module parameter 'fork_from_owner_default' to allow users
107+
to configure the default mode for vhost workers.
108+
109+
By default, `VHOST_ENABLE_FORK_OWNER_CONTROL` is set to `y`,
110+
users can change the worker thread mode as needed.
111+
If this config is disabled (n),the related IOCTLs and parameters will
112+
be unavailable.
113+
114+
If unsure, say "Y".
115+
98116
endif

‎drivers/vhost/net.c‎

Lines changed: 63 additions & 25 deletions
Original file line numberDiff line numberDiff line change
@@ -74,7 +74,8 @@ static const u64 vhost_net_features[VIRTIO_FEATURES_DWORDS] = {
7474
(1ULL << VHOST_NET_F_VIRTIO_NET_HDR) |
7575
(1ULL << VIRTIO_NET_F_MRG_RXBUF) |
7676
(1ULL << VIRTIO_F_ACCESS_PLATFORM) |
77-
(1ULL << VIRTIO_F_RING_RESET),
77+
(1ULL << VIRTIO_F_RING_RESET) |
78+
(1ULL << VIRTIO_F_IN_ORDER),
7879
VIRTIO_BIT(VIRTIO_NET_F_GUEST_UDP_TUNNEL_GSO) |
7980
VIRTIO_BIT(VIRTIO_NET_F_HOST_UDP_TUNNEL_GSO),
8081
};
@@ -376,7 +377,8 @@ static void vhost_zerocopy_signal_used(struct vhost_net *net,
376377
while (j) {
377378
add = min(UIO_MAXIOV - nvq->done_idx, j);
378379
vhost_add_used_and_signal_n(vq->dev, vq,
379-
&vq->heads[nvq->done_idx], add);
380+
&vq->heads[nvq->done_idx],
381+
NULL, add);
380382
nvq->done_idx = (nvq->done_idx + add) % UIO_MAXIOV;
381383
j -= add;
382384
}
@@ -451,15 +453,17 @@ static int vhost_net_enable_vq(struct vhost_net *n,
451453
return vhost_poll_start(poll, sock->file);
452454
}
453455

454-
static void vhost_net_signal_used(struct vhost_net_virtqueue *nvq)
456+
static void vhost_net_signal_used(struct vhost_net_virtqueue *nvq,
457+
unsigned int count)
455458
{
456459
struct vhost_virtqueue *vq = &nvq->vq;
457460
struct vhost_dev *dev = vq->dev;
458461

459462
if (!nvq->done_idx)
460463
return;
461464

462-
vhost_add_used_and_signal_n(dev, vq, vq->heads, nvq->done_idx);
465+
vhost_add_used_and_signal_n(dev, vq, vq->heads,
466+
vq->nheads, count);
463467
nvq->done_idx = 0;
464468
}
465469

@@ -468,13 +472,20 @@ static void vhost_tx_batch(struct vhost_net *net,
468472
struct socket *sock,
469473
struct msghdr *msghdr)
470474
{
475+
struct vhost_virtqueue *vq = &nvq->vq;
476+
bool in_order = vhost_has_feature(vq, VIRTIO_F_IN_ORDER);
471477
struct tun_msg_ctl ctl = {
472478
.type = TUN_MSG_PTR,
473479
.num = nvq->batched_xdp,
474480
.ptr = nvq->xdp,
475481
};
476482
int i, err;
477483

484+
if (in_order) {
485+
vq->heads[0].len = 0;
486+
vq->nheads[0] = nvq->done_idx;
487+
}
488+
478489
if (nvq->batched_xdp == 0)
479490
goto signal_used;
480491

@@ -496,7 +507,7 @@ static void vhost_tx_batch(struct vhost_net *net,
496507
}
497508

498509
signal_used:
499-
vhost_net_signal_used(nvq);
510+
vhost_net_signal_used(nvq, in_order ? 1 : nvq->done_idx);
500511
nvq->batched_xdp = 0;
501512
}
502513

@@ -750,6 +761,7 @@ static void handle_tx_copy(struct vhost_net *net, struct socket *sock)
750761
int sent_pkts = 0;
751762
bool sock_can_batch = (sock->sk->sk_sndbuf == INT_MAX);
752763
bool busyloop_intr;
764+
bool in_order = vhost_has_feature(vq, VIRTIO_F_IN_ORDER);
753765

754766
do {
755767
busyloop_intr = false;
@@ -786,11 +798,13 @@ static void handle_tx_copy(struct vhost_net *net, struct socket *sock)
786798
break;
787799
}
788800

789-
/* We can't build XDP buff, go for single
790-
* packet path but let's flush batched
791-
* packets.
792-
*/
793-
vhost_tx_batch(net, nvq, sock, &msg);
801+
if (nvq->batched_xdp) {
802+
/* We can't build XDP buff, go for single
803+
* packet path but let's flush batched
804+
* packets.
805+
*/
806+
vhost_tx_batch(net, nvq, sock, &msg);
807+
}
794808
msg.msg_control = NULL;
795809
} else {
796810
if (tx_can_batch(vq, total_len))
@@ -811,8 +825,12 @@ static void handle_tx_copy(struct vhost_net *net, struct socket *sock)
811825
pr_debug("Truncated TX packet: len %d != %zd\n",
812826
err, len);
813827
done:
814-
vq->heads[nvq->done_idx].id = cpu_to_vhost32(vq, head);
815-
vq->heads[nvq->done_idx].len = 0;
828+
if (in_order) {
829+
vq->heads[0].id = cpu_to_vhost32(vq, head);
830+
} else {
831+
vq->heads[nvq->done_idx].id = cpu_to_vhost32(vq, head);
832+
vq->heads[nvq->done_idx].len = 0;
833+
}
816834
++nvq->done_idx;
817835
} while (likely(!vhost_exceeds_weight(vq, ++sent_pkts, total_len)));
818836

@@ -991,7 +1009,7 @@ static int peek_head_len(struct vhost_net_virtqueue *rvq, struct sock *sk)
9911009
}
9921010

9931011
static int vhost_net_rx_peek_head_len(struct vhost_net *net, struct sock *sk,
994-
bool *busyloop_intr)
1012+
bool *busyloop_intr, unsigned int count)
9951013
{
9961014
struct vhost_net_virtqueue *rnvq = &net->vqs[VHOST_NET_VQ_RX];
9971015
struct vhost_net_virtqueue *tnvq = &net->vqs[VHOST_NET_VQ_TX];
@@ -1001,7 +1019,7 @@ static int vhost_net_rx_peek_head_len(struct vhost_net *net, struct sock *sk,
10011019

10021020
if (!len && rvq->busyloop_timeout) {
10031021
/* Flush batched heads first */
1004-
vhost_net_signal_used(rnvq);
1022+
vhost_net_signal_used(rnvq, count);
10051023
/* Both tx vq and rx socket were polled here */
10061024
vhost_net_busy_poll(net, rvq, tvq, busyloop_intr, true);
10071025

@@ -1013,22 +1031,25 @@ static int vhost_net_rx_peek_head_len(struct vhost_net *net, struct sock *sk,
10131031

10141032
/* This is a multi-buffer version of vhost_get_desc, that works if
10151033
* vq has read descriptors only.
1016-
* @vq - the relevant virtqueue
1034+
* @nvq - the relevant vhost_net virtqueue
10171035
* @datalen - data length we'll be reading
10181036
* @iovcount - returned count of io vectors we fill
10191037
* @log - vhost log
10201038
* @log_num - log offset
10211039
* @quota - headcount quota, 1 for big buffer
10221040
* returns number of buffer heads allocated, negative on error
10231041
*/
1024-
static int get_rx_bufs(struct vhost_virtqueue *vq,
1042+
static int get_rx_bufs(struct vhost_net_virtqueue *nvq,
10251043
struct vring_used_elem *heads,
1044+
u16 *nheads,
10261045
int datalen,
10271046
unsigned *iovcount,
10281047
struct vhost_log *log,
10291048
unsigned *log_num,
10301049
unsigned int quota)
10311050
{
1051+
struct vhost_virtqueue *vq = &nvq->vq;
1052+
bool in_order = vhost_has_feature(vq, VIRTIO_F_IN_ORDER);
10321053
unsigned int out, in;
10331054
int seg = 0;
10341055
int headcount = 0;
@@ -1065,14 +1086,16 @@ static int get_rx_bufs(struct vhost_virtqueue *vq,
10651086
nlogs += *log_num;
10661087
log += *log_num;
10671088
}
1068-
heads[headcount].id = cpu_to_vhost32(vq, d);
10691089
len = iov_length(vq->iov + seg, in);
1070-
heads[headcount].len = cpu_to_vhost32(vq, len);
1071-
datalen -= len;
1090+
if (!in_order) {
1091+
heads[headcount].id = cpu_to_vhost32(vq, d);
1092+
heads[headcount].len = cpu_to_vhost32(vq, len);
1093+
}
10721094
++headcount;
1095+
datalen -= len;
10731096
seg += in;
10741097
}
1075-
heads[headcount - 1].len = cpu_to_vhost32(vq, len + datalen);
1098+
10761099
*iovcount = seg;
10771100
if (unlikely(log))
10781101
*log_num = nlogs;
@@ -1082,6 +1105,15 @@ static int get_rx_bufs(struct vhost_virtqueue *vq,
10821105
r = UIO_MAXIOV + 1;
10831106
goto err;
10841107
}
1108+
1109+
if (!in_order)
1110+
heads[headcount - 1].len = cpu_to_vhost32(vq, len + datalen);
1111+
else {
1112+
heads[0].len = cpu_to_vhost32(vq, len + datalen);
1113+
heads[0].id = cpu_to_vhost32(vq, d);
1114+
nheads[0] = headcount;
1115+
}
1116+
10851117
return headcount;
10861118
err:
10871119
vhost_discard_vq_desc(vq, headcount);
@@ -1094,6 +1126,8 @@ static void handle_rx(struct vhost_net *net)
10941126
{
10951127
struct vhost_net_virtqueue *nvq = &net->vqs[VHOST_NET_VQ_RX];
10961128
struct vhost_virtqueue *vq = &nvq->vq;
1129+
bool in_order = vhost_has_feature(vq, VIRTIO_F_IN_ORDER);
1130+
unsigned int count = 0;
10971131
unsigned in, log;
10981132
struct vhost_log *vq_log;
10991133
struct msghdr msg = {
@@ -1141,12 +1175,13 @@ static void handle_rx(struct vhost_net *net)
11411175

11421176
do {
11431177
sock_len = vhost_net_rx_peek_head_len(net, sock->sk,
1144-
&busyloop_intr);
1178+
&busyloop_intr, count);
11451179
if (!sock_len)
11461180
break;
11471181
sock_len += sock_hlen;
11481182
vhost_len = sock_len + vhost_hlen;
1149-
headcount = get_rx_bufs(vq, vq->heads + nvq->done_idx,
1183+
headcount = get_rx_bufs(nvq, vq->heads + count,
1184+
vq->nheads + count,
11501185
vhost_len, &in, vq_log, &log,
11511186
likely(mergeable) ? UIO_MAXIOV : 1);
11521187
/* On error, stop handling until the next kick. */
@@ -1222,8 +1257,11 @@ static void handle_rx(struct vhost_net *net)
12221257
goto out;
12231258
}
12241259
nvq->done_idx += headcount;
1225-
if (nvq->done_idx > VHOST_NET_BATCH)
1226-
vhost_net_signal_used(nvq);
1260+
count += in_order ? 1 : headcount;
1261+
if (nvq->done_idx > VHOST_NET_BATCH) {
1262+
vhost_net_signal_used(nvq, count);
1263+
count = 0;
1264+
}
12271265
if (unlikely(vq_log))
12281266
vhost_log_write(vq, vq_log, log, vhost_len,
12291267
vq->iov, in);
@@ -1235,7 +1273,7 @@ static void handle_rx(struct vhost_net *net)
12351273
else if (!sock_len)
12361274
vhost_net_enable_vq(net, vq);
12371275
out:
1238-
vhost_net_signal_used(nvq);
1276+
vhost_net_signal_used(nvq, count);
12391277
mutex_unlock(&vq->mutex);
12401278
}
12411279

0 commit comments

Comments
 (0)