* [PATCH 1/4] vhost-user: Enable multiqueue
2025-11-07 14:38 [PATCH 0/4] vhost-user: Add multiqueue support Laurent Vivier
@ 2025-11-07 14:38 ` Laurent Vivier
2025-11-10 4:48 ` David Gibson
2025-11-07 14:38 ` [PATCH 2/4] vhost-user: Add queue parameter throughout the network stack Laurent Vivier
` (3 subsequent siblings)
4 siblings, 1 reply; 13+ messages in thread
From: Laurent Vivier @ 2025-11-07 14:38 UTC (permalink / raw)
To: passt-dev; +Cc: Laurent Vivier
Add the --max-queues parameter to specify the maximum number of queue
pairs supported in vhost-user mode. This enables multi-queue support
by allowing configuration of up to 16 queue pairs (32 virtqueues).
For the moment, only the first RX queue is used, the TX queue is
selected by the guest kernel.
Signed-off-by: Laurent Vivier <lvivier@redhat.com>
---
conf.c | 31 ++++++++++++++++++++++++++++++-
passt.h | 2 ++
tap.c | 15 +++++++++++++--
vhost_user.c | 38 +++++++++++++++++++++-----------------
virtio.h | 2 +-
5 files changed, 67 insertions(+), 21 deletions(-)
diff --git a/conf.c b/conf.c
index 66b9e63400ec..4738436e6245 100644
--- a/conf.c
+++ b/conf.c
@@ -862,7 +862,9 @@ static void usage(const char *name, FILE *f, int status)
" --vhost-user Enable vhost-user mode\n"
" UNIX domain socket is provided by -s option\n"
" --print-capabilities print back-end capabilities in JSON format,\n"
- " only meaningful for vhost-user mode\n");
+ " only meaningful for vhost-user mode\n"
+ " --max-queues Specify the maximum number of queue pairs\n"
+ );
FPRINTF(f,
" --repair-path PATH path for passt-repair(1)\n"
" default: append '.repair' to UNIX domain path\n");
@@ -1483,6 +1485,7 @@ void conf(struct ctx *c, int argc, char **argv)
{"migrate-exit", no_argument, NULL, 29 },
{"migrate-no-linger", no_argument, NULL, 30 },
{"stats", required_argument, NULL, 31 },
+ {"max-queues", required_argument, NULL, 32 },
{ 0 },
};
const char *optstring = "+dqfel:hs:F:I:p:P:m:a:n:M:g:i:o:D:S:H:461t:u:T:U:";
@@ -1514,6 +1517,7 @@ void conf(struct ctx *c, int argc, char **argv)
c->tcp.fwd_in.mode = c->tcp.fwd_out.mode = FWD_UNSET;
c->udp.fwd_in.mode = c->udp.fwd_out.mode = FWD_UNSET;
memcpy(c->our_tap_mac, MAC_OUR_LAA, ETH_ALEN);
+ c->max_virtqueues = 2;
optind = 0;
do {
@@ -1717,6 +1721,31 @@ void conf(struct ctx *c, int argc, char **argv)
die("Can't display statistics if not running in foreground");
c->stats = strtol(optarg, NULL, 0);
break;
+ case 32: {
+ unsigned long max_queues;
+ char *e;
+
+ if (c->mode != MODE_VU)
+ die("--max-queues is for vhost-user mode only");
+
+ errno = 0;
+ max_queues = strtoul(optarg, &e, 0);
+
+ if (errno || *e)
+ die("Invalid max-queues: %s", optarg);
+
+ if (max_queues < 1) {
+ die("max-queues %lu too small (min 1)",
+ max_queues);
+ }
+
+ if (max_queues * 2 > VHOST_USER_MAX_VQS) {
+ die("max-queues %lu too big (max %u)",
+ max_queues, VHOST_USER_MAX_VQS / 2);
+ }
+ c->max_virtqueues = max_queues * 2;
+ break;
+ }
case 'd':
c->debug = 1;
c->quiet = 0;
diff --git a/passt.h b/passt.h
index 15801b44bfa8..e73f648009de 100644
--- a/passt.h
+++ b/passt.h
@@ -205,6 +205,7 @@ struct ip6_ctx {
* @low_wmem: Low probed net.core.wmem_max
* @low_rmem: Low probed net.core.rmem_max
* @vdev: vhost-user device
+ * @max_virtqueues: vhost-user maximum number of virtqueues
* @device_state_fd: Device state migration channel
* @device_state_result: Device state migration result
* @migrate_target: Are we the target, on the next migration request?
@@ -283,6 +284,7 @@ struct ctx {
int low_rmem;
struct vu_dev *vdev;
+ unsigned int max_virtqueues;
/* Migration */
int device_state_fd;
diff --git a/tap.c b/tap.c
index bb139d647bae..e18d693a665a 100644
--- a/tap.c
+++ b/tap.c
@@ -1318,8 +1318,19 @@ static void tap_backend_show_hints(struct ctx *c)
break;
case MODE_VU:
info("You can start qemu with:");
- info(" kvm ... -chardev socket,id=chr0,path=%s -netdev vhost-user,id=netdev0,chardev=chr0 -device virtio-net,netdev=netdev0 -object memory-backend-memfd,id=memfd0,share=on,size=$RAMSIZE -numa node,memdev=memfd0\n",
- c->sock_path);
+ if (c->max_virtqueues > 2) {
+ info(" kvm ... -chardev socket,id=chr0,path=%s "
+ "-netdev vhost-user,id=netdev0,chardev=chr0,queues=%d "
+ "-device virtio-net,netdev=netdev0,mq=true "
+ "-object memory-backend-memfd,id=memfd0,share=on,size=$RAMSIZE "
+ "-numa node,memdev=memfd0\n", c->sock_path, c->max_virtqueues / 2);
+ } else {
+ info(" kvm ... -chardev socket,id=chr0,path=%s "
+ "-netdev vhost-user,id=netdev0,chardev=chr0 "
+ "-device virtio-net,netdev=netdev0 "
+ "-object memory-backend-memfd,id=memfd0,share=on,size=$RAMSIZE "
+ "-numa node,memdev=memfd0\n", c->sock_path);
+ }
break;
}
}
diff --git a/vhost_user.c b/vhost_user.c
index aa7c869d9e56..c668159e29b4 100644
--- a/vhost_user.c
+++ b/vhost_user.c
@@ -323,6 +323,7 @@ static bool vu_get_features_exec(struct vu_dev *vdev,
uint64_t features =
1ULL << VIRTIO_F_VERSION_1 |
1ULL << VIRTIO_NET_F_MRG_RXBUF |
+ 1ULL << VIRTIO_NET_F_MQ |
1ULL << VHOST_F_LOG_ALL |
1ULL << VHOST_USER_F_PROTOCOL_FEATURES;
@@ -342,9 +343,9 @@ static bool vu_get_features_exec(struct vu_dev *vdev,
*/
static void vu_set_enable_all_rings(struct vu_dev *vdev, bool enable)
{
- uint16_t i;
+ unsigned int i;
- for (i = 0; i < VHOST_USER_MAX_VQS; i++)
+ for (i = 0; i < vdev->context->max_virtqueues; i++)
vdev->vq[i].enable = enable;
}
@@ -476,7 +477,7 @@ static bool vu_set_mem_table_exec(struct vu_dev *vdev,
close(vmsg->fds[i]);
}
- for (i = 0; i < VHOST_USER_MAX_VQS; i++) {
+ for (i = 0; i < vdev->context->max_virtqueues; i++) {
if (vdev->vq[i].vring.desc) {
if (map_ring(vdev, &vdev->vq[i]))
die("remapping queue %d during setmemtable", i);
@@ -759,15 +760,18 @@ static void vu_set_watch(const struct vu_dev *vdev, int idx)
/**
* vu_check_queue_msg_file() - Check if a message is valid,
* close fds if NOFD bit is set
+ * @vdev: vhost-user device
* @vmsg: vhost-user message
*/
-static void vu_check_queue_msg_file(struct vhost_user_msg *vmsg)
+static void vu_check_queue_msg_file(const struct vu_dev *vdev,
+ struct vhost_user_msg *vmsg)
{
bool nofd = vmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK;
- int idx = vmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
+ unsigned int idx = vmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
- if (idx >= VHOST_USER_MAX_VQS)
- die("Invalid vhost-user queue index: %u", idx);
+ if (idx >= vdev->context->max_virtqueues)
+ die("Invalid vhost-user queue index: %u (max_virtqueues %u)", idx,
+ vdev->context->max_virtqueues);
if (nofd) {
vmsg_close_fds(vmsg);
@@ -794,7 +798,7 @@ static bool vu_set_vring_kick_exec(struct vu_dev *vdev,
debug("u64: 0x%016"PRIx64, vmsg->payload.u64);
- vu_check_queue_msg_file(vmsg);
+ vu_check_queue_msg_file(vdev, vmsg);
if (vdev->vq[idx].kick_fd != -1) {
epoll_del(vdev->context->epollfd, vdev->vq[idx].kick_fd);
@@ -834,7 +838,7 @@ static bool vu_set_vring_call_exec(struct vu_dev *vdev,
debug("u64: 0x%016"PRIx64, vmsg->payload.u64);
- vu_check_queue_msg_file(vmsg);
+ vu_check_queue_msg_file(vdev, vmsg);
if (vdev->vq[idx].call_fd != -1) {
close(vdev->vq[idx].call_fd);
@@ -869,7 +873,7 @@ static bool vu_set_vring_err_exec(struct vu_dev *vdev,
debug("u64: 0x%016"PRIx64, vmsg->payload.u64);
- vu_check_queue_msg_file(vmsg);
+ vu_check_queue_msg_file(vdev, vmsg);
if (vdev->vq[idx].err_fd != -1) {
close(vdev->vq[idx].err_fd);
@@ -896,7 +900,8 @@ static bool vu_get_protocol_features_exec(struct vu_dev *vdev,
uint64_t features = 1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK |
1ULL << VHOST_USER_PROTOCOL_F_LOG_SHMFD |
1ULL << VHOST_USER_PROTOCOL_F_DEVICE_STATE |
- 1ULL << VHOST_USER_PROTOCOL_F_RARP;
+ 1ULL << VHOST_USER_PROTOCOL_F_RARP |
+ 1ULL << VHOST_USER_PROTOCOL_F_MQ;
(void)vdev;
vmsg_set_reply_u64(vmsg, features);
@@ -935,10 +940,9 @@ static bool vu_get_queue_num_exec(struct vu_dev *vdev,
{
(void)vdev;
- /* NOLINTNEXTLINE(misc-redundant-expression) */
- vmsg_set_reply_u64(vmsg, VHOST_USER_MAX_VQS / 2);
+ vmsg_set_reply_u64(vmsg, vdev->context->max_virtqueues / 2);
- debug("VHOST_USER_MAX_VQS %u", VHOST_USER_MAX_VQS / 2);
+ debug("max_virtqueues %u", vdev->context->max_virtqueues / 2);
return true;
}
@@ -959,7 +963,7 @@ static bool vu_set_vring_enable_exec(struct vu_dev *vdev,
debug("State.index: %u", idx);
debug("State.enable: %u", enable);
- if (idx >= VHOST_USER_MAX_VQS)
+ if (idx >= vdev->context->max_virtqueues)
die("Invalid vring_enable index: %u", idx);
vdev->vq[idx].enable = enable;
@@ -1047,7 +1051,7 @@ static bool vu_check_device_state_exec(struct vu_dev *vdev,
*/
void vu_init(struct ctx *c)
{
- int i;
+ unsigned int i;
c->vdev = &vdev_storage;
c->vdev->context = c;
@@ -1074,7 +1078,7 @@ void vu_cleanup(struct vu_dev *vdev)
{
unsigned int i;
- for (i = 0; i < VHOST_USER_MAX_VQS; i++) {
+ for (i = 0; i < vdev->context->max_virtqueues; i++) {
struct vu_virtq *vq = &vdev->vq[i];
vq->started = false;
diff --git a/virtio.h b/virtio.h
index 12caaa0b6def..176c935cecc7 100644
--- a/virtio.h
+++ b/virtio.h
@@ -88,7 +88,7 @@ struct vu_dev_region {
uint64_t mmap_addr;
};
-#define VHOST_USER_MAX_VQS 2
+#define VHOST_USER_MAX_VQS 32
/*
* Set a reasonable maximum number of ram slots, which will be supported by
--
2.51.0
^ permalink raw reply [flat|nested] 13+ messages in thread* Re: [PATCH 1/4] vhost-user: Enable multiqueue
2025-11-07 14:38 ` [PATCH 1/4] vhost-user: Enable multiqueue Laurent Vivier
@ 2025-11-10 4:48 ` David Gibson
2025-11-17 15:26 ` Laurent Vivier
0 siblings, 1 reply; 13+ messages in thread
From: David Gibson @ 2025-11-10 4:48 UTC (permalink / raw)
To: Laurent Vivier; +Cc: passt-dev
[-- Attachment #1: Type: text/plain, Size: 10970 bytes --]
On Fri, Nov 07, 2025 at 03:38:58PM +0100, Laurent Vivier wrote:
> Add the --max-queues parameter to specify the maximum number of queue
> pairs supported in vhost-user mode. This enables multi-queue support
> by allowing configuration of up to 16 queue pairs (32 virtqueues).
>
> For the moment, only the first RX queue is used, the TX queue is
> selected by the guest kernel.
>
> Signed-off-by: Laurent Vivier <lvivier@redhat.com>
> ---
> conf.c | 31 ++++++++++++++++++++++++++++++-
> passt.h | 2 ++
> tap.c | 15 +++++++++++++--
> vhost_user.c | 38 +++++++++++++++++++++-----------------
> virtio.h | 2 +-
> 5 files changed, 67 insertions(+), 21 deletions(-)
>
> diff --git a/conf.c b/conf.c
> index 66b9e63400ec..4738436e6245 100644
> --- a/conf.c
> +++ b/conf.c
> @@ -862,7 +862,9 @@ static void usage(const char *name, FILE *f, int status)
> " --vhost-user Enable vhost-user mode\n"
> " UNIX domain socket is provided by -s option\n"
> " --print-capabilities print back-end capabilities in JSON format,\n"
> - " only meaningful for vhost-user mode\n");
> + " only meaningful for vhost-user mode\n"
> + " --max-queues Specify the maximum number of queue pairs\n"
Nit: does the "max" actually add any relevant information. Maybe
"--queues", or "--queue-pairs" or "--qpairs"?
> + );
> FPRINTF(f,
> " --repair-path PATH path for passt-repair(1)\n"
> " default: append '.repair' to UNIX domain path\n");
> @@ -1483,6 +1485,7 @@ void conf(struct ctx *c, int argc, char **argv)
> {"migrate-exit", no_argument, NULL, 29 },
> {"migrate-no-linger", no_argument, NULL, 30 },
> {"stats", required_argument, NULL, 31 },
> + {"max-queues", required_argument, NULL, 32 },
> { 0 },
> };
> const char *optstring = "+dqfel:hs:F:I:p:P:m:a:n:M:g:i:o:D:S:H:461t:u:T:U:";
> @@ -1514,6 +1517,7 @@ void conf(struct ctx *c, int argc, char **argv)
> c->tcp.fwd_in.mode = c->tcp.fwd_out.mode = FWD_UNSET;
> c->udp.fwd_in.mode = c->udp.fwd_out.mode = FWD_UNSET;
> memcpy(c->our_tap_mac, MAC_OUR_LAA, ETH_ALEN);
> + c->max_virtqueues = 2;
For the internal variable name I'd especially suggest 'qpairs', for a)
brevity and b) presumbly this will be updated if we use less than the
max.
>
> optind = 0;
> do {
> @@ -1717,6 +1721,31 @@ void conf(struct ctx *c, int argc, char **argv)
> die("Can't display statistics if not running in foreground");
> c->stats = strtol(optarg, NULL, 0);
> break;
> + case 32: {
> + unsigned long max_queues;
> + char *e;
> +
> + if (c->mode != MODE_VU)
> + die("--max-queues is for vhost-user mode only");
> +
> + errno = 0;
> + max_queues = strtoul(optarg, &e, 0);
> +
> + if (errno || *e)
> + die("Invalid max-queues: %s", optarg);
> +
> + if (max_queues < 1) {
> + die("max-queues %lu too small (min 1)",
> + max_queues);
> + }
> +
> + if (max_queues * 2 > VHOST_USER_MAX_VQS) {
> + die("max-queues %lu too big (max %u)",
> + max_queues, VHOST_USER_MAX_VQS / 2);
> + }
> + c->max_virtqueues = max_queues * 2;
Oh, right, so the varuabke is actually the max number of queues, not
queue pairs. In that case, all the more reason to give it a different
name from the command line option.
That said, IIUC, queues must always come in pairs, so storing the
number of pairs would make a bit more sense to me than the number of
queues.
> + break;
> + }
> case 'd':
> c->debug = 1;
> c->quiet = 0;
> diff --git a/passt.h b/passt.h
> index 15801b44bfa8..e73f648009de 100644
> --- a/passt.h
> +++ b/passt.h
> @@ -205,6 +205,7 @@ struct ip6_ctx {
> * @low_wmem: Low probed net.core.wmem_max
> * @low_rmem: Low probed net.core.rmem_max
> * @vdev: vhost-user device
> + * @max_virtqueues: vhost-user maximum number of virtqueues
> * @device_state_fd: Device state migration channel
> * @device_state_result: Device state migration result
> * @migrate_target: Are we the target, on the next migration request?
> @@ -283,6 +284,7 @@ struct ctx {
> int low_rmem;
>
> struct vu_dev *vdev;
> + unsigned int max_virtqueues;
>
> /* Migration */
> int device_state_fd;
> diff --git a/tap.c b/tap.c
> index bb139d647bae..e18d693a665a 100644
> --- a/tap.c
> +++ b/tap.c
> @@ -1318,8 +1318,19 @@ static void tap_backend_show_hints(struct ctx *c)
> break;
> case MODE_VU:
> info("You can start qemu with:");
> - info(" kvm ... -chardev socket,id=chr0,path=%s -netdev vhost-user,id=netdev0,chardev=chr0 -device virtio-net,netdev=netdev0 -object memory-backend-memfd,id=memfd0,share=on,size=$RAMSIZE -numa node,memdev=memfd0\n",
> - c->sock_path);
> + if (c->max_virtqueues > 2) {
> + info(" kvm ... -chardev socket,id=chr0,path=%s "
> + "-netdev vhost-user,id=netdev0,chardev=chr0,queues=%d "
> + "-device virtio-net,netdev=netdev0,mq=true "
> + "-object memory-backend-memfd,id=memfd0,share=on,size=$RAMSIZE "
> + "-numa node,memdev=memfd0\n", c->sock_path, c->max_virtqueues / 2);
> + } else {
> + info(" kvm ... -chardev socket,id=chr0,path=%s "
> + "-netdev vhost-user,id=netdev0,chardev=chr0 "
> + "-device virtio-net,netdev=netdev0 "
> + "-object memory-backend-memfd,id=memfd0,share=on,size=$RAMSIZE "
> + "-numa node,memdev=memfd0\n", c->sock_path);
> + }
> break;
> }
> }
> diff --git a/vhost_user.c b/vhost_user.c
> index aa7c869d9e56..c668159e29b4 100644
> --- a/vhost_user.c
> +++ b/vhost_user.c
> @@ -323,6 +323,7 @@ static bool vu_get_features_exec(struct vu_dev *vdev,
> uint64_t features =
> 1ULL << VIRTIO_F_VERSION_1 |
> 1ULL << VIRTIO_NET_F_MRG_RXBUF |
> + 1ULL << VIRTIO_NET_F_MQ |
Should this be conditional on max_virtqueues > 2?
> 1ULL << VHOST_F_LOG_ALL |
> 1ULL << VHOST_USER_F_PROTOCOL_FEATURES;
>
> @@ -342,9 +343,9 @@ static bool vu_get_features_exec(struct vu_dev *vdev,
> */
> static void vu_set_enable_all_rings(struct vu_dev *vdev, bool enable)
> {
> - uint16_t i;
> + unsigned int i;
>
> - for (i = 0; i < VHOST_USER_MAX_VQS; i++)
> + for (i = 0; i < vdev->context->max_virtqueues; i++)
> vdev->vq[i].enable = enable;
> }
>
> @@ -476,7 +477,7 @@ static bool vu_set_mem_table_exec(struct vu_dev *vdev,
> close(vmsg->fds[i]);
> }
>
> - for (i = 0; i < VHOST_USER_MAX_VQS; i++) {
> + for (i = 0; i < vdev->context->max_virtqueues; i++) {
> if (vdev->vq[i].vring.desc) {
> if (map_ring(vdev, &vdev->vq[i]))
> die("remapping queue %d during setmemtable", i);
> @@ -759,15 +760,18 @@ static void vu_set_watch(const struct vu_dev *vdev, int idx)
> /**
> * vu_check_queue_msg_file() - Check if a message is valid,
> * close fds if NOFD bit is set
> + * @vdev: vhost-user device
> * @vmsg: vhost-user message
> */
> -static void vu_check_queue_msg_file(struct vhost_user_msg *vmsg)
> +static void vu_check_queue_msg_file(const struct vu_dev *vdev,
> + struct vhost_user_msg *vmsg)
> {
> bool nofd = vmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK;
> - int idx = vmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
> + unsigned int idx = vmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
>
> - if (idx >= VHOST_USER_MAX_VQS)
> - die("Invalid vhost-user queue index: %u", idx);
> + if (idx >= vdev->context->max_virtqueues)
> + die("Invalid vhost-user queue index: %u (max_virtqueues %u)", idx,
> + vdev->context->max_virtqueues);
>
> if (nofd) {
> vmsg_close_fds(vmsg);
> @@ -794,7 +798,7 @@ static bool vu_set_vring_kick_exec(struct vu_dev *vdev,
>
> debug("u64: 0x%016"PRIx64, vmsg->payload.u64);
>
> - vu_check_queue_msg_file(vmsg);
> + vu_check_queue_msg_file(vdev, vmsg);
>
> if (vdev->vq[idx].kick_fd != -1) {
> epoll_del(vdev->context->epollfd, vdev->vq[idx].kick_fd);
> @@ -834,7 +838,7 @@ static bool vu_set_vring_call_exec(struct vu_dev *vdev,
>
> debug("u64: 0x%016"PRIx64, vmsg->payload.u64);
>
> - vu_check_queue_msg_file(vmsg);
> + vu_check_queue_msg_file(vdev, vmsg);
>
> if (vdev->vq[idx].call_fd != -1) {
> close(vdev->vq[idx].call_fd);
> @@ -869,7 +873,7 @@ static bool vu_set_vring_err_exec(struct vu_dev *vdev,
>
> debug("u64: 0x%016"PRIx64, vmsg->payload.u64);
>
> - vu_check_queue_msg_file(vmsg);
> + vu_check_queue_msg_file(vdev, vmsg);
>
> if (vdev->vq[idx].err_fd != -1) {
> close(vdev->vq[idx].err_fd);
> @@ -896,7 +900,8 @@ static bool vu_get_protocol_features_exec(struct vu_dev *vdev,
> uint64_t features = 1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK |
> 1ULL << VHOST_USER_PROTOCOL_F_LOG_SHMFD |
> 1ULL << VHOST_USER_PROTOCOL_F_DEVICE_STATE |
> - 1ULL << VHOST_USER_PROTOCOL_F_RARP;
> + 1ULL << VHOST_USER_PROTOCOL_F_RARP |
> + 1ULL << VHOST_USER_PROTOCOL_F_MQ;
Again, should this be conditional?
>
> (void)vdev;
> vmsg_set_reply_u64(vmsg, features);
> @@ -935,10 +940,9 @@ static bool vu_get_queue_num_exec(struct vu_dev *vdev,
> {
> (void)vdev;
>
> - /* NOLINTNEXTLINE(misc-redundant-expression) */
> - vmsg_set_reply_u64(vmsg, VHOST_USER_MAX_VQS / 2);
> + vmsg_set_reply_u64(vmsg, vdev->context->max_virtqueues / 2);
>
> - debug("VHOST_USER_MAX_VQS %u", VHOST_USER_MAX_VQS / 2);
> + debug("max_virtqueues %u", vdev->context->max_virtqueues / 2);
>
> return true;
> }
> @@ -959,7 +963,7 @@ static bool vu_set_vring_enable_exec(struct vu_dev *vdev,
> debug("State.index: %u", idx);
> debug("State.enable: %u", enable);
>
> - if (idx >= VHOST_USER_MAX_VQS)
> + if (idx >= vdev->context->max_virtqueues)
> die("Invalid vring_enable index: %u", idx);
>
> vdev->vq[idx].enable = enable;
> @@ -1047,7 +1051,7 @@ static bool vu_check_device_state_exec(struct vu_dev *vdev,
> */
> void vu_init(struct ctx *c)
> {
> - int i;
> + unsigned int i;
>
> c->vdev = &vdev_storage;
> c->vdev->context = c;
> @@ -1074,7 +1078,7 @@ void vu_cleanup(struct vu_dev *vdev)
> {
> unsigned int i;
>
> - for (i = 0; i < VHOST_USER_MAX_VQS; i++) {
> + for (i = 0; i < vdev->context->max_virtqueues; i++) {
> struct vu_virtq *vq = &vdev->vq[i];
>
> vq->started = false;
> diff --git a/virtio.h b/virtio.h
> index 12caaa0b6def..176c935cecc7 100644
> --- a/virtio.h
> +++ b/virtio.h
> @@ -88,7 +88,7 @@ struct vu_dev_region {
> uint64_t mmap_addr;
> };
>
> -#define VHOST_USER_MAX_VQS 2
> +#define VHOST_USER_MAX_VQS 32
>
> /*
> * Set a reasonable maximum number of ram slots, which will be supported by
> --
> 2.51.0
>
--
David Gibson (he or they) | I'll have my music baroque, and my code
david AT gibson.dropbear.id.au | minimalist, thank you, not the other way
| around.
http://www.ozlabs.org/~dgibson
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]
^ permalink raw reply [flat|nested] 13+ messages in thread* Re: [PATCH 1/4] vhost-user: Enable multiqueue
2025-11-10 4:48 ` David Gibson
@ 2025-11-17 15:26 ` Laurent Vivier
2025-11-18 0:16 ` David Gibson
0 siblings, 1 reply; 13+ messages in thread
From: Laurent Vivier @ 2025-11-17 15:26 UTC (permalink / raw)
To: David Gibson; +Cc: passt-dev
On 11/10/25 05:48, David Gibson wrote:
>> index aa7c869d9e56..c668159e29b4 100644
>> --- a/vhost_user.c
>> +++ b/vhost_user.c
>> @@ -323,6 +323,7 @@ static bool vu_get_features_exec(struct vu_dev *vdev,
>> uint64_t features =
>> 1ULL << VIRTIO_F_VERSION_1 |
>> 1ULL << VIRTIO_NET_F_MRG_RXBUF |
>> + 1ULL << VIRTIO_NET_F_MQ |
> Should this be conditional on max_virtqueues > 2?
I don't think it's needed as 1 qpair (2 virtqueues) is a valid value with VIRTIO_NET_F_MQ:
https://docs.oasis-open.org/virtio/virtio/v1.3/csd01/virtio-v1.3-csd01.html#x1-2410001
"The device MUST set max_virtqueue_pairs to between 1 and 0x8000 inclusive, if it offers
VIRTIO_NET_F_MQ."
Thanks,
Laurent
^ permalink raw reply [flat|nested] 13+ messages in thread
* Re: [PATCH 1/4] vhost-user: Enable multiqueue
2025-11-17 15:26 ` Laurent Vivier
@ 2025-11-18 0:16 ` David Gibson
0 siblings, 0 replies; 13+ messages in thread
From: David Gibson @ 2025-11-18 0:16 UTC (permalink / raw)
To: Laurent Vivier; +Cc: passt-dev
[-- Attachment #1: Type: text/plain, Size: 1093 bytes --]
On Mon, Nov 17, 2025 at 04:26:43PM +0100, Laurent Vivier wrote:
> On 11/10/25 05:48, David Gibson wrote:
> > > index aa7c869d9e56..c668159e29b4 100644
> > > --- a/vhost_user.c
> > > +++ b/vhost_user.c
> > > @@ -323,6 +323,7 @@ static bool vu_get_features_exec(struct vu_dev *vdev,
> > > uint64_t features =
> > > 1ULL << VIRTIO_F_VERSION_1 |
> > > 1ULL << VIRTIO_NET_F_MRG_RXBUF |
> > > + 1ULL << VIRTIO_NET_F_MQ |
> > Should this be conditional on max_virtqueues > 2?
>
> I don't think it's needed as 1 qpair (2 virtqueues) is a valid value with VIRTIO_NET_F_MQ:
>
> https://docs.oasis-open.org/virtio/virtio/v1.3/csd01/virtio-v1.3-csd01.html#x1-2410001
>
> "The device MUST set max_virtqueue_pairs to between 1 and 0x8000 inclusive,
> if it offers VIRTIO_NET_F_MQ."
Ok, that makes sense. I think I forgot this was offered features, not
required features.
--
David Gibson (he or they) | I'll have my music baroque, and my code
david AT gibson.dropbear.id.au | minimalist, thank you, not the other way
| around.
http://www.ozlabs.org/~dgibson
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]
^ permalink raw reply [flat|nested] 13+ messages in thread
* [PATCH 2/4] vhost-user: Add queue parameter throughout the network stack
2025-11-07 14:38 [PATCH 0/4] vhost-user: Add multiqueue support Laurent Vivier
2025-11-07 14:38 ` [PATCH 1/4] vhost-user: Enable multiqueue Laurent Vivier
@ 2025-11-07 14:38 ` Laurent Vivier
2025-11-10 5:19 ` David Gibson
2025-11-07 14:39 ` [PATCH 3/4] multiqueue: Add queue-aware flow management for multiqueue support Laurent Vivier
` (2 subsequent siblings)
4 siblings, 1 reply; 13+ messages in thread
From: Laurent Vivier @ 2025-11-07 14:38 UTC (permalink / raw)
To: passt-dev; +Cc: Laurent Vivier
Add a queue parameter to vu_send_single() and propagate this parameter
through the entire network stack call chain. The queue parameter specifies
which virtqueue to use for sending packets in vhost-user mode.
Functions modified to accept and propagate the queue parameter:
- Core sending: tap_send_single(), vu_send_single()
- UDP/ICMP helpers: tap_udp4_send(), tap_udp6_send(), tap_icmp4_send(),
tap_icmp6_send()
- Protocol handlers: arp(), dhcp(), dhcpv6(), ndp(), tcp_rst_no_conn()
- NDP helpers: ndp_send(), ndp_na(), ndp_ra()
- UDP error handling: udp_send_tap_icmp4(), udp_send_tap_icmp6()
All callers currently pass VHOST_USER_RX_QUEUE to preserve existing
behavior. This is a preparatory step for enabling multi-queue and
per-queue worker threads in vhost-user mode.
No functional change.
Signed-off-by: Laurent Vivier <lvivier@redhat.com>
---
arp.c | 12 +++++++-----
arp.h | 4 ++--
dhcp.c | 5 +++--
dhcp.h | 2 +-
dhcpv6.c | 12 +++++++-----
dhcpv6.h | 2 +-
fwd.c | 4 ++--
icmp.c | 6 ++++--
ndp.c | 32 +++++++++++++++++++-------------
ndp.h | 5 +++--
tap.c | 38 +++++++++++++++++++++++---------------
tap.h | 13 +++++++------
tcp.c | 15 +++++++++------
tcp.h | 11 ++++++-----
udp.c | 17 ++++++++++-------
vu_common.c | 9 ++++++---
vu_common.h | 3 ++-
17 files changed, 112 insertions(+), 78 deletions(-)
diff --git a/arp.c b/arp.c
index 33b03cf6c316..c57e0471aee8 100644
--- a/arp.c
+++ b/arp.c
@@ -63,11 +63,12 @@ static bool ignore_arp(const struct ctx *c,
/**
* arp() - Check if this is a supported ARP message, reply as needed
* @c: Execution context
+ * @queue: Queue to use to send the reply
* @data: Single packet with Ethernet buffer
*
* Return: 1 if handled, -1 on failure
*/
-int arp(const struct ctx *c, struct iov_tail *data)
+int arp(const struct ctx *c, int queue, struct iov_tail *data)
{
union inany_addr tgt;
struct {
@@ -112,7 +113,7 @@ int arp(const struct ctx *c, struct iov_tail *data)
memcpy(resp.am.tha, am->sha, sizeof(resp.am.tha));
memcpy(resp.am.tip, am->sip, sizeof(resp.am.tip));
- tap_send_single(c, &resp, sizeof(resp));
+ tap_send_single(c, queue, &resp, sizeof(resp));
return 1;
}
@@ -148,16 +149,17 @@ void arp_send_init_req(const struct ctx *c)
memcpy(req.am.tip, &c->ip4.addr, sizeof(req.am.tip));
debug("Sending initial ARP request for guest MAC address");
- tap_send_single(c, &req, sizeof(req));
+ tap_send_single(c, VHOST_USER_RX_QUEUE, &req, sizeof(req));
}
/**
* arp_announce() - Send an ARP announcement for an IPv4 host
* @c: Execution context
+ * @queue: Queue to use to send the announcement
* @ip: IPv4 address we announce as owned by @mac
* @mac: MAC address to advertise for @ip
*/
-void arp_announce(const struct ctx *c, struct in_addr *ip,
+void arp_announce(const struct ctx *c, int queue, struct in_addr *ip,
const unsigned char *mac)
{
char ip_str[INET_ADDRSTRLEN];
@@ -199,5 +201,5 @@ void arp_announce(const struct ctx *c, struct in_addr *ip,
eth_ntop(mac, mac_str, sizeof(mac_str));
debug("ARP announcement for %s / %s", ip_str, mac_str);
- tap_send_single(c, &msg, sizeof(msg));
+ tap_send_single(c, queue, &msg, sizeof(msg));
}
diff --git a/arp.h b/arp.h
index 4862e90a14ee..ba2b1253a924 100644
--- a/arp.h
+++ b/arp.h
@@ -20,9 +20,9 @@ struct arpmsg {
unsigned char tip[4];
} __attribute__((__packed__));
-int arp(const struct ctx *c, struct iov_tail *data);
+int arp(const struct ctx *c, int queue, struct iov_tail *data);
void arp_send_init_req(const struct ctx *c);
-void arp_announce(const struct ctx *c, struct in_addr *ip,
+void arp_announce(const struct ctx *c, int queue, struct in_addr *ip,
const unsigned char *mac);
#endif /* ARP_H */
diff --git a/dhcp.c b/dhcp.c
index 6b9c2e3b9e5a..3d84065db9cb 100644
--- a/dhcp.c
+++ b/dhcp.c
@@ -296,11 +296,12 @@ static void opt_set_dns_search(const struct ctx *c, size_t max_len)
/**
* dhcp() - Check if this is a DHCP message, reply as needed
* @c: Execution context
+ * @queue: Queue to use to send the reply
* @data: Single packet with Ethernet buffer
*
* Return: 0 if it's not a DHCP message, 1 if handled, -1 on failure
*/
-int dhcp(const struct ctx *c, struct iov_tail *data)
+int dhcp(const struct ctx *c, int queue, struct iov_tail *data)
{
char macstr[ETH_ADDRSTRLEN];
size_t mlen, dlen, opt_len;
@@ -471,7 +472,7 @@ int dhcp(const struct ctx *c, struct iov_tail *data)
else
dst = c->ip4.addr;
- tap_udp4_send(c, c->ip4.our_tap_addr, 67, dst, 68, &reply, dlen);
+ tap_udp4_send(c, queue, c->ip4.our_tap_addr, 67, dst, 68, &reply, dlen);
return 1;
}
diff --git a/dhcp.h b/dhcp.h
index cd50c99b8856..847d957ff7d1 100644
--- a/dhcp.h
+++ b/dhcp.h
@@ -6,7 +6,7 @@
#ifndef DHCP_H
#define DHCP_H
-int dhcp(const struct ctx *c, struct iov_tail *data);
+int dhcp(const struct ctx *c, int queue, struct iov_tail *data);
void dhcp_init(void);
#endif /* DHCP_H */
diff --git a/dhcpv6.c b/dhcpv6.c
index e4df0db562e6..f1fa1d047c86 100644
--- a/dhcpv6.c
+++ b/dhcpv6.c
@@ -369,12 +369,13 @@ notonlink:
/**
* dhcpv6_send_ia_notonlink() - Send NotOnLink status
* @c: Execution context
+ * @queue: Queue to use to send the reply
* @ia_base: Non-appropriate IA_NA or IA_TA base
* @client_id_base: Client ID message option base
* @len: Client ID length
* @xid: Transaction ID for message exchange
*/
-static void dhcpv6_send_ia_notonlink(struct ctx *c,
+static void dhcpv6_send_ia_notonlink(struct ctx *c, int queue,
const struct iov_tail *ia_base,
const struct iov_tail *client_id_base,
int len, uint32_t xid)
@@ -404,7 +405,7 @@ static void dhcpv6_send_ia_notonlink(struct ctx *c,
resp_not_on_link.hdr.xid = xid;
- tap_udp6_send(c, src, 547, tap_ip6_daddr(c, src), 546,
+ tap_udp6_send(c, queue, src, 547, tap_ip6_daddr(c, src), 546,
xid, &resp_not_on_link, n);
}
@@ -539,13 +540,14 @@ static size_t dhcpv6_client_fqdn_fill(const struct iov_tail *data,
/**
* dhcpv6() - Check if this is a DHCPv6 message, reply as needed
* @c: Execution context
+ * @queue: Queue to use to send the reply
* @data: Single packet starting from UDP header
* @saddr: Source IPv6 address of original message
* @daddr: Destination IPv6 address of original message
*
* Return: 0 if it's not a DHCPv6 message, 1 if handled, -1 on failure
*/
-int dhcpv6(struct ctx *c, struct iov_tail *data,
+int dhcpv6(struct ctx *c, int queue, struct iov_tail *data,
const struct in6_addr *saddr, const struct in6_addr *daddr)
{
const struct opt_server_id *server_id = NULL;
@@ -627,7 +629,7 @@ int dhcpv6(struct ctx *c, struct iov_tail *data,
if (dhcpv6_ia_notonlink(data, &c->ip6.addr)) {
- dhcpv6_send_ia_notonlink(c, data, &client_id_base,
+ dhcpv6_send_ia_notonlink(c, queue, data, &client_id_base,
ntohs(client_id->l), mh->xid);
return 1;
@@ -677,7 +679,7 @@ int dhcpv6(struct ctx *c, struct iov_tail *data,
resp.hdr.xid = mh->xid;
- tap_udp6_send(c, src, 547, tap_ip6_daddr(c, src), 546,
+ tap_udp6_send(c, queue, src, 547, tap_ip6_daddr(c, src), 546,
mh->xid, &resp, n);
c->ip6.addr_seen = c->ip6.addr;
diff --git a/dhcpv6.h b/dhcpv6.h
index c706dfdbb2ac..049f338fa133 100644
--- a/dhcpv6.h
+++ b/dhcpv6.h
@@ -6,7 +6,7 @@
#ifndef DHCPV6_H
#define DHCPV6_H
-int dhcpv6(struct ctx *c, struct iov_tail *data,
+int dhcpv6(struct ctx *c, int queue, struct iov_tail *data,
struct in6_addr *saddr, struct in6_addr *daddr);
void dhcpv6_init(const struct ctx *c);
diff --git a/fwd.c b/fwd.c
index 68bb11663c46..30667ab10204 100644
--- a/fwd.c
+++ b/fwd.c
@@ -147,9 +147,9 @@ void fwd_neigh_table_update(const struct ctx *c, const union inany_addr *addr,
return;
if (inany_v4(addr))
- arp_announce(c, inany_v4(addr), e->mac);
+ arp_announce(c, VHOST_USER_RX_QUEUE, inany_v4(addr), e->mac);
else
- ndp_unsolicited_na(c, &addr->a6);
+ ndp_unsolicited_na(c, VHOST_USER_RX_QUEUE, &addr->a6);
}
/**
diff --git a/icmp.c b/icmp.c
index 35faefb91870..d58499c3bf5c 100644
--- a/icmp.c
+++ b/icmp.c
@@ -132,12 +132,14 @@ void icmp_sock_handler(const struct ctx *c, union epoll_ref ref)
const struct in_addr *daddr = inany_v4(&ini->eaddr);
ASSERT(saddr && daddr); /* Must have IPv4 addresses */
- tap_icmp4_send(c, *saddr, *daddr, buf, pingf->f.tap_omac, n);
+ tap_icmp4_send(c, VHOST_USER_RX_QUEUE, *saddr, *daddr, buf,
+ pingf->f.tap_omac, n);
} else if (pingf->f.type == FLOW_PING6) {
const struct in6_addr *saddr = &ini->oaddr.a6;
const struct in6_addr *daddr = &ini->eaddr.a6;
- tap_icmp6_send(c, saddr, daddr, buf, pingf->f.tap_omac, n);
+ tap_icmp6_send(c, VHOST_USER_RX_QUEUE, saddr, daddr, buf,
+ pingf->f.tap_omac, n);
}
return;
diff --git a/ndp.c b/ndp.c
index a33239d4aa81..a5f1b946ae0b 100644
--- a/ndp.c
+++ b/ndp.c
@@ -175,25 +175,27 @@ struct ndp_ns {
/**
* ndp_send() - Send an NDP message
* @c: Execution context
+ * @queue: Queue to use to send the message
* @dst: IPv6 address to send the message to
* @buf: ICMPv6 header + message payload
* @l4len: Length of message, including ICMPv6 header
*/
-static void ndp_send(const struct ctx *c, const struct in6_addr *dst,
+static void ndp_send(const struct ctx *c, int queue, const struct in6_addr *dst,
const void *buf, size_t l4len)
{
const struct in6_addr *src = &c->ip6.our_tap_ll;
- tap_icmp6_send(c, src, dst, buf, c->our_tap_mac, l4len);
+ tap_icmp6_send(c, queue, src, dst, buf, c->our_tap_mac, l4len);
}
/**
* ndp_na() - Send an NDP Neighbour Advertisement (NA) message
* @c: Execution context
+ * @queue: Queue to use to send the NA
* @dst: IPv6 address to send the NA to
* @addr: IPv6 address to advertise
*/
-static void ndp_na(const struct ctx *c, const struct in6_addr *dst,
+static void ndp_na(const struct ctx *c, int queue, const struct in6_addr *dst,
const struct in6_addr *addr)
{
union inany_addr tgt;
@@ -217,25 +219,28 @@ static void ndp_na(const struct ctx *c, const struct in6_addr *dst,
inany_from_af(&tgt, AF_INET6, addr);
fwd_neigh_mac_get(c, &tgt, na.target_l2_addr.mac);
- ndp_send(c, dst, &na, sizeof(na));
+ ndp_send(c, queue, dst, &na, sizeof(na));
}
/**
* ndp_unsolicited_na() - Send unsolicited NA
* @c: Execution context
+ * @queue: Queue to use to send the RA
* @addr: IPv6 address to advertise
*/
-void ndp_unsolicited_na(const struct ctx *c, const struct in6_addr *addr)
+void ndp_unsolicited_na(const struct ctx *c, int queue,
+ const struct in6_addr *addr)
{
- ndp_na(c, &in6addr_ll_all_nodes, addr);
+ ndp_na(c, queue, &in6addr_ll_all_nodes, addr);
}
/**
* ndp_ra() - Send an NDP Router Advertisement (RA) message
* @c: Execution context
+ * @queue: Queue to use to send the RA
* @dst: IPv6 address to send the RA to
*/
-static void ndp_ra(const struct ctx *c, const struct in6_addr *dst)
+static void ndp_ra(const struct ctx *c, int queue, const struct in6_addr *dst)
{
struct ndp_ra ra = {
.ih = {
@@ -341,18 +346,19 @@ static void ndp_ra(const struct ctx *c, const struct in6_addr *dst)
memcpy(&ra.source_ll.mac, c->our_tap_mac, ETH_ALEN);
/* NOLINTNEXTLINE(clang-analyzer-security.PointerSub) */
- ndp_send(c, dst, &ra, ptr - (unsigned char *)&ra);
+ ndp_send(c, queue, dst, &ra, ptr - (unsigned char *)&ra);
}
/**
* ndp() - Check for NDP solicitations, reply as needed
* @c: Execution context
+ * @queue: Queue to use to send replies
* @saddr: Source IPv6 address
* @data: Single packet with ICMPv6 header
*
* Return: 0 if not handled here, 1 if handled, -1 on failure
*/
-int ndp(const struct ctx *c, const struct in6_addr *saddr,
+int ndp(const struct ctx *c, int queue, const struct in6_addr *saddr,
struct iov_tail *data)
{
struct icmp6hdr ih_storage;
@@ -381,13 +387,13 @@ int ndp(const struct ctx *c, const struct in6_addr *saddr,
info("NDP: received NS, sending NA");
- ndp_na(c, saddr, &ns->target_addr);
+ ndp_na(c, queue, saddr, &ns->target_addr);
} else if (ih->icmp6_type == RS) {
if (c->no_ra)
return 1;
info("NDP: received RS, sending RA");
- ndp_ra(c, saddr);
+ ndp_ra(c, queue, saddr);
}
return 1;
@@ -445,7 +451,7 @@ void ndp_timer(const struct ctx *c, const struct timespec *now)
info("NDP: sending unsolicited RA, next in %llds", (long long)interval);
- ndp_ra(c, &in6addr_ll_all_nodes);
+ ndp_ra(c, VHOST_USER_RX_QUEUE, &in6addr_ll_all_nodes);
first:
next_ra = now->tv_sec + interval;
@@ -468,5 +474,5 @@ void ndp_send_init_req(const struct ctx *c)
.target_addr = c->ip6.addr
};
debug("Sending initial NDP NS request for guest MAC address");
- ndp_send(c, &c->ip6.addr, &ns, sizeof(ns));
+ ndp_send(c, VHOST_USER_RX_QUEUE, &c->ip6.addr, &ns, sizeof(ns));
}
diff --git a/ndp.h b/ndp.h
index 56b756d8400b..1a9622ee17e7 100644
--- a/ndp.h
+++ b/ndp.h
@@ -8,10 +8,11 @@
struct icmp6hdr;
-int ndp(const struct ctx *c, const struct in6_addr *saddr,
+int ndp(const struct ctx *c, int queue, const struct in6_addr *saddr,
struct iov_tail *data);
void ndp_timer(const struct ctx *c, const struct timespec *now);
void ndp_send_init_req(const struct ctx *c);
-void ndp_unsolicited_na(const struct ctx *c, const struct in6_addr *addr);
+void ndp_unsolicited_na(const struct ctx *c, int queue,
+ const struct in6_addr *addr);
#endif /* NDP_H */
diff --git a/tap.c b/tap.c
index e18d693a665a..1308d49242e8 100644
--- a/tap.c
+++ b/tap.c
@@ -125,10 +125,12 @@ unsigned long tap_l2_max_len(const struct ctx *c)
/**
* tap_send_single() - Send a single frame
* @c: Execution context
+ * @queue: Queue to use to send the frame
* @data: Packet buffer
* @l2len: Total L2 packet length
*/
-void tap_send_single(const struct ctx *c, const void *data, size_t l2len)
+void tap_send_single(const struct ctx *c, int queue, const void *data,
+ size_t l2len)
{
uint32_t vnet_len = htonl(l2len);
struct iovec iov[2];
@@ -147,7 +149,7 @@ void tap_send_single(const struct ctx *c, const void *data, size_t l2len)
tap_send_frames(c, iov, iovcnt, 1);
break;
case MODE_VU:
- vu_send_single(c, data, l2len);
+ vu_send_single(c, queue, data, l2len);
break;
}
}
@@ -250,6 +252,7 @@ void *tap_push_uh4(struct udphdr *uh, struct in_addr src, in_port_t sport,
/**
* tap_udp4_send() - Send UDP over IPv4 packet
* @c: Execution context
+ * @queue: Queue to use to send packet
* @src: IPv4 source address
* @sport: UDP source port
* @dst: IPv4 destination address
@@ -257,7 +260,7 @@ void *tap_push_uh4(struct udphdr *uh, struct in_addr src, in_port_t sport,
* @in: UDP payload contents (not including UDP header)
* @dlen: UDP payload length (not including UDP header)
*/
-void tap_udp4_send(const struct ctx *c, struct in_addr src, in_port_t sport,
+void tap_udp4_send(const struct ctx *c, int queue, struct in_addr src, in_port_t sport,
struct in_addr dst, in_port_t dport,
const void *in, size_t dlen)
{
@@ -268,20 +271,22 @@ void tap_udp4_send(const struct ctx *c, struct in_addr src, in_port_t sport,
char *data = tap_push_uh4(uh, src, sport, dst, dport, in, dlen);
memcpy(data, in, dlen);
- tap_send_single(c, buf, dlen + (data - buf));
+ tap_send_single(c, queue, buf, dlen + (data - buf));
}
/**
* tap_icmp4_send() - Send ICMPv4 packet
* @c: Execution context
+ * @queue: Queue to use to send packet
* @src: IPv4 source address
* @dst: IPv4 destination address
* @in: ICMP packet, including ICMP header
* @src_mac: MAC address to be used as source for message
* @l4len: ICMP packet length, including ICMP header
*/
-void tap_icmp4_send(const struct ctx *c, struct in_addr src, struct in_addr dst,
- const void *in, const void *src_mac, size_t l4len)
+void tap_icmp4_send(const struct ctx *c, int queue, struct in_addr src,
+ struct in_addr dst, const void *in, const void *src_mac,
+ size_t l4len)
{
char buf[USHRT_MAX];
struct iphdr *ip4h = tap_push_l2h(c, buf, src_mac, ETH_P_IP);
@@ -291,7 +296,7 @@ void tap_icmp4_send(const struct ctx *c, struct in_addr src, struct in_addr dst,
memcpy(icmp4h, in, l4len);
csum_icmp4(icmp4h, icmp4h + 1, l4len - sizeof(*icmp4h));
- tap_send_single(c, buf, l4len + ((char *)icmp4h - buf));
+ tap_send_single(c, queue, buf, l4len + ((char *)icmp4h - buf));
}
/**
@@ -355,6 +360,7 @@ void *tap_push_uh6(struct udphdr *uh,
/**
* tap_udp6_send() - Send UDP over IPv6 packet
* @c: Execution context
+ * @queue: Queue to use to send packet
* @src: IPv6 source address
* @sport: UDP source port
* @dst: IPv6 destination address
@@ -363,7 +369,7 @@ void *tap_push_uh6(struct udphdr *uh,
* @in: UDP payload contents (not including UDP header)
* @dlen: UDP payload length (not including UDP header)
*/
-void tap_udp6_send(const struct ctx *c,
+void tap_udp6_send(const struct ctx *c, int queue,
const struct in6_addr *src, in_port_t sport,
const struct in6_addr *dst, in_port_t dport,
uint32_t flow, void *in, size_t dlen)
@@ -376,19 +382,20 @@ void tap_udp6_send(const struct ctx *c,
char *data = tap_push_uh6(uh, src, sport, dst, dport, in, dlen);
memcpy(data, in, dlen);
- tap_send_single(c, buf, dlen + (data - buf));
+ tap_send_single(c, queue, buf, dlen + (data - buf));
}
/**
* tap_icmp6_send() - Send ICMPv6 packet
* @c: Execution context
+ * @queue: Queue to use to send packet
* @src: IPv6 source address
* @dst: IPv6 destination address
* @in: ICMP packet, including ICMP header
* @src_mac: MAC address to be used as source for message
* @l4len: ICMP packet length, including ICMP header
*/
-void tap_icmp6_send(const struct ctx *c,
+void tap_icmp6_send(const struct ctx *c, int queue,
const struct in6_addr *src, const struct in6_addr *dst,
const void *in, const void *src_mac, size_t l4len)
{
@@ -400,7 +407,7 @@ void tap_icmp6_send(const struct ctx *c,
memcpy(icmp6h, in, l4len);
csum_icmp6(icmp6h, src, dst, icmp6h + 1, l4len - sizeof(*icmp6h));
- tap_send_single(c, buf, l4len + ((char *)icmp6h - buf));
+ tap_send_single(c, queue, buf, l4len + ((char *)icmp6h - buf));
}
/**
@@ -729,7 +736,7 @@ resume:
if (!eh)
continue;
if (ntohs(eh->h_proto) == ETH_P_ARP) {
- arp(c, &data);
+ arp(c, VHOST_USER_RX_QUEUE, &data);
continue;
}
@@ -790,7 +797,7 @@ resume:
struct iov_tail eh_data;
packet_get(in, i, &eh_data);
- if (dhcp(c, &eh_data))
+ if (dhcp(c, VHOST_USER_RX_QUEUE, &eh_data))
continue;
}
@@ -958,7 +965,7 @@ resume:
continue;
ndp_data = data;
- if (ndp(c, saddr, &ndp_data))
+ if (ndp(c, VHOST_USER_RX_QUEUE, saddr, &ndp_data))
continue;
tap_packet_debug(NULL, ip6h, NULL, proto, NULL, 1);
@@ -977,7 +984,8 @@ resume:
if (proto == IPPROTO_UDP) {
struct iov_tail uh_data = data;
- if (dhcpv6(c, &uh_data, saddr, daddr))
+ if (dhcpv6(c, VHOST_USER_RX_QUEUE, &uh_data, saddr,
+ daddr))
continue;
}
diff --git a/tap.h b/tap.h
index 1864173cc9b0..76403a43edbc 100644
--- a/tap.h
+++ b/tap.h
@@ -87,24 +87,25 @@ void *tap_push_ip6h(struct ipv6hdr *ip6h,
const struct in6_addr *src,
const struct in6_addr *dst,
size_t l4len, uint8_t proto, uint32_t flow);
-void tap_udp4_send(const struct ctx *c, struct in_addr src, in_port_t sport,
+void tap_udp4_send(const struct ctx *c, int queue, struct in_addr src, in_port_t sport,
struct in_addr dst, in_port_t dport,
const void *in, size_t dlen);
-void tap_icmp4_send(const struct ctx *c, struct in_addr src, struct in_addr dst,
- const void *in, const void *src_mac, size_t l4len);
+void tap_icmp4_send(const struct ctx *c, int queue, struct in_addr src,
+ struct in_addr dst, const void *in, const void *src_mac,
+ size_t l4len);
const struct in6_addr *tap_ip6_daddr(const struct ctx *c,
const struct in6_addr *src);
void *tap_push_ip6h(struct ipv6hdr *ip6h,
const struct in6_addr *src, const struct in6_addr *dst,
size_t l4len, uint8_t proto, uint32_t flow);
-void tap_udp6_send(const struct ctx *c,
+void tap_udp6_send(const struct ctx *c, int queue,
const struct in6_addr *src, in_port_t sport,
const struct in6_addr *dst, in_port_t dport,
uint32_t flow, void *in, size_t dlen);
-void tap_icmp6_send(const struct ctx *c,
+void tap_icmp6_send(const struct ctx *c, int queue,
const struct in6_addr *src, const struct in6_addr *dst,
const void *in, const void *src_mac, size_t l4len);
-void tap_send_single(const struct ctx *c, const void *data, size_t l2len);
+void tap_send_single(const struct ctx *c, int queue, const void *data, size_t l2len);
size_t tap_send_frames(const struct ctx *c, const struct iovec *iov,
size_t bufs_per_frame, size_t nframes);
void eth_update_mac(struct ethhdr *eh,
diff --git a/tcp.c b/tcp.c
index e91c0cf5a441..5ce34baa8a5a 100644
--- a/tcp.c
+++ b/tcp.c
@@ -1984,6 +1984,7 @@ static void tcp_conn_from_sock_finish(const struct ctx *c,
/**
* tcp_rst_no_conn() - Send RST in response to a packet with no connection
* @c: Execution context
+ * @queue: Queue to use to send the reply
* @af: Address family, AF_INET or AF_INET6
* @saddr: Source address of the packet we're responding to
* @daddr: Destination address of the packet we're responding to
@@ -1991,7 +1992,7 @@ static void tcp_conn_from_sock_finish(const struct ctx *c,
* @th: TCP header of the packet we're responding to
* @l4len: Packet length, including TCP header
*/
-static void tcp_rst_no_conn(const struct ctx *c, int af,
+static void tcp_rst_no_conn(const struct ctx *c, int queue, int af,
const void *saddr, const void *daddr,
uint32_t flow_lbl,
const struct tcphdr *th, size_t l4len)
@@ -2049,7 +2050,7 @@ static void tcp_rst_no_conn(const struct ctx *c, int af,
tcp_update_csum(psum, rsth, &payload);
rst_l2len = ((char *)rsth - buf) + sizeof(*rsth);
- tap_send_single(c, buf, rst_l2len);
+ tap_send_single(c, queue, buf, rst_l2len);
}
/**
@@ -2066,9 +2067,10 @@ static void tcp_rst_no_conn(const struct ctx *c, int af,
*
* Return: count of consumed packets
*/
-int tcp_tap_handler(const struct ctx *c, uint8_t pif, sa_family_t af,
- const void *saddr, const void *daddr, uint32_t flow_lbl,
- const struct pool *p, int idx, const struct timespec *now)
+int tcp_tap_handler(const struct ctx *c, uint8_t pif,
+ sa_family_t af, const void *saddr, const void *daddr,
+ uint32_t flow_lbl, const struct pool *p, int idx,
+ const struct timespec *now)
{
struct tcp_tap_conn *conn;
struct tcphdr th_storage;
@@ -2108,7 +2110,8 @@ int tcp_tap_handler(const struct ctx *c, uint8_t pif, sa_family_t af,
tcp_conn_from_tap(c, af, saddr, daddr, th,
opts, optlen, now);
else
- tcp_rst_no_conn(c, af, saddr, daddr, flow_lbl, th, l4len);
+ tcp_rst_no_conn(c, VHOST_USER_RX_QUEUE, af, saddr,
+ daddr, flow_lbl, th, l4len);
return 1;
}
diff --git a/tcp.h b/tcp.h
index 0082386725c2..320683ce5679 100644
--- a/tcp.h
+++ b/tcp.h
@@ -13,11 +13,12 @@ struct ctx;
void tcp_timer_handler(const struct ctx *c, union epoll_ref ref);
void tcp_listen_handler(const struct ctx *c, union epoll_ref ref,
const struct timespec *now);
-void tcp_sock_handler(const struct ctx *c, union epoll_ref ref,
- uint32_t events);
-int tcp_tap_handler(const struct ctx *c, uint8_t pif, sa_family_t af,
- const void *saddr, const void *daddr, uint32_t flow_lbl,
- const struct pool *p, int idx, const struct timespec *now);
+void tcp_sock_handler(const struct ctx *c,
+ union epoll_ref ref, uint32_t events);
+int tcp_tap_handler(const struct ctx *c, uint8_t pif,
+ sa_family_t af, const void *saddr, const void *daddr,
+ uint32_t flow_lbl, const struct pool *p, int idx,
+ const struct timespec *now);
int tcp_sock_init(const struct ctx *c, const union inany_addr *addr,
const char *ifname, in_port_t port);
int tcp_init(struct ctx *c);
diff --git a/udp.c b/udp.c
index 9c00950250a0..868ffebb5802 100644
--- a/udp.c
+++ b/udp.c
@@ -384,13 +384,14 @@ static void udp_tap_prepare(const struct mmsghdr *mmh,
/**
* udp_send_tap_icmp4() - Construct and send ICMPv4 to local peer
* @c: Execution context
+ * @queue: Queue to send the ICMPv4 packet to
* @ee: Extended error descriptor
* @toside: Destination side of flow
* @saddr: Address of ICMP generating node
* @in: First bytes (max 8) of original UDP message body
* @dlen: Length of the read part of original UDP message body
*/
-static void udp_send_tap_icmp4(const struct ctx *c,
+static void udp_send_tap_icmp4(const struct ctx *c, int queue,
const struct sock_extended_err *ee,
const struct flowside *toside,
struct in_addr saddr,
@@ -426,13 +427,14 @@ static void udp_send_tap_icmp4(const struct ctx *c,
/* Try to obtain the MAC address of the generating node */
saddr_any = inany_from_v4(saddr);
fwd_neigh_mac_get(c, &saddr_any, tap_omac);
- tap_icmp4_send(c, saddr, eaddr, &msg, tap_omac, msglen);
+ tap_icmp4_send(c, queue, saddr, eaddr, &msg, tap_omac, msglen);
}
/**
* udp_send_tap_icmp6() - Construct and send ICMPv6 to local peer
* @c: Execution context
+ * @queue: Queue to send the ICMPv6 packet to
* @ee: Extended error descriptor
* @toside: Destination side of flow
* @saddr: Address of ICMP generating node
@@ -440,7 +442,7 @@ static void udp_send_tap_icmp4(const struct ctx *c,
* @dlen: Length of the read part of original UDP message body
* @flow: IPv6 flow identifier
*/
-static void udp_send_tap_icmp6(const struct ctx *c,
+static void udp_send_tap_icmp6(const struct ctx *c, int queue,
const struct sock_extended_err *ee,
const struct flowside *toside,
const struct in6_addr *saddr,
@@ -474,7 +476,7 @@ static void udp_send_tap_icmp6(const struct ctx *c,
/* Try to obtain the MAC address of the generating node */
fwd_neigh_mac_get(c, (union inany_addr *) saddr, tap_omac);
- tap_icmp6_send(c, saddr, eaddr, &msg, tap_omac, msglen);
+ tap_icmp6_send(c, queue, saddr, eaddr, &msg, tap_omac, msglen);
}
/**
@@ -634,13 +636,14 @@ static int udp_sock_recverr(const struct ctx *c, int s, flow_sidx_t sidx,
if (hdr->cmsg_level == IPPROTO_IP &&
(o4 = inany_v4(&otap)) && inany_v4(&toside->eaddr)) {
dlen = MIN(dlen, ICMP4_MAX_DLEN);
- udp_send_tap_icmp4(c, ee, toside, *o4, data, dlen);
+ udp_send_tap_icmp4(c, VHOST_USER_RX_QUEUE, ee, toside, *o4,
+ data, dlen);
return 1;
}
if (hdr->cmsg_level == IPPROTO_IPV6 && !inany_v4(&toside->eaddr)) {
- udp_send_tap_icmp6(c, ee, toside, &otap.a6, data, dlen,
- FLOW_IDX(uflow));
+ udp_send_tap_icmp6(c, VHOST_USER_RX_QUEUE, ee, toside,
+ &otap.a6, data, dlen, FLOW_IDX(uflow));
return 1;
}
diff --git a/vu_common.c b/vu_common.c
index b13b7c308fd8..8904403e66af 100644
--- a/vu_common.c
+++ b/vu_common.c
@@ -235,23 +235,26 @@ void vu_kick_cb(struct vu_dev *vdev, union epoll_ref ref,
}
/**
- * vu_send_single() - Send a buffer to the front-end using the RX virtqueue
+ * vu_send_single() - Send a buffer to the front-end using a specified virtqueue
* @c: execution context
+ * @queue: queue to use to send the buffer
* @buf: address of the buffer
* @size: size of the buffer
*
* Return: number of bytes sent, -1 if there is an error
*/
-int vu_send_single(const struct ctx *c, const void *buf, size_t size)
+int vu_send_single(const struct ctx *c, int queue, const void *buf, size_t size)
{
struct vu_dev *vdev = c->vdev;
- struct vu_virtq *vq = &vdev->vq[VHOST_USER_RX_QUEUE];
struct vu_virtq_element elem[VIRTQUEUE_MAX_SIZE];
struct iovec in_sg[VIRTQUEUE_MAX_SIZE];
+ struct vu_virtq *vq;
size_t total;
int elem_cnt;
int i;
+ vq = &vdev->vq[queue];
+
trace("vu_send_single size %zu", size);
if (!vu_queue_enabled(vq) || !vu_queue_started(vq)) {
diff --git a/vu_common.h b/vu_common.h
index f538f237790b..2637c67f913c 100644
--- a/vu_common.h
+++ b/vu_common.h
@@ -56,6 +56,7 @@ void vu_flush(const struct vu_dev *vdev, struct vu_virtq *vq,
struct vu_virtq_element *elem, int elem_cnt);
void vu_kick_cb(struct vu_dev *vdev, union epoll_ref ref,
const struct timespec *now);
-int vu_send_single(const struct ctx *c, const void *buf, size_t size);
+int vu_send_single(const struct ctx *c, int queue, const void *buf,
+ size_t size);
#endif /* VU_COMMON_H */
--
2.51.0
^ permalink raw reply [flat|nested] 13+ messages in thread* Re: [PATCH 2/4] vhost-user: Add queue parameter throughout the network stack
2025-11-07 14:38 ` [PATCH 2/4] vhost-user: Add queue parameter throughout the network stack Laurent Vivier
@ 2025-11-10 5:19 ` David Gibson
0 siblings, 0 replies; 13+ messages in thread
From: David Gibson @ 2025-11-10 5:19 UTC (permalink / raw)
To: Laurent Vivier; +Cc: passt-dev
[-- Attachment #1: Type: text/plain, Size: 31732 bytes --]
On Fri, Nov 07, 2025 at 03:38:59PM +0100, Laurent Vivier wrote:
> Add a queue parameter to vu_send_single() and propagate this parameter
> through the entire network stack call chain. The queue parameter specifies
> which virtqueue to use for sending packets in vhost-user mode.
>
> Functions modified to accept and propagate the queue parameter:
> - Core sending: tap_send_single(), vu_send_single()
> - UDP/ICMP helpers: tap_udp4_send(), tap_udp6_send(), tap_icmp4_send(),
> tap_icmp6_send()
> - Protocol handlers: arp(), dhcp(), dhcpv6(), ndp(), tcp_rst_no_conn()
> - NDP helpers: ndp_send(), ndp_na(), ndp_ra()
> - UDP error handling: udp_send_tap_icmp4(), udp_send_tap_icmp6()
>
> All callers currently pass VHOST_USER_RX_QUEUE to preserve existing
> behavior. This is a preparatory step for enabling multi-queue and
> per-queue worker threads in vhost-user mode.
>
> No functional change.
>
> Signed-off-by: Laurent Vivier <lvivier@redhat.com>
LGTM, with some minor nits noted below.
> ---
> arp.c | 12 +++++++-----
> arp.h | 4 ++--
> dhcp.c | 5 +++--
> dhcp.h | 2 +-
> dhcpv6.c | 12 +++++++-----
> dhcpv6.h | 2 +-
> fwd.c | 4 ++--
> icmp.c | 6 ++++--
> ndp.c | 32 +++++++++++++++++++-------------
> ndp.h | 5 +++--
> tap.c | 38 +++++++++++++++++++++++---------------
> tap.h | 13 +++++++------
> tcp.c | 15 +++++++++------
> tcp.h | 11 ++++++-----
> udp.c | 17 ++++++++++-------
> vu_common.c | 9 ++++++---
> vu_common.h | 3 ++-
> 17 files changed, 112 insertions(+), 78 deletions(-)
>
> diff --git a/arp.c b/arp.c
> index 33b03cf6c316..c57e0471aee8 100644
> --- a/arp.c
> +++ b/arp.c
> @@ -63,11 +63,12 @@ static bool ignore_arp(const struct ctx *c,
> /**
> * arp() - Check if this is a supported ARP message, reply as needed
> * @c: Execution context
> + * @queue: Queue to use to send the reply
Nit: I think "Queue on which to..." reads a little better.
> * @data: Single packet with Ethernet buffer
> *
> * Return: 1 if handled, -1 on failure
> */
> -int arp(const struct ctx *c, struct iov_tail *data)
> +int arp(const struct ctx *c, int queue, struct iov_tail *data)
> {
> union inany_addr tgt;
> struct {
> @@ -112,7 +113,7 @@ int arp(const struct ctx *c, struct iov_tail *data)
> memcpy(resp.am.tha, am->sha, sizeof(resp.am.tha));
> memcpy(resp.am.tip, am->sip, sizeof(resp.am.tip));
>
> - tap_send_single(c, &resp, sizeof(resp));
> + tap_send_single(c, queue, &resp, sizeof(resp));
>
> return 1;
> }
> @@ -148,16 +149,17 @@ void arp_send_init_req(const struct ctx *c)
> memcpy(req.am.tip, &c->ip4.addr, sizeof(req.am.tip));
>
> debug("Sending initial ARP request for guest MAC address");
> - tap_send_single(c, &req, sizeof(req));
> + tap_send_single(c, VHOST_USER_RX_QUEUE, &req, sizeof(req));
Not really in scope for this patch, but I always get a little bit
tripped sending things to an "Rx" queue, until I remember that it's Rx
from the guest's perspective. I'm not sure how to improve that
without making it more confusing somewhere else. Maybe naming the
constants "TOGUEST" queue and "FROMGUEST" queue? Too verbose?
> }
>
> /**
> * arp_announce() - Send an ARP announcement for an IPv4 host
> * @c: Execution context
> + * @queue: Queue to use to send the announcement
> * @ip: IPv4 address we announce as owned by @mac
> * @mac: MAC address to advertise for @ip
> */
> -void arp_announce(const struct ctx *c, struct in_addr *ip,
> +void arp_announce(const struct ctx *c, int queue, struct in_addr *ip,
> const unsigned char *mac)
> {
> char ip_str[INET_ADDRSTRLEN];
> @@ -199,5 +201,5 @@ void arp_announce(const struct ctx *c, struct in_addr *ip,
> eth_ntop(mac, mac_str, sizeof(mac_str));
> debug("ARP announcement for %s / %s", ip_str, mac_str);
>
> - tap_send_single(c, &msg, sizeof(msg));
> + tap_send_single(c, queue, &msg, sizeof(msg));
> }
> diff --git a/arp.h b/arp.h
> index 4862e90a14ee..ba2b1253a924 100644
> --- a/arp.h
> +++ b/arp.h
> @@ -20,9 +20,9 @@ struct arpmsg {
> unsigned char tip[4];
> } __attribute__((__packed__));
>
> -int arp(const struct ctx *c, struct iov_tail *data);
> +int arp(const struct ctx *c, int queue, struct iov_tail *data);
> void arp_send_init_req(const struct ctx *c);
> -void arp_announce(const struct ctx *c, struct in_addr *ip,
> +void arp_announce(const struct ctx *c, int queue, struct in_addr *ip,
> const unsigned char *mac);
>
> #endif /* ARP_H */
> diff --git a/dhcp.c b/dhcp.c
> index 6b9c2e3b9e5a..3d84065db9cb 100644
> --- a/dhcp.c
> +++ b/dhcp.c
> @@ -296,11 +296,12 @@ static void opt_set_dns_search(const struct ctx *c, size_t max_len)
> /**
> * dhcp() - Check if this is a DHCP message, reply as needed
> * @c: Execution context
> + * @queue: Queue to use to send the reply
> * @data: Single packet with Ethernet buffer
> *
> * Return: 0 if it's not a DHCP message, 1 if handled, -1 on failure
> */
> -int dhcp(const struct ctx *c, struct iov_tail *data)
> +int dhcp(const struct ctx *c, int queue, struct iov_tail *data)
> {
> char macstr[ETH_ADDRSTRLEN];
> size_t mlen, dlen, opt_len;
> @@ -471,7 +472,7 @@ int dhcp(const struct ctx *c, struct iov_tail *data)
> else
> dst = c->ip4.addr;
>
> - tap_udp4_send(c, c->ip4.our_tap_addr, 67, dst, 68, &reply, dlen);
> + tap_udp4_send(c, queue, c->ip4.our_tap_addr, 67, dst, 68, &reply, dlen);
>
> return 1;
> }
> diff --git a/dhcp.h b/dhcp.h
> index cd50c99b8856..847d957ff7d1 100644
> --- a/dhcp.h
> +++ b/dhcp.h
> @@ -6,7 +6,7 @@
> #ifndef DHCP_H
> #define DHCP_H
>
> -int dhcp(const struct ctx *c, struct iov_tail *data);
> +int dhcp(const struct ctx *c, int queue, struct iov_tail *data);
> void dhcp_init(void);
>
> #endif /* DHCP_H */
> diff --git a/dhcpv6.c b/dhcpv6.c
> index e4df0db562e6..f1fa1d047c86 100644
> --- a/dhcpv6.c
> +++ b/dhcpv6.c
> @@ -369,12 +369,13 @@ notonlink:
> /**
> * dhcpv6_send_ia_notonlink() - Send NotOnLink status
> * @c: Execution context
> + * @queue: Queue to use to send the reply
> * @ia_base: Non-appropriate IA_NA or IA_TA base
> * @client_id_base: Client ID message option base
> * @len: Client ID length
> * @xid: Transaction ID for message exchange
> */
> -static void dhcpv6_send_ia_notonlink(struct ctx *c,
> +static void dhcpv6_send_ia_notonlink(struct ctx *c, int queue,
> const struct iov_tail *ia_base,
> const struct iov_tail *client_id_base,
> int len, uint32_t xid)
> @@ -404,7 +405,7 @@ static void dhcpv6_send_ia_notonlink(struct ctx *c,
>
> resp_not_on_link.hdr.xid = xid;
>
> - tap_udp6_send(c, src, 547, tap_ip6_daddr(c, src), 546,
> + tap_udp6_send(c, queue, src, 547, tap_ip6_daddr(c, src), 546,
> xid, &resp_not_on_link, n);
> }
>
> @@ -539,13 +540,14 @@ static size_t dhcpv6_client_fqdn_fill(const struct iov_tail *data,
> /**
> * dhcpv6() - Check if this is a DHCPv6 message, reply as needed
> * @c: Execution context
> + * @queue: Queue to use to send the reply
> * @data: Single packet starting from UDP header
> * @saddr: Source IPv6 address of original message
> * @daddr: Destination IPv6 address of original message
> *
> * Return: 0 if it's not a DHCPv6 message, 1 if handled, -1 on failure
> */
> -int dhcpv6(struct ctx *c, struct iov_tail *data,
> +int dhcpv6(struct ctx *c, int queue, struct iov_tail *data,
> const struct in6_addr *saddr, const struct in6_addr *daddr)
> {
> const struct opt_server_id *server_id = NULL;
> @@ -627,7 +629,7 @@ int dhcpv6(struct ctx *c, struct iov_tail *data,
>
> if (dhcpv6_ia_notonlink(data, &c->ip6.addr)) {
>
> - dhcpv6_send_ia_notonlink(c, data, &client_id_base,
> + dhcpv6_send_ia_notonlink(c, queue, data, &client_id_base,
> ntohs(client_id->l), mh->xid);
>
> return 1;
> @@ -677,7 +679,7 @@ int dhcpv6(struct ctx *c, struct iov_tail *data,
>
> resp.hdr.xid = mh->xid;
>
> - tap_udp6_send(c, src, 547, tap_ip6_daddr(c, src), 546,
> + tap_udp6_send(c, queue, src, 547, tap_ip6_daddr(c, src), 546,
> mh->xid, &resp, n);
> c->ip6.addr_seen = c->ip6.addr;
>
> diff --git a/dhcpv6.h b/dhcpv6.h
> index c706dfdbb2ac..049f338fa133 100644
> --- a/dhcpv6.h
> +++ b/dhcpv6.h
> @@ -6,7 +6,7 @@
> #ifndef DHCPV6_H
> #define DHCPV6_H
>
> -int dhcpv6(struct ctx *c, struct iov_tail *data,
> +int dhcpv6(struct ctx *c, int queue, struct iov_tail *data,
> struct in6_addr *saddr, struct in6_addr *daddr);
> void dhcpv6_init(const struct ctx *c);
>
> diff --git a/fwd.c b/fwd.c
> index 68bb11663c46..30667ab10204 100644
> --- a/fwd.c
> +++ b/fwd.c
> @@ -147,9 +147,9 @@ void fwd_neigh_table_update(const struct ctx *c, const union inany_addr *addr,
> return;
>
> if (inany_v4(addr))
> - arp_announce(c, inany_v4(addr), e->mac);
> + arp_announce(c, VHOST_USER_RX_QUEUE, inany_v4(addr), e->mac);
> else
> - ndp_unsolicited_na(c, &addr->a6);
> + ndp_unsolicited_na(c, VHOST_USER_RX_QUEUE, &addr->a6);
> }
>
> /**
> diff --git a/icmp.c b/icmp.c
> index 35faefb91870..d58499c3bf5c 100644
> --- a/icmp.c
> +++ b/icmp.c
> @@ -132,12 +132,14 @@ void icmp_sock_handler(const struct ctx *c, union epoll_ref ref)
> const struct in_addr *daddr = inany_v4(&ini->eaddr);
>
> ASSERT(saddr && daddr); /* Must have IPv4 addresses */
> - tap_icmp4_send(c, *saddr, *daddr, buf, pingf->f.tap_omac, n);
> + tap_icmp4_send(c, VHOST_USER_RX_QUEUE, *saddr, *daddr, buf,
> + pingf->f.tap_omac, n);
> } else if (pingf->f.type == FLOW_PING6) {
> const struct in6_addr *saddr = &ini->oaddr.a6;
> const struct in6_addr *daddr = &ini->eaddr.a6;
>
> - tap_icmp6_send(c, saddr, daddr, buf, pingf->f.tap_omac, n);
> + tap_icmp6_send(c, VHOST_USER_RX_QUEUE, saddr, daddr, buf,
> + pingf->f.tap_omac, n);
> }
> return;
>
> diff --git a/ndp.c b/ndp.c
> index a33239d4aa81..a5f1b946ae0b 100644
> --- a/ndp.c
> +++ b/ndp.c
> @@ -175,25 +175,27 @@ struct ndp_ns {
> /**
> * ndp_send() - Send an NDP message
> * @c: Execution context
> + * @queue: Queue to use to send the message
> * @dst: IPv6 address to send the message to
> * @buf: ICMPv6 header + message payload
> * @l4len: Length of message, including ICMPv6 header
> */
> -static void ndp_send(const struct ctx *c, const struct in6_addr *dst,
> +static void ndp_send(const struct ctx *c, int queue, const struct in6_addr *dst,
> const void *buf, size_t l4len)
> {
> const struct in6_addr *src = &c->ip6.our_tap_ll;
>
> - tap_icmp6_send(c, src, dst, buf, c->our_tap_mac, l4len);
> + tap_icmp6_send(c, queue, src, dst, buf, c->our_tap_mac, l4len);
> }
>
> /**
> * ndp_na() - Send an NDP Neighbour Advertisement (NA) message
> * @c: Execution context
> + * @queue: Queue to use to send the NA
> * @dst: IPv6 address to send the NA to
> * @addr: IPv6 address to advertise
> */
> -static void ndp_na(const struct ctx *c, const struct in6_addr *dst,
> +static void ndp_na(const struct ctx *c, int queue, const struct in6_addr *dst,
> const struct in6_addr *addr)
> {
> union inany_addr tgt;
> @@ -217,25 +219,28 @@ static void ndp_na(const struct ctx *c, const struct in6_addr *dst,
> inany_from_af(&tgt, AF_INET6, addr);
> fwd_neigh_mac_get(c, &tgt, na.target_l2_addr.mac);
>
> - ndp_send(c, dst, &na, sizeof(na));
> + ndp_send(c, queue, dst, &na, sizeof(na));
> }
>
> /**
> * ndp_unsolicited_na() - Send unsolicited NA
> * @c: Execution context
> + * @queue: Queue to use to send the RA
> * @addr: IPv6 address to advertise
> */
> -void ndp_unsolicited_na(const struct ctx *c, const struct in6_addr *addr)
> +void ndp_unsolicited_na(const struct ctx *c, int queue,
> + const struct in6_addr *addr)
> {
> - ndp_na(c, &in6addr_ll_all_nodes, addr);
> + ndp_na(c, queue, &in6addr_ll_all_nodes, addr);
> }
>
> /**
> * ndp_ra() - Send an NDP Router Advertisement (RA) message
> * @c: Execution context
> + * @queue: Queue to use to send the RA
> * @dst: IPv6 address to send the RA to
> */
> -static void ndp_ra(const struct ctx *c, const struct in6_addr *dst)
> +static void ndp_ra(const struct ctx *c, int queue, const struct in6_addr *dst)
> {
> struct ndp_ra ra = {
> .ih = {
> @@ -341,18 +346,19 @@ static void ndp_ra(const struct ctx *c, const struct in6_addr *dst)
> memcpy(&ra.source_ll.mac, c->our_tap_mac, ETH_ALEN);
>
> /* NOLINTNEXTLINE(clang-analyzer-security.PointerSub) */
> - ndp_send(c, dst, &ra, ptr - (unsigned char *)&ra);
> + ndp_send(c, queue, dst, &ra, ptr - (unsigned char *)&ra);
> }
>
> /**
> * ndp() - Check for NDP solicitations, reply as needed
> * @c: Execution context
> + * @queue: Queue to use to send replies
> * @saddr: Source IPv6 address
> * @data: Single packet with ICMPv6 header
> *
> * Return: 0 if not handled here, 1 if handled, -1 on failure
> */
> -int ndp(const struct ctx *c, const struct in6_addr *saddr,
> +int ndp(const struct ctx *c, int queue, const struct in6_addr *saddr,
> struct iov_tail *data)
> {
> struct icmp6hdr ih_storage;
> @@ -381,13 +387,13 @@ int ndp(const struct ctx *c, const struct in6_addr *saddr,
>
> info("NDP: received NS, sending NA");
>
> - ndp_na(c, saddr, &ns->target_addr);
> + ndp_na(c, queue, saddr, &ns->target_addr);
> } else if (ih->icmp6_type == RS) {
> if (c->no_ra)
> return 1;
>
> info("NDP: received RS, sending RA");
> - ndp_ra(c, saddr);
> + ndp_ra(c, queue, saddr);
> }
>
> return 1;
> @@ -445,7 +451,7 @@ void ndp_timer(const struct ctx *c, const struct timespec *now)
>
> info("NDP: sending unsolicited RA, next in %llds", (long long)interval);
>
> - ndp_ra(c, &in6addr_ll_all_nodes);
> + ndp_ra(c, VHOST_USER_RX_QUEUE, &in6addr_ll_all_nodes);
>
> first:
> next_ra = now->tv_sec + interval;
> @@ -468,5 +474,5 @@ void ndp_send_init_req(const struct ctx *c)
> .target_addr = c->ip6.addr
> };
> debug("Sending initial NDP NS request for guest MAC address");
> - ndp_send(c, &c->ip6.addr, &ns, sizeof(ns));
> + ndp_send(c, VHOST_USER_RX_QUEUE, &c->ip6.addr, &ns, sizeof(ns));
> }
> diff --git a/ndp.h b/ndp.h
> index 56b756d8400b..1a9622ee17e7 100644
> --- a/ndp.h
> +++ b/ndp.h
> @@ -8,10 +8,11 @@
>
> struct icmp6hdr;
>
> -int ndp(const struct ctx *c, const struct in6_addr *saddr,
> +int ndp(const struct ctx *c, int queue, const struct in6_addr *saddr,
> struct iov_tail *data);
> void ndp_timer(const struct ctx *c, const struct timespec *now);
> void ndp_send_init_req(const struct ctx *c);
> -void ndp_unsolicited_na(const struct ctx *c, const struct in6_addr *addr);
> +void ndp_unsolicited_na(const struct ctx *c, int queue,
> + const struct in6_addr *addr);
>
> #endif /* NDP_H */
> diff --git a/tap.c b/tap.c
> index e18d693a665a..1308d49242e8 100644
> --- a/tap.c
> +++ b/tap.c
> @@ -125,10 +125,12 @@ unsigned long tap_l2_max_len(const struct ctx *c)
> /**
> * tap_send_single() - Send a single frame
> * @c: Execution context
> + * @queue: Queue to use to send the frame
> * @data: Packet buffer
> * @l2len: Total L2 packet length
> */
> -void tap_send_single(const struct ctx *c, const void *data, size_t l2len)
> +void tap_send_single(const struct ctx *c, int queue, const void *data,
> + size_t l2len)
> {
> uint32_t vnet_len = htonl(l2len);
> struct iovec iov[2];
> @@ -147,7 +149,7 @@ void tap_send_single(const struct ctx *c, const void *data, size_t l2len)
> tap_send_frames(c, iov, iovcnt, 1);
> break;
> case MODE_VU:
> - vu_send_single(c, data, l2len);
> + vu_send_single(c, queue, data, l2len);
> break;
> }
> }
> @@ -250,6 +252,7 @@ void *tap_push_uh4(struct udphdr *uh, struct in_addr src, in_port_t sport,
> /**
> * tap_udp4_send() - Send UDP over IPv4 packet
> * @c: Execution context
> + * @queue: Queue to use to send packet
> * @src: IPv4 source address
> * @sport: UDP source port
> * @dst: IPv4 destination address
> @@ -257,7 +260,7 @@ void *tap_push_uh4(struct udphdr *uh, struct in_addr src, in_port_t sport,
> * @in: UDP payload contents (not including UDP header)
> * @dlen: UDP payload length (not including UDP header)
> */
> -void tap_udp4_send(const struct ctx *c, struct in_addr src, in_port_t sport,
> +void tap_udp4_send(const struct ctx *c, int queue, struct in_addr src, in_port_t sport,
> struct in_addr dst, in_port_t dport,
> const void *in, size_t dlen)
> {
> @@ -268,20 +271,22 @@ void tap_udp4_send(const struct ctx *c, struct in_addr src, in_port_t sport,
> char *data = tap_push_uh4(uh, src, sport, dst, dport, in, dlen);
>
> memcpy(data, in, dlen);
> - tap_send_single(c, buf, dlen + (data - buf));
> + tap_send_single(c, queue, buf, dlen + (data - buf));
> }
>
> /**
> * tap_icmp4_send() - Send ICMPv4 packet
> * @c: Execution context
> + * @queue: Queue to use to send packet
> * @src: IPv4 source address
> * @dst: IPv4 destination address
> * @in: ICMP packet, including ICMP header
> * @src_mac: MAC address to be used as source for message
> * @l4len: ICMP packet length, including ICMP header
> */
> -void tap_icmp4_send(const struct ctx *c, struct in_addr src, struct in_addr dst,
> - const void *in, const void *src_mac, size_t l4len)
> +void tap_icmp4_send(const struct ctx *c, int queue, struct in_addr src,
> + struct in_addr dst, const void *in, const void *src_mac,
> + size_t l4len)
> {
> char buf[USHRT_MAX];
> struct iphdr *ip4h = tap_push_l2h(c, buf, src_mac, ETH_P_IP);
> @@ -291,7 +296,7 @@ void tap_icmp4_send(const struct ctx *c, struct in_addr src, struct in_addr dst,
> memcpy(icmp4h, in, l4len);
> csum_icmp4(icmp4h, icmp4h + 1, l4len - sizeof(*icmp4h));
>
> - tap_send_single(c, buf, l4len + ((char *)icmp4h - buf));
> + tap_send_single(c, queue, buf, l4len + ((char *)icmp4h - buf));
> }
>
> /**
> @@ -355,6 +360,7 @@ void *tap_push_uh6(struct udphdr *uh,
> /**
> * tap_udp6_send() - Send UDP over IPv6 packet
> * @c: Execution context
> + * @queue: Queue to use to send packet
> * @src: IPv6 source address
> * @sport: UDP source port
> * @dst: IPv6 destination address
> @@ -363,7 +369,7 @@ void *tap_push_uh6(struct udphdr *uh,
> * @in: UDP payload contents (not including UDP header)
> * @dlen: UDP payload length (not including UDP header)
> */
> -void tap_udp6_send(const struct ctx *c,
> +void tap_udp6_send(const struct ctx *c, int queue,
> const struct in6_addr *src, in_port_t sport,
> const struct in6_addr *dst, in_port_t dport,
> uint32_t flow, void *in, size_t dlen)
> @@ -376,19 +382,20 @@ void tap_udp6_send(const struct ctx *c,
> char *data = tap_push_uh6(uh, src, sport, dst, dport, in, dlen);
>
> memcpy(data, in, dlen);
> - tap_send_single(c, buf, dlen + (data - buf));
> + tap_send_single(c, queue, buf, dlen + (data - buf));
> }
>
> /**
> * tap_icmp6_send() - Send ICMPv6 packet
> * @c: Execution context
> + * @queue: Queue to use to send packet
> * @src: IPv6 source address
> * @dst: IPv6 destination address
> * @in: ICMP packet, including ICMP header
> * @src_mac: MAC address to be used as source for message
> * @l4len: ICMP packet length, including ICMP header
> */
> -void tap_icmp6_send(const struct ctx *c,
> +void tap_icmp6_send(const struct ctx *c, int queue,
> const struct in6_addr *src, const struct in6_addr *dst,
> const void *in, const void *src_mac, size_t l4len)
> {
> @@ -400,7 +407,7 @@ void tap_icmp6_send(const struct ctx *c,
> memcpy(icmp6h, in, l4len);
> csum_icmp6(icmp6h, src, dst, icmp6h + 1, l4len - sizeof(*icmp6h));
>
> - tap_send_single(c, buf, l4len + ((char *)icmp6h - buf));
> + tap_send_single(c, queue, buf, l4len + ((char *)icmp6h - buf));
> }
>
> /**
> @@ -729,7 +736,7 @@ resume:
> if (!eh)
> continue;
> if (ntohs(eh->h_proto) == ETH_P_ARP) {
> - arp(c, &data);
> + arp(c, VHOST_USER_RX_QUEUE, &data);
> continue;
> }
>
> @@ -790,7 +797,7 @@ resume:
> struct iov_tail eh_data;
>
> packet_get(in, i, &eh_data);
> - if (dhcp(c, &eh_data))
> + if (dhcp(c, VHOST_USER_RX_QUEUE, &eh_data))
> continue;
> }
>
> @@ -958,7 +965,7 @@ resume:
> continue;
>
> ndp_data = data;
> - if (ndp(c, saddr, &ndp_data))
> + if (ndp(c, VHOST_USER_RX_QUEUE, saddr, &ndp_data))
> continue;
>
> tap_packet_debug(NULL, ip6h, NULL, proto, NULL, 1);
> @@ -977,7 +984,8 @@ resume:
> if (proto == IPPROTO_UDP) {
> struct iov_tail uh_data = data;
>
> - if (dhcpv6(c, &uh_data, saddr, daddr))
> + if (dhcpv6(c, VHOST_USER_RX_QUEUE, &uh_data, saddr,
> + daddr))
> continue;
> }
>
> diff --git a/tap.h b/tap.h
> index 1864173cc9b0..76403a43edbc 100644
> --- a/tap.h
> +++ b/tap.h
> @@ -87,24 +87,25 @@ void *tap_push_ip6h(struct ipv6hdr *ip6h,
> const struct in6_addr *src,
> const struct in6_addr *dst,
> size_t l4len, uint8_t proto, uint32_t flow);
> -void tap_udp4_send(const struct ctx *c, struct in_addr src, in_port_t sport,
> +void tap_udp4_send(const struct ctx *c, int queue, struct in_addr src, in_port_t sport,
> struct in_addr dst, in_port_t dport,
> const void *in, size_t dlen);
> -void tap_icmp4_send(const struct ctx *c, struct in_addr src, struct in_addr dst,
> - const void *in, const void *src_mac, size_t l4len);
> +void tap_icmp4_send(const struct ctx *c, int queue, struct in_addr src,
> + struct in_addr dst, const void *in, const void *src_mac,
> + size_t l4len);
> const struct in6_addr *tap_ip6_daddr(const struct ctx *c,
> const struct in6_addr *src);
> void *tap_push_ip6h(struct ipv6hdr *ip6h,
> const struct in6_addr *src, const struct in6_addr *dst,
> size_t l4len, uint8_t proto, uint32_t flow);
> -void tap_udp6_send(const struct ctx *c,
> +void tap_udp6_send(const struct ctx *c, int queue,
> const struct in6_addr *src, in_port_t sport,
> const struct in6_addr *dst, in_port_t dport,
> uint32_t flow, void *in, size_t dlen);
> -void tap_icmp6_send(const struct ctx *c,
> +void tap_icmp6_send(const struct ctx *c, int queue,
> const struct in6_addr *src, const struct in6_addr *dst,
> const void *in, const void *src_mac, size_t l4len);
> -void tap_send_single(const struct ctx *c, const void *data, size_t l2len);
> +void tap_send_single(const struct ctx *c, int queue, const void *data, size_t l2len);
> size_t tap_send_frames(const struct ctx *c, const struct iovec *iov,
> size_t bufs_per_frame, size_t nframes);
> void eth_update_mac(struct ethhdr *eh,
> diff --git a/tcp.c b/tcp.c
> index e91c0cf5a441..5ce34baa8a5a 100644
> --- a/tcp.c
> +++ b/tcp.c
> @@ -1984,6 +1984,7 @@ static void tcp_conn_from_sock_finish(const struct ctx *c,
> /**
> * tcp_rst_no_conn() - Send RST in response to a packet with no connection
> * @c: Execution context
> + * @queue: Queue to use to send the reply
> * @af: Address family, AF_INET or AF_INET6
> * @saddr: Source address of the packet we're responding to
> * @daddr: Destination address of the packet we're responding to
> @@ -1991,7 +1992,7 @@ static void tcp_conn_from_sock_finish(const struct ctx *c,
> * @th: TCP header of the packet we're responding to
> * @l4len: Packet length, including TCP header
> */
> -static void tcp_rst_no_conn(const struct ctx *c, int af,
> +static void tcp_rst_no_conn(const struct ctx *c, int queue, int af,
> const void *saddr, const void *daddr,
> uint32_t flow_lbl,
> const struct tcphdr *th, size_t l4len)
> @@ -2049,7 +2050,7 @@ static void tcp_rst_no_conn(const struct ctx *c, int af,
>
> tcp_update_csum(psum, rsth, &payload);
> rst_l2len = ((char *)rsth - buf) + sizeof(*rsth);
> - tap_send_single(c, buf, rst_l2len);
> + tap_send_single(c, queue, buf, rst_l2len);
> }
>
> /**
> @@ -2066,9 +2067,10 @@ static void tcp_rst_no_conn(const struct ctx *c, int af,
> *
> * Return: count of consumed packets
> */
> -int tcp_tap_handler(const struct ctx *c, uint8_t pif, sa_family_t af,
> - const void *saddr, const void *daddr, uint32_t flow_lbl,
> - const struct pool *p, int idx, const struct timespec *now)
> +int tcp_tap_handler(const struct ctx *c, uint8_t pif,
> + sa_family_t af, const void *saddr, const void *daddr,
> + uint32_t flow_lbl, const struct pool *p, int idx,
> + const struct timespec *now)
This seems like an unrelated change.
> {
> struct tcp_tap_conn *conn;
> struct tcphdr th_storage;
> @@ -2108,7 +2110,8 @@ int tcp_tap_handler(const struct ctx *c, uint8_t pif, sa_family_t af,
> tcp_conn_from_tap(c, af, saddr, daddr, th,
> opts, optlen, now);
> else
> - tcp_rst_no_conn(c, af, saddr, daddr, flow_lbl, th, l4len);
> + tcp_rst_no_conn(c, VHOST_USER_RX_QUEUE, af, saddr,
> + daddr, flow_lbl, th, l4len);
> return 1;
> }
>
> diff --git a/tcp.h b/tcp.h
> index 0082386725c2..320683ce5679 100644
> --- a/tcp.h
> +++ b/tcp.h
> @@ -13,11 +13,12 @@ struct ctx;
> void tcp_timer_handler(const struct ctx *c, union epoll_ref ref);
> void tcp_listen_handler(const struct ctx *c, union epoll_ref ref,
> const struct timespec *now);
> -void tcp_sock_handler(const struct ctx *c, union epoll_ref ref,
> - uint32_t events);
So does this.
> -int tcp_tap_handler(const struct ctx *c, uint8_t pif, sa_family_t af,
> - const void *saddr, const void *daddr, uint32_t flow_lbl,
> - const struct pool *p, int idx, const struct timespec *now);
> +void tcp_sock_handler(const struct ctx *c,
> + union epoll_ref ref, uint32_t events);
> +int tcp_tap_handler(const struct ctx *c, uint8_t pif,
> + sa_family_t af, const void *saddr, const void *daddr,
> + uint32_t flow_lbl, const struct pool *p, int idx,
> + const struct timespec *now);
> int tcp_sock_init(const struct ctx *c, const union inany_addr *addr,
> const char *ifname, in_port_t port);
> int tcp_init(struct ctx *c);
> diff --git a/udp.c b/udp.c
> index 9c00950250a0..868ffebb5802 100644
> --- a/udp.c
> +++ b/udp.c
> @@ -384,13 +384,14 @@ static void udp_tap_prepare(const struct mmsghdr *mmh,
> /**
> * udp_send_tap_icmp4() - Construct and send ICMPv4 to local peer
> * @c: Execution context
> + * @queue: Queue to send the ICMPv4 packet to
> * @ee: Extended error descriptor
> * @toside: Destination side of flow
> * @saddr: Address of ICMP generating node
> * @in: First bytes (max 8) of original UDP message body
> * @dlen: Length of the read part of original UDP message body
> */
> -static void udp_send_tap_icmp4(const struct ctx *c,
> +static void udp_send_tap_icmp4(const struct ctx *c, int queue,
> const struct sock_extended_err *ee,
> const struct flowside *toside,
> struct in_addr saddr,
> @@ -426,13 +427,14 @@ static void udp_send_tap_icmp4(const struct ctx *c,
> /* Try to obtain the MAC address of the generating node */
> saddr_any = inany_from_v4(saddr);
> fwd_neigh_mac_get(c, &saddr_any, tap_omac);
> - tap_icmp4_send(c, saddr, eaddr, &msg, tap_omac, msglen);
> + tap_icmp4_send(c, queue, saddr, eaddr, &msg, tap_omac, msglen);
> }
>
>
> /**
> * udp_send_tap_icmp6() - Construct and send ICMPv6 to local peer
> * @c: Execution context
> + * @queue: Queue to send the ICMPv6 packet to
> * @ee: Extended error descriptor
> * @toside: Destination side of flow
> * @saddr: Address of ICMP generating node
> @@ -440,7 +442,7 @@ static void udp_send_tap_icmp4(const struct ctx *c,
> * @dlen: Length of the read part of original UDP message body
> * @flow: IPv6 flow identifier
> */
> -static void udp_send_tap_icmp6(const struct ctx *c,
> +static void udp_send_tap_icmp6(const struct ctx *c, int queue,
> const struct sock_extended_err *ee,
> const struct flowside *toside,
> const struct in6_addr *saddr,
> @@ -474,7 +476,7 @@ static void udp_send_tap_icmp6(const struct ctx *c,
>
> /* Try to obtain the MAC address of the generating node */
> fwd_neigh_mac_get(c, (union inany_addr *) saddr, tap_omac);
> - tap_icmp6_send(c, saddr, eaddr, &msg, tap_omac, msglen);
> + tap_icmp6_send(c, queue, saddr, eaddr, &msg, tap_omac, msglen);
> }
>
> /**
> @@ -634,13 +636,14 @@ static int udp_sock_recverr(const struct ctx *c, int s, flow_sidx_t sidx,
> if (hdr->cmsg_level == IPPROTO_IP &&
> (o4 = inany_v4(&otap)) && inany_v4(&toside->eaddr)) {
> dlen = MIN(dlen, ICMP4_MAX_DLEN);
> - udp_send_tap_icmp4(c, ee, toside, *o4, data, dlen);
> + udp_send_tap_icmp4(c, VHOST_USER_RX_QUEUE, ee, toside, *o4,
> + data, dlen);
> return 1;
> }
>
> if (hdr->cmsg_level == IPPROTO_IPV6 && !inany_v4(&toside->eaddr)) {
> - udp_send_tap_icmp6(c, ee, toside, &otap.a6, data, dlen,
> - FLOW_IDX(uflow));
> + udp_send_tap_icmp6(c, VHOST_USER_RX_QUEUE, ee, toside,
> + &otap.a6, data, dlen, FLOW_IDX(uflow));
> return 1;
> }
>
> diff --git a/vu_common.c b/vu_common.c
> index b13b7c308fd8..8904403e66af 100644
> --- a/vu_common.c
> +++ b/vu_common.c
> @@ -235,23 +235,26 @@ void vu_kick_cb(struct vu_dev *vdev, union epoll_ref ref,
> }
>
> /**
> - * vu_send_single() - Send a buffer to the front-end using the RX virtqueue
> + * vu_send_single() - Send a buffer to the front-end using a specified virtqueue
> * @c: execution context
> + * @queue: queue to use to send the buffer
> * @buf: address of the buffer
> * @size: size of the buffer
> *
> * Return: number of bytes sent, -1 if there is an error
> */
> -int vu_send_single(const struct ctx *c, const void *buf, size_t size)
> +int vu_send_single(const struct ctx *c, int queue, const void *buf, size_t size)
> {
> struct vu_dev *vdev = c->vdev;
> - struct vu_virtq *vq = &vdev->vq[VHOST_USER_RX_QUEUE];
> struct vu_virtq_element elem[VIRTQUEUE_MAX_SIZE];
> struct iovec in_sg[VIRTQUEUE_MAX_SIZE];
> + struct vu_virtq *vq;
> size_t total;
> int elem_cnt;
> int i;
>
> + vq = &vdev->vq[queue];
> +
> trace("vu_send_single size %zu", size);
>
> if (!vu_queue_enabled(vq) || !vu_queue_started(vq)) {
> diff --git a/vu_common.h b/vu_common.h
> index f538f237790b..2637c67f913c 100644
> --- a/vu_common.h
> +++ b/vu_common.h
> @@ -56,6 +56,7 @@ void vu_flush(const struct vu_dev *vdev, struct vu_virtq *vq,
> struct vu_virtq_element *elem, int elem_cnt);
> void vu_kick_cb(struct vu_dev *vdev, union epoll_ref ref,
> const struct timespec *now);
> -int vu_send_single(const struct ctx *c, const void *buf, size_t size);
> +int vu_send_single(const struct ctx *c, int queue, const void *buf,
> + size_t size);
>
> #endif /* VU_COMMON_H */
> --
> 2.51.0
>
--
David Gibson (he or they) | I'll have my music baroque, and my code
david AT gibson.dropbear.id.au | minimalist, thank you, not the other way
| around.
http://www.ozlabs.org/~dgibson
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]
^ permalink raw reply [flat|nested] 13+ messages in thread
* [PATCH 3/4] multiqueue: Add queue-aware flow management for multiqueue support
2025-11-07 14:38 [PATCH 0/4] vhost-user: Add multiqueue support Laurent Vivier
2025-11-07 14:38 ` [PATCH 1/4] vhost-user: Enable multiqueue Laurent Vivier
2025-11-07 14:38 ` [PATCH 2/4] vhost-user: Add queue parameter throughout the network stack Laurent Vivier
@ 2025-11-07 14:39 ` Laurent Vivier
2025-11-10 5:54 ` David Gibson
2025-11-07 14:39 ` [PATCH 4/4] test: Add multiqueue support to vhost-user test infrastructure Laurent Vivier
2025-11-10 4:40 ` [PATCH 0/4] vhost-user: Add multiqueue support David Gibson
4 siblings, 1 reply; 13+ messages in thread
From: Laurent Vivier @ 2025-11-07 14:39 UTC (permalink / raw)
To: passt-dev; +Cc: Laurent Vivier
Packets are now routed to the correct RX queue based on which TX queue
they arrived on, rather than always using queue 0.
Note: Flows initiated from the host (via sockets, udp_flow_from_sock())
currently default to queue 0, as they don't have an associated incoming
queue.
Signed-off-by: Laurent Vivier <lvivier@redhat.com>
---
flow.c | 32 ++++++++++++++++++++++
flow.h | 10 +++++++
icmp.c | 23 +++++++++-------
icmp.h | 2 +-
tap.c | 77 +++++++++++++++++++++++++++------------------------
tap.h | 5 ++--
tcp.c | 79 +++++++++++++++++++++++++++++------------------------
tcp.h | 2 +-
tcp_vu.c | 8 ++++--
udp.c | 33 ++++++++++++----------
udp.h | 12 ++++----
udp_flow.c | 8 +++++-
udp_flow.h | 2 +-
udp_vu.c | 4 ++-
vu_common.c | 4 +--
15 files changed, 187 insertions(+), 114 deletions(-)
diff --git a/flow.c b/flow.c
index 278a9cf0ac6d..8e9d7e5e1847 100644
--- a/flow.c
+++ b/flow.c
@@ -405,6 +405,37 @@ void flow_epollid_register(int epollid, int epollfd)
epoll_id_to_fd[epollid] = epollfd;
}
+/**
+ * flow_rx_virtqueue() - Get RX (receive) queue number for a flow
+ * @f: Flow to query (may be NULL)
+ *
+ * Return: RX queue number for the flow, or 0 if flow is NULL or has no
+ * valid queue assignment
+ */
+int flow_rx_virtqueue(const struct flow_common *f)
+{
+ if (f == NULL || f->queueid == FLOW_QUEUEID_INVALID)
+ return 0;
+ return f->queueid << 1;
+}
+
+/**
+ * flow_queue_set() - Set queue pair assignment for a flow
+ * @f: Flow to update
+ * @queueid: Queue pair number to assign (even number, RX queue; TX is RX+1)
+ */
+void flow_queue_set(struct flow_common *f, int queueid)
+{
+ queueid >>= 1;
+
+ ASSERT(queueid < FLOW_QUEUEID_MAX);
+
+ flow_trace((union flow *)f, "updating queue from %d to %d",
+ f->queueid, queueid);
+
+ f->queueid = queueid;
+}
+
/**
* flow_initiate_() - Move flow to INI, setting pif[INISIDE]
* @flow: Flow to change state
@@ -609,6 +640,7 @@ union flow *flow_alloc(void)
flow_new_entry = flow;
memset(flow, 0, sizeof(*flow));
flow_epollid_clear(&flow->f);
+ flow->f.queueid = FLOW_QUEUEID_INVALID;
flow_set_state(&flow->f, FLOW_STATE_NEW);
return flow;
diff --git a/flow.h b/flow.h
index b43b0b1dd7f2..44ab4ae8fd6a 100644
--- a/flow.h
+++ b/flow.h
@@ -179,6 +179,8 @@ int flowside_connect(const struct ctx *c, int s,
* @side[]: Information for each side of the flow
* @tap_omac: MAC address of remote endpoint as seen from the guest
* @epollid: epollfd identifier, or EPOLLFD_ID_INVALID
+ * @queueid: Queue pair number assigned to this flow
+ * (FLOW_QUEUEID_INVALID if not assigned)
*/
struct flow_common {
#ifdef __GNUC__
@@ -199,6 +201,8 @@ struct flow_common {
#define EPOLLFD_ID_BITS 8
unsigned int epollid:EPOLLFD_ID_BITS;
+#define FLOW_QUEUEID_BITS 5
+ unsigned int queueid:FLOW_QUEUEID_BITS;
};
#define EPOLLFD_ID_DEFAULT 0
@@ -206,6 +210,10 @@ struct flow_common {
#define EPOLLFD_ID_MAX (EPOLLFD_ID_SIZE - 1)
#define EPOLLFD_ID_INVALID EPOLLFD_ID_MAX
+#define FLOW_QUEUEID_SIZE (1 << FLOW_QUEUEID_BITS)
+#define FLOW_QUEUEID_MAX (FLOW_QUEUEID_SIZE - 1)
+#define FLOW_QUEUEID_INVALID FLOW_QUEUEID_MAX
+
#define FLOW_INDEX_BITS 17 /* 128k - 1 */
#define FLOW_MAX MAX_FROM_BITS(FLOW_INDEX_BITS)
@@ -266,6 +274,8 @@ int flow_epollfd(const struct flow_common *f);
void flow_epollid_set(struct flow_common *f, int epollid);
void flow_epollid_clear(struct flow_common *f);
void flow_epollid_register(int epollid, int epollfd);
+int flow_rx_virtqueue(const struct flow_common *f);
+void flow_queue_set(struct flow_common *f, int queueid);
void flow_defer_handler(const struct ctx *c, const struct timespec *now);
int flow_migrate_source_early(struct ctx *c, const struct migrate_stage *stage,
int fd);
diff --git a/icmp.c b/icmp.c
index d58499c3bf5c..80e8753072fa 100644
--- a/icmp.c
+++ b/icmp.c
@@ -132,13 +132,13 @@ void icmp_sock_handler(const struct ctx *c, union epoll_ref ref)
const struct in_addr *daddr = inany_v4(&ini->eaddr);
ASSERT(saddr && daddr); /* Must have IPv4 addresses */
- tap_icmp4_send(c, VHOST_USER_RX_QUEUE, *saddr, *daddr, buf,
+ tap_icmp4_send(c, flow_rx_virtqueue(&pingf->f), *saddr, *daddr, buf,
pingf->f.tap_omac, n);
} else if (pingf->f.type == FLOW_PING6) {
const struct in6_addr *saddr = &ini->oaddr.a6;
const struct in6_addr *daddr = &ini->eaddr.a6;
- tap_icmp6_send(c, VHOST_USER_RX_QUEUE, saddr, daddr, buf,
+ tap_icmp6_send(c, flow_rx_virtqueue(&pingf->f), saddr, daddr, buf,
pingf->f.tap_omac, n);
}
return;
@@ -238,17 +238,18 @@ cancel:
/**
* icmp_tap_handler() - Handle packets from tap
- * @c: Execution context
- * @pif: pif on which the packet is arriving
- * @af: Address family, AF_INET or AF_INET6
- * @saddr: Source address
- * @daddr: Destination address
- * @data: Single packet with ICMP/ICMPv6 header
- * @now: Current timestamp
+ * @c: Execution context
+ * @incoming_queue: Incoming queue number
+ * @pif: pif on which the packet is arriving
+ * @af: Address family, AF_INET or AF_INET6
+ * @saddr: Source address
+ * @daddr: Destination address
+ * @data: Single packet with ICMP/ICMPv6 header
+ * @now: Current timestamp
*
* Return: count of consumed packets (always 1, even if malformed)
*/
-int icmp_tap_handler(const struct ctx *c, uint8_t pif, sa_family_t af,
+int icmp_tap_handler(const struct ctx *c, int incoming_queue, uint8_t pif, sa_family_t af,
const void *saddr, const void *daddr,
struct iov_tail *data, const struct timespec *now)
{
@@ -309,6 +310,8 @@ int icmp_tap_handler(const struct ctx *c, uint8_t pif, sa_family_t af,
else if (!(pingf = icmp_ping_new(c, af, id, saddr, daddr)))
return 1;
+ flow_queue_set(&pingf->f, incoming_queue);
+
tgt = &pingf->f.side[TGTSIDE];
ASSERT(flow_proto[pingf->f.type] == proto);
diff --git a/icmp.h b/icmp.h
index 1a0e6205f087..6d6d6358bb33 100644
--- a/icmp.h
+++ b/icmp.h
@@ -10,7 +10,7 @@ struct ctx;
struct icmp_ping_flow;
void icmp_sock_handler(const struct ctx *c, union epoll_ref ref);
-int icmp_tap_handler(const struct ctx *c, uint8_t pif, sa_family_t af,
+int icmp_tap_handler(const struct ctx *c, int incoming_queue, uint8_t pif, sa_family_t af,
const void *saddr, const void *daddr,
struct iov_tail *data, const struct timespec *now);
void icmp_init(void);
diff --git a/tap.c b/tap.c
index 1308d49242e8..9a4399d947a3 100644
--- a/tap.c
+++ b/tap.c
@@ -702,15 +702,17 @@ static bool tap4_is_fragment(const struct iphdr *iph,
/**
* tap4_handler() - IPv4 and ARP packet handler for tap file descriptor
- * @c: Execution context
- * @in: Ingress packet pool, packets with Ethernet headers
- * @now: Current timestamp
+ * @c: Execution context
+ * @incoming_queue: Incoming queue number
+ * @in: Ingress packet pool, packets with Ethernet headers
+ * @now: Current timestamp
*
* Return: count of packets consumed by handlers
*/
-static int tap4_handler(struct ctx *c, const struct pool *in,
- const struct timespec *now)
+static int tap4_handler(struct ctx *c, int incoming_queue,
+ const struct pool *in, const struct timespec *now)
{
+ int outgoing_queue = incoming_queue & ~1;
unsigned int i, j, seq_count;
struct tap4_l4_t *seq;
@@ -736,7 +738,7 @@ resume:
if (!eh)
continue;
if (ntohs(eh->h_proto) == ETH_P_ARP) {
- arp(c, VHOST_USER_RX_QUEUE, &data);
+ arp(c, outgoing_queue, &data);
continue;
}
@@ -783,7 +785,7 @@ resume:
tap_packet_debug(iph, NULL, NULL, 0, NULL, 1);
- icmp_tap_handler(c, PIF_TAP, AF_INET,
+ icmp_tap_handler(c, incoming_queue, PIF_TAP, AF_INET,
&iph->saddr, &iph->daddr,
&data, now);
continue;
@@ -797,7 +799,7 @@ resume:
struct iov_tail eh_data;
packet_get(in, i, &eh_data);
- if (dhcp(c, VHOST_USER_RX_QUEUE, &eh_data))
+ if (dhcp(c, outgoing_queue, &eh_data))
continue;
}
@@ -860,14 +862,14 @@ append:
if (c->no_tcp)
continue;
for (k = 0; k < p->count; )
- k += tcp_tap_handler(c, PIF_TAP, AF_INET,
+ k += tcp_tap_handler(c, incoming_queue, PIF_TAP, AF_INET,
&seq->saddr, &seq->daddr,
0, p, k, now);
} else if (seq->protocol == IPPROTO_UDP) {
if (c->no_udp)
continue;
for (k = 0; k < p->count; )
- k += udp_tap_handler(c, PIF_TAP, AF_INET,
+ k += udp_tap_handler(c, incoming_queue, PIF_TAP, AF_INET,
&seq->saddr, &seq->daddr,
seq->ttl, p, k, now);
}
@@ -881,15 +883,17 @@ append:
/**
* tap6_handler() - IPv6 packet handler for tap file descriptor
- * @c: Execution context
- * @in: Ingress packet pool, packets with Ethernet headers
- * @now: Current timestamp
+ * @c: Execution context
+ * @incoming_queue: Incoming queue number
+ * @in: Ingress packet pool, packets with Ethernet headers
+ * @now: Current timestamp
*
* Return: count of packets consumed by handlers
*/
-static int tap6_handler(struct ctx *c, const struct pool *in,
- const struct timespec *now)
+static int tap6_handler(struct ctx *c, int incoming_queue,
+ const struct pool *in, const struct timespec *now)
{
+ int outgoing_queue = incoming_queue & ~1;
unsigned int i, j, seq_count = 0;
struct tap6_l4_t *seq;
@@ -965,12 +969,12 @@ resume:
continue;
ndp_data = data;
- if (ndp(c, VHOST_USER_RX_QUEUE, saddr, &ndp_data))
+ if (ndp(c, outgoing_queue, saddr, &ndp_data))
continue;
tap_packet_debug(NULL, ip6h, NULL, proto, NULL, 1);
- icmp_tap_handler(c, PIF_TAP, AF_INET6,
+ icmp_tap_handler(c, incoming_queue, PIF_TAP, AF_INET6,
saddr, daddr, &data, now);
continue;
}
@@ -984,8 +988,7 @@ resume:
if (proto == IPPROTO_UDP) {
struct iov_tail uh_data = data;
- if (dhcpv6(c, VHOST_USER_RX_QUEUE, &uh_data, saddr,
- daddr))
+ if (dhcpv6(c, outgoing_queue, &uh_data, saddr, daddr))
continue;
}
@@ -1053,14 +1056,14 @@ append:
if (c->no_tcp)
continue;
for (k = 0; k < p->count; )
- k += tcp_tap_handler(c, PIF_TAP, AF_INET6,
+ k += tcp_tap_handler(c, incoming_queue, PIF_TAP, AF_INET6,
&seq->saddr, &seq->daddr,
seq->flow_lbl, p, k, now);
} else if (seq->protocol == IPPROTO_UDP) {
if (c->no_udp)
continue;
for (k = 0; k < p->count; )
- k += udp_tap_handler(c, PIF_TAP, AF_INET6,
+ k += udp_tap_handler(c, incoming_queue, PIF_TAP, AF_INET6,
&seq->saddr, &seq->daddr,
seq->hop_limit, p, k, now);
}
@@ -1083,22 +1086,24 @@ void tap_flush_pools(void)
/**
* tap_handler() - IPv4/IPv6 and ARP packet handler for tap file descriptor
- * @c: Execution context
- * @now: Current timestamp
+ * @c: Execution context
+ * @incoming_queue: Incoming queue number
+ * @now: Current timestamp
*/
-void tap_handler(struct ctx *c, const struct timespec *now)
+void tap_handler(struct ctx *c, int incoming_queue, const struct timespec *now)
{
- tap4_handler(c, pool_tap4, now);
- tap6_handler(c, pool_tap6, now);
+ tap4_handler(c, incoming_queue, pool_tap4, now);
+ tap6_handler(c, incoming_queue, pool_tap6, now);
}
/**
* tap_add_packet() - Queue/capture packet, update notion of guest MAC address
- * @c: Execution context
- * @data: Packet to add to the pool
- * @now: Current timestamp
+ * @c: Execution context
+ * @incoming_queue: Incoming queue number
+ * @data: Packet to add to the pool
+ * @now: Current timestamp
*/
-void tap_add_packet(struct ctx *c, struct iov_tail *data,
+void tap_add_packet(struct ctx *c, int incoming_queue, struct iov_tail *data,
const struct timespec *now)
{
struct ethhdr eh_storage;
@@ -1123,14 +1128,14 @@ void tap_add_packet(struct ctx *c, struct iov_tail *data,
case ETH_P_ARP:
case ETH_P_IP:
if (!pool_can_fit(pool_tap4, data)) {
- tap4_handler(c, pool_tap4, now);
+ tap4_handler(c, incoming_queue, pool_tap4, now);
pool_flush(pool_tap4);
}
packet_add(pool_tap4, data);
break;
case ETH_P_IPV6:
if (!pool_can_fit(pool_tap6, data)) {
- tap6_handler(c, pool_tap6, now);
+ tap6_handler(c, incoming_queue, pool_tap6, now);
pool_flush(pool_tap6);
}
packet_add(pool_tap6, data);
@@ -1217,7 +1222,7 @@ static void tap_passt_input(struct ctx *c, const struct timespec *now)
n -= sizeof(uint32_t);
data = IOV_TAIL_FROM_BUF(p, l2len, 0);
- tap_add_packet(c, &data, now);
+ tap_add_packet(c, 0, &data, now);
p += l2len;
n -= l2len;
@@ -1226,7 +1231,7 @@ static void tap_passt_input(struct ctx *c, const struct timespec *now)
partial_len = n;
partial_frame = p;
- tap_handler(c, now);
+ tap_handler(c, 0, now);
}
/**
@@ -1285,10 +1290,10 @@ static void tap_pasta_input(struct ctx *c, const struct timespec *now)
continue;
data = IOV_TAIL_FROM_BUF(pkt_buf + n, len, 0);
- tap_add_packet(c, &data, now);
+ tap_add_packet(c, 0, &data, now);
}
- tap_handler(c, now);
+ tap_handler(c, 0, now);
}
/**
diff --git a/tap.h b/tap.h
index 76403a43edbc..fe9455ffcf4b 100644
--- a/tap.h
+++ b/tap.h
@@ -119,7 +119,8 @@ int tap_sock_unix_open(char *sock_path);
void tap_sock_reset(struct ctx *c);
void tap_backend_init(struct ctx *c);
void tap_flush_pools(void);
-void tap_handler(struct ctx *c, const struct timespec *now);
-void tap_add_packet(struct ctx *c, struct iov_tail *data,
+void tap_handler(struct ctx *c, int incoming_queue,
+ const struct timespec *now);
+void tap_add_packet(struct ctx *c, int incoming_queue, struct iov_tail *data,
const struct timespec *now);
#endif /* TAP_H */
diff --git a/tcp.c b/tcp.c
index 5ce34baa8a5a..78494a2dc69f 100644
--- a/tcp.c
+++ b/tcp.c
@@ -1496,21 +1496,23 @@ static void tcp_bind_outbound(const struct ctx *c,
/**
* tcp_conn_from_tap() - Handle connection request (SYN segment) from tap
- * @c: Execution context
- * @af: Address family, AF_INET or AF_INET6
- * @saddr: Source address, pointer to in_addr or in6_addr
- * @daddr: Destination address, pointer to in_addr or in6_addr
- * @th: TCP header from tap: caller MUST ensure it's there
- * @opts: Pointer to start of options
- * @optlen: Bytes in options: caller MUST ensure available length
- * @now: Current timestamp
+ * @c: Execution context
+ * @incoming_queue: TX queue number for the flow
+ * @af: Address family, AF_INET or AF_INET6
+ * @saddr: Source address, pointer to in_addr or in6_addr
+ * @daddr: Destination address, pointer to in_addr or in6_addr
+ * @th: TCP header from tap: caller MUST ensure it's there
+ * @opts: Pointer to start of options
+ * @optlen: Bytes in options: caller MUST ensure available length
+ * @now: Current timestamp
*
* #syscalls:vu getsockname
*/
-static void tcp_conn_from_tap(const struct ctx *c, sa_family_t af,
- const void *saddr, const void *daddr,
- const struct tcphdr *th, const char *opts,
- size_t optlen, const struct timespec *now)
+static void tcp_conn_from_tap(const struct ctx *c, int incoming_queue,
+ sa_family_t af, const void *saddr,
+ const void *daddr, const struct tcphdr *th,
+ const char *opts, size_t optlen,
+ const struct timespec *now)
{
in_port_t srcport = ntohs(th->source);
in_port_t dstport = ntohs(th->dest);
@@ -1622,6 +1624,7 @@ static void tcp_conn_from_tap(const struct ctx *c, sa_family_t af,
conn_event(c, conn, TAP_SYN_ACK_SENT);
}
+ flow_queue_set(&conn->f, incoming_queue);
tcp_epoll_ctl(c, conn);
if (c->mode == MODE_VU) { /* To rebind to same oport after migration */
@@ -1983,16 +1986,16 @@ static void tcp_conn_from_sock_finish(const struct ctx *c,
/**
* tcp_rst_no_conn() - Send RST in response to a packet with no connection
- * @c: Execution context
- * @queue: Queue to use to send the reply
- * @af: Address family, AF_INET or AF_INET6
- * @saddr: Source address of the packet we're responding to
- * @daddr: Destination address of the packet we're responding to
- * @flow_lbl: IPv6 flow label (ignored for IPv4)
- * @th: TCP header of the packet we're responding to
- * @l4len: Packet length, including TCP header
- */
-static void tcp_rst_no_conn(const struct ctx *c, int queue, int af,
+ * @c: Execution context
+ * @outgoing_queue: Queue to use to send the reply
+ * @af: Address family, AF_INET or AF_INET6
+ * @saddr: Source address of the packet we're responding to
+ * @daddr: Destination address of the packet we're responding to
+ * @flow_lbl: IPv6 flow label (ignored for IPv4)
+ * @th: TCP header of the packet we're responding to
+ * @l4len: Packet length, including TCP header
+ */
+static void tcp_rst_no_conn(const struct ctx *c, int outgoing_queue, int af,
const void *saddr, const void *daddr,
uint32_t flow_lbl,
const struct tcphdr *th, size_t l4len)
@@ -2050,24 +2053,25 @@ static void tcp_rst_no_conn(const struct ctx *c, int queue, int af,
tcp_update_csum(psum, rsth, &payload);
rst_l2len = ((char *)rsth - buf) + sizeof(*rsth);
- tap_send_single(c, queue, buf, rst_l2len);
+ tap_send_single(c, outgoing_queue, buf, rst_l2len);
}
/**
* tcp_tap_handler() - Handle packets from tap and state transitions
- * @c: Execution context
- * @pif: pif on which the packet is arriving
- * @af: Address family, AF_INET or AF_INET6
- * @saddr: Source address
- * @daddr: Destination address
- * @flow_lbl: IPv6 flow label (ignored for IPv4)
- * @p: Pool of TCP packets, with TCP headers
- * @idx: Index of first packet in pool to process
- * @now: Current timestamp
+ * @c: Execution context
+ * @incoming_queue: Incoming queue number
+ * @pif: pif on which the packet is arriving
+ * @af: Address family, AF_INET or AF_INET6
+ * @saddr: Source address
+ * @daddr: Destination address
+ * @flow_lbl: IPv6 flow label (ignored for IPv4)
+ * @p: Pool of TCP packets, with TCP headers
+ * @idx: Index of first packet in pool to process
+ * @now: Current timestamp
*
* Return: count of consumed packets
*/
-int tcp_tap_handler(const struct ctx *c, uint8_t pif,
+int tcp_tap_handler(const struct ctx *c, int incoming_queue, uint8_t pif,
sa_family_t af, const void *saddr, const void *daddr,
uint32_t flow_lbl, const struct pool *p, int idx,
const struct timespec *now)
@@ -2107,11 +2111,11 @@ int tcp_tap_handler(const struct ctx *c, uint8_t pif,
/* New connection from tap */
if (!flow) {
if (opts && th->syn && !th->ack)
- tcp_conn_from_tap(c, af, saddr, daddr, th,
+ tcp_conn_from_tap(c, incoming_queue, af, saddr, daddr, th,
opts, optlen, now);
else
- tcp_rst_no_conn(c, VHOST_USER_RX_QUEUE, af, saddr,
- daddr, flow_lbl, th, l4len);
+ tcp_rst_no_conn(c, incoming_queue & ~1, af, saddr, daddr,
+ flow_lbl, th, l4len);
return 1;
}
@@ -2119,6 +2123,9 @@ int tcp_tap_handler(const struct ctx *c, uint8_t pif,
ASSERT(pif_at_sidx(sidx) == PIF_TAP);
conn = &flow->tcp;
+ /* update queue */
+ flow_queue_set(&flow->f, incoming_queue);
+
flow_trace(conn, "packet length %zu from tap", l4len);
if (th->rst) {
diff --git a/tcp.h b/tcp.h
index 320683ce5679..cddd36cadc97 100644
--- a/tcp.h
+++ b/tcp.h
@@ -15,7 +15,7 @@ void tcp_listen_handler(const struct ctx *c, union epoll_ref ref,
const struct timespec *now);
void tcp_sock_handler(const struct ctx *c,
union epoll_ref ref, uint32_t events);
-int tcp_tap_handler(const struct ctx *c, uint8_t pif,
+int tcp_tap_handler(const struct ctx *c, int incoming_queue, uint8_t pif,
sa_family_t af, const void *saddr, const void *daddr,
uint32_t flow_lbl, const struct pool *p, int idx,
const struct timespec *now);
diff --git a/tcp_vu.c b/tcp_vu.c
index 1c81ce376dad..40f552087bc5 100644
--- a/tcp_vu.c
+++ b/tcp_vu.c
@@ -70,15 +70,16 @@ static size_t tcp_vu_hdrlen(bool v6)
*/
int tcp_vu_send_flag(const struct ctx *c, struct tcp_tap_conn *conn, int flags)
{
+ int rx_queue = flow_rx_virtqueue(&conn->f);
struct vu_dev *vdev = c->vdev;
- struct vu_virtq *vq = &vdev->vq[VHOST_USER_RX_QUEUE];
- size_t optlen, hdrlen;
+ struct vu_virtq *vq = &vdev->vq[rx_queue];
struct vu_virtq_element flags_elem[2];
struct ipv6hdr *ip6h = NULL;
struct iphdr *ip4h = NULL;
struct iovec flags_iov[2];
struct tcp_syn_opts *opts;
struct iov_tail payload;
+ size_t optlen, hdrlen;
struct tcphdr *th;
struct ethhdr *eh;
uint32_t seq;
@@ -348,8 +349,9 @@ static void tcp_vu_prepare(const struct ctx *c, struct tcp_tap_conn *conn,
int tcp_vu_data_from_sock(const struct ctx *c, struct tcp_tap_conn *conn)
{
uint32_t wnd_scaled = conn->wnd_from_tap << conn->ws_from_tap;
+ int rx_queue = flow_rx_virtqueue(&conn->f);
struct vu_dev *vdev = c->vdev;
- struct vu_virtq *vq = &vdev->vq[VHOST_USER_RX_QUEUE];
+ struct vu_virtq *vq = &vdev->vq[rx_queue];
ssize_t len, previous_dlen;
int i, iov_cnt, head_cnt;
size_t hdrlen, fillsize;
diff --git a/udp.c b/udp.c
index 868ffebb5802..a0eb719888cd 100644
--- a/udp.c
+++ b/udp.c
@@ -636,14 +636,15 @@ static int udp_sock_recverr(const struct ctx *c, int s, flow_sidx_t sidx,
if (hdr->cmsg_level == IPPROTO_IP &&
(o4 = inany_v4(&otap)) && inany_v4(&toside->eaddr)) {
dlen = MIN(dlen, ICMP4_MAX_DLEN);
- udp_send_tap_icmp4(c, VHOST_USER_RX_QUEUE, ee, toside, *o4,
- data, dlen);
+ udp_send_tap_icmp4(c, flow_rx_virtqueue(&uflow->f), ee, toside,
+ *o4, data, dlen);
return 1;
}
if (hdr->cmsg_level == IPPROTO_IPV6 && !inany_v4(&toside->eaddr)) {
- udp_send_tap_icmp6(c, VHOST_USER_RX_QUEUE, ee, toside,
- &otap.a6, data, dlen, FLOW_IDX(uflow));
+ udp_send_tap_icmp6(c, flow_rx_virtqueue(&uflow->f), ee,
+ toside, &otap.a6, data, dlen,
+ FLOW_IDX(uflow));
return 1;
}
@@ -971,25 +972,27 @@ fail:
/**
* udp_tap_handler() - Handle packets from tap
- * @c: Execution context
- * @pif: pif on which the packet is arriving
- * @af: Address family, AF_INET or AF_INET6
- * @saddr: Source address
- * @daddr: Destination address
- * @ttl: TTL or hop limit for packets to be sent in this call
- * @p: Pool of UDP packets, with UDP headers
- * @idx: Index of first packet to process
- * @now: Current timestamp
+ * @c: Execution context
+ * @incoming_queue: Incoming queue number
+ * @pif: pif on which the packet is arriving
+ * @af: Address family, AF_INET or AF_INET6
+ * @saddr: Source address
+ * @daddr: Destination address
+ * @ttl: TTL or hop limit for packets to be sent in this call
+ * @p: Pool of UDP packets, with UDP headers
+ * @idx: Index of first packet to process
+ * @now: Current timestamp
*
* Return: count of consumed packets
*
* #syscalls sendmmsg
*/
-int udp_tap_handler(const struct ctx *c, uint8_t pif,
+int udp_tap_handler(const struct ctx *c, int incoming_queue, uint8_t pif,
sa_family_t af, const void *saddr, const void *daddr,
uint8_t ttl, const struct pool *p, int idx,
const struct timespec *now)
{
+ int outgoing_queue = incoming_queue & ~1;
const struct flowside *toside;
struct mmsghdr mm[UIO_MAXIOV];
union sockaddr_inany to_sa;
@@ -1019,7 +1022,7 @@ int udp_tap_handler(const struct ctx *c, uint8_t pif,
src = ntohs(uh->source);
dst = ntohs(uh->dest);
- tosidx = udp_flow_from_tap(c, pif, af, saddr, daddr, src, dst, now);
+ tosidx = udp_flow_from_tap(c, outgoing_queue, pif, af, saddr, daddr, src, dst, now);
if (!(uflow = udp_at_sidx(tosidx))) {
char sstr[INET6_ADDRSTRLEN], dstr[INET6_ADDRSTRLEN];
diff --git a/udp.h b/udp.h
index f1d83f380b3f..8ba4ccfe646a 100644
--- a/udp.h
+++ b/udp.h
@@ -7,11 +7,13 @@
#define UDP_H
void udp_portmap_clear(void);
-void udp_listen_sock_handler(const struct ctx *c, union epoll_ref ref,
- uint32_t events, const struct timespec *now);
-void udp_sock_handler(const struct ctx *c, union epoll_ref ref,
- uint32_t events, const struct timespec *now);
-int udp_tap_handler(const struct ctx *c, uint8_t pif,
+void udp_listen_sock_handler(const struct ctx *c,
+ union epoll_ref ref, uint32_t events,
+ const struct timespec *now);
+void udp_sock_handler(const struct ctx *c,
+ union epoll_ref ref, uint32_t events,
+ const struct timespec *now);
+int udp_tap_handler(const struct ctx *c, int incoming_queue, uint8_t pif,
sa_family_t af, const void *saddr, const void *daddr,
uint8_t ttl, const struct pool *p, int idx,
const struct timespec *now);
diff --git a/udp_flow.c b/udp_flow.c
index 8907f2f72741..b4a709b8d976 100644
--- a/udp_flow.c
+++ b/udp_flow.c
@@ -266,17 +266,19 @@ flow_sidx_t udp_flow_from_sock(const struct ctx *c, uint8_t pif,
/**
* udp_flow_from_tap() - Find or create UDP flow for tap packets
* @c: Execution context
+ * @queue: RX queue number for the flow
* @pif: pif on which the packet is arriving
* @af: Address family, AF_INET or AF_INET6
* @saddr: Source address on guest side
* @daddr: Destination address guest side
* @srcport: Source port on guest side
* @dstport: Destination port on guest side
+ * @now: Current timestamp
*
* Return: sidx for the destination side of the flow for this packet, or
* FLOW_SIDX_NONE if we couldn't find or create a flow.
*/
-flow_sidx_t udp_flow_from_tap(const struct ctx *c,
+flow_sidx_t udp_flow_from_tap(const struct ctx *c, int queue,
uint8_t pif, sa_family_t af,
const void *saddr, const void *daddr,
in_port_t srcport, in_port_t dstport,
@@ -293,6 +295,8 @@ flow_sidx_t udp_flow_from_tap(const struct ctx *c,
srcport, dstport);
if ((uflow = udp_at_sidx(sidx))) {
uflow->ts = now->tv_sec;
+ /* update queue */
+ flow_queue_set(&uflow->f, queue);
return flow_sidx_opposite(sidx);
}
@@ -316,6 +320,8 @@ flow_sidx_t udp_flow_from_tap(const struct ctx *c,
return FLOW_SIDX_NONE;
}
+ flow_queue_set(&flow->f, queue);
+
return udp_flow_new(c, flow, now);
}
diff --git a/udp_flow.h b/udp_flow.h
index 4c528e95ca66..4a057a9d44a8 100644
--- a/udp_flow.h
+++ b/udp_flow.h
@@ -36,7 +36,7 @@ flow_sidx_t udp_flow_from_sock(const struct ctx *c, uint8_t pif,
const union inany_addr *dst, in_port_t port,
const union sockaddr_inany *s_in,
const struct timespec *now);
-flow_sidx_t udp_flow_from_tap(const struct ctx *c,
+flow_sidx_t udp_flow_from_tap(const struct ctx *c, int queue,
uint8_t pif, sa_family_t af,
const void *saddr, const void *daddr,
in_port_t srcport, in_port_t dstport,
diff --git a/udp_vu.c b/udp_vu.c
index 099677f914e7..cd2c9c516d44 100644
--- a/udp_vu.c
+++ b/udp_vu.c
@@ -202,9 +202,11 @@ static void udp_vu_csum(const struct flowside *toside, int iov_used)
void udp_vu_sock_to_tap(const struct ctx *c, int s, int n, flow_sidx_t tosidx)
{
const struct flowside *toside = flowside_at_sidx(tosidx);
+ const struct udp_flow *uflow = udp_at_sidx(tosidx);
bool v6 = !(inany_v4(&toside->eaddr) && inany_v4(&toside->oaddr));
+ int rx_queue = flow_rx_virtqueue(&uflow->f);
struct vu_dev *vdev = c->vdev;
- struct vu_virtq *vq = &vdev->vq[VHOST_USER_RX_QUEUE];
+ struct vu_virtq *vq = &vdev->vq[rx_queue];
int i;
for (i = 0; i < n; i++) {
diff --git a/vu_common.c b/vu_common.c
index 8904403e66af..56f26317b192 100644
--- a/vu_common.c
+++ b/vu_common.c
@@ -196,11 +196,11 @@ static void vu_handle_tx(struct vu_dev *vdev, int index,
data = IOV_TAIL(elem[count].out_sg, elem[count].out_num, 0);
if (IOV_DROP_HEADER(&data, struct virtio_net_hdr_mrg_rxbuf))
- tap_add_packet(vdev->context, &data, now);
+ tap_add_packet(vdev->context, index, &data, now);
count++;
}
- tap_handler(vdev->context, now);
+ tap_handler(vdev->context, index, now);
if (count) {
int i;
--
2.51.0
^ permalink raw reply [flat|nested] 13+ messages in thread* Re: [PATCH 3/4] multiqueue: Add queue-aware flow management for multiqueue support
2025-11-07 14:39 ` [PATCH 3/4] multiqueue: Add queue-aware flow management for multiqueue support Laurent Vivier
@ 2025-11-10 5:54 ` David Gibson
0 siblings, 0 replies; 13+ messages in thread
From: David Gibson @ 2025-11-10 5:54 UTC (permalink / raw)
To: Laurent Vivier; +Cc: passt-dev
[-- Attachment #1: Type: text/plain, Size: 32027 bytes --]
On Fri, Nov 07, 2025 at 03:39:00PM +0100, Laurent Vivier wrote:
> Packets are now routed to the correct RX queue based on which TX queue
> they arrived on, rather than always using queue 0.
Am I missing something, or is this not quite accurate. The packets
are associated by flow with other packets that came on a particular
Tx queue, but AIUI the packet itself didn't arrive on a Tx queue, but
rather from a socket.
> Note: Flows initiated from the host (via sockets, udp_flow_from_sock())
> currently default to queue 0, as they don't have an associated incoming
> queue.
>
> Signed-off-by: Laurent Vivier <lvivier@redhat.com>
> ---
> flow.c | 32 ++++++++++++++++++++++
> flow.h | 10 +++++++
> icmp.c | 23 +++++++++-------
> icmp.h | 2 +-
> tap.c | 77 +++++++++++++++++++++++++++------------------------
> tap.h | 5 ++--
> tcp.c | 79 +++++++++++++++++++++++++++++------------------------
> tcp.h | 2 +-
> tcp_vu.c | 8 ++++--
> udp.c | 33 ++++++++++++----------
> udp.h | 12 ++++----
> udp_flow.c | 8 +++++-
> udp_flow.h | 2 +-
> udp_vu.c | 4 ++-
> vu_common.c | 4 +--
> 15 files changed, 187 insertions(+), 114 deletions(-)
>
> diff --git a/flow.c b/flow.c
> index 278a9cf0ac6d..8e9d7e5e1847 100644
> --- a/flow.c
> +++ b/flow.c
> @@ -405,6 +405,37 @@ void flow_epollid_register(int epollid, int epollfd)
> epoll_id_to_fd[epollid] = epollfd;
> }
>
> +/**
> + * flow_rx_virtqueue() - Get RX (receive) queue number for a flow
Maybe avoid the name "virtqueue". I believe the tap device can also
do multiqueue, and we may want to reuse this queue infrastructure for
that.
> + * @f: Flow to query (may be NULL)
> + *
> + * Return: RX queue number for the flow, or 0 if flow is NULL or has no
> + * valid queue assignment
> + */
> +int flow_rx_virtqueue(const struct flow_common *f)
> +{
> + if (f == NULL || f->queueid == FLOW_QUEUEID_INVALID)
Any reason to have the second special case here, rather than
initializing f->queueid to 0?
> + return 0;
> + return f->queueid << 1;
I generally think it's clearer to use * 2 rather than << 1 here. I
expect the compiler to turn it into the same thing.
> +}
> +
> +/**
> + * flow_queue_set() - Set queue pair assignment for a flow
> + * @f: Flow to update
> + * @queueid: Queue pair number to assign (even number, RX queue; TX is RX+1)
In comments using "from guest" / "to guest" instead of Rx/Tx is
probably a good idea. Particularly in code that's not necessarily
specific to VU.
I think it is a bit confusing that somewhere we have queue pair IDs,
sometimes absolute queue numbers. I think we should standardise on
one throughout as much of the code as possible.
> + */
> +void flow_queue_set(struct flow_common *f, int queueid)
> +{
> + queueid >>= 1;
> +
> + ASSERT(queueid < FLOW_QUEUEID_MAX);
> +
> + flow_trace((union flow *)f, "updating queue from %d to %d",
> + f->queueid, queueid);
> +
> + f->queueid = queueid;
> +}
> +
> /**
> * flow_initiate_() - Move flow to INI, setting pif[INISIDE]
> * @flow: Flow to change state
> @@ -609,6 +640,7 @@ union flow *flow_alloc(void)
> flow_new_entry = flow;
> memset(flow, 0, sizeof(*flow));
> flow_epollid_clear(&flow->f);
> + flow->f.queueid = FLOW_QUEUEID_INVALID;
> flow_set_state(&flow->f, FLOW_STATE_NEW);
>
> return flow;
> diff --git a/flow.h b/flow.h
> index b43b0b1dd7f2..44ab4ae8fd6a 100644
> --- a/flow.h
> +++ b/flow.h
> @@ -179,6 +179,8 @@ int flowside_connect(const struct ctx *c, int s,
> * @side[]: Information for each side of the flow
> * @tap_omac: MAC address of remote endpoint as seen from the guest
> * @epollid: epollfd identifier, or EPOLLFD_ID_INVALID
> + * @queueid: Queue pair number assigned to this flow
> + * (FLOW_QUEUEID_INVALID if not assigned)
> */
> struct flow_common {
> #ifdef __GNUC__
> @@ -199,6 +201,8 @@ struct flow_common {
>
> #define EPOLLFD_ID_BITS 8
> unsigned int epollid:EPOLLFD_ID_BITS;
> +#define FLOW_QUEUEID_BITS 5
> + unsigned int queueid:FLOW_QUEUEID_BITS;
> };
>
> #define EPOLLFD_ID_DEFAULT 0
> @@ -206,6 +210,10 @@ struct flow_common {
> #define EPOLLFD_ID_MAX (EPOLLFD_ID_SIZE - 1)
> #define EPOLLFD_ID_INVALID EPOLLFD_ID_MAX
>
> +#define FLOW_QUEUEID_SIZE (1 << FLOW_QUEUEID_BITS)
"SIZE" is a bit ambiguous here. Maybe "NUM" instead?
> +#define FLOW_QUEUEID_MAX (FLOW_QUEUEID_SIZE - 1)
> +#define FLOW_QUEUEID_INVALID FLOW_QUEUEID_MAX
> +
> #define FLOW_INDEX_BITS 17 /* 128k - 1 */
> #define FLOW_MAX MAX_FROM_BITS(FLOW_INDEX_BITS)
>
> @@ -266,6 +274,8 @@ int flow_epollfd(const struct flow_common *f);
> void flow_epollid_set(struct flow_common *f, int epollid);
> void flow_epollid_clear(struct flow_common *f);
> void flow_epollid_register(int epollid, int epollfd);
> +int flow_rx_virtqueue(const struct flow_common *f);
> +void flow_queue_set(struct flow_common *f, int queueid);
> void flow_defer_handler(const struct ctx *c, const struct timespec *now);
> int flow_migrate_source_early(struct ctx *c, const struct migrate_stage *stage,
> int fd);
> diff --git a/icmp.c b/icmp.c
> index d58499c3bf5c..80e8753072fa 100644
> --- a/icmp.c
> +++ b/icmp.c
> @@ -132,13 +132,13 @@ void icmp_sock_handler(const struct ctx *c, union epoll_ref ref)
> const struct in_addr *daddr = inany_v4(&ini->eaddr);
>
> ASSERT(saddr && daddr); /* Must have IPv4 addresses */
> - tap_icmp4_send(c, VHOST_USER_RX_QUEUE, *saddr, *daddr, buf,
> + tap_icmp4_send(c, flow_rx_virtqueue(&pingf->f), *saddr, *daddr, buf,
> pingf->f.tap_omac, n);
> } else if (pingf->f.type == FLOW_PING6) {
> const struct in6_addr *saddr = &ini->oaddr.a6;
> const struct in6_addr *daddr = &ini->eaddr.a6;
>
> - tap_icmp6_send(c, VHOST_USER_RX_QUEUE, saddr, daddr, buf,
> + tap_icmp6_send(c, flow_rx_virtqueue(&pingf->f), saddr, daddr, buf,
> pingf->f.tap_omac, n);
> }
> return;
> @@ -238,17 +238,18 @@ cancel:
>
> /**
> * icmp_tap_handler() - Handle packets from tap
> - * @c: Execution context
> - * @pif: pif on which the packet is arriving
> - * @af: Address family, AF_INET or AF_INET6
> - * @saddr: Source address
> - * @daddr: Destination address
> - * @data: Single packet with ICMP/ICMPv6 header
> - * @now: Current timestamp
> + * @c: Execution context
> + * @incoming_queue: Incoming queue number
> + * @pif: pif on which the packet is arriving
> + * @af: Address family, AF_INET or AF_INET6
> + * @saddr: Source address
> + * @daddr: Destination address
> + * @data: Single packet with ICMP/ICMPv6 header
> + * @now: Current timestamp
> *
> * Return: count of consumed packets (always 1, even if malformed)
> */
> -int icmp_tap_handler(const struct ctx *c, uint8_t pif, sa_family_t af,
> +int icmp_tap_handler(const struct ctx *c, int incoming_queue, uint8_t pif, sa_family_t af,
> const void *saddr, const void *daddr,
> struct iov_tail *data, const struct timespec *now)
> {
> @@ -309,6 +310,8 @@ int icmp_tap_handler(const struct ctx *c, uint8_t pif, sa_family_t af,
> else if (!(pingf = icmp_ping_new(c, af, id, saddr, daddr)))
> return 1;
>
> + flow_queue_set(&pingf->f, incoming_queue);
> +
I'd kind of like the initial setting of the flow's queue to be tied to
one of the existing flow state transitions, to ensure it can't be
forgotten.
> tgt = &pingf->f.side[TGTSIDE];
>
> ASSERT(flow_proto[pingf->f.type] == proto);
> diff --git a/icmp.h b/icmp.h
> index 1a0e6205f087..6d6d6358bb33 100644
> --- a/icmp.h
> +++ b/icmp.h
> @@ -10,7 +10,7 @@ struct ctx;
> struct icmp_ping_flow;
>
> void icmp_sock_handler(const struct ctx *c, union epoll_ref ref);
> -int icmp_tap_handler(const struct ctx *c, uint8_t pif, sa_family_t af,
> +int icmp_tap_handler(const struct ctx *c, int incoming_queue, uint8_t pif, sa_family_t af,
> const void *saddr, const void *daddr,
> struct iov_tail *data, const struct timespec *now);
> void icmp_init(void);
> diff --git a/tap.c b/tap.c
> index 1308d49242e8..9a4399d947a3 100644
> --- a/tap.c
> +++ b/tap.c
> @@ -702,15 +702,17 @@ static bool tap4_is_fragment(const struct iphdr *iph,
>
> /**
> * tap4_handler() - IPv4 and ARP packet handler for tap file descriptor
> - * @c: Execution context
> - * @in: Ingress packet pool, packets with Ethernet headers
> - * @now: Current timestamp
> + * @c: Execution context
> + * @incoming_queue: Incoming queue number
> + * @in: Ingress packet pool, packets with Ethernet headers
> + * @now: Current timestamp
> *
> * Return: count of packets consumed by handlers
> */
> -static int tap4_handler(struct ctx *c, const struct pool *in,
> - const struct timespec *now)
> +static int tap4_handler(struct ctx *c, int incoming_queue,
> + const struct pool *in, const struct timespec *now)
> {
> + int outgoing_queue = incoming_queue & ~1;
> unsigned int i, j, seq_count;
> struct tap4_l4_t *seq;
>
> @@ -736,7 +738,7 @@ resume:
> if (!eh)
> continue;
> if (ntohs(eh->h_proto) == ETH_P_ARP) {
> - arp(c, VHOST_USER_RX_QUEUE, &data);
> + arp(c, outgoing_queue, &data);
> continue;
> }
>
> @@ -783,7 +785,7 @@ resume:
>
> tap_packet_debug(iph, NULL, NULL, 0, NULL, 1);
>
> - icmp_tap_handler(c, PIF_TAP, AF_INET,
> + icmp_tap_handler(c, incoming_queue, PIF_TAP, AF_INET,
> &iph->saddr, &iph->daddr,
> &data, now);
> continue;
> @@ -797,7 +799,7 @@ resume:
> struct iov_tail eh_data;
>
> packet_get(in, i, &eh_data);
> - if (dhcp(c, VHOST_USER_RX_QUEUE, &eh_data))
> + if (dhcp(c, outgoing_queue, &eh_data))
> continue;
> }
>
> @@ -860,14 +862,14 @@ append:
> if (c->no_tcp)
> continue;
> for (k = 0; k < p->count; )
> - k += tcp_tap_handler(c, PIF_TAP, AF_INET,
> + k += tcp_tap_handler(c, incoming_queue, PIF_TAP, AF_INET,
> &seq->saddr, &seq->daddr,
> 0, p, k, now);
> } else if (seq->protocol == IPPROTO_UDP) {
> if (c->no_udp)
> continue;
> for (k = 0; k < p->count; )
> - k += udp_tap_handler(c, PIF_TAP, AF_INET,
> + k += udp_tap_handler(c, incoming_queue, PIF_TAP, AF_INET,
> &seq->saddr, &seq->daddr,
> seq->ttl, p, k, now);
> }
> @@ -881,15 +883,17 @@ append:
>
> /**
> * tap6_handler() - IPv6 packet handler for tap file descriptor
> - * @c: Execution context
> - * @in: Ingress packet pool, packets with Ethernet headers
> - * @now: Current timestamp
> + * @c: Execution context
> + * @incoming_queue: Incoming queue number
> + * @in: Ingress packet pool, packets with Ethernet headers
> + * @now: Current timestamp
> *
> * Return: count of packets consumed by handlers
> */
> -static int tap6_handler(struct ctx *c, const struct pool *in,
> - const struct timespec *now)
> +static int tap6_handler(struct ctx *c, int incoming_queue,
> + const struct pool *in, const struct timespec *now)
> {
> + int outgoing_queue = incoming_queue & ~1;
> unsigned int i, j, seq_count = 0;
> struct tap6_l4_t *seq;
>
> @@ -965,12 +969,12 @@ resume:
> continue;
>
> ndp_data = data;
> - if (ndp(c, VHOST_USER_RX_QUEUE, saddr, &ndp_data))
> + if (ndp(c, outgoing_queue, saddr, &ndp_data))
> continue;
>
> tap_packet_debug(NULL, ip6h, NULL, proto, NULL, 1);
>
> - icmp_tap_handler(c, PIF_TAP, AF_INET6,
> + icmp_tap_handler(c, incoming_queue, PIF_TAP, AF_INET6,
> saddr, daddr, &data, now);
> continue;
> }
> @@ -984,8 +988,7 @@ resume:
> if (proto == IPPROTO_UDP) {
> struct iov_tail uh_data = data;
>
> - if (dhcpv6(c, VHOST_USER_RX_QUEUE, &uh_data, saddr,
> - daddr))
> + if (dhcpv6(c, outgoing_queue, &uh_data, saddr, daddr))
> continue;
> }
>
> @@ -1053,14 +1056,14 @@ append:
> if (c->no_tcp)
> continue;
> for (k = 0; k < p->count; )
> - k += tcp_tap_handler(c, PIF_TAP, AF_INET6,
> + k += tcp_tap_handler(c, incoming_queue, PIF_TAP, AF_INET6,
> &seq->saddr, &seq->daddr,
> seq->flow_lbl, p, k, now);
> } else if (seq->protocol == IPPROTO_UDP) {
> if (c->no_udp)
> continue;
> for (k = 0; k < p->count; )
> - k += udp_tap_handler(c, PIF_TAP, AF_INET6,
> + k += udp_tap_handler(c, incoming_queue, PIF_TAP, AF_INET6,
> &seq->saddr, &seq->daddr,
> seq->hop_limit, p, k, now);
> }
> @@ -1083,22 +1086,24 @@ void tap_flush_pools(void)
>
> /**
> * tap_handler() - IPv4/IPv6 and ARP packet handler for tap file descriptor
> - * @c: Execution context
> - * @now: Current timestamp
> + * @c: Execution context
> + * @incoming_queue: Incoming queue number
> + * @now: Current timestamp
> */
> -void tap_handler(struct ctx *c, const struct timespec *now)
> +void tap_handler(struct ctx *c, int incoming_queue, const struct timespec *now)
> {
> - tap4_handler(c, pool_tap4, now);
> - tap6_handler(c, pool_tap6, now);
> + tap4_handler(c, incoming_queue, pool_tap4, now);
> + tap6_handler(c, incoming_queue, pool_tap6, now);
> }
>
> /**
> * tap_add_packet() - Queue/capture packet, update notion of guest MAC address
> - * @c: Execution context
> - * @data: Packet to add to the pool
> - * @now: Current timestamp
> + * @c: Execution context
> + * @incoming_queue: Incoming queue number
> + * @data: Packet to add to the pool
> + * @now: Current timestamp
> */
> -void tap_add_packet(struct ctx *c, struct iov_tail *data,
> +void tap_add_packet(struct ctx *c, int incoming_queue, struct iov_tail *data,
> const struct timespec *now)
> {
> struct ethhdr eh_storage;
> @@ -1123,14 +1128,14 @@ void tap_add_packet(struct ctx *c, struct iov_tail *data,
> case ETH_P_ARP:
> case ETH_P_IP:
> if (!pool_can_fit(pool_tap4, data)) {
> - tap4_handler(c, pool_tap4, now);
> + tap4_handler(c, incoming_queue, pool_tap4, now);
This is a bit murky. You're using the incoming_queue for the new
packet to process all the existiing packets in the pool. I guess that
will work out because you'll have a separate pool for each
thread/queue. But that's kind of unclear at this point. Maybe the
queue number should be more explicitly part of the pool metadata?
> pool_flush(pool_tap4);
> }
> packet_add(pool_tap4, data);
> break;
> case ETH_P_IPV6:
> if (!pool_can_fit(pool_tap6, data)) {
> - tap6_handler(c, pool_tap6, now);
> + tap6_handler(c, incoming_queue, pool_tap6, now);
> pool_flush(pool_tap6);
> }
> packet_add(pool_tap6, data);
> @@ -1217,7 +1222,7 @@ static void tap_passt_input(struct ctx *c, const struct timespec *now)
> n -= sizeof(uint32_t);
>
> data = IOV_TAIL_FROM_BUF(p, l2len, 0);
> - tap_add_packet(c, &data, now);
> + tap_add_packet(c, 0, &data, now);
>
> p += l2len;
> n -= l2len;
> @@ -1226,7 +1231,7 @@ static void tap_passt_input(struct ctx *c, const struct timespec *now)
> partial_len = n;
> partial_frame = p;
>
> - tap_handler(c, now);
> + tap_handler(c, 0, now);
> }
>
> /**
> @@ -1285,10 +1290,10 @@ static void tap_pasta_input(struct ctx *c, const struct timespec *now)
> continue;
>
> data = IOV_TAIL_FROM_BUF(pkt_buf + n, len, 0);
> - tap_add_packet(c, &data, now);
> + tap_add_packet(c, 0, &data, now);
> }
>
> - tap_handler(c, now);
> + tap_handler(c, 0, now);
> }
>
> /**
> diff --git a/tap.h b/tap.h
> index 76403a43edbc..fe9455ffcf4b 100644
> --- a/tap.h
> +++ b/tap.h
> @@ -119,7 +119,8 @@ int tap_sock_unix_open(char *sock_path);
> void tap_sock_reset(struct ctx *c);
> void tap_backend_init(struct ctx *c);
> void tap_flush_pools(void);
> -void tap_handler(struct ctx *c, const struct timespec *now);
> -void tap_add_packet(struct ctx *c, struct iov_tail *data,
> +void tap_handler(struct ctx *c, int incoming_queue,
> + const struct timespec *now);
> +void tap_add_packet(struct ctx *c, int incoming_queue, struct iov_tail *data,
> const struct timespec *now);
> #endif /* TAP_H */
> diff --git a/tcp.c b/tcp.c
> index 5ce34baa8a5a..78494a2dc69f 100644
> --- a/tcp.c
> +++ b/tcp.c
> @@ -1496,21 +1496,23 @@ static void tcp_bind_outbound(const struct ctx *c,
>
> /**
> * tcp_conn_from_tap() - Handle connection request (SYN segment) from tap
> - * @c: Execution context
> - * @af: Address family, AF_INET or AF_INET6
> - * @saddr: Source address, pointer to in_addr or in6_addr
> - * @daddr: Destination address, pointer to in_addr or in6_addr
> - * @th: TCP header from tap: caller MUST ensure it's there
> - * @opts: Pointer to start of options
> - * @optlen: Bytes in options: caller MUST ensure available length
> - * @now: Current timestamp
> + * @c: Execution context
> + * @incoming_queue: TX queue number for the flow
> + * @af: Address family, AF_INET or AF_INET6
> + * @saddr: Source address, pointer to in_addr or in6_addr
> + * @daddr: Destination address, pointer to in_addr or in6_addr
> + * @th: TCP header from tap: caller MUST ensure it's there
> + * @opts: Pointer to start of options
> + * @optlen: Bytes in options: caller MUST ensure available length
> + * @now: Current timestamp
> *
> * #syscalls:vu getsockname
> */
> -static void tcp_conn_from_tap(const struct ctx *c, sa_family_t af,
> - const void *saddr, const void *daddr,
> - const struct tcphdr *th, const char *opts,
> - size_t optlen, const struct timespec *now)
> +static void tcp_conn_from_tap(const struct ctx *c, int incoming_queue,
> + sa_family_t af, const void *saddr,
> + const void *daddr, const struct tcphdr *th,
> + const char *opts, size_t optlen,
> + const struct timespec *now)
> {
> in_port_t srcport = ntohs(th->source);
> in_port_t dstport = ntohs(th->dest);
> @@ -1622,6 +1624,7 @@ static void tcp_conn_from_tap(const struct ctx *c, sa_family_t af,
> conn_event(c, conn, TAP_SYN_ACK_SENT);
> }
>
> + flow_queue_set(&conn->f, incoming_queue);
> tcp_epoll_ctl(c, conn);
>
> if (c->mode == MODE_VU) { /* To rebind to same oport after migration */
> @@ -1983,16 +1986,16 @@ static void tcp_conn_from_sock_finish(const struct ctx *c,
>
> /**
> * tcp_rst_no_conn() - Send RST in response to a packet with no connection
> - * @c: Execution context
> - * @queue: Queue to use to send the reply
> - * @af: Address family, AF_INET or AF_INET6
> - * @saddr: Source address of the packet we're responding to
> - * @daddr: Destination address of the packet we're responding to
> - * @flow_lbl: IPv6 flow label (ignored for IPv4)
> - * @th: TCP header of the packet we're responding to
> - * @l4len: Packet length, including TCP header
> - */
> -static void tcp_rst_no_conn(const struct ctx *c, int queue, int af,
> + * @c: Execution context
> + * @outgoing_queue: Queue to use to send the reply
> + * @af: Address family, AF_INET or AF_INET6
> + * @saddr: Source address of the packet we're responding to
> + * @daddr: Destination address of the packet we're responding to
> + * @flow_lbl: IPv6 flow label (ignored for IPv4)
> + * @th: TCP header of the packet we're responding to
> + * @l4len: Packet length, including TCP header
> + */
> +static void tcp_rst_no_conn(const struct ctx *c, int outgoing_queue, int af,
> const void *saddr, const void *daddr,
> uint32_t flow_lbl,
> const struct tcphdr *th, size_t l4len)
> @@ -2050,24 +2053,25 @@ static void tcp_rst_no_conn(const struct ctx *c, int queue, int af,
>
> tcp_update_csum(psum, rsth, &payload);
> rst_l2len = ((char *)rsth - buf) + sizeof(*rsth);
> - tap_send_single(c, queue, buf, rst_l2len);
> + tap_send_single(c, outgoing_queue, buf, rst_l2len);
> }
>
> /**
> * tcp_tap_handler() - Handle packets from tap and state transitions
> - * @c: Execution context
> - * @pif: pif on which the packet is arriving
> - * @af: Address family, AF_INET or AF_INET6
> - * @saddr: Source address
> - * @daddr: Destination address
> - * @flow_lbl: IPv6 flow label (ignored for IPv4)
> - * @p: Pool of TCP packets, with TCP headers
> - * @idx: Index of first packet in pool to process
> - * @now: Current timestamp
> + * @c: Execution context
> + * @incoming_queue: Incoming queue number
> + * @pif: pif on which the packet is arriving
> + * @af: Address family, AF_INET or AF_INET6
> + * @saddr: Source address
> + * @daddr: Destination address
> + * @flow_lbl: IPv6 flow label (ignored for IPv4)
> + * @p: Pool of TCP packets, with TCP headers
> + * @idx: Index of first packet in pool to process
> + * @now: Current timestamp
> *
> * Return: count of consumed packets
> */
> -int tcp_tap_handler(const struct ctx *c, uint8_t pif,
> +int tcp_tap_handler(const struct ctx *c, int incoming_queue, uint8_t pif,
> sa_family_t af, const void *saddr, const void *daddr,
> uint32_t flow_lbl, const struct pool *p, int idx,
> const struct timespec *now)
> @@ -2107,11 +2111,11 @@ int tcp_tap_handler(const struct ctx *c, uint8_t pif,
> /* New connection from tap */
> if (!flow) {
> if (opts && th->syn && !th->ack)
> - tcp_conn_from_tap(c, af, saddr, daddr, th,
> + tcp_conn_from_tap(c, incoming_queue, af, saddr, daddr, th,
> opts, optlen, now);
> else
> - tcp_rst_no_conn(c, VHOST_USER_RX_QUEUE, af, saddr,
> - daddr, flow_lbl, th, l4len);
> + tcp_rst_no_conn(c, incoming_queue & ~1, af, saddr, daddr,
> + flow_lbl, th, l4len);
> return 1;
> }
>
> @@ -2119,6 +2123,9 @@ int tcp_tap_handler(const struct ctx *c, uint8_t pif,
> ASSERT(pif_at_sidx(sidx) == PIF_TAP);
> conn = &flow->tcp;
>
> + /* update queue */
> + flow_queue_set(&flow->f, incoming_queue);
> +
> flow_trace(conn, "packet length %zu from tap", l4len);
>
> if (th->rst) {
> diff --git a/tcp.h b/tcp.h
> index 320683ce5679..cddd36cadc97 100644
> --- a/tcp.h
> +++ b/tcp.h
> @@ -15,7 +15,7 @@ void tcp_listen_handler(const struct ctx *c, union epoll_ref ref,
> const struct timespec *now);
> void tcp_sock_handler(const struct ctx *c,
> union epoll_ref ref, uint32_t events);
> -int tcp_tap_handler(const struct ctx *c, uint8_t pif,
> +int tcp_tap_handler(const struct ctx *c, int incoming_queue, uint8_t pif,
> sa_family_t af, const void *saddr, const void *daddr,
> uint32_t flow_lbl, const struct pool *p, int idx,
> const struct timespec *now);
> diff --git a/tcp_vu.c b/tcp_vu.c
> index 1c81ce376dad..40f552087bc5 100644
> --- a/tcp_vu.c
> +++ b/tcp_vu.c
> @@ -70,15 +70,16 @@ static size_t tcp_vu_hdrlen(bool v6)
> */
> int tcp_vu_send_flag(const struct ctx *c, struct tcp_tap_conn *conn, int flags)
> {
> + int rx_queue = flow_rx_virtqueue(&conn->f);
> struct vu_dev *vdev = c->vdev;
> - struct vu_virtq *vq = &vdev->vq[VHOST_USER_RX_QUEUE];
> - size_t optlen, hdrlen;
> + struct vu_virtq *vq = &vdev->vq[rx_queue];
> struct vu_virtq_element flags_elem[2];
> struct ipv6hdr *ip6h = NULL;
> struct iphdr *ip4h = NULL;
> struct iovec flags_iov[2];
> struct tcp_syn_opts *opts;
> struct iov_tail payload;
> + size_t optlen, hdrlen;
> struct tcphdr *th;
> struct ethhdr *eh;
> uint32_t seq;
> @@ -348,8 +349,9 @@ static void tcp_vu_prepare(const struct ctx *c, struct tcp_tap_conn *conn,
> int tcp_vu_data_from_sock(const struct ctx *c, struct tcp_tap_conn *conn)
> {
> uint32_t wnd_scaled = conn->wnd_from_tap << conn->ws_from_tap;
> + int rx_queue = flow_rx_virtqueue(&conn->f);
> struct vu_dev *vdev = c->vdev;
> - struct vu_virtq *vq = &vdev->vq[VHOST_USER_RX_QUEUE];
> + struct vu_virtq *vq = &vdev->vq[rx_queue];
> ssize_t len, previous_dlen;
> int i, iov_cnt, head_cnt;
> size_t hdrlen, fillsize;
> diff --git a/udp.c b/udp.c
> index 868ffebb5802..a0eb719888cd 100644
> --- a/udp.c
> +++ b/udp.c
> @@ -636,14 +636,15 @@ static int udp_sock_recverr(const struct ctx *c, int s, flow_sidx_t sidx,
> if (hdr->cmsg_level == IPPROTO_IP &&
> (o4 = inany_v4(&otap)) && inany_v4(&toside->eaddr)) {
> dlen = MIN(dlen, ICMP4_MAX_DLEN);
> - udp_send_tap_icmp4(c, VHOST_USER_RX_QUEUE, ee, toside, *o4,
> - data, dlen);
> + udp_send_tap_icmp4(c, flow_rx_virtqueue(&uflow->f), ee, toside,
Kind of a something for an earlier patch, but given how frequently and
where it's used, it's worth having a shorter name for
flow_rx_virtqueue(). And probably a wrapper macro so it can be called
on uflow directly as well.
> + *o4, data, dlen);
> return 1;
> }
>
> if (hdr->cmsg_level == IPPROTO_IPV6 && !inany_v4(&toside->eaddr)) {
> - udp_send_tap_icmp6(c, VHOST_USER_RX_QUEUE, ee, toside,
> - &otap.a6, data, dlen, FLOW_IDX(uflow));
> + udp_send_tap_icmp6(c, flow_rx_virtqueue(&uflow->f), ee,
> + toside, &otap.a6, data, dlen,
> + FLOW_IDX(uflow));
> return 1;
> }
>
> @@ -971,25 +972,27 @@ fail:
>
> /**
> * udp_tap_handler() - Handle packets from tap
> - * @c: Execution context
> - * @pif: pif on which the packet is arriving
> - * @af: Address family, AF_INET or AF_INET6
> - * @saddr: Source address
> - * @daddr: Destination address
> - * @ttl: TTL or hop limit for packets to be sent in this call
> - * @p: Pool of UDP packets, with UDP headers
> - * @idx: Index of first packet to process
> - * @now: Current timestamp
> + * @c: Execution context
> + * @incoming_queue: Incoming queue number
> + * @pif: pif on which the packet is arriving
> + * @af: Address family, AF_INET or AF_INET6
> + * @saddr: Source address
> + * @daddr: Destination address
> + * @ttl: TTL or hop limit for packets to be sent in this call
> + * @p: Pool of UDP packets, with UDP headers
> + * @idx: Index of first packet to process
> + * @now: Current timestamp
> *
> * Return: count of consumed packets
> *
> * #syscalls sendmmsg
> */
> -int udp_tap_handler(const struct ctx *c, uint8_t pif,
> +int udp_tap_handler(const struct ctx *c, int incoming_queue, uint8_t pif,
> sa_family_t af, const void *saddr, const void *daddr,
> uint8_t ttl, const struct pool *p, int idx,
> const struct timespec *now)
> {
> + int outgoing_queue = incoming_queue & ~1;
> const struct flowside *toside;
> struct mmsghdr mm[UIO_MAXIOV];
> union sockaddr_inany to_sa;
> @@ -1019,7 +1022,7 @@ int udp_tap_handler(const struct ctx *c, uint8_t pif,
> src = ntohs(uh->source);
> dst = ntohs(uh->dest);
>
> - tosidx = udp_flow_from_tap(c, pif, af, saddr, daddr, src, dst, now);
> + tosidx = udp_flow_from_tap(c, outgoing_queue, pif, af, saddr, daddr, src, dst, now);
> if (!(uflow = udp_at_sidx(tosidx))) {
> char sstr[INET6_ADDRSTRLEN], dstr[INET6_ADDRSTRLEN];
>
> diff --git a/udp.h b/udp.h
> index f1d83f380b3f..8ba4ccfe646a 100644
> --- a/udp.h
> +++ b/udp.h
> @@ -7,11 +7,13 @@
> #define UDP_H
>
> void udp_portmap_clear(void);
> -void udp_listen_sock_handler(const struct ctx *c, union epoll_ref ref,
> - uint32_t events, const struct timespec *now);
> -void udp_sock_handler(const struct ctx *c, union epoll_ref ref,
> - uint32_t events, const struct timespec *now);
> -int udp_tap_handler(const struct ctx *c, uint8_t pif,
> +void udp_listen_sock_handler(const struct ctx *c,
> + union epoll_ref ref, uint32_t events,
> + const struct timespec *now);
> +void udp_sock_handler(const struct ctx *c,
> + union epoll_ref ref, uint32_t events,
> + const struct timespec *now);
> +int udp_tap_handler(const struct ctx *c, int incoming_queue, uint8_t pif,
> sa_family_t af, const void *saddr, const void *daddr,
> uint8_t ttl, const struct pool *p, int idx,
> const struct timespec *now);
> diff --git a/udp_flow.c b/udp_flow.c
> index 8907f2f72741..b4a709b8d976 100644
> --- a/udp_flow.c
> +++ b/udp_flow.c
> @@ -266,17 +266,19 @@ flow_sidx_t udp_flow_from_sock(const struct ctx *c, uint8_t pif,
> /**
> * udp_flow_from_tap() - Find or create UDP flow for tap packets
> * @c: Execution context
> + * @queue: RX queue number for the flow
> * @pif: pif on which the packet is arriving
> * @af: Address family, AF_INET or AF_INET6
> * @saddr: Source address on guest side
> * @daddr: Destination address guest side
> * @srcport: Source port on guest side
> * @dstport: Destination port on guest side
> + * @now: Current timestamp
> *
> * Return: sidx for the destination side of the flow for this packet, or
> * FLOW_SIDX_NONE if we couldn't find or create a flow.
> */
> -flow_sidx_t udp_flow_from_tap(const struct ctx *c,
> +flow_sidx_t udp_flow_from_tap(const struct ctx *c, int queue,
> uint8_t pif, sa_family_t af,
> const void *saddr, const void *daddr,
> in_port_t srcport, in_port_t dstport,
> @@ -293,6 +295,8 @@ flow_sidx_t udp_flow_from_tap(const struct ctx *c,
> srcport, dstport);
> if ((uflow = udp_at_sidx(sidx))) {
> uflow->ts = now->tv_sec;
> + /* update queue */
> + flow_queue_set(&uflow->f, queue);
> return flow_sidx_opposite(sidx);
> }
>
> @@ -316,6 +320,8 @@ flow_sidx_t udp_flow_from_tap(const struct ctx *c,
> return FLOW_SIDX_NONE;
> }
>
> + flow_queue_set(&flow->f, queue);
> +
> return udp_flow_new(c, flow, now);
> }
>
> diff --git a/udp_flow.h b/udp_flow.h
> index 4c528e95ca66..4a057a9d44a8 100644
> --- a/udp_flow.h
> +++ b/udp_flow.h
> @@ -36,7 +36,7 @@ flow_sidx_t udp_flow_from_sock(const struct ctx *c, uint8_t pif,
> const union inany_addr *dst, in_port_t port,
> const union sockaddr_inany *s_in,
> const struct timespec *now);
> -flow_sidx_t udp_flow_from_tap(const struct ctx *c,
> +flow_sidx_t udp_flow_from_tap(const struct ctx *c, int queue,
> uint8_t pif, sa_family_t af,
> const void *saddr, const void *daddr,
> in_port_t srcport, in_port_t dstport,
> diff --git a/udp_vu.c b/udp_vu.c
> index 099677f914e7..cd2c9c516d44 100644
> --- a/udp_vu.c
> +++ b/udp_vu.c
> @@ -202,9 +202,11 @@ static void udp_vu_csum(const struct flowside *toside, int iov_used)
> void udp_vu_sock_to_tap(const struct ctx *c, int s, int n, flow_sidx_t tosidx)
> {
> const struct flowside *toside = flowside_at_sidx(tosidx);
> + const struct udp_flow *uflow = udp_at_sidx(tosidx);
> bool v6 = !(inany_v4(&toside->eaddr) && inany_v4(&toside->oaddr));
> + int rx_queue = flow_rx_virtqueue(&uflow->f);
> struct vu_dev *vdev = c->vdev;
> - struct vu_virtq *vq = &vdev->vq[VHOST_USER_RX_QUEUE];
> + struct vu_virtq *vq = &vdev->vq[rx_queue];
> int i;
>
> for (i = 0; i < n; i++) {
> diff --git a/vu_common.c b/vu_common.c
> index 8904403e66af..56f26317b192 100644
> --- a/vu_common.c
> +++ b/vu_common.c
> @@ -196,11 +196,11 @@ static void vu_handle_tx(struct vu_dev *vdev, int index,
>
> data = IOV_TAIL(elem[count].out_sg, elem[count].out_num, 0);
> if (IOV_DROP_HEADER(&data, struct virtio_net_hdr_mrg_rxbuf))
> - tap_add_packet(vdev->context, &data, now);
> + tap_add_packet(vdev->context, index, &data, now);
>
> count++;
> }
> - tap_handler(vdev->context, now);
> + tap_handler(vdev->context, index, now);
>
> if (count) {
> int i;
> --
> 2.51.0
>
--
David Gibson (he or they) | I'll have my music baroque, and my code
david AT gibson.dropbear.id.au | minimalist, thank you, not the other way
| around.
http://www.ozlabs.org/~dgibson
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]
^ permalink raw reply [flat|nested] 13+ messages in thread
* [PATCH 4/4] test: Add multiqueue support to vhost-user test infrastructure
2025-11-07 14:38 [PATCH 0/4] vhost-user: Add multiqueue support Laurent Vivier
` (2 preceding siblings ...)
2025-11-07 14:39 ` [PATCH 3/4] multiqueue: Add queue-aware flow management for multiqueue support Laurent Vivier
@ 2025-11-07 14:39 ` Laurent Vivier
2025-11-10 5:57 ` David Gibson
2025-11-10 4:40 ` [PATCH 0/4] vhost-user: Add multiqueue support David Gibson
4 siblings, 1 reply; 13+ messages in thread
From: Laurent Vivier @ 2025-11-07 14:39 UTC (permalink / raw)
To: passt-dev; +Cc: Laurent Vivier
Extend test suite to support multiqueue testing with vhost-user
backend. This change:
- Adds VHOST_USER_MQ environment variable to control queue count
- Passes --max-queues parameter to passt when VHOST_USER_MQ > 1
- Configures QEMU with queues= parameter for vhost-user netdev
- Updates setup scripts for passt, passt_in_ns, and two_guests setups
- Adds new test runs with 8 queues for comprehensive multiqueue testing.
Note that the guest kernel will not enable more queues than the number
of vCPUs.
The test infrastructure now validates multiqueue functionality across
all test scenarios including TCP, UDP, ICMP, DHCP, NDP, and performance
tests.
Signed-off-by: Laurent Vivier <lvivier@redhat.com>
---
test/lib/setup | 60 ++++++++++++++++++++++++++++++++++++++------------
test/run | 23 +++++++++++++++++++
2 files changed, 69 insertions(+), 14 deletions(-)
diff --git a/test/lib/setup b/test/lib/setup
index 5994598744a3..69433aa89adc 100755
--- a/test/lib/setup
+++ b/test/lib/setup
@@ -45,7 +45,8 @@ setup_passt() {
[ ${PCAP} -eq 1 ] && __opts="${__opts} -p ${LOGDIR}/passt.pcap"
[ ${DEBUG} -eq 1 ] && __opts="${__opts} -d"
[ ${TRACE} -eq 1 ] && __opts="${__opts} --trace"
- [ ${VHOST_USER} -eq 1 ] && __opts="${__opts} --vhost-user"
+ [ ${VHOST_USER} -eq 1 ] && __opts="${__opts} --vhost-user" && \
+ [ ${VHOST_USER_MQ} -gt 1 ] && __opts="${__opts} --max-queues ${VHOST_USER_MQ}"
context_run passt "make clean"
context_run passt "make valgrind"
@@ -59,10 +60,18 @@ setup_passt() {
__vmem="$(((${__vmem} + 500) / 1000))G"
__qemu_netdev=" \
-chardev socket,id=c,path=${STATESETUP}/passt.socket \
- -netdev vhost-user,id=v,chardev=c \
- -device virtio-net,netdev=v \
-object memory-backend-memfd,id=m,share=on,size=${__vmem} \
-numa node,memdev=m"
+
+ if [ ${VHOST_USER_MQ} -qt 1 ]; then
+ __qemu_netdev="${__qemu_netdev} \
+ -device virtio-net,netdev=v,mq=true \
+ -netdev vhost-user,id=v,chardev=c,queues=${VHOST_USER_MQ}"
+ else
+ __qemu_netdev="${__qemu_netdev} \
+ -device virtio-net,netdev=v \
+ -netdev vhost-user,id=v,chardev=c"
+ fi
else
__qemu_netdev="-device virtio-net-pci,netdev=s \
-netdev stream,id=s,server=off,addr.type=unix,addr.path=${STATESETUP}/passt.socket"
@@ -155,7 +164,8 @@ setup_passt_in_ns() {
[ ${PCAP} -eq 1 ] && __opts="${__opts} -p ${LOGDIR}/passt_in_pasta.pcap"
[ ${DEBUG} -eq 1 ] && __opts="${__opts} -d"
[ ${TRACE} -eq 1 ] && __opts="${__opts} --trace"
- [ ${VHOST_USER} -eq 1 ] && __opts="${__opts} --vhost-user"
+ [ ${VHOST_USER} -eq 1 ] && __opts="${__opts} --vhost-user" && \
+ [ ${VHOST_USER_MQ} -gt 1 ] && __opts="${__opts} --max-queues ${VHOST_USER_MQ}"
if [ ${VALGRIND} -eq 1 ]; then
context_run passt "make clean"
@@ -173,10 +183,18 @@ setup_passt_in_ns() {
__vmem="$(((${__vmem} + 500) / 1000))G"
__qemu_netdev=" \
-chardev socket,id=c,path=${STATESETUP}/passt.socket \
- -netdev vhost-user,id=v,chardev=c \
- -device virtio-net,netdev=v \
-object memory-backend-memfd,id=m,share=on,size=${__vmem} \
-numa node,memdev=m"
+
+ if [ ${VHOST_USER_MQ} -qt 1 ]; then
+ __qemu_netdev="${__qemu_netdev} \
+ -device virtio-net,netdev=v,mq=true \
+ -netdev vhost-user,id=v,chardev=c,queues=${VHOST_USER_MQ}"
+ else
+ __qemu_netdev="${__qemu_netdev} \
+ -device virtio-net,netdev=v \
+ -netdev vhost-user,id=v,chardev=c"
+ fi
else
__qemu_netdev="-device virtio-net-pci,netdev=s \
-netdev stream,id=s,server=off,addr.type=unix,addr.path=${STATESETUP}/passt.socket"
@@ -241,18 +259,20 @@ setup_two_guests() {
[ ${PCAP} -eq 1 ] && __opts="${__opts} -p ${LOGDIR}/passt_1.pcap"
[ ${DEBUG} -eq 1 ] && __opts="${__opts} -d"
[ ${TRACE} -eq 1 ] && __opts="${__opts} --trace"
- [ ${VHOST_USER} -eq 1 ] && __opts="${__opts} --vhost-user"
+ [ ${VHOST_USER} -eq 1 ] && __opts="${__opts} --vhost-user" && \
+ [ ${VHOST_USER_MQ} -gt 1 ] && __opts="${__opts} --max-queues ${VHOST_USER_MQ}"
- context_run_bg passt_1 "./passt -s ${STATESETUP}/passt_1.socket -P ${STATESETUP}/passt_1.pid -f ${__opts} --fqdn fqdn1.passt.test -H hostname1 -t 10001 -u 10001"
+ context_run_bg passt_1 "./passt -s ${STATESETUP}/passt_1.socket -P ${STATESETUP}/passt_1.pid -f ${__opts} --fqdn fqdn1.passt.test -H hostname1 -t 10001 -u 10001 ${__passt_mq}"
wait_for [ -f "${STATESETUP}/passt_1.pid" ]
__opts=
[ ${PCAP} -eq 1 ] && __opts="${__opts} -p ${LOGDIR}/passt_2.pcap"
[ ${DEBUG} -eq 1 ] && __opts="${__opts} -d"
[ ${TRACE} -eq 1 ] && __opts="${__opts} --trace"
- [ ${VHOST_USER} -eq 1 ] && __opts="${__opts} --vhost-user"
+ [ ${VHOST_USER} -eq 1 ] && __opts="${__opts} --vhost-user" && \
+ [ ${VHOST_USER_MQ} -gt 1 ] && __opts="${__opts} --max-queues ${VHOST_USER_MQ}"
- context_run_bg passt_2 "./passt -s ${STATESETUP}/passt_2.socket -P ${STATESETUP}/passt_2.pid -f ${__opts} --hostname hostname2 --fqdn fqdn2 -t 10004 -u 10004"
+ context_run_bg passt_2 "./passt -s ${STATESETUP}/passt_2.socket -P ${STATESETUP}/passt_2.pid -f ${__opts} --hostname hostname2 --fqdn fqdn2 -t 10004 -u 10004 ${__passt_mq}"
wait_for [ -f "${STATESETUP}/passt_2.pid" ]
__vmem="$((${MEM_KIB} / 1024 / 4))"
@@ -260,16 +280,28 @@ setup_two_guests() {
__vmem="$(((${__vmem} + 500) / 1000))G"
__qemu_netdev1=" \
-chardev socket,id=c,path=${STATESETUP}/passt_1.socket \
- -netdev vhost-user,id=v,chardev=c \
- -device virtio-net,netdev=v \
-object memory-backend-memfd,id=m,share=on,size=${__vmem} \
-numa node,memdev=m"
__qemu_netdev2=" \
-chardev socket,id=c,path=${STATESETUP}/passt_2.socket \
- -netdev vhost-user,id=v,chardev=c \
- -device virtio-net,netdev=v \
-object memory-backend-memfd,id=m,share=on,size=${__vmem} \
-numa node,memdev=m"
+
+ if [ ${VHOST_USER_MQ} -qt 1 ]; then
+ __qemu_netdev1="${__qemu_netdev1} \
+ -device virtio-net,netdev=v,mq=true \
+ -netdev vhost-user,id=v,chardev=c,queues=${VHOST_USER_MQ}"
+ __qemu_netdev2="${__qemu_netdev2} \
+ -device virtio-net,netdev=v,mq=true \
+ -netdev vhost-user,id=v,chardev=c,queues=${VHOST_USER_MQ}"
+ else
+ __qemu_netdev1="${__qemu_netdev1} \
+ -device virtio-net,netdev=v \
+ -netdev vhost-user,id=v,chardev=c"
+ __qemu_netdev2="${__qemu_netdev2} \
+ -device virtio-net,netdev=v \
+ -netdev vhost-user,id=v,chardev=c"
+ fi
else
__qemu_netdev1="-device virtio-net-pci,netdev=s \
-netdev stream,id=s,server=off,addr.type=unix,addr.path=${STATESETUP}/passt_1.socket"
diff --git a/test/run b/test/run
index f858e5586847..652cc12b1234 100755
--- a/test/run
+++ b/test/run
@@ -190,6 +190,29 @@ run() {
test passt_vu_in_ns/shutdown
teardown passt_in_ns
+ VHOST_USER=1
+ VHOST_USER_MQ=8
+ setup passt_in_ns
+ test passt_vu/ndp
+ test passt_vu_in_ns/dhcp
+ test passt_vu_in_ns/icmp
+ test passt_vu_in_ns/tcp
+ test passt_vu_in_ns/udp
+ test passt_vu_in_ns/shutdown
+ teardown passt_in_ns
+
+ setup two_guests
+ test two_guests_vu/basic
+ teardown two_guests
+
+ setup passt_in_ns
+ test passt_vu/ndp
+ test passt_vu_in_ns/dhcp
+ test perf/passt_vu_tcp
+ test perf/passt_vu_udp
+ test passt_vu_in_ns/shutdown
+ teardown passt_in_ns
+
# TODO: Make those faster by at least pre-installing gcc and make on
# non-x86 images, then re-enable.
skip_distro() {
--
2.51.0
^ permalink raw reply [flat|nested] 13+ messages in thread* Re: [PATCH 4/4] test: Add multiqueue support to vhost-user test infrastructure
2025-11-07 14:39 ` [PATCH 4/4] test: Add multiqueue support to vhost-user test infrastructure Laurent Vivier
@ 2025-11-10 5:57 ` David Gibson
0 siblings, 0 replies; 13+ messages in thread
From: David Gibson @ 2025-11-10 5:57 UTC (permalink / raw)
To: Laurent Vivier; +Cc: passt-dev
[-- Attachment #1: Type: text/plain, Size: 8521 bytes --]
On Fri, Nov 07, 2025 at 03:39:01PM +0100, Laurent Vivier wrote:
> Extend test suite to support multiqueue testing with vhost-user
> backend. This change:
>
> - Adds VHOST_USER_MQ environment variable to control queue count
> - Passes --max-queues parameter to passt when VHOST_USER_MQ > 1
> - Configures QEMU with queues= parameter for vhost-user netdev
> - Updates setup scripts for passt, passt_in_ns, and two_guests setups
> - Adds new test runs with 8 queues for comprehensive multiqueue testing.
This is a good start, but it's definitely not "comprehensive" (our
existing testsuite is also far from comprehensive). That's not Claude
putting words in your mouth, is it?
> Note that the guest kernel will not enable more queues than the number
> of vCPUs.
>
> The test infrastructure now validates multiqueue functionality across
> all test scenarios including TCP, UDP, ICMP, DHCP, NDP, and performance
> tests.
>
> Signed-off-by: Laurent Vivier <lvivier@redhat.com>
> ---
> test/lib/setup | 60 ++++++++++++++++++++++++++++++++++++++------------
> test/run | 23 +++++++++++++++++++
> 2 files changed, 69 insertions(+), 14 deletions(-)
>
> diff --git a/test/lib/setup b/test/lib/setup
> index 5994598744a3..69433aa89adc 100755
> --- a/test/lib/setup
> +++ b/test/lib/setup
> @@ -45,7 +45,8 @@ setup_passt() {
> [ ${PCAP} -eq 1 ] && __opts="${__opts} -p ${LOGDIR}/passt.pcap"
> [ ${DEBUG} -eq 1 ] && __opts="${__opts} -d"
> [ ${TRACE} -eq 1 ] && __opts="${__opts} --trace"
> - [ ${VHOST_USER} -eq 1 ] && __opts="${__opts} --vhost-user"
> + [ ${VHOST_USER} -eq 1 ] && __opts="${__opts} --vhost-user" && \
> + [ ${VHOST_USER_MQ} -gt 1 ] && __opts="${__opts} --max-queues ${VHOST_USER_MQ}"
>
> context_run passt "make clean"
> context_run passt "make valgrind"
> @@ -59,10 +60,18 @@ setup_passt() {
> __vmem="$(((${__vmem} + 500) / 1000))G"
> __qemu_netdev=" \
> -chardev socket,id=c,path=${STATESETUP}/passt.socket \
> - -netdev vhost-user,id=v,chardev=c \
> - -device virtio-net,netdev=v \
> -object memory-backend-memfd,id=m,share=on,size=${__vmem} \
> -numa node,memdev=m"
> +
> + if [ ${VHOST_USER_MQ} -qt 1 ]; then
> + __qemu_netdev="${__qemu_netdev} \
> + -device virtio-net,netdev=v,mq=true \
> + -netdev vhost-user,id=v,chardev=c,queues=${VHOST_USER_MQ}"
> + else
> + __qemu_netdev="${__qemu_netdev} \
> + -device virtio-net,netdev=v \
> + -netdev vhost-user,id=v,chardev=c"
> + fi
> else
> __qemu_netdev="-device virtio-net-pci,netdev=s \
> -netdev stream,id=s,server=off,addr.type=unix,addr.path=${STATESETUP}/passt.socket"
> @@ -155,7 +164,8 @@ setup_passt_in_ns() {
> [ ${PCAP} -eq 1 ] && __opts="${__opts} -p ${LOGDIR}/passt_in_pasta.pcap"
> [ ${DEBUG} -eq 1 ] && __opts="${__opts} -d"
> [ ${TRACE} -eq 1 ] && __opts="${__opts} --trace"
> - [ ${VHOST_USER} -eq 1 ] && __opts="${__opts} --vhost-user"
> + [ ${VHOST_USER} -eq 1 ] && __opts="${__opts} --vhost-user" && \
> + [ ${VHOST_USER_MQ} -gt 1 ] && __opts="${__opts} --max-queues ${VHOST_USER_MQ}"
>
> if [ ${VALGRIND} -eq 1 ]; then
> context_run passt "make clean"
> @@ -173,10 +183,18 @@ setup_passt_in_ns() {
> __vmem="$(((${__vmem} + 500) / 1000))G"
> __qemu_netdev=" \
> -chardev socket,id=c,path=${STATESETUP}/passt.socket \
> - -netdev vhost-user,id=v,chardev=c \
> - -device virtio-net,netdev=v \
> -object memory-backend-memfd,id=m,share=on,size=${__vmem} \
> -numa node,memdev=m"
> +
> + if [ ${VHOST_USER_MQ} -qt 1 ]; then
> + __qemu_netdev="${__qemu_netdev} \
> + -device virtio-net,netdev=v,mq=true \
> + -netdev vhost-user,id=v,chardev=c,queues=${VHOST_USER_MQ}"
> + else
> + __qemu_netdev="${__qemu_netdev} \
> + -device virtio-net,netdev=v \
> + -netdev vhost-user,id=v,chardev=c"
> + fi
> else
> __qemu_netdev="-device virtio-net-pci,netdev=s \
> -netdev stream,id=s,server=off,addr.type=unix,addr.path=${STATESETUP}/passt.socket"
> @@ -241,18 +259,20 @@ setup_two_guests() {
> [ ${PCAP} -eq 1 ] && __opts="${__opts} -p ${LOGDIR}/passt_1.pcap"
> [ ${DEBUG} -eq 1 ] && __opts="${__opts} -d"
> [ ${TRACE} -eq 1 ] && __opts="${__opts} --trace"
> - [ ${VHOST_USER} -eq 1 ] && __opts="${__opts} --vhost-user"
> + [ ${VHOST_USER} -eq 1 ] && __opts="${__opts} --vhost-user" && \
> + [ ${VHOST_USER_MQ} -gt 1 ] && __opts="${__opts} --max-queues ${VHOST_USER_MQ}"
>
> - context_run_bg passt_1 "./passt -s ${STATESETUP}/passt_1.socket -P ${STATESETUP}/passt_1.pid -f ${__opts} --fqdn fqdn1.passt.test -H hostname1 -t 10001 -u 10001"
> + context_run_bg passt_1 "./passt -s ${STATESETUP}/passt_1.socket -P ${STATESETUP}/passt_1.pid -f ${__opts} --fqdn fqdn1.passt.test -H hostname1 -t 10001 -u 10001 ${__passt_mq}"
> wait_for [ -f "${STATESETUP}/passt_1.pid" ]
>
> __opts=
> [ ${PCAP} -eq 1 ] && __opts="${__opts} -p ${LOGDIR}/passt_2.pcap"
> [ ${DEBUG} -eq 1 ] && __opts="${__opts} -d"
> [ ${TRACE} -eq 1 ] && __opts="${__opts} --trace"
> - [ ${VHOST_USER} -eq 1 ] && __opts="${__opts} --vhost-user"
> + [ ${VHOST_USER} -eq 1 ] && __opts="${__opts} --vhost-user" && \
> + [ ${VHOST_USER_MQ} -gt 1 ] && __opts="${__opts} --max-queues ${VHOST_USER_MQ}"
>
> - context_run_bg passt_2 "./passt -s ${STATESETUP}/passt_2.socket -P ${STATESETUP}/passt_2.pid -f ${__opts} --hostname hostname2 --fqdn fqdn2 -t 10004 -u 10004"
> + context_run_bg passt_2 "./passt -s ${STATESETUP}/passt_2.socket -P ${STATESETUP}/passt_2.pid -f ${__opts} --hostname hostname2 --fqdn fqdn2 -t 10004 -u 10004 ${__passt_mq}"
> wait_for [ -f "${STATESETUP}/passt_2.pid" ]
>
> __vmem="$((${MEM_KIB} / 1024 / 4))"
> @@ -260,16 +280,28 @@ setup_two_guests() {
> __vmem="$(((${__vmem} + 500) / 1000))G"
> __qemu_netdev1=" \
> -chardev socket,id=c,path=${STATESETUP}/passt_1.socket \
> - -netdev vhost-user,id=v,chardev=c \
> - -device virtio-net,netdev=v \
> -object memory-backend-memfd,id=m,share=on,size=${__vmem} \
> -numa node,memdev=m"
> __qemu_netdev2=" \
> -chardev socket,id=c,path=${STATESETUP}/passt_2.socket \
> - -netdev vhost-user,id=v,chardev=c \
> - -device virtio-net,netdev=v \
> -object memory-backend-memfd,id=m,share=on,size=${__vmem} \
> -numa node,memdev=m"
> +
> + if [ ${VHOST_USER_MQ} -qt 1 ]; then
> + __qemu_netdev1="${__qemu_netdev1} \
> + -device virtio-net,netdev=v,mq=true \
> + -netdev vhost-user,id=v,chardev=c,queues=${VHOST_USER_MQ}"
> + __qemu_netdev2="${__qemu_netdev2} \
> + -device virtio-net,netdev=v,mq=true \
> + -netdev vhost-user,id=v,chardev=c,queues=${VHOST_USER_MQ}"
> + else
> + __qemu_netdev1="${__qemu_netdev1} \
> + -device virtio-net,netdev=v \
> + -netdev vhost-user,id=v,chardev=c"
> + __qemu_netdev2="${__qemu_netdev2} \
> + -device virtio-net,netdev=v \
> + -netdev vhost-user,id=v,chardev=c"
> + fi
> else
> __qemu_netdev1="-device virtio-net-pci,netdev=s \
> -netdev stream,id=s,server=off,addr.type=unix,addr.path=${STATESETUP}/passt_1.socket"
> diff --git a/test/run b/test/run
> index f858e5586847..652cc12b1234 100755
> --- a/test/run
> +++ b/test/run
> @@ -190,6 +190,29 @@ run() {
> test passt_vu_in_ns/shutdown
> teardown passt_in_ns
>
> + VHOST_USER=1
> + VHOST_USER_MQ=8
> + setup passt_in_ns
> + test passt_vu/ndp
> + test passt_vu_in_ns/dhcp
> + test passt_vu_in_ns/icmp
> + test passt_vu_in_ns/tcp
> + test passt_vu_in_ns/udp
> + test passt_vu_in_ns/shutdown
> + teardown passt_in_ns
> +
> + setup two_guests
> + test two_guests_vu/basic
> + teardown two_guests
> +
> + setup passt_in_ns
> + test passt_vu/ndp
> + test passt_vu_in_ns/dhcp
> + test perf/passt_vu_tcp
> + test perf/passt_vu_udp
> + test passt_vu_in_ns/shutdown
> + teardown passt_in_ns
> +
> # TODO: Make those faster by at least pre-installing gcc and make on
> # non-x86 images, then re-enable.
> skip_distro() {
> --
> 2.51.0
>
--
David Gibson (he or they) | I'll have my music baroque, and my code
david AT gibson.dropbear.id.au | minimalist, thank you, not the other way
| around.
http://www.ozlabs.org/~dgibson
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]
^ permalink raw reply [flat|nested] 13+ messages in thread
* Re: [PATCH 0/4] vhost-user: Add multiqueue support
2025-11-07 14:38 [PATCH 0/4] vhost-user: Add multiqueue support Laurent Vivier
` (3 preceding siblings ...)
2025-11-07 14:39 ` [PATCH 4/4] test: Add multiqueue support to vhost-user test infrastructure Laurent Vivier
@ 2025-11-10 4:40 ` David Gibson
2025-11-10 6:00 ` David Gibson
4 siblings, 1 reply; 13+ messages in thread
From: David Gibson @ 2025-11-10 4:40 UTC (permalink / raw)
To: Laurent Vivier; +Cc: passt-dev
[-- Attachment #1: Type: text/plain, Size: 3658 bytes --]
On Fri, Nov 07, 2025 at 03:38:57PM +0100, Laurent Vivier wrote:
> This series implements multiqueue support for vhost-user mode, allowing passt
> to utilize multiple queue pairs. There is no improved network performance because
> we keep using only one thread.
>
> The implementation introduces a --max-queues parameter to configure up to 16
> queue pairs (32 virtqueues) in vhost-user mode. Packets are routed to the
> appropriate RX queue based on which TX queue they originated from, enabling the
> guest kernel to distribute network traffic across multiple queues
> and vCPUs.
I find this description a little bit confusing, since at the packet
(L2) level things going to an Rx queue don't really have an
originating Tx queue. I assume you mean that each flow gets
associated to a single queue pair, so both Rx and Tx packets on that
flow will use the same queue pair (and, later, thread). Is that
correct? If so I think it would be clearer to explicitly describe
this in terms of flows.
> This series adds:
> - configuration support for multiqueue via --max-queues parameter
> - queue parameter threading throughout the network stack - a significant
> refactoring that propagates queue information through all protocol handlers
> (TCP, UDP, ICMP, ARP, DHCP, DHCPv6, NDP)
> - flow-aware queue routing that matches RX queue selection to the incoming
> TX queue, maintaining proper packet affinity
> - comprehensive test coverage with VHOST_USER_MQ environment variable to
> validate multiqueue functionality across all test scenarios
>
> Current behavior: TX queue selection is controlled by the guest kernel, while
> RX packets are routed to queues based on their associated flows. Host-initiated
> flows currently default to queue 0.
>
> The RX queue of a flow is updated on each new packet from the TX queue to
> maintain affinity.
>
> The changes maintain backward compatibility - without --max-queues, behavior
> remains unchanged with single-queue operation.
>
> Laurent Vivier (4):
> vhost-user: Enable multiqueue
> vhost-user: Add queue parameter throughout the network stack
> multiqueue: Add queue-aware flow management for multiqueue support
> test: Add multiqueue support to vhost-user test infrastructure
>
> arp.c | 12 ++---
> arp.h | 4 +-
> conf.c | 31 ++++++++++++-
> dhcp.c | 5 ++-
> dhcp.h | 2 +-
> dhcpv6.c | 12 ++---
> dhcpv6.h | 2 +-
> flow.c | 32 +++++++++++++
> flow.h | 10 +++++
> fwd.c | 4 +-
> icmp.c | 25 ++++++-----
> icmp.h | 2 +-
> ndp.c | 32 +++++++------
> ndp.h | 5 ++-
> passt.h | 2 +
> tap.c | 120 +++++++++++++++++++++++++++++--------------------
> tap.h | 18 ++++----
> tcp.c | 82 ++++++++++++++++++---------------
> tcp.h | 11 ++---
> tcp_vu.c | 8 ++--
> test/lib/setup | 60 +++++++++++++++++++------
> test/run | 23 ++++++++++
> udp.c | 40 ++++++++++-------
> udp.h | 12 ++---
> udp_flow.c | 8 +++-
> udp_flow.h | 2 +-
> udp_vu.c | 4 +-
> vhost_user.c | 38 +++++++++-------
> virtio.h | 2 +-
> vu_common.c | 13 +++---
> vu_common.h | 3 +-
> 31 files changed, 416 insertions(+), 208 deletions(-)
>
> --
> 2.51.0
>
>
--
David Gibson (he or they) | I'll have my music baroque, and my code
david AT gibson.dropbear.id.au | minimalist, thank you, not the other way
| around.
http://www.ozlabs.org/~dgibson
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]
^ permalink raw reply [flat|nested] 13+ messages in thread* Re: [PATCH 0/4] vhost-user: Add multiqueue support
2025-11-10 4:40 ` [PATCH 0/4] vhost-user: Add multiqueue support David Gibson
@ 2025-11-10 6:00 ` David Gibson
0 siblings, 0 replies; 13+ messages in thread
From: David Gibson @ 2025-11-10 6:00 UTC (permalink / raw)
To: Laurent Vivier; +Cc: passt-dev
[-- Attachment #1: Type: text/plain, Size: 2900 bytes --]
On Mon, Nov 10, 2025 at 03:40:25PM +1100, David Gibson wrote:
> On Fri, Nov 07, 2025 at 03:38:57PM +0100, Laurent Vivier wrote:
> > This series implements multiqueue support for vhost-user mode, allowing passt
> > to utilize multiple queue pairs. There is no improved network performance because
> > we keep using only one thread.
> >
> > The implementation introduces a --max-queues parameter to configure up to 16
> > queue pairs (32 virtqueues) in vhost-user mode. Packets are routed to the
> > appropriate RX queue based on which TX queue they originated from, enabling the
> > guest kernel to distribute network traffic across multiple queues
> > and vCPUs.
>
> I find this description a little bit confusing, since at the packet
> (L2) level things going to an Rx queue don't really have an
> originating Tx queue. I assume you mean that each flow gets
> associated to a single queue pair, so both Rx and Tx packets on that
> flow will use the same queue pair (and, later, thread). Is that
> correct? If so I think it would be clearer to explicitly describe
> this in terms of flows.
>
> > This series adds:
> > - configuration support for multiqueue via --max-queues parameter
> > - queue parameter threading throughout the network stack - a significant
> > refactoring that propagates queue information through all protocol handlers
> > (TCP, UDP, ICMP, ARP, DHCP, DHCPv6, NDP)
> > - flow-aware queue routing that matches RX queue selection to the incoming
> > TX queue, maintaining proper packet affinity
> > - comprehensive test coverage with VHOST_USER_MQ environment variable to
> > validate multiqueue functionality across all test scenarios
> >
> > Current behavior: TX queue selection is controlled by the guest kernel, while
> > RX packets are routed to queues based on their associated flows. Host-initiated
> > flows currently default to queue 0.
> >
> > The RX queue of a flow is updated on each new packet from the TX queue to
> > maintain affinity.
> >
> > The changes maintain backward compatibility - without --max-queues, behavior
> > remains unchanged with single-queue operation.
Having now read the whole series. I noted on an individual patch that
it's sometimes confusing what's being communicated as queue pair ID,
what as absolute queue ID. Perhaps that discussion belongs better not
attached to a specific patch.
I think we should standardize on a single way of indexing, in as much
of the code as possible - the exception should be in the low-level
code that actually deals with the vq interfaces. I _think_ queue pair
will simplify things compared to absolute queue number, but I might be
wrong.
--
David Gibson (he or they) | I'll have my music baroque, and my code
david AT gibson.dropbear.id.au | minimalist, thank you, not the other way
| around.
http://www.ozlabs.org/~dgibson
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]
^ permalink raw reply [flat|nested] 13+ messages in thread