Skip to contentSkip to conversation

vhost-user multiple queue support

PATCH Superseded Yuanhan Liu 5 participants 31 messages Sep 8, 2015 – Sep 15, 2015 Networking

v7 of 12 Superseded Latest is v12 →

lore
Share & tools
lore
#0 lore raw

v7 cover vhost-user multiple queue support

Hi,

Here is the updated patch set for enabling vhost-user multiple queue.

This patch set introduces 2 more vhost user messages: VHOST_USER_GET_QUEUE_NUM, for querying how many queues the backend supports, and VHOST_USER_SET_VRING_FLAG, for enabling/disabling a specific virt queue.

Both of th two new messages are treated as vhost protocol extension, and that's why Michaels's patch "vhost-user: add protocol feature negotiation" is also included here.

Patch 1-5 are all prepare work for actually enabling multiple queue.

Patch 6 is the key patch for enabling multiple queue, which address two major concerns from Michael: no feedback from backend if it can't support # of requested queues, and all messages are sent N time. It also fixes an hidden bug.

Patch 7 introduces the VHOST_USER_SET_VRING_FLAG message.

Note that I haven't done any formal test yet, it just passes build test and basic functional test, such as it does exit when backend doesn't support # of requested queues. Here I sent it out just for more comments, and for avoiding spending too much effort on the wrong track.

Thanks.

    --yliu

---
Changchun Ouyang (1):
vhost-user: add a new message to disable/enable a specific virt queue.

Michael S. Tsirkin (1):
vhost-user: add protocol feature negotiation

Ouyang Changchun (1):
vhost-user: add multiple queue support

Yuanhan Liu (4):
vhost-user: use VHOST_USER_XXX macro for switch statement
vhost: rename VHOST_RESET_OWNER to VHOST_RESET_DEVICE
vhost-user: add VHOST_USER_GET_QUEUE_NUM message
vhost_net: move vhost_net_set_vq_index ahead at vhost_net_init

 docs/specs/vhost-user.txt         |  75 +++++++++++++++++++-
 hw/net/vhost_net.c                |  45 +++++++++---
 hw/net/virtio-net.c               |   2 +
 hw/virtio/vhost-user.c            | 140 ++++++++++++++++++++++++++++++++------
 include/hw/virtio/vhost-backend.h |   2 +
 include/hw/virtio/vhost.h         |   2 +
 include/net/net.h                 |   1 +
 include/net/vhost_net.h           |   2 +
 linux-headers/linux/vhost.h       |   2 +-
 net/vhost-user.c                  | 135 +++++++++++++++++++++++++-----------
 qapi-schema.json                  |   6 +-
 qemu-options.hx                   |   5 +-
 tests/vhost-user-test.c           |   2 +-
 13 files changed, 340 insertions(+), 79 deletions(-)
Signature · 1 line
1.9.0
#1 lore raw

PATCH v7 1/7 vhost-user: use VHOST_USER_XXX macro for switch statement

So that we could let vhost_user_call to handle extented requests, such as VHOST_USER_GET/SET_PROTOCOL_FEATURES, instead of invoking vhost_user_read/write and constructing the msg again by ourself.

---
 hw/virtio/vhost-user.c | 37 +++++++++++++++++++++----------------
 1 file changed, 21 insertions(+), 16 deletions(-)
hw/virtio/vhost-user.c +21 -16
index e7ab829..33d1b00 100644
@@ -193,27 +193,32 @@ static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
 
     assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
 
-    msg_request = vhost_user_request_translate(request);
+    /* only translate vhost ioctl requests */
+    if (request > VHOST_USER_MAX)
+        msg_request = vhost_user_request_translate(request);
+    else
+        msg_request = request;
+
     msg.request = msg_request;
     msg.flags = VHOST_USER_VERSION;
     msg.size = 0;
 
-    switch (request) {
-    case VHOST_GET_FEATURES:
+    switch (msg_request) {
+    case VHOST_USER_GET_FEATURES:
         need_reply = 1;
         break;
 
-    case VHOST_SET_FEATURES:
-    case VHOST_SET_LOG_BASE:
+    case VHOST_USER_SET_FEATURES:
+    case VHOST_USER_SET_LOG_BASE:
         msg.u64 = *((__u64 *) arg);
         msg.size = sizeof(m.u64);
         break;
 
-    case VHOST_SET_OWNER:
-    case VHOST_RESET_OWNER:
+    case VHOST_USER_SET_OWNER:
+    case VHOST_USER_RESET_OWNER:
         break;
 
-    case VHOST_SET_MEM_TABLE:
+    case VHOST_USER_SET_MEM_TABLE:
         for (i = 0; i < dev->mem->nregions; ++i) {
             struct vhost_memory_region *reg = dev->mem->regions + i;
             ram_addr_t ram_addr;
@@ -246,30 +251,30 @@ static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
 
         break;
 
-    case VHOST_SET_LOG_FD:
+    case VHOST_USER_SET_LOG_FD:
         fds[fd_num++] = *((int *) arg);
         break;
 
-    case VHOST_SET_VRING_NUM:
-    case VHOST_SET_VRING_BASE:
+    case VHOST_USER_SET_VRING_NUM:
+    case VHOST_USER_SET_VRING_BASE:
         memcpy(&msg.state, arg, sizeof(struct vhost_vring_state));
         msg.size = sizeof(m.state);
         break;
 
-    case VHOST_GET_VRING_BASE:
+    case VHOST_USER_GET_VRING_BASE:
         memcpy(&msg.state, arg, sizeof(struct vhost_vring_state));
         msg.size = sizeof(m.state);
         need_reply = 1;
         break;
 
-    case VHOST_SET_VRING_ADDR:
+    case VHOST_USER_SET_VRING_ADDR:
         memcpy(&msg.addr, arg, sizeof(struct vhost_vring_addr));
         msg.size = sizeof(m.addr);
         break;
 
-    case VHOST_SET_VRING_KICK:
-    case VHOST_SET_VRING_CALL:
-    case VHOST_SET_VRING_ERR:
+    case VHOST_USER_SET_VRING_KICK:
+    case VHOST_USER_SET_VRING_CALL:
+    case VHOST_USER_SET_VRING_ERR:
         file = arg;
         msg.u64 = file->index & VHOST_USER_VRING_IDX_MASK;
         msg.size = sizeof(m.u64);
Signature · 1 line
1.9.0
#2 lore raw

PATCH v7 2/7 vhost-user: add protocol feature negotiation

In reply to Yuanhan Liu

From: "Michael S. Tsirkin" <[email protected]>

Support a separate bitmask for vhost-user protocol features, and messages to get/set protocol features.

Invoke them at init.

No features are defined yet.

v2: leverage vhost_user_call for request handling -- Yuanhan Liu

---
 docs/specs/vhost-user.txt | 37 +++++++++++++++++++++++++++++++++++++
 hw/net/vhost_net.c        |  2 ++
 hw/virtio/vhost-user.c    | 28 ++++++++++++++++++++++++++++
 include/hw/virtio/vhost.h |  1 +
 4 files changed, 68 insertions(+)
docs/specs/vhost-user.txt +37 -0
index 650bb18..70da3b1 100644
@@ -113,6 +113,7 @@ message replies. Most of the requests don't require replies. Here is a list of
 the ones that do:
 
  * VHOST_GET_FEATURES
+ * VHOST_GET_PROTOCOL_FEATURES
  * VHOST_GET_VRING_BASE
 
 There are several messages that the master sends with file descriptors passed
@@ -127,6 +128,13 @@ in the ancillary data:
 If Master is unable to send the full message or receives a wrong reply it will
 close the connection. An optional reconnection mechanism can be implemented.
 
+Any protocol extensions are gated by protocol feature bits,
+which allows full backwards compatibility on both master
+and slave.
+As older slaves don't support negotiating protocol features,
+a feature bit was dedicated for this purpose:
+#define VHOST_USER_F_PROTOCOL_FEATURES 30
+
 Message types
 -------------
 
@@ -138,6 +146,8 @@ Message types
       Slave payload: u64
 
       Get from the underlying vhost implementation the features bitmask.
+      Feature bit VHOST_USER_F_PROTOCOL_FEATURES signals slave support for
+      VHOST_USER_GET_PROTOCOL_FEATURES and VHOST_USER_SET_PROTOCOL_FEATURES.
 
  * VHOST_USER_SET_FEATURES
 
@@ -146,6 +156,33 @@ Message types
       Master payload: u64
 
       Enable features in the underlying vhost implementation using a bitmask.
+      Feature bit VHOST_USER_F_PROTOCOL_FEATURES signals slave support for
+      VHOST_USER_GET_PROTOCOL_FEATURES and VHOST_USER_SET_PROTOCOL_FEATURES.
+
+ * VHOST_USER_GET_PROTOCOL_FEATURES
+
+      Id: 15
+      Equivalent ioctl: VHOST_GET_FEATURES
+      Master payload: N/A
+      Slave payload: u64
+
+      Get the protocol feature bitmask from the underlying vhost implementation.
+      Only legal if feature bit VHOST_USER_F_PROTOCOL_FEATURES is present in
+      VHOST_USER_GET_FEATURES.
+      Note: slave that reported VHOST_USER_F_PROTOCOL_FEATURES must support
+      this message even before VHOST_USER_SET_FEATURES was called.
+
+ * VHOST_USER_SET_PROTOCOL_FEATURES
+
+      Id: 16
+      Ioctl: VHOST_SET_FEATURES
+      Master payload: u64
+
+      Enable protocol features in the underlying vhost implementation.
+      Only legal if feature bit VHOST_USER_F_PROTOCOL_FEATURES is present in
+      VHOST_USER_GET_FEATURES.
+      Note: slave that reported VHOST_USER_F_PROTOCOL_FEATURES must support
+      this message even before VHOST_USER_SET_FEATURES was called.
 
  * VHOST_USER_SET_OWNER
 
hw/net/vhost_net.c +2 -0
index 5c1d11f..c864237 100644
@@ -152,8 +152,10 @@ struct vhost_net *vhost_net_init(VhostNetOptions *options)
         net->dev.backend_features = qemu_has_vnet_hdr(options->net_backend)
             ? 0 : (1ULL << VHOST_NET_F_VIRTIO_NET_HDR);
         net->backend = r;
+        net->dev.protocol_features = 0;
     } else {
         net->dev.backend_features = 0;
+        net->dev.protocol_features = 0;
         net->backend = -1;
     }
     net->nc = options->net_backend;
hw/virtio/vhost-user.c +28 -0
index 33d1b00..b4d8952 100644
@@ -24,6 +24,8 @@
 #include <linux/vhost.h>
 
 #define VHOST_MEMORY_MAX_NREGIONS    8
+#define VHOST_USER_F_PROTOCOL_FEATURES 30
+#define VHOST_USER_PROTOCOL_FEATURE_MASK 0x0ULL
 
 typedef enum VhostUserRequest {
     VHOST_USER_NONE = 0,
@@ -41,6 +43,8 @@ typedef enum VhostUserRequest {
     VHOST_USER_SET_VRING_KICK = 12,
     VHOST_USER_SET_VRING_CALL = 13,
     VHOST_USER_SET_VRING_ERR = 14,
+    VHOST_USER_GET_PROTOCOL_FEATURES = 15,
+    VHOST_USER_SET_PROTOCOL_FEATURES = 16,
     VHOST_USER_MAX
 } VhostUserRequest;
 
@@ -205,11 +209,13 @@ static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
 
     switch (msg_request) {
     case VHOST_USER_GET_FEATURES:
+    case VHOST_USER_GET_PROTOCOL_FEATURES:
         need_reply = 1;
         break;
 
     case VHOST_USER_SET_FEATURES:
     case VHOST_USER_SET_LOG_BASE:
+    case VHOST_USER_SET_PROTOCOL_FEATURES:
         msg.u64 = *((__u64 *) arg);
         msg.size = sizeof(m.u64);
         break;
@@ -307,6 +313,7 @@ static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
 
         switch (msg_request) {
         case VHOST_USER_GET_FEATURES:
+        case VHOST_USER_GET_PROTOCOL_FEATURES:
             if (msg.size != sizeof(m.u64)) {
                 error_report("Received bad msg size.");
                 return -1;
@@ -332,10 +339,31 @@ static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
 
 static int vhost_user_init(struct vhost_dev *dev, void *opaque)
 {
+    unsigned long long features;
+    int err;
+
     assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
 
     dev->opaque = opaque;
 
+    err = vhost_user_call(dev, VHOST_USER_GET_FEATURES, &features);
+    if (err < 0)
+        return err;
+
+    if (__virtio_has_feature(features, VHOST_USER_F_PROTOCOL_FEATURES)) {
+        dev->backend_features |= 1ULL << VHOST_USER_F_PROTOCOL_FEATURES;
+
+        err = vhost_user_call(dev, VHOST_USER_GET_PROTOCOL_FEATURES, &features);
+        if (err < 0)
+            return err;
+
+        dev->protocol_features = features & VHOST_USER_PROTOCOL_FEATURE_MASK;
+        err = vhost_user_call(dev, VHOST_USER_SET_PROTOCOL_FEATURES,
+                              &dev->protocol_features);
+        if (err < 0)
+            return err;
+    }
+
     return 0;
 }
 
include/hw/virtio/vhost.h +1 -0
index dd51050..6467c73 100644
@@ -47,6 +47,7 @@ struct vhost_dev {
     unsigned long long features;
     unsigned long long acked_features;
     unsigned long long backend_features;
+    unsigned long long protocol_features;
     bool started;
     bool log_enabled;
     unsigned long long log_size;
Signature · 1 line
1.9.0
#3 lore raw

PATCH v7 3/7 vhost: rename VHOST_RESET_OWNER to VHOST_RESET_DEVICE

In reply to Yuanhan Liu

Quote from Michael:

We really should rename VHOST_RESET_OWNER to VHOST_RESET_DEVICE.

---
 docs/specs/vhost-user.txt   | 4 ++--
 hw/net/vhost_net.c          | 2 +-
 hw/virtio/vhost-user.c      | 6 +++---
 linux-headers/linux/vhost.h | 2 +-
 tests/vhost-user-test.c     | 2 +-
 5 files changed, 8 insertions(+), 8 deletions(-)
docs/specs/vhost-user.txt +2 -2
index 70da3b1..ccbbcbb 100644
@@ -194,10 +194,10 @@ Message types
       as an owner of the session. This can be used on the Slave as a
       "session start" flag.
 
- * VHOST_USER_RESET_OWNER
+ * VHOST_USER_RESET_DEVICE
 
       Id: 4
-      Equivalent ioctl: VHOST_RESET_OWNER
+      Equivalent ioctl: VHOST_RESET_DEVICE
       Master payload: N/A
 
       Issued when a new connection is about to be closed. The Master will no
hw/net/vhost_net.c +1 -1
index c864237..33b0e97 100644
@@ -287,7 +287,7 @@ static void vhost_net_stop_one(struct vhost_net *net,
     } else if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_VHOST_USER) {
         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
             const VhostOps *vhost_ops = net->dev.vhost_ops;
-            int r = vhost_ops->vhost_call(&net->dev, VHOST_RESET_OWNER,
+            int r = vhost_ops->vhost_call(&net->dev, VHOST_RESET_DEVICE,
                                           NULL);
             assert(r >= 0);
         }
hw/virtio/vhost-user.c +3 -3
index b4d8952..192ea4e 100644
@@ -32,7 +32,7 @@ typedef enum VhostUserRequest {
     VHOST_USER_GET_FEATURES = 1,
     VHOST_USER_SET_FEATURES = 2,
     VHOST_USER_SET_OWNER = 3,
-    VHOST_USER_RESET_OWNER = 4,
+    VHOST_USER_RESET_DEVICE = 4,
     VHOST_USER_SET_MEM_TABLE = 5,
     VHOST_USER_SET_LOG_BASE = 6,
     VHOST_USER_SET_LOG_FD = 7,
@@ -98,7 +98,7 @@ static unsigned long int ioctl_to_vhost_user_request[VHOST_USER_MAX] = {
     VHOST_GET_FEATURES,     /* VHOST_USER_GET_FEATURES */
     VHOST_SET_FEATURES,     /* VHOST_USER_SET_FEATURES */
     VHOST_SET_OWNER,        /* VHOST_USER_SET_OWNER */
-    VHOST_RESET_OWNER,      /* VHOST_USER_RESET_OWNER */
+    VHOST_RESET_DEVICE,      /* VHOST_USER_RESET_DEVICE */
     VHOST_SET_MEM_TABLE,    /* VHOST_USER_SET_MEM_TABLE */
     VHOST_SET_LOG_BASE,     /* VHOST_USER_SET_LOG_BASE */
     VHOST_SET_LOG_FD,       /* VHOST_USER_SET_LOG_FD */
@@ -221,7 +221,7 @@ static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
         break;
 
     case VHOST_USER_SET_OWNER:
-    case VHOST_USER_RESET_OWNER:
+    case VHOST_USER_RESET_DEVICE:
         break;
 
     case VHOST_USER_SET_MEM_TABLE:
linux-headers/linux/vhost.h +1 -1
index ead86db..14a0160 100644
@@ -78,7 +78,7 @@ struct vhost_memory {
 #define VHOST_SET_OWNER _IO(VHOST_VIRTIO, 0x01)
 /* Give up ownership, and reset the device to default values.
  * Allows subsequent call to VHOST_OWNER_SET to succeed. */
-#define VHOST_RESET_OWNER _IO(VHOST_VIRTIO, 0x02)
+#define VHOST_RESET_DEVICE _IO(VHOST_VIRTIO, 0x02)
 
 /* Set up/modify memory layout */
 #define VHOST_SET_MEM_TABLE	_IOW(VHOST_VIRTIO, 0x03, struct vhost_memory)
tests/vhost-user-test.c +1 -1
index 75fedf0..e301db7 100644
@@ -58,7 +58,7 @@ typedef enum VhostUserRequest {
     VHOST_USER_GET_FEATURES = 1,
     VHOST_USER_SET_FEATURES = 2,
     VHOST_USER_SET_OWNER = 3,
-    VHOST_USER_RESET_OWNER = 4,
+    VHOST_USER_RESET_DEVICE = 4,
     VHOST_USER_SET_MEM_TABLE = 5,
     VHOST_USER_SET_LOG_BASE = 6,
     VHOST_USER_SET_LOG_FD = 7,
Signature · 1 line
1.9.0
#4 lore raw

PATCH v7 4/7 vhost-user: add VHOST_USER_GET_QUEUE_NUM message

In reply to Yuanhan Liu

This is for querying how many queues the backend supports if it has mq support(when VHOST_USER_PROTOCOL_F_MQ flag is set from the quried protocol features).

vhost_net_get_max_queues() is the interface to export that value, and to tell if the backend supports # of queues user requested, which is done in the following patch.

---
 docs/specs/vhost-user.txt | 11 +++++++++++
 hw/net/vhost_net.c        |  7 +++++++
 hw/virtio/vhost-user.c    | 14 +++++++++++++-
 include/hw/virtio/vhost.h |  1 +
 include/net/vhost_net.h   |  1 +
 5 files changed, 33 insertions(+), 1 deletion(-)
docs/specs/vhost-user.txt +11 -0
index ccbbcbb..43db9b4 100644
@@ -301,3 +301,14 @@ Message types
       Bits (0-7) of the payload contain the vring index. Bit 8 is the
       invalid FD flag. This flag is set when there is no file descriptor
       in the ancillary data.
+
+ * VHOST_USER_GET_QUEUE_NUM
+
+      Id: 17
+      Equivalent ioctl: N/A
+      Master payload: N/A
+      Slave payload: u64
+
+      Query how many queues the backend supports. This request should be
+      sent only when VHOST_USER_PROTOCOL_F_MQ is set in quried protocol
+      features by VHOST_USER_GET_PROTOCOL_FEATURES.
hw/net/vhost_net.c +7 -0
index 33b0e97..f9441e9 100644
@@ -122,6 +122,11 @@ void vhost_net_ack_features(struct vhost_net *net, uint64_t features)
     vhost_ack_features(&net->dev, vhost_net_get_feature_bits(net), features);
 }
 
+uint64_t vhost_net_get_max_queues(VHostNetState *net)
+{
+    return net->dev.max_queues;
+}
+
 static int vhost_net_get_fd(NetClientState *backend)
 {
     switch (backend->info->type) {
@@ -144,6 +149,8 @@ struct vhost_net *vhost_net_init(VhostNetOptions *options)
         goto fail;
     }
 
+    net->dev.max_queues = 1;
+
     if (backend_kernel) {
         r = vhost_net_get_fd(options->net_backend);
         if (r < 0) {
hw/virtio/vhost-user.c +13 -1
index 192ea4e..8046bc0 100644
@@ -25,7 +25,9 @@
 
 #define VHOST_MEMORY_MAX_NREGIONS    8
 #define VHOST_USER_F_PROTOCOL_FEATURES 30
-#define VHOST_USER_PROTOCOL_FEATURE_MASK 0x0ULL
+#define VHOST_USER_PROTOCOL_FEATURE_MASK 0x1ULL
+
+#define VHOST_USER_PROTOCOL_F_MQ    0
 
 typedef enum VhostUserRequest {
     VHOST_USER_NONE = 0,
@@ -45,6 +47,7 @@ typedef enum VhostUserRequest {
     VHOST_USER_SET_VRING_ERR = 14,
     VHOST_USER_GET_PROTOCOL_FEATURES = 15,
     VHOST_USER_SET_PROTOCOL_FEATURES = 16,
+    VHOST_USER_GET_QUEUE_NUM = 17,
     VHOST_USER_MAX
 } VhostUserRequest;
 
@@ -210,6 +213,7 @@ static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
     switch (msg_request) {
     case VHOST_USER_GET_FEATURES:
     case VHOST_USER_GET_PROTOCOL_FEATURES:
+    case VHOST_USER_GET_QUEUE_NUM:
         need_reply = 1;
         break;
 
@@ -314,6 +318,7 @@ static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
         switch (msg_request) {
         case VHOST_USER_GET_FEATURES:
         case VHOST_USER_GET_PROTOCOL_FEATURES:
+        case VHOST_USER_GET_QUEUE_NUM:
             if (msg.size != sizeof(m.u64)) {
                 error_report("Received bad msg size.");
                 return -1;
@@ -362,6 +367,13 @@ static int vhost_user_init(struct vhost_dev *dev, void *opaque)
                               &dev->protocol_features);
         if (err < 0)
             return err;
+
+        /* query the max queues we support if backend supports Multiple Queue */
+        if (dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_MQ)) {
+            err = vhost_user_call(dev, VHOST_USER_GET_QUEUE_NUM, &dev->max_queues);
+            if (err < 0)
+                return err;
+        }
     }
 
     return 0;
include/hw/virtio/vhost.h +1 -0
index 6467c73..c3758f3 100644
@@ -48,6 +48,7 @@ struct vhost_dev {
     unsigned long long acked_features;
     unsigned long long backend_features;
     unsigned long long protocol_features;
+    unsigned long long max_queues;
     bool started;
     bool log_enabled;
     unsigned long long log_size;
include/net/vhost_net.h +1 -0
index 840d4b1..8db723e 100644
@@ -13,6 +13,7 @@ typedef struct VhostNetOptions {
     void *opaque;
 } VhostNetOptions;
 
+uint64_t vhost_net_get_max_queues(VHostNetState *net);
 struct vhost_net *vhost_net_init(VhostNetOptions *options);
 
 int vhost_net_start(VirtIODevice *dev, NetClientState *ncs, int total_queues);
Signature · 1 line
1.9.0
#5 lore raw

PATCH v7 5/7 vhost_net: move vhost_net_set_vq_index ahead at vhost_net_init

In reply to Yuanhan Liu

So that we could use the `vq_index' as well in the vhost_net_init stage, which is required when adding vhost-user multiple-queue support, where we need the vq_index to indicate which queue pair we are gonna initiate.

vhost-user has no multiple queue support yet, hence no queue_index set before. Here is a quick set to 0 at net_vhost_user_init() stage, and it will be set properly soon in the next patch.

---
 hw/net/vhost_net.c | 16 +++++++---------
 net/vhost-user.c   |  1 +
 2 files changed, 8 insertions(+), 9 deletions(-)
hw/net/vhost_net.c +7 -9
index f9441e9..141b557 100644
@@ -138,6 +138,11 @@ static int vhost_net_get_fd(NetClientState *backend)
     }
 }
 
+static void vhost_net_set_vq_index(struct vhost_net *net, int vq_index)
+{
+    net->dev.vq_index = vq_index;
+}
+
 struct vhost_net *vhost_net_init(VhostNetOptions *options)
 {
     int r;
@@ -167,6 +172,8 @@ struct vhost_net *vhost_net_init(VhostNetOptions *options)
     }
     net->nc = options->net_backend;
 
+    vhost_net_set_vq_index(net, net->nc->queue_index * 2);
+
     net->dev.nvqs = 2;
     net->dev.vqs = net->vqs;
 
@@ -196,11 +203,6 @@ fail:
     return NULL;
 }
 
-static void vhost_net_set_vq_index(struct vhost_net *net, int vq_index)
-{
-    net->dev.vq_index = vq_index;
-}
-
 static int vhost_net_set_vnet_endian(VirtIODevice *dev, NetClientState *peer,
                                      bool set)
 {
@@ -325,10 +327,6 @@ int vhost_net_start(VirtIODevice *dev, NetClientState *ncs,
         goto err;
     }
 
-    for (i = 0; i < total_queues; i++) {
-        vhost_net_set_vq_index(get_vhost_net(ncs[i].peer), i * 2);
-    }
-
     r = k->set_guest_notifiers(qbus->parent, total_queues * 2, true);
     if (r < 0) {
         error_report("Error binding guest notifier: %d", -r);
net/vhost-user.c +1 -0
index 93dcecd..2d6bbe5 100644
@@ -146,6 +146,7 @@ static int net_vhost_user_init(NetClientState *peer, const char *device,
     /* We don't provide a receive callback */
     s->nc.receive_disabled = 1;
     s->chr = chr;
+    nc->queue_index = 0;
 
     qemu_chr_add_handlers(s->chr, NULL, NULL, net_vhost_user_event, s);
 
Signature · 1 line
1.9.0
#6 lore raw

Re: [Qemu-devel] [PATCH 5/7] vhost_net: move vhost_net_set_vq_index ahead at vhost_net_init

Quoted 32 lines · Yuanhan Liu

So that we could use the `vq_index' as well in the vhost_net_init stage, which is required when adding vhost-user multiple-queue support, where we need the vq_index to indicate which queue pair we are gonna initiate.

vhost-user has no multiple queue support yet, hence no queue_index set before. Here is a quick set to 0 at net_vhost_user_init() stage, and it will be set properly soon in the next patch.

---
 hw/net/vhost_net.c | 16 +++++++---------
 net/vhost-user.c   |  1 +
 2 files changed, 8 insertions(+), 9 deletions(-)
hw/net/vhost_net.c +5 -0
index f9441e9..141b557 100644
@@ -138,6 +138,11 @@ static int vhost_net_get_fd(NetClientState *backend)
     }
 }
 
+static void vhost_net_set_vq_index(struct vhost_net *net, int vq_index)
+{
+    net->dev.vq_index = vq_index;
+}
+
 struct vhost_net *vhost_net_init(VhostNetOptions *options)
 {
     int r;

hw/net/vhost_net.c struct vhost_net *vhost_net_init(VhostNetOptions *options) · line 176

Quoted 5 lines
hw/net/vhost_net.c +2 -0
@@ -167,6 +172,8 @@ struct vhost_net *vhost_net_init(VhostNetOptions *options)
     }
     net->nc = options->net_backend;
 
+    vhost_net_set_vq_index(net, net->nc->queue_index * 2);
+

This breaks vhost kernel multiqueue since queue_index was not initialized at this time. We do this in set_netdev() instead of setting it in each kind of netdev.

Quoted 31 lines
hw/net/vhost_net.c +0 -9
     net->dev.nvqs = 2;
     net->dev.vqs = net->vqs;
 
@@ -196,11 +203,6 @@ fail:
     return NULL;
 }
 
-static void vhost_net_set_vq_index(struct vhost_net *net, int vq_index)
-{
-    net->dev.vq_index = vq_index;
-}
-
 static int vhost_net_set_vnet_endian(VirtIODevice *dev, NetClientState *peer,
                                      bool set)
 {
@@ -325,10 +327,6 @@ int vhost_net_start(VirtIODevice *dev, NetClientState *ncs,
         goto err;
     }
 
-    for (i = 0; i < total_queues; i++) {
-        vhost_net_set_vq_index(get_vhost_net(ncs[i].peer), i * 2);
-    }
-
     r = k->set_guest_notifiers(qbus->parent, total_queues * 2, true);
     if (r < 0) {
         error_report("Error binding guest notifier: %d", -r);
net/vhost-user.c +0 -0
index 93dcecd..2d6bbe5 100644

net/vhost-user.c static int net_vhost_user_init(NetClientState *peer, const char *device, · line 150

Quoted 5 lines
net/vhost-user.c +1 -0
@@ -146,6 +146,7 @@ static int net_vhost_user_init(NetClientState *peer, const char *device,
     /* We don't provide a receive callback */
     s->nc.receive_disabled = 1;
     s->chr = chr;
+    nc->queue_index = 0;
 

Fail to understand why this is needed. It will be set to correct value in set_netdev().

Quoted 2 lines
net/vhost-user.c +0 -0
     qemu_chr_add_handlers(s->chr, NULL, NULL, net_vhost_user_event, s);
 
#7 lore raw

Re: [Qemu-devel] [PATCH 5/7] vhost_net: move vhost_net_set_vq_index ahead at vhost_net_init

Quoted 38 lines · Jason Wang
Quoted 35 lines · Yuanhan Liu

So that we could use the `vq_index' as well in the vhost_net_init stage, which is required when adding vhost-user multiple-queue support, where we need the vq_index to indicate which queue pair we are gonna initiate.

vhost-user has no multiple queue support yet, hence no queue_index set before. Here is a quick set to 0 at net_vhost_user_init() stage, and it will be set properly soon in the next patch.

---
 hw/net/vhost_net.c | 16 +++++++---------
 net/vhost-user.c   |  1 +
 2 files changed, 8 insertions(+), 9 deletions(-)
hw/net/vhost_net.c +5 -0
index f9441e9..141b557 100644
@@ -138,6 +138,11 @@ static int vhost_net_get_fd(NetClientState *backend)
     }
 }
 
+static void vhost_net_set_vq_index(struct vhost_net *net, int vq_index)
+{
+    net->dev.vq_index = vq_index;
+}
+
 struct vhost_net *vhost_net_init(VhostNetOptions *options)
 {
     int r;
@@ -167,6 +172,8 @@ struct vhost_net *vhost_net_init(VhostNetOptions *options)
     }
     net->nc = options->net_backend;
 
Quoted 5 lines

hw/net/vhost_net.c struct vhost_net *vhost_net_init(VhostNetOptions *options) · line 176

Quoted 2 lines
hw/net/vhost_net.c +2 -0
+    vhost_net_set_vq_index(net, net->nc->queue_index * 2);
+

This breaks vhost kernel multiqueue since queue_index was not initialized at this time.

Right, thanks for pointing it out.

Quoted 2 lines

We do this in set_netdev() instead of setting
it in each kind of netdev.

Can we move it to net_init_tap() for setting the right queue_index for each nc?

Or, can we call vhost_net_set_vq_index twice, one at vhost_net_init(for vhost-user mq support), another one at vhost_net_start(for vhost kernel mq support)?

Or, do you have better ideas?

Quoted 32 lines
Quoted 32 lines
hw/net/vhost_net.c +0 -9
     net->dev.nvqs = 2;
     net->dev.vqs = net->vqs;
 
@@ -196,11 +203,6 @@ fail:
     return NULL;
 }
 
-static void vhost_net_set_vq_index(struct vhost_net *net, int vq_index)
-{
-    net->dev.vq_index = vq_index;
-}
-
 static int vhost_net_set_vnet_endian(VirtIODevice *dev, NetClientState *peer,
                                      bool set)
 {
@@ -325,10 +327,6 @@ int vhost_net_start(VirtIODevice *dev, NetClientState *ncs,
         goto err;
     }
 
-    for (i = 0; i < total_queues; i++) {
-        vhost_net_set_vq_index(get_vhost_net(ncs[i].peer), i * 2);
-    }
-
     r = k->set_guest_notifiers(qbus->parent, total_queues * 2, true);
     if (r < 0) {
         error_report("Error binding guest notifier: %d", -r);
net/vhost-user.c +0 -0
index 93dcecd..2d6bbe5 100644
@@ -146,6 +146,7 @@ static int net_vhost_user_init(NetClientState *peer, const char *device,
Quoted 8 lines

net/vhost-user.c static int net_vhost_user_init(NetClientState *peer, const char *device, · line 150

Quoted 5 lines
net/vhost-user.c +1 -0
     /* We don't provide a receive callback */
     s->nc.receive_disabled = 1;
     s->chr = chr;
+    nc->queue_index = 0;
 

Fail to understand why this is needed. It will be set to correct value in set_netdev().

Right, it's unnecessary.

BTW, May I ask you a quick question: why set_netdev is invoked after tap init, while it is invoked before vhost-user init?

	--yliu
Quoted 3 lines
Quoted 3 lines
net/vhost-user.c +0 -0
     qemu_chr_add_handlers(s->chr, NULL, NULL, net_vhost_user_event, s);
 
#8 lore raw

Re: [Qemu-devel] [PATCH 5/7] vhost_net: move vhost_net_set_vq_index ahead at vhost_net_init

Quoted 46 lines · Yuanhan Liu
Quoted 37 lines · Jason Wang
Quoted 35 lines · Yuanhan Liu

So that we could use the `vq_index' as well in the vhost_net_init stage, which is required when adding vhost-user multiple-queue support, where we need the vq_index to indicate which queue pair we are gonna initiate.

vhost-user has no multiple queue support yet, hence no queue_index set before. Here is a quick set to 0 at net_vhost_user_init() stage, and it will be set properly soon in the next patch.

---
 hw/net/vhost_net.c | 16 +++++++---------
 net/vhost-user.c   |  1 +
 2 files changed, 8 insertions(+), 9 deletions(-)
hw/net/vhost_net.c +5 -0
index f9441e9..141b557 100644
@@ -138,6 +138,11 @@ static int vhost_net_get_fd(NetClientState *backend)
     }
 }
 
+static void vhost_net_set_vq_index(struct vhost_net *net, int vq_index)
+{
+    net->dev.vq_index = vq_index;
+}
+
 struct vhost_net *vhost_net_init(VhostNetOptions *options)
 {
     int r;
@@ -167,6 +172,8 @@ struct vhost_net *vhost_net_init(VhostNetOptions *options)
     }
     net->nc = options->net_backend;
 
Quoted 4 lines

hw/net/vhost_net.c struct vhost_net *vhost_net_init(VhostNetOptions *options) · line 176

Quoted 2 lines
hw/net/vhost_net.c +2 -0
+    vhost_net_set_vq_index(net, net->nc->queue_index * 2);
+

This breaks vhost kernel multiqueue since queue_index was not initialized at this time.

Right, thanks for pointing it out.

Quoted 2 lines

We do this in set_netdev() instead of setting
it in each kind of netdev.

Quoted 8 lines

Can we move it to net_init_tap() for setting the right queue_index for each nc?

Or, can we call vhost_net_set_vq_index twice, one at vhost_net_init(for vhost-user mq support), another one at vhost_net_start(for vhost kernel mq support)?

Or, do you have better ideas?

I think setting queue_index in net_init_tap() looks ok. But a question is that why need we do this at so early stage? ( Even before its peers is connected.)

Quoted 39 lines
Quoted 31 lines
Quoted 31 lines
hw/net/vhost_net.c +0 -9
     net->dev.nvqs = 2;
     net->dev.vqs = net->vqs;
 
@@ -196,11 +203,6 @@ fail:
     return NULL;
 }
 
-static void vhost_net_set_vq_index(struct vhost_net *net, int vq_index)
-{
-    net->dev.vq_index = vq_index;
-}
-
 static int vhost_net_set_vnet_endian(VirtIODevice *dev, NetClientState *peer,
                                      bool set)
 {
@@ -325,10 +327,6 @@ int vhost_net_start(VirtIODevice *dev, NetClientState *ncs,
         goto err;
     }
 
-    for (i = 0; i < total_queues; i++) {
-        vhost_net_set_vq_index(get_vhost_net(ncs[i].peer), i * 2);
-    }
-
     r = k->set_guest_notifiers(qbus->parent, total_queues * 2, true);
     if (r < 0) {
         error_report("Error binding guest notifier: %d", -r);
net/vhost-user.c +0 -0
index 93dcecd..2d6bbe5 100644
Quoted 8 lines

net/vhost-user.c static int net_vhost_user_init(NetClientState *peer, const char *device, · line 150

Quoted 6 lines
net/vhost-user.c +1 -0
@@ -146,6 +146,7 @@ static int net_vhost_user_init(NetClientState *peer, const char *device,
     /* We don't provide a receive callback */
     s->nc.receive_disabled = 1;
     s->chr = chr;
+    nc->queue_index = 0;
 

Fail to understand why this is needed. It will be set to correct value in set_netdev().

Quoted 6 lines

Right, it's unnecessary.

BTW, May I ask you a quick question: why set_netdev is invoked after tap init, while it is invoked before vhost-user init?

	--yliu

I think the reason is that you must initialize all netdevs before a device tries to connect them.

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
net/vhost-user.c +0 -0
     qemu_chr_add_handlers(s->chr, NULL, NULL, net_vhost_user_event, s);
 
#9 lore raw

Re: [Qemu-devel] [PATCH 5/7] vhost_net: move vhost_net_set_vq_index ahead at vhost_net_init

Quoted 52 lines · Jason Wang
Quoted 49 lines · Yuanhan Liu
Quoted 37 lines · Jason Wang
Quoted 35 lines · Yuanhan Liu

So that we could use the `vq_index' as well in the vhost_net_init stage, which is required when adding vhost-user multiple-queue support, where we need the vq_index to indicate which queue pair we are gonna initiate.

vhost-user has no multiple queue support yet, hence no queue_index set before. Here is a quick set to 0 at net_vhost_user_init() stage, and it will be set properly soon in the next patch.

---
 hw/net/vhost_net.c | 16 +++++++---------
 net/vhost-user.c   |  1 +
 2 files changed, 8 insertions(+), 9 deletions(-)
hw/net/vhost_net.c +5 -0
index f9441e9..141b557 100644
@@ -138,6 +138,11 @@ static int vhost_net_get_fd(NetClientState *backend)
     }
 }
 
+static void vhost_net_set_vq_index(struct vhost_net *net, int vq_index)
+{
+    net->dev.vq_index = vq_index;
+}
+
 struct vhost_net *vhost_net_init(VhostNetOptions *options)
 {
     int r;
@@ -167,6 +172,8 @@ struct vhost_net *vhost_net_init(VhostNetOptions *options)
     }
     net->nc = options->net_backend;
 
Quoted 4 lines

hw/net/vhost_net.c struct vhost_net *vhost_net_init(VhostNetOptions *options) · line 176

Quoted 2 lines
hw/net/vhost_net.c +2 -0
+    vhost_net_set_vq_index(net, net->nc->queue_index * 2);
+

This breaks vhost kernel multiqueue since queue_index was not initialized at this time.

Right, thanks for pointing it out.

Quoted 2 lines

We do this in set_netdev() instead of setting
it in each kind of netdev.

Can we move it to net_init_tap() for setting the right queue_index for each nc?

Quoted 7 lines
Quoted 5 lines

Or, can we call vhost_net_set_vq_index twice, one at vhost_net_init(for vhost-user mq support), another one at vhost_net_start(for vhost kernel mq support)?

Or, do you have better ideas?

I think setting queue_index in net_init_tap() looks ok.

Good to know.

Quoted 3 lines

But a question
is that why need we do this at so early stage? ( Even before its peers is connected.)

For vhost-user multiple queues support, we will invoke vhost_net_init() N times for each queue pair, and hence we need to distinguish which queue it is while sending messages like VHOST_SET_VRING_CALL for initializing corresponding queue pair.

Does that make sense to you?

Quoted 42 lines
Quoted 42 lines
Quoted 32 lines
Quoted 32 lines
hw/net/vhost_net.c +0 -9
     net->dev.nvqs = 2;
     net->dev.vqs = net->vqs;
 
@@ -196,11 +203,6 @@ fail:
     return NULL;
 }
 
-static void vhost_net_set_vq_index(struct vhost_net *net, int vq_index)
-{
-    net->dev.vq_index = vq_index;
-}
-
 static int vhost_net_set_vnet_endian(VirtIODevice *dev, NetClientState *peer,
                                      bool set)
 {
@@ -325,10 +327,6 @@ int vhost_net_start(VirtIODevice *dev, NetClientState *ncs,
         goto err;
     }
 
-    for (i = 0; i < total_queues; i++) {
-        vhost_net_set_vq_index(get_vhost_net(ncs[i].peer), i * 2);
-    }
-
     r = k->set_guest_notifiers(qbus->parent, total_queues * 2, true);
     if (r < 0) {
         error_report("Error binding guest notifier: %d", -r);
net/vhost-user.c +0 -0
index 93dcecd..2d6bbe5 100644
Quoted 8 lines

net/vhost-user.c static int net_vhost_user_init(NetClientState *peer, const char *device, · line 150

Quoted 6 lines
net/vhost-user.c +1 -0
@@ -146,6 +146,7 @@ static int net_vhost_user_init(NetClientState *peer, const char *device,
     /* We don't provide a receive callback */
     s->nc.receive_disabled = 1;
     s->chr = chr;
+    nc->queue_index = 0;
 

Fail to understand why this is needed. It will be set to correct value in set_netdev().

Right, it's unnecessary.

Quoted 7 lines
Quoted 4 lines

BTW, May I ask you a quick question: why set_netdev is invoked after tap init, while it is invoked before vhost-user init?

	--yliu

I think the reason is that you must initialize all netdevs before a device tries to connect them.

Thanks!

	--yliu
#10 lore raw

Re: [Qemu-devel] [PATCH 5/7] vhost_net: move vhost_net_set_vq_index ahead at vhost_net_init

Quoted 65 lines · Yuanhan Liu
Quoted 52 lines · Jason Wang
Quoted 52 lines · Yuanhan Liu
Quoted 49 lines
Quoted 49 lines · Jason Wang
Quoted 37 lines
Quoted 37 lines · Yuanhan Liu
Quoted 35 lines
Quoted 35 lines

So that we could use the `vq_index' as well in the vhost_net_init stage, which is required when adding vhost-user multiple-queue support, where we need the vq_index to indicate which queue pair we are gonna initiate.

vhost-user has no multiple queue support yet, hence no queue_index set before. Here is a quick set to 0 at net_vhost_user_init() stage, and it will be set properly soon in the next patch.

---
 hw/net/vhost_net.c | 16 +++++++---------
 net/vhost-user.c   |  1 +
 2 files changed, 8 insertions(+), 9 deletions(-)
hw/net/vhost_net.c +5 -0
index f9441e9..141b557 100644
@@ -138,6 +138,11 @@ static int vhost_net_get_fd(NetClientState *backend)
     }
 }
 
+static void vhost_net_set_vq_index(struct vhost_net *net, int vq_index)
+{
+    net->dev.vq_index = vq_index;
+}
+
 struct vhost_net *vhost_net_init(VhostNetOptions *options)
 {
     int r;
@@ -167,6 +172,8 @@ struct vhost_net *vhost_net_init(VhostNetOptions *options)
     }
     net->nc = options->net_backend;
 
Quoted 4 lines
Quoted 4 lines
Quoted 2 lines

hw/net/vhost_net.c struct vhost_net *vhost_net_init(VhostNetOptions *options) · line 176

Quoted 2 lines
hw/net/vhost_net.c +2 -0
+    vhost_net_set_vq_index(net, net->nc->queue_index * 2);
+

This breaks vhost kernel multiqueue since queue_index was not initialized at this time.

Right, thanks for pointing it out.

Quoted 2 lines
Quoted 2 lines

We do this in set_netdev() instead of setting
it in each kind of netdev.

Can we move it to net_init_tap() for setting the right queue_index for each nc?

Quoted 7 lines
Quoted 7 lines
Quoted 5 lines
Quoted 5 lines

Or, can we call vhost_net_set_vq_index twice, one at vhost_net_init(for vhost-user mq support), another one at vhost_net_start(for vhost kernel mq support)?

Or, do you have better ideas?

I think setting queue_index in net_init_tap() looks ok.

Good to know.

Quoted 3 lines
Quoted 3 lines

But a question
is that why need we do this at so early stage? ( Even before its peers is connected.)

Quoted 7 lines

For vhost-user multiple queues support, we will invoke vhost_net_init() N times for each queue pair, and hence we need to distinguish which queue it is while sending messages like VHOST_SET_VRING_CALL for initializing corresponding queue pair.

Does that make sense to you?

Not sure. Since current codes works for vhost-kernel. (vhost_net_init() was also called N times). We don't want to break existed vhost-kernel API when developing multiqueue. For each virtqueue TX/RX pair, we have one vhost net device and it has no knowledge for the others (which was hide by qemu). So VHOST_SET_VRING_CALL works without any change here.

For the case here, since you still have multiple instances of vhost_net structure. Maybe the vhost-user backend can distinguish form this?

#11 lore raw

Re: [Qemu-devel] [PATCH 5/7] vhost_net: move vhost_net_set_vq_index ahead at vhost_net_init

Quoted 76 lines · Jason Wang
Quoted 65 lines · Yuanhan Liu
Quoted 52 lines · Jason Wang
Quoted 52 lines · Yuanhan Liu
Quoted 49 lines
Quoted 49 lines · Jason Wang
Quoted 37 lines
Quoted 37 lines · Yuanhan Liu
Quoted 35 lines
Quoted 35 lines

So that we could use the `vq_index' as well in the vhost_net_init stage, which is required when adding vhost-user multiple-queue support, where we need the vq_index to indicate which queue pair we are gonna initiate.

vhost-user has no multiple queue support yet, hence no queue_index set before. Here is a quick set to 0 at net_vhost_user_init() stage, and it will be set properly soon in the next patch.

---
 hw/net/vhost_net.c | 16 +++++++---------
 net/vhost-user.c   |  1 +
 2 files changed, 8 insertions(+), 9 deletions(-)
hw/net/vhost_net.c +5 -0
index f9441e9..141b557 100644
@@ -138,6 +138,11 @@ static int vhost_net_get_fd(NetClientState *backend)
     }
 }
 
+static void vhost_net_set_vq_index(struct vhost_net *net, int vq_index)
+{
+    net->dev.vq_index = vq_index;
+}
+
 struct vhost_net *vhost_net_init(VhostNetOptions *options)
 {
     int r;
@@ -167,6 +172,8 @@ struct vhost_net *vhost_net_init(VhostNetOptions *options)
     }
     net->nc = options->net_backend;
 
Quoted 4 lines
Quoted 4 lines
Quoted 2 lines

hw/net/vhost_net.c struct vhost_net *vhost_net_init(VhostNetOptions *options) · line 176

Quoted 2 lines
hw/net/vhost_net.c +2 -0
+    vhost_net_set_vq_index(net, net->nc->queue_index * 2);
+

This breaks vhost kernel multiqueue since queue_index was not initialized at this time.

Right, thanks for pointing it out.

Quoted 2 lines
Quoted 2 lines

We do this in set_netdev() instead of setting
it in each kind of netdev.

Can we move it to net_init_tap() for setting the right queue_index for each nc?

Quoted 7 lines
Quoted 7 lines
Quoted 5 lines
Quoted 5 lines

Or, can we call vhost_net_set_vq_index twice, one at vhost_net_init(for vhost-user mq support), another one at vhost_net_start(for vhost kernel mq support)?

Or, do you have better ideas?

I think setting queue_index in net_init_tap() looks ok.

Good to know.

Quoted 3 lines
Quoted 3 lines

But a question
is that why need we do this at so early stage? ( Even before its peers is connected.)

Quoted 7 lines

For vhost-user multiple queues support, we will invoke vhost_net_init() N times for each queue pair, and hence we need to distinguish which queue it is while sending messages like VHOST_SET_VRING_CALL for initializing corresponding queue pair.

Does that make sense to you?

Quoted 8 lines

Not sure. Since current codes works for vhost-kernel. (vhost_net_init() was also called N times). We don't want to break existed vhost-kernel API when developing multiqueue. For each virtqueue TX/RX pair, we have one vhost net device and it has no knowledge for the others (which was hide by qemu). So VHOST_SET_VRING_CALL works without any change here.

For the case here, since you still have multiple instances of vhost_net structure. Maybe the vhost-user backend can distinguish form this?

Yeah, I guess that's the difference between vhost-user and vhost-kernel. Vhost-kernel opens a char device(/dev/vhost-net) for each vhost_dev, hence it's distinguishable. But for vhost-user, all vhost_dev share one char device(a socket) for communication, hence, it's not distinguishable.

I was thinking maybe we could export vhost_net_set_vq_index() and invoke it at net/vhost-user.c, so that we break nothing, and in the meantime, it keeps the logic inside vhost-user.

What do you think?

	--yliu
#12 lore raw

Re: [Qemu-devel] [PATCH 5/7] vhost_net: move vhost_net_set_vq_index ahead at vhost_net_init

Quoted 78 lines · Yuanhan Liu
Quoted 77 lines · Jason Wang
Quoted 62 lines · Yuanhan Liu
Quoted 50 lines · Jason Wang
Quoted 50 lines · Yuanhan Liu
Quoted 48 lines
Quoted 48 lines · Jason Wang
Quoted 36 lines
Quoted 36 lines · Yuanhan Liu
Quoted 35 lines
Quoted 35 lines

So that we could use the `vq_index' as well in the vhost_net_init stage, which is required when adding vhost-user multiple-queue support, where we need the vq_index to indicate which queue pair we are gonna initiate.

vhost-user has no multiple queue support yet, hence no queue_index set before. Here is a quick set to 0 at net_vhost_user_init() stage, and it will be set properly soon in the next patch.

---
 hw/net/vhost_net.c | 16 +++++++---------
 net/vhost-user.c   |  1 +
 2 files changed, 8 insertions(+), 9 deletions(-)
hw/net/vhost_net.c +5 -0
index f9441e9..141b557 100644
@@ -138,6 +138,11 @@ static int vhost_net_get_fd(NetClientState *backend)
     }
 }
 
+static void vhost_net_set_vq_index(struct vhost_net *net, int vq_index)
+{
+    net->dev.vq_index = vq_index;
+}
+
 struct vhost_net *vhost_net_init(VhostNetOptions *options)
 {
     int r;
@@ -167,6 +172,8 @@ struct vhost_net *vhost_net_init(VhostNetOptions *options)
     }
     net->nc = options->net_backend;
 
Quoted 4 lines
Quoted 4 lines
Quoted 2 lines

hw/net/vhost_net.c struct vhost_net *vhost_net_init(VhostNetOptions *options) · line 176

Quoted 2 lines
hw/net/vhost_net.c +2 -0
+    vhost_net_set_vq_index(net, net->nc->queue_index * 2);
+

This breaks vhost kernel multiqueue since queue_index was not initialized at this time.

Right, thanks for pointing it out.

Quoted 2 lines
Quoted 2 lines

We do this in set_netdev() instead of setting
it in each kind of netdev.

Can we move it to net_init_tap() for setting the right queue_index for each nc?

Quoted 6 lines
Quoted 6 lines
Quoted 5 lines
Quoted 5 lines

Or, can we call vhost_net_set_vq_index twice, one at vhost_net_init(for vhost-user mq support), another one at vhost_net_start(for vhost kernel mq support)?

Or, do you have better ideas?

I think setting queue_index in net_init_tap() looks ok.

Good to know.

Quoted 3 lines
Quoted 3 lines

But a question
is that why need we do this at so early stage? ( Even before its peers is connected.)

Quoted 7 lines

For vhost-user multiple queues support, we will invoke vhost_net_init() N times for each queue pair, and hence we need to distinguish which queue it is while sending messages like VHOST_SET_VRING_CALL for initializing corresponding queue pair.

Does that make sense to you?

Not sure. Since current codes works for vhost-kernel. (vhost_net_init() was also called N times). We don't want to break existed vhost-kernel API when developing multiqueue. For each virtqueue TX/RX pair, we have one vhost net device and it has no knowledge for the others (which was hide by qemu). So VHOST_SET_VRING_CALL works without any change here.

Quoted 6 lines
Quoted 2 lines

For the case here, since you still have multiple instances of vhost_net structure. Maybe the vhost-user backend can distinguish form this?

Yeah, I guess that's the difference between vhost-user and vhost-kernel. Vhost-kernel opens a char device(/dev/vhost-net) for each vhost_dev, hence it's distinguishable. But for vhost-user, all vhost_dev share one char device(a socket) for communication, hence, it's not distinguishable.

How about using individual socket in this case? This seems can also minimize the changes of backend.

Quoted 9 lines

I was thinking maybe we could export vhost_net_set_vq_index() and invoke it at net/vhost-user.c, so that we break nothing, and in the meantime, it keeps the logic inside vhost-user.

What do you think?

	--yliu

Sounds work. Then I believe you will need to set queue_index in vhost_user initialization code?

Thanks

#13 lore raw

Re: [Qemu-devel] [PATCH 5/7] vhost_net: move vhost_net_set_vq_index ahead at vhost_net_init

Quoted 83 lines · Jason Wang
Quoted 80 lines · Yuanhan Liu
Quoted 79 lines · Jason Wang
Quoted 62 lines · Yuanhan Liu
Quoted 50 lines · Jason Wang
Quoted 50 lines · Yuanhan Liu
Quoted 48 lines
Quoted 48 lines · Jason Wang
Quoted 36 lines
Quoted 36 lines · Yuanhan Liu
Quoted 35 lines
Quoted 35 lines

So that we could use the `vq_index' as well in the vhost_net_init stage, which is required when adding vhost-user multiple-queue support, where we need the vq_index to indicate which queue pair we are gonna initiate.

vhost-user has no multiple queue support yet, hence no queue_index set before. Here is a quick set to 0 at net_vhost_user_init() stage, and it will be set properly soon in the next patch.

---
 hw/net/vhost_net.c | 16 +++++++---------
 net/vhost-user.c   |  1 +
 2 files changed, 8 insertions(+), 9 deletions(-)
hw/net/vhost_net.c +5 -0
index f9441e9..141b557 100644
@@ -138,6 +138,11 @@ static int vhost_net_get_fd(NetClientState *backend)
     }
 }
 
+static void vhost_net_set_vq_index(struct vhost_net *net, int vq_index)
+{
+    net->dev.vq_index = vq_index;
+}
+
 struct vhost_net *vhost_net_init(VhostNetOptions *options)
 {
     int r;
@@ -167,6 +172,8 @@ struct vhost_net *vhost_net_init(VhostNetOptions *options)
     }
     net->nc = options->net_backend;
 
Quoted 4 lines
Quoted 4 lines
Quoted 2 lines

hw/net/vhost_net.c struct vhost_net *vhost_net_init(VhostNetOptions *options) · line 176

Quoted 2 lines
hw/net/vhost_net.c +2 -0
+    vhost_net_set_vq_index(net, net->nc->queue_index * 2);
+

This breaks vhost kernel multiqueue since queue_index was not initialized at this time.

Right, thanks for pointing it out.

Quoted 2 lines
Quoted 2 lines

We do this in set_netdev() instead of setting
it in each kind of netdev.

Can we move it to net_init_tap() for setting the right queue_index for each nc?

Quoted 6 lines
Quoted 6 lines
Quoted 5 lines
Quoted 5 lines

Or, can we call vhost_net_set_vq_index twice, one at vhost_net_init(for vhost-user mq support), another one at vhost_net_start(for vhost kernel mq support)?

Or, do you have better ideas?

I think setting queue_index in net_init_tap() looks ok.

Good to know.

Quoted 3 lines
Quoted 3 lines

But a question
is that why need we do this at so early stage? ( Even before its peers is connected.)

Quoted 7 lines

For vhost-user multiple queues support, we will invoke vhost_net_init() N times for each queue pair, and hence we need to distinguish which queue it is while sending messages like VHOST_SET_VRING_CALL for initializing corresponding queue pair.

Does that make sense to you?

Not sure. Since current codes works for vhost-kernel. (vhost_net_init() was also called N times). We don't want to break existed vhost-kernel API when developing multiqueue. For each virtqueue TX/RX pair, we have one vhost net device and it has no knowledge for the others (which was hide by qemu). So VHOST_SET_VRING_CALL works without any change here.

For the case here, since you still have multiple instances of vhost_net structure. Maybe the vhost-user backend can distinguish form this?

Quoted 7 lines
Quoted 4 lines

Yeah, I guess that's the difference between vhost-user and vhost-kernel. Vhost-kernel opens a char device(/dev/vhost-net) for each vhost_dev, hence it's distinguishable. But for vhost-user, all vhost_dev share one char device(a socket) for communication, hence, it's not distinguishable.

How about using individual socket in this case? This seems can also minimize the changes of backend.

I doubt that will work, for the socket is given at cmd line by "-chardev" option. You can image that it's not flexible when you want to have 10 queues, or more.

Quoted 6 lines
Quoted 6 lines

I was thinking maybe we could export vhost_net_set_vq_index() and invoke it at net/vhost-user.c, so that we break nothing, and in the meantime, it keeps the logic inside vhost-user.

Quoted 6 lines
Quoted 4 lines

What do you think?

	--yliu

Sounds work.

Good.

Quoted 2 lines

Then I believe you will need to set queue_index in vhost_user initialization code?

Yes. I then will go this way if no objection.

	--yliu 
#14 lore raw

Re: [Qemu-devel] [PATCH 5/7] vhost_net: move vhost_net_set_vq_index ahead at vhost_net_init

In reply to Jason Wang
Quoted 96 lines · Jason Wang
Quoted 78 lines · Yuanhan Liu
Quoted 77 lines · Jason Wang
Quoted 62 lines · Yuanhan Liu
Quoted 50 lines · Jason Wang
Quoted 50 lines · Yuanhan Liu
Quoted 48 lines
Quoted 48 lines · Jason Wang
Quoted 36 lines
Quoted 36 lines · Yuanhan Liu
Quoted 35 lines
Quoted 35 lines

So that we could use the `vq_index' as well in the vhost_net_init stage, which is required when adding vhost-user multiple-queue support, where we need the vq_index to indicate which queue pair we are gonna initiate.

vhost-user has no multiple queue support yet, hence no queue_index set before. Here is a quick set to 0 at net_vhost_user_init() stage, and it will be set properly soon in the next patch.

---
 hw/net/vhost_net.c | 16 +++++++---------
 net/vhost-user.c   |  1 +
 2 files changed, 8 insertions(+), 9 deletions(-)
hw/net/vhost_net.c +5 -0
index f9441e9..141b557 100644
@@ -138,6 +138,11 @@ static int vhost_net_get_fd(NetClientState *backend)
     }
 }
 
+static void vhost_net_set_vq_index(struct vhost_net *net, int vq_index)
+{
+    net->dev.vq_index = vq_index;
+}
+
 struct vhost_net *vhost_net_init(VhostNetOptions *options)
 {
     int r;
@@ -167,6 +172,8 @@ struct vhost_net *vhost_net_init(VhostNetOptions *options)
     }
     net->nc = options->net_backend;
 
Quoted 4 lines
Quoted 4 lines
Quoted 2 lines

hw/net/vhost_net.c struct vhost_net *vhost_net_init(VhostNetOptions *options) · line 176

Quoted 2 lines
hw/net/vhost_net.c +2 -0
+    vhost_net_set_vq_index(net, net->nc->queue_index * 2);
+

This breaks vhost kernel multiqueue since queue_index was not initialized at this time.

Right, thanks for pointing it out.

Quoted 2 lines
Quoted 2 lines

We do this in set_netdev() instead of setting
it in each kind of netdev.

Can we move it to net_init_tap() for setting the right queue_index for each nc?

Quoted 6 lines
Quoted 6 lines
Quoted 5 lines
Quoted 5 lines

Or, can we call vhost_net_set_vq_index twice, one at vhost_net_init(for vhost-user mq support), another one at vhost_net_start(for vhost kernel mq support)?

Or, do you have better ideas?

I think setting queue_index in net_init_tap() looks ok.

Good to know.

Quoted 3 lines
Quoted 3 lines

But a question
is that why need we do this at so early stage? ( Even before its peers is connected.)

Quoted 7 lines

For vhost-user multiple queues support, we will invoke vhost_net_init() N times for each queue pair, and hence we need to distinguish which queue it is while sending messages like VHOST_SET_VRING_CALL for initializing corresponding queue pair.

Does that make sense to you?

Not sure. Since current codes works for vhost-kernel. (vhost_net_init() was also called N times). We don't want to break existed vhost-kernel API when developing multiqueue. For each virtqueue TX/RX pair, we have one vhost net device and it has no knowledge for the others (which was hide by qemu). So VHOST_SET_VRING_CALL works without any change here.

Quoted 6 lines
Quoted 2 lines

For the case here, since you still have multiple instances of vhost_net structure. Maybe the vhost-user backend can distinguish form this?

Yeah, I guess that's the difference between vhost-user and vhost-kernel. Vhost-kernel opens a char device(/dev/vhost-net) for each vhost_dev, hence it's distinguishable. But for vhost-user, all vhost_dev share one char device(a socket) for communication, hence, it's not distinguishable.

How about using individual socket in this case? This seems can also minimize the changes of backend.

Quoted 5 lines

I was thinking maybe we could export vhost_net_set_vq_index() and invoke it at net/vhost-user.c, so that we break nothing, and in the meantime, it keeps the logic inside vhost-user.

Quoted 7 lines
Quoted 4 lines

What do you think?

	--yliu

Sounds work. Then I believe you will need to set queue_index in vhost_user initialization code?

Nah, it will not work, as vhost_net_set_vq_index() needs a vhost_net as it's parameter: you can't do that before vhost_net_init(), but it's useless to do that after vhost_net_init(). My bad for not being aware it in the first time.

How about following then?

Thanks.

	--yliu

---

hw/net/vhost_net.c +4 -1
index e2985bc..7197c3f 100644
@@ -148,6 +148,7 @@ struct vhost_net *vhost_net_init(VhostNetOptions *options)
         fprintf(stderr, "vhost-net requires net backend to be setup\n");
         goto fail;
     }
+    net->nc = options->net_backend;
 
     net->dev.max_queues = 1;
 
@@ -164,8 +165,10 @@ struct vhost_net *vhost_net_init(VhostNetOptions *options)
         net->dev.backend_features = 0;
         net->dev.protocol_features = 0;
         net->backend = -1;
+
+        /* vhost-user needs vq_index to initiate a specific queue pair */
+        net->dev.vq_index = net->nc->queue_index * 2;
     }
-    net->nc = options->net_backend;
 
     net->dev.nvqs = 2;
     net->dev.vqs = net->vqs;
#15 lore raw

PATCH v7 6/7 vhost-user: add multiple queue support

In reply to Yuanhan Liu

From: Ouyang Changchun <[email protected]>

This patch is initially based a patch from Nikolay Nikolaev.

Here is the latest version for adding vhost-user multiple queue support, by creating a nc and vhost_net pair for each queue.

What differs from last version is that this patch addresses two major concerns from Michael, and fixes one hidden bug.

- Concern #1: no feedback when the backend can't support # of requested queues(by providing queues=# option).

Here we address this issue by querying VHOST_USER_PROTOCOL_F_MQ protocol features first, if not set, it means the backend don't support mq feature, and let max_queues be 1. Otherwise, we send another message, VHOST_USER_GET_QUEUE_NUM, for getting the max_queues the backend supports.

At vhost-user initiation stage(net_vhost_user_start), we then initiate one queue first, which, in the meantime, also gets the max_queues. We then do a simple compare: if requested_queues > max_queues, we exit(I guess it's safe to exit here, as the vm is not running yet).

- Concern #2: some messages are sent more times than necessary.

We came an agreement with Michael that we could categorize vhost user messages to 2 types: none-vring specific messages, which should be sent only once, and vring specific messages, which should be sent per queue.

Here I introduced a helper function vhost_user_one_time_request(), which lists following messages as none-vring specific messages:

VHOST_USER_GET_FEATURES
VHOST_USER_SET_FEATURES
VHOST_USER_GET_PROTOCOL_FEATURES
VHOST_USER_SET_PROTOCOL_FEATURES
VHOST_USER_SET_OWNER
VHOST_USER_RESET_DEVICE
VHOST_USER_SET_MEM_TABLE
VHOST_USER_GET_QUEUE_NUM

For above messages, we simply ignore them when they are not sent the first time.

I also observed a hidden bug from last version. We register the char dev event handler N times, which is not necessary, as well as buggy: A later register overwrites the former one, as qemu_chr_add_handlers() will not chain those handlers, but instead overwrites the old one. So, in theory, invoking qemu_chr_add_handlers N times will not end up with calling the handler N times.

However, the reason the handler is invoked N times is because we start the backend(the connection server) first, and hence when net_vhost_user_init() is executed, the connection is already established, and qemu_chr_add_handlers() then invoke the handler immediately, which just looks like we invoke the handler(net_vhost_user_event) directly from net_vhost_user_init().

The solution I came up with is to make VhostUserState as an upper level structure, making it includes N nc and vhost_net pairs:

   struct VhostUserNetPeer {
       NetClientState *nc;
       VHostNetState  *vhost_net;
   };

   typedef struct VhostUserState {
       CharDriverState *chr;
       bool running;
       int queues;
       struct VhostUserNetPeer peers[];
   } VhostUserState;
---
 docs/specs/vhost-user.txt |  13 +++++
 hw/virtio/vhost-user.c    |  31 ++++++++++-
 include/net/net.h         |   1 +
 net/vhost-user.c          | 136 ++++++++++++++++++++++++++++++++--------------
 qapi-schema.json          |   6 +-
 qemu-options.hx           |   5 +-
 6 files changed, 146 insertions(+), 46 deletions(-)
docs/specs/vhost-user.txt +13 -0
index 43db9b4..99d79be 100644
@@ -135,6 +135,19 @@ As older slaves don't support negotiating protocol features,
 a feature bit was dedicated for this purpose:
 #define VHOST_USER_F_PROTOCOL_FEATURES 30
 
+Multiple queue support
+-------------------
+Multiple queue is treated as a protocal extension, hence the slave has to
+implement protocol features first. Multiple queues is supported only when
+the protocol feature VHOST_USER_PROTOCOL_F_MQ(bit 0) is set.
+
+The max # of queues the slave support can be queried with message
+VHOST_USER_GET_PROTOCOL_FEATURES. Master should stop when the # of requested
+queues is bigger than that.
+
+As all queues share one connection, the master use a unique index for each
+queue in the sent message to identify one specified queue.
+
 Message types
 -------------
 
hw/virtio/vhost-user.c +30 -1
index 8046bc0..11e46b5 100644
@@ -187,6 +187,23 @@ static int vhost_user_write(struct vhost_dev *dev, VhostUserMsg *msg,
             0 : -1;
 }
 
+static bool vhost_user_one_time_request(VhostUserRequest request)
+{
+    switch (request) {
+    case VHOST_USER_GET_FEATURES:
+    case VHOST_USER_SET_FEATURES:
+    case VHOST_USER_GET_PROTOCOL_FEATURES:
+    case VHOST_USER_SET_PROTOCOL_FEATURES:
+    case VHOST_USER_SET_OWNER:
+    case VHOST_USER_RESET_DEVICE:
+    case VHOST_USER_SET_MEM_TABLE:
+    case VHOST_USER_GET_QUEUE_NUM:
+        return true;
+    default:
+        return false;
+    }
+}
+
 static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
         void *arg)
 {
@@ -206,6 +223,14 @@ static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
     else
         msg_request = request;
 
+    /*
+     * For none-vring specific requests, like VHOST_USER_GET_FEATURES,
+     * we just need send it once in the first time. For later such
+     * request, we just ignore it.
+     */
+    if (vhost_user_one_time_request(msg_request) && dev->vq_index != 0)
+        return 0;
+
     msg.request = msg_request;
     msg.flags = VHOST_USER_VERSION;
     msg.size = 0;
@@ -268,17 +293,20 @@ static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
     case VHOST_USER_SET_VRING_NUM:
     case VHOST_USER_SET_VRING_BASE:
         memcpy(&msg.state, arg, sizeof(struct vhost_vring_state));
+        msg.addr.index += dev->vq_index;
         msg.size = sizeof(m.state);
         break;
 
     case VHOST_USER_GET_VRING_BASE:
         memcpy(&msg.state, arg, sizeof(struct vhost_vring_state));
+        msg.addr.index += dev->vq_index;
         msg.size = sizeof(m.state);
         need_reply = 1;
         break;
 
     case VHOST_USER_SET_VRING_ADDR:
         memcpy(&msg.addr, arg, sizeof(struct vhost_vring_addr));
+        msg.addr.index += dev->vq_index;
         msg.size = sizeof(m.addr);
         break;
 
@@ -286,7 +314,7 @@ static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
     case VHOST_USER_SET_VRING_CALL:
     case VHOST_USER_SET_VRING_ERR:
         file = arg;
-        msg.u64 = file->index & VHOST_USER_VRING_IDX_MASK;
+        msg.u64 = (file->index + dev->vq_index) & VHOST_USER_VRING_IDX_MASK;
         msg.size = sizeof(m.u64);
         if (ioeventfd_enabled() && file->fd > 0) {
             fds[fd_num++] = file->fd;
@@ -330,6 +358,7 @@ static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
                 error_report("Received bad msg size.");
                 return -1;
             }
+            msg.state.index -= dev->vq_index;
             memcpy(arg, &msg.state, sizeof(struct vhost_vring_state));
             break;
         default:
include/net/net.h +1 -0
index 6a6cbef..6f20656 100644
@@ -92,6 +92,7 @@ struct NetClientState {
     NetClientDestructor *destructor;
     unsigned int queue_index;
     unsigned rxfilter_notify_enabled:1;
+    void *opaque;
 };
 
 typedef struct NICState {
net/vhost-user.c +94 -42

222 lines not on this page · raw message

qapi-schema.json +5 -1
index 67fef37..55c33db 100644
@@ -2480,12 +2480,16 @@
 #
 # @vhostforce: #optional vhost on for non-MSIX virtio guests (default: false).
 #
+# @queues: #optional number of queues to be created for multiqueue vhost-user
+#          (default: 1) (Since 2.5)
+#
 # Since 2.1
 ##
 { 'struct': 'NetdevVhostUserOptions',
   'data': {
     'chardev':        'str',
-    '*vhostforce':    'bool' } }
+    '*vhostforce':    'bool',
+    '*queues':        'int' } }
 
 ##
 # @NetClientOptions
qemu-options.hx +3 -2
index 77f5853..5bfa7a3 100644
@@ -1963,13 +1963,14 @@ The hubport netdev lets you connect a NIC to a QEMU "vlan" instead of a single
 netdev.  @code{-net} and @code{-device} with parameter @option{vlan} create the
 required hub automatically.
 
-@item -netdev vhost-user,chardev=@var{id}[,vhostforce=on|off]
+@item -netdev vhost-user,chardev=@var{id}[,vhostforce=on|off][,queues=n]
 
 Establish a vhost-user netdev, backed by a chardev @var{id}. The chardev should
 be a unix domain socket backed one. The vhost-user uses a specifically defined
 protocol to pass vhost ioctl replacement messages to an application on the other
 end of the socket. On non-MSIX guests, the feature can be forced with
-@var{vhostforce}.
+@var{vhostforce}. Use 'queues=@var{n}' to specify the number of queues to
+be created for multiqueue vhost-user.
 
 Example:
 @example
Signature · 1 line
1.9.0
#16 lore raw

Re: [Qemu-devel] [PATCH 6/7] vhost-user: add multiple queue support

Quoted 7 lines · Yuanhan Liu

From: Ouyang Changchun <[email protected]>

This patch is initially based a patch from Nikolay Nikolaev.

Here is the latest version for adding vhost-user multiple queue support, by creating a nc and vhost_net pair for each queue.

Reviewing grammar and interface only:

docs/specs/vhost-user.txt As older slaves don't support negotiating protocol features, · line 140

Quoted 8 lines

+++ b/docs/specs/vhost-user.txt

docs/specs/vhost-user.txt +3 -0
@@ -135,6 +135,19 @@ As older slaves don't support negotiating protocol features,
 a feature bit was dedicated for this purpose:
 #define VHOST_USER_F_PROTOCOL_FEATURES 30
 
+Multiple queue support
+-------------------
+Multiple queue is treated as a protocal extension, hence the slave has to

s/protocal/protocol/

docs/specs/vhost-user.txt As older slaves don't support negotiating protocol features, · line 144

Quoted 4 lines
docs/specs/vhost-user.txt +4 -0
+implement protocol features first. Multiple queues is supported only when
+the protocol feature VHOST_USER_PROTOCOL_F_MQ(bit 0) is set.
+
+The max # of queues the slave support can be queried with message

s/#/number/
s/support/supports/

docs/specs/vhost-user.txt As older slaves don't support negotiating protocol features, · line 145

Quoted 1 line
docs/specs/vhost-user.txt +1 -0
+VHOST_USER_GET_PROTOCOL_FEATURES. Master should stop when the # of requested

s/#/number/

docs/specs/vhost-user.txt As older slaves don't support negotiating protocol features, · line 148

Quoted 3 lines
docs/specs/vhost-user.txt +3 -0
+queues is bigger than that.
+
+As all queues share one connection, the master use a unique index for each

s/use/uses/

Quoted 9 lines
docs/specs/vhost-user.txt +4 -0
+queue in the sent message to identify one specified queue.
+
 
+++ b/qapi-schema.json
@@ -2480,12 +2480,16 @@
 #
 # @vhostforce: #optional vhost on for non-MSIX virtio guests (default: false).
 #
+# @queues: #optional number of queues to be created for multiqueue vhost-user

docs/specs/vhost-user.txt line 2492

Quoted 10 lines
docs/specs/vhost-user.txt +4 -1
+#          (default: 1) (Since 2.5)
+#
 # Since 2.1
 ##
 { 'struct': 'NetdevVhostUserOptions',
   'data': {
     'chardev':        'str',
-    '*vhostforce':    'bool' } }
+    '*vhostforce':    'bool',
+    '*queues':        'int' } }

Looks okay.

Signature · 2 lines
Eric Blake   eblake redhat com    +1-919-301-3266
Libvirt virtualization library http://libvirt.org
#17 lore raw

Re: [Qemu-devel] [PATCH 6/7] vhost-user: add multiple queue support

Quoted 10 lines · Eric Blake
Quoted 7 lines · Yuanhan Liu

From: Ouyang Changchun <[email protected]>

This patch is initially based a patch from Nikolay Nikolaev.

Here is the latest version for adding vhost-user multiple queue support, by creating a nc and vhost_net pair for each queue.

Reviewing grammar and interface only:

Thanks, and will fix them in next patch.

	--yliu
Quoted 56 lines

docs/specs/vhost-user.txt As older slaves don't support negotiating protocol features, · line 140

Quoted 9 lines

+++ b/docs/specs/vhost-user.txt

docs/specs/vhost-user.txt +3 -0
@@ -135,6 +135,19 @@ As older slaves don't support negotiating protocol features,
 a feature bit was dedicated for this purpose:
 #define VHOST_USER_F_PROTOCOL_FEATURES 30
 
+Multiple queue support
+-------------------
+Multiple queue is treated as a protocal extension, hence the slave has to

s/protocal/protocol/

docs/specs/vhost-user.txt As older slaves don't support negotiating protocol features, · line 144

Quoted 4 lines
docs/specs/vhost-user.txt +4 -0
+implement protocol features first. Multiple queues is supported only when
+the protocol feature VHOST_USER_PROTOCOL_F_MQ(bit 0) is set.
+
+The max # of queues the slave support can be queried with message

s/#/number/
s/support/supports/

docs/specs/vhost-user.txt As older slaves don't support negotiating protocol features, · line 145

Quoted 1 line
docs/specs/vhost-user.txt +1 -0
+VHOST_USER_GET_PROTOCOL_FEATURES. Master should stop when the # of requested

s/#/number/

docs/specs/vhost-user.txt As older slaves don't support negotiating protocol features, · line 148

Quoted 3 lines
docs/specs/vhost-user.txt +3 -0
+queues is bigger than that.
+
+As all queues share one connection, the master use a unique index for each

s/use/uses/

Quoted 9 lines
docs/specs/vhost-user.txt +4 -0
+queue in the sent message to identify one specified queue.
+
 
+++ b/qapi-schema.json
@@ -2480,12 +2480,16 @@
 #
 # @vhostforce: #optional vhost on for non-MSIX virtio guests (default: false).
 #
+# @queues: #optional number of queues to be created for multiqueue vhost-user

docs/specs/vhost-user.txt line 2492

Quoted 10 lines
docs/specs/vhost-user.txt +4 -1
+#          (default: 1) (Since 2.5)
+#
 # Since 2.1
 ##
 { 'struct': 'NetdevVhostUserOptions',
   'data': {
     'chardev':        'str',
-    '*vhostforce':    'bool' } }
+    '*vhostforce':    'bool',
+    '*queues':        'int' } }

Looks okay.

Signature · 2 lines
Eric Blake   eblake redhat com    +1-919-301-3266
Libvirt virtualization library http://libvirt.org
#18 lore raw

Re: [Qemu-devel] [PATCH 6/7] vhost-user: add multiple queue support

In reply to Yuanhan Liu
Quoted 28 lines

-----Original Message-----
From: Yuanhan Liu [mailto:[email protected]]
Sent: Tuesday, September 8, 2015 3:39 PM
To: [email protected]
Cc: [email protected]; Ouyang, Changchun; Yuanhan Liu
Subject: [PATCH 6/7] vhost-user: add multiple queue support

From: Ouyang Changchun <[email protected]>

This patch is initially based a patch from Nikolay Nikolaev.

Here is the latest version for adding vhost-user multiple queue support, by creating a nc and vhost_net pair for each queue.

 static int vhost_user_start(VhostUserState *s)  {
     VhostNetOptions options;
net/vhost-user.c +4 -2
+    VHostNetState *vhost_net;
+    int max_queues;
+    int i = 0;
 
-    if (vhost_user_running(s)) {
+    if (s->running)
         return 0;
-    }
 

net/vhost-user.c typedef struct VhostUserChardevProps { · line 60

Quoted 10 lines
net/vhost-user.c +5 -2
     options.backend_type = VHOST_BACKEND_TYPE_USER;
-    options.net_backend = &s->nc;
     options.opaque = s->chr;
 
-    s->vhost_net = vhost_net_init(&options);
+    options.net_backend = s->peers[i].nc;
+    vhost_net = s->peers[i++].vhost_net = vhost_net_init(&options);
+
+    max_queues = vhost_net_get_max_queues(vhost_net);
+    if (s->queues >= max_queues) {

use '>' rather than '>=' here?

#19 lore raw

Re: [Qemu-devel] [PATCH 6/7] vhost-user: add multiple queue support

Quoted 32 lines · Changchun
Quoted 32 lines

-----Original Message-----
From: Yuanhan Liu [mailto:[email protected]]
Sent: Tuesday, September 8, 2015 3:39 PM
To: [email protected]
Cc: [email protected]; Ouyang, Changchun; Yuanhan Liu
Subject: [PATCH 6/7] vhost-user: add multiple queue support

From: Ouyang Changchun <[email protected]>

This patch is initially based a patch from Nikolay Nikolaev.

Here is the latest version for adding vhost-user multiple queue support, by creating a nc and vhost_net pair for each queue.

 static int vhost_user_start(VhostUserState *s)  {
     VhostNetOptions options;
net/vhost-user.c +4 -3
+    VHostNetState *vhost_net;
+    int max_queues;
+    int i = 0;
 
-    if (vhost_user_running(s)) {
+    if (s->running)
         return 0;
-    }
 
     options.backend_type = VHOST_BACKEND_TYPE_USER;
-    options.net_backend = &s->nc;
     options.opaque = s->chr;
Quoted 8 lines

net/vhost-user.c typedef struct VhostUserChardevProps { · line 60

Quoted 6 lines
net/vhost-user.c +5 -1
-    s->vhost_net = vhost_net_init(&options);
+    options.net_backend = s->peers[i].nc;
+    vhost_net = s->peers[i++].vhost_net = vhost_net_init(&options);
+
+    max_queues = vhost_net_get_max_queues(vhost_net);
+    if (s->queues >= max_queues) {

use '>' rather than '>=' here?

Right, and thanks!

	--yliu
#20 lore raw

Re: [Qemu-devel] [PATCH 6/7] vhost-user: add multiple queue support

In reply to Yuanhan Liu
Quoted 97 lines · Yuanhan Liu

From: Ouyang Changchun <[email protected]>

This patch is initially based a patch from Nikolay Nikolaev.

Here is the latest version for adding vhost-user multiple queue support, by creating a nc and vhost_net pair for each queue.

What differs from last version is that this patch addresses two major concerns from Michael, and fixes one hidden bug.

- Concern #1: no feedback when the backend can't support # of requested queues(by providing queues=# option).

Here we address this issue by querying VHOST_USER_PROTOCOL_F_MQ protocol features first, if not set, it means the backend don't support mq feature, and let max_queues be 1. Otherwise, we send another message, VHOST_USER_GET_QUEUE_NUM, for getting the max_queues the backend supports.

At vhost-user initiation stage(net_vhost_user_start), we then initiate one queue first, which, in the meantime, also gets the max_queues. We then do a simple compare: if requested_queues > max_queues, we exit(I guess it's safe to exit here, as the vm is not running yet).

- Concern #2: some messages are sent more times than necessary.

We came an agreement with Michael that we could categorize vhost user messages to 2 types: none-vring specific messages, which should be sent only once, and vring specific messages, which should be sent per queue.

Here I introduced a helper function vhost_user_one_time_request(), which lists following messages as none-vring specific messages:

VHOST_USER_GET_FEATURES
VHOST_USER_SET_FEATURES
VHOST_USER_GET_PROTOCOL_FEATURES
VHOST_USER_SET_PROTOCOL_FEATURES
VHOST_USER_SET_OWNER
VHOST_USER_RESET_DEVICE
VHOST_USER_SET_MEM_TABLE
VHOST_USER_GET_QUEUE_NUM

For above messages, we simply ignore them when they are not sent the first time.

I also observed a hidden bug from last version. We register the char dev event handler N times, which is not necessary, as well as buggy: A later register overwrites the former one, as qemu_chr_add_handlers() will not chain those handlers, but instead overwrites the old one. So, in theory, invoking qemu_chr_add_handlers N times will not end up with calling the handler N times.

However, the reason the handler is invoked N times is because we start the backend(the connection server) first, and hence when net_vhost_user_init() is executed, the connection is already established, and qemu_chr_add_handlers() then invoke the handler immediately, which just looks like we invoke the handler(net_vhost_user_event) directly from net_vhost_user_init().

The solution I came up with is to make VhostUserState as an upper level structure, making it includes N nc and vhost_net pairs:

   struct VhostUserNetPeer {
       NetClientState *nc;
       VHostNetState  *vhost_net;
   };

   typedef struct VhostUserState {
       CharDriverState *chr;
       bool running;
       int queues;
       struct VhostUserNetPeer peers[];
   } VhostUserState;
---
 docs/specs/vhost-user.txt |  13 +++++
 hw/virtio/vhost-user.c    |  31 ++++++++++-
 include/net/net.h         |   1 +
 net/vhost-user.c          | 136 ++++++++++++++++++++++++++++++++--------------
 qapi-schema.json          |   6 +-
 qemu-options.hx           |   5 +-
 6 files changed, 146 insertions(+), 46 deletions(-)
docs/specs/vhost-user.txt +3 -0
index 43db9b4..99d79be 100644
@@ -135,6 +135,19 @@ As older slaves don't support negotiating protocol features,
 a feature bit was dedicated for this purpose:
 #define VHOST_USER_F_PROTOCOL_FEATURES 30
 
+Multiple queue support
+-------------------
+Multiple queue is treated as a protocal extension, hence the slave has to

docs/specs/vhost-user.txt As older slaves don't support negotiating protocol features, · line 150

Quoted 10 lines
docs/specs/vhost-user.txt +10 -0
+implement protocol features first. Multiple queues is supported only when
+the protocol feature VHOST_USER_PROTOCOL_F_MQ(bit 0) is set.
+
+The max # of queues the slave support can be queried with message
+VHOST_USER_GET_PROTOCOL_FEATURES. Master should stop when the # of requested
+queues is bigger than that.
+
+As all queues share one connection, the master use a unique index for each
+queue in the sent message to identify one specified queue.
+

Please also document that only 2 queues are enabled initially. More queues are enabled dynamically.
To enable more queues, the new message needs to be sent (added by patch 7).

Quoted 376 lines
docs/specs/vhost-user.txt +0 -0
 Message types
 -------------
 
hw/virtio/vhost-user.c +30 -1
index 8046bc0..11e46b5 100644
@@ -187,6 +187,23 @@ static int vhost_user_write(struct vhost_dev *dev, VhostUserMsg *msg,
             0 : -1;
 }
 
+static bool vhost_user_one_time_request(VhostUserRequest request)
+{
+    switch (request) {
+    case VHOST_USER_GET_FEATURES:
+    case VHOST_USER_SET_FEATURES:
+    case VHOST_USER_GET_PROTOCOL_FEATURES:
+    case VHOST_USER_SET_PROTOCOL_FEATURES:
+    case VHOST_USER_SET_OWNER:
+    case VHOST_USER_RESET_DEVICE:
+    case VHOST_USER_SET_MEM_TABLE:
+    case VHOST_USER_GET_QUEUE_NUM:
+        return true;
+    default:
+        return false;
+    }
+}
+
 static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
         void *arg)
 {
@@ -206,6 +223,14 @@ static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
     else
         msg_request = request;
 
+    /*
+     * For none-vring specific requests, like VHOST_USER_GET_FEATURES,
+     * we just need send it once in the first time. For later such
+     * request, we just ignore it.
+     */
+    if (vhost_user_one_time_request(msg_request) && dev->vq_index != 0)
+        return 0;
+
     msg.request = msg_request;
     msg.flags = VHOST_USER_VERSION;
     msg.size = 0;
@@ -268,17 +293,20 @@ static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
     case VHOST_USER_SET_VRING_NUM:
     case VHOST_USER_SET_VRING_BASE:
         memcpy(&msg.state, arg, sizeof(struct vhost_vring_state));
+        msg.addr.index += dev->vq_index;
         msg.size = sizeof(m.state);
         break;
 
     case VHOST_USER_GET_VRING_BASE:
         memcpy(&msg.state, arg, sizeof(struct vhost_vring_state));
+        msg.addr.index += dev->vq_index;
         msg.size = sizeof(m.state);
         need_reply = 1;
         break;
 
     case VHOST_USER_SET_VRING_ADDR:
         memcpy(&msg.addr, arg, sizeof(struct vhost_vring_addr));
+        msg.addr.index += dev->vq_index;
         msg.size = sizeof(m.addr);
         break;
 
@@ -286,7 +314,7 @@ static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
     case VHOST_USER_SET_VRING_CALL:
     case VHOST_USER_SET_VRING_ERR:
         file = arg;
-        msg.u64 = file->index & VHOST_USER_VRING_IDX_MASK;
+        msg.u64 = (file->index + dev->vq_index) & VHOST_USER_VRING_IDX_MASK;
         msg.size = sizeof(m.u64);
         if (ioeventfd_enabled() && file->fd > 0) {
             fds[fd_num++] = file->fd;
@@ -330,6 +358,7 @@ static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
                 error_report("Received bad msg size.");
                 return -1;
             }
+            msg.state.index -= dev->vq_index;
             memcpy(arg, &msg.state, sizeof(struct vhost_vring_state));
             break;
         default:
include/net/net.h +1 -0
index 6a6cbef..6f20656 100644
@@ -92,6 +92,7 @@ struct NetClientState {
     NetClientDestructor *destructor;
     unsigned int queue_index;
     unsigned rxfilter_notify_enabled:1;
+    void *opaque;
 };
 
 typedef struct NICState {
net/vhost-user.c +94 -42
index 2d6bbe5..7d4ac69 100644
@@ -15,10 +15,16 @@
 #include "qemu/config-file.h"
 #include "qemu/error-report.h"
 
+struct VhostUserNetPeer {
+    NetClientState *nc;
+    VHostNetState  *vhost_net;
+};
+
 typedef struct VhostUserState {
-    NetClientState nc;
     CharDriverState *chr;
-    VHostNetState *vhost_net;
+    bool running;
+    int queues;
+    struct VhostUserNetPeer peers[];
 } VhostUserState;
 
 typedef struct VhostUserChardevProps {
@@ -29,48 +35,75 @@ typedef struct VhostUserChardevProps {
 
 VHostNetState *vhost_user_get_vhost_net(NetClientState *nc)
 {
-    VhostUserState *s = DO_UPCAST(VhostUserState, nc, nc);
+    VhostUserState *s = nc->opaque;
     assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_VHOST_USER);
-    return s->vhost_net;
-}
-
-static int vhost_user_running(VhostUserState *s)
-{
-    return (s->vhost_net) ? 1 : 0;
+    return s->peers[nc->queue_index].vhost_net;
 }
 
 static int vhost_user_start(VhostUserState *s)
 {
     VhostNetOptions options;
+    VHostNetState *vhost_net;
+    int max_queues;
+    int i = 0;
 
-    if (vhost_user_running(s)) {
+    if (s->running)
         return 0;
-    }
 
     options.backend_type = VHOST_BACKEND_TYPE_USER;
-    options.net_backend = &s->nc;
     options.opaque = s->chr;
 
-    s->vhost_net = vhost_net_init(&options);
+    options.net_backend = s->peers[i].nc;
+    vhost_net = s->peers[i++].vhost_net = vhost_net_init(&options);
+
+    max_queues = vhost_net_get_max_queues(vhost_net);
+    if (s->queues >= max_queues) {
+        error_report("you are asking more queues than supported: %d\n",
+                     max_queues);
+        return -1;
+    }
+
+    for (; i < s->queues; i++) {
+        options.net_backend = s->peers[i].nc;
+
+        s->peers[i].vhost_net = vhost_net_init(&options);
+        if (!s->peers[i].vhost_net)
+            return -1;
+    }
+    s->running = true;
 
-    return vhost_user_running(s) ? 0 : -1;
+    return 0;
 }
 
 static void vhost_user_stop(VhostUserState *s)
 {
-    if (vhost_user_running(s)) {
-        vhost_net_cleanup(s->vhost_net);
+    int i;
+    VHostNetState *vhost_net;
+
+    if (!s->running)
+        return;
+
+    for (i = 0;  i < s->queues; i++) {
+        vhost_net = s->peers[i].vhost_net;
+        if (vhost_net)
+            vhost_net_cleanup(vhost_net);
     }
 
-    s->vhost_net = 0;
+    s->running = false;
 }
 
 static void vhost_user_cleanup(NetClientState *nc)
 {
-    VhostUserState *s = DO_UPCAST(VhostUserState, nc, nc);
+    VhostUserState *s = nc->opaque;
+    VHostNetState *vhost_net = s->peers[nc->queue_index].vhost_net;
+
+    if (vhost_net)
+        vhost_net_cleanup(vhost_net);
 
-    vhost_user_stop(s);
     qemu_purge_queued_packets(nc);
+
+    if (nc->queue_index == s->queues - 1)
+        free(s);
 }
 
 static bool vhost_user_has_vnet_hdr(NetClientState *nc)
@@ -89,7 +122,7 @@ static bool vhost_user_has_ufo(NetClientState *nc)
 
 static NetClientInfo net_vhost_user_info = {
         .type = NET_CLIENT_OPTIONS_KIND_VHOST_USER,
-        .size = sizeof(VhostUserState),
+        .size = sizeof(NetClientState),
         .cleanup = vhost_user_cleanup,
         .has_vnet_hdr = vhost_user_has_vnet_hdr,
         .has_ufo = vhost_user_has_ufo,
@@ -97,18 +130,25 @@ static NetClientInfo net_vhost_user_info = {
 
 static void net_vhost_link_down(VhostUserState *s, bool link_down)
 {
-    s->nc.link_down = link_down;
+    NetClientState *nc;
+    int i;
 
-    if (s->nc.peer) {
-        s->nc.peer->link_down = link_down;
-    }
+    for (i = 0; i < s->queues; i++) {
+        nc = s->peers[i].nc;
 
-    if (s->nc.info->link_status_changed) {
-        s->nc.info->link_status_changed(&s->nc);
-    }
+        nc->link_down = link_down;
+
+        if (nc->peer) {
+            nc->peer->link_down = link_down;
+        }
 
-    if (s->nc.peer && s->nc.peer->info->link_status_changed) {
-        s->nc.peer->info->link_status_changed(s->nc.peer);
+        if (nc->info->link_status_changed) {
+            nc->info->link_status_changed(nc);
+        }
+
+        if (nc->peer && nc->peer->info->link_status_changed) {
+            nc->peer->info->link_status_changed(nc->peer);
+        }
     }
 }
 
@@ -118,7 +158,8 @@ static void net_vhost_user_event(void *opaque, int event)
 
     switch (event) {
     case CHR_EVENT_OPENED:
-        vhost_user_start(s);
+        if (vhost_user_start(s) < 0)
+            exit(1);
         net_vhost_link_down(s, false);
         error_report("chardev \"%s\" went up", s->chr->label);
         break;
@@ -131,24 +172,28 @@ static void net_vhost_user_event(void *opaque, int event)
 }
 
 static int net_vhost_user_init(NetClientState *peer, const char *device,
-                               const char *name, CharDriverState *chr)
+                               const char *name, VhostUserState *s)
 {
     NetClientState *nc;
-    VhostUserState *s;
+    CharDriverState *chr = s->chr;
+    int i;
 
-    nc = qemu_new_net_client(&net_vhost_user_info, peer, device, name);
+    for (i = 0; i < s->queues; i++) {
+        nc = qemu_new_net_client(&net_vhost_user_info, peer, device, name);
 
-    snprintf(nc->info_str, sizeof(nc->info_str), "vhost-user to %s",
-             chr->label);
+        snprintf(nc->info_str, sizeof(nc->info_str), "vhost-user%d to %s",
+                 i, chr->label);
 
-    s = DO_UPCAST(VhostUserState, nc, nc);
+        /* We don't provide a receive callback */
+        nc->receive_disabled = 1;
 
-    /* We don't provide a receive callback */
-    s->nc.receive_disabled = 1;
-    s->chr = chr;
-    nc->queue_index = 0;
+        nc->queue_index = i;
+        nc->opaque      = s;
 
-    qemu_chr_add_handlers(s->chr, NULL, NULL, net_vhost_user_event, s);
+        s->peers[i].nc = nc;
+    }
+
+    qemu_chr_add_handlers(chr, NULL, NULL, net_vhost_user_event, s);
 
     return 0;
 }
@@ -227,8 +272,10 @@ static int net_vhost_check_net(void *opaque, QemuOpts *opts, Error **errp)
 int net_init_vhost_user(const NetClientOptions *opts, const char *name,
                         NetClientState *peer, Error **errp)
 {
+    int queues;
     const NetdevVhostUserOptions *vhost_user_opts;
     CharDriverState *chr;
+    VhostUserState *s;
 
     assert(opts->kind == NET_CLIENT_OPTIONS_KIND_VHOST_USER);
     vhost_user_opts = opts->vhost_user;
@@ -244,6 +291,11 @@ int net_init_vhost_user(const NetClientOptions *opts, const char *name,
         return -1;
     }
 
+    queues = vhost_user_opts->has_queues ? vhost_user_opts->queues : 1;
+    s = g_malloc0(sizeof(VhostUserState) +
+                  queues * sizeof(struct VhostUserNetPeer));
+    s->queues = queues;
+    s->chr    = chr;
 
-    return net_vhost_user_init(peer, "vhost_user", name, chr);
+    return net_vhost_user_init(peer, "vhost_user", name, s);
 }
qapi-schema.json +5 -1
index 67fef37..55c33db 100644
@@ -2480,12 +2480,16 @@
 #
 # @vhostforce: #optional vhost on for non-MSIX virtio guests (default: false).
 #
+# @queues: #optional number of queues to be created for multiqueue vhost-user
+#          (default: 1) (Since 2.5)
+#
 # Since 2.1
 ##
 { 'struct': 'NetdevVhostUserOptions',
   'data': {
     'chardev':        'str',
-    '*vhostforce':    'bool' } }
+    '*vhostforce':    'bool',
+    '*queues':        'int' } }
 
 ##
 # @NetClientOptions
qemu-options.hx +3 -2
index 77f5853..5bfa7a3 100644
@@ -1963,13 +1963,14 @@ The hubport netdev lets you connect a NIC to a QEMU "vlan" instead of a single
 netdev.  @code{-net} and @code{-device} with parameter @option{vlan} create the
 required hub automatically.
 
-@item -netdev vhost-user,chardev=@var{id}[,vhostforce=on|off]
+@item -netdev vhost-user,chardev=@var{id}[,vhostforce=on|off][,queues=n]
 
 Establish a vhost-user netdev, backed by a chardev @var{id}. The chardev should
 be a unix domain socket backed one. The vhost-user uses a specifically defined
 protocol to pass vhost ioctl replacement messages to an application on the other
 end of the socket. On non-MSIX guests, the feature can be forced with
-@var{vhostforce}.
+@var{vhostforce}. Use 'queues=@var{n}' to specify the number of queues to
+be created for multiqueue vhost-user.
 
 Example:
 @example
Signature · 1 line
1.9.0
#21 lore raw

Re: [Qemu-devel] [PATCH 6/7] vhost-user: add multiple queue support

Quoted 105 lines · Michael S. Tsirkin
Quoted 104 lines · Yuanhan Liu

From: Ouyang Changchun <[email protected]>

This patch is initially based a patch from Nikolay Nikolaev.

Here is the latest version for adding vhost-user multiple queue support, by creating a nc and vhost_net pair for each queue.

What differs from last version is that this patch addresses two major concerns from Michael, and fixes one hidden bug.

- Concern #1: no feedback when the backend can't support # of requested queues(by providing queues=# option).

Here we address this issue by querying VHOST_USER_PROTOCOL_F_MQ protocol features first, if not set, it means the backend don't support mq feature, and let max_queues be 1. Otherwise, we send another message, VHOST_USER_GET_QUEUE_NUM, for getting the max_queues the backend supports.

At vhost-user initiation stage(net_vhost_user_start), we then initiate one queue first, which, in the meantime, also gets the max_queues. We then do a simple compare: if requested_queues > max_queues, we exit(I guess it's safe to exit here, as the vm is not running yet).

- Concern #2: some messages are sent more times than necessary.

We came an agreement with Michael that we could categorize vhost user messages to 2 types: none-vring specific messages, which should be sent only once, and vring specific messages, which should be sent per queue.

Here I introduced a helper function vhost_user_one_time_request(), which lists following messages as none-vring specific messages:

VHOST_USER_GET_FEATURES
VHOST_USER_SET_FEATURES
VHOST_USER_GET_PROTOCOL_FEATURES
VHOST_USER_SET_PROTOCOL_FEATURES
VHOST_USER_SET_OWNER
VHOST_USER_RESET_DEVICE
VHOST_USER_SET_MEM_TABLE
VHOST_USER_GET_QUEUE_NUM

For above messages, we simply ignore them when they are not sent the first time.

I also observed a hidden bug from last version. We register the char dev event handler N times, which is not necessary, as well as buggy: A later register overwrites the former one, as qemu_chr_add_handlers() will not chain those handlers, but instead overwrites the old one. So, in theory, invoking qemu_chr_add_handlers N times will not end up with calling the handler N times.

However, the reason the handler is invoked N times is because we start the backend(the connection server) first, and hence when net_vhost_user_init() is executed, the connection is already established, and qemu_chr_add_handlers() then invoke the handler immediately, which just looks like we invoke the handler(net_vhost_user_event) directly from net_vhost_user_init().

The solution I came up with is to make VhostUserState as an upper level structure, making it includes N nc and vhost_net pairs:

   struct VhostUserNetPeer {
       NetClientState *nc;
       VHostNetState  *vhost_net;
   };

   typedef struct VhostUserState {
       CharDriverState *chr;
       bool running;
       int queues;
       struct VhostUserNetPeer peers[];
   } VhostUserState;
---
 docs/specs/vhost-user.txt |  13 +++++
 hw/virtio/vhost-user.c    |  31 ++++++++++-
 include/net/net.h         |   1 +
 net/vhost-user.c          | 136 ++++++++++++++++++++++++++++++++--------------
 qapi-schema.json          |   6 +-
 qemu-options.hx           |   5 +-
 6 files changed, 146 insertions(+), 46 deletions(-)
docs/specs/vhost-user.txt +10 -0
index 43db9b4..99d79be 100644
@@ -135,6 +135,19 @@ As older slaves don't support negotiating protocol features,
 a feature bit was dedicated for this purpose:
 #define VHOST_USER_F_PROTOCOL_FEATURES 30
 
+Multiple queue support
+-------------------
+Multiple queue is treated as a protocal extension, hence the slave has to
+implement protocol features first. Multiple queues is supported only when
+the protocol feature VHOST_USER_PROTOCOL_F_MQ(bit 0) is set.
+
+The max # of queues the slave support can be queried with message
+VHOST_USER_GET_PROTOCOL_FEATURES. Master should stop when the # of requested
+queues is bigger than that.
+
Quoted 8 lines

docs/specs/vhost-user.txt As older slaves don't support negotiating protocol features, · line 150

Quoted 3 lines
docs/specs/vhost-user.txt +3 -0
+As all queues share one connection, the master use a unique index for each
+queue in the sent message to identify one specified queue.
+

Please also document that only 2 queues are enabled initially. More queues are enabled dynamically.
To enable more queues, the new message needs to be sent (added by patch 7).

Hi Michael,

Will do that in next version. Besides that, do you have more comments? Say, does this patch address your two concerns?

	--yliu
Quoted 377 lines
Quoted 377 lines
docs/specs/vhost-user.txt +0 -0
 Message types
 -------------
 
hw/virtio/vhost-user.c +30 -1
index 8046bc0..11e46b5 100644
@@ -187,6 +187,23 @@ static int vhost_user_write(struct vhost_dev *dev, VhostUserMsg *msg,
             0 : -1;
 }
 
+static bool vhost_user_one_time_request(VhostUserRequest request)
+{
+    switch (request) {
+    case VHOST_USER_GET_FEATURES:
+    case VHOST_USER_SET_FEATURES:
+    case VHOST_USER_GET_PROTOCOL_FEATURES:
+    case VHOST_USER_SET_PROTOCOL_FEATURES:
+    case VHOST_USER_SET_OWNER:
+    case VHOST_USER_RESET_DEVICE:
+    case VHOST_USER_SET_MEM_TABLE:
+    case VHOST_USER_GET_QUEUE_NUM:
+        return true;
+    default:
+        return false;
+    }
+}
+
 static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
         void *arg)
 {
@@ -206,6 +223,14 @@ static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
     else
         msg_request = request;
 
+    /*
+     * For none-vring specific requests, like VHOST_USER_GET_FEATURES,
+     * we just need send it once in the first time. For later such
+     * request, we just ignore it.
+     */
+    if (vhost_user_one_time_request(msg_request) && dev->vq_index != 0)
+        return 0;
+
     msg.request = msg_request;
     msg.flags = VHOST_USER_VERSION;
     msg.size = 0;
@@ -268,17 +293,20 @@ static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
     case VHOST_USER_SET_VRING_NUM:
     case VHOST_USER_SET_VRING_BASE:
         memcpy(&msg.state, arg, sizeof(struct vhost_vring_state));
+        msg.addr.index += dev->vq_index;
         msg.size = sizeof(m.state);
         break;
 
     case VHOST_USER_GET_VRING_BASE:
         memcpy(&msg.state, arg, sizeof(struct vhost_vring_state));
+        msg.addr.index += dev->vq_index;
         msg.size = sizeof(m.state);
         need_reply = 1;
         break;
 
     case VHOST_USER_SET_VRING_ADDR:
         memcpy(&msg.addr, arg, sizeof(struct vhost_vring_addr));
+        msg.addr.index += dev->vq_index;
         msg.size = sizeof(m.addr);
         break;
 
@@ -286,7 +314,7 @@ static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
     case VHOST_USER_SET_VRING_CALL:
     case VHOST_USER_SET_VRING_ERR:
         file = arg;
-        msg.u64 = file->index & VHOST_USER_VRING_IDX_MASK;
+        msg.u64 = (file->index + dev->vq_index) & VHOST_USER_VRING_IDX_MASK;
         msg.size = sizeof(m.u64);
         if (ioeventfd_enabled() && file->fd > 0) {
             fds[fd_num++] = file->fd;
@@ -330,6 +358,7 @@ static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
                 error_report("Received bad msg size.");
                 return -1;
             }
+            msg.state.index -= dev->vq_index;
             memcpy(arg, &msg.state, sizeof(struct vhost_vring_state));
             break;
         default:
include/net/net.h +1 -0
index 6a6cbef..6f20656 100644
@@ -92,6 +92,7 @@ struct NetClientState {
     NetClientDestructor *destructor;
     unsigned int queue_index;
     unsigned rxfilter_notify_enabled:1;
+    void *opaque;
 };
 
 typedef struct NICState {
net/vhost-user.c +94 -42
index 2d6bbe5..7d4ac69 100644
@@ -15,10 +15,16 @@
 #include "qemu/config-file.h"
 #include "qemu/error-report.h"
 
+struct VhostUserNetPeer {
+    NetClientState *nc;
+    VHostNetState  *vhost_net;
+};
+
 typedef struct VhostUserState {
-    NetClientState nc;
     CharDriverState *chr;
-    VHostNetState *vhost_net;
+    bool running;
+    int queues;
+    struct VhostUserNetPeer peers[];
 } VhostUserState;
 
 typedef struct VhostUserChardevProps {
@@ -29,48 +35,75 @@ typedef struct VhostUserChardevProps {
 
 VHostNetState *vhost_user_get_vhost_net(NetClientState *nc)
 {
-    VhostUserState *s = DO_UPCAST(VhostUserState, nc, nc);
+    VhostUserState *s = nc->opaque;
     assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_VHOST_USER);
-    return s->vhost_net;
-}
-
-static int vhost_user_running(VhostUserState *s)
-{
-    return (s->vhost_net) ? 1 : 0;
+    return s->peers[nc->queue_index].vhost_net;
 }
 
 static int vhost_user_start(VhostUserState *s)
 {
     VhostNetOptions options;
+    VHostNetState *vhost_net;
+    int max_queues;
+    int i = 0;
 
-    if (vhost_user_running(s)) {
+    if (s->running)
         return 0;
-    }
 
     options.backend_type = VHOST_BACKEND_TYPE_USER;
-    options.net_backend = &s->nc;
     options.opaque = s->chr;
 
-    s->vhost_net = vhost_net_init(&options);
+    options.net_backend = s->peers[i].nc;
+    vhost_net = s->peers[i++].vhost_net = vhost_net_init(&options);
+
+    max_queues = vhost_net_get_max_queues(vhost_net);
+    if (s->queues >= max_queues) {
+        error_report("you are asking more queues than supported: %d\n",
+                     max_queues);
+        return -1;
+    }
+
+    for (; i < s->queues; i++) {
+        options.net_backend = s->peers[i].nc;
+
+        s->peers[i].vhost_net = vhost_net_init(&options);
+        if (!s->peers[i].vhost_net)
+            return -1;
+    }
+    s->running = true;
 
-    return vhost_user_running(s) ? 0 : -1;
+    return 0;
 }
 
 static void vhost_user_stop(VhostUserState *s)
 {
-    if (vhost_user_running(s)) {
-        vhost_net_cleanup(s->vhost_net);
+    int i;
+    VHostNetState *vhost_net;
+
+    if (!s->running)
+        return;
+
+    for (i = 0;  i < s->queues; i++) {
+        vhost_net = s->peers[i].vhost_net;
+        if (vhost_net)
+            vhost_net_cleanup(vhost_net);
     }
 
-    s->vhost_net = 0;
+    s->running = false;
 }
 
 static void vhost_user_cleanup(NetClientState *nc)
 {
-    VhostUserState *s = DO_UPCAST(VhostUserState, nc, nc);
+    VhostUserState *s = nc->opaque;
+    VHostNetState *vhost_net = s->peers[nc->queue_index].vhost_net;
+
+    if (vhost_net)
+        vhost_net_cleanup(vhost_net);
 
-    vhost_user_stop(s);
     qemu_purge_queued_packets(nc);
+
+    if (nc->queue_index == s->queues - 1)
+        free(s);
 }
 
 static bool vhost_user_has_vnet_hdr(NetClientState *nc)
@@ -89,7 +122,7 @@ static bool vhost_user_has_ufo(NetClientState *nc)
 
 static NetClientInfo net_vhost_user_info = {
         .type = NET_CLIENT_OPTIONS_KIND_VHOST_USER,
-        .size = sizeof(VhostUserState),
+        .size = sizeof(NetClientState),
         .cleanup = vhost_user_cleanup,
         .has_vnet_hdr = vhost_user_has_vnet_hdr,
         .has_ufo = vhost_user_has_ufo,
@@ -97,18 +130,25 @@ static NetClientInfo net_vhost_user_info = {
 
 static void net_vhost_link_down(VhostUserState *s, bool link_down)
 {
-    s->nc.link_down = link_down;
+    NetClientState *nc;
+    int i;
 
-    if (s->nc.peer) {
-        s->nc.peer->link_down = link_down;
-    }
+    for (i = 0; i < s->queues; i++) {
+        nc = s->peers[i].nc;
 
-    if (s->nc.info->link_status_changed) {
-        s->nc.info->link_status_changed(&s->nc);
-    }
+        nc->link_down = link_down;
+
+        if (nc->peer) {
+            nc->peer->link_down = link_down;
+        }
 
-    if (s->nc.peer && s->nc.peer->info->link_status_changed) {
-        s->nc.peer->info->link_status_changed(s->nc.peer);
+        if (nc->info->link_status_changed) {
+            nc->info->link_status_changed(nc);
+        }
+
+        if (nc->peer && nc->peer->info->link_status_changed) {
+            nc->peer->info->link_status_changed(nc->peer);
+        }
     }
 }
 
@@ -118,7 +158,8 @@ static void net_vhost_user_event(void *opaque, int event)
 
     switch (event) {
     case CHR_EVENT_OPENED:
-        vhost_user_start(s);
+        if (vhost_user_start(s) < 0)
+            exit(1);
         net_vhost_link_down(s, false);
         error_report("chardev \"%s\" went up", s->chr->label);
         break;
@@ -131,24 +172,28 @@ static void net_vhost_user_event(void *opaque, int event)
 }
 
 static int net_vhost_user_init(NetClientState *peer, const char *device,
-                               const char *name, CharDriverState *chr)
+                               const char *name, VhostUserState *s)
 {
     NetClientState *nc;
-    VhostUserState *s;
+    CharDriverState *chr = s->chr;
+    int i;
 
-    nc = qemu_new_net_client(&net_vhost_user_info, peer, device, name);
+    for (i = 0; i < s->queues; i++) {
+        nc = qemu_new_net_client(&net_vhost_user_info, peer, device, name);
 
-    snprintf(nc->info_str, sizeof(nc->info_str), "vhost-user to %s",
-             chr->label);
+        snprintf(nc->info_str, sizeof(nc->info_str), "vhost-user%d to %s",
+                 i, chr->label);
 
-    s = DO_UPCAST(VhostUserState, nc, nc);
+        /* We don't provide a receive callback */
+        nc->receive_disabled = 1;
 
-    /* We don't provide a receive callback */
-    s->nc.receive_disabled = 1;
-    s->chr = chr;
-    nc->queue_index = 0;
+        nc->queue_index = i;
+        nc->opaque      = s;
 
-    qemu_chr_add_handlers(s->chr, NULL, NULL, net_vhost_user_event, s);
+        s->peers[i].nc = nc;
+    }
+
+    qemu_chr_add_handlers(chr, NULL, NULL, net_vhost_user_event, s);
 
     return 0;
 }
@@ -227,8 +272,10 @@ static int net_vhost_check_net(void *opaque, QemuOpts *opts, Error **errp)
 int net_init_vhost_user(const NetClientOptions *opts, const char *name,
                         NetClientState *peer, Error **errp)
 {
+    int queues;
     const NetdevVhostUserOptions *vhost_user_opts;
     CharDriverState *chr;
+    VhostUserState *s;
 
     assert(opts->kind == NET_CLIENT_OPTIONS_KIND_VHOST_USER);
     vhost_user_opts = opts->vhost_user;
@@ -244,6 +291,11 @@ int net_init_vhost_user(const NetClientOptions *opts, const char *name,
         return -1;
     }
 
+    queues = vhost_user_opts->has_queues ? vhost_user_opts->queues : 1;
+    s = g_malloc0(sizeof(VhostUserState) +
+                  queues * sizeof(struct VhostUserNetPeer));
+    s->queues = queues;
+    s->chr    = chr;
 
-    return net_vhost_user_init(peer, "vhost_user", name, chr);
+    return net_vhost_user_init(peer, "vhost_user", name, s);
 }
qapi-schema.json +5 -1
index 67fef37..55c33db 100644
@@ -2480,12 +2480,16 @@
 #
 # @vhostforce: #optional vhost on for non-MSIX virtio guests (default: false).
 #
+# @queues: #optional number of queues to be created for multiqueue vhost-user
+#          (default: 1) (Since 2.5)
+#
 # Since 2.1
 ##
 { 'struct': 'NetdevVhostUserOptions',
   'data': {
     'chardev':        'str',
-    '*vhostforce':    'bool' } }
+    '*vhostforce':    'bool',
+    '*queues':        'int' } }
 
 ##
 # @NetClientOptions
qemu-options.hx +3 -2
index 77f5853..5bfa7a3 100644
@@ -1963,13 +1963,14 @@ The hubport netdev lets you connect a NIC to a QEMU "vlan" instead of a single
 netdev.  @code{-net} and @code{-device} with parameter @option{vlan} create the
 required hub automatically.
 
-@item -netdev vhost-user,chardev=@var{id}[,vhostforce=on|off]
+@item -netdev vhost-user,chardev=@var{id}[,vhostforce=on|off][,queues=n]
 
 Establish a vhost-user netdev, backed by a chardev @var{id}. The chardev should
 be a unix domain socket backed one. The vhost-user uses a specifically defined
 protocol to pass vhost ioctl replacement messages to an application on the other
 end of the socket. On non-MSIX guests, the feature can be forced with
-@var{vhostforce}.
+@var{vhostforce}. Use 'queues=@var{n}' to specify the number of queues to
+be created for multiqueue vhost-user.
 
 Example:
 @example
Signature · 1 line
1.9.0
#22 lore raw

Re: [Qemu-devel] [PATCH 6/7] vhost-user: add multiple queue support

In reply to Yuanhan Liu
Quoted 97 lines · Yuanhan Liu

From: Ouyang Changchun <[email protected]>

This patch is initially based a patch from Nikolay Nikolaev.

Here is the latest version for adding vhost-user multiple queue support, by creating a nc and vhost_net pair for each queue.

What differs from last version is that this patch addresses two major concerns from Michael, and fixes one hidden bug.

- Concern #1: no feedback when the backend can't support # of requested queues(by providing queues=# option).

Here we address this issue by querying VHOST_USER_PROTOCOL_F_MQ protocol features first, if not set, it means the backend don't support mq feature, and let max_queues be 1. Otherwise, we send another message, VHOST_USER_GET_QUEUE_NUM, for getting the max_queues the backend supports.

At vhost-user initiation stage(net_vhost_user_start), we then initiate one queue first, which, in the meantime, also gets the max_queues. We then do a simple compare: if requested_queues > max_queues, we exit(I guess it's safe to exit here, as the vm is not running yet).

- Concern #2: some messages are sent more times than necessary.

We came an agreement with Michael that we could categorize vhost user messages to 2 types: none-vring specific messages, which should be sent only once, and vring specific messages, which should be sent per queue.

Here I introduced a helper function vhost_user_one_time_request(), which lists following messages as none-vring specific messages:

VHOST_USER_GET_FEATURES
VHOST_USER_SET_FEATURES
VHOST_USER_GET_PROTOCOL_FEATURES
VHOST_USER_SET_PROTOCOL_FEATURES
VHOST_USER_SET_OWNER
VHOST_USER_RESET_DEVICE
VHOST_USER_SET_MEM_TABLE
VHOST_USER_GET_QUEUE_NUM

For above messages, we simply ignore them when they are not sent the first time.

I also observed a hidden bug from last version. We register the char dev event handler N times, which is not necessary, as well as buggy: A later register overwrites the former one, as qemu_chr_add_handlers() will not chain those handlers, but instead overwrites the old one. So, in theory, invoking qemu_chr_add_handlers N times will not end up with calling the handler N times.

However, the reason the handler is invoked N times is because we start the backend(the connection server) first, and hence when net_vhost_user_init() is executed, the connection is already established, and qemu_chr_add_handlers() then invoke the handler immediately, which just looks like we invoke the handler(net_vhost_user_event) directly from net_vhost_user_init().

The solution I came up with is to make VhostUserState as an upper level structure, making it includes N nc and vhost_net pairs:

   struct VhostUserNetPeer {
       NetClientState *nc;
       VHostNetState  *vhost_net;
   };

   typedef struct VhostUserState {
       CharDriverState *chr;
       bool running;
       int queues;
       struct VhostUserNetPeer peers[];
   } VhostUserState;
---
 docs/specs/vhost-user.txt |  13 +++++
 hw/virtio/vhost-user.c    |  31 ++++++++++-
 include/net/net.h         |   1 +
 net/vhost-user.c          | 136 ++++++++++++++++++++++++++++++++--------------
 qapi-schema.json          |   6 +-
 qemu-options.hx           |   5 +-
 6 files changed, 146 insertions(+), 46 deletions(-)
docs/specs/vhost-user.txt +3 -0
index 43db9b4..99d79be 100644
@@ -135,6 +135,19 @@ As older slaves don't support negotiating protocol features,
 a feature bit was dedicated for this purpose:
 #define VHOST_USER_F_PROTOCOL_FEATURES 30
 
+Multiple queue support
+-------------------
+Multiple queue is treated as a protocal extension, hence the slave has to

docs/specs/vhost-user.txt As older slaves don't support negotiating protocol features, · line 150

Quoted 10 lines
docs/specs/vhost-user.txt +10 -0
+implement protocol features first. Multiple queues is supported only when
+the protocol feature VHOST_USER_PROTOCOL_F_MQ(bit 0) is set.
+
+The max # of queues the slave support can be queried with message
+VHOST_USER_GET_PROTOCOL_FEATURES. Master should stop when the # of requested
+queues is bigger than that.
+
+As all queues share one connection, the master use a unique index for each
+queue in the sent message to identify one specified queue.
+

Please also document that only 2 queues are enabled initially. More queues are enabled dynamically.
To enable more queues, the new message needs to be sent (added by patch 7).

Quoted 376 lines
docs/specs/vhost-user.txt +0 -0
 Message types
 -------------
 
hw/virtio/vhost-user.c +30 -1
index 8046bc0..11e46b5 100644
@@ -187,6 +187,23 @@ static int vhost_user_write(struct vhost_dev *dev, VhostUserMsg *msg,
             0 : -1;
 }
 
+static bool vhost_user_one_time_request(VhostUserRequest request)
+{
+    switch (request) {
+    case VHOST_USER_GET_FEATURES:
+    case VHOST_USER_SET_FEATURES:
+    case VHOST_USER_GET_PROTOCOL_FEATURES:
+    case VHOST_USER_SET_PROTOCOL_FEATURES:
+    case VHOST_USER_SET_OWNER:
+    case VHOST_USER_RESET_DEVICE:
+    case VHOST_USER_SET_MEM_TABLE:
+    case VHOST_USER_GET_QUEUE_NUM:
+        return true;
+    default:
+        return false;
+    }
+}
+
 static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
         void *arg)
 {
@@ -206,6 +223,14 @@ static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
     else
         msg_request = request;
 
+    /*
+     * For none-vring specific requests, like VHOST_USER_GET_FEATURES,
+     * we just need send it once in the first time. For later such
+     * request, we just ignore it.
+     */
+    if (vhost_user_one_time_request(msg_request) && dev->vq_index != 0)
+        return 0;
+
     msg.request = msg_request;
     msg.flags = VHOST_USER_VERSION;
     msg.size = 0;
@@ -268,17 +293,20 @@ static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
     case VHOST_USER_SET_VRING_NUM:
     case VHOST_USER_SET_VRING_BASE:
         memcpy(&msg.state, arg, sizeof(struct vhost_vring_state));
+        msg.addr.index += dev->vq_index;
         msg.size = sizeof(m.state);
         break;
 
     case VHOST_USER_GET_VRING_BASE:
         memcpy(&msg.state, arg, sizeof(struct vhost_vring_state));
+        msg.addr.index += dev->vq_index;
         msg.size = sizeof(m.state);
         need_reply = 1;
         break;
 
     case VHOST_USER_SET_VRING_ADDR:
         memcpy(&msg.addr, arg, sizeof(struct vhost_vring_addr));
+        msg.addr.index += dev->vq_index;
         msg.size = sizeof(m.addr);
         break;
 
@@ -286,7 +314,7 @@ static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
     case VHOST_USER_SET_VRING_CALL:
     case VHOST_USER_SET_VRING_ERR:
         file = arg;
-        msg.u64 = file->index & VHOST_USER_VRING_IDX_MASK;
+        msg.u64 = (file->index + dev->vq_index) & VHOST_USER_VRING_IDX_MASK;
         msg.size = sizeof(m.u64);
         if (ioeventfd_enabled() && file->fd > 0) {
             fds[fd_num++] = file->fd;
@@ -330,6 +358,7 @@ static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
                 error_report("Received bad msg size.");
                 return -1;
             }
+            msg.state.index -= dev->vq_index;
             memcpy(arg, &msg.state, sizeof(struct vhost_vring_state));
             break;
         default:
include/net/net.h +1 -0
index 6a6cbef..6f20656 100644
@@ -92,6 +92,7 @@ struct NetClientState {
     NetClientDestructor *destructor;
     unsigned int queue_index;
     unsigned rxfilter_notify_enabled:1;
+    void *opaque;
 };
 
 typedef struct NICState {
net/vhost-user.c +94 -42
index 2d6bbe5..7d4ac69 100644
@@ -15,10 +15,16 @@
 #include "qemu/config-file.h"
 #include "qemu/error-report.h"
 
+struct VhostUserNetPeer {
+    NetClientState *nc;
+    VHostNetState  *vhost_net;
+};
+
 typedef struct VhostUserState {
-    NetClientState nc;
     CharDriverState *chr;
-    VHostNetState *vhost_net;
+    bool running;
+    int queues;
+    struct VhostUserNetPeer peers[];
 } VhostUserState;
 
 typedef struct VhostUserChardevProps {
@@ -29,48 +35,75 @@ typedef struct VhostUserChardevProps {
 
 VHostNetState *vhost_user_get_vhost_net(NetClientState *nc)
 {
-    VhostUserState *s = DO_UPCAST(VhostUserState, nc, nc);
+    VhostUserState *s = nc->opaque;
     assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_VHOST_USER);
-    return s->vhost_net;
-}
-
-static int vhost_user_running(VhostUserState *s)
-{
-    return (s->vhost_net) ? 1 : 0;
+    return s->peers[nc->queue_index].vhost_net;
 }
 
 static int vhost_user_start(VhostUserState *s)
 {
     VhostNetOptions options;
+    VHostNetState *vhost_net;
+    int max_queues;
+    int i = 0;
 
-    if (vhost_user_running(s)) {
+    if (s->running)
         return 0;
-    }
 
     options.backend_type = VHOST_BACKEND_TYPE_USER;
-    options.net_backend = &s->nc;
     options.opaque = s->chr;
 
-    s->vhost_net = vhost_net_init(&options);
+    options.net_backend = s->peers[i].nc;
+    vhost_net = s->peers[i++].vhost_net = vhost_net_init(&options);
+
+    max_queues = vhost_net_get_max_queues(vhost_net);
+    if (s->queues >= max_queues) {
+        error_report("you are asking more queues than supported: %d\n",
+                     max_queues);
+        return -1;
+    }
+
+    for (; i < s->queues; i++) {
+        options.net_backend = s->peers[i].nc;
+
+        s->peers[i].vhost_net = vhost_net_init(&options);
+        if (!s->peers[i].vhost_net)
+            return -1;
+    }
+    s->running = true;
 
-    return vhost_user_running(s) ? 0 : -1;
+    return 0;
 }
 
 static void vhost_user_stop(VhostUserState *s)
 {
-    if (vhost_user_running(s)) {
-        vhost_net_cleanup(s->vhost_net);
+    int i;
+    VHostNetState *vhost_net;
+
+    if (!s->running)
+        return;
+
+    for (i = 0;  i < s->queues; i++) {
+        vhost_net = s->peers[i].vhost_net;
+        if (vhost_net)
+            vhost_net_cleanup(vhost_net);
     }
 
-    s->vhost_net = 0;
+    s->running = false;
 }
 
 static void vhost_user_cleanup(NetClientState *nc)
 {
-    VhostUserState *s = DO_UPCAST(VhostUserState, nc, nc);
+    VhostUserState *s = nc->opaque;
+    VHostNetState *vhost_net = s->peers[nc->queue_index].vhost_net;
+
+    if (vhost_net)
+        vhost_net_cleanup(vhost_net);
 
-    vhost_user_stop(s);
     qemu_purge_queued_packets(nc);
+
+    if (nc->queue_index == s->queues - 1)
+        free(s);
 }
 
 static bool vhost_user_has_vnet_hdr(NetClientState *nc)
@@ -89,7 +122,7 @@ static bool vhost_user_has_ufo(NetClientState *nc)
 
 static NetClientInfo net_vhost_user_info = {
         .type = NET_CLIENT_OPTIONS_KIND_VHOST_USER,
-        .size = sizeof(VhostUserState),
+        .size = sizeof(NetClientState),
         .cleanup = vhost_user_cleanup,
         .has_vnet_hdr = vhost_user_has_vnet_hdr,
         .has_ufo = vhost_user_has_ufo,
@@ -97,18 +130,25 @@ static NetClientInfo net_vhost_user_info = {
 
 static void net_vhost_link_down(VhostUserState *s, bool link_down)
 {
-    s->nc.link_down = link_down;
+    NetClientState *nc;
+    int i;
 
-    if (s->nc.peer) {
-        s->nc.peer->link_down = link_down;
-    }
+    for (i = 0; i < s->queues; i++) {
+        nc = s->peers[i].nc;
 
-    if (s->nc.info->link_status_changed) {
-        s->nc.info->link_status_changed(&s->nc);
-    }
+        nc->link_down = link_down;
+
+        if (nc->peer) {
+            nc->peer->link_down = link_down;
+        }
 
-    if (s->nc.peer && s->nc.peer->info->link_status_changed) {
-        s->nc.peer->info->link_status_changed(s->nc.peer);
+        if (nc->info->link_status_changed) {
+            nc->info->link_status_changed(nc);
+        }
+
+        if (nc->peer && nc->peer->info->link_status_changed) {
+            nc->peer->info->link_status_changed(nc->peer);
+        }
     }
 }
 
@@ -118,7 +158,8 @@ static void net_vhost_user_event(void *opaque, int event)
 
     switch (event) {
     case CHR_EVENT_OPENED:
-        vhost_user_start(s);
+        if (vhost_user_start(s) < 0)
+            exit(1);
         net_vhost_link_down(s, false);
         error_report("chardev \"%s\" went up", s->chr->label);
         break;
@@ -131,24 +172,28 @@ static void net_vhost_user_event(void *opaque, int event)
 }
 
 static int net_vhost_user_init(NetClientState *peer, const char *device,
-                               const char *name, CharDriverState *chr)
+                               const char *name, VhostUserState *s)
 {
     NetClientState *nc;
-    VhostUserState *s;
+    CharDriverState *chr = s->chr;
+    int i;
 
-    nc = qemu_new_net_client(&net_vhost_user_info, peer, device, name);
+    for (i = 0; i < s->queues; i++) {
+        nc = qemu_new_net_client(&net_vhost_user_info, peer, device, name);
 
-    snprintf(nc->info_str, sizeof(nc->info_str), "vhost-user to %s",
-             chr->label);
+        snprintf(nc->info_str, sizeof(nc->info_str), "vhost-user%d to %s",
+                 i, chr->label);
 
-    s = DO_UPCAST(VhostUserState, nc, nc);
+        /* We don't provide a receive callback */
+        nc->receive_disabled = 1;
 
-    /* We don't provide a receive callback */
-    s->nc.receive_disabled = 1;
-    s->chr = chr;
-    nc->queue_index = 0;
+        nc->queue_index = i;
+        nc->opaque      = s;
 
-    qemu_chr_add_handlers(s->chr, NULL, NULL, net_vhost_user_event, s);
+        s->peers[i].nc = nc;
+    }
+
+    qemu_chr_add_handlers(chr, NULL, NULL, net_vhost_user_event, s);
 
     return 0;
 }
@@ -227,8 +272,10 @@ static int net_vhost_check_net(void *opaque, QemuOpts *opts, Error **errp)
 int net_init_vhost_user(const NetClientOptions *opts, const char *name,
                         NetClientState *peer, Error **errp)
 {
+    int queues;
     const NetdevVhostUserOptions *vhost_user_opts;
     CharDriverState *chr;
+    VhostUserState *s;
 
     assert(opts->kind == NET_CLIENT_OPTIONS_KIND_VHOST_USER);
     vhost_user_opts = opts->vhost_user;
@@ -244,6 +291,11 @@ int net_init_vhost_user(const NetClientOptions *opts, const char *name,
         return -1;
     }
 
+    queues = vhost_user_opts->has_queues ? vhost_user_opts->queues : 1;
+    s = g_malloc0(sizeof(VhostUserState) +
+                  queues * sizeof(struct VhostUserNetPeer));
+    s->queues = queues;
+    s->chr    = chr;
 
-    return net_vhost_user_init(peer, "vhost_user", name, chr);
+    return net_vhost_user_init(peer, "vhost_user", name, s);
 }
qapi-schema.json +5 -1
index 67fef37..55c33db 100644
@@ -2480,12 +2480,16 @@
 #
 # @vhostforce: #optional vhost on for non-MSIX virtio guests (default: false).
 #
+# @queues: #optional number of queues to be created for multiqueue vhost-user
+#          (default: 1) (Since 2.5)
+#
 # Since 2.1
 ##
 { 'struct': 'NetdevVhostUserOptions',
   'data': {
     'chardev':        'str',
-    '*vhostforce':    'bool' } }
+    '*vhostforce':    'bool',
+    '*queues':        'int' } }
 
 ##
 # @NetClientOptions
qemu-options.hx +3 -2
index 77f5853..5bfa7a3 100644
@@ -1963,13 +1963,14 @@ The hubport netdev lets you connect a NIC to a QEMU "vlan" instead of a single
 netdev.  @code{-net} and @code{-device} with parameter @option{vlan} create the
 required hub automatically.
 
-@item -netdev vhost-user,chardev=@var{id}[,vhostforce=on|off]
+@item -netdev vhost-user,chardev=@var{id}[,vhostforce=on|off][,queues=n]
 
 Establish a vhost-user netdev, backed by a chardev @var{id}. The chardev should
 be a unix domain socket backed one. The vhost-user uses a specifically defined
 protocol to pass vhost ioctl replacement messages to an application on the other
 end of the socket. On non-MSIX guests, the feature can be forced with
-@var{vhostforce}.
+@var{vhostforce}. Use 'queues=@var{n}' to specify the number of queues to
+be created for multiqueue vhost-user.
 
 Example:
 @example
Signature · 1 line
1.9.0
#23 lore raw

Re: [Qemu-devel] [PATCH 6/7] vhost-user: add multiple queue support

In reply to Yuanhan Liu
Quoted 62 lines · Yuanhan Liu

From: Ouyang Changchun <[email protected]>

This patch is initially based a patch from Nikolay Nikolaev.

Here is the latest version for adding vhost-user multiple queue support, by creating a nc and vhost_net pair for each queue.

What differs from last version is that this patch addresses two major concerns from Michael, and fixes one hidden bug.

- Concern #1: no feedback when the backend can't support # of requested queues(by providing queues=# option).

Here we address this issue by querying VHOST_USER_PROTOCOL_F_MQ protocol features first, if not set, it means the backend don't support mq feature, and let max_queues be 1. Otherwise, we send another message, VHOST_USER_GET_QUEUE_NUM, for getting the max_queues the backend supports.

At vhost-user initiation stage(net_vhost_user_start), we then initiate one queue first, which, in the meantime, also gets the max_queues. We then do a simple compare: if requested_queues > max_queues, we exit(I guess it's safe to exit here, as the vm is not running yet).

- Concern #2: some messages are sent more times than necessary.

We came an agreement with Michael that we could categorize vhost user messages to 2 types: none-vring specific messages, which should be sent only once, and vring specific messages, which should be sent per queue.

Here I introduced a helper function vhost_user_one_time_request(), which lists following messages as none-vring specific messages:

VHOST_USER_GET_FEATURES
VHOST_USER_SET_FEATURES
VHOST_USER_GET_PROTOCOL_FEATURES
VHOST_USER_SET_PROTOCOL_FEATURES
VHOST_USER_SET_OWNER
VHOST_USER_RESET_DEVICE
VHOST_USER_SET_MEM_TABLE
VHOST_USER_GET_QUEUE_NUM

For above messages, we simply ignore them when they are not sent the first time.

I also observed a hidden bug from last version. We register the char dev event handler N times, which is not necessary, as well as buggy: A later register overwrites the former one, as qemu_chr_add_handlers() will not chain those handlers, but instead overwrites the old one. So, in theory, invoking qemu_chr_add_handlers N times will not end up with calling the handler N times.

However, the reason the handler is invoked N times is because we start the backend(the connection server) first, and hence when net_vhost_user_init() is executed, the connection is already established, and qemu_chr_add_handlers() then invoke the handler immediately, which just looks like we invoke the handler(net_vhost_user_event) directly from net_vhost_user_init().

The solution I came up with is to make VhostUserState as an upper level structure, making it includes N nc and vhost_net pairs:

Quoted 11 lines
   struct VhostUserNetPeer {
       NetClientState *nc;
       VHostNetState  *vhost_net;
   };

   typedef struct VhostUserState {
       CharDriverState *chr;
       bool running;
       int queues;
       struct VhostUserNetPeer peers[];
   } VhostUserState;

Haven't looked at previous versions but alternatively, can you do this by:

- keep current VhostUserState
- allocate multiple VhostUserState
- register the char handler only on VhostUserState[0]
- enumerate all others in handler by qemu_find_net_clients_except() ?

Quoted 81 lines
---
 docs/specs/vhost-user.txt |  13 +++++
 hw/virtio/vhost-user.c    |  31 ++++++++++-
 include/net/net.h         |   1 +
 net/vhost-user.c          | 136 ++++++++++++++++++++++++++++++++--------------
 qapi-schema.json          |   6 +-
 qemu-options.hx           |   5 +-
 6 files changed, 146 insertions(+), 46 deletions(-)
docs/specs/vhost-user.txt +13 -0
index 43db9b4..99d79be 100644
@@ -135,6 +135,19 @@ As older slaves don't support negotiating protocol features,
 a feature bit was dedicated for this purpose:
 #define VHOST_USER_F_PROTOCOL_FEATURES 30
 
+Multiple queue support
+-------------------
+Multiple queue is treated as a protocal extension, hence the slave has to
+implement protocol features first. Multiple queues is supported only when
+the protocol feature VHOST_USER_PROTOCOL_F_MQ(bit 0) is set.
+
+The max # of queues the slave support can be queried with message
+VHOST_USER_GET_PROTOCOL_FEATURES. Master should stop when the # of requested
+queues is bigger than that.
+
+As all queues share one connection, the master use a unique index for each
+queue in the sent message to identify one specified queue.
+
 Message types
 -------------
 
hw/virtio/vhost-user.c +25 -0
index 8046bc0..11e46b5 100644
@@ -187,6 +187,23 @@ static int vhost_user_write(struct vhost_dev *dev, VhostUserMsg *msg,
             0 : -1;
 }
 
+static bool vhost_user_one_time_request(VhostUserRequest request)
+{
+    switch (request) {
+    case VHOST_USER_GET_FEATURES:
+    case VHOST_USER_SET_FEATURES:
+    case VHOST_USER_GET_PROTOCOL_FEATURES:
+    case VHOST_USER_SET_PROTOCOL_FEATURES:
+    case VHOST_USER_SET_OWNER:
+    case VHOST_USER_RESET_DEVICE:
+    case VHOST_USER_SET_MEM_TABLE:
+    case VHOST_USER_GET_QUEUE_NUM:
+        return true;
+    default:
+        return false;
+    }
+}
+
 static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
         void *arg)
 {
@@ -206,6 +223,14 @@ static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
     else
         msg_request = request;
 
+    /*
+     * For none-vring specific requests, like VHOST_USER_GET_FEATURES,
+     * we just need send it once in the first time. For later such
+     * request, we just ignore it.
+     */
+    if (vhost_user_one_time_request(msg_request) && dev->vq_index != 0)
+        return 0;
+
     msg.request = msg_request;
     msg.flags = VHOST_USER_VERSION;
     msg.size = 0;

hw/virtio/vhost-user.c static int vhost_user_call(struct vhost_dev *dev, unsigned long int request, · line 302

Quoted 10 lines
hw/virtio/vhost-user.c +2 -0
@@ -268,17 +293,20 @@ static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
     case VHOST_USER_SET_VRING_NUM:
     case VHOST_USER_SET_VRING_BASE:
         memcpy(&msg.state, arg, sizeof(struct vhost_vring_state));
+        msg.addr.index += dev->vq_index;
         msg.size = sizeof(m.state);
         break;
 
     case VHOST_USER_GET_VRING_BASE:
         memcpy(&msg.state, arg, sizeof(struct vhost_vring_state));
+        msg.addr.index += dev->vq_index;

Let's don't do this silently here. Since we've done a minus in vhost_virtqueue_start(), I think we could introduce a new callback and do correct calculation there.

Quoted 93 lines
hw/virtio/vhost-user.c +3 -1
         msg.size = sizeof(m.state);
         need_reply = 1;
         break;
 
     case VHOST_USER_SET_VRING_ADDR:
         memcpy(&msg.addr, arg, sizeof(struct vhost_vring_addr));
+        msg.addr.index += dev->vq_index;
         msg.size = sizeof(m.addr);
         break;
 
@@ -286,7 +314,7 @@ static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
     case VHOST_USER_SET_VRING_CALL:
     case VHOST_USER_SET_VRING_ERR:
         file = arg;
-        msg.u64 = file->index & VHOST_USER_VRING_IDX_MASK;
+        msg.u64 = (file->index + dev->vq_index) & VHOST_USER_VRING_IDX_MASK;
         msg.size = sizeof(m.u64);
         if (ioeventfd_enabled() && file->fd > 0) {
             fds[fd_num++] = file->fd;
@@ -330,6 +358,7 @@ static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
                 error_report("Received bad msg size.");
                 return -1;
             }
+            msg.state.index -= dev->vq_index;
             memcpy(arg, &msg.state, sizeof(struct vhost_vring_state));
             break;
         default:
include/net/net.h +1 -0
index 6a6cbef..6f20656 100644
@@ -92,6 +92,7 @@ struct NetClientState {
     NetClientDestructor *destructor;
     unsigned int queue_index;
     unsigned rxfilter_notify_enabled:1;
+    void *opaque;
 };
 
 typedef struct NICState {
net/vhost-user.c +14 -12
index 2d6bbe5..7d4ac69 100644
@@ -15,10 +15,16 @@
 #include "qemu/config-file.h"
 #include "qemu/error-report.h"
 
+struct VhostUserNetPeer {
+    NetClientState *nc;
+    VHostNetState  *vhost_net;
+};
+
 typedef struct VhostUserState {
-    NetClientState nc;
     CharDriverState *chr;
-    VHostNetState *vhost_net;
+    bool running;
+    int queues;
+    struct VhostUserNetPeer peers[];
 } VhostUserState;
 
 typedef struct VhostUserChardevProps {
@@ -29,48 +35,75 @@ typedef struct VhostUserChardevProps {
 
 VHostNetState *vhost_user_get_vhost_net(NetClientState *nc)
 {
-    VhostUserState *s = DO_UPCAST(VhostUserState, nc, nc);
+    VhostUserState *s = nc->opaque;
     assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_VHOST_USER);
-    return s->vhost_net;
-}
-
-static int vhost_user_running(VhostUserState *s)
-{
-    return (s->vhost_net) ? 1 : 0;
+    return s->peers[nc->queue_index].vhost_net;
 }
 
 static int vhost_user_start(VhostUserState *s)
 {
     VhostNetOptions options;
+    VHostNetState *vhost_net;
+    int max_queues;
+    int i = 0;
 
-    if (vhost_user_running(s)) {
+    if (s->running)
         return 0;
-    }
 
     options.backend_type = VHOST_BACKEND_TYPE_USER;
-    options.net_backend = &s->nc;
     options.opaque = s->chr;

net/vhost-user.c typedef struct VhostUserChardevProps { · line 63

Quoted 10 lines
net/vhost-user.c +8 -1
 
-    s->vhost_net = vhost_net_init(&options);
+    options.net_backend = s->peers[i].nc;
+    vhost_net = s->peers[i++].vhost_net = vhost_net_init(&options);
+
+    max_queues = vhost_net_get_max_queues(vhost_net);
+    if (s->queues >= max_queues) {
+        error_report("you are asking more queues than supported: %d\n",
+                     max_queues);
+        return -1;

Memory leak here?

net/vhost-user.c typedef struct VhostUserChardevProps { · line 72

Quoted 9 lines
net/vhost-user.c +9 -0
+    }
+
+    for (; i < s->queues; i++) {
+        options.net_backend = s->peers[i].nc;
+
+        s->peers[i].vhost_net = vhost_net_init(&options);
+        if (!s->peers[i].vhost_net)
+            return -1;
+    }

Maybe you can initialize all vhost_net in one loop, and check for the max_queues afterwards to simplify the code.

Quoted 121 lines
net/vhost-user.c +49 -23
+    s->running = true;
 
-    return vhost_user_running(s) ? 0 : -1;
+    return 0;
 }
 
 static void vhost_user_stop(VhostUserState *s)
 {
-    if (vhost_user_running(s)) {
-        vhost_net_cleanup(s->vhost_net);
+    int i;
+    VHostNetState *vhost_net;
+
+    if (!s->running)
+        return;
+
+    for (i = 0;  i < s->queues; i++) {
+        vhost_net = s->peers[i].vhost_net;
+        if (vhost_net)
+            vhost_net_cleanup(vhost_net);
     }
 
-    s->vhost_net = 0;
+    s->running = false;
 }
 
 static void vhost_user_cleanup(NetClientState *nc)
 {
-    VhostUserState *s = DO_UPCAST(VhostUserState, nc, nc);
+    VhostUserState *s = nc->opaque;
+    VHostNetState *vhost_net = s->peers[nc->queue_index].vhost_net;
+
+    if (vhost_net)
+        vhost_net_cleanup(vhost_net);
 
-    vhost_user_stop(s);
     qemu_purge_queued_packets(nc);
+
+    if (nc->queue_index == s->queues - 1)
+        free(s);
 }
 
 static bool vhost_user_has_vnet_hdr(NetClientState *nc)
@@ -89,7 +122,7 @@ static bool vhost_user_has_ufo(NetClientState *nc)
 
 static NetClientInfo net_vhost_user_info = {
         .type = NET_CLIENT_OPTIONS_KIND_VHOST_USER,
-        .size = sizeof(VhostUserState),
+        .size = sizeof(NetClientState),
         .cleanup = vhost_user_cleanup,
         .has_vnet_hdr = vhost_user_has_vnet_hdr,
         .has_ufo = vhost_user_has_ufo,
@@ -97,18 +130,25 @@ static NetClientInfo net_vhost_user_info = {
 
 static void net_vhost_link_down(VhostUserState *s, bool link_down)
 {
-    s->nc.link_down = link_down;
+    NetClientState *nc;
+    int i;
 
-    if (s->nc.peer) {
-        s->nc.peer->link_down = link_down;
-    }
+    for (i = 0; i < s->queues; i++) {
+        nc = s->peers[i].nc;
 
-    if (s->nc.info->link_status_changed) {
-        s->nc.info->link_status_changed(&s->nc);
-    }
+        nc->link_down = link_down;
+
+        if (nc->peer) {
+            nc->peer->link_down = link_down;
+        }
 
-    if (s->nc.peer && s->nc.peer->info->link_status_changed) {
-        s->nc.peer->info->link_status_changed(s->nc.peer);
+        if (nc->info->link_status_changed) {
+            nc->info->link_status_changed(nc);
+        }
+
+        if (nc->peer && nc->peer->info->link_status_changed) {
+            nc->peer->info->link_status_changed(nc->peer);
+        }
     }
 }
 
@@ -118,7 +158,8 @@ static void net_vhost_user_event(void *opaque, int event)
 
     switch (event) {
     case CHR_EVENT_OPENED:
-        vhost_user_start(s);
+        if (vhost_user_start(s) < 0)
+            exit(1);
         net_vhost_link_down(s, false);
         error_report("chardev \"%s\" went up", s->chr->label);
         break;
@@ -131,24 +172,28 @@ static void net_vhost_user_event(void *opaque, int event)
 }
 
 static int net_vhost_user_init(NetClientState *peer, const char *device,
-                               const char *name, CharDriverState *chr)
+                               const char *name, VhostUserState *s)
 {
     NetClientState *nc;
-    VhostUserState *s;
+    CharDriverState *chr = s->chr;
+    int i;
 
-    nc = qemu_new_net_client(&net_vhost_user_info, peer, device, name);
+    for (i = 0; i < s->queues; i++) {
+        nc = qemu_new_net_client(&net_vhost_user_info, peer, device, name);
 
-    snprintf(nc->info_str, sizeof(nc->info_str), "vhost-user to %s",
-             chr->label);
+        snprintf(nc->info_str, sizeof(nc->info_str), "vhost-user%d to %s",
+                 i, chr->label);
 
-    s = DO_UPCAST(VhostUserState, nc, nc);
+        /* We don't provide a receive callback */
+        nc->receive_disabled = 1;

net/vhost-user.c static void net_vhost_user_event(void *opaque, int event) · line 193

Quoted 10 lines
net/vhost-user.c +3 -5
 
-    /* We don't provide a receive callback */
-    s->nc.receive_disabled = 1;
-    s->chr = chr;
-    nc->queue_index = 0;
+        nc->queue_index = i;
+        nc->opaque      = s;
 
-    qemu_chr_add_handlers(s->chr, NULL, NULL, net_vhost_user_event, s);
+        s->peers[i].nc = nc;

So actually "peers" is confusing. Better rename it to "ncs" or something else.

Quoted 73 lines
net/vhost-user.c +11 -1
+    }
+
+    qemu_chr_add_handlers(chr, NULL, NULL, net_vhost_user_event, s);
 
     return 0;
 }
@@ -227,8 +272,10 @@ static int net_vhost_check_net(void *opaque, QemuOpts *opts, Error **errp)
 int net_init_vhost_user(const NetClientOptions *opts, const char *name,
                         NetClientState *peer, Error **errp)
 {
+    int queues;
     const NetdevVhostUserOptions *vhost_user_opts;
     CharDriverState *chr;
+    VhostUserState *s;
 
     assert(opts->kind == NET_CLIENT_OPTIONS_KIND_VHOST_USER);
     vhost_user_opts = opts->vhost_user;
@@ -244,6 +291,11 @@ int net_init_vhost_user(const NetClientOptions *opts, const char *name,
         return -1;
     }
 
+    queues = vhost_user_opts->has_queues ? vhost_user_opts->queues : 1;
+    s = g_malloc0(sizeof(VhostUserState) +
+                  queues * sizeof(struct VhostUserNetPeer));
+    s->queues = queues;
+    s->chr    = chr;
 
-    return net_vhost_user_init(peer, "vhost_user", name, chr);
+    return net_vhost_user_init(peer, "vhost_user", name, s);
 }
qapi-schema.json +5 -1
index 67fef37..55c33db 100644
@@ -2480,12 +2480,16 @@
 #
 # @vhostforce: #optional vhost on for non-MSIX virtio guests (default: false).
 #
+# @queues: #optional number of queues to be created for multiqueue vhost-user
+#          (default: 1) (Since 2.5)
+#
 # Since 2.1
 ##
 { 'struct': 'NetdevVhostUserOptions',
   'data': {
     'chardev':        'str',
-    '*vhostforce':    'bool' } }
+    '*vhostforce':    'bool',
+    '*queues':        'int' } }
 
 ##
 # @NetClientOptions
qemu-options.hx +3 -2
index 77f5853..5bfa7a3 100644
@@ -1963,13 +1963,14 @@ The hubport netdev lets you connect a NIC to a QEMU "vlan" instead of a single
 netdev.  @code{-net} and @code{-device} with parameter @option{vlan} create the
 required hub automatically.
 
-@item -netdev vhost-user,chardev=@var{id}[,vhostforce=on|off]
+@item -netdev vhost-user,chardev=@var{id}[,vhostforce=on|off][,queues=n]
 
 Establish a vhost-user netdev, backed by a chardev @var{id}. The chardev should
 be a unix domain socket backed one. The vhost-user uses a specifically defined
 protocol to pass vhost ioctl replacement messages to an application on the other
 end of the socket. On non-MSIX guests, the feature can be forced with
-@var{vhostforce}.
+@var{vhostforce}. Use 'queues=@var{n}' to specify the number of queues to
+be created for multiqueue vhost-user.
 
 Example:
 @example
#24 lore raw

Re: [Qemu-devel] [PATCH 6/7] vhost-user: add multiple queue support

Quoted 77 lines · Jason Wang
Quoted 62 lines · Yuanhan Liu

From: Ouyang Changchun <[email protected]>

This patch is initially based a patch from Nikolay Nikolaev.

Here is the latest version for adding vhost-user multiple queue support, by creating a nc and vhost_net pair for each queue.

What differs from last version is that this patch addresses two major concerns from Michael, and fixes one hidden bug.

- Concern #1: no feedback when the backend can't support # of requested queues(by providing queues=# option).

Here we address this issue by querying VHOST_USER_PROTOCOL_F_MQ protocol features first, if not set, it means the backend don't support mq feature, and let max_queues be 1. Otherwise, we send another message, VHOST_USER_GET_QUEUE_NUM, for getting the max_queues the backend supports.

At vhost-user initiation stage(net_vhost_user_start), we then initiate one queue first, which, in the meantime, also gets the max_queues. We then do a simple compare: if requested_queues > max_queues, we exit(I guess it's safe to exit here, as the vm is not running yet).

- Concern #2: some messages are sent more times than necessary.

We came an agreement with Michael that we could categorize vhost user messages to 2 types: none-vring specific messages, which should be sent only once, and vring specific messages, which should be sent per queue.

Here I introduced a helper function vhost_user_one_time_request(), which lists following messages as none-vring specific messages:

VHOST_USER_GET_FEATURES
VHOST_USER_SET_FEATURES
VHOST_USER_GET_PROTOCOL_FEATURES
VHOST_USER_SET_PROTOCOL_FEATURES
VHOST_USER_SET_OWNER
VHOST_USER_RESET_DEVICE
VHOST_USER_SET_MEM_TABLE
VHOST_USER_GET_QUEUE_NUM

For above messages, we simply ignore them when they are not sent the first time.

I also observed a hidden bug from last version. We register the char dev event handler N times, which is not necessary, as well as buggy: A later register overwrites the former one, as qemu_chr_add_handlers() will not chain those handlers, but instead overwrites the old one. So, in theory, invoking qemu_chr_add_handlers N times will not end up with calling the handler N times.

However, the reason the handler is invoked N times is because we start the backend(the connection server) first, and hence when net_vhost_user_init() is executed, the connection is already established, and qemu_chr_add_handlers() then invoke the handler immediately, which just looks like we invoke the handler(net_vhost_user_event) directly from net_vhost_user_init().

The solution I came up with is to make VhostUserState as an upper level structure, making it includes N nc and vhost_net pairs:

Quoted 11 lines
   struct VhostUserNetPeer {
       NetClientState *nc;
       VHostNetState  *vhost_net;
   };

   typedef struct VhostUserState {
       CharDriverState *chr;
       bool running;
       int queues;
       struct VhostUserNetPeer peers[];
   } VhostUserState;
Quoted 6 lines

Haven't looked at previous versions but alternatively, can you do this by:

- keep current VhostUserState
- allocate multiple VhostUserState
- register the char handler only on VhostUserState[0]
- enumerate all others in handler by qemu_find_net_clients_except() ?

That does sound work and much better! I will give it a try soon. Thanks!

Quoted 6 lines
Quoted 6 lines

---

...

Quoted 7 lines
Quoted 3 lines
     case VHOST_USER_GET_VRING_BASE:
         memcpy(&msg.state, arg, sizeof(struct vhost_vring_state));

+ msg.addr.index += dev->vq_index;

Let's don't do this silently here. Since we've done a minus in vhost_virtqueue_start(), I think we could introduce a new callback and do correct calculation there.

Good point, I will try it as well.

Quoted 11 lines

net/vhost-user.c typedef struct VhostUserChardevProps { · line 63

Quoted 9 lines
net/vhost-user.c +8 -1
-    s->vhost_net = vhost_net_init(&options);
+    options.net_backend = s->peers[i].nc;
+    vhost_net = s->peers[i++].vhost_net = vhost_net_init(&options);
+
+    max_queues = vhost_net_get_max_queues(vhost_net);
+    if (s->queues >= max_queues) {
+        error_report("you are asking more queues than supported: %d\n",
+                     max_queues);
+        return -1;

Memory leak here?

Yeah, it is. But Does it matter? Since we will call exit() soon, which will reclaim everything in the end. Or, even though, it's still needed?

Either way, I can do that.

Quoted 13 lines

net/vhost-user.c typedef struct VhostUserChardevProps { · line 72

Quoted 10 lines
net/vhost-user.c +9 -0
+    }
+
+    for (; i < s->queues; i++) {
+        options.net_backend = s->peers[i].nc;
+
+        s->peers[i].vhost_net = vhost_net_init(&options);
+        if (!s->peers[i].vhost_net)
+            return -1;
+    }

Maybe you can initialize all vhost_net in one loop, and check for the max_queues afterwards to simplify the code.

Okay.

Quoted 1 line
Quoted 1 line

...

Quoted 29 lines
Quoted 29 lines

static int net_vhost_user_init(NetClientState *peer, const char *device,

net/vhost-user.c +11 -10
-                               const char *name, CharDriverState *chr)
+                               const char *name, VhostUserState *s)
 {
     NetClientState *nc;
-    VhostUserState *s;
+    CharDriverState *chr = s->chr;
+    int i;
 
-    nc = qemu_new_net_client(&net_vhost_user_info, peer, device, name);
+    for (i = 0; i < s->queues; i++) {
+        nc = qemu_new_net_client(&net_vhost_user_info, peer, device, name);
 
-    snprintf(nc->info_str, sizeof(nc->info_str), "vhost-user to %s",
-             chr->label);
+        snprintf(nc->info_str, sizeof(nc->info_str), "vhost-user%d to %s",
+                 i, chr->label);
 
-    s = DO_UPCAST(VhostUserState, nc, nc);
+        /* We don't provide a receive callback */
+        nc->receive_disabled = 1;
 
-    /* We don't provide a receive callback */
-    s->nc.receive_disabled = 1;
-    s->chr = chr;
-    nc->queue_index = 0;
+        nc->queue_index = i;
+        nc->opaque      = s;
Quoted 5 lines

net/vhost-user.c static void net_vhost_user_event(void *opaque, int event) · line 193

Quoted 2 lines
net/vhost-user.c +1 -1
-    qemu_chr_add_handlers(s->chr, NULL, NULL, net_vhost_user_event, s);
+        s->peers[i].nc = nc;

So actually "peers" is confusing. Better rename it to "ncs" or something else.

Yeah, that's why I renamed it to pairs in later version. But anyway, I guess it's not needed anymore with qemu_find_net_clients_except(), right? :)

Thanks!

	--yliu
#25 lore raw

PATCH v7 7/7 vhost-user: add a new message to disable/enable a specific virt queue.

In reply to Yuanhan Liu

From: Changchun Ouyang <[email protected]>

Add a new message, VHOST_USER_SET_VRING_FLAG, to enable and disable a specific virt queue, which is similar to attach/detach queue for tap device.

virtio driver on guest doesn't have to use max virt queue pair, it could enable any number of virt queue ranging from 1 to max virt queue pair.

---
 docs/specs/vhost-user.txt         | 10 ++++++++++
 hw/net/vhost_net.c                | 18 ++++++++++++++++++
 hw/net/virtio-net.c               |  2 ++
 hw/virtio/vhost-user.c            | 32 ++++++++++++++++++++++++++++----
 include/hw/virtio/vhost-backend.h |  2 ++
 include/net/vhost_net.h           |  1 +
 6 files changed, 61 insertions(+), 4 deletions(-)
docs/specs/vhost-user.txt +10 -0
index 99d79be..c2c38f3 100644
@@ -325,3 +325,13 @@ Message types
       Query how many queues the backend supports. This request should be
       sent only when VHOST_USER_PROTOCOL_F_MQ is set in quried protocol
       features by VHOST_USER_GET_PROTOCOL_FEATURES.
+
+ * VHOST_USER_SET_VRING_FLAG
+
+      Id: 18
+      Equivalent ioctl: N/A
+      Master payload: vring state description
+
+      Set the flag(1 for enable and 0 for disable) to signal slave to enable
+      or disable corresponding virt queue. This request should be sent only
+      when the protocol feature bit VHOST_USER_PROTOCOL_F_VRING_FLAG is set.
hw/net/vhost_net.c +18 -0
index 141b557..7d9cc8d 100644
@@ -418,6 +418,19 @@ VHostNetState *get_vhost_net(NetClientState *nc)
 
     return vhost_net;
 }
+
+int vhost_set_vring_flag(NetClientState *nc, int enable)
+{
+    if (nc->info->type == NET_CLIENT_OPTIONS_KIND_VHOST_USER) {
+        VHostNetState *net = get_vhost_net(nc);
+        const VhostOps *vhost_ops = net->dev.vhost_ops;
+        if (vhost_ops->vhost_backend_mq_set_vring_flag)
+            return vhost_ops->vhost_backend_mq_set_vring_flag(&net->dev, enable);
+    }
+
+    return 0;
+}
+
 #else
 struct vhost_net *vhost_net_init(VhostNetOptions *options)
 {
@@ -463,4 +476,9 @@ VHostNetState *get_vhost_net(NetClientState *nc)
 {
     return 0;
 }
+
+int vhost_set_vring_flag(NetClientState *nc, int enable)
+{
+    return 0;
+}
 #endif
hw/net/virtio-net.c +2 -0
index 8d28e45..53f93b1 100644
@@ -407,6 +407,7 @@ static int peer_attach(VirtIONet *n, int index)
     }
 
     if (nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
+        vhost_set_vring_flag(nc->peer, 1);
         return 0;
     }
 
@@ -422,6 +423,7 @@ static int peer_detach(VirtIONet *n, int index)
     }
 
     if (nc->peer->info->type !=  NET_CLIENT_OPTIONS_KIND_TAP) {
+        vhost_set_vring_flag(nc->peer, 0);
         return 0;
     }
 
hw/virtio/vhost-user.c +28 -4
index 11e46b5..ca6f7fa 100644
@@ -25,9 +25,10 @@
 
 #define VHOST_MEMORY_MAX_NREGIONS    8
 #define VHOST_USER_F_PROTOCOL_FEATURES 30
-#define VHOST_USER_PROTOCOL_FEATURE_MASK 0x1ULL
+#define VHOST_USER_PROTOCOL_FEATURE_MASK 0xfULL
 
-#define VHOST_USER_PROTOCOL_F_MQ    0
+#define VHOST_USER_PROTOCOL_F_MQ            0
+#define VHOST_USER_PROTOCOL_F_VRING_FLAG    1
 
 typedef enum VhostUserRequest {
     VHOST_USER_NONE = 0,
@@ -48,6 +49,7 @@ typedef enum VhostUserRequest {
     VHOST_USER_GET_PROTOCOL_FEATURES = 15,
     VHOST_USER_SET_PROTOCOL_FEATURES = 16,
     VHOST_USER_GET_QUEUE_NUM = 17,
+    VHOST_USER_SET_VRING_FLAG = 18,
     VHOST_USER_MAX
 } VhostUserRequest;
 
@@ -296,6 +298,11 @@ static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
         msg.addr.index += dev->vq_index;
         msg.size = sizeof(m.state);
         break;
+    case VHOST_USER_SET_VRING_FLAG:
+        msg.state.index = dev->vq_index;
+        msg.state.num = *(int*)arg;
+        msg.size = sizeof(msg.state);
+        break;
 
     case VHOST_USER_GET_VRING_BASE:
         memcpy(&msg.state, arg, sizeof(struct vhost_vring_state));
@@ -408,6 +415,22 @@ static int vhost_user_init(struct vhost_dev *dev, void *opaque)
     return 0;
 }
 
+static int vhost_user_set_vring_flag(struct vhost_dev *dev, int enable)
+{
+    int err;
+
+    assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
+
+    if (!(dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_VRING_FLAG)))
+        return -1;
+
+    err = vhost_user_call(dev, VHOST_USER_SET_VRING_FLAG, &enable);
+    if (err < 0)
+        return err;
+
+    return 0;
+}
+
 static int vhost_user_cleanup(struct vhost_dev *dev)
 {
     assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
@@ -421,5 +444,6 @@ const VhostOps user_ops = {
         .backend_type = VHOST_BACKEND_TYPE_USER,
         .vhost_call = vhost_user_call,
         .vhost_backend_init = vhost_user_init,
-        .vhost_backend_cleanup = vhost_user_cleanup
-        };
+        .vhost_backend_cleanup = vhost_user_cleanup,
+        .vhost_backend_mq_set_vring_flag = vhost_user_set_vring_flag,
+};
include/hw/virtio/vhost-backend.h +2 -0
index e472f29..6fc76b7 100644
@@ -24,12 +24,14 @@ typedef int (*vhost_call)(struct vhost_dev *dev, unsigned long int request,
              void *arg);
 typedef int (*vhost_backend_init)(struct vhost_dev *dev, void *opaque);
 typedef int (*vhost_backend_cleanup)(struct vhost_dev *dev);
+typedef int (*vhost_backend_mq_set_vring_flag)(struct vhost_dev *dev, int enable);
 
 typedef struct VhostOps {
     VhostBackendType backend_type;
     vhost_call vhost_call;
     vhost_backend_init vhost_backend_init;
     vhost_backend_cleanup vhost_backend_cleanup;
+    vhost_backend_mq_set_vring_flag vhost_backend_mq_set_vring_flag;
 } VhostOps;
 
 extern const VhostOps user_ops;
include/net/vhost_net.h +1 -0
index 8db723e..3ec33ff 100644
@@ -28,4 +28,5 @@ bool vhost_net_virtqueue_pending(VHostNetState *net, int n);
 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
                               int idx, bool mask);
 VHostNetState *get_vhost_net(NetClientState *nc);
+int vhost_set_vring_flag(NetClientState * nc, int enable);
 #endif
Signature · 1 line
1.9.0
#26 lore raw

Re: [Qemu-devel] [PATCH 7/7] vhost-user: add a new message to disable/enable a specific virt queue.

Quoted 6 lines · Yuanhan Liu

From: Changchun Ouyang <[email protected]>

Add a new message, VHOST_USER_SET_VRING_FLAG, to enable and disable a specific virt queue, which is similar to attach/detach queue for tap device.

Quoted 6 lines

virtio driver on guest doesn't have to use max virt queue pair, it could enable any number of virt queue ranging from 1 to max virt queue pair.

Fine, but I would like to make this dependent on the MQ flag.

Quoted 14 lines
---
 docs/specs/vhost-user.txt         | 10 ++++++++++
 hw/net/vhost_net.c                | 18 ++++++++++++++++++
 hw/net/virtio-net.c               |  2 ++
 hw/virtio/vhost-user.c            | 32 ++++++++++++++++++++++++++++----
 include/hw/virtio/vhost-backend.h |  2 ++
 include/net/vhost_net.h           |  1 +
 6 files changed, 61 insertions(+), 4 deletions(-)
docs/specs/vhost-user.txt +0 -0
index 99d79be..c2c38f3 100644

docs/specs/vhost-user.txt Message types · line 329

Quoted 5 lines
docs/specs/vhost-user.txt +2 -0
@@ -325,3 +325,13 @@ Message types
       Query how many queues the backend supports. This request should be
       sent only when VHOST_USER_PROTOCOL_F_MQ is set in quried protocol
       features by VHOST_USER_GET_PROTOCOL_FEATURES.
+
+ * VHOST_USER_SET_VRING_FLAG

SET_VRIHG_ENABLE would be a better name.

docs/specs/vhost-user.txt Message types · line 335

Quoted 6 lines
docs/specs/vhost-user.txt +6 -0
+
+      Id: 18
+      Equivalent ioctl: N/A
+      Master payload: vring state description
+
+      Set the flag(1 for enable and 0 for disable) to signal slave to enable

Space before ( please.

Quoted 7 lines
docs/specs/vhost-user.txt +2 -0
+      or disable corresponding virt queue. This request should be sent only
+      when the protocol feature bit VHOST_USER_PROTOCOL_F_VRING_FLAG is set.
hw/net/vhost_net.c +0 -0
index 141b557..7d9cc8d 100644

hw/net/vhost_net.c VHostNetState *get_vhost_net(NetClientState *nc) · line 426

Quoted 9 lines
hw/net/vhost_net.c +6 -0
@@ -418,6 +418,19 @@ VHostNetState *get_vhost_net(NetClientState *nc)
 
     return vhost_net;
 }
+
+int vhost_set_vring_flag(NetClientState *nc, int enable)
+{
+    if (nc->info->type == NET_CLIENT_OPTIONS_KIND_VHOST_USER) {
+        VHostNetState *net = get_vhost_net(nc);
+        const VhostOps *vhost_ops = net->dev.vhost_ops;

add an empty line after declaration pls.

hw/net/vhost_net.c VHostNetState *get_vhost_net(NetClientState *nc) · line 428

Quoted 2 lines
hw/net/vhost_net.c +2 -0
+        if (vhost_ops->vhost_backend_mq_set_vring_flag)
+            return vhost_ops->vhost_backend_mq_set_vring_flag(&net->dev, enable);

Coding style violation:
line too long, and should use {}.

Quoted 31 lines
hw/net/vhost_net.c +10 -0
+    }
+
+    return 0;
+}
+
 #else
 struct vhost_net *vhost_net_init(VhostNetOptions *options)
 {
@@ -463,4 +476,9 @@ VHostNetState *get_vhost_net(NetClientState *nc)
 {
     return 0;
 }
+
+int vhost_set_vring_flag(NetClientState *nc, int enable)
+{
+    return 0;
+}
 #endif
hw/net/virtio-net.c +1 -0
index 8d28e45..53f93b1 100644
@@ -407,6 +407,7 @@ static int peer_attach(VirtIONet *n, int index)
     }
 
     if (nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
+        vhost_set_vring_flag(nc->peer, 1);
         return 0;
     }
 

hw/net/virtio-net.c static int peer_detach(VirtIONet *n, int index) · line 429

Quoted 7 lines
hw/net/virtio-net.c +1 -0
@@ -422,6 +423,7 @@ static int peer_detach(VirtIONet *n, int index)
     }
 
     if (nc->peer->info->type !=  NET_CLIENT_OPTIONS_KIND_TAP) {
+        vhost_set_vring_flag(nc->peer, 0);
         return 0;
     }
 

makes no sense to call it for !== tap only. Either call this unconditionally, or just for vhost user.

Quoted 50 lines
hw/virtio/vhost-user.c +18 -2
index 11e46b5..ca6f7fa 100644
@@ -25,9 +25,10 @@
 
 #define VHOST_MEMORY_MAX_NREGIONS    8
 #define VHOST_USER_F_PROTOCOL_FEATURES 30
-#define VHOST_USER_PROTOCOL_FEATURE_MASK 0x1ULL
+#define VHOST_USER_PROTOCOL_FEATURE_MASK 0xfULL
 
-#define VHOST_USER_PROTOCOL_F_MQ    0
+#define VHOST_USER_PROTOCOL_F_MQ            0
+#define VHOST_USER_PROTOCOL_F_VRING_FLAG    1
 
 typedef enum VhostUserRequest {
     VHOST_USER_NONE = 0,
@@ -48,6 +49,7 @@ typedef enum VhostUserRequest {
     VHOST_USER_GET_PROTOCOL_FEATURES = 15,
     VHOST_USER_SET_PROTOCOL_FEATURES = 16,
     VHOST_USER_GET_QUEUE_NUM = 17,
+    VHOST_USER_SET_VRING_FLAG = 18,
     VHOST_USER_MAX
 } VhostUserRequest;
 
@@ -296,6 +298,11 @@ static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
         msg.addr.index += dev->vq_index;
         msg.size = sizeof(m.state);
         break;
+    case VHOST_USER_SET_VRING_FLAG:
+        msg.state.index = dev->vq_index;
+        msg.state.num = *(int*)arg;
+        msg.size = sizeof(msg.state);
+        break;
 
     case VHOST_USER_GET_VRING_BASE:
         memcpy(&msg.state, arg, sizeof(struct vhost_vring_state));
@@ -408,6 +415,22 @@ static int vhost_user_init(struct vhost_dev *dev, void *opaque)
     return 0;
 }
 
+static int vhost_user_set_vring_flag(struct vhost_dev *dev, int enable)
+{
+    int err;
+
+    assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
+
+    if (!(dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_VRING_FLAG)))
+        return -1;
+

hw/virtio/vhost-user.c static int vhost_user_init(struct vhost_dev *dev, void *opaque) · line 436

Quoted 10 lines
hw/virtio/vhost-user.c +7 -0
+    err = vhost_user_call(dev, VHOST_USER_SET_VRING_FLAG, &enable);
+    if (err < 0)
+        return err;
+
+    return 0;
+}
+
 static int vhost_user_cleanup(struct vhost_dev *dev)
 {
     assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);

In fact, it's a per VQ pair call the way it's coded. So pls fix it, call it for all VQs.

Quoted 15 lines
hw/virtio/vhost-user.c +3 -2
@@ -421,5 +444,6 @@ const VhostOps user_ops = {
         .backend_type = VHOST_BACKEND_TYPE_USER,
         .vhost_call = vhost_user_call,
         .vhost_backend_init = vhost_user_init,
-        .vhost_backend_cleanup = vhost_user_cleanup
-        };
+        .vhost_backend_cleanup = vhost_user_cleanup,
+        .vhost_backend_mq_set_vring_flag = vhost_user_set_vring_flag,
+};
include/hw/virtio/vhost-backend.h +0 -0
index e472f29..6fc76b7 100644
@@ -24,12 +24,14 @@ typedef int (*vhost_call)(struct vhost_dev *dev, unsigned long int request,
              void *arg);

include/hw/virtio/vhost-backend.h typedef int (*vhost_call)(struct vhost_dev *dev, unsigned long int request, · line 34

Quoted 10 lines
include/hw/virtio/vhost-backend.h +2 -0
 typedef int (*vhost_backend_init)(struct vhost_dev *dev, void *opaque);
 typedef int (*vhost_backend_cleanup)(struct vhost_dev *dev);
+typedef int (*vhost_backend_mq_set_vring_flag)(struct vhost_dev *dev, int enable);
 
 typedef struct VhostOps {
     VhostBackendType backend_type;
     vhost_call vhost_call;
     vhost_backend_init vhost_backend_init;
     vhost_backend_cleanup vhost_backend_cleanup;
+    vhost_backend_mq_set_vring_flag vhost_backend_mq_set_vring_flag;

That's quite ugly: mq is a vhost net thing. Why not enable each VQ?

Quoted 15 lines
include/hw/virtio/vhost-backend.h +0 -0
 } VhostOps;
 
 extern const VhostOps user_ops;
include/net/vhost_net.h +1 -0
index 8db723e..3ec33ff 100644
@@ -28,4 +28,5 @@ bool vhost_net_virtqueue_pending(VHostNetState *net, int n);
 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
                               int idx, bool mask);
 VHostNetState *get_vhost_net(NetClientState *nc);
+int vhost_set_vring_flag(NetClientState * nc, int enable);
 #endif
Signature · 1 line
1.9.0
#27 lore raw

Re: [Qemu-devel] [PATCH 7/7] vhost-user: add a new message to disable/enable a specific virt queue.

Quoted 7 lines · Michael S. Tsirkin
Quoted 6 lines · Yuanhan Liu

From: Changchun Ouyang <[email protected]>

Add a new message, VHOST_USER_SET_VRING_FLAG, to enable and disable a specific virt queue, which is similar to attach/detach queue for tap device.

Quoted 8 lines
Quoted 6 lines

virtio driver on guest doesn't have to use max virt queue pair, it could enable any number of virt queue ranging from 1 to max virt queue pair.

Fine, but I would like to make this dependent on the MQ flag.

Reasonable.

Quoted 11 lines
Quoted 11 lines

---
 docs/specs/vhost-user.txt         | 10 ++++++++++
 hw/net/vhost_net.c                | 18 ++++++++++++++++++
 hw/net/virtio-net.c               |  2 ++
 hw/virtio/vhost-user.c            | 32 ++++++++++++++++++++++++++++----
 include/hw/virtio/vhost-backend.h |  2 ++
 include/net/vhost_net.h           |  1 +
 6 files changed, 61 insertions(+), 4 deletions(-)
Quoted 13 lines

docs/specs/vhost-user.txt Message types · line 329

Quoted 10 lines
docs/specs/vhost-user.txt +2 -0
index 99d79be..c2c38f3 100644
@@ -325,3 +325,13 @@ Message types
       Query how many queues the backend supports. This request should be
       sent only when VHOST_USER_PROTOCOL_F_MQ is set in quried protocol
       features by VHOST_USER_GET_PROTOCOL_FEATURES.
+
+ * VHOST_USER_SET_VRING_FLAG

SET_VRIHG_ENABLE would be a better name.

Okay.

Quoted 9 lines

docs/specs/vhost-user.txt Message types · line 335

Quoted 7 lines
docs/specs/vhost-user.txt +6 -0
+
+      Id: 18
+      Equivalent ioctl: N/A
+      Master payload: vring state description
+
+      Set the flag(1 for enable and 0 for disable) to signal slave to enable

Space before ( please.

Okay.

Quoted 9 lines
Quoted 9 lines
docs/specs/vhost-user.txt +2 -0
+      or disable corresponding virt queue. This request should be sent only
+      when the protocol feature bit VHOST_USER_PROTOCOL_F_VRING_FLAG is set.
hw/net/vhost_net.c +0 -0
index 141b557..7d9cc8d 100644
@@ -418,6 +418,19 @@ VHostNetState *get_vhost_net(NetClientState *nc)
 
Quoted 10 lines

hw/net/vhost_net.c VHostNetState *get_vhost_net(NetClientState *nc) · line 426

Quoted 8 lines
hw/net/vhost_net.c +6 -0
     return vhost_net;
 }
+
+int vhost_set_vring_flag(NetClientState *nc, int enable)
+{
+    if (nc->info->type == NET_CLIENT_OPTIONS_KIND_VHOST_USER) {
+        VHostNetState *net = get_vhost_net(nc);
+        const VhostOps *vhost_ops = net->dev.vhost_ops;

add an empty line after declaration pls.

Got it, thanks.

Quoted 6 lines

hw/net/vhost_net.c VHostNetState *get_vhost_net(NetClientState *nc) · line 428

Quoted 3 lines
hw/net/vhost_net.c +2 -0
+        if (vhost_ops->vhost_backend_mq_set_vring_flag)
+            return vhost_ops->vhost_backend_mq_set_vring_flag(&net->dev, enable);

Coding style violation:
line too long,

Yeah, 81 chars. I thougth that's kind of acceptable. But, anyway, since you pointed it out, I will fix it next time.

Quoted 1 line

and should use {}.

TBH, I seldom use {} for one statement, and you could see that a lot from this patch set. Is it a must in qemu community? If so, I could fix them all.

Quoted 34 lines
Quoted 34 lines
hw/net/vhost_net.c +10 -0
+    }
+
+    return 0;
+}
+
 #else
 struct vhost_net *vhost_net_init(VhostNetOptions *options)
 {
@@ -463,4 +476,9 @@ VHostNetState *get_vhost_net(NetClientState *nc)
 {
     return 0;
 }
+
+int vhost_set_vring_flag(NetClientState *nc, int enable)
+{
+    return 0;
+}
 #endif
hw/net/virtio-net.c +1 -0
index 8d28e45..53f93b1 100644
@@ -407,6 +407,7 @@ static int peer_attach(VirtIONet *n, int index)
     }
 
     if (nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
+        vhost_set_vring_flag(nc->peer, 1);
         return 0;
     }
 
@@ -422,6 +423,7 @@ static int peer_detach(VirtIONet *n, int index)
     }
 
Quoted 8 lines

hw/net/virtio-net.c static int peer_detach(VirtIONet *n, int index) · line 429

Quoted 5 lines
hw/net/virtio-net.c +1 -0
     if (nc->peer->info->type !=  NET_CLIENT_OPTIONS_KIND_TAP) {
+        vhost_set_vring_flag(nc->peer, 0);
         return 0;
     }
 

makes no sense to call it for !== tap only. Either call this unconditionally, or just for vhost user.

Agreed. I will make it vhost-usre specific then.

Quoted 57 lines
Quoted 57 lines
hw/virtio/vhost-user.c +23 -2
index 11e46b5..ca6f7fa 100644
@@ -25,9 +25,10 @@
 
 #define VHOST_MEMORY_MAX_NREGIONS    8
 #define VHOST_USER_F_PROTOCOL_FEATURES 30
-#define VHOST_USER_PROTOCOL_FEATURE_MASK 0x1ULL
+#define VHOST_USER_PROTOCOL_FEATURE_MASK 0xfULL
 
-#define VHOST_USER_PROTOCOL_F_MQ    0
+#define VHOST_USER_PROTOCOL_F_MQ            0
+#define VHOST_USER_PROTOCOL_F_VRING_FLAG    1
 
 typedef enum VhostUserRequest {
     VHOST_USER_NONE = 0,
@@ -48,6 +49,7 @@ typedef enum VhostUserRequest {
     VHOST_USER_GET_PROTOCOL_FEATURES = 15,
     VHOST_USER_SET_PROTOCOL_FEATURES = 16,
     VHOST_USER_GET_QUEUE_NUM = 17,
+    VHOST_USER_SET_VRING_FLAG = 18,
     VHOST_USER_MAX
 } VhostUserRequest;
 
@@ -296,6 +298,11 @@ static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
         msg.addr.index += dev->vq_index;
         msg.size = sizeof(m.state);
         break;
+    case VHOST_USER_SET_VRING_FLAG:
+        msg.state.index = dev->vq_index;
+        msg.state.num = *(int*)arg;
+        msg.size = sizeof(msg.state);
+        break;
 
     case VHOST_USER_GET_VRING_BASE:
         memcpy(&msg.state, arg, sizeof(struct vhost_vring_state));
@@ -408,6 +415,22 @@ static int vhost_user_init(struct vhost_dev *dev, void *opaque)
     return 0;
 }
 
+static int vhost_user_set_vring_flag(struct vhost_dev *dev, int enable)
+{
+    int err;
+
+    assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
+
+    if (!(dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_VRING_FLAG)))
+        return -1;
+
+    err = vhost_user_call(dev, VHOST_USER_SET_VRING_FLAG, &enable);
+    if (err < 0)
+        return err;
+
+    return 0;
Quoted 8 lines

hw/virtio/vhost-user.c static int vhost_user_init(struct vhost_dev *dev, void *opaque) · line 436

Quoted 5 lines
hw/virtio/vhost-user.c +2 -0
+}
+
 static int vhost_user_cleanup(struct vhost_dev *dev)
 {
     assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);

In fact, it's a per VQ pair call the way it's coded. So pls fix it, call it for all VQs.

Let me try.

Quoted 21 lines
Quoted 21 lines
hw/virtio/vhost-user.c +3 -2
@@ -421,5 +444,6 @@ const VhostOps user_ops = {
         .backend_type = VHOST_BACKEND_TYPE_USER,
         .vhost_call = vhost_user_call,
         .vhost_backend_init = vhost_user_init,
-        .vhost_backend_cleanup = vhost_user_cleanup
-        };
+        .vhost_backend_cleanup = vhost_user_cleanup,
+        .vhost_backend_mq_set_vring_flag = vhost_user_set_vring_flag,
+};
include/hw/virtio/vhost-backend.h +1 -0
index e472f29..6fc76b7 100644
@@ -24,12 +24,14 @@ typedef int (*vhost_call)(struct vhost_dev *dev, unsigned long int request,
              void *arg);
 typedef int (*vhost_backend_init)(struct vhost_dev *dev, void *opaque);
 typedef int (*vhost_backend_cleanup)(struct vhost_dev *dev);
+typedef int (*vhost_backend_mq_set_vring_flag)(struct vhost_dev *dev, int enable);
 
Quoted 9 lines

include/hw/virtio/vhost-backend.h typedef int (*vhost_call)(struct vhost_dev *dev, unsigned long int request, · line 34

Quoted 6 lines
include/hw/virtio/vhost-backend.h +1 -0
 typedef struct VhostOps {
     VhostBackendType backend_type;
     vhost_call vhost_call;
     vhost_backend_init vhost_backend_init;
     vhost_backend_cleanup vhost_backend_cleanup;
+    vhost_backend_mq_set_vring_flag vhost_backend_mq_set_vring_flag;

That's quite ugly: mq is a vhost net thing. Why not enable each VQ?

To define it as `vhost_backend_enable_vring'?

Thanks.

	--yliu
Quoted 16 lines
Quoted 16 lines
include/hw/virtio/vhost-backend.h +0 -0
 } VhostOps;
 
 extern const VhostOps user_ops;
include/net/vhost_net.h +1 -0
index 8db723e..3ec33ff 100644
@@ -28,4 +28,5 @@ bool vhost_net_virtqueue_pending(VHostNetState *net, int n);
 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
                               int idx, bool mask);
 VHostNetState *get_vhost_net(NetClientState *nc);
+int vhost_set_vring_flag(NetClientState * nc, int enable);
 #endif
Signature · 1 line
1.9.0
#28 lore raw

Re: [Qemu-devel] [PATCH 7/7] vhost-user: add a new message to disable/enable a specific virt queue.

Quoted 87 lines · Yuanhan Liu
Quoted 7 lines · Michael S. Tsirkin
Quoted 6 lines · Yuanhan Liu

From: Changchun Ouyang <[email protected]>

Add a new message, VHOST_USER_SET_VRING_FLAG, to enable and disable a specific virt queue, which is similar to attach/detach queue for tap device.

Quoted 8 lines
Quoted 6 lines

virtio driver on guest doesn't have to use max virt queue pair, it could enable any number of virt queue ranging from 1 to max virt queue pair.

Fine, but I would like to make this dependent on the MQ flag.

Reasonable.

Quoted 11 lines
Quoted 11 lines

---
 docs/specs/vhost-user.txt         | 10 ++++++++++
 hw/net/vhost_net.c                | 18 ++++++++++++++++++
 hw/net/virtio-net.c               |  2 ++
 hw/virtio/vhost-user.c            | 32 ++++++++++++++++++++++++++++----
 include/hw/virtio/vhost-backend.h |  2 ++
 include/net/vhost_net.h           |  1 +
 6 files changed, 61 insertions(+), 4 deletions(-)
Quoted 13 lines

docs/specs/vhost-user.txt Message types · line 329

Quoted 10 lines
docs/specs/vhost-user.txt +2 -0
index 99d79be..c2c38f3 100644
@@ -325,3 +325,13 @@ Message types
       Query how many queues the backend supports. This request should be
       sent only when VHOST_USER_PROTOCOL_F_MQ is set in quried protocol
       features by VHOST_USER_GET_PROTOCOL_FEATURES.
+
+ * VHOST_USER_SET_VRING_FLAG

SET_VRIHG_ENABLE would be a better name.

Okay.

Quoted 9 lines

docs/specs/vhost-user.txt Message types · line 335

Quoted 7 lines
docs/specs/vhost-user.txt +6 -0
+
+      Id: 18
+      Equivalent ioctl: N/A
+      Master payload: vring state description
+
+      Set the flag(1 for enable and 0 for disable) to signal slave to enable

Space before ( please.

Okay.

Quoted 9 lines
Quoted 9 lines
docs/specs/vhost-user.txt +2 -0
+      or disable corresponding virt queue. This request should be sent only
+      when the protocol feature bit VHOST_USER_PROTOCOL_F_VRING_FLAG is set.
hw/net/vhost_net.c +0 -0
index 141b557..7d9cc8d 100644
@@ -418,6 +418,19 @@ VHostNetState *get_vhost_net(NetClientState *nc)
 
Quoted 10 lines

hw/net/vhost_net.c VHostNetState *get_vhost_net(NetClientState *nc) · line 426

Quoted 8 lines
hw/net/vhost_net.c +6 -0
     return vhost_net;
 }
+
+int vhost_set_vring_flag(NetClientState *nc, int enable)
+{
+    if (nc->info->type == NET_CLIENT_OPTIONS_KIND_VHOST_USER) {
+        VHostNetState *net = get_vhost_net(nc);
+        const VhostOps *vhost_ops = net->dev.vhost_ops;

add an empty line after declaration pls.

Got it, thanks.

Quoted 6 lines

hw/net/vhost_net.c VHostNetState *get_vhost_net(NetClientState *nc) · line 428

Quoted 3 lines
hw/net/vhost_net.c +2 -0
+        if (vhost_ops->vhost_backend_mq_set_vring_flag)
+            return vhost_ops->vhost_backend_mq_set_vring_flag(&net->dev, enable);

Coding style violation:
line too long,

Quoted 7 lines

Yeah, 81 chars. I thougth that's kind of acceptable. But, anyway, since you pointed it out, I will fix it next time.

Quoted 1 line

and should use {}.

TBH, I seldom use {} for one statement, and you could see that a lot from this patch set. Is it a must in qemu community?

Yes it is. It's different from the kernel.

Quoted 146 lines

If so, I could fix them all.

Quoted 34 lines
Quoted 34 lines
hw/net/vhost_net.c +10 -0
+    }
+
+    return 0;
+}
+
 #else
 struct vhost_net *vhost_net_init(VhostNetOptions *options)
 {
@@ -463,4 +476,9 @@ VHostNetState *get_vhost_net(NetClientState *nc)
 {
     return 0;
 }
+
+int vhost_set_vring_flag(NetClientState *nc, int enable)
+{
+    return 0;
+}
 #endif
hw/net/virtio-net.c +1 -0
index 8d28e45..53f93b1 100644
@@ -407,6 +407,7 @@ static int peer_attach(VirtIONet *n, int index)
     }
 
     if (nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
+        vhost_set_vring_flag(nc->peer, 1);
         return 0;
     }
 
@@ -422,6 +423,7 @@ static int peer_detach(VirtIONet *n, int index)
     }
 
Quoted 8 lines

hw/net/virtio-net.c static int peer_detach(VirtIONet *n, int index) · line 429

Quoted 5 lines
hw/net/virtio-net.c +1 -0
     if (nc->peer->info->type !=  NET_CLIENT_OPTIONS_KIND_TAP) {
+        vhost_set_vring_flag(nc->peer, 0);
         return 0;
     }
 

makes no sense to call it for !== tap only. Either call this unconditionally, or just for vhost user.

Agreed. I will make it vhost-usre specific then.

Quoted 57 lines
Quoted 57 lines
hw/virtio/vhost-user.c +23 -2
index 11e46b5..ca6f7fa 100644
@@ -25,9 +25,10 @@
 
 #define VHOST_MEMORY_MAX_NREGIONS    8
 #define VHOST_USER_F_PROTOCOL_FEATURES 30
-#define VHOST_USER_PROTOCOL_FEATURE_MASK 0x1ULL
+#define VHOST_USER_PROTOCOL_FEATURE_MASK 0xfULL
 
-#define VHOST_USER_PROTOCOL_F_MQ    0
+#define VHOST_USER_PROTOCOL_F_MQ            0
+#define VHOST_USER_PROTOCOL_F_VRING_FLAG    1
 
 typedef enum VhostUserRequest {
     VHOST_USER_NONE = 0,
@@ -48,6 +49,7 @@ typedef enum VhostUserRequest {
     VHOST_USER_GET_PROTOCOL_FEATURES = 15,
     VHOST_USER_SET_PROTOCOL_FEATURES = 16,
     VHOST_USER_GET_QUEUE_NUM = 17,
+    VHOST_USER_SET_VRING_FLAG = 18,
     VHOST_USER_MAX
 } VhostUserRequest;
 
@@ -296,6 +298,11 @@ static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
         msg.addr.index += dev->vq_index;
         msg.size = sizeof(m.state);
         break;
+    case VHOST_USER_SET_VRING_FLAG:
+        msg.state.index = dev->vq_index;
+        msg.state.num = *(int*)arg;
+        msg.size = sizeof(msg.state);
+        break;
 
     case VHOST_USER_GET_VRING_BASE:
         memcpy(&msg.state, arg, sizeof(struct vhost_vring_state));
@@ -408,6 +415,22 @@ static int vhost_user_init(struct vhost_dev *dev, void *opaque)
     return 0;
 }
 
+static int vhost_user_set_vring_flag(struct vhost_dev *dev, int enable)
+{
+    int err;
+
+    assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
+
+    if (!(dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_VRING_FLAG)))
+        return -1;
+
+    err = vhost_user_call(dev, VHOST_USER_SET_VRING_FLAG, &enable);
+    if (err < 0)
+        return err;
+
+    return 0;
Quoted 8 lines

hw/virtio/vhost-user.c static int vhost_user_init(struct vhost_dev *dev, void *opaque) · line 436

Quoted 5 lines
hw/virtio/vhost-user.c +2 -0
+}
+
 static int vhost_user_cleanup(struct vhost_dev *dev)
 {
     assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);

In fact, it's a per VQ pair call the way it's coded. So pls fix it, call it for all VQs.

Let me try.

Quoted 21 lines
Quoted 21 lines
hw/virtio/vhost-user.c +3 -2
@@ -421,5 +444,6 @@ const VhostOps user_ops = {
         .backend_type = VHOST_BACKEND_TYPE_USER,
         .vhost_call = vhost_user_call,
         .vhost_backend_init = vhost_user_init,
-        .vhost_backend_cleanup = vhost_user_cleanup
-        };
+        .vhost_backend_cleanup = vhost_user_cleanup,
+        .vhost_backend_mq_set_vring_flag = vhost_user_set_vring_flag,
+};
include/hw/virtio/vhost-backend.h +1 -0
index e472f29..6fc76b7 100644
@@ -24,12 +24,14 @@ typedef int (*vhost_call)(struct vhost_dev *dev, unsigned long int request,
              void *arg);
 typedef int (*vhost_backend_init)(struct vhost_dev *dev, void *opaque);
 typedef int (*vhost_backend_cleanup)(struct vhost_dev *dev);
+typedef int (*vhost_backend_mq_set_vring_flag)(struct vhost_dev *dev, int enable);
 
Quoted 9 lines

include/hw/virtio/vhost-backend.h typedef int (*vhost_call)(struct vhost_dev *dev, unsigned long int request, · line 34

Quoted 6 lines
include/hw/virtio/vhost-backend.h +1 -0
 typedef struct VhostOps {
     VhostBackendType backend_type;
     vhost_call vhost_call;
     vhost_backend_init vhost_backend_init;
     vhost_backend_cleanup vhost_backend_cleanup;
+    vhost_backend_mq_set_vring_flag vhost_backend_mq_set_vring_flag;

That's quite ugly: mq is a vhost net thing. Why not enable each VQ?

Quoted 5 lines

To define it as `vhost_backend_enable_vring'?

Thanks.

	--yliu

Sure.

Quoted 16 lines
Quoted 16 lines
Quoted 16 lines
include/hw/virtio/vhost-backend.h +0 -0
 } VhostOps;
 
 extern const VhostOps user_ops;
include/net/vhost_net.h +1 -0
index 8db723e..3ec33ff 100644
@@ -28,4 +28,5 @@ bool vhost_net_virtqueue_pending(VHostNetState *net, int n);
 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
                               int idx, bool mask);
 VHostNetState *get_vhost_net(NetClientState *nc);
+int vhost_set_vring_flag(NetClientState * nc, int enable);
 #endif
Signature · 1 line
1.9.0
#29 lore raw

Re: [Qemu-devel] [PATCH 7/7] vhost-user: add a new message to disable/enable a specific virt queue.

In reply to Yuanhan Liu
Quoted 87 lines · Yuanhan Liu
Quoted 7 lines · Michael S. Tsirkin
Quoted 6 lines · Yuanhan Liu

From: Changchun Ouyang <[email protected]>

Add a new message, VHOST_USER_SET_VRING_FLAG, to enable and disable a specific virt queue, which is similar to attach/detach queue for tap device.

Quoted 8 lines
Quoted 6 lines

virtio driver on guest doesn't have to use max virt queue pair, it could enable any number of virt queue ranging from 1 to max virt queue pair.

Fine, but I would like to make this dependent on the MQ flag.

Reasonable.

Quoted 11 lines
Quoted 11 lines

---
 docs/specs/vhost-user.txt         | 10 ++++++++++
 hw/net/vhost_net.c                | 18 ++++++++++++++++++
 hw/net/virtio-net.c               |  2 ++
 hw/virtio/vhost-user.c            | 32 ++++++++++++++++++++++++++++----
 include/hw/virtio/vhost-backend.h |  2 ++
 include/net/vhost_net.h           |  1 +
 6 files changed, 61 insertions(+), 4 deletions(-)
Quoted 13 lines

docs/specs/vhost-user.txt Message types · line 329

Quoted 10 lines
docs/specs/vhost-user.txt +2 -0
index 99d79be..c2c38f3 100644
@@ -325,3 +325,13 @@ Message types
       Query how many queues the backend supports. This request should be
       sent only when VHOST_USER_PROTOCOL_F_MQ is set in quried protocol
       features by VHOST_USER_GET_PROTOCOL_FEATURES.
+
+ * VHOST_USER_SET_VRING_FLAG

SET_VRIHG_ENABLE would be a better name.

Okay.

Quoted 9 lines

docs/specs/vhost-user.txt Message types · line 335

Quoted 7 lines
docs/specs/vhost-user.txt +6 -0
+
+      Id: 18
+      Equivalent ioctl: N/A
+      Master payload: vring state description
+
+      Set the flag(1 for enable and 0 for disable) to signal slave to enable

Space before ( please.

Okay.

Quoted 9 lines
Quoted 9 lines
docs/specs/vhost-user.txt +2 -0
+      or disable corresponding virt queue. This request should be sent only
+      when the protocol feature bit VHOST_USER_PROTOCOL_F_VRING_FLAG is set.
hw/net/vhost_net.c +0 -0
index 141b557..7d9cc8d 100644
@@ -418,6 +418,19 @@ VHostNetState *get_vhost_net(NetClientState *nc)
 
Quoted 10 lines

hw/net/vhost_net.c VHostNetState *get_vhost_net(NetClientState *nc) · line 426

Quoted 8 lines
hw/net/vhost_net.c +6 -0
     return vhost_net;
 }
+
+int vhost_set_vring_flag(NetClientState *nc, int enable)
+{
+    if (nc->info->type == NET_CLIENT_OPTIONS_KIND_VHOST_USER) {
+        VHostNetState *net = get_vhost_net(nc);
+        const VhostOps *vhost_ops = net->dev.vhost_ops;

add an empty line after declaration pls.

Got it, thanks.

Quoted 6 lines

hw/net/vhost_net.c VHostNetState *get_vhost_net(NetClientState *nc) · line 428

Quoted 3 lines
hw/net/vhost_net.c +2 -0
+        if (vhost_ops->vhost_backend_mq_set_vring_flag)
+            return vhost_ops->vhost_backend_mq_set_vring_flag(&net->dev, enable);

Coding style violation:
line too long,

Quoted 7 lines

Yeah, 81 chars. I thougth that's kind of acceptable. But, anyway, since you pointed it out, I will fix it next time.

Quoted 1 line

and should use {}.

TBH, I seldom use {} for one statement, and you could see that a lot from this patch set. Is it a must in qemu community?

Yes it is. It's different from the kernel.

Quoted 146 lines

If so, I could fix them all.

Quoted 34 lines
Quoted 34 lines
hw/net/vhost_net.c +10 -0
+    }
+
+    return 0;
+}
+
 #else
 struct vhost_net *vhost_net_init(VhostNetOptions *options)
 {
@@ -463,4 +476,9 @@ VHostNetState *get_vhost_net(NetClientState *nc)
 {
     return 0;
 }
+
+int vhost_set_vring_flag(NetClientState *nc, int enable)
+{
+    return 0;
+}
 #endif
hw/net/virtio-net.c +1 -0
index 8d28e45..53f93b1 100644
@@ -407,6 +407,7 @@ static int peer_attach(VirtIONet *n, int index)
     }
 
     if (nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
+        vhost_set_vring_flag(nc->peer, 1);
         return 0;
     }
 
@@ -422,6 +423,7 @@ static int peer_detach(VirtIONet *n, int index)
     }
 
Quoted 8 lines

hw/net/virtio-net.c static int peer_detach(VirtIONet *n, int index) · line 429

Quoted 5 lines
hw/net/virtio-net.c +1 -0
     if (nc->peer->info->type !=  NET_CLIENT_OPTIONS_KIND_TAP) {
+        vhost_set_vring_flag(nc->peer, 0);
         return 0;
     }
 

makes no sense to call it for !== tap only. Either call this unconditionally, or just for vhost user.

Agreed. I will make it vhost-usre specific then.

Quoted 57 lines
Quoted 57 lines
hw/virtio/vhost-user.c +23 -2
index 11e46b5..ca6f7fa 100644
@@ -25,9 +25,10 @@
 
 #define VHOST_MEMORY_MAX_NREGIONS    8
 #define VHOST_USER_F_PROTOCOL_FEATURES 30
-#define VHOST_USER_PROTOCOL_FEATURE_MASK 0x1ULL
+#define VHOST_USER_PROTOCOL_FEATURE_MASK 0xfULL
 
-#define VHOST_USER_PROTOCOL_F_MQ    0
+#define VHOST_USER_PROTOCOL_F_MQ            0
+#define VHOST_USER_PROTOCOL_F_VRING_FLAG    1
 
 typedef enum VhostUserRequest {
     VHOST_USER_NONE = 0,
@@ -48,6 +49,7 @@ typedef enum VhostUserRequest {
     VHOST_USER_GET_PROTOCOL_FEATURES = 15,
     VHOST_USER_SET_PROTOCOL_FEATURES = 16,
     VHOST_USER_GET_QUEUE_NUM = 17,
+    VHOST_USER_SET_VRING_FLAG = 18,
     VHOST_USER_MAX
 } VhostUserRequest;
 
@@ -296,6 +298,11 @@ static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
         msg.addr.index += dev->vq_index;
         msg.size = sizeof(m.state);
         break;
+    case VHOST_USER_SET_VRING_FLAG:
+        msg.state.index = dev->vq_index;
+        msg.state.num = *(int*)arg;
+        msg.size = sizeof(msg.state);
+        break;
 
     case VHOST_USER_GET_VRING_BASE:
         memcpy(&msg.state, arg, sizeof(struct vhost_vring_state));
@@ -408,6 +415,22 @@ static int vhost_user_init(struct vhost_dev *dev, void *opaque)
     return 0;
 }
 
+static int vhost_user_set_vring_flag(struct vhost_dev *dev, int enable)
+{
+    int err;
+
+    assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
+
+    if (!(dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_VRING_FLAG)))
+        return -1;
+
+    err = vhost_user_call(dev, VHOST_USER_SET_VRING_FLAG, &enable);
+    if (err < 0)
+        return err;
+
+    return 0;
Quoted 8 lines

hw/virtio/vhost-user.c static int vhost_user_init(struct vhost_dev *dev, void *opaque) · line 436

Quoted 5 lines
hw/virtio/vhost-user.c +2 -0
+}
+
 static int vhost_user_cleanup(struct vhost_dev *dev)
 {
     assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);

In fact, it's a per VQ pair call the way it's coded. So pls fix it, call it for all VQs.

Let me try.

Quoted 21 lines
Quoted 21 lines
hw/virtio/vhost-user.c +3 -2
@@ -421,5 +444,6 @@ const VhostOps user_ops = {
         .backend_type = VHOST_BACKEND_TYPE_USER,
         .vhost_call = vhost_user_call,
         .vhost_backend_init = vhost_user_init,
-        .vhost_backend_cleanup = vhost_user_cleanup
-        };
+        .vhost_backend_cleanup = vhost_user_cleanup,
+        .vhost_backend_mq_set_vring_flag = vhost_user_set_vring_flag,
+};
include/hw/virtio/vhost-backend.h +1 -0
index e472f29..6fc76b7 100644
@@ -24,12 +24,14 @@ typedef int (*vhost_call)(struct vhost_dev *dev, unsigned long int request,
              void *arg);
 typedef int (*vhost_backend_init)(struct vhost_dev *dev, void *opaque);
 typedef int (*vhost_backend_cleanup)(struct vhost_dev *dev);
+typedef int (*vhost_backend_mq_set_vring_flag)(struct vhost_dev *dev, int enable);
 
Quoted 9 lines

include/hw/virtio/vhost-backend.h typedef int (*vhost_call)(struct vhost_dev *dev, unsigned long int request, · line 34

Quoted 6 lines
include/hw/virtio/vhost-backend.h +1 -0
 typedef struct VhostOps {
     VhostBackendType backend_type;
     vhost_call vhost_call;
     vhost_backend_init vhost_backend_init;
     vhost_backend_cleanup vhost_backend_cleanup;
+    vhost_backend_mq_set_vring_flag vhost_backend_mq_set_vring_flag;

That's quite ugly: mq is a vhost net thing. Why not enable each VQ?

Quoted 5 lines

To define it as `vhost_backend_enable_vring'?

Thanks.

	--yliu

Sure.

Quoted 16 lines
Quoted 16 lines
Quoted 16 lines
include/hw/virtio/vhost-backend.h +0 -0
 } VhostOps;
 
 extern const VhostOps user_ops;
include/net/vhost_net.h +1 -0
index 8db723e..3ec33ff 100644
@@ -28,4 +28,5 @@ bool vhost_net_virtqueue_pending(VHostNetState *net, int n);
 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
                               int idx, bool mask);
 VHostNetState *get_vhost_net(NetClientState *nc);
+int vhost_set_vring_flag(NetClientState * nc, int enable);
 #endif
Signature · 1 line
1.9.0
#30 lore raw

Re: [Qemu-devel] [PATCH 7/7] vhost-user: add a new message to disable/enable a specific virt queue.

In reply to Michael S. Tsirkin
Quoted 1 line · Michael S. Tsirkin

On Tue, Sep 08, 2015 at 03:38:47PM +0800, Yuanhan Liu wrote:

[snip ....]

Quoted 5 lines
Quoted 5 lines
hw/virtio/vhost-user.c +0 -0
index 11e46b5..ca6f7fa 100644
@@ -25,9 +25,10 @@

[snip ....]

Quoted 6 lines
Quoted 3 lines

static int vhost_user_cleanup(struct vhost_dev *dev)

 {
     assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);

In fact, it's a per VQ pair call the way it's coded. So pls fix it, call it for all VQs.

I'm trying to address this comment, and I'm a bit confused now: this call is based on one 'vhost_net' structure, which is related to one 'vhost_dev' structure. One 'vhost_dev', on the other hand, represents two VQs.

So, you are suggesting to call vhost_user_cleanup() twice at it's upper level?

	--yliu

---

hw/virtio/vhost.c +4 -1
index 2712c6f..6944543 100644
@@ -988,7 +988,10 @@ void vhost_dev_cleanup(struct vhost_dev *hdev)
     }
     g_free(hdev->mem);
     g_free(hdev->mem_sections);
-    hdev->vhost_ops->vhost_backend_cleanup(hdev);
+
+    for (i = 0; i < hdev->nvqs; i++) {
+        hdev->vhost_ops->vhost_backend_cleanup(hdev, i);
+    }
 }
 
 /* Stop processing guest IO notifications in qemu.