Skip to contentSkip to conversation

NIC vhost-vdpa state restore via Shadow CVQ

PATCH Superseded Eugenio Pérez 2 participants 22 messages Aug 9, 2022 – Aug 19, 2022 Networking

v8 of 10 Superseded Latest is v10 →

lore
Share & tools
lore
#0 lore raw

v8 cover NIC vhost-vdpa state restore via Shadow CVQ

CVQ of net vhost-vdpa devices can be intercepted since the addition of x-svq. The virtio-net device model is updated. The migration was blocked because although the state can be megrated between VMM it was not possible to restore on the destination NIC.

This series add support for SVQ to inject external messages without the guest's knowledge, so before the guest is resumed all the guest visible state is restored. It is done using standard CVQ messages, so the vhost-vdpa device does not need to learn how to restore it: As long as they have the feature, they know how to handle it.

This series needs fix [1] to be applied to achieve full live migration.

Thanks!

[1] https://lists.nongnu.org/archive/html/qemu-devel/2022-08/msg00325.html

v8:
- Rename NetClientInfo load to start, so is symmetrical with stop()
- Delete copy of device's in buffer at vhost_vdpa_net_load

v7:
- Remove accidental double free.

v6:
- Move map and unmap of the buffers to the start and stop of the device. This implies more callbacks on NetClientInfo, but simplifies the SVQ CVQ code.
- Not assume that in buffer is sizeof(virtio_net_ctrl_ack) in vhost_vdpa_net_cvq_add
- Reduce the number of changes from previous versions
- Delete unused memory barrier

v5:
- Rename s/start/load/
- Use independent NetClientInfo to only add load callback on cvq.
- Accept out sg instead of dev_buffers[] at vhost_vdpa_net_cvq_map_elem
- Use only out size instead of iovec dev_buffers to know if the descriptor is effectively available, allowing to delete artificial !NULL VirtQueueElement on vhost_svq_add call.

v4:
- Actually use NetClientInfo callback.

v3:
- Route vhost-vdpa start code through NetClientInfo callback.
- Delete extra vhost_net_stop_one() call.

v2:
- Fix SIGSEGV dereferencing SVQ when not in svq mode

v1 from RFC:
- Do not reorder DRIVER_OK & enable patches.
- Delete leftovers

Eugenio Pérez (12):
vhost: stop transfer elem ownership in vhost_handle_guest_kick
vhost: use SVQ element ndescs instead of opaque data for desc
validation
vhost: Delete useless read memory barrier
vhost: Do not depend on !NULL VirtQueueElement on vhost_svq_flush
vhost_net: Add NetClientInfo prepare callback
vhost_net: Add NetClientInfo stop callback
vdpa: add net_vhost_vdpa_cvq_info NetClientInfo
vdpa: Move command buffers map to start of net device
vdpa: extract vhost_vdpa_net_cvq_add from
vhost_vdpa_net_handle_ctrl_avail
vhost_net: add NetClientState->load() callback
vdpa: Add virtio-net mac address via CVQ at start
vdpa: Delete CVQ migration blocker

 include/hw/virtio/vhost-vdpa.h     |   1 -
 include/net/net.h                  |   6 +
 hw/net/vhost_net.c                 |  17 +++
 hw/virtio/vhost-shadow-virtqueue.c |  27 ++--
 hw/virtio/vhost-vdpa.c             |  14 --
 net/vhost-vdpa.c                   | 225 ++++++++++++++++++-----------
 6 files changed, 178 insertions(+), 112 deletions(-)
Signature · 1 line
2.31.1
#1 lore raw

PATCH v8 1/12 vhost: stop transfer elem ownership in vhost_handle_guest_kick

It was easier to allow vhost_svq_add to handle the memory. Now that we will allow qemu to add elements to a SVQ without the guest's knowledge, it's better to handle it in the caller.

---
 hw/virtio/vhost-shadow-virtqueue.c | 10 ++++------
 1 file changed, 4 insertions(+), 6 deletions(-)
hw/virtio/vhost-shadow-virtqueue.c +4 -6
index e4956728dd..ffd2b2c972 100644
@@ -233,9 +233,6 @@ static void vhost_svq_kick(VhostShadowVirtqueue *svq)
 /**
  * Add an element to a SVQ.
  *
- * The caller must check that there is enough slots for the new element. It
- * takes ownership of the element: In case of failure not ENOSPC, it is free.
- *
  * Return -EINVAL if element is invalid, -ENOSPC if dev queue is full
  */
 int vhost_svq_add(VhostShadowVirtqueue *svq, const struct iovec *out_sg,
@@ -252,7 +249,6 @@ int vhost_svq_add(VhostShadowVirtqueue *svq, const struct iovec *out_sg,
 
     ok = vhost_svq_add_split(svq, out_sg, out_num, in_sg, in_num, &qemu_head);
     if (unlikely(!ok)) {
-        g_free(elem);
         return -EINVAL;
     }
 
@@ -293,7 +289,7 @@ static void vhost_handle_guest_kick(VhostShadowVirtqueue *svq)
         virtio_queue_set_notification(svq->vq, false);
 
         while (true) {
-            VirtQueueElement *elem;
+            g_autofree VirtQueueElement *elem;
             int r;
 
             if (svq->next_guest_avail_elem) {
@@ -324,12 +320,14 @@ static void vhost_handle_guest_kick(VhostShadowVirtqueue *svq)
                      * queue the current guest descriptor and ignore kicks
                      * until some elements are used.
                      */
-                    svq->next_guest_avail_elem = elem;
+                    svq->next_guest_avail_elem = g_steal_pointer(&elem);
                 }
 
                 /* VQ is full or broken, just return and ignore kicks */
                 return;
             }
+            /* elem belongs to SVQ or external caller now */
+            elem = NULL;
         }
 
         virtio_queue_set_notification(svq->vq, true);
Signature · 1 line
2.31.1
#2 lore raw

PATCH v8 2/12 vhost: use SVQ element ndescs instead of opaque data for desc validation

In reply to Eugenio Pérez

Since we're going to allow SVQ to add elements without the guest's knowledge and without its own VirtQueueElement, it's easier to check if an element is a valid head checking a different thing than the VirtQueueElement.

---
 hw/virtio/vhost-shadow-virtqueue.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
hw/virtio/vhost-shadow-virtqueue.c +2 -1
index ffd2b2c972..e6eebd0e8d 100644
@@ -414,7 +414,7 @@ static VirtQueueElement *vhost_svq_get_buf(VhostShadowVirtqueue *svq,
         return NULL;
     }
 
-    if (unlikely(!svq->desc_state[used_elem.id].elem)) {
+    if (unlikely(!svq->desc_state[used_elem.id].ndescs)) {
         qemu_log_mask(LOG_GUEST_ERROR,
             "Device %s says index %u is used, but it was not available",
             svq->vdev->name, used_elem.id);
@@ -422,6 +422,7 @@ static VirtQueueElement *vhost_svq_get_buf(VhostShadowVirtqueue *svq,
     }
 
     num = svq->desc_state[used_elem.id].ndescs;
+    svq->desc_state[used_elem.id].ndescs = 0;
     last_used_chain = vhost_svq_last_desc_of_chain(svq, num, used_elem.id);
     svq->desc_next[last_used_chain] = svq->free_head;
     svq->free_head = used_elem.id;
Signature · 1 line
2.31.1
#3 lore raw

PATCH v8 3/12 vhost: Delete useless read memory barrier

In reply to Eugenio Pérez

As discussed in previous series [1], this memory barrier is useless with the atomic read of used idx at vhost_svq_more_used. Deleting it.

[1] https://lists.nongnu.org/archive/html/qemu-devel/2022-07/msg02616.html

---
 hw/virtio/vhost-shadow-virtqueue.c | 3 ---
 1 file changed, 3 deletions(-)
hw/virtio/vhost-shadow-virtqueue.c +0 -3
index e6eebd0e8d..1b49bf54f2 100644
@@ -509,9 +509,6 @@ size_t vhost_svq_poll(VhostShadowVirtqueue *svq)
         if (unlikely(g_get_monotonic_time() - start_us > 10e6)) {
             return 0;
         }
-
-        /* Make sure we read new used_idx */
-        smp_rmb();
     } while (true);
 }
 
Signature · 1 line
2.31.1
#4 lore raw

PATCH v8 4/12 vhost: Do not depend on !NULL VirtQueueElement on vhost_svq_flush

In reply to Eugenio Pérez

Since QEMU will be able to inject new elements on CVQ to restore the state, we need not to depend on a VirtQueueElement to know if a new element has been used by the device or not. Instead of check that, check if there are new elements only using used idx on vhost_svq_flush.

---
v6: Change less from the previous function

---
 hw/virtio/vhost-shadow-virtqueue.c | 11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)
hw/virtio/vhost-shadow-virtqueue.c +7 -4
index 1b49bf54f2..f863b08627 100644
@@ -499,17 +499,20 @@ static void vhost_svq_flush(VhostShadowVirtqueue *svq,
 size_t vhost_svq_poll(VhostShadowVirtqueue *svq)
 {
     int64_t start_us = g_get_monotonic_time();
+    uint32_t len;
+
     do {
-        uint32_t len;
-        VirtQueueElement *elem = vhost_svq_get_buf(svq, &len);
-        if (elem) {
-            return len;
+        if (vhost_svq_more_used(svq)) {
+            break;
         }
 
         if (unlikely(g_get_monotonic_time() - start_us > 10e6)) {
             return 0;
         }
     } while (true);
+
+    vhost_svq_get_buf(svq, &len);
+    return len;
 }
 
 /**
Signature · 1 line
2.31.1
#5 lore raw

PATCH v8 5/12 vhost_net: Add NetClientInfo prepare callback

In reply to Eugenio Pérez

This is used by the backend to perform actions before the device is started.

In particular, vdpa net use it to map CVQ buffers to the device, so it can send control commands using them.

---
v8: Rename NetClientInfo prepare callback to start, so it aligns with
future "stop"

---
 include/net/net.h  | 2 ++
 hw/net/vhost_net.c | 7 +++++++
 2 files changed, 9 insertions(+)
include/net/net.h +2 -0
index 523136c7ac..ad9e80083a 100644
@@ -44,6 +44,7 @@ typedef struct NICConf {
 
 typedef void (NetPoll)(NetClientState *, bool enable);
 typedef bool (NetCanReceive)(NetClientState *);
+typedef int (NetStart)(NetClientState *);
 typedef ssize_t (NetReceive)(NetClientState *, const uint8_t *, size_t);
 typedef ssize_t (NetReceiveIOV)(NetClientState *, const struct iovec *, int);
 typedef void (NetCleanup) (NetClientState *);
@@ -71,6 +72,7 @@ typedef struct NetClientInfo {
     NetReceive *receive_raw;
     NetReceiveIOV *receive_iov;
     NetCanReceive *can_receive;
+    NetStart *start;
     NetCleanup *cleanup;
     LinkStatusChanged *link_status_changed;
     QueryRxFilter *query_rx_filter;
hw/net/vhost_net.c +7 -0
index ccac5b7a64..2e0baeba26 100644
@@ -244,6 +244,13 @@ static int vhost_net_start_one(struct vhost_net *net,
     struct vhost_vring_file file = { };
     int r;
 
+    if (net->nc->info->start) {
+        r = net->nc->info->start(net->nc);
+        if (r < 0) {
+            return r;
+        }
+    }
+
     r = vhost_dev_enable_notifiers(&net->dev, dev);
     if (r < 0) {
         goto fail_notifiers;
Signature · 1 line
2.31.1
Acked-by on v8 5/12
#6 lore raw

Re: [PATCH v8 05/12] vhost_net: Add NetClientInfo prepare callback

Quoted 9 lines · Eugenio Pérez

This is used by the backend to perform actions before the device is started.

In particular, vdpa net use it to map CVQ buffers to the device, so it can send control commands using them.

---

So the title still calls it "prepare".

Other than this.

Quoted 48 lines

v8: Rename NetClientInfo prepare callback to start, so it aligns with
future "stop"

---
 include/net/net.h  | 2 ++
 hw/net/vhost_net.c | 7 +++++++
 2 files changed, 9 insertions(+)
include/net/net.h +2 -0
index 523136c7ac..ad9e80083a 100644
@@ -44,6 +44,7 @@ typedef struct NICConf {
 
 typedef void (NetPoll)(NetClientState *, bool enable);
 typedef bool (NetCanReceive)(NetClientState *);
+typedef int (NetStart)(NetClientState *);
 typedef ssize_t (NetReceive)(NetClientState *, const uint8_t *, size_t);
 typedef ssize_t (NetReceiveIOV)(NetClientState *, const struct iovec *, int);
 typedef void (NetCleanup) (NetClientState *);
@@ -71,6 +72,7 @@ typedef struct NetClientInfo {
     NetReceive *receive_raw;
     NetReceiveIOV *receive_iov;
     NetCanReceive *can_receive;
+    NetStart *start;
     NetCleanup *cleanup;
     LinkStatusChanged *link_status_changed;
     QueryRxFilter *query_rx_filter;
hw/net/vhost_net.c +7 -1
index ccac5b7a64..2e0baeba26 100644
@@ -244,6 +244,13 @@ static int vhost_net_start_one(struct vhost_net *net,
     struct vhost_vring_file file = { };
     int r;
 
+    if (net->nc->info->start) {
+        r = net->nc->info->start(net->nc);
+        if (r < 0) {
+            return r;
+        }
+    }
+
     r = vhost_dev_enable_notifiers(&net->dev, dev);
     if (r < 0) {
         goto fail_notifiers;
--

2.31.1

#7 lore raw

PATCH v8 6/12 vhost_net: Add NetClientInfo stop callback

In reply to Eugenio Pérez

Used by the backend to perform actions after the device is stopped.

In particular, vdpa net use it to unmap CVQ buffers to the device, cleaning the actions performend in prepare().

---
 include/net/net.h  | 2 ++
 hw/net/vhost_net.c | 3 +++
 2 files changed, 5 insertions(+)
include/net/net.h +2 -0
index ad9e80083a..476ad45b9a 100644
@@ -45,6 +45,7 @@ typedef struct NICConf {
 typedef void (NetPoll)(NetClientState *, bool enable);
 typedef bool (NetCanReceive)(NetClientState *);
 typedef int (NetStart)(NetClientState *);
+typedef void (NetStop)(NetClientState *);
 typedef ssize_t (NetReceive)(NetClientState *, const uint8_t *, size_t);
 typedef ssize_t (NetReceiveIOV)(NetClientState *, const struct iovec *, int);
 typedef void (NetCleanup) (NetClientState *);
@@ -73,6 +74,7 @@ typedef struct NetClientInfo {
     NetReceiveIOV *receive_iov;
     NetCanReceive *can_receive;
     NetStart *start;
+    NetStop *stop;
     NetCleanup *cleanup;
     LinkStatusChanged *link_status_changed;
     QueryRxFilter *query_rx_filter;
hw/net/vhost_net.c +3 -0
index 2e0baeba26..9d4b334453 100644
@@ -320,6 +320,9 @@ static void vhost_net_stop_one(struct vhost_net *net,
         net->nc->info->poll(net->nc, true);
     }
     vhost_dev_stop(&net->dev, dev);
+    if (net->nc->info->stop) {
+        net->nc->info->stop(net->nc);
+    }
     vhost_dev_disable_notifiers(&net->dev, dev);
 }
 
Signature · 1 line
2.31.1
Acked-by on v8 6/12
#8 lore raw

Re: [PATCH v8 06/12] vhost_net: Add NetClientInfo stop callback

Quoted 5 lines · Eugenio Pérez

Used by the backend to perform actions after the device is stopped.

In particular, vdpa net use it to unmap CVQ buffers to the device, cleaning the actions performend in prepare().

Should be "performed".

Other than this,

Quoted 44 lines
---
 include/net/net.h  | 2 ++
 hw/net/vhost_net.c | 3 +++
 2 files changed, 5 insertions(+)
include/net/net.h +2 -0
index ad9e80083a..476ad45b9a 100644
@@ -45,6 +45,7 @@ typedef struct NICConf {
 typedef void (NetPoll)(NetClientState *, bool enable);
 typedef bool (NetCanReceive)(NetClientState *);
 typedef int (NetStart)(NetClientState *);
+typedef void (NetStop)(NetClientState *);
 typedef ssize_t (NetReceive)(NetClientState *, const uint8_t *, size_t);
 typedef ssize_t (NetReceiveIOV)(NetClientState *, const struct iovec *, int);
 typedef void (NetCleanup) (NetClientState *);
@@ -73,6 +74,7 @@ typedef struct NetClientInfo {
     NetReceiveIOV *receive_iov;
     NetCanReceive *can_receive;
     NetStart *start;
+    NetStop *stop;
     NetCleanup *cleanup;
     LinkStatusChanged *link_status_changed;
     QueryRxFilter *query_rx_filter;
hw/net/vhost_net.c +3 -1
index 2e0baeba26..9d4b334453 100644
@@ -320,6 +320,9 @@ static void vhost_net_stop_one(struct vhost_net *net,
         net->nc->info->poll(net->nc, true);
     }
     vhost_dev_stop(&net->dev, dev);
+    if (net->nc->info->stop) {
+        net->nc->info->stop(net->nc);
+    }
     vhost_dev_disable_notifiers(&net->dev, dev);
 }
 
--

2.31.1

#9 lore raw

PATCH v8 7/12 vdpa: add net_vhost_vdpa_cvq_info NetClientInfo

In reply to Eugenio Pérez

Next patches will add a new info callback to restore NIC status through CVQ. Since only the CVQ vhost device is needed, create it with a new NetClientInfo.

---
v5: Create a new NetClientInfo instead of reusing the dataplane one.

---
 net/vhost-vdpa.c | 12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)
net/vhost-vdpa.c +11 -1
index ac1810723c..55e8a39a56 100644
@@ -334,6 +334,16 @@ static bool vhost_vdpa_net_cvq_map_elem(VhostVDPAState *s,
     return true;
 }
 
+static NetClientInfo net_vhost_vdpa_cvq_info = {
+    .type = NET_CLIENT_DRIVER_VHOST_VDPA,
+    .size = sizeof(VhostVDPAState),
+    .receive = vhost_vdpa_receive,
+    .cleanup = vhost_vdpa_cleanup,
+    .has_vnet_hdr = vhost_vdpa_has_vnet_hdr,
+    .has_ufo = vhost_vdpa_has_ufo,
+    .check_peer_type = vhost_vdpa_check_peer_type,
+};
+
 /**
  * Do not forward commands not supported by SVQ. Otherwise, the device could
  * accept it and qemu would not know how to update the device model.
@@ -475,7 +485,7 @@ static NetClientState *net_vhost_vdpa_init(NetClientState *peer,
         nc = qemu_new_net_client(&net_vhost_vdpa_info, peer, device,
                                  name);
     } else {
-        nc = qemu_new_net_control_client(&net_vhost_vdpa_info, peer,
+        nc = qemu_new_net_control_client(&net_vhost_vdpa_cvq_info, peer,
                                          device, name);
     }
     snprintf(nc->info_str, sizeof(nc->info_str), TYPE_VHOST_VDPA);
Signature · 1 line
2.31.1
Acked-by on v8 7/12
#10 lore raw

Re: [PATCH v8 07/12] vdpa: add net_vhost_vdpa_cvq_info NetClientInfo

Quoted 6 lines · Eugenio Pérez

Next patches will add a new info callback to restore NIC status through CVQ. Since only the CVQ vhost device is needed, create it with a new NetClientInfo.

Quoted 39 lines

---
v5: Create a new NetClientInfo instead of reusing the dataplane one.

---
 net/vhost-vdpa.c | 12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)
net/vhost-vdpa.c +11 -2
index ac1810723c..55e8a39a56 100644
@@ -334,6 +334,16 @@ static bool vhost_vdpa_net_cvq_map_elem(VhostVDPAState *s,
     return true;
 }
 
+static NetClientInfo net_vhost_vdpa_cvq_info = {
+    .type = NET_CLIENT_DRIVER_VHOST_VDPA,
+    .size = sizeof(VhostVDPAState),
+    .receive = vhost_vdpa_receive,
+    .cleanup = vhost_vdpa_cleanup,
+    .has_vnet_hdr = vhost_vdpa_has_vnet_hdr,
+    .has_ufo = vhost_vdpa_has_ufo,
+    .check_peer_type = vhost_vdpa_check_peer_type,
+};
+
 /**
  * Do not forward commands not supported by SVQ. Otherwise, the device could
  * accept it and qemu would not know how to update the device model.
@@ -475,7 +485,7 @@ static NetClientState *net_vhost_vdpa_init(NetClientState *peer,
         nc = qemu_new_net_client(&net_vhost_vdpa_info, peer, device,
                                  name);
     } else {
-        nc = qemu_new_net_control_client(&net_vhost_vdpa_info, peer,
+        nc = qemu_new_net_control_client(&net_vhost_vdpa_cvq_info, peer,
                                          device, name);
     }
     snprintf(nc->info_str, sizeof(nc->info_str), TYPE_VHOST_VDPA);
--

2.31.1

#11 lore raw

PATCH v8 8/12 vdpa: Move command buffers map to start of net device

In reply to Eugenio Pérez

As this series will reuse them to restore the device state at the end of a migration (or a device start), let's allocate only once at the device start so we don't duplicate their map and unmap.

---
 net/vhost-vdpa.c | 123 ++++++++++++++++++++++-------------------------
 1 file changed, 58 insertions(+), 65 deletions(-)
net/vhost-vdpa.c +58 -65
index 55e8a39a56..5254b58bdc 100644
@@ -263,29 +263,20 @@ static size_t vhost_vdpa_net_cvq_cmd_page_len(void)
     return ROUND_UP(vhost_vdpa_net_cvq_cmd_len(), qemu_real_host_page_size());
 }
 
-/** Copy and map a guest buffer. */
-static bool vhost_vdpa_cvq_map_buf(struct vhost_vdpa *v,
-                                   const struct iovec *out_data,
-                                   size_t out_num, size_t data_len, void *buf,
-                                   size_t *written, bool write)
+/** Map CVQ buffer. */
+static int vhost_vdpa_cvq_map_buf(struct vhost_vdpa *v, void *buf, size_t size,
+                                  bool write)
 {
     DMAMap map = {};
     int r;
 
-    if (unlikely(!data_len)) {
-        qemu_log_mask(LOG_GUEST_ERROR, "%s: invalid legnth of %s buffer\n",
-                      __func__, write ? "in" : "out");
-        return false;
-    }
-
-    *written = iov_to_buf(out_data, out_num, 0, buf, data_len);
     map.translated_addr = (hwaddr)(uintptr_t)buf;
-    map.size = vhost_vdpa_net_cvq_cmd_page_len() - 1;
+    map.size = size - 1;
     map.perm = write ? IOMMU_RW : IOMMU_RO,
     r = vhost_iova_tree_map_alloc(v->iova_tree, &map);
     if (unlikely(r != IOVA_OK)) {
         error_report("Cannot map injected element");
-        return false;
+        return r;
     }
 
     r = vhost_vdpa_dma_map(v, map.iova, vhost_vdpa_net_cvq_cmd_page_len(), buf,
@@ -294,50 +285,58 @@ static bool vhost_vdpa_cvq_map_buf(struct vhost_vdpa *v,
         goto dma_map_err;
     }
 
-    return true;
+    return 0;
 
 dma_map_err:
     vhost_iova_tree_remove(v->iova_tree, &map);
-    return false;
+    return r;
 }
 
-/**
- * Copy the guest element into a dedicated buffer suitable to be sent to NIC
- *
- * @iov: [0] is the out buffer, [1] is the in one
- */
-static bool vhost_vdpa_net_cvq_map_elem(VhostVDPAState *s,
-                                        VirtQueueElement *elem,
-                                        struct iovec *iov)
+static int vhost_vdpa_net_cvq_start(NetClientState *nc)
 {
-    size_t in_copied;
-    bool ok;
+    VhostVDPAState *s;
+    int r;
 
-    iov[0].iov_base = s->cvq_cmd_out_buffer;
-    ok = vhost_vdpa_cvq_map_buf(&s->vhost_vdpa, elem->out_sg, elem->out_num,
-                                vhost_vdpa_net_cvq_cmd_len(), iov[0].iov_base,
-                                &iov[0].iov_len, false);
-    if (unlikely(!ok)) {
-        return false;
+    assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA);
+
+    s = DO_UPCAST(VhostVDPAState, nc, nc);
+    if (!s->vhost_vdpa.shadow_vqs_enabled) {
+        return 0;
     }
 
-    iov[1].iov_base = s->cvq_cmd_in_buffer;
-    ok = vhost_vdpa_cvq_map_buf(&s->vhost_vdpa, NULL, 0,
-                                sizeof(virtio_net_ctrl_ack), iov[1].iov_base,
-                                &in_copied, true);
-    if (unlikely(!ok)) {
+    r = vhost_vdpa_cvq_map_buf(&s->vhost_vdpa, s->cvq_cmd_out_buffer,
+                               vhost_vdpa_net_cvq_cmd_page_len(), false);
+    if (unlikely(r < 0)) {
+        return r;
+    }
+
+    r = vhost_vdpa_cvq_map_buf(&s->vhost_vdpa, s->cvq_cmd_in_buffer,
+                               vhost_vdpa_net_cvq_cmd_page_len(), true);
+    if (unlikely(r < 0)) {
         vhost_vdpa_cvq_unmap_buf(&s->vhost_vdpa, s->cvq_cmd_out_buffer);
-        return false;
     }
 
-    iov[1].iov_len = sizeof(virtio_net_ctrl_ack);
-    return true;
+    return r;
+}
+
+static void vhost_vdpa_net_cvq_stop(NetClientState *nc)
+{
+    VhostVDPAState *s = DO_UPCAST(VhostVDPAState, nc, nc);
+
+    assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA);
+
+    if (s->vhost_vdpa.shadow_vqs_enabled) {
+        vhost_vdpa_cvq_unmap_buf(&s->vhost_vdpa, s->cvq_cmd_out_buffer);
+        vhost_vdpa_cvq_unmap_buf(&s->vhost_vdpa, s->cvq_cmd_in_buffer);
+    }
 }
 
 static NetClientInfo net_vhost_vdpa_cvq_info = {
     .type = NET_CLIENT_DRIVER_VHOST_VDPA,
     .size = sizeof(VhostVDPAState),
     .receive = vhost_vdpa_receive,
+    .start = vhost_vdpa_net_cvq_start,
+    .stop = vhost_vdpa_net_cvq_stop,
     .cleanup = vhost_vdpa_cleanup,
     .has_vnet_hdr = vhost_vdpa_has_vnet_hdr,
     .has_ufo = vhost_vdpa_has_ufo,
@@ -348,19 +347,17 @@ static NetClientInfo net_vhost_vdpa_cvq_info = {
  * Do not forward commands not supported by SVQ. Otherwise, the device could
  * accept it and qemu would not know how to update the device model.
  */
-static bool vhost_vdpa_net_cvq_validate_cmd(const struct iovec *out,
-                                            size_t out_num)
+static bool vhost_vdpa_net_cvq_validate_cmd(const void *out_buf, size_t len)
 {
     struct virtio_net_ctrl_hdr ctrl;
-    size_t n;
 
-    n = iov_to_buf(out, out_num, 0, &ctrl, sizeof(ctrl));
-    if (unlikely(n < sizeof(ctrl))) {
+    if (unlikely(len < sizeof(ctrl))) {
         qemu_log_mask(LOG_GUEST_ERROR,
-                      "%s: invalid legnth of out buffer %zu\n", __func__, n);
+                      "%s: invalid legnth of out buffer %zu\n", __func__, len);
         return false;
     }
 
+    memcpy(&ctrl, out_buf, sizeof(ctrl));
     switch (ctrl.class) {
     case VIRTIO_NET_CTRL_MAC:
         switch (ctrl.cmd) {
@@ -392,10 +389,14 @@ static int vhost_vdpa_net_handle_ctrl_avail(VhostShadowVirtqueue *svq,
     VhostVDPAState *s = opaque;
     size_t in_len, dev_written;
     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
-    /* out and in buffers sent to the device */
-    struct iovec dev_buffers[2] = {
-        { .iov_base = s->cvq_cmd_out_buffer },
-        { .iov_base = s->cvq_cmd_in_buffer },
+    /* Out buffer sent to both the vdpa device and the device model */
+    struct iovec out = {
+        .iov_base = s->cvq_cmd_out_buffer,
+    };
+    /* In buffer sent to the device */
+    const struct iovec dev_in = {
+        .iov_base = s->cvq_cmd_in_buffer,
+        .iov_len = sizeof(virtio_net_ctrl_ack),
     };
     /* in buffer used for device model */
     const struct iovec in = {
@@ -405,17 +406,15 @@ static int vhost_vdpa_net_handle_ctrl_avail(VhostShadowVirtqueue *svq,
     int r = -EINVAL;
     bool ok;
 
-    ok = vhost_vdpa_net_cvq_map_elem(s, elem, dev_buffers);
-    if (unlikely(!ok)) {
-        goto out;
-    }
-
-    ok = vhost_vdpa_net_cvq_validate_cmd(&dev_buffers[0], 1);
+    out.iov_len = iov_to_buf(elem->out_sg, elem->out_num, 0,
+                             s->cvq_cmd_out_buffer,
+                             vhost_vdpa_net_cvq_cmd_len());
+    ok = vhost_vdpa_net_cvq_validate_cmd(s->cvq_cmd_out_buffer, out.iov_len);
     if (unlikely(!ok)) {
         goto out;
     }
 
-    r = vhost_svq_add(svq, &dev_buffers[0], 1, &dev_buffers[1], 1, elem);
+    r = vhost_svq_add(svq, &out, 1, &dev_in, 1, elem);
     if (unlikely(r != 0)) {
         if (unlikely(r == -ENOSPC)) {
             qemu_log_mask(LOG_GUEST_ERROR, "%s: No space on device queue\n",
@@ -435,13 +434,13 @@ static int vhost_vdpa_net_handle_ctrl_avail(VhostShadowVirtqueue *svq,
         goto out;
     }
 
-    memcpy(&status, dev_buffers[1].iov_base, sizeof(status));
+    memcpy(&status, s->cvq_cmd_in_buffer, sizeof(status));
     if (status != VIRTIO_NET_OK) {
         goto out;
     }
 
     status = VIRTIO_NET_ERR;
-    virtio_net_handle_ctrl_iov(svq->vdev, &in, 1, dev_buffers, 1);
+    virtio_net_handle_ctrl_iov(svq->vdev, &in, 1, &out, 1);
     if (status != VIRTIO_NET_OK) {
         error_report("Bad CVQ processing in model");
     }
@@ -454,12 +453,6 @@ out:
     }
     vhost_svq_push_elem(svq, elem, MIN(in_len, sizeof(status)));
     g_free(elem);
-    if (dev_buffers[0].iov_base) {
-        vhost_vdpa_cvq_unmap_buf(&s->vhost_vdpa, dev_buffers[0].iov_base);
-    }
-    if (dev_buffers[1].iov_base) {
-        vhost_vdpa_cvq_unmap_buf(&s->vhost_vdpa, dev_buffers[1].iov_base);
-    }
     return r;
 }
 
Signature · 1 line
2.31.1
Acked-by on v8 8/12
#12 lore raw

Re: [PATCH v8 08/12] vdpa: Move command buffers map to start of net device

Quoted 6 lines · Eugenio Pérez

As this series will reuse them to restore the device state at the end of a migration (or a device start), let's allocate only once at the device start so we don't duplicate their map and unmap.

Quoted 226 lines
---
 net/vhost-vdpa.c | 123 ++++++++++++++++++++++-------------------------
 1 file changed, 58 insertions(+), 65 deletions(-)
net/vhost-vdpa.c +58 -66
index 55e8a39a56..5254b58bdc 100644
@@ -263,29 +263,20 @@ static size_t vhost_vdpa_net_cvq_cmd_page_len(void)
     return ROUND_UP(vhost_vdpa_net_cvq_cmd_len(), qemu_real_host_page_size());
 }
 
-/** Copy and map a guest buffer. */
-static bool vhost_vdpa_cvq_map_buf(struct vhost_vdpa *v,
-                                   const struct iovec *out_data,
-                                   size_t out_num, size_t data_len, void *buf,
-                                   size_t *written, bool write)
+/** Map CVQ buffer. */
+static int vhost_vdpa_cvq_map_buf(struct vhost_vdpa *v, void *buf, size_t size,
+                                  bool write)
 {
     DMAMap map = {};
     int r;
 
-    if (unlikely(!data_len)) {
-        qemu_log_mask(LOG_GUEST_ERROR, "%s: invalid legnth of %s buffer\n",
-                      __func__, write ? "in" : "out");
-        return false;
-    }
-
-    *written = iov_to_buf(out_data, out_num, 0, buf, data_len);
     map.translated_addr = (hwaddr)(uintptr_t)buf;
-    map.size = vhost_vdpa_net_cvq_cmd_page_len() - 1;
+    map.size = size - 1;
     map.perm = write ? IOMMU_RW : IOMMU_RO,
     r = vhost_iova_tree_map_alloc(v->iova_tree, &map);
     if (unlikely(r != IOVA_OK)) {
         error_report("Cannot map injected element");
-        return false;
+        return r;
     }
 
     r = vhost_vdpa_dma_map(v, map.iova, vhost_vdpa_net_cvq_cmd_page_len(), buf,
@@ -294,50 +285,58 @@ static bool vhost_vdpa_cvq_map_buf(struct vhost_vdpa *v,
         goto dma_map_err;
     }
 
-    return true;
+    return 0;
 
 dma_map_err:
     vhost_iova_tree_remove(v->iova_tree, &map);
-    return false;
+    return r;
 }
 
-/**
- * Copy the guest element into a dedicated buffer suitable to be sent to NIC
- *
- * @iov: [0] is the out buffer, [1] is the in one
- */
-static bool vhost_vdpa_net_cvq_map_elem(VhostVDPAState *s,
-                                        VirtQueueElement *elem,
-                                        struct iovec *iov)
+static int vhost_vdpa_net_cvq_start(NetClientState *nc)
 {
-    size_t in_copied;
-    bool ok;
+    VhostVDPAState *s;
+    int r;
 
-    iov[0].iov_base = s->cvq_cmd_out_buffer;
-    ok = vhost_vdpa_cvq_map_buf(&s->vhost_vdpa, elem->out_sg, elem->out_num,
-                                vhost_vdpa_net_cvq_cmd_len(), iov[0].iov_base,
-                                &iov[0].iov_len, false);
-    if (unlikely(!ok)) {
-        return false;
+    assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA);
+
+    s = DO_UPCAST(VhostVDPAState, nc, nc);
+    if (!s->vhost_vdpa.shadow_vqs_enabled) {
+        return 0;
     }
 
-    iov[1].iov_base = s->cvq_cmd_in_buffer;
-    ok = vhost_vdpa_cvq_map_buf(&s->vhost_vdpa, NULL, 0,
-                                sizeof(virtio_net_ctrl_ack), iov[1].iov_base,
-                                &in_copied, true);
-    if (unlikely(!ok)) {
+    r = vhost_vdpa_cvq_map_buf(&s->vhost_vdpa, s->cvq_cmd_out_buffer,
+                               vhost_vdpa_net_cvq_cmd_page_len(), false);
+    if (unlikely(r < 0)) {
+        return r;
+    }
+
+    r = vhost_vdpa_cvq_map_buf(&s->vhost_vdpa, s->cvq_cmd_in_buffer,
+                               vhost_vdpa_net_cvq_cmd_page_len(), true);
+    if (unlikely(r < 0)) {
         vhost_vdpa_cvq_unmap_buf(&s->vhost_vdpa, s->cvq_cmd_out_buffer);
-        return false;
     }
 
-    iov[1].iov_len = sizeof(virtio_net_ctrl_ack);
-    return true;
+    return r;
+}
+
+static void vhost_vdpa_net_cvq_stop(NetClientState *nc)
+{
+    VhostVDPAState *s = DO_UPCAST(VhostVDPAState, nc, nc);
+
+    assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA);
+
+    if (s->vhost_vdpa.shadow_vqs_enabled) {
+        vhost_vdpa_cvq_unmap_buf(&s->vhost_vdpa, s->cvq_cmd_out_buffer);
+        vhost_vdpa_cvq_unmap_buf(&s->vhost_vdpa, s->cvq_cmd_in_buffer);
+    }
 }
 
 static NetClientInfo net_vhost_vdpa_cvq_info = {
     .type = NET_CLIENT_DRIVER_VHOST_VDPA,
     .size = sizeof(VhostVDPAState),
     .receive = vhost_vdpa_receive,
+    .start = vhost_vdpa_net_cvq_start,
+    .stop = vhost_vdpa_net_cvq_stop,
     .cleanup = vhost_vdpa_cleanup,
     .has_vnet_hdr = vhost_vdpa_has_vnet_hdr,
     .has_ufo = vhost_vdpa_has_ufo,
@@ -348,19 +347,17 @@ static NetClientInfo net_vhost_vdpa_cvq_info = {
  * Do not forward commands not supported by SVQ. Otherwise, the device could
  * accept it and qemu would not know how to update the device model.
  */
-static bool vhost_vdpa_net_cvq_validate_cmd(const struct iovec *out,
-                                            size_t out_num)
+static bool vhost_vdpa_net_cvq_validate_cmd(const void *out_buf, size_t len)
 {
     struct virtio_net_ctrl_hdr ctrl;
-    size_t n;
 
-    n = iov_to_buf(out, out_num, 0, &ctrl, sizeof(ctrl));
-    if (unlikely(n < sizeof(ctrl))) {
+    if (unlikely(len < sizeof(ctrl))) {
         qemu_log_mask(LOG_GUEST_ERROR,
-                      "%s: invalid legnth of out buffer %zu\n", __func__, n);
+                      "%s: invalid legnth of out buffer %zu\n", __func__, len);
         return false;
     }
 
+    memcpy(&ctrl, out_buf, sizeof(ctrl));
     switch (ctrl.class) {
     case VIRTIO_NET_CTRL_MAC:
         switch (ctrl.cmd) {
@@ -392,10 +389,14 @@ static int vhost_vdpa_net_handle_ctrl_avail(VhostShadowVirtqueue *svq,
     VhostVDPAState *s = opaque;
     size_t in_len, dev_written;
     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
-    /* out and in buffers sent to the device */
-    struct iovec dev_buffers[2] = {
-        { .iov_base = s->cvq_cmd_out_buffer },
-        { .iov_base = s->cvq_cmd_in_buffer },
+    /* Out buffer sent to both the vdpa device and the device model */
+    struct iovec out = {
+        .iov_base = s->cvq_cmd_out_buffer,
+    };
+    /* In buffer sent to the device */
+    const struct iovec dev_in = {
+        .iov_base = s->cvq_cmd_in_buffer,
+        .iov_len = sizeof(virtio_net_ctrl_ack),
     };
     /* in buffer used for device model */
     const struct iovec in = {
@@ -405,17 +406,15 @@ static int vhost_vdpa_net_handle_ctrl_avail(VhostShadowVirtqueue *svq,
     int r = -EINVAL;
     bool ok;
 
-    ok = vhost_vdpa_net_cvq_map_elem(s, elem, dev_buffers);
-    if (unlikely(!ok)) {
-        goto out;
-    }
-
-    ok = vhost_vdpa_net_cvq_validate_cmd(&dev_buffers[0], 1);
+    out.iov_len = iov_to_buf(elem->out_sg, elem->out_num, 0,
+                             s->cvq_cmd_out_buffer,
+                             vhost_vdpa_net_cvq_cmd_len());
+    ok = vhost_vdpa_net_cvq_validate_cmd(s->cvq_cmd_out_buffer, out.iov_len);
     if (unlikely(!ok)) {
         goto out;
     }
 
-    r = vhost_svq_add(svq, &dev_buffers[0], 1, &dev_buffers[1], 1, elem);
+    r = vhost_svq_add(svq, &out, 1, &dev_in, 1, elem);
     if (unlikely(r != 0)) {
         if (unlikely(r == -ENOSPC)) {
             qemu_log_mask(LOG_GUEST_ERROR, "%s: No space on device queue\n",
@@ -435,13 +434,13 @@ static int vhost_vdpa_net_handle_ctrl_avail(VhostShadowVirtqueue *svq,
         goto out;
     }
 
-    memcpy(&status, dev_buffers[1].iov_base, sizeof(status));
+    memcpy(&status, s->cvq_cmd_in_buffer, sizeof(status));
     if (status != VIRTIO_NET_OK) {
         goto out;
     }
 
     status = VIRTIO_NET_ERR;
-    virtio_net_handle_ctrl_iov(svq->vdev, &in, 1, dev_buffers, 1);
+    virtio_net_handle_ctrl_iov(svq->vdev, &in, 1, &out, 1);
     if (status != VIRTIO_NET_OK) {
         error_report("Bad CVQ processing in model");
     }
@@ -454,12 +453,6 @@ out:
     }
     vhost_svq_push_elem(svq, elem, MIN(in_len, sizeof(status)));
     g_free(elem);
-    if (dev_buffers[0].iov_base) {
-        vhost_vdpa_cvq_unmap_buf(&s->vhost_vdpa, dev_buffers[0].iov_base);
-    }
-    if (dev_buffers[1].iov_base) {
-        vhost_vdpa_cvq_unmap_buf(&s->vhost_vdpa, dev_buffers[1].iov_base);
-    }
     return r;
 }
 
--

2.31.1

#13 lore raw

PATCH v8 9/12 vdpa: extract vhost_vdpa_net_cvq_add from vhost_vdpa_net_handle_ctrl_avail

In reply to Eugenio Pérez

So we can reuse it to inject state messages.

--
v7:
* Remove double free error

v6:
* Do not assume in buffer sent to the device is sizeof(virtio_net_ctrl_ack)

v5:
* Do not use an artificial !NULL VirtQueueElement
* Use only out size instead of iovec dev_buffers for these functions.

---
 net/vhost-vdpa.c | 59 +++++++++++++++++++++++++++++++-----------------
 1 file changed, 38 insertions(+), 21 deletions(-)
net/vhost-vdpa.c +38 -21
index 5254b58bdc..df85567cff 100644
@@ -331,6 +331,38 @@ static void vhost_vdpa_net_cvq_stop(NetClientState *nc)
     }
 }
 
+static ssize_t vhost_vdpa_net_cvq_add(VhostVDPAState *s, size_t out_len,
+                                      size_t in_len)
+{
+    /* Buffers for the device */
+    const struct iovec out = {
+        .iov_base = s->cvq_cmd_out_buffer,
+        .iov_len = out_len,
+    };
+    const struct iovec in = {
+        .iov_base = s->cvq_cmd_in_buffer,
+        .iov_len = sizeof(virtio_net_ctrl_ack),
+    };
+    VhostShadowVirtqueue *svq = g_ptr_array_index(s->vhost_vdpa.shadow_vqs, 0);
+    int r;
+
+    r = vhost_svq_add(svq, &out, 1, &in, 1, NULL);
+    if (unlikely(r != 0)) {
+        if (unlikely(r == -ENOSPC)) {
+            qemu_log_mask(LOG_GUEST_ERROR, "%s: No space on device queue\n",
+                          __func__);
+        }
+        return r;
+    }
+
+    /*
+     * We can poll here since we've had BQL from the time we sent the
+     * descriptor. Also, we need to take the answer before SVQ pulls by itself,
+     * when BQL is released
+     */
+    return vhost_svq_poll(svq);
+}
+
 static NetClientInfo net_vhost_vdpa_cvq_info = {
     .type = NET_CLIENT_DRIVER_VHOST_VDPA,
     .size = sizeof(VhostVDPAState),
@@ -387,23 +419,18 @@ static int vhost_vdpa_net_handle_ctrl_avail(VhostShadowVirtqueue *svq,
                                             void *opaque)
 {
     VhostVDPAState *s = opaque;
-    size_t in_len, dev_written;
+    size_t in_len;
     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
     /* Out buffer sent to both the vdpa device and the device model */
     struct iovec out = {
         .iov_base = s->cvq_cmd_out_buffer,
     };
-    /* In buffer sent to the device */
-    const struct iovec dev_in = {
-        .iov_base = s->cvq_cmd_in_buffer,
-        .iov_len = sizeof(virtio_net_ctrl_ack),
-    };
     /* in buffer used for device model */
     const struct iovec in = {
         .iov_base = &status,
         .iov_len = sizeof(status),
     };
-    int r = -EINVAL;
+    ssize_t dev_written = -EINVAL;
     bool ok;
 
     out.iov_len = iov_to_buf(elem->out_sg, elem->out_num, 0,
@@ -414,21 +441,11 @@ static int vhost_vdpa_net_handle_ctrl_avail(VhostShadowVirtqueue *svq,
         goto out;
     }
 
-    r = vhost_svq_add(svq, &out, 1, &dev_in, 1, elem);
-    if (unlikely(r != 0)) {
-        if (unlikely(r == -ENOSPC)) {
-            qemu_log_mask(LOG_GUEST_ERROR, "%s: No space on device queue\n",
-                          __func__);
-        }
+    dev_written = vhost_vdpa_net_cvq_add(s, out.iov_len, sizeof(status));
+    if (unlikely(dev_written < 0)) {
         goto out;
     }
 
-    /*
-     * We can poll here since we've had BQL from the time we sent the
-     * descriptor. Also, we need to take the answer before SVQ pulls by itself,
-     * when BQL is released
-     */
-    dev_written = vhost_svq_poll(svq);
     if (unlikely(dev_written < sizeof(status))) {
         error_report("Insufficient written data (%zu)", dev_written);
         goto out;
@@ -436,7 +453,7 @@ static int vhost_vdpa_net_handle_ctrl_avail(VhostShadowVirtqueue *svq,
 
     memcpy(&status, s->cvq_cmd_in_buffer, sizeof(status));
     if (status != VIRTIO_NET_OK) {
-        goto out;
+        return VIRTIO_NET_ERR;
     }
 
     status = VIRTIO_NET_ERR;
@@ -453,7 +470,7 @@ out:
     }
     vhost_svq_push_elem(svq, elem, MIN(in_len, sizeof(status)));
     g_free(elem);
-    return r;
+    return dev_written < 0 ? dev_written : 0;
 }
 
 static const VhostShadowVirtqueueOps vhost_vdpa_net_svq_ops = {
Signature · 1 line
2.31.1
#14 lore raw

PATCH v8 10/12 vhost_net: add NetClientState->load() callback

In reply to Eugenio Pérez

It allows per-net client operations right after device's successful start. In particular, to load the device status.

Vhost-vdpa net will use it to add the CVQ buffers to restore the device status.

---
v5: Rename start / load, naming it more specifically.

---
 include/net/net.h  | 2 ++
 hw/net/vhost_net.c | 7 +++++++
 2 files changed, 9 insertions(+)
include/net/net.h +2 -0
index 476ad45b9a..81d0b21def 100644
@@ -45,6 +45,7 @@ typedef struct NICConf {
 typedef void (NetPoll)(NetClientState *, bool enable);
 typedef bool (NetCanReceive)(NetClientState *);
 typedef int (NetStart)(NetClientState *);
+typedef int (NetLoad)(NetClientState *);
 typedef void (NetStop)(NetClientState *);
 typedef ssize_t (NetReceive)(NetClientState *, const uint8_t *, size_t);
 typedef ssize_t (NetReceiveIOV)(NetClientState *, const struct iovec *, int);
@@ -74,6 +75,7 @@ typedef struct NetClientInfo {
     NetReceiveIOV *receive_iov;
     NetCanReceive *can_receive;
     NetStart *start;
+    NetLoad *load;
     NetStop *stop;
     NetCleanup *cleanup;
     LinkStatusChanged *link_status_changed;
hw/net/vhost_net.c +7 -0
index 9d4b334453..d28f8b974b 100644
@@ -281,6 +281,13 @@ static int vhost_net_start_one(struct vhost_net *net,
             }
         }
     }
+
+    if (net->nc->info->load) {
+        r = net->nc->info->load(net->nc);
+        if (r < 0) {
+            goto fail;
+        }
+    }
     return 0;
 fail:
     file.fd = -1;
Signature · 1 line
2.31.1
Acked-by on v8 10/12
#15 lore raw

Re: [PATCH v8 10/12] vhost_net: add NetClientState->load() callback

Quoted 8 lines · Eugenio Pérez

It allows per-net client operations right after device's successful start. In particular, to load the device status.

Vhost-vdpa net will use it to add the CVQ buffers to restore the device status.

Quoted 48 lines

---
v5: Rename start / load, naming it more specifically.

---
 include/net/net.h  | 2 ++
 hw/net/vhost_net.c | 7 +++++++
 2 files changed, 9 insertions(+)
include/net/net.h +2 -0
index 476ad45b9a..81d0b21def 100644
@@ -45,6 +45,7 @@ typedef struct NICConf {
 typedef void (NetPoll)(NetClientState *, bool enable);
 typedef bool (NetCanReceive)(NetClientState *);
 typedef int (NetStart)(NetClientState *);
+typedef int (NetLoad)(NetClientState *);
 typedef void (NetStop)(NetClientState *);
 typedef ssize_t (NetReceive)(NetClientState *, const uint8_t *, size_t);
 typedef ssize_t (NetReceiveIOV)(NetClientState *, const struct iovec *, int);
@@ -74,6 +75,7 @@ typedef struct NetClientInfo {
     NetReceiveIOV *receive_iov;
     NetCanReceive *can_receive;
     NetStart *start;
+    NetLoad *load;
     NetStop *stop;
     NetCleanup *cleanup;
     LinkStatusChanged *link_status_changed;
hw/net/vhost_net.c +7 -1
index 9d4b334453..d28f8b974b 100644
@@ -281,6 +281,13 @@ static int vhost_net_start_one(struct vhost_net *net,
             }
         }
     }
+
+    if (net->nc->info->load) {
+        r = net->nc->info->load(net->nc);
+        if (r < 0) {
+            goto fail;
+        }
+    }
     return 0;
 fail:
     file.fd = -1;
--

2.31.1

#16 lore raw

PATCH v8 11/12 vdpa: Add virtio-net mac address via CVQ at start

In reply to Eugenio Pérez

This is needed so the destination vdpa device see the same state a the guest set in the source.

---
v8:
* Delete unneeded copy from device's in buffer.

v6:
* Map and unmap command buffers at the start and end of device usage.

v5:
* Rename s/start/load/
* Use independent NetClientInfo to only add load callback on cvq.

---
 net/vhost-vdpa.c | 41 +++++++++++++++++++++++++++++++++++++++++
 1 file changed, 41 insertions(+)
net/vhost-vdpa.c +41 -0
index df85567cff..e775957952 100644
@@ -363,11 +363,52 @@ static ssize_t vhost_vdpa_net_cvq_add(VhostVDPAState *s, size_t out_len,
     return vhost_svq_poll(svq);
 }
 
+static int vhost_vdpa_net_load(NetClientState *nc)
+{
+    VhostVDPAState *s = DO_UPCAST(VhostVDPAState, nc, nc);
+    struct vhost_vdpa *v = &s->vhost_vdpa;
+    VirtIONet *n;
+    uint64_t features;
+
+    assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA);
+
+    if (!v->shadow_vqs_enabled) {
+        return 0;
+    }
+
+    n = VIRTIO_NET(v->dev->vdev);
+    features = v->dev->vdev->host_features;
+    if (features & BIT_ULL(VIRTIO_NET_F_CTRL_MAC_ADDR)) {
+        const struct virtio_net_ctrl_hdr ctrl = {
+            .class = VIRTIO_NET_CTRL_MAC,
+            .cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET,
+        };
+        char *cursor = s->cvq_cmd_out_buffer;
+        ssize_t dev_written;
+
+        memcpy(cursor, &ctrl, sizeof(ctrl));
+        cursor += sizeof(ctrl);
+        memcpy(cursor, n->mac, sizeof(n->mac));
+        cursor += sizeof(n->mac);
+
+        dev_written = vhost_vdpa_net_cvq_add(s, sizeof(ctrl) + sizeof(n->mac),
+                                             sizeof(virtio_net_ctrl_ack));
+        if (unlikely(dev_written < 0)) {
+            return dev_written;
+        }
+
+        return *((virtio_net_ctrl_ack *)s->cvq_cmd_in_buffer) != VIRTIO_NET_OK;
+    }
+
+    return 0;
+}
+
 static NetClientInfo net_vhost_vdpa_cvq_info = {
     .type = NET_CLIENT_DRIVER_VHOST_VDPA,
     .size = sizeof(VhostVDPAState),
     .receive = vhost_vdpa_receive,
     .start = vhost_vdpa_net_cvq_start,
+    .load = vhost_vdpa_net_load,
     .stop = vhost_vdpa_net_cvq_stop,
     .cleanup = vhost_vdpa_cleanup,
     .has_vnet_hdr = vhost_vdpa_has_vnet_hdr,
Signature · 1 line
2.31.1
Acked-by on v8 11/12
#17 lore raw

Re: [PATCH v8 11/12] vdpa: Add virtio-net mac address via CVQ at start

Quoted 44 lines · Eugenio Pérez

This is needed so the destination vdpa device see the same state a the guest set in the source.

---
v8:
* Delete unneeded copy from device's in buffer.

v6:
* Map and unmap command buffers at the start and end of device usage.

v5:
* Rename s/start/load/
* Use independent NetClientInfo to only add load callback on cvq.

---
 net/vhost-vdpa.c | 41 +++++++++++++++++++++++++++++++++++++++++
 1 file changed, 41 insertions(+)
net/vhost-vdpa.c +17 -0
index df85567cff..e775957952 100644
@@ -363,11 +363,52 @@ static ssize_t vhost_vdpa_net_cvq_add(VhostVDPAState *s, size_t out_len,
     return vhost_svq_poll(svq);
 }
 
+static int vhost_vdpa_net_load(NetClientState *nc)
+{
+    VhostVDPAState *s = DO_UPCAST(VhostVDPAState, nc, nc);
+    struct vhost_vdpa *v = &s->vhost_vdpa;
+    VirtIONet *n;
+    uint64_t features;
+
+    assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA);
+
+    if (!v->shadow_vqs_enabled) {
+        return 0;
+    }
+
+    n = VIRTIO_NET(v->dev->vdev);
+    features = v->dev->vdev->host_features;
+    if (features & BIT_ULL(VIRTIO_NET_F_CTRL_MAC_ADDR)) {
+        const struct virtio_net_ctrl_hdr ctrl = {

net/vhost-vdpa.c static ssize_t vhost_vdpa_net_cvq_add(VhostVDPAState *s, size_t out_len, · line 392

Quoted 10 lines
net/vhost-vdpa.c +10 -0
+            .class = VIRTIO_NET_CTRL_MAC,
+            .cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET,
+        };
+        char *cursor = s->cvq_cmd_out_buffer;
+        ssize_t dev_written;
+
+        memcpy(cursor, &ctrl, sizeof(ctrl));
+        cursor += sizeof(ctrl);
+        memcpy(cursor, n->mac, sizeof(n->mac));
+        cursor += sizeof(n->mac);

I'd cast and initialize directly from the cmd_out_buff. But this could be done on top.

So

Quoted 25 lines
net/vhost-vdpa.c +14 -1
+
+        dev_written = vhost_vdpa_net_cvq_add(s, sizeof(ctrl) + sizeof(n->mac),
+                                             sizeof(virtio_net_ctrl_ack));
+        if (unlikely(dev_written < 0)) {
+            return dev_written;
+        }
+
+        return *((virtio_net_ctrl_ack *)s->cvq_cmd_in_buffer) != VIRTIO_NET_OK;
+    }
+
+    return 0;
+}
+
 static NetClientInfo net_vhost_vdpa_cvq_info = {
     .type = NET_CLIENT_DRIVER_VHOST_VDPA,
     .size = sizeof(VhostVDPAState),
     .receive = vhost_vdpa_receive,
     .start = vhost_vdpa_net_cvq_start,
+    .load = vhost_vdpa_net_load,
     .stop = vhost_vdpa_net_cvq_stop,
     .cleanup = vhost_vdpa_cleanup,
     .has_vnet_hdr = vhost_vdpa_has_vnet_hdr,
--

2.31.1

#18 lore raw

Re: [PATCH v8 11/12] vdpa: Add virtio-net mac address via CVQ at start

Quoted 51 lines · Jason Wang
Quoted 49 lines · Eugenio Pérez

This is needed so the destination vdpa device see the same state a the guest set in the source.

---
v8:
* Delete unneeded copy from device's in buffer.

v6:
* Map and unmap command buffers at the start and end of device usage.

v5:
* Rename s/start/load/
* Use independent NetClientInfo to only add load callback on cvq.

---
 net/vhost-vdpa.c | 41 +++++++++++++++++++++++++++++++++++++++++
 1 file changed, 41 insertions(+)
net/vhost-vdpa.c +22 -0
index df85567cff..e775957952 100644
@@ -363,11 +363,52 @@ static ssize_t vhost_vdpa_net_cvq_add(VhostVDPAState *s, size_t out_len,
     return vhost_svq_poll(svq);
 }
 
+static int vhost_vdpa_net_load(NetClientState *nc)
+{
+    VhostVDPAState *s = DO_UPCAST(VhostVDPAState, nc, nc);
+    struct vhost_vdpa *v = &s->vhost_vdpa;
+    VirtIONet *n;
+    uint64_t features;
+
+    assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA);
+
+    if (!v->shadow_vqs_enabled) {
+        return 0;
+    }
+
+    n = VIRTIO_NET(v->dev->vdev);
+    features = v->dev->vdev->host_features;
+    if (features & BIT_ULL(VIRTIO_NET_F_CTRL_MAC_ADDR)) {
+        const struct virtio_net_ctrl_hdr ctrl = {
+            .class = VIRTIO_NET_CTRL_MAC,
+            .cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET,
+        };
+        char *cursor = s->cvq_cmd_out_buffer;
+        ssize_t dev_written;
Quoted 9 lines

net/vhost-vdpa.c static ssize_t vhost_vdpa_net_cvq_add(VhostVDPAState *s, size_t out_len, · line 392

Quoted 5 lines
net/vhost-vdpa.c +5 -0
+
+        memcpy(cursor, &ctrl, sizeof(ctrl));
+        cursor += sizeof(ctrl);
+        memcpy(cursor, n->mac, sizeof(n->mac));
+        cursor += sizeof(n->mac);

I'd cast and initialize directly from the cmd_out_buff. But this could be done on top.

Actually you are right, that's valid for MAC. I think the compiler should be able to omit the memcpy call if the direct write is safe, but I haven't checked.

In a second review, either this last increment of the "cursor" variable is useless and we can delete it or we could use it to calculate the output size (cursor - s->cvq_cmd_out_buffer).

Quoted 31 lines

So

Quoted 25 lines
net/vhost-vdpa.c +14 -1
+
+        dev_written = vhost_vdpa_net_cvq_add(s, sizeof(ctrl) + sizeof(n->mac),
+                                             sizeof(virtio_net_ctrl_ack));
+        if (unlikely(dev_written < 0)) {
+            return dev_written;
+        }
+
+        return *((virtio_net_ctrl_ack *)s->cvq_cmd_in_buffer) != VIRTIO_NET_OK;
+    }
+
+    return 0;
+}
+
 static NetClientInfo net_vhost_vdpa_cvq_info = {
     .type = NET_CLIENT_DRIVER_VHOST_VDPA,
     .size = sizeof(VhostVDPAState),
     .receive = vhost_vdpa_receive,
     .start = vhost_vdpa_net_cvq_start,
+    .load = vhost_vdpa_net_load,
     .stop = vhost_vdpa_net_cvq_stop,
     .cleanup = vhost_vdpa_cleanup,
     .has_vnet_hdr = vhost_vdpa_has_vnet_hdr,
--

2.31.1

#19 lore raw

PATCH v8 12/12 vdpa: Delete CVQ migration blocker

In reply to Eugenio Pérez

We can restore the device state in the destination via CVQ now. Remove the migration blocker.

---
 include/hw/virtio/vhost-vdpa.h |  1 -
 hw/virtio/vhost-vdpa.c         | 14 --------------
 net/vhost-vdpa.c               |  2 --
 3 files changed, 17 deletions(-)
include/hw/virtio/vhost-vdpa.h +0 -1
index d10a89303e..1111d85643 100644
@@ -35,7 +35,6 @@ typedef struct vhost_vdpa {
     bool shadow_vqs_enabled;
     /* IOVA mapping used by the Shadow Virtqueue */
     VhostIOVATree *iova_tree;
-    Error *migration_blocker;
     GPtrArray *shadow_vqs;
     const VhostShadowVirtqueueOps *shadow_vq_ops;
     void *shadow_vq_ops_opaque;
hw/virtio/vhost-vdpa.c +0 -14
index 7e28d2f674..4b0cfc0f56 100644
@@ -1033,13 +1033,6 @@ static bool vhost_vdpa_svqs_start(struct vhost_dev *dev)
         return true;
     }
 
-    if (v->migration_blocker) {
-        int r = migrate_add_blocker(v->migration_blocker, &err);
-        if (unlikely(r < 0)) {
-            return false;
-        }
-    }
-
     for (i = 0; i < v->shadow_vqs->len; ++i) {
         VirtQueue *vq = virtio_get_queue(dev->vdev, dev->vq_index + i);
         VhostShadowVirtqueue *svq = g_ptr_array_index(v->shadow_vqs, i);
@@ -1082,10 +1075,6 @@ err:
         vhost_svq_stop(svq);
     }
 
-    if (v->migration_blocker) {
-        migrate_del_blocker(v->migration_blocker);
-    }
-
     return false;
 }
 
@@ -1105,9 +1094,6 @@ static bool vhost_vdpa_svqs_stop(struct vhost_dev *dev)
         }
     }
 
-    if (v->migration_blocker) {
-        migrate_del_blocker(v->migration_blocker);
-    }
     return true;
 }
 
net/vhost-vdpa.c +0 -2
index e775957952..188adb00b7 100644
@@ -556,8 +556,6 @@ static NetClientState *net_vhost_vdpa_init(NetClientState *peer,
 
         s->vhost_vdpa.shadow_vq_ops = &vhost_vdpa_net_svq_ops;
         s->vhost_vdpa.shadow_vq_ops_opaque = s;
-        error_setg(&s->vhost_vdpa.migration_blocker,
-                   "Migration disabled: vhost-vdpa uses CVQ.");
     }
     ret = vhost_vdpa_add(nc, (void *)&s->vhost_vdpa, queue_pair_index, nvqs);
     if (ret) {
Signature · 1 line
2.31.1
#20 lore raw
In reply to Eugenio Pérez
Quoted 80 lines · Eugenio Pérez

CVQ of net vhost-vdpa devices can be intercepted since the addition of x-svq. The virtio-net device model is updated. The migration was blocked because although the state can be megrated between VMM it was not possible to restore on the destination NIC.

This series add support for SVQ to inject external messages without the guest's knowledge, so before the guest is resumed all the guest visible state is restored. It is done using standard CVQ messages, so the vhost-vdpa device does not need to learn how to restore it: As long as they have the feature, they know how to handle it.

This series needs fix [1] to be applied to achieve full live migration.

Thanks!

[1] https://lists.nongnu.org/archive/html/qemu-devel/2022-08/msg00325.html

v8:
- Rename NetClientInfo load to start, so is symmetrical with stop()
- Delete copy of device's in buffer at vhost_vdpa_net_load

v7:
- Remove accidental double free.

v6:
- Move map and unmap of the buffers to the start and stop of the device. This implies more callbacks on NetClientInfo, but simplifies the SVQ CVQ code.
- Not assume that in buffer is sizeof(virtio_net_ctrl_ack) in vhost_vdpa_net_cvq_add
- Reduce the number of changes from previous versions
- Delete unused memory barrier

v5:
- Rename s/start/load/
- Use independent NetClientInfo to only add load callback on cvq.
- Accept out sg instead of dev_buffers[] at vhost_vdpa_net_cvq_map_elem
- Use only out size instead of iovec dev_buffers to know if the descriptor is effectively available, allowing to delete artificial !NULL VirtQueueElement on vhost_svq_add call.

v4:
- Actually use NetClientInfo callback.

v3:
- Route vhost-vdpa start code through NetClientInfo callback.
- Delete extra vhost_net_stop_one() call.

v2:
- Fix SIGSEGV dereferencing SVQ when not in svq mode

v1 from RFC:
- Do not reorder DRIVER_OK & enable patches.
- Delete leftovers

Eugenio Pérez (12):
vhost: stop transfer elem ownership in vhost_handle_guest_kick
vhost: use SVQ element ndescs instead of opaque data for desc
validation
vhost: Delete useless read memory barrier
vhost: Do not depend on !NULL VirtQueueElement on vhost_svq_flush
vhost_net: Add NetClientInfo prepare callback
vhost_net: Add NetClientInfo stop callback
vdpa: add net_vhost_vdpa_cvq_info NetClientInfo
vdpa: Move command buffers map to start of net device
vdpa: extract vhost_vdpa_net_cvq_add from
vhost_vdpa_net_handle_ctrl_avail
vhost_net: add NetClientState->load() callback
vdpa: Add virtio-net mac address via CVQ at start
vdpa: Delete CVQ migration blocker

 include/hw/virtio/vhost-vdpa.h     |   1 -
 include/net/net.h                  |   6 +
 hw/net/vhost_net.c                 |  17 +++
 hw/virtio/vhost-shadow-virtqueue.c |  27 ++--
 hw/virtio/vhost-vdpa.c             |  14 --
 net/vhost-vdpa.c                   | 225 ++++++++++++++++++-----------
 6 files changed, 178 insertions(+), 112 deletions(-)
Quoted 5 lines

--
2.31.1

Hi Jason,

Should I send a new version of this series with the changes you proposed, or can they be done at pull time? (Mostly changes in patch messages). Can you confirm to me that there is no other action I need to perform?

Thanks!

#21 lore raw
Quoted 88 lines · Eugenio Perez Martin
Quoted 80 lines · Eugenio Pérez

CVQ of net vhost-vdpa devices can be intercepted since the addition of x-svq. The virtio-net device model is updated. The migration was blocked because although the state can be megrated between VMM it was not possible to restore on the destination NIC.

This series add support for SVQ to inject external messages without the guest's knowledge, so before the guest is resumed all the guest visible state is restored. It is done using standard CVQ messages, so the vhost-vdpa device does not need to learn how to restore it: As long as they have the feature, they know how to handle it.

This series needs fix [1] to be applied to achieve full live migration.

Thanks!

[1] https://lists.nongnu.org/archive/html/qemu-devel/2022-08/msg00325.html

v8:
- Rename NetClientInfo load to start, so is symmetrical with stop()
- Delete copy of device's in buffer at vhost_vdpa_net_load

v7:
- Remove accidental double free.

v6:
- Move map and unmap of the buffers to the start and stop of the device. This implies more callbacks on NetClientInfo, but simplifies the SVQ CVQ code.
- Not assume that in buffer is sizeof(virtio_net_ctrl_ack) in vhost_vdpa_net_cvq_add
- Reduce the number of changes from previous versions
- Delete unused memory barrier

v5:
- Rename s/start/load/
- Use independent NetClientInfo to only add load callback on cvq.
- Accept out sg instead of dev_buffers[] at vhost_vdpa_net_cvq_map_elem
- Use only out size instead of iovec dev_buffers to know if the descriptor is effectively available, allowing to delete artificial !NULL VirtQueueElement on vhost_svq_add call.

v4:
- Actually use NetClientInfo callback.

v3:
- Route vhost-vdpa start code through NetClientInfo callback.
- Delete extra vhost_net_stop_one() call.

v2:
- Fix SIGSEGV dereferencing SVQ when not in svq mode

v1 from RFC:
- Do not reorder DRIVER_OK & enable patches.
- Delete leftovers

Eugenio Pérez (12):
vhost: stop transfer elem ownership in vhost_handle_guest_kick
vhost: use SVQ element ndescs instead of opaque data for desc
validation
vhost: Delete useless read memory barrier
vhost: Do not depend on !NULL VirtQueueElement on vhost_svq_flush
vhost_net: Add NetClientInfo prepare callback
vhost_net: Add NetClientInfo stop callback
vdpa: add net_vhost_vdpa_cvq_info NetClientInfo
vdpa: Move command buffers map to start of net device
vdpa: extract vhost_vdpa_net_cvq_add from
vhost_vdpa_net_handle_ctrl_avail
vhost_net: add NetClientState->load() callback
vdpa: Add virtio-net mac address via CVQ at start
vdpa: Delete CVQ migration blocker

 include/hw/virtio/vhost-vdpa.h     |   1 -
 include/net/net.h                  |   6 +
 hw/net/vhost_net.c                 |  17 +++
 hw/virtio/vhost-shadow-virtqueue.c |  27 ++--
 hw/virtio/vhost-vdpa.c             |  14 --
 net/vhost-vdpa.c                   | 225 ++++++++++++++++++-----------
 6 files changed, 178 insertions(+), 112 deletions(-)
Quoted 5 lines

--
2.31.1

Quoted 5 lines

Hi Jason,

Should I send a new version of this series with the changes you proposed, or can they be done at pull time? (Mostly changes in patch messages).

A new series please.

Quoted 2 lines

Can you confirm to me that there is no other action I need to perform?

No other from my side.

Thanks

Quoted 3 lines

Thanks!