Skip to contentSkip to conversation

vhost-user: Check for iotlb callback in iotlb_miss

PATCH In review Eugenio Pérez 2 participants 7 messages Jan 27, 2021 – Jan 29, 2021

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 vhost-user: Check for iotlb callback in iotlb_miss

Not registering this can lead to vhost_backend_handle_iotlb_msg and vhost_device_iotlb_miss if backend issue a miss after qemu vhost device stop.

This causes a try to access dev->vdev->dma_as with vdev == NULL.

Reproduced rebooting a guest with testpmd in txonly forward mode.
#0 0x0000559ffff94394 in vhost_device_iotlb_miss (

     dev=dev@entry=0x55a0012f6680, iova=10245279744, write=1)
     at ../hw/virtio/vhost.c:1013
 #1  0x0000559ffff9ac31 in vhost_backend_handle_iotlb_msg (
     imsg=0x7ffddcfd32c0, dev=0x55a0012f6680)
     at ../hw/virtio/vhost-backend.c:411
 #2  vhost_backend_handle_iotlb_msg (dev=dev@entry=0x55a0012f6680,
     imsg=imsg@entry=0x7ffddcfd32c0)
     at ../hw/virtio/vhost-backend.c:404
 #3  0x0000559fffeded7b in slave_read (opaque=0x55a0012f6680)
     at ../hw/virtio/vhost-user.c:1464
 #4  0x000055a0000c541b in aio_dispatch_handler (
     ctx=ctx@entry=0x55a0010a2120, node=0x55a0012d9e00)
     at ../util/aio-posix.c:329
Fixes 6dcdd06e3b ("spec/vhost-user spec: Add IOMMU support")
---
 hw/virtio/vhost-user.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)
hw/virtio/vhost-user.c +8 -2
index 2fdd5daf74..a49b2229fb 100644
@@ -238,6 +238,7 @@ struct vhost_user {
     /* Shared between vhost devs of the same virtio device */
     VhostUserState *user;
     int slave_fd;
+    bool iotlb_enabled;
     NotifierWithReturn postcopy_notifier;
     struct PostCopyFD  postcopy_fd;
     uint64_t           postcopy_client_bases[VHOST_USER_MAX_RAM_SLOTS];
@@ -1461,7 +1462,11 @@ static void slave_read(void *opaque)
 
     switch (hdr.request) {
     case VHOST_USER_SLAVE_IOTLB_MSG:
-        ret = vhost_backend_handle_iotlb_msg(dev, &payload.iotlb);
+        if (likely(u->iotlb_enabled)) {
+            ret = vhost_backend_handle_iotlb_msg(dev, &payload.iotlb);
+        } else {
+            ret = -EFAULT;
+        }
         break;
     case VHOST_USER_SLAVE_CONFIG_CHANGE_MSG :
         ret = vhost_user_slave_handle_config_change(dev);
@@ -2044,7 +2049,8 @@ static int vhost_user_send_device_iotlb_msg(struct vhost_dev *dev,
 
 static void vhost_user_set_iotlb_callback(struct vhost_dev *dev, int enabled)
 {
-    /* No-op as the receive channel is not dedicated to IOTLB messages. */
+    struct vhost_user *u = dev->opaque;
+    u->iotlb_enabled = enabled;
 }
 
 static int vhost_user_get_config(struct vhost_dev *dev, uint8_t *config,
Signature · 1 line
2.27.0
#1 lore raw
Quoted 5 lines · Eugenio Pérez

Not registering this can lead to vhost_backend_handle_iotlb_msg and vhost_device_iotlb_miss if backend issue a miss after qemu vhost device stop.

This causes a try to access dev->vdev->dma_as with vdev == NULL.

Hi Eugenio:

What condition can we get this? Did you mean we receive IOTLB_MISS before vhost_dev_start()?

If yes, it looks to me a bug somewhere else.

Thanks

Quoted 58 lines

Reproduced rebooting a guest with testpmd in txonly forward mode.
#0 0x0000559ffff94394 in vhost_device_iotlb_miss (

      dev=dev@entry=0x55a0012f6680, iova=10245279744, write=1)
      at ../hw/virtio/vhost.c:1013
  #1  0x0000559ffff9ac31 in vhost_backend_handle_iotlb_msg (
      imsg=0x7ffddcfd32c0, dev=0x55a0012f6680)
      at ../hw/virtio/vhost-backend.c:411
  #2  vhost_backend_handle_iotlb_msg (dev=dev@entry=0x55a0012f6680,
      imsg=imsg@entry=0x7ffddcfd32c0)
      at ../hw/virtio/vhost-backend.c:404
  #3  0x0000559fffeded7b in slave_read (opaque=0x55a0012f6680)
      at ../hw/virtio/vhost-user.c:1464
  #4  0x000055a0000c541b in aio_dispatch_handler (
      ctx=ctx@entry=0x55a0010a2120, node=0x55a0012d9e00)
      at ../util/aio-posix.c:329
Fixes 6dcdd06e3b ("spec/vhost-user spec: Add IOMMU support")
---
  hw/virtio/vhost-user.c | 10 ++++++++--
  1 file changed, 8 insertions(+), 2 deletions(-)
hw/virtio/vhost-user.c +8 -2
index 2fdd5daf74..a49b2229fb 100644
@@ -238,6 +238,7 @@ struct vhost_user {
      /* Shared between vhost devs of the same virtio device */
      VhostUserState *user;
      int slave_fd;
+    bool iotlb_enabled;
      NotifierWithReturn postcopy_notifier;
      struct PostCopyFD  postcopy_fd;
      uint64_t           postcopy_client_bases[VHOST_USER_MAX_RAM_SLOTS];
@@ -1461,7 +1462,11 @@ static void slave_read(void *opaque)
 
      switch (hdr.request) {
      case VHOST_USER_SLAVE_IOTLB_MSG:
-        ret = vhost_backend_handle_iotlb_msg(dev, &payload.iotlb);
+        if (likely(u->iotlb_enabled)) {
+            ret = vhost_backend_handle_iotlb_msg(dev, &payload.iotlb);
+        } else {
+            ret = -EFAULT;
+        }
          break;
      case VHOST_USER_SLAVE_CONFIG_CHANGE_MSG :
          ret = vhost_user_slave_handle_config_change(dev);
@@ -2044,7 +2049,8 @@ static int vhost_user_send_device_iotlb_msg(struct vhost_dev *dev,
 
  static void vhost_user_set_iotlb_callback(struct vhost_dev *dev, int enabled)
  {
-    /* No-op as the receive channel is not dedicated to IOTLB messages. */
+    struct vhost_user *u = dev->opaque;
+    u->iotlb_enabled = enabled;
  }
 
  static int vhost_user_get_config(struct vhost_dev *dev, uint8_t *config,
#2 lore raw

Hi Jason.

Quoted 7 lines · Jason Wang
Quoted 4 lines · Eugenio Pérez

Not registering this can lead to vhost_backend_handle_iotlb_msg and vhost_device_iotlb_miss if backend issue a miss after qemu vhost device stop.

Quoted 8 lines
Quoted 1 line

This causes a try to access dev->vdev->dma_as with vdev == NULL.

Hi Eugenio:

What condition can we get this? Did you mean we receive IOTLB_MISS before vhost_dev_start()?

In the VM reboot case, yes, between vhost_dev_stop() and vhost_dev_start(). But I can reproduce the bug in VM shutdown too, with no vhost_dev_start expected.

I didn't try to send IOTLB_MISS before starting vhost_dev, but this patch should solve that problem too.

I think we can get this with whatever malicious/buggy vhost-user backend sending unexpected iotlb misses, but I didn't test so deeply.

Quoted 1 line

If yes, it looks to me a bug somewhere else.

Where should I look for it?

Thanks!

Quoted 63 lines

Thanks

Quoted 58 lines

Reproduced rebooting a guest with testpmd in txonly forward mode.
#0 0x0000559ffff94394 in vhost_device_iotlb_miss (

      dev=dev@entry=0x55a0012f6680, iova=10245279744, write=1)
      at ../hw/virtio/vhost.c:1013
  #1  0x0000559ffff9ac31 in vhost_backend_handle_iotlb_msg (
      imsg=0x7ffddcfd32c0, dev=0x55a0012f6680)
      at ../hw/virtio/vhost-backend.c:411
  #2  vhost_backend_handle_iotlb_msg (dev=dev@entry=0x55a0012f6680,
      imsg=imsg@entry=0x7ffddcfd32c0)
      at ../hw/virtio/vhost-backend.c:404
  #3  0x0000559fffeded7b in slave_read (opaque=0x55a0012f6680)
      at ../hw/virtio/vhost-user.c:1464
  #4  0x000055a0000c541b in aio_dispatch_handler (
      ctx=ctx@entry=0x55a0010a2120, node=0x55a0012d9e00)
      at ../util/aio-posix.c:329
Fixes 6dcdd06e3b ("spec/vhost-user spec: Add IOMMU support")
---
  hw/virtio/vhost-user.c | 10 ++++++++--
  1 file changed, 8 insertions(+), 2 deletions(-)
hw/virtio/vhost-user.c +8 -2
index 2fdd5daf74..a49b2229fb 100644
@@ -238,6 +238,7 @@ struct vhost_user {
      /* Shared between vhost devs of the same virtio device */
      VhostUserState *user;
      int slave_fd;
+    bool iotlb_enabled;
      NotifierWithReturn postcopy_notifier;
      struct PostCopyFD  postcopy_fd;
      uint64_t           postcopy_client_bases[VHOST_USER_MAX_RAM_SLOTS];
@@ -1461,7 +1462,11 @@ static void slave_read(void *opaque)
 
      switch (hdr.request) {
      case VHOST_USER_SLAVE_IOTLB_MSG:
-        ret = vhost_backend_handle_iotlb_msg(dev, &payload.iotlb);
+        if (likely(u->iotlb_enabled)) {
+            ret = vhost_backend_handle_iotlb_msg(dev, &payload.iotlb);
+        } else {
+            ret = -EFAULT;
+        }
          break;
      case VHOST_USER_SLAVE_CONFIG_CHANGE_MSG :
          ret = vhost_user_slave_handle_config_change(dev);
@@ -2044,7 +2049,8 @@ static int vhost_user_send_device_iotlb_msg(struct vhost_dev *dev,
 
  static void vhost_user_set_iotlb_callback(struct vhost_dev *dev, int enabled)
  {
-    /* No-op as the receive channel is not dedicated to IOTLB messages. */
+    struct vhost_user *u = dev->opaque;
+    u->iotlb_enabled = enabled;
  }
 
  static int vhost_user_get_config(struct vhost_dev *dev, uint8_t *config,
#3 lore raw
Quoted 20 lines · Eugenio Perez Martin

Hi Jason.

Quoted 6 lines · Jason Wang
Quoted 4 lines · Eugenio Pérez

Not registering this can lead to vhost_backend_handle_iotlb_msg and vhost_device_iotlb_miss if backend issue a miss after qemu vhost device stop.

Quoted 7 lines
Quoted 1 line

This causes a try to access dev->vdev->dma_as with vdev == NULL.

Hi Eugenio:

What condition can we get this? Did you mean we receive IOTLB_MISS before vhost_dev_start()?

In the VM reboot case, yes, between vhost_dev_stop() and vhost_dev_start(). But I can reproduce the bug in VM shutdown too, with no vhost_dev_start expected.

Quoted 5 lines

I didn't try to send IOTLB_MISS before starting vhost_dev, but this patch should solve that problem too.

I think we can get this with whatever malicious/buggy vhost-user backend sending unexpected iotlb misses, but I didn't test so deeply.

I see.

Quoted 3 lines
Quoted 2 lines

If yes, it looks to me a bug somewhere else.

Where should I look for it?

So I winder whether or not we can simply ignore IOTLB message if vhost device is not started.

Thanks

Quoted 63 lines

Thanks!

Quoted 60 lines

Thanks

Quoted 57 lines

Reproduced rebooting a guest with testpmd in txonly forward mode.
#0 0x0000559ffff94394 in vhost_device_iotlb_miss (

       dev=dev@entry=0x55a0012f6680, iova=10245279744, write=1)
       at ../hw/virtio/vhost.c:1013
   #1  0x0000559ffff9ac31 in vhost_backend_handle_iotlb_msg (
       imsg=0x7ffddcfd32c0, dev=0x55a0012f6680)
       at ../hw/virtio/vhost-backend.c:411
   #2  vhost_backend_handle_iotlb_msg (dev=dev@entry=0x55a0012f6680,
       imsg=imsg@entry=0x7ffddcfd32c0)
       at ../hw/virtio/vhost-backend.c:404
   #3  0x0000559fffeded7b in slave_read (opaque=0x55a0012f6680)
       at ../hw/virtio/vhost-user.c:1464
   #4  0x000055a0000c541b in aio_dispatch_handler (
       ctx=ctx@entry=0x55a0010a2120, node=0x55a0012d9e00)
       at ../util/aio-posix.c:329
Fixes 6dcdd06e3b ("spec/vhost-user spec: Add IOMMU support")
---
   hw/virtio/vhost-user.c | 10 ++++++++--
   1 file changed, 8 insertions(+), 2 deletions(-)
hw/virtio/vhost-user.c +8 -2
index 2fdd5daf74..a49b2229fb 100644
@@ -238,6 +238,7 @@ struct vhost_user {
       /* Shared between vhost devs of the same virtio device */
       VhostUserState *user;
       int slave_fd;
+    bool iotlb_enabled;
       NotifierWithReturn postcopy_notifier;
       struct PostCopyFD  postcopy_fd;
       uint64_t           postcopy_client_bases[VHOST_USER_MAX_RAM_SLOTS];
@@ -1461,7 +1462,11 @@ static void slave_read(void *opaque)
 
       switch (hdr.request) {
       case VHOST_USER_SLAVE_IOTLB_MSG:
-        ret = vhost_backend_handle_iotlb_msg(dev, &payload.iotlb);
+        if (likely(u->iotlb_enabled)) {
+            ret = vhost_backend_handle_iotlb_msg(dev, &payload.iotlb);
+        } else {
+            ret = -EFAULT;
+        }
           break;
       case VHOST_USER_SLAVE_CONFIG_CHANGE_MSG :
           ret = vhost_user_slave_handle_config_change(dev);
@@ -2044,7 +2049,8 @@ static int vhost_user_send_device_iotlb_msg(struct vhost_dev *dev,
 
   static void vhost_user_set_iotlb_callback(struct vhost_dev *dev, int enabled)
   {
-    /* No-op as the receive channel is not dedicated to IOTLB messages. */
+    struct vhost_user *u = dev->opaque;
+    u->iotlb_enabled = enabled;
   }
 
   static int vhost_user_get_config(struct vhost_dev *dev, uint8_t *config,
#4 lore raw
Quoted 34 lines · Jason Wang
Quoted 20 lines · Eugenio Perez Martin

Hi Jason.

Quoted 6 lines · Jason Wang
Quoted 4 lines · Eugenio Pérez

Not registering this can lead to vhost_backend_handle_iotlb_msg and vhost_device_iotlb_miss if backend issue a miss after qemu vhost device stop.

Quoted 7 lines
Quoted 1 line

This causes a try to access dev->vdev->dma_as with vdev == NULL.

Hi Eugenio:

What condition can we get this? Did you mean we receive IOTLB_MISS before vhost_dev_start()?

In the VM reboot case, yes, between vhost_dev_stop() and vhost_dev_start(). But I can reproduce the bug in VM shutdown too, with no vhost_dev_start expected.

Quoted 5 lines

I didn't try to send IOTLB_MISS before starting vhost_dev, but this patch should solve that problem too.

I think we can get this with whatever malicious/buggy vhost-user backend sending unexpected iotlb misses, but I didn't test so deeply.

I see.

Quoted 1 line
Quoted 7 lines
Quoted 2 lines
Quoted 1 line

If yes, it looks to me a bug somewhere else.

Where should I look for it?

So I winder whether or not we can simply ignore IOTLB message if vhost device is not started.

Do you mean like an early return in vhost_device_iotlb_miss? That was my first first option, but this seems cleaner to me. I'm ok with both options.

Or do you mean not to return -EFAULT but 0 if !u->iotlb_enabled ?

Thanks!

Quoted 67 lines

Thanks

Quoted 63 lines

Thanks!

Quoted 60 lines

Thanks

Quoted 57 lines

Reproduced rebooting a guest with testpmd in txonly forward mode.
#0 0x0000559ffff94394 in vhost_device_iotlb_miss (

       dev=dev@entry=0x55a0012f6680, iova=10245279744, write=1)
       at ../hw/virtio/vhost.c:1013
   #1  0x0000559ffff9ac31 in vhost_backend_handle_iotlb_msg (
       imsg=0x7ffddcfd32c0, dev=0x55a0012f6680)
       at ../hw/virtio/vhost-backend.c:411
   #2  vhost_backend_handle_iotlb_msg (dev=dev@entry=0x55a0012f6680,
       imsg=imsg@entry=0x7ffddcfd32c0)
       at ../hw/virtio/vhost-backend.c:404
   #3  0x0000559fffeded7b in slave_read (opaque=0x55a0012f6680)
       at ../hw/virtio/vhost-user.c:1464
   #4  0x000055a0000c541b in aio_dispatch_handler (
       ctx=ctx@entry=0x55a0010a2120, node=0x55a0012d9e00)
       at ../util/aio-posix.c:329
Fixes 6dcdd06e3b ("spec/vhost-user spec: Add IOMMU support")
---
   hw/virtio/vhost-user.c | 10 ++++++++--
   1 file changed, 8 insertions(+), 2 deletions(-)
hw/virtio/vhost-user.c +8 -2
index 2fdd5daf74..a49b2229fb 100644
@@ -238,6 +238,7 @@ struct vhost_user {
       /* Shared between vhost devs of the same virtio device */
       VhostUserState *user;
       int slave_fd;
+    bool iotlb_enabled;
       NotifierWithReturn postcopy_notifier;
       struct PostCopyFD  postcopy_fd;
       uint64_t           postcopy_client_bases[VHOST_USER_MAX_RAM_SLOTS];
@@ -1461,7 +1462,11 @@ static void slave_read(void *opaque)
 
       switch (hdr.request) {
       case VHOST_USER_SLAVE_IOTLB_MSG:
-        ret = vhost_backend_handle_iotlb_msg(dev, &payload.iotlb);
+        if (likely(u->iotlb_enabled)) {
+            ret = vhost_backend_handle_iotlb_msg(dev, &payload.iotlb);
+        } else {
+            ret = -EFAULT;
+        }
           break;
       case VHOST_USER_SLAVE_CONFIG_CHANGE_MSG :
           ret = vhost_user_slave_handle_config_change(dev);
@@ -2044,7 +2049,8 @@ static int vhost_user_send_device_iotlb_msg(struct vhost_dev *dev,
 
   static void vhost_user_set_iotlb_callback(struct vhost_dev *dev, int enabled)
   {
-    /* No-op as the receive channel is not dedicated to IOTLB messages. */
+    struct vhost_user *u = dev->opaque;
+    u->iotlb_enabled = enabled;
   }
 
   static int vhost_user_get_config(struct vhost_dev *dev, uint8_t *config,
#5 lore raw
Quoted 30 lines · Eugenio Perez Martin
Quoted 29 lines · Jason Wang
Quoted 18 lines · Eugenio Perez Martin

Hi Jason.

Quoted 11 lines · Jason Wang
Quoted 5 lines · Eugenio Pérez

Not registering this can lead to vhost_backend_handle_iotlb_msg and vhost_device_iotlb_miss if backend issue a miss after qemu vhost device stop.

This causes a try to access dev->vdev->dma_as with vdev == NULL.

Hi Eugenio:

What condition can we get this? Did you mean we receive IOTLB_MISS before vhost_dev_start()?

In the VM reboot case, yes, between vhost_dev_stop() and vhost_dev_start(). But I can reproduce the bug in VM shutdown too, with no vhost_dev_start expected.

Quoted 5 lines

I didn't try to send IOTLB_MISS before starting vhost_dev, but this patch should solve that problem too.

I think we can get this with whatever malicious/buggy vhost-user backend sending unexpected iotlb misses, but I didn't test so deeply.

I see.

Quoted 7 lines
Quoted 6 lines
Quoted 2 lines
Quoted 1 line

If yes, it looks to me a bug somewhere else.

Where should I look for it?

So I winder whether or not we can simply ignore IOTLB message if vhost device is not started.

Do you mean like an early return in vhost_device_iotlb_miss?

Yes or probably a little bit earlier in vhost_backend_handle_iotlb_msg() which somehow a warn there.

Anyhow it's meaningless to process IOTLB message in this case and we don't need to introduce a dedicated variable for this.

Thanks

Quoted 73 lines

That was
my first first option, but this seems cleaner to me. I'm ok with both options.

Or do you mean not to return -EFAULT but 0 if !u->iotlb_enabled ?

Thanks!

Quoted 65 lines

Thanks

Quoted 62 lines

Thanks!

Quoted 60 lines

Thanks

Quoted 57 lines

Reproduced rebooting a guest with testpmd in txonly forward mode.

    #0  0x0000559ffff94394 in vhost_device_iotlb_miss (
        dev=dev@entry=0x55a0012f6680, iova=10245279744, write=1)
        at ../hw/virtio/vhost.c:1013
    #1  0x0000559ffff9ac31 in vhost_backend_handle_iotlb_msg (
        imsg=0x7ffddcfd32c0, dev=0x55a0012f6680)
        at ../hw/virtio/vhost-backend.c:411
    #2  vhost_backend_handle_iotlb_msg (dev=dev@entry=0x55a0012f6680,
        imsg=imsg@entry=0x7ffddcfd32c0)
        at ../hw/virtio/vhost-backend.c:404
    #3  0x0000559fffeded7b in slave_read (opaque=0x55a0012f6680)
        at ../hw/virtio/vhost-user.c:1464
    #4  0x000055a0000c541b in aio_dispatch_handler (
        ctx=ctx@entry=0x55a0010a2120, node=0x55a0012d9e00)
        at ../util/aio-posix.c:329
Fixes 6dcdd06e3b ("spec/vhost-user spec: Add IOMMU support")
---
    hw/virtio/vhost-user.c | 10 ++++++++--
    1 file changed, 8 insertions(+), 2 deletions(-)
hw/virtio/vhost-user.c +8 -2
index 2fdd5daf74..a49b2229fb 100644
@@ -238,6 +238,7 @@ struct vhost_user {
        /* Shared between vhost devs of the same virtio device */
        VhostUserState *user;
        int slave_fd;
+    bool iotlb_enabled;
        NotifierWithReturn postcopy_notifier;
        struct PostCopyFD  postcopy_fd;
        uint64_t           postcopy_client_bases[VHOST_USER_MAX_RAM_SLOTS];
@@ -1461,7 +1462,11 @@ static void slave_read(void *opaque)
 
        switch (hdr.request) {
        case VHOST_USER_SLAVE_IOTLB_MSG:
-        ret = vhost_backend_handle_iotlb_msg(dev, &payload.iotlb);
+        if (likely(u->iotlb_enabled)) {
+            ret = vhost_backend_handle_iotlb_msg(dev, &payload.iotlb);
+        } else {
+            ret = -EFAULT;
+        }
            break;
        case VHOST_USER_SLAVE_CONFIG_CHANGE_MSG :
            ret = vhost_user_slave_handle_config_change(dev);
@@ -2044,7 +2049,8 @@ static int vhost_user_send_device_iotlb_msg(struct vhost_dev *dev,
 
    static void vhost_user_set_iotlb_callback(struct vhost_dev *dev, int enabled)
    {
-    /* No-op as the receive channel is not dedicated to IOTLB messages. */
+    struct vhost_user *u = dev->opaque;
+    u->iotlb_enabled = enabled;
    }
 
    static int vhost_user_get_config(struct vhost_dev *dev, uint8_t *config,
#6 lore raw
Quoted 39 lines · Jason Wang
Quoted 36 lines · Eugenio Perez Martin
Quoted 35 lines · Jason Wang
Quoted 18 lines · Eugenio Perez Martin

Hi Jason.

Quoted 11 lines · Jason Wang
Quoted 5 lines · Eugenio Pérez

Not registering this can lead to vhost_backend_handle_iotlb_msg and vhost_device_iotlb_miss if backend issue a miss after qemu vhost device stop.

This causes a try to access dev->vdev->dma_as with vdev == NULL.

Hi Eugenio:

What condition can we get this? Did you mean we receive IOTLB_MISS before vhost_dev_start()?

In the VM reboot case, yes, between vhost_dev_stop() and vhost_dev_start(). But I can reproduce the bug in VM shutdown too, with no vhost_dev_start expected.

Quoted 5 lines

I didn't try to send IOTLB_MISS before starting vhost_dev, but this patch should solve that problem too.

I think we can get this with whatever malicious/buggy vhost-user backend sending unexpected iotlb misses, but I didn't test so deeply.

I see.

Quoted 2 lines
Quoted 1 line

If yes, it looks to me a bug somewhere else.

Where should I look for it?

So I winder whether or not we can simply ignore IOTLB message if vhost device is not started.

Quoted 9 lines
Quoted 1 line

Do you mean like an early return in vhost_device_iotlb_miss?

Yes or probably a little bit earlier in vhost_backend_handle_iotlb_msg() which somehow a warn there.

Anyhow it's meaningless to process IOTLB message in this case and we don't need to introduce a dedicated variable for this.

Start a new series, since the patch title does not reflect the changes anymore:

https://patchew.org/QEMU/[email protected]/

Thanks!

Quoted 77 lines

Thanks

Quoted 73 lines

That was
my first first option, but this seems cleaner to me. I'm ok with both options.

Or do you mean not to return -EFAULT but 0 if !u->iotlb_enabled ?

Thanks!

Quoted 65 lines

Thanks

Quoted 62 lines

Thanks!

Quoted 60 lines

Thanks

Quoted 57 lines

Reproduced rebooting a guest with testpmd in txonly forward mode.

    #0  0x0000559ffff94394 in vhost_device_iotlb_miss (
        dev=dev@entry=0x55a0012f6680, iova=10245279744, write=1)
        at ../hw/virtio/vhost.c:1013
    #1  0x0000559ffff9ac31 in vhost_backend_handle_iotlb_msg (
        imsg=0x7ffddcfd32c0, dev=0x55a0012f6680)
        at ../hw/virtio/vhost-backend.c:411
    #2  vhost_backend_handle_iotlb_msg (dev=dev@entry=0x55a0012f6680,
        imsg=imsg@entry=0x7ffddcfd32c0)
        at ../hw/virtio/vhost-backend.c:404
    #3  0x0000559fffeded7b in slave_read (opaque=0x55a0012f6680)
        at ../hw/virtio/vhost-user.c:1464
    #4  0x000055a0000c541b in aio_dispatch_handler (
        ctx=ctx@entry=0x55a0010a2120, node=0x55a0012d9e00)
        at ../util/aio-posix.c:329
Fixes 6dcdd06e3b ("spec/vhost-user spec: Add IOMMU support")
---
    hw/virtio/vhost-user.c | 10 ++++++++--
    1 file changed, 8 insertions(+), 2 deletions(-)
hw/virtio/vhost-user.c +8 -2
index 2fdd5daf74..a49b2229fb 100644
@@ -238,6 +238,7 @@ struct vhost_user {
        /* Shared between vhost devs of the same virtio device */
        VhostUserState *user;
        int slave_fd;
+    bool iotlb_enabled;
        NotifierWithReturn postcopy_notifier;
        struct PostCopyFD  postcopy_fd;
        uint64_t           postcopy_client_bases[VHOST_USER_MAX_RAM_SLOTS];
@@ -1461,7 +1462,11 @@ static void slave_read(void *opaque)
 
        switch (hdr.request) {
        case VHOST_USER_SLAVE_IOTLB_MSG:
-        ret = vhost_backend_handle_iotlb_msg(dev, &payload.iotlb);
+        if (likely(u->iotlb_enabled)) {
+            ret = vhost_backend_handle_iotlb_msg(dev, &payload.iotlb);
+        } else {
+            ret = -EFAULT;
+        }
            break;
        case VHOST_USER_SLAVE_CONFIG_CHANGE_MSG :
            ret = vhost_user_slave_handle_config_change(dev);
@@ -2044,7 +2049,8 @@ static int vhost_user_send_device_iotlb_msg(struct vhost_dev *dev,
 
    static void vhost_user_set_iotlb_callback(struct vhost_dev *dev, int enabled)
    {
-    /* No-op as the receive channel is not dedicated to IOTLB messages. */
+    struct vhost_user *u = dev->opaque;
+    u->iotlb_enabled = enabled;
    }
 
    static int vhost_user_get_config(struct vhost_dev *dev, uint8_t *config,