[Qemu-devel] [PATCH 1/2] docs: vhost-user: add in-band kick/call messages

[Qemu-devel] [PATCH 1/2] docs: vhost-user: add in-band kick/call messages

[Johannes Berg]

From: Johannes Berg <johannes.berg@intel.com>

For good reason, vhost-user is currently built asynchronously, that
way better performance can be obtained. However, for certain use
cases such as simulation, this is problematic.

Consider an event-based simulation in which both the device and CPU
have scheduled according to a simulation "calendar". Now, consider
the CPU sending I/O to the device, over a vring in the vhost-user
protocol. In this case, the CPU must wait for the vring interrupt
to have been processed by the device, so that the device is able to
put an entry onto the simulation calendar to obtain time to handle
the interrupt. Note that this doesn't mean the I/O is actually done
at this time, it just means that the handling of it is scheduled
before the CPU can continue running.

This cannot be done with the asynchronous eventfd based vring kick
and call design.

Extend the protocol slightly, so that a message can be used for kick
and call instead, if VHOST_USER_PROTOCOL_F_INBAND_NOTIFICATIONS is
negotiated. This in itself doesn't guarantee synchronisation, but both
sides can also negotiate VHOST_USER_PROTOCOL_F_REPLY_ACK and thus get
a reply to this message by setting the need_reply flag, and ensure
synchronisation this way.

To really use it in both directions, VHOST_USER_PROTOCOL_F_SLAVE_REQ
is also needed.

Since it is used for simulation purposes and too many messages on
the socket can lock up the virtual machine, document that this should
only be used together with the mentioned features.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
---
 docs/interop/vhost-user.rst | 120 +++++++++++++++++++++++++++++++-----
 1 file changed, 103 insertions(+), 17 deletions(-)

diff --git a/docs/interop/vhost-user.rst b/docs/interop/vhost-user.rst
index 7827b710aa0a..5f455262d55a 100644
--- a/docs/interop/vhost-user.rst
+++ b/docs/interop/vhost-user.rst
@@ -2,6 +2,7 @@
 Vhost-user Protocol
 ===================
 :Copyright: 2014 Virtual Open Systems Sarl.
+:Copyright: 2019 Intel Corporation
 :Licence: This work is licensed under the terms of the GNU GPL,
           version 2 or later. See the COPYING file in the top-level
           directory.
@@ -279,6 +280,9 @@ 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.
 
+If *slave* detects some error such as incompatible features, it may also
+close the connection. This should only happen in exceptional circumstances.
+
 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
@@ -315,7 +319,8 @@ it until ring is started, or after it has been stopped.
 
 Client must start ring upon receiving a kick (that is, detecting that
 file descriptor is readable) on the descriptor specified by
-``VHOST_USER_SET_VRING_KICK``, and stop ring upon receiving
+``VHOST_USER_SET_VRING_KICK`` or receiving the in-band message
+``VHOST_USER_VRING_KICK`` if negotiated, and stop ring upon receiving
 ``VHOST_USER_GET_VRING_BASE``.
 
 While processing the rings (whether they are enabled or not), client
@@ -767,24 +772,48 @@ When reconnecting:
 #. Resubmit inflight ``DescStatePacked`` entries in order of their
    counter value
 
+In-band notifications
+---------------------
+
+In some limited situations (e.g. for simulation) it is desirable to
+have the kick, call and error (if used) signals done via in-band
+messages instead of asynchronous eventfd notifications. This can be
+done by negotiating the ``VHOST_USER_PROTOCOL_F_INBAND_NOTIFICATIONS``
+protocol feature.
+
+Note that due to the fact that too many messages on the sockets can
+cause the sending application(s) to block, it is not advised to use
+this feature unless absolutely necessary. It is also considered an
+error to negotiate this feature without also negotiating
+``VHOST_USER_PROTOCOL_F_SLAVE_REQ`` and ``VHOST_USER_PROTOCOL_F_REPLY_ACK``,
+the former is necessary for getting a message channel from the slave
+to the master, while the latter needs to be used with the in-band
+notification messages to block until they are processed, both to avoid
+blocking later and for proper processing (at least in the simulation
+use case.) As it has no other way of signalling this error, the slave
+should close the connection as a response to a
+``VHOST_USER_SET_PROTOCOL_FEATURES`` message that sets the in-band
+notifications feature flag without the other two.
+
 Protocol features
 -----------------
 
 .. code:: c
 
-  #define VHOST_USER_PROTOCOL_F_MQ             0
-  #define VHOST_USER_PROTOCOL_F_LOG_SHMFD      1
-  #define VHOST_USER_PROTOCOL_F_RARP           2
-  #define VHOST_USER_PROTOCOL_F_REPLY_ACK      3
-  #define VHOST_USER_PROTOCOL_F_MTU            4
-  #define VHOST_USER_PROTOCOL_F_SLAVE_REQ      5
-  #define VHOST_USER_PROTOCOL_F_CROSS_ENDIAN   6
-  #define VHOST_USER_PROTOCOL_F_CRYPTO_SESSION 7
-  #define VHOST_USER_PROTOCOL_F_PAGEFAULT      8
-  #define VHOST_USER_PROTOCOL_F_CONFIG         9
-  #define VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD  10
-  #define VHOST_USER_PROTOCOL_F_HOST_NOTIFIER  11
-  #define VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD 12
+  #define VHOST_USER_PROTOCOL_F_MQ                    0
+  #define VHOST_USER_PROTOCOL_F_LOG_SHMFD             1
+  #define VHOST_USER_PROTOCOL_F_RARP                  2
+  #define VHOST_USER_PROTOCOL_F_REPLY_ACK             3
+  #define VHOST_USER_PROTOCOL_F_MTU                   4
+  #define VHOST_USER_PROTOCOL_F_SLAVE_REQ             5
+  #define VHOST_USER_PROTOCOL_F_CROSS_ENDIAN          6
+  #define VHOST_USER_PROTOCOL_F_CRYPTO_SESSION        7
+  #define VHOST_USER_PROTOCOL_F_PAGEFAULT             8
+  #define VHOST_USER_PROTOCOL_F_CONFIG                9
+  #define VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD        10
+  #define VHOST_USER_PROTOCOL_F_HOST_NOTIFIER        11
+  #define VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD       12
+  #define VHOST_USER_PROTOCOL_F_INBAND_NOTIFICATIONS 13
 
 Master message types
 --------------------
@@ -946,7 +975,12 @@ Master 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. This signals that polling should be used
-  instead of waiting for a kick.
+  instead of waiting for the kick. Note that if the protocol feature
+  ``VHOST_USER_PROTOCOL_F_INBAND_NOTIFICATIONS`` has been negotiated
+  this message isn't necessary as the ring is also started on the
+  ``VHOST_USER_VRING_KICK`` message, it may however still be used to
+  set an event file descriptor (which will be preferred over the
+  message) or to enable polling.
 
 ``VHOST_USER_SET_VRING_CALL``
   :id: 13
@@ -959,7 +993,12 @@ Master 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. This signals that polling will be used
-  instead of waiting for the call.
+  instead of waiting for the call. Note that if the protocol features
+  ``VHOST_USER_PROTOCOL_F_INBAND_NOTIFICATIONS`` and
+  ``VHOST_USER_PROTOCOL_F_SLAVE_REQ`` have been negotiated this message
+  isn't necessary as the ``VHOST_USER_SLAVE_VRING_CALL`` message can be
+  used, it may however still be used to set an event file descriptor
+  or to enable polling.
 
 ``VHOST_USER_SET_VRING_ERR``
   :id: 14
@@ -971,7 +1010,12 @@ Master 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.
+  in the ancillary data. Note that if the protocol features
+  ``VHOST_USER_PROTOCOL_F_INBAND_NOTIFICATIONS`` and
+  ``VHOST_USER_PROTOCOL_F_SLAVE_REQ`` have been negotiated this message
+  isn't necessary as the ``VHOST_USER_SLAVE_VRING_ERR`` message can be
+  used, it may however still be used to set an event file descriptor
+  (which will be preferred over the message).
 
 ``VHOST_USER_GET_QUEUE_NUM``
   :id: 17
@@ -1190,6 +1234,20 @@ Master message types
   ancillary data. The GPU protocol is used to inform the master of
   rendering state and updates. See vhost-user-gpu.rst for details.
 
+``VHOST_USER_VRING_KICK``
+  :id: 34
+  :equivalent ioctl: N/A
+  :slave payload: vring state description
+  :master payload: N/A
+
+  When the ``VHOST_USER_PROTOCOL_F_INBAND_NOTIFICATIONS`` protocol
+  feature has been successfully negotiated, this message may be
+  submitted by the master to indicate that a buffer was added to
+  the vring instead of signalling it using the vring's kick file
+  descriptor or having the slave rely on polling.
+
+  The state.num field is currently reserved and must be set to 0.
+
 Slave message types
 -------------------
 
@@ -1246,6 +1304,34 @@ Slave message types
   ``VHOST_USER_PROTOCOL_F_HOST_NOTIFIER`` protocol feature has been
   successfully negotiated.
 
+``VHOST_USER_SLAVE_VRING_CALL``
+  :id: 4
+  :equivalent ioctl: N/A
+  :slave payload: vring state description
+  :master payload: N/A
+
+  When the ``VHOST_USER_PROTOCOL_F_INBAND_NOTIFICATIONS`` protocol
+  feature has been successfully negotiated, this message may be
+  submitted by the slave to indicate that a buffer was used from
+  the vring instead of signalling this using the vring's call file
+  descriptor or having the master relying on polling.
+
+  The state.num field is currently reserved and must be set to 0.
+
+``VHOST_USER_SLAVE_VRING_ERR``
+  :id: 5
+  :equivalent ioctl: N/A
+  :slave payload: vring state description
+  :master payload: N/A
+
+  When the ``VHOST_USER_PROTOCOL_F_INBAND_NOTIFICATIONS`` protocol
+  feature has been successfully negotiated, this message may be
+  submitted by the slave to indicate that an error occurred on the
+  specific vring, instead of signalling the error file descriptor
+  set by the master via ``VHOST_USER_SET_VRING_ERR``.
+
+  The state.num field is currently reserved and must be set to 0.
+
 .. _reply_ack:
 
 VHOST_USER_PROTOCOL_F_REPLY_ACK
-- 
2.20.1

[Qemu-devel] [PATCH 2/2] libvhost-user: implement in-band notifications

[Johannes Berg]

From: Johannes Berg <johannes.berg@intel.com>

Add support for VHOST_USER_PROTOCOL_F_IN_BAND_NOTIFICATIONS, but
as it's not desired by default, don't enable it unless the device
implementation opts in by returning it from its protocol features
callback.

Note that I updated vu_set_vring_err_exec(), but didn't add any
sending of the VHOST_USER_SLAVE_VRING_ERR message as there's no
write to the err_fd today either.

This also adds vu_queue_notify_sync() which can be used to force
a synchronous notification if inband notifications are supported.
Previously, I had left out the slave->master direction handling
of F_REPLY_ACK, this now adds some code to support it as well.

Signed-off-by: Johannes Berg <johannes.berg@intel.com>
---
 contrib/libvhost-user/libvhost-user.c | 103 +++++++++++++++++++++++++-
 contrib/libvhost-user/libvhost-user.h |  14 ++++
 2 files changed, 114 insertions(+), 3 deletions(-)

diff --git a/contrib/libvhost-user/libvhost-user.c b/contrib/libvhost-user/libvhost-user.c
index 17b7833d1f6b..1e1525c1c7de 100644
--- a/contrib/libvhost-user/libvhost-user.c
+++ b/contrib/libvhost-user/libvhost-user.c
@@ -136,6 +136,7 @@ vu_request_to_string(unsigned int req)
         REQ(VHOST_USER_GET_INFLIGHT_FD),
         REQ(VHOST_USER_SET_INFLIGHT_FD),
         REQ(VHOST_USER_GPU_SET_SOCKET),
+        REQ(VHOST_USER_VRING_KICK),
         REQ(VHOST_USER_MAX),
     };
 #undef REQ
@@ -163,7 +164,10 @@ vu_panic(VuDev *dev, const char *msg, ...)
     dev->panic(dev, buf);
     free(buf);
 
-    /* FIXME: find a way to call virtio_error? */
+    /*
+     * FIXME:
+     * find a way to call virtio_error, or perhaps close the connection?
+     */
 }
 
 /* Translate guest physical address to our virtual address.  */
@@ -1172,6 +1176,14 @@ vu_set_vring_err_exec(VuDev *dev, VhostUserMsg *vmsg)
 static bool
 vu_get_protocol_features_exec(VuDev *dev, VhostUserMsg *vmsg)
 {
+    /*
+     * Note that we support, but intentionally do not set,
+     * VHOST_USER_PROTOCOL_F_INBAND_NOTIFICATIONS. This means that
+     * a device implementation can return it in its callback
+     * (get_protocol_features) if it wants to use this for
+     * simulation, but it is otherwise not desirable (if even
+     * implemented by the master.)
+     */
     uint64_t features = 1ULL << VHOST_USER_PROTOCOL_F_MQ |
                         1ULL << VHOST_USER_PROTOCOL_F_LOG_SHMFD |
                         1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ |
@@ -1204,6 +1216,25 @@ vu_set_protocol_features_exec(VuDev *dev, VhostUserMsg *vmsg)
 
     dev->protocol_features = vmsg->payload.u64;
 
+    if (vu_has_protocol_feature(dev,
+                                VHOST_USER_PROTOCOL_F_INBAND_NOTIFICATIONS) &&
+        (!vu_has_protocol_feature(dev, VHOST_USER_PROTOCOL_F_SLAVE_REQ) ||
+         !vu_has_protocol_feature(dev, VHOST_USER_PROTOCOL_F_REPLY_ACK))) {
+        /*
+         * The use case for using messages for kick/call is simulation, to make
+         * the kick and call synchronous. To actually get that behaviour, both
+         * of the other features are required.
+         * Theoretically, one could use only kick messages, or do them without
+         * having F_REPLY_ACK, but too many (possibly pending) messages on the
+         * socket will eventually cause the master to hang, to avoid this in
+         * scenarios where not desired enforce that the settings are in a way
+         * that actually enables the simulation case.
+         */
+        vu_panic(dev,
+                 "F_IN_BAND_NOTIFICATIONS requires F_SLAVE_REQ && F_REPLY_ACK");
+        return false;
+    }
+
     if (dev->iface->set_protocol_features) {
         dev->iface->set_protocol_features(dev, features);
     }
@@ -1464,6 +1495,34 @@ vu_set_inflight_fd(VuDev *dev, VhostUserMsg *vmsg)
     return false;
 }
 
+static bool
+vu_handle_vring_kick(VuDev *dev, VhostUserMsg *vmsg)
+{
+    unsigned int index = vmsg->payload.state.index;
+
+    if (index >= dev->max_queues) {
+        vu_panic(dev, "Invalid queue index: %u", index);
+        return false;
+    }
+
+    DPRINT("Got kick message: handler:%p idx:%d\n",
+           dev->vq[index].handler, index);
+
+    if (!dev->vq[index].started) {
+        dev->vq[index].started = true;
+
+        if (dev->iface->queue_set_started) {
+            dev->iface->queue_set_started(dev, index, true);
+        }
+    }
+
+    if (dev->vq[index].handler) {
+        dev->vq[index].handler(dev, index);
+    }
+
+    return false;
+}
+
 static bool
 vu_process_message(VuDev *dev, VhostUserMsg *vmsg)
 {
@@ -1546,6 +1605,8 @@ vu_process_message(VuDev *dev, VhostUserMsg *vmsg)
         return vu_get_inflight_fd(dev, vmsg);
     case VHOST_USER_SET_INFLIGHT_FD:
         return vu_set_inflight_fd(dev, vmsg);
+    case VHOST_USER_VRING_KICK:
+        return vu_handle_vring_kick(dev, vmsg);
     default:
         vmsg_close_fds(vmsg);
         vu_panic(dev, "Unhandled request: %d", vmsg->request);
@@ -2005,8 +2066,7 @@ vring_notify(VuDev *dev, VuVirtq *vq)
     return !v || vring_need_event(vring_get_used_event(vq), new, old);
 }
 
-void
-vu_queue_notify(VuDev *dev, VuVirtq *vq)
+static void _vu_queue_notify(VuDev *dev, VuVirtq *vq, bool sync)
 {
     if (unlikely(dev->broken) ||
         unlikely(!vq->vring.avail)) {
@@ -2018,11 +2078,48 @@ vu_queue_notify(VuDev *dev, VuVirtq *vq)
         return;
     }
 
+    if (vq->call_fd < 0 &&
+        vu_has_protocol_feature(dev,
+                                VHOST_USER_PROTOCOL_F_INBAND_NOTIFICATIONS) &&
+        vu_has_protocol_feature(dev, VHOST_USER_PROTOCOL_F_SLAVE_REQ)) {
+        VhostUserMsg vmsg = {
+            .request = VHOST_USER_SLAVE_VRING_CALL,
+            .flags = VHOST_USER_VERSION,
+            .size = sizeof(vmsg.payload.state),
+            .payload.state = {
+                .index = vq - dev->vq,
+            },
+        };
+        bool ack = sync &&
+                   vu_has_protocol_feature(dev,
+                                           VHOST_USER_PROTOCOL_F_REPLY_ACK);
+
+        if (ack) {
+            vmsg.flags |= VHOST_USER_NEED_REPLY_MASK;
+        }
+
+        vu_message_write(dev, dev->slave_fd, &vmsg);
+        if (ack) {
+            vu_message_read(dev, dev->slave_fd, &vmsg);
+        }
+        return;
+    }
+
     if (eventfd_write(vq->call_fd, 1) < 0) {
         vu_panic(dev, "Error writing eventfd: %s", strerror(errno));
     }
 }
 
+void vu_queue_notify(VuDev *dev, VuVirtq *vq)
+{
+    _vu_queue_notify(dev, vq, false);
+}
+
+void vu_queue_notify_sync(VuDev *dev, VuVirtq *vq)
+{
+    _vu_queue_notify(dev, vq, true);
+}
+
 static inline void
 vring_used_flags_set_bit(VuVirtq *vq, int mask)
 {
diff --git a/contrib/libvhost-user/libvhost-user.h b/contrib/libvhost-user/libvhost-user.h
index 46b600799b2e..8e780444429f 100644
--- a/contrib/libvhost-user/libvhost-user.h
+++ b/contrib/libvhost-user/libvhost-user.h
@@ -53,6 +53,7 @@ enum VhostUserProtocolFeature {
     VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD = 10,
     VHOST_USER_PROTOCOL_F_HOST_NOTIFIER = 11,
     VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD = 12,
+    VHOST_USER_PROTOCOL_F_INBAND_NOTIFICATIONS = 13,
 
     VHOST_USER_PROTOCOL_F_MAX
 };
@@ -94,6 +95,7 @@ typedef enum VhostUserRequest {
     VHOST_USER_GET_INFLIGHT_FD = 31,
     VHOST_USER_SET_INFLIGHT_FD = 32,
     VHOST_USER_GPU_SET_SOCKET = 33,
+    VHOST_USER_VRING_KICK = 34,
     VHOST_USER_MAX
 } VhostUserRequest;
 
@@ -102,6 +104,8 @@ typedef enum VhostUserSlaveRequest {
     VHOST_USER_SLAVE_IOTLB_MSG = 1,
     VHOST_USER_SLAVE_CONFIG_CHANGE_MSG = 2,
     VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG = 3,
+    VHOST_USER_SLAVE_VRING_CALL = 4,
+    VHOST_USER_SLAVE_VRING_ERR = 5,
     VHOST_USER_SLAVE_MAX
 }  VhostUserSlaveRequest;
 
@@ -522,6 +526,16 @@ bool vu_queue_empty(VuDev *dev, VuVirtq *vq);
  */
 void vu_queue_notify(VuDev *dev, VuVirtq *vq);
 
+/**
+ * vu_queue_notify_sync:
+ * @dev: a VuDev context
+ * @vq: a VuVirtq queue
+ *
+ * Request to notify the queue via callfd (skipped if unnecessary)
+ * or sync message if possible.
+ */
+void vu_queue_notify_sync(VuDev *dev, VuVirtq *vq);
+
 /**
  * vu_queue_pop:
  * @dev: a VuDev context
-- 
2.20.1

Re: [Qemu-devel] [PATCH 1/2] docs: vhost-user: add in-band kick/call messages

[Stefan Hajnoczi]

[-- Attachment #1: Type: text/plain, Size: 1943 bytes --]

On Tue, Sep 17, 2019 at 02:26:24PM +0200, Johannes Berg wrote:
> From: Johannes Berg <johannes.berg@intel.com>
> 
> For good reason, vhost-user is currently built asynchronously, that
> way better performance can be obtained. However, for certain use
> cases such as simulation, this is problematic.
> 
> Consider an event-based simulation in which both the device and CPU
> have scheduled according to a simulation "calendar". Now, consider
> the CPU sending I/O to the device, over a vring in the vhost-user
> protocol. In this case, the CPU must wait for the vring interrupt
> to have been processed by the device, so that the device is able to
> put an entry onto the simulation calendar to obtain time to handle
> the interrupt. Note that this doesn't mean the I/O is actually done
> at this time, it just means that the handling of it is scheduled
> before the CPU can continue running.
> 
> This cannot be done with the asynchronous eventfd based vring kick
> and call design.
> 
> Extend the protocol slightly, so that a message can be used for kick
> and call instead, if VHOST_USER_PROTOCOL_F_INBAND_NOTIFICATIONS is
> negotiated. This in itself doesn't guarantee synchronisation, but both
> sides can also negotiate VHOST_USER_PROTOCOL_F_REPLY_ACK and thus get
> a reply to this message by setting the need_reply flag, and ensure
> synchronisation this way.
> 
> To really use it in both directions, VHOST_USER_PROTOCOL_F_SLAVE_REQ
> is also needed.
> 
> Since it is used for simulation purposes and too many messages on
> the socket can lock up the virtual machine, document that this should
> only be used together with the mentioned features.
> 
> Signed-off-by: Johannes Berg <johannes.berg@intel.com>
> ---
>  docs/interop/vhost-user.rst | 120 +++++++++++++++++++++++++++++++-----
>  1 file changed, 103 insertions(+), 17 deletions(-)

Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]