vmbus: vmbus implementation

Add the VMBus infrastructure -- bus, devices, root bridge, vmbus state
machine, vmbus channel interactions, etc.

VMBus is a collection of technologies.  At its lowest layer, it's a message
passing and signaling mechanism, allowing efficient passing of messages to and
from guest VMs.  A layer higher, it's a mechanism for defining channels of
communication, where each channel is tagged with a type (which implies a
protocol) and a instance ID.  A layer higher than that, it's a bus driver,
serving as the basis of device enumeration within a VM, where a channel can
optionally be exposed as a paravirtual device.  When a server-side (paravirtual
back-end) component wishes to offer a channel to a guest VM, it does so by
specifying a channel type, a mode, and an instance ID.  VMBus then exposes this
in the guest.

More information about VMBus can be found in the file
vmbuskernelmodeclientlibapi.h in Microsoft's WDK.

TODO:
 - split into smaller palatable pieces
 - more comments
 - check and handle corner cases

Kudos to Evgeny Yakovlev (formerly eyakovlev@virtuozzo.com) and Andrey
Smetatin (formerly asmetanin@virtuozzo.com) for research and
prototyping.

Signed-off-by: Roman Kagan <rkagan@virtuozzo.com>
Signed-off-by: Maciej S. Szmigiero <maciej.szmigiero@oracle.com>
Signed-off-by: Jon Doron <arilou@gmail.com>
Message-Id: <20200424123444.3481728-4-arilou@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Makefile.objs

@@ -150,6 +150,7 @@ trace-events-subdirs += hw/block/dataplane
 trace-events-subdirs += hw/char
 trace-events-subdirs += hw/dma
 trace-events-subdirs += hw/hppa
+trace-events-subdirs += hw/hyperv
 trace-events-subdirs += hw/i2c
 trace-events-subdirs += hw/i386
 trace-events-subdirs += hw/i386/xen

hw/hyperv/Kconfig

@@ -6,3 +6,8 @@ config HYPERV_TESTDEV
     bool
     default y if TEST_DEVICES
     depends on HYPERV
+
+config VMBUS
+    bool
+    default y
+    depends on HYPERV

hw/hyperv/Makefile.objs

@@ -1,2 +1,3 @@
 obj-y += hyperv.o
 obj-$(CONFIG_HYPERV_TESTDEV) += hyperv_testdev.o
+obj-$(CONFIG_VMBUS) += vmbus.o

hw/hyperv/trace-events

@@ -0,0 +1,18 @@
+# vmbus
+vmbus_recv_message(uint32_t type, uint32_t size) "type %d size %d"
+vmbus_signal_event(void) ""
+vmbus_channel_notify_guest(uint32_t chan_id) "channel #%d"
+vmbus_post_msg(uint32_t type, uint32_t size) "type %d size %d"
+vmbus_msg_cb(int status) "message status %d"
+vmbus_process_incoming_message(uint32_t message_type) "type %d"
+vmbus_initiate_contact(uint16_t major, uint16_t minor, uint32_t vcpu, uint64_t monitor_page1, uint64_t monitor_page2, uint64_t interrupt_page) "version %d.%d target vp %d mon pages 0x%"PRIx64",0x%"PRIx64" int page 0x%"PRIx64
+vmbus_send_offer(uint32_t chan_id, void *dev) "channel #%d dev %p"
+vmbus_terminate_offers(void) ""
+vmbus_gpadl_header(uint32_t gpadl_id, uint16_t num_gfns) "gpadl #%d gfns %d"
+vmbus_gpadl_body(uint32_t gpadl_id) "gpadl #%d"
+vmbus_gpadl_created(uint32_t gpadl_id) "gpadl #%d"
+vmbus_gpadl_teardown(uint32_t gpadl_id) "gpadl #%d"
+vmbus_gpadl_torndown(uint32_t gpadl_id) "gpadl #%d"
+vmbus_open_channel(uint32_t chan_id, uint32_t gpadl_id, uint32_t target_vp) "channel #%d gpadl #%d target vp %d"
+vmbus_channel_open(uint32_t chan_id, uint32_t status) "channel #%d status %d"
+vmbus_close_channel(uint32_t chan_id) "channel #%d"

include/hw/hyperv/vmbus-bridge.h

@@ -0,0 +1,32 @@
+/*
+ * QEMU Hyper-V VMBus root bridge
+ *
+ * Copyright (c) 2017-2018 Virtuozzo International GmbH.
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
+ * See the COPYING file in the top-level directory.
+ */
+
+#ifndef HW_HYPERV_VMBUS_BRIDGE_H
+#define HW_HYPERV_VMBUS_BRIDGE_H
+
+#include "hw/sysbus.h"
+
+#define TYPE_VMBUS_BRIDGE "vmbus-bridge"
+
+typedef struct VMBus VMBus;
+
+typedef struct VMBusBridge {
+    SysBusDevice parent_obj;
+
+    VMBus *bus;
+} VMBusBridge;
+
+#define VMBUS_BRIDGE(obj) OBJECT_CHECK(VMBusBridge, (obj), TYPE_VMBUS_BRIDGE)
+
+static inline VMBusBridge *vmbus_bridge_find(void)
+{
+    return VMBUS_BRIDGE(object_resolve_path_type("", TYPE_VMBUS_BRIDGE, NULL));
+}
+
+#endif

include/hw/hyperv/vmbus.h

@@ -0,0 +1,227 @@
+/*
+ * QEMU Hyper-V VMBus
+ *
+ * Copyright (c) 2017-2018 Virtuozzo International GmbH.
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
+ * See the COPYING file in the top-level directory.
+ */
+
+#ifndef HW_HYPERV_VMBUS_H
+#define HW_HYPERV_VMBUS_H
+
+#include "sysemu/sysemu.h"
+#include "sysemu/dma.h"
+#include "hw/qdev-core.h"
+#include "migration/vmstate.h"
+#include "hw/hyperv/vmbus-proto.h"
+#include "qemu/uuid.h"
+
+#define TYPE_VMBUS_DEVICE "vmbus-dev"
+
+#define VMBUS_DEVICE(obj) \
+    OBJECT_CHECK(VMBusDevice, (obj), TYPE_VMBUS_DEVICE)
+#define VMBUS_DEVICE_CLASS(klass) \
+    OBJECT_CLASS_CHECK(VMBusDeviceClass, (klass), TYPE_VMBUS_DEVICE)
+#define VMBUS_DEVICE_GET_CLASS(obj) \
+    OBJECT_GET_CLASS(VMBusDeviceClass, (obj), TYPE_VMBUS_DEVICE)
+
+/*
+ * Object wrapping a GPADL -- GPA Descriptor List -- an array of guest physical
+ * pages, to be used for various buffers shared between the host and the guest.
+ */
+typedef struct VMBusGpadl VMBusGpadl;
+/*
+ * VMBus channel -- a pair of ring buffers for either direction, placed within
+ * one GPADL, and the associated notification means.
+ */
+typedef struct VMBusChannel VMBusChannel;
+/*
+ * Base class for VMBus devices.  Includes one or more channels.  Identified by
+ * class GUID and instance GUID.
+ */
+typedef struct VMBusDevice VMBusDevice;
+
+typedef void(*VMBusChannelNotifyCb)(struct VMBusChannel *chan);
+
+typedef struct VMBusDeviceClass {
+    DeviceClass parent;
+
+    QemuUUID classid;
+    QemuUUID instanceid;     /* Fixed UUID for singleton devices */
+    uint16_t channel_flags;
+    uint16_t mmio_size_mb;
+
+    /* Extentions to standard device callbacks */
+    void (*vmdev_realize)(VMBusDevice *vdev, Error **errp);
+    void (*vmdev_unrealize)(VMBusDevice *vdev);
+    void (*vmdev_reset)(VMBusDevice *vdev);
+    /*
+     * Calculate the number of channels based on the device properties.  Called
+     * at realize time.
+     **/
+    uint16_t (*num_channels)(VMBusDevice *vdev);
+    /*
+     * Device-specific actions to complete the otherwise successful process of
+     * opening a channel.
+     * Return 0 on success, -errno on failure.
+     */
+    int (*open_channel)(VMBusChannel *chan);
+    /*
+     * Device-specific actions to perform before closing a channel.
+     */
+    void (*close_channel)(VMBusChannel *chan);
+    /*
+     * Main device worker; invoked in response to notifications from either
+     * side, when there's work to do with the data in the channel ring buffers.
+     */
+    VMBusChannelNotifyCb chan_notify_cb;
+} VMBusDeviceClass;
+
+struct VMBusDevice {
+    DeviceState parent;
+    QemuUUID instanceid;
+    uint16_t num_channels;
+    VMBusChannel *channels;
+    AddressSpace *dma_as;
+};
+
+extern const VMStateDescription vmstate_vmbus_dev;
+
+/*
+ * A unit of work parsed out of a message in the receive (i.e. guest->host)
+ * ring buffer of a channel.  It's supposed to be subclassed (through
+ * embedding) by the specific devices.
+ */
+typedef struct VMBusChanReq {
+    VMBusChannel *chan;
+    uint16_t pkt_type;
+    uint32_t msglen;
+    void *msg;
+    uint64_t transaction_id;
+    bool need_comp;
+    QEMUSGList sgl;
+} VMBusChanReq;
+
+VMBusDevice *vmbus_channel_device(VMBusChannel *chan);
+VMBusChannel *vmbus_device_channel(VMBusDevice *dev, uint32_t chan_idx);
+uint32_t vmbus_channel_idx(VMBusChannel *chan);
+bool vmbus_channel_is_open(VMBusChannel *chan);
+
+/*
+ * Notify (on guest's behalf) the host side of the channel that there's data in
+ * the ringbuffer to process.
+ */
+void vmbus_channel_notify_host(VMBusChannel *chan);
+
+/*
+ * Reserve space for a packet in the send (i.e. host->guest) ringbuffer.  If
+ * there isn't enough room, indicate that to the guest, to be notified when it
+ * becomes available.
+ * Return 0 on success, negative errno on failure.
+ * The ringbuffer indices are NOT updated, the requested space indicator may.
+ */
+int vmbus_channel_reserve(VMBusChannel *chan,
+                          uint32_t desclen, uint32_t msglen);
+
+/*
+ * Send a packet to the guest.  The space for the packet MUST be reserved
+ * first.
+ * Return total number of bytes placed in the send ringbuffer on success,
+ * negative errno on failure.
+ * The ringbuffer indices are updated on success, and the guest is signaled if
+ * needed.
+ */
+ssize_t vmbus_channel_send(VMBusChannel *chan, uint16_t pkt_type,
+                           void *desc, uint32_t desclen,
+                           void *msg, uint32_t msglen,
+                           bool need_comp, uint64_t transaction_id);
+
+/*
+ * Prepare to fetch a batch of packets from the receive ring buffer.
+ * Return 0 on success, negative errno on failure.
+ */
+int vmbus_channel_recv_start(VMBusChannel *chan);
+
+/*
+ * Shortcut for a common case of sending a simple completion packet with no
+ * auxiliary descriptors.
+ */
+ssize_t vmbus_channel_send_completion(VMBusChanReq *req,
+                                      void *msg, uint32_t msglen);
+
+/*
+ * Peek at the receive (i.e. guest->host) ring buffer and extract a unit of
+ * work (a device-specific subclass of VMBusChanReq) from a packet if there's
+ * one.
+ * Return an allocated buffer, containing the request of @size with filled
+ * VMBusChanReq at the beginning, followed by the message payload, or NULL on
+ * failure.
+ * The ringbuffer indices are NOT updated, nor is the private copy of the read
+ * index.
+ */
+void *vmbus_channel_recv_peek(VMBusChannel *chan, uint32_t size);
+
+/*
+ * Update the private copy of the read index once the preceding peek is deemed
+ * successful.
+ * The ringbuffer indices are NOT updated.
+ */
+void vmbus_channel_recv_pop(VMBusChannel *chan);
+
+/*
+ * Propagate the private copy of the read index into the receive ring buffer,
+ * and thus complete the reception of a series of packets.  Notify guest if
+ * needed.
+ * Return the number of bytes popped off the receive ring buffer by the
+ * preceding recv_peek/recv_pop calls on success, negative errno on failure.
+ */
+ssize_t vmbus_channel_recv_done(VMBusChannel *chan);
+
+/*
+ * Free the request allocated by vmbus_channel_recv_peek, together with its
+ * fields.
+ */
+void vmbus_free_req(void *req);
+
+/*
+ * Find and reference a GPADL by @gpadl_id.
+ * If not found return NULL.
+ */
+VMBusGpadl *vmbus_get_gpadl(VMBusChannel *chan, uint32_t gpadl_id);
+
+/*
+ * Unreference @gpadl.  If the reference count drops to zero, free it.
+ * @gpadl may be NULL, in which case nothing is done.
+ */
+void vmbus_put_gpadl(VMBusGpadl *gpadl);
+
+/*
+ * Calculate total length in bytes of @gpadl.
+ * @gpadl must be valid.
+ */
+uint32_t vmbus_gpadl_len(VMBusGpadl *gpadl);
+
+/*
+ * Copy data from @iov to @gpadl at offset @off.
+ * Return the number of bytes copied, or a negative status on failure.
+ */
+ssize_t vmbus_iov_to_gpadl(VMBusChannel *chan, VMBusGpadl *gpadl, uint32_t off,
+                           const struct iovec *iov, size_t iov_cnt);
+
+/*
+ * Map SGList contained in the request @req, at offset @off and no more than
+ * @len bytes, for io in direction @dir, and populate @iov with the mapped
+ * iovecs.
+ * Return the number of iovecs mapped, or negative status on failure.
+ */
+int vmbus_map_sgl(VMBusChanReq *req, DMADirection dir, struct iovec *iov,
+                  unsigned iov_cnt, size_t len, size_t off);
+
+/*
+ * Unmap *iov mapped with vmbus_map_sgl, marking the number of bytes @accessed.
+ */
+void vmbus_unmap_sgl(VMBusChanReq *req, DMADirection dir, struct iovec *iov,
+                     unsigned iov_cnt, size_t accessed);
+
+#endif