[Qemu-devel] [PATCH v6 00/11] rdma: migration support

[Qemu-devel] [PATCH v6 00/11] rdma: migration support

[mrhines]

From: "Michael R. Hines" <mrhines@us.ibm.com>

Please pull.

Changes since v5:

- Removed max_size hook.
- Waiting for Signed-Off bys....

Wiki: http://wiki.qemu.org/Features/RDMALiveMigration
Github: git@github.com:hinesmr/qemu.git

Here is a brief summary of total migration time and downtime using RDMA:

Using a 40gbps infiniband link performing a worst-case stress test,
using an 8GB RAM virtual machine:
Using the following command:

$ apt-get install stress
$ stress --vm-bytes 7500M --vm 1 --vm-keep

RESULTS:

1. Migration throughput: 26 gigabits/second.
2. Downtime (stop time) varies between 15 and 100 milliseconds.

EFFECTS of memory registration on bulk phase round:

For example, in the same 8GB RAM example with all 8GB of memory in 
active use and the VM itself is completely idle using the same 40 gbps 
infiniband link:

1. x-rdma-pin-all disabled total time: approximately 7.5 seconds @ 9.5 Gbps
2. x-rdma-pin-all enabled total time: approximately 4 seconds @ 26 Gbps

These numbers would of course scale up to whatever size virtual machine
you have to migrate using RDMA.

Enabling this feature does *not* have any measurable affect on 
migration *downtime*. This is because, without this feature, all of the 
memory will have already been registered already in advance during
the bulk round and does not need to be re-registered during the successive
iteration rounds.

The following changes since commit f3aa844bbb2922a5b8393d17620eca7d7e921ab3:

  build: include config-{, all-}devices.mak after defining CONFIG_SOFTMMU and CONFIG_USER_ONLY (2013-04-24 12:18:41 -0500)

are available in the git repository at:

  git@github.com:hinesmr/qemu.git rdma_patch_v6

for you to fetch changes up to 75e6fac1f642885b93cefe6e1874d648e9850f8f:

  rdma: send pc.ram (2013-04-24 14:55:01 -0400)

----------------------------------------------------------------
Michael R. Hines (11):
      rdma: add documentation
      rdma: export yield_until_fd_readable()
      rdma: export throughput w/ MigrationStats QMP
      rdma: introduce qemu_file_mode_is_not_valid()
      rdma: export qemu_fflush()
      rdma: introduce  ram_handle_compressed()
      rdma: introduce qemu_ram_foreach_block()
      rdma: new QEMUFileOps hooks
      rdma: introduce capability x-rdma-pin-all
      rdma: core logic
      rdma: send pc.ram

 Makefile.objs                 |    1 +
 arch_init.c                   |   59 +-
 configure                     |   29 +
 docs/rdma.txt                 |  404 ++++++
 exec.c                        |    9 +
 hmp.c                         |    2 +
 include/block/coroutine.h     |    6 +
 include/exec/cpu-common.h     |    5 +
 include/migration/migration.h |   25 +
 include/migration/qemu-file.h |   30 +
 migration-rdma.c              | 2707 +++++++++++++++++++++++++++++++++++++++++
 migration.c                   |   27 +
 qapi-schema.json              |   12 +-
 qemu-coroutine-io.c           |   23 +
 savevm.c                      |  107 +-
 15 files changed, 3398 insertions(+), 48 deletions(-)
 create mode 100644 docs/rdma.txt
 create mode 100644 migration-rdma.c

-- 
1.7.10.4

[Qemu-devel] [PATCH v6 01/11] rdma: add documentation

[mrhines]

From: "Michael R. Hines" <mrhines@us.ibm.com>

docs/rdma.txt contains full documentation,
wiki links, github url and contact information.

Signed-off-by: Michael R. Hines <mrhines@us.ibm.com>
---
 docs/rdma.txt |  404 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 404 insertions(+)
 create mode 100644 docs/rdma.txt

diff --git a/docs/rdma.txt b/docs/rdma.txt
new file mode 100644
index 0000000..4f98a3b
--- /dev/null
+++ b/docs/rdma.txt
@@ -0,0 +1,404 @@
+(RDMA: Remote Direct Memory Access)
+RDMA Live Migration Specification, Version # 1
+==============================================
+Wiki: http://wiki.qemu.org/Features/RDMALiveMigration
+Github: git@github.com:hinesmr/qemu.git, 'rdma' branch
+
+Copyright (C) 2013 Michael R. Hines <mrhines@us.ibm.com>
+
+An *exhaustive* paper (2010) shows additional performance details
+linked on the QEMU wiki above.
+
+Contents:
+=========
+* Introduction
+* Before running
+* Running
+* Performance
+* RDMA Migration Protocol Description
+* Versioning and Capabilities
+* QEMUFileRDMA Interface
+* Migration of pc.ram
+* Error handling
+* TODO
+
+Introduction:
+=============
+
+RDMA helps make your migration more deterministic under heavy load because
+of the significantly lower latency and higher throughput over TCP/IP. This is
+because the RDMA I/O architecture reduces the number of interrupts and
+data copies by bypassing the host networking stack. In particular, a TCP-based
+migration, under certain types of memory-bound workloads, may take a more
+unpredicatable amount of time to complete the migration if the amount of
+memory tracked during each live migration iteration round cannot keep pace
+with the rate of dirty memory produced by the workload.
+
+RDMA currently comes in two flavors: both Ethernet based (RoCE, or RDMA
+over Convered Ethernet) as well as Infiniband-based. This implementation of
+migration using RDMA is capable of using both technologies because of
+the use of the OpenFabrics OFED software stack that abstracts out the
+programming model irrespective of the underlying hardware.
+
+Refer to openfabrics.org or your respective RDMA hardware vendor for
+an understanding on how to verify that you have the OFED software stack
+installed in your environment. You should be able to successfully link
+against the "librdmacm" and "libibverbs" libraries and development headers
+for a working build of QEMU to run successfully using RDMA Migration.
+
+BEFORE RUNNING:
+===============
+
+Use of RDMA during migration requires pinning and registering memory
+with the hardware. This means that memory must be physically resident
+before the hardware can transmit that memory to another machine.
+If this is not acceptable for your application or product, then the use
+of RDMA migration may in fact be harmful to co-located VMs or other
+software on the machine if there is not sufficient memory available to
+relocate the entire footprint of the virtual machine. If so, then the
+use of RDMA is discouraged and it is recommended to use standard TCP migration.
+
+Experimental: Next, decide if you want dynamic page registration.
+For example, if you have an 8GB RAM virtual machine, but only 1GB
+is in active use, then enabling this feature will cause all 8GB to
+be pinned and resident in memory. This feature mostly affects the
+bulk-phase round of the migration and can be enabled for extremely
+high-performance RDMA hardware using the following command:
+
+QEMU Monitor Command:
+$ migrate_set_capability x-rdma-pin-all on # disabled by default
+
+Performing this action will cause all 8GB to be pinned, so if that's
+not what you want, then please ignore this step altogether.
+
+On the other hand, this will also significantly speed up the bulk round
+of the migration, which can greatly reduce the "total" time of your migration.
+Example performance of this using an idle VM in the previous example
+can be found in the "Performance" section.
+
+RUNNING:
+========
+
+First, set the migration speed to match your hardware's capabilities:
+
+QEMU Monitor Command:
+$ migrate_set_speed 40g # or whatever is the MAX of your RDMA device
+
+Next, on the destination machine, add the following to the QEMU command line:
+
+qemu ..... -incoming x-rdma:host:port
+
+Finally, perform the actual migration on the source machine:
+
+QEMU Monitor Command:
+$ migrate -d x-rdma:host:port
+
+PERFORMANCE
+===========
+
+Here is a brief summary of total migration time and downtime using RDMA:
+Using a 40gbps infiniband link performing a worst-case stress test,
+using an 8GB RAM virtual machine:
+
+Using the following command:
+$ apt-get install stress
+$ stress --vm-bytes 7500M --vm 1 --vm-keep
+
+1. Migration throughput: 26 gigabits/second.
+2. Downtime (stop time) varies between 15 and 100 milliseconds.
+
+EFFECTS of memory registration on bulk phase round:
+
+For example, in the same 8GB RAM example with all 8GB of memory in
+active use and the VM itself is completely idle using the same 40 gbps
+infiniband link:
+
+1. x-rdma-pin-all disabled total time: approximately 7.5 seconds @ 9.5 Gbps
+2. x-rdma-pin-all enabled total time: approximately 4 seconds @ 26 Gbps
+
+These numbers would of course scale up to whatever size virtual machine
+you have to migrate using RDMA.
+
+Enabling this feature does *not* have any measurable affect on
+migration *downtime*. This is because, without this feature, all of the
+memory will have already been registered already in advance during
+the bulk round and does not need to be re-registered during the successive
+iteration rounds.
+
+RDMA Protocol Description:
+==========================
+
+Migration with RDMA is separated into two parts:
+
+1. The transmission of the pages using RDMA
+2. Everything else (a control channel is introduced)
+
+"Everything else" is transmitted using a formal
+protocol now, consisting of infiniband SEND messages.
+
+An infiniband SEND message is the standard ibverbs
+message used by applications of infiniband hardware.
+The only difference between a SEND message and an RDMA
+message is that SEND messages cause notifications
+to be posted to the completion queue (CQ) on the
+infiniband receiver side, whereas RDMA messages (used
+for pc.ram) do not (to behave like an actual DMA).
+
+Messages in infiniband require two things:
+
+1. registration of the memory that will be transmitted
+2. (SEND only) work requests to be posted on both
+   sides of the network before the actual transmission
+   can occur.
+
+RDMA messages are much easier to deal with. Once the memory
+on the receiver side is registered and pinned, we're
+basically done. All that is required is for the sender
+side to start dumping bytes onto the link.
+
+(Memory is not released from pinning until the migration
+completes, given that RDMA migrations are very fast.)
+
+SEND messages require more coordination because the
+receiver must have reserved space (using a receive
+work request) on the receive queue (RQ) before QEMUFileRDMA
+can start using them to carry all the bytes as
+a control transport for migration of device state.
+
+To begin the migration, the initial connection setup is
+as follows (migration-rdma.c):
+
+1. Receiver and Sender are started (command line or libvirt):
+2. Both sides post two RQ work requests
+3. Receiver does listen()
+4. Sender does connect()
+5. Receiver accept()
+6. Check versioning and capabilities (described later)
+
+At this point, we define a control channel on top of SEND messages
+which is described by a formal protocol. Each SEND message has a
+header portion and a data portion (but together are transmitted
+as a single SEND message).
+
+Header:
+    * Length  (of the data portion, uint32, network byte order)
+    * Type    (what command to perform, uint32, network byte order)
+    * Repeat  (Number of commands in data portion, same type only)
+
+The 'Repeat' field is here to support future multiple page registrations
+in a single message without any need to change the protocol itself
+so that the protocol is compatible against multiple versions of QEMU.
+Version #1 requires that all server implementations of the protocol must
+check this field and register all requests found in the array of commands located
+in the data portion and return an equal number of results in the response.
+The maximum number of repeats is hard-coded to 4096. This is a conservative
+limit based on the maximum size of a SEND message along with emperical
+observations on the maximum future benefit of simultaneous page registrations.
+
+The 'type' field has 9 different command values:
+    1. Unused
+    2. Error             (sent to the source during bad things)
+    3. Ready             (control-channel is available)
+    4. QEMU File         (for sending non-live device state)
+    5. RAM Blocks        (used right after connection setup)
+    6. Compress page     (zap zero page and skip registration)
+    7. Register request  (dynamic chunk registration)
+    8. Register result   ('rkey' to be used by sender)
+    9. Register finished (registration for current iteration finished)
+
+A single control message, as hinted above, can contain within the data
+portion an array of many commands of the same type. If there is more than
+one command, then the 'repeat' field will be greater than 1.
+
+After connection setup is completed, we have two protocol-level
+functions, responsible for communicating control-channel commands
+using the above list of values:
+
+Logically:
+
+qemu_rdma_exchange_recv(header, expected command type)
+
+1. We transmit a READY command to let the sender know that
+   we are *ready* to receive some data bytes on the control channel.
+2. Before attempting to receive the expected command, we post another
+   RQ work request to replace the one we just used up.
+3. Block on a CQ event channel and wait for the SEND to arrive.
+4. When the send arrives, librdmacm will unblock us.
+5. Verify that the command-type and version received matches the one we expected.
+
+qemu_rdma_exchange_send(header, data, optional response header & data):
+
+1. Block on the CQ event channel waiting for a READY command
+   from the receiver to tell us that the receiver
+   is *ready* for us to transmit some new bytes.
+2. Optionally: if we are expecting a response from the command
+   (that we have no yet transmitted), let's post an RQ
+   work request to receive that data a few moments later.
+3. When the READY arrives, librdmacm will
+   unblock us and we immediately post a RQ work request
+   to replace the one we just used up.
+4. Now, we can actually post the work request to SEND
+   the requested command type of the header we were asked for.
+5. Optionally, if we are expecting a response (as before),
+   we block again and wait for that response using the additional
+   work request we previously posted. (This is used to carry
+   'Register result' commands #6 back to the sender which
+   hold the rkey need to perform RDMA. Note that the virtual address
+   corresponding to this rkey was already exchanged at the beginning
+   of the connection (described below).
+
+All of the remaining command types (not including 'ready')
+described above all use the aformentioned two functions to do the hard work:
+
+1. After connection setup, RAMBlock information is exchanged using
+   this protocol before the actual migration begins. This information includes
+   a description of each RAMBlock on the server side as well as the virtual addresses
+   and lengths of each RAMBlock. This is used by the client to determine the
+   start and stop locations of chunks and how to register them dynamically
+   before performing the RDMA operations.
+2. During runtime, once a 'chunk' becomes full of pages ready to
+   be sent with RDMA, the registration commands are used to ask the
+   other side to register the memory for this chunk and respond
+   with the result (rkey) of the registration.
+3. Also, the QEMUFile interfaces also call these functions (described below)
+   when transmitting non-live state, such as devices or to send
+   its own protocol information during the migration process.
+4. Finally, zero pages are only checked if a page has not yet been registered
+   using chunk registration (or not checked at all and unconditionally
+   written if chunk registration is disabled. This is accomplished using
+   the "Compress" command listed above. If the page *has* been registered
+   then we check the entire chunk for zero. Only if the entire chunk is
+   zero, then we send a compress command to zap the page on the other side.
+
+Versioning and Capabilities
+===========================
+Current version of the protocol is version #1.
+
+The same version applies to both for protocol traffic and capabilities
+negotiation. (i.e. There is only one version number that is referred to
+by all communication).
+
+librdmacm provides the user with a 'private data' area to be exchanged
+at connection-setup time before any infiniband traffic is generated.
+
+Header:
+    * Version (protocol version validated before send/recv occurs), uint32, network byte order
+    * Flags   (bitwise OR of each capability), uint32, network byte order
+
+There is no data portion of this header right now, so there is
+no length field. The maximum size of the 'private data' section
+is only 192 bytes per the Infiniband specification, so it's not
+very useful for data anyway. This structure needs to remain small.
+
+This private data area is a convenient place to check for protocol
+versioning because the user does not need to register memory to
+transmit a few bytes of version information.
+
+This is also a convenient place to negotiate capabilities
+(like dynamic page registration).
+
+If the version is invalid, we throw an error.
+
+If the version is new, we only negotiate the capabilities that the
+requested version is able to perform and ignore the rest.
+
+Currently there is only *one* capability in Version #1: dynamic page registration
+
+Finally: Negotiation happens with the Flags field: If the primary-VM
+sets a flag, but the destination does not support this capability, it
+will return a zero-bit for that flag and the primary-VM will understand
+that as not being an available capability and will thus disable that
+capability on the primary-VM side.
+
+QEMUFileRDMA Interface:
+=======================
+
+QEMUFileRDMA introduces a couple of new functions:
+
+1. qemu_rdma_get_buffer()  (QEMUFileOps rdma_read_ops)
+2. qemu_rdma_put_buffer()  (QEMUFileOps rdma_write_ops)
+
+These two functions are very short and simply use the protocol
+describe above to deliver bytes without changing the upper-level
+users of QEMUFile that depend on a bytestream abstraction.
+
+Finally, how do we handoff the actual bytes to get_buffer()?
+
+Again, because we're trying to "fake" a bytestream abstraction
+using an analogy not unlike individual UDP frames, we have
+to hold on to the bytes received from control-channel's SEND
+messages in memory.
+
+Each time we receive a complete "QEMU File" control-channel
+message, the bytes from SEND are copied into a small local holding area.
+
+Then, we return the number of bytes requested by get_buffer()
+and leave the remaining bytes in the holding area until get_buffer()
+comes around for another pass.
+
+If the buffer is empty, then we follow the same steps
+listed above and issue another "QEMU File" protocol command,
+asking for a new SEND message to re-fill the buffer.
+
+Migration of pc.ram:
+====================
+
+At the beginning of the migration, (migration-rdma.c),
+the sender and the receiver populate the list of RAMBlocks
+to be registered with each other into a structure.
+Then, using the aforementioned protocol, they exchange a
+description of these blocks with each other, to be used later
+during the iteration of main memory. This description includes
+a list of all the RAMBlocks, their offsets and lengths, virtual
+addresses and possibly includes pre-registered RDMA keys in case dynamic
+page registration was disabled on the server-side, otherwise not.
+
+Main memory is not migrated with the aforementioned protocol,
+but is instead migrated with normal RDMA Write operations.
+
+Pages are migrated in "chunks" (hard-coded to 1 Megabyte right now).
+Chunk size is not dynamic, but it could be in a future implementation.
+There's nothing to indicate that this is useful right now.
+
+When a chunk is full (or a flush() occurs), the memory backed by
+the chunk is registered with librdmacm is pinned in memory on
+both sides using the aforementioned protocol.
+After pinning, an RDMA Write is generated and transmitted
+for the entire chunk.
+
+Chunks are also transmitted in batches: This means that we
+do not request that the hardware signal the completion queue
+for the completion of *every* chunk. The current batch size
+is about 64 chunks (corresponding to 64 MB of memory).
+Only the last chunk in a batch must be signaled.
+This helps keep everything as asynchronous as possible
+and helps keep the hardware busy performing RDMA operations.
+
+Error-handling:
+===============
+
+Infiniband has what is called a "Reliable, Connected"
+link (one of 4 choices). This is the mode in which
+we use for RDMA migration.
+
+If a *single* message fails,
+the decision is to abort the migration entirely and
+cleanup all the RDMA descriptors and unregister all
+the memory.
+
+After cleanup, the Virtual Machine is returned to normal
+operation the same way that would happen if the TCP
+socket is broken during a non-RDMA based migration.
+
+TODO:
+=====
+1. 'migrate x-rdma:host:port' and '-incoming x-rdma' options will be
+   renamed to 'rdma' after the experimental phase of this work has
+   completed upstream.
+2. Currently, 'ulimit -l' mlock() limits as well as cgroups swap limits
+   are not compatible with infinband memory pinning and will result in
+   an aborted migration (but with the source VM left unaffected).
+3. Use of the recent /proc/<pid>/pagemap would likely speed up
+   the use of KSM and ballooning while using RDMA.
+4. Also, some form of balloon-device usage tracking would also
+   help alleviate some issues.
-- 
1.7.10.4

[Qemu-devel] [PATCH v6 02/11] rdma: export yield_until_fd_readable()

[mrhines]

From: "Michael R. Hines" <mrhines@us.ibm.com>

The RDMA event channel can be made non-blocking just like a TCP
socket. Exporting this function allows us to yield so that the
QEMU monitor remains available.

Signed-off-by: Michael R. Hines <mrhines@us.ibm.com>
---
 include/block/coroutine.h |    6 +
 qemu-coroutine-io.c       |   23 +
 savevm.c                  |   28 -
 4 files changed, 29 insertions(+), 28 deletions(-)
 create mode 100644 migration-rdma.c

diff --git a/include/block/coroutine.h b/include/block/coroutine.h
index a978162..377805a 100644
--- a/include/block/coroutine.h
+++ b/include/block/coroutine.h
@@ -209,4 +209,10 @@ void qemu_co_rwlock_unlock(CoRwlock *lock);
  */
 void coroutine_fn co_sleep_ns(QEMUClock *clock, int64_t ns);
 
+/**
+ * Yield until a file descriptor becomes readable
+ *
+ * Note that this function clobbers the handlers for the file descriptor.
+ */
+void coroutine_fn yield_until_fd_readable(int fd);
 #endif /* QEMU_COROUTINE_H */
diff --git a/qemu-coroutine-io.c b/qemu-coroutine-io.c
index e8ad1a4..c4df35a 100644
--- a/qemu-coroutine-io.c
+++ b/qemu-coroutine-io.c
@@ -63,3 +63,26 @@ qemu_co_send_recv(int sockfd, void *buf, size_t bytes, bool do_send)
     struct iovec iov = { .iov_base = buf, .iov_len = bytes };
     return qemu_co_sendv_recvv(sockfd, &iov, 1, 0, bytes, do_send);
 }
+
+typedef struct {
+    Coroutine *co;
+    int fd;
+} FDYieldUntilData;
+
+static void fd_coroutine_enter(void *opaque)
+{
+    FDYieldUntilData *data = opaque;
+    qemu_set_fd_handler(data->fd, NULL, NULL, NULL);
+    qemu_coroutine_enter(data->co, NULL);
+}
+
+void coroutine_fn yield_until_fd_readable(int fd)
+{
+    FDYieldUntilData data;
+
+    assert(qemu_in_coroutine());
+    data.co = qemu_coroutine_self();
+    data.fd = fd;
+    qemu_set_fd_handler(fd, fd_coroutine_enter, NULL, &data);
+    qemu_coroutine_yield();
+}
diff --git a/savevm.c b/savevm.c
index 31dcce9..4dbfa16 100644
--- a/savevm.c
+++ b/savevm.c
@@ -147,34 +147,6 @@ typedef struct QEMUFileSocket
     QEMUFile *file;
 } QEMUFileSocket;
 
-typedef struct {
-    Coroutine *co;
-    int fd;
-} FDYieldUntilData;
-
-static void fd_coroutine_enter(void *opaque)
-{
-    FDYieldUntilData *data = opaque;
-    qemu_set_fd_handler(data->fd, NULL, NULL, NULL);
-    qemu_coroutine_enter(data->co, NULL);
-}
-
-/**
- * Yield until a file descriptor becomes readable
- *
- * Note that this function clobbers the handlers for the file descriptor.
- */
-static void coroutine_fn yield_until_fd_readable(int fd)
-{
-    FDYieldUntilData data;
-
-    assert(qemu_in_coroutine());
-    data.co = qemu_coroutine_self();
-    data.fd = fd;
-    qemu_set_fd_handler(fd, fd_coroutine_enter, NULL, &data);
-    qemu_coroutine_yield();
-}
-
 static ssize_t socket_writev_buffer(void *opaque, struct iovec *iov, int iovcnt,
                                     int64_t pos)
 {
-- 
1.7.10.4

[Qemu-devel] [PATCH v6 03/11] rdma: export throughput w/ MigrationStats QMP

[mrhines]

From: "Michael R. Hines" <mrhines@us.ibm.com>

This exposes throughput (in megabits/sec) through QMP.

Signed-off-by: Michael R. Hines <mrhines@us.ibm.com>
---
 hmp.c                         |    2 ++
 include/migration/migration.h |    1 +
 migration.c                   |   10 ++++++++++
 qapi-schema.json              |    5 ++++-
 4 files changed, 17 insertions(+), 1 deletion(-)

diff --git a/hmp.c b/hmp.c
index 4fb76ec..ab2c030 100644
--- a/hmp.c
+++ b/hmp.c
@@ -167,6 +167,8 @@ void hmp_info_migrate(Monitor *mon, const QDict *qdict)
     if (info->has_ram) {
         monitor_printf(mon, "transferred ram: %" PRIu64 " kbytes\n",
                        info->ram->transferred >> 10);
+        monitor_printf(mon, "throughput: %0.2f mbps\n",
+                       info->ram->mbps);
         monitor_printf(mon, "remaining ram: %" PRIu64 " kbytes\n",
                        info->ram->remaining >> 10);
         monitor_printf(mon, "total ram: %" PRIu64 " kbytes\n",
diff --git a/include/migration/migration.h b/include/migration/migration.h
index e2acec6..6aa15e0 100644
--- a/include/migration/migration.h
+++ b/include/migration/migration.h
@@ -47,6 +47,7 @@ struct MigrationState
     int64_t dirty_bytes_rate;
     bool enabled_capabilities[MIGRATION_CAPABILITY_MAX];
     int64_t xbzrle_cache_size;
+    double mbps;
 };
 
 void process_incoming_migration(QEMUFile *f);
diff --git a/migration.c b/migration.c
index 3eb0fad..4614c95 100644
--- a/migration.c
+++ b/migration.c
@@ -66,6 +66,7 @@ MigrationState *migrate_get_current(void)
         .state = MIG_STATE_SETUP,
         .bandwidth_limit = MAX_THROTTLE,
         .xbzrle_cache_size = DEFAULT_MIGRATE_CACHE_SIZE,
+        .mbps = -1,
     };
 
     return &current_migration;
@@ -201,6 +202,7 @@ MigrationInfo *qmp_query_migrate(Error **errp)
         info->ram->normal = norm_mig_pages_transferred();
         info->ram->normal_bytes = norm_mig_bytes_transferred();
         info->ram->dirty_pages_rate = s->dirty_pages_rate;
+        info->ram->mbps = s->mbps;
 
         if (blk_mig_active()) {
             info->has_disk = true;
@@ -230,6 +232,7 @@ MigrationInfo *qmp_query_migrate(Error **errp)
         info->ram->skipped = skipped_mig_pages_transferred();
         info->ram->normal = norm_mig_pages_transferred();
         info->ram->normal_bytes = norm_mig_bytes_transferred();
+        info->ram->mbps = s->mbps;
         break;
     case MIG_STATE_ERROR:
         info->has_status = true;
@@ -545,6 +548,13 @@ static void *migration_thread(void *opaque)
             double bandwidth = transferred_bytes / time_spent;
             max_size = bandwidth * migrate_max_downtime() / 1000000;
 
+            if (time_spent) {
+                s->mbps = (((double) transferred_bytes * 8.0) /
+                    ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
+            } else {
+                s->mbps = -1.0;
+            }
+
             DPRINTF("transferred %" PRIu64 " time_spent %" PRIu64
                     " bandwidth %g max_size %" PRId64 "\n",
                     transferred_bytes, time_spent, bandwidth, max_size);
diff --git a/qapi-schema.json b/qapi-schema.json
index 751d3c2..cc846c3 100644
--- a/qapi-schema.json
+++ b/qapi-schema.json
@@ -507,12 +507,15 @@
 # @dirty-pages-rate: number of pages dirtied by second by the
 #        guest (since 1.3)
 #
+# @mbps: throughput in megabits/sec. (since 1.5)
+#
 # Since: 0.14.0
 ##
 { 'type': 'MigrationStats',
   'data': {'transferred': 'int', 'remaining': 'int', 'total': 'int' ,
            'duplicate': 'int', 'skipped': 'int', 'normal': 'int',
-           'normal-bytes': 'int', 'dirty-pages-rate' : 'int' } }
+           'normal-bytes': 'int', 'dirty-pages-rate' : 'int',
+           'mbps' : 'number' } }
 
 ##
 # @XBZRLECacheStats
-- 
1.7.10.4

[Qemu-devel] [PATCH v6 04/11] rdma: introduce qemu_file_mode_is_not_valid()

[mrhines]

From: "Michael R. Hines" <mrhines@us.ibm.com>

QEMUFileRDMA also has read and write modes. This function is now
shared to reduce code duplication.

Signed-off-by: Michael R. Hines <mrhines@us.ibm.com>
---
 include/migration/qemu-file.h |    1 +
 savevm.c                      |   20 +++++++++++++-------
 2 files changed, 14 insertions(+), 7 deletions(-)

diff --git a/include/migration/qemu-file.h b/include/migration/qemu-file.h
index 7519464..7e097c1 100644
--- a/include/migration/qemu-file.h
+++ b/include/migration/qemu-file.h
@@ -80,6 +80,7 @@ void qemu_put_byte(QEMUFile *f, int v);
  * The buffer should be available till it is sent asynchronously.
  */
 void qemu_put_buffer_async(QEMUFile *f, const uint8_t *buf, int size);
+bool qemu_file_mode_is_not_valid(const char *mode);
 
 static inline void qemu_put_ubyte(QEMUFile *f, unsigned int v)
 {
diff --git a/savevm.c b/savevm.c
index 4dbfa16..28334e7 100644
--- a/savevm.c
+++ b/savevm.c
@@ -447,14 +447,23 @@ static const QEMUFileOps socket_write_ops = {
     .close =      socket_close
 };
 
-QEMUFile *qemu_fopen_socket(int fd, const char *mode)
+bool qemu_file_mode_is_not_valid(const char *mode)
 {
-    QEMUFileSocket *s = g_malloc0(sizeof(QEMUFileSocket));
-
     if (mode == NULL ||
         (mode[0] != 'r' && mode[0] != 'w') ||
         mode[1] != 'b' || mode[2] != 0) {
         fprintf(stderr, "qemu_fopen: Argument validity check failed\n");
+        return true;
+    }
+
+    return false;
+}
+
+QEMUFile *qemu_fopen_socket(int fd, const char *mode)
+{
+    QEMUFileSocket *s = g_malloc0(sizeof(QEMUFileSocket));
+
+    if (qemu_file_mode_is_not_valid(mode)) {
         return NULL;
     }
 
@@ -472,10 +481,7 @@ QEMUFile *qemu_fopen(const char *filename, const char *mode)
 {
     QEMUFileStdio *s;
 
-    if (mode == NULL ||
-	(mode[0] != 'r' && mode[0] != 'w') ||
-	mode[1] != 'b' || mode[2] != 0) {
-        fprintf(stderr, "qemu_fopen: Argument validity check failed\n");
+    if (qemu_file_mode_is_not_valid(mode)) {
         return NULL;
     }
 
-- 
1.7.10.4

[Qemu-devel] [PATCH v6 05/11] rdma: export qemu_fflush()

[mrhines]

From: "Michael R. Hines" <mrhines@us.ibm.com>

RDMA uses this to flush the control channel before sending its
own message to handle page registrations.

Signed-off-by: Michael R. Hines <mrhines@us.ibm.com>
---
 include/migration/qemu-file.h |    1 +
 savevm.c                      |    2 +-
 2 files changed, 2 insertions(+), 1 deletion(-)

diff --git a/include/migration/qemu-file.h b/include/migration/qemu-file.h
index 7e097c1..97d4277 100644
--- a/include/migration/qemu-file.h
+++ b/include/migration/qemu-file.h
@@ -111,6 +111,7 @@ void qemu_file_reset_rate_limit(QEMUFile *f);
 void qemu_file_set_rate_limit(QEMUFile *f, int64_t new_rate);
 int64_t qemu_file_get_rate_limit(QEMUFile *f);
 int qemu_file_get_error(QEMUFile *f);
+void qemu_fflush(QEMUFile *f);
 
 static inline void qemu_put_be64s(QEMUFile *f, const uint64_t *pv)
 {
diff --git a/savevm.c b/savevm.c
index 28334e7..7e6e741 100644
--- a/savevm.c
+++ b/savevm.c
@@ -586,7 +586,7 @@ static inline bool qemu_file_is_writable(QEMUFile *f)
  * If there is writev_buffer QEMUFileOps it uses it otherwise uses
  * put_buffer ops.
  */
-static void qemu_fflush(QEMUFile *f)
+void qemu_fflush(QEMUFile *f)
 {
     ssize_t ret = 0;
 
-- 
1.7.10.4

[Qemu-devel] [PATCH v6 06/11] rdma: introduce ram_handle_compressed()

[mrhines]

From: "Michael R. Hines" <mrhines@us.ibm.com>

This gives RDMA shared access to madvise() on the destination side
when an entire chunk is found to be zero.

Signed-off-by: Michael R. Hines <mrhines@us.ibm.com>
---
 arch_init.c                   |   24 ++++++++++++++++--------
 include/migration/migration.h |    2 ++
 2 files changed, 18 insertions(+), 8 deletions(-)

diff --git a/arch_init.c b/arch_init.c
index 92de1bd..e272852 100644
--- a/arch_init.c
+++ b/arch_init.c
@@ -770,6 +770,21 @@ static inline void *host_from_stream_offset(QEMUFile *f,
     return NULL;
 }
 
+/*
+ * If a page (or a whole RDMA chunk) has been
+ * determined to be zero, then zap it.
+ */
+void ram_handle_compressed(void *host, uint8_t ch, uint64_t size)
+{
+    memset(host, ch, TARGET_PAGE_SIZE);
+#ifndef _WIN32
+    if (ch == 0 && (!kvm_enabled() || kvm_has_sync_mmu()) &&
+                            getpagesize() <= TARGET_PAGE_SIZE) {
+        qemu_madvise(host, size, QEMU_MADV_DONTNEED);
+    }
+#endif
+}
+
 static int ram_load(QEMUFile *f, void *opaque, int version_id)
 {
     ram_addr_t addr;
@@ -837,14 +852,7 @@ static int ram_load(QEMUFile *f, void *opaque, int version_id)
             }
 
             ch = qemu_get_byte(f);
-            memset(host, ch, TARGET_PAGE_SIZE);
-#ifndef _WIN32
-            if (ch == 0 &&
-                (!kvm_enabled() || kvm_has_sync_mmu()) &&
-                getpagesize() <= TARGET_PAGE_SIZE) {
-                qemu_madvise(host, TARGET_PAGE_SIZE, QEMU_MADV_DONTNEED);
-            }
-#endif
+            ram_handle_compressed(host, ch, TARGET_PAGE_SIZE);
         } else if (flags & RAM_SAVE_FLAG_PAGE) {
             void *host;
 
diff --git a/include/migration/migration.h b/include/migration/migration.h
index 6aa15e0..2a5aeef 100644
--- a/include/migration/migration.h
+++ b/include/migration/migration.h
@@ -106,6 +106,8 @@ uint64_t xbzrle_mig_pages_transferred(void);
 uint64_t xbzrle_mig_pages_overflow(void);
 uint64_t xbzrle_mig_pages_cache_miss(void);
 
+void ram_handle_compressed(void *host, uint8_t ch, uint64_t size);
+
 /**
  * @migrate_add_blocker - prevent migration from proceeding
  *
-- 
1.7.10.4

[Qemu-devel] [PATCH v6 07/11] rdma: introduce qemu_ram_foreach_block()

[mrhines]

From: "Michael R. Hines" <mrhines@us.ibm.com>

This is used during RDMA initialization in order to
transmit a description of all the RAM blocks to the
peer for later dynamic chunk registration purposes.

Signed-off-by: Michael R. Hines <mrhines@us.ibm.com>
---
 exec.c                    |    9 +++++++++
 include/exec/cpu-common.h |    5 +++++
 2 files changed, 14 insertions(+)

diff --git a/exec.c b/exec.c
index fa1e0c3..0e5a2c3 100644
--- a/exec.c
+++ b/exec.c
@@ -2631,3 +2631,12 @@ bool cpu_physical_memory_is_io(hwaddr phys_addr)
              memory_region_is_romd(section->mr));
 }
 #endif
+
+void qemu_ram_foreach_block(RAMBlockIterFunc func, void *opaque)
+{
+    RAMBlock *block;
+
+    QTAILQ_FOREACH(block, &ram_list.blocks, next) {
+        func(block->host, block->offset, block->length, opaque);
+    }
+}
diff --git a/include/exec/cpu-common.h b/include/exec/cpu-common.h
index 2e5f11f..2900fd6 100644
--- a/include/exec/cpu-common.h
+++ b/include/exec/cpu-common.h
@@ -119,6 +119,11 @@ extern struct MemoryRegion io_mem_rom;
 extern struct MemoryRegion io_mem_unassigned;
 extern struct MemoryRegion io_mem_notdirty;
 
+typedef void (RAMBlockIterFunc)(void *host_addr,
+    ram_addr_t offset, ram_addr_t length, void *opaque);
+
+void qemu_ram_foreach_block(RAMBlockIterFunc func, void *opaque);
+
 #endif
 
 #endif /* !CPU_COMMON_H */
-- 
1.7.10.4

[Qemu-devel] [PATCH v6 08/11] rdma: new QEMUFileOps hooks

[mrhines]

From: "Michael R. Hines" <mrhines@us.ibm.com>

These are the prototypes and implementation of new hooks that
RDMA takes advantage of to perform dynamic page registration.

An optional hook is also introduced for a custom function
to be able to override the default save_page function.

Also included are the prototypes and accessor methods used by
arch_init.c which invoke funtions inside savevm.c to call out
to the hooks that may or may not have been overridden
inside of QEMUFileOps.

Signed-off-by: Michael R. Hines <mrhines@us.ibm.com>
---
 include/migration/migration.h |   16 ++++++++++++
 include/migration/qemu-file.h |   28 ++++++++++++++++++++
 savevm.c                      |   57 +++++++++++++++++++++++++++++++++++++++++
 3 files changed, 101 insertions(+)

diff --git a/include/migration/migration.h b/include/migration/migration.h
index 2a5aeef..dd86b1c 100644
--- a/include/migration/migration.h
+++ b/include/migration/migration.h
@@ -21,6 +21,7 @@
 #include "qapi/error.h"
 #include "migration/vmstate.h"
 #include "qapi-types.h"
+#include "exec/cpu-common.h"
 
 struct MigrationParams {
     bool blk;
@@ -130,4 +131,19 @@ int migrate_use_xbzrle(void);
 int64_t migrate_xbzrle_cache_size(void);
 
 int64_t xbzrle_cache_resize(int64_t new_size);
+
+void ram_control_before_iterate(QEMUFile *f, uint64_t flags);
+void ram_control_after_iterate(QEMUFile *f, uint64_t flags);
+void ram_control_load_hook(QEMUFile *f, uint64_t flags);
+
+/* Whenever this is found in the data stream, the flags
+ * will be passed to ram_control_load_hook in the incoming-migration
+ * side. This lets before_ram_iterate/after_ram_iterate add
+ * transport-specific sections to the RAM migration data.
+ */
+#define RAM_SAVE_FLAG_HOOK     0x80
+
+size_t ram_control_save_page(QEMUFile *f, ram_addr_t block_offset,
+                             ram_addr_t offset, size_t size);
+
 #endif
diff --git a/include/migration/qemu-file.h b/include/migration/qemu-file.h
index 97d4277..3c13f53 100644
--- a/include/migration/qemu-file.h
+++ b/include/migration/qemu-file.h
@@ -23,6 +23,7 @@
  */
 #ifndef QEMU_FILE_H
 #define QEMU_FILE_H 1
+#include "exec/cpu-common.h"
 
 /* This function writes a chunk of data to a file at the given position.
  * The pos argument can be ignored if the file is only being used for
@@ -57,12 +58,39 @@ typedef int (QEMUFileGetFD)(void *opaque);
 typedef ssize_t (QEMUFileWritevBufferFunc)(void *opaque, struct iovec *iov,
                                            int iovcnt, int64_t pos);
 
+/*
+ * This function provides hooks around different
+ * stages of RAM migration.
+ */
+typedef int (QEMURamHookFunc)(QEMUFile *f, void *opaque, uint64_t flags);
+
+/*
+ * Constants used by ram_control_* hooks
+ */
+#define RAM_CONTROL_SETUP    0
+#define RAM_CONTROL_ROUND    1
+#define RAM_CONTROL_HOOK     2
+#define RAM_CONTROL_FINISH   3
+
+/*
+ * This function allows override of where the RAM page
+ * is saved (such as RDMA, for example.)
+ */
+typedef size_t (QEMURamSaveFunc)(QEMUFile *f, void *opaque,
+                               ram_addr_t block_offset,
+                               ram_addr_t offset,
+                               size_t size);
+
 typedef struct QEMUFileOps {
     QEMUFilePutBufferFunc *put_buffer;
     QEMUFileGetBufferFunc *get_buffer;
     QEMUFileCloseFunc *close;
     QEMUFileGetFD *get_fd;
     QEMUFileWritevBufferFunc *writev_buffer;
+    QEMURamHookFunc *before_ram_iterate;
+    QEMURamHookFunc *after_ram_iterate;
+    QEMURamHookFunc *hook_ram_load;
+    QEMURamSaveFunc *save_page;
 } QEMUFileOps;
 
 QEMUFile *qemu_fopen_ops(void *opaque, const QEMUFileOps *ops);
diff --git a/savevm.c b/savevm.c
index 7e6e741..3a5dab8 100644
--- a/savevm.c
+++ b/savevm.c
@@ -613,6 +613,63 @@ void qemu_fflush(QEMUFile *f)
     }
 }
 
+void ram_control_before_iterate(QEMUFile *f, uint64_t flags)
+{
+    int ret = 0;
+
+    if (f->ops->before_ram_iterate) {
+        ret = f->ops->before_ram_iterate(f, f->opaque, flags);
+        if (ret < 0) {
+            qemu_file_set_error(f, ret);
+        }
+    }
+}
+
+void ram_control_after_iterate(QEMUFile *f, uint64_t flags)
+{
+    int ret = 0;
+
+    if (f->ops->after_ram_iterate) {
+        ret = f->ops->after_ram_iterate(f, f->opaque, flags);
+        if (ret < 0) {
+            qemu_file_set_error(f, ret);
+        }
+    }
+}
+
+void ram_control_load_hook(QEMUFile *f, uint64_t flags)
+{
+    int ret = 0;
+
+    if (f->ops->hook_ram_load) {
+        ret = f->ops->hook_ram_load(f, f->opaque, flags);
+        if (ret < 0) {
+            qemu_file_set_error(f, ret);
+        }
+    } else {
+        qemu_file_set_error(f, ret);
+    }
+}
+
+size_t ram_control_save_page(QEMUFile *f, ram_addr_t block_offset,
+                             ram_addr_t offset, size_t size)
+{
+    if (f->ops->save_page) {
+        int64_t bytes;
+        bytes = f->ops->save_page(f, f->opaque, block_offset, offset, size);
+
+        if (bytes >= 0) {
+            f->pos += bytes;
+        } else {
+            qemu_file_set_error(f, bytes);
+        }
+
+        return bytes;
+    }
+
+    return -1;
+}
+
 static void qemu_fill_buffer(QEMUFile *f)
 {
     int len;
-- 
1.7.10.4

[Qemu-devel] [PATCH v6 09/11] rdma: introduce capability x-rdma-pin-all

[mrhines]

From: "Michael R. Hines" <mrhines@us.ibm.com>

This capability allows you to disable dynamic chunk registration
for better throughput on high-performance links.

For example, using an 8GB RAM virtual machine with all 8GB of memory in
active use and the VM itself is completely idle using a 40 gbps infiniband link:

1. x-rdma-pin-all disabled total time: approximately 7.5 seconds @ 9.5 Gbps
2. x-rdma-pin-all enabled total time: approximately 4 seconds @ 26 Gbps

These numbers would of course scale up to whatever size virtual machine
you have to migrate using RDMA.

Enabling this feature does *not* have any measurable affect on
migration *downtime*. This is because, without this feature, all of the
memory will have already been registered already in advance during
the bulk round and does not need to be re-registered during the successive
iteration rounds.

Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Michael R. Hines <mrhines@us.ibm.com>
---
 include/migration/migration.h |    2 ++
 migration.c                   |    9 +++++++++
 qapi-schema.json              |    7 ++++++-
 3 files changed, 17 insertions(+), 1 deletion(-)

diff --git a/include/migration/migration.h b/include/migration/migration.h
index dd86b1c..b9b6c36 100644
--- a/include/migration/migration.h
+++ b/include/migration/migration.h
@@ -123,6 +123,8 @@ void migrate_add_blocker(Error *reason);
  */
 void migrate_del_blocker(Error *reason);
 
+bool migrate_rdma_pin_all(void);
+
 int xbzrle_encode_buffer(uint8_t *old_buf, uint8_t *new_buf, int slen,
                          uint8_t *dst, int dlen);
 int xbzrle_decode_buffer(uint8_t *src, int slen, uint8_t *dst, int dlen);
diff --git a/migration.c b/migration.c
index 4614c95..97ce6d8 100644
--- a/migration.c
+++ b/migration.c
@@ -477,6 +477,15 @@ void qmp_migrate_set_downtime(double value, Error **errp)
     max_downtime = (uint64_t)value;
 }
 
+bool migrate_rdma_pin_all(void)
+{
+    MigrationState *s;
+
+    s = migrate_get_current();
+
+    return s->enabled_capabilities[MIGRATION_CAPABILITY_X_RDMA_PIN_ALL];
+}
+
 int migrate_use_xbzrle(void)
 {
     MigrationState *s;
diff --git a/qapi-schema.json b/qapi-schema.json
index cc846c3..d5b8e20 100644
--- a/qapi-schema.json
+++ b/qapi-schema.json
@@ -602,10 +602,15 @@
 #          This feature allows us to minimize migration traffic for certain work
 #          loads, by sending compressed difference of the pages
 #
+# @x-rdma-pin-all: Controls whether or not the entire VM memory footprint is
+#          mlock()'d on demand or all at once. Refer to docs/rdma.txt for usage.
+#          Disabled by default. Experimental: may (or may not) be renamed after
+#          further testing is complete. (since 1.5)
+#
 # Since: 1.2
 ##
 { 'enum': 'MigrationCapability',
-  'data': ['xbzrle'] }
+  'data': ['xbzrle', 'x-rdma-pin-all'] }
 
 ##
 # @MigrationCapabilityStatus
-- 
1.7.10.4

[Qemu-devel] [PATCH v6 10/11] rdma: core logic

[mrhines]

From: "Michael R. Hines" <mrhines@us.ibm.com>

Code that does need to be visible is kept
well contained inside this file and this is the only
new additional file to the entire patch - good progress.

This file includes the entire protocol and interfaces
required to perform RDMA migration.

Also, the configure and Makefile modifications to link
this file are included.

Full documentation is in docs/rdma.txt

Signed-off-by: Michael R. Hines <mrhines@us.ibm.com>
---
 Makefile.objs                 |    1 +
 configure                     |   29 +++++++++++++++++++++++++++++
 include/migration/migration.h |    4 ++++
 migration.c                   |    8 ++++++++
 migration-rdma.c              | 2707 +++++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 2749 insertions(+)

diff --git a/Makefile.objs b/Makefile.objs
index a473348..d744827 100644
--- a/Makefile.objs
+++ b/Makefile.objs
@@ -49,6 +49,7 @@ common-obj-$(CONFIG_POSIX) += os-posix.o
 common-obj-$(CONFIG_LINUX) += fsdev/
 
 common-obj-y += migration.o migration-tcp.o
+common-obj-$(CONFIG_RDMA) += migration-rdma.o
 common-obj-y += qemu-char.o #aio.o
 common-obj-y += block-migration.o
 common-obj-y += page_cache.o xbzrle.o
diff --git a/configure b/configure
index 33d3354..90441cf 100755
--- a/configure
+++ b/configure
@@ -181,6 +181,7 @@ xfs=""
 vhost_net="no"
 vhost_scsi="no"
 kvm="no"
+rdma="yes"
 gprof="no"
 debug_tcg="no"
 debug="no"
@@ -926,6 +927,10 @@ for opt do
   ;;
   --enable-gtk) gtk="yes"
   ;;
+  --enable-rdma) rdma="yes"
+  ;;
+  --disable-rdma) rdma="no"
+  ;;
   --with-gtkabi=*) gtkabi="$optarg"
   ;;
   --enable-tpm) tpm="yes"
@@ -1137,6 +1142,8 @@ echo "  --enable-bluez           enable bluez stack connectivity"
 echo "  --disable-slirp          disable SLIRP userspace network connectivity"
 echo "  --disable-kvm            disable KVM acceleration support"
 echo "  --enable-kvm             enable KVM acceleration support"
+echo "  --disable-rdma           disable RDMA-based migration support"
+echo "  --enable-rdma            enable RDMA-based migration support"
 echo "  --enable-tcg-interpreter enable TCG with bytecode interpreter (TCI)"
 echo "  --disable-nptl           disable usermode NPTL support"
 echo "  --enable-nptl            enable usermode NPTL support"
@@ -1811,6 +1818,23 @@ EOF
   libs_softmmu="$sdl_libs $libs_softmmu"
 fi
 
+if test "$rdma" != "no" ; then
+  cat > $TMPC <<EOF
+#include <rdma/rdma_cma.h>
+int main(void) { return 0; }
+EOF
+  rdma_libs="-lrdmacm -libverbs"
+  if compile_prog "-Werror" "$rdma_libs" ; then
+    rdma="yes"
+    libs_softmmu="$libs_softmmu $rdma_libs"
+  else
+    if test "$rdma" = "yes" ; then
+      feature_not_found "rdma"
+    fi
+    rdma="no"
+  fi
+fi
+
 ##########################################
 # VNC TLS/WS detection
 if test "$vnc" = "yes" -a \( "$vnc_tls" != "no" -o "$vnc_ws" != "no" \) ; then
@@ -3549,6 +3573,7 @@ echo "Linux AIO support $linux_aio"
 echo "ATTR/XATTR support $attr"
 echo "Install blobs     $blobs"
 echo "KVM support       $kvm"
+echo "RDMA support      $rdma"
 echo "TCG interpreter   $tcg_interpreter"
 echo "fdt support       $fdt"
 echo "preadv support    $preadv"
@@ -4515,6 +4540,10 @@ if [ "$pixman" = "internal" ]; then
   echo "config-host.h: subdir-pixman" >> $config_host_mak
 fi
 
+if test "$rdma" = "yes" ; then
+echo "CONFIG_RDMA=y" >> $config_host_mak
+fi
+
 if [ "$dtc_internal" = "yes" ]; then
   echo "config-host.h: subdir-dtc" >> $config_host_mak
 fi
diff --git a/include/migration/migration.h b/include/migration/migration.h
index b9b6c36..5acdc69 100644
--- a/include/migration/migration.h
+++ b/include/migration/migration.h
@@ -77,6 +77,10 @@ void fd_start_incoming_migration(const char *path, Error **errp);
 
 void fd_start_outgoing_migration(MigrationState *s, const char *fdname, Error **errp);
 
+void rdma_start_outgoing_migration(void *opaque, const char *host_port, Error **errp);
+
+void rdma_start_incoming_migration(const char *host_port, Error **errp);
+
 void migrate_fd_error(MigrationState *s);
 
 void migrate_fd_connect(MigrationState *s);
diff --git a/migration.c b/migration.c
index 97ce6d8..42e43eb 100644
--- a/migration.c
+++ b/migration.c
@@ -78,6 +78,10 @@ void qemu_start_incoming_migration(const char *uri, Error **errp)
 
     if (strstart(uri, "tcp:", &p))
         tcp_start_incoming_migration(p, errp);
+#ifdef CONFIG_RDMA
+    else if (strstart(uri, "x-rdma:", &p))
+        rdma_start_incoming_migration(p, errp);
+#endif
 #if !defined(WIN32)
     else if (strstart(uri, "exec:", &p))
         exec_start_incoming_migration(p, errp);
@@ -408,6 +412,10 @@ void qmp_migrate(const char *uri, bool has_blk, bool blk,
 
     if (strstart(uri, "tcp:", &p)) {
         tcp_start_outgoing_migration(s, p, &local_err);
+#ifdef CONFIG_RDMA
+    } else if (strstart(uri, "x-rdma:", &p)) {
+        rdma_start_outgoing_migration(s, p, &local_err);
+#endif
 #if !defined(WIN32)
     } else if (strstart(uri, "exec:", &p)) {
         exec_start_outgoing_migration(s, p, &local_err);
diff --git a/migration-rdma.c b/migration-rdma.c
new file mode 100644
index 0000000..8924e94
--- /dev/null
+++ b/migration-rdma.c
@@ -0,0 +1,2707 @@
+/*
+ * RDMA protocol and interfaces
+ *
+ * Copyright IBM, Corp. 2010-2013
+ *
+ * Authors:
+ *  Michael R. Hines <mrhines@us.ibm.com>
+ *  Jiuxing Liu <jl@us.ibm.com>
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2 or
+ * later.  See the COPYING file in the top-level directory.
+ *
+ */
+#include "qemu-common.h"
+#include "migration/migration.h"
+#include "migration/qemu-file.h"
+#include "exec/cpu-common.h"
+#include "qemu/main-loop.h"
+#include "qemu/sockets.h"
+#include "block/coroutine.h"
+#include <stdio.h>
+#include <sys/types.h>
+#include <sys/socket.h>
+#include <netdb.h>
+#include <arpa/inet.h>
+#include <string.h>
+#include <rdma/rdma_cma.h>
+
+#define DEBUG_RDMA
+//#define DEBUG_RDMA_VERBOSE
+//#define DEBUG_RDMA_REALLY_VERBOSE
+
+#ifdef DEBUG_RDMA
+#define DPRINTF(fmt, ...) \
+    do { printf("rdma: " fmt, ## __VA_ARGS__); } while (0)
+#else
+#define DPRINTF(fmt, ...) \
+    do { } while (0)
+#endif
+
+#ifdef DEBUG_RDMA_VERBOSE
+#define DDPRINTF(fmt, ...) \
+    do { printf("rdma: " fmt, ## __VA_ARGS__); } while (0)
+#else
+#define DDPRINTF(fmt, ...) \
+    do { } while (0)
+#endif
+
+#ifdef DEBUG_RDMA_REALLY_VERBOSE
+#define DDDPRINTF(fmt, ...) \
+    do { printf("rdma: " fmt, ## __VA_ARGS__); } while (0)
+#else
+#define DDDPRINTF(fmt, ...) \
+    do { } while (0)
+#endif
+
+/*
+ * Print and error on both the Monitor and the Log file.
+ */
+#define ERROR(errp, fmt, ...) \
+    do { \
+        fprintf(stderr, "RDMA ERROR: " fmt, ## __VA_ARGS__); \
+        if (errp && (*(errp) == NULL)) { \
+            error_setg(errp, "RDMA ERROR: " fmt, ## __VA_ARGS__); \
+        } \
+    } while (0)
+
+#define RDMA_RESOLVE_TIMEOUT_MS 10000
+
+/* Do not merge data if larger than this. */
+#define RDMA_MERGE_MAX (4 * 1024 * 1024)
+#define RDMA_UNSIGNALED_SEND_MAX 64
+
+#define RDMA_REG_CHUNK_SHIFT 20 /* 1 MB */
+
+#define RDMA_REG_CHUNK_SIZE (1UL << (RDMA_REG_CHUNK_SHIFT))
+
+/*
+ * This is only for non-live state being migrated.
+ * Instead of RDMA_WRITE messages, we use RDMA_SEND
+ * messages for that state, which requires a different
+ * delivery design than main memory.
+ */
+#define RDMA_SEND_INCREMENT 32768
+
+/*
+ * Completion queue can be filled by both read and write work requests,
+ * so must reflect the sum of both possible queue sizes.
+ */
+#define RDMA_QP_SIZE 1000
+#define RDMA_CQ_SIZE (RDMA_QP_SIZE * 3)
+
+/*
+ * Maximum size infiniband SEND message
+ */
+#define RDMA_CONTROL_MAX_BUFFER (512 * 1024)
+#define RDMA_CONTROL_MAX_WR 2
+#define RDMA_CONTROL_MAX_COMMANDS_PER_MESSAGE 4096
+
+/*
+ * Capabilities for negotiation.
+ */
+#define RDMA_CAPABILITY_PIN_ALL 0x01
+
+/*
+ * Add the other flags above to this list of known capabilities
+ * as they are introduced.
+ */
+static uint32_t known_capabilities = RDMA_CAPABILITY_PIN_ALL;
+
+#define CHECK_ERROR_STATE() \
+    do { \
+        if (rdma->error_state) { \
+            if (!rdma->error_reported) { \
+                fprintf(stderr, "RDMA is in an error state waiting migration" \
+                                " to abort!\n"); \
+                rdma->error_reported = 1; \
+            } \
+            return rdma->error_state; \
+        } \
+    } while (0);
+/*
+ * RDMA migration protocol:
+ * 1. RDMA Writes (data messages, i.e. RAM)
+ * 2. IB Send/Recv (control channel messages)
+ */
+enum {
+    RDMA_WRID_NONE = 0,
+    RDMA_WRID_RDMA_WRITE,
+    RDMA_WRID_SEND_CONTROL = 1000,
+    RDMA_WRID_RECV_CONTROL = 2000,
+};
+
+const char *wrid_desc[] = {
+        [RDMA_WRID_NONE] = "NONE",
+        [RDMA_WRID_RDMA_WRITE] = "WRITE RDMA",
+        [RDMA_WRID_SEND_CONTROL] = "CONTROL SEND",
+        [RDMA_WRID_RECV_CONTROL] = "CONTROL RECV",
+};
+
+/*
+ * SEND/RECV IB Control Messages.
+ */
+enum {
+    RDMA_CONTROL_NONE = 0,
+    RDMA_CONTROL_ERROR,
+    RDMA_CONTROL_READY,             /* ready to receive */
+    RDMA_CONTROL_QEMU_FILE,         /* QEMUFile-transmitted bytes */
+    RDMA_CONTROL_RAM_BLOCKS,        /* RAMBlock synchronization */
+    RDMA_CONTROL_COMPRESS,          /* page contains repeat values */
+    RDMA_CONTROL_REGISTER_REQUEST,  /* dynamic page registration */
+    RDMA_CONTROL_REGISTER_RESULT,   /* key to use after registration */
+    RDMA_CONTROL_REGISTER_FINISHED, /* current iteration finished */
+};
+
+const char *control_desc[] = {
+        [RDMA_CONTROL_NONE] = "NONE",
+        [RDMA_CONTROL_ERROR] = "ERROR",
+        [RDMA_CONTROL_READY] = "READY",
+        [RDMA_CONTROL_QEMU_FILE] = "QEMU FILE",
+        [RDMA_CONTROL_RAM_BLOCKS] = "REMOTE INFO",
+        [RDMA_CONTROL_COMPRESS] = "COMPRESS",
+        [RDMA_CONTROL_REGISTER_REQUEST] = "REGISTER REQUEST",
+        [RDMA_CONTROL_REGISTER_RESULT] = "REGISTER RESULT",
+        [RDMA_CONTROL_REGISTER_FINISHED] = "REGISTER FINISHED",
+};
+
+/*
+ * Memory and MR structures used to represent an IB Send/Recv work request.
+ * This is *not* used for RDMA, only IB Send/Recv.
+ */
+typedef struct {
+    uint8_t  control[RDMA_CONTROL_MAX_BUFFER]; /* actual buffer to register */
+    struct   ibv_mr *control_mr;               /* registration metadata */
+    size_t   control_len;                      /* length of the message */
+    uint8_t *control_curr;                     /* start of unconsumed bytes */
+} RDMAWorkRequestData;
+
+/*
+ * Negotiate RDMA capabilities during connection-setup time.
+ */
+typedef struct {
+    uint32_t version;
+    uint32_t flags;
+} RDMACapabilities;
+
+static void caps_to_network(RDMACapabilities *cap)
+{
+    cap->version = htonl(cap->version);
+    cap->flags = htonl(cap->flags);
+}
+
+static void network_to_caps(RDMACapabilities *cap)
+{
+    cap->version = ntohl(cap->version);
+    cap->flags = ntohl(cap->flags);
+}
+
+/*
+ * Representation of a RAMBlock from an RDMA perspective.
+ * This is not transmitted, only local.
+ * This and subsequent structures cannot be linked lists
+ * because we're using a single IB message to transmit
+ * the information. It's small anyway, so a list is overkill.
+ */
+typedef struct RDMALocalBlock {
+    uint8_t  *local_host_addr; /* local virtual address */
+    uint64_t remote_host_addr; /* remote virtual address */
+    uint64_t offset;
+    uint64_t length;
+    struct   ibv_mr **pmr;     /* MRs for chunk-level registration */
+    struct   ibv_mr *mr;       /* MR for non-chunk-level registration */
+    uint32_t *remote_keys;     /* rkeys for chunk-level registration */
+    uint32_t remote_rkey;      /* rkeys for non-chunk-level registration */
+    int      index;            /* which block are we */
+} RDMALocalBlock;
+
+/*
+ * Also represents a RAMblock, but only on the dest.
+ * This gets transmitted by the dest during connection-time
+ * to the source / primary VM and then is used to populate the
+ * corresponding RDMALocalBlock with
+ * the information needed to perform the actual RDMA.
+ */
+typedef struct QEMU_PACKED RDMARemoteBlock {
+    uint64_t remote_host_addr;
+    uint64_t offset;
+    uint64_t length;
+    uint32_t remote_rkey;
+    uint32_t padding;
+} QEMU_PACKED RDMARemoteBlock;
+
+/*
+ * Virtual address of the above structures used for transmitting
+ * the RAMBlock descriptions at connection-time.
+ * This structure is *not* transmitted.
+ */
+typedef struct RDMALocalBlocks {
+    int num_blocks;
+    RDMALocalBlock *block;
+} RDMALocalBlocks;
+
+/*
+ * Main data structure for RDMA state.
+ * While there is only one copy of this structure being allocated right now,
+ * this is the place where one would start if you wanted to consider
+ * having more than one RDMA connection open at the same time.
+ */
+typedef struct RDMAContext {
+    char *host;
+    int port;
+
+    /* This is used by the migration protocol to transmit
+     * control messages (such as device state and registration commands)
+     *
+     * WR #0 is for control channel ready messages from the destination.
+     * WR #1 is for control channel data messages from the destination.
+     * WR #2 is for control channel send messages.
+     *
+     * We could use more WRs, but we have enough for now.
+     */
+    RDMAWorkRequestData wr_data[RDMA_CONTROL_MAX_WR + 1];
+
+    /*
+     * This is used by *_exchange_send() to figure out whether or not
+     * the initial "READY" message has already been received or not.
+     * This is because other functions may potentially poll() and detect
+     * the READY message before send() does, in which case we need to
+     * know if it completed.
+     */
+    int control_ready_expected;
+
+    /* number of outstanding unsignaled send */
+    int num_unsignaled_send;
+
+    /* number of outstanding signaled send */
+    int num_signaled_send;
+
+    /* store info about current buffer so that we can
+       merge it with future sends */
+    uint64_t current_offset;
+    uint64_t current_length;
+    /* index of ram block the current buffer belongs to */
+    int current_index;
+    /* index of the chunk in the current ram block */
+    int current_chunk;
+
+    bool pin_all;
+
+    /*
+     * infiniband-specific variables for opening the device
+     * and maintaining connection state and so forth.
+     *
+     * cm_id also has ibv_context, rdma_event_channel, and ibv_qp in
+     * cm_id->verbs, cm_id->channel, and cm_id->qp.
+     */
+    struct rdma_cm_id *cm_id;               /* connection manager ID */
+    struct rdma_cm_id *listen_id;
+
+    struct ibv_context *verbs;
+    struct rdma_event_channel *channel;
+    struct ibv_qp *qp;                      /* queue pair */
+    struct ibv_comp_channel *comp_channel;  /* completion channel */
+    struct ibv_pd *pd;                      /* protection domain */
+    struct ibv_cq *cq;                      /* completion queue */
+
+    /*
+     * If a previous write failed (perhaps because of a failed
+     * memory registration, then do not attempt any future work
+     * and remember the error state.
+     */
+    int error_state;
+    int error_reported;
+
+    /*
+     * Description of ram blocks used throughout the code.
+     */
+    RDMALocalBlocks local_ram_blocks;
+    RDMARemoteBlock *block;
+
+    /*
+     * Migration on *destination* started.
+     * Then use coroutine yield function.
+     * Source runs in a thread, so we don't care.
+     */
+    int migration_started_on_destination;
+
+    int total_registrations;
+} RDMAContext;
+
+/*
+ * Interface to the rest of the migration call stack.
+ */
+typedef struct QEMUFileRDMA {
+    RDMAContext *rdma;
+    size_t len;
+    void *file;
+} QEMUFileRDMA;
+
+#define RDMA_CONTROL_VERSION_CURRENT 1
+
+/*
+ * Main structure for IB Send/Recv control messages.
+ * This gets prepended at the beginning of every Send/Recv.
+ */
+typedef struct QEMU_PACKED {
+    uint32_t len;     /* Total length of data portion */
+    uint32_t type;    /* which control command to perform */
+    uint32_t repeat;  /* number of commands in data portion of same type */
+    uint32_t padding;
+} QEMU_PACKED RDMAControlHeader;
+
+static void control_to_network(RDMAControlHeader *control)
+{
+    control->type = htonl(control->type);
+    control->len = htonl(control->len);
+    control->repeat = htonl(control->repeat);
+}
+
+static void network_to_control(RDMAControlHeader *control)
+{
+    control->type = ntohl(control->type);
+    control->len = ntohl(control->len);
+    control->repeat = ntohl(control->repeat);
+}
+
+/*
+ * Register a single Chunk.
+ * Information sent by the primary VM to inform the dest
+ * to register an single chunk of memory before we can perform
+ * the actual RDMA operation.
+ */
+typedef struct QEMU_PACKED {
+    uint32_t len;           /* length of the chunk to be registered */
+    uint32_t current_index; /* which ramblock the chunk belongs to */
+    uint64_t offset;        /* offset into the ramblock of the chunk */
+} QEMU_PACKED RDMARegister;
+
+typedef struct QEMU_PACKED {
+    uint32_t value;     /* if zero, we will madvise() */
+    uint32_t block_idx; /* which ram block index */
+    uint64_t offset;    /* where in the remote ramblock this chunk */
+    uint64_t length;    /* length of the chunk */
+} QEMU_PACKED RDMACompress;
+
+/*
+ * The result of the dest's memory registration produces an "rkey"
+ * which the primary VM must reference in order to perform
+ * the RDMA operation.
+ */
+typedef struct QEMU_PACKED {
+    uint32_t rkey;
+    uint32_t padding;
+} QEMU_PACKED RDMARegisterResult;
+
+static inline uint64_t ram_chunk_index(uint8_t *start, uint8_t *host)
+{
+    return ((uintptr_t) host - (uintptr_t) start) >> RDMA_REG_CHUNK_SHIFT;
+}
+
+static inline uint64_t ram_chunk_count(RDMALocalBlock *rdma_ram_block)
+{
+    return ram_chunk_index(rdma_ram_block->local_host_addr,
+        rdma_ram_block->local_host_addr + rdma_ram_block->length) + 1UL;
+}
+
+static inline uint8_t *ram_chunk_start(RDMALocalBlock *rdma_ram_block,
+                                       uint64_t i)
+{
+    return (uint8_t *) (((uintptr_t) rdma_ram_block->local_host_addr)
+                                    + (i << RDMA_REG_CHUNK_SHIFT));
+}
+
+static inline uint8_t *ram_chunk_end(RDMALocalBlock *rdma_ram_block, uint64_t i)
+{
+    uint8_t *result = ram_chunk_start(rdma_ram_block, i) + RDMA_REG_CHUNK_SIZE;
+
+    if (result > (rdma_ram_block->local_host_addr + rdma_ram_block->length)) {
+        result = rdma_ram_block->local_host_addr + rdma_ram_block->length;
+    }
+
+    return result;
+}
+
+
+/*
+ * Memory regions need to be registered with the device and queue pairs setup
+ * in advanced before the migration starts. This tells us where the RAM blocks
+ * are so that we can register them individually.
+ */
+static void qemu_rdma_init_one_block(void *host_addr,
+    ram_addr_t offset, ram_addr_t length, void *opaque)
+{
+    RDMALocalBlocks *rdma_local_ram_blocks = opaque;
+    int num_blocks = rdma_local_ram_blocks->num_blocks;
+
+    rdma_local_ram_blocks->block[num_blocks].local_host_addr = host_addr;
+    rdma_local_ram_blocks->block[num_blocks].offset = (uint64_t)offset;
+    rdma_local_ram_blocks->block[num_blocks].length = (uint64_t)length;
+    rdma_local_ram_blocks->block[num_blocks].index = num_blocks;
+
+    DPRINTF("Block: %d, addr: %" PRIu64 ", offset: %" PRIu64
+           " length: %" PRIu64 " end: %" PRIu64 "\n",
+            num_blocks, (uint64_t) host_addr, offset, length,
+            (uint64_t) (host_addr + length));
+
+    rdma_local_ram_blocks->num_blocks++;
+
+}
+
+static void qemu_rdma_ram_block_counter(void *host_addr,
+            ram_addr_t offset, ram_addr_t length, void *opaque)
+{
+    int *num_blocks = opaque;
+    *num_blocks = *num_blocks + 1;
+}
+
+/*
+ * Identify the RAMBlocks and their quantity. They will be references to
+ * identify chunk boundaries inside each RAMBlock and also be referenced
+ * during dynamic page registration.
+ */
+static int qemu_rdma_init_ram_blocks(RDMALocalBlocks *rdma_local_ram_blocks)
+{
+    int num_blocks = 0;
+
+    qemu_ram_foreach_block(qemu_rdma_ram_block_counter, &num_blocks);
+
+    memset(rdma_local_ram_blocks, 0, sizeof *rdma_local_ram_blocks);
+    rdma_local_ram_blocks->block = g_malloc0(sizeof(RDMALocalBlock) *
+                                    num_blocks);
+
+    rdma_local_ram_blocks->num_blocks = 0;
+    qemu_ram_foreach_block(qemu_rdma_init_one_block, rdma_local_ram_blocks);
+
+    DPRINTF("Allocated %d local ram block structures\n",
+                    rdma_local_ram_blocks->num_blocks);
+    return 0;
+}
+
+/*
+ * Put in the log file which RDMA device was opened and the details
+ * associated with that device.
+ */
+static void qemu_rdma_dump_id(const char *who, struct ibv_context *verbs)
+{
+    printf("%s RDMA Device opened: kernel name %s "
+           "uverbs device name %s, "
+           "infiniband_verbs class device path %s,"
+           " infiniband class device path %s\n",
+                who,
+                verbs->device->name,
+                verbs->device->dev_name,
+                verbs->device->dev_path,
+                verbs->device->ibdev_path);
+}
+
+/*
+ * Put in the log file the RDMA gid addressing information,
+ * useful for folks who have trouble understanding the
+ * RDMA device hierarchy in the kernel.
+ */
+static void qemu_rdma_dump_gid(const char *who, struct rdma_cm_id *id)
+{
+    char sgid[33];
+    char dgid[33];
+    inet_ntop(AF_INET6, &id->route.addr.addr.ibaddr.sgid, sgid, sizeof sgid);
+    inet_ntop(AF_INET6, &id->route.addr.addr.ibaddr.dgid, dgid, sizeof dgid);
+    DPRINTF("%s Source GID: %s, Dest GID: %s\n", who, sgid, dgid);
+}
+
+/*
+ * Figure out which RDMA device corresponds to the requested IP hostname
+ * Also create the initial connection manager identifiers for opening
+ * the connection.
+ */
+static int qemu_rdma_resolve_host(RDMAContext *rdma, Error **errp)
+{
+    int ret;
+    struct addrinfo *res;
+    char port_str[16];
+    struct rdma_cm_event *cm_event;
+    char ip[40] = "unknown";
+
+    if (rdma->host == NULL || !strcmp(rdma->host, "")) {
+        ERROR(errp, "RDMA hostname has not been set\n");
+        return -1;
+    }
+
+    /* create CM channel */
+    rdma->channel = rdma_create_event_channel();
+    if (!rdma->channel) {
+        ERROR(errp, "could not create CM channel\n");
+        return -1;
+    }
+
+    /* create CM id */
+    ret = rdma_create_id(rdma->channel, &rdma->cm_id, NULL, RDMA_PS_TCP);
+    if (ret) {
+        ERROR(errp, "could not create channel id\n");
+        goto err_resolve_create_id;
+    }
+
+    snprintf(port_str, 16, "%d", rdma->port);
+    port_str[15] = '\0';
+
+    ret = getaddrinfo(rdma->host, port_str, NULL, &res);
+    if (ret < 0) {
+        ERROR(errp, "could not getaddrinfo address %s\n", rdma->host);
+        goto err_resolve_get_addr;
+    }
+
+    inet_ntop(AF_INET, &((struct sockaddr_in *) res->ai_addr)->sin_addr,
+                                ip, sizeof ip);
+    DPRINTF("%s => %s\n", rdma->host, ip);
+
+    /* resolve the first address */
+    ret = rdma_resolve_addr(rdma->cm_id, NULL, res->ai_addr,
+            RDMA_RESOLVE_TIMEOUT_MS);
+    if (ret) {
+        ERROR(errp, "could not resolve address %s\n", rdma->host);
+        goto err_resolve_get_addr;
+    }
+
+    qemu_rdma_dump_gid("source_resolve_addr", rdma->cm_id);
+
+    ret = rdma_get_cm_event(rdma->channel, &cm_event);
+    if (ret) {
+        ERROR(errp, "could not perform event_addr_resolved\n");
+        goto err_resolve_get_addr;
+    }
+
+    if (cm_event->event != RDMA_CM_EVENT_ADDR_RESOLVED) {
+        ERROR(errp, "result not equal to event_addr_resolved %s\n",
+                rdma_event_str(cm_event->event));
+        perror("rdma_resolve_addr");
+        goto err_resolve_get_addr;
+    }
+    rdma_ack_cm_event(cm_event);
+
+    /* resolve route */
+    ret = rdma_resolve_route(rdma->cm_id, RDMA_RESOLVE_TIMEOUT_MS);
+    if (ret) {
+        ERROR(errp, "could not resolve rdma route\n");
+        goto err_resolve_get_addr;
+    }
+
+    ret = rdma_get_cm_event(rdma->channel, &cm_event);
+    if (ret) {
+        ERROR(errp, "could not perform event_route_resolved\n");
+        goto err_resolve_get_addr;
+    }
+    if (cm_event->event != RDMA_CM_EVENT_ROUTE_RESOLVED) {
+        ERROR(errp, "result not equal to event_route_resolved: %s\n",
+                        rdma_event_str(cm_event->event));
+        rdma_ack_cm_event(cm_event);
+        goto err_resolve_get_addr;
+    }
+    rdma_ack_cm_event(cm_event);
+    rdma->verbs = rdma->cm_id->verbs;
+    qemu_rdma_dump_id("source_resolve_host", rdma->cm_id->verbs);
+    qemu_rdma_dump_gid("source_resolve_host", rdma->cm_id);
+    return 0;
+
+err_resolve_get_addr:
+    rdma_destroy_id(rdma->cm_id);
+    rdma->cm_id = 0;
+err_resolve_create_id:
+    rdma_destroy_event_channel(rdma->channel);
+    rdma->channel = NULL;
+
+    return -1;
+}
+
+/*
+ * Create protection domain and completion queues
+ */
+static int qemu_rdma_alloc_pd_cq(RDMAContext *rdma)
+{
+    /* allocate pd */
+    rdma->pd = ibv_alloc_pd(rdma->verbs);
+    if (!rdma->pd) {
+        fprintf(stderr, "failed to allocate protection domain\n");
+        return -1;
+    }
+
+    /* create completion channel */
+    rdma->comp_channel = ibv_create_comp_channel(rdma->verbs);
+    if (!rdma->comp_channel) {
+        fprintf(stderr, "failed to allocate completion channel\n");
+        goto err_alloc_pd_cq;
+    }
+
+    /* create cq */
+    rdma->cq = ibv_create_cq(rdma->verbs, RDMA_CQ_SIZE,
+            NULL, rdma->comp_channel, 0);
+    if (!rdma->cq) {
+        fprintf(stderr, "failed to allocate completion queue\n");
+        goto err_alloc_pd_cq;
+    }
+
+    return 0;
+
+err_alloc_pd_cq:
+    if (rdma->pd) {
+        ibv_dealloc_pd(rdma->pd);
+    }
+    if (rdma->comp_channel) {
+        ibv_destroy_comp_channel(rdma->comp_channel);
+    }
+    rdma->pd = NULL;
+    rdma->comp_channel = NULL;
+    return -1;
+
+}
+
+/*
+ * Create queue pairs.
+ */
+static int qemu_rdma_alloc_qp(RDMAContext *rdma)
+{
+    struct ibv_qp_init_attr attr = { 0 };
+    int ret;
+
+    attr.cap.max_send_wr = RDMA_QP_SIZE;
+    attr.cap.max_recv_wr = 3;
+    attr.cap.max_send_sge = 1;
+    attr.cap.max_recv_sge = 1;
+    attr.send_cq = rdma->cq;
+    attr.recv_cq = rdma->cq;
+    attr.qp_type = IBV_QPT_RC;
+
+    ret = rdma_create_qp(rdma->cm_id, rdma->pd, &attr);
+    if (ret) {
+        return -1;
+    }
+
+    rdma->qp = rdma->cm_id->qp;
+    return 0;
+}
+
+static int qemu_rdma_reg_whole_ram_blocks(RDMAContext *rdma,
+                                RDMALocalBlocks *rdma_local_ram_blocks)
+{
+    int i;
+    for (i = 0; i < rdma_local_ram_blocks->num_blocks; i++) {
+        rdma_local_ram_blocks->block[i].mr =
+            ibv_reg_mr(rdma->pd,
+                    rdma_local_ram_blocks->block[i].local_host_addr,
+                    rdma_local_ram_blocks->block[i].length,
+                    IBV_ACCESS_LOCAL_WRITE |
+                    IBV_ACCESS_REMOTE_WRITE
+                    );
+        if (!rdma_local_ram_blocks->block[i].mr) {
+            perror("Failed to register local dest ram block!\n");
+            break;
+        }
+        rdma->total_registrations++;
+    }
+
+    if (i >= rdma_local_ram_blocks->num_blocks) {
+        return 0;
+    }
+
+    for (i--; i >= 0; i--) {
+        ibv_dereg_mr(rdma_local_ram_blocks->block[i].mr);
+        rdma->total_registrations--;
+    }
+
+    return -1;
+
+}
+
+/*
+ * Shutdown and clean things up.
+ */
+static void qemu_rdma_dereg_ram_blocks(RDMAContext *rdma,
+                                       RDMALocalBlocks *rdma_local_ram_blocks)
+{
+    int i, j;
+    for (i = 0; i < rdma_local_ram_blocks->num_blocks; i++) {
+        int num_chunks;
+        if (!rdma_local_ram_blocks->block[i].pmr) {
+            continue;
+        }
+        num_chunks = ram_chunk_count(&(rdma_local_ram_blocks->block[i]));
+        for (j = 0; j < num_chunks; j++) {
+            if (!rdma_local_ram_blocks->block[i].pmr[j]) {
+                continue;
+            }
+            ibv_dereg_mr(rdma_local_ram_blocks->block[i].pmr[j]);
+            rdma->total_registrations--;
+        }
+        g_free(rdma_local_ram_blocks->block[i].pmr);
+        rdma_local_ram_blocks->block[i].pmr = NULL;
+    }
+    for (i = 0; i < rdma_local_ram_blocks->num_blocks; i++) {
+        if (!rdma_local_ram_blocks->block[i].mr) {
+            continue;
+        }
+        ibv_dereg_mr(rdma_local_ram_blocks->block[i].mr);
+        rdma->total_registrations--;
+        rdma_local_ram_blocks->block[i].mr = NULL;
+    }
+}
+
+/*
+ * The protocol uses two different sets of rkeys (mutually exclusive):
+ * 1. One key to represent the virtual address of the entire ram block.
+ *    (dynamic chunk registration disabled - pin everything with one rkey.)
+ * 2. One to represent individual chunks within a ram block.
+ *    (dynamic chunk registration enabled - pin individual chunks.)
+ *
+ * Once the capability is successfully negotiated, the destination transmits
+ * the keys to use (or sends them later) including the virtual addresses
+ * and then propagates the remote ram block descriptions to his local copy.
+ */
+static int qemu_rdma_process_remote_blocks(RDMAContext *rdma, int num_blocks)
+{
+    RDMALocalBlocks *local = &rdma->local_ram_blocks;
+    int i, j;
+
+    if (local->num_blocks != num_blocks) {
+        fprintf(stderr, "local %d != remote %d\n",
+            local->num_blocks, num_blocks);
+        return -1;
+    }
+
+    for (i = 0; i < num_blocks; i++) {
+        /* search local ram blocks */
+        for (j = 0; j < local->num_blocks; j++) {
+            if (rdma->block[i].offset != local->block[j].offset) {
+                continue;
+            }
+            if (rdma->block[i].length != local->block[j].length) {
+                return -1;
+            }
+            local->block[j].remote_host_addr =
+                rdma->block[i].remote_host_addr;
+            local->block[j].remote_rkey = rdma->block[i].remote_rkey;
+            break;
+        }
+        if (j >= local->num_blocks) {
+            return -1;
+        }
+    }
+
+    return 0;
+}
+
+/*
+ * Find the ram block that corresponds to the page requested to be
+ * transmitted by QEMU.
+ *
+ * Once the block is found, also identify which 'chunk' within that
+ * block that the page belongs to.
+ *
+ * This search cannot fail or the migration will fail.
+ */
+static int qemu_rdma_search_ram_block(uint64_t offset, uint64_t length,
+        RDMALocalBlocks *blocks, int *block_index, int *chunk_index)
+{
+    int i;
+    uint8_t *host_addr;
+
+    for (i = 0; i < blocks->num_blocks; i++) {
+        if (offset < blocks->block[i].offset) {
+            continue;
+        }
+        if (offset + length >
+                blocks->block[i].offset + blocks->block[i].length) {
+            continue;
+        }
+
+        *block_index = i;
+        host_addr = blocks->block[i].local_host_addr +
+                (offset - blocks->block[i].offset);
+        *chunk_index = ram_chunk_index(blocks->block[i].local_host_addr,
+                        host_addr);
+        return 0;
+    }
+    return -1;
+}
+
+/*
+ * Register a chunk with IB. If the chunk was already registered
+ * previously, then skip.
+ *
+ * Also return the keys associated with the registration needed
+ * to perform the actual RDMA operation.
+ */
+static int qemu_rdma_register_and_get_keys(RDMAContext *rdma,
+        RDMALocalBlock *block, uint8_t * host_addr,
+        uint32_t *lkey, uint32_t *rkey)
+{
+    int chunk;
+    if (block->mr) {
+        if (lkey) {
+            *lkey = block->mr->lkey;
+        }
+        if (rkey) {
+            *rkey = block->mr->rkey;
+        }
+        return 0;
+    }
+
+    /* allocate memory to store chunk MRs */
+    if (!block->pmr) {
+        int num_chunks = ram_chunk_count(block);
+        block->pmr = g_malloc0(num_chunks *
+                sizeof(struct ibv_mr *));
+        if (!block->pmr) {
+            return -1;
+        }
+    }
+
+    /*
+     * If 'rkey', then we're the destination, so grant access to the source.
+     *
+     * If 'lkey', then we're the primary VM, so grant access only to ourselves.
+     */
+    chunk = ram_chunk_index(block->local_host_addr, host_addr);
+    if (!block->pmr[chunk]) {
+        uint8_t *start_addr = ram_chunk_start(block, chunk);
+        uint8_t *end_addr = ram_chunk_end(block, chunk);
+
+        block->pmr[chunk] = ibv_reg_mr(rdma->pd,
+                start_addr,
+                end_addr - start_addr,
+                (rkey ? (IBV_ACCESS_LOCAL_WRITE |
+                        IBV_ACCESS_REMOTE_WRITE) : 0));
+
+        if (!block->pmr[chunk]) {
+            perror("Failed to register chunk!");
+            fprintf(stderr, "Chunk details: block: %d chunk index %d"
+                            " start %" PRIu64 " end %" PRIu64 " host %" PRIu64
+                            " local %" PRIu64 " registrations: %d\n",
+                            block->index, chunk, (uint64_t) start_addr,
+                            (uint64_t) end_addr, (uint64_t) host_addr,
+                            (uint64_t) block->local_host_addr,
+                            rdma->total_registrations);
+            return -1;
+        }
+        rdma->total_registrations++;
+    }
+
+    if (lkey) {
+        *lkey = block->pmr[chunk]->lkey;
+    }
+    if (rkey) {
+        *rkey = block->pmr[chunk]->rkey;
+    }
+    return 0;
+}
+
+/*
+ * Register (at connection time) the memory used for control
+ * channel messages.
+ */
+static int qemu_rdma_reg_control(RDMAContext *rdma, int idx)
+{
+    rdma->wr_data[idx].control_mr = ibv_reg_mr(rdma->pd,
+            rdma->wr_data[idx].control, RDMA_CONTROL_MAX_BUFFER,
+            IBV_ACCESS_LOCAL_WRITE | IBV_ACCESS_REMOTE_WRITE);
+    if (rdma->wr_data[idx].control_mr) {
+        rdma->total_registrations++;
+        return 0;
+    }
+    fprintf(stderr, "qemu_rdma_reg_control failed!\n");
+    return -1;
+}
+
+static int qemu_rdma_dereg_control(RDMAContext *rdma, int idx)
+{
+    rdma->total_registrations--;
+    return ibv_dereg_mr(rdma->wr_data[idx].control_mr);
+}
+
+#if defined(DEBUG_RDMA) && defined(DEBUG_RDMA_VERBOSE)
+static const char *print_wrid(int wrid)
+{
+    if (wrid >= RDMA_WRID_RECV_CONTROL) {
+        return wrid_desc[RDMA_WRID_RECV_CONTROL];
+    }
+    return wrid_desc[wrid];
+}
+#endif
+
+/*
+ * Consult the connection manager to see a work request
+ * (of any kind) has completed.
+ * Return the work request ID that completed.
+ */
+static int qemu_rdma_poll(RDMAContext *rdma)
+{
+    int ret;
+    struct ibv_wc wc;
+
+    ret = ibv_poll_cq(rdma->cq, 1, &wc);
+    if (!ret) {
+        return RDMA_WRID_NONE;
+    }
+    if (ret < 0) {
+        fprintf(stderr, "ibv_poll_cq return %d!\n", ret);
+        return ret;
+    }
+    if (wc.status != IBV_WC_SUCCESS) {
+        fprintf(stderr, "ibv_poll_cq wc.status=%d %s!\n",
+                        wc.status, ibv_wc_status_str(wc.status));
+        fprintf(stderr, "ibv_poll_cq wrid=%s!\n", wrid_desc[wc.wr_id]);
+
+        return -1;
+    }
+
+    if (rdma->control_ready_expected &&
+        (wc.wr_id >= RDMA_WRID_RECV_CONTROL)) {
+        DDDPRINTF("completion %s #%" PRId64 " received (%" PRId64 ")\n",
+            wrid_desc[RDMA_WRID_RECV_CONTROL], wc.wr_id -
+            RDMA_WRID_RECV_CONTROL, wc.wr_id);
+        rdma->control_ready_expected = 0;
+    }
+
+    if (wc.wr_id == RDMA_WRID_RDMA_WRITE) {
+        rdma->num_signaled_send--;
+        DDDPRINTF("completions %s (%" PRId64 ") left %d\n",
+            print_wrid(wc.wr_id), wc.wr_id, rdma->num_signaled_send);
+    } else {
+        DDDPRINTF("other completion %s (%" PRId64 ") received left %d\n",
+            print_wrid(wc.wr_id), wc.wr_id, rdma->num_signaled_send);
+    }
+
+    return  (int)wc.wr_id;
+}
+
+/*
+ * Block until the next work request has completed.
+ *
+ * First poll to see if a work request has already completed,
+ * otherwise block.
+ *
+ * If we encounter completed work requests for IDs other than
+ * the one we're interested in, then that's generally an error.
+ *
+ * The only exception is actual RDMA Write completions. These
+ * completions only need to be recorded, but do not actually
+ * need further processing.
+ */
+static int qemu_rdma_block_for_wrid(RDMAContext *rdma, int wrid)
+{
+    int num_cq_events = 0;
+    int r = RDMA_WRID_NONE;
+    struct ibv_cq *cq;
+    void *cq_ctx;
+
+    if (ibv_req_notify_cq(rdma->cq, 0)) {
+        return -1;
+    }
+    /* poll cq first */
+    while (r != wrid) {
+        r = qemu_rdma_poll(rdma);
+        if (r < 0) {
+            return r;
+        }
+        if (r == RDMA_WRID_NONE) {
+            break;
+        }
+        if (r != wrid) {
+            DDDPRINTF("A Wanted wrid %s (%d) but got %s (%d)\n",
+                print_wrid(wrid), wrid, print_wrid(r), r);
+        }
+    }
+    if (r == wrid) {
+        return 0;
+    }
+
+    while (1) {
+        /*
+         * Coroutine doesn't start until process_incoming_migration()
+         * so don't yield unless we know we're running inside of a coroutine.
+         */
+        if (rdma->migration_started_on_destination) {
+            yield_until_fd_readable(rdma->comp_channel->fd);
+        }
+
+        if (ibv_get_cq_event(rdma->comp_channel, &cq, &cq_ctx)) {
+            perror("ibv_get_cq_event");
+            goto err_block_for_wrid;
+        }
+
+        num_cq_events++;
+
+        if (ibv_req_notify_cq(cq, 0)) {
+            goto err_block_for_wrid;
+        }
+        /* poll cq */
+        while (r != wrid) {
+            r = qemu_rdma_poll(rdma);
+            if (r < 0) {
+                goto err_block_for_wrid;
+            }
+            if (r == RDMA_WRID_NONE) {
+                break;
+            }
+            if (r != wrid) {
+                DDDPRINTF("B Wanted wrid %s (%d) but got %s (%d)\n",
+                    print_wrid(wrid), wrid, print_wrid(r), r);
+            }
+        }
+        if (r == wrid) {
+            goto success_block_for_wrid;
+        }
+    }
+
+success_block_for_wrid:
+    if (num_cq_events) {
+        ibv_ack_cq_events(cq, num_cq_events);
+    }
+    return 0;
+
+err_block_for_wrid:
+    if (num_cq_events) {
+        ibv_ack_cq_events(cq, num_cq_events);
+    }
+    return -1;
+}
+
+/*
+ * Post a SEND message work request for the control channel
+ * containing some data and block until the post completes.
+ */
+static int qemu_rdma_post_send_control(RDMAContext *rdma, uint8_t *buf,
+                                       RDMAControlHeader *head)
+{
+    int ret = 0;
+    RDMAWorkRequestData *wr = &rdma->wr_data[RDMA_CONTROL_MAX_WR];
+    struct ibv_send_wr *bad_wr;
+    struct ibv_sge sge = {
+                           .addr = (uint64_t)(wr->control),
+                           .length = head->len + sizeof(RDMAControlHeader),
+                           .lkey = wr->control_mr->lkey,
+                         };
+    struct ibv_send_wr send_wr = {
+                                   .wr_id = RDMA_WRID_SEND_CONTROL,
+                                   .opcode = IBV_WR_SEND,
+                                   .send_flags = IBV_SEND_SIGNALED,
+                                   .sg_list = &sge,
+                                   .num_sge = 1,
+                                };
+
+    DDDPRINTF("CONTROL: sending %s..\n", control_desc[head->type]);
+
+    /*
+     * We don't actually need to do a memcpy() in here if we used
+     * the "sge" properly, but since we're only sending control messages
+     * (not RAM in a performance-critical path), then its OK for now.
+     *
+     * The copy makes the RDMAControlHeader simpler to manipulate
+     * for the time being.
+     */
+    memcpy(wr->control, head, sizeof(RDMAControlHeader));
+    control_to_network((void *) wr->control);
+
+    if (buf) {
+        memcpy(wr->control + sizeof(RDMAControlHeader), buf, head->len);
+    }
+
+
+    if (ibv_post_send(rdma->qp, &send_wr, &bad_wr)) {
+        return -1;
+    }
+
+    if (ret < 0) {
+        fprintf(stderr, "Failed to use post IB SEND for control!\n");
+        return ret;
+    }
+
+    ret = qemu_rdma_block_for_wrid(rdma, RDMA_WRID_SEND_CONTROL);
+    if (ret < 0) {
+        fprintf(stderr, "rdma migration: send polling control error!\n");
+    }
+
+    return ret;
+}
+
+/*
+ * Post a RECV work request in anticipation of some future receipt
+ * of data on the control channel.
+ */
+static int qemu_rdma_post_recv_control(RDMAContext *rdma, int idx)
+{
+    struct ibv_recv_wr *bad_wr;
+    struct ibv_sge sge = {
+                            .addr = (uint64_t)(rdma->wr_data[idx].control),
+                            .length = RDMA_CONTROL_MAX_BUFFER,
+                            .lkey = rdma->wr_data[idx].control_mr->lkey,
+                         };
+
+    struct ibv_recv_wr recv_wr = {
+                                    .wr_id = RDMA_WRID_RECV_CONTROL + idx,
+                                    .sg_list = &sge,
+                                    .num_sge = 1,
+                                 };
+
+
+    if (ibv_post_recv(rdma->qp, &recv_wr, &bad_wr)) {
+        return -1;
+    }
+
+    return 0;
+}
+
+/*
+ * Block and wait for a RECV control channel message to arrive.
+ */
+static int qemu_rdma_exchange_get_response(RDMAContext *rdma,
+                RDMAControlHeader *head, int expecting, int idx)
+{
+    int ret = qemu_rdma_block_for_wrid(rdma, RDMA_WRID_RECV_CONTROL + idx);
+
+    if (ret < 0) {
+        fprintf(stderr, "rdma migration: recv polling control error!\n");
+        return ret;
+    }
+
+    network_to_control((void *) rdma->wr_data[idx].control);
+    memcpy(head, rdma->wr_data[idx].control, sizeof(RDMAControlHeader));
+
+    DDDPRINTF("CONTROL: %s received\n", control_desc[expecting]);
+
+    if ((expecting != RDMA_CONTROL_NONE && head->type != expecting)
+            || head->type == RDMA_CONTROL_ERROR) {
+        fprintf(stderr, "Was expecting a %s (%d) control message"
+                ", but got: %s (%d), length: %d\n",
+                control_desc[expecting], expecting,
+                control_desc[head->type], head->type, head->len);
+        return -EIO;
+    }
+
+    return 0;
+}
+
+/*
+ * When a RECV work request has completed, the work request's
+ * buffer is pointed at the header.
+ *
+ * This will advance the pointer to the data portion
+ * of the control message of the work request's buffer that
+ * was populated after the work request finished.
+ */
+static void qemu_rdma_move_header(RDMAContext *rdma, int idx,
+                                  RDMAControlHeader *head)
+{
+    rdma->wr_data[idx].control_len = head->len;
+    rdma->wr_data[idx].control_curr =
+        rdma->wr_data[idx].control + sizeof(RDMAControlHeader);
+}
+
+/*
+ * This is an 'atomic' high-level operation to deliver a single, unified
+ * control-channel message.
+ *
+ * Additionally, if the user is expecting some kind of reply to this message,
+ * they can request a 'resp' response message be filled in by posting an
+ * additional work request on behalf of the user and waiting for an additional
+ * completion.
+ *
+ * The extra (optional) response is used during registration to us from having
+ * to perform an *additional* exchange of message just to provide a response by
+ * instead piggy-backing on the acknowledgement.
+ */
+static int qemu_rdma_exchange_send(RDMAContext *rdma, RDMAControlHeader *head,
+                                   uint8_t *data, RDMAControlHeader *resp,
+                                   int *resp_idx)
+{
+    int ret = 0;
+    int idx = 0;
+
+    /*
+     * Wait until the dest is ready before attempting to deliver the message
+     * by waiting for a READY message.
+     */
+    if (rdma->control_ready_expected) {
+        RDMAControlHeader resp;
+        ret = qemu_rdma_exchange_get_response(rdma,
+                                    &resp, RDMA_CONTROL_READY, idx);
+        if (ret < 0) {
+            return ret;
+        }
+    }
+
+    /*
+     * If the user is expecting a response, post a WR in anticipation of it.
+     */
+    if (resp) {
+        ret = qemu_rdma_post_recv_control(rdma, idx + 1);
+        if (ret) {
+            fprintf(stderr, "rdma migration: error posting"
+                    " extra control recv for anticipated result!");
+            return ret;
+        }
+    }
+
+    /*
+     * Post a WR to replace the one we just consumed for the READY message.
+     */
+    ret = qemu_rdma_post_recv_control(rdma, idx);
+    if (ret) {
+        fprintf(stderr, "rdma migration: error posting first control recv!");
+        return ret;
+    }
+
+    /*
+     * Deliver the control message that was requested.
+     */
+    ret = qemu_rdma_post_send_control(rdma, data, head);
+
+    if (ret < 0) {
+        fprintf(stderr, "Failed to send control buffer!\n");
+        return ret;
+    }
+
+    /*
+     * If we're expecting a response, block and wait for it.
+     */
+    if (resp) {
+        DDPRINTF("Waiting for response %s\n", control_desc[resp->type]);
+        ret = qemu_rdma_exchange_get_response(rdma, resp, resp->type, idx + 1);
+
+        if (ret < 0) {
+            return ret;
+        }
+
+        qemu_rdma_move_header(rdma, idx + 1, resp);
+        *resp_idx = idx + 1;
+        DDPRINTF("Response %s received.\n", control_desc[resp->type]);
+    }
+
+    rdma->control_ready_expected = 1;
+
+    return 0;
+}
+
+/*
+ * This is an 'atomic' high-level operation to receive a single, unified
+ * control-channel message.
+ */
+static int qemu_rdma_exchange_recv(RDMAContext *rdma, RDMAControlHeader *head,
+                                int expecting)
+{
+    RDMAControlHeader ready = {
+                                .len = 0,
+                                .type = RDMA_CONTROL_READY,
+                                .repeat = 1,
+                              };
+    int ret;
+    int idx = 0;
+
+    /*
+     * Inform the source that we're ready to receive a message.
+     */
+    ret = qemu_rdma_post_send_control(rdma, NULL, &ready);
+
+    if (ret < 0) {
+        fprintf(stderr, "Failed to send control buffer!\n");
+        return ret;
+    }
+
+    /*
+     * Block and wait for the message.
+     */
+    ret = qemu_rdma_exchange_get_response(rdma, head, expecting, idx);
+
+    if (ret < 0) {
+        return ret;
+    }
+
+    qemu_rdma_move_header(rdma, idx, head);
+
+    /*
+     * Post a new RECV work request to replace the one we just consumed.
+     */
+    ret = qemu_rdma_post_recv_control(rdma, idx);
+    if (ret) {
+        fprintf(stderr, "rdma migration: error posting second control recv!");
+        return ret;
+    }
+
+    return 0;
+}
+
+/*
+ * Write an actual chunk of memory using RDMA.
+ *
+ * If we're using dynamic registration on the dest-side, we have to
+ * send a registration command first.
+ */
+static int qemu_rdma_write_one(QEMUFile *f, RDMAContext *rdma,
+        int current_index,
+        uint64_t offset, uint64_t length,
+        uint64_t wr_id, enum ibv_send_flags flag)
+{
+    struct ibv_sge sge;
+    struct ibv_send_wr send_wr = { 0 };
+    struct ibv_send_wr *bad_wr;
+    RDMALocalBlock *block = &(rdma->local_ram_blocks.block[current_index]);
+    int chunk;
+    RDMARegister reg;
+    RDMARegisterResult *reg_result;
+    int reg_result_idx;
+    RDMAControlHeader resp = { .type = RDMA_CONTROL_REGISTER_RESULT };
+    RDMAControlHeader head = { .len = sizeof(RDMARegister),
+                               .type = RDMA_CONTROL_REGISTER_REQUEST,
+                               .repeat = 1,
+                             };
+    int ret;
+
+    sge.addr = (uint64_t)(block->local_host_addr + (offset - block->offset));
+    sge.length = length;
+
+    if (!rdma->pin_all) {
+        chunk = ram_chunk_index(block->local_host_addr, (uint8_t *) sge.addr);
+        if (!block->remote_keys[chunk]) {
+            /*
+             * This page has not yet been registered, so first check to see
+             * if the entire chunk is zero. If so, tell the other size to
+             * memset() + madvise() the entire chunk without RDMA.
+             */
+            if (can_use_buffer_find_nonzero_offset((void *)sge.addr, length)
+                   && buffer_find_nonzero_offset((void *)sge.addr,
+                                                    length) == length) {
+                RDMACompress comp = {
+                                        .offset = offset,
+                                        .value = 0,
+                                        .block_idx = current_index,
+                                        .length = length,
+                                    };
+
+                head.len = sizeof(comp);
+                head.type = RDMA_CONTROL_COMPRESS;
+
+                DDPRINTF("Entire chunk is zero, sending compress: %d for %d "
+                    "bytes, index: %d, offset: %" PRId64 "...\n",
+                    chunk, sge.length, current_index, offset);
+
+                ret = qemu_rdma_exchange_send(rdma, &head,
+                                (uint8_t *) &comp, NULL, NULL);
+
+                if (ret < 0) {
+                    return -EIO;
+                }
+
+                return 1;
+            }
+
+            /*
+             * Otherwise, tell other side to register.
+             */
+            reg.len = sge.length;
+            reg.current_index = current_index;
+            reg.offset = offset;
+
+            DDPRINTF("Sending registration request chunk %d for %d "
+                    "bytes, index: %d, offset: %" PRId64 "...\n",
+                    chunk, sge.length, current_index, offset);
+
+            ret = qemu_rdma_exchange_send(rdma, &head, (uint8_t *) &reg,
+                                    &resp, &reg_result_idx);
+            if (ret < 0) {
+                return ret;
+            }
+
+            /* try to overlap this single registration with the one we sent. */
+            if (qemu_rdma_register_and_get_keys(rdma, block,
+                                                (uint8_t *) sge.addr,
+                                                &sge.lkey, NULL)) {
+                fprintf(stderr, "cannot get lkey!\n");
+                return -EINVAL;
+            }
+
+            reg_result = (RDMARegisterResult *)
+                    rdma->wr_data[reg_result_idx].control_curr;
+
+            DDPRINTF("Received registration result:"
+                    " my key: %x their key %x, chunk %d\n",
+                    block->remote_keys[chunk], reg_result->rkey, chunk);
+
+            block->remote_keys[chunk] = reg_result->rkey;
+        } else {
+            /* already registered before */
+            if (qemu_rdma_register_and_get_keys(rdma, block,
+                                                (uint8_t *)sge.addr,
+                                                &sge.lkey, NULL)) {
+                fprintf(stderr, "cannot get lkey!\n");
+                return -EINVAL;
+            }
+        }
+
+        send_wr.wr.rdma.rkey = block->remote_keys[chunk];
+    } else {
+        send_wr.wr.rdma.rkey = block->remote_rkey;
+
+        if (qemu_rdma_register_and_get_keys(rdma, block, (uint8_t *)sge.addr,
+                                                     &sge.lkey, NULL)) {
+            fprintf(stderr, "cannot get lkey!\n");
+            return -EINVAL;
+        }
+    }
+
+
+    send_wr.wr_id = wr_id;
+    send_wr.opcode = IBV_WR_RDMA_WRITE;
+    send_wr.send_flags = flag;
+    send_wr.sg_list = &sge;
+    send_wr.num_sge = 1;
+    send_wr.wr.rdma.remote_addr = block->remote_host_addr +
+                                    (offset - block->offset);
+
+    return ibv_post_send(rdma->qp, &send_wr, &bad_wr);
+}
+
+/*
+ * Push out any unwritten RDMA operations.
+ *
+ * We support sending out multiple chunks at the same time.
+ * Not all of them need to get signaled in the completion queue.
+ */
+static int qemu_rdma_write_flush(QEMUFile *f, RDMAContext *rdma)
+{
+    int ret;
+    enum ibv_send_flags flags = 0;
+
+    if (!rdma->current_length) {
+        return 0;
+    }
+    if (rdma->num_unsignaled_send >=
+            RDMA_UNSIGNALED_SEND_MAX) {
+        flags = IBV_SEND_SIGNALED;
+    }
+
+retry:
+    ret = qemu_rdma_write_one(f, rdma,
+            rdma->current_index,
+            rdma->current_offset,
+            rdma->current_length,
+            RDMA_WRID_RDMA_WRITE, flags);
+
+    if (ret < 0) {
+        if (ret == -ENOMEM) {
+            DDPRINTF("send queue is full. wait a little....\n");
+            ret = qemu_rdma_block_for_wrid(rdma, RDMA_WRID_RDMA_WRITE);
+            if (ret >= 0) {
+                goto retry;
+            }
+            if (ret < 0) {
+                fprintf(stderr, "rdma migration: failed to make "
+                                "room in full send queue! %d\n", ret);
+                return ret;
+            }
+        }
+        perror("write flush error");
+        return ret;
+    }
+
+    if (ret == 0) {
+        if (rdma->num_unsignaled_send >=
+                RDMA_UNSIGNALED_SEND_MAX) {
+            rdma->num_unsignaled_send = 0;
+            rdma->num_signaled_send++;
+            DDDPRINTF("signaled total: %d\n", rdma->num_signaled_send);
+        } else {
+            rdma->num_unsignaled_send++;
+        }
+    }
+
+    rdma->current_length = 0;
+    rdma->current_offset = 0;
+
+    return 0;
+}
+
+static inline int qemu_rdma_in_current_block(RDMAContext *rdma,
+                uint64_t offset, uint64_t len)
+{
+    RDMALocalBlock *block =
+        &(rdma->local_ram_blocks.block[rdma->current_index]);
+    if (rdma->current_index < 0) {
+        return 0;
+    }
+    if (offset < block->offset) {
+        return 0;
+    }
+    if (offset + len > block->offset + block->length) {
+        return 0;
+    }
+    return 1;
+}
+
+static inline int qemu_rdma_in_current_chunk(RDMAContext *rdma,
+                uint64_t offset, uint64_t len)
+{
+    RDMALocalBlock *block =
+            &(rdma->local_ram_blocks.block[rdma->current_index]);
+    uint8_t *chunk_end, *host_addr;
+
+    if (rdma->current_chunk < 0) {
+        return 0;
+    }
+
+    host_addr = block->local_host_addr + (offset - block->offset);
+    chunk_end = ram_chunk_end(block, rdma->current_chunk);
+
+    if ((host_addr + len) > chunk_end) {
+        return 0;
+    }
+
+    return 1;
+}
+
+static inline int qemu_rdma_buffer_mergable(RDMAContext *rdma,
+                    uint64_t offset, uint64_t len)
+{
+    if (rdma->current_length == 0) {
+        return 0;
+    }
+    if (offset != rdma->current_offset + rdma->current_length) {
+        return 0;
+    }
+    if (!qemu_rdma_in_current_block(rdma, offset, len)) {
+        return 0;
+    }
+    if (!qemu_rdma_in_current_chunk(rdma, offset, len)) {
+        return 0;
+    }
+    return 1;
+}
+
+/*
+ * We're not actually writing here, but doing three things:
+ *
+ * 1. Identify the chunk the buffer belongs to.
+ * 2. If the chunk is full or the buffer doesn't belong to the current
+ *    chunk, then start a new chunk and flush() the old chunk.
+ * 3. To keep the hardware busy, we also group chunks into batches
+ *    and only require that a batch gets acknowledged in the completion
+ *    qeueue instead of each individual chunk.
+ */
+static int qemu_rdma_write(QEMUFile *f, RDMAContext *rdma,
+                           uint64_t offset, uint64_t len)
+{
+    int index = rdma->current_index;
+    int chunk_index = rdma->current_chunk;
+    int ret;
+
+    /* If we cannot merge it, we flush the current buffer first. */
+    if (!qemu_rdma_buffer_mergable(rdma, offset, len)) {
+        ret = qemu_rdma_write_flush(f, rdma);
+        if (ret) {
+            return ret;
+        }
+        rdma->current_length = 0;
+        rdma->current_offset = offset;
+
+        ret = qemu_rdma_search_ram_block(offset, len,
+                    &rdma->local_ram_blocks, &index, &chunk_index);
+        if (ret) {
+            fprintf(stderr, "ram block search failed\n");
+            return ret;
+        }
+        rdma->current_index = index;
+        rdma->current_chunk = chunk_index;
+    }
+
+    /* merge it */
+    rdma->current_length += len;
+
+    /* flush it if buffer is too large */
+    if (rdma->current_length >= RDMA_MERGE_MAX) {
+        return qemu_rdma_write_flush(f, rdma);
+    }
+
+    return 0;
+}
+
+static void qemu_rdma_cleanup(RDMAContext *rdma)
+{
+    struct rdma_cm_event *cm_event;
+    int ret, idx;
+
+    if (rdma->cm_id) {
+        if (rdma->error_state) {
+            RDMAControlHeader head = { .len = 0,
+                                       .type = RDMA_CONTROL_ERROR,
+                                       .repeat = 1,
+                                     };
+            fprintf(stderr, "Early error. Sending error.\n");
+            qemu_rdma_post_send_control(rdma, NULL, &head);
+        }
+
+        ret = rdma_disconnect(rdma->cm_id);
+        if (!ret) {
+            DDPRINTF("waiting for disconnect\n");
+            ret = rdma_get_cm_event(rdma->channel, &cm_event);
+            if (!ret) {
+                rdma_ack_cm_event(cm_event);
+            }
+        }
+        DDPRINTF("Disconnected.\n");
+        rdma->cm_id = 0;
+    }
+
+    g_free(rdma->block);
+    rdma->block = NULL;
+
+    for (idx = 0; idx < (RDMA_CONTROL_MAX_WR + 1); idx++) {
+        if (rdma->wr_data[idx].control_mr) {
+            qemu_rdma_dereg_control(rdma, idx);
+        }
+        rdma->wr_data[idx].control_mr = NULL;
+    }
+
+    if (rdma->local_ram_blocks.block) {
+        qemu_rdma_dereg_ram_blocks(rdma, &rdma->local_ram_blocks);
+
+        if (!rdma->pin_all) {
+            for (idx = 0; idx < rdma->local_ram_blocks.num_blocks; idx++) {
+                RDMALocalBlock *block = &(rdma->local_ram_blocks.block[idx]);
+                g_free(block->remote_keys);
+                block->remote_keys = NULL;
+            }
+        }
+        g_free(rdma->local_ram_blocks.block);
+        rdma->local_ram_blocks.block = NULL;
+    }
+
+    if (rdma->qp) {
+        ibv_destroy_qp(rdma->qp);
+        rdma->qp = NULL;
+    }
+    if (rdma->cq) {
+        ibv_destroy_cq(rdma->cq);
+        rdma->cq = NULL;
+    }
+    if (rdma->comp_channel) {
+        ibv_destroy_comp_channel(rdma->comp_channel);
+        rdma->comp_channel = NULL;
+    }
+    if (rdma->pd) {
+        ibv_dealloc_pd(rdma->pd);
+        rdma->pd = NULL;
+    }
+    if (rdma->listen_id) {
+        rdma_destroy_id(rdma->listen_id);
+        rdma->listen_id = 0;
+    }
+    if (rdma->cm_id) {
+        rdma_destroy_id(rdma->cm_id);
+        rdma->cm_id = 0;
+    }
+    if (rdma->channel) {
+        rdma_destroy_event_channel(rdma->channel);
+        rdma->channel = NULL;
+    }
+}
+
+
+static int qemu_rdma_source_init(RDMAContext *rdma, Error **errp, bool pin_all)
+{
+    int ret, idx;
+    Error *local_err = NULL, **temp = &local_err;
+
+    /*
+     * Will be validated against destination's actual capabilities
+     * after the connect() completes.
+     */
+    rdma->pin_all = pin_all;
+
+    ret = qemu_rdma_resolve_host(rdma, temp);
+    if (ret) {
+        goto err_rdma_source_init;
+    }
+
+    ret = qemu_rdma_alloc_pd_cq(rdma);
+    if (ret) {
+        ERROR(temp, "rdma migration: error allocating pd and cq! Your mlock()"
+                    " limits may be too low. Please check $ ulimit -a # and "
+                    "search for 'ulimit -l' in the output\n");
+        goto err_rdma_source_init;
+    }
+
+    ret = qemu_rdma_alloc_qp(rdma);
+    if (ret) {
+        ERROR(temp, "rdma migration: error allocating qp!\n");
+        goto err_rdma_source_init;
+    }
+
+    ret = qemu_rdma_init_ram_blocks(&rdma->local_ram_blocks);
+    if (ret) {
+        ERROR(temp, "rdma migration: error initializing ram blocks!\n");
+        goto err_rdma_source_init;
+    }
+
+    for (idx = 0; idx < (RDMA_CONTROL_MAX_WR + 1); idx++) {
+        ret = qemu_rdma_reg_control(rdma, idx);
+        if (ret) {
+            ERROR(temp, "rdma migration: error registering %d control!\n",
+                                                            idx);
+            goto err_rdma_source_init;
+        }
+    }
+
+    rdma->block = (RDMARemoteBlock *) g_malloc0(sizeof(RDMARemoteBlock) *
+                        rdma->local_ram_blocks.num_blocks);
+    return 0;
+
+err_rdma_source_init:
+    error_propagate(errp, local_err);
+    qemu_rdma_cleanup(rdma);
+    return -1;
+}
+
+static int qemu_rdma_connect(RDMAContext *rdma, Error **errp)
+{
+    RDMAControlHeader head;
+    RDMACapabilities cap = {
+                                .version = RDMA_CONTROL_VERSION_CURRENT,
+                                .flags = 0,
+                           };
+    struct rdma_conn_param conn_param = { .initiator_depth = 2,
+                                          .retry_count = 5,
+                                          .private_data = &cap,
+                                          .private_data_len = sizeof(cap),
+                                        };
+    struct rdma_cm_event *cm_event;
+    int ret;
+    int idx = 0;
+    int x;
+
+    /*
+     * Only negotiate the capability with destination if the user
+     * on the source first requested the capability.
+     */
+    if (rdma->pin_all) {
+        DPRINTF("Server pin-all memory requested.\n");
+        cap.flags |= RDMA_CAPABILITY_PIN_ALL;
+    }
+
+    caps_to_network(&cap);
+
+    ret = rdma_connect(rdma->cm_id, &conn_param);
+    if (ret) {
+        perror("rdma_connect");
+        ERROR(errp, "connecting to destination!\n");
+        rdma_destroy_id(rdma->cm_id);
+        rdma->cm_id = 0;
+        goto err_rdma_source_connect;
+    }
+
+    ret = rdma_get_cm_event(rdma->channel, &cm_event);
+    if (ret) {
+        perror("rdma_get_cm_event after rdma_connect");
+        ERROR(errp, "connecting to destination!\n");
+        rdma_ack_cm_event(cm_event);
+        rdma_destroy_id(rdma->cm_id);
+        rdma->cm_id = 0;
+        goto err_rdma_source_connect;
+    }
+
+    if (cm_event->event != RDMA_CM_EVENT_ESTABLISHED) {
+        perror("rdma_get_cm_event != EVENT_ESTABLISHED after rdma_connect");
+        ERROR(errp, "connecting to destination!\n");
+        rdma_ack_cm_event(cm_event);
+        rdma_destroy_id(rdma->cm_id);
+        rdma->cm_id = 0;
+        goto err_rdma_source_connect;
+    }
+
+    memcpy(&cap, cm_event->param.conn.private_data, sizeof(cap));
+    network_to_caps(&cap);
+
+    /*
+     * Verify that the *requested* capabilities are supported by the destination
+     * and disable them otherwise.
+     */
+    if (rdma->pin_all && !(cap.flags & RDMA_CAPABILITY_PIN_ALL)) {
+        ERROR(errp, "Server cannot support pinning all memory. "
+                        "Will register memory dynamically.\n");
+        rdma->pin_all = false;
+    }
+
+    DPRINTF("Pin all memory: %s\n", rdma->pin_all ? "enabled" : "disabled");
+
+    rdma_ack_cm_event(cm_event);
+
+    ret = qemu_rdma_post_recv_control(rdma, idx + 1);
+    if (ret) {
+        ERROR(errp, "posting first control recv!\n");
+        goto err_rdma_source_connect;
+    }
+
+    ret = qemu_rdma_post_recv_control(rdma, idx);
+    if (ret) {
+        ERROR(errp, "posting second control recv!\n");
+        goto err_rdma_source_connect;
+    }
+
+    ret = qemu_rdma_exchange_get_response(rdma,
+                            &head, RDMA_CONTROL_RAM_BLOCKS, idx + 1);
+
+    if (ret < 0) {
+        ERROR(errp, "receiving remote info!\n");
+        goto err_rdma_source_connect;
+    }
+
+    qemu_rdma_move_header(rdma, idx + 1, &head);
+    memcpy(rdma->block, rdma->wr_data[idx + 1].control_curr, head.len);
+
+    ret = qemu_rdma_process_remote_blocks(rdma,
+                        (head.len / sizeof(RDMARemoteBlock)));
+    if (ret) {
+        ERROR(errp, "processing remote ram blocks!\n");
+        goto err_rdma_source_connect;
+    }
+
+    if (!rdma->pin_all) {
+        for (x = 0; x < rdma->local_ram_blocks.num_blocks; x++) {
+            RDMALocalBlock *block = &(rdma->local_ram_blocks.block[x]);
+            int num_chunks = ram_chunk_count(block);
+            /* allocate memory to store remote rkeys */
+            block->remote_keys = g_malloc0(num_chunks * sizeof(uint32_t));
+        }
+    }
+
+    rdma->control_ready_expected = 1;
+    rdma->num_signaled_send = 0;
+    return 0;
+
+err_rdma_source_connect:
+    qemu_rdma_cleanup(rdma);
+    return -1;
+}
+
+static int qemu_rdma_dest_init(RDMAContext *rdma, Error **errp)
+{
+    int ret = -EINVAL, idx;
+    struct sockaddr_in sin;
+    struct rdma_cm_id *listen_id;
+    char ip[40] = "unknown";
+
+    for (idx = 0; idx < RDMA_CONTROL_MAX_WR; idx++) {
+        rdma->wr_data[idx].control_len = 0;
+        rdma->wr_data[idx].control_curr = NULL;
+    }
+
+    if (rdma->host == NULL) {
+        ERROR(errp, "RDMA host is not set!\n");
+        rdma->error_state = -EINVAL;
+        return -1;
+    }
+    /* create CM channel */
+    rdma->channel = rdma_create_event_channel();
+    if (!rdma->channel) {
+        ERROR(errp, "could not create rdma event channel\n");
+        rdma->error_state = -EINVAL;
+        return -1;
+    }
+
+    /* create CM id */
+    ret = rdma_create_id(rdma->channel, &listen_id, NULL, RDMA_PS_TCP);
+    if (ret) {
+        ERROR(errp, "could not create cm_id!\n");
+        goto err_dest_init_create_listen_id;
+    }
+
+    memset(&sin, 0, sizeof(sin));
+    sin.sin_family = AF_INET;
+    sin.sin_port = htons(rdma->port);
+
+    if (rdma->host && strcmp("", rdma->host)) {
+        struct hostent *dest_addr;
+        dest_addr = gethostbyname(rdma->host);
+        if (!dest_addr) {
+            ERROR(errp, "migration could not gethostbyname!\n");
+            ret = -EINVAL;
+            goto err_dest_init_bind_addr;
+        }
+        memcpy(&sin.sin_addr.s_addr, dest_addr->h_addr,
+                dest_addr->h_length);
+        inet_ntop(AF_INET, dest_addr->h_addr, ip, sizeof ip);
+    } else {
+        sin.sin_addr.s_addr = INADDR_ANY;
+    }
+
+    DPRINTF("%s => %s\n", rdma->host, ip);
+
+    ret = rdma_bind_addr(listen_id, (struct sockaddr *)&sin);
+    if (ret) {
+        ERROR(errp, "Error: could not rdma_bind_addr!\n");
+        goto err_dest_init_bind_addr;
+    }
+
+    rdma->listen_id = listen_id;
+    if (listen_id->verbs) {
+        rdma->verbs = listen_id->verbs;
+    }
+    qemu_rdma_dump_id("dest_init", rdma->verbs);
+    qemu_rdma_dump_gid("dest_init", listen_id);
+    return 0;
+
+err_dest_init_bind_addr:
+    rdma_destroy_id(listen_id);
+err_dest_init_create_listen_id:
+    rdma_destroy_event_channel(rdma->channel);
+    rdma->channel = NULL;
+    rdma->error_state = ret;
+    return ret;
+
+}
+
+static int qemu_rdma_dest_prepare(RDMAContext *rdma, Error **errp)
+{
+    int ret;
+    int idx;
+
+    if (!rdma->verbs) {
+        ERROR(errp, "no verbs context!\n");
+        return 0;
+    }
+
+    ret = qemu_rdma_alloc_pd_cq(rdma);
+    if (ret) {
+        ERROR(errp, "allocating pd and cq!\n");
+        goto err_rdma_dest_prepare;
+    }
+
+    ret = qemu_rdma_init_ram_blocks(&rdma->local_ram_blocks);
+    if (ret) {
+        ERROR(errp, "initializing ram blocks!\n");
+        goto err_rdma_dest_prepare;
+    }
+
+    rdma->block = (RDMARemoteBlock *) g_malloc0(sizeof(RDMARemoteBlock) *
+                        rdma->local_ram_blocks.num_blocks);
+
+    /* Extra one for the send buffer */
+    for (idx = 0; idx < (RDMA_CONTROL_MAX_WR + 1); idx++) {
+        ret = qemu_rdma_reg_control(rdma, idx);
+        if (ret) {
+            ERROR(errp, "registering %d control!\n", idx);
+            goto err_rdma_dest_prepare;
+        }
+    }
+
+    ret = rdma_listen(rdma->listen_id, 5);
+    if (ret) {
+        ERROR(errp, "listening on socket!\n");
+        goto err_rdma_dest_prepare;
+    }
+
+    return 0;
+
+err_rdma_dest_prepare:
+    qemu_rdma_cleanup(rdma);
+    return -1;
+}
+
+static void *qemu_rdma_data_init(const char *host_port, Error **errp)
+{
+    RDMAContext *rdma = NULL;
+    InetSocketAddress *addr;
+
+    if (host_port) {
+        rdma = g_malloc0(sizeof(RDMAContext));
+        memset(rdma, 0, sizeof(RDMAContext));
+        rdma->current_index = -1;
+        rdma->current_chunk = -1;
+
+        addr = inet_parse(host_port, NULL);
+        if (addr != NULL) {
+            rdma->port = atoi(addr->port);
+            rdma->host = g_strdup(addr->host);
+        } else {
+            ERROR(errp, "bad RDMA migration address '%s'", host_port);
+            g_free(rdma);
+            return NULL;
+        }
+    }
+
+    return rdma;
+}
+
+/*
+ * QEMUFile interface to the control channel.
+ * SEND messages for control only.
+ * pc.ram is handled with regular RDMA messages.
+ */
+static int qemu_rdma_put_buffer(void *opaque, const uint8_t *buf,
+                                int64_t pos, int size)
+{
+    QEMUFileRDMA *r = opaque;
+    QEMUFile *f = r->file;
+    RDMAContext *rdma = r->rdma;
+    size_t remaining = size;
+    uint8_t * data = (void *) buf;
+    int ret;
+
+    CHECK_ERROR_STATE();
+
+    /*
+     * Push out any writes that
+     * we're queued up for pc.ram.
+     */
+    if (qemu_rdma_write_flush(f, rdma) < 0) {
+        rdma->error_state = -EIO;
+        return rdma->error_state;
+    }
+
+    while (remaining) {
+        RDMAControlHeader head;
+
+        r->len = MIN(remaining, RDMA_SEND_INCREMENT);
+        remaining -= r->len;
+
+        head.len = r->len;
+        head.type = RDMA_CONTROL_QEMU_FILE;
+
+        ret = qemu_rdma_exchange_send(rdma, &head, data, NULL, NULL);
+
+        if (ret < 0) {
+            rdma->error_state = ret;
+            return ret;
+        }
+
+        data += r->len;
+    }
+
+    return size;
+}
+
+static size_t qemu_rdma_fill(RDMAContext *rdma, uint8_t *buf,
+                             int size, int idx)
+{
+    size_t len = 0;
+
+    if (rdma->wr_data[idx].control_len) {
+        DDDPRINTF("RDMA %" PRId64 " of %d bytes already in buffer\n",
+                    rdma->wr_data[idx].control_len, size);
+
+        len = MIN(size, rdma->wr_data[idx].control_len);
+        memcpy(buf, rdma->wr_data[idx].control_curr, len);
+        rdma->wr_data[idx].control_curr += len;
+        rdma->wr_data[idx].control_len -= len;
+    }
+
+    return len;
+}
+
+/*
+ * QEMUFile interface to the control channel.
+ * RDMA links don't use bytestreams, so we have to
+ * return bytes to QEMUFile opportunistically.
+ */
+static int qemu_rdma_get_buffer(void *opaque, uint8_t *buf,
+                                int64_t pos, int size)
+{
+    QEMUFileRDMA *r = opaque;
+    RDMAContext *rdma = r->rdma;
+    RDMAControlHeader head;
+    int ret = 0;
+
+    CHECK_ERROR_STATE();
+
+    /*
+     * First, we hold on to the last SEND message we
+     * were given and dish out the bytes until we run
+     * out of bytes.
+     */
+    r->len = qemu_rdma_fill(r->rdma, buf, size, 0);
+    if (r->len) {
+        return r->len;
+    }
+
+    /*
+     * Once we run out, we block and wait for another
+     * SEND message to arrive.
+     */
+    ret = qemu_rdma_exchange_recv(rdma, &head, RDMA_CONTROL_QEMU_FILE);
+
+    if (ret < 0) {
+        rdma->error_state = ret;
+        return ret;
+    }
+
+    /*
+     * SEND was received with new bytes, now try again.
+     */
+    return qemu_rdma_fill(r->rdma, buf, size, 0);
+}
+
+/*
+ * Block until all the outstanding chunks have been delivered by the hardware.
+ */
+static int qemu_rdma_drain_cq(QEMUFile *f, RDMAContext *rdma)
+{
+    int ret;
+
+    if (qemu_rdma_write_flush(f, rdma) < 0) {
+        return -EIO;
+    }
+
+    while (rdma->num_signaled_send) {
+        ret = qemu_rdma_block_for_wrid(rdma, RDMA_WRID_RDMA_WRITE);
+        if (ret < 0) {
+            fprintf(stderr, "rdma migration: complete polling error!\n");
+            return -EIO;
+        }
+    }
+
+    return 0;
+}
+
+static int qemu_rdma_close(void *opaque)
+{
+    QEMUFileRDMA *r = opaque;
+    if (r->rdma) {
+        qemu_rdma_cleanup(r->rdma);
+        g_free(r->rdma);
+    }
+    g_free(r);
+    return 0;
+}
+
+static size_t qemu_rdma_save_page(QEMUFile *f, void *opaque,
+                   ram_addr_t block_offset, ram_addr_t offset, size_t size)
+{
+    ram_addr_t current_addr = block_offset + offset;
+    QEMUFileRDMA *rfile = opaque;
+    RDMAContext *rdma = rfile->rdma;
+    int ret;
+
+    CHECK_ERROR_STATE();
+
+    qemu_fflush(f);
+
+    /*
+     * Add this page to the current 'chunk'. If the chunk
+     * is full, or the page doen't belong to the current chunk,
+     * an actual RDMA write will occur and a new chunk will be formed.
+     */
+    ret = qemu_rdma_write(f, rdma, current_addr, size);
+    if (ret < 0) {
+        rdma->error_state = ret;
+        fprintf(stderr, "rdma migration: write error! %d\n", ret);
+        return ret;
+    }
+
+    /*
+     * Drain the Completion Queue if possible, but do not block,
+     * just poll.
+     *
+     * If nothing to poll, the end of the iteration will do this
+     * again to make sure we don't overflow the request queue.
+     */
+    while (1) {
+        int ret = qemu_rdma_poll(rdma);
+        if (ret == RDMA_WRID_NONE) {
+            break;
+        }
+        if (ret < 0) {
+            rdma->error_state = ret;
+            fprintf(stderr, "rdma migration: polling error! %d\n", ret);
+            return ret;
+        }
+    }
+
+    return size;
+}
+
+static int qemu_rdma_accept(RDMAContext *rdma)
+{
+    RDMAControlHeader head = { .len = rdma->local_ram_blocks.num_blocks *
+                                        sizeof(RDMARemoteBlock),
+                               .type = RDMA_CONTROL_RAM_BLOCKS,
+                               .repeat = 1,
+                             };
+    RDMACapabilities cap;
+    struct rdma_conn_param conn_param = {
+                                            .responder_resources = 2,
+                                            .private_data = &cap,
+                                            .private_data_len = sizeof(cap),
+                                         };
+    struct rdma_cm_event *cm_event;
+    struct ibv_context *verbs;
+    int ret = -EINVAL;
+    RDMALocalBlocks *local = &rdma->local_ram_blocks;
+    int i;
+
+    ret = rdma_get_cm_event(rdma->channel, &cm_event);
+    if (ret) {
+        goto err_rdma_dest_wait;
+    }
+
+    if (cm_event->event != RDMA_CM_EVENT_CONNECT_REQUEST) {
+        rdma_ack_cm_event(cm_event);
+        goto err_rdma_dest_wait;
+    }
+
+    memcpy(&cap, cm_event->param.conn.private_data, sizeof(cap));
+
+    network_to_caps(&cap);
+
+    if (cap.version < 1 || cap.version > RDMA_CONTROL_VERSION_CURRENT) {
+            fprintf(stderr, "Unknown source RDMA version: %d, bailing...\n",
+                            cap.version);
+            rdma_ack_cm_event(cm_event);
+            goto err_rdma_dest_wait;
+    }
+
+    /*
+     * Respond with only the capabilities this version of QEMU knows about.
+     */
+    cap.flags &= known_capabilities;
+
+    /*
+     * Enable the ones that we do know about.
+     * Add other checks here as new ones are introduced.
+     */
+    if (cap.flags & RDMA_CAPABILITY_PIN_ALL) {
+        rdma->pin_all = true;
+    }
+
+    rdma->cm_id = cm_event->id;
+    verbs = cm_event->id->verbs;
+
+    rdma_ack_cm_event(cm_event);
+
+    DPRINTF("Memory pin all: %s\n", rdma->pin_all ? "enabled" : "disabled");
+
+    caps_to_network(&cap);
+
+    DPRINTF("verbs context after listen: %p\n", verbs);
+
+    if (!rdma->verbs) {
+        rdma->verbs = verbs;
+        /*
+         * Cannot propagate errp, as there is no error pointer
+         * to be propagated.
+         */
+        ret = qemu_rdma_dest_prepare(rdma, NULL);
+        if (ret) {
+            fprintf(stderr, "rdma migration: error preparing dest!\n");
+            goto err_rdma_dest_wait;
+        }
+    } else if (rdma->verbs != verbs) {
+            fprintf(stderr, "ibv context not matching %p, %p!\n",
+                    rdma->verbs, verbs);
+            goto err_rdma_dest_wait;
+    }
+
+    qemu_set_fd_handler2(rdma->channel->fd, NULL, NULL, NULL, NULL);
+
+    ret = qemu_rdma_alloc_qp(rdma);
+    if (ret) {
+        fprintf(stderr, "rdma migration: error allocating qp!\n");
+        goto err_rdma_dest_wait;
+    }
+
+    ret = rdma_accept(rdma->cm_id, &conn_param);
+    if (ret) {
+        fprintf(stderr, "rdma_accept returns %d!\n", ret);
+        goto err_rdma_dest_wait;
+    }
+
+    ret = rdma_get_cm_event(rdma->channel, &cm_event);
+    if (ret) {
+        fprintf(stderr, "rdma_accept get_cm_event failed %d!\n", ret);
+        goto err_rdma_dest_wait;
+    }
+
+    if (cm_event->event != RDMA_CM_EVENT_ESTABLISHED) {
+        fprintf(stderr, "rdma_accept not event established!\n");
+        rdma_ack_cm_event(cm_event);
+        goto err_rdma_dest_wait;
+    }
+
+    rdma_ack_cm_event(cm_event);
+
+    ret = qemu_rdma_post_recv_control(rdma, 0);
+    if (ret) {
+        fprintf(stderr, "rdma migration: error posting second control recv!\n");
+        goto err_rdma_dest_wait;
+    }
+
+    if (rdma->pin_all) {
+        ret = qemu_rdma_reg_whole_ram_blocks(rdma, &rdma->local_ram_blocks);
+        if (ret) {
+            fprintf(stderr, "rdma migration: error dest "
+                            "registering ram blocks!\n");
+            goto err_rdma_dest_wait;
+        }
+    }
+
+    /*
+     * Server uses this to prepare to transmit the RAMBlock descriptions
+     * to the primary VM after connection setup.
+     * Both sides use the "remote" structure to communicate and update
+     * their "local" descriptions with what was sent.
+     */
+    for (i = 0; i < local->num_blocks; i++) {
+            rdma->block[i].remote_host_addr =
+                (uint64_t)(local->block[i].local_host_addr);
+
+            if (rdma->pin_all) {
+                rdma->block[i].remote_rkey = local->block[i].mr->rkey;
+            }
+
+            rdma->block[i].offset = local->block[i].offset;
+            rdma->block[i].length = local->block[i].length;
+    }
+
+
+    ret = qemu_rdma_post_send_control(rdma, (uint8_t *) rdma->block, &head);
+
+    if (ret < 0) {
+        fprintf(stderr, "rdma migration: error sending remote info!\n");
+        goto err_rdma_dest_wait;
+    }
+
+    qemu_rdma_dump_gid("dest_connect", rdma->cm_id);
+
+    return 0;
+
+err_rdma_dest_wait:
+    rdma->error_state = ret;
+    qemu_rdma_cleanup(rdma);
+    return ret;
+}
+
+/*
+ * During each iteration of the migration, we listen for instructions
+ * by the primary VM to perform dynamic page registrations before they
+ * can perform RDMA operations.
+ *
+ * We respond with the 'rkey'.
+ *
+ * Keep doing this until the primary tells us to stop.
+ */
+static int qemu_rdma_registration_handle(QEMUFile *f, void *opaque,
+                                         uint64_t flags)
+{
+    RDMAControlHeader resp = { .len = sizeof(RDMARegisterResult),
+                               .type = RDMA_CONTROL_REGISTER_RESULT,
+                               .repeat = 0,
+                             };
+    QEMUFileRDMA *rfile = opaque;
+    RDMAContext *rdma = rfile->rdma;
+    RDMAControlHeader head;
+    RDMARegister *reg, *registers;
+    RDMACompress *comp;
+    RDMARegisterResult *reg_result;
+    static RDMARegisterResult results[RDMA_CONTROL_MAX_COMMANDS_PER_MESSAGE];
+    RDMALocalBlock *block;
+    void *host_addr;
+    int ret = 0;
+    int idx = 0;
+    int count = 0;
+
+    CHECK_ERROR_STATE();
+
+    if (rdma->pin_all) {
+        return 0;
+    }
+
+    do {
+        DDDPRINTF("Waiting for next registration %" PRIu64 "...\n", flags);
+
+        ret = qemu_rdma_exchange_recv(rdma, &head, RDMA_CONTROL_NONE);
+
+        if (ret < 0) {
+            break;
+        }
+
+        if (head.repeat > RDMA_CONTROL_MAX_COMMANDS_PER_MESSAGE) {
+            fprintf(stderr, "Too many requests in this message (%d)."
+                            "Bailing.\n", head.repeat);
+            ret = -EIO;
+            break;
+        }
+
+        switch (head.type) {
+        case RDMA_CONTROL_COMPRESS:
+            comp = (RDMACompress *) rdma->wr_data[idx].control_curr;
+
+            DDPRINTF("Zapping zero chunk: %" PRId64
+                    " bytes, index %d, offset %" PRId64 "\n",
+                    comp->length, comp->block_idx, comp->offset);
+            comp = (RDMACompress *) rdma->wr_data[idx].control_curr;
+            block = &(rdma->local_ram_blocks.block[comp->block_idx]);
+
+            host_addr = block->local_host_addr +
+                            (comp->offset - block->offset);
+
+            ram_handle_compressed(host_addr, comp->value, comp->length);
+            break;
+        case RDMA_CONTROL_REGISTER_FINISHED:
+            DDDPRINTF("Current registrations complete.\n");
+            goto out;
+        case RDMA_CONTROL_REGISTER_REQUEST:
+            DDPRINTF("There are %d registration requests\n", head.repeat);
+
+            resp.repeat = head.repeat;
+            registers = (RDMARegister *) rdma->wr_data[idx].control_curr;
+
+            for (count = 0; count < head.repeat; count++) {
+                reg = &registers[count];
+                reg_result = &results[count];
+
+                DDPRINTF("Registration request (%d): %d"
+                    " bytes, index %d, offset %" PRIu64 "\n",
+                    count, reg->len, reg->current_index, reg->offset);
+
+                block = &(rdma->local_ram_blocks.block[reg->current_index]);
+                host_addr = (block->local_host_addr +
+                            (reg->offset - block->offset));
+                if (qemu_rdma_register_and_get_keys(rdma, block,
+                            (uint8_t *)host_addr, NULL, &reg_result->rkey)) {
+                    fprintf(stderr, "cannot get rkey!\n");
+                    ret = -EINVAL;
+                    goto out;
+                }
+
+                DDPRINTF("Registered rkey for this request: %x\n",
+                                reg_result->rkey);
+            }
+
+            ret = qemu_rdma_post_send_control(rdma,
+                            (uint8_t *) results, &resp);
+
+            if (ret < 0) {
+                fprintf(stderr, "Failed to send control buffer!\n");
+                goto out;
+            }
+            break;
+        case RDMA_CONTROL_REGISTER_RESULT:
+            fprintf(stderr, "Invalid RESULT message at dest.\n");
+            ret = -EIO;
+            goto out;
+        default:
+            fprintf(stderr, "Unknown control message %s\n",
+                                control_desc[head.type]);
+            ret = -EIO;
+            goto out;
+        }
+    } while (1);
+out:
+    if (ret < 0) {
+        rdma->error_state = ret;
+    }
+    return ret;
+}
+
+static int qemu_rdma_registration_start(QEMUFile *f, void *opaque,
+                                        uint64_t flags)
+{
+    QEMUFileRDMA *rfile = opaque;
+    RDMAContext *rdma = rfile->rdma;
+
+    CHECK_ERROR_STATE();
+
+    if (rdma->pin_all) {
+        return 0;
+    }
+
+    DDDPRINTF("start section: %" PRIu64 "\n", flags);
+    qemu_put_be64(f, RAM_SAVE_FLAG_HOOK);
+    qemu_fflush(f);
+
+    return 0;
+}
+
+/*
+ * Inform dest that dynamic registrations are done for now.
+ * First, flush writes, if any.
+ */
+static int qemu_rdma_registration_stop(QEMUFile *f, void *opaque,
+                                       uint64_t flags)
+{
+    QEMUFileRDMA *rfile = opaque;
+    RDMAContext *rdma = rfile->rdma;
+    RDMAControlHeader head = { .len = 0,
+                               .type = RDMA_CONTROL_REGISTER_FINISHED,
+                               .repeat = 1,
+                             };
+
+    CHECK_ERROR_STATE();
+
+    qemu_fflush(f);
+    int ret = qemu_rdma_drain_cq(f, rdma);
+
+    if (ret < 0) {
+        goto err;
+    }
+
+    if (rdma->pin_all) {
+        return 0;
+    }
+
+    DDDPRINTF("Sending registration finish %" PRIu64 "...\n", flags);
+    ret = qemu_rdma_exchange_send(rdma, &head, NULL, NULL, NULL);
+
+    if (ret < 0) {
+        goto err;
+    }
+
+    return 0;
+err:
+    rdma->error_state = ret;
+    return ret;
+}
+
+static int qemu_rdma_get_fd(void *opaque)
+{
+    QEMUFileRDMA *rfile = opaque;
+    RDMAContext *rdma = rfile->rdma;
+
+    return rdma->comp_channel->fd;
+}
+
+const QEMUFileOps rdma_read_ops = {
+    .get_buffer    = qemu_rdma_get_buffer,
+    .get_fd        = qemu_rdma_get_fd,
+    .close         = qemu_rdma_close,
+    .hook_ram_load = qemu_rdma_registration_handle,
+};
+
+const QEMUFileOps rdma_write_ops = {
+    .put_buffer           = qemu_rdma_put_buffer,
+    .close                = qemu_rdma_close,
+    .before_ram_iterate   = qemu_rdma_registration_start,
+    .after_ram_iterate    = qemu_rdma_registration_stop,
+    .save_page            = qemu_rdma_save_page,
+};
+
+static void *qemu_fopen_rdma(RDMAContext *rdma, const char *mode)
+{
+    QEMUFileRDMA *r = g_malloc0(sizeof(QEMUFileRDMA));
+
+    if (qemu_file_mode_is_not_valid(mode)) {
+        return NULL;
+    }
+
+    r->rdma = rdma;
+
+    if (mode[0] == 'w') {
+        r->file = qemu_fopen_ops(r, &rdma_write_ops);
+    } else {
+        r->file = qemu_fopen_ops(r, &rdma_read_ops);
+    }
+
+    return r->file;
+}
+
+static void rdma_accept_incoming_migration(void *opaque)
+{
+    RDMAContext *rdma = opaque;
+    int ret;
+    QEMUFile *f;
+    Error *local_err = NULL, **errp = &local_err;
+
+    DPRINTF("Accepting rdma connection...\n");
+    ret = qemu_rdma_accept(rdma);
+
+    if (ret) {
+        ERROR(errp, "RDMA Migration initialization failed!\n");
+        goto err;
+    }
+
+    DPRINTF("Accepted migration\n");
+
+    f = qemu_fopen_rdma(rdma, "rb");
+    if (f == NULL) {
+        ERROR(errp, "could not qemu_fopen_rdma!\n");
+        goto err;
+    }
+
+    rdma->migration_started_on_destination = 1;
+    process_incoming_migration(f);
+    return;
+
+err:
+    qemu_rdma_cleanup(rdma);
+}
+
+void rdma_start_incoming_migration(const char *host_port, Error **errp)
+{
+    int ret;
+    RDMAContext *rdma;
+    Error *local_err = NULL;
+
+    DPRINTF("Starting RDMA-based incoming migration\n");
+    rdma = qemu_rdma_data_init(host_port, &local_err);
+
+    if (rdma == NULL) {
+        goto err;
+    }
+
+    ret = qemu_rdma_dest_init(rdma, &local_err);
+
+    if (ret) {
+        goto err;
+    }
+
+    DPRINTF("qemu_rdma_dest_init success\n");
+    ret = qemu_rdma_dest_prepare(rdma, &local_err);
+
+    if (ret) {
+        goto err;
+    }
+
+    DPRINTF("qemu_rdma_dest_prepare success\n");
+
+    qemu_set_fd_handler2(rdma->channel->fd, NULL,
+                         rdma_accept_incoming_migration, NULL,
+                            (void *)(intptr_t) rdma);
+    return;
+err:
+    error_propagate(errp, local_err);
+    g_free(rdma);
+}
+
+void rdma_start_outgoing_migration(void *opaque,
+                            const char *host_port, Error **errp)
+{
+    MigrationState *s = opaque;
+    Error *local_err = NULL, **temp = &local_err;
+    RDMAContext *rdma = qemu_rdma_data_init(host_port, &local_err);
+    int ret = 0;
+
+    if (rdma == NULL) {
+        ERROR(temp, "Failed to initialize RDMA data structures! %d\n", ret);
+        goto err;
+    }
+
+    ret = qemu_rdma_source_init(rdma, &local_err,
+        s->enabled_capabilities[MIGRATION_CAPABILITY_X_RDMA_PIN_ALL]);
+
+    if (ret) {
+        goto err;
+    }
+
+    DPRINTF("qemu_rdma_source_init success\n");
+    ret = qemu_rdma_connect(rdma, &local_err);
+
+    if (ret) {
+        goto err;
+    }
+
+    DPRINTF("qemu_rdma_source_connect success\n");
+
+    s->file = qemu_fopen_rdma(rdma, "wb");
+    s->total_time = qemu_get_clock_ms(rt_clock);
+    migrate_fd_connect(s);
+    return;
+err:
+    error_propagate(errp, local_err);
+    g_free(rdma);
+    migrate_fd_error(s);
+}
-- 
1.7.10.4

Re: [Qemu-devel] [PATCH v6 10/11] rdma: core logic

[Anthony Liguori]

mrhines@linux.vnet.ibm.com writes:

> From: "Michael R. Hines" <mrhines@us.ibm.com>
>
> Code that does need to be visible is kept
> well contained inside this file and this is the only
> new additional file to the entire patch - good progress.
>
> This file includes the entire protocol and interfaces
> required to perform RDMA migration.
>
> Also, the configure and Makefile modifications to link
> this file are included.
>
> Full documentation is in docs/rdma.txt
>
> Signed-off-by: Michael R. Hines <mrhines@us.ibm.com>
> ---
>  Makefile.objs                 |    1 +
>  configure                     |   29 +++++++++++++++++++++++++++++
>  include/migration/migration.h |    4 ++++
>  migration.c                   |    8 ++++++++
>  migration-rdma.c              | 2707 +++++++++++++++++++++++++++++++++++++++++++++
>  4 files changed, 2749 insertions(+)
>
> diff --git a/Makefile.objs b/Makefile.objs
> index a473348..d744827 100644
> --- a/Makefile.objs
> +++ b/Makefile.objs
> @@ -49,6 +49,7 @@ common-obj-$(CONFIG_POSIX) += os-posix.o
>  common-obj-$(CONFIG_LINUX) += fsdev/
>  
>  common-obj-y += migration.o migration-tcp.o
> +common-obj-$(CONFIG_RDMA) += migration-rdma.o
>  common-obj-y += qemu-char.o #aio.o
>  common-obj-y += block-migration.o
>  common-obj-y += page_cache.o xbzrle.o
> diff --git a/configure b/configure
> index 33d3354..90441cf 100755
> --- a/configure
> +++ b/configure
> @@ -181,6 +181,7 @@ xfs=""
>  vhost_net="no"
>  vhost_scsi="no"
>  kvm="no"
> +rdma="yes"

This should be:

rdma=""

> +if test "$rdma" != "no" ; then
> +  cat > $TMPC <<EOF
> +#include <rdma/rdma_cma.h>
> +int main(void) { return 0; }
> +EOF
> +  rdma_libs="-lrdmacm -libverbs"
> +  if compile_prog "-Werror" "$rdma_libs" ; then
> +    rdma="yes"
> +    libs_softmmu="$libs_softmmu $rdma_libs"
> +  else
> +    if test "$rdma" = "yes" ; then
> +      feature_not_found "rdma"

Otherwise if RDMA isn't available, you get:

[02:43 PM] anthony🐵 titi:~/git/qemu$ test-qemu.sh
M	roms/ipxe
M	roms/seabios
Switched to branch 'staging'

ERROR: User requested feature rdma
       configure was not able to find it

Regards,

Anthony Liguori

> +    fi
> +    rdma="no"
> +  fi
> +fi
> +
>  ##########################################
>  # VNC TLS/WS detection
>  if test "$vnc" = "yes" -a \( "$vnc_tls" != "no" -o "$vnc_ws" != "no" \) ; then
> @@ -3549,6 +3573,7 @@ echo "Linux AIO support $linux_aio"
>  echo "ATTR/XATTR support $attr"
>  echo "Install blobs     $blobs"
>  echo "KVM support       $kvm"
> +echo "RDMA support      $rdma"
>  echo "TCG interpreter   $tcg_interpreter"
>  echo "fdt support       $fdt"
>  echo "preadv support    $preadv"
> @@ -4515,6 +4540,10 @@ if [ "$pixman" = "internal" ]; then
>    echo "config-host.h: subdir-pixman" >> $config_host_mak
>  fi
>  
> +if test "$rdma" = "yes" ; then
> +echo "CONFIG_RDMA=y" >> $config_host_mak
> +fi
> +
>  if [ "$dtc_internal" = "yes" ]; then
>    echo "config-host.h: subdir-dtc" >> $config_host_mak
>  fi
> diff --git a/include/migration/migration.h b/include/migration/migration.h
> index b9b6c36..5acdc69 100644
> --- a/include/migration/migration.h
> +++ b/include/migration/migration.h
> @@ -77,6 +77,10 @@ void fd_start_incoming_migration(const char *path, Error **errp);
>  
>  void fd_start_outgoing_migration(MigrationState *s, const char *fdname, Error **errp);
>  
> +void rdma_start_outgoing_migration(void *opaque, const char *host_port, Error **errp);
> +
> +void rdma_start_incoming_migration(const char *host_port, Error **errp);
> +
>  void migrate_fd_error(MigrationState *s);
>  
>  void migrate_fd_connect(MigrationState *s);
> diff --git a/migration.c b/migration.c
> index 97ce6d8..42e43eb 100644
> --- a/migration.c
> +++ b/migration.c
> @@ -78,6 +78,10 @@ void qemu_start_incoming_migration(const char *uri, Error **errp)
>  
>      if (strstart(uri, "tcp:", &p))
>          tcp_start_incoming_migration(p, errp);
> +#ifdef CONFIG_RDMA
> +    else if (strstart(uri, "x-rdma:", &p))
> +        rdma_start_incoming_migration(p, errp);
> +#endif
>  #if !defined(WIN32)
>      else if (strstart(uri, "exec:", &p))
>          exec_start_incoming_migration(p, errp);
> @@ -408,6 +412,10 @@ void qmp_migrate(const char *uri, bool has_blk, bool blk,
>  
>      if (strstart(uri, "tcp:", &p)) {
>          tcp_start_outgoing_migration(s, p, &local_err);
> +#ifdef CONFIG_RDMA
> +    } else if (strstart(uri, "x-rdma:", &p)) {
> +        rdma_start_outgoing_migration(s, p, &local_err);
> +#endif
>  #if !defined(WIN32)
>      } else if (strstart(uri, "exec:", &p)) {
>          exec_start_outgoing_migration(s, p, &local_err);
> diff --git a/migration-rdma.c b/migration-rdma.c
> new file mode 100644
> index 0000000..8924e94
> --- /dev/null
> +++ b/migration-rdma.c
> @@ -0,0 +1,2707 @@
> +/*
> + * RDMA protocol and interfaces
> + *
> + * Copyright IBM, Corp. 2010-2013
> + *
> + * Authors:
> + *  Michael R. Hines <mrhines@us.ibm.com>
> + *  Jiuxing Liu <jl@us.ibm.com>
> + *
> + * This work is licensed under the terms of the GNU GPL, version 2 or
> + * later.  See the COPYING file in the top-level directory.
> + *
> + */
> +#include "qemu-common.h"
> +#include "migration/migration.h"
> +#include "migration/qemu-file.h"
> +#include "exec/cpu-common.h"
> +#include "qemu/main-loop.h"
> +#include "qemu/sockets.h"
> +#include "block/coroutine.h"
> +#include <stdio.h>
> +#include <sys/types.h>
> +#include <sys/socket.h>
> +#include <netdb.h>
> +#include <arpa/inet.h>
> +#include <string.h>
> +#include <rdma/rdma_cma.h>
> +
> +#define DEBUG_RDMA
> +//#define DEBUG_RDMA_VERBOSE
> +//#define DEBUG_RDMA_REALLY_VERBOSE
> +
> +#ifdef DEBUG_RDMA
> +#define DPRINTF(fmt, ...) \
> +    do { printf("rdma: " fmt, ## __VA_ARGS__); } while (0)
> +#else
> +#define DPRINTF(fmt, ...) \
> +    do { } while (0)
> +#endif
> +
> +#ifdef DEBUG_RDMA_VERBOSE
> +#define DDPRINTF(fmt, ...) \
> +    do { printf("rdma: " fmt, ## __VA_ARGS__); } while (0)
> +#else
> +#define DDPRINTF(fmt, ...) \
> +    do { } while (0)
> +#endif
> +
> +#ifdef DEBUG_RDMA_REALLY_VERBOSE
> +#define DDDPRINTF(fmt, ...) \
> +    do { printf("rdma: " fmt, ## __VA_ARGS__); } while (0)
> +#else
> +#define DDDPRINTF(fmt, ...) \
> +    do { } while (0)
> +#endif
> +
> +/*
> + * Print and error on both the Monitor and the Log file.
> + */
> +#define ERROR(errp, fmt, ...) \
> +    do { \
> +        fprintf(stderr, "RDMA ERROR: " fmt, ## __VA_ARGS__); \
> +        if (errp && (*(errp) == NULL)) { \
> +            error_setg(errp, "RDMA ERROR: " fmt, ## __VA_ARGS__); \
> +        } \
> +    } while (0)
> +
> +#define RDMA_RESOLVE_TIMEOUT_MS 10000
> +
> +/* Do not merge data if larger than this. */
> +#define RDMA_MERGE_MAX (4 * 1024 * 1024)
> +#define RDMA_UNSIGNALED_SEND_MAX 64
> +
> +#define RDMA_REG_CHUNK_SHIFT 20 /* 1 MB */
> +
> +#define RDMA_REG_CHUNK_SIZE (1UL << (RDMA_REG_CHUNK_SHIFT))
> +
> +/*
> + * This is only for non-live state being migrated.
> + * Instead of RDMA_WRITE messages, we use RDMA_SEND
> + * messages for that state, which requires a different
> + * delivery design than main memory.
> + */
> +#define RDMA_SEND_INCREMENT 32768
> +
> +/*
> + * Completion queue can be filled by both read and write work requests,
> + * so must reflect the sum of both possible queue sizes.
> + */
> +#define RDMA_QP_SIZE 1000
> +#define RDMA_CQ_SIZE (RDMA_QP_SIZE * 3)
> +
> +/*
> + * Maximum size infiniband SEND message
> + */
> +#define RDMA_CONTROL_MAX_BUFFER (512 * 1024)
> +#define RDMA_CONTROL_MAX_WR 2
> +#define RDMA_CONTROL_MAX_COMMANDS_PER_MESSAGE 4096
> +
> +/*
> + * Capabilities for negotiation.
> + */
> +#define RDMA_CAPABILITY_PIN_ALL 0x01
> +
> +/*
> + * Add the other flags above to this list of known capabilities
> + * as they are introduced.
> + */
> +static uint32_t known_capabilities = RDMA_CAPABILITY_PIN_ALL;
> +
> +#define CHECK_ERROR_STATE() \
> +    do { \
> +        if (rdma->error_state) { \
> +            if (!rdma->error_reported) { \
> +                fprintf(stderr, "RDMA is in an error state waiting migration" \
> +                                " to abort!\n"); \
> +                rdma->error_reported = 1; \
> +            } \
> +            return rdma->error_state; \
> +        } \
> +    } while (0);
> +/*
> + * RDMA migration protocol:
> + * 1. RDMA Writes (data messages, i.e. RAM)
> + * 2. IB Send/Recv (control channel messages)
> + */
> +enum {
> +    RDMA_WRID_NONE = 0,
> +    RDMA_WRID_RDMA_WRITE,
> +    RDMA_WRID_SEND_CONTROL = 1000,
> +    RDMA_WRID_RECV_CONTROL = 2000,
> +};
> +
> +const char *wrid_desc[] = {
> +        [RDMA_WRID_NONE] = "NONE",
> +        [RDMA_WRID_RDMA_WRITE] = "WRITE RDMA",
> +        [RDMA_WRID_SEND_CONTROL] = "CONTROL SEND",
> +        [RDMA_WRID_RECV_CONTROL] = "CONTROL RECV",
> +};
> +
> +/*
> + * SEND/RECV IB Control Messages.
> + */
> +enum {
> +    RDMA_CONTROL_NONE = 0,
> +    RDMA_CONTROL_ERROR,
> +    RDMA_CONTROL_READY,             /* ready to receive */
> +    RDMA_CONTROL_QEMU_FILE,         /* QEMUFile-transmitted bytes */
> +    RDMA_CONTROL_RAM_BLOCKS,        /* RAMBlock synchronization */
> +    RDMA_CONTROL_COMPRESS,          /* page contains repeat values */
> +    RDMA_CONTROL_REGISTER_REQUEST,  /* dynamic page registration */
> +    RDMA_CONTROL_REGISTER_RESULT,   /* key to use after registration */
> +    RDMA_CONTROL_REGISTER_FINISHED, /* current iteration finished */
> +};
> +
> +const char *control_desc[] = {
> +        [RDMA_CONTROL_NONE] = "NONE",
> +        [RDMA_CONTROL_ERROR] = "ERROR",
> +        [RDMA_CONTROL_READY] = "READY",
> +        [RDMA_CONTROL_QEMU_FILE] = "QEMU FILE",
> +        [RDMA_CONTROL_RAM_BLOCKS] = "REMOTE INFO",
> +        [RDMA_CONTROL_COMPRESS] = "COMPRESS",
> +        [RDMA_CONTROL_REGISTER_REQUEST] = "REGISTER REQUEST",
> +        [RDMA_CONTROL_REGISTER_RESULT] = "REGISTER RESULT",
> +        [RDMA_CONTROL_REGISTER_FINISHED] = "REGISTER FINISHED",
> +};
> +
> +/*
> + * Memory and MR structures used to represent an IB Send/Recv work request.
> + * This is *not* used for RDMA, only IB Send/Recv.
> + */
> +typedef struct {
> +    uint8_t  control[RDMA_CONTROL_MAX_BUFFER]; /* actual buffer to register */
> +    struct   ibv_mr *control_mr;               /* registration metadata */
> +    size_t   control_len;                      /* length of the message */
> +    uint8_t *control_curr;                     /* start of unconsumed bytes */
> +} RDMAWorkRequestData;
> +
> +/*
> + * Negotiate RDMA capabilities during connection-setup time.
> + */
> +typedef struct {
> +    uint32_t version;
> +    uint32_t flags;
> +} RDMACapabilities;
> +
> +static void caps_to_network(RDMACapabilities *cap)
> +{
> +    cap->version = htonl(cap->version);
> +    cap->flags = htonl(cap->flags);
> +}
> +
> +static void network_to_caps(RDMACapabilities *cap)
> +{
> +    cap->version = ntohl(cap->version);
> +    cap->flags = ntohl(cap->flags);
> +}
> +
> +/*
> + * Representation of a RAMBlock from an RDMA perspective.
> + * This is not transmitted, only local.
> + * This and subsequent structures cannot be linked lists
> + * because we're using a single IB message to transmit
> + * the information. It's small anyway, so a list is overkill.
> + */
> +typedef struct RDMALocalBlock {
> +    uint8_t  *local_host_addr; /* local virtual address */
> +    uint64_t remote_host_addr; /* remote virtual address */
> +    uint64_t offset;
> +    uint64_t length;
> +    struct   ibv_mr **pmr;     /* MRs for chunk-level registration */
> +    struct   ibv_mr *mr;       /* MR for non-chunk-level registration */
> +    uint32_t *remote_keys;     /* rkeys for chunk-level registration */
> +    uint32_t remote_rkey;      /* rkeys for non-chunk-level registration */
> +    int      index;            /* which block are we */
> +} RDMALocalBlock;
> +
> +/*
> + * Also represents a RAMblock, but only on the dest.
> + * This gets transmitted by the dest during connection-time
> + * to the source / primary VM and then is used to populate the
> + * corresponding RDMALocalBlock with
> + * the information needed to perform the actual RDMA.
> + */
> +typedef struct QEMU_PACKED RDMARemoteBlock {
> +    uint64_t remote_host_addr;
> +    uint64_t offset;
> +    uint64_t length;
> +    uint32_t remote_rkey;
> +    uint32_t padding;
> +} QEMU_PACKED RDMARemoteBlock;
> +
> +/*
> + * Virtual address of the above structures used for transmitting
> + * the RAMBlock descriptions at connection-time.
> + * This structure is *not* transmitted.
> + */
> +typedef struct RDMALocalBlocks {
> +    int num_blocks;
> +    RDMALocalBlock *block;
> +} RDMALocalBlocks;
> +
> +/*
> + * Main data structure for RDMA state.
> + * While there is only one copy of this structure being allocated right now,
> + * this is the place where one would start if you wanted to consider
> + * having more than one RDMA connection open at the same time.
> + */
> +typedef struct RDMAContext {
> +    char *host;
> +    int port;
> +
> +    /* This is used by the migration protocol to transmit
> +     * control messages (such as device state and registration commands)
> +     *
> +     * WR #0 is for control channel ready messages from the destination.
> +     * WR #1 is for control channel data messages from the destination.
> +     * WR #2 is for control channel send messages.
> +     *
> +     * We could use more WRs, but we have enough for now.
> +     */
> +    RDMAWorkRequestData wr_data[RDMA_CONTROL_MAX_WR + 1];
> +
> +    /*
> +     * This is used by *_exchange_send() to figure out whether or not
> +     * the initial "READY" message has already been received or not.
> +     * This is because other functions may potentially poll() and detect
> +     * the READY message before send() does, in which case we need to
> +     * know if it completed.
> +     */
> +    int control_ready_expected;
> +
> +    /* number of outstanding unsignaled send */
> +    int num_unsignaled_send;
> +
> +    /* number of outstanding signaled send */
> +    int num_signaled_send;
> +
> +    /* store info about current buffer so that we can
> +       merge it with future sends */
> +    uint64_t current_offset;
> +    uint64_t current_length;
> +    /* index of ram block the current buffer belongs to */
> +    int current_index;
> +    /* index of the chunk in the current ram block */
> +    int current_chunk;
> +
> +    bool pin_all;
> +
> +    /*
> +     * infiniband-specific variables for opening the device
> +     * and maintaining connection state and so forth.
> +     *
> +     * cm_id also has ibv_context, rdma_event_channel, and ibv_qp in
> +     * cm_id->verbs, cm_id->channel, and cm_id->qp.
> +     */
> +    struct rdma_cm_id *cm_id;               /* connection manager ID */
> +    struct rdma_cm_id *listen_id;
> +
> +    struct ibv_context *verbs;
> +    struct rdma_event_channel *channel;
> +    struct ibv_qp *qp;                      /* queue pair */
> +    struct ibv_comp_channel *comp_channel;  /* completion channel */
> +    struct ibv_pd *pd;                      /* protection domain */
> +    struct ibv_cq *cq;                      /* completion queue */
> +
> +    /*
> +     * If a previous write failed (perhaps because of a failed
> +     * memory registration, then do not attempt any future work
> +     * and remember the error state.
> +     */
> +    int error_state;
> +    int error_reported;
> +
> +    /*
> +     * Description of ram blocks used throughout the code.
> +     */
> +    RDMALocalBlocks local_ram_blocks;
> +    RDMARemoteBlock *block;
> +
> +    /*
> +     * Migration on *destination* started.
> +     * Then use coroutine yield function.
> +     * Source runs in a thread, so we don't care.
> +     */
> +    int migration_started_on_destination;
> +
> +    int total_registrations;
> +} RDMAContext;
> +
> +/*
> + * Interface to the rest of the migration call stack.
> + */
> +typedef struct QEMUFileRDMA {
> +    RDMAContext *rdma;
> +    size_t len;
> +    void *file;
> +} QEMUFileRDMA;
> +
> +#define RDMA_CONTROL_VERSION_CURRENT 1
> +
> +/*
> + * Main structure for IB Send/Recv control messages.
> + * This gets prepended at the beginning of every Send/Recv.
> + */
> +typedef struct QEMU_PACKED {
> +    uint32_t len;     /* Total length of data portion */
> +    uint32_t type;    /* which control command to perform */
> +    uint32_t repeat;  /* number of commands in data portion of same type */
> +    uint32_t padding;
> +} QEMU_PACKED RDMAControlHeader;
> +
> +static void control_to_network(RDMAControlHeader *control)
> +{
> +    control->type = htonl(control->type);
> +    control->len = htonl(control->len);
> +    control->repeat = htonl(control->repeat);
> +}
> +
> +static void network_to_control(RDMAControlHeader *control)
> +{
> +    control->type = ntohl(control->type);
> +    control->len = ntohl(control->len);
> +    control->repeat = ntohl(control->repeat);
> +}
> +
> +/*
> + * Register a single Chunk.
> + * Information sent by the primary VM to inform the dest
> + * to register an single chunk of memory before we can perform
> + * the actual RDMA operation.
> + */
> +typedef struct QEMU_PACKED {
> +    uint32_t len;           /* length of the chunk to be registered */
> +    uint32_t current_index; /* which ramblock the chunk belongs to */
> +    uint64_t offset;        /* offset into the ramblock of the chunk */
> +} QEMU_PACKED RDMARegister;
> +
> +typedef struct QEMU_PACKED {
> +    uint32_t value;     /* if zero, we will madvise() */
> +    uint32_t block_idx; /* which ram block index */
> +    uint64_t offset;    /* where in the remote ramblock this chunk */
> +    uint64_t length;    /* length of the chunk */
> +} QEMU_PACKED RDMACompress;
> +
> +/*
> + * The result of the dest's memory registration produces an "rkey"
> + * which the primary VM must reference in order to perform
> + * the RDMA operation.
> + */
> +typedef struct QEMU_PACKED {
> +    uint32_t rkey;
> +    uint32_t padding;
> +} QEMU_PACKED RDMARegisterResult;
> +
> +static inline uint64_t ram_chunk_index(uint8_t *start, uint8_t *host)
> +{
> +    return ((uintptr_t) host - (uintptr_t) start) >> RDMA_REG_CHUNK_SHIFT;
> +}
> +
> +static inline uint64_t ram_chunk_count(RDMALocalBlock *rdma_ram_block)
> +{
> +    return ram_chunk_index(rdma_ram_block->local_host_addr,
> +        rdma_ram_block->local_host_addr + rdma_ram_block->length) + 1UL;
> +}
> +
> +static inline uint8_t *ram_chunk_start(RDMALocalBlock *rdma_ram_block,
> +                                       uint64_t i)
> +{
> +    return (uint8_t *) (((uintptr_t) rdma_ram_block->local_host_addr)
> +                                    + (i << RDMA_REG_CHUNK_SHIFT));
> +}
> +
> +static inline uint8_t *ram_chunk_end(RDMALocalBlock *rdma_ram_block, uint64_t i)
> +{
> +    uint8_t *result = ram_chunk_start(rdma_ram_block, i) + RDMA_REG_CHUNK_SIZE;
> +
> +    if (result > (rdma_ram_block->local_host_addr + rdma_ram_block->length)) {
> +        result = rdma_ram_block->local_host_addr + rdma_ram_block->length;
> +    }
> +
> +    return result;
> +}
> +
> +
> +/*
> + * Memory regions need to be registered with the device and queue pairs setup
> + * in advanced before the migration starts. This tells us where the RAM blocks
> + * are so that we can register them individually.
> + */
> +static void qemu_rdma_init_one_block(void *host_addr,
> +    ram_addr_t offset, ram_addr_t length, void *opaque)
> +{
> +    RDMALocalBlocks *rdma_local_ram_blocks = opaque;
> +    int num_blocks = rdma_local_ram_blocks->num_blocks;
> +
> +    rdma_local_ram_blocks->block[num_blocks].local_host_addr = host_addr;
> +    rdma_local_ram_blocks->block[num_blocks].offset = (uint64_t)offset;
> +    rdma_local_ram_blocks->block[num_blocks].length = (uint64_t)length;
> +    rdma_local_ram_blocks->block[num_blocks].index = num_blocks;
> +
> +    DPRINTF("Block: %d, addr: %" PRIu64 ", offset: %" PRIu64
> +           " length: %" PRIu64 " end: %" PRIu64 "\n",
> +            num_blocks, (uint64_t) host_addr, offset, length,
> +            (uint64_t) (host_addr + length));
> +
> +    rdma_local_ram_blocks->num_blocks++;
> +
> +}
> +
> +static void qemu_rdma_ram_block_counter(void *host_addr,
> +            ram_addr_t offset, ram_addr_t length, void *opaque)
> +{
> +    int *num_blocks = opaque;
> +    *num_blocks = *num_blocks + 1;
> +}
> +
> +/*
> + * Identify the RAMBlocks and their quantity. They will be references to
> + * identify chunk boundaries inside each RAMBlock and also be referenced
> + * during dynamic page registration.
> + */
> +static int qemu_rdma_init_ram_blocks(RDMALocalBlocks *rdma_local_ram_blocks)
> +{
> +    int num_blocks = 0;
> +
> +    qemu_ram_foreach_block(qemu_rdma_ram_block_counter, &num_blocks);
> +
> +    memset(rdma_local_ram_blocks, 0, sizeof *rdma_local_ram_blocks);
> +    rdma_local_ram_blocks->block = g_malloc0(sizeof(RDMALocalBlock) *
> +                                    num_blocks);
> +
> +    rdma_local_ram_blocks->num_blocks = 0;
> +    qemu_ram_foreach_block(qemu_rdma_init_one_block, rdma_local_ram_blocks);
> +
> +    DPRINTF("Allocated %d local ram block structures\n",
> +                    rdma_local_ram_blocks->num_blocks);
> +    return 0;
> +}
> +
> +/*
> + * Put in the log file which RDMA device was opened and the details
> + * associated with that device.
> + */
> +static void qemu_rdma_dump_id(const char *who, struct ibv_context *verbs)
> +{
> +    printf("%s RDMA Device opened: kernel name %s "
> +           "uverbs device name %s, "
> +           "infiniband_verbs class device path %s,"
> +           " infiniband class device path %s\n",
> +                who,
> +                verbs->device->name,
> +                verbs->device->dev_name,
> +                verbs->device->dev_path,
> +                verbs->device->ibdev_path);
> +}
> +
> +/*
> + * Put in the log file the RDMA gid addressing information,
> + * useful for folks who have trouble understanding the
> + * RDMA device hierarchy in the kernel.
> + */
> +static void qemu_rdma_dump_gid(const char *who, struct rdma_cm_id *id)
> +{
> +    char sgid[33];
> +    char dgid[33];
> +    inet_ntop(AF_INET6, &id->route.addr.addr.ibaddr.sgid, sgid, sizeof sgid);
> +    inet_ntop(AF_INET6, &id->route.addr.addr.ibaddr.dgid, dgid, sizeof dgid);
> +    DPRINTF("%s Source GID: %s, Dest GID: %s\n", who, sgid, dgid);
> +}
> +
> +/*
> + * Figure out which RDMA device corresponds to the requested IP hostname
> + * Also create the initial connection manager identifiers for opening
> + * the connection.
> + */
> +static int qemu_rdma_resolve_host(RDMAContext *rdma, Error **errp)
> +{
> +    int ret;
> +    struct addrinfo *res;
> +    char port_str[16];
> +    struct rdma_cm_event *cm_event;
> +    char ip[40] = "unknown";
> +
> +    if (rdma->host == NULL || !strcmp(rdma->host, "")) {
> +        ERROR(errp, "RDMA hostname has not been set\n");
> +        return -1;
> +    }
> +
> +    /* create CM channel */
> +    rdma->channel = rdma_create_event_channel();
> +    if (!rdma->channel) {
> +        ERROR(errp, "could not create CM channel\n");
> +        return -1;
> +    }
> +
> +    /* create CM id */
> +    ret = rdma_create_id(rdma->channel, &rdma->cm_id, NULL, RDMA_PS_TCP);
> +    if (ret) {
> +        ERROR(errp, "could not create channel id\n");
> +        goto err_resolve_create_id;
> +    }
> +
> +    snprintf(port_str, 16, "%d", rdma->port);
> +    port_str[15] = '\0';
> +
> +    ret = getaddrinfo(rdma->host, port_str, NULL, &res);
> +    if (ret < 0) {
> +        ERROR(errp, "could not getaddrinfo address %s\n", rdma->host);
> +        goto err_resolve_get_addr;
> +    }
> +
> +    inet_ntop(AF_INET, &((struct sockaddr_in *) res->ai_addr)->sin_addr,
> +                                ip, sizeof ip);
> +    DPRINTF("%s => %s\n", rdma->host, ip);
> +
> +    /* resolve the first address */
> +    ret = rdma_resolve_addr(rdma->cm_id, NULL, res->ai_addr,
> +            RDMA_RESOLVE_TIMEOUT_MS);
> +    if (ret) {
> +        ERROR(errp, "could not resolve address %s\n", rdma->host);
> +        goto err_resolve_get_addr;
> +    }
> +
> +    qemu_rdma_dump_gid("source_resolve_addr", rdma->cm_id);
> +
> +    ret = rdma_get_cm_event(rdma->channel, &cm_event);
> +    if (ret) {
> +        ERROR(errp, "could not perform event_addr_resolved\n");
> +        goto err_resolve_get_addr;
> +    }
> +
> +    if (cm_event->event != RDMA_CM_EVENT_ADDR_RESOLVED) {
> +        ERROR(errp, "result not equal to event_addr_resolved %s\n",
> +                rdma_event_str(cm_event->event));
> +        perror("rdma_resolve_addr");
> +        goto err_resolve_get_addr;
> +    }
> +    rdma_ack_cm_event(cm_event);
> +
> +    /* resolve route */
> +    ret = rdma_resolve_route(rdma->cm_id, RDMA_RESOLVE_TIMEOUT_MS);
> +    if (ret) {
> +        ERROR(errp, "could not resolve rdma route\n");
> +        goto err_resolve_get_addr;
> +    }
> +
> +    ret = rdma_get_cm_event(rdma->channel, &cm_event);
> +    if (ret) {
> +        ERROR(errp, "could not perform event_route_resolved\n");
> +        goto err_resolve_get_addr;
> +    }
> +    if (cm_event->event != RDMA_CM_EVENT_ROUTE_RESOLVED) {
> +        ERROR(errp, "result not equal to event_route_resolved: %s\n",
> +                        rdma_event_str(cm_event->event));
> +        rdma_ack_cm_event(cm_event);
> +        goto err_resolve_get_addr;
> +    }
> +    rdma_ack_cm_event(cm_event);
> +    rdma->verbs = rdma->cm_id->verbs;
> +    qemu_rdma_dump_id("source_resolve_host", rdma->cm_id->verbs);
> +    qemu_rdma_dump_gid("source_resolve_host", rdma->cm_id);
> +    return 0;
> +
> +err_resolve_get_addr:
> +    rdma_destroy_id(rdma->cm_id);
> +    rdma->cm_id = 0;
> +err_resolve_create_id:
> +    rdma_destroy_event_channel(rdma->channel);
> +    rdma->channel = NULL;
> +
> +    return -1;
> +}
> +
> +/*
> + * Create protection domain and completion queues
> + */
> +static int qemu_rdma_alloc_pd_cq(RDMAContext *rdma)
> +{
> +    /* allocate pd */
> +    rdma->pd = ibv_alloc_pd(rdma->verbs);
> +    if (!rdma->pd) {
> +        fprintf(stderr, "failed to allocate protection domain\n");
> +        return -1;
> +    }
> +
> +    /* create completion channel */
> +    rdma->comp_channel = ibv_create_comp_channel(rdma->verbs);
> +    if (!rdma->comp_channel) {
> +        fprintf(stderr, "failed to allocate completion channel\n");
> +        goto err_alloc_pd_cq;
> +    }
> +
> +    /* create cq */
> +    rdma->cq = ibv_create_cq(rdma->verbs, RDMA_CQ_SIZE,
> +            NULL, rdma->comp_channel, 0);
> +    if (!rdma->cq) {
> +        fprintf(stderr, "failed to allocate completion queue\n");
> +        goto err_alloc_pd_cq;
> +    }
> +
> +    return 0;
> +
> +err_alloc_pd_cq:
> +    if (rdma->pd) {
> +        ibv_dealloc_pd(rdma->pd);
> +    }
> +    if (rdma->comp_channel) {
> +        ibv_destroy_comp_channel(rdma->comp_channel);
> +    }
> +    rdma->pd = NULL;
> +    rdma->comp_channel = NULL;
> +    return -1;
> +
> +}
> +
> +/*
> + * Create queue pairs.
> + */
> +static int qemu_rdma_alloc_qp(RDMAContext *rdma)
> +{
> +    struct ibv_qp_init_attr attr = { 0 };
> +    int ret;
> +
> +    attr.cap.max_send_wr = RDMA_QP_SIZE;
> +    attr.cap.max_recv_wr = 3;
> +    attr.cap.max_send_sge = 1;
> +    attr.cap.max_recv_sge = 1;
> +    attr.send_cq = rdma->cq;
> +    attr.recv_cq = rdma->cq;
> +    attr.qp_type = IBV_QPT_RC;
> +
> +    ret = rdma_create_qp(rdma->cm_id, rdma->pd, &attr);
> +    if (ret) {
> +        return -1;
> +    }
> +
> +    rdma->qp = rdma->cm_id->qp;
> +    return 0;
> +}
> +
> +static int qemu_rdma_reg_whole_ram_blocks(RDMAContext *rdma,
> +                                RDMALocalBlocks *rdma_local_ram_blocks)
> +{
> +    int i;
> +    for (i = 0; i < rdma_local_ram_blocks->num_blocks; i++) {
> +        rdma_local_ram_blocks->block[i].mr =
> +            ibv_reg_mr(rdma->pd,
> +                    rdma_local_ram_blocks->block[i].local_host_addr,
> +                    rdma_local_ram_blocks->block[i].length,
> +                    IBV_ACCESS_LOCAL_WRITE |
> +                    IBV_ACCESS_REMOTE_WRITE
> +                    );
> +        if (!rdma_local_ram_blocks->block[i].mr) {
> +            perror("Failed to register local dest ram block!\n");
> +            break;
> +        }
> +        rdma->total_registrations++;
> +    }
> +
> +    if (i >= rdma_local_ram_blocks->num_blocks) {
> +        return 0;
> +    }
> +
> +    for (i--; i >= 0; i--) {
> +        ibv_dereg_mr(rdma_local_ram_blocks->block[i].mr);
> +        rdma->total_registrations--;
> +    }
> +
> +    return -1;
> +
> +}
> +
> +/*
> + * Shutdown and clean things up.
> + */
> +static void qemu_rdma_dereg_ram_blocks(RDMAContext *rdma,
> +                                       RDMALocalBlocks *rdma_local_ram_blocks)
> +{
> +    int i, j;
> +    for (i = 0; i < rdma_local_ram_blocks->num_blocks; i++) {
> +        int num_chunks;
> +        if (!rdma_local_ram_blocks->block[i].pmr) {
> +            continue;
> +        }
> +        num_chunks = ram_chunk_count(&(rdma_local_ram_blocks->block[i]));
> +        for (j = 0; j < num_chunks; j++) {
> +            if (!rdma_local_ram_blocks->block[i].pmr[j]) {
> +                continue;
> +            }
> +            ibv_dereg_mr(rdma_local_ram_blocks->block[i].pmr[j]);
> +            rdma->total_registrations--;
> +        }
> +        g_free(rdma_local_ram_blocks->block[i].pmr);
> +        rdma_local_ram_blocks->block[i].pmr = NULL;
> +    }
> +    for (i = 0; i < rdma_local_ram_blocks->num_blocks; i++) {
> +        if (!rdma_local_ram_blocks->block[i].mr) {
> +            continue;
> +        }
> +        ibv_dereg_mr(rdma_local_ram_blocks->block[i].mr);
> +        rdma->total_registrations--;
> +        rdma_local_ram_blocks->block[i].mr = NULL;
> +    }
> +}
> +
> +/*
> + * The protocol uses two different sets of rkeys (mutually exclusive):
> + * 1. One key to represent the virtual address of the entire ram block.
> + *    (dynamic chunk registration disabled - pin everything with one rkey.)
> + * 2. One to represent individual chunks within a ram block.
> + *    (dynamic chunk registration enabled - pin individual chunks.)
> + *
> + * Once the capability is successfully negotiated, the destination transmits
> + * the keys to use (or sends them later) including the virtual addresses
> + * and then propagates the remote ram block descriptions to his local copy.
> + */
> +static int qemu_rdma_process_remote_blocks(RDMAContext *rdma, int num_blocks)
> +{
> +    RDMALocalBlocks *local = &rdma->local_ram_blocks;
> +    int i, j;
> +
> +    if (local->num_blocks != num_blocks) {
> +        fprintf(stderr, "local %d != remote %d\n",
> +            local->num_blocks, num_blocks);
> +        return -1;
> +    }
> +
> +    for (i = 0; i < num_blocks; i++) {
> +        /* search local ram blocks */
> +        for (j = 0; j < local->num_blocks; j++) {
> +            if (rdma->block[i].offset != local->block[j].offset) {
> +                continue;
> +            }
> +            if (rdma->block[i].length != local->block[j].length) {
> +                return -1;
> +            }
> +            local->block[j].remote_host_addr =
> +                rdma->block[i].remote_host_addr;
> +            local->block[j].remote_rkey = rdma->block[i].remote_rkey;
> +            break;
> +        }
> +        if (j >= local->num_blocks) {
> +            return -1;
> +        }
> +    }
> +
> +    return 0;
> +}
> +
> +/*
> + * Find the ram block that corresponds to the page requested to be
> + * transmitted by QEMU.
> + *
> + * Once the block is found, also identify which 'chunk' within that
> + * block that the page belongs to.
> + *
> + * This search cannot fail or the migration will fail.
> + */
> +static int qemu_rdma_search_ram_block(uint64_t offset, uint64_t length,
> +        RDMALocalBlocks *blocks, int *block_index, int *chunk_index)
> +{
> +    int i;
> +    uint8_t *host_addr;
> +
> +    for (i = 0; i < blocks->num_blocks; i++) {
> +        if (offset < blocks->block[i].offset) {
> +            continue;
> +        }
> +        if (offset + length >
> +                blocks->block[i].offset + blocks->block[i].length) {
> +            continue;
> +        }
> +
> +        *block_index = i;
> +        host_addr = blocks->block[i].local_host_addr +
> +                (offset - blocks->block[i].offset);
> +        *chunk_index = ram_chunk_index(blocks->block[i].local_host_addr,
> +                        host_addr);
> +        return 0;
> +    }
> +    return -1;
> +}
> +
> +/*
> + * Register a chunk with IB. If the chunk was already registered
> + * previously, then skip.
> + *
> + * Also return the keys associated with the registration needed
> + * to perform the actual RDMA operation.
> + */
> +static int qemu_rdma_register_and_get_keys(RDMAContext *rdma,
> +        RDMALocalBlock *block, uint8_t * host_addr,
> +        uint32_t *lkey, uint32_t *rkey)
> +{
> +    int chunk;
> +    if (block->mr) {
> +        if (lkey) {
> +            *lkey = block->mr->lkey;
> +        }
> +        if (rkey) {
> +            *rkey = block->mr->rkey;
> +        }
> +        return 0;
> +    }
> +
> +    /* allocate memory to store chunk MRs */
> +    if (!block->pmr) {
> +        int num_chunks = ram_chunk_count(block);
> +        block->pmr = g_malloc0(num_chunks *
> +                sizeof(struct ibv_mr *));
> +        if (!block->pmr) {
> +            return -1;
> +        }
> +    }
> +
> +    /*
> +     * If 'rkey', then we're the destination, so grant access to the source.
> +     *
> +     * If 'lkey', then we're the primary VM, so grant access only to ourselves.
> +     */
> +    chunk = ram_chunk_index(block->local_host_addr, host_addr);
> +    if (!block->pmr[chunk]) {
> +        uint8_t *start_addr = ram_chunk_start(block, chunk);
> +        uint8_t *end_addr = ram_chunk_end(block, chunk);
> +
> +        block->pmr[chunk] = ibv_reg_mr(rdma->pd,
> +                start_addr,
> +                end_addr - start_addr,
> +                (rkey ? (IBV_ACCESS_LOCAL_WRITE |
> +                        IBV_ACCESS_REMOTE_WRITE) : 0));
> +
> +        if (!block->pmr[chunk]) {
> +            perror("Failed to register chunk!");
> +            fprintf(stderr, "Chunk details: block: %d chunk index %d"
> +                            " start %" PRIu64 " end %" PRIu64 " host %" PRIu64
> +                            " local %" PRIu64 " registrations: %d\n",
> +                            block->index, chunk, (uint64_t) start_addr,
> +                            (uint64_t) end_addr, (uint64_t) host_addr,
> +                            (uint64_t) block->local_host_addr,
> +                            rdma->total_registrations);
> +            return -1;
> +        }
> +        rdma->total_registrations++;
> +    }
> +
> +    if (lkey) {
> +        *lkey = block->pmr[chunk]->lkey;
> +    }
> +    if (rkey) {
> +        *rkey = block->pmr[chunk]->rkey;
> +    }
> +    return 0;
> +}
> +
> +/*
> + * Register (at connection time) the memory used for control
> + * channel messages.
> + */
> +static int qemu_rdma_reg_control(RDMAContext *rdma, int idx)
> +{
> +    rdma->wr_data[idx].control_mr = ibv_reg_mr(rdma->pd,
> +            rdma->wr_data[idx].control, RDMA_CONTROL_MAX_BUFFER,
> +            IBV_ACCESS_LOCAL_WRITE | IBV_ACCESS_REMOTE_WRITE);
> +    if (rdma->wr_data[idx].control_mr) {
> +        rdma->total_registrations++;
> +        return 0;
> +    }
> +    fprintf(stderr, "qemu_rdma_reg_control failed!\n");
> +    return -1;
> +}
> +
> +static int qemu_rdma_dereg_control(RDMAContext *rdma, int idx)
> +{
> +    rdma->total_registrations--;
> +    return ibv_dereg_mr(rdma->wr_data[idx].control_mr);
> +}
> +
> +#if defined(DEBUG_RDMA) && defined(DEBUG_RDMA_VERBOSE)
> +static const char *print_wrid(int wrid)
> +{
> +    if (wrid >= RDMA_WRID_RECV_CONTROL) {
> +        return wrid_desc[RDMA_WRID_RECV_CONTROL];
> +    }
> +    return wrid_desc[wrid];
> +}
> +#endif
> +
> +/*
> + * Consult the connection manager to see a work request
> + * (of any kind) has completed.
> + * Return the work request ID that completed.
> + */
> +static int qemu_rdma_poll(RDMAContext *rdma)
> +{
> +    int ret;
> +    struct ibv_wc wc;
> +
> +    ret = ibv_poll_cq(rdma->cq, 1, &wc);
> +    if (!ret) {
> +        return RDMA_WRID_NONE;
> +    }
> +    if (ret < 0) {
> +        fprintf(stderr, "ibv_poll_cq return %d!\n", ret);
> +        return ret;
> +    }
> +    if (wc.status != IBV_WC_SUCCESS) {
> +        fprintf(stderr, "ibv_poll_cq wc.status=%d %s!\n",
> +                        wc.status, ibv_wc_status_str(wc.status));
> +        fprintf(stderr, "ibv_poll_cq wrid=%s!\n", wrid_desc[wc.wr_id]);
> +
> +        return -1;
> +    }
> +
> +    if (rdma->control_ready_expected &&
> +        (wc.wr_id >= RDMA_WRID_RECV_CONTROL)) {
> +        DDDPRINTF("completion %s #%" PRId64 " received (%" PRId64 ")\n",
> +            wrid_desc[RDMA_WRID_RECV_CONTROL], wc.wr_id -
> +            RDMA_WRID_RECV_CONTROL, wc.wr_id);
> +        rdma->control_ready_expected = 0;
> +    }
> +
> +    if (wc.wr_id == RDMA_WRID_RDMA_WRITE) {
> +        rdma->num_signaled_send--;
> +        DDDPRINTF("completions %s (%" PRId64 ") left %d\n",
> +            print_wrid(wc.wr_id), wc.wr_id, rdma->num_signaled_send);
> +    } else {
> +        DDDPRINTF("other completion %s (%" PRId64 ") received left %d\n",
> +            print_wrid(wc.wr_id), wc.wr_id, rdma->num_signaled_send);
> +    }
> +
> +    return  (int)wc.wr_id;
> +}
> +
> +/*
> + * Block until the next work request has completed.
> + *
> + * First poll to see if a work request has already completed,
> + * otherwise block.
> + *
> + * If we encounter completed work requests for IDs other than
> + * the one we're interested in, then that's generally an error.
> + *
> + * The only exception is actual RDMA Write completions. These
> + * completions only need to be recorded, but do not actually
> + * need further processing.
> + */
> +static int qemu_rdma_block_for_wrid(RDMAContext *rdma, int wrid)
> +{
> +    int num_cq_events = 0;
> +    int r = RDMA_WRID_NONE;
> +    struct ibv_cq *cq;
> +    void *cq_ctx;
> +
> +    if (ibv_req_notify_cq(rdma->cq, 0)) {
> +        return -1;
> +    }
> +    /* poll cq first */
> +    while (r != wrid) {
> +        r = qemu_rdma_poll(rdma);
> +        if (r < 0) {
> +            return r;
> +        }
> +        if (r == RDMA_WRID_NONE) {
> +            break;
> +        }
> +        if (r != wrid) {
> +            DDDPRINTF("A Wanted wrid %s (%d) but got %s (%d)\n",
> +                print_wrid(wrid), wrid, print_wrid(r), r);
> +        }
> +    }
> +    if (r == wrid) {
> +        return 0;
> +    }
> +
> +    while (1) {
> +        /*
> +         * Coroutine doesn't start until process_incoming_migration()
> +         * so don't yield unless we know we're running inside of a coroutine.
> +         */
> +        if (rdma->migration_started_on_destination) {
> +            yield_until_fd_readable(rdma->comp_channel->fd);
> +        }
> +
> +        if (ibv_get_cq_event(rdma->comp_channel, &cq, &cq_ctx)) {
> +            perror("ibv_get_cq_event");
> +            goto err_block_for_wrid;
> +        }
> +
> +        num_cq_events++;
> +
> +        if (ibv_req_notify_cq(cq, 0)) {
> +            goto err_block_for_wrid;
> +        }
> +        /* poll cq */
> +        while (r != wrid) {
> +            r = qemu_rdma_poll(rdma);
> +            if (r < 0) {
> +                goto err_block_for_wrid;
> +            }
> +            if (r == RDMA_WRID_NONE) {
> +                break;
> +            }
> +            if (r != wrid) {
> +                DDDPRINTF("B Wanted wrid %s (%d) but got %s (%d)\n",
> +                    print_wrid(wrid), wrid, print_wrid(r), r);
> +            }
> +        }
> +        if (r == wrid) {
> +            goto success_block_for_wrid;
> +        }
> +    }
> +
> +success_block_for_wrid:
> +    if (num_cq_events) {
> +        ibv_ack_cq_events(cq, num_cq_events);
> +    }
> +    return 0;
> +
> +err_block_for_wrid:
> +    if (num_cq_events) {
> +        ibv_ack_cq_events(cq, num_cq_events);
> +    }
> +    return -1;
> +}
> +
> +/*
> + * Post a SEND message work request for the control channel
> + * containing some data and block until the post completes.
> + */
> +static int qemu_rdma_post_send_control(RDMAContext *rdma, uint8_t *buf,
> +                                       RDMAControlHeader *head)
> +{
> +    int ret = 0;
> +    RDMAWorkRequestData *wr = &rdma->wr_data[RDMA_CONTROL_MAX_WR];
> +    struct ibv_send_wr *bad_wr;
> +    struct ibv_sge sge = {
> +                           .addr = (uint64_t)(wr->control),
> +                           .length = head->len + sizeof(RDMAControlHeader),
> +                           .lkey = wr->control_mr->lkey,
> +                         };
> +    struct ibv_send_wr send_wr = {
> +                                   .wr_id = RDMA_WRID_SEND_CONTROL,
> +                                   .opcode = IBV_WR_SEND,
> +                                   .send_flags = IBV_SEND_SIGNALED,
> +                                   .sg_list = &sge,
> +                                   .num_sge = 1,
> +                                };
> +
> +    DDDPRINTF("CONTROL: sending %s..\n", control_desc[head->type]);
> +
> +    /*
> +     * We don't actually need to do a memcpy() in here if we used
> +     * the "sge" properly, but since we're only sending control messages
> +     * (not RAM in a performance-critical path), then its OK for now.
> +     *
> +     * The copy makes the RDMAControlHeader simpler to manipulate
> +     * for the time being.
> +     */
> +    memcpy(wr->control, head, sizeof(RDMAControlHeader));
> +    control_to_network((void *) wr->control);
> +
> +    if (buf) {
> +        memcpy(wr->control + sizeof(RDMAControlHeader), buf, head->len);
> +    }
> +
> +
> +    if (ibv_post_send(rdma->qp, &send_wr, &bad_wr)) {
> +        return -1;
> +    }
> +
> +    if (ret < 0) {
> +        fprintf(stderr, "Failed to use post IB SEND for control!\n");
> +        return ret;
> +    }
> +
> +    ret = qemu_rdma_block_for_wrid(rdma, RDMA_WRID_SEND_CONTROL);
> +    if (ret < 0) {
> +        fprintf(stderr, "rdma migration: send polling control error!\n");
> +    }
> +
> +    return ret;
> +}
> +
> +/*
> + * Post a RECV work request in anticipation of some future receipt
> + * of data on the control channel.
> + */
> +static int qemu_rdma_post_recv_control(RDMAContext *rdma, int idx)
> +{
> +    struct ibv_recv_wr *bad_wr;
> +    struct ibv_sge sge = {
> +                            .addr = (uint64_t)(rdma->wr_data[idx].control),
> +                            .length = RDMA_CONTROL_MAX_BUFFER,
> +                            .lkey = rdma->wr_data[idx].control_mr->lkey,
> +                         };
> +
> +    struct ibv_recv_wr recv_wr = {
> +                                    .wr_id = RDMA_WRID_RECV_CONTROL + idx,
> +                                    .sg_list = &sge,
> +                                    .num_sge = 1,
> +                                 };
> +
> +
> +    if (ibv_post_recv(rdma->qp, &recv_wr, &bad_wr)) {
> +        return -1;
> +    }
> +
> +    return 0;
> +}
> +
> +/*
> + * Block and wait for a RECV control channel message to arrive.
> + */
> +static int qemu_rdma_exchange_get_response(RDMAContext *rdma,
> +                RDMAControlHeader *head, int expecting, int idx)
> +{
> +    int ret = qemu_rdma_block_for_wrid(rdma, RDMA_WRID_RECV_CONTROL + idx);
> +
> +    if (ret < 0) {
> +        fprintf(stderr, "rdma migration: recv polling control error!\n");
> +        return ret;
> +    }
> +
> +    network_to_control((void *) rdma->wr_data[idx].control);
> +    memcpy(head, rdma->wr_data[idx].control, sizeof(RDMAControlHeader));
> +
> +    DDDPRINTF("CONTROL: %s received\n", control_desc[expecting]);
> +
> +    if ((expecting != RDMA_CONTROL_NONE && head->type != expecting)
> +            || head->type == RDMA_CONTROL_ERROR) {
> +        fprintf(stderr, "Was expecting a %s (%d) control message"
> +                ", but got: %s (%d), length: %d\n",
> +                control_desc[expecting], expecting,
> +                control_desc[head->type], head->type, head->len);
> +        return -EIO;
> +    }
> +
> +    return 0;
> +}
> +
> +/*
> + * When a RECV work request has completed, the work request's
> + * buffer is pointed at the header.
> + *
> + * This will advance the pointer to the data portion
> + * of the control message of the work request's buffer that
> + * was populated after the work request finished.
> + */
> +static void qemu_rdma_move_header(RDMAContext *rdma, int idx,
> +                                  RDMAControlHeader *head)
> +{
> +    rdma->wr_data[idx].control_len = head->len;
> +    rdma->wr_data[idx].control_curr =
> +        rdma->wr_data[idx].control + sizeof(RDMAControlHeader);
> +}
> +
> +/*
> + * This is an 'atomic' high-level operation to deliver a single, unified
> + * control-channel message.
> + *
> + * Additionally, if the user is expecting some kind of reply to this message,
> + * they can request a 'resp' response message be filled in by posting an
> + * additional work request on behalf of the user and waiting for an additional
> + * completion.
> + *
> + * The extra (optional) response is used during registration to us from having
> + * to perform an *additional* exchange of message just to provide a response by
> + * instead piggy-backing on the acknowledgement.
> + */
> +static int qemu_rdma_exchange_send(RDMAContext *rdma, RDMAControlHeader *head,
> +                                   uint8_t *data, RDMAControlHeader *resp,
> +                                   int *resp_idx)
> +{
> +    int ret = 0;
> +    int idx = 0;
> +
> +    /*
> +     * Wait until the dest is ready before attempting to deliver the message
> +     * by waiting for a READY message.
> +     */
> +    if (rdma->control_ready_expected) {
> +        RDMAControlHeader resp;
> +        ret = qemu_rdma_exchange_get_response(rdma,
> +                                    &resp, RDMA_CONTROL_READY, idx);
> +        if (ret < 0) {
> +            return ret;
> +        }
> +    }
> +
> +    /*
> +     * If the user is expecting a response, post a WR in anticipation of it.
> +     */
> +    if (resp) {
> +        ret = qemu_rdma_post_recv_control(rdma, idx + 1);
> +        if (ret) {
> +            fprintf(stderr, "rdma migration: error posting"
> +                    " extra control recv for anticipated result!");
> +            return ret;
> +        }
> +    }
> +
> +    /*
> +     * Post a WR to replace the one we just consumed for the READY message.
> +     */
> +    ret = qemu_rdma_post_recv_control(rdma, idx);
> +    if (ret) {
> +        fprintf(stderr, "rdma migration: error posting first control recv!");
> +        return ret;
> +    }
> +
> +    /*
> +     * Deliver the control message that was requested.
> +     */
> +    ret = qemu_rdma_post_send_control(rdma, data, head);
> +
> +    if (ret < 0) {
> +        fprintf(stderr, "Failed to send control buffer!\n");
> +        return ret;
> +    }
> +
> +    /*
> +     * If we're expecting a response, block and wait for it.
> +     */
> +    if (resp) {
> +        DDPRINTF("Waiting for response %s\n", control_desc[resp->type]);
> +        ret = qemu_rdma_exchange_get_response(rdma, resp, resp->type, idx + 1);
> +
> +        if (ret < 0) {
> +            return ret;
> +        }
> +
> +        qemu_rdma_move_header(rdma, idx + 1, resp);
> +        *resp_idx = idx + 1;
> +        DDPRINTF("Response %s received.\n", control_desc[resp->type]);
> +    }
> +
> +    rdma->control_ready_expected = 1;
> +
> +    return 0;
> +}
> +
> +/*
> + * This is an 'atomic' high-level operation to receive a single, unified
> + * control-channel message.
> + */
> +static int qemu_rdma_exchange_recv(RDMAContext *rdma, RDMAControlHeader *head,
> +                                int expecting)
> +{
> +    RDMAControlHeader ready = {
> +                                .len = 0,
> +                                .type = RDMA_CONTROL_READY,
> +                                .repeat = 1,
> +                              };
> +    int ret;
> +    int idx = 0;
> +
> +    /*
> +     * Inform the source that we're ready to receive a message.
> +     */
> +    ret = qemu_rdma_post_send_control(rdma, NULL, &ready);
> +
> +    if (ret < 0) {
> +        fprintf(stderr, "Failed to send control buffer!\n");
> +        return ret;
> +    }
> +
> +    /*
> +     * Block and wait for the message.
> +     */
> +    ret = qemu_rdma_exchange_get_response(rdma, head, expecting, idx);
> +
> +    if (ret < 0) {
> +        return ret;
> +    }
> +
> +    qemu_rdma_move_header(rdma, idx, head);
> +
> +    /*
> +     * Post a new RECV work request to replace the one we just consumed.
> +     */
> +    ret = qemu_rdma_post_recv_control(rdma, idx);
> +    if (ret) {
> +        fprintf(stderr, "rdma migration: error posting second control recv!");
> +        return ret;
> +    }
> +
> +    return 0;
> +}
> +
> +/*
> + * Write an actual chunk of memory using RDMA.
> + *
> + * If we're using dynamic registration on the dest-side, we have to
> + * send a registration command first.
> + */
> +static int qemu_rdma_write_one(QEMUFile *f, RDMAContext *rdma,
> +        int current_index,
> +        uint64_t offset, uint64_t length,
> +        uint64_t wr_id, enum ibv_send_flags flag)
> +{
> +    struct ibv_sge sge;
> +    struct ibv_send_wr send_wr = { 0 };
> +    struct ibv_send_wr *bad_wr;
> +    RDMALocalBlock *block = &(rdma->local_ram_blocks.block[current_index]);
> +    int chunk;
> +    RDMARegister reg;
> +    RDMARegisterResult *reg_result;
> +    int reg_result_idx;
> +    RDMAControlHeader resp = { .type = RDMA_CONTROL_REGISTER_RESULT };
> +    RDMAControlHeader head = { .len = sizeof(RDMARegister),
> +                               .type = RDMA_CONTROL_REGISTER_REQUEST,
> +                               .repeat = 1,
> +                             };
> +    int ret;
> +
> +    sge.addr = (uint64_t)(block->local_host_addr + (offset - block->offset));
> +    sge.length = length;
> +
> +    if (!rdma->pin_all) {
> +        chunk = ram_chunk_index(block->local_host_addr, (uint8_t *) sge.addr);
> +        if (!block->remote_keys[chunk]) {
> +            /*
> +             * This page has not yet been registered, so first check to see
> +             * if the entire chunk is zero. If so, tell the other size to
> +             * memset() + madvise() the entire chunk without RDMA.
> +             */
> +            if (can_use_buffer_find_nonzero_offset((void *)sge.addr, length)
> +                   && buffer_find_nonzero_offset((void *)sge.addr,
> +                                                    length) == length) {
> +                RDMACompress comp = {
> +                                        .offset = offset,
> +                                        .value = 0,
> +                                        .block_idx = current_index,
> +                                        .length = length,
> +                                    };
> +
> +                head.len = sizeof(comp);
> +                head.type = RDMA_CONTROL_COMPRESS;
> +
> +                DDPRINTF("Entire chunk is zero, sending compress: %d for %d "
> +                    "bytes, index: %d, offset: %" PRId64 "...\n",
> +                    chunk, sge.length, current_index, offset);
> +
> +                ret = qemu_rdma_exchange_send(rdma, &head,
> +                                (uint8_t *) &comp, NULL, NULL);
> +
> +                if (ret < 0) {
> +                    return -EIO;
> +                }
> +
> +                return 1;
> +            }
> +
> +            /*
> +             * Otherwise, tell other side to register.
> +             */
> +            reg.len = sge.length;
> +            reg.current_index = current_index;
> +            reg.offset = offset;
> +
> +            DDPRINTF("Sending registration request chunk %d for %d "
> +                    "bytes, index: %d, offset: %" PRId64 "...\n",
> +                    chunk, sge.length, current_index, offset);
> +
> +            ret = qemu_rdma_exchange_send(rdma, &head, (uint8_t *) &reg,
> +                                    &resp, &reg_result_idx);
> +            if (ret < 0) {
> +                return ret;
> +            }
> +
> +            /* try to overlap this single registration with the one we sent. */
> +            if (qemu_rdma_register_and_get_keys(rdma, block,
> +                                                (uint8_t *) sge.addr,
> +                                                &sge.lkey, NULL)) {
> +                fprintf(stderr, "cannot get lkey!\n");
> +                return -EINVAL;
> +            }
> +
> +            reg_result = (RDMARegisterResult *)
> +                    rdma->wr_data[reg_result_idx].control_curr;
> +
> +            DDPRINTF("Received registration result:"
> +                    " my key: %x their key %x, chunk %d\n",
> +                    block->remote_keys[chunk], reg_result->rkey, chunk);
> +
> +            block->remote_keys[chunk] = reg_result->rkey;
> +        } else {
> +            /* already registered before */
> +            if (qemu_rdma_register_and_get_keys(rdma, block,
> +                                                (uint8_t *)sge.addr,
> +                                                &sge.lkey, NULL)) {
> +                fprintf(stderr, "cannot get lkey!\n");
> +                return -EINVAL;
> +            }
> +        }
> +
> +        send_wr.wr.rdma.rkey = block->remote_keys[chunk];
> +    } else {
> +        send_wr.wr.rdma.rkey = block->remote_rkey;
> +
> +        if (qemu_rdma_register_and_get_keys(rdma, block, (uint8_t *)sge.addr,
> +                                                     &sge.lkey, NULL)) {
> +            fprintf(stderr, "cannot get lkey!\n");
> +            return -EINVAL;
> +        }
> +    }
> +
> +
> +    send_wr.wr_id = wr_id;
> +    send_wr.opcode = IBV_WR_RDMA_WRITE;
> +    send_wr.send_flags = flag;
> +    send_wr.sg_list = &sge;
> +    send_wr.num_sge = 1;
> +    send_wr.wr.rdma.remote_addr = block->remote_host_addr +
> +                                    (offset - block->offset);
> +
> +    return ibv_post_send(rdma->qp, &send_wr, &bad_wr);
> +}
> +
> +/*
> + * Push out any unwritten RDMA operations.
> + *
> + * We support sending out multiple chunks at the same time.
> + * Not all of them need to get signaled in the completion queue.
> + */
> +static int qemu_rdma_write_flush(QEMUFile *f, RDMAContext *rdma)
> +{
> +    int ret;
> +    enum ibv_send_flags flags = 0;
> +
> +    if (!rdma->current_length) {
> +        return 0;
> +    }
> +    if (rdma->num_unsignaled_send >=
> +            RDMA_UNSIGNALED_SEND_MAX) {
> +        flags = IBV_SEND_SIGNALED;
> +    }
> +
> +retry:
> +    ret = qemu_rdma_write_one(f, rdma,
> +            rdma->current_index,
> +            rdma->current_offset,
> +            rdma->current_length,
> +            RDMA_WRID_RDMA_WRITE, flags);
> +
> +    if (ret < 0) {
> +        if (ret == -ENOMEM) {
> +            DDPRINTF("send queue is full. wait a little....\n");
> +            ret = qemu_rdma_block_for_wrid(rdma, RDMA_WRID_RDMA_WRITE);
> +            if (ret >= 0) {
> +                goto retry;
> +            }
> +            if (ret < 0) {
> +                fprintf(stderr, "rdma migration: failed to make "
> +                                "room in full send queue! %d\n", ret);
> +                return ret;
> +            }
> +        }
> +        perror("write flush error");
> +        return ret;
> +    }
> +
> +    if (ret == 0) {
> +        if (rdma->num_unsignaled_send >=
> +                RDMA_UNSIGNALED_SEND_MAX) {
> +            rdma->num_unsignaled_send = 0;
> +            rdma->num_signaled_send++;
> +            DDDPRINTF("signaled total: %d\n", rdma->num_signaled_send);
> +        } else {
> +            rdma->num_unsignaled_send++;
> +        }
> +    }
> +
> +    rdma->current_length = 0;
> +    rdma->current_offset = 0;
> +
> +    return 0;
> +}
> +
> +static inline int qemu_rdma_in_current_block(RDMAContext *rdma,
> +                uint64_t offset, uint64_t len)
> +{
> +    RDMALocalBlock *block =
> +        &(rdma->local_ram_blocks.block[rdma->current_index]);
> +    if (rdma->current_index < 0) {
> +        return 0;
> +    }
> +    if (offset < block->offset) {
> +        return 0;
> +    }
> +    if (offset + len > block->offset + block->length) {
> +        return 0;
> +    }
> +    return 1;
> +}
> +
> +static inline int qemu_rdma_in_current_chunk(RDMAContext *rdma,
> +                uint64_t offset, uint64_t len)
> +{
> +    RDMALocalBlock *block =
> +            &(rdma->local_ram_blocks.block[rdma->current_index]);
> +    uint8_t *chunk_end, *host_addr;
> +
> +    if (rdma->current_chunk < 0) {
> +        return 0;
> +    }
> +
> +    host_addr = block->local_host_addr + (offset - block->offset);
> +    chunk_end = ram_chunk_end(block, rdma->current_chunk);
> +
> +    if ((host_addr + len) > chunk_end) {
> +        return 0;
> +    }
> +
> +    return 1;
> +}
> +
> +static inline int qemu_rdma_buffer_mergable(RDMAContext *rdma,
> +                    uint64_t offset, uint64_t len)
> +{
> +    if (rdma->current_length == 0) {
> +        return 0;
> +    }
> +    if (offset != rdma->current_offset + rdma->current_length) {
> +        return 0;
> +    }
> +    if (!qemu_rdma_in_current_block(rdma, offset, len)) {
> +        return 0;
> +    }
> +    if (!qemu_rdma_in_current_chunk(rdma, offset, len)) {
> +        return 0;
> +    }
> +    return 1;
> +}
> +
> +/*
> + * We're not actually writing here, but doing three things:
> + *
> + * 1. Identify the chunk the buffer belongs to.
> + * 2. If the chunk is full or the buffer doesn't belong to the current
> + *    chunk, then start a new chunk and flush() the old chunk.
> + * 3. To keep the hardware busy, we also group chunks into batches
> + *    and only require that a batch gets acknowledged in the completion
> + *    qeueue instead of each individual chunk.
> + */
> +static int qemu_rdma_write(QEMUFile *f, RDMAContext *rdma,
> +                           uint64_t offset, uint64_t len)
> +{
> +    int index = rdma->current_index;
> +    int chunk_index = rdma->current_chunk;
> +    int ret;
> +
> +    /* If we cannot merge it, we flush the current buffer first. */
> +    if (!qemu_rdma_buffer_mergable(rdma, offset, len)) {
> +        ret = qemu_rdma_write_flush(f, rdma);
> +        if (ret) {
> +            return ret;
> +        }
> +        rdma->current_length = 0;
> +        rdma->current_offset = offset;
> +
> +        ret = qemu_rdma_search_ram_block(offset, len,
> +                    &rdma->local_ram_blocks, &index, &chunk_index);
> +        if (ret) {
> +            fprintf(stderr, "ram block search failed\n");
> +            return ret;
> +        }
> +        rdma->current_index = index;
> +        rdma->current_chunk = chunk_index;
> +    }
> +
> +    /* merge it */
> +    rdma->current_length += len;
> +
> +    /* flush it if buffer is too large */
> +    if (rdma->current_length >= RDMA_MERGE_MAX) {
> +        return qemu_rdma_write_flush(f, rdma);
> +    }
> +
> +    return 0;
> +}
> +
> +static void qemu_rdma_cleanup(RDMAContext *rdma)
> +{
> +    struct rdma_cm_event *cm_event;
> +    int ret, idx;
> +
> +    if (rdma->cm_id) {
> +        if (rdma->error_state) {
> +            RDMAControlHeader head = { .len = 0,
> +                                       .type = RDMA_CONTROL_ERROR,
> +                                       .repeat = 1,
> +                                     };
> +            fprintf(stderr, "Early error. Sending error.\n");
> +            qemu_rdma_post_send_control(rdma, NULL, &head);
> +        }
> +
> +        ret = rdma_disconnect(rdma->cm_id);
> +        if (!ret) {
> +            DDPRINTF("waiting for disconnect\n");
> +            ret = rdma_get_cm_event(rdma->channel, &cm_event);
> +            if (!ret) {
> +                rdma_ack_cm_event(cm_event);
> +            }
> +        }
> +        DDPRINTF("Disconnected.\n");
> +        rdma->cm_id = 0;
> +    }
> +
> +    g_free(rdma->block);
> +    rdma->block = NULL;
> +
> +    for (idx = 0; idx < (RDMA_CONTROL_MAX_WR + 1); idx++) {
> +        if (rdma->wr_data[idx].control_mr) {
> +            qemu_rdma_dereg_control(rdma, idx);
> +        }
> +        rdma->wr_data[idx].control_mr = NULL;
> +    }
> +
> +    if (rdma->local_ram_blocks.block) {
> +        qemu_rdma_dereg_ram_blocks(rdma, &rdma->local_ram_blocks);
> +
> +        if (!rdma->pin_all) {
> +            for (idx = 0; idx < rdma->local_ram_blocks.num_blocks; idx++) {
> +                RDMALocalBlock *block = &(rdma->local_ram_blocks.block[idx]);
> +                g_free(block->remote_keys);
> +                block->remote_keys = NULL;
> +            }
> +        }
> +        g_free(rdma->local_ram_blocks.block);
> +        rdma->local_ram_blocks.block = NULL;
> +    }
> +
> +    if (rdma->qp) {
> +        ibv_destroy_qp(rdma->qp);
> +        rdma->qp = NULL;
> +    }
> +    if (rdma->cq) {
> +        ibv_destroy_cq(rdma->cq);
> +        rdma->cq = NULL;
> +    }
> +    if (rdma->comp_channel) {
> +        ibv_destroy_comp_channel(rdma->comp_channel);
> +        rdma->comp_channel = NULL;
> +    }
> +    if (rdma->pd) {
> +        ibv_dealloc_pd(rdma->pd);
> +        rdma->pd = NULL;
> +    }
> +    if (rdma->listen_id) {
> +        rdma_destroy_id(rdma->listen_id);
> +        rdma->listen_id = 0;
> +    }
> +    if (rdma->cm_id) {
> +        rdma_destroy_id(rdma->cm_id);
> +        rdma->cm_id = 0;
> +    }
> +    if (rdma->channel) {
> +        rdma_destroy_event_channel(rdma->channel);
> +        rdma->channel = NULL;
> +    }
> +}
> +
> +
> +static int qemu_rdma_source_init(RDMAContext *rdma, Error **errp, bool pin_all)
> +{
> +    int ret, idx;
> +    Error *local_err = NULL, **temp = &local_err;
> +
> +    /*
> +     * Will be validated against destination's actual capabilities
> +     * after the connect() completes.
> +     */
> +    rdma->pin_all = pin_all;
> +
> +    ret = qemu_rdma_resolve_host(rdma, temp);
> +    if (ret) {
> +        goto err_rdma_source_init;
> +    }
> +
> +    ret = qemu_rdma_alloc_pd_cq(rdma);
> +    if (ret) {
> +        ERROR(temp, "rdma migration: error allocating pd and cq! Your mlock()"
> +                    " limits may be too low. Please check $ ulimit -a # and "
> +                    "search for 'ulimit -l' in the output\n");
> +        goto err_rdma_source_init;
> +    }
> +
> +    ret = qemu_rdma_alloc_qp(rdma);
> +    if (ret) {
> +        ERROR(temp, "rdma migration: error allocating qp!\n");
> +        goto err_rdma_source_init;
> +    }
> +
> +    ret = qemu_rdma_init_ram_blocks(&rdma->local_ram_blocks);
> +    if (ret) {
> +        ERROR(temp, "rdma migration: error initializing ram blocks!\n");
> +        goto err_rdma_source_init;
> +    }
> +
> +    for (idx = 0; idx < (RDMA_CONTROL_MAX_WR + 1); idx++) {
> +        ret = qemu_rdma_reg_control(rdma, idx);
> +        if (ret) {
> +            ERROR(temp, "rdma migration: error registering %d control!\n",
> +                                                            idx);
> +            goto err_rdma_source_init;
> +        }
> +    }
> +
> +    rdma->block = (RDMARemoteBlock *) g_malloc0(sizeof(RDMARemoteBlock) *
> +                        rdma->local_ram_blocks.num_blocks);
> +    return 0;
> +
> +err_rdma_source_init:
> +    error_propagate(errp, local_err);
> +    qemu_rdma_cleanup(rdma);
> +    return -1;
> +}
> +
> +static int qemu_rdma_connect(RDMAContext *rdma, Error **errp)
> +{
> +    RDMAControlHeader head;
> +    RDMACapabilities cap = {
> +                                .version = RDMA_CONTROL_VERSION_CURRENT,
> +                                .flags = 0,
> +                           };
> +    struct rdma_conn_param conn_param = { .initiator_depth = 2,
> +                                          .retry_count = 5,
> +                                          .private_data = &cap,
> +                                          .private_data_len = sizeof(cap),
> +                                        };
> +    struct rdma_cm_event *cm_event;
> +    int ret;
> +    int idx = 0;
> +    int x;
> +
> +    /*
> +     * Only negotiate the capability with destination if the user
> +     * on the source first requested the capability.
> +     */
> +    if (rdma->pin_all) {
> +        DPRINTF("Server pin-all memory requested.\n");
> +        cap.flags |= RDMA_CAPABILITY_PIN_ALL;
> +    }
> +
> +    caps_to_network(&cap);
> +
> +    ret = rdma_connect(rdma->cm_id, &conn_param);
> +    if (ret) {
> +        perror("rdma_connect");
> +        ERROR(errp, "connecting to destination!\n");
> +        rdma_destroy_id(rdma->cm_id);
> +        rdma->cm_id = 0;
> +        goto err_rdma_source_connect;
> +    }
> +
> +    ret = rdma_get_cm_event(rdma->channel, &cm_event);
> +    if (ret) {
> +        perror("rdma_get_cm_event after rdma_connect");
> +        ERROR(errp, "connecting to destination!\n");
> +        rdma_ack_cm_event(cm_event);
> +        rdma_destroy_id(rdma->cm_id);
> +        rdma->cm_id = 0;
> +        goto err_rdma_source_connect;
> +    }
> +
> +    if (cm_event->event != RDMA_CM_EVENT_ESTABLISHED) {
> +        perror("rdma_get_cm_event != EVENT_ESTABLISHED after rdma_connect");
> +        ERROR(errp, "connecting to destination!\n");
> +        rdma_ack_cm_event(cm_event);
> +        rdma_destroy_id(rdma->cm_id);
> +        rdma->cm_id = 0;
> +        goto err_rdma_source_connect;
> +    }
> +
> +    memcpy(&cap, cm_event->param.conn.private_data, sizeof(cap));
> +    network_to_caps(&cap);
> +
> +    /*
> +     * Verify that the *requested* capabilities are supported by the destination
> +     * and disable them otherwise.
> +     */
> +    if (rdma->pin_all && !(cap.flags & RDMA_CAPABILITY_PIN_ALL)) {
> +        ERROR(errp, "Server cannot support pinning all memory. "
> +                        "Will register memory dynamically.\n");
> +        rdma->pin_all = false;
> +    }
> +
> +    DPRINTF("Pin all memory: %s\n", rdma->pin_all ? "enabled" : "disabled");
> +
> +    rdma_ack_cm_event(cm_event);
> +
> +    ret = qemu_rdma_post_recv_control(rdma, idx + 1);
> +    if (ret) {
> +        ERROR(errp, "posting first control recv!\n");
> +        goto err_rdma_source_connect;
> +    }
> +
> +    ret = qemu_rdma_post_recv_control(rdma, idx);
> +    if (ret) {
> +        ERROR(errp, "posting second control recv!\n");
> +        goto err_rdma_source_connect;
> +    }
> +
> +    ret = qemu_rdma_exchange_get_response(rdma,
> +                            &head, RDMA_CONTROL_RAM_BLOCKS, idx + 1);
> +
> +    if (ret < 0) {
> +        ERROR(errp, "receiving remote info!\n");
> +        goto err_rdma_source_connect;
> +    }
> +
> +    qemu_rdma_move_header(rdma, idx + 1, &head);
> +    memcpy(rdma->block, rdma->wr_data[idx + 1].control_curr, head.len);
> +
> +    ret = qemu_rdma_process_remote_blocks(rdma,
> +                        (head.len / sizeof(RDMARemoteBlock)));
> +    if (ret) {
> +        ERROR(errp, "processing remote ram blocks!\n");
> +        goto err_rdma_source_connect;
> +    }
> +
> +    if (!rdma->pin_all) {
> +        for (x = 0; x < rdma->local_ram_blocks.num_blocks; x++) {
> +            RDMALocalBlock *block = &(rdma->local_ram_blocks.block[x]);
> +            int num_chunks = ram_chunk_count(block);
> +            /* allocate memory to store remote rkeys */
> +            block->remote_keys = g_malloc0(num_chunks * sizeof(uint32_t));
> +        }
> +    }
> +
> +    rdma->control_ready_expected = 1;
> +    rdma->num_signaled_send = 0;
> +    return 0;
> +
> +err_rdma_source_connect:
> +    qemu_rdma_cleanup(rdma);
> +    return -1;
> +}
> +
> +static int qemu_rdma_dest_init(RDMAContext *rdma, Error **errp)
> +{
> +    int ret = -EINVAL, idx;
> +    struct sockaddr_in sin;
> +    struct rdma_cm_id *listen_id;
> +    char ip[40] = "unknown";
> +
> +    for (idx = 0; idx < RDMA_CONTROL_MAX_WR; idx++) {
> +        rdma->wr_data[idx].control_len = 0;
> +        rdma->wr_data[idx].control_curr = NULL;
> +    }
> +
> +    if (rdma->host == NULL) {
> +        ERROR(errp, "RDMA host is not set!\n");
> +        rdma->error_state = -EINVAL;
> +        return -1;
> +    }
> +    /* create CM channel */
> +    rdma->channel = rdma_create_event_channel();
> +    if (!rdma->channel) {
> +        ERROR(errp, "could not create rdma event channel\n");
> +        rdma->error_state = -EINVAL;
> +        return -1;
> +    }
> +
> +    /* create CM id */
> +    ret = rdma_create_id(rdma->channel, &listen_id, NULL, RDMA_PS_TCP);
> +    if (ret) {
> +        ERROR(errp, "could not create cm_id!\n");
> +        goto err_dest_init_create_listen_id;
> +    }
> +
> +    memset(&sin, 0, sizeof(sin));
> +    sin.sin_family = AF_INET;
> +    sin.sin_port = htons(rdma->port);
> +
> +    if (rdma->host && strcmp("", rdma->host)) {
> +        struct hostent *dest_addr;
> +        dest_addr = gethostbyname(rdma->host);
> +        if (!dest_addr) {
> +            ERROR(errp, "migration could not gethostbyname!\n");
> +            ret = -EINVAL;
> +            goto err_dest_init_bind_addr;
> +        }
> +        memcpy(&sin.sin_addr.s_addr, dest_addr->h_addr,
> +                dest_addr->h_length);
> +        inet_ntop(AF_INET, dest_addr->h_addr, ip, sizeof ip);
> +    } else {
> +        sin.sin_addr.s_addr = INADDR_ANY;
> +    }
> +
> +    DPRINTF("%s => %s\n", rdma->host, ip);
> +
> +    ret = rdma_bind_addr(listen_id, (struct sockaddr *)&sin);
> +    if (ret) {
> +        ERROR(errp, "Error: could not rdma_bind_addr!\n");
> +        goto err_dest_init_bind_addr;
> +    }
> +
> +    rdma->listen_id = listen_id;
> +    if (listen_id->verbs) {
> +        rdma->verbs = listen_id->verbs;
> +    }
> +    qemu_rdma_dump_id("dest_init", rdma->verbs);
> +    qemu_rdma_dump_gid("dest_init", listen_id);
> +    return 0;
> +
> +err_dest_init_bind_addr:
> +    rdma_destroy_id(listen_id);
> +err_dest_init_create_listen_id:
> +    rdma_destroy_event_channel(rdma->channel);
> +    rdma->channel = NULL;
> +    rdma->error_state = ret;
> +    return ret;
> +
> +}
> +
> +static int qemu_rdma_dest_prepare(RDMAContext *rdma, Error **errp)
> +{
> +    int ret;
> +    int idx;
> +
> +    if (!rdma->verbs) {
> +        ERROR(errp, "no verbs context!\n");
> +        return 0;
> +    }
> +
> +    ret = qemu_rdma_alloc_pd_cq(rdma);
> +    if (ret) {
> +        ERROR(errp, "allocating pd and cq!\n");
> +        goto err_rdma_dest_prepare;
> +    }
> +
> +    ret = qemu_rdma_init_ram_blocks(&rdma->local_ram_blocks);
> +    if (ret) {
> +        ERROR(errp, "initializing ram blocks!\n");
> +        goto err_rdma_dest_prepare;
> +    }
> +
> +    rdma->block = (RDMARemoteBlock *) g_malloc0(sizeof(RDMARemoteBlock) *
> +                        rdma->local_ram_blocks.num_blocks);
> +
> +    /* Extra one for the send buffer */
> +    for (idx = 0; idx < (RDMA_CONTROL_MAX_WR + 1); idx++) {
> +        ret = qemu_rdma_reg_control(rdma, idx);
> +        if (ret) {
> +            ERROR(errp, "registering %d control!\n", idx);
> +            goto err_rdma_dest_prepare;
> +        }
> +    }
> +
> +    ret = rdma_listen(rdma->listen_id, 5);
> +    if (ret) {
> +        ERROR(errp, "listening on socket!\n");
> +        goto err_rdma_dest_prepare;
> +    }
> +
> +    return 0;
> +
> +err_rdma_dest_prepare:
> +    qemu_rdma_cleanup(rdma);
> +    return -1;
> +}
> +
> +static void *qemu_rdma_data_init(const char *host_port, Error **errp)
> +{
> +    RDMAContext *rdma = NULL;
> +    InetSocketAddress *addr;
> +
> +    if (host_port) {
> +        rdma = g_malloc0(sizeof(RDMAContext));
> +        memset(rdma, 0, sizeof(RDMAContext));
> +        rdma->current_index = -1;
> +        rdma->current_chunk = -1;
> +
> +        addr = inet_parse(host_port, NULL);
> +        if (addr != NULL) {
> +            rdma->port = atoi(addr->port);
> +            rdma->host = g_strdup(addr->host);
> +        } else {
> +            ERROR(errp, "bad RDMA migration address '%s'", host_port);
> +            g_free(rdma);
> +            return NULL;
> +        }
> +    }
> +
> +    return rdma;
> +}
> +
> +/*
> + * QEMUFile interface to the control channel.
> + * SEND messages for control only.
> + * pc.ram is handled with regular RDMA messages.
> + */
> +static int qemu_rdma_put_buffer(void *opaque, const uint8_t *buf,
> +                                int64_t pos, int size)
> +{
> +    QEMUFileRDMA *r = opaque;
> +    QEMUFile *f = r->file;
> +    RDMAContext *rdma = r->rdma;
> +    size_t remaining = size;
> +    uint8_t * data = (void *) buf;
> +    int ret;
> +
> +    CHECK_ERROR_STATE();
> +
> +    /*
> +     * Push out any writes that
> +     * we're queued up for pc.ram.
> +     */
> +    if (qemu_rdma_write_flush(f, rdma) < 0) {
> +        rdma->error_state = -EIO;
> +        return rdma->error_state;
> +    }
> +
> +    while (remaining) {
> +        RDMAControlHeader head;
> +
> +        r->len = MIN(remaining, RDMA_SEND_INCREMENT);
> +        remaining -= r->len;
> +
> +        head.len = r->len;
> +        head.type = RDMA_CONTROL_QEMU_FILE;
> +
> +        ret = qemu_rdma_exchange_send(rdma, &head, data, NULL, NULL);
> +
> +        if (ret < 0) {
> +            rdma->error_state = ret;
> +            return ret;
> +        }
> +
> +        data += r->len;
> +    }
> +
> +    return size;
> +}
> +
> +static size_t qemu_rdma_fill(RDMAContext *rdma, uint8_t *buf,
> +                             int size, int idx)
> +{
> +    size_t len = 0;
> +
> +    if (rdma->wr_data[idx].control_len) {
> +        DDDPRINTF("RDMA %" PRId64 " of %d bytes already in buffer\n",
> +                    rdma->wr_data[idx].control_len, size);
> +
> +        len = MIN(size, rdma->wr_data[idx].control_len);
> +        memcpy(buf, rdma->wr_data[idx].control_curr, len);
> +        rdma->wr_data[idx].control_curr += len;
> +        rdma->wr_data[idx].control_len -= len;
> +    }
> +
> +    return len;
> +}
> +
> +/*
> + * QEMUFile interface to the control channel.
> + * RDMA links don't use bytestreams, so we have to
> + * return bytes to QEMUFile opportunistically.
> + */
> +static int qemu_rdma_get_buffer(void *opaque, uint8_t *buf,
> +                                int64_t pos, int size)
> +{
> +    QEMUFileRDMA *r = opaque;
> +    RDMAContext *rdma = r->rdma;
> +    RDMAControlHeader head;
> +    int ret = 0;
> +
> +    CHECK_ERROR_STATE();
> +
> +    /*
> +     * First, we hold on to the last SEND message we
> +     * were given and dish out the bytes until we run
> +     * out of bytes.
> +     */
> +    r->len = qemu_rdma_fill(r->rdma, buf, size, 0);
> +    if (r->len) {
> +        return r->len;
> +    }
> +
> +    /*
> +     * Once we run out, we block and wait for another
> +     * SEND message to arrive.
> +     */
> +    ret = qemu_rdma_exchange_recv(rdma, &head, RDMA_CONTROL_QEMU_FILE);
> +
> +    if (ret < 0) {
> +        rdma->error_state = ret;
> +        return ret;
> +    }
> +
> +    /*
> +     * SEND was received with new bytes, now try again.
> +     */
> +    return qemu_rdma_fill(r->rdma, buf, size, 0);
> +}
> +
> +/*
> + * Block until all the outstanding chunks have been delivered by the hardware.
> + */
> +static int qemu_rdma_drain_cq(QEMUFile *f, RDMAContext *rdma)
> +{
> +    int ret;
> +
> +    if (qemu_rdma_write_flush(f, rdma) < 0) {
> +        return -EIO;
> +    }
> +
> +    while (rdma->num_signaled_send) {
> +        ret = qemu_rdma_block_for_wrid(rdma, RDMA_WRID_RDMA_WRITE);
> +        if (ret < 0) {
> +            fprintf(stderr, "rdma migration: complete polling error!\n");
> +            return -EIO;
> +        }
> +    }
> +
> +    return 0;
> +}
> +
> +static int qemu_rdma_close(void *opaque)
> +{
> +    QEMUFileRDMA *r = opaque;
> +    if (r->rdma) {
> +        qemu_rdma_cleanup(r->rdma);
> +        g_free(r->rdma);
> +    }
> +    g_free(r);
> +    return 0;
> +}
> +
> +static size_t qemu_rdma_save_page(QEMUFile *f, void *opaque,
> +                   ram_addr_t block_offset, ram_addr_t offset, size_t size)
> +{
> +    ram_addr_t current_addr = block_offset + offset;
> +    QEMUFileRDMA *rfile = opaque;
> +    RDMAContext *rdma = rfile->rdma;
> +    int ret;
> +
> +    CHECK_ERROR_STATE();
> +
> +    qemu_fflush(f);
> +
> +    /*
> +     * Add this page to the current 'chunk'. If the chunk
> +     * is full, or the page doen't belong to the current chunk,
> +     * an actual RDMA write will occur and a new chunk will be formed.
> +     */
> +    ret = qemu_rdma_write(f, rdma, current_addr, size);
> +    if (ret < 0) {
> +        rdma->error_state = ret;
> +        fprintf(stderr, "rdma migration: write error! %d\n", ret);
> +        return ret;
> +    }
> +
> +    /*
> +     * Drain the Completion Queue if possible, but do not block,
> +     * just poll.
> +     *
> +     * If nothing to poll, the end of the iteration will do this
> +     * again to make sure we don't overflow the request queue.
> +     */
> +    while (1) {
> +        int ret = qemu_rdma_poll(rdma);
> +        if (ret == RDMA_WRID_NONE) {
> +            break;
> +        }
> +        if (ret < 0) {
> +            rdma->error_state = ret;
> +            fprintf(stderr, "rdma migration: polling error! %d\n", ret);
> +            return ret;
> +        }
> +    }
> +
> +    return size;
> +}
> +
> +static int qemu_rdma_accept(RDMAContext *rdma)
> +{
> +    RDMAControlHeader head = { .len = rdma->local_ram_blocks.num_blocks *
> +                                        sizeof(RDMARemoteBlock),
> +                               .type = RDMA_CONTROL_RAM_BLOCKS,
> +                               .repeat = 1,
> +                             };
> +    RDMACapabilities cap;
> +    struct rdma_conn_param conn_param = {
> +                                            .responder_resources = 2,
> +                                            .private_data = &cap,
> +                                            .private_data_len = sizeof(cap),
> +                                         };
> +    struct rdma_cm_event *cm_event;
> +    struct ibv_context *verbs;
> +    int ret = -EINVAL;
> +    RDMALocalBlocks *local = &rdma->local_ram_blocks;
> +    int i;
> +
> +    ret = rdma_get_cm_event(rdma->channel, &cm_event);
> +    if (ret) {
> +        goto err_rdma_dest_wait;
> +    }
> +
> +    if (cm_event->event != RDMA_CM_EVENT_CONNECT_REQUEST) {
> +        rdma_ack_cm_event(cm_event);
> +        goto err_rdma_dest_wait;
> +    }
> +
> +    memcpy(&cap, cm_event->param.conn.private_data, sizeof(cap));
> +
> +    network_to_caps(&cap);
> +
> +    if (cap.version < 1 || cap.version > RDMA_CONTROL_VERSION_CURRENT) {
> +            fprintf(stderr, "Unknown source RDMA version: %d, bailing...\n",
> +                            cap.version);
> +            rdma_ack_cm_event(cm_event);
> +            goto err_rdma_dest_wait;
> +    }
> +
> +    /*
> +     * Respond with only the capabilities this version of QEMU knows about.
> +     */
> +    cap.flags &= known_capabilities;
> +
> +    /*
> +     * Enable the ones that we do know about.
> +     * Add other checks here as new ones are introduced.
> +     */
> +    if (cap.flags & RDMA_CAPABILITY_PIN_ALL) {
> +        rdma->pin_all = true;
> +    }
> +
> +    rdma->cm_id = cm_event->id;
> +    verbs = cm_event->id->verbs;
> +
> +    rdma_ack_cm_event(cm_event);
> +
> +    DPRINTF("Memory pin all: %s\n", rdma->pin_all ? "enabled" : "disabled");
> +
> +    caps_to_network(&cap);
> +
> +    DPRINTF("verbs context after listen: %p\n", verbs);
> +
> +    if (!rdma->verbs) {
> +        rdma->verbs = verbs;
> +        /*
> +         * Cannot propagate errp, as there is no error pointer
> +         * to be propagated.
> +         */
> +        ret = qemu_rdma_dest_prepare(rdma, NULL);
> +        if (ret) {
> +            fprintf(stderr, "rdma migration: error preparing dest!\n");
> +            goto err_rdma_dest_wait;
> +        }
> +    } else if (rdma->verbs != verbs) {
> +            fprintf(stderr, "ibv context not matching %p, %p!\n",
> +                    rdma->verbs, verbs);
> +            goto err_rdma_dest_wait;
> +    }
> +
> +    qemu_set_fd_handler2(rdma->channel->fd, NULL, NULL, NULL, NULL);
> +
> +    ret = qemu_rdma_alloc_qp(rdma);
> +    if (ret) {
> +        fprintf(stderr, "rdma migration: error allocating qp!\n");
> +        goto err_rdma_dest_wait;
> +    }
> +
> +    ret = rdma_accept(rdma->cm_id, &conn_param);
> +    if (ret) {
> +        fprintf(stderr, "rdma_accept returns %d!\n", ret);
> +        goto err_rdma_dest_wait;
> +    }
> +
> +    ret = rdma_get_cm_event(rdma->channel, &cm_event);
> +    if (ret) {
> +        fprintf(stderr, "rdma_accept get_cm_event failed %d!\n", ret);
> +        goto err_rdma_dest_wait;
> +    }
> +
> +    if (cm_event->event != RDMA_CM_EVENT_ESTABLISHED) {
> +        fprintf(stderr, "rdma_accept not event established!\n");
> +        rdma_ack_cm_event(cm_event);
> +        goto err_rdma_dest_wait;
> +    }
> +
> +    rdma_ack_cm_event(cm_event);
> +
> +    ret = qemu_rdma_post_recv_control(rdma, 0);
> +    if (ret) {
> +        fprintf(stderr, "rdma migration: error posting second control recv!\n");
> +        goto err_rdma_dest_wait;
> +    }
> +
> +    if (rdma->pin_all) {
> +        ret = qemu_rdma_reg_whole_ram_blocks(rdma, &rdma->local_ram_blocks);
> +        if (ret) {
> +            fprintf(stderr, "rdma migration: error dest "
> +                            "registering ram blocks!\n");
> +            goto err_rdma_dest_wait;
> +        }
> +    }
> +
> +    /*
> +     * Server uses this to prepare to transmit the RAMBlock descriptions
> +     * to the primary VM after connection setup.
> +     * Both sides use the "remote" structure to communicate and update
> +     * their "local" descriptions with what was sent.
> +     */
> +    for (i = 0; i < local->num_blocks; i++) {
> +            rdma->block[i].remote_host_addr =
> +                (uint64_t)(local->block[i].local_host_addr);
> +
> +            if (rdma->pin_all) {
> +                rdma->block[i].remote_rkey = local->block[i].mr->rkey;
> +            }
> +
> +            rdma->block[i].offset = local->block[i].offset;
> +            rdma->block[i].length = local->block[i].length;
> +    }
> +
> +
> +    ret = qemu_rdma_post_send_control(rdma, (uint8_t *) rdma->block, &head);
> +
> +    if (ret < 0) {
> +        fprintf(stderr, "rdma migration: error sending remote info!\n");
> +        goto err_rdma_dest_wait;
> +    }
> +
> +    qemu_rdma_dump_gid("dest_connect", rdma->cm_id);
> +
> +    return 0;
> +
> +err_rdma_dest_wait:
> +    rdma->error_state = ret;
> +    qemu_rdma_cleanup(rdma);
> +    return ret;
> +}
> +
> +/*
> + * During each iteration of the migration, we listen for instructions
> + * by the primary VM to perform dynamic page registrations before they
> + * can perform RDMA operations.
> + *
> + * We respond with the 'rkey'.
> + *
> + * Keep doing this until the primary tells us to stop.
> + */
> +static int qemu_rdma_registration_handle(QEMUFile *f, void *opaque,
> +                                         uint64_t flags)
> +{
> +    RDMAControlHeader resp = { .len = sizeof(RDMARegisterResult),
> +                               .type = RDMA_CONTROL_REGISTER_RESULT,
> +                               .repeat = 0,
> +                             };
> +    QEMUFileRDMA *rfile = opaque;
> +    RDMAContext *rdma = rfile->rdma;
> +    RDMAControlHeader head;
> +    RDMARegister *reg, *registers;
> +    RDMACompress *comp;
> +    RDMARegisterResult *reg_result;
> +    static RDMARegisterResult results[RDMA_CONTROL_MAX_COMMANDS_PER_MESSAGE];
> +    RDMALocalBlock *block;
> +    void *host_addr;
> +    int ret = 0;
> +    int idx = 0;
> +    int count = 0;
> +
> +    CHECK_ERROR_STATE();
> +
> +    if (rdma->pin_all) {
> +        return 0;
> +    }
> +
> +    do {
> +        DDDPRINTF("Waiting for next registration %" PRIu64 "...\n", flags);
> +
> +        ret = qemu_rdma_exchange_recv(rdma, &head, RDMA_CONTROL_NONE);
> +
> +        if (ret < 0) {
> +            break;
> +        }
> +
> +        if (head.repeat > RDMA_CONTROL_MAX_COMMANDS_PER_MESSAGE) {
> +            fprintf(stderr, "Too many requests in this message (%d)."
> +                            "Bailing.\n", head.repeat);
> +            ret = -EIO;
> +            break;
> +        }
> +
> +        switch (head.type) {
> +        case RDMA_CONTROL_COMPRESS:
> +            comp = (RDMACompress *) rdma->wr_data[idx].control_curr;
> +
> +            DDPRINTF("Zapping zero chunk: %" PRId64
> +                    " bytes, index %d, offset %" PRId64 "\n",
> +                    comp->length, comp->block_idx, comp->offset);
> +            comp = (RDMACompress *) rdma->wr_data[idx].control_curr;
> +            block = &(rdma->local_ram_blocks.block[comp->block_idx]);
> +
> +            host_addr = block->local_host_addr +
> +                            (comp->offset - block->offset);
> +
> +            ram_handle_compressed(host_addr, comp->value, comp->length);
> +            break;
> +        case RDMA_CONTROL_REGISTER_FINISHED:
> +            DDDPRINTF("Current registrations complete.\n");
> +            goto out;
> +        case RDMA_CONTROL_REGISTER_REQUEST:
> +            DDPRINTF("There are %d registration requests\n", head.repeat);
> +
> +            resp.repeat = head.repeat;
> +            registers = (RDMARegister *) rdma->wr_data[idx].control_curr;
> +
> +            for (count = 0; count < head.repeat; count++) {
> +                reg = &registers[count];
> +                reg_result = &results[count];
> +
> +                DDPRINTF("Registration request (%d): %d"
> +                    " bytes, index %d, offset %" PRIu64 "\n",
> +                    count, reg->len, reg->current_index, reg->offset);
> +
> +                block = &(rdma->local_ram_blocks.block[reg->current_index]);
> +                host_addr = (block->local_host_addr +
> +                            (reg->offset - block->offset));
> +                if (qemu_rdma_register_and_get_keys(rdma, block,
> +                            (uint8_t *)host_addr, NULL, &reg_result->rkey)) {
> +                    fprintf(stderr, "cannot get rkey!\n");
> +                    ret = -EINVAL;
> +                    goto out;
> +                }
> +
> +                DDPRINTF("Registered rkey for this request: %x\n",
> +                                reg_result->rkey);
> +            }
> +
> +            ret = qemu_rdma_post_send_control(rdma,
> +                            (uint8_t *) results, &resp);
> +
> +            if (ret < 0) {
> +                fprintf(stderr, "Failed to send control buffer!\n");
> +                goto out;
> +            }
> +            break;
> +        case RDMA_CONTROL_REGISTER_RESULT:
> +            fprintf(stderr, "Invalid RESULT message at dest.\n");
> +            ret = -EIO;
> +            goto out;
> +        default:
> +            fprintf(stderr, "Unknown control message %s\n",
> +                                control_desc[head.type]);
> +            ret = -EIO;
> +            goto out;
> +        }
> +    } while (1);
> +out:
> +    if (ret < 0) {
> +        rdma->error_state = ret;
> +    }
> +    return ret;
> +}
> +
> +static int qemu_rdma_registration_start(QEMUFile *f, void *opaque,
> +                                        uint64_t flags)
> +{
> +    QEMUFileRDMA *rfile = opaque;
> +    RDMAContext *rdma = rfile->rdma;
> +
> +    CHECK_ERROR_STATE();
> +
> +    if (rdma->pin_all) {
> +        return 0;
> +    }
> +
> +    DDDPRINTF("start section: %" PRIu64 "\n", flags);
> +    qemu_put_be64(f, RAM_SAVE_FLAG_HOOK);
> +    qemu_fflush(f);
> +
> +    return 0;
> +}
> +
> +/*
> + * Inform dest that dynamic registrations are done for now.
> + * First, flush writes, if any.
> + */
> +static int qemu_rdma_registration_stop(QEMUFile *f, void *opaque,
> +                                       uint64_t flags)
> +{
> +    QEMUFileRDMA *rfile = opaque;
> +    RDMAContext *rdma = rfile->rdma;
> +    RDMAControlHeader head = { .len = 0,
> +                               .type = RDMA_CONTROL_REGISTER_FINISHED,
> +                               .repeat = 1,
> +                             };
> +
> +    CHECK_ERROR_STATE();
> +
> +    qemu_fflush(f);
> +    int ret = qemu_rdma_drain_cq(f, rdma);
> +
> +    if (ret < 0) {
> +        goto err;
> +    }
> +
> +    if (rdma->pin_all) {
> +        return 0;
> +    }
> +
> +    DDDPRINTF("Sending registration finish %" PRIu64 "...\n", flags);
> +    ret = qemu_rdma_exchange_send(rdma, &head, NULL, NULL, NULL);
> +
> +    if (ret < 0) {
> +        goto err;
> +    }
> +
> +    return 0;
> +err:
> +    rdma->error_state = ret;
> +    return ret;
> +}
> +
> +static int qemu_rdma_get_fd(void *opaque)
> +{
> +    QEMUFileRDMA *rfile = opaque;
> +    RDMAContext *rdma = rfile->rdma;
> +
> +    return rdma->comp_channel->fd;
> +}
> +
> +const QEMUFileOps rdma_read_ops = {
> +    .get_buffer    = qemu_rdma_get_buffer,
> +    .get_fd        = qemu_rdma_get_fd,
> +    .close         = qemu_rdma_close,
> +    .hook_ram_load = qemu_rdma_registration_handle,
> +};
> +
> +const QEMUFileOps rdma_write_ops = {
> +    .put_buffer           = qemu_rdma_put_buffer,
> +    .close                = qemu_rdma_close,
> +    .before_ram_iterate   = qemu_rdma_registration_start,
> +    .after_ram_iterate    = qemu_rdma_registration_stop,
> +    .save_page            = qemu_rdma_save_page,
> +};
> +
> +static void *qemu_fopen_rdma(RDMAContext *rdma, const char *mode)
> +{
> +    QEMUFileRDMA *r = g_malloc0(sizeof(QEMUFileRDMA));
> +
> +    if (qemu_file_mode_is_not_valid(mode)) {
> +        return NULL;
> +    }
> +
> +    r->rdma = rdma;
> +
> +    if (mode[0] == 'w') {
> +        r->file = qemu_fopen_ops(r, &rdma_write_ops);
> +    } else {
> +        r->file = qemu_fopen_ops(r, &rdma_read_ops);
> +    }
> +
> +    return r->file;
> +}
> +
> +static void rdma_accept_incoming_migration(void *opaque)
> +{
> +    RDMAContext *rdma = opaque;
> +    int ret;
> +    QEMUFile *f;
> +    Error *local_err = NULL, **errp = &local_err;
> +
> +    DPRINTF("Accepting rdma connection...\n");
> +    ret = qemu_rdma_accept(rdma);
> +
> +    if (ret) {
> +        ERROR(errp, "RDMA Migration initialization failed!\n");
> +        goto err;
> +    }
> +
> +    DPRINTF("Accepted migration\n");
> +
> +    f = qemu_fopen_rdma(rdma, "rb");
> +    if (f == NULL) {
> +        ERROR(errp, "could not qemu_fopen_rdma!\n");
> +        goto err;
> +    }
> +
> +    rdma->migration_started_on_destination = 1;
> +    process_incoming_migration(f);
> +    return;
> +
> +err:
> +    qemu_rdma_cleanup(rdma);
> +}
> +
> +void rdma_start_incoming_migration(const char *host_port, Error **errp)
> +{
> +    int ret;
> +    RDMAContext *rdma;
> +    Error *local_err = NULL;
> +
> +    DPRINTF("Starting RDMA-based incoming migration\n");
> +    rdma = qemu_rdma_data_init(host_port, &local_err);
> +
> +    if (rdma == NULL) {
> +        goto err;
> +    }
> +
> +    ret = qemu_rdma_dest_init(rdma, &local_err);
> +
> +    if (ret) {
> +        goto err;
> +    }
> +
> +    DPRINTF("qemu_rdma_dest_init success\n");
> +    ret = qemu_rdma_dest_prepare(rdma, &local_err);
> +
> +    if (ret) {
> +        goto err;
> +    }
> +
> +    DPRINTF("qemu_rdma_dest_prepare success\n");
> +
> +    qemu_set_fd_handler2(rdma->channel->fd, NULL,
> +                         rdma_accept_incoming_migration, NULL,
> +                            (void *)(intptr_t) rdma);
> +    return;
> +err:
> +    error_propagate(errp, local_err);
> +    g_free(rdma);
> +}
> +
> +void rdma_start_outgoing_migration(void *opaque,
> +                            const char *host_port, Error **errp)
> +{
> +    MigrationState *s = opaque;
> +    Error *local_err = NULL, **temp = &local_err;
> +    RDMAContext *rdma = qemu_rdma_data_init(host_port, &local_err);
> +    int ret = 0;
> +
> +    if (rdma == NULL) {
> +        ERROR(temp, "Failed to initialize RDMA data structures! %d\n", ret);
> +        goto err;
> +    }
> +
> +    ret = qemu_rdma_source_init(rdma, &local_err,
> +        s->enabled_capabilities[MIGRATION_CAPABILITY_X_RDMA_PIN_ALL]);
> +
> +    if (ret) {
> +        goto err;
> +    }
> +
> +    DPRINTF("qemu_rdma_source_init success\n");
> +    ret = qemu_rdma_connect(rdma, &local_err);
> +
> +    if (ret) {
> +        goto err;
> +    }
> +
> +    DPRINTF("qemu_rdma_source_connect success\n");
> +
> +    s->file = qemu_fopen_rdma(rdma, "wb");
> +    s->total_time = qemu_get_clock_ms(rt_clock);
> +    migrate_fd_connect(s);
> +    return;
> +err:
> +    error_propagate(errp, local_err);
> +    g_free(rdma);
> +    migrate_fd_error(s);
> +}
> -- 
> 1.7.10.4

Re: [Qemu-devel] [PATCH v6 10/11] rdma: core logic

[Michael R. Hines]

On 04/25/2013 03:45 PM, Anthony Liguori wrote:
> mrhines@linux.vnet.ibm.com writes:
>
>> From: "Michael R. Hines" <mrhines@us.ibm.com>
>>
>> Code that does need to be visible is kept
>> well contained inside this file and this is the only
>> new additional file to the entire patch - good progress.
>>
>> This file includes the entire protocol and interfaces
>> required to perform RDMA migration.
>>
>> Also, the configure and Makefile modifications to link
>> this file are included.
>>
>> Full documentation is in docs/rdma.txt
>>
>> Signed-off-by: Michael R. Hines <mrhines@us.ibm.com>
>> ---
>>   Makefile.objs                 |    1 +
>>   configure                     |   29 +++++++++++++++++++++++++++++
>>   include/migration/migration.h |    4 ++++
>>   migration.c                   |    8 ++++++++
>>   migration-rdma.c              | 2707 +++++++++++++++++++++++++++++++++++++++++++++
>>   4 files changed, 2749 insertions(+)
>>
>> diff --git a/Makefile.objs b/Makefile.objs
>> index a473348..d744827 100644
>> --- a/Makefile.objs
>> +++ b/Makefile.objs
>> @@ -49,6 +49,7 @@ common-obj-$(CONFIG_POSIX) += os-posix.o
>>   common-obj-$(CONFIG_LINUX) += fsdev/
>>   
>>   common-obj-y += migration.o migration-tcp.o
>> +common-obj-$(CONFIG_RDMA) += migration-rdma.o
>>   common-obj-y += qemu-char.o #aio.o
>>   common-obj-y += block-migration.o
>>   common-obj-y += page_cache.o xbzrle.o
>> diff --git a/configure b/configure
>> index 33d3354..90441cf 100755
>> --- a/configure
>> +++ b/configure
>> @@ -181,6 +181,7 @@ xfs=""
>>   vhost_net="no"
>>   vhost_scsi="no"
>>   kvm="no"
>> +rdma="yes"
> This should be:
>
> rdma=""
>

Oooops..... Should I send out a v7 patch?

- Michael

Re: [Qemu-devel] [PATCH v6 10/11] rdma: core logic

[Anthony Liguori]

"Michael R. Hines" <mrhines@linux.vnet.ibm.com> writes:

> On 04/25/2013 03:45 PM, Anthony Liguori wrote:
>> mrhines@linux.vnet.ibm.com writes:
>>
>>> From: "Michael R. Hines" <mrhines@us.ibm.com>
>>>
>>> Code that does need to be visible is kept
>>> well contained inside this file and this is the only
>>> new additional file to the entire patch - good progress.
>>>
>>> This file includes the entire protocol and interfaces
>>> required to perform RDMA migration.
>>>
>>> Also, the configure and Makefile modifications to link
>>> this file are included.
>>>
>>> Full documentation is in docs/rdma.txt
>>>
>>> Signed-off-by: Michael R. Hines <mrhines@us.ibm.com>
>>> ---
>>>   Makefile.objs                 |    1 +
>>>   configure                     |   29 +++++++++++++++++++++++++++++
>>>   include/migration/migration.h |    4 ++++
>>>   migration.c                   |    8 ++++++++
>>>   migration-rdma.c              | 2707 +++++++++++++++++++++++++++++++++++++++++++++
>>>   4 files changed, 2749 insertions(+)
>>>
>>> diff --git a/Makefile.objs b/Makefile.objs
>>> index a473348..d744827 100644
>>> --- a/Makefile.objs
>>> +++ b/Makefile.objs
>>> @@ -49,6 +49,7 @@ common-obj-$(CONFIG_POSIX) += os-posix.o
>>>   common-obj-$(CONFIG_LINUX) += fsdev/
>>>   
>>>   common-obj-y += migration.o migration-tcp.o
>>> +common-obj-$(CONFIG_RDMA) += migration-rdma.o
>>>   common-obj-y += qemu-char.o #aio.o
>>>   common-obj-y += block-migration.o
>>>   common-obj-y += page_cache.o xbzrle.o
>>> diff --git a/configure b/configure
>>> index 33d3354..90441cf 100755
>>> --- a/configure
>>> +++ b/configure
>>> @@ -181,6 +181,7 @@ xfs=""
>>>   vhost_net="no"
>>>   vhost_scsi="no"
>>>   kvm="no"
>>> +rdma="yes"
>> This should be:
>>
>> rdma=""
>>
>
> Oooops..... Should I send out a v7 patch?

Yes.

Regards,

Anthony Liguori

>
> - Michael

[Qemu-devel] [PATCH v6 11/11] rdma: send pc.ram

[mrhines]

From: "Michael R. Hines" <mrhines@us.ibm.com>

This takes advantages of the previous patches:

1. use the new QEMUFileOps hook 'save_page'

2. call out to the right accessor methods to invoke
   the iteration hooks defined in QEMUFileOps

Signed-off-by: Michael R. Hines <mrhines@us.ibm.com>
---
 arch_init.c |   35 +++++++++++++++++++++++++++++++++--
 1 file changed, 33 insertions(+), 2 deletions(-)

diff --git a/arch_init.c b/arch_init.c
index e272852..09559f4 100644
--- a/arch_init.c
+++ b/arch_init.c
@@ -115,6 +115,7 @@ const uint32_t arch_type = QEMU_ARCH;
 #define RAM_SAVE_FLAG_EOS      0x10
 #define RAM_SAVE_FLAG_CONTINUE 0x20
 #define RAM_SAVE_FLAG_XBZRLE   0x40
+/* 0x80 is reserved in migration.h start with 0x100 next */
 
 
 static struct defconfig_file {
@@ -454,8 +455,15 @@ static int ram_save_block(QEMUFile *f, bool last_stage)
             p = memory_region_get_ram_ptr(mr) + offset;
 
             /* In doubt sent page as normal */
-            bytes_sent = -1;
-            if (is_zero_page(p)) {
+            bytes_sent = ram_control_save_page(f, block->offset,
+                                       offset, TARGET_PAGE_SIZE);
+            if (bytes_sent != -1) {
+                if (bytes_sent > 0) {
+                    acct_info.norm_pages++;
+                } else if (bytes_sent == 0) {
+                    acct_info.dup_pages++;
+                }
+            } else if (is_zero_page(p)) {
                 acct_info.dup_pages++;
                 if (!ram_bulk_stage) {
                     bytes_sent = save_block_hdr(f, block, offset, cont,
@@ -598,6 +606,15 @@ static int ram_save_setup(QEMUFile *f, void *opaque)
     }
 
     qemu_mutex_unlock_ramlist();
+
+    /*
+     * Please leave in place. These calls generate reserved messages in
+     * the RDMA protocol in order to pre-register RDMA memory in the
+     * future to before the bulk round begins.
+     */
+    ram_control_before_iterate(f, RAM_CONTROL_SETUP);
+    ram_control_after_iterate(f, RAM_CONTROL_SETUP);
+
     qemu_put_be64(f, RAM_SAVE_FLAG_EOS);
 
     return 0;
@@ -616,6 +633,8 @@ static int ram_save_iterate(QEMUFile *f, void *opaque)
         reset_ram_globals();
     }
 
+    ram_control_before_iterate(f, RAM_CONTROL_ROUND);
+
     t0 = qemu_get_clock_ns(rt_clock);
     i = 0;
     while ((ret = qemu_file_rate_limit(f)) == 0) {
@@ -646,6 +665,12 @@ static int ram_save_iterate(QEMUFile *f, void *opaque)
 
     qemu_mutex_unlock_ramlist();
 
+    /*
+     * Must occur before EOS (or any QEMUFile operation)
+     * because of RDMA protocol.
+     */
+    ram_control_after_iterate(f, RAM_CONTROL_ROUND);
+
     if (ret < 0) {
         bytes_transferred += total_sent;
         return ret;
@@ -663,6 +688,8 @@ static int ram_save_complete(QEMUFile *f, void *opaque)
     qemu_mutex_lock_ramlist();
     migration_bitmap_sync();
 
+    ram_control_before_iterate(f, RAM_CONTROL_FINISH);
+
     /* try transferring iterative blocks of memory */
 
     /* flush all remaining blocks regardless of rate limiting */
@@ -676,6 +703,8 @@ static int ram_save_complete(QEMUFile *f, void *opaque)
         }
         bytes_transferred += bytes_sent;
     }
+
+    ram_control_after_iterate(f, RAM_CONTROL_FINISH);
     migration_end();
 
     qemu_mutex_unlock_ramlist();
@@ -872,6 +901,8 @@ static int ram_load(QEMUFile *f, void *opaque, int version_id)
                 ret = -EINVAL;
                 goto done;
             }
+        } else if (flags & RAM_SAVE_FLAG_HOOK) {
+            ram_control_load_hook(f, flags);
         }
         error = qemu_file_get_error(f);
         if (error) {
-- 
1.7.10.4

Re: [Qemu-devel] [PATCH v6 00/11] rdma: migration support

[Paolo Bonzini]

Il 24/04/2013 21:00, mrhines@linux.vnet.ibm.com ha scritto:
> From: "Michael R. Hines" <mrhines@us.ibm.com>
> 
> Changes since v5:
> 
> - Removed max_size hook.

The patches look good.  I will not be very available in the next few
days due to a public holiday here, but I believe that it's okay for 1.5.
 It's clearly marked as experimental, and the changes to the internals
are safe and ok.

The one small nit is that patch 11 should come before patch 10.  It can
be fixed by whoever applies the patch.

Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>

Paolo

> 
> Wiki: http://wiki.qemu.org/Features/RDMALiveMigration
> Github: git@github.com:hinesmr/qemu.git
> 
> Here is a brief summary of total migration time and downtime using RDMA:
> 
> Using a 40gbps infiniband link performing a worst-case stress test,
> using an 8GB RAM virtual machine:
> Using the following command:
> 
> $ apt-get install stress
> $ stress --vm-bytes 7500M --vm 1 --vm-keep
> 
> RESULTS:
> 
> 1. Migration throughput: 26 gigabits/second.
> 2. Downtime (stop time) varies between 15 and 100 milliseconds.
> 
> EFFECTS of memory registration on bulk phase round:
> 
> For example, in the same 8GB RAM example with all 8GB of memory in 
> active use and the VM itself is completely idle using the same 40 gbps 
> infiniband link:
> 
> 1. x-rdma-pin-all disabled total time: approximately 7.5 seconds @ 9.5 Gbps
> 2. x-rdma-pin-all enabled total time: approximately 4 seconds @ 26 Gbps
> 
> These numbers would of course scale up to whatever size virtual machine
> you have to migrate using RDMA.
> 
> Enabling this feature does *not* have any measurable affect on 
> migration *downtime*. This is because, without this feature, all of the 
> memory will have already been registered already in advance during
> the bulk round and does not need to be re-registered during the successive
> iteration rounds.
> 
> The following changes since commit f3aa844bbb2922a5b8393d17620eca7d7e921ab3:
> 
>   build: include config-{, all-}devices.mak after defining CONFIG_SOFTMMU and CONFIG_USER_ONLY (2013-04-24 12:18:41 -0500)
> 
> are available in the git repository at:
> 
>   git@github.com:hinesmr/qemu.git rdma_patch_v6
> 
> for you to fetch changes up to 75e6fac1f642885b93cefe6e1874d648e9850f8f:
> 
>   rdma: send pc.ram (2013-04-24 14:55:01 -0400)
> 
> ----------------------------------------------------------------
> Michael R. Hines (11):
>       rdma: add documentation
>       rdma: export yield_until_fd_readable()
>       rdma: export throughput w/ MigrationStats QMP
>       rdma: introduce qemu_file_mode_is_not_valid()
>       rdma: export qemu_fflush()
>       rdma: introduce  ram_handle_compressed()
>       rdma: introduce qemu_ram_foreach_block()
>       rdma: new QEMUFileOps hooks
>       rdma: introduce capability x-rdma-pin-all
>       rdma: core logic
>       rdma: send pc.ram
> 
>  Makefile.objs                 |    1 +
>  arch_init.c                   |   59 +-
>  configure                     |   29 +
>  docs/rdma.txt                 |  404 ++++++
>  exec.c                        |    9 +
>  hmp.c                         |    2 +
>  include/block/coroutine.h     |    6 +
>  include/exec/cpu-common.h     |    5 +
>  include/migration/migration.h |   25 +
>  include/migration/qemu-file.h |   30 +
>  migration-rdma.c              | 2707 +++++++++++++++++++++++++++++++++++++++++
>  migration.c                   |   27 +
>  qapi-schema.json              |   12 +-
>  qemu-coroutine-io.c           |   23 +
>  savevm.c                      |  107 +-
>  15 files changed, 3398 insertions(+), 48 deletions(-)
>  create mode 100644 docs/rdma.txt
>  create mode 100644 migration-rdma.c
> 

Re: [Qemu-devel] [PATCH v6 00/11] rdma: migration support

[Michael R. Hines]

On 04/24/2013 05:50 PM, Paolo Bonzini wrote:
> Il 24/04/2013 21:00, mrhines@linux.vnet.ibm.com ha scritto:
>> From: "Michael R. Hines" <mrhines@us.ibm.com>
>>
>> Changes since v5:
>>
>> - Removed max_size hook.
> The patches look good.  I will not be very available in the next few
> days due to a public holiday here, but I believe that it's okay for 1.5.
>   It's clearly marked as experimental, and the changes to the internals
> are safe and ok.
>
> The one small nit is that patch 11 should come before patch 10.  It can
> be fixed by whoever applies the patch.
>
> Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
>
> Paolo

Acknowledged. Thank you.

- Michael