4 * Copyright IBM, Corp. 2008
7 * Anthony Liguori <aliguori@us.ibm.com>
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
14 #ifndef QEMU_MIGRATION_H
15 #define QEMU_MIGRATION_H
17 #include "exec/cpu-common.h"
18 #include "hw/qdev-core.h"
19 #include "qapi/qapi-types-migration.h"
20 #include "qapi/qmp/json-writer.h"
21 #include "qemu/thread.h"
22 #include "qemu/coroutine_int.h"
23 #include "io/channel.h"
24 #include "io/channel-buffer.h"
25 #include "net/announce.h"
26 #include "qom/object.h"
27 #include "postcopy-ram.h"
29 struct PostcopyBlocktimeContext
;
31 #define MIGRATION_RESUME_ACK_VALUE (1)
34 * 1<<6=64 pages -> 256K chunk when page size is 4K. This gives us
35 * the benefit that all the chunks are 64 pages aligned then the
36 * bitmaps are always aligned to LONG.
38 #define CLEAR_BITMAP_SHIFT_MIN 6
40 * 1<<18=256K pages -> 1G chunk when page size is 4K. This is the
41 * default value to use if no one specified.
43 #define CLEAR_BITMAP_SHIFT_DEFAULT 18
45 * 1<<31=2G pages -> 8T chunk when page size is 4K. This should be
46 * big enough and make sure we won't overflow easily.
48 #define CLEAR_BITMAP_SHIFT_MAX 31
50 /* This is an abstraction of a "temp huge page" for postcopy's purpose */
53 * This points to a temporary huge page as a buffer for UFFDIO_COPY. It's
54 * mmap()ed and needs to be freed when cleanup.
58 * This points to the host page we're going to install for this temp page.
59 * It tells us after we've received the whole page, where we should put it.
62 /* Number of small pages copied (in size of TARGET_PAGE_SIZE) */
63 unsigned int target_pages
;
64 /* Whether this page contains all zeros */
69 PREEMPT_THREAD_NONE
= 0,
70 PREEMPT_THREAD_CREATED
,
72 } PreemptThreadStatus
;
74 /* State for the incoming migration */
75 struct MigrationIncomingState
{
76 QEMUFile
*from_src_file
;
77 /* Previously received RAM's RAMBlock pointer */
78 RAMBlock
*last_recv_block
[RAM_CHANNEL_MAX
];
79 /* A hook to allow cleanup at the end of incoming migration */
81 void (*transport_cleanup
)(void *data
);
83 * Used to sync thread creations. Note that we can't create threads in
84 * parallel with this sem.
86 QemuSemaphore thread_sync_sem
;
88 * Free at the start of the main state load, set as the main thread finishes
91 QemuEvent main_thread_load_event
;
93 /* For network announces */
94 AnnounceTimer announce_timer
;
96 size_t largest_page_size
;
97 bool have_fault_thread
;
98 QemuThread fault_thread
;
99 /* Set this when we want the fault thread to quit */
100 bool fault_thread_quit
;
102 bool have_listen_thread
;
103 QemuThread listen_thread
;
105 /* For the kernel to send us notifications */
107 /* To notify the fault_thread to wake, e.g., when need to quit */
108 int userfault_event_fd
;
109 QEMUFile
*to_src_file
;
110 QemuMutex rp_mutex
; /* We send replies from multiple threads */
111 /* RAMBlock of last request sent to source */
114 * Number of postcopy channels including the default precopy channel, so
115 * vanilla postcopy will only contain one channel which contain both
116 * precopy and postcopy streams.
118 * This is calculated when the src requests to enable postcopy but before
119 * it starts. Its value can depend on e.g. whether postcopy preemption is
122 unsigned int postcopy_channels
;
123 /* QEMUFile for postcopy only; it'll be handled by a separate thread */
124 QEMUFile
*postcopy_qemufile_dst
;
126 * When postcopy_qemufile_dst is properly setup, this sem is posted.
127 * One can wait on this semaphore to wait until the preempt channel is
130 QemuSemaphore postcopy_qemufile_dst_done
;
131 /* Postcopy priority thread is used to receive postcopy requested pages */
132 QemuThread postcopy_prio_thread
;
134 * Always set by the main vm load thread only, but can be read by the
135 * postcopy preempt thread. "volatile" makes sure all reads will be
136 * uptodate across cores.
138 volatile PreemptThreadStatus preempt_thread_status
;
140 * Used to sync between the ram load main thread and the fast ram load
141 * thread. It protects postcopy_qemufile_dst, which is the postcopy
144 * The ram fast load thread will take it mostly for the whole lifecycle
145 * because it needs to continuously read data from the channel, and
146 * it'll only release this mutex if postcopy is interrupted, so that
147 * the ram load main thread will take this mutex over and properly
148 * release the broken channel.
150 QemuMutex postcopy_prio_thread_mutex
;
152 * An array of temp host huge pages to be used, one for each postcopy
155 PostcopyTmpPage
*postcopy_tmp_pages
;
156 /* This is shared for all postcopy channels */
157 void *postcopy_tmp_zero_page
;
158 /* PostCopyFD's for external userfaultfds & handlers of shared memory */
159 GArray
*postcopy_remote_fds
;
165 bool have_colo_incoming_thread
;
166 QemuThread colo_incoming_thread
;
167 /* The coroutine we should enter (back) after failover */
168 Coroutine
*migration_incoming_co
;
169 QemuSemaphore colo_incoming_sem
;
172 * PostcopyBlocktimeContext to keep information for postcopy
173 * live migration, to calculate vCPU block time
175 struct PostcopyBlocktimeContext
*blocktime_ctx
;
177 /* notify PAUSED postcopy incoming migrations to try to continue */
178 QemuSemaphore postcopy_pause_sem_dst
;
179 QemuSemaphore postcopy_pause_sem_fault
;
181 * This semaphore is used to allow the ram fast load thread (only when
182 * postcopy preempt is enabled) fall into sleep when there's network
183 * interruption detected. When the recovery is done, the main load
184 * thread will kick the fast ram load thread using this semaphore.
186 QemuSemaphore postcopy_pause_sem_fast_load
;
188 /* List of listening socket addresses */
189 SocketAddressList
*socket_address_list
;
191 /* A tree of pages that we requested to the source VM */
192 GTree
*page_requested
;
193 /* For debugging purpose only, but would be nice to keep */
194 int page_requested_count
;
196 * The mutex helps to maintain the requested pages that we sent to the
197 * source, IOW, to guarantee coherent between the page_requests tree and
198 * the per-ramblock receivedmap. Note! This does not guarantee consistency
199 * of the real page copy procedures (using UFFDIO_[ZERO]COPY). E.g., even
200 * if one bit in receivedmap is cleared, UFFDIO_COPY could have happened
201 * for that page already. This is intended so that the mutex won't
202 * serialize and blocked by slow operations like UFFDIO_* ioctls. However
203 * this should be enough to make sure the page_requested tree always
204 * contains valid information.
206 QemuMutex page_request_mutex
;
209 MigrationIncomingState
*migration_incoming_get_current(void);
210 void migration_incoming_state_destroy(void);
211 void migration_incoming_transport_cleanup(MigrationIncomingState
*mis
);
213 * Functions to work with blocktime context
215 void fill_destination_postcopy_migration_info(MigrationInfo
*info
);
217 #define TYPE_MIGRATION "migration"
219 typedef struct MigrationClass MigrationClass
;
220 DECLARE_OBJ_CHECKERS(MigrationState
, MigrationClass
,
221 MIGRATION_OBJ
, TYPE_MIGRATION
)
223 struct MigrationClass
{
225 DeviceClass parent_class
;
228 struct MigrationState
{
230 DeviceState parent_obj
;
236 /* Protected by qemu_file_lock */
237 QEMUFile
*to_dst_file
;
238 /* Postcopy specific transfer channel */
239 QEMUFile
*postcopy_qemufile_src
;
241 * It is posted when the preempt channel is established. Note: this is
242 * used for both the start or recover of a postcopy migration. We'll
243 * post to this sem every time a new preempt channel is created in the
244 * main thread, and we keep post() and wait() in pair.
246 QemuSemaphore postcopy_qemufile_src_sem
;
247 QIOChannelBuffer
*bioc
;
249 * Protects to_dst_file/from_dst_file pointers. We need to make sure we
250 * won't yield or hang during the critical section, since this lock will be
251 * used in OOB command handler.
253 QemuMutex qemu_file_lock
;
256 * Used to allow urgent requests to override rate limiting.
258 QemuSemaphore rate_limit_sem
;
260 /* pages already send at the beginning of current iteration */
261 uint64_t iteration_initial_pages
;
263 /* pages transferred per second */
264 double pages_per_second
;
266 /* bytes already send at the beginning of current iteration */
267 uint64_t iteration_initial_bytes
;
268 /* time at the start of current iteration */
269 int64_t iteration_start_time
;
271 * The final stage happens when the remaining data is smaller than
272 * this threshold; it's calculated from the requested downtime and
275 int64_t threshold_size
;
277 /* params from 'migrate-set-parameters' */
278 MigrationParameters parameters
;
282 /* State related to return path */
284 /* Protected by qemu_file_lock */
285 QEMUFile
*from_dst_file
;
286 QemuThread rp_thread
;
289 * We can also check non-zero of rp_thread, but there's no "official"
290 * way to do this, so this bool makes it slightly more elegant.
291 * Checking from_dst_file for this is racy because from_dst_file will
292 * be cleared in the rp_thread!
294 bool rp_thread_created
;
295 QemuSemaphore rp_sem
;
297 * We post to this when we got one PONG from dest. So far it's an
298 * easy way to know the main channel has successfully established
301 QemuSemaphore rp_pong_acks
;
305 /* Timestamp when recent migration starts (ms) */
307 /* Total time used by latest migration (ms) */
309 /* Timestamp when VM is down (ms) to migrate the last stuff */
310 int64_t downtime_start
;
312 int64_t expected_downtime
;
313 bool capabilities
[MIGRATION_CAPABILITY__MAX
];
316 * Whether guest was running when we enter the completion stage.
317 * If migration is interrupted by any reason, we need to continue
318 * running the guest on source.
322 /* Flag set once the migration has been asked to enter postcopy */
324 /* Flag set after postcopy has sent the device state */
325 bool postcopy_after_devices
;
327 /* Flag set once the migration thread is running (and needs joining) */
328 bool migration_thread_running
;
330 /* Flag set once the migration thread called bdrv_inactivate_all */
333 /* Migration is waiting for guest to unplug device */
334 QemuSemaphore wait_unplug_sem
;
336 /* Migration is paused due to pause-before-switchover */
337 QemuSemaphore pause_sem
;
339 /* The semaphore is used to notify COLO thread that failover is finished */
340 QemuSemaphore colo_exit_sem
;
342 /* The event is used to notify COLO thread to do checkpoint */
343 QemuEvent colo_checkpoint_event
;
344 int64_t colo_checkpoint_time
;
345 QEMUTimer
*colo_delay_timer
;
347 /* The first error that has occurred.
348 We used the mutex to be able to return the 1st error message */
350 /* mutex to protect errp */
351 QemuMutex error_mutex
;
353 /* Do we have to clean up -b/-i from old migrate parameters */
354 /* This feature is deprecated and will be removed */
355 bool must_remove_block_options
;
358 * Global switch on whether we need to store the global state
361 bool store_global_state
;
363 /* Whether we send QEMU_VM_CONFIGURATION during migration */
364 bool send_configuration
;
365 /* Whether we send section footer during migration */
366 bool send_section_footer
;
368 /* Needed by postcopy-pause state */
369 QemuSemaphore postcopy_pause_sem
;
370 QemuSemaphore postcopy_pause_rp_sem
;
372 * Whether we abort the migration if decompression errors are
373 * detected at the destination. It is left at false for qemu
374 * older than 3.0, since only newer qemu sends streams that
375 * do not trigger spurious decompression errors.
377 bool decompress_error_check
;
379 * This variable only affects behavior when postcopy preempt mode is
384 * - postcopy preempt src QEMU instance will generate an EOS message at
385 * the end of migration to shut the preempt channel on dest side.
387 * - postcopy preempt channel will be created at the setup phase on src
392 * - postcopy preempt src QEMU instance will _not_ generate an EOS
393 * message at the end of migration, the dest qemu will shutdown the
396 * - postcopy preempt channel will be created at the switching phase
397 * from precopy -> postcopy (to avoid race condtion of misordered
398 * creation of channels).
400 * NOTE: See message-id <ZBoShWArKDPpX/D7@work-vm> on qemu-devel
401 * mailing list for more information on the possible race. Everyone
402 * should probably just keep this value untouched after set by the
403 * machine type (or the default).
405 bool preempt_pre_7_2
;
408 * flush every channel after each section sent.
410 * This assures that we can't mix pages from one iteration through
411 * ram pages with pages for the following iteration. We really
412 * only need to do this flush after we have go through all the
413 * dirty pages. For historical reasons, we do that after each
414 * section. This is suboptimal (we flush too many times).
415 * Default value is false. (since 8.1)
417 bool multifd_flush_after_each_section
;
419 * This decides the size of guest memory chunk that will be used
420 * to track dirty bitmap clearing. The size of memory chunk will
421 * be GUEST_PAGE_SIZE << N. Say, N=0 means we will clear dirty
422 * bitmap for each page to send (1<<0=1); N=10 means we will clear
423 * dirty bitmap only once for 1<<10=1K continuous guest pages
424 * (which is in 4M chunk).
426 uint8_t clear_bitmap_shift
;
429 * This save hostname when out-going migration starts
433 /* QEMU_VM_VMDESCRIPTION content filled for all non-iterable devices. */
437 void migrate_set_state(int *state
, int old_state
, int new_state
);
439 void migration_fd_process_incoming(QEMUFile
*f
, Error
**errp
);
440 void migration_ioc_process_incoming(QIOChannel
*ioc
, Error
**errp
);
441 void migration_incoming_process(void);
443 bool migration_has_all_channels(void);
445 uint64_t migrate_max_downtime(void);
447 void migrate_set_error(MigrationState
*s
, const Error
*error
);
448 void migrate_fd_error(MigrationState
*s
, const Error
*error
);
450 void migrate_fd_connect(MigrationState
*s
, Error
*error_in
);
452 bool migration_is_setup_or_active(int state
);
453 bool migration_is_running(int state
);
455 void migrate_init(MigrationState
*s
);
456 bool migration_is_blocked(Error
**errp
);
457 /* True if outgoing migration has entered postcopy phase */
458 bool migration_in_postcopy(void);
459 MigrationState
*migrate_get_current(void);
461 uint64_t ram_get_total_transferred_pages(void);
463 /* Sending on the return path - generic and then for each message type */
464 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
466 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
468 int migrate_send_rp_req_pages(MigrationIncomingState
*mis
, RAMBlock
*rb
,
469 ram_addr_t start
, uint64_t haddr
);
470 int migrate_send_rp_message_req_pages(MigrationIncomingState
*mis
,
471 RAMBlock
*rb
, ram_addr_t start
);
472 void migrate_send_rp_recv_bitmap(MigrationIncomingState
*mis
,
474 void migrate_send_rp_resume_ack(MigrationIncomingState
*mis
, uint32_t value
);
476 void dirty_bitmap_mig_before_vm_start(void);
477 void dirty_bitmap_mig_cancel_outgoing(void);
478 void dirty_bitmap_mig_cancel_incoming(void);
479 bool check_dirty_bitmap_mig_alias_map(const BitmapMigrationNodeAliasList
*bbm
,
482 void migrate_add_address(SocketAddress
*address
);
484 int foreach_not_ignored_block(RAMBlockIterFunc func
, void *opaque
);
486 #define qemu_ram_foreach_block \
487 #warning "Use foreach_not_ignored_block in migration code"
489 void migration_make_urgent_request(void);
490 void migration_consume_urgent_request(void);
491 bool migration_rate_limit(void);
492 void migration_cancel(const Error
*error
);
494 void populate_vfio_info(MigrationInfo
*info
);
495 void postcopy_temp_page_reset(PostcopyTmpPage
*tmp_page
);