xref: /openbmc/qemu/migration/migration.h (revision a5c24e13)
1 /*
2  * QEMU live migration
3  *
4  * Copyright IBM, Corp. 2008
5  *
6  * Authors:
7  *  Anthony Liguori   <aliguori@us.ibm.com>
8  *
9  * This work is licensed under the terms of the GNU GPL, version 2.  See
10  * the COPYING file in the top-level directory.
11  *
12  */
13 
14 #ifndef QEMU_MIGRATION_H
15 #define QEMU_MIGRATION_H
16 
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.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"
28 #include "sysemu/runstate.h"
29 #include "migration/misc.h"
30 
31 struct PostcopyBlocktimeContext;
32 
33 #define  MIGRATION_RESUME_ACK_VALUE  (1)
34 
35 /*
36  * 1<<6=64 pages -> 256K chunk when page size is 4K.  This gives us
37  * the benefit that all the chunks are 64 pages aligned then the
38  * bitmaps are always aligned to LONG.
39  */
40 #define CLEAR_BITMAP_SHIFT_MIN             6
41 /*
42  * 1<<18=256K pages -> 1G chunk when page size is 4K.  This is the
43  * default value to use if no one specified.
44  */
45 #define CLEAR_BITMAP_SHIFT_DEFAULT        18
46 /*
47  * 1<<31=2G pages -> 8T chunk when page size is 4K.  This should be
48  * big enough and make sure we won't overflow easily.
49  */
50 #define CLEAR_BITMAP_SHIFT_MAX            31
51 
52 /* This is an abstraction of a "temp huge page" for postcopy's purpose */
53 typedef struct {
54     /*
55      * This points to a temporary huge page as a buffer for UFFDIO_COPY.  It's
56      * mmap()ed and needs to be freed when cleanup.
57      */
58     void *tmp_huge_page;
59     /*
60      * This points to the host page we're going to install for this temp page.
61      * It tells us after we've received the whole page, where we should put it.
62      */
63     void *host_addr;
64     /* Number of small pages copied (in size of TARGET_PAGE_SIZE) */
65     unsigned int target_pages;
66     /* Whether this page contains all zeros */
67     bool all_zero;
68 } PostcopyTmpPage;
69 
70 typedef enum {
71     PREEMPT_THREAD_NONE = 0,
72     PREEMPT_THREAD_CREATED,
73     PREEMPT_THREAD_QUIT,
74 } PreemptThreadStatus;
75 
76 /* State for the incoming migration */
77 struct MigrationIncomingState {
78     QEMUFile *from_src_file;
79     /* Previously received RAM's RAMBlock pointer */
80     RAMBlock *last_recv_block[RAM_CHANNEL_MAX];
81     /* A hook to allow cleanup at the end of incoming migration */
82     void *transport_data;
83     void (*transport_cleanup)(void *data);
84     /*
85      * Used to sync thread creations.  Note that we can't create threads in
86      * parallel with this sem.
87      */
88     QemuSemaphore  thread_sync_sem;
89     /*
90      * Free at the start of the main state load, set as the main thread finishes
91      * loading state.
92      */
93     QemuEvent main_thread_load_event;
94 
95     /* For network announces */
96     AnnounceTimer  announce_timer;
97 
98     size_t         largest_page_size;
99     bool           have_fault_thread;
100     QemuThread     fault_thread;
101     /* Set this when we want the fault thread to quit */
102     bool           fault_thread_quit;
103 
104     bool           have_listen_thread;
105     QemuThread     listen_thread;
106 
107     /* For the kernel to send us notifications */
108     int       userfault_fd;
109     /* To notify the fault_thread to wake, e.g., when need to quit */
110     int       userfault_event_fd;
111     QEMUFile *to_src_file;
112     QemuMutex rp_mutex;    /* We send replies from multiple threads */
113     /* RAMBlock of last request sent to source */
114     RAMBlock *last_rb;
115     /*
116      * Number of postcopy channels including the default precopy channel, so
117      * vanilla postcopy will only contain one channel which contain both
118      * precopy and postcopy streams.
119      *
120      * This is calculated when the src requests to enable postcopy but before
121      * it starts.  Its value can depend on e.g. whether postcopy preemption is
122      * enabled.
123      */
124     unsigned int postcopy_channels;
125     /* QEMUFile for postcopy only; it'll be handled by a separate thread */
126     QEMUFile *postcopy_qemufile_dst;
127     /*
128      * When postcopy_qemufile_dst is properly setup, this sem is posted.
129      * One can wait on this semaphore to wait until the preempt channel is
130      * properly setup.
131      */
132     QemuSemaphore postcopy_qemufile_dst_done;
133     /* Postcopy priority thread is used to receive postcopy requested pages */
134     QemuThread postcopy_prio_thread;
135     /*
136      * Always set by the main vm load thread only, but can be read by the
137      * postcopy preempt thread.  "volatile" makes sure all reads will be
138      * up-to-date across cores.
139      */
140     volatile PreemptThreadStatus preempt_thread_status;
141     /*
142      * Used to sync between the ram load main thread and the fast ram load
143      * thread.  It protects postcopy_qemufile_dst, which is the postcopy
144      * fast channel.
145      *
146      * The ram fast load thread will take it mostly for the whole lifecycle
147      * because it needs to continuously read data from the channel, and
148      * it'll only release this mutex if postcopy is interrupted, so that
149      * the ram load main thread will take this mutex over and properly
150      * release the broken channel.
151      */
152     QemuMutex postcopy_prio_thread_mutex;
153     /*
154      * An array of temp host huge pages to be used, one for each postcopy
155      * channel.
156      */
157     PostcopyTmpPage *postcopy_tmp_pages;
158     /* This is shared for all postcopy channels */
159     void     *postcopy_tmp_zero_page;
160     /* PostCopyFD's for external userfaultfds & handlers of shared memory */
161     GArray   *postcopy_remote_fds;
162 
163     MigrationStatus state;
164 
165     /*
166      * The incoming migration coroutine, non-NULL during qemu_loadvm_state().
167      * Used to wake the migration incoming coroutine from rdma code. How much is
168      * it safe - it's a question.
169      */
170     Coroutine *loadvm_co;
171 
172     /* The coroutine we should enter (back) after failover */
173     Coroutine *colo_incoming_co;
174     QemuSemaphore colo_incoming_sem;
175 
176     /*
177      * PostcopyBlocktimeContext to keep information for postcopy
178      * live migration, to calculate vCPU block time
179      * */
180     struct PostcopyBlocktimeContext *blocktime_ctx;
181 
182     /* notify PAUSED postcopy incoming migrations to try to continue */
183     QemuSemaphore postcopy_pause_sem_dst;
184     QemuSemaphore postcopy_pause_sem_fault;
185     /*
186      * This semaphore is used to allow the ram fast load thread (only when
187      * postcopy preempt is enabled) fall into sleep when there's network
188      * interruption detected.  When the recovery is done, the main load
189      * thread will kick the fast ram load thread using this semaphore.
190      */
191     QemuSemaphore postcopy_pause_sem_fast_load;
192 
193     /* List of listening socket addresses  */
194     SocketAddressList *socket_address_list;
195 
196     /* A tree of pages that we requested to the source VM */
197     GTree *page_requested;
198     /*
199      * For postcopy only, count the number of requested page faults that
200      * still haven't been resolved.
201      */
202     int page_requested_count;
203     /*
204      * The mutex helps to maintain the requested pages that we sent to the
205      * source, IOW, to guarantee coherent between the page_requests tree and
206      * the per-ramblock receivedmap.  Note! This does not guarantee consistency
207      * of the real page copy procedures (using UFFDIO_[ZERO]COPY).  E.g., even
208      * if one bit in receivedmap is cleared, UFFDIO_COPY could have happened
209      * for that page already.  This is intended so that the mutex won't
210      * serialize and blocked by slow operations like UFFDIO_* ioctls.  However
211      * this should be enough to make sure the page_requested tree always
212      * contains valid information.
213      */
214     QemuMutex page_request_mutex;
215     /*
216      * If postcopy preempt is enabled, there is a chance that the main
217      * thread finished loading its data before the preempt channel has
218      * finished loading the urgent pages.  If that happens, the two threads
219      * will use this condvar to synchronize, so the main thread will always
220      * wait until all pages received.
221      */
222     QemuCond page_request_cond;
223 
224     /*
225      * Number of devices that have yet to approve switchover. When this reaches
226      * zero an ACK that it's OK to do switchover is sent to the source. No lock
227      * is needed as this field is updated serially.
228      */
229     unsigned int switchover_ack_pending_num;
230 
231     /* Do exit on incoming migration failure */
232     bool exit_on_error;
233 };
234 
235 MigrationIncomingState *migration_incoming_get_current(void);
236 void migration_incoming_state_destroy(void);
237 void migration_incoming_transport_cleanup(MigrationIncomingState *mis);
238 /*
239  * Functions to work with blocktime context
240  */
241 void fill_destination_postcopy_migration_info(MigrationInfo *info);
242 
243 #define TYPE_MIGRATION "migration"
244 
245 typedef struct MigrationClass MigrationClass;
246 DECLARE_OBJ_CHECKERS(MigrationState, MigrationClass,
247                      MIGRATION_OBJ, TYPE_MIGRATION)
248 
249 struct MigrationClass {
250     /*< private >*/
251     DeviceClass parent_class;
252 };
253 
254 struct MigrationState {
255     /*< private >*/
256     DeviceState parent_obj;
257 
258     /*< public >*/
259     QemuThread thread;
260     /* Protected by qemu_file_lock */
261     QEMUFile *to_dst_file;
262     /* Postcopy specific transfer channel */
263     QEMUFile *postcopy_qemufile_src;
264     /*
265      * It is posted when the preempt channel is established.  Note: this is
266      * used for both the start or recover of a postcopy migration.  We'll
267      * post to this sem every time a new preempt channel is created in the
268      * main thread, and we keep post() and wait() in pair.
269      */
270     QemuSemaphore postcopy_qemufile_src_sem;
271     QIOChannelBuffer *bioc;
272     /*
273      * Protects to_dst_file/from_dst_file pointers.  We need to make sure we
274      * won't yield or hang during the critical section, since this lock will be
275      * used in OOB command handler.
276      */
277     QemuMutex qemu_file_lock;
278 
279     /*
280      * Used to allow urgent requests to override rate limiting.
281      */
282     QemuSemaphore rate_limit_sem;
283 
284     /* pages already send at the beginning of current iteration */
285     uint64_t iteration_initial_pages;
286 
287     /* pages transferred per second */
288     double pages_per_second;
289 
290     /* bytes already send at the beginning of current iteration */
291     uint64_t iteration_initial_bytes;
292     /* time at the start of current iteration */
293     int64_t iteration_start_time;
294     /*
295      * The final stage happens when the remaining data is smaller than
296      * this threshold; it's calculated from the requested downtime and
297      * measured bandwidth, or avail-switchover-bandwidth if specified.
298      */
299     uint64_t threshold_size;
300 
301     /* params from 'migrate-set-parameters' */
302     MigrationParameters parameters;
303 
304     MigrationStatus state;
305 
306     /* State related to return path */
307     struct {
308         /* Protected by qemu_file_lock */
309         QEMUFile     *from_dst_file;
310         QemuThread    rp_thread;
311         /*
312          * We can also check non-zero of rp_thread, but there's no "official"
313          * way to do this, so this bool makes it slightly more elegant.
314          * Checking from_dst_file for this is racy because from_dst_file will
315          * be cleared in the rp_thread!
316          */
317         bool          rp_thread_created;
318         /*
319          * Used to synchronize between migration main thread and return
320          * path thread.  The migration thread can wait() on this sem, while
321          * other threads (e.g., return path thread) can kick it using a
322          * post().
323          */
324         QemuSemaphore rp_sem;
325         /*
326          * We post to this when we got one PONG from dest. So far it's an
327          * easy way to know the main channel has successfully established
328          * on dest QEMU.
329          */
330         QemuSemaphore rp_pong_acks;
331     } rp_state;
332 
333     double mbps;
334     /* Timestamp when recent migration starts (ms) */
335     int64_t start_time;
336     /* Total time used by latest migration (ms) */
337     int64_t total_time;
338     /* Timestamp when VM is down (ms) to migrate the last stuff */
339     int64_t downtime_start;
340     int64_t downtime;
341     int64_t expected_downtime;
342     bool capabilities[MIGRATION_CAPABILITY__MAX];
343     int64_t setup_time;
344 
345     /*
346      * State before stopping the vm by vm_stop_force_state().
347      * If migration is interrupted by any reason, we need to continue
348      * running the guest on source if it was running or restore its stopped
349      * state.
350      */
351     RunState vm_old_state;
352 
353     /* Flag set once the migration has been asked to enter postcopy */
354     bool start_postcopy;
355 
356     /* Flag set once the migration thread is running (and needs joining) */
357     bool migration_thread_running;
358 
359     /* Flag set once the migration thread called bdrv_inactivate_all */
360     bool block_inactive;
361 
362     /* Migration is waiting for guest to unplug device */
363     QemuSemaphore wait_unplug_sem;
364 
365     /* Migration is paused due to pause-before-switchover */
366     QemuSemaphore pause_sem;
367 
368     /* The semaphore is used to notify COLO thread that failover is finished */
369     QemuSemaphore colo_exit_sem;
370 
371     /* The event is used to notify COLO thread to do checkpoint */
372     QemuEvent colo_checkpoint_event;
373     int64_t colo_checkpoint_time;
374     QEMUTimer *colo_delay_timer;
375 
376     /* The first error that has occurred.
377        We used the mutex to be able to return the 1st error message */
378     Error *error;
379     /* mutex to protect errp */
380     QemuMutex error_mutex;
381 
382     /*
383      * Global switch on whether we need to store the global state
384      * during migration.
385      */
386     bool store_global_state;
387 
388     /* Whether we send QEMU_VM_CONFIGURATION during migration */
389     bool send_configuration;
390     /* Whether we send section footer during migration */
391     bool send_section_footer;
392 
393     /* Needed by postcopy-pause state */
394     QemuSemaphore postcopy_pause_sem;
395     /*
396      * This variable only affects behavior when postcopy preempt mode is
397      * enabled.
398      *
399      * When set:
400      *
401      * - postcopy preempt src QEMU instance will generate an EOS message at
402      *   the end of migration to shut the preempt channel on dest side.
403      *
404      * - postcopy preempt channel will be created at the setup phase on src
405          QEMU.
406      *
407      * When clear:
408      *
409      * - postcopy preempt src QEMU instance will _not_ generate an EOS
410      *   message at the end of migration, the dest qemu will shutdown the
411      *   channel itself.
412      *
413      * - postcopy preempt channel will be created at the switching phase
414      *   from precopy -> postcopy (to avoid race condition of misordered
415      *   creation of channels).
416      *
417      * NOTE: See message-id <ZBoShWArKDPpX/D7@work-vm> on qemu-devel
418      * mailing list for more information on the possible race.  Everyone
419      * should probably just keep this value untouched after set by the
420      * machine type (or the default).
421      */
422     bool preempt_pre_7_2;
423 
424     /*
425      * flush every channel after each section sent.
426      *
427      * This assures that we can't mix pages from one iteration through
428      * ram pages with pages for the following iteration.  We really
429      * only need to do this flush after we have go through all the
430      * dirty pages.  For historical reasons, we do that after each
431      * section.  This is suboptimal (we flush too many times).
432      * Default value is false. (since 8.1)
433      */
434     bool multifd_flush_after_each_section;
435     /*
436      * This decides the size of guest memory chunk that will be used
437      * to track dirty bitmap clearing.  The size of memory chunk will
438      * be GUEST_PAGE_SIZE << N.  Say, N=0 means we will clear dirty
439      * bitmap for each page to send (1<<0=1); N=10 means we will clear
440      * dirty bitmap only once for 1<<10=1K continuous guest pages
441      * (which is in 4M chunk).
442      */
443     uint8_t clear_bitmap_shift;
444 
445     /*
446      * This save hostname when out-going migration starts
447      */
448     char *hostname;
449 
450     /* QEMU_VM_VMDESCRIPTION content filled for all non-iterable devices. */
451     JSONWriter *vmdesc;
452 
453     /*
454      * Indicates whether an ACK from the destination that it's OK to do
455      * switchover has been received.
456      */
457     bool switchover_acked;
458     /* Is this a rdma migration */
459     bool rdma_migration;
460 };
461 
462 void migrate_set_state(MigrationStatus *state, MigrationStatus old_state,
463                        MigrationStatus new_state);
464 
465 void migration_fd_process_incoming(QEMUFile *f);
466 void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp);
467 void migration_incoming_process(void);
468 
469 bool  migration_has_all_channels(void);
470 
471 void migrate_set_error(MigrationState *s, const Error *error);
472 bool migrate_has_error(MigrationState *s);
473 
474 void migrate_fd_connect(MigrationState *s, Error *error_in);
475 
476 int migration_call_notifiers(MigrationState *s, MigrationEventType type,
477                              Error **errp);
478 
479 int migrate_init(MigrationState *s, Error **errp);
480 bool migration_is_blocked(Error **errp);
481 /* True if outgoing migration has entered postcopy phase */
482 bool migration_in_postcopy(void);
483 bool migration_postcopy_is_alive(MigrationStatus state);
484 MigrationState *migrate_get_current(void);
485 bool migration_has_failed(MigrationState *);
486 bool migrate_mode_is_cpr(MigrationState *);
487 
488 uint64_t ram_get_total_transferred_pages(void);
489 
490 /* Sending on the return path - generic and then for each message type */
491 void migrate_send_rp_shut(MigrationIncomingState *mis,
492                           uint32_t value);
493 void migrate_send_rp_pong(MigrationIncomingState *mis,
494                           uint32_t value);
495 int migrate_send_rp_req_pages(MigrationIncomingState *mis, RAMBlock *rb,
496                               ram_addr_t start, uint64_t haddr);
497 int migrate_send_rp_message_req_pages(MigrationIncomingState *mis,
498                                       RAMBlock *rb, ram_addr_t start);
499 void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis,
500                                  char *block_name);
501 void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value);
502 int migrate_send_rp_switchover_ack(MigrationIncomingState *mis);
503 
504 void dirty_bitmap_mig_before_vm_start(void);
505 void dirty_bitmap_mig_cancel_outgoing(void);
506 void dirty_bitmap_mig_cancel_incoming(void);
507 bool check_dirty_bitmap_mig_alias_map(const BitmapMigrationNodeAliasList *bbm,
508                                       Error **errp);
509 
510 void migrate_add_address(SocketAddress *address);
511 bool migrate_uri_parse(const char *uri, MigrationChannel **channel,
512                        Error **errp);
513 int foreach_not_ignored_block(RAMBlockIterFunc func, void *opaque);
514 
515 #define qemu_ram_foreach_block \
516   #warning "Use foreach_not_ignored_block in migration code"
517 
518 void migration_make_urgent_request(void);
519 void migration_consume_urgent_request(void);
520 bool migration_rate_limit(void);
521 void migration_bh_schedule(QEMUBHFunc *cb, void *opaque);
522 void migration_cancel(const Error *error);
523 
524 void migration_populate_vfio_info(MigrationInfo *info);
525 void migration_reset_vfio_bytes_transferred(void);
526 void postcopy_temp_page_reset(PostcopyTmpPage *tmp_page);
527 
528 /*
529  * Migration thread waiting for return path thread.  Return non-zero if an
530  * error is detected.
531  */
532 int migration_rp_wait(MigrationState *s);
533 /*
534  * Kick the migration thread waiting for return path messages.  NOTE: the
535  * name can be slightly confusing (when read as "kick the rp thread"), just
536  * to remember the target is always the migration thread.
537  */
538 void migration_rp_kick(MigrationState *s);
539 
540 #endif
541