-
Notifications
You must be signed in to change notification settings - Fork 576
/
amdgpu_plugin.c
1999 lines (1658 loc) · 53.2 KB
/
amdgpu_plugin.c
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
#include <errno.h>
#include <fcntl.h>
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
#include <unistd.h>
#include <linux/limits.h>
#include <sys/ioctl.h>
#include <sys/stat.h>
#include <sys/sysmacros.h>
#include <sys/mman.h>
#include <sys/types.h>
#include <stdint.h>
#include <pthread.h>
#include <semaphore.h>
#include <xf86drm.h>
#include <libdrm/amdgpu.h>
#include <libdrm/amdgpu_drm.h>
#include "criu-plugin.h"
#include "plugin.h"
#include "criu-amdgpu.pb-c.h"
#include "kfd_ioctl.h"
#include "xmalloc.h"
#include "criu-log.h"
#include "files.h"
#include "common/list.h"
#include "amdgpu_plugin_topology.h"
#include "img-streamer.h"
#include "image.h"
#include "cr_options.h"
#define AMDGPU_KFD_DEVICE "/dev/kfd"
#define PROCPIDMEM "/proc/%d/mem"
#define HSAKMT_SHM_PATH "/dev/shm/hsakmt_shared_mem"
#define HSAKMT_SHM "/hsakmt_shared_mem"
#define HSAKMT_SEM_PATH "/dev/shm/sem.hsakmt_semaphore"
#define HSAKMT_SEM "hsakmt_semaphore"
#define KFD_IOCTL_MAJOR_VERSION 1
#define MIN_KFD_IOCTL_MINOR_VERSION 8
#define IMG_KFD_FILE "amdgpu-kfd-%d.img"
#define IMG_RENDERD_FILE "amdgpu-renderD-%d.img"
#define IMG_PAGES_FILE "amdgpu-pages-%d-%04x.img"
#ifndef _GNU_SOURCE
#define _GNU_SOURCE 1
#endif
#ifdef LOG_PREFIX
#undef LOG_PREFIX
#endif
#define LOG_PREFIX "amdgpu_plugin: "
#ifdef DEBUG
#define plugin_log_msg(fmt, ...) pr_debug(fmt, ##__VA_ARGS__)
#else
#define plugin_log_msg(fmt, ...) \
{ \
}
#endif
#define SDMA_PACKET(op, sub_op, e) ((((e)&0xFFFF) << 16) | (((sub_op)&0xFF) << 8) | (((op)&0xFF) << 0))
#define SDMA_OPCODE_COPY 1
#define SDMA_COPY_SUB_OPCODE_LINEAR 0
#define SDMA_NOP 0
#define SDMA_LINEAR_COPY_MAX_SIZE (1ULL << 21)
enum sdma_op_type {
SDMA_OP_VRAM_READ,
SDMA_OP_VRAM_WRITE,
};
struct vma_metadata {
struct list_head list;
uint64_t old_pgoff;
uint64_t new_pgoff;
uint64_t vma_entry;
uint32_t new_minor;
int fd;
};
/************************************ Global Variables ********************************************/
struct tp_system src_topology;
struct tp_system dest_topology;
struct device_maps checkpoint_maps;
struct device_maps restore_maps;
extern int fd_next;
static LIST_HEAD(update_vma_info_list);
extern bool kfd_fw_version_check;
extern bool kfd_sdma_fw_version_check;
extern bool kfd_caches_count_check;
extern bool kfd_num_gws_check;
extern bool kfd_vram_size_check;
extern bool kfd_numa_check;
extern bool kfd_capability_check;
size_t kfd_max_buffer_size;
/**************************************************************************************************/
int write_fp(FILE *fp, const void *buf, const size_t buf_len)
{
size_t len_write;
len_write = fwrite(buf, 1, buf_len, fp);
if (len_write != buf_len) {
pr_perror("Unable to write file (wrote:%ld buf_len:%ld)", len_write, buf_len);
return -EIO;
}
return 0;
}
int read_fp(FILE *fp, void *buf, const size_t buf_len)
{
size_t len_read;
len_read = fread(buf, 1, buf_len, fp);
if (len_read != buf_len) {
pr_perror("Unable to read file (read:%ld buf_len:%ld)", len_read, buf_len);
return -EIO;
}
return 0;
}
/**
* @brief Open an image file
*
* We store the size of the actual contents in the first 8-bytes of the file. This allows us to
* determine the file size when using criu_image_streamer when fseek and fstat are not available.
* The FILE * returned is already at the location of the first actual contents.
*
* @param path The file path
* @param write False for read, true for write
* @param size Size of actual contents
* @return FILE *if successful, NULL if failed
*/
FILE *open_img_file(char *path, bool write, size_t *size)
{
FILE *fp = NULL;
int fd, ret;
if (opts.stream)
fd = img_streamer_open(path, write ? O_DUMP : O_RSTR);
else
fd = openat(criu_get_image_dir(), path, write ? (O_WRONLY | O_CREAT) : O_RDONLY, 0600);
if (fd < 0) {
pr_perror("%s: Failed to open for %s", path, write ? "write" : "read");
return NULL;
}
fp = fdopen(fd, write ? "w" : "r");
if (!fp) {
pr_perror("%s: Failed get pointer for %s", path, write ? "write" : "read");
return NULL;
}
if (write)
ret = write_fp(fp, size, sizeof(*size));
else
ret = read_fp(fp, size, sizeof(*size));
if (ret) {
pr_perror("%s:Failed to access file size", path);
fclose(fp);
return NULL;
}
pr_debug("%s:Opened file for %s with size:%ld\n", path, write ? "write" : "read", *size);
return fp;
}
/**
* @brief Write an image file
*
* We store the size of the actual contents in the first 8-bytes of the file. This allows us to
* determine the file size when using criu_image_streamer when fseek and fstat are not available.
*
* @param path The file path
* @param buf pointer to data to be written
* @param buf_len size of buf
* @return 0 if successful. -errno on failure
*/
int write_img_file(char *path, const void *buf, const size_t buf_len)
{
int ret;
FILE *fp;
size_t len = buf_len;
fp = open_img_file(path, true, &len);
if (!fp)
return -errno;
ret = write_fp(fp, buf, buf_len);
fclose(fp); /* this will also close fd */
return ret;
}
int read_file(const char *file_path, void *buf, const size_t buf_len)
{
int ret;
FILE *fp;
fp = fopen(file_path, "r");
if (!fp) {
pr_perror("Cannot fopen %s", file_path);
return -errno;
}
ret = read_fp(fp, buf, buf_len);
fclose(fp); /* this will also close fd */
return ret;
}
/* Call ioctl, restarting if it is interrupted */
int kmtIoctl(int fd, unsigned long request, void *arg)
{
int ret, max_retries = 200;
do {
ret = ioctl(fd, request, arg);
} while (ret == -1 && max_retries-- > 0 && (errno == EINTR || errno == EAGAIN));
if (ret == -1 && errno == EBADF)
/* In case pthread_atfork didn't catch it, this will
* make any subsequent hsaKmt calls fail in CHECK_KFD_OPEN.
*/
pr_perror("KFD file descriptor not valid in this process");
return ret;
}
static void free_e(CriuKfd *e)
{
for (int i = 0; i < e->n_bo_entries; i++) {
if (e->bo_entries[i])
xfree(e->bo_entries[i]);
}
for (int i = 0; i < e->n_device_entries; i++) {
if (e->device_entries[i]) {
for (int j = 0; j < e->device_entries[i]->n_iolinks; j++)
xfree(e->device_entries[i]->iolinks[j]);
xfree(e->device_entries[i]);
}
}
xfree(e);
}
static int allocate_device_entries(CriuKfd *e, int num_of_devices)
{
e->device_entries = xmalloc(sizeof(DeviceEntry *) * num_of_devices);
if (!e->device_entries) {
pr_err("Failed to allocate device_entries\n");
return -ENOMEM;
}
for (int i = 0; i < num_of_devices; i++) {
DeviceEntry *entry = xzalloc(sizeof(*entry));
if (!entry) {
pr_err("Failed to allocate entry\n");
return -ENOMEM;
}
device_entry__init(entry);
e->device_entries[i] = entry;
e->n_device_entries++;
}
return 0;
}
static int allocate_bo_entries(CriuKfd *e, int num_bos, struct kfd_criu_bo_bucket *bo_bucket_ptr)
{
e->bo_entries = xmalloc(sizeof(BoEntry *) * num_bos);
if (!e->bo_entries) {
pr_err("Failed to allocate bo_info\n");
return -ENOMEM;
}
for (int i = 0; i < num_bos; i++) {
BoEntry *entry = xzalloc(sizeof(*entry));
if (!entry) {
pr_err("Failed to allocate botest\n");
return -ENOMEM;
}
bo_entry__init(entry);
e->bo_entries[i] = entry;
e->n_bo_entries++;
}
return 0;
}
int topology_to_devinfo(struct tp_system *sys, struct device_maps *maps, DeviceEntry **deviceEntries)
{
uint32_t devinfo_index = 0;
struct tp_node *node;
list_for_each_entry(node, &sys->nodes, listm_system) {
DeviceEntry *devinfo = deviceEntries[devinfo_index++];
devinfo->node_id = node->id;
if (NODE_IS_GPU(node)) {
devinfo->gpu_id = maps_get_dest_gpu(maps, node->gpu_id);
if (!devinfo->gpu_id)
return -EINVAL;
devinfo->simd_count = node->simd_count;
devinfo->mem_banks_count = node->mem_banks_count;
devinfo->caches_count = node->caches_count;
devinfo->io_links_count = node->io_links_count;
devinfo->max_waves_per_simd = node->max_waves_per_simd;
devinfo->lds_size_in_kb = node->lds_size_in_kb;
devinfo->num_gws = node->num_gws;
devinfo->wave_front_size = node->wave_front_size;
devinfo->array_count = node->array_count;
devinfo->simd_arrays_per_engine = node->simd_arrays_per_engine;
devinfo->cu_per_simd_array = node->cu_per_simd_array;
devinfo->simd_per_cu = node->simd_per_cu;
devinfo->max_slots_scratch_cu = node->max_slots_scratch_cu;
devinfo->vendor_id = node->vendor_id;
devinfo->device_id = node->device_id;
devinfo->domain = node->domain;
devinfo->drm_render_minor = node->drm_render_minor;
devinfo->hive_id = node->hive_id;
devinfo->num_sdma_engines = node->num_sdma_engines;
devinfo->num_sdma_xgmi_engines = node->num_sdma_xgmi_engines;
devinfo->num_sdma_queues_per_engine = node->num_sdma_queues_per_engine;
devinfo->num_cp_queues = node->num_cp_queues;
devinfo->fw_version = node->fw_version;
devinfo->capability = node->capability;
devinfo->sdma_fw_version = node->sdma_fw_version;
devinfo->vram_public = node->vram_public;
devinfo->vram_size = node->vram_size;
} else {
devinfo->cpu_cores_count = node->cpu_cores_count;
}
if (node->num_valid_iolinks) {
struct tp_iolink *iolink;
uint32_t iolink_index = 0;
devinfo->iolinks = xmalloc(sizeof(DevIolink *) * node->num_valid_iolinks);
if (!devinfo->iolinks)
return -ENOMEM;
list_for_each_entry(iolink, &node->iolinks, listm) {
if (!iolink->valid)
continue;
devinfo->iolinks[iolink_index] = xmalloc(sizeof(DevIolink));
if (!devinfo->iolinks[iolink_index])
return -ENOMEM;
dev_iolink__init(devinfo->iolinks[iolink_index]);
devinfo->iolinks[iolink_index]->type = iolink->type;
devinfo->iolinks[iolink_index]->node_to_id = iolink->node_to_id;
iolink_index++;
}
devinfo->n_iolinks = iolink_index;
}
}
return 0;
}
int devinfo_to_topology(DeviceEntry *devinfos[], uint32_t num_devices, struct tp_system *sys)
{
for (int i = 0; i < num_devices; i++) {
struct tp_node *node;
DeviceEntry *devinfo = devinfos[i];
node = sys_add_node(sys, devinfo->node_id, devinfo->gpu_id);
if (!node)
return -ENOMEM;
if (devinfo->cpu_cores_count) {
node->cpu_cores_count = devinfo->cpu_cores_count;
} else {
node->simd_count = devinfo->simd_count;
node->mem_banks_count = devinfo->mem_banks_count;
node->caches_count = devinfo->caches_count;
node->io_links_count = devinfo->io_links_count;
node->max_waves_per_simd = devinfo->max_waves_per_simd;
node->lds_size_in_kb = devinfo->lds_size_in_kb;
node->num_gws = devinfo->num_gws;
node->wave_front_size = devinfo->wave_front_size;
node->array_count = devinfo->array_count;
node->simd_arrays_per_engine = devinfo->simd_arrays_per_engine;
node->cu_per_simd_array = devinfo->cu_per_simd_array;
node->simd_per_cu = devinfo->simd_per_cu;
node->max_slots_scratch_cu = devinfo->max_slots_scratch_cu;
node->vendor_id = devinfo->vendor_id;
node->device_id = devinfo->device_id;
node->domain = devinfo->domain;
node->drm_render_minor = devinfo->drm_render_minor;
node->hive_id = devinfo->hive_id;
node->num_sdma_engines = devinfo->num_sdma_engines;
node->num_sdma_xgmi_engines = devinfo->num_sdma_xgmi_engines;
node->num_sdma_queues_per_engine = devinfo->num_sdma_queues_per_engine;
node->num_cp_queues = devinfo->num_cp_queues;
node->fw_version = devinfo->fw_version;
node->capability = devinfo->capability;
node->sdma_fw_version = devinfo->sdma_fw_version;
node->vram_public = devinfo->vram_public;
node->vram_size = devinfo->vram_size;
}
for (int j = 0; j < devinfo->n_iolinks; j++) {
struct tp_iolink *iolink;
DevIolink *devlink = (devinfo->iolinks[j]);
iolink = node_add_iolink(node, devlink->type, devlink->node_to_id);
if (!iolink)
return -ENOMEM;
}
}
return 0;
}
void getenv_bool(const char *var, bool *value)
{
char *value_str = getenv(var);
if (value_str) {
if (!strcmp(value_str, "0") || !strcasecmp(value_str, "NO"))
*value = false;
else if (!strcmp(value_str, "1") || !strcasecmp(value_str, "YES"))
*value = true;
else
pr_err("Ignoring invalid value for %s=%s, expecting (YES/NO)\n", var, value_str);
}
pr_info("param: %s:%s\n", var, *value ? "Y" : "N");
}
void getenv_size_t(const char *var, size_t *value)
{
char *value_str = getenv(var);
char *endp = value_str;
int sh = 0;
size_t size;
pr_info("Value str: %s\n", value_str);
if (value_str) {
size = (size_t)strtoul(value_str, &endp, 0);
if (errno || value_str == endp) {
pr_err("Ignoring invalid value for %s=%s, expecting a positive integer\n", var, value_str);
return;
}
switch (*endp) {
case 'k':
case 'K':
sh = 10;
break;
case 'M':
sh = 20;
break;
case 'G':
sh = 30;
break;
case '\0':
sh = 0;
break;
default:
pr_err("Ignoring invalid size suffix for %s=%s, expecting 'K'/k', 'M', or 'G'\n", var, value_str);
return;
}
if (SIZE_MAX >> sh < size) {
pr_err("Ignoring invalid value for %s=%s, exceeds SIZE_MAX\n", var, value_str);
return;
}
*value = size << sh;
}
pr_info("param: %s:0x%lx\n", var, *value);
}
int amdgpu_plugin_init(int stage)
{
pr_info("initialized: %s (AMDGPU/KFD)\n", CR_PLUGIN_DESC.name);
topology_init(&src_topology);
topology_init(&dest_topology);
maps_init(&checkpoint_maps);
maps_init(&restore_maps);
if (stage == CR_PLUGIN_STAGE__RESTORE) {
/* Default Values */
kfd_fw_version_check = true;
kfd_sdma_fw_version_check = true;
kfd_caches_count_check = true;
kfd_num_gws_check = true;
kfd_vram_size_check = true;
kfd_numa_check = true;
kfd_capability_check = true;
getenv_bool("KFD_FW_VER_CHECK", &kfd_fw_version_check);
getenv_bool("KFD_SDMA_FW_VER_CHECK", &kfd_sdma_fw_version_check);
getenv_bool("KFD_CACHES_COUNT_CHECK", &kfd_caches_count_check);
getenv_bool("KFD_NUM_GWS_CHECK", &kfd_num_gws_check);
getenv_bool("KFD_VRAM_SIZE_CHECK", &kfd_vram_size_check);
getenv_bool("KFD_NUMA_CHECK", &kfd_numa_check);
getenv_bool("KFD_CAPABILITY_CHECK", &kfd_capability_check);
}
kfd_max_buffer_size = 0;
getenv_size_t("KFD_MAX_BUFFER_SIZE", &kfd_max_buffer_size);
return 0;
}
void amdgpu_plugin_fini(int stage, int ret)
{
pr_info("finished %s (AMDGPU/KFD)\n", CR_PLUGIN_DESC.name);
if (stage == CR_PLUGIN_STAGE__RESTORE)
sys_close_drm_render_devices(&dest_topology);
maps_free(&checkpoint_maps);
maps_free(&restore_maps);
topology_free(&src_topology);
topology_free(&dest_topology);
}
CR_PLUGIN_REGISTER("amdgpu_plugin", amdgpu_plugin_init, amdgpu_plugin_fini)
struct thread_data {
pthread_t thread;
uint64_t num_of_bos;
uint32_t gpu_id;
pid_t pid;
struct kfd_criu_bo_bucket *bo_buckets;
BoEntry **bo_entries;
int drm_fd;
int ret;
int id; /* File ID used by CRIU to identify KFD image for this process */
};
int amdgpu_plugin_handle_device_vma(int fd, const struct stat *st_buf)
{
struct stat st_kfd, st_dri_min;
char img_path[128];
int ret = 0;
pr_debug("Enter %s\n", __func__);
ret = stat(AMDGPU_KFD_DEVICE, &st_kfd);
if (ret == -1) {
pr_perror("stat error for /dev/kfd");
return ret;
}
snprintf(img_path, sizeof(img_path), "/dev/dri/renderD%d", DRM_FIRST_RENDER_NODE);
ret = stat(img_path, &st_dri_min);
if (ret == -1) {
pr_perror("stat error for %s", img_path);
return ret;
}
if (major(st_buf->st_rdev) == major(st_kfd.st_rdev) || ((major(st_buf->st_rdev) == major(st_dri_min.st_rdev)) &&
(minor(st_buf->st_rdev) >= minor(st_dri_min.st_rdev) &&
minor(st_buf->st_rdev) >= DRM_FIRST_RENDER_NODE))) {
pr_debug("Known non-regular mapping, kfd-renderD%d -> OK\n", minor(st_buf->st_rdev));
pr_debug("AMD KFD(maj) = %d, DRI(maj,min) = %d:%d VMA Device fd(maj,min) = %d:%d\n",
major(st_kfd.st_rdev), major(st_dri_min.st_rdev), minor(st_dri_min.st_rdev),
major(st_buf->st_rdev), minor(st_buf->st_rdev));
/* VMA belongs to kfd */
return 0;
}
pr_perror("Can't handle the VMA mapping");
return -ENOTSUP;
}
CR_PLUGIN_REGISTER_HOOK(CR_PLUGIN_HOOK__HANDLE_DEVICE_VMA, amdgpu_plugin_handle_device_vma)
int alloc_and_map(amdgpu_device_handle h_dev, uint64_t size, uint32_t domain, amdgpu_bo_handle *ph_bo,
amdgpu_va_handle *ph_va, uint64_t *p_gpu_addr, void **p_cpu_addr)
{
struct amdgpu_bo_alloc_request alloc_req;
amdgpu_bo_handle h_bo;
amdgpu_va_handle h_va;
uint64_t gpu_addr;
void *cpu_addr;
int err;
memset(&alloc_req, 0, sizeof(alloc_req));
alloc_req.alloc_size = size;
alloc_req.phys_alignment = 0x1000;
alloc_req.preferred_heap = domain;
alloc_req.flags = 0;
err = amdgpu_bo_alloc(h_dev, &alloc_req, &h_bo);
if (err) {
pr_perror("failed to alloc BO");
return err;
}
err = amdgpu_va_range_alloc(h_dev, amdgpu_gpu_va_range_general, size, 0x1000, 0, &gpu_addr, &h_va, 0);
if (err) {
pr_perror("failed to alloc VA");
goto err_va;
}
err = amdgpu_bo_va_op(h_bo, 0, size, gpu_addr, 0, AMDGPU_VA_OP_MAP);
if (err) {
pr_perror("failed to GPU map BO");
goto err_gpu_map;
}
if (p_cpu_addr) {
err = amdgpu_bo_cpu_map(h_bo, &cpu_addr);
if (err) {
pr_perror("failed to CPU map BO");
goto err_cpu_map;
}
*p_cpu_addr = cpu_addr;
}
*ph_bo = h_bo;
*ph_va = h_va;
*p_gpu_addr = gpu_addr;
return 0;
err_cpu_map:
amdgpu_bo_va_op(h_bo, 0, size, gpu_addr, 0, AMDGPU_VA_OP_UNMAP);
err_gpu_map:
amdgpu_va_range_free(h_va);
err_va:
amdgpu_bo_free(h_bo);
return err;
}
void free_and_unmap(uint64_t size, amdgpu_bo_handle h_bo, amdgpu_va_handle h_va, uint64_t gpu_addr, void *cpu_addr)
{
if (cpu_addr)
amdgpu_bo_cpu_unmap(h_bo);
amdgpu_bo_va_op(h_bo, 0, size, gpu_addr, 0, AMDGPU_VA_OP_UNMAP);
amdgpu_va_range_free(h_va);
amdgpu_bo_free(h_bo);
}
int sdma_copy_bo(struct kfd_criu_bo_bucket bo_bucket, FILE *storage_fp, void *buffer, size_t buffer_size,
amdgpu_device_handle h_dev, uint64_t max_copy_size, enum sdma_op_type type)
{
uint64_t size, src_bo_size, dst_bo_size, buffer_bo_size, bytes_remain, buffer_space_remain;
uint64_t gpu_addr_src, gpu_addr_dst, gpu_addr_ib, copy_src, copy_dst, copy_size;
amdgpu_va_handle h_va_src, h_va_dst, h_va_ib;
amdgpu_bo_handle h_bo_src, h_bo_dst, h_bo_ib;
struct amdgpu_bo_import_result res = { 0 };
struct amdgpu_cs_ib_info ib_info;
amdgpu_bo_list_handle h_bo_list;
struct amdgpu_cs_request cs_req;
amdgpu_bo_handle resources[3];
struct amdgpu_cs_fence fence;
uint32_t expired;
amdgpu_context_handle h_ctx;
uint32_t *ib = NULL;
int j, err, shared_fd, packets_per_buffer;
shared_fd = bo_bucket.dmabuf_fd;
size = bo_bucket.size;
buffer_bo_size = min(size, buffer_size);
packets_per_buffer = ((buffer_bo_size - 1) / max_copy_size) + 1;
src_bo_size = (type == SDMA_OP_VRAM_WRITE) ? buffer_bo_size : size;
dst_bo_size = (type == SDMA_OP_VRAM_READ) ? buffer_bo_size : size;
plugin_log_msg("Enter %s\n", __func__);
/* prepare src buffer */
switch (type) {
case SDMA_OP_VRAM_WRITE:
err = amdgpu_create_bo_from_user_mem(h_dev, buffer, src_bo_size, &h_bo_src);
if (err) {
pr_perror("failed to create userptr for sdma");
return -EFAULT;
}
break;
case SDMA_OP_VRAM_READ:
err = amdgpu_bo_import(h_dev, amdgpu_bo_handle_type_dma_buf_fd, shared_fd, &res);
if (err) {
pr_perror("failed to import dmabuf handle from libdrm");
return -EFAULT;
}
h_bo_src = res.buf_handle;
break;
default:
pr_perror("Invalid sdma operation");
return -EINVAL;
}
err = amdgpu_va_range_alloc(h_dev, amdgpu_gpu_va_range_general, src_bo_size, 0x1000, 0, &gpu_addr_src,
&h_va_src, 0);
if (err) {
pr_perror("failed to alloc VA for src bo");
goto err_src_va;
}
err = amdgpu_bo_va_op(h_bo_src, 0, src_bo_size, gpu_addr_src, 0, AMDGPU_VA_OP_MAP);
if (err) {
pr_perror("failed to GPU map the src BO");
goto err_src_bo_map;
}
plugin_log_msg("Source BO: GPU VA: %lx, size: %lx\n", gpu_addr_src, src_bo_size);
/* prepare dest buffer */
switch (type) {
case SDMA_OP_VRAM_WRITE:
err = amdgpu_bo_import(h_dev, amdgpu_bo_handle_type_dma_buf_fd, shared_fd, &res);
if (err) {
pr_perror("failed to import dmabuf handle from libdrm");
goto err_dst_bo_prep;
}
h_bo_dst = res.buf_handle;
break;
case SDMA_OP_VRAM_READ:
err = amdgpu_create_bo_from_user_mem(h_dev, buffer, dst_bo_size, &h_bo_dst);
if (err) {
pr_perror("failed to create userptr for sdma");
goto err_dst_bo_prep;
}
break;
default:
pr_perror("Invalid sdma operation");
goto err_dst_bo_prep;
}
err = amdgpu_va_range_alloc(h_dev, amdgpu_gpu_va_range_general, dst_bo_size, 0x1000, 0, &gpu_addr_dst,
&h_va_dst, 0);
if (err) {
pr_perror("failed to alloc VA for dest bo");
goto err_dst_va;
}
err = amdgpu_bo_va_op(h_bo_dst, 0, dst_bo_size, gpu_addr_dst, 0, AMDGPU_VA_OP_MAP);
if (err) {
pr_perror("failed to GPU map the dest BO");
goto err_dst_bo_map;
}
plugin_log_msg("Dest BO: GPU VA: %lx, size: %lx\n", gpu_addr_dst, dst_bo_size);
/* prepare ring buffer/indirect buffer for command submission
* each copy packet is 7 dwords so we need to alloc 28x size for ib
*/
err = alloc_and_map(h_dev, packets_per_buffer * 28, AMDGPU_GEM_DOMAIN_GTT, &h_bo_ib, &h_va_ib, &gpu_addr_ib,
(void **)&ib);
if (err) {
pr_perror("failed to allocate and map ib/rb");
goto err_ib_gpu_alloc;
}
plugin_log_msg("Indirect BO: GPU VA: %lx, size: %lx\n", gpu_addr_ib, packets_per_buffer * 28);
resources[0] = h_bo_src;
resources[1] = h_bo_dst;
resources[2] = h_bo_ib;
err = amdgpu_bo_list_create(h_dev, 3, resources, NULL, &h_bo_list);
if (err) {
pr_perror("failed to create BO resources list");
goto err_bo_list;
}
bytes_remain = size;
if (type == SDMA_OP_VRAM_WRITE)
copy_dst = gpu_addr_dst;
else
copy_src = gpu_addr_src;
while (bytes_remain > 0) {
memset(&cs_req, 0, sizeof(cs_req));
memset(&fence, 0, sizeof(fence));
memset(ib, 0, packets_per_buffer * 28);
if (type == SDMA_OP_VRAM_WRITE) {
err = read_fp(storage_fp, buffer, min(bytes_remain, buffer_bo_size));
if (err) {
pr_perror("failed to read from storage");
goto err_bo_list;
}
}
buffer_space_remain = buffer_bo_size;
if (type == SDMA_OP_VRAM_WRITE)
copy_src = gpu_addr_src;
else
copy_dst = gpu_addr_dst;
j = 0;
while (bytes_remain > 0 && buffer_space_remain > 0) {
copy_size = min(min(bytes_remain, max_copy_size), buffer_space_remain);
ib[j++] = SDMA_PACKET(SDMA_OPCODE_COPY, SDMA_COPY_SUB_OPCODE_LINEAR, 0);
ib[j++] = copy_size;
ib[j++] = 0;
ib[j++] = 0xffffffff & copy_src;
ib[j++] = (0xffffffff00000000 & copy_src) >> 32;
ib[j++] = 0xffffffff & copy_dst;
ib[j++] = (0xffffffff00000000 & copy_dst) >> 32;
copy_src += copy_size;
copy_dst += copy_size;
bytes_remain -= copy_size;
buffer_space_remain -= copy_size;
}
/* pad the IB to the required number of dw with SDMA_NOP */
while (j & 7)
ib[j++] = SDMA_NOP;
ib_info.ib_mc_address = gpu_addr_ib;
ib_info.size = j;
cs_req.ip_type = AMDGPU_HW_IP_DMA;
/* possible future optimization: may use other rings, info available in
* amdgpu_query_hw_ip_info()
*/
cs_req.ring = 0;
cs_req.number_of_ibs = 1;
cs_req.ibs = &ib_info;
cs_req.resources = h_bo_list;
cs_req.fence_info.handle = NULL;
err = amdgpu_cs_ctx_create(h_dev, &h_ctx);
if (err) {
pr_perror("failed to create context for SDMA command submission");
goto err_ctx;
}
err = amdgpu_cs_submit(h_ctx, 0, &cs_req, 1);
if (err) {
pr_perror("failed to submit command for SDMA IB");
goto err_cs_submit_ib;
}
fence.context = h_ctx;
fence.ip_type = AMDGPU_HW_IP_DMA;
fence.ip_instance = 0;
fence.ring = 0;
fence.fence = cs_req.seq_no;
err = amdgpu_cs_query_fence_status(&fence, AMDGPU_TIMEOUT_INFINITE, 0, &expired);
if (err) {
pr_perror("failed to query fence status");
goto err_cs_submit_ib;
}
if (!expired) {
pr_err("IB execution did not complete\n");
err = -EBUSY;
goto err_cs_submit_ib;
}
if (type == SDMA_OP_VRAM_READ) {
err = write_fp(storage_fp, buffer, buffer_bo_size - buffer_space_remain);
if (err) {
pr_perror("failed to write out to storage");
goto err_cs_submit_ib;
}
}
err_cs_submit_ib:
amdgpu_cs_ctx_free(h_ctx);
if (err)
break;
}
err_ctx:
amdgpu_bo_list_destroy(h_bo_list);
err_bo_list:
free_and_unmap(packets_per_buffer * 28, h_bo_ib, h_va_ib, gpu_addr_ib, ib);
err_ib_gpu_alloc:
err = amdgpu_bo_va_op(h_bo_dst, 0, size, gpu_addr_dst, 0, AMDGPU_VA_OP_UNMAP);
if (err)
pr_perror("failed to GPU unmap the dest BO %lx, size = %lx", gpu_addr_dst, size);
err_dst_bo_map:
err = amdgpu_va_range_free(h_va_dst);
if (err)
pr_perror("dest range free failed");
err_dst_va:
err = amdgpu_bo_free(h_bo_dst);
if (err)
pr_perror("dest bo free failed");
err_dst_bo_prep:
err = amdgpu_bo_va_op(h_bo_src, 0, size, gpu_addr_src, 0, AMDGPU_VA_OP_UNMAP);
if (err)
pr_perror("failed to GPU unmap the src BO %lx, size = %lx", gpu_addr_src, size);
err_src_bo_map:
err = amdgpu_va_range_free(h_va_src);
if (err)
pr_perror("src range free failed");
err_src_va:
err = amdgpu_bo_free(h_bo_src);
if (err)
pr_perror("src bo free failed");
plugin_log_msg("Leaving sdma_copy_bo, err = %d\n", err);
return err;
}
void *dump_bo_contents(void *_thread_data)
{
struct thread_data *thread_data = (struct thread_data *)_thread_data;
struct kfd_criu_bo_bucket *bo_buckets = thread_data->bo_buckets;
struct amdgpu_gpu_info gpu_info = { 0 };
amdgpu_device_handle h_dev;
size_t max_bo_size = 0, image_size = 0, buffer_size;
uint64_t max_copy_size;
uint32_t major, minor;
int num_bos = 0;
int i, ret = 0;
FILE *bo_contents_fp = NULL;
void *buffer;
char img_path[40];
pr_info("Thread[0x%x] started\n", thread_data->gpu_id);
ret = amdgpu_device_initialize(thread_data->drm_fd, &major, &minor, &h_dev);
if (ret) {
pr_perror("failed to initialize device");
goto exit;
}
plugin_log_msg("libdrm initialized successfully\n");
ret = amdgpu_query_gpu_info(h_dev, &gpu_info);
if (ret) {
pr_perror("failed to query gpuinfo via libdrm");
goto exit;
}
max_copy_size = (gpu_info.family_id >= AMDGPU_FAMILY_AI) ? SDMA_LINEAR_COPY_MAX_SIZE :
SDMA_LINEAR_COPY_MAX_SIZE - 1;
for (i = 0; i < thread_data->num_of_bos; i++) {
if (bo_buckets[i].gpu_id == thread_data->gpu_id &&
(bo_buckets[i].alloc_flags & (KFD_IOC_ALLOC_MEM_FLAGS_VRAM | KFD_IOC_ALLOC_MEM_FLAGS_GTT))) {
image_size += bo_buckets[i].size;
if (bo_buckets[i].size > max_bo_size)
max_bo_size = bo_buckets[i].size;
}
}
buffer_size = kfd_max_buffer_size > 0 ? min(kfd_max_buffer_size, max_bo_size) : max_bo_size;
posix_memalign(&buffer, sysconf(_SC_PAGE_SIZE), buffer_size);
if (!buffer) {
pr_perror("Failed to alloc aligned memory. Consider setting KFD_MAX_BUFFER_SIZE.");
ret = -ENOMEM;
goto exit;
}
snprintf(img_path, sizeof(img_path), IMG_PAGES_FILE, thread_data->id, thread_data->gpu_id);
bo_contents_fp = open_img_file(img_path, true, &image_size);
if (!bo_contents_fp) {
pr_perror("Cannot fopen %s", img_path);
ret = -EIO;
goto exit;
}
for (i = 0; i < thread_data->num_of_bos; i++) {
if (bo_buckets[i].gpu_id != thread_data->gpu_id)
continue;
if (!(bo_buckets[i].alloc_flags & (KFD_IOC_ALLOC_MEM_FLAGS_VRAM | KFD_IOC_ALLOC_MEM_FLAGS_GTT)))
continue;
num_bos++;
/* perform sDMA based vram copy */
ret = sdma_copy_bo(bo_buckets[i], bo_contents_fp, buffer, buffer_size, h_dev, max_copy_size,
SDMA_OP_VRAM_READ);
if (ret) {
pr_err("Failed to drain the BO using sDMA: bo_buckets[%d]\n", i);
break;
}
}
exit:
pr_info("Thread[0x%x] done num_bos:%d ret:%d\n", thread_data->gpu_id, num_bos, ret);
if (bo_contents_fp)
fclose(bo_contents_fp);
xfree(buffer);
amdgpu_device_deinitialize(h_dev);
thread_data->ret = ret;
return NULL;
};
void *restore_bo_contents(void *_thread_data)
{
struct thread_data *thread_data = (struct thread_data *)_thread_data;
struct kfd_criu_bo_bucket *bo_buckets = thread_data->bo_buckets;
size_t image_size = 0, total_bo_size = 0, max_bo_size = 0, buffer_size;