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
|
// SPDX-License-Identifier: GPL-2.0
/*
* Copyright 2016-2019 HabanaLabs, Ltd.
* All Rights Reserved.
*/
#include "habanalabs.h"
#include <linux/slab.h>
/*
* hl_queue_add_ptr - add to pi or ci and checks if it wraps around
*
* @ptr: the current pi/ci value
* @val: the amount to add
*
* Add val to ptr. It can go until twice the queue length.
*/
inline u32 hl_hw_queue_add_ptr(u32 ptr, u16 val)
{
ptr += val;
ptr &= ((HL_QUEUE_LENGTH << 1) - 1);
return ptr;
}
static inline int queue_free_slots(struct hl_hw_queue *q, u32 queue_len)
{
int delta = (q->pi - q->ci);
if (delta >= 0)
return (queue_len - delta);
else
return (abs(delta) - queue_len);
}
void hl_int_hw_queue_update_ci(struct hl_cs *cs)
{
struct hl_device *hdev = cs->ctx->hdev;
struct hl_hw_queue *q;
int i;
hdev->asic_funcs->hw_queues_lock(hdev);
if (hdev->disabled)
goto out;
q = &hdev->kernel_queues[0];
for (i = 0 ; i < HL_MAX_QUEUES ; i++, q++) {
if (q->queue_type == QUEUE_TYPE_INT) {
q->ci += cs->jobs_in_queue_cnt[i];
q->ci &= ((q->int_queue_len << 1) - 1);
}
}
out:
hdev->asic_funcs->hw_queues_unlock(hdev);
}
/*
* ext_queue_submit_bd - Submit a buffer descriptor to an external queue
*
* @hdev: pointer to habanalabs device structure
* @q: pointer to habanalabs queue structure
* @ctl: BD's control word
* @len: BD's length
* @ptr: BD's pointer
*
* This function assumes there is enough space on the queue to submit a new
* BD to it. It initializes the next BD and calls the device specific
* function to set the pi (and doorbell)
*
* This function must be called when the scheduler mutex is taken
*
*/
static void ext_queue_submit_bd(struct hl_device *hdev, struct hl_hw_queue *q,
u32 ctl, u32 len, u64 ptr)
{
struct hl_bd *bd;
bd = (struct hl_bd *) (uintptr_t) q->kernel_address;
bd += hl_pi_2_offset(q->pi);
bd->ctl = cpu_to_le32(ctl);
bd->len = cpu_to_le32(len);
bd->ptr = cpu_to_le64(ptr);
q->pi = hl_queue_inc_ptr(q->pi);
hdev->asic_funcs->ring_doorbell(hdev, q->hw_queue_id, q->pi);
}
/*
* ext_queue_sanity_checks - perform some sanity checks on external queue
*
* @hdev : pointer to hl_device structure
* @q : pointer to hl_hw_queue structure
* @num_of_entries : how many entries to check for space
* @reserve_cq_entry : whether to reserve an entry in the cq
*
* H/W queues spinlock should be taken before calling this function
*
* Perform the following:
* - Make sure we have enough space in the h/w queue
* - Make sure we have enough space in the completion queue
* - Reserve space in the completion queue (needs to be reversed if there
* is a failure down the road before the actual submission of work). Only
* do this action if reserve_cq_entry is true
*
*/
static int ext_queue_sanity_checks(struct hl_device *hdev,
struct hl_hw_queue *q, int num_of_entries,
bool reserve_cq_entry)
{
atomic_t *free_slots =
&hdev->completion_queue[q->hw_queue_id].free_slots_cnt;
int free_slots_cnt;
/* Check we have enough space in the queue */
free_slots_cnt = queue_free_slots(q, HL_QUEUE_LENGTH);
if (free_slots_cnt < num_of_entries) {
dev_dbg(hdev->dev, "Queue %d doesn't have room for %d CBs\n",
q->hw_queue_id, num_of_entries);
return -EAGAIN;
}
if (reserve_cq_entry) {
/*
* Check we have enough space in the completion queue
* Add -1 to counter (decrement) unless counter was already 0
* In that case, CQ is full so we can't submit a new CB because
* we won't get ack on its completion
* atomic_add_unless will return 0 if counter was already 0
*/
if (atomic_add_negative(num_of_entries * -1, free_slots)) {
dev_dbg(hdev->dev, "No space for %d on CQ %d\n",
num_of_entries, q->hw_queue_id);
atomic_add(num_of_entries, free_slots);
return -EAGAIN;
}
}
return 0;
}
/*
* int_queue_sanity_checks - perform some sanity checks on internal queue
*
* @hdev : pointer to hl_device structure
* @q : pointer to hl_hw_queue structure
* @num_of_entries : how many entries to check for space
*
* H/W queues spinlock should be taken before calling this function
*
* Perform the following:
* - Make sure we have enough space in the h/w queue
*
*/
static int int_queue_sanity_checks(struct hl_device *hdev,
struct hl_hw_queue *q,
int num_of_entries)
{
int free_slots_cnt;
/* Check we have enough space in the queue */
free_slots_cnt = queue_free_slots(q, q->int_queue_len);
if (free_slots_cnt < num_of_entries) {
dev_dbg(hdev->dev, "Queue %d doesn't have room for %d CBs\n",
q->hw_queue_id, num_of_entries);
return -EAGAIN;
}
return 0;
}
/*
* hl_hw_queue_send_cb_no_cmpl - send a single CB (not a JOB) without completion
*
* @hdev: pointer to hl_device structure
* @hw_queue_id: Queue's type
* @cb_size: size of CB
* @cb_ptr: pointer to CB location
*
* This function sends a single CB, that must NOT generate a completion entry
*
*/
int hl_hw_queue_send_cb_no_cmpl(struct hl_device *hdev, u32 hw_queue_id,
u32 cb_size, u64 cb_ptr)
{
struct hl_hw_queue *q = &hdev->kernel_queues[hw_queue_id];
int rc;
/*
* The CPU queue is a synchronous queue with an effective depth of
* a single entry (although it is allocated with room for multiple
* entries). Therefore, there is a different lock, called
* send_cpu_message_lock, that serializes accesses to the CPU queue.
* As a result, we don't need to lock the access to the entire H/W
* queues module when submitting a JOB to the CPU queue
*/
if (q->queue_type != QUEUE_TYPE_CPU)
hdev->asic_funcs->hw_queues_lock(hdev);
if (hdev->disabled) {
rc = -EPERM;
goto out;
}
rc = ext_queue_sanity_checks(hdev, q, 1, false);
if (rc)
goto out;
ext_queue_submit_bd(hdev, q, 0, cb_size, cb_ptr);
out:
if (q->queue_type != QUEUE_TYPE_CPU)
hdev->asic_funcs->hw_queues_unlock(hdev);
return rc;
}
/*
* ext_hw_queue_schedule_job - submit an JOB to an external queue
*
* @job: pointer to the job that needs to be submitted to the queue
*
* This function must be called when the scheduler mutex is taken
*
*/
static void ext_hw_queue_schedule_job(struct hl_cs_job *job)
{
struct hl_device *hdev = job->cs->ctx->hdev;
struct hl_hw_queue *q = &hdev->kernel_queues[job->hw_queue_id];
struct hl_cq_entry cq_pkt;
struct hl_cq *cq;
u64 cq_addr;
struct hl_cb *cb;
u32 ctl;
u32 len;
u64 ptr;
/*
* Update the JOB ID inside the BD CTL so the device would know what
* to write in the completion queue
*/
ctl = ((q->pi << BD_CTL_SHADOW_INDEX_SHIFT) & BD_CTL_SHADOW_INDEX_MASK);
cb = job->patched_cb;
len = job->job_cb_size;
ptr = cb->bus_address;
cq_pkt.data = cpu_to_le32(
((q->pi << CQ_ENTRY_SHADOW_INDEX_SHIFT)
& CQ_ENTRY_SHADOW_INDEX_MASK) |
(1 << CQ_ENTRY_SHADOW_INDEX_VALID_SHIFT) |
(1 << CQ_ENTRY_READY_SHIFT));
/*
* No need to protect pi_offset because scheduling to the
* H/W queues is done under the scheduler mutex
|