summaryrefslogtreecommitdiffstats
path: root/drivers/misc/habanalabs/common/irq.c
blob: 8500e15ef743f0b196f9d9e9a35c1d29a9f052ef (plain)
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
// SPDX-License-Identifier: GPL-2.0

/*
 * Copyright 2016-2019 HabanaLabs, Ltd.
 * All Rights Reserved.
 */

#include "habanalabs.h"

#include <linux/slab.h>

/**
 * struct hl_eqe_work - This structure is used to schedule work of EQ
 *                      entry and cpucp_reset event
 *
 * @eq_work:          workqueue object to run when EQ entry is received
 * @hdev:             pointer to device structure
 * @eq_entry:         copy of the EQ entry
 */
struct hl_eqe_work {
	struct work_struct	eq_work;
	struct hl_device	*hdev;
	struct hl_eq_entry	eq_entry;
};

/**
 * hl_cq_inc_ptr - increment ci or pi of cq
 *
 * @ptr: the current ci or pi value of the completion queue
 *
 * Increment ptr by 1. If it reaches the number of completion queue
 * entries, set it to 0
 */
inline u32 hl_cq_inc_ptr(u32 ptr)
{
	ptr++;
	if (unlikely(ptr == HL_CQ_LENGTH))
		ptr = 0;
	return ptr;
}

/**
 * hl_eq_inc_ptr - increment ci of eq
 *
 * @ptr: the current ci value of the event queue
 *
 * Increment ptr by 1. If it reaches the number of event queue
 * entries, set it to 0
 */
static inline u32 hl_eq_inc_ptr(u32 ptr)
{
	ptr++;
	if (unlikely(ptr == HL_EQ_LENGTH))
		ptr = 0;
	return ptr;
}

static void irq_handle_eqe(struct work_struct *work)
{
	struct hl_eqe_work *eqe_work = container_of(work, struct hl_eqe_work,
							eq_work);
	struct hl_device *hdev = eqe_work->hdev;

	hdev->asic_funcs->handle_eqe(hdev, &eqe_work->eq_entry);

	kfree(eqe_work);
}

/**
 * hl_irq_handler_cq - irq handler for completion queue
 *
 * @irq: irq number
 * @arg: pointer to completion queue structure
 *
 */
irqreturn_t hl_irq_handler_cq(int irq, void *arg)
{
	struct hl_cq *cq = arg;
	struct hl_device *hdev = cq->hdev;
	struct hl_hw_queue *queue;
	struct hl_cs_job *job;
	bool shadow_index_valid;
	u16 shadow_index;
	struct hl_cq_entry *cq_entry, *cq_base;

	if (hdev->disabled) {
		dev_dbg(hdev->dev,
			"Device disabled but received IRQ %d for CQ %d\n",
			irq, cq->hw_queue_id);
		return IRQ_HANDLED;
	}

	cq_base = cq->kernel_address;

	while (1) {
		bool entry_ready = ((le32_to_cpu(cq_base[cq->ci].data) &
					CQ_ENTRY_READY_MASK)
						>> CQ_ENTRY_READY_SHIFT);

		if (!entry_ready)
			break;

		cq_entry = (struct hl_cq_entry *) &cq_base[cq->ci];

		/* Make sure we read CQ entry contents after we've
		 * checked the ownership bit.
		 */
		dma_rmb();

		shadow_index_valid = ((le32_to_cpu(cq_entry->data) &
					CQ_ENTRY_SHADOW_INDEX_VALID_MASK)
					>> CQ_ENTRY_SHADOW_INDEX_VALID_SHIFT);

		shadow_index = (u16) ((le32_to_cpu(cq_entry->data) &
					CQ_ENTRY_SHADOW_INDEX_MASK)
					>> CQ_ENTRY_SHADOW_INDEX_SHIFT);

		queue = &hdev->kernel_queues[cq->hw_queue_id];

		if ((shadow_index_valid) && (!hdev->disabled)) {
			job = queue->shadow_queue[hl_pi_2_offset(shadow_index)];
			queue_work(hdev->cq_wq[cq->cq_idx], &job->finish_work);
		}

		atomic_inc(&queue->ci);

		/* Clear CQ entry ready bit */
		cq_entry->data = cpu_to_le32(le32_to_cpu(cq_entry->data) &
						~CQ_ENTRY_READY_MASK);

		cq->ci = hl_cq_inc_ptr(cq->ci);

		/* Increment free slots */
		atomic_inc(&cq->free_slots_cnt);
	}

	return IRQ_HANDLED;
}

/*
 * hl_ts_free_objects - handler of the free objects workqueue.
 * This function should put refcount to objects that the registration node
 * took refcount to them.
 * @work: workqueue object pointer
 */
static void hl_ts_free_objects(struct work_struct *work)
{
	struct timestamp_reg_work_obj *job =
			container_of(work, struct timestamp_reg_work_obj, free_obj);
	struct timestamp_reg_free_node *free_obj, *temp_free_obj;
	struct list_head *free_list_head = job->free_obj_head;
	struct hl_device *hdev = job->hdev;

	list_for_each_entry_safe(free_obj, temp_free_obj, free_list_head, free_objects_node) {
		dev_dbg(hdev->dev, "About to put refcount to buf (%p) cq_cb(%p)\n",
					free_obj->buf,
					free_obj->cq_cb);

		hl_mmap_mem_buf_put(free_obj->buf);
		hl_cb_put(free_obj->cq_cb);
		kfree(free_obj);
	}

	kfree(free_list_head);
	kfree(job);
}

/*
 * This function called with spin_lock of wait_list_lock taken
 * This function will set timestamp and delete the registration node from the
 * wait_list_lock.
 * and since we're protected with spin_lock here, so we cannot just put the refcount
 * for the objects here, since the release function may be called and it's also a long
 * logic (which might sleep also) that cannot be handled in irq context.
 * so here we'll be filling a list with nodes of "put" jobs and then will send this
 * list to a dedicated workqueue to do the actual put.
 */
static int handle_registration_node(struct hl_device *hdev, struct hl_user_pending_interrupt *pend,
						struct list_head **free_list)
{
	struct timestamp_reg_free_node *free_node;
	u64 timestamp;

	if (!(*free_list)) {
		/* Alloc/Init the timestamp registration free objects list */
		*free_list = kmalloc(sizeof(struct list_head), GFP_ATOMIC);
		if (!(*free_list))
			return -ENOMEM;

		INIT_LIST_HEAD(*free_list);
	}

	free_node = kmalloc(sizeof(*free_node), GFP_ATOMIC);
	if (!free_node)
		return -ENOMEM;

	timestamp = ktime_get_ns();

	*pend->ts_reg_info.timestamp_kernel_addr = timestamp;

	dev_dbg(hdev->dev, "Timestamp is set to ts cb address (%p), ts: 0x%llx\n",
			pend->ts_reg_info.timestamp_kernel_addr,
			*(u64 *)pend->ts_reg_info.timestamp_kernel_addr);

	list_del(&pend->wait_list_node);

	/* Mark kernel CB node as free */
	pend->ts_reg_info.in_use = 0;

	/* Putting the refcount for ts_buff and cq_cb objects will be handled
	 * in workqueue context, just add job to free_list.
	 */
	free_node->buf = pend->ts_reg_info.buf;
	free_node->cq_cb = pend->ts_reg_info.cq_cb;
	list_add(&free_node->free_objects_node, *free_list);

	return 0;
}

static void handle_user_cq(struct hl_device *hdev,
			struct hl_user_interrupt *user_cq)
{
	struct hl_user_pending_interrupt *pend, *temp_pend;
	struct list_head *ts_reg_free_list_head = NULL;
	struct timestamp_reg_work_obj *job;
	bool reg_node_handle_fail = false;
	ktime_t now = ktime_get();
	int rc;

	/* For registration nodes:
	 * As part of handling the registration nodes, we should put refcount to
	 * some objects. the problem is that we cannot do that under spinlock
	 * or in irq handler context at all (since release functions are long and
	 * might sleep), so we will need to handle that part in workqueue context.
	 * To avoid handling kmalloc failure which compels us rolling back actions
	 * and move nodes hanged on the free list back to the interrupt wait list
	 * we always alloc the job of the WQ at the beginning.
	 */
	job = kmalloc(sizeof(*job), GFP_ATOMIC);
	if (!job)
		return;

	spin_lock(&user_cq->wait_list_lock);
	list_for_each_entry_safe(pend, temp_pend, &user_cq->wait_list_head, wait_list_node) {
		if ((pend->cq_kernel_addr && *(pend->cq_kernel_addr) >= pend->cq_target_value) ||
				!pend->cq_kernel_addr) {
			if (pend->ts_reg_info.buf) {
				if (!reg_node_handle_fail) {
					rc = handle_registration_node(hdev, pend,
									&ts_reg_free_list_head);
					if (rc)
						reg_node_handle_fail = true;
				}
			} else {
				/* Handle wait target value node */
				pend->fence.timestamp = now;
				complete_all(&pend->fence.completion);
			}
		}
	}
	spin_unlock(&user_cq->wait_list_lock);

	if (ts_reg_free_list_head) {
		INIT_WORK(&job->free_obj, hl_ts_free_objects);
		job->free_obj_head = ts_reg_free_list_head;
		job->hdev = hdev;
		queue_work(hdev->ts_free_obj_wq, &job->free_obj);
	} else {
		kfree(job);
	}
}

/**
 * hl_irq_handler_user_cq - irq handler for user completion queues
 *
 * @irq: irq number
 * @arg: pointer to user interrupt structure
 *
 */
irqreturn_t hl_irq_handler_user_cq(int irq, void *arg)
{
	struct hl_user_interrupt *user_cq = arg;
	struct hl_device *hdev = user_cq->hdev;

	/* Handle user cq interrupts registered on all interrupts */
	handle_user_cq(hdev, &hdev->common_user_interrupt);

	/* Handle user cq interrupts registered on this specific interrupt */
	handle_user_cq(hdev, user_cq);

	return IRQ_HANDLED;
}

/**
 * hl_irq_handler_default - default irq handler
 *
 * @irq: irq number
 * @arg: pointer to user interrupt structure
 *
 */
irqreturn_t hl_irq_handler_default(int irq, void *arg)
{
	struct hl_user_interrupt *user_interrupt = arg;
	struct hl_device *hdev = user_interrupt->hdev;
	u32 interrupt_id = user_interrupt->interrupt_id;

	dev_err(hdev->dev,
		"got invalid user interrupt %u",
		interrupt_id);

	return IRQ_HANDLED;
}

/**
 * hl_irq_handler_eq - irq handler for event queue
 *
 * @irq: irq number
 * @arg: pointer to event queue structure
 *
 */
irqreturn_t hl_irq_handler_eq(int irq, void *arg)
{
	struct hl_eq *eq = arg;
	struct hl_device *hdev = eq->hdev;
	struct hl_eq_entry *eq_entry;
	struct hl_eq_entry *eq_base;
	struct hl_eqe_work *handle_eqe_work;
	bool entry_ready;
	u32 cur_eqe;
	u16 cur_eqe_index;

	eq_base = eq->kernel_address;

	while (1) {
		cur_eqe = le32_to_cpu(eq_base[eq->ci].hdr.ctl);
		entry_ready = !!FIELD_GET(EQ_CTL_READY_MASK, cur_eqe);

		if (!entry_ready)
			break;

		cur_eqe_index = FIELD_GET(EQ_CTL_INDEX_MASK, cur_eqe);
		if ((hdev->event_queue.check_eqe_index) &&
				(((eq->prev_eqe_index + 1) & EQ_CTL_INDEX_MASK)
							!= cur_eqe_index)) {
			dev_dbg(hdev->dev,
				"EQE 0x%x in queue is ready but index does not match %d!=%d",
				eq_base[eq->ci].hdr.ctl,
				((eq->prev_eqe_index + 1) & EQ_CTL_INDEX_MASK),
				cur_eqe_index);
			break;
		}

		eq->prev_eqe_index++;

		eq_entry = &eq_base[eq->ci];

		/*
		 * Make sure we read EQ entry contents after we've
		 * checked the ownership bit.
		 */
		dma_rmb();

		if (hdev->disabled && !hdev->reset_info.is_in_soft_reset) {
			dev_warn(hdev->dev, "Device disabled but received an EQ event\n");
			goto skip_irq;
		}

		handle_eqe_work = kmalloc(sizeof(*handle_eqe_work), GFP_ATOMIC);
		if (handle_eqe_work) {
			INIT_WORK(&handle_eqe_work->eq_work, irq_handle_eqe);
			handle_eqe_work->hdev = hdev;

			memcpy(&handle_eqe_work->eq_entry, eq_entry,
					sizeof(*eq_entry));

			queue_work(hdev->eq_wq, &handle_eqe_work->eq_work);
		}
skip_irq:
		/* Clear EQ entry ready bit */
		eq_entry->hdr.ctl =
			cpu_to_le32(le32_to_cpu(eq_entry->hdr.ctl) &
							~EQ_CTL_READY_MASK);

		eq->ci = hl_eq_inc_ptr(eq->ci);

		hdev->asic_funcs->update_eq_ci(hdev, eq->ci);
	}

	return IRQ_HANDLED;
}

/**
 * hl_cq_init - main initialization function for an cq object
 *
 * @hdev: pointer to device structure
 * @q: pointer to cq structure
 * @hw_queue_id: The H/W queue ID this completion queue belongs to
 *
 * Allocate dma-able memory for the completion queue and initialize fields
 * Returns 0 on success
 */
int hl_cq_init(struct hl_device *hdev, struct hl_cq *q, u32 hw_queue_id)
{
	void *p;

	p = hdev->asic_funcs->asic_dma_alloc_coherent(hdev, HL_CQ_SIZE_IN_BYTES,
				&q->bus_address, GFP_KERNEL | __GFP_ZERO);
	if (!p)
		return -ENOMEM;

	q->hdev = hdev;
	q->kernel_address = p;
	q->hw_queue_id = hw_queue_id;
	q->ci = 0;
	q->pi = 0;

	atomic_set(&q->free_slots_cnt, HL_CQ_LENGTH);

	return 0;
}

/**
 * hl_cq_fini - destroy completion queue
 *
 * @hdev: pointer to device structure
 * @q: pointer to cq structure
 *
 * Free the completion queue memory
 */
void hl_cq_fini(struct hl_device *hdev, struct hl_cq *q)
{
	hdev->asic_funcs->asic_dma_free_coherent(hdev, HL_CQ_SIZE_IN_BYTES,
						 q->kernel_address,
						 q->bus_address);
}

void hl_cq_reset(struct hl_device *hdev, struct hl_cq *q)
{
	q->ci = 0;
	q->pi = 0;

	atomic_set(&q->free_slots_cnt, HL_CQ_LENGTH);

	/*
	 * It's not enough to just reset the PI/CI because the H/W may have
	 * written valid completion entries before it was halted and therefore
	 * we need to clean the actual queues so we won't process old entries
	 * when the device is operational again
	 */

	memset(q->kernel_address, 0, HL_CQ_SIZE_IN_BYTES);
}

/**
 * hl_eq_init - main initialization function for an event queue object
 *
 * @hdev: pointer to device structure
 * @q: pointer to eq structure
 *
 * Allocate dma-able memory for the event queue and initialize fields
 * Returns 0 on success
 */
int hl_eq_init(struct hl_device *hdev, struct hl_eq *q)
{
	void *p;

	p = hdev->asic_funcs->cpu_accessible_dma_pool_alloc(hdev,
							HL_EQ_SIZE_IN_BYTES,
							&q->bus_address);
	if (!p)
		return -ENOMEM;

	q->hdev = hdev;
	q->kernel_address = p;
	q->ci = 0;
	q->prev_eqe_index = 0;

	return 0;
}

/**
 * hl_eq_fini - destroy event queue
 *
 * @hdev: pointer to device structure
 * @q: pointer to eq structure
 *
 * Free the event queue memory
 */
void hl_eq_fini(struct hl_device *hdev, struct hl_eq *q)
{
	flush_workqueue(hdev->eq_wq);

	hdev->asic_funcs->cpu_accessible_dma_pool_free(hdev,
					HL_EQ_SIZE_IN_BYTES,
					q->kernel_address);
}

void hl_eq_reset(struct hl_device *hdev, struct hl_eq *q)
{
	q->ci = 0;
	q->prev_eqe_index = 0;

	/*
	 * It's not enough to just reset the PI/CI because the H/W may have
	 * written valid completion entries before it was halted and therefore
	 * we need to clean the actual queues so we won't process old entries
	 * when the device is operational again
	 */

	memset(q->kernel_address, 0, HL_EQ_SIZE_IN_BYTES);
}