summaryrefslogtreecommitdiffstats
path: root/drivers/misc/habanalabs/command_buffer.c
blob: e495f44064fa8d89462d614287f5ddc4f1857190 (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
// SPDX-License-Identifier: GPL-2.0

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

#include <uapi/misc/habanalabs.h>
#include "habanalabs.h"

#include <linux/mm.h>
#include <linux/slab.h>

static void cb_fini(struct hl_device *hdev, struct hl_cb *cb)
{
	hdev->asic_funcs->asic_dma_free_coherent(hdev, cb->size,
			(void *) (uintptr_t) cb->kernel_address,
			cb->bus_address);
	kfree(cb);
}

static void cb_do_release(struct hl_device *hdev, struct hl_cb *cb)
{
	if (cb->is_pool) {
		spin_lock(&hdev->cb_pool_lock);
		list_add(&cb->pool_list, &hdev->cb_pool);
		spin_unlock(&hdev->cb_pool_lock);
	} else {
		cb_fini(hdev, cb);
	}
}

static void cb_release(struct kref *ref)
{
	struct hl_device *hdev;
	struct hl_cb *cb;

	cb = container_of(ref, struct hl_cb, refcount);
	hdev = cb->hdev;

	hl_debugfs_remove_cb(cb);

	cb_do_release(hdev, cb);
}

static struct hl_cb *hl_cb_alloc(struct hl_device *hdev, u32 cb_size,
					int ctx_id)
{
	struct hl_cb *cb;
	void *p;

	/*
	 * We use of GFP_ATOMIC here because this function can be called from
	 * the latency-sensitive code path for command submission. Due to H/W
	 * limitations in some of the ASICs, the kernel must copy the user CB
	 * that is designated for an external queue and actually enqueue
	 * the kernel's copy. Hence, we must never sleep in this code section
	 * and must use GFP_ATOMIC for all memory allocations.
	 */
	if (ctx_id == HL_KERNEL_ASID_ID)
		cb = kzalloc(sizeof(*cb), GFP_ATOMIC);
	else
		cb = kzalloc(sizeof(*cb), GFP_KERNEL);

	if (!cb)
		return NULL;

	if (ctx_id == HL_KERNEL_ASID_ID)
		p = hdev->asic_funcs->asic_dma_alloc_coherent(hdev, cb_size,
						&cb->bus_address, GFP_ATOMIC);
	else
		p = hdev->asic_funcs->asic_dma_alloc_coherent(hdev, cb_size,
						&cb->bus_address,
						GFP_USER | __GFP_ZERO);
	if (!p) {
		dev_err(hdev->dev,
			"failed to allocate %d of dma memory for CB\n",
			cb_size);
		kfree(cb);
		return NULL;
	}

	cb->kernel_address = (u64) (uintptr_t) p;
	cb->size = cb_size;

	return cb;
}

int hl_cb_create(struct hl_device *hdev, struct hl_cb_mgr *mgr,
			u32 cb_size, u64 *handle, int ctx_id)
{
	struct hl_cb *cb;
	bool alloc_new_cb = true;
	int rc;

	/*
	 * Can't use generic function to check this because of special case
	 * where we create a CB as part of the reset process
	 */
	if ((hdev->disabled) || ((atomic_read(&hdev->in_reset)) &&
					(ctx_id != HL_KERNEL_ASID_ID))) {
		dev_warn_ratelimited(hdev->dev,
			"Device is disabled or in reset. Can't create new CBs\n");
		rc = -EBUSY;
		goto out_err;
	}

	if (cb_size > HL_MAX_CB_SIZE) {
		dev_err(hdev->dev,
			"CB size %d must be less then %d\n",
			cb_size, HL_MAX_CB_SIZE);
		rc = -EINVAL;
		goto out_err;
	}

	/* Minimum allocation must be PAGE SIZE */
	if (cb_size < PAGE_SIZE)
		cb_size = PAGE_SIZE;

	if (ctx_id == HL_KERNEL_ASID_ID &&
			cb_size <= hdev->asic_prop.cb_pool_cb_size) {

		spin_lock(&hdev->cb_pool_lock);
		if (!list_empty(&hdev->cb_pool)) {
			cb = list_first_entry(&hdev->cb_pool, typeof(*cb),
					pool_list);
			list_del(&cb->pool_list);
			spin_unlock(&hdev->cb_pool_lock);
			alloc_new_cb = false;
		} else {
			spin_unlock(&hdev->cb_pool_lock);
			dev_dbg(hdev->dev, "CB pool is empty\n");
		}
	}

	if (alloc_new_cb) {
		cb = hl_cb_alloc(hdev, cb_size, ctx_id);
		if (!cb) {
			rc = -ENOMEM;
			goto out_err;
		}
	}

	cb->hdev = hdev;
	cb->ctx_id = ctx_id;

	spin_lock(&mgr->cb_lock);
	rc = idr_alloc(&mgr->cb_handles, cb, 1, 0, GFP_ATOMIC);
	spin_unlock(&mgr->cb_lock);

	if (rc < 0) {
		dev_err(hdev->dev, "Failed to allocate IDR for a new CB\n");
		goto release_cb;
	}

	cb->id = rc;

	kref_init(&cb->refcount);
	spin_lock_init(&cb->lock);

	/*
	 * idr is 32-bit so we can safely OR it with a mask that is above
	 * 32 bit
	 */
	*handle = cb->id | HL_MMAP_CB_MASK;
	*handle <<= PAGE_SHIFT;

	hl_debugfs_add_cb(cb);

	return 0;

release_cb:
	cb_do_release(hdev, cb);
out_err:
	*handle = 0;

	return rc;
}

int hl_cb_destroy(struct hl_device *hdev, struct hl_cb_mgr *mgr, u64 cb_handle)
{
	struct hl_cb *cb;
	u32 handle;
	int rc = 0;

	/*
	 * handle was given to user to do mmap, I need to shift it back to
	 * how the idr module gave it to me
	 */
	cb_handle >>= PAGE_SHIFT;
	handle = (u32) cb_handle;

	spin_lock(&mgr->cb_lock);

	cb = idr_find(&mgr->cb_handles, handle);
	if (cb) {
		idr_remove(&mgr->cb_handles, handle);
		spin_unlock(&mgr->cb_lock);
		kref_put(&cb->refcount, cb_release);
	} else {
		spin_unlock(&mgr->cb_lock);
		dev_err(hdev->dev,
			"CB destroy failed, no match to handle 0x%x\n", handle);
		rc = -EINVAL;
	}

	return rc;
}

int hl_cb_ioctl(struct hl_fpriv *hpriv, void *data)
{
	union hl_cb_args *args = data;
	struct hl_device *hdev = hpriv->hdev;
	u64 handle;
	int rc;

	if (hl_device_disabled_or_in_reset(hdev)) {
		dev_warn_ratelimited(hdev->dev,
			"Device is %s. Can't execute CB IOCTL\n",
			atomic_read(&hdev->in_reset) ? "in_reset" : "disabled");
		return -EBUSY;
	}

	switch (args->in.op) {
	case HL_CB_OP_CREATE:
		rc = hl_cb_create(hdev, &hpriv->cb_mgr, args->in.cb_size,
					&handle, hpriv->ctx->asid);
		memset(args, 0, sizeof(*args));
		args->out.cb_handle = handle;
		break;
	case HL_CB_OP_DESTROY:
		rc = hl_cb_destroy(hdev, &hpriv->cb_mgr,
					args->in.cb_handle);
		break;
	default:
		rc = -ENOTTY;
		break;
	}

	return rc;
}

static void cb_vm_close(struct vm_area_struct *vma)
{
	struct hl_cb *cb = (struct hl_cb *) vma->vm_private_data;
	long new_mmap_size;

	new_mmap_size = cb->mmap_size - (vma->vm_end - vma->vm_start);

	if (new_mmap_size > 0) {
		cb->mmap_size = new_mmap_size;
		return;
	}

	spin_lock(&cb->lock);
	cb->mmap = false;
	spin_unlock(&cb->lock);

	hl_cb_put(cb);
	vma->vm_private_data = NULL;
}

static const struct vm_operations_struct cb_vm_ops = {
	.close = cb_vm_close
};

int hl_cb_mmap(struct hl_fpriv *hpriv, struct vm_area_struct *vma)
{
	struct hl_device *hdev = hpriv->hdev;
	struct hl_cb *cb;
	phys_addr_t address;
	u32 handle;
	int rc;

	handle = vma->vm_pgoff;

	/* reference was taken here */
	cb = hl_cb_get(hdev, &hpriv->cb_mgr, handle);
	if (!cb) {
		dev_err(hdev->dev,
			"CB mmap failed, no match to handle %d\n", handle);
		return -EINVAL;
	}

	/* Validation check */
	if ((vma->vm_end - vma->vm_start) != ALIGN(cb->size, PAGE_SIZE)) {
		dev_err(hdev->dev,
			"CB mmap failed, mmap size 0x%lx != 0x%x cb size\n",
			vma->vm_end - vma->vm_start, cb->size);
		rc = -EINVAL;
		goto put_cb;
	}

	spin_lock(&cb->lock);

	if (cb->mmap) {
		dev_err(hdev->dev,
			"CB mmap failed, CB already mmaped to user\n");
		rc = -EINVAL;
		goto release_lock;
	}

	cb->mmap = true;

	spin_unlock(&cb->lock);

	vma->vm_ops = &cb_vm_ops;

	/*
	 * Note: We're transferring the cb reference to
	 * vma->vm_private_data here.
	 */

	vma->vm_private_data = cb;

	/* Calculate address for CB */
	address = virt_to_phys((void *) (uintptr_t) cb->kernel_address);

	rc = hdev->asic_funcs->cb_mmap(hdev, vma, cb->kernel_address,
					address, cb->size);

	if (rc) {
		spin_lock(&cb->lock);
		cb->mmap = false;
		goto release_lock;
	}

	cb->mmap_size = cb->size;

	return 0;

release_lock:
	spin_unlock(&cb->lock);
put_cb:
	hl_cb_put(cb);
	return rc;
}

struct hl_cb *hl_cb_get(struct hl_device *hdev, struct hl_cb_mgr *mgr,
			u32 handle)
{
	struct hl_cb *cb;

	spin_lock(&mgr->cb_lock);
	cb = idr_find(&mgr->cb_handles, handle);

	if (!cb) {
		spin_unlock(&mgr->cb_lock);
		dev_warn(hdev->dev,
			"CB get failed, no match to handle %d\n", handle);
		return NULL;
	}

	kref_get(&cb->refcount);

	spin_unlock(&mgr->cb_lock);

	return cb;

}

void hl_cb_put(struct hl_cb *cb)
{
	kref_put(&cb->refcount, cb_release);
}

void hl_cb_mgr_init(struct hl_cb_mgr *mgr)
{
	spin_lock_init(&mgr->cb_lock);
	idr_init(&mgr->cb_handles);
}

void hl_cb_mgr_fini(struct hl_device *hdev, struct hl_cb_mgr *mgr)
{
	struct hl_cb *cb;
	struct idr *idp;
	u32 id;

	idp = &mgr->cb_handles;

	idr_for_each_entry(idp, cb, id) {
		if (kref_put(&cb->refcount, cb_release) != 1)
			dev_err(hdev->dev,
				"CB %d for CTX ID %d is still alive\n",
				id, cb->ctx_id);
	}

	idr_destroy(&mgr->cb_handles);
}

struct hl_cb *hl_cb_kernel_create(struct hl_device *hdev, u32 cb_size)
{
	u64 cb_handle;
	struct hl_cb *cb;
	int rc;

	rc = hl_cb_create(hdev, &hdev->kernel_cb_mgr, cb_size, &cb_handle,
			HL_KERNEL_ASID_ID);
	if (rc) {
		dev_err(hdev->dev, "Failed to allocate CB for KMD %d\n", rc);
		return NULL;
	}

	cb_handle >>= PAGE_SHIFT;
	cb = hl_cb_get(hdev, &hdev->kernel_cb_mgr, (u32) cb_handle);
	/* hl_cb_get should never fail here so use kernel WARN */
	WARN(!cb, "Kernel CB handle invalid 0x%x\n", (u32) cb_handle);
	if (!cb)
		goto destroy_cb;

	return cb;

destroy_cb:
	hl_cb_destroy(hdev, &hdev->kernel_cb_mgr, cb_handle << PAGE_SHIFT);

	return NULL;
}

int hl_cb_pool_init(struct hl_device *hdev)
{
	struct hl_cb *cb;
	int i;

	INIT_LIST_HEAD(&hdev->cb_pool);
	spin_lock_init(&hdev->cb_pool_lock);

	for (i = 0 ; i < hdev->asic_prop.cb_pool_cb_cnt ; i++) {
		cb = hl_cb_alloc(hdev, hdev->asic_prop.cb_pool_cb_size,
				HL_KERNEL_ASID_ID);
		if (cb) {
			cb->is_pool = true;
			list_add(&cb->pool_list, &hdev->cb_pool);
		} else {
			hl_cb_pool_fini(hdev);
			return -ENOMEM;
		}
	}

	return 0;
}

int hl_cb_pool_fini(struct hl_device *hdev)
{
	struct hl_cb *cb, *tmp;

	list_for_each_entry_safe(cb, tmp, &hdev->cb_pool, pool_list) {
		list_del(&cb->pool_list);
		cb_fini(hdev, cb);
	}

	return 0;
}