summaryrefslogtreecommitdiffstats
path: root/drivers/misc/habanalabs/common/mmu.c
blob: 28a4638741d8881d4589a3023297e6e0718ece2f (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
512
513
514
// SPDX-License-Identifier: GPL-2.0

/*
 * Copyright 2016-2020 HabanaLabs, Ltd.
 * All Rights Reserved.
 */

#include <linux/slab.h>

#include "habanalabs.h"

bool hl_is_dram_va(struct hl_device *hdev, u64 virt_addr)
{
	struct asic_fixed_properties *prop = &hdev->asic_prop;

	return hl_mem_area_inside_range(virt_addr, prop->dmmu.page_size,
					prop->dmmu.start_addr,
					prop->dmmu.end_addr);
}

/**
 * hl_mmu_init() - initialize the MMU module.
 * @hdev: habanalabs device structure.
 *
 * Return: 0 for success, non-zero for failure.
 */
int hl_mmu_init(struct hl_device *hdev)
{
	int rc = -EOPNOTSUPP;

	if (!hdev->mmu_enable)
		return 0;

	if (hdev->mmu_func[MMU_DR_PGT].init != NULL) {
		rc = hdev->mmu_func[MMU_DR_PGT].init(hdev);
		if (rc)
			return rc;
	}

	if (hdev->mmu_func[MMU_HR_PGT].init != NULL)
		rc = hdev->mmu_func[MMU_HR_PGT].init(hdev);

	return rc;
}

/**
 * hl_mmu_fini() - release the MMU module.
 * @hdev: habanalabs device structure.
 *
 * This function does the following:
 * - Disable MMU in H/W.
 * - Free the pgt_infos pool.
 *
 * All contexts should be freed before calling this function.
 */
void hl_mmu_fini(struct hl_device *hdev)
{
	if (!hdev->mmu_enable)
		return;

	if (hdev->mmu_func[MMU_DR_PGT].fini != NULL)
		hdev->mmu_func[MMU_DR_PGT].fini(hdev);

	if (hdev->mmu_func[MMU_HR_PGT].fini != NULL)
		hdev->mmu_func[MMU_HR_PGT].fini(hdev);
}

/**
 * hl_mmu_ctx_init() - initialize a context for using the MMU module.
 * @ctx: pointer to the context structure to initialize.
 *
 * Initialize a mutex to protect the concurrent mapping flow, a hash to hold all
 * page tables hops related to this context.
 * Return: 0 on success, non-zero otherwise.
 */
int hl_mmu_ctx_init(struct hl_ctx *ctx)
{
	struct hl_device *hdev = ctx->hdev;
	int rc = -EOPNOTSUPP;

	if (!hdev->mmu_enable)
		return 0;

	mutex_init(&ctx->mmu_lock);

	if (hdev->mmu_func[MMU_DR_PGT].ctx_init != NULL) {
		rc = hdev->mmu_func[MMU_DR_PGT].ctx_init(ctx);
		if (rc)
			return rc;
	}

	if (hdev->mmu_func[MMU_HR_PGT].ctx_init != NULL)
		rc = hdev->mmu_func[MMU_HR_PGT].ctx_init(ctx);

	return rc;
}

/*
 * hl_mmu_ctx_fini - disable a ctx from using the mmu module
 *
 * @ctx: pointer to the context structure
 *
 * This function does the following:
 * - Free any pgts which were not freed yet
 * - Free the mutex
 * - Free DRAM default page mapping hops
 */
void hl_mmu_ctx_fini(struct hl_ctx *ctx)
{
	struct hl_device *hdev = ctx->hdev;

	if (!hdev->mmu_enable)
		return;

	if (hdev->mmu_func[MMU_DR_PGT].ctx_fini != NULL)
		hdev->mmu_func[MMU_DR_PGT].ctx_fini(ctx);

	if (hdev->mmu_func[MMU_HR_PGT].ctx_fini != NULL)
		hdev->mmu_func[MMU_HR_PGT].ctx_fini(ctx);

	mutex_destroy(&ctx->mmu_lock);
}

/*
 * hl_mmu_unmap_page - unmaps a virtual addr
 *
 * @ctx: pointer to the context structure
 * @virt_addr: virt addr to map from
 * @page_size: size of the page to unmap
 * @flush_pte: whether to do a PCI flush
 *
 * This function does the following:
 * - Check that the virt addr is mapped
 * - Unmap the virt addr and frees pgts if possible
 * - Returns 0 on success, -EINVAL if the given addr is not mapped
 *
 * Because this function changes the page tables in the device and because it
 * changes the MMU hash, it must be protected by a lock.
 * However, because it maps only a single page, the lock should be implemented
 * in a higher level in order to protect the entire mapping of the memory area
 *
 * For optimization reasons PCI flush may be requested once after unmapping of
 * large area.
 */
int hl_mmu_unmap_page(struct hl_ctx *ctx, u64 virt_addr, u32 page_size,
		bool flush_pte)
{
	struct hl_device *hdev = ctx->hdev;
	struct asic_fixed_properties *prop = &hdev->asic_prop;
	struct hl_mmu_properties *mmu_prop;
	u64 real_virt_addr;
	u32 real_page_size, npages;
	int i, rc = 0, pgt_residency;
	bool is_dram_addr;

	if (!hdev->mmu_enable)
		return 0;

	is_dram_addr = hl_is_dram_va(hdev, virt_addr);

	if (is_dram_addr)
		mmu_prop = &prop->dmmu;
	else if ((page_size % prop->pmmu_huge.page_size) == 0)
		mmu_prop = &prop->pmmu_huge;
	else
		mmu_prop = &prop->pmmu;

	pgt_residency = mmu_prop->host_resident ? MMU_HR_PGT : MMU_DR_PGT;

	/*
	 * The H/W handles mapping of specific page sizes. Hence if the page
	 * size is bigger, we break it to sub-pages and unmap them separately.
	 */
	if ((page_size % mmu_prop->page_size) == 0) {
		real_page_size = mmu_prop->page_size;
	} else {
		dev_err(hdev->dev,
			"page size of %u is not %uKB aligned, can't unmap\n",
			page_size, mmu_prop->page_size >> 10);

		return -EFAULT;
	}

	npages = page_size / real_page_size;
	real_virt_addr = virt_addr;

	for (i = 0 ; i < npages ; i++) {
		rc = hdev->mmu_func[pgt_residency].unmap(ctx,
						real_virt_addr, is_dram_addr);
		if (rc)
			break;

		real_virt_addr += real_page_size;
	}

	if (flush_pte)
		hdev->mmu_func[pgt_residency].flush(ctx);

	return rc;
}

/*
 * hl_mmu_map_page - maps a virtual addr to physical addr
 *
 * @ctx: pointer to the context structure
 * @virt_addr: virt addr to map from
 * @phys_addr: phys addr to map to
 * @page_size: physical page size
 * @flush_pte: whether to do a PCI flush
 *
 * This function does the following:
 * - Check that the virt addr is not mapped
 * - Allocate pgts as necessary in order to map the virt addr to the phys
 * - Returns 0 on success, -EINVAL if addr is already mapped, or -ENOMEM.
 *
 * Because this function changes the page tables in the device and because it
 * changes the MMU hash, it must be protected by a lock.
 * However, because it maps only a single page, the lock should be implemented
 * in a higher level in order to protect the entire mapping of the memory area
 *
 * For optimization reasons PCI flush may be requested once after mapping of
 * large area.
 */
int hl_mmu_map_page(struct hl_ctx *ctx, u64 virt_addr, u64 phys_addr,
		u32 page_size, bool flush_pte)
{
	struct hl_device *hdev = ctx->hdev;
	struct asic_fixed_properties *prop = &hdev->asic_prop;
	struct hl_mmu_properties *mmu_prop;
	u64 real_virt_addr, real_phys_addr;
	u32 real_page_size, npages;
	int i, rc, pgt_residency, mapped_cnt = 0;
	bool is_dram_addr;


	if (!hdev->mmu_enable)
		return 0;

	is_dram_addr = hl_is_dram_va(hdev, virt_addr);

	if (is_dram_addr)
		mmu_prop = &prop->dmmu;
	else if ((page_size % prop->pmmu_huge.page_size) == 0)
		mmu_prop = &prop->pmmu_huge;
	else
		mmu_prop = &prop->pmmu;

	pgt_residency = mmu_prop->host_resident ? MMU_HR_PGT : MMU_DR_PGT;

	/*
	 * The H/W handles mapping of specific page sizes. Hence if the page
	 * size is bigger, we break it to sub-pages and map them separately.
	 */
	if ((page_size % mmu_prop->page_size) == 0) {
		real_page_size = mmu_prop->page_size;
	} else {
		dev_err(hdev->dev,
			"page size of %u is not %uKB aligned, can't map\n",
			page_size, mmu_prop->page_size >> 10);

		return -EFAULT;
	}

	WARN_ONCE((phys_addr & (real_page_size - 1)),
		"Mapping 0x%llx with page size of 0x%x is erroneous! Address must be divisible by page size",
		phys_addr, real_page_size);

	npages = page_size / real_page_size;
	real_virt_addr = virt_addr;
	real_phys_addr = phys_addr;

	for (i = 0 ; i < npages ; i++) {
		rc = hdev->mmu_func[pgt_residency].map(ctx,
						real_virt_addr, real_phys_addr,
						real_page_size, is_dram_addr);
		if (rc)
			goto err;

		real_virt_addr += real_page_size;
		real_phys_addr += real_page_size;
		mapped_cnt++;
	}

	if (flush_pte)
		hdev->mmu_func[pgt_residency].flush(ctx);

	return 0;

err:
	real_virt_addr = virt_addr;
	for (i = 0 ; i < mapped_cnt ; i++) {
		if (hdev->mmu_func[pgt_residency].unmap(ctx,
						real_virt_addr, is_dram_addr))
			dev_warn_ratelimited(hdev->dev,
				"failed to unmap va: 0x%llx\n", real_virt_addr);

		real_virt_addr += real_page_size;
	}

	hdev->mmu_func[pgt_residency].flush(ctx);

	return rc;
}

/*
 * hl_mmu_map_contiguous - implements a wrapper for hl_mmu_map_page
 *                         for mapping contiguous physical memory
 *
 * @ctx: pointer to the context structure
 * @virt_addr: virt addr to map from
 * @phys_addr: phys addr to map to
 * @size: size to map
 *
 */
int hl_mmu_map_contiguous(struct hl_ctx *ctx, u64 virt_addr,
					u64 phys_addr, u32 size)
{
	struct hl_device *hdev = ctx->hdev;
	struct asic_fixed_properties *prop = &hdev->asic_prop;
	u64 curr_va, curr_pa;
	u32 page_size;
	bool flush_pte;
	int rc = 0, off;

	if (hl_mem_area_inside_range(virt_addr, size,
			prop->dmmu.start_addr, prop->dmmu.end_addr))
		page_size = prop->dmmu.page_size;
	else if (hl_mem_area_inside_range(virt_addr, size,
			prop->pmmu.start_addr, prop->pmmu.end_addr))
		page_size = prop->pmmu.page_size;
	else if (hl_mem_area_inside_range(virt_addr, size,
			prop->pmmu_huge.start_addr, prop->pmmu_huge.end_addr))
		page_size = prop->pmmu_huge.page_size;
	else
		return -EINVAL;

	for (off = 0 ; off < size ; off += page_size) {
		curr_va = virt_addr + off;
		curr_pa = phys_addr + off;
		flush_pte = (off + page_size) >= size;
		rc = hl_mmu_map_page(ctx, curr_va, curr_pa, page_size,
								flush_pte);
		if (rc) {
			dev_err(hdev->dev,
				"Map failed for va 0x%llx to pa 0x%llx\n",
				curr_va, curr_pa);
			goto unmap;
		}
	}

	return rc;

unmap:
	for (; off >= 0 ; off -= page_size) {
		curr_va = virt_addr + off;
		flush_pte = (off - (s32) page_size) < 0;
		if (hl_mmu_unmap_page(ctx, curr_va, page_size, flush_pte))
			dev_warn_ratelimited(hdev->dev,
				"failed to unmap va 0x%llx\n", curr_va);
	}

	return rc;
}

/*
 * hl_mmu_unmap_contiguous - implements a wrapper for hl_mmu_unmap_page
 *                           for unmapping contiguous physical memory
 *
 * @ctx: pointer to the context structure
 * @virt_addr: virt addr to unmap
 * @size: size to unmap
 *
 */
int hl_mmu_unmap_contiguous(struct hl_ctx *ctx, u64 virt_addr, u32 size)
{
	struct hl_device *hdev = ctx->hdev;
	struct asic_fixed_properties *prop = &hdev->asic_prop;
	u64 curr_va;
	u32 page_size;
	bool flush_pte;
	int rc = 0, off;

	if (hl_mem_area_inside_range(virt_addr, size,
			prop->dmmu.start_addr, prop->dmmu.end_addr))
		page_size = prop->dmmu.page_size;
	else if (hl_mem_area_inside_range(virt_addr, size,
			prop->pmmu.start_addr, prop->pmmu.end_addr))
		page_size = prop->pmmu.page_size;
	else if (hl_mem_area_inside_range(virt_addr, size,
			prop->pmmu_huge.start_addr, prop->pmmu_huge.end_addr))
		page_size = prop->pmmu_huge.page_size;
	else
		return -EINVAL;

	for (off = 0 ; off < size ; off += page_size) {
		curr_va = virt_addr + off;
		flush_pte = (off + page_size) >= size;
		rc = hl_mmu_unmap_page(ctx, curr_va, page_size, flush_pte);
		if (rc)
			dev_warn_ratelimited(hdev->dev,
				"Unmap failed for va 0x%llx\n", curr_va);
	}

	return rc;
}

/*
 * hl_mmu_swap_out - marks all mapping of the given ctx as swapped out
 *
 * @ctx: pointer to the context structure
 *
 */
void hl_mmu_swap_out(struct hl_ctx *ctx)
{
	struct hl_device *hdev = ctx->hdev;

	if (!hdev->mmu_enable)
		return;

	if (hdev->mmu_func[MMU_DR_PGT].swap_out != NULL)
		hdev->mmu_func[MMU_DR_PGT].swap_out(ctx);

	if (hdev->mmu_func[MMU_HR_PGT].swap_out != NULL)
		hdev->mmu_func[MMU_HR_PGT].swap_out(ctx);
}

/*
 * hl_mmu_swap_in - marks all mapping of the given ctx as swapped in
 *
 * @ctx: pointer to the context structure
 *
 */
void hl_mmu_swap_in(struct hl_ctx *ctx)
{
	struct hl_device *hdev = ctx->hdev;

	if (!hdev->mmu_enable)
		return;

	if (hdev->mmu_func[MMU_DR_PGT].swap_in != NULL)
		hdev->mmu_func[MMU_DR_PGT].swap_in(ctx);

	if (hdev->mmu_func[MMU_HR_PGT].swap_in != NULL)
		hdev->mmu_func[MMU_HR_PGT].swap_in(ctx);
}

int hl_mmu_va_to_pa(struct hl_ctx *ctx, u64 virt_addr, u64 *phys_addr)
{
	struct hl_mmu_hop_info hops;
	u64 tmp_addr;
	int rc;

	rc = hl_mmu_get_tlb_info(ctx, virt_addr, &hops);
	if (rc)
		return rc;

	/* last hop holds the phys address and flags */
	tmp_addr = hops.hop_info[hops.used_hops - 1].hop_pte_val;
	*phys_addr = (tmp_addr & HOP_PHYS_ADDR_MASK) | (virt_addr & FLAGS_MASK);

	return 0;
}

int hl_mmu_get_tlb_info(struct hl_ctx *ctx, u64 virt_addr,
			struct hl_mmu_hop_info *hops)
{
	struct hl_device *hdev = ctx->hdev;
	struct asic_fixed_properties *prop = &hdev->asic_prop;
	struct hl_mmu_properties *mmu_prop;
	int rc;
	bool is_dram_addr;

	if (!hdev->mmu_enable)
		return -EOPNOTSUPP;

	is_dram_addr = hl_mem_area_inside_range(virt_addr, prop->dmmu.page_size,
						prop->dmmu.start_addr,
						prop->dmmu.end_addr);

	/* host-residency is the same in PMMU and HPMMU, use one of them */
	mmu_prop = is_dram_addr ? &prop->dmmu : &prop->pmmu;

	mutex_lock(&ctx->mmu_lock);

	if (mmu_prop->host_resident)
		rc = hdev->mmu_func[MMU_HR_PGT].get_tlb_info(ctx,
							virt_addr, hops);
	else
		rc = hdev->mmu_func[MMU_DR_PGT].get_tlb_info(ctx,
							virt_addr, hops);

	mutex_unlock(&ctx->mmu_lock);

	return rc;
}

int hl_mmu_if_set_funcs(struct hl_device *hdev)
{
	if (!hdev->mmu_enable)
		return 0;

	switch (hdev->asic_type) {
	case ASIC_GOYA:
	case ASIC_GAUDI:
		hl_mmu_v1_set_funcs(hdev, &hdev->mmu_func[MMU_DR_PGT]);
		break;
	default:
		dev_err(hdev->dev, "Unrecognized ASIC type %d\n",
			hdev->asic_type);
		return -EOPNOTSUPP;
	}

	return 0;
}