xref: /openbmc/linux/drivers/iommu/tegra-smmu.c (revision efe4a1ac)
1 /*
2  * Copyright (C) 2011-2014 NVIDIA CORPORATION.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of the GNU General Public License version 2 as
6  * published by the Free Software Foundation.
7  */
8 
9 #include <linux/bitops.h>
10 #include <linux/debugfs.h>
11 #include <linux/err.h>
12 #include <linux/iommu.h>
13 #include <linux/kernel.h>
14 #include <linux/of.h>
15 #include <linux/of_device.h>
16 #include <linux/platform_device.h>
17 #include <linux/slab.h>
18 #include <linux/dma-mapping.h>
19 
20 #include <soc/tegra/ahb.h>
21 #include <soc/tegra/mc.h>
22 
23 struct tegra_smmu {
24 	void __iomem *regs;
25 	struct device *dev;
26 
27 	struct tegra_mc *mc;
28 	const struct tegra_smmu_soc *soc;
29 
30 	unsigned long pfn_mask;
31 	unsigned long tlb_mask;
32 
33 	unsigned long *asids;
34 	struct mutex lock;
35 
36 	struct list_head list;
37 
38 	struct dentry *debugfs;
39 };
40 
41 struct tegra_smmu_as {
42 	struct iommu_domain domain;
43 	struct tegra_smmu *smmu;
44 	unsigned int use_count;
45 	u32 *count;
46 	struct page **pts;
47 	struct page *pd;
48 	dma_addr_t pd_dma;
49 	unsigned id;
50 	u32 attr;
51 };
52 
53 static struct tegra_smmu_as *to_smmu_as(struct iommu_domain *dom)
54 {
55 	return container_of(dom, struct tegra_smmu_as, domain);
56 }
57 
58 static inline void smmu_writel(struct tegra_smmu *smmu, u32 value,
59 			       unsigned long offset)
60 {
61 	writel(value, smmu->regs + offset);
62 }
63 
64 static inline u32 smmu_readl(struct tegra_smmu *smmu, unsigned long offset)
65 {
66 	return readl(smmu->regs + offset);
67 }
68 
69 #define SMMU_CONFIG 0x010
70 #define  SMMU_CONFIG_ENABLE (1 << 0)
71 
72 #define SMMU_TLB_CONFIG 0x14
73 #define  SMMU_TLB_CONFIG_HIT_UNDER_MISS (1 << 29)
74 #define  SMMU_TLB_CONFIG_ROUND_ROBIN_ARBITRATION (1 << 28)
75 #define  SMMU_TLB_CONFIG_ACTIVE_LINES(smmu) \
76 	((smmu)->soc->num_tlb_lines & (smmu)->tlb_mask)
77 
78 #define SMMU_PTC_CONFIG 0x18
79 #define  SMMU_PTC_CONFIG_ENABLE (1 << 29)
80 #define  SMMU_PTC_CONFIG_REQ_LIMIT(x) (((x) & 0x0f) << 24)
81 #define  SMMU_PTC_CONFIG_INDEX_MAP(x) ((x) & 0x3f)
82 
83 #define SMMU_PTB_ASID 0x01c
84 #define  SMMU_PTB_ASID_VALUE(x) ((x) & 0x7f)
85 
86 #define SMMU_PTB_DATA 0x020
87 #define  SMMU_PTB_DATA_VALUE(dma, attr) ((dma) >> 12 | (attr))
88 
89 #define SMMU_MK_PDE(dma, attr) ((dma) >> SMMU_PTE_SHIFT | (attr))
90 
91 #define SMMU_TLB_FLUSH 0x030
92 #define  SMMU_TLB_FLUSH_VA_MATCH_ALL     (0 << 0)
93 #define  SMMU_TLB_FLUSH_VA_MATCH_SECTION (2 << 0)
94 #define  SMMU_TLB_FLUSH_VA_MATCH_GROUP   (3 << 0)
95 #define  SMMU_TLB_FLUSH_ASID(x)          (((x) & 0x7f) << 24)
96 #define  SMMU_TLB_FLUSH_VA_SECTION(addr) ((((addr) & 0xffc00000) >> 12) | \
97 					  SMMU_TLB_FLUSH_VA_MATCH_SECTION)
98 #define  SMMU_TLB_FLUSH_VA_GROUP(addr)   ((((addr) & 0xffffc000) >> 12) | \
99 					  SMMU_TLB_FLUSH_VA_MATCH_GROUP)
100 #define  SMMU_TLB_FLUSH_ASID_MATCH       (1 << 31)
101 
102 #define SMMU_PTC_FLUSH 0x034
103 #define  SMMU_PTC_FLUSH_TYPE_ALL (0 << 0)
104 #define  SMMU_PTC_FLUSH_TYPE_ADR (1 << 0)
105 
106 #define SMMU_PTC_FLUSH_HI 0x9b8
107 #define  SMMU_PTC_FLUSH_HI_MASK 0x3
108 
109 /* per-SWGROUP SMMU_*_ASID register */
110 #define SMMU_ASID_ENABLE (1 << 31)
111 #define SMMU_ASID_MASK 0x7f
112 #define SMMU_ASID_VALUE(x) ((x) & SMMU_ASID_MASK)
113 
114 /* page table definitions */
115 #define SMMU_NUM_PDE 1024
116 #define SMMU_NUM_PTE 1024
117 
118 #define SMMU_SIZE_PD (SMMU_NUM_PDE * 4)
119 #define SMMU_SIZE_PT (SMMU_NUM_PTE * 4)
120 
121 #define SMMU_PDE_SHIFT 22
122 #define SMMU_PTE_SHIFT 12
123 
124 #define SMMU_PD_READABLE	(1 << 31)
125 #define SMMU_PD_WRITABLE	(1 << 30)
126 #define SMMU_PD_NONSECURE	(1 << 29)
127 
128 #define SMMU_PDE_READABLE	(1 << 31)
129 #define SMMU_PDE_WRITABLE	(1 << 30)
130 #define SMMU_PDE_NONSECURE	(1 << 29)
131 #define SMMU_PDE_NEXT		(1 << 28)
132 
133 #define SMMU_PTE_READABLE	(1 << 31)
134 #define SMMU_PTE_WRITABLE	(1 << 30)
135 #define SMMU_PTE_NONSECURE	(1 << 29)
136 
137 #define SMMU_PDE_ATTR		(SMMU_PDE_READABLE | SMMU_PDE_WRITABLE | \
138 				 SMMU_PDE_NONSECURE)
139 #define SMMU_PTE_ATTR		(SMMU_PTE_READABLE | SMMU_PTE_WRITABLE | \
140 				 SMMU_PTE_NONSECURE)
141 
142 static unsigned int iova_pd_index(unsigned long iova)
143 {
144 	return (iova >> SMMU_PDE_SHIFT) & (SMMU_NUM_PDE - 1);
145 }
146 
147 static unsigned int iova_pt_index(unsigned long iova)
148 {
149 	return (iova >> SMMU_PTE_SHIFT) & (SMMU_NUM_PTE - 1);
150 }
151 
152 static bool smmu_dma_addr_valid(struct tegra_smmu *smmu, dma_addr_t addr)
153 {
154 	addr >>= 12;
155 	return (addr & smmu->pfn_mask) == addr;
156 }
157 
158 static dma_addr_t smmu_pde_to_dma(u32 pde)
159 {
160 	return pde << 12;
161 }
162 
163 static void smmu_flush_ptc_all(struct tegra_smmu *smmu)
164 {
165 	smmu_writel(smmu, SMMU_PTC_FLUSH_TYPE_ALL, SMMU_PTC_FLUSH);
166 }
167 
168 static inline void smmu_flush_ptc(struct tegra_smmu *smmu, dma_addr_t dma,
169 				  unsigned long offset)
170 {
171 	u32 value;
172 
173 	offset &= ~(smmu->mc->soc->atom_size - 1);
174 
175 	if (smmu->mc->soc->num_address_bits > 32) {
176 #ifdef CONFIG_ARCH_DMA_ADDR_T_64BIT
177 		value = (dma >> 32) & SMMU_PTC_FLUSH_HI_MASK;
178 #else
179 		value = 0;
180 #endif
181 		smmu_writel(smmu, value, SMMU_PTC_FLUSH_HI);
182 	}
183 
184 	value = (dma + offset) | SMMU_PTC_FLUSH_TYPE_ADR;
185 	smmu_writel(smmu, value, SMMU_PTC_FLUSH);
186 }
187 
188 static inline void smmu_flush_tlb(struct tegra_smmu *smmu)
189 {
190 	smmu_writel(smmu, SMMU_TLB_FLUSH_VA_MATCH_ALL, SMMU_TLB_FLUSH);
191 }
192 
193 static inline void smmu_flush_tlb_asid(struct tegra_smmu *smmu,
194 				       unsigned long asid)
195 {
196 	u32 value;
197 
198 	value = SMMU_TLB_FLUSH_ASID_MATCH | SMMU_TLB_FLUSH_ASID(asid) |
199 		SMMU_TLB_FLUSH_VA_MATCH_ALL;
200 	smmu_writel(smmu, value, SMMU_TLB_FLUSH);
201 }
202 
203 static inline void smmu_flush_tlb_section(struct tegra_smmu *smmu,
204 					  unsigned long asid,
205 					  unsigned long iova)
206 {
207 	u32 value;
208 
209 	value = SMMU_TLB_FLUSH_ASID_MATCH | SMMU_TLB_FLUSH_ASID(asid) |
210 		SMMU_TLB_FLUSH_VA_SECTION(iova);
211 	smmu_writel(smmu, value, SMMU_TLB_FLUSH);
212 }
213 
214 static inline void smmu_flush_tlb_group(struct tegra_smmu *smmu,
215 					unsigned long asid,
216 					unsigned long iova)
217 {
218 	u32 value;
219 
220 	value = SMMU_TLB_FLUSH_ASID_MATCH | SMMU_TLB_FLUSH_ASID(asid) |
221 		SMMU_TLB_FLUSH_VA_GROUP(iova);
222 	smmu_writel(smmu, value, SMMU_TLB_FLUSH);
223 }
224 
225 static inline void smmu_flush(struct tegra_smmu *smmu)
226 {
227 	smmu_readl(smmu, SMMU_CONFIG);
228 }
229 
230 static int tegra_smmu_alloc_asid(struct tegra_smmu *smmu, unsigned int *idp)
231 {
232 	unsigned long id;
233 
234 	mutex_lock(&smmu->lock);
235 
236 	id = find_first_zero_bit(smmu->asids, smmu->soc->num_asids);
237 	if (id >= smmu->soc->num_asids) {
238 		mutex_unlock(&smmu->lock);
239 		return -ENOSPC;
240 	}
241 
242 	set_bit(id, smmu->asids);
243 	*idp = id;
244 
245 	mutex_unlock(&smmu->lock);
246 	return 0;
247 }
248 
249 static void tegra_smmu_free_asid(struct tegra_smmu *smmu, unsigned int id)
250 {
251 	mutex_lock(&smmu->lock);
252 	clear_bit(id, smmu->asids);
253 	mutex_unlock(&smmu->lock);
254 }
255 
256 static bool tegra_smmu_capable(enum iommu_cap cap)
257 {
258 	return false;
259 }
260 
261 static struct iommu_domain *tegra_smmu_domain_alloc(unsigned type)
262 {
263 	struct tegra_smmu_as *as;
264 
265 	if (type != IOMMU_DOMAIN_UNMANAGED)
266 		return NULL;
267 
268 	as = kzalloc(sizeof(*as), GFP_KERNEL);
269 	if (!as)
270 		return NULL;
271 
272 	as->attr = SMMU_PD_READABLE | SMMU_PD_WRITABLE | SMMU_PD_NONSECURE;
273 
274 	as->pd = alloc_page(GFP_KERNEL | __GFP_DMA | __GFP_ZERO);
275 	if (!as->pd) {
276 		kfree(as);
277 		return NULL;
278 	}
279 
280 	as->count = kcalloc(SMMU_NUM_PDE, sizeof(u32), GFP_KERNEL);
281 	if (!as->count) {
282 		__free_page(as->pd);
283 		kfree(as);
284 		return NULL;
285 	}
286 
287 	as->pts = kcalloc(SMMU_NUM_PDE, sizeof(*as->pts), GFP_KERNEL);
288 	if (!as->pts) {
289 		kfree(as->count);
290 		__free_page(as->pd);
291 		kfree(as);
292 		return NULL;
293 	}
294 
295 	/* setup aperture */
296 	as->domain.geometry.aperture_start = 0;
297 	as->domain.geometry.aperture_end = 0xffffffff;
298 	as->domain.geometry.force_aperture = true;
299 
300 	return &as->domain;
301 }
302 
303 static void tegra_smmu_domain_free(struct iommu_domain *domain)
304 {
305 	struct tegra_smmu_as *as = to_smmu_as(domain);
306 
307 	/* TODO: free page directory and page tables */
308 
309 	kfree(as);
310 }
311 
312 static const struct tegra_smmu_swgroup *
313 tegra_smmu_find_swgroup(struct tegra_smmu *smmu, unsigned int swgroup)
314 {
315 	const struct tegra_smmu_swgroup *group = NULL;
316 	unsigned int i;
317 
318 	for (i = 0; i < smmu->soc->num_swgroups; i++) {
319 		if (smmu->soc->swgroups[i].swgroup == swgroup) {
320 			group = &smmu->soc->swgroups[i];
321 			break;
322 		}
323 	}
324 
325 	return group;
326 }
327 
328 static void tegra_smmu_enable(struct tegra_smmu *smmu, unsigned int swgroup,
329 			      unsigned int asid)
330 {
331 	const struct tegra_smmu_swgroup *group;
332 	unsigned int i;
333 	u32 value;
334 
335 	for (i = 0; i < smmu->soc->num_clients; i++) {
336 		const struct tegra_mc_client *client = &smmu->soc->clients[i];
337 
338 		if (client->swgroup != swgroup)
339 			continue;
340 
341 		value = smmu_readl(smmu, client->smmu.reg);
342 		value |= BIT(client->smmu.bit);
343 		smmu_writel(smmu, value, client->smmu.reg);
344 	}
345 
346 	group = tegra_smmu_find_swgroup(smmu, swgroup);
347 	if (group) {
348 		value = smmu_readl(smmu, group->reg);
349 		value &= ~SMMU_ASID_MASK;
350 		value |= SMMU_ASID_VALUE(asid);
351 		value |= SMMU_ASID_ENABLE;
352 		smmu_writel(smmu, value, group->reg);
353 	}
354 }
355 
356 static void tegra_smmu_disable(struct tegra_smmu *smmu, unsigned int swgroup,
357 			       unsigned int asid)
358 {
359 	const struct tegra_smmu_swgroup *group;
360 	unsigned int i;
361 	u32 value;
362 
363 	group = tegra_smmu_find_swgroup(smmu, swgroup);
364 	if (group) {
365 		value = smmu_readl(smmu, group->reg);
366 		value &= ~SMMU_ASID_MASK;
367 		value |= SMMU_ASID_VALUE(asid);
368 		value &= ~SMMU_ASID_ENABLE;
369 		smmu_writel(smmu, value, group->reg);
370 	}
371 
372 	for (i = 0; i < smmu->soc->num_clients; i++) {
373 		const struct tegra_mc_client *client = &smmu->soc->clients[i];
374 
375 		if (client->swgroup != swgroup)
376 			continue;
377 
378 		value = smmu_readl(smmu, client->smmu.reg);
379 		value &= ~BIT(client->smmu.bit);
380 		smmu_writel(smmu, value, client->smmu.reg);
381 	}
382 }
383 
384 static int tegra_smmu_as_prepare(struct tegra_smmu *smmu,
385 				 struct tegra_smmu_as *as)
386 {
387 	u32 value;
388 	int err;
389 
390 	if (as->use_count > 0) {
391 		as->use_count++;
392 		return 0;
393 	}
394 
395 	as->pd_dma = dma_map_page(smmu->dev, as->pd, 0, SMMU_SIZE_PD,
396 				  DMA_TO_DEVICE);
397 	if (dma_mapping_error(smmu->dev, as->pd_dma))
398 		return -ENOMEM;
399 
400 	/* We can't handle 64-bit DMA addresses */
401 	if (!smmu_dma_addr_valid(smmu, as->pd_dma)) {
402 		err = -ENOMEM;
403 		goto err_unmap;
404 	}
405 
406 	err = tegra_smmu_alloc_asid(smmu, &as->id);
407 	if (err < 0)
408 		goto err_unmap;
409 
410 	smmu_flush_ptc(smmu, as->pd_dma, 0);
411 	smmu_flush_tlb_asid(smmu, as->id);
412 
413 	smmu_writel(smmu, as->id & 0x7f, SMMU_PTB_ASID);
414 	value = SMMU_PTB_DATA_VALUE(as->pd_dma, as->attr);
415 	smmu_writel(smmu, value, SMMU_PTB_DATA);
416 	smmu_flush(smmu);
417 
418 	as->smmu = smmu;
419 	as->use_count++;
420 
421 	return 0;
422 
423 err_unmap:
424 	dma_unmap_page(smmu->dev, as->pd_dma, SMMU_SIZE_PD, DMA_TO_DEVICE);
425 	return err;
426 }
427 
428 static void tegra_smmu_as_unprepare(struct tegra_smmu *smmu,
429 				    struct tegra_smmu_as *as)
430 {
431 	if (--as->use_count > 0)
432 		return;
433 
434 	tegra_smmu_free_asid(smmu, as->id);
435 
436 	dma_unmap_page(smmu->dev, as->pd_dma, SMMU_SIZE_PD, DMA_TO_DEVICE);
437 
438 	as->smmu = NULL;
439 }
440 
441 static int tegra_smmu_attach_dev(struct iommu_domain *domain,
442 				 struct device *dev)
443 {
444 	struct tegra_smmu *smmu = dev->archdata.iommu;
445 	struct tegra_smmu_as *as = to_smmu_as(domain);
446 	struct device_node *np = dev->of_node;
447 	struct of_phandle_args args;
448 	unsigned int index = 0;
449 	int err = 0;
450 
451 	while (!of_parse_phandle_with_args(np, "iommus", "#iommu-cells", index,
452 					   &args)) {
453 		unsigned int swgroup = args.args[0];
454 
455 		if (args.np != smmu->dev->of_node) {
456 			of_node_put(args.np);
457 			continue;
458 		}
459 
460 		of_node_put(args.np);
461 
462 		err = tegra_smmu_as_prepare(smmu, as);
463 		if (err < 0)
464 			return err;
465 
466 		tegra_smmu_enable(smmu, swgroup, as->id);
467 		index++;
468 	}
469 
470 	if (index == 0)
471 		return -ENODEV;
472 
473 	return 0;
474 }
475 
476 static void tegra_smmu_detach_dev(struct iommu_domain *domain, struct device *dev)
477 {
478 	struct tegra_smmu_as *as = to_smmu_as(domain);
479 	struct device_node *np = dev->of_node;
480 	struct tegra_smmu *smmu = as->smmu;
481 	struct of_phandle_args args;
482 	unsigned int index = 0;
483 
484 	while (!of_parse_phandle_with_args(np, "iommus", "#iommu-cells", index,
485 					   &args)) {
486 		unsigned int swgroup = args.args[0];
487 
488 		if (args.np != smmu->dev->of_node) {
489 			of_node_put(args.np);
490 			continue;
491 		}
492 
493 		of_node_put(args.np);
494 
495 		tegra_smmu_disable(smmu, swgroup, as->id);
496 		tegra_smmu_as_unprepare(smmu, as);
497 		index++;
498 	}
499 }
500 
501 static void tegra_smmu_set_pde(struct tegra_smmu_as *as, unsigned long iova,
502 			       u32 value)
503 {
504 	unsigned int pd_index = iova_pd_index(iova);
505 	struct tegra_smmu *smmu = as->smmu;
506 	u32 *pd = page_address(as->pd);
507 	unsigned long offset = pd_index * sizeof(*pd);
508 
509 	/* Set the page directory entry first */
510 	pd[pd_index] = value;
511 
512 	/* The flush the page directory entry from caches */
513 	dma_sync_single_range_for_device(smmu->dev, as->pd_dma, offset,
514 					 sizeof(*pd), DMA_TO_DEVICE);
515 
516 	/* And flush the iommu */
517 	smmu_flush_ptc(smmu, as->pd_dma, offset);
518 	smmu_flush_tlb_section(smmu, as->id, iova);
519 	smmu_flush(smmu);
520 }
521 
522 static u32 *tegra_smmu_pte_offset(struct page *pt_page, unsigned long iova)
523 {
524 	u32 *pt = page_address(pt_page);
525 
526 	return pt + iova_pt_index(iova);
527 }
528 
529 static u32 *tegra_smmu_pte_lookup(struct tegra_smmu_as *as, unsigned long iova,
530 				  dma_addr_t *dmap)
531 {
532 	unsigned int pd_index = iova_pd_index(iova);
533 	struct page *pt_page;
534 	u32 *pd;
535 
536 	pt_page = as->pts[pd_index];
537 	if (!pt_page)
538 		return NULL;
539 
540 	pd = page_address(as->pd);
541 	*dmap = smmu_pde_to_dma(pd[pd_index]);
542 
543 	return tegra_smmu_pte_offset(pt_page, iova);
544 }
545 
546 static u32 *as_get_pte(struct tegra_smmu_as *as, dma_addr_t iova,
547 		       dma_addr_t *dmap)
548 {
549 	unsigned int pde = iova_pd_index(iova);
550 	struct tegra_smmu *smmu = as->smmu;
551 
552 	if (!as->pts[pde]) {
553 		struct page *page;
554 		dma_addr_t dma;
555 
556 		page = alloc_page(GFP_KERNEL | __GFP_DMA | __GFP_ZERO);
557 		if (!page)
558 			return NULL;
559 
560 		dma = dma_map_page(smmu->dev, page, 0, SMMU_SIZE_PT,
561 				   DMA_TO_DEVICE);
562 		if (dma_mapping_error(smmu->dev, dma)) {
563 			__free_page(page);
564 			return NULL;
565 		}
566 
567 		if (!smmu_dma_addr_valid(smmu, dma)) {
568 			dma_unmap_page(smmu->dev, dma, SMMU_SIZE_PT,
569 				       DMA_TO_DEVICE);
570 			__free_page(page);
571 			return NULL;
572 		}
573 
574 		as->pts[pde] = page;
575 
576 		tegra_smmu_set_pde(as, iova, SMMU_MK_PDE(dma, SMMU_PDE_ATTR |
577 							      SMMU_PDE_NEXT));
578 
579 		*dmap = dma;
580 	} else {
581 		u32 *pd = page_address(as->pd);
582 
583 		*dmap = smmu_pde_to_dma(pd[pde]);
584 	}
585 
586 	return tegra_smmu_pte_offset(as->pts[pde], iova);
587 }
588 
589 static void tegra_smmu_pte_get_use(struct tegra_smmu_as *as, unsigned long iova)
590 {
591 	unsigned int pd_index = iova_pd_index(iova);
592 
593 	as->count[pd_index]++;
594 }
595 
596 static void tegra_smmu_pte_put_use(struct tegra_smmu_as *as, unsigned long iova)
597 {
598 	unsigned int pde = iova_pd_index(iova);
599 	struct page *page = as->pts[pde];
600 
601 	/*
602 	 * When no entries in this page table are used anymore, return the
603 	 * memory page to the system.
604 	 */
605 	if (--as->count[pde] == 0) {
606 		struct tegra_smmu *smmu = as->smmu;
607 		u32 *pd = page_address(as->pd);
608 		dma_addr_t pte_dma = smmu_pde_to_dma(pd[pde]);
609 
610 		tegra_smmu_set_pde(as, iova, 0);
611 
612 		dma_unmap_page(smmu->dev, pte_dma, SMMU_SIZE_PT, DMA_TO_DEVICE);
613 		__free_page(page);
614 		as->pts[pde] = NULL;
615 	}
616 }
617 
618 static void tegra_smmu_set_pte(struct tegra_smmu_as *as, unsigned long iova,
619 			       u32 *pte, dma_addr_t pte_dma, u32 val)
620 {
621 	struct tegra_smmu *smmu = as->smmu;
622 	unsigned long offset = offset_in_page(pte);
623 
624 	*pte = val;
625 
626 	dma_sync_single_range_for_device(smmu->dev, pte_dma, offset,
627 					 4, DMA_TO_DEVICE);
628 	smmu_flush_ptc(smmu, pte_dma, offset);
629 	smmu_flush_tlb_group(smmu, as->id, iova);
630 	smmu_flush(smmu);
631 }
632 
633 static int tegra_smmu_map(struct iommu_domain *domain, unsigned long iova,
634 			  phys_addr_t paddr, size_t size, int prot)
635 {
636 	struct tegra_smmu_as *as = to_smmu_as(domain);
637 	dma_addr_t pte_dma;
638 	u32 *pte;
639 
640 	pte = as_get_pte(as, iova, &pte_dma);
641 	if (!pte)
642 		return -ENOMEM;
643 
644 	/* If we aren't overwriting a pre-existing entry, increment use */
645 	if (*pte == 0)
646 		tegra_smmu_pte_get_use(as, iova);
647 
648 	tegra_smmu_set_pte(as, iova, pte, pte_dma,
649 			   __phys_to_pfn(paddr) | SMMU_PTE_ATTR);
650 
651 	return 0;
652 }
653 
654 static size_t tegra_smmu_unmap(struct iommu_domain *domain, unsigned long iova,
655 			       size_t size)
656 {
657 	struct tegra_smmu_as *as = to_smmu_as(domain);
658 	dma_addr_t pte_dma;
659 	u32 *pte;
660 
661 	pte = tegra_smmu_pte_lookup(as, iova, &pte_dma);
662 	if (!pte || !*pte)
663 		return 0;
664 
665 	tegra_smmu_set_pte(as, iova, pte, pte_dma, 0);
666 	tegra_smmu_pte_put_use(as, iova);
667 
668 	return size;
669 }
670 
671 static phys_addr_t tegra_smmu_iova_to_phys(struct iommu_domain *domain,
672 					   dma_addr_t iova)
673 {
674 	struct tegra_smmu_as *as = to_smmu_as(domain);
675 	unsigned long pfn;
676 	dma_addr_t pte_dma;
677 	u32 *pte;
678 
679 	pte = tegra_smmu_pte_lookup(as, iova, &pte_dma);
680 	if (!pte || !*pte)
681 		return 0;
682 
683 	pfn = *pte & as->smmu->pfn_mask;
684 
685 	return PFN_PHYS(pfn);
686 }
687 
688 static struct tegra_smmu *tegra_smmu_find(struct device_node *np)
689 {
690 	struct platform_device *pdev;
691 	struct tegra_mc *mc;
692 
693 	pdev = of_find_device_by_node(np);
694 	if (!pdev)
695 		return NULL;
696 
697 	mc = platform_get_drvdata(pdev);
698 	if (!mc)
699 		return NULL;
700 
701 	return mc->smmu;
702 }
703 
704 static int tegra_smmu_add_device(struct device *dev)
705 {
706 	struct device_node *np = dev->of_node;
707 	struct of_phandle_args args;
708 	unsigned int index = 0;
709 
710 	while (of_parse_phandle_with_args(np, "iommus", "#iommu-cells", index,
711 					  &args) == 0) {
712 		struct tegra_smmu *smmu;
713 
714 		smmu = tegra_smmu_find(args.np);
715 		if (smmu) {
716 			/*
717 			 * Only a single IOMMU master interface is currently
718 			 * supported by the Linux kernel, so abort after the
719 			 * first match.
720 			 */
721 			dev->archdata.iommu = smmu;
722 			break;
723 		}
724 
725 		index++;
726 	}
727 
728 	return 0;
729 }
730 
731 static void tegra_smmu_remove_device(struct device *dev)
732 {
733 	dev->archdata.iommu = NULL;
734 }
735 
736 static const struct iommu_ops tegra_smmu_ops = {
737 	.capable = tegra_smmu_capable,
738 	.domain_alloc = tegra_smmu_domain_alloc,
739 	.domain_free = tegra_smmu_domain_free,
740 	.attach_dev = tegra_smmu_attach_dev,
741 	.detach_dev = tegra_smmu_detach_dev,
742 	.add_device = tegra_smmu_add_device,
743 	.remove_device = tegra_smmu_remove_device,
744 	.map = tegra_smmu_map,
745 	.unmap = tegra_smmu_unmap,
746 	.map_sg = default_iommu_map_sg,
747 	.iova_to_phys = tegra_smmu_iova_to_phys,
748 
749 	.pgsize_bitmap = SZ_4K,
750 };
751 
752 static void tegra_smmu_ahb_enable(void)
753 {
754 	static const struct of_device_id ahb_match[] = {
755 		{ .compatible = "nvidia,tegra30-ahb", },
756 		{ }
757 	};
758 	struct device_node *ahb;
759 
760 	ahb = of_find_matching_node(NULL, ahb_match);
761 	if (ahb) {
762 		tegra_ahb_enable_smmu(ahb);
763 		of_node_put(ahb);
764 	}
765 }
766 
767 static int tegra_smmu_swgroups_show(struct seq_file *s, void *data)
768 {
769 	struct tegra_smmu *smmu = s->private;
770 	unsigned int i;
771 	u32 value;
772 
773 	seq_printf(s, "swgroup    enabled  ASID\n");
774 	seq_printf(s, "------------------------\n");
775 
776 	for (i = 0; i < smmu->soc->num_swgroups; i++) {
777 		const struct tegra_smmu_swgroup *group = &smmu->soc->swgroups[i];
778 		const char *status;
779 		unsigned int asid;
780 
781 		value = smmu_readl(smmu, group->reg);
782 
783 		if (value & SMMU_ASID_ENABLE)
784 			status = "yes";
785 		else
786 			status = "no";
787 
788 		asid = value & SMMU_ASID_MASK;
789 
790 		seq_printf(s, "%-9s  %-7s  %#04x\n", group->name, status,
791 			   asid);
792 	}
793 
794 	return 0;
795 }
796 
797 static int tegra_smmu_swgroups_open(struct inode *inode, struct file *file)
798 {
799 	return single_open(file, tegra_smmu_swgroups_show, inode->i_private);
800 }
801 
802 static const struct file_operations tegra_smmu_swgroups_fops = {
803 	.open = tegra_smmu_swgroups_open,
804 	.read = seq_read,
805 	.llseek = seq_lseek,
806 	.release = single_release,
807 };
808 
809 static int tegra_smmu_clients_show(struct seq_file *s, void *data)
810 {
811 	struct tegra_smmu *smmu = s->private;
812 	unsigned int i;
813 	u32 value;
814 
815 	seq_printf(s, "client       enabled\n");
816 	seq_printf(s, "--------------------\n");
817 
818 	for (i = 0; i < smmu->soc->num_clients; i++) {
819 		const struct tegra_mc_client *client = &smmu->soc->clients[i];
820 		const char *status;
821 
822 		value = smmu_readl(smmu, client->smmu.reg);
823 
824 		if (value & BIT(client->smmu.bit))
825 			status = "yes";
826 		else
827 			status = "no";
828 
829 		seq_printf(s, "%-12s %s\n", client->name, status);
830 	}
831 
832 	return 0;
833 }
834 
835 static int tegra_smmu_clients_open(struct inode *inode, struct file *file)
836 {
837 	return single_open(file, tegra_smmu_clients_show, inode->i_private);
838 }
839 
840 static const struct file_operations tegra_smmu_clients_fops = {
841 	.open = tegra_smmu_clients_open,
842 	.read = seq_read,
843 	.llseek = seq_lseek,
844 	.release = single_release,
845 };
846 
847 static void tegra_smmu_debugfs_init(struct tegra_smmu *smmu)
848 {
849 	smmu->debugfs = debugfs_create_dir("smmu", NULL);
850 	if (!smmu->debugfs)
851 		return;
852 
853 	debugfs_create_file("swgroups", S_IRUGO, smmu->debugfs, smmu,
854 			    &tegra_smmu_swgroups_fops);
855 	debugfs_create_file("clients", S_IRUGO, smmu->debugfs, smmu,
856 			    &tegra_smmu_clients_fops);
857 }
858 
859 static void tegra_smmu_debugfs_exit(struct tegra_smmu *smmu)
860 {
861 	debugfs_remove_recursive(smmu->debugfs);
862 }
863 
864 struct tegra_smmu *tegra_smmu_probe(struct device *dev,
865 				    const struct tegra_smmu_soc *soc,
866 				    struct tegra_mc *mc)
867 {
868 	struct tegra_smmu *smmu;
869 	size_t size;
870 	u32 value;
871 	int err;
872 
873 	/* This can happen on Tegra20 which doesn't have an SMMU */
874 	if (!soc)
875 		return NULL;
876 
877 	smmu = devm_kzalloc(dev, sizeof(*smmu), GFP_KERNEL);
878 	if (!smmu)
879 		return ERR_PTR(-ENOMEM);
880 
881 	/*
882 	 * This is a bit of a hack. Ideally we'd want to simply return this
883 	 * value. However the IOMMU registration process will attempt to add
884 	 * all devices to the IOMMU when bus_set_iommu() is called. In order
885 	 * not to rely on global variables to track the IOMMU instance, we
886 	 * set it here so that it can be looked up from the .add_device()
887 	 * callback via the IOMMU device's .drvdata field.
888 	 */
889 	mc->smmu = smmu;
890 
891 	size = BITS_TO_LONGS(soc->num_asids) * sizeof(long);
892 
893 	smmu->asids = devm_kzalloc(dev, size, GFP_KERNEL);
894 	if (!smmu->asids)
895 		return ERR_PTR(-ENOMEM);
896 
897 	mutex_init(&smmu->lock);
898 
899 	smmu->regs = mc->regs;
900 	smmu->soc = soc;
901 	smmu->dev = dev;
902 	smmu->mc = mc;
903 
904 	smmu->pfn_mask = BIT_MASK(mc->soc->num_address_bits - PAGE_SHIFT) - 1;
905 	dev_dbg(dev, "address bits: %u, PFN mask: %#lx\n",
906 		mc->soc->num_address_bits, smmu->pfn_mask);
907 	smmu->tlb_mask = (smmu->soc->num_tlb_lines << 1) - 1;
908 	dev_dbg(dev, "TLB lines: %u, mask: %#lx\n", smmu->soc->num_tlb_lines,
909 		smmu->tlb_mask);
910 
911 	value = SMMU_PTC_CONFIG_ENABLE | SMMU_PTC_CONFIG_INDEX_MAP(0x3f);
912 
913 	if (soc->supports_request_limit)
914 		value |= SMMU_PTC_CONFIG_REQ_LIMIT(8);
915 
916 	smmu_writel(smmu, value, SMMU_PTC_CONFIG);
917 
918 	value = SMMU_TLB_CONFIG_HIT_UNDER_MISS |
919 		SMMU_TLB_CONFIG_ACTIVE_LINES(smmu);
920 
921 	if (soc->supports_round_robin_arbitration)
922 		value |= SMMU_TLB_CONFIG_ROUND_ROBIN_ARBITRATION;
923 
924 	smmu_writel(smmu, value, SMMU_TLB_CONFIG);
925 
926 	smmu_flush_ptc_all(smmu);
927 	smmu_flush_tlb(smmu);
928 	smmu_writel(smmu, SMMU_CONFIG_ENABLE, SMMU_CONFIG);
929 	smmu_flush(smmu);
930 
931 	tegra_smmu_ahb_enable();
932 
933 	err = bus_set_iommu(&platform_bus_type, &tegra_smmu_ops);
934 	if (err < 0)
935 		return ERR_PTR(err);
936 
937 	if (IS_ENABLED(CONFIG_DEBUG_FS))
938 		tegra_smmu_debugfs_init(smmu);
939 
940 	return smmu;
941 }
942 
943 void tegra_smmu_remove(struct tegra_smmu *smmu)
944 {
945 	if (IS_ENABLED(CONFIG_DEBUG_FS))
946 		tegra_smmu_debugfs_exit(smmu);
947 }
948