xref: /linux/drivers/iommu/io-pgtable-arm.c (revision e3b9f1e81de2083f359bacd2a94bf1c024f2ede0)
1 /*
2  * CPU-agnostic ARM page table allocator.
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  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
11  * GNU General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public License
14  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
15  *
16  * Copyright (C) 2014 ARM Limited
17  *
18  * Author: Will Deacon <will.deacon@arm.com>
19  */
20 
21 #define pr_fmt(fmt)	"arm-lpae io-pgtable: " fmt
22 
23 #include <linux/atomic.h>
24 #include <linux/iommu.h>
25 #include <linux/kernel.h>
26 #include <linux/sizes.h>
27 #include <linux/slab.h>
28 #include <linux/types.h>
29 #include <linux/dma-mapping.h>
30 
31 #include <asm/barrier.h>
32 
33 #include "io-pgtable.h"
34 
35 #define ARM_LPAE_MAX_ADDR_BITS		48
36 #define ARM_LPAE_S2_MAX_CONCAT_PAGES	16
37 #define ARM_LPAE_MAX_LEVELS		4
38 
39 /* Struct accessors */
40 #define io_pgtable_to_data(x)						\
41 	container_of((x), struct arm_lpae_io_pgtable, iop)
42 
43 #define io_pgtable_ops_to_data(x)					\
44 	io_pgtable_to_data(io_pgtable_ops_to_pgtable(x))
45 
46 /*
47  * For consistency with the architecture, we always consider
48  * ARM_LPAE_MAX_LEVELS levels, with the walk starting at level n >=0
49  */
50 #define ARM_LPAE_START_LVL(d)		(ARM_LPAE_MAX_LEVELS - (d)->levels)
51 
52 /*
53  * Calculate the right shift amount to get to the portion describing level l
54  * in a virtual address mapped by the pagetable in d.
55  */
56 #define ARM_LPAE_LVL_SHIFT(l,d)						\
57 	((((d)->levels - ((l) - ARM_LPAE_START_LVL(d) + 1))		\
58 	  * (d)->bits_per_level) + (d)->pg_shift)
59 
60 #define ARM_LPAE_GRANULE(d)		(1UL << (d)->pg_shift)
61 
62 #define ARM_LPAE_PAGES_PER_PGD(d)					\
63 	DIV_ROUND_UP((d)->pgd_size, ARM_LPAE_GRANULE(d))
64 
65 /*
66  * Calculate the index at level l used to map virtual address a using the
67  * pagetable in d.
68  */
69 #define ARM_LPAE_PGD_IDX(l,d)						\
70 	((l) == ARM_LPAE_START_LVL(d) ? ilog2(ARM_LPAE_PAGES_PER_PGD(d)) : 0)
71 
72 #define ARM_LPAE_LVL_IDX(a,l,d)						\
73 	(((u64)(a) >> ARM_LPAE_LVL_SHIFT(l,d)) &			\
74 	 ((1 << ((d)->bits_per_level + ARM_LPAE_PGD_IDX(l,d))) - 1))
75 
76 /* Calculate the block/page mapping size at level l for pagetable in d. */
77 #define ARM_LPAE_BLOCK_SIZE(l,d)					\
78 	(1ULL << (ilog2(sizeof(arm_lpae_iopte)) +			\
79 		((ARM_LPAE_MAX_LEVELS - (l)) * (d)->bits_per_level)))
80 
81 /* Page table bits */
82 #define ARM_LPAE_PTE_TYPE_SHIFT		0
83 #define ARM_LPAE_PTE_TYPE_MASK		0x3
84 
85 #define ARM_LPAE_PTE_TYPE_BLOCK		1
86 #define ARM_LPAE_PTE_TYPE_TABLE		3
87 #define ARM_LPAE_PTE_TYPE_PAGE		3
88 
89 #define ARM_LPAE_PTE_NSTABLE		(((arm_lpae_iopte)1) << 63)
90 #define ARM_LPAE_PTE_XN			(((arm_lpae_iopte)3) << 53)
91 #define ARM_LPAE_PTE_AF			(((arm_lpae_iopte)1) << 10)
92 #define ARM_LPAE_PTE_SH_NS		(((arm_lpae_iopte)0) << 8)
93 #define ARM_LPAE_PTE_SH_OS		(((arm_lpae_iopte)2) << 8)
94 #define ARM_LPAE_PTE_SH_IS		(((arm_lpae_iopte)3) << 8)
95 #define ARM_LPAE_PTE_NS			(((arm_lpae_iopte)1) << 5)
96 #define ARM_LPAE_PTE_VALID		(((arm_lpae_iopte)1) << 0)
97 
98 #define ARM_LPAE_PTE_ATTR_LO_MASK	(((arm_lpae_iopte)0x3ff) << 2)
99 /* Ignore the contiguous bit for block splitting */
100 #define ARM_LPAE_PTE_ATTR_HI_MASK	(((arm_lpae_iopte)6) << 52)
101 #define ARM_LPAE_PTE_ATTR_MASK		(ARM_LPAE_PTE_ATTR_LO_MASK |	\
102 					 ARM_LPAE_PTE_ATTR_HI_MASK)
103 /* Software bit for solving coherency races */
104 #define ARM_LPAE_PTE_SW_SYNC		(((arm_lpae_iopte)1) << 55)
105 
106 /* Stage-1 PTE */
107 #define ARM_LPAE_PTE_AP_UNPRIV		(((arm_lpae_iopte)1) << 6)
108 #define ARM_LPAE_PTE_AP_RDONLY		(((arm_lpae_iopte)2) << 6)
109 #define ARM_LPAE_PTE_ATTRINDX_SHIFT	2
110 #define ARM_LPAE_PTE_nG			(((arm_lpae_iopte)1) << 11)
111 
112 /* Stage-2 PTE */
113 #define ARM_LPAE_PTE_HAP_FAULT		(((arm_lpae_iopte)0) << 6)
114 #define ARM_LPAE_PTE_HAP_READ		(((arm_lpae_iopte)1) << 6)
115 #define ARM_LPAE_PTE_HAP_WRITE		(((arm_lpae_iopte)2) << 6)
116 #define ARM_LPAE_PTE_MEMATTR_OIWB	(((arm_lpae_iopte)0xf) << 2)
117 #define ARM_LPAE_PTE_MEMATTR_NC		(((arm_lpae_iopte)0x5) << 2)
118 #define ARM_LPAE_PTE_MEMATTR_DEV	(((arm_lpae_iopte)0x1) << 2)
119 
120 /* Register bits */
121 #define ARM_32_LPAE_TCR_EAE		(1 << 31)
122 #define ARM_64_LPAE_S2_TCR_RES1		(1 << 31)
123 
124 #define ARM_LPAE_TCR_EPD1		(1 << 23)
125 
126 #define ARM_LPAE_TCR_TG0_4K		(0 << 14)
127 #define ARM_LPAE_TCR_TG0_64K		(1 << 14)
128 #define ARM_LPAE_TCR_TG0_16K		(2 << 14)
129 
130 #define ARM_LPAE_TCR_SH0_SHIFT		12
131 #define ARM_LPAE_TCR_SH0_MASK		0x3
132 #define ARM_LPAE_TCR_SH_NS		0
133 #define ARM_LPAE_TCR_SH_OS		2
134 #define ARM_LPAE_TCR_SH_IS		3
135 
136 #define ARM_LPAE_TCR_ORGN0_SHIFT	10
137 #define ARM_LPAE_TCR_IRGN0_SHIFT	8
138 #define ARM_LPAE_TCR_RGN_MASK		0x3
139 #define ARM_LPAE_TCR_RGN_NC		0
140 #define ARM_LPAE_TCR_RGN_WBWA		1
141 #define ARM_LPAE_TCR_RGN_WT		2
142 #define ARM_LPAE_TCR_RGN_WB		3
143 
144 #define ARM_LPAE_TCR_SL0_SHIFT		6
145 #define ARM_LPAE_TCR_SL0_MASK		0x3
146 
147 #define ARM_LPAE_TCR_T0SZ_SHIFT		0
148 #define ARM_LPAE_TCR_SZ_MASK		0xf
149 
150 #define ARM_LPAE_TCR_PS_SHIFT		16
151 #define ARM_LPAE_TCR_PS_MASK		0x7
152 
153 #define ARM_LPAE_TCR_IPS_SHIFT		32
154 #define ARM_LPAE_TCR_IPS_MASK		0x7
155 
156 #define ARM_LPAE_TCR_PS_32_BIT		0x0ULL
157 #define ARM_LPAE_TCR_PS_36_BIT		0x1ULL
158 #define ARM_LPAE_TCR_PS_40_BIT		0x2ULL
159 #define ARM_LPAE_TCR_PS_42_BIT		0x3ULL
160 #define ARM_LPAE_TCR_PS_44_BIT		0x4ULL
161 #define ARM_LPAE_TCR_PS_48_BIT		0x5ULL
162 
163 #define ARM_LPAE_MAIR_ATTR_SHIFT(n)	((n) << 3)
164 #define ARM_LPAE_MAIR_ATTR_MASK		0xff
165 #define ARM_LPAE_MAIR_ATTR_DEVICE	0x04
166 #define ARM_LPAE_MAIR_ATTR_NC		0x44
167 #define ARM_LPAE_MAIR_ATTR_WBRWA	0xff
168 #define ARM_LPAE_MAIR_ATTR_IDX_NC	0
169 #define ARM_LPAE_MAIR_ATTR_IDX_CACHE	1
170 #define ARM_LPAE_MAIR_ATTR_IDX_DEV	2
171 
172 /* IOPTE accessors */
173 #define iopte_deref(pte,d)					\
174 	(__va((pte) & ((1ULL << ARM_LPAE_MAX_ADDR_BITS) - 1)	\
175 	& ~(ARM_LPAE_GRANULE(d) - 1ULL)))
176 
177 #define iopte_type(pte,l)					\
178 	(((pte) >> ARM_LPAE_PTE_TYPE_SHIFT) & ARM_LPAE_PTE_TYPE_MASK)
179 
180 #define iopte_prot(pte)	((pte) & ARM_LPAE_PTE_ATTR_MASK)
181 
182 #define iopte_leaf(pte,l)					\
183 	(l == (ARM_LPAE_MAX_LEVELS - 1) ?			\
184 		(iopte_type(pte,l) == ARM_LPAE_PTE_TYPE_PAGE) :	\
185 		(iopte_type(pte,l) == ARM_LPAE_PTE_TYPE_BLOCK))
186 
187 #define iopte_to_pfn(pte,d)					\
188 	(((pte) & ((1ULL << ARM_LPAE_MAX_ADDR_BITS) - 1)) >> (d)->pg_shift)
189 
190 #define pfn_to_iopte(pfn,d)					\
191 	(((pfn) << (d)->pg_shift) & ((1ULL << ARM_LPAE_MAX_ADDR_BITS) - 1))
192 
193 struct arm_lpae_io_pgtable {
194 	struct io_pgtable	iop;
195 
196 	int			levels;
197 	size_t			pgd_size;
198 	unsigned long		pg_shift;
199 	unsigned long		bits_per_level;
200 
201 	void			*pgd;
202 };
203 
204 typedef u64 arm_lpae_iopte;
205 
206 static bool selftest_running = false;
207 
208 static dma_addr_t __arm_lpae_dma_addr(void *pages)
209 {
210 	return (dma_addr_t)virt_to_phys(pages);
211 }
212 
213 static void *__arm_lpae_alloc_pages(size_t size, gfp_t gfp,
214 				    struct io_pgtable_cfg *cfg)
215 {
216 	struct device *dev = cfg->iommu_dev;
217 	dma_addr_t dma;
218 	void *pages = alloc_pages_exact(size, gfp | __GFP_ZERO);
219 
220 	if (!pages)
221 		return NULL;
222 
223 	if (!(cfg->quirks & IO_PGTABLE_QUIRK_NO_DMA)) {
224 		dma = dma_map_single(dev, pages, size, DMA_TO_DEVICE);
225 		if (dma_mapping_error(dev, dma))
226 			goto out_free;
227 		/*
228 		 * We depend on the IOMMU being able to work with any physical
229 		 * address directly, so if the DMA layer suggests otherwise by
230 		 * translating or truncating them, that bodes very badly...
231 		 */
232 		if (dma != virt_to_phys(pages))
233 			goto out_unmap;
234 	}
235 
236 	return pages;
237 
238 out_unmap:
239 	dev_err(dev, "Cannot accommodate DMA translation for IOMMU page tables\n");
240 	dma_unmap_single(dev, dma, size, DMA_TO_DEVICE);
241 out_free:
242 	free_pages_exact(pages, size);
243 	return NULL;
244 }
245 
246 static void __arm_lpae_free_pages(void *pages, size_t size,
247 				  struct io_pgtable_cfg *cfg)
248 {
249 	if (!(cfg->quirks & IO_PGTABLE_QUIRK_NO_DMA))
250 		dma_unmap_single(cfg->iommu_dev, __arm_lpae_dma_addr(pages),
251 				 size, DMA_TO_DEVICE);
252 	free_pages_exact(pages, size);
253 }
254 
255 static void __arm_lpae_sync_pte(arm_lpae_iopte *ptep,
256 				struct io_pgtable_cfg *cfg)
257 {
258 	dma_sync_single_for_device(cfg->iommu_dev, __arm_lpae_dma_addr(ptep),
259 				   sizeof(*ptep), DMA_TO_DEVICE);
260 }
261 
262 static void __arm_lpae_set_pte(arm_lpae_iopte *ptep, arm_lpae_iopte pte,
263 			       struct io_pgtable_cfg *cfg)
264 {
265 	*ptep = pte;
266 
267 	if (!(cfg->quirks & IO_PGTABLE_QUIRK_NO_DMA))
268 		__arm_lpae_sync_pte(ptep, cfg);
269 }
270 
271 static int __arm_lpae_unmap(struct arm_lpae_io_pgtable *data,
272 			    unsigned long iova, size_t size, int lvl,
273 			    arm_lpae_iopte *ptep);
274 
275 static void __arm_lpae_init_pte(struct arm_lpae_io_pgtable *data,
276 				phys_addr_t paddr, arm_lpae_iopte prot,
277 				int lvl, arm_lpae_iopte *ptep)
278 {
279 	arm_lpae_iopte pte = prot;
280 
281 	if (data->iop.cfg.quirks & IO_PGTABLE_QUIRK_ARM_NS)
282 		pte |= ARM_LPAE_PTE_NS;
283 
284 	if (lvl == ARM_LPAE_MAX_LEVELS - 1)
285 		pte |= ARM_LPAE_PTE_TYPE_PAGE;
286 	else
287 		pte |= ARM_LPAE_PTE_TYPE_BLOCK;
288 
289 	pte |= ARM_LPAE_PTE_AF | ARM_LPAE_PTE_SH_IS;
290 	pte |= pfn_to_iopte(paddr >> data->pg_shift, data);
291 
292 	__arm_lpae_set_pte(ptep, pte, &data->iop.cfg);
293 }
294 
295 static int arm_lpae_init_pte(struct arm_lpae_io_pgtable *data,
296 			     unsigned long iova, phys_addr_t paddr,
297 			     arm_lpae_iopte prot, int lvl,
298 			     arm_lpae_iopte *ptep)
299 {
300 	arm_lpae_iopte pte = *ptep;
301 
302 	if (iopte_leaf(pte, lvl)) {
303 		/* We require an unmap first */
304 		WARN_ON(!selftest_running);
305 		return -EEXIST;
306 	} else if (iopte_type(pte, lvl) == ARM_LPAE_PTE_TYPE_TABLE) {
307 		/*
308 		 * We need to unmap and free the old table before
309 		 * overwriting it with a block entry.
310 		 */
311 		arm_lpae_iopte *tblp;
312 		size_t sz = ARM_LPAE_BLOCK_SIZE(lvl, data);
313 
314 		tblp = ptep - ARM_LPAE_LVL_IDX(iova, lvl, data);
315 		if (WARN_ON(__arm_lpae_unmap(data, iova, sz, lvl, tblp) != sz))
316 			return -EINVAL;
317 	}
318 
319 	__arm_lpae_init_pte(data, paddr, prot, lvl, ptep);
320 	return 0;
321 }
322 
323 static arm_lpae_iopte arm_lpae_install_table(arm_lpae_iopte *table,
324 					     arm_lpae_iopte *ptep,
325 					     arm_lpae_iopte curr,
326 					     struct io_pgtable_cfg *cfg)
327 {
328 	arm_lpae_iopte old, new;
329 
330 	new = __pa(table) | ARM_LPAE_PTE_TYPE_TABLE;
331 	if (cfg->quirks & IO_PGTABLE_QUIRK_ARM_NS)
332 		new |= ARM_LPAE_PTE_NSTABLE;
333 
334 	/*
335 	 * Ensure the table itself is visible before its PTE can be.
336 	 * Whilst we could get away with cmpxchg64_release below, this
337 	 * doesn't have any ordering semantics when !CONFIG_SMP.
338 	 */
339 	dma_wmb();
340 
341 	old = cmpxchg64_relaxed(ptep, curr, new);
342 
343 	if ((cfg->quirks & IO_PGTABLE_QUIRK_NO_DMA) ||
344 	    (old & ARM_LPAE_PTE_SW_SYNC))
345 		return old;
346 
347 	/* Even if it's not ours, there's no point waiting; just kick it */
348 	__arm_lpae_sync_pte(ptep, cfg);
349 	if (old == curr)
350 		WRITE_ONCE(*ptep, new | ARM_LPAE_PTE_SW_SYNC);
351 
352 	return old;
353 }
354 
355 static int __arm_lpae_map(struct arm_lpae_io_pgtable *data, unsigned long iova,
356 			  phys_addr_t paddr, size_t size, arm_lpae_iopte prot,
357 			  int lvl, arm_lpae_iopte *ptep)
358 {
359 	arm_lpae_iopte *cptep, pte;
360 	size_t block_size = ARM_LPAE_BLOCK_SIZE(lvl, data);
361 	size_t tblsz = ARM_LPAE_GRANULE(data);
362 	struct io_pgtable_cfg *cfg = &data->iop.cfg;
363 
364 	/* Find our entry at the current level */
365 	ptep += ARM_LPAE_LVL_IDX(iova, lvl, data);
366 
367 	/* If we can install a leaf entry at this level, then do so */
368 	if (size == block_size && (size & cfg->pgsize_bitmap))
369 		return arm_lpae_init_pte(data, iova, paddr, prot, lvl, ptep);
370 
371 	/* We can't allocate tables at the final level */
372 	if (WARN_ON(lvl >= ARM_LPAE_MAX_LEVELS - 1))
373 		return -EINVAL;
374 
375 	/* Grab a pointer to the next level */
376 	pte = READ_ONCE(*ptep);
377 	if (!pte) {
378 		cptep = __arm_lpae_alloc_pages(tblsz, GFP_ATOMIC, cfg);
379 		if (!cptep)
380 			return -ENOMEM;
381 
382 		pte = arm_lpae_install_table(cptep, ptep, 0, cfg);
383 		if (pte)
384 			__arm_lpae_free_pages(cptep, tblsz, cfg);
385 	} else if (!(cfg->quirks & IO_PGTABLE_QUIRK_NO_DMA) &&
386 		   !(pte & ARM_LPAE_PTE_SW_SYNC)) {
387 		__arm_lpae_sync_pte(ptep, cfg);
388 	}
389 
390 	if (pte && !iopte_leaf(pte, lvl)) {
391 		cptep = iopte_deref(pte, data);
392 	} else if (pte) {
393 		/* We require an unmap first */
394 		WARN_ON(!selftest_running);
395 		return -EEXIST;
396 	}
397 
398 	/* Rinse, repeat */
399 	return __arm_lpae_map(data, iova, paddr, size, prot, lvl + 1, cptep);
400 }
401 
402 static arm_lpae_iopte arm_lpae_prot_to_pte(struct arm_lpae_io_pgtable *data,
403 					   int prot)
404 {
405 	arm_lpae_iopte pte;
406 
407 	if (data->iop.fmt == ARM_64_LPAE_S1 ||
408 	    data->iop.fmt == ARM_32_LPAE_S1) {
409 		pte = ARM_LPAE_PTE_nG;
410 
411 		if (!(prot & IOMMU_WRITE) && (prot & IOMMU_READ))
412 			pte |= ARM_LPAE_PTE_AP_RDONLY;
413 
414 		if (!(prot & IOMMU_PRIV))
415 			pte |= ARM_LPAE_PTE_AP_UNPRIV;
416 
417 		if (prot & IOMMU_MMIO)
418 			pte |= (ARM_LPAE_MAIR_ATTR_IDX_DEV
419 				<< ARM_LPAE_PTE_ATTRINDX_SHIFT);
420 		else if (prot & IOMMU_CACHE)
421 			pte |= (ARM_LPAE_MAIR_ATTR_IDX_CACHE
422 				<< ARM_LPAE_PTE_ATTRINDX_SHIFT);
423 	} else {
424 		pte = ARM_LPAE_PTE_HAP_FAULT;
425 		if (prot & IOMMU_READ)
426 			pte |= ARM_LPAE_PTE_HAP_READ;
427 		if (prot & IOMMU_WRITE)
428 			pte |= ARM_LPAE_PTE_HAP_WRITE;
429 		if (prot & IOMMU_MMIO)
430 			pte |= ARM_LPAE_PTE_MEMATTR_DEV;
431 		else if (prot & IOMMU_CACHE)
432 			pte |= ARM_LPAE_PTE_MEMATTR_OIWB;
433 		else
434 			pte |= ARM_LPAE_PTE_MEMATTR_NC;
435 	}
436 
437 	if (prot & IOMMU_NOEXEC)
438 		pte |= ARM_LPAE_PTE_XN;
439 
440 	return pte;
441 }
442 
443 static int arm_lpae_map(struct io_pgtable_ops *ops, unsigned long iova,
444 			phys_addr_t paddr, size_t size, int iommu_prot)
445 {
446 	struct arm_lpae_io_pgtable *data = io_pgtable_ops_to_data(ops);
447 	arm_lpae_iopte *ptep = data->pgd;
448 	int ret, lvl = ARM_LPAE_START_LVL(data);
449 	arm_lpae_iopte prot;
450 
451 	/* If no access, then nothing to do */
452 	if (!(iommu_prot & (IOMMU_READ | IOMMU_WRITE)))
453 		return 0;
454 
455 	if (WARN_ON(iova >= (1ULL << data->iop.cfg.ias) ||
456 		    paddr >= (1ULL << data->iop.cfg.oas)))
457 		return -ERANGE;
458 
459 	prot = arm_lpae_prot_to_pte(data, iommu_prot);
460 	ret = __arm_lpae_map(data, iova, paddr, size, prot, lvl, ptep);
461 	/*
462 	 * Synchronise all PTE updates for the new mapping before there's
463 	 * a chance for anything to kick off a table walk for the new iova.
464 	 */
465 	wmb();
466 
467 	return ret;
468 }
469 
470 static void __arm_lpae_free_pgtable(struct arm_lpae_io_pgtable *data, int lvl,
471 				    arm_lpae_iopte *ptep)
472 {
473 	arm_lpae_iopte *start, *end;
474 	unsigned long table_size;
475 
476 	if (lvl == ARM_LPAE_START_LVL(data))
477 		table_size = data->pgd_size;
478 	else
479 		table_size = ARM_LPAE_GRANULE(data);
480 
481 	start = ptep;
482 
483 	/* Only leaf entries at the last level */
484 	if (lvl == ARM_LPAE_MAX_LEVELS - 1)
485 		end = ptep;
486 	else
487 		end = (void *)ptep + table_size;
488 
489 	while (ptep != end) {
490 		arm_lpae_iopte pte = *ptep++;
491 
492 		if (!pte || iopte_leaf(pte, lvl))
493 			continue;
494 
495 		__arm_lpae_free_pgtable(data, lvl + 1, iopte_deref(pte, data));
496 	}
497 
498 	__arm_lpae_free_pages(start, table_size, &data->iop.cfg);
499 }
500 
501 static void arm_lpae_free_pgtable(struct io_pgtable *iop)
502 {
503 	struct arm_lpae_io_pgtable *data = io_pgtable_to_data(iop);
504 
505 	__arm_lpae_free_pgtable(data, ARM_LPAE_START_LVL(data), data->pgd);
506 	kfree(data);
507 }
508 
509 static int arm_lpae_split_blk_unmap(struct arm_lpae_io_pgtable *data,
510 				    unsigned long iova, size_t size,
511 				    arm_lpae_iopte blk_pte, int lvl,
512 				    arm_lpae_iopte *ptep)
513 {
514 	struct io_pgtable_cfg *cfg = &data->iop.cfg;
515 	arm_lpae_iopte pte, *tablep;
516 	phys_addr_t blk_paddr;
517 	size_t tablesz = ARM_LPAE_GRANULE(data);
518 	size_t split_sz = ARM_LPAE_BLOCK_SIZE(lvl, data);
519 	int i, unmap_idx = -1;
520 
521 	if (WARN_ON(lvl == ARM_LPAE_MAX_LEVELS))
522 		return 0;
523 
524 	tablep = __arm_lpae_alloc_pages(tablesz, GFP_ATOMIC, cfg);
525 	if (!tablep)
526 		return 0; /* Bytes unmapped */
527 
528 	if (size == split_sz)
529 		unmap_idx = ARM_LPAE_LVL_IDX(iova, lvl, data);
530 
531 	blk_paddr = iopte_to_pfn(blk_pte, data) << data->pg_shift;
532 	pte = iopte_prot(blk_pte);
533 
534 	for (i = 0; i < tablesz / sizeof(pte); i++, blk_paddr += split_sz) {
535 		/* Unmap! */
536 		if (i == unmap_idx)
537 			continue;
538 
539 		__arm_lpae_init_pte(data, blk_paddr, pte, lvl, &tablep[i]);
540 	}
541 
542 	pte = arm_lpae_install_table(tablep, ptep, blk_pte, cfg);
543 	if (pte != blk_pte) {
544 		__arm_lpae_free_pages(tablep, tablesz, cfg);
545 		/*
546 		 * We may race against someone unmapping another part of this
547 		 * block, but anything else is invalid. We can't misinterpret
548 		 * a page entry here since we're never at the last level.
549 		 */
550 		if (iopte_type(pte, lvl - 1) != ARM_LPAE_PTE_TYPE_TABLE)
551 			return 0;
552 
553 		tablep = iopte_deref(pte, data);
554 	}
555 
556 	if (unmap_idx < 0)
557 		return __arm_lpae_unmap(data, iova, size, lvl, tablep);
558 
559 	io_pgtable_tlb_add_flush(&data->iop, iova, size, size, true);
560 	return size;
561 }
562 
563 static int __arm_lpae_unmap(struct arm_lpae_io_pgtable *data,
564 			    unsigned long iova, size_t size, int lvl,
565 			    arm_lpae_iopte *ptep)
566 {
567 	arm_lpae_iopte pte;
568 	struct io_pgtable *iop = &data->iop;
569 
570 	/* Something went horribly wrong and we ran out of page table */
571 	if (WARN_ON(lvl == ARM_LPAE_MAX_LEVELS))
572 		return 0;
573 
574 	ptep += ARM_LPAE_LVL_IDX(iova, lvl, data);
575 	pte = READ_ONCE(*ptep);
576 	if (WARN_ON(!pte))
577 		return 0;
578 
579 	/* If the size matches this level, we're in the right place */
580 	if (size == ARM_LPAE_BLOCK_SIZE(lvl, data)) {
581 		__arm_lpae_set_pte(ptep, 0, &iop->cfg);
582 
583 		if (!iopte_leaf(pte, lvl)) {
584 			/* Also flush any partial walks */
585 			io_pgtable_tlb_add_flush(iop, iova, size,
586 						ARM_LPAE_GRANULE(data), false);
587 			io_pgtable_tlb_sync(iop);
588 			ptep = iopte_deref(pte, data);
589 			__arm_lpae_free_pgtable(data, lvl + 1, ptep);
590 		} else {
591 			io_pgtable_tlb_add_flush(iop, iova, size, size, true);
592 		}
593 
594 		return size;
595 	} else if (iopte_leaf(pte, lvl)) {
596 		/*
597 		 * Insert a table at the next level to map the old region,
598 		 * minus the part we want to unmap
599 		 */
600 		return arm_lpae_split_blk_unmap(data, iova, size, pte,
601 						lvl + 1, ptep);
602 	}
603 
604 	/* Keep on walkin' */
605 	ptep = iopte_deref(pte, data);
606 	return __arm_lpae_unmap(data, iova, size, lvl + 1, ptep);
607 }
608 
609 static int arm_lpae_unmap(struct io_pgtable_ops *ops, unsigned long iova,
610 			  size_t size)
611 {
612 	struct arm_lpae_io_pgtable *data = io_pgtable_ops_to_data(ops);
613 	arm_lpae_iopte *ptep = data->pgd;
614 	int lvl = ARM_LPAE_START_LVL(data);
615 
616 	if (WARN_ON(iova >= (1ULL << data->iop.cfg.ias)))
617 		return 0;
618 
619 	return __arm_lpae_unmap(data, iova, size, lvl, ptep);
620 }
621 
622 static phys_addr_t arm_lpae_iova_to_phys(struct io_pgtable_ops *ops,
623 					 unsigned long iova)
624 {
625 	struct arm_lpae_io_pgtable *data = io_pgtable_ops_to_data(ops);
626 	arm_lpae_iopte pte, *ptep = data->pgd;
627 	int lvl = ARM_LPAE_START_LVL(data);
628 
629 	do {
630 		/* Valid IOPTE pointer? */
631 		if (!ptep)
632 			return 0;
633 
634 		/* Grab the IOPTE we're interested in */
635 		ptep += ARM_LPAE_LVL_IDX(iova, lvl, data);
636 		pte = READ_ONCE(*ptep);
637 
638 		/* Valid entry? */
639 		if (!pte)
640 			return 0;
641 
642 		/* Leaf entry? */
643 		if (iopte_leaf(pte,lvl))
644 			goto found_translation;
645 
646 		/* Take it to the next level */
647 		ptep = iopte_deref(pte, data);
648 	} while (++lvl < ARM_LPAE_MAX_LEVELS);
649 
650 	/* Ran out of page tables to walk */
651 	return 0;
652 
653 found_translation:
654 	iova &= (ARM_LPAE_BLOCK_SIZE(lvl, data) - 1);
655 	return ((phys_addr_t)iopte_to_pfn(pte,data) << data->pg_shift) | iova;
656 }
657 
658 static void arm_lpae_restrict_pgsizes(struct io_pgtable_cfg *cfg)
659 {
660 	unsigned long granule;
661 
662 	/*
663 	 * We need to restrict the supported page sizes to match the
664 	 * translation regime for a particular granule. Aim to match
665 	 * the CPU page size if possible, otherwise prefer smaller sizes.
666 	 * While we're at it, restrict the block sizes to match the
667 	 * chosen granule.
668 	 */
669 	if (cfg->pgsize_bitmap & PAGE_SIZE)
670 		granule = PAGE_SIZE;
671 	else if (cfg->pgsize_bitmap & ~PAGE_MASK)
672 		granule = 1UL << __fls(cfg->pgsize_bitmap & ~PAGE_MASK);
673 	else if (cfg->pgsize_bitmap & PAGE_MASK)
674 		granule = 1UL << __ffs(cfg->pgsize_bitmap & PAGE_MASK);
675 	else
676 		granule = 0;
677 
678 	switch (granule) {
679 	case SZ_4K:
680 		cfg->pgsize_bitmap &= (SZ_4K | SZ_2M | SZ_1G);
681 		break;
682 	case SZ_16K:
683 		cfg->pgsize_bitmap &= (SZ_16K | SZ_32M);
684 		break;
685 	case SZ_64K:
686 		cfg->pgsize_bitmap &= (SZ_64K | SZ_512M);
687 		break;
688 	default:
689 		cfg->pgsize_bitmap = 0;
690 	}
691 }
692 
693 static struct arm_lpae_io_pgtable *
694 arm_lpae_alloc_pgtable(struct io_pgtable_cfg *cfg)
695 {
696 	unsigned long va_bits, pgd_bits;
697 	struct arm_lpae_io_pgtable *data;
698 
699 	arm_lpae_restrict_pgsizes(cfg);
700 
701 	if (!(cfg->pgsize_bitmap & (SZ_4K | SZ_16K | SZ_64K)))
702 		return NULL;
703 
704 	if (cfg->ias > ARM_LPAE_MAX_ADDR_BITS)
705 		return NULL;
706 
707 	if (cfg->oas > ARM_LPAE_MAX_ADDR_BITS)
708 		return NULL;
709 
710 	if (!selftest_running && cfg->iommu_dev->dma_pfn_offset) {
711 		dev_err(cfg->iommu_dev, "Cannot accommodate DMA offset for IOMMU page tables\n");
712 		return NULL;
713 	}
714 
715 	data = kmalloc(sizeof(*data), GFP_KERNEL);
716 	if (!data)
717 		return NULL;
718 
719 	data->pg_shift = __ffs(cfg->pgsize_bitmap);
720 	data->bits_per_level = data->pg_shift - ilog2(sizeof(arm_lpae_iopte));
721 
722 	va_bits = cfg->ias - data->pg_shift;
723 	data->levels = DIV_ROUND_UP(va_bits, data->bits_per_level);
724 
725 	/* Calculate the actual size of our pgd (without concatenation) */
726 	pgd_bits = va_bits - (data->bits_per_level * (data->levels - 1));
727 	data->pgd_size = 1UL << (pgd_bits + ilog2(sizeof(arm_lpae_iopte)));
728 
729 	data->iop.ops = (struct io_pgtable_ops) {
730 		.map		= arm_lpae_map,
731 		.unmap		= arm_lpae_unmap,
732 		.iova_to_phys	= arm_lpae_iova_to_phys,
733 	};
734 
735 	return data;
736 }
737 
738 static struct io_pgtable *
739 arm_64_lpae_alloc_pgtable_s1(struct io_pgtable_cfg *cfg, void *cookie)
740 {
741 	u64 reg;
742 	struct arm_lpae_io_pgtable *data;
743 
744 	if (cfg->quirks & ~(IO_PGTABLE_QUIRK_ARM_NS | IO_PGTABLE_QUIRK_NO_DMA))
745 		return NULL;
746 
747 	data = arm_lpae_alloc_pgtable(cfg);
748 	if (!data)
749 		return NULL;
750 
751 	/* TCR */
752 	reg = (ARM_LPAE_TCR_SH_IS << ARM_LPAE_TCR_SH0_SHIFT) |
753 	      (ARM_LPAE_TCR_RGN_WBWA << ARM_LPAE_TCR_IRGN0_SHIFT) |
754 	      (ARM_LPAE_TCR_RGN_WBWA << ARM_LPAE_TCR_ORGN0_SHIFT);
755 
756 	switch (ARM_LPAE_GRANULE(data)) {
757 	case SZ_4K:
758 		reg |= ARM_LPAE_TCR_TG0_4K;
759 		break;
760 	case SZ_16K:
761 		reg |= ARM_LPAE_TCR_TG0_16K;
762 		break;
763 	case SZ_64K:
764 		reg |= ARM_LPAE_TCR_TG0_64K;
765 		break;
766 	}
767 
768 	switch (cfg->oas) {
769 	case 32:
770 		reg |= (ARM_LPAE_TCR_PS_32_BIT << ARM_LPAE_TCR_IPS_SHIFT);
771 		break;
772 	case 36:
773 		reg |= (ARM_LPAE_TCR_PS_36_BIT << ARM_LPAE_TCR_IPS_SHIFT);
774 		break;
775 	case 40:
776 		reg |= (ARM_LPAE_TCR_PS_40_BIT << ARM_LPAE_TCR_IPS_SHIFT);
777 		break;
778 	case 42:
779 		reg |= (ARM_LPAE_TCR_PS_42_BIT << ARM_LPAE_TCR_IPS_SHIFT);
780 		break;
781 	case 44:
782 		reg |= (ARM_LPAE_TCR_PS_44_BIT << ARM_LPAE_TCR_IPS_SHIFT);
783 		break;
784 	case 48:
785 		reg |= (ARM_LPAE_TCR_PS_48_BIT << ARM_LPAE_TCR_IPS_SHIFT);
786 		break;
787 	default:
788 		goto out_free_data;
789 	}
790 
791 	reg |= (64ULL - cfg->ias) << ARM_LPAE_TCR_T0SZ_SHIFT;
792 
793 	/* Disable speculative walks through TTBR1 */
794 	reg |= ARM_LPAE_TCR_EPD1;
795 	cfg->arm_lpae_s1_cfg.tcr = reg;
796 
797 	/* MAIRs */
798 	reg = (ARM_LPAE_MAIR_ATTR_NC
799 	       << ARM_LPAE_MAIR_ATTR_SHIFT(ARM_LPAE_MAIR_ATTR_IDX_NC)) |
800 	      (ARM_LPAE_MAIR_ATTR_WBRWA
801 	       << ARM_LPAE_MAIR_ATTR_SHIFT(ARM_LPAE_MAIR_ATTR_IDX_CACHE)) |
802 	      (ARM_LPAE_MAIR_ATTR_DEVICE
803 	       << ARM_LPAE_MAIR_ATTR_SHIFT(ARM_LPAE_MAIR_ATTR_IDX_DEV));
804 
805 	cfg->arm_lpae_s1_cfg.mair[0] = reg;
806 	cfg->arm_lpae_s1_cfg.mair[1] = 0;
807 
808 	/* Looking good; allocate a pgd */
809 	data->pgd = __arm_lpae_alloc_pages(data->pgd_size, GFP_KERNEL, cfg);
810 	if (!data->pgd)
811 		goto out_free_data;
812 
813 	/* Ensure the empty pgd is visible before any actual TTBR write */
814 	wmb();
815 
816 	/* TTBRs */
817 	cfg->arm_lpae_s1_cfg.ttbr[0] = virt_to_phys(data->pgd);
818 	cfg->arm_lpae_s1_cfg.ttbr[1] = 0;
819 	return &data->iop;
820 
821 out_free_data:
822 	kfree(data);
823 	return NULL;
824 }
825 
826 static struct io_pgtable *
827 arm_64_lpae_alloc_pgtable_s2(struct io_pgtable_cfg *cfg, void *cookie)
828 {
829 	u64 reg, sl;
830 	struct arm_lpae_io_pgtable *data;
831 
832 	/* The NS quirk doesn't apply at stage 2 */
833 	if (cfg->quirks & ~IO_PGTABLE_QUIRK_NO_DMA)
834 		return NULL;
835 
836 	data = arm_lpae_alloc_pgtable(cfg);
837 	if (!data)
838 		return NULL;
839 
840 	/*
841 	 * Concatenate PGDs at level 1 if possible in order to reduce
842 	 * the depth of the stage-2 walk.
843 	 */
844 	if (data->levels == ARM_LPAE_MAX_LEVELS) {
845 		unsigned long pgd_pages;
846 
847 		pgd_pages = data->pgd_size >> ilog2(sizeof(arm_lpae_iopte));
848 		if (pgd_pages <= ARM_LPAE_S2_MAX_CONCAT_PAGES) {
849 			data->pgd_size = pgd_pages << data->pg_shift;
850 			data->levels--;
851 		}
852 	}
853 
854 	/* VTCR */
855 	reg = ARM_64_LPAE_S2_TCR_RES1 |
856 	     (ARM_LPAE_TCR_SH_IS << ARM_LPAE_TCR_SH0_SHIFT) |
857 	     (ARM_LPAE_TCR_RGN_WBWA << ARM_LPAE_TCR_IRGN0_SHIFT) |
858 	     (ARM_LPAE_TCR_RGN_WBWA << ARM_LPAE_TCR_ORGN0_SHIFT);
859 
860 	sl = ARM_LPAE_START_LVL(data);
861 
862 	switch (ARM_LPAE_GRANULE(data)) {
863 	case SZ_4K:
864 		reg |= ARM_LPAE_TCR_TG0_4K;
865 		sl++; /* SL0 format is different for 4K granule size */
866 		break;
867 	case SZ_16K:
868 		reg |= ARM_LPAE_TCR_TG0_16K;
869 		break;
870 	case SZ_64K:
871 		reg |= ARM_LPAE_TCR_TG0_64K;
872 		break;
873 	}
874 
875 	switch (cfg->oas) {
876 	case 32:
877 		reg |= (ARM_LPAE_TCR_PS_32_BIT << ARM_LPAE_TCR_PS_SHIFT);
878 		break;
879 	case 36:
880 		reg |= (ARM_LPAE_TCR_PS_36_BIT << ARM_LPAE_TCR_PS_SHIFT);
881 		break;
882 	case 40:
883 		reg |= (ARM_LPAE_TCR_PS_40_BIT << ARM_LPAE_TCR_PS_SHIFT);
884 		break;
885 	case 42:
886 		reg |= (ARM_LPAE_TCR_PS_42_BIT << ARM_LPAE_TCR_PS_SHIFT);
887 		break;
888 	case 44:
889 		reg |= (ARM_LPAE_TCR_PS_44_BIT << ARM_LPAE_TCR_PS_SHIFT);
890 		break;
891 	case 48:
892 		reg |= (ARM_LPAE_TCR_PS_48_BIT << ARM_LPAE_TCR_PS_SHIFT);
893 		break;
894 	default:
895 		goto out_free_data;
896 	}
897 
898 	reg |= (64ULL - cfg->ias) << ARM_LPAE_TCR_T0SZ_SHIFT;
899 	reg |= (~sl & ARM_LPAE_TCR_SL0_MASK) << ARM_LPAE_TCR_SL0_SHIFT;
900 	cfg->arm_lpae_s2_cfg.vtcr = reg;
901 
902 	/* Allocate pgd pages */
903 	data->pgd = __arm_lpae_alloc_pages(data->pgd_size, GFP_KERNEL, cfg);
904 	if (!data->pgd)
905 		goto out_free_data;
906 
907 	/* Ensure the empty pgd is visible before any actual TTBR write */
908 	wmb();
909 
910 	/* VTTBR */
911 	cfg->arm_lpae_s2_cfg.vttbr = virt_to_phys(data->pgd);
912 	return &data->iop;
913 
914 out_free_data:
915 	kfree(data);
916 	return NULL;
917 }
918 
919 static struct io_pgtable *
920 arm_32_lpae_alloc_pgtable_s1(struct io_pgtable_cfg *cfg, void *cookie)
921 {
922 	struct io_pgtable *iop;
923 
924 	if (cfg->ias > 32 || cfg->oas > 40)
925 		return NULL;
926 
927 	cfg->pgsize_bitmap &= (SZ_4K | SZ_2M | SZ_1G);
928 	iop = arm_64_lpae_alloc_pgtable_s1(cfg, cookie);
929 	if (iop) {
930 		cfg->arm_lpae_s1_cfg.tcr |= ARM_32_LPAE_TCR_EAE;
931 		cfg->arm_lpae_s1_cfg.tcr &= 0xffffffff;
932 	}
933 
934 	return iop;
935 }
936 
937 static struct io_pgtable *
938 arm_32_lpae_alloc_pgtable_s2(struct io_pgtable_cfg *cfg, void *cookie)
939 {
940 	struct io_pgtable *iop;
941 
942 	if (cfg->ias > 40 || cfg->oas > 40)
943 		return NULL;
944 
945 	cfg->pgsize_bitmap &= (SZ_4K | SZ_2M | SZ_1G);
946 	iop = arm_64_lpae_alloc_pgtable_s2(cfg, cookie);
947 	if (iop)
948 		cfg->arm_lpae_s2_cfg.vtcr &= 0xffffffff;
949 
950 	return iop;
951 }
952 
953 struct io_pgtable_init_fns io_pgtable_arm_64_lpae_s1_init_fns = {
954 	.alloc	= arm_64_lpae_alloc_pgtable_s1,
955 	.free	= arm_lpae_free_pgtable,
956 };
957 
958 struct io_pgtable_init_fns io_pgtable_arm_64_lpae_s2_init_fns = {
959 	.alloc	= arm_64_lpae_alloc_pgtable_s2,
960 	.free	= arm_lpae_free_pgtable,
961 };
962 
963 struct io_pgtable_init_fns io_pgtable_arm_32_lpae_s1_init_fns = {
964 	.alloc	= arm_32_lpae_alloc_pgtable_s1,
965 	.free	= arm_lpae_free_pgtable,
966 };
967 
968 struct io_pgtable_init_fns io_pgtable_arm_32_lpae_s2_init_fns = {
969 	.alloc	= arm_32_lpae_alloc_pgtable_s2,
970 	.free	= arm_lpae_free_pgtable,
971 };
972 
973 #ifdef CONFIG_IOMMU_IO_PGTABLE_LPAE_SELFTEST
974 
975 static struct io_pgtable_cfg *cfg_cookie;
976 
977 static void dummy_tlb_flush_all(void *cookie)
978 {
979 	WARN_ON(cookie != cfg_cookie);
980 }
981 
982 static void dummy_tlb_add_flush(unsigned long iova, size_t size,
983 				size_t granule, bool leaf, void *cookie)
984 {
985 	WARN_ON(cookie != cfg_cookie);
986 	WARN_ON(!(size & cfg_cookie->pgsize_bitmap));
987 }
988 
989 static void dummy_tlb_sync(void *cookie)
990 {
991 	WARN_ON(cookie != cfg_cookie);
992 }
993 
994 static const struct iommu_gather_ops dummy_tlb_ops __initconst = {
995 	.tlb_flush_all	= dummy_tlb_flush_all,
996 	.tlb_add_flush	= dummy_tlb_add_flush,
997 	.tlb_sync	= dummy_tlb_sync,
998 };
999 
1000 static void __init arm_lpae_dump_ops(struct io_pgtable_ops *ops)
1001 {
1002 	struct arm_lpae_io_pgtable *data = io_pgtable_ops_to_data(ops);
1003 	struct io_pgtable_cfg *cfg = &data->iop.cfg;
1004 
1005 	pr_err("cfg: pgsize_bitmap 0x%lx, ias %u-bit\n",
1006 		cfg->pgsize_bitmap, cfg->ias);
1007 	pr_err("data: %d levels, 0x%zx pgd_size, %lu pg_shift, %lu bits_per_level, pgd @ %p\n",
1008 		data->levels, data->pgd_size, data->pg_shift,
1009 		data->bits_per_level, data->pgd);
1010 }
1011 
1012 #define __FAIL(ops, i)	({						\
1013 		WARN(1, "selftest: test failed for fmt idx %d\n", (i));	\
1014 		arm_lpae_dump_ops(ops);					\
1015 		selftest_running = false;				\
1016 		-EFAULT;						\
1017 })
1018 
1019 static int __init arm_lpae_run_tests(struct io_pgtable_cfg *cfg)
1020 {
1021 	static const enum io_pgtable_fmt fmts[] = {
1022 		ARM_64_LPAE_S1,
1023 		ARM_64_LPAE_S2,
1024 	};
1025 
1026 	int i, j;
1027 	unsigned long iova;
1028 	size_t size;
1029 	struct io_pgtable_ops *ops;
1030 
1031 	selftest_running = true;
1032 
1033 	for (i = 0; i < ARRAY_SIZE(fmts); ++i) {
1034 		cfg_cookie = cfg;
1035 		ops = alloc_io_pgtable_ops(fmts[i], cfg, cfg);
1036 		if (!ops) {
1037 			pr_err("selftest: failed to allocate io pgtable ops\n");
1038 			return -ENOMEM;
1039 		}
1040 
1041 		/*
1042 		 * Initial sanity checks.
1043 		 * Empty page tables shouldn't provide any translations.
1044 		 */
1045 		if (ops->iova_to_phys(ops, 42))
1046 			return __FAIL(ops, i);
1047 
1048 		if (ops->iova_to_phys(ops, SZ_1G + 42))
1049 			return __FAIL(ops, i);
1050 
1051 		if (ops->iova_to_phys(ops, SZ_2G + 42))
1052 			return __FAIL(ops, i);
1053 
1054 		/*
1055 		 * Distinct mappings of different granule sizes.
1056 		 */
1057 		iova = 0;
1058 		for_each_set_bit(j, &cfg->pgsize_bitmap, BITS_PER_LONG) {
1059 			size = 1UL << j;
1060 
1061 			if (ops->map(ops, iova, iova, size, IOMMU_READ |
1062 							    IOMMU_WRITE |
1063 							    IOMMU_NOEXEC |
1064 							    IOMMU_CACHE))
1065 				return __FAIL(ops, i);
1066 
1067 			/* Overlapping mappings */
1068 			if (!ops->map(ops, iova, iova + size, size,
1069 				      IOMMU_READ | IOMMU_NOEXEC))
1070 				return __FAIL(ops, i);
1071 
1072 			if (ops->iova_to_phys(ops, iova + 42) != (iova + 42))
1073 				return __FAIL(ops, i);
1074 
1075 			iova += SZ_1G;
1076 		}
1077 
1078 		/* Partial unmap */
1079 		size = 1UL << __ffs(cfg->pgsize_bitmap);
1080 		if (ops->unmap(ops, SZ_1G + size, size) != size)
1081 			return __FAIL(ops, i);
1082 
1083 		/* Remap of partial unmap */
1084 		if (ops->map(ops, SZ_1G + size, size, size, IOMMU_READ))
1085 			return __FAIL(ops, i);
1086 
1087 		if (ops->iova_to_phys(ops, SZ_1G + size + 42) != (size + 42))
1088 			return __FAIL(ops, i);
1089 
1090 		/* Full unmap */
1091 		iova = 0;
1092 		j = find_first_bit(&cfg->pgsize_bitmap, BITS_PER_LONG);
1093 		while (j != BITS_PER_LONG) {
1094 			size = 1UL << j;
1095 
1096 			if (ops->unmap(ops, iova, size) != size)
1097 				return __FAIL(ops, i);
1098 
1099 			if (ops->iova_to_phys(ops, iova + 42))
1100 				return __FAIL(ops, i);
1101 
1102 			/* Remap full block */
1103 			if (ops->map(ops, iova, iova, size, IOMMU_WRITE))
1104 				return __FAIL(ops, i);
1105 
1106 			if (ops->iova_to_phys(ops, iova + 42) != (iova + 42))
1107 				return __FAIL(ops, i);
1108 
1109 			iova += SZ_1G;
1110 			j++;
1111 			j = find_next_bit(&cfg->pgsize_bitmap, BITS_PER_LONG, j);
1112 		}
1113 
1114 		free_io_pgtable_ops(ops);
1115 	}
1116 
1117 	selftest_running = false;
1118 	return 0;
1119 }
1120 
1121 static int __init arm_lpae_do_selftests(void)
1122 {
1123 	static const unsigned long pgsize[] = {
1124 		SZ_4K | SZ_2M | SZ_1G,
1125 		SZ_16K | SZ_32M,
1126 		SZ_64K | SZ_512M,
1127 	};
1128 
1129 	static const unsigned int ias[] = {
1130 		32, 36, 40, 42, 44, 48,
1131 	};
1132 
1133 	int i, j, pass = 0, fail = 0;
1134 	struct io_pgtable_cfg cfg = {
1135 		.tlb = &dummy_tlb_ops,
1136 		.oas = 48,
1137 		.quirks = IO_PGTABLE_QUIRK_NO_DMA,
1138 	};
1139 
1140 	for (i = 0; i < ARRAY_SIZE(pgsize); ++i) {
1141 		for (j = 0; j < ARRAY_SIZE(ias); ++j) {
1142 			cfg.pgsize_bitmap = pgsize[i];
1143 			cfg.ias = ias[j];
1144 			pr_info("selftest: pgsize_bitmap 0x%08lx, IAS %u\n",
1145 				pgsize[i], ias[j]);
1146 			if (arm_lpae_run_tests(&cfg))
1147 				fail++;
1148 			else
1149 				pass++;
1150 		}
1151 	}
1152 
1153 	pr_info("selftest: completed with %d PASS %d FAIL\n", pass, fail);
1154 	return fail ? -EFAULT : 0;
1155 }
1156 subsys_initcall(arm_lpae_do_selftests);
1157 #endif
1158