• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * CPU-agnostic ARM page table allocator.
4  *
5  * Copyright (C) 2014 ARM Limited
6  *
7  * Author: Will Deacon <will.deacon@arm.com>
8  */
9 
10 #define pr_fmt(fmt)	"arm-lpae io-pgtable: " fmt
11 
12 #include <linux/atomic.h>
13 #include <linux/bitops.h>
14 #include <linux/io-pgtable.h>
15 #include <linux/kernel.h>
16 #include <linux/sizes.h>
17 #include <linux/slab.h>
18 #include <linux/types.h>
19 #include <linux/dma-mapping.h>
20 
21 #include <asm/barrier.h>
22 
23 #include "io-pgtable-arm.h"
24 
25 #define ARM_LPAE_MAX_ADDR_BITS		52
26 #define ARM_LPAE_S2_MAX_CONCAT_PAGES	16
27 #define ARM_LPAE_MAX_LEVELS		4
28 
29 /* Struct accessors */
30 #define io_pgtable_to_data(x)						\
31 	container_of((x), struct arm_lpae_io_pgtable, iop)
32 
33 #define io_pgtable_ops_to_data(x)					\
34 	io_pgtable_to_data(io_pgtable_ops_to_pgtable(x))
35 
36 /*
37  * Calculate the right shift amount to get to the portion describing level l
38  * in a virtual address mapped by the pagetable in d.
39  */
40 #define ARM_LPAE_LVL_SHIFT(l,d)						\
41 	(((ARM_LPAE_MAX_LEVELS - (l)) * (d)->bits_per_level) +		\
42 	ilog2(sizeof(arm_lpae_iopte)))
43 
44 #define ARM_LPAE_GRANULE(d)						\
45 	(sizeof(arm_lpae_iopte) << (d)->bits_per_level)
46 #define ARM_LPAE_PGD_SIZE(d)						\
47 	(sizeof(arm_lpae_iopte) << (d)->pgd_bits)
48 
49 #define ARM_LPAE_PTES_PER_TABLE(d)					\
50 	(ARM_LPAE_GRANULE(d) >> ilog2(sizeof(arm_lpae_iopte)))
51 
52 /*
53  * Calculate the index at level l used to map virtual address a using the
54  * pagetable in d.
55  */
56 #define ARM_LPAE_PGD_IDX(l,d)						\
57 	((l) == (d)->start_level ? (d)->pgd_bits - (d)->bits_per_level : 0)
58 
59 #define ARM_LPAE_LVL_IDX(a,l,d)						\
60 	(((u64)(a) >> ARM_LPAE_LVL_SHIFT(l,d)) &			\
61 	 ((1 << ((d)->bits_per_level + ARM_LPAE_PGD_IDX(l,d))) - 1))
62 
63 /* Calculate the block/page mapping size at level l for pagetable in d. */
64 #define ARM_LPAE_BLOCK_SIZE(l,d)	(1ULL << ARM_LPAE_LVL_SHIFT(l,d))
65 
66 /* Page table bits */
67 #define ARM_LPAE_PTE_TYPE_SHIFT		0
68 #define ARM_LPAE_PTE_TYPE_MASK		0x3
69 
70 #define ARM_LPAE_PTE_TYPE_BLOCK		1
71 #define ARM_LPAE_PTE_TYPE_TABLE		3
72 #define ARM_LPAE_PTE_TYPE_PAGE		3
73 
74 #define ARM_LPAE_PTE_ADDR_MASK		GENMASK_ULL(47,12)
75 
76 #define ARM_LPAE_PTE_NSTABLE		(((arm_lpae_iopte)1) << 63)
77 #define ARM_LPAE_PTE_XN			(((arm_lpae_iopte)3) << 53)
78 #define ARM_LPAE_PTE_AF			(((arm_lpae_iopte)1) << 10)
79 #define ARM_LPAE_PTE_SH_NS		(((arm_lpae_iopte)0) << 8)
80 #define ARM_LPAE_PTE_SH_OS		(((arm_lpae_iopte)2) << 8)
81 #define ARM_LPAE_PTE_SH_IS		(((arm_lpae_iopte)3) << 8)
82 #define ARM_LPAE_PTE_NS			(((arm_lpae_iopte)1) << 5)
83 #define ARM_LPAE_PTE_VALID		(((arm_lpae_iopte)1) << 0)
84 
85 #define ARM_LPAE_PTE_ATTR_LO_MASK	(((arm_lpae_iopte)0x3ff) << 2)
86 /* Ignore the contiguous bit for block splitting */
87 #define ARM_LPAE_PTE_ATTR_HI_MASK	(((arm_lpae_iopte)6) << 52)
88 #define ARM_LPAE_PTE_ATTR_MASK		(ARM_LPAE_PTE_ATTR_LO_MASK |	\
89 					 ARM_LPAE_PTE_ATTR_HI_MASK)
90 /* Software bit for solving coherency races */
91 #define ARM_LPAE_PTE_SW_SYNC		(((arm_lpae_iopte)1) << 55)
92 
93 /* Stage-1 PTE */
94 #define ARM_LPAE_PTE_AP_UNPRIV		(((arm_lpae_iopte)1) << 6)
95 #define ARM_LPAE_PTE_AP_RDONLY		(((arm_lpae_iopte)2) << 6)
96 #define ARM_LPAE_PTE_ATTRINDX_SHIFT	2
97 #define ARM_LPAE_PTE_nG			(((arm_lpae_iopte)1) << 11)
98 
99 /* Stage-2 PTE */
100 #define ARM_LPAE_PTE_HAP_FAULT		(((arm_lpae_iopte)0) << 6)
101 #define ARM_LPAE_PTE_HAP_READ		(((arm_lpae_iopte)1) << 6)
102 #define ARM_LPAE_PTE_HAP_WRITE		(((arm_lpae_iopte)2) << 6)
103 #define ARM_LPAE_PTE_MEMATTR_OIWB	(((arm_lpae_iopte)0xf) << 2)
104 #define ARM_LPAE_PTE_MEMATTR_NC		(((arm_lpae_iopte)0x5) << 2)
105 #define ARM_LPAE_PTE_MEMATTR_DEV	(((arm_lpae_iopte)0x1) << 2)
106 
107 /* Register bits */
108 #define ARM_LPAE_VTCR_SL0_MASK		0x3
109 
110 #define ARM_LPAE_TCR_T0SZ_SHIFT		0
111 
112 #define ARM_LPAE_VTCR_PS_SHIFT		16
113 #define ARM_LPAE_VTCR_PS_MASK		0x7
114 
115 #define ARM_LPAE_MAIR_ATTR_SHIFT(n)	((n) << 3)
116 #define ARM_LPAE_MAIR_ATTR_MASK		0xff
117 #define ARM_LPAE_MAIR_ATTR_DEVICE	0x04
118 #define ARM_LPAE_MAIR_ATTR_NC		0x44
119 #define ARM_LPAE_MAIR_ATTR_INC_OWBRWA	0xf4
120 #define ARM_LPAE_MAIR_ATTR_WBRWA	0xff
121 #define ARM_LPAE_MAIR_ATTR_INC_OWBRANWA		0xe4ULL
122 #define ARM_LPAE_MAIR_ATTR_IWBRWA_OWBRANWA	0xefULL
123 #define ARM_LPAE_MAIR_ATTR_IDX_NC	0
124 #define ARM_LPAE_MAIR_ATTR_IDX_CACHE	1
125 #define ARM_LPAE_MAIR_ATTR_IDX_DEV	2
126 #define ARM_LPAE_MAIR_ATTR_IDX_INC_OCACHE	3
127 #define ARM_LPAE_MAIR_ATTR_IDX_INC_OCACHE_NWA		4
128 #define ARM_LPAE_MAIR_ATTR_IDX_ICACHE_OCACHE_NWA	5
129 
130 #define ARM_MALI_LPAE_TTBR_ADRMODE_TABLE (3u << 0)
131 #define ARM_MALI_LPAE_TTBR_READ_INNER	BIT(2)
132 #define ARM_MALI_LPAE_TTBR_SHARE_OUTER	BIT(4)
133 
134 #define ARM_MALI_LPAE_MEMATTR_IMP_DEF	0x88ULL
135 #define ARM_MALI_LPAE_MEMATTR_WRITE_ALLOC 0x8DULL
136 
137 /* IOPTE accessors */
138 #define iopte_deref(pte,d) __va(iopte_to_paddr(pte, d))
139 
140 #define iopte_type(pte)					\
141 	(((pte) >> ARM_LPAE_PTE_TYPE_SHIFT) & ARM_LPAE_PTE_TYPE_MASK)
142 
143 #define iopte_prot(pte)	((pte) & ARM_LPAE_PTE_ATTR_MASK)
144 
145 struct arm_lpae_io_pgtable {
146 	struct io_pgtable	iop;
147 
148 	int			pgd_bits;
149 	int			start_level;
150 	int			bits_per_level;
151 
152 	void			*pgd;
153 };
154 
155 typedef u64 arm_lpae_iopte;
156 
iopte_leaf(arm_lpae_iopte pte,int lvl,enum io_pgtable_fmt fmt)157 static inline bool iopte_leaf(arm_lpae_iopte pte, int lvl,
158 			      enum io_pgtable_fmt fmt)
159 {
160 	if (lvl == (ARM_LPAE_MAX_LEVELS - 1) && fmt != ARM_MALI_LPAE)
161 		return iopte_type(pte) == ARM_LPAE_PTE_TYPE_PAGE;
162 
163 	return iopte_type(pte) == ARM_LPAE_PTE_TYPE_BLOCK;
164 }
165 
paddr_to_iopte(phys_addr_t paddr,struct arm_lpae_io_pgtable * data)166 static arm_lpae_iopte paddr_to_iopte(phys_addr_t paddr,
167 				     struct arm_lpae_io_pgtable *data)
168 {
169 	arm_lpae_iopte pte = paddr;
170 
171 	/* Of the bits which overlap, either 51:48 or 15:12 are always RES0 */
172 	return (pte | (pte >> (48 - 12))) & ARM_LPAE_PTE_ADDR_MASK;
173 }
174 
iopte_to_paddr(arm_lpae_iopte pte,struct arm_lpae_io_pgtable * data)175 static phys_addr_t iopte_to_paddr(arm_lpae_iopte pte,
176 				  struct arm_lpae_io_pgtable *data)
177 {
178 	u64 paddr = pte & ARM_LPAE_PTE_ADDR_MASK;
179 
180 	if (ARM_LPAE_GRANULE(data) < SZ_64K)
181 		return paddr;
182 
183 	/* Rotate the packed high-order bits back to the top */
184 	return (paddr | (paddr << (48 - 12))) & (ARM_LPAE_PTE_ADDR_MASK << 4);
185 }
186 
187 static bool selftest_running = false;
188 
__arm_lpae_dma_addr(void * pages)189 static dma_addr_t __arm_lpae_dma_addr(void *pages)
190 {
191 	return (dma_addr_t)virt_to_phys(pages);
192 }
193 
__arm_lpae_alloc_pages(size_t size,gfp_t gfp,struct io_pgtable_cfg * cfg)194 static void *__arm_lpae_alloc_pages(size_t size, gfp_t gfp,
195 				    struct io_pgtable_cfg *cfg)
196 {
197 	struct device *dev = cfg->iommu_dev;
198 	int order = get_order(size);
199 	struct page *p;
200 	dma_addr_t dma;
201 	void *pages;
202 
203 	VM_BUG_ON((gfp & __GFP_HIGHMEM));
204 	p = alloc_pages_node(dev_to_node(dev), gfp | __GFP_ZERO, order);
205 	if (!p)
206 		return NULL;
207 
208 	pages = page_address(p);
209 	if (!cfg->coherent_walk) {
210 		dma = dma_map_single(dev, pages, size, DMA_TO_DEVICE);
211 		if (dma_mapping_error(dev, dma))
212 			goto out_free;
213 		/*
214 		 * We depend on the IOMMU being able to work with any physical
215 		 * address directly, so if the DMA layer suggests otherwise by
216 		 * translating or truncating them, that bodes very badly...
217 		 */
218 		if (dma != virt_to_phys(pages))
219 			goto out_unmap;
220 	}
221 
222 	return pages;
223 
224 out_unmap:
225 	dev_err(dev, "Cannot accommodate DMA translation for IOMMU page tables\n");
226 	dma_unmap_single(dev, dma, size, DMA_TO_DEVICE);
227 out_free:
228 	__free_pages(p, order);
229 	return NULL;
230 }
231 
__arm_lpae_free_pages(void * pages,size_t size,struct io_pgtable_cfg * cfg)232 static void __arm_lpae_free_pages(void *pages, size_t size,
233 				  struct io_pgtable_cfg *cfg)
234 {
235 	if (!cfg->coherent_walk)
236 		dma_unmap_single(cfg->iommu_dev, __arm_lpae_dma_addr(pages),
237 				 size, DMA_TO_DEVICE);
238 	free_pages((unsigned long)pages, get_order(size));
239 }
240 
__arm_lpae_sync_pte(arm_lpae_iopte * ptep,int num_entries,struct io_pgtable_cfg * cfg)241 static void __arm_lpae_sync_pte(arm_lpae_iopte *ptep, int num_entries,
242 				struct io_pgtable_cfg *cfg)
243 {
244 	dma_sync_single_for_device(cfg->iommu_dev, __arm_lpae_dma_addr(ptep),
245 				   sizeof(*ptep) * num_entries, DMA_TO_DEVICE);
246 }
247 
__arm_lpae_clear_pte(arm_lpae_iopte * ptep,struct io_pgtable_cfg * cfg)248 static void __arm_lpae_clear_pte(arm_lpae_iopte *ptep, struct io_pgtable_cfg *cfg)
249 {
250 
251 	*ptep = 0;
252 
253 	if (!cfg->coherent_walk)
254 		__arm_lpae_sync_pte(ptep, 1, cfg);
255 }
256 
257 static size_t __arm_lpae_unmap(struct arm_lpae_io_pgtable *data,
258 			       struct iommu_iotlb_gather *gather,
259 			       unsigned long iova, size_t size, size_t pgcount,
260 			       int lvl, arm_lpae_iopte *ptep);
261 
__arm_lpae_init_pte(struct arm_lpae_io_pgtable * data,phys_addr_t paddr,arm_lpae_iopte prot,int lvl,int num_entries,arm_lpae_iopte * ptep)262 static void __arm_lpae_init_pte(struct arm_lpae_io_pgtable *data,
263 				phys_addr_t paddr, arm_lpae_iopte prot,
264 				int lvl, int num_entries, arm_lpae_iopte *ptep)
265 {
266 	arm_lpae_iopte pte = prot;
267 	struct io_pgtable_cfg *cfg = &data->iop.cfg;
268 	size_t sz = ARM_LPAE_BLOCK_SIZE(lvl, data);
269 	int i;
270 
271 	if (data->iop.fmt != ARM_MALI_LPAE && lvl == ARM_LPAE_MAX_LEVELS - 1)
272 		pte |= ARM_LPAE_PTE_TYPE_PAGE;
273 	else
274 		pte |= ARM_LPAE_PTE_TYPE_BLOCK;
275 
276 	for (i = 0; i < num_entries; i++)
277 		ptep[i] = pte | paddr_to_iopte(paddr + i * sz, data);
278 
279 	if (!cfg->coherent_walk)
280 		__arm_lpae_sync_pte(ptep, num_entries, cfg);
281 }
282 
arm_lpae_init_pte(struct arm_lpae_io_pgtable * data,unsigned long iova,phys_addr_t paddr,arm_lpae_iopte prot,int lvl,int num_entries,arm_lpae_iopte * ptep)283 static int arm_lpae_init_pte(struct arm_lpae_io_pgtable *data,
284 			     unsigned long iova, phys_addr_t paddr,
285 			     arm_lpae_iopte prot, int lvl, int num_entries,
286 			     arm_lpae_iopte *ptep)
287 {
288 	int i;
289 
290 	for (i = 0; i < num_entries; i++)
291 		if (iopte_leaf(ptep[i], lvl, data->iop.fmt)) {
292 			/* We require an unmap first */
293 			WARN_ON(!selftest_running);
294 			return -EEXIST;
295 		} else if (iopte_type(ptep[i]) == ARM_LPAE_PTE_TYPE_TABLE) {
296 			/*
297 			 * We need to unmap and free the old table before
298 			 * overwriting it with a block entry.
299 			 */
300 			arm_lpae_iopte *tblp;
301 			size_t sz = ARM_LPAE_BLOCK_SIZE(lvl, data);
302 
303 			tblp = ptep - ARM_LPAE_LVL_IDX(iova, lvl, data);
304 			if (__arm_lpae_unmap(data, NULL, iova + i * sz, sz, 1,
305 					     lvl, tblp) != sz) {
306 				WARN_ON(1);
307 				return -EINVAL;
308 			}
309 		}
310 
311 	__arm_lpae_init_pte(data, paddr, prot, lvl, num_entries, ptep);
312 	return 0;
313 }
314 
arm_lpae_install_table(arm_lpae_iopte * table,arm_lpae_iopte * ptep,arm_lpae_iopte curr,struct arm_lpae_io_pgtable * data)315 static arm_lpae_iopte arm_lpae_install_table(arm_lpae_iopte *table,
316 					     arm_lpae_iopte *ptep,
317 					     arm_lpae_iopte curr,
318 					     struct arm_lpae_io_pgtable *data)
319 {
320 	arm_lpae_iopte old, new;
321 	struct io_pgtable_cfg *cfg = &data->iop.cfg;
322 
323 	new = paddr_to_iopte(__pa(table), data) | ARM_LPAE_PTE_TYPE_TABLE;
324 	if (cfg->quirks & IO_PGTABLE_QUIRK_ARM_NS)
325 		new |= ARM_LPAE_PTE_NSTABLE;
326 
327 	/*
328 	 * Ensure the table itself is visible before its PTE can be.
329 	 * Whilst we could get away with cmpxchg64_release below, this
330 	 * doesn't have any ordering semantics when !CONFIG_SMP.
331 	 */
332 	dma_wmb();
333 
334 	old = cmpxchg64_relaxed(ptep, curr, new);
335 
336 	if (cfg->coherent_walk || (old & ARM_LPAE_PTE_SW_SYNC))
337 		return old;
338 
339 	/* Even if it's not ours, there's no point waiting; just kick it */
340 	__arm_lpae_sync_pte(ptep, 1, cfg);
341 	if (old == curr)
342 		WRITE_ONCE(*ptep, new | ARM_LPAE_PTE_SW_SYNC);
343 
344 	return old;
345 }
346 
__arm_lpae_map(struct arm_lpae_io_pgtable * data,unsigned long iova,phys_addr_t paddr,size_t size,size_t pgcount,arm_lpae_iopte prot,int lvl,arm_lpae_iopte * ptep,gfp_t gfp,size_t * mapped)347 static int __arm_lpae_map(struct arm_lpae_io_pgtable *data, unsigned long iova,
348 			  phys_addr_t paddr, size_t size, size_t pgcount,
349 			  arm_lpae_iopte prot, int lvl, arm_lpae_iopte *ptep,
350 			  gfp_t gfp, size_t *mapped)
351 {
352 	arm_lpae_iopte *cptep, pte;
353 	size_t block_size = ARM_LPAE_BLOCK_SIZE(lvl, data);
354 	size_t tblsz = ARM_LPAE_GRANULE(data);
355 	struct io_pgtable_cfg *cfg = &data->iop.cfg;
356 	int ret = 0, num_entries, max_entries, map_idx_start;
357 
358 	/* Find our entry at the current level */
359 	map_idx_start = ARM_LPAE_LVL_IDX(iova, lvl, data);
360 	ptep += map_idx_start;
361 
362 	/* If we can install a leaf entry at this level, then do so */
363 	if (size == block_size) {
364 		max_entries = ARM_LPAE_PTES_PER_TABLE(data) - map_idx_start;
365 		num_entries = min_t(int, pgcount, max_entries);
366 		ret = arm_lpae_init_pte(data, iova, paddr, prot, lvl, num_entries, ptep);
367 		if (!ret && mapped)
368 			*mapped += num_entries * size;
369 
370 		return ret;
371 	}
372 
373 	/* We can't allocate tables at the final level */
374 	if (WARN_ON(lvl >= ARM_LPAE_MAX_LEVELS - 1))
375 		return -EINVAL;
376 
377 	/* Grab a pointer to the next level */
378 	pte = READ_ONCE(*ptep);
379 	if (!pte) {
380 		cptep = __arm_lpae_alloc_pages(tblsz, gfp, cfg);
381 		if (!cptep)
382 			return -ENOMEM;
383 
384 		pte = arm_lpae_install_table(cptep, ptep, 0, data);
385 		if (pte)
386 			__arm_lpae_free_pages(cptep, tblsz, cfg);
387 	} else if (!cfg->coherent_walk && !(pte & ARM_LPAE_PTE_SW_SYNC)) {
388 		__arm_lpae_sync_pte(ptep, 1, cfg);
389 	}
390 
391 	if (pte && !iopte_leaf(pte, lvl, data->iop.fmt)) {
392 		cptep = iopte_deref(pte, data);
393 	} else if (pte) {
394 		/* We require an unmap first */
395 		WARN_ON(!selftest_running);
396 		return -EEXIST;
397 	}
398 
399 	/* Rinse, repeat */
400 	return __arm_lpae_map(data, iova, paddr, size, pgcount, prot, lvl + 1,
401 			      cptep, gfp, mapped);
402 }
403 
arm_lpae_prot_to_pte(struct arm_lpae_io_pgtable * data,int prot)404 static arm_lpae_iopte arm_lpae_prot_to_pte(struct arm_lpae_io_pgtable *data,
405 					   int prot)
406 {
407 	arm_lpae_iopte pte;
408 
409 	if (data->iop.fmt == ARM_64_LPAE_S1 ||
410 	    data->iop.fmt == ARM_32_LPAE_S1) {
411 		pte = ARM_LPAE_PTE_nG;
412 		if (!(prot & IOMMU_WRITE) && (prot & IOMMU_READ))
413 			pte |= ARM_LPAE_PTE_AP_RDONLY;
414 		if (!(prot & IOMMU_PRIV))
415 			pte |= ARM_LPAE_PTE_AP_UNPRIV;
416 	} else {
417 		pte = ARM_LPAE_PTE_HAP_FAULT;
418 		if (prot & IOMMU_READ)
419 			pte |= ARM_LPAE_PTE_HAP_READ;
420 		if (prot & IOMMU_WRITE)
421 			pte |= ARM_LPAE_PTE_HAP_WRITE;
422 	}
423 
424 	/*
425 	 * Note that this logic is structured to accommodate Mali LPAE
426 	 * having stage-1-like attributes but stage-2-like permissions.
427 	 */
428 	if (data->iop.fmt == ARM_64_LPAE_S2 ||
429 	    data->iop.fmt == ARM_32_LPAE_S2) {
430 		if (prot & IOMMU_MMIO)
431 			pte |= ARM_LPAE_PTE_MEMATTR_DEV;
432 		else if (prot & IOMMU_CACHE)
433 			pte |= ARM_LPAE_PTE_MEMATTR_OIWB;
434 		else
435 			pte |= ARM_LPAE_PTE_MEMATTR_NC;
436 	} else {
437 		if (prot & IOMMU_MMIO)
438 			pte |= (ARM_LPAE_MAIR_ATTR_IDX_DEV
439 				<< ARM_LPAE_PTE_ATTRINDX_SHIFT);
440 		else if ((prot & IOMMU_CACHE) && (prot & IOMMU_SYS_CACHE_NWA))
441 			pte |= (ARM_LPAE_MAIR_ATTR_IDX_ICACHE_OCACHE_NWA
442 				<< ARM_LPAE_PTE_ATTRINDX_SHIFT);
443 		/* IOMMU_CACHE + IOMMU_SYS_CACHE equivalent to IOMMU_CACHE */
444 		else if (prot & IOMMU_CACHE)
445 			pte |= (ARM_LPAE_MAIR_ATTR_IDX_CACHE
446 				<< ARM_LPAE_PTE_ATTRINDX_SHIFT);
447 		else if (prot & IOMMU_SYS_CACHE)
448 			pte |= (ARM_LPAE_MAIR_ATTR_IDX_INC_OCACHE
449 				<< ARM_LPAE_PTE_ATTRINDX_SHIFT);
450 		else if (prot & IOMMU_SYS_CACHE_NWA)
451 			pte |= (ARM_LPAE_MAIR_ATTR_IDX_INC_OCACHE_NWA
452 				<< ARM_LPAE_PTE_ATTRINDX_SHIFT);
453 	}
454 
455 	/*
456 	 * Also Mali has its own notions of shareability wherein its Inner
457 	 * domain covers the cores within the GPU, and its Outer domain is
458 	 * "outside the GPU" (i.e. either the Inner or System domain in CPU
459 	 * terms, depending on coherency).
460 	 */
461 	if (prot & IOMMU_CACHE && data->iop.fmt != ARM_MALI_LPAE)
462 		pte |= ARM_LPAE_PTE_SH_IS;
463 	else
464 		pte |= ARM_LPAE_PTE_SH_OS;
465 
466 	if (prot & IOMMU_NOEXEC)
467 		pte |= ARM_LPAE_PTE_XN;
468 
469 	if (data->iop.cfg.quirks & IO_PGTABLE_QUIRK_ARM_NS)
470 		pte |= ARM_LPAE_PTE_NS;
471 
472 	if (data->iop.fmt != ARM_MALI_LPAE)
473 		pte |= ARM_LPAE_PTE_AF;
474 
475 	return pte;
476 }
477 
arm_lpae_map_pages(struct io_pgtable_ops * ops,unsigned long iova,phys_addr_t paddr,size_t pgsize,size_t pgcount,int iommu_prot,gfp_t gfp,size_t * mapped)478 static int arm_lpae_map_pages(struct io_pgtable_ops *ops, unsigned long iova,
479 			      phys_addr_t paddr, size_t pgsize, size_t pgcount,
480 			      int iommu_prot, gfp_t gfp, size_t *mapped)
481 {
482 	struct arm_lpae_io_pgtable *data = io_pgtable_ops_to_data(ops);
483 	struct io_pgtable_cfg *cfg = &data->iop.cfg;
484 	arm_lpae_iopte *ptep = data->pgd;
485 	int ret, lvl = data->start_level;
486 	arm_lpae_iopte prot;
487 	long iaext = (s64)iova >> cfg->ias;
488 
489 	if (WARN_ON(!pgsize || (pgsize & cfg->pgsize_bitmap) != pgsize))
490 		return -EINVAL;
491 
492 	if (cfg->quirks & IO_PGTABLE_QUIRK_ARM_TTBR1)
493 		iaext = ~iaext;
494 	if (WARN_ON(iaext || paddr >> cfg->oas))
495 		return -ERANGE;
496 
497 	/* If no access, then nothing to do */
498 	if (!(iommu_prot & (IOMMU_READ | IOMMU_WRITE)))
499 		return 0;
500 
501 	prot = arm_lpae_prot_to_pte(data, iommu_prot);
502 	ret = __arm_lpae_map(data, iova, paddr, pgsize, pgcount, prot, lvl,
503 			     ptep, gfp, mapped);
504 	/*
505 	 * Synchronise all PTE updates for the new mapping before there's
506 	 * a chance for anything to kick off a table walk for the new iova.
507 	 */
508 	wmb();
509 
510 	return ret;
511 }
512 
arm_lpae_map(struct io_pgtable_ops * ops,unsigned long iova,phys_addr_t paddr,size_t size,int iommu_prot,gfp_t gfp)513 static int arm_lpae_map(struct io_pgtable_ops *ops, unsigned long iova,
514 			phys_addr_t paddr, size_t size, int iommu_prot, gfp_t gfp)
515 {
516 	return arm_lpae_map_pages(ops, iova, paddr, size, 1, iommu_prot, gfp,
517 				  NULL);
518 }
519 
__arm_lpae_free_pgtable(struct arm_lpae_io_pgtable * data,int lvl,arm_lpae_iopte * ptep)520 static void __arm_lpae_free_pgtable(struct arm_lpae_io_pgtable *data, int lvl,
521 				    arm_lpae_iopte *ptep)
522 {
523 	arm_lpae_iopte *start, *end;
524 	unsigned long table_size;
525 
526 	if (lvl == data->start_level)
527 		table_size = ARM_LPAE_PGD_SIZE(data);
528 	else
529 		table_size = ARM_LPAE_GRANULE(data);
530 
531 	start = ptep;
532 
533 	/* Only leaf entries at the last level */
534 	if (lvl == ARM_LPAE_MAX_LEVELS - 1)
535 		end = ptep;
536 	else
537 		end = (void *)ptep + table_size;
538 
539 	while (ptep != end) {
540 		arm_lpae_iopte pte = *ptep++;
541 
542 		if (!pte || iopte_leaf(pte, lvl, data->iop.fmt))
543 			continue;
544 
545 		__arm_lpae_free_pgtable(data, lvl + 1, iopte_deref(pte, data));
546 	}
547 
548 	__arm_lpae_free_pages(start, table_size, &data->iop.cfg);
549 }
550 
arm_lpae_free_pgtable(struct io_pgtable * iop)551 static void arm_lpae_free_pgtable(struct io_pgtable *iop)
552 {
553 	struct arm_lpae_io_pgtable *data = io_pgtable_to_data(iop);
554 
555 	__arm_lpae_free_pgtable(data, data->start_level, data->pgd);
556 	kfree(data);
557 }
558 
arm_lpae_split_blk_unmap(struct arm_lpae_io_pgtable * data,struct iommu_iotlb_gather * gather,unsigned long iova,size_t size,arm_lpae_iopte blk_pte,int lvl,arm_lpae_iopte * ptep,size_t pgcount)559 static size_t arm_lpae_split_blk_unmap(struct arm_lpae_io_pgtable *data,
560 				       struct iommu_iotlb_gather *gather,
561 				       unsigned long iova, size_t size,
562 				       arm_lpae_iopte blk_pte, int lvl,
563 				       arm_lpae_iopte *ptep, size_t pgcount)
564 {
565 	struct io_pgtable_cfg *cfg = &data->iop.cfg;
566 	arm_lpae_iopte pte, *tablep;
567 	phys_addr_t blk_paddr;
568 	size_t tablesz = ARM_LPAE_GRANULE(data);
569 	size_t split_sz = ARM_LPAE_BLOCK_SIZE(lvl, data);
570 	int ptes_per_table = ARM_LPAE_PTES_PER_TABLE(data);
571 	int i, unmap_idx_start = -1, num_entries = 0, max_entries;
572 
573 	if (WARN_ON(lvl == ARM_LPAE_MAX_LEVELS))
574 		return 0;
575 
576 	tablep = __arm_lpae_alloc_pages(tablesz, GFP_ATOMIC, cfg);
577 	if (!tablep)
578 		return 0; /* Bytes unmapped */
579 
580 	if (size == split_sz) {
581 		unmap_idx_start = ARM_LPAE_LVL_IDX(iova, lvl, data);
582 		max_entries = ptes_per_table - unmap_idx_start;
583 		num_entries = min_t(int, pgcount, max_entries);
584 	}
585 
586 	blk_paddr = iopte_to_paddr(blk_pte, data);
587 	pte = iopte_prot(blk_pte);
588 
589 	for (i = 0; i < ptes_per_table; i++, blk_paddr += split_sz) {
590 		/* Unmap! */
591 		if (i >= unmap_idx_start && i < (unmap_idx_start + num_entries))
592 			continue;
593 
594 		__arm_lpae_init_pte(data, blk_paddr, pte, lvl, 1, &tablep[i]);
595 	}
596 
597 	pte = arm_lpae_install_table(tablep, ptep, blk_pte, data);
598 	if (pte != blk_pte) {
599 		__arm_lpae_free_pages(tablep, tablesz, cfg);
600 		/*
601 		 * We may race against someone unmapping another part of this
602 		 * block, but anything else is invalid. We can't misinterpret
603 		 * a page entry here since we're never at the last level.
604 		 */
605 		if (iopte_type(pte) != ARM_LPAE_PTE_TYPE_TABLE)
606 			return 0;
607 
608 		tablep = iopte_deref(pte, data);
609 	} else if (unmap_idx_start >= 0) {
610 		for (i = 0; i < num_entries; i++)
611 			io_pgtable_tlb_add_page(&data->iop, gather, iova + i * size, size);
612 
613 		return num_entries * size;
614 	}
615 
616 	return __arm_lpae_unmap(data, gather, iova, size, pgcount, lvl, tablep);
617 }
618 
__arm_lpae_unmap(struct arm_lpae_io_pgtable * data,struct iommu_iotlb_gather * gather,unsigned long iova,size_t size,size_t pgcount,int lvl,arm_lpae_iopte * ptep)619 static size_t __arm_lpae_unmap(struct arm_lpae_io_pgtable *data,
620 			       struct iommu_iotlb_gather *gather,
621 			       unsigned long iova, size_t size, size_t pgcount,
622 			       int lvl, arm_lpae_iopte *ptep)
623 {
624 	arm_lpae_iopte pte;
625 	struct io_pgtable *iop = &data->iop;
626 	int i = 0, num_entries, max_entries, unmap_idx_start;
627 
628 	/* Something went horribly wrong and we ran out of page table */
629 	if (WARN_ON(lvl == ARM_LPAE_MAX_LEVELS))
630 		return 0;
631 
632 	unmap_idx_start = ARM_LPAE_LVL_IDX(iova, lvl, data);
633 	ptep += unmap_idx_start;
634 	pte = READ_ONCE(*ptep);
635 	if (WARN_ON(!pte))
636 		return 0;
637 
638 	/* If the size matches this level, we're in the right place */
639 	if (size == ARM_LPAE_BLOCK_SIZE(lvl, data)) {
640 		max_entries = ARM_LPAE_PTES_PER_TABLE(data) - unmap_idx_start;
641 		num_entries = min_t(int, pgcount, max_entries);
642 
643 		while (i < num_entries) {
644 			pte = READ_ONCE(*ptep);
645 			if (WARN_ON(!pte))
646 				break;
647 
648 			__arm_lpae_clear_pte(ptep, &iop->cfg);
649 
650 			if (!iopte_leaf(pte, lvl, iop->fmt)) {
651 				/* Also flush any partial walks */
652 				io_pgtable_tlb_flush_walk(iop, iova + i * size, size,
653 							  ARM_LPAE_GRANULE(data));
654 				__arm_lpae_free_pgtable(data, lvl + 1, iopte_deref(pte, data));
655 			} else if (!iommu_iotlb_gather_queued(gather)) {
656 				io_pgtable_tlb_add_page(iop, gather, iova + i * size, size);
657 			}
658 
659 			ptep++;
660 			i++;
661 		}
662 
663 		return i * size;
664 	} else if (iopte_leaf(pte, lvl, iop->fmt)) {
665 		/*
666 		 * Insert a table at the next level to map the old region,
667 		 * minus the part we want to unmap
668 		 */
669 		return arm_lpae_split_blk_unmap(data, gather, iova, size, pte,
670 						lvl + 1, ptep, pgcount);
671 	}
672 
673 	/* Keep on walkin' */
674 	ptep = iopte_deref(pte, data);
675 	return __arm_lpae_unmap(data, gather, iova, size, pgcount, lvl + 1, ptep);
676 }
677 
arm_lpae_unmap_pages(struct io_pgtable_ops * ops,unsigned long iova,size_t pgsize,size_t pgcount,struct iommu_iotlb_gather * gather)678 static size_t arm_lpae_unmap_pages(struct io_pgtable_ops *ops, unsigned long iova,
679 				   size_t pgsize, size_t pgcount,
680 				   struct iommu_iotlb_gather *gather)
681 {
682 	struct arm_lpae_io_pgtable *data = io_pgtable_ops_to_data(ops);
683 	struct io_pgtable_cfg *cfg = &data->iop.cfg;
684 	arm_lpae_iopte *ptep = data->pgd;
685 	long iaext = (s64)iova >> cfg->ias;
686 
687 	if (WARN_ON(!pgsize || (pgsize & cfg->pgsize_bitmap) != pgsize || !pgcount))
688 		return 0;
689 
690 	if (cfg->quirks & IO_PGTABLE_QUIRK_ARM_TTBR1)
691 		iaext = ~iaext;
692 	if (WARN_ON(iaext))
693 		return 0;
694 
695 	return __arm_lpae_unmap(data, gather, iova, pgsize, pgcount,
696 				data->start_level, ptep);
697 }
698 
arm_lpae_unmap(struct io_pgtable_ops * ops,unsigned long iova,size_t size,struct iommu_iotlb_gather * gather)699 static size_t arm_lpae_unmap(struct io_pgtable_ops *ops, unsigned long iova,
700 			     size_t size, struct iommu_iotlb_gather *gather)
701 {
702 	return arm_lpae_unmap_pages(ops, iova, size, 1, gather);
703 }
704 
arm_lpae_iova_to_phys(struct io_pgtable_ops * ops,unsigned long iova)705 static phys_addr_t arm_lpae_iova_to_phys(struct io_pgtable_ops *ops,
706 					 unsigned long iova)
707 {
708 	struct arm_lpae_io_pgtable *data = io_pgtable_ops_to_data(ops);
709 	arm_lpae_iopte pte, *ptep = data->pgd;
710 	int lvl = data->start_level;
711 
712 	do {
713 		/* Valid IOPTE pointer? */
714 		if (!ptep)
715 			return 0;
716 
717 		/* Grab the IOPTE we're interested in */
718 		ptep += ARM_LPAE_LVL_IDX(iova, lvl, data);
719 		pte = READ_ONCE(*ptep);
720 
721 		/* Valid entry? */
722 		if (!pte)
723 			return 0;
724 
725 		/* Leaf entry? */
726 		if (iopte_leaf(pte, lvl, data->iop.fmt))
727 			goto found_translation;
728 
729 		/* Take it to the next level */
730 		ptep = iopte_deref(pte, data);
731 	} while (++lvl < ARM_LPAE_MAX_LEVELS);
732 
733 	/* Ran out of page tables to walk */
734 	return 0;
735 
736 found_translation:
737 	iova &= (ARM_LPAE_BLOCK_SIZE(lvl, data) - 1);
738 	return iopte_to_paddr(pte, data) | iova;
739 }
740 
arm_lpae_restrict_pgsizes(struct io_pgtable_cfg * cfg)741 static void arm_lpae_restrict_pgsizes(struct io_pgtable_cfg *cfg)
742 {
743 	unsigned long granule, page_sizes;
744 	unsigned int max_addr_bits = 48;
745 
746 	/*
747 	 * We need to restrict the supported page sizes to match the
748 	 * translation regime for a particular granule. Aim to match
749 	 * the CPU page size if possible, otherwise prefer smaller sizes.
750 	 * While we're at it, restrict the block sizes to match the
751 	 * chosen granule.
752 	 */
753 	if (cfg->pgsize_bitmap & PAGE_SIZE)
754 		granule = PAGE_SIZE;
755 	else if (cfg->pgsize_bitmap & ~PAGE_MASK)
756 		granule = 1UL << __fls(cfg->pgsize_bitmap & ~PAGE_MASK);
757 	else if (cfg->pgsize_bitmap & PAGE_MASK)
758 		granule = 1UL << __ffs(cfg->pgsize_bitmap & PAGE_MASK);
759 	else
760 		granule = 0;
761 
762 	switch (granule) {
763 	case SZ_4K:
764 		page_sizes = (SZ_4K | SZ_2M | SZ_1G);
765 		break;
766 	case SZ_16K:
767 		page_sizes = (SZ_16K | SZ_32M);
768 		break;
769 	case SZ_64K:
770 		max_addr_bits = 52;
771 		page_sizes = (SZ_64K | SZ_512M);
772 		if (cfg->oas > 48)
773 			page_sizes |= 1ULL << 42; /* 4TB */
774 		break;
775 	default:
776 		page_sizes = 0;
777 	}
778 
779 	cfg->pgsize_bitmap &= page_sizes;
780 	cfg->ias = min(cfg->ias, max_addr_bits);
781 	cfg->oas = min(cfg->oas, max_addr_bits);
782 }
783 
784 static struct arm_lpae_io_pgtable *
arm_lpae_alloc_pgtable(struct io_pgtable_cfg * cfg)785 arm_lpae_alloc_pgtable(struct io_pgtable_cfg *cfg)
786 {
787 	struct arm_lpae_io_pgtable *data;
788 	int levels, va_bits, pg_shift;
789 
790 	arm_lpae_restrict_pgsizes(cfg);
791 
792 	if (!(cfg->pgsize_bitmap & (SZ_4K | SZ_16K | SZ_64K)))
793 		return NULL;
794 
795 	if (cfg->ias > ARM_LPAE_MAX_ADDR_BITS)
796 		return NULL;
797 
798 	if (cfg->oas > ARM_LPAE_MAX_ADDR_BITS)
799 		return NULL;
800 
801 	data = kmalloc(sizeof(*data), GFP_KERNEL);
802 	if (!data)
803 		return NULL;
804 
805 	pg_shift = __ffs(cfg->pgsize_bitmap);
806 	data->bits_per_level = pg_shift - ilog2(sizeof(arm_lpae_iopte));
807 
808 	va_bits = cfg->ias - pg_shift;
809 	levels = DIV_ROUND_UP(va_bits, data->bits_per_level);
810 	data->start_level = ARM_LPAE_MAX_LEVELS - levels;
811 
812 	/* Calculate the actual size of our pgd (without concatenation) */
813 	data->pgd_bits = va_bits - (data->bits_per_level * (levels - 1));
814 
815 	data->iop.ops = (struct io_pgtable_ops) {
816 		.map		= arm_lpae_map,
817 		.map_pages	= arm_lpae_map_pages,
818 		.unmap		= arm_lpae_unmap,
819 		.unmap_pages	= arm_lpae_unmap_pages,
820 		.iova_to_phys	= arm_lpae_iova_to_phys,
821 	};
822 
823 	return data;
824 }
825 
826 static struct io_pgtable *
arm_64_lpae_alloc_pgtable_s1(struct io_pgtable_cfg * cfg,void * cookie)827 arm_64_lpae_alloc_pgtable_s1(struct io_pgtable_cfg *cfg, void *cookie)
828 {
829 	u64 reg;
830 	struct arm_lpae_io_pgtable *data;
831 	typeof(&cfg->arm_lpae_s1_cfg.tcr) tcr = &cfg->arm_lpae_s1_cfg.tcr;
832 	bool tg1;
833 
834 	if (cfg->quirks & ~(IO_PGTABLE_QUIRK_ARM_NS |
835 			    IO_PGTABLE_QUIRK_ARM_TTBR1 |
836 			    IO_PGTABLE_QUIRK_ARM_OUTER_WBWA))
837 		return NULL;
838 
839 	data = arm_lpae_alloc_pgtable(cfg);
840 	if (!data)
841 		return NULL;
842 
843 	/* TCR */
844 	if (cfg->coherent_walk) {
845 		tcr->sh = ARM_LPAE_TCR_SH_IS;
846 		tcr->irgn = ARM_LPAE_TCR_RGN_WBWA;
847 		tcr->orgn = ARM_LPAE_TCR_RGN_WBWA;
848 		if (cfg->quirks & IO_PGTABLE_QUIRK_ARM_OUTER_WBWA)
849 			goto out_free_data;
850 	} else {
851 		tcr->sh = ARM_LPAE_TCR_SH_OS;
852 		tcr->irgn = ARM_LPAE_TCR_RGN_NC;
853 		if (!(cfg->quirks & IO_PGTABLE_QUIRK_ARM_OUTER_WBWA))
854 			tcr->orgn = ARM_LPAE_TCR_RGN_NC;
855 		else
856 			tcr->orgn = ARM_LPAE_TCR_RGN_WBWA;
857 	}
858 
859 	tg1 = cfg->quirks & IO_PGTABLE_QUIRK_ARM_TTBR1;
860 	switch (ARM_LPAE_GRANULE(data)) {
861 	case SZ_4K:
862 		tcr->tg = tg1 ? ARM_LPAE_TCR_TG1_4K : ARM_LPAE_TCR_TG0_4K;
863 		break;
864 	case SZ_16K:
865 		tcr->tg = tg1 ? ARM_LPAE_TCR_TG1_16K : ARM_LPAE_TCR_TG0_16K;
866 		break;
867 	case SZ_64K:
868 		tcr->tg = tg1 ? ARM_LPAE_TCR_TG1_64K : ARM_LPAE_TCR_TG0_64K;
869 		break;
870 	}
871 
872 	switch (cfg->oas) {
873 	case 32:
874 		tcr->ips = ARM_LPAE_TCR_PS_32_BIT;
875 		break;
876 	case 36:
877 		tcr->ips = ARM_LPAE_TCR_PS_36_BIT;
878 		break;
879 	case 40:
880 		tcr->ips = ARM_LPAE_TCR_PS_40_BIT;
881 		break;
882 	case 42:
883 		tcr->ips = ARM_LPAE_TCR_PS_42_BIT;
884 		break;
885 	case 44:
886 		tcr->ips = ARM_LPAE_TCR_PS_44_BIT;
887 		break;
888 	case 48:
889 		tcr->ips = ARM_LPAE_TCR_PS_48_BIT;
890 		break;
891 	case 52:
892 		tcr->ips = ARM_LPAE_TCR_PS_52_BIT;
893 		break;
894 	default:
895 		goto out_free_data;
896 	}
897 
898 	tcr->tsz = 64ULL - cfg->ias;
899 
900 	/* MAIRs */
901 	reg = (ARM_LPAE_MAIR_ATTR_NC
902 	       << ARM_LPAE_MAIR_ATTR_SHIFT(ARM_LPAE_MAIR_ATTR_IDX_NC)) |
903 	      (ARM_LPAE_MAIR_ATTR_WBRWA
904 	       << ARM_LPAE_MAIR_ATTR_SHIFT(ARM_LPAE_MAIR_ATTR_IDX_CACHE)) |
905 	      (ARM_LPAE_MAIR_ATTR_DEVICE
906 	       << ARM_LPAE_MAIR_ATTR_SHIFT(ARM_LPAE_MAIR_ATTR_IDX_DEV)) |
907 	      (ARM_LPAE_MAIR_ATTR_INC_OWBRWA
908 	       << ARM_LPAE_MAIR_ATTR_SHIFT(ARM_LPAE_MAIR_ATTR_IDX_INC_OCACHE)) |
909 	      (ARM_LPAE_MAIR_ATTR_INC_OWBRANWA
910 	       << ARM_LPAE_MAIR_ATTR_SHIFT(ARM_LPAE_MAIR_ATTR_IDX_INC_OCACHE_NWA)) |
911 	      (ARM_LPAE_MAIR_ATTR_IWBRWA_OWBRANWA
912 	       << ARM_LPAE_MAIR_ATTR_SHIFT(ARM_LPAE_MAIR_ATTR_IDX_ICACHE_OCACHE_NWA));
913 
914 	cfg->arm_lpae_s1_cfg.mair = reg;
915 
916 	/* Looking good; allocate a pgd */
917 	data->pgd = __arm_lpae_alloc_pages(ARM_LPAE_PGD_SIZE(data),
918 					   GFP_KERNEL, cfg);
919 	if (!data->pgd)
920 		goto out_free_data;
921 
922 	/* Ensure the empty pgd is visible before any actual TTBR write */
923 	wmb();
924 
925 	/* TTBR */
926 	cfg->arm_lpae_s1_cfg.ttbr = virt_to_phys(data->pgd);
927 	return &data->iop;
928 
929 out_free_data:
930 	kfree(data);
931 	return NULL;
932 }
933 
934 static struct io_pgtable *
arm_64_lpae_alloc_pgtable_s2(struct io_pgtable_cfg * cfg,void * cookie)935 arm_64_lpae_alloc_pgtable_s2(struct io_pgtable_cfg *cfg, void *cookie)
936 {
937 	u64 sl;
938 	struct arm_lpae_io_pgtable *data;
939 	typeof(&cfg->arm_lpae_s2_cfg.vtcr) vtcr = &cfg->arm_lpae_s2_cfg.vtcr;
940 
941 	/* The NS quirk doesn't apply at stage 2 */
942 	if (cfg->quirks)
943 		return NULL;
944 
945 	data = arm_lpae_alloc_pgtable(cfg);
946 	if (!data)
947 		return NULL;
948 
949 	/*
950 	 * Concatenate PGDs at level 1 if possible in order to reduce
951 	 * the depth of the stage-2 walk.
952 	 */
953 	if (data->start_level == 0) {
954 		unsigned long pgd_pages;
955 
956 		pgd_pages = ARM_LPAE_PGD_SIZE(data) / sizeof(arm_lpae_iopte);
957 		if (pgd_pages <= ARM_LPAE_S2_MAX_CONCAT_PAGES) {
958 			data->pgd_bits += data->bits_per_level;
959 			data->start_level++;
960 		}
961 	}
962 
963 	/* VTCR */
964 	if (cfg->coherent_walk) {
965 		vtcr->sh = ARM_LPAE_TCR_SH_IS;
966 		vtcr->irgn = ARM_LPAE_TCR_RGN_WBWA;
967 		vtcr->orgn = ARM_LPAE_TCR_RGN_WBWA;
968 	} else {
969 		vtcr->sh = ARM_LPAE_TCR_SH_OS;
970 		vtcr->irgn = ARM_LPAE_TCR_RGN_NC;
971 		vtcr->orgn = ARM_LPAE_TCR_RGN_NC;
972 	}
973 
974 	sl = data->start_level;
975 
976 	switch (ARM_LPAE_GRANULE(data)) {
977 	case SZ_4K:
978 		vtcr->tg = ARM_LPAE_TCR_TG0_4K;
979 		sl++; /* SL0 format is different for 4K granule size */
980 		break;
981 	case SZ_16K:
982 		vtcr->tg = ARM_LPAE_TCR_TG0_16K;
983 		break;
984 	case SZ_64K:
985 		vtcr->tg = ARM_LPAE_TCR_TG0_64K;
986 		break;
987 	}
988 
989 	switch (cfg->oas) {
990 	case 32:
991 		vtcr->ps = ARM_LPAE_TCR_PS_32_BIT;
992 		break;
993 	case 36:
994 		vtcr->ps = ARM_LPAE_TCR_PS_36_BIT;
995 		break;
996 	case 40:
997 		vtcr->ps = ARM_LPAE_TCR_PS_40_BIT;
998 		break;
999 	case 42:
1000 		vtcr->ps = ARM_LPAE_TCR_PS_42_BIT;
1001 		break;
1002 	case 44:
1003 		vtcr->ps = ARM_LPAE_TCR_PS_44_BIT;
1004 		break;
1005 	case 48:
1006 		vtcr->ps = ARM_LPAE_TCR_PS_48_BIT;
1007 		break;
1008 	case 52:
1009 		vtcr->ps = ARM_LPAE_TCR_PS_52_BIT;
1010 		break;
1011 	default:
1012 		goto out_free_data;
1013 	}
1014 
1015 	vtcr->tsz = 64ULL - cfg->ias;
1016 	vtcr->sl = ~sl & ARM_LPAE_VTCR_SL0_MASK;
1017 
1018 	/* Allocate pgd pages */
1019 	data->pgd = __arm_lpae_alloc_pages(ARM_LPAE_PGD_SIZE(data),
1020 					   GFP_KERNEL, cfg);
1021 	if (!data->pgd)
1022 		goto out_free_data;
1023 
1024 	/* Ensure the empty pgd is visible before any actual TTBR write */
1025 	wmb();
1026 
1027 	/* VTTBR */
1028 	cfg->arm_lpae_s2_cfg.vttbr = virt_to_phys(data->pgd);
1029 	return &data->iop;
1030 
1031 out_free_data:
1032 	kfree(data);
1033 	return NULL;
1034 }
1035 
1036 static struct io_pgtable *
arm_32_lpae_alloc_pgtable_s1(struct io_pgtable_cfg * cfg,void * cookie)1037 arm_32_lpae_alloc_pgtable_s1(struct io_pgtable_cfg *cfg, void *cookie)
1038 {
1039 	if (cfg->ias > 32 || cfg->oas > 40)
1040 		return NULL;
1041 
1042 	cfg->pgsize_bitmap &= (SZ_4K | SZ_2M | SZ_1G);
1043 	return arm_64_lpae_alloc_pgtable_s1(cfg, cookie);
1044 }
1045 
1046 static struct io_pgtable *
arm_32_lpae_alloc_pgtable_s2(struct io_pgtable_cfg * cfg,void * cookie)1047 arm_32_lpae_alloc_pgtable_s2(struct io_pgtable_cfg *cfg, void *cookie)
1048 {
1049 	if (cfg->ias > 40 || cfg->oas > 40)
1050 		return NULL;
1051 
1052 	cfg->pgsize_bitmap &= (SZ_4K | SZ_2M | SZ_1G);
1053 	return arm_64_lpae_alloc_pgtable_s2(cfg, cookie);
1054 }
1055 
1056 static struct io_pgtable *
arm_mali_lpae_alloc_pgtable(struct io_pgtable_cfg * cfg,void * cookie)1057 arm_mali_lpae_alloc_pgtable(struct io_pgtable_cfg *cfg, void *cookie)
1058 {
1059 	struct arm_lpae_io_pgtable *data;
1060 
1061 	/* No quirks for Mali (hopefully) */
1062 	if (cfg->quirks)
1063 		return NULL;
1064 
1065 	if (cfg->ias > 48 || cfg->oas > 40)
1066 		return NULL;
1067 
1068 	cfg->pgsize_bitmap &= (SZ_4K | SZ_2M | SZ_1G);
1069 
1070 	data = arm_lpae_alloc_pgtable(cfg);
1071 	if (!data)
1072 		return NULL;
1073 
1074 	/* Mali seems to need a full 4-level table regardless of IAS */
1075 	if (data->start_level > 0) {
1076 		data->start_level = 0;
1077 		data->pgd_bits = 0;
1078 	}
1079 	/*
1080 	 * MEMATTR: Mali has no actual notion of a non-cacheable type, so the
1081 	 * best we can do is mimic the out-of-tree driver and hope that the
1082 	 * "implementation-defined caching policy" is good enough. Similarly,
1083 	 * we'll use it for the sake of a valid attribute for our 'device'
1084 	 * index, although callers should never request that in practice.
1085 	 */
1086 	cfg->arm_mali_lpae_cfg.memattr =
1087 		(ARM_MALI_LPAE_MEMATTR_IMP_DEF
1088 		 << ARM_LPAE_MAIR_ATTR_SHIFT(ARM_LPAE_MAIR_ATTR_IDX_NC)) |
1089 		(ARM_MALI_LPAE_MEMATTR_WRITE_ALLOC
1090 		 << ARM_LPAE_MAIR_ATTR_SHIFT(ARM_LPAE_MAIR_ATTR_IDX_CACHE)) |
1091 		(ARM_MALI_LPAE_MEMATTR_IMP_DEF
1092 		 << ARM_LPAE_MAIR_ATTR_SHIFT(ARM_LPAE_MAIR_ATTR_IDX_DEV));
1093 
1094 	data->pgd = __arm_lpae_alloc_pages(ARM_LPAE_PGD_SIZE(data), GFP_KERNEL,
1095 					   cfg);
1096 	if (!data->pgd)
1097 		goto out_free_data;
1098 
1099 	/* Ensure the empty pgd is visible before TRANSTAB can be written */
1100 	wmb();
1101 
1102 	cfg->arm_mali_lpae_cfg.transtab = virt_to_phys(data->pgd) |
1103 					  ARM_MALI_LPAE_TTBR_READ_INNER |
1104 					  ARM_MALI_LPAE_TTBR_ADRMODE_TABLE;
1105 	if (cfg->coherent_walk)
1106 		cfg->arm_mali_lpae_cfg.transtab |= ARM_MALI_LPAE_TTBR_SHARE_OUTER;
1107 
1108 	return &data->iop;
1109 
1110 out_free_data:
1111 	kfree(data);
1112 	return NULL;
1113 }
1114 
1115 struct io_pgtable_init_fns io_pgtable_arm_64_lpae_s1_init_fns = {
1116 	.alloc	= arm_64_lpae_alloc_pgtable_s1,
1117 	.free	= arm_lpae_free_pgtable,
1118 };
1119 
1120 struct io_pgtable_init_fns io_pgtable_arm_64_lpae_s2_init_fns = {
1121 	.alloc	= arm_64_lpae_alloc_pgtable_s2,
1122 	.free	= arm_lpae_free_pgtable,
1123 };
1124 
1125 struct io_pgtable_init_fns io_pgtable_arm_32_lpae_s1_init_fns = {
1126 	.alloc	= arm_32_lpae_alloc_pgtable_s1,
1127 	.free	= arm_lpae_free_pgtable,
1128 };
1129 
1130 struct io_pgtable_init_fns io_pgtable_arm_32_lpae_s2_init_fns = {
1131 	.alloc	= arm_32_lpae_alloc_pgtable_s2,
1132 	.free	= arm_lpae_free_pgtable,
1133 };
1134 
1135 struct io_pgtable_init_fns io_pgtable_arm_mali_lpae_init_fns = {
1136 	.alloc	= arm_mali_lpae_alloc_pgtable,
1137 	.free	= arm_lpae_free_pgtable,
1138 };
1139 
1140 #ifdef CONFIG_IOMMU_IO_PGTABLE_LPAE_SELFTEST
1141 
1142 static struct io_pgtable_cfg *cfg_cookie __initdata;
1143 
dummy_tlb_flush_all(void * cookie)1144 static void __init dummy_tlb_flush_all(void *cookie)
1145 {
1146 	WARN_ON(cookie != cfg_cookie);
1147 }
1148 
dummy_tlb_flush(unsigned long iova,size_t size,size_t granule,void * cookie)1149 static void __init dummy_tlb_flush(unsigned long iova, size_t size,
1150 				   size_t granule, void *cookie)
1151 {
1152 	WARN_ON(cookie != cfg_cookie);
1153 	WARN_ON(!(size & cfg_cookie->pgsize_bitmap));
1154 }
1155 
dummy_tlb_add_page(struct iommu_iotlb_gather * gather,unsigned long iova,size_t granule,void * cookie)1156 static void __init dummy_tlb_add_page(struct iommu_iotlb_gather *gather,
1157 				      unsigned long iova, size_t granule,
1158 				      void *cookie)
1159 {
1160 	dummy_tlb_flush(iova, granule, granule, cookie);
1161 }
1162 
1163 static const struct iommu_flush_ops dummy_tlb_ops __initconst = {
1164 	.tlb_flush_all	= dummy_tlb_flush_all,
1165 	.tlb_flush_walk	= dummy_tlb_flush,
1166 	.tlb_add_page	= dummy_tlb_add_page,
1167 };
1168 
arm_lpae_dump_ops(struct io_pgtable_ops * ops)1169 static void __init arm_lpae_dump_ops(struct io_pgtable_ops *ops)
1170 {
1171 	struct arm_lpae_io_pgtable *data = io_pgtable_ops_to_data(ops);
1172 	struct io_pgtable_cfg *cfg = &data->iop.cfg;
1173 
1174 	pr_err("cfg: pgsize_bitmap 0x%lx, ias %u-bit\n",
1175 		cfg->pgsize_bitmap, cfg->ias);
1176 	pr_err("data: %d levels, 0x%zx pgd_size, %u pg_shift, %u bits_per_level, pgd @ %p\n",
1177 		ARM_LPAE_MAX_LEVELS - data->start_level, ARM_LPAE_PGD_SIZE(data),
1178 		ilog2(ARM_LPAE_GRANULE(data)), data->bits_per_level, data->pgd);
1179 }
1180 
1181 #define __FAIL(ops, i)	({						\
1182 		WARN(1, "selftest: test failed for fmt idx %d\n", (i));	\
1183 		arm_lpae_dump_ops(ops);					\
1184 		selftest_running = false;				\
1185 		-EFAULT;						\
1186 })
1187 
arm_lpae_run_tests(struct io_pgtable_cfg * cfg)1188 static int __init arm_lpae_run_tests(struct io_pgtable_cfg *cfg)
1189 {
1190 	static const enum io_pgtable_fmt fmts[] __initconst = {
1191 		ARM_64_LPAE_S1,
1192 		ARM_64_LPAE_S2,
1193 	};
1194 
1195 	int i, j;
1196 	unsigned long iova;
1197 	size_t size;
1198 	struct io_pgtable_ops *ops;
1199 
1200 	selftest_running = true;
1201 
1202 	for (i = 0; i < ARRAY_SIZE(fmts); ++i) {
1203 		cfg_cookie = cfg;
1204 		ops = alloc_io_pgtable_ops(fmts[i], cfg, cfg);
1205 		if (!ops) {
1206 			pr_err("selftest: failed to allocate io pgtable ops\n");
1207 			return -ENOMEM;
1208 		}
1209 
1210 		/*
1211 		 * Initial sanity checks.
1212 		 * Empty page tables shouldn't provide any translations.
1213 		 */
1214 		if (ops->iova_to_phys(ops, 42))
1215 			return __FAIL(ops, i);
1216 
1217 		if (ops->iova_to_phys(ops, SZ_1G + 42))
1218 			return __FAIL(ops, i);
1219 
1220 		if (ops->iova_to_phys(ops, SZ_2G + 42))
1221 			return __FAIL(ops, i);
1222 
1223 		/*
1224 		 * Distinct mappings of different granule sizes.
1225 		 */
1226 		iova = 0;
1227 		for_each_set_bit(j, &cfg->pgsize_bitmap, BITS_PER_LONG) {
1228 			size = 1UL << j;
1229 
1230 			if (ops->map(ops, iova, iova, size, IOMMU_READ |
1231 							    IOMMU_WRITE |
1232 							    IOMMU_NOEXEC |
1233 							    IOMMU_CACHE, GFP_KERNEL))
1234 				return __FAIL(ops, i);
1235 
1236 			/* Overlapping mappings */
1237 			if (!ops->map(ops, iova, iova + size, size,
1238 				      IOMMU_READ | IOMMU_NOEXEC, GFP_KERNEL))
1239 				return __FAIL(ops, i);
1240 
1241 			if (ops->iova_to_phys(ops, iova + 42) != (iova + 42))
1242 				return __FAIL(ops, i);
1243 
1244 			iova += SZ_1G;
1245 		}
1246 
1247 		/* Partial unmap */
1248 		size = 1UL << __ffs(cfg->pgsize_bitmap);
1249 		if (ops->unmap(ops, SZ_1G + size, size, NULL) != size)
1250 			return __FAIL(ops, i);
1251 
1252 		/* Remap of partial unmap */
1253 		if (ops->map(ops, SZ_1G + size, size, size, IOMMU_READ, GFP_KERNEL))
1254 			return __FAIL(ops, i);
1255 
1256 		if (ops->iova_to_phys(ops, SZ_1G + size + 42) != (size + 42))
1257 			return __FAIL(ops, i);
1258 
1259 		/* Full unmap */
1260 		iova = 0;
1261 		for_each_set_bit(j, &cfg->pgsize_bitmap, BITS_PER_LONG) {
1262 			size = 1UL << j;
1263 
1264 			if (ops->unmap(ops, iova, size, NULL) != size)
1265 				return __FAIL(ops, i);
1266 
1267 			if (ops->iova_to_phys(ops, iova + 42))
1268 				return __FAIL(ops, i);
1269 
1270 			/* Remap full block */
1271 			if (ops->map(ops, iova, iova, size, IOMMU_WRITE, GFP_KERNEL))
1272 				return __FAIL(ops, i);
1273 
1274 			if (ops->iova_to_phys(ops, iova + 42) != (iova + 42))
1275 				return __FAIL(ops, i);
1276 
1277 			iova += SZ_1G;
1278 		}
1279 
1280 		free_io_pgtable_ops(ops);
1281 	}
1282 
1283 	selftest_running = false;
1284 	return 0;
1285 }
1286 
arm_lpae_do_selftests(void)1287 static int __init arm_lpae_do_selftests(void)
1288 {
1289 	static const unsigned long pgsize[] __initconst = {
1290 		SZ_4K | SZ_2M | SZ_1G,
1291 		SZ_16K | SZ_32M,
1292 		SZ_64K | SZ_512M,
1293 	};
1294 
1295 	static const unsigned int ias[] __initconst = {
1296 		32, 36, 40, 42, 44, 48,
1297 	};
1298 
1299 	int i, j, pass = 0, fail = 0;
1300 	struct device dev;
1301 	struct io_pgtable_cfg cfg = {
1302 		.tlb = &dummy_tlb_ops,
1303 		.oas = 48,
1304 		.coherent_walk = true,
1305 		.iommu_dev = &dev,
1306 	};
1307 
1308 	/* __arm_lpae_alloc_pages() merely needs dev_to_node() to work */
1309 	set_dev_node(&dev, NUMA_NO_NODE);
1310 
1311 	for (i = 0; i < ARRAY_SIZE(pgsize); ++i) {
1312 		for (j = 0; j < ARRAY_SIZE(ias); ++j) {
1313 			cfg.pgsize_bitmap = pgsize[i];
1314 			cfg.ias = ias[j];
1315 			pr_info("selftest: pgsize_bitmap 0x%08lx, IAS %u\n",
1316 				pgsize[i], ias[j]);
1317 			if (arm_lpae_run_tests(&cfg))
1318 				fail++;
1319 			else
1320 				pass++;
1321 		}
1322 	}
1323 
1324 	pr_info("selftest: completed with %d PASS %d FAIL\n", pass, fail);
1325 	return fail ? -EFAULT : 0;
1326 }
1327 subsys_initcall(arm_lpae_do_selftests);
1328 #endif
1329