2 * CPU-agnostic ARM page table allocator.
4 * ARMv7 Short-descriptor format, supporting
5 * - Basic memory attributes
6 * - Simplified access permissions (AP[2:1] model)
7 * - Backwards-compatible TEX remap
8 * - Large pages/supersections (if indicated by the caller)
11 * - Legacy access permissions (AP[2:0] model)
13 * Almost certainly never supporting:
17 * This program is free software; you can redistribute it and/or modify
18 * it under the terms of the GNU General Public License version 2 as
19 * published by the Free Software Foundation.
21 * This program is distributed in the hope that it will be useful,
22 * but WITHOUT ANY WARRANTY; without even the implied warranty of
23 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
24 * GNU General Public License for more details.
26 * You should have received a copy of the GNU General Public License
27 * along with this program. If not, see <http://www.gnu.org/licenses/>.
29 * Copyright (C) 2014-2015 ARM Limited
30 * Copyright (c) 2014-2015 MediaTek Inc.
33 #define pr_fmt(fmt) "arm-v7s io-pgtable: " fmt
35 #include <linux/dma-mapping.h>
36 #include <linux/gfp.h>
37 #include <linux/iommu.h>
38 #include <linux/kernel.h>
39 #include <linux/kmemleak.h>
40 #include <linux/sizes.h>
41 #include <linux/slab.h>
42 #include <linux/types.h>
44 #include <asm/barrier.h>
46 #include "io-pgtable.h"
48 /* Struct accessors */
49 #define io_pgtable_to_data(x) \
50 container_of((x), struct arm_v7s_io_pgtable, iop)
52 #define io_pgtable_ops_to_data(x) \
53 io_pgtable_to_data(io_pgtable_ops_to_pgtable(x))
56 * We have 32 bits total; 12 bits resolved at level 1, 8 bits at level 2,
57 * and 12 bits in a page. With some carefully-chosen coefficients we can
58 * hide the ugly inconsistencies behind these macros and at least let the
59 * rest of the code pretend to be somewhat sane.
61 #define ARM_V7S_ADDR_BITS 32
62 #define _ARM_V7S_LVL_BITS(lvl) (16 - (lvl) * 4)
63 #define ARM_V7S_LVL_SHIFT(lvl) (ARM_V7S_ADDR_BITS - (4 + 8 * (lvl)))
64 #define ARM_V7S_TABLE_SHIFT 10
66 #define ARM_V7S_PTES_PER_LVL(lvl) (1 << _ARM_V7S_LVL_BITS(lvl))
67 #define ARM_V7S_TABLE_SIZE(lvl) \
68 (ARM_V7S_PTES_PER_LVL(lvl) * sizeof(arm_v7s_iopte))
70 #define ARM_V7S_BLOCK_SIZE(lvl) (1UL << ARM_V7S_LVL_SHIFT(lvl))
71 #define ARM_V7S_LVL_MASK(lvl) ((u32)(~0U << ARM_V7S_LVL_SHIFT(lvl)))
72 #define ARM_V7S_TABLE_MASK ((u32)(~0U << ARM_V7S_TABLE_SHIFT))
73 #define _ARM_V7S_IDX_MASK(lvl) (ARM_V7S_PTES_PER_LVL(lvl) - 1)
74 #define ARM_V7S_LVL_IDX(addr, lvl) ({ \
76 ((u32)(addr) >> ARM_V7S_LVL_SHIFT(_l)) & _ARM_V7S_IDX_MASK(_l); \
80 * Large page/supersection entries are effectively a block of 16 page/section
81 * entries, along the lines of the LPAE contiguous hint, but all with the
82 * same output address. For want of a better common name we'll call them
83 * "contiguous" versions of their respective page/section entries here, but
84 * noting the distinction (WRT to TLB maintenance) that they represent *one*
85 * entry repeated 16 times, not 16 separate entries (as in the LPAE case).
87 #define ARM_V7S_CONT_PAGES 16
89 /* PTE type bits: these are all mixed up with XN/PXN bits in most cases */
90 #define ARM_V7S_PTE_TYPE_TABLE 0x1
91 #define ARM_V7S_PTE_TYPE_PAGE 0x2
92 #define ARM_V7S_PTE_TYPE_CONT_PAGE 0x1
94 #define ARM_V7S_PTE_IS_VALID(pte) (((pte) & 0x3) != 0)
95 #define ARM_V7S_PTE_IS_TABLE(pte, lvl) (lvl == 1 && ((pte) & ARM_V7S_PTE_TYPE_TABLE))
98 #define ARM_V7S_ATTR_XN(lvl) BIT(4 * (2 - (lvl)))
99 #define ARM_V7S_ATTR_B BIT(2)
100 #define ARM_V7S_ATTR_C BIT(3)
101 #define ARM_V7S_ATTR_NS_TABLE BIT(3)
102 #define ARM_V7S_ATTR_NS_SECTION BIT(19)
104 #define ARM_V7S_CONT_SECTION BIT(18)
105 #define ARM_V7S_CONT_PAGE_XN_SHIFT 15
108 * The attribute bits are consistently ordered*, but occupy bits [17:10] of
109 * a level 1 PTE vs. bits [11:4] at level 2. Thus we define the individual
110 * fields relative to that 8-bit block, plus a total shift relative to the PTE.
112 #define ARM_V7S_ATTR_SHIFT(lvl) (16 - (lvl) * 6)
114 #define ARM_V7S_ATTR_MASK 0xff
115 #define ARM_V7S_ATTR_AP0 BIT(0)
116 #define ARM_V7S_ATTR_AP1 BIT(1)
117 #define ARM_V7S_ATTR_AP2 BIT(5)
118 #define ARM_V7S_ATTR_S BIT(6)
119 #define ARM_V7S_ATTR_NG BIT(7)
120 #define ARM_V7S_TEX_SHIFT 2
121 #define ARM_V7S_TEX_MASK 0x7
122 #define ARM_V7S_ATTR_TEX(val) (((val) & ARM_V7S_TEX_MASK) << ARM_V7S_TEX_SHIFT)
124 /* *well, except for TEX on level 2 large pages, of course :( */
125 #define ARM_V7S_CONT_PAGE_TEX_SHIFT 6
126 #define ARM_V7S_CONT_PAGE_TEX_MASK (ARM_V7S_TEX_MASK << ARM_V7S_CONT_PAGE_TEX_SHIFT)
128 /* Simplified access permissions */
129 #define ARM_V7S_PTE_AF ARM_V7S_ATTR_AP0
130 #define ARM_V7S_PTE_AP_UNPRIV ARM_V7S_ATTR_AP1
131 #define ARM_V7S_PTE_AP_RDONLY ARM_V7S_ATTR_AP2
134 #define ARM_V7S_RGN_NC 0
135 #define ARM_V7S_RGN_WBWA 1
136 #define ARM_V7S_RGN_WT 2
137 #define ARM_V7S_RGN_WB 3
139 #define ARM_V7S_PRRR_TYPE_DEVICE 1
140 #define ARM_V7S_PRRR_TYPE_NORMAL 2
141 #define ARM_V7S_PRRR_TR(n, type) (((type) & 0x3) << ((n) * 2))
142 #define ARM_V7S_PRRR_DS0 BIT(16)
143 #define ARM_V7S_PRRR_DS1 BIT(17)
144 #define ARM_V7S_PRRR_NS0 BIT(18)
145 #define ARM_V7S_PRRR_NS1 BIT(19)
146 #define ARM_V7S_PRRR_NOS(n) BIT((n) + 24)
148 #define ARM_V7S_NMRR_IR(n, attr) (((attr) & 0x3) << ((n) * 2))
149 #define ARM_V7S_NMRR_OR(n, attr) (((attr) & 0x3) << ((n) * 2 + 16))
151 #define ARM_V7S_TTBR_S BIT(1)
152 #define ARM_V7S_TTBR_NOS BIT(5)
153 #define ARM_V7S_TTBR_ORGN_ATTR(attr) (((attr) & 0x3) << 3)
154 #define ARM_V7S_TTBR_IRGN_ATTR(attr) \
155 ((((attr) & 0x1) << 6) | (((attr) & 0x2) >> 1))
157 #define ARM_V7S_TCR_PD1 BIT(5)
159 typedef u32 arm_v7s_iopte;
161 static bool selftest_running;
163 struct arm_v7s_io_pgtable {
164 struct io_pgtable iop;
167 struct kmem_cache *l2_tables;
170 static dma_addr_t __arm_v7s_dma_addr(void *pages)
172 return (dma_addr_t)virt_to_phys(pages);
175 static arm_v7s_iopte *iopte_deref(arm_v7s_iopte pte, int lvl)
177 if (ARM_V7S_PTE_IS_TABLE(pte, lvl))
178 pte &= ARM_V7S_TABLE_MASK;
180 pte &= ARM_V7S_LVL_MASK(lvl);
181 return phys_to_virt(pte);
184 static void *__arm_v7s_alloc_table(int lvl, gfp_t gfp,
185 struct arm_v7s_io_pgtable *data)
187 struct device *dev = data->iop.cfg.iommu_dev;
189 size_t size = ARM_V7S_TABLE_SIZE(lvl);
193 table = (void *)__get_dma_pages(__GFP_ZERO, get_order(size));
195 table = kmem_cache_zalloc(data->l2_tables, gfp | GFP_DMA);
196 if (table && !selftest_running) {
197 dma = dma_map_single(dev, table, size, DMA_TO_DEVICE);
198 if (dma_mapping_error(dev, dma))
201 * We depend on the IOMMU being able to work with any physical
202 * address directly, so if the DMA layer suggests otherwise by
203 * translating or truncating them, that bodes very badly...
205 if (dma != virt_to_phys(table))
208 kmemleak_ignore(table);
212 dev_err(dev, "Cannot accommodate DMA translation for IOMMU page tables\n");
213 dma_unmap_single(dev, dma, size, DMA_TO_DEVICE);
216 free_pages((unsigned long)table, get_order(size));
218 kmem_cache_free(data->l2_tables, table);
222 static void __arm_v7s_free_table(void *table, int lvl,
223 struct arm_v7s_io_pgtable *data)
225 struct device *dev = data->iop.cfg.iommu_dev;
226 size_t size = ARM_V7S_TABLE_SIZE(lvl);
228 if (!selftest_running)
229 dma_unmap_single(dev, __arm_v7s_dma_addr(table), size,
232 free_pages((unsigned long)table, get_order(size));
234 kmem_cache_free(data->l2_tables, table);
237 static void __arm_v7s_pte_sync(arm_v7s_iopte *ptep, int num_entries,
238 struct io_pgtable_cfg *cfg)
240 if (selftest_running)
243 dma_sync_single_for_device(cfg->iommu_dev, __arm_v7s_dma_addr(ptep),
244 num_entries * sizeof(*ptep), DMA_TO_DEVICE);
246 static void __arm_v7s_set_pte(arm_v7s_iopte *ptep, arm_v7s_iopte pte,
247 int num_entries, struct io_pgtable_cfg *cfg)
251 for (i = 0; i < num_entries; i++)
254 __arm_v7s_pte_sync(ptep, num_entries, cfg);
257 static arm_v7s_iopte arm_v7s_prot_to_pte(int prot, int lvl,
258 struct io_pgtable_cfg *cfg)
260 bool ap = !(cfg->quirks & IO_PGTABLE_QUIRK_NO_PERMS);
261 arm_v7s_iopte pte = ARM_V7S_ATTR_NG | ARM_V7S_ATTR_S |
265 pte |= ARM_V7S_PTE_AF | ARM_V7S_PTE_AP_UNPRIV;
266 if (!(prot & IOMMU_WRITE))
267 pte |= ARM_V7S_PTE_AP_RDONLY;
269 pte <<= ARM_V7S_ATTR_SHIFT(lvl);
271 if ((prot & IOMMU_NOEXEC) && ap)
272 pte |= ARM_V7S_ATTR_XN(lvl);
273 if (prot & IOMMU_CACHE)
274 pte |= ARM_V7S_ATTR_B | ARM_V7S_ATTR_C;
279 static int arm_v7s_pte_to_prot(arm_v7s_iopte pte, int lvl)
281 int prot = IOMMU_READ;
283 if (pte & (ARM_V7S_PTE_AP_RDONLY << ARM_V7S_ATTR_SHIFT(lvl)))
285 if (pte & ARM_V7S_ATTR_C)
291 static arm_v7s_iopte arm_v7s_pte_to_cont(arm_v7s_iopte pte, int lvl)
294 pte |= ARM_V7S_CONT_SECTION;
295 } else if (lvl == 2) {
296 arm_v7s_iopte xn = pte & ARM_V7S_ATTR_XN(lvl);
297 arm_v7s_iopte tex = pte & ARM_V7S_CONT_PAGE_TEX_MASK;
299 pte ^= xn | tex | ARM_V7S_PTE_TYPE_PAGE;
300 pte |= (xn << ARM_V7S_CONT_PAGE_XN_SHIFT) |
301 (tex << ARM_V7S_CONT_PAGE_TEX_SHIFT) |
302 ARM_V7S_PTE_TYPE_CONT_PAGE;
307 static arm_v7s_iopte arm_v7s_cont_to_pte(arm_v7s_iopte pte, int lvl)
310 pte &= ~ARM_V7S_CONT_SECTION;
311 } else if (lvl == 2) {
312 arm_v7s_iopte xn = pte & BIT(ARM_V7S_CONT_PAGE_XN_SHIFT);
313 arm_v7s_iopte tex = pte & (ARM_V7S_CONT_PAGE_TEX_MASK <<
314 ARM_V7S_CONT_PAGE_TEX_SHIFT);
316 pte ^= xn | tex | ARM_V7S_PTE_TYPE_CONT_PAGE;
317 pte |= (xn >> ARM_V7S_CONT_PAGE_XN_SHIFT) |
318 (tex >> ARM_V7S_CONT_PAGE_TEX_SHIFT) |
319 ARM_V7S_PTE_TYPE_PAGE;
324 static bool arm_v7s_pte_is_cont(arm_v7s_iopte pte, int lvl)
326 if (lvl == 1 && !ARM_V7S_PTE_IS_TABLE(pte, lvl))
327 return pte & ARM_V7S_CONT_SECTION;
329 return !(pte & ARM_V7S_PTE_TYPE_PAGE);
333 static int __arm_v7s_unmap(struct arm_v7s_io_pgtable *, unsigned long,
334 size_t, int, arm_v7s_iopte *);
336 static int arm_v7s_init_pte(struct arm_v7s_io_pgtable *data,
337 unsigned long iova, phys_addr_t paddr, int prot,
338 int lvl, int num_entries, arm_v7s_iopte *ptep)
340 struct io_pgtable_cfg *cfg = &data->iop.cfg;
341 arm_v7s_iopte pte = arm_v7s_prot_to_pte(prot, lvl, cfg);
344 for (i = 0; i < num_entries; i++)
345 if (ARM_V7S_PTE_IS_TABLE(ptep[i], lvl)) {
347 * We need to unmap and free the old table before
348 * overwriting it with a block entry.
351 size_t sz = ARM_V7S_BLOCK_SIZE(lvl);
353 tblp = ptep - ARM_V7S_LVL_IDX(iova, lvl);
354 if (WARN_ON(__arm_v7s_unmap(data, iova + i * sz,
355 sz, lvl, tblp) != sz))
357 } else if (ptep[i]) {
358 /* We require an unmap first */
359 WARN_ON(!selftest_running);
363 pte |= ARM_V7S_PTE_TYPE_PAGE;
364 if (lvl == 1 && (cfg->quirks & IO_PGTABLE_QUIRK_ARM_NS))
365 pte |= ARM_V7S_ATTR_NS_SECTION;
368 pte = arm_v7s_pte_to_cont(pte, lvl);
370 pte |= paddr & ARM_V7S_LVL_MASK(lvl);
372 __arm_v7s_set_pte(ptep, pte, num_entries, cfg);
376 static int __arm_v7s_map(struct arm_v7s_io_pgtable *data, unsigned long iova,
377 phys_addr_t paddr, size_t size, int prot,
378 int lvl, arm_v7s_iopte *ptep)
380 struct io_pgtable_cfg *cfg = &data->iop.cfg;
381 arm_v7s_iopte pte, *cptep;
382 int num_entries = size >> ARM_V7S_LVL_SHIFT(lvl);
384 /* Find our entry at the current level */
385 ptep += ARM_V7S_LVL_IDX(iova, lvl);
387 /* If we can install a leaf entry at this level, then do so */
389 return arm_v7s_init_pte(data, iova, paddr, prot,
390 lvl, num_entries, ptep);
392 /* We can't allocate tables at the final level */
393 if (WARN_ON(lvl == 2))
396 /* Grab a pointer to the next level */
399 cptep = __arm_v7s_alloc_table(lvl + 1, GFP_ATOMIC, data);
403 pte = virt_to_phys(cptep) | ARM_V7S_PTE_TYPE_TABLE;
404 if (cfg->quirks & IO_PGTABLE_QUIRK_ARM_NS)
405 pte |= ARM_V7S_ATTR_NS_TABLE;
407 __arm_v7s_set_pte(ptep, pte, 1, cfg);
409 cptep = iopte_deref(pte, lvl);
413 return __arm_v7s_map(data, iova, paddr, size, prot, lvl + 1, cptep);
416 static int arm_v7s_map(struct io_pgtable_ops *ops, unsigned long iova,
417 phys_addr_t paddr, size_t size, int prot)
419 struct arm_v7s_io_pgtable *data = io_pgtable_ops_to_data(ops);
420 struct io_pgtable *iop = &data->iop;
423 /* If no access, then nothing to do */
424 if (!(prot & (IOMMU_READ | IOMMU_WRITE)))
427 ret = __arm_v7s_map(data, iova, paddr, size, prot, 1, data->pgd);
429 * Synchronise all PTE updates for the new mapping before there's
430 * a chance for anything to kick off a table walk for the new iova.
432 if (iop->cfg.quirks & IO_PGTABLE_QUIRK_TLBI_ON_MAP) {
433 io_pgtable_tlb_add_flush(iop, iova, size,
434 ARM_V7S_BLOCK_SIZE(2), false);
435 io_pgtable_tlb_sync(iop);
443 static void arm_v7s_free_pgtable(struct io_pgtable *iop)
445 struct arm_v7s_io_pgtable *data = io_pgtable_to_data(iop);
448 for (i = 0; i < ARM_V7S_PTES_PER_LVL(1); i++) {
449 arm_v7s_iopte pte = data->pgd[i];
451 if (ARM_V7S_PTE_IS_TABLE(pte, 1))
452 __arm_v7s_free_table(iopte_deref(pte, 1), 2, data);
454 __arm_v7s_free_table(data->pgd, 1, data);
455 kmem_cache_destroy(data->l2_tables);
459 static void arm_v7s_split_cont(struct arm_v7s_io_pgtable *data,
460 unsigned long iova, int idx, int lvl,
463 struct io_pgtable *iop = &data->iop;
465 size_t size = ARM_V7S_BLOCK_SIZE(lvl);
468 ptep -= idx & (ARM_V7S_CONT_PAGES - 1);
469 pte = arm_v7s_cont_to_pte(*ptep, lvl);
470 for (i = 0; i < ARM_V7S_CONT_PAGES; i++) {
475 __arm_v7s_pte_sync(ptep, ARM_V7S_CONT_PAGES, &iop->cfg);
477 size *= ARM_V7S_CONT_PAGES;
478 io_pgtable_tlb_add_flush(iop, iova, size, size, true);
479 io_pgtable_tlb_sync(iop);
482 static int arm_v7s_split_blk_unmap(struct arm_v7s_io_pgtable *data,
483 unsigned long iova, size_t size,
486 unsigned long blk_start, blk_end, blk_size;
487 phys_addr_t blk_paddr;
488 arm_v7s_iopte table = 0;
489 int prot = arm_v7s_pte_to_prot(*ptep, 1);
491 blk_size = ARM_V7S_BLOCK_SIZE(1);
492 blk_start = iova & ARM_V7S_LVL_MASK(1);
493 blk_end = blk_start + ARM_V7S_BLOCK_SIZE(1);
494 blk_paddr = *ptep & ARM_V7S_LVL_MASK(1);
496 for (; blk_start < blk_end; blk_start += size, blk_paddr += size) {
497 arm_v7s_iopte *tablep;
500 if (blk_start == iova)
503 /* __arm_v7s_map expects a pointer to the start of the table */
504 tablep = &table - ARM_V7S_LVL_IDX(blk_start, 1);
505 if (__arm_v7s_map(data, blk_start, blk_paddr, size, prot, 1,
508 /* Free the table we allocated */
509 tablep = iopte_deref(table, 1);
510 __arm_v7s_free_table(tablep, 2, data);
512 return 0; /* Bytes unmapped */
516 __arm_v7s_set_pte(ptep, table, 1, &data->iop.cfg);
517 iova &= ~(blk_size - 1);
518 io_pgtable_tlb_add_flush(&data->iop, iova, blk_size, blk_size, true);
522 static int __arm_v7s_unmap(struct arm_v7s_io_pgtable *data,
523 unsigned long iova, size_t size, int lvl,
526 arm_v7s_iopte pte[ARM_V7S_CONT_PAGES];
527 struct io_pgtable *iop = &data->iop;
528 int idx, i = 0, num_entries = size >> ARM_V7S_LVL_SHIFT(lvl);
530 /* Something went horribly wrong and we ran out of page table */
531 if (WARN_ON(lvl > 2))
534 idx = ARM_V7S_LVL_IDX(iova, lvl);
537 if (WARN_ON(!ARM_V7S_PTE_IS_VALID(ptep[i])))
540 } while (++i < num_entries);
543 * If we've hit a contiguous 'large page' entry at this level, it
544 * needs splitting first, unless we're unmapping the whole lot.
546 if (num_entries <= 1 && arm_v7s_pte_is_cont(pte[0], lvl))
547 arm_v7s_split_cont(data, iova, idx, lvl, ptep);
549 /* If the size matches this level, we're in the right place */
551 size_t blk_size = ARM_V7S_BLOCK_SIZE(lvl);
553 __arm_v7s_set_pte(ptep, 0, num_entries, &iop->cfg);
555 for (i = 0; i < num_entries; i++) {
556 if (ARM_V7S_PTE_IS_TABLE(pte[i], lvl)) {
557 /* Also flush any partial walks */
558 io_pgtable_tlb_add_flush(iop, iova, blk_size,
559 ARM_V7S_BLOCK_SIZE(lvl + 1), false);
560 io_pgtable_tlb_sync(iop);
561 ptep = iopte_deref(pte[i], lvl);
562 __arm_v7s_free_table(ptep, lvl + 1, data);
564 io_pgtable_tlb_add_flush(iop, iova, blk_size,
570 } else if (lvl == 1 && !ARM_V7S_PTE_IS_TABLE(pte[0], lvl)) {
572 * Insert a table at the next level to map the old region,
573 * minus the part we want to unmap
575 return arm_v7s_split_blk_unmap(data, iova, size, ptep);
578 /* Keep on walkin' */
579 ptep = iopte_deref(pte[0], lvl);
580 return __arm_v7s_unmap(data, iova, size, lvl + 1, ptep);
583 static int arm_v7s_unmap(struct io_pgtable_ops *ops, unsigned long iova,
586 struct arm_v7s_io_pgtable *data = io_pgtable_ops_to_data(ops);
589 unmapped = __arm_v7s_unmap(data, iova, size, 1, data->pgd);
591 io_pgtable_tlb_sync(&data->iop);
596 static phys_addr_t arm_v7s_iova_to_phys(struct io_pgtable_ops *ops,
599 struct arm_v7s_io_pgtable *data = io_pgtable_ops_to_data(ops);
600 arm_v7s_iopte *ptep = data->pgd, pte;
605 pte = ptep[ARM_V7S_LVL_IDX(iova, ++lvl)];
606 ptep = iopte_deref(pte, lvl);
607 } while (ARM_V7S_PTE_IS_TABLE(pte, lvl));
609 if (!ARM_V7S_PTE_IS_VALID(pte))
612 mask = ARM_V7S_LVL_MASK(lvl);
613 if (arm_v7s_pte_is_cont(pte, lvl))
614 mask *= ARM_V7S_CONT_PAGES;
615 return (pte & mask) | (iova & ~mask);
618 static struct io_pgtable *arm_v7s_alloc_pgtable(struct io_pgtable_cfg *cfg,
621 struct arm_v7s_io_pgtable *data;
623 if (cfg->ias > ARM_V7S_ADDR_BITS || cfg->oas > ARM_V7S_ADDR_BITS)
626 if (cfg->quirks & ~(IO_PGTABLE_QUIRK_ARM_NS |
627 IO_PGTABLE_QUIRK_NO_PERMS |
628 IO_PGTABLE_QUIRK_TLBI_ON_MAP))
631 data = kmalloc(sizeof(*data), GFP_KERNEL);
635 data->l2_tables = kmem_cache_create("io-pgtable_armv7s_l2",
636 ARM_V7S_TABLE_SIZE(2),
637 ARM_V7S_TABLE_SIZE(2),
638 SLAB_CACHE_DMA, NULL);
639 if (!data->l2_tables)
642 data->iop.ops = (struct io_pgtable_ops) {
644 .unmap = arm_v7s_unmap,
645 .iova_to_phys = arm_v7s_iova_to_phys,
648 /* We have to do this early for __arm_v7s_alloc_table to work... */
649 data->iop.cfg = *cfg;
652 * Unless the IOMMU driver indicates supersection support by
653 * having SZ_16M set in the initial bitmap, they won't be used.
655 cfg->pgsize_bitmap &= SZ_4K | SZ_64K | SZ_1M | SZ_16M;
657 /* TCR: T0SZ=0, disable TTBR1 */
658 cfg->arm_v7s_cfg.tcr = ARM_V7S_TCR_PD1;
661 * TEX remap: the indices used map to the closest equivalent types
662 * under the non-TEX-remap interpretation of those attribute bits,
663 * excepting various implementation-defined aspects of shareability.
665 cfg->arm_v7s_cfg.prrr = ARM_V7S_PRRR_TR(1, ARM_V7S_PRRR_TYPE_DEVICE) |
666 ARM_V7S_PRRR_TR(4, ARM_V7S_PRRR_TYPE_NORMAL) |
667 ARM_V7S_PRRR_TR(7, ARM_V7S_PRRR_TYPE_NORMAL) |
668 ARM_V7S_PRRR_DS0 | ARM_V7S_PRRR_DS1 |
669 ARM_V7S_PRRR_NS1 | ARM_V7S_PRRR_NOS(7);
670 cfg->arm_v7s_cfg.nmrr = ARM_V7S_NMRR_IR(7, ARM_V7S_RGN_WBWA) |
671 ARM_V7S_NMRR_OR(7, ARM_V7S_RGN_WBWA);
673 /* Looking good; allocate a pgd */
674 data->pgd = __arm_v7s_alloc_table(1, GFP_KERNEL, data);
678 /* Ensure the empty pgd is visible before any actual TTBR write */
682 cfg->arm_v7s_cfg.ttbr[0] = virt_to_phys(data->pgd) |
683 ARM_V7S_TTBR_S | ARM_V7S_TTBR_NOS |
684 ARM_V7S_TTBR_IRGN_ATTR(ARM_V7S_RGN_WBWA) |
685 ARM_V7S_TTBR_ORGN_ATTR(ARM_V7S_RGN_WBWA);
686 cfg->arm_v7s_cfg.ttbr[1] = 0;
690 kmem_cache_destroy(data->l2_tables);
695 struct io_pgtable_init_fns io_pgtable_arm_v7s_init_fns = {
696 .alloc = arm_v7s_alloc_pgtable,
697 .free = arm_v7s_free_pgtable,
700 #ifdef CONFIG_IOMMU_IO_PGTABLE_ARMV7S_SELFTEST
702 static struct io_pgtable_cfg *cfg_cookie;
704 static void dummy_tlb_flush_all(void *cookie)
706 WARN_ON(cookie != cfg_cookie);
709 static void dummy_tlb_add_flush(unsigned long iova, size_t size,
710 size_t granule, bool leaf, void *cookie)
712 WARN_ON(cookie != cfg_cookie);
713 WARN_ON(!(size & cfg_cookie->pgsize_bitmap));
716 static void dummy_tlb_sync(void *cookie)
718 WARN_ON(cookie != cfg_cookie);
721 static struct iommu_gather_ops dummy_tlb_ops = {
722 .tlb_flush_all = dummy_tlb_flush_all,
723 .tlb_add_flush = dummy_tlb_add_flush,
724 .tlb_sync = dummy_tlb_sync,
727 #define __FAIL(ops) ({ \
728 WARN(1, "selftest: test failed\n"); \
729 selftest_running = false; \
733 static int __init arm_v7s_do_selftests(void)
735 struct io_pgtable_ops *ops;
736 struct io_pgtable_cfg cfg = {
737 .tlb = &dummy_tlb_ops,
740 .quirks = IO_PGTABLE_QUIRK_ARM_NS,
741 .pgsize_bitmap = SZ_4K | SZ_64K | SZ_1M | SZ_16M,
743 unsigned int iova, size, iova_start;
744 unsigned int i, loopnr = 0;
746 selftest_running = true;
750 ops = alloc_io_pgtable_ops(ARM_V7S, &cfg, &cfg);
752 pr_err("selftest: failed to allocate io pgtable ops\n");
757 * Initial sanity checks.
758 * Empty page tables shouldn't provide any translations.
760 if (ops->iova_to_phys(ops, 42))
763 if (ops->iova_to_phys(ops, SZ_1G + 42))
766 if (ops->iova_to_phys(ops, SZ_2G + 42))
770 * Distinct mappings of different granule sizes.
773 i = find_first_bit(&cfg.pgsize_bitmap, BITS_PER_LONG);
774 while (i != BITS_PER_LONG) {
776 if (ops->map(ops, iova, iova, size, IOMMU_READ |
782 /* Overlapping mappings */
783 if (!ops->map(ops, iova, iova + size, size,
784 IOMMU_READ | IOMMU_NOEXEC))
787 if (ops->iova_to_phys(ops, iova + 42) != (iova + 42))
792 i = find_next_bit(&cfg.pgsize_bitmap, BITS_PER_LONG, i);
798 size = 1UL << __ffs(cfg.pgsize_bitmap);
800 iova_start = i * SZ_16M;
801 if (ops->unmap(ops, iova_start + size, size) != size)
804 /* Remap of partial unmap */
805 if (ops->map(ops, iova_start + size, size, size, IOMMU_READ))
808 if (ops->iova_to_phys(ops, iova_start + size + 42)
816 i = find_first_bit(&cfg.pgsize_bitmap, BITS_PER_LONG);
817 while (i != BITS_PER_LONG) {
820 if (ops->unmap(ops, iova, size) != size)
823 if (ops->iova_to_phys(ops, iova + 42))
826 /* Remap full block */
827 if (ops->map(ops, iova, iova, size, IOMMU_WRITE))
830 if (ops->iova_to_phys(ops, iova + 42) != (iova + 42))
835 i = find_next_bit(&cfg.pgsize_bitmap, BITS_PER_LONG, i);
838 free_io_pgtable_ops(ops);
840 selftest_running = false;
842 pr_info("self test ok\n");
845 subsys_initcall(arm_v7s_do_selftests);