blob: def8ca1c982d5869a358ca69467a6f13763c46fb [file] [log] [blame]
Robin Murphye5fc9752016-01-26 17:13:13 +00001/*
2 * CPU-agnostic ARM page table allocator.
3 *
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)
9 *
10 * Not supporting:
11 * - Legacy access permissions (AP[2:0] model)
12 *
13 * Almost certainly never supporting:
14 * - PXN
15 * - Domains
16 *
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.
20 *
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.
25 *
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/>.
28 *
29 * Copyright (C) 2014-2015 ARM Limited
30 * Copyright (c) 2014-2015 MediaTek Inc.
31 */
32
33#define pr_fmt(fmt) "arm-v7s io-pgtable: " fmt
34
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>
43
44#include <asm/barrier.h>
45
46#include "io-pgtable.h"
47
48/* Struct accessors */
49#define io_pgtable_to_data(x) \
50 container_of((x), struct arm_v7s_io_pgtable, iop)
51
52#define io_pgtable_ops_to_data(x) \
53 io_pgtable_to_data(io_pgtable_ops_to_pgtable(x))
54
55/*
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.
60 */
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
65
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))
69
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) ({ \
75 int _l = lvl; \
76 ((u32)(addr) >> ARM_V7S_LVL_SHIFT(_l)) & _ARM_V7S_IDX_MASK(_l); \
77})
78
79/*
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).
86 */
87#define ARM_V7S_CONT_PAGES 16
88
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
93
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))
96
97/* Page table bits */
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)
103
104#define ARM_V7S_CONT_SECTION BIT(18)
105#define ARM_V7S_CONT_PAGE_XN_SHIFT 15
106
107/*
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.
111 */
112#define ARM_V7S_ATTR_SHIFT(lvl) (16 - (lvl) * 6)
113
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)
123
Yong Wu1afe2312016-03-14 06:01:10 +0800124#define ARM_V7S_ATTR_MTK_4GB BIT(9) /* MTK extend it for 4GB mode */
125
Robin Murphye5fc9752016-01-26 17:13:13 +0000126/* *well, except for TEX on level 2 large pages, of course :( */
127#define ARM_V7S_CONT_PAGE_TEX_SHIFT 6
128#define ARM_V7S_CONT_PAGE_TEX_MASK (ARM_V7S_TEX_MASK << ARM_V7S_CONT_PAGE_TEX_SHIFT)
129
130/* Simplified access permissions */
131#define ARM_V7S_PTE_AF ARM_V7S_ATTR_AP0
132#define ARM_V7S_PTE_AP_UNPRIV ARM_V7S_ATTR_AP1
133#define ARM_V7S_PTE_AP_RDONLY ARM_V7S_ATTR_AP2
134
135/* Register bits */
136#define ARM_V7S_RGN_NC 0
137#define ARM_V7S_RGN_WBWA 1
138#define ARM_V7S_RGN_WT 2
139#define ARM_V7S_RGN_WB 3
140
141#define ARM_V7S_PRRR_TYPE_DEVICE 1
142#define ARM_V7S_PRRR_TYPE_NORMAL 2
143#define ARM_V7S_PRRR_TR(n, type) (((type) & 0x3) << ((n) * 2))
144#define ARM_V7S_PRRR_DS0 BIT(16)
145#define ARM_V7S_PRRR_DS1 BIT(17)
146#define ARM_V7S_PRRR_NS0 BIT(18)
147#define ARM_V7S_PRRR_NS1 BIT(19)
148#define ARM_V7S_PRRR_NOS(n) BIT((n) + 24)
149
150#define ARM_V7S_NMRR_IR(n, attr) (((attr) & 0x3) << ((n) * 2))
151#define ARM_V7S_NMRR_OR(n, attr) (((attr) & 0x3) << ((n) * 2 + 16))
152
153#define ARM_V7S_TTBR_S BIT(1)
154#define ARM_V7S_TTBR_NOS BIT(5)
155#define ARM_V7S_TTBR_ORGN_ATTR(attr) (((attr) & 0x3) << 3)
156#define ARM_V7S_TTBR_IRGN_ATTR(attr) \
157 ((((attr) & 0x1) << 6) | (((attr) & 0x2) >> 1))
158
159#define ARM_V7S_TCR_PD1 BIT(5)
160
161typedef u32 arm_v7s_iopte;
162
163static bool selftest_running;
164
165struct arm_v7s_io_pgtable {
166 struct io_pgtable iop;
167
168 arm_v7s_iopte *pgd;
169 struct kmem_cache *l2_tables;
170};
171
172static dma_addr_t __arm_v7s_dma_addr(void *pages)
173{
174 return (dma_addr_t)virt_to_phys(pages);
175}
176
177static arm_v7s_iopte *iopte_deref(arm_v7s_iopte pte, int lvl)
178{
179 if (ARM_V7S_PTE_IS_TABLE(pte, lvl))
180 pte &= ARM_V7S_TABLE_MASK;
181 else
182 pte &= ARM_V7S_LVL_MASK(lvl);
183 return phys_to_virt(pte);
184}
185
186static void *__arm_v7s_alloc_table(int lvl, gfp_t gfp,
187 struct arm_v7s_io_pgtable *data)
188{
189 struct device *dev = data->iop.cfg.iommu_dev;
190 dma_addr_t dma;
191 size_t size = ARM_V7S_TABLE_SIZE(lvl);
192 void *table = NULL;
193
194 if (lvl == 1)
195 table = (void *)__get_dma_pages(__GFP_ZERO, get_order(size));
196 else if (lvl == 2)
Robin Murphy048b31c2016-03-01 19:07:03 +0000197 table = kmem_cache_zalloc(data->l2_tables, gfp | GFP_DMA);
Robin Murphye5fc9752016-01-26 17:13:13 +0000198 if (table && !selftest_running) {
199 dma = dma_map_single(dev, table, size, DMA_TO_DEVICE);
200 if (dma_mapping_error(dev, dma))
201 goto out_free;
202 /*
203 * We depend on the IOMMU being able to work with any physical
204 * address directly, so if the DMA layer suggests otherwise by
205 * translating or truncating them, that bodes very badly...
206 */
207 if (dma != virt_to_phys(table))
208 goto out_unmap;
209 }
210 kmemleak_ignore(table);
211 return table;
212
213out_unmap:
214 dev_err(dev, "Cannot accommodate DMA translation for IOMMU page tables\n");
215 dma_unmap_single(dev, dma, size, DMA_TO_DEVICE);
216out_free:
217 if (lvl == 1)
218 free_pages((unsigned long)table, get_order(size));
219 else
220 kmem_cache_free(data->l2_tables, table);
221 return NULL;
222}
223
224static void __arm_v7s_free_table(void *table, int lvl,
225 struct arm_v7s_io_pgtable *data)
226{
227 struct device *dev = data->iop.cfg.iommu_dev;
228 size_t size = ARM_V7S_TABLE_SIZE(lvl);
229
230 if (!selftest_running)
231 dma_unmap_single(dev, __arm_v7s_dma_addr(table), size,
232 DMA_TO_DEVICE);
233 if (lvl == 1)
234 free_pages((unsigned long)table, get_order(size));
235 else
236 kmem_cache_free(data->l2_tables, table);
237}
238
239static void __arm_v7s_pte_sync(arm_v7s_iopte *ptep, int num_entries,
240 struct io_pgtable_cfg *cfg)
241{
242 if (selftest_running)
243 return;
244
245 dma_sync_single_for_device(cfg->iommu_dev, __arm_v7s_dma_addr(ptep),
246 num_entries * sizeof(*ptep), DMA_TO_DEVICE);
247}
248static void __arm_v7s_set_pte(arm_v7s_iopte *ptep, arm_v7s_iopte pte,
249 int num_entries, struct io_pgtable_cfg *cfg)
250{
251 int i;
252
253 for (i = 0; i < num_entries; i++)
254 ptep[i] = pte;
255
256 __arm_v7s_pte_sync(ptep, num_entries, cfg);
257}
258
259static arm_v7s_iopte arm_v7s_prot_to_pte(int prot, int lvl,
260 struct io_pgtable_cfg *cfg)
261{
262 bool ap = !(cfg->quirks & IO_PGTABLE_QUIRK_NO_PERMS);
Robin Murphye88ccab2016-04-05 12:39:32 +0100263 arm_v7s_iopte pte = ARM_V7S_ATTR_NG | ARM_V7S_ATTR_S;
Robin Murphye5fc9752016-01-26 17:13:13 +0000264
Robin Murphye88ccab2016-04-05 12:39:32 +0100265 if (!(prot & IOMMU_MMIO))
266 pte |= ARM_V7S_ATTR_TEX(1);
Robin Murphye5fc9752016-01-26 17:13:13 +0000267 if (ap) {
268 pte |= ARM_V7S_PTE_AF | ARM_V7S_PTE_AP_UNPRIV;
269 if (!(prot & IOMMU_WRITE))
270 pte |= ARM_V7S_PTE_AP_RDONLY;
271 }
272 pte <<= ARM_V7S_ATTR_SHIFT(lvl);
273
274 if ((prot & IOMMU_NOEXEC) && ap)
275 pte |= ARM_V7S_ATTR_XN(lvl);
Robin Murphye88ccab2016-04-05 12:39:32 +0100276 if (prot & IOMMU_MMIO)
277 pte |= ARM_V7S_ATTR_B;
278 else if (prot & IOMMU_CACHE)
Robin Murphye5fc9752016-01-26 17:13:13 +0000279 pte |= ARM_V7S_ATTR_B | ARM_V7S_ATTR_C;
280
281 return pte;
282}
283
284static int arm_v7s_pte_to_prot(arm_v7s_iopte pte, int lvl)
285{
286 int prot = IOMMU_READ;
Robin Murphye88ccab2016-04-05 12:39:32 +0100287 arm_v7s_iopte attr = pte >> ARM_V7S_ATTR_SHIFT(lvl);
Robin Murphye5fc9752016-01-26 17:13:13 +0000288
Robin Murphye633fc72016-08-11 17:44:05 +0100289 if (!(attr & ARM_V7S_PTE_AP_RDONLY))
Robin Murphye5fc9752016-01-26 17:13:13 +0000290 prot |= IOMMU_WRITE;
Robin Murphye88ccab2016-04-05 12:39:32 +0100291 if ((attr & (ARM_V7S_TEX_MASK << ARM_V7S_TEX_SHIFT)) == 0)
292 prot |= IOMMU_MMIO;
293 else if (pte & ARM_V7S_ATTR_C)
Robin Murphye5fc9752016-01-26 17:13:13 +0000294 prot |= IOMMU_CACHE;
Robin Murphye633fc72016-08-11 17:44:05 +0100295 if (pte & ARM_V7S_ATTR_XN(lvl))
296 prot |= IOMMU_NOEXEC;
Robin Murphye5fc9752016-01-26 17:13:13 +0000297
298 return prot;
299}
300
301static arm_v7s_iopte arm_v7s_pte_to_cont(arm_v7s_iopte pte, int lvl)
302{
303 if (lvl == 1) {
304 pte |= ARM_V7S_CONT_SECTION;
305 } else if (lvl == 2) {
306 arm_v7s_iopte xn = pte & ARM_V7S_ATTR_XN(lvl);
307 arm_v7s_iopte tex = pte & ARM_V7S_CONT_PAGE_TEX_MASK;
308
309 pte ^= xn | tex | ARM_V7S_PTE_TYPE_PAGE;
310 pte |= (xn << ARM_V7S_CONT_PAGE_XN_SHIFT) |
311 (tex << ARM_V7S_CONT_PAGE_TEX_SHIFT) |
312 ARM_V7S_PTE_TYPE_CONT_PAGE;
313 }
314 return pte;
315}
316
317static arm_v7s_iopte arm_v7s_cont_to_pte(arm_v7s_iopte pte, int lvl)
318{
319 if (lvl == 1) {
320 pte &= ~ARM_V7S_CONT_SECTION;
321 } else if (lvl == 2) {
322 arm_v7s_iopte xn = pte & BIT(ARM_V7S_CONT_PAGE_XN_SHIFT);
323 arm_v7s_iopte tex = pte & (ARM_V7S_CONT_PAGE_TEX_MASK <<
324 ARM_V7S_CONT_PAGE_TEX_SHIFT);
325
326 pte ^= xn | tex | ARM_V7S_PTE_TYPE_CONT_PAGE;
327 pte |= (xn >> ARM_V7S_CONT_PAGE_XN_SHIFT) |
328 (tex >> ARM_V7S_CONT_PAGE_TEX_SHIFT) |
329 ARM_V7S_PTE_TYPE_PAGE;
330 }
331 return pte;
332}
333
334static bool arm_v7s_pte_is_cont(arm_v7s_iopte pte, int lvl)
335{
336 if (lvl == 1 && !ARM_V7S_PTE_IS_TABLE(pte, lvl))
337 return pte & ARM_V7S_CONT_SECTION;
338 else if (lvl == 2)
339 return !(pte & ARM_V7S_PTE_TYPE_PAGE);
340 return false;
341}
342
343static int __arm_v7s_unmap(struct arm_v7s_io_pgtable *, unsigned long,
344 size_t, int, arm_v7s_iopte *);
345
346static int arm_v7s_init_pte(struct arm_v7s_io_pgtable *data,
347 unsigned long iova, phys_addr_t paddr, int prot,
348 int lvl, int num_entries, arm_v7s_iopte *ptep)
349{
350 struct io_pgtable_cfg *cfg = &data->iop.cfg;
351 arm_v7s_iopte pte = arm_v7s_prot_to_pte(prot, lvl, cfg);
352 int i;
353
354 for (i = 0; i < num_entries; i++)
355 if (ARM_V7S_PTE_IS_TABLE(ptep[i], lvl)) {
356 /*
357 * We need to unmap and free the old table before
358 * overwriting it with a block entry.
359 */
360 arm_v7s_iopte *tblp;
361 size_t sz = ARM_V7S_BLOCK_SIZE(lvl);
362
363 tblp = ptep - ARM_V7S_LVL_IDX(iova, lvl);
364 if (WARN_ON(__arm_v7s_unmap(data, iova + i * sz,
365 sz, lvl, tblp) != sz))
366 return -EINVAL;
367 } else if (ptep[i]) {
368 /* We require an unmap first */
369 WARN_ON(!selftest_running);
370 return -EEXIST;
371 }
372
373 pte |= ARM_V7S_PTE_TYPE_PAGE;
374 if (lvl == 1 && (cfg->quirks & IO_PGTABLE_QUIRK_ARM_NS))
375 pte |= ARM_V7S_ATTR_NS_SECTION;
376
Yong Wu1afe2312016-03-14 06:01:10 +0800377 if (cfg->quirks & IO_PGTABLE_QUIRK_ARM_MTK_4GB)
378 pte |= ARM_V7S_ATTR_MTK_4GB;
379
Robin Murphye5fc9752016-01-26 17:13:13 +0000380 if (num_entries > 1)
381 pte = arm_v7s_pte_to_cont(pte, lvl);
382
383 pte |= paddr & ARM_V7S_LVL_MASK(lvl);
384
385 __arm_v7s_set_pte(ptep, pte, num_entries, cfg);
386 return 0;
387}
388
389static int __arm_v7s_map(struct arm_v7s_io_pgtable *data, unsigned long iova,
390 phys_addr_t paddr, size_t size, int prot,
391 int lvl, arm_v7s_iopte *ptep)
392{
393 struct io_pgtable_cfg *cfg = &data->iop.cfg;
394 arm_v7s_iopte pte, *cptep;
395 int num_entries = size >> ARM_V7S_LVL_SHIFT(lvl);
396
397 /* Find our entry at the current level */
398 ptep += ARM_V7S_LVL_IDX(iova, lvl);
399
400 /* If we can install a leaf entry at this level, then do so */
401 if (num_entries)
402 return arm_v7s_init_pte(data, iova, paddr, prot,
403 lvl, num_entries, ptep);
404
405 /* We can't allocate tables at the final level */
406 if (WARN_ON(lvl == 2))
407 return -EINVAL;
408
409 /* Grab a pointer to the next level */
410 pte = *ptep;
411 if (!pte) {
412 cptep = __arm_v7s_alloc_table(lvl + 1, GFP_ATOMIC, data);
413 if (!cptep)
414 return -ENOMEM;
415
416 pte = virt_to_phys(cptep) | ARM_V7S_PTE_TYPE_TABLE;
417 if (cfg->quirks & IO_PGTABLE_QUIRK_ARM_NS)
418 pte |= ARM_V7S_ATTR_NS_TABLE;
419
420 __arm_v7s_set_pte(ptep, pte, 1, cfg);
421 } else {
422 cptep = iopte_deref(pte, lvl);
423 }
424
425 /* Rinse, repeat */
426 return __arm_v7s_map(data, iova, paddr, size, prot, lvl + 1, cptep);
427}
428
429static int arm_v7s_map(struct io_pgtable_ops *ops, unsigned long iova,
430 phys_addr_t paddr, size_t size, int prot)
431{
432 struct arm_v7s_io_pgtable *data = io_pgtable_ops_to_data(ops);
Robin Murphy507e4c92016-01-26 17:13:14 +0000433 struct io_pgtable *iop = &data->iop;
Robin Murphye5fc9752016-01-26 17:13:13 +0000434 int ret;
435
436 /* If no access, then nothing to do */
437 if (!(prot & (IOMMU_READ | IOMMU_WRITE)))
438 return 0;
439
440 ret = __arm_v7s_map(data, iova, paddr, size, prot, 1, data->pgd);
441 /*
442 * Synchronise all PTE updates for the new mapping before there's
443 * a chance for anything to kick off a table walk for the new iova.
444 */
Robin Murphy507e4c92016-01-26 17:13:14 +0000445 if (iop->cfg.quirks & IO_PGTABLE_QUIRK_TLBI_ON_MAP) {
446 io_pgtable_tlb_add_flush(iop, iova, size,
447 ARM_V7S_BLOCK_SIZE(2), false);
448 io_pgtable_tlb_sync(iop);
Robin Murphye5fc9752016-01-26 17:13:13 +0000449 } else {
450 wmb();
451 }
452
453 return ret;
454}
455
456static void arm_v7s_free_pgtable(struct io_pgtable *iop)
457{
458 struct arm_v7s_io_pgtable *data = io_pgtable_to_data(iop);
459 int i;
460
461 for (i = 0; i < ARM_V7S_PTES_PER_LVL(1); i++) {
462 arm_v7s_iopte pte = data->pgd[i];
463
464 if (ARM_V7S_PTE_IS_TABLE(pte, 1))
465 __arm_v7s_free_table(iopte_deref(pte, 1), 2, data);
466 }
467 __arm_v7s_free_table(data->pgd, 1, data);
468 kmem_cache_destroy(data->l2_tables);
469 kfree(data);
470}
471
472static void arm_v7s_split_cont(struct arm_v7s_io_pgtable *data,
473 unsigned long iova, int idx, int lvl,
474 arm_v7s_iopte *ptep)
475{
Robin Murphy507e4c92016-01-26 17:13:14 +0000476 struct io_pgtable *iop = &data->iop;
Robin Murphye5fc9752016-01-26 17:13:13 +0000477 arm_v7s_iopte pte;
478 size_t size = ARM_V7S_BLOCK_SIZE(lvl);
479 int i;
480
481 ptep -= idx & (ARM_V7S_CONT_PAGES - 1);
482 pte = arm_v7s_cont_to_pte(*ptep, lvl);
483 for (i = 0; i < ARM_V7S_CONT_PAGES; i++) {
484 ptep[i] = pte;
485 pte += size;
486 }
487
Robin Murphy507e4c92016-01-26 17:13:14 +0000488 __arm_v7s_pte_sync(ptep, ARM_V7S_CONT_PAGES, &iop->cfg);
Robin Murphye5fc9752016-01-26 17:13:13 +0000489
490 size *= ARM_V7S_CONT_PAGES;
Robin Murphy507e4c92016-01-26 17:13:14 +0000491 io_pgtable_tlb_add_flush(iop, iova, size, size, true);
492 io_pgtable_tlb_sync(iop);
Robin Murphye5fc9752016-01-26 17:13:13 +0000493}
494
495static int arm_v7s_split_blk_unmap(struct arm_v7s_io_pgtable *data,
496 unsigned long iova, size_t size,
497 arm_v7s_iopte *ptep)
498{
499 unsigned long blk_start, blk_end, blk_size;
500 phys_addr_t blk_paddr;
501 arm_v7s_iopte table = 0;
Robin Murphye5fc9752016-01-26 17:13:13 +0000502 int prot = arm_v7s_pte_to_prot(*ptep, 1);
503
504 blk_size = ARM_V7S_BLOCK_SIZE(1);
505 blk_start = iova & ARM_V7S_LVL_MASK(1);
506 blk_end = blk_start + ARM_V7S_BLOCK_SIZE(1);
507 blk_paddr = *ptep & ARM_V7S_LVL_MASK(1);
508
509 for (; blk_start < blk_end; blk_start += size, blk_paddr += size) {
510 arm_v7s_iopte *tablep;
511
512 /* Unmap! */
513 if (blk_start == iova)
514 continue;
515
516 /* __arm_v7s_map expects a pointer to the start of the table */
517 tablep = &table - ARM_V7S_LVL_IDX(blk_start, 1);
518 if (__arm_v7s_map(data, blk_start, blk_paddr, size, prot, 1,
519 tablep) < 0) {
520 if (table) {
521 /* Free the table we allocated */
522 tablep = iopte_deref(table, 1);
523 __arm_v7s_free_table(tablep, 2, data);
524 }
525 return 0; /* Bytes unmapped */
526 }
527 }
528
Robin Murphy507e4c92016-01-26 17:13:14 +0000529 __arm_v7s_set_pte(ptep, table, 1, &data->iop.cfg);
Robin Murphye5fc9752016-01-26 17:13:13 +0000530 iova &= ~(blk_size - 1);
Robin Murphy507e4c92016-01-26 17:13:14 +0000531 io_pgtable_tlb_add_flush(&data->iop, iova, blk_size, blk_size, true);
Robin Murphye5fc9752016-01-26 17:13:13 +0000532 return size;
533}
534
535static int __arm_v7s_unmap(struct arm_v7s_io_pgtable *data,
536 unsigned long iova, size_t size, int lvl,
537 arm_v7s_iopte *ptep)
538{
539 arm_v7s_iopte pte[ARM_V7S_CONT_PAGES];
Robin Murphy507e4c92016-01-26 17:13:14 +0000540 struct io_pgtable *iop = &data->iop;
Robin Murphye5fc9752016-01-26 17:13:13 +0000541 int idx, i = 0, num_entries = size >> ARM_V7S_LVL_SHIFT(lvl);
542
543 /* Something went horribly wrong and we ran out of page table */
544 if (WARN_ON(lvl > 2))
545 return 0;
546
547 idx = ARM_V7S_LVL_IDX(iova, lvl);
548 ptep += idx;
549 do {
550 if (WARN_ON(!ARM_V7S_PTE_IS_VALID(ptep[i])))
551 return 0;
552 pte[i] = ptep[i];
553 } while (++i < num_entries);
554
555 /*
556 * If we've hit a contiguous 'large page' entry at this level, it
557 * needs splitting first, unless we're unmapping the whole lot.
558 */
559 if (num_entries <= 1 && arm_v7s_pte_is_cont(pte[0], lvl))
560 arm_v7s_split_cont(data, iova, idx, lvl, ptep);
561
562 /* If the size matches this level, we're in the right place */
563 if (num_entries) {
564 size_t blk_size = ARM_V7S_BLOCK_SIZE(lvl);
565
Robin Murphy507e4c92016-01-26 17:13:14 +0000566 __arm_v7s_set_pte(ptep, 0, num_entries, &iop->cfg);
Robin Murphye5fc9752016-01-26 17:13:13 +0000567
568 for (i = 0; i < num_entries; i++) {
569 if (ARM_V7S_PTE_IS_TABLE(pte[i], lvl)) {
570 /* Also flush any partial walks */
Robin Murphy507e4c92016-01-26 17:13:14 +0000571 io_pgtable_tlb_add_flush(iop, iova, blk_size,
572 ARM_V7S_BLOCK_SIZE(lvl + 1), false);
573 io_pgtable_tlb_sync(iop);
Robin Murphye5fc9752016-01-26 17:13:13 +0000574 ptep = iopte_deref(pte[i], lvl);
575 __arm_v7s_free_table(ptep, lvl + 1, data);
576 } else {
Robin Murphy507e4c92016-01-26 17:13:14 +0000577 io_pgtable_tlb_add_flush(iop, iova, blk_size,
578 blk_size, true);
Robin Murphye5fc9752016-01-26 17:13:13 +0000579 }
580 iova += blk_size;
581 }
582 return size;
583 } else if (lvl == 1 && !ARM_V7S_PTE_IS_TABLE(pte[0], lvl)) {
584 /*
585 * Insert a table at the next level to map the old region,
586 * minus the part we want to unmap
587 */
588 return arm_v7s_split_blk_unmap(data, iova, size, ptep);
589 }
590
591 /* Keep on walkin' */
592 ptep = iopte_deref(pte[0], lvl);
593 return __arm_v7s_unmap(data, iova, size, lvl + 1, ptep);
594}
595
596static int arm_v7s_unmap(struct io_pgtable_ops *ops, unsigned long iova,
597 size_t size)
598{
Robin Murphye5fc9752016-01-26 17:13:13 +0000599 struct arm_v7s_io_pgtable *data = io_pgtable_ops_to_data(ops);
Robin Murphy507e4c92016-01-26 17:13:14 +0000600 size_t unmapped;
Robin Murphye5fc9752016-01-26 17:13:13 +0000601
602 unmapped = __arm_v7s_unmap(data, iova, size, 1, data->pgd);
603 if (unmapped)
Robin Murphy507e4c92016-01-26 17:13:14 +0000604 io_pgtable_tlb_sync(&data->iop);
Robin Murphye5fc9752016-01-26 17:13:13 +0000605
606 return unmapped;
607}
608
609static phys_addr_t arm_v7s_iova_to_phys(struct io_pgtable_ops *ops,
610 unsigned long iova)
611{
612 struct arm_v7s_io_pgtable *data = io_pgtable_ops_to_data(ops);
613 arm_v7s_iopte *ptep = data->pgd, pte;
614 int lvl = 0;
615 u32 mask;
616
617 do {
618 pte = ptep[ARM_V7S_LVL_IDX(iova, ++lvl)];
619 ptep = iopte_deref(pte, lvl);
620 } while (ARM_V7S_PTE_IS_TABLE(pte, lvl));
621
622 if (!ARM_V7S_PTE_IS_VALID(pte))
623 return 0;
624
625 mask = ARM_V7S_LVL_MASK(lvl);
626 if (arm_v7s_pte_is_cont(pte, lvl))
627 mask *= ARM_V7S_CONT_PAGES;
628 return (pte & mask) | (iova & ~mask);
629}
630
631static struct io_pgtable *arm_v7s_alloc_pgtable(struct io_pgtable_cfg *cfg,
632 void *cookie)
633{
634 struct arm_v7s_io_pgtable *data;
635
636 if (cfg->ias > ARM_V7S_ADDR_BITS || cfg->oas > ARM_V7S_ADDR_BITS)
637 return NULL;
638
Robin Murphy3850db42016-02-12 17:09:46 +0000639 if (cfg->quirks & ~(IO_PGTABLE_QUIRK_ARM_NS |
640 IO_PGTABLE_QUIRK_NO_PERMS |
Yong Wu1afe2312016-03-14 06:01:10 +0800641 IO_PGTABLE_QUIRK_TLBI_ON_MAP |
642 IO_PGTABLE_QUIRK_ARM_MTK_4GB))
Robin Murphy3850db42016-02-12 17:09:46 +0000643 return NULL;
644
Yong Wu1afe2312016-03-14 06:01:10 +0800645 /* If ARM_MTK_4GB is enabled, the NO_PERMS is also expected. */
646 if (cfg->quirks & IO_PGTABLE_QUIRK_ARM_MTK_4GB &&
647 !(cfg->quirks & IO_PGTABLE_QUIRK_NO_PERMS))
648 return NULL;
649
Robin Murphye5fc9752016-01-26 17:13:13 +0000650 data = kmalloc(sizeof(*data), GFP_KERNEL);
651 if (!data)
652 return NULL;
653
654 data->l2_tables = kmem_cache_create("io-pgtable_armv7s_l2",
655 ARM_V7S_TABLE_SIZE(2),
656 ARM_V7S_TABLE_SIZE(2),
657 SLAB_CACHE_DMA, NULL);
658 if (!data->l2_tables)
659 goto out_free_data;
660
661 data->iop.ops = (struct io_pgtable_ops) {
662 .map = arm_v7s_map,
663 .unmap = arm_v7s_unmap,
664 .iova_to_phys = arm_v7s_iova_to_phys,
665 };
666
667 /* We have to do this early for __arm_v7s_alloc_table to work... */
668 data->iop.cfg = *cfg;
669
670 /*
671 * Unless the IOMMU driver indicates supersection support by
672 * having SZ_16M set in the initial bitmap, they won't be used.
673 */
674 cfg->pgsize_bitmap &= SZ_4K | SZ_64K | SZ_1M | SZ_16M;
675
676 /* TCR: T0SZ=0, disable TTBR1 */
677 cfg->arm_v7s_cfg.tcr = ARM_V7S_TCR_PD1;
678
679 /*
680 * TEX remap: the indices used map to the closest equivalent types
681 * under the non-TEX-remap interpretation of those attribute bits,
682 * excepting various implementation-defined aspects of shareability.
683 */
684 cfg->arm_v7s_cfg.prrr = ARM_V7S_PRRR_TR(1, ARM_V7S_PRRR_TYPE_DEVICE) |
685 ARM_V7S_PRRR_TR(4, ARM_V7S_PRRR_TYPE_NORMAL) |
686 ARM_V7S_PRRR_TR(7, ARM_V7S_PRRR_TYPE_NORMAL) |
687 ARM_V7S_PRRR_DS0 | ARM_V7S_PRRR_DS1 |
688 ARM_V7S_PRRR_NS1 | ARM_V7S_PRRR_NOS(7);
689 cfg->arm_v7s_cfg.nmrr = ARM_V7S_NMRR_IR(7, ARM_V7S_RGN_WBWA) |
690 ARM_V7S_NMRR_OR(7, ARM_V7S_RGN_WBWA);
691
692 /* Looking good; allocate a pgd */
693 data->pgd = __arm_v7s_alloc_table(1, GFP_KERNEL, data);
694 if (!data->pgd)
695 goto out_free_data;
696
697 /* Ensure the empty pgd is visible before any actual TTBR write */
698 wmb();
699
700 /* TTBRs */
701 cfg->arm_v7s_cfg.ttbr[0] = virt_to_phys(data->pgd) |
702 ARM_V7S_TTBR_S | ARM_V7S_TTBR_NOS |
703 ARM_V7S_TTBR_IRGN_ATTR(ARM_V7S_RGN_WBWA) |
704 ARM_V7S_TTBR_ORGN_ATTR(ARM_V7S_RGN_WBWA);
705 cfg->arm_v7s_cfg.ttbr[1] = 0;
706 return &data->iop;
707
708out_free_data:
709 kmem_cache_destroy(data->l2_tables);
710 kfree(data);
711 return NULL;
712}
713
714struct io_pgtable_init_fns io_pgtable_arm_v7s_init_fns = {
715 .alloc = arm_v7s_alloc_pgtable,
716 .free = arm_v7s_free_pgtable,
717};
718
719#ifdef CONFIG_IOMMU_IO_PGTABLE_ARMV7S_SELFTEST
720
721static struct io_pgtable_cfg *cfg_cookie;
722
723static void dummy_tlb_flush_all(void *cookie)
724{
725 WARN_ON(cookie != cfg_cookie);
726}
727
728static void dummy_tlb_add_flush(unsigned long iova, size_t size,
729 size_t granule, bool leaf, void *cookie)
730{
731 WARN_ON(cookie != cfg_cookie);
732 WARN_ON(!(size & cfg_cookie->pgsize_bitmap));
733}
734
735static void dummy_tlb_sync(void *cookie)
736{
737 WARN_ON(cookie != cfg_cookie);
738}
739
740static struct iommu_gather_ops dummy_tlb_ops = {
741 .tlb_flush_all = dummy_tlb_flush_all,
742 .tlb_add_flush = dummy_tlb_add_flush,
743 .tlb_sync = dummy_tlb_sync,
744};
745
746#define __FAIL(ops) ({ \
747 WARN(1, "selftest: test failed\n"); \
748 selftest_running = false; \
749 -EFAULT; \
750})
751
752static int __init arm_v7s_do_selftests(void)
753{
754 struct io_pgtable_ops *ops;
755 struct io_pgtable_cfg cfg = {
756 .tlb = &dummy_tlb_ops,
757 .oas = 32,
758 .ias = 32,
759 .quirks = IO_PGTABLE_QUIRK_ARM_NS,
760 .pgsize_bitmap = SZ_4K | SZ_64K | SZ_1M | SZ_16M,
761 };
762 unsigned int iova, size, iova_start;
763 unsigned int i, loopnr = 0;
764
765 selftest_running = true;
766
767 cfg_cookie = &cfg;
768
769 ops = alloc_io_pgtable_ops(ARM_V7S, &cfg, &cfg);
770 if (!ops) {
771 pr_err("selftest: failed to allocate io pgtable ops\n");
772 return -EINVAL;
773 }
774
775 /*
776 * Initial sanity checks.
777 * Empty page tables shouldn't provide any translations.
778 */
779 if (ops->iova_to_phys(ops, 42))
780 return __FAIL(ops);
781
782 if (ops->iova_to_phys(ops, SZ_1G + 42))
783 return __FAIL(ops);
784
785 if (ops->iova_to_phys(ops, SZ_2G + 42))
786 return __FAIL(ops);
787
788 /*
789 * Distinct mappings of different granule sizes.
790 */
791 iova = 0;
792 i = find_first_bit(&cfg.pgsize_bitmap, BITS_PER_LONG);
793 while (i != BITS_PER_LONG) {
794 size = 1UL << i;
795 if (ops->map(ops, iova, iova, size, IOMMU_READ |
796 IOMMU_WRITE |
797 IOMMU_NOEXEC |
798 IOMMU_CACHE))
799 return __FAIL(ops);
800
801 /* Overlapping mappings */
802 if (!ops->map(ops, iova, iova + size, size,
803 IOMMU_READ | IOMMU_NOEXEC))
804 return __FAIL(ops);
805
806 if (ops->iova_to_phys(ops, iova + 42) != (iova + 42))
807 return __FAIL(ops);
808
809 iova += SZ_16M;
810 i++;
811 i = find_next_bit(&cfg.pgsize_bitmap, BITS_PER_LONG, i);
812 loopnr++;
813 }
814
815 /* Partial unmap */
816 i = 1;
817 size = 1UL << __ffs(cfg.pgsize_bitmap);
818 while (i < loopnr) {
819 iova_start = i * SZ_16M;
820 if (ops->unmap(ops, iova_start + size, size) != size)
821 return __FAIL(ops);
822
823 /* Remap of partial unmap */
824 if (ops->map(ops, iova_start + size, size, size, IOMMU_READ))
825 return __FAIL(ops);
826
827 if (ops->iova_to_phys(ops, iova_start + size + 42)
828 != (size + 42))
829 return __FAIL(ops);
830 i++;
831 }
832
833 /* Full unmap */
834 iova = 0;
835 i = find_first_bit(&cfg.pgsize_bitmap, BITS_PER_LONG);
836 while (i != BITS_PER_LONG) {
837 size = 1UL << i;
838
839 if (ops->unmap(ops, iova, size) != size)
840 return __FAIL(ops);
841
842 if (ops->iova_to_phys(ops, iova + 42))
843 return __FAIL(ops);
844
845 /* Remap full block */
846 if (ops->map(ops, iova, iova, size, IOMMU_WRITE))
847 return __FAIL(ops);
848
849 if (ops->iova_to_phys(ops, iova + 42) != (iova + 42))
850 return __FAIL(ops);
851
852 iova += SZ_16M;
853 i++;
854 i = find_next_bit(&cfg.pgsize_bitmap, BITS_PER_LONG, i);
855 }
856
857 free_io_pgtable_ops(ops);
858
859 selftest_running = false;
860
861 pr_info("self test ok\n");
862 return 0;
863}
864subsys_initcall(arm_v7s_do_selftests);
865#endif