Merge tag 'v3.17-rc3' into next
[cascardo/linux.git] / drivers / iommu / arm-smmu.c
1 /*
2  * IOMMU API for ARM architected SMMU implementations.
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of the GNU General Public License version 2 as
6  * published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
11  * GNU General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public License
14  * along with this program; if not, write to the Free Software
15  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
16  *
17  * Copyright (C) 2013 ARM Limited
18  *
19  * Author: Will Deacon <will.deacon@arm.com>
20  *
21  * This driver currently supports:
22  *      - SMMUv1 and v2 implementations
23  *      - Stream-matching and stream-indexing
24  *      - v7/v8 long-descriptor format
25  *      - Non-secure access to the SMMU
26  *      - 4k and 64k pages, with contiguous pte hints.
27  *      - Up to 42-bit addressing (dependent on VA_BITS)
28  *      - Context fault reporting
29  */
30
31 #define pr_fmt(fmt) "arm-smmu: " fmt
32
33 #include <linux/delay.h>
34 #include <linux/dma-mapping.h>
35 #include <linux/err.h>
36 #include <linux/interrupt.h>
37 #include <linux/io.h>
38 #include <linux/iommu.h>
39 #include <linux/mm.h>
40 #include <linux/module.h>
41 #include <linux/of.h>
42 #include <linux/pci.h>
43 #include <linux/platform_device.h>
44 #include <linux/slab.h>
45 #include <linux/spinlock.h>
46
47 #include <linux/amba/bus.h>
48
49 #include <asm/pgalloc.h>
50
51 /* Maximum number of stream IDs assigned to a single device */
52 #define MAX_MASTER_STREAMIDS            MAX_PHANDLE_ARGS
53
54 /* Maximum number of context banks per SMMU */
55 #define ARM_SMMU_MAX_CBS                128
56
57 /* Maximum number of mapping groups per SMMU */
58 #define ARM_SMMU_MAX_SMRS               128
59
60 /* SMMU global address space */
61 #define ARM_SMMU_GR0(smmu)              ((smmu)->base)
62 #define ARM_SMMU_GR1(smmu)              ((smmu)->base + (smmu)->pagesize)
63
64 /*
65  * SMMU global address space with conditional offset to access secure
66  * aliases of non-secure registers (e.g. nsCR0: 0x400, nsGFSR: 0x448,
67  * nsGFSYNR0: 0x450)
68  */
69 #define ARM_SMMU_GR0_NS(smmu)                                           \
70         ((smmu)->base +                                                 \
71                 ((smmu->options & ARM_SMMU_OPT_SECURE_CFG_ACCESS)       \
72                         ? 0x400 : 0))
73
74 /* Page table bits */
75 #define ARM_SMMU_PTE_XN                 (((pteval_t)3) << 53)
76 #define ARM_SMMU_PTE_CONT               (((pteval_t)1) << 52)
77 #define ARM_SMMU_PTE_AF                 (((pteval_t)1) << 10)
78 #define ARM_SMMU_PTE_SH_NS              (((pteval_t)0) << 8)
79 #define ARM_SMMU_PTE_SH_OS              (((pteval_t)2) << 8)
80 #define ARM_SMMU_PTE_SH_IS              (((pteval_t)3) << 8)
81 #define ARM_SMMU_PTE_PAGE               (((pteval_t)3) << 0)
82
83 #if PAGE_SIZE == SZ_4K
84 #define ARM_SMMU_PTE_CONT_ENTRIES       16
85 #elif PAGE_SIZE == SZ_64K
86 #define ARM_SMMU_PTE_CONT_ENTRIES       32
87 #else
88 #define ARM_SMMU_PTE_CONT_ENTRIES       1
89 #endif
90
91 #define ARM_SMMU_PTE_CONT_SIZE          (PAGE_SIZE * ARM_SMMU_PTE_CONT_ENTRIES)
92 #define ARM_SMMU_PTE_CONT_MASK          (~(ARM_SMMU_PTE_CONT_SIZE - 1))
93
94 /* Stage-1 PTE */
95 #define ARM_SMMU_PTE_AP_UNPRIV          (((pteval_t)1) << 6)
96 #define ARM_SMMU_PTE_AP_RDONLY          (((pteval_t)2) << 6)
97 #define ARM_SMMU_PTE_ATTRINDX_SHIFT     2
98 #define ARM_SMMU_PTE_nG                 (((pteval_t)1) << 11)
99
100 /* Stage-2 PTE */
101 #define ARM_SMMU_PTE_HAP_FAULT          (((pteval_t)0) << 6)
102 #define ARM_SMMU_PTE_HAP_READ           (((pteval_t)1) << 6)
103 #define ARM_SMMU_PTE_HAP_WRITE          (((pteval_t)2) << 6)
104 #define ARM_SMMU_PTE_MEMATTR_OIWB       (((pteval_t)0xf) << 2)
105 #define ARM_SMMU_PTE_MEMATTR_NC         (((pteval_t)0x5) << 2)
106 #define ARM_SMMU_PTE_MEMATTR_DEV        (((pteval_t)0x1) << 2)
107
108 /* Configuration registers */
109 #define ARM_SMMU_GR0_sCR0               0x0
110 #define sCR0_CLIENTPD                   (1 << 0)
111 #define sCR0_GFRE                       (1 << 1)
112 #define sCR0_GFIE                       (1 << 2)
113 #define sCR0_GCFGFRE                    (1 << 4)
114 #define sCR0_GCFGFIE                    (1 << 5)
115 #define sCR0_USFCFG                     (1 << 10)
116 #define sCR0_VMIDPNE                    (1 << 11)
117 #define sCR0_PTM                        (1 << 12)
118 #define sCR0_FB                         (1 << 13)
119 #define sCR0_BSU_SHIFT                  14
120 #define sCR0_BSU_MASK                   0x3
121
122 /* Identification registers */
123 #define ARM_SMMU_GR0_ID0                0x20
124 #define ARM_SMMU_GR0_ID1                0x24
125 #define ARM_SMMU_GR0_ID2                0x28
126 #define ARM_SMMU_GR0_ID3                0x2c
127 #define ARM_SMMU_GR0_ID4                0x30
128 #define ARM_SMMU_GR0_ID5                0x34
129 #define ARM_SMMU_GR0_ID6                0x38
130 #define ARM_SMMU_GR0_ID7                0x3c
131 #define ARM_SMMU_GR0_sGFSR              0x48
132 #define ARM_SMMU_GR0_sGFSYNR0           0x50
133 #define ARM_SMMU_GR0_sGFSYNR1           0x54
134 #define ARM_SMMU_GR0_sGFSYNR2           0x58
135 #define ARM_SMMU_GR0_PIDR0              0xfe0
136 #define ARM_SMMU_GR0_PIDR1              0xfe4
137 #define ARM_SMMU_GR0_PIDR2              0xfe8
138
139 #define ID0_S1TS                        (1 << 30)
140 #define ID0_S2TS                        (1 << 29)
141 #define ID0_NTS                         (1 << 28)
142 #define ID0_SMS                         (1 << 27)
143 #define ID0_PTFS_SHIFT                  24
144 #define ID0_PTFS_MASK                   0x2
145 #define ID0_PTFS_V8_ONLY                0x2
146 #define ID0_CTTW                        (1 << 14)
147 #define ID0_NUMIRPT_SHIFT               16
148 #define ID0_NUMIRPT_MASK                0xff
149 #define ID0_NUMSMRG_SHIFT               0
150 #define ID0_NUMSMRG_MASK                0xff
151
152 #define ID1_PAGESIZE                    (1 << 31)
153 #define ID1_NUMPAGENDXB_SHIFT           28
154 #define ID1_NUMPAGENDXB_MASK            7
155 #define ID1_NUMS2CB_SHIFT               16
156 #define ID1_NUMS2CB_MASK                0xff
157 #define ID1_NUMCB_SHIFT                 0
158 #define ID1_NUMCB_MASK                  0xff
159
160 #define ID2_OAS_SHIFT                   4
161 #define ID2_OAS_MASK                    0xf
162 #define ID2_IAS_SHIFT                   0
163 #define ID2_IAS_MASK                    0xf
164 #define ID2_UBS_SHIFT                   8
165 #define ID2_UBS_MASK                    0xf
166 #define ID2_PTFS_4K                     (1 << 12)
167 #define ID2_PTFS_16K                    (1 << 13)
168 #define ID2_PTFS_64K                    (1 << 14)
169
170 #define PIDR2_ARCH_SHIFT                4
171 #define PIDR2_ARCH_MASK                 0xf
172
173 /* Global TLB invalidation */
174 #define ARM_SMMU_GR0_STLBIALL           0x60
175 #define ARM_SMMU_GR0_TLBIVMID           0x64
176 #define ARM_SMMU_GR0_TLBIALLNSNH        0x68
177 #define ARM_SMMU_GR0_TLBIALLH           0x6c
178 #define ARM_SMMU_GR0_sTLBGSYNC          0x70
179 #define ARM_SMMU_GR0_sTLBGSTATUS        0x74
180 #define sTLBGSTATUS_GSACTIVE            (1 << 0)
181 #define TLB_LOOP_TIMEOUT                1000000 /* 1s! */
182
183 /* Stream mapping registers */
184 #define ARM_SMMU_GR0_SMR(n)             (0x800 + ((n) << 2))
185 #define SMR_VALID                       (1 << 31)
186 #define SMR_MASK_SHIFT                  16
187 #define SMR_MASK_MASK                   0x7fff
188 #define SMR_ID_SHIFT                    0
189 #define SMR_ID_MASK                     0x7fff
190
191 #define ARM_SMMU_GR0_S2CR(n)            (0xc00 + ((n) << 2))
192 #define S2CR_CBNDX_SHIFT                0
193 #define S2CR_CBNDX_MASK                 0xff
194 #define S2CR_TYPE_SHIFT                 16
195 #define S2CR_TYPE_MASK                  0x3
196 #define S2CR_TYPE_TRANS                 (0 << S2CR_TYPE_SHIFT)
197 #define S2CR_TYPE_BYPASS                (1 << S2CR_TYPE_SHIFT)
198 #define S2CR_TYPE_FAULT                 (2 << S2CR_TYPE_SHIFT)
199
200 /* Context bank attribute registers */
201 #define ARM_SMMU_GR1_CBAR(n)            (0x0 + ((n) << 2))
202 #define CBAR_VMID_SHIFT                 0
203 #define CBAR_VMID_MASK                  0xff
204 #define CBAR_S1_BPSHCFG_SHIFT           8
205 #define CBAR_S1_BPSHCFG_MASK            3
206 #define CBAR_S1_BPSHCFG_NSH             3
207 #define CBAR_S1_MEMATTR_SHIFT           12
208 #define CBAR_S1_MEMATTR_MASK            0xf
209 #define CBAR_S1_MEMATTR_WB              0xf
210 #define CBAR_TYPE_SHIFT                 16
211 #define CBAR_TYPE_MASK                  0x3
212 #define CBAR_TYPE_S2_TRANS              (0 << CBAR_TYPE_SHIFT)
213 #define CBAR_TYPE_S1_TRANS_S2_BYPASS    (1 << CBAR_TYPE_SHIFT)
214 #define CBAR_TYPE_S1_TRANS_S2_FAULT     (2 << CBAR_TYPE_SHIFT)
215 #define CBAR_TYPE_S1_TRANS_S2_TRANS     (3 << CBAR_TYPE_SHIFT)
216 #define CBAR_IRPTNDX_SHIFT              24
217 #define CBAR_IRPTNDX_MASK               0xff
218
219 #define ARM_SMMU_GR1_CBA2R(n)           (0x800 + ((n) << 2))
220 #define CBA2R_RW64_32BIT                (0 << 0)
221 #define CBA2R_RW64_64BIT                (1 << 0)
222
223 /* Translation context bank */
224 #define ARM_SMMU_CB_BASE(smmu)          ((smmu)->base + ((smmu)->size >> 1))
225 #define ARM_SMMU_CB(smmu, n)            ((n) * (smmu)->pagesize)
226
227 #define ARM_SMMU_CB_SCTLR               0x0
228 #define ARM_SMMU_CB_RESUME              0x8
229 #define ARM_SMMU_CB_TTBCR2              0x10
230 #define ARM_SMMU_CB_TTBR0_LO            0x20
231 #define ARM_SMMU_CB_TTBR0_HI            0x24
232 #define ARM_SMMU_CB_TTBCR               0x30
233 #define ARM_SMMU_CB_S1_MAIR0            0x38
234 #define ARM_SMMU_CB_FSR                 0x58
235 #define ARM_SMMU_CB_FAR_LO              0x60
236 #define ARM_SMMU_CB_FAR_HI              0x64
237 #define ARM_SMMU_CB_FSYNR0              0x68
238 #define ARM_SMMU_CB_S1_TLBIASID         0x610
239
240 #define SCTLR_S1_ASIDPNE                (1 << 12)
241 #define SCTLR_CFCFG                     (1 << 7)
242 #define SCTLR_CFIE                      (1 << 6)
243 #define SCTLR_CFRE                      (1 << 5)
244 #define SCTLR_E                         (1 << 4)
245 #define SCTLR_AFE                       (1 << 2)
246 #define SCTLR_TRE                       (1 << 1)
247 #define SCTLR_M                         (1 << 0)
248 #define SCTLR_EAE_SBOP                  (SCTLR_AFE | SCTLR_TRE)
249
250 #define RESUME_RETRY                    (0 << 0)
251 #define RESUME_TERMINATE                (1 << 0)
252
253 #define TTBCR_EAE                       (1 << 31)
254
255 #define TTBCR_PASIZE_SHIFT              16
256 #define TTBCR_PASIZE_MASK               0x7
257
258 #define TTBCR_TG0_4K                    (0 << 14)
259 #define TTBCR_TG0_64K                   (1 << 14)
260
261 #define TTBCR_SH0_SHIFT                 12
262 #define TTBCR_SH0_MASK                  0x3
263 #define TTBCR_SH_NS                     0
264 #define TTBCR_SH_OS                     2
265 #define TTBCR_SH_IS                     3
266
267 #define TTBCR_ORGN0_SHIFT               10
268 #define TTBCR_IRGN0_SHIFT               8
269 #define TTBCR_RGN_MASK                  0x3
270 #define TTBCR_RGN_NC                    0
271 #define TTBCR_RGN_WBWA                  1
272 #define TTBCR_RGN_WT                    2
273 #define TTBCR_RGN_WB                    3
274
275 #define TTBCR_SL0_SHIFT                 6
276 #define TTBCR_SL0_MASK                  0x3
277 #define TTBCR_SL0_LVL_2                 0
278 #define TTBCR_SL0_LVL_1                 1
279
280 #define TTBCR_T1SZ_SHIFT                16
281 #define TTBCR_T0SZ_SHIFT                0
282 #define TTBCR_SZ_MASK                   0xf
283
284 #define TTBCR2_SEP_SHIFT                15
285 #define TTBCR2_SEP_MASK                 0x7
286
287 #define TTBCR2_PASIZE_SHIFT             0
288 #define TTBCR2_PASIZE_MASK              0x7
289
290 /* Common definitions for PASize and SEP fields */
291 #define TTBCR2_ADDR_32                  0
292 #define TTBCR2_ADDR_36                  1
293 #define TTBCR2_ADDR_40                  2
294 #define TTBCR2_ADDR_42                  3
295 #define TTBCR2_ADDR_44                  4
296 #define TTBCR2_ADDR_48                  5
297
298 #define TTBRn_HI_ASID_SHIFT             16
299
300 #define MAIR_ATTR_SHIFT(n)              ((n) << 3)
301 #define MAIR_ATTR_MASK                  0xff
302 #define MAIR_ATTR_DEVICE                0x04
303 #define MAIR_ATTR_NC                    0x44
304 #define MAIR_ATTR_WBRWA                 0xff
305 #define MAIR_ATTR_IDX_NC                0
306 #define MAIR_ATTR_IDX_CACHE             1
307 #define MAIR_ATTR_IDX_DEV               2
308
309 #define FSR_MULTI                       (1 << 31)
310 #define FSR_SS                          (1 << 30)
311 #define FSR_UUT                         (1 << 8)
312 #define FSR_ASF                         (1 << 7)
313 #define FSR_TLBLKF                      (1 << 6)
314 #define FSR_TLBMCF                      (1 << 5)
315 #define FSR_EF                          (1 << 4)
316 #define FSR_PF                          (1 << 3)
317 #define FSR_AFF                         (1 << 2)
318 #define FSR_TF                          (1 << 1)
319
320 #define FSR_IGN                         (FSR_AFF | FSR_ASF | \
321                                          FSR_TLBMCF | FSR_TLBLKF)
322 #define FSR_FAULT                       (FSR_MULTI | FSR_SS | FSR_UUT | \
323                                          FSR_EF | FSR_PF | FSR_TF | FSR_IGN)
324
325 #define FSYNR0_WNR                      (1 << 4)
326
327 struct arm_smmu_smr {
328         u8                              idx;
329         u16                             mask;
330         u16                             id;
331 };
332
333 struct arm_smmu_master_cfg {
334         int                             num_streamids;
335         u16                             streamids[MAX_MASTER_STREAMIDS];
336         struct arm_smmu_smr             *smrs;
337 };
338
339 struct arm_smmu_master {
340         struct device_node              *of_node;
341         struct rb_node                  node;
342         struct arm_smmu_master_cfg      cfg;
343 };
344
345 struct arm_smmu_device {
346         struct device                   *dev;
347
348         void __iomem                    *base;
349         unsigned long                   size;
350         unsigned long                   pagesize;
351
352 #define ARM_SMMU_FEAT_COHERENT_WALK     (1 << 0)
353 #define ARM_SMMU_FEAT_STREAM_MATCH      (1 << 1)
354 #define ARM_SMMU_FEAT_TRANS_S1          (1 << 2)
355 #define ARM_SMMU_FEAT_TRANS_S2          (1 << 3)
356 #define ARM_SMMU_FEAT_TRANS_NESTED      (1 << 4)
357         u32                             features;
358
359 #define ARM_SMMU_OPT_SECURE_CFG_ACCESS (1 << 0)
360         u32                             options;
361         int                             version;
362
363         u32                             num_context_banks;
364         u32                             num_s2_context_banks;
365         DECLARE_BITMAP(context_map, ARM_SMMU_MAX_CBS);
366         atomic_t                        irptndx;
367
368         u32                             num_mapping_groups;
369         DECLARE_BITMAP(smr_map, ARM_SMMU_MAX_SMRS);
370
371         unsigned long                   input_size;
372         unsigned long                   s1_output_size;
373         unsigned long                   s2_output_size;
374
375         u32                             num_global_irqs;
376         u32                             num_context_irqs;
377         unsigned int                    *irqs;
378
379         struct list_head                list;
380         struct rb_root                  masters;
381 };
382
383 struct arm_smmu_cfg {
384         u8                              cbndx;
385         u8                              irptndx;
386         u32                             cbar;
387         pgd_t                           *pgd;
388 };
389 #define INVALID_IRPTNDX                 0xff
390
391 #define ARM_SMMU_CB_ASID(cfg)           ((cfg)->cbndx)
392 #define ARM_SMMU_CB_VMID(cfg)           ((cfg)->cbndx + 1)
393
394 struct arm_smmu_domain {
395         struct arm_smmu_device          *smmu;
396         struct arm_smmu_cfg             cfg;
397         spinlock_t                      lock;
398 };
399
400 static DEFINE_SPINLOCK(arm_smmu_devices_lock);
401 static LIST_HEAD(arm_smmu_devices);
402
403 struct arm_smmu_option_prop {
404         u32 opt;
405         const char *prop;
406 };
407
408 static struct arm_smmu_option_prop arm_smmu_options[] = {
409         { ARM_SMMU_OPT_SECURE_CFG_ACCESS, "calxeda,smmu-secure-config-access" },
410         { 0, NULL},
411 };
412
413 static void parse_driver_options(struct arm_smmu_device *smmu)
414 {
415         int i = 0;
416
417         do {
418                 if (of_property_read_bool(smmu->dev->of_node,
419                                                 arm_smmu_options[i].prop)) {
420                         smmu->options |= arm_smmu_options[i].opt;
421                         dev_notice(smmu->dev, "option %s\n",
422                                 arm_smmu_options[i].prop);
423                 }
424         } while (arm_smmu_options[++i].opt);
425 }
426
427 static struct device *dev_get_master_dev(struct device *dev)
428 {
429         if (dev_is_pci(dev)) {
430                 struct pci_bus *bus = to_pci_dev(dev)->bus;
431
432                 while (!pci_is_root_bus(bus))
433                         bus = bus->parent;
434                 return bus->bridge->parent;
435         }
436
437         return dev;
438 }
439
440 static struct arm_smmu_master *find_smmu_master(struct arm_smmu_device *smmu,
441                                                 struct device_node *dev_node)
442 {
443         struct rb_node *node = smmu->masters.rb_node;
444
445         while (node) {
446                 struct arm_smmu_master *master;
447
448                 master = container_of(node, struct arm_smmu_master, node);
449
450                 if (dev_node < master->of_node)
451                         node = node->rb_left;
452                 else if (dev_node > master->of_node)
453                         node = node->rb_right;
454                 else
455                         return master;
456         }
457
458         return NULL;
459 }
460
461 static struct arm_smmu_master_cfg *
462 find_smmu_master_cfg(struct arm_smmu_device *smmu, struct device *dev)
463 {
464         struct arm_smmu_master *master;
465
466         if (dev_is_pci(dev))
467                 return dev->archdata.iommu;
468
469         master = find_smmu_master(smmu, dev->of_node);
470         return master ? &master->cfg : NULL;
471 }
472
473 static int insert_smmu_master(struct arm_smmu_device *smmu,
474                               struct arm_smmu_master *master)
475 {
476         struct rb_node **new, *parent;
477
478         new = &smmu->masters.rb_node;
479         parent = NULL;
480         while (*new) {
481                 struct arm_smmu_master *this
482                         = container_of(*new, struct arm_smmu_master, node);
483
484                 parent = *new;
485                 if (master->of_node < this->of_node)
486                         new = &((*new)->rb_left);
487                 else if (master->of_node > this->of_node)
488                         new = &((*new)->rb_right);
489                 else
490                         return -EEXIST;
491         }
492
493         rb_link_node(&master->node, parent, new);
494         rb_insert_color(&master->node, &smmu->masters);
495         return 0;
496 }
497
498 static int register_smmu_master(struct arm_smmu_device *smmu,
499                                 struct device *dev,
500                                 struct of_phandle_args *masterspec)
501 {
502         int i;
503         struct arm_smmu_master *master;
504
505         master = find_smmu_master(smmu, masterspec->np);
506         if (master) {
507                 dev_err(dev,
508                         "rejecting multiple registrations for master device %s\n",
509                         masterspec->np->name);
510                 return -EBUSY;
511         }
512
513         if (masterspec->args_count > MAX_MASTER_STREAMIDS) {
514                 dev_err(dev,
515                         "reached maximum number (%d) of stream IDs for master device %s\n",
516                         MAX_MASTER_STREAMIDS, masterspec->np->name);
517                 return -ENOSPC;
518         }
519
520         master = devm_kzalloc(dev, sizeof(*master), GFP_KERNEL);
521         if (!master)
522                 return -ENOMEM;
523
524         master->of_node                 = masterspec->np;
525         master->cfg.num_streamids       = masterspec->args_count;
526
527         for (i = 0; i < master->cfg.num_streamids; ++i)
528                 master->cfg.streamids[i] = masterspec->args[i];
529
530         return insert_smmu_master(smmu, master);
531 }
532
533 static struct arm_smmu_device *find_smmu_for_device(struct device *dev)
534 {
535         struct arm_smmu_device *smmu;
536         struct arm_smmu_master *master = NULL;
537         struct device_node *dev_node = dev_get_master_dev(dev)->of_node;
538
539         spin_lock(&arm_smmu_devices_lock);
540         list_for_each_entry(smmu, &arm_smmu_devices, list) {
541                 master = find_smmu_master(smmu, dev_node);
542                 if (master)
543                         break;
544         }
545         spin_unlock(&arm_smmu_devices_lock);
546
547         return master ? smmu : NULL;
548 }
549
550 static int __arm_smmu_alloc_bitmap(unsigned long *map, int start, int end)
551 {
552         int idx;
553
554         do {
555                 idx = find_next_zero_bit(map, end, start);
556                 if (idx == end)
557                         return -ENOSPC;
558         } while (test_and_set_bit(idx, map));
559
560         return idx;
561 }
562
563 static void __arm_smmu_free_bitmap(unsigned long *map, int idx)
564 {
565         clear_bit(idx, map);
566 }
567
568 /* Wait for any pending TLB invalidations to complete */
569 static void arm_smmu_tlb_sync(struct arm_smmu_device *smmu)
570 {
571         int count = 0;
572         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
573
574         writel_relaxed(0, gr0_base + ARM_SMMU_GR0_sTLBGSYNC);
575         while (readl_relaxed(gr0_base + ARM_SMMU_GR0_sTLBGSTATUS)
576                & sTLBGSTATUS_GSACTIVE) {
577                 cpu_relax();
578                 if (++count == TLB_LOOP_TIMEOUT) {
579                         dev_err_ratelimited(smmu->dev,
580                         "TLB sync timed out -- SMMU may be deadlocked\n");
581                         return;
582                 }
583                 udelay(1);
584         }
585 }
586
587 static void arm_smmu_tlb_inv_context(struct arm_smmu_domain *smmu_domain)
588 {
589         struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
590         struct arm_smmu_device *smmu = smmu_domain->smmu;
591         void __iomem *base = ARM_SMMU_GR0(smmu);
592         bool stage1 = cfg->cbar != CBAR_TYPE_S2_TRANS;
593
594         if (stage1) {
595                 base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
596                 writel_relaxed(ARM_SMMU_CB_ASID(cfg),
597                                base + ARM_SMMU_CB_S1_TLBIASID);
598         } else {
599                 base = ARM_SMMU_GR0(smmu);
600                 writel_relaxed(ARM_SMMU_CB_VMID(cfg),
601                                base + ARM_SMMU_GR0_TLBIVMID);
602         }
603
604         arm_smmu_tlb_sync(smmu);
605 }
606
607 static irqreturn_t arm_smmu_context_fault(int irq, void *dev)
608 {
609         int flags, ret;
610         u32 fsr, far, fsynr, resume;
611         unsigned long iova;
612         struct iommu_domain *domain = dev;
613         struct arm_smmu_domain *smmu_domain = domain->priv;
614         struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
615         struct arm_smmu_device *smmu = smmu_domain->smmu;
616         void __iomem *cb_base;
617
618         cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
619         fsr = readl_relaxed(cb_base + ARM_SMMU_CB_FSR);
620
621         if (!(fsr & FSR_FAULT))
622                 return IRQ_NONE;
623
624         if (fsr & FSR_IGN)
625                 dev_err_ratelimited(smmu->dev,
626                                     "Unexpected context fault (fsr 0x%u)\n",
627                                     fsr);
628
629         fsynr = readl_relaxed(cb_base + ARM_SMMU_CB_FSYNR0);
630         flags = fsynr & FSYNR0_WNR ? IOMMU_FAULT_WRITE : IOMMU_FAULT_READ;
631
632         far = readl_relaxed(cb_base + ARM_SMMU_CB_FAR_LO);
633         iova = far;
634 #ifdef CONFIG_64BIT
635         far = readl_relaxed(cb_base + ARM_SMMU_CB_FAR_HI);
636         iova |= ((unsigned long)far << 32);
637 #endif
638
639         if (!report_iommu_fault(domain, smmu->dev, iova, flags)) {
640                 ret = IRQ_HANDLED;
641                 resume = RESUME_RETRY;
642         } else {
643                 dev_err_ratelimited(smmu->dev,
644                     "Unhandled context fault: iova=0x%08lx, fsynr=0x%x, cb=%d\n",
645                     iova, fsynr, cfg->cbndx);
646                 ret = IRQ_NONE;
647                 resume = RESUME_TERMINATE;
648         }
649
650         /* Clear the faulting FSR */
651         writel(fsr, cb_base + ARM_SMMU_CB_FSR);
652
653         /* Retry or terminate any stalled transactions */
654         if (fsr & FSR_SS)
655                 writel_relaxed(resume, cb_base + ARM_SMMU_CB_RESUME);
656
657         return ret;
658 }
659
660 static irqreturn_t arm_smmu_global_fault(int irq, void *dev)
661 {
662         u32 gfsr, gfsynr0, gfsynr1, gfsynr2;
663         struct arm_smmu_device *smmu = dev;
664         void __iomem *gr0_base = ARM_SMMU_GR0_NS(smmu);
665
666         gfsr = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSR);
667         gfsynr0 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSYNR0);
668         gfsynr1 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSYNR1);
669         gfsynr2 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSYNR2);
670
671         if (!gfsr)
672                 return IRQ_NONE;
673
674         dev_err_ratelimited(smmu->dev,
675                 "Unexpected global fault, this could be serious\n");
676         dev_err_ratelimited(smmu->dev,
677                 "\tGFSR 0x%08x, GFSYNR0 0x%08x, GFSYNR1 0x%08x, GFSYNR2 0x%08x\n",
678                 gfsr, gfsynr0, gfsynr1, gfsynr2);
679
680         writel(gfsr, gr0_base + ARM_SMMU_GR0_sGFSR);
681         return IRQ_HANDLED;
682 }
683
684 static void arm_smmu_flush_pgtable(struct arm_smmu_device *smmu, void *addr,
685                                    size_t size)
686 {
687         unsigned long offset = (unsigned long)addr & ~PAGE_MASK;
688
689
690         /* Ensure new page tables are visible to the hardware walker */
691         if (smmu->features & ARM_SMMU_FEAT_COHERENT_WALK) {
692                 dsb(ishst);
693         } else {
694                 /*
695                  * If the SMMU can't walk tables in the CPU caches, treat them
696                  * like non-coherent DMA since we need to flush the new entries
697                  * all the way out to memory. There's no possibility of
698                  * recursion here as the SMMU table walker will not be wired
699                  * through another SMMU.
700                  */
701                 dma_map_page(smmu->dev, virt_to_page(addr), offset, size,
702                                 DMA_TO_DEVICE);
703         }
704 }
705
706 static void arm_smmu_init_context_bank(struct arm_smmu_domain *smmu_domain)
707 {
708         u32 reg;
709         bool stage1;
710         struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
711         struct arm_smmu_device *smmu = smmu_domain->smmu;
712         void __iomem *cb_base, *gr0_base, *gr1_base;
713
714         gr0_base = ARM_SMMU_GR0(smmu);
715         gr1_base = ARM_SMMU_GR1(smmu);
716         stage1 = cfg->cbar != CBAR_TYPE_S2_TRANS;
717         cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
718
719         /* CBAR */
720         reg = cfg->cbar;
721         if (smmu->version == 1)
722                 reg |= cfg->irptndx << CBAR_IRPTNDX_SHIFT;
723
724         /*
725          * Use the weakest shareability/memory types, so they are
726          * overridden by the ttbcr/pte.
727          */
728         if (stage1) {
729                 reg |= (CBAR_S1_BPSHCFG_NSH << CBAR_S1_BPSHCFG_SHIFT) |
730                         (CBAR_S1_MEMATTR_WB << CBAR_S1_MEMATTR_SHIFT);
731         } else {
732                 reg |= ARM_SMMU_CB_VMID(cfg) << CBAR_VMID_SHIFT;
733         }
734         writel_relaxed(reg, gr1_base + ARM_SMMU_GR1_CBAR(cfg->cbndx));
735
736         if (smmu->version > 1) {
737                 /* CBA2R */
738 #ifdef CONFIG_64BIT
739                 reg = CBA2R_RW64_64BIT;
740 #else
741                 reg = CBA2R_RW64_32BIT;
742 #endif
743                 writel_relaxed(reg,
744                                gr1_base + ARM_SMMU_GR1_CBA2R(cfg->cbndx));
745
746                 /* TTBCR2 */
747                 switch (smmu->input_size) {
748                 case 32:
749                         reg = (TTBCR2_ADDR_32 << TTBCR2_SEP_SHIFT);
750                         break;
751                 case 36:
752                         reg = (TTBCR2_ADDR_36 << TTBCR2_SEP_SHIFT);
753                         break;
754                 case 39:
755                         reg = (TTBCR2_ADDR_40 << TTBCR2_SEP_SHIFT);
756                         break;
757                 case 42:
758                         reg = (TTBCR2_ADDR_42 << TTBCR2_SEP_SHIFT);
759                         break;
760                 case 44:
761                         reg = (TTBCR2_ADDR_44 << TTBCR2_SEP_SHIFT);
762                         break;
763                 case 48:
764                         reg = (TTBCR2_ADDR_48 << TTBCR2_SEP_SHIFT);
765                         break;
766                 }
767
768                 switch (smmu->s1_output_size) {
769                 case 32:
770                         reg |= (TTBCR2_ADDR_32 << TTBCR2_PASIZE_SHIFT);
771                         break;
772                 case 36:
773                         reg |= (TTBCR2_ADDR_36 << TTBCR2_PASIZE_SHIFT);
774                         break;
775                 case 39:
776                         reg |= (TTBCR2_ADDR_40 << TTBCR2_PASIZE_SHIFT);
777                         break;
778                 case 42:
779                         reg |= (TTBCR2_ADDR_42 << TTBCR2_PASIZE_SHIFT);
780                         break;
781                 case 44:
782                         reg |= (TTBCR2_ADDR_44 << TTBCR2_PASIZE_SHIFT);
783                         break;
784                 case 48:
785                         reg |= (TTBCR2_ADDR_48 << TTBCR2_PASIZE_SHIFT);
786                         break;
787                 }
788
789                 if (stage1)
790                         writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBCR2);
791         }
792
793         /* TTBR0 */
794         arm_smmu_flush_pgtable(smmu, cfg->pgd,
795                                PTRS_PER_PGD * sizeof(pgd_t));
796         reg = __pa(cfg->pgd);
797         writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBR0_LO);
798         reg = (phys_addr_t)__pa(cfg->pgd) >> 32;
799         if (stage1)
800                 reg |= ARM_SMMU_CB_ASID(cfg) << TTBRn_HI_ASID_SHIFT;
801         writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBR0_HI);
802
803         /*
804          * TTBCR
805          * We use long descriptor, with inner-shareable WBWA tables in TTBR0.
806          */
807         if (smmu->version > 1) {
808                 if (PAGE_SIZE == SZ_4K)
809                         reg = TTBCR_TG0_4K;
810                 else
811                         reg = TTBCR_TG0_64K;
812
813                 if (!stage1) {
814                         reg |= (64 - smmu->s1_output_size) << TTBCR_T0SZ_SHIFT;
815
816                         switch (smmu->s2_output_size) {
817                         case 32:
818                                 reg |= (TTBCR2_ADDR_32 << TTBCR_PASIZE_SHIFT);
819                                 break;
820                         case 36:
821                                 reg |= (TTBCR2_ADDR_36 << TTBCR_PASIZE_SHIFT);
822                                 break;
823                         case 40:
824                                 reg |= (TTBCR2_ADDR_40 << TTBCR_PASIZE_SHIFT);
825                                 break;
826                         case 42:
827                                 reg |= (TTBCR2_ADDR_42 << TTBCR_PASIZE_SHIFT);
828                                 break;
829                         case 44:
830                                 reg |= (TTBCR2_ADDR_44 << TTBCR_PASIZE_SHIFT);
831                                 break;
832                         case 48:
833                                 reg |= (TTBCR2_ADDR_48 << TTBCR_PASIZE_SHIFT);
834                                 break;
835                         }
836                 } else {
837                         reg |= (64 - smmu->input_size) << TTBCR_T0SZ_SHIFT;
838                 }
839         } else {
840                 reg = 0;
841         }
842
843         reg |= TTBCR_EAE |
844               (TTBCR_SH_IS << TTBCR_SH0_SHIFT) |
845               (TTBCR_RGN_WBWA << TTBCR_ORGN0_SHIFT) |
846               (TTBCR_RGN_WBWA << TTBCR_IRGN0_SHIFT) |
847               (TTBCR_SL0_LVL_1 << TTBCR_SL0_SHIFT);
848         writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBCR);
849
850         /* MAIR0 (stage-1 only) */
851         if (stage1) {
852                 reg = (MAIR_ATTR_NC << MAIR_ATTR_SHIFT(MAIR_ATTR_IDX_NC)) |
853                       (MAIR_ATTR_WBRWA << MAIR_ATTR_SHIFT(MAIR_ATTR_IDX_CACHE)) |
854                       (MAIR_ATTR_DEVICE << MAIR_ATTR_SHIFT(MAIR_ATTR_IDX_DEV));
855                 writel_relaxed(reg, cb_base + ARM_SMMU_CB_S1_MAIR0);
856         }
857
858         /* SCTLR */
859         reg = SCTLR_CFCFG | SCTLR_CFIE | SCTLR_CFRE | SCTLR_M | SCTLR_EAE_SBOP;
860         if (stage1)
861                 reg |= SCTLR_S1_ASIDPNE;
862 #ifdef __BIG_ENDIAN
863         reg |= SCTLR_E;
864 #endif
865         writel_relaxed(reg, cb_base + ARM_SMMU_CB_SCTLR);
866 }
867
868 static int arm_smmu_init_domain_context(struct iommu_domain *domain,
869                                         struct arm_smmu_device *smmu)
870 {
871         int irq, ret, start;
872         struct arm_smmu_domain *smmu_domain = domain->priv;
873         struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
874
875         if (smmu->features & ARM_SMMU_FEAT_TRANS_NESTED) {
876                 /*
877                  * We will likely want to change this if/when KVM gets
878                  * involved.
879                  */
880                 cfg->cbar = CBAR_TYPE_S1_TRANS_S2_BYPASS;
881                 start = smmu->num_s2_context_banks;
882         } else if (smmu->features & ARM_SMMU_FEAT_TRANS_S1) {
883                 cfg->cbar = CBAR_TYPE_S1_TRANS_S2_BYPASS;
884                 start = smmu->num_s2_context_banks;
885         } else {
886                 cfg->cbar = CBAR_TYPE_S2_TRANS;
887                 start = 0;
888         }
889
890         ret = __arm_smmu_alloc_bitmap(smmu->context_map, start,
891                                       smmu->num_context_banks);
892         if (IS_ERR_VALUE(ret))
893                 return ret;
894
895         cfg->cbndx = ret;
896         if (smmu->version == 1) {
897                 cfg->irptndx = atomic_inc_return(&smmu->irptndx);
898                 cfg->irptndx %= smmu->num_context_irqs;
899         } else {
900                 cfg->irptndx = cfg->cbndx;
901         }
902
903         irq = smmu->irqs[smmu->num_global_irqs + cfg->irptndx];
904         ret = request_irq(irq, arm_smmu_context_fault, IRQF_SHARED,
905                           "arm-smmu-context-fault", domain);
906         if (IS_ERR_VALUE(ret)) {
907                 dev_err(smmu->dev, "failed to request context IRQ %d (%u)\n",
908                         cfg->irptndx, irq);
909                 cfg->irptndx = INVALID_IRPTNDX;
910                 goto out_free_context;
911         }
912
913         smmu_domain->smmu = smmu;
914         arm_smmu_init_context_bank(smmu_domain);
915         return 0;
916
917 out_free_context:
918         __arm_smmu_free_bitmap(smmu->context_map, cfg->cbndx);
919         return ret;
920 }
921
922 static void arm_smmu_destroy_domain_context(struct iommu_domain *domain)
923 {
924         struct arm_smmu_domain *smmu_domain = domain->priv;
925         struct arm_smmu_device *smmu = smmu_domain->smmu;
926         struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
927         void __iomem *cb_base;
928         int irq;
929
930         if (!smmu)
931                 return;
932
933         /* Disable the context bank and nuke the TLB before freeing it. */
934         cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
935         writel_relaxed(0, cb_base + ARM_SMMU_CB_SCTLR);
936         arm_smmu_tlb_inv_context(smmu_domain);
937
938         if (cfg->irptndx != INVALID_IRPTNDX) {
939                 irq = smmu->irqs[smmu->num_global_irqs + cfg->irptndx];
940                 free_irq(irq, domain);
941         }
942
943         __arm_smmu_free_bitmap(smmu->context_map, cfg->cbndx);
944 }
945
946 static int arm_smmu_domain_init(struct iommu_domain *domain)
947 {
948         struct arm_smmu_domain *smmu_domain;
949         pgd_t *pgd;
950
951         /*
952          * Allocate the domain and initialise some of its data structures.
953          * We can't really do anything meaningful until we've added a
954          * master.
955          */
956         smmu_domain = kzalloc(sizeof(*smmu_domain), GFP_KERNEL);
957         if (!smmu_domain)
958                 return -ENOMEM;
959
960         pgd = kcalloc(PTRS_PER_PGD, sizeof(pgd_t), GFP_KERNEL);
961         if (!pgd)
962                 goto out_free_domain;
963         smmu_domain->cfg.pgd = pgd;
964
965         spin_lock_init(&smmu_domain->lock);
966         domain->priv = smmu_domain;
967         return 0;
968
969 out_free_domain:
970         kfree(smmu_domain);
971         return -ENOMEM;
972 }
973
974 static void arm_smmu_free_ptes(pmd_t *pmd)
975 {
976         pgtable_t table = pmd_pgtable(*pmd);
977
978         pgtable_page_dtor(table);
979         __free_page(table);
980 }
981
982 static void arm_smmu_free_pmds(pud_t *pud)
983 {
984         int i;
985         pmd_t *pmd, *pmd_base = pmd_offset(pud, 0);
986
987         pmd = pmd_base;
988         for (i = 0; i < PTRS_PER_PMD; ++i) {
989                 if (pmd_none(*pmd))
990                         continue;
991
992                 arm_smmu_free_ptes(pmd);
993                 pmd++;
994         }
995
996         pmd_free(NULL, pmd_base);
997 }
998
999 static void arm_smmu_free_puds(pgd_t *pgd)
1000 {
1001         int i;
1002         pud_t *pud, *pud_base = pud_offset(pgd, 0);
1003
1004         pud = pud_base;
1005         for (i = 0; i < PTRS_PER_PUD; ++i) {
1006                 if (pud_none(*pud))
1007                         continue;
1008
1009                 arm_smmu_free_pmds(pud);
1010                 pud++;
1011         }
1012
1013         pud_free(NULL, pud_base);
1014 }
1015
1016 static void arm_smmu_free_pgtables(struct arm_smmu_domain *smmu_domain)
1017 {
1018         int i;
1019         struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
1020         pgd_t *pgd, *pgd_base = cfg->pgd;
1021
1022         /*
1023          * Recursively free the page tables for this domain. We don't
1024          * care about speculative TLB filling because the tables should
1025          * not be active in any context bank at this point (SCTLR.M is 0).
1026          */
1027         pgd = pgd_base;
1028         for (i = 0; i < PTRS_PER_PGD; ++i) {
1029                 if (pgd_none(*pgd))
1030                         continue;
1031                 arm_smmu_free_puds(pgd);
1032                 pgd++;
1033         }
1034
1035         kfree(pgd_base);
1036 }
1037
1038 static void arm_smmu_domain_destroy(struct iommu_domain *domain)
1039 {
1040         struct arm_smmu_domain *smmu_domain = domain->priv;
1041
1042         /*
1043          * Free the domain resources. We assume that all devices have
1044          * already been detached.
1045          */
1046         arm_smmu_destroy_domain_context(domain);
1047         arm_smmu_free_pgtables(smmu_domain);
1048         kfree(smmu_domain);
1049 }
1050
1051 static int arm_smmu_master_configure_smrs(struct arm_smmu_device *smmu,
1052                                           struct arm_smmu_master_cfg *cfg)
1053 {
1054         int i;
1055         struct arm_smmu_smr *smrs;
1056         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1057
1058         if (!(smmu->features & ARM_SMMU_FEAT_STREAM_MATCH))
1059                 return 0;
1060
1061         if (cfg->smrs)
1062                 return -EEXIST;
1063
1064         smrs = kmalloc_array(cfg->num_streamids, sizeof(*smrs), GFP_KERNEL);
1065         if (!smrs) {
1066                 dev_err(smmu->dev, "failed to allocate %d SMRs\n",
1067                         cfg->num_streamids);
1068                 return -ENOMEM;
1069         }
1070
1071         /* Allocate the SMRs on the SMMU */
1072         for (i = 0; i < cfg->num_streamids; ++i) {
1073                 int idx = __arm_smmu_alloc_bitmap(smmu->smr_map, 0,
1074                                                   smmu->num_mapping_groups);
1075                 if (IS_ERR_VALUE(idx)) {
1076                         dev_err(smmu->dev, "failed to allocate free SMR\n");
1077                         goto err_free_smrs;
1078                 }
1079
1080                 smrs[i] = (struct arm_smmu_smr) {
1081                         .idx    = idx,
1082                         .mask   = 0, /* We don't currently share SMRs */
1083                         .id     = cfg->streamids[i],
1084                 };
1085         }
1086
1087         /* It worked! Now, poke the actual hardware */
1088         for (i = 0; i < cfg->num_streamids; ++i) {
1089                 u32 reg = SMR_VALID | smrs[i].id << SMR_ID_SHIFT |
1090                           smrs[i].mask << SMR_MASK_SHIFT;
1091                 writel_relaxed(reg, gr0_base + ARM_SMMU_GR0_SMR(smrs[i].idx));
1092         }
1093
1094         cfg->smrs = smrs;
1095         return 0;
1096
1097 err_free_smrs:
1098         while (--i >= 0)
1099                 __arm_smmu_free_bitmap(smmu->smr_map, smrs[i].idx);
1100         kfree(smrs);
1101         return -ENOSPC;
1102 }
1103
1104 static void arm_smmu_master_free_smrs(struct arm_smmu_device *smmu,
1105                                       struct arm_smmu_master_cfg *cfg)
1106 {
1107         int i;
1108         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1109         struct arm_smmu_smr *smrs = cfg->smrs;
1110
1111         /* Invalidate the SMRs before freeing back to the allocator */
1112         for (i = 0; i < cfg->num_streamids; ++i) {
1113                 u8 idx = smrs[i].idx;
1114
1115                 writel_relaxed(~SMR_VALID, gr0_base + ARM_SMMU_GR0_SMR(idx));
1116                 __arm_smmu_free_bitmap(smmu->smr_map, idx);
1117         }
1118
1119         cfg->smrs = NULL;
1120         kfree(smrs);
1121 }
1122
1123 static void arm_smmu_bypass_stream_mapping(struct arm_smmu_device *smmu,
1124                                            struct arm_smmu_master_cfg *cfg)
1125 {
1126         int i;
1127         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1128
1129         for (i = 0; i < cfg->num_streamids; ++i) {
1130                 u16 sid = cfg->streamids[i];
1131
1132                 writel_relaxed(S2CR_TYPE_BYPASS,
1133                                gr0_base + ARM_SMMU_GR0_S2CR(sid));
1134         }
1135 }
1136
1137 static int arm_smmu_domain_add_master(struct arm_smmu_domain *smmu_domain,
1138                                       struct arm_smmu_master_cfg *cfg)
1139 {
1140         int i, ret;
1141         struct arm_smmu_device *smmu = smmu_domain->smmu;
1142         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1143
1144         ret = arm_smmu_master_configure_smrs(smmu, cfg);
1145         if (ret)
1146                 return ret;
1147
1148         for (i = 0; i < cfg->num_streamids; ++i) {
1149                 u32 idx, s2cr;
1150
1151                 idx = cfg->smrs ? cfg->smrs[i].idx : cfg->streamids[i];
1152                 s2cr = S2CR_TYPE_TRANS |
1153                        (smmu_domain->cfg.cbndx << S2CR_CBNDX_SHIFT);
1154                 writel_relaxed(s2cr, gr0_base + ARM_SMMU_GR0_S2CR(idx));
1155         }
1156
1157         return 0;
1158 }
1159
1160 static void arm_smmu_domain_remove_master(struct arm_smmu_domain *smmu_domain,
1161                                           struct arm_smmu_master_cfg *cfg)
1162 {
1163         struct arm_smmu_device *smmu = smmu_domain->smmu;
1164
1165         /*
1166          * We *must* clear the S2CR first, because freeing the SMR means
1167          * that it can be re-allocated immediately.
1168          */
1169         arm_smmu_bypass_stream_mapping(smmu, cfg);
1170         arm_smmu_master_free_smrs(smmu, cfg);
1171 }
1172
1173 static int arm_smmu_attach_dev(struct iommu_domain *domain, struct device *dev)
1174 {
1175         int ret = -EINVAL;
1176         struct arm_smmu_domain *smmu_domain = domain->priv;
1177         struct arm_smmu_device *smmu;
1178         struct arm_smmu_master_cfg *cfg;
1179         unsigned long flags;
1180
1181         smmu = dev_get_master_dev(dev)->archdata.iommu;
1182         if (!smmu) {
1183                 dev_err(dev, "cannot attach to SMMU, is it on the same bus?\n");
1184                 return -ENXIO;
1185         }
1186
1187         /*
1188          * Sanity check the domain. We don't support domains across
1189          * different SMMUs.
1190          */
1191         spin_lock_irqsave(&smmu_domain->lock, flags);
1192         if (!smmu_domain->smmu) {
1193                 /* Now that we have a master, we can finalise the domain */
1194                 ret = arm_smmu_init_domain_context(domain, smmu);
1195                 if (IS_ERR_VALUE(ret))
1196                         goto err_unlock;
1197         } else if (smmu_domain->smmu != smmu) {
1198                 dev_err(dev,
1199                         "cannot attach to SMMU %s whilst already attached to domain on SMMU %s\n",
1200                         dev_name(smmu_domain->smmu->dev),
1201                         dev_name(smmu->dev));
1202                 goto err_unlock;
1203         }
1204         spin_unlock_irqrestore(&smmu_domain->lock, flags);
1205
1206         /* Looks ok, so add the device to the domain */
1207         cfg = find_smmu_master_cfg(smmu_domain->smmu, dev);
1208         if (!cfg)
1209                 return -ENODEV;
1210
1211         return arm_smmu_domain_add_master(smmu_domain, cfg);
1212
1213 err_unlock:
1214         spin_unlock_irqrestore(&smmu_domain->lock, flags);
1215         return ret;
1216 }
1217
1218 static void arm_smmu_detach_dev(struct iommu_domain *domain, struct device *dev)
1219 {
1220         struct arm_smmu_domain *smmu_domain = domain->priv;
1221         struct arm_smmu_master_cfg *cfg;
1222
1223         cfg = find_smmu_master_cfg(smmu_domain->smmu, dev);
1224         if (cfg)
1225                 arm_smmu_domain_remove_master(smmu_domain, cfg);
1226 }
1227
1228 static bool arm_smmu_pte_is_contiguous_range(unsigned long addr,
1229                                              unsigned long end)
1230 {
1231         return !(addr & ~ARM_SMMU_PTE_CONT_MASK) &&
1232                 (addr + ARM_SMMU_PTE_CONT_SIZE <= end);
1233 }
1234
1235 static int arm_smmu_alloc_init_pte(struct arm_smmu_device *smmu, pmd_t *pmd,
1236                                    unsigned long addr, unsigned long end,
1237                                    unsigned long pfn, int prot, int stage)
1238 {
1239         pte_t *pte, *start;
1240         pteval_t pteval = ARM_SMMU_PTE_PAGE | ARM_SMMU_PTE_AF | ARM_SMMU_PTE_XN;
1241
1242         if (pmd_none(*pmd)) {
1243                 /* Allocate a new set of tables */
1244                 pgtable_t table = alloc_page(GFP_ATOMIC|__GFP_ZERO);
1245
1246                 if (!table)
1247                         return -ENOMEM;
1248
1249                 arm_smmu_flush_pgtable(smmu, page_address(table), PAGE_SIZE);
1250                 if (!pgtable_page_ctor(table)) {
1251                         __free_page(table);
1252                         return -ENOMEM;
1253                 }
1254                 pmd_populate(NULL, pmd, table);
1255                 arm_smmu_flush_pgtable(smmu, pmd, sizeof(*pmd));
1256         }
1257
1258         if (stage == 1) {
1259                 pteval |= ARM_SMMU_PTE_AP_UNPRIV | ARM_SMMU_PTE_nG;
1260                 if (!(prot & IOMMU_WRITE) && (prot & IOMMU_READ))
1261                         pteval |= ARM_SMMU_PTE_AP_RDONLY;
1262
1263                 if (prot & IOMMU_CACHE)
1264                         pteval |= (MAIR_ATTR_IDX_CACHE <<
1265                                    ARM_SMMU_PTE_ATTRINDX_SHIFT);
1266         } else {
1267                 pteval |= ARM_SMMU_PTE_HAP_FAULT;
1268                 if (prot & IOMMU_READ)
1269                         pteval |= ARM_SMMU_PTE_HAP_READ;
1270                 if (prot & IOMMU_WRITE)
1271                         pteval |= ARM_SMMU_PTE_HAP_WRITE;
1272                 if (prot & IOMMU_CACHE)
1273                         pteval |= ARM_SMMU_PTE_MEMATTR_OIWB;
1274                 else
1275                         pteval |= ARM_SMMU_PTE_MEMATTR_NC;
1276         }
1277
1278         /* If no access, create a faulting entry to avoid TLB fills */
1279         if (prot & IOMMU_EXEC)
1280                 pteval &= ~ARM_SMMU_PTE_XN;
1281         else if (!(prot & (IOMMU_READ | IOMMU_WRITE)))
1282                 pteval &= ~ARM_SMMU_PTE_PAGE;
1283
1284         pteval |= ARM_SMMU_PTE_SH_IS;
1285         start = pmd_page_vaddr(*pmd) + pte_index(addr);
1286         pte = start;
1287
1288         /*
1289          * Install the page table entries. This is fairly complicated
1290          * since we attempt to make use of the contiguous hint in the
1291          * ptes where possible. The contiguous hint indicates a series
1292          * of ARM_SMMU_PTE_CONT_ENTRIES ptes mapping a physically
1293          * contiguous region with the following constraints:
1294          *
1295          *   - The region start is aligned to ARM_SMMU_PTE_CONT_SIZE
1296          *   - Each pte in the region has the contiguous hint bit set
1297          *
1298          * This complicates unmapping (also handled by this code, when
1299          * neither IOMMU_READ or IOMMU_WRITE are set) because it is
1300          * possible, yet highly unlikely, that a client may unmap only
1301          * part of a contiguous range. This requires clearing of the
1302          * contiguous hint bits in the range before installing the new
1303          * faulting entries.
1304          *
1305          * Note that re-mapping an address range without first unmapping
1306          * it is not supported, so TLB invalidation is not required here
1307          * and is instead performed at unmap and domain-init time.
1308          */
1309         do {
1310                 int i = 1;
1311
1312                 pteval &= ~ARM_SMMU_PTE_CONT;
1313
1314                 if (arm_smmu_pte_is_contiguous_range(addr, end)) {
1315                         i = ARM_SMMU_PTE_CONT_ENTRIES;
1316                         pteval |= ARM_SMMU_PTE_CONT;
1317                 } else if (pte_val(*pte) &
1318                            (ARM_SMMU_PTE_CONT | ARM_SMMU_PTE_PAGE)) {
1319                         int j;
1320                         pte_t *cont_start;
1321                         unsigned long idx = pte_index(addr);
1322
1323                         idx &= ~(ARM_SMMU_PTE_CONT_ENTRIES - 1);
1324                         cont_start = pmd_page_vaddr(*pmd) + idx;
1325                         for (j = 0; j < ARM_SMMU_PTE_CONT_ENTRIES; ++j)
1326                                 pte_val(*(cont_start + j)) &=
1327                                         ~ARM_SMMU_PTE_CONT;
1328
1329                         arm_smmu_flush_pgtable(smmu, cont_start,
1330                                                sizeof(*pte) *
1331                                                ARM_SMMU_PTE_CONT_ENTRIES);
1332                 }
1333
1334                 do {
1335                         *pte = pfn_pte(pfn, __pgprot(pteval));
1336                 } while (pte++, pfn++, addr += PAGE_SIZE, --i);
1337         } while (addr != end);
1338
1339         arm_smmu_flush_pgtable(smmu, start, sizeof(*pte) * (pte - start));
1340         return 0;
1341 }
1342
1343 static int arm_smmu_alloc_init_pmd(struct arm_smmu_device *smmu, pud_t *pud,
1344                                    unsigned long addr, unsigned long end,
1345                                    phys_addr_t phys, int prot, int stage)
1346 {
1347         int ret;
1348         pmd_t *pmd;
1349         unsigned long next, pfn = __phys_to_pfn(phys);
1350
1351 #ifndef __PAGETABLE_PMD_FOLDED
1352         if (pud_none(*pud)) {
1353                 pmd = (pmd_t *)get_zeroed_page(GFP_ATOMIC);
1354                 if (!pmd)
1355                         return -ENOMEM;
1356
1357                 arm_smmu_flush_pgtable(smmu, pmd, PAGE_SIZE);
1358                 pud_populate(NULL, pud, pmd);
1359                 arm_smmu_flush_pgtable(smmu, pud, sizeof(*pud));
1360
1361                 pmd += pmd_index(addr);
1362         } else
1363 #endif
1364                 pmd = pmd_offset(pud, addr);
1365
1366         do {
1367                 next = pmd_addr_end(addr, end);
1368                 ret = arm_smmu_alloc_init_pte(smmu, pmd, addr, next, pfn,
1369                                               prot, stage);
1370                 phys += next - addr;
1371         } while (pmd++, addr = next, addr < end);
1372
1373         return ret;
1374 }
1375
1376 static int arm_smmu_alloc_init_pud(struct arm_smmu_device *smmu, pgd_t *pgd,
1377                                    unsigned long addr, unsigned long end,
1378                                    phys_addr_t phys, int prot, int stage)
1379 {
1380         int ret = 0;
1381         pud_t *pud;
1382         unsigned long next;
1383
1384 #ifndef __PAGETABLE_PUD_FOLDED
1385         if (pgd_none(*pgd)) {
1386                 pud = (pud_t *)get_zeroed_page(GFP_ATOMIC);
1387                 if (!pud)
1388                         return -ENOMEM;
1389
1390                 arm_smmu_flush_pgtable(smmu, pud, PAGE_SIZE);
1391                 pgd_populate(NULL, pgd, pud);
1392                 arm_smmu_flush_pgtable(smmu, pgd, sizeof(*pgd));
1393
1394                 pud += pud_index(addr);
1395         } else
1396 #endif
1397                 pud = pud_offset(pgd, addr);
1398
1399         do {
1400                 next = pud_addr_end(addr, end);
1401                 ret = arm_smmu_alloc_init_pmd(smmu, pud, addr, next, phys,
1402                                               prot, stage);
1403                 phys += next - addr;
1404         } while (pud++, addr = next, addr < end);
1405
1406         return ret;
1407 }
1408
1409 static int arm_smmu_handle_mapping(struct arm_smmu_domain *smmu_domain,
1410                                    unsigned long iova, phys_addr_t paddr,
1411                                    size_t size, int prot)
1412 {
1413         int ret, stage;
1414         unsigned long end;
1415         phys_addr_t input_mask, output_mask;
1416         struct arm_smmu_device *smmu = smmu_domain->smmu;
1417         struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
1418         pgd_t *pgd = cfg->pgd;
1419         unsigned long flags;
1420
1421         if (cfg->cbar == CBAR_TYPE_S2_TRANS) {
1422                 stage = 2;
1423                 output_mask = (1ULL << smmu->s2_output_size) - 1;
1424         } else {
1425                 stage = 1;
1426                 output_mask = (1ULL << smmu->s1_output_size) - 1;
1427         }
1428
1429         if (!pgd)
1430                 return -EINVAL;
1431
1432         if (size & ~PAGE_MASK)
1433                 return -EINVAL;
1434
1435         input_mask = (1ULL << smmu->input_size) - 1;
1436         if ((phys_addr_t)iova & ~input_mask)
1437                 return -ERANGE;
1438
1439         if (paddr & ~output_mask)
1440                 return -ERANGE;
1441
1442         spin_lock_irqsave(&smmu_domain->lock, flags);
1443         pgd += pgd_index(iova);
1444         end = iova + size;
1445         do {
1446                 unsigned long next = pgd_addr_end(iova, end);
1447
1448                 ret = arm_smmu_alloc_init_pud(smmu, pgd, iova, next, paddr,
1449                                               prot, stage);
1450                 if (ret)
1451                         goto out_unlock;
1452
1453                 paddr += next - iova;
1454                 iova = next;
1455         } while (pgd++, iova != end);
1456
1457 out_unlock:
1458         spin_unlock_irqrestore(&smmu_domain->lock, flags);
1459
1460         return ret;
1461 }
1462
1463 static int arm_smmu_map(struct iommu_domain *domain, unsigned long iova,
1464                         phys_addr_t paddr, size_t size, int prot)
1465 {
1466         struct arm_smmu_domain *smmu_domain = domain->priv;
1467
1468         if (!smmu_domain)
1469                 return -ENODEV;
1470
1471         return arm_smmu_handle_mapping(smmu_domain, iova, paddr, size, prot);
1472 }
1473
1474 static size_t arm_smmu_unmap(struct iommu_domain *domain, unsigned long iova,
1475                              size_t size)
1476 {
1477         int ret;
1478         struct arm_smmu_domain *smmu_domain = domain->priv;
1479
1480         ret = arm_smmu_handle_mapping(smmu_domain, iova, 0, size, 0);
1481         arm_smmu_tlb_inv_context(smmu_domain);
1482         return ret ? 0 : size;
1483 }
1484
1485 static phys_addr_t arm_smmu_iova_to_phys(struct iommu_domain *domain,
1486                                          dma_addr_t iova)
1487 {
1488         pgd_t *pgdp, pgd;
1489         pud_t pud;
1490         pmd_t pmd;
1491         pte_t pte;
1492         struct arm_smmu_domain *smmu_domain = domain->priv;
1493         struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
1494
1495         pgdp = cfg->pgd;
1496         if (!pgdp)
1497                 return 0;
1498
1499         pgd = *(pgdp + pgd_index(iova));
1500         if (pgd_none(pgd))
1501                 return 0;
1502
1503         pud = *pud_offset(&pgd, iova);
1504         if (pud_none(pud))
1505                 return 0;
1506
1507         pmd = *pmd_offset(&pud, iova);
1508         if (pmd_none(pmd))
1509                 return 0;
1510
1511         pte = *(pmd_page_vaddr(pmd) + pte_index(iova));
1512         if (pte_none(pte))
1513                 return 0;
1514
1515         return __pfn_to_phys(pte_pfn(pte)) | (iova & ~PAGE_MASK);
1516 }
1517
1518 static int arm_smmu_domain_has_cap(struct iommu_domain *domain,
1519                                    unsigned long cap)
1520 {
1521         struct arm_smmu_domain *smmu_domain = domain->priv;
1522         struct arm_smmu_device *smmu = smmu_domain->smmu;
1523         u32 features = smmu ? smmu->features : 0;
1524
1525         switch (cap) {
1526         case IOMMU_CAP_CACHE_COHERENCY:
1527                 return features & ARM_SMMU_FEAT_COHERENT_WALK;
1528         case IOMMU_CAP_INTR_REMAP:
1529                 return 1; /* MSIs are just memory writes */
1530         default:
1531                 return 0;
1532         }
1533 }
1534
1535 static int __arm_smmu_get_pci_sid(struct pci_dev *pdev, u16 alias, void *data)
1536 {
1537         *((u16 *)data) = alias;
1538         return 0; /* Continue walking */
1539 }
1540
1541 static int arm_smmu_add_device(struct device *dev)
1542 {
1543         struct arm_smmu_device *smmu;
1544         struct iommu_group *group;
1545         int ret;
1546
1547         if (dev->archdata.iommu) {
1548                 dev_warn(dev, "IOMMU driver already assigned to device\n");
1549                 return -EINVAL;
1550         }
1551
1552         smmu = find_smmu_for_device(dev);
1553         if (!smmu)
1554                 return -ENODEV;
1555
1556         group = iommu_group_alloc();
1557         if (IS_ERR(group)) {
1558                 dev_err(dev, "Failed to allocate IOMMU group\n");
1559                 return PTR_ERR(group);
1560         }
1561
1562         if (dev_is_pci(dev)) {
1563                 struct arm_smmu_master_cfg *cfg;
1564                 struct pci_dev *pdev = to_pci_dev(dev);
1565
1566                 cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1567                 if (!cfg) {
1568                         ret = -ENOMEM;
1569                         goto out_put_group;
1570                 }
1571
1572                 cfg->num_streamids = 1;
1573                 /*
1574                  * Assume Stream ID == Requester ID for now.
1575                  * We need a way to describe the ID mappings in FDT.
1576                  */
1577                 pci_for_each_dma_alias(pdev, __arm_smmu_get_pci_sid,
1578                                        &cfg->streamids[0]);
1579                 dev->archdata.iommu = cfg;
1580         } else {
1581                 dev->archdata.iommu = smmu;
1582         }
1583
1584         ret = iommu_group_add_device(group, dev);
1585
1586 out_put_group:
1587         iommu_group_put(group);
1588         return ret;
1589 }
1590
1591 static void arm_smmu_remove_device(struct device *dev)
1592 {
1593         if (dev_is_pci(dev))
1594                 kfree(dev->archdata.iommu);
1595
1596         dev->archdata.iommu = NULL;
1597         iommu_group_remove_device(dev);
1598 }
1599
1600 static const struct iommu_ops arm_smmu_ops = {
1601         .domain_init    = arm_smmu_domain_init,
1602         .domain_destroy = arm_smmu_domain_destroy,
1603         .attach_dev     = arm_smmu_attach_dev,
1604         .detach_dev     = arm_smmu_detach_dev,
1605         .map            = arm_smmu_map,
1606         .unmap          = arm_smmu_unmap,
1607         .iova_to_phys   = arm_smmu_iova_to_phys,
1608         .domain_has_cap = arm_smmu_domain_has_cap,
1609         .add_device     = arm_smmu_add_device,
1610         .remove_device  = arm_smmu_remove_device,
1611         .pgsize_bitmap  = (SECTION_SIZE |
1612                            ARM_SMMU_PTE_CONT_SIZE |
1613                            PAGE_SIZE),
1614 };
1615
1616 static void arm_smmu_device_reset(struct arm_smmu_device *smmu)
1617 {
1618         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1619         void __iomem *cb_base;
1620         int i = 0;
1621         u32 reg;
1622
1623         /* clear global FSR */
1624         reg = readl_relaxed(ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sGFSR);
1625         writel(reg, ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sGFSR);
1626
1627         /* Mark all SMRn as invalid and all S2CRn as bypass */
1628         for (i = 0; i < smmu->num_mapping_groups; ++i) {
1629                 writel_relaxed(~SMR_VALID, gr0_base + ARM_SMMU_GR0_SMR(i));
1630                 writel_relaxed(S2CR_TYPE_BYPASS,
1631                         gr0_base + ARM_SMMU_GR0_S2CR(i));
1632         }
1633
1634         /* Make sure all context banks are disabled and clear CB_FSR  */
1635         for (i = 0; i < smmu->num_context_banks; ++i) {
1636                 cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, i);
1637                 writel_relaxed(0, cb_base + ARM_SMMU_CB_SCTLR);
1638                 writel_relaxed(FSR_FAULT, cb_base + ARM_SMMU_CB_FSR);
1639         }
1640
1641         /* Invalidate the TLB, just in case */
1642         writel_relaxed(0, gr0_base + ARM_SMMU_GR0_STLBIALL);
1643         writel_relaxed(0, gr0_base + ARM_SMMU_GR0_TLBIALLH);
1644         writel_relaxed(0, gr0_base + ARM_SMMU_GR0_TLBIALLNSNH);
1645
1646         reg = readl_relaxed(ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sCR0);
1647
1648         /* Enable fault reporting */
1649         reg |= (sCR0_GFRE | sCR0_GFIE | sCR0_GCFGFRE | sCR0_GCFGFIE);
1650
1651         /* Disable TLB broadcasting. */
1652         reg |= (sCR0_VMIDPNE | sCR0_PTM);
1653
1654         /* Enable client access, but bypass when no mapping is found */
1655         reg &= ~(sCR0_CLIENTPD | sCR0_USFCFG);
1656
1657         /* Disable forced broadcasting */
1658         reg &= ~sCR0_FB;
1659
1660         /* Don't upgrade barriers */
1661         reg &= ~(sCR0_BSU_MASK << sCR0_BSU_SHIFT);
1662
1663         /* Push the button */
1664         arm_smmu_tlb_sync(smmu);
1665         writel(reg, ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sCR0);
1666 }
1667
1668 static int arm_smmu_id_size_to_bits(int size)
1669 {
1670         switch (size) {
1671         case 0:
1672                 return 32;
1673         case 1:
1674                 return 36;
1675         case 2:
1676                 return 40;
1677         case 3:
1678                 return 42;
1679         case 4:
1680                 return 44;
1681         case 5:
1682         default:
1683                 return 48;
1684         }
1685 }
1686
1687 static int arm_smmu_device_cfg_probe(struct arm_smmu_device *smmu)
1688 {
1689         unsigned long size;
1690         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1691         u32 id;
1692
1693         dev_notice(smmu->dev, "probing hardware configuration...\n");
1694
1695         /* Primecell ID */
1696         id = readl_relaxed(gr0_base + ARM_SMMU_GR0_PIDR2);
1697         smmu->version = ((id >> PIDR2_ARCH_SHIFT) & PIDR2_ARCH_MASK) + 1;
1698         dev_notice(smmu->dev, "SMMUv%d with:\n", smmu->version);
1699
1700         /* ID0 */
1701         id = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID0);
1702 #ifndef CONFIG_64BIT
1703         if (((id >> ID0_PTFS_SHIFT) & ID0_PTFS_MASK) == ID0_PTFS_V8_ONLY) {
1704                 dev_err(smmu->dev, "\tno v7 descriptor support!\n");
1705                 return -ENODEV;
1706         }
1707 #endif
1708         if (id & ID0_S1TS) {
1709                 smmu->features |= ARM_SMMU_FEAT_TRANS_S1;
1710                 dev_notice(smmu->dev, "\tstage 1 translation\n");
1711         }
1712
1713         if (id & ID0_S2TS) {
1714                 smmu->features |= ARM_SMMU_FEAT_TRANS_S2;
1715                 dev_notice(smmu->dev, "\tstage 2 translation\n");
1716         }
1717
1718         if (id & ID0_NTS) {
1719                 smmu->features |= ARM_SMMU_FEAT_TRANS_NESTED;
1720                 dev_notice(smmu->dev, "\tnested translation\n");
1721         }
1722
1723         if (!(smmu->features &
1724                 (ARM_SMMU_FEAT_TRANS_S1 | ARM_SMMU_FEAT_TRANS_S2 |
1725                  ARM_SMMU_FEAT_TRANS_NESTED))) {
1726                 dev_err(smmu->dev, "\tno translation support!\n");
1727                 return -ENODEV;
1728         }
1729
1730         if (id & ID0_CTTW) {
1731                 smmu->features |= ARM_SMMU_FEAT_COHERENT_WALK;
1732                 dev_notice(smmu->dev, "\tcoherent table walk\n");
1733         }
1734
1735         if (id & ID0_SMS) {
1736                 u32 smr, sid, mask;
1737
1738                 smmu->features |= ARM_SMMU_FEAT_STREAM_MATCH;
1739                 smmu->num_mapping_groups = (id >> ID0_NUMSMRG_SHIFT) &
1740                                            ID0_NUMSMRG_MASK;
1741                 if (smmu->num_mapping_groups == 0) {
1742                         dev_err(smmu->dev,
1743                                 "stream-matching supported, but no SMRs present!\n");
1744                         return -ENODEV;
1745                 }
1746
1747                 smr = SMR_MASK_MASK << SMR_MASK_SHIFT;
1748                 smr |= (SMR_ID_MASK << SMR_ID_SHIFT);
1749                 writel_relaxed(smr, gr0_base + ARM_SMMU_GR0_SMR(0));
1750                 smr = readl_relaxed(gr0_base + ARM_SMMU_GR0_SMR(0));
1751
1752                 mask = (smr >> SMR_MASK_SHIFT) & SMR_MASK_MASK;
1753                 sid = (smr >> SMR_ID_SHIFT) & SMR_ID_MASK;
1754                 if ((mask & sid) != sid) {
1755                         dev_err(smmu->dev,
1756                                 "SMR mask bits (0x%x) insufficient for ID field (0x%x)\n",
1757                                 mask, sid);
1758                         return -ENODEV;
1759                 }
1760
1761                 dev_notice(smmu->dev,
1762                            "\tstream matching with %u register groups, mask 0x%x",
1763                            smmu->num_mapping_groups, mask);
1764         }
1765
1766         /* ID1 */
1767         id = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID1);
1768         smmu->pagesize = (id & ID1_PAGESIZE) ? SZ_64K : SZ_4K;
1769
1770         /* Check for size mismatch of SMMU address space from mapped region */
1771         size = 1 <<
1772                 (((id >> ID1_NUMPAGENDXB_SHIFT) & ID1_NUMPAGENDXB_MASK) + 1);
1773         size *= (smmu->pagesize << 1);
1774         if (smmu->size != size)
1775                 dev_warn(smmu->dev,
1776                         "SMMU address space size (0x%lx) differs from mapped region size (0x%lx)!\n",
1777                         size, smmu->size);
1778
1779         smmu->num_s2_context_banks = (id >> ID1_NUMS2CB_SHIFT) &
1780                                       ID1_NUMS2CB_MASK;
1781         smmu->num_context_banks = (id >> ID1_NUMCB_SHIFT) & ID1_NUMCB_MASK;
1782         if (smmu->num_s2_context_banks > smmu->num_context_banks) {
1783                 dev_err(smmu->dev, "impossible number of S2 context banks!\n");
1784                 return -ENODEV;
1785         }
1786         dev_notice(smmu->dev, "\t%u context banks (%u stage-2 only)\n",
1787                    smmu->num_context_banks, smmu->num_s2_context_banks);
1788
1789         /* ID2 */
1790         id = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID2);
1791         size = arm_smmu_id_size_to_bits((id >> ID2_IAS_SHIFT) & ID2_IAS_MASK);
1792
1793         /*
1794          * Stage-1 output limited by stage-2 input size due to pgd
1795          * allocation (PTRS_PER_PGD).
1796          */
1797 #ifdef CONFIG_64BIT
1798         smmu->s1_output_size = min_t(unsigned long, VA_BITS, size);
1799 #else
1800         smmu->s1_output_size = min(32UL, size);
1801 #endif
1802
1803         /* The stage-2 output mask is also applied for bypass */
1804         size = arm_smmu_id_size_to_bits((id >> ID2_OAS_SHIFT) & ID2_OAS_MASK);
1805         smmu->s2_output_size = min_t(unsigned long, PHYS_MASK_SHIFT, size);
1806
1807         if (smmu->version == 1) {
1808                 smmu->input_size = 32;
1809         } else {
1810 #ifdef CONFIG_64BIT
1811                 size = (id >> ID2_UBS_SHIFT) & ID2_UBS_MASK;
1812                 size = min(VA_BITS, arm_smmu_id_size_to_bits(size));
1813 #else
1814                 size = 32;
1815 #endif
1816                 smmu->input_size = size;
1817
1818                 if ((PAGE_SIZE == SZ_4K && !(id & ID2_PTFS_4K)) ||
1819                     (PAGE_SIZE == SZ_64K && !(id & ID2_PTFS_64K)) ||
1820                     (PAGE_SIZE != SZ_4K && PAGE_SIZE != SZ_64K)) {
1821                         dev_err(smmu->dev, "CPU page size 0x%lx unsupported\n",
1822                                 PAGE_SIZE);
1823                         return -ENODEV;
1824                 }
1825         }
1826
1827         dev_notice(smmu->dev,
1828                    "\t%lu-bit VA, %lu-bit IPA, %lu-bit PA\n",
1829                    smmu->input_size, smmu->s1_output_size,
1830                    smmu->s2_output_size);
1831         return 0;
1832 }
1833
1834 static int arm_smmu_device_dt_probe(struct platform_device *pdev)
1835 {
1836         struct resource *res;
1837         struct arm_smmu_device *smmu;
1838         struct device *dev = &pdev->dev;
1839         struct rb_node *node;
1840         struct of_phandle_args masterspec;
1841         int num_irqs, i, err;
1842
1843         smmu = devm_kzalloc(dev, sizeof(*smmu), GFP_KERNEL);
1844         if (!smmu) {
1845                 dev_err(dev, "failed to allocate arm_smmu_device\n");
1846                 return -ENOMEM;
1847         }
1848         smmu->dev = dev;
1849
1850         res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1851         smmu->base = devm_ioremap_resource(dev, res);
1852         if (IS_ERR(smmu->base))
1853                 return PTR_ERR(smmu->base);
1854         smmu->size = resource_size(res);
1855
1856         if (of_property_read_u32(dev->of_node, "#global-interrupts",
1857                                  &smmu->num_global_irqs)) {
1858                 dev_err(dev, "missing #global-interrupts property\n");
1859                 return -ENODEV;
1860         }
1861
1862         num_irqs = 0;
1863         while ((res = platform_get_resource(pdev, IORESOURCE_IRQ, num_irqs))) {
1864                 num_irqs++;
1865                 if (num_irqs > smmu->num_global_irqs)
1866                         smmu->num_context_irqs++;
1867         }
1868
1869         if (!smmu->num_context_irqs) {
1870                 dev_err(dev, "found %d interrupts but expected at least %d\n",
1871                         num_irqs, smmu->num_global_irqs + 1);
1872                 return -ENODEV;
1873         }
1874
1875         smmu->irqs = devm_kzalloc(dev, sizeof(*smmu->irqs) * num_irqs,
1876                                   GFP_KERNEL);
1877         if (!smmu->irqs) {
1878                 dev_err(dev, "failed to allocate %d irqs\n", num_irqs);
1879                 return -ENOMEM;
1880         }
1881
1882         for (i = 0; i < num_irqs; ++i) {
1883                 int irq = platform_get_irq(pdev, i);
1884
1885                 if (irq < 0) {
1886                         dev_err(dev, "failed to get irq index %d\n", i);
1887                         return -ENODEV;
1888                 }
1889                 smmu->irqs[i] = irq;
1890         }
1891
1892         i = 0;
1893         smmu->masters = RB_ROOT;
1894         while (!of_parse_phandle_with_args(dev->of_node, "mmu-masters",
1895                                            "#stream-id-cells", i,
1896                                            &masterspec)) {
1897                 err = register_smmu_master(smmu, dev, &masterspec);
1898                 if (err) {
1899                         dev_err(dev, "failed to add master %s\n",
1900                                 masterspec.np->name);
1901                         goto out_put_masters;
1902                 }
1903
1904                 i++;
1905         }
1906         dev_notice(dev, "registered %d master devices\n", i);
1907
1908         err = arm_smmu_device_cfg_probe(smmu);
1909         if (err)
1910                 goto out_put_masters;
1911
1912         parse_driver_options(smmu);
1913
1914         if (smmu->version > 1 &&
1915             smmu->num_context_banks != smmu->num_context_irqs) {
1916                 dev_err(dev,
1917                         "found only %d context interrupt(s) but %d required\n",
1918                         smmu->num_context_irqs, smmu->num_context_banks);
1919                 err = -ENODEV;
1920                 goto out_put_masters;
1921         }
1922
1923         for (i = 0; i < smmu->num_global_irqs; ++i) {
1924                 err = request_irq(smmu->irqs[i],
1925                                   arm_smmu_global_fault,
1926                                   IRQF_SHARED,
1927                                   "arm-smmu global fault",
1928                                   smmu);
1929                 if (err) {
1930                         dev_err(dev, "failed to request global IRQ %d (%u)\n",
1931                                 i, smmu->irqs[i]);
1932                         goto out_free_irqs;
1933                 }
1934         }
1935
1936         INIT_LIST_HEAD(&smmu->list);
1937         spin_lock(&arm_smmu_devices_lock);
1938         list_add(&smmu->list, &arm_smmu_devices);
1939         spin_unlock(&arm_smmu_devices_lock);
1940
1941         arm_smmu_device_reset(smmu);
1942         return 0;
1943
1944 out_free_irqs:
1945         while (i--)
1946                 free_irq(smmu->irqs[i], smmu);
1947
1948 out_put_masters:
1949         for (node = rb_first(&smmu->masters); node; node = rb_next(node)) {
1950                 struct arm_smmu_master *master
1951                         = container_of(node, struct arm_smmu_master, node);
1952                 of_node_put(master->of_node);
1953         }
1954
1955         return err;
1956 }
1957
1958 static int arm_smmu_device_remove(struct platform_device *pdev)
1959 {
1960         int i;
1961         struct device *dev = &pdev->dev;
1962         struct arm_smmu_device *curr, *smmu = NULL;
1963         struct rb_node *node;
1964
1965         spin_lock(&arm_smmu_devices_lock);
1966         list_for_each_entry(curr, &arm_smmu_devices, list) {
1967                 if (curr->dev == dev) {
1968                         smmu = curr;
1969                         list_del(&smmu->list);
1970                         break;
1971                 }
1972         }
1973         spin_unlock(&arm_smmu_devices_lock);
1974
1975         if (!smmu)
1976                 return -ENODEV;
1977
1978         for (node = rb_first(&smmu->masters); node; node = rb_next(node)) {
1979                 struct arm_smmu_master *master
1980                         = container_of(node, struct arm_smmu_master, node);
1981                 of_node_put(master->of_node);
1982         }
1983
1984         if (!bitmap_empty(smmu->context_map, ARM_SMMU_MAX_CBS))
1985                 dev_err(dev, "removing device with active domains!\n");
1986
1987         for (i = 0; i < smmu->num_global_irqs; ++i)
1988                 free_irq(smmu->irqs[i], smmu);
1989
1990         /* Turn the thing off */
1991         writel(sCR0_CLIENTPD, ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sCR0);
1992         return 0;
1993 }
1994
1995 #ifdef CONFIG_OF
1996 static struct of_device_id arm_smmu_of_match[] = {
1997         { .compatible = "arm,smmu-v1", },
1998         { .compatible = "arm,smmu-v2", },
1999         { .compatible = "arm,mmu-400", },
2000         { .compatible = "arm,mmu-500", },
2001         { },
2002 };
2003 MODULE_DEVICE_TABLE(of, arm_smmu_of_match);
2004 #endif
2005
2006 static struct platform_driver arm_smmu_driver = {
2007         .driver = {
2008                 .owner          = THIS_MODULE,
2009                 .name           = "arm-smmu",
2010                 .of_match_table = of_match_ptr(arm_smmu_of_match),
2011         },
2012         .probe  = arm_smmu_device_dt_probe,
2013         .remove = arm_smmu_device_remove,
2014 };
2015
2016 static int __init arm_smmu_init(void)
2017 {
2018         int ret;
2019
2020         ret = platform_driver_register(&arm_smmu_driver);
2021         if (ret)
2022                 return ret;
2023
2024         /* Oh, for a proper bus abstraction */
2025         if (!iommu_present(&platform_bus_type))
2026                 bus_set_iommu(&platform_bus_type, &arm_smmu_ops);
2027
2028 #ifdef CONFIG_ARM_AMBA
2029         if (!iommu_present(&amba_bustype))
2030                 bus_set_iommu(&amba_bustype, &arm_smmu_ops);
2031 #endif
2032
2033 #ifdef CONFIG_PCI
2034         if (!iommu_present(&pci_bus_type))
2035                 bus_set_iommu(&pci_bus_type, &arm_smmu_ops);
2036 #endif
2037
2038         return 0;
2039 }
2040
2041 static void __exit arm_smmu_exit(void)
2042 {
2043         return platform_driver_unregister(&arm_smmu_driver);
2044 }
2045
2046 subsys_initcall(arm_smmu_init);
2047 module_exit(arm_smmu_exit);
2048
2049 MODULE_DESCRIPTION("IOMMU API for ARM architected SMMU implementations");
2050 MODULE_AUTHOR("Will Deacon <will.deacon@arm.com>");
2051 MODULE_LICENSE("GPL v2");