iommu/arm-smmu: fix capability checking prior to device attach
[firefly-linux-kernel-4.4.55.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 | FSR_TLBMCF |       \
321                                          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         do {
417                 if (of_property_read_bool(smmu->dev->of_node,
418                                                 arm_smmu_options[i].prop)) {
419                         smmu->options |= arm_smmu_options[i].opt;
420                         dev_notice(smmu->dev, "option %s\n",
421                                 arm_smmu_options[i].prop);
422                 }
423         } while (arm_smmu_options[++i].opt);
424 }
425
426 static struct device *dev_get_master_dev(struct device *dev)
427 {
428         if (dev_is_pci(dev)) {
429                 struct pci_bus *bus = to_pci_dev(dev)->bus;
430                 while (!pci_is_root_bus(bus))
431                         bus = bus->parent;
432                 return bus->bridge->parent;
433         }
434
435         return dev;
436 }
437
438 static struct arm_smmu_master *find_smmu_master(struct arm_smmu_device *smmu,
439                                                 struct device_node *dev_node)
440 {
441         struct rb_node *node = smmu->masters.rb_node;
442
443         while (node) {
444                 struct arm_smmu_master *master;
445                 master = container_of(node, struct arm_smmu_master, node);
446
447                 if (dev_node < master->of_node)
448                         node = node->rb_left;
449                 else if (dev_node > master->of_node)
450                         node = node->rb_right;
451                 else
452                         return master;
453         }
454
455         return NULL;
456 }
457
458 static struct arm_smmu_master_cfg *
459 find_smmu_master_cfg(struct arm_smmu_device *smmu, struct device *dev)
460 {
461         struct arm_smmu_master *master;
462
463         if (dev_is_pci(dev))
464                 return dev->archdata.iommu;
465
466         master = find_smmu_master(smmu, dev->of_node);
467         return master ? &master->cfg : NULL;
468 }
469
470 static int insert_smmu_master(struct arm_smmu_device *smmu,
471                               struct arm_smmu_master *master)
472 {
473         struct rb_node **new, *parent;
474
475         new = &smmu->masters.rb_node;
476         parent = NULL;
477         while (*new) {
478                 struct arm_smmu_master *this;
479                 this = container_of(*new, struct arm_smmu_master, node);
480
481                 parent = *new;
482                 if (master->of_node < this->of_node)
483                         new = &((*new)->rb_left);
484                 else if (master->of_node > this->of_node)
485                         new = &((*new)->rb_right);
486                 else
487                         return -EEXIST;
488         }
489
490         rb_link_node(&master->node, parent, new);
491         rb_insert_color(&master->node, &smmu->masters);
492         return 0;
493 }
494
495 static int register_smmu_master(struct arm_smmu_device *smmu,
496                                 struct device *dev,
497                                 struct of_phandle_args *masterspec)
498 {
499         int i;
500         struct arm_smmu_master *master;
501
502         master = find_smmu_master(smmu, masterspec->np);
503         if (master) {
504                 dev_err(dev,
505                         "rejecting multiple registrations for master device %s\n",
506                         masterspec->np->name);
507                 return -EBUSY;
508         }
509
510         if (masterspec->args_count > MAX_MASTER_STREAMIDS) {
511                 dev_err(dev,
512                         "reached maximum number (%d) of stream IDs for master device %s\n",
513                         MAX_MASTER_STREAMIDS, masterspec->np->name);
514                 return -ENOSPC;
515         }
516
517         master = devm_kzalloc(dev, sizeof(*master), GFP_KERNEL);
518         if (!master)
519                 return -ENOMEM;
520
521         master->of_node                 = masterspec->np;
522         master->cfg.num_streamids       = masterspec->args_count;
523
524         for (i = 0; i < master->cfg.num_streamids; ++i)
525                 master->cfg.streamids[i] = masterspec->args[i];
526
527         return insert_smmu_master(smmu, master);
528 }
529
530 static struct arm_smmu_device *find_smmu_for_device(struct device *dev)
531 {
532         struct arm_smmu_device *smmu;
533         struct arm_smmu_master *master = NULL;
534         struct device_node *dev_node = dev_get_master_dev(dev)->of_node;
535
536         spin_lock(&arm_smmu_devices_lock);
537         list_for_each_entry(smmu, &arm_smmu_devices, list) {
538                 master = find_smmu_master(smmu, dev_node);
539                 if (master)
540                         break;
541         }
542         spin_unlock(&arm_smmu_devices_lock);
543
544         return master ? smmu : NULL;
545 }
546
547 static int __arm_smmu_alloc_bitmap(unsigned long *map, int start, int end)
548 {
549         int idx;
550
551         do {
552                 idx = find_next_zero_bit(map, end, start);
553                 if (idx == end)
554                         return -ENOSPC;
555         } while (test_and_set_bit(idx, map));
556
557         return idx;
558 }
559
560 static void __arm_smmu_free_bitmap(unsigned long *map, int idx)
561 {
562         clear_bit(idx, map);
563 }
564
565 /* Wait for any pending TLB invalidations to complete */
566 static void arm_smmu_tlb_sync(struct arm_smmu_device *smmu)
567 {
568         int count = 0;
569         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
570
571         writel_relaxed(0, gr0_base + ARM_SMMU_GR0_sTLBGSYNC);
572         while (readl_relaxed(gr0_base + ARM_SMMU_GR0_sTLBGSTATUS)
573                & sTLBGSTATUS_GSACTIVE) {
574                 cpu_relax();
575                 if (++count == TLB_LOOP_TIMEOUT) {
576                         dev_err_ratelimited(smmu->dev,
577                         "TLB sync timed out -- SMMU may be deadlocked\n");
578                         return;
579                 }
580                 udelay(1);
581         }
582 }
583
584 static void arm_smmu_tlb_inv_context(struct arm_smmu_domain *smmu_domain)
585 {
586         struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
587         struct arm_smmu_device *smmu = smmu_domain->smmu;
588         void __iomem *base = ARM_SMMU_GR0(smmu);
589         bool stage1 = cfg->cbar != CBAR_TYPE_S2_TRANS;
590
591         if (stage1) {
592                 base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
593                 writel_relaxed(ARM_SMMU_CB_ASID(cfg),
594                                base + ARM_SMMU_CB_S1_TLBIASID);
595         } else {
596                 base = ARM_SMMU_GR0(smmu);
597                 writel_relaxed(ARM_SMMU_CB_VMID(cfg),
598                                base + ARM_SMMU_GR0_TLBIVMID);
599         }
600
601         arm_smmu_tlb_sync(smmu);
602 }
603
604 static irqreturn_t arm_smmu_context_fault(int irq, void *dev)
605 {
606         int flags, ret;
607         u32 fsr, far, fsynr, resume;
608         unsigned long iova;
609         struct iommu_domain *domain = dev;
610         struct arm_smmu_domain *smmu_domain = domain->priv;
611         struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
612         struct arm_smmu_device *smmu = smmu_domain->smmu;
613         void __iomem *cb_base;
614
615         cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
616         fsr = readl_relaxed(cb_base + ARM_SMMU_CB_FSR);
617
618         if (!(fsr & FSR_FAULT))
619                 return IRQ_NONE;
620
621         if (fsr & FSR_IGN)
622                 dev_err_ratelimited(smmu->dev,
623                                     "Unexpected context fault (fsr 0x%u)\n",
624                                     fsr);
625
626         fsynr = readl_relaxed(cb_base + ARM_SMMU_CB_FSYNR0);
627         flags = fsynr & FSYNR0_WNR ? IOMMU_FAULT_WRITE : IOMMU_FAULT_READ;
628
629         far = readl_relaxed(cb_base + ARM_SMMU_CB_FAR_LO);
630         iova = far;
631 #ifdef CONFIG_64BIT
632         far = readl_relaxed(cb_base + ARM_SMMU_CB_FAR_HI);
633         iova |= ((unsigned long)far << 32);
634 #endif
635
636         if (!report_iommu_fault(domain, smmu->dev, iova, flags)) {
637                 ret = IRQ_HANDLED;
638                 resume = RESUME_RETRY;
639         } else {
640                 dev_err_ratelimited(smmu->dev,
641                     "Unhandled context fault: iova=0x%08lx, fsynr=0x%x, cb=%d\n",
642                     iova, fsynr, cfg->cbndx);
643                 ret = IRQ_NONE;
644                 resume = RESUME_TERMINATE;
645         }
646
647         /* Clear the faulting FSR */
648         writel(fsr, cb_base + ARM_SMMU_CB_FSR);
649
650         /* Retry or terminate any stalled transactions */
651         if (fsr & FSR_SS)
652                 writel_relaxed(resume, cb_base + ARM_SMMU_CB_RESUME);
653
654         return ret;
655 }
656
657 static irqreturn_t arm_smmu_global_fault(int irq, void *dev)
658 {
659         u32 gfsr, gfsynr0, gfsynr1, gfsynr2;
660         struct arm_smmu_device *smmu = dev;
661         void __iomem *gr0_base = ARM_SMMU_GR0_NS(smmu);
662
663         gfsr = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSR);
664         gfsynr0 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSYNR0);
665         gfsynr1 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSYNR1);
666         gfsynr2 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSYNR2);
667
668         if (!gfsr)
669                 return IRQ_NONE;
670
671         dev_err_ratelimited(smmu->dev,
672                 "Unexpected global fault, this could be serious\n");
673         dev_err_ratelimited(smmu->dev,
674                 "\tGFSR 0x%08x, GFSYNR0 0x%08x, GFSYNR1 0x%08x, GFSYNR2 0x%08x\n",
675                 gfsr, gfsynr0, gfsynr1, gfsynr2);
676
677         writel(gfsr, gr0_base + ARM_SMMU_GR0_sGFSR);
678         return IRQ_HANDLED;
679 }
680
681 static void arm_smmu_flush_pgtable(struct arm_smmu_device *smmu, void *addr,
682                                    size_t size)
683 {
684         unsigned long offset = (unsigned long)addr & ~PAGE_MASK;
685
686
687         /* Ensure new page tables are visible to the hardware walker */
688         if (smmu->features & ARM_SMMU_FEAT_COHERENT_WALK) {
689                 dsb(ishst);
690         } else {
691                 /*
692                  * If the SMMU can't walk tables in the CPU caches, treat them
693                  * like non-coherent DMA since we need to flush the new entries
694                  * all the way out to memory. There's no possibility of
695                  * recursion here as the SMMU table walker will not be wired
696                  * through another SMMU.
697                  */
698                 dma_map_page(smmu->dev, virt_to_page(addr), offset, size,
699                                 DMA_TO_DEVICE);
700         }
701 }
702
703 static void arm_smmu_init_context_bank(struct arm_smmu_domain *smmu_domain)
704 {
705         u32 reg;
706         bool stage1;
707         struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
708         struct arm_smmu_device *smmu = smmu_domain->smmu;
709         void __iomem *cb_base, *gr0_base, *gr1_base;
710
711         gr0_base = ARM_SMMU_GR0(smmu);
712         gr1_base = ARM_SMMU_GR1(smmu);
713         stage1 = cfg->cbar != CBAR_TYPE_S2_TRANS;
714         cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
715
716         /* CBAR */
717         reg = cfg->cbar;
718         if (smmu->version == 1)
719               reg |= cfg->irptndx << CBAR_IRPTNDX_SHIFT;
720
721         /*
722          * Use the weakest shareability/memory types, so they are
723          * overridden by the ttbcr/pte.
724          */
725         if (stage1) {
726                 reg |= (CBAR_S1_BPSHCFG_NSH << CBAR_S1_BPSHCFG_SHIFT) |
727                         (CBAR_S1_MEMATTR_WB << CBAR_S1_MEMATTR_SHIFT);
728         } else {
729                 reg |= ARM_SMMU_CB_VMID(cfg) << CBAR_VMID_SHIFT;
730         }
731         writel_relaxed(reg, gr1_base + ARM_SMMU_GR1_CBAR(cfg->cbndx));
732
733         if (smmu->version > 1) {
734                 /* CBA2R */
735 #ifdef CONFIG_64BIT
736                 reg = CBA2R_RW64_64BIT;
737 #else
738                 reg = CBA2R_RW64_32BIT;
739 #endif
740                 writel_relaxed(reg,
741                                gr1_base + ARM_SMMU_GR1_CBA2R(cfg->cbndx));
742
743                 /* TTBCR2 */
744                 switch (smmu->input_size) {
745                 case 32:
746                         reg = (TTBCR2_ADDR_32 << TTBCR2_SEP_SHIFT);
747                         break;
748                 case 36:
749                         reg = (TTBCR2_ADDR_36 << TTBCR2_SEP_SHIFT);
750                         break;
751                 case 39:
752                         reg = (TTBCR2_ADDR_40 << TTBCR2_SEP_SHIFT);
753                         break;
754                 case 42:
755                         reg = (TTBCR2_ADDR_42 << TTBCR2_SEP_SHIFT);
756                         break;
757                 case 44:
758                         reg = (TTBCR2_ADDR_44 << TTBCR2_SEP_SHIFT);
759                         break;
760                 case 48:
761                         reg = (TTBCR2_ADDR_48 << TTBCR2_SEP_SHIFT);
762                         break;
763                 }
764
765                 switch (smmu->s1_output_size) {
766                 case 32:
767                         reg |= (TTBCR2_ADDR_32 << TTBCR2_PASIZE_SHIFT);
768                         break;
769                 case 36:
770                         reg |= (TTBCR2_ADDR_36 << TTBCR2_PASIZE_SHIFT);
771                         break;
772                 case 39:
773                         reg |= (TTBCR2_ADDR_40 << TTBCR2_PASIZE_SHIFT);
774                         break;
775                 case 42:
776                         reg |= (TTBCR2_ADDR_42 << TTBCR2_PASIZE_SHIFT);
777                         break;
778                 case 44:
779                         reg |= (TTBCR2_ADDR_44 << TTBCR2_PASIZE_SHIFT);
780                         break;
781                 case 48:
782                         reg |= (TTBCR2_ADDR_48 << TTBCR2_PASIZE_SHIFT);
783                         break;
784                 }
785
786                 if (stage1)
787                         writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBCR2);
788         }
789
790         /* TTBR0 */
791         arm_smmu_flush_pgtable(smmu, cfg->pgd,
792                                PTRS_PER_PGD * sizeof(pgd_t));
793         reg = __pa(cfg->pgd);
794         writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBR0_LO);
795         reg = (phys_addr_t)__pa(cfg->pgd) >> 32;
796         if (stage1)
797                 reg |= ARM_SMMU_CB_ASID(cfg) << TTBRn_HI_ASID_SHIFT;
798         writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBR0_HI);
799
800         /*
801          * TTBCR
802          * We use long descriptor, with inner-shareable WBWA tables in TTBR0.
803          */
804         if (smmu->version > 1) {
805                 if (PAGE_SIZE == SZ_4K)
806                         reg = TTBCR_TG0_4K;
807                 else
808                         reg = TTBCR_TG0_64K;
809
810                 if (!stage1) {
811                         reg |= (64 - smmu->s1_output_size) << TTBCR_T0SZ_SHIFT;
812
813                         switch (smmu->s2_output_size) {
814                         case 32:
815                                 reg |= (TTBCR2_ADDR_32 << TTBCR_PASIZE_SHIFT);
816                                 break;
817                         case 36:
818                                 reg |= (TTBCR2_ADDR_36 << TTBCR_PASIZE_SHIFT);
819                                 break;
820                         case 40:
821                                 reg |= (TTBCR2_ADDR_40 << TTBCR_PASIZE_SHIFT);
822                                 break;
823                         case 42:
824                                 reg |= (TTBCR2_ADDR_42 << TTBCR_PASIZE_SHIFT);
825                                 break;
826                         case 44:
827                                 reg |= (TTBCR2_ADDR_44 << TTBCR_PASIZE_SHIFT);
828                                 break;
829                         case 48:
830                                 reg |= (TTBCR2_ADDR_48 << TTBCR_PASIZE_SHIFT);
831                                 break;
832                         }
833                 } else {
834                         reg |= (64 - smmu->input_size) << TTBCR_T0SZ_SHIFT;
835                 }
836         } else {
837                 reg = 0;
838         }
839
840         reg |= TTBCR_EAE |
841               (TTBCR_SH_IS << TTBCR_SH0_SHIFT) |
842               (TTBCR_RGN_WBWA << TTBCR_ORGN0_SHIFT) |
843               (TTBCR_RGN_WBWA << TTBCR_IRGN0_SHIFT) |
844               (TTBCR_SL0_LVL_1 << TTBCR_SL0_SHIFT);
845         writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBCR);
846
847         /* MAIR0 (stage-1 only) */
848         if (stage1) {
849                 reg = (MAIR_ATTR_NC << MAIR_ATTR_SHIFT(MAIR_ATTR_IDX_NC)) |
850                       (MAIR_ATTR_WBRWA << MAIR_ATTR_SHIFT(MAIR_ATTR_IDX_CACHE)) |
851                       (MAIR_ATTR_DEVICE << MAIR_ATTR_SHIFT(MAIR_ATTR_IDX_DEV));
852                 writel_relaxed(reg, cb_base + ARM_SMMU_CB_S1_MAIR0);
853         }
854
855         /* SCTLR */
856         reg = SCTLR_CFCFG | SCTLR_CFIE | SCTLR_CFRE | SCTLR_M | SCTLR_EAE_SBOP;
857         if (stage1)
858                 reg |= SCTLR_S1_ASIDPNE;
859 #ifdef __BIG_ENDIAN
860         reg |= SCTLR_E;
861 #endif
862         writel_relaxed(reg, cb_base + ARM_SMMU_CB_SCTLR);
863 }
864
865 static int arm_smmu_init_domain_context(struct iommu_domain *domain,
866                                         struct arm_smmu_device *smmu)
867 {
868         int irq, ret, start;
869         struct arm_smmu_domain *smmu_domain = domain->priv;
870         struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
871
872         if (smmu->features & ARM_SMMU_FEAT_TRANS_NESTED) {
873                 /*
874                  * We will likely want to change this if/when KVM gets
875                  * involved.
876                  */
877                 cfg->cbar = CBAR_TYPE_S1_TRANS_S2_BYPASS;
878                 start = smmu->num_s2_context_banks;
879         } else if (smmu->features & ARM_SMMU_FEAT_TRANS_S1) {
880                 cfg->cbar = CBAR_TYPE_S1_TRANS_S2_BYPASS;
881                 start = smmu->num_s2_context_banks;
882         } else {
883                 cfg->cbar = CBAR_TYPE_S2_TRANS;
884                 start = 0;
885         }
886
887         ret = __arm_smmu_alloc_bitmap(smmu->context_map, start,
888                                       smmu->num_context_banks);
889         if (IS_ERR_VALUE(ret))
890                 return ret;
891
892         cfg->cbndx = ret;
893         if (smmu->version == 1) {
894                 cfg->irptndx = atomic_inc_return(&smmu->irptndx);
895                 cfg->irptndx %= smmu->num_context_irqs;
896         } else {
897                 cfg->irptndx = cfg->cbndx;
898         }
899
900         irq = smmu->irqs[smmu->num_global_irqs + cfg->irptndx];
901         ret = request_irq(irq, arm_smmu_context_fault, IRQF_SHARED,
902                           "arm-smmu-context-fault", domain);
903         if (IS_ERR_VALUE(ret)) {
904                 dev_err(smmu->dev, "failed to request context IRQ %d (%u)\n",
905                         cfg->irptndx, irq);
906                 cfg->irptndx = INVALID_IRPTNDX;
907                 goto out_free_context;
908         }
909
910         smmu_domain->smmu = smmu;
911         arm_smmu_init_context_bank(smmu_domain);
912         return 0;
913
914 out_free_context:
915         __arm_smmu_free_bitmap(smmu->context_map, cfg->cbndx);
916         return ret;
917 }
918
919 static void arm_smmu_destroy_domain_context(struct iommu_domain *domain)
920 {
921         struct arm_smmu_domain *smmu_domain = domain->priv;
922         struct arm_smmu_device *smmu = smmu_domain->smmu;
923         struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
924         void __iomem *cb_base;
925         int irq;
926
927         if (!smmu)
928                 return;
929
930         /* Disable the context bank and nuke the TLB before freeing it. */
931         cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
932         writel_relaxed(0, cb_base + ARM_SMMU_CB_SCTLR);
933         arm_smmu_tlb_inv_context(smmu_domain);
934
935         if (cfg->irptndx != INVALID_IRPTNDX) {
936                 irq = smmu->irqs[smmu->num_global_irqs + cfg->irptndx];
937                 free_irq(irq, domain);
938         }
939
940         __arm_smmu_free_bitmap(smmu->context_map, cfg->cbndx);
941 }
942
943 static int arm_smmu_domain_init(struct iommu_domain *domain)
944 {
945         struct arm_smmu_domain *smmu_domain;
946         pgd_t *pgd;
947
948         /*
949          * Allocate the domain and initialise some of its data structures.
950          * We can't really do anything meaningful until we've added a
951          * master.
952          */
953         smmu_domain = kzalloc(sizeof(*smmu_domain), GFP_KERNEL);
954         if (!smmu_domain)
955                 return -ENOMEM;
956
957         pgd = kzalloc(PTRS_PER_PGD * sizeof(pgd_t), GFP_KERNEL);
958         if (!pgd)
959                 goto out_free_domain;
960         smmu_domain->cfg.pgd = pgd;
961
962         spin_lock_init(&smmu_domain->lock);
963         domain->priv = smmu_domain;
964         return 0;
965
966 out_free_domain:
967         kfree(smmu_domain);
968         return -ENOMEM;
969 }
970
971 static void arm_smmu_free_ptes(pmd_t *pmd)
972 {
973         pgtable_t table = pmd_pgtable(*pmd);
974         pgtable_page_dtor(table);
975         __free_page(table);
976 }
977
978 static void arm_smmu_free_pmds(pud_t *pud)
979 {
980         int i;
981         pmd_t *pmd, *pmd_base = pmd_offset(pud, 0);
982
983         pmd = pmd_base;
984         for (i = 0; i < PTRS_PER_PMD; ++i) {
985                 if (pmd_none(*pmd))
986                         continue;
987
988                 arm_smmu_free_ptes(pmd);
989                 pmd++;
990         }
991
992         pmd_free(NULL, pmd_base);
993 }
994
995 static void arm_smmu_free_puds(pgd_t *pgd)
996 {
997         int i;
998         pud_t *pud, *pud_base = pud_offset(pgd, 0);
999
1000         pud = pud_base;
1001         for (i = 0; i < PTRS_PER_PUD; ++i) {
1002                 if (pud_none(*pud))
1003                         continue;
1004
1005                 arm_smmu_free_pmds(pud);
1006                 pud++;
1007         }
1008
1009         pud_free(NULL, pud_base);
1010 }
1011
1012 static void arm_smmu_free_pgtables(struct arm_smmu_domain *smmu_domain)
1013 {
1014         int i;
1015         struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
1016         pgd_t *pgd, *pgd_base = cfg->pgd;
1017
1018         /*
1019          * Recursively free the page tables for this domain. We don't
1020          * care about speculative TLB filling because the tables should
1021          * not be active in any context bank at this point (SCTLR.M is 0).
1022          */
1023         pgd = pgd_base;
1024         for (i = 0; i < PTRS_PER_PGD; ++i) {
1025                 if (pgd_none(*pgd))
1026                         continue;
1027                 arm_smmu_free_puds(pgd);
1028                 pgd++;
1029         }
1030
1031         kfree(pgd_base);
1032 }
1033
1034 static void arm_smmu_domain_destroy(struct iommu_domain *domain)
1035 {
1036         struct arm_smmu_domain *smmu_domain = domain->priv;
1037
1038         /*
1039          * Free the domain resources. We assume that all devices have
1040          * already been detached.
1041          */
1042         arm_smmu_destroy_domain_context(domain);
1043         arm_smmu_free_pgtables(smmu_domain);
1044         kfree(smmu_domain);
1045 }
1046
1047 static int arm_smmu_master_configure_smrs(struct arm_smmu_device *smmu,
1048                                           struct arm_smmu_master_cfg *cfg)
1049 {
1050         int i;
1051         struct arm_smmu_smr *smrs;
1052         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1053
1054         if (!(smmu->features & ARM_SMMU_FEAT_STREAM_MATCH))
1055                 return 0;
1056
1057         if (cfg->smrs)
1058                 return -EEXIST;
1059
1060         smrs = kmalloc(sizeof(*smrs) * cfg->num_streamids, GFP_KERNEL);
1061         if (!smrs) {
1062                 dev_err(smmu->dev, "failed to allocate %d SMRs\n",
1063                         cfg->num_streamids);
1064                 return -ENOMEM;
1065         }
1066
1067         /* Allocate the SMRs on the SMMU */
1068         for (i = 0; i < cfg->num_streamids; ++i) {
1069                 int idx = __arm_smmu_alloc_bitmap(smmu->smr_map, 0,
1070                                                   smmu->num_mapping_groups);
1071                 if (IS_ERR_VALUE(idx)) {
1072                         dev_err(smmu->dev, "failed to allocate free SMR\n");
1073                         goto err_free_smrs;
1074                 }
1075
1076                 smrs[i] = (struct arm_smmu_smr) {
1077                         .idx    = idx,
1078                         .mask   = 0, /* We don't currently share SMRs */
1079                         .id     = cfg->streamids[i],
1080                 };
1081         }
1082
1083         /* It worked! Now, poke the actual hardware */
1084         for (i = 0; i < cfg->num_streamids; ++i) {
1085                 u32 reg = SMR_VALID | smrs[i].id << SMR_ID_SHIFT |
1086                           smrs[i].mask << SMR_MASK_SHIFT;
1087                 writel_relaxed(reg, gr0_base + ARM_SMMU_GR0_SMR(smrs[i].idx));
1088         }
1089
1090         cfg->smrs = smrs;
1091         return 0;
1092
1093 err_free_smrs:
1094         while (--i >= 0)
1095                 __arm_smmu_free_bitmap(smmu->smr_map, smrs[i].idx);
1096         kfree(smrs);
1097         return -ENOSPC;
1098 }
1099
1100 static void arm_smmu_master_free_smrs(struct arm_smmu_device *smmu,
1101                                       struct arm_smmu_master_cfg *cfg)
1102 {
1103         int i;
1104         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1105         struct arm_smmu_smr *smrs = cfg->smrs;
1106
1107         /* Invalidate the SMRs before freeing back to the allocator */
1108         for (i = 0; i < cfg->num_streamids; ++i) {
1109                 u8 idx = smrs[i].idx;
1110                 writel_relaxed(~SMR_VALID, gr0_base + ARM_SMMU_GR0_SMR(idx));
1111                 __arm_smmu_free_bitmap(smmu->smr_map, idx);
1112         }
1113
1114         cfg->smrs = NULL;
1115         kfree(smrs);
1116 }
1117
1118 static void arm_smmu_bypass_stream_mapping(struct arm_smmu_device *smmu,
1119                                            struct arm_smmu_master_cfg *cfg)
1120 {
1121         int i;
1122         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1123
1124         for (i = 0; i < cfg->num_streamids; ++i) {
1125                 u16 sid = cfg->streamids[i];
1126                 writel_relaxed(S2CR_TYPE_BYPASS,
1127                                gr0_base + ARM_SMMU_GR0_S2CR(sid));
1128         }
1129 }
1130
1131 static int arm_smmu_domain_add_master(struct arm_smmu_domain *smmu_domain,
1132                                       struct arm_smmu_master_cfg *cfg)
1133 {
1134         int i, ret;
1135         struct arm_smmu_device *smmu = smmu_domain->smmu;
1136         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1137
1138         ret = arm_smmu_master_configure_smrs(smmu, cfg);
1139         if (ret)
1140                 return ret;
1141
1142         for (i = 0; i < cfg->num_streamids; ++i) {
1143                 u32 idx, s2cr;
1144                 idx = cfg->smrs ? cfg->smrs[i].idx : cfg->streamids[i];
1145                 s2cr = S2CR_TYPE_TRANS |
1146                        (smmu_domain->cfg.cbndx << S2CR_CBNDX_SHIFT);
1147                 writel_relaxed(s2cr, gr0_base + ARM_SMMU_GR0_S2CR(idx));
1148         }
1149
1150         return 0;
1151 }
1152
1153 static void arm_smmu_domain_remove_master(struct arm_smmu_domain *smmu_domain,
1154                                           struct arm_smmu_master_cfg *cfg)
1155 {
1156         struct arm_smmu_device *smmu = smmu_domain->smmu;
1157
1158         /*
1159          * We *must* clear the S2CR first, because freeing the SMR means
1160          * that it can be re-allocated immediately.
1161          */
1162         arm_smmu_bypass_stream_mapping(smmu, cfg);
1163         arm_smmu_master_free_smrs(smmu, cfg);
1164 }
1165
1166 static int arm_smmu_attach_dev(struct iommu_domain *domain, struct device *dev)
1167 {
1168         int ret = -EINVAL;
1169         struct arm_smmu_domain *smmu_domain = domain->priv;
1170         struct arm_smmu_device *smmu;
1171         struct arm_smmu_master_cfg *cfg;
1172         unsigned long flags;
1173
1174         smmu = dev_get_master_dev(dev)->archdata.iommu;
1175         if (!smmu) {
1176                 dev_err(dev, "cannot attach to SMMU, is it on the same bus?\n");
1177                 return -ENXIO;
1178         }
1179
1180         /*
1181          * Sanity check the domain. We don't support domains across
1182          * different SMMUs.
1183          */
1184         spin_lock_irqsave(&smmu_domain->lock, flags);
1185         if (!smmu_domain->smmu) {
1186                 /* Now that we have a master, we can finalise the domain */
1187                 ret = arm_smmu_init_domain_context(domain, smmu);
1188                 if (IS_ERR_VALUE(ret))
1189                         goto err_unlock;
1190         } else if (smmu_domain->smmu != smmu) {
1191                 dev_err(dev,
1192                         "cannot attach to SMMU %s whilst already attached to domain on SMMU %s\n",
1193                         dev_name(smmu_domain->smmu->dev),
1194                         dev_name(smmu->dev));
1195                 goto err_unlock;
1196         }
1197         spin_unlock_irqrestore(&smmu_domain->lock, flags);
1198
1199         /* Looks ok, so add the device to the domain */
1200         cfg = find_smmu_master_cfg(smmu_domain->smmu, dev);
1201         if (!cfg)
1202                 return -ENODEV;
1203
1204         return arm_smmu_domain_add_master(smmu_domain, cfg);
1205
1206 err_unlock:
1207         spin_unlock_irqrestore(&smmu_domain->lock, flags);
1208         return ret;
1209 }
1210
1211 static void arm_smmu_detach_dev(struct iommu_domain *domain, struct device *dev)
1212 {
1213         struct arm_smmu_domain *smmu_domain = domain->priv;
1214         struct arm_smmu_master_cfg *cfg;
1215
1216         cfg = find_smmu_master_cfg(smmu_domain->smmu, dev);
1217         if (cfg)
1218                 arm_smmu_domain_remove_master(smmu_domain, cfg);
1219 }
1220
1221 static bool arm_smmu_pte_is_contiguous_range(unsigned long addr,
1222                                              unsigned long end)
1223 {
1224         return !(addr & ~ARM_SMMU_PTE_CONT_MASK) &&
1225                 (addr + ARM_SMMU_PTE_CONT_SIZE <= end);
1226 }
1227
1228 static int arm_smmu_alloc_init_pte(struct arm_smmu_device *smmu, pmd_t *pmd,
1229                                    unsigned long addr, unsigned long end,
1230                                    unsigned long pfn, int prot, int stage)
1231 {
1232         pte_t *pte, *start;
1233         pteval_t pteval = ARM_SMMU_PTE_PAGE | ARM_SMMU_PTE_AF | ARM_SMMU_PTE_XN;
1234
1235         if (pmd_none(*pmd)) {
1236                 /* Allocate a new set of tables */
1237                 pgtable_t table = alloc_page(GFP_ATOMIC|__GFP_ZERO);
1238                 if (!table)
1239                         return -ENOMEM;
1240
1241                 arm_smmu_flush_pgtable(smmu, page_address(table), PAGE_SIZE);
1242                 if (!pgtable_page_ctor(table)) {
1243                         __free_page(table);
1244                         return -ENOMEM;
1245                 }
1246                 pmd_populate(NULL, pmd, table);
1247                 arm_smmu_flush_pgtable(smmu, pmd, sizeof(*pmd));
1248         }
1249
1250         if (stage == 1) {
1251                 pteval |= ARM_SMMU_PTE_AP_UNPRIV | ARM_SMMU_PTE_nG;
1252                 if (!(prot & IOMMU_WRITE) && (prot & IOMMU_READ))
1253                         pteval |= ARM_SMMU_PTE_AP_RDONLY;
1254
1255                 if (prot & IOMMU_CACHE)
1256                         pteval |= (MAIR_ATTR_IDX_CACHE <<
1257                                    ARM_SMMU_PTE_ATTRINDX_SHIFT);
1258         } else {
1259                 pteval |= ARM_SMMU_PTE_HAP_FAULT;
1260                 if (prot & IOMMU_READ)
1261                         pteval |= ARM_SMMU_PTE_HAP_READ;
1262                 if (prot & IOMMU_WRITE)
1263                         pteval |= ARM_SMMU_PTE_HAP_WRITE;
1264                 if (prot & IOMMU_CACHE)
1265                         pteval |= ARM_SMMU_PTE_MEMATTR_OIWB;
1266                 else
1267                         pteval |= ARM_SMMU_PTE_MEMATTR_NC;
1268         }
1269
1270         /* If no access, create a faulting entry to avoid TLB fills */
1271         if (prot & IOMMU_EXEC)
1272                 pteval &= ~ARM_SMMU_PTE_XN;
1273         else if (!(prot & (IOMMU_READ | IOMMU_WRITE)))
1274                 pteval &= ~ARM_SMMU_PTE_PAGE;
1275
1276         pteval |= ARM_SMMU_PTE_SH_IS;
1277         start = pmd_page_vaddr(*pmd) + pte_index(addr);
1278         pte = start;
1279
1280         /*
1281          * Install the page table entries. This is fairly complicated
1282          * since we attempt to make use of the contiguous hint in the
1283          * ptes where possible. The contiguous hint indicates a series
1284          * of ARM_SMMU_PTE_CONT_ENTRIES ptes mapping a physically
1285          * contiguous region with the following constraints:
1286          *
1287          *   - The region start is aligned to ARM_SMMU_PTE_CONT_SIZE
1288          *   - Each pte in the region has the contiguous hint bit set
1289          *
1290          * This complicates unmapping (also handled by this code, when
1291          * neither IOMMU_READ or IOMMU_WRITE are set) because it is
1292          * possible, yet highly unlikely, that a client may unmap only
1293          * part of a contiguous range. This requires clearing of the
1294          * contiguous hint bits in the range before installing the new
1295          * faulting entries.
1296          *
1297          * Note that re-mapping an address range without first unmapping
1298          * it is not supported, so TLB invalidation is not required here
1299          * and is instead performed at unmap and domain-init time.
1300          */
1301         do {
1302                 int i = 1;
1303                 pteval &= ~ARM_SMMU_PTE_CONT;
1304
1305                 if (arm_smmu_pte_is_contiguous_range(addr, end)) {
1306                         i = ARM_SMMU_PTE_CONT_ENTRIES;
1307                         pteval |= ARM_SMMU_PTE_CONT;
1308                 } else if (pte_val(*pte) &
1309                            (ARM_SMMU_PTE_CONT | ARM_SMMU_PTE_PAGE)) {
1310                         int j;
1311                         pte_t *cont_start;
1312                         unsigned long idx = pte_index(addr);
1313
1314                         idx &= ~(ARM_SMMU_PTE_CONT_ENTRIES - 1);
1315                         cont_start = pmd_page_vaddr(*pmd) + idx;
1316                         for (j = 0; j < ARM_SMMU_PTE_CONT_ENTRIES; ++j)
1317                                 pte_val(*(cont_start + j)) &= ~ARM_SMMU_PTE_CONT;
1318
1319                         arm_smmu_flush_pgtable(smmu, cont_start,
1320                                                sizeof(*pte) *
1321                                                ARM_SMMU_PTE_CONT_ENTRIES);
1322                 }
1323
1324                 do {
1325                         *pte = pfn_pte(pfn, __pgprot(pteval));
1326                 } while (pte++, pfn++, addr += PAGE_SIZE, --i);
1327         } while (addr != end);
1328
1329         arm_smmu_flush_pgtable(smmu, start, sizeof(*pte) * (pte - start));
1330         return 0;
1331 }
1332
1333 static int arm_smmu_alloc_init_pmd(struct arm_smmu_device *smmu, pud_t *pud,
1334                                    unsigned long addr, unsigned long end,
1335                                    phys_addr_t phys, int prot, int stage)
1336 {
1337         int ret;
1338         pmd_t *pmd;
1339         unsigned long next, pfn = __phys_to_pfn(phys);
1340
1341 #ifndef __PAGETABLE_PMD_FOLDED
1342         if (pud_none(*pud)) {
1343                 pmd = (pmd_t *)get_zeroed_page(GFP_ATOMIC);
1344                 if (!pmd)
1345                         return -ENOMEM;
1346
1347                 arm_smmu_flush_pgtable(smmu, pmd, PAGE_SIZE);
1348                 pud_populate(NULL, pud, pmd);
1349                 arm_smmu_flush_pgtable(smmu, pud, sizeof(*pud));
1350
1351                 pmd += pmd_index(addr);
1352         } else
1353 #endif
1354                 pmd = pmd_offset(pud, addr);
1355
1356         do {
1357                 next = pmd_addr_end(addr, end);
1358                 ret = arm_smmu_alloc_init_pte(smmu, pmd, addr, next, pfn,
1359                                               prot, stage);
1360                 phys += next - addr;
1361         } while (pmd++, addr = next, addr < end);
1362
1363         return ret;
1364 }
1365
1366 static int arm_smmu_alloc_init_pud(struct arm_smmu_device *smmu, pgd_t *pgd,
1367                                    unsigned long addr, unsigned long end,
1368                                    phys_addr_t phys, int prot, int stage)
1369 {
1370         int ret = 0;
1371         pud_t *pud;
1372         unsigned long next;
1373
1374 #ifndef __PAGETABLE_PUD_FOLDED
1375         if (pgd_none(*pgd)) {
1376                 pud = (pud_t *)get_zeroed_page(GFP_ATOMIC);
1377                 if (!pud)
1378                         return -ENOMEM;
1379
1380                 arm_smmu_flush_pgtable(smmu, pud, PAGE_SIZE);
1381                 pgd_populate(NULL, pgd, pud);
1382                 arm_smmu_flush_pgtable(smmu, pgd, sizeof(*pgd));
1383
1384                 pud += pud_index(addr);
1385         } else
1386 #endif
1387                 pud = pud_offset(pgd, addr);
1388
1389         do {
1390                 next = pud_addr_end(addr, end);
1391                 ret = arm_smmu_alloc_init_pmd(smmu, pud, addr, next, phys,
1392                                               prot, stage);
1393                 phys += next - addr;
1394         } while (pud++, addr = next, addr < end);
1395
1396         return ret;
1397 }
1398
1399 static int arm_smmu_handle_mapping(struct arm_smmu_domain *smmu_domain,
1400                                    unsigned long iova, phys_addr_t paddr,
1401                                    size_t size, int prot)
1402 {
1403         int ret, stage;
1404         unsigned long end;
1405         phys_addr_t input_mask, output_mask;
1406         struct arm_smmu_device *smmu = smmu_domain->smmu;
1407         struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
1408         pgd_t *pgd = cfg->pgd;
1409         unsigned long flags;
1410
1411         if (cfg->cbar == CBAR_TYPE_S2_TRANS) {
1412                 stage = 2;
1413                 output_mask = (1ULL << smmu->s2_output_size) - 1;
1414         } else {
1415                 stage = 1;
1416                 output_mask = (1ULL << smmu->s1_output_size) - 1;
1417         }
1418
1419         if (!pgd)
1420                 return -EINVAL;
1421
1422         if (size & ~PAGE_MASK)
1423                 return -EINVAL;
1424
1425         input_mask = (1ULL << smmu->input_size) - 1;
1426         if ((phys_addr_t)iova & ~input_mask)
1427                 return -ERANGE;
1428
1429         if (paddr & ~output_mask)
1430                 return -ERANGE;
1431
1432         spin_lock_irqsave(&smmu_domain->lock, flags);
1433         pgd += pgd_index(iova);
1434         end = iova + size;
1435         do {
1436                 unsigned long next = pgd_addr_end(iova, end);
1437
1438                 ret = arm_smmu_alloc_init_pud(smmu, pgd, iova, next, paddr,
1439                                               prot, stage);
1440                 if (ret)
1441                         goto out_unlock;
1442
1443                 paddr += next - iova;
1444                 iova = next;
1445         } while (pgd++, iova != end);
1446
1447 out_unlock:
1448         spin_unlock_irqrestore(&smmu_domain->lock, flags);
1449
1450         return ret;
1451 }
1452
1453 static int arm_smmu_map(struct iommu_domain *domain, unsigned long iova,
1454                         phys_addr_t paddr, size_t size, int prot)
1455 {
1456         struct arm_smmu_domain *smmu_domain = domain->priv;
1457
1458         if (!smmu_domain)
1459                 return -ENODEV;
1460
1461         return arm_smmu_handle_mapping(smmu_domain, iova, paddr, size, prot);
1462 }
1463
1464 static size_t arm_smmu_unmap(struct iommu_domain *domain, unsigned long iova,
1465                              size_t size)
1466 {
1467         int ret;
1468         struct arm_smmu_domain *smmu_domain = domain->priv;
1469
1470         ret = arm_smmu_handle_mapping(smmu_domain, iova, 0, size, 0);
1471         arm_smmu_tlb_inv_context(smmu_domain);
1472         return ret ? 0 : size;
1473 }
1474
1475 static phys_addr_t arm_smmu_iova_to_phys(struct iommu_domain *domain,
1476                                          dma_addr_t iova)
1477 {
1478         pgd_t *pgdp, pgd;
1479         pud_t pud;
1480         pmd_t pmd;
1481         pte_t pte;
1482         struct arm_smmu_domain *smmu_domain = domain->priv;
1483         struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
1484
1485         pgdp = cfg->pgd;
1486         if (!pgdp)
1487                 return 0;
1488
1489         pgd = *(pgdp + pgd_index(iova));
1490         if (pgd_none(pgd))
1491                 return 0;
1492
1493         pud = *pud_offset(&pgd, iova);
1494         if (pud_none(pud))
1495                 return 0;
1496
1497         pmd = *pmd_offset(&pud, iova);
1498         if (pmd_none(pmd))
1499                 return 0;
1500
1501         pte = *(pmd_page_vaddr(pmd) + pte_index(iova));
1502         if (pte_none(pte))
1503                 return 0;
1504
1505         return __pfn_to_phys(pte_pfn(pte)) | (iova & ~PAGE_MASK);
1506 }
1507
1508 static int arm_smmu_domain_has_cap(struct iommu_domain *domain,
1509                                    unsigned long cap)
1510 {
1511         struct arm_smmu_domain *smmu_domain = domain->priv;
1512         struct arm_smmu_device *smmu = smmu_domain->smmu;
1513         u32 features = smmu ? smmu->features : 0;
1514
1515         switch (cap) {
1516         case IOMMU_CAP_CACHE_COHERENCY:
1517                 return features & ARM_SMMU_FEAT_COHERENT_WALK;
1518         case IOMMU_CAP_INTR_REMAP:
1519                 return 1; /* MSIs are just memory writes */
1520         default:
1521                 return 0;
1522         }
1523 }
1524
1525 static int __arm_smmu_get_pci_sid(struct pci_dev *pdev, u16 alias, void *data)
1526 {
1527         *((u16 *)data) = alias;
1528         return 0; /* Continue walking */
1529 }
1530
1531 static int arm_smmu_add_device(struct device *dev)
1532 {
1533         struct arm_smmu_device *smmu;
1534         struct iommu_group *group;
1535         int ret;
1536
1537         if (dev->archdata.iommu) {
1538                 dev_warn(dev, "IOMMU driver already assigned to device\n");
1539                 return -EINVAL;
1540         }
1541
1542         smmu = find_smmu_for_device(dev);
1543         if (!smmu)
1544                 return -ENODEV;
1545
1546         group = iommu_group_alloc();
1547         if (IS_ERR(group)) {
1548                 dev_err(dev, "Failed to allocate IOMMU group\n");
1549                 return PTR_ERR(group);
1550         }
1551
1552         if (dev_is_pci(dev)) {
1553                 struct arm_smmu_master_cfg *cfg;
1554                 struct pci_dev *pdev = to_pci_dev(dev);
1555
1556                 cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1557                 if (!cfg) {
1558                         ret = -ENOMEM;
1559                         goto out_put_group;
1560                 }
1561
1562                 cfg->num_streamids = 1;
1563                 /*
1564                  * Assume Stream ID == Requester ID for now.
1565                  * We need a way to describe the ID mappings in FDT.
1566                  */
1567                 pci_for_each_dma_alias(pdev, __arm_smmu_get_pci_sid,
1568                                        &cfg->streamids[0]);
1569                 dev->archdata.iommu = cfg;
1570         } else {
1571                 dev->archdata.iommu = smmu;
1572         }
1573
1574         ret = iommu_group_add_device(group, dev);
1575
1576 out_put_group:
1577         iommu_group_put(group);
1578         return ret;
1579 }
1580
1581 static void arm_smmu_remove_device(struct device *dev)
1582 {
1583         if (dev_is_pci(dev))
1584                 kfree(dev->archdata.iommu);
1585
1586         dev->archdata.iommu = NULL;
1587         iommu_group_remove_device(dev);
1588 }
1589
1590 static struct iommu_ops arm_smmu_ops = {
1591         .domain_init    = arm_smmu_domain_init,
1592         .domain_destroy = arm_smmu_domain_destroy,
1593         .attach_dev     = arm_smmu_attach_dev,
1594         .detach_dev     = arm_smmu_detach_dev,
1595         .map            = arm_smmu_map,
1596         .unmap          = arm_smmu_unmap,
1597         .iova_to_phys   = arm_smmu_iova_to_phys,
1598         .domain_has_cap = arm_smmu_domain_has_cap,
1599         .add_device     = arm_smmu_add_device,
1600         .remove_device  = arm_smmu_remove_device,
1601         .pgsize_bitmap  = (SECTION_SIZE |
1602                            ARM_SMMU_PTE_CONT_SIZE |
1603                            PAGE_SIZE),
1604 };
1605
1606 static void arm_smmu_device_reset(struct arm_smmu_device *smmu)
1607 {
1608         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1609         void __iomem *cb_base;
1610         int i = 0;
1611         u32 reg;
1612
1613         /* clear global FSR */
1614         reg = readl_relaxed(ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sGFSR);
1615         writel(reg, ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sGFSR);
1616
1617         /* Mark all SMRn as invalid and all S2CRn as bypass */
1618         for (i = 0; i < smmu->num_mapping_groups; ++i) {
1619                 writel_relaxed(~SMR_VALID, gr0_base + ARM_SMMU_GR0_SMR(i));
1620                 writel_relaxed(S2CR_TYPE_BYPASS, gr0_base + ARM_SMMU_GR0_S2CR(i));
1621         }
1622
1623         /* Make sure all context banks are disabled and clear CB_FSR  */
1624         for (i = 0; i < smmu->num_context_banks; ++i) {
1625                 cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, i);
1626                 writel_relaxed(0, cb_base + ARM_SMMU_CB_SCTLR);
1627                 writel_relaxed(FSR_FAULT, cb_base + ARM_SMMU_CB_FSR);
1628         }
1629
1630         /* Invalidate the TLB, just in case */
1631         writel_relaxed(0, gr0_base + ARM_SMMU_GR0_STLBIALL);
1632         writel_relaxed(0, gr0_base + ARM_SMMU_GR0_TLBIALLH);
1633         writel_relaxed(0, gr0_base + ARM_SMMU_GR0_TLBIALLNSNH);
1634
1635         reg = readl_relaxed(ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sCR0);
1636
1637         /* Enable fault reporting */
1638         reg |= (sCR0_GFRE | sCR0_GFIE | sCR0_GCFGFRE | sCR0_GCFGFIE);
1639
1640         /* Disable TLB broadcasting. */
1641         reg |= (sCR0_VMIDPNE | sCR0_PTM);
1642
1643         /* Enable client access, but bypass when no mapping is found */
1644         reg &= ~(sCR0_CLIENTPD | sCR0_USFCFG);
1645
1646         /* Disable forced broadcasting */
1647         reg &= ~sCR0_FB;
1648
1649         /* Don't upgrade barriers */
1650         reg &= ~(sCR0_BSU_MASK << sCR0_BSU_SHIFT);
1651
1652         /* Push the button */
1653         arm_smmu_tlb_sync(smmu);
1654         writel(reg, ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sCR0);
1655 }
1656
1657 static int arm_smmu_id_size_to_bits(int size)
1658 {
1659         switch (size) {
1660         case 0:
1661                 return 32;
1662         case 1:
1663                 return 36;
1664         case 2:
1665                 return 40;
1666         case 3:
1667                 return 42;
1668         case 4:
1669                 return 44;
1670         case 5:
1671         default:
1672                 return 48;
1673         }
1674 }
1675
1676 static int arm_smmu_device_cfg_probe(struct arm_smmu_device *smmu)
1677 {
1678         unsigned long size;
1679         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1680         u32 id;
1681
1682         dev_notice(smmu->dev, "probing hardware configuration...\n");
1683
1684         /* Primecell ID */
1685         id = readl_relaxed(gr0_base + ARM_SMMU_GR0_PIDR2);
1686         smmu->version = ((id >> PIDR2_ARCH_SHIFT) & PIDR2_ARCH_MASK) + 1;
1687         dev_notice(smmu->dev, "SMMUv%d with:\n", smmu->version);
1688
1689         /* ID0 */
1690         id = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID0);
1691 #ifndef CONFIG_64BIT
1692         if (((id >> ID0_PTFS_SHIFT) & ID0_PTFS_MASK) == ID0_PTFS_V8_ONLY) {
1693                 dev_err(smmu->dev, "\tno v7 descriptor support!\n");
1694                 return -ENODEV;
1695         }
1696 #endif
1697         if (id & ID0_S1TS) {
1698                 smmu->features |= ARM_SMMU_FEAT_TRANS_S1;
1699                 dev_notice(smmu->dev, "\tstage 1 translation\n");
1700         }
1701
1702         if (id & ID0_S2TS) {
1703                 smmu->features |= ARM_SMMU_FEAT_TRANS_S2;
1704                 dev_notice(smmu->dev, "\tstage 2 translation\n");
1705         }
1706
1707         if (id & ID0_NTS) {
1708                 smmu->features |= ARM_SMMU_FEAT_TRANS_NESTED;
1709                 dev_notice(smmu->dev, "\tnested translation\n");
1710         }
1711
1712         if (!(smmu->features &
1713                 (ARM_SMMU_FEAT_TRANS_S1 | ARM_SMMU_FEAT_TRANS_S2 |
1714                  ARM_SMMU_FEAT_TRANS_NESTED))) {
1715                 dev_err(smmu->dev, "\tno translation support!\n");
1716                 return -ENODEV;
1717         }
1718
1719         if (id & ID0_CTTW) {
1720                 smmu->features |= ARM_SMMU_FEAT_COHERENT_WALK;
1721                 dev_notice(smmu->dev, "\tcoherent table walk\n");
1722         }
1723
1724         if (id & ID0_SMS) {
1725                 u32 smr, sid, mask;
1726
1727                 smmu->features |= ARM_SMMU_FEAT_STREAM_MATCH;
1728                 smmu->num_mapping_groups = (id >> ID0_NUMSMRG_SHIFT) &
1729                                            ID0_NUMSMRG_MASK;
1730                 if (smmu->num_mapping_groups == 0) {
1731                         dev_err(smmu->dev,
1732                                 "stream-matching supported, but no SMRs present!\n");
1733                         return -ENODEV;
1734                 }
1735
1736                 smr = SMR_MASK_MASK << SMR_MASK_SHIFT;
1737                 smr |= (SMR_ID_MASK << SMR_ID_SHIFT);
1738                 writel_relaxed(smr, gr0_base + ARM_SMMU_GR0_SMR(0));
1739                 smr = readl_relaxed(gr0_base + ARM_SMMU_GR0_SMR(0));
1740
1741                 mask = (smr >> SMR_MASK_SHIFT) & SMR_MASK_MASK;
1742                 sid = (smr >> SMR_ID_SHIFT) & SMR_ID_MASK;
1743                 if ((mask & sid) != sid) {
1744                         dev_err(smmu->dev,
1745                                 "SMR mask bits (0x%x) insufficient for ID field (0x%x)\n",
1746                                 mask, sid);
1747                         return -ENODEV;
1748                 }
1749
1750                 dev_notice(smmu->dev,
1751                            "\tstream matching with %u register groups, mask 0x%x",
1752                            smmu->num_mapping_groups, mask);
1753         }
1754
1755         /* ID1 */
1756         id = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID1);
1757         smmu->pagesize = (id & ID1_PAGESIZE) ? SZ_64K : SZ_4K;
1758
1759         /* Check for size mismatch of SMMU address space from mapped region */
1760         size = 1 << (((id >> ID1_NUMPAGENDXB_SHIFT) & ID1_NUMPAGENDXB_MASK) + 1);
1761         size *= (smmu->pagesize << 1);
1762         if (smmu->size != size)
1763                 dev_warn(smmu->dev, "SMMU address space size (0x%lx) differs "
1764                         "from mapped region size (0x%lx)!\n", size, smmu->size);
1765
1766         smmu->num_s2_context_banks = (id >> ID1_NUMS2CB_SHIFT) &
1767                                       ID1_NUMS2CB_MASK;
1768         smmu->num_context_banks = (id >> ID1_NUMCB_SHIFT) & ID1_NUMCB_MASK;
1769         if (smmu->num_s2_context_banks > smmu->num_context_banks) {
1770                 dev_err(smmu->dev, "impossible number of S2 context banks!\n");
1771                 return -ENODEV;
1772         }
1773         dev_notice(smmu->dev, "\t%u context banks (%u stage-2 only)\n",
1774                    smmu->num_context_banks, smmu->num_s2_context_banks);
1775
1776         /* ID2 */
1777         id = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID2);
1778         size = arm_smmu_id_size_to_bits((id >> ID2_IAS_SHIFT) & ID2_IAS_MASK);
1779
1780         /*
1781          * Stage-1 output limited by stage-2 input size due to pgd
1782          * allocation (PTRS_PER_PGD).
1783          */
1784 #ifdef CONFIG_64BIT
1785         smmu->s1_output_size = min((unsigned long)VA_BITS, size);
1786 #else
1787         smmu->s1_output_size = min(32UL, size);
1788 #endif
1789
1790         /* The stage-2 output mask is also applied for bypass */
1791         size = arm_smmu_id_size_to_bits((id >> ID2_OAS_SHIFT) & ID2_OAS_MASK);
1792         smmu->s2_output_size = min((unsigned long)PHYS_MASK_SHIFT, size);
1793
1794         if (smmu->version == 1) {
1795                 smmu->input_size = 32;
1796         } else {
1797 #ifdef CONFIG_64BIT
1798                 size = (id >> ID2_UBS_SHIFT) & ID2_UBS_MASK;
1799                 size = min(VA_BITS, arm_smmu_id_size_to_bits(size));
1800 #else
1801                 size = 32;
1802 #endif
1803                 smmu->input_size = size;
1804
1805                 if ((PAGE_SIZE == SZ_4K && !(id & ID2_PTFS_4K)) ||
1806                     (PAGE_SIZE == SZ_64K && !(id & ID2_PTFS_64K)) ||
1807                     (PAGE_SIZE != SZ_4K && PAGE_SIZE != SZ_64K)) {
1808                         dev_err(smmu->dev, "CPU page size 0x%lx unsupported\n",
1809                                 PAGE_SIZE);
1810                         return -ENODEV;
1811                 }
1812         }
1813
1814         dev_notice(smmu->dev,
1815                    "\t%lu-bit VA, %lu-bit IPA, %lu-bit PA\n",
1816                    smmu->input_size, smmu->s1_output_size, smmu->s2_output_size);
1817         return 0;
1818 }
1819
1820 static int arm_smmu_device_dt_probe(struct platform_device *pdev)
1821 {
1822         struct resource *res;
1823         struct arm_smmu_device *smmu;
1824         struct device *dev = &pdev->dev;
1825         struct rb_node *node;
1826         struct of_phandle_args masterspec;
1827         int num_irqs, i, err;
1828
1829         smmu = devm_kzalloc(dev, sizeof(*smmu), GFP_KERNEL);
1830         if (!smmu) {
1831                 dev_err(dev, "failed to allocate arm_smmu_device\n");
1832                 return -ENOMEM;
1833         }
1834         smmu->dev = dev;
1835
1836         res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1837         smmu->base = devm_ioremap_resource(dev, res);
1838         if (IS_ERR(smmu->base))
1839                 return PTR_ERR(smmu->base);
1840         smmu->size = resource_size(res);
1841
1842         if (of_property_read_u32(dev->of_node, "#global-interrupts",
1843                                  &smmu->num_global_irqs)) {
1844                 dev_err(dev, "missing #global-interrupts property\n");
1845                 return -ENODEV;
1846         }
1847
1848         num_irqs = 0;
1849         while ((res = platform_get_resource(pdev, IORESOURCE_IRQ, num_irqs))) {
1850                 num_irqs++;
1851                 if (num_irqs > smmu->num_global_irqs)
1852                         smmu->num_context_irqs++;
1853         }
1854
1855         if (!smmu->num_context_irqs) {
1856                 dev_err(dev, "found %d interrupts but expected at least %d\n",
1857                         num_irqs, smmu->num_global_irqs + 1);
1858                 return -ENODEV;
1859         }
1860
1861         smmu->irqs = devm_kzalloc(dev, sizeof(*smmu->irqs) * num_irqs,
1862                                   GFP_KERNEL);
1863         if (!smmu->irqs) {
1864                 dev_err(dev, "failed to allocate %d irqs\n", num_irqs);
1865                 return -ENOMEM;
1866         }
1867
1868         for (i = 0; i < num_irqs; ++i) {
1869                 int irq = platform_get_irq(pdev, i);
1870                 if (irq < 0) {
1871                         dev_err(dev, "failed to get irq index %d\n", i);
1872                         return -ENODEV;
1873                 }
1874                 smmu->irqs[i] = irq;
1875         }
1876
1877         i = 0;
1878         smmu->masters = RB_ROOT;
1879         while (!of_parse_phandle_with_args(dev->of_node, "mmu-masters",
1880                                            "#stream-id-cells", i,
1881                                            &masterspec)) {
1882                 err = register_smmu_master(smmu, dev, &masterspec);
1883                 if (err) {
1884                         dev_err(dev, "failed to add master %s\n",
1885                                 masterspec.np->name);
1886                         goto out_put_masters;
1887                 }
1888
1889                 i++;
1890         }
1891         dev_notice(dev, "registered %d master devices\n", i);
1892
1893         err = arm_smmu_device_cfg_probe(smmu);
1894         if (err)
1895                 goto out_put_masters;
1896
1897         parse_driver_options(smmu);
1898
1899         if (smmu->version > 1 &&
1900             smmu->num_context_banks != smmu->num_context_irqs) {
1901                 dev_err(dev,
1902                         "found only %d context interrupt(s) but %d required\n",
1903                         smmu->num_context_irqs, smmu->num_context_banks);
1904                 err = -ENODEV;
1905                 goto out_put_masters;
1906         }
1907
1908         for (i = 0; i < smmu->num_global_irqs; ++i) {
1909                 err = request_irq(smmu->irqs[i],
1910                                   arm_smmu_global_fault,
1911                                   IRQF_SHARED,
1912                                   "arm-smmu global fault",
1913                                   smmu);
1914                 if (err) {
1915                         dev_err(dev, "failed to request global IRQ %d (%u)\n",
1916                                 i, smmu->irqs[i]);
1917                         goto out_free_irqs;
1918                 }
1919         }
1920
1921         INIT_LIST_HEAD(&smmu->list);
1922         spin_lock(&arm_smmu_devices_lock);
1923         list_add(&smmu->list, &arm_smmu_devices);
1924         spin_unlock(&arm_smmu_devices_lock);
1925
1926         arm_smmu_device_reset(smmu);
1927         return 0;
1928
1929 out_free_irqs:
1930         while (i--)
1931                 free_irq(smmu->irqs[i], smmu);
1932
1933 out_put_masters:
1934         for (node = rb_first(&smmu->masters); node; node = rb_next(node)) {
1935                 struct arm_smmu_master *master;
1936                 master = container_of(node, struct arm_smmu_master, node);
1937                 of_node_put(master->of_node);
1938         }
1939
1940         return err;
1941 }
1942
1943 static int arm_smmu_device_remove(struct platform_device *pdev)
1944 {
1945         int i;
1946         struct device *dev = &pdev->dev;
1947         struct arm_smmu_device *curr, *smmu = NULL;
1948         struct rb_node *node;
1949
1950         spin_lock(&arm_smmu_devices_lock);
1951         list_for_each_entry(curr, &arm_smmu_devices, list) {
1952                 if (curr->dev == dev) {
1953                         smmu = curr;
1954                         list_del(&smmu->list);
1955                         break;
1956                 }
1957         }
1958         spin_unlock(&arm_smmu_devices_lock);
1959
1960         if (!smmu)
1961                 return -ENODEV;
1962
1963         for (node = rb_first(&smmu->masters); node; node = rb_next(node)) {
1964                 struct arm_smmu_master *master;
1965                 master = container_of(node, struct arm_smmu_master, node);
1966                 of_node_put(master->of_node);
1967         }
1968
1969         if (!bitmap_empty(smmu->context_map, ARM_SMMU_MAX_CBS))
1970                 dev_err(dev, "removing device with active domains!\n");
1971
1972         for (i = 0; i < smmu->num_global_irqs; ++i)
1973                 free_irq(smmu->irqs[i], smmu);
1974
1975         /* Turn the thing off */
1976         writel(sCR0_CLIENTPD,ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sCR0);
1977         return 0;
1978 }
1979
1980 #ifdef CONFIG_OF
1981 static struct of_device_id arm_smmu_of_match[] = {
1982         { .compatible = "arm,smmu-v1", },
1983         { .compatible = "arm,smmu-v2", },
1984         { .compatible = "arm,mmu-400", },
1985         { .compatible = "arm,mmu-500", },
1986         { },
1987 };
1988 MODULE_DEVICE_TABLE(of, arm_smmu_of_match);
1989 #endif
1990
1991 static struct platform_driver arm_smmu_driver = {
1992         .driver = {
1993                 .owner          = THIS_MODULE,
1994                 .name           = "arm-smmu",
1995                 .of_match_table = of_match_ptr(arm_smmu_of_match),
1996         },
1997         .probe  = arm_smmu_device_dt_probe,
1998         .remove = arm_smmu_device_remove,
1999 };
2000
2001 static int __init arm_smmu_init(void)
2002 {
2003         int ret;
2004
2005         ret = platform_driver_register(&arm_smmu_driver);
2006         if (ret)
2007                 return ret;
2008
2009         /* Oh, for a proper bus abstraction */
2010         if (!iommu_present(&platform_bus_type))
2011                 bus_set_iommu(&platform_bus_type, &arm_smmu_ops);
2012
2013 #ifdef CONFIG_ARM_AMBA
2014         if (!iommu_present(&amba_bustype))
2015                 bus_set_iommu(&amba_bustype, &arm_smmu_ops);
2016 #endif
2017
2018 #ifdef CONFIG_PCI
2019         if (!iommu_present(&pci_bus_type))
2020                 bus_set_iommu(&pci_bus_type, &arm_smmu_ops);
2021 #endif
2022
2023         return 0;
2024 }
2025
2026 static void __exit arm_smmu_exit(void)
2027 {
2028         return platform_driver_unregister(&arm_smmu_driver);
2029 }
2030
2031 subsys_initcall(arm_smmu_init);
2032 module_exit(arm_smmu_exit);
2033
2034 MODULE_DESCRIPTION("IOMMU API for ARM architected SMMU implementations");
2035 MODULE_AUTHOR("Will Deacon <will.deacon@arm.com>");
2036 MODULE_LICENSE("GPL v2");