]> git.kernelconcepts.de Git - karo-tx-linux.git/blob - drivers/iommu/arm-smmu.c
arm: imx6: defconfig: update tx6 defconfigs
[karo-tx-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 39-bit addressing
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/platform_device.h>
43 #include <linux/slab.h>
44 #include <linux/spinlock.h>
45
46 #include <linux/amba/bus.h>
47
48 #include <asm/pgalloc.h>
49
50 /* Maximum number of stream IDs assigned to a single device */
51 #define MAX_MASTER_STREAMIDS            8
52
53 /* Maximum number of context banks per SMMU */
54 #define ARM_SMMU_MAX_CBS                128
55
56 /* Maximum number of mapping groups per SMMU */
57 #define ARM_SMMU_MAX_SMRS               128
58
59 /* SMMU global address space */
60 #define ARM_SMMU_GR0(smmu)              ((smmu)->base)
61 #define ARM_SMMU_GR1(smmu)              ((smmu)->base + (smmu)->pagesize)
62
63 /* Page table bits */
64 #define ARM_SMMU_PTE_PAGE               (((pteval_t)3) << 0)
65 #define ARM_SMMU_PTE_CONT               (((pteval_t)1) << 52)
66 #define ARM_SMMU_PTE_AF                 (((pteval_t)1) << 10)
67 #define ARM_SMMU_PTE_SH_NS              (((pteval_t)0) << 8)
68 #define ARM_SMMU_PTE_SH_OS              (((pteval_t)2) << 8)
69 #define ARM_SMMU_PTE_SH_IS              (((pteval_t)3) << 8)
70
71 #if PAGE_SIZE == SZ_4K
72 #define ARM_SMMU_PTE_CONT_ENTRIES       16
73 #elif PAGE_SIZE == SZ_64K
74 #define ARM_SMMU_PTE_CONT_ENTRIES       32
75 #else
76 #define ARM_SMMU_PTE_CONT_ENTRIES       1
77 #endif
78
79 #define ARM_SMMU_PTE_CONT_SIZE          (PAGE_SIZE * ARM_SMMU_PTE_CONT_ENTRIES)
80 #define ARM_SMMU_PTE_CONT_MASK          (~(ARM_SMMU_PTE_CONT_SIZE - 1))
81 #define ARM_SMMU_PTE_HWTABLE_SIZE       (PTRS_PER_PTE * sizeof(pte_t))
82
83 /* Stage-1 PTE */
84 #define ARM_SMMU_PTE_AP_UNPRIV          (((pteval_t)1) << 6)
85 #define ARM_SMMU_PTE_AP_RDONLY          (((pteval_t)2) << 6)
86 #define ARM_SMMU_PTE_ATTRINDX_SHIFT     2
87 #define ARM_SMMU_PTE_nG                 (((pteval_t)1) << 11)
88
89 /* Stage-2 PTE */
90 #define ARM_SMMU_PTE_HAP_FAULT          (((pteval_t)0) << 6)
91 #define ARM_SMMU_PTE_HAP_READ           (((pteval_t)1) << 6)
92 #define ARM_SMMU_PTE_HAP_WRITE          (((pteval_t)2) << 6)
93 #define ARM_SMMU_PTE_MEMATTR_OIWB       (((pteval_t)0xf) << 2)
94 #define ARM_SMMU_PTE_MEMATTR_NC         (((pteval_t)0x5) << 2)
95 #define ARM_SMMU_PTE_MEMATTR_DEV        (((pteval_t)0x1) << 2)
96
97 /* Configuration registers */
98 #define ARM_SMMU_GR0_sCR0               0x0
99 #define sCR0_CLIENTPD                   (1 << 0)
100 #define sCR0_GFRE                       (1 << 1)
101 #define sCR0_GFIE                       (1 << 2)
102 #define sCR0_GCFGFRE                    (1 << 4)
103 #define sCR0_GCFGFIE                    (1 << 5)
104 #define sCR0_USFCFG                     (1 << 10)
105 #define sCR0_VMIDPNE                    (1 << 11)
106 #define sCR0_PTM                        (1 << 12)
107 #define sCR0_FB                         (1 << 13)
108 #define sCR0_BSU_SHIFT                  14
109 #define sCR0_BSU_MASK                   0x3
110
111 /* Identification registers */
112 #define ARM_SMMU_GR0_ID0                0x20
113 #define ARM_SMMU_GR0_ID1                0x24
114 #define ARM_SMMU_GR0_ID2                0x28
115 #define ARM_SMMU_GR0_ID3                0x2c
116 #define ARM_SMMU_GR0_ID4                0x30
117 #define ARM_SMMU_GR0_ID5                0x34
118 #define ARM_SMMU_GR0_ID6                0x38
119 #define ARM_SMMU_GR0_ID7                0x3c
120 #define ARM_SMMU_GR0_sGFSR              0x48
121 #define ARM_SMMU_GR0_sGFSYNR0           0x50
122 #define ARM_SMMU_GR0_sGFSYNR1           0x54
123 #define ARM_SMMU_GR0_sGFSYNR2           0x58
124 #define ARM_SMMU_GR0_PIDR0              0xfe0
125 #define ARM_SMMU_GR0_PIDR1              0xfe4
126 #define ARM_SMMU_GR0_PIDR2              0xfe8
127
128 #define ID0_S1TS                        (1 << 30)
129 #define ID0_S2TS                        (1 << 29)
130 #define ID0_NTS                         (1 << 28)
131 #define ID0_SMS                         (1 << 27)
132 #define ID0_PTFS_SHIFT                  24
133 #define ID0_PTFS_MASK                   0x2
134 #define ID0_PTFS_V8_ONLY                0x2
135 #define ID0_CTTW                        (1 << 14)
136 #define ID0_NUMIRPT_SHIFT               16
137 #define ID0_NUMIRPT_MASK                0xff
138 #define ID0_NUMSMRG_SHIFT               0
139 #define ID0_NUMSMRG_MASK                0xff
140
141 #define ID1_PAGESIZE                    (1 << 31)
142 #define ID1_NUMPAGENDXB_SHIFT           28
143 #define ID1_NUMPAGENDXB_MASK            7
144 #define ID1_NUMS2CB_SHIFT               16
145 #define ID1_NUMS2CB_MASK                0xff
146 #define ID1_NUMCB_SHIFT                 0
147 #define ID1_NUMCB_MASK                  0xff
148
149 #define ID2_OAS_SHIFT                   4
150 #define ID2_OAS_MASK                    0xf
151 #define ID2_IAS_SHIFT                   0
152 #define ID2_IAS_MASK                    0xf
153 #define ID2_UBS_SHIFT                   8
154 #define ID2_UBS_MASK                    0xf
155 #define ID2_PTFS_4K                     (1 << 12)
156 #define ID2_PTFS_16K                    (1 << 13)
157 #define ID2_PTFS_64K                    (1 << 14)
158
159 #define PIDR2_ARCH_SHIFT                4
160 #define PIDR2_ARCH_MASK                 0xf
161
162 /* Global TLB invalidation */
163 #define ARM_SMMU_GR0_STLBIALL           0x60
164 #define ARM_SMMU_GR0_TLBIVMID           0x64
165 #define ARM_SMMU_GR0_TLBIALLNSNH        0x68
166 #define ARM_SMMU_GR0_TLBIALLH           0x6c
167 #define ARM_SMMU_GR0_sTLBGSYNC          0x70
168 #define ARM_SMMU_GR0_sTLBGSTATUS        0x74
169 #define sTLBGSTATUS_GSACTIVE            (1 << 0)
170 #define TLB_LOOP_TIMEOUT                1000000 /* 1s! */
171
172 /* Stream mapping registers */
173 #define ARM_SMMU_GR0_SMR(n)             (0x800 + ((n) << 2))
174 #define SMR_VALID                       (1 << 31)
175 #define SMR_MASK_SHIFT                  16
176 #define SMR_MASK_MASK                   0x7fff
177 #define SMR_ID_SHIFT                    0
178 #define SMR_ID_MASK                     0x7fff
179
180 #define ARM_SMMU_GR0_S2CR(n)            (0xc00 + ((n) << 2))
181 #define S2CR_CBNDX_SHIFT                0
182 #define S2CR_CBNDX_MASK                 0xff
183 #define S2CR_TYPE_SHIFT                 16
184 #define S2CR_TYPE_MASK                  0x3
185 #define S2CR_TYPE_TRANS                 (0 << S2CR_TYPE_SHIFT)
186 #define S2CR_TYPE_BYPASS                (1 << S2CR_TYPE_SHIFT)
187 #define S2CR_TYPE_FAULT                 (2 << S2CR_TYPE_SHIFT)
188
189 /* Context bank attribute registers */
190 #define ARM_SMMU_GR1_CBAR(n)            (0x0 + ((n) << 2))
191 #define CBAR_VMID_SHIFT                 0
192 #define CBAR_VMID_MASK                  0xff
193 #define CBAR_S1_MEMATTR_SHIFT           12
194 #define CBAR_S1_MEMATTR_MASK            0xf
195 #define CBAR_S1_MEMATTR_WB              0xf
196 #define CBAR_TYPE_SHIFT                 16
197 #define CBAR_TYPE_MASK                  0x3
198 #define CBAR_TYPE_S2_TRANS              (0 << CBAR_TYPE_SHIFT)
199 #define CBAR_TYPE_S1_TRANS_S2_BYPASS    (1 << CBAR_TYPE_SHIFT)
200 #define CBAR_TYPE_S1_TRANS_S2_FAULT     (2 << CBAR_TYPE_SHIFT)
201 #define CBAR_TYPE_S1_TRANS_S2_TRANS     (3 << CBAR_TYPE_SHIFT)
202 #define CBAR_IRPTNDX_SHIFT              24
203 #define CBAR_IRPTNDX_MASK               0xff
204
205 #define ARM_SMMU_GR1_CBA2R(n)           (0x800 + ((n) << 2))
206 #define CBA2R_RW64_32BIT                (0 << 0)
207 #define CBA2R_RW64_64BIT                (1 << 0)
208
209 /* Translation context bank */
210 #define ARM_SMMU_CB_BASE(smmu)          ((smmu)->base + ((smmu)->size >> 1))
211 #define ARM_SMMU_CB(smmu, n)            ((n) * (smmu)->pagesize)
212
213 #define ARM_SMMU_CB_SCTLR               0x0
214 #define ARM_SMMU_CB_RESUME              0x8
215 #define ARM_SMMU_CB_TTBCR2              0x10
216 #define ARM_SMMU_CB_TTBR0_LO            0x20
217 #define ARM_SMMU_CB_TTBR0_HI            0x24
218 #define ARM_SMMU_CB_TTBCR               0x30
219 #define ARM_SMMU_CB_S1_MAIR0            0x38
220 #define ARM_SMMU_CB_FSR                 0x58
221 #define ARM_SMMU_CB_FAR_LO              0x60
222 #define ARM_SMMU_CB_FAR_HI              0x64
223 #define ARM_SMMU_CB_FSYNR0              0x68
224 #define ARM_SMMU_CB_S1_TLBIASID         0x610
225
226 #define SCTLR_S1_ASIDPNE                (1 << 12)
227 #define SCTLR_CFCFG                     (1 << 7)
228 #define SCTLR_CFIE                      (1 << 6)
229 #define SCTLR_CFRE                      (1 << 5)
230 #define SCTLR_E                         (1 << 4)
231 #define SCTLR_AFE                       (1 << 2)
232 #define SCTLR_TRE                       (1 << 1)
233 #define SCTLR_M                         (1 << 0)
234 #define SCTLR_EAE_SBOP                  (SCTLR_AFE | SCTLR_TRE)
235
236 #define RESUME_RETRY                    (0 << 0)
237 #define RESUME_TERMINATE                (1 << 0)
238
239 #define TTBCR_EAE                       (1 << 31)
240
241 #define TTBCR_PASIZE_SHIFT              16
242 #define TTBCR_PASIZE_MASK               0x7
243
244 #define TTBCR_TG0_4K                    (0 << 14)
245 #define TTBCR_TG0_64K                   (1 << 14)
246
247 #define TTBCR_SH0_SHIFT                 12
248 #define TTBCR_SH0_MASK                  0x3
249 #define TTBCR_SH_NS                     0
250 #define TTBCR_SH_OS                     2
251 #define TTBCR_SH_IS                     3
252
253 #define TTBCR_ORGN0_SHIFT               10
254 #define TTBCR_IRGN0_SHIFT               8
255 #define TTBCR_RGN_MASK                  0x3
256 #define TTBCR_RGN_NC                    0
257 #define TTBCR_RGN_WBWA                  1
258 #define TTBCR_RGN_WT                    2
259 #define TTBCR_RGN_WB                    3
260
261 #define TTBCR_SL0_SHIFT                 6
262 #define TTBCR_SL0_MASK                  0x3
263 #define TTBCR_SL0_LVL_2                 0
264 #define TTBCR_SL0_LVL_1                 1
265
266 #define TTBCR_T1SZ_SHIFT                16
267 #define TTBCR_T0SZ_SHIFT                0
268 #define TTBCR_SZ_MASK                   0xf
269
270 #define TTBCR2_SEP_SHIFT                15
271 #define TTBCR2_SEP_MASK                 0x7
272
273 #define TTBCR2_PASIZE_SHIFT             0
274 #define TTBCR2_PASIZE_MASK              0x7
275
276 /* Common definitions for PASize and SEP fields */
277 #define TTBCR2_ADDR_32                  0
278 #define TTBCR2_ADDR_36                  1
279 #define TTBCR2_ADDR_40                  2
280 #define TTBCR2_ADDR_42                  3
281 #define TTBCR2_ADDR_44                  4
282 #define TTBCR2_ADDR_48                  5
283
284 #define TTBRn_HI_ASID_SHIFT             16
285
286 #define MAIR_ATTR_SHIFT(n)              ((n) << 3)
287 #define MAIR_ATTR_MASK                  0xff
288 #define MAIR_ATTR_DEVICE                0x04
289 #define MAIR_ATTR_NC                    0x44
290 #define MAIR_ATTR_WBRWA                 0xff
291 #define MAIR_ATTR_IDX_NC                0
292 #define MAIR_ATTR_IDX_CACHE             1
293 #define MAIR_ATTR_IDX_DEV               2
294
295 #define FSR_MULTI                       (1 << 31)
296 #define FSR_SS                          (1 << 30)
297 #define FSR_UUT                         (1 << 8)
298 #define FSR_ASF                         (1 << 7)
299 #define FSR_TLBLKF                      (1 << 6)
300 #define FSR_TLBMCF                      (1 << 5)
301 #define FSR_EF                          (1 << 4)
302 #define FSR_PF                          (1 << 3)
303 #define FSR_AFF                         (1 << 2)
304 #define FSR_TF                          (1 << 1)
305
306 #define FSR_IGN                         (FSR_AFF | FSR_ASF | FSR_TLBMCF |       \
307                                          FSR_TLBLKF)
308 #define FSR_FAULT                       (FSR_MULTI | FSR_SS | FSR_UUT |         \
309                                          FSR_EF | FSR_PF | FSR_TF | FSR_IGN)
310
311 #define FSYNR0_WNR                      (1 << 4)
312
313 struct arm_smmu_smr {
314         u8                              idx;
315         u16                             mask;
316         u16                             id;
317 };
318
319 struct arm_smmu_master {
320         struct device_node              *of_node;
321
322         /*
323          * The following is specific to the master's position in the
324          * SMMU chain.
325          */
326         struct rb_node                  node;
327         int                             num_streamids;
328         u16                             streamids[MAX_MASTER_STREAMIDS];
329
330         /*
331          * We only need to allocate these on the root SMMU, as we
332          * configure unmatched streams to bypass translation.
333          */
334         struct arm_smmu_smr             *smrs;
335 };
336
337 struct arm_smmu_device {
338         struct device                   *dev;
339         struct device_node              *parent_of_node;
340
341         void __iomem                    *base;
342         unsigned long                   size;
343         unsigned long                   pagesize;
344
345 #define ARM_SMMU_FEAT_COHERENT_WALK     (1 << 0)
346 #define ARM_SMMU_FEAT_STREAM_MATCH      (1 << 1)
347 #define ARM_SMMU_FEAT_TRANS_S1          (1 << 2)
348 #define ARM_SMMU_FEAT_TRANS_S2          (1 << 3)
349 #define ARM_SMMU_FEAT_TRANS_NESTED      (1 << 4)
350         u32                             features;
351         int                             version;
352
353         u32                             num_context_banks;
354         u32                             num_s2_context_banks;
355         DECLARE_BITMAP(context_map, ARM_SMMU_MAX_CBS);
356         atomic_t                        irptndx;
357
358         u32                             num_mapping_groups;
359         DECLARE_BITMAP(smr_map, ARM_SMMU_MAX_SMRS);
360
361         unsigned long                   input_size;
362         unsigned long                   s1_output_size;
363         unsigned long                   s2_output_size;
364
365         u32                             num_global_irqs;
366         u32                             num_context_irqs;
367         unsigned int                    *irqs;
368
369         struct list_head                list;
370         struct rb_root                  masters;
371 };
372
373 struct arm_smmu_cfg {
374         struct arm_smmu_device          *smmu;
375         u8                              cbndx;
376         u8                              irptndx;
377         u32                             cbar;
378         pgd_t                           *pgd;
379 };
380 #define INVALID_IRPTNDX                 0xff
381
382 #define ARM_SMMU_CB_ASID(cfg)           ((cfg)->cbndx)
383 #define ARM_SMMU_CB_VMID(cfg)           ((cfg)->cbndx + 1)
384
385 struct arm_smmu_domain {
386         /*
387          * A domain can span across multiple, chained SMMUs and requires
388          * all devices within the domain to follow the same translation
389          * path.
390          */
391         struct arm_smmu_device          *leaf_smmu;
392         struct arm_smmu_cfg             root_cfg;
393         phys_addr_t                     output_mask;
394
395         spinlock_t                      lock;
396 };
397
398 static DEFINE_SPINLOCK(arm_smmu_devices_lock);
399 static LIST_HEAD(arm_smmu_devices);
400
401 static struct arm_smmu_master *find_smmu_master(struct arm_smmu_device *smmu,
402                                                 struct device_node *dev_node)
403 {
404         struct rb_node *node = smmu->masters.rb_node;
405
406         while (node) {
407                 struct arm_smmu_master *master;
408                 master = container_of(node, struct arm_smmu_master, node);
409
410                 if (dev_node < master->of_node)
411                         node = node->rb_left;
412                 else if (dev_node > master->of_node)
413                         node = node->rb_right;
414                 else
415                         return master;
416         }
417
418         return NULL;
419 }
420
421 static int insert_smmu_master(struct arm_smmu_device *smmu,
422                               struct arm_smmu_master *master)
423 {
424         struct rb_node **new, *parent;
425
426         new = &smmu->masters.rb_node;
427         parent = NULL;
428         while (*new) {
429                 struct arm_smmu_master *this;
430                 this = container_of(*new, struct arm_smmu_master, node);
431
432                 parent = *new;
433                 if (master->of_node < this->of_node)
434                         new = &((*new)->rb_left);
435                 else if (master->of_node > this->of_node)
436                         new = &((*new)->rb_right);
437                 else
438                         return -EEXIST;
439         }
440
441         rb_link_node(&master->node, parent, new);
442         rb_insert_color(&master->node, &smmu->masters);
443         return 0;
444 }
445
446 static int register_smmu_master(struct arm_smmu_device *smmu,
447                                 struct device *dev,
448                                 struct of_phandle_args *masterspec)
449 {
450         int i;
451         struct arm_smmu_master *master;
452
453         master = find_smmu_master(smmu, masterspec->np);
454         if (master) {
455                 dev_err(dev,
456                         "rejecting multiple registrations for master device %s\n",
457                         masterspec->np->name);
458                 return -EBUSY;
459         }
460
461         if (masterspec->args_count > MAX_MASTER_STREAMIDS) {
462                 dev_err(dev,
463                         "reached maximum number (%d) of stream IDs for master device %s\n",
464                         MAX_MASTER_STREAMIDS, masterspec->np->name);
465                 return -ENOSPC;
466         }
467
468         master = devm_kzalloc(dev, sizeof(*master), GFP_KERNEL);
469         if (!master)
470                 return -ENOMEM;
471
472         master->of_node         = masterspec->np;
473         master->num_streamids   = masterspec->args_count;
474
475         for (i = 0; i < master->num_streamids; ++i)
476                 master->streamids[i] = masterspec->args[i];
477
478         return insert_smmu_master(smmu, master);
479 }
480
481 static struct arm_smmu_device *find_parent_smmu(struct arm_smmu_device *smmu)
482 {
483         struct arm_smmu_device *parent;
484
485         if (!smmu->parent_of_node)
486                 return NULL;
487
488         spin_lock(&arm_smmu_devices_lock);
489         list_for_each_entry(parent, &arm_smmu_devices, list)
490                 if (parent->dev->of_node == smmu->parent_of_node)
491                         goto out_unlock;
492
493         parent = NULL;
494         dev_warn(smmu->dev,
495                  "Failed to find SMMU parent despite parent in DT\n");
496 out_unlock:
497         spin_unlock(&arm_smmu_devices_lock);
498         return parent;
499 }
500
501 static int __arm_smmu_alloc_bitmap(unsigned long *map, int start, int end)
502 {
503         int idx;
504
505         do {
506                 idx = find_next_zero_bit(map, end, start);
507                 if (idx == end)
508                         return -ENOSPC;
509         } while (test_and_set_bit(idx, map));
510
511         return idx;
512 }
513
514 static void __arm_smmu_free_bitmap(unsigned long *map, int idx)
515 {
516         clear_bit(idx, map);
517 }
518
519 /* Wait for any pending TLB invalidations to complete */
520 static void arm_smmu_tlb_sync(struct arm_smmu_device *smmu)
521 {
522         int count = 0;
523         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
524
525         writel_relaxed(0, gr0_base + ARM_SMMU_GR0_sTLBGSYNC);
526         while (readl_relaxed(gr0_base + ARM_SMMU_GR0_sTLBGSTATUS)
527                & sTLBGSTATUS_GSACTIVE) {
528                 cpu_relax();
529                 if (++count == TLB_LOOP_TIMEOUT) {
530                         dev_err_ratelimited(smmu->dev,
531                         "TLB sync timed out -- SMMU may be deadlocked\n");
532                         return;
533                 }
534                 udelay(1);
535         }
536 }
537
538 static void arm_smmu_tlb_inv_context(struct arm_smmu_cfg *cfg)
539 {
540         struct arm_smmu_device *smmu = cfg->smmu;
541         void __iomem *base = ARM_SMMU_GR0(smmu);
542         bool stage1 = cfg->cbar != CBAR_TYPE_S2_TRANS;
543
544         if (stage1) {
545                 base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
546                 writel_relaxed(ARM_SMMU_CB_ASID(cfg),
547                                base + ARM_SMMU_CB_S1_TLBIASID);
548         } else {
549                 base = ARM_SMMU_GR0(smmu);
550                 writel_relaxed(ARM_SMMU_CB_VMID(cfg),
551                                base + ARM_SMMU_GR0_TLBIVMID);
552         }
553
554         arm_smmu_tlb_sync(smmu);
555 }
556
557 static irqreturn_t arm_smmu_context_fault(int irq, void *dev)
558 {
559         int flags, ret;
560         u32 fsr, far, fsynr, resume;
561         unsigned long iova;
562         struct iommu_domain *domain = dev;
563         struct arm_smmu_domain *smmu_domain = domain->priv;
564         struct arm_smmu_cfg *root_cfg = &smmu_domain->root_cfg;
565         struct arm_smmu_device *smmu = root_cfg->smmu;
566         void __iomem *cb_base;
567
568         cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, root_cfg->cbndx);
569         fsr = readl_relaxed(cb_base + ARM_SMMU_CB_FSR);
570
571         if (!(fsr & FSR_FAULT))
572                 return IRQ_NONE;
573
574         if (fsr & FSR_IGN)
575                 dev_err_ratelimited(smmu->dev,
576                                     "Unexpected context fault (fsr 0x%u)\n",
577                                     fsr);
578
579         fsynr = readl_relaxed(cb_base + ARM_SMMU_CB_FSYNR0);
580         flags = fsynr & FSYNR0_WNR ? IOMMU_FAULT_WRITE : IOMMU_FAULT_READ;
581
582         far = readl_relaxed(cb_base + ARM_SMMU_CB_FAR_LO);
583         iova = far;
584 #ifdef CONFIG_64BIT
585         far = readl_relaxed(cb_base + ARM_SMMU_CB_FAR_HI);
586         iova |= ((unsigned long)far << 32);
587 #endif
588
589         if (!report_iommu_fault(domain, smmu->dev, iova, flags)) {
590                 ret = IRQ_HANDLED;
591                 resume = RESUME_RETRY;
592         } else {
593                 ret = IRQ_NONE;
594                 resume = RESUME_TERMINATE;
595         }
596
597         /* Clear the faulting FSR */
598         writel(fsr, cb_base + ARM_SMMU_CB_FSR);
599
600         /* Retry or terminate any stalled transactions */
601         if (fsr & FSR_SS)
602                 writel_relaxed(resume, cb_base + ARM_SMMU_CB_RESUME);
603
604         return ret;
605 }
606
607 static irqreturn_t arm_smmu_global_fault(int irq, void *dev)
608 {
609         u32 gfsr, gfsynr0, gfsynr1, gfsynr2;
610         struct arm_smmu_device *smmu = dev;
611         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
612
613         gfsr = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSR);
614         if (!gfsr)
615                 return IRQ_NONE;
616
617         gfsynr0 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSYNR0);
618         gfsynr1 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSYNR1);
619         gfsynr2 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSYNR2);
620
621         dev_err_ratelimited(smmu->dev,
622                 "Unexpected global fault, this could be serious\n");
623         dev_err_ratelimited(smmu->dev,
624                 "\tGFSR 0x%08x, GFSYNR0 0x%08x, GFSYNR1 0x%08x, GFSYNR2 0x%08x\n",
625                 gfsr, gfsynr0, gfsynr1, gfsynr2);
626
627         writel(gfsr, gr0_base + ARM_SMMU_GR0_sGFSR);
628         return IRQ_HANDLED;
629 }
630
631 static void arm_smmu_init_context_bank(struct arm_smmu_domain *smmu_domain)
632 {
633         u32 reg;
634         bool stage1;
635         struct arm_smmu_cfg *root_cfg = &smmu_domain->root_cfg;
636         struct arm_smmu_device *smmu = root_cfg->smmu;
637         void __iomem *cb_base, *gr0_base, *gr1_base;
638
639         gr0_base = ARM_SMMU_GR0(smmu);
640         gr1_base = ARM_SMMU_GR1(smmu);
641         stage1 = root_cfg->cbar != CBAR_TYPE_S2_TRANS;
642         cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, root_cfg->cbndx);
643
644         /* CBAR */
645         reg = root_cfg->cbar;
646         if (smmu->version == 1)
647               reg |= root_cfg->irptndx << CBAR_IRPTNDX_SHIFT;
648
649         /* Use the weakest memory type, so it is overridden by the pte */
650         if (stage1)
651                 reg |= (CBAR_S1_MEMATTR_WB << CBAR_S1_MEMATTR_SHIFT);
652         else
653                 reg |= ARM_SMMU_CB_VMID(root_cfg) << CBAR_VMID_SHIFT;
654         writel_relaxed(reg, gr1_base + ARM_SMMU_GR1_CBAR(root_cfg->cbndx));
655
656         if (smmu->version > 1) {
657                 /* CBA2R */
658 #ifdef CONFIG_64BIT
659                 reg = CBA2R_RW64_64BIT;
660 #else
661                 reg = CBA2R_RW64_32BIT;
662 #endif
663                 writel_relaxed(reg,
664                                gr1_base + ARM_SMMU_GR1_CBA2R(root_cfg->cbndx));
665
666                 /* TTBCR2 */
667                 switch (smmu->input_size) {
668                 case 32:
669                         reg = (TTBCR2_ADDR_32 << TTBCR2_SEP_SHIFT);
670                         break;
671                 case 36:
672                         reg = (TTBCR2_ADDR_36 << TTBCR2_SEP_SHIFT);
673                         break;
674                 case 39:
675                         reg = (TTBCR2_ADDR_40 << TTBCR2_SEP_SHIFT);
676                         break;
677                 case 42:
678                         reg = (TTBCR2_ADDR_42 << TTBCR2_SEP_SHIFT);
679                         break;
680                 case 44:
681                         reg = (TTBCR2_ADDR_44 << TTBCR2_SEP_SHIFT);
682                         break;
683                 case 48:
684                         reg = (TTBCR2_ADDR_48 << TTBCR2_SEP_SHIFT);
685                         break;
686                 }
687
688                 switch (smmu->s1_output_size) {
689                 case 32:
690                         reg |= (TTBCR2_ADDR_32 << TTBCR2_PASIZE_SHIFT);
691                         break;
692                 case 36:
693                         reg |= (TTBCR2_ADDR_36 << TTBCR2_PASIZE_SHIFT);
694                         break;
695                 case 39:
696                         reg |= (TTBCR2_ADDR_40 << TTBCR2_PASIZE_SHIFT);
697                         break;
698                 case 42:
699                         reg |= (TTBCR2_ADDR_42 << TTBCR2_PASIZE_SHIFT);
700                         break;
701                 case 44:
702                         reg |= (TTBCR2_ADDR_44 << TTBCR2_PASIZE_SHIFT);
703                         break;
704                 case 48:
705                         reg |= (TTBCR2_ADDR_48 << TTBCR2_PASIZE_SHIFT);
706                         break;
707                 }
708
709                 if (stage1)
710                         writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBCR2);
711         }
712
713         /* TTBR0 */
714         reg = __pa(root_cfg->pgd);
715         writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBR0_LO);
716         reg = (phys_addr_t)__pa(root_cfg->pgd) >> 32;
717         if (stage1)
718                 reg |= ARM_SMMU_CB_ASID(root_cfg) << TTBRn_HI_ASID_SHIFT;
719         writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBR0_HI);
720
721         /*
722          * TTBCR
723          * We use long descriptor, with inner-shareable WBWA tables in TTBR0.
724          */
725         if (smmu->version > 1) {
726                 if (PAGE_SIZE == SZ_4K)
727                         reg = TTBCR_TG0_4K;
728                 else
729                         reg = TTBCR_TG0_64K;
730
731                 if (!stage1) {
732                         switch (smmu->s2_output_size) {
733                         case 32:
734                                 reg |= (TTBCR2_ADDR_32 << TTBCR_PASIZE_SHIFT);
735                                 break;
736                         case 36:
737                                 reg |= (TTBCR2_ADDR_36 << TTBCR_PASIZE_SHIFT);
738                                 break;
739                         case 40:
740                                 reg |= (TTBCR2_ADDR_40 << TTBCR_PASIZE_SHIFT);
741                                 break;
742                         case 42:
743                                 reg |= (TTBCR2_ADDR_42 << TTBCR_PASIZE_SHIFT);
744                                 break;
745                         case 44:
746                                 reg |= (TTBCR2_ADDR_44 << TTBCR_PASIZE_SHIFT);
747                                 break;
748                         case 48:
749                                 reg |= (TTBCR2_ADDR_48 << TTBCR_PASIZE_SHIFT);
750                                 break;
751                         }
752                 } else {
753                         reg |= (64 - smmu->s1_output_size) << TTBCR_T0SZ_SHIFT;
754                 }
755         } else {
756                 reg = 0;
757         }
758
759         reg |= TTBCR_EAE |
760               (TTBCR_SH_IS << TTBCR_SH0_SHIFT) |
761               (TTBCR_RGN_WBWA << TTBCR_ORGN0_SHIFT) |
762               (TTBCR_RGN_WBWA << TTBCR_IRGN0_SHIFT) |
763               (TTBCR_SL0_LVL_1 << TTBCR_SL0_SHIFT);
764         writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBCR);
765
766         /* MAIR0 (stage-1 only) */
767         if (stage1) {
768                 reg = (MAIR_ATTR_NC << MAIR_ATTR_SHIFT(MAIR_ATTR_IDX_NC)) |
769                       (MAIR_ATTR_WBRWA << MAIR_ATTR_SHIFT(MAIR_ATTR_IDX_CACHE)) |
770                       (MAIR_ATTR_DEVICE << MAIR_ATTR_SHIFT(MAIR_ATTR_IDX_DEV));
771                 writel_relaxed(reg, cb_base + ARM_SMMU_CB_S1_MAIR0);
772         }
773
774         /* SCTLR */
775         reg = SCTLR_CFCFG | SCTLR_CFIE | SCTLR_CFRE | SCTLR_M | SCTLR_EAE_SBOP;
776         if (stage1)
777                 reg |= SCTLR_S1_ASIDPNE;
778 #ifdef __BIG_ENDIAN
779         reg |= SCTLR_E;
780 #endif
781         writel(reg, cb_base + ARM_SMMU_CB_SCTLR);
782 }
783
784 static int arm_smmu_init_domain_context(struct iommu_domain *domain,
785                                         struct device *dev)
786 {
787         int irq, ret, start;
788         struct arm_smmu_domain *smmu_domain = domain->priv;
789         struct arm_smmu_cfg *root_cfg = &smmu_domain->root_cfg;
790         struct arm_smmu_device *smmu, *parent;
791
792         /*
793          * Walk the SMMU chain to find the root device for this chain.
794          * We assume that no masters have translations which terminate
795          * early, and therefore check that the root SMMU does indeed have
796          * a StreamID for the master in question.
797          */
798         parent = dev->archdata.iommu;
799         smmu_domain->output_mask = -1;
800         do {
801                 smmu = parent;
802                 smmu_domain->output_mask &= (1ULL << smmu->s2_output_size) - 1;
803         } while ((parent = find_parent_smmu(smmu)));
804
805         if (!find_smmu_master(smmu, dev->of_node)) {
806                 dev_err(dev, "unable to find root SMMU for device\n");
807                 return -ENODEV;
808         }
809
810         if (smmu->features & ARM_SMMU_FEAT_TRANS_NESTED) {
811                 /*
812                  * We will likely want to change this if/when KVM gets
813                  * involved.
814                  */
815                 root_cfg->cbar = CBAR_TYPE_S1_TRANS_S2_BYPASS;
816                 start = smmu->num_s2_context_banks;
817         } else if (smmu->features & ARM_SMMU_FEAT_TRANS_S2) {
818                 root_cfg->cbar = CBAR_TYPE_S2_TRANS;
819                 start = 0;
820         } else {
821                 root_cfg->cbar = CBAR_TYPE_S1_TRANS_S2_BYPASS;
822                 start = smmu->num_s2_context_banks;
823         }
824
825         ret = __arm_smmu_alloc_bitmap(smmu->context_map, start,
826                                       smmu->num_context_banks);
827         if (IS_ERR_VALUE(ret))
828                 return ret;
829
830         root_cfg->cbndx = ret;
831         if (smmu->version == 1) {
832                 root_cfg->irptndx = atomic_inc_return(&smmu->irptndx);
833                 root_cfg->irptndx %= smmu->num_context_irqs;
834         } else {
835                 root_cfg->irptndx = root_cfg->cbndx;
836         }
837
838         irq = smmu->irqs[smmu->num_global_irqs + root_cfg->irptndx];
839         ret = request_irq(irq, arm_smmu_context_fault, IRQF_SHARED,
840                           "arm-smmu-context-fault", domain);
841         if (IS_ERR_VALUE(ret)) {
842                 dev_err(smmu->dev, "failed to request context IRQ %d (%u)\n",
843                         root_cfg->irptndx, irq);
844                 root_cfg->irptndx = INVALID_IRPTNDX;
845                 goto out_free_context;
846         }
847
848         root_cfg->smmu = smmu;
849         arm_smmu_init_context_bank(smmu_domain);
850         return ret;
851
852 out_free_context:
853         __arm_smmu_free_bitmap(smmu->context_map, root_cfg->cbndx);
854         return ret;
855 }
856
857 static void arm_smmu_destroy_domain_context(struct iommu_domain *domain)
858 {
859         struct arm_smmu_domain *smmu_domain = domain->priv;
860         struct arm_smmu_cfg *root_cfg = &smmu_domain->root_cfg;
861         struct arm_smmu_device *smmu = root_cfg->smmu;
862         void __iomem *cb_base;
863         int irq;
864
865         if (!smmu)
866                 return;
867
868         /* Disable the context bank and nuke the TLB before freeing it. */
869         cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, root_cfg->cbndx);
870         writel_relaxed(0, cb_base + ARM_SMMU_CB_SCTLR);
871         arm_smmu_tlb_inv_context(root_cfg);
872
873         if (root_cfg->irptndx != INVALID_IRPTNDX) {
874                 irq = smmu->irqs[smmu->num_global_irqs + root_cfg->irptndx];
875                 free_irq(irq, domain);
876         }
877
878         __arm_smmu_free_bitmap(smmu->context_map, root_cfg->cbndx);
879 }
880
881 static int arm_smmu_domain_init(struct iommu_domain *domain)
882 {
883         struct arm_smmu_domain *smmu_domain;
884         pgd_t *pgd;
885
886         /*
887          * Allocate the domain and initialise some of its data structures.
888          * We can't really do anything meaningful until we've added a
889          * master.
890          */
891         smmu_domain = kzalloc(sizeof(*smmu_domain), GFP_KERNEL);
892         if (!smmu_domain)
893                 return -ENOMEM;
894
895         pgd = kzalloc(PTRS_PER_PGD * sizeof(pgd_t), GFP_KERNEL);
896         if (!pgd)
897                 goto out_free_domain;
898         smmu_domain->root_cfg.pgd = pgd;
899
900         spin_lock_init(&smmu_domain->lock);
901         domain->priv = smmu_domain;
902         return 0;
903
904 out_free_domain:
905         kfree(smmu_domain);
906         return -ENOMEM;
907 }
908
909 static void arm_smmu_free_ptes(pmd_t *pmd)
910 {
911         pgtable_t table = pmd_pgtable(*pmd);
912         pgtable_page_dtor(table);
913         __free_page(table);
914 }
915
916 static void arm_smmu_free_pmds(pud_t *pud)
917 {
918         int i;
919         pmd_t *pmd, *pmd_base = pmd_offset(pud, 0);
920
921         pmd = pmd_base;
922         for (i = 0; i < PTRS_PER_PMD; ++i) {
923                 if (pmd_none(*pmd))
924                         continue;
925
926                 arm_smmu_free_ptes(pmd);
927                 pmd++;
928         }
929
930         pmd_free(NULL, pmd_base);
931 }
932
933 static void arm_smmu_free_puds(pgd_t *pgd)
934 {
935         int i;
936         pud_t *pud, *pud_base = pud_offset(pgd, 0);
937
938         pud = pud_base;
939         for (i = 0; i < PTRS_PER_PUD; ++i) {
940                 if (pud_none(*pud))
941                         continue;
942
943                 arm_smmu_free_pmds(pud);
944                 pud++;
945         }
946
947         pud_free(NULL, pud_base);
948 }
949
950 static void arm_smmu_free_pgtables(struct arm_smmu_domain *smmu_domain)
951 {
952         int i;
953         struct arm_smmu_cfg *root_cfg = &smmu_domain->root_cfg;
954         pgd_t *pgd, *pgd_base = root_cfg->pgd;
955
956         /*
957          * Recursively free the page tables for this domain. We don't
958          * care about speculative TLB filling, because the TLB will be
959          * nuked next time this context bank is re-allocated and no devices
960          * currently map to these tables.
961          */
962         pgd = pgd_base;
963         for (i = 0; i < PTRS_PER_PGD; ++i) {
964                 if (pgd_none(*pgd))
965                         continue;
966                 arm_smmu_free_puds(pgd);
967                 pgd++;
968         }
969
970         kfree(pgd_base);
971 }
972
973 static void arm_smmu_domain_destroy(struct iommu_domain *domain)
974 {
975         struct arm_smmu_domain *smmu_domain = domain->priv;
976
977         /*
978          * Free the domain resources. We assume that all devices have
979          * already been detached.
980          */
981         arm_smmu_destroy_domain_context(domain);
982         arm_smmu_free_pgtables(smmu_domain);
983         kfree(smmu_domain);
984 }
985
986 static int arm_smmu_master_configure_smrs(struct arm_smmu_device *smmu,
987                                           struct arm_smmu_master *master)
988 {
989         int i;
990         struct arm_smmu_smr *smrs;
991         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
992
993         if (!(smmu->features & ARM_SMMU_FEAT_STREAM_MATCH))
994                 return 0;
995
996         if (master->smrs)
997                 return -EEXIST;
998
999         smrs = kmalloc(sizeof(*smrs) * master->num_streamids, GFP_KERNEL);
1000         if (!smrs) {
1001                 dev_err(smmu->dev, "failed to allocate %d SMRs for master %s\n",
1002                         master->num_streamids, master->of_node->name);
1003                 return -ENOMEM;
1004         }
1005
1006         /* Allocate the SMRs on the root SMMU */
1007         for (i = 0; i < master->num_streamids; ++i) {
1008                 int idx = __arm_smmu_alloc_bitmap(smmu->smr_map, 0,
1009                                                   smmu->num_mapping_groups);
1010                 if (IS_ERR_VALUE(idx)) {
1011                         dev_err(smmu->dev, "failed to allocate free SMR\n");
1012                         goto err_free_smrs;
1013                 }
1014
1015                 smrs[i] = (struct arm_smmu_smr) {
1016                         .idx    = idx,
1017                         .mask   = 0, /* We don't currently share SMRs */
1018                         .id     = master->streamids[i],
1019                 };
1020         }
1021
1022         /* It worked! Now, poke the actual hardware */
1023         for (i = 0; i < master->num_streamids; ++i) {
1024                 u32 reg = SMR_VALID | smrs[i].id << SMR_ID_SHIFT |
1025                           smrs[i].mask << SMR_MASK_SHIFT;
1026                 writel_relaxed(reg, gr0_base + ARM_SMMU_GR0_SMR(smrs[i].idx));
1027         }
1028
1029         master->smrs = smrs;
1030         return 0;
1031
1032 err_free_smrs:
1033         while (--i >= 0)
1034                 __arm_smmu_free_bitmap(smmu->smr_map, smrs[i].idx);
1035         kfree(smrs);
1036         return -ENOSPC;
1037 }
1038
1039 static void arm_smmu_master_free_smrs(struct arm_smmu_device *smmu,
1040                                       struct arm_smmu_master *master)
1041 {
1042         int i;
1043         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1044         struct arm_smmu_smr *smrs = master->smrs;
1045
1046         /* Invalidate the SMRs before freeing back to the allocator */
1047         for (i = 0; i < master->num_streamids; ++i) {
1048                 u8 idx = smrs[i].idx;
1049                 writel_relaxed(~SMR_VALID, gr0_base + ARM_SMMU_GR0_SMR(idx));
1050                 __arm_smmu_free_bitmap(smmu->smr_map, idx);
1051         }
1052
1053         master->smrs = NULL;
1054         kfree(smrs);
1055 }
1056
1057 static void arm_smmu_bypass_stream_mapping(struct arm_smmu_device *smmu,
1058                                            struct arm_smmu_master *master)
1059 {
1060         int i;
1061         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1062
1063         for (i = 0; i < master->num_streamids; ++i) {
1064                 u16 sid = master->streamids[i];
1065                 writel_relaxed(S2CR_TYPE_BYPASS,
1066                                gr0_base + ARM_SMMU_GR0_S2CR(sid));
1067         }
1068 }
1069
1070 static int arm_smmu_domain_add_master(struct arm_smmu_domain *smmu_domain,
1071                                       struct arm_smmu_master *master)
1072 {
1073         int i, ret;
1074         struct arm_smmu_device *parent, *smmu = smmu_domain->root_cfg.smmu;
1075         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1076
1077         ret = arm_smmu_master_configure_smrs(smmu, master);
1078         if (ret)
1079                 return ret;
1080
1081         /* Bypass the leaves */
1082         smmu = smmu_domain->leaf_smmu;
1083         while ((parent = find_parent_smmu(smmu))) {
1084                 /*
1085                  * We won't have a StreamID match for anything but the root
1086                  * smmu, so we only need to worry about StreamID indexing,
1087                  * where we must install bypass entries in the S2CRs.
1088                  */
1089                 if (smmu->features & ARM_SMMU_FEAT_STREAM_MATCH)
1090                         continue;
1091
1092                 arm_smmu_bypass_stream_mapping(smmu, master);
1093                 smmu = parent;
1094         }
1095
1096         /* Now we're at the root, time to point at our context bank */
1097         for (i = 0; i < master->num_streamids; ++i) {
1098                 u32 idx, s2cr;
1099                 idx = master->smrs ? master->smrs[i].idx : master->streamids[i];
1100                 s2cr = (S2CR_TYPE_TRANS << S2CR_TYPE_SHIFT) |
1101                        (smmu_domain->root_cfg.cbndx << S2CR_CBNDX_SHIFT);
1102                 writel_relaxed(s2cr, gr0_base + ARM_SMMU_GR0_S2CR(idx));
1103         }
1104
1105         return 0;
1106 }
1107
1108 static void arm_smmu_domain_remove_master(struct arm_smmu_domain *smmu_domain,
1109                                           struct arm_smmu_master *master)
1110 {
1111         struct arm_smmu_device *smmu = smmu_domain->root_cfg.smmu;
1112
1113         /*
1114          * We *must* clear the S2CR first, because freeing the SMR means
1115          * that it can be re-allocated immediately.
1116          */
1117         arm_smmu_bypass_stream_mapping(smmu, master);
1118         arm_smmu_master_free_smrs(smmu, master);
1119 }
1120
1121 static int arm_smmu_attach_dev(struct iommu_domain *domain, struct device *dev)
1122 {
1123         int ret = -EINVAL;
1124         struct arm_smmu_domain *smmu_domain = domain->priv;
1125         struct arm_smmu_device *device_smmu = dev->archdata.iommu;
1126         struct arm_smmu_master *master;
1127
1128         if (!device_smmu) {
1129                 dev_err(dev, "cannot attach to SMMU, is it on the same bus?\n");
1130                 return -ENXIO;
1131         }
1132
1133         /*
1134          * Sanity check the domain. We don't currently support domains
1135          * that cross between different SMMU chains.
1136          */
1137         spin_lock(&smmu_domain->lock);
1138         if (!smmu_domain->leaf_smmu) {
1139                 /* Now that we have a master, we can finalise the domain */
1140                 ret = arm_smmu_init_domain_context(domain, dev);
1141                 if (IS_ERR_VALUE(ret))
1142                         goto err_unlock;
1143
1144                 smmu_domain->leaf_smmu = device_smmu;
1145         } else if (smmu_domain->leaf_smmu != device_smmu) {
1146                 dev_err(dev,
1147                         "cannot attach to SMMU %s whilst already attached to domain on SMMU %s\n",
1148                         dev_name(smmu_domain->leaf_smmu->dev),
1149                         dev_name(device_smmu->dev));
1150                 goto err_unlock;
1151         }
1152         spin_unlock(&smmu_domain->lock);
1153
1154         /* Looks ok, so add the device to the domain */
1155         master = find_smmu_master(smmu_domain->leaf_smmu, dev->of_node);
1156         if (!master)
1157                 return -ENODEV;
1158
1159         return arm_smmu_domain_add_master(smmu_domain, master);
1160
1161 err_unlock:
1162         spin_unlock(&smmu_domain->lock);
1163         return ret;
1164 }
1165
1166 static void arm_smmu_detach_dev(struct iommu_domain *domain, struct device *dev)
1167 {
1168         struct arm_smmu_domain *smmu_domain = domain->priv;
1169         struct arm_smmu_master *master;
1170
1171         master = find_smmu_master(smmu_domain->leaf_smmu, dev->of_node);
1172         if (master)
1173                 arm_smmu_domain_remove_master(smmu_domain, master);
1174 }
1175
1176 static void arm_smmu_flush_pgtable(struct arm_smmu_device *smmu, void *addr,
1177                                    size_t size)
1178 {
1179         unsigned long offset = (unsigned long)addr & ~PAGE_MASK;
1180
1181         /*
1182          * If the SMMU can't walk tables in the CPU caches, treat them
1183          * like non-coherent DMA since we need to flush the new entries
1184          * all the way out to memory. There's no possibility of recursion
1185          * here as the SMMU table walker will not be wired through another
1186          * SMMU.
1187          */
1188         if (!(smmu->features & ARM_SMMU_FEAT_COHERENT_WALK))
1189                 dma_map_page(smmu->dev, virt_to_page(addr), offset, size,
1190                              DMA_TO_DEVICE);
1191 }
1192
1193 static bool arm_smmu_pte_is_contiguous_range(unsigned long addr,
1194                                              unsigned long end)
1195 {
1196         return !(addr & ~ARM_SMMU_PTE_CONT_MASK) &&
1197                 (addr + ARM_SMMU_PTE_CONT_SIZE <= end);
1198 }
1199
1200 static int arm_smmu_alloc_init_pte(struct arm_smmu_device *smmu, pmd_t *pmd,
1201                                    unsigned long addr, unsigned long end,
1202                                    unsigned long pfn, int flags, int stage)
1203 {
1204         pte_t *pte, *start;
1205         pteval_t pteval = ARM_SMMU_PTE_PAGE | ARM_SMMU_PTE_AF;
1206
1207         if (pmd_none(*pmd)) {
1208                 /* Allocate a new set of tables */
1209                 pgtable_t table = alloc_page(PGALLOC_GFP);
1210                 if (!table)
1211                         return -ENOMEM;
1212
1213                 arm_smmu_flush_pgtable(smmu, page_address(table),
1214                                        ARM_SMMU_PTE_HWTABLE_SIZE);
1215                 pgtable_page_ctor(table);
1216                 pmd_populate(NULL, pmd, table);
1217                 arm_smmu_flush_pgtable(smmu, pmd, sizeof(*pmd));
1218         }
1219
1220         if (stage == 1) {
1221                 pteval |= ARM_SMMU_PTE_AP_UNPRIV | ARM_SMMU_PTE_nG;
1222                 if (!(flags & IOMMU_WRITE) && (flags & IOMMU_READ))
1223                         pteval |= ARM_SMMU_PTE_AP_RDONLY;
1224
1225                 if (flags & IOMMU_CACHE)
1226                         pteval |= (MAIR_ATTR_IDX_CACHE <<
1227                                    ARM_SMMU_PTE_ATTRINDX_SHIFT);
1228         } else {
1229                 pteval |= ARM_SMMU_PTE_HAP_FAULT;
1230                 if (flags & IOMMU_READ)
1231                         pteval |= ARM_SMMU_PTE_HAP_READ;
1232                 if (flags & IOMMU_WRITE)
1233                         pteval |= ARM_SMMU_PTE_HAP_WRITE;
1234                 if (flags & IOMMU_CACHE)
1235                         pteval |= ARM_SMMU_PTE_MEMATTR_OIWB;
1236                 else
1237                         pteval |= ARM_SMMU_PTE_MEMATTR_NC;
1238         }
1239
1240         /* If no access, create a faulting entry to avoid TLB fills */
1241         if (!(flags & (IOMMU_READ | IOMMU_WRITE)))
1242                 pteval &= ~ARM_SMMU_PTE_PAGE;
1243
1244         pteval |= ARM_SMMU_PTE_SH_IS;
1245         start = pmd_page_vaddr(*pmd) + pte_index(addr);
1246         pte = start;
1247
1248         /*
1249          * Install the page table entries. This is fairly complicated
1250          * since we attempt to make use of the contiguous hint in the
1251          * ptes where possible. The contiguous hint indicates a series
1252          * of ARM_SMMU_PTE_CONT_ENTRIES ptes mapping a physically
1253          * contiguous region with the following constraints:
1254          *
1255          *   - The region start is aligned to ARM_SMMU_PTE_CONT_SIZE
1256          *   - Each pte in the region has the contiguous hint bit set
1257          *
1258          * This complicates unmapping (also handled by this code, when
1259          * neither IOMMU_READ or IOMMU_WRITE are set) because it is
1260          * possible, yet highly unlikely, that a client may unmap only
1261          * part of a contiguous range. This requires clearing of the
1262          * contiguous hint bits in the range before installing the new
1263          * faulting entries.
1264          *
1265          * Note that re-mapping an address range without first unmapping
1266          * it is not supported, so TLB invalidation is not required here
1267          * and is instead performed at unmap and domain-init time.
1268          */
1269         do {
1270                 int i = 1;
1271                 pteval &= ~ARM_SMMU_PTE_CONT;
1272
1273                 if (arm_smmu_pte_is_contiguous_range(addr, end)) {
1274                         i = ARM_SMMU_PTE_CONT_ENTRIES;
1275                         pteval |= ARM_SMMU_PTE_CONT;
1276                 } else if (pte_val(*pte) &
1277                            (ARM_SMMU_PTE_CONT | ARM_SMMU_PTE_PAGE)) {
1278                         int j;
1279                         pte_t *cont_start;
1280                         unsigned long idx = pte_index(addr);
1281
1282                         idx &= ~(ARM_SMMU_PTE_CONT_ENTRIES - 1);
1283                         cont_start = pmd_page_vaddr(*pmd) + idx;
1284                         for (j = 0; j < ARM_SMMU_PTE_CONT_ENTRIES; ++j)
1285                                 pte_val(*(cont_start + j)) &= ~ARM_SMMU_PTE_CONT;
1286
1287                         arm_smmu_flush_pgtable(smmu, cont_start,
1288                                                sizeof(*pte) *
1289                                                ARM_SMMU_PTE_CONT_ENTRIES);
1290                 }
1291
1292                 do {
1293                         *pte = pfn_pte(pfn, __pgprot(pteval));
1294                 } while (pte++, pfn++, addr += PAGE_SIZE, --i);
1295         } while (addr != end);
1296
1297         arm_smmu_flush_pgtable(smmu, start, sizeof(*pte) * (pte - start));
1298         return 0;
1299 }
1300
1301 static int arm_smmu_alloc_init_pmd(struct arm_smmu_device *smmu, pud_t *pud,
1302                                    unsigned long addr, unsigned long end,
1303                                    phys_addr_t phys, int flags, int stage)
1304 {
1305         int ret;
1306         pmd_t *pmd;
1307         unsigned long next, pfn = __phys_to_pfn(phys);
1308
1309 #ifndef __PAGETABLE_PMD_FOLDED
1310         if (pud_none(*pud)) {
1311                 pmd = pmd_alloc_one(NULL, addr);
1312                 if (!pmd)
1313                         return -ENOMEM;
1314         } else
1315 #endif
1316                 pmd = pmd_offset(pud, addr);
1317
1318         do {
1319                 next = pmd_addr_end(addr, end);
1320                 ret = arm_smmu_alloc_init_pte(smmu, pmd, addr, end, pfn,
1321                                               flags, stage);
1322                 pud_populate(NULL, pud, pmd);
1323                 arm_smmu_flush_pgtable(smmu, pud, sizeof(*pud));
1324                 phys += next - addr;
1325         } while (pmd++, addr = next, addr < end);
1326
1327         return ret;
1328 }
1329
1330 static int arm_smmu_alloc_init_pud(struct arm_smmu_device *smmu, pgd_t *pgd,
1331                                    unsigned long addr, unsigned long end,
1332                                    phys_addr_t phys, int flags, int stage)
1333 {
1334         int ret = 0;
1335         pud_t *pud;
1336         unsigned long next;
1337
1338 #ifndef __PAGETABLE_PUD_FOLDED
1339         if (pgd_none(*pgd)) {
1340                 pud = pud_alloc_one(NULL, addr);
1341                 if (!pud)
1342                         return -ENOMEM;
1343         } else
1344 #endif
1345                 pud = pud_offset(pgd, addr);
1346
1347         do {
1348                 next = pud_addr_end(addr, end);
1349                 ret = arm_smmu_alloc_init_pmd(smmu, pud, addr, next, phys,
1350                                               flags, stage);
1351                 pgd_populate(NULL, pud, pgd);
1352                 arm_smmu_flush_pgtable(smmu, pgd, sizeof(*pgd));
1353                 phys += next - addr;
1354         } while (pud++, addr = next, addr < end);
1355
1356         return ret;
1357 }
1358
1359 static int arm_smmu_handle_mapping(struct arm_smmu_domain *smmu_domain,
1360                                    unsigned long iova, phys_addr_t paddr,
1361                                    size_t size, int flags)
1362 {
1363         int ret, stage;
1364         unsigned long end;
1365         phys_addr_t input_mask, output_mask;
1366         struct arm_smmu_cfg *root_cfg = &smmu_domain->root_cfg;
1367         pgd_t *pgd = root_cfg->pgd;
1368         struct arm_smmu_device *smmu = root_cfg->smmu;
1369
1370         if (root_cfg->cbar == CBAR_TYPE_S2_TRANS) {
1371                 stage = 2;
1372                 output_mask = (1ULL << smmu->s2_output_size) - 1;
1373         } else {
1374                 stage = 1;
1375                 output_mask = (1ULL << smmu->s1_output_size) - 1;
1376         }
1377
1378         if (!pgd)
1379                 return -EINVAL;
1380
1381         if (size & ~PAGE_MASK)
1382                 return -EINVAL;
1383
1384         input_mask = (1ULL << smmu->input_size) - 1;
1385         if ((phys_addr_t)iova & ~input_mask)
1386                 return -ERANGE;
1387
1388         if (paddr & ~output_mask)
1389                 return -ERANGE;
1390
1391         spin_lock(&smmu_domain->lock);
1392         pgd += pgd_index(iova);
1393         end = iova + size;
1394         do {
1395                 unsigned long next = pgd_addr_end(iova, end);
1396
1397                 ret = arm_smmu_alloc_init_pud(smmu, pgd, iova, next, paddr,
1398                                               flags, stage);
1399                 if (ret)
1400                         goto out_unlock;
1401
1402                 paddr += next - iova;
1403                 iova = next;
1404         } while (pgd++, iova != end);
1405
1406 out_unlock:
1407         spin_unlock(&smmu_domain->lock);
1408
1409         /* Ensure new page tables are visible to the hardware walker */
1410         if (smmu->features & ARM_SMMU_FEAT_COHERENT_WALK)
1411                 dsb();
1412
1413         return ret;
1414 }
1415
1416 static int arm_smmu_map(struct iommu_domain *domain, unsigned long iova,
1417                         phys_addr_t paddr, size_t size, int flags)
1418 {
1419         struct arm_smmu_domain *smmu_domain = domain->priv;
1420         struct arm_smmu_device *smmu = smmu_domain->leaf_smmu;
1421
1422         if (!smmu_domain || !smmu)
1423                 return -ENODEV;
1424
1425         /* Check for silent address truncation up the SMMU chain. */
1426         if ((phys_addr_t)iova & ~smmu_domain->output_mask)
1427                 return -ERANGE;
1428
1429         return arm_smmu_handle_mapping(smmu_domain, iova, paddr, size, flags);
1430 }
1431
1432 static size_t arm_smmu_unmap(struct iommu_domain *domain, unsigned long iova,
1433                              size_t size)
1434 {
1435         int ret;
1436         struct arm_smmu_domain *smmu_domain = domain->priv;
1437
1438         ret = arm_smmu_handle_mapping(smmu_domain, iova, 0, size, 0);
1439         arm_smmu_tlb_inv_context(&smmu_domain->root_cfg);
1440         return ret ? ret : size;
1441 }
1442
1443 static phys_addr_t arm_smmu_iova_to_phys(struct iommu_domain *domain,
1444                                          dma_addr_t iova)
1445 {
1446         pgd_t *pgd;
1447         pud_t *pud;
1448         pmd_t *pmd;
1449         pte_t *pte;
1450         struct arm_smmu_domain *smmu_domain = domain->priv;
1451         struct arm_smmu_cfg *root_cfg = &smmu_domain->root_cfg;
1452         struct arm_smmu_device *smmu = root_cfg->smmu;
1453
1454         spin_lock(&smmu_domain->lock);
1455         pgd = root_cfg->pgd;
1456         if (!pgd)
1457                 goto err_unlock;
1458
1459         pgd += pgd_index(iova);
1460         if (pgd_none_or_clear_bad(pgd))
1461                 goto err_unlock;
1462
1463         pud = pud_offset(pgd, iova);
1464         if (pud_none_or_clear_bad(pud))
1465                 goto err_unlock;
1466
1467         pmd = pmd_offset(pud, iova);
1468         if (pmd_none_or_clear_bad(pmd))
1469                 goto err_unlock;
1470
1471         pte = pmd_page_vaddr(*pmd) + pte_index(iova);
1472         if (pte_none(pte))
1473                 goto err_unlock;
1474
1475         spin_unlock(&smmu_domain->lock);
1476         return __pfn_to_phys(pte_pfn(*pte)) | (iova & ~PAGE_MASK);
1477
1478 err_unlock:
1479         spin_unlock(&smmu_domain->lock);
1480         dev_warn(smmu->dev,
1481                  "invalid (corrupt?) page tables detected for iova 0x%llx\n",
1482                  (unsigned long long)iova);
1483         return -EINVAL;
1484 }
1485
1486 static int arm_smmu_domain_has_cap(struct iommu_domain *domain,
1487                                    unsigned long cap)
1488 {
1489         unsigned long caps = 0;
1490         struct arm_smmu_domain *smmu_domain = domain->priv;
1491
1492         if (smmu_domain->root_cfg.smmu->features & ARM_SMMU_FEAT_COHERENT_WALK)
1493                 caps |= IOMMU_CAP_CACHE_COHERENCY;
1494
1495         return !!(cap & caps);
1496 }
1497
1498 static int arm_smmu_add_device(struct device *dev)
1499 {
1500         struct arm_smmu_device *child, *parent, *smmu;
1501         struct arm_smmu_master *master = NULL;
1502
1503         spin_lock(&arm_smmu_devices_lock);
1504         list_for_each_entry(parent, &arm_smmu_devices, list) {
1505                 smmu = parent;
1506
1507                 /* Try to find a child of the current SMMU. */
1508                 list_for_each_entry(child, &arm_smmu_devices, list) {
1509                         if (child->parent_of_node == parent->dev->of_node) {
1510                                 /* Does the child sit above our master? */
1511                                 master = find_smmu_master(child, dev->of_node);
1512                                 if (master) {
1513                                         smmu = NULL;
1514                                         break;
1515                                 }
1516                         }
1517                 }
1518
1519                 /* We found some children, so keep searching. */
1520                 if (!smmu) {
1521                         master = NULL;
1522                         continue;
1523                 }
1524
1525                 master = find_smmu_master(smmu, dev->of_node);
1526                 if (master)
1527                         break;
1528         }
1529         spin_unlock(&arm_smmu_devices_lock);
1530
1531         if (!master)
1532                 return -ENODEV;
1533
1534         dev->archdata.iommu = smmu;
1535         return 0;
1536 }
1537
1538 static void arm_smmu_remove_device(struct device *dev)
1539 {
1540         dev->archdata.iommu = NULL;
1541 }
1542
1543 static struct iommu_ops arm_smmu_ops = {
1544         .domain_init    = arm_smmu_domain_init,
1545         .domain_destroy = arm_smmu_domain_destroy,
1546         .attach_dev     = arm_smmu_attach_dev,
1547         .detach_dev     = arm_smmu_detach_dev,
1548         .map            = arm_smmu_map,
1549         .unmap          = arm_smmu_unmap,
1550         .iova_to_phys   = arm_smmu_iova_to_phys,
1551         .domain_has_cap = arm_smmu_domain_has_cap,
1552         .add_device     = arm_smmu_add_device,
1553         .remove_device  = arm_smmu_remove_device,
1554         .pgsize_bitmap  = (SECTION_SIZE |
1555                            ARM_SMMU_PTE_CONT_SIZE |
1556                            PAGE_SIZE),
1557 };
1558
1559 static void arm_smmu_device_reset(struct arm_smmu_device *smmu)
1560 {
1561         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1562         void __iomem *sctlr_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB_SCTLR;
1563         int i = 0;
1564         u32 scr0 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sCR0);
1565
1566         /* Mark all SMRn as invalid and all S2CRn as bypass */
1567         for (i = 0; i < smmu->num_mapping_groups; ++i) {
1568                 writel_relaxed(~SMR_VALID, gr0_base + ARM_SMMU_GR0_SMR(i));
1569                 writel_relaxed(S2CR_TYPE_BYPASS, gr0_base + ARM_SMMU_GR0_S2CR(i));
1570         }
1571
1572         /* Make sure all context banks are disabled */
1573         for (i = 0; i < smmu->num_context_banks; ++i)
1574                 writel_relaxed(0, sctlr_base + ARM_SMMU_CB(smmu, i));
1575
1576         /* Invalidate the TLB, just in case */
1577         writel_relaxed(0, gr0_base + ARM_SMMU_GR0_STLBIALL);
1578         writel_relaxed(0, gr0_base + ARM_SMMU_GR0_TLBIALLH);
1579         writel_relaxed(0, gr0_base + ARM_SMMU_GR0_TLBIALLNSNH);
1580
1581         /* Enable fault reporting */
1582         scr0 |= (sCR0_GFRE | sCR0_GFIE | sCR0_GCFGFRE | sCR0_GCFGFIE);
1583
1584         /* Disable TLB broadcasting. */
1585         scr0 |= (sCR0_VMIDPNE | sCR0_PTM);
1586
1587         /* Enable client access, but bypass when no mapping is found */
1588         scr0 &= ~(sCR0_CLIENTPD | sCR0_USFCFG);
1589
1590         /* Disable forced broadcasting */
1591         scr0 &= ~sCR0_FB;
1592
1593         /* Don't upgrade barriers */
1594         scr0 &= ~(sCR0_BSU_MASK << sCR0_BSU_SHIFT);
1595
1596         /* Push the button */
1597         arm_smmu_tlb_sync(smmu);
1598         writel(scr0, gr0_base + ARM_SMMU_GR0_sCR0);
1599 }
1600
1601 static int arm_smmu_id_size_to_bits(int size)
1602 {
1603         switch (size) {
1604         case 0:
1605                 return 32;
1606         case 1:
1607                 return 36;
1608         case 2:
1609                 return 40;
1610         case 3:
1611                 return 42;
1612         case 4:
1613                 return 44;
1614         case 5:
1615         default:
1616                 return 48;
1617         }
1618 }
1619
1620 static int arm_smmu_device_cfg_probe(struct arm_smmu_device *smmu)
1621 {
1622         unsigned long size;
1623         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1624         u32 id;
1625
1626         dev_notice(smmu->dev, "probing hardware configuration...\n");
1627
1628         /* Primecell ID */
1629         id = readl_relaxed(gr0_base + ARM_SMMU_GR0_PIDR2);
1630         smmu->version = ((id >> PIDR2_ARCH_SHIFT) & PIDR2_ARCH_MASK) + 1;
1631         dev_notice(smmu->dev, "SMMUv%d with:\n", smmu->version);
1632
1633         /* ID0 */
1634         id = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID0);
1635 #ifndef CONFIG_64BIT
1636         if (((id >> ID0_PTFS_SHIFT) & ID0_PTFS_MASK) == ID0_PTFS_V8_ONLY) {
1637                 dev_err(smmu->dev, "\tno v7 descriptor support!\n");
1638                 return -ENODEV;
1639         }
1640 #endif
1641         if (id & ID0_S1TS) {
1642                 smmu->features |= ARM_SMMU_FEAT_TRANS_S1;
1643                 dev_notice(smmu->dev, "\tstage 1 translation\n");
1644         }
1645
1646         if (id & ID0_S2TS) {
1647                 smmu->features |= ARM_SMMU_FEAT_TRANS_S2;
1648                 dev_notice(smmu->dev, "\tstage 2 translation\n");
1649         }
1650
1651         if (id & ID0_NTS) {
1652                 smmu->features |= ARM_SMMU_FEAT_TRANS_NESTED;
1653                 dev_notice(smmu->dev, "\tnested translation\n");
1654         }
1655
1656         if (!(smmu->features &
1657                 (ARM_SMMU_FEAT_TRANS_S1 | ARM_SMMU_FEAT_TRANS_S2 |
1658                  ARM_SMMU_FEAT_TRANS_NESTED))) {
1659                 dev_err(smmu->dev, "\tno translation support!\n");
1660                 return -ENODEV;
1661         }
1662
1663         if (id & ID0_CTTW) {
1664                 smmu->features |= ARM_SMMU_FEAT_COHERENT_WALK;
1665                 dev_notice(smmu->dev, "\tcoherent table walk\n");
1666         }
1667
1668         if (id & ID0_SMS) {
1669                 u32 smr, sid, mask;
1670
1671                 smmu->features |= ARM_SMMU_FEAT_STREAM_MATCH;
1672                 smmu->num_mapping_groups = (id >> ID0_NUMSMRG_SHIFT) &
1673                                            ID0_NUMSMRG_MASK;
1674                 if (smmu->num_mapping_groups == 0) {
1675                         dev_err(smmu->dev,
1676                                 "stream-matching supported, but no SMRs present!\n");
1677                         return -ENODEV;
1678                 }
1679
1680                 smr = SMR_MASK_MASK << SMR_MASK_SHIFT;
1681                 smr |= (SMR_ID_MASK << SMR_ID_SHIFT);
1682                 writel_relaxed(smr, gr0_base + ARM_SMMU_GR0_SMR(0));
1683                 smr = readl_relaxed(gr0_base + ARM_SMMU_GR0_SMR(0));
1684
1685                 mask = (smr >> SMR_MASK_SHIFT) & SMR_MASK_MASK;
1686                 sid = (smr >> SMR_ID_SHIFT) & SMR_ID_MASK;
1687                 if ((mask & sid) != sid) {
1688                         dev_err(smmu->dev,
1689                                 "SMR mask bits (0x%x) insufficient for ID field (0x%x)\n",
1690                                 mask, sid);
1691                         return -ENODEV;
1692                 }
1693
1694                 dev_notice(smmu->dev,
1695                            "\tstream matching with %u register groups, mask 0x%x",
1696                            smmu->num_mapping_groups, mask);
1697         }
1698
1699         /* ID1 */
1700         id = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID1);
1701         smmu->pagesize = (id & ID1_PAGESIZE) ? SZ_64K : SZ_4K;
1702
1703         /* Check that we ioremapped enough */
1704         size = 1 << (((id >> ID1_NUMPAGENDXB_SHIFT) & ID1_NUMPAGENDXB_MASK) + 1);
1705         size *= (smmu->pagesize << 1);
1706         if (smmu->size < size)
1707                 dev_warn(smmu->dev,
1708                          "device is 0x%lx bytes but only mapped 0x%lx!\n",
1709                          size, smmu->size);
1710
1711         smmu->num_s2_context_banks = (id >> ID1_NUMS2CB_SHIFT) &
1712                                       ID1_NUMS2CB_MASK;
1713         smmu->num_context_banks = (id >> ID1_NUMCB_SHIFT) & ID1_NUMCB_MASK;
1714         if (smmu->num_s2_context_banks > smmu->num_context_banks) {
1715                 dev_err(smmu->dev, "impossible number of S2 context banks!\n");
1716                 return -ENODEV;
1717         }
1718         dev_notice(smmu->dev, "\t%u context banks (%u stage-2 only)\n",
1719                    smmu->num_context_banks, smmu->num_s2_context_banks);
1720
1721         /* ID2 */
1722         id = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID2);
1723         size = arm_smmu_id_size_to_bits((id >> ID2_IAS_SHIFT) & ID2_IAS_MASK);
1724
1725         /*
1726          * Stage-1 output limited by stage-2 input size due to pgd
1727          * allocation (PTRS_PER_PGD).
1728          */
1729 #ifdef CONFIG_64BIT
1730         /* Current maximum output size of 39 bits */
1731         smmu->s1_output_size = min(39UL, size);
1732 #else
1733         smmu->s1_output_size = min(32UL, size);
1734 #endif
1735
1736         /* The stage-2 output mask is also applied for bypass */
1737         size = arm_smmu_id_size_to_bits((id >> ID2_OAS_SHIFT) & ID2_OAS_MASK);
1738         smmu->s2_output_size = min((unsigned long)PHYS_MASK_SHIFT, size);
1739
1740         if (smmu->version == 1) {
1741                 smmu->input_size = 32;
1742         } else {
1743 #ifdef CONFIG_64BIT
1744                 size = (id >> ID2_UBS_SHIFT) & ID2_UBS_MASK;
1745                 size = min(39, arm_smmu_id_size_to_bits(size));
1746 #else
1747                 size = 32;
1748 #endif
1749                 smmu->input_size = size;
1750
1751                 if ((PAGE_SIZE == SZ_4K && !(id & ID2_PTFS_4K)) ||
1752                     (PAGE_SIZE == SZ_64K && !(id & ID2_PTFS_64K)) ||
1753                     (PAGE_SIZE != SZ_4K && PAGE_SIZE != SZ_64K)) {
1754                         dev_err(smmu->dev, "CPU page size 0x%lx unsupported\n",
1755                                 PAGE_SIZE);
1756                         return -ENODEV;
1757                 }
1758         }
1759
1760         dev_notice(smmu->dev,
1761                    "\t%lu-bit VA, %lu-bit IPA, %lu-bit PA\n",
1762                    smmu->input_size, smmu->s1_output_size, smmu->s2_output_size);
1763         return 0;
1764 }
1765
1766 static int arm_smmu_device_dt_probe(struct platform_device *pdev)
1767 {
1768         struct resource *res;
1769         struct arm_smmu_device *smmu;
1770         struct device_node *dev_node;
1771         struct device *dev = &pdev->dev;
1772         struct rb_node *node;
1773         struct of_phandle_args masterspec;
1774         int num_irqs, i, err;
1775
1776         smmu = devm_kzalloc(dev, sizeof(*smmu), GFP_KERNEL);
1777         if (!smmu) {
1778                 dev_err(dev, "failed to allocate arm_smmu_device\n");
1779                 return -ENOMEM;
1780         }
1781         smmu->dev = dev;
1782
1783         res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1784         if (!res) {
1785                 dev_err(dev, "missing base address/size\n");
1786                 return -ENODEV;
1787         }
1788
1789         smmu->size = resource_size(res);
1790         smmu->base = devm_request_and_ioremap(dev, res);
1791         if (!smmu->base)
1792                 return -EADDRNOTAVAIL;
1793
1794         if (of_property_read_u32(dev->of_node, "#global-interrupts",
1795                                  &smmu->num_global_irqs)) {
1796                 dev_err(dev, "missing #global-interrupts property\n");
1797                 return -ENODEV;
1798         }
1799
1800         num_irqs = 0;
1801         while ((res = platform_get_resource(pdev, IORESOURCE_IRQ, num_irqs))) {
1802                 num_irqs++;
1803                 if (num_irqs > smmu->num_global_irqs)
1804                         smmu->num_context_irqs++;
1805         }
1806
1807         if (num_irqs < smmu->num_global_irqs) {
1808                 dev_warn(dev, "found %d interrupts but expected at least %d\n",
1809                          num_irqs, smmu->num_global_irqs);
1810                 smmu->num_global_irqs = num_irqs;
1811         }
1812         smmu->num_context_irqs = num_irqs - smmu->num_global_irqs;
1813
1814         smmu->irqs = devm_kzalloc(dev, sizeof(*smmu->irqs) * num_irqs,
1815                                   GFP_KERNEL);
1816         if (!smmu->irqs) {
1817                 dev_err(dev, "failed to allocate %d irqs\n", num_irqs);
1818                 return -ENOMEM;
1819         }
1820
1821         for (i = 0; i < num_irqs; ++i) {
1822                 int irq = platform_get_irq(pdev, i);
1823                 if (irq < 0) {
1824                         dev_err(dev, "failed to get irq index %d\n", i);
1825                         return -ENODEV;
1826                 }
1827                 smmu->irqs[i] = irq;
1828         }
1829
1830         i = 0;
1831         smmu->masters = RB_ROOT;
1832         while (!of_parse_phandle_with_args(dev->of_node, "mmu-masters",
1833                                            "#stream-id-cells", i,
1834                                            &masterspec)) {
1835                 err = register_smmu_master(smmu, dev, &masterspec);
1836                 if (err) {
1837                         dev_err(dev, "failed to add master %s\n",
1838                                 masterspec.np->name);
1839                         goto out_put_masters;
1840                 }
1841
1842                 i++;
1843         }
1844         dev_notice(dev, "registered %d master devices\n", i);
1845
1846         if ((dev_node = of_parse_phandle(dev->of_node, "smmu-parent", 0)))
1847                 smmu->parent_of_node = dev_node;
1848
1849         err = arm_smmu_device_cfg_probe(smmu);
1850         if (err)
1851                 goto out_put_parent;
1852
1853         if (smmu->version > 1 &&
1854             smmu->num_context_banks != smmu->num_context_irqs) {
1855                 dev_err(dev,
1856                         "found only %d context interrupt(s) but %d required\n",
1857                         smmu->num_context_irqs, smmu->num_context_banks);
1858                 goto out_put_parent;
1859         }
1860
1861         for (i = 0; i < smmu->num_global_irqs; ++i) {
1862                 err = request_irq(smmu->irqs[i],
1863                                   arm_smmu_global_fault,
1864                                   IRQF_SHARED,
1865                                   "arm-smmu global fault",
1866                                   smmu);
1867                 if (err) {
1868                         dev_err(dev, "failed to request global IRQ %d (%u)\n",
1869                                 i, smmu->irqs[i]);
1870                         goto out_free_irqs;
1871                 }
1872         }
1873
1874         INIT_LIST_HEAD(&smmu->list);
1875         spin_lock(&arm_smmu_devices_lock);
1876         list_add(&smmu->list, &arm_smmu_devices);
1877         spin_unlock(&arm_smmu_devices_lock);
1878
1879         arm_smmu_device_reset(smmu);
1880         return 0;
1881
1882 out_free_irqs:
1883         while (i--)
1884                 free_irq(smmu->irqs[i], smmu);
1885
1886 out_put_parent:
1887         if (smmu->parent_of_node)
1888                 of_node_put(smmu->parent_of_node);
1889
1890 out_put_masters:
1891         for (node = rb_first(&smmu->masters); node; node = rb_next(node)) {
1892                 struct arm_smmu_master *master;
1893                 master = container_of(node, struct arm_smmu_master, node);
1894                 of_node_put(master->of_node);
1895         }
1896
1897         return err;
1898 }
1899
1900 static int arm_smmu_device_remove(struct platform_device *pdev)
1901 {
1902         int i;
1903         struct device *dev = &pdev->dev;
1904         struct arm_smmu_device *curr, *smmu = NULL;
1905         struct rb_node *node;
1906
1907         spin_lock(&arm_smmu_devices_lock);
1908         list_for_each_entry(curr, &arm_smmu_devices, list) {
1909                 if (curr->dev == dev) {
1910                         smmu = curr;
1911                         list_del(&smmu->list);
1912                         break;
1913                 }
1914         }
1915         spin_unlock(&arm_smmu_devices_lock);
1916
1917         if (!smmu)
1918                 return -ENODEV;
1919
1920         if (smmu->parent_of_node)
1921                 of_node_put(smmu->parent_of_node);
1922
1923         for (node = rb_first(&smmu->masters); node; node = rb_next(node)) {
1924                 struct arm_smmu_master *master;
1925                 master = container_of(node, struct arm_smmu_master, node);
1926                 of_node_put(master->of_node);
1927         }
1928
1929         if (!bitmap_empty(smmu->context_map, ARM_SMMU_MAX_CBS))
1930                 dev_err(dev, "removing device with active domains!\n");
1931
1932         for (i = 0; i < smmu->num_global_irqs; ++i)
1933                 free_irq(smmu->irqs[i], smmu);
1934
1935         /* Turn the thing off */
1936         writel(sCR0_CLIENTPD, ARM_SMMU_GR0(smmu) + ARM_SMMU_GR0_sCR0);
1937         return 0;
1938 }
1939
1940 #ifdef CONFIG_OF
1941 static struct of_device_id arm_smmu_of_match[] = {
1942         { .compatible = "arm,smmu-v1", },
1943         { .compatible = "arm,smmu-v2", },
1944         { .compatible = "arm,mmu-400", },
1945         { .compatible = "arm,mmu-500", },
1946         { },
1947 };
1948 MODULE_DEVICE_TABLE(of, arm_smmu_of_match);
1949 #endif
1950
1951 static struct platform_driver arm_smmu_driver = {
1952         .driver = {
1953                 .owner          = THIS_MODULE,
1954                 .name           = "arm-smmu",
1955                 .of_match_table = of_match_ptr(arm_smmu_of_match),
1956         },
1957         .probe  = arm_smmu_device_dt_probe,
1958         .remove = arm_smmu_device_remove,
1959 };
1960
1961 static int __init arm_smmu_init(void)
1962 {
1963         int ret;
1964
1965         ret = platform_driver_register(&arm_smmu_driver);
1966         if (ret)
1967                 return ret;
1968
1969         /* Oh, for a proper bus abstraction */
1970         if (!iommu_present(&platform_bus_type))
1971                 bus_set_iommu(&platform_bus_type, &arm_smmu_ops);
1972
1973         if (!iommu_present(&amba_bustype))
1974                 bus_set_iommu(&amba_bustype, &arm_smmu_ops);
1975
1976         return 0;
1977 }
1978
1979 static void __exit arm_smmu_exit(void)
1980 {
1981         return platform_driver_unregister(&arm_smmu_driver);
1982 }
1983
1984 module_init(arm_smmu_init);
1985 module_exit(arm_smmu_exit);
1986
1987 MODULE_DESCRIPTION("IOMMU API for ARM architected SMMU implementations");
1988 MODULE_AUTHOR("Will Deacon <will.deacon@arm.com>");
1989 MODULE_LICENSE("GPL v2");