]> git.karo-electronics.de Git - karo-tx-linux.git/blob - drivers/iommu/arm-smmu.c
Merge tag 'tegra-for-3.15-dt' of git://git.kernel.org/pub/scm/linux/kernel/git/tegra...
[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 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/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_XN                 (((pteval_t)3) << 53)
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 #define ARM_SMMU_PTE_PAGE               (((pteval_t)3) << 0)
71
72 #if PAGE_SIZE == SZ_4K
73 #define ARM_SMMU_PTE_CONT_ENTRIES       16
74 #elif PAGE_SIZE == SZ_64K
75 #define ARM_SMMU_PTE_CONT_ENTRIES       32
76 #else
77 #define ARM_SMMU_PTE_CONT_ENTRIES       1
78 #endif
79
80 #define ARM_SMMU_PTE_CONT_SIZE          (PAGE_SIZE * ARM_SMMU_PTE_CONT_ENTRIES)
81 #define ARM_SMMU_PTE_CONT_MASK          (~(ARM_SMMU_PTE_CONT_SIZE - 1))
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_BPSHCFG_SHIFT           8
194 #define CBAR_S1_BPSHCFG_MASK            3
195 #define CBAR_S1_BPSHCFG_NSH             3
196 #define CBAR_S1_MEMATTR_SHIFT           12
197 #define CBAR_S1_MEMATTR_MASK            0xf
198 #define CBAR_S1_MEMATTR_WB              0xf
199 #define CBAR_TYPE_SHIFT                 16
200 #define CBAR_TYPE_MASK                  0x3
201 #define CBAR_TYPE_S2_TRANS              (0 << CBAR_TYPE_SHIFT)
202 #define CBAR_TYPE_S1_TRANS_S2_BYPASS    (1 << CBAR_TYPE_SHIFT)
203 #define CBAR_TYPE_S1_TRANS_S2_FAULT     (2 << CBAR_TYPE_SHIFT)
204 #define CBAR_TYPE_S1_TRANS_S2_TRANS     (3 << CBAR_TYPE_SHIFT)
205 #define CBAR_IRPTNDX_SHIFT              24
206 #define CBAR_IRPTNDX_MASK               0xff
207
208 #define ARM_SMMU_GR1_CBA2R(n)           (0x800 + ((n) << 2))
209 #define CBA2R_RW64_32BIT                (0 << 0)
210 #define CBA2R_RW64_64BIT                (1 << 0)
211
212 /* Translation context bank */
213 #define ARM_SMMU_CB_BASE(smmu)          ((smmu)->base + ((smmu)->size >> 1))
214 #define ARM_SMMU_CB(smmu, n)            ((n) * (smmu)->pagesize)
215
216 #define ARM_SMMU_CB_SCTLR               0x0
217 #define ARM_SMMU_CB_RESUME              0x8
218 #define ARM_SMMU_CB_TTBCR2              0x10
219 #define ARM_SMMU_CB_TTBR0_LO            0x20
220 #define ARM_SMMU_CB_TTBR0_HI            0x24
221 #define ARM_SMMU_CB_TTBCR               0x30
222 #define ARM_SMMU_CB_S1_MAIR0            0x38
223 #define ARM_SMMU_CB_FSR                 0x58
224 #define ARM_SMMU_CB_FAR_LO              0x60
225 #define ARM_SMMU_CB_FAR_HI              0x64
226 #define ARM_SMMU_CB_FSYNR0              0x68
227 #define ARM_SMMU_CB_S1_TLBIASID         0x610
228
229 #define SCTLR_S1_ASIDPNE                (1 << 12)
230 #define SCTLR_CFCFG                     (1 << 7)
231 #define SCTLR_CFIE                      (1 << 6)
232 #define SCTLR_CFRE                      (1 << 5)
233 #define SCTLR_E                         (1 << 4)
234 #define SCTLR_AFE                       (1 << 2)
235 #define SCTLR_TRE                       (1 << 1)
236 #define SCTLR_M                         (1 << 0)
237 #define SCTLR_EAE_SBOP                  (SCTLR_AFE | SCTLR_TRE)
238
239 #define RESUME_RETRY                    (0 << 0)
240 #define RESUME_TERMINATE                (1 << 0)
241
242 #define TTBCR_EAE                       (1 << 31)
243
244 #define TTBCR_PASIZE_SHIFT              16
245 #define TTBCR_PASIZE_MASK               0x7
246
247 #define TTBCR_TG0_4K                    (0 << 14)
248 #define TTBCR_TG0_64K                   (1 << 14)
249
250 #define TTBCR_SH0_SHIFT                 12
251 #define TTBCR_SH0_MASK                  0x3
252 #define TTBCR_SH_NS                     0
253 #define TTBCR_SH_OS                     2
254 #define TTBCR_SH_IS                     3
255
256 #define TTBCR_ORGN0_SHIFT               10
257 #define TTBCR_IRGN0_SHIFT               8
258 #define TTBCR_RGN_MASK                  0x3
259 #define TTBCR_RGN_NC                    0
260 #define TTBCR_RGN_WBWA                  1
261 #define TTBCR_RGN_WT                    2
262 #define TTBCR_RGN_WB                    3
263
264 #define TTBCR_SL0_SHIFT                 6
265 #define TTBCR_SL0_MASK                  0x3
266 #define TTBCR_SL0_LVL_2                 0
267 #define TTBCR_SL0_LVL_1                 1
268
269 #define TTBCR_T1SZ_SHIFT                16
270 #define TTBCR_T0SZ_SHIFT                0
271 #define TTBCR_SZ_MASK                   0xf
272
273 #define TTBCR2_SEP_SHIFT                15
274 #define TTBCR2_SEP_MASK                 0x7
275
276 #define TTBCR2_PASIZE_SHIFT             0
277 #define TTBCR2_PASIZE_MASK              0x7
278
279 /* Common definitions for PASize and SEP fields */
280 #define TTBCR2_ADDR_32                  0
281 #define TTBCR2_ADDR_36                  1
282 #define TTBCR2_ADDR_40                  2
283 #define TTBCR2_ADDR_42                  3
284 #define TTBCR2_ADDR_44                  4
285 #define TTBCR2_ADDR_48                  5
286
287 #define TTBRn_HI_ASID_SHIFT             16
288
289 #define MAIR_ATTR_SHIFT(n)              ((n) << 3)
290 #define MAIR_ATTR_MASK                  0xff
291 #define MAIR_ATTR_DEVICE                0x04
292 #define MAIR_ATTR_NC                    0x44
293 #define MAIR_ATTR_WBRWA                 0xff
294 #define MAIR_ATTR_IDX_NC                0
295 #define MAIR_ATTR_IDX_CACHE             1
296 #define MAIR_ATTR_IDX_DEV               2
297
298 #define FSR_MULTI                       (1 << 31)
299 #define FSR_SS                          (1 << 30)
300 #define FSR_UUT                         (1 << 8)
301 #define FSR_ASF                         (1 << 7)
302 #define FSR_TLBLKF                      (1 << 6)
303 #define FSR_TLBMCF                      (1 << 5)
304 #define FSR_EF                          (1 << 4)
305 #define FSR_PF                          (1 << 3)
306 #define FSR_AFF                         (1 << 2)
307 #define FSR_TF                          (1 << 1)
308
309 #define FSR_IGN                         (FSR_AFF | FSR_ASF | FSR_TLBMCF |       \
310                                          FSR_TLBLKF)
311 #define FSR_FAULT                       (FSR_MULTI | FSR_SS | FSR_UUT |         \
312                                          FSR_EF | FSR_PF | FSR_TF | FSR_IGN)
313
314 #define FSYNR0_WNR                      (1 << 4)
315
316 struct arm_smmu_smr {
317         u8                              idx;
318         u16                             mask;
319         u16                             id;
320 };
321
322 struct arm_smmu_master {
323         struct device_node              *of_node;
324
325         /*
326          * The following is specific to the master's position in the
327          * SMMU chain.
328          */
329         struct rb_node                  node;
330         int                             num_streamids;
331         u16                             streamids[MAX_MASTER_STREAMIDS];
332
333         /*
334          * We only need to allocate these on the root SMMU, as we
335          * configure unmatched streams to bypass translation.
336          */
337         struct arm_smmu_smr             *smrs;
338 };
339
340 struct arm_smmu_device {
341         struct device                   *dev;
342         struct device_node              *parent_of_node;
343
344         void __iomem                    *base;
345         unsigned long                   size;
346         unsigned long                   pagesize;
347
348 #define ARM_SMMU_FEAT_COHERENT_WALK     (1 << 0)
349 #define ARM_SMMU_FEAT_STREAM_MATCH      (1 << 1)
350 #define ARM_SMMU_FEAT_TRANS_S1          (1 << 2)
351 #define ARM_SMMU_FEAT_TRANS_S2          (1 << 3)
352 #define ARM_SMMU_FEAT_TRANS_NESTED      (1 << 4)
353         u32                             features;
354         int                             version;
355
356         u32                             num_context_banks;
357         u32                             num_s2_context_banks;
358         DECLARE_BITMAP(context_map, ARM_SMMU_MAX_CBS);
359         atomic_t                        irptndx;
360
361         u32                             num_mapping_groups;
362         DECLARE_BITMAP(smr_map, ARM_SMMU_MAX_SMRS);
363
364         unsigned long                   input_size;
365         unsigned long                   s1_output_size;
366         unsigned long                   s2_output_size;
367
368         u32                             num_global_irqs;
369         u32                             num_context_irqs;
370         unsigned int                    *irqs;
371
372         struct list_head                list;
373         struct rb_root                  masters;
374 };
375
376 struct arm_smmu_cfg {
377         struct arm_smmu_device          *smmu;
378         u8                              cbndx;
379         u8                              irptndx;
380         u32                             cbar;
381         pgd_t                           *pgd;
382 };
383 #define INVALID_IRPTNDX                 0xff
384
385 #define ARM_SMMU_CB_ASID(cfg)           ((cfg)->cbndx)
386 #define ARM_SMMU_CB_VMID(cfg)           ((cfg)->cbndx + 1)
387
388 struct arm_smmu_domain {
389         /*
390          * A domain can span across multiple, chained SMMUs and requires
391          * all devices within the domain to follow the same translation
392          * path.
393          */
394         struct arm_smmu_device          *leaf_smmu;
395         struct arm_smmu_cfg             root_cfg;
396         phys_addr_t                     output_mask;
397
398         spinlock_t                      lock;
399 };
400
401 static DEFINE_SPINLOCK(arm_smmu_devices_lock);
402 static LIST_HEAD(arm_smmu_devices);
403
404 static struct arm_smmu_master *find_smmu_master(struct arm_smmu_device *smmu,
405                                                 struct device_node *dev_node)
406 {
407         struct rb_node *node = smmu->masters.rb_node;
408
409         while (node) {
410                 struct arm_smmu_master *master;
411                 master = container_of(node, struct arm_smmu_master, node);
412
413                 if (dev_node < master->of_node)
414                         node = node->rb_left;
415                 else if (dev_node > master->of_node)
416                         node = node->rb_right;
417                 else
418                         return master;
419         }
420
421         return NULL;
422 }
423
424 static int insert_smmu_master(struct arm_smmu_device *smmu,
425                               struct arm_smmu_master *master)
426 {
427         struct rb_node **new, *parent;
428
429         new = &smmu->masters.rb_node;
430         parent = NULL;
431         while (*new) {
432                 struct arm_smmu_master *this;
433                 this = container_of(*new, struct arm_smmu_master, node);
434
435                 parent = *new;
436                 if (master->of_node < this->of_node)
437                         new = &((*new)->rb_left);
438                 else if (master->of_node > this->of_node)
439                         new = &((*new)->rb_right);
440                 else
441                         return -EEXIST;
442         }
443
444         rb_link_node(&master->node, parent, new);
445         rb_insert_color(&master->node, &smmu->masters);
446         return 0;
447 }
448
449 static int register_smmu_master(struct arm_smmu_device *smmu,
450                                 struct device *dev,
451                                 struct of_phandle_args *masterspec)
452 {
453         int i;
454         struct arm_smmu_master *master;
455
456         master = find_smmu_master(smmu, masterspec->np);
457         if (master) {
458                 dev_err(dev,
459                         "rejecting multiple registrations for master device %s\n",
460                         masterspec->np->name);
461                 return -EBUSY;
462         }
463
464         if (masterspec->args_count > MAX_MASTER_STREAMIDS) {
465                 dev_err(dev,
466                         "reached maximum number (%d) of stream IDs for master device %s\n",
467                         MAX_MASTER_STREAMIDS, masterspec->np->name);
468                 return -ENOSPC;
469         }
470
471         master = devm_kzalloc(dev, sizeof(*master), GFP_KERNEL);
472         if (!master)
473                 return -ENOMEM;
474
475         master->of_node         = masterspec->np;
476         master->num_streamids   = masterspec->args_count;
477
478         for (i = 0; i < master->num_streamids; ++i)
479                 master->streamids[i] = masterspec->args[i];
480
481         return insert_smmu_master(smmu, master);
482 }
483
484 static struct arm_smmu_device *find_parent_smmu(struct arm_smmu_device *smmu)
485 {
486         struct arm_smmu_device *parent;
487
488         if (!smmu->parent_of_node)
489                 return NULL;
490
491         spin_lock(&arm_smmu_devices_lock);
492         list_for_each_entry(parent, &arm_smmu_devices, list)
493                 if (parent->dev->of_node == smmu->parent_of_node)
494                         goto out_unlock;
495
496         parent = NULL;
497         dev_warn(smmu->dev,
498                  "Failed to find SMMU parent despite parent in DT\n");
499 out_unlock:
500         spin_unlock(&arm_smmu_devices_lock);
501         return parent;
502 }
503
504 static int __arm_smmu_alloc_bitmap(unsigned long *map, int start, int end)
505 {
506         int idx;
507
508         do {
509                 idx = find_next_zero_bit(map, end, start);
510                 if (idx == end)
511                         return -ENOSPC;
512         } while (test_and_set_bit(idx, map));
513
514         return idx;
515 }
516
517 static void __arm_smmu_free_bitmap(unsigned long *map, int idx)
518 {
519         clear_bit(idx, map);
520 }
521
522 /* Wait for any pending TLB invalidations to complete */
523 static void arm_smmu_tlb_sync(struct arm_smmu_device *smmu)
524 {
525         int count = 0;
526         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
527
528         writel_relaxed(0, gr0_base + ARM_SMMU_GR0_sTLBGSYNC);
529         while (readl_relaxed(gr0_base + ARM_SMMU_GR0_sTLBGSTATUS)
530                & sTLBGSTATUS_GSACTIVE) {
531                 cpu_relax();
532                 if (++count == TLB_LOOP_TIMEOUT) {
533                         dev_err_ratelimited(smmu->dev,
534                         "TLB sync timed out -- SMMU may be deadlocked\n");
535                         return;
536                 }
537                 udelay(1);
538         }
539 }
540
541 static void arm_smmu_tlb_inv_context(struct arm_smmu_cfg *cfg)
542 {
543         struct arm_smmu_device *smmu = cfg->smmu;
544         void __iomem *base = ARM_SMMU_GR0(smmu);
545         bool stage1 = cfg->cbar != CBAR_TYPE_S2_TRANS;
546
547         if (stage1) {
548                 base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
549                 writel_relaxed(ARM_SMMU_CB_ASID(cfg),
550                                base + ARM_SMMU_CB_S1_TLBIASID);
551         } else {
552                 base = ARM_SMMU_GR0(smmu);
553                 writel_relaxed(ARM_SMMU_CB_VMID(cfg),
554                                base + ARM_SMMU_GR0_TLBIVMID);
555         }
556
557         arm_smmu_tlb_sync(smmu);
558 }
559
560 static irqreturn_t arm_smmu_context_fault(int irq, void *dev)
561 {
562         int flags, ret;
563         u32 fsr, far, fsynr, resume;
564         unsigned long iova;
565         struct iommu_domain *domain = dev;
566         struct arm_smmu_domain *smmu_domain = domain->priv;
567         struct arm_smmu_cfg *root_cfg = &smmu_domain->root_cfg;
568         struct arm_smmu_device *smmu = root_cfg->smmu;
569         void __iomem *cb_base;
570
571         cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, root_cfg->cbndx);
572         fsr = readl_relaxed(cb_base + ARM_SMMU_CB_FSR);
573
574         if (!(fsr & FSR_FAULT))
575                 return IRQ_NONE;
576
577         if (fsr & FSR_IGN)
578                 dev_err_ratelimited(smmu->dev,
579                                     "Unexpected context fault (fsr 0x%u)\n",
580                                     fsr);
581
582         fsynr = readl_relaxed(cb_base + ARM_SMMU_CB_FSYNR0);
583         flags = fsynr & FSYNR0_WNR ? IOMMU_FAULT_WRITE : IOMMU_FAULT_READ;
584
585         far = readl_relaxed(cb_base + ARM_SMMU_CB_FAR_LO);
586         iova = far;
587 #ifdef CONFIG_64BIT
588         far = readl_relaxed(cb_base + ARM_SMMU_CB_FAR_HI);
589         iova |= ((unsigned long)far << 32);
590 #endif
591
592         if (!report_iommu_fault(domain, smmu->dev, iova, flags)) {
593                 ret = IRQ_HANDLED;
594                 resume = RESUME_RETRY;
595         } else {
596                 dev_err_ratelimited(smmu->dev,
597                     "Unhandled context fault: iova=0x%08lx, fsynr=0x%x, cb=%d\n",
598                     iova, fsynr, root_cfg->cbndx);
599                 ret = IRQ_NONE;
600                 resume = RESUME_TERMINATE;
601         }
602
603         /* Clear the faulting FSR */
604         writel(fsr, cb_base + ARM_SMMU_CB_FSR);
605
606         /* Retry or terminate any stalled transactions */
607         if (fsr & FSR_SS)
608                 writel_relaxed(resume, cb_base + ARM_SMMU_CB_RESUME);
609
610         return ret;
611 }
612
613 static irqreturn_t arm_smmu_global_fault(int irq, void *dev)
614 {
615         u32 gfsr, gfsynr0, gfsynr1, gfsynr2;
616         struct arm_smmu_device *smmu = dev;
617         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
618
619         gfsr = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSR);
620         if (!gfsr)
621                 return IRQ_NONE;
622
623         gfsynr0 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSYNR0);
624         gfsynr1 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSYNR1);
625         gfsynr2 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSYNR2);
626
627         dev_err_ratelimited(smmu->dev,
628                 "Unexpected global fault, this could be serious\n");
629         dev_err_ratelimited(smmu->dev,
630                 "\tGFSR 0x%08x, GFSYNR0 0x%08x, GFSYNR1 0x%08x, GFSYNR2 0x%08x\n",
631                 gfsr, gfsynr0, gfsynr1, gfsynr2);
632
633         writel(gfsr, gr0_base + ARM_SMMU_GR0_sGFSR);
634         return IRQ_HANDLED;
635 }
636
637 static void arm_smmu_flush_pgtable(struct arm_smmu_device *smmu, void *addr,
638                                    size_t size)
639 {
640         unsigned long offset = (unsigned long)addr & ~PAGE_MASK;
641
642
643         /* Ensure new page tables are visible to the hardware walker */
644         if (smmu->features & ARM_SMMU_FEAT_COHERENT_WALK) {
645                 dsb();
646         } else {
647                 /*
648                  * If the SMMU can't walk tables in the CPU caches, treat them
649                  * like non-coherent DMA since we need to flush the new entries
650                  * all the way out to memory. There's no possibility of
651                  * recursion here as the SMMU table walker will not be wired
652                  * through another SMMU.
653                  */
654                 dma_map_page(smmu->dev, virt_to_page(addr), offset, size,
655                                 DMA_TO_DEVICE);
656         }
657 }
658
659 static void arm_smmu_init_context_bank(struct arm_smmu_domain *smmu_domain)
660 {
661         u32 reg;
662         bool stage1;
663         struct arm_smmu_cfg *root_cfg = &smmu_domain->root_cfg;
664         struct arm_smmu_device *smmu = root_cfg->smmu;
665         void __iomem *cb_base, *gr0_base, *gr1_base;
666
667         gr0_base = ARM_SMMU_GR0(smmu);
668         gr1_base = ARM_SMMU_GR1(smmu);
669         stage1 = root_cfg->cbar != CBAR_TYPE_S2_TRANS;
670         cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, root_cfg->cbndx);
671
672         /* CBAR */
673         reg = root_cfg->cbar;
674         if (smmu->version == 1)
675               reg |= root_cfg->irptndx << CBAR_IRPTNDX_SHIFT;
676
677         /*
678          * Use the weakest shareability/memory types, so they are
679          * overridden by the ttbcr/pte.
680          */
681         if (stage1) {
682                 reg |= (CBAR_S1_BPSHCFG_NSH << CBAR_S1_BPSHCFG_SHIFT) |
683                         (CBAR_S1_MEMATTR_WB << CBAR_S1_MEMATTR_SHIFT);
684         } else {
685                 reg |= ARM_SMMU_CB_VMID(root_cfg) << CBAR_VMID_SHIFT;
686         }
687         writel_relaxed(reg, gr1_base + ARM_SMMU_GR1_CBAR(root_cfg->cbndx));
688
689         if (smmu->version > 1) {
690                 /* CBA2R */
691 #ifdef CONFIG_64BIT
692                 reg = CBA2R_RW64_64BIT;
693 #else
694                 reg = CBA2R_RW64_32BIT;
695 #endif
696                 writel_relaxed(reg,
697                                gr1_base + ARM_SMMU_GR1_CBA2R(root_cfg->cbndx));
698
699                 /* TTBCR2 */
700                 switch (smmu->input_size) {
701                 case 32:
702                         reg = (TTBCR2_ADDR_32 << TTBCR2_SEP_SHIFT);
703                         break;
704                 case 36:
705                         reg = (TTBCR2_ADDR_36 << TTBCR2_SEP_SHIFT);
706                         break;
707                 case 39:
708                         reg = (TTBCR2_ADDR_40 << TTBCR2_SEP_SHIFT);
709                         break;
710                 case 42:
711                         reg = (TTBCR2_ADDR_42 << TTBCR2_SEP_SHIFT);
712                         break;
713                 case 44:
714                         reg = (TTBCR2_ADDR_44 << TTBCR2_SEP_SHIFT);
715                         break;
716                 case 48:
717                         reg = (TTBCR2_ADDR_48 << TTBCR2_SEP_SHIFT);
718                         break;
719                 }
720
721                 switch (smmu->s1_output_size) {
722                 case 32:
723                         reg |= (TTBCR2_ADDR_32 << TTBCR2_PASIZE_SHIFT);
724                         break;
725                 case 36:
726                         reg |= (TTBCR2_ADDR_36 << TTBCR2_PASIZE_SHIFT);
727                         break;
728                 case 39:
729                         reg |= (TTBCR2_ADDR_40 << TTBCR2_PASIZE_SHIFT);
730                         break;
731                 case 42:
732                         reg |= (TTBCR2_ADDR_42 << TTBCR2_PASIZE_SHIFT);
733                         break;
734                 case 44:
735                         reg |= (TTBCR2_ADDR_44 << TTBCR2_PASIZE_SHIFT);
736                         break;
737                 case 48:
738                         reg |= (TTBCR2_ADDR_48 << TTBCR2_PASIZE_SHIFT);
739                         break;
740                 }
741
742                 if (stage1)
743                         writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBCR2);
744         }
745
746         /* TTBR0 */
747         arm_smmu_flush_pgtable(smmu, root_cfg->pgd,
748                                PTRS_PER_PGD * sizeof(pgd_t));
749         reg = __pa(root_cfg->pgd);
750         writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBR0_LO);
751         reg = (phys_addr_t)__pa(root_cfg->pgd) >> 32;
752         if (stage1)
753                 reg |= ARM_SMMU_CB_ASID(root_cfg) << TTBRn_HI_ASID_SHIFT;
754         writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBR0_HI);
755
756         /*
757          * TTBCR
758          * We use long descriptor, with inner-shareable WBWA tables in TTBR0.
759          */
760         if (smmu->version > 1) {
761                 if (PAGE_SIZE == SZ_4K)
762                         reg = TTBCR_TG0_4K;
763                 else
764                         reg = TTBCR_TG0_64K;
765
766                 if (!stage1) {
767                         switch (smmu->s2_output_size) {
768                         case 32:
769                                 reg |= (TTBCR2_ADDR_32 << TTBCR_PASIZE_SHIFT);
770                                 break;
771                         case 36:
772                                 reg |= (TTBCR2_ADDR_36 << TTBCR_PASIZE_SHIFT);
773                                 break;
774                         case 40:
775                                 reg |= (TTBCR2_ADDR_40 << TTBCR_PASIZE_SHIFT);
776                                 break;
777                         case 42:
778                                 reg |= (TTBCR2_ADDR_42 << TTBCR_PASIZE_SHIFT);
779                                 break;
780                         case 44:
781                                 reg |= (TTBCR2_ADDR_44 << TTBCR_PASIZE_SHIFT);
782                                 break;
783                         case 48:
784                                 reg |= (TTBCR2_ADDR_48 << TTBCR_PASIZE_SHIFT);
785                                 break;
786                         }
787                 } else {
788                         reg |= (64 - smmu->s1_output_size) << TTBCR_T0SZ_SHIFT;
789                 }
790         } else {
791                 reg = 0;
792         }
793
794         reg |= TTBCR_EAE |
795               (TTBCR_SH_IS << TTBCR_SH0_SHIFT) |
796               (TTBCR_RGN_WBWA << TTBCR_ORGN0_SHIFT) |
797               (TTBCR_RGN_WBWA << TTBCR_IRGN0_SHIFT) |
798               (TTBCR_SL0_LVL_1 << TTBCR_SL0_SHIFT);
799         writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBCR);
800
801         /* MAIR0 (stage-1 only) */
802         if (stage1) {
803                 reg = (MAIR_ATTR_NC << MAIR_ATTR_SHIFT(MAIR_ATTR_IDX_NC)) |
804                       (MAIR_ATTR_WBRWA << MAIR_ATTR_SHIFT(MAIR_ATTR_IDX_CACHE)) |
805                       (MAIR_ATTR_DEVICE << MAIR_ATTR_SHIFT(MAIR_ATTR_IDX_DEV));
806                 writel_relaxed(reg, cb_base + ARM_SMMU_CB_S1_MAIR0);
807         }
808
809         /* SCTLR */
810         reg = SCTLR_CFCFG | SCTLR_CFIE | SCTLR_CFRE | SCTLR_M | SCTLR_EAE_SBOP;
811         if (stage1)
812                 reg |= SCTLR_S1_ASIDPNE;
813 #ifdef __BIG_ENDIAN
814         reg |= SCTLR_E;
815 #endif
816         writel_relaxed(reg, cb_base + ARM_SMMU_CB_SCTLR);
817 }
818
819 static int arm_smmu_init_domain_context(struct iommu_domain *domain,
820                                         struct device *dev)
821 {
822         int irq, ret, start;
823         struct arm_smmu_domain *smmu_domain = domain->priv;
824         struct arm_smmu_cfg *root_cfg = &smmu_domain->root_cfg;
825         struct arm_smmu_device *smmu, *parent;
826
827         /*
828          * Walk the SMMU chain to find the root device for this chain.
829          * We assume that no masters have translations which terminate
830          * early, and therefore check that the root SMMU does indeed have
831          * a StreamID for the master in question.
832          */
833         parent = dev->archdata.iommu;
834         smmu_domain->output_mask = -1;
835         do {
836                 smmu = parent;
837                 smmu_domain->output_mask &= (1ULL << smmu->s2_output_size) - 1;
838         } while ((parent = find_parent_smmu(smmu)));
839
840         if (!find_smmu_master(smmu, dev->of_node)) {
841                 dev_err(dev, "unable to find root SMMU for device\n");
842                 return -ENODEV;
843         }
844
845         if (smmu->features & ARM_SMMU_FEAT_TRANS_NESTED) {
846                 /*
847                  * We will likely want to change this if/when KVM gets
848                  * involved.
849                  */
850                 root_cfg->cbar = CBAR_TYPE_S1_TRANS_S2_BYPASS;
851                 start = smmu->num_s2_context_banks;
852         } else if (smmu->features & ARM_SMMU_FEAT_TRANS_S2) {
853                 root_cfg->cbar = CBAR_TYPE_S2_TRANS;
854                 start = 0;
855         } else {
856                 root_cfg->cbar = CBAR_TYPE_S1_TRANS_S2_BYPASS;
857                 start = smmu->num_s2_context_banks;
858         }
859
860         ret = __arm_smmu_alloc_bitmap(smmu->context_map, start,
861                                       smmu->num_context_banks);
862         if (IS_ERR_VALUE(ret))
863                 return ret;
864
865         root_cfg->cbndx = ret;
866         if (smmu->version == 1) {
867                 root_cfg->irptndx = atomic_inc_return(&smmu->irptndx);
868                 root_cfg->irptndx %= smmu->num_context_irqs;
869         } else {
870                 root_cfg->irptndx = root_cfg->cbndx;
871         }
872
873         irq = smmu->irqs[smmu->num_global_irqs + root_cfg->irptndx];
874         ret = request_irq(irq, arm_smmu_context_fault, IRQF_SHARED,
875                           "arm-smmu-context-fault", domain);
876         if (IS_ERR_VALUE(ret)) {
877                 dev_err(smmu->dev, "failed to request context IRQ %d (%u)\n",
878                         root_cfg->irptndx, irq);
879                 root_cfg->irptndx = INVALID_IRPTNDX;
880                 goto out_free_context;
881         }
882
883         root_cfg->smmu = smmu;
884         arm_smmu_init_context_bank(smmu_domain);
885         return ret;
886
887 out_free_context:
888         __arm_smmu_free_bitmap(smmu->context_map, root_cfg->cbndx);
889         return ret;
890 }
891
892 static void arm_smmu_destroy_domain_context(struct iommu_domain *domain)
893 {
894         struct arm_smmu_domain *smmu_domain = domain->priv;
895         struct arm_smmu_cfg *root_cfg = &smmu_domain->root_cfg;
896         struct arm_smmu_device *smmu = root_cfg->smmu;
897         void __iomem *cb_base;
898         int irq;
899
900         if (!smmu)
901                 return;
902
903         /* Disable the context bank and nuke the TLB before freeing it. */
904         cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, root_cfg->cbndx);
905         writel_relaxed(0, cb_base + ARM_SMMU_CB_SCTLR);
906         arm_smmu_tlb_inv_context(root_cfg);
907
908         if (root_cfg->irptndx != INVALID_IRPTNDX) {
909                 irq = smmu->irqs[smmu->num_global_irqs + root_cfg->irptndx];
910                 free_irq(irq, domain);
911         }
912
913         __arm_smmu_free_bitmap(smmu->context_map, root_cfg->cbndx);
914 }
915
916 static int arm_smmu_domain_init(struct iommu_domain *domain)
917 {
918         struct arm_smmu_domain *smmu_domain;
919         pgd_t *pgd;
920
921         /*
922          * Allocate the domain and initialise some of its data structures.
923          * We can't really do anything meaningful until we've added a
924          * master.
925          */
926         smmu_domain = kzalloc(sizeof(*smmu_domain), GFP_KERNEL);
927         if (!smmu_domain)
928                 return -ENOMEM;
929
930         pgd = kzalloc(PTRS_PER_PGD * sizeof(pgd_t), GFP_KERNEL);
931         if (!pgd)
932                 goto out_free_domain;
933         smmu_domain->root_cfg.pgd = pgd;
934
935         spin_lock_init(&smmu_domain->lock);
936         domain->priv = smmu_domain;
937         return 0;
938
939 out_free_domain:
940         kfree(smmu_domain);
941         return -ENOMEM;
942 }
943
944 static void arm_smmu_free_ptes(pmd_t *pmd)
945 {
946         pgtable_t table = pmd_pgtable(*pmd);
947         pgtable_page_dtor(table);
948         __free_page(table);
949 }
950
951 static void arm_smmu_free_pmds(pud_t *pud)
952 {
953         int i;
954         pmd_t *pmd, *pmd_base = pmd_offset(pud, 0);
955
956         pmd = pmd_base;
957         for (i = 0; i < PTRS_PER_PMD; ++i) {
958                 if (pmd_none(*pmd))
959                         continue;
960
961                 arm_smmu_free_ptes(pmd);
962                 pmd++;
963         }
964
965         pmd_free(NULL, pmd_base);
966 }
967
968 static void arm_smmu_free_puds(pgd_t *pgd)
969 {
970         int i;
971         pud_t *pud, *pud_base = pud_offset(pgd, 0);
972
973         pud = pud_base;
974         for (i = 0; i < PTRS_PER_PUD; ++i) {
975                 if (pud_none(*pud))
976                         continue;
977
978                 arm_smmu_free_pmds(pud);
979                 pud++;
980         }
981
982         pud_free(NULL, pud_base);
983 }
984
985 static void arm_smmu_free_pgtables(struct arm_smmu_domain *smmu_domain)
986 {
987         int i;
988         struct arm_smmu_cfg *root_cfg = &smmu_domain->root_cfg;
989         pgd_t *pgd, *pgd_base = root_cfg->pgd;
990
991         /*
992          * Recursively free the page tables for this domain. We don't
993          * care about speculative TLB filling, because the TLB will be
994          * nuked next time this context bank is re-allocated and no devices
995          * currently map to these tables.
996          */
997         pgd = pgd_base;
998         for (i = 0; i < PTRS_PER_PGD; ++i) {
999                 if (pgd_none(*pgd))
1000                         continue;
1001                 arm_smmu_free_puds(pgd);
1002                 pgd++;
1003         }
1004
1005         kfree(pgd_base);
1006 }
1007
1008 static void arm_smmu_domain_destroy(struct iommu_domain *domain)
1009 {
1010         struct arm_smmu_domain *smmu_domain = domain->priv;
1011
1012         /*
1013          * Free the domain resources. We assume that all devices have
1014          * already been detached.
1015          */
1016         arm_smmu_destroy_domain_context(domain);
1017         arm_smmu_free_pgtables(smmu_domain);
1018         kfree(smmu_domain);
1019 }
1020
1021 static int arm_smmu_master_configure_smrs(struct arm_smmu_device *smmu,
1022                                           struct arm_smmu_master *master)
1023 {
1024         int i;
1025         struct arm_smmu_smr *smrs;
1026         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1027
1028         if (!(smmu->features & ARM_SMMU_FEAT_STREAM_MATCH))
1029                 return 0;
1030
1031         if (master->smrs)
1032                 return -EEXIST;
1033
1034         smrs = kmalloc(sizeof(*smrs) * master->num_streamids, GFP_KERNEL);
1035         if (!smrs) {
1036                 dev_err(smmu->dev, "failed to allocate %d SMRs for master %s\n",
1037                         master->num_streamids, master->of_node->name);
1038                 return -ENOMEM;
1039         }
1040
1041         /* Allocate the SMRs on the root SMMU */
1042         for (i = 0; i < master->num_streamids; ++i) {
1043                 int idx = __arm_smmu_alloc_bitmap(smmu->smr_map, 0,
1044                                                   smmu->num_mapping_groups);
1045                 if (IS_ERR_VALUE(idx)) {
1046                         dev_err(smmu->dev, "failed to allocate free SMR\n");
1047                         goto err_free_smrs;
1048                 }
1049
1050                 smrs[i] = (struct arm_smmu_smr) {
1051                         .idx    = idx,
1052                         .mask   = 0, /* We don't currently share SMRs */
1053                         .id     = master->streamids[i],
1054                 };
1055         }
1056
1057         /* It worked! Now, poke the actual hardware */
1058         for (i = 0; i < master->num_streamids; ++i) {
1059                 u32 reg = SMR_VALID | smrs[i].id << SMR_ID_SHIFT |
1060                           smrs[i].mask << SMR_MASK_SHIFT;
1061                 writel_relaxed(reg, gr0_base + ARM_SMMU_GR0_SMR(smrs[i].idx));
1062         }
1063
1064         master->smrs = smrs;
1065         return 0;
1066
1067 err_free_smrs:
1068         while (--i >= 0)
1069                 __arm_smmu_free_bitmap(smmu->smr_map, smrs[i].idx);
1070         kfree(smrs);
1071         return -ENOSPC;
1072 }
1073
1074 static void arm_smmu_master_free_smrs(struct arm_smmu_device *smmu,
1075                                       struct arm_smmu_master *master)
1076 {
1077         int i;
1078         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1079         struct arm_smmu_smr *smrs = master->smrs;
1080
1081         /* Invalidate the SMRs before freeing back to the allocator */
1082         for (i = 0; i < master->num_streamids; ++i) {
1083                 u8 idx = smrs[i].idx;
1084                 writel_relaxed(~SMR_VALID, gr0_base + ARM_SMMU_GR0_SMR(idx));
1085                 __arm_smmu_free_bitmap(smmu->smr_map, idx);
1086         }
1087
1088         master->smrs = NULL;
1089         kfree(smrs);
1090 }
1091
1092 static void arm_smmu_bypass_stream_mapping(struct arm_smmu_device *smmu,
1093                                            struct arm_smmu_master *master)
1094 {
1095         int i;
1096         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1097
1098         for (i = 0; i < master->num_streamids; ++i) {
1099                 u16 sid = master->streamids[i];
1100                 writel_relaxed(S2CR_TYPE_BYPASS,
1101                                gr0_base + ARM_SMMU_GR0_S2CR(sid));
1102         }
1103 }
1104
1105 static int arm_smmu_domain_add_master(struct arm_smmu_domain *smmu_domain,
1106                                       struct arm_smmu_master *master)
1107 {
1108         int i, ret;
1109         struct arm_smmu_device *parent, *smmu = smmu_domain->root_cfg.smmu;
1110         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1111
1112         ret = arm_smmu_master_configure_smrs(smmu, master);
1113         if (ret)
1114                 return ret;
1115
1116         /* Bypass the leaves */
1117         smmu = smmu_domain->leaf_smmu;
1118         while ((parent = find_parent_smmu(smmu))) {
1119                 /*
1120                  * We won't have a StreamID match for anything but the root
1121                  * smmu, so we only need to worry about StreamID indexing,
1122                  * where we must install bypass entries in the S2CRs.
1123                  */
1124                 if (smmu->features & ARM_SMMU_FEAT_STREAM_MATCH)
1125                         continue;
1126
1127                 arm_smmu_bypass_stream_mapping(smmu, master);
1128                 smmu = parent;
1129         }
1130
1131         /* Now we're at the root, time to point at our context bank */
1132         for (i = 0; i < master->num_streamids; ++i) {
1133                 u32 idx, s2cr;
1134                 idx = master->smrs ? master->smrs[i].idx : master->streamids[i];
1135                 s2cr = (S2CR_TYPE_TRANS << S2CR_TYPE_SHIFT) |
1136                        (smmu_domain->root_cfg.cbndx << S2CR_CBNDX_SHIFT);
1137                 writel_relaxed(s2cr, gr0_base + ARM_SMMU_GR0_S2CR(idx));
1138         }
1139
1140         return 0;
1141 }
1142
1143 static void arm_smmu_domain_remove_master(struct arm_smmu_domain *smmu_domain,
1144                                           struct arm_smmu_master *master)
1145 {
1146         struct arm_smmu_device *smmu = smmu_domain->root_cfg.smmu;
1147
1148         /*
1149          * We *must* clear the S2CR first, because freeing the SMR means
1150          * that it can be re-allocated immediately.
1151          */
1152         arm_smmu_bypass_stream_mapping(smmu, master);
1153         arm_smmu_master_free_smrs(smmu, master);
1154 }
1155
1156 static int arm_smmu_attach_dev(struct iommu_domain *domain, struct device *dev)
1157 {
1158         int ret = -EINVAL;
1159         struct arm_smmu_domain *smmu_domain = domain->priv;
1160         struct arm_smmu_device *device_smmu = dev->archdata.iommu;
1161         struct arm_smmu_master *master;
1162         unsigned long flags;
1163
1164         if (!device_smmu) {
1165                 dev_err(dev, "cannot attach to SMMU, is it on the same bus?\n");
1166                 return -ENXIO;
1167         }
1168
1169         /*
1170          * Sanity check the domain. We don't currently support domains
1171          * that cross between different SMMU chains.
1172          */
1173         spin_lock_irqsave(&smmu_domain->lock, flags);
1174         if (!smmu_domain->leaf_smmu) {
1175                 /* Now that we have a master, we can finalise the domain */
1176                 ret = arm_smmu_init_domain_context(domain, dev);
1177                 if (IS_ERR_VALUE(ret))
1178                         goto err_unlock;
1179
1180                 smmu_domain->leaf_smmu = device_smmu;
1181         } else if (smmu_domain->leaf_smmu != device_smmu) {
1182                 dev_err(dev,
1183                         "cannot attach to SMMU %s whilst already attached to domain on SMMU %s\n",
1184                         dev_name(smmu_domain->leaf_smmu->dev),
1185                         dev_name(device_smmu->dev));
1186                 goto err_unlock;
1187         }
1188         spin_unlock_irqrestore(&smmu_domain->lock, flags);
1189
1190         /* Looks ok, so add the device to the domain */
1191         master = find_smmu_master(smmu_domain->leaf_smmu, dev->of_node);
1192         if (!master)
1193                 return -ENODEV;
1194
1195         return arm_smmu_domain_add_master(smmu_domain, master);
1196
1197 err_unlock:
1198         spin_unlock_irqrestore(&smmu_domain->lock, flags);
1199         return ret;
1200 }
1201
1202 static void arm_smmu_detach_dev(struct iommu_domain *domain, struct device *dev)
1203 {
1204         struct arm_smmu_domain *smmu_domain = domain->priv;
1205         struct arm_smmu_master *master;
1206
1207         master = find_smmu_master(smmu_domain->leaf_smmu, dev->of_node);
1208         if (master)
1209                 arm_smmu_domain_remove_master(smmu_domain, master);
1210 }
1211
1212 static bool arm_smmu_pte_is_contiguous_range(unsigned long addr,
1213                                              unsigned long end)
1214 {
1215         return !(addr & ~ARM_SMMU_PTE_CONT_MASK) &&
1216                 (addr + ARM_SMMU_PTE_CONT_SIZE <= end);
1217 }
1218
1219 static int arm_smmu_alloc_init_pte(struct arm_smmu_device *smmu, pmd_t *pmd,
1220                                    unsigned long addr, unsigned long end,
1221                                    unsigned long pfn, int flags, int stage)
1222 {
1223         pte_t *pte, *start;
1224         pteval_t pteval = ARM_SMMU_PTE_PAGE | ARM_SMMU_PTE_AF | ARM_SMMU_PTE_XN;
1225
1226         if (pmd_none(*pmd)) {
1227                 /* Allocate a new set of tables */
1228                 pgtable_t table = alloc_page(GFP_ATOMIC|__GFP_ZERO);
1229                 if (!table)
1230                         return -ENOMEM;
1231
1232                 arm_smmu_flush_pgtable(smmu, page_address(table), PAGE_SIZE);
1233                 if (!pgtable_page_ctor(table)) {
1234                         __free_page(table);
1235                         return -ENOMEM;
1236                 }
1237                 pmd_populate(NULL, pmd, table);
1238                 arm_smmu_flush_pgtable(smmu, pmd, sizeof(*pmd));
1239         }
1240
1241         if (stage == 1) {
1242                 pteval |= ARM_SMMU_PTE_AP_UNPRIV | ARM_SMMU_PTE_nG;
1243                 if (!(flags & IOMMU_WRITE) && (flags & IOMMU_READ))
1244                         pteval |= ARM_SMMU_PTE_AP_RDONLY;
1245
1246                 if (flags & IOMMU_CACHE)
1247                         pteval |= (MAIR_ATTR_IDX_CACHE <<
1248                                    ARM_SMMU_PTE_ATTRINDX_SHIFT);
1249         } else {
1250                 pteval |= ARM_SMMU_PTE_HAP_FAULT;
1251                 if (flags & IOMMU_READ)
1252                         pteval |= ARM_SMMU_PTE_HAP_READ;
1253                 if (flags & IOMMU_WRITE)
1254                         pteval |= ARM_SMMU_PTE_HAP_WRITE;
1255                 if (flags & IOMMU_CACHE)
1256                         pteval |= ARM_SMMU_PTE_MEMATTR_OIWB;
1257                 else
1258                         pteval |= ARM_SMMU_PTE_MEMATTR_NC;
1259         }
1260
1261         /* If no access, create a faulting entry to avoid TLB fills */
1262         if (flags & IOMMU_EXEC)
1263                 pteval &= ~ARM_SMMU_PTE_XN;
1264         else if (!(flags & (IOMMU_READ | IOMMU_WRITE)))
1265                 pteval &= ~ARM_SMMU_PTE_PAGE;
1266
1267         pteval |= ARM_SMMU_PTE_SH_IS;
1268         start = pmd_page_vaddr(*pmd) + pte_index(addr);
1269         pte = start;
1270
1271         /*
1272          * Install the page table entries. This is fairly complicated
1273          * since we attempt to make use of the contiguous hint in the
1274          * ptes where possible. The contiguous hint indicates a series
1275          * of ARM_SMMU_PTE_CONT_ENTRIES ptes mapping a physically
1276          * contiguous region with the following constraints:
1277          *
1278          *   - The region start is aligned to ARM_SMMU_PTE_CONT_SIZE
1279          *   - Each pte in the region has the contiguous hint bit set
1280          *
1281          * This complicates unmapping (also handled by this code, when
1282          * neither IOMMU_READ or IOMMU_WRITE are set) because it is
1283          * possible, yet highly unlikely, that a client may unmap only
1284          * part of a contiguous range. This requires clearing of the
1285          * contiguous hint bits in the range before installing the new
1286          * faulting entries.
1287          *
1288          * Note that re-mapping an address range without first unmapping
1289          * it is not supported, so TLB invalidation is not required here
1290          * and is instead performed at unmap and domain-init time.
1291          */
1292         do {
1293                 int i = 1;
1294                 pteval &= ~ARM_SMMU_PTE_CONT;
1295
1296                 if (arm_smmu_pte_is_contiguous_range(addr, end)) {
1297                         i = ARM_SMMU_PTE_CONT_ENTRIES;
1298                         pteval |= ARM_SMMU_PTE_CONT;
1299                 } else if (pte_val(*pte) &
1300                            (ARM_SMMU_PTE_CONT | ARM_SMMU_PTE_PAGE)) {
1301                         int j;
1302                         pte_t *cont_start;
1303                         unsigned long idx = pte_index(addr);
1304
1305                         idx &= ~(ARM_SMMU_PTE_CONT_ENTRIES - 1);
1306                         cont_start = pmd_page_vaddr(*pmd) + idx;
1307                         for (j = 0; j < ARM_SMMU_PTE_CONT_ENTRIES; ++j)
1308                                 pte_val(*(cont_start + j)) &= ~ARM_SMMU_PTE_CONT;
1309
1310                         arm_smmu_flush_pgtable(smmu, cont_start,
1311                                                sizeof(*pte) *
1312                                                ARM_SMMU_PTE_CONT_ENTRIES);
1313                 }
1314
1315                 do {
1316                         *pte = pfn_pte(pfn, __pgprot(pteval));
1317                 } while (pte++, pfn++, addr += PAGE_SIZE, --i);
1318         } while (addr != end);
1319
1320         arm_smmu_flush_pgtable(smmu, start, sizeof(*pte) * (pte - start));
1321         return 0;
1322 }
1323
1324 static int arm_smmu_alloc_init_pmd(struct arm_smmu_device *smmu, pud_t *pud,
1325                                    unsigned long addr, unsigned long end,
1326                                    phys_addr_t phys, int flags, int stage)
1327 {
1328         int ret;
1329         pmd_t *pmd;
1330         unsigned long next, pfn = __phys_to_pfn(phys);
1331
1332 #ifndef __PAGETABLE_PMD_FOLDED
1333         if (pud_none(*pud)) {
1334                 pmd = (pmd_t *)get_zeroed_page(GFP_ATOMIC);
1335                 if (!pmd)
1336                         return -ENOMEM;
1337
1338                 arm_smmu_flush_pgtable(smmu, pmd, PAGE_SIZE);
1339                 pud_populate(NULL, pud, pmd);
1340                 arm_smmu_flush_pgtable(smmu, pud, sizeof(*pud));
1341
1342                 pmd += pmd_index(addr);
1343         } else
1344 #endif
1345                 pmd = pmd_offset(pud, addr);
1346
1347         do {
1348                 next = pmd_addr_end(addr, end);
1349                 ret = arm_smmu_alloc_init_pte(smmu, pmd, addr, end, pfn,
1350                                               flags, stage);
1351                 phys += next - addr;
1352         } while (pmd++, addr = next, addr < end);
1353
1354         return ret;
1355 }
1356
1357 static int arm_smmu_alloc_init_pud(struct arm_smmu_device *smmu, pgd_t *pgd,
1358                                    unsigned long addr, unsigned long end,
1359                                    phys_addr_t phys, int flags, int stage)
1360 {
1361         int ret = 0;
1362         pud_t *pud;
1363         unsigned long next;
1364
1365 #ifndef __PAGETABLE_PUD_FOLDED
1366         if (pgd_none(*pgd)) {
1367                 pud = (pud_t *)get_zeroed_page(GFP_ATOMIC);
1368                 if (!pud)
1369                         return -ENOMEM;
1370
1371                 arm_smmu_flush_pgtable(smmu, pud, PAGE_SIZE);
1372                 pgd_populate(NULL, pgd, pud);
1373                 arm_smmu_flush_pgtable(smmu, pgd, sizeof(*pgd));
1374
1375                 pud += pud_index(addr);
1376         } else
1377 #endif
1378                 pud = pud_offset(pgd, addr);
1379
1380         do {
1381                 next = pud_addr_end(addr, end);
1382                 ret = arm_smmu_alloc_init_pmd(smmu, pud, addr, next, phys,
1383                                               flags, stage);
1384                 phys += next - addr;
1385         } while (pud++, addr = next, addr < end);
1386
1387         return ret;
1388 }
1389
1390 static int arm_smmu_handle_mapping(struct arm_smmu_domain *smmu_domain,
1391                                    unsigned long iova, phys_addr_t paddr,
1392                                    size_t size, int flags)
1393 {
1394         int ret, stage;
1395         unsigned long end;
1396         phys_addr_t input_mask, output_mask;
1397         struct arm_smmu_cfg *root_cfg = &smmu_domain->root_cfg;
1398         pgd_t *pgd = root_cfg->pgd;
1399         struct arm_smmu_device *smmu = root_cfg->smmu;
1400         unsigned long irqflags;
1401
1402         if (root_cfg->cbar == CBAR_TYPE_S2_TRANS) {
1403                 stage = 2;
1404                 output_mask = (1ULL << smmu->s2_output_size) - 1;
1405         } else {
1406                 stage = 1;
1407                 output_mask = (1ULL << smmu->s1_output_size) - 1;
1408         }
1409
1410         if (!pgd)
1411                 return -EINVAL;
1412
1413         if (size & ~PAGE_MASK)
1414                 return -EINVAL;
1415
1416         input_mask = (1ULL << smmu->input_size) - 1;
1417         if ((phys_addr_t)iova & ~input_mask)
1418                 return -ERANGE;
1419
1420         if (paddr & ~output_mask)
1421                 return -ERANGE;
1422
1423         spin_lock_irqsave(&smmu_domain->lock, irqflags);
1424         pgd += pgd_index(iova);
1425         end = iova + size;
1426         do {
1427                 unsigned long next = pgd_addr_end(iova, end);
1428
1429                 ret = arm_smmu_alloc_init_pud(smmu, pgd, iova, next, paddr,
1430                                               flags, stage);
1431                 if (ret)
1432                         goto out_unlock;
1433
1434                 paddr += next - iova;
1435                 iova = next;
1436         } while (pgd++, iova != end);
1437
1438 out_unlock:
1439         spin_unlock_irqrestore(&smmu_domain->lock, irqflags);
1440
1441         return ret;
1442 }
1443
1444 static int arm_smmu_map(struct iommu_domain *domain, unsigned long iova,
1445                         phys_addr_t paddr, size_t size, int flags)
1446 {
1447         struct arm_smmu_domain *smmu_domain = domain->priv;
1448
1449         if (!smmu_domain)
1450                 return -ENODEV;
1451
1452         /* Check for silent address truncation up the SMMU chain. */
1453         if ((phys_addr_t)iova & ~smmu_domain->output_mask)
1454                 return -ERANGE;
1455
1456         return arm_smmu_handle_mapping(smmu_domain, iova, paddr, size, flags);
1457 }
1458
1459 static size_t arm_smmu_unmap(struct iommu_domain *domain, unsigned long iova,
1460                              size_t size)
1461 {
1462         int ret;
1463         struct arm_smmu_domain *smmu_domain = domain->priv;
1464
1465         ret = arm_smmu_handle_mapping(smmu_domain, iova, 0, size, 0);
1466         arm_smmu_tlb_inv_context(&smmu_domain->root_cfg);
1467         return ret ? ret : size;
1468 }
1469
1470 static phys_addr_t arm_smmu_iova_to_phys(struct iommu_domain *domain,
1471                                          dma_addr_t iova)
1472 {
1473         pgd_t *pgdp, pgd;
1474         pud_t pud;
1475         pmd_t pmd;
1476         pte_t pte;
1477         struct arm_smmu_domain *smmu_domain = domain->priv;
1478         struct arm_smmu_cfg *root_cfg = &smmu_domain->root_cfg;
1479
1480         pgdp = root_cfg->pgd;
1481         if (!pgdp)
1482                 return 0;
1483
1484         pgd = *(pgdp + pgd_index(iova));
1485         if (pgd_none(pgd))
1486                 return 0;
1487
1488         pud = *pud_offset(&pgd, iova);
1489         if (pud_none(pud))
1490                 return 0;
1491
1492         pmd = *pmd_offset(&pud, iova);
1493         if (pmd_none(pmd))
1494                 return 0;
1495
1496         pte = *(pmd_page_vaddr(pmd) + pte_index(iova));
1497         if (pte_none(pte))
1498                 return 0;
1499
1500         return __pfn_to_phys(pte_pfn(pte)) | (iova & ~PAGE_MASK);
1501 }
1502
1503 static int arm_smmu_domain_has_cap(struct iommu_domain *domain,
1504                                    unsigned long cap)
1505 {
1506         unsigned long caps = 0;
1507         struct arm_smmu_domain *smmu_domain = domain->priv;
1508
1509         if (smmu_domain->root_cfg.smmu->features & ARM_SMMU_FEAT_COHERENT_WALK)
1510                 caps |= IOMMU_CAP_CACHE_COHERENCY;
1511
1512         return !!(cap & caps);
1513 }
1514
1515 static int arm_smmu_add_device(struct device *dev)
1516 {
1517         struct arm_smmu_device *child, *parent, *smmu;
1518         struct arm_smmu_master *master = NULL;
1519         struct iommu_group *group;
1520         int ret;
1521
1522         if (dev->archdata.iommu) {
1523                 dev_warn(dev, "IOMMU driver already assigned to device\n");
1524                 return -EINVAL;
1525         }
1526
1527         spin_lock(&arm_smmu_devices_lock);
1528         list_for_each_entry(parent, &arm_smmu_devices, list) {
1529                 smmu = parent;
1530
1531                 /* Try to find a child of the current SMMU. */
1532                 list_for_each_entry(child, &arm_smmu_devices, list) {
1533                         if (child->parent_of_node == parent->dev->of_node) {
1534                                 /* Does the child sit above our master? */
1535                                 master = find_smmu_master(child, dev->of_node);
1536                                 if (master) {
1537                                         smmu = NULL;
1538                                         break;
1539                                 }
1540                         }
1541                 }
1542
1543                 /* We found some children, so keep searching. */
1544                 if (!smmu) {
1545                         master = NULL;
1546                         continue;
1547                 }
1548
1549                 master = find_smmu_master(smmu, dev->of_node);
1550                 if (master)
1551                         break;
1552         }
1553         spin_unlock(&arm_smmu_devices_lock);
1554
1555         if (!master)
1556                 return -ENODEV;
1557
1558         group = iommu_group_alloc();
1559         if (IS_ERR(group)) {
1560                 dev_err(dev, "Failed to allocate IOMMU group\n");
1561                 return PTR_ERR(group);
1562         }
1563
1564         ret = iommu_group_add_device(group, dev);
1565         iommu_group_put(group);
1566         dev->archdata.iommu = smmu;
1567
1568         return ret;
1569 }
1570
1571 static void arm_smmu_remove_device(struct device *dev)
1572 {
1573         dev->archdata.iommu = NULL;
1574         iommu_group_remove_device(dev);
1575 }
1576
1577 static struct iommu_ops arm_smmu_ops = {
1578         .domain_init    = arm_smmu_domain_init,
1579         .domain_destroy = arm_smmu_domain_destroy,
1580         .attach_dev     = arm_smmu_attach_dev,
1581         .detach_dev     = arm_smmu_detach_dev,
1582         .map            = arm_smmu_map,
1583         .unmap          = arm_smmu_unmap,
1584         .iova_to_phys   = arm_smmu_iova_to_phys,
1585         .domain_has_cap = arm_smmu_domain_has_cap,
1586         .add_device     = arm_smmu_add_device,
1587         .remove_device  = arm_smmu_remove_device,
1588         .pgsize_bitmap  = (SECTION_SIZE |
1589                            ARM_SMMU_PTE_CONT_SIZE |
1590                            PAGE_SIZE),
1591 };
1592
1593 static void arm_smmu_device_reset(struct arm_smmu_device *smmu)
1594 {
1595         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1596         void __iomem *cb_base;
1597         int i = 0;
1598         u32 reg;
1599
1600         /* Clear Global FSR */
1601         reg = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSR);
1602         writel(reg, gr0_base + ARM_SMMU_GR0_sGFSR);
1603
1604         /* Mark all SMRn as invalid and all S2CRn as bypass */
1605         for (i = 0; i < smmu->num_mapping_groups; ++i) {
1606                 writel_relaxed(~SMR_VALID, gr0_base + ARM_SMMU_GR0_SMR(i));
1607                 writel_relaxed(S2CR_TYPE_BYPASS, gr0_base + ARM_SMMU_GR0_S2CR(i));
1608         }
1609
1610         /* Make sure all context banks are disabled and clear CB_FSR  */
1611         for (i = 0; i < smmu->num_context_banks; ++i) {
1612                 cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, i);
1613                 writel_relaxed(0, cb_base + ARM_SMMU_CB_SCTLR);
1614                 writel_relaxed(FSR_FAULT, cb_base + ARM_SMMU_CB_FSR);
1615         }
1616
1617         /* Invalidate the TLB, just in case */
1618         writel_relaxed(0, gr0_base + ARM_SMMU_GR0_STLBIALL);
1619         writel_relaxed(0, gr0_base + ARM_SMMU_GR0_TLBIALLH);
1620         writel_relaxed(0, gr0_base + ARM_SMMU_GR0_TLBIALLNSNH);
1621
1622         reg = readl_relaxed(gr0_base + ARM_SMMU_GR0_sCR0);
1623
1624         /* Enable fault reporting */
1625         reg |= (sCR0_GFRE | sCR0_GFIE | sCR0_GCFGFRE | sCR0_GCFGFIE);
1626
1627         /* Disable TLB broadcasting. */
1628         reg |= (sCR0_VMIDPNE | sCR0_PTM);
1629
1630         /* Enable client access, but bypass when no mapping is found */
1631         reg &= ~(sCR0_CLIENTPD | sCR0_USFCFG);
1632
1633         /* Disable forced broadcasting */
1634         reg &= ~sCR0_FB;
1635
1636         /* Don't upgrade barriers */
1637         reg &= ~(sCR0_BSU_MASK << sCR0_BSU_SHIFT);
1638
1639         /* Push the button */
1640         arm_smmu_tlb_sync(smmu);
1641         writel_relaxed(reg, gr0_base + ARM_SMMU_GR0_sCR0);
1642 }
1643
1644 static int arm_smmu_id_size_to_bits(int size)
1645 {
1646         switch (size) {
1647         case 0:
1648                 return 32;
1649         case 1:
1650                 return 36;
1651         case 2:
1652                 return 40;
1653         case 3:
1654                 return 42;
1655         case 4:
1656                 return 44;
1657         case 5:
1658         default:
1659                 return 48;
1660         }
1661 }
1662
1663 static int arm_smmu_device_cfg_probe(struct arm_smmu_device *smmu)
1664 {
1665         unsigned long size;
1666         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1667         u32 id;
1668
1669         dev_notice(smmu->dev, "probing hardware configuration...\n");
1670
1671         /* Primecell ID */
1672         id = readl_relaxed(gr0_base + ARM_SMMU_GR0_PIDR2);
1673         smmu->version = ((id >> PIDR2_ARCH_SHIFT) & PIDR2_ARCH_MASK) + 1;
1674         dev_notice(smmu->dev, "SMMUv%d with:\n", smmu->version);
1675
1676         /* ID0 */
1677         id = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID0);
1678 #ifndef CONFIG_64BIT
1679         if (((id >> ID0_PTFS_SHIFT) & ID0_PTFS_MASK) == ID0_PTFS_V8_ONLY) {
1680                 dev_err(smmu->dev, "\tno v7 descriptor support!\n");
1681                 return -ENODEV;
1682         }
1683 #endif
1684         if (id & ID0_S1TS) {
1685                 smmu->features |= ARM_SMMU_FEAT_TRANS_S1;
1686                 dev_notice(smmu->dev, "\tstage 1 translation\n");
1687         }
1688
1689         if (id & ID0_S2TS) {
1690                 smmu->features |= ARM_SMMU_FEAT_TRANS_S2;
1691                 dev_notice(smmu->dev, "\tstage 2 translation\n");
1692         }
1693
1694         if (id & ID0_NTS) {
1695                 smmu->features |= ARM_SMMU_FEAT_TRANS_NESTED;
1696                 dev_notice(smmu->dev, "\tnested translation\n");
1697         }
1698
1699         if (!(smmu->features &
1700                 (ARM_SMMU_FEAT_TRANS_S1 | ARM_SMMU_FEAT_TRANS_S2 |
1701                  ARM_SMMU_FEAT_TRANS_NESTED))) {
1702                 dev_err(smmu->dev, "\tno translation support!\n");
1703                 return -ENODEV;
1704         }
1705
1706         if (id & ID0_CTTW) {
1707                 smmu->features |= ARM_SMMU_FEAT_COHERENT_WALK;
1708                 dev_notice(smmu->dev, "\tcoherent table walk\n");
1709         }
1710
1711         if (id & ID0_SMS) {
1712                 u32 smr, sid, mask;
1713
1714                 smmu->features |= ARM_SMMU_FEAT_STREAM_MATCH;
1715                 smmu->num_mapping_groups = (id >> ID0_NUMSMRG_SHIFT) &
1716                                            ID0_NUMSMRG_MASK;
1717                 if (smmu->num_mapping_groups == 0) {
1718                         dev_err(smmu->dev,
1719                                 "stream-matching supported, but no SMRs present!\n");
1720                         return -ENODEV;
1721                 }
1722
1723                 smr = SMR_MASK_MASK << SMR_MASK_SHIFT;
1724                 smr |= (SMR_ID_MASK << SMR_ID_SHIFT);
1725                 writel_relaxed(smr, gr0_base + ARM_SMMU_GR0_SMR(0));
1726                 smr = readl_relaxed(gr0_base + ARM_SMMU_GR0_SMR(0));
1727
1728                 mask = (smr >> SMR_MASK_SHIFT) & SMR_MASK_MASK;
1729                 sid = (smr >> SMR_ID_SHIFT) & SMR_ID_MASK;
1730                 if ((mask & sid) != sid) {
1731                         dev_err(smmu->dev,
1732                                 "SMR mask bits (0x%x) insufficient for ID field (0x%x)\n",
1733                                 mask, sid);
1734                         return -ENODEV;
1735                 }
1736
1737                 dev_notice(smmu->dev,
1738                            "\tstream matching with %u register groups, mask 0x%x",
1739                            smmu->num_mapping_groups, mask);
1740         }
1741
1742         /* ID1 */
1743         id = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID1);
1744         smmu->pagesize = (id & ID1_PAGESIZE) ? SZ_64K : SZ_4K;
1745
1746         /* Check for size mismatch of SMMU address space from mapped region */
1747         size = 1 << (((id >> ID1_NUMPAGENDXB_SHIFT) & ID1_NUMPAGENDXB_MASK) + 1);
1748         size *= (smmu->pagesize << 1);
1749         if (smmu->size != size)
1750                 dev_warn(smmu->dev, "SMMU address space size (0x%lx) differs "
1751                         "from mapped region size (0x%lx)!\n", size, smmu->size);
1752
1753         smmu->num_s2_context_banks = (id >> ID1_NUMS2CB_SHIFT) &
1754                                       ID1_NUMS2CB_MASK;
1755         smmu->num_context_banks = (id >> ID1_NUMCB_SHIFT) & ID1_NUMCB_MASK;
1756         if (smmu->num_s2_context_banks > smmu->num_context_banks) {
1757                 dev_err(smmu->dev, "impossible number of S2 context banks!\n");
1758                 return -ENODEV;
1759         }
1760         dev_notice(smmu->dev, "\t%u context banks (%u stage-2 only)\n",
1761                    smmu->num_context_banks, smmu->num_s2_context_banks);
1762
1763         /* ID2 */
1764         id = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID2);
1765         size = arm_smmu_id_size_to_bits((id >> ID2_IAS_SHIFT) & ID2_IAS_MASK);
1766
1767         /*
1768          * Stage-1 output limited by stage-2 input size due to pgd
1769          * allocation (PTRS_PER_PGD).
1770          */
1771 #ifdef CONFIG_64BIT
1772         smmu->s1_output_size = min(39UL, size);
1773 #else
1774         smmu->s1_output_size = min(32UL, size);
1775 #endif
1776
1777         /* The stage-2 output mask is also applied for bypass */
1778         size = arm_smmu_id_size_to_bits((id >> ID2_OAS_SHIFT) & ID2_OAS_MASK);
1779         smmu->s2_output_size = min((unsigned long)PHYS_MASK_SHIFT, size);
1780
1781         if (smmu->version == 1) {
1782                 smmu->input_size = 32;
1783         } else {
1784 #ifdef CONFIG_64BIT
1785                 size = (id >> ID2_UBS_SHIFT) & ID2_UBS_MASK;
1786                 size = min(VA_BITS, arm_smmu_id_size_to_bits(size));
1787 #else
1788                 size = 32;
1789 #endif
1790                 smmu->input_size = size;
1791
1792                 if ((PAGE_SIZE == SZ_4K && !(id & ID2_PTFS_4K)) ||
1793                     (PAGE_SIZE == SZ_64K && !(id & ID2_PTFS_64K)) ||
1794                     (PAGE_SIZE != SZ_4K && PAGE_SIZE != SZ_64K)) {
1795                         dev_err(smmu->dev, "CPU page size 0x%lx unsupported\n",
1796                                 PAGE_SIZE);
1797                         return -ENODEV;
1798                 }
1799         }
1800
1801         dev_notice(smmu->dev,
1802                    "\t%lu-bit VA, %lu-bit IPA, %lu-bit PA\n",
1803                    smmu->input_size, smmu->s1_output_size, smmu->s2_output_size);
1804         return 0;
1805 }
1806
1807 static int arm_smmu_device_dt_probe(struct platform_device *pdev)
1808 {
1809         struct resource *res;
1810         struct arm_smmu_device *smmu;
1811         struct device_node *dev_node;
1812         struct device *dev = &pdev->dev;
1813         struct rb_node *node;
1814         struct of_phandle_args masterspec;
1815         int num_irqs, i, err;
1816
1817         smmu = devm_kzalloc(dev, sizeof(*smmu), GFP_KERNEL);
1818         if (!smmu) {
1819                 dev_err(dev, "failed to allocate arm_smmu_device\n");
1820                 return -ENOMEM;
1821         }
1822         smmu->dev = dev;
1823
1824         res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1825         smmu->base = devm_ioremap_resource(dev, res);
1826         if (IS_ERR(smmu->base))
1827                 return PTR_ERR(smmu->base);
1828         smmu->size = resource_size(res);
1829
1830         if (of_property_read_u32(dev->of_node, "#global-interrupts",
1831                                  &smmu->num_global_irqs)) {
1832                 dev_err(dev, "missing #global-interrupts property\n");
1833                 return -ENODEV;
1834         }
1835
1836         num_irqs = 0;
1837         while ((res = platform_get_resource(pdev, IORESOURCE_IRQ, num_irqs))) {
1838                 num_irqs++;
1839                 if (num_irqs > smmu->num_global_irqs)
1840                         smmu->num_context_irqs++;
1841         }
1842
1843         if (!smmu->num_context_irqs) {
1844                 dev_err(dev, "found %d interrupts but expected at least %d\n",
1845                         num_irqs, smmu->num_global_irqs + 1);
1846                 return -ENODEV;
1847         }
1848
1849         smmu->irqs = devm_kzalloc(dev, sizeof(*smmu->irqs) * num_irqs,
1850                                   GFP_KERNEL);
1851         if (!smmu->irqs) {
1852                 dev_err(dev, "failed to allocate %d irqs\n", num_irqs);
1853                 return -ENOMEM;
1854         }
1855
1856         for (i = 0; i < num_irqs; ++i) {
1857                 int irq = platform_get_irq(pdev, i);
1858                 if (irq < 0) {
1859                         dev_err(dev, "failed to get irq index %d\n", i);
1860                         return -ENODEV;
1861                 }
1862                 smmu->irqs[i] = irq;
1863         }
1864
1865         i = 0;
1866         smmu->masters = RB_ROOT;
1867         while (!of_parse_phandle_with_args(dev->of_node, "mmu-masters",
1868                                            "#stream-id-cells", i,
1869                                            &masterspec)) {
1870                 err = register_smmu_master(smmu, dev, &masterspec);
1871                 if (err) {
1872                         dev_err(dev, "failed to add master %s\n",
1873                                 masterspec.np->name);
1874                         goto out_put_masters;
1875                 }
1876
1877                 i++;
1878         }
1879         dev_notice(dev, "registered %d master devices\n", i);
1880
1881         if ((dev_node = of_parse_phandle(dev->of_node, "smmu-parent", 0)))
1882                 smmu->parent_of_node = dev_node;
1883
1884         err = arm_smmu_device_cfg_probe(smmu);
1885         if (err)
1886                 goto out_put_parent;
1887
1888         if (smmu->version > 1 &&
1889             smmu->num_context_banks != smmu->num_context_irqs) {
1890                 dev_err(dev,
1891                         "found only %d context interrupt(s) but %d required\n",
1892                         smmu->num_context_irqs, smmu->num_context_banks);
1893                 err = -ENODEV;
1894                 goto out_put_parent;
1895         }
1896
1897         for (i = 0; i < smmu->num_global_irqs; ++i) {
1898                 err = request_irq(smmu->irqs[i],
1899                                   arm_smmu_global_fault,
1900                                   IRQF_SHARED,
1901                                   "arm-smmu global fault",
1902                                   smmu);
1903                 if (err) {
1904                         dev_err(dev, "failed to request global IRQ %d (%u)\n",
1905                                 i, smmu->irqs[i]);
1906                         goto out_free_irqs;
1907                 }
1908         }
1909
1910         INIT_LIST_HEAD(&smmu->list);
1911         spin_lock(&arm_smmu_devices_lock);
1912         list_add(&smmu->list, &arm_smmu_devices);
1913         spin_unlock(&arm_smmu_devices_lock);
1914
1915         arm_smmu_device_reset(smmu);
1916         return 0;
1917
1918 out_free_irqs:
1919         while (i--)
1920                 free_irq(smmu->irqs[i], smmu);
1921
1922 out_put_parent:
1923         if (smmu->parent_of_node)
1924                 of_node_put(smmu->parent_of_node);
1925
1926 out_put_masters:
1927         for (node = rb_first(&smmu->masters); node; node = rb_next(node)) {
1928                 struct arm_smmu_master *master;
1929                 master = container_of(node, struct arm_smmu_master, node);
1930                 of_node_put(master->of_node);
1931         }
1932
1933         return err;
1934 }
1935
1936 static int arm_smmu_device_remove(struct platform_device *pdev)
1937 {
1938         int i;
1939         struct device *dev = &pdev->dev;
1940         struct arm_smmu_device *curr, *smmu = NULL;
1941         struct rb_node *node;
1942
1943         spin_lock(&arm_smmu_devices_lock);
1944         list_for_each_entry(curr, &arm_smmu_devices, list) {
1945                 if (curr->dev == dev) {
1946                         smmu = curr;
1947                         list_del(&smmu->list);
1948                         break;
1949                 }
1950         }
1951         spin_unlock(&arm_smmu_devices_lock);
1952
1953         if (!smmu)
1954                 return -ENODEV;
1955
1956         if (smmu->parent_of_node)
1957                 of_node_put(smmu->parent_of_node);
1958
1959         for (node = rb_first(&smmu->masters); node; node = rb_next(node)) {
1960                 struct arm_smmu_master *master;
1961                 master = container_of(node, struct arm_smmu_master, node);
1962                 of_node_put(master->of_node);
1963         }
1964
1965         if (!bitmap_empty(smmu->context_map, ARM_SMMU_MAX_CBS))
1966                 dev_err(dev, "removing device with active domains!\n");
1967
1968         for (i = 0; i < smmu->num_global_irqs; ++i)
1969                 free_irq(smmu->irqs[i], smmu);
1970
1971         /* Turn the thing off */
1972         writel_relaxed(sCR0_CLIENTPD, ARM_SMMU_GR0(smmu) + ARM_SMMU_GR0_sCR0);
1973         return 0;
1974 }
1975
1976 #ifdef CONFIG_OF
1977 static struct of_device_id arm_smmu_of_match[] = {
1978         { .compatible = "arm,smmu-v1", },
1979         { .compatible = "arm,smmu-v2", },
1980         { .compatible = "arm,mmu-400", },
1981         { .compatible = "arm,mmu-500", },
1982         { },
1983 };
1984 MODULE_DEVICE_TABLE(of, arm_smmu_of_match);
1985 #endif
1986
1987 static struct platform_driver arm_smmu_driver = {
1988         .driver = {
1989                 .owner          = THIS_MODULE,
1990                 .name           = "arm-smmu",
1991                 .of_match_table = of_match_ptr(arm_smmu_of_match),
1992         },
1993         .probe  = arm_smmu_device_dt_probe,
1994         .remove = arm_smmu_device_remove,
1995 };
1996
1997 static int __init arm_smmu_init(void)
1998 {
1999         int ret;
2000
2001         ret = platform_driver_register(&arm_smmu_driver);
2002         if (ret)
2003                 return ret;
2004
2005         /* Oh, for a proper bus abstraction */
2006         if (!iommu_present(&platform_bus_type))
2007                 bus_set_iommu(&platform_bus_type, &arm_smmu_ops);
2008
2009 #ifdef CONFIG_ARM_AMBA
2010         if (!iommu_present(&amba_bustype))
2011                 bus_set_iommu(&amba_bustype, &arm_smmu_ops);
2012 #endif
2013
2014         return 0;
2015 }
2016
2017 static void __exit arm_smmu_exit(void)
2018 {
2019         return platform_driver_unregister(&arm_smmu_driver);
2020 }
2021
2022 subsys_initcall(arm_smmu_init);
2023 module_exit(arm_smmu_exit);
2024
2025 MODULE_DESCRIPTION("IOMMU API for ARM architected SMMU implementations");
2026 MODULE_AUTHOR("Will Deacon <will.deacon@arm.com>");
2027 MODULE_LICENSE("GPL v2");