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 * - Context fault reporting
27 */
28
29 #define pr_fmt(fmt) "arm-smmu: " fmt
30
31 #include <linux/delay.h>
32 #include <linux/dma-mapping.h>
33 #include <linux/err.h>
34 #include <linux/interrupt.h>
35 #include <linux/io.h>
36 #include <linux/iommu.h>
37 #include <linux/iopoll.h>
38 #include <linux/module.h>
39 #include <linux/of.h>
40 #include <linux/of_address.h>
41 #include <linux/pci.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 "io-pgtable.h"
49
50 /* Maximum number of stream IDs assigned to a single device */
51 #define MAX_MASTER_STREAMIDS MAX_PHANDLE_ARGS
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 + (1 << (smmu)->pgshift))
62
63 /*
64 * SMMU global address space with conditional offset to access secure
65 * aliases of non-secure registers (e.g. nsCR0: 0x400, nsGFSR: 0x448,
66 * nsGFSYNR0: 0x450)
67 */
68 #define ARM_SMMU_GR0_NS(smmu) \
69 ((smmu)->base + \
70 ((smmu->options & ARM_SMMU_OPT_SECURE_CFG_ACCESS) \
71 ? 0x400 : 0))
72
73 #ifdef CONFIG_64BIT
74 #define smmu_writeq writeq_relaxed
75 #else
76 #define smmu_writeq(reg64, addr) \
77 do { \
78 u64 __val = (reg64); \
79 void __iomem *__addr = (addr); \
80 writel_relaxed(__val >> 32, __addr + 4); \
81 writel_relaxed(__val, __addr); \
82 } while (0)
83 #endif
84
85 /* Configuration registers */
86 #define ARM_SMMU_GR0_sCR0 0x0
87 #define sCR0_CLIENTPD (1 << 0)
88 #define sCR0_GFRE (1 << 1)
89 #define sCR0_GFIE (1 << 2)
90 #define sCR0_GCFGFRE (1 << 4)
91 #define sCR0_GCFGFIE (1 << 5)
92 #define sCR0_USFCFG (1 << 10)
93 #define sCR0_VMIDPNE (1 << 11)
94 #define sCR0_PTM (1 << 12)
95 #define sCR0_FB (1 << 13)
96 #define sCR0_BSU_SHIFT 14
97 #define sCR0_BSU_MASK 0x3
98
99 /* Identification registers */
100 #define ARM_SMMU_GR0_ID0 0x20
101 #define ARM_SMMU_GR0_ID1 0x24
102 #define ARM_SMMU_GR0_ID2 0x28
103 #define ARM_SMMU_GR0_ID3 0x2c
104 #define ARM_SMMU_GR0_ID4 0x30
105 #define ARM_SMMU_GR0_ID5 0x34
106 #define ARM_SMMU_GR0_ID6 0x38
107 #define ARM_SMMU_GR0_ID7 0x3c
108 #define ARM_SMMU_GR0_sGFSR 0x48
109 #define ARM_SMMU_GR0_sGFSYNR0 0x50
110 #define ARM_SMMU_GR0_sGFSYNR1 0x54
111 #define ARM_SMMU_GR0_sGFSYNR2 0x58
112
113 #define ID0_S1TS (1 << 30)
114 #define ID0_S2TS (1 << 29)
115 #define ID0_NTS (1 << 28)
116 #define ID0_SMS (1 << 27)
117 #define ID0_ATOSNS (1 << 26)
118 #define ID0_CTTW (1 << 14)
119 #define ID0_NUMIRPT_SHIFT 16
120 #define ID0_NUMIRPT_MASK 0xff
121 #define ID0_NUMSIDB_SHIFT 9
122 #define ID0_NUMSIDB_MASK 0xf
123 #define ID0_NUMSMRG_SHIFT 0
124 #define ID0_NUMSMRG_MASK 0xff
125
126 #define ID1_PAGESIZE (1 << 31)
127 #define ID1_NUMPAGENDXB_SHIFT 28
128 #define ID1_NUMPAGENDXB_MASK 7
129 #define ID1_NUMS2CB_SHIFT 16
130 #define ID1_NUMS2CB_MASK 0xff
131 #define ID1_NUMCB_SHIFT 0
132 #define ID1_NUMCB_MASK 0xff
133
134 #define ID2_OAS_SHIFT 4
135 #define ID2_OAS_MASK 0xf
136 #define ID2_IAS_SHIFT 0
137 #define ID2_IAS_MASK 0xf
138 #define ID2_UBS_SHIFT 8
139 #define ID2_UBS_MASK 0xf
140 #define ID2_PTFS_4K (1 << 12)
141 #define ID2_PTFS_16K (1 << 13)
142 #define ID2_PTFS_64K (1 << 14)
143
144 /* Global TLB invalidation */
145 #define ARM_SMMU_GR0_TLBIVMID 0x64
146 #define ARM_SMMU_GR0_TLBIALLNSNH 0x68
147 #define ARM_SMMU_GR0_TLBIALLH 0x6c
148 #define ARM_SMMU_GR0_sTLBGSYNC 0x70
149 #define ARM_SMMU_GR0_sTLBGSTATUS 0x74
150 #define sTLBGSTATUS_GSACTIVE (1 << 0)
151 #define TLB_LOOP_TIMEOUT 1000000 /* 1s! */
152
153 /* Stream mapping registers */
154 #define ARM_SMMU_GR0_SMR(n) (0x800 + ((n) << 2))
155 #define SMR_VALID (1 << 31)
156 #define SMR_MASK_SHIFT 16
157 #define SMR_MASK_MASK 0x7fff
158 #define SMR_ID_SHIFT 0
159 #define SMR_ID_MASK 0x7fff
160
161 #define ARM_SMMU_GR0_S2CR(n) (0xc00 + ((n) << 2))
162 #define S2CR_CBNDX_SHIFT 0
163 #define S2CR_CBNDX_MASK 0xff
164 #define S2CR_TYPE_SHIFT 16
165 #define S2CR_TYPE_MASK 0x3
166 #define S2CR_TYPE_TRANS (0 << S2CR_TYPE_SHIFT)
167 #define S2CR_TYPE_BYPASS (1 << S2CR_TYPE_SHIFT)
168 #define S2CR_TYPE_FAULT (2 << S2CR_TYPE_SHIFT)
169
170 /* Context bank attribute registers */
171 #define ARM_SMMU_GR1_CBAR(n) (0x0 + ((n) << 2))
172 #define CBAR_VMID_SHIFT 0
173 #define CBAR_VMID_MASK 0xff
174 #define CBAR_S1_BPSHCFG_SHIFT 8
175 #define CBAR_S1_BPSHCFG_MASK 3
176 #define CBAR_S1_BPSHCFG_NSH 3
177 #define CBAR_S1_MEMATTR_SHIFT 12
178 #define CBAR_S1_MEMATTR_MASK 0xf
179 #define CBAR_S1_MEMATTR_WB 0xf
180 #define CBAR_TYPE_SHIFT 16
181 #define CBAR_TYPE_MASK 0x3
182 #define CBAR_TYPE_S2_TRANS (0 << CBAR_TYPE_SHIFT)
183 #define CBAR_TYPE_S1_TRANS_S2_BYPASS (1 << CBAR_TYPE_SHIFT)
184 #define CBAR_TYPE_S1_TRANS_S2_FAULT (2 << CBAR_TYPE_SHIFT)
185 #define CBAR_TYPE_S1_TRANS_S2_TRANS (3 << CBAR_TYPE_SHIFT)
186 #define CBAR_IRPTNDX_SHIFT 24
187 #define CBAR_IRPTNDX_MASK 0xff
188
189 #define ARM_SMMU_GR1_CBA2R(n) (0x800 + ((n) << 2))
190 #define CBA2R_RW64_32BIT (0 << 0)
191 #define CBA2R_RW64_64BIT (1 << 0)
192
193 /* Translation context bank */
194 #define ARM_SMMU_CB_BASE(smmu) ((smmu)->base + ((smmu)->size >> 1))
195 #define ARM_SMMU_CB(smmu, n) ((n) * (1 << (smmu)->pgshift))
196
197 #define ARM_SMMU_CB_SCTLR 0x0
198 #define ARM_SMMU_CB_RESUME 0x8
199 #define ARM_SMMU_CB_TTBCR2 0x10
200 #define ARM_SMMU_CB_TTBR0 0x20
201 #define ARM_SMMU_CB_TTBR1 0x28
202 #define ARM_SMMU_CB_TTBCR 0x30
203 #define ARM_SMMU_CB_S1_MAIR0 0x38
204 #define ARM_SMMU_CB_S1_MAIR1 0x3c
205 #define ARM_SMMU_CB_PAR_LO 0x50
206 #define ARM_SMMU_CB_PAR_HI 0x54
207 #define ARM_SMMU_CB_FSR 0x58
208 #define ARM_SMMU_CB_FAR_LO 0x60
209 #define ARM_SMMU_CB_FAR_HI 0x64
210 #define ARM_SMMU_CB_FSYNR0 0x68
211 #define ARM_SMMU_CB_S1_TLBIVA 0x600
212 #define ARM_SMMU_CB_S1_TLBIASID 0x610
213 #define ARM_SMMU_CB_S1_TLBIVAL 0x620
214 #define ARM_SMMU_CB_S2_TLBIIPAS2 0x630
215 #define ARM_SMMU_CB_S2_TLBIIPAS2L 0x638
216 #define ARM_SMMU_CB_ATS1PR 0x800
217 #define ARM_SMMU_CB_ATSR 0x8f0
218
219 #define SCTLR_S1_ASIDPNE (1 << 12)
220 #define SCTLR_CFCFG (1 << 7)
221 #define SCTLR_CFIE (1 << 6)
222 #define SCTLR_CFRE (1 << 5)
223 #define SCTLR_E (1 << 4)
224 #define SCTLR_AFE (1 << 2)
225 #define SCTLR_TRE (1 << 1)
226 #define SCTLR_M (1 << 0)
227 #define SCTLR_EAE_SBOP (SCTLR_AFE | SCTLR_TRE)
228
229 #define CB_PAR_F (1 << 0)
230
231 #define ATSR_ACTIVE (1 << 0)
232
233 #define RESUME_RETRY (0 << 0)
234 #define RESUME_TERMINATE (1 << 0)
235
236 #define TTBCR2_SEP_SHIFT 15
237 #define TTBCR2_SEP_UPSTREAM (0x7 << TTBCR2_SEP_SHIFT)
238
239 #define TTBRn_ASID_SHIFT 48
240
241 #define FSR_MULTI (1 << 31)
242 #define FSR_SS (1 << 30)
243 #define FSR_UUT (1 << 8)
244 #define FSR_ASF (1 << 7)
245 #define FSR_TLBLKF (1 << 6)
246 #define FSR_TLBMCF (1 << 5)
247 #define FSR_EF (1 << 4)
248 #define FSR_PF (1 << 3)
249 #define FSR_AFF (1 << 2)
250 #define FSR_TF (1 << 1)
251
252 #define FSR_IGN (FSR_AFF | FSR_ASF | \
253 FSR_TLBMCF | FSR_TLBLKF)
254 #define FSR_FAULT (FSR_MULTI | FSR_SS | FSR_UUT | \
255 FSR_EF | FSR_PF | FSR_TF | FSR_IGN)
256
257 #define FSYNR0_WNR (1 << 4)
258
259 static int force_stage;
260 module_param_named(force_stage, force_stage, int, S_IRUGO);
261 MODULE_PARM_DESC(force_stage,
262 "Force SMMU mappings to be installed at a particular stage of translation. A value of '1' or '2' forces the corresponding stage. All other values are ignored (i.e. no stage is forced). Note that selecting a specific stage will disable support for nested translation.");
263
264 enum arm_smmu_arch_version {
265 ARM_SMMU_V1 = 1,
266 ARM_SMMU_V2,
267 };
268
269 struct arm_smmu_smr {
270 u8 idx;
271 u16 mask;
272 u16 id;
273 };
274
275 struct arm_smmu_master_cfg {
276 int num_streamids;
277 u16 streamids[MAX_MASTER_STREAMIDS];
278 struct arm_smmu_smr *smrs;
279 };
280
281 struct arm_smmu_master {
282 struct device_node *of_node;
283 struct rb_node node;
284 struct arm_smmu_master_cfg cfg;
285 };
286
287 struct arm_smmu_device {
288 struct device *dev;
289
290 void __iomem *base;
291 unsigned long size;
292 unsigned long pgshift;
293
294 #define ARM_SMMU_FEAT_COHERENT_WALK (1 << 0)
295 #define ARM_SMMU_FEAT_STREAM_MATCH (1 << 1)
296 #define ARM_SMMU_FEAT_TRANS_S1 (1 << 2)
297 #define ARM_SMMU_FEAT_TRANS_S2 (1 << 3)
298 #define ARM_SMMU_FEAT_TRANS_NESTED (1 << 4)
299 #define ARM_SMMU_FEAT_TRANS_OPS (1 << 5)
300 u32 features;
301
302 #define ARM_SMMU_OPT_SECURE_CFG_ACCESS (1 << 0)
303 u32 options;
304 enum arm_smmu_arch_version version;
305
306 u32 num_context_banks;
307 u32 num_s2_context_banks;
308 DECLARE_BITMAP(context_map, ARM_SMMU_MAX_CBS);
309 atomic_t irptndx;
310
311 u32 num_mapping_groups;
312 DECLARE_BITMAP(smr_map, ARM_SMMU_MAX_SMRS);
313
314 unsigned long va_size;
315 unsigned long ipa_size;
316 unsigned long pa_size;
317
318 u32 num_global_irqs;
319 u32 num_context_irqs;
320 unsigned int *irqs;
321
322 struct list_head list;
323 struct rb_root masters;
324 };
325
326 struct arm_smmu_cfg {
327 u8 cbndx;
328 u8 irptndx;
329 u32 cbar;
330 };
331 #define INVALID_IRPTNDX 0xff
332
333 #define ARM_SMMU_CB_ASID(cfg) ((cfg)->cbndx)
334 #define ARM_SMMU_CB_VMID(cfg) ((cfg)->cbndx + 1)
335
336 enum arm_smmu_domain_stage {
337 ARM_SMMU_DOMAIN_S1 = 0,
338 ARM_SMMU_DOMAIN_S2,
339 ARM_SMMU_DOMAIN_NESTED,
340 };
341
342 struct arm_smmu_domain {
343 struct arm_smmu_device *smmu;
344 struct io_pgtable_ops *pgtbl_ops;
345 spinlock_t pgtbl_lock;
346 struct arm_smmu_cfg cfg;
347 enum arm_smmu_domain_stage stage;
348 struct mutex init_mutex; /* Protects smmu pointer */
349 struct iommu_domain domain;
350 };
351
352 static struct iommu_ops arm_smmu_ops;
353
354 static DEFINE_SPINLOCK(arm_smmu_devices_lock);
355 static LIST_HEAD(arm_smmu_devices);
356
357 struct arm_smmu_option_prop {
358 u32 opt;
359 const char *prop;
360 };
361
362 static struct arm_smmu_option_prop arm_smmu_options[] = {
363 { ARM_SMMU_OPT_SECURE_CFG_ACCESS, "calxeda,smmu-secure-config-access" },
364 { 0, NULL},
365 };
366
to_smmu_domain(struct iommu_domain * dom)367 static struct arm_smmu_domain *to_smmu_domain(struct iommu_domain *dom)
368 {
369 return container_of(dom, struct arm_smmu_domain, domain);
370 }
371
parse_driver_options(struct arm_smmu_device * smmu)372 static void parse_driver_options(struct arm_smmu_device *smmu)
373 {
374 int i = 0;
375
376 do {
377 if (of_property_read_bool(smmu->dev->of_node,
378 arm_smmu_options[i].prop)) {
379 smmu->options |= arm_smmu_options[i].opt;
380 dev_notice(smmu->dev, "option %s\n",
381 arm_smmu_options[i].prop);
382 }
383 } while (arm_smmu_options[++i].opt);
384 }
385
dev_get_dev_node(struct device * dev)386 static struct device_node *dev_get_dev_node(struct device *dev)
387 {
388 if (dev_is_pci(dev)) {
389 struct pci_bus *bus = to_pci_dev(dev)->bus;
390
391 while (!pci_is_root_bus(bus))
392 bus = bus->parent;
393 return bus->bridge->parent->of_node;
394 }
395
396 return dev->of_node;
397 }
398
find_smmu_master(struct arm_smmu_device * smmu,struct device_node * dev_node)399 static struct arm_smmu_master *find_smmu_master(struct arm_smmu_device *smmu,
400 struct device_node *dev_node)
401 {
402 struct rb_node *node = smmu->masters.rb_node;
403
404 while (node) {
405 struct arm_smmu_master *master;
406
407 master = container_of(node, struct arm_smmu_master, node);
408
409 if (dev_node < master->of_node)
410 node = node->rb_left;
411 else if (dev_node > master->of_node)
412 node = node->rb_right;
413 else
414 return master;
415 }
416
417 return NULL;
418 }
419
420 static struct arm_smmu_master_cfg *
find_smmu_master_cfg(struct device * dev)421 find_smmu_master_cfg(struct device *dev)
422 {
423 struct arm_smmu_master_cfg *cfg = NULL;
424 struct iommu_group *group = iommu_group_get(dev);
425
426 if (group) {
427 cfg = iommu_group_get_iommudata(group);
428 iommu_group_put(group);
429 }
430
431 return cfg;
432 }
433
insert_smmu_master(struct arm_smmu_device * smmu,struct arm_smmu_master * master)434 static int insert_smmu_master(struct arm_smmu_device *smmu,
435 struct arm_smmu_master *master)
436 {
437 struct rb_node **new, *parent;
438
439 new = &smmu->masters.rb_node;
440 parent = NULL;
441 while (*new) {
442 struct arm_smmu_master *this
443 = container_of(*new, struct arm_smmu_master, node);
444
445 parent = *new;
446 if (master->of_node < this->of_node)
447 new = &((*new)->rb_left);
448 else if (master->of_node > this->of_node)
449 new = &((*new)->rb_right);
450 else
451 return -EEXIST;
452 }
453
454 rb_link_node(&master->node, parent, new);
455 rb_insert_color(&master->node, &smmu->masters);
456 return 0;
457 }
458
register_smmu_master(struct arm_smmu_device * smmu,struct device * dev,struct of_phandle_args * masterspec)459 static int register_smmu_master(struct arm_smmu_device *smmu,
460 struct device *dev,
461 struct of_phandle_args *masterspec)
462 {
463 int i;
464 struct arm_smmu_master *master;
465
466 master = find_smmu_master(smmu, masterspec->np);
467 if (master) {
468 dev_err(dev,
469 "rejecting multiple registrations for master device %s\n",
470 masterspec->np->name);
471 return -EBUSY;
472 }
473
474 if (masterspec->args_count > MAX_MASTER_STREAMIDS) {
475 dev_err(dev,
476 "reached maximum number (%d) of stream IDs for master device %s\n",
477 MAX_MASTER_STREAMIDS, masterspec->np->name);
478 return -ENOSPC;
479 }
480
481 master = devm_kzalloc(dev, sizeof(*master), GFP_KERNEL);
482 if (!master)
483 return -ENOMEM;
484
485 master->of_node = masterspec->np;
486 master->cfg.num_streamids = masterspec->args_count;
487
488 for (i = 0; i < master->cfg.num_streamids; ++i) {
489 u16 streamid = masterspec->args[i];
490
491 if (!(smmu->features & ARM_SMMU_FEAT_STREAM_MATCH) &&
492 (streamid >= smmu->num_mapping_groups)) {
493 dev_err(dev,
494 "stream ID for master device %s greater than maximum allowed (%d)\n",
495 masterspec->np->name, smmu->num_mapping_groups);
496 return -ERANGE;
497 }
498 master->cfg.streamids[i] = streamid;
499 }
500 return insert_smmu_master(smmu, master);
501 }
502
find_smmu_for_device(struct device * dev)503 static struct arm_smmu_device *find_smmu_for_device(struct device *dev)
504 {
505 struct arm_smmu_device *smmu;
506 struct arm_smmu_master *master = NULL;
507 struct device_node *dev_node = dev_get_dev_node(dev);
508
509 spin_lock(&arm_smmu_devices_lock);
510 list_for_each_entry(smmu, &arm_smmu_devices, list) {
511 master = find_smmu_master(smmu, dev_node);
512 if (master)
513 break;
514 }
515 spin_unlock(&arm_smmu_devices_lock);
516
517 return master ? smmu : NULL;
518 }
519
__arm_smmu_alloc_bitmap(unsigned long * map,int start,int end)520 static int __arm_smmu_alloc_bitmap(unsigned long *map, int start, int end)
521 {
522 int idx;
523
524 do {
525 idx = find_next_zero_bit(map, end, start);
526 if (idx == end)
527 return -ENOSPC;
528 } while (test_and_set_bit(idx, map));
529
530 return idx;
531 }
532
__arm_smmu_free_bitmap(unsigned long * map,int idx)533 static void __arm_smmu_free_bitmap(unsigned long *map, int idx)
534 {
535 clear_bit(idx, map);
536 }
537
538 /* Wait for any pending TLB invalidations to complete */
__arm_smmu_tlb_sync(struct arm_smmu_device * smmu)539 static void __arm_smmu_tlb_sync(struct arm_smmu_device *smmu)
540 {
541 int count = 0;
542 void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
543
544 writel_relaxed(0, gr0_base + ARM_SMMU_GR0_sTLBGSYNC);
545 while (readl_relaxed(gr0_base + ARM_SMMU_GR0_sTLBGSTATUS)
546 & sTLBGSTATUS_GSACTIVE) {
547 cpu_relax();
548 if (++count == TLB_LOOP_TIMEOUT) {
549 dev_err_ratelimited(smmu->dev,
550 "TLB sync timed out -- SMMU may be deadlocked\n");
551 return;
552 }
553 udelay(1);
554 }
555 }
556
arm_smmu_tlb_sync(void * cookie)557 static void arm_smmu_tlb_sync(void *cookie)
558 {
559 struct arm_smmu_domain *smmu_domain = cookie;
560 __arm_smmu_tlb_sync(smmu_domain->smmu);
561 }
562
arm_smmu_tlb_inv_context(void * cookie)563 static void arm_smmu_tlb_inv_context(void *cookie)
564 {
565 struct arm_smmu_domain *smmu_domain = cookie;
566 struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
567 struct arm_smmu_device *smmu = smmu_domain->smmu;
568 bool stage1 = cfg->cbar != CBAR_TYPE_S2_TRANS;
569 void __iomem *base;
570
571 if (stage1) {
572 base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
573 writel_relaxed(ARM_SMMU_CB_ASID(cfg),
574 base + ARM_SMMU_CB_S1_TLBIASID);
575 } else {
576 base = ARM_SMMU_GR0(smmu);
577 writel_relaxed(ARM_SMMU_CB_VMID(cfg),
578 base + ARM_SMMU_GR0_TLBIVMID);
579 }
580
581 __arm_smmu_tlb_sync(smmu);
582 }
583
arm_smmu_tlb_inv_range_nosync(unsigned long iova,size_t size,bool leaf,void * cookie)584 static void arm_smmu_tlb_inv_range_nosync(unsigned long iova, size_t size,
585 bool leaf, void *cookie)
586 {
587 struct arm_smmu_domain *smmu_domain = cookie;
588 struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
589 struct arm_smmu_device *smmu = smmu_domain->smmu;
590 bool stage1 = cfg->cbar != CBAR_TYPE_S2_TRANS;
591 void __iomem *reg;
592
593 if (stage1) {
594 reg = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
595 reg += leaf ? ARM_SMMU_CB_S1_TLBIVAL : ARM_SMMU_CB_S1_TLBIVA;
596
597 if (!IS_ENABLED(CONFIG_64BIT) || smmu->version == ARM_SMMU_V1) {
598 iova &= ~12UL;
599 iova |= ARM_SMMU_CB_ASID(cfg);
600 writel_relaxed(iova, reg);
601 #ifdef CONFIG_64BIT
602 } else {
603 iova >>= 12;
604 iova |= (u64)ARM_SMMU_CB_ASID(cfg) << 48;
605 writeq_relaxed(iova, reg);
606 #endif
607 }
608 #ifdef CONFIG_64BIT
609 } else if (smmu->version == ARM_SMMU_V2) {
610 reg = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
611 reg += leaf ? ARM_SMMU_CB_S2_TLBIIPAS2L :
612 ARM_SMMU_CB_S2_TLBIIPAS2;
613 writeq_relaxed(iova >> 12, reg);
614 #endif
615 } else {
616 reg = ARM_SMMU_GR0(smmu) + ARM_SMMU_GR0_TLBIVMID;
617 writel_relaxed(ARM_SMMU_CB_VMID(cfg), reg);
618 }
619 }
620
621 static struct iommu_gather_ops arm_smmu_gather_ops = {
622 .tlb_flush_all = arm_smmu_tlb_inv_context,
623 .tlb_add_flush = arm_smmu_tlb_inv_range_nosync,
624 .tlb_sync = arm_smmu_tlb_sync,
625 };
626
arm_smmu_context_fault(int irq,void * dev)627 static irqreturn_t arm_smmu_context_fault(int irq, void *dev)
628 {
629 int flags, ret;
630 u32 fsr, far, fsynr, resume;
631 unsigned long iova;
632 struct iommu_domain *domain = dev;
633 struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
634 struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
635 struct arm_smmu_device *smmu = smmu_domain->smmu;
636 void __iomem *cb_base;
637
638 cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
639 fsr = readl_relaxed(cb_base + ARM_SMMU_CB_FSR);
640
641 if (!(fsr & FSR_FAULT))
642 return IRQ_NONE;
643
644 if (fsr & FSR_IGN)
645 dev_err_ratelimited(smmu->dev,
646 "Unexpected context fault (fsr 0x%x)\n",
647 fsr);
648
649 fsynr = readl_relaxed(cb_base + ARM_SMMU_CB_FSYNR0);
650 flags = fsynr & FSYNR0_WNR ? IOMMU_FAULT_WRITE : IOMMU_FAULT_READ;
651
652 far = readl_relaxed(cb_base + ARM_SMMU_CB_FAR_LO);
653 iova = far;
654 #ifdef CONFIG_64BIT
655 far = readl_relaxed(cb_base + ARM_SMMU_CB_FAR_HI);
656 iova |= ((unsigned long)far << 32);
657 #endif
658
659 if (!report_iommu_fault(domain, smmu->dev, iova, flags)) {
660 ret = IRQ_HANDLED;
661 resume = RESUME_RETRY;
662 } else {
663 dev_err_ratelimited(smmu->dev,
664 "Unhandled context fault: iova=0x%08lx, fsynr=0x%x, cb=%d\n",
665 iova, fsynr, cfg->cbndx);
666 ret = IRQ_NONE;
667 resume = RESUME_TERMINATE;
668 }
669
670 /* Clear the faulting FSR */
671 writel(fsr, cb_base + ARM_SMMU_CB_FSR);
672
673 /* Retry or terminate any stalled transactions */
674 if (fsr & FSR_SS)
675 writel_relaxed(resume, cb_base + ARM_SMMU_CB_RESUME);
676
677 return ret;
678 }
679
arm_smmu_global_fault(int irq,void * dev)680 static irqreturn_t arm_smmu_global_fault(int irq, void *dev)
681 {
682 u32 gfsr, gfsynr0, gfsynr1, gfsynr2;
683 struct arm_smmu_device *smmu = dev;
684 void __iomem *gr0_base = ARM_SMMU_GR0_NS(smmu);
685
686 gfsr = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSR);
687 gfsynr0 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSYNR0);
688 gfsynr1 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSYNR1);
689 gfsynr2 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSYNR2);
690
691 if (!gfsr)
692 return IRQ_NONE;
693
694 dev_err_ratelimited(smmu->dev,
695 "Unexpected global fault, this could be serious\n");
696 dev_err_ratelimited(smmu->dev,
697 "\tGFSR 0x%08x, GFSYNR0 0x%08x, GFSYNR1 0x%08x, GFSYNR2 0x%08x\n",
698 gfsr, gfsynr0, gfsynr1, gfsynr2);
699
700 writel(gfsr, gr0_base + ARM_SMMU_GR0_sGFSR);
701 return IRQ_HANDLED;
702 }
703
arm_smmu_init_context_bank(struct arm_smmu_domain * smmu_domain,struct io_pgtable_cfg * pgtbl_cfg)704 static void arm_smmu_init_context_bank(struct arm_smmu_domain *smmu_domain,
705 struct io_pgtable_cfg *pgtbl_cfg)
706 {
707 u32 reg;
708 u64 reg64;
709 bool stage1;
710 struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
711 struct arm_smmu_device *smmu = smmu_domain->smmu;
712 void __iomem *cb_base, *gr1_base;
713
714 gr1_base = ARM_SMMU_GR1(smmu);
715 stage1 = cfg->cbar != CBAR_TYPE_S2_TRANS;
716 cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
717
718 if (smmu->version > ARM_SMMU_V1) {
719 /*
720 * CBA2R.
721 * *Must* be initialised before CBAR thanks to VMID16
722 * architectural oversight affected some implementations.
723 */
724 #ifdef CONFIG_64BIT
725 reg = CBA2R_RW64_64BIT;
726 #else
727 reg = CBA2R_RW64_32BIT;
728 #endif
729 writel_relaxed(reg, gr1_base + ARM_SMMU_GR1_CBA2R(cfg->cbndx));
730 }
731
732 /* CBAR */
733 reg = cfg->cbar;
734 if (smmu->version == ARM_SMMU_V1)
735 reg |= cfg->irptndx << CBAR_IRPTNDX_SHIFT;
736
737 /*
738 * Use the weakest shareability/memory types, so they are
739 * overridden by the ttbcr/pte.
740 */
741 if (stage1) {
742 reg |= (CBAR_S1_BPSHCFG_NSH << CBAR_S1_BPSHCFG_SHIFT) |
743 (CBAR_S1_MEMATTR_WB << CBAR_S1_MEMATTR_SHIFT);
744 } else {
745 reg |= ARM_SMMU_CB_VMID(cfg) << CBAR_VMID_SHIFT;
746 }
747 writel_relaxed(reg, gr1_base + ARM_SMMU_GR1_CBAR(cfg->cbndx));
748
749 /* TTBRs */
750 if (stage1) {
751 reg64 = pgtbl_cfg->arm_lpae_s1_cfg.ttbr[0];
752
753 reg64 |= ((u64)ARM_SMMU_CB_ASID(cfg)) << TTBRn_ASID_SHIFT;
754 smmu_writeq(reg64, cb_base + ARM_SMMU_CB_TTBR0);
755
756 reg64 = pgtbl_cfg->arm_lpae_s1_cfg.ttbr[1];
757 reg64 |= ((u64)ARM_SMMU_CB_ASID(cfg)) << TTBRn_ASID_SHIFT;
758 smmu_writeq(reg64, cb_base + ARM_SMMU_CB_TTBR1);
759 } else {
760 reg64 = pgtbl_cfg->arm_lpae_s2_cfg.vttbr;
761 smmu_writeq(reg64, cb_base + ARM_SMMU_CB_TTBR0);
762 }
763
764 /* TTBCR */
765 if (stage1) {
766 reg = pgtbl_cfg->arm_lpae_s1_cfg.tcr;
767 writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBCR);
768 if (smmu->version > ARM_SMMU_V1) {
769 reg = pgtbl_cfg->arm_lpae_s1_cfg.tcr >> 32;
770 reg |= TTBCR2_SEP_UPSTREAM;
771 writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBCR2);
772 }
773 } else {
774 reg = pgtbl_cfg->arm_lpae_s2_cfg.vtcr;
775 writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBCR);
776 }
777
778 /* MAIRs (stage-1 only) */
779 if (stage1) {
780 reg = pgtbl_cfg->arm_lpae_s1_cfg.mair[0];
781 writel_relaxed(reg, cb_base + ARM_SMMU_CB_S1_MAIR0);
782 reg = pgtbl_cfg->arm_lpae_s1_cfg.mair[1];
783 writel_relaxed(reg, cb_base + ARM_SMMU_CB_S1_MAIR1);
784 }
785
786 /* SCTLR */
787 reg = SCTLR_CFCFG | SCTLR_CFIE | SCTLR_CFRE | SCTLR_M | SCTLR_EAE_SBOP;
788 if (stage1)
789 reg |= SCTLR_S1_ASIDPNE;
790 #ifdef __BIG_ENDIAN
791 reg |= SCTLR_E;
792 #endif
793 writel_relaxed(reg, cb_base + ARM_SMMU_CB_SCTLR);
794 }
795
arm_smmu_init_domain_context(struct iommu_domain * domain,struct arm_smmu_device * smmu)796 static int arm_smmu_init_domain_context(struct iommu_domain *domain,
797 struct arm_smmu_device *smmu)
798 {
799 int irq, start, ret = 0;
800 unsigned long ias, oas;
801 struct io_pgtable_ops *pgtbl_ops;
802 struct io_pgtable_cfg pgtbl_cfg;
803 enum io_pgtable_fmt fmt;
804 struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
805 struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
806
807 mutex_lock(&smmu_domain->init_mutex);
808 if (smmu_domain->smmu)
809 goto out_unlock;
810
811 /*
812 * Mapping the requested stage onto what we support is surprisingly
813 * complicated, mainly because the spec allows S1+S2 SMMUs without
814 * support for nested translation. That means we end up with the
815 * following table:
816 *
817 * Requested Supported Actual
818 * S1 N S1
819 * S1 S1+S2 S1
820 * S1 S2 S2
821 * S1 S1 S1
822 * N N N
823 * N S1+S2 S2
824 * N S2 S2
825 * N S1 S1
826 *
827 * Note that you can't actually request stage-2 mappings.
828 */
829 if (!(smmu->features & ARM_SMMU_FEAT_TRANS_S1))
830 smmu_domain->stage = ARM_SMMU_DOMAIN_S2;
831 if (!(smmu->features & ARM_SMMU_FEAT_TRANS_S2))
832 smmu_domain->stage = ARM_SMMU_DOMAIN_S1;
833
834 switch (smmu_domain->stage) {
835 case ARM_SMMU_DOMAIN_S1:
836 cfg->cbar = CBAR_TYPE_S1_TRANS_S2_BYPASS;
837 start = smmu->num_s2_context_banks;
838 ias = smmu->va_size;
839 oas = smmu->ipa_size;
840 if (IS_ENABLED(CONFIG_64BIT))
841 fmt = ARM_64_LPAE_S1;
842 else
843 fmt = ARM_32_LPAE_S1;
844 break;
845 case ARM_SMMU_DOMAIN_NESTED:
846 /*
847 * We will likely want to change this if/when KVM gets
848 * involved.
849 */
850 case ARM_SMMU_DOMAIN_S2:
851 cfg->cbar = CBAR_TYPE_S2_TRANS;
852 start = 0;
853 ias = smmu->ipa_size;
854 oas = smmu->pa_size;
855 if (IS_ENABLED(CONFIG_64BIT))
856 fmt = ARM_64_LPAE_S2;
857 else
858 fmt = ARM_32_LPAE_S2;
859 break;
860 default:
861 ret = -EINVAL;
862 goto out_unlock;
863 }
864
865 ret = __arm_smmu_alloc_bitmap(smmu->context_map, start,
866 smmu->num_context_banks);
867 if (IS_ERR_VALUE(ret))
868 goto out_unlock;
869
870 cfg->cbndx = ret;
871 if (smmu->version == ARM_SMMU_V1) {
872 cfg->irptndx = atomic_inc_return(&smmu->irptndx);
873 cfg->irptndx %= smmu->num_context_irqs;
874 } else {
875 cfg->irptndx = cfg->cbndx;
876 }
877
878 pgtbl_cfg = (struct io_pgtable_cfg) {
879 .pgsize_bitmap = arm_smmu_ops.pgsize_bitmap,
880 .ias = ias,
881 .oas = oas,
882 .tlb = &arm_smmu_gather_ops,
883 .iommu_dev = smmu->dev,
884 };
885
886 smmu_domain->smmu = smmu;
887 pgtbl_ops = alloc_io_pgtable_ops(fmt, &pgtbl_cfg, smmu_domain);
888 if (!pgtbl_ops) {
889 ret = -ENOMEM;
890 goto out_clear_smmu;
891 }
892
893 /* Update our support page sizes to reflect the page table format */
894 arm_smmu_ops.pgsize_bitmap = pgtbl_cfg.pgsize_bitmap;
895
896 /* Initialise the context bank with our page table cfg */
897 arm_smmu_init_context_bank(smmu_domain, &pgtbl_cfg);
898
899 /*
900 * Request context fault interrupt. Do this last to avoid the
901 * handler seeing a half-initialised domain state.
902 */
903 irq = smmu->irqs[smmu->num_global_irqs + cfg->irptndx];
904 ret = request_irq(irq, arm_smmu_context_fault, IRQF_SHARED,
905 "arm-smmu-context-fault", domain);
906 if (IS_ERR_VALUE(ret)) {
907 dev_err(smmu->dev, "failed to request context IRQ %d (%u)\n",
908 cfg->irptndx, irq);
909 cfg->irptndx = INVALID_IRPTNDX;
910 }
911
912 mutex_unlock(&smmu_domain->init_mutex);
913
914 /* Publish page table ops for map/unmap */
915 smmu_domain->pgtbl_ops = pgtbl_ops;
916 return 0;
917
918 out_clear_smmu:
919 smmu_domain->smmu = NULL;
920 out_unlock:
921 mutex_unlock(&smmu_domain->init_mutex);
922 return ret;
923 }
924
arm_smmu_destroy_domain_context(struct iommu_domain * domain)925 static void arm_smmu_destroy_domain_context(struct iommu_domain *domain)
926 {
927 struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
928 struct arm_smmu_device *smmu = smmu_domain->smmu;
929 struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
930 void __iomem *cb_base;
931 int irq;
932
933 if (!smmu)
934 return;
935
936 /*
937 * Disable the context bank and free the page tables before freeing
938 * it.
939 */
940 cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
941 writel_relaxed(0, cb_base + ARM_SMMU_CB_SCTLR);
942
943 if (cfg->irptndx != INVALID_IRPTNDX) {
944 irq = smmu->irqs[smmu->num_global_irqs + cfg->irptndx];
945 free_irq(irq, domain);
946 }
947
948 if (smmu_domain->pgtbl_ops)
949 free_io_pgtable_ops(smmu_domain->pgtbl_ops);
950
951 __arm_smmu_free_bitmap(smmu->context_map, cfg->cbndx);
952 }
953
arm_smmu_domain_alloc(unsigned type)954 static struct iommu_domain *arm_smmu_domain_alloc(unsigned type)
955 {
956 struct arm_smmu_domain *smmu_domain;
957
958 if (type != IOMMU_DOMAIN_UNMANAGED)
959 return NULL;
960 /*
961 * Allocate the domain and initialise some of its data structures.
962 * We can't really do anything meaningful until we've added a
963 * master.
964 */
965 smmu_domain = kzalloc(sizeof(*smmu_domain), GFP_KERNEL);
966 if (!smmu_domain)
967 return NULL;
968
969 mutex_init(&smmu_domain->init_mutex);
970 spin_lock_init(&smmu_domain->pgtbl_lock);
971
972 return &smmu_domain->domain;
973 }
974
arm_smmu_domain_free(struct iommu_domain * domain)975 static void arm_smmu_domain_free(struct iommu_domain *domain)
976 {
977 struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
978
979 /*
980 * Free the domain resources. We assume that all devices have
981 * already been detached.
982 */
983 arm_smmu_destroy_domain_context(domain);
984 kfree(smmu_domain);
985 }
986
arm_smmu_master_configure_smrs(struct arm_smmu_device * smmu,struct arm_smmu_master_cfg * cfg)987 static int arm_smmu_master_configure_smrs(struct arm_smmu_device *smmu,
988 struct arm_smmu_master_cfg *cfg)
989 {
990 int i;
991 struct arm_smmu_smr *smrs;
992 void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
993
994 if (!(smmu->features & ARM_SMMU_FEAT_STREAM_MATCH))
995 return 0;
996
997 if (cfg->smrs)
998 return -EEXIST;
999
1000 smrs = kmalloc_array(cfg->num_streamids, sizeof(*smrs), GFP_KERNEL);
1001 if (!smrs) {
1002 dev_err(smmu->dev, "failed to allocate %d SMRs\n",
1003 cfg->num_streamids);
1004 return -ENOMEM;
1005 }
1006
1007 /* Allocate the SMRs on the SMMU */
1008 for (i = 0; i < cfg->num_streamids; ++i) {
1009 int idx = __arm_smmu_alloc_bitmap(smmu->smr_map, 0,
1010 smmu->num_mapping_groups);
1011 if (IS_ERR_VALUE(idx)) {
1012 dev_err(smmu->dev, "failed to allocate free SMR\n");
1013 goto err_free_smrs;
1014 }
1015
1016 smrs[i] = (struct arm_smmu_smr) {
1017 .idx = idx,
1018 .mask = 0, /* We don't currently share SMRs */
1019 .id = cfg->streamids[i],
1020 };
1021 }
1022
1023 /* It worked! Now, poke the actual hardware */
1024 for (i = 0; i < cfg->num_streamids; ++i) {
1025 u32 reg = SMR_VALID | smrs[i].id << SMR_ID_SHIFT |
1026 smrs[i].mask << SMR_MASK_SHIFT;
1027 writel_relaxed(reg, gr0_base + ARM_SMMU_GR0_SMR(smrs[i].idx));
1028 }
1029
1030 cfg->smrs = smrs;
1031 return 0;
1032
1033 err_free_smrs:
1034 while (--i >= 0)
1035 __arm_smmu_free_bitmap(smmu->smr_map, smrs[i].idx);
1036 kfree(smrs);
1037 return -ENOSPC;
1038 }
1039
arm_smmu_master_free_smrs(struct arm_smmu_device * smmu,struct arm_smmu_master_cfg * cfg)1040 static void arm_smmu_master_free_smrs(struct arm_smmu_device *smmu,
1041 struct arm_smmu_master_cfg *cfg)
1042 {
1043 int i;
1044 void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1045 struct arm_smmu_smr *smrs = cfg->smrs;
1046
1047 if (!smrs)
1048 return;
1049
1050 /* Invalidate the SMRs before freeing back to the allocator */
1051 for (i = 0; i < cfg->num_streamids; ++i) {
1052 u8 idx = smrs[i].idx;
1053
1054 writel_relaxed(~SMR_VALID, gr0_base + ARM_SMMU_GR0_SMR(idx));
1055 __arm_smmu_free_bitmap(smmu->smr_map, idx);
1056 }
1057
1058 cfg->smrs = NULL;
1059 kfree(smrs);
1060 }
1061
arm_smmu_domain_add_master(struct arm_smmu_domain * smmu_domain,struct arm_smmu_master_cfg * cfg)1062 static int arm_smmu_domain_add_master(struct arm_smmu_domain *smmu_domain,
1063 struct arm_smmu_master_cfg *cfg)
1064 {
1065 int i, ret;
1066 struct arm_smmu_device *smmu = smmu_domain->smmu;
1067 void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1068
1069 /* Devices in an IOMMU group may already be configured */
1070 ret = arm_smmu_master_configure_smrs(smmu, cfg);
1071 if (ret)
1072 return ret == -EEXIST ? 0 : ret;
1073
1074 for (i = 0; i < cfg->num_streamids; ++i) {
1075 u32 idx, s2cr;
1076
1077 idx = cfg->smrs ? cfg->smrs[i].idx : cfg->streamids[i];
1078 s2cr = S2CR_TYPE_TRANS |
1079 (smmu_domain->cfg.cbndx << S2CR_CBNDX_SHIFT);
1080 writel_relaxed(s2cr, gr0_base + ARM_SMMU_GR0_S2CR(idx));
1081 }
1082
1083 return 0;
1084 }
1085
arm_smmu_domain_remove_master(struct arm_smmu_domain * smmu_domain,struct arm_smmu_master_cfg * cfg)1086 static void arm_smmu_domain_remove_master(struct arm_smmu_domain *smmu_domain,
1087 struct arm_smmu_master_cfg *cfg)
1088 {
1089 int i;
1090 struct arm_smmu_device *smmu = smmu_domain->smmu;
1091 void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1092
1093 /* An IOMMU group is torn down by the first device to be removed */
1094 if ((smmu->features & ARM_SMMU_FEAT_STREAM_MATCH) && !cfg->smrs)
1095 return;
1096
1097 /*
1098 * We *must* clear the S2CR first, because freeing the SMR means
1099 * that it can be re-allocated immediately.
1100 */
1101 for (i = 0; i < cfg->num_streamids; ++i) {
1102 u32 idx = cfg->smrs ? cfg->smrs[i].idx : cfg->streamids[i];
1103
1104 writel_relaxed(S2CR_TYPE_BYPASS,
1105 gr0_base + ARM_SMMU_GR0_S2CR(idx));
1106 }
1107
1108 arm_smmu_master_free_smrs(smmu, cfg);
1109 }
1110
arm_smmu_attach_dev(struct iommu_domain * domain,struct device * dev)1111 static int arm_smmu_attach_dev(struct iommu_domain *domain, struct device *dev)
1112 {
1113 int ret;
1114 struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
1115 struct arm_smmu_device *smmu;
1116 struct arm_smmu_master_cfg *cfg;
1117
1118 smmu = find_smmu_for_device(dev);
1119 if (!smmu) {
1120 dev_err(dev, "cannot attach to SMMU, is it on the same bus?\n");
1121 return -ENXIO;
1122 }
1123
1124 if (dev->archdata.iommu) {
1125 dev_err(dev, "already attached to IOMMU domain\n");
1126 return -EEXIST;
1127 }
1128
1129 /* Ensure that the domain is finalised */
1130 ret = arm_smmu_init_domain_context(domain, smmu);
1131 if (IS_ERR_VALUE(ret))
1132 return ret;
1133
1134 /*
1135 * Sanity check the domain. We don't support domains across
1136 * different SMMUs.
1137 */
1138 if (smmu_domain->smmu != smmu) {
1139 dev_err(dev,
1140 "cannot attach to SMMU %s whilst already attached to domain on SMMU %s\n",
1141 dev_name(smmu_domain->smmu->dev), dev_name(smmu->dev));
1142 return -EINVAL;
1143 }
1144
1145 /* Looks ok, so add the device to the domain */
1146 cfg = find_smmu_master_cfg(dev);
1147 if (!cfg)
1148 return -ENODEV;
1149
1150 ret = arm_smmu_domain_add_master(smmu_domain, cfg);
1151 if (!ret)
1152 dev->archdata.iommu = domain;
1153 return ret;
1154 }
1155
arm_smmu_detach_dev(struct iommu_domain * domain,struct device * dev)1156 static void arm_smmu_detach_dev(struct iommu_domain *domain, struct device *dev)
1157 {
1158 struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
1159 struct arm_smmu_master_cfg *cfg;
1160
1161 cfg = find_smmu_master_cfg(dev);
1162 if (!cfg)
1163 return;
1164
1165 dev->archdata.iommu = NULL;
1166 arm_smmu_domain_remove_master(smmu_domain, cfg);
1167 }
1168
arm_smmu_map(struct iommu_domain * domain,unsigned long iova,phys_addr_t paddr,size_t size,int prot)1169 static int arm_smmu_map(struct iommu_domain *domain, unsigned long iova,
1170 phys_addr_t paddr, size_t size, int prot)
1171 {
1172 int ret;
1173 unsigned long flags;
1174 struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
1175 struct io_pgtable_ops *ops= smmu_domain->pgtbl_ops;
1176
1177 if (!ops)
1178 return -ENODEV;
1179
1180 spin_lock_irqsave(&smmu_domain->pgtbl_lock, flags);
1181 ret = ops->map(ops, iova, paddr, size, prot);
1182 spin_unlock_irqrestore(&smmu_domain->pgtbl_lock, flags);
1183 return ret;
1184 }
1185
arm_smmu_unmap(struct iommu_domain * domain,unsigned long iova,size_t size)1186 static size_t arm_smmu_unmap(struct iommu_domain *domain, unsigned long iova,
1187 size_t size)
1188 {
1189 size_t ret;
1190 unsigned long flags;
1191 struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
1192 struct io_pgtable_ops *ops= smmu_domain->pgtbl_ops;
1193
1194 if (!ops)
1195 return 0;
1196
1197 spin_lock_irqsave(&smmu_domain->pgtbl_lock, flags);
1198 ret = ops->unmap(ops, iova, size);
1199 spin_unlock_irqrestore(&smmu_domain->pgtbl_lock, flags);
1200 return ret;
1201 }
1202
arm_smmu_iova_to_phys_hard(struct iommu_domain * domain,dma_addr_t iova)1203 static phys_addr_t arm_smmu_iova_to_phys_hard(struct iommu_domain *domain,
1204 dma_addr_t iova)
1205 {
1206 struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
1207 struct arm_smmu_device *smmu = smmu_domain->smmu;
1208 struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
1209 struct io_pgtable_ops *ops= smmu_domain->pgtbl_ops;
1210 struct device *dev = smmu->dev;
1211 void __iomem *cb_base;
1212 u32 tmp;
1213 u64 phys;
1214 unsigned long va;
1215
1216 cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
1217
1218 /* ATS1 registers can only be written atomically */
1219 va = iova & ~0xfffUL;
1220 if (smmu->version == ARM_SMMU_V2)
1221 smmu_writeq(va, cb_base + ARM_SMMU_CB_ATS1PR);
1222 else
1223 writel_relaxed(va, cb_base + ARM_SMMU_CB_ATS1PR);
1224
1225 if (readl_poll_timeout_atomic(cb_base + ARM_SMMU_CB_ATSR, tmp,
1226 !(tmp & ATSR_ACTIVE), 5, 50)) {
1227 dev_err(dev,
1228 "iova to phys timed out on %pad. Falling back to software table walk.\n",
1229 &iova);
1230 return ops->iova_to_phys(ops, iova);
1231 }
1232
1233 phys = readl_relaxed(cb_base + ARM_SMMU_CB_PAR_LO);
1234 phys |= ((u64)readl_relaxed(cb_base + ARM_SMMU_CB_PAR_HI)) << 32;
1235
1236 if (phys & CB_PAR_F) {
1237 dev_err(dev, "translation fault!\n");
1238 dev_err(dev, "PAR = 0x%llx\n", phys);
1239 return 0;
1240 }
1241
1242 return (phys & GENMASK_ULL(39, 12)) | (iova & 0xfff);
1243 }
1244
arm_smmu_iova_to_phys(struct iommu_domain * domain,dma_addr_t iova)1245 static phys_addr_t arm_smmu_iova_to_phys(struct iommu_domain *domain,
1246 dma_addr_t iova)
1247 {
1248 phys_addr_t ret;
1249 unsigned long flags;
1250 struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
1251 struct io_pgtable_ops *ops= smmu_domain->pgtbl_ops;
1252
1253 if (!ops)
1254 return 0;
1255
1256 spin_lock_irqsave(&smmu_domain->pgtbl_lock, flags);
1257 if (smmu_domain->smmu->features & ARM_SMMU_FEAT_TRANS_OPS &&
1258 smmu_domain->stage == ARM_SMMU_DOMAIN_S1) {
1259 ret = arm_smmu_iova_to_phys_hard(domain, iova);
1260 } else {
1261 ret = ops->iova_to_phys(ops, iova);
1262 }
1263
1264 spin_unlock_irqrestore(&smmu_domain->pgtbl_lock, flags);
1265
1266 return ret;
1267 }
1268
arm_smmu_capable(enum iommu_cap cap)1269 static bool arm_smmu_capable(enum iommu_cap cap)
1270 {
1271 switch (cap) {
1272 case IOMMU_CAP_CACHE_COHERENCY:
1273 /*
1274 * Return true here as the SMMU can always send out coherent
1275 * requests.
1276 */
1277 return true;
1278 case IOMMU_CAP_INTR_REMAP:
1279 return true; /* MSIs are just memory writes */
1280 case IOMMU_CAP_NOEXEC:
1281 return true;
1282 default:
1283 return false;
1284 }
1285 }
1286
__arm_smmu_get_pci_sid(struct pci_dev * pdev,u16 alias,void * data)1287 static int __arm_smmu_get_pci_sid(struct pci_dev *pdev, u16 alias, void *data)
1288 {
1289 *((u16 *)data) = alias;
1290 return 0; /* Continue walking */
1291 }
1292
__arm_smmu_release_pci_iommudata(void * data)1293 static void __arm_smmu_release_pci_iommudata(void *data)
1294 {
1295 kfree(data);
1296 }
1297
arm_smmu_init_pci_device(struct pci_dev * pdev,struct iommu_group * group)1298 static int arm_smmu_init_pci_device(struct pci_dev *pdev,
1299 struct iommu_group *group)
1300 {
1301 struct arm_smmu_master_cfg *cfg;
1302 u16 sid;
1303 int i;
1304
1305 cfg = iommu_group_get_iommudata(group);
1306 if (!cfg) {
1307 cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1308 if (!cfg)
1309 return -ENOMEM;
1310
1311 iommu_group_set_iommudata(group, cfg,
1312 __arm_smmu_release_pci_iommudata);
1313 }
1314
1315 if (cfg->num_streamids >= MAX_MASTER_STREAMIDS)
1316 return -ENOSPC;
1317
1318 /*
1319 * Assume Stream ID == Requester ID for now.
1320 * We need a way to describe the ID mappings in FDT.
1321 */
1322 pci_for_each_dma_alias(pdev, __arm_smmu_get_pci_sid, &sid);
1323 for (i = 0; i < cfg->num_streamids; ++i)
1324 if (cfg->streamids[i] == sid)
1325 break;
1326
1327 /* Avoid duplicate SIDs, as this can lead to SMR conflicts */
1328 if (i == cfg->num_streamids)
1329 cfg->streamids[cfg->num_streamids++] = sid;
1330
1331 return 0;
1332 }
1333
arm_smmu_init_platform_device(struct device * dev,struct iommu_group * group)1334 static int arm_smmu_init_platform_device(struct device *dev,
1335 struct iommu_group *group)
1336 {
1337 struct arm_smmu_device *smmu = find_smmu_for_device(dev);
1338 struct arm_smmu_master *master;
1339
1340 if (!smmu)
1341 return -ENODEV;
1342
1343 master = find_smmu_master(smmu, dev->of_node);
1344 if (!master)
1345 return -ENODEV;
1346
1347 iommu_group_set_iommudata(group, &master->cfg, NULL);
1348
1349 return 0;
1350 }
1351
arm_smmu_add_device(struct device * dev)1352 static int arm_smmu_add_device(struct device *dev)
1353 {
1354 struct iommu_group *group;
1355
1356 group = iommu_group_get_for_dev(dev);
1357 if (IS_ERR(group))
1358 return PTR_ERR(group);
1359
1360 return 0;
1361 }
1362
arm_smmu_remove_device(struct device * dev)1363 static void arm_smmu_remove_device(struct device *dev)
1364 {
1365 iommu_group_remove_device(dev);
1366 }
1367
arm_smmu_device_group(struct device * dev)1368 static struct iommu_group *arm_smmu_device_group(struct device *dev)
1369 {
1370 struct iommu_group *group;
1371 int ret;
1372
1373 if (dev_is_pci(dev))
1374 group = pci_device_group(dev);
1375 else
1376 group = generic_device_group(dev);
1377
1378 if (IS_ERR(group))
1379 return group;
1380
1381 if (dev_is_pci(dev))
1382 ret = arm_smmu_init_pci_device(to_pci_dev(dev), group);
1383 else
1384 ret = arm_smmu_init_platform_device(dev, group);
1385
1386 if (ret) {
1387 iommu_group_put(group);
1388 group = ERR_PTR(ret);
1389 }
1390
1391 return group;
1392 }
1393
arm_smmu_domain_get_attr(struct iommu_domain * domain,enum iommu_attr attr,void * data)1394 static int arm_smmu_domain_get_attr(struct iommu_domain *domain,
1395 enum iommu_attr attr, void *data)
1396 {
1397 struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
1398
1399 switch (attr) {
1400 case DOMAIN_ATTR_NESTING:
1401 *(int *)data = (smmu_domain->stage == ARM_SMMU_DOMAIN_NESTED);
1402 return 0;
1403 default:
1404 return -ENODEV;
1405 }
1406 }
1407
arm_smmu_domain_set_attr(struct iommu_domain * domain,enum iommu_attr attr,void * data)1408 static int arm_smmu_domain_set_attr(struct iommu_domain *domain,
1409 enum iommu_attr attr, void *data)
1410 {
1411 int ret = 0;
1412 struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
1413
1414 mutex_lock(&smmu_domain->init_mutex);
1415
1416 switch (attr) {
1417 case DOMAIN_ATTR_NESTING:
1418 if (smmu_domain->smmu) {
1419 ret = -EPERM;
1420 goto out_unlock;
1421 }
1422
1423 if (*(int *)data)
1424 smmu_domain->stage = ARM_SMMU_DOMAIN_NESTED;
1425 else
1426 smmu_domain->stage = ARM_SMMU_DOMAIN_S1;
1427
1428 break;
1429 default:
1430 ret = -ENODEV;
1431 }
1432
1433 out_unlock:
1434 mutex_unlock(&smmu_domain->init_mutex);
1435 return ret;
1436 }
1437
1438 static struct iommu_ops arm_smmu_ops = {
1439 .capable = arm_smmu_capable,
1440 .domain_alloc = arm_smmu_domain_alloc,
1441 .domain_free = arm_smmu_domain_free,
1442 .attach_dev = arm_smmu_attach_dev,
1443 .detach_dev = arm_smmu_detach_dev,
1444 .map = arm_smmu_map,
1445 .unmap = arm_smmu_unmap,
1446 .map_sg = default_iommu_map_sg,
1447 .iova_to_phys = arm_smmu_iova_to_phys,
1448 .add_device = arm_smmu_add_device,
1449 .remove_device = arm_smmu_remove_device,
1450 .device_group = arm_smmu_device_group,
1451 .domain_get_attr = arm_smmu_domain_get_attr,
1452 .domain_set_attr = arm_smmu_domain_set_attr,
1453 .pgsize_bitmap = -1UL, /* Restricted during device attach */
1454 };
1455
arm_smmu_device_reset(struct arm_smmu_device * smmu)1456 static void arm_smmu_device_reset(struct arm_smmu_device *smmu)
1457 {
1458 void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1459 void __iomem *cb_base;
1460 int i = 0;
1461 u32 reg;
1462
1463 /* clear global FSR */
1464 reg = readl_relaxed(ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sGFSR);
1465 writel(reg, ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sGFSR);
1466
1467 /* Mark all SMRn as invalid and all S2CRn as bypass */
1468 for (i = 0; i < smmu->num_mapping_groups; ++i) {
1469 writel_relaxed(0, gr0_base + ARM_SMMU_GR0_SMR(i));
1470 writel_relaxed(S2CR_TYPE_BYPASS,
1471 gr0_base + ARM_SMMU_GR0_S2CR(i));
1472 }
1473
1474 /* Make sure all context banks are disabled and clear CB_FSR */
1475 for (i = 0; i < smmu->num_context_banks; ++i) {
1476 cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, i);
1477 writel_relaxed(0, cb_base + ARM_SMMU_CB_SCTLR);
1478 writel_relaxed(FSR_FAULT, cb_base + ARM_SMMU_CB_FSR);
1479 }
1480
1481 /* Invalidate the TLB, just in case */
1482 writel_relaxed(0, gr0_base + ARM_SMMU_GR0_TLBIALLH);
1483 writel_relaxed(0, gr0_base + ARM_SMMU_GR0_TLBIALLNSNH);
1484
1485 reg = readl_relaxed(ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sCR0);
1486
1487 /* Enable fault reporting */
1488 reg |= (sCR0_GFRE | sCR0_GFIE | sCR0_GCFGFRE | sCR0_GCFGFIE);
1489
1490 /* Disable TLB broadcasting. */
1491 reg |= (sCR0_VMIDPNE | sCR0_PTM);
1492
1493 /* Enable client access, but bypass when no mapping is found */
1494 reg &= ~(sCR0_CLIENTPD | sCR0_USFCFG);
1495
1496 /* Disable forced broadcasting */
1497 reg &= ~sCR0_FB;
1498
1499 /* Don't upgrade barriers */
1500 reg &= ~(sCR0_BSU_MASK << sCR0_BSU_SHIFT);
1501
1502 /* Push the button */
1503 __arm_smmu_tlb_sync(smmu);
1504 writel(reg, ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sCR0);
1505 }
1506
arm_smmu_id_size_to_bits(int size)1507 static int arm_smmu_id_size_to_bits(int size)
1508 {
1509 switch (size) {
1510 case 0:
1511 return 32;
1512 case 1:
1513 return 36;
1514 case 2:
1515 return 40;
1516 case 3:
1517 return 42;
1518 case 4:
1519 return 44;
1520 case 5:
1521 default:
1522 return 48;
1523 }
1524 }
1525
arm_smmu_device_cfg_probe(struct arm_smmu_device * smmu)1526 static int arm_smmu_device_cfg_probe(struct arm_smmu_device *smmu)
1527 {
1528 unsigned long size;
1529 void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1530 u32 id;
1531 bool cttw_dt, cttw_reg;
1532
1533 dev_notice(smmu->dev, "probing hardware configuration...\n");
1534 dev_notice(smmu->dev, "SMMUv%d with:\n", smmu->version);
1535
1536 /* ID0 */
1537 id = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID0);
1538
1539 /* Restrict available stages based on module parameter */
1540 if (force_stage == 1)
1541 id &= ~(ID0_S2TS | ID0_NTS);
1542 else if (force_stage == 2)
1543 id &= ~(ID0_S1TS | ID0_NTS);
1544
1545 if (id & ID0_S1TS) {
1546 smmu->features |= ARM_SMMU_FEAT_TRANS_S1;
1547 dev_notice(smmu->dev, "\tstage 1 translation\n");
1548 }
1549
1550 if (id & ID0_S2TS) {
1551 smmu->features |= ARM_SMMU_FEAT_TRANS_S2;
1552 dev_notice(smmu->dev, "\tstage 2 translation\n");
1553 }
1554
1555 if (id & ID0_NTS) {
1556 smmu->features |= ARM_SMMU_FEAT_TRANS_NESTED;
1557 dev_notice(smmu->dev, "\tnested translation\n");
1558 }
1559
1560 if (!(smmu->features &
1561 (ARM_SMMU_FEAT_TRANS_S1 | ARM_SMMU_FEAT_TRANS_S2))) {
1562 dev_err(smmu->dev, "\tno translation support!\n");
1563 return -ENODEV;
1564 }
1565
1566 if ((id & ID0_S1TS) && ((smmu->version == 1) || !(id & ID0_ATOSNS))) {
1567 smmu->features |= ARM_SMMU_FEAT_TRANS_OPS;
1568 dev_notice(smmu->dev, "\taddress translation ops\n");
1569 }
1570
1571 /*
1572 * In order for DMA API calls to work properly, we must defer to what
1573 * the DT says about coherency, regardless of what the hardware claims.
1574 * Fortunately, this also opens up a workaround for systems where the
1575 * ID register value has ended up configured incorrectly.
1576 */
1577 cttw_dt = of_dma_is_coherent(smmu->dev->of_node);
1578 cttw_reg = !!(id & ID0_CTTW);
1579 if (cttw_dt)
1580 smmu->features |= ARM_SMMU_FEAT_COHERENT_WALK;
1581 if (cttw_dt || cttw_reg)
1582 dev_notice(smmu->dev, "\t%scoherent table walk\n",
1583 cttw_dt ? "" : "non-");
1584 if (cttw_dt != cttw_reg)
1585 dev_notice(smmu->dev,
1586 "\t(IDR0.CTTW overridden by dma-coherent property)\n");
1587
1588 if (id & ID0_SMS) {
1589 u32 smr, sid, mask;
1590
1591 smmu->features |= ARM_SMMU_FEAT_STREAM_MATCH;
1592 smmu->num_mapping_groups = (id >> ID0_NUMSMRG_SHIFT) &
1593 ID0_NUMSMRG_MASK;
1594 if (smmu->num_mapping_groups == 0) {
1595 dev_err(smmu->dev,
1596 "stream-matching supported, but no SMRs present!\n");
1597 return -ENODEV;
1598 }
1599
1600 smr = SMR_MASK_MASK << SMR_MASK_SHIFT;
1601 smr |= (SMR_ID_MASK << SMR_ID_SHIFT);
1602 writel_relaxed(smr, gr0_base + ARM_SMMU_GR0_SMR(0));
1603 smr = readl_relaxed(gr0_base + ARM_SMMU_GR0_SMR(0));
1604
1605 mask = (smr >> SMR_MASK_SHIFT) & SMR_MASK_MASK;
1606 sid = (smr >> SMR_ID_SHIFT) & SMR_ID_MASK;
1607 if ((mask & sid) != sid) {
1608 dev_err(smmu->dev,
1609 "SMR mask bits (0x%x) insufficient for ID field (0x%x)\n",
1610 mask, sid);
1611 return -ENODEV;
1612 }
1613
1614 dev_notice(smmu->dev,
1615 "\tstream matching with %u register groups, mask 0x%x",
1616 smmu->num_mapping_groups, mask);
1617 } else {
1618 smmu->num_mapping_groups = (id >> ID0_NUMSIDB_SHIFT) &
1619 ID0_NUMSIDB_MASK;
1620 }
1621
1622 /* ID1 */
1623 id = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID1);
1624 smmu->pgshift = (id & ID1_PAGESIZE) ? 16 : 12;
1625
1626 /* Check for size mismatch of SMMU address space from mapped region */
1627 size = 1 << (((id >> ID1_NUMPAGENDXB_SHIFT) & ID1_NUMPAGENDXB_MASK) + 1);
1628 size *= 2 << smmu->pgshift;
1629 if (smmu->size != size)
1630 dev_warn(smmu->dev,
1631 "SMMU address space size (0x%lx) differs from mapped region size (0x%lx)!\n",
1632 size, smmu->size);
1633
1634 smmu->num_s2_context_banks = (id >> ID1_NUMS2CB_SHIFT) & ID1_NUMS2CB_MASK;
1635 smmu->num_context_banks = (id >> ID1_NUMCB_SHIFT) & ID1_NUMCB_MASK;
1636 if (smmu->num_s2_context_banks > smmu->num_context_banks) {
1637 dev_err(smmu->dev, "impossible number of S2 context banks!\n");
1638 return -ENODEV;
1639 }
1640 dev_notice(smmu->dev, "\t%u context banks (%u stage-2 only)\n",
1641 smmu->num_context_banks, smmu->num_s2_context_banks);
1642
1643 /* ID2 */
1644 id = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID2);
1645 size = arm_smmu_id_size_to_bits((id >> ID2_IAS_SHIFT) & ID2_IAS_MASK);
1646 smmu->ipa_size = size;
1647
1648 /* The output mask is also applied for bypass */
1649 size = arm_smmu_id_size_to_bits((id >> ID2_OAS_SHIFT) & ID2_OAS_MASK);
1650 smmu->pa_size = size;
1651
1652 /*
1653 * What the page table walker can address actually depends on which
1654 * descriptor format is in use, but since a) we don't know that yet,
1655 * and b) it can vary per context bank, this will have to do...
1656 */
1657 if (dma_set_mask_and_coherent(smmu->dev, DMA_BIT_MASK(size)))
1658 dev_warn(smmu->dev,
1659 "failed to set DMA mask for table walker\n");
1660
1661 if (smmu->version == ARM_SMMU_V1) {
1662 smmu->va_size = smmu->ipa_size;
1663 size = SZ_4K | SZ_2M | SZ_1G;
1664 } else {
1665 size = (id >> ID2_UBS_SHIFT) & ID2_UBS_MASK;
1666 smmu->va_size = arm_smmu_id_size_to_bits(size);
1667 #ifndef CONFIG_64BIT
1668 smmu->va_size = min(32UL, smmu->va_size);
1669 #endif
1670 size = 0;
1671 if (id & ID2_PTFS_4K)
1672 size |= SZ_4K | SZ_2M | SZ_1G;
1673 if (id & ID2_PTFS_16K)
1674 size |= SZ_16K | SZ_32M;
1675 if (id & ID2_PTFS_64K)
1676 size |= SZ_64K | SZ_512M;
1677 }
1678
1679 arm_smmu_ops.pgsize_bitmap &= size;
1680 dev_notice(smmu->dev, "\tSupported page sizes: 0x%08lx\n", size);
1681
1682 if (smmu->features & ARM_SMMU_FEAT_TRANS_S1)
1683 dev_notice(smmu->dev, "\tStage-1: %lu-bit VA -> %lu-bit IPA\n",
1684 smmu->va_size, smmu->ipa_size);
1685
1686 if (smmu->features & ARM_SMMU_FEAT_TRANS_S2)
1687 dev_notice(smmu->dev, "\tStage-2: %lu-bit IPA -> %lu-bit PA\n",
1688 smmu->ipa_size, smmu->pa_size);
1689
1690 return 0;
1691 }
1692
1693 static const struct of_device_id arm_smmu_of_match[] = {
1694 { .compatible = "arm,smmu-v1", .data = (void *)ARM_SMMU_V1 },
1695 { .compatible = "arm,smmu-v2", .data = (void *)ARM_SMMU_V2 },
1696 { .compatible = "arm,mmu-400", .data = (void *)ARM_SMMU_V1 },
1697 { .compatible = "arm,mmu-401", .data = (void *)ARM_SMMU_V1 },
1698 { .compatible = "arm,mmu-500", .data = (void *)ARM_SMMU_V2 },
1699 { },
1700 };
1701 MODULE_DEVICE_TABLE(of, arm_smmu_of_match);
1702
arm_smmu_device_dt_probe(struct platform_device * pdev)1703 static int arm_smmu_device_dt_probe(struct platform_device *pdev)
1704 {
1705 const struct of_device_id *of_id;
1706 struct resource *res;
1707 struct arm_smmu_device *smmu;
1708 struct device *dev = &pdev->dev;
1709 struct rb_node *node;
1710 struct of_phandle_args masterspec;
1711 int num_irqs, i, err;
1712
1713 smmu = devm_kzalloc(dev, sizeof(*smmu), GFP_KERNEL);
1714 if (!smmu) {
1715 dev_err(dev, "failed to allocate arm_smmu_device\n");
1716 return -ENOMEM;
1717 }
1718 smmu->dev = dev;
1719
1720 of_id = of_match_node(arm_smmu_of_match, dev->of_node);
1721 smmu->version = (enum arm_smmu_arch_version)of_id->data;
1722
1723 res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1724 smmu->base = devm_ioremap_resource(dev, res);
1725 if (IS_ERR(smmu->base))
1726 return PTR_ERR(smmu->base);
1727 smmu->size = resource_size(res);
1728
1729 if (of_property_read_u32(dev->of_node, "#global-interrupts",
1730 &smmu->num_global_irqs)) {
1731 dev_err(dev, "missing #global-interrupts property\n");
1732 return -ENODEV;
1733 }
1734
1735 num_irqs = 0;
1736 while ((res = platform_get_resource(pdev, IORESOURCE_IRQ, num_irqs))) {
1737 num_irqs++;
1738 if (num_irqs > smmu->num_global_irqs)
1739 smmu->num_context_irqs++;
1740 }
1741
1742 if (!smmu->num_context_irqs) {
1743 dev_err(dev, "found %d interrupts but expected at least %d\n",
1744 num_irqs, smmu->num_global_irqs + 1);
1745 return -ENODEV;
1746 }
1747
1748 smmu->irqs = devm_kzalloc(dev, sizeof(*smmu->irqs) * num_irqs,
1749 GFP_KERNEL);
1750 if (!smmu->irqs) {
1751 dev_err(dev, "failed to allocate %d irqs\n", num_irqs);
1752 return -ENOMEM;
1753 }
1754
1755 for (i = 0; i < num_irqs; ++i) {
1756 int irq = platform_get_irq(pdev, i);
1757
1758 if (irq < 0) {
1759 dev_err(dev, "failed to get irq index %d\n", i);
1760 return -ENODEV;
1761 }
1762 smmu->irqs[i] = irq;
1763 }
1764
1765 err = arm_smmu_device_cfg_probe(smmu);
1766 if (err)
1767 return err;
1768
1769 i = 0;
1770 smmu->masters = RB_ROOT;
1771 while (!of_parse_phandle_with_args(dev->of_node, "mmu-masters",
1772 "#stream-id-cells", i,
1773 &masterspec)) {
1774 err = register_smmu_master(smmu, dev, &masterspec);
1775 if (err) {
1776 dev_err(dev, "failed to add master %s\n",
1777 masterspec.np->name);
1778 goto out_put_masters;
1779 }
1780
1781 i++;
1782 }
1783 dev_notice(dev, "registered %d master devices\n", i);
1784
1785 parse_driver_options(smmu);
1786
1787 if (smmu->version > ARM_SMMU_V1 &&
1788 smmu->num_context_banks != smmu->num_context_irqs) {
1789 dev_err(dev,
1790 "found only %d context interrupt(s) but %d required\n",
1791 smmu->num_context_irqs, smmu->num_context_banks);
1792 err = -ENODEV;
1793 goto out_put_masters;
1794 }
1795
1796 for (i = 0; i < smmu->num_global_irqs; ++i) {
1797 err = request_irq(smmu->irqs[i],
1798 arm_smmu_global_fault,
1799 IRQF_SHARED,
1800 "arm-smmu global fault",
1801 smmu);
1802 if (err) {
1803 dev_err(dev, "failed to request global IRQ %d (%u)\n",
1804 i, smmu->irqs[i]);
1805 goto out_free_irqs;
1806 }
1807 }
1808
1809 INIT_LIST_HEAD(&smmu->list);
1810 spin_lock(&arm_smmu_devices_lock);
1811 list_add(&smmu->list, &arm_smmu_devices);
1812 spin_unlock(&arm_smmu_devices_lock);
1813
1814 arm_smmu_device_reset(smmu);
1815 return 0;
1816
1817 out_free_irqs:
1818 while (i--)
1819 free_irq(smmu->irqs[i], smmu);
1820
1821 out_put_masters:
1822 for (node = rb_first(&smmu->masters); node; node = rb_next(node)) {
1823 struct arm_smmu_master *master
1824 = container_of(node, struct arm_smmu_master, node);
1825 of_node_put(master->of_node);
1826 }
1827
1828 return err;
1829 }
1830
arm_smmu_device_remove(struct platform_device * pdev)1831 static int arm_smmu_device_remove(struct platform_device *pdev)
1832 {
1833 int i;
1834 struct device *dev = &pdev->dev;
1835 struct arm_smmu_device *curr, *smmu = NULL;
1836 struct rb_node *node;
1837
1838 spin_lock(&arm_smmu_devices_lock);
1839 list_for_each_entry(curr, &arm_smmu_devices, list) {
1840 if (curr->dev == dev) {
1841 smmu = curr;
1842 list_del(&smmu->list);
1843 break;
1844 }
1845 }
1846 spin_unlock(&arm_smmu_devices_lock);
1847
1848 if (!smmu)
1849 return -ENODEV;
1850
1851 for (node = rb_first(&smmu->masters); node; node = rb_next(node)) {
1852 struct arm_smmu_master *master
1853 = container_of(node, struct arm_smmu_master, node);
1854 of_node_put(master->of_node);
1855 }
1856
1857 if (!bitmap_empty(smmu->context_map, ARM_SMMU_MAX_CBS))
1858 dev_err(dev, "removing device with active domains!\n");
1859
1860 for (i = 0; i < smmu->num_global_irqs; ++i)
1861 free_irq(smmu->irqs[i], smmu);
1862
1863 /* Turn the thing off */
1864 writel(sCR0_CLIENTPD, ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sCR0);
1865 return 0;
1866 }
1867
1868 static struct platform_driver arm_smmu_driver = {
1869 .driver = {
1870 .name = "arm-smmu",
1871 .of_match_table = of_match_ptr(arm_smmu_of_match),
1872 },
1873 .probe = arm_smmu_device_dt_probe,
1874 .remove = arm_smmu_device_remove,
1875 };
1876
arm_smmu_init(void)1877 static int __init arm_smmu_init(void)
1878 {
1879 struct device_node *np;
1880 int ret;
1881
1882 /*
1883 * Play nice with systems that don't have an ARM SMMU by checking that
1884 * an ARM SMMU exists in the system before proceeding with the driver
1885 * and IOMMU bus operation registration.
1886 */
1887 np = of_find_matching_node(NULL, arm_smmu_of_match);
1888 if (!np)
1889 return 0;
1890
1891 of_node_put(np);
1892
1893 ret = platform_driver_register(&arm_smmu_driver);
1894 if (ret)
1895 return ret;
1896
1897 /* Oh, for a proper bus abstraction */
1898 if (!iommu_present(&platform_bus_type))
1899 bus_set_iommu(&platform_bus_type, &arm_smmu_ops);
1900
1901 #ifdef CONFIG_ARM_AMBA
1902 if (!iommu_present(&amba_bustype))
1903 bus_set_iommu(&amba_bustype, &arm_smmu_ops);
1904 #endif
1905
1906 #ifdef CONFIG_PCI
1907 if (!iommu_present(&pci_bus_type))
1908 bus_set_iommu(&pci_bus_type, &arm_smmu_ops);
1909 #endif
1910
1911 return 0;
1912 }
1913
arm_smmu_exit(void)1914 static void __exit arm_smmu_exit(void)
1915 {
1916 return platform_driver_unregister(&arm_smmu_driver);
1917 }
1918
1919 subsys_initcall(arm_smmu_init);
1920 module_exit(arm_smmu_exit);
1921
1922 MODULE_DESCRIPTION("IOMMU API for ARM architected SMMU implementations");
1923 MODULE_AUTHOR("Will Deacon <will.deacon@arm.com>");
1924 MODULE_LICENSE("GPL v2");
1925