2 * IOMMU API for ARM architected SMMU implementations.
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.
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.
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.
17 * Copyright (C) 2013 ARM Limited
19 * Author: Will Deacon <will.deacon@arm.com>
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
29 #define pr_fmt(fmt) "arm-smmu: " fmt
31 #include <linux/delay.h>
32 #include <linux/dma-mapping.h>
33 #include <linux/err.h>
34 #include <linux/interrupt.h>
36 #include <linux/iommu.h>
37 #include <linux/iopoll.h>
38 #include <linux/module.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>
46 #include <linux/amba/bus.h>
48 #include "io-pgtable.h"
50 /* Maximum number of stream IDs assigned to a single device */
51 #define MAX_MASTER_STREAMIDS MAX_PHANDLE_ARGS
53 /* Maximum number of context banks per SMMU */
54 #define ARM_SMMU_MAX_CBS 128
56 /* Maximum number of mapping groups per SMMU */
57 #define ARM_SMMU_MAX_SMRS 128
59 /* SMMU global address space */
60 #define ARM_SMMU_GR0(smmu) ((smmu)->base)
61 #define ARM_SMMU_GR1(smmu) ((smmu)->base + (1 << (smmu)->pgshift))
64 * SMMU global address space with conditional offset to access secure
65 * aliases of non-secure registers (e.g. nsCR0: 0x400, nsGFSR: 0x448,
68 #define ARM_SMMU_GR0_NS(smmu) \
70 ((smmu->options & ARM_SMMU_OPT_SECURE_CFG_ACCESS) \
74 #define smmu_writeq writeq_relaxed
76 #define smmu_writeq(reg64, addr) \
78 u64 __val = (reg64); \
79 void __iomem *__addr = (addr); \
80 writel_relaxed(__val >> 32, __addr + 4); \
81 writel_relaxed(__val, __addr); \
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
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
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
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
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)
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! */
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
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)
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
189 #define ARM_SMMU_GR1_CBA2R(n) (0x800 + ((n) << 2))
190 #define CBA2R_RW64_32BIT (0 << 0)
191 #define CBA2R_RW64_64BIT (1 << 0)
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))
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
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)
229 #define CB_PAR_F (1 << 0)
231 #define ATSR_ACTIVE (1 << 0)
233 #define RESUME_RETRY (0 << 0)
234 #define RESUME_TERMINATE (1 << 0)
236 #define TTBCR2_SEP_SHIFT 15
237 #define TTBCR2_SEP_UPSTREAM (0x7 << TTBCR2_SEP_SHIFT)
239 #define TTBRn_ASID_SHIFT 48
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)
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)
257 #define FSYNR0_WNR (1 << 4)
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.");
264 enum arm_smmu_arch_version {
269 struct arm_smmu_smr {
275 struct arm_smmu_master_cfg {
277 u16 streamids[MAX_MASTER_STREAMIDS];
278 struct arm_smmu_smr *smrs;
281 struct arm_smmu_master {
282 struct device_node *of_node;
284 struct arm_smmu_master_cfg cfg;
287 struct arm_smmu_device {
292 unsigned long pgshift;
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)
302 #define ARM_SMMU_OPT_SECURE_CFG_ACCESS (1 << 0)
304 enum arm_smmu_arch_version version;
306 u32 num_context_banks;
307 u32 num_s2_context_banks;
308 DECLARE_BITMAP(context_map, ARM_SMMU_MAX_CBS);
311 u32 num_mapping_groups;
312 DECLARE_BITMAP(smr_map, ARM_SMMU_MAX_SMRS);
314 unsigned long va_size;
315 unsigned long ipa_size;
316 unsigned long pa_size;
319 u32 num_context_irqs;
322 struct list_head list;
323 struct rb_root masters;
326 struct arm_smmu_cfg {
331 #define INVALID_IRPTNDX 0xff
333 #define ARM_SMMU_CB_ASID(cfg) ((cfg)->cbndx)
334 #define ARM_SMMU_CB_VMID(cfg) ((cfg)->cbndx + 1)
336 enum arm_smmu_domain_stage {
337 ARM_SMMU_DOMAIN_S1 = 0,
339 ARM_SMMU_DOMAIN_NESTED,
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;
352 static struct iommu_ops arm_smmu_ops;
354 static DEFINE_SPINLOCK(arm_smmu_devices_lock);
355 static LIST_HEAD(arm_smmu_devices);
357 struct arm_smmu_option_prop {
362 static struct arm_smmu_option_prop arm_smmu_options[] = {
363 { ARM_SMMU_OPT_SECURE_CFG_ACCESS, "calxeda,smmu-secure-config-access" },
367 static struct arm_smmu_domain *to_smmu_domain(struct iommu_domain *dom)
369 return container_of(dom, struct arm_smmu_domain, domain);
372 static void parse_driver_options(struct arm_smmu_device *smmu)
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);
383 } while (arm_smmu_options[++i].opt);
386 static struct device_node *dev_get_dev_node(struct device *dev)
388 if (dev_is_pci(dev)) {
389 struct pci_bus *bus = to_pci_dev(dev)->bus;
391 while (!pci_is_root_bus(bus))
393 return bus->bridge->parent->of_node;
399 static struct arm_smmu_master *find_smmu_master(struct arm_smmu_device *smmu,
400 struct device_node *dev_node)
402 struct rb_node *node = smmu->masters.rb_node;
405 struct arm_smmu_master *master;
407 master = container_of(node, struct arm_smmu_master, node);
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;
420 static struct arm_smmu_master_cfg *
421 find_smmu_master_cfg(struct device *dev)
423 struct arm_smmu_master_cfg *cfg = NULL;
424 struct iommu_group *group = iommu_group_get(dev);
427 cfg = iommu_group_get_iommudata(group);
428 iommu_group_put(group);
434 static int insert_smmu_master(struct arm_smmu_device *smmu,
435 struct arm_smmu_master *master)
437 struct rb_node **new, *parent;
439 new = &smmu->masters.rb_node;
442 struct arm_smmu_master *this
443 = container_of(*new, struct arm_smmu_master, node);
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);
454 rb_link_node(&master->node, parent, new);
455 rb_insert_color(&master->node, &smmu->masters);
459 static int register_smmu_master(struct arm_smmu_device *smmu,
461 struct of_phandle_args *masterspec)
464 struct arm_smmu_master *master;
466 master = find_smmu_master(smmu, masterspec->np);
469 "rejecting multiple registrations for master device %s\n",
470 masterspec->np->name);
474 if (masterspec->args_count > MAX_MASTER_STREAMIDS) {
476 "reached maximum number (%d) of stream IDs for master device %s\n",
477 MAX_MASTER_STREAMIDS, masterspec->np->name);
481 master = devm_kzalloc(dev, sizeof(*master), GFP_KERNEL);
485 master->of_node = masterspec->np;
486 master->cfg.num_streamids = masterspec->args_count;
488 for (i = 0; i < master->cfg.num_streamids; ++i) {
489 u16 streamid = masterspec->args[i];
491 if (!(smmu->features & ARM_SMMU_FEAT_STREAM_MATCH) &&
492 (streamid >= smmu->num_mapping_groups)) {
494 "stream ID for master device %s greater than maximum allowed (%d)\n",
495 masterspec->np->name, smmu->num_mapping_groups);
498 master->cfg.streamids[i] = streamid;
500 return insert_smmu_master(smmu, master);
503 static struct arm_smmu_device *find_smmu_for_device(struct device *dev)
505 struct arm_smmu_device *smmu;
506 struct arm_smmu_master *master = NULL;
507 struct device_node *dev_node = dev_get_dev_node(dev);
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);
515 spin_unlock(&arm_smmu_devices_lock);
517 return master ? smmu : NULL;
520 static int __arm_smmu_alloc_bitmap(unsigned long *map, int start, int end)
525 idx = find_next_zero_bit(map, end, start);
528 } while (test_and_set_bit(idx, map));
533 static void __arm_smmu_free_bitmap(unsigned long *map, int idx)
538 /* Wait for any pending TLB invalidations to complete */
539 static void __arm_smmu_tlb_sync(struct arm_smmu_device *smmu)
542 void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
544 writel_relaxed(0, gr0_base + ARM_SMMU_GR0_sTLBGSYNC);
545 while (readl_relaxed(gr0_base + ARM_SMMU_GR0_sTLBGSTATUS)
546 & sTLBGSTATUS_GSACTIVE) {
548 if (++count == TLB_LOOP_TIMEOUT) {
549 dev_err_ratelimited(smmu->dev,
550 "TLB sync timed out -- SMMU may be deadlocked\n");
557 static void arm_smmu_tlb_sync(void *cookie)
559 struct arm_smmu_domain *smmu_domain = cookie;
560 __arm_smmu_tlb_sync(smmu_domain->smmu);
563 static void arm_smmu_tlb_inv_context(void *cookie)
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;
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);
576 base = ARM_SMMU_GR0(smmu);
577 writel_relaxed(ARM_SMMU_CB_VMID(cfg),
578 base + ARM_SMMU_GR0_TLBIVMID);
581 __arm_smmu_tlb_sync(smmu);
584 static void arm_smmu_tlb_inv_range_nosync(unsigned long iova, size_t size,
585 bool leaf, void *cookie)
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;
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;
597 if (!IS_ENABLED(CONFIG_64BIT) || smmu->version == ARM_SMMU_V1) {
599 iova |= ARM_SMMU_CB_ASID(cfg);
600 writel_relaxed(iova, reg);
604 iova |= (u64)ARM_SMMU_CB_ASID(cfg) << 48;
605 writeq_relaxed(iova, reg);
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);
616 reg = ARM_SMMU_GR0(smmu) + ARM_SMMU_GR0_TLBIVMID;
617 writel_relaxed(ARM_SMMU_CB_VMID(cfg), reg);
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,
627 static irqreturn_t arm_smmu_context_fault(int irq, void *dev)
630 u32 fsr, far, fsynr, resume;
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;
638 cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
639 fsr = readl_relaxed(cb_base + ARM_SMMU_CB_FSR);
641 if (!(fsr & FSR_FAULT))
645 dev_err_ratelimited(smmu->dev,
646 "Unexpected context fault (fsr 0x%x)\n",
649 fsynr = readl_relaxed(cb_base + ARM_SMMU_CB_FSYNR0);
650 flags = fsynr & FSYNR0_WNR ? IOMMU_FAULT_WRITE : IOMMU_FAULT_READ;
652 far = readl_relaxed(cb_base + ARM_SMMU_CB_FAR_LO);
655 far = readl_relaxed(cb_base + ARM_SMMU_CB_FAR_HI);
656 iova |= ((unsigned long)far << 32);
659 if (!report_iommu_fault(domain, smmu->dev, iova, flags)) {
661 resume = RESUME_RETRY;
663 dev_err_ratelimited(smmu->dev,
664 "Unhandled context fault: iova=0x%08lx, fsynr=0x%x, cb=%d\n",
665 iova, fsynr, cfg->cbndx);
667 resume = RESUME_TERMINATE;
670 /* Clear the faulting FSR */
671 writel(fsr, cb_base + ARM_SMMU_CB_FSR);
673 /* Retry or terminate any stalled transactions */
675 writel_relaxed(resume, cb_base + ARM_SMMU_CB_RESUME);
680 static irqreturn_t arm_smmu_global_fault(int irq, void *dev)
682 u32 gfsr, gfsynr0, gfsynr1, gfsynr2;
683 struct arm_smmu_device *smmu = dev;
684 void __iomem *gr0_base = ARM_SMMU_GR0_NS(smmu);
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);
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);
700 writel(gfsr, gr0_base + ARM_SMMU_GR0_sGFSR);
704 static void arm_smmu_init_context_bank(struct arm_smmu_domain *smmu_domain,
705 struct io_pgtable_cfg *pgtbl_cfg)
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;
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);
718 if (smmu->version > ARM_SMMU_V1) {
721 * *Must* be initialised before CBAR thanks to VMID16
722 * architectural oversight affected some implementations.
725 reg = CBA2R_RW64_64BIT;
727 reg = CBA2R_RW64_32BIT;
729 writel_relaxed(reg, gr1_base + ARM_SMMU_GR1_CBA2R(cfg->cbndx));
734 if (smmu->version == ARM_SMMU_V1)
735 reg |= cfg->irptndx << CBAR_IRPTNDX_SHIFT;
738 * Use the weakest shareability/memory types, so they are
739 * overridden by the ttbcr/pte.
742 reg |= (CBAR_S1_BPSHCFG_NSH << CBAR_S1_BPSHCFG_SHIFT) |
743 (CBAR_S1_MEMATTR_WB << CBAR_S1_MEMATTR_SHIFT);
745 reg |= ARM_SMMU_CB_VMID(cfg) << CBAR_VMID_SHIFT;
747 writel_relaxed(reg, gr1_base + ARM_SMMU_GR1_CBAR(cfg->cbndx));
751 reg64 = pgtbl_cfg->arm_lpae_s1_cfg.ttbr[0];
753 reg64 |= ((u64)ARM_SMMU_CB_ASID(cfg)) << TTBRn_ASID_SHIFT;
754 smmu_writeq(reg64, cb_base + ARM_SMMU_CB_TTBR0);
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);
760 reg64 = pgtbl_cfg->arm_lpae_s2_cfg.vttbr;
761 smmu_writeq(reg64, cb_base + ARM_SMMU_CB_TTBR0);
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);
774 reg = pgtbl_cfg->arm_lpae_s2_cfg.vtcr;
775 writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBCR);
778 /* MAIRs (stage-1 only) */
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);
787 reg = SCTLR_CFCFG | SCTLR_CFIE | SCTLR_CFRE | SCTLR_M | SCTLR_EAE_SBOP;
789 reg |= SCTLR_S1_ASIDPNE;
793 writel_relaxed(reg, cb_base + ARM_SMMU_CB_SCTLR);
796 static int arm_smmu_init_domain_context(struct iommu_domain *domain,
797 struct arm_smmu_device *smmu)
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;
807 mutex_lock(&smmu_domain->init_mutex);
808 if (smmu_domain->smmu)
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
817 * Requested Supported Actual
827 * Note that you can't actually request stage-2 mappings.
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;
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;
839 oas = smmu->ipa_size;
840 if (IS_ENABLED(CONFIG_64BIT))
841 fmt = ARM_64_LPAE_S1;
843 fmt = ARM_32_LPAE_S1;
845 case ARM_SMMU_DOMAIN_NESTED:
847 * We will likely want to change this if/when KVM gets
850 case ARM_SMMU_DOMAIN_S2:
851 cfg->cbar = CBAR_TYPE_S2_TRANS;
853 ias = smmu->ipa_size;
855 if (IS_ENABLED(CONFIG_64BIT))
856 fmt = ARM_64_LPAE_S2;
858 fmt = ARM_32_LPAE_S2;
865 ret = __arm_smmu_alloc_bitmap(smmu->context_map, start,
866 smmu->num_context_banks);
867 if (IS_ERR_VALUE(ret))
871 if (smmu->version == ARM_SMMU_V1) {
872 cfg->irptndx = atomic_inc_return(&smmu->irptndx);
873 cfg->irptndx %= smmu->num_context_irqs;
875 cfg->irptndx = cfg->cbndx;
878 pgtbl_cfg = (struct io_pgtable_cfg) {
879 .pgsize_bitmap = arm_smmu_ops.pgsize_bitmap,
882 .tlb = &arm_smmu_gather_ops,
883 .iommu_dev = smmu->dev,
886 smmu_domain->smmu = smmu;
887 pgtbl_ops = alloc_io_pgtable_ops(fmt, &pgtbl_cfg, smmu_domain);
893 /* Update our support page sizes to reflect the page table format */
894 arm_smmu_ops.pgsize_bitmap = pgtbl_cfg.pgsize_bitmap;
896 /* Initialise the context bank with our page table cfg */
897 arm_smmu_init_context_bank(smmu_domain, &pgtbl_cfg);
900 * Request context fault interrupt. Do this last to avoid the
901 * handler seeing a half-initialised domain state.
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",
909 cfg->irptndx = INVALID_IRPTNDX;
912 mutex_unlock(&smmu_domain->init_mutex);
914 /* Publish page table ops for map/unmap */
915 smmu_domain->pgtbl_ops = pgtbl_ops;
919 smmu_domain->smmu = NULL;
921 mutex_unlock(&smmu_domain->init_mutex);
925 static void arm_smmu_destroy_domain_context(struct iommu_domain *domain)
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;
937 * Disable the context bank and free the page tables before freeing
940 cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
941 writel_relaxed(0, cb_base + ARM_SMMU_CB_SCTLR);
943 if (cfg->irptndx != INVALID_IRPTNDX) {
944 irq = smmu->irqs[smmu->num_global_irqs + cfg->irptndx];
945 free_irq(irq, domain);
948 free_io_pgtable_ops(smmu_domain->pgtbl_ops);
949 __arm_smmu_free_bitmap(smmu->context_map, cfg->cbndx);
952 static struct iommu_domain *arm_smmu_domain_alloc(unsigned type)
954 struct arm_smmu_domain *smmu_domain;
956 if (type != IOMMU_DOMAIN_UNMANAGED)
959 * Allocate the domain and initialise some of its data structures.
960 * We can't really do anything meaningful until we've added a
963 smmu_domain = kzalloc(sizeof(*smmu_domain), GFP_KERNEL);
967 mutex_init(&smmu_domain->init_mutex);
968 spin_lock_init(&smmu_domain->pgtbl_lock);
970 return &smmu_domain->domain;
973 static void arm_smmu_domain_free(struct iommu_domain *domain)
975 struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
978 * Free the domain resources. We assume that all devices have
979 * already been detached.
981 arm_smmu_destroy_domain_context(domain);
985 static int arm_smmu_master_configure_smrs(struct arm_smmu_device *smmu,
986 struct arm_smmu_master_cfg *cfg)
989 struct arm_smmu_smr *smrs;
990 void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
992 if (!(smmu->features & ARM_SMMU_FEAT_STREAM_MATCH))
998 smrs = kmalloc_array(cfg->num_streamids, sizeof(*smrs), GFP_KERNEL);
1000 dev_err(smmu->dev, "failed to allocate %d SMRs\n",
1001 cfg->num_streamids);
1005 /* Allocate the SMRs on the SMMU */
1006 for (i = 0; i < cfg->num_streamids; ++i) {
1007 int idx = __arm_smmu_alloc_bitmap(smmu->smr_map, 0,
1008 smmu->num_mapping_groups);
1009 if (IS_ERR_VALUE(idx)) {
1010 dev_err(smmu->dev, "failed to allocate free SMR\n");
1014 smrs[i] = (struct arm_smmu_smr) {
1016 .mask = 0, /* We don't currently share SMRs */
1017 .id = cfg->streamids[i],
1021 /* It worked! Now, poke the actual hardware */
1022 for (i = 0; i < cfg->num_streamids; ++i) {
1023 u32 reg = SMR_VALID | smrs[i].id << SMR_ID_SHIFT |
1024 smrs[i].mask << SMR_MASK_SHIFT;
1025 writel_relaxed(reg, gr0_base + ARM_SMMU_GR0_SMR(smrs[i].idx));
1033 __arm_smmu_free_bitmap(smmu->smr_map, smrs[i].idx);
1038 static void arm_smmu_master_free_smrs(struct arm_smmu_device *smmu,
1039 struct arm_smmu_master_cfg *cfg)
1042 void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1043 struct arm_smmu_smr *smrs = cfg->smrs;
1048 /* Invalidate the SMRs before freeing back to the allocator */
1049 for (i = 0; i < cfg->num_streamids; ++i) {
1050 u8 idx = smrs[i].idx;
1052 writel_relaxed(~SMR_VALID, gr0_base + ARM_SMMU_GR0_SMR(idx));
1053 __arm_smmu_free_bitmap(smmu->smr_map, idx);
1060 static int arm_smmu_domain_add_master(struct arm_smmu_domain *smmu_domain,
1061 struct arm_smmu_master_cfg *cfg)
1064 struct arm_smmu_device *smmu = smmu_domain->smmu;
1065 void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1067 /* Devices in an IOMMU group may already be configured */
1068 ret = arm_smmu_master_configure_smrs(smmu, cfg);
1070 return ret == -EEXIST ? 0 : ret;
1072 for (i = 0; i < cfg->num_streamids; ++i) {
1075 idx = cfg->smrs ? cfg->smrs[i].idx : cfg->streamids[i];
1076 s2cr = S2CR_TYPE_TRANS |
1077 (smmu_domain->cfg.cbndx << S2CR_CBNDX_SHIFT);
1078 writel_relaxed(s2cr, gr0_base + ARM_SMMU_GR0_S2CR(idx));
1084 static void arm_smmu_domain_remove_master(struct arm_smmu_domain *smmu_domain,
1085 struct arm_smmu_master_cfg *cfg)
1088 struct arm_smmu_device *smmu = smmu_domain->smmu;
1089 void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1091 /* An IOMMU group is torn down by the first device to be removed */
1092 if ((smmu->features & ARM_SMMU_FEAT_STREAM_MATCH) && !cfg->smrs)
1096 * We *must* clear the S2CR first, because freeing the SMR means
1097 * that it can be re-allocated immediately.
1099 for (i = 0; i < cfg->num_streamids; ++i) {
1100 u32 idx = cfg->smrs ? cfg->smrs[i].idx : cfg->streamids[i];
1102 writel_relaxed(S2CR_TYPE_BYPASS,
1103 gr0_base + ARM_SMMU_GR0_S2CR(idx));
1106 arm_smmu_master_free_smrs(smmu, cfg);
1109 static int arm_smmu_attach_dev(struct iommu_domain *domain, struct device *dev)
1112 struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
1113 struct arm_smmu_device *smmu;
1114 struct arm_smmu_master_cfg *cfg;
1116 smmu = find_smmu_for_device(dev);
1118 dev_err(dev, "cannot attach to SMMU, is it on the same bus?\n");
1122 if (dev->archdata.iommu) {
1123 dev_err(dev, "already attached to IOMMU domain\n");
1127 /* Ensure that the domain is finalised */
1128 ret = arm_smmu_init_domain_context(domain, smmu);
1129 if (IS_ERR_VALUE(ret))
1133 * Sanity check the domain. We don't support domains across
1136 if (smmu_domain->smmu != smmu) {
1138 "cannot attach to SMMU %s whilst already attached to domain on SMMU %s\n",
1139 dev_name(smmu_domain->smmu->dev), dev_name(smmu->dev));
1143 /* Looks ok, so add the device to the domain */
1144 cfg = find_smmu_master_cfg(dev);
1148 ret = arm_smmu_domain_add_master(smmu_domain, cfg);
1150 dev->archdata.iommu = domain;
1154 static void arm_smmu_detach_dev(struct iommu_domain *domain, struct device *dev)
1156 struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
1157 struct arm_smmu_master_cfg *cfg;
1159 cfg = find_smmu_master_cfg(dev);
1163 dev->archdata.iommu = NULL;
1164 arm_smmu_domain_remove_master(smmu_domain, cfg);
1167 static int arm_smmu_map(struct iommu_domain *domain, unsigned long iova,
1168 phys_addr_t paddr, size_t size, int prot)
1171 unsigned long flags;
1172 struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
1173 struct io_pgtable_ops *ops= smmu_domain->pgtbl_ops;
1178 spin_lock_irqsave(&smmu_domain->pgtbl_lock, flags);
1179 ret = ops->map(ops, iova, paddr, size, prot);
1180 spin_unlock_irqrestore(&smmu_domain->pgtbl_lock, flags);
1184 static size_t arm_smmu_unmap(struct iommu_domain *domain, unsigned long iova,
1188 unsigned long flags;
1189 struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
1190 struct io_pgtable_ops *ops= smmu_domain->pgtbl_ops;
1195 spin_lock_irqsave(&smmu_domain->pgtbl_lock, flags);
1196 ret = ops->unmap(ops, iova, size);
1197 spin_unlock_irqrestore(&smmu_domain->pgtbl_lock, flags);
1201 static phys_addr_t arm_smmu_iova_to_phys_hard(struct iommu_domain *domain,
1204 struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
1205 struct arm_smmu_device *smmu = smmu_domain->smmu;
1206 struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
1207 struct io_pgtable_ops *ops= smmu_domain->pgtbl_ops;
1208 struct device *dev = smmu->dev;
1209 void __iomem *cb_base;
1214 cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
1216 /* ATS1 registers can only be written atomically */
1217 va = iova & ~0xfffUL;
1218 if (smmu->version == ARM_SMMU_V2)
1219 smmu_writeq(va, cb_base + ARM_SMMU_CB_ATS1PR);
1221 writel_relaxed(va, cb_base + ARM_SMMU_CB_ATS1PR);
1223 if (readl_poll_timeout_atomic(cb_base + ARM_SMMU_CB_ATSR, tmp,
1224 !(tmp & ATSR_ACTIVE), 5, 50)) {
1226 "iova to phys timed out on %pad. Falling back to software table walk.\n",
1228 return ops->iova_to_phys(ops, iova);
1231 phys = readl_relaxed(cb_base + ARM_SMMU_CB_PAR_LO);
1232 phys |= ((u64)readl_relaxed(cb_base + ARM_SMMU_CB_PAR_HI)) << 32;
1234 if (phys & CB_PAR_F) {
1235 dev_err(dev, "translation fault!\n");
1236 dev_err(dev, "PAR = 0x%llx\n", phys);
1240 return (phys & GENMASK_ULL(39, 12)) | (iova & 0xfff);
1243 static phys_addr_t arm_smmu_iova_to_phys(struct iommu_domain *domain,
1247 unsigned long flags;
1248 struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
1249 struct io_pgtable_ops *ops= smmu_domain->pgtbl_ops;
1254 spin_lock_irqsave(&smmu_domain->pgtbl_lock, flags);
1255 if (smmu_domain->smmu->features & ARM_SMMU_FEAT_TRANS_OPS &&
1256 smmu_domain->stage == ARM_SMMU_DOMAIN_S1) {
1257 ret = arm_smmu_iova_to_phys_hard(domain, iova);
1259 ret = ops->iova_to_phys(ops, iova);
1262 spin_unlock_irqrestore(&smmu_domain->pgtbl_lock, flags);
1267 static bool arm_smmu_capable(enum iommu_cap cap)
1270 case IOMMU_CAP_CACHE_COHERENCY:
1272 * Return true here as the SMMU can always send out coherent
1276 case IOMMU_CAP_INTR_REMAP:
1277 return true; /* MSIs are just memory writes */
1278 case IOMMU_CAP_NOEXEC:
1285 static int __arm_smmu_get_pci_sid(struct pci_dev *pdev, u16 alias, void *data)
1287 *((u16 *)data) = alias;
1288 return 0; /* Continue walking */
1291 static void __arm_smmu_release_pci_iommudata(void *data)
1296 static int arm_smmu_init_pci_device(struct pci_dev *pdev,
1297 struct iommu_group *group)
1299 struct arm_smmu_master_cfg *cfg;
1303 cfg = iommu_group_get_iommudata(group);
1305 cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1309 iommu_group_set_iommudata(group, cfg,
1310 __arm_smmu_release_pci_iommudata);
1313 if (cfg->num_streamids >= MAX_MASTER_STREAMIDS)
1317 * Assume Stream ID == Requester ID for now.
1318 * We need a way to describe the ID mappings in FDT.
1320 pci_for_each_dma_alias(pdev, __arm_smmu_get_pci_sid, &sid);
1321 for (i = 0; i < cfg->num_streamids; ++i)
1322 if (cfg->streamids[i] == sid)
1325 /* Avoid duplicate SIDs, as this can lead to SMR conflicts */
1326 if (i == cfg->num_streamids)
1327 cfg->streamids[cfg->num_streamids++] = sid;
1332 static int arm_smmu_init_platform_device(struct device *dev,
1333 struct iommu_group *group)
1335 struct arm_smmu_device *smmu = find_smmu_for_device(dev);
1336 struct arm_smmu_master *master;
1341 master = find_smmu_master(smmu, dev->of_node);
1345 iommu_group_set_iommudata(group, &master->cfg, NULL);
1350 static int arm_smmu_add_device(struct device *dev)
1352 struct iommu_group *group;
1354 group = iommu_group_get_for_dev(dev);
1356 return PTR_ERR(group);
1358 iommu_group_put(group);
1362 static void arm_smmu_remove_device(struct device *dev)
1364 iommu_group_remove_device(dev);
1367 static struct iommu_group *arm_smmu_device_group(struct device *dev)
1369 struct iommu_group *group;
1372 if (dev_is_pci(dev))
1373 group = pci_device_group(dev);
1375 group = generic_device_group(dev);
1380 if (dev_is_pci(dev))
1381 ret = arm_smmu_init_pci_device(to_pci_dev(dev), group);
1383 ret = arm_smmu_init_platform_device(dev, group);
1386 iommu_group_put(group);
1387 group = ERR_PTR(ret);
1393 static int arm_smmu_domain_get_attr(struct iommu_domain *domain,
1394 enum iommu_attr attr, void *data)
1396 struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
1399 case DOMAIN_ATTR_NESTING:
1400 *(int *)data = (smmu_domain->stage == ARM_SMMU_DOMAIN_NESTED);
1407 static int arm_smmu_domain_set_attr(struct iommu_domain *domain,
1408 enum iommu_attr attr, void *data)
1411 struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
1413 mutex_lock(&smmu_domain->init_mutex);
1416 case DOMAIN_ATTR_NESTING:
1417 if (smmu_domain->smmu) {
1423 smmu_domain->stage = ARM_SMMU_DOMAIN_NESTED;
1425 smmu_domain->stage = ARM_SMMU_DOMAIN_S1;
1433 mutex_unlock(&smmu_domain->init_mutex);
1437 static struct iommu_ops arm_smmu_ops = {
1438 .capable = arm_smmu_capable,
1439 .domain_alloc = arm_smmu_domain_alloc,
1440 .domain_free = arm_smmu_domain_free,
1441 .attach_dev = arm_smmu_attach_dev,
1442 .detach_dev = arm_smmu_detach_dev,
1443 .map = arm_smmu_map,
1444 .unmap = arm_smmu_unmap,
1445 .map_sg = default_iommu_map_sg,
1446 .iova_to_phys = arm_smmu_iova_to_phys,
1447 .add_device = arm_smmu_add_device,
1448 .remove_device = arm_smmu_remove_device,
1449 .device_group = arm_smmu_device_group,
1450 .domain_get_attr = arm_smmu_domain_get_attr,
1451 .domain_set_attr = arm_smmu_domain_set_attr,
1452 .pgsize_bitmap = -1UL, /* Restricted during device attach */
1455 static void arm_smmu_device_reset(struct arm_smmu_device *smmu)
1457 void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1458 void __iomem *cb_base;
1462 /* clear global FSR */
1463 reg = readl_relaxed(ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sGFSR);
1464 writel(reg, ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sGFSR);
1466 /* Mark all SMRn as invalid and all S2CRn as bypass */
1467 for (i = 0; i < smmu->num_mapping_groups; ++i) {
1468 writel_relaxed(0, gr0_base + ARM_SMMU_GR0_SMR(i));
1469 writel_relaxed(S2CR_TYPE_BYPASS,
1470 gr0_base + ARM_SMMU_GR0_S2CR(i));
1473 /* Make sure all context banks are disabled and clear CB_FSR */
1474 for (i = 0; i < smmu->num_context_banks; ++i) {
1475 cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, i);
1476 writel_relaxed(0, cb_base + ARM_SMMU_CB_SCTLR);
1477 writel_relaxed(FSR_FAULT, cb_base + ARM_SMMU_CB_FSR);
1480 /* Invalidate the TLB, just in case */
1481 writel_relaxed(0, gr0_base + ARM_SMMU_GR0_TLBIALLH);
1482 writel_relaxed(0, gr0_base + ARM_SMMU_GR0_TLBIALLNSNH);
1484 reg = readl_relaxed(ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sCR0);
1486 /* Enable fault reporting */
1487 reg |= (sCR0_GFRE | sCR0_GFIE | sCR0_GCFGFRE | sCR0_GCFGFIE);
1489 /* Disable TLB broadcasting. */
1490 reg |= (sCR0_VMIDPNE | sCR0_PTM);
1492 /* Enable client access, but bypass when no mapping is found */
1493 reg &= ~(sCR0_CLIENTPD | sCR0_USFCFG);
1495 /* Disable forced broadcasting */
1498 /* Don't upgrade barriers */
1499 reg &= ~(sCR0_BSU_MASK << sCR0_BSU_SHIFT);
1501 /* Push the button */
1502 __arm_smmu_tlb_sync(smmu);
1503 writel(reg, ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sCR0);
1506 static int arm_smmu_id_size_to_bits(int size)
1525 static int arm_smmu_device_cfg_probe(struct arm_smmu_device *smmu)
1528 void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1530 bool cttw_dt, cttw_reg;
1532 dev_notice(smmu->dev, "probing hardware configuration...\n");
1533 dev_notice(smmu->dev, "SMMUv%d with:\n", smmu->version);
1536 id = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID0);
1538 /* Restrict available stages based on module parameter */
1539 if (force_stage == 1)
1540 id &= ~(ID0_S2TS | ID0_NTS);
1541 else if (force_stage == 2)
1542 id &= ~(ID0_S1TS | ID0_NTS);
1544 if (id & ID0_S1TS) {
1545 smmu->features |= ARM_SMMU_FEAT_TRANS_S1;
1546 dev_notice(smmu->dev, "\tstage 1 translation\n");
1549 if (id & ID0_S2TS) {
1550 smmu->features |= ARM_SMMU_FEAT_TRANS_S2;
1551 dev_notice(smmu->dev, "\tstage 2 translation\n");
1555 smmu->features |= ARM_SMMU_FEAT_TRANS_NESTED;
1556 dev_notice(smmu->dev, "\tnested translation\n");
1559 if (!(smmu->features &
1560 (ARM_SMMU_FEAT_TRANS_S1 | ARM_SMMU_FEAT_TRANS_S2))) {
1561 dev_err(smmu->dev, "\tno translation support!\n");
1565 if ((id & ID0_S1TS) && ((smmu->version == 1) || !(id & ID0_ATOSNS))) {
1566 smmu->features |= ARM_SMMU_FEAT_TRANS_OPS;
1567 dev_notice(smmu->dev, "\taddress translation ops\n");
1571 * In order for DMA API calls to work properly, we must defer to what
1572 * the DT says about coherency, regardless of what the hardware claims.
1573 * Fortunately, this also opens up a workaround for systems where the
1574 * ID register value has ended up configured incorrectly.
1576 cttw_dt = of_dma_is_coherent(smmu->dev->of_node);
1577 cttw_reg = !!(id & ID0_CTTW);
1579 smmu->features |= ARM_SMMU_FEAT_COHERENT_WALK;
1580 if (cttw_dt || cttw_reg)
1581 dev_notice(smmu->dev, "\t%scoherent table walk\n",
1582 cttw_dt ? "" : "non-");
1583 if (cttw_dt != cttw_reg)
1584 dev_notice(smmu->dev,
1585 "\t(IDR0.CTTW overridden by dma-coherent property)\n");
1590 smmu->features |= ARM_SMMU_FEAT_STREAM_MATCH;
1591 smmu->num_mapping_groups = (id >> ID0_NUMSMRG_SHIFT) &
1593 if (smmu->num_mapping_groups == 0) {
1595 "stream-matching supported, but no SMRs present!\n");
1599 smr = SMR_MASK_MASK << SMR_MASK_SHIFT;
1600 smr |= (SMR_ID_MASK << SMR_ID_SHIFT);
1601 writel_relaxed(smr, gr0_base + ARM_SMMU_GR0_SMR(0));
1602 smr = readl_relaxed(gr0_base + ARM_SMMU_GR0_SMR(0));
1604 mask = (smr >> SMR_MASK_SHIFT) & SMR_MASK_MASK;
1605 sid = (smr >> SMR_ID_SHIFT) & SMR_ID_MASK;
1606 if ((mask & sid) != sid) {
1608 "SMR mask bits (0x%x) insufficient for ID field (0x%x)\n",
1613 dev_notice(smmu->dev,
1614 "\tstream matching with %u register groups, mask 0x%x",
1615 smmu->num_mapping_groups, mask);
1617 smmu->num_mapping_groups = (id >> ID0_NUMSIDB_SHIFT) &
1622 id = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID1);
1623 smmu->pgshift = (id & ID1_PAGESIZE) ? 16 : 12;
1625 /* Check for size mismatch of SMMU address space from mapped region */
1626 size = 1 << (((id >> ID1_NUMPAGENDXB_SHIFT) & ID1_NUMPAGENDXB_MASK) + 1);
1627 size *= 2 << smmu->pgshift;
1628 if (smmu->size != size)
1630 "SMMU address space size (0x%lx) differs from mapped region size (0x%lx)!\n",
1633 smmu->num_s2_context_banks = (id >> ID1_NUMS2CB_SHIFT) & ID1_NUMS2CB_MASK;
1634 smmu->num_context_banks = (id >> ID1_NUMCB_SHIFT) & ID1_NUMCB_MASK;
1635 if (smmu->num_s2_context_banks > smmu->num_context_banks) {
1636 dev_err(smmu->dev, "impossible number of S2 context banks!\n");
1639 dev_notice(smmu->dev, "\t%u context banks (%u stage-2 only)\n",
1640 smmu->num_context_banks, smmu->num_s2_context_banks);
1643 id = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID2);
1644 size = arm_smmu_id_size_to_bits((id >> ID2_IAS_SHIFT) & ID2_IAS_MASK);
1645 smmu->ipa_size = size;
1647 /* The output mask is also applied for bypass */
1648 size = arm_smmu_id_size_to_bits((id >> ID2_OAS_SHIFT) & ID2_OAS_MASK);
1649 smmu->pa_size = size;
1652 * What the page table walker can address actually depends on which
1653 * descriptor format is in use, but since a) we don't know that yet,
1654 * and b) it can vary per context bank, this will have to do...
1656 if (dma_set_mask_and_coherent(smmu->dev, DMA_BIT_MASK(size)))
1658 "failed to set DMA mask for table walker\n");
1660 if (smmu->version == ARM_SMMU_V1) {
1661 smmu->va_size = smmu->ipa_size;
1662 size = SZ_4K | SZ_2M | SZ_1G;
1664 size = (id >> ID2_UBS_SHIFT) & ID2_UBS_MASK;
1665 smmu->va_size = arm_smmu_id_size_to_bits(size);
1666 #ifndef CONFIG_64BIT
1667 smmu->va_size = min(32UL, smmu->va_size);
1670 if (id & ID2_PTFS_4K)
1671 size |= SZ_4K | SZ_2M | SZ_1G;
1672 if (id & ID2_PTFS_16K)
1673 size |= SZ_16K | SZ_32M;
1674 if (id & ID2_PTFS_64K)
1675 size |= SZ_64K | SZ_512M;
1678 arm_smmu_ops.pgsize_bitmap &= size;
1679 dev_notice(smmu->dev, "\tSupported page sizes: 0x%08lx\n", size);
1681 if (smmu->features & ARM_SMMU_FEAT_TRANS_S1)
1682 dev_notice(smmu->dev, "\tStage-1: %lu-bit VA -> %lu-bit IPA\n",
1683 smmu->va_size, smmu->ipa_size);
1685 if (smmu->features & ARM_SMMU_FEAT_TRANS_S2)
1686 dev_notice(smmu->dev, "\tStage-2: %lu-bit IPA -> %lu-bit PA\n",
1687 smmu->ipa_size, smmu->pa_size);
1692 static const struct of_device_id arm_smmu_of_match[] = {
1693 { .compatible = "arm,smmu-v1", .data = (void *)ARM_SMMU_V1 },
1694 { .compatible = "arm,smmu-v2", .data = (void *)ARM_SMMU_V2 },
1695 { .compatible = "arm,mmu-400", .data = (void *)ARM_SMMU_V1 },
1696 { .compatible = "arm,mmu-401", .data = (void *)ARM_SMMU_V1 },
1697 { .compatible = "arm,mmu-500", .data = (void *)ARM_SMMU_V2 },
1700 MODULE_DEVICE_TABLE(of, arm_smmu_of_match);
1702 static int arm_smmu_device_dt_probe(struct platform_device *pdev)
1704 const struct of_device_id *of_id;
1705 struct resource *res;
1706 struct arm_smmu_device *smmu;
1707 struct device *dev = &pdev->dev;
1708 struct rb_node *node;
1709 struct of_phandle_args masterspec;
1710 int num_irqs, i, err;
1712 smmu = devm_kzalloc(dev, sizeof(*smmu), GFP_KERNEL);
1714 dev_err(dev, "failed to allocate arm_smmu_device\n");
1719 of_id = of_match_node(arm_smmu_of_match, dev->of_node);
1720 smmu->version = (enum arm_smmu_arch_version)of_id->data;
1722 res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1723 smmu->base = devm_ioremap_resource(dev, res);
1724 if (IS_ERR(smmu->base))
1725 return PTR_ERR(smmu->base);
1726 smmu->size = resource_size(res);
1728 if (of_property_read_u32(dev->of_node, "#global-interrupts",
1729 &smmu->num_global_irqs)) {
1730 dev_err(dev, "missing #global-interrupts property\n");
1735 while ((res = platform_get_resource(pdev, IORESOURCE_IRQ, num_irqs))) {
1737 if (num_irqs > smmu->num_global_irqs)
1738 smmu->num_context_irqs++;
1741 if (!smmu->num_context_irqs) {
1742 dev_err(dev, "found %d interrupts but expected at least %d\n",
1743 num_irqs, smmu->num_global_irqs + 1);
1747 smmu->irqs = devm_kzalloc(dev, sizeof(*smmu->irqs) * num_irqs,
1750 dev_err(dev, "failed to allocate %d irqs\n", num_irqs);
1754 for (i = 0; i < num_irqs; ++i) {
1755 int irq = platform_get_irq(pdev, i);
1758 dev_err(dev, "failed to get irq index %d\n", i);
1761 smmu->irqs[i] = irq;
1764 err = arm_smmu_device_cfg_probe(smmu);
1769 smmu->masters = RB_ROOT;
1770 while (!of_parse_phandle_with_args(dev->of_node, "mmu-masters",
1771 "#stream-id-cells", i,
1773 err = register_smmu_master(smmu, dev, &masterspec);
1775 dev_err(dev, "failed to add master %s\n",
1776 masterspec.np->name);
1777 goto out_put_masters;
1782 dev_notice(dev, "registered %d master devices\n", i);
1784 parse_driver_options(smmu);
1786 if (smmu->version > ARM_SMMU_V1 &&
1787 smmu->num_context_banks != smmu->num_context_irqs) {
1789 "found only %d context interrupt(s) but %d required\n",
1790 smmu->num_context_irqs, smmu->num_context_banks);
1792 goto out_put_masters;
1795 for (i = 0; i < smmu->num_global_irqs; ++i) {
1796 err = request_irq(smmu->irqs[i],
1797 arm_smmu_global_fault,
1799 "arm-smmu global fault",
1802 dev_err(dev, "failed to request global IRQ %d (%u)\n",
1808 INIT_LIST_HEAD(&smmu->list);
1809 spin_lock(&arm_smmu_devices_lock);
1810 list_add(&smmu->list, &arm_smmu_devices);
1811 spin_unlock(&arm_smmu_devices_lock);
1813 arm_smmu_device_reset(smmu);
1818 free_irq(smmu->irqs[i], smmu);
1821 for (node = rb_first(&smmu->masters); node; node = rb_next(node)) {
1822 struct arm_smmu_master *master
1823 = container_of(node, struct arm_smmu_master, node);
1824 of_node_put(master->of_node);
1830 static int arm_smmu_device_remove(struct platform_device *pdev)
1833 struct device *dev = &pdev->dev;
1834 struct arm_smmu_device *curr, *smmu = NULL;
1835 struct rb_node *node;
1837 spin_lock(&arm_smmu_devices_lock);
1838 list_for_each_entry(curr, &arm_smmu_devices, list) {
1839 if (curr->dev == dev) {
1841 list_del(&smmu->list);
1845 spin_unlock(&arm_smmu_devices_lock);
1850 for (node = rb_first(&smmu->masters); node; node = rb_next(node)) {
1851 struct arm_smmu_master *master
1852 = container_of(node, struct arm_smmu_master, node);
1853 of_node_put(master->of_node);
1856 if (!bitmap_empty(smmu->context_map, ARM_SMMU_MAX_CBS))
1857 dev_err(dev, "removing device with active domains!\n");
1859 for (i = 0; i < smmu->num_global_irqs; ++i)
1860 free_irq(smmu->irqs[i], smmu);
1862 /* Turn the thing off */
1863 writel(sCR0_CLIENTPD, ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sCR0);
1867 static struct platform_driver arm_smmu_driver = {
1870 .of_match_table = of_match_ptr(arm_smmu_of_match),
1872 .probe = arm_smmu_device_dt_probe,
1873 .remove = arm_smmu_device_remove,
1876 static int __init arm_smmu_init(void)
1878 struct device_node *np;
1882 * Play nice with systems that don't have an ARM SMMU by checking that
1883 * an ARM SMMU exists in the system before proceeding with the driver
1884 * and IOMMU bus operation registration.
1886 np = of_find_matching_node(NULL, arm_smmu_of_match);
1892 ret = platform_driver_register(&arm_smmu_driver);
1896 /* Oh, for a proper bus abstraction */
1897 if (!iommu_present(&platform_bus_type))
1898 bus_set_iommu(&platform_bus_type, &arm_smmu_ops);
1900 #ifdef CONFIG_ARM_AMBA
1901 if (!iommu_present(&amba_bustype))
1902 bus_set_iommu(&amba_bustype, &arm_smmu_ops);
1906 if (!iommu_present(&pci_bus_type))
1907 bus_set_iommu(&pci_bus_type, &arm_smmu_ops);
1913 static void __exit arm_smmu_exit(void)
1915 return platform_driver_unregister(&arm_smmu_driver);
1918 subsys_initcall(arm_smmu_init);
1919 module_exit(arm_smmu_exit);
1921 MODULE_DESCRIPTION("IOMMU API for ARM architected SMMU implementations");
1922 MODULE_AUTHOR("Will Deacon <will.deacon@arm.com>");
1923 MODULE_LICENSE("GPL v2");