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/pci.h>
41 #include <linux/platform_device.h>
42 #include <linux/slab.h>
43 #include <linux/spinlock.h>
45 #include <linux/amba/bus.h>
47 #include "io-pgtable.h"
49 /* Maximum number of stream IDs assigned to a single device */
50 #define MAX_MASTER_STREAMIDS MAX_PHANDLE_ARGS
52 /* Maximum number of context banks per SMMU */
53 #define ARM_SMMU_MAX_CBS 128
55 /* Maximum number of mapping groups per SMMU */
56 #define ARM_SMMU_MAX_SMRS 128
58 /* SMMU global address space */
59 #define ARM_SMMU_GR0(smmu) ((smmu)->base)
60 #define ARM_SMMU_GR1(smmu) ((smmu)->base + (1 << (smmu)->pgshift))
63 * SMMU global address space with conditional offset to access secure
64 * aliases of non-secure registers (e.g. nsCR0: 0x400, nsGFSR: 0x448,
67 #define ARM_SMMU_GR0_NS(smmu) \
69 ((smmu->options & ARM_SMMU_OPT_SECURE_CFG_ACCESS) \
72 /* Configuration registers */
73 #define ARM_SMMU_GR0_sCR0 0x0
74 #define sCR0_CLIENTPD (1 << 0)
75 #define sCR0_GFRE (1 << 1)
76 #define sCR0_GFIE (1 << 2)
77 #define sCR0_GCFGFRE (1 << 4)
78 #define sCR0_GCFGFIE (1 << 5)
79 #define sCR0_USFCFG (1 << 10)
80 #define sCR0_VMIDPNE (1 << 11)
81 #define sCR0_PTM (1 << 12)
82 #define sCR0_FB (1 << 13)
83 #define sCR0_BSU_SHIFT 14
84 #define sCR0_BSU_MASK 0x3
86 /* Identification registers */
87 #define ARM_SMMU_GR0_ID0 0x20
88 #define ARM_SMMU_GR0_ID1 0x24
89 #define ARM_SMMU_GR0_ID2 0x28
90 #define ARM_SMMU_GR0_ID3 0x2c
91 #define ARM_SMMU_GR0_ID4 0x30
92 #define ARM_SMMU_GR0_ID5 0x34
93 #define ARM_SMMU_GR0_ID6 0x38
94 #define ARM_SMMU_GR0_ID7 0x3c
95 #define ARM_SMMU_GR0_sGFSR 0x48
96 #define ARM_SMMU_GR0_sGFSYNR0 0x50
97 #define ARM_SMMU_GR0_sGFSYNR1 0x54
98 #define ARM_SMMU_GR0_sGFSYNR2 0x58
100 #define ID0_S1TS (1 << 30)
101 #define ID0_S2TS (1 << 29)
102 #define ID0_NTS (1 << 28)
103 #define ID0_SMS (1 << 27)
104 #define ID0_ATOSNS (1 << 26)
105 #define ID0_CTTW (1 << 14)
106 #define ID0_NUMIRPT_SHIFT 16
107 #define ID0_NUMIRPT_MASK 0xff
108 #define ID0_NUMSIDB_SHIFT 9
109 #define ID0_NUMSIDB_MASK 0xf
110 #define ID0_NUMSMRG_SHIFT 0
111 #define ID0_NUMSMRG_MASK 0xff
113 #define ID1_PAGESIZE (1 << 31)
114 #define ID1_NUMPAGENDXB_SHIFT 28
115 #define ID1_NUMPAGENDXB_MASK 7
116 #define ID1_NUMS2CB_SHIFT 16
117 #define ID1_NUMS2CB_MASK 0xff
118 #define ID1_NUMCB_SHIFT 0
119 #define ID1_NUMCB_MASK 0xff
121 #define ID2_OAS_SHIFT 4
122 #define ID2_OAS_MASK 0xf
123 #define ID2_IAS_SHIFT 0
124 #define ID2_IAS_MASK 0xf
125 #define ID2_UBS_SHIFT 8
126 #define ID2_UBS_MASK 0xf
127 #define ID2_PTFS_4K (1 << 12)
128 #define ID2_PTFS_16K (1 << 13)
129 #define ID2_PTFS_64K (1 << 14)
131 /* Global TLB invalidation */
132 #define ARM_SMMU_GR0_TLBIVMID 0x64
133 #define ARM_SMMU_GR0_TLBIALLNSNH 0x68
134 #define ARM_SMMU_GR0_TLBIALLH 0x6c
135 #define ARM_SMMU_GR0_sTLBGSYNC 0x70
136 #define ARM_SMMU_GR0_sTLBGSTATUS 0x74
137 #define sTLBGSTATUS_GSACTIVE (1 << 0)
138 #define TLB_LOOP_TIMEOUT 1000000 /* 1s! */
140 /* Stream mapping registers */
141 #define ARM_SMMU_GR0_SMR(n) (0x800 + ((n) << 2))
142 #define SMR_VALID (1 << 31)
143 #define SMR_MASK_SHIFT 16
144 #define SMR_MASK_MASK 0x7fff
145 #define SMR_ID_SHIFT 0
146 #define SMR_ID_MASK 0x7fff
148 #define ARM_SMMU_GR0_S2CR(n) (0xc00 + ((n) << 2))
149 #define S2CR_CBNDX_SHIFT 0
150 #define S2CR_CBNDX_MASK 0xff
151 #define S2CR_TYPE_SHIFT 16
152 #define S2CR_TYPE_MASK 0x3
153 #define S2CR_TYPE_TRANS (0 << S2CR_TYPE_SHIFT)
154 #define S2CR_TYPE_BYPASS (1 << S2CR_TYPE_SHIFT)
155 #define S2CR_TYPE_FAULT (2 << S2CR_TYPE_SHIFT)
157 /* Context bank attribute registers */
158 #define ARM_SMMU_GR1_CBAR(n) (0x0 + ((n) << 2))
159 #define CBAR_VMID_SHIFT 0
160 #define CBAR_VMID_MASK 0xff
161 #define CBAR_S1_BPSHCFG_SHIFT 8
162 #define CBAR_S1_BPSHCFG_MASK 3
163 #define CBAR_S1_BPSHCFG_NSH 3
164 #define CBAR_S1_MEMATTR_SHIFT 12
165 #define CBAR_S1_MEMATTR_MASK 0xf
166 #define CBAR_S1_MEMATTR_WB 0xf
167 #define CBAR_TYPE_SHIFT 16
168 #define CBAR_TYPE_MASK 0x3
169 #define CBAR_TYPE_S2_TRANS (0 << CBAR_TYPE_SHIFT)
170 #define CBAR_TYPE_S1_TRANS_S2_BYPASS (1 << CBAR_TYPE_SHIFT)
171 #define CBAR_TYPE_S1_TRANS_S2_FAULT (2 << CBAR_TYPE_SHIFT)
172 #define CBAR_TYPE_S1_TRANS_S2_TRANS (3 << CBAR_TYPE_SHIFT)
173 #define CBAR_IRPTNDX_SHIFT 24
174 #define CBAR_IRPTNDX_MASK 0xff
176 #define ARM_SMMU_GR1_CBA2R(n) (0x800 + ((n) << 2))
177 #define CBA2R_RW64_32BIT (0 << 0)
178 #define CBA2R_RW64_64BIT (1 << 0)
180 /* Translation context bank */
181 #define ARM_SMMU_CB_BASE(smmu) ((smmu)->base + ((smmu)->size >> 1))
182 #define ARM_SMMU_CB(smmu, n) ((n) * (1 << (smmu)->pgshift))
184 #define ARM_SMMU_CB_SCTLR 0x0
185 #define ARM_SMMU_CB_RESUME 0x8
186 #define ARM_SMMU_CB_TTBCR2 0x10
187 #define ARM_SMMU_CB_TTBR0_LO 0x20
188 #define ARM_SMMU_CB_TTBR0_HI 0x24
189 #define ARM_SMMU_CB_TTBR1_LO 0x28
190 #define ARM_SMMU_CB_TTBR1_HI 0x2c
191 #define ARM_SMMU_CB_TTBCR 0x30
192 #define ARM_SMMU_CB_S1_MAIR0 0x38
193 #define ARM_SMMU_CB_S1_MAIR1 0x3c
194 #define ARM_SMMU_CB_PAR_LO 0x50
195 #define ARM_SMMU_CB_PAR_HI 0x54
196 #define ARM_SMMU_CB_FSR 0x58
197 #define ARM_SMMU_CB_FAR_LO 0x60
198 #define ARM_SMMU_CB_FAR_HI 0x64
199 #define ARM_SMMU_CB_FSYNR0 0x68
200 #define ARM_SMMU_CB_S1_TLBIVA 0x600
201 #define ARM_SMMU_CB_S1_TLBIASID 0x610
202 #define ARM_SMMU_CB_S1_TLBIVAL 0x620
203 #define ARM_SMMU_CB_S2_TLBIIPAS2 0x630
204 #define ARM_SMMU_CB_S2_TLBIIPAS2L 0x638
205 #define ARM_SMMU_CB_ATS1PR_LO 0x800
206 #define ARM_SMMU_CB_ATS1PR_HI 0x804
207 #define ARM_SMMU_CB_ATSR 0x8f0
209 #define SCTLR_S1_ASIDPNE (1 << 12)
210 #define SCTLR_CFCFG (1 << 7)
211 #define SCTLR_CFIE (1 << 6)
212 #define SCTLR_CFRE (1 << 5)
213 #define SCTLR_E (1 << 4)
214 #define SCTLR_AFE (1 << 2)
215 #define SCTLR_TRE (1 << 1)
216 #define SCTLR_M (1 << 0)
217 #define SCTLR_EAE_SBOP (SCTLR_AFE | SCTLR_TRE)
219 #define CB_PAR_F (1 << 0)
221 #define ATSR_ACTIVE (1 << 0)
223 #define RESUME_RETRY (0 << 0)
224 #define RESUME_TERMINATE (1 << 0)
226 #define TTBCR2_SEP_SHIFT 15
227 #define TTBCR2_SEP_MASK 0x7
229 #define TTBCR2_ADDR_32 0
230 #define TTBCR2_ADDR_36 1
231 #define TTBCR2_ADDR_40 2
232 #define TTBCR2_ADDR_42 3
233 #define TTBCR2_ADDR_44 4
234 #define TTBCR2_ADDR_48 5
236 #define TTBRn_HI_ASID_SHIFT 16
238 #define FSR_MULTI (1 << 31)
239 #define FSR_SS (1 << 30)
240 #define FSR_UUT (1 << 8)
241 #define FSR_ASF (1 << 7)
242 #define FSR_TLBLKF (1 << 6)
243 #define FSR_TLBMCF (1 << 5)
244 #define FSR_EF (1 << 4)
245 #define FSR_PF (1 << 3)
246 #define FSR_AFF (1 << 2)
247 #define FSR_TF (1 << 1)
249 #define FSR_IGN (FSR_AFF | FSR_ASF | \
250 FSR_TLBMCF | FSR_TLBLKF)
251 #define FSR_FAULT (FSR_MULTI | FSR_SS | FSR_UUT | \
252 FSR_EF | FSR_PF | FSR_TF | FSR_IGN)
254 #define FSYNR0_WNR (1 << 4)
256 static int force_stage;
257 module_param_named(force_stage, force_stage, int, S_IRUGO | S_IWUSR);
258 MODULE_PARM_DESC(force_stage,
259 "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.");
261 enum arm_smmu_arch_version {
266 struct arm_smmu_smr {
272 struct arm_smmu_master_cfg {
274 u16 streamids[MAX_MASTER_STREAMIDS];
275 struct arm_smmu_smr *smrs;
278 struct arm_smmu_master {
279 struct device_node *of_node;
281 struct arm_smmu_master_cfg cfg;
284 struct arm_smmu_device {
289 unsigned long pgshift;
291 #define ARM_SMMU_FEAT_COHERENT_WALK (1 << 0)
292 #define ARM_SMMU_FEAT_STREAM_MATCH (1 << 1)
293 #define ARM_SMMU_FEAT_TRANS_S1 (1 << 2)
294 #define ARM_SMMU_FEAT_TRANS_S2 (1 << 3)
295 #define ARM_SMMU_FEAT_TRANS_NESTED (1 << 4)
296 #define ARM_SMMU_FEAT_TRANS_OPS (1 << 5)
299 #define ARM_SMMU_OPT_SECURE_CFG_ACCESS (1 << 0)
301 enum arm_smmu_arch_version version;
303 u32 num_context_banks;
304 u32 num_s2_context_banks;
305 DECLARE_BITMAP(context_map, ARM_SMMU_MAX_CBS);
308 u32 num_mapping_groups;
309 DECLARE_BITMAP(smr_map, ARM_SMMU_MAX_SMRS);
311 unsigned long va_size;
312 unsigned long ipa_size;
313 unsigned long pa_size;
316 u32 num_context_irqs;
319 struct list_head list;
320 struct rb_root masters;
323 struct arm_smmu_cfg {
328 #define INVALID_IRPTNDX 0xff
330 #define ARM_SMMU_CB_ASID(cfg) ((cfg)->cbndx)
331 #define ARM_SMMU_CB_VMID(cfg) ((cfg)->cbndx + 1)
333 enum arm_smmu_domain_stage {
334 ARM_SMMU_DOMAIN_S1 = 0,
336 ARM_SMMU_DOMAIN_NESTED,
339 struct arm_smmu_domain {
340 struct arm_smmu_device *smmu;
341 struct io_pgtable_ops *pgtbl_ops;
342 spinlock_t pgtbl_lock;
343 struct arm_smmu_cfg cfg;
344 enum arm_smmu_domain_stage stage;
345 struct mutex init_mutex; /* Protects smmu pointer */
348 static struct iommu_ops arm_smmu_ops;
350 static DEFINE_SPINLOCK(arm_smmu_devices_lock);
351 static LIST_HEAD(arm_smmu_devices);
353 struct arm_smmu_option_prop {
358 static struct arm_smmu_option_prop arm_smmu_options[] = {
359 { ARM_SMMU_OPT_SECURE_CFG_ACCESS, "calxeda,smmu-secure-config-access" },
363 static void parse_driver_options(struct arm_smmu_device *smmu)
368 if (of_property_read_bool(smmu->dev->of_node,
369 arm_smmu_options[i].prop)) {
370 smmu->options |= arm_smmu_options[i].opt;
371 dev_notice(smmu->dev, "option %s\n",
372 arm_smmu_options[i].prop);
374 } while (arm_smmu_options[++i].opt);
377 static struct device_node *dev_get_dev_node(struct device *dev)
379 if (dev_is_pci(dev)) {
380 struct pci_bus *bus = to_pci_dev(dev)->bus;
382 while (!pci_is_root_bus(bus))
384 return bus->bridge->parent->of_node;
390 static struct arm_smmu_master *find_smmu_master(struct arm_smmu_device *smmu,
391 struct device_node *dev_node)
393 struct rb_node *node = smmu->masters.rb_node;
396 struct arm_smmu_master *master;
398 master = container_of(node, struct arm_smmu_master, node);
400 if (dev_node < master->of_node)
401 node = node->rb_left;
402 else if (dev_node > master->of_node)
403 node = node->rb_right;
411 static struct arm_smmu_master_cfg *
412 find_smmu_master_cfg(struct device *dev)
414 struct arm_smmu_master_cfg *cfg = NULL;
415 struct iommu_group *group = iommu_group_get(dev);
418 cfg = iommu_group_get_iommudata(group);
419 iommu_group_put(group);
425 static int insert_smmu_master(struct arm_smmu_device *smmu,
426 struct arm_smmu_master *master)
428 struct rb_node **new, *parent;
430 new = &smmu->masters.rb_node;
433 struct arm_smmu_master *this
434 = container_of(*new, struct arm_smmu_master, node);
437 if (master->of_node < this->of_node)
438 new = &((*new)->rb_left);
439 else if (master->of_node > this->of_node)
440 new = &((*new)->rb_right);
445 rb_link_node(&master->node, parent, new);
446 rb_insert_color(&master->node, &smmu->masters);
450 static int register_smmu_master(struct arm_smmu_device *smmu,
452 struct of_phandle_args *masterspec)
455 struct arm_smmu_master *master;
457 master = find_smmu_master(smmu, masterspec->np);
460 "rejecting multiple registrations for master device %s\n",
461 masterspec->np->name);
465 if (masterspec->args_count > MAX_MASTER_STREAMIDS) {
467 "reached maximum number (%d) of stream IDs for master device %s\n",
468 MAX_MASTER_STREAMIDS, masterspec->np->name);
472 master = devm_kzalloc(dev, sizeof(*master), GFP_KERNEL);
476 master->of_node = masterspec->np;
477 master->cfg.num_streamids = masterspec->args_count;
479 for (i = 0; i < master->cfg.num_streamids; ++i) {
480 u16 streamid = masterspec->args[i];
482 if (!(smmu->features & ARM_SMMU_FEAT_STREAM_MATCH) &&
483 (streamid >= smmu->num_mapping_groups)) {
485 "stream ID for master device %s greater than maximum allowed (%d)\n",
486 masterspec->np->name, smmu->num_mapping_groups);
489 master->cfg.streamids[i] = streamid;
491 return insert_smmu_master(smmu, master);
494 static struct arm_smmu_device *find_smmu_for_device(struct device *dev)
496 struct arm_smmu_device *smmu;
497 struct arm_smmu_master *master = NULL;
498 struct device_node *dev_node = dev_get_dev_node(dev);
500 spin_lock(&arm_smmu_devices_lock);
501 list_for_each_entry(smmu, &arm_smmu_devices, list) {
502 master = find_smmu_master(smmu, dev_node);
506 spin_unlock(&arm_smmu_devices_lock);
508 return master ? smmu : NULL;
511 static int __arm_smmu_alloc_bitmap(unsigned long *map, int start, int end)
516 idx = find_next_zero_bit(map, end, start);
519 } while (test_and_set_bit(idx, map));
524 static void __arm_smmu_free_bitmap(unsigned long *map, int idx)
529 /* Wait for any pending TLB invalidations to complete */
530 static void __arm_smmu_tlb_sync(struct arm_smmu_device *smmu)
533 void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
535 writel_relaxed(0, gr0_base + ARM_SMMU_GR0_sTLBGSYNC);
536 while (readl_relaxed(gr0_base + ARM_SMMU_GR0_sTLBGSTATUS)
537 & sTLBGSTATUS_GSACTIVE) {
539 if (++count == TLB_LOOP_TIMEOUT) {
540 dev_err_ratelimited(smmu->dev,
541 "TLB sync timed out -- SMMU may be deadlocked\n");
548 static void arm_smmu_tlb_sync(void *cookie)
550 struct arm_smmu_domain *smmu_domain = cookie;
551 __arm_smmu_tlb_sync(smmu_domain->smmu);
554 static void arm_smmu_tlb_inv_context(void *cookie)
556 struct arm_smmu_domain *smmu_domain = cookie;
557 struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
558 struct arm_smmu_device *smmu = smmu_domain->smmu;
559 bool stage1 = cfg->cbar != CBAR_TYPE_S2_TRANS;
563 base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
564 writel_relaxed(ARM_SMMU_CB_ASID(cfg),
565 base + ARM_SMMU_CB_S1_TLBIASID);
567 base = ARM_SMMU_GR0(smmu);
568 writel_relaxed(ARM_SMMU_CB_VMID(cfg),
569 base + ARM_SMMU_GR0_TLBIVMID);
572 __arm_smmu_tlb_sync(smmu);
575 static void arm_smmu_tlb_inv_range_nosync(unsigned long iova, size_t size,
576 bool leaf, void *cookie)
578 struct arm_smmu_domain *smmu_domain = cookie;
579 struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
580 struct arm_smmu_device *smmu = smmu_domain->smmu;
581 bool stage1 = cfg->cbar != CBAR_TYPE_S2_TRANS;
585 reg = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
586 reg += leaf ? ARM_SMMU_CB_S1_TLBIVAL : ARM_SMMU_CB_S1_TLBIVA;
588 if (!IS_ENABLED(CONFIG_64BIT) || smmu->version == ARM_SMMU_V1) {
590 iova |= ARM_SMMU_CB_ASID(cfg);
591 writel_relaxed(iova, reg);
595 iova |= (u64)ARM_SMMU_CB_ASID(cfg) << 48;
596 writeq_relaxed(iova, reg);
600 } else if (smmu->version == ARM_SMMU_V2) {
601 reg = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
602 reg += leaf ? ARM_SMMU_CB_S2_TLBIIPAS2L :
603 ARM_SMMU_CB_S2_TLBIIPAS2;
604 writeq_relaxed(iova >> 12, reg);
607 reg = ARM_SMMU_GR0(smmu) + ARM_SMMU_GR0_TLBIVMID;
608 writel_relaxed(ARM_SMMU_CB_VMID(cfg), reg);
612 static void arm_smmu_flush_pgtable(void *addr, size_t size, void *cookie)
614 struct arm_smmu_domain *smmu_domain = cookie;
615 struct arm_smmu_device *smmu = smmu_domain->smmu;
616 unsigned long offset = (unsigned long)addr & ~PAGE_MASK;
619 /* Ensure new page tables are visible to the hardware walker */
620 if (smmu->features & ARM_SMMU_FEAT_COHERENT_WALK) {
624 * If the SMMU can't walk tables in the CPU caches, treat them
625 * like non-coherent DMA since we need to flush the new entries
626 * all the way out to memory. There's no possibility of
627 * recursion here as the SMMU table walker will not be wired
628 * through another SMMU.
630 dma_map_page(smmu->dev, virt_to_page(addr), offset, size,
635 static struct iommu_gather_ops arm_smmu_gather_ops = {
636 .tlb_flush_all = arm_smmu_tlb_inv_context,
637 .tlb_add_flush = arm_smmu_tlb_inv_range_nosync,
638 .tlb_sync = arm_smmu_tlb_sync,
639 .flush_pgtable = arm_smmu_flush_pgtable,
642 static irqreturn_t arm_smmu_context_fault(int irq, void *dev)
645 u32 fsr, far, fsynr, resume;
647 struct iommu_domain *domain = dev;
648 struct arm_smmu_domain *smmu_domain = domain->priv;
649 struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
650 struct arm_smmu_device *smmu = smmu_domain->smmu;
651 void __iomem *cb_base;
653 cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
654 fsr = readl_relaxed(cb_base + ARM_SMMU_CB_FSR);
656 if (!(fsr & FSR_FAULT))
660 dev_err_ratelimited(smmu->dev,
661 "Unexpected context fault (fsr 0x%x)\n",
664 fsynr = readl_relaxed(cb_base + ARM_SMMU_CB_FSYNR0);
665 flags = fsynr & FSYNR0_WNR ? IOMMU_FAULT_WRITE : IOMMU_FAULT_READ;
667 far = readl_relaxed(cb_base + ARM_SMMU_CB_FAR_LO);
670 far = readl_relaxed(cb_base + ARM_SMMU_CB_FAR_HI);
671 iova |= ((unsigned long)far << 32);
674 if (!report_iommu_fault(domain, smmu->dev, iova, flags)) {
676 resume = RESUME_RETRY;
678 dev_err_ratelimited(smmu->dev,
679 "Unhandled context fault: iova=0x%08lx, fsynr=0x%x, cb=%d\n",
680 iova, fsynr, cfg->cbndx);
682 resume = RESUME_TERMINATE;
685 /* Clear the faulting FSR */
686 writel(fsr, cb_base + ARM_SMMU_CB_FSR);
688 /* Retry or terminate any stalled transactions */
690 writel_relaxed(resume, cb_base + ARM_SMMU_CB_RESUME);
695 static irqreturn_t arm_smmu_global_fault(int irq, void *dev)
697 u32 gfsr, gfsynr0, gfsynr1, gfsynr2;
698 struct arm_smmu_device *smmu = dev;
699 void __iomem *gr0_base = ARM_SMMU_GR0_NS(smmu);
701 gfsr = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSR);
702 gfsynr0 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSYNR0);
703 gfsynr1 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSYNR1);
704 gfsynr2 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSYNR2);
709 dev_err_ratelimited(smmu->dev,
710 "Unexpected global fault, this could be serious\n");
711 dev_err_ratelimited(smmu->dev,
712 "\tGFSR 0x%08x, GFSYNR0 0x%08x, GFSYNR1 0x%08x, GFSYNR2 0x%08x\n",
713 gfsr, gfsynr0, gfsynr1, gfsynr2);
715 writel(gfsr, gr0_base + ARM_SMMU_GR0_sGFSR);
719 static void arm_smmu_init_context_bank(struct arm_smmu_domain *smmu_domain,
720 struct io_pgtable_cfg *pgtbl_cfg)
724 struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
725 struct arm_smmu_device *smmu = smmu_domain->smmu;
726 void __iomem *cb_base, *gr0_base, *gr1_base;
728 gr0_base = ARM_SMMU_GR0(smmu);
729 gr1_base = ARM_SMMU_GR1(smmu);
730 stage1 = cfg->cbar != CBAR_TYPE_S2_TRANS;
731 cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
733 if (smmu->version > ARM_SMMU_V1) {
736 * *Must* be initialised before CBAR thanks to VMID16
737 * architectural oversight affected some implementations.
740 reg = CBA2R_RW64_64BIT;
742 reg = CBA2R_RW64_32BIT;
744 writel_relaxed(reg, gr1_base + ARM_SMMU_GR1_CBA2R(cfg->cbndx));
749 if (smmu->version == ARM_SMMU_V1)
750 reg |= cfg->irptndx << CBAR_IRPTNDX_SHIFT;
753 * Use the weakest shareability/memory types, so they are
754 * overridden by the ttbcr/pte.
757 reg |= (CBAR_S1_BPSHCFG_NSH << CBAR_S1_BPSHCFG_SHIFT) |
758 (CBAR_S1_MEMATTR_WB << CBAR_S1_MEMATTR_SHIFT);
760 reg |= ARM_SMMU_CB_VMID(cfg) << CBAR_VMID_SHIFT;
762 writel_relaxed(reg, gr1_base + ARM_SMMU_GR1_CBAR(cfg->cbndx));
766 reg = pgtbl_cfg->arm_lpae_s1_cfg.ttbr[0];
767 writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBR0_LO);
768 reg = pgtbl_cfg->arm_lpae_s1_cfg.ttbr[0] >> 32;
769 reg |= ARM_SMMU_CB_ASID(cfg) << TTBRn_HI_ASID_SHIFT;
770 writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBR0_HI);
772 reg = pgtbl_cfg->arm_lpae_s1_cfg.ttbr[1];
773 writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBR1_LO);
774 reg = pgtbl_cfg->arm_lpae_s1_cfg.ttbr[1] >> 32;
775 reg |= ARM_SMMU_CB_ASID(cfg) << TTBRn_HI_ASID_SHIFT;
776 writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBR1_HI);
778 reg = pgtbl_cfg->arm_lpae_s2_cfg.vttbr;
779 writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBR0_LO);
780 reg = pgtbl_cfg->arm_lpae_s2_cfg.vttbr >> 32;
781 writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBR0_HI);
786 reg = pgtbl_cfg->arm_lpae_s1_cfg.tcr;
787 writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBCR);
788 if (smmu->version > ARM_SMMU_V1) {
789 reg = pgtbl_cfg->arm_lpae_s1_cfg.tcr >> 32;
790 switch (smmu->va_size) {
792 reg |= (TTBCR2_ADDR_32 << TTBCR2_SEP_SHIFT);
795 reg |= (TTBCR2_ADDR_36 << TTBCR2_SEP_SHIFT);
798 reg |= (TTBCR2_ADDR_40 << TTBCR2_SEP_SHIFT);
801 reg |= (TTBCR2_ADDR_42 << TTBCR2_SEP_SHIFT);
804 reg |= (TTBCR2_ADDR_44 << TTBCR2_SEP_SHIFT);
807 reg |= (TTBCR2_ADDR_48 << TTBCR2_SEP_SHIFT);
810 writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBCR2);
813 reg = pgtbl_cfg->arm_lpae_s2_cfg.vtcr;
814 writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBCR);
817 /* MAIRs (stage-1 only) */
819 reg = pgtbl_cfg->arm_lpae_s1_cfg.mair[0];
820 writel_relaxed(reg, cb_base + ARM_SMMU_CB_S1_MAIR0);
821 reg = pgtbl_cfg->arm_lpae_s1_cfg.mair[1];
822 writel_relaxed(reg, cb_base + ARM_SMMU_CB_S1_MAIR1);
826 reg = SCTLR_CFCFG | SCTLR_CFIE | SCTLR_CFRE | SCTLR_M | SCTLR_EAE_SBOP;
828 reg |= SCTLR_S1_ASIDPNE;
832 writel_relaxed(reg, cb_base + ARM_SMMU_CB_SCTLR);
835 static int arm_smmu_init_domain_context(struct iommu_domain *domain,
836 struct arm_smmu_device *smmu)
838 int irq, start, ret = 0;
839 unsigned long ias, oas;
840 struct io_pgtable_ops *pgtbl_ops;
841 struct io_pgtable_cfg pgtbl_cfg;
842 enum io_pgtable_fmt fmt;
843 struct arm_smmu_domain *smmu_domain = domain->priv;
844 struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
846 mutex_lock(&smmu_domain->init_mutex);
847 if (smmu_domain->smmu)
851 * Mapping the requested stage onto what we support is surprisingly
852 * complicated, mainly because the spec allows S1+S2 SMMUs without
853 * support for nested translation. That means we end up with the
856 * Requested Supported Actual
866 * Note that you can't actually request stage-2 mappings.
868 if (!(smmu->features & ARM_SMMU_FEAT_TRANS_S1))
869 smmu_domain->stage = ARM_SMMU_DOMAIN_S2;
870 if (!(smmu->features & ARM_SMMU_FEAT_TRANS_S2))
871 smmu_domain->stage = ARM_SMMU_DOMAIN_S1;
873 switch (smmu_domain->stage) {
874 case ARM_SMMU_DOMAIN_S1:
875 cfg->cbar = CBAR_TYPE_S1_TRANS_S2_BYPASS;
876 start = smmu->num_s2_context_banks;
878 oas = smmu->ipa_size;
879 if (IS_ENABLED(CONFIG_64BIT))
880 fmt = ARM_64_LPAE_S1;
882 fmt = ARM_32_LPAE_S1;
884 case ARM_SMMU_DOMAIN_NESTED:
886 * We will likely want to change this if/when KVM gets
889 case ARM_SMMU_DOMAIN_S2:
890 cfg->cbar = CBAR_TYPE_S2_TRANS;
892 ias = smmu->ipa_size;
894 if (IS_ENABLED(CONFIG_64BIT))
895 fmt = ARM_64_LPAE_S2;
897 fmt = ARM_32_LPAE_S2;
904 ret = __arm_smmu_alloc_bitmap(smmu->context_map, start,
905 smmu->num_context_banks);
906 if (IS_ERR_VALUE(ret))
910 if (smmu->version == ARM_SMMU_V1) {
911 cfg->irptndx = atomic_inc_return(&smmu->irptndx);
912 cfg->irptndx %= smmu->num_context_irqs;
914 cfg->irptndx = cfg->cbndx;
917 pgtbl_cfg = (struct io_pgtable_cfg) {
918 .pgsize_bitmap = arm_smmu_ops.pgsize_bitmap,
921 .tlb = &arm_smmu_gather_ops,
924 smmu_domain->smmu = smmu;
925 pgtbl_ops = alloc_io_pgtable_ops(fmt, &pgtbl_cfg, smmu_domain);
931 /* Update our support page sizes to reflect the page table format */
932 arm_smmu_ops.pgsize_bitmap = pgtbl_cfg.pgsize_bitmap;
934 /* Initialise the context bank with our page table cfg */
935 arm_smmu_init_context_bank(smmu_domain, &pgtbl_cfg);
938 * Request context fault interrupt. Do this last to avoid the
939 * handler seeing a half-initialised domain state.
941 irq = smmu->irqs[smmu->num_global_irqs + cfg->irptndx];
942 ret = request_irq(irq, arm_smmu_context_fault, IRQF_SHARED,
943 "arm-smmu-context-fault", domain);
944 if (IS_ERR_VALUE(ret)) {
945 dev_err(smmu->dev, "failed to request context IRQ %d (%u)\n",
947 cfg->irptndx = INVALID_IRPTNDX;
950 mutex_unlock(&smmu_domain->init_mutex);
952 /* Publish page table ops for map/unmap */
953 smmu_domain->pgtbl_ops = pgtbl_ops;
957 smmu_domain->smmu = NULL;
959 mutex_unlock(&smmu_domain->init_mutex);
963 static void arm_smmu_destroy_domain_context(struct iommu_domain *domain)
965 struct arm_smmu_domain *smmu_domain = domain->priv;
966 struct arm_smmu_device *smmu = smmu_domain->smmu;
967 struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
968 void __iomem *cb_base;
975 * Disable the context bank and free the page tables before freeing
978 cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
979 writel_relaxed(0, cb_base + ARM_SMMU_CB_SCTLR);
981 if (cfg->irptndx != INVALID_IRPTNDX) {
982 irq = smmu->irqs[smmu->num_global_irqs + cfg->irptndx];
983 free_irq(irq, domain);
986 if (smmu_domain->pgtbl_ops)
987 free_io_pgtable_ops(smmu_domain->pgtbl_ops);
989 __arm_smmu_free_bitmap(smmu->context_map, cfg->cbndx);
992 static int arm_smmu_domain_init(struct iommu_domain *domain)
994 struct arm_smmu_domain *smmu_domain;
997 * Allocate the domain and initialise some of its data structures.
998 * We can't really do anything meaningful until we've added a
1001 smmu_domain = kzalloc(sizeof(*smmu_domain), GFP_KERNEL);
1005 mutex_init(&smmu_domain->init_mutex);
1006 spin_lock_init(&smmu_domain->pgtbl_lock);
1007 domain->priv = smmu_domain;
1011 static void arm_smmu_domain_destroy(struct iommu_domain *domain)
1013 struct arm_smmu_domain *smmu_domain = domain->priv;
1016 * Free the domain resources. We assume that all devices have
1017 * already been detached.
1019 arm_smmu_destroy_domain_context(domain);
1023 static int arm_smmu_master_configure_smrs(struct arm_smmu_device *smmu,
1024 struct arm_smmu_master_cfg *cfg)
1027 struct arm_smmu_smr *smrs;
1028 void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1030 if (!(smmu->features & ARM_SMMU_FEAT_STREAM_MATCH))
1036 smrs = kmalloc_array(cfg->num_streamids, sizeof(*smrs), GFP_KERNEL);
1038 dev_err(smmu->dev, "failed to allocate %d SMRs\n",
1039 cfg->num_streamids);
1043 /* Allocate the SMRs on the SMMU */
1044 for (i = 0; i < cfg->num_streamids; ++i) {
1045 int idx = __arm_smmu_alloc_bitmap(smmu->smr_map, 0,
1046 smmu->num_mapping_groups);
1047 if (IS_ERR_VALUE(idx)) {
1048 dev_err(smmu->dev, "failed to allocate free SMR\n");
1052 smrs[i] = (struct arm_smmu_smr) {
1054 .mask = 0, /* We don't currently share SMRs */
1055 .id = cfg->streamids[i],
1059 /* It worked! Now, poke the actual hardware */
1060 for (i = 0; i < cfg->num_streamids; ++i) {
1061 u32 reg = SMR_VALID | smrs[i].id << SMR_ID_SHIFT |
1062 smrs[i].mask << SMR_MASK_SHIFT;
1063 writel_relaxed(reg, gr0_base + ARM_SMMU_GR0_SMR(smrs[i].idx));
1071 __arm_smmu_free_bitmap(smmu->smr_map, smrs[i].idx);
1076 static void arm_smmu_master_free_smrs(struct arm_smmu_device *smmu,
1077 struct arm_smmu_master_cfg *cfg)
1080 void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1081 struct arm_smmu_smr *smrs = cfg->smrs;
1086 /* Invalidate the SMRs before freeing back to the allocator */
1087 for (i = 0; i < cfg->num_streamids; ++i) {
1088 u8 idx = smrs[i].idx;
1090 writel_relaxed(~SMR_VALID, gr0_base + ARM_SMMU_GR0_SMR(idx));
1091 __arm_smmu_free_bitmap(smmu->smr_map, idx);
1098 static int arm_smmu_domain_add_master(struct arm_smmu_domain *smmu_domain,
1099 struct arm_smmu_master_cfg *cfg)
1102 struct arm_smmu_device *smmu = smmu_domain->smmu;
1103 void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1105 /* Devices in an IOMMU group may already be configured */
1106 ret = arm_smmu_master_configure_smrs(smmu, cfg);
1108 return ret == -EEXIST ? 0 : ret;
1110 for (i = 0; i < cfg->num_streamids; ++i) {
1113 idx = cfg->smrs ? cfg->smrs[i].idx : cfg->streamids[i];
1114 s2cr = S2CR_TYPE_TRANS |
1115 (smmu_domain->cfg.cbndx << S2CR_CBNDX_SHIFT);
1116 writel_relaxed(s2cr, gr0_base + ARM_SMMU_GR0_S2CR(idx));
1122 static void arm_smmu_domain_remove_master(struct arm_smmu_domain *smmu_domain,
1123 struct arm_smmu_master_cfg *cfg)
1126 struct arm_smmu_device *smmu = smmu_domain->smmu;
1127 void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1129 /* An IOMMU group is torn down by the first device to be removed */
1130 if ((smmu->features & ARM_SMMU_FEAT_STREAM_MATCH) && !cfg->smrs)
1134 * We *must* clear the S2CR first, because freeing the SMR means
1135 * that it can be re-allocated immediately.
1137 for (i = 0; i < cfg->num_streamids; ++i) {
1138 u32 idx = cfg->smrs ? cfg->smrs[i].idx : cfg->streamids[i];
1140 writel_relaxed(S2CR_TYPE_BYPASS,
1141 gr0_base + ARM_SMMU_GR0_S2CR(idx));
1144 arm_smmu_master_free_smrs(smmu, cfg);
1147 static int arm_smmu_attach_dev(struct iommu_domain *domain, struct device *dev)
1150 struct arm_smmu_domain *smmu_domain = domain->priv;
1151 struct arm_smmu_device *smmu;
1152 struct arm_smmu_master_cfg *cfg;
1154 smmu = find_smmu_for_device(dev);
1156 dev_err(dev, "cannot attach to SMMU, is it on the same bus?\n");
1160 if (dev->archdata.iommu) {
1161 dev_err(dev, "already attached to IOMMU domain\n");
1165 /* Ensure that the domain is finalised */
1166 ret = arm_smmu_init_domain_context(domain, smmu);
1167 if (IS_ERR_VALUE(ret))
1171 * Sanity check the domain. We don't support domains across
1174 if (smmu_domain->smmu != smmu) {
1176 "cannot attach to SMMU %s whilst already attached to domain on SMMU %s\n",
1177 dev_name(smmu_domain->smmu->dev), dev_name(smmu->dev));
1181 /* Looks ok, so add the device to the domain */
1182 cfg = find_smmu_master_cfg(dev);
1186 ret = arm_smmu_domain_add_master(smmu_domain, cfg);
1188 dev->archdata.iommu = domain;
1192 static void arm_smmu_detach_dev(struct iommu_domain *domain, struct device *dev)
1194 struct arm_smmu_domain *smmu_domain = domain->priv;
1195 struct arm_smmu_master_cfg *cfg;
1197 cfg = find_smmu_master_cfg(dev);
1201 dev->archdata.iommu = NULL;
1202 arm_smmu_domain_remove_master(smmu_domain, cfg);
1205 static int arm_smmu_map(struct iommu_domain *domain, unsigned long iova,
1206 phys_addr_t paddr, size_t size, int prot)
1209 unsigned long flags;
1210 struct arm_smmu_domain *smmu_domain = domain->priv;
1211 struct io_pgtable_ops *ops= smmu_domain->pgtbl_ops;
1216 spin_lock_irqsave(&smmu_domain->pgtbl_lock, flags);
1217 ret = ops->map(ops, iova, paddr, size, prot);
1218 spin_unlock_irqrestore(&smmu_domain->pgtbl_lock, flags);
1222 static size_t arm_smmu_unmap(struct iommu_domain *domain, unsigned long iova,
1226 unsigned long flags;
1227 struct arm_smmu_domain *smmu_domain = domain->priv;
1228 struct io_pgtable_ops *ops= smmu_domain->pgtbl_ops;
1233 spin_lock_irqsave(&smmu_domain->pgtbl_lock, flags);
1234 ret = ops->unmap(ops, iova, size);
1235 spin_unlock_irqrestore(&smmu_domain->pgtbl_lock, flags);
1239 static phys_addr_t arm_smmu_iova_to_phys_hard(struct iommu_domain *domain,
1242 struct arm_smmu_domain *smmu_domain = domain->priv;
1243 struct arm_smmu_device *smmu = smmu_domain->smmu;
1244 struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
1245 struct io_pgtable_ops *ops= smmu_domain->pgtbl_ops;
1246 struct device *dev = smmu->dev;
1247 void __iomem *cb_base;
1251 cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
1253 if (smmu->version == 1) {
1254 u32 reg = iova & ~0xfff;
1255 writel_relaxed(reg, cb_base + ARM_SMMU_CB_ATS1PR_LO);
1257 u32 reg = iova & ~0xfff;
1258 writel_relaxed(reg, cb_base + ARM_SMMU_CB_ATS1PR_LO);
1259 reg = ((u64)iova & ~0xfff) >> 32;
1260 writel_relaxed(reg, cb_base + ARM_SMMU_CB_ATS1PR_HI);
1263 if (readl_poll_timeout_atomic(cb_base + ARM_SMMU_CB_ATSR, tmp,
1264 !(tmp & ATSR_ACTIVE), 5, 50)) {
1266 "iova to phys timed out on 0x%pad. Falling back to software table walk.\n",
1268 return ops->iova_to_phys(ops, iova);
1271 phys = readl_relaxed(cb_base + ARM_SMMU_CB_PAR_LO);
1272 phys |= ((u64)readl_relaxed(cb_base + ARM_SMMU_CB_PAR_HI)) << 32;
1274 if (phys & CB_PAR_F) {
1275 dev_err(dev, "translation fault!\n");
1276 dev_err(dev, "PAR = 0x%llx\n", phys);
1280 return (phys & GENMASK_ULL(39, 12)) | (iova & 0xfff);
1283 static phys_addr_t arm_smmu_iova_to_phys(struct iommu_domain *domain,
1287 unsigned long flags;
1288 struct arm_smmu_domain *smmu_domain = domain->priv;
1289 struct io_pgtable_ops *ops= smmu_domain->pgtbl_ops;
1294 spin_lock_irqsave(&smmu_domain->pgtbl_lock, flags);
1295 if (smmu_domain->smmu->features & ARM_SMMU_FEAT_TRANS_OPS)
1296 ret = arm_smmu_iova_to_phys_hard(domain, iova);
1298 ret = ops->iova_to_phys(ops, iova);
1299 spin_unlock_irqrestore(&smmu_domain->pgtbl_lock, flags);
1304 static bool arm_smmu_capable(enum iommu_cap cap)
1307 case IOMMU_CAP_CACHE_COHERENCY:
1309 * Return true here as the SMMU can always send out coherent
1313 case IOMMU_CAP_INTR_REMAP:
1314 return true; /* MSIs are just memory writes */
1315 case IOMMU_CAP_NOEXEC:
1322 static int __arm_smmu_get_pci_sid(struct pci_dev *pdev, u16 alias, void *data)
1324 *((u16 *)data) = alias;
1325 return 0; /* Continue walking */
1328 static void __arm_smmu_release_pci_iommudata(void *data)
1333 static int arm_smmu_add_device(struct device *dev)
1335 struct arm_smmu_device *smmu;
1336 struct arm_smmu_master_cfg *cfg;
1337 struct iommu_group *group;
1338 void (*releasefn)(void *) = NULL;
1341 smmu = find_smmu_for_device(dev);
1345 group = iommu_group_alloc();
1346 if (IS_ERR(group)) {
1347 dev_err(dev, "Failed to allocate IOMMU group\n");
1348 return PTR_ERR(group);
1351 if (dev_is_pci(dev)) {
1352 struct pci_dev *pdev = to_pci_dev(dev);
1354 cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1360 cfg->num_streamids = 1;
1362 * Assume Stream ID == Requester ID for now.
1363 * We need a way to describe the ID mappings in FDT.
1365 pci_for_each_dma_alias(pdev, __arm_smmu_get_pci_sid,
1366 &cfg->streamids[0]);
1367 releasefn = __arm_smmu_release_pci_iommudata;
1369 struct arm_smmu_master *master;
1371 master = find_smmu_master(smmu, dev->of_node);
1380 iommu_group_set_iommudata(group, cfg, releasefn);
1381 ret = iommu_group_add_device(group, dev);
1384 iommu_group_put(group);
1388 static void arm_smmu_remove_device(struct device *dev)
1390 iommu_group_remove_device(dev);
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 = domain->priv;
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 = domain->priv;
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_init = arm_smmu_domain_init,
1440 .domain_destroy = arm_smmu_domain_destroy,
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 .domain_get_attr = arm_smmu_domain_get_attr,
1450 .domain_set_attr = arm_smmu_domain_set_attr,
1451 .pgsize_bitmap = -1UL, /* Restricted during device attach */
1454 static void arm_smmu_device_reset(struct arm_smmu_device *smmu)
1456 void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1457 void __iomem *cb_base;
1461 /* clear global FSR */
1462 reg = readl_relaxed(ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sGFSR);
1463 writel(reg, ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sGFSR);
1465 /* Mark all SMRn as invalid and all S2CRn as bypass */
1466 for (i = 0; i < smmu->num_mapping_groups; ++i) {
1467 writel_relaxed(0, gr0_base + ARM_SMMU_GR0_SMR(i));
1468 writel_relaxed(S2CR_TYPE_BYPASS,
1469 gr0_base + ARM_SMMU_GR0_S2CR(i));
1472 /* Make sure all context banks are disabled and clear CB_FSR */
1473 for (i = 0; i < smmu->num_context_banks; ++i) {
1474 cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, i);
1475 writel_relaxed(0, cb_base + ARM_SMMU_CB_SCTLR);
1476 writel_relaxed(FSR_FAULT, cb_base + ARM_SMMU_CB_FSR);
1479 /* Invalidate the TLB, just in case */
1480 writel_relaxed(0, gr0_base + ARM_SMMU_GR0_TLBIALLH);
1481 writel_relaxed(0, gr0_base + ARM_SMMU_GR0_TLBIALLNSNH);
1483 reg = readl_relaxed(ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sCR0);
1485 /* Enable fault reporting */
1486 reg |= (sCR0_GFRE | sCR0_GFIE | sCR0_GCFGFRE | sCR0_GCFGFIE);
1488 /* Disable TLB broadcasting. */
1489 reg |= (sCR0_VMIDPNE | sCR0_PTM);
1491 /* Enable client access, but bypass when no mapping is found */
1492 reg &= ~(sCR0_CLIENTPD | sCR0_USFCFG);
1494 /* Disable forced broadcasting */
1497 /* Don't upgrade barriers */
1498 reg &= ~(sCR0_BSU_MASK << sCR0_BSU_SHIFT);
1500 /* Push the button */
1501 __arm_smmu_tlb_sync(smmu);
1502 writel(reg, ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sCR0);
1505 static int arm_smmu_id_size_to_bits(int size)
1524 static int arm_smmu_device_cfg_probe(struct arm_smmu_device *smmu)
1527 void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1530 dev_notice(smmu->dev, "probing hardware configuration...\n");
1531 dev_notice(smmu->dev, "SMMUv%d with:\n", smmu->version);
1534 id = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID0);
1536 /* Restrict available stages based on module parameter */
1537 if (force_stage == 1)
1538 id &= ~(ID0_S2TS | ID0_NTS);
1539 else if (force_stage == 2)
1540 id &= ~(ID0_S1TS | ID0_NTS);
1542 if (id & ID0_S1TS) {
1543 smmu->features |= ARM_SMMU_FEAT_TRANS_S1;
1544 dev_notice(smmu->dev, "\tstage 1 translation\n");
1547 if (id & ID0_S2TS) {
1548 smmu->features |= ARM_SMMU_FEAT_TRANS_S2;
1549 dev_notice(smmu->dev, "\tstage 2 translation\n");
1553 smmu->features |= ARM_SMMU_FEAT_TRANS_NESTED;
1554 dev_notice(smmu->dev, "\tnested translation\n");
1557 if (!(smmu->features &
1558 (ARM_SMMU_FEAT_TRANS_S1 | ARM_SMMU_FEAT_TRANS_S2))) {
1559 dev_err(smmu->dev, "\tno translation support!\n");
1563 if (smmu->version == 1 || (!(id & ID0_ATOSNS) && (id & ID0_S1TS))) {
1564 smmu->features |= ARM_SMMU_FEAT_TRANS_OPS;
1565 dev_notice(smmu->dev, "\taddress translation ops\n");
1568 if (id & ID0_CTTW) {
1569 smmu->features |= ARM_SMMU_FEAT_COHERENT_WALK;
1570 dev_notice(smmu->dev, "\tcoherent table walk\n");
1576 smmu->features |= ARM_SMMU_FEAT_STREAM_MATCH;
1577 smmu->num_mapping_groups = (id >> ID0_NUMSMRG_SHIFT) &
1579 if (smmu->num_mapping_groups == 0) {
1581 "stream-matching supported, but no SMRs present!\n");
1585 smr = SMR_MASK_MASK << SMR_MASK_SHIFT;
1586 smr |= (SMR_ID_MASK << SMR_ID_SHIFT);
1587 writel_relaxed(smr, gr0_base + ARM_SMMU_GR0_SMR(0));
1588 smr = readl_relaxed(gr0_base + ARM_SMMU_GR0_SMR(0));
1590 mask = (smr >> SMR_MASK_SHIFT) & SMR_MASK_MASK;
1591 sid = (smr >> SMR_ID_SHIFT) & SMR_ID_MASK;
1592 if ((mask & sid) != sid) {
1594 "SMR mask bits (0x%x) insufficient for ID field (0x%x)\n",
1599 dev_notice(smmu->dev,
1600 "\tstream matching with %u register groups, mask 0x%x",
1601 smmu->num_mapping_groups, mask);
1603 smmu->num_mapping_groups = (id >> ID0_NUMSIDB_SHIFT) &
1608 id = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID1);
1609 smmu->pgshift = (id & ID1_PAGESIZE) ? 16 : 12;
1611 /* Check for size mismatch of SMMU address space from mapped region */
1612 size = 1 << (((id >> ID1_NUMPAGENDXB_SHIFT) & ID1_NUMPAGENDXB_MASK) + 1);
1613 size *= 2 << smmu->pgshift;
1614 if (smmu->size != size)
1616 "SMMU address space size (0x%lx) differs from mapped region size (0x%lx)!\n",
1619 smmu->num_s2_context_banks = (id >> ID1_NUMS2CB_SHIFT) & ID1_NUMS2CB_MASK;
1620 smmu->num_context_banks = (id >> ID1_NUMCB_SHIFT) & ID1_NUMCB_MASK;
1621 if (smmu->num_s2_context_banks > smmu->num_context_banks) {
1622 dev_err(smmu->dev, "impossible number of S2 context banks!\n");
1625 dev_notice(smmu->dev, "\t%u context banks (%u stage-2 only)\n",
1626 smmu->num_context_banks, smmu->num_s2_context_banks);
1629 id = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID2);
1630 size = arm_smmu_id_size_to_bits((id >> ID2_IAS_SHIFT) & ID2_IAS_MASK);
1631 smmu->ipa_size = size;
1633 /* The output mask is also applied for bypass */
1634 size = arm_smmu_id_size_to_bits((id >> ID2_OAS_SHIFT) & ID2_OAS_MASK);
1635 smmu->pa_size = size;
1637 if (smmu->version == ARM_SMMU_V1) {
1638 smmu->va_size = smmu->ipa_size;
1639 size = SZ_4K | SZ_2M | SZ_1G;
1641 size = (id >> ID2_UBS_SHIFT) & ID2_UBS_MASK;
1642 smmu->va_size = arm_smmu_id_size_to_bits(size);
1643 #ifndef CONFIG_64BIT
1644 smmu->va_size = min(32UL, smmu->va_size);
1647 if (id & ID2_PTFS_4K)
1648 size |= SZ_4K | SZ_2M | SZ_1G;
1649 if (id & ID2_PTFS_16K)
1650 size |= SZ_16K | SZ_32M;
1651 if (id & ID2_PTFS_64K)
1652 size |= SZ_64K | SZ_512M;
1655 arm_smmu_ops.pgsize_bitmap &= size;
1656 dev_notice(smmu->dev, "\tSupported page sizes: 0x%08lx\n", size);
1658 if (smmu->features & ARM_SMMU_FEAT_TRANS_S1)
1659 dev_notice(smmu->dev, "\tStage-1: %lu-bit VA -> %lu-bit IPA\n",
1660 smmu->va_size, smmu->ipa_size);
1662 if (smmu->features & ARM_SMMU_FEAT_TRANS_S2)
1663 dev_notice(smmu->dev, "\tStage-2: %lu-bit IPA -> %lu-bit PA\n",
1664 smmu->ipa_size, smmu->pa_size);
1669 static const struct of_device_id arm_smmu_of_match[] = {
1670 { .compatible = "arm,smmu-v1", .data = (void *)ARM_SMMU_V1 },
1671 { .compatible = "arm,smmu-v2", .data = (void *)ARM_SMMU_V2 },
1672 { .compatible = "arm,mmu-400", .data = (void *)ARM_SMMU_V1 },
1673 { .compatible = "arm,mmu-401", .data = (void *)ARM_SMMU_V1 },
1674 { .compatible = "arm,mmu-500", .data = (void *)ARM_SMMU_V2 },
1677 MODULE_DEVICE_TABLE(of, arm_smmu_of_match);
1679 static int arm_smmu_device_dt_probe(struct platform_device *pdev)
1681 const struct of_device_id *of_id;
1682 struct resource *res;
1683 struct arm_smmu_device *smmu;
1684 struct device *dev = &pdev->dev;
1685 struct rb_node *node;
1686 struct of_phandle_args masterspec;
1687 int num_irqs, i, err;
1689 smmu = devm_kzalloc(dev, sizeof(*smmu), GFP_KERNEL);
1691 dev_err(dev, "failed to allocate arm_smmu_device\n");
1696 of_id = of_match_node(arm_smmu_of_match, dev->of_node);
1697 smmu->version = (enum arm_smmu_arch_version)of_id->data;
1699 res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1700 smmu->base = devm_ioremap_resource(dev, res);
1701 if (IS_ERR(smmu->base))
1702 return PTR_ERR(smmu->base);
1703 smmu->size = resource_size(res);
1705 if (of_property_read_u32(dev->of_node, "#global-interrupts",
1706 &smmu->num_global_irqs)) {
1707 dev_err(dev, "missing #global-interrupts property\n");
1712 while ((res = platform_get_resource(pdev, IORESOURCE_IRQ, num_irqs))) {
1714 if (num_irqs > smmu->num_global_irqs)
1715 smmu->num_context_irqs++;
1718 if (!smmu->num_context_irqs) {
1719 dev_err(dev, "found %d interrupts but expected at least %d\n",
1720 num_irqs, smmu->num_global_irqs + 1);
1724 smmu->irqs = devm_kzalloc(dev, sizeof(*smmu->irqs) * num_irqs,
1727 dev_err(dev, "failed to allocate %d irqs\n", num_irqs);
1731 for (i = 0; i < num_irqs; ++i) {
1732 int irq = platform_get_irq(pdev, i);
1735 dev_err(dev, "failed to get irq index %d\n", i);
1738 smmu->irqs[i] = irq;
1741 err = arm_smmu_device_cfg_probe(smmu);
1746 smmu->masters = RB_ROOT;
1747 while (!of_parse_phandle_with_args(dev->of_node, "mmu-masters",
1748 "#stream-id-cells", i,
1750 err = register_smmu_master(smmu, dev, &masterspec);
1752 dev_err(dev, "failed to add master %s\n",
1753 masterspec.np->name);
1754 goto out_put_masters;
1759 dev_notice(dev, "registered %d master devices\n", i);
1761 parse_driver_options(smmu);
1763 if (smmu->version > ARM_SMMU_V1 &&
1764 smmu->num_context_banks != smmu->num_context_irqs) {
1766 "found only %d context interrupt(s) but %d required\n",
1767 smmu->num_context_irqs, smmu->num_context_banks);
1769 goto out_put_masters;
1772 for (i = 0; i < smmu->num_global_irqs; ++i) {
1773 err = request_irq(smmu->irqs[i],
1774 arm_smmu_global_fault,
1776 "arm-smmu global fault",
1779 dev_err(dev, "failed to request global IRQ %d (%u)\n",
1785 INIT_LIST_HEAD(&smmu->list);
1786 spin_lock(&arm_smmu_devices_lock);
1787 list_add(&smmu->list, &arm_smmu_devices);
1788 spin_unlock(&arm_smmu_devices_lock);
1790 arm_smmu_device_reset(smmu);
1795 free_irq(smmu->irqs[i], smmu);
1798 for (node = rb_first(&smmu->masters); node; node = rb_next(node)) {
1799 struct arm_smmu_master *master
1800 = container_of(node, struct arm_smmu_master, node);
1801 of_node_put(master->of_node);
1807 static int arm_smmu_device_remove(struct platform_device *pdev)
1810 struct device *dev = &pdev->dev;
1811 struct arm_smmu_device *curr, *smmu = NULL;
1812 struct rb_node *node;
1814 spin_lock(&arm_smmu_devices_lock);
1815 list_for_each_entry(curr, &arm_smmu_devices, list) {
1816 if (curr->dev == dev) {
1818 list_del(&smmu->list);
1822 spin_unlock(&arm_smmu_devices_lock);
1827 for (node = rb_first(&smmu->masters); node; node = rb_next(node)) {
1828 struct arm_smmu_master *master
1829 = container_of(node, struct arm_smmu_master, node);
1830 of_node_put(master->of_node);
1833 if (!bitmap_empty(smmu->context_map, ARM_SMMU_MAX_CBS))
1834 dev_err(dev, "removing device with active domains!\n");
1836 for (i = 0; i < smmu->num_global_irqs; ++i)
1837 free_irq(smmu->irqs[i], smmu);
1839 /* Turn the thing off */
1840 writel(sCR0_CLIENTPD, ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sCR0);
1844 static struct platform_driver arm_smmu_driver = {
1847 .of_match_table = of_match_ptr(arm_smmu_of_match),
1849 .probe = arm_smmu_device_dt_probe,
1850 .remove = arm_smmu_device_remove,
1853 static int __init arm_smmu_init(void)
1855 struct device_node *np;
1859 * Play nice with systems that don't have an ARM SMMU by checking that
1860 * an ARM SMMU exists in the system before proceeding with the driver
1861 * and IOMMU bus operation registration.
1863 np = of_find_matching_node(NULL, arm_smmu_of_match);
1869 ret = platform_driver_register(&arm_smmu_driver);
1873 /* Oh, for a proper bus abstraction */
1874 if (!iommu_present(&platform_bus_type))
1875 bus_set_iommu(&platform_bus_type, &arm_smmu_ops);
1877 #ifdef CONFIG_ARM_AMBA
1878 if (!iommu_present(&amba_bustype))
1879 bus_set_iommu(&amba_bustype, &arm_smmu_ops);
1883 if (!iommu_present(&pci_bus_type))
1884 bus_set_iommu(&pci_bus_type, &arm_smmu_ops);
1890 static void __exit arm_smmu_exit(void)
1892 return platform_driver_unregister(&arm_smmu_driver);
1895 subsys_initcall(arm_smmu_init);
1896 module_exit(arm_smmu_exit);
1898 MODULE_DESCRIPTION("IOMMU API for ARM architected SMMU implementations");
1899 MODULE_AUTHOR("Will Deacon <will.deacon@arm.com>");
1900 MODULE_LICENSE("GPL v2");