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 * - 4k and 64k pages, with contiguous pte hints.
27 * - Up to 39-bit addressing
28 * - Context fault reporting
31 #define pr_fmt(fmt) "arm-smmu: " fmt
33 #include <linux/delay.h>
34 #include <linux/dma-mapping.h>
35 #include <linux/err.h>
36 #include <linux/interrupt.h>
38 #include <linux/iommu.h>
40 #include <linux/module.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 <asm/pgalloc.h>
50 /* Maximum number of stream IDs assigned to a single device */
51 #define MAX_MASTER_STREAMIDS 8
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 + (smmu)->pagesize)
64 #define ARM_SMMU_PTE_PAGE (((pteval_t)3) << 0)
65 #define ARM_SMMU_PTE_CONT (((pteval_t)1) << 52)
66 #define ARM_SMMU_PTE_AF (((pteval_t)1) << 10)
67 #define ARM_SMMU_PTE_SH_NS (((pteval_t)0) << 8)
68 #define ARM_SMMU_PTE_SH_OS (((pteval_t)2) << 8)
69 #define ARM_SMMU_PTE_SH_IS (((pteval_t)3) << 8)
71 #if PAGE_SIZE == SZ_4K
72 #define ARM_SMMU_PTE_CONT_ENTRIES 16
73 #elif PAGE_SIZE == SZ_64K
74 #define ARM_SMMU_PTE_CONT_ENTRIES 32
76 #define ARM_SMMU_PTE_CONT_ENTRIES 1
79 #define ARM_SMMU_PTE_CONT_SIZE (PAGE_SIZE * ARM_SMMU_PTE_CONT_ENTRIES)
80 #define ARM_SMMU_PTE_CONT_MASK (~(ARM_SMMU_PTE_CONT_SIZE - 1))
81 #define ARM_SMMU_PTE_HWTABLE_SIZE (PTRS_PER_PTE * sizeof(pte_t))
84 #define ARM_SMMU_PTE_AP_UNPRIV (((pteval_t)1) << 6)
85 #define ARM_SMMU_PTE_AP_RDONLY (((pteval_t)2) << 6)
86 #define ARM_SMMU_PTE_ATTRINDX_SHIFT 2
87 #define ARM_SMMU_PTE_nG (((pteval_t)1) << 11)
90 #define ARM_SMMU_PTE_HAP_FAULT (((pteval_t)0) << 6)
91 #define ARM_SMMU_PTE_HAP_READ (((pteval_t)1) << 6)
92 #define ARM_SMMU_PTE_HAP_WRITE (((pteval_t)2) << 6)
93 #define ARM_SMMU_PTE_MEMATTR_OIWB (((pteval_t)0xf) << 2)
94 #define ARM_SMMU_PTE_MEMATTR_NC (((pteval_t)0x5) << 2)
95 #define ARM_SMMU_PTE_MEMATTR_DEV (((pteval_t)0x1) << 2)
97 /* Configuration registers */
98 #define ARM_SMMU_GR0_sCR0 0x0
99 #define sCR0_CLIENTPD (1 << 0)
100 #define sCR0_GFRE (1 << 1)
101 #define sCR0_GFIE (1 << 2)
102 #define sCR0_GCFGFRE (1 << 4)
103 #define sCR0_GCFGFIE (1 << 5)
104 #define sCR0_USFCFG (1 << 10)
105 #define sCR0_VMIDPNE (1 << 11)
106 #define sCR0_PTM (1 << 12)
107 #define sCR0_FB (1 << 13)
108 #define sCR0_BSU_SHIFT 14
109 #define sCR0_BSU_MASK 0x3
111 /* Identification registers */
112 #define ARM_SMMU_GR0_ID0 0x20
113 #define ARM_SMMU_GR0_ID1 0x24
114 #define ARM_SMMU_GR0_ID2 0x28
115 #define ARM_SMMU_GR0_ID3 0x2c
116 #define ARM_SMMU_GR0_ID4 0x30
117 #define ARM_SMMU_GR0_ID5 0x34
118 #define ARM_SMMU_GR0_ID6 0x38
119 #define ARM_SMMU_GR0_ID7 0x3c
120 #define ARM_SMMU_GR0_sGFSR 0x48
121 #define ARM_SMMU_GR0_sGFSYNR0 0x50
122 #define ARM_SMMU_GR0_sGFSYNR1 0x54
123 #define ARM_SMMU_GR0_sGFSYNR2 0x58
124 #define ARM_SMMU_GR0_PIDR0 0xfe0
125 #define ARM_SMMU_GR0_PIDR1 0xfe4
126 #define ARM_SMMU_GR0_PIDR2 0xfe8
128 #define ID0_S1TS (1 << 30)
129 #define ID0_S2TS (1 << 29)
130 #define ID0_NTS (1 << 28)
131 #define ID0_SMS (1 << 27)
132 #define ID0_PTFS_SHIFT 24
133 #define ID0_PTFS_MASK 0x2
134 #define ID0_PTFS_V8_ONLY 0x2
135 #define ID0_CTTW (1 << 14)
136 #define ID0_NUMIRPT_SHIFT 16
137 #define ID0_NUMIRPT_MASK 0xff
138 #define ID0_NUMSMRG_SHIFT 0
139 #define ID0_NUMSMRG_MASK 0xff
141 #define ID1_PAGESIZE (1 << 31)
142 #define ID1_NUMPAGENDXB_SHIFT 28
143 #define ID1_NUMPAGENDXB_MASK 7
144 #define ID1_NUMS2CB_SHIFT 16
145 #define ID1_NUMS2CB_MASK 0xff
146 #define ID1_NUMCB_SHIFT 0
147 #define ID1_NUMCB_MASK 0xff
149 #define ID2_OAS_SHIFT 4
150 #define ID2_OAS_MASK 0xf
151 #define ID2_IAS_SHIFT 0
152 #define ID2_IAS_MASK 0xf
153 #define ID2_UBS_SHIFT 8
154 #define ID2_UBS_MASK 0xf
155 #define ID2_PTFS_4K (1 << 12)
156 #define ID2_PTFS_16K (1 << 13)
157 #define ID2_PTFS_64K (1 << 14)
159 #define PIDR2_ARCH_SHIFT 4
160 #define PIDR2_ARCH_MASK 0xf
162 /* Global TLB invalidation */
163 #define ARM_SMMU_GR0_STLBIALL 0x60
164 #define ARM_SMMU_GR0_TLBIVMID 0x64
165 #define ARM_SMMU_GR0_TLBIALLNSNH 0x68
166 #define ARM_SMMU_GR0_TLBIALLH 0x6c
167 #define ARM_SMMU_GR0_sTLBGSYNC 0x70
168 #define ARM_SMMU_GR0_sTLBGSTATUS 0x74
169 #define sTLBGSTATUS_GSACTIVE (1 << 0)
170 #define TLB_LOOP_TIMEOUT 1000000 /* 1s! */
172 /* Stream mapping registers */
173 #define ARM_SMMU_GR0_SMR(n) (0x800 + ((n) << 2))
174 #define SMR_VALID (1 << 31)
175 #define SMR_MASK_SHIFT 16
176 #define SMR_MASK_MASK 0x7fff
177 #define SMR_ID_SHIFT 0
178 #define SMR_ID_MASK 0x7fff
180 #define ARM_SMMU_GR0_S2CR(n) (0xc00 + ((n) << 2))
181 #define S2CR_CBNDX_SHIFT 0
182 #define S2CR_CBNDX_MASK 0xff
183 #define S2CR_TYPE_SHIFT 16
184 #define S2CR_TYPE_MASK 0x3
185 #define S2CR_TYPE_TRANS (0 << S2CR_TYPE_SHIFT)
186 #define S2CR_TYPE_BYPASS (1 << S2CR_TYPE_SHIFT)
187 #define S2CR_TYPE_FAULT (2 << S2CR_TYPE_SHIFT)
189 /* Context bank attribute registers */
190 #define ARM_SMMU_GR1_CBAR(n) (0x0 + ((n) << 2))
191 #define CBAR_VMID_SHIFT 0
192 #define CBAR_VMID_MASK 0xff
193 #define CBAR_S1_MEMATTR_SHIFT 12
194 #define CBAR_S1_MEMATTR_MASK 0xf
195 #define CBAR_S1_MEMATTR_WB 0xf
196 #define CBAR_TYPE_SHIFT 16
197 #define CBAR_TYPE_MASK 0x3
198 #define CBAR_TYPE_S2_TRANS (0 << CBAR_TYPE_SHIFT)
199 #define CBAR_TYPE_S1_TRANS_S2_BYPASS (1 << CBAR_TYPE_SHIFT)
200 #define CBAR_TYPE_S1_TRANS_S2_FAULT (2 << CBAR_TYPE_SHIFT)
201 #define CBAR_TYPE_S1_TRANS_S2_TRANS (3 << CBAR_TYPE_SHIFT)
202 #define CBAR_IRPTNDX_SHIFT 24
203 #define CBAR_IRPTNDX_MASK 0xff
205 #define ARM_SMMU_GR1_CBA2R(n) (0x800 + ((n) << 2))
206 #define CBA2R_RW64_32BIT (0 << 0)
207 #define CBA2R_RW64_64BIT (1 << 0)
209 /* Translation context bank */
210 #define ARM_SMMU_CB_BASE(smmu) ((smmu)->base + ((smmu)->size >> 1))
211 #define ARM_SMMU_CB(smmu, n) ((n) * (smmu)->pagesize)
213 #define ARM_SMMU_CB_SCTLR 0x0
214 #define ARM_SMMU_CB_RESUME 0x8
215 #define ARM_SMMU_CB_TTBCR2 0x10
216 #define ARM_SMMU_CB_TTBR0_LO 0x20
217 #define ARM_SMMU_CB_TTBR0_HI 0x24
218 #define ARM_SMMU_CB_TTBCR 0x30
219 #define ARM_SMMU_CB_S1_MAIR0 0x38
220 #define ARM_SMMU_CB_FSR 0x58
221 #define ARM_SMMU_CB_FAR_LO 0x60
222 #define ARM_SMMU_CB_FAR_HI 0x64
223 #define ARM_SMMU_CB_FSYNR0 0x68
224 #define ARM_SMMU_CB_S1_TLBIASID 0x610
226 #define SCTLR_S1_ASIDPNE (1 << 12)
227 #define SCTLR_CFCFG (1 << 7)
228 #define SCTLR_CFIE (1 << 6)
229 #define SCTLR_CFRE (1 << 5)
230 #define SCTLR_E (1 << 4)
231 #define SCTLR_AFE (1 << 2)
232 #define SCTLR_TRE (1 << 1)
233 #define SCTLR_M (1 << 0)
234 #define SCTLR_EAE_SBOP (SCTLR_AFE | SCTLR_TRE)
236 #define RESUME_RETRY (0 << 0)
237 #define RESUME_TERMINATE (1 << 0)
239 #define TTBCR_EAE (1 << 31)
241 #define TTBCR_PASIZE_SHIFT 16
242 #define TTBCR_PASIZE_MASK 0x7
244 #define TTBCR_TG0_4K (0 << 14)
245 #define TTBCR_TG0_64K (1 << 14)
247 #define TTBCR_SH0_SHIFT 12
248 #define TTBCR_SH0_MASK 0x3
249 #define TTBCR_SH_NS 0
250 #define TTBCR_SH_OS 2
251 #define TTBCR_SH_IS 3
253 #define TTBCR_ORGN0_SHIFT 10
254 #define TTBCR_IRGN0_SHIFT 8
255 #define TTBCR_RGN_MASK 0x3
256 #define TTBCR_RGN_NC 0
257 #define TTBCR_RGN_WBWA 1
258 #define TTBCR_RGN_WT 2
259 #define TTBCR_RGN_WB 3
261 #define TTBCR_SL0_SHIFT 6
262 #define TTBCR_SL0_MASK 0x3
263 #define TTBCR_SL0_LVL_2 0
264 #define TTBCR_SL0_LVL_1 1
266 #define TTBCR_T1SZ_SHIFT 16
267 #define TTBCR_T0SZ_SHIFT 0
268 #define TTBCR_SZ_MASK 0xf
270 #define TTBCR2_SEP_SHIFT 15
271 #define TTBCR2_SEP_MASK 0x7
273 #define TTBCR2_PASIZE_SHIFT 0
274 #define TTBCR2_PASIZE_MASK 0x7
276 /* Common definitions for PASize and SEP fields */
277 #define TTBCR2_ADDR_32 0
278 #define TTBCR2_ADDR_36 1
279 #define TTBCR2_ADDR_40 2
280 #define TTBCR2_ADDR_42 3
281 #define TTBCR2_ADDR_44 4
282 #define TTBCR2_ADDR_48 5
284 #define TTBRn_HI_ASID_SHIFT 16
286 #define MAIR_ATTR_SHIFT(n) ((n) << 3)
287 #define MAIR_ATTR_MASK 0xff
288 #define MAIR_ATTR_DEVICE 0x04
289 #define MAIR_ATTR_NC 0x44
290 #define MAIR_ATTR_WBRWA 0xff
291 #define MAIR_ATTR_IDX_NC 0
292 #define MAIR_ATTR_IDX_CACHE 1
293 #define MAIR_ATTR_IDX_DEV 2
295 #define FSR_MULTI (1 << 31)
296 #define FSR_SS (1 << 30)
297 #define FSR_UUT (1 << 8)
298 #define FSR_ASF (1 << 7)
299 #define FSR_TLBLKF (1 << 6)
300 #define FSR_TLBMCF (1 << 5)
301 #define FSR_EF (1 << 4)
302 #define FSR_PF (1 << 3)
303 #define FSR_AFF (1 << 2)
304 #define FSR_TF (1 << 1)
306 #define FSR_IGN (FSR_AFF | FSR_ASF | FSR_TLBMCF | \
308 #define FSR_FAULT (FSR_MULTI | FSR_SS | FSR_UUT | \
309 FSR_EF | FSR_PF | FSR_TF | FSR_IGN)
311 #define FSYNR0_WNR (1 << 4)
313 struct arm_smmu_smr
{
319 struct arm_smmu_master
{
320 struct device_node
*of_node
;
323 * The following is specific to the master's position in the
328 u16 streamids
[MAX_MASTER_STREAMIDS
];
331 * We only need to allocate these on the root SMMU, as we
332 * configure unmatched streams to bypass translation.
334 struct arm_smmu_smr
*smrs
;
337 struct arm_smmu_device
{
339 struct device_node
*parent_of_node
;
343 unsigned long pagesize
;
345 #define ARM_SMMU_FEAT_COHERENT_WALK (1 << 0)
346 #define ARM_SMMU_FEAT_STREAM_MATCH (1 << 1)
347 #define ARM_SMMU_FEAT_TRANS_S1 (1 << 2)
348 #define ARM_SMMU_FEAT_TRANS_S2 (1 << 3)
349 #define ARM_SMMU_FEAT_TRANS_NESTED (1 << 4)
353 u32 num_context_banks
;
354 u32 num_s2_context_banks
;
355 DECLARE_BITMAP(context_map
, ARM_SMMU_MAX_CBS
);
358 u32 num_mapping_groups
;
359 DECLARE_BITMAP(smr_map
, ARM_SMMU_MAX_SMRS
);
361 unsigned long input_size
;
362 unsigned long s1_output_size
;
363 unsigned long s2_output_size
;
366 u32 num_context_irqs
;
369 struct list_head list
;
370 struct rb_root masters
;
373 struct arm_smmu_cfg
{
374 struct arm_smmu_device
*smmu
;
380 #define INVALID_IRPTNDX 0xff
382 #define ARM_SMMU_CB_ASID(cfg) ((cfg)->cbndx)
383 #define ARM_SMMU_CB_VMID(cfg) ((cfg)->cbndx + 1)
385 struct arm_smmu_domain
{
387 * A domain can span across multiple, chained SMMUs and requires
388 * all devices within the domain to follow the same translation
391 struct arm_smmu_device
*leaf_smmu
;
392 struct arm_smmu_cfg root_cfg
;
393 phys_addr_t output_mask
;
398 static DEFINE_SPINLOCK(arm_smmu_devices_lock
);
399 static LIST_HEAD(arm_smmu_devices
);
401 static struct arm_smmu_master
*find_smmu_master(struct arm_smmu_device
*smmu
,
402 struct device_node
*dev_node
)
404 struct rb_node
*node
= smmu
->masters
.rb_node
;
407 struct arm_smmu_master
*master
;
408 master
= container_of(node
, struct arm_smmu_master
, node
);
410 if (dev_node
< master
->of_node
)
411 node
= node
->rb_left
;
412 else if (dev_node
> master
->of_node
)
413 node
= node
->rb_right
;
421 static int insert_smmu_master(struct arm_smmu_device
*smmu
,
422 struct arm_smmu_master
*master
)
424 struct rb_node
**new, *parent
;
426 new = &smmu
->masters
.rb_node
;
429 struct arm_smmu_master
*this;
430 this = container_of(*new, struct arm_smmu_master
, node
);
433 if (master
->of_node
< this->of_node
)
434 new = &((*new)->rb_left
);
435 else if (master
->of_node
> this->of_node
)
436 new = &((*new)->rb_right
);
441 rb_link_node(&master
->node
, parent
, new);
442 rb_insert_color(&master
->node
, &smmu
->masters
);
446 static int register_smmu_master(struct arm_smmu_device
*smmu
,
448 struct of_phandle_args
*masterspec
)
451 struct arm_smmu_master
*master
;
453 master
= find_smmu_master(smmu
, masterspec
->np
);
456 "rejecting multiple registrations for master device %s\n",
457 masterspec
->np
->name
);
461 if (masterspec
->args_count
> MAX_MASTER_STREAMIDS
) {
463 "reached maximum number (%d) of stream IDs for master device %s\n",
464 MAX_MASTER_STREAMIDS
, masterspec
->np
->name
);
468 master
= devm_kzalloc(dev
, sizeof(*master
), GFP_KERNEL
);
472 master
->of_node
= masterspec
->np
;
473 master
->num_streamids
= masterspec
->args_count
;
475 for (i
= 0; i
< master
->num_streamids
; ++i
)
476 master
->streamids
[i
] = masterspec
->args
[i
];
478 return insert_smmu_master(smmu
, master
);
481 static struct arm_smmu_device
*find_parent_smmu(struct arm_smmu_device
*smmu
)
483 struct arm_smmu_device
*parent
;
485 if (!smmu
->parent_of_node
)
488 spin_lock(&arm_smmu_devices_lock
);
489 list_for_each_entry(parent
, &arm_smmu_devices
, list
)
490 if (parent
->dev
->of_node
== smmu
->parent_of_node
)
495 "Failed to find SMMU parent despite parent in DT\n");
497 spin_unlock(&arm_smmu_devices_lock
);
501 static int __arm_smmu_alloc_bitmap(unsigned long *map
, int start
, int end
)
506 idx
= find_next_zero_bit(map
, end
, start
);
509 } while (test_and_set_bit(idx
, map
));
514 static void __arm_smmu_free_bitmap(unsigned long *map
, int idx
)
519 /* Wait for any pending TLB invalidations to complete */
520 static void arm_smmu_tlb_sync(struct arm_smmu_device
*smmu
)
523 void __iomem
*gr0_base
= ARM_SMMU_GR0(smmu
);
525 writel_relaxed(0, gr0_base
+ ARM_SMMU_GR0_sTLBGSYNC
);
526 while (readl_relaxed(gr0_base
+ ARM_SMMU_GR0_sTLBGSTATUS
)
527 & sTLBGSTATUS_GSACTIVE
) {
529 if (++count
== TLB_LOOP_TIMEOUT
) {
530 dev_err_ratelimited(smmu
->dev
,
531 "TLB sync timed out -- SMMU may be deadlocked\n");
538 static void arm_smmu_tlb_inv_context(struct arm_smmu_cfg
*cfg
)
540 struct arm_smmu_device
*smmu
= cfg
->smmu
;
541 void __iomem
*base
= ARM_SMMU_GR0(smmu
);
542 bool stage1
= cfg
->cbar
!= CBAR_TYPE_S2_TRANS
;
545 base
= ARM_SMMU_CB_BASE(smmu
) + ARM_SMMU_CB(smmu
, cfg
->cbndx
);
546 writel_relaxed(ARM_SMMU_CB_ASID(cfg
),
547 base
+ ARM_SMMU_CB_S1_TLBIASID
);
549 base
= ARM_SMMU_GR0(smmu
);
550 writel_relaxed(ARM_SMMU_CB_VMID(cfg
),
551 base
+ ARM_SMMU_GR0_TLBIVMID
);
554 arm_smmu_tlb_sync(smmu
);
557 static irqreturn_t
arm_smmu_context_fault(int irq
, void *dev
)
560 u32 fsr
, far
, fsynr
, resume
;
562 struct iommu_domain
*domain
= dev
;
563 struct arm_smmu_domain
*smmu_domain
= domain
->priv
;
564 struct arm_smmu_cfg
*root_cfg
= &smmu_domain
->root_cfg
;
565 struct arm_smmu_device
*smmu
= root_cfg
->smmu
;
566 void __iomem
*cb_base
;
568 cb_base
= ARM_SMMU_CB_BASE(smmu
) + ARM_SMMU_CB(smmu
, root_cfg
->cbndx
);
569 fsr
= readl_relaxed(cb_base
+ ARM_SMMU_CB_FSR
);
571 if (!(fsr
& FSR_FAULT
))
575 dev_err_ratelimited(smmu
->dev
,
576 "Unexpected context fault (fsr 0x%u)\n",
579 fsynr
= readl_relaxed(cb_base
+ ARM_SMMU_CB_FSYNR0
);
580 flags
= fsynr
& FSYNR0_WNR
? IOMMU_FAULT_WRITE
: IOMMU_FAULT_READ
;
582 far
= readl_relaxed(cb_base
+ ARM_SMMU_CB_FAR_LO
);
585 far
= readl_relaxed(cb_base
+ ARM_SMMU_CB_FAR_HI
);
586 iova
|= ((unsigned long)far
<< 32);
589 if (!report_iommu_fault(domain
, smmu
->dev
, iova
, flags
)) {
591 resume
= RESUME_RETRY
;
593 dev_err_ratelimited(smmu
->dev
,
594 "Unhandled context fault: iova=0x%08lx, fsynr=0x%x, cb=%d\n",
595 iova
, fsynr
, root_cfg
->cbndx
);
597 resume
= RESUME_TERMINATE
;
600 /* Clear the faulting FSR */
601 writel(fsr
, cb_base
+ ARM_SMMU_CB_FSR
);
603 /* Retry or terminate any stalled transactions */
605 writel_relaxed(resume
, cb_base
+ ARM_SMMU_CB_RESUME
);
610 static irqreturn_t
arm_smmu_global_fault(int irq
, void *dev
)
612 u32 gfsr
, gfsynr0
, gfsynr1
, gfsynr2
;
613 struct arm_smmu_device
*smmu
= dev
;
614 void __iomem
*gr0_base
= ARM_SMMU_GR0(smmu
);
616 gfsr
= readl_relaxed(gr0_base
+ ARM_SMMU_GR0_sGFSR
);
620 gfsynr0
= readl_relaxed(gr0_base
+ ARM_SMMU_GR0_sGFSYNR0
);
621 gfsynr1
= readl_relaxed(gr0_base
+ ARM_SMMU_GR0_sGFSYNR1
);
622 gfsynr2
= readl_relaxed(gr0_base
+ ARM_SMMU_GR0_sGFSYNR2
);
624 dev_err_ratelimited(smmu
->dev
,
625 "Unexpected global fault, this could be serious\n");
626 dev_err_ratelimited(smmu
->dev
,
627 "\tGFSR 0x%08x, GFSYNR0 0x%08x, GFSYNR1 0x%08x, GFSYNR2 0x%08x\n",
628 gfsr
, gfsynr0
, gfsynr1
, gfsynr2
);
630 writel(gfsr
, gr0_base
+ ARM_SMMU_GR0_sGFSR
);
634 static void arm_smmu_init_context_bank(struct arm_smmu_domain
*smmu_domain
)
638 struct arm_smmu_cfg
*root_cfg
= &smmu_domain
->root_cfg
;
639 struct arm_smmu_device
*smmu
= root_cfg
->smmu
;
640 void __iomem
*cb_base
, *gr0_base
, *gr1_base
;
642 gr0_base
= ARM_SMMU_GR0(smmu
);
643 gr1_base
= ARM_SMMU_GR1(smmu
);
644 stage1
= root_cfg
->cbar
!= CBAR_TYPE_S2_TRANS
;
645 cb_base
= ARM_SMMU_CB_BASE(smmu
) + ARM_SMMU_CB(smmu
, root_cfg
->cbndx
);
648 reg
= root_cfg
->cbar
;
649 if (smmu
->version
== 1)
650 reg
|= root_cfg
->irptndx
<< CBAR_IRPTNDX_SHIFT
;
652 /* Use the weakest memory type, so it is overridden by the pte */
654 reg
|= (CBAR_S1_MEMATTR_WB
<< CBAR_S1_MEMATTR_SHIFT
);
656 reg
|= ARM_SMMU_CB_VMID(root_cfg
) << CBAR_VMID_SHIFT
;
657 writel_relaxed(reg
, gr1_base
+ ARM_SMMU_GR1_CBAR(root_cfg
->cbndx
));
659 if (smmu
->version
> 1) {
662 reg
= CBA2R_RW64_64BIT
;
664 reg
= CBA2R_RW64_32BIT
;
667 gr1_base
+ ARM_SMMU_GR1_CBA2R(root_cfg
->cbndx
));
670 switch (smmu
->input_size
) {
672 reg
= (TTBCR2_ADDR_32
<< TTBCR2_SEP_SHIFT
);
675 reg
= (TTBCR2_ADDR_36
<< TTBCR2_SEP_SHIFT
);
678 reg
= (TTBCR2_ADDR_40
<< TTBCR2_SEP_SHIFT
);
681 reg
= (TTBCR2_ADDR_42
<< TTBCR2_SEP_SHIFT
);
684 reg
= (TTBCR2_ADDR_44
<< TTBCR2_SEP_SHIFT
);
687 reg
= (TTBCR2_ADDR_48
<< TTBCR2_SEP_SHIFT
);
691 switch (smmu
->s1_output_size
) {
693 reg
|= (TTBCR2_ADDR_32
<< TTBCR2_PASIZE_SHIFT
);
696 reg
|= (TTBCR2_ADDR_36
<< TTBCR2_PASIZE_SHIFT
);
699 reg
|= (TTBCR2_ADDR_40
<< TTBCR2_PASIZE_SHIFT
);
702 reg
|= (TTBCR2_ADDR_42
<< TTBCR2_PASIZE_SHIFT
);
705 reg
|= (TTBCR2_ADDR_44
<< TTBCR2_PASIZE_SHIFT
);
708 reg
|= (TTBCR2_ADDR_48
<< TTBCR2_PASIZE_SHIFT
);
713 writel_relaxed(reg
, cb_base
+ ARM_SMMU_CB_TTBCR2
);
717 reg
= __pa(root_cfg
->pgd
);
718 writel_relaxed(reg
, cb_base
+ ARM_SMMU_CB_TTBR0_LO
);
719 reg
= (phys_addr_t
)__pa(root_cfg
->pgd
) >> 32;
721 reg
|= ARM_SMMU_CB_ASID(root_cfg
) << TTBRn_HI_ASID_SHIFT
;
722 writel_relaxed(reg
, cb_base
+ ARM_SMMU_CB_TTBR0_HI
);
726 * We use long descriptor, with inner-shareable WBWA tables in TTBR0.
728 if (smmu
->version
> 1) {
729 if (PAGE_SIZE
== SZ_4K
)
735 switch (smmu
->s2_output_size
) {
737 reg
|= (TTBCR2_ADDR_32
<< TTBCR_PASIZE_SHIFT
);
740 reg
|= (TTBCR2_ADDR_36
<< TTBCR_PASIZE_SHIFT
);
743 reg
|= (TTBCR2_ADDR_40
<< TTBCR_PASIZE_SHIFT
);
746 reg
|= (TTBCR2_ADDR_42
<< TTBCR_PASIZE_SHIFT
);
749 reg
|= (TTBCR2_ADDR_44
<< TTBCR_PASIZE_SHIFT
);
752 reg
|= (TTBCR2_ADDR_48
<< TTBCR_PASIZE_SHIFT
);
756 reg
|= (64 - smmu
->s1_output_size
) << TTBCR_T0SZ_SHIFT
;
763 (TTBCR_SH_IS
<< TTBCR_SH0_SHIFT
) |
764 (TTBCR_RGN_WBWA
<< TTBCR_ORGN0_SHIFT
) |
765 (TTBCR_RGN_WBWA
<< TTBCR_IRGN0_SHIFT
) |
766 (TTBCR_SL0_LVL_1
<< TTBCR_SL0_SHIFT
);
767 writel_relaxed(reg
, cb_base
+ ARM_SMMU_CB_TTBCR
);
769 /* MAIR0 (stage-1 only) */
771 reg
= (MAIR_ATTR_NC
<< MAIR_ATTR_SHIFT(MAIR_ATTR_IDX_NC
)) |
772 (MAIR_ATTR_WBRWA
<< MAIR_ATTR_SHIFT(MAIR_ATTR_IDX_CACHE
)) |
773 (MAIR_ATTR_DEVICE
<< MAIR_ATTR_SHIFT(MAIR_ATTR_IDX_DEV
));
774 writel_relaxed(reg
, cb_base
+ ARM_SMMU_CB_S1_MAIR0
);
778 reg
= SCTLR_CFCFG
| SCTLR_CFIE
| SCTLR_CFRE
| SCTLR_M
| SCTLR_EAE_SBOP
;
780 reg
|= SCTLR_S1_ASIDPNE
;
784 writel_relaxed(reg
, cb_base
+ ARM_SMMU_CB_SCTLR
);
787 static int arm_smmu_init_domain_context(struct iommu_domain
*domain
,
791 struct arm_smmu_domain
*smmu_domain
= domain
->priv
;
792 struct arm_smmu_cfg
*root_cfg
= &smmu_domain
->root_cfg
;
793 struct arm_smmu_device
*smmu
, *parent
;
796 * Walk the SMMU chain to find the root device for this chain.
797 * We assume that no masters have translations which terminate
798 * early, and therefore check that the root SMMU does indeed have
799 * a StreamID for the master in question.
801 parent
= dev
->archdata
.iommu
;
802 smmu_domain
->output_mask
= -1;
805 smmu_domain
->output_mask
&= (1ULL << smmu
->s2_output_size
) - 1;
806 } while ((parent
= find_parent_smmu(smmu
)));
808 if (!find_smmu_master(smmu
, dev
->of_node
)) {
809 dev_err(dev
, "unable to find root SMMU for device\n");
813 if (smmu
->features
& ARM_SMMU_FEAT_TRANS_NESTED
) {
815 * We will likely want to change this if/when KVM gets
818 root_cfg
->cbar
= CBAR_TYPE_S1_TRANS_S2_BYPASS
;
819 start
= smmu
->num_s2_context_banks
;
820 } else if (smmu
->features
& ARM_SMMU_FEAT_TRANS_S2
) {
821 root_cfg
->cbar
= CBAR_TYPE_S2_TRANS
;
824 root_cfg
->cbar
= CBAR_TYPE_S1_TRANS_S2_BYPASS
;
825 start
= smmu
->num_s2_context_banks
;
828 ret
= __arm_smmu_alloc_bitmap(smmu
->context_map
, start
,
829 smmu
->num_context_banks
);
830 if (IS_ERR_VALUE(ret
))
833 root_cfg
->cbndx
= ret
;
834 if (smmu
->version
== 1) {
835 root_cfg
->irptndx
= atomic_inc_return(&smmu
->irptndx
);
836 root_cfg
->irptndx
%= smmu
->num_context_irqs
;
838 root_cfg
->irptndx
= root_cfg
->cbndx
;
841 irq
= smmu
->irqs
[smmu
->num_global_irqs
+ root_cfg
->irptndx
];
842 ret
= request_irq(irq
, arm_smmu_context_fault
, IRQF_SHARED
,
843 "arm-smmu-context-fault", domain
);
844 if (IS_ERR_VALUE(ret
)) {
845 dev_err(smmu
->dev
, "failed to request context IRQ %d (%u)\n",
846 root_cfg
->irptndx
, irq
);
847 root_cfg
->irptndx
= INVALID_IRPTNDX
;
848 goto out_free_context
;
851 root_cfg
->smmu
= smmu
;
852 arm_smmu_init_context_bank(smmu_domain
);
856 __arm_smmu_free_bitmap(smmu
->context_map
, root_cfg
->cbndx
);
860 static void arm_smmu_destroy_domain_context(struct iommu_domain
*domain
)
862 struct arm_smmu_domain
*smmu_domain
= domain
->priv
;
863 struct arm_smmu_cfg
*root_cfg
= &smmu_domain
->root_cfg
;
864 struct arm_smmu_device
*smmu
= root_cfg
->smmu
;
865 void __iomem
*cb_base
;
871 /* Disable the context bank and nuke the TLB before freeing it. */
872 cb_base
= ARM_SMMU_CB_BASE(smmu
) + ARM_SMMU_CB(smmu
, root_cfg
->cbndx
);
873 writel_relaxed(0, cb_base
+ ARM_SMMU_CB_SCTLR
);
874 arm_smmu_tlb_inv_context(root_cfg
);
876 if (root_cfg
->irptndx
!= INVALID_IRPTNDX
) {
877 irq
= smmu
->irqs
[smmu
->num_global_irqs
+ root_cfg
->irptndx
];
878 free_irq(irq
, domain
);
881 __arm_smmu_free_bitmap(smmu
->context_map
, root_cfg
->cbndx
);
884 static int arm_smmu_domain_init(struct iommu_domain
*domain
)
886 struct arm_smmu_domain
*smmu_domain
;
890 * Allocate the domain and initialise some of its data structures.
891 * We can't really do anything meaningful until we've added a
894 smmu_domain
= kzalloc(sizeof(*smmu_domain
), GFP_KERNEL
);
898 pgd
= kzalloc(PTRS_PER_PGD
* sizeof(pgd_t
), GFP_KERNEL
);
900 goto out_free_domain
;
901 smmu_domain
->root_cfg
.pgd
= pgd
;
903 spin_lock_init(&smmu_domain
->lock
);
904 domain
->priv
= smmu_domain
;
912 static void arm_smmu_free_ptes(pmd_t
*pmd
)
914 pgtable_t table
= pmd_pgtable(*pmd
);
915 pgtable_page_dtor(table
);
919 static void arm_smmu_free_pmds(pud_t
*pud
)
922 pmd_t
*pmd
, *pmd_base
= pmd_offset(pud
, 0);
925 for (i
= 0; i
< PTRS_PER_PMD
; ++i
) {
929 arm_smmu_free_ptes(pmd
);
933 pmd_free(NULL
, pmd_base
);
936 static void arm_smmu_free_puds(pgd_t
*pgd
)
939 pud_t
*pud
, *pud_base
= pud_offset(pgd
, 0);
942 for (i
= 0; i
< PTRS_PER_PUD
; ++i
) {
946 arm_smmu_free_pmds(pud
);
950 pud_free(NULL
, pud_base
);
953 static void arm_smmu_free_pgtables(struct arm_smmu_domain
*smmu_domain
)
956 struct arm_smmu_cfg
*root_cfg
= &smmu_domain
->root_cfg
;
957 pgd_t
*pgd
, *pgd_base
= root_cfg
->pgd
;
960 * Recursively free the page tables for this domain. We don't
961 * care about speculative TLB filling, because the TLB will be
962 * nuked next time this context bank is re-allocated and no devices
963 * currently map to these tables.
966 for (i
= 0; i
< PTRS_PER_PGD
; ++i
) {
969 arm_smmu_free_puds(pgd
);
976 static void arm_smmu_domain_destroy(struct iommu_domain
*domain
)
978 struct arm_smmu_domain
*smmu_domain
= domain
->priv
;
981 * Free the domain resources. We assume that all devices have
982 * already been detached.
984 arm_smmu_destroy_domain_context(domain
);
985 arm_smmu_free_pgtables(smmu_domain
);
989 static int arm_smmu_master_configure_smrs(struct arm_smmu_device
*smmu
,
990 struct arm_smmu_master
*master
)
993 struct arm_smmu_smr
*smrs
;
994 void __iomem
*gr0_base
= ARM_SMMU_GR0(smmu
);
996 if (!(smmu
->features
& ARM_SMMU_FEAT_STREAM_MATCH
))
1002 smrs
= kmalloc(sizeof(*smrs
) * master
->num_streamids
, GFP_KERNEL
);
1004 dev_err(smmu
->dev
, "failed to allocate %d SMRs for master %s\n",
1005 master
->num_streamids
, master
->of_node
->name
);
1009 /* Allocate the SMRs on the root SMMU */
1010 for (i
= 0; i
< master
->num_streamids
; ++i
) {
1011 int idx
= __arm_smmu_alloc_bitmap(smmu
->smr_map
, 0,
1012 smmu
->num_mapping_groups
);
1013 if (IS_ERR_VALUE(idx
)) {
1014 dev_err(smmu
->dev
, "failed to allocate free SMR\n");
1018 smrs
[i
] = (struct arm_smmu_smr
) {
1020 .mask
= 0, /* We don't currently share SMRs */
1021 .id
= master
->streamids
[i
],
1025 /* It worked! Now, poke the actual hardware */
1026 for (i
= 0; i
< master
->num_streamids
; ++i
) {
1027 u32 reg
= SMR_VALID
| smrs
[i
].id
<< SMR_ID_SHIFT
|
1028 smrs
[i
].mask
<< SMR_MASK_SHIFT
;
1029 writel_relaxed(reg
, gr0_base
+ ARM_SMMU_GR0_SMR(smrs
[i
].idx
));
1032 master
->smrs
= smrs
;
1037 __arm_smmu_free_bitmap(smmu
->smr_map
, smrs
[i
].idx
);
1042 static void arm_smmu_master_free_smrs(struct arm_smmu_device
*smmu
,
1043 struct arm_smmu_master
*master
)
1046 void __iomem
*gr0_base
= ARM_SMMU_GR0(smmu
);
1047 struct arm_smmu_smr
*smrs
= master
->smrs
;
1049 /* Invalidate the SMRs before freeing back to the allocator */
1050 for (i
= 0; i
< master
->num_streamids
; ++i
) {
1051 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
);
1056 master
->smrs
= NULL
;
1060 static void arm_smmu_bypass_stream_mapping(struct arm_smmu_device
*smmu
,
1061 struct arm_smmu_master
*master
)
1064 void __iomem
*gr0_base
= ARM_SMMU_GR0(smmu
);
1066 for (i
= 0; i
< master
->num_streamids
; ++i
) {
1067 u16 sid
= master
->streamids
[i
];
1068 writel_relaxed(S2CR_TYPE_BYPASS
,
1069 gr0_base
+ ARM_SMMU_GR0_S2CR(sid
));
1073 static int arm_smmu_domain_add_master(struct arm_smmu_domain
*smmu_domain
,
1074 struct arm_smmu_master
*master
)
1077 struct arm_smmu_device
*parent
, *smmu
= smmu_domain
->root_cfg
.smmu
;
1078 void __iomem
*gr0_base
= ARM_SMMU_GR0(smmu
);
1080 ret
= arm_smmu_master_configure_smrs(smmu
, master
);
1084 /* Bypass the leaves */
1085 smmu
= smmu_domain
->leaf_smmu
;
1086 while ((parent
= find_parent_smmu(smmu
))) {
1088 * We won't have a StreamID match for anything but the root
1089 * smmu, so we only need to worry about StreamID indexing,
1090 * where we must install bypass entries in the S2CRs.
1092 if (smmu
->features
& ARM_SMMU_FEAT_STREAM_MATCH
)
1095 arm_smmu_bypass_stream_mapping(smmu
, master
);
1099 /* Now we're at the root, time to point at our context bank */
1100 for (i
= 0; i
< master
->num_streamids
; ++i
) {
1102 idx
= master
->smrs
? master
->smrs
[i
].idx
: master
->streamids
[i
];
1103 s2cr
= (S2CR_TYPE_TRANS
<< S2CR_TYPE_SHIFT
) |
1104 (smmu_domain
->root_cfg
.cbndx
<< S2CR_CBNDX_SHIFT
);
1105 writel_relaxed(s2cr
, gr0_base
+ ARM_SMMU_GR0_S2CR(idx
));
1111 static void arm_smmu_domain_remove_master(struct arm_smmu_domain
*smmu_domain
,
1112 struct arm_smmu_master
*master
)
1114 struct arm_smmu_device
*smmu
= smmu_domain
->root_cfg
.smmu
;
1117 * We *must* clear the S2CR first, because freeing the SMR means
1118 * that it can be re-allocated immediately.
1120 arm_smmu_bypass_stream_mapping(smmu
, master
);
1121 arm_smmu_master_free_smrs(smmu
, master
);
1124 static int arm_smmu_attach_dev(struct iommu_domain
*domain
, struct device
*dev
)
1127 struct arm_smmu_domain
*smmu_domain
= domain
->priv
;
1128 struct arm_smmu_device
*device_smmu
= dev
->archdata
.iommu
;
1129 struct arm_smmu_master
*master
;
1132 dev_err(dev
, "cannot attach to SMMU, is it on the same bus?\n");
1137 * Sanity check the domain. We don't currently support domains
1138 * that cross between different SMMU chains.
1140 spin_lock(&smmu_domain
->lock
);
1141 if (!smmu_domain
->leaf_smmu
) {
1142 /* Now that we have a master, we can finalise the domain */
1143 ret
= arm_smmu_init_domain_context(domain
, dev
);
1144 if (IS_ERR_VALUE(ret
))
1147 smmu_domain
->leaf_smmu
= device_smmu
;
1148 } else if (smmu_domain
->leaf_smmu
!= device_smmu
) {
1150 "cannot attach to SMMU %s whilst already attached to domain on SMMU %s\n",
1151 dev_name(smmu_domain
->leaf_smmu
->dev
),
1152 dev_name(device_smmu
->dev
));
1155 spin_unlock(&smmu_domain
->lock
);
1157 /* Looks ok, so add the device to the domain */
1158 master
= find_smmu_master(smmu_domain
->leaf_smmu
, dev
->of_node
);
1162 return arm_smmu_domain_add_master(smmu_domain
, master
);
1165 spin_unlock(&smmu_domain
->lock
);
1169 static void arm_smmu_detach_dev(struct iommu_domain
*domain
, struct device
*dev
)
1171 struct arm_smmu_domain
*smmu_domain
= domain
->priv
;
1172 struct arm_smmu_master
*master
;
1174 master
= find_smmu_master(smmu_domain
->leaf_smmu
, dev
->of_node
);
1176 arm_smmu_domain_remove_master(smmu_domain
, master
);
1179 static void arm_smmu_flush_pgtable(struct arm_smmu_device
*smmu
, void *addr
,
1182 unsigned long offset
= (unsigned long)addr
& ~PAGE_MASK
;
1185 * If the SMMU can't walk tables in the CPU caches, treat them
1186 * like non-coherent DMA since we need to flush the new entries
1187 * all the way out to memory. There's no possibility of recursion
1188 * here as the SMMU table walker will not be wired through another
1191 if (!(smmu
->features
& ARM_SMMU_FEAT_COHERENT_WALK
))
1192 dma_map_page(smmu
->dev
, virt_to_page(addr
), offset
, size
,
1196 static bool arm_smmu_pte_is_contiguous_range(unsigned long addr
,
1199 return !(addr
& ~ARM_SMMU_PTE_CONT_MASK
) &&
1200 (addr
+ ARM_SMMU_PTE_CONT_SIZE
<= end
);
1203 static int arm_smmu_alloc_init_pte(struct arm_smmu_device
*smmu
, pmd_t
*pmd
,
1204 unsigned long addr
, unsigned long end
,
1205 unsigned long pfn
, int flags
, int stage
)
1208 pteval_t pteval
= ARM_SMMU_PTE_PAGE
| ARM_SMMU_PTE_AF
;
1210 if (pmd_none(*pmd
)) {
1211 /* Allocate a new set of tables */
1212 pgtable_t table
= alloc_page(PGALLOC_GFP
);
1216 arm_smmu_flush_pgtable(smmu
, page_address(table
),
1217 ARM_SMMU_PTE_HWTABLE_SIZE
);
1218 if (!pgtable_page_ctor(table
)) {
1222 pmd_populate(NULL
, pmd
, table
);
1223 arm_smmu_flush_pgtable(smmu
, pmd
, sizeof(*pmd
));
1227 pteval
|= ARM_SMMU_PTE_AP_UNPRIV
| ARM_SMMU_PTE_nG
;
1228 if (!(flags
& IOMMU_WRITE
) && (flags
& IOMMU_READ
))
1229 pteval
|= ARM_SMMU_PTE_AP_RDONLY
;
1231 if (flags
& IOMMU_CACHE
)
1232 pteval
|= (MAIR_ATTR_IDX_CACHE
<<
1233 ARM_SMMU_PTE_ATTRINDX_SHIFT
);
1235 pteval
|= ARM_SMMU_PTE_HAP_FAULT
;
1236 if (flags
& IOMMU_READ
)
1237 pteval
|= ARM_SMMU_PTE_HAP_READ
;
1238 if (flags
& IOMMU_WRITE
)
1239 pteval
|= ARM_SMMU_PTE_HAP_WRITE
;
1240 if (flags
& IOMMU_CACHE
)
1241 pteval
|= ARM_SMMU_PTE_MEMATTR_OIWB
;
1243 pteval
|= ARM_SMMU_PTE_MEMATTR_NC
;
1246 /* If no access, create a faulting entry to avoid TLB fills */
1247 if (!(flags
& (IOMMU_READ
| IOMMU_WRITE
)))
1248 pteval
&= ~ARM_SMMU_PTE_PAGE
;
1250 pteval
|= ARM_SMMU_PTE_SH_IS
;
1251 start
= pmd_page_vaddr(*pmd
) + pte_index(addr
);
1255 * Install the page table entries. This is fairly complicated
1256 * since we attempt to make use of the contiguous hint in the
1257 * ptes where possible. The contiguous hint indicates a series
1258 * of ARM_SMMU_PTE_CONT_ENTRIES ptes mapping a physically
1259 * contiguous region with the following constraints:
1261 * - The region start is aligned to ARM_SMMU_PTE_CONT_SIZE
1262 * - Each pte in the region has the contiguous hint bit set
1264 * This complicates unmapping (also handled by this code, when
1265 * neither IOMMU_READ or IOMMU_WRITE are set) because it is
1266 * possible, yet highly unlikely, that a client may unmap only
1267 * part of a contiguous range. This requires clearing of the
1268 * contiguous hint bits in the range before installing the new
1271 * Note that re-mapping an address range without first unmapping
1272 * it is not supported, so TLB invalidation is not required here
1273 * and is instead performed at unmap and domain-init time.
1277 pteval
&= ~ARM_SMMU_PTE_CONT
;
1279 if (arm_smmu_pte_is_contiguous_range(addr
, end
)) {
1280 i
= ARM_SMMU_PTE_CONT_ENTRIES
;
1281 pteval
|= ARM_SMMU_PTE_CONT
;
1282 } else if (pte_val(*pte
) &
1283 (ARM_SMMU_PTE_CONT
| ARM_SMMU_PTE_PAGE
)) {
1286 unsigned long idx
= pte_index(addr
);
1288 idx
&= ~(ARM_SMMU_PTE_CONT_ENTRIES
- 1);
1289 cont_start
= pmd_page_vaddr(*pmd
) + idx
;
1290 for (j
= 0; j
< ARM_SMMU_PTE_CONT_ENTRIES
; ++j
)
1291 pte_val(*(cont_start
+ j
)) &= ~ARM_SMMU_PTE_CONT
;
1293 arm_smmu_flush_pgtable(smmu
, cont_start
,
1295 ARM_SMMU_PTE_CONT_ENTRIES
);
1299 *pte
= pfn_pte(pfn
, __pgprot(pteval
));
1300 } while (pte
++, pfn
++, addr
+= PAGE_SIZE
, --i
);
1301 } while (addr
!= end
);
1303 arm_smmu_flush_pgtable(smmu
, start
, sizeof(*pte
) * (pte
- start
));
1307 static int arm_smmu_alloc_init_pmd(struct arm_smmu_device
*smmu
, pud_t
*pud
,
1308 unsigned long addr
, unsigned long end
,
1309 phys_addr_t phys
, int flags
, int stage
)
1313 unsigned long next
, pfn
= __phys_to_pfn(phys
);
1315 #ifndef __PAGETABLE_PMD_FOLDED
1316 if (pud_none(*pud
)) {
1317 pmd
= pmd_alloc_one(NULL
, addr
);
1322 pmd
= pmd_offset(pud
, addr
);
1325 next
= pmd_addr_end(addr
, end
);
1326 ret
= arm_smmu_alloc_init_pte(smmu
, pmd
, addr
, end
, pfn
,
1328 pud_populate(NULL
, pud
, pmd
);
1329 arm_smmu_flush_pgtable(smmu
, pud
, sizeof(*pud
));
1330 phys
+= next
- addr
;
1331 } while (pmd
++, addr
= next
, addr
< end
);
1336 static int arm_smmu_alloc_init_pud(struct arm_smmu_device
*smmu
, pgd_t
*pgd
,
1337 unsigned long addr
, unsigned long end
,
1338 phys_addr_t phys
, int flags
, int stage
)
1344 #ifndef __PAGETABLE_PUD_FOLDED
1345 if (pgd_none(*pgd
)) {
1346 pud
= pud_alloc_one(NULL
, addr
);
1351 pud
= pud_offset(pgd
, addr
);
1354 next
= pud_addr_end(addr
, end
);
1355 ret
= arm_smmu_alloc_init_pmd(smmu
, pud
, addr
, next
, phys
,
1357 pgd_populate(NULL
, pud
, pgd
);
1358 arm_smmu_flush_pgtable(smmu
, pgd
, sizeof(*pgd
));
1359 phys
+= next
- addr
;
1360 } while (pud
++, addr
= next
, addr
< end
);
1365 static int arm_smmu_handle_mapping(struct arm_smmu_domain
*smmu_domain
,
1366 unsigned long iova
, phys_addr_t paddr
,
1367 size_t size
, int flags
)
1371 phys_addr_t input_mask
, output_mask
;
1372 struct arm_smmu_cfg
*root_cfg
= &smmu_domain
->root_cfg
;
1373 pgd_t
*pgd
= root_cfg
->pgd
;
1374 struct arm_smmu_device
*smmu
= root_cfg
->smmu
;
1376 if (root_cfg
->cbar
== CBAR_TYPE_S2_TRANS
) {
1378 output_mask
= (1ULL << smmu
->s2_output_size
) - 1;
1381 output_mask
= (1ULL << smmu
->s1_output_size
) - 1;
1387 if (size
& ~PAGE_MASK
)
1390 input_mask
= (1ULL << smmu
->input_size
) - 1;
1391 if ((phys_addr_t
)iova
& ~input_mask
)
1394 if (paddr
& ~output_mask
)
1397 spin_lock(&smmu_domain
->lock
);
1398 pgd
+= pgd_index(iova
);
1401 unsigned long next
= pgd_addr_end(iova
, end
);
1403 ret
= arm_smmu_alloc_init_pud(smmu
, pgd
, iova
, next
, paddr
,
1408 paddr
+= next
- iova
;
1410 } while (pgd
++, iova
!= end
);
1413 spin_unlock(&smmu_domain
->lock
);
1415 /* Ensure new page tables are visible to the hardware walker */
1416 if (smmu
->features
& ARM_SMMU_FEAT_COHERENT_WALK
)
1422 static int arm_smmu_map(struct iommu_domain
*domain
, unsigned long iova
,
1423 phys_addr_t paddr
, size_t size
, int flags
)
1425 struct arm_smmu_domain
*smmu_domain
= domain
->priv
;
1426 struct arm_smmu_device
*smmu
= smmu_domain
->leaf_smmu
;
1428 if (!smmu_domain
|| !smmu
)
1431 /* Check for silent address truncation up the SMMU chain. */
1432 if ((phys_addr_t
)iova
& ~smmu_domain
->output_mask
)
1435 return arm_smmu_handle_mapping(smmu_domain
, iova
, paddr
, size
, flags
);
1438 static size_t arm_smmu_unmap(struct iommu_domain
*domain
, unsigned long iova
,
1442 struct arm_smmu_domain
*smmu_domain
= domain
->priv
;
1444 ret
= arm_smmu_handle_mapping(smmu_domain
, iova
, 0, size
, 0);
1445 arm_smmu_tlb_inv_context(&smmu_domain
->root_cfg
);
1446 return ret
? ret
: size
;
1449 static phys_addr_t
arm_smmu_iova_to_phys(struct iommu_domain
*domain
,
1456 struct arm_smmu_domain
*smmu_domain
= domain
->priv
;
1457 struct arm_smmu_cfg
*root_cfg
= &smmu_domain
->root_cfg
;
1458 struct arm_smmu_device
*smmu
= root_cfg
->smmu
;
1460 spin_lock(&smmu_domain
->lock
);
1461 pgd
= root_cfg
->pgd
;
1465 pgd
+= pgd_index(iova
);
1466 if (pgd_none_or_clear_bad(pgd
))
1469 pud
= pud_offset(pgd
, iova
);
1470 if (pud_none_or_clear_bad(pud
))
1473 pmd
= pmd_offset(pud
, iova
);
1474 if (pmd_none_or_clear_bad(pmd
))
1477 pte
= pmd_page_vaddr(*pmd
) + pte_index(iova
);
1481 spin_unlock(&smmu_domain
->lock
);
1482 return __pfn_to_phys(pte_pfn(*pte
)) | (iova
& ~PAGE_MASK
);
1485 spin_unlock(&smmu_domain
->lock
);
1487 "invalid (corrupt?) page tables detected for iova 0x%llx\n",
1488 (unsigned long long)iova
);
1492 static int arm_smmu_domain_has_cap(struct iommu_domain
*domain
,
1495 unsigned long caps
= 0;
1496 struct arm_smmu_domain
*smmu_domain
= domain
->priv
;
1498 if (smmu_domain
->root_cfg
.smmu
->features
& ARM_SMMU_FEAT_COHERENT_WALK
)
1499 caps
|= IOMMU_CAP_CACHE_COHERENCY
;
1501 return !!(cap
& caps
);
1504 static int arm_smmu_add_device(struct device
*dev
)
1506 struct arm_smmu_device
*child
, *parent
, *smmu
;
1507 struct arm_smmu_master
*master
= NULL
;
1509 spin_lock(&arm_smmu_devices_lock
);
1510 list_for_each_entry(parent
, &arm_smmu_devices
, list
) {
1513 /* Try to find a child of the current SMMU. */
1514 list_for_each_entry(child
, &arm_smmu_devices
, list
) {
1515 if (child
->parent_of_node
== parent
->dev
->of_node
) {
1516 /* Does the child sit above our master? */
1517 master
= find_smmu_master(child
, dev
->of_node
);
1525 /* We found some children, so keep searching. */
1531 master
= find_smmu_master(smmu
, dev
->of_node
);
1535 spin_unlock(&arm_smmu_devices_lock
);
1540 dev
->archdata
.iommu
= smmu
;
1544 static void arm_smmu_remove_device(struct device
*dev
)
1546 dev
->archdata
.iommu
= NULL
;
1549 static struct iommu_ops arm_smmu_ops
= {
1550 .domain_init
= arm_smmu_domain_init
,
1551 .domain_destroy
= arm_smmu_domain_destroy
,
1552 .attach_dev
= arm_smmu_attach_dev
,
1553 .detach_dev
= arm_smmu_detach_dev
,
1554 .map
= arm_smmu_map
,
1555 .unmap
= arm_smmu_unmap
,
1556 .iova_to_phys
= arm_smmu_iova_to_phys
,
1557 .domain_has_cap
= arm_smmu_domain_has_cap
,
1558 .add_device
= arm_smmu_add_device
,
1559 .remove_device
= arm_smmu_remove_device
,
1560 .pgsize_bitmap
= (SECTION_SIZE
|
1561 ARM_SMMU_PTE_CONT_SIZE
|
1565 static void arm_smmu_device_reset(struct arm_smmu_device
*smmu
)
1567 void __iomem
*gr0_base
= ARM_SMMU_GR0(smmu
);
1568 void __iomem
*cb_base
;
1572 /* Clear Global FSR */
1573 reg
= readl_relaxed(gr0_base
+ ARM_SMMU_GR0_sGFSR
);
1574 writel(reg
, gr0_base
+ ARM_SMMU_GR0_sGFSR
);
1576 /* Mark all SMRn as invalid and all S2CRn as bypass */
1577 for (i
= 0; i
< smmu
->num_mapping_groups
; ++i
) {
1578 writel_relaxed(~SMR_VALID
, gr0_base
+ ARM_SMMU_GR0_SMR(i
));
1579 writel_relaxed(S2CR_TYPE_BYPASS
, gr0_base
+ ARM_SMMU_GR0_S2CR(i
));
1582 /* Make sure all context banks are disabled and clear CB_FSR */
1583 for (i
= 0; i
< smmu
->num_context_banks
; ++i
) {
1584 cb_base
= ARM_SMMU_CB_BASE(smmu
) + ARM_SMMU_CB(smmu
, i
);
1585 writel_relaxed(0, cb_base
+ ARM_SMMU_CB_SCTLR
);
1586 writel_relaxed(FSR_FAULT
, cb_base
+ ARM_SMMU_CB_FSR
);
1589 /* Invalidate the TLB, just in case */
1590 writel_relaxed(0, gr0_base
+ ARM_SMMU_GR0_STLBIALL
);
1591 writel_relaxed(0, gr0_base
+ ARM_SMMU_GR0_TLBIALLH
);
1592 writel_relaxed(0, gr0_base
+ ARM_SMMU_GR0_TLBIALLNSNH
);
1594 reg
= readl_relaxed(gr0_base
+ ARM_SMMU_GR0_sCR0
);
1596 /* Enable fault reporting */
1597 reg
|= (sCR0_GFRE
| sCR0_GFIE
| sCR0_GCFGFRE
| sCR0_GCFGFIE
);
1599 /* Disable TLB broadcasting. */
1600 reg
|= (sCR0_VMIDPNE
| sCR0_PTM
);
1602 /* Enable client access, but bypass when no mapping is found */
1603 reg
&= ~(sCR0_CLIENTPD
| sCR0_USFCFG
);
1605 /* Disable forced broadcasting */
1608 /* Don't upgrade barriers */
1609 reg
&= ~(sCR0_BSU_MASK
<< sCR0_BSU_SHIFT
);
1611 /* Push the button */
1612 arm_smmu_tlb_sync(smmu
);
1613 writel_relaxed(reg
, gr0_base
+ ARM_SMMU_GR0_sCR0
);
1616 static int arm_smmu_id_size_to_bits(int size
)
1635 static int arm_smmu_device_cfg_probe(struct arm_smmu_device
*smmu
)
1638 void __iomem
*gr0_base
= ARM_SMMU_GR0(smmu
);
1641 dev_notice(smmu
->dev
, "probing hardware configuration...\n");
1644 id
= readl_relaxed(gr0_base
+ ARM_SMMU_GR0_PIDR2
);
1645 smmu
->version
= ((id
>> PIDR2_ARCH_SHIFT
) & PIDR2_ARCH_MASK
) + 1;
1646 dev_notice(smmu
->dev
, "SMMUv%d with:\n", smmu
->version
);
1649 id
= readl_relaxed(gr0_base
+ ARM_SMMU_GR0_ID0
);
1650 #ifndef CONFIG_64BIT
1651 if (((id
>> ID0_PTFS_SHIFT
) & ID0_PTFS_MASK
) == ID0_PTFS_V8_ONLY
) {
1652 dev_err(smmu
->dev
, "\tno v7 descriptor support!\n");
1656 if (id
& ID0_S1TS
) {
1657 smmu
->features
|= ARM_SMMU_FEAT_TRANS_S1
;
1658 dev_notice(smmu
->dev
, "\tstage 1 translation\n");
1661 if (id
& ID0_S2TS
) {
1662 smmu
->features
|= ARM_SMMU_FEAT_TRANS_S2
;
1663 dev_notice(smmu
->dev
, "\tstage 2 translation\n");
1667 smmu
->features
|= ARM_SMMU_FEAT_TRANS_NESTED
;
1668 dev_notice(smmu
->dev
, "\tnested translation\n");
1671 if (!(smmu
->features
&
1672 (ARM_SMMU_FEAT_TRANS_S1
| ARM_SMMU_FEAT_TRANS_S2
|
1673 ARM_SMMU_FEAT_TRANS_NESTED
))) {
1674 dev_err(smmu
->dev
, "\tno translation support!\n");
1678 if (id
& ID0_CTTW
) {
1679 smmu
->features
|= ARM_SMMU_FEAT_COHERENT_WALK
;
1680 dev_notice(smmu
->dev
, "\tcoherent table walk\n");
1686 smmu
->features
|= ARM_SMMU_FEAT_STREAM_MATCH
;
1687 smmu
->num_mapping_groups
= (id
>> ID0_NUMSMRG_SHIFT
) &
1689 if (smmu
->num_mapping_groups
== 0) {
1691 "stream-matching supported, but no SMRs present!\n");
1695 smr
= SMR_MASK_MASK
<< SMR_MASK_SHIFT
;
1696 smr
|= (SMR_ID_MASK
<< SMR_ID_SHIFT
);
1697 writel_relaxed(smr
, gr0_base
+ ARM_SMMU_GR0_SMR(0));
1698 smr
= readl_relaxed(gr0_base
+ ARM_SMMU_GR0_SMR(0));
1700 mask
= (smr
>> SMR_MASK_SHIFT
) & SMR_MASK_MASK
;
1701 sid
= (smr
>> SMR_ID_SHIFT
) & SMR_ID_MASK
;
1702 if ((mask
& sid
) != sid
) {
1704 "SMR mask bits (0x%x) insufficient for ID field (0x%x)\n",
1709 dev_notice(smmu
->dev
,
1710 "\tstream matching with %u register groups, mask 0x%x",
1711 smmu
->num_mapping_groups
, mask
);
1715 id
= readl_relaxed(gr0_base
+ ARM_SMMU_GR0_ID1
);
1716 smmu
->pagesize
= (id
& ID1_PAGESIZE
) ? SZ_64K
: SZ_4K
;
1718 /* Check for size mismatch of SMMU address space from mapped region */
1719 size
= 1 << (((id
>> ID1_NUMPAGENDXB_SHIFT
) & ID1_NUMPAGENDXB_MASK
) + 1);
1720 size
*= (smmu
->pagesize
<< 1);
1721 if (smmu
->size
!= size
)
1722 dev_warn(smmu
->dev
, "SMMU address space size (0x%lx) differs "
1723 "from mapped region size (0x%lx)!\n", size
, smmu
->size
);
1725 smmu
->num_s2_context_banks
= (id
>> ID1_NUMS2CB_SHIFT
) &
1727 smmu
->num_context_banks
= (id
>> ID1_NUMCB_SHIFT
) & ID1_NUMCB_MASK
;
1728 if (smmu
->num_s2_context_banks
> smmu
->num_context_banks
) {
1729 dev_err(smmu
->dev
, "impossible number of S2 context banks!\n");
1732 dev_notice(smmu
->dev
, "\t%u context banks (%u stage-2 only)\n",
1733 smmu
->num_context_banks
, smmu
->num_s2_context_banks
);
1736 id
= readl_relaxed(gr0_base
+ ARM_SMMU_GR0_ID2
);
1737 size
= arm_smmu_id_size_to_bits((id
>> ID2_IAS_SHIFT
) & ID2_IAS_MASK
);
1740 * Stage-1 output limited by stage-2 input size due to pgd
1741 * allocation (PTRS_PER_PGD).
1744 /* Current maximum output size of 39 bits */
1745 smmu
->s1_output_size
= min(39UL, size
);
1747 smmu
->s1_output_size
= min(32UL, size
);
1750 /* The stage-2 output mask is also applied for bypass */
1751 size
= arm_smmu_id_size_to_bits((id
>> ID2_OAS_SHIFT
) & ID2_OAS_MASK
);
1752 smmu
->s2_output_size
= min((unsigned long)PHYS_MASK_SHIFT
, size
);
1754 if (smmu
->version
== 1) {
1755 smmu
->input_size
= 32;
1758 size
= (id
>> ID2_UBS_SHIFT
) & ID2_UBS_MASK
;
1759 size
= min(39, arm_smmu_id_size_to_bits(size
));
1763 smmu
->input_size
= size
;
1765 if ((PAGE_SIZE
== SZ_4K
&& !(id
& ID2_PTFS_4K
)) ||
1766 (PAGE_SIZE
== SZ_64K
&& !(id
& ID2_PTFS_64K
)) ||
1767 (PAGE_SIZE
!= SZ_4K
&& PAGE_SIZE
!= SZ_64K
)) {
1768 dev_err(smmu
->dev
, "CPU page size 0x%lx unsupported\n",
1774 dev_notice(smmu
->dev
,
1775 "\t%lu-bit VA, %lu-bit IPA, %lu-bit PA\n",
1776 smmu
->input_size
, smmu
->s1_output_size
, smmu
->s2_output_size
);
1780 static int arm_smmu_device_dt_probe(struct platform_device
*pdev
)
1782 struct resource
*res
;
1783 struct arm_smmu_device
*smmu
;
1784 struct device_node
*dev_node
;
1785 struct device
*dev
= &pdev
->dev
;
1786 struct rb_node
*node
;
1787 struct of_phandle_args masterspec
;
1788 int num_irqs
, i
, err
;
1790 smmu
= devm_kzalloc(dev
, sizeof(*smmu
), GFP_KERNEL
);
1792 dev_err(dev
, "failed to allocate arm_smmu_device\n");
1797 res
= platform_get_resource(pdev
, IORESOURCE_MEM
, 0);
1798 smmu
->base
= devm_ioremap_resource(dev
, res
);
1799 if (IS_ERR(smmu
->base
))
1800 return PTR_ERR(smmu
->base
);
1801 smmu
->size
= resource_size(res
);
1803 if (of_property_read_u32(dev
->of_node
, "#global-interrupts",
1804 &smmu
->num_global_irqs
)) {
1805 dev_err(dev
, "missing #global-interrupts property\n");
1810 while ((res
= platform_get_resource(pdev
, IORESOURCE_IRQ
, num_irqs
))) {
1812 if (num_irqs
> smmu
->num_global_irqs
)
1813 smmu
->num_context_irqs
++;
1816 if (!smmu
->num_context_irqs
) {
1817 dev_err(dev
, "found %d interrupts but expected at least %d\n",
1818 num_irqs
, smmu
->num_global_irqs
+ 1);
1822 smmu
->irqs
= devm_kzalloc(dev
, sizeof(*smmu
->irqs
) * num_irqs
,
1825 dev_err(dev
, "failed to allocate %d irqs\n", num_irqs
);
1829 for (i
= 0; i
< num_irqs
; ++i
) {
1830 int irq
= platform_get_irq(pdev
, i
);
1832 dev_err(dev
, "failed to get irq index %d\n", i
);
1835 smmu
->irqs
[i
] = irq
;
1839 smmu
->masters
= RB_ROOT
;
1840 while (!of_parse_phandle_with_args(dev
->of_node
, "mmu-masters",
1841 "#stream-id-cells", i
,
1843 err
= register_smmu_master(smmu
, dev
, &masterspec
);
1845 dev_err(dev
, "failed to add master %s\n",
1846 masterspec
.np
->name
);
1847 goto out_put_masters
;
1852 dev_notice(dev
, "registered %d master devices\n", i
);
1854 if ((dev_node
= of_parse_phandle(dev
->of_node
, "smmu-parent", 0)))
1855 smmu
->parent_of_node
= dev_node
;
1857 err
= arm_smmu_device_cfg_probe(smmu
);
1859 goto out_put_parent
;
1861 if (smmu
->version
> 1 &&
1862 smmu
->num_context_banks
!= smmu
->num_context_irqs
) {
1864 "found only %d context interrupt(s) but %d required\n",
1865 smmu
->num_context_irqs
, smmu
->num_context_banks
);
1866 goto out_put_parent
;
1869 for (i
= 0; i
< smmu
->num_global_irqs
; ++i
) {
1870 err
= request_irq(smmu
->irqs
[i
],
1871 arm_smmu_global_fault
,
1873 "arm-smmu global fault",
1876 dev_err(dev
, "failed to request global IRQ %d (%u)\n",
1882 INIT_LIST_HEAD(&smmu
->list
);
1883 spin_lock(&arm_smmu_devices_lock
);
1884 list_add(&smmu
->list
, &arm_smmu_devices
);
1885 spin_unlock(&arm_smmu_devices_lock
);
1887 arm_smmu_device_reset(smmu
);
1892 free_irq(smmu
->irqs
[i
], smmu
);
1895 if (smmu
->parent_of_node
)
1896 of_node_put(smmu
->parent_of_node
);
1899 for (node
= rb_first(&smmu
->masters
); node
; node
= rb_next(node
)) {
1900 struct arm_smmu_master
*master
;
1901 master
= container_of(node
, struct arm_smmu_master
, node
);
1902 of_node_put(master
->of_node
);
1908 static int arm_smmu_device_remove(struct platform_device
*pdev
)
1911 struct device
*dev
= &pdev
->dev
;
1912 struct arm_smmu_device
*curr
, *smmu
= NULL
;
1913 struct rb_node
*node
;
1915 spin_lock(&arm_smmu_devices_lock
);
1916 list_for_each_entry(curr
, &arm_smmu_devices
, list
) {
1917 if (curr
->dev
== dev
) {
1919 list_del(&smmu
->list
);
1923 spin_unlock(&arm_smmu_devices_lock
);
1928 if (smmu
->parent_of_node
)
1929 of_node_put(smmu
->parent_of_node
);
1931 for (node
= rb_first(&smmu
->masters
); node
; node
= rb_next(node
)) {
1932 struct arm_smmu_master
*master
;
1933 master
= container_of(node
, struct arm_smmu_master
, node
);
1934 of_node_put(master
->of_node
);
1937 if (!bitmap_empty(smmu
->context_map
, ARM_SMMU_MAX_CBS
))
1938 dev_err(dev
, "removing device with active domains!\n");
1940 for (i
= 0; i
< smmu
->num_global_irqs
; ++i
)
1941 free_irq(smmu
->irqs
[i
], smmu
);
1943 /* Turn the thing off */
1944 writel_relaxed(sCR0_CLIENTPD
, ARM_SMMU_GR0(smmu
) + ARM_SMMU_GR0_sCR0
);
1949 static struct of_device_id arm_smmu_of_match
[] = {
1950 { .compatible
= "arm,smmu-v1", },
1951 { .compatible
= "arm,smmu-v2", },
1952 { .compatible
= "arm,mmu-400", },
1953 { .compatible
= "arm,mmu-500", },
1956 MODULE_DEVICE_TABLE(of
, arm_smmu_of_match
);
1959 static struct platform_driver arm_smmu_driver
= {
1961 .owner
= THIS_MODULE
,
1963 .of_match_table
= of_match_ptr(arm_smmu_of_match
),
1965 .probe
= arm_smmu_device_dt_probe
,
1966 .remove
= arm_smmu_device_remove
,
1969 static int __init
arm_smmu_init(void)
1973 ret
= platform_driver_register(&arm_smmu_driver
);
1977 /* Oh, for a proper bus abstraction */
1978 if (!iommu_present(&platform_bus_type
))
1979 bus_set_iommu(&platform_bus_type
, &arm_smmu_ops
);
1981 if (!iommu_present(&amba_bustype
))
1982 bus_set_iommu(&amba_bustype
, &arm_smmu_ops
);
1987 static void __exit
arm_smmu_exit(void)
1989 return platform_driver_unregister(&arm_smmu_driver
);
1992 subsys_initcall(arm_smmu_init
);
1993 module_exit(arm_smmu_exit
);
1995 MODULE_DESCRIPTION("IOMMU API for ARM architected SMMU implementations");
1996 MODULE_AUTHOR("Will Deacon <will.deacon@arm.com>");
1997 MODULE_LICENSE("GPL v2");