Commit | Line | Data |
---|---|---|
45ae7cff WD |
1 | /* |
2 | * IOMMU API for ARM architected SMMU implementations. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or modify | |
5 | * it under the terms of the GNU General Public License version 2 as | |
6 | * published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
11 | * GNU General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public License | |
14 | * along with this program; if not, write to the Free Software | |
15 | * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. | |
16 | * | |
17 | * Copyright (C) 2013 ARM Limited | |
18 | * | |
19 | * Author: Will Deacon <will.deacon@arm.com> | |
20 | * | |
21 | * This driver currently supports: | |
22 | * - SMMUv1 and v2 implementations | |
23 | * - Stream-matching and stream-indexing | |
24 | * - v7/v8 long-descriptor format | |
25 | * - Non-secure access to the SMMU | |
26 | * - 4k and 64k pages, with contiguous pte hints. | |
27 | * - Up to 39-bit addressing | |
28 | * - Context fault reporting | |
29 | */ | |
30 | ||
31 | #define pr_fmt(fmt) "arm-smmu: " fmt | |
32 | ||
33 | #include <linux/delay.h> | |
34 | #include <linux/dma-mapping.h> | |
35 | #include <linux/err.h> | |
36 | #include <linux/interrupt.h> | |
37 | #include <linux/io.h> | |
38 | #include <linux/iommu.h> | |
39 | #include <linux/mm.h> | |
40 | #include <linux/module.h> | |
41 | #include <linux/of.h> | |
42 | #include <linux/platform_device.h> | |
43 | #include <linux/slab.h> | |
44 | #include <linux/spinlock.h> | |
45 | ||
46 | #include <linux/amba/bus.h> | |
47 | ||
48 | #include <asm/pgalloc.h> | |
49 | ||
50 | /* Maximum number of stream IDs assigned to a single device */ | |
51 | #define MAX_MASTER_STREAMIDS 8 | |
52 | ||
53 | /* Maximum number of context banks per SMMU */ | |
54 | #define ARM_SMMU_MAX_CBS 128 | |
55 | ||
56 | /* Maximum number of mapping groups per SMMU */ | |
57 | #define ARM_SMMU_MAX_SMRS 128 | |
58 | ||
59 | /* Number of VMIDs per SMMU */ | |
60 | #define ARM_SMMU_NUM_VMIDS 256 | |
61 | ||
62 | /* SMMU global address space */ | |
63 | #define ARM_SMMU_GR0(smmu) ((smmu)->base) | |
64 | #define ARM_SMMU_GR1(smmu) ((smmu)->base + (smmu)->pagesize) | |
65 | ||
66 | /* Page table bits */ | |
67 | #define ARM_SMMU_PTE_PAGE (((pteval_t)3) << 0) | |
68 | #define ARM_SMMU_PTE_CONT (((pteval_t)1) << 52) | |
69 | #define ARM_SMMU_PTE_AF (((pteval_t)1) << 10) | |
70 | #define ARM_SMMU_PTE_SH_NS (((pteval_t)0) << 8) | |
71 | #define ARM_SMMU_PTE_SH_OS (((pteval_t)2) << 8) | |
72 | #define ARM_SMMU_PTE_SH_IS (((pteval_t)3) << 8) | |
73 | ||
74 | #if PAGE_SIZE == SZ_4K | |
75 | #define ARM_SMMU_PTE_CONT_ENTRIES 16 | |
76 | #elif PAGE_SIZE == SZ_64K | |
77 | #define ARM_SMMU_PTE_CONT_ENTRIES 32 | |
78 | #else | |
79 | #define ARM_SMMU_PTE_CONT_ENTRIES 1 | |
80 | #endif | |
81 | ||
82 | #define ARM_SMMU_PTE_CONT_SIZE (PAGE_SIZE * ARM_SMMU_PTE_CONT_ENTRIES) | |
83 | #define ARM_SMMU_PTE_CONT_MASK (~(ARM_SMMU_PTE_CONT_SIZE - 1)) | |
84 | #define ARM_SMMU_PTE_HWTABLE_SIZE (PTRS_PER_PTE * sizeof(pte_t)) | |
85 | ||
86 | /* Stage-1 PTE */ | |
87 | #define ARM_SMMU_PTE_AP_UNPRIV (((pteval_t)1) << 6) | |
88 | #define ARM_SMMU_PTE_AP_RDONLY (((pteval_t)2) << 6) | |
89 | #define ARM_SMMU_PTE_ATTRINDX_SHIFT 2 | |
90 | ||
91 | /* Stage-2 PTE */ | |
92 | #define ARM_SMMU_PTE_HAP_FAULT (((pteval_t)0) << 6) | |
93 | #define ARM_SMMU_PTE_HAP_READ (((pteval_t)1) << 6) | |
94 | #define ARM_SMMU_PTE_HAP_WRITE (((pteval_t)2) << 6) | |
95 | #define ARM_SMMU_PTE_MEMATTR_OIWB (((pteval_t)0xf) << 2) | |
96 | #define ARM_SMMU_PTE_MEMATTR_NC (((pteval_t)0x5) << 2) | |
97 | #define ARM_SMMU_PTE_MEMATTR_DEV (((pteval_t)0x1) << 2) | |
98 | ||
99 | /* Configuration registers */ | |
100 | #define ARM_SMMU_GR0_sCR0 0x0 | |
101 | #define sCR0_CLIENTPD (1 << 0) | |
102 | #define sCR0_GFRE (1 << 1) | |
103 | #define sCR0_GFIE (1 << 2) | |
104 | #define sCR0_GCFGFRE (1 << 4) | |
105 | #define sCR0_GCFGFIE (1 << 5) | |
106 | #define sCR0_USFCFG (1 << 10) | |
107 | #define sCR0_VMIDPNE (1 << 11) | |
108 | #define sCR0_PTM (1 << 12) | |
109 | #define sCR0_FB (1 << 13) | |
110 | #define sCR0_BSU_SHIFT 14 | |
111 | #define sCR0_BSU_MASK 0x3 | |
112 | ||
113 | /* Identification registers */ | |
114 | #define ARM_SMMU_GR0_ID0 0x20 | |
115 | #define ARM_SMMU_GR0_ID1 0x24 | |
116 | #define ARM_SMMU_GR0_ID2 0x28 | |
117 | #define ARM_SMMU_GR0_ID3 0x2c | |
118 | #define ARM_SMMU_GR0_ID4 0x30 | |
119 | #define ARM_SMMU_GR0_ID5 0x34 | |
120 | #define ARM_SMMU_GR0_ID6 0x38 | |
121 | #define ARM_SMMU_GR0_ID7 0x3c | |
122 | #define ARM_SMMU_GR0_sGFSR 0x48 | |
123 | #define ARM_SMMU_GR0_sGFSYNR0 0x50 | |
124 | #define ARM_SMMU_GR0_sGFSYNR1 0x54 | |
125 | #define ARM_SMMU_GR0_sGFSYNR2 0x58 | |
126 | #define ARM_SMMU_GR0_PIDR0 0xfe0 | |
127 | #define ARM_SMMU_GR0_PIDR1 0xfe4 | |
128 | #define ARM_SMMU_GR0_PIDR2 0xfe8 | |
129 | ||
130 | #define ID0_S1TS (1 << 30) | |
131 | #define ID0_S2TS (1 << 29) | |
132 | #define ID0_NTS (1 << 28) | |
133 | #define ID0_SMS (1 << 27) | |
134 | #define ID0_PTFS_SHIFT 24 | |
135 | #define ID0_PTFS_MASK 0x2 | |
136 | #define ID0_PTFS_V8_ONLY 0x2 | |
137 | #define ID0_CTTW (1 << 14) | |
138 | #define ID0_NUMIRPT_SHIFT 16 | |
139 | #define ID0_NUMIRPT_MASK 0xff | |
140 | #define ID0_NUMSMRG_SHIFT 0 | |
141 | #define ID0_NUMSMRG_MASK 0xff | |
142 | ||
143 | #define ID1_PAGESIZE (1 << 31) | |
144 | #define ID1_NUMPAGENDXB_SHIFT 28 | |
145 | #define ID1_NUMPAGENDXB_MASK 7 | |
146 | #define ID1_NUMS2CB_SHIFT 16 | |
147 | #define ID1_NUMS2CB_MASK 0xff | |
148 | #define ID1_NUMCB_SHIFT 0 | |
149 | #define ID1_NUMCB_MASK 0xff | |
150 | ||
151 | #define ID2_OAS_SHIFT 4 | |
152 | #define ID2_OAS_MASK 0xf | |
153 | #define ID2_IAS_SHIFT 0 | |
154 | #define ID2_IAS_MASK 0xf | |
155 | #define ID2_UBS_SHIFT 8 | |
156 | #define ID2_UBS_MASK 0xf | |
157 | #define ID2_PTFS_4K (1 << 12) | |
158 | #define ID2_PTFS_16K (1 << 13) | |
159 | #define ID2_PTFS_64K (1 << 14) | |
160 | ||
161 | #define PIDR2_ARCH_SHIFT 4 | |
162 | #define PIDR2_ARCH_MASK 0xf | |
163 | ||
164 | /* Global TLB invalidation */ | |
165 | #define ARM_SMMU_GR0_STLBIALL 0x60 | |
166 | #define ARM_SMMU_GR0_TLBIVMID 0x64 | |
167 | #define ARM_SMMU_GR0_TLBIALLNSNH 0x68 | |
168 | #define ARM_SMMU_GR0_TLBIALLH 0x6c | |
169 | #define ARM_SMMU_GR0_sTLBGSYNC 0x70 | |
170 | #define ARM_SMMU_GR0_sTLBGSTATUS 0x74 | |
171 | #define sTLBGSTATUS_GSACTIVE (1 << 0) | |
172 | #define TLB_LOOP_TIMEOUT 1000000 /* 1s! */ | |
173 | ||
174 | /* Stream mapping registers */ | |
175 | #define ARM_SMMU_GR0_SMR(n) (0x800 + ((n) << 2)) | |
176 | #define SMR_VALID (1 << 31) | |
177 | #define SMR_MASK_SHIFT 16 | |
178 | #define SMR_MASK_MASK 0x7fff | |
179 | #define SMR_ID_SHIFT 0 | |
180 | #define SMR_ID_MASK 0x7fff | |
181 | ||
182 | #define ARM_SMMU_GR0_S2CR(n) (0xc00 + ((n) << 2)) | |
183 | #define S2CR_CBNDX_SHIFT 0 | |
184 | #define S2CR_CBNDX_MASK 0xff | |
185 | #define S2CR_TYPE_SHIFT 16 | |
186 | #define S2CR_TYPE_MASK 0x3 | |
187 | #define S2CR_TYPE_TRANS (0 << S2CR_TYPE_SHIFT) | |
188 | #define S2CR_TYPE_BYPASS (1 << S2CR_TYPE_SHIFT) | |
189 | #define S2CR_TYPE_FAULT (2 << S2CR_TYPE_SHIFT) | |
190 | ||
191 | /* Context bank attribute registers */ | |
192 | #define ARM_SMMU_GR1_CBAR(n) (0x0 + ((n) << 2)) | |
193 | #define CBAR_VMID_SHIFT 0 | |
194 | #define CBAR_VMID_MASK 0xff | |
195 | #define CBAR_S1_MEMATTR_SHIFT 12 | |
196 | #define CBAR_S1_MEMATTR_MASK 0xf | |
197 | #define CBAR_S1_MEMATTR_WB 0xf | |
198 | #define CBAR_TYPE_SHIFT 16 | |
199 | #define CBAR_TYPE_MASK 0x3 | |
200 | #define CBAR_TYPE_S2_TRANS (0 << CBAR_TYPE_SHIFT) | |
201 | #define CBAR_TYPE_S1_TRANS_S2_BYPASS (1 << CBAR_TYPE_SHIFT) | |
202 | #define CBAR_TYPE_S1_TRANS_S2_FAULT (2 << CBAR_TYPE_SHIFT) | |
203 | #define CBAR_TYPE_S1_TRANS_S2_TRANS (3 << CBAR_TYPE_SHIFT) | |
204 | #define CBAR_IRPTNDX_SHIFT 24 | |
205 | #define CBAR_IRPTNDX_MASK 0xff | |
206 | ||
207 | #define ARM_SMMU_GR1_CBA2R(n) (0x800 + ((n) << 2)) | |
208 | #define CBA2R_RW64_32BIT (0 << 0) | |
209 | #define CBA2R_RW64_64BIT (1 << 0) | |
210 | ||
211 | /* Translation context bank */ | |
212 | #define ARM_SMMU_CB_BASE(smmu) ((smmu)->base + ((smmu)->size >> 1)) | |
213 | #define ARM_SMMU_CB(smmu, n) ((n) * (smmu)->pagesize) | |
214 | ||
215 | #define ARM_SMMU_CB_SCTLR 0x0 | |
216 | #define ARM_SMMU_CB_RESUME 0x8 | |
217 | #define ARM_SMMU_CB_TTBCR2 0x10 | |
218 | #define ARM_SMMU_CB_TTBR0_LO 0x20 | |
219 | #define ARM_SMMU_CB_TTBR0_HI 0x24 | |
220 | #define ARM_SMMU_CB_TTBCR 0x30 | |
221 | #define ARM_SMMU_CB_S1_MAIR0 0x38 | |
222 | #define ARM_SMMU_CB_FSR 0x58 | |
223 | #define ARM_SMMU_CB_FAR_LO 0x60 | |
224 | #define ARM_SMMU_CB_FAR_HI 0x64 | |
225 | #define ARM_SMMU_CB_FSYNR0 0x68 | |
226 | ||
227 | #define SCTLR_S1_ASIDPNE (1 << 12) | |
228 | #define SCTLR_CFCFG (1 << 7) | |
229 | #define SCTLR_CFIE (1 << 6) | |
230 | #define SCTLR_CFRE (1 << 5) | |
231 | #define SCTLR_E (1 << 4) | |
232 | #define SCTLR_AFE (1 << 2) | |
233 | #define SCTLR_TRE (1 << 1) | |
234 | #define SCTLR_M (1 << 0) | |
235 | #define SCTLR_EAE_SBOP (SCTLR_AFE | SCTLR_TRE) | |
236 | ||
237 | #define RESUME_RETRY (0 << 0) | |
238 | #define RESUME_TERMINATE (1 << 0) | |
239 | ||
240 | #define TTBCR_EAE (1 << 31) | |
241 | ||
242 | #define TTBCR_PASIZE_SHIFT 16 | |
243 | #define TTBCR_PASIZE_MASK 0x7 | |
244 | ||
245 | #define TTBCR_TG0_4K (0 << 14) | |
246 | #define TTBCR_TG0_64K (1 << 14) | |
247 | ||
248 | #define TTBCR_SH0_SHIFT 12 | |
249 | #define TTBCR_SH0_MASK 0x3 | |
250 | #define TTBCR_SH_NS 0 | |
251 | #define TTBCR_SH_OS 2 | |
252 | #define TTBCR_SH_IS 3 | |
253 | ||
254 | #define TTBCR_ORGN0_SHIFT 10 | |
255 | #define TTBCR_IRGN0_SHIFT 8 | |
256 | #define TTBCR_RGN_MASK 0x3 | |
257 | #define TTBCR_RGN_NC 0 | |
258 | #define TTBCR_RGN_WBWA 1 | |
259 | #define TTBCR_RGN_WT 2 | |
260 | #define TTBCR_RGN_WB 3 | |
261 | ||
262 | #define TTBCR_SL0_SHIFT 6 | |
263 | #define TTBCR_SL0_MASK 0x3 | |
264 | #define TTBCR_SL0_LVL_2 0 | |
265 | #define TTBCR_SL0_LVL_1 1 | |
266 | ||
267 | #define TTBCR_T1SZ_SHIFT 16 | |
268 | #define TTBCR_T0SZ_SHIFT 0 | |
269 | #define TTBCR_SZ_MASK 0xf | |
270 | ||
271 | #define TTBCR2_SEP_SHIFT 15 | |
272 | #define TTBCR2_SEP_MASK 0x7 | |
273 | ||
274 | #define TTBCR2_PASIZE_SHIFT 0 | |
275 | #define TTBCR2_PASIZE_MASK 0x7 | |
276 | ||
277 | /* Common definitions for PASize and SEP fields */ | |
278 | #define TTBCR2_ADDR_32 0 | |
279 | #define TTBCR2_ADDR_36 1 | |
280 | #define TTBCR2_ADDR_40 2 | |
281 | #define TTBCR2_ADDR_42 3 | |
282 | #define TTBCR2_ADDR_44 4 | |
283 | #define TTBCR2_ADDR_48 5 | |
284 | ||
285 | #define MAIR_ATTR_SHIFT(n) ((n) << 3) | |
286 | #define MAIR_ATTR_MASK 0xff | |
287 | #define MAIR_ATTR_DEVICE 0x04 | |
288 | #define MAIR_ATTR_NC 0x44 | |
289 | #define MAIR_ATTR_WBRWA 0xff | |
290 | #define MAIR_ATTR_IDX_NC 0 | |
291 | #define MAIR_ATTR_IDX_CACHE 1 | |
292 | #define MAIR_ATTR_IDX_DEV 2 | |
293 | ||
294 | #define FSR_MULTI (1 << 31) | |
295 | #define FSR_SS (1 << 30) | |
296 | #define FSR_UUT (1 << 8) | |
297 | #define FSR_ASF (1 << 7) | |
298 | #define FSR_TLBLKF (1 << 6) | |
299 | #define FSR_TLBMCF (1 << 5) | |
300 | #define FSR_EF (1 << 4) | |
301 | #define FSR_PF (1 << 3) | |
302 | #define FSR_AFF (1 << 2) | |
303 | #define FSR_TF (1 << 1) | |
304 | ||
305 | #define FSR_IGN (FSR_AFF | FSR_ASF | FSR_TLBMCF | \ | |
306 | FSR_TLBLKF) | |
307 | #define FSR_FAULT (FSR_MULTI | FSR_SS | FSR_UUT | \ | |
308 | FSR_EF | FSR_PF | FSR_TF) | |
309 | ||
310 | #define FSYNR0_WNR (1 << 4) | |
311 | ||
312 | struct arm_smmu_smr { | |
313 | u8 idx; | |
314 | u16 mask; | |
315 | u16 id; | |
316 | }; | |
317 | ||
318 | struct arm_smmu_master { | |
319 | struct device_node *of_node; | |
320 | ||
321 | /* | |
322 | * The following is specific to the master's position in the | |
323 | * SMMU chain. | |
324 | */ | |
325 | struct rb_node node; | |
326 | int num_streamids; | |
327 | u16 streamids[MAX_MASTER_STREAMIDS]; | |
328 | ||
329 | /* | |
330 | * We only need to allocate these on the root SMMU, as we | |
331 | * configure unmatched streams to bypass translation. | |
332 | */ | |
333 | struct arm_smmu_smr *smrs; | |
334 | }; | |
335 | ||
336 | struct arm_smmu_device { | |
337 | struct device *dev; | |
338 | struct device_node *parent_of_node; | |
339 | ||
340 | void __iomem *base; | |
341 | unsigned long size; | |
342 | unsigned long pagesize; | |
343 | ||
344 | #define ARM_SMMU_FEAT_COHERENT_WALK (1 << 0) | |
345 | #define ARM_SMMU_FEAT_STREAM_MATCH (1 << 1) | |
346 | #define ARM_SMMU_FEAT_TRANS_S1 (1 << 2) | |
347 | #define ARM_SMMU_FEAT_TRANS_S2 (1 << 3) | |
348 | #define ARM_SMMU_FEAT_TRANS_NESTED (1 << 4) | |
349 | u32 features; | |
350 | int version; | |
351 | ||
352 | u32 num_context_banks; | |
353 | u32 num_s2_context_banks; | |
354 | DECLARE_BITMAP(context_map, ARM_SMMU_MAX_CBS); | |
355 | atomic_t irptndx; | |
356 | ||
357 | u32 num_mapping_groups; | |
358 | DECLARE_BITMAP(smr_map, ARM_SMMU_MAX_SMRS); | |
359 | ||
360 | unsigned long input_size; | |
361 | unsigned long s1_output_size; | |
362 | unsigned long s2_output_size; | |
363 | ||
364 | u32 num_global_irqs; | |
365 | u32 num_context_irqs; | |
366 | unsigned int *irqs; | |
367 | ||
368 | DECLARE_BITMAP(vmid_map, ARM_SMMU_NUM_VMIDS); | |
369 | ||
370 | struct list_head list; | |
371 | struct rb_root masters; | |
372 | }; | |
373 | ||
374 | struct arm_smmu_cfg { | |
375 | struct arm_smmu_device *smmu; | |
376 | u8 vmid; | |
377 | u8 cbndx; | |
378 | u8 irptndx; | |
379 | u32 cbar; | |
380 | pgd_t *pgd; | |
381 | }; | |
382 | ||
383 | struct arm_smmu_domain { | |
384 | /* | |
385 | * A domain can span across multiple, chained SMMUs and requires | |
386 | * all devices within the domain to follow the same translation | |
387 | * path. | |
388 | */ | |
389 | struct arm_smmu_device *leaf_smmu; | |
390 | struct arm_smmu_cfg root_cfg; | |
391 | phys_addr_t output_mask; | |
392 | ||
393 | spinlock_t lock; | |
394 | }; | |
395 | ||
396 | static DEFINE_SPINLOCK(arm_smmu_devices_lock); | |
397 | static LIST_HEAD(arm_smmu_devices); | |
398 | ||
399 | static struct arm_smmu_master *find_smmu_master(struct arm_smmu_device *smmu, | |
400 | struct device_node *dev_node) | |
401 | { | |
402 | struct rb_node *node = smmu->masters.rb_node; | |
403 | ||
404 | while (node) { | |
405 | struct arm_smmu_master *master; | |
406 | master = container_of(node, struct arm_smmu_master, node); | |
407 | ||
408 | if (dev_node < master->of_node) | |
409 | node = node->rb_left; | |
410 | else if (dev_node > master->of_node) | |
411 | node = node->rb_right; | |
412 | else | |
413 | return master; | |
414 | } | |
415 | ||
416 | return NULL; | |
417 | } | |
418 | ||
419 | static int insert_smmu_master(struct arm_smmu_device *smmu, | |
420 | struct arm_smmu_master *master) | |
421 | { | |
422 | struct rb_node **new, *parent; | |
423 | ||
424 | new = &smmu->masters.rb_node; | |
425 | parent = NULL; | |
426 | while (*new) { | |
427 | struct arm_smmu_master *this; | |
428 | this = container_of(*new, struct arm_smmu_master, node); | |
429 | ||
430 | parent = *new; | |
431 | if (master->of_node < this->of_node) | |
432 | new = &((*new)->rb_left); | |
433 | else if (master->of_node > this->of_node) | |
434 | new = &((*new)->rb_right); | |
435 | else | |
436 | return -EEXIST; | |
437 | } | |
438 | ||
439 | rb_link_node(&master->node, parent, new); | |
440 | rb_insert_color(&master->node, &smmu->masters); | |
441 | return 0; | |
442 | } | |
443 | ||
444 | static int register_smmu_master(struct arm_smmu_device *smmu, | |
445 | struct device *dev, | |
446 | struct of_phandle_args *masterspec) | |
447 | { | |
448 | int i; | |
449 | struct arm_smmu_master *master; | |
450 | ||
451 | master = find_smmu_master(smmu, masterspec->np); | |
452 | if (master) { | |
453 | dev_err(dev, | |
454 | "rejecting multiple registrations for master device %s\n", | |
455 | masterspec->np->name); | |
456 | return -EBUSY; | |
457 | } | |
458 | ||
459 | if (masterspec->args_count > MAX_MASTER_STREAMIDS) { | |
460 | dev_err(dev, | |
461 | "reached maximum number (%d) of stream IDs for master device %s\n", | |
462 | MAX_MASTER_STREAMIDS, masterspec->np->name); | |
463 | return -ENOSPC; | |
464 | } | |
465 | ||
466 | master = devm_kzalloc(dev, sizeof(*master), GFP_KERNEL); | |
467 | if (!master) | |
468 | return -ENOMEM; | |
469 | ||
470 | master->of_node = masterspec->np; | |
471 | master->num_streamids = masterspec->args_count; | |
472 | ||
473 | for (i = 0; i < master->num_streamids; ++i) | |
474 | master->streamids[i] = masterspec->args[i]; | |
475 | ||
476 | return insert_smmu_master(smmu, master); | |
477 | } | |
478 | ||
479 | static struct arm_smmu_device *find_parent_smmu(struct arm_smmu_device *smmu) | |
480 | { | |
481 | struct arm_smmu_device *parent; | |
482 | ||
483 | if (!smmu->parent_of_node) | |
484 | return NULL; | |
485 | ||
486 | spin_lock(&arm_smmu_devices_lock); | |
487 | list_for_each_entry(parent, &arm_smmu_devices, list) | |
488 | if (parent->dev->of_node == smmu->parent_of_node) | |
489 | goto out_unlock; | |
490 | ||
491 | parent = NULL; | |
492 | dev_warn(smmu->dev, | |
493 | "Failed to find SMMU parent despite parent in DT\n"); | |
494 | out_unlock: | |
495 | spin_unlock(&arm_smmu_devices_lock); | |
496 | return parent; | |
497 | } | |
498 | ||
499 | static int __arm_smmu_alloc_bitmap(unsigned long *map, int start, int end) | |
500 | { | |
501 | int idx; | |
502 | ||
503 | do { | |
504 | idx = find_next_zero_bit(map, end, start); | |
505 | if (idx == end) | |
506 | return -ENOSPC; | |
507 | } while (test_and_set_bit(idx, map)); | |
508 | ||
509 | return idx; | |
510 | } | |
511 | ||
512 | static void __arm_smmu_free_bitmap(unsigned long *map, int idx) | |
513 | { | |
514 | clear_bit(idx, map); | |
515 | } | |
516 | ||
517 | /* Wait for any pending TLB invalidations to complete */ | |
518 | static void arm_smmu_tlb_sync(struct arm_smmu_device *smmu) | |
519 | { | |
520 | int count = 0; | |
521 | void __iomem *gr0_base = ARM_SMMU_GR0(smmu); | |
522 | ||
523 | writel_relaxed(0, gr0_base + ARM_SMMU_GR0_sTLBGSYNC); | |
524 | while (readl_relaxed(gr0_base + ARM_SMMU_GR0_sTLBGSTATUS) | |
525 | & sTLBGSTATUS_GSACTIVE) { | |
526 | cpu_relax(); | |
527 | if (++count == TLB_LOOP_TIMEOUT) { | |
528 | dev_err_ratelimited(smmu->dev, | |
529 | "TLB sync timed out -- SMMU may be deadlocked\n"); | |
530 | return; | |
531 | } | |
532 | udelay(1); | |
533 | } | |
534 | } | |
535 | ||
536 | static irqreturn_t arm_smmu_context_fault(int irq, void *dev) | |
537 | { | |
538 | int flags, ret; | |
539 | u32 fsr, far, fsynr, resume; | |
540 | unsigned long iova; | |
541 | struct iommu_domain *domain = dev; | |
542 | struct arm_smmu_domain *smmu_domain = domain->priv; | |
543 | struct arm_smmu_cfg *root_cfg = &smmu_domain->root_cfg; | |
544 | struct arm_smmu_device *smmu = root_cfg->smmu; | |
545 | void __iomem *cb_base; | |
546 | ||
547 | cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, root_cfg->cbndx); | |
548 | fsr = readl_relaxed(cb_base + ARM_SMMU_CB_FSR); | |
549 | ||
550 | if (!(fsr & FSR_FAULT)) | |
551 | return IRQ_NONE; | |
552 | ||
553 | if (fsr & FSR_IGN) | |
554 | dev_err_ratelimited(smmu->dev, | |
555 | "Unexpected context fault (fsr 0x%u)\n", | |
556 | fsr); | |
557 | ||
558 | fsynr = readl_relaxed(cb_base + ARM_SMMU_CB_FSYNR0); | |
559 | flags = fsynr & FSYNR0_WNR ? IOMMU_FAULT_WRITE : IOMMU_FAULT_READ; | |
560 | ||
561 | far = readl_relaxed(cb_base + ARM_SMMU_CB_FAR_LO); | |
562 | iova = far; | |
563 | #ifdef CONFIG_64BIT | |
564 | far = readl_relaxed(cb_base + ARM_SMMU_CB_FAR_HI); | |
565 | iova |= ((unsigned long)far << 32); | |
566 | #endif | |
567 | ||
568 | if (!report_iommu_fault(domain, smmu->dev, iova, flags)) { | |
569 | ret = IRQ_HANDLED; | |
570 | resume = RESUME_RETRY; | |
571 | } else { | |
572 | ret = IRQ_NONE; | |
573 | resume = RESUME_TERMINATE; | |
574 | } | |
575 | ||
576 | /* Clear the faulting FSR */ | |
577 | writel(fsr, cb_base + ARM_SMMU_CB_FSR); | |
578 | ||
579 | /* Retry or terminate any stalled transactions */ | |
580 | if (fsr & FSR_SS) | |
581 | writel_relaxed(resume, cb_base + ARM_SMMU_CB_RESUME); | |
582 | ||
583 | return ret; | |
584 | } | |
585 | ||
586 | static irqreturn_t arm_smmu_global_fault(int irq, void *dev) | |
587 | { | |
588 | u32 gfsr, gfsynr0, gfsynr1, gfsynr2; | |
589 | struct arm_smmu_device *smmu = dev; | |
590 | void __iomem *gr0_base = ARM_SMMU_GR0(smmu); | |
591 | ||
592 | gfsr = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSR); | |
593 | gfsynr0 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSYNR0); | |
594 | gfsynr1 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSYNR1); | |
595 | gfsynr2 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSYNR2); | |
596 | ||
597 | dev_err_ratelimited(smmu->dev, | |
598 | "Unexpected global fault, this could be serious\n"); | |
599 | dev_err_ratelimited(smmu->dev, | |
600 | "\tGFSR 0x%08x, GFSYNR0 0x%08x, GFSYNR1 0x%08x, GFSYNR2 0x%08x\n", | |
601 | gfsr, gfsynr0, gfsynr1, gfsynr2); | |
602 | ||
603 | writel(gfsr, gr0_base + ARM_SMMU_GR0_sGFSR); | |
604 | return IRQ_NONE; | |
605 | } | |
606 | ||
607 | static void arm_smmu_init_context_bank(struct arm_smmu_domain *smmu_domain) | |
608 | { | |
609 | u32 reg; | |
610 | bool stage1; | |
611 | struct arm_smmu_cfg *root_cfg = &smmu_domain->root_cfg; | |
612 | struct arm_smmu_device *smmu = root_cfg->smmu; | |
613 | void __iomem *cb_base, *gr0_base, *gr1_base; | |
614 | ||
615 | gr0_base = ARM_SMMU_GR0(smmu); | |
616 | gr1_base = ARM_SMMU_GR1(smmu); | |
617 | stage1 = root_cfg->cbar != CBAR_TYPE_S2_TRANS; | |
618 | cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, root_cfg->cbndx); | |
619 | ||
620 | /* CBAR */ | |
621 | reg = root_cfg->cbar | | |
622 | (root_cfg->vmid << CBAR_VMID_SHIFT); | |
623 | if (smmu->version == 1) | |
624 | reg |= root_cfg->irptndx << CBAR_IRPTNDX_SHIFT; | |
625 | ||
626 | /* Use the weakest memory type, so it is overridden by the pte */ | |
627 | if (stage1) | |
628 | reg |= (CBAR_S1_MEMATTR_WB << CBAR_S1_MEMATTR_SHIFT); | |
629 | writel_relaxed(reg, gr1_base + ARM_SMMU_GR1_CBAR(root_cfg->cbndx)); | |
630 | ||
631 | if (smmu->version > 1) { | |
632 | /* CBA2R */ | |
633 | #ifdef CONFIG_64BIT | |
634 | reg = CBA2R_RW64_64BIT; | |
635 | #else | |
636 | reg = CBA2R_RW64_32BIT; | |
637 | #endif | |
638 | writel_relaxed(reg, | |
639 | gr1_base + ARM_SMMU_GR1_CBA2R(root_cfg->cbndx)); | |
640 | ||
641 | /* TTBCR2 */ | |
642 | switch (smmu->input_size) { | |
643 | case 32: | |
644 | reg = (TTBCR2_ADDR_32 << TTBCR2_SEP_SHIFT); | |
645 | break; | |
646 | case 36: | |
647 | reg = (TTBCR2_ADDR_36 << TTBCR2_SEP_SHIFT); | |
648 | break; | |
649 | case 39: | |
650 | reg = (TTBCR2_ADDR_40 << TTBCR2_SEP_SHIFT); | |
651 | break; | |
652 | case 42: | |
653 | reg = (TTBCR2_ADDR_42 << TTBCR2_SEP_SHIFT); | |
654 | break; | |
655 | case 44: | |
656 | reg = (TTBCR2_ADDR_44 << TTBCR2_SEP_SHIFT); | |
657 | break; | |
658 | case 48: | |
659 | reg = (TTBCR2_ADDR_48 << TTBCR2_SEP_SHIFT); | |
660 | break; | |
661 | } | |
662 | ||
663 | switch (smmu->s1_output_size) { | |
664 | case 32: | |
665 | reg |= (TTBCR2_ADDR_32 << TTBCR2_PASIZE_SHIFT); | |
666 | break; | |
667 | case 36: | |
668 | reg |= (TTBCR2_ADDR_36 << TTBCR2_PASIZE_SHIFT); | |
669 | break; | |
670 | case 39: | |
671 | reg |= (TTBCR2_ADDR_40 << TTBCR2_PASIZE_SHIFT); | |
672 | break; | |
673 | case 42: | |
674 | reg |= (TTBCR2_ADDR_42 << TTBCR2_PASIZE_SHIFT); | |
675 | break; | |
676 | case 44: | |
677 | reg |= (TTBCR2_ADDR_44 << TTBCR2_PASIZE_SHIFT); | |
678 | break; | |
679 | case 48: | |
680 | reg |= (TTBCR2_ADDR_48 << TTBCR2_PASIZE_SHIFT); | |
681 | break; | |
682 | } | |
683 | ||
684 | if (stage1) | |
685 | writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBCR2); | |
686 | } | |
687 | ||
688 | /* TTBR0 */ | |
689 | reg = __pa(root_cfg->pgd); | |
690 | #ifndef __BIG_ENDIAN | |
691 | writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBR0_LO); | |
692 | reg = (phys_addr_t)__pa(root_cfg->pgd) >> 32; | |
693 | writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBR0_HI); | |
694 | #else | |
695 | writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBR0_HI); | |
696 | reg = (phys_addr_t)__pa(root_cfg->pgd) >> 32; | |
697 | writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBR0_LO); | |
698 | #endif | |
699 | ||
700 | /* | |
701 | * TTBCR | |
702 | * We use long descriptor, with inner-shareable WBWA tables in TTBR0. | |
703 | */ | |
704 | if (smmu->version > 1) { | |
705 | if (PAGE_SIZE == SZ_4K) | |
706 | reg = TTBCR_TG0_4K; | |
707 | else | |
708 | reg = TTBCR_TG0_64K; | |
709 | ||
710 | if (!stage1) { | |
711 | switch (smmu->s2_output_size) { | |
712 | case 32: | |
713 | reg |= (TTBCR2_ADDR_32 << TTBCR_PASIZE_SHIFT); | |
714 | break; | |
715 | case 36: | |
716 | reg |= (TTBCR2_ADDR_36 << TTBCR_PASIZE_SHIFT); | |
717 | break; | |
718 | case 40: | |
719 | reg |= (TTBCR2_ADDR_40 << TTBCR_PASIZE_SHIFT); | |
720 | break; | |
721 | case 42: | |
722 | reg |= (TTBCR2_ADDR_42 << TTBCR_PASIZE_SHIFT); | |
723 | break; | |
724 | case 44: | |
725 | reg |= (TTBCR2_ADDR_44 << TTBCR_PASIZE_SHIFT); | |
726 | break; | |
727 | case 48: | |
728 | reg |= (TTBCR2_ADDR_48 << TTBCR_PASIZE_SHIFT); | |
729 | break; | |
730 | } | |
731 | } else { | |
732 | reg |= (64 - smmu->s1_output_size) << TTBCR_T0SZ_SHIFT; | |
733 | } | |
734 | } else { | |
735 | reg = 0; | |
736 | } | |
737 | ||
738 | reg |= TTBCR_EAE | | |
739 | (TTBCR_SH_IS << TTBCR_SH0_SHIFT) | | |
740 | (TTBCR_RGN_WBWA << TTBCR_ORGN0_SHIFT) | | |
741 | (TTBCR_RGN_WBWA << TTBCR_IRGN0_SHIFT) | | |
742 | (TTBCR_SL0_LVL_1 << TTBCR_SL0_SHIFT); | |
743 | writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBCR); | |
744 | ||
745 | /* MAIR0 (stage-1 only) */ | |
746 | if (stage1) { | |
747 | reg = (MAIR_ATTR_NC << MAIR_ATTR_SHIFT(MAIR_ATTR_IDX_NC)) | | |
748 | (MAIR_ATTR_WBRWA << MAIR_ATTR_SHIFT(MAIR_ATTR_IDX_CACHE)) | | |
749 | (MAIR_ATTR_DEVICE << MAIR_ATTR_SHIFT(MAIR_ATTR_IDX_DEV)); | |
750 | writel_relaxed(reg, cb_base + ARM_SMMU_CB_S1_MAIR0); | |
751 | } | |
752 | ||
753 | /* Nuke the TLB */ | |
754 | writel_relaxed(root_cfg->vmid, gr0_base + ARM_SMMU_GR0_TLBIVMID); | |
755 | arm_smmu_tlb_sync(smmu); | |
756 | ||
757 | /* SCTLR */ | |
758 | reg = SCTLR_CFCFG | SCTLR_CFIE | SCTLR_CFRE | SCTLR_M | SCTLR_EAE_SBOP; | |
759 | if (stage1) | |
760 | reg |= SCTLR_S1_ASIDPNE; | |
761 | #ifdef __BIG_ENDIAN | |
762 | reg |= SCTLR_E; | |
763 | #endif | |
764 | writel(reg, cb_base + ARM_SMMU_CB_SCTLR); | |
765 | } | |
766 | ||
767 | static int arm_smmu_init_domain_context(struct iommu_domain *domain, | |
768 | struct device *dev) | |
769 | { | |
770 | int irq, ret, start; | |
771 | struct arm_smmu_domain *smmu_domain = domain->priv; | |
772 | struct arm_smmu_cfg *root_cfg = &smmu_domain->root_cfg; | |
773 | struct arm_smmu_device *smmu, *parent; | |
774 | ||
775 | /* | |
776 | * Walk the SMMU chain to find the root device for this chain. | |
777 | * We assume that no masters have translations which terminate | |
778 | * early, and therefore check that the root SMMU does indeed have | |
779 | * a StreamID for the master in question. | |
780 | */ | |
781 | parent = dev->archdata.iommu; | |
782 | smmu_domain->output_mask = -1; | |
783 | do { | |
784 | smmu = parent; | |
785 | smmu_domain->output_mask &= (1ULL << smmu->s2_output_size) - 1; | |
786 | } while ((parent = find_parent_smmu(smmu))); | |
787 | ||
788 | if (!find_smmu_master(smmu, dev->of_node)) { | |
789 | dev_err(dev, "unable to find root SMMU for device\n"); | |
790 | return -ENODEV; | |
791 | } | |
792 | ||
793 | ret = __arm_smmu_alloc_bitmap(smmu->vmid_map, 0, ARM_SMMU_NUM_VMIDS); | |
794 | if (IS_ERR_VALUE(ret)) | |
795 | return ret; | |
796 | ||
797 | root_cfg->vmid = ret; | |
798 | if (smmu->features & ARM_SMMU_FEAT_TRANS_NESTED) { | |
799 | /* | |
800 | * We will likely want to change this if/when KVM gets | |
801 | * involved. | |
802 | */ | |
803 | root_cfg->cbar = CBAR_TYPE_S1_TRANS_S2_BYPASS; | |
804 | start = smmu->num_s2_context_banks; | |
805 | } else if (smmu->features & ARM_SMMU_FEAT_TRANS_S2) { | |
806 | root_cfg->cbar = CBAR_TYPE_S2_TRANS; | |
807 | start = 0; | |
808 | } else { | |
809 | root_cfg->cbar = CBAR_TYPE_S1_TRANS_S2_BYPASS; | |
810 | start = smmu->num_s2_context_banks; | |
811 | } | |
812 | ||
813 | ret = __arm_smmu_alloc_bitmap(smmu->context_map, start, | |
814 | smmu->num_context_banks); | |
815 | if (IS_ERR_VALUE(ret)) | |
816 | goto out_free_vmid; | |
817 | ||
818 | root_cfg->cbndx = ret; | |
819 | ||
820 | if (smmu->version == 1) { | |
821 | root_cfg->irptndx = atomic_inc_return(&smmu->irptndx); | |
822 | root_cfg->irptndx %= smmu->num_context_irqs; | |
823 | } else { | |
824 | root_cfg->irptndx = root_cfg->cbndx; | |
825 | } | |
826 | ||
827 | irq = smmu->irqs[smmu->num_global_irqs + root_cfg->irptndx]; | |
828 | ret = request_irq(irq, arm_smmu_context_fault, IRQF_SHARED, | |
829 | "arm-smmu-context-fault", domain); | |
830 | if (IS_ERR_VALUE(ret)) { | |
831 | dev_err(smmu->dev, "failed to request context IRQ %d (%u)\n", | |
832 | root_cfg->irptndx, irq); | |
833 | root_cfg->irptndx = -1; | |
834 | goto out_free_context; | |
835 | } | |
836 | ||
837 | root_cfg->smmu = smmu; | |
838 | arm_smmu_init_context_bank(smmu_domain); | |
839 | return ret; | |
840 | ||
841 | out_free_context: | |
842 | __arm_smmu_free_bitmap(smmu->context_map, root_cfg->cbndx); | |
843 | out_free_vmid: | |
844 | __arm_smmu_free_bitmap(smmu->vmid_map, root_cfg->vmid); | |
845 | return ret; | |
846 | } | |
847 | ||
848 | static void arm_smmu_destroy_domain_context(struct iommu_domain *domain) | |
849 | { | |
850 | struct arm_smmu_domain *smmu_domain = domain->priv; | |
851 | struct arm_smmu_cfg *root_cfg = &smmu_domain->root_cfg; | |
852 | struct arm_smmu_device *smmu = root_cfg->smmu; | |
853 | int irq; | |
854 | ||
855 | if (!smmu) | |
856 | return; | |
857 | ||
858 | if (root_cfg->irptndx != -1) { | |
859 | irq = smmu->irqs[smmu->num_global_irqs + root_cfg->irptndx]; | |
860 | free_irq(irq, domain); | |
861 | } | |
862 | ||
863 | __arm_smmu_free_bitmap(smmu->vmid_map, root_cfg->vmid); | |
864 | __arm_smmu_free_bitmap(smmu->context_map, root_cfg->cbndx); | |
865 | } | |
866 | ||
867 | static int arm_smmu_domain_init(struct iommu_domain *domain) | |
868 | { | |
869 | struct arm_smmu_domain *smmu_domain; | |
870 | pgd_t *pgd; | |
871 | ||
872 | /* | |
873 | * Allocate the domain and initialise some of its data structures. | |
874 | * We can't really do anything meaningful until we've added a | |
875 | * master. | |
876 | */ | |
877 | smmu_domain = kzalloc(sizeof(*smmu_domain), GFP_KERNEL); | |
878 | if (!smmu_domain) | |
879 | return -ENOMEM; | |
880 | ||
881 | pgd = kzalloc(PTRS_PER_PGD * sizeof(pgd_t), GFP_KERNEL); | |
882 | if (!pgd) | |
883 | goto out_free_domain; | |
884 | smmu_domain->root_cfg.pgd = pgd; | |
885 | ||
886 | spin_lock_init(&smmu_domain->lock); | |
887 | domain->priv = smmu_domain; | |
888 | return 0; | |
889 | ||
890 | out_free_domain: | |
891 | kfree(smmu_domain); | |
892 | return -ENOMEM; | |
893 | } | |
894 | ||
895 | static void arm_smmu_free_ptes(pmd_t *pmd) | |
896 | { | |
897 | pgtable_t table = pmd_pgtable(*pmd); | |
898 | pgtable_page_dtor(table); | |
899 | __free_page(table); | |
900 | } | |
901 | ||
902 | static void arm_smmu_free_pmds(pud_t *pud) | |
903 | { | |
904 | int i; | |
905 | pmd_t *pmd, *pmd_base = pmd_offset(pud, 0); | |
906 | ||
907 | pmd = pmd_base; | |
908 | for (i = 0; i < PTRS_PER_PMD; ++i) { | |
909 | if (pmd_none(*pmd)) | |
910 | continue; | |
911 | ||
912 | arm_smmu_free_ptes(pmd); | |
913 | pmd++; | |
914 | } | |
915 | ||
916 | pmd_free(NULL, pmd_base); | |
917 | } | |
918 | ||
919 | static void arm_smmu_free_puds(pgd_t *pgd) | |
920 | { | |
921 | int i; | |
922 | pud_t *pud, *pud_base = pud_offset(pgd, 0); | |
923 | ||
924 | pud = pud_base; | |
925 | for (i = 0; i < PTRS_PER_PUD; ++i) { | |
926 | if (pud_none(*pud)) | |
927 | continue; | |
928 | ||
929 | arm_smmu_free_pmds(pud); | |
930 | pud++; | |
931 | } | |
932 | ||
933 | pud_free(NULL, pud_base); | |
934 | } | |
935 | ||
936 | static void arm_smmu_free_pgtables(struct arm_smmu_domain *smmu_domain) | |
937 | { | |
938 | int i; | |
939 | struct arm_smmu_cfg *root_cfg = &smmu_domain->root_cfg; | |
940 | pgd_t *pgd, *pgd_base = root_cfg->pgd; | |
941 | ||
942 | /* | |
943 | * Recursively free the page tables for this domain. We don't | |
944 | * care about speculative TLB filling, because the TLB will be | |
945 | * nuked next time this context bank is re-allocated and no devices | |
946 | * currently map to these tables. | |
947 | */ | |
948 | pgd = pgd_base; | |
949 | for (i = 0; i < PTRS_PER_PGD; ++i) { | |
950 | if (pgd_none(*pgd)) | |
951 | continue; | |
952 | arm_smmu_free_puds(pgd); | |
953 | pgd++; | |
954 | } | |
955 | ||
956 | kfree(pgd_base); | |
957 | } | |
958 | ||
959 | static void arm_smmu_domain_destroy(struct iommu_domain *domain) | |
960 | { | |
961 | struct arm_smmu_domain *smmu_domain = domain->priv; | |
962 | arm_smmu_destroy_domain_context(domain); | |
963 | arm_smmu_free_pgtables(smmu_domain); | |
964 | kfree(smmu_domain); | |
965 | } | |
966 | ||
967 | static int arm_smmu_master_configure_smrs(struct arm_smmu_device *smmu, | |
968 | struct arm_smmu_master *master) | |
969 | { | |
970 | int i; | |
971 | struct arm_smmu_smr *smrs; | |
972 | void __iomem *gr0_base = ARM_SMMU_GR0(smmu); | |
973 | ||
974 | if (!(smmu->features & ARM_SMMU_FEAT_STREAM_MATCH)) | |
975 | return 0; | |
976 | ||
977 | if (master->smrs) | |
978 | return -EEXIST; | |
979 | ||
980 | smrs = kmalloc(sizeof(*smrs) * master->num_streamids, GFP_KERNEL); | |
981 | if (!smrs) { | |
982 | dev_err(smmu->dev, "failed to allocate %d SMRs for master %s\n", | |
983 | master->num_streamids, master->of_node->name); | |
984 | return -ENOMEM; | |
985 | } | |
986 | ||
987 | /* Allocate the SMRs on the root SMMU */ | |
988 | for (i = 0; i < master->num_streamids; ++i) { | |
989 | int idx = __arm_smmu_alloc_bitmap(smmu->smr_map, 0, | |
990 | smmu->num_mapping_groups); | |
991 | if (IS_ERR_VALUE(idx)) { | |
992 | dev_err(smmu->dev, "failed to allocate free SMR\n"); | |
993 | goto err_free_smrs; | |
994 | } | |
995 | ||
996 | smrs[i] = (struct arm_smmu_smr) { | |
997 | .idx = idx, | |
998 | .mask = 0, /* We don't currently share SMRs */ | |
999 | .id = master->streamids[i], | |
1000 | }; | |
1001 | } | |
1002 | ||
1003 | /* It worked! Now, poke the actual hardware */ | |
1004 | for (i = 0; i < master->num_streamids; ++i) { | |
1005 | u32 reg = SMR_VALID | smrs[i].id << SMR_ID_SHIFT | | |
1006 | smrs[i].mask << SMR_MASK_SHIFT; | |
1007 | writel_relaxed(reg, gr0_base + ARM_SMMU_GR0_SMR(smrs[i].idx)); | |
1008 | } | |
1009 | ||
1010 | master->smrs = smrs; | |
1011 | return 0; | |
1012 | ||
1013 | err_free_smrs: | |
1014 | while (--i >= 0) | |
1015 | __arm_smmu_free_bitmap(smmu->smr_map, smrs[i].idx); | |
1016 | kfree(smrs); | |
1017 | return -ENOSPC; | |
1018 | } | |
1019 | ||
1020 | static void arm_smmu_master_free_smrs(struct arm_smmu_device *smmu, | |
1021 | struct arm_smmu_master *master) | |
1022 | { | |
1023 | int i; | |
1024 | void __iomem *gr0_base = ARM_SMMU_GR0(smmu); | |
1025 | struct arm_smmu_smr *smrs = master->smrs; | |
1026 | ||
1027 | /* Invalidate the SMRs before freeing back to the allocator */ | |
1028 | for (i = 0; i < master->num_streamids; ++i) { | |
1029 | u8 idx = smrs[i].idx; | |
1030 | writel_relaxed(~SMR_VALID, gr0_base + ARM_SMMU_GR0_SMR(idx)); | |
1031 | __arm_smmu_free_bitmap(smmu->smr_map, idx); | |
1032 | } | |
1033 | ||
1034 | master->smrs = NULL; | |
1035 | kfree(smrs); | |
1036 | } | |
1037 | ||
1038 | static void arm_smmu_bypass_stream_mapping(struct arm_smmu_device *smmu, | |
1039 | struct arm_smmu_master *master) | |
1040 | { | |
1041 | int i; | |
1042 | void __iomem *gr0_base = ARM_SMMU_GR0(smmu); | |
1043 | ||
1044 | for (i = 0; i < master->num_streamids; ++i) { | |
1045 | u16 sid = master->streamids[i]; | |
1046 | writel_relaxed(S2CR_TYPE_BYPASS, | |
1047 | gr0_base + ARM_SMMU_GR0_S2CR(sid)); | |
1048 | } | |
1049 | } | |
1050 | ||
1051 | static int arm_smmu_domain_add_master(struct arm_smmu_domain *smmu_domain, | |
1052 | struct arm_smmu_master *master) | |
1053 | { | |
1054 | int i, ret; | |
1055 | struct arm_smmu_device *parent, *smmu = smmu_domain->root_cfg.smmu; | |
1056 | void __iomem *gr0_base = ARM_SMMU_GR0(smmu); | |
1057 | ||
1058 | ret = arm_smmu_master_configure_smrs(smmu, master); | |
1059 | if (ret) | |
1060 | return ret; | |
1061 | ||
1062 | /* Bypass the leaves */ | |
1063 | smmu = smmu_domain->leaf_smmu; | |
1064 | while ((parent = find_parent_smmu(smmu))) { | |
1065 | /* | |
1066 | * We won't have a StreamID match for anything but the root | |
1067 | * smmu, so we only need to worry about StreamID indexing, | |
1068 | * where we must install bypass entries in the S2CRs. | |
1069 | */ | |
1070 | if (smmu->features & ARM_SMMU_FEAT_STREAM_MATCH) | |
1071 | continue; | |
1072 | ||
1073 | arm_smmu_bypass_stream_mapping(smmu, master); | |
1074 | smmu = parent; | |
1075 | } | |
1076 | ||
1077 | /* Now we're at the root, time to point at our context bank */ | |
1078 | for (i = 0; i < master->num_streamids; ++i) { | |
1079 | u32 idx, s2cr; | |
1080 | idx = master->smrs ? master->smrs[i].idx : master->streamids[i]; | |
1081 | s2cr = (S2CR_TYPE_TRANS << S2CR_TYPE_SHIFT) | | |
1082 | (smmu_domain->root_cfg.cbndx << S2CR_CBNDX_SHIFT); | |
1083 | writel_relaxed(s2cr, gr0_base + ARM_SMMU_GR0_S2CR(idx)); | |
1084 | } | |
1085 | ||
1086 | return 0; | |
1087 | } | |
1088 | ||
1089 | static void arm_smmu_domain_remove_master(struct arm_smmu_domain *smmu_domain, | |
1090 | struct arm_smmu_master *master) | |
1091 | { | |
1092 | struct arm_smmu_device *smmu = smmu_domain->root_cfg.smmu; | |
1093 | ||
1094 | /* | |
1095 | * We *must* clear the S2CR first, because freeing the SMR means | |
1096 | * that it can be re-allocated immediately. | |
1097 | */ | |
1098 | arm_smmu_bypass_stream_mapping(smmu, master); | |
1099 | arm_smmu_master_free_smrs(smmu, master); | |
1100 | } | |
1101 | ||
1102 | static int arm_smmu_attach_dev(struct iommu_domain *domain, struct device *dev) | |
1103 | { | |
1104 | int ret = -EINVAL; | |
1105 | struct arm_smmu_domain *smmu_domain = domain->priv; | |
1106 | struct arm_smmu_device *device_smmu = dev->archdata.iommu; | |
1107 | struct arm_smmu_master *master; | |
1108 | ||
1109 | if (!device_smmu) { | |
1110 | dev_err(dev, "cannot attach to SMMU, is it on the same bus?\n"); | |
1111 | return -ENXIO; | |
1112 | } | |
1113 | ||
1114 | /* | |
1115 | * Sanity check the domain. We don't currently support domains | |
1116 | * that cross between different SMMU chains. | |
1117 | */ | |
1118 | spin_lock(&smmu_domain->lock); | |
1119 | if (!smmu_domain->leaf_smmu) { | |
1120 | /* Now that we have a master, we can finalise the domain */ | |
1121 | ret = arm_smmu_init_domain_context(domain, dev); | |
1122 | if (IS_ERR_VALUE(ret)) | |
1123 | goto err_unlock; | |
1124 | ||
1125 | smmu_domain->leaf_smmu = device_smmu; | |
1126 | } else if (smmu_domain->leaf_smmu != device_smmu) { | |
1127 | dev_err(dev, | |
1128 | "cannot attach to SMMU %s whilst already attached to domain on SMMU %s\n", | |
1129 | dev_name(smmu_domain->leaf_smmu->dev), | |
1130 | dev_name(device_smmu->dev)); | |
1131 | goto err_unlock; | |
1132 | } | |
1133 | spin_unlock(&smmu_domain->lock); | |
1134 | ||
1135 | /* Looks ok, so add the device to the domain */ | |
1136 | master = find_smmu_master(smmu_domain->leaf_smmu, dev->of_node); | |
1137 | if (!master) | |
1138 | return -ENODEV; | |
1139 | ||
1140 | return arm_smmu_domain_add_master(smmu_domain, master); | |
1141 | ||
1142 | err_unlock: | |
1143 | spin_unlock(&smmu_domain->lock); | |
1144 | return ret; | |
1145 | } | |
1146 | ||
1147 | static void arm_smmu_detach_dev(struct iommu_domain *domain, struct device *dev) | |
1148 | { | |
1149 | struct arm_smmu_domain *smmu_domain = domain->priv; | |
1150 | struct arm_smmu_master *master; | |
1151 | ||
1152 | master = find_smmu_master(smmu_domain->leaf_smmu, dev->of_node); | |
1153 | if (master) | |
1154 | arm_smmu_domain_remove_master(smmu_domain, master); | |
1155 | } | |
1156 | ||
1157 | static void arm_smmu_flush_pgtable(struct arm_smmu_device *smmu, void *addr, | |
1158 | size_t size) | |
1159 | { | |
1160 | unsigned long offset = (unsigned long)addr & ~PAGE_MASK; | |
1161 | ||
1162 | /* | |
1163 | * If the SMMU can't walk tables in the CPU caches, treat them | |
1164 | * like non-coherent DMA since we need to flush the new entries | |
1165 | * all the way out to memory. There's no possibility of recursion | |
1166 | * here as the SMMU table walker will not be wired through another | |
1167 | * SMMU. | |
1168 | */ | |
1169 | if (!(smmu->features & ARM_SMMU_FEAT_COHERENT_WALK)) | |
1170 | dma_map_page(smmu->dev, virt_to_page(addr), offset, size, | |
1171 | DMA_TO_DEVICE); | |
1172 | } | |
1173 | ||
1174 | static bool arm_smmu_pte_is_contiguous_range(unsigned long addr, | |
1175 | unsigned long end) | |
1176 | { | |
1177 | return !(addr & ~ARM_SMMU_PTE_CONT_MASK) && | |
1178 | (addr + ARM_SMMU_PTE_CONT_SIZE <= end); | |
1179 | } | |
1180 | ||
1181 | static int arm_smmu_alloc_init_pte(struct arm_smmu_device *smmu, pmd_t *pmd, | |
1182 | unsigned long addr, unsigned long end, | |
1183 | unsigned long pfn, int flags, int stage) | |
1184 | { | |
1185 | pte_t *pte, *start; | |
1186 | pteval_t pteval = ARM_SMMU_PTE_PAGE | ARM_SMMU_PTE_AF; | |
1187 | ||
1188 | if (pmd_none(*pmd)) { | |
1189 | /* Allocate a new set of tables */ | |
1190 | pgtable_t table = alloc_page(PGALLOC_GFP); | |
1191 | if (!table) | |
1192 | return -ENOMEM; | |
1193 | ||
1194 | arm_smmu_flush_pgtable(smmu, page_address(table), | |
1195 | ARM_SMMU_PTE_HWTABLE_SIZE); | |
1196 | pgtable_page_ctor(table); | |
1197 | pmd_populate(NULL, pmd, table); | |
1198 | arm_smmu_flush_pgtable(smmu, pmd, sizeof(*pmd)); | |
1199 | } | |
1200 | ||
1201 | if (stage == 1) { | |
1202 | pteval |= ARM_SMMU_PTE_AP_UNPRIV; | |
1203 | if (!(flags & IOMMU_WRITE) && (flags & IOMMU_READ)) | |
1204 | pteval |= ARM_SMMU_PTE_AP_RDONLY; | |
1205 | ||
1206 | if (flags & IOMMU_CACHE) | |
1207 | pteval |= (MAIR_ATTR_IDX_CACHE << | |
1208 | ARM_SMMU_PTE_ATTRINDX_SHIFT); | |
1209 | } else { | |
1210 | pteval |= ARM_SMMU_PTE_HAP_FAULT; | |
1211 | if (flags & IOMMU_READ) | |
1212 | pteval |= ARM_SMMU_PTE_HAP_READ; | |
1213 | if (flags & IOMMU_WRITE) | |
1214 | pteval |= ARM_SMMU_PTE_HAP_WRITE; | |
1215 | if (flags & IOMMU_CACHE) | |
1216 | pteval |= ARM_SMMU_PTE_MEMATTR_OIWB; | |
1217 | else | |
1218 | pteval |= ARM_SMMU_PTE_MEMATTR_NC; | |
1219 | } | |
1220 | ||
1221 | /* If no access, create a faulting entry to avoid TLB fills */ | |
1222 | if (!(flags & (IOMMU_READ | IOMMU_WRITE))) | |
1223 | pteval &= ~ARM_SMMU_PTE_PAGE; | |
1224 | ||
1225 | pteval |= ARM_SMMU_PTE_SH_IS; | |
1226 | start = pmd_page_vaddr(*pmd) + pte_index(addr); | |
1227 | pte = start; | |
1228 | ||
1229 | /* | |
1230 | * Install the page table entries. This is fairly complicated | |
1231 | * since we attempt to make use of the contiguous hint in the | |
1232 | * ptes where possible. The contiguous hint indicates a series | |
1233 | * of ARM_SMMU_PTE_CONT_ENTRIES ptes mapping a physically | |
1234 | * contiguous region with the following constraints: | |
1235 | * | |
1236 | * - The region start is aligned to ARM_SMMU_PTE_CONT_SIZE | |
1237 | * - Each pte in the region has the contiguous hint bit set | |
1238 | * | |
1239 | * This complicates unmapping (also handled by this code, when | |
1240 | * neither IOMMU_READ or IOMMU_WRITE are set) because it is | |
1241 | * possible, yet highly unlikely, that a client may unmap only | |
1242 | * part of a contiguous range. This requires clearing of the | |
1243 | * contiguous hint bits in the range before installing the new | |
1244 | * faulting entries. | |
1245 | * | |
1246 | * Note that re-mapping an address range without first unmapping | |
1247 | * it is not supported, so TLB invalidation is not required here | |
1248 | * and is instead performed at unmap and domain-init time. | |
1249 | */ | |
1250 | do { | |
1251 | int i = 1; | |
1252 | pteval &= ~ARM_SMMU_PTE_CONT; | |
1253 | ||
1254 | if (arm_smmu_pte_is_contiguous_range(addr, end)) { | |
1255 | i = ARM_SMMU_PTE_CONT_ENTRIES; | |
1256 | pteval |= ARM_SMMU_PTE_CONT; | |
1257 | } else if (pte_val(*pte) & | |
1258 | (ARM_SMMU_PTE_CONT | ARM_SMMU_PTE_PAGE)) { | |
1259 | int j; | |
1260 | pte_t *cont_start; | |
1261 | unsigned long idx = pte_index(addr); | |
1262 | ||
1263 | idx &= ~(ARM_SMMU_PTE_CONT_ENTRIES - 1); | |
1264 | cont_start = pmd_page_vaddr(*pmd) + idx; | |
1265 | for (j = 0; j < ARM_SMMU_PTE_CONT_ENTRIES; ++j) | |
1266 | pte_val(*(cont_start + j)) &= ~ARM_SMMU_PTE_CONT; | |
1267 | ||
1268 | arm_smmu_flush_pgtable(smmu, cont_start, | |
1269 | sizeof(*pte) * | |
1270 | ARM_SMMU_PTE_CONT_ENTRIES); | |
1271 | } | |
1272 | ||
1273 | do { | |
1274 | *pte = pfn_pte(pfn, __pgprot(pteval)); | |
1275 | } while (pte++, pfn++, addr += PAGE_SIZE, --i); | |
1276 | } while (addr != end); | |
1277 | ||
1278 | arm_smmu_flush_pgtable(smmu, start, sizeof(*pte) * (pte - start)); | |
1279 | return 0; | |
1280 | } | |
1281 | ||
1282 | static int arm_smmu_alloc_init_pmd(struct arm_smmu_device *smmu, pud_t *pud, | |
1283 | unsigned long addr, unsigned long end, | |
1284 | phys_addr_t phys, int flags, int stage) | |
1285 | { | |
1286 | int ret; | |
1287 | pmd_t *pmd; | |
1288 | unsigned long next, pfn = __phys_to_pfn(phys); | |
1289 | ||
1290 | #ifndef __PAGETABLE_PMD_FOLDED | |
1291 | if (pud_none(*pud)) { | |
1292 | pmd = pmd_alloc_one(NULL, addr); | |
1293 | if (!pmd) | |
1294 | return -ENOMEM; | |
1295 | } else | |
1296 | #endif | |
1297 | pmd = pmd_offset(pud, addr); | |
1298 | ||
1299 | do { | |
1300 | next = pmd_addr_end(addr, end); | |
1301 | ret = arm_smmu_alloc_init_pte(smmu, pmd, addr, end, pfn, | |
1302 | flags, stage); | |
1303 | pud_populate(NULL, pud, pmd); | |
1304 | arm_smmu_flush_pgtable(smmu, pud, sizeof(*pud)); | |
1305 | phys += next - addr; | |
1306 | } while (pmd++, addr = next, addr < end); | |
1307 | ||
1308 | return ret; | |
1309 | } | |
1310 | ||
1311 | static int arm_smmu_alloc_init_pud(struct arm_smmu_device *smmu, pgd_t *pgd, | |
1312 | unsigned long addr, unsigned long end, | |
1313 | phys_addr_t phys, int flags, int stage) | |
1314 | { | |
1315 | int ret = 0; | |
1316 | pud_t *pud; | |
1317 | unsigned long next; | |
1318 | ||
1319 | #ifndef __PAGETABLE_PUD_FOLDED | |
1320 | if (pgd_none(*pgd)) { | |
1321 | pud = pud_alloc_one(NULL, addr); | |
1322 | if (!pud) | |
1323 | return -ENOMEM; | |
1324 | } else | |
1325 | #endif | |
1326 | pud = pud_offset(pgd, addr); | |
1327 | ||
1328 | do { | |
1329 | next = pud_addr_end(addr, end); | |
1330 | ret = arm_smmu_alloc_init_pmd(smmu, pud, addr, next, phys, | |
1331 | flags, stage); | |
1332 | pgd_populate(NULL, pud, pgd); | |
1333 | arm_smmu_flush_pgtable(smmu, pgd, sizeof(*pgd)); | |
1334 | phys += next - addr; | |
1335 | } while (pud++, addr = next, addr < end); | |
1336 | ||
1337 | return ret; | |
1338 | } | |
1339 | ||
1340 | static int arm_smmu_handle_mapping(struct arm_smmu_domain *smmu_domain, | |
1341 | unsigned long iova, phys_addr_t paddr, | |
1342 | size_t size, int flags) | |
1343 | { | |
1344 | int ret, stage; | |
1345 | unsigned long end; | |
1346 | phys_addr_t input_mask, output_mask; | |
1347 | struct arm_smmu_cfg *root_cfg = &smmu_domain->root_cfg; | |
1348 | pgd_t *pgd = root_cfg->pgd; | |
1349 | struct arm_smmu_device *smmu = root_cfg->smmu; | |
1350 | ||
1351 | if (root_cfg->cbar == CBAR_TYPE_S2_TRANS) { | |
1352 | stage = 2; | |
1353 | output_mask = (1ULL << smmu->s2_output_size) - 1; | |
1354 | } else { | |
1355 | stage = 1; | |
1356 | output_mask = (1ULL << smmu->s1_output_size) - 1; | |
1357 | } | |
1358 | ||
1359 | if (!pgd) | |
1360 | return -EINVAL; | |
1361 | ||
1362 | if (size & ~PAGE_MASK) | |
1363 | return -EINVAL; | |
1364 | ||
1365 | input_mask = (1ULL << smmu->input_size) - 1; | |
1366 | if ((phys_addr_t)iova & ~input_mask) | |
1367 | return -ERANGE; | |
1368 | ||
1369 | if (paddr & ~output_mask) | |
1370 | return -ERANGE; | |
1371 | ||
1372 | spin_lock(&smmu_domain->lock); | |
1373 | pgd += pgd_index(iova); | |
1374 | end = iova + size; | |
1375 | do { | |
1376 | unsigned long next = pgd_addr_end(iova, end); | |
1377 | ||
1378 | ret = arm_smmu_alloc_init_pud(smmu, pgd, iova, next, paddr, | |
1379 | flags, stage); | |
1380 | if (ret) | |
1381 | goto out_unlock; | |
1382 | ||
1383 | paddr += next - iova; | |
1384 | iova = next; | |
1385 | } while (pgd++, iova != end); | |
1386 | ||
1387 | out_unlock: | |
1388 | spin_unlock(&smmu_domain->lock); | |
1389 | ||
1390 | /* Ensure new page tables are visible to the hardware walker */ | |
1391 | if (smmu->features & ARM_SMMU_FEAT_COHERENT_WALK) | |
1392 | dsb(); | |
1393 | ||
1394 | return ret; | |
1395 | } | |
1396 | ||
1397 | static int arm_smmu_map(struct iommu_domain *domain, unsigned long iova, | |
1398 | phys_addr_t paddr, size_t size, int flags) | |
1399 | { | |
1400 | struct arm_smmu_domain *smmu_domain = domain->priv; | |
1401 | struct arm_smmu_device *smmu = smmu_domain->leaf_smmu; | |
1402 | ||
1403 | if (!smmu_domain || !smmu) | |
1404 | return -ENODEV; | |
1405 | ||
1406 | /* Check for silent address truncation up the SMMU chain. */ | |
1407 | if ((phys_addr_t)iova & ~smmu_domain->output_mask) | |
1408 | return -ERANGE; | |
1409 | ||
1410 | return arm_smmu_handle_mapping(smmu_domain, iova, paddr, size, flags); | |
1411 | } | |
1412 | ||
1413 | static size_t arm_smmu_unmap(struct iommu_domain *domain, unsigned long iova, | |
1414 | size_t size) | |
1415 | { | |
1416 | int ret; | |
1417 | struct arm_smmu_domain *smmu_domain = domain->priv; | |
1418 | struct arm_smmu_cfg *root_cfg = &smmu_domain->root_cfg; | |
1419 | struct arm_smmu_device *smmu = root_cfg->smmu; | |
1420 | void __iomem *gr0_base = ARM_SMMU_GR0(smmu); | |
1421 | ||
1422 | ret = arm_smmu_handle_mapping(smmu_domain, iova, 0, size, 0); | |
1423 | writel_relaxed(root_cfg->vmid, gr0_base + ARM_SMMU_GR0_TLBIVMID); | |
1424 | arm_smmu_tlb_sync(smmu); | |
1425 | return ret ? ret : size; | |
1426 | } | |
1427 | ||
1428 | static phys_addr_t arm_smmu_iova_to_phys(struct iommu_domain *domain, | |
1429 | dma_addr_t iova) | |
1430 | { | |
1431 | pgd_t *pgd; | |
1432 | pud_t *pud; | |
1433 | pmd_t *pmd; | |
1434 | pte_t *pte; | |
1435 | struct arm_smmu_domain *smmu_domain = domain->priv; | |
1436 | struct arm_smmu_cfg *root_cfg = &smmu_domain->root_cfg; | |
1437 | struct arm_smmu_device *smmu = root_cfg->smmu; | |
1438 | ||
1439 | spin_lock(&smmu_domain->lock); | |
1440 | pgd = root_cfg->pgd; | |
1441 | if (!pgd) | |
1442 | goto err_unlock; | |
1443 | ||
1444 | pgd += pgd_index(iova); | |
1445 | if (pgd_none_or_clear_bad(pgd)) | |
1446 | goto err_unlock; | |
1447 | ||
1448 | pud = pud_offset(pgd, iova); | |
1449 | if (pud_none_or_clear_bad(pud)) | |
1450 | goto err_unlock; | |
1451 | ||
1452 | pmd = pmd_offset(pud, iova); | |
1453 | if (pmd_none_or_clear_bad(pmd)) | |
1454 | goto err_unlock; | |
1455 | ||
1456 | pte = pmd_page_vaddr(*pmd) + pte_index(iova); | |
1457 | if (pte_none(pte)) | |
1458 | goto err_unlock; | |
1459 | ||
1460 | spin_unlock(&smmu_domain->lock); | |
1461 | return __pfn_to_phys(pte_pfn(*pte)) | (iova & ~PAGE_MASK); | |
1462 | ||
1463 | err_unlock: | |
1464 | spin_unlock(&smmu_domain->lock); | |
1465 | dev_warn(smmu->dev, | |
1466 | "invalid (corrupt?) page tables detected for iova 0x%llx\n", | |
1467 | (unsigned long long)iova); | |
1468 | return -EINVAL; | |
1469 | } | |
1470 | ||
1471 | static int arm_smmu_domain_has_cap(struct iommu_domain *domain, | |
1472 | unsigned long cap) | |
1473 | { | |
1474 | unsigned long caps = 0; | |
1475 | struct arm_smmu_domain *smmu_domain = domain->priv; | |
1476 | ||
1477 | if (smmu_domain->root_cfg.smmu->features & ARM_SMMU_FEAT_COHERENT_WALK) | |
1478 | caps |= IOMMU_CAP_CACHE_COHERENCY; | |
1479 | ||
1480 | return !!(cap & caps); | |
1481 | } | |
1482 | ||
1483 | static int arm_smmu_add_device(struct device *dev) | |
1484 | { | |
1485 | struct arm_smmu_device *child, *parent, *smmu; | |
1486 | struct arm_smmu_master *master = NULL; | |
1487 | ||
1488 | spin_lock(&arm_smmu_devices_lock); | |
1489 | list_for_each_entry(parent, &arm_smmu_devices, list) { | |
1490 | smmu = parent; | |
1491 | ||
1492 | /* Try to find a child of the current SMMU. */ | |
1493 | list_for_each_entry(child, &arm_smmu_devices, list) { | |
1494 | if (child->parent_of_node == parent->dev->of_node) { | |
1495 | /* Does the child sit above our master? */ | |
1496 | master = find_smmu_master(child, dev->of_node); | |
1497 | if (master) { | |
1498 | smmu = NULL; | |
1499 | break; | |
1500 | } | |
1501 | } | |
1502 | } | |
1503 | ||
1504 | /* We found some children, so keep searching. */ | |
1505 | if (!smmu) { | |
1506 | master = NULL; | |
1507 | continue; | |
1508 | } | |
1509 | ||
1510 | master = find_smmu_master(smmu, dev->of_node); | |
1511 | if (master) | |
1512 | break; | |
1513 | } | |
1514 | spin_unlock(&arm_smmu_devices_lock); | |
1515 | ||
1516 | if (!master) | |
1517 | return -ENODEV; | |
1518 | ||
1519 | dev->archdata.iommu = smmu; | |
1520 | return 0; | |
1521 | } | |
1522 | ||
1523 | static void arm_smmu_remove_device(struct device *dev) | |
1524 | { | |
1525 | dev->archdata.iommu = NULL; | |
1526 | } | |
1527 | ||
1528 | static struct iommu_ops arm_smmu_ops = { | |
1529 | .domain_init = arm_smmu_domain_init, | |
1530 | .domain_destroy = arm_smmu_domain_destroy, | |
1531 | .attach_dev = arm_smmu_attach_dev, | |
1532 | .detach_dev = arm_smmu_detach_dev, | |
1533 | .map = arm_smmu_map, | |
1534 | .unmap = arm_smmu_unmap, | |
1535 | .iova_to_phys = arm_smmu_iova_to_phys, | |
1536 | .domain_has_cap = arm_smmu_domain_has_cap, | |
1537 | .add_device = arm_smmu_add_device, | |
1538 | .remove_device = arm_smmu_remove_device, | |
1539 | .pgsize_bitmap = (SECTION_SIZE | | |
1540 | ARM_SMMU_PTE_CONT_SIZE | | |
1541 | PAGE_SIZE), | |
1542 | }; | |
1543 | ||
1544 | static void arm_smmu_device_reset(struct arm_smmu_device *smmu) | |
1545 | { | |
1546 | void __iomem *gr0_base = ARM_SMMU_GR0(smmu); | |
1547 | int i = 0; | |
1548 | u32 scr0 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sCR0); | |
1549 | ||
1550 | /* Mark all SMRn as invalid and all S2CRn as bypass */ | |
1551 | for (i = 0; i < smmu->num_mapping_groups; ++i) { | |
1552 | writel_relaxed(~SMR_VALID, gr0_base + ARM_SMMU_GR0_SMR(i)); | |
1553 | writel_relaxed(S2CR_TYPE_BYPASS, gr0_base + ARM_SMMU_GR0_S2CR(i)); | |
1554 | } | |
1555 | ||
1556 | /* Invalidate the TLB, just in case */ | |
1557 | writel_relaxed(0, gr0_base + ARM_SMMU_GR0_STLBIALL); | |
1558 | writel_relaxed(0, gr0_base + ARM_SMMU_GR0_TLBIALLH); | |
1559 | writel_relaxed(0, gr0_base + ARM_SMMU_GR0_TLBIALLNSNH); | |
1560 | ||
1561 | /* Enable fault reporting */ | |
1562 | scr0 |= (sCR0_GFRE | sCR0_GFIE | sCR0_GCFGFRE | sCR0_GCFGFIE); | |
1563 | ||
1564 | /* Disable TLB broadcasting. */ | |
1565 | scr0 |= (sCR0_VMIDPNE | sCR0_PTM); | |
1566 | ||
1567 | /* Enable client access, but bypass when no mapping is found */ | |
1568 | scr0 &= ~(sCR0_CLIENTPD | sCR0_USFCFG); | |
1569 | ||
1570 | /* Disable forced broadcasting */ | |
1571 | scr0 &= ~sCR0_FB; | |
1572 | ||
1573 | /* Don't upgrade barriers */ | |
1574 | scr0 &= ~(sCR0_BSU_MASK << sCR0_BSU_SHIFT); | |
1575 | ||
1576 | /* Push the button */ | |
1577 | arm_smmu_tlb_sync(smmu); | |
1578 | writel(scr0, gr0_base + ARM_SMMU_GR0_sCR0); | |
1579 | } | |
1580 | ||
1581 | static int arm_smmu_id_size_to_bits(int size) | |
1582 | { | |
1583 | switch (size) { | |
1584 | case 0: | |
1585 | return 32; | |
1586 | case 1: | |
1587 | return 36; | |
1588 | case 2: | |
1589 | return 40; | |
1590 | case 3: | |
1591 | return 42; | |
1592 | case 4: | |
1593 | return 44; | |
1594 | case 5: | |
1595 | default: | |
1596 | return 48; | |
1597 | } | |
1598 | } | |
1599 | ||
1600 | static int arm_smmu_device_cfg_probe(struct arm_smmu_device *smmu) | |
1601 | { | |
1602 | unsigned long size; | |
1603 | void __iomem *gr0_base = ARM_SMMU_GR0(smmu); | |
1604 | u32 id; | |
1605 | ||
1606 | dev_notice(smmu->dev, "probing hardware configuration...\n"); | |
1607 | ||
1608 | /* Primecell ID */ | |
1609 | id = readl_relaxed(gr0_base + ARM_SMMU_GR0_PIDR2); | |
1610 | smmu->version = ((id >> PIDR2_ARCH_SHIFT) & PIDR2_ARCH_MASK) + 1; | |
1611 | dev_notice(smmu->dev, "SMMUv%d with:\n", smmu->version); | |
1612 | ||
1613 | /* ID0 */ | |
1614 | id = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID0); | |
1615 | #ifndef CONFIG_64BIT | |
1616 | if (((id >> ID0_PTFS_SHIFT) & ID0_PTFS_MASK) == ID0_PTFS_V8_ONLY) { | |
1617 | dev_err(smmu->dev, "\tno v7 descriptor support!\n"); | |
1618 | return -ENODEV; | |
1619 | } | |
1620 | #endif | |
1621 | if (id & ID0_S1TS) { | |
1622 | smmu->features |= ARM_SMMU_FEAT_TRANS_S1; | |
1623 | dev_notice(smmu->dev, "\tstage 1 translation\n"); | |
1624 | } | |
1625 | ||
1626 | if (id & ID0_S2TS) { | |
1627 | smmu->features |= ARM_SMMU_FEAT_TRANS_S2; | |
1628 | dev_notice(smmu->dev, "\tstage 2 translation\n"); | |
1629 | } | |
1630 | ||
1631 | if (id & ID0_NTS) { | |
1632 | smmu->features |= ARM_SMMU_FEAT_TRANS_NESTED; | |
1633 | dev_notice(smmu->dev, "\tnested translation\n"); | |
1634 | } | |
1635 | ||
1636 | if (!(smmu->features & | |
1637 | (ARM_SMMU_FEAT_TRANS_S1 | ARM_SMMU_FEAT_TRANS_S2 | | |
1638 | ARM_SMMU_FEAT_TRANS_NESTED))) { | |
1639 | dev_err(smmu->dev, "\tno translation support!\n"); | |
1640 | return -ENODEV; | |
1641 | } | |
1642 | ||
1643 | if (id & ID0_CTTW) { | |
1644 | smmu->features |= ARM_SMMU_FEAT_COHERENT_WALK; | |
1645 | dev_notice(smmu->dev, "\tcoherent table walk\n"); | |
1646 | } | |
1647 | ||
1648 | if (id & ID0_SMS) { | |
1649 | u32 smr, sid, mask; | |
1650 | ||
1651 | smmu->features |= ARM_SMMU_FEAT_STREAM_MATCH; | |
1652 | smmu->num_mapping_groups = (id >> ID0_NUMSMRG_SHIFT) & | |
1653 | ID0_NUMSMRG_MASK; | |
1654 | if (smmu->num_mapping_groups == 0) { | |
1655 | dev_err(smmu->dev, | |
1656 | "stream-matching supported, but no SMRs present!\n"); | |
1657 | return -ENODEV; | |
1658 | } | |
1659 | ||
1660 | smr = SMR_MASK_MASK << SMR_MASK_SHIFT; | |
1661 | smr |= (SMR_ID_MASK << SMR_ID_SHIFT); | |
1662 | writel_relaxed(smr, gr0_base + ARM_SMMU_GR0_SMR(0)); | |
1663 | smr = readl_relaxed(gr0_base + ARM_SMMU_GR0_SMR(0)); | |
1664 | ||
1665 | mask = (smr >> SMR_MASK_SHIFT) & SMR_MASK_MASK; | |
1666 | sid = (smr >> SMR_ID_SHIFT) & SMR_ID_MASK; | |
1667 | if ((mask & sid) != sid) { | |
1668 | dev_err(smmu->dev, | |
1669 | "SMR mask bits (0x%x) insufficient for ID field (0x%x)\n", | |
1670 | mask, sid); | |
1671 | return -ENODEV; | |
1672 | } | |
1673 | ||
1674 | dev_notice(smmu->dev, | |
1675 | "\tstream matching with %u register groups, mask 0x%x", | |
1676 | smmu->num_mapping_groups, mask); | |
1677 | } | |
1678 | ||
1679 | /* ID1 */ | |
1680 | id = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID1); | |
1681 | smmu->pagesize = (id & ID1_PAGESIZE) ? SZ_64K : SZ_4K; | |
1682 | ||
1683 | /* Check that we ioremapped enough */ | |
1684 | size = 1 << (((id >> ID1_NUMPAGENDXB_SHIFT) & ID1_NUMPAGENDXB_MASK) + 1); | |
1685 | size *= (smmu->pagesize << 1); | |
1686 | if (smmu->size < size) | |
1687 | dev_warn(smmu->dev, | |
1688 | "device is 0x%lx bytes but only mapped 0x%lx!\n", | |
1689 | size, smmu->size); | |
1690 | ||
1691 | smmu->num_s2_context_banks = (id >> ID1_NUMS2CB_SHIFT) & | |
1692 | ID1_NUMS2CB_MASK; | |
1693 | smmu->num_context_banks = (id >> ID1_NUMCB_SHIFT) & ID1_NUMCB_MASK; | |
1694 | if (smmu->num_s2_context_banks > smmu->num_context_banks) { | |
1695 | dev_err(smmu->dev, "impossible number of S2 context banks!\n"); | |
1696 | return -ENODEV; | |
1697 | } | |
1698 | dev_notice(smmu->dev, "\t%u context banks (%u stage-2 only)\n", | |
1699 | smmu->num_context_banks, smmu->num_s2_context_banks); | |
1700 | ||
1701 | /* ID2 */ | |
1702 | id = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID2); | |
1703 | size = arm_smmu_id_size_to_bits((id >> ID2_IAS_SHIFT) & ID2_IAS_MASK); | |
1704 | ||
1705 | /* | |
1706 | * Stage-1 output limited by stage-2 input size due to pgd | |
1707 | * allocation (PTRS_PER_PGD). | |
1708 | */ | |
1709 | #ifdef CONFIG_64BIT | |
1710 | /* Current maximum output size of 39 bits */ | |
1711 | smmu->s1_output_size = min(39UL, size); | |
1712 | #else | |
1713 | smmu->s1_output_size = min(32UL, size); | |
1714 | #endif | |
1715 | ||
1716 | /* The stage-2 output mask is also applied for bypass */ | |
1717 | size = arm_smmu_id_size_to_bits((id >> ID2_OAS_SHIFT) & ID2_OAS_MASK); | |
1718 | smmu->s2_output_size = min((unsigned long)PHYS_MASK_SHIFT, size); | |
1719 | ||
1720 | if (smmu->version == 1) { | |
1721 | smmu->input_size = 32; | |
1722 | } else { | |
1723 | #ifdef CONFIG_64BIT | |
1724 | size = (id >> ID2_UBS_SHIFT) & ID2_UBS_MASK; | |
1725 | size = min(39, arm_smmu_id_size_to_bits(size)); | |
1726 | #else | |
1727 | size = 32; | |
1728 | #endif | |
1729 | smmu->input_size = size; | |
1730 | ||
1731 | if ((PAGE_SIZE == SZ_4K && !(id & ID2_PTFS_4K)) || | |
1732 | (PAGE_SIZE == SZ_64K && !(id & ID2_PTFS_64K)) || | |
1733 | (PAGE_SIZE != SZ_4K && PAGE_SIZE != SZ_64K)) { | |
1734 | dev_err(smmu->dev, "CPU page size 0x%lx unsupported\n", | |
1735 | PAGE_SIZE); | |
1736 | return -ENODEV; | |
1737 | } | |
1738 | } | |
1739 | ||
1740 | dev_notice(smmu->dev, | |
1741 | "\t%lu-bit VA, %lu-bit IPA, %lu-bit PA\n", | |
1742 | smmu->input_size, smmu->s1_output_size, smmu->s2_output_size); | |
1743 | return 0; | |
1744 | } | |
1745 | ||
1746 | static int arm_smmu_device_dt_probe(struct platform_device *pdev) | |
1747 | { | |
1748 | struct resource *res; | |
1749 | struct arm_smmu_device *smmu; | |
1750 | struct device_node *dev_node; | |
1751 | struct device *dev = &pdev->dev; | |
1752 | struct rb_node *node; | |
1753 | struct of_phandle_args masterspec; | |
1754 | int num_irqs, i, err; | |
1755 | ||
1756 | smmu = devm_kzalloc(dev, sizeof(*smmu), GFP_KERNEL); | |
1757 | if (!smmu) { | |
1758 | dev_err(dev, "failed to allocate arm_smmu_device\n"); | |
1759 | return -ENOMEM; | |
1760 | } | |
1761 | smmu->dev = dev; | |
1762 | ||
1763 | res = platform_get_resource(pdev, IORESOURCE_MEM, 0); | |
1764 | if (!res) { | |
1765 | dev_err(dev, "missing base address/size\n"); | |
1766 | return -ENODEV; | |
1767 | } | |
1768 | ||
1769 | smmu->size = resource_size(res); | |
1770 | smmu->base = devm_request_and_ioremap(dev, res); | |
1771 | if (!smmu->base) | |
1772 | return -EADDRNOTAVAIL; | |
1773 | ||
1774 | if (of_property_read_u32(dev->of_node, "#global-interrupts", | |
1775 | &smmu->num_global_irqs)) { | |
1776 | dev_err(dev, "missing #global-interrupts property\n"); | |
1777 | return -ENODEV; | |
1778 | } | |
1779 | ||
1780 | num_irqs = 0; | |
1781 | while ((res = platform_get_resource(pdev, IORESOURCE_IRQ, num_irqs))) { | |
1782 | num_irqs++; | |
1783 | if (num_irqs > smmu->num_global_irqs) | |
1784 | smmu->num_context_irqs++; | |
1785 | } | |
1786 | ||
1787 | if (num_irqs < smmu->num_global_irqs) { | |
1788 | dev_warn(dev, "found %d interrupts but expected at least %d\n", | |
1789 | num_irqs, smmu->num_global_irqs); | |
1790 | smmu->num_global_irqs = num_irqs; | |
1791 | } | |
1792 | smmu->num_context_irqs = num_irqs - smmu->num_global_irqs; | |
1793 | ||
1794 | smmu->irqs = devm_kzalloc(dev, sizeof(*smmu->irqs) * num_irqs, | |
1795 | GFP_KERNEL); | |
1796 | if (!smmu->irqs) { | |
1797 | dev_err(dev, "failed to allocate %d irqs\n", num_irqs); | |
1798 | return -ENOMEM; | |
1799 | } | |
1800 | ||
1801 | for (i = 0; i < num_irqs; ++i) { | |
1802 | int irq = platform_get_irq(pdev, i); | |
1803 | if (irq < 0) { | |
1804 | dev_err(dev, "failed to get irq index %d\n", i); | |
1805 | return -ENODEV; | |
1806 | } | |
1807 | smmu->irqs[i] = irq; | |
1808 | } | |
1809 | ||
1810 | i = 0; | |
1811 | smmu->masters = RB_ROOT; | |
1812 | while (!of_parse_phandle_with_args(dev->of_node, "mmu-masters", | |
1813 | "#stream-id-cells", i, | |
1814 | &masterspec)) { | |
1815 | err = register_smmu_master(smmu, dev, &masterspec); | |
1816 | if (err) { | |
1817 | dev_err(dev, "failed to add master %s\n", | |
1818 | masterspec.np->name); | |
1819 | goto out_put_masters; | |
1820 | } | |
1821 | ||
1822 | i++; | |
1823 | } | |
1824 | dev_notice(dev, "registered %d master devices\n", i); | |
1825 | ||
1826 | if ((dev_node = of_parse_phandle(dev->of_node, "smmu-parent", 0))) | |
1827 | smmu->parent_of_node = dev_node; | |
1828 | ||
1829 | err = arm_smmu_device_cfg_probe(smmu); | |
1830 | if (err) | |
1831 | goto out_put_parent; | |
1832 | ||
1833 | if (smmu->version > 1 && | |
1834 | smmu->num_context_banks != smmu->num_context_irqs) { | |
1835 | dev_err(dev, | |
1836 | "found only %d context interrupt(s) but %d required\n", | |
1837 | smmu->num_context_irqs, smmu->num_context_banks); | |
1838 | goto out_put_parent; | |
1839 | } | |
1840 | ||
1841 | arm_smmu_device_reset(smmu); | |
1842 | ||
1843 | for (i = 0; i < smmu->num_global_irqs; ++i) { | |
1844 | err = request_irq(smmu->irqs[i], | |
1845 | arm_smmu_global_fault, | |
1846 | IRQF_SHARED, | |
1847 | "arm-smmu global fault", | |
1848 | smmu); | |
1849 | if (err) { | |
1850 | dev_err(dev, "failed to request global IRQ %d (%u)\n", | |
1851 | i, smmu->irqs[i]); | |
1852 | goto out_free_irqs; | |
1853 | } | |
1854 | } | |
1855 | ||
1856 | INIT_LIST_HEAD(&smmu->list); | |
1857 | spin_lock(&arm_smmu_devices_lock); | |
1858 | list_add(&smmu->list, &arm_smmu_devices); | |
1859 | spin_unlock(&arm_smmu_devices_lock); | |
1860 | return 0; | |
1861 | ||
1862 | out_free_irqs: | |
1863 | while (i--) | |
1864 | free_irq(smmu->irqs[i], smmu); | |
1865 | ||
1866 | out_put_parent: | |
1867 | if (smmu->parent_of_node) | |
1868 | of_node_put(smmu->parent_of_node); | |
1869 | ||
1870 | out_put_masters: | |
1871 | for (node = rb_first(&smmu->masters); node; node = rb_next(node)) { | |
1872 | struct arm_smmu_master *master; | |
1873 | master = container_of(node, struct arm_smmu_master, node); | |
1874 | of_node_put(master->of_node); | |
1875 | } | |
1876 | ||
1877 | return err; | |
1878 | } | |
1879 | ||
1880 | static int arm_smmu_device_remove(struct platform_device *pdev) | |
1881 | { | |
1882 | int i; | |
1883 | struct device *dev = &pdev->dev; | |
1884 | struct arm_smmu_device *curr, *smmu = NULL; | |
1885 | struct rb_node *node; | |
1886 | ||
1887 | spin_lock(&arm_smmu_devices_lock); | |
1888 | list_for_each_entry(curr, &arm_smmu_devices, list) { | |
1889 | if (curr->dev == dev) { | |
1890 | smmu = curr; | |
1891 | list_del(&smmu->list); | |
1892 | break; | |
1893 | } | |
1894 | } | |
1895 | spin_unlock(&arm_smmu_devices_lock); | |
1896 | ||
1897 | if (!smmu) | |
1898 | return -ENODEV; | |
1899 | ||
1900 | if (smmu->parent_of_node) | |
1901 | of_node_put(smmu->parent_of_node); | |
1902 | ||
1903 | for (node = rb_first(&smmu->masters); node; node = rb_next(node)) { | |
1904 | struct arm_smmu_master *master; | |
1905 | master = container_of(node, struct arm_smmu_master, node); | |
1906 | of_node_put(master->of_node); | |
1907 | } | |
1908 | ||
1909 | if (!bitmap_empty(smmu->vmid_map, ARM_SMMU_NUM_VMIDS)) | |
1910 | dev_err(dev, "removing device with active domains!\n"); | |
1911 | ||
1912 | for (i = 0; i < smmu->num_global_irqs; ++i) | |
1913 | free_irq(smmu->irqs[i], smmu); | |
1914 | ||
1915 | /* Turn the thing off */ | |
1916 | writel(sCR0_CLIENTPD, ARM_SMMU_GR0(smmu) + ARM_SMMU_GR0_sCR0); | |
1917 | return 0; | |
1918 | } | |
1919 | ||
1920 | #ifdef CONFIG_OF | |
1921 | static struct of_device_id arm_smmu_of_match[] = { | |
1922 | { .compatible = "arm,smmu-v1", }, | |
1923 | { .compatible = "arm,smmu-v2", }, | |
1924 | { .compatible = "arm,mmu-400", }, | |
1925 | { .compatible = "arm,mmu-500", }, | |
1926 | { }, | |
1927 | }; | |
1928 | MODULE_DEVICE_TABLE(of, arm_smmu_of_match); | |
1929 | #endif | |
1930 | ||
1931 | static struct platform_driver arm_smmu_driver = { | |
1932 | .driver = { | |
1933 | .owner = THIS_MODULE, | |
1934 | .name = "arm-smmu", | |
1935 | .of_match_table = of_match_ptr(arm_smmu_of_match), | |
1936 | }, | |
1937 | .probe = arm_smmu_device_dt_probe, | |
1938 | .remove = arm_smmu_device_remove, | |
1939 | }; | |
1940 | ||
1941 | static int __init arm_smmu_init(void) | |
1942 | { | |
1943 | int ret; | |
1944 | ||
1945 | ret = platform_driver_register(&arm_smmu_driver); | |
1946 | if (ret) | |
1947 | return ret; | |
1948 | ||
1949 | /* Oh, for a proper bus abstraction */ | |
1950 | if (!iommu_present(&platform_bus_type)); | |
1951 | bus_set_iommu(&platform_bus_type, &arm_smmu_ops); | |
1952 | ||
1953 | if (!iommu_present(&amba_bustype)); | |
1954 | bus_set_iommu(&amba_bustype, &arm_smmu_ops); | |
1955 | ||
1956 | return 0; | |
1957 | } | |
1958 | ||
1959 | static void __exit arm_smmu_exit(void) | |
1960 | { | |
1961 | return platform_driver_unregister(&arm_smmu_driver); | |
1962 | } | |
1963 | ||
1964 | module_init(arm_smmu_init); | |
1965 | module_exit(arm_smmu_exit); | |
1966 | ||
1967 | MODULE_DESCRIPTION("IOMMU API for ARM architected SMMU implementations"); | |
1968 | MODULE_AUTHOR("Will Deacon <will.deacon@arm.com>"); | |
1969 | MODULE_LICENSE("GPL v2"); |