drivers/iommu/arm-smmu.c

   1 /*
   2  * IOMMU API for ARM architected SMMU implementations.
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License version 2 as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  11  * GNU General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  16  *
  17  * Copyright (C) 2013 ARM Limited
  18  *
  19  * Author: Will Deacon <will.deacon@arm.com>
  20  *
  21  * This driver currently supports:
  22  *      - SMMUv1 and v2 implementations
  23  *      - Stream-matching and stream-indexing
  24  *      - v7/v8 long-descriptor format
  25  *      - Non-secure access to the SMMU
  26  *      - Context fault reporting
  27  */
  28
  29 #define pr_fmt(fmt) "arm-smmu: " fmt
  30
  31 #include <linux/delay.h>
  32 #include <linux/dma-mapping.h>
  33 #include <linux/err.h>
  34 #include <linux/interrupt.h>
  35 #include <linux/io.h>
  36 #include <linux/iommu.h>
  37 #include <linux/iopoll.h>
  38 #include <linux/module.h>
  39 #include <linux/of.h>
  40 #include <linux/pci.h>
  41 #include <linux/platform_device.h>
  42 #include <linux/slab.h>
  43 #include <linux/spinlock.h>
  44
  45 #include <linux/amba/bus.h>
  46
  47 #include "io-pgtable.h"
  48
  49 /* Maximum number of stream IDs assigned to a single device */
  50 #define MAX_MASTER_STREAMIDS            MAX_PHANDLE_ARGS
  51
  52 /* Maximum number of context banks per SMMU */
  53 #define ARM_SMMU_MAX_CBS                128
  54
  55 /* Maximum number of mapping groups per SMMU */
  56 #define ARM_SMMU_MAX_SMRS               128
  57
  58 /* SMMU global address space */
  59 #define ARM_SMMU_GR0(smmu)              ((smmu)->base)
  60 #define ARM_SMMU_GR1(smmu)              ((smmu)->base + (1 << (smmu)->pgshift))
  61
  62 /*
  63  * SMMU global address space with conditional offset to access secure
  64  * aliases of non-secure registers (e.g. nsCR0: 0x400, nsGFSR: 0x448,
  65  * nsGFSYNR0: 0x450)
  66  */
  67 #define ARM_SMMU_GR0_NS(smmu)                                           \
  68         ((smmu)->base +                                                 \
  69                 ((smmu->options & ARM_SMMU_OPT_SECURE_CFG_ACCESS)       \
  70                         ? 0x400 : 0))
  71
  72 /* Configuration registers */
  73 #define ARM_SMMU_GR0_sCR0               0x0
  74 #define sCR0_CLIENTPD                   (1 << 0)
  75 #define sCR0_GFRE                       (1 << 1)
  76 #define sCR0_GFIE                       (1 << 2)
  77 #define sCR0_GCFGFRE                    (1 << 4)
  78 #define sCR0_GCFGFIE                    (1 << 5)
  79 #define sCR0_USFCFG                     (1 << 10)
  80 #define sCR0_VMIDPNE                    (1 << 11)
  81 #define sCR0_PTM                        (1 << 12)
  82 #define sCR0_FB                         (1 << 13)
  83 #define sCR0_BSU_SHIFT                  14
  84 #define sCR0_BSU_MASK                   0x3
  85
  86 /* Identification registers */
  87 #define ARM_SMMU_GR0_ID0                0x20
  88 #define ARM_SMMU_GR0_ID1                0x24
  89 #define ARM_SMMU_GR0_ID2                0x28
  90 #define ARM_SMMU_GR0_ID3                0x2c
  91 #define ARM_SMMU_GR0_ID4                0x30
  92 #define ARM_SMMU_GR0_ID5                0x34
  93 #define ARM_SMMU_GR0_ID6                0x38
  94 #define ARM_SMMU_GR0_ID7                0x3c
  95 #define ARM_SMMU_GR0_sGFSR              0x48
  96 #define ARM_SMMU_GR0_sGFSYNR0           0x50
  97 #define ARM_SMMU_GR0_sGFSYNR1           0x54
  98 #define ARM_SMMU_GR0_sGFSYNR2           0x58
  99
 100 #define ID0_S1TS                        (1 << 30)
 101 #define ID0_S2TS                        (1 << 29)
 102 #define ID0_NTS                         (1 << 28)
 103 #define ID0_SMS                         (1 << 27)
 104 #define ID0_ATOSNS                      (1 << 26)
 105 #define ID0_CTTW                        (1 << 14)
 106 #define ID0_NUMIRPT_SHIFT               16
 107 #define ID0_NUMIRPT_MASK                0xff
 108 #define ID0_NUMSIDB_SHIFT               9
 109 #define ID0_NUMSIDB_MASK                0xf
 110 #define ID0_NUMSMRG_SHIFT               0
 111 #define ID0_NUMSMRG_MASK                0xff
 112
 113 #define ID1_PAGESIZE                    (1 << 31)
 114 #define ID1_NUMPAGENDXB_SHIFT           28
 115 #define ID1_NUMPAGENDXB_MASK            7
 116 #define ID1_NUMS2CB_SHIFT               16
 117 #define ID1_NUMS2CB_MASK                0xff
 118 #define ID1_NUMCB_SHIFT                 0
 119 #define ID1_NUMCB_MASK                  0xff
 120
 121 #define ID2_OAS_SHIFT                   4
 122 #define ID2_OAS_MASK                    0xf
 123 #define ID2_IAS_SHIFT                   0
 124 #define ID2_IAS_MASK                    0xf
 125 #define ID2_UBS_SHIFT                   8
 126 #define ID2_UBS_MASK                    0xf
 127 #define ID2_PTFS_4K                     (1 << 12)
 128 #define ID2_PTFS_16K                    (1 << 13)
 129 #define ID2_PTFS_64K                    (1 << 14)
 130
 131 /* Global TLB invalidation */
 132 #define ARM_SMMU_GR0_TLBIVMID           0x64
 133 #define ARM_SMMU_GR0_TLBIALLNSNH        0x68
 134 #define ARM_SMMU_GR0_TLBIALLH           0x6c
 135 #define ARM_SMMU_GR0_sTLBGSYNC          0x70
 136 #define ARM_SMMU_GR0_sTLBGSTATUS        0x74
 137 #define sTLBGSTATUS_GSACTIVE            (1 << 0)
 138 #define TLB_LOOP_TIMEOUT                1000000 /* 1s! */
 139
 140 /* Stream mapping registers */
 141 #define ARM_SMMU_GR0_SMR(n)             (0x800 + ((n) << 2))
 142 #define SMR_VALID                       (1 << 31)
 143 #define SMR_MASK_SHIFT                  16
 144 #define SMR_MASK_MASK                   0x7fff
 145 #define SMR_ID_SHIFT                    0
 146 #define SMR_ID_MASK                     0x7fff
 147
 148 #define ARM_SMMU_GR0_S2CR(n)            (0xc00 + ((n) << 2))
 149 #define S2CR_CBNDX_SHIFT                0
 150 #define S2CR_CBNDX_MASK                 0xff
 151 #define S2CR_TYPE_SHIFT                 16
 152 #define S2CR_TYPE_MASK                  0x3
 153 #define S2CR_TYPE_TRANS                 (0 << S2CR_TYPE_SHIFT)
 154 #define S2CR_TYPE_BYPASS                (1 << S2CR_TYPE_SHIFT)
 155 #define S2CR_TYPE_FAULT                 (2 << S2CR_TYPE_SHIFT)
 156
 157 /* Context bank attribute registers */
 158 #define ARM_SMMU_GR1_CBAR(n)            (0x0 + ((n) << 2))
 159 #define CBAR_VMID_SHIFT                 0
 160 #define CBAR_VMID_MASK                  0xff
 161 #define CBAR_S1_BPSHCFG_SHIFT           8
 162 #define CBAR_S1_BPSHCFG_MASK            3
 163 #define CBAR_S1_BPSHCFG_NSH             3
 164 #define CBAR_S1_MEMATTR_SHIFT           12
 165 #define CBAR_S1_MEMATTR_MASK            0xf
 166 #define CBAR_S1_MEMATTR_WB              0xf
 167 #define CBAR_TYPE_SHIFT                 16
 168 #define CBAR_TYPE_MASK                  0x3
 169 #define CBAR_TYPE_S2_TRANS              (0 << CBAR_TYPE_SHIFT)
 170 #define CBAR_TYPE_S1_TRANS_S2_BYPASS    (1 << CBAR_TYPE_SHIFT)
 171 #define CBAR_TYPE_S1_TRANS_S2_FAULT     (2 << CBAR_TYPE_SHIFT)
 172 #define CBAR_TYPE_S1_TRANS_S2_TRANS     (3 << CBAR_TYPE_SHIFT)
 173 #define CBAR_IRPTNDX_SHIFT              24
 174 #define CBAR_IRPTNDX_MASK               0xff
 175
 176 #define ARM_SMMU_GR1_CBA2R(n)           (0x800 + ((n) << 2))
 177 #define CBA2R_RW64_32BIT                (0 << 0)
 178 #define CBA2R_RW64_64BIT                (1 << 0)
 179
 180 /* Translation context bank */
 181 #define ARM_SMMU_CB_BASE(smmu)          ((smmu)->base + ((smmu)->size >> 1))
 182 #define ARM_SMMU_CB(smmu, n)            ((n) * (1 << (smmu)->pgshift))
 183
 184 #define ARM_SMMU_CB_SCTLR               0x0
 185 #define ARM_SMMU_CB_RESUME              0x8
 186 #define ARM_SMMU_CB_TTBCR2              0x10
 187 #define ARM_SMMU_CB_TTBR0_LO            0x20
 188 #define ARM_SMMU_CB_TTBR0_HI            0x24
 189 #define ARM_SMMU_CB_TTBR1_LO            0x28
 190 #define ARM_SMMU_CB_TTBR1_HI            0x2c
 191 #define ARM_SMMU_CB_TTBCR               0x30
 192 #define ARM_SMMU_CB_S1_MAIR0            0x38
 193 #define ARM_SMMU_CB_S1_MAIR1            0x3c
 194 #define ARM_SMMU_CB_PAR_LO              0x50
 195 #define ARM_SMMU_CB_PAR_HI              0x54
 196 #define ARM_SMMU_CB_FSR                 0x58
 197 #define ARM_SMMU_CB_FAR_LO              0x60
 198 #define ARM_SMMU_CB_FAR_HI              0x64
 199 #define ARM_SMMU_CB_FSYNR0              0x68
 200 #define ARM_SMMU_CB_S1_TLBIVA           0x600
 201 #define ARM_SMMU_CB_S1_TLBIASID         0x610
 202 #define ARM_SMMU_CB_S1_TLBIVAL          0x620
 203 #define ARM_SMMU_CB_S2_TLBIIPAS2        0x630
 204 #define ARM_SMMU_CB_S2_TLBIIPAS2L       0x638
 205 #define ARM_SMMU_CB_ATS1PR_LO           0x800
 206 #define ARM_SMMU_CB_ATS1PR_HI           0x804
 207 #define ARM_SMMU_CB_ATSR                0x8f0
 208
 209 #define SCTLR_S1_ASIDPNE                (1 << 12)
 210 #define SCTLR_CFCFG                     (1 << 7)
 211 #define SCTLR_CFIE                      (1 << 6)
 212 #define SCTLR_CFRE                      (1 << 5)
 213 #define SCTLR_E                         (1 << 4)
 214 #define SCTLR_AFE                       (1 << 2)
 215 #define SCTLR_TRE                       (1 << 1)
 216 #define SCTLR_M                         (1 << 0)
 217 #define SCTLR_EAE_SBOP                  (SCTLR_AFE | SCTLR_TRE)
 218
 219 #define CB_PAR_F                        (1 << 0)
 220
 221 #define ATSR_ACTIVE                     (1 << 0)
 222
 223 #define RESUME_RETRY                    (0 << 0)
 224 #define RESUME_TERMINATE                (1 << 0)
 225
 226 #define TTBCR2_SEP_SHIFT                15
 227 #define TTBCR2_SEP_MASK                 0x7
 228
 229 #define TTBCR2_ADDR_32                  0
 230 #define TTBCR2_ADDR_36                  1
 231 #define TTBCR2_ADDR_40                  2
 232 #define TTBCR2_ADDR_42                  3
 233 #define TTBCR2_ADDR_44                  4
 234 #define TTBCR2_ADDR_48                  5
 235
 236 #define TTBRn_HI_ASID_SHIFT            16
 237
 238 #define FSR_MULTI                       (1 << 31)
 239 #define FSR_SS                          (1 << 30)
 240 #define FSR_UUT                         (1 << 8)
 241 #define FSR_ASF                         (1 << 7)
 242 #define FSR_TLBLKF                      (1 << 6)
 243 #define FSR_TLBMCF                      (1 << 5)
 244 #define FSR_EF                          (1 << 4)
 245 #define FSR_PF                          (1 << 3)
 246 #define FSR_AFF                         (1 << 2)
 247 #define FSR_TF                          (1 << 1)
 248
 249 #define FSR_IGN                         (FSR_AFF | FSR_ASF | \
 250                                          FSR_TLBMCF | FSR_TLBLKF)
 251 #define FSR_FAULT                       (FSR_MULTI | FSR_SS | FSR_UUT | \
 252                                          FSR_EF | FSR_PF | FSR_TF | FSR_IGN)
 253
 254 #define FSYNR0_WNR                      (1 << 4)
 255
 256 static int force_stage;
 257 module_param_named(force_stage, force_stage, int, S_IRUGO | S_IWUSR);
 258 MODULE_PARM_DESC(force_stage,
 259         "Force SMMU mappings to be installed at a particular stage of translation. A value of '1' or '2' forces the corresponding stage. All other values are ignored (i.e. no stage is forced). Note that selecting a specific stage will disable support for nested translation.");
 260
 261 enum arm_smmu_arch_version {
 262         ARM_SMMU_V1 = 1,
 263         ARM_SMMU_V2,
 264 };
 265
 266 struct arm_smmu_smr {
 267         u8                              idx;
 268         u16                             mask;
 269         u16                             id;
 270 };
 271
 272 struct arm_smmu_master_cfg {
 273         int                             num_streamids;
 274         u16                             streamids[MAX_MASTER_STREAMIDS];
 275         struct arm_smmu_smr             *smrs;
 276 };
 277
 278 struct arm_smmu_master {
 279         struct device_node              *of_node;
 280         struct rb_node                  node;
 281         struct arm_smmu_master_cfg      cfg;
 282 };
 283
 284 struct arm_smmu_device {
 285         struct device                   *dev;
 286
 287         void __iomem                    *base;
 288         unsigned long                   size;
 289         unsigned long                   pgshift;
 290
 291 #define ARM_SMMU_FEAT_COHERENT_WALK     (1 << 0)
 292 #define ARM_SMMU_FEAT_STREAM_MATCH      (1 << 1)
 293 #define ARM_SMMU_FEAT_TRANS_S1          (1 << 2)
 294 #define ARM_SMMU_FEAT_TRANS_S2          (1 << 3)
 295 #define ARM_SMMU_FEAT_TRANS_NESTED      (1 << 4)
 296 #define ARM_SMMU_FEAT_TRANS_OPS         (1 << 5)
 297         u32                             features;
 298
 299 #define ARM_SMMU_OPT_SECURE_CFG_ACCESS (1 << 0)
 300         u32                             options;
 301         enum arm_smmu_arch_version      version;
 302
 303         u32                             num_context_banks;
 304         u32                             num_s2_context_banks;
 305         DECLARE_BITMAP(context_map, ARM_SMMU_MAX_CBS);
 306         atomic_t                        irptndx;
 307
 308         u32                             num_mapping_groups;
 309         DECLARE_BITMAP(smr_map, ARM_SMMU_MAX_SMRS);
 310
 311         unsigned long                   va_size;
 312         unsigned long                   ipa_size;
 313         unsigned long                   pa_size;
 314
 315         u32                             num_global_irqs;
 316         u32                             num_context_irqs;
 317         unsigned int                    *irqs;
 318
 319         struct list_head                list;
 320         struct rb_root                  masters;
 321 };
 322
 323 struct arm_smmu_cfg {
 324         u8                              cbndx;
 325         u8                              irptndx;
 326         u32                             cbar;
 327 };
 328 #define INVALID_IRPTNDX                 0xff
 329
 330 #define ARM_SMMU_CB_ASID(cfg)           ((cfg)->cbndx)
 331 #define ARM_SMMU_CB_VMID(cfg)           ((cfg)->cbndx + 1)
 332
 333 enum arm_smmu_domain_stage {
 334         ARM_SMMU_DOMAIN_S1 = 0,
 335         ARM_SMMU_DOMAIN_S2,
 336         ARM_SMMU_DOMAIN_NESTED,
 337 };
 338
 339 struct arm_smmu_domain {
 340         struct arm_smmu_device          *smmu;
 341         struct io_pgtable_ops           *pgtbl_ops;
 342         spinlock_t                      pgtbl_lock;
 343         struct arm_smmu_cfg             cfg;
 344         enum arm_smmu_domain_stage      stage;
 345         struct mutex                    init_mutex; /* Protects smmu pointer */
 346 };
 347
 348 static struct iommu_ops arm_smmu_ops;
 349
 350 static DEFINE_SPINLOCK(arm_smmu_devices_lock);
 351 static LIST_HEAD(arm_smmu_devices);
 352
 353 struct arm_smmu_option_prop {
 354         u32 opt;
 355         const char *prop;
 356 };
 357
 358 static struct arm_smmu_option_prop arm_smmu_options[] = {
 359         { ARM_SMMU_OPT_SECURE_CFG_ACCESS, "calxeda,smmu-secure-config-access" },
 360         { 0, NULL},
 361 };
 362
 363 static void parse_driver_options(struct arm_smmu_device *smmu)
 364 {
 365         int i = 0;
 366
 367         do {
 368                 if (of_property_read_bool(smmu->dev->of_node,
 369                                                 arm_smmu_options[i].prop)) {
 370                         smmu->options |= arm_smmu_options[i].opt;
 371                         dev_notice(smmu->dev, "option %s\n",
 372                                 arm_smmu_options[i].prop);
 373                 }
 374         } while (arm_smmu_options[++i].opt);
 375 }
 376
 377 static struct device_node *dev_get_dev_node(struct device *dev)
 378 {
 379         if (dev_is_pci(dev)) {
 380                 struct pci_bus *bus = to_pci_dev(dev)->bus;
 381
 382                 while (!pci_is_root_bus(bus))
 383                         bus = bus->parent;
 384                 return bus->bridge->parent->of_node;
 385         }
 386
 387         return dev->of_node;
 388 }
 389
 390 static struct arm_smmu_master *find_smmu_master(struct arm_smmu_device *smmu,
 391                                                 struct device_node *dev_node)
 392 {
 393         struct rb_node *node = smmu->masters.rb_node;
 394
 395         while (node) {
 396                 struct arm_smmu_master *master;
 397
 398                 master = container_of(node, struct arm_smmu_master, node);
 399
 400                 if (dev_node < master->of_node)
 401                         node = node->rb_left;
 402                 else if (dev_node > master->of_node)
 403                         node = node->rb_right;
 404                 else
 405                         return master;
 406         }
 407
 408         return NULL;
 409 }
 410
 411 static struct arm_smmu_master_cfg *
 412 find_smmu_master_cfg(struct device *dev)
 413 {
 414         struct arm_smmu_master_cfg *cfg = NULL;
 415         struct iommu_group *group = iommu_group_get(dev);
 416
 417         if (group) {
 418                 cfg = iommu_group_get_iommudata(group);
 419                 iommu_group_put(group);
 420         }
 421
 422         return cfg;
 423 }
 424
 425 static int insert_smmu_master(struct arm_smmu_device *smmu,
 426                               struct arm_smmu_master *master)
 427 {
 428         struct rb_node **new, *parent;
 429
 430         new = &smmu->masters.rb_node;
 431         parent = NULL;
 432         while (*new) {
 433                 struct arm_smmu_master *this
 434                         = container_of(*new, struct arm_smmu_master, node);
 435
 436                 parent = *new;
 437                 if (master->of_node < this->of_node)
 438                         new = &((*new)->rb_left);
 439                 else if (master->of_node > this->of_node)
 440                         new = &((*new)->rb_right);
 441                 else
 442                         return -EEXIST;
 443         }
 444
 445         rb_link_node(&master->node, parent, new);
 446         rb_insert_color(&master->node, &smmu->masters);
 447         return 0;
 448 }
 449
 450 static int register_smmu_master(struct arm_smmu_device *smmu,
 451                                 struct device *dev,
 452                                 struct of_phandle_args *masterspec)
 453 {
 454         int i;
 455         struct arm_smmu_master *master;
 456
 457         master = find_smmu_master(smmu, masterspec->np);
 458         if (master) {
 459                 dev_err(dev,
 460                         "rejecting multiple registrations for master device %s\n",
 461                         masterspec->np->name);
 462                 return -EBUSY;
 463         }
 464
 465         if (masterspec->args_count > MAX_MASTER_STREAMIDS) {
 466                 dev_err(dev,
 467                         "reached maximum number (%d) of stream IDs for master device %s\n",
 468                         MAX_MASTER_STREAMIDS, masterspec->np->name);
 469                 return -ENOSPC;
 470         }
 471
 472         master = devm_kzalloc(dev, sizeof(*master), GFP_KERNEL);
 473         if (!master)
 474                 return -ENOMEM;
 475
 476         master->of_node                 = masterspec->np;
 477         master->cfg.num_streamids       = masterspec->args_count;
 478
 479         for (i = 0; i < master->cfg.num_streamids; ++i) {
 480                 u16 streamid = masterspec->args[i];
 481
 482                 if (!(smmu->features & ARM_SMMU_FEAT_STREAM_MATCH) &&
 483                      (streamid >= smmu->num_mapping_groups)) {
 484                         dev_err(dev,
 485                                 "stream ID for master device %s greater than maximum allowed (%d)\n",
 486                                 masterspec->np->name, smmu->num_mapping_groups);
 487                         return -ERANGE;
 488                 }
 489                 master->cfg.streamids[i] = streamid;
 490         }
 491         return insert_smmu_master(smmu, master);
 492 }
 493
 494 static struct arm_smmu_device *find_smmu_for_device(struct device *dev)
 495 {
 496         struct arm_smmu_device *smmu;
 497         struct arm_smmu_master *master = NULL;
 498         struct device_node *dev_node = dev_get_dev_node(dev);
 499
 500         spin_lock(&arm_smmu_devices_lock);
 501         list_for_each_entry(smmu, &arm_smmu_devices, list) {
 502                 master = find_smmu_master(smmu, dev_node);
 503                 if (master)
 504                         break;
 505         }
 506         spin_unlock(&arm_smmu_devices_lock);
 507
 508         return master ? smmu : NULL;
 509 }
 510
 511 static int __arm_smmu_alloc_bitmap(unsigned long *map, int start, int end)
 512 {
 513         int idx;
 514
 515         do {
 516                 idx = find_next_zero_bit(map, end, start);
 517                 if (idx == end)
 518                         return -ENOSPC;
 519         } while (test_and_set_bit(idx, map));
 520
 521         return idx;
 522 }
 523
 524 static void __arm_smmu_free_bitmap(unsigned long *map, int idx)
 525 {
 526         clear_bit(idx, map);
 527 }
 528
 529 /* Wait for any pending TLB invalidations to complete */
 530 static void __arm_smmu_tlb_sync(struct arm_smmu_device *smmu)
 531 {
 532         int count = 0;
 533         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
 534
 535         writel_relaxed(0, gr0_base + ARM_SMMU_GR0_sTLBGSYNC);
 536         while (readl_relaxed(gr0_base + ARM_SMMU_GR0_sTLBGSTATUS)
 537                & sTLBGSTATUS_GSACTIVE) {
 538                 cpu_relax();
 539                 if (++count == TLB_LOOP_TIMEOUT) {
 540                         dev_err_ratelimited(smmu->dev,
 541                         "TLB sync timed out -- SMMU may be deadlocked\n");
 542                         return;
 543                 }
 544                 udelay(1);
 545         }
 546 }
 547
 548 static void arm_smmu_tlb_sync(void *cookie)
 549 {
 550         struct arm_smmu_domain *smmu_domain = cookie;
 551         __arm_smmu_tlb_sync(smmu_domain->smmu);
 552 }
 553
 554 static void arm_smmu_tlb_inv_context(void *cookie)
 555 {
 556         struct arm_smmu_domain *smmu_domain = cookie;
 557         struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
 558         struct arm_smmu_device *smmu = smmu_domain->smmu;
 559         bool stage1 = cfg->cbar != CBAR_TYPE_S2_TRANS;
 560         void __iomem *base;
 561
 562         if (stage1) {
 563                 base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
 564                 writel_relaxed(ARM_SMMU_CB_ASID(cfg),
 565                                base + ARM_SMMU_CB_S1_TLBIASID);
 566         } else {
 567                 base = ARM_SMMU_GR0(smmu);
 568                 writel_relaxed(ARM_SMMU_CB_VMID(cfg),
 569                                base + ARM_SMMU_GR0_TLBIVMID);
 570         }
 571
 572         __arm_smmu_tlb_sync(smmu);
 573 }
 574
 575 static void arm_smmu_tlb_inv_range_nosync(unsigned long iova, size_t size,
 576                                           bool leaf, void *cookie)
 577 {
 578         struct arm_smmu_domain *smmu_domain = cookie;
 579         struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
 580         struct arm_smmu_device *smmu = smmu_domain->smmu;
 581         bool stage1 = cfg->cbar != CBAR_TYPE_S2_TRANS;
 582         void __iomem *reg;
 583
 584         if (stage1) {
 585                 reg = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
 586                 reg += leaf ? ARM_SMMU_CB_S1_TLBIVAL : ARM_SMMU_CB_S1_TLBIVA;
 587
 588                 if (!IS_ENABLED(CONFIG_64BIT) || smmu->version == ARM_SMMU_V1) {
 589                         iova &= ~12UL;
 590                         iova |= ARM_SMMU_CB_ASID(cfg);
 591                         writel_relaxed(iova, reg);
 592 #ifdef CONFIG_64BIT
 593                 } else {
 594                         iova >>= 12;
 595                         iova |= (u64)ARM_SMMU_CB_ASID(cfg) << 48;
 596                         writeq_relaxed(iova, reg);
 597 #endif
 598                 }
 599 #ifdef CONFIG_64BIT
 600         } else if (smmu->version == ARM_SMMU_V2) {
 601                 reg = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
 602                 reg += leaf ? ARM_SMMU_CB_S2_TLBIIPAS2L :
 603                               ARM_SMMU_CB_S2_TLBIIPAS2;
 604                 writeq_relaxed(iova >> 12, reg);
 605 #endif
 606         } else {
 607                 reg = ARM_SMMU_GR0(smmu) + ARM_SMMU_GR0_TLBIVMID;
 608                 writel_relaxed(ARM_SMMU_CB_VMID(cfg), reg);
 609         }
 610 }
 611
 612 static void arm_smmu_flush_pgtable(void *addr, size_t size, void *cookie)
 613 {
 614         struct arm_smmu_domain *smmu_domain = cookie;
 615         struct arm_smmu_device *smmu = smmu_domain->smmu;
 616         unsigned long offset = (unsigned long)addr & ~PAGE_MASK;
 617
 618
 619         /* Ensure new page tables are visible to the hardware walker */
 620         if (smmu->features & ARM_SMMU_FEAT_COHERENT_WALK) {
 621                 dsb(ishst);
 622         } else {
 623                 /*
 624                  * If the SMMU can't walk tables in the CPU caches, treat them
 625                  * like non-coherent DMA since we need to flush the new entries
 626                  * all the way out to memory. There's no possibility of
 627                  * recursion here as the SMMU table walker will not be wired
 628                  * through another SMMU.
 629                  */
 630                 dma_map_page(smmu->dev, virt_to_page(addr), offset, size,
 631                              DMA_TO_DEVICE);
 632         }
 633 }
 634
 635 static struct iommu_gather_ops arm_smmu_gather_ops = {
 636         .tlb_flush_all  = arm_smmu_tlb_inv_context,
 637         .tlb_add_flush  = arm_smmu_tlb_inv_range_nosync,
 638         .tlb_sync       = arm_smmu_tlb_sync,
 639         .flush_pgtable  = arm_smmu_flush_pgtable,
 640 };
 641
 642 static irqreturn_t arm_smmu_context_fault(int irq, void *dev)
 643 {
 644         int flags, ret;
 645         u32 fsr, far, fsynr, resume;
 646         unsigned long iova;
 647         struct iommu_domain *domain = dev;
 648         struct arm_smmu_domain *smmu_domain = domain->priv;
 649         struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
 650         struct arm_smmu_device *smmu = smmu_domain->smmu;
 651         void __iomem *cb_base;
 652
 653         cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
 654         fsr = readl_relaxed(cb_base + ARM_SMMU_CB_FSR);
 655
 656         if (!(fsr & FSR_FAULT))
 657                 return IRQ_NONE;
 658
 659         if (fsr & FSR_IGN)
 660                 dev_err_ratelimited(smmu->dev,
 661                                     "Unexpected context fault (fsr 0x%x)\n",
 662                                     fsr);
 663
 664         fsynr = readl_relaxed(cb_base + ARM_SMMU_CB_FSYNR0);
 665         flags = fsynr & FSYNR0_WNR ? IOMMU_FAULT_WRITE : IOMMU_FAULT_READ;
 666
 667         far = readl_relaxed(cb_base + ARM_SMMU_CB_FAR_LO);
 668         iova = far;
 669 #ifdef CONFIG_64BIT
 670         far = readl_relaxed(cb_base + ARM_SMMU_CB_FAR_HI);
 671         iova |= ((unsigned long)far << 32);
 672 #endif
 673
 674         if (!report_iommu_fault(domain, smmu->dev, iova, flags)) {
 675                 ret = IRQ_HANDLED;
 676                 resume = RESUME_RETRY;
 677         } else {
 678                 dev_err_ratelimited(smmu->dev,
 679                     "Unhandled context fault: iova=0x%08lx, fsynr=0x%x, cb=%d\n",
 680                     iova, fsynr, cfg->cbndx);
 681                 ret = IRQ_NONE;
 682                 resume = RESUME_TERMINATE;
 683         }
 684
 685         /* Clear the faulting FSR */
 686         writel(fsr, cb_base + ARM_SMMU_CB_FSR);
 687
 688         /* Retry or terminate any stalled transactions */
 689         if (fsr & FSR_SS)
 690                 writel_relaxed(resume, cb_base + ARM_SMMU_CB_RESUME);
 691
 692         return ret;
 693 }
 694
 695 static irqreturn_t arm_smmu_global_fault(int irq, void *dev)
 696 {
 697         u32 gfsr, gfsynr0, gfsynr1, gfsynr2;
 698         struct arm_smmu_device *smmu = dev;
 699         void __iomem *gr0_base = ARM_SMMU_GR0_NS(smmu);
 700
 701         gfsr = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSR);
 702         gfsynr0 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSYNR0);
 703         gfsynr1 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSYNR1);
 704         gfsynr2 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSYNR2);
 705
 706         if (!gfsr)
 707                 return IRQ_NONE;
 708
 709         dev_err_ratelimited(smmu->dev,
 710                 "Unexpected global fault, this could be serious\n");
 711         dev_err_ratelimited(smmu->dev,
 712                 "\tGFSR 0x%08x, GFSYNR0 0x%08x, GFSYNR1 0x%08x, GFSYNR2 0x%08x\n",
 713                 gfsr, gfsynr0, gfsynr1, gfsynr2);
 714
 715         writel(gfsr, gr0_base + ARM_SMMU_GR0_sGFSR);
 716         return IRQ_HANDLED;
 717 }
 718
 719 static void arm_smmu_init_context_bank(struct arm_smmu_domain *smmu_domain,
 720                                        struct io_pgtable_cfg *pgtbl_cfg)
 721 {
 722         u32 reg;
 723         bool stage1;
 724         struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
 725         struct arm_smmu_device *smmu = smmu_domain->smmu;
 726         void __iomem *cb_base, *gr0_base, *gr1_base;
 727
 728         gr0_base = ARM_SMMU_GR0(smmu);
 729         gr1_base = ARM_SMMU_GR1(smmu);
 730         stage1 = cfg->cbar != CBAR_TYPE_S2_TRANS;
 731         cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
 732
 733         /* CBAR */
 734         reg = cfg->cbar;
 735         if (smmu->version == ARM_SMMU_V1)
 736                 reg |= cfg->irptndx << CBAR_IRPTNDX_SHIFT;
 737
 738         /*
 739          * Use the weakest shareability/memory types, so they are
 740          * overridden by the ttbcr/pte.
 741          */
 742         if (stage1) {
 743                 reg |= (CBAR_S1_BPSHCFG_NSH << CBAR_S1_BPSHCFG_SHIFT) |
 744                         (CBAR_S1_MEMATTR_WB << CBAR_S1_MEMATTR_SHIFT);
 745         } else {
 746                 reg |= ARM_SMMU_CB_VMID(cfg) << CBAR_VMID_SHIFT;
 747         }
 748         writel_relaxed(reg, gr1_base + ARM_SMMU_GR1_CBAR(cfg->cbndx));
 749
 750         if (smmu->version > ARM_SMMU_V1) {
 751                 /* CBA2R */
 752 #ifdef CONFIG_64BIT
 753                 reg = CBA2R_RW64_64BIT;
 754 #else
 755                 reg = CBA2R_RW64_32BIT;
 756 #endif
 757                 writel_relaxed(reg, gr1_base + ARM_SMMU_GR1_CBA2R(cfg->cbndx));
 758         }
 759
 760         /* TTBRs */
 761         if (stage1) {
 762                 reg = pgtbl_cfg->arm_lpae_s1_cfg.ttbr[0];
 763                 writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBR0_LO);
 764                 reg = pgtbl_cfg->arm_lpae_s1_cfg.ttbr[0] >> 32;
 765                 reg |= ARM_SMMU_CB_ASID(cfg) << TTBRn_HI_ASID_SHIFT;
 766                 writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBR0_HI);
 767
 768                 reg = pgtbl_cfg->arm_lpae_s1_cfg.ttbr[1];
 769                 writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBR1_LO);
 770                 reg = pgtbl_cfg->arm_lpae_s1_cfg.ttbr[1] >> 32;
 771                 reg |= ARM_SMMU_CB_ASID(cfg) << TTBRn_HI_ASID_SHIFT;
 772                 writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBR1_HI);
 773         } else {
 774                 reg = pgtbl_cfg->arm_lpae_s2_cfg.vttbr;
 775                 writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBR0_LO);
 776                 reg = pgtbl_cfg->arm_lpae_s2_cfg.vttbr >> 32;
 777                 writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBR0_HI);
 778         }
 779
 780         /* TTBCR */
 781         if (stage1) {
 782                 reg = pgtbl_cfg->arm_lpae_s1_cfg.tcr;
 783                 writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBCR);
 784                 if (smmu->version > ARM_SMMU_V1) {
 785                         reg = pgtbl_cfg->arm_lpae_s1_cfg.tcr >> 32;
 786                         switch (smmu->va_size) {
 787                         case 32:
 788                                 reg |= (TTBCR2_ADDR_32 << TTBCR2_SEP_SHIFT);
 789                                 break;
 790                         case 36:
 791                                 reg |= (TTBCR2_ADDR_36 << TTBCR2_SEP_SHIFT);
 792                                 break;
 793                         case 40:
 794                                 reg |= (TTBCR2_ADDR_40 << TTBCR2_SEP_SHIFT);
 795                                 break;
 796                         case 42:
 797                                 reg |= (TTBCR2_ADDR_42 << TTBCR2_SEP_SHIFT);
 798                                 break;
 799                         case 44:
 800                                 reg |= (TTBCR2_ADDR_44 << TTBCR2_SEP_SHIFT);
 801                                 break;
 802                         case 48:
 803                                 reg |= (TTBCR2_ADDR_48 << TTBCR2_SEP_SHIFT);
 804                                 break;
 805                         }
 806                         writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBCR2);
 807                 }
 808         } else {
 809                 reg = pgtbl_cfg->arm_lpae_s2_cfg.vtcr;
 810                 writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBCR);
 811         }
 812
 813         /* MAIRs (stage-1 only) */
 814         if (stage1) {
 815                 reg = pgtbl_cfg->arm_lpae_s1_cfg.mair[0];
 816                 writel_relaxed(reg, cb_base + ARM_SMMU_CB_S1_MAIR0);
 817                 reg = pgtbl_cfg->arm_lpae_s1_cfg.mair[1];
 818                 writel_relaxed(reg, cb_base + ARM_SMMU_CB_S1_MAIR1);
 819         }
 820
 821         /* SCTLR */
 822         reg = SCTLR_CFCFG | SCTLR_CFIE | SCTLR_CFRE | SCTLR_M | SCTLR_EAE_SBOP;
 823         if (stage1)
 824                 reg |= SCTLR_S1_ASIDPNE;
 825 #ifdef __BIG_ENDIAN
 826         reg |= SCTLR_E;
 827 #endif
 828         writel_relaxed(reg, cb_base + ARM_SMMU_CB_SCTLR);
 829 }
 830
 831 static int arm_smmu_init_domain_context(struct iommu_domain *domain,
 832                                         struct arm_smmu_device *smmu)
 833 {
 834         int irq, start, ret = 0;
 835         unsigned long ias, oas;
 836         struct io_pgtable_ops *pgtbl_ops;
 837         struct io_pgtable_cfg pgtbl_cfg;
 838         enum io_pgtable_fmt fmt;
 839         struct arm_smmu_domain *smmu_domain = domain->priv;
 840         struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
 841
 842         mutex_lock(&smmu_domain->init_mutex);
 843         if (smmu_domain->smmu)
 844                 goto out_unlock;
 845
 846         /*
 847          * Mapping the requested stage onto what we support is surprisingly
 848          * complicated, mainly because the spec allows S1+S2 SMMUs without
 849          * support for nested translation. That means we end up with the
 850          * following table:
 851          *
 852          * Requested        Supported        Actual
 853          *     S1               N              S1
 854          *     S1             S1+S2            S1
 855          *     S1               S2             S2
 856          *     S1               S1             S1
 857          *     N                N              N
 858          *     N              S1+S2            S2
 859          *     N                S2             S2
 860          *     N                S1             S1
 861          *
 862          * Note that you can't actually request stage-2 mappings.
 863          */
 864         if (!(smmu->features & ARM_SMMU_FEAT_TRANS_S1))
 865                 smmu_domain->stage = ARM_SMMU_DOMAIN_S2;
 866         if (!(smmu->features & ARM_SMMU_FEAT_TRANS_S2))
 867                 smmu_domain->stage = ARM_SMMU_DOMAIN_S1;
 868
 869         switch (smmu_domain->stage) {
 870         case ARM_SMMU_DOMAIN_S1:
 871                 cfg->cbar = CBAR_TYPE_S1_TRANS_S2_BYPASS;
 872                 start = smmu->num_s2_context_banks;
 873                 ias = smmu->va_size;
 874                 oas = smmu->ipa_size;
 875                 if (IS_ENABLED(CONFIG_64BIT))
 876                         fmt = ARM_64_LPAE_S1;
 877                 else
 878                         fmt = ARM_32_LPAE_S1;
 879                 break;
 880         case ARM_SMMU_DOMAIN_NESTED:
 881                 /*
 882                  * We will likely want to change this if/when KVM gets
 883                  * involved.
 884                  */
 885         case ARM_SMMU_DOMAIN_S2:
 886                 cfg->cbar = CBAR_TYPE_S2_TRANS;
 887                 start = 0;
 888                 ias = smmu->ipa_size;
 889                 oas = smmu->pa_size;
 890                 if (IS_ENABLED(CONFIG_64BIT))
 891                         fmt = ARM_64_LPAE_S2;
 892                 else
 893                         fmt = ARM_32_LPAE_S2;
 894                 break;
 895         default:
 896                 ret = -EINVAL;
 897                 goto out_unlock;
 898         }
 899
 900         ret = __arm_smmu_alloc_bitmap(smmu->context_map, start,
 901                                       smmu->num_context_banks);
 902         if (IS_ERR_VALUE(ret))
 903                 goto out_unlock;
 904
 905         cfg->cbndx = ret;
 906         if (smmu->version == ARM_SMMU_V1) {
 907                 cfg->irptndx = atomic_inc_return(&smmu->irptndx);
 908                 cfg->irptndx %= smmu->num_context_irqs;
 909         } else {
 910                 cfg->irptndx = cfg->cbndx;
 911         }
 912
 913         pgtbl_cfg = (struct io_pgtable_cfg) {
 914                 .pgsize_bitmap  = arm_smmu_ops.pgsize_bitmap,
 915                 .ias            = ias,
 916                 .oas            = oas,
 917                 .tlb            = &arm_smmu_gather_ops,
 918         };
 919
 920         smmu_domain->smmu = smmu;
 921         pgtbl_ops = alloc_io_pgtable_ops(fmt, &pgtbl_cfg, smmu_domain);
 922         if (!pgtbl_ops) {
 923                 ret = -ENOMEM;
 924                 goto out_clear_smmu;
 925         }
 926
 927         /* Update our support page sizes to reflect the page table format */
 928         arm_smmu_ops.pgsize_bitmap = pgtbl_cfg.pgsize_bitmap;
 929
 930         /* Initialise the context bank with our page table cfg */
 931         arm_smmu_init_context_bank(smmu_domain, &pgtbl_cfg);
 932
 933         /*
 934          * Request context fault interrupt. Do this last to avoid the
 935          * handler seeing a half-initialised domain state.
 936          */
 937         irq = smmu->irqs[smmu->num_global_irqs + cfg->irptndx];
 938         ret = request_irq(irq, arm_smmu_context_fault, IRQF_SHARED,
 939                           "arm-smmu-context-fault", domain);
 940         if (IS_ERR_VALUE(ret)) {
 941                 dev_err(smmu->dev, "failed to request context IRQ %d (%u)\n",
 942                         cfg->irptndx, irq);
 943                 cfg->irptndx = INVALID_IRPTNDX;
 944         }
 945
 946         mutex_unlock(&smmu_domain->init_mutex);
 947
 948         /* Publish page table ops for map/unmap */
 949         smmu_domain->pgtbl_ops = pgtbl_ops;
 950         return 0;
 951
 952 out_clear_smmu:
 953         smmu_domain->smmu = NULL;
 954 out_unlock:
 955         mutex_unlock(&smmu_domain->init_mutex);
 956         return ret;
 957 }
 958
 959 static void arm_smmu_destroy_domain_context(struct iommu_domain *domain)
 960 {
 961         struct arm_smmu_domain *smmu_domain = domain->priv;
 962         struct arm_smmu_device *smmu = smmu_domain->smmu;
 963         struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
 964         void __iomem *cb_base;
 965         int irq;
 966
 967         if (!smmu)
 968                 return;
 969
 970         /*
 971          * Disable the context bank and free the page tables before freeing
 972          * it.
 973          */
 974         cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
 975         writel_relaxed(0, cb_base + ARM_SMMU_CB_SCTLR);
 976
 977         if (cfg->irptndx != INVALID_IRPTNDX) {
 978                 irq = smmu->irqs[smmu->num_global_irqs + cfg->irptndx];
 979                 free_irq(irq, domain);
 980         }
 981
 982         if (smmu_domain->pgtbl_ops)
 983                 free_io_pgtable_ops(smmu_domain->pgtbl_ops);
 984
 985         __arm_smmu_free_bitmap(smmu->context_map, cfg->cbndx);
 986 }
 987
 988 static int arm_smmu_domain_init(struct iommu_domain *domain)
 989 {
 990         struct arm_smmu_domain *smmu_domain;
 991
 992         /*
 993          * Allocate the domain and initialise some of its data structures.
 994          * We can't really do anything meaningful until we've added a
 995          * master.
 996          */
 997         smmu_domain = kzalloc(sizeof(*smmu_domain), GFP_KERNEL);
 998         if (!smmu_domain)
 999                 return -ENOMEM;
1000
1001         mutex_init(&smmu_domain->init_mutex);
1002         spin_lock_init(&smmu_domain->pgtbl_lock);
1003         domain->priv = smmu_domain;
1004         return 0;
1005 }
1006
1007 static void arm_smmu_domain_destroy(struct iommu_domain *domain)
1008 {
1009         struct arm_smmu_domain *smmu_domain = domain->priv;
1010
1011         /*
1012          * Free the domain resources. We assume that all devices have
1013          * already been detached.
1014          */
1015         arm_smmu_destroy_domain_context(domain);
1016         kfree(smmu_domain);
1017 }
1018
1019 static int arm_smmu_master_configure_smrs(struct arm_smmu_device *smmu,
1020                                           struct arm_smmu_master_cfg *cfg)
1021 {
1022         int i;
1023         struct arm_smmu_smr *smrs;
1024         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1025
1026         if (!(smmu->features & ARM_SMMU_FEAT_STREAM_MATCH))
1027                 return 0;
1028
1029         if (cfg->smrs)
1030                 return -EEXIST;
1031
1032         smrs = kmalloc_array(cfg->num_streamids, sizeof(*smrs), GFP_KERNEL);
1033         if (!smrs) {
1034                 dev_err(smmu->dev, "failed to allocate %d SMRs\n",
1035                         cfg->num_streamids);
1036                 return -ENOMEM;
1037         }
1038
1039         /* Allocate the SMRs on the SMMU */
1040         for (i = 0; i < cfg->num_streamids; ++i) {
1041                 int idx = __arm_smmu_alloc_bitmap(smmu->smr_map, 0,
1042                                                   smmu->num_mapping_groups);
1043                 if (IS_ERR_VALUE(idx)) {
1044                         dev_err(smmu->dev, "failed to allocate free SMR\n");
1045                         goto err_free_smrs;
1046                 }
1047
1048                 smrs[i] = (struct arm_smmu_smr) {
1049                         .idx    = idx,
1050                         .mask   = 0, /* We don't currently share SMRs */
1051                         .id     = cfg->streamids[i],
1052                 };
1053         }
1054
1055         /* It worked! Now, poke the actual hardware */
1056         for (i = 0; i < cfg->num_streamids; ++i) {
1057                 u32 reg = SMR_VALID | smrs[i].id << SMR_ID_SHIFT |
1058                           smrs[i].mask << SMR_MASK_SHIFT;
1059                 writel_relaxed(reg, gr0_base + ARM_SMMU_GR0_SMR(smrs[i].idx));
1060         }
1061
1062         cfg->smrs = smrs;
1063         return 0;
1064
1065 err_free_smrs:
1066         while (--i >= 0)
1067                 __arm_smmu_free_bitmap(smmu->smr_map, smrs[i].idx);
1068         kfree(smrs);
1069         return -ENOSPC;
1070 }
1071
1072 static void arm_smmu_master_free_smrs(struct arm_smmu_device *smmu,
1073                                       struct arm_smmu_master_cfg *cfg)
1074 {
1075         int i;
1076         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1077         struct arm_smmu_smr *smrs = cfg->smrs;
1078
1079         if (!smrs)
1080                 return;
1081
1082         /* Invalidate the SMRs before freeing back to the allocator */
1083         for (i = 0; i < cfg->num_streamids; ++i) {
1084                 u8 idx = smrs[i].idx;
1085
1086                 writel_relaxed(~SMR_VALID, gr0_base + ARM_SMMU_GR0_SMR(idx));
1087                 __arm_smmu_free_bitmap(smmu->smr_map, idx);
1088         }
1089
1090         cfg->smrs = NULL;
1091         kfree(smrs);
1092 }
1093
1094 static int arm_smmu_domain_add_master(struct arm_smmu_domain *smmu_domain,
1095                                       struct arm_smmu_master_cfg *cfg)
1096 {
1097         int i, ret;
1098         struct arm_smmu_device *smmu = smmu_domain->smmu;
1099         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1100
1101         /* Devices in an IOMMU group may already be configured */
1102         ret = arm_smmu_master_configure_smrs(smmu, cfg);
1103         if (ret)
1104                 return ret == -EEXIST ? 0 : ret;
1105
1106         for (i = 0; i < cfg->num_streamids; ++i) {
1107                 u32 idx, s2cr;
1108
1109                 idx = cfg->smrs ? cfg->smrs[i].idx : cfg->streamids[i];
1110                 s2cr = S2CR_TYPE_TRANS |
1111                        (smmu_domain->cfg.cbndx << S2CR_CBNDX_SHIFT);
1112                 writel_relaxed(s2cr, gr0_base + ARM_SMMU_GR0_S2CR(idx));
1113         }
1114
1115         return 0;
1116 }
1117
1118 static void arm_smmu_domain_remove_master(struct arm_smmu_domain *smmu_domain,
1119                                           struct arm_smmu_master_cfg *cfg)
1120 {
1121         int i;
1122         struct arm_smmu_device *smmu = smmu_domain->smmu;
1123         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1124
1125         /* An IOMMU group is torn down by the first device to be removed */
1126         if ((smmu->features & ARM_SMMU_FEAT_STREAM_MATCH) && !cfg->smrs)
1127                 return;
1128
1129         /*
1130          * We *must* clear the S2CR first, because freeing the SMR means
1131          * that it can be re-allocated immediately.
1132          */
1133         for (i = 0; i < cfg->num_streamids; ++i) {
1134                 u32 idx = cfg->smrs ? cfg->smrs[i].idx : cfg->streamids[i];
1135
1136                 writel_relaxed(S2CR_TYPE_BYPASS,
1137                                gr0_base + ARM_SMMU_GR0_S2CR(idx));
1138         }
1139
1140         arm_smmu_master_free_smrs(smmu, cfg);
1141 }
1142
1143 static int arm_smmu_attach_dev(struct iommu_domain *domain, struct device *dev)
1144 {
1145         int ret;
1146         struct arm_smmu_domain *smmu_domain = domain->priv;
1147         struct arm_smmu_device *smmu;
1148         struct arm_smmu_master_cfg *cfg;
1149
1150         smmu = find_smmu_for_device(dev);
1151         if (!smmu) {
1152                 dev_err(dev, "cannot attach to SMMU, is it on the same bus?\n");
1153                 return -ENXIO;
1154         }
1155
1156         if (dev->archdata.iommu) {
1157                 dev_err(dev, "already attached to IOMMU domain\n");
1158                 return -EEXIST;
1159         }
1160
1161         /* Ensure that the domain is finalised */
1162         ret = arm_smmu_init_domain_context(domain, smmu);
1163         if (IS_ERR_VALUE(ret))
1164                 return ret;
1165
1166         /*
1167          * Sanity check the domain. We don't support domains across
1168          * different SMMUs.
1169          */
1170         if (smmu_domain->smmu != smmu) {
1171                 dev_err(dev,
1172                         "cannot attach to SMMU %s whilst already attached to domain on SMMU %s\n",
1173                         dev_name(smmu_domain->smmu->dev), dev_name(smmu->dev));
1174                 return -EINVAL;
1175         }
1176
1177         /* Looks ok, so add the device to the domain */
1178         cfg = find_smmu_master_cfg(dev);
1179         if (!cfg)
1180                 return -ENODEV;
1181
1182         ret = arm_smmu_domain_add_master(smmu_domain, cfg);
1183         if (!ret)
1184                 dev->archdata.iommu = domain;
1185         return ret;
1186 }
1187
1188 static void arm_smmu_detach_dev(struct iommu_domain *domain, struct device *dev)
1189 {
1190         struct arm_smmu_domain *smmu_domain = domain->priv;
1191         struct arm_smmu_master_cfg *cfg;
1192
1193         cfg = find_smmu_master_cfg(dev);
1194         if (!cfg)
1195                 return;
1196
1197         dev->archdata.iommu = NULL;
1198         arm_smmu_domain_remove_master(smmu_domain, cfg);
1199 }
1200
1201 static int arm_smmu_map(struct iommu_domain *domain, unsigned long iova,
1202                         phys_addr_t paddr, size_t size, int prot)
1203 {
1204         int ret;
1205         unsigned long flags;
1206         struct arm_smmu_domain *smmu_domain = domain->priv;
1207         struct io_pgtable_ops *ops= smmu_domain->pgtbl_ops;
1208
1209         if (!ops)
1210                 return -ENODEV;
1211
1212         spin_lock_irqsave(&smmu_domain->pgtbl_lock, flags);
1213         ret = ops->map(ops, iova, paddr, size, prot);
1214         spin_unlock_irqrestore(&smmu_domain->pgtbl_lock, flags);
1215         return ret;
1216 }
1217
1218 static size_t arm_smmu_unmap(struct iommu_domain *domain, unsigned long iova,
1219                              size_t size)
1220 {
1221         size_t ret;
1222         unsigned long flags;
1223         struct arm_smmu_domain *smmu_domain = domain->priv;
1224         struct io_pgtable_ops *ops= smmu_domain->pgtbl_ops;
1225
1226         if (!ops)
1227                 return 0;
1228
1229         spin_lock_irqsave(&smmu_domain->pgtbl_lock, flags);
1230         ret = ops->unmap(ops, iova, size);
1231         spin_unlock_irqrestore(&smmu_domain->pgtbl_lock, flags);
1232         return ret;
1233 }
1234
1235 static phys_addr_t arm_smmu_iova_to_phys_hard(struct iommu_domain *domain,
1236                                               dma_addr_t iova)
1237 {
1238         struct arm_smmu_domain *smmu_domain = domain->priv;
1239         struct arm_smmu_device *smmu = smmu_domain->smmu;
1240         struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
1241         struct io_pgtable_ops *ops= smmu_domain->pgtbl_ops;
1242         struct device *dev = smmu->dev;
1243         void __iomem *cb_base;
1244         u32 tmp;
1245         u64 phys;
1246
1247         cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
1248
1249         if (smmu->version == 1) {
1250                 u32 reg = iova & ~0xfff;
1251                 writel_relaxed(reg, cb_base + ARM_SMMU_CB_ATS1PR_LO);
1252         } else {
1253                 u32 reg = iova & ~0xfff;
1254                 writel_relaxed(reg, cb_base + ARM_SMMU_CB_ATS1PR_LO);
1255                 reg = ((u64)iova & ~0xfff) >> 32;
1256                 writel_relaxed(reg, cb_base + ARM_SMMU_CB_ATS1PR_HI);
1257         }
1258
1259         if (readl_poll_timeout_atomic(cb_base + ARM_SMMU_CB_ATSR, tmp,
1260                                       !(tmp & ATSR_ACTIVE), 5, 50)) {
1261                 dev_err(dev,
1262                         "iova to phys timed out on 0x%pad. Falling back to software table walk.\n",
1263                         &iova);
1264                 return ops->iova_to_phys(ops, iova);
1265         }
1266
1267         phys = readl_relaxed(cb_base + ARM_SMMU_CB_PAR_LO);
1268         phys |= ((u64)readl_relaxed(cb_base + ARM_SMMU_CB_PAR_HI)) << 32;
1269
1270         if (phys & CB_PAR_F) {
1271                 dev_err(dev, "translation fault!\n");
1272                 dev_err(dev, "PAR = 0x%llx\n", phys);
1273                 return 0;
1274         }
1275
1276         return (phys & GENMASK_ULL(39, 12)) | (iova & 0xfff);
1277 }
1278
1279 static phys_addr_t arm_smmu_iova_to_phys(struct iommu_domain *domain,
1280                                         dma_addr_t iova)
1281 {
1282         phys_addr_t ret;
1283         unsigned long flags;
1284         struct arm_smmu_domain *smmu_domain = domain->priv;
1285         struct io_pgtable_ops *ops= smmu_domain->pgtbl_ops;
1286
1287         if (!ops)
1288                 return 0;
1289
1290         spin_lock_irqsave(&smmu_domain->pgtbl_lock, flags);
1291         if (smmu_domain->smmu->features & ARM_SMMU_FEAT_TRANS_OPS &&
1292                         smmu_domain->stage == ARM_SMMU_DOMAIN_S1) {
1293                 ret = arm_smmu_iova_to_phys_hard(domain, iova);
1294         } else {
1295                 ret = ops->iova_to_phys(ops, iova);
1296         }
1297
1298         spin_unlock_irqrestore(&smmu_domain->pgtbl_lock, flags);
1299
1300         return ret;
1301 }
1302
1303 static bool arm_smmu_capable(enum iommu_cap cap)
1304 {
1305         switch (cap) {
1306         case IOMMU_CAP_CACHE_COHERENCY:
1307                 /*
1308                  * Return true here as the SMMU can always send out coherent
1309                  * requests.
1310                  */
1311                 return true;
1312         case IOMMU_CAP_INTR_REMAP:
1313                 return true; /* MSIs are just memory writes */
1314         case IOMMU_CAP_NOEXEC:
1315                 return true;
1316         default:
1317                 return false;
1318         }
1319 }
1320
1321 static int __arm_smmu_get_pci_sid(struct pci_dev *pdev, u16 alias, void *data)
1322 {
1323         *((u16 *)data) = alias;
1324         return 0; /* Continue walking */
1325 }
1326
1327 static void __arm_smmu_release_pci_iommudata(void *data)
1328 {
1329         kfree(data);
1330 }
1331
1332 static int arm_smmu_add_device(struct device *dev)
1333 {
1334         struct arm_smmu_device *smmu;
1335         struct arm_smmu_master_cfg *cfg;
1336         struct iommu_group *group;
1337         void (*releasefn)(void *) = NULL;
1338         int ret;
1339
1340         smmu = find_smmu_for_device(dev);
1341         if (!smmu)
1342                 return -ENODEV;
1343
1344         group = iommu_group_alloc();
1345         if (IS_ERR(group)) {
1346                 dev_err(dev, "Failed to allocate IOMMU group\n");
1347                 return PTR_ERR(group);
1348         }
1349
1350         if (dev_is_pci(dev)) {
1351                 struct pci_dev *pdev = to_pci_dev(dev);
1352
1353                 cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1354                 if (!cfg) {
1355                         ret = -ENOMEM;
1356                         goto out_put_group;
1357                 }
1358
1359                 cfg->num_streamids = 1;
1360                 /*
1361                  * Assume Stream ID == Requester ID for now.
1362                  * We need a way to describe the ID mappings in FDT.
1363                  */
1364                 pci_for_each_dma_alias(pdev, __arm_smmu_get_pci_sid,
1365                                        &cfg->streamids[0]);
1366                 releasefn = __arm_smmu_release_pci_iommudata;
1367         } else {
1368                 struct arm_smmu_master *master;
1369
1370                 master = find_smmu_master(smmu, dev->of_node);
1371                 if (!master) {
1372                         ret = -ENODEV;
1373                         goto out_put_group;
1374                 }
1375
1376                 cfg = &master->cfg;
1377         }
1378
1379         iommu_group_set_iommudata(group, cfg, releasefn);
1380         ret = iommu_group_add_device(group, dev);
1381
1382 out_put_group:
1383         iommu_group_put(group);
1384         return ret;
1385 }
1386
1387 static void arm_smmu_remove_device(struct device *dev)
1388 {
1389         iommu_group_remove_device(dev);
1390 }
1391
1392 static int arm_smmu_domain_get_attr(struct iommu_domain *domain,
1393                                     enum iommu_attr attr, void *data)
1394 {
1395         struct arm_smmu_domain *smmu_domain = domain->priv;
1396
1397         switch (attr) {
1398         case DOMAIN_ATTR_NESTING:
1399                 *(int *)data = (smmu_domain->stage == ARM_SMMU_DOMAIN_NESTED);
1400                 return 0;
1401         default:
1402                 return -ENODEV;
1403         }
1404 }
1405
1406 static int arm_smmu_domain_set_attr(struct iommu_domain *domain,
1407                                     enum iommu_attr attr, void *data)
1408 {
1409         int ret = 0;
1410         struct arm_smmu_domain *smmu_domain = domain->priv;
1411
1412         mutex_lock(&smmu_domain->init_mutex);
1413
1414         switch (attr) {
1415         case DOMAIN_ATTR_NESTING:
1416                 if (smmu_domain->smmu) {
1417                         ret = -EPERM;
1418                         goto out_unlock;
1419                 }
1420
1421                 if (*(int *)data)
1422                         smmu_domain->stage = ARM_SMMU_DOMAIN_NESTED;
1423                 else
1424                         smmu_domain->stage = ARM_SMMU_DOMAIN_S1;
1425
1426                 break;
1427         default:
1428                 ret = -ENODEV;
1429         }
1430
1431 out_unlock:
1432         mutex_unlock(&smmu_domain->init_mutex);
1433         return ret;
1434 }
1435
1436 static struct iommu_ops arm_smmu_ops = {
1437         .capable                = arm_smmu_capable,
1438         .domain_init            = arm_smmu_domain_init,
1439         .domain_destroy         = arm_smmu_domain_destroy,
1440         .attach_dev             = arm_smmu_attach_dev,
1441         .detach_dev             = arm_smmu_detach_dev,
1442         .map                    = arm_smmu_map,
1443         .unmap                  = arm_smmu_unmap,
1444         .map_sg                 = default_iommu_map_sg,
1445         .iova_to_phys           = arm_smmu_iova_to_phys,
1446         .add_device             = arm_smmu_add_device,
1447         .remove_device          = arm_smmu_remove_device,
1448         .domain_get_attr        = arm_smmu_domain_get_attr,
1449         .domain_set_attr        = arm_smmu_domain_set_attr,
1450         .pgsize_bitmap          = -1UL, /* Restricted during device attach */
1451 };
1452
1453 static void arm_smmu_device_reset(struct arm_smmu_device *smmu)
1454 {
1455         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1456         void __iomem *cb_base;
1457         int i = 0;
1458         u32 reg;
1459
1460         /* clear global FSR */
1461         reg = readl_relaxed(ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sGFSR);
1462         writel(reg, ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sGFSR);
1463
1464         /* Mark all SMRn as invalid and all S2CRn as bypass */
1465         for (i = 0; i < smmu->num_mapping_groups; ++i) {
1466                 writel_relaxed(0, gr0_base + ARM_SMMU_GR0_SMR(i));
1467                 writel_relaxed(S2CR_TYPE_BYPASS,
1468                         gr0_base + ARM_SMMU_GR0_S2CR(i));
1469         }
1470
1471         /* Make sure all context banks are disabled and clear CB_FSR  */
1472         for (i = 0; i < smmu->num_context_banks; ++i) {
1473                 cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, i);
1474                 writel_relaxed(0, cb_base + ARM_SMMU_CB_SCTLR);
1475                 writel_relaxed(FSR_FAULT, cb_base + ARM_SMMU_CB_FSR);
1476         }
1477
1478         /* Invalidate the TLB, just in case */
1479         writel_relaxed(0, gr0_base + ARM_SMMU_GR0_TLBIALLH);
1480         writel_relaxed(0, gr0_base + ARM_SMMU_GR0_TLBIALLNSNH);
1481
1482         reg = readl_relaxed(ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sCR0);
1483
1484         /* Enable fault reporting */
1485         reg |= (sCR0_GFRE | sCR0_GFIE | sCR0_GCFGFRE | sCR0_GCFGFIE);
1486
1487         /* Disable TLB broadcasting. */
1488         reg |= (sCR0_VMIDPNE | sCR0_PTM);
1489
1490         /* Enable client access, but bypass when no mapping is found */
1491         reg &= ~(sCR0_CLIENTPD | sCR0_USFCFG);
1492
1493         /* Disable forced broadcasting */
1494         reg &= ~sCR0_FB;
1495
1496         /* Don't upgrade barriers */
1497         reg &= ~(sCR0_BSU_MASK << sCR0_BSU_SHIFT);
1498
1499         /* Push the button */
1500         __arm_smmu_tlb_sync(smmu);
1501         writel(reg, ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sCR0);
1502 }
1503
1504 static int arm_smmu_id_size_to_bits(int size)
1505 {
1506         switch (size) {
1507         case 0:
1508                 return 32;
1509         case 1:
1510                 return 36;
1511         case 2:
1512                 return 40;
1513         case 3:
1514                 return 42;
1515         case 4:
1516                 return 44;
1517         case 5:
1518         default:
1519                 return 48;
1520         }
1521 }
1522
1523 static int arm_smmu_device_cfg_probe(struct arm_smmu_device *smmu)
1524 {
1525         unsigned long size;
1526         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1527         u32 id;
1528
1529         dev_notice(smmu->dev, "probing hardware configuration...\n");
1530         dev_notice(smmu->dev, "SMMUv%d with:\n", smmu->version);
1531
1532         /* ID0 */
1533         id = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID0);
1534
1535         /* Restrict available stages based on module parameter */
1536         if (force_stage == 1)
1537                 id &= ~(ID0_S2TS | ID0_NTS);
1538         else if (force_stage == 2)
1539                 id &= ~(ID0_S1TS | ID0_NTS);
1540
1541         if (id & ID0_S1TS) {
1542                 smmu->features |= ARM_SMMU_FEAT_TRANS_S1;
1543                 dev_notice(smmu->dev, "\tstage 1 translation\n");
1544         }
1545
1546         if (id & ID0_S2TS) {
1547                 smmu->features |= ARM_SMMU_FEAT_TRANS_S2;
1548                 dev_notice(smmu->dev, "\tstage 2 translation\n");
1549         }
1550
1551         if (id & ID0_NTS) {
1552                 smmu->features |= ARM_SMMU_FEAT_TRANS_NESTED;
1553                 dev_notice(smmu->dev, "\tnested translation\n");
1554         }
1555
1556         if (!(smmu->features &
1557                 (ARM_SMMU_FEAT_TRANS_S1 | ARM_SMMU_FEAT_TRANS_S2))) {
1558                 dev_err(smmu->dev, "\tno translation support!\n");
1559                 return -ENODEV;
1560         }
1561
1562         if ((id & ID0_S1TS) && ((smmu->version == 1) || (id & ID0_ATOSNS))) {
1563                 smmu->features |= ARM_SMMU_FEAT_TRANS_OPS;
1564                 dev_notice(smmu->dev, "\taddress translation ops\n");
1565         }
1566
1567         if (id & ID0_CTTW) {
1568                 smmu->features |= ARM_SMMU_FEAT_COHERENT_WALK;
1569                 dev_notice(smmu->dev, "\tcoherent table walk\n");
1570         }
1571
1572         if (id & ID0_SMS) {
1573                 u32 smr, sid, mask;
1574
1575                 smmu->features |= ARM_SMMU_FEAT_STREAM_MATCH;
1576                 smmu->num_mapping_groups = (id >> ID0_NUMSMRG_SHIFT) &
1577                                            ID0_NUMSMRG_MASK;
1578                 if (smmu->num_mapping_groups == 0) {
1579                         dev_err(smmu->dev,
1580                                 "stream-matching supported, but no SMRs present!\n");
1581                         return -ENODEV;
1582                 }
1583
1584                 smr = SMR_MASK_MASK << SMR_MASK_SHIFT;
1585                 smr |= (SMR_ID_MASK << SMR_ID_SHIFT);
1586                 writel_relaxed(smr, gr0_base + ARM_SMMU_GR0_SMR(0));
1587                 smr = readl_relaxed(gr0_base + ARM_SMMU_GR0_SMR(0));
1588
1589                 mask = (smr >> SMR_MASK_SHIFT) & SMR_MASK_MASK;
1590                 sid = (smr >> SMR_ID_SHIFT) & SMR_ID_MASK;
1591                 if ((mask & sid) != sid) {
1592                         dev_err(smmu->dev,
1593                                 "SMR mask bits (0x%x) insufficient for ID field (0x%x)\n",
1594                                 mask, sid);
1595                         return -ENODEV;
1596                 }
1597
1598                 dev_notice(smmu->dev,
1599                            "\tstream matching with %u register groups, mask 0x%x",
1600                            smmu->num_mapping_groups, mask);
1601         } else {
1602                 smmu->num_mapping_groups = (id >> ID0_NUMSIDB_SHIFT) &
1603                                            ID0_NUMSIDB_MASK;
1604         }
1605
1606         /* ID1 */
1607         id = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID1);
1608         smmu->pgshift = (id & ID1_PAGESIZE) ? 16 : 12;
1609
1610         /* Check for size mismatch of SMMU address space from mapped region */
1611         size = 1 << (((id >> ID1_NUMPAGENDXB_SHIFT) & ID1_NUMPAGENDXB_MASK) + 1);
1612         size *= 2 << smmu->pgshift;
1613         if (smmu->size != size)
1614                 dev_warn(smmu->dev,
1615                         "SMMU address space size (0x%lx) differs from mapped region size (0x%lx)!\n",
1616                         size, smmu->size);
1617
1618         smmu->num_s2_context_banks = (id >> ID1_NUMS2CB_SHIFT) & ID1_NUMS2CB_MASK;
1619         smmu->num_context_banks = (id >> ID1_NUMCB_SHIFT) & ID1_NUMCB_MASK;
1620         if (smmu->num_s2_context_banks > smmu->num_context_banks) {
1621                 dev_err(smmu->dev, "impossible number of S2 context banks!\n");
1622                 return -ENODEV;
1623         }
1624         dev_notice(smmu->dev, "\t%u context banks (%u stage-2 only)\n",
1625                    smmu->num_context_banks, smmu->num_s2_context_banks);
1626
1627         /* ID2 */
1628         id = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID2);
1629         size = arm_smmu_id_size_to_bits((id >> ID2_IAS_SHIFT) & ID2_IAS_MASK);
1630         smmu->ipa_size = size;
1631
1632         /* The output mask is also applied for bypass */
1633         size = arm_smmu_id_size_to_bits((id >> ID2_OAS_SHIFT) & ID2_OAS_MASK);
1634         smmu->pa_size = size;
1635
1636         if (smmu->version == ARM_SMMU_V1) {
1637                 smmu->va_size = smmu->ipa_size;
1638                 size = SZ_4K | SZ_2M | SZ_1G;
1639         } else {
1640                 size = (id >> ID2_UBS_SHIFT) & ID2_UBS_MASK;
1641                 smmu->va_size = arm_smmu_id_size_to_bits(size);
1642 #ifndef CONFIG_64BIT
1643                 smmu->va_size = min(32UL, smmu->va_size);
1644 #endif
1645                 size = 0;
1646                 if (id & ID2_PTFS_4K)
1647                         size |= SZ_4K | SZ_2M | SZ_1G;
1648                 if (id & ID2_PTFS_16K)
1649                         size |= SZ_16K | SZ_32M;
1650                 if (id & ID2_PTFS_64K)
1651                         size |= SZ_64K | SZ_512M;
1652         }
1653
1654         arm_smmu_ops.pgsize_bitmap &= size;
1655         dev_notice(smmu->dev, "\tSupported page sizes: 0x%08lx\n", size);
1656
1657         if (smmu->features & ARM_SMMU_FEAT_TRANS_S1)
1658                 dev_notice(smmu->dev, "\tStage-1: %lu-bit VA -> %lu-bit IPA\n",
1659                            smmu->va_size, smmu->ipa_size);
1660
1661         if (smmu->features & ARM_SMMU_FEAT_TRANS_S2)
1662                 dev_notice(smmu->dev, "\tStage-2: %lu-bit IPA -> %lu-bit PA\n",
1663                            smmu->ipa_size, smmu->pa_size);
1664
1665         return 0;
1666 }
1667
1668 static const struct of_device_id arm_smmu_of_match[] = {
1669         { .compatible = "arm,smmu-v1", .data = (void *)ARM_SMMU_V1 },
1670         { .compatible = "arm,smmu-v2", .data = (void *)ARM_SMMU_V2 },
1671         { .compatible = "arm,mmu-400", .data = (void *)ARM_SMMU_V1 },
1672         { .compatible = "arm,mmu-401", .data = (void *)ARM_SMMU_V1 },
1673         { .compatible = "arm,mmu-500", .data = (void *)ARM_SMMU_V2 },
1674         { },
1675 };
1676 MODULE_DEVICE_TABLE(of, arm_smmu_of_match);
1677
1678 static int arm_smmu_device_dt_probe(struct platform_device *pdev)
1679 {
1680         const struct of_device_id *of_id;
1681         struct resource *res;
1682         struct arm_smmu_device *smmu;
1683         struct device *dev = &pdev->dev;
1684         struct rb_node *node;
1685         struct of_phandle_args masterspec;
1686         int num_irqs, i, err;
1687
1688         smmu = devm_kzalloc(dev, sizeof(*smmu), GFP_KERNEL);
1689         if (!smmu) {
1690                 dev_err(dev, "failed to allocate arm_smmu_device\n");
1691                 return -ENOMEM;
1692         }
1693         smmu->dev = dev;
1694
1695         of_id = of_match_node(arm_smmu_of_match, dev->of_node);
1696         smmu->version = (enum arm_smmu_arch_version)of_id->data;
1697
1698         res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1699         smmu->base = devm_ioremap_resource(dev, res);
1700         if (IS_ERR(smmu->base))
1701                 return PTR_ERR(smmu->base);
1702         smmu->size = resource_size(res);
1703
1704         if (of_property_read_u32(dev->of_node, "#global-interrupts",
1705                                  &smmu->num_global_irqs)) {
1706                 dev_err(dev, "missing #global-interrupts property\n");
1707                 return -ENODEV;
1708         }
1709
1710         num_irqs = 0;
1711         while ((res = platform_get_resource(pdev, IORESOURCE_IRQ, num_irqs))) {
1712                 num_irqs++;
1713                 if (num_irqs > smmu->num_global_irqs)
1714                         smmu->num_context_irqs++;
1715         }
1716
1717         if (!smmu->num_context_irqs) {
1718                 dev_err(dev, "found %d interrupts but expected at least %d\n",
1719                         num_irqs, smmu->num_global_irqs + 1);
1720                 return -ENODEV;
1721         }
1722
1723         smmu->irqs = devm_kzalloc(dev, sizeof(*smmu->irqs) * num_irqs,
1724                                   GFP_KERNEL);
1725         if (!smmu->irqs) {
1726                 dev_err(dev, "failed to allocate %d irqs\n", num_irqs);
1727                 return -ENOMEM;
1728         }
1729
1730         for (i = 0; i < num_irqs; ++i) {
1731                 int irq = platform_get_irq(pdev, i);
1732
1733                 if (irq < 0) {
1734                         dev_err(dev, "failed to get irq index %d\n", i);
1735                         return -ENODEV;
1736                 }
1737                 smmu->irqs[i] = irq;
1738         }
1739
1740         err = arm_smmu_device_cfg_probe(smmu);
1741         if (err)
1742                 return err;
1743
1744         i = 0;
1745         smmu->masters = RB_ROOT;
1746         while (!of_parse_phandle_with_args(dev->of_node, "mmu-masters",
1747                                            "#stream-id-cells", i,
1748                                            &masterspec)) {
1749                 err = register_smmu_master(smmu, dev, &masterspec);
1750                 if (err) {
1751                         dev_err(dev, "failed to add master %s\n",
1752                                 masterspec.np->name);
1753                         goto out_put_masters;
1754                 }
1755
1756                 i++;
1757         }
1758         dev_notice(dev, "registered %d master devices\n", i);
1759
1760         parse_driver_options(smmu);
1761
1762         if (smmu->version > ARM_SMMU_V1 &&
1763             smmu->num_context_banks != smmu->num_context_irqs) {
1764                 dev_err(dev,
1765                         "found only %d context interrupt(s) but %d required\n",
1766                         smmu->num_context_irqs, smmu->num_context_banks);
1767                 err = -ENODEV;
1768                 goto out_put_masters;
1769         }
1770
1771         for (i = 0; i < smmu->num_global_irqs; ++i) {
1772                 err = request_irq(smmu->irqs[i],
1773                                   arm_smmu_global_fault,
1774                                   IRQF_SHARED,
1775                                   "arm-smmu global fault",
1776                                   smmu);
1777                 if (err) {
1778                         dev_err(dev, "failed to request global IRQ %d (%u)\n",
1779                                 i, smmu->irqs[i]);
1780                         goto out_free_irqs;
1781                 }
1782         }
1783
1784         INIT_LIST_HEAD(&smmu->list);
1785         spin_lock(&arm_smmu_devices_lock);
1786         list_add(&smmu->list, &arm_smmu_devices);
1787         spin_unlock(&arm_smmu_devices_lock);
1788
1789         arm_smmu_device_reset(smmu);
1790         return 0;
1791
1792 out_free_irqs:
1793         while (i--)
1794                 free_irq(smmu->irqs[i], smmu);
1795
1796 out_put_masters:
1797         for (node = rb_first(&smmu->masters); node; node = rb_next(node)) {
1798                 struct arm_smmu_master *master
1799                         = container_of(node, struct arm_smmu_master, node);
1800                 of_node_put(master->of_node);
1801         }
1802
1803         return err;
1804 }
1805
1806 static int arm_smmu_device_remove(struct platform_device *pdev)
1807 {
1808         int i;
1809         struct device *dev = &pdev->dev;
1810         struct arm_smmu_device *curr, *smmu = NULL;
1811         struct rb_node *node;
1812
1813         spin_lock(&arm_smmu_devices_lock);
1814         list_for_each_entry(curr, &arm_smmu_devices, list) {
1815                 if (curr->dev == dev) {
1816                         smmu = curr;
1817                         list_del(&smmu->list);
1818                         break;
1819                 }
1820         }
1821         spin_unlock(&arm_smmu_devices_lock);
1822
1823         if (!smmu)
1824                 return -ENODEV;
1825
1826         for (node = rb_first(&smmu->masters); node; node = rb_next(node)) {
1827                 struct arm_smmu_master *master
1828                         = container_of(node, struct arm_smmu_master, node);
1829                 of_node_put(master->of_node);
1830         }
1831
1832         if (!bitmap_empty(smmu->context_map, ARM_SMMU_MAX_CBS))
1833                 dev_err(dev, "removing device with active domains!\n");
1834
1835         for (i = 0; i < smmu->num_global_irqs; ++i)
1836                 free_irq(smmu->irqs[i], smmu);
1837
1838         /* Turn the thing off */
1839         writel(sCR0_CLIENTPD, ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sCR0);
1840         return 0;
1841 }
1842
1843 static struct platform_driver arm_smmu_driver = {
1844         .driver = {
1845                 .name           = "arm-smmu",
1846                 .of_match_table = of_match_ptr(arm_smmu_of_match),
1847         },
1848         .probe  = arm_smmu_device_dt_probe,
1849         .remove = arm_smmu_device_remove,
1850 };
1851
1852 static int __init arm_smmu_init(void)
1853 {
1854         struct device_node *np;
1855         int ret;
1856
1857         /*
1858          * Play nice with systems that don't have an ARM SMMU by checking that
1859          * an ARM SMMU exists in the system before proceeding with the driver
1860          * and IOMMU bus operation registration.
1861          */
1862         np = of_find_matching_node(NULL, arm_smmu_of_match);
1863         if (!np)
1864                 return 0;
1865
1866         of_node_put(np);
1867
1868         ret = platform_driver_register(&arm_smmu_driver);
1869         if (ret)
1870                 return ret;
1871
1872         /* Oh, for a proper bus abstraction */
1873         if (!iommu_present(&platform_bus_type))
1874                 bus_set_iommu(&platform_bus_type, &arm_smmu_ops);
1875
1876 #ifdef CONFIG_ARM_AMBA
1877         if (!iommu_present(&amba_bustype))
1878                 bus_set_iommu(&amba_bustype, &arm_smmu_ops);
1879 #endif
1880
1881 #ifdef CONFIG_PCI
1882         if (!iommu_present(&pci_bus_type))
1883                 bus_set_iommu(&pci_bus_type, &arm_smmu_ops);
1884 #endif
1885
1886         return 0;
1887 }
1888
1889 static void __exit arm_smmu_exit(void)
1890 {
1891         return platform_driver_unregister(&arm_smmu_driver);
1892 }
1893
1894 subsys_initcall(arm_smmu_init);
1895 module_exit(arm_smmu_exit);
1896
1897 MODULE_DESCRIPTION("IOMMU API for ARM architected SMMU implementations");
1898 MODULE_AUTHOR("Will Deacon <will.deacon@arm.com>");
1899 MODULE_LICENSE("GPL v2");