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-iommu.h>
  33 #include <linux/dma-mapping.h>
  34 #include <linux/err.h>
  35 #include <linux/interrupt.h>
  36 #include <linux/io.h>
  37 #include <linux/iommu.h>
  38 #include <linux/iopoll.h>
  39 #include <linux/module.h>
  40 #include <linux/of.h>
  41 #include <linux/of_address.h>
  42 #include <linux/pci.h>
  43 #include <linux/platform_device.h>
  44 #include <linux/slab.h>
  45 #include <linux/spinlock.h>
  46
  47 #include <linux/amba/bus.h>
  48
  49 #include "io-pgtable.h"
  50
  51 /* Maximum number of stream IDs assigned to a single device */
  52 #define MAX_MASTER_STREAMIDS            MAX_PHANDLE_ARGS
  53
  54 /* Maximum number of context banks per SMMU */
  55 #define ARM_SMMU_MAX_CBS                128
  56
  57 /* Maximum number of mapping groups per SMMU */
  58 #define ARM_SMMU_MAX_SMRS               128
  59
  60 /* SMMU global address space */
  61 #define ARM_SMMU_GR0(smmu)              ((smmu)->base)
  62 #define ARM_SMMU_GR1(smmu)              ((smmu)->base + (1 << (smmu)->pgshift))
  63
  64 /*
  65  * SMMU global address space with conditional offset to access secure
  66  * aliases of non-secure registers (e.g. nsCR0: 0x400, nsGFSR: 0x448,
  67  * nsGFSYNR0: 0x450)
  68  */
  69 #define ARM_SMMU_GR0_NS(smmu)                                           \
  70         ((smmu)->base +                                                 \
  71                 ((smmu->options & ARM_SMMU_OPT_SECURE_CFG_ACCESS)       \
  72                         ? 0x400 : 0))
  73
  74 #ifdef CONFIG_64BIT
  75 #define smmu_writeq     writeq_relaxed
  76 #else
  77 #define smmu_writeq(reg64, addr)                                \
  78         do {                                                    \
  79                 u64 __val = (reg64);                            \
  80                 void __iomem *__addr = (addr);                  \
  81                 writel_relaxed(__val >> 32, __addr + 4);        \
  82                 writel_relaxed(__val, __addr);                  \
  83         } while (0)
  84 #endif
  85
  86 /* Configuration registers */
  87 #define ARM_SMMU_GR0_sCR0               0x0
  88 #define sCR0_CLIENTPD                   (1 << 0)
  89 #define sCR0_GFRE                       (1 << 1)
  90 #define sCR0_GFIE                       (1 << 2)
  91 #define sCR0_GCFGFRE                    (1 << 4)
  92 #define sCR0_GCFGFIE                    (1 << 5)
  93 #define sCR0_USFCFG                     (1 << 10)
  94 #define sCR0_VMIDPNE                    (1 << 11)
  95 #define sCR0_PTM                        (1 << 12)
  96 #define sCR0_FB                         (1 << 13)
  97 #define sCR0_BSU_SHIFT                  14
  98 #define sCR0_BSU_MASK                   0x3
  99
 100 /* Identification registers */
 101 #define ARM_SMMU_GR0_ID0                0x20
 102 #define ARM_SMMU_GR0_ID1                0x24
 103 #define ARM_SMMU_GR0_ID2                0x28
 104 #define ARM_SMMU_GR0_ID3                0x2c
 105 #define ARM_SMMU_GR0_ID4                0x30
 106 #define ARM_SMMU_GR0_ID5                0x34
 107 #define ARM_SMMU_GR0_ID6                0x38
 108 #define ARM_SMMU_GR0_ID7                0x3c
 109 #define ARM_SMMU_GR0_sGFSR              0x48
 110 #define ARM_SMMU_GR0_sGFSYNR0           0x50
 111 #define ARM_SMMU_GR0_sGFSYNR1           0x54
 112 #define ARM_SMMU_GR0_sGFSYNR2           0x58
 113
 114 #define ID0_S1TS                        (1 << 30)
 115 #define ID0_S2TS                        (1 << 29)
 116 #define ID0_NTS                         (1 << 28)
 117 #define ID0_SMS                         (1 << 27)
 118 #define ID0_ATOSNS                      (1 << 26)
 119 #define ID0_CTTW                        (1 << 14)
 120 #define ID0_NUMIRPT_SHIFT               16
 121 #define ID0_NUMIRPT_MASK                0xff
 122 #define ID0_NUMSIDB_SHIFT               9
 123 #define ID0_NUMSIDB_MASK                0xf
 124 #define ID0_NUMSMRG_SHIFT               0
 125 #define ID0_NUMSMRG_MASK                0xff
 126
 127 #define ID1_PAGESIZE                    (1 << 31)
 128 #define ID1_NUMPAGENDXB_SHIFT           28
 129 #define ID1_NUMPAGENDXB_MASK            7
 130 #define ID1_NUMS2CB_SHIFT               16
 131 #define ID1_NUMS2CB_MASK                0xff
 132 #define ID1_NUMCB_SHIFT                 0
 133 #define ID1_NUMCB_MASK                  0xff
 134
 135 #define ID2_OAS_SHIFT                   4
 136 #define ID2_OAS_MASK                    0xf
 137 #define ID2_IAS_SHIFT                   0
 138 #define ID2_IAS_MASK                    0xf
 139 #define ID2_UBS_SHIFT                   8
 140 #define ID2_UBS_MASK                    0xf
 141 #define ID2_PTFS_4K                     (1 << 12)
 142 #define ID2_PTFS_16K                    (1 << 13)
 143 #define ID2_PTFS_64K                    (1 << 14)
 144
 145 /* Global TLB invalidation */
 146 #define ARM_SMMU_GR0_TLBIVMID           0x64
 147 #define ARM_SMMU_GR0_TLBIALLNSNH        0x68
 148 #define ARM_SMMU_GR0_TLBIALLH           0x6c
 149 #define ARM_SMMU_GR0_sTLBGSYNC          0x70
 150 #define ARM_SMMU_GR0_sTLBGSTATUS        0x74
 151 #define sTLBGSTATUS_GSACTIVE            (1 << 0)
 152 #define TLB_LOOP_TIMEOUT                1000000 /* 1s! */
 153
 154 /* Stream mapping registers */
 155 #define ARM_SMMU_GR0_SMR(n)             (0x800 + ((n) << 2))
 156 #define SMR_VALID                       (1 << 31)
 157 #define SMR_MASK_SHIFT                  16
 158 #define SMR_MASK_MASK                   0x7fff
 159 #define SMR_ID_SHIFT                    0
 160 #define SMR_ID_MASK                     0x7fff
 161
 162 #define ARM_SMMU_GR0_S2CR(n)            (0xc00 + ((n) << 2))
 163 #define S2CR_CBNDX_SHIFT                0
 164 #define S2CR_CBNDX_MASK                 0xff
 165 #define S2CR_TYPE_SHIFT                 16
 166 #define S2CR_TYPE_MASK                  0x3
 167 #define S2CR_TYPE_TRANS                 (0 << S2CR_TYPE_SHIFT)
 168 #define S2CR_TYPE_BYPASS                (1 << S2CR_TYPE_SHIFT)
 169 #define S2CR_TYPE_FAULT                 (2 << S2CR_TYPE_SHIFT)
 170
 171 #define S2CR_PRIVCFG_SHIFT              24
 172 #define S2CR_PRIVCFG_UNPRIV             (2 << S2CR_PRIVCFG_SHIFT)
 173
 174 /* Context bank attribute registers */
 175 #define ARM_SMMU_GR1_CBAR(n)            (0x0 + ((n) << 2))
 176 #define CBAR_VMID_SHIFT                 0
 177 #define CBAR_VMID_MASK                  0xff
 178 #define CBAR_S1_BPSHCFG_SHIFT           8
 179 #define CBAR_S1_BPSHCFG_MASK            3
 180 #define CBAR_S1_BPSHCFG_NSH             3
 181 #define CBAR_S1_MEMATTR_SHIFT           12
 182 #define CBAR_S1_MEMATTR_MASK            0xf
 183 #define CBAR_S1_MEMATTR_WB              0xf
 184 #define CBAR_TYPE_SHIFT                 16
 185 #define CBAR_TYPE_MASK                  0x3
 186 #define CBAR_TYPE_S2_TRANS              (0 << CBAR_TYPE_SHIFT)
 187 #define CBAR_TYPE_S1_TRANS_S2_BYPASS    (1 << CBAR_TYPE_SHIFT)
 188 #define CBAR_TYPE_S1_TRANS_S2_FAULT     (2 << CBAR_TYPE_SHIFT)
 189 #define CBAR_TYPE_S1_TRANS_S2_TRANS     (3 << CBAR_TYPE_SHIFT)
 190 #define CBAR_IRPTNDX_SHIFT              24
 191 #define CBAR_IRPTNDX_MASK               0xff
 192
 193 #define ARM_SMMU_GR1_CBA2R(n)           (0x800 + ((n) << 2))
 194 #define CBA2R_RW64_32BIT                (0 << 0)
 195 #define CBA2R_RW64_64BIT                (1 << 0)
 196
 197 /* Translation context bank */
 198 #define ARM_SMMU_CB_BASE(smmu)          ((smmu)->base + ((smmu)->size >> 1))
 199 #define ARM_SMMU_CB(smmu, n)            ((n) * (1 << (smmu)->pgshift))
 200
 201 #define ARM_SMMU_CB_SCTLR               0x0
 202 #define ARM_SMMU_CB_RESUME              0x8
 203 #define ARM_SMMU_CB_TTBCR2              0x10
 204 #define ARM_SMMU_CB_TTBR0               0x20
 205 #define ARM_SMMU_CB_TTBR1               0x28
 206 #define ARM_SMMU_CB_TTBCR               0x30
 207 #define ARM_SMMU_CB_S1_MAIR0            0x38
 208 #define ARM_SMMU_CB_S1_MAIR1            0x3c
 209 #define ARM_SMMU_CB_PAR_LO              0x50
 210 #define ARM_SMMU_CB_PAR_HI              0x54
 211 #define ARM_SMMU_CB_FSR                 0x58
 212 #define ARM_SMMU_CB_FAR_LO              0x60
 213 #define ARM_SMMU_CB_FAR_HI              0x64
 214 #define ARM_SMMU_CB_FSYNR0              0x68
 215 #define ARM_SMMU_CB_S1_TLBIVA           0x600
 216 #define ARM_SMMU_CB_S1_TLBIASID         0x610
 217 #define ARM_SMMU_CB_S1_TLBIVAL          0x620
 218 #define ARM_SMMU_CB_S2_TLBIIPAS2        0x630
 219 #define ARM_SMMU_CB_S2_TLBIIPAS2L       0x638
 220 #define ARM_SMMU_CB_ATS1PR              0x800
 221 #define ARM_SMMU_CB_ATSR                0x8f0
 222
 223 #define SCTLR_S1_ASIDPNE                (1 << 12)
 224 #define SCTLR_CFCFG                     (1 << 7)
 225 #define SCTLR_CFIE                      (1 << 6)
 226 #define SCTLR_CFRE                      (1 << 5)
 227 #define SCTLR_E                         (1 << 4)
 228 #define SCTLR_AFE                       (1 << 2)
 229 #define SCTLR_TRE                       (1 << 1)
 230 #define SCTLR_M                         (1 << 0)
 231 #define SCTLR_EAE_SBOP                  (SCTLR_AFE | SCTLR_TRE)
 232
 233 #define CB_PAR_F                        (1 << 0)
 234
 235 #define ATSR_ACTIVE                     (1 << 0)
 236
 237 #define RESUME_RETRY                    (0 << 0)
 238 #define RESUME_TERMINATE                (1 << 0)
 239
 240 #define TTBCR2_SEP_SHIFT                15
 241 #define TTBCR2_SEP_UPSTREAM             (0x7 << TTBCR2_SEP_SHIFT)
 242
 243 #define TTBRn_ASID_SHIFT                48
 244
 245 #define FSR_MULTI                       (1 << 31)
 246 #define FSR_SS                          (1 << 30)
 247 #define FSR_UUT                         (1 << 8)
 248 #define FSR_ASF                         (1 << 7)
 249 #define FSR_TLBLKF                      (1 << 6)
 250 #define FSR_TLBMCF                      (1 << 5)
 251 #define FSR_EF                          (1 << 4)
 252 #define FSR_PF                          (1 << 3)
 253 #define FSR_AFF                         (1 << 2)
 254 #define FSR_TF                          (1 << 1)
 255
 256 #define FSR_IGN                         (FSR_AFF | FSR_ASF | \
 257                                          FSR_TLBMCF | FSR_TLBLKF)
 258 #define FSR_FAULT                       (FSR_MULTI | FSR_SS | FSR_UUT | \
 259                                          FSR_EF | FSR_PF | FSR_TF | FSR_IGN)
 260
 261 #define FSYNR0_WNR                      (1 << 4)
 262
 263 static int force_stage;
 264 module_param(force_stage, int, S_IRUGO);
 265 MODULE_PARM_DESC(force_stage,
 266         "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.");
 267 static bool disable_bypass;
 268 module_param(disable_bypass, bool, S_IRUGO);
 269 MODULE_PARM_DESC(disable_bypass,
 270         "Disable bypass streams such that incoming transactions from devices that are not attached to an iommu domain will report an abort back to the device and will not be allowed to pass through the SMMU.");
 271
 272 enum arm_smmu_arch_version {
 273         ARM_SMMU_V1 = 1,
 274         ARM_SMMU_V2,
 275 };
 276
 277 struct arm_smmu_smr {
 278         u8                              idx;
 279         u16                             mask;
 280         u16                             id;
 281 };
 282
 283 struct arm_smmu_master_cfg {
 284         int                             num_streamids;
 285         u16                             streamids[MAX_MASTER_STREAMIDS];
 286         struct arm_smmu_smr             *smrs;
 287 };
 288
 289 struct arm_smmu_master {
 290         struct device_node              *of_node;
 291         struct rb_node                  node;
 292         struct arm_smmu_master_cfg      cfg;
 293 };
 294
 295 struct arm_smmu_device {
 296         struct device                   *dev;
 297
 298         void __iomem                    *base;
 299         unsigned long                   size;
 300         unsigned long                   pgshift;
 301
 302 #define ARM_SMMU_FEAT_COHERENT_WALK     (1 << 0)
 303 #define ARM_SMMU_FEAT_STREAM_MATCH      (1 << 1)
 304 #define ARM_SMMU_FEAT_TRANS_S1          (1 << 2)
 305 #define ARM_SMMU_FEAT_TRANS_S2          (1 << 3)
 306 #define ARM_SMMU_FEAT_TRANS_NESTED      (1 << 4)
 307 #define ARM_SMMU_FEAT_TRANS_OPS         (1 << 5)
 308         u32                             features;
 309
 310 #define ARM_SMMU_OPT_SECURE_CFG_ACCESS (1 << 0)
 311         u32                             options;
 312         enum arm_smmu_arch_version      version;
 313
 314         u32                             num_context_banks;
 315         u32                             num_s2_context_banks;
 316         DECLARE_BITMAP(context_map, ARM_SMMU_MAX_CBS);
 317         atomic_t                        irptndx;
 318
 319         u32                             num_mapping_groups;
 320         DECLARE_BITMAP(smr_map, ARM_SMMU_MAX_SMRS);
 321
 322         unsigned long                   va_size;
 323         unsigned long                   ipa_size;
 324         unsigned long                   pa_size;
 325
 326         u32                             num_global_irqs;
 327         u32                             num_context_irqs;
 328         unsigned int                    *irqs;
 329
 330         struct list_head                list;
 331         struct rb_root                  masters;
 332 };
 333
 334 struct arm_smmu_cfg {
 335         u8                              cbndx;
 336         u8                              irptndx;
 337         u32                             cbar;
 338 };
 339 #define INVALID_IRPTNDX                 0xff
 340
 341 #define ARM_SMMU_CB_ASID(cfg)           ((cfg)->cbndx)
 342 #define ARM_SMMU_CB_VMID(cfg)           ((cfg)->cbndx + 1)
 343
 344 enum arm_smmu_domain_stage {
 345         ARM_SMMU_DOMAIN_S1 = 0,
 346         ARM_SMMU_DOMAIN_S2,
 347         ARM_SMMU_DOMAIN_NESTED,
 348 };
 349
 350 struct arm_smmu_domain {
 351         struct arm_smmu_device          *smmu;
 352         struct io_pgtable_ops           *pgtbl_ops;
 353         spinlock_t                      pgtbl_lock;
 354         struct arm_smmu_cfg             cfg;
 355         enum arm_smmu_domain_stage      stage;
 356         struct mutex                    init_mutex; /* Protects smmu pointer */
 357         struct iommu_domain             domain;
 358 };
 359
 360 static struct iommu_ops arm_smmu_ops;
 361
 362 static DEFINE_SPINLOCK(arm_smmu_devices_lock);
 363 static LIST_HEAD(arm_smmu_devices);
 364
 365 struct arm_smmu_option_prop {
 366         u32 opt;
 367         const char *prop;
 368 };
 369
 370 static struct arm_smmu_option_prop arm_smmu_options[] = {
 371         { ARM_SMMU_OPT_SECURE_CFG_ACCESS, "calxeda,smmu-secure-config-access" },
 372         { 0, NULL},
 373 };
 374
 375 static struct arm_smmu_domain *to_smmu_domain(struct iommu_domain *dom)
 376 {
 377         return container_of(dom, struct arm_smmu_domain, domain);
 378 }
 379
 380 static void parse_driver_options(struct arm_smmu_device *smmu)
 381 {
 382         int i = 0;
 383
 384         do {
 385                 if (of_property_read_bool(smmu->dev->of_node,
 386                                                 arm_smmu_options[i].prop)) {
 387                         smmu->options |= arm_smmu_options[i].opt;
 388                         dev_notice(smmu->dev, "option %s\n",
 389                                 arm_smmu_options[i].prop);
 390                 }
 391         } while (arm_smmu_options[++i].opt);
 392 }
 393
 394 static struct device_node *dev_get_dev_node(struct device *dev)
 395 {
 396         if (dev_is_pci(dev)) {
 397                 struct pci_bus *bus = to_pci_dev(dev)->bus;
 398
 399                 while (!pci_is_root_bus(bus))
 400                         bus = bus->parent;
 401                 return bus->bridge->parent->of_node;
 402         }
 403
 404         return dev->of_node;
 405 }
 406
 407 static struct arm_smmu_master *find_smmu_master(struct arm_smmu_device *smmu,
 408                                                 struct device_node *dev_node)
 409 {
 410         struct rb_node *node = smmu->masters.rb_node;
 411
 412         while (node) {
 413                 struct arm_smmu_master *master;
 414
 415                 master = container_of(node, struct arm_smmu_master, node);
 416
 417                 if (dev_node < master->of_node)
 418                         node = node->rb_left;
 419                 else if (dev_node > master->of_node)
 420                         node = node->rb_right;
 421                 else
 422                         return master;
 423         }
 424
 425         return NULL;
 426 }
 427
 428 static struct arm_smmu_master_cfg *
 429 find_smmu_master_cfg(struct device *dev)
 430 {
 431         struct arm_smmu_master_cfg *cfg = NULL;
 432         struct iommu_group *group = iommu_group_get(dev);
 433
 434         if (group) {
 435                 cfg = iommu_group_get_iommudata(group);
 436                 iommu_group_put(group);
 437         }
 438
 439         return cfg;
 440 }
 441
 442 static int insert_smmu_master(struct arm_smmu_device *smmu,
 443                               struct arm_smmu_master *master)
 444 {
 445         struct rb_node **new, *parent;
 446
 447         new = &smmu->masters.rb_node;
 448         parent = NULL;
 449         while (*new) {
 450                 struct arm_smmu_master *this
 451                         = container_of(*new, struct arm_smmu_master, node);
 452
 453                 parent = *new;
 454                 if (master->of_node < this->of_node)
 455                         new = &((*new)->rb_left);
 456                 else if (master->of_node > this->of_node)
 457                         new = &((*new)->rb_right);
 458                 else
 459                         return -EEXIST;
 460         }
 461
 462         rb_link_node(&master->node, parent, new);
 463         rb_insert_color(&master->node, &smmu->masters);
 464         return 0;
 465 }
 466
 467 static int register_smmu_master(struct arm_smmu_device *smmu,
 468                                 struct device *dev,
 469                                 struct of_phandle_args *masterspec)
 470 {
 471         int i;
 472         struct arm_smmu_master *master;
 473
 474         master = find_smmu_master(smmu, masterspec->np);
 475         if (master) {
 476                 dev_err(dev,
 477                         "rejecting multiple registrations for master device %s\n",
 478                         masterspec->np->name);
 479                 return -EBUSY;
 480         }
 481
 482         if (masterspec->args_count > MAX_MASTER_STREAMIDS) {
 483                 dev_err(dev,
 484                         "reached maximum number (%d) of stream IDs for master device %s\n",
 485                         MAX_MASTER_STREAMIDS, masterspec->np->name);
 486                 return -ENOSPC;
 487         }
 488
 489         master = devm_kzalloc(dev, sizeof(*master), GFP_KERNEL);
 490         if (!master)
 491                 return -ENOMEM;
 492
 493         master->of_node                 = masterspec->np;
 494         master->cfg.num_streamids       = masterspec->args_count;
 495
 496         for (i = 0; i < master->cfg.num_streamids; ++i) {
 497                 u16 streamid = masterspec->args[i];
 498
 499                 if (!(smmu->features & ARM_SMMU_FEAT_STREAM_MATCH) &&
 500                      (streamid >= smmu->num_mapping_groups)) {
 501                         dev_err(dev,
 502                                 "stream ID for master device %s greater than maximum allowed (%d)\n",
 503                                 masterspec->np->name, smmu->num_mapping_groups);
 504                         return -ERANGE;
 505                 }
 506                 master->cfg.streamids[i] = streamid;
 507         }
 508         return insert_smmu_master(smmu, master);
 509 }
 510
 511 static struct arm_smmu_device *find_smmu_for_device(struct device *dev)
 512 {
 513         struct arm_smmu_device *smmu;
 514         struct arm_smmu_master *master = NULL;
 515         struct device_node *dev_node = dev_get_dev_node(dev);
 516
 517         spin_lock(&arm_smmu_devices_lock);
 518         list_for_each_entry(smmu, &arm_smmu_devices, list) {
 519                 master = find_smmu_master(smmu, dev_node);
 520                 if (master)
 521                         break;
 522         }
 523         spin_unlock(&arm_smmu_devices_lock);
 524
 525         return master ? smmu : NULL;
 526 }
 527
 528 static int __arm_smmu_alloc_bitmap(unsigned long *map, int start, int end)
 529 {
 530         int idx;
 531
 532         do {
 533                 idx = find_next_zero_bit(map, end, start);
 534                 if (idx == end)
 535                         return -ENOSPC;
 536         } while (test_and_set_bit(idx, map));
 537
 538         return idx;
 539 }
 540
 541 static void __arm_smmu_free_bitmap(unsigned long *map, int idx)
 542 {
 543         clear_bit(idx, map);
 544 }
 545
 546 /* Wait for any pending TLB invalidations to complete */
 547 static void __arm_smmu_tlb_sync(struct arm_smmu_device *smmu)
 548 {
 549         int count = 0;
 550         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
 551
 552         writel_relaxed(0, gr0_base + ARM_SMMU_GR0_sTLBGSYNC);
 553         while (readl_relaxed(gr0_base + ARM_SMMU_GR0_sTLBGSTATUS)
 554                & sTLBGSTATUS_GSACTIVE) {
 555                 cpu_relax();
 556                 if (++count == TLB_LOOP_TIMEOUT) {
 557                         dev_err_ratelimited(smmu->dev,
 558                         "TLB sync timed out -- SMMU may be deadlocked\n");
 559                         return;
 560                 }
 561                 udelay(1);
 562         }
 563 }
 564
 565 static void arm_smmu_tlb_sync(void *cookie)
 566 {
 567         struct arm_smmu_domain *smmu_domain = cookie;
 568         __arm_smmu_tlb_sync(smmu_domain->smmu);
 569 }
 570
 571 static void arm_smmu_tlb_inv_context(void *cookie)
 572 {
 573         struct arm_smmu_domain *smmu_domain = cookie;
 574         struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
 575         struct arm_smmu_device *smmu = smmu_domain->smmu;
 576         bool stage1 = cfg->cbar != CBAR_TYPE_S2_TRANS;
 577         void __iomem *base;
 578
 579         if (stage1) {
 580                 base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
 581                 writel_relaxed(ARM_SMMU_CB_ASID(cfg),
 582                                base + ARM_SMMU_CB_S1_TLBIASID);
 583         } else {
 584                 base = ARM_SMMU_GR0(smmu);
 585                 writel_relaxed(ARM_SMMU_CB_VMID(cfg),
 586                                base + ARM_SMMU_GR0_TLBIVMID);
 587         }
 588
 589         __arm_smmu_tlb_sync(smmu);
 590 }
 591
 592 static void arm_smmu_tlb_inv_range_nosync(unsigned long iova, size_t size,
 593                                           size_t granule, bool leaf, void *cookie)
 594 {
 595         struct arm_smmu_domain *smmu_domain = cookie;
 596         struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
 597         struct arm_smmu_device *smmu = smmu_domain->smmu;
 598         bool stage1 = cfg->cbar != CBAR_TYPE_S2_TRANS;
 599         void __iomem *reg;
 600
 601         if (stage1) {
 602                 reg = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
 603                 reg += leaf ? ARM_SMMU_CB_S1_TLBIVAL : ARM_SMMU_CB_S1_TLBIVA;
 604
 605                 if (!IS_ENABLED(CONFIG_64BIT) || smmu->version == ARM_SMMU_V1) {
 606                         iova &= ~12UL;
 607                         iova |= ARM_SMMU_CB_ASID(cfg);
 608                         do {
 609                                 writel_relaxed(iova, reg);
 610                                 iova += granule;
 611                         } while (size -= granule);
 612 #ifdef CONFIG_64BIT
 613                 } else {
 614                         iova >>= 12;
 615                         iova |= (u64)ARM_SMMU_CB_ASID(cfg) << 48;
 616                         do {
 617                                 writeq_relaxed(iova, reg);
 618                                 iova += granule >> 12;
 619                         } while (size -= granule);
 620 #endif
 621                 }
 622 #ifdef CONFIG_64BIT
 623         } else if (smmu->version == ARM_SMMU_V2) {
 624                 reg = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
 625                 reg += leaf ? ARM_SMMU_CB_S2_TLBIIPAS2L :
 626                               ARM_SMMU_CB_S2_TLBIIPAS2;
 627                 iova >>= 12;
 628                 do {
 629                         writeq_relaxed(iova, reg);
 630                         iova += granule >> 12;
 631                 } while (size -= granule);
 632 #endif
 633         } else {
 634                 reg = ARM_SMMU_GR0(smmu) + ARM_SMMU_GR0_TLBIVMID;
 635                 writel_relaxed(ARM_SMMU_CB_VMID(cfg), reg);
 636         }
 637 }
 638
 639 static struct iommu_gather_ops arm_smmu_gather_ops = {
 640         .tlb_flush_all  = arm_smmu_tlb_inv_context,
 641         .tlb_add_flush  = arm_smmu_tlb_inv_range_nosync,
 642         .tlb_sync       = arm_smmu_tlb_sync,
 643 };
 644
 645 static irqreturn_t arm_smmu_context_fault(int irq, void *dev)
 646 {
 647         int flags, ret;
 648         u32 fsr, far, fsynr, resume;
 649         unsigned long iova;
 650         struct iommu_domain *domain = dev;
 651         struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
 652         struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
 653         struct arm_smmu_device *smmu = smmu_domain->smmu;
 654         void __iomem *cb_base;
 655
 656         cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
 657         fsr = readl_relaxed(cb_base + ARM_SMMU_CB_FSR);
 658
 659         if (!(fsr & FSR_FAULT))
 660                 return IRQ_NONE;
 661
 662         if (fsr & FSR_IGN)
 663                 dev_err_ratelimited(smmu->dev,
 664                                     "Unexpected context fault (fsr 0x%x)\n",
 665                                     fsr);
 666
 667         fsynr = readl_relaxed(cb_base + ARM_SMMU_CB_FSYNR0);
 668         flags = fsynr & FSYNR0_WNR ? IOMMU_FAULT_WRITE : IOMMU_FAULT_READ;
 669
 670         far = readl_relaxed(cb_base + ARM_SMMU_CB_FAR_LO);
 671         iova = far;
 672 #ifdef CONFIG_64BIT
 673         far = readl_relaxed(cb_base + ARM_SMMU_CB_FAR_HI);
 674         iova |= ((unsigned long)far << 32);
 675 #endif
 676
 677         if (!report_iommu_fault(domain, smmu->dev, iova, flags)) {
 678                 ret = IRQ_HANDLED;
 679                 resume = RESUME_RETRY;
 680         } else {
 681                 dev_err_ratelimited(smmu->dev,
 682                     "Unhandled context fault: iova=0x%08lx, fsynr=0x%x, cb=%d\n",
 683                     iova, fsynr, cfg->cbndx);
 684                 ret = IRQ_NONE;
 685                 resume = RESUME_TERMINATE;
 686         }
 687
 688         /* Clear the faulting FSR */
 689         writel(fsr, cb_base + ARM_SMMU_CB_FSR);
 690
 691         /* Retry or terminate any stalled transactions */
 692         if (fsr & FSR_SS)
 693                 writel_relaxed(resume, cb_base + ARM_SMMU_CB_RESUME);
 694
 695         return ret;
 696 }
 697
 698 static irqreturn_t arm_smmu_global_fault(int irq, void *dev)
 699 {
 700         u32 gfsr, gfsynr0, gfsynr1, gfsynr2;
 701         struct arm_smmu_device *smmu = dev;
 702         void __iomem *gr0_base = ARM_SMMU_GR0_NS(smmu);
 703
 704         gfsr = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSR);
 705         gfsynr0 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSYNR0);
 706         gfsynr1 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSYNR1);
 707         gfsynr2 = readl_relaxed(gr0_base + ARM_SMMU_GR0_sGFSYNR2);
 708
 709         if (!gfsr)
 710                 return IRQ_NONE;
 711
 712         dev_err_ratelimited(smmu->dev,
 713                 "Unexpected global fault, this could be serious\n");
 714         dev_err_ratelimited(smmu->dev,
 715                 "\tGFSR 0x%08x, GFSYNR0 0x%08x, GFSYNR1 0x%08x, GFSYNR2 0x%08x\n",
 716                 gfsr, gfsynr0, gfsynr1, gfsynr2);
 717
 718         writel(gfsr, gr0_base + ARM_SMMU_GR0_sGFSR);
 719         return IRQ_HANDLED;
 720 }
 721
 722 static void arm_smmu_init_context_bank(struct arm_smmu_domain *smmu_domain,
 723                                        struct io_pgtable_cfg *pgtbl_cfg)
 724 {
 725         u32 reg;
 726         u64 reg64;
 727         bool stage1;
 728         struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
 729         struct arm_smmu_device *smmu = smmu_domain->smmu;
 730         void __iomem *cb_base, *gr1_base;
 731
 732         gr1_base = ARM_SMMU_GR1(smmu);
 733         stage1 = cfg->cbar != CBAR_TYPE_S2_TRANS;
 734         cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
 735
 736         if (smmu->version > ARM_SMMU_V1) {
 737                 /*
 738                  * CBA2R.
 739                  * *Must* be initialised before CBAR thanks to VMID16
 740                  * architectural oversight affected some implementations.
 741                  */
 742 #ifdef CONFIG_64BIT
 743                 reg = CBA2R_RW64_64BIT;
 744 #else
 745                 reg = CBA2R_RW64_32BIT;
 746 #endif
 747                 writel_relaxed(reg, gr1_base + ARM_SMMU_GR1_CBA2R(cfg->cbndx));
 748         }
 749
 750         /* CBAR */
 751         reg = cfg->cbar;
 752         if (smmu->version == ARM_SMMU_V1)
 753                 reg |= cfg->irptndx << CBAR_IRPTNDX_SHIFT;
 754
 755         /*
 756          * Use the weakest shareability/memory types, so they are
 757          * overridden by the ttbcr/pte.
 758          */
 759         if (stage1) {
 760                 reg |= (CBAR_S1_BPSHCFG_NSH << CBAR_S1_BPSHCFG_SHIFT) |
 761                         (CBAR_S1_MEMATTR_WB << CBAR_S1_MEMATTR_SHIFT);
 762         } else {
 763                 reg |= ARM_SMMU_CB_VMID(cfg) << CBAR_VMID_SHIFT;
 764         }
 765         writel_relaxed(reg, gr1_base + ARM_SMMU_GR1_CBAR(cfg->cbndx));
 766
 767         /* TTBRs */
 768         if (stage1) {
 769                 reg64 = pgtbl_cfg->arm_lpae_s1_cfg.ttbr[0];
 770
 771                 reg64 |= ((u64)ARM_SMMU_CB_ASID(cfg)) << TTBRn_ASID_SHIFT;
 772                 smmu_writeq(reg64, cb_base + ARM_SMMU_CB_TTBR0);
 773
 774                 reg64 = pgtbl_cfg->arm_lpae_s1_cfg.ttbr[1];
 775                 reg64 |= ((u64)ARM_SMMU_CB_ASID(cfg)) << TTBRn_ASID_SHIFT;
 776                 smmu_writeq(reg64, cb_base + ARM_SMMU_CB_TTBR1);
 777         } else {
 778                 reg64 = pgtbl_cfg->arm_lpae_s2_cfg.vttbr;
 779                 smmu_writeq(reg64, cb_base + ARM_SMMU_CB_TTBR0);
 780         }
 781
 782         /* TTBCR */
 783         if (stage1) {
 784                 reg = pgtbl_cfg->arm_lpae_s1_cfg.tcr;
 785                 writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBCR);
 786                 if (smmu->version > ARM_SMMU_V1) {
 787                         reg = pgtbl_cfg->arm_lpae_s1_cfg.tcr >> 32;
 788                         reg |= TTBCR2_SEP_UPSTREAM;
 789                         writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBCR2);
 790                 }
 791         } else {
 792                 reg = pgtbl_cfg->arm_lpae_s2_cfg.vtcr;
 793                 writel_relaxed(reg, cb_base + ARM_SMMU_CB_TTBCR);
 794         }
 795
 796         /* MAIRs (stage-1 only) */
 797         if (stage1) {
 798                 reg = pgtbl_cfg->arm_lpae_s1_cfg.mair[0];
 799                 writel_relaxed(reg, cb_base + ARM_SMMU_CB_S1_MAIR0);
 800                 reg = pgtbl_cfg->arm_lpae_s1_cfg.mair[1];
 801                 writel_relaxed(reg, cb_base + ARM_SMMU_CB_S1_MAIR1);
 802         }
 803
 804         /* SCTLR */
 805         reg = SCTLR_CFCFG | SCTLR_CFIE | SCTLR_CFRE | SCTLR_M | SCTLR_EAE_SBOP;
 806         if (stage1)
 807                 reg |= SCTLR_S1_ASIDPNE;
 808 #ifdef __BIG_ENDIAN
 809         reg |= SCTLR_E;
 810 #endif
 811         writel_relaxed(reg, cb_base + ARM_SMMU_CB_SCTLR);
 812 }
 813
 814 static int arm_smmu_init_domain_context(struct iommu_domain *domain,
 815                                         struct arm_smmu_device *smmu)
 816 {
 817         int irq, start, ret = 0;
 818         unsigned long ias, oas;
 819         struct io_pgtable_ops *pgtbl_ops;
 820         struct io_pgtable_cfg pgtbl_cfg;
 821         enum io_pgtable_fmt fmt;
 822         struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
 823         struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
 824
 825         mutex_lock(&smmu_domain->init_mutex);
 826         if (smmu_domain->smmu)
 827                 goto out_unlock;
 828
 829         /*
 830          * Mapping the requested stage onto what we support is surprisingly
 831          * complicated, mainly because the spec allows S1+S2 SMMUs without
 832          * support for nested translation. That means we end up with the
 833          * following table:
 834          *
 835          * Requested        Supported        Actual
 836          *     S1               N              S1
 837          *     S1             S1+S2            S1
 838          *     S1               S2             S2
 839          *     S1               S1             S1
 840          *     N                N              N
 841          *     N              S1+S2            S2
 842          *     N                S2             S2
 843          *     N                S1             S1
 844          *
 845          * Note that you can't actually request stage-2 mappings.
 846          */
 847         if (!(smmu->features & ARM_SMMU_FEAT_TRANS_S1))
 848                 smmu_domain->stage = ARM_SMMU_DOMAIN_S2;
 849         if (!(smmu->features & ARM_SMMU_FEAT_TRANS_S2))
 850                 smmu_domain->stage = ARM_SMMU_DOMAIN_S1;
 851
 852         switch (smmu_domain->stage) {
 853         case ARM_SMMU_DOMAIN_S1:
 854                 cfg->cbar = CBAR_TYPE_S1_TRANS_S2_BYPASS;
 855                 start = smmu->num_s2_context_banks;
 856                 ias = smmu->va_size;
 857                 oas = smmu->ipa_size;
 858                 if (IS_ENABLED(CONFIG_64BIT))
 859                         fmt = ARM_64_LPAE_S1;
 860                 else
 861                         fmt = ARM_32_LPAE_S1;
 862                 break;
 863         case ARM_SMMU_DOMAIN_NESTED:
 864                 /*
 865                  * We will likely want to change this if/when KVM gets
 866                  * involved.
 867                  */
 868         case ARM_SMMU_DOMAIN_S2:
 869                 cfg->cbar = CBAR_TYPE_S2_TRANS;
 870                 start = 0;
 871                 ias = smmu->ipa_size;
 872                 oas = smmu->pa_size;
 873                 if (IS_ENABLED(CONFIG_64BIT))
 874                         fmt = ARM_64_LPAE_S2;
 875                 else
 876                         fmt = ARM_32_LPAE_S2;
 877                 break;
 878         default:
 879                 ret = -EINVAL;
 880                 goto out_unlock;
 881         }
 882
 883         ret = __arm_smmu_alloc_bitmap(smmu->context_map, start,
 884                                       smmu->num_context_banks);
 885         if (IS_ERR_VALUE(ret))
 886                 goto out_unlock;
 887
 888         cfg->cbndx = ret;
 889         if (smmu->version == ARM_SMMU_V1) {
 890                 cfg->irptndx = atomic_inc_return(&smmu->irptndx);
 891                 cfg->irptndx %= smmu->num_context_irqs;
 892         } else {
 893                 cfg->irptndx = cfg->cbndx;
 894         }
 895
 896         pgtbl_cfg = (struct io_pgtable_cfg) {
 897                 .pgsize_bitmap  = arm_smmu_ops.pgsize_bitmap,
 898                 .ias            = ias,
 899                 .oas            = oas,
 900                 .tlb            = &arm_smmu_gather_ops,
 901                 .iommu_dev      = smmu->dev,
 902         };
 903
 904         smmu_domain->smmu = smmu;
 905         pgtbl_ops = alloc_io_pgtable_ops(fmt, &pgtbl_cfg, smmu_domain);
 906         if (!pgtbl_ops) {
 907                 ret = -ENOMEM;
 908                 goto out_clear_smmu;
 909         }
 910
 911         /* Update our support page sizes to reflect the page table format */
 912         arm_smmu_ops.pgsize_bitmap = pgtbl_cfg.pgsize_bitmap;
 913
 914         /* Initialise the context bank with our page table cfg */
 915         arm_smmu_init_context_bank(smmu_domain, &pgtbl_cfg);
 916
 917         /*
 918          * Request context fault interrupt. Do this last to avoid the
 919          * handler seeing a half-initialised domain state.
 920          */
 921         irq = smmu->irqs[smmu->num_global_irqs + cfg->irptndx];
 922         ret = request_irq(irq, arm_smmu_context_fault, IRQF_SHARED,
 923                           "arm-smmu-context-fault", domain);
 924         if (IS_ERR_VALUE(ret)) {
 925                 dev_err(smmu->dev, "failed to request context IRQ %d (%u)\n",
 926                         cfg->irptndx, irq);
 927                 cfg->irptndx = INVALID_IRPTNDX;
 928         }
 929
 930         mutex_unlock(&smmu_domain->init_mutex);
 931
 932         /* Publish page table ops for map/unmap */
 933         smmu_domain->pgtbl_ops = pgtbl_ops;
 934         return 0;
 935
 936 out_clear_smmu:
 937         smmu_domain->smmu = NULL;
 938 out_unlock:
 939         mutex_unlock(&smmu_domain->init_mutex);
 940         return ret;
 941 }
 942
 943 static void arm_smmu_destroy_domain_context(struct iommu_domain *domain)
 944 {
 945         struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
 946         struct arm_smmu_device *smmu = smmu_domain->smmu;
 947         struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
 948         void __iomem *cb_base;
 949         int irq;
 950
 951         if (!smmu)
 952                 return;
 953
 954         /*
 955          * Disable the context bank and free the page tables before freeing
 956          * it.
 957          */
 958         cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
 959         writel_relaxed(0, cb_base + ARM_SMMU_CB_SCTLR);
 960
 961         if (cfg->irptndx != INVALID_IRPTNDX) {
 962                 irq = smmu->irqs[smmu->num_global_irqs + cfg->irptndx];
 963                 free_irq(irq, domain);
 964         }
 965
 966         free_io_pgtable_ops(smmu_domain->pgtbl_ops);
 967         __arm_smmu_free_bitmap(smmu->context_map, cfg->cbndx);
 968 }
 969
 970 static struct iommu_domain *arm_smmu_domain_alloc(unsigned type)
 971 {
 972         struct arm_smmu_domain *smmu_domain;
 973
 974         if (type != IOMMU_DOMAIN_UNMANAGED && type != IOMMU_DOMAIN_DMA)
 975                 return NULL;
 976         /*
 977          * Allocate the domain and initialise some of its data structures.
 978          * We can't really do anything meaningful until we've added a
 979          * master.
 980          */
 981         smmu_domain = kzalloc(sizeof(*smmu_domain), GFP_KERNEL);
 982         if (!smmu_domain)
 983                 return NULL;
 984
 985         if (type == IOMMU_DOMAIN_DMA &&
 986             iommu_get_dma_cookie(&smmu_domain->domain)) {
 987                 kfree(smmu_domain);
 988                 return NULL;
 989         }
 990
 991         mutex_init(&smmu_domain->init_mutex);
 992         spin_lock_init(&smmu_domain->pgtbl_lock);
 993
 994         return &smmu_domain->domain;
 995 }
 996
 997 static void arm_smmu_domain_free(struct iommu_domain *domain)
 998 {
 999         struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
1000
1001         /*
1002          * Free the domain resources. We assume that all devices have
1003          * already been detached.
1004          */
1005         iommu_put_dma_cookie(domain);
1006         arm_smmu_destroy_domain_context(domain);
1007         kfree(smmu_domain);
1008 }
1009
1010 static int arm_smmu_master_configure_smrs(struct arm_smmu_device *smmu,
1011                                           struct arm_smmu_master_cfg *cfg)
1012 {
1013         int i;
1014         struct arm_smmu_smr *smrs;
1015         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1016
1017         if (!(smmu->features & ARM_SMMU_FEAT_STREAM_MATCH))
1018                 return 0;
1019
1020         if (cfg->smrs)
1021                 return -EEXIST;
1022
1023         smrs = kmalloc_array(cfg->num_streamids, sizeof(*smrs), GFP_KERNEL);
1024         if (!smrs) {
1025                 dev_err(smmu->dev, "failed to allocate %d SMRs\n",
1026                         cfg->num_streamids);
1027                 return -ENOMEM;
1028         }
1029
1030         /* Allocate the SMRs on the SMMU */
1031         for (i = 0; i < cfg->num_streamids; ++i) {
1032                 int idx = __arm_smmu_alloc_bitmap(smmu->smr_map, 0,
1033                                                   smmu->num_mapping_groups);
1034                 if (IS_ERR_VALUE(idx)) {
1035                         dev_err(smmu->dev, "failed to allocate free SMR\n");
1036                         goto err_free_smrs;
1037                 }
1038
1039                 smrs[i] = (struct arm_smmu_smr) {
1040                         .idx    = idx,
1041                         .mask   = 0, /* We don't currently share SMRs */
1042                         .id     = cfg->streamids[i],
1043                 };
1044         }
1045
1046         /* It worked! Now, poke the actual hardware */
1047         for (i = 0; i < cfg->num_streamids; ++i) {
1048                 u32 reg = SMR_VALID | smrs[i].id << SMR_ID_SHIFT |
1049                           smrs[i].mask << SMR_MASK_SHIFT;
1050                 writel_relaxed(reg, gr0_base + ARM_SMMU_GR0_SMR(smrs[i].idx));
1051         }
1052
1053         cfg->smrs = smrs;
1054         return 0;
1055
1056 err_free_smrs:
1057         while (--i >= 0)
1058                 __arm_smmu_free_bitmap(smmu->smr_map, smrs[i].idx);
1059         kfree(smrs);
1060         return -ENOSPC;
1061 }
1062
1063 static void arm_smmu_master_free_smrs(struct arm_smmu_device *smmu,
1064                                       struct arm_smmu_master_cfg *cfg)
1065 {
1066         int i;
1067         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1068         struct arm_smmu_smr *smrs = cfg->smrs;
1069
1070         if (!smrs)
1071                 return;
1072
1073         /* Invalidate the SMRs before freeing back to the allocator */
1074         for (i = 0; i < cfg->num_streamids; ++i) {
1075                 u8 idx = smrs[i].idx;
1076
1077                 writel_relaxed(~SMR_VALID, gr0_base + ARM_SMMU_GR0_SMR(idx));
1078                 __arm_smmu_free_bitmap(smmu->smr_map, idx);
1079         }
1080
1081         cfg->smrs = NULL;
1082         kfree(smrs);
1083 }
1084
1085 static int arm_smmu_domain_add_master(struct arm_smmu_domain *smmu_domain,
1086                                       struct arm_smmu_master_cfg *cfg)
1087 {
1088         int i, ret;
1089         struct arm_smmu_device *smmu = smmu_domain->smmu;
1090         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1091
1092         /* Devices in an IOMMU group may already be configured */
1093         ret = arm_smmu_master_configure_smrs(smmu, cfg);
1094         if (ret)
1095                 return ret == -EEXIST ? 0 : ret;
1096
1097         for (i = 0; i < cfg->num_streamids; ++i) {
1098                 u32 idx, s2cr;
1099
1100                 idx = cfg->smrs ? cfg->smrs[i].idx : cfg->streamids[i];
1101                 s2cr = S2CR_TYPE_TRANS | S2CR_PRIVCFG_UNPRIV |
1102                        (smmu_domain->cfg.cbndx << S2CR_CBNDX_SHIFT);
1103                 writel_relaxed(s2cr, gr0_base + ARM_SMMU_GR0_S2CR(idx));
1104         }
1105
1106         return 0;
1107 }
1108
1109 static void arm_smmu_domain_remove_master(struct arm_smmu_domain *smmu_domain,
1110                                           struct arm_smmu_master_cfg *cfg)
1111 {
1112         int i;
1113         struct arm_smmu_device *smmu = smmu_domain->smmu;
1114         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1115
1116         /* An IOMMU group is torn down by the first device to be removed */
1117         if ((smmu->features & ARM_SMMU_FEAT_STREAM_MATCH) && !cfg->smrs)
1118                 return;
1119
1120         /*
1121          * We *must* clear the S2CR first, because freeing the SMR means
1122          * that it can be re-allocated immediately.
1123          */
1124         for (i = 0; i < cfg->num_streamids; ++i) {
1125                 u32 idx = cfg->smrs ? cfg->smrs[i].idx : cfg->streamids[i];
1126                 u32 reg = disable_bypass ? S2CR_TYPE_FAULT : S2CR_TYPE_BYPASS;
1127
1128                 writel_relaxed(reg, gr0_base + ARM_SMMU_GR0_S2CR(idx));
1129         }
1130
1131         arm_smmu_master_free_smrs(smmu, cfg);
1132 }
1133
1134 static int arm_smmu_attach_dev(struct iommu_domain *domain, struct device *dev)
1135 {
1136         int ret;
1137         struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
1138         struct arm_smmu_device *smmu;
1139         struct arm_smmu_master_cfg *cfg;
1140
1141         smmu = find_smmu_for_device(dev);
1142         if (!smmu) {
1143                 dev_err(dev, "cannot attach to SMMU, is it on the same bus?\n");
1144                 return -ENXIO;
1145         }
1146
1147         if (dev->archdata.iommu) {
1148                 dev_err(dev, "already attached to IOMMU domain\n");
1149                 return -EEXIST;
1150         }
1151
1152         /* Ensure that the domain is finalised */
1153         ret = arm_smmu_init_domain_context(domain, smmu);
1154         if (IS_ERR_VALUE(ret))
1155                 return ret;
1156
1157         /*
1158          * Sanity check the domain. We don't support domains across
1159          * different SMMUs.
1160          */
1161         if (smmu_domain->smmu != smmu) {
1162                 dev_err(dev,
1163                         "cannot attach to SMMU %s whilst already attached to domain on SMMU %s\n",
1164                         dev_name(smmu_domain->smmu->dev), dev_name(smmu->dev));
1165                 return -EINVAL;
1166         }
1167
1168         /* Looks ok, so add the device to the domain */
1169         cfg = find_smmu_master_cfg(dev);
1170         if (!cfg)
1171                 return -ENODEV;
1172
1173         ret = arm_smmu_domain_add_master(smmu_domain, cfg);
1174         if (!ret)
1175                 dev->archdata.iommu = domain;
1176         return ret;
1177 }
1178
1179 static void arm_smmu_detach_dev(struct iommu_domain *domain, struct device *dev)
1180 {
1181         struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
1182         struct arm_smmu_master_cfg *cfg;
1183
1184         cfg = find_smmu_master_cfg(dev);
1185         if (!cfg)
1186                 return;
1187
1188         dev->archdata.iommu = NULL;
1189         arm_smmu_domain_remove_master(smmu_domain, cfg);
1190 }
1191
1192 static int arm_smmu_map(struct iommu_domain *domain, unsigned long iova,
1193                         phys_addr_t paddr, size_t size, int prot)
1194 {
1195         int ret;
1196         unsigned long flags;
1197         struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
1198         struct io_pgtable_ops *ops= smmu_domain->pgtbl_ops;
1199
1200         if (!ops)
1201                 return -ENODEV;
1202
1203         spin_lock_irqsave(&smmu_domain->pgtbl_lock, flags);
1204         ret = ops->map(ops, iova, paddr, size, prot);
1205         spin_unlock_irqrestore(&smmu_domain->pgtbl_lock, flags);
1206         return ret;
1207 }
1208
1209 static size_t arm_smmu_unmap(struct iommu_domain *domain, unsigned long iova,
1210                              size_t size)
1211 {
1212         size_t ret;
1213         unsigned long flags;
1214         struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
1215         struct io_pgtable_ops *ops= smmu_domain->pgtbl_ops;
1216
1217         if (!ops)
1218                 return 0;
1219
1220         spin_lock_irqsave(&smmu_domain->pgtbl_lock, flags);
1221         ret = ops->unmap(ops, iova, size);
1222         spin_unlock_irqrestore(&smmu_domain->pgtbl_lock, flags);
1223         return ret;
1224 }
1225
1226 static phys_addr_t arm_smmu_iova_to_phys_hard(struct iommu_domain *domain,
1227                                               dma_addr_t iova)
1228 {
1229         struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
1230         struct arm_smmu_device *smmu = smmu_domain->smmu;
1231         struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
1232         struct io_pgtable_ops *ops= smmu_domain->pgtbl_ops;
1233         struct device *dev = smmu->dev;
1234         void __iomem *cb_base;
1235         u32 tmp;
1236         u64 phys;
1237         unsigned long va;
1238
1239         cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
1240
1241         /* ATS1 registers can only be written atomically */
1242         va = iova & ~0xfffUL;
1243         if (smmu->version == ARM_SMMU_V2)
1244                 smmu_writeq(va, cb_base + ARM_SMMU_CB_ATS1PR);
1245         else
1246                 writel_relaxed(va, cb_base + ARM_SMMU_CB_ATS1PR);
1247
1248         if (readl_poll_timeout_atomic(cb_base + ARM_SMMU_CB_ATSR, tmp,
1249                                       !(tmp & ATSR_ACTIVE), 5, 50)) {
1250                 dev_err(dev,
1251                         "iova to phys timed out on %pad. Falling back to software table walk.\n",
1252                         &iova);
1253                 return ops->iova_to_phys(ops, iova);
1254         }
1255
1256         phys = readl_relaxed(cb_base + ARM_SMMU_CB_PAR_LO);
1257         phys |= ((u64)readl_relaxed(cb_base + ARM_SMMU_CB_PAR_HI)) << 32;
1258
1259         if (phys & CB_PAR_F) {
1260                 dev_err(dev, "translation fault!\n");
1261                 dev_err(dev, "PAR = 0x%llx\n", phys);
1262                 return 0;
1263         }
1264
1265         return (phys & GENMASK_ULL(39, 12)) | (iova & 0xfff);
1266 }
1267
1268 static phys_addr_t arm_smmu_iova_to_phys(struct iommu_domain *domain,
1269                                         dma_addr_t iova)
1270 {
1271         phys_addr_t ret;
1272         unsigned long flags;
1273         struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
1274         struct io_pgtable_ops *ops= smmu_domain->pgtbl_ops;
1275
1276         if (!ops)
1277                 return 0;
1278
1279         spin_lock_irqsave(&smmu_domain->pgtbl_lock, flags);
1280         if (smmu_domain->smmu->features & ARM_SMMU_FEAT_TRANS_OPS &&
1281                         smmu_domain->stage == ARM_SMMU_DOMAIN_S1) {
1282                 ret = arm_smmu_iova_to_phys_hard(domain, iova);
1283         } else {
1284                 ret = ops->iova_to_phys(ops, iova);
1285         }
1286
1287         spin_unlock_irqrestore(&smmu_domain->pgtbl_lock, flags);
1288
1289         return ret;
1290 }
1291
1292 static bool arm_smmu_capable(enum iommu_cap cap)
1293 {
1294         switch (cap) {
1295         case IOMMU_CAP_CACHE_COHERENCY:
1296                 /*
1297                  * Return true here as the SMMU can always send out coherent
1298                  * requests.
1299                  */
1300                 return true;
1301         case IOMMU_CAP_INTR_REMAP:
1302                 return true; /* MSIs are just memory writes */
1303         case IOMMU_CAP_NOEXEC:
1304                 return true;
1305         default:
1306                 return false;
1307         }
1308 }
1309
1310 static int __arm_smmu_get_pci_sid(struct pci_dev *pdev, u16 alias, void *data)
1311 {
1312         *((u16 *)data) = alias;
1313         return 0; /* Continue walking */
1314 }
1315
1316 static void __arm_smmu_release_pci_iommudata(void *data)
1317 {
1318         kfree(data);
1319 }
1320
1321 static int arm_smmu_init_pci_device(struct pci_dev *pdev,
1322                                     struct iommu_group *group)
1323 {
1324         struct arm_smmu_master_cfg *cfg;
1325         u16 sid;
1326         int i;
1327
1328         cfg = iommu_group_get_iommudata(group);
1329         if (!cfg) {
1330                 cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1331                 if (!cfg)
1332                         return -ENOMEM;
1333
1334                 iommu_group_set_iommudata(group, cfg,
1335                                           __arm_smmu_release_pci_iommudata);
1336         }
1337
1338         if (cfg->num_streamids >= MAX_MASTER_STREAMIDS)
1339                 return -ENOSPC;
1340
1341         /*
1342          * Assume Stream ID == Requester ID for now.
1343          * We need a way to describe the ID mappings in FDT.
1344          */
1345         pci_for_each_dma_alias(pdev, __arm_smmu_get_pci_sid, &sid);
1346         for (i = 0; i < cfg->num_streamids; ++i)
1347                 if (cfg->streamids[i] == sid)
1348                         break;
1349
1350         /* Avoid duplicate SIDs, as this can lead to SMR conflicts */
1351         if (i == cfg->num_streamids)
1352                 cfg->streamids[cfg->num_streamids++] = sid;
1353
1354         return 0;
1355 }
1356
1357 static int arm_smmu_init_platform_device(struct device *dev,
1358                                          struct iommu_group *group)
1359 {
1360         struct arm_smmu_device *smmu = find_smmu_for_device(dev);
1361         struct arm_smmu_master *master;
1362
1363         if (!smmu)
1364                 return -ENODEV;
1365
1366         master = find_smmu_master(smmu, dev->of_node);
1367         if (!master)
1368                 return -ENODEV;
1369
1370         iommu_group_set_iommudata(group, &master->cfg, NULL);
1371
1372         return 0;
1373 }
1374
1375 static int arm_smmu_add_device(struct device *dev)
1376 {
1377         struct iommu_group *group;
1378
1379         group = iommu_group_get_for_dev(dev);
1380         if (IS_ERR(group))
1381                 return PTR_ERR(group);
1382
1383         iommu_group_put(group);
1384         return 0;
1385 }
1386
1387 static void arm_smmu_remove_device(struct device *dev)
1388 {
1389         iommu_group_remove_device(dev);
1390 }
1391
1392 static struct iommu_group *arm_smmu_device_group(struct device *dev)
1393 {
1394         struct iommu_group *group;
1395         int ret;
1396
1397         if (dev_is_pci(dev))
1398                 group = pci_device_group(dev);
1399         else
1400                 group = generic_device_group(dev);
1401
1402         if (IS_ERR(group))
1403                 return group;
1404
1405         if (dev_is_pci(dev))
1406                 ret = arm_smmu_init_pci_device(to_pci_dev(dev), group);
1407         else
1408                 ret = arm_smmu_init_platform_device(dev, group);
1409
1410         if (ret) {
1411                 iommu_group_put(group);
1412                 group = ERR_PTR(ret);
1413         }
1414
1415         return group;
1416 }
1417
1418 static int arm_smmu_domain_get_attr(struct iommu_domain *domain,
1419                                     enum iommu_attr attr, void *data)
1420 {
1421         struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
1422
1423         switch (attr) {
1424         case DOMAIN_ATTR_NESTING:
1425                 *(int *)data = (smmu_domain->stage == ARM_SMMU_DOMAIN_NESTED);
1426                 return 0;
1427         default:
1428                 return -ENODEV;
1429         }
1430 }
1431
1432 static int arm_smmu_domain_set_attr(struct iommu_domain *domain,
1433                                     enum iommu_attr attr, void *data)
1434 {
1435         int ret = 0;
1436         struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
1437
1438         mutex_lock(&smmu_domain->init_mutex);
1439
1440         switch (attr) {
1441         case DOMAIN_ATTR_NESTING:
1442                 if (smmu_domain->smmu) {
1443                         ret = -EPERM;
1444                         goto out_unlock;
1445                 }
1446
1447                 if (*(int *)data)
1448                         smmu_domain->stage = ARM_SMMU_DOMAIN_NESTED;
1449                 else
1450                         smmu_domain->stage = ARM_SMMU_DOMAIN_S1;
1451
1452                 break;
1453         default:
1454                 ret = -ENODEV;
1455         }
1456
1457 out_unlock:
1458         mutex_unlock(&smmu_domain->init_mutex);
1459         return ret;
1460 }
1461
1462 static struct iommu_ops arm_smmu_ops = {
1463         .capable                = arm_smmu_capable,
1464         .domain_alloc           = arm_smmu_domain_alloc,
1465         .domain_free            = arm_smmu_domain_free,
1466         .attach_dev             = arm_smmu_attach_dev,
1467         .detach_dev             = arm_smmu_detach_dev,
1468         .map                    = arm_smmu_map,
1469         .unmap                  = arm_smmu_unmap,
1470         .map_sg                 = default_iommu_map_sg,
1471         .iova_to_phys           = arm_smmu_iova_to_phys,
1472         .add_device             = arm_smmu_add_device,
1473         .remove_device          = arm_smmu_remove_device,
1474         .device_group           = arm_smmu_device_group,
1475         .domain_get_attr        = arm_smmu_domain_get_attr,
1476         .domain_set_attr        = arm_smmu_domain_set_attr,
1477         .pgsize_bitmap          = -1UL, /* Restricted during device attach */
1478 };
1479
1480 static void arm_smmu_device_reset(struct arm_smmu_device *smmu)
1481 {
1482         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1483         void __iomem *cb_base;
1484         int i = 0;
1485         u32 reg;
1486
1487         /* clear global FSR */
1488         reg = readl_relaxed(ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sGFSR);
1489         writel(reg, ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sGFSR);
1490
1491         /* Mark all SMRn as invalid and all S2CRn as bypass unless overridden */
1492         reg = disable_bypass ? S2CR_TYPE_FAULT : S2CR_TYPE_BYPASS;
1493         for (i = 0; i < smmu->num_mapping_groups; ++i) {
1494                 writel_relaxed(0, gr0_base + ARM_SMMU_GR0_SMR(i));
1495                 writel_relaxed(reg, gr0_base + ARM_SMMU_GR0_S2CR(i));
1496         }
1497
1498         /* Make sure all context banks are disabled and clear CB_FSR  */
1499         for (i = 0; i < smmu->num_context_banks; ++i) {
1500                 cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, i);
1501                 writel_relaxed(0, cb_base + ARM_SMMU_CB_SCTLR);
1502                 writel_relaxed(FSR_FAULT, cb_base + ARM_SMMU_CB_FSR);
1503         }
1504
1505         /* Invalidate the TLB, just in case */
1506         writel_relaxed(0, gr0_base + ARM_SMMU_GR0_TLBIALLH);
1507         writel_relaxed(0, gr0_base + ARM_SMMU_GR0_TLBIALLNSNH);
1508
1509         reg = readl_relaxed(ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sCR0);
1510
1511         /* Enable fault reporting */
1512         reg |= (sCR0_GFRE | sCR0_GFIE | sCR0_GCFGFRE | sCR0_GCFGFIE);
1513
1514         /* Disable TLB broadcasting. */
1515         reg |= (sCR0_VMIDPNE | sCR0_PTM);
1516
1517         /* Enable client access, handling unmatched streams as appropriate */
1518         reg &= ~sCR0_CLIENTPD;
1519         if (disable_bypass)
1520                 reg |= sCR0_USFCFG;
1521         else
1522                 reg &= ~sCR0_USFCFG;
1523
1524         /* Disable forced broadcasting */
1525         reg &= ~sCR0_FB;
1526
1527         /* Don't upgrade barriers */
1528         reg &= ~(sCR0_BSU_MASK << sCR0_BSU_SHIFT);
1529
1530         /* Push the button */
1531         __arm_smmu_tlb_sync(smmu);
1532         writel(reg, ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sCR0);
1533 }
1534
1535 static int arm_smmu_id_size_to_bits(int size)
1536 {
1537         switch (size) {
1538         case 0:
1539                 return 32;
1540         case 1:
1541                 return 36;
1542         case 2:
1543                 return 40;
1544         case 3:
1545                 return 42;
1546         case 4:
1547                 return 44;
1548         case 5:
1549         default:
1550                 return 48;
1551         }
1552 }
1553
1554 static int arm_smmu_device_cfg_probe(struct arm_smmu_device *smmu)
1555 {
1556         unsigned long size;
1557         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1558         u32 id;
1559         bool cttw_dt, cttw_reg;
1560
1561         dev_notice(smmu->dev, "probing hardware configuration...\n");
1562         dev_notice(smmu->dev, "SMMUv%d with:\n", smmu->version);
1563
1564         /* ID0 */
1565         id = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID0);
1566
1567         /* Restrict available stages based on module parameter */
1568         if (force_stage == 1)
1569                 id &= ~(ID0_S2TS | ID0_NTS);
1570         else if (force_stage == 2)
1571                 id &= ~(ID0_S1TS | ID0_NTS);
1572
1573         if (id & ID0_S1TS) {
1574                 smmu->features |= ARM_SMMU_FEAT_TRANS_S1;
1575                 dev_notice(smmu->dev, "\tstage 1 translation\n");
1576         }
1577
1578         if (id & ID0_S2TS) {
1579                 smmu->features |= ARM_SMMU_FEAT_TRANS_S2;
1580                 dev_notice(smmu->dev, "\tstage 2 translation\n");
1581         }
1582
1583         if (id & ID0_NTS) {
1584                 smmu->features |= ARM_SMMU_FEAT_TRANS_NESTED;
1585                 dev_notice(smmu->dev, "\tnested translation\n");
1586         }
1587
1588         if (!(smmu->features &
1589                 (ARM_SMMU_FEAT_TRANS_S1 | ARM_SMMU_FEAT_TRANS_S2))) {
1590                 dev_err(smmu->dev, "\tno translation support!\n");
1591                 return -ENODEV;
1592         }
1593
1594         if ((id & ID0_S1TS) && ((smmu->version == 1) || !(id & ID0_ATOSNS))) {
1595                 smmu->features |= ARM_SMMU_FEAT_TRANS_OPS;
1596                 dev_notice(smmu->dev, "\taddress translation ops\n");
1597         }
1598
1599         /*
1600          * In order for DMA API calls to work properly, we must defer to what
1601          * the DT says about coherency, regardless of what the hardware claims.
1602          * Fortunately, this also opens up a workaround for systems where the
1603          * ID register value has ended up configured incorrectly.
1604          */
1605         cttw_dt = of_dma_is_coherent(smmu->dev->of_node);
1606         cttw_reg = !!(id & ID0_CTTW);
1607         if (cttw_dt)
1608                 smmu->features |= ARM_SMMU_FEAT_COHERENT_WALK;
1609         if (cttw_dt || cttw_reg)
1610                 dev_notice(smmu->dev, "\t%scoherent table walk\n",
1611                            cttw_dt ? "" : "non-");
1612         if (cttw_dt != cttw_reg)
1613                 dev_notice(smmu->dev,
1614                            "\t(IDR0.CTTW overridden by dma-coherent property)\n");
1615
1616         if (id & ID0_SMS) {
1617                 u32 smr, sid, mask;
1618
1619                 smmu->features |= ARM_SMMU_FEAT_STREAM_MATCH;
1620                 smmu->num_mapping_groups = (id >> ID0_NUMSMRG_SHIFT) &
1621                                            ID0_NUMSMRG_MASK;
1622                 if (smmu->num_mapping_groups == 0) {
1623                         dev_err(smmu->dev,
1624                                 "stream-matching supported, but no SMRs present!\n");
1625                         return -ENODEV;
1626                 }
1627
1628                 smr = SMR_MASK_MASK << SMR_MASK_SHIFT;
1629                 smr |= (SMR_ID_MASK << SMR_ID_SHIFT);
1630                 writel_relaxed(smr, gr0_base + ARM_SMMU_GR0_SMR(0));
1631                 smr = readl_relaxed(gr0_base + ARM_SMMU_GR0_SMR(0));
1632
1633                 mask = (smr >> SMR_MASK_SHIFT) & SMR_MASK_MASK;
1634                 sid = (smr >> SMR_ID_SHIFT) & SMR_ID_MASK;
1635                 if ((mask & sid) != sid) {
1636                         dev_err(smmu->dev,
1637                                 "SMR mask bits (0x%x) insufficient for ID field (0x%x)\n",
1638                                 mask, sid);
1639                         return -ENODEV;
1640                 }
1641
1642                 dev_notice(smmu->dev,
1643                            "\tstream matching with %u register groups, mask 0x%x",
1644                            smmu->num_mapping_groups, mask);
1645         } else {
1646                 smmu->num_mapping_groups = (id >> ID0_NUMSIDB_SHIFT) &
1647                                            ID0_NUMSIDB_MASK;
1648         }
1649
1650         /* ID1 */
1651         id = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID1);
1652         smmu->pgshift = (id & ID1_PAGESIZE) ? 16 : 12;
1653
1654         /* Check for size mismatch of SMMU address space from mapped region */
1655         size = 1 << (((id >> ID1_NUMPAGENDXB_SHIFT) & ID1_NUMPAGENDXB_MASK) + 1);
1656         size *= 2 << smmu->pgshift;
1657         if (smmu->size != size)
1658                 dev_warn(smmu->dev,
1659                         "SMMU address space size (0x%lx) differs from mapped region size (0x%lx)!\n",
1660                         size, smmu->size);
1661
1662         smmu->num_s2_context_banks = (id >> ID1_NUMS2CB_SHIFT) & ID1_NUMS2CB_MASK;
1663         smmu->num_context_banks = (id >> ID1_NUMCB_SHIFT) & ID1_NUMCB_MASK;
1664         if (smmu->num_s2_context_banks > smmu->num_context_banks) {
1665                 dev_err(smmu->dev, "impossible number of S2 context banks!\n");
1666                 return -ENODEV;
1667         }
1668         dev_notice(smmu->dev, "\t%u context banks (%u stage-2 only)\n",
1669                    smmu->num_context_banks, smmu->num_s2_context_banks);
1670
1671         /* ID2 */
1672         id = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID2);
1673         size = arm_smmu_id_size_to_bits((id >> ID2_IAS_SHIFT) & ID2_IAS_MASK);
1674         smmu->ipa_size = size;
1675
1676         /* The output mask is also applied for bypass */
1677         size = arm_smmu_id_size_to_bits((id >> ID2_OAS_SHIFT) & ID2_OAS_MASK);
1678         smmu->pa_size = size;
1679
1680         /*
1681          * What the page table walker can address actually depends on which
1682          * descriptor format is in use, but since a) we don't know that yet,
1683          * and b) it can vary per context bank, this will have to do...
1684          */
1685         if (dma_set_mask_and_coherent(smmu->dev, DMA_BIT_MASK(size)))
1686                 dev_warn(smmu->dev,
1687                          "failed to set DMA mask for table walker\n");
1688
1689         if (smmu->version == ARM_SMMU_V1) {
1690                 smmu->va_size = smmu->ipa_size;
1691                 size = SZ_4K | SZ_2M | SZ_1G;
1692         } else {
1693                 size = (id >> ID2_UBS_SHIFT) & ID2_UBS_MASK;
1694                 smmu->va_size = arm_smmu_id_size_to_bits(size);
1695 #ifndef CONFIG_64BIT
1696                 smmu->va_size = min(32UL, smmu->va_size);
1697 #endif
1698                 size = 0;
1699                 if (id & ID2_PTFS_4K)
1700                         size |= SZ_4K | SZ_2M | SZ_1G;
1701                 if (id & ID2_PTFS_16K)
1702                         size |= SZ_16K | SZ_32M;
1703                 if (id & ID2_PTFS_64K)
1704                         size |= SZ_64K | SZ_512M;
1705         }
1706
1707         arm_smmu_ops.pgsize_bitmap &= size;
1708         dev_notice(smmu->dev, "\tSupported page sizes: 0x%08lx\n", size);
1709
1710         if (smmu->features & ARM_SMMU_FEAT_TRANS_S1)
1711                 dev_notice(smmu->dev, "\tStage-1: %lu-bit VA -> %lu-bit IPA\n",
1712                            smmu->va_size, smmu->ipa_size);
1713
1714         if (smmu->features & ARM_SMMU_FEAT_TRANS_S2)
1715                 dev_notice(smmu->dev, "\tStage-2: %lu-bit IPA -> %lu-bit PA\n",
1716                            smmu->ipa_size, smmu->pa_size);
1717
1718         return 0;
1719 }
1720
1721 static const struct of_device_id arm_smmu_of_match[] = {
1722         { .compatible = "arm,smmu-v1", .data = (void *)ARM_SMMU_V1 },
1723         { .compatible = "arm,smmu-v2", .data = (void *)ARM_SMMU_V2 },
1724         { .compatible = "arm,mmu-400", .data = (void *)ARM_SMMU_V1 },
1725         { .compatible = "arm,mmu-401", .data = (void *)ARM_SMMU_V1 },
1726         { .compatible = "arm,mmu-500", .data = (void *)ARM_SMMU_V2 },
1727         { },
1728 };
1729 MODULE_DEVICE_TABLE(of, arm_smmu_of_match);
1730
1731 static int arm_smmu_device_dt_probe(struct platform_device *pdev)
1732 {
1733         const struct of_device_id *of_id;
1734         struct resource *res;
1735         struct arm_smmu_device *smmu;
1736         struct device *dev = &pdev->dev;
1737         struct rb_node *node;
1738         struct of_phandle_args masterspec;
1739         int num_irqs, i, err;
1740
1741         smmu = devm_kzalloc(dev, sizeof(*smmu), GFP_KERNEL);
1742         if (!smmu) {
1743                 dev_err(dev, "failed to allocate arm_smmu_device\n");
1744                 return -ENOMEM;
1745         }
1746         smmu->dev = dev;
1747
1748         of_id = of_match_node(arm_smmu_of_match, dev->of_node);
1749         smmu->version = (enum arm_smmu_arch_version)of_id->data;
1750
1751         res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1752         smmu->base = devm_ioremap_resource(dev, res);
1753         if (IS_ERR(smmu->base))
1754                 return PTR_ERR(smmu->base);
1755         smmu->size = resource_size(res);
1756
1757         if (of_property_read_u32(dev->of_node, "#global-interrupts",
1758                                  &smmu->num_global_irqs)) {
1759                 dev_err(dev, "missing #global-interrupts property\n");
1760                 return -ENODEV;
1761         }
1762
1763         num_irqs = 0;
1764         while ((res = platform_get_resource(pdev, IORESOURCE_IRQ, num_irqs))) {
1765                 num_irqs++;
1766                 if (num_irqs > smmu->num_global_irqs)
1767                         smmu->num_context_irqs++;
1768         }
1769
1770         if (!smmu->num_context_irqs) {
1771                 dev_err(dev, "found %d interrupts but expected at least %d\n",
1772                         num_irqs, smmu->num_global_irqs + 1);
1773                 return -ENODEV;
1774         }
1775
1776         smmu->irqs = devm_kzalloc(dev, sizeof(*smmu->irqs) * num_irqs,
1777                                   GFP_KERNEL);
1778         if (!smmu->irqs) {
1779                 dev_err(dev, "failed to allocate %d irqs\n", num_irqs);
1780                 return -ENOMEM;
1781         }
1782
1783         for (i = 0; i < num_irqs; ++i) {
1784                 int irq = platform_get_irq(pdev, i);
1785
1786                 if (irq < 0) {
1787                         dev_err(dev, "failed to get irq index %d\n", i);
1788                         return -ENODEV;
1789                 }
1790                 smmu->irqs[i] = irq;
1791         }
1792
1793         err = arm_smmu_device_cfg_probe(smmu);
1794         if (err)
1795                 return err;
1796
1797         i = 0;
1798         smmu->masters = RB_ROOT;
1799         while (!of_parse_phandle_with_args(dev->of_node, "mmu-masters",
1800                                            "#stream-id-cells", i,
1801                                            &masterspec)) {
1802                 err = register_smmu_master(smmu, dev, &masterspec);
1803                 if (err) {
1804                         dev_err(dev, "failed to add master %s\n",
1805                                 masterspec.np->name);
1806                         goto out_put_masters;
1807                 }
1808
1809                 i++;
1810         }
1811         dev_notice(dev, "registered %d master devices\n", i);
1812
1813         parse_driver_options(smmu);
1814
1815         if (smmu->version > ARM_SMMU_V1 &&
1816             smmu->num_context_banks != smmu->num_context_irqs) {
1817                 dev_err(dev,
1818                         "found only %d context interrupt(s) but %d required\n",
1819                         smmu->num_context_irqs, smmu->num_context_banks);
1820                 err = -ENODEV;
1821                 goto out_put_masters;
1822         }
1823
1824         for (i = 0; i < smmu->num_global_irqs; ++i) {
1825                 err = request_irq(smmu->irqs[i],
1826                                   arm_smmu_global_fault,
1827                                   IRQF_SHARED,
1828                                   "arm-smmu global fault",
1829                                   smmu);
1830                 if (err) {
1831                         dev_err(dev, "failed to request global IRQ %d (%u)\n",
1832                                 i, smmu->irqs[i]);
1833                         goto out_free_irqs;
1834                 }
1835         }
1836
1837         INIT_LIST_HEAD(&smmu->list);
1838         spin_lock(&arm_smmu_devices_lock);
1839         list_add(&smmu->list, &arm_smmu_devices);
1840         spin_unlock(&arm_smmu_devices_lock);
1841
1842         arm_smmu_device_reset(smmu);
1843         return 0;
1844
1845 out_free_irqs:
1846         while (i--)
1847                 free_irq(smmu->irqs[i], smmu);
1848
1849 out_put_masters:
1850         for (node = rb_first(&smmu->masters); node; node = rb_next(node)) {
1851                 struct arm_smmu_master *master
1852                         = container_of(node, struct arm_smmu_master, node);
1853                 of_node_put(master->of_node);
1854         }
1855
1856         return err;
1857 }
1858
1859 static int arm_smmu_device_remove(struct platform_device *pdev)
1860 {
1861         int i;
1862         struct device *dev = &pdev->dev;
1863         struct arm_smmu_device *curr, *smmu = NULL;
1864         struct rb_node *node;
1865
1866         spin_lock(&arm_smmu_devices_lock);
1867         list_for_each_entry(curr, &arm_smmu_devices, list) {
1868                 if (curr->dev == dev) {
1869                         smmu = curr;
1870                         list_del(&smmu->list);
1871                         break;
1872                 }
1873         }
1874         spin_unlock(&arm_smmu_devices_lock);
1875
1876         if (!smmu)
1877                 return -ENODEV;
1878
1879         for (node = rb_first(&smmu->masters); node; node = rb_next(node)) {
1880                 struct arm_smmu_master *master
1881                         = container_of(node, struct arm_smmu_master, node);
1882                 of_node_put(master->of_node);
1883         }
1884
1885         if (!bitmap_empty(smmu->context_map, ARM_SMMU_MAX_CBS))
1886                 dev_err(dev, "removing device with active domains!\n");
1887
1888         for (i = 0; i < smmu->num_global_irqs; ++i)
1889                 free_irq(smmu->irqs[i], smmu);
1890
1891         /* Turn the thing off */
1892         writel(sCR0_CLIENTPD, ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sCR0);
1893         return 0;
1894 }
1895
1896 static struct platform_driver arm_smmu_driver = {
1897         .driver = {
1898                 .name           = "arm-smmu",
1899                 .of_match_table = of_match_ptr(arm_smmu_of_match),
1900         },
1901         .probe  = arm_smmu_device_dt_probe,
1902         .remove = arm_smmu_device_remove,
1903 };
1904
1905 static int __init arm_smmu_init(void)
1906 {
1907         struct device_node *np;
1908         int ret;
1909
1910         /*
1911          * Play nice with systems that don't have an ARM SMMU by checking that
1912          * an ARM SMMU exists in the system before proceeding with the driver
1913          * and IOMMU bus operation registration.
1914          */
1915         np = of_find_matching_node(NULL, arm_smmu_of_match);
1916         if (!np)
1917                 return 0;
1918
1919         of_node_put(np);
1920
1921         ret = platform_driver_register(&arm_smmu_driver);
1922         if (ret)
1923                 return ret;
1924
1925         /* Oh, for a proper bus abstraction */
1926         if (!iommu_present(&platform_bus_type))
1927                 bus_set_iommu(&platform_bus_type, &arm_smmu_ops);
1928
1929 #ifdef CONFIG_ARM_AMBA
1930         if (!iommu_present(&amba_bustype))
1931                 bus_set_iommu(&amba_bustype, &arm_smmu_ops);
1932 #endif
1933
1934 #ifdef CONFIG_PCI
1935         if (!iommu_present(&pci_bus_type))
1936                 bus_set_iommu(&pci_bus_type, &arm_smmu_ops);
1937 #endif
1938
1939         return 0;
1940 }
1941
1942 static void __exit arm_smmu_exit(void)
1943 {
1944         return platform_driver_unregister(&arm_smmu_driver);
1945 }
1946
1947 subsys_initcall(arm_smmu_init);
1948 module_exit(arm_smmu_exit);
1949
1950 MODULE_DESCRIPTION("IOMMU API for ARM architected SMMU implementations");
1951 MODULE_AUTHOR("Will Deacon <will.deacon@arm.com>");
1952 MODULE_LICENSE("GPL v2");