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         /* We're bypassing these SIDs, so don't allocate an actual context */
 830         if (domain->type == IOMMU_DOMAIN_DMA) {
 831                 smmu_domain->smmu = smmu;
 832                 goto out_unlock;
 833         }
 834
 835         /*
 836          * Mapping the requested stage onto what we support is surprisingly
 837          * complicated, mainly because the spec allows S1+S2 SMMUs without
 838          * support for nested translation. That means we end up with the
 839          * following table:
 840          *
 841          * Requested        Supported        Actual
 842          *     S1               N              S1
 843          *     S1             S1+S2            S1
 844          *     S1               S2             S2
 845          *     S1               S1             S1
 846          *     N                N              N
 847          *     N              S1+S2            S2
 848          *     N                S2             S2
 849          *     N                S1             S1
 850          *
 851          * Note that you can't actually request stage-2 mappings.
 852          */
 853         if (!(smmu->features & ARM_SMMU_FEAT_TRANS_S1))
 854                 smmu_domain->stage = ARM_SMMU_DOMAIN_S2;
 855         if (!(smmu->features & ARM_SMMU_FEAT_TRANS_S2))
 856                 smmu_domain->stage = ARM_SMMU_DOMAIN_S1;
 857
 858         switch (smmu_domain->stage) {
 859         case ARM_SMMU_DOMAIN_S1:
 860                 cfg->cbar = CBAR_TYPE_S1_TRANS_S2_BYPASS;
 861                 start = smmu->num_s2_context_banks;
 862                 ias = smmu->va_size;
 863                 oas = smmu->ipa_size;
 864                 if (IS_ENABLED(CONFIG_64BIT))
 865                         fmt = ARM_64_LPAE_S1;
 866                 else
 867                         fmt = ARM_32_LPAE_S1;
 868                 break;
 869         case ARM_SMMU_DOMAIN_NESTED:
 870                 /*
 871                  * We will likely want to change this if/when KVM gets
 872                  * involved.
 873                  */
 874         case ARM_SMMU_DOMAIN_S2:
 875                 cfg->cbar = CBAR_TYPE_S2_TRANS;
 876                 start = 0;
 877                 ias = smmu->ipa_size;
 878                 oas = smmu->pa_size;
 879                 if (IS_ENABLED(CONFIG_64BIT))
 880                         fmt = ARM_64_LPAE_S2;
 881                 else
 882                         fmt = ARM_32_LPAE_S2;
 883                 break;
 884         default:
 885                 ret = -EINVAL;
 886                 goto out_unlock;
 887         }
 888
 889         ret = __arm_smmu_alloc_bitmap(smmu->context_map, start,
 890                                       smmu->num_context_banks);
 891         if (IS_ERR_VALUE(ret))
 892                 goto out_unlock;
 893
 894         cfg->cbndx = ret;
 895         if (smmu->version == ARM_SMMU_V1) {
 896                 cfg->irptndx = atomic_inc_return(&smmu->irptndx);
 897                 cfg->irptndx %= smmu->num_context_irqs;
 898         } else {
 899                 cfg->irptndx = cfg->cbndx;
 900         }
 901
 902         pgtbl_cfg = (struct io_pgtable_cfg) {
 903                 .pgsize_bitmap  = arm_smmu_ops.pgsize_bitmap,
 904                 .ias            = ias,
 905                 .oas            = oas,
 906                 .tlb            = &arm_smmu_gather_ops,
 907                 .iommu_dev      = smmu->dev,
 908         };
 909
 910         smmu_domain->smmu = smmu;
 911         pgtbl_ops = alloc_io_pgtable_ops(fmt, &pgtbl_cfg, smmu_domain);
 912         if (!pgtbl_ops) {
 913                 ret = -ENOMEM;
 914                 goto out_clear_smmu;
 915         }
 916
 917         /* Update our support page sizes to reflect the page table format */
 918         arm_smmu_ops.pgsize_bitmap = pgtbl_cfg.pgsize_bitmap;
 919
 920         /* Initialise the context bank with our page table cfg */
 921         arm_smmu_init_context_bank(smmu_domain, &pgtbl_cfg);
 922
 923         /*
 924          * Request context fault interrupt. Do this last to avoid the
 925          * handler seeing a half-initialised domain state.
 926          */
 927         irq = smmu->irqs[smmu->num_global_irqs + cfg->irptndx];
 928         ret = request_irq(irq, arm_smmu_context_fault, IRQF_SHARED,
 929                           "arm-smmu-context-fault", domain);
 930         if (IS_ERR_VALUE(ret)) {
 931                 dev_err(smmu->dev, "failed to request context IRQ %d (%u)\n",
 932                         cfg->irptndx, irq);
 933                 cfg->irptndx = INVALID_IRPTNDX;
 934         }
 935
 936         mutex_unlock(&smmu_domain->init_mutex);
 937
 938         /* Publish page table ops for map/unmap */
 939         smmu_domain->pgtbl_ops = pgtbl_ops;
 940         return 0;
 941
 942 out_clear_smmu:
 943         smmu_domain->smmu = NULL;
 944 out_unlock:
 945         mutex_unlock(&smmu_domain->init_mutex);
 946         return ret;
 947 }
 948
 949 static void arm_smmu_destroy_domain_context(struct iommu_domain *domain)
 950 {
 951         struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
 952         struct arm_smmu_device *smmu = smmu_domain->smmu;
 953         struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
 954         void __iomem *cb_base;
 955         int irq;
 956
 957         if (!smmu || domain->type == IOMMU_DOMAIN_DMA)
 958                 return;
 959
 960         /*
 961          * Disable the context bank and free the page tables before freeing
 962          * it.
 963          */
 964         cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
 965         writel_relaxed(0, cb_base + ARM_SMMU_CB_SCTLR);
 966
 967         if (cfg->irptndx != INVALID_IRPTNDX) {
 968                 irq = smmu->irqs[smmu->num_global_irqs + cfg->irptndx];
 969                 free_irq(irq, domain);
 970         }
 971
 972         free_io_pgtable_ops(smmu_domain->pgtbl_ops);
 973         __arm_smmu_free_bitmap(smmu->context_map, cfg->cbndx);
 974 }
 975
 976 static struct iommu_domain *arm_smmu_domain_alloc(unsigned type)
 977 {
 978         struct arm_smmu_domain *smmu_domain;
 979
 980         if (type != IOMMU_DOMAIN_UNMANAGED && type != IOMMU_DOMAIN_DMA)
 981                 return NULL;
 982         /*
 983          * Allocate the domain and initialise some of its data structures.
 984          * We can't really do anything meaningful until we've added a
 985          * master.
 986          */
 987         smmu_domain = kzalloc(sizeof(*smmu_domain), GFP_KERNEL);
 988         if (!smmu_domain)
 989                 return NULL;
 990
 991         if (type == IOMMU_DOMAIN_DMA &&
 992             iommu_get_dma_cookie(&smmu_domain->domain)) {
 993                 kfree(smmu_domain);
 994                 return NULL;
 995         }
 996
 997         mutex_init(&smmu_domain->init_mutex);
 998         spin_lock_init(&smmu_domain->pgtbl_lock);
 999
1000         return &smmu_domain->domain;
1001 }
1002
1003 static void arm_smmu_domain_free(struct iommu_domain *domain)
1004 {
1005         struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
1006
1007         /*
1008          * Free the domain resources. We assume that all devices have
1009          * already been detached.
1010          */
1011         iommu_put_dma_cookie(domain);
1012         arm_smmu_destroy_domain_context(domain);
1013         kfree(smmu_domain);
1014 }
1015
1016 static int arm_smmu_master_configure_smrs(struct arm_smmu_device *smmu,
1017                                           struct arm_smmu_master_cfg *cfg)
1018 {
1019         int i;
1020         struct arm_smmu_smr *smrs;
1021         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1022
1023         if (!(smmu->features & ARM_SMMU_FEAT_STREAM_MATCH))
1024                 return 0;
1025
1026         if (cfg->smrs)
1027                 return -EEXIST;
1028
1029         smrs = kmalloc_array(cfg->num_streamids, sizeof(*smrs), GFP_KERNEL);
1030         if (!smrs) {
1031                 dev_err(smmu->dev, "failed to allocate %d SMRs\n",
1032                         cfg->num_streamids);
1033                 return -ENOMEM;
1034         }
1035
1036         /* Allocate the SMRs on the SMMU */
1037         for (i = 0; i < cfg->num_streamids; ++i) {
1038                 int idx = __arm_smmu_alloc_bitmap(smmu->smr_map, 0,
1039                                                   smmu->num_mapping_groups);
1040                 if (IS_ERR_VALUE(idx)) {
1041                         dev_err(smmu->dev, "failed to allocate free SMR\n");
1042                         goto err_free_smrs;
1043                 }
1044
1045                 smrs[i] = (struct arm_smmu_smr) {
1046                         .idx    = idx,
1047                         .mask   = 0, /* We don't currently share SMRs */
1048                         .id     = cfg->streamids[i],
1049                 };
1050         }
1051
1052         /* It worked! Now, poke the actual hardware */
1053         for (i = 0; i < cfg->num_streamids; ++i) {
1054                 u32 reg = SMR_VALID | smrs[i].id << SMR_ID_SHIFT |
1055                           smrs[i].mask << SMR_MASK_SHIFT;
1056                 writel_relaxed(reg, gr0_base + ARM_SMMU_GR0_SMR(smrs[i].idx));
1057         }
1058
1059         cfg->smrs = smrs;
1060         return 0;
1061
1062 err_free_smrs:
1063         while (--i >= 0)
1064                 __arm_smmu_free_bitmap(smmu->smr_map, smrs[i].idx);
1065         kfree(smrs);
1066         return -ENOSPC;
1067 }
1068
1069 static void arm_smmu_master_free_smrs(struct arm_smmu_device *smmu,
1070                                       struct arm_smmu_master_cfg *cfg)
1071 {
1072         int i;
1073         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1074         struct arm_smmu_smr *smrs = cfg->smrs;
1075
1076         if (!smrs)
1077                 return;
1078
1079         /* Invalidate the SMRs before freeing back to the allocator */
1080         for (i = 0; i < cfg->num_streamids; ++i) {
1081                 u8 idx = smrs[i].idx;
1082
1083                 writel_relaxed(~SMR_VALID, gr0_base + ARM_SMMU_GR0_SMR(idx));
1084                 __arm_smmu_free_bitmap(smmu->smr_map, idx);
1085         }
1086
1087         cfg->smrs = NULL;
1088         kfree(smrs);
1089 }
1090
1091 static int arm_smmu_domain_add_master(struct arm_smmu_domain *smmu_domain,
1092                                       struct arm_smmu_master_cfg *cfg)
1093 {
1094         int i, ret;
1095         struct arm_smmu_device *smmu = smmu_domain->smmu;
1096         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1097
1098         /*
1099          * FIXME: This won't be needed once we have IOMMU-backed DMA ops
1100          * for all devices behind the SMMU. Note that we need to take
1101          * care configuring SMRs for devices both a platform_device and
1102          * and a PCI device (i.e. a PCI host controller)
1103          */
1104         if (smmu_domain->domain.type == IOMMU_DOMAIN_DMA)
1105                 return 0;
1106
1107         /* Devices in an IOMMU group may already be configured */
1108         ret = arm_smmu_master_configure_smrs(smmu, cfg);
1109         if (ret)
1110                 return ret == -EEXIST ? 0 : ret;
1111
1112         for (i = 0; i < cfg->num_streamids; ++i) {
1113                 u32 idx, s2cr;
1114
1115                 idx = cfg->smrs ? cfg->smrs[i].idx : cfg->streamids[i];
1116                 s2cr = S2CR_TYPE_TRANS | S2CR_PRIVCFG_UNPRIV |
1117                        (smmu_domain->cfg.cbndx << S2CR_CBNDX_SHIFT);
1118                 writel_relaxed(s2cr, gr0_base + ARM_SMMU_GR0_S2CR(idx));
1119         }
1120
1121         return 0;
1122 }
1123
1124 static void arm_smmu_domain_remove_master(struct arm_smmu_domain *smmu_domain,
1125                                           struct arm_smmu_master_cfg *cfg)
1126 {
1127         int i;
1128         struct arm_smmu_device *smmu = smmu_domain->smmu;
1129         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1130
1131         /* An IOMMU group is torn down by the first device to be removed */
1132         if ((smmu->features & ARM_SMMU_FEAT_STREAM_MATCH) && !cfg->smrs)
1133                 return;
1134
1135         /*
1136          * We *must* clear the S2CR first, because freeing the SMR means
1137          * that it can be re-allocated immediately.
1138          */
1139         for (i = 0; i < cfg->num_streamids; ++i) {
1140                 u32 idx = cfg->smrs ? cfg->smrs[i].idx : cfg->streamids[i];
1141                 u32 reg = disable_bypass ? S2CR_TYPE_FAULT : S2CR_TYPE_BYPASS;
1142
1143                 writel_relaxed(reg, gr0_base + ARM_SMMU_GR0_S2CR(idx));
1144         }
1145
1146         arm_smmu_master_free_smrs(smmu, cfg);
1147 }
1148
1149 static void arm_smmu_detach_dev(struct device *dev,
1150                                 struct arm_smmu_master_cfg *cfg)
1151 {
1152         struct iommu_domain *domain = dev->archdata.iommu;
1153         struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
1154
1155         dev->archdata.iommu = NULL;
1156         arm_smmu_domain_remove_master(smmu_domain, cfg);
1157 }
1158
1159 static int arm_smmu_attach_dev(struct iommu_domain *domain, struct device *dev)
1160 {
1161         int ret;
1162         struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
1163         struct arm_smmu_device *smmu;
1164         struct arm_smmu_master_cfg *cfg;
1165
1166         smmu = find_smmu_for_device(dev);
1167         if (!smmu) {
1168                 dev_err(dev, "cannot attach to SMMU, is it on the same bus?\n");
1169                 return -ENXIO;
1170         }
1171
1172         /* Ensure that the domain is finalised */
1173         ret = arm_smmu_init_domain_context(domain, smmu);
1174         if (IS_ERR_VALUE(ret))
1175                 return ret;
1176
1177         /*
1178          * Sanity check the domain. We don't support domains across
1179          * different SMMUs.
1180          */
1181         if (smmu_domain->smmu != smmu) {
1182                 dev_err(dev,
1183                         "cannot attach to SMMU %s whilst already attached to domain on SMMU %s\n",
1184                         dev_name(smmu_domain->smmu->dev), dev_name(smmu->dev));
1185                 return -EINVAL;
1186         }
1187
1188         /* Looks ok, so add the device to the domain */
1189         cfg = find_smmu_master_cfg(dev);
1190         if (!cfg)
1191                 return -ENODEV;
1192
1193         /* Detach the dev from its current domain */
1194         if (dev->archdata.iommu)
1195                 arm_smmu_detach_dev(dev, cfg);
1196
1197         ret = arm_smmu_domain_add_master(smmu_domain, cfg);
1198         if (!ret)
1199                 dev->archdata.iommu = domain;
1200         return ret;
1201 }
1202
1203 static int arm_smmu_map(struct iommu_domain *domain, unsigned long iova,
1204                         phys_addr_t paddr, size_t size, int prot)
1205 {
1206         int ret;
1207         unsigned long flags;
1208         struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
1209         struct io_pgtable_ops *ops= smmu_domain->pgtbl_ops;
1210
1211         if (!ops)
1212                 return -ENODEV;
1213
1214         spin_lock_irqsave(&smmu_domain->pgtbl_lock, flags);
1215         ret = ops->map(ops, iova, paddr, size, prot);
1216         spin_unlock_irqrestore(&smmu_domain->pgtbl_lock, flags);
1217         return ret;
1218 }
1219
1220 static size_t arm_smmu_unmap(struct iommu_domain *domain, unsigned long iova,
1221                              size_t size)
1222 {
1223         size_t ret;
1224         unsigned long flags;
1225         struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
1226         struct io_pgtable_ops *ops= smmu_domain->pgtbl_ops;
1227
1228         if (!ops)
1229                 return 0;
1230
1231         spin_lock_irqsave(&smmu_domain->pgtbl_lock, flags);
1232         ret = ops->unmap(ops, iova, size);
1233         spin_unlock_irqrestore(&smmu_domain->pgtbl_lock, flags);
1234         return ret;
1235 }
1236
1237 static phys_addr_t arm_smmu_iova_to_phys_hard(struct iommu_domain *domain,
1238                                               dma_addr_t iova)
1239 {
1240         struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
1241         struct arm_smmu_device *smmu = smmu_domain->smmu;
1242         struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
1243         struct io_pgtable_ops *ops= smmu_domain->pgtbl_ops;
1244         struct device *dev = smmu->dev;
1245         void __iomem *cb_base;
1246         u32 tmp;
1247         u64 phys;
1248         unsigned long va;
1249
1250         cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, cfg->cbndx);
1251
1252         /* ATS1 registers can only be written atomically */
1253         va = iova & ~0xfffUL;
1254         if (smmu->version == ARM_SMMU_V2)
1255                 smmu_writeq(va, cb_base + ARM_SMMU_CB_ATS1PR);
1256         else
1257                 writel_relaxed(va, cb_base + ARM_SMMU_CB_ATS1PR);
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 %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 = to_smmu_domain(domain);
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_init_pci_device(struct pci_dev *pdev,
1333                                     struct iommu_group *group)
1334 {
1335         struct arm_smmu_master_cfg *cfg;
1336         u16 sid;
1337         int i;
1338
1339         cfg = iommu_group_get_iommudata(group);
1340         if (!cfg) {
1341                 cfg = kzalloc(sizeof(*cfg), GFP_KERNEL);
1342                 if (!cfg)
1343                         return -ENOMEM;
1344
1345                 iommu_group_set_iommudata(group, cfg,
1346                                           __arm_smmu_release_pci_iommudata);
1347         }
1348
1349         if (cfg->num_streamids >= MAX_MASTER_STREAMIDS)
1350                 return -ENOSPC;
1351
1352         /*
1353          * Assume Stream ID == Requester ID for now.
1354          * We need a way to describe the ID mappings in FDT.
1355          */
1356         pci_for_each_dma_alias(pdev, __arm_smmu_get_pci_sid, &sid);
1357         for (i = 0; i < cfg->num_streamids; ++i)
1358                 if (cfg->streamids[i] == sid)
1359                         break;
1360
1361         /* Avoid duplicate SIDs, as this can lead to SMR conflicts */
1362         if (i == cfg->num_streamids)
1363                 cfg->streamids[cfg->num_streamids++] = sid;
1364
1365         return 0;
1366 }
1367
1368 static int arm_smmu_init_platform_device(struct device *dev,
1369                                          struct iommu_group *group)
1370 {
1371         struct arm_smmu_device *smmu = find_smmu_for_device(dev);
1372         struct arm_smmu_master *master;
1373
1374         if (!smmu)
1375                 return -ENODEV;
1376
1377         master = find_smmu_master(smmu, dev->of_node);
1378         if (!master)
1379                 return -ENODEV;
1380
1381         iommu_group_set_iommudata(group, &master->cfg, NULL);
1382
1383         return 0;
1384 }
1385
1386 static int arm_smmu_add_device(struct device *dev)
1387 {
1388         struct iommu_group *group;
1389
1390         group = iommu_group_get_for_dev(dev);
1391         if (IS_ERR(group))
1392                 return PTR_ERR(group);
1393
1394         iommu_group_put(group);
1395         return 0;
1396 }
1397
1398 static void arm_smmu_remove_device(struct device *dev)
1399 {
1400         iommu_group_remove_device(dev);
1401 }
1402
1403 static struct iommu_group *arm_smmu_device_group(struct device *dev)
1404 {
1405         struct iommu_group *group;
1406         int ret;
1407
1408         if (dev_is_pci(dev))
1409                 group = pci_device_group(dev);
1410         else
1411                 group = generic_device_group(dev);
1412
1413         if (IS_ERR(group))
1414                 return group;
1415
1416         if (dev_is_pci(dev))
1417                 ret = arm_smmu_init_pci_device(to_pci_dev(dev), group);
1418         else
1419                 ret = arm_smmu_init_platform_device(dev, group);
1420
1421         if (ret) {
1422                 iommu_group_put(group);
1423                 group = ERR_PTR(ret);
1424         }
1425
1426         return group;
1427 }
1428
1429 static int arm_smmu_domain_get_attr(struct iommu_domain *domain,
1430                                     enum iommu_attr attr, void *data)
1431 {
1432         struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
1433
1434         switch (attr) {
1435         case DOMAIN_ATTR_NESTING:
1436                 *(int *)data = (smmu_domain->stage == ARM_SMMU_DOMAIN_NESTED);
1437                 return 0;
1438         default:
1439                 return -ENODEV;
1440         }
1441 }
1442
1443 static int arm_smmu_domain_set_attr(struct iommu_domain *domain,
1444                                     enum iommu_attr attr, void *data)
1445 {
1446         int ret = 0;
1447         struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
1448
1449         mutex_lock(&smmu_domain->init_mutex);
1450
1451         switch (attr) {
1452         case DOMAIN_ATTR_NESTING:
1453                 if (smmu_domain->smmu) {
1454                         ret = -EPERM;
1455                         goto out_unlock;
1456                 }
1457
1458                 if (*(int *)data)
1459                         smmu_domain->stage = ARM_SMMU_DOMAIN_NESTED;
1460                 else
1461                         smmu_domain->stage = ARM_SMMU_DOMAIN_S1;
1462
1463                 break;
1464         default:
1465                 ret = -ENODEV;
1466         }
1467
1468 out_unlock:
1469         mutex_unlock(&smmu_domain->init_mutex);
1470         return ret;
1471 }
1472
1473 static struct iommu_ops arm_smmu_ops = {
1474         .capable                = arm_smmu_capable,
1475         .domain_alloc           = arm_smmu_domain_alloc,
1476         .domain_free            = arm_smmu_domain_free,
1477         .attach_dev             = arm_smmu_attach_dev,
1478         .map                    = arm_smmu_map,
1479         .unmap                  = arm_smmu_unmap,
1480         .map_sg                 = default_iommu_map_sg,
1481         .iova_to_phys           = arm_smmu_iova_to_phys,
1482         .add_device             = arm_smmu_add_device,
1483         .remove_device          = arm_smmu_remove_device,
1484         .device_group           = arm_smmu_device_group,
1485         .domain_get_attr        = arm_smmu_domain_get_attr,
1486         .domain_set_attr        = arm_smmu_domain_set_attr,
1487         .pgsize_bitmap          = -1UL, /* Restricted during device attach */
1488 };
1489
1490 static void arm_smmu_device_reset(struct arm_smmu_device *smmu)
1491 {
1492         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1493         void __iomem *cb_base;
1494         int i = 0;
1495         u32 reg;
1496
1497         /* clear global FSR */
1498         reg = readl_relaxed(ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sGFSR);
1499         writel(reg, ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sGFSR);
1500
1501         /* Mark all SMRn as invalid and all S2CRn as bypass unless overridden */
1502         reg = disable_bypass ? S2CR_TYPE_FAULT : S2CR_TYPE_BYPASS;
1503         for (i = 0; i < smmu->num_mapping_groups; ++i) {
1504                 writel_relaxed(0, gr0_base + ARM_SMMU_GR0_SMR(i));
1505                 writel_relaxed(reg, gr0_base + ARM_SMMU_GR0_S2CR(i));
1506         }
1507
1508         /* Make sure all context banks are disabled and clear CB_FSR  */
1509         for (i = 0; i < smmu->num_context_banks; ++i) {
1510                 cb_base = ARM_SMMU_CB_BASE(smmu) + ARM_SMMU_CB(smmu, i);
1511                 writel_relaxed(0, cb_base + ARM_SMMU_CB_SCTLR);
1512                 writel_relaxed(FSR_FAULT, cb_base + ARM_SMMU_CB_FSR);
1513         }
1514
1515         /* Invalidate the TLB, just in case */
1516         writel_relaxed(0, gr0_base + ARM_SMMU_GR0_TLBIALLH);
1517         writel_relaxed(0, gr0_base + ARM_SMMU_GR0_TLBIALLNSNH);
1518
1519         reg = readl_relaxed(ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sCR0);
1520
1521         /* Enable fault reporting */
1522         reg |= (sCR0_GFRE | sCR0_GFIE | sCR0_GCFGFRE | sCR0_GCFGFIE);
1523
1524         /* Disable TLB broadcasting. */
1525         reg |= (sCR0_VMIDPNE | sCR0_PTM);
1526
1527         /* Enable client access, handling unmatched streams as appropriate */
1528         reg &= ~sCR0_CLIENTPD;
1529         if (disable_bypass)
1530                 reg |= sCR0_USFCFG;
1531         else
1532                 reg &= ~sCR0_USFCFG;
1533
1534         /* Disable forced broadcasting */
1535         reg &= ~sCR0_FB;
1536
1537         /* Don't upgrade barriers */
1538         reg &= ~(sCR0_BSU_MASK << sCR0_BSU_SHIFT);
1539
1540         /* Push the button */
1541         __arm_smmu_tlb_sync(smmu);
1542         writel(reg, ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sCR0);
1543 }
1544
1545 static int arm_smmu_id_size_to_bits(int size)
1546 {
1547         switch (size) {
1548         case 0:
1549                 return 32;
1550         case 1:
1551                 return 36;
1552         case 2:
1553                 return 40;
1554         case 3:
1555                 return 42;
1556         case 4:
1557                 return 44;
1558         case 5:
1559         default:
1560                 return 48;
1561         }
1562 }
1563
1564 static int arm_smmu_device_cfg_probe(struct arm_smmu_device *smmu)
1565 {
1566         unsigned long size;
1567         void __iomem *gr0_base = ARM_SMMU_GR0(smmu);
1568         u32 id;
1569         bool cttw_dt, cttw_reg;
1570
1571         dev_notice(smmu->dev, "probing hardware configuration...\n");
1572         dev_notice(smmu->dev, "SMMUv%d with:\n", smmu->version);
1573
1574         /* ID0 */
1575         id = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID0);
1576
1577         /* Restrict available stages based on module parameter */
1578         if (force_stage == 1)
1579                 id &= ~(ID0_S2TS | ID0_NTS);
1580         else if (force_stage == 2)
1581                 id &= ~(ID0_S1TS | ID0_NTS);
1582
1583         if (id & ID0_S1TS) {
1584                 smmu->features |= ARM_SMMU_FEAT_TRANS_S1;
1585                 dev_notice(smmu->dev, "\tstage 1 translation\n");
1586         }
1587
1588         if (id & ID0_S2TS) {
1589                 smmu->features |= ARM_SMMU_FEAT_TRANS_S2;
1590                 dev_notice(smmu->dev, "\tstage 2 translation\n");
1591         }
1592
1593         if (id & ID0_NTS) {
1594                 smmu->features |= ARM_SMMU_FEAT_TRANS_NESTED;
1595                 dev_notice(smmu->dev, "\tnested translation\n");
1596         }
1597
1598         if (!(smmu->features &
1599                 (ARM_SMMU_FEAT_TRANS_S1 | ARM_SMMU_FEAT_TRANS_S2))) {
1600                 dev_err(smmu->dev, "\tno translation support!\n");
1601                 return -ENODEV;
1602         }
1603
1604         if ((id & ID0_S1TS) && ((smmu->version == 1) || !(id & ID0_ATOSNS))) {
1605                 smmu->features |= ARM_SMMU_FEAT_TRANS_OPS;
1606                 dev_notice(smmu->dev, "\taddress translation ops\n");
1607         }
1608
1609         /*
1610          * In order for DMA API calls to work properly, we must defer to what
1611          * the DT says about coherency, regardless of what the hardware claims.
1612          * Fortunately, this also opens up a workaround for systems where the
1613          * ID register value has ended up configured incorrectly.
1614          */
1615         cttw_dt = of_dma_is_coherent(smmu->dev->of_node);
1616         cttw_reg = !!(id & ID0_CTTW);
1617         if (cttw_dt)
1618                 smmu->features |= ARM_SMMU_FEAT_COHERENT_WALK;
1619         if (cttw_dt || cttw_reg)
1620                 dev_notice(smmu->dev, "\t%scoherent table walk\n",
1621                            cttw_dt ? "" : "non-");
1622         if (cttw_dt != cttw_reg)
1623                 dev_notice(smmu->dev,
1624                            "\t(IDR0.CTTW overridden by dma-coherent property)\n");
1625
1626         if (id & ID0_SMS) {
1627                 u32 smr, sid, mask;
1628
1629                 smmu->features |= ARM_SMMU_FEAT_STREAM_MATCH;
1630                 smmu->num_mapping_groups = (id >> ID0_NUMSMRG_SHIFT) &
1631                                            ID0_NUMSMRG_MASK;
1632                 if (smmu->num_mapping_groups == 0) {
1633                         dev_err(smmu->dev,
1634                                 "stream-matching supported, but no SMRs present!\n");
1635                         return -ENODEV;
1636                 }
1637
1638                 smr = SMR_MASK_MASK << SMR_MASK_SHIFT;
1639                 smr |= (SMR_ID_MASK << SMR_ID_SHIFT);
1640                 writel_relaxed(smr, gr0_base + ARM_SMMU_GR0_SMR(0));
1641                 smr = readl_relaxed(gr0_base + ARM_SMMU_GR0_SMR(0));
1642
1643                 mask = (smr >> SMR_MASK_SHIFT) & SMR_MASK_MASK;
1644                 sid = (smr >> SMR_ID_SHIFT) & SMR_ID_MASK;
1645                 if ((mask & sid) != sid) {
1646                         dev_err(smmu->dev,
1647                                 "SMR mask bits (0x%x) insufficient for ID field (0x%x)\n",
1648                                 mask, sid);
1649                         return -ENODEV;
1650                 }
1651
1652                 dev_notice(smmu->dev,
1653                            "\tstream matching with %u register groups, mask 0x%x",
1654                            smmu->num_mapping_groups, mask);
1655         } else {
1656                 smmu->num_mapping_groups = (id >> ID0_NUMSIDB_SHIFT) &
1657                                            ID0_NUMSIDB_MASK;
1658         }
1659
1660         /* ID1 */
1661         id = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID1);
1662         smmu->pgshift = (id & ID1_PAGESIZE) ? 16 : 12;
1663
1664         /* Check for size mismatch of SMMU address space from mapped region */
1665         size = 1 << (((id >> ID1_NUMPAGENDXB_SHIFT) & ID1_NUMPAGENDXB_MASK) + 1);
1666         size *= 2 << smmu->pgshift;
1667         if (smmu->size != size)
1668                 dev_warn(smmu->dev,
1669                         "SMMU address space size (0x%lx) differs from mapped region size (0x%lx)!\n",
1670                         size, smmu->size);
1671
1672         smmu->num_s2_context_banks = (id >> ID1_NUMS2CB_SHIFT) & ID1_NUMS2CB_MASK;
1673         smmu->num_context_banks = (id >> ID1_NUMCB_SHIFT) & ID1_NUMCB_MASK;
1674         if (smmu->num_s2_context_banks > smmu->num_context_banks) {
1675                 dev_err(smmu->dev, "impossible number of S2 context banks!\n");
1676                 return -ENODEV;
1677         }
1678         dev_notice(smmu->dev, "\t%u context banks (%u stage-2 only)\n",
1679                    smmu->num_context_banks, smmu->num_s2_context_banks);
1680
1681         /* ID2 */
1682         id = readl_relaxed(gr0_base + ARM_SMMU_GR0_ID2);
1683         size = arm_smmu_id_size_to_bits((id >> ID2_IAS_SHIFT) & ID2_IAS_MASK);
1684         smmu->ipa_size = size;
1685
1686         /* The output mask is also applied for bypass */
1687         size = arm_smmu_id_size_to_bits((id >> ID2_OAS_SHIFT) & ID2_OAS_MASK);
1688         smmu->pa_size = size;
1689
1690         /*
1691          * What the page table walker can address actually depends on which
1692          * descriptor format is in use, but since a) we don't know that yet,
1693          * and b) it can vary per context bank, this will have to do...
1694          */
1695         if (dma_set_mask_and_coherent(smmu->dev, DMA_BIT_MASK(size)))
1696                 dev_warn(smmu->dev,
1697                          "failed to set DMA mask for table walker\n");
1698
1699         if (smmu->version == ARM_SMMU_V1) {
1700                 smmu->va_size = smmu->ipa_size;
1701                 size = SZ_4K | SZ_2M | SZ_1G;
1702         } else {
1703                 size = (id >> ID2_UBS_SHIFT) & ID2_UBS_MASK;
1704                 smmu->va_size = arm_smmu_id_size_to_bits(size);
1705 #ifndef CONFIG_64BIT
1706                 smmu->va_size = min(32UL, smmu->va_size);
1707 #endif
1708                 size = 0;
1709                 if (id & ID2_PTFS_4K)
1710                         size |= SZ_4K | SZ_2M | SZ_1G;
1711                 if (id & ID2_PTFS_16K)
1712                         size |= SZ_16K | SZ_32M;
1713                 if (id & ID2_PTFS_64K)
1714                         size |= SZ_64K | SZ_512M;
1715         }
1716
1717         arm_smmu_ops.pgsize_bitmap &= size;
1718         dev_notice(smmu->dev, "\tSupported page sizes: 0x%08lx\n", size);
1719
1720         if (smmu->features & ARM_SMMU_FEAT_TRANS_S1)
1721                 dev_notice(smmu->dev, "\tStage-1: %lu-bit VA -> %lu-bit IPA\n",
1722                            smmu->va_size, smmu->ipa_size);
1723
1724         if (smmu->features & ARM_SMMU_FEAT_TRANS_S2)
1725                 dev_notice(smmu->dev, "\tStage-2: %lu-bit IPA -> %lu-bit PA\n",
1726                            smmu->ipa_size, smmu->pa_size);
1727
1728         return 0;
1729 }
1730
1731 static const struct of_device_id arm_smmu_of_match[] = {
1732         { .compatible = "arm,smmu-v1", .data = (void *)ARM_SMMU_V1 },
1733         { .compatible = "arm,smmu-v2", .data = (void *)ARM_SMMU_V2 },
1734         { .compatible = "arm,mmu-400", .data = (void *)ARM_SMMU_V1 },
1735         { .compatible = "arm,mmu-401", .data = (void *)ARM_SMMU_V1 },
1736         { .compatible = "arm,mmu-500", .data = (void *)ARM_SMMU_V2 },
1737         { },
1738 };
1739 MODULE_DEVICE_TABLE(of, arm_smmu_of_match);
1740
1741 static int arm_smmu_device_dt_probe(struct platform_device *pdev)
1742 {
1743         const struct of_device_id *of_id;
1744         struct resource *res;
1745         struct arm_smmu_device *smmu;
1746         struct device *dev = &pdev->dev;
1747         struct rb_node *node;
1748         struct of_phandle_args masterspec;
1749         int num_irqs, i, err;
1750
1751         smmu = devm_kzalloc(dev, sizeof(*smmu), GFP_KERNEL);
1752         if (!smmu) {
1753                 dev_err(dev, "failed to allocate arm_smmu_device\n");
1754                 return -ENOMEM;
1755         }
1756         smmu->dev = dev;
1757
1758         of_id = of_match_node(arm_smmu_of_match, dev->of_node);
1759         smmu->version = (enum arm_smmu_arch_version)of_id->data;
1760
1761         res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1762         smmu->base = devm_ioremap_resource(dev, res);
1763         if (IS_ERR(smmu->base))
1764                 return PTR_ERR(smmu->base);
1765         smmu->size = resource_size(res);
1766
1767         if (of_property_read_u32(dev->of_node, "#global-interrupts",
1768                                  &smmu->num_global_irqs)) {
1769                 dev_err(dev, "missing #global-interrupts property\n");
1770                 return -ENODEV;
1771         }
1772
1773         num_irqs = 0;
1774         while ((res = platform_get_resource(pdev, IORESOURCE_IRQ, num_irqs))) {
1775                 num_irqs++;
1776                 if (num_irqs > smmu->num_global_irqs)
1777                         smmu->num_context_irqs++;
1778         }
1779
1780         if (!smmu->num_context_irqs) {
1781                 dev_err(dev, "found %d interrupts but expected at least %d\n",
1782                         num_irqs, smmu->num_global_irqs + 1);
1783                 return -ENODEV;
1784         }
1785
1786         smmu->irqs = devm_kzalloc(dev, sizeof(*smmu->irqs) * num_irqs,
1787                                   GFP_KERNEL);
1788         if (!smmu->irqs) {
1789                 dev_err(dev, "failed to allocate %d irqs\n", num_irqs);
1790                 return -ENOMEM;
1791         }
1792
1793         for (i = 0; i < num_irqs; ++i) {
1794                 int irq = platform_get_irq(pdev, i);
1795
1796                 if (irq < 0) {
1797                         dev_err(dev, "failed to get irq index %d\n", i);
1798                         return -ENODEV;
1799                 }
1800                 smmu->irqs[i] = irq;
1801         }
1802
1803         err = arm_smmu_device_cfg_probe(smmu);
1804         if (err)
1805                 return err;
1806
1807         i = 0;
1808         smmu->masters = RB_ROOT;
1809         while (!of_parse_phandle_with_args(dev->of_node, "mmu-masters",
1810                                            "#stream-id-cells", i,
1811                                            &masterspec)) {
1812                 err = register_smmu_master(smmu, dev, &masterspec);
1813                 if (err) {
1814                         dev_err(dev, "failed to add master %s\n",
1815                                 masterspec.np->name);
1816                         goto out_put_masters;
1817                 }
1818
1819                 i++;
1820         }
1821         dev_notice(dev, "registered %d master devices\n", i);
1822
1823         parse_driver_options(smmu);
1824
1825         if (smmu->version > ARM_SMMU_V1 &&
1826             smmu->num_context_banks != smmu->num_context_irqs) {
1827                 dev_err(dev,
1828                         "found only %d context interrupt(s) but %d required\n",
1829                         smmu->num_context_irqs, smmu->num_context_banks);
1830                 err = -ENODEV;
1831                 goto out_put_masters;
1832         }
1833
1834         for (i = 0; i < smmu->num_global_irqs; ++i) {
1835                 err = request_irq(smmu->irqs[i],
1836                                   arm_smmu_global_fault,
1837                                   IRQF_SHARED,
1838                                   "arm-smmu global fault",
1839                                   smmu);
1840                 if (err) {
1841                         dev_err(dev, "failed to request global IRQ %d (%u)\n",
1842                                 i, smmu->irqs[i]);
1843                         goto out_free_irqs;
1844                 }
1845         }
1846
1847         INIT_LIST_HEAD(&smmu->list);
1848         spin_lock(&arm_smmu_devices_lock);
1849         list_add(&smmu->list, &arm_smmu_devices);
1850         spin_unlock(&arm_smmu_devices_lock);
1851
1852         arm_smmu_device_reset(smmu);
1853         return 0;
1854
1855 out_free_irqs:
1856         while (i--)
1857                 free_irq(smmu->irqs[i], smmu);
1858
1859 out_put_masters:
1860         for (node = rb_first(&smmu->masters); node; node = rb_next(node)) {
1861                 struct arm_smmu_master *master
1862                         = container_of(node, struct arm_smmu_master, node);
1863                 of_node_put(master->of_node);
1864         }
1865
1866         return err;
1867 }
1868
1869 static int arm_smmu_device_remove(struct platform_device *pdev)
1870 {
1871         int i;
1872         struct device *dev = &pdev->dev;
1873         struct arm_smmu_device *curr, *smmu = NULL;
1874         struct rb_node *node;
1875
1876         spin_lock(&arm_smmu_devices_lock);
1877         list_for_each_entry(curr, &arm_smmu_devices, list) {
1878                 if (curr->dev == dev) {
1879                         smmu = curr;
1880                         list_del(&smmu->list);
1881                         break;
1882                 }
1883         }
1884         spin_unlock(&arm_smmu_devices_lock);
1885
1886         if (!smmu)
1887                 return -ENODEV;
1888
1889         for (node = rb_first(&smmu->masters); node; node = rb_next(node)) {
1890                 struct arm_smmu_master *master
1891                         = container_of(node, struct arm_smmu_master, node);
1892                 of_node_put(master->of_node);
1893         }
1894
1895         if (!bitmap_empty(smmu->context_map, ARM_SMMU_MAX_CBS))
1896                 dev_err(dev, "removing device with active domains!\n");
1897
1898         for (i = 0; i < smmu->num_global_irqs; ++i)
1899                 free_irq(smmu->irqs[i], smmu);
1900
1901         /* Turn the thing off */
1902         writel(sCR0_CLIENTPD, ARM_SMMU_GR0_NS(smmu) + ARM_SMMU_GR0_sCR0);
1903         return 0;
1904 }
1905
1906 static struct platform_driver arm_smmu_driver = {
1907         .driver = {
1908                 .name           = "arm-smmu",
1909                 .of_match_table = of_match_ptr(arm_smmu_of_match),
1910         },
1911         .probe  = arm_smmu_device_dt_probe,
1912         .remove = arm_smmu_device_remove,
1913 };
1914
1915 static int __init arm_smmu_init(void)
1916 {
1917         struct device_node *np;
1918         int ret;
1919
1920         /*
1921          * Play nice with systems that don't have an ARM SMMU by checking that
1922          * an ARM SMMU exists in the system before proceeding with the driver
1923          * and IOMMU bus operation registration.
1924          */
1925         np = of_find_matching_node(NULL, arm_smmu_of_match);
1926         if (!np)
1927                 return 0;
1928
1929         of_node_put(np);
1930
1931         ret = platform_driver_register(&arm_smmu_driver);
1932         if (ret)
1933                 return ret;
1934
1935         /* Oh, for a proper bus abstraction */
1936         if (!iommu_present(&platform_bus_type))
1937                 bus_set_iommu(&platform_bus_type, &arm_smmu_ops);
1938
1939 #ifdef CONFIG_ARM_AMBA
1940         if (!iommu_present(&amba_bustype))
1941                 bus_set_iommu(&amba_bustype, &arm_smmu_ops);
1942 #endif
1943
1944 #ifdef CONFIG_PCI
1945         if (!iommu_present(&pci_bus_type))
1946                 bus_set_iommu(&pci_bus_type, &arm_smmu_ops);
1947 #endif
1948
1949         return 0;
1950 }
1951
1952 static void __exit arm_smmu_exit(void)
1953 {
1954         return platform_driver_unregister(&arm_smmu_driver);
1955 }
1956
1957 subsys_initcall(arm_smmu_init);
1958 module_exit(arm_smmu_exit);
1959
1960 MODULE_DESCRIPTION("IOMMU API for ARM architected SMMU implementations");
1961 MODULE_AUTHOR("Will Deacon <will.deacon@arm.com>");
1962 MODULE_LICENSE("GPL v2");