drivers/iommu/intel_irq_remapping.c

   1 #include <linux/interrupt.h>
   2 #include <linux/dmar.h>
   3 #include <linux/spinlock.h>
   4 #include <linux/slab.h>
   5 #include <linux/jiffies.h>
   6 #include <linux/hpet.h>
   7 #include <linux/pci.h>
   8 #include <linux/irq.h>
   9 #include <asm/io_apic.h>
  10 #include <asm/smp.h>
  11 #include <asm/cpu.h>
  12 #include <linux/intel-iommu.h>
  13 #include <acpi/acpi.h>
  14 #include <asm/irq_remapping.h>
  15 #include <asm/pci-direct.h>
  16 #include <asm/msidef.h>
  17
  18 #include "irq_remapping.h"
  19
  20 struct ioapic_scope {
  21         struct intel_iommu *iommu;
  22         unsigned int id;
  23         unsigned int bus;       /* PCI bus number */
  24         unsigned int devfn;     /* PCI devfn number */
  25 };
  26
  27 struct hpet_scope {
  28         struct intel_iommu *iommu;
  29         u8 id;
  30         unsigned int bus;
  31         unsigned int devfn;
  32 };
  33
  34 #define IR_X2APIC_MODE(mode) (mode ? (1 << 11) : 0)
  35 #define IRTE_DEST(dest) ((x2apic_mode) ? dest : dest << 8)
  36
  37 static struct ioapic_scope ir_ioapic[MAX_IO_APICS];
  38 static struct hpet_scope ir_hpet[MAX_HPET_TBS];
  39 static int ir_ioapic_num, ir_hpet_num;
  40
  41 static DEFINE_RAW_SPINLOCK(irq_2_ir_lock);
  42
  43 static struct irq_2_iommu *irq_2_iommu(unsigned int irq)
  44 {
  45         struct irq_cfg *cfg = irq_get_chip_data(irq);
  46         return cfg ? &cfg->irq_2_iommu : NULL;
  47 }
  48
  49 int get_irte(int irq, struct irte *entry)
  50 {
  51         struct irq_2_iommu *irq_iommu = irq_2_iommu(irq);
  52         unsigned long flags;
  53         int index;
  54
  55         if (!entry || !irq_iommu)
  56                 return -1;
  57
  58         raw_spin_lock_irqsave(&irq_2_ir_lock, flags);
  59
  60         index = irq_iommu->irte_index + irq_iommu->sub_handle;
  61         *entry = *(irq_iommu->iommu->ir_table->base + index);
  62
  63         raw_spin_unlock_irqrestore(&irq_2_ir_lock, flags);
  64         return 0;
  65 }
  66
  67 static int alloc_irte(struct intel_iommu *iommu, int irq, u16 count)
  68 {
  69         struct ir_table *table = iommu->ir_table;
  70         struct irq_2_iommu *irq_iommu = irq_2_iommu(irq);
  71         struct irq_cfg *cfg = irq_get_chip_data(irq);
  72         u16 index, start_index;
  73         unsigned int mask = 0;
  74         unsigned long flags;
  75         int i;
  76
  77         if (!count || !irq_iommu)
  78                 return -1;
  79
  80         /*
  81          * start the IRTE search from index 0.
  82          */
  83         index = start_index = 0;
  84
  85         if (count > 1) {
  86                 count = __roundup_pow_of_two(count);
  87                 mask = ilog2(count);
  88         }
  89
  90         if (mask > ecap_max_handle_mask(iommu->ecap)) {
  91                 printk(KERN_ERR
  92                        "Requested mask %x exceeds the max invalidation handle"
  93                        " mask value %Lx\n", mask,
  94                        ecap_max_handle_mask(iommu->ecap));
  95                 return -1;
  96         }
  97
  98         raw_spin_lock_irqsave(&irq_2_ir_lock, flags);
  99         do {
 100                 for (i = index; i < index + count; i++)
 101                         if  (table->base[i].present)
 102                                 break;
 103                 /* empty index found */
 104                 if (i == index + count)
 105                         break;
 106
 107                 index = (index + count) % INTR_REMAP_TABLE_ENTRIES;
 108
 109                 if (index == start_index) {
 110                         raw_spin_unlock_irqrestore(&irq_2_ir_lock, flags);
 111                         printk(KERN_ERR "can't allocate an IRTE\n");
 112                         return -1;
 113                 }
 114         } while (1);
 115
 116         for (i = index; i < index + count; i++)
 117                 table->base[i].present = 1;
 118
 119         cfg->remapped = 1;
 120         irq_iommu->iommu = iommu;
 121         irq_iommu->irte_index =  index;
 122         irq_iommu->sub_handle = 0;
 123         irq_iommu->irte_mask = mask;
 124
 125         raw_spin_unlock_irqrestore(&irq_2_ir_lock, flags);
 126
 127         return index;
 128 }
 129
 130 static int qi_flush_iec(struct intel_iommu *iommu, int index, int mask)
 131 {
 132         struct qi_desc desc;
 133
 134         desc.low = QI_IEC_IIDEX(index) | QI_IEC_TYPE | QI_IEC_IM(mask)
 135                    | QI_IEC_SELECTIVE;
 136         desc.high = 0;
 137
 138         return qi_submit_sync(&desc, iommu);
 139 }
 140
 141 static int map_irq_to_irte_handle(int irq, u16 *sub_handle)
 142 {
 143         struct irq_2_iommu *irq_iommu = irq_2_iommu(irq);
 144         unsigned long flags;
 145         int index;
 146
 147         if (!irq_iommu)
 148                 return -1;
 149
 150         raw_spin_lock_irqsave(&irq_2_ir_lock, flags);
 151         *sub_handle = irq_iommu->sub_handle;
 152         index = irq_iommu->irte_index;
 153         raw_spin_unlock_irqrestore(&irq_2_ir_lock, flags);
 154         return index;
 155 }
 156
 157 static int set_irte_irq(int irq, struct intel_iommu *iommu, u16 index, u16 subhandle)
 158 {
 159         struct irq_2_iommu *irq_iommu = irq_2_iommu(irq);
 160         struct irq_cfg *cfg = irq_get_chip_data(irq);
 161         unsigned long flags;
 162
 163         if (!irq_iommu)
 164                 return -1;
 165
 166         raw_spin_lock_irqsave(&irq_2_ir_lock, flags);
 167
 168         cfg->remapped = 1;
 169         irq_iommu->iommu = iommu;
 170         irq_iommu->irte_index = index;
 171         irq_iommu->sub_handle = subhandle;
 172         irq_iommu->irte_mask = 0;
 173
 174         raw_spin_unlock_irqrestore(&irq_2_ir_lock, flags);
 175
 176         return 0;
 177 }
 178
 179 static int modify_irte(int irq, struct irte *irte_modified)
 180 {
 181         struct irq_2_iommu *irq_iommu = irq_2_iommu(irq);
 182         struct intel_iommu *iommu;
 183         unsigned long flags;
 184         struct irte *irte;
 185         int rc, index;
 186
 187         if (!irq_iommu)
 188                 return -1;
 189
 190         raw_spin_lock_irqsave(&irq_2_ir_lock, flags);
 191
 192         iommu = irq_iommu->iommu;
 193
 194         index = irq_iommu->irte_index + irq_iommu->sub_handle;
 195         irte = &iommu->ir_table->base[index];
 196
 197         set_64bit(&irte->low, irte_modified->low);
 198         set_64bit(&irte->high, irte_modified->high);
 199         __iommu_flush_cache(iommu, irte, sizeof(*irte));
 200
 201         rc = qi_flush_iec(iommu, index, 0);
 202         raw_spin_unlock_irqrestore(&irq_2_ir_lock, flags);
 203
 204         return rc;
 205 }
 206
 207 static struct intel_iommu *map_hpet_to_ir(u8 hpet_id)
 208 {
 209         int i;
 210
 211         for (i = 0; i < MAX_HPET_TBS; i++)
 212                 if (ir_hpet[i].id == hpet_id)
 213                         return ir_hpet[i].iommu;
 214         return NULL;
 215 }
 216
 217 static struct intel_iommu *map_ioapic_to_ir(int apic)
 218 {
 219         int i;
 220
 221         for (i = 0; i < MAX_IO_APICS; i++)
 222                 if (ir_ioapic[i].id == apic)
 223                         return ir_ioapic[i].iommu;
 224         return NULL;
 225 }
 226
 227 static struct intel_iommu *map_dev_to_ir(struct pci_dev *dev)
 228 {
 229         struct dmar_drhd_unit *drhd;
 230
 231         drhd = dmar_find_matched_drhd_unit(dev);
 232         if (!drhd)
 233                 return NULL;
 234
 235         return drhd->iommu;
 236 }
 237
 238 static int clear_entries(struct irq_2_iommu *irq_iommu)
 239 {
 240         struct irte *start, *entry, *end;
 241         struct intel_iommu *iommu;
 242         int index;
 243
 244         if (irq_iommu->sub_handle)
 245                 return 0;
 246
 247         iommu = irq_iommu->iommu;
 248         index = irq_iommu->irte_index + irq_iommu->sub_handle;
 249
 250         start = iommu->ir_table->base + index;
 251         end = start + (1 << irq_iommu->irte_mask);
 252
 253         for (entry = start; entry < end; entry++) {
 254                 set_64bit(&entry->low, 0);
 255                 set_64bit(&entry->high, 0);
 256         }
 257
 258         return qi_flush_iec(iommu, index, irq_iommu->irte_mask);
 259 }
 260
 261 static int free_irte(int irq)
 262 {
 263         struct irq_2_iommu *irq_iommu = irq_2_iommu(irq);
 264         unsigned long flags;
 265         int rc;
 266
 267         if (!irq_iommu)
 268                 return -1;
 269
 270         raw_spin_lock_irqsave(&irq_2_ir_lock, flags);
 271
 272         rc = clear_entries(irq_iommu);
 273
 274         irq_iommu->iommu = NULL;
 275         irq_iommu->irte_index = 0;
 276         irq_iommu->sub_handle = 0;
 277         irq_iommu->irte_mask = 0;
 278
 279         raw_spin_unlock_irqrestore(&irq_2_ir_lock, flags);
 280
 281         return rc;
 282 }
 283
 284 /*
 285  * source validation type
 286  */
 287 #define SVT_NO_VERIFY           0x0  /* no verification is required */
 288 #define SVT_VERIFY_SID_SQ       0x1  /* verify using SID and SQ fields */
 289 #define SVT_VERIFY_BUS          0x2  /* verify bus of request-id */
 290
 291 /*
 292  * source-id qualifier
 293  */
 294 #define SQ_ALL_16       0x0  /* verify all 16 bits of request-id */
 295 #define SQ_13_IGNORE_1  0x1  /* verify most significant 13 bits, ignore
 296                               * the third least significant bit
 297                               */
 298 #define SQ_13_IGNORE_2  0x2  /* verify most significant 13 bits, ignore
 299                               * the second and third least significant bits
 300                               */
 301 #define SQ_13_IGNORE_3  0x3  /* verify most significant 13 bits, ignore
 302                               * the least three significant bits
 303                               */
 304
 305 /*
 306  * set SVT, SQ and SID fields of irte to verify
 307  * source ids of interrupt requests
 308  */
 309 static void set_irte_sid(struct irte *irte, unsigned int svt,
 310                          unsigned int sq, unsigned int sid)
 311 {
 312         if (disable_sourceid_checking)
 313                 svt = SVT_NO_VERIFY;
 314         irte->svt = svt;
 315         irte->sq = sq;
 316         irte->sid = sid;
 317 }
 318
 319 static int set_ioapic_sid(struct irte *irte, int apic)
 320 {
 321         int i;
 322         u16 sid = 0;
 323
 324         if (!irte)
 325                 return -1;
 326
 327         for (i = 0; i < MAX_IO_APICS; i++) {
 328                 if (ir_ioapic[i].id == apic) {
 329                         sid = (ir_ioapic[i].bus << 8) | ir_ioapic[i].devfn;
 330                         break;
 331                 }
 332         }
 333
 334         if (sid == 0) {
 335                 pr_warning("Failed to set source-id of IOAPIC (%d)\n", apic);
 336                 return -1;
 337         }
 338
 339         set_irte_sid(irte, 1, 0, sid);
 340
 341         return 0;
 342 }
 343
 344 static int set_hpet_sid(struct irte *irte, u8 id)
 345 {
 346         int i;
 347         u16 sid = 0;
 348
 349         if (!irte)
 350                 return -1;
 351
 352         for (i = 0; i < MAX_HPET_TBS; i++) {
 353                 if (ir_hpet[i].id == id) {
 354                         sid = (ir_hpet[i].bus << 8) | ir_hpet[i].devfn;
 355                         break;
 356                 }
 357         }
 358
 359         if (sid == 0) {
 360                 pr_warning("Failed to set source-id of HPET block (%d)\n", id);
 361                 return -1;
 362         }
 363
 364         /*
 365          * Should really use SQ_ALL_16. Some platforms are broken.
 366          * While we figure out the right quirks for these broken platforms, use
 367          * SQ_13_IGNORE_3 for now.
 368          */
 369         set_irte_sid(irte, SVT_VERIFY_SID_SQ, SQ_13_IGNORE_3, sid);
 370
 371         return 0;
 372 }
 373
 374 static int set_msi_sid(struct irte *irte, struct pci_dev *dev)
 375 {
 376         struct pci_dev *bridge;
 377
 378         if (!irte || !dev)
 379                 return -1;
 380
 381         /* PCIe device or Root Complex integrated PCI device */
 382         if (pci_is_pcie(dev) || !dev->bus->parent) {
 383                 set_irte_sid(irte, SVT_VERIFY_SID_SQ, SQ_ALL_16,
 384                              (dev->bus->number << 8) | dev->devfn);
 385                 return 0;
 386         }
 387
 388         bridge = pci_find_upstream_pcie_bridge(dev);
 389         if (bridge) {
 390                 if (pci_is_pcie(bridge))/* this is a PCIe-to-PCI/PCIX bridge */
 391                         set_irte_sid(irte, SVT_VERIFY_BUS, SQ_ALL_16,
 392                                 (bridge->bus->number << 8) | dev->bus->number);
 393                 else /* this is a legacy PCI bridge */
 394                         set_irte_sid(irte, SVT_VERIFY_SID_SQ, SQ_ALL_16,
 395                                 (bridge->bus->number << 8) | bridge->devfn);
 396         }
 397
 398         return 0;
 399 }
 400
 401 static void iommu_set_irq_remapping(struct intel_iommu *iommu, int mode)
 402 {
 403         u64 addr;
 404         u32 sts;
 405         unsigned long flags;
 406
 407         addr = virt_to_phys((void *)iommu->ir_table->base);
 408
 409         raw_spin_lock_irqsave(&iommu->register_lock, flags);
 410
 411         dmar_writeq(iommu->reg + DMAR_IRTA_REG,
 412                     (addr) | IR_X2APIC_MODE(mode) | INTR_REMAP_TABLE_REG_SIZE);
 413
 414         /* Set interrupt-remapping table pointer */
 415         iommu->gcmd |= DMA_GCMD_SIRTP;
 416         writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
 417
 418         IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
 419                       readl, (sts & DMA_GSTS_IRTPS), sts);
 420         raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
 421
 422         /*
 423          * global invalidation of interrupt entry cache before enabling
 424          * interrupt-remapping.
 425          */
 426         qi_global_iec(iommu);
 427
 428         raw_spin_lock_irqsave(&iommu->register_lock, flags);
 429
 430         /* Enable interrupt-remapping */
 431         iommu->gcmd |= DMA_GCMD_IRE;
 432         iommu->gcmd &= ~DMA_GCMD_CFI;  /* Block compatibility-format MSIs */
 433         writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
 434
 435         IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
 436                       readl, (sts & DMA_GSTS_IRES), sts);
 437
 438         /*
 439          * With CFI clear in the Global Command register, we should be
 440          * protected from dangerous (i.e. compatibility) interrupts
 441          * regardless of x2apic status.  Check just to be sure.
 442          */
 443         if (sts & DMA_GSTS_CFIS)
 444                 WARN(1, KERN_WARNING
 445                         "Compatibility-format IRQs enabled despite intr remapping;\n"
 446                         "you are vulnerable to IRQ injection.\n");
 447
 448         raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
 449 }
 450
 451
 452 static int intel_setup_irq_remapping(struct intel_iommu *iommu, int mode)
 453 {
 454         struct ir_table *ir_table;
 455         struct page *pages;
 456
 457         ir_table = iommu->ir_table = kzalloc(sizeof(struct ir_table),
 458                                              GFP_ATOMIC);
 459
 460         if (!iommu->ir_table)
 461                 return -ENOMEM;
 462
 463         pages = alloc_pages_node(iommu->node, GFP_ATOMIC | __GFP_ZERO,
 464                                  INTR_REMAP_PAGE_ORDER);
 465
 466         if (!pages) {
 467                 printk(KERN_ERR "failed to allocate pages of order %d\n",
 468                        INTR_REMAP_PAGE_ORDER);
 469                 kfree(iommu->ir_table);
 470                 return -ENOMEM;
 471         }
 472
 473         ir_table->base = page_address(pages);
 474
 475         iommu_set_irq_remapping(iommu, mode);
 476         return 0;
 477 }
 478
 479 /*
 480  * Disable Interrupt Remapping.
 481  */
 482 static void iommu_disable_irq_remapping(struct intel_iommu *iommu)
 483 {
 484         unsigned long flags;
 485         u32 sts;
 486
 487         if (!ecap_ir_support(iommu->ecap))
 488                 return;
 489
 490         /*
 491          * global invalidation of interrupt entry cache before disabling
 492          * interrupt-remapping.
 493          */
 494         qi_global_iec(iommu);
 495
 496         raw_spin_lock_irqsave(&iommu->register_lock, flags);
 497
 498         sts = dmar_readq(iommu->reg + DMAR_GSTS_REG);
 499         if (!(sts & DMA_GSTS_IRES))
 500                 goto end;
 501
 502         iommu->gcmd &= ~DMA_GCMD_IRE;
 503         writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
 504
 505         IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
 506                       readl, !(sts & DMA_GSTS_IRES), sts);
 507
 508 end:
 509         raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
 510 }
 511
 512 static int __init dmar_x2apic_optout(void)
 513 {
 514         struct acpi_table_dmar *dmar;
 515         dmar = (struct acpi_table_dmar *)dmar_tbl;
 516         if (!dmar || no_x2apic_optout)
 517                 return 0;
 518         return dmar->flags & DMAR_X2APIC_OPT_OUT;
 519 }
 520
 521 static int __init intel_irq_remapping_supported(void)
 522 {
 523         struct dmar_drhd_unit *drhd;
 524
 525         if (disable_irq_remap)
 526                 return 0;
 527         if (irq_remap_broken) {
 528                 printk(KERN_WARNING
 529                         "This system BIOS has enabled interrupt remapping\n"
 530                         "on a chipset that contains an erratum making that\n"
 531                         "feature unstable.  To maintain system stability\n"
 532                         "interrupt remapping is being disabled.  Please\n"
 533                         "contact your BIOS vendor for an update\n");
 534                 add_taint(TAINT_FIRMWARE_WORKAROUND, LOCKDEP_STILL_OK);
 535                 disable_irq_remap = 1;
 536                 return 0;
 537         }
 538
 539         if (!dmar_ir_support())
 540                 return 0;
 541
 542         for_each_drhd_unit(drhd) {
 543                 struct intel_iommu *iommu = drhd->iommu;
 544
 545                 if (!ecap_ir_support(iommu->ecap))
 546                         return 0;
 547         }
 548
 549         return 1;
 550 }
 551
 552 static int __init intel_enable_irq_remapping(void)
 553 {
 554         struct dmar_drhd_unit *drhd;
 555         bool x2apic_present;
 556         int setup = 0;
 557         int eim = 0;
 558
 559         x2apic_present = x2apic_supported();
 560
 561         if (parse_ioapics_under_ir() != 1) {
 562                 printk(KERN_INFO "Not enable interrupt remapping\n");
 563                 goto error;
 564         }
 565
 566         if (x2apic_present) {
 567                 eim = !dmar_x2apic_optout();
 568                 if (!eim)
 569                         printk(KERN_WARNING
 570                                 "Your BIOS is broken and requested that x2apic be disabled.\n"
 571                                 "This will slightly decrease performance.\n"
 572                                 "Use 'intremap=no_x2apic_optout' to override BIOS request.\n");
 573         }
 574
 575         for_each_drhd_unit(drhd) {
 576                 struct intel_iommu *iommu = drhd->iommu;
 577
 578                 /*
 579                  * If the queued invalidation is already initialized,
 580                  * shouldn't disable it.
 581                  */
 582                 if (iommu->qi)
 583                         continue;
 584
 585                 /*
 586                  * Clear previous faults.
 587                  */
 588                 dmar_fault(-1, iommu);
 589
 590                 /*
 591                  * Disable intr remapping and queued invalidation, if already
 592                  * enabled prior to OS handover.
 593                  */
 594                 iommu_disable_irq_remapping(iommu);
 595
 596                 dmar_disable_qi(iommu);
 597         }
 598
 599         /*
 600          * check for the Interrupt-remapping support
 601          */
 602         for_each_drhd_unit(drhd) {
 603                 struct intel_iommu *iommu = drhd->iommu;
 604
 605                 if (!ecap_ir_support(iommu->ecap))
 606                         continue;
 607
 608                 if (eim && !ecap_eim_support(iommu->ecap)) {
 609                         printk(KERN_INFO "DRHD %Lx: EIM not supported by DRHD, "
 610                                " ecap %Lx\n", drhd->reg_base_addr, iommu->ecap);
 611                         goto error;
 612                 }
 613         }
 614
 615         /*
 616          * Enable queued invalidation for all the DRHD's.
 617          */
 618         for_each_drhd_unit(drhd) {
 619                 int ret;
 620                 struct intel_iommu *iommu = drhd->iommu;
 621                 ret = dmar_enable_qi(iommu);
 622
 623                 if (ret) {
 624                         printk(KERN_ERR "DRHD %Lx: failed to enable queued, "
 625                                " invalidation, ecap %Lx, ret %d\n",
 626                                drhd->reg_base_addr, iommu->ecap, ret);
 627                         goto error;
 628                 }
 629         }
 630
 631         /*
 632          * Setup Interrupt-remapping for all the DRHD's now.
 633          */
 634         for_each_drhd_unit(drhd) {
 635                 struct intel_iommu *iommu = drhd->iommu;
 636
 637                 if (!ecap_ir_support(iommu->ecap))
 638                         continue;
 639
 640                 if (intel_setup_irq_remapping(iommu, eim))
 641                         goto error;
 642
 643                 setup = 1;
 644         }
 645
 646         if (!setup)
 647                 goto error;
 648
 649         irq_remapping_enabled = 1;
 650
 651         /*
 652          * VT-d has a different layout for IO-APIC entries when
 653          * interrupt remapping is enabled. So it needs a special routine
 654          * to print IO-APIC entries for debugging purposes too.
 655          */
 656         x86_io_apic_ops.print_entries = intel_ir_io_apic_print_entries;
 657
 658         pr_info("Enabled IRQ remapping in %s mode\n", eim ? "x2apic" : "xapic");
 659
 660         return eim ? IRQ_REMAP_X2APIC_MODE : IRQ_REMAP_XAPIC_MODE;
 661
 662 error:
 663         /*
 664          * handle error condition gracefully here!
 665          */
 666
 667         if (x2apic_present)
 668                 pr_warn("Failed to enable irq remapping.  You are vulnerable to irq-injection attacks.\n");
 669
 670         return -1;
 671 }
 672
 673 static void ir_parse_one_hpet_scope(struct acpi_dmar_device_scope *scope,
 674                                       struct intel_iommu *iommu)
 675 {
 676         struct acpi_dmar_pci_path *path;
 677         u8 bus;
 678         int count;
 679
 680         bus = scope->bus;
 681         path = (struct acpi_dmar_pci_path *)(scope + 1);
 682         count = (scope->length - sizeof(struct acpi_dmar_device_scope))
 683                 / sizeof(struct acpi_dmar_pci_path);
 684
 685         while (--count > 0) {
 686                 /*
 687                  * Access PCI directly due to the PCI
 688                  * subsystem isn't initialized yet.
 689                  */
 690                 bus = read_pci_config_byte(bus, path->dev, path->fn,
 691                                            PCI_SECONDARY_BUS);
 692                 path++;
 693         }
 694         ir_hpet[ir_hpet_num].bus   = bus;
 695         ir_hpet[ir_hpet_num].devfn = PCI_DEVFN(path->dev, path->fn);
 696         ir_hpet[ir_hpet_num].iommu = iommu;
 697         ir_hpet[ir_hpet_num].id    = scope->enumeration_id;
 698         ir_hpet_num++;
 699 }
 700
 701 static void ir_parse_one_ioapic_scope(struct acpi_dmar_device_scope *scope,
 702                                       struct intel_iommu *iommu)
 703 {
 704         struct acpi_dmar_pci_path *path;
 705         u8 bus;
 706         int count;
 707
 708         bus = scope->bus;
 709         path = (struct acpi_dmar_pci_path *)(scope + 1);
 710         count = (scope->length - sizeof(struct acpi_dmar_device_scope))
 711                 / sizeof(struct acpi_dmar_pci_path);
 712
 713         while (--count > 0) {
 714                 /*
 715                  * Access PCI directly due to the PCI
 716                  * subsystem isn't initialized yet.
 717                  */
 718                 bus = read_pci_config_byte(bus, path->dev, path->fn,
 719                                            PCI_SECONDARY_BUS);
 720                 path++;
 721         }
 722
 723         ir_ioapic[ir_ioapic_num].bus   = bus;
 724         ir_ioapic[ir_ioapic_num].devfn = PCI_DEVFN(path->dev, path->fn);
 725         ir_ioapic[ir_ioapic_num].iommu = iommu;
 726         ir_ioapic[ir_ioapic_num].id    = scope->enumeration_id;
 727         ir_ioapic_num++;
 728 }
 729
 730 static int ir_parse_ioapic_hpet_scope(struct acpi_dmar_header *header,
 731                                       struct intel_iommu *iommu)
 732 {
 733         struct acpi_dmar_hardware_unit *drhd;
 734         struct acpi_dmar_device_scope *scope;
 735         void *start, *end;
 736
 737         drhd = (struct acpi_dmar_hardware_unit *)header;
 738
 739         start = (void *)(drhd + 1);
 740         end = ((void *)drhd) + header->length;
 741
 742         while (start < end) {
 743                 scope = start;
 744                 if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_IOAPIC) {
 745                         if (ir_ioapic_num == MAX_IO_APICS) {
 746                                 printk(KERN_WARNING "Exceeded Max IO APICS\n");
 747                                 return -1;
 748                         }
 749
 750                         printk(KERN_INFO "IOAPIC id %d under DRHD base "
 751                                " 0x%Lx IOMMU %d\n", scope->enumeration_id,
 752                                drhd->address, iommu->seq_id);
 753
 754                         ir_parse_one_ioapic_scope(scope, iommu);
 755                 } else if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_HPET) {
 756                         if (ir_hpet_num == MAX_HPET_TBS) {
 757                                 printk(KERN_WARNING "Exceeded Max HPET blocks\n");
 758                                 return -1;
 759                         }
 760
 761                         printk(KERN_INFO "HPET id %d under DRHD base"
 762                                " 0x%Lx\n", scope->enumeration_id,
 763                                drhd->address);
 764
 765                         ir_parse_one_hpet_scope(scope, iommu);
 766                 }
 767                 start += scope->length;
 768         }
 769
 770         return 0;
 771 }
 772
 773 /*
 774  * Finds the assocaition between IOAPIC's and its Interrupt-remapping
 775  * hardware unit.
 776  */
 777 int __init parse_ioapics_under_ir(void)
 778 {
 779         struct dmar_drhd_unit *drhd;
 780         int ir_supported = 0;
 781         int ioapic_idx;
 782
 783         for_each_drhd_unit(drhd) {
 784                 struct intel_iommu *iommu = drhd->iommu;
 785
 786                 if (ecap_ir_support(iommu->ecap)) {
 787                         if (ir_parse_ioapic_hpet_scope(drhd->hdr, iommu))
 788                                 return -1;
 789
 790                         ir_supported = 1;
 791                 }
 792         }
 793
 794         if (!ir_supported)
 795                 return 0;
 796
 797         for (ioapic_idx = 0; ioapic_idx < nr_ioapics; ioapic_idx++) {
 798                 int ioapic_id = mpc_ioapic_id(ioapic_idx);
 799                 if (!map_ioapic_to_ir(ioapic_id)) {
 800                         pr_err(FW_BUG "ioapic %d has no mapping iommu, "
 801                                "interrupt remapping will be disabled\n",
 802                                ioapic_id);
 803                         return -1;
 804                 }
 805         }
 806
 807         return 1;
 808 }
 809
 810 int __init ir_dev_scope_init(void)
 811 {
 812         if (!irq_remapping_enabled)
 813                 return 0;
 814
 815         return dmar_dev_scope_init();
 816 }
 817 rootfs_initcall(ir_dev_scope_init);
 818
 819 static void disable_irq_remapping(void)
 820 {
 821         struct dmar_drhd_unit *drhd;
 822         struct intel_iommu *iommu = NULL;
 823
 824         /*
 825          * Disable Interrupt-remapping for all the DRHD's now.
 826          */
 827         for_each_iommu(iommu, drhd) {
 828                 if (!ecap_ir_support(iommu->ecap))
 829                         continue;
 830
 831                 iommu_disable_irq_remapping(iommu);
 832         }
 833 }
 834
 835 static int reenable_irq_remapping(int eim)
 836 {
 837         struct dmar_drhd_unit *drhd;
 838         int setup = 0;
 839         struct intel_iommu *iommu = NULL;
 840
 841         for_each_iommu(iommu, drhd)
 842                 if (iommu->qi)
 843                         dmar_reenable_qi(iommu);
 844
 845         /*
 846          * Setup Interrupt-remapping for all the DRHD's now.
 847          */
 848         for_each_iommu(iommu, drhd) {
 849                 if (!ecap_ir_support(iommu->ecap))
 850                         continue;
 851
 852                 /* Set up interrupt remapping for iommu.*/
 853                 iommu_set_irq_remapping(iommu, eim);
 854                 setup = 1;
 855         }
 856
 857         if (!setup)
 858                 goto error;
 859
 860         return 0;
 861
 862 error:
 863         /*
 864          * handle error condition gracefully here!
 865          */
 866         return -1;
 867 }
 868
 869 static void prepare_irte(struct irte *irte, int vector,
 870                          unsigned int dest)
 871 {
 872         memset(irte, 0, sizeof(*irte));
 873
 874         irte->present = 1;
 875         irte->dst_mode = apic->irq_dest_mode;
 876         /*
 877          * Trigger mode in the IRTE will always be edge, and for IO-APIC, the
 878          * actual level or edge trigger will be setup in the IO-APIC
 879          * RTE. This will help simplify level triggered irq migration.
 880          * For more details, see the comments (in io_apic.c) explainig IO-APIC
 881          * irq migration in the presence of interrupt-remapping.
 882         */
 883         irte->trigger_mode = 0;
 884         irte->dlvry_mode = apic->irq_delivery_mode;
 885         irte->vector = vector;
 886         irte->dest_id = IRTE_DEST(dest);
 887         irte->redir_hint = 1;
 888 }
 889
 890 static int intel_setup_ioapic_entry(int irq,
 891                                     struct IO_APIC_route_entry *route_entry,
 892                                     unsigned int destination, int vector,
 893                                     struct io_apic_irq_attr *attr)
 894 {
 895         int ioapic_id = mpc_ioapic_id(attr->ioapic);
 896         struct intel_iommu *iommu = map_ioapic_to_ir(ioapic_id);
 897         struct IR_IO_APIC_route_entry *entry;
 898         struct irte irte;
 899         int index;
 900
 901         if (!iommu) {
 902                 pr_warn("No mapping iommu for ioapic %d\n", ioapic_id);
 903                 return -ENODEV;
 904         }
 905
 906         entry = (struct IR_IO_APIC_route_entry *)route_entry;
 907
 908         index = alloc_irte(iommu, irq, 1);
 909         if (index < 0) {
 910                 pr_warn("Failed to allocate IRTE for ioapic %d\n", ioapic_id);
 911                 return -ENOMEM;
 912         }
 913
 914         prepare_irte(&irte, vector, destination);
 915
 916         /* Set source-id of interrupt request */
 917         set_ioapic_sid(&irte, ioapic_id);
 918
 919         modify_irte(irq, &irte);
 920
 921         apic_printk(APIC_VERBOSE, KERN_DEBUG "IOAPIC[%d]: "
 922                 "Set IRTE entry (P:%d FPD:%d Dst_Mode:%d "
 923                 "Redir_hint:%d Trig_Mode:%d Dlvry_Mode:%X "
 924                 "Avail:%X Vector:%02X Dest:%08X "
 925                 "SID:%04X SQ:%X SVT:%X)\n",
 926                 attr->ioapic, irte.present, irte.fpd, irte.dst_mode,
 927                 irte.redir_hint, irte.trigger_mode, irte.dlvry_mode,
 928                 irte.avail, irte.vector, irte.dest_id,
 929                 irte.sid, irte.sq, irte.svt);
 930
 931         memset(entry, 0, sizeof(*entry));
 932
 933         entry->index2   = (index >> 15) & 0x1;
 934         entry->zero     = 0;
 935         entry->format   = 1;
 936         entry->index    = (index & 0x7fff);
 937         /*
 938          * IO-APIC RTE will be configured with virtual vector.
 939          * irq handler will do the explicit EOI to the io-apic.
 940          */
 941         entry->vector   = attr->ioapic_pin;
 942         entry->mask     = 0;                    /* enable IRQ */
 943         entry->trigger  = attr->trigger;
 944         entry->polarity = attr->polarity;
 945
 946         /* Mask level triggered irqs.
 947          * Use IRQ_DELAYED_DISABLE for edge triggered irqs.
 948          */
 949         if (attr->trigger)
 950                 entry->mask = 1;
 951
 952         return 0;
 953 }
 954
 955 /*
 956  * Migrate the IO-APIC irq in the presence of intr-remapping.
 957  *
 958  * For both level and edge triggered, irq migration is a simple atomic
 959  * update(of vector and cpu destination) of IRTE and flush the hardware cache.
 960  *
 961  * For level triggered, we eliminate the io-apic RTE modification (with the
 962  * updated vector information), by using a virtual vector (io-apic pin number).
 963  * Real vector that is used for interrupting cpu will be coming from
 964  * the interrupt-remapping table entry.
 965  *
 966  * As the migration is a simple atomic update of IRTE, the same mechanism
 967  * is used to migrate MSI irq's in the presence of interrupt-remapping.
 968  */
 969 static int
 970 intel_ioapic_set_affinity(struct irq_data *data, const struct cpumask *mask,
 971                           bool force)
 972 {
 973         struct irq_cfg *cfg = data->chip_data;
 974         unsigned int dest, irq = data->irq;
 975         struct irte irte;
 976         int err;
 977
 978         if (!config_enabled(CONFIG_SMP))
 979                 return -EINVAL;
 980
 981         if (!cpumask_intersects(mask, cpu_online_mask))
 982                 return -EINVAL;
 983
 984         if (get_irte(irq, &irte))
 985                 return -EBUSY;
 986
 987         err = assign_irq_vector(irq, cfg, mask);
 988         if (err)
 989                 return err;
 990
 991         err = apic->cpu_mask_to_apicid_and(cfg->domain, mask, &dest);
 992         if (err) {
 993                 if (assign_irq_vector(irq, cfg, data->affinity))
 994                         pr_err("Failed to recover vector for irq %d\n", irq);
 995                 return err;
 996         }
 997
 998         irte.vector = cfg->vector;
 999         irte.dest_id = IRTE_DEST(dest);
1000
1001         /*
1002          * Atomically updates the IRTE with the new destination, vector
1003          * and flushes the interrupt entry cache.
1004          */
1005         modify_irte(irq, &irte);
1006
1007         /*
1008          * After this point, all the interrupts will start arriving
1009          * at the new destination. So, time to cleanup the previous
1010          * vector allocation.
1011          */
1012         if (cfg->move_in_progress)
1013                 send_cleanup_vector(cfg);
1014
1015         cpumask_copy(data->affinity, mask);
1016         return 0;
1017 }
1018
1019 static void intel_compose_msi_msg(struct pci_dev *pdev,
1020                                   unsigned int irq, unsigned int dest,
1021                                   struct msi_msg *msg, u8 hpet_id)
1022 {
1023         struct irq_cfg *cfg;
1024         struct irte irte;
1025         u16 sub_handle = 0;
1026         int ir_index;
1027
1028         cfg = irq_get_chip_data(irq);
1029
1030         ir_index = map_irq_to_irte_handle(irq, &sub_handle);
1031         BUG_ON(ir_index == -1);
1032
1033         prepare_irte(&irte, cfg->vector, dest);
1034
1035         /* Set source-id of interrupt request */
1036         if (pdev)
1037                 set_msi_sid(&irte, pdev);
1038         else
1039                 set_hpet_sid(&irte, hpet_id);
1040
1041         modify_irte(irq, &irte);
1042
1043         msg->address_hi = MSI_ADDR_BASE_HI;
1044         msg->data = sub_handle;
1045         msg->address_lo = MSI_ADDR_BASE_LO | MSI_ADDR_IR_EXT_INT |
1046                           MSI_ADDR_IR_SHV |
1047                           MSI_ADDR_IR_INDEX1(ir_index) |
1048                           MSI_ADDR_IR_INDEX2(ir_index);
1049 }
1050
1051 /*
1052  * Map the PCI dev to the corresponding remapping hardware unit
1053  * and allocate 'nvec' consecutive interrupt-remapping table entries
1054  * in it.
1055  */
1056 static int intel_msi_alloc_irq(struct pci_dev *dev, int irq, int nvec)
1057 {
1058         struct intel_iommu *iommu;
1059         int index;
1060
1061         iommu = map_dev_to_ir(dev);
1062         if (!iommu) {
1063                 printk(KERN_ERR
1064                        "Unable to map PCI %s to iommu\n", pci_name(dev));
1065                 return -ENOENT;
1066         }
1067
1068         index = alloc_irte(iommu, irq, nvec);
1069         if (index < 0) {
1070                 printk(KERN_ERR
1071                        "Unable to allocate %d IRTE for PCI %s\n", nvec,
1072                        pci_name(dev));
1073                 return -ENOSPC;
1074         }
1075         return index;
1076 }
1077
1078 static int intel_msi_setup_irq(struct pci_dev *pdev, unsigned int irq,
1079                                int index, int sub_handle)
1080 {
1081         struct intel_iommu *iommu;
1082
1083         iommu = map_dev_to_ir(pdev);
1084         if (!iommu)
1085                 return -ENOENT;
1086         /*
1087          * setup the mapping between the irq and the IRTE
1088          * base index, the sub_handle pointing to the
1089          * appropriate interrupt remap table entry.
1090          */
1091         set_irte_irq(irq, iommu, index, sub_handle);
1092
1093         return 0;
1094 }
1095
1096 static int intel_setup_hpet_msi(unsigned int irq, unsigned int id)
1097 {
1098         struct intel_iommu *iommu = map_hpet_to_ir(id);
1099         int index;
1100
1101         if (!iommu)
1102                 return -1;
1103
1104         index = alloc_irte(iommu, irq, 1);
1105         if (index < 0)
1106                 return -1;
1107
1108         return 0;
1109 }
1110
1111 struct irq_remap_ops intel_irq_remap_ops = {
1112         .supported              = intel_irq_remapping_supported,
1113         .prepare                = dmar_table_init,
1114         .enable                 = intel_enable_irq_remapping,
1115         .disable                = disable_irq_remapping,
1116         .reenable               = reenable_irq_remapping,
1117         .enable_faulting        = enable_drhd_fault_handling,
1118         .setup_ioapic_entry     = intel_setup_ioapic_entry,
1119         .set_affinity           = intel_ioapic_set_affinity,
1120         .free_irq               = free_irte,
1121         .compose_msi_msg        = intel_compose_msi_msg,
1122         .msi_alloc_irq          = intel_msi_alloc_irq,
1123         .msi_setup_irq          = intel_msi_setup_irq,
1124         .setup_hpet_msi         = intel_setup_hpet_msi,
1125 };