linux/drivers/iommu/intel_irq_remapping.c
<<
>>
Prefs
   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
  20struct 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
  27struct 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
  37static struct ioapic_scope ir_ioapic[MAX_IO_APICS];
  38static struct hpet_scope ir_hpet[MAX_HPET_TBS];
  39static int ir_ioapic_num, ir_hpet_num;
  40
  41static DEFINE_RAW_SPINLOCK(irq_2_ir_lock);
  42
  43static 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
  49int 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
  67static 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
 130static 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
 141static 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
 157static 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
 179static 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
 207static 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
 217static 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
 227static 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
 238static 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
 261static 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 */
 309static 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
 319static 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
 344static 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
 374static 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
 401static 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
 452static 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 */
 482static 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
 508end:
 509        raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
 510}
 511
 512static 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
 521static int __init intel_irq_remapping_supported(void)
 522{
 523        struct dmar_drhd_unit *drhd;
 524
 525        if (disable_irq_remap)
 526                return 0;
 527
 528        if (!dmar_ir_support())
 529                return 0;
 530
 531        for_each_drhd_unit(drhd) {
 532                struct intel_iommu *iommu = drhd->iommu;
 533
 534                if (!ecap_ir_support(iommu->ecap))
 535                        return 0;
 536        }
 537
 538        return 1;
 539}
 540
 541static int __init intel_enable_irq_remapping(void)
 542{
 543        struct dmar_drhd_unit *drhd;
 544        bool x2apic_present;
 545        int setup = 0;
 546        int eim = 0;
 547
 548        x2apic_present = x2apic_supported();
 549
 550        if (parse_ioapics_under_ir() != 1) {
 551                printk(KERN_INFO "Not enable interrupt remapping\n");
 552                goto error;
 553        }
 554
 555        if (x2apic_present) {
 556                eim = !dmar_x2apic_optout();
 557                if (!eim)
 558                        printk(KERN_WARNING
 559                                "Your BIOS is broken and requested that x2apic be disabled.\n"
 560                                "This will slightly decrease performance.\n"
 561                                "Use 'intremap=no_x2apic_optout' to override BIOS request.\n");
 562        }
 563
 564        for_each_drhd_unit(drhd) {
 565                struct intel_iommu *iommu = drhd->iommu;
 566
 567                /*
 568                 * If the queued invalidation is already initialized,
 569                 * shouldn't disable it.
 570                 */
 571                if (iommu->qi)
 572                        continue;
 573
 574                /*
 575                 * Clear previous faults.
 576                 */
 577                dmar_fault(-1, iommu);
 578
 579                /*
 580                 * Disable intr remapping and queued invalidation, if already
 581                 * enabled prior to OS handover.
 582                 */
 583                iommu_disable_irq_remapping(iommu);
 584
 585                dmar_disable_qi(iommu);
 586        }
 587
 588        /*
 589         * check for the Interrupt-remapping support
 590         */
 591        for_each_drhd_unit(drhd) {
 592                struct intel_iommu *iommu = drhd->iommu;
 593
 594                if (!ecap_ir_support(iommu->ecap))
 595                        continue;
 596
 597                if (eim && !ecap_eim_support(iommu->ecap)) {
 598                        printk(KERN_INFO "DRHD %Lx: EIM not supported by DRHD, "
 599                               " ecap %Lx\n", drhd->reg_base_addr, iommu->ecap);
 600                        goto error;
 601                }
 602        }
 603
 604        /*
 605         * Enable queued invalidation for all the DRHD's.
 606         */
 607        for_each_drhd_unit(drhd) {
 608                int ret;
 609                struct intel_iommu *iommu = drhd->iommu;
 610                ret = dmar_enable_qi(iommu);
 611
 612                if (ret) {
 613                        printk(KERN_ERR "DRHD %Lx: failed to enable queued, "
 614                               " invalidation, ecap %Lx, ret %d\n",
 615                               drhd->reg_base_addr, iommu->ecap, ret);
 616                        goto error;
 617                }
 618        }
 619
 620        /*
 621         * Setup Interrupt-remapping for all the DRHD's now.
 622         */
 623        for_each_drhd_unit(drhd) {
 624                struct intel_iommu *iommu = drhd->iommu;
 625
 626                if (!ecap_ir_support(iommu->ecap))
 627                        continue;
 628
 629                if (intel_setup_irq_remapping(iommu, eim))
 630                        goto error;
 631
 632                setup = 1;
 633        }
 634
 635        if (!setup)
 636                goto error;
 637
 638        irq_remapping_enabled = 1;
 639
 640        /*
 641         * VT-d has a different layout for IO-APIC entries when
 642         * interrupt remapping is enabled. So it needs a special routine
 643         * to print IO-APIC entries for debugging purposes too.
 644         */
 645        x86_io_apic_ops.print_entries = intel_ir_io_apic_print_entries;
 646
 647        pr_info("Enabled IRQ remapping in %s mode\n", eim ? "x2apic" : "xapic");
 648
 649        return eim ? IRQ_REMAP_X2APIC_MODE : IRQ_REMAP_XAPIC_MODE;
 650
 651error:
 652        /*
 653         * handle error condition gracefully here!
 654         */
 655
 656        if (x2apic_present)
 657                WARN(1, KERN_WARNING
 658                        "Failed to enable irq remapping.  You are vulnerable to irq-injection attacks.\n");
 659
 660        return -1;
 661}
 662
 663static void ir_parse_one_hpet_scope(struct acpi_dmar_device_scope *scope,
 664                                      struct intel_iommu *iommu)
 665{
 666        struct acpi_dmar_pci_path *path;
 667        u8 bus;
 668        int count;
 669
 670        bus = scope->bus;
 671        path = (struct acpi_dmar_pci_path *)(scope + 1);
 672        count = (scope->length - sizeof(struct acpi_dmar_device_scope))
 673                / sizeof(struct acpi_dmar_pci_path);
 674
 675        while (--count > 0) {
 676                /*
 677                 * Access PCI directly due to the PCI
 678                 * subsystem isn't initialized yet.
 679                 */
 680                bus = read_pci_config_byte(bus, path->dev, path->fn,
 681                                           PCI_SECONDARY_BUS);
 682                path++;
 683        }
 684        ir_hpet[ir_hpet_num].bus   = bus;
 685        ir_hpet[ir_hpet_num].devfn = PCI_DEVFN(path->dev, path->fn);
 686        ir_hpet[ir_hpet_num].iommu = iommu;
 687        ir_hpet[ir_hpet_num].id    = scope->enumeration_id;
 688        ir_hpet_num++;
 689}
 690
 691static void ir_parse_one_ioapic_scope(struct acpi_dmar_device_scope *scope,
 692                                      struct intel_iommu *iommu)
 693{
 694        struct acpi_dmar_pci_path *path;
 695        u8 bus;
 696        int count;
 697
 698        bus = scope->bus;
 699        path = (struct acpi_dmar_pci_path *)(scope + 1);
 700        count = (scope->length - sizeof(struct acpi_dmar_device_scope))
 701                / sizeof(struct acpi_dmar_pci_path);
 702
 703        while (--count > 0) {
 704                /*
 705                 * Access PCI directly due to the PCI
 706                 * subsystem isn't initialized yet.
 707                 */
 708                bus = read_pci_config_byte(bus, path->dev, path->fn,
 709                                           PCI_SECONDARY_BUS);
 710                path++;
 711        }
 712
 713        ir_ioapic[ir_ioapic_num].bus   = bus;
 714        ir_ioapic[ir_ioapic_num].devfn = PCI_DEVFN(path->dev, path->fn);
 715        ir_ioapic[ir_ioapic_num].iommu = iommu;
 716        ir_ioapic[ir_ioapic_num].id    = scope->enumeration_id;
 717        ir_ioapic_num++;
 718}
 719
 720static int ir_parse_ioapic_hpet_scope(struct acpi_dmar_header *header,
 721                                      struct intel_iommu *iommu)
 722{
 723        struct acpi_dmar_hardware_unit *drhd;
 724        struct acpi_dmar_device_scope *scope;
 725        void *start, *end;
 726
 727        drhd = (struct acpi_dmar_hardware_unit *)header;
 728
 729        start = (void *)(drhd + 1);
 730        end = ((void *)drhd) + header->length;
 731
 732        while (start < end) {
 733                scope = start;
 734                if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_IOAPIC) {
 735                        if (ir_ioapic_num == MAX_IO_APICS) {
 736                                printk(KERN_WARNING "Exceeded Max IO APICS\n");
 737                                return -1;
 738                        }
 739
 740                        printk(KERN_INFO "IOAPIC id %d under DRHD base "
 741                               " 0x%Lx IOMMU %d\n", scope->enumeration_id,
 742                               drhd->address, iommu->seq_id);
 743
 744                        ir_parse_one_ioapic_scope(scope, iommu);
 745                } else if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_HPET) {
 746                        if (ir_hpet_num == MAX_HPET_TBS) {
 747                                printk(KERN_WARNING "Exceeded Max HPET blocks\n");
 748                                return -1;
 749                        }
 750
 751                        printk(KERN_INFO "HPET id %d under DRHD base"
 752                               " 0x%Lx\n", scope->enumeration_id,
 753                               drhd->address);
 754
 755                        ir_parse_one_hpet_scope(scope, iommu);
 756                }
 757                start += scope->length;
 758        }
 759
 760        return 0;
 761}
 762
 763/*
 764 * Finds the assocaition between IOAPIC's and its Interrupt-remapping
 765 * hardware unit.
 766 */
 767int __init parse_ioapics_under_ir(void)
 768{
 769        struct dmar_drhd_unit *drhd;
 770        int ir_supported = 0;
 771        int ioapic_idx;
 772
 773        for_each_drhd_unit(drhd) {
 774                struct intel_iommu *iommu = drhd->iommu;
 775
 776                if (ecap_ir_support(iommu->ecap)) {
 777                        if (ir_parse_ioapic_hpet_scope(drhd->hdr, iommu))
 778                                return -1;
 779
 780                        ir_supported = 1;
 781                }
 782        }
 783
 784        if (!ir_supported)
 785                return 0;
 786
 787        for (ioapic_idx = 0; ioapic_idx < nr_ioapics; ioapic_idx++) {
 788                int ioapic_id = mpc_ioapic_id(ioapic_idx);
 789                if (!map_ioapic_to_ir(ioapic_id)) {
 790                        pr_err(FW_BUG "ioapic %d has no mapping iommu, "
 791                               "interrupt remapping will be disabled\n",
 792                               ioapic_id);
 793                        return -1;
 794                }
 795        }
 796
 797        return 1;
 798}
 799
 800int __init ir_dev_scope_init(void)
 801{
 802        if (!irq_remapping_enabled)
 803                return 0;
 804
 805        return dmar_dev_scope_init();
 806}
 807rootfs_initcall(ir_dev_scope_init);
 808
 809static void disable_irq_remapping(void)
 810{
 811        struct dmar_drhd_unit *drhd;
 812        struct intel_iommu *iommu = NULL;
 813
 814        /*
 815         * Disable Interrupt-remapping for all the DRHD's now.
 816         */
 817        for_each_iommu(iommu, drhd) {
 818                if (!ecap_ir_support(iommu->ecap))
 819                        continue;
 820
 821                iommu_disable_irq_remapping(iommu);
 822        }
 823}
 824
 825static int reenable_irq_remapping(int eim)
 826{
 827        struct dmar_drhd_unit *drhd;
 828        int setup = 0;
 829        struct intel_iommu *iommu = NULL;
 830
 831        for_each_iommu(iommu, drhd)
 832                if (iommu->qi)
 833                        dmar_reenable_qi(iommu);
 834
 835        /*
 836         * Setup Interrupt-remapping for all the DRHD's now.
 837         */
 838        for_each_iommu(iommu, drhd) {
 839                if (!ecap_ir_support(iommu->ecap))
 840                        continue;
 841
 842                /* Set up interrupt remapping for iommu.*/
 843                iommu_set_irq_remapping(iommu, eim);
 844                setup = 1;
 845        }
 846
 847        if (!setup)
 848                goto error;
 849
 850        return 0;
 851
 852error:
 853        /*
 854         * handle error condition gracefully here!
 855         */
 856        return -1;
 857}
 858
 859static void prepare_irte(struct irte *irte, int vector,
 860                         unsigned int dest)
 861{
 862        memset(irte, 0, sizeof(*irte));
 863
 864        irte->present = 1;
 865        irte->dst_mode = apic->irq_dest_mode;
 866        /*
 867         * Trigger mode in the IRTE will always be edge, and for IO-APIC, the
 868         * actual level or edge trigger will be setup in the IO-APIC
 869         * RTE. This will help simplify level triggered irq migration.
 870         * For more details, see the comments (in io_apic.c) explainig IO-APIC
 871         * irq migration in the presence of interrupt-remapping.
 872        */
 873        irte->trigger_mode = 0;
 874        irte->dlvry_mode = apic->irq_delivery_mode;
 875        irte->vector = vector;
 876        irte->dest_id = IRTE_DEST(dest);
 877        irte->redir_hint = 1;
 878}
 879
 880static int intel_setup_ioapic_entry(int irq,
 881                                    struct IO_APIC_route_entry *route_entry,
 882                                    unsigned int destination, int vector,
 883                                    struct io_apic_irq_attr *attr)
 884{
 885        int ioapic_id = mpc_ioapic_id(attr->ioapic);
 886        struct intel_iommu *iommu = map_ioapic_to_ir(ioapic_id);
 887        struct IR_IO_APIC_route_entry *entry;
 888        struct irte irte;
 889        int index;
 890
 891        if (!iommu) {
 892                pr_warn("No mapping iommu for ioapic %d\n", ioapic_id);
 893                return -ENODEV;
 894        }
 895
 896        entry = (struct IR_IO_APIC_route_entry *)route_entry;
 897
 898        index = alloc_irte(iommu, irq, 1);
 899        if (index < 0) {
 900                pr_warn("Failed to allocate IRTE for ioapic %d\n", ioapic_id);
 901                return -ENOMEM;
 902        }
 903
 904        prepare_irte(&irte, vector, destination);
 905
 906        /* Set source-id of interrupt request */
 907        set_ioapic_sid(&irte, ioapic_id);
 908
 909        modify_irte(irq, &irte);
 910
 911        apic_printk(APIC_VERBOSE, KERN_DEBUG "IOAPIC[%d]: "
 912                "Set IRTE entry (P:%d FPD:%d Dst_Mode:%d "
 913                "Redir_hint:%d Trig_Mode:%d Dlvry_Mode:%X "
 914                "Avail:%X Vector:%02X Dest:%08X "
 915                "SID:%04X SQ:%X SVT:%X)\n",
 916                attr->ioapic, irte.present, irte.fpd, irte.dst_mode,
 917                irte.redir_hint, irte.trigger_mode, irte.dlvry_mode,
 918                irte.avail, irte.vector, irte.dest_id,
 919                irte.sid, irte.sq, irte.svt);
 920
 921        memset(entry, 0, sizeof(*entry));
 922
 923        entry->index2   = (index >> 15) & 0x1;
 924        entry->zero     = 0;
 925        entry->format   = 1;
 926        entry->index    = (index & 0x7fff);
 927        /*
 928         * IO-APIC RTE will be configured with virtual vector.
 929         * irq handler will do the explicit EOI to the io-apic.
 930         */
 931        entry->vector   = attr->ioapic_pin;
 932        entry->mask     = 0;                    /* enable IRQ */
 933        entry->trigger  = attr->trigger;
 934        entry->polarity = attr->polarity;
 935
 936        /* Mask level triggered irqs.
 937         * Use IRQ_DELAYED_DISABLE for edge triggered irqs.
 938         */
 939        if (attr->trigger)
 940                entry->mask = 1;
 941
 942        return 0;
 943}
 944
 945/*
 946 * Migrate the IO-APIC irq in the presence of intr-remapping.
 947 *
 948 * For both level and edge triggered, irq migration is a simple atomic
 949 * update(of vector and cpu destination) of IRTE and flush the hardware cache.
 950 *
 951 * For level triggered, we eliminate the io-apic RTE modification (with the
 952 * updated vector information), by using a virtual vector (io-apic pin number).
 953 * Real vector that is used for interrupting cpu will be coming from
 954 * the interrupt-remapping table entry.
 955 *
 956 * As the migration is a simple atomic update of IRTE, the same mechanism
 957 * is used to migrate MSI irq's in the presence of interrupt-remapping.
 958 */
 959static int
 960intel_ioapic_set_affinity(struct irq_data *data, const struct cpumask *mask,
 961                          bool force)
 962{
 963        struct irq_cfg *cfg = data->chip_data;
 964        unsigned int dest, irq = data->irq;
 965        struct irte irte;
 966        int err;
 967
 968        if (!config_enabled(CONFIG_SMP))
 969                return -EINVAL;
 970
 971        if (!cpumask_intersects(mask, cpu_online_mask))
 972                return -EINVAL;
 973
 974        if (get_irte(irq, &irte))
 975                return -EBUSY;
 976
 977        err = assign_irq_vector(irq, cfg, mask);
 978        if (err)
 979                return err;
 980
 981        err = apic->cpu_mask_to_apicid_and(cfg->domain, mask, &dest);
 982        if (err) {
 983                if (assign_irq_vector(irq, cfg, data->affinity))
 984                        pr_err("Failed to recover vector for irq %d\n", irq);
 985                return err;
 986        }
 987
 988        irte.vector = cfg->vector;
 989        irte.dest_id = IRTE_DEST(dest);
 990
 991        /*
 992         * Atomically updates the IRTE with the new destination, vector
 993         * and flushes the interrupt entry cache.
 994         */
 995        modify_irte(irq, &irte);
 996
 997        /*
 998         * After this point, all the interrupts will start arriving
 999         * at the new destination. So, time to cleanup the previous
1000         * vector allocation.
1001         */
1002        if (cfg->move_in_progress)
1003                send_cleanup_vector(cfg);
1004
1005        cpumask_copy(data->affinity, mask);
1006        return 0;
1007}
1008
1009static void intel_compose_msi_msg(struct pci_dev *pdev,
1010                                  unsigned int irq, unsigned int dest,
1011                                  struct msi_msg *msg, u8 hpet_id)
1012{
1013        struct irq_cfg *cfg;
1014        struct irte irte;
1015        u16 sub_handle = 0;
1016        int ir_index;
1017
1018        cfg = irq_get_chip_data(irq);
1019
1020        ir_index = map_irq_to_irte_handle(irq, &sub_handle);
1021        BUG_ON(ir_index == -1);
1022
1023        prepare_irte(&irte, cfg->vector, dest);
1024
1025        /* Set source-id of interrupt request */
1026        if (pdev)
1027                set_msi_sid(&irte, pdev);
1028        else
1029                set_hpet_sid(&irte, hpet_id);
1030
1031        modify_irte(irq, &irte);
1032
1033        msg->address_hi = MSI_ADDR_BASE_HI;
1034        msg->data = sub_handle;
1035        msg->address_lo = MSI_ADDR_BASE_LO | MSI_ADDR_IR_EXT_INT |
1036                          MSI_ADDR_IR_SHV |
1037                          MSI_ADDR_IR_INDEX1(ir_index) |
1038                          MSI_ADDR_IR_INDEX2(ir_index);
1039}
1040
1041/*
1042 * Map the PCI dev to the corresponding remapping hardware unit
1043 * and allocate 'nvec' consecutive interrupt-remapping table entries
1044 * in it.
1045 */
1046static int intel_msi_alloc_irq(struct pci_dev *dev, int irq, int nvec)
1047{
1048        struct intel_iommu *iommu;
1049        int index;
1050
1051        iommu = map_dev_to_ir(dev);
1052        if (!iommu) {
1053                printk(KERN_ERR
1054                       "Unable to map PCI %s to iommu\n", pci_name(dev));
1055                return -ENOENT;
1056        }
1057
1058        index = alloc_irte(iommu, irq, nvec);
1059        if (index < 0) {
1060                printk(KERN_ERR
1061                       "Unable to allocate %d IRTE for PCI %s\n", nvec,
1062                       pci_name(dev));
1063                return -ENOSPC;
1064        }
1065        return index;
1066}
1067
1068static int intel_msi_setup_irq(struct pci_dev *pdev, unsigned int irq,
1069                               int index, int sub_handle)
1070{
1071        struct intel_iommu *iommu;
1072
1073        iommu = map_dev_to_ir(pdev);
1074        if (!iommu)
1075                return -ENOENT;
1076        /*
1077         * setup the mapping between the irq and the IRTE
1078         * base index, the sub_handle pointing to the
1079         * appropriate interrupt remap table entry.
1080         */
1081        set_irte_irq(irq, iommu, index, sub_handle);
1082
1083        return 0;
1084}
1085
1086static int intel_setup_hpet_msi(unsigned int irq, unsigned int id)
1087{
1088        struct intel_iommu *iommu = map_hpet_to_ir(id);
1089        int index;
1090
1091        if (!iommu)
1092                return -1;
1093
1094        index = alloc_irte(iommu, irq, 1);
1095        if (index < 0)
1096                return -1;
1097
1098        return 0;
1099}
1100
1101struct irq_remap_ops intel_irq_remap_ops = {
1102        .supported              = intel_irq_remapping_supported,
1103        .prepare                = dmar_table_init,
1104        .enable                 = intel_enable_irq_remapping,
1105        .disable                = disable_irq_remapping,
1106        .reenable               = reenable_irq_remapping,
1107        .enable_faulting        = enable_drhd_fault_handling,
1108        .setup_ioapic_entry     = intel_setup_ioapic_entry,
1109        .set_affinity           = intel_ioapic_set_affinity,
1110        .free_irq               = free_irte,
1111        .compose_msi_msg        = intel_compose_msi_msg,
1112        .msi_alloc_irq          = intel_msi_alloc_irq,
1113        .msi_setup_irq          = intel_msi_setup_irq,
1114        .setup_hpet_msi         = intel_setup_hpet_msi,
1115};
1116
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.