drivers/pci/intr_remapping.c

   1 #include <linux/interrupt.h>
   2 #include <linux/dmar.h>
   3 #include <linux/spinlock.h>
   4 #include <linux/jiffies.h>
   5 #include <linux/pci.h>
   6 #include <linux/irq.h>
   7 #include <asm/io_apic.h>
   8 #include <asm/smp.h>
   9 #include <linux/intel-iommu.h>
  10 #include "intr_remapping.h"
  11
  12 static struct ioapic_scope ir_ioapic[MAX_IO_APICS];
  13 static int ir_ioapic_num;
  14 int intr_remapping_enabled;
  15
  16 struct irq_2_iommu {
  17         struct intel_iommu *iommu;
  18         u16 irte_index;
  19         u16 sub_handle;
  20         u8  irte_mask;
  21 };
  22
  23 #ifdef CONFIG_SPARSE_IRQ
  24 static struct irq_2_iommu *get_one_free_irq_2_iommu(int cpu)
  25 {
  26         struct irq_2_iommu *iommu;
  27         int node;
  28
  29         node = cpu_to_node(cpu);
  30
  31         iommu = kzalloc_node(sizeof(*iommu), GFP_ATOMIC, node);
  32         printk(KERN_DEBUG "alloc irq_2_iommu on cpu %d node %d\n", cpu, node);
  33
  34         return iommu;
  35 }
  36
  37 static struct irq_2_iommu *irq_2_iommu(unsigned int irq)
  38 {
  39         struct irq_desc *desc;
  40
  41         desc = irq_to_desc(irq);
  42
  43         if (WARN_ON_ONCE(!desc))
  44                 return NULL;
  45
  46         return desc->irq_2_iommu;
  47 }
  48
  49 static struct irq_2_iommu *irq_2_iommu_alloc_cpu(unsigned int irq, int cpu)
  50 {
  51         struct irq_desc *desc;
  52         struct irq_2_iommu *irq_iommu;
  53
  54         /*
  55          * alloc irq desc if not allocated already.
  56          */
  57         desc = irq_to_desc_alloc_cpu(irq, cpu);
  58         if (!desc) {
  59                 printk(KERN_INFO "can not get irq_desc for %d\n", irq);
  60                 return NULL;
  61         }
  62
  63         irq_iommu = desc->irq_2_iommu;
  64
  65         if (!irq_iommu)
  66                 desc->irq_2_iommu = get_one_free_irq_2_iommu(cpu);
  67
  68         return desc->irq_2_iommu;
  69 }
  70
  71 static struct irq_2_iommu *irq_2_iommu_alloc(unsigned int irq)
  72 {
  73         return irq_2_iommu_alloc_cpu(irq, boot_cpu_id);
  74 }
  75
  76 #else /* !CONFIG_SPARSE_IRQ */
  77
  78 static struct irq_2_iommu irq_2_iommuX[NR_IRQS];
  79
  80 static struct irq_2_iommu *irq_2_iommu(unsigned int irq)
  81 {
  82         if (irq < nr_irqs)
  83                 return &irq_2_iommuX[irq];
  84
  85         return NULL;
  86 }
  87 static struct irq_2_iommu *irq_2_iommu_alloc(unsigned int irq)
  88 {
  89         return irq_2_iommu(irq);
  90 }
  91 #endif
  92
  93 static DEFINE_SPINLOCK(irq_2_ir_lock);
  94
  95 static struct irq_2_iommu *valid_irq_2_iommu(unsigned int irq)
  96 {
  97         struct irq_2_iommu *irq_iommu;
  98
  99         irq_iommu = irq_2_iommu(irq);
 100
 101         if (!irq_iommu)
 102                 return NULL;
 103
 104         if (!irq_iommu->iommu)
 105                 return NULL;
 106
 107         return irq_iommu;
 108 }
 109
 110 int irq_remapped(int irq)
 111 {
 112         return valid_irq_2_iommu(irq) != NULL;
 113 }
 114
 115 int get_irte(int irq, struct irte *entry)
 116 {
 117         int index;
 118         struct irq_2_iommu *irq_iommu;
 119
 120         if (!entry)
 121                 return -1;
 122
 123         spin_lock(&irq_2_ir_lock);
 124         irq_iommu = valid_irq_2_iommu(irq);
 125         if (!irq_iommu) {
 126                 spin_unlock(&irq_2_ir_lock);
 127                 return -1;
 128         }
 129
 130         index = irq_iommu->irte_index + irq_iommu->sub_handle;
 131         *entry = *(irq_iommu->iommu->ir_table->base + index);
 132
 133         spin_unlock(&irq_2_ir_lock);
 134         return 0;
 135 }
 136
 137 int alloc_irte(struct intel_iommu *iommu, int irq, u16 count)
 138 {
 139         struct ir_table *table = iommu->ir_table;
 140         struct irq_2_iommu *irq_iommu;
 141         u16 index, start_index;
 142         unsigned int mask = 0;
 143         int i;
 144
 145         if (!count)
 146                 return -1;
 147
 148 #ifndef CONFIG_SPARSE_IRQ
 149         /* protect irq_2_iommu_alloc later */
 150         if (irq >= nr_irqs)
 151                 return -1;
 152 #endif
 153
 154         /*
 155          * start the IRTE search from index 0.
 156          */
 157         index = start_index = 0;
 158
 159         if (count > 1) {
 160                 count = __roundup_pow_of_two(count);
 161                 mask = ilog2(count);
 162         }
 163
 164         if (mask > ecap_max_handle_mask(iommu->ecap)) {
 165                 printk(KERN_ERR
 166                        "Requested mask %x exceeds the max invalidation handle"
 167                        " mask value %Lx\n", mask,
 168                        ecap_max_handle_mask(iommu->ecap));
 169                 return -1;
 170         }
 171
 172         spin_lock(&irq_2_ir_lock);
 173         do {
 174                 for (i = index; i < index + count; i++)
 175                         if  (table->base[i].present)
 176                                 break;
 177                 /* empty index found */
 178                 if (i == index + count)
 179                         break;
 180
 181                 index = (index + count) % INTR_REMAP_TABLE_ENTRIES;
 182
 183                 if (index == start_index) {
 184                         spin_unlock(&irq_2_ir_lock);
 185                         printk(KERN_ERR "can't allocate an IRTE\n");
 186                         return -1;
 187                 }
 188         } while (1);
 189
 190         for (i = index; i < index + count; i++)
 191                 table->base[i].present = 1;
 192
 193         irq_iommu = irq_2_iommu_alloc(irq);
 194         if (!irq_iommu) {
 195                 spin_unlock(&irq_2_ir_lock);
 196                 printk(KERN_ERR "can't allocate irq_2_iommu\n");
 197                 return -1;
 198         }
 199
 200         irq_iommu->iommu = iommu;
 201         irq_iommu->irte_index =  index;
 202         irq_iommu->sub_handle = 0;
 203         irq_iommu->irte_mask = mask;
 204
 205         spin_unlock(&irq_2_ir_lock);
 206
 207         return index;
 208 }
 209
 210 static void qi_flush_iec(struct intel_iommu *iommu, int index, int mask)
 211 {
 212         struct qi_desc desc;
 213
 214         desc.low = QI_IEC_IIDEX(index) | QI_IEC_TYPE | QI_IEC_IM(mask)
 215                    | QI_IEC_SELECTIVE;
 216         desc.high = 0;
 217
 218         qi_submit_sync(&desc, iommu);
 219 }
 220
 221 int map_irq_to_irte_handle(int irq, u16 *sub_handle)
 222 {
 223         int index;
 224         struct irq_2_iommu *irq_iommu;
 225
 226         spin_lock(&irq_2_ir_lock);
 227         irq_iommu = valid_irq_2_iommu(irq);
 228         if (!irq_iommu) {
 229                 spin_unlock(&irq_2_ir_lock);
 230                 return -1;
 231         }
 232
 233         *sub_handle = irq_iommu->sub_handle;
 234         index = irq_iommu->irte_index;
 235         spin_unlock(&irq_2_ir_lock);
 236         return index;
 237 }
 238
 239 int set_irte_irq(int irq, struct intel_iommu *iommu, u16 index, u16 subhandle)
 240 {
 241         struct irq_2_iommu *irq_iommu;
 242
 243         spin_lock(&irq_2_ir_lock);
 244
 245         irq_iommu = irq_2_iommu_alloc(irq);
 246
 247         if (!irq_iommu) {
 248                 spin_unlock(&irq_2_ir_lock);
 249                 printk(KERN_ERR "can't allocate irq_2_iommu\n");
 250                 return -1;
 251         }
 252
 253         irq_iommu->iommu = iommu;
 254         irq_iommu->irte_index = index;
 255         irq_iommu->sub_handle = subhandle;
 256         irq_iommu->irte_mask = 0;
 257
 258         spin_unlock(&irq_2_ir_lock);
 259
 260         return 0;
 261 }
 262
 263 int clear_irte_irq(int irq, struct intel_iommu *iommu, u16 index)
 264 {
 265         struct irq_2_iommu *irq_iommu;
 266
 267         spin_lock(&irq_2_ir_lock);
 268         irq_iommu = valid_irq_2_iommu(irq);
 269         if (!irq_iommu) {
 270                 spin_unlock(&irq_2_ir_lock);
 271                 return -1;
 272         }
 273
 274         irq_iommu->iommu = NULL;
 275         irq_iommu->irte_index = 0;
 276         irq_iommu->sub_handle = 0;
 277         irq_2_iommu(irq)->irte_mask = 0;
 278
 279         spin_unlock(&irq_2_ir_lock);
 280
 281         return 0;
 282 }
 283
 284 int modify_irte(int irq, struct irte *irte_modified)
 285 {
 286         int index;
 287         struct irte *irte;
 288         struct intel_iommu *iommu;
 289         struct irq_2_iommu *irq_iommu;
 290
 291         spin_lock(&irq_2_ir_lock);
 292         irq_iommu = valid_irq_2_iommu(irq);
 293         if (!irq_iommu) {
 294                 spin_unlock(&irq_2_ir_lock);
 295                 return -1;
 296         }
 297
 298         iommu = irq_iommu->iommu;
 299
 300         index = irq_iommu->irte_index + irq_iommu->sub_handle;
 301         irte = &iommu->ir_table->base[index];
 302
 303         set_64bit((unsigned long *)irte, irte_modified->low | (1 << 1));
 304         __iommu_flush_cache(iommu, irte, sizeof(*irte));
 305
 306         qi_flush_iec(iommu, index, 0);
 307
 308         spin_unlock(&irq_2_ir_lock);
 309         return 0;
 310 }
 311
 312 int flush_irte(int irq)
 313 {
 314         int index;
 315         struct intel_iommu *iommu;
 316         struct irq_2_iommu *irq_iommu;
 317
 318         spin_lock(&irq_2_ir_lock);
 319         irq_iommu = valid_irq_2_iommu(irq);
 320         if (!irq_iommu) {
 321                 spin_unlock(&irq_2_ir_lock);
 322                 return -1;
 323         }
 324
 325         iommu = irq_iommu->iommu;
 326
 327         index = irq_iommu->irte_index + irq_iommu->sub_handle;
 328
 329         qi_flush_iec(iommu, index, irq_iommu->irte_mask);
 330         spin_unlock(&irq_2_ir_lock);
 331
 332         return 0;
 333 }
 334
 335 struct intel_iommu *map_ioapic_to_ir(int apic)
 336 {
 337         int i;
 338
 339         for (i = 0; i < MAX_IO_APICS; i++)
 340                 if (ir_ioapic[i].id == apic)
 341                         return ir_ioapic[i].iommu;
 342         return NULL;
 343 }
 344
 345 struct intel_iommu *map_dev_to_ir(struct pci_dev *dev)
 346 {
 347         struct dmar_drhd_unit *drhd;
 348
 349         drhd = dmar_find_matched_drhd_unit(dev);
 350         if (!drhd)
 351                 return NULL;
 352
 353         return drhd->iommu;
 354 }
 355
 356 int free_irte(int irq)
 357 {
 358         int index, i;
 359         struct irte *irte;
 360         struct intel_iommu *iommu;
 361         struct irq_2_iommu *irq_iommu;
 362
 363         spin_lock(&irq_2_ir_lock);
 364         irq_iommu = valid_irq_2_iommu(irq);
 365         if (!irq_iommu) {
 366                 spin_unlock(&irq_2_ir_lock);
 367                 return -1;
 368         }
 369
 370         iommu = irq_iommu->iommu;
 371
 372         index = irq_iommu->irte_index + irq_iommu->sub_handle;
 373         irte = &iommu->ir_table->base[index];
 374
 375         if (!irq_iommu->sub_handle) {
 376                 for (i = 0; i < (1 << irq_iommu->irte_mask); i++)
 377                         set_64bit((unsigned long *)irte, 0);
 378                 qi_flush_iec(iommu, index, irq_iommu->irte_mask);
 379         }
 380
 381         irq_iommu->iommu = NULL;
 382         irq_iommu->irte_index = 0;
 383         irq_iommu->sub_handle = 0;
 384         irq_iommu->irte_mask = 0;
 385
 386         spin_unlock(&irq_2_ir_lock);
 387
 388         return 0;
 389 }
 390
 391 static void iommu_set_intr_remapping(struct intel_iommu *iommu, int mode)
 392 {
 393         u64 addr;
 394         u32 cmd, sts;
 395         unsigned long flags;
 396
 397         addr = virt_to_phys((void *)iommu->ir_table->base);
 398
 399         spin_lock_irqsave(&iommu->register_lock, flags);
 400
 401         dmar_writeq(iommu->reg + DMAR_IRTA_REG,
 402                     (addr) | IR_X2APIC_MODE(mode) | INTR_REMAP_TABLE_REG_SIZE);
 403
 404         /* Set interrupt-remapping table pointer */
 405         cmd = iommu->gcmd | DMA_GCMD_SIRTP;
 406         writel(cmd, iommu->reg + DMAR_GCMD_REG);
 407
 408         IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
 409                       readl, (sts & DMA_GSTS_IRTPS), sts);
 410         spin_unlock_irqrestore(&iommu->register_lock, flags);
 411
 412         /*
 413          * global invalidation of interrupt entry cache before enabling
 414          * interrupt-remapping.
 415          */
 416         qi_global_iec(iommu);
 417
 418         spin_lock_irqsave(&iommu->register_lock, flags);
 419
 420         /* Enable interrupt-remapping */
 421         cmd = iommu->gcmd | DMA_GCMD_IRE;
 422         iommu->gcmd |= DMA_GCMD_IRE;
 423         writel(cmd, iommu->reg + DMAR_GCMD_REG);
 424
 425         IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG,
 426                       readl, (sts & DMA_GSTS_IRES), sts);
 427
 428         spin_unlock_irqrestore(&iommu->register_lock, flags);
 429 }
 430
 431
 432 static int setup_intr_remapping(struct intel_iommu *iommu, int mode)
 433 {
 434         struct ir_table *ir_table;
 435         struct page *pages;
 436
 437         ir_table = iommu->ir_table = kzalloc(sizeof(struct ir_table),
 438                                              GFP_KERNEL);
 439
 440         if (!iommu->ir_table)
 441                 return -ENOMEM;
 442
 443         pages = alloc_pages(GFP_KERNEL | __GFP_ZERO, INTR_REMAP_PAGE_ORDER);
 444
 445         if (!pages) {
 446                 printk(KERN_ERR "failed to allocate pages of order %d\n",
 447                        INTR_REMAP_PAGE_ORDER);
 448                 kfree(iommu->ir_table);
 449                 return -ENOMEM;
 450         }
 451
 452         ir_table->base = page_address(pages);
 453
 454         iommu_set_intr_remapping(iommu, mode);
 455         return 0;
 456 }
 457
 458 int __init enable_intr_remapping(int eim)
 459 {
 460         struct dmar_drhd_unit *drhd;
 461         int setup = 0;
 462
 463         /*
 464          * check for the Interrupt-remapping support
 465          */
 466         for_each_drhd_unit(drhd) {
 467                 struct intel_iommu *iommu = drhd->iommu;
 468
 469                 if (!ecap_ir_support(iommu->ecap))
 470                         continue;
 471
 472                 if (eim && !ecap_eim_support(iommu->ecap)) {
 473                         printk(KERN_INFO "DRHD %Lx: EIM not supported by DRHD, "
 474                                " ecap %Lx\n", drhd->reg_base_addr, iommu->ecap);
 475                         return -1;
 476                 }
 477         }
 478
 479         /*
 480          * Enable queued invalidation for all the DRHD's.
 481          */
 482         for_each_drhd_unit(drhd) {
 483                 int ret;
 484                 struct intel_iommu *iommu = drhd->iommu;
 485                 ret = dmar_enable_qi(iommu);
 486
 487                 if (ret) {
 488                         printk(KERN_ERR "DRHD %Lx: failed to enable queued, "
 489                                " invalidation, ecap %Lx, ret %d\n",
 490                                drhd->reg_base_addr, iommu->ecap, ret);
 491                         return -1;
 492                 }
 493         }
 494
 495         /*
 496          * Setup Interrupt-remapping for all the DRHD's now.
 497          */
 498         for_each_drhd_unit(drhd) {
 499                 struct intel_iommu *iommu = drhd->iommu;
 500
 501                 if (!ecap_ir_support(iommu->ecap))
 502                         continue;
 503
 504                 if (setup_intr_remapping(iommu, eim))
 505                         goto error;
 506
 507                 setup = 1;
 508         }
 509
 510         if (!setup)
 511                 goto error;
 512
 513         intr_remapping_enabled = 1;
 514
 515         return 0;
 516
 517 error:
 518         /*
 519          * handle error condition gracefully here!
 520          */
 521         return -1;
 522 }
 523
 524 static int ir_parse_ioapic_scope(struct acpi_dmar_header *header,
 525                                  struct intel_iommu *iommu)
 526 {
 527         struct acpi_dmar_hardware_unit *drhd;
 528         struct acpi_dmar_device_scope *scope;
 529         void *start, *end;
 530
 531         drhd = (struct acpi_dmar_hardware_unit *)header;
 532
 533         start = (void *)(drhd + 1);
 534         end = ((void *)drhd) + header->length;
 535
 536         while (start < end) {
 537                 scope = start;
 538                 if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_IOAPIC) {
 539                         if (ir_ioapic_num == MAX_IO_APICS) {
 540                                 printk(KERN_WARNING "Exceeded Max IO APICS\n");
 541                                 return -1;
 542                         }
 543
 544                         printk(KERN_INFO "IOAPIC id %d under DRHD base"
 545                                " 0x%Lx\n", scope->enumeration_id,
 546                                drhd->address);
 547
 548                         ir_ioapic[ir_ioapic_num].iommu = iommu;
 549                         ir_ioapic[ir_ioapic_num].id = scope->enumeration_id;
 550                         ir_ioapic_num++;
 551                 }
 552                 start += scope->length;
 553         }
 554
 555         return 0;
 556 }
 557
 558 /*
 559  * Finds the assocaition between IOAPIC's and its Interrupt-remapping
 560  * hardware unit.
 561  */
 562 int __init parse_ioapics_under_ir(void)
 563 {
 564         struct dmar_drhd_unit *drhd;
 565         int ir_supported = 0;
 566
 567         for_each_drhd_unit(drhd) {
 568                 struct intel_iommu *iommu = drhd->iommu;
 569
 570                 if (ecap_ir_support(iommu->ecap)) {
 571                         if (ir_parse_ioapic_scope(drhd->hdr, iommu))
 572                                 return -1;
 573
 574                         ir_supported = 1;
 575                 }
 576         }
 577
 578         if (ir_supported && ir_ioapic_num != nr_ioapics) {
 579                 printk(KERN_WARNING
 580                        "Not all IO-APIC's listed under remapping hardware\n");
 581                 return -1;
 582         }
 583
 584         return ir_supported;
 585 }