drivers/pci/host/vmd.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Volume Management Device driver
   4  * Copyright (c) 2015, Intel Corporation.
   5  */
   6
   7 #include <linux/device.h>
   8 #include <linux/interrupt.h>
   9 #include <linux/irq.h>
  10 #include <linux/kernel.h>
  11 #include <linux/module.h>
  12 #include <linux/msi.h>
  13 #include <linux/pci.h>
  14 #include <linux/srcu.h>
  15 #include <linux/rculist.h>
  16 #include <linux/rcupdate.h>
  17
  18 #include <asm/irqdomain.h>
  19 #include <asm/device.h>
  20 #include <asm/msi.h>
  21 #include <asm/msidef.h>
  22
  23 #define VMD_CFGBAR      0
  24 #define VMD_MEMBAR1     2
  25 #define VMD_MEMBAR2     4
  26
  27 /*
  28  * Lock for manipulating VMD IRQ lists.
  29  */
  30 static DEFINE_RAW_SPINLOCK(list_lock);
  31
  32 /**
  33  * struct vmd_irq - private data to map driver IRQ to the VMD shared vector
  34  * @node:       list item for parent traversal.
  35  * @irq:        back pointer to parent.
  36  * @enabled:    true if driver enabled IRQ
  37  * @virq:       the virtual IRQ value provided to the requesting driver.
  38  *
  39  * Every MSI/MSI-X IRQ requested for a device in a VMD domain will be mapped to
  40  * a VMD IRQ using this structure.
  41  */
  42 struct vmd_irq {
  43         struct list_head        node;
  44         struct vmd_irq_list     *irq;
  45         bool                    enabled;
  46         unsigned int            virq;
  47 };
  48
  49 /**
  50  * struct vmd_irq_list - list of driver requested IRQs mapping to a VMD vector
  51  * @irq_list:   the list of irq's the VMD one demuxes to.
  52  * @srcu:       SRCU struct for local synchronization.
  53  * @count:      number of child IRQs assigned to this vector; used to track
  54  *              sharing.
  55  */
  56 struct vmd_irq_list {
  57         struct list_head        irq_list;
  58         struct srcu_struct      srcu;
  59         unsigned int            count;
  60 };
  61
  62 struct vmd_dev {
  63         struct pci_dev          *dev;
  64
  65         spinlock_t              cfg_lock;
  66         char __iomem            *cfgbar;
  67
  68         int msix_count;
  69         struct vmd_irq_list     *irqs;
  70
  71         struct pci_sysdata      sysdata;
  72         struct resource         resources[3];
  73         struct irq_domain       *irq_domain;
  74         struct pci_bus          *bus;
  75
  76 #ifdef CONFIG_X86_DEV_DMA_OPS
  77         struct dma_map_ops      dma_ops;
  78         struct dma_domain       dma_domain;
  79 #endif
  80 };
  81
  82 static inline struct vmd_dev *vmd_from_bus(struct pci_bus *bus)
  83 {
  84         return container_of(bus->sysdata, struct vmd_dev, sysdata);
  85 }
  86
  87 static inline unsigned int index_from_irqs(struct vmd_dev *vmd,
  88                                            struct vmd_irq_list *irqs)
  89 {
  90         return irqs - vmd->irqs;
  91 }
  92
  93 /*
  94  * Drivers managing a device in a VMD domain allocate their own IRQs as before,
  95  * but the MSI entry for the hardware it's driving will be programmed with a
  96  * destination ID for the VMD MSI-X table.  The VMD muxes interrupts in its
  97  * domain into one of its own, and the VMD driver de-muxes these for the
  98  * handlers sharing that VMD IRQ.  The vmd irq_domain provides the operations
  99  * and irq_chip to set this up.
 100  */
 101 static void vmd_compose_msi_msg(struct irq_data *data, struct msi_msg *msg)
 102 {
 103         struct vmd_irq *vmdirq = data->chip_data;
 104         struct vmd_irq_list *irq = vmdirq->irq;
 105         struct vmd_dev *vmd = irq_data_get_irq_handler_data(data);
 106
 107         msg->address_hi = MSI_ADDR_BASE_HI;
 108         msg->address_lo = MSI_ADDR_BASE_LO |
 109                           MSI_ADDR_DEST_ID(index_from_irqs(vmd, irq));
 110         msg->data = 0;
 111 }
 112
 113 /*
 114  * We rely on MSI_FLAG_USE_DEF_CHIP_OPS to set the IRQ mask/unmask ops.
 115  */
 116 static void vmd_irq_enable(struct irq_data *data)
 117 {
 118         struct vmd_irq *vmdirq = data->chip_data;
 119         unsigned long flags;
 120
 121         raw_spin_lock_irqsave(&list_lock, flags);
 122         WARN_ON(vmdirq->enabled);
 123         list_add_tail_rcu(&vmdirq->node, &vmdirq->irq->irq_list);
 124         vmdirq->enabled = true;
 125         raw_spin_unlock_irqrestore(&list_lock, flags);
 126
 127         data->chip->irq_unmask(data);
 128 }
 129
 130 static void vmd_irq_disable(struct irq_data *data)
 131 {
 132         struct vmd_irq *vmdirq = data->chip_data;
 133         unsigned long flags;
 134
 135         data->chip->irq_mask(data);
 136
 137         raw_spin_lock_irqsave(&list_lock, flags);
 138         if (vmdirq->enabled) {
 139                 list_del_rcu(&vmdirq->node);
 140                 vmdirq->enabled = false;
 141         }
 142         raw_spin_unlock_irqrestore(&list_lock, flags);
 143 }
 144
 145 /*
 146  * XXX: Stubbed until we develop acceptable way to not create conflicts with
 147  * other devices sharing the same vector.
 148  */
 149 static int vmd_irq_set_affinity(struct irq_data *data,
 150                                 const struct cpumask *dest, bool force)
 151 {
 152         return -EINVAL;
 153 }
 154
 155 static struct irq_chip vmd_msi_controller = {
 156         .name                   = "VMD-MSI",
 157         .irq_enable             = vmd_irq_enable,
 158         .irq_disable            = vmd_irq_disable,
 159         .irq_compose_msi_msg    = vmd_compose_msi_msg,
 160         .irq_set_affinity       = vmd_irq_set_affinity,
 161 };
 162
 163 static irq_hw_number_t vmd_get_hwirq(struct msi_domain_info *info,
 164                                      msi_alloc_info_t *arg)
 165 {
 166         return 0;
 167 }
 168
 169 /*
 170  * XXX: We can be even smarter selecting the best IRQ once we solve the
 171  * affinity problem.
 172  */
 173 static struct vmd_irq_list *vmd_next_irq(struct vmd_dev *vmd, struct msi_desc *desc)
 174 {
 175         int i, best = 1;
 176         unsigned long flags;
 177
 178         if (pci_is_bridge(msi_desc_to_pci_dev(desc)) || vmd->msix_count == 1)
 179                 return &vmd->irqs[0];
 180
 181         raw_spin_lock_irqsave(&list_lock, flags);
 182         for (i = 1; i < vmd->msix_count; i++)
 183                 if (vmd->irqs[i].count < vmd->irqs[best].count)
 184                         best = i;
 185         vmd->irqs[best].count++;
 186         raw_spin_unlock_irqrestore(&list_lock, flags);
 187
 188         return &vmd->irqs[best];
 189 }
 190
 191 static int vmd_msi_init(struct irq_domain *domain, struct msi_domain_info *info,
 192                         unsigned int virq, irq_hw_number_t hwirq,
 193                         msi_alloc_info_t *arg)
 194 {
 195         struct msi_desc *desc = arg->desc;
 196         struct vmd_dev *vmd = vmd_from_bus(msi_desc_to_pci_dev(desc)->bus);
 197         struct vmd_irq *vmdirq = kzalloc(sizeof(*vmdirq), GFP_KERNEL);
 198         unsigned int index, vector;
 199
 200         if (!vmdirq)
 201                 return -ENOMEM;
 202
 203         INIT_LIST_HEAD(&vmdirq->node);
 204         vmdirq->irq = vmd_next_irq(vmd, desc);
 205         vmdirq->virq = virq;
 206         index = index_from_irqs(vmd, vmdirq->irq);
 207         vector = pci_irq_vector(vmd->dev, index);
 208
 209         irq_domain_set_info(domain, virq, vector, info->chip, vmdirq,
 210                             handle_untracked_irq, vmd, NULL);
 211         return 0;
 212 }
 213
 214 static void vmd_msi_free(struct irq_domain *domain,
 215                         struct msi_domain_info *info, unsigned int virq)
 216 {
 217         struct vmd_irq *vmdirq = irq_get_chip_data(virq);
 218         unsigned long flags;
 219
 220         synchronize_srcu(&vmdirq->irq->srcu);
 221
 222         /* XXX: Potential optimization to rebalance */
 223         raw_spin_lock_irqsave(&list_lock, flags);
 224         vmdirq->irq->count--;
 225         raw_spin_unlock_irqrestore(&list_lock, flags);
 226
 227         kfree(vmdirq);
 228 }
 229
 230 static int vmd_msi_prepare(struct irq_domain *domain, struct device *dev,
 231                            int nvec, msi_alloc_info_t *arg)
 232 {
 233         struct pci_dev *pdev = to_pci_dev(dev);
 234         struct vmd_dev *vmd = vmd_from_bus(pdev->bus);
 235
 236         if (nvec > vmd->msix_count)
 237                 return vmd->msix_count;
 238
 239         memset(arg, 0, sizeof(*arg));
 240         return 0;
 241 }
 242
 243 static void vmd_set_desc(msi_alloc_info_t *arg, struct msi_desc *desc)
 244 {
 245         arg->desc = desc;
 246 }
 247
 248 static struct msi_domain_ops vmd_msi_domain_ops = {
 249         .get_hwirq      = vmd_get_hwirq,
 250         .msi_init       = vmd_msi_init,
 251         .msi_free       = vmd_msi_free,
 252         .msi_prepare    = vmd_msi_prepare,
 253         .set_desc       = vmd_set_desc,
 254 };
 255
 256 static struct msi_domain_info vmd_msi_domain_info = {
 257         .flags          = MSI_FLAG_USE_DEF_DOM_OPS | MSI_FLAG_USE_DEF_CHIP_OPS |
 258                           MSI_FLAG_PCI_MSIX,
 259         .ops            = &vmd_msi_domain_ops,
 260         .chip           = &vmd_msi_controller,
 261 };
 262
 263 #ifdef CONFIG_X86_DEV_DMA_OPS
 264 /*
 265  * VMD replaces the requester ID with its own.  DMA mappings for devices in a
 266  * VMD domain need to be mapped for the VMD, not the device requiring
 267  * the mapping.
 268  */
 269 static struct device *to_vmd_dev(struct device *dev)
 270 {
 271         struct pci_dev *pdev = to_pci_dev(dev);
 272         struct vmd_dev *vmd = vmd_from_bus(pdev->bus);
 273
 274         return &vmd->dev->dev;
 275 }
 276
 277 static const struct dma_map_ops *vmd_dma_ops(struct device *dev)
 278 {
 279         return get_dma_ops(to_vmd_dev(dev));
 280 }
 281
 282 static void *vmd_alloc(struct device *dev, size_t size, dma_addr_t *addr,
 283                        gfp_t flag, unsigned long attrs)
 284 {
 285         return vmd_dma_ops(dev)->alloc(to_vmd_dev(dev), size, addr, flag,
 286                                        attrs);
 287 }
 288
 289 static void vmd_free(struct device *dev, size_t size, void *vaddr,
 290                      dma_addr_t addr, unsigned long attrs)
 291 {
 292         return vmd_dma_ops(dev)->free(to_vmd_dev(dev), size, vaddr, addr,
 293                                       attrs);
 294 }
 295
 296 static int vmd_mmap(struct device *dev, struct vm_area_struct *vma,
 297                     void *cpu_addr, dma_addr_t addr, size_t size,
 298                     unsigned long attrs)
 299 {
 300         return vmd_dma_ops(dev)->mmap(to_vmd_dev(dev), vma, cpu_addr, addr,
 301                                       size, attrs);
 302 }
 303
 304 static int vmd_get_sgtable(struct device *dev, struct sg_table *sgt,
 305                            void *cpu_addr, dma_addr_t addr, size_t size,
 306                            unsigned long attrs)
 307 {
 308         return vmd_dma_ops(dev)->get_sgtable(to_vmd_dev(dev), sgt, cpu_addr,
 309                                              addr, size, attrs);
 310 }
 311
 312 static dma_addr_t vmd_map_page(struct device *dev, struct page *page,
 313                                unsigned long offset, size_t size,
 314                                enum dma_data_direction dir,
 315                                unsigned long attrs)
 316 {
 317         return vmd_dma_ops(dev)->map_page(to_vmd_dev(dev), page, offset, size,
 318                                           dir, attrs);
 319 }
 320
 321 static void vmd_unmap_page(struct device *dev, dma_addr_t addr, size_t size,
 322                            enum dma_data_direction dir, unsigned long attrs)
 323 {
 324         vmd_dma_ops(dev)->unmap_page(to_vmd_dev(dev), addr, size, dir, attrs);
 325 }
 326
 327 static int vmd_map_sg(struct device *dev, struct scatterlist *sg, int nents,
 328                       enum dma_data_direction dir, unsigned long attrs)
 329 {
 330         return vmd_dma_ops(dev)->map_sg(to_vmd_dev(dev), sg, nents, dir, attrs);
 331 }
 332
 333 static void vmd_unmap_sg(struct device *dev, struct scatterlist *sg, int nents,
 334                          enum dma_data_direction dir, unsigned long attrs)
 335 {
 336         vmd_dma_ops(dev)->unmap_sg(to_vmd_dev(dev), sg, nents, dir, attrs);
 337 }
 338
 339 static void vmd_sync_single_for_cpu(struct device *dev, dma_addr_t addr,
 340                                     size_t size, enum dma_data_direction dir)
 341 {
 342         vmd_dma_ops(dev)->sync_single_for_cpu(to_vmd_dev(dev), addr, size, dir);
 343 }
 344
 345 static void vmd_sync_single_for_device(struct device *dev, dma_addr_t addr,
 346                                        size_t size, enum dma_data_direction dir)
 347 {
 348         vmd_dma_ops(dev)->sync_single_for_device(to_vmd_dev(dev), addr, size,
 349                                                  dir);
 350 }
 351
 352 static void vmd_sync_sg_for_cpu(struct device *dev, struct scatterlist *sg,
 353                                 int nents, enum dma_data_direction dir)
 354 {
 355         vmd_dma_ops(dev)->sync_sg_for_cpu(to_vmd_dev(dev), sg, nents, dir);
 356 }
 357
 358 static void vmd_sync_sg_for_device(struct device *dev, struct scatterlist *sg,
 359                                    int nents, enum dma_data_direction dir)
 360 {
 361         vmd_dma_ops(dev)->sync_sg_for_device(to_vmd_dev(dev), sg, nents, dir);
 362 }
 363
 364 static int vmd_mapping_error(struct device *dev, dma_addr_t addr)
 365 {
 366         return vmd_dma_ops(dev)->mapping_error(to_vmd_dev(dev), addr);
 367 }
 368
 369 static int vmd_dma_supported(struct device *dev, u64 mask)
 370 {
 371         return vmd_dma_ops(dev)->dma_supported(to_vmd_dev(dev), mask);
 372 }
 373
 374 #ifdef ARCH_HAS_DMA_GET_REQUIRED_MASK
 375 static u64 vmd_get_required_mask(struct device *dev)
 376 {
 377         return vmd_dma_ops(dev)->get_required_mask(to_vmd_dev(dev));
 378 }
 379 #endif
 380
 381 static void vmd_teardown_dma_ops(struct vmd_dev *vmd)
 382 {
 383         struct dma_domain *domain = &vmd->dma_domain;
 384
 385         if (get_dma_ops(&vmd->dev->dev))
 386                 del_dma_domain(domain);
 387 }
 388
 389 #define ASSIGN_VMD_DMA_OPS(source, dest, fn)    \
 390         do {                                    \
 391                 if (source->fn)                 \
 392                         dest->fn = vmd_##fn;    \
 393         } while (0)
 394
 395 static void vmd_setup_dma_ops(struct vmd_dev *vmd)
 396 {
 397         const struct dma_map_ops *source = get_dma_ops(&vmd->dev->dev);
 398         struct dma_map_ops *dest = &vmd->dma_ops;
 399         struct dma_domain *domain = &vmd->dma_domain;
 400
 401         domain->domain_nr = vmd->sysdata.domain;
 402         domain->dma_ops = dest;
 403
 404         if (!source)
 405                 return;
 406         ASSIGN_VMD_DMA_OPS(source, dest, alloc);
 407         ASSIGN_VMD_DMA_OPS(source, dest, free);
 408         ASSIGN_VMD_DMA_OPS(source, dest, mmap);
 409         ASSIGN_VMD_DMA_OPS(source, dest, get_sgtable);
 410         ASSIGN_VMD_DMA_OPS(source, dest, map_page);
 411         ASSIGN_VMD_DMA_OPS(source, dest, unmap_page);
 412         ASSIGN_VMD_DMA_OPS(source, dest, map_sg);
 413         ASSIGN_VMD_DMA_OPS(source, dest, unmap_sg);
 414         ASSIGN_VMD_DMA_OPS(source, dest, sync_single_for_cpu);
 415         ASSIGN_VMD_DMA_OPS(source, dest, sync_single_for_device);
 416         ASSIGN_VMD_DMA_OPS(source, dest, sync_sg_for_cpu);
 417         ASSIGN_VMD_DMA_OPS(source, dest, sync_sg_for_device);
 418         ASSIGN_VMD_DMA_OPS(source, dest, mapping_error);
 419         ASSIGN_VMD_DMA_OPS(source, dest, dma_supported);
 420 #ifdef ARCH_HAS_DMA_GET_REQUIRED_MASK
 421         ASSIGN_VMD_DMA_OPS(source, dest, get_required_mask);
 422 #endif
 423         add_dma_domain(domain);
 424 }
 425 #undef ASSIGN_VMD_DMA_OPS
 426 #else
 427 static void vmd_teardown_dma_ops(struct vmd_dev *vmd) {}
 428 static void vmd_setup_dma_ops(struct vmd_dev *vmd) {}
 429 #endif
 430
 431 static char __iomem *vmd_cfg_addr(struct vmd_dev *vmd, struct pci_bus *bus,
 432                                   unsigned int devfn, int reg, int len)
 433 {
 434         char __iomem *addr = vmd->cfgbar +
 435                              (bus->number << 20) + (devfn << 12) + reg;
 436
 437         if ((addr - vmd->cfgbar) + len >=
 438             resource_size(&vmd->dev->resource[VMD_CFGBAR]))
 439                 return NULL;
 440
 441         return addr;
 442 }
 443
 444 /*
 445  * CPU may deadlock if config space is not serialized on some versions of this
 446  * hardware, so all config space access is done under a spinlock.
 447  */
 448 static int vmd_pci_read(struct pci_bus *bus, unsigned int devfn, int reg,
 449                         int len, u32 *value)
 450 {
 451         struct vmd_dev *vmd = vmd_from_bus(bus);
 452         char __iomem *addr = vmd_cfg_addr(vmd, bus, devfn, reg, len);
 453         unsigned long flags;
 454         int ret = 0;
 455
 456         if (!addr)
 457                 return -EFAULT;
 458
 459         spin_lock_irqsave(&vmd->cfg_lock, flags);
 460         switch (len) {
 461         case 1:
 462                 *value = readb(addr);
 463                 break;
 464         case 2:
 465                 *value = readw(addr);
 466                 break;
 467         case 4:
 468                 *value = readl(addr);
 469                 break;
 470         default:
 471                 ret = -EINVAL;
 472                 break;
 473         }
 474         spin_unlock_irqrestore(&vmd->cfg_lock, flags);
 475         return ret;
 476 }
 477
 478 /*
 479  * VMD h/w converts non-posted config writes to posted memory writes. The
 480  * read-back in this function forces the completion so it returns only after
 481  * the config space was written, as expected.
 482  */
 483 static int vmd_pci_write(struct pci_bus *bus, unsigned int devfn, int reg,
 484                          int len, u32 value)
 485 {
 486         struct vmd_dev *vmd = vmd_from_bus(bus);
 487         char __iomem *addr = vmd_cfg_addr(vmd, bus, devfn, reg, len);
 488         unsigned long flags;
 489         int ret = 0;
 490
 491         if (!addr)
 492                 return -EFAULT;
 493
 494         spin_lock_irqsave(&vmd->cfg_lock, flags);
 495         switch (len) {
 496         case 1:
 497                 writeb(value, addr);
 498                 readb(addr);
 499                 break;
 500         case 2:
 501                 writew(value, addr);
 502                 readw(addr);
 503                 break;
 504         case 4:
 505                 writel(value, addr);
 506                 readl(addr);
 507                 break;
 508         default:
 509                 ret = -EINVAL;
 510                 break;
 511         }
 512         spin_unlock_irqrestore(&vmd->cfg_lock, flags);
 513         return ret;
 514 }
 515
 516 static struct pci_ops vmd_ops = {
 517         .read           = vmd_pci_read,
 518         .write          = vmd_pci_write,
 519 };
 520
 521 static void vmd_attach_resources(struct vmd_dev *vmd)
 522 {
 523         vmd->dev->resource[VMD_MEMBAR1].child = &vmd->resources[1];
 524         vmd->dev->resource[VMD_MEMBAR2].child = &vmd->resources[2];
 525 }
 526
 527 static void vmd_detach_resources(struct vmd_dev *vmd)
 528 {
 529         vmd->dev->resource[VMD_MEMBAR1].child = NULL;
 530         vmd->dev->resource[VMD_MEMBAR2].child = NULL;
 531 }
 532
 533 /*
 534  * VMD domains start at 0x10000 to not clash with ACPI _SEG domains.
 535  * Per ACPI r6.0, sec 6.5.6,  _SEG returns an integer, of which the lower
 536  * 16 bits are the PCI Segment Group (domain) number.  Other bits are
 537  * currently reserved.
 538  */
 539 static int vmd_find_free_domain(void)
 540 {
 541         int domain = 0xffff;
 542         struct pci_bus *bus = NULL;
 543
 544         while ((bus = pci_find_next_bus(bus)) != NULL)
 545                 domain = max_t(int, domain, pci_domain_nr(bus));
 546         return domain + 1;
 547 }
 548
 549 static int vmd_enable_domain(struct vmd_dev *vmd)
 550 {
 551         struct pci_sysdata *sd = &vmd->sysdata;
 552         struct fwnode_handle *fn;
 553         struct resource *res;
 554         u32 upper_bits;
 555         unsigned long flags;
 556         LIST_HEAD(resources);
 557
 558         res = &vmd->dev->resource[VMD_CFGBAR];
 559         vmd->resources[0] = (struct resource) {
 560                 .name  = "VMD CFGBAR",
 561                 .start = 0,
 562                 .end   = (resource_size(res) >> 20) - 1,
 563                 .flags = IORESOURCE_BUS | IORESOURCE_PCI_FIXED,
 564         };
 565
 566         /*
 567          * If the window is below 4GB, clear IORESOURCE_MEM_64 so we can
 568          * put 32-bit resources in the window.
 569          *
 570          * There's no hardware reason why a 64-bit window *couldn't*
 571          * contain a 32-bit resource, but pbus_size_mem() computes the
 572          * bridge window size assuming a 64-bit window will contain no
 573          * 32-bit resources.  __pci_assign_resource() enforces that
 574          * artificial restriction to make sure everything will fit.
 575          *
 576          * The only way we could use a 64-bit non-prefechable MEMBAR is
 577          * if its address is <4GB so that we can convert it to a 32-bit
 578          * resource.  To be visible to the host OS, all VMD endpoints must
 579          * be initially configured by platform BIOS, which includes setting
 580          * up these resources.  We can assume the device is configured
 581          * according to the platform needs.
 582          */
 583         res = &vmd->dev->resource[VMD_MEMBAR1];
 584         upper_bits = upper_32_bits(res->end);
 585         flags = res->flags & ~IORESOURCE_SIZEALIGN;
 586         if (!upper_bits)
 587                 flags &= ~IORESOURCE_MEM_64;
 588         vmd->resources[1] = (struct resource) {
 589                 .name  = "VMD MEMBAR1",
 590                 .start = res->start,
 591                 .end   = res->end,
 592                 .flags = flags,
 593                 .parent = res,
 594         };
 595
 596         res = &vmd->dev->resource[VMD_MEMBAR2];
 597         upper_bits = upper_32_bits(res->end);
 598         flags = res->flags & ~IORESOURCE_SIZEALIGN;
 599         if (!upper_bits)
 600                 flags &= ~IORESOURCE_MEM_64;
 601         vmd->resources[2] = (struct resource) {
 602                 .name  = "VMD MEMBAR2",
 603                 .start = res->start + 0x2000,
 604                 .end   = res->end,
 605                 .flags = flags,
 606                 .parent = res,
 607         };
 608
 609         sd->vmd_domain = true;
 610         sd->domain = vmd_find_free_domain();
 611         if (sd->domain < 0)
 612                 return sd->domain;
 613
 614         sd->node = pcibus_to_node(vmd->dev->bus);
 615
 616         fn = irq_domain_alloc_named_id_fwnode("VMD-MSI", vmd->sysdata.domain);
 617         if (!fn)
 618                 return -ENODEV;
 619
 620         vmd->irq_domain = pci_msi_create_irq_domain(fn, &vmd_msi_domain_info,
 621                                                     x86_vector_domain);
 622         irq_domain_free_fwnode(fn);
 623         if (!vmd->irq_domain)
 624                 return -ENODEV;
 625
 626         pci_add_resource(&resources, &vmd->resources[0]);
 627         pci_add_resource(&resources, &vmd->resources[1]);
 628         pci_add_resource(&resources, &vmd->resources[2]);
 629         vmd->bus = pci_create_root_bus(&vmd->dev->dev, 0, &vmd_ops, sd,
 630                                        &resources);
 631         if (!vmd->bus) {
 632                 pci_free_resource_list(&resources);
 633                 irq_domain_remove(vmd->irq_domain);
 634                 return -ENODEV;
 635         }
 636
 637         vmd_attach_resources(vmd);
 638         vmd_setup_dma_ops(vmd);
 639         dev_set_msi_domain(&vmd->bus->dev, vmd->irq_domain);
 640         pci_rescan_bus(vmd->bus);
 641
 642         WARN(sysfs_create_link(&vmd->dev->dev.kobj, &vmd->bus->dev.kobj,
 643                                "domain"), "Can't create symlink to domain\n");
 644         return 0;
 645 }
 646
 647 static irqreturn_t vmd_irq(int irq, void *data)
 648 {
 649         struct vmd_irq_list *irqs = data;
 650         struct vmd_irq *vmdirq;
 651         int idx;
 652
 653         idx = srcu_read_lock(&irqs->srcu);
 654         list_for_each_entry_rcu(vmdirq, &irqs->irq_list, node)
 655                 generic_handle_irq(vmdirq->virq);
 656         srcu_read_unlock(&irqs->srcu, idx);
 657
 658         return IRQ_HANDLED;
 659 }
 660
 661 static int vmd_probe(struct pci_dev *dev, const struct pci_device_id *id)
 662 {
 663         struct vmd_dev *vmd;
 664         int i, err;
 665
 666         if (resource_size(&dev->resource[VMD_CFGBAR]) < (1 << 20))
 667                 return -ENOMEM;
 668
 669         vmd = devm_kzalloc(&dev->dev, sizeof(*vmd), GFP_KERNEL);
 670         if (!vmd)
 671                 return -ENOMEM;
 672
 673         vmd->dev = dev;
 674         err = pcim_enable_device(dev);
 675         if (err < 0)
 676                 return err;
 677
 678         vmd->cfgbar = pcim_iomap(dev, VMD_CFGBAR, 0);
 679         if (!vmd->cfgbar)
 680                 return -ENOMEM;
 681
 682         pci_set_master(dev);
 683         if (dma_set_mask_and_coherent(&dev->dev, DMA_BIT_MASK(64)) &&
 684             dma_set_mask_and_coherent(&dev->dev, DMA_BIT_MASK(32)))
 685                 return -ENODEV;
 686
 687         vmd->msix_count = pci_msix_vec_count(dev);
 688         if (vmd->msix_count < 0)
 689                 return -ENODEV;
 690
 691         vmd->msix_count = pci_alloc_irq_vectors(dev, 1, vmd->msix_count,
 692                                         PCI_IRQ_MSIX);
 693         if (vmd->msix_count < 0)
 694                 return vmd->msix_count;
 695
 696         vmd->irqs = devm_kcalloc(&dev->dev, vmd->msix_count, sizeof(*vmd->irqs),
 697                                  GFP_KERNEL);
 698         if (!vmd->irqs)
 699                 return -ENOMEM;
 700
 701         for (i = 0; i < vmd->msix_count; i++) {
 702                 err = init_srcu_struct(&vmd->irqs[i].srcu);
 703                 if (err)
 704                         return err;
 705
 706                 INIT_LIST_HEAD(&vmd->irqs[i].irq_list);
 707                 err = devm_request_irq(&dev->dev, pci_irq_vector(dev, i),
 708                                        vmd_irq, IRQF_NO_THREAD,
 709                                        "vmd", &vmd->irqs[i]);
 710                 if (err)
 711                         return err;
 712         }
 713
 714         spin_lock_init(&vmd->cfg_lock);
 715         pci_set_drvdata(dev, vmd);
 716         err = vmd_enable_domain(vmd);
 717         if (err)
 718                 return err;
 719
 720         dev_info(&vmd->dev->dev, "Bound to PCI domain %04x\n",
 721                  vmd->sysdata.domain);
 722         return 0;
 723 }
 724
 725 static void vmd_cleanup_srcu(struct vmd_dev *vmd)
 726 {
 727         int i;
 728
 729         for (i = 0; i < vmd->msix_count; i++)
 730                 cleanup_srcu_struct(&vmd->irqs[i].srcu);
 731 }
 732
 733 static void vmd_remove(struct pci_dev *dev)
 734 {
 735         struct vmd_dev *vmd = pci_get_drvdata(dev);
 736
 737         vmd_detach_resources(vmd);
 738         sysfs_remove_link(&vmd->dev->dev.kobj, "domain");
 739         pci_stop_root_bus(vmd->bus);
 740         pci_remove_root_bus(vmd->bus);
 741         vmd_cleanup_srcu(vmd);
 742         vmd_teardown_dma_ops(vmd);
 743         irq_domain_remove(vmd->irq_domain);
 744 }
 745
 746 #ifdef CONFIG_PM_SLEEP
 747 static int vmd_suspend(struct device *dev)
 748 {
 749         struct pci_dev *pdev = to_pci_dev(dev);
 750         struct vmd_dev *vmd = pci_get_drvdata(pdev);
 751         int i;
 752
 753         for (i = 0; i < vmd->msix_count; i++)
 754                 devm_free_irq(dev, pci_irq_vector(pdev, i), &vmd->irqs[i]);
 755
 756         pci_save_state(pdev);
 757         return 0;
 758 }
 759
 760 static int vmd_resume(struct device *dev)
 761 {
 762         struct pci_dev *pdev = to_pci_dev(dev);
 763         struct vmd_dev *vmd = pci_get_drvdata(pdev);
 764         int err, i;
 765
 766         for (i = 0; i < vmd->msix_count; i++) {
 767                 err = devm_request_irq(dev, pci_irq_vector(pdev, i),
 768                                        vmd_irq, IRQF_NO_THREAD,
 769                                        "vmd", &vmd->irqs[i]);
 770                 if (err)
 771                         return err;
 772         }
 773
 774         pci_restore_state(pdev);
 775         return 0;
 776 }
 777 #endif
 778 static SIMPLE_DEV_PM_OPS(vmd_dev_pm_ops, vmd_suspend, vmd_resume);
 779
 780 static const struct pci_device_id vmd_ids[] = {
 781         {PCI_DEVICE(PCI_VENDOR_ID_INTEL, 0x201d),},
 782         {0,}
 783 };
 784 MODULE_DEVICE_TABLE(pci, vmd_ids);
 785
 786 static struct pci_driver vmd_drv = {
 787         .name           = "vmd",
 788         .id_table       = vmd_ids,
 789         .probe          = vmd_probe,
 790         .remove         = vmd_remove,
 791         .driver         = {
 792                 .pm     = &vmd_dev_pm_ops,
 793         },
 794 };
 795 module_pci_driver(vmd_drv);
 796
 797 MODULE_AUTHOR("Intel Corporation");
 798 MODULE_LICENSE("GPL v2");
 799 MODULE_VERSION("0.6");