arch/x86/pci/vmd.c

   1 /*
   2  * Volume Management Device driver
   3  * Copyright (c) 2015, Intel Corporation.
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms and conditions of the GNU General Public License,
   7  * version 2, as published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  12  * more details.
  13  */
  14
  15 #include <linux/device.h>
  16 #include <linux/interrupt.h>
  17 #include <linux/irq.h>
  18 #include <linux/kernel.h>
  19 #include <linux/module.h>
  20 #include <linux/msi.h>
  21 #include <linux/pci.h>
  22 #include <linux/rculist.h>
  23 #include <linux/rcupdate.h>
  24
  25 #include <asm/irqdomain.h>
  26 #include <asm/device.h>
  27 #include <asm/msi.h>
  28 #include <asm/msidef.h>
  29
  30 #define VMD_CFGBAR      0
  31 #define VMD_MEMBAR1     2
  32 #define VMD_MEMBAR2     4
  33
  34 /*
  35  * Lock for manipulating VMD IRQ lists.
  36  */
  37 static DEFINE_RAW_SPINLOCK(list_lock);
  38
  39 /**
  40  * struct vmd_irq - private data to map driver IRQ to the VMD shared vector
  41  * @node:       list item for parent traversal.
  42  * @rcu:        RCU callback item for freeing.
  43  * @irq:        back pointer to parent.
  44  * @enabled:    true if driver enabled IRQ
  45  * @virq:       the virtual IRQ value provided to the requesting driver.
  46  *
  47  * Every MSI/MSI-X IRQ requested for a device in a VMD domain will be mapped to
  48  * a VMD IRQ using this structure.
  49  */
  50 struct vmd_irq {
  51         struct list_head        node;
  52         struct rcu_head         rcu;
  53         struct vmd_irq_list     *irq;
  54         bool                    enabled;
  55         unsigned int            virq;
  56 };
  57
  58 /**
  59  * struct vmd_irq_list - list of driver requested IRQs mapping to a VMD vector
  60  * @irq_list:   the list of irq's the VMD one demuxes to.
  61  * @vmd_vector: the h/w IRQ assigned to the VMD.
  62  * @index:      index into the VMD MSI-X table; used for message routing.
  63  * @count:      number of child IRQs assigned to this vector; used to track
  64  *              sharing.
  65  */
  66 struct vmd_irq_list {
  67         struct list_head        irq_list;
  68         struct vmd_dev          *vmd;
  69         unsigned int            vmd_vector;
  70         unsigned int            index;
  71         unsigned int            count;
  72 };
  73
  74 struct vmd_dev {
  75         struct pci_dev          *dev;
  76
  77         spinlock_t              cfg_lock;
  78         char __iomem            *cfgbar;
  79
  80         int msix_count;
  81         struct msix_entry       *msix_entries;
  82         struct vmd_irq_list     *irqs;
  83
  84         struct pci_sysdata      sysdata;
  85         struct resource         resources[3];
  86         struct irq_domain       *irq_domain;
  87         struct pci_bus          *bus;
  88
  89 #ifdef CONFIG_X86_DEV_DMA_OPS
  90         struct dma_map_ops      dma_ops;
  91         struct dma_domain       dma_domain;
  92 #endif
  93 };
  94
  95 static inline struct vmd_dev *vmd_from_bus(struct pci_bus *bus)
  96 {
  97         return container_of(bus->sysdata, struct vmd_dev, sysdata);
  98 }
  99
 100 /*
 101  * Drivers managing a device in a VMD domain allocate their own IRQs as before,
 102  * but the MSI entry for the hardware it's driving will be programmed with a
 103  * destination ID for the VMD MSI-X table.  The VMD muxes interrupts in its
 104  * domain into one of its own, and the VMD driver de-muxes these for the
 105  * handlers sharing that VMD IRQ.  The vmd irq_domain provides the operations
 106  * and irq_chip to set this up.
 107  */
 108 static void vmd_compose_msi_msg(struct irq_data *data, struct msi_msg *msg)
 109 {
 110         struct vmd_irq *vmdirq = data->chip_data;
 111         struct vmd_irq_list *irq = vmdirq->irq;
 112
 113         msg->address_hi = MSI_ADDR_BASE_HI;
 114         msg->address_lo = MSI_ADDR_BASE_LO | MSI_ADDR_DEST_ID(irq->index);
 115         msg->data = 0;
 116 }
 117
 118 /*
 119  * We rely on MSI_FLAG_USE_DEF_CHIP_OPS to set the IRQ mask/unmask ops.
 120  */
 121 static void vmd_irq_enable(struct irq_data *data)
 122 {
 123         struct vmd_irq *vmdirq = data->chip_data;
 124         unsigned long flags;
 125
 126         raw_spin_lock_irqsave(&list_lock, flags);
 127         WARN_ON(vmdirq->enabled);
 128         list_add_tail_rcu(&vmdirq->node, &vmdirq->irq->irq_list);
 129         vmdirq->enabled = true;
 130         raw_spin_unlock_irqrestore(&list_lock, flags);
 131
 132         data->chip->irq_unmask(data);
 133 }
 134
 135 static void vmd_irq_disable(struct irq_data *data)
 136 {
 137         struct vmd_irq *vmdirq = data->chip_data;
 138         unsigned long flags;
 139
 140         data->chip->irq_mask(data);
 141
 142         raw_spin_lock_irqsave(&list_lock, flags);
 143         if (vmdirq->enabled) {
 144                 list_del_rcu(&vmdirq->node);
 145                 vmdirq->enabled = false;
 146         }
 147         raw_spin_unlock_irqrestore(&list_lock, flags);
 148 }
 149
 150 /*
 151  * XXX: Stubbed until we develop acceptable way to not create conflicts with
 152  * other devices sharing the same vector.
 153  */
 154 static int vmd_irq_set_affinity(struct irq_data *data,
 155                                 const struct cpumask *dest, bool force)
 156 {
 157         return -EINVAL;
 158 }
 159
 160 static struct irq_chip vmd_msi_controller = {
 161         .name                   = "VMD-MSI",
 162         .irq_enable             = vmd_irq_enable,
 163         .irq_disable            = vmd_irq_disable,
 164         .irq_compose_msi_msg    = vmd_compose_msi_msg,
 165         .irq_set_affinity       = vmd_irq_set_affinity,
 166 };
 167
 168 static irq_hw_number_t vmd_get_hwirq(struct msi_domain_info *info,
 169                                      msi_alloc_info_t *arg)
 170 {
 171         return 0;
 172 }
 173
 174 /*
 175  * XXX: We can be even smarter selecting the best IRQ once we solve the
 176  * affinity problem.
 177  */
 178 static struct vmd_irq_list *vmd_next_irq(struct vmd_dev *vmd, struct msi_desc *desc)
 179 {
 180         int i, best = 1;
 181         unsigned long flags;
 182
 183         if (!desc->msi_attrib.is_msix || vmd->msix_count == 1)
 184                 return &vmd->irqs[0];
 185
 186         raw_spin_lock_irqsave(&list_lock, flags);
 187         for (i = 1; i < vmd->msix_count; i++)
 188                 if (vmd->irqs[i].count < vmd->irqs[best].count)
 189                         best = i;
 190         vmd->irqs[best].count++;
 191         raw_spin_unlock_irqrestore(&list_lock, flags);
 192
 193         return &vmd->irqs[best];
 194 }
 195
 196 static int vmd_msi_init(struct irq_domain *domain, struct msi_domain_info *info,
 197                         unsigned int virq, irq_hw_number_t hwirq,
 198                         msi_alloc_info_t *arg)
 199 {
 200         struct msi_desc *desc = arg->desc;
 201         struct vmd_dev *vmd = vmd_from_bus(msi_desc_to_pci_dev(desc)->bus);
 202         struct vmd_irq *vmdirq = kzalloc(sizeof(*vmdirq), GFP_KERNEL);
 203
 204         if (!vmdirq)
 205                 return -ENOMEM;
 206
 207         INIT_LIST_HEAD(&vmdirq->node);
 208         vmdirq->irq = vmd_next_irq(vmd, desc);
 209         vmdirq->virq = virq;
 210
 211         irq_domain_set_info(domain, virq, vmdirq->irq->vmd_vector, info->chip,
 212                             vmdirq, handle_untracked_irq, vmd, NULL);
 213         return 0;
 214 }
 215
 216 static void vmd_msi_free(struct irq_domain *domain,
 217                         struct msi_domain_info *info, unsigned int virq)
 218 {
 219         struct vmd_irq *vmdirq = irq_get_chip_data(virq);
 220         unsigned long flags;
 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_rcu(vmdirq, rcu);
 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 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 0x1000 to not clash with ACPI _SEG domains.
 535  */
 536 static int vmd_find_free_domain(void)
 537 {
 538         int domain = 0xffff;
 539         struct pci_bus *bus = NULL;
 540
 541         while ((bus = pci_find_next_bus(bus)) != NULL)
 542                 domain = max_t(int, domain, pci_domain_nr(bus));
 543         return domain + 1;
 544 }
 545
 546 static int vmd_enable_domain(struct vmd_dev *vmd)
 547 {
 548         struct pci_sysdata *sd = &vmd->sysdata;
 549         struct resource *res;
 550         u32 upper_bits;
 551         unsigned long flags;
 552         LIST_HEAD(resources);
 553
 554         res = &vmd->dev->resource[VMD_CFGBAR];
 555         vmd->resources[0] = (struct resource) {
 556                 .name  = "VMD CFGBAR",
 557                 .start = 0,
 558                 .end   = (resource_size(res) >> 20) - 1,
 559                 .flags = IORESOURCE_BUS | IORESOURCE_PCI_FIXED,
 560         };
 561
 562         /*
 563          * If the window is below 4GB, clear IORESOURCE_MEM_64 so we can
 564          * put 32-bit resources in the window.
 565          *
 566          * There's no hardware reason why a 64-bit window *couldn't*
 567          * contain a 32-bit resource, but pbus_size_mem() computes the
 568          * bridge window size assuming a 64-bit window will contain no
 569          * 32-bit resources.  __pci_assign_resource() enforces that
 570          * artificial restriction to make sure everything will fit.
 571          *
 572          * The only way we could use a 64-bit non-prefechable MEMBAR is
 573          * if its address is <4GB so that we can convert it to a 32-bit
 574          * resource.  To be visible to the host OS, all VMD endpoints must
 575          * be initially configured by platform BIOS, which includes setting
 576          * up these resources.  We can assume the device is configured
 577          * according to the platform needs.
 578          */
 579         res = &vmd->dev->resource[VMD_MEMBAR1];
 580         upper_bits = upper_32_bits(res->end);
 581         flags = res->flags & ~IORESOURCE_SIZEALIGN;
 582         if (!upper_bits)
 583                 flags &= ~IORESOURCE_MEM_64;
 584         vmd->resources[1] = (struct resource) {
 585                 .name  = "VMD MEMBAR1",
 586                 .start = res->start,
 587                 .end   = res->end,
 588                 .flags = flags,
 589                 .parent = res,
 590         };
 591
 592         res = &vmd->dev->resource[VMD_MEMBAR2];
 593         upper_bits = upper_32_bits(res->end);
 594         flags = res->flags & ~IORESOURCE_SIZEALIGN;
 595         if (!upper_bits)
 596                 flags &= ~IORESOURCE_MEM_64;
 597         vmd->resources[2] = (struct resource) {
 598                 .name  = "VMD MEMBAR2",
 599                 .start = res->start + 0x2000,
 600                 .end   = res->end,
 601                 .flags = flags,
 602                 .parent = res,
 603         };
 604
 605         sd->domain = vmd_find_free_domain();
 606         if (sd->domain < 0)
 607                 return sd->domain;
 608
 609         sd->node = pcibus_to_node(vmd->dev->bus);
 610
 611         vmd->irq_domain = pci_msi_create_irq_domain(NULL, &vmd_msi_domain_info,
 612                                                     x86_vector_domain);
 613         if (!vmd->irq_domain)
 614                 return -ENODEV;
 615
 616         pci_add_resource(&resources, &vmd->resources[0]);
 617         pci_add_resource(&resources, &vmd->resources[1]);
 618         pci_add_resource(&resources, &vmd->resources[2]);
 619         vmd->bus = pci_create_root_bus(&vmd->dev->dev, 0, &vmd_ops, sd,
 620                                        &resources);
 621         if (!vmd->bus) {
 622                 pci_free_resource_list(&resources);
 623                 irq_domain_remove(vmd->irq_domain);
 624                 return -ENODEV;
 625         }
 626
 627         vmd_attach_resources(vmd);
 628         vmd_setup_dma_ops(vmd);
 629         dev_set_msi_domain(&vmd->bus->dev, vmd->irq_domain);
 630         pci_rescan_bus(vmd->bus);
 631
 632         WARN(sysfs_create_link(&vmd->dev->dev.kobj, &vmd->bus->dev.kobj,
 633                                "domain"), "Can't create symlink to domain\n");
 634         return 0;
 635 }
 636
 637 static irqreturn_t vmd_irq(int irq, void *data)
 638 {
 639         struct vmd_irq_list *irqs = data;
 640         struct vmd_irq *vmdirq;
 641
 642         rcu_read_lock();
 643         list_for_each_entry_rcu(vmdirq, &irqs->irq_list, node)
 644                 generic_handle_irq(vmdirq->virq);
 645         rcu_read_unlock();
 646
 647         return IRQ_HANDLED;
 648 }
 649
 650 static int vmd_probe(struct pci_dev *dev, const struct pci_device_id *id)
 651 {
 652         struct vmd_dev *vmd;
 653         int i, err;
 654
 655         if (resource_size(&dev->resource[VMD_CFGBAR]) < (1 << 20))
 656                 return -ENOMEM;
 657
 658         vmd = devm_kzalloc(&dev->dev, sizeof(*vmd), GFP_KERNEL);
 659         if (!vmd)
 660                 return -ENOMEM;
 661
 662         vmd->dev = dev;
 663         err = pcim_enable_device(dev);
 664         if (err < 0)
 665                 return err;
 666
 667         vmd->cfgbar = pcim_iomap(dev, VMD_CFGBAR, 0);
 668         if (!vmd->cfgbar)
 669                 return -ENOMEM;
 670
 671         pci_set_master(dev);
 672         if (dma_set_mask_and_coherent(&dev->dev, DMA_BIT_MASK(64)) &&
 673             dma_set_mask_and_coherent(&dev->dev, DMA_BIT_MASK(32)))
 674                 return -ENODEV;
 675
 676         vmd->msix_count = pci_msix_vec_count(dev);
 677         if (vmd->msix_count < 0)
 678                 return -ENODEV;
 679
 680         vmd->irqs = devm_kcalloc(&dev->dev, vmd->msix_count, sizeof(*vmd->irqs),
 681                                  GFP_KERNEL);
 682         if (!vmd->irqs)
 683                 return -ENOMEM;
 684
 685         vmd->msix_entries = devm_kcalloc(&dev->dev, vmd->msix_count,
 686                                          sizeof(*vmd->msix_entries),
 687                                          GFP_KERNEL);
 688         if (!vmd->msix_entries)
 689                 return -ENOMEM;
 690         for (i = 0; i < vmd->msix_count; i++)
 691                 vmd->msix_entries[i].entry = i;
 692
 693         vmd->msix_count = pci_enable_msix_range(vmd->dev, vmd->msix_entries, 1,
 694                                                 vmd->msix_count);
 695         if (vmd->msix_count < 0)
 696                 return vmd->msix_count;
 697
 698         for (i = 0; i < vmd->msix_count; i++) {
 699                 INIT_LIST_HEAD(&vmd->irqs[i].irq_list);
 700                 vmd->irqs[i].vmd_vector = vmd->msix_entries[i].vector;
 701                 vmd->irqs[i].index = i;
 702
 703                 err = devm_request_irq(&dev->dev, vmd->irqs[i].vmd_vector,
 704                                        vmd_irq, 0, "vmd", &vmd->irqs[i]);
 705                 if (err)
 706                         return err;
 707         }
 708
 709         spin_lock_init(&vmd->cfg_lock);
 710         pci_set_drvdata(dev, vmd);
 711         err = vmd_enable_domain(vmd);
 712         if (err)
 713                 return err;
 714
 715         dev_info(&vmd->dev->dev, "Bound to PCI domain %04x\n",
 716                  vmd->sysdata.domain);
 717         return 0;
 718 }
 719
 720 static void vmd_remove(struct pci_dev *dev)
 721 {
 722         struct vmd_dev *vmd = pci_get_drvdata(dev);
 723
 724         vmd_detach_resources(vmd);
 725         pci_set_drvdata(dev, NULL);
 726         sysfs_remove_link(&vmd->dev->dev.kobj, "domain");
 727         pci_stop_root_bus(vmd->bus);
 728         pci_remove_root_bus(vmd->bus);
 729         vmd_teardown_dma_ops(vmd);
 730         irq_domain_remove(vmd->irq_domain);
 731 }
 732
 733 #ifdef CONFIG_PM
 734 static int vmd_suspend(struct device *dev)
 735 {
 736         struct pci_dev *pdev = to_pci_dev(dev);
 737
 738         pci_save_state(pdev);
 739         return 0;
 740 }
 741
 742 static int vmd_resume(struct device *dev)
 743 {
 744         struct pci_dev *pdev = to_pci_dev(dev);
 745
 746         pci_restore_state(pdev);
 747         return 0;
 748 }
 749 #endif
 750 static SIMPLE_DEV_PM_OPS(vmd_dev_pm_ops, vmd_suspend, vmd_resume);
 751
 752 static const struct pci_device_id vmd_ids[] = {
 753         {PCI_DEVICE(PCI_VENDOR_ID_INTEL, 0x201d),},
 754         {0,}
 755 };
 756 MODULE_DEVICE_TABLE(pci, vmd_ids);
 757
 758 static struct pci_driver vmd_drv = {
 759         .name           = "vmd",
 760         .id_table       = vmd_ids,
 761         .probe          = vmd_probe,
 762         .remove         = vmd_remove,
 763         .driver         = {
 764                 .pm     = &vmd_dev_pm_ops,
 765         },
 766 };
 767 module_pci_driver(vmd_drv);
 768
 769 MODULE_AUTHOR("Intel Corporation");
 770 MODULE_LICENSE("GPL v2");
 771 MODULE_VERSION("0.6");