linux/drivers/pci/controller/vmd.c
<<
>>
Prefs
   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#define PCI_REG_VMCAP           0x40
  28#define BUS_RESTRICT_CAP(vmcap) (vmcap & 0x1)
  29#define PCI_REG_VMCONFIG        0x44
  30#define BUS_RESTRICT_CFG(vmcfg) ((vmcfg >> 8) & 0x3)
  31#define PCI_REG_VMLOCK          0x70
  32#define MB2_SHADOW_EN(vmlock)   (vmlock & 0x2)
  33
  34enum vmd_features {
  35        /*
  36         * Device may contain registers which hint the physical location of the
  37         * membars, in order to allow proper address translation during
  38         * resource assignment to enable guest virtualization
  39         */
  40        VMD_FEAT_HAS_MEMBAR_SHADOW      = (1 << 0),
  41
  42        /*
  43         * Device may provide root port configuration information which limits
  44         * bus numbering
  45         */
  46        VMD_FEAT_HAS_BUS_RESTRICTIONS   = (1 << 1),
  47};
  48
  49/*
  50 * Lock for manipulating VMD IRQ lists.
  51 */
  52static DEFINE_RAW_SPINLOCK(list_lock);
  53
  54/**
  55 * struct vmd_irq - private data to map driver IRQ to the VMD shared vector
  56 * @node:       list item for parent traversal.
  57 * @irq:        back pointer to parent.
  58 * @enabled:    true if driver enabled IRQ
  59 * @virq:       the virtual IRQ value provided to the requesting driver.
  60 *
  61 * Every MSI/MSI-X IRQ requested for a device in a VMD domain will be mapped to
  62 * a VMD IRQ using this structure.
  63 */
  64struct vmd_irq {
  65        struct list_head        node;
  66        struct vmd_irq_list     *irq;
  67        bool                    enabled;
  68        unsigned int            virq;
  69};
  70
  71/**
  72 * struct vmd_irq_list - list of driver requested IRQs mapping to a VMD vector
  73 * @irq_list:   the list of irq's the VMD one demuxes to.
  74 * @srcu:       SRCU struct for local synchronization.
  75 * @count:      number of child IRQs assigned to this vector; used to track
  76 *              sharing.
  77 */
  78struct vmd_irq_list {
  79        struct list_head        irq_list;
  80        struct srcu_struct      srcu;
  81        unsigned int            count;
  82};
  83
  84struct vmd_dev {
  85        struct pci_dev          *dev;
  86
  87        spinlock_t              cfg_lock;
  88        char __iomem            *cfgbar;
  89
  90        int msix_count;
  91        struct vmd_irq_list     *irqs;
  92
  93        struct pci_sysdata      sysdata;
  94        struct resource         resources[3];
  95        struct irq_domain       *irq_domain;
  96        struct pci_bus          *bus;
  97
  98        struct dma_map_ops      dma_ops;
  99        struct dma_domain       dma_domain;
 100};
 101
 102static inline struct vmd_dev *vmd_from_bus(struct pci_bus *bus)
 103{
 104        return container_of(bus->sysdata, struct vmd_dev, sysdata);
 105}
 106
 107static inline unsigned int index_from_irqs(struct vmd_dev *vmd,
 108                                           struct vmd_irq_list *irqs)
 109{
 110        return irqs - vmd->irqs;
 111}
 112
 113/*
 114 * Drivers managing a device in a VMD domain allocate their own IRQs as before,
 115 * but the MSI entry for the hardware it's driving will be programmed with a
 116 * destination ID for the VMD MSI-X table.  The VMD muxes interrupts in its
 117 * domain into one of its own, and the VMD driver de-muxes these for the
 118 * handlers sharing that VMD IRQ.  The vmd irq_domain provides the operations
 119 * and irq_chip to set this up.
 120 */
 121static void vmd_compose_msi_msg(struct irq_data *data, struct msi_msg *msg)
 122{
 123        struct vmd_irq *vmdirq = data->chip_data;
 124        struct vmd_irq_list *irq = vmdirq->irq;
 125        struct vmd_dev *vmd = irq_data_get_irq_handler_data(data);
 126
 127        msg->address_hi = MSI_ADDR_BASE_HI;
 128        msg->address_lo = MSI_ADDR_BASE_LO |
 129                          MSI_ADDR_DEST_ID(index_from_irqs(vmd, irq));
 130        msg->data = 0;
 131}
 132
 133/*
 134 * We rely on MSI_FLAG_USE_DEF_CHIP_OPS to set the IRQ mask/unmask ops.
 135 */
 136static void vmd_irq_enable(struct irq_data *data)
 137{
 138        struct vmd_irq *vmdirq = data->chip_data;
 139        unsigned long flags;
 140
 141        raw_spin_lock_irqsave(&list_lock, flags);
 142        WARN_ON(vmdirq->enabled);
 143        list_add_tail_rcu(&vmdirq->node, &vmdirq->irq->irq_list);
 144        vmdirq->enabled = true;
 145        raw_spin_unlock_irqrestore(&list_lock, flags);
 146
 147        data->chip->irq_unmask(data);
 148}
 149
 150static void vmd_irq_disable(struct irq_data *data)
 151{
 152        struct vmd_irq *vmdirq = data->chip_data;
 153        unsigned long flags;
 154
 155        data->chip->irq_mask(data);
 156
 157        raw_spin_lock_irqsave(&list_lock, flags);
 158        if (vmdirq->enabled) {
 159                list_del_rcu(&vmdirq->node);
 160                vmdirq->enabled = false;
 161        }
 162        raw_spin_unlock_irqrestore(&list_lock, flags);
 163}
 164
 165/*
 166 * XXX: Stubbed until we develop acceptable way to not create conflicts with
 167 * other devices sharing the same vector.
 168 */
 169static int vmd_irq_set_affinity(struct irq_data *data,
 170                                const struct cpumask *dest, bool force)
 171{
 172        return -EINVAL;
 173}
 174
 175static struct irq_chip vmd_msi_controller = {
 176        .name                   = "VMD-MSI",
 177        .irq_enable             = vmd_irq_enable,
 178        .irq_disable            = vmd_irq_disable,
 179        .irq_compose_msi_msg    = vmd_compose_msi_msg,
 180        .irq_set_affinity       = vmd_irq_set_affinity,
 181};
 182
 183static irq_hw_number_t vmd_get_hwirq(struct msi_domain_info *info,
 184                                     msi_alloc_info_t *arg)
 185{
 186        return 0;
 187}
 188
 189/*
 190 * XXX: We can be even smarter selecting the best IRQ once we solve the
 191 * affinity problem.
 192 */
 193static struct vmd_irq_list *vmd_next_irq(struct vmd_dev *vmd, struct msi_desc *desc)
 194{
 195        int i, best = 1;
 196        unsigned long flags;
 197
 198        if (vmd->msix_count == 1)
 199                return &vmd->irqs[0];
 200
 201        /*
 202         * White list for fast-interrupt handlers. All others will share the
 203         * "slow" interrupt vector.
 204         */
 205        switch (msi_desc_to_pci_dev(desc)->class) {
 206        case PCI_CLASS_STORAGE_EXPRESS:
 207                break;
 208        default:
 209                return &vmd->irqs[0];
 210        }
 211
 212        raw_spin_lock_irqsave(&list_lock, flags);
 213        for (i = 1; i < vmd->msix_count; i++)
 214                if (vmd->irqs[i].count < vmd->irqs[best].count)
 215                        best = i;
 216        vmd->irqs[best].count++;
 217        raw_spin_unlock_irqrestore(&list_lock, flags);
 218
 219        return &vmd->irqs[best];
 220}
 221
 222static int vmd_msi_init(struct irq_domain *domain, struct msi_domain_info *info,
 223                        unsigned int virq, irq_hw_number_t hwirq,
 224                        msi_alloc_info_t *arg)
 225{
 226        struct msi_desc *desc = arg->desc;
 227        struct vmd_dev *vmd = vmd_from_bus(msi_desc_to_pci_dev(desc)->bus);
 228        struct vmd_irq *vmdirq = kzalloc(sizeof(*vmdirq), GFP_KERNEL);
 229        unsigned int index, vector;
 230
 231        if (!vmdirq)
 232                return -ENOMEM;
 233
 234        INIT_LIST_HEAD(&vmdirq->node);
 235        vmdirq->irq = vmd_next_irq(vmd, desc);
 236        vmdirq->virq = virq;
 237        index = index_from_irqs(vmd, vmdirq->irq);
 238        vector = pci_irq_vector(vmd->dev, index);
 239
 240        irq_domain_set_info(domain, virq, vector, info->chip, vmdirq,
 241                            handle_untracked_irq, vmd, NULL);
 242        return 0;
 243}
 244
 245static void vmd_msi_free(struct irq_domain *domain,
 246                        struct msi_domain_info *info, unsigned int virq)
 247{
 248        struct vmd_irq *vmdirq = irq_get_chip_data(virq);
 249        unsigned long flags;
 250
 251        synchronize_srcu(&vmdirq->irq->srcu);
 252
 253        /* XXX: Potential optimization to rebalance */
 254        raw_spin_lock_irqsave(&list_lock, flags);
 255        vmdirq->irq->count--;
 256        raw_spin_unlock_irqrestore(&list_lock, flags);
 257
 258        kfree(vmdirq);
 259}
 260
 261static int vmd_msi_prepare(struct irq_domain *domain, struct device *dev,
 262                           int nvec, msi_alloc_info_t *arg)
 263{
 264        struct pci_dev *pdev = to_pci_dev(dev);
 265        struct vmd_dev *vmd = vmd_from_bus(pdev->bus);
 266
 267        if (nvec > vmd->msix_count)
 268                return vmd->msix_count;
 269
 270        memset(arg, 0, sizeof(*arg));
 271        return 0;
 272}
 273
 274static void vmd_set_desc(msi_alloc_info_t *arg, struct msi_desc *desc)
 275{
 276        arg->desc = desc;
 277}
 278
 279static struct msi_domain_ops vmd_msi_domain_ops = {
 280        .get_hwirq      = vmd_get_hwirq,
 281        .msi_init       = vmd_msi_init,
 282        .msi_free       = vmd_msi_free,
 283        .msi_prepare    = vmd_msi_prepare,
 284        .set_desc       = vmd_set_desc,
 285};
 286
 287static struct msi_domain_info vmd_msi_domain_info = {
 288        .flags          = MSI_FLAG_USE_DEF_DOM_OPS | MSI_FLAG_USE_DEF_CHIP_OPS |
 289                          MSI_FLAG_PCI_MSIX,
 290        .ops            = &vmd_msi_domain_ops,
 291        .chip           = &vmd_msi_controller,
 292};
 293
 294/*
 295 * VMD replaces the requester ID with its own.  DMA mappings for devices in a
 296 * VMD domain need to be mapped for the VMD, not the device requiring
 297 * the mapping.
 298 */
 299static struct device *to_vmd_dev(struct device *dev)
 300{
 301        struct pci_dev *pdev = to_pci_dev(dev);
 302        struct vmd_dev *vmd = vmd_from_bus(pdev->bus);
 303
 304        return &vmd->dev->dev;
 305}
 306
 307static void *vmd_alloc(struct device *dev, size_t size, dma_addr_t *addr,
 308                       gfp_t flag, unsigned long attrs)
 309{
 310        return dma_alloc_attrs(to_vmd_dev(dev), size, addr, flag, attrs);
 311}
 312
 313static void vmd_free(struct device *dev, size_t size, void *vaddr,
 314                     dma_addr_t addr, unsigned long attrs)
 315{
 316        return dma_free_attrs(to_vmd_dev(dev), size, vaddr, addr, attrs);
 317}
 318
 319static int vmd_mmap(struct device *dev, struct vm_area_struct *vma,
 320                    void *cpu_addr, dma_addr_t addr, size_t size,
 321                    unsigned long attrs)
 322{
 323        return dma_mmap_attrs(to_vmd_dev(dev), vma, cpu_addr, addr, size,
 324                        attrs);
 325}
 326
 327static int vmd_get_sgtable(struct device *dev, struct sg_table *sgt,
 328                           void *cpu_addr, dma_addr_t addr, size_t size,
 329                           unsigned long attrs)
 330{
 331        return dma_get_sgtable_attrs(to_vmd_dev(dev), sgt, cpu_addr, addr, size,
 332                        attrs);
 333}
 334
 335static dma_addr_t vmd_map_page(struct device *dev, struct page *page,
 336                               unsigned long offset, size_t size,
 337                               enum dma_data_direction dir,
 338                               unsigned long attrs)
 339{
 340        return dma_map_page_attrs(to_vmd_dev(dev), page, offset, size, dir,
 341                        attrs);
 342}
 343
 344static void vmd_unmap_page(struct device *dev, dma_addr_t addr, size_t size,
 345                           enum dma_data_direction dir, unsigned long attrs)
 346{
 347        dma_unmap_page_attrs(to_vmd_dev(dev), addr, size, dir, attrs);
 348}
 349
 350static int vmd_map_sg(struct device *dev, struct scatterlist *sg, int nents,
 351                      enum dma_data_direction dir, unsigned long attrs)
 352{
 353        return dma_map_sg_attrs(to_vmd_dev(dev), sg, nents, dir, attrs);
 354}
 355
 356static void vmd_unmap_sg(struct device *dev, struct scatterlist *sg, int nents,
 357                         enum dma_data_direction dir, unsigned long attrs)
 358{
 359        dma_unmap_sg_attrs(to_vmd_dev(dev), sg, nents, dir, attrs);
 360}
 361
 362static void vmd_sync_single_for_cpu(struct device *dev, dma_addr_t addr,
 363                                    size_t size, enum dma_data_direction dir)
 364{
 365        dma_sync_single_for_cpu(to_vmd_dev(dev), addr, size, dir);
 366}
 367
 368static void vmd_sync_single_for_device(struct device *dev, dma_addr_t addr,
 369                                       size_t size, enum dma_data_direction dir)
 370{
 371        dma_sync_single_for_device(to_vmd_dev(dev), addr, size, dir);
 372}
 373
 374static void vmd_sync_sg_for_cpu(struct device *dev, struct scatterlist *sg,
 375                                int nents, enum dma_data_direction dir)
 376{
 377        dma_sync_sg_for_cpu(to_vmd_dev(dev), sg, nents, dir);
 378}
 379
 380static void vmd_sync_sg_for_device(struct device *dev, struct scatterlist *sg,
 381                                   int nents, enum dma_data_direction dir)
 382{
 383        dma_sync_sg_for_device(to_vmd_dev(dev), sg, nents, dir);
 384}
 385
 386static int vmd_dma_supported(struct device *dev, u64 mask)
 387{
 388        return dma_supported(to_vmd_dev(dev), mask);
 389}
 390
 391static u64 vmd_get_required_mask(struct device *dev)
 392{
 393        return dma_get_required_mask(to_vmd_dev(dev));
 394}
 395
 396static void vmd_teardown_dma_ops(struct vmd_dev *vmd)
 397{
 398        struct dma_domain *domain = &vmd->dma_domain;
 399
 400        if (get_dma_ops(&vmd->dev->dev))
 401                del_dma_domain(domain);
 402}
 403
 404#define ASSIGN_VMD_DMA_OPS(source, dest, fn)    \
 405        do {                                    \
 406                if (source->fn)                 \
 407                        dest->fn = vmd_##fn;    \
 408        } while (0)
 409
 410static void vmd_setup_dma_ops(struct vmd_dev *vmd)
 411{
 412        const struct dma_map_ops *source = get_dma_ops(&vmd->dev->dev);
 413        struct dma_map_ops *dest = &vmd->dma_ops;
 414        struct dma_domain *domain = &vmd->dma_domain;
 415
 416        domain->domain_nr = vmd->sysdata.domain;
 417        domain->dma_ops = dest;
 418
 419        if (!source)
 420                return;
 421        ASSIGN_VMD_DMA_OPS(source, dest, alloc);
 422        ASSIGN_VMD_DMA_OPS(source, dest, free);
 423        ASSIGN_VMD_DMA_OPS(source, dest, mmap);
 424        ASSIGN_VMD_DMA_OPS(source, dest, get_sgtable);
 425        ASSIGN_VMD_DMA_OPS(source, dest, map_page);
 426        ASSIGN_VMD_DMA_OPS(source, dest, unmap_page);
 427        ASSIGN_VMD_DMA_OPS(source, dest, map_sg);
 428        ASSIGN_VMD_DMA_OPS(source, dest, unmap_sg);
 429        ASSIGN_VMD_DMA_OPS(source, dest, sync_single_for_cpu);
 430        ASSIGN_VMD_DMA_OPS(source, dest, sync_single_for_device);
 431        ASSIGN_VMD_DMA_OPS(source, dest, sync_sg_for_cpu);
 432        ASSIGN_VMD_DMA_OPS(source, dest, sync_sg_for_device);
 433        ASSIGN_VMD_DMA_OPS(source, dest, dma_supported);
 434        ASSIGN_VMD_DMA_OPS(source, dest, get_required_mask);
 435        add_dma_domain(domain);
 436}
 437#undef ASSIGN_VMD_DMA_OPS
 438
 439static char __iomem *vmd_cfg_addr(struct vmd_dev *vmd, struct pci_bus *bus,
 440                                  unsigned int devfn, int reg, int len)
 441{
 442        char __iomem *addr = vmd->cfgbar +
 443                             (bus->number << 20) + (devfn << 12) + reg;
 444
 445        if ((addr - vmd->cfgbar) + len >=
 446            resource_size(&vmd->dev->resource[VMD_CFGBAR]))
 447                return NULL;
 448
 449        return addr;
 450}
 451
 452/*
 453 * CPU may deadlock if config space is not serialized on some versions of this
 454 * hardware, so all config space access is done under a spinlock.
 455 */
 456static int vmd_pci_read(struct pci_bus *bus, unsigned int devfn, int reg,
 457                        int len, u32 *value)
 458{
 459        struct vmd_dev *vmd = vmd_from_bus(bus);
 460        char __iomem *addr = vmd_cfg_addr(vmd, bus, devfn, reg, len);
 461        unsigned long flags;
 462        int ret = 0;
 463
 464        if (!addr)
 465                return -EFAULT;
 466
 467        spin_lock_irqsave(&vmd->cfg_lock, flags);
 468        switch (len) {
 469        case 1:
 470                *value = readb(addr);
 471                break;
 472        case 2:
 473                *value = readw(addr);
 474                break;
 475        case 4:
 476                *value = readl(addr);
 477                break;
 478        default:
 479                ret = -EINVAL;
 480                break;
 481        }
 482        spin_unlock_irqrestore(&vmd->cfg_lock, flags);
 483        return ret;
 484}
 485
 486/*
 487 * VMD h/w converts non-posted config writes to posted memory writes. The
 488 * read-back in this function forces the completion so it returns only after
 489 * the config space was written, as expected.
 490 */
 491static int vmd_pci_write(struct pci_bus *bus, unsigned int devfn, int reg,
 492                         int len, u32 value)
 493{
 494        struct vmd_dev *vmd = vmd_from_bus(bus);
 495        char __iomem *addr = vmd_cfg_addr(vmd, bus, devfn, reg, len);
 496        unsigned long flags;
 497        int ret = 0;
 498
 499        if (!addr)
 500                return -EFAULT;
 501
 502        spin_lock_irqsave(&vmd->cfg_lock, flags);
 503        switch (len) {
 504        case 1:
 505                writeb(value, addr);
 506                readb(addr);
 507                break;
 508        case 2:
 509                writew(value, addr);
 510                readw(addr);
 511                break;
 512        case 4:
 513                writel(value, addr);
 514                readl(addr);
 515                break;
 516        default:
 517                ret = -EINVAL;
 518                break;
 519        }
 520        spin_unlock_irqrestore(&vmd->cfg_lock, flags);
 521        return ret;
 522}
 523
 524static struct pci_ops vmd_ops = {
 525        .read           = vmd_pci_read,
 526        .write          = vmd_pci_write,
 527};
 528
 529static void vmd_attach_resources(struct vmd_dev *vmd)
 530{
 531        vmd->dev->resource[VMD_MEMBAR1].child = &vmd->resources[1];
 532        vmd->dev->resource[VMD_MEMBAR2].child = &vmd->resources[2];
 533}
 534
 535static void vmd_detach_resources(struct vmd_dev *vmd)
 536{
 537        vmd->dev->resource[VMD_MEMBAR1].child = NULL;
 538        vmd->dev->resource[VMD_MEMBAR2].child = NULL;
 539}
 540
 541/*
 542 * VMD domains start at 0x10000 to not clash with ACPI _SEG domains.
 543 * Per ACPI r6.0, sec 6.5.6,  _SEG returns an integer, of which the lower
 544 * 16 bits are the PCI Segment Group (domain) number.  Other bits are
 545 * currently reserved.
 546 */
 547static int vmd_find_free_domain(void)
 548{
 549        int domain = 0xffff;
 550        struct pci_bus *bus = NULL;
 551
 552        while ((bus = pci_find_next_bus(bus)) != NULL)
 553                domain = max_t(int, domain, pci_domain_nr(bus));
 554        return domain + 1;
 555}
 556
 557static int vmd_enable_domain(struct vmd_dev *vmd, unsigned long features)
 558{
 559        struct pci_sysdata *sd = &vmd->sysdata;
 560        struct fwnode_handle *fn;
 561        struct resource *res;
 562        u32 upper_bits;
 563        unsigned long flags;
 564        LIST_HEAD(resources);
 565        resource_size_t offset[2] = {0};
 566        resource_size_t membar2_offset = 0x2000, busn_start = 0;
 567        struct pci_bus *child;
 568
 569        /*
 570         * Shadow registers may exist in certain VMD device ids which allow
 571         * guests to correctly assign host physical addresses to the root ports
 572         * and child devices. These registers will either return the host value
 573         * or 0, depending on an enable bit in the VMD device.
 574         */
 575        if (features & VMD_FEAT_HAS_MEMBAR_SHADOW) {
 576                u32 vmlock;
 577                int ret;
 578
 579                membar2_offset = 0x2018;
 580                ret = pci_read_config_dword(vmd->dev, PCI_REG_VMLOCK, &vmlock);
 581                if (ret || vmlock == ~0)
 582                        return -ENODEV;
 583
 584                if (MB2_SHADOW_EN(vmlock)) {
 585                        void __iomem *membar2;
 586
 587                        membar2 = pci_iomap(vmd->dev, VMD_MEMBAR2, 0);
 588                        if (!membar2)
 589                                return -ENOMEM;
 590                        offset[0] = vmd->dev->resource[VMD_MEMBAR1].start -
 591                                                readq(membar2 + 0x2008);
 592                        offset[1] = vmd->dev->resource[VMD_MEMBAR2].start -
 593                                                readq(membar2 + 0x2010);
 594                        pci_iounmap(vmd->dev, membar2);
 595                }
 596        }
 597
 598        /*
 599         * Certain VMD devices may have a root port configuration option which
 600         * limits the bus range to between 0-127 or 128-255
 601         */
 602        if (features & VMD_FEAT_HAS_BUS_RESTRICTIONS) {
 603                u32 vmcap, vmconfig;
 604
 605                pci_read_config_dword(vmd->dev, PCI_REG_VMCAP, &vmcap);
 606                pci_read_config_dword(vmd->dev, PCI_REG_VMCONFIG, &vmconfig);
 607                if (BUS_RESTRICT_CAP(vmcap) &&
 608                    (BUS_RESTRICT_CFG(vmconfig) == 0x1))
 609                        busn_start = 128;
 610        }
 611
 612        res = &vmd->dev->resource[VMD_CFGBAR];
 613        vmd->resources[0] = (struct resource) {
 614                .name  = "VMD CFGBAR",
 615                .start = busn_start,
 616                .end   = busn_start + (resource_size(res) >> 20) - 1,
 617                .flags = IORESOURCE_BUS | IORESOURCE_PCI_FIXED,
 618        };
 619
 620        /*
 621         * If the window is below 4GB, clear IORESOURCE_MEM_64 so we can
 622         * put 32-bit resources in the window.
 623         *
 624         * There's no hardware reason why a 64-bit window *couldn't*
 625         * contain a 32-bit resource, but pbus_size_mem() computes the
 626         * bridge window size assuming a 64-bit window will contain no
 627         * 32-bit resources.  __pci_assign_resource() enforces that
 628         * artificial restriction to make sure everything will fit.
 629         *
 630         * The only way we could use a 64-bit non-prefechable MEMBAR is
 631         * if its address is <4GB so that we can convert it to a 32-bit
 632         * resource.  To be visible to the host OS, all VMD endpoints must
 633         * be initially configured by platform BIOS, which includes setting
 634         * up these resources.  We can assume the device is configured
 635         * according to the platform needs.
 636         */
 637        res = &vmd->dev->resource[VMD_MEMBAR1];
 638        upper_bits = upper_32_bits(res->end);
 639        flags = res->flags & ~IORESOURCE_SIZEALIGN;
 640        if (!upper_bits)
 641                flags &= ~IORESOURCE_MEM_64;
 642        vmd->resources[1] = (struct resource) {
 643                .name  = "VMD MEMBAR1",
 644                .start = res->start,
 645                .end   = res->end,
 646                .flags = flags,
 647                .parent = res,
 648        };
 649
 650        res = &vmd->dev->resource[VMD_MEMBAR2];
 651        upper_bits = upper_32_bits(res->end);
 652        flags = res->flags & ~IORESOURCE_SIZEALIGN;
 653        if (!upper_bits)
 654                flags &= ~IORESOURCE_MEM_64;
 655        vmd->resources[2] = (struct resource) {
 656                .name  = "VMD MEMBAR2",
 657                .start = res->start + membar2_offset,
 658                .end   = res->end,
 659                .flags = flags,
 660                .parent = res,
 661        };
 662
 663        sd->vmd_domain = true;
 664        sd->domain = vmd_find_free_domain();
 665        if (sd->domain < 0)
 666                return sd->domain;
 667
 668        sd->node = pcibus_to_node(vmd->dev->bus);
 669
 670        fn = irq_domain_alloc_named_id_fwnode("VMD-MSI", vmd->sysdata.domain);
 671        if (!fn)
 672                return -ENODEV;
 673
 674        vmd->irq_domain = pci_msi_create_irq_domain(fn, &vmd_msi_domain_info,
 675                                                    x86_vector_domain);
 676        irq_domain_free_fwnode(fn);
 677        if (!vmd->irq_domain)
 678                return -ENODEV;
 679
 680        pci_add_resource(&resources, &vmd->resources[0]);
 681        pci_add_resource_offset(&resources, &vmd->resources[1], offset[0]);
 682        pci_add_resource_offset(&resources, &vmd->resources[2], offset[1]);
 683
 684        vmd->bus = pci_create_root_bus(&vmd->dev->dev, busn_start, &vmd_ops,
 685                                       sd, &resources);
 686        if (!vmd->bus) {
 687                pci_free_resource_list(&resources);
 688                irq_domain_remove(vmd->irq_domain);
 689                return -ENODEV;
 690        }
 691
 692        vmd_attach_resources(vmd);
 693        vmd_setup_dma_ops(vmd);
 694        dev_set_msi_domain(&vmd->bus->dev, vmd->irq_domain);
 695
 696        pci_scan_child_bus(vmd->bus);
 697        pci_assign_unassigned_bus_resources(vmd->bus);
 698
 699        /*
 700         * VMD root buses are virtual and don't return true on pci_is_pcie()
 701         * and will fail pcie_bus_configure_settings() early. It can instead be
 702         * run on each of the real root ports.
 703         */
 704        list_for_each_entry(child, &vmd->bus->children, node)
 705                pcie_bus_configure_settings(child);
 706
 707        pci_bus_add_devices(vmd->bus);
 708
 709        WARN(sysfs_create_link(&vmd->dev->dev.kobj, &vmd->bus->dev.kobj,
 710                               "domain"), "Can't create symlink to domain\n");
 711        return 0;
 712}
 713
 714static irqreturn_t vmd_irq(int irq, void *data)
 715{
 716        struct vmd_irq_list *irqs = data;
 717        struct vmd_irq *vmdirq;
 718        int idx;
 719
 720        idx = srcu_read_lock(&irqs->srcu);
 721        list_for_each_entry_rcu(vmdirq, &irqs->irq_list, node)
 722                generic_handle_irq(vmdirq->virq);
 723        srcu_read_unlock(&irqs->srcu, idx);
 724
 725        return IRQ_HANDLED;
 726}
 727
 728static int vmd_probe(struct pci_dev *dev, const struct pci_device_id *id)
 729{
 730        struct vmd_dev *vmd;
 731        int i, err;
 732
 733        if (resource_size(&dev->resource[VMD_CFGBAR]) < (1 << 20))
 734                return -ENOMEM;
 735
 736        vmd = devm_kzalloc(&dev->dev, sizeof(*vmd), GFP_KERNEL);
 737        if (!vmd)
 738                return -ENOMEM;
 739
 740        vmd->dev = dev;
 741        err = pcim_enable_device(dev);
 742        if (err < 0)
 743                return err;
 744
 745        vmd->cfgbar = pcim_iomap(dev, VMD_CFGBAR, 0);
 746        if (!vmd->cfgbar)
 747                return -ENOMEM;
 748
 749        pci_set_master(dev);
 750        if (dma_set_mask_and_coherent(&dev->dev, DMA_BIT_MASK(64)) &&
 751            dma_set_mask_and_coherent(&dev->dev, DMA_BIT_MASK(32)))
 752                return -ENODEV;
 753
 754        vmd->msix_count = pci_msix_vec_count(dev);
 755        if (vmd->msix_count < 0)
 756                return -ENODEV;
 757
 758        vmd->msix_count = pci_alloc_irq_vectors(dev, 1, vmd->msix_count,
 759                                        PCI_IRQ_MSIX);
 760        if (vmd->msix_count < 0)
 761                return vmd->msix_count;
 762
 763        vmd->irqs = devm_kcalloc(&dev->dev, vmd->msix_count, sizeof(*vmd->irqs),
 764                                 GFP_KERNEL);
 765        if (!vmd->irqs)
 766                return -ENOMEM;
 767
 768        for (i = 0; i < vmd->msix_count; i++) {
 769                err = init_srcu_struct(&vmd->irqs[i].srcu);
 770                if (err)
 771                        return err;
 772
 773                INIT_LIST_HEAD(&vmd->irqs[i].irq_list);
 774                err = devm_request_irq(&dev->dev, pci_irq_vector(dev, i),
 775                                       vmd_irq, IRQF_NO_THREAD,
 776                                       "vmd", &vmd->irqs[i]);
 777                if (err)
 778                        return err;
 779        }
 780
 781        spin_lock_init(&vmd->cfg_lock);
 782        pci_set_drvdata(dev, vmd);
 783        err = vmd_enable_domain(vmd, (unsigned long) id->driver_data);
 784        if (err)
 785                return err;
 786
 787        dev_info(&vmd->dev->dev, "Bound to PCI domain %04x\n",
 788                 vmd->sysdata.domain);
 789        return 0;
 790}
 791
 792static void vmd_cleanup_srcu(struct vmd_dev *vmd)
 793{
 794        int i;
 795
 796        for (i = 0; i < vmd->msix_count; i++)
 797                cleanup_srcu_struct(&vmd->irqs[i].srcu);
 798}
 799
 800static void vmd_remove(struct pci_dev *dev)
 801{
 802        struct vmd_dev *vmd = pci_get_drvdata(dev);
 803
 804        sysfs_remove_link(&vmd->dev->dev.kobj, "domain");
 805        pci_stop_root_bus(vmd->bus);
 806        pci_remove_root_bus(vmd->bus);
 807        vmd_cleanup_srcu(vmd);
 808        vmd_teardown_dma_ops(vmd);
 809        vmd_detach_resources(vmd);
 810        irq_domain_remove(vmd->irq_domain);
 811}
 812
 813#ifdef CONFIG_PM_SLEEP
 814static int vmd_suspend(struct device *dev)
 815{
 816        struct pci_dev *pdev = to_pci_dev(dev);
 817        struct vmd_dev *vmd = pci_get_drvdata(pdev);
 818        int i;
 819
 820        for (i = 0; i < vmd->msix_count; i++)
 821                devm_free_irq(dev, pci_irq_vector(pdev, i), &vmd->irqs[i]);
 822
 823        pci_save_state(pdev);
 824        return 0;
 825}
 826
 827static int vmd_resume(struct device *dev)
 828{
 829        struct pci_dev *pdev = to_pci_dev(dev);
 830        struct vmd_dev *vmd = pci_get_drvdata(pdev);
 831        int err, i;
 832
 833        for (i = 0; i < vmd->msix_count; i++) {
 834                err = devm_request_irq(dev, pci_irq_vector(pdev, i),
 835                                       vmd_irq, IRQF_NO_THREAD,
 836                                       "vmd", &vmd->irqs[i]);
 837                if (err)
 838                        return err;
 839        }
 840
 841        pci_restore_state(pdev);
 842        return 0;
 843}
 844#endif
 845static SIMPLE_DEV_PM_OPS(vmd_dev_pm_ops, vmd_suspend, vmd_resume);
 846
 847static const struct pci_device_id vmd_ids[] = {
 848        {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_VMD_201D),},
 849        {PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_VMD_28C0),
 850                .driver_data = VMD_FEAT_HAS_MEMBAR_SHADOW |
 851                                VMD_FEAT_HAS_BUS_RESTRICTIONS,},
 852        {0,}
 853};
 854MODULE_DEVICE_TABLE(pci, vmd_ids);
 855
 856static struct pci_driver vmd_drv = {
 857        .name           = "vmd",
 858        .id_table       = vmd_ids,
 859        .probe          = vmd_probe,
 860        .remove         = vmd_remove,
 861        .driver         = {
 862                .pm     = &vmd_dev_pm_ops,
 863        },
 864};
 865module_pci_driver(vmd_drv);
 866
 867MODULE_AUTHOR("Intel Corporation");
 868MODULE_LICENSE("GPL v2");
 869MODULE_VERSION("0.6");
 870