drivers/pci/msi.c

   1 /*
   2  * File:        msi.c
   3  * Purpose:     PCI Message Signaled Interrupt (MSI)
   4  *
   5  * Copyright (C) 2003-2004 Intel
   6  * Copyright (C) Tom Long Nguyen (tom.l.nguyen@intel.com)
   7  */
   8
   9 #include <linux/err.h>
  10 #include <linux/mm.h>
  11 #include <linux/irq.h>
  12 #include <linux/interrupt.h>
  13 #include <linux/init.h>
  14 #include <linux/ioport.h>
  15 #include <linux/pci.h>
  16 #include <linux/proc_fs.h>
  17 #include <linux/msi.h>
  18 #include <linux/smp.h>
  19 #include <linux/errno.h>
  20 #include <linux/io.h>
  21 #include <linux/slab.h>
  22
  23 #include "pci.h"
  24 #include "msi.h"
  25
  26 static int pci_msi_enable = 1;
  27
  28 /* Arch hooks */
  29
  30 #ifndef arch_msi_check_device
  31 int arch_msi_check_device(struct pci_dev *dev, int nvec, int type)
  32 {
  33         return 0;
  34 }
  35 #endif
  36
  37 #ifndef arch_setup_msi_irqs
  38 # define arch_setup_msi_irqs default_setup_msi_irqs
  39 # define HAVE_DEFAULT_MSI_SETUP_IRQS
  40 #endif
  41
  42 #ifdef HAVE_DEFAULT_MSI_SETUP_IRQS
  43 int default_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
  44 {
  45         struct msi_desc *entry;
  46         int ret;
  47
  48         /*
  49          * If an architecture wants to support multiple MSI, it needs to
  50          * override arch_setup_msi_irqs()
  51          */
  52         if (type == PCI_CAP_ID_MSI && nvec > 1)
  53                 return 1;
  54
  55         list_for_each_entry(entry, &dev->msi_list, list) {
  56                 ret = arch_setup_msi_irq(dev, entry);
  57                 if (ret < 0)
  58                         return ret;
  59                 if (ret > 0)
  60                         return -ENOSPC;
  61         }
  62
  63         return 0;
  64 }
  65 #endif
  66
  67 #ifndef arch_teardown_msi_irqs
  68 # define arch_teardown_msi_irqs default_teardown_msi_irqs
  69 # define HAVE_DEFAULT_MSI_TEARDOWN_IRQS
  70 #endif
  71
  72 #ifdef HAVE_DEFAULT_MSI_TEARDOWN_IRQS
  73 void default_teardown_msi_irqs(struct pci_dev *dev)
  74 {
  75         struct msi_desc *entry;
  76
  77         list_for_each_entry(entry, &dev->msi_list, list) {
  78                 int i, nvec;
  79                 if (entry->irq == 0)
  80                         continue;
  81                 nvec = 1 << entry->msi_attrib.multiple;
  82                 for (i = 0; i < nvec; i++)
  83                         arch_teardown_msi_irq(entry->irq + i);
  84         }
  85 }
  86 #endif
  87
  88 static void msi_set_enable(struct pci_dev *dev, int pos, int enable)
  89 {
  90         u16 control;
  91
  92         BUG_ON(!pos);
  93
  94         pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &control);
  95         control &= ~PCI_MSI_FLAGS_ENABLE;
  96         if (enable)
  97                 control |= PCI_MSI_FLAGS_ENABLE;
  98         pci_write_config_word(dev, pos + PCI_MSI_FLAGS, control);
  99 }
 100
 101 static void msix_set_enable(struct pci_dev *dev, int enable)
 102 {
 103         int pos;
 104         u16 control;
 105
 106         pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
 107         if (pos) {
 108                 pci_read_config_word(dev, pos + PCI_MSIX_FLAGS, &control);
 109                 control &= ~PCI_MSIX_FLAGS_ENABLE;
 110                 if (enable)
 111                         control |= PCI_MSIX_FLAGS_ENABLE;
 112                 pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
 113         }
 114 }
 115
 116 static inline __attribute_const__ u32 msi_mask(unsigned x)
 117 {
 118         /* Don't shift by >= width of type */
 119         if (x >= 5)
 120                 return 0xffffffff;
 121         return (1 << (1 << x)) - 1;
 122 }
 123
 124 static inline __attribute_const__ u32 msi_capable_mask(u16 control)
 125 {
 126         return msi_mask((control >> 1) & 7);
 127 }
 128
 129 static inline __attribute_const__ u32 msi_enabled_mask(u16 control)
 130 {
 131         return msi_mask((control >> 4) & 7);
 132 }
 133
 134 /*
 135  * PCI 2.3 does not specify mask bits for each MSI interrupt.  Attempting to
 136  * mask all MSI interrupts by clearing the MSI enable bit does not work
 137  * reliably as devices without an INTx disable bit will then generate a
 138  * level IRQ which will never be cleared.
 139  */
 140 static u32 __msi_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
 141 {
 142         u32 mask_bits = desc->masked;
 143
 144         if (!desc->msi_attrib.maskbit)
 145                 return 0;
 146
 147         mask_bits &= ~mask;
 148         mask_bits |= flag;
 149         pci_write_config_dword(desc->dev, desc->mask_pos, mask_bits);
 150
 151         return mask_bits;
 152 }
 153
 154 static void msi_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
 155 {
 156         desc->masked = __msi_mask_irq(desc, mask, flag);
 157 }
 158
 159 /*
 160  * This internal function does not flush PCI writes to the device.
 161  * All users must ensure that they read from the device before either
 162  * assuming that the device state is up to date, or returning out of this
 163  * file.  This saves a few milliseconds when initialising devices with lots
 164  * of MSI-X interrupts.
 165  */
 166 static u32 __msix_mask_irq(struct msi_desc *desc, u32 flag)
 167 {
 168         u32 mask_bits = desc->masked;
 169         unsigned offset = desc->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE +
 170                                                 PCI_MSIX_ENTRY_VECTOR_CTRL;
 171         mask_bits &= ~PCI_MSIX_ENTRY_CTRL_MASKBIT;
 172         if (flag)
 173                 mask_bits |= PCI_MSIX_ENTRY_CTRL_MASKBIT;
 174         writel(mask_bits, desc->mask_base + offset);
 175
 176         return mask_bits;
 177 }
 178
 179 static void msix_mask_irq(struct msi_desc *desc, u32 flag)
 180 {
 181         desc->masked = __msix_mask_irq(desc, flag);
 182 }
 183
 184 static void msi_set_mask_bit(struct irq_data *data, u32 flag)
 185 {
 186         struct msi_desc *desc = irq_data_get_msi(data);
 187
 188         if (desc->msi_attrib.is_msix) {
 189                 msix_mask_irq(desc, flag);
 190                 readl(desc->mask_base);         /* Flush write to device */
 191         } else {
 192                 unsigned offset = data->irq - desc->dev->irq;
 193                 msi_mask_irq(desc, 1 << offset, flag << offset);
 194         }
 195 }
 196
 197 void mask_msi_irq(struct irq_data *data)
 198 {
 199         msi_set_mask_bit(data, 1);
 200 }
 201
 202 void unmask_msi_irq(struct irq_data *data)
 203 {
 204         msi_set_mask_bit(data, 0);
 205 }
 206
 207 void __read_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
 208 {
 209         BUG_ON(entry->dev->current_state != PCI_D0);
 210
 211         if (entry->msi_attrib.is_msix) {
 212                 void __iomem *base = entry->mask_base +
 213                         entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
 214
 215                 msg->address_lo = readl(base + PCI_MSIX_ENTRY_LOWER_ADDR);
 216                 msg->address_hi = readl(base + PCI_MSIX_ENTRY_UPPER_ADDR);
 217                 msg->data = readl(base + PCI_MSIX_ENTRY_DATA);
 218         } else {
 219                 struct pci_dev *dev = entry->dev;
 220                 int pos = entry->msi_attrib.pos;
 221                 u16 data;
 222
 223                 pci_read_config_dword(dev, msi_lower_address_reg(pos),
 224                                         &msg->address_lo);
 225                 if (entry->msi_attrib.is_64) {
 226                         pci_read_config_dword(dev, msi_upper_address_reg(pos),
 227                                                 &msg->address_hi);
 228                         pci_read_config_word(dev, msi_data_reg(pos, 1), &data);
 229                 } else {
 230                         msg->address_hi = 0;
 231                         pci_read_config_word(dev, msi_data_reg(pos, 0), &data);
 232                 }
 233                 msg->data = data;
 234         }
 235 }
 236
 237 void read_msi_msg(unsigned int irq, struct msi_msg *msg)
 238 {
 239         struct msi_desc *entry = irq_get_msi_desc(irq);
 240
 241         __read_msi_msg(entry, msg);
 242 }
 243
 244 void __get_cached_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
 245 {
 246         /* Assert that the cache is valid, assuming that
 247          * valid messages are not all-zeroes. */
 248         BUG_ON(!(entry->msg.address_hi | entry->msg.address_lo |
 249                  entry->msg.data));
 250
 251         *msg = entry->msg;
 252 }
 253
 254 void get_cached_msi_msg(unsigned int irq, struct msi_msg *msg)
 255 {
 256         struct msi_desc *entry = irq_get_msi_desc(irq);
 257
 258         __get_cached_msi_msg(entry, msg);
 259 }
 260
 261 void __write_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
 262 {
 263         if (entry->dev->current_state != PCI_D0) {
 264                 /* Don't touch the hardware now */
 265         } else if (entry->msi_attrib.is_msix) {
 266                 void __iomem *base;
 267                 base = entry->mask_base +
 268                         entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
 269
 270                 writel(msg->address_lo, base + PCI_MSIX_ENTRY_LOWER_ADDR);
 271                 writel(msg->address_hi, base + PCI_MSIX_ENTRY_UPPER_ADDR);
 272                 writel(msg->data, base + PCI_MSIX_ENTRY_DATA);
 273         } else {
 274                 struct pci_dev *dev = entry->dev;
 275                 int pos = entry->msi_attrib.pos;
 276                 u16 msgctl;
 277
 278                 pci_read_config_word(dev, msi_control_reg(pos), &msgctl);
 279                 msgctl &= ~PCI_MSI_FLAGS_QSIZE;
 280                 msgctl |= entry->msi_attrib.multiple << 4;
 281                 pci_write_config_word(dev, msi_control_reg(pos), msgctl);
 282
 283                 pci_write_config_dword(dev, msi_lower_address_reg(pos),
 284                                         msg->address_lo);
 285                 if (entry->msi_attrib.is_64) {
 286                         pci_write_config_dword(dev, msi_upper_address_reg(pos),
 287                                                 msg->address_hi);
 288                         pci_write_config_word(dev, msi_data_reg(pos, 1),
 289                                                 msg->data);
 290                 } else {
 291                         pci_write_config_word(dev, msi_data_reg(pos, 0),
 292                                                 msg->data);
 293                 }
 294         }
 295         entry->msg = *msg;
 296 }
 297
 298 void write_msi_msg(unsigned int irq, struct msi_msg *msg)
 299 {
 300         struct msi_desc *entry = irq_get_msi_desc(irq);
 301
 302         __write_msi_msg(entry, msg);
 303 }
 304
 305 static void free_msi_irqs(struct pci_dev *dev)
 306 {
 307         struct msi_desc *entry, *tmp;
 308
 309         list_for_each_entry(entry, &dev->msi_list, list) {
 310                 int i, nvec;
 311                 if (!entry->irq)
 312                         continue;
 313                 nvec = 1 << entry->msi_attrib.multiple;
 314                 for (i = 0; i < nvec; i++)
 315                         BUG_ON(irq_has_action(entry->irq + i));
 316         }
 317
 318         arch_teardown_msi_irqs(dev);
 319
 320         list_for_each_entry_safe(entry, tmp, &dev->msi_list, list) {
 321                 if (entry->msi_attrib.is_msix) {
 322                         if (list_is_last(&entry->list, &dev->msi_list))
 323                                 iounmap(entry->mask_base);
 324                 }
 325                 list_del(&entry->list);
 326                 kfree(entry);
 327         }
 328 }
 329
 330 static struct msi_desc *alloc_msi_entry(struct pci_dev *dev)
 331 {
 332         struct msi_desc *desc = kzalloc(sizeof(*desc), GFP_KERNEL);
 333         if (!desc)
 334                 return NULL;
 335
 336         INIT_LIST_HEAD(&desc->list);
 337         desc->dev = dev;
 338
 339         return desc;
 340 }
 341
 342 static void pci_intx_for_msi(struct pci_dev *dev, int enable)
 343 {
 344         if (!(dev->dev_flags & PCI_DEV_FLAGS_MSI_INTX_DISABLE_BUG))
 345                 pci_intx(dev, enable);
 346 }
 347
 348 static void __pci_restore_msi_state(struct pci_dev *dev)
 349 {
 350         int pos;
 351         u16 control;
 352         struct msi_desc *entry;
 353
 354         if (!dev->msi_enabled)
 355                 return;
 356
 357         entry = irq_get_msi_desc(dev->irq);
 358         pos = entry->msi_attrib.pos;
 359
 360         pci_intx_for_msi(dev, 0);
 361         msi_set_enable(dev, pos, 0);
 362         write_msi_msg(dev->irq, &entry->msg);
 363
 364         pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &control);
 365         msi_mask_irq(entry, msi_capable_mask(control), entry->masked);
 366         control &= ~PCI_MSI_FLAGS_QSIZE;
 367         control |= (entry->msi_attrib.multiple << 4) | PCI_MSI_FLAGS_ENABLE;
 368         pci_write_config_word(dev, pos + PCI_MSI_FLAGS, control);
 369 }
 370
 371 static void __pci_restore_msix_state(struct pci_dev *dev)
 372 {
 373         int pos;
 374         struct msi_desc *entry;
 375         u16 control;
 376
 377         if (!dev->msix_enabled)
 378                 return;
 379         BUG_ON(list_empty(&dev->msi_list));
 380         entry = list_first_entry(&dev->msi_list, struct msi_desc, list);
 381         pos = entry->msi_attrib.pos;
 382         pci_read_config_word(dev, pos + PCI_MSIX_FLAGS, &control);
 383
 384         /* route the table */
 385         pci_intx_for_msi(dev, 0);
 386         control |= PCI_MSIX_FLAGS_ENABLE | PCI_MSIX_FLAGS_MASKALL;
 387         pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
 388
 389         list_for_each_entry(entry, &dev->msi_list, list) {
 390                 write_msi_msg(entry->irq, &entry->msg);
 391                 msix_mask_irq(entry, entry->masked);
 392         }
 393
 394         control &= ~PCI_MSIX_FLAGS_MASKALL;
 395         pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
 396 }
 397
 398 void pci_restore_msi_state(struct pci_dev *dev)
 399 {
 400         __pci_restore_msi_state(dev);
 401         __pci_restore_msix_state(dev);
 402 }
 403 EXPORT_SYMBOL_GPL(pci_restore_msi_state);
 404
 405 /**
 406  * msi_capability_init - configure device's MSI capability structure
 407  * @dev: pointer to the pci_dev data structure of MSI device function
 408  * @nvec: number of interrupts to allocate
 409  *
 410  * Setup the MSI capability structure of the device with the requested
 411  * number of interrupts.  A return value of zero indicates the successful
 412  * setup of an entry with the new MSI irq.  A negative return value indicates
 413  * an error, and a positive return value indicates the number of interrupts
 414  * which could have been allocated.
 415  */
 416 static int msi_capability_init(struct pci_dev *dev, int nvec)
 417 {
 418         struct msi_desc *entry;
 419         int pos, ret;
 420         u16 control;
 421         unsigned mask;
 422
 423         pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
 424         msi_set_enable(dev, pos, 0);    /* Disable MSI during set up */
 425
 426         pci_read_config_word(dev, msi_control_reg(pos), &control);
 427         /* MSI Entry Initialization */
 428         entry = alloc_msi_entry(dev);
 429         if (!entry)
 430                 return -ENOMEM;
 431
 432         entry->msi_attrib.is_msix       = 0;
 433         entry->msi_attrib.is_64         = is_64bit_address(control);
 434         entry->msi_attrib.entry_nr      = 0;
 435         entry->msi_attrib.maskbit       = is_mask_bit_support(control);
 436         entry->msi_attrib.default_irq   = dev->irq;     /* Save IOAPIC IRQ */
 437         entry->msi_attrib.pos           = pos;
 438
 439         entry->mask_pos = msi_mask_reg(pos, entry->msi_attrib.is_64);
 440         /* All MSIs are unmasked by default, Mask them all */
 441         if (entry->msi_attrib.maskbit)
 442                 pci_read_config_dword(dev, entry->mask_pos, &entry->masked);
 443         mask = msi_capable_mask(control);
 444         msi_mask_irq(entry, mask, mask);
 445
 446         list_add_tail(&entry->list, &dev->msi_list);
 447
 448         /* Configure MSI capability structure */
 449         ret = arch_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSI);
 450         if (ret) {
 451                 msi_mask_irq(entry, mask, ~mask);
 452                 free_msi_irqs(dev);
 453                 return ret;
 454         }
 455
 456         /* Set MSI enabled bits  */
 457         pci_intx_for_msi(dev, 0);
 458         msi_set_enable(dev, pos, 1);
 459         dev->msi_enabled = 1;
 460
 461         dev->irq = entry->irq;
 462         return 0;
 463 }
 464
 465 static void __iomem *msix_map_region(struct pci_dev *dev, unsigned pos,
 466                                                         unsigned nr_entries)
 467 {
 468         resource_size_t phys_addr;
 469         u32 table_offset;
 470         u8 bir;
 471
 472         pci_read_config_dword(dev, msix_table_offset_reg(pos), &table_offset);
 473         bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
 474         table_offset &= ~PCI_MSIX_FLAGS_BIRMASK;
 475         phys_addr = pci_resource_start(dev, bir) + table_offset;
 476
 477         return ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
 478 }
 479
 480 static int msix_setup_entries(struct pci_dev *dev, unsigned pos,
 481                                 void __iomem *base, struct msix_entry *entries,
 482                                 int nvec)
 483 {
 484         struct msi_desc *entry;
 485         int i;
 486
 487         for (i = 0; i < nvec; i++) {
 488                 entry = alloc_msi_entry(dev);
 489                 if (!entry) {
 490                         if (!i)
 491                                 iounmap(base);
 492                         else
 493                                 free_msi_irqs(dev);
 494                         /* No enough memory. Don't try again */
 495                         return -ENOMEM;
 496                 }
 497
 498                 entry->msi_attrib.is_msix       = 1;
 499                 entry->msi_attrib.is_64         = 1;
 500                 entry->msi_attrib.entry_nr      = entries[i].entry;
 501                 entry->msi_attrib.default_irq   = dev->irq;
 502                 entry->msi_attrib.pos           = pos;
 503                 entry->mask_base                = base;
 504
 505                 list_add_tail(&entry->list, &dev->msi_list);
 506         }
 507
 508         return 0;
 509 }
 510
 511 static void msix_program_entries(struct pci_dev *dev,
 512                                         struct msix_entry *entries)
 513 {
 514         struct msi_desc *entry;
 515         int i = 0;
 516
 517         list_for_each_entry(entry, &dev->msi_list, list) {
 518                 int offset = entries[i].entry * PCI_MSIX_ENTRY_SIZE +
 519                                                 PCI_MSIX_ENTRY_VECTOR_CTRL;
 520
 521                 entries[i].vector = entry->irq;
 522                 irq_set_msi_desc(entry->irq, entry);
 523                 entry->masked = readl(entry->mask_base + offset);
 524                 msix_mask_irq(entry, 1);
 525                 i++;
 526         }
 527 }
 528
 529 /**
 530  * msix_capability_init - configure device's MSI-X capability
 531  * @dev: pointer to the pci_dev data structure of MSI-X device function
 532  * @entries: pointer to an array of struct msix_entry entries
 533  * @nvec: number of @entries
 534  *
 535  * Setup the MSI-X capability structure of device function with a
 536  * single MSI-X irq. A return of zero indicates the successful setup of
 537  * requested MSI-X entries with allocated irqs or non-zero for otherwise.
 538  **/
 539 static int msix_capability_init(struct pci_dev *dev,
 540                                 struct msix_entry *entries, int nvec)
 541 {
 542         int pos, ret;
 543         u16 control;
 544         void __iomem *base;
 545
 546         pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
 547         pci_read_config_word(dev, pos + PCI_MSIX_FLAGS, &control);
 548
 549         /* Ensure MSI-X is disabled while it is set up */
 550         control &= ~PCI_MSIX_FLAGS_ENABLE;
 551         pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
 552
 553         /* Request & Map MSI-X table region */
 554         base = msix_map_region(dev, pos, multi_msix_capable(control));
 555         if (!base)
 556                 return -ENOMEM;
 557
 558         ret = msix_setup_entries(dev, pos, base, entries, nvec);
 559         if (ret)
 560                 return ret;
 561
 562         ret = arch_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSIX);
 563         if (ret)
 564                 goto error;
 565
 566         /*
 567          * Some devices require MSI-X to be enabled before we can touch the
 568          * MSI-X registers.  We need to mask all the vectors to prevent
 569          * interrupts coming in before they're fully set up.
 570          */
 571         control |= PCI_MSIX_FLAGS_MASKALL | PCI_MSIX_FLAGS_ENABLE;
 572         pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
 573
 574         msix_program_entries(dev, entries);
 575
 576         /* Set MSI-X enabled bits and unmask the function */
 577         pci_intx_for_msi(dev, 0);
 578         dev->msix_enabled = 1;
 579
 580         control &= ~PCI_MSIX_FLAGS_MASKALL;
 581         pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
 582
 583         return 0;
 584
 585 error:
 586         if (ret < 0) {
 587                 /*
 588                  * If we had some success, report the number of irqs
 589                  * we succeeded in setting up.
 590                  */
 591                 struct msi_desc *entry;
 592                 int avail = 0;
 593
 594                 list_for_each_entry(entry, &dev->msi_list, list) {
 595                         if (entry->irq != 0)
 596                                 avail++;
 597                 }
 598                 if (avail != 0)
 599                         ret = avail;
 600         }
 601
 602         free_msi_irqs(dev);
 603
 604         return ret;
 605 }
 606
 607 /**
 608  * pci_msi_check_device - check whether MSI may be enabled on a device
 609  * @dev: pointer to the pci_dev data structure of MSI device function
 610  * @nvec: how many MSIs have been requested ?
 611  * @type: are we checking for MSI or MSI-X ?
 612  *
 613  * Look at global flags, the device itself, and its parent busses
 614  * to determine if MSI/-X are supported for the device. If MSI/-X is
 615  * supported return 0, else return an error code.
 616  **/
 617 static int pci_msi_check_device(struct pci_dev *dev, int nvec, int type)
 618 {
 619         struct pci_bus *bus;
 620         int ret;
 621
 622         /* MSI must be globally enabled and supported by the device */
 623         if (!pci_msi_enable || !dev || dev->no_msi)
 624                 return -EINVAL;
 625
 626         /*
 627          * You can't ask to have 0 or less MSIs configured.
 628          *  a) it's stupid ..
 629          *  b) the list manipulation code assumes nvec >= 1.
 630          */
 631         if (nvec < 1)
 632                 return -ERANGE;
 633
 634         /*
 635          * Any bridge which does NOT route MSI transactions from its
 636          * secondary bus to its primary bus must set NO_MSI flag on
 637          * the secondary pci_bus.
 638          * We expect only arch-specific PCI host bus controller driver
 639          * or quirks for specific PCI bridges to be setting NO_MSI.
 640          */
 641         for (bus = dev->bus; bus; bus = bus->parent)
 642                 if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
 643                         return -EINVAL;
 644
 645         ret = arch_msi_check_device(dev, nvec, type);
 646         if (ret)
 647                 return ret;
 648
 649         if (!pci_find_capability(dev, type))
 650                 return -EINVAL;
 651
 652         return 0;
 653 }
 654
 655 /**
 656  * pci_enable_msi_block - configure device's MSI capability structure
 657  * @dev: device to configure
 658  * @nvec: number of interrupts to configure
 659  *
 660  * Allocate IRQs for a device with the MSI capability.
 661  * This function returns a negative errno if an error occurs.  If it
 662  * is unable to allocate the number of interrupts requested, it returns
 663  * the number of interrupts it might be able to allocate.  If it successfully
 664  * allocates at least the number of interrupts requested, it returns 0 and
 665  * updates the @dev's irq member to the lowest new interrupt number; the
 666  * other interrupt numbers allocated to this device are consecutive.
 667  */
 668 int pci_enable_msi_block(struct pci_dev *dev, unsigned int nvec)
 669 {
 670         int status, pos, maxvec;
 671         u16 msgctl;
 672
 673         pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
 674         if (!pos)
 675                 return -EINVAL;
 676         pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &msgctl);
 677         maxvec = 1 << ((msgctl & PCI_MSI_FLAGS_QMASK) >> 1);
 678         if (nvec > maxvec)
 679                 return maxvec;
 680
 681         status = pci_msi_check_device(dev, nvec, PCI_CAP_ID_MSI);
 682         if (status)
 683                 return status;
 684
 685         WARN_ON(!!dev->msi_enabled);
 686
 687         /* Check whether driver already requested MSI-X irqs */
 688         if (dev->msix_enabled) {
 689                 dev_info(&dev->dev, "can't enable MSI "
 690                          "(MSI-X already enabled)\n");
 691                 return -EINVAL;
 692         }
 693
 694         status = msi_capability_init(dev, nvec);
 695         return status;
 696 }
 697 EXPORT_SYMBOL(pci_enable_msi_block);
 698
 699 void pci_msi_shutdown(struct pci_dev *dev)
 700 {
 701         struct msi_desc *desc;
 702         u32 mask;
 703         u16 ctrl;
 704         unsigned pos;
 705
 706         if (!pci_msi_enable || !dev || !dev->msi_enabled)
 707                 return;
 708
 709         BUG_ON(list_empty(&dev->msi_list));
 710         desc = list_first_entry(&dev->msi_list, struct msi_desc, list);
 711         pos = desc->msi_attrib.pos;
 712
 713         msi_set_enable(dev, pos, 0);
 714         pci_intx_for_msi(dev, 1);
 715         dev->msi_enabled = 0;
 716
 717         /* Return the device with MSI unmasked as initial states */
 718         pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &ctrl);
 719         mask = msi_capable_mask(ctrl);
 720         /* Keep cached state to be restored */
 721         __msi_mask_irq(desc, mask, ~mask);
 722
 723         /* Restore dev->irq to its default pin-assertion irq */
 724         dev->irq = desc->msi_attrib.default_irq;
 725 }
 726
 727 void pci_disable_msi(struct pci_dev *dev)
 728 {
 729         if (!pci_msi_enable || !dev || !dev->msi_enabled)
 730                 return;
 731
 732         pci_msi_shutdown(dev);
 733         free_msi_irqs(dev);
 734 }
 735 EXPORT_SYMBOL(pci_disable_msi);
 736
 737 /**
 738  * pci_msix_table_size - return the number of device's MSI-X table entries
 739  * @dev: pointer to the pci_dev data structure of MSI-X device function
 740  */
 741 int pci_msix_table_size(struct pci_dev *dev)
 742 {
 743         int pos;
 744         u16 control;
 745
 746         pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
 747         if (!pos)
 748                 return 0;
 749
 750         pci_read_config_word(dev, msi_control_reg(pos), &control);
 751         return multi_msix_capable(control);
 752 }
 753
 754 /**
 755  * pci_enable_msix - configure device's MSI-X capability structure
 756  * @dev: pointer to the pci_dev data structure of MSI-X device function
 757  * @entries: pointer to an array of MSI-X entries
 758  * @nvec: number of MSI-X irqs requested for allocation by device driver
 759  *
 760  * Setup the MSI-X capability structure of device function with the number
 761  * of requested irqs upon its software driver call to request for
 762  * MSI-X mode enabled on its hardware device function. A return of zero
 763  * indicates the successful configuration of MSI-X capability structure
 764  * with new allocated MSI-X irqs. A return of < 0 indicates a failure.
 765  * Or a return of > 0 indicates that driver request is exceeding the number
 766  * of irqs or MSI-X vectors available. Driver should use the returned value to
 767  * re-send its request.
 768  **/
 769 int pci_enable_msix(struct pci_dev *dev, struct msix_entry *entries, int nvec)
 770 {
 771         int status, nr_entries;
 772         int i, j;
 773
 774         if (!entries)
 775                 return -EINVAL;
 776
 777         status = pci_msi_check_device(dev, nvec, PCI_CAP_ID_MSIX);
 778         if (status)
 779                 return status;
 780
 781         nr_entries = pci_msix_table_size(dev);
 782         if (nvec > nr_entries)
 783                 return nr_entries;
 784
 785         /* Check for any invalid entries */
 786         for (i = 0; i < nvec; i++) {
 787                 if (entries[i].entry >= nr_entries)
 788                         return -EINVAL;         /* invalid entry */
 789                 for (j = i + 1; j < nvec; j++) {
 790                         if (entries[i].entry == entries[j].entry)
 791                                 return -EINVAL; /* duplicate entry */
 792                 }
 793         }
 794         WARN_ON(!!dev->msix_enabled);
 795
 796         /* Check whether driver already requested for MSI irq */
 797         if (dev->msi_enabled) {
 798                 dev_info(&dev->dev, "can't enable MSI-X "
 799                        "(MSI IRQ already assigned)\n");
 800                 return -EINVAL;
 801         }
 802         status = msix_capability_init(dev, entries, nvec);
 803         return status;
 804 }
 805 EXPORT_SYMBOL(pci_enable_msix);
 806
 807 void pci_msix_shutdown(struct pci_dev *dev)
 808 {
 809         struct msi_desc *entry;
 810
 811         if (!pci_msi_enable || !dev || !dev->msix_enabled)
 812                 return;
 813
 814         /* Return the device with MSI-X masked as initial states */
 815         list_for_each_entry(entry, &dev->msi_list, list) {
 816                 /* Keep cached states to be restored */
 817                 __msix_mask_irq(entry, 1);
 818         }
 819
 820         msix_set_enable(dev, 0);
 821         pci_intx_for_msi(dev, 1);
 822         dev->msix_enabled = 0;
 823 }
 824
 825 void pci_disable_msix(struct pci_dev *dev)
 826 {
 827         if (!pci_msi_enable || !dev || !dev->msix_enabled)
 828                 return;
 829
 830         pci_msix_shutdown(dev);
 831         free_msi_irqs(dev);
 832 }
 833 EXPORT_SYMBOL(pci_disable_msix);
 834
 835 /**
 836  * msi_remove_pci_irq_vectors - reclaim MSI(X) irqs to unused state
 837  * @dev: pointer to the pci_dev data structure of MSI(X) device function
 838  *
 839  * Being called during hotplug remove, from which the device function
 840  * is hot-removed. All previous assigned MSI/MSI-X irqs, if
 841  * allocated for this device function, are reclaimed to unused state,
 842  * which may be used later on.
 843  **/
 844 void msi_remove_pci_irq_vectors(struct pci_dev *dev)
 845 {
 846         if (!pci_msi_enable || !dev)
 847                 return;
 848
 849         if (dev->msi_enabled || dev->msix_enabled)
 850                 free_msi_irqs(dev);
 851 }
 852
 853 void pci_no_msi(void)
 854 {
 855         pci_msi_enable = 0;
 856 }
 857
 858 /**
 859  * pci_msi_enabled - is MSI enabled?
 860  *
 861  * Returns true if MSI has not been disabled by the command-line option
 862  * pci=nomsi.
 863  **/
 864 int pci_msi_enabled(void)
 865 {
 866         return pci_msi_enable;
 867 }
 868 EXPORT_SYMBOL(pci_msi_enabled);
 869
 870 void pci_msi_init_pci_dev(struct pci_dev *dev)
 871 {
 872         INIT_LIST_HEAD(&dev->msi_list);
 873 }