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/mm.h>
  10 #include <linux/irq.h>
  11 #include <linux/interrupt.h>
  12 #include <linux/init.h>
  13 #include <linux/config.h>
  14 #include <linux/ioport.h>
  15 #include <linux/smp_lock.h>
  16 #include <linux/pci.h>
  17 #include <linux/proc_fs.h>
  18
  19 #include <asm/errno.h>
  20 #include <asm/io.h>
  21 #include <asm/smp.h>
  22
  23 #include "pci.h"
  24 #include "msi.h"
  25
  26 static DEFINE_SPINLOCK(msi_lock);
  27 static struct msi_desc* msi_desc[NR_IRQS] = { [0 ... NR_IRQS-1] = NULL };
  28 static kmem_cache_t* msi_cachep;
  29
  30 static int pci_msi_enable = 1;
  31 static int last_alloc_vector;
  32 static int nr_released_vectors;
  33 static int nr_reserved_vectors = NR_HP_RESERVED_VECTORS;
  34 static int nr_msix_devices;
  35
  36 #ifndef CONFIG_X86_IO_APIC
  37 int vector_irq[NR_VECTORS] = { [0 ... NR_VECTORS - 1] = -1};
  38 u8 irq_vector[NR_IRQ_VECTORS] = { FIRST_DEVICE_VECTOR , 0 };
  39 #endif
  40
  41 static void msi_cache_ctor(void *p, kmem_cache_t *cache, unsigned long flags)
  42 {
  43         memset(p, 0, NR_IRQS * sizeof(struct msi_desc));
  44 }
  45
  46 static int msi_cache_init(void)
  47 {
  48         msi_cachep = kmem_cache_create("msi_cache",
  49                         NR_IRQS * sizeof(struct msi_desc),
  50                         0, SLAB_HWCACHE_ALIGN, msi_cache_ctor, NULL);
  51         if (!msi_cachep)
  52                 return -ENOMEM;
  53
  54         return 0;
  55 }
  56
  57 static void msi_set_mask_bit(unsigned int vector, int flag)
  58 {
  59         struct msi_desc *entry;
  60
  61         entry = (struct msi_desc *)msi_desc[vector];
  62         if (!entry || !entry->dev || !entry->mask_base)
  63                 return;
  64         switch (entry->msi_attrib.type) {
  65         case PCI_CAP_ID_MSI:
  66         {
  67                 int             pos;
  68                 u32             mask_bits;
  69
  70                 pos = (long)entry->mask_base;
  71                 pci_read_config_dword(entry->dev, pos, &mask_bits);
  72                 mask_bits &= ~(1);
  73                 mask_bits |= flag;
  74                 pci_write_config_dword(entry->dev, pos, mask_bits);
  75                 break;
  76         }
  77         case PCI_CAP_ID_MSIX:
  78         {
  79                 int offset = entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE +
  80                         PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET;
  81                 writel(flag, entry->mask_base + offset);
  82                 break;
  83         }
  84         default:
  85                 break;
  86         }
  87 }
  88
  89 #ifdef CONFIG_SMP
  90 static void set_msi_affinity(unsigned int vector, cpumask_t cpu_mask)
  91 {
  92         struct msi_desc *entry;
  93         struct msg_address address;
  94         unsigned int irq = vector;
  95
  96         entry = (struct msi_desc *)msi_desc[vector];
  97         if (!entry || !entry->dev)
  98                 return;
  99
 100         switch (entry->msi_attrib.type) {
 101         case PCI_CAP_ID_MSI:
 102         {
 103                 int pos;
 104
 105                 if (!(pos = pci_find_capability(entry->dev, PCI_CAP_ID_MSI)))
 106                         return;
 107
 108                 pci_read_config_dword(entry->dev, msi_lower_address_reg(pos),
 109                         &address.lo_address.value);
 110                 address.lo_address.value &= MSI_ADDRESS_DEST_ID_MASK;
 111                 address.lo_address.value |= (cpu_mask_to_apicid(cpu_mask) <<
 112                         MSI_TARGET_CPU_SHIFT);
 113                 entry->msi_attrib.current_cpu = cpu_mask_to_apicid(cpu_mask);
 114                 pci_write_config_dword(entry->dev, msi_lower_address_reg(pos),
 115                         address.lo_address.value);
 116                 set_native_irq_info(irq, cpu_mask);
 117                 break;
 118         }
 119         case PCI_CAP_ID_MSIX:
 120         {
 121                 int offset = entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE +
 122                         PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET;
 123
 124                 address.lo_address.value = readl(entry->mask_base + offset);
 125                 address.lo_address.value &= MSI_ADDRESS_DEST_ID_MASK;
 126                 address.lo_address.value |= (cpu_mask_to_apicid(cpu_mask) <<
 127                         MSI_TARGET_CPU_SHIFT);
 128                 entry->msi_attrib.current_cpu = cpu_mask_to_apicid(cpu_mask);
 129                 writel(address.lo_address.value, entry->mask_base + offset);
 130                 set_native_irq_info(irq, cpu_mask);
 131                 break;
 132         }
 133         default:
 134                 break;
 135         }
 136 }
 137 #endif /* CONFIG_SMP */
 138
 139 static void mask_MSI_irq(unsigned int vector)
 140 {
 141         msi_set_mask_bit(vector, 1);
 142 }
 143
 144 static void unmask_MSI_irq(unsigned int vector)
 145 {
 146         msi_set_mask_bit(vector, 0);
 147 }
 148
 149 static unsigned int startup_msi_irq_wo_maskbit(unsigned int vector)
 150 {
 151         struct msi_desc *entry;
 152         unsigned long flags;
 153
 154         spin_lock_irqsave(&msi_lock, flags);
 155         entry = msi_desc[vector];
 156         if (!entry || !entry->dev) {
 157                 spin_unlock_irqrestore(&msi_lock, flags);
 158                 return 0;
 159         }
 160         entry->msi_attrib.state = 1;    /* Mark it active */
 161         spin_unlock_irqrestore(&msi_lock, flags);
 162
 163         return 0;       /* never anything pending */
 164 }
 165
 166 static unsigned int startup_msi_irq_w_maskbit(unsigned int vector)
 167 {
 168         startup_msi_irq_wo_maskbit(vector);
 169         unmask_MSI_irq(vector);
 170         return 0;       /* never anything pending */
 171 }
 172
 173 static void shutdown_msi_irq(unsigned int vector)
 174 {
 175         struct msi_desc *entry;
 176         unsigned long flags;
 177
 178         spin_lock_irqsave(&msi_lock, flags);
 179         entry = msi_desc[vector];
 180         if (entry && entry->dev)
 181                 entry->msi_attrib.state = 0;    /* Mark it not active */
 182         spin_unlock_irqrestore(&msi_lock, flags);
 183 }
 184
 185 static void end_msi_irq_wo_maskbit(unsigned int vector)
 186 {
 187         move_native_irq(vector);
 188         ack_APIC_irq();
 189 }
 190
 191 static void end_msi_irq_w_maskbit(unsigned int vector)
 192 {
 193         move_native_irq(vector);
 194         unmask_MSI_irq(vector);
 195         ack_APIC_irq();
 196 }
 197
 198 static void do_nothing(unsigned int vector)
 199 {
 200 }
 201
 202 /*
 203  * Interrupt Type for MSI-X PCI/PCI-X/PCI-Express Devices,
 204  * which implement the MSI-X Capability Structure.
 205  */
 206 static struct hw_interrupt_type msix_irq_type = {
 207         .typename       = "PCI-MSI-X",
 208         .startup        = startup_msi_irq_w_maskbit,
 209         .shutdown       = shutdown_msi_irq,
 210         .enable         = unmask_MSI_irq,
 211         .disable        = mask_MSI_irq,
 212         .ack            = mask_MSI_irq,
 213         .end            = end_msi_irq_w_maskbit,
 214         .set_affinity   = set_msi_irq_affinity
 215 };
 216
 217 /*
 218  * Interrupt Type for MSI PCI/PCI-X/PCI-Express Devices,
 219  * which implement the MSI Capability Structure with
 220  * Mask-and-Pending Bits.
 221  */
 222 static struct hw_interrupt_type msi_irq_w_maskbit_type = {
 223         .typename       = "PCI-MSI",
 224         .startup        = startup_msi_irq_w_maskbit,
 225         .shutdown       = shutdown_msi_irq,
 226         .enable         = unmask_MSI_irq,
 227         .disable        = mask_MSI_irq,
 228         .ack            = mask_MSI_irq,
 229         .end            = end_msi_irq_w_maskbit,
 230         .set_affinity   = set_msi_irq_affinity
 231 };
 232
 233 /*
 234  * Interrupt Type for MSI PCI/PCI-X/PCI-Express Devices,
 235  * which implement the MSI Capability Structure without
 236  * Mask-and-Pending Bits.
 237  */
 238 static struct hw_interrupt_type msi_irq_wo_maskbit_type = {
 239         .typename       = "PCI-MSI",
 240         .startup        = startup_msi_irq_wo_maskbit,
 241         .shutdown       = shutdown_msi_irq,
 242         .enable         = do_nothing,
 243         .disable        = do_nothing,
 244         .ack            = do_nothing,
 245         .end            = end_msi_irq_wo_maskbit,
 246         .set_affinity   = set_msi_irq_affinity
 247 };
 248
 249 static void msi_data_init(struct msg_data *msi_data,
 250                           unsigned int vector)
 251 {
 252         memset(msi_data, 0, sizeof(struct msg_data));
 253         msi_data->vector = (u8)vector;
 254         msi_data->delivery_mode = MSI_DELIVERY_MODE;
 255         msi_data->level = MSI_LEVEL_MODE;
 256         msi_data->trigger = MSI_TRIGGER_MODE;
 257 }
 258
 259 static void msi_address_init(struct msg_address *msi_address)
 260 {
 261         unsigned int    dest_id;
 262
 263         memset(msi_address, 0, sizeof(struct msg_address));
 264         msi_address->hi_address = (u32)0;
 265         dest_id = (MSI_ADDRESS_HEADER << MSI_ADDRESS_HEADER_SHIFT);
 266         msi_address->lo_address.u.dest_mode = MSI_DEST_MODE;
 267         msi_address->lo_address.u.redirection_hint = MSI_REDIRECTION_HINT_MODE;
 268         msi_address->lo_address.u.dest_id = dest_id;
 269         msi_address->lo_address.value |= (MSI_TARGET_CPU << MSI_TARGET_CPU_SHIFT);
 270 }
 271
 272 static int msi_free_vector(struct pci_dev* dev, int vector, int reassign);
 273 static int assign_msi_vector(void)
 274 {
 275         static int new_vector_avail = 1;
 276         int vector;
 277         unsigned long flags;
 278
 279         /*
 280          * msi_lock is provided to ensure that successful allocation of MSI
 281          * vector is assigned unique among drivers.
 282          */
 283         spin_lock_irqsave(&msi_lock, flags);
 284
 285         if (!new_vector_avail) {
 286                 int free_vector = 0;
 287
 288                 /*
 289                  * vector_irq[] = -1 indicates that this specific vector is:
 290                  * - assigned for MSI (since MSI have no associated IRQ) or
 291                  * - assigned for legacy if less than 16, or
 292                  * - having no corresponding 1:1 vector-to-IOxAPIC IRQ mapping
 293                  * vector_irq[] = 0 indicates that this vector, previously
 294                  * assigned for MSI, is freed by hotplug removed operations.
 295                  * This vector will be reused for any subsequent hotplug added
 296                  * operations.
 297                  * vector_irq[] > 0 indicates that this vector is assigned for
 298                  * IOxAPIC IRQs. This vector and its value provides a 1-to-1
 299                  * vector-to-IOxAPIC IRQ mapping.
 300                  */
 301                 for (vector = FIRST_DEVICE_VECTOR; vector < NR_IRQS; vector++) {
 302                         if (vector_irq[vector] != 0)
 303                                 continue;
 304                         free_vector = vector;
 305                         if (!msi_desc[vector])
 306                                 break;
 307                         else
 308                                 continue;
 309                 }
 310                 if (!free_vector) {
 311                         spin_unlock_irqrestore(&msi_lock, flags);
 312                         return -EBUSY;
 313                 }
 314                 vector_irq[free_vector] = -1;
 315                 nr_released_vectors--;
 316                 spin_unlock_irqrestore(&msi_lock, flags);
 317                 if (msi_desc[free_vector] != NULL) {
 318                         struct pci_dev *dev;
 319                         int tail;
 320
 321                         /* free all linked vectors before re-assign */
 322                         do {
 323                                 spin_lock_irqsave(&msi_lock, flags);
 324                                 dev = msi_desc[free_vector]->dev;
 325                                 tail = msi_desc[free_vector]->link.tail;
 326                                 spin_unlock_irqrestore(&msi_lock, flags);
 327                                 msi_free_vector(dev, tail, 1);
 328                         } while (free_vector != tail);
 329                 }
 330
 331                 return free_vector;
 332         }
 333         vector = assign_irq_vector(AUTO_ASSIGN);
 334         last_alloc_vector = vector;
 335         if (vector  == LAST_DEVICE_VECTOR)
 336                 new_vector_avail = 0;
 337
 338         spin_unlock_irqrestore(&msi_lock, flags);
 339         return vector;
 340 }
 341
 342 static int get_new_vector(void)
 343 {
 344         int vector;
 345
 346         if ((vector = assign_msi_vector()) > 0)
 347                 set_intr_gate(vector, interrupt[vector]);
 348
 349         return vector;
 350 }
 351
 352 static int msi_init(void)
 353 {
 354         static int status = -ENOMEM;
 355
 356         if (!status)
 357                 return status;
 358
 359         if (pci_msi_quirk) {
 360                 pci_msi_enable = 0;
 361                 printk(KERN_WARNING "PCI: MSI quirk detected. MSI disabled.\n");
 362                 status = -EINVAL;
 363                 return status;
 364         }
 365
 366         if ((status = msi_cache_init()) < 0) {
 367                 pci_msi_enable = 0;
 368                 printk(KERN_WARNING "PCI: MSI cache init failed\n");
 369                 return status;
 370         }
 371         last_alloc_vector = assign_irq_vector(AUTO_ASSIGN);
 372         if (last_alloc_vector < 0) {
 373                 pci_msi_enable = 0;
 374                 printk(KERN_WARNING "PCI: No interrupt vectors available for MSI\n");
 375                 status = -EBUSY;
 376                 return status;
 377         }
 378         vector_irq[last_alloc_vector] = 0;
 379         nr_released_vectors++;
 380
 381         return status;
 382 }
 383
 384 static int get_msi_vector(struct pci_dev *dev)
 385 {
 386         return get_new_vector();
 387 }
 388
 389 static struct msi_desc* alloc_msi_entry(void)
 390 {
 391         struct msi_desc *entry;
 392
 393         entry = kmem_cache_alloc(msi_cachep, SLAB_KERNEL);
 394         if (!entry)
 395                 return NULL;
 396
 397         memset(entry, 0, sizeof(struct msi_desc));
 398         entry->link.tail = entry->link.head = 0;        /* single message */
 399         entry->dev = NULL;
 400
 401         return entry;
 402 }
 403
 404 static void attach_msi_entry(struct msi_desc *entry, int vector)
 405 {
 406         unsigned long flags;
 407
 408         spin_lock_irqsave(&msi_lock, flags);
 409         msi_desc[vector] = entry;
 410         spin_unlock_irqrestore(&msi_lock, flags);
 411 }
 412
 413 static void irq_handler_init(int cap_id, int pos, int mask)
 414 {
 415         spin_lock(&irq_desc[pos].lock);
 416         if (cap_id == PCI_CAP_ID_MSIX)
 417                 irq_desc[pos].handler = &msix_irq_type;
 418         else {
 419                 if (!mask)
 420                         irq_desc[pos].handler = &msi_irq_wo_maskbit_type;
 421                 else
 422                         irq_desc[pos].handler = &msi_irq_w_maskbit_type;
 423         }
 424         spin_unlock(&irq_desc[pos].lock);
 425 }
 426
 427 static void enable_msi_mode(struct pci_dev *dev, int pos, int type)
 428 {
 429         u16 control;
 430
 431         pci_read_config_word(dev, msi_control_reg(pos), &control);
 432         if (type == PCI_CAP_ID_MSI) {
 433                 /* Set enabled bits to single MSI & enable MSI_enable bit */
 434                 msi_enable(control, 1);
 435                 pci_write_config_word(dev, msi_control_reg(pos), control);
 436         } else {
 437                 msix_enable(control);
 438                 pci_write_config_word(dev, msi_control_reg(pos), control);
 439         }
 440         if (pci_find_capability(dev, PCI_CAP_ID_EXP)) {
 441                 /* PCI Express Endpoint device detected */
 442                 u16 cmd;
 443                 pci_read_config_word(dev, PCI_COMMAND, &cmd);
 444                 cmd |= PCI_COMMAND_INTX_DISABLE;
 445                 pci_write_config_word(dev, PCI_COMMAND, cmd);
 446         }
 447 }
 448
 449 void disable_msi_mode(struct pci_dev *dev, int pos, int type)
 450 {
 451         u16 control;
 452
 453         pci_read_config_word(dev, msi_control_reg(pos), &control);
 454         if (type == PCI_CAP_ID_MSI) {
 455                 /* Set enabled bits to single MSI & enable MSI_enable bit */
 456                 msi_disable(control);
 457                 pci_write_config_word(dev, msi_control_reg(pos), control);
 458         } else {
 459                 msix_disable(control);
 460                 pci_write_config_word(dev, msi_control_reg(pos), control);
 461         }
 462         if (pci_find_capability(dev, PCI_CAP_ID_EXP)) {
 463                 /* PCI Express Endpoint device detected */
 464                 u16 cmd;
 465                 pci_read_config_word(dev, PCI_COMMAND, &cmd);
 466                 cmd &= ~PCI_COMMAND_INTX_DISABLE;
 467                 pci_write_config_word(dev, PCI_COMMAND, cmd);
 468         }
 469 }
 470
 471 static int msi_lookup_vector(struct pci_dev *dev, int type)
 472 {
 473         int vector;
 474         unsigned long flags;
 475
 476         spin_lock_irqsave(&msi_lock, flags);
 477         for (vector = FIRST_DEVICE_VECTOR; vector < NR_IRQS; vector++) {
 478                 if (!msi_desc[vector] || msi_desc[vector]->dev != dev ||
 479                         msi_desc[vector]->msi_attrib.type != type ||
 480                         msi_desc[vector]->msi_attrib.default_vector != dev->irq)
 481                         continue;
 482                 spin_unlock_irqrestore(&msi_lock, flags);
 483                 /* This pre-assigned MSI vector for this device
 484                    already exits. Override dev->irq with this vector */
 485                 dev->irq = vector;
 486                 return 0;
 487         }
 488         spin_unlock_irqrestore(&msi_lock, flags);
 489
 490         return -EACCES;
 491 }
 492
 493 void pci_scan_msi_device(struct pci_dev *dev)
 494 {
 495         if (!dev)
 496                 return;
 497
 498         if (pci_find_capability(dev, PCI_CAP_ID_MSIX) > 0)
 499                 nr_msix_devices++;
 500         else if (pci_find_capability(dev, PCI_CAP_ID_MSI) > 0)
 501                 nr_reserved_vectors++;
 502 }
 503
 504 /**
 505  * msi_capability_init - configure device's MSI capability structure
 506  * @dev: pointer to the pci_dev data structure of MSI device function
 507  *
 508  * Setup the MSI capability structure of device function with a single
 509  * MSI vector, regardless of device function is capable of handling
 510  * multiple messages. A return of zero indicates the successful setup
 511  * of an entry zero with the new MSI vector or non-zero for otherwise.
 512  **/
 513 static int msi_capability_init(struct pci_dev *dev)
 514 {
 515         struct msi_desc *entry;
 516         struct msg_address address;
 517         struct msg_data data;
 518         int pos, vector;
 519         u16 control;
 520
 521         pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
 522         pci_read_config_word(dev, msi_control_reg(pos), &control);
 523         /* MSI Entry Initialization */
 524         if (!(entry = alloc_msi_entry()))
 525                 return -ENOMEM;
 526
 527         if ((vector = get_msi_vector(dev)) < 0) {
 528                 kmem_cache_free(msi_cachep, entry);
 529                 return -EBUSY;
 530         }
 531         entry->link.head = vector;
 532         entry->link.tail = vector;
 533         entry->msi_attrib.type = PCI_CAP_ID_MSI;
 534         entry->msi_attrib.state = 0;                    /* Mark it not active */
 535         entry->msi_attrib.entry_nr = 0;
 536         entry->msi_attrib.maskbit = is_mask_bit_support(control);
 537         entry->msi_attrib.default_vector = dev->irq;    /* Save IOAPIC IRQ */
 538         dev->irq = vector;
 539         entry->dev = dev;
 540         if (is_mask_bit_support(control)) {
 541                 entry->mask_base = (void __iomem *)(long)msi_mask_bits_reg(pos,
 542                                 is_64bit_address(control));
 543         }
 544         /* Replace with MSI handler */
 545         irq_handler_init(PCI_CAP_ID_MSI, vector, entry->msi_attrib.maskbit);
 546         /* Configure MSI capability structure */
 547         msi_address_init(&address);
 548         msi_data_init(&data, vector);
 549         entry->msi_attrib.current_cpu = ((address.lo_address.u.dest_id >>
 550                                 MSI_TARGET_CPU_SHIFT) & MSI_TARGET_CPU_MASK);
 551         pci_write_config_dword(dev, msi_lower_address_reg(pos),
 552                         address.lo_address.value);
 553         if (is_64bit_address(control)) {
 554                 pci_write_config_dword(dev,
 555                         msi_upper_address_reg(pos), address.hi_address);
 556                 pci_write_config_word(dev,
 557                         msi_data_reg(pos, 1), *((u32*)&data));
 558         } else
 559                 pci_write_config_word(dev,
 560                         msi_data_reg(pos, 0), *((u32*)&data));
 561         if (entry->msi_attrib.maskbit) {
 562                 unsigned int maskbits, temp;
 563                 /* All MSIs are unmasked by default, Mask them all */
 564                 pci_read_config_dword(dev,
 565                         msi_mask_bits_reg(pos, is_64bit_address(control)),
 566                         &maskbits);
 567                 temp = (1 << multi_msi_capable(control));
 568                 temp = ((temp - 1) & ~temp);
 569                 maskbits |= temp;
 570                 pci_write_config_dword(dev,
 571                         msi_mask_bits_reg(pos, is_64bit_address(control)),
 572                         maskbits);
 573         }
 574         attach_msi_entry(entry, vector);
 575         /* Set MSI enabled bits  */
 576         enable_msi_mode(dev, pos, PCI_CAP_ID_MSI);
 577
 578         return 0;
 579 }
 580
 581 /**
 582  * msix_capability_init - configure device's MSI-X capability
 583  * @dev: pointer to the pci_dev data structure of MSI-X device function
 584  *
 585  * Setup the MSI-X capability structure of device function with a
 586  * single MSI-X vector. A return of zero indicates the successful setup of
 587  * requested MSI-X entries with allocated vectors or non-zero for otherwise.
 588  **/
 589 static int msix_capability_init(struct pci_dev *dev,
 590                                 struct msix_entry *entries, int nvec)
 591 {
 592         struct msi_desc *head = NULL, *tail = NULL, *entry = NULL;
 593         struct msg_address address;
 594         struct msg_data data;
 595         int vector, pos, i, j, nr_entries, temp = 0;
 596         u32 phys_addr, table_offset;
 597         u16 control;
 598         u8 bir;
 599         void __iomem *base;
 600
 601         pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
 602         /* Request & Map MSI-X table region */
 603         pci_read_config_word(dev, msi_control_reg(pos), &control);
 604         nr_entries = multi_msix_capable(control);
 605         pci_read_config_dword(dev, msix_table_offset_reg(pos),
 606                 &table_offset);
 607         bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
 608         phys_addr = pci_resource_start (dev, bir);
 609         phys_addr += (u32)(table_offset & ~PCI_MSIX_FLAGS_BIRMASK);
 610         base = ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
 611         if (base == NULL)
 612                 return -ENOMEM;
 613
 614         /* MSI-X Table Initialization */
 615         for (i = 0; i < nvec; i++) {
 616                 entry = alloc_msi_entry();
 617                 if (!entry)
 618                         break;
 619                 if ((vector = get_msi_vector(dev)) < 0)
 620                         break;
 621
 622                 j = entries[i].entry;
 623                 entries[i].vector = vector;
 624                 entry->msi_attrib.type = PCI_CAP_ID_MSIX;
 625                 entry->msi_attrib.state = 0;            /* Mark it not active */
 626                 entry->msi_attrib.entry_nr = j;
 627                 entry->msi_attrib.maskbit = 1;
 628                 entry->msi_attrib.default_vector = dev->irq;
 629                 entry->dev = dev;
 630                 entry->mask_base = base;
 631                 if (!head) {
 632                         entry->link.head = vector;
 633                         entry->link.tail = vector;
 634                         head = entry;
 635                 } else {
 636                         entry->link.head = temp;
 637                         entry->link.tail = tail->link.tail;
 638                         tail->link.tail = vector;
 639                         head->link.head = vector;
 640                 }
 641                 temp = vector;
 642                 tail = entry;
 643                 /* Replace with MSI-X handler */
 644                 irq_handler_init(PCI_CAP_ID_MSIX, vector, 1);
 645                 /* Configure MSI-X capability structure */
 646                 msi_address_init(&address);
 647                 msi_data_init(&data, vector);
 648                 entry->msi_attrib.current_cpu =
 649                         ((address.lo_address.u.dest_id >>
 650                         MSI_TARGET_CPU_SHIFT) & MSI_TARGET_CPU_MASK);
 651                 writel(address.lo_address.value,
 652                         base + j * PCI_MSIX_ENTRY_SIZE +
 653                         PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
 654                 writel(address.hi_address,
 655                         base + j * PCI_MSIX_ENTRY_SIZE +
 656                         PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
 657                 writel(*(u32*)&data,
 658                         base + j * PCI_MSIX_ENTRY_SIZE +
 659                         PCI_MSIX_ENTRY_DATA_OFFSET);
 660                 attach_msi_entry(entry, vector);
 661         }
 662         if (i != nvec) {
 663                 i--;
 664                 for (; i >= 0; i--) {
 665                         vector = (entries + i)->vector;
 666                         msi_free_vector(dev, vector, 0);
 667                         (entries + i)->vector = 0;
 668                 }
 669                 return -EBUSY;
 670         }
 671         /* Set MSI-X enabled bits */
 672         enable_msi_mode(dev, pos, PCI_CAP_ID_MSIX);
 673
 674         return 0;
 675 }
 676
 677 /**
 678  * pci_enable_msi - configure device's MSI capability structure
 679  * @dev: pointer to the pci_dev data structure of MSI device function
 680  *
 681  * Setup the MSI capability structure of device function with
 682  * a single MSI vector upon its software driver call to request for
 683  * MSI mode enabled on its hardware device function. A return of zero
 684  * indicates the successful setup of an entry zero with the new MSI
 685  * vector or non-zero for otherwise.
 686  **/
 687 int pci_enable_msi(struct pci_dev* dev)
 688 {
 689         int pos, temp, status = -EINVAL;
 690         u16 control;
 691
 692         if (!pci_msi_enable || !dev)
 693                 return status;
 694
 695         if (dev->no_msi)
 696                 return status;
 697
 698         temp = dev->irq;
 699
 700         if ((status = msi_init()) < 0)
 701                 return status;
 702
 703         if (!(pos = pci_find_capability(dev, PCI_CAP_ID_MSI)))
 704                 return -EINVAL;
 705
 706         pci_read_config_word(dev, msi_control_reg(pos), &control);
 707         if (control & PCI_MSI_FLAGS_ENABLE)
 708                 return 0;                       /* Already in MSI mode */
 709
 710         if (!msi_lookup_vector(dev, PCI_CAP_ID_MSI)) {
 711                 /* Lookup Sucess */
 712                 unsigned long flags;
 713
 714                 spin_lock_irqsave(&msi_lock, flags);
 715                 if (!vector_irq[dev->irq]) {
 716                         msi_desc[dev->irq]->msi_attrib.state = 0;
 717                         vector_irq[dev->irq] = -1;
 718                         nr_released_vectors--;
 719                         spin_unlock_irqrestore(&msi_lock, flags);
 720                         enable_msi_mode(dev, pos, PCI_CAP_ID_MSI);
 721                         return 0;
 722                 }
 723                 spin_unlock_irqrestore(&msi_lock, flags);
 724                 dev->irq = temp;
 725         }
 726         /* Check whether driver already requested for MSI-X vectors */
 727         if ((pos = pci_find_capability(dev, PCI_CAP_ID_MSIX)) > 0 &&
 728                 !msi_lookup_vector(dev, PCI_CAP_ID_MSIX)) {
 729                         printk(KERN_INFO "PCI: %s: Can't enable MSI.  "
 730                                "Device already has MSI-X vectors assigned\n",
 731                                pci_name(dev));
 732                         dev->irq = temp;
 733                         return -EINVAL;
 734         }
 735         status = msi_capability_init(dev);
 736         if (!status) {
 737                 if (!pos)
 738                         nr_reserved_vectors--;  /* Only MSI capable */
 739                 else if (nr_msix_devices > 0)
 740                         nr_msix_devices--;      /* Both MSI and MSI-X capable,
 741                                                    but choose enabling MSI */
 742         }
 743
 744         return status;
 745 }
 746
 747 void pci_disable_msi(struct pci_dev* dev)
 748 {
 749         struct msi_desc *entry;
 750         int pos, default_vector;
 751         u16 control;
 752         unsigned long flags;
 753
 754         if (!dev || !(pos = pci_find_capability(dev, PCI_CAP_ID_MSI)))
 755                 return;
 756
 757         pci_read_config_word(dev, msi_control_reg(pos), &control);
 758         if (!(control & PCI_MSI_FLAGS_ENABLE))
 759                 return;
 760
 761         spin_lock_irqsave(&msi_lock, flags);
 762         entry = msi_desc[dev->irq];
 763         if (!entry || !entry->dev || entry->msi_attrib.type != PCI_CAP_ID_MSI) {
 764                 spin_unlock_irqrestore(&msi_lock, flags);
 765                 return;
 766         }
 767         if (entry->msi_attrib.state) {
 768                 spin_unlock_irqrestore(&msi_lock, flags);
 769                 printk(KERN_WARNING "PCI: %s: pci_disable_msi() called without "
 770                        "free_irq() on MSI vector %d\n",
 771                        pci_name(dev), dev->irq);
 772                 BUG_ON(entry->msi_attrib.state > 0);
 773         } else {
 774                 vector_irq[dev->irq] = 0; /* free it */
 775                 nr_released_vectors++;
 776                 default_vector = entry->msi_attrib.default_vector;
 777                 spin_unlock_irqrestore(&msi_lock, flags);
 778                 /* Restore dev->irq to its default pin-assertion vector */
 779                 dev->irq = default_vector;
 780                 disable_msi_mode(dev, pci_find_capability(dev, PCI_CAP_ID_MSI),
 781                                         PCI_CAP_ID_MSI);
 782         }
 783 }
 784
 785 static int msi_free_vector(struct pci_dev* dev, int vector, int reassign)
 786 {
 787         struct msi_desc *entry;
 788         int head, entry_nr, type;
 789         void __iomem *base;
 790         unsigned long flags;
 791
 792         spin_lock_irqsave(&msi_lock, flags);
 793         entry = msi_desc[vector];
 794         if (!entry || entry->dev != dev) {
 795                 spin_unlock_irqrestore(&msi_lock, flags);
 796                 return -EINVAL;
 797         }
 798         type = entry->msi_attrib.type;
 799         entry_nr = entry->msi_attrib.entry_nr;
 800         head = entry->link.head;
 801         base = entry->mask_base;
 802         msi_desc[entry->link.head]->link.tail = entry->link.tail;
 803         msi_desc[entry->link.tail]->link.head = entry->link.head;
 804         entry->dev = NULL;
 805         if (!reassign) {
 806                 vector_irq[vector] = 0;
 807                 nr_released_vectors++;
 808         }
 809         msi_desc[vector] = NULL;
 810         spin_unlock_irqrestore(&msi_lock, flags);
 811
 812         kmem_cache_free(msi_cachep, entry);
 813
 814         if (type == PCI_CAP_ID_MSIX) {
 815                 if (!reassign)
 816                         writel(1, base +
 817                                 entry_nr * PCI_MSIX_ENTRY_SIZE +
 818                                 PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET);
 819
 820                 if (head == vector) {
 821                         /*
 822                          * Detect last MSI-X vector to be released.
 823                          * Release the MSI-X memory-mapped table.
 824                          */
 825                         int pos, nr_entries;
 826                         u32 phys_addr, table_offset;
 827                         u16 control;
 828                         u8 bir;
 829
 830                         pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
 831                         pci_read_config_word(dev, msi_control_reg(pos),
 832                                 &control);
 833                         nr_entries = multi_msix_capable(control);
 834                         pci_read_config_dword(dev, msix_table_offset_reg(pos),
 835                                 &table_offset);
 836                         bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
 837                         phys_addr = pci_resource_start (dev, bir);
 838                         phys_addr += (u32)(table_offset &
 839                                 ~PCI_MSIX_FLAGS_BIRMASK);
 840                         iounmap(base);
 841                 }
 842         }
 843
 844         return 0;
 845 }
 846
 847 static int reroute_msix_table(int head, struct msix_entry *entries, int *nvec)
 848 {
 849         int vector = head, tail = 0;
 850         int i, j = 0, nr_entries = 0;
 851         void __iomem *base;
 852         unsigned long flags;
 853
 854         spin_lock_irqsave(&msi_lock, flags);
 855         while (head != tail) {
 856                 nr_entries++;
 857                 tail = msi_desc[vector]->link.tail;
 858                 if (entries[0].entry == msi_desc[vector]->msi_attrib.entry_nr)
 859                         j = vector;
 860                 vector = tail;
 861         }
 862         if (*nvec > nr_entries) {
 863                 spin_unlock_irqrestore(&msi_lock, flags);
 864                 *nvec = nr_entries;
 865                 return -EINVAL;
 866         }
 867         vector = ((j > 0) ? j : head);
 868         for (i = 0; i < *nvec; i++) {
 869                 j = msi_desc[vector]->msi_attrib.entry_nr;
 870                 msi_desc[vector]->msi_attrib.state = 0; /* Mark it not active */
 871                 vector_irq[vector] = -1;                /* Mark it busy */
 872                 nr_released_vectors--;
 873                 entries[i].vector = vector;
 874                 if (j != (entries + i)->entry) {
 875                         base = msi_desc[vector]->mask_base;
 876                         msi_desc[vector]->msi_attrib.entry_nr =
 877                                 (entries + i)->entry;
 878                         writel( readl(base + j * PCI_MSIX_ENTRY_SIZE +
 879                                 PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET), base +
 880                                 (entries + i)->entry * PCI_MSIX_ENTRY_SIZE +
 881                                 PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
 882                         writel( readl(base + j * PCI_MSIX_ENTRY_SIZE +
 883                                 PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET), base +
 884                                 (entries + i)->entry * PCI_MSIX_ENTRY_SIZE +
 885                                 PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
 886                         writel( (readl(base + j * PCI_MSIX_ENTRY_SIZE +
 887                                 PCI_MSIX_ENTRY_DATA_OFFSET) & 0xff00) | vector,
 888                                 base + (entries+i)->entry*PCI_MSIX_ENTRY_SIZE +
 889                                 PCI_MSIX_ENTRY_DATA_OFFSET);
 890                 }
 891                 vector = msi_desc[vector]->link.tail;
 892         }
 893         spin_unlock_irqrestore(&msi_lock, flags);
 894
 895         return 0;
 896 }
 897
 898 /**
 899  * pci_enable_msix - configure device's MSI-X capability structure
 900  * @dev: pointer to the pci_dev data structure of MSI-X device function
 901  * @entries: pointer to an array of MSI-X entries
 902  * @nvec: number of MSI-X vectors requested for allocation by device driver
 903  *
 904  * Setup the MSI-X capability structure of device function with the number
 905  * of requested vectors upon its software driver call to request for
 906  * MSI-X mode enabled on its hardware device function. A return of zero
 907  * indicates the successful configuration of MSI-X capability structure
 908  * with new allocated MSI-X vectors. A return of < 0 indicates a failure.
 909  * Or a return of > 0 indicates that driver request is exceeding the number
 910  * of vectors available. Driver should use the returned value to re-send
 911  * its request.
 912  **/
 913 int pci_enable_msix(struct pci_dev* dev, struct msix_entry *entries, int nvec)
 914 {
 915         int status, pos, nr_entries, free_vectors;
 916         int i, j, temp;
 917         u16 control;
 918         unsigned long flags;
 919
 920         if (!pci_msi_enable || !dev || !entries)
 921                 return -EINVAL;
 922
 923         if ((status = msi_init()) < 0)
 924                 return status;
 925
 926         if (!(pos = pci_find_capability(dev, PCI_CAP_ID_MSIX)))
 927                 return -EINVAL;
 928
 929         pci_read_config_word(dev, msi_control_reg(pos), &control);
 930         if (control & PCI_MSIX_FLAGS_ENABLE)
 931                 return -EINVAL;                 /* Already in MSI-X mode */
 932
 933         nr_entries = multi_msix_capable(control);
 934         if (nvec > nr_entries)
 935                 return -EINVAL;
 936
 937         /* Check for any invalid entries */
 938         for (i = 0; i < nvec; i++) {
 939                 if (entries[i].entry >= nr_entries)
 940                         return -EINVAL;         /* invalid entry */
 941                 for (j = i + 1; j < nvec; j++) {
 942                         if (entries[i].entry == entries[j].entry)
 943                                 return -EINVAL; /* duplicate entry */
 944                 }
 945         }
 946         temp = dev->irq;
 947         if (!msi_lookup_vector(dev, PCI_CAP_ID_MSIX)) {
 948                 /* Lookup Sucess */
 949                 nr_entries = nvec;
 950                 /* Reroute MSI-X table */
 951                 if (reroute_msix_table(dev->irq, entries, &nr_entries)) {
 952                         /* #requested > #previous-assigned */
 953                         dev->irq = temp;
 954                         return nr_entries;
 955                 }
 956                 dev->irq = temp;
 957                 enable_msi_mode(dev, pos, PCI_CAP_ID_MSIX);
 958                 return 0;
 959         }
 960         /* Check whether driver already requested for MSI vector */
 961         if (pci_find_capability(dev, PCI_CAP_ID_MSI) > 0 &&
 962                 !msi_lookup_vector(dev, PCI_CAP_ID_MSI)) {
 963                 printk(KERN_INFO "PCI: %s: Can't enable MSI-X.  "
 964                        "Device already has an MSI vector assigned\n",
 965                        pci_name(dev));
 966                 dev->irq = temp;
 967                 return -EINVAL;
 968         }
 969
 970         spin_lock_irqsave(&msi_lock, flags);
 971         /*
 972          * msi_lock is provided to ensure that enough vectors resources are
 973          * available before granting.
 974          */
 975         free_vectors = pci_vector_resources(last_alloc_vector,
 976                                 nr_released_vectors);
 977         /* Ensure that each MSI/MSI-X device has one vector reserved by
 978            default to avoid any MSI-X driver to take all available
 979            resources */
 980         free_vectors -= nr_reserved_vectors;
 981         /* Find the average of free vectors among MSI-X devices */
 982         if (nr_msix_devices > 0)
 983                 free_vectors /= nr_msix_devices;
 984         spin_unlock_irqrestore(&msi_lock, flags);
 985
 986         if (nvec > free_vectors) {
 987                 if (free_vectors > 0)
 988                         return free_vectors;
 989                 else
 990                         return -EBUSY;
 991         }
 992
 993         status = msix_capability_init(dev, entries, nvec);
 994         if (!status && nr_msix_devices > 0)
 995                 nr_msix_devices--;
 996
 997         return status;
 998 }
 999
1000 void pci_disable_msix(struct pci_dev* dev)
1001 {
1002         int pos, temp;
1003         u16 control;
1004
1005         if (!dev || !(pos = pci_find_capability(dev, PCI_CAP_ID_MSIX)))
1006                 return;
1007
1008         pci_read_config_word(dev, msi_control_reg(pos), &control);
1009         if (!(control & PCI_MSIX_FLAGS_ENABLE))
1010                 return;
1011
1012         temp = dev->irq;
1013         if (!msi_lookup_vector(dev, PCI_CAP_ID_MSIX)) {
1014                 int state, vector, head, tail = 0, warning = 0;
1015                 unsigned long flags;
1016
1017                 vector = head = dev->irq;
1018                 spin_lock_irqsave(&msi_lock, flags);
1019                 while (head != tail) {
1020                         state = msi_desc[vector]->msi_attrib.state;
1021                         if (state)
1022                                 warning = 1;
1023                         else {
1024                                 vector_irq[vector] = 0; /* free it */
1025                                 nr_released_vectors++;
1026                         }
1027                         tail = msi_desc[vector]->link.tail;
1028                         vector = tail;
1029                 }
1030                 spin_unlock_irqrestore(&msi_lock, flags);
1031                 if (warning) {
1032                         dev->irq = temp;
1033                         printk(KERN_WARNING "PCI: %s: pci_disable_msix() called without "
1034                                "free_irq() on all MSI-X vectors\n",
1035                                pci_name(dev));
1036                         BUG_ON(warning > 0);
1037                 } else {
1038                         dev->irq = temp;
1039                         disable_msi_mode(dev,
1040                                 pci_find_capability(dev, PCI_CAP_ID_MSIX),
1041                                 PCI_CAP_ID_MSIX);
1042
1043                 }
1044         }
1045 }
1046
1047 /**
1048  * msi_remove_pci_irq_vectors - reclaim MSI(X) vectors to unused state
1049  * @dev: pointer to the pci_dev data structure of MSI(X) device function
1050  *
1051  * Being called during hotplug remove, from which the device function
1052  * is hot-removed. All previous assigned MSI/MSI-X vectors, if
1053  * allocated for this device function, are reclaimed to unused state,
1054  * which may be used later on.
1055  **/
1056 void msi_remove_pci_irq_vectors(struct pci_dev* dev)
1057 {
1058         int state, pos, temp;
1059         unsigned long flags;
1060
1061         if (!pci_msi_enable || !dev)
1062                 return;
1063
1064         temp = dev->irq;                /* Save IOAPIC IRQ */
1065         if ((pos = pci_find_capability(dev, PCI_CAP_ID_MSI)) > 0 &&
1066                 !msi_lookup_vector(dev, PCI_CAP_ID_MSI)) {
1067                 spin_lock_irqsave(&msi_lock, flags);
1068                 state = msi_desc[dev->irq]->msi_attrib.state;
1069                 spin_unlock_irqrestore(&msi_lock, flags);
1070                 if (state) {
1071                         printk(KERN_WARNING "PCI: %s: msi_remove_pci_irq_vectors() "
1072                                "called without free_irq() on MSI vector %d\n",
1073                                pci_name(dev), dev->irq);
1074                         BUG_ON(state > 0);
1075                 } else /* Release MSI vector assigned to this device */
1076                         msi_free_vector(dev, dev->irq, 0);
1077                 dev->irq = temp;                /* Restore IOAPIC IRQ */
1078         }
1079         if ((pos = pci_find_capability(dev, PCI_CAP_ID_MSIX)) > 0 &&
1080                 !msi_lookup_vector(dev, PCI_CAP_ID_MSIX)) {
1081                 int vector, head, tail = 0, warning = 0;
1082                 void __iomem *base = NULL;
1083
1084                 vector = head = dev->irq;
1085                 while (head != tail) {
1086                         spin_lock_irqsave(&msi_lock, flags);
1087                         state = msi_desc[vector]->msi_attrib.state;
1088                         tail = msi_desc[vector]->link.tail;
1089                         base = msi_desc[vector]->mask_base;
1090                         spin_unlock_irqrestore(&msi_lock, flags);
1091                         if (state)
1092                                 warning = 1;
1093                         else if (vector != head) /* Release MSI-X vector */
1094                                 msi_free_vector(dev, vector, 0);
1095                         vector = tail;
1096                 }
1097                 msi_free_vector(dev, vector, 0);
1098                 if (warning) {
1099                         /* Force to release the MSI-X memory-mapped table */
1100                         u32 phys_addr, table_offset;
1101                         u16 control;
1102                         u8 bir;
1103
1104                         pci_read_config_word(dev, msi_control_reg(pos),
1105                                 &control);
1106                         pci_read_config_dword(dev, msix_table_offset_reg(pos),
1107                                 &table_offset);
1108                         bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
1109                         phys_addr = pci_resource_start (dev, bir);
1110                         phys_addr += (u32)(table_offset &
1111                                 ~PCI_MSIX_FLAGS_BIRMASK);
1112                         iounmap(base);
1113                         printk(KERN_WARNING "PCI: %s: msi_remove_pci_irq_vectors() "
1114                                "called without free_irq() on all MSI-X vectors\n",
1115                                pci_name(dev));
1116                         BUG_ON(warning > 0);
1117                 }
1118                 dev->irq = temp;                /* Restore IOAPIC IRQ */
1119         }
1120 }
1121
1122 EXPORT_SYMBOL(pci_enable_msi);
1123 EXPORT_SYMBOL(pci_disable_msi);
1124 EXPORT_SYMBOL(pci_enable_msix);
1125 EXPORT_SYMBOL(pci_disable_msix);