msi.c revision eaae4b3a84a3781543a32bcaf0a33306ae915574
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 26static DEFINE_SPINLOCK(msi_lock); 27static struct msi_desc* msi_desc[NR_IRQS] = { [0 ... NR_IRQS-1] = NULL }; 28static kmem_cache_t* msi_cachep; 29 30static int pci_msi_enable = 1; 31static int last_alloc_vector = 0; 32static int nr_released_vectors = 0; 33static int nr_reserved_vectors = NR_HP_RESERVED_VECTORS; 34static int nr_msix_devices = 0; 35 36#ifndef CONFIG_X86_IO_APIC 37int vector_irq[NR_VECTORS] = { [0 ... NR_VECTORS - 1] = -1}; 38u8 irq_vector[NR_IRQ_VECTORS] = { FIRST_DEVICE_VECTOR , 0 }; 39#endif 40 41static 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 46static 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 57static 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 90static void set_msi_affinity(unsigned int vector, cpumask_t cpu_mask) 91{ 92 struct msi_desc *entry; 93 struct msg_address address; 94 95 entry = (struct msi_desc *)msi_desc[vector]; 96 if (!entry || !entry->dev) 97 return; 98 99 switch (entry->msi_attrib.type) { 100 case PCI_CAP_ID_MSI: 101 { 102 int pos; 103 104 if (!(pos = pci_find_capability(entry->dev, PCI_CAP_ID_MSI))) 105 return; 106 107 pci_read_config_dword(entry->dev, msi_lower_address_reg(pos), 108 &address.lo_address.value); 109 address.lo_address.value &= MSI_ADDRESS_DEST_ID_MASK; 110 address.lo_address.value |= (cpu_mask_to_apicid(cpu_mask) << 111 MSI_TARGET_CPU_SHIFT); 112 entry->msi_attrib.current_cpu = cpu_mask_to_apicid(cpu_mask); 113 pci_write_config_dword(entry->dev, msi_lower_address_reg(pos), 114 address.lo_address.value); 115 break; 116 } 117 case PCI_CAP_ID_MSIX: 118 { 119 int offset = entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE + 120 PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET; 121 122 address.lo_address.value = readl(entry->mask_base + offset); 123 address.lo_address.value &= MSI_ADDRESS_DEST_ID_MASK; 124 address.lo_address.value |= (cpu_mask_to_apicid(cpu_mask) << 125 MSI_TARGET_CPU_SHIFT); 126 entry->msi_attrib.current_cpu = cpu_mask_to_apicid(cpu_mask); 127 writel(address.lo_address.value, entry->mask_base + offset); 128 break; 129 } 130 default: 131 break; 132 } 133} 134 135#ifdef CONFIG_IRQBALANCE 136static inline void move_msi(int vector) 137{ 138 if (!cpus_empty(pending_irq_balance_cpumask[vector])) { 139 set_msi_affinity(vector, pending_irq_balance_cpumask[vector]); 140 cpus_clear(pending_irq_balance_cpumask[vector]); 141 } 142} 143#endif /* CONFIG_IRQBALANCE */ 144#endif /* CONFIG_SMP */ 145 146static void mask_MSI_irq(unsigned int vector) 147{ 148 msi_set_mask_bit(vector, 1); 149} 150 151static void unmask_MSI_irq(unsigned int vector) 152{ 153 msi_set_mask_bit(vector, 0); 154} 155 156static unsigned int startup_msi_irq_wo_maskbit(unsigned int vector) 157{ 158 struct msi_desc *entry; 159 unsigned long flags; 160 161 spin_lock_irqsave(&msi_lock, flags); 162 entry = msi_desc[vector]; 163 if (!entry || !entry->dev) { 164 spin_unlock_irqrestore(&msi_lock, flags); 165 return 0; 166 } 167 entry->msi_attrib.state = 1; /* Mark it active */ 168 spin_unlock_irqrestore(&msi_lock, flags); 169 170 return 0; /* never anything pending */ 171} 172 173static void release_msi(unsigned int vector); 174static void shutdown_msi_irq(unsigned int vector) 175{ 176 release_msi(vector); 177} 178 179#define shutdown_msi_irq_wo_maskbit shutdown_msi_irq 180static void enable_msi_irq_wo_maskbit(unsigned int vector) {} 181static void disable_msi_irq_wo_maskbit(unsigned int vector) {} 182static void ack_msi_irq_wo_maskbit(unsigned int vector) {} 183static void end_msi_irq_wo_maskbit(unsigned int vector) 184{ 185 move_msi(vector); 186 ack_APIC_irq(); 187} 188 189static unsigned int startup_msi_irq_w_maskbit(unsigned int vector) 190{ 191 struct msi_desc *entry; 192 unsigned long flags; 193 194 spin_lock_irqsave(&msi_lock, flags); 195 entry = msi_desc[vector]; 196 if (!entry || !entry->dev) { 197 spin_unlock_irqrestore(&msi_lock, flags); 198 return 0; 199 } 200 entry->msi_attrib.state = 1; /* Mark it active */ 201 spin_unlock_irqrestore(&msi_lock, flags); 202 203 unmask_MSI_irq(vector); 204 return 0; /* never anything pending */ 205} 206 207#define shutdown_msi_irq_w_maskbit shutdown_msi_irq 208#define enable_msi_irq_w_maskbit unmask_MSI_irq 209#define disable_msi_irq_w_maskbit mask_MSI_irq 210#define ack_msi_irq_w_maskbit mask_MSI_irq 211 212static void end_msi_irq_w_maskbit(unsigned int vector) 213{ 214 move_msi(vector); 215 unmask_MSI_irq(vector); 216 ack_APIC_irq(); 217} 218 219/* 220 * Interrupt Type for MSI-X PCI/PCI-X/PCI-Express Devices, 221 * which implement the MSI-X Capability Structure. 222 */ 223static struct hw_interrupt_type msix_irq_type = { 224 .typename = "PCI-MSI-X", 225 .startup = startup_msi_irq_w_maskbit, 226 .shutdown = shutdown_msi_irq_w_maskbit, 227 .enable = enable_msi_irq_w_maskbit, 228 .disable = disable_msi_irq_w_maskbit, 229 .ack = ack_msi_irq_w_maskbit, 230 .end = end_msi_irq_w_maskbit, 231 .set_affinity = set_msi_irq_affinity 232}; 233 234/* 235 * Interrupt Type for MSI PCI/PCI-X/PCI-Express Devices, 236 * which implement the MSI Capability Structure with 237 * Mask-and-Pending Bits. 238 */ 239static struct hw_interrupt_type msi_irq_w_maskbit_type = { 240 .typename = "PCI-MSI", 241 .startup = startup_msi_irq_w_maskbit, 242 .shutdown = shutdown_msi_irq_w_maskbit, 243 .enable = enable_msi_irq_w_maskbit, 244 .disable = disable_msi_irq_w_maskbit, 245 .ack = ack_msi_irq_w_maskbit, 246 .end = end_msi_irq_w_maskbit, 247 .set_affinity = set_msi_irq_affinity 248}; 249 250/* 251 * Interrupt Type for MSI PCI/PCI-X/PCI-Express Devices, 252 * which implement the MSI Capability Structure without 253 * Mask-and-Pending Bits. 254 */ 255static struct hw_interrupt_type msi_irq_wo_maskbit_type = { 256 .typename = "PCI-MSI", 257 .startup = startup_msi_irq_wo_maskbit, 258 .shutdown = shutdown_msi_irq_wo_maskbit, 259 .enable = enable_msi_irq_wo_maskbit, 260 .disable = disable_msi_irq_wo_maskbit, 261 .ack = ack_msi_irq_wo_maskbit, 262 .end = end_msi_irq_wo_maskbit, 263 .set_affinity = set_msi_irq_affinity 264}; 265 266static void msi_data_init(struct msg_data *msi_data, 267 unsigned int vector) 268{ 269 memset(msi_data, 0, sizeof(struct msg_data)); 270 msi_data->vector = (u8)vector; 271 msi_data->delivery_mode = MSI_DELIVERY_MODE; 272 msi_data->level = MSI_LEVEL_MODE; 273 msi_data->trigger = MSI_TRIGGER_MODE; 274} 275 276static void msi_address_init(struct msg_address *msi_address) 277{ 278 unsigned int dest_id; 279 280 memset(msi_address, 0, sizeof(struct msg_address)); 281 msi_address->hi_address = (u32)0; 282 dest_id = (MSI_ADDRESS_HEADER << MSI_ADDRESS_HEADER_SHIFT); 283 msi_address->lo_address.u.dest_mode = MSI_DEST_MODE; 284 msi_address->lo_address.u.redirection_hint = MSI_REDIRECTION_HINT_MODE; 285 msi_address->lo_address.u.dest_id = dest_id; 286 msi_address->lo_address.value |= (MSI_TARGET_CPU << MSI_TARGET_CPU_SHIFT); 287} 288 289static int msi_free_vector(struct pci_dev* dev, int vector, int reassign); 290static int assign_msi_vector(void) 291{ 292 static int new_vector_avail = 1; 293 int vector; 294 unsigned long flags; 295 296 /* 297 * msi_lock is provided to ensure that successful allocation of MSI 298 * vector is assigned unique among drivers. 299 */ 300 spin_lock_irqsave(&msi_lock, flags); 301 302 if (!new_vector_avail) { 303 int free_vector = 0; 304 305 /* 306 * vector_irq[] = -1 indicates that this specific vector is: 307 * - assigned for MSI (since MSI have no associated IRQ) or 308 * - assigned for legacy if less than 16, or 309 * - having no corresponding 1:1 vector-to-IOxAPIC IRQ mapping 310 * vector_irq[] = 0 indicates that this vector, previously 311 * assigned for MSI, is freed by hotplug removed operations. 312 * This vector will be reused for any subsequent hotplug added 313 * operations. 314 * vector_irq[] > 0 indicates that this vector is assigned for 315 * IOxAPIC IRQs. This vector and its value provides a 1-to-1 316 * vector-to-IOxAPIC IRQ mapping. 317 */ 318 for (vector = FIRST_DEVICE_VECTOR; vector < NR_IRQS; vector++) { 319 if (vector_irq[vector] != 0) 320 continue; 321 free_vector = vector; 322 if (!msi_desc[vector]) 323 break; 324 else 325 continue; 326 } 327 if (!free_vector) { 328 spin_unlock_irqrestore(&msi_lock, flags); 329 return -EBUSY; 330 } 331 vector_irq[free_vector] = -1; 332 nr_released_vectors--; 333 spin_unlock_irqrestore(&msi_lock, flags); 334 if (msi_desc[free_vector] != NULL) { 335 struct pci_dev *dev; 336 int tail; 337 338 /* free all linked vectors before re-assign */ 339 do { 340 spin_lock_irqsave(&msi_lock, flags); 341 dev = msi_desc[free_vector]->dev; 342 tail = msi_desc[free_vector]->link.tail; 343 spin_unlock_irqrestore(&msi_lock, flags); 344 msi_free_vector(dev, tail, 1); 345 } while (free_vector != tail); 346 } 347 348 return free_vector; 349 } 350 vector = assign_irq_vector(AUTO_ASSIGN); 351 last_alloc_vector = vector; 352 if (vector == LAST_DEVICE_VECTOR) 353 new_vector_avail = 0; 354 355 spin_unlock_irqrestore(&msi_lock, flags); 356 return vector; 357} 358 359static int get_new_vector(void) 360{ 361 int vector; 362 363 if ((vector = assign_msi_vector()) > 0) 364 set_intr_gate(vector, interrupt[vector]); 365 366 return vector; 367} 368 369static int msi_init(void) 370{ 371 static int status = -ENOMEM; 372 373 if (!status) 374 return status; 375 376 if (pci_msi_quirk) { 377 pci_msi_enable = 0; 378 printk(KERN_WARNING "PCI: MSI quirk detected. MSI disabled.\n"); 379 status = -EINVAL; 380 return status; 381 } 382 383 if ((status = msi_cache_init()) < 0) { 384 pci_msi_enable = 0; 385 printk(KERN_WARNING "PCI: MSI cache init failed\n"); 386 return status; 387 } 388 last_alloc_vector = assign_irq_vector(AUTO_ASSIGN); 389 if (last_alloc_vector < 0) { 390 pci_msi_enable = 0; 391 printk(KERN_WARNING "PCI: No interrupt vectors available for MSI\n"); 392 status = -EBUSY; 393 return status; 394 } 395 vector_irq[last_alloc_vector] = 0; 396 nr_released_vectors++; 397 398 return status; 399} 400 401static int get_msi_vector(struct pci_dev *dev) 402{ 403 return get_new_vector(); 404} 405 406static struct msi_desc* alloc_msi_entry(void) 407{ 408 struct msi_desc *entry; 409 410 entry = (struct msi_desc*) kmem_cache_alloc(msi_cachep, SLAB_KERNEL); 411 if (!entry) 412 return NULL; 413 414 memset(entry, 0, sizeof(struct msi_desc)); 415 entry->link.tail = entry->link.head = 0; /* single message */ 416 entry->dev = NULL; 417 418 return entry; 419} 420 421static void attach_msi_entry(struct msi_desc *entry, int vector) 422{ 423 unsigned long flags; 424 425 spin_lock_irqsave(&msi_lock, flags); 426 msi_desc[vector] = entry; 427 spin_unlock_irqrestore(&msi_lock, flags); 428} 429 430static void irq_handler_init(int cap_id, int pos, int mask) 431{ 432 spin_lock(&irq_desc[pos].lock); 433 if (cap_id == PCI_CAP_ID_MSIX) 434 irq_desc[pos].handler = &msix_irq_type; 435 else { 436 if (!mask) 437 irq_desc[pos].handler = &msi_irq_wo_maskbit_type; 438 else 439 irq_desc[pos].handler = &msi_irq_w_maskbit_type; 440 } 441 spin_unlock(&irq_desc[pos].lock); 442} 443 444static void enable_msi_mode(struct pci_dev *dev, int pos, int type) 445{ 446 u16 control; 447 448 pci_read_config_word(dev, msi_control_reg(pos), &control); 449 if (type == PCI_CAP_ID_MSI) { 450 /* Set enabled bits to single MSI & enable MSI_enable bit */ 451 msi_enable(control, 1); 452 pci_write_config_word(dev, msi_control_reg(pos), control); 453 } else { 454 msix_enable(control); 455 pci_write_config_word(dev, msi_control_reg(pos), control); 456 } 457 if (pci_find_capability(dev, PCI_CAP_ID_EXP)) { 458 /* PCI Express Endpoint device detected */ 459 u16 cmd; 460 pci_read_config_word(dev, PCI_COMMAND, &cmd); 461 cmd |= PCI_COMMAND_INTX_DISABLE; 462 pci_write_config_word(dev, PCI_COMMAND, cmd); 463 } 464} 465 466static void disable_msi_mode(struct pci_dev *dev, int pos, int type) 467{ 468 u16 control; 469 470 pci_read_config_word(dev, msi_control_reg(pos), &control); 471 if (type == PCI_CAP_ID_MSI) { 472 /* Set enabled bits to single MSI & enable MSI_enable bit */ 473 msi_disable(control); 474 pci_write_config_word(dev, msi_control_reg(pos), control); 475 } else { 476 msix_disable(control); 477 pci_write_config_word(dev, msi_control_reg(pos), control); 478 } 479 if (pci_find_capability(dev, PCI_CAP_ID_EXP)) { 480 /* PCI Express Endpoint device detected */ 481 u16 cmd; 482 pci_read_config_word(dev, PCI_COMMAND, &cmd); 483 cmd &= ~PCI_COMMAND_INTX_DISABLE; 484 pci_write_config_word(dev, PCI_COMMAND, cmd); 485 } 486} 487 488static int msi_lookup_vector(struct pci_dev *dev, int type) 489{ 490 int vector; 491 unsigned long flags; 492 493 spin_lock_irqsave(&msi_lock, flags); 494 for (vector = FIRST_DEVICE_VECTOR; vector < NR_IRQS; vector++) { 495 if (!msi_desc[vector] || msi_desc[vector]->dev != dev || 496 msi_desc[vector]->msi_attrib.type != type || 497 msi_desc[vector]->msi_attrib.default_vector != dev->irq) 498 continue; 499 spin_unlock_irqrestore(&msi_lock, flags); 500 /* This pre-assigned MSI vector for this device 501 already exits. Override dev->irq with this vector */ 502 dev->irq = vector; 503 return 0; 504 } 505 spin_unlock_irqrestore(&msi_lock, flags); 506 507 return -EACCES; 508} 509 510void pci_scan_msi_device(struct pci_dev *dev) 511{ 512 if (!dev) 513 return; 514 515 if (pci_find_capability(dev, PCI_CAP_ID_MSIX) > 0) 516 nr_msix_devices++; 517 else if (pci_find_capability(dev, PCI_CAP_ID_MSI) > 0) 518 nr_reserved_vectors++; 519} 520 521/** 522 * msi_capability_init - configure device's MSI capability structure 523 * @dev: pointer to the pci_dev data structure of MSI device function 524 * 525 * Setup the MSI capability structure of device function with a single 526 * MSI vector, regardless of device function is capable of handling 527 * multiple messages. A return of zero indicates the successful setup 528 * of an entry zero with the new MSI vector or non-zero for otherwise. 529 **/ 530static int msi_capability_init(struct pci_dev *dev) 531{ 532 struct msi_desc *entry; 533 struct msg_address address; 534 struct msg_data data; 535 int pos, vector; 536 u16 control; 537 538 pos = pci_find_capability(dev, PCI_CAP_ID_MSI); 539 pci_read_config_word(dev, msi_control_reg(pos), &control); 540 /* MSI Entry Initialization */ 541 if (!(entry = alloc_msi_entry())) 542 return -ENOMEM; 543 544 if ((vector = get_msi_vector(dev)) < 0) { 545 kmem_cache_free(msi_cachep, entry); 546 return -EBUSY; 547 } 548 entry->link.head = vector; 549 entry->link.tail = vector; 550 entry->msi_attrib.type = PCI_CAP_ID_MSI; 551 entry->msi_attrib.state = 0; /* Mark it not active */ 552 entry->msi_attrib.entry_nr = 0; 553 entry->msi_attrib.maskbit = is_mask_bit_support(control); 554 entry->msi_attrib.default_vector = dev->irq; /* Save IOAPIC IRQ */ 555 dev->irq = vector; 556 entry->dev = dev; 557 if (is_mask_bit_support(control)) { 558 entry->mask_base = (void __iomem *)(long)msi_mask_bits_reg(pos, 559 is_64bit_address(control)); 560 } 561 /* Replace with MSI handler */ 562 irq_handler_init(PCI_CAP_ID_MSI, vector, entry->msi_attrib.maskbit); 563 /* Configure MSI capability structure */ 564 msi_address_init(&address); 565 msi_data_init(&data, vector); 566 entry->msi_attrib.current_cpu = ((address.lo_address.u.dest_id >> 567 MSI_TARGET_CPU_SHIFT) & MSI_TARGET_CPU_MASK); 568 pci_write_config_dword(dev, msi_lower_address_reg(pos), 569 address.lo_address.value); 570 if (is_64bit_address(control)) { 571 pci_write_config_dword(dev, 572 msi_upper_address_reg(pos), address.hi_address); 573 pci_write_config_word(dev, 574 msi_data_reg(pos, 1), *((u32*)&data)); 575 } else 576 pci_write_config_word(dev, 577 msi_data_reg(pos, 0), *((u32*)&data)); 578 if (entry->msi_attrib.maskbit) { 579 unsigned int maskbits, temp; 580 /* All MSIs are unmasked by default, Mask them all */ 581 pci_read_config_dword(dev, 582 msi_mask_bits_reg(pos, is_64bit_address(control)), 583 &maskbits); 584 temp = (1 << multi_msi_capable(control)); 585 temp = ((temp - 1) & ~temp); 586 maskbits |= temp; 587 pci_write_config_dword(dev, 588 msi_mask_bits_reg(pos, is_64bit_address(control)), 589 maskbits); 590 } 591 attach_msi_entry(entry, vector); 592 /* Set MSI enabled bits */ 593 enable_msi_mode(dev, pos, PCI_CAP_ID_MSI); 594 595 return 0; 596} 597 598/** 599 * msix_capability_init - configure device's MSI-X capability 600 * @dev: pointer to the pci_dev data structure of MSI-X device function 601 * 602 * Setup the MSI-X capability structure of device function with a 603 * single MSI-X vector. A return of zero indicates the successful setup of 604 * requested MSI-X entries with allocated vectors or non-zero for otherwise. 605 **/ 606static int msix_capability_init(struct pci_dev *dev, 607 struct msix_entry *entries, int nvec) 608{ 609 struct msi_desc *head = NULL, *tail = NULL, *entry = NULL; 610 struct msg_address address; 611 struct msg_data data; 612 int vector, pos, i, j, nr_entries, temp = 0; 613 u32 phys_addr, table_offset; 614 u16 control; 615 u8 bir; 616 void __iomem *base; 617 618 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX); 619 /* Request & Map MSI-X table region */ 620 pci_read_config_word(dev, msi_control_reg(pos), &control); 621 nr_entries = multi_msix_capable(control); 622 pci_read_config_dword(dev, msix_table_offset_reg(pos), 623 &table_offset); 624 bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK); 625 phys_addr = pci_resource_start (dev, bir); 626 phys_addr += (u32)(table_offset & ~PCI_MSIX_FLAGS_BIRMASK); 627 base = ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE); 628 if (base == NULL) 629 return -ENOMEM; 630 631 /* MSI-X Table Initialization */ 632 for (i = 0; i < nvec; i++) { 633 entry = alloc_msi_entry(); 634 if (!entry) 635 break; 636 if ((vector = get_msi_vector(dev)) < 0) 637 break; 638 639 j = entries[i].entry; 640 entries[i].vector = vector; 641 entry->msi_attrib.type = PCI_CAP_ID_MSIX; 642 entry->msi_attrib.state = 0; /* Mark it not active */ 643 entry->msi_attrib.entry_nr = j; 644 entry->msi_attrib.maskbit = 1; 645 entry->msi_attrib.default_vector = dev->irq; 646 entry->dev = dev; 647 entry->mask_base = base; 648 if (!head) { 649 entry->link.head = vector; 650 entry->link.tail = vector; 651 head = entry; 652 } else { 653 entry->link.head = temp; 654 entry->link.tail = tail->link.tail; 655 tail->link.tail = vector; 656 head->link.head = vector; 657 } 658 temp = vector; 659 tail = entry; 660 /* Replace with MSI-X handler */ 661 irq_handler_init(PCI_CAP_ID_MSIX, vector, 1); 662 /* Configure MSI-X capability structure */ 663 msi_address_init(&address); 664 msi_data_init(&data, vector); 665 entry->msi_attrib.current_cpu = 666 ((address.lo_address.u.dest_id >> 667 MSI_TARGET_CPU_SHIFT) & MSI_TARGET_CPU_MASK); 668 writel(address.lo_address.value, 669 base + j * PCI_MSIX_ENTRY_SIZE + 670 PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET); 671 writel(address.hi_address, 672 base + j * PCI_MSIX_ENTRY_SIZE + 673 PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET); 674 writel(*(u32*)&data, 675 base + j * PCI_MSIX_ENTRY_SIZE + 676 PCI_MSIX_ENTRY_DATA_OFFSET); 677 attach_msi_entry(entry, vector); 678 } 679 if (i != nvec) { 680 i--; 681 for (; i >= 0; i--) { 682 vector = (entries + i)->vector; 683 msi_free_vector(dev, vector, 0); 684 (entries + i)->vector = 0; 685 } 686 return -EBUSY; 687 } 688 /* Set MSI-X enabled bits */ 689 enable_msi_mode(dev, pos, PCI_CAP_ID_MSIX); 690 691 return 0; 692} 693 694/** 695 * pci_enable_msi - configure device's MSI capability structure 696 * @dev: pointer to the pci_dev data structure of MSI device function 697 * 698 * Setup the MSI capability structure of device function with 699 * a single MSI vector upon its software driver call to request for 700 * MSI mode enabled on its hardware device function. A return of zero 701 * indicates the successful setup of an entry zero with the new MSI 702 * vector or non-zero for otherwise. 703 **/ 704int pci_enable_msi(struct pci_dev* dev) 705{ 706 int pos, temp, status = -EINVAL; 707 u16 control; 708 709 if (!pci_msi_enable || !dev) 710 return status; 711 712 temp = dev->irq; 713 714 if ((status = msi_init()) < 0) 715 return status; 716 717 if (!(pos = pci_find_capability(dev, PCI_CAP_ID_MSI))) 718 return -EINVAL; 719 720 pci_read_config_word(dev, msi_control_reg(pos), &control); 721 if (control & PCI_MSI_FLAGS_ENABLE) 722 return 0; /* Already in MSI mode */ 723 724 if (!msi_lookup_vector(dev, PCI_CAP_ID_MSI)) { 725 /* Lookup Sucess */ 726 unsigned long flags; 727 728 spin_lock_irqsave(&msi_lock, flags); 729 if (!vector_irq[dev->irq]) { 730 msi_desc[dev->irq]->msi_attrib.state = 0; 731 vector_irq[dev->irq] = -1; 732 nr_released_vectors--; 733 spin_unlock_irqrestore(&msi_lock, flags); 734 enable_msi_mode(dev, pos, PCI_CAP_ID_MSI); 735 return 0; 736 } 737 spin_unlock_irqrestore(&msi_lock, flags); 738 dev->irq = temp; 739 } 740 /* Check whether driver already requested for MSI-X vectors */ 741 if ((pos = pci_find_capability(dev, PCI_CAP_ID_MSIX)) > 0 && 742 !msi_lookup_vector(dev, PCI_CAP_ID_MSIX)) { 743 printk(KERN_INFO "PCI: %s: Can't enable MSI. " 744 "Device already has MSI-X vectors assigned\n", 745 pci_name(dev)); 746 dev->irq = temp; 747 return -EINVAL; 748 } 749 status = msi_capability_init(dev); 750 if (!status) { 751 if (!pos) 752 nr_reserved_vectors--; /* Only MSI capable */ 753 else if (nr_msix_devices > 0) 754 nr_msix_devices--; /* Both MSI and MSI-X capable, 755 but choose enabling MSI */ 756 } 757 758 return status; 759} 760 761void pci_disable_msi(struct pci_dev* dev) 762{ 763 struct msi_desc *entry; 764 int pos, default_vector; 765 u16 control; 766 unsigned long flags; 767 768 if (!dev || !(pos = pci_find_capability(dev, PCI_CAP_ID_MSI))) 769 return; 770 771 pci_read_config_word(dev, msi_control_reg(pos), &control); 772 if (!(control & PCI_MSI_FLAGS_ENABLE)) 773 return; 774 775 spin_lock_irqsave(&msi_lock, flags); 776 entry = msi_desc[dev->irq]; 777 if (!entry || !entry->dev || entry->msi_attrib.type != PCI_CAP_ID_MSI) { 778 spin_unlock_irqrestore(&msi_lock, flags); 779 return; 780 } 781 if (entry->msi_attrib.state) { 782 spin_unlock_irqrestore(&msi_lock, flags); 783 printk(KERN_WARNING "PCI: %s: pci_disable_msi() called without " 784 "free_irq() on MSI vector %d\n", 785 pci_name(dev), dev->irq); 786 BUG_ON(entry->msi_attrib.state > 0); 787 } else { 788 vector_irq[dev->irq] = 0; /* free it */ 789 nr_released_vectors++; 790 default_vector = entry->msi_attrib.default_vector; 791 spin_unlock_irqrestore(&msi_lock, flags); 792 /* Restore dev->irq to its default pin-assertion vector */ 793 dev->irq = default_vector; 794 disable_msi_mode(dev, pci_find_capability(dev, PCI_CAP_ID_MSI), 795 PCI_CAP_ID_MSI); 796 } 797} 798 799static void release_msi(unsigned int vector) 800{ 801 struct msi_desc *entry; 802 unsigned long flags; 803 804 spin_lock_irqsave(&msi_lock, flags); 805 entry = msi_desc[vector]; 806 if (entry && entry->dev) 807 entry->msi_attrib.state = 0; /* Mark it not active */ 808 spin_unlock_irqrestore(&msi_lock, flags); 809} 810 811static int msi_free_vector(struct pci_dev* dev, int vector, int reassign) 812{ 813 struct msi_desc *entry; 814 int head, entry_nr, type; 815 void __iomem *base; 816 unsigned long flags; 817 818 spin_lock_irqsave(&msi_lock, flags); 819 entry = msi_desc[vector]; 820 if (!entry || entry->dev != dev) { 821 spin_unlock_irqrestore(&msi_lock, flags); 822 return -EINVAL; 823 } 824 type = entry->msi_attrib.type; 825 entry_nr = entry->msi_attrib.entry_nr; 826 head = entry->link.head; 827 base = entry->mask_base; 828 msi_desc[entry->link.head]->link.tail = entry->link.tail; 829 msi_desc[entry->link.tail]->link.head = entry->link.head; 830 entry->dev = NULL; 831 if (!reassign) { 832 vector_irq[vector] = 0; 833 nr_released_vectors++; 834 } 835 msi_desc[vector] = NULL; 836 spin_unlock_irqrestore(&msi_lock, flags); 837 838 kmem_cache_free(msi_cachep, entry); 839 840 if (type == PCI_CAP_ID_MSIX) { 841 if (!reassign) 842 writel(1, base + 843 entry_nr * PCI_MSIX_ENTRY_SIZE + 844 PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET); 845 846 if (head == vector) { 847 /* 848 * Detect last MSI-X vector to be released. 849 * Release the MSI-X memory-mapped table. 850 */ 851 int pos, nr_entries; 852 u32 phys_addr, table_offset; 853 u16 control; 854 u8 bir; 855 856 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX); 857 pci_read_config_word(dev, msi_control_reg(pos), 858 &control); 859 nr_entries = multi_msix_capable(control); 860 pci_read_config_dword(dev, msix_table_offset_reg(pos), 861 &table_offset); 862 bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK); 863 phys_addr = pci_resource_start (dev, bir); 864 phys_addr += (u32)(table_offset & 865 ~PCI_MSIX_FLAGS_BIRMASK); 866 iounmap(base); 867 } 868 } 869 870 return 0; 871} 872 873static int reroute_msix_table(int head, struct msix_entry *entries, int *nvec) 874{ 875 int vector = head, tail = 0; 876 int i, j = 0, nr_entries = 0; 877 void __iomem *base; 878 unsigned long flags; 879 880 spin_lock_irqsave(&msi_lock, flags); 881 while (head != tail) { 882 nr_entries++; 883 tail = msi_desc[vector]->link.tail; 884 if (entries[0].entry == msi_desc[vector]->msi_attrib.entry_nr) 885 j = vector; 886 vector = tail; 887 } 888 if (*nvec > nr_entries) { 889 spin_unlock_irqrestore(&msi_lock, flags); 890 *nvec = nr_entries; 891 return -EINVAL; 892 } 893 vector = ((j > 0) ? j : head); 894 for (i = 0; i < *nvec; i++) { 895 j = msi_desc[vector]->msi_attrib.entry_nr; 896 msi_desc[vector]->msi_attrib.state = 0; /* Mark it not active */ 897 vector_irq[vector] = -1; /* Mark it busy */ 898 nr_released_vectors--; 899 entries[i].vector = vector; 900 if (j != (entries + i)->entry) { 901 base = msi_desc[vector]->mask_base; 902 msi_desc[vector]->msi_attrib.entry_nr = 903 (entries + i)->entry; 904 writel( readl(base + j * PCI_MSIX_ENTRY_SIZE + 905 PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET), base + 906 (entries + i)->entry * PCI_MSIX_ENTRY_SIZE + 907 PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET); 908 writel( readl(base + j * PCI_MSIX_ENTRY_SIZE + 909 PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET), base + 910 (entries + i)->entry * PCI_MSIX_ENTRY_SIZE + 911 PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET); 912 writel( (readl(base + j * PCI_MSIX_ENTRY_SIZE + 913 PCI_MSIX_ENTRY_DATA_OFFSET) & 0xff00) | vector, 914 base + (entries+i)->entry*PCI_MSIX_ENTRY_SIZE + 915 PCI_MSIX_ENTRY_DATA_OFFSET); 916 } 917 vector = msi_desc[vector]->link.tail; 918 } 919 spin_unlock_irqrestore(&msi_lock, flags); 920 921 return 0; 922} 923 924/** 925 * pci_enable_msix - configure device's MSI-X capability structure 926 * @dev: pointer to the pci_dev data structure of MSI-X device function 927 * @data: pointer to an array of MSI-X entries 928 * @nvec: number of MSI-X vectors requested for allocation by device driver 929 * 930 * Setup the MSI-X capability structure of device function with the number 931 * of requested vectors upon its software driver call to request for 932 * MSI-X mode enabled on its hardware device function. A return of zero 933 * indicates the successful configuration of MSI-X capability structure 934 * with new allocated MSI-X vectors. A return of < 0 indicates a failure. 935 * Or a return of > 0 indicates that driver request is exceeding the number 936 * of vectors available. Driver should use the returned value to re-send 937 * its request. 938 **/ 939int pci_enable_msix(struct pci_dev* dev, struct msix_entry *entries, int nvec) 940{ 941 int status, pos, nr_entries, free_vectors; 942 int i, j, temp; 943 u16 control; 944 unsigned long flags; 945 946 if (!pci_msi_enable || !dev || !entries) 947 return -EINVAL; 948 949 if ((status = msi_init()) < 0) 950 return status; 951 952 if (!(pos = pci_find_capability(dev, PCI_CAP_ID_MSIX))) 953 return -EINVAL; 954 955 pci_read_config_word(dev, msi_control_reg(pos), &control); 956 if (control & PCI_MSIX_FLAGS_ENABLE) 957 return -EINVAL; /* Already in MSI-X mode */ 958 959 nr_entries = multi_msix_capable(control); 960 if (nvec > nr_entries) 961 return -EINVAL; 962 963 /* Check for any invalid entries */ 964 for (i = 0; i < nvec; i++) { 965 if (entries[i].entry >= nr_entries) 966 return -EINVAL; /* invalid entry */ 967 for (j = i + 1; j < nvec; j++) { 968 if (entries[i].entry == entries[j].entry) 969 return -EINVAL; /* duplicate entry */ 970 } 971 } 972 temp = dev->irq; 973 if (!msi_lookup_vector(dev, PCI_CAP_ID_MSIX)) { 974 /* Lookup Sucess */ 975 nr_entries = nvec; 976 /* Reroute MSI-X table */ 977 if (reroute_msix_table(dev->irq, entries, &nr_entries)) { 978 /* #requested > #previous-assigned */ 979 dev->irq = temp; 980 return nr_entries; 981 } 982 dev->irq = temp; 983 enable_msi_mode(dev, pos, PCI_CAP_ID_MSIX); 984 return 0; 985 } 986 /* Check whether driver already requested for MSI vector */ 987 if (pci_find_capability(dev, PCI_CAP_ID_MSI) > 0 && 988 !msi_lookup_vector(dev, PCI_CAP_ID_MSI)) { 989 printk(KERN_INFO "PCI: %s: Can't enable MSI-X. " 990 "Device already has an MSI vector assigned\n", 991 pci_name(dev)); 992 dev->irq = temp; 993 return -EINVAL; 994 } 995 996 spin_lock_irqsave(&msi_lock, flags); 997 /* 998 * msi_lock is provided to ensure that enough vectors resources are 999 * available before granting. 1000 */ 1001 free_vectors = pci_vector_resources(last_alloc_vector, 1002 nr_released_vectors); 1003 /* Ensure that each MSI/MSI-X device has one vector reserved by 1004 default to avoid any MSI-X driver to take all available 1005 resources */ 1006 free_vectors -= nr_reserved_vectors; 1007 /* Find the average of free vectors among MSI-X devices */ 1008 if (nr_msix_devices > 0) 1009 free_vectors /= nr_msix_devices; 1010 spin_unlock_irqrestore(&msi_lock, flags); 1011 1012 if (nvec > free_vectors) { 1013 if (free_vectors > 0) 1014 return free_vectors; 1015 else 1016 return -EBUSY; 1017 } 1018 1019 status = msix_capability_init(dev, entries, nvec); 1020 if (!status && nr_msix_devices > 0) 1021 nr_msix_devices--; 1022 1023 return status; 1024} 1025 1026void pci_disable_msix(struct pci_dev* dev) 1027{ 1028 int pos, temp; 1029 u16 control; 1030 1031 if (!dev || !(pos = pci_find_capability(dev, PCI_CAP_ID_MSIX))) 1032 return; 1033 1034 pci_read_config_word(dev, msi_control_reg(pos), &control); 1035 if (!(control & PCI_MSIX_FLAGS_ENABLE)) 1036 return; 1037 1038 temp = dev->irq; 1039 if (!msi_lookup_vector(dev, PCI_CAP_ID_MSIX)) { 1040 int state, vector, head, tail = 0, warning = 0; 1041 unsigned long flags; 1042 1043 vector = head = dev->irq; 1044 spin_lock_irqsave(&msi_lock, flags); 1045 while (head != tail) { 1046 state = msi_desc[vector]->msi_attrib.state; 1047 if (state) 1048 warning = 1; 1049 else { 1050 vector_irq[vector] = 0; /* free it */ 1051 nr_released_vectors++; 1052 } 1053 tail = msi_desc[vector]->link.tail; 1054 vector = tail; 1055 } 1056 spin_unlock_irqrestore(&msi_lock, flags); 1057 if (warning) { 1058 dev->irq = temp; 1059 printk(KERN_WARNING "PCI: %s: pci_disable_msix() called without " 1060 "free_irq() on all MSI-X vectors\n", 1061 pci_name(dev)); 1062 BUG_ON(warning > 0); 1063 } else { 1064 dev->irq = temp; 1065 disable_msi_mode(dev, 1066 pci_find_capability(dev, PCI_CAP_ID_MSIX), 1067 PCI_CAP_ID_MSIX); 1068 1069 } 1070 } 1071} 1072 1073/** 1074 * msi_remove_pci_irq_vectors - reclaim MSI(X) vectors to unused state 1075 * @dev: pointer to the pci_dev data structure of MSI(X) device function 1076 * 1077 * Being called during hotplug remove, from which the device function 1078 * is hot-removed. All previous assigned MSI/MSI-X vectors, if 1079 * allocated for this device function, are reclaimed to unused state, 1080 * which may be used later on. 1081 **/ 1082void msi_remove_pci_irq_vectors(struct pci_dev* dev) 1083{ 1084 int state, pos, temp; 1085 unsigned long flags; 1086 1087 if (!pci_msi_enable || !dev) 1088 return; 1089 1090 temp = dev->irq; /* Save IOAPIC IRQ */ 1091 if ((pos = pci_find_capability(dev, PCI_CAP_ID_MSI)) > 0 && 1092 !msi_lookup_vector(dev, PCI_CAP_ID_MSI)) { 1093 spin_lock_irqsave(&msi_lock, flags); 1094 state = msi_desc[dev->irq]->msi_attrib.state; 1095 spin_unlock_irqrestore(&msi_lock, flags); 1096 if (state) { 1097 printk(KERN_WARNING "PCI: %s: msi_remove_pci_irq_vectors() " 1098 "called without free_irq() on MSI vector %d\n", 1099 pci_name(dev), dev->irq); 1100 BUG_ON(state > 0); 1101 } else /* Release MSI vector assigned to this device */ 1102 msi_free_vector(dev, dev->irq, 0); 1103 dev->irq = temp; /* Restore IOAPIC IRQ */ 1104 } 1105 if ((pos = pci_find_capability(dev, PCI_CAP_ID_MSIX)) > 0 && 1106 !msi_lookup_vector(dev, PCI_CAP_ID_MSIX)) { 1107 int vector, head, tail = 0, warning = 0; 1108 void __iomem *base = NULL; 1109 1110 vector = head = dev->irq; 1111 while (head != tail) { 1112 spin_lock_irqsave(&msi_lock, flags); 1113 state = msi_desc[vector]->msi_attrib.state; 1114 tail = msi_desc[vector]->link.tail; 1115 base = msi_desc[vector]->mask_base; 1116 spin_unlock_irqrestore(&msi_lock, flags); 1117 if (state) 1118 warning = 1; 1119 else if (vector != head) /* Release MSI-X vector */ 1120 msi_free_vector(dev, vector, 0); 1121 vector = tail; 1122 } 1123 msi_free_vector(dev, vector, 0); 1124 if (warning) { 1125 /* Force to release the MSI-X memory-mapped table */ 1126 u32 phys_addr, table_offset; 1127 u16 control; 1128 u8 bir; 1129 1130 pci_read_config_word(dev, msi_control_reg(pos), 1131 &control); 1132 pci_read_config_dword(dev, msix_table_offset_reg(pos), 1133 &table_offset); 1134 bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK); 1135 phys_addr = pci_resource_start (dev, bir); 1136 phys_addr += (u32)(table_offset & 1137 ~PCI_MSIX_FLAGS_BIRMASK); 1138 iounmap(base); 1139 printk(KERN_WARNING "PCI: %s: msi_remove_pci_irq_vectors() " 1140 "called without free_irq() on all MSI-X vectors\n", 1141 pci_name(dev)); 1142 BUG_ON(warning > 0); 1143 } 1144 dev->irq = temp; /* Restore IOAPIC IRQ */ 1145 } 1146} 1147 1148EXPORT_SYMBOL(pci_enable_msi); 1149EXPORT_SYMBOL(pci_disable_msi); 1150EXPORT_SYMBOL(pci_enable_msix); 1151EXPORT_SYMBOL(pci_disable_msix); 1152