1 /* 2 * Copyright (C) 2012 Red Hat, Inc. All rights reserved. 3 * Author: Alex Williamson <alex.williamson@redhat.com> 4 * 5 * This program is free software; you can redistribute it and/or modify 6 * it under the terms of the GNU General Public License version 2 as 7 * published by the Free Software Foundation. 8 * 9 * Derived from original vfio: 10 * Copyright 2010 Cisco Systems, Inc. All rights reserved. 11 * Author: Tom Lyon, pugs@cisco.com 12 */ 13 14 #include <linux/device.h> 15 #include <linux/eventfd.h> 16 #include <linux/interrupt.h> 17 #include <linux/iommu.h> 18 #include <linux/module.h> 19 #include <linux/mutex.h> 20 #include <linux/notifier.h> 21 #include <linux/pci.h> 22 #include <linux/pm_runtime.h> 23 #include <linux/slab.h> 24 #include <linux/types.h> 25 #include <linux/uaccess.h> 26 #include <linux/vfio.h> 27 28 #include "vfio_pci_private.h" 29 30 #define DRIVER_VERSION "0.2" 31 #define DRIVER_AUTHOR "Alex Williamson <alex.williamson@redhat.com>" 32 #define DRIVER_DESC "VFIO PCI - User Level meta-driver" 33 34 static bool nointxmask; 35 module_param_named(nointxmask, nointxmask, bool, S_IRUGO | S_IWUSR); 36 MODULE_PARM_DESC(nointxmask, 37 "Disable support for PCI 2.3 style INTx masking. If this resolves problems for specific devices, report lspci -vvvxxx to linux-pci@vger.kernel.org so the device can be fixed automatically via the broken_intx_masking flag."); 38 39 static int vfio_pci_enable(struct vfio_pci_device *vdev) 40 { 41 struct pci_dev *pdev = vdev->pdev; 42 int ret; 43 u16 cmd; 44 u8 msix_pos; 45 46 ret = pci_enable_device(pdev); 47 if (ret) 48 return ret; 49 50 vdev->reset_works = (pci_reset_function(pdev) == 0); 51 pci_save_state(pdev); 52 vdev->pci_saved_state = pci_store_saved_state(pdev); 53 if (!vdev->pci_saved_state) 54 pr_debug("%s: Couldn't store %s saved state\n", 55 __func__, dev_name(&pdev->dev)); 56 57 ret = vfio_config_init(vdev); 58 if (ret) { 59 pci_load_and_free_saved_state(pdev, &vdev->pci_saved_state); 60 pci_disable_device(pdev); 61 return ret; 62 } 63 64 if (likely(!nointxmask)) 65 vdev->pci_2_3 = pci_intx_mask_supported(pdev); 66 67 pci_read_config_word(pdev, PCI_COMMAND, &cmd); 68 if (vdev->pci_2_3 && (cmd & PCI_COMMAND_INTX_DISABLE)) { 69 cmd &= ~PCI_COMMAND_INTX_DISABLE; 70 pci_write_config_word(pdev, PCI_COMMAND, cmd); 71 } 72 73 msix_pos = pdev->msix_cap; 74 if (msix_pos) { 75 u16 flags; 76 u32 table; 77 78 pci_read_config_word(pdev, msix_pos + PCI_MSIX_FLAGS, &flags); 79 pci_read_config_dword(pdev, msix_pos + PCI_MSIX_TABLE, &table); 80 81 vdev->msix_bar = table & PCI_MSIX_TABLE_BIR; 82 vdev->msix_offset = table & PCI_MSIX_TABLE_OFFSET; 83 vdev->msix_size = ((flags & PCI_MSIX_FLAGS_QSIZE) + 1) * 16; 84 } else 85 vdev->msix_bar = 0xFF; 86 87 #ifdef CONFIG_VFIO_PCI_VGA 88 if ((pdev->class >> 8) == PCI_CLASS_DISPLAY_VGA) 89 vdev->has_vga = true; 90 #endif 91 92 return 0; 93 } 94 95 static void vfio_pci_disable(struct vfio_pci_device *vdev) 96 { 97 struct pci_dev *pdev = vdev->pdev; 98 int bar; 99 100 pci_disable_device(pdev); 101 102 vfio_pci_set_irqs_ioctl(vdev, VFIO_IRQ_SET_DATA_NONE | 103 VFIO_IRQ_SET_ACTION_TRIGGER, 104 vdev->irq_type, 0, 0, NULL); 105 106 vdev->virq_disabled = false; 107 108 vfio_config_free(vdev); 109 110 for (bar = PCI_STD_RESOURCES; bar <= PCI_STD_RESOURCE_END; bar++) { 111 if (!vdev->barmap[bar]) 112 continue; 113 pci_iounmap(pdev, vdev->barmap[bar]); 114 pci_release_selected_regions(pdev, 1 << bar); 115 vdev->barmap[bar] = NULL; 116 } 117 118 /* 119 * If we have saved state, restore it. If we can reset the device, 120 * even better. Resetting with current state seems better than 121 * nothing, but saving and restoring current state without reset 122 * is just busy work. 123 */ 124 if (pci_load_and_free_saved_state(pdev, &vdev->pci_saved_state)) { 125 pr_info("%s: Couldn't reload %s saved state\n", 126 __func__, dev_name(&pdev->dev)); 127 128 if (!vdev->reset_works) 129 return; 130 131 pci_save_state(pdev); 132 } 133 134 /* 135 * Disable INTx and MSI, presumably to avoid spurious interrupts 136 * during reset. Stolen from pci_reset_function() 137 */ 138 pci_write_config_word(pdev, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE); 139 140 if (vdev->reset_works) 141 __pci_reset_function(pdev); 142 143 pci_restore_state(pdev); 144 } 145 146 static void vfio_pci_release(void *device_data) 147 { 148 struct vfio_pci_device *vdev = device_data; 149 150 if (atomic_dec_and_test(&vdev->refcnt)) 151 vfio_pci_disable(vdev); 152 153 module_put(THIS_MODULE); 154 } 155 156 static int vfio_pci_open(void *device_data) 157 { 158 struct vfio_pci_device *vdev = device_data; 159 160 if (!try_module_get(THIS_MODULE)) 161 return -ENODEV; 162 163 if (atomic_inc_return(&vdev->refcnt) == 1) { 164 int ret = vfio_pci_enable(vdev); 165 if (ret) { 166 module_put(THIS_MODULE); 167 return ret; 168 } 169 } 170 171 return 0; 172 } 173 174 static int vfio_pci_get_irq_count(struct vfio_pci_device *vdev, int irq_type) 175 { 176 if (irq_type == VFIO_PCI_INTX_IRQ_INDEX) { 177 u8 pin; 178 pci_read_config_byte(vdev->pdev, PCI_INTERRUPT_PIN, &pin); 179 if (pin) 180 return 1; 181 182 } else if (irq_type == VFIO_PCI_MSI_IRQ_INDEX) { 183 u8 pos; 184 u16 flags; 185 186 pos = vdev->pdev->msi_cap; 187 if (pos) { 188 pci_read_config_word(vdev->pdev, 189 pos + PCI_MSI_FLAGS, &flags); 190 191 return 1 << (flags & PCI_MSI_FLAGS_QMASK); 192 } 193 } else if (irq_type == VFIO_PCI_MSIX_IRQ_INDEX) { 194 u8 pos; 195 u16 flags; 196 197 pos = vdev->pdev->msix_cap; 198 if (pos) { 199 pci_read_config_word(vdev->pdev, 200 pos + PCI_MSIX_FLAGS, &flags); 201 202 return (flags & PCI_MSIX_FLAGS_QSIZE) + 1; 203 } 204 } else if (irq_type == VFIO_PCI_ERR_IRQ_INDEX) 205 if (pci_is_pcie(vdev->pdev)) 206 return 1; 207 208 return 0; 209 } 210 211 static long vfio_pci_ioctl(void *device_data, 212 unsigned int cmd, unsigned long arg) 213 { 214 struct vfio_pci_device *vdev = device_data; 215 unsigned long minsz; 216 217 if (cmd == VFIO_DEVICE_GET_INFO) { 218 struct vfio_device_info info; 219 220 minsz = offsetofend(struct vfio_device_info, num_irqs); 221 222 if (copy_from_user(&info, (void __user *)arg, minsz)) 223 return -EFAULT; 224 225 if (info.argsz < minsz) 226 return -EINVAL; 227 228 info.flags = VFIO_DEVICE_FLAGS_PCI; 229 230 if (vdev->reset_works) 231 info.flags |= VFIO_DEVICE_FLAGS_RESET; 232 233 info.num_regions = VFIO_PCI_NUM_REGIONS; 234 info.num_irqs = VFIO_PCI_NUM_IRQS; 235 236 return copy_to_user((void __user *)arg, &info, minsz); 237 238 } else if (cmd == VFIO_DEVICE_GET_REGION_INFO) { 239 struct pci_dev *pdev = vdev->pdev; 240 struct vfio_region_info info; 241 242 minsz = offsetofend(struct vfio_region_info, offset); 243 244 if (copy_from_user(&info, (void __user *)arg, minsz)) 245 return -EFAULT; 246 247 if (info.argsz < minsz) 248 return -EINVAL; 249 250 switch (info.index) { 251 case VFIO_PCI_CONFIG_REGION_INDEX: 252 info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index); 253 info.size = pdev->cfg_size; 254 info.flags = VFIO_REGION_INFO_FLAG_READ | 255 VFIO_REGION_INFO_FLAG_WRITE; 256 break; 257 case VFIO_PCI_BAR0_REGION_INDEX ... VFIO_PCI_BAR5_REGION_INDEX: 258 info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index); 259 info.size = pci_resource_len(pdev, info.index); 260 if (!info.size) { 261 info.flags = 0; 262 break; 263 } 264 265 info.flags = VFIO_REGION_INFO_FLAG_READ | 266 VFIO_REGION_INFO_FLAG_WRITE; 267 if (pci_resource_flags(pdev, info.index) & 268 IORESOURCE_MEM && info.size >= PAGE_SIZE) 269 info.flags |= VFIO_REGION_INFO_FLAG_MMAP; 270 break; 271 case VFIO_PCI_ROM_REGION_INDEX: 272 { 273 void __iomem *io; 274 size_t size; 275 276 info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index); 277 info.flags = 0; 278 279 /* Report the BAR size, not the ROM size */ 280 info.size = pci_resource_len(pdev, info.index); 281 if (!info.size) 282 break; 283 284 /* Is it really there? */ 285 io = pci_map_rom(pdev, &size); 286 if (!io || !size) { 287 info.size = 0; 288 break; 289 } 290 pci_unmap_rom(pdev, io); 291 292 info.flags = VFIO_REGION_INFO_FLAG_READ; 293 break; 294 } 295 case VFIO_PCI_VGA_REGION_INDEX: 296 if (!vdev->has_vga) 297 return -EINVAL; 298 299 info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index); 300 info.size = 0xc0000; 301 info.flags = VFIO_REGION_INFO_FLAG_READ | 302 VFIO_REGION_INFO_FLAG_WRITE; 303 304 break; 305 default: 306 return -EINVAL; 307 } 308 309 return copy_to_user((void __user *)arg, &info, minsz); 310 311 } else if (cmd == VFIO_DEVICE_GET_IRQ_INFO) { 312 struct vfio_irq_info info; 313 314 minsz = offsetofend(struct vfio_irq_info, count); 315 316 if (copy_from_user(&info, (void __user *)arg, minsz)) 317 return -EFAULT; 318 319 if (info.argsz < minsz || info.index >= VFIO_PCI_NUM_IRQS) 320 return -EINVAL; 321 322 switch (info.index) { 323 case VFIO_PCI_INTX_IRQ_INDEX ... VFIO_PCI_MSIX_IRQ_INDEX: 324 break; 325 case VFIO_PCI_ERR_IRQ_INDEX: 326 if (pci_is_pcie(vdev->pdev)) 327 break; 328 /* pass thru to return error */ 329 default: 330 return -EINVAL; 331 } 332 333 info.flags = VFIO_IRQ_INFO_EVENTFD; 334 335 info.count = vfio_pci_get_irq_count(vdev, info.index); 336 337 if (info.index == VFIO_PCI_INTX_IRQ_INDEX) 338 info.flags |= (VFIO_IRQ_INFO_MASKABLE | 339 VFIO_IRQ_INFO_AUTOMASKED); 340 else 341 info.flags |= VFIO_IRQ_INFO_NORESIZE; 342 343 return copy_to_user((void __user *)arg, &info, minsz); 344 345 } else if (cmd == VFIO_DEVICE_SET_IRQS) { 346 struct vfio_irq_set hdr; 347 u8 *data = NULL; 348 int ret = 0; 349 350 minsz = offsetofend(struct vfio_irq_set, count); 351 352 if (copy_from_user(&hdr, (void __user *)arg, minsz)) 353 return -EFAULT; 354 355 if (hdr.argsz < minsz || hdr.index >= VFIO_PCI_NUM_IRQS || 356 hdr.flags & ~(VFIO_IRQ_SET_DATA_TYPE_MASK | 357 VFIO_IRQ_SET_ACTION_TYPE_MASK)) 358 return -EINVAL; 359 360 if (!(hdr.flags & VFIO_IRQ_SET_DATA_NONE)) { 361 size_t size; 362 int max = vfio_pci_get_irq_count(vdev, hdr.index); 363 364 if (hdr.flags & VFIO_IRQ_SET_DATA_BOOL) 365 size = sizeof(uint8_t); 366 else if (hdr.flags & VFIO_IRQ_SET_DATA_EVENTFD) 367 size = sizeof(int32_t); 368 else 369 return -EINVAL; 370 371 if (hdr.argsz - minsz < hdr.count * size || 372 hdr.start >= max || hdr.start + hdr.count > max) 373 return -EINVAL; 374 375 data = memdup_user((void __user *)(arg + minsz), 376 hdr.count * size); 377 if (IS_ERR(data)) 378 return PTR_ERR(data); 379 } 380 381 mutex_lock(&vdev->igate); 382 383 ret = vfio_pci_set_irqs_ioctl(vdev, hdr.flags, hdr.index, 384 hdr.start, hdr.count, data); 385 386 mutex_unlock(&vdev->igate); 387 kfree(data); 388 389 return ret; 390 391 } else if (cmd == VFIO_DEVICE_RESET) 392 return vdev->reset_works ? 393 pci_reset_function(vdev->pdev) : -EINVAL; 394 395 return -ENOTTY; 396 } 397 398 static ssize_t vfio_pci_rw(void *device_data, char __user *buf, 399 size_t count, loff_t *ppos, bool iswrite) 400 { 401 unsigned int index = VFIO_PCI_OFFSET_TO_INDEX(*ppos); 402 struct vfio_pci_device *vdev = device_data; 403 404 if (index >= VFIO_PCI_NUM_REGIONS) 405 return -EINVAL; 406 407 switch (index) { 408 case VFIO_PCI_CONFIG_REGION_INDEX: 409 return vfio_pci_config_rw(vdev, buf, count, ppos, iswrite); 410 411 case VFIO_PCI_ROM_REGION_INDEX: 412 if (iswrite) 413 return -EINVAL; 414 return vfio_pci_bar_rw(vdev, buf, count, ppos, false); 415 416 case VFIO_PCI_BAR0_REGION_INDEX ... VFIO_PCI_BAR5_REGION_INDEX: 417 return vfio_pci_bar_rw(vdev, buf, count, ppos, iswrite); 418 419 case VFIO_PCI_VGA_REGION_INDEX: 420 return vfio_pci_vga_rw(vdev, buf, count, ppos, iswrite); 421 } 422 423 return -EINVAL; 424 } 425 426 static ssize_t vfio_pci_read(void *device_data, char __user *buf, 427 size_t count, loff_t *ppos) 428 { 429 if (!count) 430 return 0; 431 432 return vfio_pci_rw(device_data, buf, count, ppos, false); 433 } 434 435 static ssize_t vfio_pci_write(void *device_data, const char __user *buf, 436 size_t count, loff_t *ppos) 437 { 438 if (!count) 439 return 0; 440 441 return vfio_pci_rw(device_data, (char __user *)buf, count, ppos, true); 442 } 443 444 static int vfio_pci_mmap(void *device_data, struct vm_area_struct *vma) 445 { 446 struct vfio_pci_device *vdev = device_data; 447 struct pci_dev *pdev = vdev->pdev; 448 unsigned int index; 449 u64 phys_len, req_len, pgoff, req_start; 450 int ret; 451 452 index = vma->vm_pgoff >> (VFIO_PCI_OFFSET_SHIFT - PAGE_SHIFT); 453 454 if (vma->vm_end < vma->vm_start) 455 return -EINVAL; 456 if ((vma->vm_flags & VM_SHARED) == 0) 457 return -EINVAL; 458 if (index >= VFIO_PCI_ROM_REGION_INDEX) 459 return -EINVAL; 460 if (!(pci_resource_flags(pdev, index) & IORESOURCE_MEM)) 461 return -EINVAL; 462 463 phys_len = pci_resource_len(pdev, index); 464 req_len = vma->vm_end - vma->vm_start; 465 pgoff = vma->vm_pgoff & 466 ((1U << (VFIO_PCI_OFFSET_SHIFT - PAGE_SHIFT)) - 1); 467 req_start = pgoff << PAGE_SHIFT; 468 469 if (phys_len < PAGE_SIZE || req_start + req_len > phys_len) 470 return -EINVAL; 471 472 if (index == vdev->msix_bar) { 473 /* 474 * Disallow mmaps overlapping the MSI-X table; users don't 475 * get to touch this directly. We could find somewhere 476 * else to map the overlap, but page granularity is only 477 * a recommendation, not a requirement, so the user needs 478 * to know which bits are real. Requiring them to mmap 479 * around the table makes that clear. 480 */ 481 482 /* If neither entirely above nor below, then it overlaps */ 483 if (!(req_start >= vdev->msix_offset + vdev->msix_size || 484 req_start + req_len <= vdev->msix_offset)) 485 return -EINVAL; 486 } 487 488 /* 489 * Even though we don't make use of the barmap for the mmap, 490 * we need to request the region and the barmap tracks that. 491 */ 492 if (!vdev->barmap[index]) { 493 ret = pci_request_selected_regions(pdev, 494 1 << index, "vfio-pci"); 495 if (ret) 496 return ret; 497 498 vdev->barmap[index] = pci_iomap(pdev, index, 0); 499 } 500 501 vma->vm_private_data = vdev; 502 vma->vm_flags |= VM_IO | VM_DONTEXPAND | VM_DONTDUMP; 503 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot); 504 vma->vm_pgoff = (pci_resource_start(pdev, index) >> PAGE_SHIFT) + pgoff; 505 506 return remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff, 507 req_len, vma->vm_page_prot); 508 } 509 510 static const struct vfio_device_ops vfio_pci_ops = { 511 .name = "vfio-pci", 512 .open = vfio_pci_open, 513 .release = vfio_pci_release, 514 .ioctl = vfio_pci_ioctl, 515 .read = vfio_pci_read, 516 .write = vfio_pci_write, 517 .mmap = vfio_pci_mmap, 518 }; 519 520 static int vfio_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id) 521 { 522 u8 type; 523 struct vfio_pci_device *vdev; 524 struct iommu_group *group; 525 int ret; 526 527 pci_read_config_byte(pdev, PCI_HEADER_TYPE, &type); 528 if ((type & PCI_HEADER_TYPE) != PCI_HEADER_TYPE_NORMAL) 529 return -EINVAL; 530 531 group = iommu_group_get(&pdev->dev); 532 if (!group) 533 return -EINVAL; 534 535 vdev = kzalloc(sizeof(*vdev), GFP_KERNEL); 536 if (!vdev) { 537 iommu_group_put(group); 538 return -ENOMEM; 539 } 540 541 vdev->pdev = pdev; 542 vdev->irq_type = VFIO_PCI_NUM_IRQS; 543 mutex_init(&vdev->igate); 544 spin_lock_init(&vdev->irqlock); 545 atomic_set(&vdev->refcnt, 0); 546 547 ret = vfio_add_group_dev(&pdev->dev, &vfio_pci_ops, vdev); 548 if (ret) { 549 iommu_group_put(group); 550 kfree(vdev); 551 } 552 553 return ret; 554 } 555 556 static void vfio_pci_remove(struct pci_dev *pdev) 557 { 558 struct vfio_pci_device *vdev; 559 560 vdev = vfio_del_group_dev(&pdev->dev); 561 if (!vdev) 562 return; 563 564 iommu_group_put(pdev->dev.iommu_group); 565 kfree(vdev); 566 } 567 568 static pci_ers_result_t vfio_pci_aer_err_detected(struct pci_dev *pdev, 569 pci_channel_state_t state) 570 { 571 struct vfio_pci_device *vdev; 572 struct vfio_device *device; 573 574 device = vfio_device_get_from_dev(&pdev->dev); 575 if (device == NULL) 576 return PCI_ERS_RESULT_DISCONNECT; 577 578 vdev = vfio_device_data(device); 579 if (vdev == NULL) { 580 vfio_device_put(device); 581 return PCI_ERS_RESULT_DISCONNECT; 582 } 583 584 if (vdev->err_trigger) 585 eventfd_signal(vdev->err_trigger, 1); 586 587 vfio_device_put(device); 588 589 return PCI_ERS_RESULT_CAN_RECOVER; 590 } 591 592 static struct pci_error_handlers vfio_err_handlers = { 593 .error_detected = vfio_pci_aer_err_detected, 594 }; 595 596 static struct pci_driver vfio_pci_driver = { 597 .name = "vfio-pci", 598 .id_table = NULL, /* only dynamic ids */ 599 .probe = vfio_pci_probe, 600 .remove = vfio_pci_remove, 601 .err_handler = &vfio_err_handlers, 602 }; 603 604 static void __exit vfio_pci_cleanup(void) 605 { 606 pci_unregister_driver(&vfio_pci_driver); 607 vfio_pci_virqfd_exit(); 608 vfio_pci_uninit_perm_bits(); 609 } 610 611 static int __init vfio_pci_init(void) 612 { 613 int ret; 614 615 /* Allocate shared config space permision data used by all devices */ 616 ret = vfio_pci_init_perm_bits(); 617 if (ret) 618 return ret; 619 620 /* Start the virqfd cleanup handler */ 621 ret = vfio_pci_virqfd_init(); 622 if (ret) 623 goto out_virqfd; 624 625 /* Register and scan for devices */ 626 ret = pci_register_driver(&vfio_pci_driver); 627 if (ret) 628 goto out_driver; 629 630 return 0; 631 632 out_driver: 633 vfio_pci_virqfd_exit(); 634 out_virqfd: 635 vfio_pci_uninit_perm_bits(); 636 return ret; 637 } 638 639 module_init(vfio_pci_init); 640 module_exit(vfio_pci_cleanup); 641 642 MODULE_VERSION(DRIVER_VERSION); 643 MODULE_LICENSE("GPL v2"); 644 MODULE_AUTHOR(DRIVER_AUTHOR); 645 MODULE_DESCRIPTION(DRIVER_DESC); 646