1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (c) 2023, NVIDIA CORPORATION & AFFILIATES. All rights reserved 4 */ 5 6 #include <linux/device.h> 7 #include <linux/module.h> 8 #include <linux/mutex.h> 9 #include <linux/pci.h> 10 #include <linux/pm_runtime.h> 11 #include <linux/types.h> 12 #include <linux/uaccess.h> 13 #include <linux/vfio.h> 14 #include <linux/vfio_pci_core.h> 15 #include <linux/virtio_pci.h> 16 #include <linux/virtio_net.h> 17 #include <linux/virtio_pci_admin.h> 18 19 struct virtiovf_pci_core_device { 20 struct vfio_pci_core_device core_device; 21 u8 *bar0_virtual_buf; 22 /* synchronize access to the virtual buf */ 23 struct mutex bar_mutex; 24 void __iomem *notify_addr; 25 u64 notify_offset; 26 __le32 pci_base_addr_0; 27 __le16 pci_cmd; 28 u8 bar0_virtual_buf_size; 29 u8 notify_bar; 30 }; 31 32 static int 33 virtiovf_issue_legacy_rw_cmd(struct virtiovf_pci_core_device *virtvdev, 34 loff_t pos, char __user *buf, 35 size_t count, bool read) 36 { 37 bool msix_enabled = 38 (virtvdev->core_device.irq_type == VFIO_PCI_MSIX_IRQ_INDEX); 39 struct pci_dev *pdev = virtvdev->core_device.pdev; 40 u8 *bar0_buf = virtvdev->bar0_virtual_buf; 41 bool common; 42 u8 offset; 43 int ret; 44 45 common = pos < VIRTIO_PCI_CONFIG_OFF(msix_enabled); 46 /* offset within the relevant configuration area */ 47 offset = common ? pos : pos - VIRTIO_PCI_CONFIG_OFF(msix_enabled); 48 mutex_lock(&virtvdev->bar_mutex); 49 if (read) { 50 if (common) 51 ret = virtio_pci_admin_legacy_common_io_read(pdev, offset, 52 count, bar0_buf + pos); 53 else 54 ret = virtio_pci_admin_legacy_device_io_read(pdev, offset, 55 count, bar0_buf + pos); 56 if (ret) 57 goto out; 58 if (copy_to_user(buf, bar0_buf + pos, count)) 59 ret = -EFAULT; 60 } else { 61 if (copy_from_user(bar0_buf + pos, buf, count)) { 62 ret = -EFAULT; 63 goto out; 64 } 65 66 if (common) 67 ret = virtio_pci_admin_legacy_common_io_write(pdev, offset, 68 count, bar0_buf + pos); 69 else 70 ret = virtio_pci_admin_legacy_device_io_write(pdev, offset, 71 count, bar0_buf + pos); 72 } 73 out: 74 mutex_unlock(&virtvdev->bar_mutex); 75 return ret; 76 } 77 78 static int 79 virtiovf_pci_bar0_rw(struct virtiovf_pci_core_device *virtvdev, 80 loff_t pos, char __user *buf, 81 size_t count, bool read) 82 { 83 struct vfio_pci_core_device *core_device = &virtvdev->core_device; 84 struct pci_dev *pdev = core_device->pdev; 85 u16 queue_notify; 86 int ret; 87 88 if (!(le16_to_cpu(virtvdev->pci_cmd) & PCI_COMMAND_IO)) 89 return -EIO; 90 91 if (pos + count > virtvdev->bar0_virtual_buf_size) 92 return -EINVAL; 93 94 ret = pm_runtime_resume_and_get(&pdev->dev); 95 if (ret) { 96 pci_info_ratelimited(pdev, "runtime resume failed %d\n", ret); 97 return -EIO; 98 } 99 100 switch (pos) { 101 case VIRTIO_PCI_QUEUE_NOTIFY: 102 if (count != sizeof(queue_notify)) { 103 ret = -EINVAL; 104 goto end; 105 } 106 if (read) { 107 ret = vfio_pci_core_ioread16(core_device, true, &queue_notify, 108 virtvdev->notify_addr); 109 if (ret) 110 goto end; 111 if (copy_to_user(buf, &queue_notify, 112 sizeof(queue_notify))) { 113 ret = -EFAULT; 114 goto end; 115 } 116 } else { 117 if (copy_from_user(&queue_notify, buf, count)) { 118 ret = -EFAULT; 119 goto end; 120 } 121 ret = vfio_pci_core_iowrite16(core_device, true, queue_notify, 122 virtvdev->notify_addr); 123 } 124 break; 125 default: 126 ret = virtiovf_issue_legacy_rw_cmd(virtvdev, pos, buf, count, 127 read); 128 } 129 130 end: 131 pm_runtime_put(&pdev->dev); 132 return ret ? ret : count; 133 } 134 135 static bool range_intersect_range(loff_t range1_start, size_t count1, 136 loff_t range2_start, size_t count2, 137 loff_t *start_offset, 138 size_t *intersect_count, 139 size_t *register_offset) 140 { 141 if (range1_start <= range2_start && 142 range1_start + count1 > range2_start) { 143 *start_offset = range2_start - range1_start; 144 *intersect_count = min_t(size_t, count2, 145 range1_start + count1 - range2_start); 146 *register_offset = 0; 147 return true; 148 } 149 150 if (range1_start > range2_start && 151 range1_start < range2_start + count2) { 152 *start_offset = 0; 153 *intersect_count = min_t(size_t, count1, 154 range2_start + count2 - range1_start); 155 *register_offset = range1_start - range2_start; 156 return true; 157 } 158 159 return false; 160 } 161 162 static ssize_t virtiovf_pci_read_config(struct vfio_device *core_vdev, 163 char __user *buf, size_t count, 164 loff_t *ppos) 165 { 166 struct virtiovf_pci_core_device *virtvdev = container_of( 167 core_vdev, struct virtiovf_pci_core_device, core_device.vdev); 168 loff_t pos = *ppos & VFIO_PCI_OFFSET_MASK; 169 size_t register_offset; 170 loff_t copy_offset; 171 size_t copy_count; 172 __le32 val32; 173 __le16 val16; 174 u8 val8; 175 int ret; 176 177 ret = vfio_pci_core_read(core_vdev, buf, count, ppos); 178 if (ret < 0) 179 return ret; 180 181 if (range_intersect_range(pos, count, PCI_DEVICE_ID, sizeof(val16), 182 ©_offset, ©_count, ®ister_offset)) { 183 val16 = cpu_to_le16(VIRTIO_TRANS_ID_NET); 184 if (copy_to_user(buf + copy_offset, (void *)&val16 + register_offset, copy_count)) 185 return -EFAULT; 186 } 187 188 if ((le16_to_cpu(virtvdev->pci_cmd) & PCI_COMMAND_IO) && 189 range_intersect_range(pos, count, PCI_COMMAND, sizeof(val16), 190 ©_offset, ©_count, ®ister_offset)) { 191 if (copy_from_user((void *)&val16 + register_offset, buf + copy_offset, 192 copy_count)) 193 return -EFAULT; 194 val16 |= cpu_to_le16(PCI_COMMAND_IO); 195 if (copy_to_user(buf + copy_offset, (void *)&val16 + register_offset, 196 copy_count)) 197 return -EFAULT; 198 } 199 200 if (range_intersect_range(pos, count, PCI_REVISION_ID, sizeof(val8), 201 ©_offset, ©_count, ®ister_offset)) { 202 /* Transional needs to have revision 0 */ 203 val8 = 0; 204 if (copy_to_user(buf + copy_offset, &val8, copy_count)) 205 return -EFAULT; 206 } 207 208 if (range_intersect_range(pos, count, PCI_BASE_ADDRESS_0, sizeof(val32), 209 ©_offset, ©_count, ®ister_offset)) { 210 u32 bar_mask = ~(virtvdev->bar0_virtual_buf_size - 1); 211 u32 pci_base_addr_0 = le32_to_cpu(virtvdev->pci_base_addr_0); 212 213 val32 = cpu_to_le32((pci_base_addr_0 & bar_mask) | PCI_BASE_ADDRESS_SPACE_IO); 214 if (copy_to_user(buf + copy_offset, (void *)&val32 + register_offset, copy_count)) 215 return -EFAULT; 216 } 217 218 if (range_intersect_range(pos, count, PCI_SUBSYSTEM_ID, sizeof(val16), 219 ©_offset, ©_count, ®ister_offset)) { 220 /* 221 * Transitional devices use the PCI subsystem device id as 222 * virtio device id, same as legacy driver always did. 223 */ 224 val16 = cpu_to_le16(VIRTIO_ID_NET); 225 if (copy_to_user(buf + copy_offset, (void *)&val16 + register_offset, 226 copy_count)) 227 return -EFAULT; 228 } 229 230 if (range_intersect_range(pos, count, PCI_SUBSYSTEM_VENDOR_ID, sizeof(val16), 231 ©_offset, ©_count, ®ister_offset)) { 232 val16 = cpu_to_le16(PCI_VENDOR_ID_REDHAT_QUMRANET); 233 if (copy_to_user(buf + copy_offset, (void *)&val16 + register_offset, 234 copy_count)) 235 return -EFAULT; 236 } 237 238 return count; 239 } 240 241 static ssize_t 242 virtiovf_pci_core_read(struct vfio_device *core_vdev, char __user *buf, 243 size_t count, loff_t *ppos) 244 { 245 struct virtiovf_pci_core_device *virtvdev = container_of( 246 core_vdev, struct virtiovf_pci_core_device, core_device.vdev); 247 unsigned int index = VFIO_PCI_OFFSET_TO_INDEX(*ppos); 248 loff_t pos = *ppos & VFIO_PCI_OFFSET_MASK; 249 250 if (!count) 251 return 0; 252 253 if (index == VFIO_PCI_CONFIG_REGION_INDEX) 254 return virtiovf_pci_read_config(core_vdev, buf, count, ppos); 255 256 if (index == VFIO_PCI_BAR0_REGION_INDEX) 257 return virtiovf_pci_bar0_rw(virtvdev, pos, buf, count, true); 258 259 return vfio_pci_core_read(core_vdev, buf, count, ppos); 260 } 261 262 static ssize_t virtiovf_pci_write_config(struct vfio_device *core_vdev, 263 const char __user *buf, size_t count, 264 loff_t *ppos) 265 { 266 struct virtiovf_pci_core_device *virtvdev = container_of( 267 core_vdev, struct virtiovf_pci_core_device, core_device.vdev); 268 loff_t pos = *ppos & VFIO_PCI_OFFSET_MASK; 269 size_t register_offset; 270 loff_t copy_offset; 271 size_t copy_count; 272 273 if (range_intersect_range(pos, count, PCI_COMMAND, sizeof(virtvdev->pci_cmd), 274 ©_offset, ©_count, 275 ®ister_offset)) { 276 if (copy_from_user((void *)&virtvdev->pci_cmd + register_offset, 277 buf + copy_offset, 278 copy_count)) 279 return -EFAULT; 280 } 281 282 if (range_intersect_range(pos, count, PCI_BASE_ADDRESS_0, 283 sizeof(virtvdev->pci_base_addr_0), 284 ©_offset, ©_count, 285 ®ister_offset)) { 286 if (copy_from_user((void *)&virtvdev->pci_base_addr_0 + register_offset, 287 buf + copy_offset, 288 copy_count)) 289 return -EFAULT; 290 } 291 292 return vfio_pci_core_write(core_vdev, buf, count, ppos); 293 } 294 295 static ssize_t 296 virtiovf_pci_core_write(struct vfio_device *core_vdev, const char __user *buf, 297 size_t count, loff_t *ppos) 298 { 299 struct virtiovf_pci_core_device *virtvdev = container_of( 300 core_vdev, struct virtiovf_pci_core_device, core_device.vdev); 301 unsigned int index = VFIO_PCI_OFFSET_TO_INDEX(*ppos); 302 loff_t pos = *ppos & VFIO_PCI_OFFSET_MASK; 303 304 if (!count) 305 return 0; 306 307 if (index == VFIO_PCI_CONFIG_REGION_INDEX) 308 return virtiovf_pci_write_config(core_vdev, buf, count, ppos); 309 310 if (index == VFIO_PCI_BAR0_REGION_INDEX) 311 return virtiovf_pci_bar0_rw(virtvdev, pos, (char __user *)buf, count, false); 312 313 return vfio_pci_core_write(core_vdev, buf, count, ppos); 314 } 315 316 static int 317 virtiovf_pci_ioctl_get_region_info(struct vfio_device *core_vdev, 318 unsigned int cmd, unsigned long arg) 319 { 320 struct virtiovf_pci_core_device *virtvdev = container_of( 321 core_vdev, struct virtiovf_pci_core_device, core_device.vdev); 322 unsigned long minsz = offsetofend(struct vfio_region_info, offset); 323 void __user *uarg = (void __user *)arg; 324 struct vfio_region_info info = {}; 325 326 if (copy_from_user(&info, uarg, minsz)) 327 return -EFAULT; 328 329 if (info.argsz < minsz) 330 return -EINVAL; 331 332 switch (info.index) { 333 case VFIO_PCI_BAR0_REGION_INDEX: 334 info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index); 335 info.size = virtvdev->bar0_virtual_buf_size; 336 info.flags = VFIO_REGION_INFO_FLAG_READ | 337 VFIO_REGION_INFO_FLAG_WRITE; 338 return copy_to_user(uarg, &info, minsz) ? -EFAULT : 0; 339 default: 340 return vfio_pci_core_ioctl(core_vdev, cmd, arg); 341 } 342 } 343 344 static long 345 virtiovf_vfio_pci_core_ioctl(struct vfio_device *core_vdev, unsigned int cmd, 346 unsigned long arg) 347 { 348 switch (cmd) { 349 case VFIO_DEVICE_GET_REGION_INFO: 350 return virtiovf_pci_ioctl_get_region_info(core_vdev, cmd, arg); 351 default: 352 return vfio_pci_core_ioctl(core_vdev, cmd, arg); 353 } 354 } 355 356 static int 357 virtiovf_set_notify_addr(struct virtiovf_pci_core_device *virtvdev) 358 { 359 struct vfio_pci_core_device *core_device = &virtvdev->core_device; 360 int ret; 361 362 /* 363 * Setup the BAR where the 'notify' exists to be used by vfio as well 364 * This will let us mmap it only once and use it when needed. 365 */ 366 ret = vfio_pci_core_setup_barmap(core_device, 367 virtvdev->notify_bar); 368 if (ret) 369 return ret; 370 371 virtvdev->notify_addr = core_device->barmap[virtvdev->notify_bar] + 372 virtvdev->notify_offset; 373 return 0; 374 } 375 376 static int virtiovf_pci_open_device(struct vfio_device *core_vdev) 377 { 378 struct virtiovf_pci_core_device *virtvdev = container_of( 379 core_vdev, struct virtiovf_pci_core_device, core_device.vdev); 380 struct vfio_pci_core_device *vdev = &virtvdev->core_device; 381 int ret; 382 383 ret = vfio_pci_core_enable(vdev); 384 if (ret) 385 return ret; 386 387 if (virtvdev->bar0_virtual_buf) { 388 /* 389 * Upon close_device() the vfio_pci_core_disable() is called 390 * and will close all the previous mmaps, so it seems that the 391 * valid life cycle for the 'notify' addr is per open/close. 392 */ 393 ret = virtiovf_set_notify_addr(virtvdev); 394 if (ret) { 395 vfio_pci_core_disable(vdev); 396 return ret; 397 } 398 } 399 400 vfio_pci_core_finish_enable(vdev); 401 return 0; 402 } 403 404 static int virtiovf_get_device_config_size(unsigned short device) 405 { 406 /* Network card */ 407 return offsetofend(struct virtio_net_config, status); 408 } 409 410 static int virtiovf_read_notify_info(struct virtiovf_pci_core_device *virtvdev) 411 { 412 u64 offset; 413 int ret; 414 u8 bar; 415 416 ret = virtio_pci_admin_legacy_io_notify_info(virtvdev->core_device.pdev, 417 VIRTIO_ADMIN_CMD_NOTIFY_INFO_FLAGS_OWNER_MEM, 418 &bar, &offset); 419 if (ret) 420 return ret; 421 422 virtvdev->notify_bar = bar; 423 virtvdev->notify_offset = offset; 424 return 0; 425 } 426 427 static int virtiovf_pci_init_device(struct vfio_device *core_vdev) 428 { 429 struct virtiovf_pci_core_device *virtvdev = container_of( 430 core_vdev, struct virtiovf_pci_core_device, core_device.vdev); 431 struct pci_dev *pdev; 432 int ret; 433 434 ret = vfio_pci_core_init_dev(core_vdev); 435 if (ret) 436 return ret; 437 438 pdev = virtvdev->core_device.pdev; 439 ret = virtiovf_read_notify_info(virtvdev); 440 if (ret) 441 return ret; 442 443 virtvdev->bar0_virtual_buf_size = VIRTIO_PCI_CONFIG_OFF(true) + 444 virtiovf_get_device_config_size(pdev->device); 445 BUILD_BUG_ON(!is_power_of_2(virtvdev->bar0_virtual_buf_size)); 446 virtvdev->bar0_virtual_buf = kzalloc(virtvdev->bar0_virtual_buf_size, 447 GFP_KERNEL); 448 if (!virtvdev->bar0_virtual_buf) 449 return -ENOMEM; 450 mutex_init(&virtvdev->bar_mutex); 451 return 0; 452 } 453 454 static void virtiovf_pci_core_release_dev(struct vfio_device *core_vdev) 455 { 456 struct virtiovf_pci_core_device *virtvdev = container_of( 457 core_vdev, struct virtiovf_pci_core_device, core_device.vdev); 458 459 kfree(virtvdev->bar0_virtual_buf); 460 vfio_pci_core_release_dev(core_vdev); 461 } 462 463 static const struct vfio_device_ops virtiovf_vfio_pci_tran_ops = { 464 .name = "virtio-vfio-pci-trans", 465 .init = virtiovf_pci_init_device, 466 .release = virtiovf_pci_core_release_dev, 467 .open_device = virtiovf_pci_open_device, 468 .close_device = vfio_pci_core_close_device, 469 .ioctl = virtiovf_vfio_pci_core_ioctl, 470 .device_feature = vfio_pci_core_ioctl_feature, 471 .read = virtiovf_pci_core_read, 472 .write = virtiovf_pci_core_write, 473 .mmap = vfio_pci_core_mmap, 474 .request = vfio_pci_core_request, 475 .match = vfio_pci_core_match, 476 .bind_iommufd = vfio_iommufd_physical_bind, 477 .unbind_iommufd = vfio_iommufd_physical_unbind, 478 .attach_ioas = vfio_iommufd_physical_attach_ioas, 479 .detach_ioas = vfio_iommufd_physical_detach_ioas, 480 }; 481 482 static const struct vfio_device_ops virtiovf_vfio_pci_ops = { 483 .name = "virtio-vfio-pci", 484 .init = vfio_pci_core_init_dev, 485 .release = vfio_pci_core_release_dev, 486 .open_device = virtiovf_pci_open_device, 487 .close_device = vfio_pci_core_close_device, 488 .ioctl = vfio_pci_core_ioctl, 489 .device_feature = vfio_pci_core_ioctl_feature, 490 .read = vfio_pci_core_read, 491 .write = vfio_pci_core_write, 492 .mmap = vfio_pci_core_mmap, 493 .request = vfio_pci_core_request, 494 .match = vfio_pci_core_match, 495 .bind_iommufd = vfio_iommufd_physical_bind, 496 .unbind_iommufd = vfio_iommufd_physical_unbind, 497 .attach_ioas = vfio_iommufd_physical_attach_ioas, 498 .detach_ioas = vfio_iommufd_physical_detach_ioas, 499 }; 500 501 static bool virtiovf_bar0_exists(struct pci_dev *pdev) 502 { 503 struct resource *res = pdev->resource; 504 505 return res->flags; 506 } 507 508 static int virtiovf_pci_probe(struct pci_dev *pdev, 509 const struct pci_device_id *id) 510 { 511 const struct vfio_device_ops *ops = &virtiovf_vfio_pci_ops; 512 struct virtiovf_pci_core_device *virtvdev; 513 int ret; 514 515 if (pdev->is_virtfn && virtio_pci_admin_has_legacy_io(pdev) && 516 !virtiovf_bar0_exists(pdev)) 517 ops = &virtiovf_vfio_pci_tran_ops; 518 519 virtvdev = vfio_alloc_device(virtiovf_pci_core_device, core_device.vdev, 520 &pdev->dev, ops); 521 if (IS_ERR(virtvdev)) 522 return PTR_ERR(virtvdev); 523 524 dev_set_drvdata(&pdev->dev, &virtvdev->core_device); 525 ret = vfio_pci_core_register_device(&virtvdev->core_device); 526 if (ret) 527 goto out; 528 return 0; 529 out: 530 vfio_put_device(&virtvdev->core_device.vdev); 531 return ret; 532 } 533 534 static void virtiovf_pci_remove(struct pci_dev *pdev) 535 { 536 struct virtiovf_pci_core_device *virtvdev = dev_get_drvdata(&pdev->dev); 537 538 vfio_pci_core_unregister_device(&virtvdev->core_device); 539 vfio_put_device(&virtvdev->core_device.vdev); 540 } 541 542 static const struct pci_device_id virtiovf_pci_table[] = { 543 /* Only virtio-net is supported/tested so far */ 544 { PCI_DRIVER_OVERRIDE_DEVICE_VFIO(PCI_VENDOR_ID_REDHAT_QUMRANET, 0x1041) }, 545 {} 546 }; 547 548 MODULE_DEVICE_TABLE(pci, virtiovf_pci_table); 549 550 static void virtiovf_pci_aer_reset_done(struct pci_dev *pdev) 551 { 552 struct virtiovf_pci_core_device *virtvdev = dev_get_drvdata(&pdev->dev); 553 554 virtvdev->pci_cmd = 0; 555 } 556 557 static const struct pci_error_handlers virtiovf_err_handlers = { 558 .reset_done = virtiovf_pci_aer_reset_done, 559 .error_detected = vfio_pci_core_aer_err_detected, 560 }; 561 562 static struct pci_driver virtiovf_pci_driver = { 563 .name = KBUILD_MODNAME, 564 .id_table = virtiovf_pci_table, 565 .probe = virtiovf_pci_probe, 566 .remove = virtiovf_pci_remove, 567 .err_handler = &virtiovf_err_handlers, 568 .driver_managed_dma = true, 569 }; 570 571 module_pci_driver(virtiovf_pci_driver); 572 573 MODULE_LICENSE("GPL"); 574 MODULE_AUTHOR("Yishai Hadas <yishaih@nvidia.com>"); 575 MODULE_DESCRIPTION( 576 "VIRTIO VFIO PCI - User Level meta-driver for VIRTIO NET devices"); 577