1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (C) 2018-2020 Intel Corporation. 4 * Copyright (C) 2020 Red Hat, Inc. 5 * 6 * Author: Tiwei Bie <tiwei.bie@intel.com> 7 * Jason Wang <jasowang@redhat.com> 8 * 9 * Thanks Michael S. Tsirkin for the valuable comments and 10 * suggestions. And thanks to Cunming Liang and Zhihong Wang for all 11 * their supports. 12 */ 13 14 #include <linux/kernel.h> 15 #include <linux/module.h> 16 #include <linux/cdev.h> 17 #include <linux/device.h> 18 #include <linux/mm.h> 19 #include <linux/iommu.h> 20 #include <linux/uuid.h> 21 #include <linux/vdpa.h> 22 #include <linux/nospec.h> 23 #include <linux/vhost.h> 24 #include <linux/virtio_net.h> 25 26 #include "vhost.h" 27 28 enum { 29 VHOST_VDPA_BACKEND_FEATURES = 30 (1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2) | 31 (1ULL << VHOST_BACKEND_F_IOTLB_BATCH), 32 }; 33 34 #define VHOST_VDPA_DEV_MAX (1U << MINORBITS) 35 36 struct vhost_vdpa { 37 struct vhost_dev vdev; 38 struct iommu_domain *domain; 39 struct vhost_virtqueue *vqs; 40 struct completion completion; 41 struct vdpa_device *vdpa; 42 struct device dev; 43 struct cdev cdev; 44 atomic_t opened; 45 int nvqs; 46 int virtio_id; 47 int minor; 48 struct eventfd_ctx *config_ctx; 49 int in_batch; 50 }; 51 52 static DEFINE_IDA(vhost_vdpa_ida); 53 54 static dev_t vhost_vdpa_major; 55 56 static void handle_vq_kick(struct vhost_work *work) 57 { 58 struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue, 59 poll.work); 60 struct vhost_vdpa *v = container_of(vq->dev, struct vhost_vdpa, vdev); 61 const struct vdpa_config_ops *ops = v->vdpa->config; 62 63 ops->kick_vq(v->vdpa, vq - v->vqs); 64 } 65 66 static irqreturn_t vhost_vdpa_virtqueue_cb(void *private) 67 { 68 struct vhost_virtqueue *vq = private; 69 struct eventfd_ctx *call_ctx = vq->call_ctx.ctx; 70 71 if (call_ctx) 72 eventfd_signal(call_ctx, 1); 73 74 return IRQ_HANDLED; 75 } 76 77 static irqreturn_t vhost_vdpa_config_cb(void *private) 78 { 79 struct vhost_vdpa *v = private; 80 struct eventfd_ctx *config_ctx = v->config_ctx; 81 82 if (config_ctx) 83 eventfd_signal(config_ctx, 1); 84 85 return IRQ_HANDLED; 86 } 87 88 static void vhost_vdpa_setup_vq_irq(struct vhost_vdpa *v, u16 qid) 89 { 90 struct vhost_virtqueue *vq = &v->vqs[qid]; 91 const struct vdpa_config_ops *ops = v->vdpa->config; 92 struct vdpa_device *vdpa = v->vdpa; 93 int ret, irq; 94 95 if (!ops->get_vq_irq) 96 return; 97 98 irq = ops->get_vq_irq(vdpa, qid); 99 irq_bypass_unregister_producer(&vq->call_ctx.producer); 100 if (!vq->call_ctx.ctx || irq < 0) 101 return; 102 103 vq->call_ctx.producer.token = vq->call_ctx.ctx; 104 vq->call_ctx.producer.irq = irq; 105 ret = irq_bypass_register_producer(&vq->call_ctx.producer); 106 } 107 108 static void vhost_vdpa_unsetup_vq_irq(struct vhost_vdpa *v, u16 qid) 109 { 110 struct vhost_virtqueue *vq = &v->vqs[qid]; 111 112 irq_bypass_unregister_producer(&vq->call_ctx.producer); 113 } 114 115 static void vhost_vdpa_reset(struct vhost_vdpa *v) 116 { 117 struct vdpa_device *vdpa = v->vdpa; 118 119 vdpa_reset(vdpa); 120 v->in_batch = 0; 121 } 122 123 static long vhost_vdpa_get_device_id(struct vhost_vdpa *v, u8 __user *argp) 124 { 125 struct vdpa_device *vdpa = v->vdpa; 126 const struct vdpa_config_ops *ops = vdpa->config; 127 u32 device_id; 128 129 device_id = ops->get_device_id(vdpa); 130 131 if (copy_to_user(argp, &device_id, sizeof(device_id))) 132 return -EFAULT; 133 134 return 0; 135 } 136 137 static long vhost_vdpa_get_status(struct vhost_vdpa *v, u8 __user *statusp) 138 { 139 struct vdpa_device *vdpa = v->vdpa; 140 const struct vdpa_config_ops *ops = vdpa->config; 141 u8 status; 142 143 status = ops->get_status(vdpa); 144 145 if (copy_to_user(statusp, &status, sizeof(status))) 146 return -EFAULT; 147 148 return 0; 149 } 150 151 static long vhost_vdpa_set_status(struct vhost_vdpa *v, u8 __user *statusp) 152 { 153 struct vdpa_device *vdpa = v->vdpa; 154 const struct vdpa_config_ops *ops = vdpa->config; 155 u8 status, status_old; 156 int nvqs = v->nvqs; 157 u16 i; 158 159 if (copy_from_user(&status, statusp, sizeof(status))) 160 return -EFAULT; 161 162 status_old = ops->get_status(vdpa); 163 164 /* 165 * Userspace shouldn't remove status bits unless reset the 166 * status to 0. 167 */ 168 if (status != 0 && (ops->get_status(vdpa) & ~status) != 0) 169 return -EINVAL; 170 171 ops->set_status(vdpa, status); 172 173 if ((status & VIRTIO_CONFIG_S_DRIVER_OK) && !(status_old & VIRTIO_CONFIG_S_DRIVER_OK)) 174 for (i = 0; i < nvqs; i++) 175 vhost_vdpa_setup_vq_irq(v, i); 176 177 if ((status_old & VIRTIO_CONFIG_S_DRIVER_OK) && !(status & VIRTIO_CONFIG_S_DRIVER_OK)) 178 for (i = 0; i < nvqs; i++) 179 vhost_vdpa_unsetup_vq_irq(v, i); 180 181 return 0; 182 } 183 184 static int vhost_vdpa_config_validate(struct vhost_vdpa *v, 185 struct vhost_vdpa_config *c) 186 { 187 long size = 0; 188 189 switch (v->virtio_id) { 190 case VIRTIO_ID_NET: 191 size = sizeof(struct virtio_net_config); 192 break; 193 } 194 195 if (c->len == 0) 196 return -EINVAL; 197 198 if (c->len > size - c->off) 199 return -E2BIG; 200 201 return 0; 202 } 203 204 static long vhost_vdpa_get_config(struct vhost_vdpa *v, 205 struct vhost_vdpa_config __user *c) 206 { 207 struct vdpa_device *vdpa = v->vdpa; 208 struct vhost_vdpa_config config; 209 unsigned long size = offsetof(struct vhost_vdpa_config, buf); 210 u8 *buf; 211 212 if (copy_from_user(&config, c, size)) 213 return -EFAULT; 214 if (vhost_vdpa_config_validate(v, &config)) 215 return -EINVAL; 216 buf = kvzalloc(config.len, GFP_KERNEL); 217 if (!buf) 218 return -ENOMEM; 219 220 vdpa_get_config(vdpa, config.off, buf, config.len); 221 222 if (copy_to_user(c->buf, buf, config.len)) { 223 kvfree(buf); 224 return -EFAULT; 225 } 226 227 kvfree(buf); 228 return 0; 229 } 230 231 static long vhost_vdpa_set_config(struct vhost_vdpa *v, 232 struct vhost_vdpa_config __user *c) 233 { 234 struct vdpa_device *vdpa = v->vdpa; 235 const struct vdpa_config_ops *ops = vdpa->config; 236 struct vhost_vdpa_config config; 237 unsigned long size = offsetof(struct vhost_vdpa_config, buf); 238 u8 *buf; 239 240 if (copy_from_user(&config, c, size)) 241 return -EFAULT; 242 if (vhost_vdpa_config_validate(v, &config)) 243 return -EINVAL; 244 buf = kvzalloc(config.len, GFP_KERNEL); 245 if (!buf) 246 return -ENOMEM; 247 248 if (copy_from_user(buf, c->buf, config.len)) { 249 kvfree(buf); 250 return -EFAULT; 251 } 252 253 ops->set_config(vdpa, config.off, buf, config.len); 254 255 kvfree(buf); 256 return 0; 257 } 258 259 static long vhost_vdpa_get_features(struct vhost_vdpa *v, u64 __user *featurep) 260 { 261 struct vdpa_device *vdpa = v->vdpa; 262 const struct vdpa_config_ops *ops = vdpa->config; 263 u64 features; 264 265 features = ops->get_features(vdpa); 266 267 if (copy_to_user(featurep, &features, sizeof(features))) 268 return -EFAULT; 269 270 return 0; 271 } 272 273 static long vhost_vdpa_set_features(struct vhost_vdpa *v, u64 __user *featurep) 274 { 275 struct vdpa_device *vdpa = v->vdpa; 276 const struct vdpa_config_ops *ops = vdpa->config; 277 u64 features; 278 279 /* 280 * It's not allowed to change the features after they have 281 * been negotiated. 282 */ 283 if (ops->get_status(vdpa) & VIRTIO_CONFIG_S_FEATURES_OK) 284 return -EBUSY; 285 286 if (copy_from_user(&features, featurep, sizeof(features))) 287 return -EFAULT; 288 289 if (vdpa_set_features(vdpa, features)) 290 return -EINVAL; 291 292 return 0; 293 } 294 295 static long vhost_vdpa_get_vring_num(struct vhost_vdpa *v, u16 __user *argp) 296 { 297 struct vdpa_device *vdpa = v->vdpa; 298 const struct vdpa_config_ops *ops = vdpa->config; 299 u16 num; 300 301 num = ops->get_vq_num_max(vdpa); 302 303 if (copy_to_user(argp, &num, sizeof(num))) 304 return -EFAULT; 305 306 return 0; 307 } 308 309 static void vhost_vdpa_config_put(struct vhost_vdpa *v) 310 { 311 if (v->config_ctx) 312 eventfd_ctx_put(v->config_ctx); 313 } 314 315 static long vhost_vdpa_set_config_call(struct vhost_vdpa *v, u32 __user *argp) 316 { 317 struct vdpa_callback cb; 318 int fd; 319 struct eventfd_ctx *ctx; 320 321 cb.callback = vhost_vdpa_config_cb; 322 cb.private = v->vdpa; 323 if (copy_from_user(&fd, argp, sizeof(fd))) 324 return -EFAULT; 325 326 ctx = fd == VHOST_FILE_UNBIND ? NULL : eventfd_ctx_fdget(fd); 327 swap(ctx, v->config_ctx); 328 329 if (!IS_ERR_OR_NULL(ctx)) 330 eventfd_ctx_put(ctx); 331 332 if (IS_ERR(v->config_ctx)) 333 return PTR_ERR(v->config_ctx); 334 335 v->vdpa->config->set_config_cb(v->vdpa, &cb); 336 337 return 0; 338 } 339 340 static long vhost_vdpa_vring_ioctl(struct vhost_vdpa *v, unsigned int cmd, 341 void __user *argp) 342 { 343 struct vdpa_device *vdpa = v->vdpa; 344 const struct vdpa_config_ops *ops = vdpa->config; 345 struct vdpa_vq_state vq_state; 346 struct vdpa_callback cb; 347 struct vhost_virtqueue *vq; 348 struct vhost_vring_state s; 349 u32 idx; 350 long r; 351 352 r = get_user(idx, (u32 __user *)argp); 353 if (r < 0) 354 return r; 355 356 if (idx >= v->nvqs) 357 return -ENOBUFS; 358 359 idx = array_index_nospec(idx, v->nvqs); 360 vq = &v->vqs[idx]; 361 362 switch (cmd) { 363 case VHOST_VDPA_SET_VRING_ENABLE: 364 if (copy_from_user(&s, argp, sizeof(s))) 365 return -EFAULT; 366 ops->set_vq_ready(vdpa, idx, s.num); 367 return 0; 368 case VHOST_GET_VRING_BASE: 369 r = ops->get_vq_state(v->vdpa, idx, &vq_state); 370 if (r) 371 return r; 372 373 vq->last_avail_idx = vq_state.avail_index; 374 break; 375 } 376 377 r = vhost_vring_ioctl(&v->vdev, cmd, argp); 378 if (r) 379 return r; 380 381 switch (cmd) { 382 case VHOST_SET_VRING_ADDR: 383 if (ops->set_vq_address(vdpa, idx, 384 (u64)(uintptr_t)vq->desc, 385 (u64)(uintptr_t)vq->avail, 386 (u64)(uintptr_t)vq->used)) 387 r = -EINVAL; 388 break; 389 390 case VHOST_SET_VRING_BASE: 391 vq_state.avail_index = vq->last_avail_idx; 392 if (ops->set_vq_state(vdpa, idx, &vq_state)) 393 r = -EINVAL; 394 break; 395 396 case VHOST_SET_VRING_CALL: 397 if (vq->call_ctx.ctx) { 398 cb.callback = vhost_vdpa_virtqueue_cb; 399 cb.private = vq; 400 } else { 401 cb.callback = NULL; 402 cb.private = NULL; 403 } 404 ops->set_vq_cb(vdpa, idx, &cb); 405 vhost_vdpa_setup_vq_irq(v, idx); 406 break; 407 408 case VHOST_SET_VRING_NUM: 409 ops->set_vq_num(vdpa, idx, vq->num); 410 break; 411 } 412 413 return r; 414 } 415 416 static long vhost_vdpa_unlocked_ioctl(struct file *filep, 417 unsigned int cmd, unsigned long arg) 418 { 419 struct vhost_vdpa *v = filep->private_data; 420 struct vhost_dev *d = &v->vdev; 421 void __user *argp = (void __user *)arg; 422 u64 __user *featurep = argp; 423 u64 features; 424 long r; 425 426 if (cmd == VHOST_SET_BACKEND_FEATURES) { 427 r = copy_from_user(&features, featurep, sizeof(features)); 428 if (r) 429 return r; 430 if (features & ~VHOST_VDPA_BACKEND_FEATURES) 431 return -EOPNOTSUPP; 432 vhost_set_backend_features(&v->vdev, features); 433 return 0; 434 } 435 436 mutex_lock(&d->mutex); 437 438 switch (cmd) { 439 case VHOST_VDPA_GET_DEVICE_ID: 440 r = vhost_vdpa_get_device_id(v, argp); 441 break; 442 case VHOST_VDPA_GET_STATUS: 443 r = vhost_vdpa_get_status(v, argp); 444 break; 445 case VHOST_VDPA_SET_STATUS: 446 r = vhost_vdpa_set_status(v, argp); 447 break; 448 case VHOST_VDPA_GET_CONFIG: 449 r = vhost_vdpa_get_config(v, argp); 450 break; 451 case VHOST_VDPA_SET_CONFIG: 452 r = vhost_vdpa_set_config(v, argp); 453 break; 454 case VHOST_GET_FEATURES: 455 r = vhost_vdpa_get_features(v, argp); 456 break; 457 case VHOST_SET_FEATURES: 458 r = vhost_vdpa_set_features(v, argp); 459 break; 460 case VHOST_VDPA_GET_VRING_NUM: 461 r = vhost_vdpa_get_vring_num(v, argp); 462 break; 463 case VHOST_SET_LOG_BASE: 464 case VHOST_SET_LOG_FD: 465 r = -ENOIOCTLCMD; 466 break; 467 case VHOST_VDPA_SET_CONFIG_CALL: 468 r = vhost_vdpa_set_config_call(v, argp); 469 break; 470 case VHOST_GET_BACKEND_FEATURES: 471 features = VHOST_VDPA_BACKEND_FEATURES; 472 r = copy_to_user(featurep, &features, sizeof(features)); 473 break; 474 default: 475 r = vhost_dev_ioctl(&v->vdev, cmd, argp); 476 if (r == -ENOIOCTLCMD) 477 r = vhost_vdpa_vring_ioctl(v, cmd, argp); 478 break; 479 } 480 481 mutex_unlock(&d->mutex); 482 return r; 483 } 484 485 static void vhost_vdpa_iotlb_unmap(struct vhost_vdpa *v, u64 start, u64 last) 486 { 487 struct vhost_dev *dev = &v->vdev; 488 struct vhost_iotlb *iotlb = dev->iotlb; 489 struct vhost_iotlb_map *map; 490 struct page *page; 491 unsigned long pfn, pinned; 492 493 while ((map = vhost_iotlb_itree_first(iotlb, start, last)) != NULL) { 494 pinned = map->size >> PAGE_SHIFT; 495 for (pfn = map->addr >> PAGE_SHIFT; 496 pinned > 0; pfn++, pinned--) { 497 page = pfn_to_page(pfn); 498 if (map->perm & VHOST_ACCESS_WO) 499 set_page_dirty_lock(page); 500 unpin_user_page(page); 501 } 502 atomic64_sub(map->size >> PAGE_SHIFT, &dev->mm->pinned_vm); 503 vhost_iotlb_map_free(iotlb, map); 504 } 505 } 506 507 static void vhost_vdpa_iotlb_free(struct vhost_vdpa *v) 508 { 509 struct vhost_dev *dev = &v->vdev; 510 511 vhost_vdpa_iotlb_unmap(v, 0ULL, 0ULL - 1); 512 kfree(dev->iotlb); 513 dev->iotlb = NULL; 514 } 515 516 static int perm_to_iommu_flags(u32 perm) 517 { 518 int flags = 0; 519 520 switch (perm) { 521 case VHOST_ACCESS_WO: 522 flags |= IOMMU_WRITE; 523 break; 524 case VHOST_ACCESS_RO: 525 flags |= IOMMU_READ; 526 break; 527 case VHOST_ACCESS_RW: 528 flags |= (IOMMU_WRITE | IOMMU_READ); 529 break; 530 default: 531 WARN(1, "invalidate vhost IOTLB permission\n"); 532 break; 533 } 534 535 return flags | IOMMU_CACHE; 536 } 537 538 static int vhost_vdpa_map(struct vhost_vdpa *v, 539 u64 iova, u64 size, u64 pa, u32 perm) 540 { 541 struct vhost_dev *dev = &v->vdev; 542 struct vdpa_device *vdpa = v->vdpa; 543 const struct vdpa_config_ops *ops = vdpa->config; 544 int r = 0; 545 546 r = vhost_iotlb_add_range(dev->iotlb, iova, iova + size - 1, 547 pa, perm); 548 if (r) 549 return r; 550 551 if (ops->dma_map) { 552 r = ops->dma_map(vdpa, iova, size, pa, perm); 553 } else if (ops->set_map) { 554 if (!v->in_batch) 555 r = ops->set_map(vdpa, dev->iotlb); 556 } else { 557 r = iommu_map(v->domain, iova, pa, size, 558 perm_to_iommu_flags(perm)); 559 } 560 561 if (r) 562 vhost_iotlb_del_range(dev->iotlb, iova, iova + size - 1); 563 564 return r; 565 } 566 567 static void vhost_vdpa_unmap(struct vhost_vdpa *v, u64 iova, u64 size) 568 { 569 struct vhost_dev *dev = &v->vdev; 570 struct vdpa_device *vdpa = v->vdpa; 571 const struct vdpa_config_ops *ops = vdpa->config; 572 573 vhost_vdpa_iotlb_unmap(v, iova, iova + size - 1); 574 575 if (ops->dma_map) { 576 ops->dma_unmap(vdpa, iova, size); 577 } else if (ops->set_map) { 578 if (!v->in_batch) 579 ops->set_map(vdpa, dev->iotlb); 580 } else { 581 iommu_unmap(v->domain, iova, size); 582 } 583 } 584 585 static int vhost_vdpa_process_iotlb_update(struct vhost_vdpa *v, 586 struct vhost_iotlb_msg *msg) 587 { 588 struct vhost_dev *dev = &v->vdev; 589 struct vhost_iotlb *iotlb = dev->iotlb; 590 struct page **page_list; 591 struct vm_area_struct **vmas; 592 unsigned int gup_flags = FOLL_LONGTERM; 593 unsigned long map_pfn, last_pfn = 0; 594 unsigned long npages, lock_limit; 595 unsigned long i, nmap = 0; 596 u64 iova = msg->iova; 597 long pinned; 598 int ret = 0; 599 600 if (vhost_iotlb_itree_first(iotlb, msg->iova, 601 msg->iova + msg->size - 1)) 602 return -EEXIST; 603 604 if (msg->perm & VHOST_ACCESS_WO) 605 gup_flags |= FOLL_WRITE; 606 607 npages = PAGE_ALIGN(msg->size + (iova & ~PAGE_MASK)) >> PAGE_SHIFT; 608 if (!npages) 609 return -EINVAL; 610 611 page_list = kvmalloc_array(npages, sizeof(struct page *), GFP_KERNEL); 612 vmas = kvmalloc_array(npages, sizeof(struct vm_area_struct *), 613 GFP_KERNEL); 614 if (!page_list || !vmas) { 615 ret = -ENOMEM; 616 goto free; 617 } 618 619 mmap_read_lock(dev->mm); 620 621 lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; 622 if (npages + atomic64_read(&dev->mm->pinned_vm) > lock_limit) { 623 ret = -ENOMEM; 624 goto unlock; 625 } 626 627 pinned = pin_user_pages(msg->uaddr & PAGE_MASK, npages, gup_flags, 628 page_list, vmas); 629 if (npages != pinned) { 630 if (pinned < 0) { 631 ret = pinned; 632 } else { 633 unpin_user_pages(page_list, pinned); 634 ret = -ENOMEM; 635 } 636 goto unlock; 637 } 638 639 iova &= PAGE_MASK; 640 map_pfn = page_to_pfn(page_list[0]); 641 642 /* One more iteration to avoid extra vdpa_map() call out of loop. */ 643 for (i = 0; i <= npages; i++) { 644 unsigned long this_pfn; 645 u64 csize; 646 647 /* The last chunk may have no valid PFN next to it */ 648 this_pfn = i < npages ? page_to_pfn(page_list[i]) : -1UL; 649 650 if (last_pfn && (this_pfn == -1UL || 651 this_pfn != last_pfn + 1)) { 652 /* Pin a contiguous chunk of memory */ 653 csize = last_pfn - map_pfn + 1; 654 ret = vhost_vdpa_map(v, iova, csize << PAGE_SHIFT, 655 map_pfn << PAGE_SHIFT, 656 msg->perm); 657 if (ret) { 658 /* 659 * Unpin the rest chunks of memory on the 660 * flight with no corresponding vdpa_map() 661 * calls having been made yet. On the other 662 * hand, vdpa_unmap() in the failure path 663 * is in charge of accounting the number of 664 * pinned pages for its own. 665 * This asymmetrical pattern of accounting 666 * is for efficiency to pin all pages at 667 * once, while there is no other callsite 668 * of vdpa_map() than here above. 669 */ 670 unpin_user_pages(&page_list[nmap], 671 npages - nmap); 672 goto out; 673 } 674 atomic64_add(csize, &dev->mm->pinned_vm); 675 nmap += csize; 676 iova += csize << PAGE_SHIFT; 677 map_pfn = this_pfn; 678 } 679 last_pfn = this_pfn; 680 } 681 682 WARN_ON(nmap != npages); 683 out: 684 if (ret) 685 vhost_vdpa_unmap(v, msg->iova, msg->size); 686 unlock: 687 mmap_read_unlock(dev->mm); 688 free: 689 kvfree(vmas); 690 kvfree(page_list); 691 return ret; 692 } 693 694 static int vhost_vdpa_process_iotlb_msg(struct vhost_dev *dev, 695 struct vhost_iotlb_msg *msg) 696 { 697 struct vhost_vdpa *v = container_of(dev, struct vhost_vdpa, vdev); 698 struct vdpa_device *vdpa = v->vdpa; 699 const struct vdpa_config_ops *ops = vdpa->config; 700 int r = 0; 701 702 r = vhost_dev_check_owner(dev); 703 if (r) 704 return r; 705 706 switch (msg->type) { 707 case VHOST_IOTLB_UPDATE: 708 r = vhost_vdpa_process_iotlb_update(v, msg); 709 break; 710 case VHOST_IOTLB_INVALIDATE: 711 vhost_vdpa_unmap(v, msg->iova, msg->size); 712 break; 713 case VHOST_IOTLB_BATCH_BEGIN: 714 v->in_batch = true; 715 break; 716 case VHOST_IOTLB_BATCH_END: 717 if (v->in_batch && ops->set_map) 718 ops->set_map(vdpa, dev->iotlb); 719 v->in_batch = false; 720 break; 721 default: 722 r = -EINVAL; 723 break; 724 } 725 726 return r; 727 } 728 729 static ssize_t vhost_vdpa_chr_write_iter(struct kiocb *iocb, 730 struct iov_iter *from) 731 { 732 struct file *file = iocb->ki_filp; 733 struct vhost_vdpa *v = file->private_data; 734 struct vhost_dev *dev = &v->vdev; 735 736 return vhost_chr_write_iter(dev, from); 737 } 738 739 static int vhost_vdpa_alloc_domain(struct vhost_vdpa *v) 740 { 741 struct vdpa_device *vdpa = v->vdpa; 742 const struct vdpa_config_ops *ops = vdpa->config; 743 struct device *dma_dev = vdpa_get_dma_dev(vdpa); 744 struct bus_type *bus; 745 int ret; 746 747 /* Device want to do DMA by itself */ 748 if (ops->set_map || ops->dma_map) 749 return 0; 750 751 bus = dma_dev->bus; 752 if (!bus) 753 return -EFAULT; 754 755 if (!iommu_capable(bus, IOMMU_CAP_CACHE_COHERENCY)) 756 return -ENOTSUPP; 757 758 v->domain = iommu_domain_alloc(bus); 759 if (!v->domain) 760 return -EIO; 761 762 ret = iommu_attach_device(v->domain, dma_dev); 763 if (ret) 764 goto err_attach; 765 766 return 0; 767 768 err_attach: 769 iommu_domain_free(v->domain); 770 return ret; 771 } 772 773 static void vhost_vdpa_free_domain(struct vhost_vdpa *v) 774 { 775 struct vdpa_device *vdpa = v->vdpa; 776 struct device *dma_dev = vdpa_get_dma_dev(vdpa); 777 778 if (v->domain) { 779 iommu_detach_device(v->domain, dma_dev); 780 iommu_domain_free(v->domain); 781 } 782 783 v->domain = NULL; 784 } 785 786 static int vhost_vdpa_open(struct inode *inode, struct file *filep) 787 { 788 struct vhost_vdpa *v; 789 struct vhost_dev *dev; 790 struct vhost_virtqueue **vqs; 791 int nvqs, i, r, opened; 792 793 v = container_of(inode->i_cdev, struct vhost_vdpa, cdev); 794 795 opened = atomic_cmpxchg(&v->opened, 0, 1); 796 if (opened) 797 return -EBUSY; 798 799 nvqs = v->nvqs; 800 vhost_vdpa_reset(v); 801 802 vqs = kmalloc_array(nvqs, sizeof(*vqs), GFP_KERNEL); 803 if (!vqs) { 804 r = -ENOMEM; 805 goto err; 806 } 807 808 dev = &v->vdev; 809 for (i = 0; i < nvqs; i++) { 810 vqs[i] = &v->vqs[i]; 811 vqs[i]->handle_kick = handle_vq_kick; 812 } 813 vhost_dev_init(dev, vqs, nvqs, 0, 0, 0, false, 814 vhost_vdpa_process_iotlb_msg); 815 816 dev->iotlb = vhost_iotlb_alloc(0, 0); 817 if (!dev->iotlb) { 818 r = -ENOMEM; 819 goto err_init_iotlb; 820 } 821 822 r = vhost_vdpa_alloc_domain(v); 823 if (r) 824 goto err_init_iotlb; 825 826 filep->private_data = v; 827 828 return 0; 829 830 err_init_iotlb: 831 vhost_dev_cleanup(&v->vdev); 832 kfree(vqs); 833 err: 834 atomic_dec(&v->opened); 835 return r; 836 } 837 838 static void vhost_vdpa_clean_irq(struct vhost_vdpa *v) 839 { 840 struct vhost_virtqueue *vq; 841 int i; 842 843 for (i = 0; i < v->nvqs; i++) { 844 vq = &v->vqs[i]; 845 if (vq->call_ctx.producer.irq) 846 irq_bypass_unregister_producer(&vq->call_ctx.producer); 847 } 848 } 849 850 static int vhost_vdpa_release(struct inode *inode, struct file *filep) 851 { 852 struct vhost_vdpa *v = filep->private_data; 853 struct vhost_dev *d = &v->vdev; 854 855 mutex_lock(&d->mutex); 856 filep->private_data = NULL; 857 vhost_vdpa_reset(v); 858 vhost_dev_stop(&v->vdev); 859 vhost_vdpa_iotlb_free(v); 860 vhost_vdpa_free_domain(v); 861 vhost_vdpa_config_put(v); 862 vhost_vdpa_clean_irq(v); 863 vhost_dev_cleanup(&v->vdev); 864 kfree(v->vdev.vqs); 865 mutex_unlock(&d->mutex); 866 867 atomic_dec(&v->opened); 868 complete(&v->completion); 869 870 return 0; 871 } 872 873 #ifdef CONFIG_MMU 874 static vm_fault_t vhost_vdpa_fault(struct vm_fault *vmf) 875 { 876 struct vhost_vdpa *v = vmf->vma->vm_file->private_data; 877 struct vdpa_device *vdpa = v->vdpa; 878 const struct vdpa_config_ops *ops = vdpa->config; 879 struct vdpa_notification_area notify; 880 struct vm_area_struct *vma = vmf->vma; 881 u16 index = vma->vm_pgoff; 882 883 notify = ops->get_vq_notification(vdpa, index); 884 885 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot); 886 if (remap_pfn_range(vma, vmf->address & PAGE_MASK, 887 notify.addr >> PAGE_SHIFT, PAGE_SIZE, 888 vma->vm_page_prot)) 889 return VM_FAULT_SIGBUS; 890 891 return VM_FAULT_NOPAGE; 892 } 893 894 static const struct vm_operations_struct vhost_vdpa_vm_ops = { 895 .fault = vhost_vdpa_fault, 896 }; 897 898 static int vhost_vdpa_mmap(struct file *file, struct vm_area_struct *vma) 899 { 900 struct vhost_vdpa *v = vma->vm_file->private_data; 901 struct vdpa_device *vdpa = v->vdpa; 902 const struct vdpa_config_ops *ops = vdpa->config; 903 struct vdpa_notification_area notify; 904 unsigned long index = vma->vm_pgoff; 905 906 if (vma->vm_end - vma->vm_start != PAGE_SIZE) 907 return -EINVAL; 908 if ((vma->vm_flags & VM_SHARED) == 0) 909 return -EINVAL; 910 if (vma->vm_flags & VM_READ) 911 return -EINVAL; 912 if (index > 65535) 913 return -EINVAL; 914 if (!ops->get_vq_notification) 915 return -ENOTSUPP; 916 917 /* To be safe and easily modelled by userspace, We only 918 * support the doorbell which sits on the page boundary and 919 * does not share the page with other registers. 920 */ 921 notify = ops->get_vq_notification(vdpa, index); 922 if (notify.addr & (PAGE_SIZE - 1)) 923 return -EINVAL; 924 if (vma->vm_end - vma->vm_start != notify.size) 925 return -ENOTSUPP; 926 927 vma->vm_ops = &vhost_vdpa_vm_ops; 928 return 0; 929 } 930 #endif /* CONFIG_MMU */ 931 932 static const struct file_operations vhost_vdpa_fops = { 933 .owner = THIS_MODULE, 934 .open = vhost_vdpa_open, 935 .release = vhost_vdpa_release, 936 .write_iter = vhost_vdpa_chr_write_iter, 937 .unlocked_ioctl = vhost_vdpa_unlocked_ioctl, 938 #ifdef CONFIG_MMU 939 .mmap = vhost_vdpa_mmap, 940 #endif /* CONFIG_MMU */ 941 .compat_ioctl = compat_ptr_ioctl, 942 }; 943 944 static void vhost_vdpa_release_dev(struct device *device) 945 { 946 struct vhost_vdpa *v = 947 container_of(device, struct vhost_vdpa, dev); 948 949 ida_simple_remove(&vhost_vdpa_ida, v->minor); 950 kfree(v->vqs); 951 kfree(v); 952 } 953 954 static int vhost_vdpa_probe(struct vdpa_device *vdpa) 955 { 956 const struct vdpa_config_ops *ops = vdpa->config; 957 struct vhost_vdpa *v; 958 int minor; 959 int r; 960 961 /* Currently, we only accept the network devices. */ 962 if (ops->get_device_id(vdpa) != VIRTIO_ID_NET) 963 return -ENOTSUPP; 964 965 v = kzalloc(sizeof(*v), GFP_KERNEL | __GFP_RETRY_MAYFAIL); 966 if (!v) 967 return -ENOMEM; 968 969 minor = ida_simple_get(&vhost_vdpa_ida, 0, 970 VHOST_VDPA_DEV_MAX, GFP_KERNEL); 971 if (minor < 0) { 972 kfree(v); 973 return minor; 974 } 975 976 atomic_set(&v->opened, 0); 977 v->minor = minor; 978 v->vdpa = vdpa; 979 v->nvqs = vdpa->nvqs; 980 v->virtio_id = ops->get_device_id(vdpa); 981 982 device_initialize(&v->dev); 983 v->dev.release = vhost_vdpa_release_dev; 984 v->dev.parent = &vdpa->dev; 985 v->dev.devt = MKDEV(MAJOR(vhost_vdpa_major), minor); 986 v->vqs = kmalloc_array(v->nvqs, sizeof(struct vhost_virtqueue), 987 GFP_KERNEL); 988 if (!v->vqs) { 989 r = -ENOMEM; 990 goto err; 991 } 992 993 r = dev_set_name(&v->dev, "vhost-vdpa-%u", minor); 994 if (r) 995 goto err; 996 997 cdev_init(&v->cdev, &vhost_vdpa_fops); 998 v->cdev.owner = THIS_MODULE; 999 1000 r = cdev_device_add(&v->cdev, &v->dev); 1001 if (r) 1002 goto err; 1003 1004 init_completion(&v->completion); 1005 vdpa_set_drvdata(vdpa, v); 1006 1007 return 0; 1008 1009 err: 1010 put_device(&v->dev); 1011 return r; 1012 } 1013 1014 static void vhost_vdpa_remove(struct vdpa_device *vdpa) 1015 { 1016 struct vhost_vdpa *v = vdpa_get_drvdata(vdpa); 1017 int opened; 1018 1019 cdev_device_del(&v->cdev, &v->dev); 1020 1021 do { 1022 opened = atomic_cmpxchg(&v->opened, 0, 1); 1023 if (!opened) 1024 break; 1025 wait_for_completion(&v->completion); 1026 } while (1); 1027 1028 put_device(&v->dev); 1029 } 1030 1031 static struct vdpa_driver vhost_vdpa_driver = { 1032 .driver = { 1033 .name = "vhost_vdpa", 1034 }, 1035 .probe = vhost_vdpa_probe, 1036 .remove = vhost_vdpa_remove, 1037 }; 1038 1039 static int __init vhost_vdpa_init(void) 1040 { 1041 int r; 1042 1043 r = alloc_chrdev_region(&vhost_vdpa_major, 0, VHOST_VDPA_DEV_MAX, 1044 "vhost-vdpa"); 1045 if (r) 1046 goto err_alloc_chrdev; 1047 1048 r = vdpa_register_driver(&vhost_vdpa_driver); 1049 if (r) 1050 goto err_vdpa_register_driver; 1051 1052 return 0; 1053 1054 err_vdpa_register_driver: 1055 unregister_chrdev_region(vhost_vdpa_major, VHOST_VDPA_DEV_MAX); 1056 err_alloc_chrdev: 1057 return r; 1058 } 1059 module_init(vhost_vdpa_init); 1060 1061 static void __exit vhost_vdpa_exit(void) 1062 { 1063 vdpa_unregister_driver(&vhost_vdpa_driver); 1064 unregister_chrdev_region(vhost_vdpa_major, VHOST_VDPA_DEV_MAX); 1065 } 1066 module_exit(vhost_vdpa_exit); 1067 1068 MODULE_VERSION("0.0.1"); 1069 MODULE_LICENSE("GPL v2"); 1070 MODULE_AUTHOR("Intel Corporation"); 1071 MODULE_DESCRIPTION("vDPA-based vhost backend for virtio"); 1072