drivers/vhost/vdpa.c

   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         struct vdpa_iova_range range;
  51 };
  52
  53 static DEFINE_IDA(vhost_vdpa_ida);
  54
  55 static dev_t vhost_vdpa_major;
  56
  57 static void handle_vq_kick(struct vhost_work *work)
  58 {
  59         struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
  60                                                   poll.work);
  61         struct vhost_vdpa *v = container_of(vq->dev, struct vhost_vdpa, vdev);
  62         const struct vdpa_config_ops *ops = v->vdpa->config;
  63
  64         ops->kick_vq(v->vdpa, vq - v->vqs);
  65 }
  66
  67 static irqreturn_t vhost_vdpa_virtqueue_cb(void *private)
  68 {
  69         struct vhost_virtqueue *vq = private;
  70         struct eventfd_ctx *call_ctx = vq->call_ctx.ctx;
  71
  72         if (call_ctx)
  73                 eventfd_signal(call_ctx, 1);
  74
  75         return IRQ_HANDLED;
  76 }
  77
  78 static irqreturn_t vhost_vdpa_config_cb(void *private)
  79 {
  80         struct vhost_vdpa *v = private;
  81         struct eventfd_ctx *config_ctx = v->config_ctx;
  82
  83         if (config_ctx)
  84                 eventfd_signal(config_ctx, 1);
  85
  86         return IRQ_HANDLED;
  87 }
  88
  89 static void vhost_vdpa_setup_vq_irq(struct vhost_vdpa *v, u16 qid)
  90 {
  91         struct vhost_virtqueue *vq = &v->vqs[qid];
  92         const struct vdpa_config_ops *ops = v->vdpa->config;
  93         struct vdpa_device *vdpa = v->vdpa;
  94         int ret, irq;
  95
  96         if (!ops->get_vq_irq)
  97                 return;
  98
  99         irq = ops->get_vq_irq(vdpa, qid);
 100         irq_bypass_unregister_producer(&vq->call_ctx.producer);
 101         if (!vq->call_ctx.ctx || irq < 0)
 102                 return;
 103
 104         vq->call_ctx.producer.token = vq->call_ctx.ctx;
 105         vq->call_ctx.producer.irq = irq;
 106         ret = irq_bypass_register_producer(&vq->call_ctx.producer);
 107         if (unlikely(ret))
 108                 dev_info(&v->dev, "vq %u, irq bypass producer (token %p) registration fails, ret =  %d\n",
 109                          qid, vq->call_ctx.producer.token, ret);
 110 }
 111
 112 static void vhost_vdpa_unsetup_vq_irq(struct vhost_vdpa *v, u16 qid)
 113 {
 114         struct vhost_virtqueue *vq = &v->vqs[qid];
 115
 116         irq_bypass_unregister_producer(&vq->call_ctx.producer);
 117 }
 118
 119 static void vhost_vdpa_reset(struct vhost_vdpa *v)
 120 {
 121         struct vdpa_device *vdpa = v->vdpa;
 122
 123         vdpa_reset(vdpa);
 124         v->in_batch = 0;
 125 }
 126
 127 static long vhost_vdpa_get_device_id(struct vhost_vdpa *v, u8 __user *argp)
 128 {
 129         struct vdpa_device *vdpa = v->vdpa;
 130         const struct vdpa_config_ops *ops = vdpa->config;
 131         u32 device_id;
 132
 133         device_id = ops->get_device_id(vdpa);
 134
 135         if (copy_to_user(argp, &device_id, sizeof(device_id)))
 136                 return -EFAULT;
 137
 138         return 0;
 139 }
 140
 141 static long vhost_vdpa_get_status(struct vhost_vdpa *v, u8 __user *statusp)
 142 {
 143         struct vdpa_device *vdpa = v->vdpa;
 144         const struct vdpa_config_ops *ops = vdpa->config;
 145         u8 status;
 146
 147         status = ops->get_status(vdpa);
 148
 149         if (copy_to_user(statusp, &status, sizeof(status)))
 150                 return -EFAULT;
 151
 152         return 0;
 153 }
 154
 155 static long vhost_vdpa_set_status(struct vhost_vdpa *v, u8 __user *statusp)
 156 {
 157         struct vdpa_device *vdpa = v->vdpa;
 158         const struct vdpa_config_ops *ops = vdpa->config;
 159         u8 status, status_old;
 160         int nvqs = v->nvqs;
 161         u16 i;
 162
 163         if (copy_from_user(&status, statusp, sizeof(status)))
 164                 return -EFAULT;
 165
 166         status_old = ops->get_status(vdpa);
 167
 168         /*
 169          * Userspace shouldn't remove status bits unless reset the
 170          * status to 0.
 171          */
 172         if (status != 0 && (ops->get_status(vdpa) & ~status) != 0)
 173                 return -EINVAL;
 174
 175         ops->set_status(vdpa, status);
 176
 177         if ((status & VIRTIO_CONFIG_S_DRIVER_OK) && !(status_old & VIRTIO_CONFIG_S_DRIVER_OK))
 178                 for (i = 0; i < nvqs; i++)
 179                         vhost_vdpa_setup_vq_irq(v, i);
 180
 181         if ((status_old & VIRTIO_CONFIG_S_DRIVER_OK) && !(status & VIRTIO_CONFIG_S_DRIVER_OK))
 182                 for (i = 0; i < nvqs; i++)
 183                         vhost_vdpa_unsetup_vq_irq(v, i);
 184
 185         return 0;
 186 }
 187
 188 static int vhost_vdpa_config_validate(struct vhost_vdpa *v,
 189                                       struct vhost_vdpa_config *c)
 190 {
 191         long size = 0;
 192
 193         switch (v->virtio_id) {
 194         case VIRTIO_ID_NET:
 195                 size = sizeof(struct virtio_net_config);
 196                 break;
 197         }
 198
 199         if (c->len == 0)
 200                 return -EINVAL;
 201
 202         if (c->len > size - c->off)
 203                 return -E2BIG;
 204
 205         return 0;
 206 }
 207
 208 static long vhost_vdpa_get_config(struct vhost_vdpa *v,
 209                                   struct vhost_vdpa_config __user *c)
 210 {
 211         struct vdpa_device *vdpa = v->vdpa;
 212         struct vhost_vdpa_config config;
 213         unsigned long size = offsetof(struct vhost_vdpa_config, buf);
 214         u8 *buf;
 215
 216         if (copy_from_user(&config, c, size))
 217                 return -EFAULT;
 218         if (vhost_vdpa_config_validate(v, &config))
 219                 return -EINVAL;
 220         buf = kvzalloc(config.len, GFP_KERNEL);
 221         if (!buf)
 222                 return -ENOMEM;
 223
 224         vdpa_get_config(vdpa, config.off, buf, config.len);
 225
 226         if (copy_to_user(c->buf, buf, config.len)) {
 227                 kvfree(buf);
 228                 return -EFAULT;
 229         }
 230
 231         kvfree(buf);
 232         return 0;
 233 }
 234
 235 static long vhost_vdpa_set_config(struct vhost_vdpa *v,
 236                                   struct vhost_vdpa_config __user *c)
 237 {
 238         struct vdpa_device *vdpa = v->vdpa;
 239         const struct vdpa_config_ops *ops = vdpa->config;
 240         struct vhost_vdpa_config config;
 241         unsigned long size = offsetof(struct vhost_vdpa_config, buf);
 242         u8 *buf;
 243
 244         if (copy_from_user(&config, c, size))
 245                 return -EFAULT;
 246         if (vhost_vdpa_config_validate(v, &config))
 247                 return -EINVAL;
 248
 249         buf = vmemdup_user(c->buf, config.len);
 250         if (IS_ERR(buf))
 251                 return PTR_ERR(buf);
 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_get_iova_range(struct vhost_vdpa *v, u32 __user *argp)
 341 {
 342         struct vhost_vdpa_iova_range range = {
 343                 .first = v->range.first,
 344                 .last = v->range.last,
 345         };
 346
 347         if (copy_to_user(argp, &range, sizeof(range)))
 348                 return -EFAULT;
 349         return 0;
 350 }
 351
 352 static long vhost_vdpa_vring_ioctl(struct vhost_vdpa *v, unsigned int cmd,
 353                                    void __user *argp)
 354 {
 355         struct vdpa_device *vdpa = v->vdpa;
 356         const struct vdpa_config_ops *ops = vdpa->config;
 357         struct vdpa_vq_state vq_state;
 358         struct vdpa_callback cb;
 359         struct vhost_virtqueue *vq;
 360         struct vhost_vring_state s;
 361         u32 idx;
 362         long r;
 363
 364         r = get_user(idx, (u32 __user *)argp);
 365         if (r < 0)
 366                 return r;
 367
 368         if (idx >= v->nvqs)
 369                 return -ENOBUFS;
 370
 371         idx = array_index_nospec(idx, v->nvqs);
 372         vq = &v->vqs[idx];
 373
 374         switch (cmd) {
 375         case VHOST_VDPA_SET_VRING_ENABLE:
 376                 if (copy_from_user(&s, argp, sizeof(s)))
 377                         return -EFAULT;
 378                 ops->set_vq_ready(vdpa, idx, s.num);
 379                 return 0;
 380         case VHOST_GET_VRING_BASE:
 381                 r = ops->get_vq_state(v->vdpa, idx, &vq_state);
 382                 if (r)
 383                         return r;
 384
 385                 vq->last_avail_idx = vq_state.avail_index;
 386                 break;
 387         }
 388
 389         r = vhost_vring_ioctl(&v->vdev, cmd, argp);
 390         if (r)
 391                 return r;
 392
 393         switch (cmd) {
 394         case VHOST_SET_VRING_ADDR:
 395                 if (ops->set_vq_address(vdpa, idx,
 396                                         (u64)(uintptr_t)vq->desc,
 397                                         (u64)(uintptr_t)vq->avail,
 398                                         (u64)(uintptr_t)vq->used))
 399                         r = -EINVAL;
 400                 break;
 401
 402         case VHOST_SET_VRING_BASE:
 403                 vq_state.avail_index = vq->last_avail_idx;
 404                 if (ops->set_vq_state(vdpa, idx, &vq_state))
 405                         r = -EINVAL;
 406                 break;
 407
 408         case VHOST_SET_VRING_CALL:
 409                 if (vq->call_ctx.ctx) {
 410                         cb.callback = vhost_vdpa_virtqueue_cb;
 411                         cb.private = vq;
 412                 } else {
 413                         cb.callback = NULL;
 414                         cb.private = NULL;
 415                 }
 416                 ops->set_vq_cb(vdpa, idx, &cb);
 417                 vhost_vdpa_setup_vq_irq(v, idx);
 418                 break;
 419
 420         case VHOST_SET_VRING_NUM:
 421                 ops->set_vq_num(vdpa, idx, vq->num);
 422                 break;
 423         }
 424
 425         return r;
 426 }
 427
 428 static long vhost_vdpa_unlocked_ioctl(struct file *filep,
 429                                       unsigned int cmd, unsigned long arg)
 430 {
 431         struct vhost_vdpa *v = filep->private_data;
 432         struct vhost_dev *d = &v->vdev;
 433         void __user *argp = (void __user *)arg;
 434         u64 __user *featurep = argp;
 435         u64 features;
 436         long r = 0;
 437
 438         if (cmd == VHOST_SET_BACKEND_FEATURES) {
 439                 if (copy_from_user(&features, featurep, sizeof(features)))
 440                         return -EFAULT;
 441                 if (features & ~VHOST_VDPA_BACKEND_FEATURES)
 442                         return -EOPNOTSUPP;
 443                 vhost_set_backend_features(&v->vdev, features);
 444                 return 0;
 445         }
 446
 447         mutex_lock(&d->mutex);
 448
 449         switch (cmd) {
 450         case VHOST_VDPA_GET_DEVICE_ID:
 451                 r = vhost_vdpa_get_device_id(v, argp);
 452                 break;
 453         case VHOST_VDPA_GET_STATUS:
 454                 r = vhost_vdpa_get_status(v, argp);
 455                 break;
 456         case VHOST_VDPA_SET_STATUS:
 457                 r = vhost_vdpa_set_status(v, argp);
 458                 break;
 459         case VHOST_VDPA_GET_CONFIG:
 460                 r = vhost_vdpa_get_config(v, argp);
 461                 break;
 462         case VHOST_VDPA_SET_CONFIG:
 463                 r = vhost_vdpa_set_config(v, argp);
 464                 break;
 465         case VHOST_GET_FEATURES:
 466                 r = vhost_vdpa_get_features(v, argp);
 467                 break;
 468         case VHOST_SET_FEATURES:
 469                 r = vhost_vdpa_set_features(v, argp);
 470                 break;
 471         case VHOST_VDPA_GET_VRING_NUM:
 472                 r = vhost_vdpa_get_vring_num(v, argp);
 473                 break;
 474         case VHOST_SET_LOG_BASE:
 475         case VHOST_SET_LOG_FD:
 476                 r = -ENOIOCTLCMD;
 477                 break;
 478         case VHOST_VDPA_SET_CONFIG_CALL:
 479                 r = vhost_vdpa_set_config_call(v, argp);
 480                 break;
 481         case VHOST_GET_BACKEND_FEATURES:
 482                 features = VHOST_VDPA_BACKEND_FEATURES;
 483                 if (copy_to_user(featurep, &features, sizeof(features)))
 484                         r = -EFAULT;
 485                 break;
 486         case VHOST_VDPA_GET_IOVA_RANGE:
 487                 r = vhost_vdpa_get_iova_range(v, argp);
 488                 break;
 489         default:
 490                 r = vhost_dev_ioctl(&v->vdev, cmd, argp);
 491                 if (r == -ENOIOCTLCMD)
 492                         r = vhost_vdpa_vring_ioctl(v, cmd, argp);
 493                 break;
 494         }
 495
 496         mutex_unlock(&d->mutex);
 497         return r;
 498 }
 499
 500 static void vhost_vdpa_iotlb_unmap(struct vhost_vdpa *v, u64 start, u64 last)
 501 {
 502         struct vhost_dev *dev = &v->vdev;
 503         struct vhost_iotlb *iotlb = dev->iotlb;
 504         struct vhost_iotlb_map *map;
 505         struct page *page;
 506         unsigned long pfn, pinned;
 507
 508         while ((map = vhost_iotlb_itree_first(iotlb, start, last)) != NULL) {
 509                 pinned = map->size >> PAGE_SHIFT;
 510                 for (pfn = map->addr >> PAGE_SHIFT;
 511                      pinned > 0; pfn++, pinned--) {
 512                         page = pfn_to_page(pfn);
 513                         if (map->perm & VHOST_ACCESS_WO)
 514                                 set_page_dirty_lock(page);
 515                         unpin_user_page(page);
 516                 }
 517                 atomic64_sub(map->size >> PAGE_SHIFT, &dev->mm->pinned_vm);
 518                 vhost_iotlb_map_free(iotlb, map);
 519         }
 520 }
 521
 522 static void vhost_vdpa_iotlb_free(struct vhost_vdpa *v)
 523 {
 524         struct vhost_dev *dev = &v->vdev;
 525
 526         vhost_vdpa_iotlb_unmap(v, 0ULL, 0ULL - 1);
 527         kfree(dev->iotlb);
 528         dev->iotlb = NULL;
 529 }
 530
 531 static int perm_to_iommu_flags(u32 perm)
 532 {
 533         int flags = 0;
 534
 535         switch (perm) {
 536         case VHOST_ACCESS_WO:
 537                 flags |= IOMMU_WRITE;
 538                 break;
 539         case VHOST_ACCESS_RO:
 540                 flags |= IOMMU_READ;
 541                 break;
 542         case VHOST_ACCESS_RW:
 543                 flags |= (IOMMU_WRITE | IOMMU_READ);
 544                 break;
 545         default:
 546                 WARN(1, "invalidate vhost IOTLB permission\n");
 547                 break;
 548         }
 549
 550         return flags | IOMMU_CACHE;
 551 }
 552
 553 static int vhost_vdpa_map(struct vhost_vdpa *v,
 554                           u64 iova, u64 size, u64 pa, u32 perm)
 555 {
 556         struct vhost_dev *dev = &v->vdev;
 557         struct vdpa_device *vdpa = v->vdpa;
 558         const struct vdpa_config_ops *ops = vdpa->config;
 559         int r = 0;
 560
 561         r = vhost_iotlb_add_range(dev->iotlb, iova, iova + size - 1,
 562                                   pa, perm);
 563         if (r)
 564                 return r;
 565
 566         if (ops->dma_map) {
 567                 r = ops->dma_map(vdpa, iova, size, pa, perm);
 568         } else if (ops->set_map) {
 569                 if (!v->in_batch)
 570                         r = ops->set_map(vdpa, dev->iotlb);
 571         } else {
 572                 r = iommu_map(v->domain, iova, pa, size,
 573                               perm_to_iommu_flags(perm));
 574         }
 575
 576         if (r)
 577                 vhost_iotlb_del_range(dev->iotlb, iova, iova + size - 1);
 578         else
 579                 atomic64_add(size >> PAGE_SHIFT, &dev->mm->pinned_vm);
 580
 581         return r;
 582 }
 583
 584 static void vhost_vdpa_unmap(struct vhost_vdpa *v, u64 iova, u64 size)
 585 {
 586         struct vhost_dev *dev = &v->vdev;
 587         struct vdpa_device *vdpa = v->vdpa;
 588         const struct vdpa_config_ops *ops = vdpa->config;
 589
 590         vhost_vdpa_iotlb_unmap(v, iova, iova + size - 1);
 591
 592         if (ops->dma_map) {
 593                 ops->dma_unmap(vdpa, iova, size);
 594         } else if (ops->set_map) {
 595                 if (!v->in_batch)
 596                         ops->set_map(vdpa, dev->iotlb);
 597         } else {
 598                 iommu_unmap(v->domain, iova, size);
 599         }
 600 }
 601
 602 static int vhost_vdpa_process_iotlb_update(struct vhost_vdpa *v,
 603                                            struct vhost_iotlb_msg *msg)
 604 {
 605         struct vhost_dev *dev = &v->vdev;
 606         struct vhost_iotlb *iotlb = dev->iotlb;
 607         struct page **page_list;
 608         unsigned long list_size = PAGE_SIZE / sizeof(struct page *);
 609         unsigned int gup_flags = FOLL_LONGTERM;
 610         unsigned long npages, cur_base, map_pfn, last_pfn = 0;
 611         unsigned long lock_limit, sz2pin, nchunks, i;
 612         u64 iova = msg->iova;
 613         long pinned;
 614         int ret = 0;
 615
 616         if (msg->iova < v->range.first ||
 617             msg->iova + msg->size - 1 > v->range.last)
 618                 return -EINVAL;
 619
 620         if (vhost_iotlb_itree_first(iotlb, msg->iova,
 621                                     msg->iova + msg->size - 1))
 622                 return -EEXIST;
 623
 624         /* Limit the use of memory for bookkeeping */
 625         page_list = (struct page **) __get_free_page(GFP_KERNEL);
 626         if (!page_list)
 627                 return -ENOMEM;
 628
 629         if (msg->perm & VHOST_ACCESS_WO)
 630                 gup_flags |= FOLL_WRITE;
 631
 632         npages = PAGE_ALIGN(msg->size + (iova & ~PAGE_MASK)) >> PAGE_SHIFT;
 633         if (!npages) {
 634                 ret = -EINVAL;
 635                 goto free;
 636         }
 637
 638         mmap_read_lock(dev->mm);
 639
 640         lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
 641         if (npages + atomic64_read(&dev->mm->pinned_vm) > lock_limit) {
 642                 ret = -ENOMEM;
 643                 goto unlock;
 644         }
 645
 646         cur_base = msg->uaddr & PAGE_MASK;
 647         iova &= PAGE_MASK;
 648         nchunks = 0;
 649
 650         while (npages) {
 651                 sz2pin = min_t(unsigned long, npages, list_size);
 652                 pinned = pin_user_pages(cur_base, sz2pin,
 653                                         gup_flags, page_list, NULL);
 654                 if (sz2pin != pinned) {
 655                         if (pinned < 0) {
 656                                 ret = pinned;
 657                         } else {
 658                                 unpin_user_pages(page_list, pinned);
 659                                 ret = -ENOMEM;
 660                         }
 661                         goto out;
 662                 }
 663                 nchunks++;
 664
 665                 if (!last_pfn)
 666                         map_pfn = page_to_pfn(page_list[0]);
 667
 668                 for (i = 0; i < pinned; i++) {
 669                         unsigned long this_pfn = page_to_pfn(page_list[i]);
 670                         u64 csize;
 671
 672                         if (last_pfn && (this_pfn != last_pfn + 1)) {
 673                                 /* Pin a contiguous chunk of memory */
 674                                 csize = (last_pfn - map_pfn + 1) << PAGE_SHIFT;
 675                                 ret = vhost_vdpa_map(v, iova, csize,
 676                                                      map_pfn << PAGE_SHIFT,
 677                                                      msg->perm);
 678                                 if (ret) {
 679                                         /*
 680                                          * Unpin the pages that are left unmapped
 681                                          * from this point on in the current
 682                                          * page_list. The remaining outstanding
 683                                          * ones which may stride across several
 684                                          * chunks will be covered in the common
 685                                          * error path subsequently.
 686                                          */
 687                                         unpin_user_pages(&page_list[i],
 688                                                          pinned - i);
 689                                         goto out;
 690                                 }
 691
 692                                 map_pfn = this_pfn;
 693                                 iova += csize;
 694                                 nchunks = 0;
 695                         }
 696
 697                         last_pfn = this_pfn;
 698                 }
 699
 700                 cur_base += pinned << PAGE_SHIFT;
 701                 npages -= pinned;
 702         }
 703
 704         /* Pin the rest chunk */
 705         ret = vhost_vdpa_map(v, iova, (last_pfn - map_pfn + 1) << PAGE_SHIFT,
 706                              map_pfn << PAGE_SHIFT, msg->perm);
 707 out:
 708         if (ret) {
 709                 if (nchunks) {
 710                         unsigned long pfn;
 711
 712                         /*
 713                          * Unpin the outstanding pages which are yet to be
 714                          * mapped but haven't due to vdpa_map() or
 715                          * pin_user_pages() failure.
 716                          *
 717                          * Mapped pages are accounted in vdpa_map(), hence
 718                          * the corresponding unpinning will be handled by
 719                          * vdpa_unmap().
 720                          */
 721                         WARN_ON(!last_pfn);
 722                         for (pfn = map_pfn; pfn <= last_pfn; pfn++)
 723                                 unpin_user_page(pfn_to_page(pfn));
 724                 }
 725                 vhost_vdpa_unmap(v, msg->iova, msg->size);
 726         }
 727 unlock:
 728         mmap_read_unlock(dev->mm);
 729 free:
 730         free_page((unsigned long)page_list);
 731         return ret;
 732 }
 733
 734 static int vhost_vdpa_process_iotlb_msg(struct vhost_dev *dev,
 735                                         struct vhost_iotlb_msg *msg)
 736 {
 737         struct vhost_vdpa *v = container_of(dev, struct vhost_vdpa, vdev);
 738         struct vdpa_device *vdpa = v->vdpa;
 739         const struct vdpa_config_ops *ops = vdpa->config;
 740         int r = 0;
 741
 742         r = vhost_dev_check_owner(dev);
 743         if (r)
 744                 return r;
 745
 746         switch (msg->type) {
 747         case VHOST_IOTLB_UPDATE:
 748                 r = vhost_vdpa_process_iotlb_update(v, msg);
 749                 break;
 750         case VHOST_IOTLB_INVALIDATE:
 751                 vhost_vdpa_unmap(v, msg->iova, msg->size);
 752                 break;
 753         case VHOST_IOTLB_BATCH_BEGIN:
 754                 v->in_batch = true;
 755                 break;
 756         case VHOST_IOTLB_BATCH_END:
 757                 if (v->in_batch && ops->set_map)
 758                         ops->set_map(vdpa, dev->iotlb);
 759                 v->in_batch = false;
 760                 break;
 761         default:
 762                 r = -EINVAL;
 763                 break;
 764         }
 765
 766         return r;
 767 }
 768
 769 static ssize_t vhost_vdpa_chr_write_iter(struct kiocb *iocb,
 770                                          struct iov_iter *from)
 771 {
 772         struct file *file = iocb->ki_filp;
 773         struct vhost_vdpa *v = file->private_data;
 774         struct vhost_dev *dev = &v->vdev;
 775
 776         return vhost_chr_write_iter(dev, from);
 777 }
 778
 779 static int vhost_vdpa_alloc_domain(struct vhost_vdpa *v)
 780 {
 781         struct vdpa_device *vdpa = v->vdpa;
 782         const struct vdpa_config_ops *ops = vdpa->config;
 783         struct device *dma_dev = vdpa_get_dma_dev(vdpa);
 784         struct bus_type *bus;
 785         int ret;
 786
 787         /* Device want to do DMA by itself */
 788         if (ops->set_map || ops->dma_map)
 789                 return 0;
 790
 791         bus = dma_dev->bus;
 792         if (!bus)
 793                 return -EFAULT;
 794
 795         if (!iommu_capable(bus, IOMMU_CAP_CACHE_COHERENCY))
 796                 return -ENOTSUPP;
 797
 798         v->domain = iommu_domain_alloc(bus);
 799         if (!v->domain)
 800                 return -EIO;
 801
 802         ret = iommu_attach_device(v->domain, dma_dev);
 803         if (ret)
 804                 goto err_attach;
 805
 806         return 0;
 807
 808 err_attach:
 809         iommu_domain_free(v->domain);
 810         return ret;
 811 }
 812
 813 static void vhost_vdpa_free_domain(struct vhost_vdpa *v)
 814 {
 815         struct vdpa_device *vdpa = v->vdpa;
 816         struct device *dma_dev = vdpa_get_dma_dev(vdpa);
 817
 818         if (v->domain) {
 819                 iommu_detach_device(v->domain, dma_dev);
 820                 iommu_domain_free(v->domain);
 821         }
 822
 823         v->domain = NULL;
 824 }
 825
 826 static void vhost_vdpa_set_iova_range(struct vhost_vdpa *v)
 827 {
 828         struct vdpa_iova_range *range = &v->range;
 829         struct iommu_domain_geometry geo;
 830         struct vdpa_device *vdpa = v->vdpa;
 831         const struct vdpa_config_ops *ops = vdpa->config;
 832
 833         if (ops->get_iova_range) {
 834                 *range = ops->get_iova_range(vdpa);
 835         } else if (v->domain &&
 836                    !iommu_domain_get_attr(v->domain,
 837                    DOMAIN_ATTR_GEOMETRY, &geo) &&
 838                    geo.force_aperture) {
 839                 range->first = geo.aperture_start;
 840                 range->last = geo.aperture_end;
 841         } else {
 842                 range->first = 0;
 843                 range->last = ULLONG_MAX;
 844         }
 845 }
 846
 847 static int vhost_vdpa_open(struct inode *inode, struct file *filep)
 848 {
 849         struct vhost_vdpa *v;
 850         struct vhost_dev *dev;
 851         struct vhost_virtqueue **vqs;
 852         int nvqs, i, r, opened;
 853
 854         v = container_of(inode->i_cdev, struct vhost_vdpa, cdev);
 855
 856         opened = atomic_cmpxchg(&v->opened, 0, 1);
 857         if (opened)
 858                 return -EBUSY;
 859
 860         nvqs = v->nvqs;
 861         vhost_vdpa_reset(v);
 862
 863         vqs = kmalloc_array(nvqs, sizeof(*vqs), GFP_KERNEL);
 864         if (!vqs) {
 865                 r = -ENOMEM;
 866                 goto err;
 867         }
 868
 869         dev = &v->vdev;
 870         for (i = 0; i < nvqs; i++) {
 871                 vqs[i] = &v->vqs[i];
 872                 vqs[i]->handle_kick = handle_vq_kick;
 873         }
 874         vhost_dev_init(dev, vqs, nvqs, 0, 0, 0, false,
 875                        vhost_vdpa_process_iotlb_msg);
 876
 877         dev->iotlb = vhost_iotlb_alloc(0, 0);
 878         if (!dev->iotlb) {
 879                 r = -ENOMEM;
 880                 goto err_init_iotlb;
 881         }
 882
 883         r = vhost_vdpa_alloc_domain(v);
 884         if (r)
 885                 goto err_init_iotlb;
 886
 887         vhost_vdpa_set_iova_range(v);
 888
 889         filep->private_data = v;
 890
 891         return 0;
 892
 893 err_init_iotlb:
 894         vhost_dev_cleanup(&v->vdev);
 895         kfree(vqs);
 896 err:
 897         atomic_dec(&v->opened);
 898         return r;
 899 }
 900
 901 static void vhost_vdpa_clean_irq(struct vhost_vdpa *v)
 902 {
 903         struct vhost_virtqueue *vq;
 904         int i;
 905
 906         for (i = 0; i < v->nvqs; i++) {
 907                 vq = &v->vqs[i];
 908                 if (vq->call_ctx.producer.irq)
 909                         irq_bypass_unregister_producer(&vq->call_ctx.producer);
 910         }
 911 }
 912
 913 static int vhost_vdpa_release(struct inode *inode, struct file *filep)
 914 {
 915         struct vhost_vdpa *v = filep->private_data;
 916         struct vhost_dev *d = &v->vdev;
 917
 918         mutex_lock(&d->mutex);
 919         filep->private_data = NULL;
 920         vhost_vdpa_reset(v);
 921         vhost_dev_stop(&v->vdev);
 922         vhost_vdpa_iotlb_free(v);
 923         vhost_vdpa_free_domain(v);
 924         vhost_vdpa_config_put(v);
 925         vhost_vdpa_clean_irq(v);
 926         vhost_dev_cleanup(&v->vdev);
 927         kfree(v->vdev.vqs);
 928         mutex_unlock(&d->mutex);
 929
 930         atomic_dec(&v->opened);
 931         complete(&v->completion);
 932
 933         return 0;
 934 }
 935
 936 #ifdef CONFIG_MMU
 937 static vm_fault_t vhost_vdpa_fault(struct vm_fault *vmf)
 938 {
 939         struct vhost_vdpa *v = vmf->vma->vm_file->private_data;
 940         struct vdpa_device *vdpa = v->vdpa;
 941         const struct vdpa_config_ops *ops = vdpa->config;
 942         struct vdpa_notification_area notify;
 943         struct vm_area_struct *vma = vmf->vma;
 944         u16 index = vma->vm_pgoff;
 945
 946         notify = ops->get_vq_notification(vdpa, index);
 947
 948         vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
 949         if (remap_pfn_range(vma, vmf->address & PAGE_MASK,
 950                             notify.addr >> PAGE_SHIFT, PAGE_SIZE,
 951                             vma->vm_page_prot))
 952                 return VM_FAULT_SIGBUS;
 953
 954         return VM_FAULT_NOPAGE;
 955 }
 956
 957 static const struct vm_operations_struct vhost_vdpa_vm_ops = {
 958         .fault = vhost_vdpa_fault,
 959 };
 960
 961 static int vhost_vdpa_mmap(struct file *file, struct vm_area_struct *vma)
 962 {
 963         struct vhost_vdpa *v = vma->vm_file->private_data;
 964         struct vdpa_device *vdpa = v->vdpa;
 965         const struct vdpa_config_ops *ops = vdpa->config;
 966         struct vdpa_notification_area notify;
 967         unsigned long index = vma->vm_pgoff;
 968
 969         if (vma->vm_end - vma->vm_start != PAGE_SIZE)
 970                 return -EINVAL;
 971         if ((vma->vm_flags & VM_SHARED) == 0)
 972                 return -EINVAL;
 973         if (vma->vm_flags & VM_READ)
 974                 return -EINVAL;
 975         if (index > 65535)
 976                 return -EINVAL;
 977         if (!ops->get_vq_notification)
 978                 return -ENOTSUPP;
 979
 980         /* To be safe and easily modelled by userspace, We only
 981          * support the doorbell which sits on the page boundary and
 982          * does not share the page with other registers.
 983          */
 984         notify = ops->get_vq_notification(vdpa, index);
 985         if (notify.addr & (PAGE_SIZE - 1))
 986                 return -EINVAL;
 987         if (vma->vm_end - vma->vm_start != notify.size)
 988                 return -ENOTSUPP;
 989
 990         vma->vm_ops = &vhost_vdpa_vm_ops;
 991         return 0;
 992 }
 993 #endif /* CONFIG_MMU */
 994
 995 static const struct file_operations vhost_vdpa_fops = {
 996         .owner          = THIS_MODULE,
 997         .open           = vhost_vdpa_open,
 998         .release        = vhost_vdpa_release,
 999         .write_iter     = vhost_vdpa_chr_write_iter,
1000         .unlocked_ioctl = vhost_vdpa_unlocked_ioctl,
1001 #ifdef CONFIG_MMU
1002         .mmap           = vhost_vdpa_mmap,
1003 #endif /* CONFIG_MMU */
1004         .compat_ioctl   = compat_ptr_ioctl,
1005 };
1006
1007 static void vhost_vdpa_release_dev(struct device *device)
1008 {
1009         struct vhost_vdpa *v =
1010                container_of(device, struct vhost_vdpa, dev);
1011
1012         ida_simple_remove(&vhost_vdpa_ida, v->minor);
1013         kfree(v->vqs);
1014         kfree(v);
1015 }
1016
1017 static int vhost_vdpa_probe(struct vdpa_device *vdpa)
1018 {
1019         const struct vdpa_config_ops *ops = vdpa->config;
1020         struct vhost_vdpa *v;
1021         int minor;
1022         int r;
1023
1024         /* Currently, we only accept the network devices. */
1025         if (ops->get_device_id(vdpa) != VIRTIO_ID_NET)
1026                 return -ENOTSUPP;
1027
1028         v = kzalloc(sizeof(*v), GFP_KERNEL | __GFP_RETRY_MAYFAIL);
1029         if (!v)
1030                 return -ENOMEM;
1031
1032         minor = ida_simple_get(&vhost_vdpa_ida, 0,
1033                                VHOST_VDPA_DEV_MAX, GFP_KERNEL);
1034         if (minor < 0) {
1035                 kfree(v);
1036                 return minor;
1037         }
1038
1039         atomic_set(&v->opened, 0);
1040         v->minor = minor;
1041         v->vdpa = vdpa;
1042         v->nvqs = vdpa->nvqs;
1043         v->virtio_id = ops->get_device_id(vdpa);
1044
1045         device_initialize(&v->dev);
1046         v->dev.release = vhost_vdpa_release_dev;
1047         v->dev.parent = &vdpa->dev;
1048         v->dev.devt = MKDEV(MAJOR(vhost_vdpa_major), minor);
1049         v->vqs = kmalloc_array(v->nvqs, sizeof(struct vhost_virtqueue),
1050                                GFP_KERNEL);
1051         if (!v->vqs) {
1052                 r = -ENOMEM;
1053                 goto err;
1054         }
1055
1056         r = dev_set_name(&v->dev, "vhost-vdpa-%u", minor);
1057         if (r)
1058                 goto err;
1059
1060         cdev_init(&v->cdev, &vhost_vdpa_fops);
1061         v->cdev.owner = THIS_MODULE;
1062
1063         r = cdev_device_add(&v->cdev, &v->dev);
1064         if (r)
1065                 goto err;
1066
1067         init_completion(&v->completion);
1068         vdpa_set_drvdata(vdpa, v);
1069
1070         return 0;
1071
1072 err:
1073         put_device(&v->dev);
1074         return r;
1075 }
1076
1077 static void vhost_vdpa_remove(struct vdpa_device *vdpa)
1078 {
1079         struct vhost_vdpa *v = vdpa_get_drvdata(vdpa);
1080         int opened;
1081
1082         cdev_device_del(&v->cdev, &v->dev);
1083
1084         do {
1085                 opened = atomic_cmpxchg(&v->opened, 0, 1);
1086                 if (!opened)
1087                         break;
1088                 wait_for_completion(&v->completion);
1089         } while (1);
1090
1091         put_device(&v->dev);
1092 }
1093
1094 static struct vdpa_driver vhost_vdpa_driver = {
1095         .driver = {
1096                 .name   = "vhost_vdpa",
1097         },
1098         .probe  = vhost_vdpa_probe,
1099         .remove = vhost_vdpa_remove,
1100 };
1101
1102 static int __init vhost_vdpa_init(void)
1103 {
1104         int r;
1105
1106         r = alloc_chrdev_region(&vhost_vdpa_major, 0, VHOST_VDPA_DEV_MAX,
1107                                 "vhost-vdpa");
1108         if (r)
1109                 goto err_alloc_chrdev;
1110
1111         r = vdpa_register_driver(&vhost_vdpa_driver);
1112         if (r)
1113                 goto err_vdpa_register_driver;
1114
1115         return 0;
1116
1117 err_vdpa_register_driver:
1118         unregister_chrdev_region(vhost_vdpa_major, VHOST_VDPA_DEV_MAX);
1119 err_alloc_chrdev:
1120         return r;
1121 }
1122 module_init(vhost_vdpa_init);
1123
1124 static void __exit vhost_vdpa_exit(void)
1125 {
1126         vdpa_unregister_driver(&vhost_vdpa_driver);
1127         unregister_chrdev_region(vhost_vdpa_major, VHOST_VDPA_DEV_MAX);
1128 }
1129 module_exit(vhost_vdpa_exit);
1130
1131 MODULE_VERSION("0.0.1");
1132 MODULE_LICENSE("GPL v2");
1133 MODULE_AUTHOR("Intel Corporation");
1134 MODULE_DESCRIPTION("vDPA-based vhost backend for virtio");