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/slab.h>
  20 #include <linux/iommu.h>
  21 #include <linux/uuid.h>
  22 #include <linux/vdpa.h>
  23 #include <linux/nospec.h>
  24 #include <linux/vhost.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 int vhost_vdpa_reset(struct vhost_vdpa *v)
 120 {
 121         struct vdpa_device *vdpa = v->vdpa;
 122
 123         v->in_batch = 0;
 124
 125         return vdpa_reset(vdpa);
 126 }
 127
 128 static long vhost_vdpa_get_device_id(struct vhost_vdpa *v, u8 __user *argp)
 129 {
 130         struct vdpa_device *vdpa = v->vdpa;
 131         const struct vdpa_config_ops *ops = vdpa->config;
 132         u32 device_id;
 133
 134         device_id = ops->get_device_id(vdpa);
 135
 136         if (copy_to_user(argp, &device_id, sizeof(device_id)))
 137                 return -EFAULT;
 138
 139         return 0;
 140 }
 141
 142 static long vhost_vdpa_get_status(struct vhost_vdpa *v, u8 __user *statusp)
 143 {
 144         struct vdpa_device *vdpa = v->vdpa;
 145         const struct vdpa_config_ops *ops = vdpa->config;
 146         u8 status;
 147
 148         status = ops->get_status(vdpa);
 149
 150         if (copy_to_user(statusp, &status, sizeof(status)))
 151                 return -EFAULT;
 152
 153         return 0;
 154 }
 155
 156 static long vhost_vdpa_set_status(struct vhost_vdpa *v, u8 __user *statusp)
 157 {
 158         struct vdpa_device *vdpa = v->vdpa;
 159         const struct vdpa_config_ops *ops = vdpa->config;
 160         u8 status, status_old;
 161         int ret, nvqs = v->nvqs;
 162         u16 i;
 163
 164         if (copy_from_user(&status, statusp, sizeof(status)))
 165                 return -EFAULT;
 166
 167         status_old = ops->get_status(vdpa);
 168
 169         /*
 170          * Userspace shouldn't remove status bits unless reset the
 171          * status to 0.
 172          */
 173         if (status != 0 && (ops->get_status(vdpa) & ~status) != 0)
 174                 return -EINVAL;
 175
 176         if (status == 0) {
 177                 ret = ops->reset(vdpa);
 178                 if (ret)
 179                         return ret;
 180         } else
 181                 ops->set_status(vdpa, status);
 182
 183         if ((status & VIRTIO_CONFIG_S_DRIVER_OK) && !(status_old & VIRTIO_CONFIG_S_DRIVER_OK))
 184                 for (i = 0; i < nvqs; i++)
 185                         vhost_vdpa_setup_vq_irq(v, i);
 186
 187         if ((status_old & VIRTIO_CONFIG_S_DRIVER_OK) && !(status & VIRTIO_CONFIG_S_DRIVER_OK))
 188                 for (i = 0; i < nvqs; i++)
 189                         vhost_vdpa_unsetup_vq_irq(v, i);
 190
 191         return 0;
 192 }
 193
 194 static int vhost_vdpa_config_validate(struct vhost_vdpa *v,
 195                                       struct vhost_vdpa_config *c)
 196 {
 197         struct vdpa_device *vdpa = v->vdpa;
 198         long size = vdpa->config->get_config_size(vdpa);
 199
 200         if (c->len == 0)
 201                 return -EINVAL;
 202
 203         if (c->len > size - c->off)
 204                 return -E2BIG;
 205
 206         return 0;
 207 }
 208
 209 static long vhost_vdpa_get_config(struct vhost_vdpa *v,
 210                                   struct vhost_vdpa_config __user *c)
 211 {
 212         struct vdpa_device *vdpa = v->vdpa;
 213         struct vhost_vdpa_config config;
 214         unsigned long size = offsetof(struct vhost_vdpa_config, buf);
 215         u8 *buf;
 216
 217         if (copy_from_user(&config, c, size))
 218                 return -EFAULT;
 219         if (vhost_vdpa_config_validate(v, &config))
 220                 return -EINVAL;
 221         buf = kvzalloc(config.len, GFP_KERNEL);
 222         if (!buf)
 223                 return -ENOMEM;
 224
 225         vdpa_get_config(vdpa, config.off, buf, config.len);
 226
 227         if (copy_to_user(c->buf, buf, config.len)) {
 228                 kvfree(buf);
 229                 return -EFAULT;
 230         }
 231
 232         kvfree(buf);
 233         return 0;
 234 }
 235
 236 static long vhost_vdpa_set_config(struct vhost_vdpa *v,
 237                                   struct vhost_vdpa_config __user *c)
 238 {
 239         struct vdpa_device *vdpa = v->vdpa;
 240         const struct vdpa_config_ops *ops = vdpa->config;
 241         struct vhost_vdpa_config config;
 242         unsigned long size = offsetof(struct vhost_vdpa_config, buf);
 243         u8 *buf;
 244
 245         if (copy_from_user(&config, c, size))
 246                 return -EFAULT;
 247         if (vhost_vdpa_config_validate(v, &config))
 248                 return -EINVAL;
 249
 250         buf = vmemdup_user(c->buf, config.len);
 251         if (IS_ERR(buf))
 252                 return PTR_ERR(buf);
 253
 254         ops->set_config(vdpa, config.off, buf, config.len);
 255
 256         kvfree(buf);
 257         return 0;
 258 }
 259
 260 static long vhost_vdpa_get_features(struct vhost_vdpa *v, u64 __user *featurep)
 261 {
 262         struct vdpa_device *vdpa = v->vdpa;
 263         const struct vdpa_config_ops *ops = vdpa->config;
 264         u64 features;
 265
 266         features = ops->get_features(vdpa);
 267
 268         if (copy_to_user(featurep, &features, sizeof(features)))
 269                 return -EFAULT;
 270
 271         return 0;
 272 }
 273
 274 static long vhost_vdpa_set_features(struct vhost_vdpa *v, u64 __user *featurep)
 275 {
 276         struct vdpa_device *vdpa = v->vdpa;
 277         const struct vdpa_config_ops *ops = vdpa->config;
 278         u64 features;
 279
 280         /*
 281          * It's not allowed to change the features after they have
 282          * been negotiated.
 283          */
 284         if (ops->get_status(vdpa) & VIRTIO_CONFIG_S_FEATURES_OK)
 285                 return -EBUSY;
 286
 287         if (copy_from_user(&features, featurep, sizeof(features)))
 288                 return -EFAULT;
 289
 290         if (vdpa_set_features(vdpa, features))
 291                 return -EINVAL;
 292
 293         return 0;
 294 }
 295
 296 static long vhost_vdpa_get_vring_num(struct vhost_vdpa *v, u16 __user *argp)
 297 {
 298         struct vdpa_device *vdpa = v->vdpa;
 299         const struct vdpa_config_ops *ops = vdpa->config;
 300         u16 num;
 301
 302         num = ops->get_vq_num_max(vdpa);
 303
 304         if (copy_to_user(argp, &num, sizeof(num)))
 305                 return -EFAULT;
 306
 307         return 0;
 308 }
 309
 310 static void vhost_vdpa_config_put(struct vhost_vdpa *v)
 311 {
 312         if (v->config_ctx) {
 313                 eventfd_ctx_put(v->config_ctx);
 314                 v->config_ctx = NULL;
 315         }
 316 }
 317
 318 static long vhost_vdpa_set_config_call(struct vhost_vdpa *v, u32 __user *argp)
 319 {
 320         struct vdpa_callback cb;
 321         int fd;
 322         struct eventfd_ctx *ctx;
 323
 324         cb.callback = vhost_vdpa_config_cb;
 325         cb.private = v->vdpa;
 326         if (copy_from_user(&fd, argp, sizeof(fd)))
 327                 return  -EFAULT;
 328
 329         ctx = fd == VHOST_FILE_UNBIND ? NULL : eventfd_ctx_fdget(fd);
 330         swap(ctx, v->config_ctx);
 331
 332         if (!IS_ERR_OR_NULL(ctx))
 333                 eventfd_ctx_put(ctx);
 334
 335         if (IS_ERR(v->config_ctx)) {
 336                 long ret = PTR_ERR(v->config_ctx);
 337
 338                 v->config_ctx = NULL;
 339                 return ret;
 340         }
 341
 342         v->vdpa->config->set_config_cb(v->vdpa, &cb);
 343
 344         return 0;
 345 }
 346
 347 static long vhost_vdpa_get_iova_range(struct vhost_vdpa *v, u32 __user *argp)
 348 {
 349         struct vhost_vdpa_iova_range range = {
 350                 .first = v->range.first,
 351                 .last = v->range.last,
 352         };
 353
 354         if (copy_to_user(argp, &range, sizeof(range)))
 355                 return -EFAULT;
 356         return 0;
 357 }
 358
 359 static long vhost_vdpa_vring_ioctl(struct vhost_vdpa *v, unsigned int cmd,
 360                                    void __user *argp)
 361 {
 362         struct vdpa_device *vdpa = v->vdpa;
 363         const struct vdpa_config_ops *ops = vdpa->config;
 364         struct vdpa_vq_state vq_state;
 365         struct vdpa_callback cb;
 366         struct vhost_virtqueue *vq;
 367         struct vhost_vring_state s;
 368         u32 idx;
 369         long r;
 370
 371         r = get_user(idx, (u32 __user *)argp);
 372         if (r < 0)
 373                 return r;
 374
 375         if (idx >= v->nvqs)
 376                 return -ENOBUFS;
 377
 378         idx = array_index_nospec(idx, v->nvqs);
 379         vq = &v->vqs[idx];
 380
 381         switch (cmd) {
 382         case VHOST_VDPA_SET_VRING_ENABLE:
 383                 if (copy_from_user(&s, argp, sizeof(s)))
 384                         return -EFAULT;
 385                 ops->set_vq_ready(vdpa, idx, s.num);
 386                 return 0;
 387         case VHOST_GET_VRING_BASE:
 388                 r = ops->get_vq_state(v->vdpa, idx, &vq_state);
 389                 if (r)
 390                         return r;
 391
 392                 vq->last_avail_idx = vq_state.split.avail_index;
 393                 break;
 394         }
 395
 396         r = vhost_vring_ioctl(&v->vdev, cmd, argp);
 397         if (r)
 398                 return r;
 399
 400         switch (cmd) {
 401         case VHOST_SET_VRING_ADDR:
 402                 if (ops->set_vq_address(vdpa, idx,
 403                                         (u64)(uintptr_t)vq->desc,
 404                                         (u64)(uintptr_t)vq->avail,
 405                                         (u64)(uintptr_t)vq->used))
 406                         r = -EINVAL;
 407                 break;
 408
 409         case VHOST_SET_VRING_BASE:
 410                 vq_state.split.avail_index = vq->last_avail_idx;
 411                 if (ops->set_vq_state(vdpa, idx, &vq_state))
 412                         r = -EINVAL;
 413                 break;
 414
 415         case VHOST_SET_VRING_CALL:
 416                 if (vq->call_ctx.ctx) {
 417                         cb.callback = vhost_vdpa_virtqueue_cb;
 418                         cb.private = vq;
 419                 } else {
 420                         cb.callback = NULL;
 421                         cb.private = NULL;
 422                 }
 423                 ops->set_vq_cb(vdpa, idx, &cb);
 424                 vhost_vdpa_setup_vq_irq(v, idx);
 425                 break;
 426
 427         case VHOST_SET_VRING_NUM:
 428                 ops->set_vq_num(vdpa, idx, vq->num);
 429                 break;
 430         }
 431
 432         return r;
 433 }
 434
 435 static long vhost_vdpa_unlocked_ioctl(struct file *filep,
 436                                       unsigned int cmd, unsigned long arg)
 437 {
 438         struct vhost_vdpa *v = filep->private_data;
 439         struct vhost_dev *d = &v->vdev;
 440         void __user *argp = (void __user *)arg;
 441         u64 __user *featurep = argp;
 442         u64 features;
 443         long r = 0;
 444
 445         if (cmd == VHOST_SET_BACKEND_FEATURES) {
 446                 if (copy_from_user(&features, featurep, sizeof(features)))
 447                         return -EFAULT;
 448                 if (features & ~VHOST_VDPA_BACKEND_FEATURES)
 449                         return -EOPNOTSUPP;
 450                 vhost_set_backend_features(&v->vdev, features);
 451                 return 0;
 452         }
 453
 454         mutex_lock(&d->mutex);
 455
 456         switch (cmd) {
 457         case VHOST_VDPA_GET_DEVICE_ID:
 458                 r = vhost_vdpa_get_device_id(v, argp);
 459                 break;
 460         case VHOST_VDPA_GET_STATUS:
 461                 r = vhost_vdpa_get_status(v, argp);
 462                 break;
 463         case VHOST_VDPA_SET_STATUS:
 464                 r = vhost_vdpa_set_status(v, argp);
 465                 break;
 466         case VHOST_VDPA_GET_CONFIG:
 467                 r = vhost_vdpa_get_config(v, argp);
 468                 break;
 469         case VHOST_VDPA_SET_CONFIG:
 470                 r = vhost_vdpa_set_config(v, argp);
 471                 break;
 472         case VHOST_GET_FEATURES:
 473                 r = vhost_vdpa_get_features(v, argp);
 474                 break;
 475         case VHOST_SET_FEATURES:
 476                 r = vhost_vdpa_set_features(v, argp);
 477                 break;
 478         case VHOST_VDPA_GET_VRING_NUM:
 479                 r = vhost_vdpa_get_vring_num(v, argp);
 480                 break;
 481         case VHOST_SET_LOG_BASE:
 482         case VHOST_SET_LOG_FD:
 483                 r = -ENOIOCTLCMD;
 484                 break;
 485         case VHOST_VDPA_SET_CONFIG_CALL:
 486                 r = vhost_vdpa_set_config_call(v, argp);
 487                 break;
 488         case VHOST_GET_BACKEND_FEATURES:
 489                 features = VHOST_VDPA_BACKEND_FEATURES;
 490                 if (copy_to_user(featurep, &features, sizeof(features)))
 491                         r = -EFAULT;
 492                 break;
 493         case VHOST_VDPA_GET_IOVA_RANGE:
 494                 r = vhost_vdpa_get_iova_range(v, argp);
 495                 break;
 496         default:
 497                 r = vhost_dev_ioctl(&v->vdev, cmd, argp);
 498                 if (r == -ENOIOCTLCMD)
 499                         r = vhost_vdpa_vring_ioctl(v, cmd, argp);
 500                 break;
 501         }
 502
 503         mutex_unlock(&d->mutex);
 504         return r;
 505 }
 506
 507 static void vhost_vdpa_pa_unmap(struct vhost_vdpa *v, u64 start, u64 last)
 508 {
 509         struct vhost_dev *dev = &v->vdev;
 510         struct vhost_iotlb *iotlb = dev->iotlb;
 511         struct vhost_iotlb_map *map;
 512         struct page *page;
 513         unsigned long pfn, pinned;
 514
 515         while ((map = vhost_iotlb_itree_first(iotlb, start, last)) != NULL) {
 516                 pinned = PFN_DOWN(map->size);
 517                 for (pfn = PFN_DOWN(map->addr);
 518                      pinned > 0; pfn++, pinned--) {
 519                         page = pfn_to_page(pfn);
 520                         if (map->perm & VHOST_ACCESS_WO)
 521                                 set_page_dirty_lock(page);
 522                         unpin_user_page(page);
 523                 }
 524                 atomic64_sub(PFN_DOWN(map->size), &dev->mm->pinned_vm);
 525                 vhost_iotlb_map_free(iotlb, map);
 526         }
 527 }
 528
 529 static void vhost_vdpa_va_unmap(struct vhost_vdpa *v, u64 start, u64 last)
 530 {
 531         struct vhost_dev *dev = &v->vdev;
 532         struct vhost_iotlb *iotlb = dev->iotlb;
 533         struct vhost_iotlb_map *map;
 534         struct vdpa_map_file *map_file;
 535
 536         while ((map = vhost_iotlb_itree_first(iotlb, start, last)) != NULL) {
 537                 map_file = (struct vdpa_map_file *)map->opaque;
 538                 fput(map_file->file);
 539                 kfree(map_file);
 540                 vhost_iotlb_map_free(iotlb, map);
 541         }
 542 }
 543
 544 static void vhost_vdpa_iotlb_unmap(struct vhost_vdpa *v, u64 start, u64 last)
 545 {
 546         struct vdpa_device *vdpa = v->vdpa;
 547
 548         if (vdpa->use_va)
 549                 return vhost_vdpa_va_unmap(v, start, last);
 550
 551         return vhost_vdpa_pa_unmap(v, start, last);
 552 }
 553
 554 static void vhost_vdpa_iotlb_free(struct vhost_vdpa *v)
 555 {
 556         struct vhost_dev *dev = &v->vdev;
 557
 558         vhost_vdpa_iotlb_unmap(v, 0ULL, 0ULL - 1);
 559         kfree(dev->iotlb);
 560         dev->iotlb = NULL;
 561 }
 562
 563 static int perm_to_iommu_flags(u32 perm)
 564 {
 565         int flags = 0;
 566
 567         switch (perm) {
 568         case VHOST_ACCESS_WO:
 569                 flags |= IOMMU_WRITE;
 570                 break;
 571         case VHOST_ACCESS_RO:
 572                 flags |= IOMMU_READ;
 573                 break;
 574         case VHOST_ACCESS_RW:
 575                 flags |= (IOMMU_WRITE | IOMMU_READ);
 576                 break;
 577         default:
 578                 WARN(1, "invalidate vhost IOTLB permission\n");
 579                 break;
 580         }
 581
 582         return flags | IOMMU_CACHE;
 583 }
 584
 585 static int vhost_vdpa_map(struct vhost_vdpa *v, u64 iova,
 586                           u64 size, u64 pa, u32 perm, void *opaque)
 587 {
 588         struct vhost_dev *dev = &v->vdev;
 589         struct vdpa_device *vdpa = v->vdpa;
 590         const struct vdpa_config_ops *ops = vdpa->config;
 591         int r = 0;
 592
 593         r = vhost_iotlb_add_range_ctx(dev->iotlb, iova, iova + size - 1,
 594                                       pa, perm, opaque);
 595         if (r)
 596                 return r;
 597
 598         if (ops->dma_map) {
 599                 r = ops->dma_map(vdpa, iova, size, pa, perm, opaque);
 600         } else if (ops->set_map) {
 601                 if (!v->in_batch)
 602                         r = ops->set_map(vdpa, dev->iotlb);
 603         } else {
 604                 r = iommu_map(v->domain, iova, pa, size,
 605                               perm_to_iommu_flags(perm));
 606         }
 607         if (r) {
 608                 vhost_iotlb_del_range(dev->iotlb, iova, iova + size - 1);
 609                 return r;
 610         }
 611
 612         if (!vdpa->use_va)
 613                 atomic64_add(PFN_DOWN(size), &dev->mm->pinned_vm);
 614
 615         return 0;
 616 }
 617
 618 static void vhost_vdpa_unmap(struct vhost_vdpa *v, u64 iova, u64 size)
 619 {
 620         struct vhost_dev *dev = &v->vdev;
 621         struct vdpa_device *vdpa = v->vdpa;
 622         const struct vdpa_config_ops *ops = vdpa->config;
 623
 624         vhost_vdpa_iotlb_unmap(v, iova, iova + size - 1);
 625
 626         if (ops->dma_map) {
 627                 ops->dma_unmap(vdpa, iova, size);
 628         } else if (ops->set_map) {
 629                 if (!v->in_batch)
 630                         ops->set_map(vdpa, dev->iotlb);
 631         } else {
 632                 iommu_unmap(v->domain, iova, size);
 633         }
 634 }
 635
 636 static int vhost_vdpa_va_map(struct vhost_vdpa *v,
 637                              u64 iova, u64 size, u64 uaddr, u32 perm)
 638 {
 639         struct vhost_dev *dev = &v->vdev;
 640         u64 offset, map_size, map_iova = iova;
 641         struct vdpa_map_file *map_file;
 642         struct vm_area_struct *vma;
 643         int ret;
 644
 645         mmap_read_lock(dev->mm);
 646
 647         while (size) {
 648                 vma = find_vma(dev->mm, uaddr);
 649                 if (!vma) {
 650                         ret = -EINVAL;
 651                         break;
 652                 }
 653                 map_size = min(size, vma->vm_end - uaddr);
 654                 if (!(vma->vm_file && (vma->vm_flags & VM_SHARED) &&
 655                         !(vma->vm_flags & (VM_IO | VM_PFNMAP))))
 656                         goto next;
 657
 658                 map_file = kzalloc(sizeof(*map_file), GFP_KERNEL);
 659                 if (!map_file) {
 660                         ret = -ENOMEM;
 661                         break;
 662                 }
 663                 offset = (vma->vm_pgoff << PAGE_SHIFT) + uaddr - vma->vm_start;
 664                 map_file->offset = offset;
 665                 map_file->file = get_file(vma->vm_file);
 666                 ret = vhost_vdpa_map(v, map_iova, map_size, uaddr,
 667                                      perm, map_file);
 668                 if (ret) {
 669                         fput(map_file->file);
 670                         kfree(map_file);
 671                         break;
 672                 }
 673 next:
 674                 size -= map_size;
 675                 uaddr += map_size;
 676                 map_iova += map_size;
 677         }
 678         if (ret)
 679                 vhost_vdpa_unmap(v, iova, map_iova - iova);
 680
 681         mmap_read_unlock(dev->mm);
 682
 683         return ret;
 684 }
 685
 686 static int vhost_vdpa_pa_map(struct vhost_vdpa *v,
 687                              u64 iova, u64 size, u64 uaddr, u32 perm)
 688 {
 689         struct vhost_dev *dev = &v->vdev;
 690         struct page **page_list;
 691         unsigned long list_size = PAGE_SIZE / sizeof(struct page *);
 692         unsigned int gup_flags = FOLL_LONGTERM;
 693         unsigned long npages, cur_base, map_pfn, last_pfn = 0;
 694         unsigned long lock_limit, sz2pin, nchunks, i;
 695         u64 start = iova;
 696         long pinned;
 697         int ret = 0;
 698
 699         /* Limit the use of memory for bookkeeping */
 700         page_list = (struct page **) __get_free_page(GFP_KERNEL);
 701         if (!page_list)
 702                 return -ENOMEM;
 703
 704         if (perm & VHOST_ACCESS_WO)
 705                 gup_flags |= FOLL_WRITE;
 706
 707         npages = PFN_UP(size + (iova & ~PAGE_MASK));
 708         if (!npages) {
 709                 ret = -EINVAL;
 710                 goto free;
 711         }
 712
 713         mmap_read_lock(dev->mm);
 714
 715         lock_limit = PFN_DOWN(rlimit(RLIMIT_MEMLOCK));
 716         if (npages + atomic64_read(&dev->mm->pinned_vm) > lock_limit) {
 717                 ret = -ENOMEM;
 718                 goto unlock;
 719         }
 720
 721         cur_base = uaddr & PAGE_MASK;
 722         iova &= PAGE_MASK;
 723         nchunks = 0;
 724
 725         while (npages) {
 726                 sz2pin = min_t(unsigned long, npages, list_size);
 727                 pinned = pin_user_pages(cur_base, sz2pin,
 728                                         gup_flags, page_list, NULL);
 729                 if (sz2pin != pinned) {
 730                         if (pinned < 0) {
 731                                 ret = pinned;
 732                         } else {
 733                                 unpin_user_pages(page_list, pinned);
 734                                 ret = -ENOMEM;
 735                         }
 736                         goto out;
 737                 }
 738                 nchunks++;
 739
 740                 if (!last_pfn)
 741                         map_pfn = page_to_pfn(page_list[0]);
 742
 743                 for (i = 0; i < pinned; i++) {
 744                         unsigned long this_pfn = page_to_pfn(page_list[i]);
 745                         u64 csize;
 746
 747                         if (last_pfn && (this_pfn != last_pfn + 1)) {
 748                                 /* Pin a contiguous chunk of memory */
 749                                 csize = PFN_PHYS(last_pfn - map_pfn + 1);
 750                                 ret = vhost_vdpa_map(v, iova, csize,
 751                                                      PFN_PHYS(map_pfn),
 752                                                      perm, NULL);
 753                                 if (ret) {
 754                                         /*
 755                                          * Unpin the pages that are left unmapped
 756                                          * from this point on in the current
 757                                          * page_list. The remaining outstanding
 758                                          * ones which may stride across several
 759                                          * chunks will be covered in the common
 760                                          * error path subsequently.
 761                                          */
 762                                         unpin_user_pages(&page_list[i],
 763                                                          pinned - i);
 764                                         goto out;
 765                                 }
 766
 767                                 map_pfn = this_pfn;
 768                                 iova += csize;
 769                                 nchunks = 0;
 770                         }
 771
 772                         last_pfn = this_pfn;
 773                 }
 774
 775                 cur_base += PFN_PHYS(pinned);
 776                 npages -= pinned;
 777         }
 778
 779         /* Pin the rest chunk */
 780         ret = vhost_vdpa_map(v, iova, PFN_PHYS(last_pfn - map_pfn + 1),
 781                              PFN_PHYS(map_pfn), perm, NULL);
 782 out:
 783         if (ret) {
 784                 if (nchunks) {
 785                         unsigned long pfn;
 786
 787                         /*
 788                          * Unpin the outstanding pages which are yet to be
 789                          * mapped but haven't due to vdpa_map() or
 790                          * pin_user_pages() failure.
 791                          *
 792                          * Mapped pages are accounted in vdpa_map(), hence
 793                          * the corresponding unpinning will be handled by
 794                          * vdpa_unmap().
 795                          */
 796                         WARN_ON(!last_pfn);
 797                         for (pfn = map_pfn; pfn <= last_pfn; pfn++)
 798                                 unpin_user_page(pfn_to_page(pfn));
 799                 }
 800                 vhost_vdpa_unmap(v, start, size);
 801         }
 802 unlock:
 803         mmap_read_unlock(dev->mm);
 804 free:
 805         free_page((unsigned long)page_list);
 806         return ret;
 807
 808 }
 809
 810 static int vhost_vdpa_process_iotlb_update(struct vhost_vdpa *v,
 811                                            struct vhost_iotlb_msg *msg)
 812 {
 813         struct vhost_dev *dev = &v->vdev;
 814         struct vdpa_device *vdpa = v->vdpa;
 815         struct vhost_iotlb *iotlb = dev->iotlb;
 816
 817         if (msg->iova < v->range.first || !msg->size ||
 818             msg->iova > U64_MAX - msg->size + 1 ||
 819             msg->iova + msg->size - 1 > v->range.last)
 820                 return -EINVAL;
 821
 822         if (vhost_iotlb_itree_first(iotlb, msg->iova,
 823                                     msg->iova + msg->size - 1))
 824                 return -EEXIST;
 825
 826         if (vdpa->use_va)
 827                 return vhost_vdpa_va_map(v, msg->iova, msg->size,
 828                                          msg->uaddr, msg->perm);
 829
 830         return vhost_vdpa_pa_map(v, msg->iova, msg->size, msg->uaddr,
 831                                  msg->perm);
 832 }
 833
 834 static int vhost_vdpa_process_iotlb_msg(struct vhost_dev *dev,
 835                                         struct vhost_iotlb_msg *msg)
 836 {
 837         struct vhost_vdpa *v = container_of(dev, struct vhost_vdpa, vdev);
 838         struct vdpa_device *vdpa = v->vdpa;
 839         const struct vdpa_config_ops *ops = vdpa->config;
 840         int r = 0;
 841
 842         mutex_lock(&dev->mutex);
 843
 844         r = vhost_dev_check_owner(dev);
 845         if (r)
 846                 goto unlock;
 847
 848         switch (msg->type) {
 849         case VHOST_IOTLB_UPDATE:
 850                 r = vhost_vdpa_process_iotlb_update(v, msg);
 851                 break;
 852         case VHOST_IOTLB_INVALIDATE:
 853                 vhost_vdpa_unmap(v, msg->iova, msg->size);
 854                 break;
 855         case VHOST_IOTLB_BATCH_BEGIN:
 856                 v->in_batch = true;
 857                 break;
 858         case VHOST_IOTLB_BATCH_END:
 859                 if (v->in_batch && ops->set_map)
 860                         ops->set_map(vdpa, dev->iotlb);
 861                 v->in_batch = false;
 862                 break;
 863         default:
 864                 r = -EINVAL;
 865                 break;
 866         }
 867 unlock:
 868         mutex_unlock(&dev->mutex);
 869
 870         return r;
 871 }
 872
 873 static ssize_t vhost_vdpa_chr_write_iter(struct kiocb *iocb,
 874                                          struct iov_iter *from)
 875 {
 876         struct file *file = iocb->ki_filp;
 877         struct vhost_vdpa *v = file->private_data;
 878         struct vhost_dev *dev = &v->vdev;
 879
 880         return vhost_chr_write_iter(dev, from);
 881 }
 882
 883 static int vhost_vdpa_alloc_domain(struct vhost_vdpa *v)
 884 {
 885         struct vdpa_device *vdpa = v->vdpa;
 886         const struct vdpa_config_ops *ops = vdpa->config;
 887         struct device *dma_dev = vdpa_get_dma_dev(vdpa);
 888         struct bus_type *bus;
 889         int ret;
 890
 891         /* Device want to do DMA by itself */
 892         if (ops->set_map || ops->dma_map)
 893                 return 0;
 894
 895         bus = dma_dev->bus;
 896         if (!bus)
 897                 return -EFAULT;
 898
 899         if (!iommu_capable(bus, IOMMU_CAP_CACHE_COHERENCY))
 900                 return -ENOTSUPP;
 901
 902         v->domain = iommu_domain_alloc(bus);
 903         if (!v->domain)
 904                 return -EIO;
 905
 906         ret = iommu_attach_device(v->domain, dma_dev);
 907         if (ret)
 908                 goto err_attach;
 909
 910         return 0;
 911
 912 err_attach:
 913         iommu_domain_free(v->domain);
 914         return ret;
 915 }
 916
 917 static void vhost_vdpa_free_domain(struct vhost_vdpa *v)
 918 {
 919         struct vdpa_device *vdpa = v->vdpa;
 920         struct device *dma_dev = vdpa_get_dma_dev(vdpa);
 921
 922         if (v->domain) {
 923                 iommu_detach_device(v->domain, dma_dev);
 924                 iommu_domain_free(v->domain);
 925         }
 926
 927         v->domain = NULL;
 928 }
 929
 930 static void vhost_vdpa_set_iova_range(struct vhost_vdpa *v)
 931 {
 932         struct vdpa_iova_range *range = &v->range;
 933         struct vdpa_device *vdpa = v->vdpa;
 934         const struct vdpa_config_ops *ops = vdpa->config;
 935
 936         if (ops->get_iova_range) {
 937                 *range = ops->get_iova_range(vdpa);
 938         } else if (v->domain && v->domain->geometry.force_aperture) {
 939                 range->first = v->domain->geometry.aperture_start;
 940                 range->last = v->domain->geometry.aperture_end;
 941         } else {
 942                 range->first = 0;
 943                 range->last = ULLONG_MAX;
 944         }
 945 }
 946
 947 static int vhost_vdpa_open(struct inode *inode, struct file *filep)
 948 {
 949         struct vhost_vdpa *v;
 950         struct vhost_dev *dev;
 951         struct vhost_virtqueue **vqs;
 952         int nvqs, i, r, opened;
 953
 954         v = container_of(inode->i_cdev, struct vhost_vdpa, cdev);
 955
 956         opened = atomic_cmpxchg(&v->opened, 0, 1);
 957         if (opened)
 958                 return -EBUSY;
 959
 960         nvqs = v->nvqs;
 961         r = vhost_vdpa_reset(v);
 962         if (r)
 963                 goto err;
 964
 965         vqs = kmalloc_array(nvqs, sizeof(*vqs), GFP_KERNEL);
 966         if (!vqs) {
 967                 r = -ENOMEM;
 968                 goto err;
 969         }
 970
 971         dev = &v->vdev;
 972         for (i = 0; i < nvqs; i++) {
 973                 vqs[i] = &v->vqs[i];
 974                 vqs[i]->handle_kick = handle_vq_kick;
 975         }
 976         vhost_dev_init(dev, vqs, nvqs, 0, 0, 0, false,
 977                        vhost_vdpa_process_iotlb_msg);
 978
 979         dev->iotlb = vhost_iotlb_alloc(0, 0);
 980         if (!dev->iotlb) {
 981                 r = -ENOMEM;
 982                 goto err_init_iotlb;
 983         }
 984
 985         r = vhost_vdpa_alloc_domain(v);
 986         if (r)
 987                 goto err_init_iotlb;
 988
 989         vhost_vdpa_set_iova_range(v);
 990
 991         filep->private_data = v;
 992
 993         return 0;
 994
 995 err_init_iotlb:
 996         vhost_dev_cleanup(&v->vdev);
 997         kfree(vqs);
 998 err:
 999         atomic_dec(&v->opened);
1000         return r;
1001 }
1002
1003 static void vhost_vdpa_clean_irq(struct vhost_vdpa *v)
1004 {
1005         int i;
1006
1007         for (i = 0; i < v->nvqs; i++)
1008                 vhost_vdpa_unsetup_vq_irq(v, i);
1009 }
1010
1011 static int vhost_vdpa_release(struct inode *inode, struct file *filep)
1012 {
1013         struct vhost_vdpa *v = filep->private_data;
1014         struct vhost_dev *d = &v->vdev;
1015
1016         mutex_lock(&d->mutex);
1017         filep->private_data = NULL;
1018         vhost_vdpa_reset(v);
1019         vhost_dev_stop(&v->vdev);
1020         vhost_vdpa_iotlb_free(v);
1021         vhost_vdpa_free_domain(v);
1022         vhost_vdpa_config_put(v);
1023         vhost_vdpa_clean_irq(v);
1024         vhost_dev_cleanup(&v->vdev);
1025         kfree(v->vdev.vqs);
1026         mutex_unlock(&d->mutex);
1027
1028         atomic_dec(&v->opened);
1029         complete(&v->completion);
1030
1031         return 0;
1032 }
1033
1034 #ifdef CONFIG_MMU
1035 static vm_fault_t vhost_vdpa_fault(struct vm_fault *vmf)
1036 {
1037         struct vhost_vdpa *v = vmf->vma->vm_file->private_data;
1038         struct vdpa_device *vdpa = v->vdpa;
1039         const struct vdpa_config_ops *ops = vdpa->config;
1040         struct vdpa_notification_area notify;
1041         struct vm_area_struct *vma = vmf->vma;
1042         u16 index = vma->vm_pgoff;
1043
1044         notify = ops->get_vq_notification(vdpa, index);
1045
1046         vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1047         if (remap_pfn_range(vma, vmf->address & PAGE_MASK,
1048                             PFN_DOWN(notify.addr), PAGE_SIZE,
1049                             vma->vm_page_prot))
1050                 return VM_FAULT_SIGBUS;
1051
1052         return VM_FAULT_NOPAGE;
1053 }
1054
1055 static const struct vm_operations_struct vhost_vdpa_vm_ops = {
1056         .fault = vhost_vdpa_fault,
1057 };
1058
1059 static int vhost_vdpa_mmap(struct file *file, struct vm_area_struct *vma)
1060 {
1061         struct vhost_vdpa *v = vma->vm_file->private_data;
1062         struct vdpa_device *vdpa = v->vdpa;
1063         const struct vdpa_config_ops *ops = vdpa->config;
1064         struct vdpa_notification_area notify;
1065         unsigned long index = vma->vm_pgoff;
1066
1067         if (vma->vm_end - vma->vm_start != PAGE_SIZE)
1068                 return -EINVAL;
1069         if ((vma->vm_flags & VM_SHARED) == 0)
1070                 return -EINVAL;
1071         if (vma->vm_flags & VM_READ)
1072                 return -EINVAL;
1073         if (index > 65535)
1074                 return -EINVAL;
1075         if (!ops->get_vq_notification)
1076                 return -ENOTSUPP;
1077
1078         /* To be safe and easily modelled by userspace, We only
1079          * support the doorbell which sits on the page boundary and
1080          * does not share the page with other registers.
1081          */
1082         notify = ops->get_vq_notification(vdpa, index);
1083         if (notify.addr & (PAGE_SIZE - 1))
1084                 return -EINVAL;
1085         if (vma->vm_end - vma->vm_start != notify.size)
1086                 return -ENOTSUPP;
1087
1088         vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
1089         vma->vm_ops = &vhost_vdpa_vm_ops;
1090         return 0;
1091 }
1092 #endif /* CONFIG_MMU */
1093
1094 static const struct file_operations vhost_vdpa_fops = {
1095         .owner          = THIS_MODULE,
1096         .open           = vhost_vdpa_open,
1097         .release        = vhost_vdpa_release,
1098         .write_iter     = vhost_vdpa_chr_write_iter,
1099         .unlocked_ioctl = vhost_vdpa_unlocked_ioctl,
1100 #ifdef CONFIG_MMU
1101         .mmap           = vhost_vdpa_mmap,
1102 #endif /* CONFIG_MMU */
1103         .compat_ioctl   = compat_ptr_ioctl,
1104 };
1105
1106 static void vhost_vdpa_release_dev(struct device *device)
1107 {
1108         struct vhost_vdpa *v =
1109                container_of(device, struct vhost_vdpa, dev);
1110
1111         ida_simple_remove(&vhost_vdpa_ida, v->minor);
1112         kfree(v->vqs);
1113         kfree(v);
1114 }
1115
1116 static int vhost_vdpa_probe(struct vdpa_device *vdpa)
1117 {
1118         const struct vdpa_config_ops *ops = vdpa->config;
1119         struct vhost_vdpa *v;
1120         int minor;
1121         int r;
1122
1123         v = kzalloc(sizeof(*v), GFP_KERNEL | __GFP_RETRY_MAYFAIL);
1124         if (!v)
1125                 return -ENOMEM;
1126
1127         minor = ida_simple_get(&vhost_vdpa_ida, 0,
1128                                VHOST_VDPA_DEV_MAX, GFP_KERNEL);
1129         if (minor < 0) {
1130                 kfree(v);
1131                 return minor;
1132         }
1133
1134         atomic_set(&v->opened, 0);
1135         v->minor = minor;
1136         v->vdpa = vdpa;
1137         v->nvqs = vdpa->nvqs;
1138         v->virtio_id = ops->get_device_id(vdpa);
1139
1140         device_initialize(&v->dev);
1141         v->dev.release = vhost_vdpa_release_dev;
1142         v->dev.parent = &vdpa->dev;
1143         v->dev.devt = MKDEV(MAJOR(vhost_vdpa_major), minor);
1144         v->vqs = kmalloc_array(v->nvqs, sizeof(struct vhost_virtqueue),
1145                                GFP_KERNEL);
1146         if (!v->vqs) {
1147                 r = -ENOMEM;
1148                 goto err;
1149         }
1150
1151         r = dev_set_name(&v->dev, "vhost-vdpa-%u", minor);
1152         if (r)
1153                 goto err;
1154
1155         cdev_init(&v->cdev, &vhost_vdpa_fops);
1156         v->cdev.owner = THIS_MODULE;
1157
1158         r = cdev_device_add(&v->cdev, &v->dev);
1159         if (r)
1160                 goto err;
1161
1162         init_completion(&v->completion);
1163         vdpa_set_drvdata(vdpa, v);
1164
1165         return 0;
1166
1167 err:
1168         put_device(&v->dev);
1169         return r;
1170 }
1171
1172 static void vhost_vdpa_remove(struct vdpa_device *vdpa)
1173 {
1174         struct vhost_vdpa *v = vdpa_get_drvdata(vdpa);
1175         int opened;
1176
1177         cdev_device_del(&v->cdev, &v->dev);
1178
1179         do {
1180                 opened = atomic_cmpxchg(&v->opened, 0, 1);
1181                 if (!opened)
1182                         break;
1183                 wait_for_completion(&v->completion);
1184         } while (1);
1185
1186         put_device(&v->dev);
1187 }
1188
1189 static struct vdpa_driver vhost_vdpa_driver = {
1190         .driver = {
1191                 .name   = "vhost_vdpa",
1192         },
1193         .probe  = vhost_vdpa_probe,
1194         .remove = vhost_vdpa_remove,
1195 };
1196
1197 static int __init vhost_vdpa_init(void)
1198 {
1199         int r;
1200
1201         r = alloc_chrdev_region(&vhost_vdpa_major, 0, VHOST_VDPA_DEV_MAX,
1202                                 "vhost-vdpa");
1203         if (r)
1204                 goto err_alloc_chrdev;
1205
1206         r = vdpa_register_driver(&vhost_vdpa_driver);
1207         if (r)
1208                 goto err_vdpa_register_driver;
1209
1210         return 0;
1211
1212 err_vdpa_register_driver:
1213         unregister_chrdev_region(vhost_vdpa_major, VHOST_VDPA_DEV_MAX);
1214 err_alloc_chrdev:
1215         return r;
1216 }
1217 module_init(vhost_vdpa_init);
1218
1219 static void __exit vhost_vdpa_exit(void)
1220 {
1221         vdpa_unregister_driver(&vhost_vdpa_driver);
1222         unregister_chrdev_region(vhost_vdpa_major, VHOST_VDPA_DEV_MAX);
1223 }
1224 module_exit(vhost_vdpa_exit);
1225
1226 MODULE_VERSION("0.0.1");
1227 MODULE_LICENSE("GPL v2");
1228 MODULE_AUTHOR("Intel Corporation");
1229 MODULE_DESCRIPTION("vDPA-based vhost backend for virtio");