arch/x86/kvm/svm/sev.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Kernel-based Virtual Machine driver for Linux
   4  *
   5  * AMD SVM-SEV support
   6  *
   7  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
   8  */
   9
  10 #include <linux/kvm_types.h>
  11 #include <linux/kvm_host.h>
  12 #include <linux/kernel.h>
  13 #include <linux/highmem.h>
  14 #include <linux/psp-sev.h>
  15 #include <linux/pagemap.h>
  16 #include <linux/swap.h>
  17
  18 #include "x86.h"
  19 #include "svm.h"
  20
  21 static int sev_flush_asids(void);
  22 static DECLARE_RWSEM(sev_deactivate_lock);
  23 static DEFINE_MUTEX(sev_bitmap_lock);
  24 unsigned int max_sev_asid;
  25 static unsigned int min_sev_asid;
  26 static unsigned long *sev_asid_bitmap;
  27 static unsigned long *sev_reclaim_asid_bitmap;
  28 #define __sme_page_pa(x) __sme_set(page_to_pfn(x) << PAGE_SHIFT)
  29
  30 struct enc_region {
  31         struct list_head list;
  32         unsigned long npages;
  33         struct page **pages;
  34         unsigned long uaddr;
  35         unsigned long size;
  36 };
  37
  38 static int sev_flush_asids(void)
  39 {
  40         int ret, error = 0;
  41
  42         /*
  43          * DEACTIVATE will clear the WBINVD indicator causing DF_FLUSH to fail,
  44          * so it must be guarded.
  45          */
  46         down_write(&sev_deactivate_lock);
  47
  48         wbinvd_on_all_cpus();
  49         ret = sev_guest_df_flush(&error);
  50
  51         up_write(&sev_deactivate_lock);
  52
  53         if (ret)
  54                 pr_err("SEV: DF_FLUSH failed, ret=%d, error=%#x\n", ret, error);
  55
  56         return ret;
  57 }
  58
  59 /* Must be called with the sev_bitmap_lock held */
  60 static bool __sev_recycle_asids(void)
  61 {
  62         int pos;
  63
  64         /* Check if there are any ASIDs to reclaim before performing a flush */
  65         pos = find_next_bit(sev_reclaim_asid_bitmap,
  66                             max_sev_asid, min_sev_asid - 1);
  67         if (pos >= max_sev_asid)
  68                 return false;
  69
  70         if (sev_flush_asids())
  71                 return false;
  72
  73         bitmap_xor(sev_asid_bitmap, sev_asid_bitmap, sev_reclaim_asid_bitmap,
  74                    max_sev_asid);
  75         bitmap_zero(sev_reclaim_asid_bitmap, max_sev_asid);
  76
  77         return true;
  78 }
  79
  80 static int sev_asid_new(void)
  81 {
  82         bool retry = true;
  83         int pos;
  84
  85         mutex_lock(&sev_bitmap_lock);
  86
  87         /*
  88          * SEV-enabled guest must use asid from min_sev_asid to max_sev_asid.
  89          */
  90 again:
  91         pos = find_next_zero_bit(sev_asid_bitmap, max_sev_asid, min_sev_asid - 1);
  92         if (pos >= max_sev_asid) {
  93                 if (retry && __sev_recycle_asids()) {
  94                         retry = false;
  95                         goto again;
  96                 }
  97                 mutex_unlock(&sev_bitmap_lock);
  98                 return -EBUSY;
  99         }
 100
 101         __set_bit(pos, sev_asid_bitmap);
 102
 103         mutex_unlock(&sev_bitmap_lock);
 104
 105         return pos + 1;
 106 }
 107
 108 static int sev_get_asid(struct kvm *kvm)
 109 {
 110         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 111
 112         return sev->asid;
 113 }
 114
 115 static void sev_asid_free(int asid)
 116 {
 117         struct svm_cpu_data *sd;
 118         int cpu, pos;
 119
 120         mutex_lock(&sev_bitmap_lock);
 121
 122         pos = asid - 1;
 123         __set_bit(pos, sev_reclaim_asid_bitmap);
 124
 125         for_each_possible_cpu(cpu) {
 126                 sd = per_cpu(svm_data, cpu);
 127                 sd->sev_vmcbs[pos] = NULL;
 128         }
 129
 130         mutex_unlock(&sev_bitmap_lock);
 131 }
 132
 133 static void sev_unbind_asid(struct kvm *kvm, unsigned int handle)
 134 {
 135         struct sev_data_decommission *decommission;
 136         struct sev_data_deactivate *data;
 137
 138         if (!handle)
 139                 return;
 140
 141         data = kzalloc(sizeof(*data), GFP_KERNEL);
 142         if (!data)
 143                 return;
 144
 145         /* deactivate handle */
 146         data->handle = handle;
 147
 148         /* Guard DEACTIVATE against WBINVD/DF_FLUSH used in ASID recycling */
 149         down_read(&sev_deactivate_lock);
 150         sev_guest_deactivate(data, NULL);
 151         up_read(&sev_deactivate_lock);
 152
 153         kfree(data);
 154
 155         decommission = kzalloc(sizeof(*decommission), GFP_KERNEL);
 156         if (!decommission)
 157                 return;
 158
 159         /* decommission handle */
 160         decommission->handle = handle;
 161         sev_guest_decommission(decommission, NULL);
 162
 163         kfree(decommission);
 164 }
 165
 166 static int sev_guest_init(struct kvm *kvm, struct kvm_sev_cmd *argp)
 167 {
 168         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 169         int asid, ret;
 170
 171         ret = -EBUSY;
 172         if (unlikely(sev->active))
 173                 return ret;
 174
 175         asid = sev_asid_new();
 176         if (asid < 0)
 177                 return ret;
 178
 179         ret = sev_platform_init(&argp->error);
 180         if (ret)
 181                 goto e_free;
 182
 183         sev->active = true;
 184         sev->asid = asid;
 185         INIT_LIST_HEAD(&sev->regions_list);
 186
 187         return 0;
 188
 189 e_free:
 190         sev_asid_free(asid);
 191         return ret;
 192 }
 193
 194 static int sev_bind_asid(struct kvm *kvm, unsigned int handle, int *error)
 195 {
 196         struct sev_data_activate *data;
 197         int asid = sev_get_asid(kvm);
 198         int ret;
 199
 200         data = kzalloc(sizeof(*data), GFP_KERNEL_ACCOUNT);
 201         if (!data)
 202                 return -ENOMEM;
 203
 204         /* activate ASID on the given handle */
 205         data->handle = handle;
 206         data->asid   = asid;
 207         ret = sev_guest_activate(data, error);
 208         kfree(data);
 209
 210         return ret;
 211 }
 212
 213 static int __sev_issue_cmd(int fd, int id, void *data, int *error)
 214 {
 215         struct fd f;
 216         int ret;
 217
 218         f = fdget(fd);
 219         if (!f.file)
 220                 return -EBADF;
 221
 222         ret = sev_issue_cmd_external_user(f.file, id, data, error);
 223
 224         fdput(f);
 225         return ret;
 226 }
 227
 228 static int sev_issue_cmd(struct kvm *kvm, int id, void *data, int *error)
 229 {
 230         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 231
 232         return __sev_issue_cmd(sev->fd, id, data, error);
 233 }
 234
 235 static int sev_launch_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
 236 {
 237         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 238         struct sev_data_launch_start *start;
 239         struct kvm_sev_launch_start params;
 240         void *dh_blob, *session_blob;
 241         int *error = &argp->error;
 242         int ret;
 243
 244         if (!sev_guest(kvm))
 245                 return -ENOTTY;
 246
 247         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data, sizeof(params)))
 248                 return -EFAULT;
 249
 250         start = kzalloc(sizeof(*start), GFP_KERNEL_ACCOUNT);
 251         if (!start)
 252                 return -ENOMEM;
 253
 254         dh_blob = NULL;
 255         if (params.dh_uaddr) {
 256                 dh_blob = psp_copy_user_blob(params.dh_uaddr, params.dh_len);
 257                 if (IS_ERR(dh_blob)) {
 258                         ret = PTR_ERR(dh_blob);
 259                         goto e_free;
 260                 }
 261
 262                 start->dh_cert_address = __sme_set(__pa(dh_blob));
 263                 start->dh_cert_len = params.dh_len;
 264         }
 265
 266         session_blob = NULL;
 267         if (params.session_uaddr) {
 268                 session_blob = psp_copy_user_blob(params.session_uaddr, params.session_len);
 269                 if (IS_ERR(session_blob)) {
 270                         ret = PTR_ERR(session_blob);
 271                         goto e_free_dh;
 272                 }
 273
 274                 start->session_address = __sme_set(__pa(session_blob));
 275                 start->session_len = params.session_len;
 276         }
 277
 278         start->handle = params.handle;
 279         start->policy = params.policy;
 280
 281         /* create memory encryption context */
 282         ret = __sev_issue_cmd(argp->sev_fd, SEV_CMD_LAUNCH_START, start, error);
 283         if (ret)
 284                 goto e_free_session;
 285
 286         /* Bind ASID to this guest */
 287         ret = sev_bind_asid(kvm, start->handle, error);
 288         if (ret)
 289                 goto e_free_session;
 290
 291         /* return handle to userspace */
 292         params.handle = start->handle;
 293         if (copy_to_user((void __user *)(uintptr_t)argp->data, &params, sizeof(params))) {
 294                 sev_unbind_asid(kvm, start->handle);
 295                 ret = -EFAULT;
 296                 goto e_free_session;
 297         }
 298
 299         sev->handle = start->handle;
 300         sev->fd = argp->sev_fd;
 301
 302 e_free_session:
 303         kfree(session_blob);
 304 e_free_dh:
 305         kfree(dh_blob);
 306 e_free:
 307         kfree(start);
 308         return ret;
 309 }
 310
 311 static struct page **sev_pin_memory(struct kvm *kvm, unsigned long uaddr,
 312                                     unsigned long ulen, unsigned long *n,
 313                                     int write)
 314 {
 315         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 316         unsigned long npages, size;
 317         int npinned;
 318         unsigned long locked, lock_limit;
 319         struct page **pages;
 320         unsigned long first, last;
 321         int ret;
 322
 323         if (ulen == 0 || uaddr + ulen < uaddr)
 324                 return ERR_PTR(-EINVAL);
 325
 326         /* Calculate number of pages. */
 327         first = (uaddr & PAGE_MASK) >> PAGE_SHIFT;
 328         last = ((uaddr + ulen - 1) & PAGE_MASK) >> PAGE_SHIFT;
 329         npages = (last - first + 1);
 330
 331         locked = sev->pages_locked + npages;
 332         lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
 333         if (locked > lock_limit && !capable(CAP_IPC_LOCK)) {
 334                 pr_err("SEV: %lu locked pages exceed the lock limit of %lu.\n", locked, lock_limit);
 335                 return ERR_PTR(-ENOMEM);
 336         }
 337
 338         if (WARN_ON_ONCE(npages > INT_MAX))
 339                 return ERR_PTR(-EINVAL);
 340
 341         /* Avoid using vmalloc for smaller buffers. */
 342         size = npages * sizeof(struct page *);
 343         if (size > PAGE_SIZE)
 344                 pages = __vmalloc(size, GFP_KERNEL_ACCOUNT | __GFP_ZERO);
 345         else
 346                 pages = kmalloc(size, GFP_KERNEL_ACCOUNT);
 347
 348         if (!pages)
 349                 return ERR_PTR(-ENOMEM);
 350
 351         /* Pin the user virtual address. */
 352         npinned = pin_user_pages_fast(uaddr, npages, write ? FOLL_WRITE : 0, pages);
 353         if (npinned != npages) {
 354                 pr_err("SEV: Failure locking %lu pages.\n", npages);
 355                 ret = -ENOMEM;
 356                 goto err;
 357         }
 358
 359         *n = npages;
 360         sev->pages_locked = locked;
 361
 362         return pages;
 363
 364 err:
 365         if (npinned > 0)
 366                 unpin_user_pages(pages, npinned);
 367
 368         kvfree(pages);
 369         return ERR_PTR(ret);
 370 }
 371
 372 static void sev_unpin_memory(struct kvm *kvm, struct page **pages,
 373                              unsigned long npages)
 374 {
 375         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 376
 377         unpin_user_pages(pages, npages);
 378         kvfree(pages);
 379         sev->pages_locked -= npages;
 380 }
 381
 382 static void sev_clflush_pages(struct page *pages[], unsigned long npages)
 383 {
 384         uint8_t *page_virtual;
 385         unsigned long i;
 386
 387         if (this_cpu_has(X86_FEATURE_SME_COHERENT) || npages == 0 ||
 388             pages == NULL)
 389                 return;
 390
 391         for (i = 0; i < npages; i++) {
 392                 page_virtual = kmap_atomic(pages[i]);
 393                 clflush_cache_range(page_virtual, PAGE_SIZE);
 394                 kunmap_atomic(page_virtual);
 395         }
 396 }
 397
 398 static unsigned long get_num_contig_pages(unsigned long idx,
 399                                 struct page **inpages, unsigned long npages)
 400 {
 401         unsigned long paddr, next_paddr;
 402         unsigned long i = idx + 1, pages = 1;
 403
 404         /* find the number of contiguous pages starting from idx */
 405         paddr = __sme_page_pa(inpages[idx]);
 406         while (i < npages) {
 407                 next_paddr = __sme_page_pa(inpages[i++]);
 408                 if ((paddr + PAGE_SIZE) == next_paddr) {
 409                         pages++;
 410                         paddr = next_paddr;
 411                         continue;
 412                 }
 413                 break;
 414         }
 415
 416         return pages;
 417 }
 418
 419 static int sev_launch_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
 420 {
 421         unsigned long vaddr, vaddr_end, next_vaddr, npages, pages, size, i;
 422         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 423         struct kvm_sev_launch_update_data params;
 424         struct sev_data_launch_update_data *data;
 425         struct page **inpages;
 426         int ret;
 427
 428         if (!sev_guest(kvm))
 429                 return -ENOTTY;
 430
 431         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data, sizeof(params)))
 432                 return -EFAULT;
 433
 434         data = kzalloc(sizeof(*data), GFP_KERNEL_ACCOUNT);
 435         if (!data)
 436                 return -ENOMEM;
 437
 438         vaddr = params.uaddr;
 439         size = params.len;
 440         vaddr_end = vaddr + size;
 441
 442         /* Lock the user memory. */
 443         inpages = sev_pin_memory(kvm, vaddr, size, &npages, 1);
 444         if (IS_ERR(inpages)) {
 445                 ret = PTR_ERR(inpages);
 446                 goto e_free;
 447         }
 448
 449         /*
 450          * The LAUNCH_UPDATE command will perform in-place encryption of the
 451          * memory content (i.e it will write the same memory region with C=1).
 452          * It's possible that the cache may contain the data with C=0, i.e.,
 453          * unencrypted so invalidate it first.
 454          */
 455         sev_clflush_pages(inpages, npages);
 456
 457         for (i = 0; vaddr < vaddr_end; vaddr = next_vaddr, i += pages) {
 458                 int offset, len;
 459
 460                 /*
 461                  * If the user buffer is not page-aligned, calculate the offset
 462                  * within the page.
 463                  */
 464                 offset = vaddr & (PAGE_SIZE - 1);
 465
 466                 /* Calculate the number of pages that can be encrypted in one go. */
 467                 pages = get_num_contig_pages(i, inpages, npages);
 468
 469                 len = min_t(size_t, ((pages * PAGE_SIZE) - offset), size);
 470
 471                 data->handle = sev->handle;
 472                 data->len = len;
 473                 data->address = __sme_page_pa(inpages[i]) + offset;
 474                 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_DATA, data, &argp->error);
 475                 if (ret)
 476                         goto e_unpin;
 477
 478                 size -= len;
 479                 next_vaddr = vaddr + len;
 480         }
 481
 482 e_unpin:
 483         /* content of memory is updated, mark pages dirty */
 484         for (i = 0; i < npages; i++) {
 485                 set_page_dirty_lock(inpages[i]);
 486                 mark_page_accessed(inpages[i]);
 487         }
 488         /* unlock the user pages */
 489         sev_unpin_memory(kvm, inpages, npages);
 490 e_free:
 491         kfree(data);
 492         return ret;
 493 }
 494
 495 static int sev_launch_measure(struct kvm *kvm, struct kvm_sev_cmd *argp)
 496 {
 497         void __user *measure = (void __user *)(uintptr_t)argp->data;
 498         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 499         struct sev_data_launch_measure *data;
 500         struct kvm_sev_launch_measure params;
 501         void __user *p = NULL;
 502         void *blob = NULL;
 503         int ret;
 504
 505         if (!sev_guest(kvm))
 506                 return -ENOTTY;
 507
 508         if (copy_from_user(&params, measure, sizeof(params)))
 509                 return -EFAULT;
 510
 511         data = kzalloc(sizeof(*data), GFP_KERNEL_ACCOUNT);
 512         if (!data)
 513                 return -ENOMEM;
 514
 515         /* User wants to query the blob length */
 516         if (!params.len)
 517                 goto cmd;
 518
 519         p = (void __user *)(uintptr_t)params.uaddr;
 520         if (p) {
 521                 if (params.len > SEV_FW_BLOB_MAX_SIZE) {
 522                         ret = -EINVAL;
 523                         goto e_free;
 524                 }
 525
 526                 ret = -ENOMEM;
 527                 blob = kmalloc(params.len, GFP_KERNEL);
 528                 if (!blob)
 529                         goto e_free;
 530
 531                 data->address = __psp_pa(blob);
 532                 data->len = params.len;
 533         }
 534
 535 cmd:
 536         data->handle = sev->handle;
 537         ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_MEASURE, data, &argp->error);
 538
 539         /*
 540          * If we query the session length, FW responded with expected data.
 541          */
 542         if (!params.len)
 543                 goto done;
 544
 545         if (ret)
 546                 goto e_free_blob;
 547
 548         if (blob) {
 549                 if (copy_to_user(p, blob, params.len))
 550                         ret = -EFAULT;
 551         }
 552
 553 done:
 554         params.len = data->len;
 555         if (copy_to_user(measure, &params, sizeof(params)))
 556                 ret = -EFAULT;
 557 e_free_blob:
 558         kfree(blob);
 559 e_free:
 560         kfree(data);
 561         return ret;
 562 }
 563
 564 static int sev_launch_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
 565 {
 566         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 567         struct sev_data_launch_finish *data;
 568         int ret;
 569
 570         if (!sev_guest(kvm))
 571                 return -ENOTTY;
 572
 573         data = kzalloc(sizeof(*data), GFP_KERNEL_ACCOUNT);
 574         if (!data)
 575                 return -ENOMEM;
 576
 577         data->handle = sev->handle;
 578         ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_FINISH, data, &argp->error);
 579
 580         kfree(data);
 581         return ret;
 582 }
 583
 584 static int sev_guest_status(struct kvm *kvm, struct kvm_sev_cmd *argp)
 585 {
 586         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 587         struct kvm_sev_guest_status params;
 588         struct sev_data_guest_status *data;
 589         int ret;
 590
 591         if (!sev_guest(kvm))
 592                 return -ENOTTY;
 593
 594         data = kzalloc(sizeof(*data), GFP_KERNEL_ACCOUNT);
 595         if (!data)
 596                 return -ENOMEM;
 597
 598         data->handle = sev->handle;
 599         ret = sev_issue_cmd(kvm, SEV_CMD_GUEST_STATUS, data, &argp->error);
 600         if (ret)
 601                 goto e_free;
 602
 603         params.policy = data->policy;
 604         params.state = data->state;
 605         params.handle = data->handle;
 606
 607         if (copy_to_user((void __user *)(uintptr_t)argp->data, &params, sizeof(params)))
 608                 ret = -EFAULT;
 609 e_free:
 610         kfree(data);
 611         return ret;
 612 }
 613
 614 static int __sev_issue_dbg_cmd(struct kvm *kvm, unsigned long src,
 615                                unsigned long dst, int size,
 616                                int *error, bool enc)
 617 {
 618         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 619         struct sev_data_dbg *data;
 620         int ret;
 621
 622         data = kzalloc(sizeof(*data), GFP_KERNEL_ACCOUNT);
 623         if (!data)
 624                 return -ENOMEM;
 625
 626         data->handle = sev->handle;
 627         data->dst_addr = dst;
 628         data->src_addr = src;
 629         data->len = size;
 630
 631         ret = sev_issue_cmd(kvm,
 632                             enc ? SEV_CMD_DBG_ENCRYPT : SEV_CMD_DBG_DECRYPT,
 633                             data, error);
 634         kfree(data);
 635         return ret;
 636 }
 637
 638 static int __sev_dbg_decrypt(struct kvm *kvm, unsigned long src_paddr,
 639                              unsigned long dst_paddr, int sz, int *err)
 640 {
 641         int offset;
 642
 643         /*
 644          * Its safe to read more than we are asked, caller should ensure that
 645          * destination has enough space.
 646          */
 647         src_paddr = round_down(src_paddr, 16);
 648         offset = src_paddr & 15;
 649         sz = round_up(sz + offset, 16);
 650
 651         return __sev_issue_dbg_cmd(kvm, src_paddr, dst_paddr, sz, err, false);
 652 }
 653
 654 static int __sev_dbg_decrypt_user(struct kvm *kvm, unsigned long paddr,
 655                                   unsigned long __user dst_uaddr,
 656                                   unsigned long dst_paddr,
 657                                   int size, int *err)
 658 {
 659         struct page *tpage = NULL;
 660         int ret, offset;
 661
 662         /* if inputs are not 16-byte then use intermediate buffer */
 663         if (!IS_ALIGNED(dst_paddr, 16) ||
 664             !IS_ALIGNED(paddr,     16) ||
 665             !IS_ALIGNED(size,      16)) {
 666                 tpage = (void *)alloc_page(GFP_KERNEL);
 667                 if (!tpage)
 668                         return -ENOMEM;
 669
 670                 dst_paddr = __sme_page_pa(tpage);
 671         }
 672
 673         ret = __sev_dbg_decrypt(kvm, paddr, dst_paddr, size, err);
 674         if (ret)
 675                 goto e_free;
 676
 677         if (tpage) {
 678                 offset = paddr & 15;
 679                 if (copy_to_user((void __user *)(uintptr_t)dst_uaddr,
 680                                  page_address(tpage) + offset, size))
 681                         ret = -EFAULT;
 682         }
 683
 684 e_free:
 685         if (tpage)
 686                 __free_page(tpage);
 687
 688         return ret;
 689 }
 690
 691 static int __sev_dbg_encrypt_user(struct kvm *kvm, unsigned long paddr,
 692                                   unsigned long __user vaddr,
 693                                   unsigned long dst_paddr,
 694                                   unsigned long __user dst_vaddr,
 695                                   int size, int *error)
 696 {
 697         struct page *src_tpage = NULL;
 698         struct page *dst_tpage = NULL;
 699         int ret, len = size;
 700
 701         /* If source buffer is not aligned then use an intermediate buffer */
 702         if (!IS_ALIGNED(vaddr, 16)) {
 703                 src_tpage = alloc_page(GFP_KERNEL);
 704                 if (!src_tpage)
 705                         return -ENOMEM;
 706
 707                 if (copy_from_user(page_address(src_tpage),
 708                                 (void __user *)(uintptr_t)vaddr, size)) {
 709                         __free_page(src_tpage);
 710                         return -EFAULT;
 711                 }
 712
 713                 paddr = __sme_page_pa(src_tpage);
 714         }
 715
 716         /*
 717          *  If destination buffer or length is not aligned then do read-modify-write:
 718          *   - decrypt destination in an intermediate buffer
 719          *   - copy the source buffer in an intermediate buffer
 720          *   - use the intermediate buffer as source buffer
 721          */
 722         if (!IS_ALIGNED(dst_vaddr, 16) || !IS_ALIGNED(size, 16)) {
 723                 int dst_offset;
 724
 725                 dst_tpage = alloc_page(GFP_KERNEL);
 726                 if (!dst_tpage) {
 727                         ret = -ENOMEM;
 728                         goto e_free;
 729                 }
 730
 731                 ret = __sev_dbg_decrypt(kvm, dst_paddr,
 732                                         __sme_page_pa(dst_tpage), size, error);
 733                 if (ret)
 734                         goto e_free;
 735
 736                 /*
 737                  *  If source is kernel buffer then use memcpy() otherwise
 738                  *  copy_from_user().
 739                  */
 740                 dst_offset = dst_paddr & 15;
 741
 742                 if (src_tpage)
 743                         memcpy(page_address(dst_tpage) + dst_offset,
 744                                page_address(src_tpage), size);
 745                 else {
 746                         if (copy_from_user(page_address(dst_tpage) + dst_offset,
 747                                            (void __user *)(uintptr_t)vaddr, size)) {
 748                                 ret = -EFAULT;
 749                                 goto e_free;
 750                         }
 751                 }
 752
 753                 paddr = __sme_page_pa(dst_tpage);
 754                 dst_paddr = round_down(dst_paddr, 16);
 755                 len = round_up(size, 16);
 756         }
 757
 758         ret = __sev_issue_dbg_cmd(kvm, paddr, dst_paddr, len, error, true);
 759
 760 e_free:
 761         if (src_tpage)
 762                 __free_page(src_tpage);
 763         if (dst_tpage)
 764                 __free_page(dst_tpage);
 765         return ret;
 766 }
 767
 768 static int sev_dbg_crypt(struct kvm *kvm, struct kvm_sev_cmd *argp, bool dec)
 769 {
 770         unsigned long vaddr, vaddr_end, next_vaddr;
 771         unsigned long dst_vaddr;
 772         struct page **src_p, **dst_p;
 773         struct kvm_sev_dbg debug;
 774         unsigned long n;
 775         unsigned int size;
 776         int ret;
 777
 778         if (!sev_guest(kvm))
 779                 return -ENOTTY;
 780
 781         if (copy_from_user(&debug, (void __user *)(uintptr_t)argp->data, sizeof(debug)))
 782                 return -EFAULT;
 783
 784         if (!debug.len || debug.src_uaddr + debug.len < debug.src_uaddr)
 785                 return -EINVAL;
 786         if (!debug.dst_uaddr)
 787                 return -EINVAL;
 788
 789         vaddr = debug.src_uaddr;
 790         size = debug.len;
 791         vaddr_end = vaddr + size;
 792         dst_vaddr = debug.dst_uaddr;
 793
 794         for (; vaddr < vaddr_end; vaddr = next_vaddr) {
 795                 int len, s_off, d_off;
 796
 797                 /* lock userspace source and destination page */
 798                 src_p = sev_pin_memory(kvm, vaddr & PAGE_MASK, PAGE_SIZE, &n, 0);
 799                 if (IS_ERR(src_p))
 800                         return PTR_ERR(src_p);
 801
 802                 dst_p = sev_pin_memory(kvm, dst_vaddr & PAGE_MASK, PAGE_SIZE, &n, 1);
 803                 if (IS_ERR(dst_p)) {
 804                         sev_unpin_memory(kvm, src_p, n);
 805                         return PTR_ERR(dst_p);
 806                 }
 807
 808                 /*
 809                  * The DBG_{DE,EN}CRYPT commands will perform {dec,en}cryption of the
 810                  * memory content (i.e it will write the same memory region with C=1).
 811                  * It's possible that the cache may contain the data with C=0, i.e.,
 812                  * unencrypted so invalidate it first.
 813                  */
 814                 sev_clflush_pages(src_p, 1);
 815                 sev_clflush_pages(dst_p, 1);
 816
 817                 /*
 818                  * Since user buffer may not be page aligned, calculate the
 819                  * offset within the page.
 820                  */
 821                 s_off = vaddr & ~PAGE_MASK;
 822                 d_off = dst_vaddr & ~PAGE_MASK;
 823                 len = min_t(size_t, (PAGE_SIZE - s_off), size);
 824
 825                 if (dec)
 826                         ret = __sev_dbg_decrypt_user(kvm,
 827                                                      __sme_page_pa(src_p[0]) + s_off,
 828                                                      dst_vaddr,
 829                                                      __sme_page_pa(dst_p[0]) + d_off,
 830                                                      len, &argp->error);
 831                 else
 832                         ret = __sev_dbg_encrypt_user(kvm,
 833                                                      __sme_page_pa(src_p[0]) + s_off,
 834                                                      vaddr,
 835                                                      __sme_page_pa(dst_p[0]) + d_off,
 836                                                      dst_vaddr,
 837                                                      len, &argp->error);
 838
 839                 sev_unpin_memory(kvm, src_p, n);
 840                 sev_unpin_memory(kvm, dst_p, n);
 841
 842                 if (ret)
 843                         goto err;
 844
 845                 next_vaddr = vaddr + len;
 846                 dst_vaddr = dst_vaddr + len;
 847                 size -= len;
 848         }
 849 err:
 850         return ret;
 851 }
 852
 853 static int sev_launch_secret(struct kvm *kvm, struct kvm_sev_cmd *argp)
 854 {
 855         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 856         struct sev_data_launch_secret *data;
 857         struct kvm_sev_launch_secret params;
 858         struct page **pages;
 859         void *blob, *hdr;
 860         unsigned long n;
 861         int ret, offset;
 862
 863         if (!sev_guest(kvm))
 864                 return -ENOTTY;
 865
 866         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data, sizeof(params)))
 867                 return -EFAULT;
 868
 869         pages = sev_pin_memory(kvm, params.guest_uaddr, params.guest_len, &n, 1);
 870         if (IS_ERR(pages))
 871                 return PTR_ERR(pages);
 872
 873         /*
 874          * The secret must be copied into contiguous memory region, lets verify
 875          * that userspace memory pages are contiguous before we issue command.
 876          */
 877         if (get_num_contig_pages(0, pages, n) != n) {
 878                 ret = -EINVAL;
 879                 goto e_unpin_memory;
 880         }
 881
 882         ret = -ENOMEM;
 883         data = kzalloc(sizeof(*data), GFP_KERNEL_ACCOUNT);
 884         if (!data)
 885                 goto e_unpin_memory;
 886
 887         offset = params.guest_uaddr & (PAGE_SIZE - 1);
 888         data->guest_address = __sme_page_pa(pages[0]) + offset;
 889         data->guest_len = params.guest_len;
 890
 891         blob = psp_copy_user_blob(params.trans_uaddr, params.trans_len);
 892         if (IS_ERR(blob)) {
 893                 ret = PTR_ERR(blob);
 894                 goto e_free;
 895         }
 896
 897         data->trans_address = __psp_pa(blob);
 898         data->trans_len = params.trans_len;
 899
 900         hdr = psp_copy_user_blob(params.hdr_uaddr, params.hdr_len);
 901         if (IS_ERR(hdr)) {
 902                 ret = PTR_ERR(hdr);
 903                 goto e_free_blob;
 904         }
 905         data->hdr_address = __psp_pa(hdr);
 906         data->hdr_len = params.hdr_len;
 907
 908         data->handle = sev->handle;
 909         ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_SECRET, data, &argp->error);
 910
 911         kfree(hdr);
 912
 913 e_free_blob:
 914         kfree(blob);
 915 e_free:
 916         kfree(data);
 917 e_unpin_memory:
 918         sev_unpin_memory(kvm, pages, n);
 919         return ret;
 920 }
 921
 922 int svm_mem_enc_op(struct kvm *kvm, void __user *argp)
 923 {
 924         struct kvm_sev_cmd sev_cmd;
 925         int r;
 926
 927         if (!svm_sev_enabled())
 928                 return -ENOTTY;
 929
 930         if (!argp)
 931                 return 0;
 932
 933         if (copy_from_user(&sev_cmd, argp, sizeof(struct kvm_sev_cmd)))
 934                 return -EFAULT;
 935
 936         mutex_lock(&kvm->lock);
 937
 938         switch (sev_cmd.id) {
 939         case KVM_SEV_INIT:
 940                 r = sev_guest_init(kvm, &sev_cmd);
 941                 break;
 942         case KVM_SEV_LAUNCH_START:
 943                 r = sev_launch_start(kvm, &sev_cmd);
 944                 break;
 945         case KVM_SEV_LAUNCH_UPDATE_DATA:
 946                 r = sev_launch_update_data(kvm, &sev_cmd);
 947                 break;
 948         case KVM_SEV_LAUNCH_MEASURE:
 949                 r = sev_launch_measure(kvm, &sev_cmd);
 950                 break;
 951         case KVM_SEV_LAUNCH_FINISH:
 952                 r = sev_launch_finish(kvm, &sev_cmd);
 953                 break;
 954         case KVM_SEV_GUEST_STATUS:
 955                 r = sev_guest_status(kvm, &sev_cmd);
 956                 break;
 957         case KVM_SEV_DBG_DECRYPT:
 958                 r = sev_dbg_crypt(kvm, &sev_cmd, true);
 959                 break;
 960         case KVM_SEV_DBG_ENCRYPT:
 961                 r = sev_dbg_crypt(kvm, &sev_cmd, false);
 962                 break;
 963         case KVM_SEV_LAUNCH_SECRET:
 964                 r = sev_launch_secret(kvm, &sev_cmd);
 965                 break;
 966         default:
 967                 r = -EINVAL;
 968                 goto out;
 969         }
 970
 971         if (copy_to_user(argp, &sev_cmd, sizeof(struct kvm_sev_cmd)))
 972                 r = -EFAULT;
 973
 974 out:
 975         mutex_unlock(&kvm->lock);
 976         return r;
 977 }
 978
 979 int svm_register_enc_region(struct kvm *kvm,
 980                             struct kvm_enc_region *range)
 981 {
 982         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 983         struct enc_region *region;
 984         int ret = 0;
 985
 986         if (!sev_guest(kvm))
 987                 return -ENOTTY;
 988
 989         if (range->addr > ULONG_MAX || range->size > ULONG_MAX)
 990                 return -EINVAL;
 991
 992         region = kzalloc(sizeof(*region), GFP_KERNEL_ACCOUNT);
 993         if (!region)
 994                 return -ENOMEM;
 995
 996         region->pages = sev_pin_memory(kvm, range->addr, range->size, &region->npages, 1);
 997         if (IS_ERR(region->pages)) {
 998                 ret = PTR_ERR(region->pages);
 999                 goto e_free;
1000         }
1001
1002         /*
1003          * The guest may change the memory encryption attribute from C=0 -> C=1
1004          * or vice versa for this memory range. Lets make sure caches are
1005          * flushed to ensure that guest data gets written into memory with
1006          * correct C-bit.
1007          */
1008         sev_clflush_pages(region->pages, region->npages);
1009
1010         region->uaddr = range->addr;
1011         region->size = range->size;
1012
1013         mutex_lock(&kvm->lock);
1014         list_add_tail(&region->list, &sev->regions_list);
1015         mutex_unlock(&kvm->lock);
1016
1017         return ret;
1018
1019 e_free:
1020         kfree(region);
1021         return ret;
1022 }
1023
1024 static struct enc_region *
1025 find_enc_region(struct kvm *kvm, struct kvm_enc_region *range)
1026 {
1027         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1028         struct list_head *head = &sev->regions_list;
1029         struct enc_region *i;
1030
1031         list_for_each_entry(i, head, list) {
1032                 if (i->uaddr == range->addr &&
1033                     i->size == range->size)
1034                         return i;
1035         }
1036
1037         return NULL;
1038 }
1039
1040 static void __unregister_enc_region_locked(struct kvm *kvm,
1041                                            struct enc_region *region)
1042 {
1043         sev_unpin_memory(kvm, region->pages, region->npages);
1044         list_del(&region->list);
1045         kfree(region);
1046 }
1047
1048 int svm_unregister_enc_region(struct kvm *kvm,
1049                               struct kvm_enc_region *range)
1050 {
1051         struct enc_region *region;
1052         int ret;
1053
1054         mutex_lock(&kvm->lock);
1055
1056         if (!sev_guest(kvm)) {
1057                 ret = -ENOTTY;
1058                 goto failed;
1059         }
1060
1061         region = find_enc_region(kvm, range);
1062         if (!region) {
1063                 ret = -EINVAL;
1064                 goto failed;
1065         }
1066
1067         /*
1068          * Ensure that all guest tagged cache entries are flushed before
1069          * releasing the pages back to the system for use. CLFLUSH will
1070          * not do this, so issue a WBINVD.
1071          */
1072         wbinvd_on_all_cpus();
1073
1074         __unregister_enc_region_locked(kvm, region);
1075
1076         mutex_unlock(&kvm->lock);
1077         return 0;
1078
1079 failed:
1080         mutex_unlock(&kvm->lock);
1081         return ret;
1082 }
1083
1084 void sev_vm_destroy(struct kvm *kvm)
1085 {
1086         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1087         struct list_head *head = &sev->regions_list;
1088         struct list_head *pos, *q;
1089
1090         if (!sev_guest(kvm))
1091                 return;
1092
1093         mutex_lock(&kvm->lock);
1094
1095         /*
1096          * Ensure that all guest tagged cache entries are flushed before
1097          * releasing the pages back to the system for use. CLFLUSH will
1098          * not do this, so issue a WBINVD.
1099          */
1100         wbinvd_on_all_cpus();
1101
1102         /*
1103          * if userspace was terminated before unregistering the memory regions
1104          * then lets unpin all the registered memory.
1105          */
1106         if (!list_empty(head)) {
1107                 list_for_each_safe(pos, q, head) {
1108                         __unregister_enc_region_locked(kvm,
1109                                 list_entry(pos, struct enc_region, list));
1110                         cond_resched();
1111                 }
1112         }
1113
1114         mutex_unlock(&kvm->lock);
1115
1116         sev_unbind_asid(kvm, sev->handle);
1117         sev_asid_free(sev->asid);
1118 }
1119
1120 int __init sev_hardware_setup(void)
1121 {
1122         struct sev_user_data_status *status;
1123         int rc;
1124
1125         /* Maximum number of encrypted guests supported simultaneously */
1126         max_sev_asid = cpuid_ecx(0x8000001F);
1127
1128         if (!svm_sev_enabled())
1129                 return 1;
1130
1131         /* Minimum ASID value that should be used for SEV guest */
1132         min_sev_asid = cpuid_edx(0x8000001F);
1133
1134         /* Initialize SEV ASID bitmaps */
1135         sev_asid_bitmap = bitmap_zalloc(max_sev_asid, GFP_KERNEL);
1136         if (!sev_asid_bitmap)
1137                 return 1;
1138
1139         sev_reclaim_asid_bitmap = bitmap_zalloc(max_sev_asid, GFP_KERNEL);
1140         if (!sev_reclaim_asid_bitmap)
1141                 return 1;
1142
1143         status = kmalloc(sizeof(*status), GFP_KERNEL);
1144         if (!status)
1145                 return 1;
1146
1147         /*
1148          * Check SEV platform status.
1149          *
1150          * PLATFORM_STATUS can be called in any state, if we failed to query
1151          * the PLATFORM status then either PSP firmware does not support SEV
1152          * feature or SEV firmware is dead.
1153          */
1154         rc = sev_platform_status(status, NULL);
1155         if (rc)
1156                 goto err;
1157
1158         pr_info("SEV supported\n");
1159
1160 err:
1161         kfree(status);
1162         return rc;
1163 }
1164
1165 void sev_hardware_teardown(void)
1166 {
1167         if (!svm_sev_enabled())
1168                 return;
1169
1170         bitmap_free(sev_asid_bitmap);
1171         bitmap_free(sev_reclaim_asid_bitmap);
1172
1173         sev_flush_asids();
1174 }
1175
1176 void pre_sev_run(struct vcpu_svm *svm, int cpu)
1177 {
1178         struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
1179         int asid = sev_get_asid(svm->vcpu.kvm);
1180
1181         /* Assign the asid allocated with this SEV guest */
1182         svm->vmcb->control.asid = asid;
1183
1184         /*
1185          * Flush guest TLB:
1186          *
1187          * 1) when different VMCB for the same ASID is to be run on the same host CPU.
1188          * 2) or this VMCB was executed on different host CPU in previous VMRUNs.
1189          */
1190         if (sd->sev_vmcbs[asid] == svm->vmcb &&
1191             svm->vcpu.arch.last_vmentry_cpu == cpu)
1192                 return;
1193
1194         sd->sev_vmcbs[asid] = svm->vmcb;
1195         svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
1196         vmcb_mark_dirty(svm->vmcb, VMCB_ASID);
1197 }