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, npinned, size;
 317         unsigned long locked, lock_limit;
 318         struct page **pages;
 319         unsigned long first, last;
 320
 321         if (ulen == 0 || uaddr + ulen < uaddr)
 322                 return NULL;
 323
 324         /* Calculate number of pages. */
 325         first = (uaddr & PAGE_MASK) >> PAGE_SHIFT;
 326         last = ((uaddr + ulen - 1) & PAGE_MASK) >> PAGE_SHIFT;
 327         npages = (last - first + 1);
 328
 329         locked = sev->pages_locked + npages;
 330         lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
 331         if (locked > lock_limit && !capable(CAP_IPC_LOCK)) {
 332                 pr_err("SEV: %lu locked pages exceed the lock limit of %lu.\n", locked, lock_limit);
 333                 return NULL;
 334         }
 335
 336         /* Avoid using vmalloc for smaller buffers. */
 337         size = npages * sizeof(struct page *);
 338         if (size > PAGE_SIZE)
 339                 pages = __vmalloc(size, GFP_KERNEL_ACCOUNT | __GFP_ZERO);
 340         else
 341                 pages = kmalloc(size, GFP_KERNEL_ACCOUNT);
 342
 343         if (!pages)
 344                 return NULL;
 345
 346         /* Pin the user virtual address. */
 347         npinned = get_user_pages_fast(uaddr, npages, write ? FOLL_WRITE : 0, pages);
 348         if (npinned != npages) {
 349                 pr_err("SEV: Failure locking %lu pages.\n", npages);
 350                 goto err;
 351         }
 352
 353         *n = npages;
 354         sev->pages_locked = locked;
 355
 356         return pages;
 357
 358 err:
 359         if (npinned > 0)
 360                 release_pages(pages, npinned);
 361
 362         kvfree(pages);
 363         return NULL;
 364 }
 365
 366 static void sev_unpin_memory(struct kvm *kvm, struct page **pages,
 367                              unsigned long npages)
 368 {
 369         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 370
 371         release_pages(pages, npages);
 372         kvfree(pages);
 373         sev->pages_locked -= npages;
 374 }
 375
 376 static void sev_clflush_pages(struct page *pages[], unsigned long npages)
 377 {
 378         uint8_t *page_virtual;
 379         unsigned long i;
 380
 381         if (npages == 0 || pages == NULL)
 382                 return;
 383
 384         for (i = 0; i < npages; i++) {
 385                 page_virtual = kmap_atomic(pages[i]);
 386                 clflush_cache_range(page_virtual, PAGE_SIZE);
 387                 kunmap_atomic(page_virtual);
 388         }
 389 }
 390
 391 static unsigned long get_num_contig_pages(unsigned long idx,
 392                                 struct page **inpages, unsigned long npages)
 393 {
 394         unsigned long paddr, next_paddr;
 395         unsigned long i = idx + 1, pages = 1;
 396
 397         /* find the number of contiguous pages starting from idx */
 398         paddr = __sme_page_pa(inpages[idx]);
 399         while (i < npages) {
 400                 next_paddr = __sme_page_pa(inpages[i++]);
 401                 if ((paddr + PAGE_SIZE) == next_paddr) {
 402                         pages++;
 403                         paddr = next_paddr;
 404                         continue;
 405                 }
 406                 break;
 407         }
 408
 409         return pages;
 410 }
 411
 412 static int sev_launch_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
 413 {
 414         unsigned long vaddr, vaddr_end, next_vaddr, npages, pages, size, i;
 415         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 416         struct kvm_sev_launch_update_data params;
 417         struct sev_data_launch_update_data *data;
 418         struct page **inpages;
 419         int ret;
 420
 421         if (!sev_guest(kvm))
 422                 return -ENOTTY;
 423
 424         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data, sizeof(params)))
 425                 return -EFAULT;
 426
 427         data = kzalloc(sizeof(*data), GFP_KERNEL_ACCOUNT);
 428         if (!data)
 429                 return -ENOMEM;
 430
 431         vaddr = params.uaddr;
 432         size = params.len;
 433         vaddr_end = vaddr + size;
 434
 435         /* Lock the user memory. */
 436         inpages = sev_pin_memory(kvm, vaddr, size, &npages, 1);
 437         if (!inpages) {
 438                 ret = -ENOMEM;
 439                 goto e_free;
 440         }
 441
 442         /*
 443          * The LAUNCH_UPDATE command will perform in-place encryption of the
 444          * memory content (i.e it will write the same memory region with C=1).
 445          * It's possible that the cache may contain the data with C=0, i.e.,
 446          * unencrypted so invalidate it first.
 447          */
 448         sev_clflush_pages(inpages, npages);
 449
 450         for (i = 0; vaddr < vaddr_end; vaddr = next_vaddr, i += pages) {
 451                 int offset, len;
 452
 453                 /*
 454                  * If the user buffer is not page-aligned, calculate the offset
 455                  * within the page.
 456                  */
 457                 offset = vaddr & (PAGE_SIZE - 1);
 458
 459                 /* Calculate the number of pages that can be encrypted in one go. */
 460                 pages = get_num_contig_pages(i, inpages, npages);
 461
 462                 len = min_t(size_t, ((pages * PAGE_SIZE) - offset), size);
 463
 464                 data->handle = sev->handle;
 465                 data->len = len;
 466                 data->address = __sme_page_pa(inpages[i]) + offset;
 467                 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_DATA, data, &argp->error);
 468                 if (ret)
 469                         goto e_unpin;
 470
 471                 size -= len;
 472                 next_vaddr = vaddr + len;
 473         }
 474
 475 e_unpin:
 476         /* content of memory is updated, mark pages dirty */
 477         for (i = 0; i < npages; i++) {
 478                 set_page_dirty_lock(inpages[i]);
 479                 mark_page_accessed(inpages[i]);
 480         }
 481         /* unlock the user pages */
 482         sev_unpin_memory(kvm, inpages, npages);
 483 e_free:
 484         kfree(data);
 485         return ret;
 486 }
 487
 488 static int sev_launch_measure(struct kvm *kvm, struct kvm_sev_cmd *argp)
 489 {
 490         void __user *measure = (void __user *)(uintptr_t)argp->data;
 491         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 492         struct sev_data_launch_measure *data;
 493         struct kvm_sev_launch_measure params;
 494         void __user *p = NULL;
 495         void *blob = NULL;
 496         int ret;
 497
 498         if (!sev_guest(kvm))
 499                 return -ENOTTY;
 500
 501         if (copy_from_user(&params, measure, sizeof(params)))
 502                 return -EFAULT;
 503
 504         data = kzalloc(sizeof(*data), GFP_KERNEL_ACCOUNT);
 505         if (!data)
 506                 return -ENOMEM;
 507
 508         /* User wants to query the blob length */
 509         if (!params.len)
 510                 goto cmd;
 511
 512         p = (void __user *)(uintptr_t)params.uaddr;
 513         if (p) {
 514                 if (params.len > SEV_FW_BLOB_MAX_SIZE) {
 515                         ret = -EINVAL;
 516                         goto e_free;
 517                 }
 518
 519                 ret = -ENOMEM;
 520                 blob = kmalloc(params.len, GFP_KERNEL);
 521                 if (!blob)
 522                         goto e_free;
 523
 524                 data->address = __psp_pa(blob);
 525                 data->len = params.len;
 526         }
 527
 528 cmd:
 529         data->handle = sev->handle;
 530         ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_MEASURE, data, &argp->error);
 531
 532         /*
 533          * If we query the session length, FW responded with expected data.
 534          */
 535         if (!params.len)
 536                 goto done;
 537
 538         if (ret)
 539                 goto e_free_blob;
 540
 541         if (blob) {
 542                 if (copy_to_user(p, blob, params.len))
 543                         ret = -EFAULT;
 544         }
 545
 546 done:
 547         params.len = data->len;
 548         if (copy_to_user(measure, &params, sizeof(params)))
 549                 ret = -EFAULT;
 550 e_free_blob:
 551         kfree(blob);
 552 e_free:
 553         kfree(data);
 554         return ret;
 555 }
 556
 557 static int sev_launch_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
 558 {
 559         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 560         struct sev_data_launch_finish *data;
 561         int ret;
 562
 563         if (!sev_guest(kvm))
 564                 return -ENOTTY;
 565
 566         data = kzalloc(sizeof(*data), GFP_KERNEL_ACCOUNT);
 567         if (!data)
 568                 return -ENOMEM;
 569
 570         data->handle = sev->handle;
 571         ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_FINISH, data, &argp->error);
 572
 573         kfree(data);
 574         return ret;
 575 }
 576
 577 static int sev_guest_status(struct kvm *kvm, struct kvm_sev_cmd *argp)
 578 {
 579         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 580         struct kvm_sev_guest_status params;
 581         struct sev_data_guest_status *data;
 582         int ret;
 583
 584         if (!sev_guest(kvm))
 585                 return -ENOTTY;
 586
 587         data = kzalloc(sizeof(*data), GFP_KERNEL_ACCOUNT);
 588         if (!data)
 589                 return -ENOMEM;
 590
 591         data->handle = sev->handle;
 592         ret = sev_issue_cmd(kvm, SEV_CMD_GUEST_STATUS, data, &argp->error);
 593         if (ret)
 594                 goto e_free;
 595
 596         params.policy = data->policy;
 597         params.state = data->state;
 598         params.handle = data->handle;
 599
 600         if (copy_to_user((void __user *)(uintptr_t)argp->data, &params, sizeof(params)))
 601                 ret = -EFAULT;
 602 e_free:
 603         kfree(data);
 604         return ret;
 605 }
 606
 607 static int __sev_issue_dbg_cmd(struct kvm *kvm, unsigned long src,
 608                                unsigned long dst, int size,
 609                                int *error, bool enc)
 610 {
 611         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 612         struct sev_data_dbg *data;
 613         int ret;
 614
 615         data = kzalloc(sizeof(*data), GFP_KERNEL_ACCOUNT);
 616         if (!data)
 617                 return -ENOMEM;
 618
 619         data->handle = sev->handle;
 620         data->dst_addr = dst;
 621         data->src_addr = src;
 622         data->len = size;
 623
 624         ret = sev_issue_cmd(kvm,
 625                             enc ? SEV_CMD_DBG_ENCRYPT : SEV_CMD_DBG_DECRYPT,
 626                             data, error);
 627         kfree(data);
 628         return ret;
 629 }
 630
 631 static int __sev_dbg_decrypt(struct kvm *kvm, unsigned long src_paddr,
 632                              unsigned long dst_paddr, int sz, int *err)
 633 {
 634         int offset;
 635
 636         /*
 637          * Its safe to read more than we are asked, caller should ensure that
 638          * destination has enough space.
 639          */
 640         src_paddr = round_down(src_paddr, 16);
 641         offset = src_paddr & 15;
 642         sz = round_up(sz + offset, 16);
 643
 644         return __sev_issue_dbg_cmd(kvm, src_paddr, dst_paddr, sz, err, false);
 645 }
 646
 647 static int __sev_dbg_decrypt_user(struct kvm *kvm, unsigned long paddr,
 648                                   unsigned long __user dst_uaddr,
 649                                   unsigned long dst_paddr,
 650                                   int size, int *err)
 651 {
 652         struct page *tpage = NULL;
 653         int ret, offset;
 654
 655         /* if inputs are not 16-byte then use intermediate buffer */
 656         if (!IS_ALIGNED(dst_paddr, 16) ||
 657             !IS_ALIGNED(paddr,     16) ||
 658             !IS_ALIGNED(size,      16)) {
 659                 tpage = (void *)alloc_page(GFP_KERNEL);
 660                 if (!tpage)
 661                         return -ENOMEM;
 662
 663                 dst_paddr = __sme_page_pa(tpage);
 664         }
 665
 666         ret = __sev_dbg_decrypt(kvm, paddr, dst_paddr, size, err);
 667         if (ret)
 668                 goto e_free;
 669
 670         if (tpage) {
 671                 offset = paddr & 15;
 672                 if (copy_to_user((void __user *)(uintptr_t)dst_uaddr,
 673                                  page_address(tpage) + offset, size))
 674                         ret = -EFAULT;
 675         }
 676
 677 e_free:
 678         if (tpage)
 679                 __free_page(tpage);
 680
 681         return ret;
 682 }
 683
 684 static int __sev_dbg_encrypt_user(struct kvm *kvm, unsigned long paddr,
 685                                   unsigned long __user vaddr,
 686                                   unsigned long dst_paddr,
 687                                   unsigned long __user dst_vaddr,
 688                                   int size, int *error)
 689 {
 690         struct page *src_tpage = NULL;
 691         struct page *dst_tpage = NULL;
 692         int ret, len = size;
 693
 694         /* If source buffer is not aligned then use an intermediate buffer */
 695         if (!IS_ALIGNED(vaddr, 16)) {
 696                 src_tpage = alloc_page(GFP_KERNEL);
 697                 if (!src_tpage)
 698                         return -ENOMEM;
 699
 700                 if (copy_from_user(page_address(src_tpage),
 701                                 (void __user *)(uintptr_t)vaddr, size)) {
 702                         __free_page(src_tpage);
 703                         return -EFAULT;
 704                 }
 705
 706                 paddr = __sme_page_pa(src_tpage);
 707         }
 708
 709         /*
 710          *  If destination buffer or length is not aligned then do read-modify-write:
 711          *   - decrypt destination in an intermediate buffer
 712          *   - copy the source buffer in an intermediate buffer
 713          *   - use the intermediate buffer as source buffer
 714          */
 715         if (!IS_ALIGNED(dst_vaddr, 16) || !IS_ALIGNED(size, 16)) {
 716                 int dst_offset;
 717
 718                 dst_tpage = alloc_page(GFP_KERNEL);
 719                 if (!dst_tpage) {
 720                         ret = -ENOMEM;
 721                         goto e_free;
 722                 }
 723
 724                 ret = __sev_dbg_decrypt(kvm, dst_paddr,
 725                                         __sme_page_pa(dst_tpage), size, error);
 726                 if (ret)
 727                         goto e_free;
 728
 729                 /*
 730                  *  If source is kernel buffer then use memcpy() otherwise
 731                  *  copy_from_user().
 732                  */
 733                 dst_offset = dst_paddr & 15;
 734
 735                 if (src_tpage)
 736                         memcpy(page_address(dst_tpage) + dst_offset,
 737                                page_address(src_tpage), size);
 738                 else {
 739                         if (copy_from_user(page_address(dst_tpage) + dst_offset,
 740                                            (void __user *)(uintptr_t)vaddr, size)) {
 741                                 ret = -EFAULT;
 742                                 goto e_free;
 743                         }
 744                 }
 745
 746                 paddr = __sme_page_pa(dst_tpage);
 747                 dst_paddr = round_down(dst_paddr, 16);
 748                 len = round_up(size, 16);
 749         }
 750
 751         ret = __sev_issue_dbg_cmd(kvm, paddr, dst_paddr, len, error, true);
 752
 753 e_free:
 754         if (src_tpage)
 755                 __free_page(src_tpage);
 756         if (dst_tpage)
 757                 __free_page(dst_tpage);
 758         return ret;
 759 }
 760
 761 static int sev_dbg_crypt(struct kvm *kvm, struct kvm_sev_cmd *argp, bool dec)
 762 {
 763         unsigned long vaddr, vaddr_end, next_vaddr;
 764         unsigned long dst_vaddr;
 765         struct page **src_p, **dst_p;
 766         struct kvm_sev_dbg debug;
 767         unsigned long n;
 768         unsigned int size;
 769         int ret;
 770
 771         if (!sev_guest(kvm))
 772                 return -ENOTTY;
 773
 774         if (copy_from_user(&debug, (void __user *)(uintptr_t)argp->data, sizeof(debug)))
 775                 return -EFAULT;
 776
 777         if (!debug.len || debug.src_uaddr + debug.len < debug.src_uaddr)
 778                 return -EINVAL;
 779         if (!debug.dst_uaddr)
 780                 return -EINVAL;
 781
 782         vaddr = debug.src_uaddr;
 783         size = debug.len;
 784         vaddr_end = vaddr + size;
 785         dst_vaddr = debug.dst_uaddr;
 786
 787         for (; vaddr < vaddr_end; vaddr = next_vaddr) {
 788                 int len, s_off, d_off;
 789
 790                 /* lock userspace source and destination page */
 791                 src_p = sev_pin_memory(kvm, vaddr & PAGE_MASK, PAGE_SIZE, &n, 0);
 792                 if (!src_p)
 793                         return -EFAULT;
 794
 795                 dst_p = sev_pin_memory(kvm, dst_vaddr & PAGE_MASK, PAGE_SIZE, &n, 1);
 796                 if (!dst_p) {
 797                         sev_unpin_memory(kvm, src_p, n);
 798                         return -EFAULT;
 799                 }
 800
 801                 /*
 802                  * The DBG_{DE,EN}CRYPT commands will perform {dec,en}cryption of the
 803                  * memory content (i.e it will write the same memory region with C=1).
 804                  * It's possible that the cache may contain the data with C=0, i.e.,
 805                  * unencrypted so invalidate it first.
 806                  */
 807                 sev_clflush_pages(src_p, 1);
 808                 sev_clflush_pages(dst_p, 1);
 809
 810                 /*
 811                  * Since user buffer may not be page aligned, calculate the
 812                  * offset within the page.
 813                  */
 814                 s_off = vaddr & ~PAGE_MASK;
 815                 d_off = dst_vaddr & ~PAGE_MASK;
 816                 len = min_t(size_t, (PAGE_SIZE - s_off), size);
 817
 818                 if (dec)
 819                         ret = __sev_dbg_decrypt_user(kvm,
 820                                                      __sme_page_pa(src_p[0]) + s_off,
 821                                                      dst_vaddr,
 822                                                      __sme_page_pa(dst_p[0]) + d_off,
 823                                                      len, &argp->error);
 824                 else
 825                         ret = __sev_dbg_encrypt_user(kvm,
 826                                                      __sme_page_pa(src_p[0]) + s_off,
 827                                                      vaddr,
 828                                                      __sme_page_pa(dst_p[0]) + d_off,
 829                                                      dst_vaddr,
 830                                                      len, &argp->error);
 831
 832                 sev_unpin_memory(kvm, src_p, n);
 833                 sev_unpin_memory(kvm, dst_p, n);
 834
 835                 if (ret)
 836                         goto err;
 837
 838                 next_vaddr = vaddr + len;
 839                 dst_vaddr = dst_vaddr + len;
 840                 size -= len;
 841         }
 842 err:
 843         return ret;
 844 }
 845
 846 static int sev_launch_secret(struct kvm *kvm, struct kvm_sev_cmd *argp)
 847 {
 848         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 849         struct sev_data_launch_secret *data;
 850         struct kvm_sev_launch_secret params;
 851         struct page **pages;
 852         void *blob, *hdr;
 853         unsigned long n;
 854         int ret, offset;
 855
 856         if (!sev_guest(kvm))
 857                 return -ENOTTY;
 858
 859         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data, sizeof(params)))
 860                 return -EFAULT;
 861
 862         pages = sev_pin_memory(kvm, params.guest_uaddr, params.guest_len, &n, 1);
 863         if (!pages)
 864                 return -ENOMEM;
 865
 866         /*
 867          * The secret must be copied into contiguous memory region, lets verify
 868          * that userspace memory pages are contiguous before we issue command.
 869          */
 870         if (get_num_contig_pages(0, pages, n) != n) {
 871                 ret = -EINVAL;
 872                 goto e_unpin_memory;
 873         }
 874
 875         ret = -ENOMEM;
 876         data = kzalloc(sizeof(*data), GFP_KERNEL_ACCOUNT);
 877         if (!data)
 878                 goto e_unpin_memory;
 879
 880         offset = params.guest_uaddr & (PAGE_SIZE - 1);
 881         data->guest_address = __sme_page_pa(pages[0]) + offset;
 882         data->guest_len = params.guest_len;
 883
 884         blob = psp_copy_user_blob(params.trans_uaddr, params.trans_len);
 885         if (IS_ERR(blob)) {
 886                 ret = PTR_ERR(blob);
 887                 goto e_free;
 888         }
 889
 890         data->trans_address = __psp_pa(blob);
 891         data->trans_len = params.trans_len;
 892
 893         hdr = psp_copy_user_blob(params.hdr_uaddr, params.hdr_len);
 894         if (IS_ERR(hdr)) {
 895                 ret = PTR_ERR(hdr);
 896                 goto e_free_blob;
 897         }
 898         data->hdr_address = __psp_pa(hdr);
 899         data->hdr_len = params.hdr_len;
 900
 901         data->handle = sev->handle;
 902         ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_SECRET, data, &argp->error);
 903
 904         kfree(hdr);
 905
 906 e_free_blob:
 907         kfree(blob);
 908 e_free:
 909         kfree(data);
 910 e_unpin_memory:
 911         sev_unpin_memory(kvm, pages, n);
 912         return ret;
 913 }
 914
 915 int svm_mem_enc_op(struct kvm *kvm, void __user *argp)
 916 {
 917         struct kvm_sev_cmd sev_cmd;
 918         int r;
 919
 920         if (!svm_sev_enabled())
 921                 return -ENOTTY;
 922
 923         if (!argp)
 924                 return 0;
 925
 926         if (copy_from_user(&sev_cmd, argp, sizeof(struct kvm_sev_cmd)))
 927                 return -EFAULT;
 928
 929         mutex_lock(&kvm->lock);
 930
 931         switch (sev_cmd.id) {
 932         case KVM_SEV_INIT:
 933                 r = sev_guest_init(kvm, &sev_cmd);
 934                 break;
 935         case KVM_SEV_LAUNCH_START:
 936                 r = sev_launch_start(kvm, &sev_cmd);
 937                 break;
 938         case KVM_SEV_LAUNCH_UPDATE_DATA:
 939                 r = sev_launch_update_data(kvm, &sev_cmd);
 940                 break;
 941         case KVM_SEV_LAUNCH_MEASURE:
 942                 r = sev_launch_measure(kvm, &sev_cmd);
 943                 break;
 944         case KVM_SEV_LAUNCH_FINISH:
 945                 r = sev_launch_finish(kvm, &sev_cmd);
 946                 break;
 947         case KVM_SEV_GUEST_STATUS:
 948                 r = sev_guest_status(kvm, &sev_cmd);
 949                 break;
 950         case KVM_SEV_DBG_DECRYPT:
 951                 r = sev_dbg_crypt(kvm, &sev_cmd, true);
 952                 break;
 953         case KVM_SEV_DBG_ENCRYPT:
 954                 r = sev_dbg_crypt(kvm, &sev_cmd, false);
 955                 break;
 956         case KVM_SEV_LAUNCH_SECRET:
 957                 r = sev_launch_secret(kvm, &sev_cmd);
 958                 break;
 959         default:
 960                 r = -EINVAL;
 961                 goto out;
 962         }
 963
 964         if (copy_to_user(argp, &sev_cmd, sizeof(struct kvm_sev_cmd)))
 965                 r = -EFAULT;
 966
 967 out:
 968         mutex_unlock(&kvm->lock);
 969         return r;
 970 }
 971
 972 int svm_register_enc_region(struct kvm *kvm,
 973                             struct kvm_enc_region *range)
 974 {
 975         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 976         struct enc_region *region;
 977         int ret = 0;
 978
 979         if (!sev_guest(kvm))
 980                 return -ENOTTY;
 981
 982         if (range->addr > ULONG_MAX || range->size > ULONG_MAX)
 983                 return -EINVAL;
 984
 985         region = kzalloc(sizeof(*region), GFP_KERNEL_ACCOUNT);
 986         if (!region)
 987                 return -ENOMEM;
 988
 989         region->pages = sev_pin_memory(kvm, range->addr, range->size, &region->npages, 1);
 990         if (!region->pages) {
 991                 ret = -ENOMEM;
 992                 goto e_free;
 993         }
 994
 995         /*
 996          * The guest may change the memory encryption attribute from C=0 -> C=1
 997          * or vice versa for this memory range. Lets make sure caches are
 998          * flushed to ensure that guest data gets written into memory with
 999          * correct C-bit.
1000          */
1001         sev_clflush_pages(region->pages, region->npages);
1002
1003         region->uaddr = range->addr;
1004         region->size = range->size;
1005
1006         mutex_lock(&kvm->lock);
1007         list_add_tail(&region->list, &sev->regions_list);
1008         mutex_unlock(&kvm->lock);
1009
1010         return ret;
1011
1012 e_free:
1013         kfree(region);
1014         return ret;
1015 }
1016
1017 static struct enc_region *
1018 find_enc_region(struct kvm *kvm, struct kvm_enc_region *range)
1019 {
1020         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1021         struct list_head *head = &sev->regions_list;
1022         struct enc_region *i;
1023
1024         list_for_each_entry(i, head, list) {
1025                 if (i->uaddr == range->addr &&
1026                     i->size == range->size)
1027                         return i;
1028         }
1029
1030         return NULL;
1031 }
1032
1033 static void __unregister_enc_region_locked(struct kvm *kvm,
1034                                            struct enc_region *region)
1035 {
1036         sev_unpin_memory(kvm, region->pages, region->npages);
1037         list_del(&region->list);
1038         kfree(region);
1039 }
1040
1041 int svm_unregister_enc_region(struct kvm *kvm,
1042                               struct kvm_enc_region *range)
1043 {
1044         struct enc_region *region;
1045         int ret;
1046
1047         mutex_lock(&kvm->lock);
1048
1049         if (!sev_guest(kvm)) {
1050                 ret = -ENOTTY;
1051                 goto failed;
1052         }
1053
1054         region = find_enc_region(kvm, range);
1055         if (!region) {
1056                 ret = -EINVAL;
1057                 goto failed;
1058         }
1059
1060         /*
1061          * Ensure that all guest tagged cache entries are flushed before
1062          * releasing the pages back to the system for use. CLFLUSH will
1063          * not do this, so issue a WBINVD.
1064          */
1065         wbinvd_on_all_cpus();
1066
1067         __unregister_enc_region_locked(kvm, region);
1068
1069         mutex_unlock(&kvm->lock);
1070         return 0;
1071
1072 failed:
1073         mutex_unlock(&kvm->lock);
1074         return ret;
1075 }
1076
1077 void sev_vm_destroy(struct kvm *kvm)
1078 {
1079         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1080         struct list_head *head = &sev->regions_list;
1081         struct list_head *pos, *q;
1082
1083         if (!sev_guest(kvm))
1084                 return;
1085
1086         mutex_lock(&kvm->lock);
1087
1088         /*
1089          * Ensure that all guest tagged cache entries are flushed before
1090          * releasing the pages back to the system for use. CLFLUSH will
1091          * not do this, so issue a WBINVD.
1092          */
1093         wbinvd_on_all_cpus();
1094
1095         /*
1096          * if userspace was terminated before unregistering the memory regions
1097          * then lets unpin all the registered memory.
1098          */
1099         if (!list_empty(head)) {
1100                 list_for_each_safe(pos, q, head) {
1101                         __unregister_enc_region_locked(kvm,
1102                                 list_entry(pos, struct enc_region, list));
1103                 }
1104         }
1105
1106         mutex_unlock(&kvm->lock);
1107
1108         sev_unbind_asid(kvm, sev->handle);
1109         sev_asid_free(sev->asid);
1110 }
1111
1112 int __init sev_hardware_setup(void)
1113 {
1114         struct sev_user_data_status *status;
1115         int rc;
1116
1117         /* Maximum number of encrypted guests supported simultaneously */
1118         max_sev_asid = cpuid_ecx(0x8000001F);
1119
1120         if (!svm_sev_enabled())
1121                 return 1;
1122
1123         /* Minimum ASID value that should be used for SEV guest */
1124         min_sev_asid = cpuid_edx(0x8000001F);
1125
1126         /* Initialize SEV ASID bitmaps */
1127         sev_asid_bitmap = bitmap_zalloc(max_sev_asid, GFP_KERNEL);
1128         if (!sev_asid_bitmap)
1129                 return 1;
1130
1131         sev_reclaim_asid_bitmap = bitmap_zalloc(max_sev_asid, GFP_KERNEL);
1132         if (!sev_reclaim_asid_bitmap)
1133                 return 1;
1134
1135         status = kmalloc(sizeof(*status), GFP_KERNEL);
1136         if (!status)
1137                 return 1;
1138
1139         /*
1140          * Check SEV platform status.
1141          *
1142          * PLATFORM_STATUS can be called in any state, if we failed to query
1143          * the PLATFORM status then either PSP firmware does not support SEV
1144          * feature or SEV firmware is dead.
1145          */
1146         rc = sev_platform_status(status, NULL);
1147         if (rc)
1148                 goto err;
1149
1150         pr_info("SEV supported\n");
1151
1152 err:
1153         kfree(status);
1154         return rc;
1155 }
1156
1157 void sev_hardware_teardown(void)
1158 {
1159         if (!svm_sev_enabled())
1160                 return;
1161
1162         bitmap_free(sev_asid_bitmap);
1163         bitmap_free(sev_reclaim_asid_bitmap);
1164
1165         sev_flush_asids();
1166 }
1167
1168 void pre_sev_run(struct vcpu_svm *svm, int cpu)
1169 {
1170         struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
1171         int asid = sev_get_asid(svm->vcpu.kvm);
1172
1173         /* Assign the asid allocated with this SEV guest */
1174         svm->vmcb->control.asid = asid;
1175
1176         /*
1177          * Flush guest TLB:
1178          *
1179          * 1) when different VMCB for the same ASID is to be run on the same host CPU.
1180          * 2) or this VMCB was executed on different host CPU in previous VMRUNs.
1181          */
1182         if (sd->sev_vmcbs[asid] == svm->vmcb &&
1183             svm->last_cpu == cpu)
1184                 return;
1185
1186         svm->last_cpu = cpu;
1187         sd->sev_vmcbs[asid] = svm->vmcb;
1188         svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
1189         mark_dirty(svm->vmcb, VMCB_ASID);
1190 }