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 #include <linux/processor.h>
  18 #include <linux/trace_events.h>
  19 #include <asm/fpu/internal.h>
  20
  21 #include <asm/trapnr.h>
  22
  23 #include "x86.h"
  24 #include "svm.h"
  25 #include "svm_ops.h"
  26 #include "cpuid.h"
  27 #include "trace.h"
  28
  29 #define __ex(x) __kvm_handle_fault_on_reboot(x)
  30
  31 static u8 sev_enc_bit;
  32 static int sev_flush_asids(void);
  33 static DECLARE_RWSEM(sev_deactivate_lock);
  34 static DEFINE_MUTEX(sev_bitmap_lock);
  35 unsigned int max_sev_asid;
  36 static unsigned int min_sev_asid;
  37 static unsigned long sev_me_mask;
  38 static unsigned long *sev_asid_bitmap;
  39 static unsigned long *sev_reclaim_asid_bitmap;
  40
  41 struct enc_region {
  42         struct list_head list;
  43         unsigned long npages;
  44         struct page **pages;
  45         unsigned long uaddr;
  46         unsigned long size;
  47 };
  48
  49 static int sev_flush_asids(void)
  50 {
  51         int ret, error = 0;
  52
  53         /*
  54          * DEACTIVATE will clear the WBINVD indicator causing DF_FLUSH to fail,
  55          * so it must be guarded.
  56          */
  57         down_write(&sev_deactivate_lock);
  58
  59         wbinvd_on_all_cpus();
  60         ret = sev_guest_df_flush(&error);
  61
  62         up_write(&sev_deactivate_lock);
  63
  64         if (ret)
  65                 pr_err("SEV: DF_FLUSH failed, ret=%d, error=%#x\n", ret, error);
  66
  67         return ret;
  68 }
  69
  70 static inline bool is_mirroring_enc_context(struct kvm *kvm)
  71 {
  72         return !!to_kvm_svm(kvm)->sev_info.enc_context_owner;
  73 }
  74
  75 /* Must be called with the sev_bitmap_lock held */
  76 static bool __sev_recycle_asids(int min_asid, int max_asid)
  77 {
  78         int pos;
  79
  80         /* Check if there are any ASIDs to reclaim before performing a flush */
  81         pos = find_next_bit(sev_reclaim_asid_bitmap, max_sev_asid, min_asid);
  82         if (pos >= max_asid)
  83                 return false;
  84
  85         if (sev_flush_asids())
  86                 return false;
  87
  88         /* The flush process will flush all reclaimable SEV and SEV-ES ASIDs */
  89         bitmap_xor(sev_asid_bitmap, sev_asid_bitmap, sev_reclaim_asid_bitmap,
  90                    max_sev_asid);
  91         bitmap_zero(sev_reclaim_asid_bitmap, max_sev_asid);
  92
  93         return true;
  94 }
  95
  96 static int sev_asid_new(bool es_active)
  97 {
  98         int pos, min_asid, max_asid;
  99         bool retry = true;
 100
 101         mutex_lock(&sev_bitmap_lock);
 102
 103         /*
 104          * SEV-enabled guests must use asid from min_sev_asid to max_sev_asid.
 105          * SEV-ES-enabled guest can use from 1 to min_sev_asid - 1.
 106          */
 107         min_asid = es_active ? 0 : min_sev_asid - 1;
 108         max_asid = es_active ? min_sev_asid - 1 : max_sev_asid;
 109 again:
 110         pos = find_next_zero_bit(sev_asid_bitmap, max_sev_asid, min_asid);
 111         if (pos >= max_asid) {
 112                 if (retry && __sev_recycle_asids(min_asid, max_asid)) {
 113                         retry = false;
 114                         goto again;
 115                 }
 116                 mutex_unlock(&sev_bitmap_lock);
 117                 return -EBUSY;
 118         }
 119
 120         __set_bit(pos, sev_asid_bitmap);
 121
 122         mutex_unlock(&sev_bitmap_lock);
 123
 124         return pos + 1;
 125 }
 126
 127 static int sev_get_asid(struct kvm *kvm)
 128 {
 129         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 130
 131         return sev->asid;
 132 }
 133
 134 static void sev_asid_free(int asid)
 135 {
 136         struct svm_cpu_data *sd;
 137         int cpu, pos;
 138
 139         mutex_lock(&sev_bitmap_lock);
 140
 141         pos = asid - 1;
 142         __set_bit(pos, sev_reclaim_asid_bitmap);
 143
 144         for_each_possible_cpu(cpu) {
 145                 sd = per_cpu(svm_data, cpu);
 146                 sd->sev_vmcbs[pos] = NULL;
 147         }
 148
 149         mutex_unlock(&sev_bitmap_lock);
 150 }
 151
 152 static void sev_unbind_asid(struct kvm *kvm, unsigned int handle)
 153 {
 154         struct sev_data_decommission decommission;
 155         struct sev_data_deactivate deactivate;
 156
 157         if (!handle)
 158                 return;
 159
 160         deactivate.handle = handle;
 161
 162         /* Guard DEACTIVATE against WBINVD/DF_FLUSH used in ASID recycling */
 163         down_read(&sev_deactivate_lock);
 164         sev_guest_deactivate(&deactivate, NULL);
 165         up_read(&sev_deactivate_lock);
 166
 167         /* decommission handle */
 168         decommission.handle = handle;
 169         sev_guest_decommission(&decommission, NULL);
 170 }
 171
 172 static int sev_guest_init(struct kvm *kvm, struct kvm_sev_cmd *argp)
 173 {
 174         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 175         bool es_active = argp->id == KVM_SEV_ES_INIT;
 176         int asid, ret;
 177
 178         if (kvm->created_vcpus)
 179                 return -EINVAL;
 180
 181         ret = -EBUSY;
 182         if (unlikely(sev->active))
 183                 return ret;
 184
 185         asid = sev_asid_new(es_active);
 186         if (asid < 0)
 187                 return ret;
 188
 189         ret = sev_platform_init(&argp->error);
 190         if (ret)
 191                 goto e_free;
 192
 193         sev->active = true;
 194         sev->es_active = es_active;
 195         sev->asid = asid;
 196         INIT_LIST_HEAD(&sev->regions_list);
 197
 198         return 0;
 199
 200 e_free:
 201         sev_asid_free(asid);
 202         return ret;
 203 }
 204
 205 static int sev_bind_asid(struct kvm *kvm, unsigned int handle, int *error)
 206 {
 207         struct sev_data_activate activate;
 208         int asid = sev_get_asid(kvm);
 209         int ret;
 210
 211         /* activate ASID on the given handle */
 212         activate.handle = handle;
 213         activate.asid   = asid;
 214         ret = sev_guest_activate(&activate, error);
 215
 216         return ret;
 217 }
 218
 219 static int __sev_issue_cmd(int fd, int id, void *data, int *error)
 220 {
 221         struct fd f;
 222         int ret;
 223
 224         f = fdget(fd);
 225         if (!f.file)
 226                 return -EBADF;
 227
 228         ret = sev_issue_cmd_external_user(f.file, id, data, error);
 229
 230         fdput(f);
 231         return ret;
 232 }
 233
 234 static int sev_issue_cmd(struct kvm *kvm, int id, void *data, int *error)
 235 {
 236         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 237
 238         return __sev_issue_cmd(sev->fd, id, data, error);
 239 }
 240
 241 static int sev_launch_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
 242 {
 243         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 244         struct sev_data_launch_start start;
 245         struct kvm_sev_launch_start params;
 246         void *dh_blob, *session_blob;
 247         int *error = &argp->error;
 248         int ret;
 249
 250         if (!sev_guest(kvm))
 251                 return -ENOTTY;
 252
 253         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data, sizeof(params)))
 254                 return -EFAULT;
 255
 256         memset(&start, 0, sizeof(start));
 257
 258         dh_blob = NULL;
 259         if (params.dh_uaddr) {
 260                 dh_blob = psp_copy_user_blob(params.dh_uaddr, params.dh_len);
 261                 if (IS_ERR(dh_blob))
 262                         return PTR_ERR(dh_blob);
 263
 264                 start.dh_cert_address = __sme_set(__pa(dh_blob));
 265                 start.dh_cert_len = params.dh_len;
 266         }
 267
 268         session_blob = NULL;
 269         if (params.session_uaddr) {
 270                 session_blob = psp_copy_user_blob(params.session_uaddr, params.session_len);
 271                 if (IS_ERR(session_blob)) {
 272                         ret = PTR_ERR(session_blob);
 273                         goto e_free_dh;
 274                 }
 275
 276                 start.session_address = __sme_set(__pa(session_blob));
 277                 start.session_len = params.session_len;
 278         }
 279
 280         start.handle = params.handle;
 281         start.policy = params.policy;
 282
 283         /* create memory encryption context */
 284         ret = __sev_issue_cmd(argp->sev_fd, SEV_CMD_LAUNCH_START, &start, error);
 285         if (ret)
 286                 goto e_free_session;
 287
 288         /* Bind ASID to this guest */
 289         ret = sev_bind_asid(kvm, start.handle, error);
 290         if (ret)
 291                 goto e_free_session;
 292
 293         /* return handle to userspace */
 294         params.handle = start.handle;
 295         if (copy_to_user((void __user *)(uintptr_t)argp->data, &params, sizeof(params))) {
 296                 sev_unbind_asid(kvm, start.handle);
 297                 ret = -EFAULT;
 298                 goto e_free_session;
 299         }
 300
 301         sev->handle = start.handle;
 302         sev->fd = argp->sev_fd;
 303
 304 e_free_session:
 305         kfree(session_blob);
 306 e_free_dh:
 307         kfree(dh_blob);
 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         lockdep_assert_held(&kvm->lock);
 324
 325         if (ulen == 0 || uaddr + ulen < uaddr)
 326                 return ERR_PTR(-EINVAL);
 327
 328         /* Calculate number of pages. */
 329         first = (uaddr & PAGE_MASK) >> PAGE_SHIFT;
 330         last = ((uaddr + ulen - 1) & PAGE_MASK) >> PAGE_SHIFT;
 331         npages = (last - first + 1);
 332
 333         locked = sev->pages_locked + npages;
 334         lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
 335         if (locked > lock_limit && !capable(CAP_IPC_LOCK)) {
 336                 pr_err("SEV: %lu locked pages exceed the lock limit of %lu.\n", locked, lock_limit);
 337                 return ERR_PTR(-ENOMEM);
 338         }
 339
 340         if (WARN_ON_ONCE(npages > INT_MAX))
 341                 return ERR_PTR(-EINVAL);
 342
 343         /* Avoid using vmalloc for smaller buffers. */
 344         size = npages * sizeof(struct page *);
 345         if (size > PAGE_SIZE)
 346                 pages = __vmalloc(size, GFP_KERNEL_ACCOUNT | __GFP_ZERO);
 347         else
 348                 pages = kmalloc(size, GFP_KERNEL_ACCOUNT);
 349
 350         if (!pages)
 351                 return ERR_PTR(-ENOMEM);
 352
 353         /* Pin the user virtual address. */
 354         npinned = pin_user_pages_fast(uaddr, npages, write ? FOLL_WRITE : 0, pages);
 355         if (npinned != npages) {
 356                 pr_err("SEV: Failure locking %lu pages.\n", npages);
 357                 ret = -ENOMEM;
 358                 goto err;
 359         }
 360
 361         *n = npages;
 362         sev->pages_locked = locked;
 363
 364         return pages;
 365
 366 err:
 367         if (npinned > 0)
 368                 unpin_user_pages(pages, npinned);
 369
 370         kvfree(pages);
 371         return ERR_PTR(ret);
 372 }
 373
 374 static void sev_unpin_memory(struct kvm *kvm, struct page **pages,
 375                              unsigned long npages)
 376 {
 377         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 378
 379         unpin_user_pages(pages, npages);
 380         kvfree(pages);
 381         sev->pages_locked -= npages;
 382 }
 383
 384 static void sev_clflush_pages(struct page *pages[], unsigned long npages)
 385 {
 386         uint8_t *page_virtual;
 387         unsigned long i;
 388
 389         if (this_cpu_has(X86_FEATURE_SME_COHERENT) || npages == 0 ||
 390             pages == NULL)
 391                 return;
 392
 393         for (i = 0; i < npages; i++) {
 394                 page_virtual = kmap_atomic(pages[i]);
 395                 clflush_cache_range(page_virtual, PAGE_SIZE);
 396                 kunmap_atomic(page_virtual);
 397         }
 398 }
 399
 400 static unsigned long get_num_contig_pages(unsigned long idx,
 401                                 struct page **inpages, unsigned long npages)
 402 {
 403         unsigned long paddr, next_paddr;
 404         unsigned long i = idx + 1, pages = 1;
 405
 406         /* find the number of contiguous pages starting from idx */
 407         paddr = __sme_page_pa(inpages[idx]);
 408         while (i < npages) {
 409                 next_paddr = __sme_page_pa(inpages[i++]);
 410                 if ((paddr + PAGE_SIZE) == next_paddr) {
 411                         pages++;
 412                         paddr = next_paddr;
 413                         continue;
 414                 }
 415                 break;
 416         }
 417
 418         return pages;
 419 }
 420
 421 static int sev_launch_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
 422 {
 423         unsigned long vaddr, vaddr_end, next_vaddr, npages, pages, size, i;
 424         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 425         struct kvm_sev_launch_update_data params;
 426         struct sev_data_launch_update_data data;
 427         struct page **inpages;
 428         int ret;
 429
 430         if (!sev_guest(kvm))
 431                 return -ENOTTY;
 432
 433         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data, sizeof(params)))
 434                 return -EFAULT;
 435
 436         vaddr = params.uaddr;
 437         size = params.len;
 438         vaddr_end = vaddr + size;
 439
 440         /* Lock the user memory. */
 441         inpages = sev_pin_memory(kvm, vaddr, size, &npages, 1);
 442         if (IS_ERR(inpages))
 443                 return PTR_ERR(inpages);
 444
 445         /*
 446          * Flush (on non-coherent CPUs) before LAUNCH_UPDATE encrypts pages in
 447          * place; the cache may contain the data that was written unencrypted.
 448          */
 449         sev_clflush_pages(inpages, npages);
 450
 451         data.reserved = 0;
 452         data.handle = sev->handle;
 453
 454         for (i = 0; vaddr < vaddr_end; vaddr = next_vaddr, i += pages) {
 455                 int offset, len;
 456
 457                 /*
 458                  * If the user buffer is not page-aligned, calculate the offset
 459                  * within the page.
 460                  */
 461                 offset = vaddr & (PAGE_SIZE - 1);
 462
 463                 /* Calculate the number of pages that can be encrypted in one go. */
 464                 pages = get_num_contig_pages(i, inpages, npages);
 465
 466                 len = min_t(size_t, ((pages * PAGE_SIZE) - offset), size);
 467
 468                 data.len = len;
 469                 data.address = __sme_page_pa(inpages[i]) + offset;
 470                 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_DATA, &data, &argp->error);
 471                 if (ret)
 472                         goto e_unpin;
 473
 474                 size -= len;
 475                 next_vaddr = vaddr + len;
 476         }
 477
 478 e_unpin:
 479         /* content of memory is updated, mark pages dirty */
 480         for (i = 0; i < npages; i++) {
 481                 set_page_dirty_lock(inpages[i]);
 482                 mark_page_accessed(inpages[i]);
 483         }
 484         /* unlock the user pages */
 485         sev_unpin_memory(kvm, inpages, npages);
 486         return ret;
 487 }
 488
 489 static int sev_es_sync_vmsa(struct vcpu_svm *svm)
 490 {
 491         struct vmcb_save_area *save = &svm->vmcb->save;
 492
 493         /* Check some debug related fields before encrypting the VMSA */
 494         if (svm->vcpu.guest_debug || (save->dr7 & ~DR7_FIXED_1))
 495                 return -EINVAL;
 496
 497         /* Sync registgers */
 498         save->rax = svm->vcpu.arch.regs[VCPU_REGS_RAX];
 499         save->rbx = svm->vcpu.arch.regs[VCPU_REGS_RBX];
 500         save->rcx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
 501         save->rdx = svm->vcpu.arch.regs[VCPU_REGS_RDX];
 502         save->rsp = svm->vcpu.arch.regs[VCPU_REGS_RSP];
 503         save->rbp = svm->vcpu.arch.regs[VCPU_REGS_RBP];
 504         save->rsi = svm->vcpu.arch.regs[VCPU_REGS_RSI];
 505         save->rdi = svm->vcpu.arch.regs[VCPU_REGS_RDI];
 506 #ifdef CONFIG_X86_64
 507         save->r8  = svm->vcpu.arch.regs[VCPU_REGS_R8];
 508         save->r9  = svm->vcpu.arch.regs[VCPU_REGS_R9];
 509         save->r10 = svm->vcpu.arch.regs[VCPU_REGS_R10];
 510         save->r11 = svm->vcpu.arch.regs[VCPU_REGS_R11];
 511         save->r12 = svm->vcpu.arch.regs[VCPU_REGS_R12];
 512         save->r13 = svm->vcpu.arch.regs[VCPU_REGS_R13];
 513         save->r14 = svm->vcpu.arch.regs[VCPU_REGS_R14];
 514         save->r15 = svm->vcpu.arch.regs[VCPU_REGS_R15];
 515 #endif
 516         save->rip = svm->vcpu.arch.regs[VCPU_REGS_RIP];
 517
 518         /* Sync some non-GPR registers before encrypting */
 519         save->xcr0 = svm->vcpu.arch.xcr0;
 520         save->pkru = svm->vcpu.arch.pkru;
 521         save->xss  = svm->vcpu.arch.ia32_xss;
 522
 523         /*
 524          * SEV-ES will use a VMSA that is pointed to by the VMCB, not
 525          * the traditional VMSA that is part of the VMCB. Copy the
 526          * traditional VMSA as it has been built so far (in prep
 527          * for LAUNCH_UPDATE_VMSA) to be the initial SEV-ES state.
 528          */
 529         memcpy(svm->vmsa, save, sizeof(*save));
 530
 531         return 0;
 532 }
 533
 534 static int sev_launch_update_vmsa(struct kvm *kvm, struct kvm_sev_cmd *argp)
 535 {
 536         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 537         struct sev_data_launch_update_vmsa vmsa;
 538         struct kvm_vcpu *vcpu;
 539         int i, ret;
 540
 541         if (!sev_es_guest(kvm))
 542                 return -ENOTTY;
 543
 544         vmsa.reserved = 0;
 545
 546         kvm_for_each_vcpu(i, vcpu, kvm) {
 547                 struct vcpu_svm *svm = to_svm(vcpu);
 548
 549                 /* Perform some pre-encryption checks against the VMSA */
 550                 ret = sev_es_sync_vmsa(svm);
 551                 if (ret)
 552                         return ret;
 553
 554                 /*
 555                  * The LAUNCH_UPDATE_VMSA command will perform in-place
 556                  * encryption of the VMSA memory content (i.e it will write
 557                  * the same memory region with the guest's key), so invalidate
 558                  * it first.
 559                  */
 560                 clflush_cache_range(svm->vmsa, PAGE_SIZE);
 561
 562                 vmsa.handle = sev->handle;
 563                 vmsa.address = __sme_pa(svm->vmsa);
 564                 vmsa.len = PAGE_SIZE;
 565                 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_VMSA, &vmsa,
 566                                     &argp->error);
 567                 if (ret)
 568                         return ret;
 569
 570                 svm->vcpu.arch.guest_state_protected = true;
 571         }
 572
 573         return 0;
 574 }
 575
 576 static int sev_launch_measure(struct kvm *kvm, struct kvm_sev_cmd *argp)
 577 {
 578         void __user *measure = (void __user *)(uintptr_t)argp->data;
 579         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 580         struct sev_data_launch_measure data;
 581         struct kvm_sev_launch_measure params;
 582         void __user *p = NULL;
 583         void *blob = NULL;
 584         int ret;
 585
 586         if (!sev_guest(kvm))
 587                 return -ENOTTY;
 588
 589         if (copy_from_user(&params, measure, sizeof(params)))
 590                 return -EFAULT;
 591
 592         memset(&data, 0, sizeof(data));
 593
 594         /* User wants to query the blob length */
 595         if (!params.len)
 596                 goto cmd;
 597
 598         p = (void __user *)(uintptr_t)params.uaddr;
 599         if (p) {
 600                 if (params.len > SEV_FW_BLOB_MAX_SIZE)
 601                         return -EINVAL;
 602
 603                 blob = kmalloc(params.len, GFP_KERNEL_ACCOUNT);
 604                 if (!blob)
 605                         return -ENOMEM;
 606
 607                 data.address = __psp_pa(blob);
 608                 data.len = params.len;
 609         }
 610
 611 cmd:
 612         data.handle = sev->handle;
 613         ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_MEASURE, &data, &argp->error);
 614
 615         /*
 616          * If we query the session length, FW responded with expected data.
 617          */
 618         if (!params.len)
 619                 goto done;
 620
 621         if (ret)
 622                 goto e_free_blob;
 623
 624         if (blob) {
 625                 if (copy_to_user(p, blob, params.len))
 626                         ret = -EFAULT;
 627         }
 628
 629 done:
 630         params.len = data.len;
 631         if (copy_to_user(measure, &params, sizeof(params)))
 632                 ret = -EFAULT;
 633 e_free_blob:
 634         kfree(blob);
 635         return ret;
 636 }
 637
 638 static int sev_launch_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
 639 {
 640         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 641         struct sev_data_launch_finish data;
 642
 643         if (!sev_guest(kvm))
 644                 return -ENOTTY;
 645
 646         data.handle = sev->handle;
 647         return sev_issue_cmd(kvm, SEV_CMD_LAUNCH_FINISH, &data, &argp->error);
 648 }
 649
 650 static int sev_guest_status(struct kvm *kvm, struct kvm_sev_cmd *argp)
 651 {
 652         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 653         struct kvm_sev_guest_status params;
 654         struct sev_data_guest_status data;
 655         int ret;
 656
 657         if (!sev_guest(kvm))
 658                 return -ENOTTY;
 659
 660         memset(&data, 0, sizeof(data));
 661
 662         data.handle = sev->handle;
 663         ret = sev_issue_cmd(kvm, SEV_CMD_GUEST_STATUS, &data, &argp->error);
 664         if (ret)
 665                 return ret;
 666
 667         params.policy = data.policy;
 668         params.state = data.state;
 669         params.handle = data.handle;
 670
 671         if (copy_to_user((void __user *)(uintptr_t)argp->data, &params, sizeof(params)))
 672                 ret = -EFAULT;
 673
 674         return ret;
 675 }
 676
 677 static int __sev_issue_dbg_cmd(struct kvm *kvm, unsigned long src,
 678                                unsigned long dst, int size,
 679                                int *error, bool enc)
 680 {
 681         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 682         struct sev_data_dbg data;
 683
 684         data.reserved = 0;
 685         data.handle = sev->handle;
 686         data.dst_addr = dst;
 687         data.src_addr = src;
 688         data.len = size;
 689
 690         return sev_issue_cmd(kvm,
 691                              enc ? SEV_CMD_DBG_ENCRYPT : SEV_CMD_DBG_DECRYPT,
 692                              &data, error);
 693 }
 694
 695 static int __sev_dbg_decrypt(struct kvm *kvm, unsigned long src_paddr,
 696                              unsigned long dst_paddr, int sz, int *err)
 697 {
 698         int offset;
 699
 700         /*
 701          * Its safe to read more than we are asked, caller should ensure that
 702          * destination has enough space.
 703          */
 704         offset = src_paddr & 15;
 705         src_paddr = round_down(src_paddr, 16);
 706         sz = round_up(sz + offset, 16);
 707
 708         return __sev_issue_dbg_cmd(kvm, src_paddr, dst_paddr, sz, err, false);
 709 }
 710
 711 static int __sev_dbg_decrypt_user(struct kvm *kvm, unsigned long paddr,
 712                                   unsigned long __user dst_uaddr,
 713                                   unsigned long dst_paddr,
 714                                   int size, int *err)
 715 {
 716         struct page *tpage = NULL;
 717         int ret, offset;
 718
 719         /* if inputs are not 16-byte then use intermediate buffer */
 720         if (!IS_ALIGNED(dst_paddr, 16) ||
 721             !IS_ALIGNED(paddr,     16) ||
 722             !IS_ALIGNED(size,      16)) {
 723                 tpage = (void *)alloc_page(GFP_KERNEL);
 724                 if (!tpage)
 725                         return -ENOMEM;
 726
 727                 dst_paddr = __sme_page_pa(tpage);
 728         }
 729
 730         ret = __sev_dbg_decrypt(kvm, paddr, dst_paddr, size, err);
 731         if (ret)
 732                 goto e_free;
 733
 734         if (tpage) {
 735                 offset = paddr & 15;
 736                 if (copy_to_user((void __user *)(uintptr_t)dst_uaddr,
 737                                  page_address(tpage) + offset, size))
 738                         ret = -EFAULT;
 739         }
 740
 741 e_free:
 742         if (tpage)
 743                 __free_page(tpage);
 744
 745         return ret;
 746 }
 747
 748 static int __sev_dbg_encrypt_user(struct kvm *kvm, unsigned long paddr,
 749                                   unsigned long __user vaddr,
 750                                   unsigned long dst_paddr,
 751                                   unsigned long __user dst_vaddr,
 752                                   int size, int *error)
 753 {
 754         struct page *src_tpage = NULL;
 755         struct page *dst_tpage = NULL;
 756         int ret, len = size;
 757
 758         /* If source buffer is not aligned then use an intermediate buffer */
 759         if (!IS_ALIGNED(vaddr, 16)) {
 760                 src_tpage = alloc_page(GFP_KERNEL);
 761                 if (!src_tpage)
 762                         return -ENOMEM;
 763
 764                 if (copy_from_user(page_address(src_tpage),
 765                                 (void __user *)(uintptr_t)vaddr, size)) {
 766                         __free_page(src_tpage);
 767                         return -EFAULT;
 768                 }
 769
 770                 paddr = __sme_page_pa(src_tpage);
 771         }
 772
 773         /*
 774          *  If destination buffer or length is not aligned then do read-modify-write:
 775          *   - decrypt destination in an intermediate buffer
 776          *   - copy the source buffer in an intermediate buffer
 777          *   - use the intermediate buffer as source buffer
 778          */
 779         if (!IS_ALIGNED(dst_vaddr, 16) || !IS_ALIGNED(size, 16)) {
 780                 int dst_offset;
 781
 782                 dst_tpage = alloc_page(GFP_KERNEL);
 783                 if (!dst_tpage) {
 784                         ret = -ENOMEM;
 785                         goto e_free;
 786                 }
 787
 788                 ret = __sev_dbg_decrypt(kvm, dst_paddr,
 789                                         __sme_page_pa(dst_tpage), size, error);
 790                 if (ret)
 791                         goto e_free;
 792
 793                 /*
 794                  *  If source is kernel buffer then use memcpy() otherwise
 795                  *  copy_from_user().
 796                  */
 797                 dst_offset = dst_paddr & 15;
 798
 799                 if (src_tpage)
 800                         memcpy(page_address(dst_tpage) + dst_offset,
 801                                page_address(src_tpage), size);
 802                 else {
 803                         if (copy_from_user(page_address(dst_tpage) + dst_offset,
 804                                            (void __user *)(uintptr_t)vaddr, size)) {
 805                                 ret = -EFAULT;
 806                                 goto e_free;
 807                         }
 808                 }
 809
 810                 paddr = __sme_page_pa(dst_tpage);
 811                 dst_paddr = round_down(dst_paddr, 16);
 812                 len = round_up(size, 16);
 813         }
 814
 815         ret = __sev_issue_dbg_cmd(kvm, paddr, dst_paddr, len, error, true);
 816
 817 e_free:
 818         if (src_tpage)
 819                 __free_page(src_tpage);
 820         if (dst_tpage)
 821                 __free_page(dst_tpage);
 822         return ret;
 823 }
 824
 825 static int sev_dbg_crypt(struct kvm *kvm, struct kvm_sev_cmd *argp, bool dec)
 826 {
 827         unsigned long vaddr, vaddr_end, next_vaddr;
 828         unsigned long dst_vaddr;
 829         struct page **src_p, **dst_p;
 830         struct kvm_sev_dbg debug;
 831         unsigned long n;
 832         unsigned int size;
 833         int ret;
 834
 835         if (!sev_guest(kvm))
 836                 return -ENOTTY;
 837
 838         if (copy_from_user(&debug, (void __user *)(uintptr_t)argp->data, sizeof(debug)))
 839                 return -EFAULT;
 840
 841         if (!debug.len || debug.src_uaddr + debug.len < debug.src_uaddr)
 842                 return -EINVAL;
 843         if (!debug.dst_uaddr)
 844                 return -EINVAL;
 845
 846         vaddr = debug.src_uaddr;
 847         size = debug.len;
 848         vaddr_end = vaddr + size;
 849         dst_vaddr = debug.dst_uaddr;
 850
 851         for (; vaddr < vaddr_end; vaddr = next_vaddr) {
 852                 int len, s_off, d_off;
 853
 854                 /* lock userspace source and destination page */
 855                 src_p = sev_pin_memory(kvm, vaddr & PAGE_MASK, PAGE_SIZE, &n, 0);
 856                 if (IS_ERR(src_p))
 857                         return PTR_ERR(src_p);
 858
 859                 dst_p = sev_pin_memory(kvm, dst_vaddr & PAGE_MASK, PAGE_SIZE, &n, 1);
 860                 if (IS_ERR(dst_p)) {
 861                         sev_unpin_memory(kvm, src_p, n);
 862                         return PTR_ERR(dst_p);
 863                 }
 864
 865                 /*
 866                  * Flush (on non-coherent CPUs) before DBG_{DE,EN}CRYPT read or modify
 867                  * the pages; flush the destination too so that future accesses do not
 868                  * see stale data.
 869                  */
 870                 sev_clflush_pages(src_p, 1);
 871                 sev_clflush_pages(dst_p, 1);
 872
 873                 /*
 874                  * Since user buffer may not be page aligned, calculate the
 875                  * offset within the page.
 876                  */
 877                 s_off = vaddr & ~PAGE_MASK;
 878                 d_off = dst_vaddr & ~PAGE_MASK;
 879                 len = min_t(size_t, (PAGE_SIZE - s_off), size);
 880
 881                 if (dec)
 882                         ret = __sev_dbg_decrypt_user(kvm,
 883                                                      __sme_page_pa(src_p[0]) + s_off,
 884                                                      dst_vaddr,
 885                                                      __sme_page_pa(dst_p[0]) + d_off,
 886                                                      len, &argp->error);
 887                 else
 888                         ret = __sev_dbg_encrypt_user(kvm,
 889                                                      __sme_page_pa(src_p[0]) + s_off,
 890                                                      vaddr,
 891                                                      __sme_page_pa(dst_p[0]) + d_off,
 892                                                      dst_vaddr,
 893                                                      len, &argp->error);
 894
 895                 sev_unpin_memory(kvm, src_p, n);
 896                 sev_unpin_memory(kvm, dst_p, n);
 897
 898                 if (ret)
 899                         goto err;
 900
 901                 next_vaddr = vaddr + len;
 902                 dst_vaddr = dst_vaddr + len;
 903                 size -= len;
 904         }
 905 err:
 906         return ret;
 907 }
 908
 909 static int sev_launch_secret(struct kvm *kvm, struct kvm_sev_cmd *argp)
 910 {
 911         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 912         struct sev_data_launch_secret data;
 913         struct kvm_sev_launch_secret params;
 914         struct page **pages;
 915         void *blob, *hdr;
 916         unsigned long n, i;
 917         int ret, offset;
 918
 919         if (!sev_guest(kvm))
 920                 return -ENOTTY;
 921
 922         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data, sizeof(params)))
 923                 return -EFAULT;
 924
 925         pages = sev_pin_memory(kvm, params.guest_uaddr, params.guest_len, &n, 1);
 926         if (IS_ERR(pages))
 927                 return PTR_ERR(pages);
 928
 929         /*
 930          * Flush (on non-coherent CPUs) before LAUNCH_SECRET encrypts pages in
 931          * place; the cache may contain the data that was written unencrypted.
 932          */
 933         sev_clflush_pages(pages, n);
 934
 935         /*
 936          * The secret must be copied into contiguous memory region, lets verify
 937          * that userspace memory pages are contiguous before we issue command.
 938          */
 939         if (get_num_contig_pages(0, pages, n) != n) {
 940                 ret = -EINVAL;
 941                 goto e_unpin_memory;
 942         }
 943
 944         memset(&data, 0, sizeof(data));
 945
 946         offset = params.guest_uaddr & (PAGE_SIZE - 1);
 947         data.guest_address = __sme_page_pa(pages[0]) + offset;
 948         data.guest_len = params.guest_len;
 949
 950         blob = psp_copy_user_blob(params.trans_uaddr, params.trans_len);
 951         if (IS_ERR(blob)) {
 952                 ret = PTR_ERR(blob);
 953                 goto e_unpin_memory;
 954         }
 955
 956         data.trans_address = __psp_pa(blob);
 957         data.trans_len = params.trans_len;
 958
 959         hdr = psp_copy_user_blob(params.hdr_uaddr, params.hdr_len);
 960         if (IS_ERR(hdr)) {
 961                 ret = PTR_ERR(hdr);
 962                 goto e_free_blob;
 963         }
 964         data.hdr_address = __psp_pa(hdr);
 965         data.hdr_len = params.hdr_len;
 966
 967         data.handle = sev->handle;
 968         ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_SECRET, &data, &argp->error);
 969
 970         kfree(hdr);
 971
 972 e_free_blob:
 973         kfree(blob);
 974 e_unpin_memory:
 975         /* content of memory is updated, mark pages dirty */
 976         for (i = 0; i < n; i++) {
 977                 set_page_dirty_lock(pages[i]);
 978                 mark_page_accessed(pages[i]);
 979         }
 980         sev_unpin_memory(kvm, pages, n);
 981         return ret;
 982 }
 983
 984 static int sev_get_attestation_report(struct kvm *kvm, struct kvm_sev_cmd *argp)
 985 {
 986         void __user *report = (void __user *)(uintptr_t)argp->data;
 987         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 988         struct sev_data_attestation_report data;
 989         struct kvm_sev_attestation_report params;
 990         void __user *p;
 991         void *blob = NULL;
 992         int ret;
 993
 994         if (!sev_guest(kvm))
 995                 return -ENOTTY;
 996
 997         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data, sizeof(params)))
 998                 return -EFAULT;
 999
1000         memset(&data, 0, sizeof(data));
1001
1002         /* User wants to query the blob length */
1003         if (!params.len)
1004                 goto cmd;
1005
1006         p = (void __user *)(uintptr_t)params.uaddr;
1007         if (p) {
1008                 if (params.len > SEV_FW_BLOB_MAX_SIZE)
1009                         return -EINVAL;
1010
1011                 blob = kmalloc(params.len, GFP_KERNEL_ACCOUNT);
1012                 if (!blob)
1013                         return -ENOMEM;
1014
1015                 data.address = __psp_pa(blob);
1016                 data.len = params.len;
1017                 memcpy(data.mnonce, params.mnonce, sizeof(params.mnonce));
1018         }
1019 cmd:
1020         data.handle = sev->handle;
1021         ret = sev_issue_cmd(kvm, SEV_CMD_ATTESTATION_REPORT, &data, &argp->error);
1022         /*
1023          * If we query the session length, FW responded with expected data.
1024          */
1025         if (!params.len)
1026                 goto done;
1027
1028         if (ret)
1029                 goto e_free_blob;
1030
1031         if (blob) {
1032                 if (copy_to_user(p, blob, params.len))
1033                         ret = -EFAULT;
1034         }
1035
1036 done:
1037         params.len = data.len;
1038         if (copy_to_user(report, &params, sizeof(params)))
1039                 ret = -EFAULT;
1040 e_free_blob:
1041         kfree(blob);
1042         return ret;
1043 }
1044
1045 /* Userspace wants to query session length. */
1046 static int
1047 __sev_send_start_query_session_length(struct kvm *kvm, struct kvm_sev_cmd *argp,
1048                                       struct kvm_sev_send_start *params)
1049 {
1050         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1051         struct sev_data_send_start data;
1052         int ret;
1053
1054         data.handle = sev->handle;
1055         ret = sev_issue_cmd(kvm, SEV_CMD_SEND_START, &data, &argp->error);
1056         if (ret < 0)
1057                 return ret;
1058
1059         params->session_len = data.session_len;
1060         if (copy_to_user((void __user *)(uintptr_t)argp->data, params,
1061                                 sizeof(struct kvm_sev_send_start)))
1062                 ret = -EFAULT;
1063
1064         return ret;
1065 }
1066
1067 static int sev_send_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
1068 {
1069         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1070         struct sev_data_send_start data;
1071         struct kvm_sev_send_start params;
1072         void *amd_certs, *session_data;
1073         void *pdh_cert, *plat_certs;
1074         int ret;
1075
1076         if (!sev_guest(kvm))
1077                 return -ENOTTY;
1078
1079         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data,
1080                                 sizeof(struct kvm_sev_send_start)))
1081                 return -EFAULT;
1082
1083         /* if session_len is zero, userspace wants to query the session length */
1084         if (!params.session_len)
1085                 return __sev_send_start_query_session_length(kvm, argp,
1086                                 &params);
1087
1088         /* some sanity checks */
1089         if (!params.pdh_cert_uaddr || !params.pdh_cert_len ||
1090             !params.session_uaddr || params.session_len > SEV_FW_BLOB_MAX_SIZE)
1091                 return -EINVAL;
1092
1093         /* allocate the memory to hold the session data blob */
1094         session_data = kmalloc(params.session_len, GFP_KERNEL_ACCOUNT);
1095         if (!session_data)
1096                 return -ENOMEM;
1097
1098         /* copy the certificate blobs from userspace */
1099         pdh_cert = psp_copy_user_blob(params.pdh_cert_uaddr,
1100                                 params.pdh_cert_len);
1101         if (IS_ERR(pdh_cert)) {
1102                 ret = PTR_ERR(pdh_cert);
1103                 goto e_free_session;
1104         }
1105
1106         plat_certs = psp_copy_user_blob(params.plat_certs_uaddr,
1107                                 params.plat_certs_len);
1108         if (IS_ERR(plat_certs)) {
1109                 ret = PTR_ERR(plat_certs);
1110                 goto e_free_pdh;
1111         }
1112
1113         amd_certs = psp_copy_user_blob(params.amd_certs_uaddr,
1114                                 params.amd_certs_len);
1115         if (IS_ERR(amd_certs)) {
1116                 ret = PTR_ERR(amd_certs);
1117                 goto e_free_plat_cert;
1118         }
1119
1120         /* populate the FW SEND_START field with system physical address */
1121         memset(&data, 0, sizeof(data));
1122         data.pdh_cert_address = __psp_pa(pdh_cert);
1123         data.pdh_cert_len = params.pdh_cert_len;
1124         data.plat_certs_address = __psp_pa(plat_certs);
1125         data.plat_certs_len = params.plat_certs_len;
1126         data.amd_certs_address = __psp_pa(amd_certs);
1127         data.amd_certs_len = params.amd_certs_len;
1128         data.session_address = __psp_pa(session_data);
1129         data.session_len = params.session_len;
1130         data.handle = sev->handle;
1131
1132         ret = sev_issue_cmd(kvm, SEV_CMD_SEND_START, &data, &argp->error);
1133
1134         if (!ret && copy_to_user((void __user *)(uintptr_t)params.session_uaddr,
1135                         session_data, params.session_len)) {
1136                 ret = -EFAULT;
1137                 goto e_free_amd_cert;
1138         }
1139
1140         params.policy = data.policy;
1141         params.session_len = data.session_len;
1142         if (copy_to_user((void __user *)(uintptr_t)argp->data, &params,
1143                                 sizeof(struct kvm_sev_send_start)))
1144                 ret = -EFAULT;
1145
1146 e_free_amd_cert:
1147         kfree(amd_certs);
1148 e_free_plat_cert:
1149         kfree(plat_certs);
1150 e_free_pdh:
1151         kfree(pdh_cert);
1152 e_free_session:
1153         kfree(session_data);
1154         return ret;
1155 }
1156
1157 /* Userspace wants to query either header or trans length. */
1158 static int
1159 __sev_send_update_data_query_lengths(struct kvm *kvm, struct kvm_sev_cmd *argp,
1160                                      struct kvm_sev_send_update_data *params)
1161 {
1162         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1163         struct sev_data_send_update_data data;
1164         int ret;
1165
1166         data.handle = sev->handle;
1167         ret = sev_issue_cmd(kvm, SEV_CMD_SEND_UPDATE_DATA, &data, &argp->error);
1168         if (ret < 0)
1169                 return ret;
1170
1171         params->hdr_len = data.hdr_len;
1172         params->trans_len = data.trans_len;
1173
1174         if (copy_to_user((void __user *)(uintptr_t)argp->data, params,
1175                          sizeof(struct kvm_sev_send_update_data)))
1176                 ret = -EFAULT;
1177
1178         return ret;
1179 }
1180
1181 static int sev_send_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
1182 {
1183         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1184         struct sev_data_send_update_data data;
1185         struct kvm_sev_send_update_data params;
1186         void *hdr, *trans_data;
1187         struct page **guest_page;
1188         unsigned long n;
1189         int ret, offset;
1190
1191         if (!sev_guest(kvm))
1192                 return -ENOTTY;
1193
1194         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data,
1195                         sizeof(struct kvm_sev_send_update_data)))
1196                 return -EFAULT;
1197
1198         /* userspace wants to query either header or trans length */
1199         if (!params.trans_len || !params.hdr_len)
1200                 return __sev_send_update_data_query_lengths(kvm, argp, &params);
1201
1202         if (!params.trans_uaddr || !params.guest_uaddr ||
1203             !params.guest_len || !params.hdr_uaddr)
1204                 return -EINVAL;
1205
1206         /* Check if we are crossing the page boundary */
1207         offset = params.guest_uaddr & (PAGE_SIZE - 1);
1208         if ((params.guest_len + offset > PAGE_SIZE))
1209                 return -EINVAL;
1210
1211         /* Pin guest memory */
1212         guest_page = sev_pin_memory(kvm, params.guest_uaddr & PAGE_MASK,
1213                                     PAGE_SIZE, &n, 0);
1214         if (!guest_page)
1215                 return -EFAULT;
1216
1217         /* allocate memory for header and transport buffer */
1218         ret = -ENOMEM;
1219         hdr = kmalloc(params.hdr_len, GFP_KERNEL_ACCOUNT);
1220         if (!hdr)
1221                 goto e_unpin;
1222
1223         trans_data = kmalloc(params.trans_len, GFP_KERNEL_ACCOUNT);
1224         if (!trans_data)
1225                 goto e_free_hdr;
1226
1227         memset(&data, 0, sizeof(data));
1228         data.hdr_address = __psp_pa(hdr);
1229         data.hdr_len = params.hdr_len;
1230         data.trans_address = __psp_pa(trans_data);
1231         data.trans_len = params.trans_len;
1232
1233         /* The SEND_UPDATE_DATA command requires C-bit to be always set. */
1234         data.guest_address = (page_to_pfn(guest_page[0]) << PAGE_SHIFT) + offset;
1235         data.guest_address |= sev_me_mask;
1236         data.guest_len = params.guest_len;
1237         data.handle = sev->handle;
1238
1239         ret = sev_issue_cmd(kvm, SEV_CMD_SEND_UPDATE_DATA, &data, &argp->error);
1240
1241         if (ret)
1242                 goto e_free_trans_data;
1243
1244         /* copy transport buffer to user space */
1245         if (copy_to_user((void __user *)(uintptr_t)params.trans_uaddr,
1246                          trans_data, params.trans_len)) {
1247                 ret = -EFAULT;
1248                 goto e_free_trans_data;
1249         }
1250
1251         /* Copy packet header to userspace. */
1252         ret = copy_to_user((void __user *)(uintptr_t)params.hdr_uaddr, hdr,
1253                                 params.hdr_len);
1254
1255 e_free_trans_data:
1256         kfree(trans_data);
1257 e_free_hdr:
1258         kfree(hdr);
1259 e_unpin:
1260         sev_unpin_memory(kvm, guest_page, n);
1261
1262         return ret;
1263 }
1264
1265 static int sev_send_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
1266 {
1267         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1268         struct sev_data_send_finish data;
1269
1270         if (!sev_guest(kvm))
1271                 return -ENOTTY;
1272
1273         data.handle = sev->handle;
1274         return sev_issue_cmd(kvm, SEV_CMD_SEND_FINISH, &data, &argp->error);
1275 }
1276
1277 static int sev_send_cancel(struct kvm *kvm, struct kvm_sev_cmd *argp)
1278 {
1279         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1280         struct sev_data_send_cancel data;
1281
1282         if (!sev_guest(kvm))
1283                 return -ENOTTY;
1284
1285         data.handle = sev->handle;
1286         return sev_issue_cmd(kvm, SEV_CMD_SEND_CANCEL, &data, &argp->error);
1287 }
1288
1289 static int sev_receive_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
1290 {
1291         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1292         struct sev_data_receive_start start;
1293         struct kvm_sev_receive_start params;
1294         int *error = &argp->error;
1295         void *session_data;
1296         void *pdh_data;
1297         int ret;
1298
1299         if (!sev_guest(kvm))
1300                 return -ENOTTY;
1301
1302         /* Get parameter from the userspace */
1303         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data,
1304                         sizeof(struct kvm_sev_receive_start)))
1305                 return -EFAULT;
1306
1307         /* some sanity checks */
1308         if (!params.pdh_uaddr || !params.pdh_len ||
1309             !params.session_uaddr || !params.session_len)
1310                 return -EINVAL;
1311
1312         pdh_data = psp_copy_user_blob(params.pdh_uaddr, params.pdh_len);
1313         if (IS_ERR(pdh_data))
1314                 return PTR_ERR(pdh_data);
1315
1316         session_data = psp_copy_user_blob(params.session_uaddr,
1317                         params.session_len);
1318         if (IS_ERR(session_data)) {
1319                 ret = PTR_ERR(session_data);
1320                 goto e_free_pdh;
1321         }
1322
1323         memset(&start, 0, sizeof(start));
1324         start.handle = params.handle;
1325         start.policy = params.policy;
1326         start.pdh_cert_address = __psp_pa(pdh_data);
1327         start.pdh_cert_len = params.pdh_len;
1328         start.session_address = __psp_pa(session_data);
1329         start.session_len = params.session_len;
1330
1331         /* create memory encryption context */
1332         ret = __sev_issue_cmd(argp->sev_fd, SEV_CMD_RECEIVE_START, &start,
1333                                 error);
1334         if (ret)
1335                 goto e_free_session;
1336
1337         /* Bind ASID to this guest */
1338         ret = sev_bind_asid(kvm, start.handle, error);
1339         if (ret)
1340                 goto e_free_session;
1341
1342         params.handle = start.handle;
1343         if (copy_to_user((void __user *)(uintptr_t)argp->data,
1344                          &params, sizeof(struct kvm_sev_receive_start))) {
1345                 ret = -EFAULT;
1346                 sev_unbind_asid(kvm, start.handle);
1347                 goto e_free_session;
1348         }
1349
1350         sev->handle = start.handle;
1351         sev->fd = argp->sev_fd;
1352
1353 e_free_session:
1354         kfree(session_data);
1355 e_free_pdh:
1356         kfree(pdh_data);
1357
1358         return ret;
1359 }
1360
1361 static int sev_receive_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
1362 {
1363         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1364         struct kvm_sev_receive_update_data params;
1365         struct sev_data_receive_update_data data;
1366         void *hdr = NULL, *trans = NULL;
1367         struct page **guest_page;
1368         unsigned long n;
1369         int ret, offset;
1370
1371         if (!sev_guest(kvm))
1372                 return -EINVAL;
1373
1374         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data,
1375                         sizeof(struct kvm_sev_receive_update_data)))
1376                 return -EFAULT;
1377
1378         if (!params.hdr_uaddr || !params.hdr_len ||
1379             !params.guest_uaddr || !params.guest_len ||
1380             !params.trans_uaddr || !params.trans_len)
1381                 return -EINVAL;
1382
1383         /* Check if we are crossing the page boundary */
1384         offset = params.guest_uaddr & (PAGE_SIZE - 1);
1385         if ((params.guest_len + offset > PAGE_SIZE))
1386                 return -EINVAL;
1387
1388         hdr = psp_copy_user_blob(params.hdr_uaddr, params.hdr_len);
1389         if (IS_ERR(hdr))
1390                 return PTR_ERR(hdr);
1391
1392         trans = psp_copy_user_blob(params.trans_uaddr, params.trans_len);
1393         if (IS_ERR(trans)) {
1394                 ret = PTR_ERR(trans);
1395                 goto e_free_hdr;
1396         }
1397
1398         memset(&data, 0, sizeof(data));
1399         data.hdr_address = __psp_pa(hdr);
1400         data.hdr_len = params.hdr_len;
1401         data.trans_address = __psp_pa(trans);
1402         data.trans_len = params.trans_len;
1403
1404         /* Pin guest memory */
1405         ret = -EFAULT;
1406         guest_page = sev_pin_memory(kvm, params.guest_uaddr & PAGE_MASK,
1407                                     PAGE_SIZE, &n, 0);
1408         if (!guest_page)
1409                 goto e_free_trans;
1410
1411         /* The RECEIVE_UPDATE_DATA command requires C-bit to be always set. */
1412         data.guest_address = (page_to_pfn(guest_page[0]) << PAGE_SHIFT) + offset;
1413         data.guest_address |= sev_me_mask;
1414         data.guest_len = params.guest_len;
1415         data.handle = sev->handle;
1416
1417         ret = sev_issue_cmd(kvm, SEV_CMD_RECEIVE_UPDATE_DATA, &data,
1418                                 &argp->error);
1419
1420         sev_unpin_memory(kvm, guest_page, n);
1421
1422 e_free_trans:
1423         kfree(trans);
1424 e_free_hdr:
1425         kfree(hdr);
1426
1427         return ret;
1428 }
1429
1430 static int sev_receive_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
1431 {
1432         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1433         struct sev_data_receive_finish data;
1434
1435         if (!sev_guest(kvm))
1436                 return -ENOTTY;
1437
1438         data.handle = sev->handle;
1439         return sev_issue_cmd(kvm, SEV_CMD_RECEIVE_FINISH, &data, &argp->error);
1440 }
1441
1442 int svm_mem_enc_op(struct kvm *kvm, void __user *argp)
1443 {
1444         struct kvm_sev_cmd sev_cmd;
1445         int r;
1446
1447         if (!svm_sev_enabled() || !sev)
1448                 return -ENOTTY;
1449
1450         if (!argp)
1451                 return 0;
1452
1453         if (copy_from_user(&sev_cmd, argp, sizeof(struct kvm_sev_cmd)))
1454                 return -EFAULT;
1455
1456         mutex_lock(&kvm->lock);
1457
1458         /* enc_context_owner handles all memory enc operations */
1459         if (is_mirroring_enc_context(kvm)) {
1460                 r = -EINVAL;
1461                 goto out;
1462         }
1463
1464         switch (sev_cmd.id) {
1465         case KVM_SEV_ES_INIT:
1466                 if (!sev_es) {
1467                         r = -ENOTTY;
1468                         goto out;
1469                 }
1470                 fallthrough;
1471         case KVM_SEV_INIT:
1472                 r = sev_guest_init(kvm, &sev_cmd);
1473                 break;
1474         case KVM_SEV_LAUNCH_START:
1475                 r = sev_launch_start(kvm, &sev_cmd);
1476                 break;
1477         case KVM_SEV_LAUNCH_UPDATE_DATA:
1478                 r = sev_launch_update_data(kvm, &sev_cmd);
1479                 break;
1480         case KVM_SEV_LAUNCH_UPDATE_VMSA:
1481                 r = sev_launch_update_vmsa(kvm, &sev_cmd);
1482                 break;
1483         case KVM_SEV_LAUNCH_MEASURE:
1484                 r = sev_launch_measure(kvm, &sev_cmd);
1485                 break;
1486         case KVM_SEV_LAUNCH_FINISH:
1487                 r = sev_launch_finish(kvm, &sev_cmd);
1488                 break;
1489         case KVM_SEV_GUEST_STATUS:
1490                 r = sev_guest_status(kvm, &sev_cmd);
1491                 break;
1492         case KVM_SEV_DBG_DECRYPT:
1493                 r = sev_dbg_crypt(kvm, &sev_cmd, true);
1494                 break;
1495         case KVM_SEV_DBG_ENCRYPT:
1496                 r = sev_dbg_crypt(kvm, &sev_cmd, false);
1497                 break;
1498         case KVM_SEV_LAUNCH_SECRET:
1499                 r = sev_launch_secret(kvm, &sev_cmd);
1500                 break;
1501         case KVM_SEV_GET_ATTESTATION_REPORT:
1502                 r = sev_get_attestation_report(kvm, &sev_cmd);
1503                 break;
1504         case KVM_SEV_SEND_START:
1505                 r = sev_send_start(kvm, &sev_cmd);
1506                 break;
1507         case KVM_SEV_SEND_UPDATE_DATA:
1508                 r = sev_send_update_data(kvm, &sev_cmd);
1509                 break;
1510         case KVM_SEV_SEND_FINISH:
1511                 r = sev_send_finish(kvm, &sev_cmd);
1512                 break;
1513         case KVM_SEV_SEND_CANCEL:
1514                 r = sev_send_cancel(kvm, &sev_cmd);
1515                 break;
1516         case KVM_SEV_RECEIVE_START:
1517                 r = sev_receive_start(kvm, &sev_cmd);
1518                 break;
1519         case KVM_SEV_RECEIVE_UPDATE_DATA:
1520                 r = sev_receive_update_data(kvm, &sev_cmd);
1521                 break;
1522         case KVM_SEV_RECEIVE_FINISH:
1523                 r = sev_receive_finish(kvm, &sev_cmd);
1524                 break;
1525         default:
1526                 r = -EINVAL;
1527                 goto out;
1528         }
1529
1530         if (copy_to_user(argp, &sev_cmd, sizeof(struct kvm_sev_cmd)))
1531                 r = -EFAULT;
1532
1533 out:
1534         mutex_unlock(&kvm->lock);
1535         return r;
1536 }
1537
1538 int svm_register_enc_region(struct kvm *kvm,
1539                             struct kvm_enc_region *range)
1540 {
1541         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1542         struct enc_region *region;
1543         int ret = 0;
1544
1545         if (!sev_guest(kvm))
1546                 return -ENOTTY;
1547
1548         /* If kvm is mirroring encryption context it isn't responsible for it */
1549         if (is_mirroring_enc_context(kvm))
1550                 return -EINVAL;
1551
1552         if (range->addr > ULONG_MAX || range->size > ULONG_MAX)
1553                 return -EINVAL;
1554
1555         region = kzalloc(sizeof(*region), GFP_KERNEL_ACCOUNT);
1556         if (!region)
1557                 return -ENOMEM;
1558
1559         mutex_lock(&kvm->lock);
1560         region->pages = sev_pin_memory(kvm, range->addr, range->size, &region->npages, 1);
1561         if (IS_ERR(region->pages)) {
1562                 ret = PTR_ERR(region->pages);
1563                 mutex_unlock(&kvm->lock);
1564                 goto e_free;
1565         }
1566
1567         region->uaddr = range->addr;
1568         region->size = range->size;
1569
1570         list_add_tail(&region->list, &sev->regions_list);
1571         mutex_unlock(&kvm->lock);
1572
1573         /*
1574          * The guest may change the memory encryption attribute from C=0 -> C=1
1575          * or vice versa for this memory range. Lets make sure caches are
1576          * flushed to ensure that guest data gets written into memory with
1577          * correct C-bit.
1578          */
1579         sev_clflush_pages(region->pages, region->npages);
1580
1581         return ret;
1582
1583 e_free:
1584         kfree(region);
1585         return ret;
1586 }
1587
1588 static struct enc_region *
1589 find_enc_region(struct kvm *kvm, struct kvm_enc_region *range)
1590 {
1591         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1592         struct list_head *head = &sev->regions_list;
1593         struct enc_region *i;
1594
1595         list_for_each_entry(i, head, list) {
1596                 if (i->uaddr == range->addr &&
1597                     i->size == range->size)
1598                         return i;
1599         }
1600
1601         return NULL;
1602 }
1603
1604 static void __unregister_enc_region_locked(struct kvm *kvm,
1605                                            struct enc_region *region)
1606 {
1607         sev_unpin_memory(kvm, region->pages, region->npages);
1608         list_del(&region->list);
1609         kfree(region);
1610 }
1611
1612 int svm_unregister_enc_region(struct kvm *kvm,
1613                               struct kvm_enc_region *range)
1614 {
1615         struct enc_region *region;
1616         int ret;
1617
1618         /* If kvm is mirroring encryption context it isn't responsible for it */
1619         if (is_mirroring_enc_context(kvm))
1620                 return -EINVAL;
1621
1622         mutex_lock(&kvm->lock);
1623
1624         if (!sev_guest(kvm)) {
1625                 ret = -ENOTTY;
1626                 goto failed;
1627         }
1628
1629         region = find_enc_region(kvm, range);
1630         if (!region) {
1631                 ret = -EINVAL;
1632                 goto failed;
1633         }
1634
1635         /*
1636          * Ensure that all guest tagged cache entries are flushed before
1637          * releasing the pages back to the system for use. CLFLUSH will
1638          * not do this, so issue a WBINVD.
1639          */
1640         wbinvd_on_all_cpus();
1641
1642         __unregister_enc_region_locked(kvm, region);
1643
1644         mutex_unlock(&kvm->lock);
1645         return 0;
1646
1647 failed:
1648         mutex_unlock(&kvm->lock);
1649         return ret;
1650 }
1651
1652 int svm_vm_copy_asid_from(struct kvm *kvm, unsigned int source_fd)
1653 {
1654         struct file *source_kvm_file;
1655         struct kvm *source_kvm;
1656         struct kvm_sev_info *mirror_sev;
1657         unsigned int asid;
1658         int ret;
1659
1660         source_kvm_file = fget(source_fd);
1661         if (!file_is_kvm(source_kvm_file)) {
1662                 ret = -EBADF;
1663                 goto e_source_put;
1664         }
1665
1666         source_kvm = source_kvm_file->private_data;
1667         mutex_lock(&source_kvm->lock);
1668
1669         if (!sev_guest(source_kvm)) {
1670                 ret = -EINVAL;
1671                 goto e_source_unlock;
1672         }
1673
1674         /* Mirrors of mirrors should work, but let's not get silly */
1675         if (is_mirroring_enc_context(source_kvm) || source_kvm == kvm) {
1676                 ret = -EINVAL;
1677                 goto e_source_unlock;
1678         }
1679
1680         asid = to_kvm_svm(source_kvm)->sev_info.asid;
1681
1682         /*
1683          * The mirror kvm holds an enc_context_owner ref so its asid can't
1684          * disappear until we're done with it
1685          */
1686         kvm_get_kvm(source_kvm);
1687
1688         fput(source_kvm_file);
1689         mutex_unlock(&source_kvm->lock);
1690         mutex_lock(&kvm->lock);
1691
1692         if (sev_guest(kvm)) {
1693                 ret = -EINVAL;
1694                 goto e_mirror_unlock;
1695         }
1696
1697         /* Set enc_context_owner and copy its encryption context over */
1698         mirror_sev = &to_kvm_svm(kvm)->sev_info;
1699         mirror_sev->enc_context_owner = source_kvm;
1700         mirror_sev->asid = asid;
1701         mirror_sev->active = true;
1702
1703         mutex_unlock(&kvm->lock);
1704         return 0;
1705
1706 e_mirror_unlock:
1707         mutex_unlock(&kvm->lock);
1708         kvm_put_kvm(source_kvm);
1709         return ret;
1710 e_source_unlock:
1711         mutex_unlock(&source_kvm->lock);
1712 e_source_put:
1713         fput(source_kvm_file);
1714         return ret;
1715 }
1716
1717 void sev_vm_destroy(struct kvm *kvm)
1718 {
1719         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1720         struct list_head *head = &sev->regions_list;
1721         struct list_head *pos, *q;
1722
1723         if (!sev_guest(kvm))
1724                 return;
1725
1726         /* If this is a mirror_kvm release the enc_context_owner and skip sev cleanup */
1727         if (is_mirroring_enc_context(kvm)) {
1728                 kvm_put_kvm(sev->enc_context_owner);
1729                 return;
1730         }
1731
1732         mutex_lock(&kvm->lock);
1733
1734         /*
1735          * Ensure that all guest tagged cache entries are flushed before
1736          * releasing the pages back to the system for use. CLFLUSH will
1737          * not do this, so issue a WBINVD.
1738          */
1739         wbinvd_on_all_cpus();
1740
1741         /*
1742          * if userspace was terminated before unregistering the memory regions
1743          * then lets unpin all the registered memory.
1744          */
1745         if (!list_empty(head)) {
1746                 list_for_each_safe(pos, q, head) {
1747                         __unregister_enc_region_locked(kvm,
1748                                 list_entry(pos, struct enc_region, list));
1749                         cond_resched();
1750                 }
1751         }
1752
1753         mutex_unlock(&kvm->lock);
1754
1755         sev_unbind_asid(kvm, sev->handle);
1756         sev_asid_free(sev->asid);
1757 }
1758
1759 void __init sev_hardware_setup(void)
1760 {
1761         unsigned int eax, ebx, ecx, edx;
1762         bool sev_es_supported = false;
1763         bool sev_supported = false;
1764
1765         /* Does the CPU support SEV? */
1766         if (!boot_cpu_has(X86_FEATURE_SEV))
1767                 goto out;
1768
1769         /* Retrieve SEV CPUID information */
1770         cpuid(0x8000001f, &eax, &ebx, &ecx, &edx);
1771
1772         /* Set encryption bit location for SEV-ES guests */
1773         sev_enc_bit = ebx & 0x3f;
1774
1775         /* Maximum number of encrypted guests supported simultaneously */
1776         max_sev_asid = ecx;
1777
1778         if (!svm_sev_enabled())
1779                 goto out;
1780
1781         /* Minimum ASID value that should be used for SEV guest */
1782         min_sev_asid = edx;
1783         sev_me_mask = 1UL << (ebx & 0x3f);
1784
1785         /* Initialize SEV ASID bitmaps */
1786         sev_asid_bitmap = bitmap_zalloc(max_sev_asid, GFP_KERNEL);
1787         if (!sev_asid_bitmap)
1788                 goto out;
1789
1790         sev_reclaim_asid_bitmap = bitmap_zalloc(max_sev_asid, GFP_KERNEL);
1791         if (!sev_reclaim_asid_bitmap)
1792                 goto out;
1793
1794         pr_info("SEV supported: %u ASIDs\n", max_sev_asid - min_sev_asid + 1);
1795         sev_supported = true;
1796
1797         /* SEV-ES support requested? */
1798         if (!sev_es)
1799                 goto out;
1800
1801         /* Does the CPU support SEV-ES? */
1802         if (!boot_cpu_has(X86_FEATURE_SEV_ES))
1803                 goto out;
1804
1805         /* Has the system been allocated ASIDs for SEV-ES? */
1806         if (min_sev_asid == 1)
1807                 goto out;
1808
1809         pr_info("SEV-ES supported: %u ASIDs\n", min_sev_asid - 1);
1810         sev_es_supported = true;
1811
1812 out:
1813         sev = sev_supported;
1814         sev_es = sev_es_supported;
1815 }
1816
1817 void sev_hardware_teardown(void)
1818 {
1819         if (!svm_sev_enabled())
1820                 return;
1821
1822         bitmap_free(sev_asid_bitmap);
1823         bitmap_free(sev_reclaim_asid_bitmap);
1824
1825         sev_flush_asids();
1826 }
1827
1828 /*
1829  * Pages used by hardware to hold guest encrypted state must be flushed before
1830  * returning them to the system.
1831  */
1832 static void sev_flush_guest_memory(struct vcpu_svm *svm, void *va,
1833                                    unsigned long len)
1834 {
1835         /*
1836          * If hardware enforced cache coherency for encrypted mappings of the
1837          * same physical page is supported, nothing to do.
1838          */
1839         if (boot_cpu_has(X86_FEATURE_SME_COHERENT))
1840                 return;
1841
1842         /*
1843          * If the VM Page Flush MSR is supported, use it to flush the page
1844          * (using the page virtual address and the guest ASID).
1845          */
1846         if (boot_cpu_has(X86_FEATURE_VM_PAGE_FLUSH)) {
1847                 struct kvm_sev_info *sev;
1848                 unsigned long va_start;
1849                 u64 start, stop;
1850
1851                 /* Align start and stop to page boundaries. */
1852                 va_start = (unsigned long)va;
1853                 start = (u64)va_start & PAGE_MASK;
1854                 stop = PAGE_ALIGN((u64)va_start + len);
1855
1856                 if (start < stop) {
1857                         sev = &to_kvm_svm(svm->vcpu.kvm)->sev_info;
1858
1859                         while (start < stop) {
1860                                 wrmsrl(MSR_AMD64_VM_PAGE_FLUSH,
1861                                        start | sev->asid);
1862
1863                                 start += PAGE_SIZE;
1864                         }
1865
1866                         return;
1867                 }
1868
1869                 WARN(1, "Address overflow, using WBINVD\n");
1870         }
1871
1872         /*
1873          * Hardware should always have one of the above features,
1874          * but if not, use WBINVD and issue a warning.
1875          */
1876         WARN_ONCE(1, "Using WBINVD to flush guest memory\n");
1877         wbinvd_on_all_cpus();
1878 }
1879
1880 void sev_free_vcpu(struct kvm_vcpu *vcpu)
1881 {
1882         struct vcpu_svm *svm;
1883
1884         if (!sev_es_guest(vcpu->kvm))
1885                 return;
1886
1887         svm = to_svm(vcpu);
1888
1889         if (vcpu->arch.guest_state_protected)
1890                 sev_flush_guest_memory(svm, svm->vmsa, PAGE_SIZE);
1891         __free_page(virt_to_page(svm->vmsa));
1892
1893         if (svm->ghcb_sa_free)
1894                 kfree(svm->ghcb_sa);
1895 }
1896
1897 static void dump_ghcb(struct vcpu_svm *svm)
1898 {
1899         struct ghcb *ghcb = svm->ghcb;
1900         unsigned int nbits;
1901
1902         /* Re-use the dump_invalid_vmcb module parameter */
1903         if (!dump_invalid_vmcb) {
1904                 pr_warn_ratelimited("set kvm_amd.dump_invalid_vmcb=1 to dump internal KVM state.\n");
1905                 return;
1906         }
1907
1908         nbits = sizeof(ghcb->save.valid_bitmap) * 8;
1909
1910         pr_err("GHCB (GPA=%016llx):\n", svm->vmcb->control.ghcb_gpa);
1911         pr_err("%-20s%016llx is_valid: %u\n", "sw_exit_code",
1912                ghcb->save.sw_exit_code, ghcb_sw_exit_code_is_valid(ghcb));
1913         pr_err("%-20s%016llx is_valid: %u\n", "sw_exit_info_1",
1914                ghcb->save.sw_exit_info_1, ghcb_sw_exit_info_1_is_valid(ghcb));
1915         pr_err("%-20s%016llx is_valid: %u\n", "sw_exit_info_2",
1916                ghcb->save.sw_exit_info_2, ghcb_sw_exit_info_2_is_valid(ghcb));
1917         pr_err("%-20s%016llx is_valid: %u\n", "sw_scratch",
1918                ghcb->save.sw_scratch, ghcb_sw_scratch_is_valid(ghcb));
1919         pr_err("%-20s%*pb\n", "valid_bitmap", nbits, ghcb->save.valid_bitmap);
1920 }
1921
1922 static void sev_es_sync_to_ghcb(struct vcpu_svm *svm)
1923 {
1924         struct kvm_vcpu *vcpu = &svm->vcpu;
1925         struct ghcb *ghcb = svm->ghcb;
1926
1927         /*
1928          * The GHCB protocol so far allows for the following data
1929          * to be returned:
1930          *   GPRs RAX, RBX, RCX, RDX
1931          *
1932          * Copy their values, even if they may not have been written during the
1933          * VM-Exit.  It's the guest's responsibility to not consume random data.
1934          */
1935         ghcb_set_rax(ghcb, vcpu->arch.regs[VCPU_REGS_RAX]);
1936         ghcb_set_rbx(ghcb, vcpu->arch.regs[VCPU_REGS_RBX]);
1937         ghcb_set_rcx(ghcb, vcpu->arch.regs[VCPU_REGS_RCX]);
1938         ghcb_set_rdx(ghcb, vcpu->arch.regs[VCPU_REGS_RDX]);
1939 }
1940
1941 static void sev_es_sync_from_ghcb(struct vcpu_svm *svm)
1942 {
1943         struct vmcb_control_area *control = &svm->vmcb->control;
1944         struct kvm_vcpu *vcpu = &svm->vcpu;
1945         struct ghcb *ghcb = svm->ghcb;
1946         u64 exit_code;
1947
1948         /*
1949          * The GHCB protocol so far allows for the following data
1950          * to be supplied:
1951          *   GPRs RAX, RBX, RCX, RDX
1952          *   XCR0
1953          *   CPL
1954          *
1955          * VMMCALL allows the guest to provide extra registers. KVM also
1956          * expects RSI for hypercalls, so include that, too.
1957          *
1958          * Copy their values to the appropriate location if supplied.
1959          */
1960         memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
1961
1962         vcpu->arch.regs[VCPU_REGS_RAX] = ghcb_get_rax_if_valid(ghcb);
1963         vcpu->arch.regs[VCPU_REGS_RBX] = ghcb_get_rbx_if_valid(ghcb);
1964         vcpu->arch.regs[VCPU_REGS_RCX] = ghcb_get_rcx_if_valid(ghcb);
1965         vcpu->arch.regs[VCPU_REGS_RDX] = ghcb_get_rdx_if_valid(ghcb);
1966         vcpu->arch.regs[VCPU_REGS_RSI] = ghcb_get_rsi_if_valid(ghcb);
1967
1968         svm->vmcb->save.cpl = ghcb_get_cpl_if_valid(ghcb);
1969
1970         if (ghcb_xcr0_is_valid(ghcb)) {
1971                 vcpu->arch.xcr0 = ghcb_get_xcr0(ghcb);
1972                 kvm_update_cpuid_runtime(vcpu);
1973         }
1974
1975         /* Copy the GHCB exit information into the VMCB fields */
1976         exit_code = ghcb_get_sw_exit_code(ghcb);
1977         control->exit_code = lower_32_bits(exit_code);
1978         control->exit_code_hi = upper_32_bits(exit_code);
1979         control->exit_info_1 = ghcb_get_sw_exit_info_1(ghcb);
1980         control->exit_info_2 = ghcb_get_sw_exit_info_2(ghcb);
1981
1982         /* Clear the valid entries fields */
1983         memset(ghcb->save.valid_bitmap, 0, sizeof(ghcb->save.valid_bitmap));
1984 }
1985
1986 static int sev_es_validate_vmgexit(struct vcpu_svm *svm)
1987 {
1988         struct kvm_vcpu *vcpu;
1989         struct ghcb *ghcb;
1990         u64 exit_code = 0;
1991
1992         ghcb = svm->ghcb;
1993
1994         /* Only GHCB Usage code 0 is supported */
1995         if (ghcb->ghcb_usage)
1996                 goto vmgexit_err;
1997
1998         /*
1999          * Retrieve the exit code now even though is may not be marked valid
2000          * as it could help with debugging.
2001          */
2002         exit_code = ghcb_get_sw_exit_code(ghcb);
2003
2004         if (!ghcb_sw_exit_code_is_valid(ghcb) ||
2005             !ghcb_sw_exit_info_1_is_valid(ghcb) ||
2006             !ghcb_sw_exit_info_2_is_valid(ghcb))
2007                 goto vmgexit_err;
2008
2009         switch (ghcb_get_sw_exit_code(ghcb)) {
2010         case SVM_EXIT_READ_DR7:
2011                 break;
2012         case SVM_EXIT_WRITE_DR7:
2013                 if (!ghcb_rax_is_valid(ghcb))
2014                         goto vmgexit_err;
2015                 break;
2016         case SVM_EXIT_RDTSC:
2017                 break;
2018         case SVM_EXIT_RDPMC:
2019                 if (!ghcb_rcx_is_valid(ghcb))
2020                         goto vmgexit_err;
2021                 break;
2022         case SVM_EXIT_CPUID:
2023                 if (!ghcb_rax_is_valid(ghcb) ||
2024                     !ghcb_rcx_is_valid(ghcb))
2025                         goto vmgexit_err;
2026                 if (ghcb_get_rax(ghcb) == 0xd)
2027                         if (!ghcb_xcr0_is_valid(ghcb))
2028                                 goto vmgexit_err;
2029                 break;
2030         case SVM_EXIT_INVD:
2031                 break;
2032         case SVM_EXIT_IOIO:
2033                 if (ghcb_get_sw_exit_info_1(ghcb) & SVM_IOIO_STR_MASK) {
2034                         if (!ghcb_sw_scratch_is_valid(ghcb))
2035                                 goto vmgexit_err;
2036                 } else {
2037                         if (!(ghcb_get_sw_exit_info_1(ghcb) & SVM_IOIO_TYPE_MASK))
2038                                 if (!ghcb_rax_is_valid(ghcb))
2039                                         goto vmgexit_err;
2040                 }
2041                 break;
2042         case SVM_EXIT_MSR:
2043                 if (!ghcb_rcx_is_valid(ghcb))
2044                         goto vmgexit_err;
2045                 if (ghcb_get_sw_exit_info_1(ghcb)) {
2046                         if (!ghcb_rax_is_valid(ghcb) ||
2047                             !ghcb_rdx_is_valid(ghcb))
2048                                 goto vmgexit_err;
2049                 }
2050                 break;
2051         case SVM_EXIT_VMMCALL:
2052                 if (!ghcb_rax_is_valid(ghcb) ||
2053                     !ghcb_cpl_is_valid(ghcb))
2054                         goto vmgexit_err;
2055                 break;
2056         case SVM_EXIT_RDTSCP:
2057                 break;
2058         case SVM_EXIT_WBINVD:
2059                 break;
2060         case SVM_EXIT_MONITOR:
2061                 if (!ghcb_rax_is_valid(ghcb) ||
2062                     !ghcb_rcx_is_valid(ghcb) ||
2063                     !ghcb_rdx_is_valid(ghcb))
2064                         goto vmgexit_err;
2065                 break;
2066         case SVM_EXIT_MWAIT:
2067                 if (!ghcb_rax_is_valid(ghcb) ||
2068                     !ghcb_rcx_is_valid(ghcb))
2069                         goto vmgexit_err;
2070                 break;
2071         case SVM_VMGEXIT_MMIO_READ:
2072         case SVM_VMGEXIT_MMIO_WRITE:
2073                 if (!ghcb_sw_scratch_is_valid(ghcb))
2074                         goto vmgexit_err;
2075                 break;
2076         case SVM_VMGEXIT_NMI_COMPLETE:
2077         case SVM_VMGEXIT_AP_HLT_LOOP:
2078         case SVM_VMGEXIT_AP_JUMP_TABLE:
2079         case SVM_VMGEXIT_UNSUPPORTED_EVENT:
2080                 break;
2081         default:
2082                 goto vmgexit_err;
2083         }
2084
2085         return 0;
2086
2087 vmgexit_err:
2088         vcpu = &svm->vcpu;
2089
2090         if (ghcb->ghcb_usage) {
2091                 vcpu_unimpl(vcpu, "vmgexit: ghcb usage %#x is not valid\n",
2092                             ghcb->ghcb_usage);
2093         } else {
2094                 vcpu_unimpl(vcpu, "vmgexit: exit reason %#llx is not valid\n",
2095                             exit_code);
2096                 dump_ghcb(svm);
2097         }
2098
2099         vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2100         vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON;
2101         vcpu->run->internal.ndata = 2;
2102         vcpu->run->internal.data[0] = exit_code;
2103         vcpu->run->internal.data[1] = vcpu->arch.last_vmentry_cpu;
2104
2105         return -EINVAL;
2106 }
2107
2108 static void pre_sev_es_run(struct vcpu_svm *svm)
2109 {
2110         if (!svm->ghcb)
2111                 return;
2112
2113         if (svm->ghcb_sa_free) {
2114                 /*
2115                  * The scratch area lives outside the GHCB, so there is a
2116                  * buffer that, depending on the operation performed, may
2117                  * need to be synced, then freed.
2118                  */
2119                 if (svm->ghcb_sa_sync) {
2120                         kvm_write_guest(svm->vcpu.kvm,
2121                                         ghcb_get_sw_scratch(svm->ghcb),
2122                                         svm->ghcb_sa, svm->ghcb_sa_len);
2123                         svm->ghcb_sa_sync = false;
2124                 }
2125
2126                 kfree(svm->ghcb_sa);
2127                 svm->ghcb_sa = NULL;
2128                 svm->ghcb_sa_free = false;
2129         }
2130
2131         trace_kvm_vmgexit_exit(svm->vcpu.vcpu_id, svm->ghcb);
2132
2133         sev_es_sync_to_ghcb(svm);
2134
2135         kvm_vcpu_unmap(&svm->vcpu, &svm->ghcb_map, true);
2136         svm->ghcb = NULL;
2137 }
2138
2139 void pre_sev_run(struct vcpu_svm *svm, int cpu)
2140 {
2141         struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
2142         int asid = sev_get_asid(svm->vcpu.kvm);
2143
2144         /* Perform any SEV-ES pre-run actions */
2145         pre_sev_es_run(svm);
2146
2147         /* Assign the asid allocated with this SEV guest */
2148         svm->asid = asid;
2149
2150         /*
2151          * Flush guest TLB:
2152          *
2153          * 1) when different VMCB for the same ASID is to be run on the same host CPU.
2154          * 2) or this VMCB was executed on different host CPU in previous VMRUNs.
2155          */
2156         if (sd->sev_vmcbs[asid] == svm->vmcb &&
2157             svm->vcpu.arch.last_vmentry_cpu == cpu)
2158                 return;
2159
2160         sd->sev_vmcbs[asid] = svm->vmcb;
2161         svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
2162         vmcb_mark_dirty(svm->vmcb, VMCB_ASID);
2163 }
2164
2165 #define GHCB_SCRATCH_AREA_LIMIT         (16ULL * PAGE_SIZE)
2166 static bool setup_vmgexit_scratch(struct vcpu_svm *svm, bool sync, u64 len)
2167 {
2168         struct vmcb_control_area *control = &svm->vmcb->control;
2169         struct ghcb *ghcb = svm->ghcb;
2170         u64 ghcb_scratch_beg, ghcb_scratch_end;
2171         u64 scratch_gpa_beg, scratch_gpa_end;
2172         void *scratch_va;
2173
2174         scratch_gpa_beg = ghcb_get_sw_scratch(ghcb);
2175         if (!scratch_gpa_beg) {
2176                 pr_err("vmgexit: scratch gpa not provided\n");
2177                 return false;
2178         }
2179
2180         scratch_gpa_end = scratch_gpa_beg + len;
2181         if (scratch_gpa_end < scratch_gpa_beg) {
2182                 pr_err("vmgexit: scratch length (%#llx) not valid for scratch address (%#llx)\n",
2183                        len, scratch_gpa_beg);
2184                 return false;
2185         }
2186
2187         if ((scratch_gpa_beg & PAGE_MASK) == control->ghcb_gpa) {
2188                 /* Scratch area begins within GHCB */
2189                 ghcb_scratch_beg = control->ghcb_gpa +
2190                                    offsetof(struct ghcb, shared_buffer);
2191                 ghcb_scratch_end = control->ghcb_gpa +
2192                                    offsetof(struct ghcb, reserved_1);
2193
2194                 /*
2195                  * If the scratch area begins within the GHCB, it must be
2196                  * completely contained in the GHCB shared buffer area.
2197                  */
2198                 if (scratch_gpa_beg < ghcb_scratch_beg ||
2199                     scratch_gpa_end > ghcb_scratch_end) {
2200                         pr_err("vmgexit: scratch area is outside of GHCB shared buffer area (%#llx - %#llx)\n",
2201                                scratch_gpa_beg, scratch_gpa_end);
2202                         return false;
2203                 }
2204
2205                 scratch_va = (void *)svm->ghcb;
2206                 scratch_va += (scratch_gpa_beg - control->ghcb_gpa);
2207         } else {
2208                 /*
2209                  * The guest memory must be read into a kernel buffer, so
2210                  * limit the size
2211                  */
2212                 if (len > GHCB_SCRATCH_AREA_LIMIT) {
2213                         pr_err("vmgexit: scratch area exceeds KVM limits (%#llx requested, %#llx limit)\n",
2214                                len, GHCB_SCRATCH_AREA_LIMIT);
2215                         return false;
2216                 }
2217                 scratch_va = kzalloc(len, GFP_KERNEL_ACCOUNT);
2218                 if (!scratch_va)
2219                         return false;
2220
2221                 if (kvm_read_guest(svm->vcpu.kvm, scratch_gpa_beg, scratch_va, len)) {
2222                         /* Unable to copy scratch area from guest */
2223                         pr_err("vmgexit: kvm_read_guest for scratch area failed\n");
2224
2225                         kfree(scratch_va);
2226                         return false;
2227                 }
2228
2229                 /*
2230                  * The scratch area is outside the GHCB. The operation will
2231                  * dictate whether the buffer needs to be synced before running
2232                  * the vCPU next time (i.e. a read was requested so the data
2233                  * must be written back to the guest memory).
2234                  */
2235                 svm->ghcb_sa_sync = sync;
2236                 svm->ghcb_sa_free = true;
2237         }
2238
2239         svm->ghcb_sa = scratch_va;
2240         svm->ghcb_sa_len = len;
2241
2242         return true;
2243 }
2244
2245 static void set_ghcb_msr_bits(struct vcpu_svm *svm, u64 value, u64 mask,
2246                               unsigned int pos)
2247 {
2248         svm->vmcb->control.ghcb_gpa &= ~(mask << pos);
2249         svm->vmcb->control.ghcb_gpa |= (value & mask) << pos;
2250 }
2251
2252 static u64 get_ghcb_msr_bits(struct vcpu_svm *svm, u64 mask, unsigned int pos)
2253 {
2254         return (svm->vmcb->control.ghcb_gpa >> pos) & mask;
2255 }
2256
2257 static void set_ghcb_msr(struct vcpu_svm *svm, u64 value)
2258 {
2259         svm->vmcb->control.ghcb_gpa = value;
2260 }
2261
2262 static int sev_handle_vmgexit_msr_protocol(struct vcpu_svm *svm)
2263 {
2264         struct vmcb_control_area *control = &svm->vmcb->control;
2265         struct kvm_vcpu *vcpu = &svm->vcpu;
2266         u64 ghcb_info;
2267         int ret = 1;
2268
2269         ghcb_info = control->ghcb_gpa & GHCB_MSR_INFO_MASK;
2270
2271         trace_kvm_vmgexit_msr_protocol_enter(svm->vcpu.vcpu_id,
2272                                              control->ghcb_gpa);
2273
2274         switch (ghcb_info) {
2275         case GHCB_MSR_SEV_INFO_REQ:
2276                 set_ghcb_msr(svm, GHCB_MSR_SEV_INFO(GHCB_VERSION_MAX,
2277                                                     GHCB_VERSION_MIN,
2278                                                     sev_enc_bit));
2279                 break;
2280         case GHCB_MSR_CPUID_REQ: {
2281                 u64 cpuid_fn, cpuid_reg, cpuid_value;
2282
2283                 cpuid_fn = get_ghcb_msr_bits(svm,
2284                                              GHCB_MSR_CPUID_FUNC_MASK,
2285                                              GHCB_MSR_CPUID_FUNC_POS);
2286
2287                 /* Initialize the registers needed by the CPUID intercept */
2288                 vcpu->arch.regs[VCPU_REGS_RAX] = cpuid_fn;
2289                 vcpu->arch.regs[VCPU_REGS_RCX] = 0;
2290
2291                 ret = svm_invoke_exit_handler(vcpu, SVM_EXIT_CPUID);
2292                 if (!ret) {
2293                         ret = -EINVAL;
2294                         break;
2295                 }
2296
2297                 cpuid_reg = get_ghcb_msr_bits(svm,
2298                                               GHCB_MSR_CPUID_REG_MASK,
2299                                               GHCB_MSR_CPUID_REG_POS);
2300                 if (cpuid_reg == 0)
2301                         cpuid_value = vcpu->arch.regs[VCPU_REGS_RAX];
2302                 else if (cpuid_reg == 1)
2303                         cpuid_value = vcpu->arch.regs[VCPU_REGS_RBX];
2304                 else if (cpuid_reg == 2)
2305                         cpuid_value = vcpu->arch.regs[VCPU_REGS_RCX];
2306                 else
2307                         cpuid_value = vcpu->arch.regs[VCPU_REGS_RDX];
2308
2309                 set_ghcb_msr_bits(svm, cpuid_value,
2310                                   GHCB_MSR_CPUID_VALUE_MASK,
2311                                   GHCB_MSR_CPUID_VALUE_POS);
2312
2313                 set_ghcb_msr_bits(svm, GHCB_MSR_CPUID_RESP,
2314                                   GHCB_MSR_INFO_MASK,
2315                                   GHCB_MSR_INFO_POS);
2316                 break;
2317         }
2318         case GHCB_MSR_TERM_REQ: {
2319                 u64 reason_set, reason_code;
2320
2321                 reason_set = get_ghcb_msr_bits(svm,
2322                                                GHCB_MSR_TERM_REASON_SET_MASK,
2323                                                GHCB_MSR_TERM_REASON_SET_POS);
2324                 reason_code = get_ghcb_msr_bits(svm,
2325                                                 GHCB_MSR_TERM_REASON_MASK,
2326                                                 GHCB_MSR_TERM_REASON_POS);
2327                 pr_info("SEV-ES guest requested termination: %#llx:%#llx\n",
2328                         reason_set, reason_code);
2329                 fallthrough;
2330         }
2331         default:
2332                 ret = -EINVAL;
2333         }
2334
2335         trace_kvm_vmgexit_msr_protocol_exit(svm->vcpu.vcpu_id,
2336                                             control->ghcb_gpa, ret);
2337
2338         return ret;
2339 }
2340
2341 int sev_handle_vmgexit(struct kvm_vcpu *vcpu)
2342 {
2343         struct vcpu_svm *svm = to_svm(vcpu);
2344         struct vmcb_control_area *control = &svm->vmcb->control;
2345         u64 ghcb_gpa, exit_code;
2346         struct ghcb *ghcb;
2347         int ret;
2348
2349         /* Validate the GHCB */
2350         ghcb_gpa = control->ghcb_gpa;
2351         if (ghcb_gpa & GHCB_MSR_INFO_MASK)
2352                 return sev_handle_vmgexit_msr_protocol(svm);
2353
2354         if (!ghcb_gpa) {
2355                 vcpu_unimpl(vcpu, "vmgexit: GHCB gpa is not set\n");
2356                 return -EINVAL;
2357         }
2358
2359         if (kvm_vcpu_map(vcpu, ghcb_gpa >> PAGE_SHIFT, &svm->ghcb_map)) {
2360                 /* Unable to map GHCB from guest */
2361                 vcpu_unimpl(vcpu, "vmgexit: error mapping GHCB [%#llx] from guest\n",
2362                             ghcb_gpa);
2363                 return -EINVAL;
2364         }
2365
2366         svm->ghcb = svm->ghcb_map.hva;
2367         ghcb = svm->ghcb_map.hva;
2368
2369         trace_kvm_vmgexit_enter(vcpu->vcpu_id, ghcb);
2370
2371         exit_code = ghcb_get_sw_exit_code(ghcb);
2372
2373         ret = sev_es_validate_vmgexit(svm);
2374         if (ret)
2375                 return ret;
2376
2377         sev_es_sync_from_ghcb(svm);
2378         ghcb_set_sw_exit_info_1(ghcb, 0);
2379         ghcb_set_sw_exit_info_2(ghcb, 0);
2380
2381         ret = -EINVAL;
2382         switch (exit_code) {
2383         case SVM_VMGEXIT_MMIO_READ:
2384                 if (!setup_vmgexit_scratch(svm, true, control->exit_info_2))
2385                         break;
2386
2387                 ret = kvm_sev_es_mmio_read(vcpu,
2388                                            control->exit_info_1,
2389                                            control->exit_info_2,
2390                                            svm->ghcb_sa);
2391                 break;
2392         case SVM_VMGEXIT_MMIO_WRITE:
2393                 if (!setup_vmgexit_scratch(svm, false, control->exit_info_2))
2394                         break;
2395
2396                 ret = kvm_sev_es_mmio_write(vcpu,
2397                                             control->exit_info_1,
2398                                             control->exit_info_2,
2399                                             svm->ghcb_sa);
2400                 break;
2401         case SVM_VMGEXIT_NMI_COMPLETE:
2402                 ret = svm_invoke_exit_handler(vcpu, SVM_EXIT_IRET);
2403                 break;
2404         case SVM_VMGEXIT_AP_HLT_LOOP:
2405                 ret = kvm_emulate_ap_reset_hold(vcpu);
2406                 break;
2407         case SVM_VMGEXIT_AP_JUMP_TABLE: {
2408                 struct kvm_sev_info *sev = &to_kvm_svm(vcpu->kvm)->sev_info;
2409
2410                 switch (control->exit_info_1) {
2411                 case 0:
2412                         /* Set AP jump table address */
2413                         sev->ap_jump_table = control->exit_info_2;
2414                         break;
2415                 case 1:
2416                         /* Get AP jump table address */
2417                         ghcb_set_sw_exit_info_2(ghcb, sev->ap_jump_table);
2418                         break;
2419                 default:
2420                         pr_err("svm: vmgexit: unsupported AP jump table request - exit_info_1=%#llx\n",
2421                                control->exit_info_1);
2422                         ghcb_set_sw_exit_info_1(ghcb, 1);
2423                         ghcb_set_sw_exit_info_2(ghcb,
2424                                                 X86_TRAP_UD |
2425                                                 SVM_EVTINJ_TYPE_EXEPT |
2426                                                 SVM_EVTINJ_VALID);
2427                 }
2428
2429                 ret = 1;
2430                 break;
2431         }
2432         case SVM_VMGEXIT_UNSUPPORTED_EVENT:
2433                 vcpu_unimpl(vcpu,
2434                             "vmgexit: unsupported event - exit_info_1=%#llx, exit_info_2=%#llx\n",
2435                             control->exit_info_1, control->exit_info_2);
2436                 break;
2437         default:
2438                 ret = svm_invoke_exit_handler(vcpu, exit_code);
2439         }
2440
2441         return ret;
2442 }
2443
2444 int sev_es_string_io(struct vcpu_svm *svm, int size, unsigned int port, int in)
2445 {
2446         if (!setup_vmgexit_scratch(svm, in, svm->vmcb->control.exit_info_2))
2447                 return -EINVAL;
2448
2449         return kvm_sev_es_string_io(&svm->vcpu, size, port,
2450                                     svm->ghcb_sa, svm->ghcb_sa_len, in);
2451 }
2452
2453 void sev_es_init_vmcb(struct vcpu_svm *svm)
2454 {
2455         struct kvm_vcpu *vcpu = &svm->vcpu;
2456
2457         svm->vmcb->control.nested_ctl |= SVM_NESTED_CTL_SEV_ES_ENABLE;
2458         svm->vmcb->control.virt_ext |= LBR_CTL_ENABLE_MASK;
2459
2460         /*
2461          * An SEV-ES guest requires a VMSA area that is a separate from the
2462          * VMCB page. Do not include the encryption mask on the VMSA physical
2463          * address since hardware will access it using the guest key.
2464          */
2465         svm->vmcb->control.vmsa_pa = __pa(svm->vmsa);
2466
2467         /* Can't intercept CR register access, HV can't modify CR registers */
2468         svm_clr_intercept(svm, INTERCEPT_CR0_READ);
2469         svm_clr_intercept(svm, INTERCEPT_CR4_READ);
2470         svm_clr_intercept(svm, INTERCEPT_CR8_READ);
2471         svm_clr_intercept(svm, INTERCEPT_CR0_WRITE);
2472         svm_clr_intercept(svm, INTERCEPT_CR4_WRITE);
2473         svm_clr_intercept(svm, INTERCEPT_CR8_WRITE);
2474
2475         svm_clr_intercept(svm, INTERCEPT_SELECTIVE_CR0);
2476
2477         /* Track EFER/CR register changes */
2478         svm_set_intercept(svm, TRAP_EFER_WRITE);
2479         svm_set_intercept(svm, TRAP_CR0_WRITE);
2480         svm_set_intercept(svm, TRAP_CR4_WRITE);
2481         svm_set_intercept(svm, TRAP_CR8_WRITE);
2482
2483         /* No support for enable_vmware_backdoor */
2484         clr_exception_intercept(svm, GP_VECTOR);
2485
2486         /* Can't intercept XSETBV, HV can't modify XCR0 directly */
2487         svm_clr_intercept(svm, INTERCEPT_XSETBV);
2488
2489         /* Clear intercepts on selected MSRs */
2490         set_msr_interception(vcpu, svm->msrpm, MSR_EFER, 1, 1);
2491         set_msr_interception(vcpu, svm->msrpm, MSR_IA32_CR_PAT, 1, 1);
2492         set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
2493         set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
2494         set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
2495         set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
2496 }
2497
2498 void sev_es_create_vcpu(struct vcpu_svm *svm)
2499 {
2500         /*
2501          * Set the GHCB MSR value as per the GHCB specification when creating
2502          * a vCPU for an SEV-ES guest.
2503          */
2504         set_ghcb_msr(svm, GHCB_MSR_SEV_INFO(GHCB_VERSION_MAX,
2505                                             GHCB_VERSION_MIN,
2506                                             sev_enc_bit));
2507 }
2508
2509 void sev_es_prepare_guest_switch(struct vcpu_svm *svm, unsigned int cpu)
2510 {
2511         struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
2512         struct vmcb_save_area *hostsa;
2513
2514         /*
2515          * As an SEV-ES guest, hardware will restore the host state on VMEXIT,
2516          * of which one step is to perform a VMLOAD. Since hardware does not
2517          * perform a VMSAVE on VMRUN, the host savearea must be updated.
2518          */
2519         vmsave(__sme_page_pa(sd->save_area));
2520
2521         /* XCR0 is restored on VMEXIT, save the current host value */
2522         hostsa = (struct vmcb_save_area *)(page_address(sd->save_area) + 0x400);
2523         hostsa->xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
2524
2525         /* PKRU is restored on VMEXIT, save the curent host value */
2526         hostsa->pkru = read_pkru();
2527
2528         /* MSR_IA32_XSS is restored on VMEXIT, save the currnet host value */
2529         hostsa->xss = host_xss;
2530 }
2531
2532 void sev_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
2533 {
2534         struct vcpu_svm *svm = to_svm(vcpu);
2535
2536         /* First SIPI: Use the values as initially set by the VMM */
2537         if (!svm->received_first_sipi) {
2538                 svm->received_first_sipi = true;
2539                 return;
2540         }
2541
2542         /*
2543          * Subsequent SIPI: Return from an AP Reset Hold VMGEXIT, where
2544          * the guest will set the CS and RIP. Set SW_EXIT_INFO_2 to a
2545          * non-zero value.
2546          */
2547         if (!svm->ghcb)
2548                 return;
2549
2550         ghcb_set_sw_exit_info_2(svm->ghcb, 1);
2551 }