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/misc_cgroup.h>
  18 #include <linux/processor.h>
  19 #include <linux/trace_events.h>
  20 #include <asm/fpu/internal.h>
  21
  22 #include <asm/trapnr.h>
  23
  24 #include "x86.h"
  25 #include "svm.h"
  26 #include "svm_ops.h"
  27 #include "cpuid.h"
  28 #include "trace.h"
  29
  30 #define __ex(x) __kvm_handle_fault_on_reboot(x)
  31
  32 #ifndef CONFIG_KVM_AMD_SEV
  33 /*
  34  * When this config is not defined, SEV feature is not supported and APIs in
  35  * this file are not used but this file still gets compiled into the KVM AMD
  36  * module.
  37  *
  38  * We will not have MISC_CG_RES_SEV and MISC_CG_RES_SEV_ES entries in the enum
  39  * misc_res_type {} defined in linux/misc_cgroup.h.
  40  *
  41  * Below macros allow compilation to succeed.
  42  */
  43 #define MISC_CG_RES_SEV MISC_CG_RES_TYPES
  44 #define MISC_CG_RES_SEV_ES MISC_CG_RES_TYPES
  45 #endif
  46
  47 #ifdef CONFIG_KVM_AMD_SEV
  48 /* enable/disable SEV support */
  49 static bool sev_enabled = true;
  50 module_param_named(sev, sev_enabled, bool, 0444);
  51
  52 /* enable/disable SEV-ES support */
  53 static bool sev_es_enabled = true;
  54 module_param_named(sev_es, sev_es_enabled, bool, 0444);
  55 #else
  56 #define sev_enabled false
  57 #define sev_es_enabled false
  58 #endif /* CONFIG_KVM_AMD_SEV */
  59
  60 static u8 sev_enc_bit;
  61 static DECLARE_RWSEM(sev_deactivate_lock);
  62 static DEFINE_MUTEX(sev_bitmap_lock);
  63 unsigned int max_sev_asid;
  64 static unsigned int min_sev_asid;
  65 static unsigned long sev_me_mask;
  66 static unsigned long *sev_asid_bitmap;
  67 static unsigned long *sev_reclaim_asid_bitmap;
  68
  69 struct enc_region {
  70         struct list_head list;
  71         unsigned long npages;
  72         struct page **pages;
  73         unsigned long uaddr;
  74         unsigned long size;
  75 };
  76
  77 /* Called with the sev_bitmap_lock held, or on shutdown  */
  78 static int sev_flush_asids(int min_asid, int max_asid)
  79 {
  80         int ret, pos, error = 0;
  81
  82         /* Check if there are any ASIDs to reclaim before performing a flush */
  83         pos = find_next_bit(sev_reclaim_asid_bitmap, max_asid, min_asid);
  84         if (pos >= max_asid)
  85                 return -EBUSY;
  86
  87         /*
  88          * DEACTIVATE will clear the WBINVD indicator causing DF_FLUSH to fail,
  89          * so it must be guarded.
  90          */
  91         down_write(&sev_deactivate_lock);
  92
  93         wbinvd_on_all_cpus();
  94         ret = sev_guest_df_flush(&error);
  95
  96         up_write(&sev_deactivate_lock);
  97
  98         if (ret)
  99                 pr_err("SEV: DF_FLUSH failed, ret=%d, error=%#x\n", ret, error);
 100
 101         return ret;
 102 }
 103
 104 static inline bool is_mirroring_enc_context(struct kvm *kvm)
 105 {
 106         return !!to_kvm_svm(kvm)->sev_info.enc_context_owner;
 107 }
 108
 109 /* Must be called with the sev_bitmap_lock held */
 110 static bool __sev_recycle_asids(int min_asid, int max_asid)
 111 {
 112         if (sev_flush_asids(min_asid, max_asid))
 113                 return false;
 114
 115         /* The flush process will flush all reclaimable SEV and SEV-ES ASIDs */
 116         bitmap_xor(sev_asid_bitmap, sev_asid_bitmap, sev_reclaim_asid_bitmap,
 117                    max_sev_asid);
 118         bitmap_zero(sev_reclaim_asid_bitmap, max_sev_asid);
 119
 120         return true;
 121 }
 122
 123 static int sev_asid_new(struct kvm_sev_info *sev)
 124 {
 125         int pos, min_asid, max_asid, ret;
 126         bool retry = true;
 127         enum misc_res_type type;
 128
 129         type = sev->es_active ? MISC_CG_RES_SEV_ES : MISC_CG_RES_SEV;
 130         WARN_ON(sev->misc_cg);
 131         sev->misc_cg = get_current_misc_cg();
 132         ret = misc_cg_try_charge(type, sev->misc_cg, 1);
 133         if (ret) {
 134                 put_misc_cg(sev->misc_cg);
 135                 sev->misc_cg = NULL;
 136                 return ret;
 137         }
 138
 139         mutex_lock(&sev_bitmap_lock);
 140
 141         /*
 142          * SEV-enabled guests must use asid from min_sev_asid to max_sev_asid.
 143          * SEV-ES-enabled guest can use from 1 to min_sev_asid - 1.
 144          */
 145         min_asid = sev->es_active ? 0 : min_sev_asid - 1;
 146         max_asid = sev->es_active ? min_sev_asid - 1 : max_sev_asid;
 147 again:
 148         pos = find_next_zero_bit(sev_asid_bitmap, max_sev_asid, min_asid);
 149         if (pos >= max_asid) {
 150                 if (retry && __sev_recycle_asids(min_asid, max_asid)) {
 151                         retry = false;
 152                         goto again;
 153                 }
 154                 mutex_unlock(&sev_bitmap_lock);
 155                 ret = -EBUSY;
 156                 goto e_uncharge;
 157         }
 158
 159         __set_bit(pos, sev_asid_bitmap);
 160
 161         mutex_unlock(&sev_bitmap_lock);
 162
 163         return pos + 1;
 164 e_uncharge:
 165         misc_cg_uncharge(type, sev->misc_cg, 1);
 166         put_misc_cg(sev->misc_cg);
 167         sev->misc_cg = NULL;
 168         return ret;
 169 }
 170
 171 static int sev_get_asid(struct kvm *kvm)
 172 {
 173         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 174
 175         return sev->asid;
 176 }
 177
 178 static void sev_asid_free(struct kvm_sev_info *sev)
 179 {
 180         struct svm_cpu_data *sd;
 181         int cpu, pos;
 182         enum misc_res_type type;
 183
 184         mutex_lock(&sev_bitmap_lock);
 185
 186         pos = sev->asid - 1;
 187         __set_bit(pos, sev_reclaim_asid_bitmap);
 188
 189         for_each_possible_cpu(cpu) {
 190                 sd = per_cpu(svm_data, cpu);
 191                 sd->sev_vmcbs[pos] = NULL;
 192         }
 193
 194         mutex_unlock(&sev_bitmap_lock);
 195
 196         type = sev->es_active ? MISC_CG_RES_SEV_ES : MISC_CG_RES_SEV;
 197         misc_cg_uncharge(type, sev->misc_cg, 1);
 198         put_misc_cg(sev->misc_cg);
 199         sev->misc_cg = NULL;
 200 }
 201
 202 static void sev_decommission(unsigned int handle)
 203 {
 204         struct sev_data_decommission decommission;
 205
 206         if (!handle)
 207                 return;
 208
 209         decommission.handle = handle;
 210         sev_guest_decommission(&decommission, NULL);
 211 }
 212
 213 static void sev_unbind_asid(struct kvm *kvm, unsigned int handle)
 214 {
 215         struct sev_data_deactivate deactivate;
 216
 217         if (!handle)
 218                 return;
 219
 220         deactivate.handle = handle;
 221
 222         /* Guard DEACTIVATE against WBINVD/DF_FLUSH used in ASID recycling */
 223         down_read(&sev_deactivate_lock);
 224         sev_guest_deactivate(&deactivate, NULL);
 225         up_read(&sev_deactivate_lock);
 226
 227         sev_decommission(handle);
 228 }
 229
 230 static int sev_guest_init(struct kvm *kvm, struct kvm_sev_cmd *argp)
 231 {
 232         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 233         bool es_active = argp->id == KVM_SEV_ES_INIT;
 234         int asid, ret;
 235
 236         if (kvm->created_vcpus)
 237                 return -EINVAL;
 238
 239         ret = -EBUSY;
 240         if (unlikely(sev->active))
 241                 return ret;
 242
 243         sev->es_active = es_active;
 244         asid = sev_asid_new(sev);
 245         if (asid < 0)
 246                 goto e_no_asid;
 247         sev->asid = asid;
 248
 249         ret = sev_platform_init(&argp->error);
 250         if (ret)
 251                 goto e_free;
 252
 253         sev->active = true;
 254         sev->asid = asid;
 255         INIT_LIST_HEAD(&sev->regions_list);
 256
 257         return 0;
 258
 259 e_free:
 260         sev_asid_free(sev);
 261         sev->asid = 0;
 262 e_no_asid:
 263         sev->es_active = false;
 264         return ret;
 265 }
 266
 267 static int sev_bind_asid(struct kvm *kvm, unsigned int handle, int *error)
 268 {
 269         struct sev_data_activate activate;
 270         int asid = sev_get_asid(kvm);
 271         int ret;
 272
 273         /* activate ASID on the given handle */
 274         activate.handle = handle;
 275         activate.asid   = asid;
 276         ret = sev_guest_activate(&activate, error);
 277
 278         return ret;
 279 }
 280
 281 static int __sev_issue_cmd(int fd, int id, void *data, int *error)
 282 {
 283         struct fd f;
 284         int ret;
 285
 286         f = fdget(fd);
 287         if (!f.file)
 288                 return -EBADF;
 289
 290         ret = sev_issue_cmd_external_user(f.file, id, data, error);
 291
 292         fdput(f);
 293         return ret;
 294 }
 295
 296 static int sev_issue_cmd(struct kvm *kvm, int id, void *data, int *error)
 297 {
 298         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 299
 300         return __sev_issue_cmd(sev->fd, id, data, error);
 301 }
 302
 303 static int sev_launch_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
 304 {
 305         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 306         struct sev_data_launch_start start;
 307         struct kvm_sev_launch_start params;
 308         void *dh_blob, *session_blob;
 309         int *error = &argp->error;
 310         int ret;
 311
 312         if (!sev_guest(kvm))
 313                 return -ENOTTY;
 314
 315         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data, sizeof(params)))
 316                 return -EFAULT;
 317
 318         memset(&start, 0, sizeof(start));
 319
 320         dh_blob = NULL;
 321         if (params.dh_uaddr) {
 322                 dh_blob = psp_copy_user_blob(params.dh_uaddr, params.dh_len);
 323                 if (IS_ERR(dh_blob))
 324                         return PTR_ERR(dh_blob);
 325
 326                 start.dh_cert_address = __sme_set(__pa(dh_blob));
 327                 start.dh_cert_len = params.dh_len;
 328         }
 329
 330         session_blob = NULL;
 331         if (params.session_uaddr) {
 332                 session_blob = psp_copy_user_blob(params.session_uaddr, params.session_len);
 333                 if (IS_ERR(session_blob)) {
 334                         ret = PTR_ERR(session_blob);
 335                         goto e_free_dh;
 336                 }
 337
 338                 start.session_address = __sme_set(__pa(session_blob));
 339                 start.session_len = params.session_len;
 340         }
 341
 342         start.handle = params.handle;
 343         start.policy = params.policy;
 344
 345         /* create memory encryption context */
 346         ret = __sev_issue_cmd(argp->sev_fd, SEV_CMD_LAUNCH_START, &start, error);
 347         if (ret)
 348                 goto e_free_session;
 349
 350         /* Bind ASID to this guest */
 351         ret = sev_bind_asid(kvm, start.handle, error);
 352         if (ret) {
 353                 sev_decommission(start.handle);
 354                 goto e_free_session;
 355         }
 356
 357         /* return handle to userspace */
 358         params.handle = start.handle;
 359         if (copy_to_user((void __user *)(uintptr_t)argp->data, &params, sizeof(params))) {
 360                 sev_unbind_asid(kvm, start.handle);
 361                 ret = -EFAULT;
 362                 goto e_free_session;
 363         }
 364
 365         sev->handle = start.handle;
 366         sev->fd = argp->sev_fd;
 367
 368 e_free_session:
 369         kfree(session_blob);
 370 e_free_dh:
 371         kfree(dh_blob);
 372         return ret;
 373 }
 374
 375 static struct page **sev_pin_memory(struct kvm *kvm, unsigned long uaddr,
 376                                     unsigned long ulen, unsigned long *n,
 377                                     int write)
 378 {
 379         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 380         unsigned long npages, size;
 381         int npinned;
 382         unsigned long locked, lock_limit;
 383         struct page **pages;
 384         unsigned long first, last;
 385         int ret;
 386
 387         lockdep_assert_held(&kvm->lock);
 388
 389         if (ulen == 0 || uaddr + ulen < uaddr)
 390                 return ERR_PTR(-EINVAL);
 391
 392         /* Calculate number of pages. */
 393         first = (uaddr & PAGE_MASK) >> PAGE_SHIFT;
 394         last = ((uaddr + ulen - 1) & PAGE_MASK) >> PAGE_SHIFT;
 395         npages = (last - first + 1);
 396
 397         locked = sev->pages_locked + npages;
 398         lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
 399         if (locked > lock_limit && !capable(CAP_IPC_LOCK)) {
 400                 pr_err("SEV: %lu locked pages exceed the lock limit of %lu.\n", locked, lock_limit);
 401                 return ERR_PTR(-ENOMEM);
 402         }
 403
 404         if (WARN_ON_ONCE(npages > INT_MAX))
 405                 return ERR_PTR(-EINVAL);
 406
 407         /* Avoid using vmalloc for smaller buffers. */
 408         size = npages * sizeof(struct page *);
 409         if (size > PAGE_SIZE)
 410                 pages = __vmalloc(size, GFP_KERNEL_ACCOUNT | __GFP_ZERO);
 411         else
 412                 pages = kmalloc(size, GFP_KERNEL_ACCOUNT);
 413
 414         if (!pages)
 415                 return ERR_PTR(-ENOMEM);
 416
 417         /* Pin the user virtual address. */
 418         npinned = pin_user_pages_fast(uaddr, npages, write ? FOLL_WRITE : 0, pages);
 419         if (npinned != npages) {
 420                 pr_err("SEV: Failure locking %lu pages.\n", npages);
 421                 ret = -ENOMEM;
 422                 goto err;
 423         }
 424
 425         *n = npages;
 426         sev->pages_locked = locked;
 427
 428         return pages;
 429
 430 err:
 431         if (npinned > 0)
 432                 unpin_user_pages(pages, npinned);
 433
 434         kvfree(pages);
 435         return ERR_PTR(ret);
 436 }
 437
 438 static void sev_unpin_memory(struct kvm *kvm, struct page **pages,
 439                              unsigned long npages)
 440 {
 441         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 442
 443         unpin_user_pages(pages, npages);
 444         kvfree(pages);
 445         sev->pages_locked -= npages;
 446 }
 447
 448 static void sev_clflush_pages(struct page *pages[], unsigned long npages)
 449 {
 450         uint8_t *page_virtual;
 451         unsigned long i;
 452
 453         if (this_cpu_has(X86_FEATURE_SME_COHERENT) || npages == 0 ||
 454             pages == NULL)
 455                 return;
 456
 457         for (i = 0; i < npages; i++) {
 458                 page_virtual = kmap_atomic(pages[i]);
 459                 clflush_cache_range(page_virtual, PAGE_SIZE);
 460                 kunmap_atomic(page_virtual);
 461         }
 462 }
 463
 464 static unsigned long get_num_contig_pages(unsigned long idx,
 465                                 struct page **inpages, unsigned long npages)
 466 {
 467         unsigned long paddr, next_paddr;
 468         unsigned long i = idx + 1, pages = 1;
 469
 470         /* find the number of contiguous pages starting from idx */
 471         paddr = __sme_page_pa(inpages[idx]);
 472         while (i < npages) {
 473                 next_paddr = __sme_page_pa(inpages[i++]);
 474                 if ((paddr + PAGE_SIZE) == next_paddr) {
 475                         pages++;
 476                         paddr = next_paddr;
 477                         continue;
 478                 }
 479                 break;
 480         }
 481
 482         return pages;
 483 }
 484
 485 static int sev_launch_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
 486 {
 487         unsigned long vaddr, vaddr_end, next_vaddr, npages, pages, size, i;
 488         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 489         struct kvm_sev_launch_update_data params;
 490         struct sev_data_launch_update_data data;
 491         struct page **inpages;
 492         int ret;
 493
 494         if (!sev_guest(kvm))
 495                 return -ENOTTY;
 496
 497         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data, sizeof(params)))
 498                 return -EFAULT;
 499
 500         vaddr = params.uaddr;
 501         size = params.len;
 502         vaddr_end = vaddr + size;
 503
 504         /* Lock the user memory. */
 505         inpages = sev_pin_memory(kvm, vaddr, size, &npages, 1);
 506         if (IS_ERR(inpages))
 507                 return PTR_ERR(inpages);
 508
 509         /*
 510          * Flush (on non-coherent CPUs) before LAUNCH_UPDATE encrypts pages in
 511          * place; the cache may contain the data that was written unencrypted.
 512          */
 513         sev_clflush_pages(inpages, npages);
 514
 515         data.reserved = 0;
 516         data.handle = sev->handle;
 517
 518         for (i = 0; vaddr < vaddr_end; vaddr = next_vaddr, i += pages) {
 519                 int offset, len;
 520
 521                 /*
 522                  * If the user buffer is not page-aligned, calculate the offset
 523                  * within the page.
 524                  */
 525                 offset = vaddr & (PAGE_SIZE - 1);
 526
 527                 /* Calculate the number of pages that can be encrypted in one go. */
 528                 pages = get_num_contig_pages(i, inpages, npages);
 529
 530                 len = min_t(size_t, ((pages * PAGE_SIZE) - offset), size);
 531
 532                 data.len = len;
 533                 data.address = __sme_page_pa(inpages[i]) + offset;
 534                 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_DATA, &data, &argp->error);
 535                 if (ret)
 536                         goto e_unpin;
 537
 538                 size -= len;
 539                 next_vaddr = vaddr + len;
 540         }
 541
 542 e_unpin:
 543         /* content of memory is updated, mark pages dirty */
 544         for (i = 0; i < npages; i++) {
 545                 set_page_dirty_lock(inpages[i]);
 546                 mark_page_accessed(inpages[i]);
 547         }
 548         /* unlock the user pages */
 549         sev_unpin_memory(kvm, inpages, npages);
 550         return ret;
 551 }
 552
 553 static int sev_es_sync_vmsa(struct vcpu_svm *svm)
 554 {
 555         struct vmcb_save_area *save = &svm->vmcb->save;
 556
 557         /* Check some debug related fields before encrypting the VMSA */
 558         if (svm->vcpu.guest_debug || (save->dr7 & ~DR7_FIXED_1))
 559                 return -EINVAL;
 560
 561         /* Sync registgers */
 562         save->rax = svm->vcpu.arch.regs[VCPU_REGS_RAX];
 563         save->rbx = svm->vcpu.arch.regs[VCPU_REGS_RBX];
 564         save->rcx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
 565         save->rdx = svm->vcpu.arch.regs[VCPU_REGS_RDX];
 566         save->rsp = svm->vcpu.arch.regs[VCPU_REGS_RSP];
 567         save->rbp = svm->vcpu.arch.regs[VCPU_REGS_RBP];
 568         save->rsi = svm->vcpu.arch.regs[VCPU_REGS_RSI];
 569         save->rdi = svm->vcpu.arch.regs[VCPU_REGS_RDI];
 570 #ifdef CONFIG_X86_64
 571         save->r8  = svm->vcpu.arch.regs[VCPU_REGS_R8];
 572         save->r9  = svm->vcpu.arch.regs[VCPU_REGS_R9];
 573         save->r10 = svm->vcpu.arch.regs[VCPU_REGS_R10];
 574         save->r11 = svm->vcpu.arch.regs[VCPU_REGS_R11];
 575         save->r12 = svm->vcpu.arch.regs[VCPU_REGS_R12];
 576         save->r13 = svm->vcpu.arch.regs[VCPU_REGS_R13];
 577         save->r14 = svm->vcpu.arch.regs[VCPU_REGS_R14];
 578         save->r15 = svm->vcpu.arch.regs[VCPU_REGS_R15];
 579 #endif
 580         save->rip = svm->vcpu.arch.regs[VCPU_REGS_RIP];
 581
 582         /* Sync some non-GPR registers before encrypting */
 583         save->xcr0 = svm->vcpu.arch.xcr0;
 584         save->pkru = svm->vcpu.arch.pkru;
 585         save->xss  = svm->vcpu.arch.ia32_xss;
 586
 587         /*
 588          * SEV-ES will use a VMSA that is pointed to by the VMCB, not
 589          * the traditional VMSA that is part of the VMCB. Copy the
 590          * traditional VMSA as it has been built so far (in prep
 591          * for LAUNCH_UPDATE_VMSA) to be the initial SEV-ES state.
 592          */
 593         memcpy(svm->vmsa, save, sizeof(*save));
 594
 595         return 0;
 596 }
 597
 598 static int sev_launch_update_vmsa(struct kvm *kvm, struct kvm_sev_cmd *argp)
 599 {
 600         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 601         struct sev_data_launch_update_vmsa vmsa;
 602         struct kvm_vcpu *vcpu;
 603         int i, ret;
 604
 605         if (!sev_es_guest(kvm))
 606                 return -ENOTTY;
 607
 608         vmsa.reserved = 0;
 609
 610         kvm_for_each_vcpu(i, vcpu, kvm) {
 611                 struct vcpu_svm *svm = to_svm(vcpu);
 612
 613                 /* Perform some pre-encryption checks against the VMSA */
 614                 ret = sev_es_sync_vmsa(svm);
 615                 if (ret)
 616                         return ret;
 617
 618                 /*
 619                  * The LAUNCH_UPDATE_VMSA command will perform in-place
 620                  * encryption of the VMSA memory content (i.e it will write
 621                  * the same memory region with the guest's key), so invalidate
 622                  * it first.
 623                  */
 624                 clflush_cache_range(svm->vmsa, PAGE_SIZE);
 625
 626                 vmsa.handle = sev->handle;
 627                 vmsa.address = __sme_pa(svm->vmsa);
 628                 vmsa.len = PAGE_SIZE;
 629                 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_VMSA, &vmsa,
 630                                     &argp->error);
 631                 if (ret)
 632                         return ret;
 633
 634                 svm->vcpu.arch.guest_state_protected = true;
 635         }
 636
 637         return 0;
 638 }
 639
 640 static int sev_launch_measure(struct kvm *kvm, struct kvm_sev_cmd *argp)
 641 {
 642         void __user *measure = (void __user *)(uintptr_t)argp->data;
 643         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 644         struct sev_data_launch_measure data;
 645         struct kvm_sev_launch_measure params;
 646         void __user *p = NULL;
 647         void *blob = NULL;
 648         int ret;
 649
 650         if (!sev_guest(kvm))
 651                 return -ENOTTY;
 652
 653         if (copy_from_user(&params, measure, sizeof(params)))
 654                 return -EFAULT;
 655
 656         memset(&data, 0, sizeof(data));
 657
 658         /* User wants to query the blob length */
 659         if (!params.len)
 660                 goto cmd;
 661
 662         p = (void __user *)(uintptr_t)params.uaddr;
 663         if (p) {
 664                 if (params.len > SEV_FW_BLOB_MAX_SIZE)
 665                         return -EINVAL;
 666
 667                 blob = kmalloc(params.len, GFP_KERNEL_ACCOUNT);
 668                 if (!blob)
 669                         return -ENOMEM;
 670
 671                 data.address = __psp_pa(blob);
 672                 data.len = params.len;
 673         }
 674
 675 cmd:
 676         data.handle = sev->handle;
 677         ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_MEASURE, &data, &argp->error);
 678
 679         /*
 680          * If we query the session length, FW responded with expected data.
 681          */
 682         if (!params.len)
 683                 goto done;
 684
 685         if (ret)
 686                 goto e_free_blob;
 687
 688         if (blob) {
 689                 if (copy_to_user(p, blob, params.len))
 690                         ret = -EFAULT;
 691         }
 692
 693 done:
 694         params.len = data.len;
 695         if (copy_to_user(measure, &params, sizeof(params)))
 696                 ret = -EFAULT;
 697 e_free_blob:
 698         kfree(blob);
 699         return ret;
 700 }
 701
 702 static int sev_launch_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
 703 {
 704         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 705         struct sev_data_launch_finish data;
 706
 707         if (!sev_guest(kvm))
 708                 return -ENOTTY;
 709
 710         data.handle = sev->handle;
 711         return sev_issue_cmd(kvm, SEV_CMD_LAUNCH_FINISH, &data, &argp->error);
 712 }
 713
 714 static int sev_guest_status(struct kvm *kvm, struct kvm_sev_cmd *argp)
 715 {
 716         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 717         struct kvm_sev_guest_status params;
 718         struct sev_data_guest_status data;
 719         int ret;
 720
 721         if (!sev_guest(kvm))
 722                 return -ENOTTY;
 723
 724         memset(&data, 0, sizeof(data));
 725
 726         data.handle = sev->handle;
 727         ret = sev_issue_cmd(kvm, SEV_CMD_GUEST_STATUS, &data, &argp->error);
 728         if (ret)
 729                 return ret;
 730
 731         params.policy = data.policy;
 732         params.state = data.state;
 733         params.handle = data.handle;
 734
 735         if (copy_to_user((void __user *)(uintptr_t)argp->data, &params, sizeof(params)))
 736                 ret = -EFAULT;
 737
 738         return ret;
 739 }
 740
 741 static int __sev_issue_dbg_cmd(struct kvm *kvm, unsigned long src,
 742                                unsigned long dst, int size,
 743                                int *error, bool enc)
 744 {
 745         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 746         struct sev_data_dbg data;
 747
 748         data.reserved = 0;
 749         data.handle = sev->handle;
 750         data.dst_addr = dst;
 751         data.src_addr = src;
 752         data.len = size;
 753
 754         return sev_issue_cmd(kvm,
 755                              enc ? SEV_CMD_DBG_ENCRYPT : SEV_CMD_DBG_DECRYPT,
 756                              &data, error);
 757 }
 758
 759 static int __sev_dbg_decrypt(struct kvm *kvm, unsigned long src_paddr,
 760                              unsigned long dst_paddr, int sz, int *err)
 761 {
 762         int offset;
 763
 764         /*
 765          * Its safe to read more than we are asked, caller should ensure that
 766          * destination has enough space.
 767          */
 768         offset = src_paddr & 15;
 769         src_paddr = round_down(src_paddr, 16);
 770         sz = round_up(sz + offset, 16);
 771
 772         return __sev_issue_dbg_cmd(kvm, src_paddr, dst_paddr, sz, err, false);
 773 }
 774
 775 static int __sev_dbg_decrypt_user(struct kvm *kvm, unsigned long paddr,
 776                                   void __user *dst_uaddr,
 777                                   unsigned long dst_paddr,
 778                                   int size, int *err)
 779 {
 780         struct page *tpage = NULL;
 781         int ret, offset;
 782
 783         /* if inputs are not 16-byte then use intermediate buffer */
 784         if (!IS_ALIGNED(dst_paddr, 16) ||
 785             !IS_ALIGNED(paddr,     16) ||
 786             !IS_ALIGNED(size,      16)) {
 787                 tpage = (void *)alloc_page(GFP_KERNEL);
 788                 if (!tpage)
 789                         return -ENOMEM;
 790
 791                 dst_paddr = __sme_page_pa(tpage);
 792         }
 793
 794         ret = __sev_dbg_decrypt(kvm, paddr, dst_paddr, size, err);
 795         if (ret)
 796                 goto e_free;
 797
 798         if (tpage) {
 799                 offset = paddr & 15;
 800                 if (copy_to_user(dst_uaddr, page_address(tpage) + offset, size))
 801                         ret = -EFAULT;
 802         }
 803
 804 e_free:
 805         if (tpage)
 806                 __free_page(tpage);
 807
 808         return ret;
 809 }
 810
 811 static int __sev_dbg_encrypt_user(struct kvm *kvm, unsigned long paddr,
 812                                   void __user *vaddr,
 813                                   unsigned long dst_paddr,
 814                                   void __user *dst_vaddr,
 815                                   int size, int *error)
 816 {
 817         struct page *src_tpage = NULL;
 818         struct page *dst_tpage = NULL;
 819         int ret, len = size;
 820
 821         /* If source buffer is not aligned then use an intermediate buffer */
 822         if (!IS_ALIGNED((unsigned long)vaddr, 16)) {
 823                 src_tpage = alloc_page(GFP_KERNEL);
 824                 if (!src_tpage)
 825                         return -ENOMEM;
 826
 827                 if (copy_from_user(page_address(src_tpage), vaddr, size)) {
 828                         __free_page(src_tpage);
 829                         return -EFAULT;
 830                 }
 831
 832                 paddr = __sme_page_pa(src_tpage);
 833         }
 834
 835         /*
 836          *  If destination buffer or length is not aligned then do read-modify-write:
 837          *   - decrypt destination in an intermediate buffer
 838          *   - copy the source buffer in an intermediate buffer
 839          *   - use the intermediate buffer as source buffer
 840          */
 841         if (!IS_ALIGNED((unsigned long)dst_vaddr, 16) || !IS_ALIGNED(size, 16)) {
 842                 int dst_offset;
 843
 844                 dst_tpage = alloc_page(GFP_KERNEL);
 845                 if (!dst_tpage) {
 846                         ret = -ENOMEM;
 847                         goto e_free;
 848                 }
 849
 850                 ret = __sev_dbg_decrypt(kvm, dst_paddr,
 851                                         __sme_page_pa(dst_tpage), size, error);
 852                 if (ret)
 853                         goto e_free;
 854
 855                 /*
 856                  *  If source is kernel buffer then use memcpy() otherwise
 857                  *  copy_from_user().
 858                  */
 859                 dst_offset = dst_paddr & 15;
 860
 861                 if (src_tpage)
 862                         memcpy(page_address(dst_tpage) + dst_offset,
 863                                page_address(src_tpage), size);
 864                 else {
 865                         if (copy_from_user(page_address(dst_tpage) + dst_offset,
 866                                            vaddr, size)) {
 867                                 ret = -EFAULT;
 868                                 goto e_free;
 869                         }
 870                 }
 871
 872                 paddr = __sme_page_pa(dst_tpage);
 873                 dst_paddr = round_down(dst_paddr, 16);
 874                 len = round_up(size, 16);
 875         }
 876
 877         ret = __sev_issue_dbg_cmd(kvm, paddr, dst_paddr, len, error, true);
 878
 879 e_free:
 880         if (src_tpage)
 881                 __free_page(src_tpage);
 882         if (dst_tpage)
 883                 __free_page(dst_tpage);
 884         return ret;
 885 }
 886
 887 static int sev_dbg_crypt(struct kvm *kvm, struct kvm_sev_cmd *argp, bool dec)
 888 {
 889         unsigned long vaddr, vaddr_end, next_vaddr;
 890         unsigned long dst_vaddr;
 891         struct page **src_p, **dst_p;
 892         struct kvm_sev_dbg debug;
 893         unsigned long n;
 894         unsigned int size;
 895         int ret;
 896
 897         if (!sev_guest(kvm))
 898                 return -ENOTTY;
 899
 900         if (copy_from_user(&debug, (void __user *)(uintptr_t)argp->data, sizeof(debug)))
 901                 return -EFAULT;
 902
 903         if (!debug.len || debug.src_uaddr + debug.len < debug.src_uaddr)
 904                 return -EINVAL;
 905         if (!debug.dst_uaddr)
 906                 return -EINVAL;
 907
 908         vaddr = debug.src_uaddr;
 909         size = debug.len;
 910         vaddr_end = vaddr + size;
 911         dst_vaddr = debug.dst_uaddr;
 912
 913         for (; vaddr < vaddr_end; vaddr = next_vaddr) {
 914                 int len, s_off, d_off;
 915
 916                 /* lock userspace source and destination page */
 917                 src_p = sev_pin_memory(kvm, vaddr & PAGE_MASK, PAGE_SIZE, &n, 0);
 918                 if (IS_ERR(src_p))
 919                         return PTR_ERR(src_p);
 920
 921                 dst_p = sev_pin_memory(kvm, dst_vaddr & PAGE_MASK, PAGE_SIZE, &n, 1);
 922                 if (IS_ERR(dst_p)) {
 923                         sev_unpin_memory(kvm, src_p, n);
 924                         return PTR_ERR(dst_p);
 925                 }
 926
 927                 /*
 928                  * Flush (on non-coherent CPUs) before DBG_{DE,EN}CRYPT read or modify
 929                  * the pages; flush the destination too so that future accesses do not
 930                  * see stale data.
 931                  */
 932                 sev_clflush_pages(src_p, 1);
 933                 sev_clflush_pages(dst_p, 1);
 934
 935                 /*
 936                  * Since user buffer may not be page aligned, calculate the
 937                  * offset within the page.
 938                  */
 939                 s_off = vaddr & ~PAGE_MASK;
 940                 d_off = dst_vaddr & ~PAGE_MASK;
 941                 len = min_t(size_t, (PAGE_SIZE - s_off), size);
 942
 943                 if (dec)
 944                         ret = __sev_dbg_decrypt_user(kvm,
 945                                                      __sme_page_pa(src_p[0]) + s_off,
 946                                                      (void __user *)dst_vaddr,
 947                                                      __sme_page_pa(dst_p[0]) + d_off,
 948                                                      len, &argp->error);
 949                 else
 950                         ret = __sev_dbg_encrypt_user(kvm,
 951                                                      __sme_page_pa(src_p[0]) + s_off,
 952                                                      (void __user *)vaddr,
 953                                                      __sme_page_pa(dst_p[0]) + d_off,
 954                                                      (void __user *)dst_vaddr,
 955                                                      len, &argp->error);
 956
 957                 sev_unpin_memory(kvm, src_p, n);
 958                 sev_unpin_memory(kvm, dst_p, n);
 959
 960                 if (ret)
 961                         goto err;
 962
 963                 next_vaddr = vaddr + len;
 964                 dst_vaddr = dst_vaddr + len;
 965                 size -= len;
 966         }
 967 err:
 968         return ret;
 969 }
 970
 971 static int sev_launch_secret(struct kvm *kvm, struct kvm_sev_cmd *argp)
 972 {
 973         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 974         struct sev_data_launch_secret data;
 975         struct kvm_sev_launch_secret params;
 976         struct page **pages;
 977         void *blob, *hdr;
 978         unsigned long n, i;
 979         int ret, offset;
 980
 981         if (!sev_guest(kvm))
 982                 return -ENOTTY;
 983
 984         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data, sizeof(params)))
 985                 return -EFAULT;
 986
 987         pages = sev_pin_memory(kvm, params.guest_uaddr, params.guest_len, &n, 1);
 988         if (IS_ERR(pages))
 989                 return PTR_ERR(pages);
 990
 991         /*
 992          * Flush (on non-coherent CPUs) before LAUNCH_SECRET encrypts pages in
 993          * place; the cache may contain the data that was written unencrypted.
 994          */
 995         sev_clflush_pages(pages, n);
 996
 997         /*
 998          * The secret must be copied into contiguous memory region, lets verify
 999          * that userspace memory pages are contiguous before we issue command.
1000          */
1001         if (get_num_contig_pages(0, pages, n) != n) {
1002                 ret = -EINVAL;
1003                 goto e_unpin_memory;
1004         }
1005
1006         memset(&data, 0, sizeof(data));
1007
1008         offset = params.guest_uaddr & (PAGE_SIZE - 1);
1009         data.guest_address = __sme_page_pa(pages[0]) + offset;
1010         data.guest_len = params.guest_len;
1011
1012         blob = psp_copy_user_blob(params.trans_uaddr, params.trans_len);
1013         if (IS_ERR(blob)) {
1014                 ret = PTR_ERR(blob);
1015                 goto e_unpin_memory;
1016         }
1017
1018         data.trans_address = __psp_pa(blob);
1019         data.trans_len = params.trans_len;
1020
1021         hdr = psp_copy_user_blob(params.hdr_uaddr, params.hdr_len);
1022         if (IS_ERR(hdr)) {
1023                 ret = PTR_ERR(hdr);
1024                 goto e_free_blob;
1025         }
1026         data.hdr_address = __psp_pa(hdr);
1027         data.hdr_len = params.hdr_len;
1028
1029         data.handle = sev->handle;
1030         ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_SECRET, &data, &argp->error);
1031
1032         kfree(hdr);
1033
1034 e_free_blob:
1035         kfree(blob);
1036 e_unpin_memory:
1037         /* content of memory is updated, mark pages dirty */
1038         for (i = 0; i < n; i++) {
1039                 set_page_dirty_lock(pages[i]);
1040                 mark_page_accessed(pages[i]);
1041         }
1042         sev_unpin_memory(kvm, pages, n);
1043         return ret;
1044 }
1045
1046 static int sev_get_attestation_report(struct kvm *kvm, struct kvm_sev_cmd *argp)
1047 {
1048         void __user *report = (void __user *)(uintptr_t)argp->data;
1049         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1050         struct sev_data_attestation_report data;
1051         struct kvm_sev_attestation_report params;
1052         void __user *p;
1053         void *blob = NULL;
1054         int ret;
1055
1056         if (!sev_guest(kvm))
1057                 return -ENOTTY;
1058
1059         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data, sizeof(params)))
1060                 return -EFAULT;
1061
1062         memset(&data, 0, sizeof(data));
1063
1064         /* User wants to query the blob length */
1065         if (!params.len)
1066                 goto cmd;
1067
1068         p = (void __user *)(uintptr_t)params.uaddr;
1069         if (p) {
1070                 if (params.len > SEV_FW_BLOB_MAX_SIZE)
1071                         return -EINVAL;
1072
1073                 blob = kmalloc(params.len, GFP_KERNEL_ACCOUNT);
1074                 if (!blob)
1075                         return -ENOMEM;
1076
1077                 data.address = __psp_pa(blob);
1078                 data.len = params.len;
1079                 memcpy(data.mnonce, params.mnonce, sizeof(params.mnonce));
1080         }
1081 cmd:
1082         data.handle = sev->handle;
1083         ret = sev_issue_cmd(kvm, SEV_CMD_ATTESTATION_REPORT, &data, &argp->error);
1084         /*
1085          * If we query the session length, FW responded with expected data.
1086          */
1087         if (!params.len)
1088                 goto done;
1089
1090         if (ret)
1091                 goto e_free_blob;
1092
1093         if (blob) {
1094                 if (copy_to_user(p, blob, params.len))
1095                         ret = -EFAULT;
1096         }
1097
1098 done:
1099         params.len = data.len;
1100         if (copy_to_user(report, &params, sizeof(params)))
1101                 ret = -EFAULT;
1102 e_free_blob:
1103         kfree(blob);
1104         return ret;
1105 }
1106
1107 /* Userspace wants to query session length. */
1108 static int
1109 __sev_send_start_query_session_length(struct kvm *kvm, struct kvm_sev_cmd *argp,
1110                                       struct kvm_sev_send_start *params)
1111 {
1112         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1113         struct sev_data_send_start data;
1114         int ret;
1115
1116         memset(&data, 0, sizeof(data));
1117         data.handle = sev->handle;
1118         ret = sev_issue_cmd(kvm, SEV_CMD_SEND_START, &data, &argp->error);
1119
1120         params->session_len = data.session_len;
1121         if (copy_to_user((void __user *)(uintptr_t)argp->data, params,
1122                                 sizeof(struct kvm_sev_send_start)))
1123                 ret = -EFAULT;
1124
1125         return ret;
1126 }
1127
1128 static int sev_send_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
1129 {
1130         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1131         struct sev_data_send_start data;
1132         struct kvm_sev_send_start params;
1133         void *amd_certs, *session_data;
1134         void *pdh_cert, *plat_certs;
1135         int ret;
1136
1137         if (!sev_guest(kvm))
1138                 return -ENOTTY;
1139
1140         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data,
1141                                 sizeof(struct kvm_sev_send_start)))
1142                 return -EFAULT;
1143
1144         /* if session_len is zero, userspace wants to query the session length */
1145         if (!params.session_len)
1146                 return __sev_send_start_query_session_length(kvm, argp,
1147                                 &params);
1148
1149         /* some sanity checks */
1150         if (!params.pdh_cert_uaddr || !params.pdh_cert_len ||
1151             !params.session_uaddr || params.session_len > SEV_FW_BLOB_MAX_SIZE)
1152                 return -EINVAL;
1153
1154         /* allocate the memory to hold the session data blob */
1155         session_data = kmalloc(params.session_len, GFP_KERNEL_ACCOUNT);
1156         if (!session_data)
1157                 return -ENOMEM;
1158
1159         /* copy the certificate blobs from userspace */
1160         pdh_cert = psp_copy_user_blob(params.pdh_cert_uaddr,
1161                                 params.pdh_cert_len);
1162         if (IS_ERR(pdh_cert)) {
1163                 ret = PTR_ERR(pdh_cert);
1164                 goto e_free_session;
1165         }
1166
1167         plat_certs = psp_copy_user_blob(params.plat_certs_uaddr,
1168                                 params.plat_certs_len);
1169         if (IS_ERR(plat_certs)) {
1170                 ret = PTR_ERR(plat_certs);
1171                 goto e_free_pdh;
1172         }
1173
1174         amd_certs = psp_copy_user_blob(params.amd_certs_uaddr,
1175                                 params.amd_certs_len);
1176         if (IS_ERR(amd_certs)) {
1177                 ret = PTR_ERR(amd_certs);
1178                 goto e_free_plat_cert;
1179         }
1180
1181         /* populate the FW SEND_START field with system physical address */
1182         memset(&data, 0, sizeof(data));
1183         data.pdh_cert_address = __psp_pa(pdh_cert);
1184         data.pdh_cert_len = params.pdh_cert_len;
1185         data.plat_certs_address = __psp_pa(plat_certs);
1186         data.plat_certs_len = params.plat_certs_len;
1187         data.amd_certs_address = __psp_pa(amd_certs);
1188         data.amd_certs_len = params.amd_certs_len;
1189         data.session_address = __psp_pa(session_data);
1190         data.session_len = params.session_len;
1191         data.handle = sev->handle;
1192
1193         ret = sev_issue_cmd(kvm, SEV_CMD_SEND_START, &data, &argp->error);
1194
1195         if (!ret && copy_to_user((void __user *)(uintptr_t)params.session_uaddr,
1196                         session_data, params.session_len)) {
1197                 ret = -EFAULT;
1198                 goto e_free_amd_cert;
1199         }
1200
1201         params.policy = data.policy;
1202         params.session_len = data.session_len;
1203         if (copy_to_user((void __user *)(uintptr_t)argp->data, &params,
1204                                 sizeof(struct kvm_sev_send_start)))
1205                 ret = -EFAULT;
1206
1207 e_free_amd_cert:
1208         kfree(amd_certs);
1209 e_free_plat_cert:
1210         kfree(plat_certs);
1211 e_free_pdh:
1212         kfree(pdh_cert);
1213 e_free_session:
1214         kfree(session_data);
1215         return ret;
1216 }
1217
1218 /* Userspace wants to query either header or trans length. */
1219 static int
1220 __sev_send_update_data_query_lengths(struct kvm *kvm, struct kvm_sev_cmd *argp,
1221                                      struct kvm_sev_send_update_data *params)
1222 {
1223         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1224         struct sev_data_send_update_data data;
1225         int ret;
1226
1227         memset(&data, 0, sizeof(data));
1228         data.handle = sev->handle;
1229         ret = sev_issue_cmd(kvm, SEV_CMD_SEND_UPDATE_DATA, &data, &argp->error);
1230
1231         params->hdr_len = data.hdr_len;
1232         params->trans_len = data.trans_len;
1233
1234         if (copy_to_user((void __user *)(uintptr_t)argp->data, params,
1235                          sizeof(struct kvm_sev_send_update_data)))
1236                 ret = -EFAULT;
1237
1238         return ret;
1239 }
1240
1241 static int sev_send_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
1242 {
1243         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1244         struct sev_data_send_update_data data;
1245         struct kvm_sev_send_update_data params;
1246         void *hdr, *trans_data;
1247         struct page **guest_page;
1248         unsigned long n;
1249         int ret, offset;
1250
1251         if (!sev_guest(kvm))
1252                 return -ENOTTY;
1253
1254         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data,
1255                         sizeof(struct kvm_sev_send_update_data)))
1256                 return -EFAULT;
1257
1258         /* userspace wants to query either header or trans length */
1259         if (!params.trans_len || !params.hdr_len)
1260                 return __sev_send_update_data_query_lengths(kvm, argp, &params);
1261
1262         if (!params.trans_uaddr || !params.guest_uaddr ||
1263             !params.guest_len || !params.hdr_uaddr)
1264                 return -EINVAL;
1265
1266         /* Check if we are crossing the page boundary */
1267         offset = params.guest_uaddr & (PAGE_SIZE - 1);
1268         if ((params.guest_len + offset > PAGE_SIZE))
1269                 return -EINVAL;
1270
1271         /* Pin guest memory */
1272         guest_page = sev_pin_memory(kvm, params.guest_uaddr & PAGE_MASK,
1273                                     PAGE_SIZE, &n, 0);
1274         if (!guest_page)
1275                 return -EFAULT;
1276
1277         /* allocate memory for header and transport buffer */
1278         ret = -ENOMEM;
1279         hdr = kmalloc(params.hdr_len, GFP_KERNEL_ACCOUNT);
1280         if (!hdr)
1281                 goto e_unpin;
1282
1283         trans_data = kmalloc(params.trans_len, GFP_KERNEL_ACCOUNT);
1284         if (!trans_data)
1285                 goto e_free_hdr;
1286
1287         memset(&data, 0, sizeof(data));
1288         data.hdr_address = __psp_pa(hdr);
1289         data.hdr_len = params.hdr_len;
1290         data.trans_address = __psp_pa(trans_data);
1291         data.trans_len = params.trans_len;
1292
1293         /* The SEND_UPDATE_DATA command requires C-bit to be always set. */
1294         data.guest_address = (page_to_pfn(guest_page[0]) << PAGE_SHIFT) + offset;
1295         data.guest_address |= sev_me_mask;
1296         data.guest_len = params.guest_len;
1297         data.handle = sev->handle;
1298
1299         ret = sev_issue_cmd(kvm, SEV_CMD_SEND_UPDATE_DATA, &data, &argp->error);
1300
1301         if (ret)
1302                 goto e_free_trans_data;
1303
1304         /* copy transport buffer to user space */
1305         if (copy_to_user((void __user *)(uintptr_t)params.trans_uaddr,
1306                          trans_data, params.trans_len)) {
1307                 ret = -EFAULT;
1308                 goto e_free_trans_data;
1309         }
1310
1311         /* Copy packet header to userspace. */
1312         ret = copy_to_user((void __user *)(uintptr_t)params.hdr_uaddr, hdr,
1313                                 params.hdr_len);
1314
1315 e_free_trans_data:
1316         kfree(trans_data);
1317 e_free_hdr:
1318         kfree(hdr);
1319 e_unpin:
1320         sev_unpin_memory(kvm, guest_page, n);
1321
1322         return ret;
1323 }
1324
1325 static int sev_send_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
1326 {
1327         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1328         struct sev_data_send_finish data;
1329
1330         if (!sev_guest(kvm))
1331                 return -ENOTTY;
1332
1333         data.handle = sev->handle;
1334         return sev_issue_cmd(kvm, SEV_CMD_SEND_FINISH, &data, &argp->error);
1335 }
1336
1337 static int sev_send_cancel(struct kvm *kvm, struct kvm_sev_cmd *argp)
1338 {
1339         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1340         struct sev_data_send_cancel data;
1341
1342         if (!sev_guest(kvm))
1343                 return -ENOTTY;
1344
1345         data.handle = sev->handle;
1346         return sev_issue_cmd(kvm, SEV_CMD_SEND_CANCEL, &data, &argp->error);
1347 }
1348
1349 static int sev_receive_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
1350 {
1351         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1352         struct sev_data_receive_start start;
1353         struct kvm_sev_receive_start params;
1354         int *error = &argp->error;
1355         void *session_data;
1356         void *pdh_data;
1357         int ret;
1358
1359         if (!sev_guest(kvm))
1360                 return -ENOTTY;
1361
1362         /* Get parameter from the userspace */
1363         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data,
1364                         sizeof(struct kvm_sev_receive_start)))
1365                 return -EFAULT;
1366
1367         /* some sanity checks */
1368         if (!params.pdh_uaddr || !params.pdh_len ||
1369             !params.session_uaddr || !params.session_len)
1370                 return -EINVAL;
1371
1372         pdh_data = psp_copy_user_blob(params.pdh_uaddr, params.pdh_len);
1373         if (IS_ERR(pdh_data))
1374                 return PTR_ERR(pdh_data);
1375
1376         session_data = psp_copy_user_blob(params.session_uaddr,
1377                         params.session_len);
1378         if (IS_ERR(session_data)) {
1379                 ret = PTR_ERR(session_data);
1380                 goto e_free_pdh;
1381         }
1382
1383         memset(&start, 0, sizeof(start));
1384         start.handle = params.handle;
1385         start.policy = params.policy;
1386         start.pdh_cert_address = __psp_pa(pdh_data);
1387         start.pdh_cert_len = params.pdh_len;
1388         start.session_address = __psp_pa(session_data);
1389         start.session_len = params.session_len;
1390
1391         /* create memory encryption context */
1392         ret = __sev_issue_cmd(argp->sev_fd, SEV_CMD_RECEIVE_START, &start,
1393                                 error);
1394         if (ret)
1395                 goto e_free_session;
1396
1397         /* Bind ASID to this guest */
1398         ret = sev_bind_asid(kvm, start.handle, error);
1399         if (ret)
1400                 goto e_free_session;
1401
1402         params.handle = start.handle;
1403         if (copy_to_user((void __user *)(uintptr_t)argp->data,
1404                          &params, sizeof(struct kvm_sev_receive_start))) {
1405                 ret = -EFAULT;
1406                 sev_unbind_asid(kvm, start.handle);
1407                 goto e_free_session;
1408         }
1409
1410         sev->handle = start.handle;
1411         sev->fd = argp->sev_fd;
1412
1413 e_free_session:
1414         kfree(session_data);
1415 e_free_pdh:
1416         kfree(pdh_data);
1417
1418         return ret;
1419 }
1420
1421 static int sev_receive_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
1422 {
1423         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1424         struct kvm_sev_receive_update_data params;
1425         struct sev_data_receive_update_data data;
1426         void *hdr = NULL, *trans = NULL;
1427         struct page **guest_page;
1428         unsigned long n;
1429         int ret, offset;
1430
1431         if (!sev_guest(kvm))
1432                 return -EINVAL;
1433
1434         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data,
1435                         sizeof(struct kvm_sev_receive_update_data)))
1436                 return -EFAULT;
1437
1438         if (!params.hdr_uaddr || !params.hdr_len ||
1439             !params.guest_uaddr || !params.guest_len ||
1440             !params.trans_uaddr || !params.trans_len)
1441                 return -EINVAL;
1442
1443         /* Check if we are crossing the page boundary */
1444         offset = params.guest_uaddr & (PAGE_SIZE - 1);
1445         if ((params.guest_len + offset > PAGE_SIZE))
1446                 return -EINVAL;
1447
1448         hdr = psp_copy_user_blob(params.hdr_uaddr, params.hdr_len);
1449         if (IS_ERR(hdr))
1450                 return PTR_ERR(hdr);
1451
1452         trans = psp_copy_user_blob(params.trans_uaddr, params.trans_len);
1453         if (IS_ERR(trans)) {
1454                 ret = PTR_ERR(trans);
1455                 goto e_free_hdr;
1456         }
1457
1458         memset(&data, 0, sizeof(data));
1459         data.hdr_address = __psp_pa(hdr);
1460         data.hdr_len = params.hdr_len;
1461         data.trans_address = __psp_pa(trans);
1462         data.trans_len = params.trans_len;
1463
1464         /* Pin guest memory */
1465         ret = -EFAULT;
1466         guest_page = sev_pin_memory(kvm, params.guest_uaddr & PAGE_MASK,
1467                                     PAGE_SIZE, &n, 0);
1468         if (!guest_page)
1469                 goto e_free_trans;
1470
1471         /* The RECEIVE_UPDATE_DATA command requires C-bit to be always set. */
1472         data.guest_address = (page_to_pfn(guest_page[0]) << PAGE_SHIFT) + offset;
1473         data.guest_address |= sev_me_mask;
1474         data.guest_len = params.guest_len;
1475         data.handle = sev->handle;
1476
1477         ret = sev_issue_cmd(kvm, SEV_CMD_RECEIVE_UPDATE_DATA, &data,
1478                                 &argp->error);
1479
1480         sev_unpin_memory(kvm, guest_page, n);
1481
1482 e_free_trans:
1483         kfree(trans);
1484 e_free_hdr:
1485         kfree(hdr);
1486
1487         return ret;
1488 }
1489
1490 static int sev_receive_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
1491 {
1492         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1493         struct sev_data_receive_finish data;
1494
1495         if (!sev_guest(kvm))
1496                 return -ENOTTY;
1497
1498         data.handle = sev->handle;
1499         return sev_issue_cmd(kvm, SEV_CMD_RECEIVE_FINISH, &data, &argp->error);
1500 }
1501
1502 int svm_mem_enc_op(struct kvm *kvm, void __user *argp)
1503 {
1504         struct kvm_sev_cmd sev_cmd;
1505         int r;
1506
1507         if (!sev_enabled)
1508                 return -ENOTTY;
1509
1510         if (!argp)
1511                 return 0;
1512
1513         if (copy_from_user(&sev_cmd, argp, sizeof(struct kvm_sev_cmd)))
1514                 return -EFAULT;
1515
1516         mutex_lock(&kvm->lock);
1517
1518         /* enc_context_owner handles all memory enc operations */
1519         if (is_mirroring_enc_context(kvm)) {
1520                 r = -EINVAL;
1521                 goto out;
1522         }
1523
1524         switch (sev_cmd.id) {
1525         case KVM_SEV_ES_INIT:
1526                 if (!sev_es_enabled) {
1527                         r = -ENOTTY;
1528                         goto out;
1529                 }
1530                 fallthrough;
1531         case KVM_SEV_INIT:
1532                 r = sev_guest_init(kvm, &sev_cmd);
1533                 break;
1534         case KVM_SEV_LAUNCH_START:
1535                 r = sev_launch_start(kvm, &sev_cmd);
1536                 break;
1537         case KVM_SEV_LAUNCH_UPDATE_DATA:
1538                 r = sev_launch_update_data(kvm, &sev_cmd);
1539                 break;
1540         case KVM_SEV_LAUNCH_UPDATE_VMSA:
1541                 r = sev_launch_update_vmsa(kvm, &sev_cmd);
1542                 break;
1543         case KVM_SEV_LAUNCH_MEASURE:
1544                 r = sev_launch_measure(kvm, &sev_cmd);
1545                 break;
1546         case KVM_SEV_LAUNCH_FINISH:
1547                 r = sev_launch_finish(kvm, &sev_cmd);
1548                 break;
1549         case KVM_SEV_GUEST_STATUS:
1550                 r = sev_guest_status(kvm, &sev_cmd);
1551                 break;
1552         case KVM_SEV_DBG_DECRYPT:
1553                 r = sev_dbg_crypt(kvm, &sev_cmd, true);
1554                 break;
1555         case KVM_SEV_DBG_ENCRYPT:
1556                 r = sev_dbg_crypt(kvm, &sev_cmd, false);
1557                 break;
1558         case KVM_SEV_LAUNCH_SECRET:
1559                 r = sev_launch_secret(kvm, &sev_cmd);
1560                 break;
1561         case KVM_SEV_GET_ATTESTATION_REPORT:
1562                 r = sev_get_attestation_report(kvm, &sev_cmd);
1563                 break;
1564         case KVM_SEV_SEND_START:
1565                 r = sev_send_start(kvm, &sev_cmd);
1566                 break;
1567         case KVM_SEV_SEND_UPDATE_DATA:
1568                 r = sev_send_update_data(kvm, &sev_cmd);
1569                 break;
1570         case KVM_SEV_SEND_FINISH:
1571                 r = sev_send_finish(kvm, &sev_cmd);
1572                 break;
1573         case KVM_SEV_SEND_CANCEL:
1574                 r = sev_send_cancel(kvm, &sev_cmd);
1575                 break;
1576         case KVM_SEV_RECEIVE_START:
1577                 r = sev_receive_start(kvm, &sev_cmd);
1578                 break;
1579         case KVM_SEV_RECEIVE_UPDATE_DATA:
1580                 r = sev_receive_update_data(kvm, &sev_cmd);
1581                 break;
1582         case KVM_SEV_RECEIVE_FINISH:
1583                 r = sev_receive_finish(kvm, &sev_cmd);
1584                 break;
1585         default:
1586                 r = -EINVAL;
1587                 goto out;
1588         }
1589
1590         if (copy_to_user(argp, &sev_cmd, sizeof(struct kvm_sev_cmd)))
1591                 r = -EFAULT;
1592
1593 out:
1594         mutex_unlock(&kvm->lock);
1595         return r;
1596 }
1597
1598 int svm_register_enc_region(struct kvm *kvm,
1599                             struct kvm_enc_region *range)
1600 {
1601         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1602         struct enc_region *region;
1603         int ret = 0;
1604
1605         if (!sev_guest(kvm))
1606                 return -ENOTTY;
1607
1608         /* If kvm is mirroring encryption context it isn't responsible for it */
1609         if (is_mirroring_enc_context(kvm))
1610                 return -EINVAL;
1611
1612         if (range->addr > ULONG_MAX || range->size > ULONG_MAX)
1613                 return -EINVAL;
1614
1615         region = kzalloc(sizeof(*region), GFP_KERNEL_ACCOUNT);
1616         if (!region)
1617                 return -ENOMEM;
1618
1619         mutex_lock(&kvm->lock);
1620         region->pages = sev_pin_memory(kvm, range->addr, range->size, &region->npages, 1);
1621         if (IS_ERR(region->pages)) {
1622                 ret = PTR_ERR(region->pages);
1623                 mutex_unlock(&kvm->lock);
1624                 goto e_free;
1625         }
1626
1627         region->uaddr = range->addr;
1628         region->size = range->size;
1629
1630         list_add_tail(&region->list, &sev->regions_list);
1631         mutex_unlock(&kvm->lock);
1632
1633         /*
1634          * The guest may change the memory encryption attribute from C=0 -> C=1
1635          * or vice versa for this memory range. Lets make sure caches are
1636          * flushed to ensure that guest data gets written into memory with
1637          * correct C-bit.
1638          */
1639         sev_clflush_pages(region->pages, region->npages);
1640
1641         return ret;
1642
1643 e_free:
1644         kfree(region);
1645         return ret;
1646 }
1647
1648 static struct enc_region *
1649 find_enc_region(struct kvm *kvm, struct kvm_enc_region *range)
1650 {
1651         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1652         struct list_head *head = &sev->regions_list;
1653         struct enc_region *i;
1654
1655         list_for_each_entry(i, head, list) {
1656                 if (i->uaddr == range->addr &&
1657                     i->size == range->size)
1658                         return i;
1659         }
1660
1661         return NULL;
1662 }
1663
1664 static void __unregister_enc_region_locked(struct kvm *kvm,
1665                                            struct enc_region *region)
1666 {
1667         sev_unpin_memory(kvm, region->pages, region->npages);
1668         list_del(&region->list);
1669         kfree(region);
1670 }
1671
1672 int svm_unregister_enc_region(struct kvm *kvm,
1673                               struct kvm_enc_region *range)
1674 {
1675         struct enc_region *region;
1676         int ret;
1677
1678         /* If kvm is mirroring encryption context it isn't responsible for it */
1679         if (is_mirroring_enc_context(kvm))
1680                 return -EINVAL;
1681
1682         mutex_lock(&kvm->lock);
1683
1684         if (!sev_guest(kvm)) {
1685                 ret = -ENOTTY;
1686                 goto failed;
1687         }
1688
1689         region = find_enc_region(kvm, range);
1690         if (!region) {
1691                 ret = -EINVAL;
1692                 goto failed;
1693         }
1694
1695         /*
1696          * Ensure that all guest tagged cache entries are flushed before
1697          * releasing the pages back to the system for use. CLFLUSH will
1698          * not do this, so issue a WBINVD.
1699          */
1700         wbinvd_on_all_cpus();
1701
1702         __unregister_enc_region_locked(kvm, region);
1703
1704         mutex_unlock(&kvm->lock);
1705         return 0;
1706
1707 failed:
1708         mutex_unlock(&kvm->lock);
1709         return ret;
1710 }
1711
1712 int svm_vm_copy_asid_from(struct kvm *kvm, unsigned int source_fd)
1713 {
1714         struct file *source_kvm_file;
1715         struct kvm *source_kvm;
1716         struct kvm_sev_info *mirror_sev;
1717         unsigned int asid;
1718         int ret;
1719
1720         source_kvm_file = fget(source_fd);
1721         if (!file_is_kvm(source_kvm_file)) {
1722                 ret = -EBADF;
1723                 goto e_source_put;
1724         }
1725
1726         source_kvm = source_kvm_file->private_data;
1727         mutex_lock(&source_kvm->lock);
1728
1729         if (!sev_guest(source_kvm)) {
1730                 ret = -EINVAL;
1731                 goto e_source_unlock;
1732         }
1733
1734         /* Mirrors of mirrors should work, but let's not get silly */
1735         if (is_mirroring_enc_context(source_kvm) || source_kvm == kvm) {
1736                 ret = -EINVAL;
1737                 goto e_source_unlock;
1738         }
1739
1740         asid = to_kvm_svm(source_kvm)->sev_info.asid;
1741
1742         /*
1743          * The mirror kvm holds an enc_context_owner ref so its asid can't
1744          * disappear until we're done with it
1745          */
1746         kvm_get_kvm(source_kvm);
1747
1748         fput(source_kvm_file);
1749         mutex_unlock(&source_kvm->lock);
1750         mutex_lock(&kvm->lock);
1751
1752         if (sev_guest(kvm)) {
1753                 ret = -EINVAL;
1754                 goto e_mirror_unlock;
1755         }
1756
1757         /* Set enc_context_owner and copy its encryption context over */
1758         mirror_sev = &to_kvm_svm(kvm)->sev_info;
1759         mirror_sev->enc_context_owner = source_kvm;
1760         mirror_sev->asid = asid;
1761         mirror_sev->active = true;
1762
1763         mutex_unlock(&kvm->lock);
1764         return 0;
1765
1766 e_mirror_unlock:
1767         mutex_unlock(&kvm->lock);
1768         kvm_put_kvm(source_kvm);
1769         return ret;
1770 e_source_unlock:
1771         mutex_unlock(&source_kvm->lock);
1772 e_source_put:
1773         if (source_kvm_file)
1774                 fput(source_kvm_file);
1775         return ret;
1776 }
1777
1778 void sev_vm_destroy(struct kvm *kvm)
1779 {
1780         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1781         struct list_head *head = &sev->regions_list;
1782         struct list_head *pos, *q;
1783
1784         if (!sev_guest(kvm))
1785                 return;
1786
1787         /* If this is a mirror_kvm release the enc_context_owner and skip sev cleanup */
1788         if (is_mirroring_enc_context(kvm)) {
1789                 kvm_put_kvm(sev->enc_context_owner);
1790                 return;
1791         }
1792
1793         mutex_lock(&kvm->lock);
1794
1795         /*
1796          * Ensure that all guest tagged cache entries are flushed before
1797          * releasing the pages back to the system for use. CLFLUSH will
1798          * not do this, so issue a WBINVD.
1799          */
1800         wbinvd_on_all_cpus();
1801
1802         /*
1803          * if userspace was terminated before unregistering the memory regions
1804          * then lets unpin all the registered memory.
1805          */
1806         if (!list_empty(head)) {
1807                 list_for_each_safe(pos, q, head) {
1808                         __unregister_enc_region_locked(kvm,
1809                                 list_entry(pos, struct enc_region, list));
1810                         cond_resched();
1811                 }
1812         }
1813
1814         mutex_unlock(&kvm->lock);
1815
1816         sev_unbind_asid(kvm, sev->handle);
1817         sev_asid_free(sev);
1818 }
1819
1820 void __init sev_set_cpu_caps(void)
1821 {
1822         if (!sev_enabled)
1823                 kvm_cpu_cap_clear(X86_FEATURE_SEV);
1824         if (!sev_es_enabled)
1825                 kvm_cpu_cap_clear(X86_FEATURE_SEV_ES);
1826 }
1827
1828 void __init sev_hardware_setup(void)
1829 {
1830 #ifdef CONFIG_KVM_AMD_SEV
1831         unsigned int eax, ebx, ecx, edx, sev_asid_count, sev_es_asid_count;
1832         bool sev_es_supported = false;
1833         bool sev_supported = false;
1834
1835         if (!sev_enabled || !npt_enabled)
1836                 goto out;
1837
1838         /* Does the CPU support SEV? */
1839         if (!boot_cpu_has(X86_FEATURE_SEV))
1840                 goto out;
1841
1842         /* Retrieve SEV CPUID information */
1843         cpuid(0x8000001f, &eax, &ebx, &ecx, &edx);
1844
1845         /* Set encryption bit location for SEV-ES guests */
1846         sev_enc_bit = ebx & 0x3f;
1847
1848         /* Maximum number of encrypted guests supported simultaneously */
1849         max_sev_asid = ecx;
1850         if (!max_sev_asid)
1851                 goto out;
1852
1853         /* Minimum ASID value that should be used for SEV guest */
1854         min_sev_asid = edx;
1855         sev_me_mask = 1UL << (ebx & 0x3f);
1856
1857         /* Initialize SEV ASID bitmaps */
1858         sev_asid_bitmap = bitmap_zalloc(max_sev_asid, GFP_KERNEL);
1859         if (!sev_asid_bitmap)
1860                 goto out;
1861
1862         sev_reclaim_asid_bitmap = bitmap_zalloc(max_sev_asid, GFP_KERNEL);
1863         if (!sev_reclaim_asid_bitmap) {
1864                 bitmap_free(sev_asid_bitmap);
1865                 sev_asid_bitmap = NULL;
1866                 goto out;
1867         }
1868
1869         sev_asid_count = max_sev_asid - min_sev_asid + 1;
1870         if (misc_cg_set_capacity(MISC_CG_RES_SEV, sev_asid_count))
1871                 goto out;
1872
1873         pr_info("SEV supported: %u ASIDs\n", sev_asid_count);
1874         sev_supported = true;
1875
1876         /* SEV-ES support requested? */
1877         if (!sev_es_enabled)
1878                 goto out;
1879
1880         /* Does the CPU support SEV-ES? */
1881         if (!boot_cpu_has(X86_FEATURE_SEV_ES))
1882                 goto out;
1883
1884         /* Has the system been allocated ASIDs for SEV-ES? */
1885         if (min_sev_asid == 1)
1886                 goto out;
1887
1888         sev_es_asid_count = min_sev_asid - 1;
1889         if (misc_cg_set_capacity(MISC_CG_RES_SEV_ES, sev_es_asid_count))
1890                 goto out;
1891
1892         pr_info("SEV-ES supported: %u ASIDs\n", sev_es_asid_count);
1893         sev_es_supported = true;
1894
1895 out:
1896         sev_enabled = sev_supported;
1897         sev_es_enabled = sev_es_supported;
1898 #endif
1899 }
1900
1901 void sev_hardware_teardown(void)
1902 {
1903         if (!sev_enabled)
1904                 return;
1905
1906         /* No need to take sev_bitmap_lock, all VMs have been destroyed. */
1907         sev_flush_asids(0, max_sev_asid);
1908
1909         bitmap_free(sev_asid_bitmap);
1910         bitmap_free(sev_reclaim_asid_bitmap);
1911
1912         misc_cg_set_capacity(MISC_CG_RES_SEV, 0);
1913         misc_cg_set_capacity(MISC_CG_RES_SEV_ES, 0);
1914 }
1915
1916 int sev_cpu_init(struct svm_cpu_data *sd)
1917 {
1918         if (!sev_enabled)
1919                 return 0;
1920
1921         sd->sev_vmcbs = kcalloc(max_sev_asid + 1, sizeof(void *), GFP_KERNEL);
1922         if (!sd->sev_vmcbs)
1923                 return -ENOMEM;
1924
1925         return 0;
1926 }
1927
1928 /*
1929  * Pages used by hardware to hold guest encrypted state must be flushed before
1930  * returning them to the system.
1931  */
1932 static void sev_flush_guest_memory(struct vcpu_svm *svm, void *va,
1933                                    unsigned long len)
1934 {
1935         /*
1936          * If hardware enforced cache coherency for encrypted mappings of the
1937          * same physical page is supported, nothing to do.
1938          */
1939         if (boot_cpu_has(X86_FEATURE_SME_COHERENT))
1940                 return;
1941
1942         /*
1943          * If the VM Page Flush MSR is supported, use it to flush the page
1944          * (using the page virtual address and the guest ASID).
1945          */
1946         if (boot_cpu_has(X86_FEATURE_VM_PAGE_FLUSH)) {
1947                 struct kvm_sev_info *sev;
1948                 unsigned long va_start;
1949                 u64 start, stop;
1950
1951                 /* Align start and stop to page boundaries. */
1952                 va_start = (unsigned long)va;
1953                 start = (u64)va_start & PAGE_MASK;
1954                 stop = PAGE_ALIGN((u64)va_start + len);
1955
1956                 if (start < stop) {
1957                         sev = &to_kvm_svm(svm->vcpu.kvm)->sev_info;
1958
1959                         while (start < stop) {
1960                                 wrmsrl(MSR_AMD64_VM_PAGE_FLUSH,
1961                                        start | sev->asid);
1962
1963                                 start += PAGE_SIZE;
1964                         }
1965
1966                         return;
1967                 }
1968
1969                 WARN(1, "Address overflow, using WBINVD\n");
1970         }
1971
1972         /*
1973          * Hardware should always have one of the above features,
1974          * but if not, use WBINVD and issue a warning.
1975          */
1976         WARN_ONCE(1, "Using WBINVD to flush guest memory\n");
1977         wbinvd_on_all_cpus();
1978 }
1979
1980 void sev_free_vcpu(struct kvm_vcpu *vcpu)
1981 {
1982         struct vcpu_svm *svm;
1983
1984         if (!sev_es_guest(vcpu->kvm))
1985                 return;
1986
1987         svm = to_svm(vcpu);
1988
1989         if (vcpu->arch.guest_state_protected)
1990                 sev_flush_guest_memory(svm, svm->vmsa, PAGE_SIZE);
1991         __free_page(virt_to_page(svm->vmsa));
1992
1993         if (svm->ghcb_sa_free)
1994                 kfree(svm->ghcb_sa);
1995 }
1996
1997 static void dump_ghcb(struct vcpu_svm *svm)
1998 {
1999         struct ghcb *ghcb = svm->ghcb;
2000         unsigned int nbits;
2001
2002         /* Re-use the dump_invalid_vmcb module parameter */
2003         if (!dump_invalid_vmcb) {
2004                 pr_warn_ratelimited("set kvm_amd.dump_invalid_vmcb=1 to dump internal KVM state.\n");
2005                 return;
2006         }
2007
2008         nbits = sizeof(ghcb->save.valid_bitmap) * 8;
2009
2010         pr_err("GHCB (GPA=%016llx):\n", svm->vmcb->control.ghcb_gpa);
2011         pr_err("%-20s%016llx is_valid: %u\n", "sw_exit_code",
2012                ghcb->save.sw_exit_code, ghcb_sw_exit_code_is_valid(ghcb));
2013         pr_err("%-20s%016llx is_valid: %u\n", "sw_exit_info_1",
2014                ghcb->save.sw_exit_info_1, ghcb_sw_exit_info_1_is_valid(ghcb));
2015         pr_err("%-20s%016llx is_valid: %u\n", "sw_exit_info_2",
2016                ghcb->save.sw_exit_info_2, ghcb_sw_exit_info_2_is_valid(ghcb));
2017         pr_err("%-20s%016llx is_valid: %u\n", "sw_scratch",
2018                ghcb->save.sw_scratch, ghcb_sw_scratch_is_valid(ghcb));
2019         pr_err("%-20s%*pb\n", "valid_bitmap", nbits, ghcb->save.valid_bitmap);
2020 }
2021
2022 static void sev_es_sync_to_ghcb(struct vcpu_svm *svm)
2023 {
2024         struct kvm_vcpu *vcpu = &svm->vcpu;
2025         struct ghcb *ghcb = svm->ghcb;
2026
2027         /*
2028          * The GHCB protocol so far allows for the following data
2029          * to be returned:
2030          *   GPRs RAX, RBX, RCX, RDX
2031          *
2032          * Copy their values, even if they may not have been written during the
2033          * VM-Exit.  It's the guest's responsibility to not consume random data.
2034          */
2035         ghcb_set_rax(ghcb, vcpu->arch.regs[VCPU_REGS_RAX]);
2036         ghcb_set_rbx(ghcb, vcpu->arch.regs[VCPU_REGS_RBX]);
2037         ghcb_set_rcx(ghcb, vcpu->arch.regs[VCPU_REGS_RCX]);
2038         ghcb_set_rdx(ghcb, vcpu->arch.regs[VCPU_REGS_RDX]);
2039 }
2040
2041 static void sev_es_sync_from_ghcb(struct vcpu_svm *svm)
2042 {
2043         struct vmcb_control_area *control = &svm->vmcb->control;
2044         struct kvm_vcpu *vcpu = &svm->vcpu;
2045         struct ghcb *ghcb = svm->ghcb;
2046         u64 exit_code;
2047
2048         /*
2049          * The GHCB protocol so far allows for the following data
2050          * to be supplied:
2051          *   GPRs RAX, RBX, RCX, RDX
2052          *   XCR0
2053          *   CPL
2054          *
2055          * VMMCALL allows the guest to provide extra registers. KVM also
2056          * expects RSI for hypercalls, so include that, too.
2057          *
2058          * Copy their values to the appropriate location if supplied.
2059          */
2060         memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
2061
2062         vcpu->arch.regs[VCPU_REGS_RAX] = ghcb_get_rax_if_valid(ghcb);
2063         vcpu->arch.regs[VCPU_REGS_RBX] = ghcb_get_rbx_if_valid(ghcb);
2064         vcpu->arch.regs[VCPU_REGS_RCX] = ghcb_get_rcx_if_valid(ghcb);
2065         vcpu->arch.regs[VCPU_REGS_RDX] = ghcb_get_rdx_if_valid(ghcb);
2066         vcpu->arch.regs[VCPU_REGS_RSI] = ghcb_get_rsi_if_valid(ghcb);
2067
2068         svm->vmcb->save.cpl = ghcb_get_cpl_if_valid(ghcb);
2069
2070         if (ghcb_xcr0_is_valid(ghcb)) {
2071                 vcpu->arch.xcr0 = ghcb_get_xcr0(ghcb);
2072                 kvm_update_cpuid_runtime(vcpu);
2073         }
2074
2075         /* Copy the GHCB exit information into the VMCB fields */
2076         exit_code = ghcb_get_sw_exit_code(ghcb);
2077         control->exit_code = lower_32_bits(exit_code);
2078         control->exit_code_hi = upper_32_bits(exit_code);
2079         control->exit_info_1 = ghcb_get_sw_exit_info_1(ghcb);
2080         control->exit_info_2 = ghcb_get_sw_exit_info_2(ghcb);
2081
2082         /* Clear the valid entries fields */
2083         memset(ghcb->save.valid_bitmap, 0, sizeof(ghcb->save.valid_bitmap));
2084 }
2085
2086 static int sev_es_validate_vmgexit(struct vcpu_svm *svm)
2087 {
2088         struct kvm_vcpu *vcpu;
2089         struct ghcb *ghcb;
2090         u64 exit_code = 0;
2091
2092         ghcb = svm->ghcb;
2093
2094         /* Only GHCB Usage code 0 is supported */
2095         if (ghcb->ghcb_usage)
2096                 goto vmgexit_err;
2097
2098         /*
2099          * Retrieve the exit code now even though is may not be marked valid
2100          * as it could help with debugging.
2101          */
2102         exit_code = ghcb_get_sw_exit_code(ghcb);
2103
2104         if (!ghcb_sw_exit_code_is_valid(ghcb) ||
2105             !ghcb_sw_exit_info_1_is_valid(ghcb) ||
2106             !ghcb_sw_exit_info_2_is_valid(ghcb))
2107                 goto vmgexit_err;
2108
2109         switch (ghcb_get_sw_exit_code(ghcb)) {
2110         case SVM_EXIT_READ_DR7:
2111                 break;
2112         case SVM_EXIT_WRITE_DR7:
2113                 if (!ghcb_rax_is_valid(ghcb))
2114                         goto vmgexit_err;
2115                 break;
2116         case SVM_EXIT_RDTSC:
2117                 break;
2118         case SVM_EXIT_RDPMC:
2119                 if (!ghcb_rcx_is_valid(ghcb))
2120                         goto vmgexit_err;
2121                 break;
2122         case SVM_EXIT_CPUID:
2123                 if (!ghcb_rax_is_valid(ghcb) ||
2124                     !ghcb_rcx_is_valid(ghcb))
2125                         goto vmgexit_err;
2126                 if (ghcb_get_rax(ghcb) == 0xd)
2127                         if (!ghcb_xcr0_is_valid(ghcb))
2128                                 goto vmgexit_err;
2129                 break;
2130         case SVM_EXIT_INVD:
2131                 break;
2132         case SVM_EXIT_IOIO:
2133                 if (ghcb_get_sw_exit_info_1(ghcb) & SVM_IOIO_STR_MASK) {
2134                         if (!ghcb_sw_scratch_is_valid(ghcb))
2135                                 goto vmgexit_err;
2136                 } else {
2137                         if (!(ghcb_get_sw_exit_info_1(ghcb) & SVM_IOIO_TYPE_MASK))
2138                                 if (!ghcb_rax_is_valid(ghcb))
2139                                         goto vmgexit_err;
2140                 }
2141                 break;
2142         case SVM_EXIT_MSR:
2143                 if (!ghcb_rcx_is_valid(ghcb))
2144                         goto vmgexit_err;
2145                 if (ghcb_get_sw_exit_info_1(ghcb)) {
2146                         if (!ghcb_rax_is_valid(ghcb) ||
2147                             !ghcb_rdx_is_valid(ghcb))
2148                                 goto vmgexit_err;
2149                 }
2150                 break;
2151         case SVM_EXIT_VMMCALL:
2152                 if (!ghcb_rax_is_valid(ghcb) ||
2153                     !ghcb_cpl_is_valid(ghcb))
2154                         goto vmgexit_err;
2155                 break;
2156         case SVM_EXIT_RDTSCP:
2157                 break;
2158         case SVM_EXIT_WBINVD:
2159                 break;
2160         case SVM_EXIT_MONITOR:
2161                 if (!ghcb_rax_is_valid(ghcb) ||
2162                     !ghcb_rcx_is_valid(ghcb) ||
2163                     !ghcb_rdx_is_valid(ghcb))
2164                         goto vmgexit_err;
2165                 break;
2166         case SVM_EXIT_MWAIT:
2167                 if (!ghcb_rax_is_valid(ghcb) ||
2168                     !ghcb_rcx_is_valid(ghcb))
2169                         goto vmgexit_err;
2170                 break;
2171         case SVM_VMGEXIT_MMIO_READ:
2172         case SVM_VMGEXIT_MMIO_WRITE:
2173                 if (!ghcb_sw_scratch_is_valid(ghcb))
2174                         goto vmgexit_err;
2175                 break;
2176         case SVM_VMGEXIT_NMI_COMPLETE:
2177         case SVM_VMGEXIT_AP_HLT_LOOP:
2178         case SVM_VMGEXIT_AP_JUMP_TABLE:
2179         case SVM_VMGEXIT_UNSUPPORTED_EVENT:
2180                 break;
2181         default:
2182                 goto vmgexit_err;
2183         }
2184
2185         return 0;
2186
2187 vmgexit_err:
2188         vcpu = &svm->vcpu;
2189
2190         if (ghcb->ghcb_usage) {
2191                 vcpu_unimpl(vcpu, "vmgexit: ghcb usage %#x is not valid\n",
2192                             ghcb->ghcb_usage);
2193         } else {
2194                 vcpu_unimpl(vcpu, "vmgexit: exit reason %#llx is not valid\n",
2195                             exit_code);
2196                 dump_ghcb(svm);
2197         }
2198
2199         vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2200         vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON;
2201         vcpu->run->internal.ndata = 2;
2202         vcpu->run->internal.data[0] = exit_code;
2203         vcpu->run->internal.data[1] = vcpu->arch.last_vmentry_cpu;
2204
2205         return -EINVAL;
2206 }
2207
2208 void sev_es_unmap_ghcb(struct vcpu_svm *svm)
2209 {
2210         if (!svm->ghcb)
2211                 return;
2212
2213         if (svm->ghcb_sa_free) {
2214                 /*
2215                  * The scratch area lives outside the GHCB, so there is a
2216                  * buffer that, depending on the operation performed, may
2217                  * need to be synced, then freed.
2218                  */
2219                 if (svm->ghcb_sa_sync) {
2220                         kvm_write_guest(svm->vcpu.kvm,
2221                                         ghcb_get_sw_scratch(svm->ghcb),
2222                                         svm->ghcb_sa, svm->ghcb_sa_len);
2223                         svm->ghcb_sa_sync = false;
2224                 }
2225
2226                 kfree(svm->ghcb_sa);
2227                 svm->ghcb_sa = NULL;
2228                 svm->ghcb_sa_free = false;
2229         }
2230
2231         trace_kvm_vmgexit_exit(svm->vcpu.vcpu_id, svm->ghcb);
2232
2233         sev_es_sync_to_ghcb(svm);
2234
2235         kvm_vcpu_unmap(&svm->vcpu, &svm->ghcb_map, true);
2236         svm->ghcb = NULL;
2237 }
2238
2239 void pre_sev_run(struct vcpu_svm *svm, int cpu)
2240 {
2241         struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
2242         int asid = sev_get_asid(svm->vcpu.kvm);
2243
2244         /* Assign the asid allocated with this SEV guest */
2245         svm->asid = asid;
2246
2247         /*
2248          * Flush guest TLB:
2249          *
2250          * 1) when different VMCB for the same ASID is to be run on the same host CPU.
2251          * 2) or this VMCB was executed on different host CPU in previous VMRUNs.
2252          */
2253         if (sd->sev_vmcbs[asid] == svm->vmcb &&
2254             svm->vcpu.arch.last_vmentry_cpu == cpu)
2255                 return;
2256
2257         sd->sev_vmcbs[asid] = svm->vmcb;
2258         svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
2259         vmcb_mark_dirty(svm->vmcb, VMCB_ASID);
2260 }
2261
2262 #define GHCB_SCRATCH_AREA_LIMIT         (16ULL * PAGE_SIZE)
2263 static bool setup_vmgexit_scratch(struct vcpu_svm *svm, bool sync, u64 len)
2264 {
2265         struct vmcb_control_area *control = &svm->vmcb->control;
2266         struct ghcb *ghcb = svm->ghcb;
2267         u64 ghcb_scratch_beg, ghcb_scratch_end;
2268         u64 scratch_gpa_beg, scratch_gpa_end;
2269         void *scratch_va;
2270
2271         scratch_gpa_beg = ghcb_get_sw_scratch(ghcb);
2272         if (!scratch_gpa_beg) {
2273                 pr_err("vmgexit: scratch gpa not provided\n");
2274                 return false;
2275         }
2276
2277         scratch_gpa_end = scratch_gpa_beg + len;
2278         if (scratch_gpa_end < scratch_gpa_beg) {
2279                 pr_err("vmgexit: scratch length (%#llx) not valid for scratch address (%#llx)\n",
2280                        len, scratch_gpa_beg);
2281                 return false;
2282         }
2283
2284         if ((scratch_gpa_beg & PAGE_MASK) == control->ghcb_gpa) {
2285                 /* Scratch area begins within GHCB */
2286                 ghcb_scratch_beg = control->ghcb_gpa +
2287                                    offsetof(struct ghcb, shared_buffer);
2288                 ghcb_scratch_end = control->ghcb_gpa +
2289                                    offsetof(struct ghcb, reserved_1);
2290
2291                 /*
2292                  * If the scratch area begins within the GHCB, it must be
2293                  * completely contained in the GHCB shared buffer area.
2294                  */
2295                 if (scratch_gpa_beg < ghcb_scratch_beg ||
2296                     scratch_gpa_end > ghcb_scratch_end) {
2297                         pr_err("vmgexit: scratch area is outside of GHCB shared buffer area (%#llx - %#llx)\n",
2298                                scratch_gpa_beg, scratch_gpa_end);
2299                         return false;
2300                 }
2301
2302                 scratch_va = (void *)svm->ghcb;
2303                 scratch_va += (scratch_gpa_beg - control->ghcb_gpa);
2304         } else {
2305                 /*
2306                  * The guest memory must be read into a kernel buffer, so
2307                  * limit the size
2308                  */
2309                 if (len > GHCB_SCRATCH_AREA_LIMIT) {
2310                         pr_err("vmgexit: scratch area exceeds KVM limits (%#llx requested, %#llx limit)\n",
2311                                len, GHCB_SCRATCH_AREA_LIMIT);
2312                         return false;
2313                 }
2314                 scratch_va = kzalloc(len, GFP_KERNEL_ACCOUNT);
2315                 if (!scratch_va)
2316                         return false;
2317
2318                 if (kvm_read_guest(svm->vcpu.kvm, scratch_gpa_beg, scratch_va, len)) {
2319                         /* Unable to copy scratch area from guest */
2320                         pr_err("vmgexit: kvm_read_guest for scratch area failed\n");
2321
2322                         kfree(scratch_va);
2323                         return false;
2324                 }
2325
2326                 /*
2327                  * The scratch area is outside the GHCB. The operation will
2328                  * dictate whether the buffer needs to be synced before running
2329                  * the vCPU next time (i.e. a read was requested so the data
2330                  * must be written back to the guest memory).
2331                  */
2332                 svm->ghcb_sa_sync = sync;
2333                 svm->ghcb_sa_free = true;
2334         }
2335
2336         svm->ghcb_sa = scratch_va;
2337         svm->ghcb_sa_len = len;
2338
2339         return true;
2340 }
2341
2342 static void set_ghcb_msr_bits(struct vcpu_svm *svm, u64 value, u64 mask,
2343                               unsigned int pos)
2344 {
2345         svm->vmcb->control.ghcb_gpa &= ~(mask << pos);
2346         svm->vmcb->control.ghcb_gpa |= (value & mask) << pos;
2347 }
2348
2349 static u64 get_ghcb_msr_bits(struct vcpu_svm *svm, u64 mask, unsigned int pos)
2350 {
2351         return (svm->vmcb->control.ghcb_gpa >> pos) & mask;
2352 }
2353
2354 static void set_ghcb_msr(struct vcpu_svm *svm, u64 value)
2355 {
2356         svm->vmcb->control.ghcb_gpa = value;
2357 }
2358
2359 static int sev_handle_vmgexit_msr_protocol(struct vcpu_svm *svm)
2360 {
2361         struct vmcb_control_area *control = &svm->vmcb->control;
2362         struct kvm_vcpu *vcpu = &svm->vcpu;
2363         u64 ghcb_info;
2364         int ret = 1;
2365
2366         ghcb_info = control->ghcb_gpa & GHCB_MSR_INFO_MASK;
2367
2368         trace_kvm_vmgexit_msr_protocol_enter(svm->vcpu.vcpu_id,
2369                                              control->ghcb_gpa);
2370
2371         switch (ghcb_info) {
2372         case GHCB_MSR_SEV_INFO_REQ:
2373                 set_ghcb_msr(svm, GHCB_MSR_SEV_INFO(GHCB_VERSION_MAX,
2374                                                     GHCB_VERSION_MIN,
2375                                                     sev_enc_bit));
2376                 break;
2377         case GHCB_MSR_CPUID_REQ: {
2378                 u64 cpuid_fn, cpuid_reg, cpuid_value;
2379
2380                 cpuid_fn = get_ghcb_msr_bits(svm,
2381                                              GHCB_MSR_CPUID_FUNC_MASK,
2382                                              GHCB_MSR_CPUID_FUNC_POS);
2383
2384                 /* Initialize the registers needed by the CPUID intercept */
2385                 vcpu->arch.regs[VCPU_REGS_RAX] = cpuid_fn;
2386                 vcpu->arch.regs[VCPU_REGS_RCX] = 0;
2387
2388                 ret = svm_invoke_exit_handler(vcpu, SVM_EXIT_CPUID);
2389                 if (!ret) {
2390                         ret = -EINVAL;
2391                         break;
2392                 }
2393
2394                 cpuid_reg = get_ghcb_msr_bits(svm,
2395                                               GHCB_MSR_CPUID_REG_MASK,
2396                                               GHCB_MSR_CPUID_REG_POS);
2397                 if (cpuid_reg == 0)
2398                         cpuid_value = vcpu->arch.regs[VCPU_REGS_RAX];
2399                 else if (cpuid_reg == 1)
2400                         cpuid_value = vcpu->arch.regs[VCPU_REGS_RBX];
2401                 else if (cpuid_reg == 2)
2402                         cpuid_value = vcpu->arch.regs[VCPU_REGS_RCX];
2403                 else
2404                         cpuid_value = vcpu->arch.regs[VCPU_REGS_RDX];
2405
2406                 set_ghcb_msr_bits(svm, cpuid_value,
2407                                   GHCB_MSR_CPUID_VALUE_MASK,
2408                                   GHCB_MSR_CPUID_VALUE_POS);
2409
2410                 set_ghcb_msr_bits(svm, GHCB_MSR_CPUID_RESP,
2411                                   GHCB_MSR_INFO_MASK,
2412                                   GHCB_MSR_INFO_POS);
2413                 break;
2414         }
2415         case GHCB_MSR_TERM_REQ: {
2416                 u64 reason_set, reason_code;
2417
2418                 reason_set = get_ghcb_msr_bits(svm,
2419                                                GHCB_MSR_TERM_REASON_SET_MASK,
2420                                                GHCB_MSR_TERM_REASON_SET_POS);
2421                 reason_code = get_ghcb_msr_bits(svm,
2422                                                 GHCB_MSR_TERM_REASON_MASK,
2423                                                 GHCB_MSR_TERM_REASON_POS);
2424                 pr_info("SEV-ES guest requested termination: %#llx:%#llx\n",
2425                         reason_set, reason_code);
2426                 fallthrough;
2427         }
2428         default:
2429                 ret = -EINVAL;
2430         }
2431
2432         trace_kvm_vmgexit_msr_protocol_exit(svm->vcpu.vcpu_id,
2433                                             control->ghcb_gpa, ret);
2434
2435         return ret;
2436 }
2437
2438 int sev_handle_vmgexit(struct kvm_vcpu *vcpu)
2439 {
2440         struct vcpu_svm *svm = to_svm(vcpu);
2441         struct vmcb_control_area *control = &svm->vmcb->control;
2442         u64 ghcb_gpa, exit_code;
2443         struct ghcb *ghcb;
2444         int ret;
2445
2446         /* Validate the GHCB */
2447         ghcb_gpa = control->ghcb_gpa;
2448         if (ghcb_gpa & GHCB_MSR_INFO_MASK)
2449                 return sev_handle_vmgexit_msr_protocol(svm);
2450
2451         if (!ghcb_gpa) {
2452                 vcpu_unimpl(vcpu, "vmgexit: GHCB gpa is not set\n");
2453                 return -EINVAL;
2454         }
2455
2456         if (kvm_vcpu_map(vcpu, ghcb_gpa >> PAGE_SHIFT, &svm->ghcb_map)) {
2457                 /* Unable to map GHCB from guest */
2458                 vcpu_unimpl(vcpu, "vmgexit: error mapping GHCB [%#llx] from guest\n",
2459                             ghcb_gpa);
2460                 return -EINVAL;
2461         }
2462
2463         svm->ghcb = svm->ghcb_map.hva;
2464         ghcb = svm->ghcb_map.hva;
2465
2466         trace_kvm_vmgexit_enter(vcpu->vcpu_id, ghcb);
2467
2468         exit_code = ghcb_get_sw_exit_code(ghcb);
2469
2470         ret = sev_es_validate_vmgexit(svm);
2471         if (ret)
2472                 return ret;
2473
2474         sev_es_sync_from_ghcb(svm);
2475         ghcb_set_sw_exit_info_1(ghcb, 0);
2476         ghcb_set_sw_exit_info_2(ghcb, 0);
2477
2478         ret = -EINVAL;
2479         switch (exit_code) {
2480         case SVM_VMGEXIT_MMIO_READ:
2481                 if (!setup_vmgexit_scratch(svm, true, control->exit_info_2))
2482                         break;
2483
2484                 ret = kvm_sev_es_mmio_read(vcpu,
2485                                            control->exit_info_1,
2486                                            control->exit_info_2,
2487                                            svm->ghcb_sa);
2488                 break;
2489         case SVM_VMGEXIT_MMIO_WRITE:
2490                 if (!setup_vmgexit_scratch(svm, false, control->exit_info_2))
2491                         break;
2492
2493                 ret = kvm_sev_es_mmio_write(vcpu,
2494                                             control->exit_info_1,
2495                                             control->exit_info_2,
2496                                             svm->ghcb_sa);
2497                 break;
2498         case SVM_VMGEXIT_NMI_COMPLETE:
2499                 ret = svm_invoke_exit_handler(vcpu, SVM_EXIT_IRET);
2500                 break;
2501         case SVM_VMGEXIT_AP_HLT_LOOP:
2502                 ret = kvm_emulate_ap_reset_hold(vcpu);
2503                 break;
2504         case SVM_VMGEXIT_AP_JUMP_TABLE: {
2505                 struct kvm_sev_info *sev = &to_kvm_svm(vcpu->kvm)->sev_info;
2506
2507                 switch (control->exit_info_1) {
2508                 case 0:
2509                         /* Set AP jump table address */
2510                         sev->ap_jump_table = control->exit_info_2;
2511                         break;
2512                 case 1:
2513                         /* Get AP jump table address */
2514                         ghcb_set_sw_exit_info_2(ghcb, sev->ap_jump_table);
2515                         break;
2516                 default:
2517                         pr_err("svm: vmgexit: unsupported AP jump table request - exit_info_1=%#llx\n",
2518                                control->exit_info_1);
2519                         ghcb_set_sw_exit_info_1(ghcb, 1);
2520                         ghcb_set_sw_exit_info_2(ghcb,
2521                                                 X86_TRAP_UD |
2522                                                 SVM_EVTINJ_TYPE_EXEPT |
2523                                                 SVM_EVTINJ_VALID);
2524                 }
2525
2526                 ret = 1;
2527                 break;
2528         }
2529         case SVM_VMGEXIT_UNSUPPORTED_EVENT:
2530                 vcpu_unimpl(vcpu,
2531                             "vmgexit: unsupported event - exit_info_1=%#llx, exit_info_2=%#llx\n",
2532                             control->exit_info_1, control->exit_info_2);
2533                 break;
2534         default:
2535                 ret = svm_invoke_exit_handler(vcpu, exit_code);
2536         }
2537
2538         return ret;
2539 }
2540
2541 int sev_es_string_io(struct vcpu_svm *svm, int size, unsigned int port, int in)
2542 {
2543         if (!setup_vmgexit_scratch(svm, in, svm->vmcb->control.exit_info_2))
2544                 return -EINVAL;
2545
2546         return kvm_sev_es_string_io(&svm->vcpu, size, port,
2547                                     svm->ghcb_sa, svm->ghcb_sa_len, in);
2548 }
2549
2550 void sev_es_init_vmcb(struct vcpu_svm *svm)
2551 {
2552         struct kvm_vcpu *vcpu = &svm->vcpu;
2553
2554         svm->vmcb->control.nested_ctl |= SVM_NESTED_CTL_SEV_ES_ENABLE;
2555         svm->vmcb->control.virt_ext |= LBR_CTL_ENABLE_MASK;
2556
2557         /*
2558          * An SEV-ES guest requires a VMSA area that is a separate from the
2559          * VMCB page. Do not include the encryption mask on the VMSA physical
2560          * address since hardware will access it using the guest key.
2561          */
2562         svm->vmcb->control.vmsa_pa = __pa(svm->vmsa);
2563
2564         /* Can't intercept CR register access, HV can't modify CR registers */
2565         svm_clr_intercept(svm, INTERCEPT_CR0_READ);
2566         svm_clr_intercept(svm, INTERCEPT_CR4_READ);
2567         svm_clr_intercept(svm, INTERCEPT_CR8_READ);
2568         svm_clr_intercept(svm, INTERCEPT_CR0_WRITE);
2569         svm_clr_intercept(svm, INTERCEPT_CR4_WRITE);
2570         svm_clr_intercept(svm, INTERCEPT_CR8_WRITE);
2571
2572         svm_clr_intercept(svm, INTERCEPT_SELECTIVE_CR0);
2573
2574         /* Track EFER/CR register changes */
2575         svm_set_intercept(svm, TRAP_EFER_WRITE);
2576         svm_set_intercept(svm, TRAP_CR0_WRITE);
2577         svm_set_intercept(svm, TRAP_CR4_WRITE);
2578         svm_set_intercept(svm, TRAP_CR8_WRITE);
2579
2580         /* No support for enable_vmware_backdoor */
2581         clr_exception_intercept(svm, GP_VECTOR);
2582
2583         /* Can't intercept XSETBV, HV can't modify XCR0 directly */
2584         svm_clr_intercept(svm, INTERCEPT_XSETBV);
2585
2586         /* Clear intercepts on selected MSRs */
2587         set_msr_interception(vcpu, svm->msrpm, MSR_EFER, 1, 1);
2588         set_msr_interception(vcpu, svm->msrpm, MSR_IA32_CR_PAT, 1, 1);
2589         set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
2590         set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
2591         set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
2592         set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
2593 }
2594
2595 void sev_es_create_vcpu(struct vcpu_svm *svm)
2596 {
2597         /*
2598          * Set the GHCB MSR value as per the GHCB specification when creating
2599          * a vCPU for an SEV-ES guest.
2600          */
2601         set_ghcb_msr(svm, GHCB_MSR_SEV_INFO(GHCB_VERSION_MAX,
2602                                             GHCB_VERSION_MIN,
2603                                             sev_enc_bit));
2604 }
2605
2606 void sev_es_prepare_guest_switch(struct vcpu_svm *svm, unsigned int cpu)
2607 {
2608         struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
2609         struct vmcb_save_area *hostsa;
2610
2611         /*
2612          * As an SEV-ES guest, hardware will restore the host state on VMEXIT,
2613          * of which one step is to perform a VMLOAD. Since hardware does not
2614          * perform a VMSAVE on VMRUN, the host savearea must be updated.
2615          */
2616         vmsave(__sme_page_pa(sd->save_area));
2617
2618         /* XCR0 is restored on VMEXIT, save the current host value */
2619         hostsa = (struct vmcb_save_area *)(page_address(sd->save_area) + 0x400);
2620         hostsa->xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
2621
2622         /* PKRU is restored on VMEXIT, save the current host value */
2623         hostsa->pkru = read_pkru();
2624
2625         /* MSR_IA32_XSS is restored on VMEXIT, save the currnet host value */
2626         hostsa->xss = host_xss;
2627 }
2628
2629 void sev_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
2630 {
2631         struct vcpu_svm *svm = to_svm(vcpu);
2632
2633         /* First SIPI: Use the values as initially set by the VMM */
2634         if (!svm->received_first_sipi) {
2635                 svm->received_first_sipi = true;
2636                 return;
2637         }
2638
2639         /*
2640          * Subsequent SIPI: Return from an AP Reset Hold VMGEXIT, where
2641          * the guest will set the CS and RIP. Set SW_EXIT_INFO_2 to a
2642          * non-zero value.
2643          */
2644         if (!svm->ghcb)
2645                 return;
2646
2647         ghcb_set_sw_exit_info_2(svm->ghcb, 1);
2648 }