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