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/pkru.h>
  23 #include <asm/trapnr.h>
  24
  25 #include "x86.h"
  26 #include "svm.h"
  27 #include "svm_ops.h"
  28 #include "cpuid.h"
  29 #include "trace.h"
  30
  31 #define __ex(x) __kvm_handle_fault_on_reboot(x)
  32
  33 #ifndef CONFIG_KVM_AMD_SEV
  34 /*
  35  * When this config is not defined, SEV feature is not supported and APIs in
  36  * this file are not used but this file still gets compiled into the KVM AMD
  37  * module.
  38  *
  39  * We will not have MISC_CG_RES_SEV and MISC_CG_RES_SEV_ES entries in the enum
  40  * misc_res_type {} defined in linux/misc_cgroup.h.
  41  *
  42  * Below macros allow compilation to succeed.
  43  */
  44 #define MISC_CG_RES_SEV MISC_CG_RES_TYPES
  45 #define MISC_CG_RES_SEV_ES MISC_CG_RES_TYPES
  46 #endif
  47
  48 #ifdef CONFIG_KVM_AMD_SEV
  49 /* enable/disable SEV support */
  50 static bool sev_enabled = true;
  51 module_param_named(sev, sev_enabled, bool, 0444);
  52
  53 /* enable/disable SEV-ES support */
  54 static bool sev_es_enabled = true;
  55 module_param_named(sev_es, sev_es_enabled, bool, 0444);
  56 #else
  57 #define sev_enabled false
  58 #define sev_es_enabled false
  59 #endif /* CONFIG_KVM_AMD_SEV */
  60
  61 static u8 sev_enc_bit;
  62 static DECLARE_RWSEM(sev_deactivate_lock);
  63 static DEFINE_MUTEX(sev_bitmap_lock);
  64 unsigned int max_sev_asid;
  65 static unsigned int min_sev_asid;
  66 static unsigned long sev_me_mask;
  67 static unsigned int nr_asids;
  68 static unsigned long *sev_asid_bitmap;
  69 static unsigned long *sev_reclaim_asid_bitmap;
  70
  71 struct enc_region {
  72         struct list_head list;
  73         unsigned long npages;
  74         struct page **pages;
  75         unsigned long uaddr;
  76         unsigned long size;
  77 };
  78
  79 /* Called with the sev_bitmap_lock held, or on shutdown  */
  80 static int sev_flush_asids(int min_asid, int max_asid)
  81 {
  82         int ret, asid, error = 0;
  83
  84         /* Check if there are any ASIDs to reclaim before performing a flush */
  85         asid = find_next_bit(sev_reclaim_asid_bitmap, nr_asids, min_asid);
  86         if (asid > max_asid)
  87                 return -EBUSY;
  88
  89         /*
  90          * DEACTIVATE will clear the WBINVD indicator causing DF_FLUSH to fail,
  91          * so it must be guarded.
  92          */
  93         down_write(&sev_deactivate_lock);
  94
  95         wbinvd_on_all_cpus();
  96         ret = sev_guest_df_flush(&error);
  97
  98         up_write(&sev_deactivate_lock);
  99
 100         if (ret)
 101                 pr_err("SEV: DF_FLUSH failed, ret=%d, error=%#x\n", ret, error);
 102
 103         return ret;
 104 }
 105
 106 static inline bool is_mirroring_enc_context(struct kvm *kvm)
 107 {
 108         return !!to_kvm_svm(kvm)->sev_info.enc_context_owner;
 109 }
 110
 111 /* Must be called with the sev_bitmap_lock held */
 112 static bool __sev_recycle_asids(int min_asid, int max_asid)
 113 {
 114         if (sev_flush_asids(min_asid, max_asid))
 115                 return false;
 116
 117         /* The flush process will flush all reclaimable SEV and SEV-ES ASIDs */
 118         bitmap_xor(sev_asid_bitmap, sev_asid_bitmap, sev_reclaim_asid_bitmap,
 119                    nr_asids);
 120         bitmap_zero(sev_reclaim_asid_bitmap, nr_asids);
 121
 122         return true;
 123 }
 124
 125 static int sev_asid_new(struct kvm_sev_info *sev)
 126 {
 127         int asid, min_asid, max_asid, ret;
 128         bool retry = true;
 129         enum misc_res_type type;
 130
 131         type = sev->es_active ? MISC_CG_RES_SEV_ES : MISC_CG_RES_SEV;
 132         WARN_ON(sev->misc_cg);
 133         sev->misc_cg = get_current_misc_cg();
 134         ret = misc_cg_try_charge(type, sev->misc_cg, 1);
 135         if (ret) {
 136                 put_misc_cg(sev->misc_cg);
 137                 sev->misc_cg = NULL;
 138                 return ret;
 139         }
 140
 141         mutex_lock(&sev_bitmap_lock);
 142
 143         /*
 144          * SEV-enabled guests must use asid from min_sev_asid to max_sev_asid.
 145          * SEV-ES-enabled guest can use from 1 to min_sev_asid - 1.
 146          */
 147         min_asid = sev->es_active ? 1 : min_sev_asid;
 148         max_asid = sev->es_active ? min_sev_asid - 1 : max_sev_asid;
 149 again:
 150         asid = find_next_zero_bit(sev_asid_bitmap, max_asid + 1, min_asid);
 151         if (asid > max_asid) {
 152                 if (retry && __sev_recycle_asids(min_asid, max_asid)) {
 153                         retry = false;
 154                         goto again;
 155                 }
 156                 mutex_unlock(&sev_bitmap_lock);
 157                 ret = -EBUSY;
 158                 goto e_uncharge;
 159         }
 160
 161         __set_bit(asid, sev_asid_bitmap);
 162
 163         mutex_unlock(&sev_bitmap_lock);
 164
 165         return asid;
 166 e_uncharge:
 167         misc_cg_uncharge(type, sev->misc_cg, 1);
 168         put_misc_cg(sev->misc_cg);
 169         sev->misc_cg = NULL;
 170         return ret;
 171 }
 172
 173 static int sev_get_asid(struct kvm *kvm)
 174 {
 175         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 176
 177         return sev->asid;
 178 }
 179
 180 static void sev_asid_free(struct kvm_sev_info *sev)
 181 {
 182         struct svm_cpu_data *sd;
 183         int cpu;
 184         enum misc_res_type type;
 185
 186         mutex_lock(&sev_bitmap_lock);
 187
 188         __set_bit(sev->asid, sev_reclaim_asid_bitmap);
 189
 190         for_each_possible_cpu(cpu) {
 191                 sd = per_cpu(svm_data, cpu);
 192                 sd->sev_vmcbs[sev->asid] = NULL;
 193         }
 194
 195         mutex_unlock(&sev_bitmap_lock);
 196
 197         type = sev->es_active ? MISC_CG_RES_SEV_ES : MISC_CG_RES_SEV;
 198         misc_cg_uncharge(type, sev->misc_cg, 1);
 199         put_misc_cg(sev->misc_cg);
 200         sev->misc_cg = NULL;
 201 }
 202
 203 static void sev_decommission(unsigned int handle)
 204 {
 205         struct sev_data_decommission decommission;
 206
 207         if (!handle)
 208                 return;
 209
 210         decommission.handle = handle;
 211         sev_guest_decommission(&decommission, NULL);
 212 }
 213
 214 static void sev_unbind_asid(struct kvm *kvm, unsigned int handle)
 215 {
 216         struct sev_data_deactivate deactivate;
 217
 218         if (!handle)
 219                 return;
 220
 221         deactivate.handle = handle;
 222
 223         /* Guard DEACTIVATE against WBINVD/DF_FLUSH used in ASID recycling */
 224         down_read(&sev_deactivate_lock);
 225         sev_guest_deactivate(&deactivate, NULL);
 226         up_read(&sev_deactivate_lock);
 227
 228         sev_decommission(handle);
 229 }
 230
 231 static int sev_guest_init(struct kvm *kvm, struct kvm_sev_cmd *argp)
 232 {
 233         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 234         bool es_active = argp->id == KVM_SEV_ES_INIT;
 235         int asid, ret;
 236
 237         if (kvm->created_vcpus)
 238                 return -EINVAL;
 239
 240         ret = -EBUSY;
 241         if (unlikely(sev->active))
 242                 return ret;
 243
 244         sev->es_active = es_active;
 245         asid = sev_asid_new(sev);
 246         if (asid < 0)
 247                 goto e_no_asid;
 248         sev->asid = asid;
 249
 250         ret = sev_platform_init(&argp->error);
 251         if (ret)
 252                 goto e_free;
 253
 254         sev->active = true;
 255         sev->asid = asid;
 256         INIT_LIST_HEAD(&sev->regions_list);
 257
 258         return 0;
 259
 260 e_free:
 261         sev_asid_free(sev);
 262         sev->asid = 0;
 263 e_no_asid:
 264         sev->es_active = false;
 265         return ret;
 266 }
 267
 268 static int sev_bind_asid(struct kvm *kvm, unsigned int handle, int *error)
 269 {
 270         struct sev_data_activate activate;
 271         int asid = sev_get_asid(kvm);
 272         int ret;
 273
 274         /* activate ASID on the given handle */
 275         activate.handle = handle;
 276         activate.asid   = asid;
 277         ret = sev_guest_activate(&activate, error);
 278
 279         return ret;
 280 }
 281
 282 static int __sev_issue_cmd(int fd, int id, void *data, int *error)
 283 {
 284         struct fd f;
 285         int ret;
 286
 287         f = fdget(fd);
 288         if (!f.file)
 289                 return -EBADF;
 290
 291         ret = sev_issue_cmd_external_user(f.file, id, data, error);
 292
 293         fdput(f);
 294         return ret;
 295 }
 296
 297 static int sev_issue_cmd(struct kvm *kvm, int id, void *data, int *error)
 298 {
 299         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 300
 301         return __sev_issue_cmd(sev->fd, id, data, error);
 302 }
 303
 304 static int sev_launch_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
 305 {
 306         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 307         struct sev_data_launch_start start;
 308         struct kvm_sev_launch_start params;
 309         void *dh_blob, *session_blob;
 310         int *error = &argp->error;
 311         int ret;
 312
 313         if (!sev_guest(kvm))
 314                 return -ENOTTY;
 315
 316         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data, sizeof(params)))
 317                 return -EFAULT;
 318
 319         memset(&start, 0, sizeof(start));
 320
 321         dh_blob = NULL;
 322         if (params.dh_uaddr) {
 323                 dh_blob = psp_copy_user_blob(params.dh_uaddr, params.dh_len);
 324                 if (IS_ERR(dh_blob))
 325                         return PTR_ERR(dh_blob);
 326
 327                 start.dh_cert_address = __sme_set(__pa(dh_blob));
 328                 start.dh_cert_len = params.dh_len;
 329         }
 330
 331         session_blob = NULL;
 332         if (params.session_uaddr) {
 333                 session_blob = psp_copy_user_blob(params.session_uaddr, params.session_len);
 334                 if (IS_ERR(session_blob)) {
 335                         ret = PTR_ERR(session_blob);
 336                         goto e_free_dh;
 337                 }
 338
 339                 start.session_address = __sme_set(__pa(session_blob));
 340                 start.session_len = params.session_len;
 341         }
 342
 343         start.handle = params.handle;
 344         start.policy = params.policy;
 345
 346         /* create memory encryption context */
 347         ret = __sev_issue_cmd(argp->sev_fd, SEV_CMD_LAUNCH_START, &start, error);
 348         if (ret)
 349                 goto e_free_session;
 350
 351         /* Bind ASID to this guest */
 352         ret = sev_bind_asid(kvm, start.handle, error);
 353         if (ret) {
 354                 sev_decommission(start.handle);
 355                 goto e_free_session;
 356         }
 357
 358         /* return handle to userspace */
 359         params.handle = start.handle;
 360         if (copy_to_user((void __user *)(uintptr_t)argp->data, &params, sizeof(params))) {
 361                 sev_unbind_asid(kvm, start.handle);
 362                 ret = -EFAULT;
 363                 goto e_free_session;
 364         }
 365
 366         sev->handle = start.handle;
 367         sev->fd = argp->sev_fd;
 368
 369 e_free_session:
 370         kfree(session_blob);
 371 e_free_dh:
 372         kfree(dh_blob);
 373         return ret;
 374 }
 375
 376 static struct page **sev_pin_memory(struct kvm *kvm, unsigned long uaddr,
 377                                     unsigned long ulen, unsigned long *n,
 378                                     int write)
 379 {
 380         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 381         unsigned long npages, size;
 382         int npinned;
 383         unsigned long locked, lock_limit;
 384         struct page **pages;
 385         unsigned long first, last;
 386         int ret;
 387
 388         lockdep_assert_held(&kvm->lock);
 389
 390         if (ulen == 0 || uaddr + ulen < uaddr)
 391                 return ERR_PTR(-EINVAL);
 392
 393         /* Calculate number of pages. */
 394         first = (uaddr & PAGE_MASK) >> PAGE_SHIFT;
 395         last = ((uaddr + ulen - 1) & PAGE_MASK) >> PAGE_SHIFT;
 396         npages = (last - first + 1);
 397
 398         locked = sev->pages_locked + npages;
 399         lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
 400         if (locked > lock_limit && !capable(CAP_IPC_LOCK)) {
 401                 pr_err("SEV: %lu locked pages exceed the lock limit of %lu.\n", locked, lock_limit);
 402                 return ERR_PTR(-ENOMEM);
 403         }
 404
 405         if (WARN_ON_ONCE(npages > INT_MAX))
 406                 return ERR_PTR(-EINVAL);
 407
 408         /* Avoid using vmalloc for smaller buffers. */
 409         size = npages * sizeof(struct page *);
 410         if (size > PAGE_SIZE)
 411                 pages = __vmalloc(size, GFP_KERNEL_ACCOUNT | __GFP_ZERO);
 412         else
 413                 pages = kmalloc(size, GFP_KERNEL_ACCOUNT);
 414
 415         if (!pages)
 416                 return ERR_PTR(-ENOMEM);
 417
 418         /* Pin the user virtual address. */
 419         npinned = pin_user_pages_fast(uaddr, npages, write ? FOLL_WRITE : 0, pages);
 420         if (npinned != npages) {
 421                 pr_err("SEV: Failure locking %lu pages.\n", npages);
 422                 ret = -ENOMEM;
 423                 goto err;
 424         }
 425
 426         *n = npages;
 427         sev->pages_locked = locked;
 428
 429         return pages;
 430
 431 err:
 432         if (npinned > 0)
 433                 unpin_user_pages(pages, npinned);
 434
 435         kvfree(pages);
 436         return ERR_PTR(ret);
 437 }
 438
 439 static void sev_unpin_memory(struct kvm *kvm, struct page **pages,
 440                              unsigned long npages)
 441 {
 442         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 443
 444         unpin_user_pages(pages, npages);
 445         kvfree(pages);
 446         sev->pages_locked -= npages;
 447 }
 448
 449 static void sev_clflush_pages(struct page *pages[], unsigned long npages)
 450 {
 451         uint8_t *page_virtual;
 452         unsigned long i;
 453
 454         if (this_cpu_has(X86_FEATURE_SME_COHERENT) || npages == 0 ||
 455             pages == NULL)
 456                 return;
 457
 458         for (i = 0; i < npages; i++) {
 459                 page_virtual = kmap_atomic(pages[i]);
 460                 clflush_cache_range(page_virtual, PAGE_SIZE);
 461                 kunmap_atomic(page_virtual);
 462         }
 463 }
 464
 465 static unsigned long get_num_contig_pages(unsigned long idx,
 466                                 struct page **inpages, unsigned long npages)
 467 {
 468         unsigned long paddr, next_paddr;
 469         unsigned long i = idx + 1, pages = 1;
 470
 471         /* find the number of contiguous pages starting from idx */
 472         paddr = __sme_page_pa(inpages[idx]);
 473         while (i < npages) {
 474                 next_paddr = __sme_page_pa(inpages[i++]);
 475                 if ((paddr + PAGE_SIZE) == next_paddr) {
 476                         pages++;
 477                         paddr = next_paddr;
 478                         continue;
 479                 }
 480                 break;
 481         }
 482
 483         return pages;
 484 }
 485
 486 static int sev_launch_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
 487 {
 488         unsigned long vaddr, vaddr_end, next_vaddr, npages, pages, size, i;
 489         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 490         struct kvm_sev_launch_update_data params;
 491         struct sev_data_launch_update_data data;
 492         struct page **inpages;
 493         int ret;
 494
 495         if (!sev_guest(kvm))
 496                 return -ENOTTY;
 497
 498         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data, sizeof(params)))
 499                 return -EFAULT;
 500
 501         vaddr = params.uaddr;
 502         size = params.len;
 503         vaddr_end = vaddr + size;
 504
 505         /* Lock the user memory. */
 506         inpages = sev_pin_memory(kvm, vaddr, size, &npages, 1);
 507         if (IS_ERR(inpages))
 508                 return PTR_ERR(inpages);
 509
 510         /*
 511          * Flush (on non-coherent CPUs) before LAUNCH_UPDATE encrypts pages in
 512          * place; the cache may contain the data that was written unencrypted.
 513          */
 514         sev_clflush_pages(inpages, npages);
 515
 516         data.reserved = 0;
 517         data.handle = sev->handle;
 518
 519         for (i = 0; vaddr < vaddr_end; vaddr = next_vaddr, i += pages) {
 520                 int offset, len;
 521
 522                 /*
 523                  * If the user buffer is not page-aligned, calculate the offset
 524                  * within the page.
 525                  */
 526                 offset = vaddr & (PAGE_SIZE - 1);
 527
 528                 /* Calculate the number of pages that can be encrypted in one go. */
 529                 pages = get_num_contig_pages(i, inpages, npages);
 530
 531                 len = min_t(size_t, ((pages * PAGE_SIZE) - offset), size);
 532
 533                 data.len = len;
 534                 data.address = __sme_page_pa(inpages[i]) + offset;
 535                 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_DATA, &data, &argp->error);
 536                 if (ret)
 537                         goto e_unpin;
 538
 539                 size -= len;
 540                 next_vaddr = vaddr + len;
 541         }
 542
 543 e_unpin:
 544         /* content of memory is updated, mark pages dirty */
 545         for (i = 0; i < npages; i++) {
 546                 set_page_dirty_lock(inpages[i]);
 547                 mark_page_accessed(inpages[i]);
 548         }
 549         /* unlock the user pages */
 550         sev_unpin_memory(kvm, inpages, npages);
 551         return ret;
 552 }
 553
 554 static int sev_es_sync_vmsa(struct vcpu_svm *svm)
 555 {
 556         struct vmcb_save_area *save = &svm->vmcb->save;
 557
 558         /* Check some debug related fields before encrypting the VMSA */
 559         if (svm->vcpu.guest_debug || (save->dr7 & ~DR7_FIXED_1))
 560                 return -EINVAL;
 561
 562         /* Sync registgers */
 563         save->rax = svm->vcpu.arch.regs[VCPU_REGS_RAX];
 564         save->rbx = svm->vcpu.arch.regs[VCPU_REGS_RBX];
 565         save->rcx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
 566         save->rdx = svm->vcpu.arch.regs[VCPU_REGS_RDX];
 567         save->rsp = svm->vcpu.arch.regs[VCPU_REGS_RSP];
 568         save->rbp = svm->vcpu.arch.regs[VCPU_REGS_RBP];
 569         save->rsi = svm->vcpu.arch.regs[VCPU_REGS_RSI];
 570         save->rdi = svm->vcpu.arch.regs[VCPU_REGS_RDI];
 571 #ifdef CONFIG_X86_64
 572         save->r8  = svm->vcpu.arch.regs[VCPU_REGS_R8];
 573         save->r9  = svm->vcpu.arch.regs[VCPU_REGS_R9];
 574         save->r10 = svm->vcpu.arch.regs[VCPU_REGS_R10];
 575         save->r11 = svm->vcpu.arch.regs[VCPU_REGS_R11];
 576         save->r12 = svm->vcpu.arch.regs[VCPU_REGS_R12];
 577         save->r13 = svm->vcpu.arch.regs[VCPU_REGS_R13];
 578         save->r14 = svm->vcpu.arch.regs[VCPU_REGS_R14];
 579         save->r15 = svm->vcpu.arch.regs[VCPU_REGS_R15];
 580 #endif
 581         save->rip = svm->vcpu.arch.regs[VCPU_REGS_RIP];
 582
 583         /* Sync some non-GPR registers before encrypting */
 584         save->xcr0 = svm->vcpu.arch.xcr0;
 585         save->pkru = svm->vcpu.arch.pkru;
 586         save->xss  = svm->vcpu.arch.ia32_xss;
 587
 588         /*
 589          * SEV-ES will use a VMSA that is pointed to by the VMCB, not
 590          * the traditional VMSA that is part of the VMCB. Copy the
 591          * traditional VMSA as it has been built so far (in prep
 592          * for LAUNCH_UPDATE_VMSA) to be the initial SEV-ES state.
 593          */
 594         memcpy(svm->vmsa, save, sizeof(*save));
 595
 596         return 0;
 597 }
 598
 599 static int sev_launch_update_vmsa(struct kvm *kvm, struct kvm_sev_cmd *argp)
 600 {
 601         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 602         struct sev_data_launch_update_vmsa vmsa;
 603         struct kvm_vcpu *vcpu;
 604         int i, ret;
 605
 606         if (!sev_es_guest(kvm))
 607                 return -ENOTTY;
 608
 609         vmsa.reserved = 0;
 610
 611         kvm_for_each_vcpu(i, vcpu, kvm) {
 612                 struct vcpu_svm *svm = to_svm(vcpu);
 613
 614                 /* Perform some pre-encryption checks against the VMSA */
 615                 ret = sev_es_sync_vmsa(svm);
 616                 if (ret)
 617                         return ret;
 618
 619                 /*
 620                  * The LAUNCH_UPDATE_VMSA command will perform in-place
 621                  * encryption of the VMSA memory content (i.e it will write
 622                  * the same memory region with the guest's key), so invalidate
 623                  * it first.
 624                  */
 625                 clflush_cache_range(svm->vmsa, PAGE_SIZE);
 626
 627                 vmsa.handle = sev->handle;
 628                 vmsa.address = __sme_pa(svm->vmsa);
 629                 vmsa.len = PAGE_SIZE;
 630                 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_VMSA, &vmsa,
 631                                     &argp->error);
 632                 if (ret)
 633                         return ret;
 634
 635                 svm->vcpu.arch.guest_state_protected = true;
 636         }
 637
 638         return 0;
 639 }
 640
 641 static int sev_launch_measure(struct kvm *kvm, struct kvm_sev_cmd *argp)
 642 {
 643         void __user *measure = (void __user *)(uintptr_t)argp->data;
 644         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 645         struct sev_data_launch_measure data;
 646         struct kvm_sev_launch_measure params;
 647         void __user *p = NULL;
 648         void *blob = NULL;
 649         int ret;
 650
 651         if (!sev_guest(kvm))
 652                 return -ENOTTY;
 653
 654         if (copy_from_user(&params, measure, sizeof(params)))
 655                 return -EFAULT;
 656
 657         memset(&data, 0, sizeof(data));
 658
 659         /* User wants to query the blob length */
 660         if (!params.len)
 661                 goto cmd;
 662
 663         p = (void __user *)(uintptr_t)params.uaddr;
 664         if (p) {
 665                 if (params.len > SEV_FW_BLOB_MAX_SIZE)
 666                         return -EINVAL;
 667
 668                 blob = kmalloc(params.len, GFP_KERNEL_ACCOUNT);
 669                 if (!blob)
 670                         return -ENOMEM;
 671
 672                 data.address = __psp_pa(blob);
 673                 data.len = params.len;
 674         }
 675
 676 cmd:
 677         data.handle = sev->handle;
 678         ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_MEASURE, &data, &argp->error);
 679
 680         /*
 681          * If we query the session length, FW responded with expected data.
 682          */
 683         if (!params.len)
 684                 goto done;
 685
 686         if (ret)
 687                 goto e_free_blob;
 688
 689         if (blob) {
 690                 if (copy_to_user(p, blob, params.len))
 691                         ret = -EFAULT;
 692         }
 693
 694 done:
 695         params.len = data.len;
 696         if (copy_to_user(measure, &params, sizeof(params)))
 697                 ret = -EFAULT;
 698 e_free_blob:
 699         kfree(blob);
 700         return ret;
 701 }
 702
 703 static int sev_launch_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
 704 {
 705         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 706         struct sev_data_launch_finish data;
 707
 708         if (!sev_guest(kvm))
 709                 return -ENOTTY;
 710
 711         data.handle = sev->handle;
 712         return sev_issue_cmd(kvm, SEV_CMD_LAUNCH_FINISH, &data, &argp->error);
 713 }
 714
 715 static int sev_guest_status(struct kvm *kvm, struct kvm_sev_cmd *argp)
 716 {
 717         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 718         struct kvm_sev_guest_status params;
 719         struct sev_data_guest_status data;
 720         int ret;
 721
 722         if (!sev_guest(kvm))
 723                 return -ENOTTY;
 724
 725         memset(&data, 0, sizeof(data));
 726
 727         data.handle = sev->handle;
 728         ret = sev_issue_cmd(kvm, SEV_CMD_GUEST_STATUS, &data, &argp->error);
 729         if (ret)
 730                 return ret;
 731
 732         params.policy = data.policy;
 733         params.state = data.state;
 734         params.handle = data.handle;
 735
 736         if (copy_to_user((void __user *)(uintptr_t)argp->data, &params, sizeof(params)))
 737                 ret = -EFAULT;
 738
 739         return ret;
 740 }
 741
 742 static int __sev_issue_dbg_cmd(struct kvm *kvm, unsigned long src,
 743                                unsigned long dst, int size,
 744                                int *error, bool enc)
 745 {
 746         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 747         struct sev_data_dbg data;
 748
 749         data.reserved = 0;
 750         data.handle = sev->handle;
 751         data.dst_addr = dst;
 752         data.src_addr = src;
 753         data.len = size;
 754
 755         return sev_issue_cmd(kvm,
 756                              enc ? SEV_CMD_DBG_ENCRYPT : SEV_CMD_DBG_DECRYPT,
 757                              &data, error);
 758 }
 759
 760 static int __sev_dbg_decrypt(struct kvm *kvm, unsigned long src_paddr,
 761                              unsigned long dst_paddr, int sz, int *err)
 762 {
 763         int offset;
 764
 765         /*
 766          * Its safe to read more than we are asked, caller should ensure that
 767          * destination has enough space.
 768          */
 769         offset = src_paddr & 15;
 770         src_paddr = round_down(src_paddr, 16);
 771         sz = round_up(sz + offset, 16);
 772
 773         return __sev_issue_dbg_cmd(kvm, src_paddr, dst_paddr, sz, err, false);
 774 }
 775
 776 static int __sev_dbg_decrypt_user(struct kvm *kvm, unsigned long paddr,
 777                                   void __user *dst_uaddr,
 778                                   unsigned long dst_paddr,
 779                                   int size, int *err)
 780 {
 781         struct page *tpage = NULL;
 782         int ret, offset;
 783
 784         /* if inputs are not 16-byte then use intermediate buffer */
 785         if (!IS_ALIGNED(dst_paddr, 16) ||
 786             !IS_ALIGNED(paddr,     16) ||
 787             !IS_ALIGNED(size,      16)) {
 788                 tpage = (void *)alloc_page(GFP_KERNEL);
 789                 if (!tpage)
 790                         return -ENOMEM;
 791
 792                 dst_paddr = __sme_page_pa(tpage);
 793         }
 794
 795         ret = __sev_dbg_decrypt(kvm, paddr, dst_paddr, size, err);
 796         if (ret)
 797                 goto e_free;
 798
 799         if (tpage) {
 800                 offset = paddr & 15;
 801                 if (copy_to_user(dst_uaddr, page_address(tpage) + offset, size))
 802                         ret = -EFAULT;
 803         }
 804
 805 e_free:
 806         if (tpage)
 807                 __free_page(tpage);
 808
 809         return ret;
 810 }
 811
 812 static int __sev_dbg_encrypt_user(struct kvm *kvm, unsigned long paddr,
 813                                   void __user *vaddr,
 814                                   unsigned long dst_paddr,
 815                                   void __user *dst_vaddr,
 816                                   int size, int *error)
 817 {
 818         struct page *src_tpage = NULL;
 819         struct page *dst_tpage = NULL;
 820         int ret, len = size;
 821
 822         /* If source buffer is not aligned then use an intermediate buffer */
 823         if (!IS_ALIGNED((unsigned long)vaddr, 16)) {
 824                 src_tpage = alloc_page(GFP_KERNEL);
 825                 if (!src_tpage)
 826                         return -ENOMEM;
 827
 828                 if (copy_from_user(page_address(src_tpage), vaddr, size)) {
 829                         __free_page(src_tpage);
 830                         return -EFAULT;
 831                 }
 832
 833                 paddr = __sme_page_pa(src_tpage);
 834         }
 835
 836         /*
 837          *  If destination buffer or length is not aligned then do read-modify-write:
 838          *   - decrypt destination in an intermediate buffer
 839          *   - copy the source buffer in an intermediate buffer
 840          *   - use the intermediate buffer as source buffer
 841          */
 842         if (!IS_ALIGNED((unsigned long)dst_vaddr, 16) || !IS_ALIGNED(size, 16)) {
 843                 int dst_offset;
 844
 845                 dst_tpage = alloc_page(GFP_KERNEL);
 846                 if (!dst_tpage) {
 847                         ret = -ENOMEM;
 848                         goto e_free;
 849                 }
 850
 851                 ret = __sev_dbg_decrypt(kvm, dst_paddr,
 852                                         __sme_page_pa(dst_tpage), size, error);
 853                 if (ret)
 854                         goto e_free;
 855
 856                 /*
 857                  *  If source is kernel buffer then use memcpy() otherwise
 858                  *  copy_from_user().
 859                  */
 860                 dst_offset = dst_paddr & 15;
 861
 862                 if (src_tpage)
 863                         memcpy(page_address(dst_tpage) + dst_offset,
 864                                page_address(src_tpage), size);
 865                 else {
 866                         if (copy_from_user(page_address(dst_tpage) + dst_offset,
 867                                            vaddr, size)) {
 868                                 ret = -EFAULT;
 869                                 goto e_free;
 870                         }
 871                 }
 872
 873                 paddr = __sme_page_pa(dst_tpage);
 874                 dst_paddr = round_down(dst_paddr, 16);
 875                 len = round_up(size, 16);
 876         }
 877
 878         ret = __sev_issue_dbg_cmd(kvm, paddr, dst_paddr, len, error, true);
 879
 880 e_free:
 881         if (src_tpage)
 882                 __free_page(src_tpage);
 883         if (dst_tpage)
 884                 __free_page(dst_tpage);
 885         return ret;
 886 }
 887
 888 static int sev_dbg_crypt(struct kvm *kvm, struct kvm_sev_cmd *argp, bool dec)
 889 {
 890         unsigned long vaddr, vaddr_end, next_vaddr;
 891         unsigned long dst_vaddr;
 892         struct page **src_p, **dst_p;
 893         struct kvm_sev_dbg debug;
 894         unsigned long n;
 895         unsigned int size;
 896         int ret;
 897
 898         if (!sev_guest(kvm))
 899                 return -ENOTTY;
 900
 901         if (copy_from_user(&debug, (void __user *)(uintptr_t)argp->data, sizeof(debug)))
 902                 return -EFAULT;
 903
 904         if (!debug.len || debug.src_uaddr + debug.len < debug.src_uaddr)
 905                 return -EINVAL;
 906         if (!debug.dst_uaddr)
 907                 return -EINVAL;
 908
 909         vaddr = debug.src_uaddr;
 910         size = debug.len;
 911         vaddr_end = vaddr + size;
 912         dst_vaddr = debug.dst_uaddr;
 913
 914         for (; vaddr < vaddr_end; vaddr = next_vaddr) {
 915                 int len, s_off, d_off;
 916
 917                 /* lock userspace source and destination page */
 918                 src_p = sev_pin_memory(kvm, vaddr & PAGE_MASK, PAGE_SIZE, &n, 0);
 919                 if (IS_ERR(src_p))
 920                         return PTR_ERR(src_p);
 921
 922                 dst_p = sev_pin_memory(kvm, dst_vaddr & PAGE_MASK, PAGE_SIZE, &n, 1);
 923                 if (IS_ERR(dst_p)) {
 924                         sev_unpin_memory(kvm, src_p, n);
 925                         return PTR_ERR(dst_p);
 926                 }
 927
 928                 /*
 929                  * Flush (on non-coherent CPUs) before DBG_{DE,EN}CRYPT read or modify
 930                  * the pages; flush the destination too so that future accesses do not
 931                  * see stale data.
 932                  */
 933                 sev_clflush_pages(src_p, 1);
 934                 sev_clflush_pages(dst_p, 1);
 935
 936                 /*
 937                  * Since user buffer may not be page aligned, calculate the
 938                  * offset within the page.
 939                  */
 940                 s_off = vaddr & ~PAGE_MASK;
 941                 d_off = dst_vaddr & ~PAGE_MASK;
 942                 len = min_t(size_t, (PAGE_SIZE - s_off), size);
 943
 944                 if (dec)
 945                         ret = __sev_dbg_decrypt_user(kvm,
 946                                                      __sme_page_pa(src_p[0]) + s_off,
 947                                                      (void __user *)dst_vaddr,
 948                                                      __sme_page_pa(dst_p[0]) + d_off,
 949                                                      len, &argp->error);
 950                 else
 951                         ret = __sev_dbg_encrypt_user(kvm,
 952                                                      __sme_page_pa(src_p[0]) + s_off,
 953                                                      (void __user *)vaddr,
 954                                                      __sme_page_pa(dst_p[0]) + d_off,
 955                                                      (void __user *)dst_vaddr,
 956                                                      len, &argp->error);
 957
 958                 sev_unpin_memory(kvm, src_p, n);
 959                 sev_unpin_memory(kvm, dst_p, n);
 960
 961                 if (ret)
 962                         goto err;
 963
 964                 next_vaddr = vaddr + len;
 965                 dst_vaddr = dst_vaddr + len;
 966                 size -= len;
 967         }
 968 err:
 969         return ret;
 970 }
 971
 972 static int sev_launch_secret(struct kvm *kvm, struct kvm_sev_cmd *argp)
 973 {
 974         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 975         struct sev_data_launch_secret data;
 976         struct kvm_sev_launch_secret params;
 977         struct page **pages;
 978         void *blob, *hdr;
 979         unsigned long n, i;
 980         int ret, offset;
 981
 982         if (!sev_guest(kvm))
 983                 return -ENOTTY;
 984
 985         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data, sizeof(params)))
 986                 return -EFAULT;
 987
 988         pages = sev_pin_memory(kvm, params.guest_uaddr, params.guest_len, &n, 1);
 989         if (IS_ERR(pages))
 990                 return PTR_ERR(pages);
 991
 992         /*
 993          * Flush (on non-coherent CPUs) before LAUNCH_SECRET encrypts pages in
 994          * place; the cache may contain the data that was written unencrypted.
 995          */
 996         sev_clflush_pages(pages, n);
 997
 998         /*
 999          * The secret must be copied into contiguous memory region, lets verify
1000          * that userspace memory pages are contiguous before we issue command.
1001          */
1002         if (get_num_contig_pages(0, pages, n) != n) {
1003                 ret = -EINVAL;
1004                 goto e_unpin_memory;
1005         }
1006
1007         memset(&data, 0, sizeof(data));
1008
1009         offset = params.guest_uaddr & (PAGE_SIZE - 1);
1010         data.guest_address = __sme_page_pa(pages[0]) + offset;
1011         data.guest_len = params.guest_len;
1012
1013         blob = psp_copy_user_blob(params.trans_uaddr, params.trans_len);
1014         if (IS_ERR(blob)) {
1015                 ret = PTR_ERR(blob);
1016                 goto e_unpin_memory;
1017         }
1018
1019         data.trans_address = __psp_pa(blob);
1020         data.trans_len = params.trans_len;
1021
1022         hdr = psp_copy_user_blob(params.hdr_uaddr, params.hdr_len);
1023         if (IS_ERR(hdr)) {
1024                 ret = PTR_ERR(hdr);
1025                 goto e_free_blob;
1026         }
1027         data.hdr_address = __psp_pa(hdr);
1028         data.hdr_len = params.hdr_len;
1029
1030         data.handle = sev->handle;
1031         ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_SECRET, &data, &argp->error);
1032
1033         kfree(hdr);
1034
1035 e_free_blob:
1036         kfree(blob);
1037 e_unpin_memory:
1038         /* content of memory is updated, mark pages dirty */
1039         for (i = 0; i < n; i++) {
1040                 set_page_dirty_lock(pages[i]);
1041                 mark_page_accessed(pages[i]);
1042         }
1043         sev_unpin_memory(kvm, pages, n);
1044         return ret;
1045 }
1046
1047 static int sev_get_attestation_report(struct kvm *kvm, struct kvm_sev_cmd *argp)
1048 {
1049         void __user *report = (void __user *)(uintptr_t)argp->data;
1050         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1051         struct sev_data_attestation_report data;
1052         struct kvm_sev_attestation_report params;
1053         void __user *p;
1054         void *blob = NULL;
1055         int ret;
1056
1057         if (!sev_guest(kvm))
1058                 return -ENOTTY;
1059
1060         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data, sizeof(params)))
1061                 return -EFAULT;
1062
1063         memset(&data, 0, sizeof(data));
1064
1065         /* User wants to query the blob length */
1066         if (!params.len)
1067                 goto cmd;
1068
1069         p = (void __user *)(uintptr_t)params.uaddr;
1070         if (p) {
1071                 if (params.len > SEV_FW_BLOB_MAX_SIZE)
1072                         return -EINVAL;
1073
1074                 blob = kmalloc(params.len, GFP_KERNEL_ACCOUNT);
1075                 if (!blob)
1076                         return -ENOMEM;
1077
1078                 data.address = __psp_pa(blob);
1079                 data.len = params.len;
1080                 memcpy(data.mnonce, params.mnonce, sizeof(params.mnonce));
1081         }
1082 cmd:
1083         data.handle = sev->handle;
1084         ret = sev_issue_cmd(kvm, SEV_CMD_ATTESTATION_REPORT, &data, &argp->error);
1085         /*
1086          * If we query the session length, FW responded with expected data.
1087          */
1088         if (!params.len)
1089                 goto done;
1090
1091         if (ret)
1092                 goto e_free_blob;
1093
1094         if (blob) {
1095                 if (copy_to_user(p, blob, params.len))
1096                         ret = -EFAULT;
1097         }
1098
1099 done:
1100         params.len = data.len;
1101         if (copy_to_user(report, &params, sizeof(params)))
1102                 ret = -EFAULT;
1103 e_free_blob:
1104         kfree(blob);
1105         return ret;
1106 }
1107
1108 /* Userspace wants to query session length. */
1109 static int
1110 __sev_send_start_query_session_length(struct kvm *kvm, struct kvm_sev_cmd *argp,
1111                                       struct kvm_sev_send_start *params)
1112 {
1113         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1114         struct sev_data_send_start data;
1115         int ret;
1116
1117         memset(&data, 0, sizeof(data));
1118         data.handle = sev->handle;
1119         ret = sev_issue_cmd(kvm, SEV_CMD_SEND_START, &data, &argp->error);
1120
1121         params->session_len = data.session_len;
1122         if (copy_to_user((void __user *)(uintptr_t)argp->data, params,
1123                                 sizeof(struct kvm_sev_send_start)))
1124                 ret = -EFAULT;
1125
1126         return ret;
1127 }
1128
1129 static int sev_send_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
1130 {
1131         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1132         struct sev_data_send_start data;
1133         struct kvm_sev_send_start params;
1134         void *amd_certs, *session_data;
1135         void *pdh_cert, *plat_certs;
1136         int ret;
1137
1138         if (!sev_guest(kvm))
1139                 return -ENOTTY;
1140
1141         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data,
1142                                 sizeof(struct kvm_sev_send_start)))
1143                 return -EFAULT;
1144
1145         /* if session_len is zero, userspace wants to query the session length */
1146         if (!params.session_len)
1147                 return __sev_send_start_query_session_length(kvm, argp,
1148                                 &params);
1149
1150         /* some sanity checks */
1151         if (!params.pdh_cert_uaddr || !params.pdh_cert_len ||
1152             !params.session_uaddr || params.session_len > SEV_FW_BLOB_MAX_SIZE)
1153                 return -EINVAL;
1154
1155         /* allocate the memory to hold the session data blob */
1156         session_data = kmalloc(params.session_len, GFP_KERNEL_ACCOUNT);
1157         if (!session_data)
1158                 return -ENOMEM;
1159
1160         /* copy the certificate blobs from userspace */
1161         pdh_cert = psp_copy_user_blob(params.pdh_cert_uaddr,
1162                                 params.pdh_cert_len);
1163         if (IS_ERR(pdh_cert)) {
1164                 ret = PTR_ERR(pdh_cert);
1165                 goto e_free_session;
1166         }
1167
1168         plat_certs = psp_copy_user_blob(params.plat_certs_uaddr,
1169                                 params.plat_certs_len);
1170         if (IS_ERR(plat_certs)) {
1171                 ret = PTR_ERR(plat_certs);
1172                 goto e_free_pdh;
1173         }
1174
1175         amd_certs = psp_copy_user_blob(params.amd_certs_uaddr,
1176                                 params.amd_certs_len);
1177         if (IS_ERR(amd_certs)) {
1178                 ret = PTR_ERR(amd_certs);
1179                 goto e_free_plat_cert;
1180         }
1181
1182         /* populate the FW SEND_START field with system physical address */
1183         memset(&data, 0, sizeof(data));
1184         data.pdh_cert_address = __psp_pa(pdh_cert);
1185         data.pdh_cert_len = params.pdh_cert_len;
1186         data.plat_certs_address = __psp_pa(plat_certs);
1187         data.plat_certs_len = params.plat_certs_len;
1188         data.amd_certs_address = __psp_pa(amd_certs);
1189         data.amd_certs_len = params.amd_certs_len;
1190         data.session_address = __psp_pa(session_data);
1191         data.session_len = params.session_len;
1192         data.handle = sev->handle;
1193
1194         ret = sev_issue_cmd(kvm, SEV_CMD_SEND_START, &data, &argp->error);
1195
1196         if (!ret && copy_to_user((void __user *)(uintptr_t)params.session_uaddr,
1197                         session_data, params.session_len)) {
1198                 ret = -EFAULT;
1199                 goto e_free_amd_cert;
1200         }
1201
1202         params.policy = data.policy;
1203         params.session_len = data.session_len;
1204         if (copy_to_user((void __user *)(uintptr_t)argp->data, &params,
1205                                 sizeof(struct kvm_sev_send_start)))
1206                 ret = -EFAULT;
1207
1208 e_free_amd_cert:
1209         kfree(amd_certs);
1210 e_free_plat_cert:
1211         kfree(plat_certs);
1212 e_free_pdh:
1213         kfree(pdh_cert);
1214 e_free_session:
1215         kfree(session_data);
1216         return ret;
1217 }
1218
1219 /* Userspace wants to query either header or trans length. */
1220 static int
1221 __sev_send_update_data_query_lengths(struct kvm *kvm, struct kvm_sev_cmd *argp,
1222                                      struct kvm_sev_send_update_data *params)
1223 {
1224         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1225         struct sev_data_send_update_data data;
1226         int ret;
1227
1228         memset(&data, 0, sizeof(data));
1229         data.handle = sev->handle;
1230         ret = sev_issue_cmd(kvm, SEV_CMD_SEND_UPDATE_DATA, &data, &argp->error);
1231
1232         params->hdr_len = data.hdr_len;
1233         params->trans_len = data.trans_len;
1234
1235         if (copy_to_user((void __user *)(uintptr_t)argp->data, params,
1236                          sizeof(struct kvm_sev_send_update_data)))
1237                 ret = -EFAULT;
1238
1239         return ret;
1240 }
1241
1242 static int sev_send_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
1243 {
1244         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1245         struct sev_data_send_update_data data;
1246         struct kvm_sev_send_update_data params;
1247         void *hdr, *trans_data;
1248         struct page **guest_page;
1249         unsigned long n;
1250         int ret, offset;
1251
1252         if (!sev_guest(kvm))
1253                 return -ENOTTY;
1254
1255         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data,
1256                         sizeof(struct kvm_sev_send_update_data)))
1257                 return -EFAULT;
1258
1259         /* userspace wants to query either header or trans length */
1260         if (!params.trans_len || !params.hdr_len)
1261                 return __sev_send_update_data_query_lengths(kvm, argp, &params);
1262
1263         if (!params.trans_uaddr || !params.guest_uaddr ||
1264             !params.guest_len || !params.hdr_uaddr)
1265                 return -EINVAL;
1266
1267         /* Check if we are crossing the page boundary */
1268         offset = params.guest_uaddr & (PAGE_SIZE - 1);
1269         if ((params.guest_len + offset > PAGE_SIZE))
1270                 return -EINVAL;
1271
1272         /* Pin guest memory */
1273         guest_page = sev_pin_memory(kvm, params.guest_uaddr & PAGE_MASK,
1274                                     PAGE_SIZE, &n, 0);
1275         if (IS_ERR(guest_page))
1276                 return PTR_ERR(guest_page);
1277
1278         /* allocate memory for header and transport buffer */
1279         ret = -ENOMEM;
1280         hdr = kmalloc(params.hdr_len, GFP_KERNEL_ACCOUNT);
1281         if (!hdr)
1282                 goto e_unpin;
1283
1284         trans_data = kmalloc(params.trans_len, GFP_KERNEL_ACCOUNT);
1285         if (!trans_data)
1286                 goto e_free_hdr;
1287
1288         memset(&data, 0, sizeof(data));
1289         data.hdr_address = __psp_pa(hdr);
1290         data.hdr_len = params.hdr_len;
1291         data.trans_address = __psp_pa(trans_data);
1292         data.trans_len = params.trans_len;
1293
1294         /* The SEND_UPDATE_DATA command requires C-bit to be always set. */
1295         data.guest_address = (page_to_pfn(guest_page[0]) << PAGE_SHIFT) + offset;
1296         data.guest_address |= sev_me_mask;
1297         data.guest_len = params.guest_len;
1298         data.handle = sev->handle;
1299
1300         ret = sev_issue_cmd(kvm, SEV_CMD_SEND_UPDATE_DATA, &data, &argp->error);
1301
1302         if (ret)
1303                 goto e_free_trans_data;
1304
1305         /* copy transport buffer to user space */
1306         if (copy_to_user((void __user *)(uintptr_t)params.trans_uaddr,
1307                          trans_data, params.trans_len)) {
1308                 ret = -EFAULT;
1309                 goto e_free_trans_data;
1310         }
1311
1312         /* Copy packet header to userspace. */
1313         if (copy_to_user((void __user *)(uintptr_t)params.hdr_uaddr, hdr,
1314                          params.hdr_len))
1315                 ret = -EFAULT;
1316
1317 e_free_trans_data:
1318         kfree(trans_data);
1319 e_free_hdr:
1320         kfree(hdr);
1321 e_unpin:
1322         sev_unpin_memory(kvm, guest_page, n);
1323
1324         return ret;
1325 }
1326
1327 static int sev_send_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
1328 {
1329         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1330         struct sev_data_send_finish data;
1331
1332         if (!sev_guest(kvm))
1333                 return -ENOTTY;
1334
1335         data.handle = sev->handle;
1336         return sev_issue_cmd(kvm, SEV_CMD_SEND_FINISH, &data, &argp->error);
1337 }
1338
1339 static int sev_send_cancel(struct kvm *kvm, struct kvm_sev_cmd *argp)
1340 {
1341         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1342         struct sev_data_send_cancel data;
1343
1344         if (!sev_guest(kvm))
1345                 return -ENOTTY;
1346
1347         data.handle = sev->handle;
1348         return sev_issue_cmd(kvm, SEV_CMD_SEND_CANCEL, &data, &argp->error);
1349 }
1350
1351 static int sev_receive_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
1352 {
1353         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1354         struct sev_data_receive_start start;
1355         struct kvm_sev_receive_start params;
1356         int *error = &argp->error;
1357         void *session_data;
1358         void *pdh_data;
1359         int ret;
1360
1361         if (!sev_guest(kvm))
1362                 return -ENOTTY;
1363
1364         /* Get parameter from the userspace */
1365         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data,
1366                         sizeof(struct kvm_sev_receive_start)))
1367                 return -EFAULT;
1368
1369         /* some sanity checks */
1370         if (!params.pdh_uaddr || !params.pdh_len ||
1371             !params.session_uaddr || !params.session_len)
1372                 return -EINVAL;
1373
1374         pdh_data = psp_copy_user_blob(params.pdh_uaddr, params.pdh_len);
1375         if (IS_ERR(pdh_data))
1376                 return PTR_ERR(pdh_data);
1377
1378         session_data = psp_copy_user_blob(params.session_uaddr,
1379                         params.session_len);
1380         if (IS_ERR(session_data)) {
1381                 ret = PTR_ERR(session_data);
1382                 goto e_free_pdh;
1383         }
1384
1385         memset(&start, 0, sizeof(start));
1386         start.handle = params.handle;
1387         start.policy = params.policy;
1388         start.pdh_cert_address = __psp_pa(pdh_data);
1389         start.pdh_cert_len = params.pdh_len;
1390         start.session_address = __psp_pa(session_data);
1391         start.session_len = params.session_len;
1392
1393         /* create memory encryption context */
1394         ret = __sev_issue_cmd(argp->sev_fd, SEV_CMD_RECEIVE_START, &start,
1395                                 error);
1396         if (ret)
1397                 goto e_free_session;
1398
1399         /* Bind ASID to this guest */
1400         ret = sev_bind_asid(kvm, start.handle, error);
1401         if (ret)
1402                 goto e_free_session;
1403
1404         params.handle = start.handle;
1405         if (copy_to_user((void __user *)(uintptr_t)argp->data,
1406                          &params, sizeof(struct kvm_sev_receive_start))) {
1407                 ret = -EFAULT;
1408                 sev_unbind_asid(kvm, start.handle);
1409                 goto e_free_session;
1410         }
1411
1412         sev->handle = start.handle;
1413         sev->fd = argp->sev_fd;
1414
1415 e_free_session:
1416         kfree(session_data);
1417 e_free_pdh:
1418         kfree(pdh_data);
1419
1420         return ret;
1421 }
1422
1423 static int sev_receive_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
1424 {
1425         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1426         struct kvm_sev_receive_update_data params;
1427         struct sev_data_receive_update_data data;
1428         void *hdr = NULL, *trans = NULL;
1429         struct page **guest_page;
1430         unsigned long n;
1431         int ret, offset;
1432
1433         if (!sev_guest(kvm))
1434                 return -EINVAL;
1435
1436         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data,
1437                         sizeof(struct kvm_sev_receive_update_data)))
1438                 return -EFAULT;
1439
1440         if (!params.hdr_uaddr || !params.hdr_len ||
1441             !params.guest_uaddr || !params.guest_len ||
1442             !params.trans_uaddr || !params.trans_len)
1443                 return -EINVAL;
1444
1445         /* Check if we are crossing the page boundary */
1446         offset = params.guest_uaddr & (PAGE_SIZE - 1);
1447         if ((params.guest_len + offset > PAGE_SIZE))
1448                 return -EINVAL;
1449
1450         hdr = psp_copy_user_blob(params.hdr_uaddr, params.hdr_len);
1451         if (IS_ERR(hdr))
1452                 return PTR_ERR(hdr);
1453
1454         trans = psp_copy_user_blob(params.trans_uaddr, params.trans_len);
1455         if (IS_ERR(trans)) {
1456                 ret = PTR_ERR(trans);
1457                 goto e_free_hdr;
1458         }
1459
1460         memset(&data, 0, sizeof(data));
1461         data.hdr_address = __psp_pa(hdr);
1462         data.hdr_len = params.hdr_len;
1463         data.trans_address = __psp_pa(trans);
1464         data.trans_len = params.trans_len;
1465
1466         /* Pin guest memory */
1467         guest_page = sev_pin_memory(kvm, params.guest_uaddr & PAGE_MASK,
1468                                     PAGE_SIZE, &n, 0);
1469         if (IS_ERR(guest_page)) {
1470                 ret = PTR_ERR(guest_page);
1471                 goto e_free_trans;
1472         }
1473
1474         /* The RECEIVE_UPDATE_DATA command requires C-bit to be always set. */
1475         data.guest_address = (page_to_pfn(guest_page[0]) << PAGE_SHIFT) + offset;
1476         data.guest_address |= sev_me_mask;
1477         data.guest_len = params.guest_len;
1478         data.handle = sev->handle;
1479
1480         ret = sev_issue_cmd(kvm, SEV_CMD_RECEIVE_UPDATE_DATA, &data,
1481                                 &argp->error);
1482
1483         sev_unpin_memory(kvm, guest_page, n);
1484
1485 e_free_trans:
1486         kfree(trans);
1487 e_free_hdr:
1488         kfree(hdr);
1489
1490         return ret;
1491 }
1492
1493 static int sev_receive_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
1494 {
1495         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1496         struct sev_data_receive_finish data;
1497
1498         if (!sev_guest(kvm))
1499                 return -ENOTTY;
1500
1501         data.handle = sev->handle;
1502         return sev_issue_cmd(kvm, SEV_CMD_RECEIVE_FINISH, &data, &argp->error);
1503 }
1504
1505 int svm_mem_enc_op(struct kvm *kvm, void __user *argp)
1506 {
1507         struct kvm_sev_cmd sev_cmd;
1508         int r;
1509
1510         if (!sev_enabled)
1511                 return -ENOTTY;
1512
1513         if (!argp)
1514                 return 0;
1515
1516         if (copy_from_user(&sev_cmd, argp, sizeof(struct kvm_sev_cmd)))
1517                 return -EFAULT;
1518
1519         mutex_lock(&kvm->lock);
1520
1521         /* enc_context_owner handles all memory enc operations */
1522         if (is_mirroring_enc_context(kvm)) {
1523                 r = -EINVAL;
1524                 goto out;
1525         }
1526
1527         switch (sev_cmd.id) {
1528         case KVM_SEV_ES_INIT:
1529                 if (!sev_es_enabled) {
1530                         r = -ENOTTY;
1531                         goto out;
1532                 }
1533                 fallthrough;
1534         case KVM_SEV_INIT:
1535                 r = sev_guest_init(kvm, &sev_cmd);
1536                 break;
1537         case KVM_SEV_LAUNCH_START:
1538                 r = sev_launch_start(kvm, &sev_cmd);
1539                 break;
1540         case KVM_SEV_LAUNCH_UPDATE_DATA:
1541                 r = sev_launch_update_data(kvm, &sev_cmd);
1542                 break;
1543         case KVM_SEV_LAUNCH_UPDATE_VMSA:
1544                 r = sev_launch_update_vmsa(kvm, &sev_cmd);
1545                 break;
1546         case KVM_SEV_LAUNCH_MEASURE:
1547                 r = sev_launch_measure(kvm, &sev_cmd);
1548                 break;
1549         case KVM_SEV_LAUNCH_FINISH:
1550                 r = sev_launch_finish(kvm, &sev_cmd);
1551                 break;
1552         case KVM_SEV_GUEST_STATUS:
1553                 r = sev_guest_status(kvm, &sev_cmd);
1554                 break;
1555         case KVM_SEV_DBG_DECRYPT:
1556                 r = sev_dbg_crypt(kvm, &sev_cmd, true);
1557                 break;
1558         case KVM_SEV_DBG_ENCRYPT:
1559                 r = sev_dbg_crypt(kvm, &sev_cmd, false);
1560                 break;
1561         case KVM_SEV_LAUNCH_SECRET:
1562                 r = sev_launch_secret(kvm, &sev_cmd);
1563                 break;
1564         case KVM_SEV_GET_ATTESTATION_REPORT:
1565                 r = sev_get_attestation_report(kvm, &sev_cmd);
1566                 break;
1567         case KVM_SEV_SEND_START:
1568                 r = sev_send_start(kvm, &sev_cmd);
1569                 break;
1570         case KVM_SEV_SEND_UPDATE_DATA:
1571                 r = sev_send_update_data(kvm, &sev_cmd);
1572                 break;
1573         case KVM_SEV_SEND_FINISH:
1574                 r = sev_send_finish(kvm, &sev_cmd);
1575                 break;
1576         case KVM_SEV_SEND_CANCEL:
1577                 r = sev_send_cancel(kvm, &sev_cmd);
1578                 break;
1579         case KVM_SEV_RECEIVE_START:
1580                 r = sev_receive_start(kvm, &sev_cmd);
1581                 break;
1582         case KVM_SEV_RECEIVE_UPDATE_DATA:
1583                 r = sev_receive_update_data(kvm, &sev_cmd);
1584                 break;
1585         case KVM_SEV_RECEIVE_FINISH:
1586                 r = sev_receive_finish(kvm, &sev_cmd);
1587                 break;
1588         default:
1589                 r = -EINVAL;
1590                 goto out;
1591         }
1592
1593         if (copy_to_user(argp, &sev_cmd, sizeof(struct kvm_sev_cmd)))
1594                 r = -EFAULT;
1595
1596 out:
1597         mutex_unlock(&kvm->lock);
1598         return r;
1599 }
1600
1601 int svm_register_enc_region(struct kvm *kvm,
1602                             struct kvm_enc_region *range)
1603 {
1604         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1605         struct enc_region *region;
1606         int ret = 0;
1607
1608         if (!sev_guest(kvm))
1609                 return -ENOTTY;
1610
1611         /* If kvm is mirroring encryption context it isn't responsible for it */
1612         if (is_mirroring_enc_context(kvm))
1613                 return -EINVAL;
1614
1615         if (range->addr > ULONG_MAX || range->size > ULONG_MAX)
1616                 return -EINVAL;
1617
1618         region = kzalloc(sizeof(*region), GFP_KERNEL_ACCOUNT);
1619         if (!region)
1620                 return -ENOMEM;
1621
1622         mutex_lock(&kvm->lock);
1623         region->pages = sev_pin_memory(kvm, range->addr, range->size, &region->npages, 1);
1624         if (IS_ERR(region->pages)) {
1625                 ret = PTR_ERR(region->pages);
1626                 mutex_unlock(&kvm->lock);
1627                 goto e_free;
1628         }
1629
1630         region->uaddr = range->addr;
1631         region->size = range->size;
1632
1633         list_add_tail(&region->list, &sev->regions_list);
1634         mutex_unlock(&kvm->lock);
1635
1636         /*
1637          * The guest may change the memory encryption attribute from C=0 -> C=1
1638          * or vice versa for this memory range. Lets make sure caches are
1639          * flushed to ensure that guest data gets written into memory with
1640          * correct C-bit.
1641          */
1642         sev_clflush_pages(region->pages, region->npages);
1643
1644         return ret;
1645
1646 e_free:
1647         kfree(region);
1648         return ret;
1649 }
1650
1651 static struct enc_region *
1652 find_enc_region(struct kvm *kvm, struct kvm_enc_region *range)
1653 {
1654         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1655         struct list_head *head = &sev->regions_list;
1656         struct enc_region *i;
1657
1658         list_for_each_entry(i, head, list) {
1659                 if (i->uaddr == range->addr &&
1660                     i->size == range->size)
1661                         return i;
1662         }
1663
1664         return NULL;
1665 }
1666
1667 static void __unregister_enc_region_locked(struct kvm *kvm,
1668                                            struct enc_region *region)
1669 {
1670         sev_unpin_memory(kvm, region->pages, region->npages);
1671         list_del(&region->list);
1672         kfree(region);
1673 }
1674
1675 int svm_unregister_enc_region(struct kvm *kvm,
1676                               struct kvm_enc_region *range)
1677 {
1678         struct enc_region *region;
1679         int ret;
1680
1681         /* If kvm is mirroring encryption context it isn't responsible for it */
1682         if (is_mirroring_enc_context(kvm))
1683                 return -EINVAL;
1684
1685         mutex_lock(&kvm->lock);
1686
1687         if (!sev_guest(kvm)) {
1688                 ret = -ENOTTY;
1689                 goto failed;
1690         }
1691
1692         region = find_enc_region(kvm, range);
1693         if (!region) {
1694                 ret = -EINVAL;
1695                 goto failed;
1696         }
1697
1698         /*
1699          * Ensure that all guest tagged cache entries are flushed before
1700          * releasing the pages back to the system for use. CLFLUSH will
1701          * not do this, so issue a WBINVD.
1702          */
1703         wbinvd_on_all_cpus();
1704
1705         __unregister_enc_region_locked(kvm, region);
1706
1707         mutex_unlock(&kvm->lock);
1708         return 0;
1709
1710 failed:
1711         mutex_unlock(&kvm->lock);
1712         return ret;
1713 }
1714
1715 int svm_vm_copy_asid_from(struct kvm *kvm, unsigned int source_fd)
1716 {
1717         struct file *source_kvm_file;
1718         struct kvm *source_kvm;
1719         struct kvm_sev_info *mirror_sev;
1720         unsigned int asid;
1721         int ret;
1722
1723         source_kvm_file = fget(source_fd);
1724         if (!file_is_kvm(source_kvm_file)) {
1725                 ret = -EBADF;
1726                 goto e_source_put;
1727         }
1728
1729         source_kvm = source_kvm_file->private_data;
1730         mutex_lock(&source_kvm->lock);
1731
1732         if (!sev_guest(source_kvm)) {
1733                 ret = -EINVAL;
1734                 goto e_source_unlock;
1735         }
1736
1737         /* Mirrors of mirrors should work, but let's not get silly */
1738         if (is_mirroring_enc_context(source_kvm) || source_kvm == kvm) {
1739                 ret = -EINVAL;
1740                 goto e_source_unlock;
1741         }
1742
1743         asid = to_kvm_svm(source_kvm)->sev_info.asid;
1744
1745         /*
1746          * The mirror kvm holds an enc_context_owner ref so its asid can't
1747          * disappear until we're done with it
1748          */
1749         kvm_get_kvm(source_kvm);
1750
1751         fput(source_kvm_file);
1752         mutex_unlock(&source_kvm->lock);
1753         mutex_lock(&kvm->lock);
1754
1755         if (sev_guest(kvm)) {
1756                 ret = -EINVAL;
1757                 goto e_mirror_unlock;
1758         }
1759
1760         /* Set enc_context_owner and copy its encryption context over */
1761         mirror_sev = &to_kvm_svm(kvm)->sev_info;
1762         mirror_sev->enc_context_owner = source_kvm;
1763         mirror_sev->asid = asid;
1764         mirror_sev->active = true;
1765
1766         mutex_unlock(&kvm->lock);
1767         return 0;
1768
1769 e_mirror_unlock:
1770         mutex_unlock(&kvm->lock);
1771         kvm_put_kvm(source_kvm);
1772         return ret;
1773 e_source_unlock:
1774         mutex_unlock(&source_kvm->lock);
1775 e_source_put:
1776         if (source_kvm_file)
1777                 fput(source_kvm_file);
1778         return ret;
1779 }
1780
1781 void sev_vm_destroy(struct kvm *kvm)
1782 {
1783         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1784         struct list_head *head = &sev->regions_list;
1785         struct list_head *pos, *q;
1786
1787         if (!sev_guest(kvm))
1788                 return;
1789
1790         /* If this is a mirror_kvm release the enc_context_owner and skip sev cleanup */
1791         if (is_mirroring_enc_context(kvm)) {
1792                 kvm_put_kvm(sev->enc_context_owner);
1793                 return;
1794         }
1795
1796         mutex_lock(&kvm->lock);
1797
1798         /*
1799          * Ensure that all guest tagged cache entries are flushed before
1800          * releasing the pages back to the system for use. CLFLUSH will
1801          * not do this, so issue a WBINVD.
1802          */
1803         wbinvd_on_all_cpus();
1804
1805         /*
1806          * if userspace was terminated before unregistering the memory regions
1807          * then lets unpin all the registered memory.
1808          */
1809         if (!list_empty(head)) {
1810                 list_for_each_safe(pos, q, head) {
1811                         __unregister_enc_region_locked(kvm,
1812                                 list_entry(pos, struct enc_region, list));
1813                         cond_resched();
1814                 }
1815         }
1816
1817         mutex_unlock(&kvm->lock);
1818
1819         sev_unbind_asid(kvm, sev->handle);
1820         sev_asid_free(sev);
1821 }
1822
1823 void __init sev_set_cpu_caps(void)
1824 {
1825         if (!sev_enabled)
1826                 kvm_cpu_cap_clear(X86_FEATURE_SEV);
1827         if (!sev_es_enabled)
1828                 kvm_cpu_cap_clear(X86_FEATURE_SEV_ES);
1829 }
1830
1831 void __init sev_hardware_setup(void)
1832 {
1833 #ifdef CONFIG_KVM_AMD_SEV
1834         unsigned int eax, ebx, ecx, edx, sev_asid_count, sev_es_asid_count;
1835         bool sev_es_supported = false;
1836         bool sev_supported = false;
1837
1838         if (!sev_enabled || !npt_enabled)
1839                 goto out;
1840
1841         /* Does the CPU support SEV? */
1842         if (!boot_cpu_has(X86_FEATURE_SEV))
1843                 goto out;
1844
1845         /* Retrieve SEV CPUID information */
1846         cpuid(0x8000001f, &eax, &ebx, &ecx, &edx);
1847
1848         /* Set encryption bit location for SEV-ES guests */
1849         sev_enc_bit = ebx & 0x3f;
1850
1851         /* Maximum number of encrypted guests supported simultaneously */
1852         max_sev_asid = ecx;
1853         if (!max_sev_asid)
1854                 goto out;
1855
1856         /* Minimum ASID value that should be used for SEV guest */
1857         min_sev_asid = edx;
1858         sev_me_mask = 1UL << (ebx & 0x3f);
1859
1860         /*
1861          * Initialize SEV ASID bitmaps. Allocate space for ASID 0 in the bitmap,
1862          * even though it's never used, so that the bitmap is indexed by the
1863          * actual ASID.
1864          */
1865         nr_asids = max_sev_asid + 1;
1866         sev_asid_bitmap = bitmap_zalloc(nr_asids, GFP_KERNEL);
1867         if (!sev_asid_bitmap)
1868                 goto out;
1869
1870         sev_reclaim_asid_bitmap = bitmap_zalloc(nr_asids, GFP_KERNEL);
1871         if (!sev_reclaim_asid_bitmap) {
1872                 bitmap_free(sev_asid_bitmap);
1873                 sev_asid_bitmap = NULL;
1874                 goto out;
1875         }
1876
1877         sev_asid_count = max_sev_asid - min_sev_asid + 1;
1878         if (misc_cg_set_capacity(MISC_CG_RES_SEV, sev_asid_count))
1879                 goto out;
1880
1881         pr_info("SEV supported: %u ASIDs\n", sev_asid_count);
1882         sev_supported = true;
1883
1884         /* SEV-ES support requested? */
1885         if (!sev_es_enabled)
1886                 goto out;
1887
1888         /* Does the CPU support SEV-ES? */
1889         if (!boot_cpu_has(X86_FEATURE_SEV_ES))
1890                 goto out;
1891
1892         /* Has the system been allocated ASIDs for SEV-ES? */
1893         if (min_sev_asid == 1)
1894                 goto out;
1895
1896         sev_es_asid_count = min_sev_asid - 1;
1897         if (misc_cg_set_capacity(MISC_CG_RES_SEV_ES, sev_es_asid_count))
1898                 goto out;
1899
1900         pr_info("SEV-ES supported: %u ASIDs\n", sev_es_asid_count);
1901         sev_es_supported = true;
1902
1903 out:
1904         sev_enabled = sev_supported;
1905         sev_es_enabled = sev_es_supported;
1906 #endif
1907 }
1908
1909 void sev_hardware_teardown(void)
1910 {
1911         if (!sev_enabled)
1912                 return;
1913
1914         /* No need to take sev_bitmap_lock, all VMs have been destroyed. */
1915         sev_flush_asids(1, max_sev_asid);
1916
1917         bitmap_free(sev_asid_bitmap);
1918         bitmap_free(sev_reclaim_asid_bitmap);
1919
1920         misc_cg_set_capacity(MISC_CG_RES_SEV, 0);
1921         misc_cg_set_capacity(MISC_CG_RES_SEV_ES, 0);
1922 }
1923
1924 int sev_cpu_init(struct svm_cpu_data *sd)
1925 {
1926         if (!sev_enabled)
1927                 return 0;
1928
1929         sd->sev_vmcbs = kcalloc(nr_asids, sizeof(void *), GFP_KERNEL);
1930         if (!sd->sev_vmcbs)
1931                 return -ENOMEM;
1932
1933         return 0;
1934 }
1935
1936 /*
1937  * Pages used by hardware to hold guest encrypted state must be flushed before
1938  * returning them to the system.
1939  */
1940 static void sev_flush_guest_memory(struct vcpu_svm *svm, void *va,
1941                                    unsigned long len)
1942 {
1943         /*
1944          * If hardware enforced cache coherency for encrypted mappings of the
1945          * same physical page is supported, nothing to do.
1946          */
1947         if (boot_cpu_has(X86_FEATURE_SME_COHERENT))
1948                 return;
1949
1950         /*
1951          * If the VM Page Flush MSR is supported, use it to flush the page
1952          * (using the page virtual address and the guest ASID).
1953          */
1954         if (boot_cpu_has(X86_FEATURE_VM_PAGE_FLUSH)) {
1955                 struct kvm_sev_info *sev;
1956                 unsigned long va_start;
1957                 u64 start, stop;
1958
1959                 /* Align start and stop to page boundaries. */
1960                 va_start = (unsigned long)va;
1961                 start = (u64)va_start & PAGE_MASK;
1962                 stop = PAGE_ALIGN((u64)va_start + len);
1963
1964                 if (start < stop) {
1965                         sev = &to_kvm_svm(svm->vcpu.kvm)->sev_info;
1966
1967                         while (start < stop) {
1968                                 wrmsrl(MSR_AMD64_VM_PAGE_FLUSH,
1969                                        start | sev->asid);
1970
1971                                 start += PAGE_SIZE;
1972                         }
1973
1974                         return;
1975                 }
1976
1977                 WARN(1, "Address overflow, using WBINVD\n");
1978         }
1979
1980         /*
1981          * Hardware should always have one of the above features,
1982          * but if not, use WBINVD and issue a warning.
1983          */
1984         WARN_ONCE(1, "Using WBINVD to flush guest memory\n");
1985         wbinvd_on_all_cpus();
1986 }
1987
1988 void sev_free_vcpu(struct kvm_vcpu *vcpu)
1989 {
1990         struct vcpu_svm *svm;
1991
1992         if (!sev_es_guest(vcpu->kvm))
1993                 return;
1994
1995         svm = to_svm(vcpu);
1996
1997         if (vcpu->arch.guest_state_protected)
1998                 sev_flush_guest_memory(svm, svm->vmsa, PAGE_SIZE);
1999         __free_page(virt_to_page(svm->vmsa));
2000
2001         if (svm->ghcb_sa_free)
2002                 kfree(svm->ghcb_sa);
2003 }
2004
2005 static void dump_ghcb(struct vcpu_svm *svm)
2006 {
2007         struct ghcb *ghcb = svm->ghcb;
2008         unsigned int nbits;
2009
2010         /* Re-use the dump_invalid_vmcb module parameter */
2011         if (!dump_invalid_vmcb) {
2012                 pr_warn_ratelimited("set kvm_amd.dump_invalid_vmcb=1 to dump internal KVM state.\n");
2013                 return;
2014         }
2015
2016         nbits = sizeof(ghcb->save.valid_bitmap) * 8;
2017
2018         pr_err("GHCB (GPA=%016llx):\n", svm->vmcb->control.ghcb_gpa);
2019         pr_err("%-20s%016llx is_valid: %u\n", "sw_exit_code",
2020                ghcb->save.sw_exit_code, ghcb_sw_exit_code_is_valid(ghcb));
2021         pr_err("%-20s%016llx is_valid: %u\n", "sw_exit_info_1",
2022                ghcb->save.sw_exit_info_1, ghcb_sw_exit_info_1_is_valid(ghcb));
2023         pr_err("%-20s%016llx is_valid: %u\n", "sw_exit_info_2",
2024                ghcb->save.sw_exit_info_2, ghcb_sw_exit_info_2_is_valid(ghcb));
2025         pr_err("%-20s%016llx is_valid: %u\n", "sw_scratch",
2026                ghcb->save.sw_scratch, ghcb_sw_scratch_is_valid(ghcb));
2027         pr_err("%-20s%*pb\n", "valid_bitmap", nbits, ghcb->save.valid_bitmap);
2028 }
2029
2030 static void sev_es_sync_to_ghcb(struct vcpu_svm *svm)
2031 {
2032         struct kvm_vcpu *vcpu = &svm->vcpu;
2033         struct ghcb *ghcb = svm->ghcb;
2034
2035         /*
2036          * The GHCB protocol so far allows for the following data
2037          * to be returned:
2038          *   GPRs RAX, RBX, RCX, RDX
2039          *
2040          * Copy their values, even if they may not have been written during the
2041          * VM-Exit.  It's the guest's responsibility to not consume random data.
2042          */
2043         ghcb_set_rax(ghcb, vcpu->arch.regs[VCPU_REGS_RAX]);
2044         ghcb_set_rbx(ghcb, vcpu->arch.regs[VCPU_REGS_RBX]);
2045         ghcb_set_rcx(ghcb, vcpu->arch.regs[VCPU_REGS_RCX]);
2046         ghcb_set_rdx(ghcb, vcpu->arch.regs[VCPU_REGS_RDX]);
2047 }
2048
2049 static void sev_es_sync_from_ghcb(struct vcpu_svm *svm)
2050 {
2051         struct vmcb_control_area *control = &svm->vmcb->control;
2052         struct kvm_vcpu *vcpu = &svm->vcpu;
2053         struct ghcb *ghcb = svm->ghcb;
2054         u64 exit_code;
2055
2056         /*
2057          * The GHCB protocol so far allows for the following data
2058          * to be supplied:
2059          *   GPRs RAX, RBX, RCX, RDX
2060          *   XCR0
2061          *   CPL
2062          *
2063          * VMMCALL allows the guest to provide extra registers. KVM also
2064          * expects RSI for hypercalls, so include that, too.
2065          *
2066          * Copy their values to the appropriate location if supplied.
2067          */
2068         memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
2069
2070         vcpu->arch.regs[VCPU_REGS_RAX] = ghcb_get_rax_if_valid(ghcb);
2071         vcpu->arch.regs[VCPU_REGS_RBX] = ghcb_get_rbx_if_valid(ghcb);
2072         vcpu->arch.regs[VCPU_REGS_RCX] = ghcb_get_rcx_if_valid(ghcb);
2073         vcpu->arch.regs[VCPU_REGS_RDX] = ghcb_get_rdx_if_valid(ghcb);
2074         vcpu->arch.regs[VCPU_REGS_RSI] = ghcb_get_rsi_if_valid(ghcb);
2075
2076         svm->vmcb->save.cpl = ghcb_get_cpl_if_valid(ghcb);
2077
2078         if (ghcb_xcr0_is_valid(ghcb)) {
2079                 vcpu->arch.xcr0 = ghcb_get_xcr0(ghcb);
2080                 kvm_update_cpuid_runtime(vcpu);
2081         }
2082
2083         /* Copy the GHCB exit information into the VMCB fields */
2084         exit_code = ghcb_get_sw_exit_code(ghcb);
2085         control->exit_code = lower_32_bits(exit_code);
2086         control->exit_code_hi = upper_32_bits(exit_code);
2087         control->exit_info_1 = ghcb_get_sw_exit_info_1(ghcb);
2088         control->exit_info_2 = ghcb_get_sw_exit_info_2(ghcb);
2089
2090         /* Clear the valid entries fields */
2091         memset(ghcb->save.valid_bitmap, 0, sizeof(ghcb->save.valid_bitmap));
2092 }
2093
2094 static int sev_es_validate_vmgexit(struct vcpu_svm *svm)
2095 {
2096         struct kvm_vcpu *vcpu;
2097         struct ghcb *ghcb;
2098         u64 exit_code = 0;
2099
2100         ghcb = svm->ghcb;
2101
2102         /* Only GHCB Usage code 0 is supported */
2103         if (ghcb->ghcb_usage)
2104                 goto vmgexit_err;
2105
2106         /*
2107          * Retrieve the exit code now even though is may not be marked valid
2108          * as it could help with debugging.
2109          */
2110         exit_code = ghcb_get_sw_exit_code(ghcb);
2111
2112         if (!ghcb_sw_exit_code_is_valid(ghcb) ||
2113             !ghcb_sw_exit_info_1_is_valid(ghcb) ||
2114             !ghcb_sw_exit_info_2_is_valid(ghcb))
2115                 goto vmgexit_err;
2116
2117         switch (ghcb_get_sw_exit_code(ghcb)) {
2118         case SVM_EXIT_READ_DR7:
2119                 break;
2120         case SVM_EXIT_WRITE_DR7:
2121                 if (!ghcb_rax_is_valid(ghcb))
2122                         goto vmgexit_err;
2123                 break;
2124         case SVM_EXIT_RDTSC:
2125                 break;
2126         case SVM_EXIT_RDPMC:
2127                 if (!ghcb_rcx_is_valid(ghcb))
2128                         goto vmgexit_err;
2129                 break;
2130         case SVM_EXIT_CPUID:
2131                 if (!ghcb_rax_is_valid(ghcb) ||
2132                     !ghcb_rcx_is_valid(ghcb))
2133                         goto vmgexit_err;
2134                 if (ghcb_get_rax(ghcb) == 0xd)
2135                         if (!ghcb_xcr0_is_valid(ghcb))
2136                                 goto vmgexit_err;
2137                 break;
2138         case SVM_EXIT_INVD:
2139                 break;
2140         case SVM_EXIT_IOIO:
2141                 if (ghcb_get_sw_exit_info_1(ghcb) & SVM_IOIO_STR_MASK) {
2142                         if (!ghcb_sw_scratch_is_valid(ghcb))
2143                                 goto vmgexit_err;
2144                 } else {
2145                         if (!(ghcb_get_sw_exit_info_1(ghcb) & SVM_IOIO_TYPE_MASK))
2146                                 if (!ghcb_rax_is_valid(ghcb))
2147                                         goto vmgexit_err;
2148                 }
2149                 break;
2150         case SVM_EXIT_MSR:
2151                 if (!ghcb_rcx_is_valid(ghcb))
2152                         goto vmgexit_err;
2153                 if (ghcb_get_sw_exit_info_1(ghcb)) {
2154                         if (!ghcb_rax_is_valid(ghcb) ||
2155                             !ghcb_rdx_is_valid(ghcb))
2156                                 goto vmgexit_err;
2157                 }
2158                 break;
2159         case SVM_EXIT_VMMCALL:
2160                 if (!ghcb_rax_is_valid(ghcb) ||
2161                     !ghcb_cpl_is_valid(ghcb))
2162                         goto vmgexit_err;
2163                 break;
2164         case SVM_EXIT_RDTSCP:
2165                 break;
2166         case SVM_EXIT_WBINVD:
2167                 break;
2168         case SVM_EXIT_MONITOR:
2169                 if (!ghcb_rax_is_valid(ghcb) ||
2170                     !ghcb_rcx_is_valid(ghcb) ||
2171                     !ghcb_rdx_is_valid(ghcb))
2172                         goto vmgexit_err;
2173                 break;
2174         case SVM_EXIT_MWAIT:
2175                 if (!ghcb_rax_is_valid(ghcb) ||
2176                     !ghcb_rcx_is_valid(ghcb))
2177                         goto vmgexit_err;
2178                 break;
2179         case SVM_VMGEXIT_MMIO_READ:
2180         case SVM_VMGEXIT_MMIO_WRITE:
2181                 if (!ghcb_sw_scratch_is_valid(ghcb))
2182                         goto vmgexit_err;
2183                 break;
2184         case SVM_VMGEXIT_NMI_COMPLETE:
2185         case SVM_VMGEXIT_AP_HLT_LOOP:
2186         case SVM_VMGEXIT_AP_JUMP_TABLE:
2187         case SVM_VMGEXIT_UNSUPPORTED_EVENT:
2188                 break;
2189         default:
2190                 goto vmgexit_err;
2191         }
2192
2193         return 0;
2194
2195 vmgexit_err:
2196         vcpu = &svm->vcpu;
2197
2198         if (ghcb->ghcb_usage) {
2199                 vcpu_unimpl(vcpu, "vmgexit: ghcb usage %#x is not valid\n",
2200                             ghcb->ghcb_usage);
2201         } else {
2202                 vcpu_unimpl(vcpu, "vmgexit: exit reason %#llx is not valid\n",
2203                             exit_code);
2204                 dump_ghcb(svm);
2205         }
2206
2207         vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2208         vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON;
2209         vcpu->run->internal.ndata = 2;
2210         vcpu->run->internal.data[0] = exit_code;
2211         vcpu->run->internal.data[1] = vcpu->arch.last_vmentry_cpu;
2212
2213         return -EINVAL;
2214 }
2215
2216 void sev_es_unmap_ghcb(struct vcpu_svm *svm)
2217 {
2218         if (!svm->ghcb)
2219                 return;
2220
2221         if (svm->ghcb_sa_free) {
2222                 /*
2223                  * The scratch area lives outside the GHCB, so there is a
2224                  * buffer that, depending on the operation performed, may
2225                  * need to be synced, then freed.
2226                  */
2227                 if (svm->ghcb_sa_sync) {
2228                         kvm_write_guest(svm->vcpu.kvm,
2229                                         ghcb_get_sw_scratch(svm->ghcb),
2230                                         svm->ghcb_sa, svm->ghcb_sa_len);
2231                         svm->ghcb_sa_sync = false;
2232                 }
2233
2234                 kfree(svm->ghcb_sa);
2235                 svm->ghcb_sa = NULL;
2236                 svm->ghcb_sa_free = false;
2237         }
2238
2239         trace_kvm_vmgexit_exit(svm->vcpu.vcpu_id, svm->ghcb);
2240
2241         sev_es_sync_to_ghcb(svm);
2242
2243         kvm_vcpu_unmap(&svm->vcpu, &svm->ghcb_map, true);
2244         svm->ghcb = NULL;
2245 }
2246
2247 void pre_sev_run(struct vcpu_svm *svm, int cpu)
2248 {
2249         struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
2250         int asid = sev_get_asid(svm->vcpu.kvm);
2251
2252         /* Assign the asid allocated with this SEV guest */
2253         svm->asid = asid;
2254
2255         /*
2256          * Flush guest TLB:
2257          *
2258          * 1) when different VMCB for the same ASID is to be run on the same host CPU.
2259          * 2) or this VMCB was executed on different host CPU in previous VMRUNs.
2260          */
2261         if (sd->sev_vmcbs[asid] == svm->vmcb &&
2262             svm->vcpu.arch.last_vmentry_cpu == cpu)
2263                 return;
2264
2265         sd->sev_vmcbs[asid] = svm->vmcb;
2266         svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
2267         vmcb_mark_dirty(svm->vmcb, VMCB_ASID);
2268 }
2269
2270 #define GHCB_SCRATCH_AREA_LIMIT         (16ULL * PAGE_SIZE)
2271 static bool setup_vmgexit_scratch(struct vcpu_svm *svm, bool sync, u64 len)
2272 {
2273         struct vmcb_control_area *control = &svm->vmcb->control;
2274         struct ghcb *ghcb = svm->ghcb;
2275         u64 ghcb_scratch_beg, ghcb_scratch_end;
2276         u64 scratch_gpa_beg, scratch_gpa_end;
2277         void *scratch_va;
2278
2279         scratch_gpa_beg = ghcb_get_sw_scratch(ghcb);
2280         if (!scratch_gpa_beg) {
2281                 pr_err("vmgexit: scratch gpa not provided\n");
2282                 return false;
2283         }
2284
2285         scratch_gpa_end = scratch_gpa_beg + len;
2286         if (scratch_gpa_end < scratch_gpa_beg) {
2287                 pr_err("vmgexit: scratch length (%#llx) not valid for scratch address (%#llx)\n",
2288                        len, scratch_gpa_beg);
2289                 return false;
2290         }
2291
2292         if ((scratch_gpa_beg & PAGE_MASK) == control->ghcb_gpa) {
2293                 /* Scratch area begins within GHCB */
2294                 ghcb_scratch_beg = control->ghcb_gpa +
2295                                    offsetof(struct ghcb, shared_buffer);
2296                 ghcb_scratch_end = control->ghcb_gpa +
2297                                    offsetof(struct ghcb, reserved_1);
2298
2299                 /*
2300                  * If the scratch area begins within the GHCB, it must be
2301                  * completely contained in the GHCB shared buffer area.
2302                  */
2303                 if (scratch_gpa_beg < ghcb_scratch_beg ||
2304                     scratch_gpa_end > ghcb_scratch_end) {
2305                         pr_err("vmgexit: scratch area is outside of GHCB shared buffer area (%#llx - %#llx)\n",
2306                                scratch_gpa_beg, scratch_gpa_end);
2307                         return false;
2308                 }
2309
2310                 scratch_va = (void *)svm->ghcb;
2311                 scratch_va += (scratch_gpa_beg - control->ghcb_gpa);
2312         } else {
2313                 /*
2314                  * The guest memory must be read into a kernel buffer, so
2315                  * limit the size
2316                  */
2317                 if (len > GHCB_SCRATCH_AREA_LIMIT) {
2318                         pr_err("vmgexit: scratch area exceeds KVM limits (%#llx requested, %#llx limit)\n",
2319                                len, GHCB_SCRATCH_AREA_LIMIT);
2320                         return false;
2321                 }
2322                 scratch_va = kzalloc(len, GFP_KERNEL_ACCOUNT);
2323                 if (!scratch_va)
2324                         return false;
2325
2326                 if (kvm_read_guest(svm->vcpu.kvm, scratch_gpa_beg, scratch_va, len)) {
2327                         /* Unable to copy scratch area from guest */
2328                         pr_err("vmgexit: kvm_read_guest for scratch area failed\n");
2329
2330                         kfree(scratch_va);
2331                         return false;
2332                 }
2333
2334                 /*
2335                  * The scratch area is outside the GHCB. The operation will
2336                  * dictate whether the buffer needs to be synced before running
2337                  * the vCPU next time (i.e. a read was requested so the data
2338                  * must be written back to the guest memory).
2339                  */
2340                 svm->ghcb_sa_sync = sync;
2341                 svm->ghcb_sa_free = true;
2342         }
2343
2344         svm->ghcb_sa = scratch_va;
2345         svm->ghcb_sa_len = len;
2346
2347         return true;
2348 }
2349
2350 static void set_ghcb_msr_bits(struct vcpu_svm *svm, u64 value, u64 mask,
2351                               unsigned int pos)
2352 {
2353         svm->vmcb->control.ghcb_gpa &= ~(mask << pos);
2354         svm->vmcb->control.ghcb_gpa |= (value & mask) << pos;
2355 }
2356
2357 static u64 get_ghcb_msr_bits(struct vcpu_svm *svm, u64 mask, unsigned int pos)
2358 {
2359         return (svm->vmcb->control.ghcb_gpa >> pos) & mask;
2360 }
2361
2362 static void set_ghcb_msr(struct vcpu_svm *svm, u64 value)
2363 {
2364         svm->vmcb->control.ghcb_gpa = value;
2365 }
2366
2367 static int sev_handle_vmgexit_msr_protocol(struct vcpu_svm *svm)
2368 {
2369         struct vmcb_control_area *control = &svm->vmcb->control;
2370         struct kvm_vcpu *vcpu = &svm->vcpu;
2371         u64 ghcb_info;
2372         int ret = 1;
2373
2374         ghcb_info = control->ghcb_gpa & GHCB_MSR_INFO_MASK;
2375
2376         trace_kvm_vmgexit_msr_protocol_enter(svm->vcpu.vcpu_id,
2377                                              control->ghcb_gpa);
2378
2379         switch (ghcb_info) {
2380         case GHCB_MSR_SEV_INFO_REQ:
2381                 set_ghcb_msr(svm, GHCB_MSR_SEV_INFO(GHCB_VERSION_MAX,
2382                                                     GHCB_VERSION_MIN,
2383                                                     sev_enc_bit));
2384                 break;
2385         case GHCB_MSR_CPUID_REQ: {
2386                 u64 cpuid_fn, cpuid_reg, cpuid_value;
2387
2388                 cpuid_fn = get_ghcb_msr_bits(svm,
2389                                              GHCB_MSR_CPUID_FUNC_MASK,
2390                                              GHCB_MSR_CPUID_FUNC_POS);
2391
2392                 /* Initialize the registers needed by the CPUID intercept */
2393                 vcpu->arch.regs[VCPU_REGS_RAX] = cpuid_fn;
2394                 vcpu->arch.regs[VCPU_REGS_RCX] = 0;
2395
2396                 ret = svm_invoke_exit_handler(vcpu, SVM_EXIT_CPUID);
2397                 if (!ret) {
2398                         ret = -EINVAL;
2399                         break;
2400                 }
2401
2402                 cpuid_reg = get_ghcb_msr_bits(svm,
2403                                               GHCB_MSR_CPUID_REG_MASK,
2404                                               GHCB_MSR_CPUID_REG_POS);
2405                 if (cpuid_reg == 0)
2406                         cpuid_value = vcpu->arch.regs[VCPU_REGS_RAX];
2407                 else if (cpuid_reg == 1)
2408                         cpuid_value = vcpu->arch.regs[VCPU_REGS_RBX];
2409                 else if (cpuid_reg == 2)
2410                         cpuid_value = vcpu->arch.regs[VCPU_REGS_RCX];
2411                 else
2412                         cpuid_value = vcpu->arch.regs[VCPU_REGS_RDX];
2413
2414                 set_ghcb_msr_bits(svm, cpuid_value,
2415                                   GHCB_MSR_CPUID_VALUE_MASK,
2416                                   GHCB_MSR_CPUID_VALUE_POS);
2417
2418                 set_ghcb_msr_bits(svm, GHCB_MSR_CPUID_RESP,
2419                                   GHCB_MSR_INFO_MASK,
2420                                   GHCB_MSR_INFO_POS);
2421                 break;
2422         }
2423         case GHCB_MSR_TERM_REQ: {
2424                 u64 reason_set, reason_code;
2425
2426                 reason_set = get_ghcb_msr_bits(svm,
2427                                                GHCB_MSR_TERM_REASON_SET_MASK,
2428                                                GHCB_MSR_TERM_REASON_SET_POS);
2429                 reason_code = get_ghcb_msr_bits(svm,
2430                                                 GHCB_MSR_TERM_REASON_MASK,
2431                                                 GHCB_MSR_TERM_REASON_POS);
2432                 pr_info("SEV-ES guest requested termination: %#llx:%#llx\n",
2433                         reason_set, reason_code);
2434                 fallthrough;
2435         }
2436         default:
2437                 ret = -EINVAL;
2438         }
2439
2440         trace_kvm_vmgexit_msr_protocol_exit(svm->vcpu.vcpu_id,
2441                                             control->ghcb_gpa, ret);
2442
2443         return ret;
2444 }
2445
2446 int sev_handle_vmgexit(struct kvm_vcpu *vcpu)
2447 {
2448         struct vcpu_svm *svm = to_svm(vcpu);
2449         struct vmcb_control_area *control = &svm->vmcb->control;
2450         u64 ghcb_gpa, exit_code;
2451         struct ghcb *ghcb;
2452         int ret;
2453
2454         /* Validate the GHCB */
2455         ghcb_gpa = control->ghcb_gpa;
2456         if (ghcb_gpa & GHCB_MSR_INFO_MASK)
2457                 return sev_handle_vmgexit_msr_protocol(svm);
2458
2459         if (!ghcb_gpa) {
2460                 vcpu_unimpl(vcpu, "vmgexit: GHCB gpa is not set\n");
2461                 return -EINVAL;
2462         }
2463
2464         if (kvm_vcpu_map(vcpu, ghcb_gpa >> PAGE_SHIFT, &svm->ghcb_map)) {
2465                 /* Unable to map GHCB from guest */
2466                 vcpu_unimpl(vcpu, "vmgexit: error mapping GHCB [%#llx] from guest\n",
2467                             ghcb_gpa);
2468                 return -EINVAL;
2469         }
2470
2471         svm->ghcb = svm->ghcb_map.hva;
2472         ghcb = svm->ghcb_map.hva;
2473
2474         trace_kvm_vmgexit_enter(vcpu->vcpu_id, ghcb);
2475
2476         exit_code = ghcb_get_sw_exit_code(ghcb);
2477
2478         ret = sev_es_validate_vmgexit(svm);
2479         if (ret)
2480                 return ret;
2481
2482         sev_es_sync_from_ghcb(svm);
2483         ghcb_set_sw_exit_info_1(ghcb, 0);
2484         ghcb_set_sw_exit_info_2(ghcb, 0);
2485
2486         ret = -EINVAL;
2487         switch (exit_code) {
2488         case SVM_VMGEXIT_MMIO_READ:
2489                 if (!setup_vmgexit_scratch(svm, true, control->exit_info_2))
2490                         break;
2491
2492                 ret = kvm_sev_es_mmio_read(vcpu,
2493                                            control->exit_info_1,
2494                                            control->exit_info_2,
2495                                            svm->ghcb_sa);
2496                 break;
2497         case SVM_VMGEXIT_MMIO_WRITE:
2498                 if (!setup_vmgexit_scratch(svm, false, control->exit_info_2))
2499                         break;
2500
2501                 ret = kvm_sev_es_mmio_write(vcpu,
2502                                             control->exit_info_1,
2503                                             control->exit_info_2,
2504                                             svm->ghcb_sa);
2505                 break;
2506         case SVM_VMGEXIT_NMI_COMPLETE:
2507                 ret = svm_invoke_exit_handler(vcpu, SVM_EXIT_IRET);
2508                 break;
2509         case SVM_VMGEXIT_AP_HLT_LOOP:
2510                 ret = kvm_emulate_ap_reset_hold(vcpu);
2511                 break;
2512         case SVM_VMGEXIT_AP_JUMP_TABLE: {
2513                 struct kvm_sev_info *sev = &to_kvm_svm(vcpu->kvm)->sev_info;
2514
2515                 switch (control->exit_info_1) {
2516                 case 0:
2517                         /* Set AP jump table address */
2518                         sev->ap_jump_table = control->exit_info_2;
2519                         break;
2520                 case 1:
2521                         /* Get AP jump table address */
2522                         ghcb_set_sw_exit_info_2(ghcb, sev->ap_jump_table);
2523                         break;
2524                 default:
2525                         pr_err("svm: vmgexit: unsupported AP jump table request - exit_info_1=%#llx\n",
2526                                control->exit_info_1);
2527                         ghcb_set_sw_exit_info_1(ghcb, 1);
2528                         ghcb_set_sw_exit_info_2(ghcb,
2529                                                 X86_TRAP_UD |
2530                                                 SVM_EVTINJ_TYPE_EXEPT |
2531                                                 SVM_EVTINJ_VALID);
2532                 }
2533
2534                 ret = 1;
2535                 break;
2536         }
2537         case SVM_VMGEXIT_UNSUPPORTED_EVENT:
2538                 vcpu_unimpl(vcpu,
2539                             "vmgexit: unsupported event - exit_info_1=%#llx, exit_info_2=%#llx\n",
2540                             control->exit_info_1, control->exit_info_2);
2541                 break;
2542         default:
2543                 ret = svm_invoke_exit_handler(vcpu, exit_code);
2544         }
2545
2546         return ret;
2547 }
2548
2549 int sev_es_string_io(struct vcpu_svm *svm, int size, unsigned int port, int in)
2550 {
2551         if (!setup_vmgexit_scratch(svm, in, svm->vmcb->control.exit_info_2))
2552                 return -EINVAL;
2553
2554         return kvm_sev_es_string_io(&svm->vcpu, size, port,
2555                                     svm->ghcb_sa, svm->ghcb_sa_len, in);
2556 }
2557
2558 void sev_es_init_vmcb(struct vcpu_svm *svm)
2559 {
2560         struct kvm_vcpu *vcpu = &svm->vcpu;
2561
2562         svm->vmcb->control.nested_ctl |= SVM_NESTED_CTL_SEV_ES_ENABLE;
2563         svm->vmcb->control.virt_ext |= LBR_CTL_ENABLE_MASK;
2564
2565         /*
2566          * An SEV-ES guest requires a VMSA area that is a separate from the
2567          * VMCB page. Do not include the encryption mask on the VMSA physical
2568          * address since hardware will access it using the guest key.
2569          */
2570         svm->vmcb->control.vmsa_pa = __pa(svm->vmsa);
2571
2572         /* Can't intercept CR register access, HV can't modify CR registers */
2573         svm_clr_intercept(svm, INTERCEPT_CR0_READ);
2574         svm_clr_intercept(svm, INTERCEPT_CR4_READ);
2575         svm_clr_intercept(svm, INTERCEPT_CR8_READ);
2576         svm_clr_intercept(svm, INTERCEPT_CR0_WRITE);
2577         svm_clr_intercept(svm, INTERCEPT_CR4_WRITE);
2578         svm_clr_intercept(svm, INTERCEPT_CR8_WRITE);
2579
2580         svm_clr_intercept(svm, INTERCEPT_SELECTIVE_CR0);
2581
2582         /* Track EFER/CR register changes */
2583         svm_set_intercept(svm, TRAP_EFER_WRITE);
2584         svm_set_intercept(svm, TRAP_CR0_WRITE);
2585         svm_set_intercept(svm, TRAP_CR4_WRITE);
2586         svm_set_intercept(svm, TRAP_CR8_WRITE);
2587
2588         /* No support for enable_vmware_backdoor */
2589         clr_exception_intercept(svm, GP_VECTOR);
2590
2591         /* Can't intercept XSETBV, HV can't modify XCR0 directly */
2592         svm_clr_intercept(svm, INTERCEPT_XSETBV);
2593
2594         /* Clear intercepts on selected MSRs */
2595         set_msr_interception(vcpu, svm->msrpm, MSR_EFER, 1, 1);
2596         set_msr_interception(vcpu, svm->msrpm, MSR_IA32_CR_PAT, 1, 1);
2597         set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
2598         set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
2599         set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
2600         set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
2601 }
2602
2603 void sev_es_create_vcpu(struct vcpu_svm *svm)
2604 {
2605         /*
2606          * Set the GHCB MSR value as per the GHCB specification when creating
2607          * a vCPU for an SEV-ES guest.
2608          */
2609         set_ghcb_msr(svm, GHCB_MSR_SEV_INFO(GHCB_VERSION_MAX,
2610                                             GHCB_VERSION_MIN,
2611                                             sev_enc_bit));
2612 }
2613
2614 void sev_es_prepare_guest_switch(struct vcpu_svm *svm, unsigned int cpu)
2615 {
2616         struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
2617         struct vmcb_save_area *hostsa;
2618
2619         /*
2620          * As an SEV-ES guest, hardware will restore the host state on VMEXIT,
2621          * of which one step is to perform a VMLOAD. Since hardware does not
2622          * perform a VMSAVE on VMRUN, the host savearea must be updated.
2623          */
2624         vmsave(__sme_page_pa(sd->save_area));
2625
2626         /* XCR0 is restored on VMEXIT, save the current host value */
2627         hostsa = (struct vmcb_save_area *)(page_address(sd->save_area) + 0x400);
2628         hostsa->xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
2629
2630         /* PKRU is restored on VMEXIT, save the current host value */
2631         hostsa->pkru = read_pkru();
2632
2633         /* MSR_IA32_XSS is restored on VMEXIT, save the currnet host value */
2634         hostsa->xss = host_xss;
2635 }
2636
2637 void sev_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
2638 {
2639         struct vcpu_svm *svm = to_svm(vcpu);
2640
2641         /* First SIPI: Use the values as initially set by the VMM */
2642         if (!svm->received_first_sipi) {
2643                 svm->received_first_sipi = true;
2644                 return;
2645         }
2646
2647         /*
2648          * Subsequent SIPI: Return from an AP Reset Hold VMGEXIT, where
2649          * the guest will set the CS and RIP. Set SW_EXIT_INFO_2 to a
2650          * non-zero value.
2651          */
2652         if (!svm->ghcb)
2653                 return;
2654
2655         ghcb_set_sw_exit_info_2(svm->ghcb, 1);
2656 }