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 #ifndef CONFIG_KVM_AMD_SEV
  32 /*
  33  * When this config is not defined, SEV feature is not supported and APIs in
  34  * this file are not used but this file still gets compiled into the KVM AMD
  35  * module.
  36  *
  37  * We will not have MISC_CG_RES_SEV and MISC_CG_RES_SEV_ES entries in the enum
  38  * misc_res_type {} defined in linux/misc_cgroup.h.
  39  *
  40  * Below macros allow compilation to succeed.
  41  */
  42 #define MISC_CG_RES_SEV MISC_CG_RES_TYPES
  43 #define MISC_CG_RES_SEV_ES MISC_CG_RES_TYPES
  44 #endif
  45
  46 #ifdef CONFIG_KVM_AMD_SEV
  47 /* enable/disable SEV support */
  48 static bool sev_enabled = true;
  49 module_param_named(sev, sev_enabled, bool, 0444);
  50
  51 /* enable/disable SEV-ES support */
  52 static bool sev_es_enabled = true;
  53 module_param_named(sev_es, sev_es_enabled, bool, 0444);
  54 #else
  55 #define sev_enabled false
  56 #define sev_es_enabled false
  57 #endif /* CONFIG_KVM_AMD_SEV */
  58
  59 static u8 sev_enc_bit;
  60 static DECLARE_RWSEM(sev_deactivate_lock);
  61 static DEFINE_MUTEX(sev_bitmap_lock);
  62 unsigned int max_sev_asid;
  63 static unsigned int min_sev_asid;
  64 static unsigned long sev_me_mask;
  65 static unsigned int nr_asids;
  66 static unsigned long *sev_asid_bitmap;
  67 static unsigned long *sev_reclaim_asid_bitmap;
  68
  69 struct enc_region {
  70         struct list_head list;
  71         unsigned long npages;
  72         struct page **pages;
  73         unsigned long uaddr;
  74         unsigned long size;
  75 };
  76
  77 /* Called with the sev_bitmap_lock held, or on shutdown  */
  78 static int sev_flush_asids(int min_asid, int max_asid)
  79 {
  80         int ret, asid, error = 0;
  81
  82         /* Check if there are any ASIDs to reclaim before performing a flush */
  83         asid = find_next_bit(sev_reclaim_asid_bitmap, nr_asids, min_asid);
  84         if (asid > max_asid)
  85                 return -EBUSY;
  86
  87         /*
  88          * DEACTIVATE will clear the WBINVD indicator causing DF_FLUSH to fail,
  89          * so it must be guarded.
  90          */
  91         down_write(&sev_deactivate_lock);
  92
  93         wbinvd_on_all_cpus();
  94         ret = sev_guest_df_flush(&error);
  95
  96         up_write(&sev_deactivate_lock);
  97
  98         if (ret)
  99                 pr_err("SEV: DF_FLUSH failed, ret=%d, error=%#x\n", ret, error);
 100
 101         return ret;
 102 }
 103
 104 static inline bool is_mirroring_enc_context(struct kvm *kvm)
 105 {
 106         return !!to_kvm_svm(kvm)->sev_info.enc_context_owner;
 107 }
 108
 109 /* Must be called with the sev_bitmap_lock held */
 110 static bool __sev_recycle_asids(int min_asid, int max_asid)
 111 {
 112         if (sev_flush_asids(min_asid, max_asid))
 113                 return false;
 114
 115         /* The flush process will flush all reclaimable SEV and SEV-ES ASIDs */
 116         bitmap_xor(sev_asid_bitmap, sev_asid_bitmap, sev_reclaim_asid_bitmap,
 117                    nr_asids);
 118         bitmap_zero(sev_reclaim_asid_bitmap, nr_asids);
 119
 120         return true;
 121 }
 122
 123 static int sev_asid_new(struct kvm_sev_info *sev)
 124 {
 125         int asid, min_asid, max_asid, ret;
 126         bool retry = true;
 127         enum misc_res_type type;
 128
 129         type = sev->es_active ? MISC_CG_RES_SEV_ES : MISC_CG_RES_SEV;
 130         WARN_ON(sev->misc_cg);
 131         sev->misc_cg = get_current_misc_cg();
 132         ret = misc_cg_try_charge(type, sev->misc_cg, 1);
 133         if (ret) {
 134                 put_misc_cg(sev->misc_cg);
 135                 sev->misc_cg = NULL;
 136                 return ret;
 137         }
 138
 139         mutex_lock(&sev_bitmap_lock);
 140
 141         /*
 142          * SEV-enabled guests must use asid from min_sev_asid to max_sev_asid.
 143          * SEV-ES-enabled guest can use from 1 to min_sev_asid - 1.
 144          */
 145         min_asid = sev->es_active ? 1 : min_sev_asid;
 146         max_asid = sev->es_active ? min_sev_asid - 1 : max_sev_asid;
 147 again:
 148         asid = find_next_zero_bit(sev_asid_bitmap, max_asid + 1, min_asid);
 149         if (asid > max_asid) {
 150                 if (retry && __sev_recycle_asids(min_asid, max_asid)) {
 151                         retry = false;
 152                         goto again;
 153                 }
 154                 mutex_unlock(&sev_bitmap_lock);
 155                 ret = -EBUSY;
 156                 goto e_uncharge;
 157         }
 158
 159         __set_bit(asid, sev_asid_bitmap);
 160
 161         mutex_unlock(&sev_bitmap_lock);
 162
 163         return asid;
 164 e_uncharge:
 165         misc_cg_uncharge(type, sev->misc_cg, 1);
 166         put_misc_cg(sev->misc_cg);
 167         sev->misc_cg = NULL;
 168         return ret;
 169 }
 170
 171 static int sev_get_asid(struct kvm *kvm)
 172 {
 173         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 174
 175         return sev->asid;
 176 }
 177
 178 static void sev_asid_free(struct kvm_sev_info *sev)
 179 {
 180         struct svm_cpu_data *sd;
 181         int cpu;
 182         enum misc_res_type type;
 183
 184         mutex_lock(&sev_bitmap_lock);
 185
 186         __set_bit(sev->asid, sev_reclaim_asid_bitmap);
 187
 188         for_each_possible_cpu(cpu) {
 189                 sd = per_cpu(svm_data, cpu);
 190                 sd->sev_vmcbs[sev->asid] = NULL;
 191         }
 192
 193         mutex_unlock(&sev_bitmap_lock);
 194
 195         type = sev->es_active ? MISC_CG_RES_SEV_ES : MISC_CG_RES_SEV;
 196         misc_cg_uncharge(type, sev->misc_cg, 1);
 197         put_misc_cg(sev->misc_cg);
 198         sev->misc_cg = NULL;
 199 }
 200
 201 static void sev_decommission(unsigned int handle)
 202 {
 203         struct sev_data_decommission decommission;
 204
 205         if (!handle)
 206                 return;
 207
 208         decommission.handle = handle;
 209         sev_guest_decommission(&decommission, NULL);
 210 }
 211
 212 static void sev_unbind_asid(struct kvm *kvm, unsigned int handle)
 213 {
 214         struct sev_data_deactivate deactivate;
 215
 216         if (!handle)
 217                 return;
 218
 219         deactivate.handle = handle;
 220
 221         /* Guard DEACTIVATE against WBINVD/DF_FLUSH used in ASID recycling */
 222         down_read(&sev_deactivate_lock);
 223         sev_guest_deactivate(&deactivate, NULL);
 224         up_read(&sev_deactivate_lock);
 225
 226         sev_decommission(handle);
 227 }
 228
 229 static int sev_guest_init(struct kvm *kvm, struct kvm_sev_cmd *argp)
 230 {
 231         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 232         bool es_active = argp->id == KVM_SEV_ES_INIT;
 233         int asid, ret;
 234
 235         if (kvm->created_vcpus)
 236                 return -EINVAL;
 237
 238         ret = -EBUSY;
 239         if (unlikely(sev->active))
 240                 return ret;
 241
 242         sev->es_active = es_active;
 243         asid = sev_asid_new(sev);
 244         if (asid < 0)
 245                 goto e_no_asid;
 246         sev->asid = asid;
 247
 248         ret = sev_platform_init(&argp->error);
 249         if (ret)
 250                 goto e_free;
 251
 252         sev->active = true;
 253         sev->asid = asid;
 254         INIT_LIST_HEAD(&sev->regions_list);
 255
 256         return 0;
 257
 258 e_free:
 259         sev_asid_free(sev);
 260         sev->asid = 0;
 261 e_no_asid:
 262         sev->es_active = false;
 263         return ret;
 264 }
 265
 266 static int sev_bind_asid(struct kvm *kvm, unsigned int handle, int *error)
 267 {
 268         struct sev_data_activate activate;
 269         int asid = sev_get_asid(kvm);
 270         int ret;
 271
 272         /* activate ASID on the given handle */
 273         activate.handle = handle;
 274         activate.asid   = asid;
 275         ret = sev_guest_activate(&activate, error);
 276
 277         return ret;
 278 }
 279
 280 static int __sev_issue_cmd(int fd, int id, void *data, int *error)
 281 {
 282         struct fd f;
 283         int ret;
 284
 285         f = fdget(fd);
 286         if (!f.file)
 287                 return -EBADF;
 288
 289         ret = sev_issue_cmd_external_user(f.file, id, data, error);
 290
 291         fdput(f);
 292         return ret;
 293 }
 294
 295 static int sev_issue_cmd(struct kvm *kvm, int id, void *data, int *error)
 296 {
 297         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 298
 299         return __sev_issue_cmd(sev->fd, id, data, error);
 300 }
 301
 302 static int sev_launch_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
 303 {
 304         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 305         struct sev_data_launch_start start;
 306         struct kvm_sev_launch_start params;
 307         void *dh_blob, *session_blob;
 308         int *error = &argp->error;
 309         int ret;
 310
 311         if (!sev_guest(kvm))
 312                 return -ENOTTY;
 313
 314         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data, sizeof(params)))
 315                 return -EFAULT;
 316
 317         memset(&start, 0, sizeof(start));
 318
 319         dh_blob = NULL;
 320         if (params.dh_uaddr) {
 321                 dh_blob = psp_copy_user_blob(params.dh_uaddr, params.dh_len);
 322                 if (IS_ERR(dh_blob))
 323                         return PTR_ERR(dh_blob);
 324
 325                 start.dh_cert_address = __sme_set(__pa(dh_blob));
 326                 start.dh_cert_len = params.dh_len;
 327         }
 328
 329         session_blob = NULL;
 330         if (params.session_uaddr) {
 331                 session_blob = psp_copy_user_blob(params.session_uaddr, params.session_len);
 332                 if (IS_ERR(session_blob)) {
 333                         ret = PTR_ERR(session_blob);
 334                         goto e_free_dh;
 335                 }
 336
 337                 start.session_address = __sme_set(__pa(session_blob));
 338                 start.session_len = params.session_len;
 339         }
 340
 341         start.handle = params.handle;
 342         start.policy = params.policy;
 343
 344         /* create memory encryption context */
 345         ret = __sev_issue_cmd(argp->sev_fd, SEV_CMD_LAUNCH_START, &start, error);
 346         if (ret)
 347                 goto e_free_session;
 348
 349         /* Bind ASID to this guest */
 350         ret = sev_bind_asid(kvm, start.handle, error);
 351         if (ret) {
 352                 sev_decommission(start.handle);
 353                 goto e_free_session;
 354         }
 355
 356         /* return handle to userspace */
 357         params.handle = start.handle;
 358         if (copy_to_user((void __user *)(uintptr_t)argp->data, &params, sizeof(params))) {
 359                 sev_unbind_asid(kvm, start.handle);
 360                 ret = -EFAULT;
 361                 goto e_free_session;
 362         }
 363
 364         sev->handle = start.handle;
 365         sev->fd = argp->sev_fd;
 366
 367 e_free_session:
 368         kfree(session_blob);
 369 e_free_dh:
 370         kfree(dh_blob);
 371         return ret;
 372 }
 373
 374 static struct page **sev_pin_memory(struct kvm *kvm, unsigned long uaddr,
 375                                     unsigned long ulen, unsigned long *n,
 376                                     int write)
 377 {
 378         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 379         unsigned long npages, size;
 380         int npinned;
 381         unsigned long locked, lock_limit;
 382         struct page **pages;
 383         unsigned long first, last;
 384         int ret;
 385
 386         lockdep_assert_held(&kvm->lock);
 387
 388         if (ulen == 0 || uaddr + ulen < uaddr)
 389                 return ERR_PTR(-EINVAL);
 390
 391         /* Calculate number of pages. */
 392         first = (uaddr & PAGE_MASK) >> PAGE_SHIFT;
 393         last = ((uaddr + ulen - 1) & PAGE_MASK) >> PAGE_SHIFT;
 394         npages = (last - first + 1);
 395
 396         locked = sev->pages_locked + npages;
 397         lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
 398         if (locked > lock_limit && !capable(CAP_IPC_LOCK)) {
 399                 pr_err("SEV: %lu locked pages exceed the lock limit of %lu.\n", locked, lock_limit);
 400                 return ERR_PTR(-ENOMEM);
 401         }
 402
 403         if (WARN_ON_ONCE(npages > INT_MAX))
 404                 return ERR_PTR(-EINVAL);
 405
 406         /* Avoid using vmalloc for smaller buffers. */
 407         size = npages * sizeof(struct page *);
 408         if (size > PAGE_SIZE)
 409                 pages = __vmalloc(size, GFP_KERNEL_ACCOUNT | __GFP_ZERO);
 410         else
 411                 pages = kmalloc(size, GFP_KERNEL_ACCOUNT);
 412
 413         if (!pages)
 414                 return ERR_PTR(-ENOMEM);
 415
 416         /* Pin the user virtual address. */
 417         npinned = pin_user_pages_fast(uaddr, npages, write ? FOLL_WRITE : 0, pages);
 418         if (npinned != npages) {
 419                 pr_err("SEV: Failure locking %lu pages.\n", npages);
 420                 ret = -ENOMEM;
 421                 goto err;
 422         }
 423
 424         *n = npages;
 425         sev->pages_locked = locked;
 426
 427         return pages;
 428
 429 err:
 430         if (npinned > 0)
 431                 unpin_user_pages(pages, npinned);
 432
 433         kvfree(pages);
 434         return ERR_PTR(ret);
 435 }
 436
 437 static void sev_unpin_memory(struct kvm *kvm, struct page **pages,
 438                              unsigned long npages)
 439 {
 440         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 441
 442         unpin_user_pages(pages, npages);
 443         kvfree(pages);
 444         sev->pages_locked -= npages;
 445 }
 446
 447 static void sev_clflush_pages(struct page *pages[], unsigned long npages)
 448 {
 449         uint8_t *page_virtual;
 450         unsigned long i;
 451
 452         if (this_cpu_has(X86_FEATURE_SME_COHERENT) || npages == 0 ||
 453             pages == NULL)
 454                 return;
 455
 456         for (i = 0; i < npages; i++) {
 457                 page_virtual = kmap_atomic(pages[i]);
 458                 clflush_cache_range(page_virtual, PAGE_SIZE);
 459                 kunmap_atomic(page_virtual);
 460         }
 461 }
 462
 463 static unsigned long get_num_contig_pages(unsigned long idx,
 464                                 struct page **inpages, unsigned long npages)
 465 {
 466         unsigned long paddr, next_paddr;
 467         unsigned long i = idx + 1, pages = 1;
 468
 469         /* find the number of contiguous pages starting from idx */
 470         paddr = __sme_page_pa(inpages[idx]);
 471         while (i < npages) {
 472                 next_paddr = __sme_page_pa(inpages[i++]);
 473                 if ((paddr + PAGE_SIZE) == next_paddr) {
 474                         pages++;
 475                         paddr = next_paddr;
 476                         continue;
 477                 }
 478                 break;
 479         }
 480
 481         return pages;
 482 }
 483
 484 static int sev_launch_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
 485 {
 486         unsigned long vaddr, vaddr_end, next_vaddr, npages, pages, size, i;
 487         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 488         struct kvm_sev_launch_update_data params;
 489         struct sev_data_launch_update_data data;
 490         struct page **inpages;
 491         int ret;
 492
 493         if (!sev_guest(kvm))
 494                 return -ENOTTY;
 495
 496         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data, sizeof(params)))
 497                 return -EFAULT;
 498
 499         vaddr = params.uaddr;
 500         size = params.len;
 501         vaddr_end = vaddr + size;
 502
 503         /* Lock the user memory. */
 504         inpages = sev_pin_memory(kvm, vaddr, size, &npages, 1);
 505         if (IS_ERR(inpages))
 506                 return PTR_ERR(inpages);
 507
 508         /*
 509          * Flush (on non-coherent CPUs) before LAUNCH_UPDATE encrypts pages in
 510          * place; the cache may contain the data that was written unencrypted.
 511          */
 512         sev_clflush_pages(inpages, npages);
 513
 514         data.reserved = 0;
 515         data.handle = sev->handle;
 516
 517         for (i = 0; vaddr < vaddr_end; vaddr = next_vaddr, i += pages) {
 518                 int offset, len;
 519
 520                 /*
 521                  * If the user buffer is not page-aligned, calculate the offset
 522                  * within the page.
 523                  */
 524                 offset = vaddr & (PAGE_SIZE - 1);
 525
 526                 /* Calculate the number of pages that can be encrypted in one go. */
 527                 pages = get_num_contig_pages(i, inpages, npages);
 528
 529                 len = min_t(size_t, ((pages * PAGE_SIZE) - offset), size);
 530
 531                 data.len = len;
 532                 data.address = __sme_page_pa(inpages[i]) + offset;
 533                 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_DATA, &data, &argp->error);
 534                 if (ret)
 535                         goto e_unpin;
 536
 537                 size -= len;
 538                 next_vaddr = vaddr + len;
 539         }
 540
 541 e_unpin:
 542         /* content of memory is updated, mark pages dirty */
 543         for (i = 0; i < npages; i++) {
 544                 set_page_dirty_lock(inpages[i]);
 545                 mark_page_accessed(inpages[i]);
 546         }
 547         /* unlock the user pages */
 548         sev_unpin_memory(kvm, inpages, npages);
 549         return ret;
 550 }
 551
 552 static int sev_es_sync_vmsa(struct vcpu_svm *svm)
 553 {
 554         struct vmcb_save_area *save = &svm->vmcb->save;
 555
 556         /* Check some debug related fields before encrypting the VMSA */
 557         if (svm->vcpu.guest_debug || (save->dr7 & ~DR7_FIXED_1))
 558                 return -EINVAL;
 559
 560         /* Sync registgers */
 561         save->rax = svm->vcpu.arch.regs[VCPU_REGS_RAX];
 562         save->rbx = svm->vcpu.arch.regs[VCPU_REGS_RBX];
 563         save->rcx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
 564         save->rdx = svm->vcpu.arch.regs[VCPU_REGS_RDX];
 565         save->rsp = svm->vcpu.arch.regs[VCPU_REGS_RSP];
 566         save->rbp = svm->vcpu.arch.regs[VCPU_REGS_RBP];
 567         save->rsi = svm->vcpu.arch.regs[VCPU_REGS_RSI];
 568         save->rdi = svm->vcpu.arch.regs[VCPU_REGS_RDI];
 569 #ifdef CONFIG_X86_64
 570         save->r8  = svm->vcpu.arch.regs[VCPU_REGS_R8];
 571         save->r9  = svm->vcpu.arch.regs[VCPU_REGS_R9];
 572         save->r10 = svm->vcpu.arch.regs[VCPU_REGS_R10];
 573         save->r11 = svm->vcpu.arch.regs[VCPU_REGS_R11];
 574         save->r12 = svm->vcpu.arch.regs[VCPU_REGS_R12];
 575         save->r13 = svm->vcpu.arch.regs[VCPU_REGS_R13];
 576         save->r14 = svm->vcpu.arch.regs[VCPU_REGS_R14];
 577         save->r15 = svm->vcpu.arch.regs[VCPU_REGS_R15];
 578 #endif
 579         save->rip = svm->vcpu.arch.regs[VCPU_REGS_RIP];
 580
 581         /* Sync some non-GPR registers before encrypting */
 582         save->xcr0 = svm->vcpu.arch.xcr0;
 583         save->pkru = svm->vcpu.arch.pkru;
 584         save->xss  = svm->vcpu.arch.ia32_xss;
 585         save->dr6  = svm->vcpu.arch.dr6;
 586
 587         /*
 588          * SEV-ES will use a VMSA that is pointed to by the VMCB, not
 589          * the traditional VMSA that is part of the VMCB. Copy the
 590          * traditional VMSA as it has been built so far (in prep
 591          * for LAUNCH_UPDATE_VMSA) to be the initial SEV-ES state.
 592          */
 593         memcpy(svm->vmsa, save, sizeof(*save));
 594
 595         return 0;
 596 }
 597
 598 static int sev_launch_update_vmsa(struct kvm *kvm, struct kvm_sev_cmd *argp)
 599 {
 600         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 601         struct sev_data_launch_update_vmsa vmsa;
 602         struct kvm_vcpu *vcpu;
 603         int i, ret;
 604
 605         if (!sev_es_guest(kvm))
 606                 return -ENOTTY;
 607
 608         vmsa.reserved = 0;
 609
 610         kvm_for_each_vcpu(i, vcpu, kvm) {
 611                 struct vcpu_svm *svm = to_svm(vcpu);
 612
 613                 /* Perform some pre-encryption checks against the VMSA */
 614                 ret = sev_es_sync_vmsa(svm);
 615                 if (ret)
 616                         return ret;
 617
 618                 /*
 619                  * The LAUNCH_UPDATE_VMSA command will perform in-place
 620                  * encryption of the VMSA memory content (i.e it will write
 621                  * the same memory region with the guest's key), so invalidate
 622                  * it first.
 623                  */
 624                 clflush_cache_range(svm->vmsa, PAGE_SIZE);
 625
 626                 vmsa.handle = sev->handle;
 627                 vmsa.address = __sme_pa(svm->vmsa);
 628                 vmsa.len = PAGE_SIZE;
 629                 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_VMSA, &vmsa,
 630                                     &argp->error);
 631                 if (ret)
 632                         return ret;
 633
 634                 svm->vcpu.arch.guest_state_protected = true;
 635         }
 636
 637         return 0;
 638 }
 639
 640 static int sev_launch_measure(struct kvm *kvm, struct kvm_sev_cmd *argp)
 641 {
 642         void __user *measure = (void __user *)(uintptr_t)argp->data;
 643         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 644         struct sev_data_launch_measure data;
 645         struct kvm_sev_launch_measure params;
 646         void __user *p = NULL;
 647         void *blob = NULL;
 648         int ret;
 649
 650         if (!sev_guest(kvm))
 651                 return -ENOTTY;
 652
 653         if (copy_from_user(&params, measure, sizeof(params)))
 654                 return -EFAULT;
 655
 656         memset(&data, 0, sizeof(data));
 657
 658         /* User wants to query the blob length */
 659         if (!params.len)
 660                 goto cmd;
 661
 662         p = (void __user *)(uintptr_t)params.uaddr;
 663         if (p) {
 664                 if (params.len > SEV_FW_BLOB_MAX_SIZE)
 665                         return -EINVAL;
 666
 667                 blob = kmalloc(params.len, GFP_KERNEL_ACCOUNT);
 668                 if (!blob)
 669                         return -ENOMEM;
 670
 671                 data.address = __psp_pa(blob);
 672                 data.len = params.len;
 673         }
 674
 675 cmd:
 676         data.handle = sev->handle;
 677         ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_MEASURE, &data, &argp->error);
 678
 679         /*
 680          * If we query the session length, FW responded with expected data.
 681          */
 682         if (!params.len)
 683                 goto done;
 684
 685         if (ret)
 686                 goto e_free_blob;
 687
 688         if (blob) {
 689                 if (copy_to_user(p, blob, params.len))
 690                         ret = -EFAULT;
 691         }
 692
 693 done:
 694         params.len = data.len;
 695         if (copy_to_user(measure, &params, sizeof(params)))
 696                 ret = -EFAULT;
 697 e_free_blob:
 698         kfree(blob);
 699         return ret;
 700 }
 701
 702 static int sev_launch_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
 703 {
 704         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 705         struct sev_data_launch_finish data;
 706
 707         if (!sev_guest(kvm))
 708                 return -ENOTTY;
 709
 710         data.handle = sev->handle;
 711         return sev_issue_cmd(kvm, SEV_CMD_LAUNCH_FINISH, &data, &argp->error);
 712 }
 713
 714 static int sev_guest_status(struct kvm *kvm, struct kvm_sev_cmd *argp)
 715 {
 716         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 717         struct kvm_sev_guest_status params;
 718         struct sev_data_guest_status data;
 719         int ret;
 720
 721         if (!sev_guest(kvm))
 722                 return -ENOTTY;
 723
 724         memset(&data, 0, sizeof(data));
 725
 726         data.handle = sev->handle;
 727         ret = sev_issue_cmd(kvm, SEV_CMD_GUEST_STATUS, &data, &argp->error);
 728         if (ret)
 729                 return ret;
 730
 731         params.policy = data.policy;
 732         params.state = data.state;
 733         params.handle = data.handle;
 734
 735         if (copy_to_user((void __user *)(uintptr_t)argp->data, &params, sizeof(params)))
 736                 ret = -EFAULT;
 737
 738         return ret;
 739 }
 740
 741 static int __sev_issue_dbg_cmd(struct kvm *kvm, unsigned long src,
 742                                unsigned long dst, int size,
 743                                int *error, bool enc)
 744 {
 745         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 746         struct sev_data_dbg data;
 747
 748         data.reserved = 0;
 749         data.handle = sev->handle;
 750         data.dst_addr = dst;
 751         data.src_addr = src;
 752         data.len = size;
 753
 754         return sev_issue_cmd(kvm,
 755                              enc ? SEV_CMD_DBG_ENCRYPT : SEV_CMD_DBG_DECRYPT,
 756                              &data, error);
 757 }
 758
 759 static int __sev_dbg_decrypt(struct kvm *kvm, unsigned long src_paddr,
 760                              unsigned long dst_paddr, int sz, int *err)
 761 {
 762         int offset;
 763
 764         /*
 765          * Its safe to read more than we are asked, caller should ensure that
 766          * destination has enough space.
 767          */
 768         offset = src_paddr & 15;
 769         src_paddr = round_down(src_paddr, 16);
 770         sz = round_up(sz + offset, 16);
 771
 772         return __sev_issue_dbg_cmd(kvm, src_paddr, dst_paddr, sz, err, false);
 773 }
 774
 775 static int __sev_dbg_decrypt_user(struct kvm *kvm, unsigned long paddr,
 776                                   void __user *dst_uaddr,
 777                                   unsigned long dst_paddr,
 778                                   int size, int *err)
 779 {
 780         struct page *tpage = NULL;
 781         int ret, offset;
 782
 783         /* if inputs are not 16-byte then use intermediate buffer */
 784         if (!IS_ALIGNED(dst_paddr, 16) ||
 785             !IS_ALIGNED(paddr,     16) ||
 786             !IS_ALIGNED(size,      16)) {
 787                 tpage = (void *)alloc_page(GFP_KERNEL);
 788                 if (!tpage)
 789                         return -ENOMEM;
 790
 791                 dst_paddr = __sme_page_pa(tpage);
 792         }
 793
 794         ret = __sev_dbg_decrypt(kvm, paddr, dst_paddr, size, err);
 795         if (ret)
 796                 goto e_free;
 797
 798         if (tpage) {
 799                 offset = paddr & 15;
 800                 if (copy_to_user(dst_uaddr, page_address(tpage) + offset, size))
 801                         ret = -EFAULT;
 802         }
 803
 804 e_free:
 805         if (tpage)
 806                 __free_page(tpage);
 807
 808         return ret;
 809 }
 810
 811 static int __sev_dbg_encrypt_user(struct kvm *kvm, unsigned long paddr,
 812                                   void __user *vaddr,
 813                                   unsigned long dst_paddr,
 814                                   void __user *dst_vaddr,
 815                                   int size, int *error)
 816 {
 817         struct page *src_tpage = NULL;
 818         struct page *dst_tpage = NULL;
 819         int ret, len = size;
 820
 821         /* If source buffer is not aligned then use an intermediate buffer */
 822         if (!IS_ALIGNED((unsigned long)vaddr, 16)) {
 823                 src_tpage = alloc_page(GFP_KERNEL);
 824                 if (!src_tpage)
 825                         return -ENOMEM;
 826
 827                 if (copy_from_user(page_address(src_tpage), vaddr, size)) {
 828                         __free_page(src_tpage);
 829                         return -EFAULT;
 830                 }
 831
 832                 paddr = __sme_page_pa(src_tpage);
 833         }
 834
 835         /*
 836          *  If destination buffer or length is not aligned then do read-modify-write:
 837          *   - decrypt destination in an intermediate buffer
 838          *   - copy the source buffer in an intermediate buffer
 839          *   - use the intermediate buffer as source buffer
 840          */
 841         if (!IS_ALIGNED((unsigned long)dst_vaddr, 16) || !IS_ALIGNED(size, 16)) {
 842                 int dst_offset;
 843
 844                 dst_tpage = alloc_page(GFP_KERNEL);
 845                 if (!dst_tpage) {
 846                         ret = -ENOMEM;
 847                         goto e_free;
 848                 }
 849
 850                 ret = __sev_dbg_decrypt(kvm, dst_paddr,
 851                                         __sme_page_pa(dst_tpage), size, error);
 852                 if (ret)
 853                         goto e_free;
 854
 855                 /*
 856                  *  If source is kernel buffer then use memcpy() otherwise
 857                  *  copy_from_user().
 858                  */
 859                 dst_offset = dst_paddr & 15;
 860
 861                 if (src_tpage)
 862                         memcpy(page_address(dst_tpage) + dst_offset,
 863                                page_address(src_tpage), size);
 864                 else {
 865                         if (copy_from_user(page_address(dst_tpage) + dst_offset,
 866                                            vaddr, size)) {
 867                                 ret = -EFAULT;
 868                                 goto e_free;
 869                         }
 870                 }
 871
 872                 paddr = __sme_page_pa(dst_tpage);
 873                 dst_paddr = round_down(dst_paddr, 16);
 874                 len = round_up(size, 16);
 875         }
 876
 877         ret = __sev_issue_dbg_cmd(kvm, paddr, dst_paddr, len, error, true);
 878
 879 e_free:
 880         if (src_tpage)
 881                 __free_page(src_tpage);
 882         if (dst_tpage)
 883                 __free_page(dst_tpage);
 884         return ret;
 885 }
 886
 887 static int sev_dbg_crypt(struct kvm *kvm, struct kvm_sev_cmd *argp, bool dec)
 888 {
 889         unsigned long vaddr, vaddr_end, next_vaddr;
 890         unsigned long dst_vaddr;
 891         struct page **src_p, **dst_p;
 892         struct kvm_sev_dbg debug;
 893         unsigned long n;
 894         unsigned int size;
 895         int ret;
 896
 897         if (!sev_guest(kvm))
 898                 return -ENOTTY;
 899
 900         if (copy_from_user(&debug, (void __user *)(uintptr_t)argp->data, sizeof(debug)))
 901                 return -EFAULT;
 902
 903         if (!debug.len || debug.src_uaddr + debug.len < debug.src_uaddr)
 904                 return -EINVAL;
 905         if (!debug.dst_uaddr)
 906                 return -EINVAL;
 907
 908         vaddr = debug.src_uaddr;
 909         size = debug.len;
 910         vaddr_end = vaddr + size;
 911         dst_vaddr = debug.dst_uaddr;
 912
 913         for (; vaddr < vaddr_end; vaddr = next_vaddr) {
 914                 int len, s_off, d_off;
 915
 916                 /* lock userspace source and destination page */
 917                 src_p = sev_pin_memory(kvm, vaddr & PAGE_MASK, PAGE_SIZE, &n, 0);
 918                 if (IS_ERR(src_p))
 919                         return PTR_ERR(src_p);
 920
 921                 dst_p = sev_pin_memory(kvm, dst_vaddr & PAGE_MASK, PAGE_SIZE, &n, 1);
 922                 if (IS_ERR(dst_p)) {
 923                         sev_unpin_memory(kvm, src_p, n);
 924                         return PTR_ERR(dst_p);
 925                 }
 926
 927                 /*
 928                  * Flush (on non-coherent CPUs) before DBG_{DE,EN}CRYPT read or modify
 929                  * the pages; flush the destination too so that future accesses do not
 930                  * see stale data.
 931                  */
 932                 sev_clflush_pages(src_p, 1);
 933                 sev_clflush_pages(dst_p, 1);
 934
 935                 /*
 936                  * Since user buffer may not be page aligned, calculate the
 937                  * offset within the page.
 938                  */
 939                 s_off = vaddr & ~PAGE_MASK;
 940                 d_off = dst_vaddr & ~PAGE_MASK;
 941                 len = min_t(size_t, (PAGE_SIZE - s_off), size);
 942
 943                 if (dec)
 944                         ret = __sev_dbg_decrypt_user(kvm,
 945                                                      __sme_page_pa(src_p[0]) + s_off,
 946                                                      (void __user *)dst_vaddr,
 947                                                      __sme_page_pa(dst_p[0]) + d_off,
 948                                                      len, &argp->error);
 949                 else
 950                         ret = __sev_dbg_encrypt_user(kvm,
 951                                                      __sme_page_pa(src_p[0]) + s_off,
 952                                                      (void __user *)vaddr,
 953                                                      __sme_page_pa(dst_p[0]) + d_off,
 954                                                      (void __user *)dst_vaddr,
 955                                                      len, &argp->error);
 956
 957                 sev_unpin_memory(kvm, src_p, n);
 958                 sev_unpin_memory(kvm, dst_p, n);
 959
 960                 if (ret)
 961                         goto err;
 962
 963                 next_vaddr = vaddr + len;
 964                 dst_vaddr = dst_vaddr + len;
 965                 size -= len;
 966         }
 967 err:
 968         return ret;
 969 }
 970
 971 static int sev_launch_secret(struct kvm *kvm, struct kvm_sev_cmd *argp)
 972 {
 973         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 974         struct sev_data_launch_secret data;
 975         struct kvm_sev_launch_secret params;
 976         struct page **pages;
 977         void *blob, *hdr;
 978         unsigned long n, i;
 979         int ret, offset;
 980
 981         if (!sev_guest(kvm))
 982                 return -ENOTTY;
 983
 984         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data, sizeof(params)))
 985                 return -EFAULT;
 986
 987         pages = sev_pin_memory(kvm, params.guest_uaddr, params.guest_len, &n, 1);
 988         if (IS_ERR(pages))
 989                 return PTR_ERR(pages);
 990
 991         /*
 992          * Flush (on non-coherent CPUs) before LAUNCH_SECRET encrypts pages in
 993          * place; the cache may contain the data that was written unencrypted.
 994          */
 995         sev_clflush_pages(pages, n);
 996
 997         /*
 998          * The secret must be copied into contiguous memory region, lets verify
 999          * that userspace memory pages are contiguous before we issue command.
1000          */
1001         if (get_num_contig_pages(0, pages, n) != n) {
1002                 ret = -EINVAL;
1003                 goto e_unpin_memory;
1004         }
1005
1006         memset(&data, 0, sizeof(data));
1007
1008         offset = params.guest_uaddr & (PAGE_SIZE - 1);
1009         data.guest_address = __sme_page_pa(pages[0]) + offset;
1010         data.guest_len = params.guest_len;
1011
1012         blob = psp_copy_user_blob(params.trans_uaddr, params.trans_len);
1013         if (IS_ERR(blob)) {
1014                 ret = PTR_ERR(blob);
1015                 goto e_unpin_memory;
1016         }
1017
1018         data.trans_address = __psp_pa(blob);
1019         data.trans_len = params.trans_len;
1020
1021         hdr = psp_copy_user_blob(params.hdr_uaddr, params.hdr_len);
1022         if (IS_ERR(hdr)) {
1023                 ret = PTR_ERR(hdr);
1024                 goto e_free_blob;
1025         }
1026         data.hdr_address = __psp_pa(hdr);
1027         data.hdr_len = params.hdr_len;
1028
1029         data.handle = sev->handle;
1030         ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_SECRET, &data, &argp->error);
1031
1032         kfree(hdr);
1033
1034 e_free_blob:
1035         kfree(blob);
1036 e_unpin_memory:
1037         /* content of memory is updated, mark pages dirty */
1038         for (i = 0; i < n; i++) {
1039                 set_page_dirty_lock(pages[i]);
1040                 mark_page_accessed(pages[i]);
1041         }
1042         sev_unpin_memory(kvm, pages, n);
1043         return ret;
1044 }
1045
1046 static int sev_get_attestation_report(struct kvm *kvm, struct kvm_sev_cmd *argp)
1047 {
1048         void __user *report = (void __user *)(uintptr_t)argp->data;
1049         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1050         struct sev_data_attestation_report data;
1051         struct kvm_sev_attestation_report params;
1052         void __user *p;
1053         void *blob = NULL;
1054         int ret;
1055
1056         if (!sev_guest(kvm))
1057                 return -ENOTTY;
1058
1059         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data, sizeof(params)))
1060                 return -EFAULT;
1061
1062         memset(&data, 0, sizeof(data));
1063
1064         /* User wants to query the blob length */
1065         if (!params.len)
1066                 goto cmd;
1067
1068         p = (void __user *)(uintptr_t)params.uaddr;
1069         if (p) {
1070                 if (params.len > SEV_FW_BLOB_MAX_SIZE)
1071                         return -EINVAL;
1072
1073                 blob = kmalloc(params.len, GFP_KERNEL_ACCOUNT);
1074                 if (!blob)
1075                         return -ENOMEM;
1076
1077                 data.address = __psp_pa(blob);
1078                 data.len = params.len;
1079                 memcpy(data.mnonce, params.mnonce, sizeof(params.mnonce));
1080         }
1081 cmd:
1082         data.handle = sev->handle;
1083         ret = sev_issue_cmd(kvm, SEV_CMD_ATTESTATION_REPORT, &data, &argp->error);
1084         /*
1085          * If we query the session length, FW responded with expected data.
1086          */
1087         if (!params.len)
1088                 goto done;
1089
1090         if (ret)
1091                 goto e_free_blob;
1092
1093         if (blob) {
1094                 if (copy_to_user(p, blob, params.len))
1095                         ret = -EFAULT;
1096         }
1097
1098 done:
1099         params.len = data.len;
1100         if (copy_to_user(report, &params, sizeof(params)))
1101                 ret = -EFAULT;
1102 e_free_blob:
1103         kfree(blob);
1104         return ret;
1105 }
1106
1107 /* Userspace wants to query session length. */
1108 static int
1109 __sev_send_start_query_session_length(struct kvm *kvm, struct kvm_sev_cmd *argp,
1110                                       struct kvm_sev_send_start *params)
1111 {
1112         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1113         struct sev_data_send_start data;
1114         int ret;
1115
1116         memset(&data, 0, sizeof(data));
1117         data.handle = sev->handle;
1118         ret = sev_issue_cmd(kvm, SEV_CMD_SEND_START, &data, &argp->error);
1119
1120         params->session_len = data.session_len;
1121         if (copy_to_user((void __user *)(uintptr_t)argp->data, params,
1122                                 sizeof(struct kvm_sev_send_start)))
1123                 ret = -EFAULT;
1124
1125         return ret;
1126 }
1127
1128 static int sev_send_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
1129 {
1130         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1131         struct sev_data_send_start data;
1132         struct kvm_sev_send_start params;
1133         void *amd_certs, *session_data;
1134         void *pdh_cert, *plat_certs;
1135         int ret;
1136
1137         if (!sev_guest(kvm))
1138                 return -ENOTTY;
1139
1140         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data,
1141                                 sizeof(struct kvm_sev_send_start)))
1142                 return -EFAULT;
1143
1144         /* if session_len is zero, userspace wants to query the session length */
1145         if (!params.session_len)
1146                 return __sev_send_start_query_session_length(kvm, argp,
1147                                 &params);
1148
1149         /* some sanity checks */
1150         if (!params.pdh_cert_uaddr || !params.pdh_cert_len ||
1151             !params.session_uaddr || params.session_len > SEV_FW_BLOB_MAX_SIZE)
1152                 return -EINVAL;
1153
1154         /* allocate the memory to hold the session data blob */
1155         session_data = kmalloc(params.session_len, GFP_KERNEL_ACCOUNT);
1156         if (!session_data)
1157                 return -ENOMEM;
1158
1159         /* copy the certificate blobs from userspace */
1160         pdh_cert = psp_copy_user_blob(params.pdh_cert_uaddr,
1161                                 params.pdh_cert_len);
1162         if (IS_ERR(pdh_cert)) {
1163                 ret = PTR_ERR(pdh_cert);
1164                 goto e_free_session;
1165         }
1166
1167         plat_certs = psp_copy_user_blob(params.plat_certs_uaddr,
1168                                 params.plat_certs_len);
1169         if (IS_ERR(plat_certs)) {
1170                 ret = PTR_ERR(plat_certs);
1171                 goto e_free_pdh;
1172         }
1173
1174         amd_certs = psp_copy_user_blob(params.amd_certs_uaddr,
1175                                 params.amd_certs_len);
1176         if (IS_ERR(amd_certs)) {
1177                 ret = PTR_ERR(amd_certs);
1178                 goto e_free_plat_cert;
1179         }
1180
1181         /* populate the FW SEND_START field with system physical address */
1182         memset(&data, 0, sizeof(data));
1183         data.pdh_cert_address = __psp_pa(pdh_cert);
1184         data.pdh_cert_len = params.pdh_cert_len;
1185         data.plat_certs_address = __psp_pa(plat_certs);
1186         data.plat_certs_len = params.plat_certs_len;
1187         data.amd_certs_address = __psp_pa(amd_certs);
1188         data.amd_certs_len = params.amd_certs_len;
1189         data.session_address = __psp_pa(session_data);
1190         data.session_len = params.session_len;
1191         data.handle = sev->handle;
1192
1193         ret = sev_issue_cmd(kvm, SEV_CMD_SEND_START, &data, &argp->error);
1194
1195         if (!ret && copy_to_user((void __user *)(uintptr_t)params.session_uaddr,
1196                         session_data, params.session_len)) {
1197                 ret = -EFAULT;
1198                 goto e_free_amd_cert;
1199         }
1200
1201         params.policy = data.policy;
1202         params.session_len = data.session_len;
1203         if (copy_to_user((void __user *)(uintptr_t)argp->data, &params,
1204                                 sizeof(struct kvm_sev_send_start)))
1205                 ret = -EFAULT;
1206
1207 e_free_amd_cert:
1208         kfree(amd_certs);
1209 e_free_plat_cert:
1210         kfree(plat_certs);
1211 e_free_pdh:
1212         kfree(pdh_cert);
1213 e_free_session:
1214         kfree(session_data);
1215         return ret;
1216 }
1217
1218 /* Userspace wants to query either header or trans length. */
1219 static int
1220 __sev_send_update_data_query_lengths(struct kvm *kvm, struct kvm_sev_cmd *argp,
1221                                      struct kvm_sev_send_update_data *params)
1222 {
1223         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1224         struct sev_data_send_update_data data;
1225         int ret;
1226
1227         memset(&data, 0, sizeof(data));
1228         data.handle = sev->handle;
1229         ret = sev_issue_cmd(kvm, SEV_CMD_SEND_UPDATE_DATA, &data, &argp->error);
1230
1231         params->hdr_len = data.hdr_len;
1232         params->trans_len = data.trans_len;
1233
1234         if (copy_to_user((void __user *)(uintptr_t)argp->data, params,
1235                          sizeof(struct kvm_sev_send_update_data)))
1236                 ret = -EFAULT;
1237
1238         return ret;
1239 }
1240
1241 static int sev_send_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
1242 {
1243         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1244         struct sev_data_send_update_data data;
1245         struct kvm_sev_send_update_data params;
1246         void *hdr, *trans_data;
1247         struct page **guest_page;
1248         unsigned long n;
1249         int ret, offset;
1250
1251         if (!sev_guest(kvm))
1252                 return -ENOTTY;
1253
1254         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data,
1255                         sizeof(struct kvm_sev_send_update_data)))
1256                 return -EFAULT;
1257
1258         /* userspace wants to query either header or trans length */
1259         if (!params.trans_len || !params.hdr_len)
1260                 return __sev_send_update_data_query_lengths(kvm, argp, &params);
1261
1262         if (!params.trans_uaddr || !params.guest_uaddr ||
1263             !params.guest_len || !params.hdr_uaddr)
1264                 return -EINVAL;
1265
1266         /* Check if we are crossing the page boundary */
1267         offset = params.guest_uaddr & (PAGE_SIZE - 1);
1268         if ((params.guest_len + offset > PAGE_SIZE))
1269                 return -EINVAL;
1270
1271         /* Pin guest memory */
1272         guest_page = sev_pin_memory(kvm, params.guest_uaddr & PAGE_MASK,
1273                                     PAGE_SIZE, &n, 0);
1274         if (IS_ERR(guest_page))
1275                 return PTR_ERR(guest_page);
1276
1277         /* allocate memory for header and transport buffer */
1278         ret = -ENOMEM;
1279         hdr = kmalloc(params.hdr_len, GFP_KERNEL_ACCOUNT);
1280         if (!hdr)
1281                 goto e_unpin;
1282
1283         trans_data = kmalloc(params.trans_len, GFP_KERNEL_ACCOUNT);
1284         if (!trans_data)
1285                 goto e_free_hdr;
1286
1287         memset(&data, 0, sizeof(data));
1288         data.hdr_address = __psp_pa(hdr);
1289         data.hdr_len = params.hdr_len;
1290         data.trans_address = __psp_pa(trans_data);
1291         data.trans_len = params.trans_len;
1292
1293         /* The SEND_UPDATE_DATA command requires C-bit to be always set. */
1294         data.guest_address = (page_to_pfn(guest_page[0]) << PAGE_SHIFT) + offset;
1295         data.guest_address |= sev_me_mask;
1296         data.guest_len = params.guest_len;
1297         data.handle = sev->handle;
1298
1299         ret = sev_issue_cmd(kvm, SEV_CMD_SEND_UPDATE_DATA, &data, &argp->error);
1300
1301         if (ret)
1302                 goto e_free_trans_data;
1303
1304         /* copy transport buffer to user space */
1305         if (copy_to_user((void __user *)(uintptr_t)params.trans_uaddr,
1306                          trans_data, params.trans_len)) {
1307                 ret = -EFAULT;
1308                 goto e_free_trans_data;
1309         }
1310
1311         /* Copy packet header to userspace. */
1312         if (copy_to_user((void __user *)(uintptr_t)params.hdr_uaddr, hdr,
1313                          params.hdr_len))
1314                 ret = -EFAULT;
1315
1316 e_free_trans_data:
1317         kfree(trans_data);
1318 e_free_hdr:
1319         kfree(hdr);
1320 e_unpin:
1321         sev_unpin_memory(kvm, guest_page, n);
1322
1323         return ret;
1324 }
1325
1326 static int sev_send_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
1327 {
1328         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1329         struct sev_data_send_finish data;
1330
1331         if (!sev_guest(kvm))
1332                 return -ENOTTY;
1333
1334         data.handle = sev->handle;
1335         return sev_issue_cmd(kvm, SEV_CMD_SEND_FINISH, &data, &argp->error);
1336 }
1337
1338 static int sev_send_cancel(struct kvm *kvm, struct kvm_sev_cmd *argp)
1339 {
1340         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1341         struct sev_data_send_cancel data;
1342
1343         if (!sev_guest(kvm))
1344                 return -ENOTTY;
1345
1346         data.handle = sev->handle;
1347         return sev_issue_cmd(kvm, SEV_CMD_SEND_CANCEL, &data, &argp->error);
1348 }
1349
1350 static int sev_receive_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
1351 {
1352         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1353         struct sev_data_receive_start start;
1354         struct kvm_sev_receive_start params;
1355         int *error = &argp->error;
1356         void *session_data;
1357         void *pdh_data;
1358         int ret;
1359
1360         if (!sev_guest(kvm))
1361                 return -ENOTTY;
1362
1363         /* Get parameter from the userspace */
1364         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data,
1365                         sizeof(struct kvm_sev_receive_start)))
1366                 return -EFAULT;
1367
1368         /* some sanity checks */
1369         if (!params.pdh_uaddr || !params.pdh_len ||
1370             !params.session_uaddr || !params.session_len)
1371                 return -EINVAL;
1372
1373         pdh_data = psp_copy_user_blob(params.pdh_uaddr, params.pdh_len);
1374         if (IS_ERR(pdh_data))
1375                 return PTR_ERR(pdh_data);
1376
1377         session_data = psp_copy_user_blob(params.session_uaddr,
1378                         params.session_len);
1379         if (IS_ERR(session_data)) {
1380                 ret = PTR_ERR(session_data);
1381                 goto e_free_pdh;
1382         }
1383
1384         memset(&start, 0, sizeof(start));
1385         start.handle = params.handle;
1386         start.policy = params.policy;
1387         start.pdh_cert_address = __psp_pa(pdh_data);
1388         start.pdh_cert_len = params.pdh_len;
1389         start.session_address = __psp_pa(session_data);
1390         start.session_len = params.session_len;
1391
1392         /* create memory encryption context */
1393         ret = __sev_issue_cmd(argp->sev_fd, SEV_CMD_RECEIVE_START, &start,
1394                                 error);
1395         if (ret)
1396                 goto e_free_session;
1397
1398         /* Bind ASID to this guest */
1399         ret = sev_bind_asid(kvm, start.handle, error);
1400         if (ret)
1401                 goto e_free_session;
1402
1403         params.handle = start.handle;
1404         if (copy_to_user((void __user *)(uintptr_t)argp->data,
1405                          &params, sizeof(struct kvm_sev_receive_start))) {
1406                 ret = -EFAULT;
1407                 sev_unbind_asid(kvm, start.handle);
1408                 goto e_free_session;
1409         }
1410
1411         sev->handle = start.handle;
1412         sev->fd = argp->sev_fd;
1413
1414 e_free_session:
1415         kfree(session_data);
1416 e_free_pdh:
1417         kfree(pdh_data);
1418
1419         return ret;
1420 }
1421
1422 static int sev_receive_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
1423 {
1424         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1425         struct kvm_sev_receive_update_data params;
1426         struct sev_data_receive_update_data data;
1427         void *hdr = NULL, *trans = NULL;
1428         struct page **guest_page;
1429         unsigned long n;
1430         int ret, offset;
1431
1432         if (!sev_guest(kvm))
1433                 return -EINVAL;
1434
1435         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data,
1436                         sizeof(struct kvm_sev_receive_update_data)))
1437                 return -EFAULT;
1438
1439         if (!params.hdr_uaddr || !params.hdr_len ||
1440             !params.guest_uaddr || !params.guest_len ||
1441             !params.trans_uaddr || !params.trans_len)
1442                 return -EINVAL;
1443
1444         /* Check if we are crossing the page boundary */
1445         offset = params.guest_uaddr & (PAGE_SIZE - 1);
1446         if ((params.guest_len + offset > PAGE_SIZE))
1447                 return -EINVAL;
1448
1449         hdr = psp_copy_user_blob(params.hdr_uaddr, params.hdr_len);
1450         if (IS_ERR(hdr))
1451                 return PTR_ERR(hdr);
1452
1453         trans = psp_copy_user_blob(params.trans_uaddr, params.trans_len);
1454         if (IS_ERR(trans)) {
1455                 ret = PTR_ERR(trans);
1456                 goto e_free_hdr;
1457         }
1458
1459         memset(&data, 0, sizeof(data));
1460         data.hdr_address = __psp_pa(hdr);
1461         data.hdr_len = params.hdr_len;
1462         data.trans_address = __psp_pa(trans);
1463         data.trans_len = params.trans_len;
1464
1465         /* Pin guest memory */
1466         guest_page = sev_pin_memory(kvm, params.guest_uaddr & PAGE_MASK,
1467                                     PAGE_SIZE, &n, 0);
1468         if (IS_ERR(guest_page)) {
1469                 ret = PTR_ERR(guest_page);
1470                 goto e_free_trans;
1471         }
1472
1473         /* The RECEIVE_UPDATE_DATA command requires C-bit to be always set. */
1474         data.guest_address = (page_to_pfn(guest_page[0]) << PAGE_SHIFT) + offset;
1475         data.guest_address |= sev_me_mask;
1476         data.guest_len = params.guest_len;
1477         data.handle = sev->handle;
1478
1479         ret = sev_issue_cmd(kvm, SEV_CMD_RECEIVE_UPDATE_DATA, &data,
1480                                 &argp->error);
1481
1482         sev_unpin_memory(kvm, guest_page, n);
1483
1484 e_free_trans:
1485         kfree(trans);
1486 e_free_hdr:
1487         kfree(hdr);
1488
1489         return ret;
1490 }
1491
1492 static int sev_receive_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
1493 {
1494         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1495         struct sev_data_receive_finish data;
1496
1497         if (!sev_guest(kvm))
1498                 return -ENOTTY;
1499
1500         data.handle = sev->handle;
1501         return sev_issue_cmd(kvm, SEV_CMD_RECEIVE_FINISH, &data, &argp->error);
1502 }
1503
1504 int svm_mem_enc_op(struct kvm *kvm, void __user *argp)
1505 {
1506         struct kvm_sev_cmd sev_cmd;
1507         int r;
1508
1509         if (!sev_enabled)
1510                 return -ENOTTY;
1511
1512         if (!argp)
1513                 return 0;
1514
1515         if (copy_from_user(&sev_cmd, argp, sizeof(struct kvm_sev_cmd)))
1516                 return -EFAULT;
1517
1518         mutex_lock(&kvm->lock);
1519
1520         /* enc_context_owner handles all memory enc operations */
1521         if (is_mirroring_enc_context(kvm)) {
1522                 r = -EINVAL;
1523                 goto out;
1524         }
1525
1526         switch (sev_cmd.id) {
1527         case KVM_SEV_ES_INIT:
1528                 if (!sev_es_enabled) {
1529                         r = -ENOTTY;
1530                         goto out;
1531                 }
1532                 fallthrough;
1533         case KVM_SEV_INIT:
1534                 r = sev_guest_init(kvm, &sev_cmd);
1535                 break;
1536         case KVM_SEV_LAUNCH_START:
1537                 r = sev_launch_start(kvm, &sev_cmd);
1538                 break;
1539         case KVM_SEV_LAUNCH_UPDATE_DATA:
1540                 r = sev_launch_update_data(kvm, &sev_cmd);
1541                 break;
1542         case KVM_SEV_LAUNCH_UPDATE_VMSA:
1543                 r = sev_launch_update_vmsa(kvm, &sev_cmd);
1544                 break;
1545         case KVM_SEV_LAUNCH_MEASURE:
1546                 r = sev_launch_measure(kvm, &sev_cmd);
1547                 break;
1548         case KVM_SEV_LAUNCH_FINISH:
1549                 r = sev_launch_finish(kvm, &sev_cmd);
1550                 break;
1551         case KVM_SEV_GUEST_STATUS:
1552                 r = sev_guest_status(kvm, &sev_cmd);
1553                 break;
1554         case KVM_SEV_DBG_DECRYPT:
1555                 r = sev_dbg_crypt(kvm, &sev_cmd, true);
1556                 break;
1557         case KVM_SEV_DBG_ENCRYPT:
1558                 r = sev_dbg_crypt(kvm, &sev_cmd, false);
1559                 break;
1560         case KVM_SEV_LAUNCH_SECRET:
1561                 r = sev_launch_secret(kvm, &sev_cmd);
1562                 break;
1563         case KVM_SEV_GET_ATTESTATION_REPORT:
1564                 r = sev_get_attestation_report(kvm, &sev_cmd);
1565                 break;
1566         case KVM_SEV_SEND_START:
1567                 r = sev_send_start(kvm, &sev_cmd);
1568                 break;
1569         case KVM_SEV_SEND_UPDATE_DATA:
1570                 r = sev_send_update_data(kvm, &sev_cmd);
1571                 break;
1572         case KVM_SEV_SEND_FINISH:
1573                 r = sev_send_finish(kvm, &sev_cmd);
1574                 break;
1575         case KVM_SEV_SEND_CANCEL:
1576                 r = sev_send_cancel(kvm, &sev_cmd);
1577                 break;
1578         case KVM_SEV_RECEIVE_START:
1579                 r = sev_receive_start(kvm, &sev_cmd);
1580                 break;
1581         case KVM_SEV_RECEIVE_UPDATE_DATA:
1582                 r = sev_receive_update_data(kvm, &sev_cmd);
1583                 break;
1584         case KVM_SEV_RECEIVE_FINISH:
1585                 r = sev_receive_finish(kvm, &sev_cmd);
1586                 break;
1587         default:
1588                 r = -EINVAL;
1589                 goto out;
1590         }
1591
1592         if (copy_to_user(argp, &sev_cmd, sizeof(struct kvm_sev_cmd)))
1593                 r = -EFAULT;
1594
1595 out:
1596         mutex_unlock(&kvm->lock);
1597         return r;
1598 }
1599
1600 int svm_register_enc_region(struct kvm *kvm,
1601                             struct kvm_enc_region *range)
1602 {
1603         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1604         struct enc_region *region;
1605         int ret = 0;
1606
1607         if (!sev_guest(kvm))
1608                 return -ENOTTY;
1609
1610         /* If kvm is mirroring encryption context it isn't responsible for it */
1611         if (is_mirroring_enc_context(kvm))
1612                 return -EINVAL;
1613
1614         if (range->addr > ULONG_MAX || range->size > ULONG_MAX)
1615                 return -EINVAL;
1616
1617         region = kzalloc(sizeof(*region), GFP_KERNEL_ACCOUNT);
1618         if (!region)
1619                 return -ENOMEM;
1620
1621         mutex_lock(&kvm->lock);
1622         region->pages = sev_pin_memory(kvm, range->addr, range->size, &region->npages, 1);
1623         if (IS_ERR(region->pages)) {
1624                 ret = PTR_ERR(region->pages);
1625                 mutex_unlock(&kvm->lock);
1626                 goto e_free;
1627         }
1628
1629         region->uaddr = range->addr;
1630         region->size = range->size;
1631
1632         list_add_tail(&region->list, &sev->regions_list);
1633         mutex_unlock(&kvm->lock);
1634
1635         /*
1636          * The guest may change the memory encryption attribute from C=0 -> C=1
1637          * or vice versa for this memory range. Lets make sure caches are
1638          * flushed to ensure that guest data gets written into memory with
1639          * correct C-bit.
1640          */
1641         sev_clflush_pages(region->pages, region->npages);
1642
1643         return ret;
1644
1645 e_free:
1646         kfree(region);
1647         return ret;
1648 }
1649
1650 static struct enc_region *
1651 find_enc_region(struct kvm *kvm, struct kvm_enc_region *range)
1652 {
1653         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1654         struct list_head *head = &sev->regions_list;
1655         struct enc_region *i;
1656
1657         list_for_each_entry(i, head, list) {
1658                 if (i->uaddr == range->addr &&
1659                     i->size == range->size)
1660                         return i;
1661         }
1662
1663         return NULL;
1664 }
1665
1666 static void __unregister_enc_region_locked(struct kvm *kvm,
1667                                            struct enc_region *region)
1668 {
1669         sev_unpin_memory(kvm, region->pages, region->npages);
1670         list_del(&region->list);
1671         kfree(region);
1672 }
1673
1674 int svm_unregister_enc_region(struct kvm *kvm,
1675                               struct kvm_enc_region *range)
1676 {
1677         struct enc_region *region;
1678         int ret;
1679
1680         /* If kvm is mirroring encryption context it isn't responsible for it */
1681         if (is_mirroring_enc_context(kvm))
1682                 return -EINVAL;
1683
1684         mutex_lock(&kvm->lock);
1685
1686         if (!sev_guest(kvm)) {
1687                 ret = -ENOTTY;
1688                 goto failed;
1689         }
1690
1691         region = find_enc_region(kvm, range);
1692         if (!region) {
1693                 ret = -EINVAL;
1694                 goto failed;
1695         }
1696
1697         /*
1698          * Ensure that all guest tagged cache entries are flushed before
1699          * releasing the pages back to the system for use. CLFLUSH will
1700          * not do this, so issue a WBINVD.
1701          */
1702         wbinvd_on_all_cpus();
1703
1704         __unregister_enc_region_locked(kvm, region);
1705
1706         mutex_unlock(&kvm->lock);
1707         return 0;
1708
1709 failed:
1710         mutex_unlock(&kvm->lock);
1711         return ret;
1712 }
1713
1714 int svm_vm_copy_asid_from(struct kvm *kvm, unsigned int source_fd)
1715 {
1716         struct file *source_kvm_file;
1717         struct kvm *source_kvm;
1718         struct kvm_sev_info *mirror_sev;
1719         unsigned int asid;
1720         int ret;
1721
1722         source_kvm_file = fget(source_fd);
1723         if (!file_is_kvm(source_kvm_file)) {
1724                 ret = -EBADF;
1725                 goto e_source_put;
1726         }
1727
1728         source_kvm = source_kvm_file->private_data;
1729         mutex_lock(&source_kvm->lock);
1730
1731         if (!sev_guest(source_kvm)) {
1732                 ret = -EINVAL;
1733                 goto e_source_unlock;
1734         }
1735
1736         /* Mirrors of mirrors should work, but let's not get silly */
1737         if (is_mirroring_enc_context(source_kvm) || source_kvm == kvm) {
1738                 ret = -EINVAL;
1739                 goto e_source_unlock;
1740         }
1741
1742         asid = to_kvm_svm(source_kvm)->sev_info.asid;
1743
1744         /*
1745          * The mirror kvm holds an enc_context_owner ref so its asid can't
1746          * disappear until we're done with it
1747          */
1748         kvm_get_kvm(source_kvm);
1749
1750         fput(source_kvm_file);
1751         mutex_unlock(&source_kvm->lock);
1752         mutex_lock(&kvm->lock);
1753
1754         if (sev_guest(kvm)) {
1755                 ret = -EINVAL;
1756                 goto e_mirror_unlock;
1757         }
1758
1759         /* Set enc_context_owner and copy its encryption context over */
1760         mirror_sev = &to_kvm_svm(kvm)->sev_info;
1761         mirror_sev->enc_context_owner = source_kvm;
1762         mirror_sev->asid = asid;
1763         mirror_sev->active = true;
1764
1765         mutex_unlock(&kvm->lock);
1766         return 0;
1767
1768 e_mirror_unlock:
1769         mutex_unlock(&kvm->lock);
1770         kvm_put_kvm(source_kvm);
1771         return ret;
1772 e_source_unlock:
1773         mutex_unlock(&source_kvm->lock);
1774 e_source_put:
1775         if (source_kvm_file)
1776                 fput(source_kvm_file);
1777         return ret;
1778 }
1779
1780 void sev_vm_destroy(struct kvm *kvm)
1781 {
1782         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1783         struct list_head *head = &sev->regions_list;
1784         struct list_head *pos, *q;
1785
1786         if (!sev_guest(kvm))
1787                 return;
1788
1789         /* If this is a mirror_kvm release the enc_context_owner and skip sev cleanup */
1790         if (is_mirroring_enc_context(kvm)) {
1791                 kvm_put_kvm(sev->enc_context_owner);
1792                 return;
1793         }
1794
1795         mutex_lock(&kvm->lock);
1796
1797         /*
1798          * Ensure that all guest tagged cache entries are flushed before
1799          * releasing the pages back to the system for use. CLFLUSH will
1800          * not do this, so issue a WBINVD.
1801          */
1802         wbinvd_on_all_cpus();
1803
1804         /*
1805          * if userspace was terminated before unregistering the memory regions
1806          * then lets unpin all the registered memory.
1807          */
1808         if (!list_empty(head)) {
1809                 list_for_each_safe(pos, q, head) {
1810                         __unregister_enc_region_locked(kvm,
1811                                 list_entry(pos, struct enc_region, list));
1812                         cond_resched();
1813                 }
1814         }
1815
1816         mutex_unlock(&kvm->lock);
1817
1818         sev_unbind_asid(kvm, sev->handle);
1819         sev_asid_free(sev);
1820 }
1821
1822 void __init sev_set_cpu_caps(void)
1823 {
1824         if (!sev_enabled)
1825                 kvm_cpu_cap_clear(X86_FEATURE_SEV);
1826         if (!sev_es_enabled)
1827                 kvm_cpu_cap_clear(X86_FEATURE_SEV_ES);
1828 }
1829
1830 void __init sev_hardware_setup(void)
1831 {
1832 #ifdef CONFIG_KVM_AMD_SEV
1833         unsigned int eax, ebx, ecx, edx, sev_asid_count, sev_es_asid_count;
1834         bool sev_es_supported = false;
1835         bool sev_supported = false;
1836
1837         if (!sev_enabled || !npt_enabled)
1838                 goto out;
1839
1840         /* Does the CPU support SEV? */
1841         if (!boot_cpu_has(X86_FEATURE_SEV))
1842                 goto out;
1843
1844         /* Retrieve SEV CPUID information */
1845         cpuid(0x8000001f, &eax, &ebx, &ecx, &edx);
1846
1847         /* Set encryption bit location for SEV-ES guests */
1848         sev_enc_bit = ebx & 0x3f;
1849
1850         /* Maximum number of encrypted guests supported simultaneously */
1851         max_sev_asid = ecx;
1852         if (!max_sev_asid)
1853                 goto out;
1854
1855         /* Minimum ASID value that should be used for SEV guest */
1856         min_sev_asid = edx;
1857         sev_me_mask = 1UL << (ebx & 0x3f);
1858
1859         /*
1860          * Initialize SEV ASID bitmaps. Allocate space for ASID 0 in the bitmap,
1861          * even though it's never used, so that the bitmap is indexed by the
1862          * actual ASID.
1863          */
1864         nr_asids = max_sev_asid + 1;
1865         sev_asid_bitmap = bitmap_zalloc(nr_asids, GFP_KERNEL);
1866         if (!sev_asid_bitmap)
1867                 goto out;
1868
1869         sev_reclaim_asid_bitmap = bitmap_zalloc(nr_asids, GFP_KERNEL);
1870         if (!sev_reclaim_asid_bitmap) {
1871                 bitmap_free(sev_asid_bitmap);
1872                 sev_asid_bitmap = NULL;
1873                 goto out;
1874         }
1875
1876         sev_asid_count = max_sev_asid - min_sev_asid + 1;
1877         if (misc_cg_set_capacity(MISC_CG_RES_SEV, sev_asid_count))
1878                 goto out;
1879
1880         pr_info("SEV supported: %u ASIDs\n", sev_asid_count);
1881         sev_supported = true;
1882
1883         /* SEV-ES support requested? */
1884         if (!sev_es_enabled)
1885                 goto out;
1886
1887         /* Does the CPU support SEV-ES? */
1888         if (!boot_cpu_has(X86_FEATURE_SEV_ES))
1889                 goto out;
1890
1891         /* Has the system been allocated ASIDs for SEV-ES? */
1892         if (min_sev_asid == 1)
1893                 goto out;
1894
1895         sev_es_asid_count = min_sev_asid - 1;
1896         if (misc_cg_set_capacity(MISC_CG_RES_SEV_ES, sev_es_asid_count))
1897                 goto out;
1898
1899         pr_info("SEV-ES supported: %u ASIDs\n", sev_es_asid_count);
1900         sev_es_supported = true;
1901
1902 out:
1903         sev_enabled = sev_supported;
1904         sev_es_enabled = sev_es_supported;
1905 #endif
1906 }
1907
1908 void sev_hardware_teardown(void)
1909 {
1910         if (!sev_enabled)
1911                 return;
1912
1913         /* No need to take sev_bitmap_lock, all VMs have been destroyed. */
1914         sev_flush_asids(1, max_sev_asid);
1915
1916         bitmap_free(sev_asid_bitmap);
1917         bitmap_free(sev_reclaim_asid_bitmap);
1918
1919         misc_cg_set_capacity(MISC_CG_RES_SEV, 0);
1920         misc_cg_set_capacity(MISC_CG_RES_SEV_ES, 0);
1921 }
1922
1923 int sev_cpu_init(struct svm_cpu_data *sd)
1924 {
1925         if (!sev_enabled)
1926                 return 0;
1927
1928         sd->sev_vmcbs = kcalloc(nr_asids, sizeof(void *), GFP_KERNEL);
1929         if (!sd->sev_vmcbs)
1930                 return -ENOMEM;
1931
1932         return 0;
1933 }
1934
1935 /*
1936  * Pages used by hardware to hold guest encrypted state must be flushed before
1937  * returning them to the system.
1938  */
1939 static void sev_flush_guest_memory(struct vcpu_svm *svm, void *va,
1940                                    unsigned long len)
1941 {
1942         /*
1943          * If hardware enforced cache coherency for encrypted mappings of the
1944          * same physical page is supported, nothing to do.
1945          */
1946         if (boot_cpu_has(X86_FEATURE_SME_COHERENT))
1947                 return;
1948
1949         /*
1950          * If the VM Page Flush MSR is supported, use it to flush the page
1951          * (using the page virtual address and the guest ASID).
1952          */
1953         if (boot_cpu_has(X86_FEATURE_VM_PAGE_FLUSH)) {
1954                 struct kvm_sev_info *sev;
1955                 unsigned long va_start;
1956                 u64 start, stop;
1957
1958                 /* Align start and stop to page boundaries. */
1959                 va_start = (unsigned long)va;
1960                 start = (u64)va_start & PAGE_MASK;
1961                 stop = PAGE_ALIGN((u64)va_start + len);
1962
1963                 if (start < stop) {
1964                         sev = &to_kvm_svm(svm->vcpu.kvm)->sev_info;
1965
1966                         while (start < stop) {
1967                                 wrmsrl(MSR_AMD64_VM_PAGE_FLUSH,
1968                                        start | sev->asid);
1969
1970                                 start += PAGE_SIZE;
1971                         }
1972
1973                         return;
1974                 }
1975
1976                 WARN(1, "Address overflow, using WBINVD\n");
1977         }
1978
1979         /*
1980          * Hardware should always have one of the above features,
1981          * but if not, use WBINVD and issue a warning.
1982          */
1983         WARN_ONCE(1, "Using WBINVD to flush guest memory\n");
1984         wbinvd_on_all_cpus();
1985 }
1986
1987 void sev_free_vcpu(struct kvm_vcpu *vcpu)
1988 {
1989         struct vcpu_svm *svm;
1990
1991         if (!sev_es_guest(vcpu->kvm))
1992                 return;
1993
1994         svm = to_svm(vcpu);
1995
1996         if (vcpu->arch.guest_state_protected)
1997                 sev_flush_guest_memory(svm, svm->vmsa, PAGE_SIZE);
1998         __free_page(virt_to_page(svm->vmsa));
1999
2000         if (svm->ghcb_sa_free)
2001                 kfree(svm->ghcb_sa);
2002 }
2003
2004 static void dump_ghcb(struct vcpu_svm *svm)
2005 {
2006         struct ghcb *ghcb = svm->ghcb;
2007         unsigned int nbits;
2008
2009         /* Re-use the dump_invalid_vmcb module parameter */
2010         if (!dump_invalid_vmcb) {
2011                 pr_warn_ratelimited("set kvm_amd.dump_invalid_vmcb=1 to dump internal KVM state.\n");
2012                 return;
2013         }
2014
2015         nbits = sizeof(ghcb->save.valid_bitmap) * 8;
2016
2017         pr_err("GHCB (GPA=%016llx):\n", svm->vmcb->control.ghcb_gpa);
2018         pr_err("%-20s%016llx is_valid: %u\n", "sw_exit_code",
2019                ghcb->save.sw_exit_code, ghcb_sw_exit_code_is_valid(ghcb));
2020         pr_err("%-20s%016llx is_valid: %u\n", "sw_exit_info_1",
2021                ghcb->save.sw_exit_info_1, ghcb_sw_exit_info_1_is_valid(ghcb));
2022         pr_err("%-20s%016llx is_valid: %u\n", "sw_exit_info_2",
2023                ghcb->save.sw_exit_info_2, ghcb_sw_exit_info_2_is_valid(ghcb));
2024         pr_err("%-20s%016llx is_valid: %u\n", "sw_scratch",
2025                ghcb->save.sw_scratch, ghcb_sw_scratch_is_valid(ghcb));
2026         pr_err("%-20s%*pb\n", "valid_bitmap", nbits, ghcb->save.valid_bitmap);
2027 }
2028
2029 static void sev_es_sync_to_ghcb(struct vcpu_svm *svm)
2030 {
2031         struct kvm_vcpu *vcpu = &svm->vcpu;
2032         struct ghcb *ghcb = svm->ghcb;
2033
2034         /*
2035          * The GHCB protocol so far allows for the following data
2036          * to be returned:
2037          *   GPRs RAX, RBX, RCX, RDX
2038          *
2039          * Copy their values, even if they may not have been written during the
2040          * VM-Exit.  It's the guest's responsibility to not consume random data.
2041          */
2042         ghcb_set_rax(ghcb, vcpu->arch.regs[VCPU_REGS_RAX]);
2043         ghcb_set_rbx(ghcb, vcpu->arch.regs[VCPU_REGS_RBX]);
2044         ghcb_set_rcx(ghcb, vcpu->arch.regs[VCPU_REGS_RCX]);
2045         ghcb_set_rdx(ghcb, vcpu->arch.regs[VCPU_REGS_RDX]);
2046 }
2047
2048 static void sev_es_sync_from_ghcb(struct vcpu_svm *svm)
2049 {
2050         struct vmcb_control_area *control = &svm->vmcb->control;
2051         struct kvm_vcpu *vcpu = &svm->vcpu;
2052         struct ghcb *ghcb = svm->ghcb;
2053         u64 exit_code;
2054
2055         /*
2056          * The GHCB protocol so far allows for the following data
2057          * to be supplied:
2058          *   GPRs RAX, RBX, RCX, RDX
2059          *   XCR0
2060          *   CPL
2061          *
2062          * VMMCALL allows the guest to provide extra registers. KVM also
2063          * expects RSI for hypercalls, so include that, too.
2064          *
2065          * Copy their values to the appropriate location if supplied.
2066          */
2067         memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
2068
2069         vcpu->arch.regs[VCPU_REGS_RAX] = ghcb_get_rax_if_valid(ghcb);
2070         vcpu->arch.regs[VCPU_REGS_RBX] = ghcb_get_rbx_if_valid(ghcb);
2071         vcpu->arch.regs[VCPU_REGS_RCX] = ghcb_get_rcx_if_valid(ghcb);
2072         vcpu->arch.regs[VCPU_REGS_RDX] = ghcb_get_rdx_if_valid(ghcb);
2073         vcpu->arch.regs[VCPU_REGS_RSI] = ghcb_get_rsi_if_valid(ghcb);
2074
2075         svm->vmcb->save.cpl = ghcb_get_cpl_if_valid(ghcb);
2076
2077         if (ghcb_xcr0_is_valid(ghcb)) {
2078                 vcpu->arch.xcr0 = ghcb_get_xcr0(ghcb);
2079                 kvm_update_cpuid_runtime(vcpu);
2080         }
2081
2082         /* Copy the GHCB exit information into the VMCB fields */
2083         exit_code = ghcb_get_sw_exit_code(ghcb);
2084         control->exit_code = lower_32_bits(exit_code);
2085         control->exit_code_hi = upper_32_bits(exit_code);
2086         control->exit_info_1 = ghcb_get_sw_exit_info_1(ghcb);
2087         control->exit_info_2 = ghcb_get_sw_exit_info_2(ghcb);
2088
2089         /* Clear the valid entries fields */
2090         memset(ghcb->save.valid_bitmap, 0, sizeof(ghcb->save.valid_bitmap));
2091 }
2092
2093 static int sev_es_validate_vmgexit(struct vcpu_svm *svm)
2094 {
2095         struct kvm_vcpu *vcpu;
2096         struct ghcb *ghcb;
2097         u64 exit_code = 0;
2098
2099         ghcb = svm->ghcb;
2100
2101         /* Only GHCB Usage code 0 is supported */
2102         if (ghcb->ghcb_usage)
2103                 goto vmgexit_err;
2104
2105         /*
2106          * Retrieve the exit code now even though is may not be marked valid
2107          * as it could help with debugging.
2108          */
2109         exit_code = ghcb_get_sw_exit_code(ghcb);
2110
2111         if (!ghcb_sw_exit_code_is_valid(ghcb) ||
2112             !ghcb_sw_exit_info_1_is_valid(ghcb) ||
2113             !ghcb_sw_exit_info_2_is_valid(ghcb))
2114                 goto vmgexit_err;
2115
2116         switch (ghcb_get_sw_exit_code(ghcb)) {
2117         case SVM_EXIT_READ_DR7:
2118                 break;
2119         case SVM_EXIT_WRITE_DR7:
2120                 if (!ghcb_rax_is_valid(ghcb))
2121                         goto vmgexit_err;
2122                 break;
2123         case SVM_EXIT_RDTSC:
2124                 break;
2125         case SVM_EXIT_RDPMC:
2126                 if (!ghcb_rcx_is_valid(ghcb))
2127                         goto vmgexit_err;
2128                 break;
2129         case SVM_EXIT_CPUID:
2130                 if (!ghcb_rax_is_valid(ghcb) ||
2131                     !ghcb_rcx_is_valid(ghcb))
2132                         goto vmgexit_err;
2133                 if (ghcb_get_rax(ghcb) == 0xd)
2134                         if (!ghcb_xcr0_is_valid(ghcb))
2135                                 goto vmgexit_err;
2136                 break;
2137         case SVM_EXIT_INVD:
2138                 break;
2139         case SVM_EXIT_IOIO:
2140                 if (ghcb_get_sw_exit_info_1(ghcb) & SVM_IOIO_STR_MASK) {
2141                         if (!ghcb_sw_scratch_is_valid(ghcb))
2142                                 goto vmgexit_err;
2143                 } else {
2144                         if (!(ghcb_get_sw_exit_info_1(ghcb) & SVM_IOIO_TYPE_MASK))
2145                                 if (!ghcb_rax_is_valid(ghcb))
2146                                         goto vmgexit_err;
2147                 }
2148                 break;
2149         case SVM_EXIT_MSR:
2150                 if (!ghcb_rcx_is_valid(ghcb))
2151                         goto vmgexit_err;
2152                 if (ghcb_get_sw_exit_info_1(ghcb)) {
2153                         if (!ghcb_rax_is_valid(ghcb) ||
2154                             !ghcb_rdx_is_valid(ghcb))
2155                                 goto vmgexit_err;
2156                 }
2157                 break;
2158         case SVM_EXIT_VMMCALL:
2159                 if (!ghcb_rax_is_valid(ghcb) ||
2160                     !ghcb_cpl_is_valid(ghcb))
2161                         goto vmgexit_err;
2162                 break;
2163         case SVM_EXIT_RDTSCP:
2164                 break;
2165         case SVM_EXIT_WBINVD:
2166                 break;
2167         case SVM_EXIT_MONITOR:
2168                 if (!ghcb_rax_is_valid(ghcb) ||
2169                     !ghcb_rcx_is_valid(ghcb) ||
2170                     !ghcb_rdx_is_valid(ghcb))
2171                         goto vmgexit_err;
2172                 break;
2173         case SVM_EXIT_MWAIT:
2174                 if (!ghcb_rax_is_valid(ghcb) ||
2175                     !ghcb_rcx_is_valid(ghcb))
2176                         goto vmgexit_err;
2177                 break;
2178         case SVM_VMGEXIT_MMIO_READ:
2179         case SVM_VMGEXIT_MMIO_WRITE:
2180                 if (!ghcb_sw_scratch_is_valid(ghcb))
2181                         goto vmgexit_err;
2182                 break;
2183         case SVM_VMGEXIT_NMI_COMPLETE:
2184         case SVM_VMGEXIT_AP_HLT_LOOP:
2185         case SVM_VMGEXIT_AP_JUMP_TABLE:
2186         case SVM_VMGEXIT_UNSUPPORTED_EVENT:
2187                 break;
2188         default:
2189                 goto vmgexit_err;
2190         }
2191
2192         return 0;
2193
2194 vmgexit_err:
2195         vcpu = &svm->vcpu;
2196
2197         if (ghcb->ghcb_usage) {
2198                 vcpu_unimpl(vcpu, "vmgexit: ghcb usage %#x is not valid\n",
2199                             ghcb->ghcb_usage);
2200         } else {
2201                 vcpu_unimpl(vcpu, "vmgexit: exit reason %#llx is not valid\n",
2202                             exit_code);
2203                 dump_ghcb(svm);
2204         }
2205
2206         vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2207         vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON;
2208         vcpu->run->internal.ndata = 2;
2209         vcpu->run->internal.data[0] = exit_code;
2210         vcpu->run->internal.data[1] = vcpu->arch.last_vmentry_cpu;
2211
2212         return -EINVAL;
2213 }
2214
2215 void sev_es_unmap_ghcb(struct vcpu_svm *svm)
2216 {
2217         if (!svm->ghcb)
2218                 return;
2219
2220         if (svm->ghcb_sa_free) {
2221                 /*
2222                  * The scratch area lives outside the GHCB, so there is a
2223                  * buffer that, depending on the operation performed, may
2224                  * need to be synced, then freed.
2225                  */
2226                 if (svm->ghcb_sa_sync) {
2227                         kvm_write_guest(svm->vcpu.kvm,
2228                                         ghcb_get_sw_scratch(svm->ghcb),
2229                                         svm->ghcb_sa, svm->ghcb_sa_len);
2230                         svm->ghcb_sa_sync = false;
2231                 }
2232
2233                 kfree(svm->ghcb_sa);
2234                 svm->ghcb_sa = NULL;
2235                 svm->ghcb_sa_free = false;
2236         }
2237
2238         trace_kvm_vmgexit_exit(svm->vcpu.vcpu_id, svm->ghcb);
2239
2240         sev_es_sync_to_ghcb(svm);
2241
2242         kvm_vcpu_unmap(&svm->vcpu, &svm->ghcb_map, true);
2243         svm->ghcb = NULL;
2244 }
2245
2246 void pre_sev_run(struct vcpu_svm *svm, int cpu)
2247 {
2248         struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
2249         int asid = sev_get_asid(svm->vcpu.kvm);
2250
2251         /* Assign the asid allocated with this SEV guest */
2252         svm->asid = asid;
2253
2254         /*
2255          * Flush guest TLB:
2256          *
2257          * 1) when different VMCB for the same ASID is to be run on the same host CPU.
2258          * 2) or this VMCB was executed on different host CPU in previous VMRUNs.
2259          */
2260         if (sd->sev_vmcbs[asid] == svm->vmcb &&
2261             svm->vcpu.arch.last_vmentry_cpu == cpu)
2262                 return;
2263
2264         sd->sev_vmcbs[asid] = svm->vmcb;
2265         svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
2266         vmcb_mark_dirty(svm->vmcb, VMCB_ASID);
2267 }
2268
2269 #define GHCB_SCRATCH_AREA_LIMIT         (16ULL * PAGE_SIZE)
2270 static bool setup_vmgexit_scratch(struct vcpu_svm *svm, bool sync, u64 len)
2271 {
2272         struct vmcb_control_area *control = &svm->vmcb->control;
2273         struct ghcb *ghcb = svm->ghcb;
2274         u64 ghcb_scratch_beg, ghcb_scratch_end;
2275         u64 scratch_gpa_beg, scratch_gpa_end;
2276         void *scratch_va;
2277
2278         scratch_gpa_beg = ghcb_get_sw_scratch(ghcb);
2279         if (!scratch_gpa_beg) {
2280                 pr_err("vmgexit: scratch gpa not provided\n");
2281                 return false;
2282         }
2283
2284         scratch_gpa_end = scratch_gpa_beg + len;
2285         if (scratch_gpa_end < scratch_gpa_beg) {
2286                 pr_err("vmgexit: scratch length (%#llx) not valid for scratch address (%#llx)\n",
2287                        len, scratch_gpa_beg);
2288                 return false;
2289         }
2290
2291         if ((scratch_gpa_beg & PAGE_MASK) == control->ghcb_gpa) {
2292                 /* Scratch area begins within GHCB */
2293                 ghcb_scratch_beg = control->ghcb_gpa +
2294                                    offsetof(struct ghcb, shared_buffer);
2295                 ghcb_scratch_end = control->ghcb_gpa +
2296                                    offsetof(struct ghcb, reserved_1);
2297
2298                 /*
2299                  * If the scratch area begins within the GHCB, it must be
2300                  * completely contained in the GHCB shared buffer area.
2301                  */
2302                 if (scratch_gpa_beg < ghcb_scratch_beg ||
2303                     scratch_gpa_end > ghcb_scratch_end) {
2304                         pr_err("vmgexit: scratch area is outside of GHCB shared buffer area (%#llx - %#llx)\n",
2305                                scratch_gpa_beg, scratch_gpa_end);
2306                         return false;
2307                 }
2308
2309                 scratch_va = (void *)svm->ghcb;
2310                 scratch_va += (scratch_gpa_beg - control->ghcb_gpa);
2311         } else {
2312                 /*
2313                  * The guest memory must be read into a kernel buffer, so
2314                  * limit the size
2315                  */
2316                 if (len > GHCB_SCRATCH_AREA_LIMIT) {
2317                         pr_err("vmgexit: scratch area exceeds KVM limits (%#llx requested, %#llx limit)\n",
2318                                len, GHCB_SCRATCH_AREA_LIMIT);
2319                         return false;
2320                 }
2321                 scratch_va = kzalloc(len, GFP_KERNEL_ACCOUNT);
2322                 if (!scratch_va)
2323                         return false;
2324
2325                 if (kvm_read_guest(svm->vcpu.kvm, scratch_gpa_beg, scratch_va, len)) {
2326                         /* Unable to copy scratch area from guest */
2327                         pr_err("vmgexit: kvm_read_guest for scratch area failed\n");
2328
2329                         kfree(scratch_va);
2330                         return false;
2331                 }
2332
2333                 /*
2334                  * The scratch area is outside the GHCB. The operation will
2335                  * dictate whether the buffer needs to be synced before running
2336                  * the vCPU next time (i.e. a read was requested so the data
2337                  * must be written back to the guest memory).
2338                  */
2339                 svm->ghcb_sa_sync = sync;
2340                 svm->ghcb_sa_free = true;
2341         }
2342
2343         svm->ghcb_sa = scratch_va;
2344         svm->ghcb_sa_len = len;
2345
2346         return true;
2347 }
2348
2349 static void set_ghcb_msr_bits(struct vcpu_svm *svm, u64 value, u64 mask,
2350                               unsigned int pos)
2351 {
2352         svm->vmcb->control.ghcb_gpa &= ~(mask << pos);
2353         svm->vmcb->control.ghcb_gpa |= (value & mask) << pos;
2354 }
2355
2356 static u64 get_ghcb_msr_bits(struct vcpu_svm *svm, u64 mask, unsigned int pos)
2357 {
2358         return (svm->vmcb->control.ghcb_gpa >> pos) & mask;
2359 }
2360
2361 static void set_ghcb_msr(struct vcpu_svm *svm, u64 value)
2362 {
2363         svm->vmcb->control.ghcb_gpa = value;
2364 }
2365
2366 static int sev_handle_vmgexit_msr_protocol(struct vcpu_svm *svm)
2367 {
2368         struct vmcb_control_area *control = &svm->vmcb->control;
2369         struct kvm_vcpu *vcpu = &svm->vcpu;
2370         u64 ghcb_info;
2371         int ret = 1;
2372
2373         ghcb_info = control->ghcb_gpa & GHCB_MSR_INFO_MASK;
2374
2375         trace_kvm_vmgexit_msr_protocol_enter(svm->vcpu.vcpu_id,
2376                                              control->ghcb_gpa);
2377
2378         switch (ghcb_info) {
2379         case GHCB_MSR_SEV_INFO_REQ:
2380                 set_ghcb_msr(svm, GHCB_MSR_SEV_INFO(GHCB_VERSION_MAX,
2381                                                     GHCB_VERSION_MIN,
2382                                                     sev_enc_bit));
2383                 break;
2384         case GHCB_MSR_CPUID_REQ: {
2385                 u64 cpuid_fn, cpuid_reg, cpuid_value;
2386
2387                 cpuid_fn = get_ghcb_msr_bits(svm,
2388                                              GHCB_MSR_CPUID_FUNC_MASK,
2389                                              GHCB_MSR_CPUID_FUNC_POS);
2390
2391                 /* Initialize the registers needed by the CPUID intercept */
2392                 vcpu->arch.regs[VCPU_REGS_RAX] = cpuid_fn;
2393                 vcpu->arch.regs[VCPU_REGS_RCX] = 0;
2394
2395                 ret = svm_invoke_exit_handler(vcpu, SVM_EXIT_CPUID);
2396                 if (!ret) {
2397                         ret = -EINVAL;
2398                         break;
2399                 }
2400
2401                 cpuid_reg = get_ghcb_msr_bits(svm,
2402                                               GHCB_MSR_CPUID_REG_MASK,
2403                                               GHCB_MSR_CPUID_REG_POS);
2404                 if (cpuid_reg == 0)
2405                         cpuid_value = vcpu->arch.regs[VCPU_REGS_RAX];
2406                 else if (cpuid_reg == 1)
2407                         cpuid_value = vcpu->arch.regs[VCPU_REGS_RBX];
2408                 else if (cpuid_reg == 2)
2409                         cpuid_value = vcpu->arch.regs[VCPU_REGS_RCX];
2410                 else
2411                         cpuid_value = vcpu->arch.regs[VCPU_REGS_RDX];
2412
2413                 set_ghcb_msr_bits(svm, cpuid_value,
2414                                   GHCB_MSR_CPUID_VALUE_MASK,
2415                                   GHCB_MSR_CPUID_VALUE_POS);
2416
2417                 set_ghcb_msr_bits(svm, GHCB_MSR_CPUID_RESP,
2418                                   GHCB_MSR_INFO_MASK,
2419                                   GHCB_MSR_INFO_POS);
2420                 break;
2421         }
2422         case GHCB_MSR_TERM_REQ: {
2423                 u64 reason_set, reason_code;
2424
2425                 reason_set = get_ghcb_msr_bits(svm,
2426                                                GHCB_MSR_TERM_REASON_SET_MASK,
2427                                                GHCB_MSR_TERM_REASON_SET_POS);
2428                 reason_code = get_ghcb_msr_bits(svm,
2429                                                 GHCB_MSR_TERM_REASON_MASK,
2430                                                 GHCB_MSR_TERM_REASON_POS);
2431                 pr_info("SEV-ES guest requested termination: %#llx:%#llx\n",
2432                         reason_set, reason_code);
2433                 fallthrough;
2434         }
2435         default:
2436                 ret = -EINVAL;
2437         }
2438
2439         trace_kvm_vmgexit_msr_protocol_exit(svm->vcpu.vcpu_id,
2440                                             control->ghcb_gpa, ret);
2441
2442         return ret;
2443 }
2444
2445 int sev_handle_vmgexit(struct kvm_vcpu *vcpu)
2446 {
2447         struct vcpu_svm *svm = to_svm(vcpu);
2448         struct vmcb_control_area *control = &svm->vmcb->control;
2449         u64 ghcb_gpa, exit_code;
2450         struct ghcb *ghcb;
2451         int ret;
2452
2453         /* Validate the GHCB */
2454         ghcb_gpa = control->ghcb_gpa;
2455         if (ghcb_gpa & GHCB_MSR_INFO_MASK)
2456                 return sev_handle_vmgexit_msr_protocol(svm);
2457
2458         if (!ghcb_gpa) {
2459                 vcpu_unimpl(vcpu, "vmgexit: GHCB gpa is not set\n");
2460                 return -EINVAL;
2461         }
2462
2463         if (kvm_vcpu_map(vcpu, ghcb_gpa >> PAGE_SHIFT, &svm->ghcb_map)) {
2464                 /* Unable to map GHCB from guest */
2465                 vcpu_unimpl(vcpu, "vmgexit: error mapping GHCB [%#llx] from guest\n",
2466                             ghcb_gpa);
2467                 return -EINVAL;
2468         }
2469
2470         svm->ghcb = svm->ghcb_map.hva;
2471         ghcb = svm->ghcb_map.hva;
2472
2473         trace_kvm_vmgexit_enter(vcpu->vcpu_id, ghcb);
2474
2475         exit_code = ghcb_get_sw_exit_code(ghcb);
2476
2477         ret = sev_es_validate_vmgexit(svm);
2478         if (ret)
2479                 return ret;
2480
2481         sev_es_sync_from_ghcb(svm);
2482         ghcb_set_sw_exit_info_1(ghcb, 0);
2483         ghcb_set_sw_exit_info_2(ghcb, 0);
2484
2485         ret = -EINVAL;
2486         switch (exit_code) {
2487         case SVM_VMGEXIT_MMIO_READ:
2488                 if (!setup_vmgexit_scratch(svm, true, control->exit_info_2))
2489                         break;
2490
2491                 ret = kvm_sev_es_mmio_read(vcpu,
2492                                            control->exit_info_1,
2493                                            control->exit_info_2,
2494                                            svm->ghcb_sa);
2495                 break;
2496         case SVM_VMGEXIT_MMIO_WRITE:
2497                 if (!setup_vmgexit_scratch(svm, false, control->exit_info_2))
2498                         break;
2499
2500                 ret = kvm_sev_es_mmio_write(vcpu,
2501                                             control->exit_info_1,
2502                                             control->exit_info_2,
2503                                             svm->ghcb_sa);
2504                 break;
2505         case SVM_VMGEXIT_NMI_COMPLETE:
2506                 ret = svm_invoke_exit_handler(vcpu, SVM_EXIT_IRET);
2507                 break;
2508         case SVM_VMGEXIT_AP_HLT_LOOP:
2509                 ret = kvm_emulate_ap_reset_hold(vcpu);
2510                 break;
2511         case SVM_VMGEXIT_AP_JUMP_TABLE: {
2512                 struct kvm_sev_info *sev = &to_kvm_svm(vcpu->kvm)->sev_info;
2513
2514                 switch (control->exit_info_1) {
2515                 case 0:
2516                         /* Set AP jump table address */
2517                         sev->ap_jump_table = control->exit_info_2;
2518                         break;
2519                 case 1:
2520                         /* Get AP jump table address */
2521                         ghcb_set_sw_exit_info_2(ghcb, sev->ap_jump_table);
2522                         break;
2523                 default:
2524                         pr_err("svm: vmgexit: unsupported AP jump table request - exit_info_1=%#llx\n",
2525                                control->exit_info_1);
2526                         ghcb_set_sw_exit_info_1(ghcb, 1);
2527                         ghcb_set_sw_exit_info_2(ghcb,
2528                                                 X86_TRAP_UD |
2529                                                 SVM_EVTINJ_TYPE_EXEPT |
2530                                                 SVM_EVTINJ_VALID);
2531                 }
2532
2533                 ret = 1;
2534                 break;
2535         }
2536         case SVM_VMGEXIT_UNSUPPORTED_EVENT:
2537                 vcpu_unimpl(vcpu,
2538                             "vmgexit: unsupported event - exit_info_1=%#llx, exit_info_2=%#llx\n",
2539                             control->exit_info_1, control->exit_info_2);
2540                 break;
2541         default:
2542                 ret = svm_invoke_exit_handler(vcpu, exit_code);
2543         }
2544
2545         return ret;
2546 }
2547
2548 int sev_es_string_io(struct vcpu_svm *svm, int size, unsigned int port, int in)
2549 {
2550         if (!setup_vmgexit_scratch(svm, in, svm->vmcb->control.exit_info_2))
2551                 return -EINVAL;
2552
2553         return kvm_sev_es_string_io(&svm->vcpu, size, port,
2554                                     svm->ghcb_sa, svm->ghcb_sa_len, in);
2555 }
2556
2557 void sev_es_init_vmcb(struct vcpu_svm *svm)
2558 {
2559         struct kvm_vcpu *vcpu = &svm->vcpu;
2560
2561         svm->vmcb->control.nested_ctl |= SVM_NESTED_CTL_SEV_ES_ENABLE;
2562         svm->vmcb->control.virt_ext |= LBR_CTL_ENABLE_MASK;
2563
2564         /*
2565          * An SEV-ES guest requires a VMSA area that is a separate from the
2566          * VMCB page. Do not include the encryption mask on the VMSA physical
2567          * address since hardware will access it using the guest key.
2568          */
2569         svm->vmcb->control.vmsa_pa = __pa(svm->vmsa);
2570
2571         /* Can't intercept CR register access, HV can't modify CR registers */
2572         svm_clr_intercept(svm, INTERCEPT_CR0_READ);
2573         svm_clr_intercept(svm, INTERCEPT_CR4_READ);
2574         svm_clr_intercept(svm, INTERCEPT_CR8_READ);
2575         svm_clr_intercept(svm, INTERCEPT_CR0_WRITE);
2576         svm_clr_intercept(svm, INTERCEPT_CR4_WRITE);
2577         svm_clr_intercept(svm, INTERCEPT_CR8_WRITE);
2578
2579         svm_clr_intercept(svm, INTERCEPT_SELECTIVE_CR0);
2580
2581         /* Track EFER/CR register changes */
2582         svm_set_intercept(svm, TRAP_EFER_WRITE);
2583         svm_set_intercept(svm, TRAP_CR0_WRITE);
2584         svm_set_intercept(svm, TRAP_CR4_WRITE);
2585         svm_set_intercept(svm, TRAP_CR8_WRITE);
2586
2587         /* No support for enable_vmware_backdoor */
2588         clr_exception_intercept(svm, GP_VECTOR);
2589
2590         /* Can't intercept XSETBV, HV can't modify XCR0 directly */
2591         svm_clr_intercept(svm, INTERCEPT_XSETBV);
2592
2593         /* Clear intercepts on selected MSRs */
2594         set_msr_interception(vcpu, svm->msrpm, MSR_EFER, 1, 1);
2595         set_msr_interception(vcpu, svm->msrpm, MSR_IA32_CR_PAT, 1, 1);
2596         set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
2597         set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
2598         set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
2599         set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
2600 }
2601
2602 void sev_es_create_vcpu(struct vcpu_svm *svm)
2603 {
2604         /*
2605          * Set the GHCB MSR value as per the GHCB specification when creating
2606          * a vCPU for an SEV-ES guest.
2607          */
2608         set_ghcb_msr(svm, GHCB_MSR_SEV_INFO(GHCB_VERSION_MAX,
2609                                             GHCB_VERSION_MIN,
2610                                             sev_enc_bit));
2611 }
2612
2613 void sev_es_prepare_guest_switch(struct vcpu_svm *svm, unsigned int cpu)
2614 {
2615         struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
2616         struct vmcb_save_area *hostsa;
2617
2618         /*
2619          * As an SEV-ES guest, hardware will restore the host state on VMEXIT,
2620          * of which one step is to perform a VMLOAD. Since hardware does not
2621          * perform a VMSAVE on VMRUN, the host savearea must be updated.
2622          */
2623         vmsave(__sme_page_pa(sd->save_area));
2624
2625         /* XCR0 is restored on VMEXIT, save the current host value */
2626         hostsa = (struct vmcb_save_area *)(page_address(sd->save_area) + 0x400);
2627         hostsa->xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
2628
2629         /* PKRU is restored on VMEXIT, save the current host value */
2630         hostsa->pkru = read_pkru();
2631
2632         /* MSR_IA32_XSS is restored on VMEXIT, save the currnet host value */
2633         hostsa->xss = host_xss;
2634 }
2635
2636 void sev_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
2637 {
2638         struct vcpu_svm *svm = to_svm(vcpu);
2639
2640         /* First SIPI: Use the values as initially set by the VMM */
2641         if (!svm->received_first_sipi) {
2642                 svm->received_first_sipi = true;
2643                 return;
2644         }
2645
2646         /*
2647          * Subsequent SIPI: Return from an AP Reset Hold VMGEXIT, where
2648          * the guest will set the CS and RIP. Set SW_EXIT_INFO_2 to a
2649          * non-zero value.
2650          */
2651         if (!svm->ghcb)
2652                 return;
2653
2654         ghcb_set_sw_exit_info_2(svm->ghcb, 1);
2655 }