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_vcpu *vcpu,
 599                                     int *error)
 600 {
 601         struct sev_data_launch_update_vmsa vmsa;
 602         struct vcpu_svm *svm = to_svm(vcpu);
 603         int ret;
 604
 605         /* Perform some pre-encryption checks against the VMSA */
 606         ret = sev_es_sync_vmsa(svm);
 607         if (ret)
 608                 return ret;
 609
 610         /*
 611          * The LAUNCH_UPDATE_VMSA command will perform in-place encryption of
 612          * the VMSA memory content (i.e it will write the same memory region
 613          * with the guest's key), so invalidate it first.
 614          */
 615         clflush_cache_range(svm->vmsa, PAGE_SIZE);
 616
 617         vmsa.reserved = 0;
 618         vmsa.handle = to_kvm_svm(kvm)->sev_info.handle;
 619         vmsa.address = __sme_pa(svm->vmsa);
 620         vmsa.len = PAGE_SIZE;
 621         return sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_VMSA, &vmsa, error);
 622 }
 623
 624 static int sev_launch_update_vmsa(struct kvm *kvm, struct kvm_sev_cmd *argp)
 625 {
 626         struct kvm_vcpu *vcpu;
 627         int i, ret;
 628
 629         if (!sev_es_guest(kvm))
 630                 return -ENOTTY;
 631
 632         kvm_for_each_vcpu(i, vcpu, kvm) {
 633                 ret = mutex_lock_killable(&vcpu->mutex);
 634                 if (ret)
 635                         return ret;
 636
 637                 ret = __sev_launch_update_vmsa(kvm, vcpu, &argp->error);
 638
 639                 mutex_unlock(&vcpu->mutex);
 640                 if (ret)
 641                         return ret;
 642         }
 643
 644         return 0;
 645 }
 646
 647 static int sev_launch_measure(struct kvm *kvm, struct kvm_sev_cmd *argp)
 648 {
 649         void __user *measure = (void __user *)(uintptr_t)argp->data;
 650         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 651         struct sev_data_launch_measure data;
 652         struct kvm_sev_launch_measure params;
 653         void __user *p = NULL;
 654         void *blob = NULL;
 655         int ret;
 656
 657         if (!sev_guest(kvm))
 658                 return -ENOTTY;
 659
 660         if (copy_from_user(&params, measure, sizeof(params)))
 661                 return -EFAULT;
 662
 663         memset(&data, 0, sizeof(data));
 664
 665         /* User wants to query the blob length */
 666         if (!params.len)
 667                 goto cmd;
 668
 669         p = (void __user *)(uintptr_t)params.uaddr;
 670         if (p) {
 671                 if (params.len > SEV_FW_BLOB_MAX_SIZE)
 672                         return -EINVAL;
 673
 674                 blob = kmalloc(params.len, GFP_KERNEL_ACCOUNT);
 675                 if (!blob)
 676                         return -ENOMEM;
 677
 678                 data.address = __psp_pa(blob);
 679                 data.len = params.len;
 680         }
 681
 682 cmd:
 683         data.handle = sev->handle;
 684         ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_MEASURE, &data, &argp->error);
 685
 686         /*
 687          * If we query the session length, FW responded with expected data.
 688          */
 689         if (!params.len)
 690                 goto done;
 691
 692         if (ret)
 693                 goto e_free_blob;
 694
 695         if (blob) {
 696                 if (copy_to_user(p, blob, params.len))
 697                         ret = -EFAULT;
 698         }
 699
 700 done:
 701         params.len = data.len;
 702         if (copy_to_user(measure, &params, sizeof(params)))
 703                 ret = -EFAULT;
 704 e_free_blob:
 705         kfree(blob);
 706         return ret;
 707 }
 708
 709 static int sev_launch_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
 710 {
 711         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 712         struct sev_data_launch_finish data;
 713
 714         if (!sev_guest(kvm))
 715                 return -ENOTTY;
 716
 717         data.handle = sev->handle;
 718         return sev_issue_cmd(kvm, SEV_CMD_LAUNCH_FINISH, &data, &argp->error);
 719 }
 720
 721 static int sev_guest_status(struct kvm *kvm, struct kvm_sev_cmd *argp)
 722 {
 723         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 724         struct kvm_sev_guest_status params;
 725         struct sev_data_guest_status data;
 726         int ret;
 727
 728         if (!sev_guest(kvm))
 729                 return -ENOTTY;
 730
 731         memset(&data, 0, sizeof(data));
 732
 733         data.handle = sev->handle;
 734         ret = sev_issue_cmd(kvm, SEV_CMD_GUEST_STATUS, &data, &argp->error);
 735         if (ret)
 736                 return ret;
 737
 738         params.policy = data.policy;
 739         params.state = data.state;
 740         params.handle = data.handle;
 741
 742         if (copy_to_user((void __user *)(uintptr_t)argp->data, &params, sizeof(params)))
 743                 ret = -EFAULT;
 744
 745         return ret;
 746 }
 747
 748 static int __sev_issue_dbg_cmd(struct kvm *kvm, unsigned long src,
 749                                unsigned long dst, int size,
 750                                int *error, bool enc)
 751 {
 752         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 753         struct sev_data_dbg data;
 754
 755         data.reserved = 0;
 756         data.handle = sev->handle;
 757         data.dst_addr = dst;
 758         data.src_addr = src;
 759         data.len = size;
 760
 761         return sev_issue_cmd(kvm,
 762                              enc ? SEV_CMD_DBG_ENCRYPT : SEV_CMD_DBG_DECRYPT,
 763                              &data, error);
 764 }
 765
 766 static int __sev_dbg_decrypt(struct kvm *kvm, unsigned long src_paddr,
 767                              unsigned long dst_paddr, int sz, int *err)
 768 {
 769         int offset;
 770
 771         /*
 772          * Its safe to read more than we are asked, caller should ensure that
 773          * destination has enough space.
 774          */
 775         offset = src_paddr & 15;
 776         src_paddr = round_down(src_paddr, 16);
 777         sz = round_up(sz + offset, 16);
 778
 779         return __sev_issue_dbg_cmd(kvm, src_paddr, dst_paddr, sz, err, false);
 780 }
 781
 782 static int __sev_dbg_decrypt_user(struct kvm *kvm, unsigned long paddr,
 783                                   void __user *dst_uaddr,
 784                                   unsigned long dst_paddr,
 785                                   int size, int *err)
 786 {
 787         struct page *tpage = NULL;
 788         int ret, offset;
 789
 790         /* if inputs are not 16-byte then use intermediate buffer */
 791         if (!IS_ALIGNED(dst_paddr, 16) ||
 792             !IS_ALIGNED(paddr,     16) ||
 793             !IS_ALIGNED(size,      16)) {
 794                 tpage = (void *)alloc_page(GFP_KERNEL);
 795                 if (!tpage)
 796                         return -ENOMEM;
 797
 798                 dst_paddr = __sme_page_pa(tpage);
 799         }
 800
 801         ret = __sev_dbg_decrypt(kvm, paddr, dst_paddr, size, err);
 802         if (ret)
 803                 goto e_free;
 804
 805         if (tpage) {
 806                 offset = paddr & 15;
 807                 if (copy_to_user(dst_uaddr, page_address(tpage) + offset, size))
 808                         ret = -EFAULT;
 809         }
 810
 811 e_free:
 812         if (tpage)
 813                 __free_page(tpage);
 814
 815         return ret;
 816 }
 817
 818 static int __sev_dbg_encrypt_user(struct kvm *kvm, unsigned long paddr,
 819                                   void __user *vaddr,
 820                                   unsigned long dst_paddr,
 821                                   void __user *dst_vaddr,
 822                                   int size, int *error)
 823 {
 824         struct page *src_tpage = NULL;
 825         struct page *dst_tpage = NULL;
 826         int ret, len = size;
 827
 828         /* If source buffer is not aligned then use an intermediate buffer */
 829         if (!IS_ALIGNED((unsigned long)vaddr, 16)) {
 830                 src_tpage = alloc_page(GFP_KERNEL);
 831                 if (!src_tpage)
 832                         return -ENOMEM;
 833
 834                 if (copy_from_user(page_address(src_tpage), vaddr, size)) {
 835                         __free_page(src_tpage);
 836                         return -EFAULT;
 837                 }
 838
 839                 paddr = __sme_page_pa(src_tpage);
 840         }
 841
 842         /*
 843          *  If destination buffer or length is not aligned then do read-modify-write:
 844          *   - decrypt destination in an intermediate buffer
 845          *   - copy the source buffer in an intermediate buffer
 846          *   - use the intermediate buffer as source buffer
 847          */
 848         if (!IS_ALIGNED((unsigned long)dst_vaddr, 16) || !IS_ALIGNED(size, 16)) {
 849                 int dst_offset;
 850
 851                 dst_tpage = alloc_page(GFP_KERNEL);
 852                 if (!dst_tpage) {
 853                         ret = -ENOMEM;
 854                         goto e_free;
 855                 }
 856
 857                 ret = __sev_dbg_decrypt(kvm, dst_paddr,
 858                                         __sme_page_pa(dst_tpage), size, error);
 859                 if (ret)
 860                         goto e_free;
 861
 862                 /*
 863                  *  If source is kernel buffer then use memcpy() otherwise
 864                  *  copy_from_user().
 865                  */
 866                 dst_offset = dst_paddr & 15;
 867
 868                 if (src_tpage)
 869                         memcpy(page_address(dst_tpage) + dst_offset,
 870                                page_address(src_tpage), size);
 871                 else {
 872                         if (copy_from_user(page_address(dst_tpage) + dst_offset,
 873                                            vaddr, size)) {
 874                                 ret = -EFAULT;
 875                                 goto e_free;
 876                         }
 877                 }
 878
 879                 paddr = __sme_page_pa(dst_tpage);
 880                 dst_paddr = round_down(dst_paddr, 16);
 881                 len = round_up(size, 16);
 882         }
 883
 884         ret = __sev_issue_dbg_cmd(kvm, paddr, dst_paddr, len, error, true);
 885
 886 e_free:
 887         if (src_tpage)
 888                 __free_page(src_tpage);
 889         if (dst_tpage)
 890                 __free_page(dst_tpage);
 891         return ret;
 892 }
 893
 894 static int sev_dbg_crypt(struct kvm *kvm, struct kvm_sev_cmd *argp, bool dec)
 895 {
 896         unsigned long vaddr, vaddr_end, next_vaddr;
 897         unsigned long dst_vaddr;
 898         struct page **src_p, **dst_p;
 899         struct kvm_sev_dbg debug;
 900         unsigned long n;
 901         unsigned int size;
 902         int ret;
 903
 904         if (!sev_guest(kvm))
 905                 return -ENOTTY;
 906
 907         if (copy_from_user(&debug, (void __user *)(uintptr_t)argp->data, sizeof(debug)))
 908                 return -EFAULT;
 909
 910         if (!debug.len || debug.src_uaddr + debug.len < debug.src_uaddr)
 911                 return -EINVAL;
 912         if (!debug.dst_uaddr)
 913                 return -EINVAL;
 914
 915         vaddr = debug.src_uaddr;
 916         size = debug.len;
 917         vaddr_end = vaddr + size;
 918         dst_vaddr = debug.dst_uaddr;
 919
 920         for (; vaddr < vaddr_end; vaddr = next_vaddr) {
 921                 int len, s_off, d_off;
 922
 923                 /* lock userspace source and destination page */
 924                 src_p = sev_pin_memory(kvm, vaddr & PAGE_MASK, PAGE_SIZE, &n, 0);
 925                 if (IS_ERR(src_p))
 926                         return PTR_ERR(src_p);
 927
 928                 dst_p = sev_pin_memory(kvm, dst_vaddr & PAGE_MASK, PAGE_SIZE, &n, 1);
 929                 if (IS_ERR(dst_p)) {
 930                         sev_unpin_memory(kvm, src_p, n);
 931                         return PTR_ERR(dst_p);
 932                 }
 933
 934                 /*
 935                  * Flush (on non-coherent CPUs) before DBG_{DE,EN}CRYPT read or modify
 936                  * the pages; flush the destination too so that future accesses do not
 937                  * see stale data.
 938                  */
 939                 sev_clflush_pages(src_p, 1);
 940                 sev_clflush_pages(dst_p, 1);
 941
 942                 /*
 943                  * Since user buffer may not be page aligned, calculate the
 944                  * offset within the page.
 945                  */
 946                 s_off = vaddr & ~PAGE_MASK;
 947                 d_off = dst_vaddr & ~PAGE_MASK;
 948                 len = min_t(size_t, (PAGE_SIZE - s_off), size);
 949
 950                 if (dec)
 951                         ret = __sev_dbg_decrypt_user(kvm,
 952                                                      __sme_page_pa(src_p[0]) + s_off,
 953                                                      (void __user *)dst_vaddr,
 954                                                      __sme_page_pa(dst_p[0]) + d_off,
 955                                                      len, &argp->error);
 956                 else
 957                         ret = __sev_dbg_encrypt_user(kvm,
 958                                                      __sme_page_pa(src_p[0]) + s_off,
 959                                                      (void __user *)vaddr,
 960                                                      __sme_page_pa(dst_p[0]) + d_off,
 961                                                      (void __user *)dst_vaddr,
 962                                                      len, &argp->error);
 963
 964                 sev_unpin_memory(kvm, src_p, n);
 965                 sev_unpin_memory(kvm, dst_p, n);
 966
 967                 if (ret)
 968                         goto err;
 969
 970                 next_vaddr = vaddr + len;
 971                 dst_vaddr = dst_vaddr + len;
 972                 size -= len;
 973         }
 974 err:
 975         return ret;
 976 }
 977
 978 static int sev_launch_secret(struct kvm *kvm, struct kvm_sev_cmd *argp)
 979 {
 980         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
 981         struct sev_data_launch_secret data;
 982         struct kvm_sev_launch_secret params;
 983         struct page **pages;
 984         void *blob, *hdr;
 985         unsigned long n, i;
 986         int ret, offset;
 987
 988         if (!sev_guest(kvm))
 989                 return -ENOTTY;
 990
 991         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data, sizeof(params)))
 992                 return -EFAULT;
 993
 994         pages = sev_pin_memory(kvm, params.guest_uaddr, params.guest_len, &n, 1);
 995         if (IS_ERR(pages))
 996                 return PTR_ERR(pages);
 997
 998         /*
 999          * Flush (on non-coherent CPUs) before LAUNCH_SECRET encrypts pages in
1000          * place; the cache may contain the data that was written unencrypted.
1001          */
1002         sev_clflush_pages(pages, n);
1003
1004         /*
1005          * The secret must be copied into contiguous memory region, lets verify
1006          * that userspace memory pages are contiguous before we issue command.
1007          */
1008         if (get_num_contig_pages(0, pages, n) != n) {
1009                 ret = -EINVAL;
1010                 goto e_unpin_memory;
1011         }
1012
1013         memset(&data, 0, sizeof(data));
1014
1015         offset = params.guest_uaddr & (PAGE_SIZE - 1);
1016         data.guest_address = __sme_page_pa(pages[0]) + offset;
1017         data.guest_len = params.guest_len;
1018
1019         blob = psp_copy_user_blob(params.trans_uaddr, params.trans_len);
1020         if (IS_ERR(blob)) {
1021                 ret = PTR_ERR(blob);
1022                 goto e_unpin_memory;
1023         }
1024
1025         data.trans_address = __psp_pa(blob);
1026         data.trans_len = params.trans_len;
1027
1028         hdr = psp_copy_user_blob(params.hdr_uaddr, params.hdr_len);
1029         if (IS_ERR(hdr)) {
1030                 ret = PTR_ERR(hdr);
1031                 goto e_free_blob;
1032         }
1033         data.hdr_address = __psp_pa(hdr);
1034         data.hdr_len = params.hdr_len;
1035
1036         data.handle = sev->handle;
1037         ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_SECRET, &data, &argp->error);
1038
1039         kfree(hdr);
1040
1041 e_free_blob:
1042         kfree(blob);
1043 e_unpin_memory:
1044         /* content of memory is updated, mark pages dirty */
1045         for (i = 0; i < n; i++) {
1046                 set_page_dirty_lock(pages[i]);
1047                 mark_page_accessed(pages[i]);
1048         }
1049         sev_unpin_memory(kvm, pages, n);
1050         return ret;
1051 }
1052
1053 static int sev_get_attestation_report(struct kvm *kvm, struct kvm_sev_cmd *argp)
1054 {
1055         void __user *report = (void __user *)(uintptr_t)argp->data;
1056         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1057         struct sev_data_attestation_report data;
1058         struct kvm_sev_attestation_report params;
1059         void __user *p;
1060         void *blob = NULL;
1061         int ret;
1062
1063         if (!sev_guest(kvm))
1064                 return -ENOTTY;
1065
1066         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data, sizeof(params)))
1067                 return -EFAULT;
1068
1069         memset(&data, 0, sizeof(data));
1070
1071         /* User wants to query the blob length */
1072         if (!params.len)
1073                 goto cmd;
1074
1075         p = (void __user *)(uintptr_t)params.uaddr;
1076         if (p) {
1077                 if (params.len > SEV_FW_BLOB_MAX_SIZE)
1078                         return -EINVAL;
1079
1080                 blob = kmalloc(params.len, GFP_KERNEL_ACCOUNT);
1081                 if (!blob)
1082                         return -ENOMEM;
1083
1084                 data.address = __psp_pa(blob);
1085                 data.len = params.len;
1086                 memcpy(data.mnonce, params.mnonce, sizeof(params.mnonce));
1087         }
1088 cmd:
1089         data.handle = sev->handle;
1090         ret = sev_issue_cmd(kvm, SEV_CMD_ATTESTATION_REPORT, &data, &argp->error);
1091         /*
1092          * If we query the session length, FW responded with expected data.
1093          */
1094         if (!params.len)
1095                 goto done;
1096
1097         if (ret)
1098                 goto e_free_blob;
1099
1100         if (blob) {
1101                 if (copy_to_user(p, blob, params.len))
1102                         ret = -EFAULT;
1103         }
1104
1105 done:
1106         params.len = data.len;
1107         if (copy_to_user(report, &params, sizeof(params)))
1108                 ret = -EFAULT;
1109 e_free_blob:
1110         kfree(blob);
1111         return ret;
1112 }
1113
1114 /* Userspace wants to query session length. */
1115 static int
1116 __sev_send_start_query_session_length(struct kvm *kvm, struct kvm_sev_cmd *argp,
1117                                       struct kvm_sev_send_start *params)
1118 {
1119         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1120         struct sev_data_send_start data;
1121         int ret;
1122
1123         memset(&data, 0, sizeof(data));
1124         data.handle = sev->handle;
1125         ret = sev_issue_cmd(kvm, SEV_CMD_SEND_START, &data, &argp->error);
1126
1127         params->session_len = data.session_len;
1128         if (copy_to_user((void __user *)(uintptr_t)argp->data, params,
1129                                 sizeof(struct kvm_sev_send_start)))
1130                 ret = -EFAULT;
1131
1132         return ret;
1133 }
1134
1135 static int sev_send_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
1136 {
1137         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1138         struct sev_data_send_start data;
1139         struct kvm_sev_send_start params;
1140         void *amd_certs, *session_data;
1141         void *pdh_cert, *plat_certs;
1142         int ret;
1143
1144         if (!sev_guest(kvm))
1145                 return -ENOTTY;
1146
1147         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data,
1148                                 sizeof(struct kvm_sev_send_start)))
1149                 return -EFAULT;
1150
1151         /* if session_len is zero, userspace wants to query the session length */
1152         if (!params.session_len)
1153                 return __sev_send_start_query_session_length(kvm, argp,
1154                                 &params);
1155
1156         /* some sanity checks */
1157         if (!params.pdh_cert_uaddr || !params.pdh_cert_len ||
1158             !params.session_uaddr || params.session_len > SEV_FW_BLOB_MAX_SIZE)
1159                 return -EINVAL;
1160
1161         /* allocate the memory to hold the session data blob */
1162         session_data = kmalloc(params.session_len, GFP_KERNEL_ACCOUNT);
1163         if (!session_data)
1164                 return -ENOMEM;
1165
1166         /* copy the certificate blobs from userspace */
1167         pdh_cert = psp_copy_user_blob(params.pdh_cert_uaddr,
1168                                 params.pdh_cert_len);
1169         if (IS_ERR(pdh_cert)) {
1170                 ret = PTR_ERR(pdh_cert);
1171                 goto e_free_session;
1172         }
1173
1174         plat_certs = psp_copy_user_blob(params.plat_certs_uaddr,
1175                                 params.plat_certs_len);
1176         if (IS_ERR(plat_certs)) {
1177                 ret = PTR_ERR(plat_certs);
1178                 goto e_free_pdh;
1179         }
1180
1181         amd_certs = psp_copy_user_blob(params.amd_certs_uaddr,
1182                                 params.amd_certs_len);
1183         if (IS_ERR(amd_certs)) {
1184                 ret = PTR_ERR(amd_certs);
1185                 goto e_free_plat_cert;
1186         }
1187
1188         /* populate the FW SEND_START field with system physical address */
1189         memset(&data, 0, sizeof(data));
1190         data.pdh_cert_address = __psp_pa(pdh_cert);
1191         data.pdh_cert_len = params.pdh_cert_len;
1192         data.plat_certs_address = __psp_pa(plat_certs);
1193         data.plat_certs_len = params.plat_certs_len;
1194         data.amd_certs_address = __psp_pa(amd_certs);
1195         data.amd_certs_len = params.amd_certs_len;
1196         data.session_address = __psp_pa(session_data);
1197         data.session_len = params.session_len;
1198         data.handle = sev->handle;
1199
1200         ret = sev_issue_cmd(kvm, SEV_CMD_SEND_START, &data, &argp->error);
1201
1202         if (!ret && copy_to_user((void __user *)(uintptr_t)params.session_uaddr,
1203                         session_data, params.session_len)) {
1204                 ret = -EFAULT;
1205                 goto e_free_amd_cert;
1206         }
1207
1208         params.policy = data.policy;
1209         params.session_len = data.session_len;
1210         if (copy_to_user((void __user *)(uintptr_t)argp->data, &params,
1211                                 sizeof(struct kvm_sev_send_start)))
1212                 ret = -EFAULT;
1213
1214 e_free_amd_cert:
1215         kfree(amd_certs);
1216 e_free_plat_cert:
1217         kfree(plat_certs);
1218 e_free_pdh:
1219         kfree(pdh_cert);
1220 e_free_session:
1221         kfree(session_data);
1222         return ret;
1223 }
1224
1225 /* Userspace wants to query either header or trans length. */
1226 static int
1227 __sev_send_update_data_query_lengths(struct kvm *kvm, struct kvm_sev_cmd *argp,
1228                                      struct kvm_sev_send_update_data *params)
1229 {
1230         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1231         struct sev_data_send_update_data data;
1232         int ret;
1233
1234         memset(&data, 0, sizeof(data));
1235         data.handle = sev->handle;
1236         ret = sev_issue_cmd(kvm, SEV_CMD_SEND_UPDATE_DATA, &data, &argp->error);
1237
1238         params->hdr_len = data.hdr_len;
1239         params->trans_len = data.trans_len;
1240
1241         if (copy_to_user((void __user *)(uintptr_t)argp->data, params,
1242                          sizeof(struct kvm_sev_send_update_data)))
1243                 ret = -EFAULT;
1244
1245         return ret;
1246 }
1247
1248 static int sev_send_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
1249 {
1250         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1251         struct sev_data_send_update_data data;
1252         struct kvm_sev_send_update_data params;
1253         void *hdr, *trans_data;
1254         struct page **guest_page;
1255         unsigned long n;
1256         int ret, offset;
1257
1258         if (!sev_guest(kvm))
1259                 return -ENOTTY;
1260
1261         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data,
1262                         sizeof(struct kvm_sev_send_update_data)))
1263                 return -EFAULT;
1264
1265         /* userspace wants to query either header or trans length */
1266         if (!params.trans_len || !params.hdr_len)
1267                 return __sev_send_update_data_query_lengths(kvm, argp, &params);
1268
1269         if (!params.trans_uaddr || !params.guest_uaddr ||
1270             !params.guest_len || !params.hdr_uaddr)
1271                 return -EINVAL;
1272
1273         /* Check if we are crossing the page boundary */
1274         offset = params.guest_uaddr & (PAGE_SIZE - 1);
1275         if ((params.guest_len + offset > PAGE_SIZE))
1276                 return -EINVAL;
1277
1278         /* Pin guest memory */
1279         guest_page = sev_pin_memory(kvm, params.guest_uaddr & PAGE_MASK,
1280                                     PAGE_SIZE, &n, 0);
1281         if (IS_ERR(guest_page))
1282                 return PTR_ERR(guest_page);
1283
1284         /* allocate memory for header and transport buffer */
1285         ret = -ENOMEM;
1286         hdr = kmalloc(params.hdr_len, GFP_KERNEL_ACCOUNT);
1287         if (!hdr)
1288                 goto e_unpin;
1289
1290         trans_data = kmalloc(params.trans_len, GFP_KERNEL_ACCOUNT);
1291         if (!trans_data)
1292                 goto e_free_hdr;
1293
1294         memset(&data, 0, sizeof(data));
1295         data.hdr_address = __psp_pa(hdr);
1296         data.hdr_len = params.hdr_len;
1297         data.trans_address = __psp_pa(trans_data);
1298         data.trans_len = params.trans_len;
1299
1300         /* The SEND_UPDATE_DATA command requires C-bit to be always set. */
1301         data.guest_address = (page_to_pfn(guest_page[0]) << PAGE_SHIFT) + offset;
1302         data.guest_address |= sev_me_mask;
1303         data.guest_len = params.guest_len;
1304         data.handle = sev->handle;
1305
1306         ret = sev_issue_cmd(kvm, SEV_CMD_SEND_UPDATE_DATA, &data, &argp->error);
1307
1308         if (ret)
1309                 goto e_free_trans_data;
1310
1311         /* copy transport buffer to user space */
1312         if (copy_to_user((void __user *)(uintptr_t)params.trans_uaddr,
1313                          trans_data, params.trans_len)) {
1314                 ret = -EFAULT;
1315                 goto e_free_trans_data;
1316         }
1317
1318         /* Copy packet header to userspace. */
1319         if (copy_to_user((void __user *)(uintptr_t)params.hdr_uaddr, hdr,
1320                          params.hdr_len))
1321                 ret = -EFAULT;
1322
1323 e_free_trans_data:
1324         kfree(trans_data);
1325 e_free_hdr:
1326         kfree(hdr);
1327 e_unpin:
1328         sev_unpin_memory(kvm, guest_page, n);
1329
1330         return ret;
1331 }
1332
1333 static int sev_send_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
1334 {
1335         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1336         struct sev_data_send_finish data;
1337
1338         if (!sev_guest(kvm))
1339                 return -ENOTTY;
1340
1341         data.handle = sev->handle;
1342         return sev_issue_cmd(kvm, SEV_CMD_SEND_FINISH, &data, &argp->error);
1343 }
1344
1345 static int sev_send_cancel(struct kvm *kvm, struct kvm_sev_cmd *argp)
1346 {
1347         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1348         struct sev_data_send_cancel data;
1349
1350         if (!sev_guest(kvm))
1351                 return -ENOTTY;
1352
1353         data.handle = sev->handle;
1354         return sev_issue_cmd(kvm, SEV_CMD_SEND_CANCEL, &data, &argp->error);
1355 }
1356
1357 static int sev_receive_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
1358 {
1359         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1360         struct sev_data_receive_start start;
1361         struct kvm_sev_receive_start params;
1362         int *error = &argp->error;
1363         void *session_data;
1364         void *pdh_data;
1365         int ret;
1366
1367         if (!sev_guest(kvm))
1368                 return -ENOTTY;
1369
1370         /* Get parameter from the userspace */
1371         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data,
1372                         sizeof(struct kvm_sev_receive_start)))
1373                 return -EFAULT;
1374
1375         /* some sanity checks */
1376         if (!params.pdh_uaddr || !params.pdh_len ||
1377             !params.session_uaddr || !params.session_len)
1378                 return -EINVAL;
1379
1380         pdh_data = psp_copy_user_blob(params.pdh_uaddr, params.pdh_len);
1381         if (IS_ERR(pdh_data))
1382                 return PTR_ERR(pdh_data);
1383
1384         session_data = psp_copy_user_blob(params.session_uaddr,
1385                         params.session_len);
1386         if (IS_ERR(session_data)) {
1387                 ret = PTR_ERR(session_data);
1388                 goto e_free_pdh;
1389         }
1390
1391         memset(&start, 0, sizeof(start));
1392         start.handle = params.handle;
1393         start.policy = params.policy;
1394         start.pdh_cert_address = __psp_pa(pdh_data);
1395         start.pdh_cert_len = params.pdh_len;
1396         start.session_address = __psp_pa(session_data);
1397         start.session_len = params.session_len;
1398
1399         /* create memory encryption context */
1400         ret = __sev_issue_cmd(argp->sev_fd, SEV_CMD_RECEIVE_START, &start,
1401                                 error);
1402         if (ret)
1403                 goto e_free_session;
1404
1405         /* Bind ASID to this guest */
1406         ret = sev_bind_asid(kvm, start.handle, error);
1407         if (ret) {
1408                 sev_decommission(start.handle);
1409                 goto e_free_session;
1410         }
1411
1412         params.handle = start.handle;
1413         if (copy_to_user((void __user *)(uintptr_t)argp->data,
1414                          &params, sizeof(struct kvm_sev_receive_start))) {
1415                 ret = -EFAULT;
1416                 sev_unbind_asid(kvm, start.handle);
1417                 goto e_free_session;
1418         }
1419
1420         sev->handle = start.handle;
1421         sev->fd = argp->sev_fd;
1422
1423 e_free_session:
1424         kfree(session_data);
1425 e_free_pdh:
1426         kfree(pdh_data);
1427
1428         return ret;
1429 }
1430
1431 static int sev_receive_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
1432 {
1433         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1434         struct kvm_sev_receive_update_data params;
1435         struct sev_data_receive_update_data data;
1436         void *hdr = NULL, *trans = NULL;
1437         struct page **guest_page;
1438         unsigned long n;
1439         int ret, offset;
1440
1441         if (!sev_guest(kvm))
1442                 return -EINVAL;
1443
1444         if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data,
1445                         sizeof(struct kvm_sev_receive_update_data)))
1446                 return -EFAULT;
1447
1448         if (!params.hdr_uaddr || !params.hdr_len ||
1449             !params.guest_uaddr || !params.guest_len ||
1450             !params.trans_uaddr || !params.trans_len)
1451                 return -EINVAL;
1452
1453         /* Check if we are crossing the page boundary */
1454         offset = params.guest_uaddr & (PAGE_SIZE - 1);
1455         if ((params.guest_len + offset > PAGE_SIZE))
1456                 return -EINVAL;
1457
1458         hdr = psp_copy_user_blob(params.hdr_uaddr, params.hdr_len);
1459         if (IS_ERR(hdr))
1460                 return PTR_ERR(hdr);
1461
1462         trans = psp_copy_user_blob(params.trans_uaddr, params.trans_len);
1463         if (IS_ERR(trans)) {
1464                 ret = PTR_ERR(trans);
1465                 goto e_free_hdr;
1466         }
1467
1468         memset(&data, 0, sizeof(data));
1469         data.hdr_address = __psp_pa(hdr);
1470         data.hdr_len = params.hdr_len;
1471         data.trans_address = __psp_pa(trans);
1472         data.trans_len = params.trans_len;
1473
1474         /* Pin guest memory */
1475         guest_page = sev_pin_memory(kvm, params.guest_uaddr & PAGE_MASK,
1476                                     PAGE_SIZE, &n, 1);
1477         if (IS_ERR(guest_page)) {
1478                 ret = PTR_ERR(guest_page);
1479                 goto e_free_trans;
1480         }
1481
1482         /* The RECEIVE_UPDATE_DATA command requires C-bit to be always set. */
1483         data.guest_address = (page_to_pfn(guest_page[0]) << PAGE_SHIFT) + offset;
1484         data.guest_address |= sev_me_mask;
1485         data.guest_len = params.guest_len;
1486         data.handle = sev->handle;
1487
1488         ret = sev_issue_cmd(kvm, SEV_CMD_RECEIVE_UPDATE_DATA, &data,
1489                                 &argp->error);
1490
1491         sev_unpin_memory(kvm, guest_page, n);
1492
1493 e_free_trans:
1494         kfree(trans);
1495 e_free_hdr:
1496         kfree(hdr);
1497
1498         return ret;
1499 }
1500
1501 static int sev_receive_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
1502 {
1503         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1504         struct sev_data_receive_finish data;
1505
1506         if (!sev_guest(kvm))
1507                 return -ENOTTY;
1508
1509         data.handle = sev->handle;
1510         return sev_issue_cmd(kvm, SEV_CMD_RECEIVE_FINISH, &data, &argp->error);
1511 }
1512
1513 static bool cmd_allowed_from_miror(u32 cmd_id)
1514 {
1515         /*
1516          * Allow mirrors VM to call KVM_SEV_LAUNCH_UPDATE_VMSA to enable SEV-ES
1517          * active mirror VMs. Also allow the debugging and status commands.
1518          */
1519         if (cmd_id == KVM_SEV_LAUNCH_UPDATE_VMSA ||
1520             cmd_id == KVM_SEV_GUEST_STATUS || cmd_id == KVM_SEV_DBG_DECRYPT ||
1521             cmd_id == KVM_SEV_DBG_ENCRYPT)
1522                 return true;
1523
1524         return false;
1525 }
1526
1527 int svm_mem_enc_op(struct kvm *kvm, void __user *argp)
1528 {
1529         struct kvm_sev_cmd sev_cmd;
1530         int r;
1531
1532         if (!sev_enabled)
1533                 return -ENOTTY;
1534
1535         if (!argp)
1536                 return 0;
1537
1538         if (copy_from_user(&sev_cmd, argp, sizeof(struct kvm_sev_cmd)))
1539                 return -EFAULT;
1540
1541         mutex_lock(&kvm->lock);
1542
1543         /* Only the enc_context_owner handles some memory enc operations. */
1544         if (is_mirroring_enc_context(kvm) &&
1545             !cmd_allowed_from_miror(sev_cmd.id)) {
1546                 r = -EINVAL;
1547                 goto out;
1548         }
1549
1550         switch (sev_cmd.id) {
1551         case KVM_SEV_ES_INIT:
1552                 if (!sev_es_enabled) {
1553                         r = -ENOTTY;
1554                         goto out;
1555                 }
1556                 fallthrough;
1557         case KVM_SEV_INIT:
1558                 r = sev_guest_init(kvm, &sev_cmd);
1559                 break;
1560         case KVM_SEV_LAUNCH_START:
1561                 r = sev_launch_start(kvm, &sev_cmd);
1562                 break;
1563         case KVM_SEV_LAUNCH_UPDATE_DATA:
1564                 r = sev_launch_update_data(kvm, &sev_cmd);
1565                 break;
1566         case KVM_SEV_LAUNCH_UPDATE_VMSA:
1567                 r = sev_launch_update_vmsa(kvm, &sev_cmd);
1568                 break;
1569         case KVM_SEV_LAUNCH_MEASURE:
1570                 r = sev_launch_measure(kvm, &sev_cmd);
1571                 break;
1572         case KVM_SEV_LAUNCH_FINISH:
1573                 r = sev_launch_finish(kvm, &sev_cmd);
1574                 break;
1575         case KVM_SEV_GUEST_STATUS:
1576                 r = sev_guest_status(kvm, &sev_cmd);
1577                 break;
1578         case KVM_SEV_DBG_DECRYPT:
1579                 r = sev_dbg_crypt(kvm, &sev_cmd, true);
1580                 break;
1581         case KVM_SEV_DBG_ENCRYPT:
1582                 r = sev_dbg_crypt(kvm, &sev_cmd, false);
1583                 break;
1584         case KVM_SEV_LAUNCH_SECRET:
1585                 r = sev_launch_secret(kvm, &sev_cmd);
1586                 break;
1587         case KVM_SEV_GET_ATTESTATION_REPORT:
1588                 r = sev_get_attestation_report(kvm, &sev_cmd);
1589                 break;
1590         case KVM_SEV_SEND_START:
1591                 r = sev_send_start(kvm, &sev_cmd);
1592                 break;
1593         case KVM_SEV_SEND_UPDATE_DATA:
1594                 r = sev_send_update_data(kvm, &sev_cmd);
1595                 break;
1596         case KVM_SEV_SEND_FINISH:
1597                 r = sev_send_finish(kvm, &sev_cmd);
1598                 break;
1599         case KVM_SEV_SEND_CANCEL:
1600                 r = sev_send_cancel(kvm, &sev_cmd);
1601                 break;
1602         case KVM_SEV_RECEIVE_START:
1603                 r = sev_receive_start(kvm, &sev_cmd);
1604                 break;
1605         case KVM_SEV_RECEIVE_UPDATE_DATA:
1606                 r = sev_receive_update_data(kvm, &sev_cmd);
1607                 break;
1608         case KVM_SEV_RECEIVE_FINISH:
1609                 r = sev_receive_finish(kvm, &sev_cmd);
1610                 break;
1611         default:
1612                 r = -EINVAL;
1613                 goto out;
1614         }
1615
1616         if (copy_to_user(argp, &sev_cmd, sizeof(struct kvm_sev_cmd)))
1617                 r = -EFAULT;
1618
1619 out:
1620         mutex_unlock(&kvm->lock);
1621         return r;
1622 }
1623
1624 int svm_register_enc_region(struct kvm *kvm,
1625                             struct kvm_enc_region *range)
1626 {
1627         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1628         struct enc_region *region;
1629         int ret = 0;
1630
1631         if (!sev_guest(kvm))
1632                 return -ENOTTY;
1633
1634         /* If kvm is mirroring encryption context it isn't responsible for it */
1635         if (is_mirroring_enc_context(kvm))
1636                 return -EINVAL;
1637
1638         if (range->addr > ULONG_MAX || range->size > ULONG_MAX)
1639                 return -EINVAL;
1640
1641         region = kzalloc(sizeof(*region), GFP_KERNEL_ACCOUNT);
1642         if (!region)
1643                 return -ENOMEM;
1644
1645         mutex_lock(&kvm->lock);
1646         region->pages = sev_pin_memory(kvm, range->addr, range->size, &region->npages, 1);
1647         if (IS_ERR(region->pages)) {
1648                 ret = PTR_ERR(region->pages);
1649                 mutex_unlock(&kvm->lock);
1650                 goto e_free;
1651         }
1652
1653         region->uaddr = range->addr;
1654         region->size = range->size;
1655
1656         list_add_tail(&region->list, &sev->regions_list);
1657         mutex_unlock(&kvm->lock);
1658
1659         /*
1660          * The guest may change the memory encryption attribute from C=0 -> C=1
1661          * or vice versa for this memory range. Lets make sure caches are
1662          * flushed to ensure that guest data gets written into memory with
1663          * correct C-bit.
1664          */
1665         sev_clflush_pages(region->pages, region->npages);
1666
1667         return ret;
1668
1669 e_free:
1670         kfree(region);
1671         return ret;
1672 }
1673
1674 static struct enc_region *
1675 find_enc_region(struct kvm *kvm, struct kvm_enc_region *range)
1676 {
1677         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1678         struct list_head *head = &sev->regions_list;
1679         struct enc_region *i;
1680
1681         list_for_each_entry(i, head, list) {
1682                 if (i->uaddr == range->addr &&
1683                     i->size == range->size)
1684                         return i;
1685         }
1686
1687         return NULL;
1688 }
1689
1690 static void __unregister_enc_region_locked(struct kvm *kvm,
1691                                            struct enc_region *region)
1692 {
1693         sev_unpin_memory(kvm, region->pages, region->npages);
1694         list_del(&region->list);
1695         kfree(region);
1696 }
1697
1698 int svm_unregister_enc_region(struct kvm *kvm,
1699                               struct kvm_enc_region *range)
1700 {
1701         struct enc_region *region;
1702         int ret;
1703
1704         /* If kvm is mirroring encryption context it isn't responsible for it */
1705         if (is_mirroring_enc_context(kvm))
1706                 return -EINVAL;
1707
1708         mutex_lock(&kvm->lock);
1709
1710         if (!sev_guest(kvm)) {
1711                 ret = -ENOTTY;
1712                 goto failed;
1713         }
1714
1715         region = find_enc_region(kvm, range);
1716         if (!region) {
1717                 ret = -EINVAL;
1718                 goto failed;
1719         }
1720
1721         /*
1722          * Ensure that all guest tagged cache entries are flushed before
1723          * releasing the pages back to the system for use. CLFLUSH will
1724          * not do this, so issue a WBINVD.
1725          */
1726         wbinvd_on_all_cpus();
1727
1728         __unregister_enc_region_locked(kvm, region);
1729
1730         mutex_unlock(&kvm->lock);
1731         return 0;
1732
1733 failed:
1734         mutex_unlock(&kvm->lock);
1735         return ret;
1736 }
1737
1738 int svm_vm_copy_asid_from(struct kvm *kvm, unsigned int source_fd)
1739 {
1740         struct file *source_kvm_file;
1741         struct kvm *source_kvm;
1742         struct kvm_sev_info source_sev, *mirror_sev;
1743         int ret;
1744
1745         source_kvm_file = fget(source_fd);
1746         if (!file_is_kvm(source_kvm_file)) {
1747                 ret = -EBADF;
1748                 goto e_source_put;
1749         }
1750
1751         source_kvm = source_kvm_file->private_data;
1752         mutex_lock(&source_kvm->lock);
1753
1754         if (!sev_guest(source_kvm)) {
1755                 ret = -EINVAL;
1756                 goto e_source_unlock;
1757         }
1758
1759         /* Mirrors of mirrors should work, but let's not get silly */
1760         if (is_mirroring_enc_context(source_kvm) || source_kvm == kvm) {
1761                 ret = -EINVAL;
1762                 goto e_source_unlock;
1763         }
1764
1765         memcpy(&source_sev, &to_kvm_svm(source_kvm)->sev_info,
1766                sizeof(source_sev));
1767
1768         /*
1769          * The mirror kvm holds an enc_context_owner ref so its asid can't
1770          * disappear until we're done with it
1771          */
1772         kvm_get_kvm(source_kvm);
1773
1774         fput(source_kvm_file);
1775         mutex_unlock(&source_kvm->lock);
1776         mutex_lock(&kvm->lock);
1777
1778         if (sev_guest(kvm)) {
1779                 ret = -EINVAL;
1780                 goto e_mirror_unlock;
1781         }
1782
1783         /* Set enc_context_owner and copy its encryption context over */
1784         mirror_sev = &to_kvm_svm(kvm)->sev_info;
1785         mirror_sev->enc_context_owner = source_kvm;
1786         mirror_sev->active = true;
1787         mirror_sev->asid = source_sev.asid;
1788         mirror_sev->fd = source_sev.fd;
1789         mirror_sev->es_active = source_sev.es_active;
1790         mirror_sev->handle = source_sev.handle;
1791         /*
1792          * Do not copy ap_jump_table. Since the mirror does not share the same
1793          * KVM contexts as the original, and they may have different
1794          * memory-views.
1795          */
1796
1797         mutex_unlock(&kvm->lock);
1798         return 0;
1799
1800 e_mirror_unlock:
1801         mutex_unlock(&kvm->lock);
1802         kvm_put_kvm(source_kvm);
1803         return ret;
1804 e_source_unlock:
1805         mutex_unlock(&source_kvm->lock);
1806 e_source_put:
1807         if (source_kvm_file)
1808                 fput(source_kvm_file);
1809         return ret;
1810 }
1811
1812 void sev_vm_destroy(struct kvm *kvm)
1813 {
1814         struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1815         struct list_head *head = &sev->regions_list;
1816         struct list_head *pos, *q;
1817
1818         if (!sev_guest(kvm))
1819                 return;
1820
1821         /* If this is a mirror_kvm release the enc_context_owner and skip sev cleanup */
1822         if (is_mirroring_enc_context(kvm)) {
1823                 kvm_put_kvm(sev->enc_context_owner);
1824                 return;
1825         }
1826
1827         mutex_lock(&kvm->lock);
1828
1829         /*
1830          * Ensure that all guest tagged cache entries are flushed before
1831          * releasing the pages back to the system for use. CLFLUSH will
1832          * not do this, so issue a WBINVD.
1833          */
1834         wbinvd_on_all_cpus();
1835
1836         /*
1837          * if userspace was terminated before unregistering the memory regions
1838          * then lets unpin all the registered memory.
1839          */
1840         if (!list_empty(head)) {
1841                 list_for_each_safe(pos, q, head) {
1842                         __unregister_enc_region_locked(kvm,
1843                                 list_entry(pos, struct enc_region, list));
1844                         cond_resched();
1845                 }
1846         }
1847
1848         mutex_unlock(&kvm->lock);
1849
1850         sev_unbind_asid(kvm, sev->handle);
1851         sev_asid_free(sev);
1852 }
1853
1854 void __init sev_set_cpu_caps(void)
1855 {
1856         if (!sev_enabled)
1857                 kvm_cpu_cap_clear(X86_FEATURE_SEV);
1858         if (!sev_es_enabled)
1859                 kvm_cpu_cap_clear(X86_FEATURE_SEV_ES);
1860 }
1861
1862 void __init sev_hardware_setup(void)
1863 {
1864 #ifdef CONFIG_KVM_AMD_SEV
1865         unsigned int eax, ebx, ecx, edx, sev_asid_count, sev_es_asid_count;
1866         bool sev_es_supported = false;
1867         bool sev_supported = false;
1868
1869         if (!sev_enabled || !npt_enabled)
1870                 goto out;
1871
1872         /* Does the CPU support SEV? */
1873         if (!boot_cpu_has(X86_FEATURE_SEV))
1874                 goto out;
1875
1876         /* Retrieve SEV CPUID information */
1877         cpuid(0x8000001f, &eax, &ebx, &ecx, &edx);
1878
1879         /* Set encryption bit location for SEV-ES guests */
1880         sev_enc_bit = ebx & 0x3f;
1881
1882         /* Maximum number of encrypted guests supported simultaneously */
1883         max_sev_asid = ecx;
1884         if (!max_sev_asid)
1885                 goto out;
1886
1887         /* Minimum ASID value that should be used for SEV guest */
1888         min_sev_asid = edx;
1889         sev_me_mask = 1UL << (ebx & 0x3f);
1890
1891         /*
1892          * Initialize SEV ASID bitmaps. Allocate space for ASID 0 in the bitmap,
1893          * even though it's never used, so that the bitmap is indexed by the
1894          * actual ASID.
1895          */
1896         nr_asids = max_sev_asid + 1;
1897         sev_asid_bitmap = bitmap_zalloc(nr_asids, GFP_KERNEL);
1898         if (!sev_asid_bitmap)
1899                 goto out;
1900
1901         sev_reclaim_asid_bitmap = bitmap_zalloc(nr_asids, GFP_KERNEL);
1902         if (!sev_reclaim_asid_bitmap) {
1903                 bitmap_free(sev_asid_bitmap);
1904                 sev_asid_bitmap = NULL;
1905                 goto out;
1906         }
1907
1908         sev_asid_count = max_sev_asid - min_sev_asid + 1;
1909         if (misc_cg_set_capacity(MISC_CG_RES_SEV, sev_asid_count))
1910                 goto out;
1911
1912         pr_info("SEV supported: %u ASIDs\n", sev_asid_count);
1913         sev_supported = true;
1914
1915         /* SEV-ES support requested? */
1916         if (!sev_es_enabled)
1917                 goto out;
1918
1919         /* Does the CPU support SEV-ES? */
1920         if (!boot_cpu_has(X86_FEATURE_SEV_ES))
1921                 goto out;
1922
1923         /* Has the system been allocated ASIDs for SEV-ES? */
1924         if (min_sev_asid == 1)
1925                 goto out;
1926
1927         sev_es_asid_count = min_sev_asid - 1;
1928         if (misc_cg_set_capacity(MISC_CG_RES_SEV_ES, sev_es_asid_count))
1929                 goto out;
1930
1931         pr_info("SEV-ES supported: %u ASIDs\n", sev_es_asid_count);
1932         sev_es_supported = true;
1933
1934 out:
1935         sev_enabled = sev_supported;
1936         sev_es_enabled = sev_es_supported;
1937 #endif
1938 }
1939
1940 void sev_hardware_teardown(void)
1941 {
1942         if (!sev_enabled)
1943                 return;
1944
1945         /* No need to take sev_bitmap_lock, all VMs have been destroyed. */
1946         sev_flush_asids(1, max_sev_asid);
1947
1948         bitmap_free(sev_asid_bitmap);
1949         bitmap_free(sev_reclaim_asid_bitmap);
1950
1951         misc_cg_set_capacity(MISC_CG_RES_SEV, 0);
1952         misc_cg_set_capacity(MISC_CG_RES_SEV_ES, 0);
1953 }
1954
1955 int sev_cpu_init(struct svm_cpu_data *sd)
1956 {
1957         if (!sev_enabled)
1958                 return 0;
1959
1960         sd->sev_vmcbs = kcalloc(nr_asids, sizeof(void *), GFP_KERNEL);
1961         if (!sd->sev_vmcbs)
1962                 return -ENOMEM;
1963
1964         return 0;
1965 }
1966
1967 /*
1968  * Pages used by hardware to hold guest encrypted state must be flushed before
1969  * returning them to the system.
1970  */
1971 static void sev_flush_guest_memory(struct vcpu_svm *svm, void *va,
1972                                    unsigned long len)
1973 {
1974         /*
1975          * If hardware enforced cache coherency for encrypted mappings of the
1976          * same physical page is supported, nothing to do.
1977          */
1978         if (boot_cpu_has(X86_FEATURE_SME_COHERENT))
1979                 return;
1980
1981         /*
1982          * If the VM Page Flush MSR is supported, use it to flush the page
1983          * (using the page virtual address and the guest ASID).
1984          */
1985         if (boot_cpu_has(X86_FEATURE_VM_PAGE_FLUSH)) {
1986                 struct kvm_sev_info *sev;
1987                 unsigned long va_start;
1988                 u64 start, stop;
1989
1990                 /* Align start and stop to page boundaries. */
1991                 va_start = (unsigned long)va;
1992                 start = (u64)va_start & PAGE_MASK;
1993                 stop = PAGE_ALIGN((u64)va_start + len);
1994
1995                 if (start < stop) {
1996                         sev = &to_kvm_svm(svm->vcpu.kvm)->sev_info;
1997
1998                         while (start < stop) {
1999                                 wrmsrl(MSR_AMD64_VM_PAGE_FLUSH,
2000                                        start | sev->asid);
2001
2002                                 start += PAGE_SIZE;
2003                         }
2004
2005                         return;
2006                 }
2007
2008                 WARN(1, "Address overflow, using WBINVD\n");
2009         }
2010
2011         /*
2012          * Hardware should always have one of the above features,
2013          * but if not, use WBINVD and issue a warning.
2014          */
2015         WARN_ONCE(1, "Using WBINVD to flush guest memory\n");
2016         wbinvd_on_all_cpus();
2017 }
2018
2019 void sev_free_vcpu(struct kvm_vcpu *vcpu)
2020 {
2021         struct vcpu_svm *svm;
2022
2023         if (!sev_es_guest(vcpu->kvm))
2024                 return;
2025
2026         svm = to_svm(vcpu);
2027
2028         if (vcpu->arch.guest_state_protected)
2029                 sev_flush_guest_memory(svm, svm->vmsa, PAGE_SIZE);
2030         __free_page(virt_to_page(svm->vmsa));
2031
2032         if (svm->ghcb_sa_free)
2033                 kfree(svm->ghcb_sa);
2034 }
2035
2036 static void dump_ghcb(struct vcpu_svm *svm)
2037 {
2038         struct ghcb *ghcb = svm->ghcb;
2039         unsigned int nbits;
2040
2041         /* Re-use the dump_invalid_vmcb module parameter */
2042         if (!dump_invalid_vmcb) {
2043                 pr_warn_ratelimited("set kvm_amd.dump_invalid_vmcb=1 to dump internal KVM state.\n");
2044                 return;
2045         }
2046
2047         nbits = sizeof(ghcb->save.valid_bitmap) * 8;
2048
2049         pr_err("GHCB (GPA=%016llx):\n", svm->vmcb->control.ghcb_gpa);
2050         pr_err("%-20s%016llx is_valid: %u\n", "sw_exit_code",
2051                ghcb->save.sw_exit_code, ghcb_sw_exit_code_is_valid(ghcb));
2052         pr_err("%-20s%016llx is_valid: %u\n", "sw_exit_info_1",
2053                ghcb->save.sw_exit_info_1, ghcb_sw_exit_info_1_is_valid(ghcb));
2054         pr_err("%-20s%016llx is_valid: %u\n", "sw_exit_info_2",
2055                ghcb->save.sw_exit_info_2, ghcb_sw_exit_info_2_is_valid(ghcb));
2056         pr_err("%-20s%016llx is_valid: %u\n", "sw_scratch",
2057                ghcb->save.sw_scratch, ghcb_sw_scratch_is_valid(ghcb));
2058         pr_err("%-20s%*pb\n", "valid_bitmap", nbits, ghcb->save.valid_bitmap);
2059 }
2060
2061 static void sev_es_sync_to_ghcb(struct vcpu_svm *svm)
2062 {
2063         struct kvm_vcpu *vcpu = &svm->vcpu;
2064         struct ghcb *ghcb = svm->ghcb;
2065
2066         /*
2067          * The GHCB protocol so far allows for the following data
2068          * to be returned:
2069          *   GPRs RAX, RBX, RCX, RDX
2070          *
2071          * Copy their values, even if they may not have been written during the
2072          * VM-Exit.  It's the guest's responsibility to not consume random data.
2073          */
2074         ghcb_set_rax(ghcb, vcpu->arch.regs[VCPU_REGS_RAX]);
2075         ghcb_set_rbx(ghcb, vcpu->arch.regs[VCPU_REGS_RBX]);
2076         ghcb_set_rcx(ghcb, vcpu->arch.regs[VCPU_REGS_RCX]);
2077         ghcb_set_rdx(ghcb, vcpu->arch.regs[VCPU_REGS_RDX]);
2078 }
2079
2080 static void sev_es_sync_from_ghcb(struct vcpu_svm *svm)
2081 {
2082         struct vmcb_control_area *control = &svm->vmcb->control;
2083         struct kvm_vcpu *vcpu = &svm->vcpu;
2084         struct ghcb *ghcb = svm->ghcb;
2085         u64 exit_code;
2086
2087         /*
2088          * The GHCB protocol so far allows for the following data
2089          * to be supplied:
2090          *   GPRs RAX, RBX, RCX, RDX
2091          *   XCR0
2092          *   CPL
2093          *
2094          * VMMCALL allows the guest to provide extra registers. KVM also
2095          * expects RSI for hypercalls, so include that, too.
2096          *
2097          * Copy their values to the appropriate location if supplied.
2098          */
2099         memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
2100
2101         vcpu->arch.regs[VCPU_REGS_RAX] = ghcb_get_rax_if_valid(ghcb);
2102         vcpu->arch.regs[VCPU_REGS_RBX] = ghcb_get_rbx_if_valid(ghcb);
2103         vcpu->arch.regs[VCPU_REGS_RCX] = ghcb_get_rcx_if_valid(ghcb);
2104         vcpu->arch.regs[VCPU_REGS_RDX] = ghcb_get_rdx_if_valid(ghcb);
2105         vcpu->arch.regs[VCPU_REGS_RSI] = ghcb_get_rsi_if_valid(ghcb);
2106
2107         svm->vmcb->save.cpl = ghcb_get_cpl_if_valid(ghcb);
2108
2109         if (ghcb_xcr0_is_valid(ghcb)) {
2110                 vcpu->arch.xcr0 = ghcb_get_xcr0(ghcb);
2111                 kvm_update_cpuid_runtime(vcpu);
2112         }
2113
2114         /* Copy the GHCB exit information into the VMCB fields */
2115         exit_code = ghcb_get_sw_exit_code(ghcb);
2116         control->exit_code = lower_32_bits(exit_code);
2117         control->exit_code_hi = upper_32_bits(exit_code);
2118         control->exit_info_1 = ghcb_get_sw_exit_info_1(ghcb);
2119         control->exit_info_2 = ghcb_get_sw_exit_info_2(ghcb);
2120
2121         /* Clear the valid entries fields */
2122         memset(ghcb->save.valid_bitmap, 0, sizeof(ghcb->save.valid_bitmap));
2123 }
2124
2125 static int sev_es_validate_vmgexit(struct vcpu_svm *svm)
2126 {
2127         struct kvm_vcpu *vcpu;
2128         struct ghcb *ghcb;
2129         u64 exit_code = 0;
2130
2131         ghcb = svm->ghcb;
2132
2133         /* Only GHCB Usage code 0 is supported */
2134         if (ghcb->ghcb_usage)
2135                 goto vmgexit_err;
2136
2137         /*
2138          * Retrieve the exit code now even though is may not be marked valid
2139          * as it could help with debugging.
2140          */
2141         exit_code = ghcb_get_sw_exit_code(ghcb);
2142
2143         if (!ghcb_sw_exit_code_is_valid(ghcb) ||
2144             !ghcb_sw_exit_info_1_is_valid(ghcb) ||
2145             !ghcb_sw_exit_info_2_is_valid(ghcb))
2146                 goto vmgexit_err;
2147
2148         switch (ghcb_get_sw_exit_code(ghcb)) {
2149         case SVM_EXIT_READ_DR7:
2150                 break;
2151         case SVM_EXIT_WRITE_DR7:
2152                 if (!ghcb_rax_is_valid(ghcb))
2153                         goto vmgexit_err;
2154                 break;
2155         case SVM_EXIT_RDTSC:
2156                 break;
2157         case SVM_EXIT_RDPMC:
2158                 if (!ghcb_rcx_is_valid(ghcb))
2159                         goto vmgexit_err;
2160                 break;
2161         case SVM_EXIT_CPUID:
2162                 if (!ghcb_rax_is_valid(ghcb) ||
2163                     !ghcb_rcx_is_valid(ghcb))
2164                         goto vmgexit_err;
2165                 if (ghcb_get_rax(ghcb) == 0xd)
2166                         if (!ghcb_xcr0_is_valid(ghcb))
2167                                 goto vmgexit_err;
2168                 break;
2169         case SVM_EXIT_INVD:
2170                 break;
2171         case SVM_EXIT_IOIO:
2172                 if (ghcb_get_sw_exit_info_1(ghcb) & SVM_IOIO_STR_MASK) {
2173                         if (!ghcb_sw_scratch_is_valid(ghcb))
2174                                 goto vmgexit_err;
2175                 } else {
2176                         if (!(ghcb_get_sw_exit_info_1(ghcb) & SVM_IOIO_TYPE_MASK))
2177                                 if (!ghcb_rax_is_valid(ghcb))
2178                                         goto vmgexit_err;
2179                 }
2180                 break;
2181         case SVM_EXIT_MSR:
2182                 if (!ghcb_rcx_is_valid(ghcb))
2183                         goto vmgexit_err;
2184                 if (ghcb_get_sw_exit_info_1(ghcb)) {
2185                         if (!ghcb_rax_is_valid(ghcb) ||
2186                             !ghcb_rdx_is_valid(ghcb))
2187                                 goto vmgexit_err;
2188                 }
2189                 break;
2190         case SVM_EXIT_VMMCALL:
2191                 if (!ghcb_rax_is_valid(ghcb) ||
2192                     !ghcb_cpl_is_valid(ghcb))
2193                         goto vmgexit_err;
2194                 break;
2195         case SVM_EXIT_RDTSCP:
2196                 break;
2197         case SVM_EXIT_WBINVD:
2198                 break;
2199         case SVM_EXIT_MONITOR:
2200                 if (!ghcb_rax_is_valid(ghcb) ||
2201                     !ghcb_rcx_is_valid(ghcb) ||
2202                     !ghcb_rdx_is_valid(ghcb))
2203                         goto vmgexit_err;
2204                 break;
2205         case SVM_EXIT_MWAIT:
2206                 if (!ghcb_rax_is_valid(ghcb) ||
2207                     !ghcb_rcx_is_valid(ghcb))
2208                         goto vmgexit_err;
2209                 break;
2210         case SVM_VMGEXIT_MMIO_READ:
2211         case SVM_VMGEXIT_MMIO_WRITE:
2212                 if (!ghcb_sw_scratch_is_valid(ghcb))
2213                         goto vmgexit_err;
2214                 break;
2215         case SVM_VMGEXIT_NMI_COMPLETE:
2216         case SVM_VMGEXIT_AP_HLT_LOOP:
2217         case SVM_VMGEXIT_AP_JUMP_TABLE:
2218         case SVM_VMGEXIT_UNSUPPORTED_EVENT:
2219                 break;
2220         default:
2221                 goto vmgexit_err;
2222         }
2223
2224         return 0;
2225
2226 vmgexit_err:
2227         vcpu = &svm->vcpu;
2228
2229         if (ghcb->ghcb_usage) {
2230                 vcpu_unimpl(vcpu, "vmgexit: ghcb usage %#x is not valid\n",
2231                             ghcb->ghcb_usage);
2232         } else {
2233                 vcpu_unimpl(vcpu, "vmgexit: exit reason %#llx is not valid\n",
2234                             exit_code);
2235                 dump_ghcb(svm);
2236         }
2237
2238         vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2239         vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON;
2240         vcpu->run->internal.ndata = 2;
2241         vcpu->run->internal.data[0] = exit_code;
2242         vcpu->run->internal.data[1] = vcpu->arch.last_vmentry_cpu;
2243
2244         return -EINVAL;
2245 }
2246
2247 void sev_es_unmap_ghcb(struct vcpu_svm *svm)
2248 {
2249         if (!svm->ghcb)
2250                 return;
2251
2252         if (svm->ghcb_sa_free) {
2253                 /*
2254                  * The scratch area lives outside the GHCB, so there is a
2255                  * buffer that, depending on the operation performed, may
2256                  * need to be synced, then freed.
2257                  */
2258                 if (svm->ghcb_sa_sync) {
2259                         kvm_write_guest(svm->vcpu.kvm,
2260                                         ghcb_get_sw_scratch(svm->ghcb),
2261                                         svm->ghcb_sa, svm->ghcb_sa_len);
2262                         svm->ghcb_sa_sync = false;
2263                 }
2264
2265                 kfree(svm->ghcb_sa);
2266                 svm->ghcb_sa = NULL;
2267                 svm->ghcb_sa_free = false;
2268         }
2269
2270         trace_kvm_vmgexit_exit(svm->vcpu.vcpu_id, svm->ghcb);
2271
2272         sev_es_sync_to_ghcb(svm);
2273
2274         kvm_vcpu_unmap(&svm->vcpu, &svm->ghcb_map, true);
2275         svm->ghcb = NULL;
2276 }
2277
2278 void pre_sev_run(struct vcpu_svm *svm, int cpu)
2279 {
2280         struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
2281         int asid = sev_get_asid(svm->vcpu.kvm);
2282
2283         /* Assign the asid allocated with this SEV guest */
2284         svm->asid = asid;
2285
2286         /*
2287          * Flush guest TLB:
2288          *
2289          * 1) when different VMCB for the same ASID is to be run on the same host CPU.
2290          * 2) or this VMCB was executed on different host CPU in previous VMRUNs.
2291          */
2292         if (sd->sev_vmcbs[asid] == svm->vmcb &&
2293             svm->vcpu.arch.last_vmentry_cpu == cpu)
2294                 return;
2295
2296         sd->sev_vmcbs[asid] = svm->vmcb;
2297         svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
2298         vmcb_mark_dirty(svm->vmcb, VMCB_ASID);
2299 }
2300
2301 #define GHCB_SCRATCH_AREA_LIMIT         (16ULL * PAGE_SIZE)
2302 static bool setup_vmgexit_scratch(struct vcpu_svm *svm, bool sync, u64 len)
2303 {
2304         struct vmcb_control_area *control = &svm->vmcb->control;
2305         struct ghcb *ghcb = svm->ghcb;
2306         u64 ghcb_scratch_beg, ghcb_scratch_end;
2307         u64 scratch_gpa_beg, scratch_gpa_end;
2308         void *scratch_va;
2309
2310         scratch_gpa_beg = ghcb_get_sw_scratch(ghcb);
2311         if (!scratch_gpa_beg) {
2312                 pr_err("vmgexit: scratch gpa not provided\n");
2313                 return false;
2314         }
2315
2316         scratch_gpa_end = scratch_gpa_beg + len;
2317         if (scratch_gpa_end < scratch_gpa_beg) {
2318                 pr_err("vmgexit: scratch length (%#llx) not valid for scratch address (%#llx)\n",
2319                        len, scratch_gpa_beg);
2320                 return false;
2321         }
2322
2323         if ((scratch_gpa_beg & PAGE_MASK) == control->ghcb_gpa) {
2324                 /* Scratch area begins within GHCB */
2325                 ghcb_scratch_beg = control->ghcb_gpa +
2326                                    offsetof(struct ghcb, shared_buffer);
2327                 ghcb_scratch_end = control->ghcb_gpa +
2328                                    offsetof(struct ghcb, reserved_1);
2329
2330                 /*
2331                  * If the scratch area begins within the GHCB, it must be
2332                  * completely contained in the GHCB shared buffer area.
2333                  */
2334                 if (scratch_gpa_beg < ghcb_scratch_beg ||
2335                     scratch_gpa_end > ghcb_scratch_end) {
2336                         pr_err("vmgexit: scratch area is outside of GHCB shared buffer area (%#llx - %#llx)\n",
2337                                scratch_gpa_beg, scratch_gpa_end);
2338                         return false;
2339                 }
2340
2341                 scratch_va = (void *)svm->ghcb;
2342                 scratch_va += (scratch_gpa_beg - control->ghcb_gpa);
2343         } else {
2344                 /*
2345                  * The guest memory must be read into a kernel buffer, so
2346                  * limit the size
2347                  */
2348                 if (len > GHCB_SCRATCH_AREA_LIMIT) {
2349                         pr_err("vmgexit: scratch area exceeds KVM limits (%#llx requested, %#llx limit)\n",
2350                                len, GHCB_SCRATCH_AREA_LIMIT);
2351                         return false;
2352                 }
2353                 scratch_va = kzalloc(len, GFP_KERNEL_ACCOUNT);
2354                 if (!scratch_va)
2355                         return false;
2356
2357                 if (kvm_read_guest(svm->vcpu.kvm, scratch_gpa_beg, scratch_va, len)) {
2358                         /* Unable to copy scratch area from guest */
2359                         pr_err("vmgexit: kvm_read_guest for scratch area failed\n");
2360
2361                         kfree(scratch_va);
2362                         return false;
2363                 }
2364
2365                 /*
2366                  * The scratch area is outside the GHCB. The operation will
2367                  * dictate whether the buffer needs to be synced before running
2368                  * the vCPU next time (i.e. a read was requested so the data
2369                  * must be written back to the guest memory).
2370                  */
2371                 svm->ghcb_sa_sync = sync;
2372                 svm->ghcb_sa_free = true;
2373         }
2374
2375         svm->ghcb_sa = scratch_va;
2376         svm->ghcb_sa_len = len;
2377
2378         return true;
2379 }
2380
2381 static void set_ghcb_msr_bits(struct vcpu_svm *svm, u64 value, u64 mask,
2382                               unsigned int pos)
2383 {
2384         svm->vmcb->control.ghcb_gpa &= ~(mask << pos);
2385         svm->vmcb->control.ghcb_gpa |= (value & mask) << pos;
2386 }
2387
2388 static u64 get_ghcb_msr_bits(struct vcpu_svm *svm, u64 mask, unsigned int pos)
2389 {
2390         return (svm->vmcb->control.ghcb_gpa >> pos) & mask;
2391 }
2392
2393 static void set_ghcb_msr(struct vcpu_svm *svm, u64 value)
2394 {
2395         svm->vmcb->control.ghcb_gpa = value;
2396 }
2397
2398 static int sev_handle_vmgexit_msr_protocol(struct vcpu_svm *svm)
2399 {
2400         struct vmcb_control_area *control = &svm->vmcb->control;
2401         struct kvm_vcpu *vcpu = &svm->vcpu;
2402         u64 ghcb_info;
2403         int ret = 1;
2404
2405         ghcb_info = control->ghcb_gpa & GHCB_MSR_INFO_MASK;
2406
2407         trace_kvm_vmgexit_msr_protocol_enter(svm->vcpu.vcpu_id,
2408                                              control->ghcb_gpa);
2409
2410         switch (ghcb_info) {
2411         case GHCB_MSR_SEV_INFO_REQ:
2412                 set_ghcb_msr(svm, GHCB_MSR_SEV_INFO(GHCB_VERSION_MAX,
2413                                                     GHCB_VERSION_MIN,
2414                                                     sev_enc_bit));
2415                 break;
2416         case GHCB_MSR_CPUID_REQ: {
2417                 u64 cpuid_fn, cpuid_reg, cpuid_value;
2418
2419                 cpuid_fn = get_ghcb_msr_bits(svm,
2420                                              GHCB_MSR_CPUID_FUNC_MASK,
2421                                              GHCB_MSR_CPUID_FUNC_POS);
2422
2423                 /* Initialize the registers needed by the CPUID intercept */
2424                 vcpu->arch.regs[VCPU_REGS_RAX] = cpuid_fn;
2425                 vcpu->arch.regs[VCPU_REGS_RCX] = 0;
2426
2427                 ret = svm_invoke_exit_handler(vcpu, SVM_EXIT_CPUID);
2428                 if (!ret) {
2429                         ret = -EINVAL;
2430                         break;
2431                 }
2432
2433                 cpuid_reg = get_ghcb_msr_bits(svm,
2434                                               GHCB_MSR_CPUID_REG_MASK,
2435                                               GHCB_MSR_CPUID_REG_POS);
2436                 if (cpuid_reg == 0)
2437                         cpuid_value = vcpu->arch.regs[VCPU_REGS_RAX];
2438                 else if (cpuid_reg == 1)
2439                         cpuid_value = vcpu->arch.regs[VCPU_REGS_RBX];
2440                 else if (cpuid_reg == 2)
2441                         cpuid_value = vcpu->arch.regs[VCPU_REGS_RCX];
2442                 else
2443                         cpuid_value = vcpu->arch.regs[VCPU_REGS_RDX];
2444
2445                 set_ghcb_msr_bits(svm, cpuid_value,
2446                                   GHCB_MSR_CPUID_VALUE_MASK,
2447                                   GHCB_MSR_CPUID_VALUE_POS);
2448
2449                 set_ghcb_msr_bits(svm, GHCB_MSR_CPUID_RESP,
2450                                   GHCB_MSR_INFO_MASK,
2451                                   GHCB_MSR_INFO_POS);
2452                 break;
2453         }
2454         case GHCB_MSR_TERM_REQ: {
2455                 u64 reason_set, reason_code;
2456
2457                 reason_set = get_ghcb_msr_bits(svm,
2458                                                GHCB_MSR_TERM_REASON_SET_MASK,
2459                                                GHCB_MSR_TERM_REASON_SET_POS);
2460                 reason_code = get_ghcb_msr_bits(svm,
2461                                                 GHCB_MSR_TERM_REASON_MASK,
2462                                                 GHCB_MSR_TERM_REASON_POS);
2463                 pr_info("SEV-ES guest requested termination: %#llx:%#llx\n",
2464                         reason_set, reason_code);
2465                 fallthrough;
2466         }
2467         default:
2468                 ret = -EINVAL;
2469         }
2470
2471         trace_kvm_vmgexit_msr_protocol_exit(svm->vcpu.vcpu_id,
2472                                             control->ghcb_gpa, ret);
2473
2474         return ret;
2475 }
2476
2477 int sev_handle_vmgexit(struct kvm_vcpu *vcpu)
2478 {
2479         struct vcpu_svm *svm = to_svm(vcpu);
2480         struct vmcb_control_area *control = &svm->vmcb->control;
2481         u64 ghcb_gpa, exit_code;
2482         struct ghcb *ghcb;
2483         int ret;
2484
2485         /* Validate the GHCB */
2486         ghcb_gpa = control->ghcb_gpa;
2487         if (ghcb_gpa & GHCB_MSR_INFO_MASK)
2488                 return sev_handle_vmgexit_msr_protocol(svm);
2489
2490         if (!ghcb_gpa) {
2491                 vcpu_unimpl(vcpu, "vmgexit: GHCB gpa is not set\n");
2492                 return -EINVAL;
2493         }
2494
2495         if (kvm_vcpu_map(vcpu, ghcb_gpa >> PAGE_SHIFT, &svm->ghcb_map)) {
2496                 /* Unable to map GHCB from guest */
2497                 vcpu_unimpl(vcpu, "vmgexit: error mapping GHCB [%#llx] from guest\n",
2498                             ghcb_gpa);
2499                 return -EINVAL;
2500         }
2501
2502         svm->ghcb = svm->ghcb_map.hva;
2503         ghcb = svm->ghcb_map.hva;
2504
2505         trace_kvm_vmgexit_enter(vcpu->vcpu_id, ghcb);
2506
2507         exit_code = ghcb_get_sw_exit_code(ghcb);
2508
2509         ret = sev_es_validate_vmgexit(svm);
2510         if (ret)
2511                 return ret;
2512
2513         sev_es_sync_from_ghcb(svm);
2514         ghcb_set_sw_exit_info_1(ghcb, 0);
2515         ghcb_set_sw_exit_info_2(ghcb, 0);
2516
2517         ret = -EINVAL;
2518         switch (exit_code) {
2519         case SVM_VMGEXIT_MMIO_READ:
2520                 if (!setup_vmgexit_scratch(svm, true, control->exit_info_2))
2521                         break;
2522
2523                 ret = kvm_sev_es_mmio_read(vcpu,
2524                                            control->exit_info_1,
2525                                            control->exit_info_2,
2526                                            svm->ghcb_sa);
2527                 break;
2528         case SVM_VMGEXIT_MMIO_WRITE:
2529                 if (!setup_vmgexit_scratch(svm, false, control->exit_info_2))
2530                         break;
2531
2532                 ret = kvm_sev_es_mmio_write(vcpu,
2533                                             control->exit_info_1,
2534                                             control->exit_info_2,
2535                                             svm->ghcb_sa);
2536                 break;
2537         case SVM_VMGEXIT_NMI_COMPLETE:
2538                 ret = svm_invoke_exit_handler(vcpu, SVM_EXIT_IRET);
2539                 break;
2540         case SVM_VMGEXIT_AP_HLT_LOOP:
2541                 ret = kvm_emulate_ap_reset_hold(vcpu);
2542                 break;
2543         case SVM_VMGEXIT_AP_JUMP_TABLE: {
2544                 struct kvm_sev_info *sev = &to_kvm_svm(vcpu->kvm)->sev_info;
2545
2546                 switch (control->exit_info_1) {
2547                 case 0:
2548                         /* Set AP jump table address */
2549                         sev->ap_jump_table = control->exit_info_2;
2550                         break;
2551                 case 1:
2552                         /* Get AP jump table address */
2553                         ghcb_set_sw_exit_info_2(ghcb, sev->ap_jump_table);
2554                         break;
2555                 default:
2556                         pr_err("svm: vmgexit: unsupported AP jump table request - exit_info_1=%#llx\n",
2557                                control->exit_info_1);
2558                         ghcb_set_sw_exit_info_1(ghcb, 1);
2559                         ghcb_set_sw_exit_info_2(ghcb,
2560                                                 X86_TRAP_UD |
2561                                                 SVM_EVTINJ_TYPE_EXEPT |
2562                                                 SVM_EVTINJ_VALID);
2563                 }
2564
2565                 ret = 1;
2566                 break;
2567         }
2568         case SVM_VMGEXIT_UNSUPPORTED_EVENT:
2569                 vcpu_unimpl(vcpu,
2570                             "vmgexit: unsupported event - exit_info_1=%#llx, exit_info_2=%#llx\n",
2571                             control->exit_info_1, control->exit_info_2);
2572                 break;
2573         default:
2574                 ret = svm_invoke_exit_handler(vcpu, exit_code);
2575         }
2576
2577         return ret;
2578 }
2579
2580 int sev_es_string_io(struct vcpu_svm *svm, int size, unsigned int port, int in)
2581 {
2582         if (!setup_vmgexit_scratch(svm, in, svm->vmcb->control.exit_info_2))
2583                 return -EINVAL;
2584
2585         return kvm_sev_es_string_io(&svm->vcpu, size, port,
2586                                     svm->ghcb_sa, svm->ghcb_sa_len, in);
2587 }
2588
2589 void sev_es_init_vmcb(struct vcpu_svm *svm)
2590 {
2591         struct kvm_vcpu *vcpu = &svm->vcpu;
2592
2593         svm->vmcb->control.nested_ctl |= SVM_NESTED_CTL_SEV_ES_ENABLE;
2594         svm->vmcb->control.virt_ext |= LBR_CTL_ENABLE_MASK;
2595
2596         /*
2597          * An SEV-ES guest requires a VMSA area that is a separate from the
2598          * VMCB page. Do not include the encryption mask on the VMSA physical
2599          * address since hardware will access it using the guest key.
2600          */
2601         svm->vmcb->control.vmsa_pa = __pa(svm->vmsa);
2602
2603         /* Can't intercept CR register access, HV can't modify CR registers */
2604         svm_clr_intercept(svm, INTERCEPT_CR0_READ);
2605         svm_clr_intercept(svm, INTERCEPT_CR4_READ);
2606         svm_clr_intercept(svm, INTERCEPT_CR8_READ);
2607         svm_clr_intercept(svm, INTERCEPT_CR0_WRITE);
2608         svm_clr_intercept(svm, INTERCEPT_CR4_WRITE);
2609         svm_clr_intercept(svm, INTERCEPT_CR8_WRITE);
2610
2611         svm_clr_intercept(svm, INTERCEPT_SELECTIVE_CR0);
2612
2613         /* Track EFER/CR register changes */
2614         svm_set_intercept(svm, TRAP_EFER_WRITE);
2615         svm_set_intercept(svm, TRAP_CR0_WRITE);
2616         svm_set_intercept(svm, TRAP_CR4_WRITE);
2617         svm_set_intercept(svm, TRAP_CR8_WRITE);
2618
2619         /* No support for enable_vmware_backdoor */
2620         clr_exception_intercept(svm, GP_VECTOR);
2621
2622         /* Can't intercept XSETBV, HV can't modify XCR0 directly */
2623         svm_clr_intercept(svm, INTERCEPT_XSETBV);
2624
2625         /* Clear intercepts on selected MSRs */
2626         set_msr_interception(vcpu, svm->msrpm, MSR_EFER, 1, 1);
2627         set_msr_interception(vcpu, svm->msrpm, MSR_IA32_CR_PAT, 1, 1);
2628         set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
2629         set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
2630         set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
2631         set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
2632 }
2633
2634 void sev_es_create_vcpu(struct vcpu_svm *svm)
2635 {
2636         /*
2637          * Set the GHCB MSR value as per the GHCB specification when creating
2638          * a vCPU for an SEV-ES guest.
2639          */
2640         set_ghcb_msr(svm, GHCB_MSR_SEV_INFO(GHCB_VERSION_MAX,
2641                                             GHCB_VERSION_MIN,
2642                                             sev_enc_bit));
2643 }
2644
2645 void sev_es_prepare_guest_switch(struct vcpu_svm *svm, unsigned int cpu)
2646 {
2647         struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
2648         struct vmcb_save_area *hostsa;
2649
2650         /*
2651          * As an SEV-ES guest, hardware will restore the host state on VMEXIT,
2652          * of which one step is to perform a VMLOAD. Since hardware does not
2653          * perform a VMSAVE on VMRUN, the host savearea must be updated.
2654          */
2655         vmsave(__sme_page_pa(sd->save_area));
2656
2657         /* XCR0 is restored on VMEXIT, save the current host value */
2658         hostsa = (struct vmcb_save_area *)(page_address(sd->save_area) + 0x400);
2659         hostsa->xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
2660
2661         /* PKRU is restored on VMEXIT, save the current host value */
2662         hostsa->pkru = read_pkru();
2663
2664         /* MSR_IA32_XSS is restored on VMEXIT, save the currnet host value */
2665         hostsa->xss = host_xss;
2666 }
2667
2668 void sev_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
2669 {
2670         struct vcpu_svm *svm = to_svm(vcpu);
2671
2672         /* First SIPI: Use the values as initially set by the VMM */
2673         if (!svm->received_first_sipi) {
2674                 svm->received_first_sipi = true;
2675                 return;
2676         }
2677
2678         /*
2679          * Subsequent SIPI: Return from an AP Reset Hold VMGEXIT, where
2680          * the guest will set the CS and RIP. Set SW_EXIT_INFO_2 to a
2681          * non-zero value.
2682          */
2683         if (!svm->ghcb)
2684                 return;
2685
2686         ghcb_set_sw_exit_info_2(svm->ghcb, 1);
2687 }