| 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | /* |
| 3 | * AMD Encrypted Register State Support |
| 4 | * |
| 5 | * Author: Joerg Roedel <jroedel@suse.de> |
| 6 | */ |
| 7 | |
| 8 | /* |
| 9 | * misc.h needs to be first because it knows how to include the other kernel |
| 10 | * headers in the pre-decompression code in a way that does not break |
| 11 | * compilation. |
| 12 | */ |
| 13 | #include "misc.h" |
| 14 | |
| 15 | #include <asm/bootparam.h> |
| 16 | #include <asm/pgtable_types.h> |
| 17 | #include <asm/shared/msr.h> |
| 18 | #include <asm/sev.h> |
| 19 | #include <asm/trapnr.h> |
| 20 | #include <asm/trap_pf.h> |
| 21 | #include <asm/msr-index.h> |
| 22 | #include <asm/fpu/xcr.h> |
| 23 | #include <asm/ptrace.h> |
| 24 | #include <asm/svm.h> |
| 25 | #include <asm/cpuid/api.h> |
| 26 | |
| 27 | #include "error.h" |
| 28 | #include "sev.h" |
| 29 | |
| 30 | static struct ghcb boot_ghcb_page __aligned(PAGE_SIZE); |
| 31 | struct ghcb *boot_ghcb; |
| 32 | |
| 33 | #undef __init |
| 34 | #define __init |
| 35 | |
| 36 | #define __BOOT_COMPRESSED |
| 37 | |
| 38 | u8 snp_vmpl; |
| 39 | u16 ghcb_version; |
| 40 | |
| 41 | u64 boot_svsm_caa_pa; |
| 42 | |
| 43 | /* Include code for early handlers */ |
| 44 | #include "../../boot/startup/sev-shared.c" |
| 45 | |
| 46 | static bool sev_snp_enabled(void) |
| 47 | { |
| 48 | return sev_status & MSR_AMD64_SEV_SNP_ENABLED; |
| 49 | } |
| 50 | |
| 51 | void snp_set_page_private(unsigned long paddr) |
| 52 | { |
| 53 | struct psc_desc d = { |
| 54 | SNP_PAGE_STATE_PRIVATE, |
| 55 | (struct svsm_ca *)boot_svsm_caa_pa, |
| 56 | boot_svsm_caa_pa |
| 57 | }; |
| 58 | |
| 59 | if (!sev_snp_enabled()) |
| 60 | return; |
| 61 | |
| 62 | __page_state_change(vaddr: paddr, paddr, desc: &d); |
| 63 | } |
| 64 | |
| 65 | void snp_set_page_shared(unsigned long paddr) |
| 66 | { |
| 67 | struct psc_desc d = { |
| 68 | SNP_PAGE_STATE_SHARED, |
| 69 | (struct svsm_ca *)boot_svsm_caa_pa, |
| 70 | boot_svsm_caa_pa |
| 71 | }; |
| 72 | |
| 73 | if (!sev_snp_enabled()) |
| 74 | return; |
| 75 | |
| 76 | __page_state_change(vaddr: paddr, paddr, desc: &d); |
| 77 | } |
| 78 | |
| 79 | bool early_setup_ghcb(void) |
| 80 | { |
| 81 | if (set_page_decrypted((unsigned long)&boot_ghcb_page)) |
| 82 | return false; |
| 83 | |
| 84 | /* Page is now mapped decrypted, clear it */ |
| 85 | memset(s: &boot_ghcb_page, c: 0, n: sizeof(boot_ghcb_page)); |
| 86 | |
| 87 | boot_ghcb = &boot_ghcb_page; |
| 88 | |
| 89 | /* Initialize lookup tables for the instruction decoder */ |
| 90 | sev_insn_decode_init(); |
| 91 | |
| 92 | /* SNP guest requires the GHCB GPA must be registered */ |
| 93 | if (sev_snp_enabled()) |
| 94 | snp_register_ghcb_early(__pa(&boot_ghcb_page)); |
| 95 | |
| 96 | return true; |
| 97 | } |
| 98 | |
| 99 | void snp_accept_memory(phys_addr_t start, phys_addr_t end) |
| 100 | { |
| 101 | struct psc_desc d = { |
| 102 | SNP_PAGE_STATE_PRIVATE, |
| 103 | (struct svsm_ca *)boot_svsm_caa_pa, |
| 104 | boot_svsm_caa_pa |
| 105 | }; |
| 106 | |
| 107 | for (phys_addr_t pa = start; pa < end; pa += PAGE_SIZE) |
| 108 | __page_state_change(vaddr: pa, paddr: pa, desc: &d); |
| 109 | } |
| 110 | |
| 111 | void sev_es_shutdown_ghcb(void) |
| 112 | { |
| 113 | if (!boot_ghcb) |
| 114 | return; |
| 115 | |
| 116 | if (!sev_es_check_cpu_features()) |
| 117 | error(m: "SEV-ES CPU Features missing." ); |
| 118 | |
| 119 | /* |
| 120 | * This denotes whether to use the GHCB MSR protocol or the GHCB |
| 121 | * shared page to perform a GHCB request. Since the GHCB page is |
| 122 | * being changed to encrypted, it can't be used to perform GHCB |
| 123 | * requests. Clear the boot_ghcb variable so that the GHCB MSR |
| 124 | * protocol is used to change the GHCB page over to an encrypted |
| 125 | * page. |
| 126 | */ |
| 127 | boot_ghcb = NULL; |
| 128 | |
| 129 | /* |
| 130 | * GHCB Page must be flushed from the cache and mapped encrypted again. |
| 131 | * Otherwise the running kernel will see strange cache effects when |
| 132 | * trying to use that page. |
| 133 | */ |
| 134 | if (set_page_encrypted((unsigned long)&boot_ghcb_page)) |
| 135 | error(m: "Can't map GHCB page encrypted" ); |
| 136 | |
| 137 | /* |
| 138 | * GHCB page is mapped encrypted again and flushed from the cache. |
| 139 | * Mark it non-present now to catch bugs when #VC exceptions trigger |
| 140 | * after this point. |
| 141 | */ |
| 142 | if (set_page_non_present((unsigned long)&boot_ghcb_page)) |
| 143 | error(m: "Can't unmap GHCB page" ); |
| 144 | } |
| 145 | |
| 146 | static void __noreturn sev_es_ghcb_terminate(struct ghcb *ghcb, unsigned int set, |
| 147 | unsigned int reason, u64 exit_info_2) |
| 148 | { |
| 149 | u64 exit_info_1 = SVM_VMGEXIT_TERM_REASON(set, reason); |
| 150 | |
| 151 | vc_ghcb_invalidate(ghcb); |
| 152 | ghcb_set_sw_exit_code(ghcb, SVM_VMGEXIT_TERM_REQUEST); |
| 153 | ghcb_set_sw_exit_info_1(ghcb, value: exit_info_1); |
| 154 | ghcb_set_sw_exit_info_2(ghcb, value: exit_info_2); |
| 155 | |
| 156 | sev_es_wr_ghcb_msr(__pa(ghcb)); |
| 157 | VMGEXIT(); |
| 158 | |
| 159 | while (true) |
| 160 | asm volatile("hlt\n" : : : "memory" ); |
| 161 | } |
| 162 | |
| 163 | bool sev_es_check_ghcb_fault(unsigned long address) |
| 164 | { |
| 165 | /* Check whether the fault was on the GHCB page */ |
| 166 | return ((address & PAGE_MASK) == (unsigned long)&boot_ghcb_page); |
| 167 | } |
| 168 | |
| 169 | /* |
| 170 | * SNP_FEATURES_IMPL_REQ is the mask of SNP features that will need |
| 171 | * guest side implementation for proper functioning of the guest. If any |
| 172 | * of these features are enabled in the hypervisor but are lacking guest |
| 173 | * side implementation, the behavior of the guest will be undefined. The |
| 174 | * guest could fail in non-obvious way making it difficult to debug. |
| 175 | * |
| 176 | * As the behavior of reserved feature bits is unknown to be on the |
| 177 | * safe side add them to the required features mask. |
| 178 | */ |
| 179 | #define SNP_FEATURES_IMPL_REQ (MSR_AMD64_SNP_VTOM | \ |
| 180 | MSR_AMD64_SNP_REFLECT_VC | \ |
| 181 | MSR_AMD64_SNP_RESTRICTED_INJ | \ |
| 182 | MSR_AMD64_SNP_ALT_INJ | \ |
| 183 | MSR_AMD64_SNP_DEBUG_SWAP | \ |
| 184 | MSR_AMD64_SNP_VMPL_SSS | \ |
| 185 | MSR_AMD64_SNP_SECURE_TSC | \ |
| 186 | MSR_AMD64_SNP_VMGEXIT_PARAM | \ |
| 187 | MSR_AMD64_SNP_VMSA_REG_PROT | \ |
| 188 | MSR_AMD64_SNP_RESERVED_BIT13 | \ |
| 189 | MSR_AMD64_SNP_RESERVED_BIT15 | \ |
| 190 | MSR_AMD64_SNP_SECURE_AVIC | \ |
| 191 | MSR_AMD64_SNP_RESERVED_MASK) |
| 192 | |
| 193 | #ifdef CONFIG_AMD_SECURE_AVIC |
| 194 | #define SNP_FEATURE_SECURE_AVIC MSR_AMD64_SNP_SECURE_AVIC |
| 195 | #else |
| 196 | #define SNP_FEATURE_SECURE_AVIC 0 |
| 197 | #endif |
| 198 | |
| 199 | /* |
| 200 | * SNP_FEATURES_PRESENT is the mask of SNP features that are implemented |
| 201 | * by the guest kernel. As and when a new feature is implemented in the |
| 202 | * guest kernel, a corresponding bit should be added to the mask. |
| 203 | */ |
| 204 | #define SNP_FEATURES_PRESENT (MSR_AMD64_SNP_DEBUG_SWAP | \ |
| 205 | MSR_AMD64_SNP_SECURE_TSC | \ |
| 206 | SNP_FEATURE_SECURE_AVIC) |
| 207 | |
| 208 | u64 snp_get_unsupported_features(u64 status) |
| 209 | { |
| 210 | if (!(status & MSR_AMD64_SEV_SNP_ENABLED)) |
| 211 | return 0; |
| 212 | |
| 213 | return status & SNP_FEATURES_IMPL_REQ & ~SNP_FEATURES_PRESENT; |
| 214 | } |
| 215 | |
| 216 | void snp_check_features(void) |
| 217 | { |
| 218 | u64 unsupported; |
| 219 | |
| 220 | /* |
| 221 | * Terminate the boot if hypervisor has enabled any feature lacking |
| 222 | * guest side implementation. Pass on the unsupported features mask through |
| 223 | * EXIT_INFO_2 of the GHCB protocol so that those features can be reported |
| 224 | * as part of the guest boot failure. |
| 225 | */ |
| 226 | unsupported = snp_get_unsupported_features(status: sev_status); |
| 227 | if (unsupported) { |
| 228 | if (ghcb_version < 2 || (!boot_ghcb && !early_setup_ghcb())) |
| 229 | sev_es_terminate(SEV_TERM_SET_GEN, GHCB_SNP_UNSUPPORTED); |
| 230 | |
| 231 | sev_es_ghcb_terminate(ghcb: boot_ghcb, SEV_TERM_SET_GEN, |
| 232 | GHCB_SNP_UNSUPPORTED, exit_info_2: unsupported); |
| 233 | } |
| 234 | } |
| 235 | |
| 236 | /* Search for Confidential Computing blob in the EFI config table. */ |
| 237 | static struct cc_blob_sev_info *find_cc_blob_efi(struct boot_params *bp) |
| 238 | { |
| 239 | unsigned long cfg_table_pa; |
| 240 | unsigned int cfg_table_len; |
| 241 | int ret; |
| 242 | |
| 243 | ret = efi_get_conf_table(bp, cfg_tbl_pa: &cfg_table_pa, cfg_tbl_len: &cfg_table_len); |
| 244 | if (ret) |
| 245 | return NULL; |
| 246 | |
| 247 | return (struct cc_blob_sev_info *)efi_find_vendor_table(bp, cfg_tbl_pa: cfg_table_pa, |
| 248 | cfg_tbl_len: cfg_table_len, |
| 249 | EFI_CC_BLOB_GUID); |
| 250 | } |
| 251 | |
| 252 | /* |
| 253 | * Initial set up of SNP relies on information provided by the |
| 254 | * Confidential Computing blob, which can be passed to the boot kernel |
| 255 | * by firmware/bootloader in the following ways: |
| 256 | * |
| 257 | * - via an entry in the EFI config table |
| 258 | * - via a setup_data structure, as defined by the Linux Boot Protocol |
| 259 | * |
| 260 | * Scan for the blob in that order. |
| 261 | */ |
| 262 | static struct cc_blob_sev_info *find_cc_blob(struct boot_params *bp) |
| 263 | { |
| 264 | struct cc_blob_sev_info *cc_info; |
| 265 | |
| 266 | cc_info = find_cc_blob_efi(bp); |
| 267 | if (cc_info) |
| 268 | goto found_cc_info; |
| 269 | |
| 270 | cc_info = find_cc_blob_setup_data(bp); |
| 271 | if (!cc_info) |
| 272 | return NULL; |
| 273 | |
| 274 | found_cc_info: |
| 275 | if (cc_info->magic != CC_BLOB_SEV_HDR_MAGIC) |
| 276 | sev_es_terminate(SEV_TERM_SET_GEN, GHCB_SNP_UNSUPPORTED); |
| 277 | |
| 278 | return cc_info; |
| 279 | } |
| 280 | |
| 281 | /* |
| 282 | * Indicate SNP based on presence of SNP-specific CC blob. Subsequent checks |
| 283 | * will verify the SNP CPUID/MSR bits. |
| 284 | */ |
| 285 | static bool early_snp_init(struct boot_params *bp) |
| 286 | { |
| 287 | struct cc_blob_sev_info *cc_info; |
| 288 | |
| 289 | if (!bp) |
| 290 | return false; |
| 291 | |
| 292 | cc_info = find_cc_blob(bp); |
| 293 | if (!cc_info) |
| 294 | return false; |
| 295 | |
| 296 | /* |
| 297 | * If a SNP-specific Confidential Computing blob is present, then |
| 298 | * firmware/bootloader have indicated SNP support. Verifying this |
| 299 | * involves CPUID checks which will be more reliable if the SNP |
| 300 | * CPUID table is used. See comments over snp_setup_cpuid_table() for |
| 301 | * more details. |
| 302 | */ |
| 303 | setup_cpuid_table(cc_info); |
| 304 | |
| 305 | /* |
| 306 | * Record the SVSM Calling Area (CA) address if the guest is not |
| 307 | * running at VMPL0. The CA will be used to communicate with the |
| 308 | * SVSM and request its services. |
| 309 | */ |
| 310 | svsm_setup_ca(cc_info, page: rip_rel_ptr(p: &boot_ghcb_page)); |
| 311 | |
| 312 | /* |
| 313 | * Pass run-time kernel a pointer to CC info via boot_params so EFI |
| 314 | * config table doesn't need to be searched again during early startup |
| 315 | * phase. |
| 316 | */ |
| 317 | bp->cc_blob_address = (u32)(unsigned long)cc_info; |
| 318 | |
| 319 | return true; |
| 320 | } |
| 321 | |
| 322 | /* |
| 323 | * sev_check_cpu_support - Check for SEV support in the CPU capabilities |
| 324 | * |
| 325 | * Returns < 0 if SEV is not supported, otherwise the position of the |
| 326 | * encryption bit in the page table descriptors. |
| 327 | */ |
| 328 | static int sev_check_cpu_support(void) |
| 329 | { |
| 330 | unsigned int eax, ebx, ecx, edx; |
| 331 | |
| 332 | /* Check for the SME/SEV support leaf */ |
| 333 | eax = 0x80000000; |
| 334 | ecx = 0; |
| 335 | native_cpuid(eax: &eax, ebx: &ebx, ecx: &ecx, edx: &edx); |
| 336 | if (eax < 0x8000001f) |
| 337 | return -ENODEV; |
| 338 | |
| 339 | /* |
| 340 | * Check for the SME/SEV feature: |
| 341 | * CPUID Fn8000_001F[EAX] |
| 342 | * - Bit 0 - Secure Memory Encryption support |
| 343 | * - Bit 1 - Secure Encrypted Virtualization support |
| 344 | * CPUID Fn8000_001F[EBX] |
| 345 | * - Bits 5:0 - Pagetable bit position used to indicate encryption |
| 346 | */ |
| 347 | eax = 0x8000001f; |
| 348 | ecx = 0; |
| 349 | native_cpuid(eax: &eax, ebx: &ebx, ecx: &ecx, edx: &edx); |
| 350 | /* Check whether SEV is supported */ |
| 351 | if (!(eax & BIT(1))) |
| 352 | return -ENODEV; |
| 353 | |
| 354 | sev_snp_needs_sfw = !(ebx & BIT(31)); |
| 355 | |
| 356 | return ebx & 0x3f; |
| 357 | } |
| 358 | |
| 359 | void sev_enable(struct boot_params *bp) |
| 360 | { |
| 361 | struct msr m; |
| 362 | int bitpos; |
| 363 | bool snp; |
| 364 | |
| 365 | /* |
| 366 | * bp->cc_blob_address should only be set by boot/compressed kernel. |
| 367 | * Initialize it to 0 to ensure that uninitialized values from |
| 368 | * buggy bootloaders aren't propagated. |
| 369 | */ |
| 370 | if (bp) |
| 371 | bp->cc_blob_address = 0; |
| 372 | |
| 373 | /* |
| 374 | * Do an initial SEV capability check before early_snp_init() which |
| 375 | * loads the CPUID page and the same checks afterwards are done |
| 376 | * without the hypervisor and are trustworthy. |
| 377 | * |
| 378 | * If the HV fakes SEV support, the guest will crash'n'burn |
| 379 | * which is good enough. |
| 380 | */ |
| 381 | |
| 382 | if (sev_check_cpu_support() < 0) |
| 383 | return; |
| 384 | |
| 385 | /* |
| 386 | * Setup/preliminary detection of SNP. This will be sanity-checked |
| 387 | * against CPUID/MSR values later. |
| 388 | */ |
| 389 | snp = early_snp_init(bp); |
| 390 | |
| 391 | /* Now repeat the checks with the SNP CPUID table. */ |
| 392 | |
| 393 | bitpos = sev_check_cpu_support(); |
| 394 | if (bitpos < 0) { |
| 395 | if (snp) |
| 396 | error(m: "SEV-SNP support indicated by CC blob, but not CPUID." ); |
| 397 | return; |
| 398 | } |
| 399 | |
| 400 | /* Set the SME mask if this is an SEV guest. */ |
| 401 | raw_rdmsr(MSR_AMD64_SEV, m: &m); |
| 402 | sev_status = m.q; |
| 403 | if (!(sev_status & MSR_AMD64_SEV_ENABLED)) |
| 404 | return; |
| 405 | |
| 406 | /* Negotiate the GHCB protocol version. */ |
| 407 | if (sev_status & MSR_AMD64_SEV_ES_ENABLED) { |
| 408 | if (!sev_es_negotiate_protocol()) |
| 409 | sev_es_terminate(SEV_TERM_SET_GEN, GHCB_SEV_ES_PROT_UNSUPPORTED); |
| 410 | } |
| 411 | |
| 412 | /* |
| 413 | * SNP is supported in v2 of the GHCB spec which mandates support for HV |
| 414 | * features. |
| 415 | */ |
| 416 | if (sev_status & MSR_AMD64_SEV_SNP_ENABLED) { |
| 417 | u64 hv_features; |
| 418 | |
| 419 | hv_features = get_hv_features(); |
| 420 | if (!(hv_features & GHCB_HV_FT_SNP)) |
| 421 | sev_es_terminate(SEV_TERM_SET_GEN, GHCB_SNP_UNSUPPORTED); |
| 422 | |
| 423 | /* |
| 424 | * Running at VMPL0 is required unless an SVSM is present and |
| 425 | * the hypervisor supports the required SVSM GHCB events. |
| 426 | */ |
| 427 | if (snp_vmpl && !(hv_features & GHCB_HV_FT_SNP_MULTI_VMPL)) |
| 428 | sev_es_terminate(SEV_TERM_SET_LINUX, GHCB_TERM_NOT_VMPL0); |
| 429 | } |
| 430 | |
| 431 | if (snp && !(sev_status & MSR_AMD64_SEV_SNP_ENABLED)) |
| 432 | error(m: "SEV-SNP supported indicated by CC blob, but not SEV status MSR." ); |
| 433 | |
| 434 | sme_me_mask = BIT_ULL(bitpos); |
| 435 | } |
| 436 | |
| 437 | /* |
| 438 | * sev_get_status - Retrieve the SEV status mask |
| 439 | * |
| 440 | * Returns 0 if the CPU is not SEV capable, otherwise the value of the |
| 441 | * AMD64_SEV MSR. |
| 442 | */ |
| 443 | u64 sev_get_status(void) |
| 444 | { |
| 445 | struct msr m; |
| 446 | |
| 447 | if (sev_check_cpu_support() < 0) |
| 448 | return 0; |
| 449 | |
| 450 | raw_rdmsr(MSR_AMD64_SEV, m: &m); |
| 451 | return m.q; |
| 452 | } |
| 453 | |
| 454 | void sev_prep_identity_maps(unsigned long top_level_pgt) |
| 455 | { |
| 456 | /* |
| 457 | * The Confidential Computing blob is used very early in uncompressed |
| 458 | * kernel to find the in-memory CPUID table to handle CPUID |
| 459 | * instructions. Make sure an identity-mapping exists so it can be |
| 460 | * accessed after switchover. |
| 461 | */ |
| 462 | if (sev_snp_enabled()) { |
| 463 | unsigned long cc_info_pa = boot_params_ptr->cc_blob_address; |
| 464 | struct cc_blob_sev_info *cc_info; |
| 465 | |
| 466 | kernel_add_identity_map(start: cc_info_pa, end: cc_info_pa + sizeof(*cc_info)); |
| 467 | |
| 468 | cc_info = (struct cc_blob_sev_info *)cc_info_pa; |
| 469 | kernel_add_identity_map(start: cc_info->cpuid_phys, end: cc_info->cpuid_phys + cc_info->cpuid_len); |
| 470 | } |
| 471 | |
| 472 | sev_verify_cbit(cr3: top_level_pgt); |
| 473 | } |
| 474 | |
| 475 | bool early_is_sevsnp_guest(void) |
| 476 | { |
| 477 | static bool sevsnp; |
| 478 | |
| 479 | if (sevsnp) |
| 480 | return true; |
| 481 | |
| 482 | if (!(sev_get_status() & MSR_AMD64_SEV_SNP_ENABLED)) |
| 483 | return false; |
| 484 | |
| 485 | sevsnp = true; |
| 486 | |
| 487 | if (!snp_vmpl) { |
| 488 | unsigned int eax, ebx, ecx, edx; |
| 489 | |
| 490 | /* |
| 491 | * CPUID Fn8000_001F_EAX[28] - SVSM support |
| 492 | */ |
| 493 | eax = 0x8000001f; |
| 494 | ecx = 0; |
| 495 | native_cpuid(eax: &eax, ebx: &ebx, ecx: &ecx, edx: &edx); |
| 496 | if (eax & BIT(28)) { |
| 497 | struct msr m; |
| 498 | |
| 499 | /* Obtain the address of the calling area to use */ |
| 500 | raw_rdmsr(MSR_SVSM_CAA, m: &m); |
| 501 | boot_svsm_caa_pa = m.q; |
| 502 | |
| 503 | /* |
| 504 | * The real VMPL level cannot be discovered, but the |
| 505 | * memory acceptance routines make no use of that so |
| 506 | * any non-zero value suffices here. |
| 507 | */ |
| 508 | snp_vmpl = U8_MAX; |
| 509 | } |
| 510 | } |
| 511 | return true; |
| 512 | } |
| 513 | |