| 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | /* |
| 3 | * Hyper-V Isolation VM interface with paravisor and hypervisor |
| 4 | * |
| 5 | * Author: |
| 6 | * Tianyu Lan <Tianyu.Lan@microsoft.com> |
| 7 | */ |
| 8 | |
| 9 | #include <linux/bitfield.h> |
| 10 | #include <linux/types.h> |
| 11 | #include <linux/slab.h> |
| 12 | #include <linux/cpu.h> |
| 13 | #include <asm/svm.h> |
| 14 | #include <asm/sev.h> |
| 15 | #include <asm/io.h> |
| 16 | #include <asm/coco.h> |
| 17 | #include <asm/mem_encrypt.h> |
| 18 | #include <asm/set_memory.h> |
| 19 | #include <asm/mshyperv.h> |
| 20 | #include <asm/hypervisor.h> |
| 21 | #include <asm/mtrr.h> |
| 22 | #include <asm/io_apic.h> |
| 23 | #include <asm/realmode.h> |
| 24 | #include <asm/e820/api.h> |
| 25 | #include <asm/desc.h> |
| 26 | #include <asm/msr.h> |
| 27 | #include <uapi/asm/vmx.h> |
| 28 | |
| 29 | #ifdef CONFIG_AMD_MEM_ENCRYPT |
| 30 | |
| 31 | #define GHCB_USAGE_HYPERV_CALL 1 |
| 32 | |
| 33 | union hv_ghcb { |
| 34 | struct ghcb ghcb; |
| 35 | struct { |
| 36 | u64 hypercalldata[509]; |
| 37 | u64 outputgpa; |
| 38 | union { |
| 39 | union { |
| 40 | struct { |
| 41 | u32 callcode : 16; |
| 42 | u32 isfast : 1; |
| 43 | u32 reserved1 : 14; |
| 44 | u32 isnested : 1; |
| 45 | u32 countofelements : 12; |
| 46 | u32 reserved2 : 4; |
| 47 | u32 repstartindex : 12; |
| 48 | u32 reserved3 : 4; |
| 49 | }; |
| 50 | u64 asuint64; |
| 51 | } hypercallinput; |
| 52 | union { |
| 53 | struct { |
| 54 | u16 callstatus; |
| 55 | u16 reserved1; |
| 56 | u32 elementsprocessed : 12; |
| 57 | u32 reserved2 : 20; |
| 58 | }; |
| 59 | u64 asunit64; |
| 60 | } hypercalloutput; |
| 61 | }; |
| 62 | u64 reserved2; |
| 63 | } hypercall; |
| 64 | } __packed __aligned(HV_HYP_PAGE_SIZE); |
| 65 | |
| 66 | /* Only used in an SNP VM with the paravisor */ |
| 67 | static u16 hv_ghcb_version __ro_after_init; |
| 68 | |
| 69 | /* Functions only used in an SNP VM with the paravisor go here. */ |
| 70 | u64 hv_ghcb_hypercall(u64 control, void *input, void *output, u32 input_size) |
| 71 | { |
| 72 | union hv_ghcb *hv_ghcb; |
| 73 | void **ghcb_base; |
| 74 | unsigned long flags; |
| 75 | u64 status; |
| 76 | |
| 77 | if (!hv_ghcb_pg) |
| 78 | return -EFAULT; |
| 79 | |
| 80 | WARN_ON(in_nmi()); |
| 81 | |
| 82 | local_irq_save(flags); |
| 83 | ghcb_base = (void **)this_cpu_ptr(hv_ghcb_pg); |
| 84 | hv_ghcb = (union hv_ghcb *)*ghcb_base; |
| 85 | if (!hv_ghcb) { |
| 86 | local_irq_restore(flags); |
| 87 | return -EFAULT; |
| 88 | } |
| 89 | |
| 90 | hv_ghcb->ghcb.protocol_version = GHCB_PROTOCOL_MAX; |
| 91 | hv_ghcb->ghcb.ghcb_usage = GHCB_USAGE_HYPERV_CALL; |
| 92 | |
| 93 | hv_ghcb->hypercall.outputgpa = (u64)output; |
| 94 | hv_ghcb->hypercall.hypercallinput.asuint64 = 0; |
| 95 | hv_ghcb->hypercall.hypercallinput.callcode = control; |
| 96 | |
| 97 | if (input_size) |
| 98 | memcpy(hv_ghcb->hypercall.hypercalldata, input, input_size); |
| 99 | |
| 100 | VMGEXIT(); |
| 101 | |
| 102 | hv_ghcb->ghcb.ghcb_usage = 0xffffffff; |
| 103 | memset(hv_ghcb->ghcb.save.valid_bitmap, 0, |
| 104 | sizeof(hv_ghcb->ghcb.save.valid_bitmap)); |
| 105 | |
| 106 | status = hv_ghcb->hypercall.hypercalloutput.callstatus; |
| 107 | |
| 108 | local_irq_restore(flags); |
| 109 | |
| 110 | return status; |
| 111 | } |
| 112 | |
| 113 | static inline u64 rd_ghcb_msr(void) |
| 114 | { |
| 115 | return native_rdmsrq(MSR_AMD64_SEV_ES_GHCB); |
| 116 | } |
| 117 | |
| 118 | static inline void wr_ghcb_msr(u64 val) |
| 119 | { |
| 120 | native_wrmsrq(MSR_AMD64_SEV_ES_GHCB, val); |
| 121 | } |
| 122 | |
| 123 | static enum es_result hv_ghcb_hv_call(struct ghcb *ghcb, u64 exit_code, |
| 124 | u64 exit_info_1, u64 exit_info_2) |
| 125 | { |
| 126 | /* Fill in protocol and format specifiers */ |
| 127 | ghcb->protocol_version = hv_ghcb_version; |
| 128 | ghcb->ghcb_usage = GHCB_DEFAULT_USAGE; |
| 129 | |
| 130 | ghcb_set_sw_exit_code(ghcb, value: exit_code); |
| 131 | ghcb_set_sw_exit_info_1(ghcb, value: exit_info_1); |
| 132 | ghcb_set_sw_exit_info_2(ghcb, value: exit_info_2); |
| 133 | |
| 134 | VMGEXIT(); |
| 135 | |
| 136 | if (ghcb->save.sw_exit_info_1 & GENMASK_ULL(31, 0)) |
| 137 | return ES_VMM_ERROR; |
| 138 | else |
| 139 | return ES_OK; |
| 140 | } |
| 141 | |
| 142 | void __noreturn hv_ghcb_terminate(unsigned int set, unsigned int reason) |
| 143 | { |
| 144 | u64 val = GHCB_MSR_TERM_REQ; |
| 145 | |
| 146 | /* Tell the hypervisor what went wrong. */ |
| 147 | val |= GHCB_SEV_TERM_REASON(set, reason); |
| 148 | |
| 149 | /* Request Guest Termination from Hypervisor */ |
| 150 | wr_ghcb_msr(val); |
| 151 | VMGEXIT(); |
| 152 | |
| 153 | while (true) |
| 154 | asm volatile("hlt\n" : : : "memory" ); |
| 155 | } |
| 156 | |
| 157 | bool hv_ghcb_negotiate_protocol(void) |
| 158 | { |
| 159 | u64 ghcb_gpa; |
| 160 | u64 val; |
| 161 | |
| 162 | /* Save ghcb page gpa. */ |
| 163 | ghcb_gpa = rd_ghcb_msr(); |
| 164 | |
| 165 | /* Do the GHCB protocol version negotiation */ |
| 166 | wr_ghcb_msr(GHCB_MSR_SEV_INFO_REQ); |
| 167 | VMGEXIT(); |
| 168 | val = rd_ghcb_msr(); |
| 169 | |
| 170 | if (GHCB_MSR_INFO(val) != GHCB_MSR_SEV_INFO_RESP) |
| 171 | return false; |
| 172 | |
| 173 | if (GHCB_MSR_PROTO_MAX(val) < GHCB_PROTOCOL_MIN || |
| 174 | GHCB_MSR_PROTO_MIN(val) > GHCB_PROTOCOL_MAX) |
| 175 | return false; |
| 176 | |
| 177 | hv_ghcb_version = min_t(size_t, GHCB_MSR_PROTO_MAX(val), |
| 178 | GHCB_PROTOCOL_MAX); |
| 179 | |
| 180 | /* Write ghcb page back after negotiating protocol. */ |
| 181 | wr_ghcb_msr(val: ghcb_gpa); |
| 182 | VMGEXIT(); |
| 183 | |
| 184 | return true; |
| 185 | } |
| 186 | |
| 187 | static void hv_ghcb_msr_write(u64 msr, u64 value) |
| 188 | { |
| 189 | union hv_ghcb *hv_ghcb; |
| 190 | void **ghcb_base; |
| 191 | unsigned long flags; |
| 192 | |
| 193 | if (!hv_ghcb_pg) |
| 194 | return; |
| 195 | |
| 196 | WARN_ON(in_nmi()); |
| 197 | |
| 198 | local_irq_save(flags); |
| 199 | ghcb_base = (void **)this_cpu_ptr(hv_ghcb_pg); |
| 200 | hv_ghcb = (union hv_ghcb *)*ghcb_base; |
| 201 | if (!hv_ghcb) { |
| 202 | local_irq_restore(flags); |
| 203 | return; |
| 204 | } |
| 205 | |
| 206 | ghcb_set_rcx(ghcb: &hv_ghcb->ghcb, value: msr); |
| 207 | ghcb_set_rax(ghcb: &hv_ghcb->ghcb, lower_32_bits(value)); |
| 208 | ghcb_set_rdx(ghcb: &hv_ghcb->ghcb, upper_32_bits(value)); |
| 209 | |
| 210 | if (hv_ghcb_hv_call(ghcb: &hv_ghcb->ghcb, SVM_EXIT_MSR, exit_info_1: 1, exit_info_2: 0)) |
| 211 | pr_warn("Fail to write msr via ghcb %llx.\n" , msr); |
| 212 | |
| 213 | local_irq_restore(flags); |
| 214 | } |
| 215 | |
| 216 | static void hv_ghcb_msr_read(u64 msr, u64 *value) |
| 217 | { |
| 218 | union hv_ghcb *hv_ghcb; |
| 219 | void **ghcb_base; |
| 220 | unsigned long flags; |
| 221 | |
| 222 | /* Check size of union hv_ghcb here. */ |
| 223 | BUILD_BUG_ON(sizeof(union hv_ghcb) != HV_HYP_PAGE_SIZE); |
| 224 | |
| 225 | if (!hv_ghcb_pg) |
| 226 | return; |
| 227 | |
| 228 | WARN_ON(in_nmi()); |
| 229 | |
| 230 | local_irq_save(flags); |
| 231 | ghcb_base = (void **)this_cpu_ptr(hv_ghcb_pg); |
| 232 | hv_ghcb = (union hv_ghcb *)*ghcb_base; |
| 233 | if (!hv_ghcb) { |
| 234 | local_irq_restore(flags); |
| 235 | return; |
| 236 | } |
| 237 | |
| 238 | ghcb_set_rcx(ghcb: &hv_ghcb->ghcb, value: msr); |
| 239 | if (hv_ghcb_hv_call(ghcb: &hv_ghcb->ghcb, SVM_EXIT_MSR, exit_info_1: 0, exit_info_2: 0)) |
| 240 | pr_warn("Fail to read msr via ghcb %llx.\n" , msr); |
| 241 | else |
| 242 | *value = (u64)lower_32_bits(hv_ghcb->ghcb.save.rax) |
| 243 | | ((u64)lower_32_bits(hv_ghcb->ghcb.save.rdx) << 32); |
| 244 | local_irq_restore(flags); |
| 245 | } |
| 246 | |
| 247 | /* Only used in a fully enlightened SNP VM, i.e. without the paravisor */ |
| 248 | static u8 ap_start_input_arg[PAGE_SIZE] __bss_decrypted __aligned(PAGE_SIZE); |
| 249 | static u8 ap_start_stack[PAGE_SIZE] __aligned(PAGE_SIZE); |
| 250 | static DEFINE_PER_CPU(struct sev_es_save_area *, hv_sev_vmsa); |
| 251 | |
| 252 | /* Functions only used in an SNP VM without the paravisor go here. */ |
| 253 | |
| 254 | #define hv_populate_vmcb_seg(seg, gdtr_base) \ |
| 255 | do { \ |
| 256 | if (seg.selector) { \ |
| 257 | seg.base = 0; \ |
| 258 | seg.limit = HV_AP_SEGMENT_LIMIT; \ |
| 259 | seg.attrib = *(u16 *)(gdtr_base + seg.selector + 5); \ |
| 260 | seg.attrib = (seg.attrib & 0xFF) | ((seg.attrib >> 4) & 0xF00); \ |
| 261 | } \ |
| 262 | } while (0) \ |
| 263 | |
| 264 | static int snp_set_vmsa(void *va, bool vmsa) |
| 265 | { |
| 266 | u64 attrs; |
| 267 | |
| 268 | /* |
| 269 | * Running at VMPL0 allows the kernel to change the VMSA bit for a page |
| 270 | * using the RMPADJUST instruction. However, for the instruction to |
| 271 | * succeed it must target the permissions of a lesser privileged |
| 272 | * (higher numbered) VMPL level, so use VMPL1 (refer to the RMPADJUST |
| 273 | * instruction in the AMD64 APM Volume 3). |
| 274 | */ |
| 275 | attrs = 1; |
| 276 | if (vmsa) |
| 277 | attrs |= RMPADJUST_VMSA_PAGE_BIT; |
| 278 | |
| 279 | return rmpadjust(vaddr: (unsigned long)va, RMP_PG_SIZE_4K, attrs); |
| 280 | } |
| 281 | |
| 282 | static void snp_cleanup_vmsa(struct sev_es_save_area *vmsa) |
| 283 | { |
| 284 | int err; |
| 285 | |
| 286 | err = snp_set_vmsa(va: vmsa, vmsa: false); |
| 287 | if (err) |
| 288 | pr_err("clear VMSA page failed (%u), leaking page\n" , err); |
| 289 | else |
| 290 | free_page((unsigned long)vmsa); |
| 291 | } |
| 292 | |
| 293 | int hv_snp_boot_ap(u32 apic_id, unsigned long start_ip, unsigned int cpu) |
| 294 | { |
| 295 | struct sev_es_save_area *vmsa = (struct sev_es_save_area *) |
| 296 | __get_free_page(GFP_KERNEL | __GFP_ZERO); |
| 297 | struct sev_es_save_area *cur_vmsa; |
| 298 | struct desc_ptr gdtr; |
| 299 | u64 ret, retry = 5; |
| 300 | struct hv_enable_vp_vtl *start_vp_input; |
| 301 | unsigned long flags; |
| 302 | int vp_index; |
| 303 | |
| 304 | if (!vmsa) |
| 305 | return -ENOMEM; |
| 306 | |
| 307 | /* Find the Hyper-V VP index which might be not the same as APIC ID */ |
| 308 | vp_index = hv_apicid_to_vp_index(apic_id); |
| 309 | if (vp_index < 0 || vp_index > ms_hyperv.max_vp_index) |
| 310 | return -EINVAL; |
| 311 | |
| 312 | native_store_gdt(dtr: &gdtr); |
| 313 | |
| 314 | vmsa->gdtr.base = gdtr.address; |
| 315 | vmsa->gdtr.limit = gdtr.size; |
| 316 | |
| 317 | asm volatile("movl %%es, %%eax;" : "=a" (vmsa->es.selector)); |
| 318 | hv_populate_vmcb_seg(vmsa->es, vmsa->gdtr.base); |
| 319 | |
| 320 | asm volatile("movl %%cs, %%eax;" : "=a" (vmsa->cs.selector)); |
| 321 | hv_populate_vmcb_seg(vmsa->cs, vmsa->gdtr.base); |
| 322 | |
| 323 | asm volatile("movl %%ss, %%eax;" : "=a" (vmsa->ss.selector)); |
| 324 | hv_populate_vmcb_seg(vmsa->ss, vmsa->gdtr.base); |
| 325 | |
| 326 | asm volatile("movl %%ds, %%eax;" : "=a" (vmsa->ds.selector)); |
| 327 | hv_populate_vmcb_seg(vmsa->ds, vmsa->gdtr.base); |
| 328 | |
| 329 | vmsa->efer = native_read_msr(MSR_EFER); |
| 330 | |
| 331 | vmsa->cr4 = native_read_cr4(); |
| 332 | vmsa->cr3 = __native_read_cr3(); |
| 333 | vmsa->cr0 = native_read_cr0(); |
| 334 | |
| 335 | vmsa->xcr0 = 1; |
| 336 | vmsa->g_pat = HV_AP_INIT_GPAT_DEFAULT; |
| 337 | vmsa->rip = (u64)secondary_startup_64_no_verify; |
| 338 | vmsa->rsp = (u64)&ap_start_stack[PAGE_SIZE]; |
| 339 | |
| 340 | /* |
| 341 | * Set the SNP-specific fields for this VMSA: |
| 342 | * VMPL level |
| 343 | * SEV_FEATURES (matches the SEV STATUS MSR right shifted 2 bits) |
| 344 | */ |
| 345 | vmsa->vmpl = 0; |
| 346 | vmsa->sev_features = sev_status >> 2; |
| 347 | |
| 348 | ret = snp_set_vmsa(va: vmsa, vmsa: true); |
| 349 | if (ret) { |
| 350 | pr_err("RMPADJUST(%llx) failed: %llx\n" , (u64)vmsa, ret); |
| 351 | free_page((u64)vmsa); |
| 352 | return ret; |
| 353 | } |
| 354 | |
| 355 | local_irq_save(flags); |
| 356 | start_vp_input = (struct hv_enable_vp_vtl *)ap_start_input_arg; |
| 357 | memset(start_vp_input, 0, sizeof(*start_vp_input)); |
| 358 | start_vp_input->partition_id = -1; |
| 359 | start_vp_input->vp_index = vp_index; |
| 360 | start_vp_input->target_vtl.target_vtl = ms_hyperv.vtl; |
| 361 | *(u64 *)&start_vp_input->vp_context = __pa(vmsa) | 1; |
| 362 | |
| 363 | do { |
| 364 | ret = hv_do_hypercall(HVCALL_START_VP, |
| 365 | inputaddr: start_vp_input, NULL); |
| 366 | } while (hv_result(status: ret) == HV_STATUS_TIME_OUT && retry--); |
| 367 | |
| 368 | local_irq_restore(flags); |
| 369 | |
| 370 | if (!hv_result_success(status: ret)) { |
| 371 | pr_err("HvCallStartVirtualProcessor failed: %llx\n" , ret); |
| 372 | snp_cleanup_vmsa(vmsa); |
| 373 | vmsa = NULL; |
| 374 | } |
| 375 | |
| 376 | cur_vmsa = per_cpu(hv_sev_vmsa, cpu); |
| 377 | /* Free up any previous VMSA page */ |
| 378 | if (cur_vmsa) |
| 379 | snp_cleanup_vmsa(vmsa: cur_vmsa); |
| 380 | |
| 381 | /* Record the current VMSA page */ |
| 382 | per_cpu(hv_sev_vmsa, cpu) = vmsa; |
| 383 | |
| 384 | return ret; |
| 385 | } |
| 386 | |
| 387 | #else |
| 388 | static inline void hv_ghcb_msr_write(u64 msr, u64 value) {} |
| 389 | static inline void hv_ghcb_msr_read(u64 msr, u64 *value) {} |
| 390 | #endif /* CONFIG_AMD_MEM_ENCRYPT */ |
| 391 | |
| 392 | #ifdef CONFIG_INTEL_TDX_GUEST |
| 393 | static void hv_tdx_msr_write(u64 msr, u64 val) |
| 394 | { |
| 395 | struct tdx_module_args args = { |
| 396 | .r10 = TDX_HYPERCALL_STANDARD, |
| 397 | .r11 = EXIT_REASON_MSR_WRITE, |
| 398 | .r12 = msr, |
| 399 | .r13 = val, |
| 400 | }; |
| 401 | |
| 402 | u64 ret = __tdx_hypercall(args: &args); |
| 403 | |
| 404 | WARN_ONCE(ret, "Failed to emulate MSR write: %lld\n" , ret); |
| 405 | } |
| 406 | |
| 407 | static void hv_tdx_msr_read(u64 msr, u64 *val) |
| 408 | { |
| 409 | struct tdx_module_args args = { |
| 410 | .r10 = TDX_HYPERCALL_STANDARD, |
| 411 | .r11 = EXIT_REASON_MSR_READ, |
| 412 | .r12 = msr, |
| 413 | }; |
| 414 | |
| 415 | u64 ret = __tdx_hypercall(args: &args); |
| 416 | |
| 417 | if (WARN_ONCE(ret, "Failed to emulate MSR read: %lld\n" , ret)) |
| 418 | *val = 0; |
| 419 | else |
| 420 | *val = args.r11; |
| 421 | } |
| 422 | |
| 423 | u64 hv_tdx_hypercall(u64 control, u64 param1, u64 param2) |
| 424 | { |
| 425 | struct tdx_module_args args = { }; |
| 426 | |
| 427 | args.r10 = control; |
| 428 | args.rdx = param1; |
| 429 | args.r8 = param2; |
| 430 | |
| 431 | (void)__tdx_hypercall(args: &args); |
| 432 | |
| 433 | return args.r11; |
| 434 | } |
| 435 | |
| 436 | #else |
| 437 | static inline void hv_tdx_msr_write(u64 msr, u64 value) {} |
| 438 | static inline void hv_tdx_msr_read(u64 msr, u64 *value) {} |
| 439 | #endif /* CONFIG_INTEL_TDX_GUEST */ |
| 440 | |
| 441 | #if defined(CONFIG_AMD_MEM_ENCRYPT) || defined(CONFIG_INTEL_TDX_GUEST) |
| 442 | void hv_ivm_msr_write(u64 msr, u64 value) |
| 443 | { |
| 444 | if (!ms_hyperv.paravisor_present) |
| 445 | return; |
| 446 | |
| 447 | if (hv_isolation_type_tdx()) |
| 448 | hv_tdx_msr_write(msr, val: value); |
| 449 | else if (hv_isolation_type_snp()) |
| 450 | hv_ghcb_msr_write(msr, value); |
| 451 | } |
| 452 | |
| 453 | void hv_ivm_msr_read(u64 msr, u64 *value) |
| 454 | { |
| 455 | if (!ms_hyperv.paravisor_present) |
| 456 | return; |
| 457 | |
| 458 | if (hv_isolation_type_tdx()) |
| 459 | hv_tdx_msr_read(msr, val: value); |
| 460 | else if (hv_isolation_type_snp()) |
| 461 | hv_ghcb_msr_read(msr, value); |
| 462 | } |
| 463 | |
| 464 | /* |
| 465 | * hv_mark_gpa_visibility - Set pages visible to host via hvcall. |
| 466 | * |
| 467 | * In Isolation VM, all guest memory is encrypted from host and guest |
| 468 | * needs to set memory visible to host via hvcall before sharing memory |
| 469 | * with host. |
| 470 | */ |
| 471 | static int hv_mark_gpa_visibility(u16 count, const u64 pfn[], |
| 472 | enum hv_mem_host_visibility visibility) |
| 473 | { |
| 474 | struct hv_gpa_range_for_visibility *input; |
| 475 | u64 hv_status; |
| 476 | unsigned long flags; |
| 477 | |
| 478 | /* no-op if partition isolation is not enabled */ |
| 479 | if (!hv_is_isolation_supported()) |
| 480 | return 0; |
| 481 | |
| 482 | if (count > HV_MAX_MODIFY_GPA_REP_COUNT) { |
| 483 | pr_err("Hyper-V: GPA count:%d exceeds supported:%lu\n" , count, |
| 484 | HV_MAX_MODIFY_GPA_REP_COUNT); |
| 485 | return -EINVAL; |
| 486 | } |
| 487 | |
| 488 | local_irq_save(flags); |
| 489 | input = *this_cpu_ptr(hyperv_pcpu_input_arg); |
| 490 | |
| 491 | if (unlikely(!input)) { |
| 492 | local_irq_restore(flags); |
| 493 | return -EINVAL; |
| 494 | } |
| 495 | |
| 496 | input->partition_id = HV_PARTITION_ID_SELF; |
| 497 | input->host_visibility = visibility; |
| 498 | input->reserved0 = 0; |
| 499 | input->reserved1 = 0; |
| 500 | memcpy((void *)input->gpa_page_list, pfn, count * sizeof(*pfn)); |
| 501 | hv_status = hv_do_rep_hypercall( |
| 502 | HVCALL_MODIFY_SPARSE_GPA_PAGE_HOST_VISIBILITY, rep_count: count, |
| 503 | varhead_size: 0, input, NULL); |
| 504 | local_irq_restore(flags); |
| 505 | |
| 506 | if (hv_result_success(status: hv_status)) |
| 507 | return 0; |
| 508 | else |
| 509 | return -EFAULT; |
| 510 | } |
| 511 | |
| 512 | /* |
| 513 | * When transitioning memory between encrypted and decrypted, the caller |
| 514 | * of set_memory_encrypted() or set_memory_decrypted() is responsible for |
| 515 | * ensuring that the memory isn't in use and isn't referenced while the |
| 516 | * transition is in progress. The transition has multiple steps, and the |
| 517 | * memory is in an inconsistent state until all steps are complete. A |
| 518 | * reference while the state is inconsistent could result in an exception |
| 519 | * that can't be cleanly fixed up. |
| 520 | * |
| 521 | * But the Linux kernel load_unaligned_zeropad() mechanism could cause a |
| 522 | * stray reference that can't be prevented by the caller, so Linux has |
| 523 | * specific code to handle this case. But when the #VC and #VE exceptions |
| 524 | * routed to a paravisor, the specific code doesn't work. To avoid this |
| 525 | * problem, mark the pages as "not present" while the transition is in |
| 526 | * progress. If load_unaligned_zeropad() causes a stray reference, a normal |
| 527 | * page fault is generated instead of #VC or #VE, and the page-fault-based |
| 528 | * handlers for load_unaligned_zeropad() resolve the reference. When the |
| 529 | * transition is complete, hv_vtom_set_host_visibility() marks the pages |
| 530 | * as "present" again. |
| 531 | */ |
| 532 | static int hv_vtom_clear_present(unsigned long kbuffer, int pagecount, bool enc) |
| 533 | { |
| 534 | return set_memory_np(addr: kbuffer, numpages: pagecount); |
| 535 | } |
| 536 | |
| 537 | /* |
| 538 | * hv_vtom_set_host_visibility - Set specified memory visible to host. |
| 539 | * |
| 540 | * In Isolation VM, all guest memory is encrypted from host and guest |
| 541 | * needs to set memory visible to host via hvcall before sharing memory |
| 542 | * with host. This function works as wrap of hv_mark_gpa_visibility() |
| 543 | * with memory base and size. |
| 544 | */ |
| 545 | static int hv_vtom_set_host_visibility(unsigned long kbuffer, int pagecount, bool enc) |
| 546 | { |
| 547 | enum hv_mem_host_visibility visibility = enc ? |
| 548 | VMBUS_PAGE_NOT_VISIBLE : VMBUS_PAGE_VISIBLE_READ_WRITE; |
| 549 | u64 *pfn_array; |
| 550 | phys_addr_t paddr; |
| 551 | int i, pfn, err; |
| 552 | void *vaddr; |
| 553 | int ret = 0; |
| 554 | |
| 555 | pfn_array = kmalloc(HV_HYP_PAGE_SIZE, GFP_KERNEL); |
| 556 | if (!pfn_array) { |
| 557 | ret = -ENOMEM; |
| 558 | goto err_set_memory_p; |
| 559 | } |
| 560 | |
| 561 | for (i = 0, pfn = 0; i < pagecount; i++) { |
| 562 | /* |
| 563 | * Use slow_virt_to_phys() because the PRESENT bit has been |
| 564 | * temporarily cleared in the PTEs. slow_virt_to_phys() works |
| 565 | * without the PRESENT bit while virt_to_hvpfn() or similar |
| 566 | * does not. |
| 567 | */ |
| 568 | vaddr = (void *)kbuffer + (i * HV_HYP_PAGE_SIZE); |
| 569 | paddr = slow_virt_to_phys(address: vaddr); |
| 570 | pfn_array[pfn] = paddr >> HV_HYP_PAGE_SHIFT; |
| 571 | pfn++; |
| 572 | |
| 573 | if (pfn == HV_MAX_MODIFY_GPA_REP_COUNT || i == pagecount - 1) { |
| 574 | ret = hv_mark_gpa_visibility(count: pfn, pfn: pfn_array, |
| 575 | visibility); |
| 576 | if (ret) |
| 577 | goto err_free_pfn_array; |
| 578 | pfn = 0; |
| 579 | } |
| 580 | } |
| 581 | |
| 582 | err_free_pfn_array: |
| 583 | kfree(objp: pfn_array); |
| 584 | |
| 585 | err_set_memory_p: |
| 586 | /* |
| 587 | * Set the PTE PRESENT bits again to revert what hv_vtom_clear_present() |
| 588 | * did. Do this even if there is an error earlier in this function in |
| 589 | * order to avoid leaving the memory range in a "broken" state. Setting |
| 590 | * the PRESENT bits shouldn't fail, but return an error if it does. |
| 591 | */ |
| 592 | err = set_memory_p(addr: kbuffer, numpages: pagecount); |
| 593 | if (err && !ret) |
| 594 | ret = err; |
| 595 | |
| 596 | return ret; |
| 597 | } |
| 598 | |
| 599 | static bool hv_vtom_tlb_flush_required(bool private) |
| 600 | { |
| 601 | /* |
| 602 | * Since hv_vtom_clear_present() marks the PTEs as "not present" |
| 603 | * and flushes the TLB, they can't be in the TLB. That makes the |
| 604 | * flush controlled by this function redundant, so return "false". |
| 605 | */ |
| 606 | return false; |
| 607 | } |
| 608 | |
| 609 | static bool hv_vtom_cache_flush_required(void) |
| 610 | { |
| 611 | return false; |
| 612 | } |
| 613 | |
| 614 | static bool hv_is_private_mmio(u64 addr) |
| 615 | { |
| 616 | /* |
| 617 | * Hyper-V always provides a single IO-APIC in a guest VM. |
| 618 | * When a paravisor is used, it is emulated by the paravisor |
| 619 | * in the guest context and must be mapped private. |
| 620 | */ |
| 621 | if (addr >= HV_IOAPIC_BASE_ADDRESS && |
| 622 | addr < (HV_IOAPIC_BASE_ADDRESS + PAGE_SIZE)) |
| 623 | return true; |
| 624 | |
| 625 | /* Same with a vTPM */ |
| 626 | if (addr >= VTPM_BASE_ADDRESS && |
| 627 | addr < (VTPM_BASE_ADDRESS + PAGE_SIZE)) |
| 628 | return true; |
| 629 | |
| 630 | return false; |
| 631 | } |
| 632 | |
| 633 | void __init hv_vtom_init(void) |
| 634 | { |
| 635 | enum hv_isolation_type type = hv_get_isolation_type(); |
| 636 | |
| 637 | switch (type) { |
| 638 | case HV_ISOLATION_TYPE_VBS: |
| 639 | fallthrough; |
| 640 | /* |
| 641 | * By design, a VM using vTOM doesn't see the SEV setting, |
| 642 | * so SEV initialization is bypassed and sev_status isn't set. |
| 643 | * Set it here to indicate a vTOM VM. |
| 644 | * |
| 645 | * Note: if CONFIG_AMD_MEM_ENCRYPT is not set, sev_status is |
| 646 | * defined as 0ULL, to which we can't assigned a value. |
| 647 | */ |
| 648 | #ifdef CONFIG_AMD_MEM_ENCRYPT |
| 649 | case HV_ISOLATION_TYPE_SNP: |
| 650 | sev_status = MSR_AMD64_SNP_VTOM; |
| 651 | cc_vendor = CC_VENDOR_AMD; |
| 652 | break; |
| 653 | #endif |
| 654 | |
| 655 | case HV_ISOLATION_TYPE_TDX: |
| 656 | cc_vendor = CC_VENDOR_INTEL; |
| 657 | break; |
| 658 | |
| 659 | default: |
| 660 | panic(fmt: "hv_vtom_init: unsupported isolation type %d\n" , type); |
| 661 | } |
| 662 | |
| 663 | cc_set_mask(mask: ms_hyperv.shared_gpa_boundary); |
| 664 | physical_mask &= ms_hyperv.shared_gpa_boundary - 1; |
| 665 | |
| 666 | x86_platform.hyper.is_private_mmio = hv_is_private_mmio; |
| 667 | x86_platform.guest.enc_cache_flush_required = hv_vtom_cache_flush_required; |
| 668 | x86_platform.guest.enc_tlb_flush_required = hv_vtom_tlb_flush_required; |
| 669 | x86_platform.guest.enc_status_change_prepare = hv_vtom_clear_present; |
| 670 | x86_platform.guest.enc_status_change_finish = hv_vtom_set_host_visibility; |
| 671 | |
| 672 | /* Set WB as the default cache mode. */ |
| 673 | guest_force_mtrr_state(NULL, num_var: 0, MTRR_TYPE_WRBACK); |
| 674 | } |
| 675 | |
| 676 | #endif /* defined(CONFIG_AMD_MEM_ENCRYPT) || defined(CONFIG_INTEL_TDX_GUEST) */ |
| 677 | |
| 678 | enum hv_isolation_type hv_get_isolation_type(void) |
| 679 | { |
| 680 | if (!(ms_hyperv.priv_high & HV_ISOLATION)) |
| 681 | return HV_ISOLATION_TYPE_NONE; |
| 682 | return FIELD_GET(HV_ISOLATION_TYPE, ms_hyperv.isolation_config_b); |
| 683 | } |
| 684 | EXPORT_SYMBOL_GPL(hv_get_isolation_type); |
| 685 | |
| 686 | /* |
| 687 | * hv_is_isolation_supported - Check system runs in the Hyper-V |
| 688 | * isolation VM. |
| 689 | */ |
| 690 | bool hv_is_isolation_supported(void) |
| 691 | { |
| 692 | if (!cpu_feature_enabled(X86_FEATURE_HYPERVISOR)) |
| 693 | return false; |
| 694 | |
| 695 | if (!hypervisor_is_type(type: X86_HYPER_MS_HYPERV)) |
| 696 | return false; |
| 697 | |
| 698 | return hv_get_isolation_type() != HV_ISOLATION_TYPE_NONE; |
| 699 | } |
| 700 | |
| 701 | DEFINE_STATIC_KEY_FALSE(isolation_type_snp); |
| 702 | |
| 703 | /* |
| 704 | * hv_isolation_type_snp - Check if the system runs in an AMD SEV-SNP based |
| 705 | * isolation VM. |
| 706 | */ |
| 707 | bool hv_isolation_type_snp(void) |
| 708 | { |
| 709 | return static_branch_unlikely(&isolation_type_snp); |
| 710 | } |
| 711 | |
| 712 | DEFINE_STATIC_KEY_FALSE(isolation_type_tdx); |
| 713 | /* |
| 714 | * hv_isolation_type_tdx - Check if the system runs in an Intel TDX based |
| 715 | * isolated VM. |
| 716 | */ |
| 717 | bool hv_isolation_type_tdx(void) |
| 718 | { |
| 719 | return static_branch_unlikely(&isolation_type_tdx); |
| 720 | } |
| 721 | |