linux/arch/x86/kernel/sev.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * AMD Memory Encryption Support
   4 *
   5 * Copyright (C) 2019 SUSE
   6 *
   7 * Author: Joerg Roedel <jroedel@suse.de>
   8 */
   9
  10#define pr_fmt(fmt)     "SEV: " fmt
  11
  12#include <linux/sched/debug.h>  /* For show_regs() */
  13#include <linux/percpu-defs.h>
  14#include <linux/mem_encrypt.h>
  15#include <linux/printk.h>
  16#include <linux/mm_types.h>
  17#include <linux/set_memory.h>
  18#include <linux/memblock.h>
  19#include <linux/kernel.h>
  20#include <linux/mm.h>
  21
  22#include <asm/cpu_entry_area.h>
  23#include <asm/stacktrace.h>
  24#include <asm/sev.h>
  25#include <asm/insn-eval.h>
  26#include <asm/fpu/internal.h>
  27#include <asm/processor.h>
  28#include <asm/realmode.h>
  29#include <asm/traps.h>
  30#include <asm/svm.h>
  31#include <asm/smp.h>
  32#include <asm/cpu.h>
  33
  34#define DR7_RESET_VALUE        0x400
  35
  36/* For early boot hypervisor communication in SEV-ES enabled guests */
  37static struct ghcb boot_ghcb_page __bss_decrypted __aligned(PAGE_SIZE);
  38
  39/*
  40 * Needs to be in the .data section because we need it NULL before bss is
  41 * cleared
  42 */
  43static struct ghcb __initdata *boot_ghcb;
  44
  45/* #VC handler runtime per-CPU data */
  46struct sev_es_runtime_data {
  47        struct ghcb ghcb_page;
  48
  49        /* Physical storage for the per-CPU IST stack of the #VC handler */
  50        char ist_stack[EXCEPTION_STKSZ] __aligned(PAGE_SIZE);
  51
  52        /*
  53         * Physical storage for the per-CPU fall-back stack of the #VC handler.
  54         * The fall-back stack is used when it is not safe to switch back to the
  55         * interrupted stack in the #VC entry code.
  56         */
  57        char fallback_stack[EXCEPTION_STKSZ] __aligned(PAGE_SIZE);
  58
  59        /*
  60         * Reserve one page per CPU as backup storage for the unencrypted GHCB.
  61         * It is needed when an NMI happens while the #VC handler uses the real
  62         * GHCB, and the NMI handler itself is causing another #VC exception. In
  63         * that case the GHCB content of the first handler needs to be backed up
  64         * and restored.
  65         */
  66        struct ghcb backup_ghcb;
  67
  68        /*
  69         * Mark the per-cpu GHCBs as in-use to detect nested #VC exceptions.
  70         * There is no need for it to be atomic, because nothing is written to
  71         * the GHCB between the read and the write of ghcb_active. So it is safe
  72         * to use it when a nested #VC exception happens before the write.
  73         *
  74         * This is necessary for example in the #VC->NMI->#VC case when the NMI
  75         * happens while the first #VC handler uses the GHCB. When the NMI code
  76         * raises a second #VC handler it might overwrite the contents of the
  77         * GHCB written by the first handler. To avoid this the content of the
  78         * GHCB is saved and restored when the GHCB is detected to be in use
  79         * already.
  80         */
  81        bool ghcb_active;
  82        bool backup_ghcb_active;
  83
  84        /*
  85         * Cached DR7 value - write it on DR7 writes and return it on reads.
  86         * That value will never make it to the real hardware DR7 as debugging
  87         * is currently unsupported in SEV-ES guests.
  88         */
  89        unsigned long dr7;
  90};
  91
  92struct ghcb_state {
  93        struct ghcb *ghcb;
  94};
  95
  96static DEFINE_PER_CPU(struct sev_es_runtime_data*, runtime_data);
  97DEFINE_STATIC_KEY_FALSE(sev_es_enable_key);
  98
  99/* Needed in vc_early_forward_exception */
 100void do_early_exception(struct pt_regs *regs, int trapnr);
 101
 102static void __init setup_vc_stacks(int cpu)
 103{
 104        struct sev_es_runtime_data *data;
 105        struct cpu_entry_area *cea;
 106        unsigned long vaddr;
 107        phys_addr_t pa;
 108
 109        data = per_cpu(runtime_data, cpu);
 110        cea  = get_cpu_entry_area(cpu);
 111
 112        /* Map #VC IST stack */
 113        vaddr = CEA_ESTACK_BOT(&cea->estacks, VC);
 114        pa    = __pa(data->ist_stack);
 115        cea_set_pte((void *)vaddr, pa, PAGE_KERNEL);
 116
 117        /* Map VC fall-back stack */
 118        vaddr = CEA_ESTACK_BOT(&cea->estacks, VC2);
 119        pa    = __pa(data->fallback_stack);
 120        cea_set_pte((void *)vaddr, pa, PAGE_KERNEL);
 121}
 122
 123static __always_inline bool on_vc_stack(struct pt_regs *regs)
 124{
 125        unsigned long sp = regs->sp;
 126
 127        /* User-mode RSP is not trusted */
 128        if (user_mode(regs))
 129                return false;
 130
 131        /* SYSCALL gap still has user-mode RSP */
 132        if (ip_within_syscall_gap(regs))
 133                return false;
 134
 135        return ((sp >= __this_cpu_ist_bottom_va(VC)) && (sp < __this_cpu_ist_top_va(VC)));
 136}
 137
 138/*
 139 * This function handles the case when an NMI is raised in the #VC
 140 * exception handler entry code, before the #VC handler has switched off
 141 * its IST stack. In this case, the IST entry for #VC must be adjusted,
 142 * so that any nested #VC exception will not overwrite the stack
 143 * contents of the interrupted #VC handler.
 144 *
 145 * The IST entry is adjusted unconditionally so that it can be also be
 146 * unconditionally adjusted back in __sev_es_ist_exit(). Otherwise a
 147 * nested sev_es_ist_exit() call may adjust back the IST entry too
 148 * early.
 149 *
 150 * The __sev_es_ist_enter() and __sev_es_ist_exit() functions always run
 151 * on the NMI IST stack, as they are only called from NMI handling code
 152 * right now.
 153 */
 154void noinstr __sev_es_ist_enter(struct pt_regs *regs)
 155{
 156        unsigned long old_ist, new_ist;
 157
 158        /* Read old IST entry */
 159        new_ist = old_ist = __this_cpu_read(cpu_tss_rw.x86_tss.ist[IST_INDEX_VC]);
 160
 161        /*
 162         * If NMI happened while on the #VC IST stack, set the new IST
 163         * value below regs->sp, so that the interrupted stack frame is
 164         * not overwritten by subsequent #VC exceptions.
 165         */
 166        if (on_vc_stack(regs))
 167                new_ist = regs->sp;
 168
 169        /*
 170         * Reserve additional 8 bytes and store old IST value so this
 171         * adjustment can be unrolled in __sev_es_ist_exit().
 172         */
 173        new_ist -= sizeof(old_ist);
 174        *(unsigned long *)new_ist = old_ist;
 175
 176        /* Set new IST entry */
 177        this_cpu_write(cpu_tss_rw.x86_tss.ist[IST_INDEX_VC], new_ist);
 178}
 179
 180void noinstr __sev_es_ist_exit(void)
 181{
 182        unsigned long ist;
 183
 184        /* Read IST entry */
 185        ist = __this_cpu_read(cpu_tss_rw.x86_tss.ist[IST_INDEX_VC]);
 186
 187        if (WARN_ON(ist == __this_cpu_ist_top_va(VC)))
 188                return;
 189
 190        /* Read back old IST entry and write it to the TSS */
 191        this_cpu_write(cpu_tss_rw.x86_tss.ist[IST_INDEX_VC], *(unsigned long *)ist);
 192}
 193
 194/*
 195 * Nothing shall interrupt this code path while holding the per-CPU
 196 * GHCB. The backup GHCB is only for NMIs interrupting this path.
 197 *
 198 * Callers must disable local interrupts around it.
 199 */
 200static noinstr struct ghcb *__sev_get_ghcb(struct ghcb_state *state)
 201{
 202        struct sev_es_runtime_data *data;
 203        struct ghcb *ghcb;
 204
 205        WARN_ON(!irqs_disabled());
 206
 207        data = this_cpu_read(runtime_data);
 208        ghcb = &data->ghcb_page;
 209
 210        if (unlikely(data->ghcb_active)) {
 211                /* GHCB is already in use - save its contents */
 212
 213                if (unlikely(data->backup_ghcb_active)) {
 214                        /*
 215                         * Backup-GHCB is also already in use. There is no way
 216                         * to continue here so just kill the machine. To make
 217                         * panic() work, mark GHCBs inactive so that messages
 218                         * can be printed out.
 219                         */
 220                        data->ghcb_active        = false;
 221                        data->backup_ghcb_active = false;
 222
 223                        instrumentation_begin();
 224                        panic("Unable to handle #VC exception! GHCB and Backup GHCB are already in use");
 225                        instrumentation_end();
 226                }
 227
 228                /* Mark backup_ghcb active before writing to it */
 229                data->backup_ghcb_active = true;
 230
 231                state->ghcb = &data->backup_ghcb;
 232
 233                /* Backup GHCB content */
 234                *state->ghcb = *ghcb;
 235        } else {
 236                state->ghcb = NULL;
 237                data->ghcb_active = true;
 238        }
 239
 240        return ghcb;
 241}
 242
 243/* Needed in vc_early_forward_exception */
 244void do_early_exception(struct pt_regs *regs, int trapnr);
 245
 246static inline u64 sev_es_rd_ghcb_msr(void)
 247{
 248        return __rdmsr(MSR_AMD64_SEV_ES_GHCB);
 249}
 250
 251static __always_inline void sev_es_wr_ghcb_msr(u64 val)
 252{
 253        u32 low, high;
 254
 255        low  = (u32)(val);
 256        high = (u32)(val >> 32);
 257
 258        native_wrmsr(MSR_AMD64_SEV_ES_GHCB, low, high);
 259}
 260
 261static int vc_fetch_insn_kernel(struct es_em_ctxt *ctxt,
 262                                unsigned char *buffer)
 263{
 264        return copy_from_kernel_nofault(buffer, (unsigned char *)ctxt->regs->ip, MAX_INSN_SIZE);
 265}
 266
 267static enum es_result __vc_decode_user_insn(struct es_em_ctxt *ctxt)
 268{
 269        char buffer[MAX_INSN_SIZE];
 270        int insn_bytes;
 271
 272        insn_bytes = insn_fetch_from_user_inatomic(ctxt->regs, buffer);
 273        if (insn_bytes == 0) {
 274                /* Nothing could be copied */
 275                ctxt->fi.vector     = X86_TRAP_PF;
 276                ctxt->fi.error_code = X86_PF_INSTR | X86_PF_USER;
 277                ctxt->fi.cr2        = ctxt->regs->ip;
 278                return ES_EXCEPTION;
 279        } else if (insn_bytes == -EINVAL) {
 280                /* Effective RIP could not be calculated */
 281                ctxt->fi.vector     = X86_TRAP_GP;
 282                ctxt->fi.error_code = 0;
 283                ctxt->fi.cr2        = 0;
 284                return ES_EXCEPTION;
 285        }
 286
 287        if (!insn_decode_from_regs(&ctxt->insn, ctxt->regs, buffer, insn_bytes))
 288                return ES_DECODE_FAILED;
 289
 290        if (ctxt->insn.immediate.got)
 291                return ES_OK;
 292        else
 293                return ES_DECODE_FAILED;
 294}
 295
 296static enum es_result __vc_decode_kern_insn(struct es_em_ctxt *ctxt)
 297{
 298        char buffer[MAX_INSN_SIZE];
 299        int res, ret;
 300
 301        res = vc_fetch_insn_kernel(ctxt, buffer);
 302        if (res) {
 303                ctxt->fi.vector     = X86_TRAP_PF;
 304                ctxt->fi.error_code = X86_PF_INSTR;
 305                ctxt->fi.cr2        = ctxt->regs->ip;
 306                return ES_EXCEPTION;
 307        }
 308
 309        ret = insn_decode(&ctxt->insn, buffer, MAX_INSN_SIZE, INSN_MODE_64);
 310        if (ret < 0)
 311                return ES_DECODE_FAILED;
 312        else
 313                return ES_OK;
 314}
 315
 316static enum es_result vc_decode_insn(struct es_em_ctxt *ctxt)
 317{
 318        if (user_mode(ctxt->regs))
 319                return __vc_decode_user_insn(ctxt);
 320        else
 321                return __vc_decode_kern_insn(ctxt);
 322}
 323
 324static enum es_result vc_write_mem(struct es_em_ctxt *ctxt,
 325                                   char *dst, char *buf, size_t size)
 326{
 327        unsigned long error_code = X86_PF_PROT | X86_PF_WRITE;
 328        char __user *target = (char __user *)dst;
 329        u64 d8;
 330        u32 d4;
 331        u16 d2;
 332        u8  d1;
 333
 334        /*
 335         * This function uses __put_user() independent of whether kernel or user
 336         * memory is accessed. This works fine because __put_user() does no
 337         * sanity checks of the pointer being accessed. All that it does is
 338         * to report when the access failed.
 339         *
 340         * Also, this function runs in atomic context, so __put_user() is not
 341         * allowed to sleep. The page-fault handler detects that it is running
 342         * in atomic context and will not try to take mmap_sem and handle the
 343         * fault, so additional pagefault_enable()/disable() calls are not
 344         * needed.
 345         *
 346         * The access can't be done via copy_to_user() here because
 347         * vc_write_mem() must not use string instructions to access unsafe
 348         * memory. The reason is that MOVS is emulated by the #VC handler by
 349         * splitting the move up into a read and a write and taking a nested #VC
 350         * exception on whatever of them is the MMIO access. Using string
 351         * instructions here would cause infinite nesting.
 352         */
 353        switch (size) {
 354        case 1:
 355                memcpy(&d1, buf, 1);
 356                if (__put_user(d1, target))
 357                        goto fault;
 358                break;
 359        case 2:
 360                memcpy(&d2, buf, 2);
 361                if (__put_user(d2, target))
 362                        goto fault;
 363                break;
 364        case 4:
 365                memcpy(&d4, buf, 4);
 366                if (__put_user(d4, target))
 367                        goto fault;
 368                break;
 369        case 8:
 370                memcpy(&d8, buf, 8);
 371                if (__put_user(d8, target))
 372                        goto fault;
 373                break;
 374        default:
 375                WARN_ONCE(1, "%s: Invalid size: %zu\n", __func__, size);
 376                return ES_UNSUPPORTED;
 377        }
 378
 379        return ES_OK;
 380
 381fault:
 382        if (user_mode(ctxt->regs))
 383                error_code |= X86_PF_USER;
 384
 385        ctxt->fi.vector = X86_TRAP_PF;
 386        ctxt->fi.error_code = error_code;
 387        ctxt->fi.cr2 = (unsigned long)dst;
 388
 389        return ES_EXCEPTION;
 390}
 391
 392static enum es_result vc_read_mem(struct es_em_ctxt *ctxt,
 393                                  char *src, char *buf, size_t size)
 394{
 395        unsigned long error_code = X86_PF_PROT;
 396        char __user *s = (char __user *)src;
 397        u64 d8;
 398        u32 d4;
 399        u16 d2;
 400        u8  d1;
 401
 402        /*
 403         * This function uses __get_user() independent of whether kernel or user
 404         * memory is accessed. This works fine because __get_user() does no
 405         * sanity checks of the pointer being accessed. All that it does is
 406         * to report when the access failed.
 407         *
 408         * Also, this function runs in atomic context, so __get_user() is not
 409         * allowed to sleep. The page-fault handler detects that it is running
 410         * in atomic context and will not try to take mmap_sem and handle the
 411         * fault, so additional pagefault_enable()/disable() calls are not
 412         * needed.
 413         *
 414         * The access can't be done via copy_from_user() here because
 415         * vc_read_mem() must not use string instructions to access unsafe
 416         * memory. The reason is that MOVS is emulated by the #VC handler by
 417         * splitting the move up into a read and a write and taking a nested #VC
 418         * exception on whatever of them is the MMIO access. Using string
 419         * instructions here would cause infinite nesting.
 420         */
 421        switch (size) {
 422        case 1:
 423                if (__get_user(d1, s))
 424                        goto fault;
 425                memcpy(buf, &d1, 1);
 426                break;
 427        case 2:
 428                if (__get_user(d2, s))
 429                        goto fault;
 430                memcpy(buf, &d2, 2);
 431                break;
 432        case 4:
 433                if (__get_user(d4, s))
 434                        goto fault;
 435                memcpy(buf, &d4, 4);
 436                break;
 437        case 8:
 438                if (__get_user(d8, s))
 439                        goto fault;
 440                memcpy(buf, &d8, 8);
 441                break;
 442        default:
 443                WARN_ONCE(1, "%s: Invalid size: %zu\n", __func__, size);
 444                return ES_UNSUPPORTED;
 445        }
 446
 447        return ES_OK;
 448
 449fault:
 450        if (user_mode(ctxt->regs))
 451                error_code |= X86_PF_USER;
 452
 453        ctxt->fi.vector = X86_TRAP_PF;
 454        ctxt->fi.error_code = error_code;
 455        ctxt->fi.cr2 = (unsigned long)src;
 456
 457        return ES_EXCEPTION;
 458}
 459
 460static enum es_result vc_slow_virt_to_phys(struct ghcb *ghcb, struct es_em_ctxt *ctxt,
 461                                           unsigned long vaddr, phys_addr_t *paddr)
 462{
 463        unsigned long va = (unsigned long)vaddr;
 464        unsigned int level;
 465        phys_addr_t pa;
 466        pgd_t *pgd;
 467        pte_t *pte;
 468
 469        pgd = __va(read_cr3_pa());
 470        pgd = &pgd[pgd_index(va)];
 471        pte = lookup_address_in_pgd(pgd, va, &level);
 472        if (!pte) {
 473                ctxt->fi.vector     = X86_TRAP_PF;
 474                ctxt->fi.cr2        = vaddr;
 475                ctxt->fi.error_code = 0;
 476
 477                if (user_mode(ctxt->regs))
 478                        ctxt->fi.error_code |= X86_PF_USER;
 479
 480                return ES_EXCEPTION;
 481        }
 482
 483        if (WARN_ON_ONCE(pte_val(*pte) & _PAGE_ENC))
 484                /* Emulated MMIO to/from encrypted memory not supported */
 485                return ES_UNSUPPORTED;
 486
 487        pa = (phys_addr_t)pte_pfn(*pte) << PAGE_SHIFT;
 488        pa |= va & ~page_level_mask(level);
 489
 490        *paddr = pa;
 491
 492        return ES_OK;
 493}
 494
 495/* Include code shared with pre-decompression boot stage */
 496#include "sev-shared.c"
 497
 498static noinstr void __sev_put_ghcb(struct ghcb_state *state)
 499{
 500        struct sev_es_runtime_data *data;
 501        struct ghcb *ghcb;
 502
 503        WARN_ON(!irqs_disabled());
 504
 505        data = this_cpu_read(runtime_data);
 506        ghcb = &data->ghcb_page;
 507
 508        if (state->ghcb) {
 509                /* Restore GHCB from Backup */
 510                *ghcb = *state->ghcb;
 511                data->backup_ghcb_active = false;
 512                state->ghcb = NULL;
 513        } else {
 514                /*
 515                 * Invalidate the GHCB so a VMGEXIT instruction issued
 516                 * from userspace won't appear to be valid.
 517                 */
 518                vc_ghcb_invalidate(ghcb);
 519                data->ghcb_active = false;
 520        }
 521}
 522
 523void noinstr __sev_es_nmi_complete(void)
 524{
 525        struct ghcb_state state;
 526        struct ghcb *ghcb;
 527
 528        ghcb = __sev_get_ghcb(&state);
 529
 530        vc_ghcb_invalidate(ghcb);
 531        ghcb_set_sw_exit_code(ghcb, SVM_VMGEXIT_NMI_COMPLETE);
 532        ghcb_set_sw_exit_info_1(ghcb, 0);
 533        ghcb_set_sw_exit_info_2(ghcb, 0);
 534
 535        sev_es_wr_ghcb_msr(__pa_nodebug(ghcb));
 536        VMGEXIT();
 537
 538        __sev_put_ghcb(&state);
 539}
 540
 541static u64 get_jump_table_addr(void)
 542{
 543        struct ghcb_state state;
 544        unsigned long flags;
 545        struct ghcb *ghcb;
 546        u64 ret = 0;
 547
 548        local_irq_save(flags);
 549
 550        ghcb = __sev_get_ghcb(&state);
 551
 552        vc_ghcb_invalidate(ghcb);
 553        ghcb_set_sw_exit_code(ghcb, SVM_VMGEXIT_AP_JUMP_TABLE);
 554        ghcb_set_sw_exit_info_1(ghcb, SVM_VMGEXIT_GET_AP_JUMP_TABLE);
 555        ghcb_set_sw_exit_info_2(ghcb, 0);
 556
 557        sev_es_wr_ghcb_msr(__pa(ghcb));
 558        VMGEXIT();
 559
 560        if (ghcb_sw_exit_info_1_is_valid(ghcb) &&
 561            ghcb_sw_exit_info_2_is_valid(ghcb))
 562                ret = ghcb->save.sw_exit_info_2;
 563
 564        __sev_put_ghcb(&state);
 565
 566        local_irq_restore(flags);
 567
 568        return ret;
 569}
 570
 571int sev_es_setup_ap_jump_table(struct real_mode_header *rmh)
 572{
 573        u16 startup_cs, startup_ip;
 574        phys_addr_t jump_table_pa;
 575        u64 jump_table_addr;
 576        u16 __iomem *jump_table;
 577
 578        jump_table_addr = get_jump_table_addr();
 579
 580        /* On UP guests there is no jump table so this is not a failure */
 581        if (!jump_table_addr)
 582                return 0;
 583
 584        /* Check if AP Jump Table is page-aligned */
 585        if (jump_table_addr & ~PAGE_MASK)
 586                return -EINVAL;
 587
 588        jump_table_pa = jump_table_addr & PAGE_MASK;
 589
 590        startup_cs = (u16)(rmh->trampoline_start >> 4);
 591        startup_ip = (u16)(rmh->sev_es_trampoline_start -
 592                           rmh->trampoline_start);
 593
 594        jump_table = ioremap_encrypted(jump_table_pa, PAGE_SIZE);
 595        if (!jump_table)
 596                return -EIO;
 597
 598        writew(startup_ip, &jump_table[0]);
 599        writew(startup_cs, &jump_table[1]);
 600
 601        iounmap(jump_table);
 602
 603        return 0;
 604}
 605
 606/*
 607 * This is needed by the OVMF UEFI firmware which will use whatever it finds in
 608 * the GHCB MSR as its GHCB to talk to the hypervisor. So make sure the per-cpu
 609 * runtime GHCBs used by the kernel are also mapped in the EFI page-table.
 610 */
 611int __init sev_es_efi_map_ghcbs(pgd_t *pgd)
 612{
 613        struct sev_es_runtime_data *data;
 614        unsigned long address, pflags;
 615        int cpu;
 616        u64 pfn;
 617
 618        if (!sev_es_active())
 619                return 0;
 620
 621        pflags = _PAGE_NX | _PAGE_RW;
 622
 623        for_each_possible_cpu(cpu) {
 624                data = per_cpu(runtime_data, cpu);
 625
 626                address = __pa(&data->ghcb_page);
 627                pfn = address >> PAGE_SHIFT;
 628
 629                if (kernel_map_pages_in_pgd(pgd, pfn, address, 1, pflags))
 630                        return 1;
 631        }
 632
 633        return 0;
 634}
 635
 636static enum es_result vc_handle_msr(struct ghcb *ghcb, struct es_em_ctxt *ctxt)
 637{
 638        struct pt_regs *regs = ctxt->regs;
 639        enum es_result ret;
 640        u64 exit_info_1;
 641
 642        /* Is it a WRMSR? */
 643        exit_info_1 = (ctxt->insn.opcode.bytes[1] == 0x30) ? 1 : 0;
 644
 645        ghcb_set_rcx(ghcb, regs->cx);
 646        if (exit_info_1) {
 647                ghcb_set_rax(ghcb, regs->ax);
 648                ghcb_set_rdx(ghcb, regs->dx);
 649        }
 650
 651        ret = sev_es_ghcb_hv_call(ghcb, ctxt, SVM_EXIT_MSR, exit_info_1, 0);
 652
 653        if ((ret == ES_OK) && (!exit_info_1)) {
 654                regs->ax = ghcb->save.rax;
 655                regs->dx = ghcb->save.rdx;
 656        }
 657
 658        return ret;
 659}
 660
 661/*
 662 * This function runs on the first #VC exception after the kernel
 663 * switched to virtual addresses.
 664 */
 665static bool __init sev_es_setup_ghcb(void)
 666{
 667        /* First make sure the hypervisor talks a supported protocol. */
 668        if (!sev_es_negotiate_protocol())
 669                return false;
 670
 671        /*
 672         * Clear the boot_ghcb. The first exception comes in before the bss
 673         * section is cleared.
 674         */
 675        memset(&boot_ghcb_page, 0, PAGE_SIZE);
 676
 677        /* Alright - Make the boot-ghcb public */
 678        boot_ghcb = &boot_ghcb_page;
 679
 680        return true;
 681}
 682
 683#ifdef CONFIG_HOTPLUG_CPU
 684static void sev_es_ap_hlt_loop(void)
 685{
 686        struct ghcb_state state;
 687        struct ghcb *ghcb;
 688
 689        ghcb = __sev_get_ghcb(&state);
 690
 691        while (true) {
 692                vc_ghcb_invalidate(ghcb);
 693                ghcb_set_sw_exit_code(ghcb, SVM_VMGEXIT_AP_HLT_LOOP);
 694                ghcb_set_sw_exit_info_1(ghcb, 0);
 695                ghcb_set_sw_exit_info_2(ghcb, 0);
 696
 697                sev_es_wr_ghcb_msr(__pa(ghcb));
 698                VMGEXIT();
 699
 700                /* Wakeup signal? */
 701                if (ghcb_sw_exit_info_2_is_valid(ghcb) &&
 702                    ghcb->save.sw_exit_info_2)
 703                        break;
 704        }
 705
 706        __sev_put_ghcb(&state);
 707}
 708
 709/*
 710 * Play_dead handler when running under SEV-ES. This is needed because
 711 * the hypervisor can't deliver an SIPI request to restart the AP.
 712 * Instead the kernel has to issue a VMGEXIT to halt the VCPU until the
 713 * hypervisor wakes it up again.
 714 */
 715static void sev_es_play_dead(void)
 716{
 717        play_dead_common();
 718
 719        /* IRQs now disabled */
 720
 721        sev_es_ap_hlt_loop();
 722
 723        /*
 724         * If we get here, the VCPU was woken up again. Jump to CPU
 725         * startup code to get it back online.
 726         */
 727        start_cpu0();
 728}
 729#else  /* CONFIG_HOTPLUG_CPU */
 730#define sev_es_play_dead        native_play_dead
 731#endif /* CONFIG_HOTPLUG_CPU */
 732
 733#ifdef CONFIG_SMP
 734static void __init sev_es_setup_play_dead(void)
 735{
 736        smp_ops.play_dead = sev_es_play_dead;
 737}
 738#else
 739static inline void sev_es_setup_play_dead(void) { }
 740#endif
 741
 742static void __init alloc_runtime_data(int cpu)
 743{
 744        struct sev_es_runtime_data *data;
 745
 746        data = memblock_alloc(sizeof(*data), PAGE_SIZE);
 747        if (!data)
 748                panic("Can't allocate SEV-ES runtime data");
 749
 750        per_cpu(runtime_data, cpu) = data;
 751}
 752
 753static void __init init_ghcb(int cpu)
 754{
 755        struct sev_es_runtime_data *data;
 756        int err;
 757
 758        data = per_cpu(runtime_data, cpu);
 759
 760        err = early_set_memory_decrypted((unsigned long)&data->ghcb_page,
 761                                         sizeof(data->ghcb_page));
 762        if (err)
 763                panic("Can't map GHCBs unencrypted");
 764
 765        memset(&data->ghcb_page, 0, sizeof(data->ghcb_page));
 766
 767        data->ghcb_active = false;
 768        data->backup_ghcb_active = false;
 769}
 770
 771void __init sev_es_init_vc_handling(void)
 772{
 773        int cpu;
 774
 775        BUILD_BUG_ON(offsetof(struct sev_es_runtime_data, ghcb_page) % PAGE_SIZE);
 776
 777        if (!sev_es_active())
 778                return;
 779
 780        if (!sev_es_check_cpu_features())
 781                panic("SEV-ES CPU Features missing");
 782
 783        /* Enable SEV-ES special handling */
 784        static_branch_enable(&sev_es_enable_key);
 785
 786        /* Initialize per-cpu GHCB pages */
 787        for_each_possible_cpu(cpu) {
 788                alloc_runtime_data(cpu);
 789                init_ghcb(cpu);
 790                setup_vc_stacks(cpu);
 791        }
 792
 793        sev_es_setup_play_dead();
 794
 795        /* Secondary CPUs use the runtime #VC handler */
 796        initial_vc_handler = (unsigned long)kernel_exc_vmm_communication;
 797}
 798
 799static void __init vc_early_forward_exception(struct es_em_ctxt *ctxt)
 800{
 801        int trapnr = ctxt->fi.vector;
 802
 803        if (trapnr == X86_TRAP_PF)
 804                native_write_cr2(ctxt->fi.cr2);
 805
 806        ctxt->regs->orig_ax = ctxt->fi.error_code;
 807        do_early_exception(ctxt->regs, trapnr);
 808}
 809
 810static long *vc_insn_get_reg(struct es_em_ctxt *ctxt)
 811{
 812        long *reg_array;
 813        int offset;
 814
 815        reg_array = (long *)ctxt->regs;
 816        offset    = insn_get_modrm_reg_off(&ctxt->insn, ctxt->regs);
 817
 818        if (offset < 0)
 819                return NULL;
 820
 821        offset /= sizeof(long);
 822
 823        return reg_array + offset;
 824}
 825
 826static long *vc_insn_get_rm(struct es_em_ctxt *ctxt)
 827{
 828        long *reg_array;
 829        int offset;
 830
 831        reg_array = (long *)ctxt->regs;
 832        offset    = insn_get_modrm_rm_off(&ctxt->insn, ctxt->regs);
 833
 834        if (offset < 0)
 835                return NULL;
 836
 837        offset /= sizeof(long);
 838
 839        return reg_array + offset;
 840}
 841static enum es_result vc_do_mmio(struct ghcb *ghcb, struct es_em_ctxt *ctxt,
 842                                 unsigned int bytes, bool read)
 843{
 844        u64 exit_code, exit_info_1, exit_info_2;
 845        unsigned long ghcb_pa = __pa(ghcb);
 846        enum es_result res;
 847        phys_addr_t paddr;
 848        void __user *ref;
 849
 850        ref = insn_get_addr_ref(&ctxt->insn, ctxt->regs);
 851        if (ref == (void __user *)-1L)
 852                return ES_UNSUPPORTED;
 853
 854        exit_code = read ? SVM_VMGEXIT_MMIO_READ : SVM_VMGEXIT_MMIO_WRITE;
 855
 856        res = vc_slow_virt_to_phys(ghcb, ctxt, (unsigned long)ref, &paddr);
 857        if (res != ES_OK) {
 858                if (res == ES_EXCEPTION && !read)
 859                        ctxt->fi.error_code |= X86_PF_WRITE;
 860
 861                return res;
 862        }
 863
 864        exit_info_1 = paddr;
 865        /* Can never be greater than 8 */
 866        exit_info_2 = bytes;
 867
 868        ghcb_set_sw_scratch(ghcb, ghcb_pa + offsetof(struct ghcb, shared_buffer));
 869
 870        return sev_es_ghcb_hv_call(ghcb, ctxt, exit_code, exit_info_1, exit_info_2);
 871}
 872
 873static enum es_result vc_handle_mmio_twobyte_ops(struct ghcb *ghcb,
 874                                                 struct es_em_ctxt *ctxt)
 875{
 876        struct insn *insn = &ctxt->insn;
 877        unsigned int bytes = 0;
 878        enum es_result ret;
 879        int sign_byte;
 880        long *reg_data;
 881
 882        switch (insn->opcode.bytes[1]) {
 883                /* MMIO Read w/ zero-extension */
 884        case 0xb6:
 885                bytes = 1;
 886                fallthrough;
 887        case 0xb7:
 888                if (!bytes)
 889                        bytes = 2;
 890
 891                ret = vc_do_mmio(ghcb, ctxt, bytes, true);
 892                if (ret)
 893                        break;
 894
 895                /* Zero extend based on operand size */
 896                reg_data = vc_insn_get_reg(ctxt);
 897                if (!reg_data)
 898                        return ES_DECODE_FAILED;
 899
 900                memset(reg_data, 0, insn->opnd_bytes);
 901
 902                memcpy(reg_data, ghcb->shared_buffer, bytes);
 903                break;
 904
 905                /* MMIO Read w/ sign-extension */
 906        case 0xbe:
 907                bytes = 1;
 908                fallthrough;
 909        case 0xbf:
 910                if (!bytes)
 911                        bytes = 2;
 912
 913                ret = vc_do_mmio(ghcb, ctxt, bytes, true);
 914                if (ret)
 915                        break;
 916
 917                /* Sign extend based on operand size */
 918                reg_data = vc_insn_get_reg(ctxt);
 919                if (!reg_data)
 920                        return ES_DECODE_FAILED;
 921
 922                if (bytes == 1) {
 923                        u8 *val = (u8 *)ghcb->shared_buffer;
 924
 925                        sign_byte = (*val & 0x80) ? 0xff : 0x00;
 926                } else {
 927                        u16 *val = (u16 *)ghcb->shared_buffer;
 928
 929                        sign_byte = (*val & 0x8000) ? 0xff : 0x00;
 930                }
 931                memset(reg_data, sign_byte, insn->opnd_bytes);
 932
 933                memcpy(reg_data, ghcb->shared_buffer, bytes);
 934                break;
 935
 936        default:
 937                ret = ES_UNSUPPORTED;
 938        }
 939
 940        return ret;
 941}
 942
 943/*
 944 * The MOVS instruction has two memory operands, which raises the
 945 * problem that it is not known whether the access to the source or the
 946 * destination caused the #VC exception (and hence whether an MMIO read
 947 * or write operation needs to be emulated).
 948 *
 949 * Instead of playing games with walking page-tables and trying to guess
 950 * whether the source or destination is an MMIO range, split the move
 951 * into two operations, a read and a write with only one memory operand.
 952 * This will cause a nested #VC exception on the MMIO address which can
 953 * then be handled.
 954 *
 955 * This implementation has the benefit that it also supports MOVS where
 956 * source _and_ destination are MMIO regions.
 957 *
 958 * It will slow MOVS on MMIO down a lot, but in SEV-ES guests it is a
 959 * rare operation. If it turns out to be a performance problem the split
 960 * operations can be moved to memcpy_fromio() and memcpy_toio().
 961 */
 962static enum es_result vc_handle_mmio_movs(struct es_em_ctxt *ctxt,
 963                                          unsigned int bytes)
 964{
 965        unsigned long ds_base, es_base;
 966        unsigned char *src, *dst;
 967        unsigned char buffer[8];
 968        enum es_result ret;
 969        bool rep;
 970        int off;
 971
 972        ds_base = insn_get_seg_base(ctxt->regs, INAT_SEG_REG_DS);
 973        es_base = insn_get_seg_base(ctxt->regs, INAT_SEG_REG_ES);
 974
 975        if (ds_base == -1L || es_base == -1L) {
 976                ctxt->fi.vector = X86_TRAP_GP;
 977                ctxt->fi.error_code = 0;
 978                return ES_EXCEPTION;
 979        }
 980
 981        src = ds_base + (unsigned char *)ctxt->regs->si;
 982        dst = es_base + (unsigned char *)ctxt->regs->di;
 983
 984        ret = vc_read_mem(ctxt, src, buffer, bytes);
 985        if (ret != ES_OK)
 986                return ret;
 987
 988        ret = vc_write_mem(ctxt, dst, buffer, bytes);
 989        if (ret != ES_OK)
 990                return ret;
 991
 992        if (ctxt->regs->flags & X86_EFLAGS_DF)
 993                off = -bytes;
 994        else
 995                off =  bytes;
 996
 997        ctxt->regs->si += off;
 998        ctxt->regs->di += off;
 999
1000        rep = insn_has_rep_prefix(&ctxt->insn);
1001        if (rep)
1002                ctxt->regs->cx -= 1;
1003
1004        if (!rep || ctxt->regs->cx == 0)
1005                return ES_OK;
1006        else
1007                return ES_RETRY;
1008}
1009
1010static enum es_result vc_handle_mmio(struct ghcb *ghcb,
1011                                     struct es_em_ctxt *ctxt)
1012{
1013        struct insn *insn = &ctxt->insn;
1014        unsigned int bytes = 0;
1015        enum es_result ret;
1016        long *reg_data;
1017
1018        switch (insn->opcode.bytes[0]) {
1019        /* MMIO Write */
1020        case 0x88:
1021                bytes = 1;
1022                fallthrough;
1023        case 0x89:
1024                if (!bytes)
1025                        bytes = insn->opnd_bytes;
1026
1027                reg_data = vc_insn_get_reg(ctxt);
1028                if (!reg_data)
1029                        return ES_DECODE_FAILED;
1030
1031                memcpy(ghcb->shared_buffer, reg_data, bytes);
1032
1033                ret = vc_do_mmio(ghcb, ctxt, bytes, false);
1034                break;
1035
1036        case 0xc6:
1037                bytes = 1;
1038                fallthrough;
1039        case 0xc7:
1040                if (!bytes)
1041                        bytes = insn->opnd_bytes;
1042
1043                memcpy(ghcb->shared_buffer, insn->immediate1.bytes, bytes);
1044
1045                ret = vc_do_mmio(ghcb, ctxt, bytes, false);
1046                break;
1047
1048                /* MMIO Read */
1049        case 0x8a:
1050                bytes = 1;
1051                fallthrough;
1052        case 0x8b:
1053                if (!bytes)
1054                        bytes = insn->opnd_bytes;
1055
1056                ret = vc_do_mmio(ghcb, ctxt, bytes, true);
1057                if (ret)
1058                        break;
1059
1060                reg_data = vc_insn_get_reg(ctxt);
1061                if (!reg_data)
1062                        return ES_DECODE_FAILED;
1063
1064                /* Zero-extend for 32-bit operation */
1065                if (bytes == 4)
1066                        *reg_data = 0;
1067
1068                memcpy(reg_data, ghcb->shared_buffer, bytes);
1069                break;
1070
1071                /* MOVS instruction */
1072        case 0xa4:
1073                bytes = 1;
1074                fallthrough;
1075        case 0xa5:
1076                if (!bytes)
1077                        bytes = insn->opnd_bytes;
1078
1079                ret = vc_handle_mmio_movs(ctxt, bytes);
1080                break;
1081                /* Two-Byte Opcodes */
1082        case 0x0f:
1083                ret = vc_handle_mmio_twobyte_ops(ghcb, ctxt);
1084                break;
1085        default:
1086                ret = ES_UNSUPPORTED;
1087        }
1088
1089        return ret;
1090}
1091
1092static enum es_result vc_handle_dr7_write(struct ghcb *ghcb,
1093                                          struct es_em_ctxt *ctxt)
1094{
1095        struct sev_es_runtime_data *data = this_cpu_read(runtime_data);
1096        long val, *reg = vc_insn_get_rm(ctxt);
1097        enum es_result ret;
1098
1099        if (!reg)
1100                return ES_DECODE_FAILED;
1101
1102        val = *reg;
1103
1104        /* Upper 32 bits must be written as zeroes */
1105        if (val >> 32) {
1106                ctxt->fi.vector = X86_TRAP_GP;
1107                ctxt->fi.error_code = 0;
1108                return ES_EXCEPTION;
1109        }
1110
1111        /* Clear out other reserved bits and set bit 10 */
1112        val = (val & 0xffff23ffL) | BIT(10);
1113
1114        /* Early non-zero writes to DR7 are not supported */
1115        if (!data && (val & ~DR7_RESET_VALUE))
1116                return ES_UNSUPPORTED;
1117
1118        /* Using a value of 0 for ExitInfo1 means RAX holds the value */
1119        ghcb_set_rax(ghcb, val);
1120        ret = sev_es_ghcb_hv_call(ghcb, ctxt, SVM_EXIT_WRITE_DR7, 0, 0);
1121        if (ret != ES_OK)
1122                return ret;
1123
1124        if (data)
1125                data->dr7 = val;
1126
1127        return ES_OK;
1128}
1129
1130static enum es_result vc_handle_dr7_read(struct ghcb *ghcb,
1131                                         struct es_em_ctxt *ctxt)
1132{
1133        struct sev_es_runtime_data *data = this_cpu_read(runtime_data);
1134        long *reg = vc_insn_get_rm(ctxt);
1135
1136        if (!reg)
1137                return ES_DECODE_FAILED;
1138
1139        if (data)
1140                *reg = data->dr7;
1141        else
1142                *reg = DR7_RESET_VALUE;
1143
1144        return ES_OK;
1145}
1146
1147static enum es_result vc_handle_wbinvd(struct ghcb *ghcb,
1148                                       struct es_em_ctxt *ctxt)
1149{
1150        return sev_es_ghcb_hv_call(ghcb, ctxt, SVM_EXIT_WBINVD, 0, 0);
1151}
1152
1153static enum es_result vc_handle_rdpmc(struct ghcb *ghcb, struct es_em_ctxt *ctxt)
1154{
1155        enum es_result ret;
1156
1157        ghcb_set_rcx(ghcb, ctxt->regs->cx);
1158
1159        ret = sev_es_ghcb_hv_call(ghcb, ctxt, SVM_EXIT_RDPMC, 0, 0);
1160        if (ret != ES_OK)
1161                return ret;
1162
1163        if (!(ghcb_rax_is_valid(ghcb) && ghcb_rdx_is_valid(ghcb)))
1164                return ES_VMM_ERROR;
1165
1166        ctxt->regs->ax = ghcb->save.rax;
1167        ctxt->regs->dx = ghcb->save.rdx;
1168
1169        return ES_OK;
1170}
1171
1172static enum es_result vc_handle_monitor(struct ghcb *ghcb,
1173                                        struct es_em_ctxt *ctxt)
1174{
1175        /*
1176         * Treat it as a NOP and do not leak a physical address to the
1177         * hypervisor.
1178         */
1179        return ES_OK;
1180}
1181
1182static enum es_result vc_handle_mwait(struct ghcb *ghcb,
1183                                      struct es_em_ctxt *ctxt)
1184{
1185        /* Treat the same as MONITOR/MONITORX */
1186        return ES_OK;
1187}
1188
1189static enum es_result vc_handle_vmmcall(struct ghcb *ghcb,
1190                                        struct es_em_ctxt *ctxt)
1191{
1192        enum es_result ret;
1193
1194        ghcb_set_rax(ghcb, ctxt->regs->ax);
1195        ghcb_set_cpl(ghcb, user_mode(ctxt->regs) ? 3 : 0);
1196
1197        if (x86_platform.hyper.sev_es_hcall_prepare)
1198                x86_platform.hyper.sev_es_hcall_prepare(ghcb, ctxt->regs);
1199
1200        ret = sev_es_ghcb_hv_call(ghcb, ctxt, SVM_EXIT_VMMCALL, 0, 0);
1201        if (ret != ES_OK)
1202                return ret;
1203
1204        if (!ghcb_rax_is_valid(ghcb))
1205                return ES_VMM_ERROR;
1206
1207        ctxt->regs->ax = ghcb->save.rax;
1208
1209        /*
1210         * Call sev_es_hcall_finish() after regs->ax is already set.
1211         * This allows the hypervisor handler to overwrite it again if
1212         * necessary.
1213         */
1214        if (x86_platform.hyper.sev_es_hcall_finish &&
1215            !x86_platform.hyper.sev_es_hcall_finish(ghcb, ctxt->regs))
1216                return ES_VMM_ERROR;
1217
1218        return ES_OK;
1219}
1220
1221static enum es_result vc_handle_trap_ac(struct ghcb *ghcb,
1222                                        struct es_em_ctxt *ctxt)
1223{
1224        /*
1225         * Calling ecx_alignment_check() directly does not work, because it
1226         * enables IRQs and the GHCB is active. Forward the exception and call
1227         * it later from vc_forward_exception().
1228         */
1229        ctxt->fi.vector = X86_TRAP_AC;
1230        ctxt->fi.error_code = 0;
1231        return ES_EXCEPTION;
1232}
1233
1234static enum es_result vc_handle_exitcode(struct es_em_ctxt *ctxt,
1235                                         struct ghcb *ghcb,
1236                                         unsigned long exit_code)
1237{
1238        enum es_result result;
1239
1240        switch (exit_code) {
1241        case SVM_EXIT_READ_DR7:
1242                result = vc_handle_dr7_read(ghcb, ctxt);
1243                break;
1244        case SVM_EXIT_WRITE_DR7:
1245                result = vc_handle_dr7_write(ghcb, ctxt);
1246                break;
1247        case SVM_EXIT_EXCP_BASE + X86_TRAP_AC:
1248                result = vc_handle_trap_ac(ghcb, ctxt);
1249                break;
1250        case SVM_EXIT_RDTSC:
1251        case SVM_EXIT_RDTSCP:
1252                result = vc_handle_rdtsc(ghcb, ctxt, exit_code);
1253                break;
1254        case SVM_EXIT_RDPMC:
1255                result = vc_handle_rdpmc(ghcb, ctxt);
1256                break;
1257        case SVM_EXIT_INVD:
1258                pr_err_ratelimited("#VC exception for INVD??? Seriously???\n");
1259                result = ES_UNSUPPORTED;
1260                break;
1261        case SVM_EXIT_CPUID:
1262                result = vc_handle_cpuid(ghcb, ctxt);
1263                break;
1264        case SVM_EXIT_IOIO:
1265                result = vc_handle_ioio(ghcb, ctxt);
1266                break;
1267        case SVM_EXIT_MSR:
1268                result = vc_handle_msr(ghcb, ctxt);
1269                break;
1270        case SVM_EXIT_VMMCALL:
1271                result = vc_handle_vmmcall(ghcb, ctxt);
1272                break;
1273        case SVM_EXIT_WBINVD:
1274                result = vc_handle_wbinvd(ghcb, ctxt);
1275                break;
1276        case SVM_EXIT_MONITOR:
1277                result = vc_handle_monitor(ghcb, ctxt);
1278                break;
1279        case SVM_EXIT_MWAIT:
1280                result = vc_handle_mwait(ghcb, ctxt);
1281                break;
1282        case SVM_EXIT_NPF:
1283                result = vc_handle_mmio(ghcb, ctxt);
1284                break;
1285        default:
1286                /*
1287                 * Unexpected #VC exception
1288                 */
1289                result = ES_UNSUPPORTED;
1290        }
1291
1292        return result;
1293}
1294
1295static __always_inline void vc_forward_exception(struct es_em_ctxt *ctxt)
1296{
1297        long error_code = ctxt->fi.error_code;
1298        int trapnr = ctxt->fi.vector;
1299
1300        ctxt->regs->orig_ax = ctxt->fi.error_code;
1301
1302        switch (trapnr) {
1303        case X86_TRAP_GP:
1304                exc_general_protection(ctxt->regs, error_code);
1305                break;
1306        case X86_TRAP_UD:
1307                exc_invalid_op(ctxt->regs);
1308                break;
1309        case X86_TRAP_PF:
1310                write_cr2(ctxt->fi.cr2);
1311                exc_page_fault(ctxt->regs, error_code);
1312                break;
1313        case X86_TRAP_AC:
1314                exc_alignment_check(ctxt->regs, error_code);
1315                break;
1316        default:
1317                pr_emerg("Unsupported exception in #VC instruction emulation - can't continue\n");
1318                BUG();
1319        }
1320}
1321
1322static __always_inline bool on_vc_fallback_stack(struct pt_regs *regs)
1323{
1324        unsigned long sp = (unsigned long)regs;
1325
1326        return (sp >= __this_cpu_ist_bottom_va(VC2) && sp < __this_cpu_ist_top_va(VC2));
1327}
1328
1329static bool vc_raw_handle_exception(struct pt_regs *regs, unsigned long error_code)
1330{
1331        struct ghcb_state state;
1332        struct es_em_ctxt ctxt;
1333        enum es_result result;
1334        struct ghcb *ghcb;
1335        bool ret = true;
1336
1337        ghcb = __sev_get_ghcb(&state);
1338
1339        vc_ghcb_invalidate(ghcb);
1340        result = vc_init_em_ctxt(&ctxt, regs, error_code);
1341
1342        if (result == ES_OK)
1343                result = vc_handle_exitcode(&ctxt, ghcb, error_code);
1344
1345        __sev_put_ghcb(&state);
1346
1347        /* Done - now check the result */
1348        switch (result) {
1349        case ES_OK:
1350                vc_finish_insn(&ctxt);
1351                break;
1352        case ES_UNSUPPORTED:
1353                pr_err_ratelimited("Unsupported exit-code 0x%02lx in #VC exception (IP: 0x%lx)\n",
1354                                   error_code, regs->ip);
1355                ret = false;
1356                break;
1357        case ES_VMM_ERROR:
1358                pr_err_ratelimited("Failure in communication with VMM (exit-code 0x%02lx IP: 0x%lx)\n",
1359                                   error_code, regs->ip);
1360                ret = false;
1361                break;
1362        case ES_DECODE_FAILED:
1363                pr_err_ratelimited("Failed to decode instruction (exit-code 0x%02lx IP: 0x%lx)\n",
1364                                   error_code, regs->ip);
1365                ret = false;
1366                break;
1367        case ES_EXCEPTION:
1368                vc_forward_exception(&ctxt);
1369                break;
1370        case ES_RETRY:
1371                /* Nothing to do */
1372                break;
1373        default:
1374                pr_emerg("Unknown result in %s():%d\n", __func__, result);
1375                /*
1376                 * Emulating the instruction which caused the #VC exception
1377                 * failed - can't continue so print debug information
1378                 */
1379                BUG();
1380        }
1381
1382        return ret;
1383}
1384
1385static __always_inline bool vc_is_db(unsigned long error_code)
1386{
1387        return error_code == SVM_EXIT_EXCP_BASE + X86_TRAP_DB;
1388}
1389
1390/*
1391 * Runtime #VC exception handler when raised from kernel mode. Runs in NMI mode
1392 * and will panic when an error happens.
1393 */
1394DEFINE_IDTENTRY_VC_KERNEL(exc_vmm_communication)
1395{
1396        irqentry_state_t irq_state;
1397
1398        /*
1399         * With the current implementation it is always possible to switch to a
1400         * safe stack because #VC exceptions only happen at known places, like
1401         * intercepted instructions or accesses to MMIO areas/IO ports. They can
1402         * also happen with code instrumentation when the hypervisor intercepts
1403         * #DB, but the critical paths are forbidden to be instrumented, so #DB
1404         * exceptions currently also only happen in safe places.
1405         *
1406         * But keep this here in case the noinstr annotations are violated due
1407         * to bug elsewhere.
1408         */
1409        if (unlikely(on_vc_fallback_stack(regs))) {
1410                instrumentation_begin();
1411                panic("Can't handle #VC exception from unsupported context\n");
1412                instrumentation_end();
1413        }
1414
1415        /*
1416         * Handle #DB before calling into !noinstr code to avoid recursive #DB.
1417         */
1418        if (vc_is_db(error_code)) {
1419                exc_debug(regs);
1420                return;
1421        }
1422
1423        irq_state = irqentry_nmi_enter(regs);
1424
1425        instrumentation_begin();
1426
1427        if (!vc_raw_handle_exception(regs, error_code)) {
1428                /* Show some debug info */
1429                show_regs(regs);
1430
1431                /* Ask hypervisor to sev_es_terminate */
1432                sev_es_terminate(GHCB_SEV_ES_REASON_GENERAL_REQUEST);
1433
1434                /* If that fails and we get here - just panic */
1435                panic("Returned from Terminate-Request to Hypervisor\n");
1436        }
1437
1438        instrumentation_end();
1439        irqentry_nmi_exit(regs, irq_state);
1440}
1441
1442/*
1443 * Runtime #VC exception handler when raised from user mode. Runs in IRQ mode
1444 * and will kill the current task with SIGBUS when an error happens.
1445 */
1446DEFINE_IDTENTRY_VC_USER(exc_vmm_communication)
1447{
1448        /*
1449         * Handle #DB before calling into !noinstr code to avoid recursive #DB.
1450         */
1451        if (vc_is_db(error_code)) {
1452                noist_exc_debug(regs);
1453                return;
1454        }
1455
1456        irqentry_enter_from_user_mode(regs);
1457        instrumentation_begin();
1458
1459        if (!vc_raw_handle_exception(regs, error_code)) {
1460                /*
1461                 * Do not kill the machine if user-space triggered the
1462                 * exception. Send SIGBUS instead and let user-space deal with
1463                 * it.
1464                 */
1465                force_sig_fault(SIGBUS, BUS_OBJERR, (void __user *)0);
1466        }
1467
1468        instrumentation_end();
1469        irqentry_exit_to_user_mode(regs);
1470}
1471
1472bool __init handle_vc_boot_ghcb(struct pt_regs *regs)
1473{
1474        unsigned long exit_code = regs->orig_ax;
1475        struct es_em_ctxt ctxt;
1476        enum es_result result;
1477
1478        /* Do initial setup or terminate the guest */
1479        if (unlikely(boot_ghcb == NULL && !sev_es_setup_ghcb()))
1480                sev_es_terminate(GHCB_SEV_ES_REASON_GENERAL_REQUEST);
1481
1482        vc_ghcb_invalidate(boot_ghcb);
1483
1484        result = vc_init_em_ctxt(&ctxt, regs, exit_code);
1485        if (result == ES_OK)
1486                result = vc_handle_exitcode(&ctxt, boot_ghcb, exit_code);
1487
1488        /* Done - now check the result */
1489        switch (result) {
1490        case ES_OK:
1491                vc_finish_insn(&ctxt);
1492                break;
1493        case ES_UNSUPPORTED:
1494                early_printk("PANIC: Unsupported exit-code 0x%02lx in early #VC exception (IP: 0x%lx)\n",
1495                                exit_code, regs->ip);
1496                goto fail;
1497        case ES_VMM_ERROR:
1498                early_printk("PANIC: Failure in communication with VMM (exit-code 0x%02lx IP: 0x%lx)\n",
1499                                exit_code, regs->ip);
1500                goto fail;
1501        case ES_DECODE_FAILED:
1502                early_printk("PANIC: Failed to decode instruction (exit-code 0x%02lx IP: 0x%lx)\n",
1503                                exit_code, regs->ip);
1504                goto fail;
1505        case ES_EXCEPTION:
1506                vc_early_forward_exception(&ctxt);
1507                break;
1508        case ES_RETRY:
1509                /* Nothing to do */
1510                break;
1511        default:
1512                BUG();
1513        }
1514
1515        return true;
1516
1517fail:
1518        show_regs(regs);
1519
1520        while (true)
1521                halt();
1522}
1523
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.