1 // SPDX-License-Identifier: GPL-2.0-only
3 * AMD Memory Encryption Support
5 * Copyright (C) 2019 SUSE
7 * Author: Joerg Roedel <jroedel@suse.de>
10 #define pr_fmt(fmt) "SEV-ES: " fmt
12 #include <linux/sched/debug.h> /* For show_regs() */
13 #include <linux/percpu-defs.h>
14 #include <linux/mem_encrypt.h>
15 #include <linux/lockdep.h>
16 #include <linux/printk.h>
17 #include <linux/mm_types.h>
18 #include <linux/set_memory.h>
19 #include <linux/memblock.h>
20 #include <linux/kernel.h>
23 #include <asm/cpu_entry_area.h>
24 #include <asm/stacktrace.h>
25 #include <asm/sev-es.h>
26 #include <asm/insn-eval.h>
27 #include <asm/fpu/internal.h>
28 #include <asm/processor.h>
29 #include <asm/realmode.h>
30 #include <asm/traps.h>
35 #define DR7_RESET_VALUE 0x400
37 /* For early boot hypervisor communication in SEV-ES enabled guests */
38 static struct ghcb boot_ghcb_page __bss_decrypted __aligned(PAGE_SIZE);
41 * Needs to be in the .data section because we need it NULL before bss is
44 static struct ghcb __initdata *boot_ghcb;
46 /* #VC handler runtime per-CPU data */
47 struct sev_es_runtime_data {
48 struct ghcb ghcb_page;
50 /* Physical storage for the per-CPU IST stack of the #VC handler */
51 char ist_stack[EXCEPTION_STKSZ] __aligned(PAGE_SIZE);
54 * Physical storage for the per-CPU fall-back stack of the #VC handler.
55 * The fall-back stack is used when it is not safe to switch back to the
56 * interrupted stack in the #VC entry code.
58 char fallback_stack[EXCEPTION_STKSZ] __aligned(PAGE_SIZE);
61 * Reserve one page per CPU as backup storage for the unencrypted GHCB.
62 * It is needed when an NMI happens while the #VC handler uses the real
63 * GHCB, and the NMI handler itself is causing another #VC exception. In
64 * that case the GHCB content of the first handler needs to be backed up
67 struct ghcb backup_ghcb;
70 * Mark the per-cpu GHCBs as in-use to detect nested #VC exceptions.
71 * There is no need for it to be atomic, because nothing is written to
72 * the GHCB between the read and the write of ghcb_active. So it is safe
73 * to use it when a nested #VC exception happens before the write.
75 * This is necessary for example in the #VC->NMI->#VC case when the NMI
76 * happens while the first #VC handler uses the GHCB. When the NMI code
77 * raises a second #VC handler it might overwrite the contents of the
78 * GHCB written by the first handler. To avoid this the content of the
79 * GHCB is saved and restored when the GHCB is detected to be in use
83 bool backup_ghcb_active;
86 * Cached DR7 value - write it on DR7 writes and return it on reads.
87 * That value will never make it to the real hardware DR7 as debugging
88 * is currently unsupported in SEV-ES guests.
97 static DEFINE_PER_CPU(struct sev_es_runtime_data*, runtime_data);
98 DEFINE_STATIC_KEY_FALSE(sev_es_enable_key);
100 /* Needed in vc_early_forward_exception */
101 void do_early_exception(struct pt_regs *regs, int trapnr);
103 static void __init setup_vc_stacks(int cpu)
105 struct sev_es_runtime_data *data;
106 struct cpu_entry_area *cea;
110 data = per_cpu(runtime_data, cpu);
111 cea = get_cpu_entry_area(cpu);
113 /* Map #VC IST stack */
114 vaddr = CEA_ESTACK_BOT(&cea->estacks, VC);
115 pa = __pa(data->ist_stack);
116 cea_set_pte((void *)vaddr, pa, PAGE_KERNEL);
118 /* Map VC fall-back stack */
119 vaddr = CEA_ESTACK_BOT(&cea->estacks, VC2);
120 pa = __pa(data->fallback_stack);
121 cea_set_pte((void *)vaddr, pa, PAGE_KERNEL);
124 static __always_inline bool on_vc_stack(unsigned long sp)
126 return ((sp >= __this_cpu_ist_bottom_va(VC)) && (sp < __this_cpu_ist_top_va(VC)));
130 * This function handles the case when an NMI is raised in the #VC exception
131 * handler entry code. In this case, the IST entry for #VC must be adjusted, so
132 * that any subsequent #VC exception will not overwrite the stack contents of the
133 * interrupted #VC handler.
135 * The IST entry is adjusted unconditionally so that it can be also be
136 * unconditionally adjusted back in sev_es_ist_exit(). Otherwise a nested
137 * sev_es_ist_exit() call may adjust back the IST entry too early.
139 void noinstr __sev_es_ist_enter(struct pt_regs *regs)
141 unsigned long old_ist, new_ist;
143 /* Read old IST entry */
144 old_ist = __this_cpu_read(cpu_tss_rw.x86_tss.ist[IST_INDEX_VC]);
146 /* Make room on the IST stack */
147 if (on_vc_stack(regs->sp))
148 new_ist = ALIGN_DOWN(regs->sp, 8) - sizeof(old_ist);
150 new_ist = old_ist - sizeof(old_ist);
152 /* Store old IST entry */
153 *(unsigned long *)new_ist = old_ist;
155 /* Set new IST entry */
156 this_cpu_write(cpu_tss_rw.x86_tss.ist[IST_INDEX_VC], new_ist);
159 void noinstr __sev_es_ist_exit(void)
164 ist = __this_cpu_read(cpu_tss_rw.x86_tss.ist[IST_INDEX_VC]);
166 if (WARN_ON(ist == __this_cpu_ist_top_va(VC)))
169 /* Read back old IST entry and write it to the TSS */
170 this_cpu_write(cpu_tss_rw.x86_tss.ist[IST_INDEX_VC], *(unsigned long *)ist);
173 static __always_inline struct ghcb *sev_es_get_ghcb(struct ghcb_state *state)
175 struct sev_es_runtime_data *data;
178 data = this_cpu_read(runtime_data);
179 ghcb = &data->ghcb_page;
181 if (unlikely(data->ghcb_active)) {
182 /* GHCB is already in use - save its contents */
184 if (unlikely(data->backup_ghcb_active))
187 /* Mark backup_ghcb active before writing to it */
188 data->backup_ghcb_active = true;
190 state->ghcb = &data->backup_ghcb;
192 /* Backup GHCB content */
193 *state->ghcb = *ghcb;
196 data->ghcb_active = true;
202 static __always_inline void sev_es_put_ghcb(struct ghcb_state *state)
204 struct sev_es_runtime_data *data;
207 data = this_cpu_read(runtime_data);
208 ghcb = &data->ghcb_page;
211 /* Restore GHCB from Backup */
212 *ghcb = *state->ghcb;
213 data->backup_ghcb_active = false;
216 data->ghcb_active = false;
220 /* Needed in vc_early_forward_exception */
221 void do_early_exception(struct pt_regs *regs, int trapnr);
223 static inline u64 sev_es_rd_ghcb_msr(void)
225 return __rdmsr(MSR_AMD64_SEV_ES_GHCB);
228 static inline void sev_es_wr_ghcb_msr(u64 val)
233 high = (u32)(val >> 32);
235 native_wrmsr(MSR_AMD64_SEV_ES_GHCB, low, high);
238 static int vc_fetch_insn_kernel(struct es_em_ctxt *ctxt,
239 unsigned char *buffer)
241 return copy_from_kernel_nofault(buffer, (unsigned char *)ctxt->regs->ip, MAX_INSN_SIZE);
244 static enum es_result vc_decode_insn(struct es_em_ctxt *ctxt)
246 char buffer[MAX_INSN_SIZE];
250 if (user_mode(ctxt->regs)) {
251 res = insn_fetch_from_user(ctxt->regs, buffer);
253 ctxt->fi.vector = X86_TRAP_PF;
254 ctxt->fi.error_code = X86_PF_INSTR | X86_PF_USER;
255 ctxt->fi.cr2 = ctxt->regs->ip;
259 if (!insn_decode(&ctxt->insn, ctxt->regs, buffer, res))
260 return ES_DECODE_FAILED;
262 res = vc_fetch_insn_kernel(ctxt, buffer);
264 ctxt->fi.vector = X86_TRAP_PF;
265 ctxt->fi.error_code = X86_PF_INSTR;
266 ctxt->fi.cr2 = ctxt->regs->ip;
270 insn_init(&ctxt->insn, buffer, MAX_INSN_SIZE - res, 1);
271 insn_get_length(&ctxt->insn);
274 ret = ctxt->insn.immediate.got ? ES_OK : ES_DECODE_FAILED;
279 static enum es_result vc_write_mem(struct es_em_ctxt *ctxt,
280 char *dst, char *buf, size_t size)
282 unsigned long error_code = X86_PF_PROT | X86_PF_WRITE;
283 char __user *target = (char __user *)dst;
292 if (put_user(d1, target))
297 if (put_user(d2, target))
302 if (put_user(d4, target))
307 if (put_user(d8, target))
311 WARN_ONCE(1, "%s: Invalid size: %zu\n", __func__, size);
312 return ES_UNSUPPORTED;
318 if (user_mode(ctxt->regs))
319 error_code |= X86_PF_USER;
321 ctxt->fi.vector = X86_TRAP_PF;
322 ctxt->fi.error_code = error_code;
323 ctxt->fi.cr2 = (unsigned long)dst;
328 static enum es_result vc_read_mem(struct es_em_ctxt *ctxt,
329 char *src, char *buf, size_t size)
331 unsigned long error_code = X86_PF_PROT;
332 char __user *s = (char __user *)src;
360 WARN_ONCE(1, "%s: Invalid size: %zu\n", __func__, size);
361 return ES_UNSUPPORTED;
367 if (user_mode(ctxt->regs))
368 error_code |= X86_PF_USER;
370 ctxt->fi.vector = X86_TRAP_PF;
371 ctxt->fi.error_code = error_code;
372 ctxt->fi.cr2 = (unsigned long)src;
377 static enum es_result vc_slow_virt_to_phys(struct ghcb *ghcb, struct es_em_ctxt *ctxt,
378 unsigned long vaddr, phys_addr_t *paddr)
380 unsigned long va = (unsigned long)vaddr;
386 pgd = __va(read_cr3_pa());
387 pgd = &pgd[pgd_index(va)];
388 pte = lookup_address_in_pgd(pgd, va, &level);
390 ctxt->fi.vector = X86_TRAP_PF;
391 ctxt->fi.cr2 = vaddr;
392 ctxt->fi.error_code = 0;
394 if (user_mode(ctxt->regs))
395 ctxt->fi.error_code |= X86_PF_USER;
400 if (WARN_ON_ONCE(pte_val(*pte) & _PAGE_ENC))
401 /* Emulated MMIO to/from encrypted memory not supported */
402 return ES_UNSUPPORTED;
404 pa = (phys_addr_t)pte_pfn(*pte) << PAGE_SHIFT;
405 pa |= va & ~page_level_mask(level);
412 /* Include code shared with pre-decompression boot stage */
413 #include "sev-es-shared.c"
415 void noinstr __sev_es_nmi_complete(void)
417 struct ghcb_state state;
420 ghcb = sev_es_get_ghcb(&state);
422 vc_ghcb_invalidate(ghcb);
423 ghcb_set_sw_exit_code(ghcb, SVM_VMGEXIT_NMI_COMPLETE);
424 ghcb_set_sw_exit_info_1(ghcb, 0);
425 ghcb_set_sw_exit_info_2(ghcb, 0);
427 sev_es_wr_ghcb_msr(__pa_nodebug(ghcb));
430 sev_es_put_ghcb(&state);
433 static u64 get_jump_table_addr(void)
435 struct ghcb_state state;
440 local_irq_save(flags);
442 ghcb = sev_es_get_ghcb(&state);
444 vc_ghcb_invalidate(ghcb);
445 ghcb_set_sw_exit_code(ghcb, SVM_VMGEXIT_AP_JUMP_TABLE);
446 ghcb_set_sw_exit_info_1(ghcb, SVM_VMGEXIT_GET_AP_JUMP_TABLE);
447 ghcb_set_sw_exit_info_2(ghcb, 0);
449 sev_es_wr_ghcb_msr(__pa(ghcb));
452 if (ghcb_sw_exit_info_1_is_valid(ghcb) &&
453 ghcb_sw_exit_info_2_is_valid(ghcb))
454 ret = ghcb->save.sw_exit_info_2;
456 sev_es_put_ghcb(&state);
458 local_irq_restore(flags);
463 int sev_es_setup_ap_jump_table(struct real_mode_header *rmh)
465 u16 startup_cs, startup_ip;
466 phys_addr_t jump_table_pa;
468 u16 __iomem *jump_table;
470 jump_table_addr = get_jump_table_addr();
472 /* On UP guests there is no jump table so this is not a failure */
473 if (!jump_table_addr)
476 /* Check if AP Jump Table is page-aligned */
477 if (jump_table_addr & ~PAGE_MASK)
480 jump_table_pa = jump_table_addr & PAGE_MASK;
482 startup_cs = (u16)(rmh->trampoline_start >> 4);
483 startup_ip = (u16)(rmh->sev_es_trampoline_start -
484 rmh->trampoline_start);
486 jump_table = ioremap_encrypted(jump_table_pa, PAGE_SIZE);
490 writew(startup_ip, &jump_table[0]);
491 writew(startup_cs, &jump_table[1]);
499 * This is needed by the OVMF UEFI firmware which will use whatever it finds in
500 * the GHCB MSR as its GHCB to talk to the hypervisor. So make sure the per-cpu
501 * runtime GHCBs used by the kernel are also mapped in the EFI page-table.
503 int __init sev_es_efi_map_ghcbs(pgd_t *pgd)
505 struct sev_es_runtime_data *data;
506 unsigned long address, pflags;
510 if (!sev_es_active())
513 pflags = _PAGE_NX | _PAGE_RW;
515 for_each_possible_cpu(cpu) {
516 data = per_cpu(runtime_data, cpu);
518 address = __pa(&data->ghcb_page);
519 pfn = address >> PAGE_SHIFT;
521 if (kernel_map_pages_in_pgd(pgd, pfn, address, 1, pflags))
528 static enum es_result vc_handle_msr(struct ghcb *ghcb, struct es_em_ctxt *ctxt)
530 struct pt_regs *regs = ctxt->regs;
535 exit_info_1 = (ctxt->insn.opcode.bytes[1] == 0x30) ? 1 : 0;
537 ghcb_set_rcx(ghcb, regs->cx);
539 ghcb_set_rax(ghcb, regs->ax);
540 ghcb_set_rdx(ghcb, regs->dx);
543 ret = sev_es_ghcb_hv_call(ghcb, ctxt, SVM_EXIT_MSR, exit_info_1, 0);
545 if ((ret == ES_OK) && (!exit_info_1)) {
546 regs->ax = ghcb->save.rax;
547 regs->dx = ghcb->save.rdx;
554 * This function runs on the first #VC exception after the kernel
555 * switched to virtual addresses.
557 static bool __init sev_es_setup_ghcb(void)
559 /* First make sure the hypervisor talks a supported protocol. */
560 if (!sev_es_negotiate_protocol())
564 * Clear the boot_ghcb. The first exception comes in before the bss
565 * section is cleared.
567 memset(&boot_ghcb_page, 0, PAGE_SIZE);
569 /* Alright - Make the boot-ghcb public */
570 boot_ghcb = &boot_ghcb_page;
575 #ifdef CONFIG_HOTPLUG_CPU
576 static void sev_es_ap_hlt_loop(void)
578 struct ghcb_state state;
581 ghcb = sev_es_get_ghcb(&state);
584 vc_ghcb_invalidate(ghcb);
585 ghcb_set_sw_exit_code(ghcb, SVM_VMGEXIT_AP_HLT_LOOP);
586 ghcb_set_sw_exit_info_1(ghcb, 0);
587 ghcb_set_sw_exit_info_2(ghcb, 0);
589 sev_es_wr_ghcb_msr(__pa(ghcb));
593 if (ghcb_sw_exit_info_2_is_valid(ghcb) &&
594 ghcb->save.sw_exit_info_2)
598 sev_es_put_ghcb(&state);
602 * Play_dead handler when running under SEV-ES. This is needed because
603 * the hypervisor can't deliver an SIPI request to restart the AP.
604 * Instead the kernel has to issue a VMGEXIT to halt the VCPU until the
605 * hypervisor wakes it up again.
607 static void sev_es_play_dead(void)
611 /* IRQs now disabled */
613 sev_es_ap_hlt_loop();
616 * If we get here, the VCPU was woken up again. Jump to CPU
617 * startup code to get it back online.
621 #else /* CONFIG_HOTPLUG_CPU */
622 #define sev_es_play_dead native_play_dead
623 #endif /* CONFIG_HOTPLUG_CPU */
626 static void __init sev_es_setup_play_dead(void)
628 smp_ops.play_dead = sev_es_play_dead;
631 static inline void sev_es_setup_play_dead(void) { }
634 static void __init alloc_runtime_data(int cpu)
636 struct sev_es_runtime_data *data;
638 data = memblock_alloc(sizeof(*data), PAGE_SIZE);
640 panic("Can't allocate SEV-ES runtime data");
642 per_cpu(runtime_data, cpu) = data;
645 static void __init init_ghcb(int cpu)
647 struct sev_es_runtime_data *data;
650 data = per_cpu(runtime_data, cpu);
652 err = early_set_memory_decrypted((unsigned long)&data->ghcb_page,
653 sizeof(data->ghcb_page));
655 panic("Can't map GHCBs unencrypted");
657 memset(&data->ghcb_page, 0, sizeof(data->ghcb_page));
659 data->ghcb_active = false;
660 data->backup_ghcb_active = false;
663 void __init sev_es_init_vc_handling(void)
667 BUILD_BUG_ON(offsetof(struct sev_es_runtime_data, ghcb_page) % PAGE_SIZE);
669 if (!sev_es_active())
672 if (!sev_es_check_cpu_features())
673 panic("SEV-ES CPU Features missing");
675 /* Enable SEV-ES special handling */
676 static_branch_enable(&sev_es_enable_key);
678 /* Initialize per-cpu GHCB pages */
679 for_each_possible_cpu(cpu) {
680 alloc_runtime_data(cpu);
682 setup_vc_stacks(cpu);
685 sev_es_setup_play_dead();
687 /* Secondary CPUs use the runtime #VC handler */
688 initial_vc_handler = (unsigned long)safe_stack_exc_vmm_communication;
691 static void __init vc_early_forward_exception(struct es_em_ctxt *ctxt)
693 int trapnr = ctxt->fi.vector;
695 if (trapnr == X86_TRAP_PF)
696 native_write_cr2(ctxt->fi.cr2);
698 ctxt->regs->orig_ax = ctxt->fi.error_code;
699 do_early_exception(ctxt->regs, trapnr);
702 static long *vc_insn_get_reg(struct es_em_ctxt *ctxt)
707 reg_array = (long *)ctxt->regs;
708 offset = insn_get_modrm_reg_off(&ctxt->insn, ctxt->regs);
713 offset /= sizeof(long);
715 return reg_array + offset;
718 static long *vc_insn_get_rm(struct es_em_ctxt *ctxt)
723 reg_array = (long *)ctxt->regs;
724 offset = insn_get_modrm_rm_off(&ctxt->insn, ctxt->regs);
729 offset /= sizeof(long);
731 return reg_array + offset;
733 static enum es_result vc_do_mmio(struct ghcb *ghcb, struct es_em_ctxt *ctxt,
734 unsigned int bytes, bool read)
736 u64 exit_code, exit_info_1, exit_info_2;
737 unsigned long ghcb_pa = __pa(ghcb);
742 ref = insn_get_addr_ref(&ctxt->insn, ctxt->regs);
743 if (ref == (void __user *)-1L)
744 return ES_UNSUPPORTED;
746 exit_code = read ? SVM_VMGEXIT_MMIO_READ : SVM_VMGEXIT_MMIO_WRITE;
748 res = vc_slow_virt_to_phys(ghcb, ctxt, (unsigned long)ref, &paddr);
750 if (res == ES_EXCEPTION && !read)
751 ctxt->fi.error_code |= X86_PF_WRITE;
757 /* Can never be greater than 8 */
760 ghcb_set_sw_scratch(ghcb, ghcb_pa + offsetof(struct ghcb, shared_buffer));
762 return sev_es_ghcb_hv_call(ghcb, ctxt, exit_code, exit_info_1, exit_info_2);
765 static enum es_result vc_handle_mmio_twobyte_ops(struct ghcb *ghcb,
766 struct es_em_ctxt *ctxt)
768 struct insn *insn = &ctxt->insn;
769 unsigned int bytes = 0;
774 switch (insn->opcode.bytes[1]) {
775 /* MMIO Read w/ zero-extension */
783 ret = vc_do_mmio(ghcb, ctxt, bytes, true);
787 /* Zero extend based on operand size */
788 reg_data = vc_insn_get_reg(ctxt);
790 return ES_DECODE_FAILED;
792 memset(reg_data, 0, insn->opnd_bytes);
794 memcpy(reg_data, ghcb->shared_buffer, bytes);
797 /* MMIO Read w/ sign-extension */
805 ret = vc_do_mmio(ghcb, ctxt, bytes, true);
809 /* Sign extend based on operand size */
810 reg_data = vc_insn_get_reg(ctxt);
812 return ES_DECODE_FAILED;
815 u8 *val = (u8 *)ghcb->shared_buffer;
817 sign_byte = (*val & 0x80) ? 0xff : 0x00;
819 u16 *val = (u16 *)ghcb->shared_buffer;
821 sign_byte = (*val & 0x8000) ? 0xff : 0x00;
823 memset(reg_data, sign_byte, insn->opnd_bytes);
825 memcpy(reg_data, ghcb->shared_buffer, bytes);
829 ret = ES_UNSUPPORTED;
836 * The MOVS instruction has two memory operands, which raises the
837 * problem that it is not known whether the access to the source or the
838 * destination caused the #VC exception (and hence whether an MMIO read
839 * or write operation needs to be emulated).
841 * Instead of playing games with walking page-tables and trying to guess
842 * whether the source or destination is an MMIO range, split the move
843 * into two operations, a read and a write with only one memory operand.
844 * This will cause a nested #VC exception on the MMIO address which can
847 * This implementation has the benefit that it also supports MOVS where
848 * source _and_ destination are MMIO regions.
850 * It will slow MOVS on MMIO down a lot, but in SEV-ES guests it is a
851 * rare operation. If it turns out to be a performance problem the split
852 * operations can be moved to memcpy_fromio() and memcpy_toio().
854 static enum es_result vc_handle_mmio_movs(struct es_em_ctxt *ctxt,
857 unsigned long ds_base, es_base;
858 unsigned char *src, *dst;
859 unsigned char buffer[8];
864 ds_base = insn_get_seg_base(ctxt->regs, INAT_SEG_REG_DS);
865 es_base = insn_get_seg_base(ctxt->regs, INAT_SEG_REG_ES);
867 if (ds_base == -1L || es_base == -1L) {
868 ctxt->fi.vector = X86_TRAP_GP;
869 ctxt->fi.error_code = 0;
873 src = ds_base + (unsigned char *)ctxt->regs->si;
874 dst = es_base + (unsigned char *)ctxt->regs->di;
876 ret = vc_read_mem(ctxt, src, buffer, bytes);
880 ret = vc_write_mem(ctxt, dst, buffer, bytes);
884 if (ctxt->regs->flags & X86_EFLAGS_DF)
889 ctxt->regs->si += off;
890 ctxt->regs->di += off;
892 rep = insn_has_rep_prefix(&ctxt->insn);
896 if (!rep || ctxt->regs->cx == 0)
902 static enum es_result vc_handle_mmio(struct ghcb *ghcb,
903 struct es_em_ctxt *ctxt)
905 struct insn *insn = &ctxt->insn;
906 unsigned int bytes = 0;
910 switch (insn->opcode.bytes[0]) {
917 bytes = insn->opnd_bytes;
919 reg_data = vc_insn_get_reg(ctxt);
921 return ES_DECODE_FAILED;
923 memcpy(ghcb->shared_buffer, reg_data, bytes);
925 ret = vc_do_mmio(ghcb, ctxt, bytes, false);
933 bytes = insn->opnd_bytes;
935 memcpy(ghcb->shared_buffer, insn->immediate1.bytes, bytes);
937 ret = vc_do_mmio(ghcb, ctxt, bytes, false);
946 bytes = insn->opnd_bytes;
948 ret = vc_do_mmio(ghcb, ctxt, bytes, true);
952 reg_data = vc_insn_get_reg(ctxt);
954 return ES_DECODE_FAILED;
956 /* Zero-extend for 32-bit operation */
960 memcpy(reg_data, ghcb->shared_buffer, bytes);
963 /* MOVS instruction */
969 bytes = insn->opnd_bytes;
971 ret = vc_handle_mmio_movs(ctxt, bytes);
973 /* Two-Byte Opcodes */
975 ret = vc_handle_mmio_twobyte_ops(ghcb, ctxt);
978 ret = ES_UNSUPPORTED;
984 static enum es_result vc_handle_dr7_write(struct ghcb *ghcb,
985 struct es_em_ctxt *ctxt)
987 struct sev_es_runtime_data *data = this_cpu_read(runtime_data);
988 long val, *reg = vc_insn_get_rm(ctxt);
992 return ES_DECODE_FAILED;
996 /* Upper 32 bits must be written as zeroes */
998 ctxt->fi.vector = X86_TRAP_GP;
999 ctxt->fi.error_code = 0;
1000 return ES_EXCEPTION;
1003 /* Clear out other reserved bits and set bit 10 */
1004 val = (val & 0xffff23ffL) | BIT(10);
1006 /* Early non-zero writes to DR7 are not supported */
1007 if (!data && (val & ~DR7_RESET_VALUE))
1008 return ES_UNSUPPORTED;
1010 /* Using a value of 0 for ExitInfo1 means RAX holds the value */
1011 ghcb_set_rax(ghcb, val);
1012 ret = sev_es_ghcb_hv_call(ghcb, ctxt, SVM_EXIT_WRITE_DR7, 0, 0);
1022 static enum es_result vc_handle_dr7_read(struct ghcb *ghcb,
1023 struct es_em_ctxt *ctxt)
1025 struct sev_es_runtime_data *data = this_cpu_read(runtime_data);
1026 long *reg = vc_insn_get_rm(ctxt);
1029 return ES_DECODE_FAILED;
1034 *reg = DR7_RESET_VALUE;
1039 static enum es_result vc_handle_wbinvd(struct ghcb *ghcb,
1040 struct es_em_ctxt *ctxt)
1042 return sev_es_ghcb_hv_call(ghcb, ctxt, SVM_EXIT_WBINVD, 0, 0);
1045 static enum es_result vc_handle_rdpmc(struct ghcb *ghcb, struct es_em_ctxt *ctxt)
1049 ghcb_set_rcx(ghcb, ctxt->regs->cx);
1051 ret = sev_es_ghcb_hv_call(ghcb, ctxt, SVM_EXIT_RDPMC, 0, 0);
1055 if (!(ghcb_rax_is_valid(ghcb) && ghcb_rdx_is_valid(ghcb)))
1056 return ES_VMM_ERROR;
1058 ctxt->regs->ax = ghcb->save.rax;
1059 ctxt->regs->dx = ghcb->save.rdx;
1064 static enum es_result vc_handle_monitor(struct ghcb *ghcb,
1065 struct es_em_ctxt *ctxt)
1068 * Treat it as a NOP and do not leak a physical address to the
1074 static enum es_result vc_handle_mwait(struct ghcb *ghcb,
1075 struct es_em_ctxt *ctxt)
1077 /* Treat the same as MONITOR/MONITORX */
1081 static enum es_result vc_handle_vmmcall(struct ghcb *ghcb,
1082 struct es_em_ctxt *ctxt)
1086 ghcb_set_rax(ghcb, ctxt->regs->ax);
1087 ghcb_set_cpl(ghcb, user_mode(ctxt->regs) ? 3 : 0);
1089 if (x86_platform.hyper.sev_es_hcall_prepare)
1090 x86_platform.hyper.sev_es_hcall_prepare(ghcb, ctxt->regs);
1092 ret = sev_es_ghcb_hv_call(ghcb, ctxt, SVM_EXIT_VMMCALL, 0, 0);
1096 if (!ghcb_rax_is_valid(ghcb))
1097 return ES_VMM_ERROR;
1099 ctxt->regs->ax = ghcb->save.rax;
1102 * Call sev_es_hcall_finish() after regs->ax is already set.
1103 * This allows the hypervisor handler to overwrite it again if
1106 if (x86_platform.hyper.sev_es_hcall_finish &&
1107 !x86_platform.hyper.sev_es_hcall_finish(ghcb, ctxt->regs))
1108 return ES_VMM_ERROR;
1113 static enum es_result vc_handle_trap_ac(struct ghcb *ghcb,
1114 struct es_em_ctxt *ctxt)
1117 * Calling ecx_alignment_check() directly does not work, because it
1118 * enables IRQs and the GHCB is active. Forward the exception and call
1119 * it later from vc_forward_exception().
1121 ctxt->fi.vector = X86_TRAP_AC;
1122 ctxt->fi.error_code = 0;
1123 return ES_EXCEPTION;
1126 static __always_inline void vc_handle_trap_db(struct pt_regs *regs)
1128 if (user_mode(regs))
1129 noist_exc_debug(regs);
1134 static enum es_result vc_handle_exitcode(struct es_em_ctxt *ctxt,
1136 unsigned long exit_code)
1138 enum es_result result;
1140 switch (exit_code) {
1141 case SVM_EXIT_READ_DR7:
1142 result = vc_handle_dr7_read(ghcb, ctxt);
1144 case SVM_EXIT_WRITE_DR7:
1145 result = vc_handle_dr7_write(ghcb, ctxt);
1147 case SVM_EXIT_EXCP_BASE + X86_TRAP_AC:
1148 result = vc_handle_trap_ac(ghcb, ctxt);
1150 case SVM_EXIT_RDTSC:
1151 case SVM_EXIT_RDTSCP:
1152 result = vc_handle_rdtsc(ghcb, ctxt, exit_code);
1154 case SVM_EXIT_RDPMC:
1155 result = vc_handle_rdpmc(ghcb, ctxt);
1158 pr_err_ratelimited("#VC exception for INVD??? Seriously???\n");
1159 result = ES_UNSUPPORTED;
1161 case SVM_EXIT_CPUID:
1162 result = vc_handle_cpuid(ghcb, ctxt);
1165 result = vc_handle_ioio(ghcb, ctxt);
1168 result = vc_handle_msr(ghcb, ctxt);
1170 case SVM_EXIT_VMMCALL:
1171 result = vc_handle_vmmcall(ghcb, ctxt);
1173 case SVM_EXIT_WBINVD:
1174 result = vc_handle_wbinvd(ghcb, ctxt);
1176 case SVM_EXIT_MONITOR:
1177 result = vc_handle_monitor(ghcb, ctxt);
1179 case SVM_EXIT_MWAIT:
1180 result = vc_handle_mwait(ghcb, ctxt);
1183 result = vc_handle_mmio(ghcb, ctxt);
1187 * Unexpected #VC exception
1189 result = ES_UNSUPPORTED;
1195 static __always_inline void vc_forward_exception(struct es_em_ctxt *ctxt)
1197 long error_code = ctxt->fi.error_code;
1198 int trapnr = ctxt->fi.vector;
1200 ctxt->regs->orig_ax = ctxt->fi.error_code;
1204 exc_general_protection(ctxt->regs, error_code);
1207 exc_invalid_op(ctxt->regs);
1210 exc_alignment_check(ctxt->regs, error_code);
1213 pr_emerg("Unsupported exception in #VC instruction emulation - can't continue\n");
1218 static __always_inline bool on_vc_fallback_stack(struct pt_regs *regs)
1220 unsigned long sp = (unsigned long)regs;
1222 return (sp >= __this_cpu_ist_bottom_va(VC2) && sp < __this_cpu_ist_top_va(VC2));
1226 * Main #VC exception handler. It is called when the entry code was able to
1227 * switch off the IST to a safe kernel stack.
1229 * With the current implementation it is always possible to switch to a safe
1230 * stack because #VC exceptions only happen at known places, like intercepted
1231 * instructions or accesses to MMIO areas/IO ports. They can also happen with
1232 * code instrumentation when the hypervisor intercepts #DB, but the critical
1233 * paths are forbidden to be instrumented, so #DB exceptions currently also
1234 * only happen in safe places.
1236 DEFINE_IDTENTRY_VC_SAFE_STACK(exc_vmm_communication)
1238 struct sev_es_runtime_data *data = this_cpu_read(runtime_data);
1239 struct ghcb_state state;
1240 struct es_em_ctxt ctxt;
1241 enum es_result result;
1244 lockdep_assert_irqs_disabled();
1247 * Handle #DB before calling into !noinstr code to avoid recursive #DB.
1249 if (error_code == SVM_EXIT_EXCP_BASE + X86_TRAP_DB) {
1250 vc_handle_trap_db(regs);
1254 instrumentation_begin();
1257 * This is invoked through an interrupt gate, so IRQs are disabled. The
1258 * code below might walk page-tables for user or kernel addresses, so
1259 * keep the IRQs disabled to protect us against concurrent TLB flushes.
1262 ghcb = sev_es_get_ghcb(&state);
1265 * Mark GHCBs inactive so that panic() is able to print the
1268 data->ghcb_active = false;
1269 data->backup_ghcb_active = false;
1271 panic("Unable to handle #VC exception! GHCB and Backup GHCB are already in use");
1274 vc_ghcb_invalidate(ghcb);
1275 result = vc_init_em_ctxt(&ctxt, regs, error_code);
1277 if (result == ES_OK)
1278 result = vc_handle_exitcode(&ctxt, ghcb, error_code);
1280 sev_es_put_ghcb(&state);
1282 /* Done - now check the result */
1285 vc_finish_insn(&ctxt);
1287 case ES_UNSUPPORTED:
1288 pr_err_ratelimited("Unsupported exit-code 0x%02lx in early #VC exception (IP: 0x%lx)\n",
1289 error_code, regs->ip);
1292 pr_err_ratelimited("Failure in communication with VMM (exit-code 0x%02lx IP: 0x%lx)\n",
1293 error_code, regs->ip);
1295 case ES_DECODE_FAILED:
1296 pr_err_ratelimited("Failed to decode instruction (exit-code 0x%02lx IP: 0x%lx)\n",
1297 error_code, regs->ip);
1300 vc_forward_exception(&ctxt);
1306 pr_emerg("Unknown result in %s():%d\n", __func__, result);
1308 * Emulating the instruction which caused the #VC exception
1309 * failed - can't continue so print debug information
1315 instrumentation_end();
1320 if (user_mode(regs)) {
1322 * Do not kill the machine if user-space triggered the
1323 * exception. Send SIGBUS instead and let user-space deal with
1326 force_sig_fault(SIGBUS, BUS_OBJERR, (void __user *)0);
1328 pr_emerg("PANIC: Unhandled #VC exception in kernel space (result=%d)\n",
1331 /* Show some debug info */
1334 /* Ask hypervisor to sev_es_terminate */
1335 sev_es_terminate(GHCB_SEV_ES_REASON_GENERAL_REQUEST);
1337 /* If that fails and we get here - just panic */
1338 panic("Returned from Terminate-Request to Hypervisor\n");
1344 /* This handler runs on the #VC fall-back stack. It can cause further #VC exceptions */
1345 DEFINE_IDTENTRY_VC_IST(exc_vmm_communication)
1347 instrumentation_begin();
1348 panic("Can't handle #VC exception from unsupported context\n");
1349 instrumentation_end();
1352 DEFINE_IDTENTRY_VC(exc_vmm_communication)
1354 if (likely(!on_vc_fallback_stack(regs)))
1355 safe_stack_exc_vmm_communication(regs, error_code);
1357 ist_exc_vmm_communication(regs, error_code);
1360 bool __init handle_vc_boot_ghcb(struct pt_regs *regs)
1362 unsigned long exit_code = regs->orig_ax;
1363 struct es_em_ctxt ctxt;
1364 enum es_result result;
1366 /* Do initial setup or terminate the guest */
1367 if (unlikely(boot_ghcb == NULL && !sev_es_setup_ghcb()))
1368 sev_es_terminate(GHCB_SEV_ES_REASON_GENERAL_REQUEST);
1370 vc_ghcb_invalidate(boot_ghcb);
1372 result = vc_init_em_ctxt(&ctxt, regs, exit_code);
1373 if (result == ES_OK)
1374 result = vc_handle_exitcode(&ctxt, boot_ghcb, exit_code);
1376 /* Done - now check the result */
1379 vc_finish_insn(&ctxt);
1381 case ES_UNSUPPORTED:
1382 early_printk("PANIC: Unsupported exit-code 0x%02lx in early #VC exception (IP: 0x%lx)\n",
1383 exit_code, regs->ip);
1386 early_printk("PANIC: Failure in communication with VMM (exit-code 0x%02lx IP: 0x%lx)\n",
1387 exit_code, regs->ip);
1389 case ES_DECODE_FAILED:
1390 early_printk("PANIC: Failed to decode instruction (exit-code 0x%02lx IP: 0x%lx)\n",
1391 exit_code, regs->ip);
1394 vc_early_forward_exception(&ctxt);