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 __always_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;
289 /* If instruction ran in kernel mode and the I/O buffer is in kernel space */
290 if (!user_mode(ctxt->regs) && !access_ok(target, size)) {
291 memcpy(dst, buf, size);
298 if (put_user(d1, target))
303 if (put_user(d2, target))
308 if (put_user(d4, target))
313 if (put_user(d8, target))
317 WARN_ONCE(1, "%s: Invalid size: %zu\n", __func__, size);
318 return ES_UNSUPPORTED;
324 if (user_mode(ctxt->regs))
325 error_code |= X86_PF_USER;
327 ctxt->fi.vector = X86_TRAP_PF;
328 ctxt->fi.error_code = error_code;
329 ctxt->fi.cr2 = (unsigned long)dst;
334 static enum es_result vc_read_mem(struct es_em_ctxt *ctxt,
335 char *src, char *buf, size_t size)
337 unsigned long error_code = X86_PF_PROT;
338 char __user *s = (char __user *)src;
344 /* If instruction ran in kernel mode and the I/O buffer is in kernel space */
345 if (!user_mode(ctxt->regs) && !access_ok(s, size)) {
346 memcpy(buf, src, size);
372 WARN_ONCE(1, "%s: Invalid size: %zu\n", __func__, size);
373 return ES_UNSUPPORTED;
379 if (user_mode(ctxt->regs))
380 error_code |= X86_PF_USER;
382 ctxt->fi.vector = X86_TRAP_PF;
383 ctxt->fi.error_code = error_code;
384 ctxt->fi.cr2 = (unsigned long)src;
389 static enum es_result vc_slow_virt_to_phys(struct ghcb *ghcb, struct es_em_ctxt *ctxt,
390 unsigned long vaddr, phys_addr_t *paddr)
392 unsigned long va = (unsigned long)vaddr;
398 pgd = __va(read_cr3_pa());
399 pgd = &pgd[pgd_index(va)];
400 pte = lookup_address_in_pgd(pgd, va, &level);
402 ctxt->fi.vector = X86_TRAP_PF;
403 ctxt->fi.cr2 = vaddr;
404 ctxt->fi.error_code = 0;
406 if (user_mode(ctxt->regs))
407 ctxt->fi.error_code |= X86_PF_USER;
412 if (WARN_ON_ONCE(pte_val(*pte) & _PAGE_ENC))
413 /* Emulated MMIO to/from encrypted memory not supported */
414 return ES_UNSUPPORTED;
416 pa = (phys_addr_t)pte_pfn(*pte) << PAGE_SHIFT;
417 pa |= va & ~page_level_mask(level);
424 /* Include code shared with pre-decompression boot stage */
425 #include "sev-es-shared.c"
427 void noinstr __sev_es_nmi_complete(void)
429 struct ghcb_state state;
432 ghcb = sev_es_get_ghcb(&state);
434 vc_ghcb_invalidate(ghcb);
435 ghcb_set_sw_exit_code(ghcb, SVM_VMGEXIT_NMI_COMPLETE);
436 ghcb_set_sw_exit_info_1(ghcb, 0);
437 ghcb_set_sw_exit_info_2(ghcb, 0);
439 sev_es_wr_ghcb_msr(__pa_nodebug(ghcb));
442 sev_es_put_ghcb(&state);
445 static u64 get_jump_table_addr(void)
447 struct ghcb_state state;
452 local_irq_save(flags);
454 ghcb = sev_es_get_ghcb(&state);
456 vc_ghcb_invalidate(ghcb);
457 ghcb_set_sw_exit_code(ghcb, SVM_VMGEXIT_AP_JUMP_TABLE);
458 ghcb_set_sw_exit_info_1(ghcb, SVM_VMGEXIT_GET_AP_JUMP_TABLE);
459 ghcb_set_sw_exit_info_2(ghcb, 0);
461 sev_es_wr_ghcb_msr(__pa(ghcb));
464 if (ghcb_sw_exit_info_1_is_valid(ghcb) &&
465 ghcb_sw_exit_info_2_is_valid(ghcb))
466 ret = ghcb->save.sw_exit_info_2;
468 sev_es_put_ghcb(&state);
470 local_irq_restore(flags);
475 int sev_es_setup_ap_jump_table(struct real_mode_header *rmh)
477 u16 startup_cs, startup_ip;
478 phys_addr_t jump_table_pa;
480 u16 __iomem *jump_table;
482 jump_table_addr = get_jump_table_addr();
484 /* On UP guests there is no jump table so this is not a failure */
485 if (!jump_table_addr)
488 /* Check if AP Jump Table is page-aligned */
489 if (jump_table_addr & ~PAGE_MASK)
492 jump_table_pa = jump_table_addr & PAGE_MASK;
494 startup_cs = (u16)(rmh->trampoline_start >> 4);
495 startup_ip = (u16)(rmh->sev_es_trampoline_start -
496 rmh->trampoline_start);
498 jump_table = ioremap_encrypted(jump_table_pa, PAGE_SIZE);
502 writew(startup_ip, &jump_table[0]);
503 writew(startup_cs, &jump_table[1]);
511 * This is needed by the OVMF UEFI firmware which will use whatever it finds in
512 * the GHCB MSR as its GHCB to talk to the hypervisor. So make sure the per-cpu
513 * runtime GHCBs used by the kernel are also mapped in the EFI page-table.
515 int __init sev_es_efi_map_ghcbs(pgd_t *pgd)
517 struct sev_es_runtime_data *data;
518 unsigned long address, pflags;
522 if (!sev_es_active())
525 pflags = _PAGE_NX | _PAGE_RW;
527 for_each_possible_cpu(cpu) {
528 data = per_cpu(runtime_data, cpu);
530 address = __pa(&data->ghcb_page);
531 pfn = address >> PAGE_SHIFT;
533 if (kernel_map_pages_in_pgd(pgd, pfn, address, 1, pflags))
540 static enum es_result vc_handle_msr(struct ghcb *ghcb, struct es_em_ctxt *ctxt)
542 struct pt_regs *regs = ctxt->regs;
547 exit_info_1 = (ctxt->insn.opcode.bytes[1] == 0x30) ? 1 : 0;
549 ghcb_set_rcx(ghcb, regs->cx);
551 ghcb_set_rax(ghcb, regs->ax);
552 ghcb_set_rdx(ghcb, regs->dx);
555 ret = sev_es_ghcb_hv_call(ghcb, ctxt, SVM_EXIT_MSR, exit_info_1, 0);
557 if ((ret == ES_OK) && (!exit_info_1)) {
558 regs->ax = ghcb->save.rax;
559 regs->dx = ghcb->save.rdx;
566 * This function runs on the first #VC exception after the kernel
567 * switched to virtual addresses.
569 static bool __init sev_es_setup_ghcb(void)
571 /* First make sure the hypervisor talks a supported protocol. */
572 if (!sev_es_negotiate_protocol())
576 * Clear the boot_ghcb. The first exception comes in before the bss
577 * section is cleared.
579 memset(&boot_ghcb_page, 0, PAGE_SIZE);
581 /* Alright - Make the boot-ghcb public */
582 boot_ghcb = &boot_ghcb_page;
587 #ifdef CONFIG_HOTPLUG_CPU
588 static void sev_es_ap_hlt_loop(void)
590 struct ghcb_state state;
593 ghcb = sev_es_get_ghcb(&state);
596 vc_ghcb_invalidate(ghcb);
597 ghcb_set_sw_exit_code(ghcb, SVM_VMGEXIT_AP_HLT_LOOP);
598 ghcb_set_sw_exit_info_1(ghcb, 0);
599 ghcb_set_sw_exit_info_2(ghcb, 0);
601 sev_es_wr_ghcb_msr(__pa(ghcb));
605 if (ghcb_sw_exit_info_2_is_valid(ghcb) &&
606 ghcb->save.sw_exit_info_2)
610 sev_es_put_ghcb(&state);
614 * Play_dead handler when running under SEV-ES. This is needed because
615 * the hypervisor can't deliver an SIPI request to restart the AP.
616 * Instead the kernel has to issue a VMGEXIT to halt the VCPU until the
617 * hypervisor wakes it up again.
619 static void sev_es_play_dead(void)
623 /* IRQs now disabled */
625 sev_es_ap_hlt_loop();
628 * If we get here, the VCPU was woken up again. Jump to CPU
629 * startup code to get it back online.
633 #else /* CONFIG_HOTPLUG_CPU */
634 #define sev_es_play_dead native_play_dead
635 #endif /* CONFIG_HOTPLUG_CPU */
638 static void __init sev_es_setup_play_dead(void)
640 smp_ops.play_dead = sev_es_play_dead;
643 static inline void sev_es_setup_play_dead(void) { }
646 static void __init alloc_runtime_data(int cpu)
648 struct sev_es_runtime_data *data;
650 data = memblock_alloc(sizeof(*data), PAGE_SIZE);
652 panic("Can't allocate SEV-ES runtime data");
654 per_cpu(runtime_data, cpu) = data;
657 static void __init init_ghcb(int cpu)
659 struct sev_es_runtime_data *data;
662 data = per_cpu(runtime_data, cpu);
664 err = early_set_memory_decrypted((unsigned long)&data->ghcb_page,
665 sizeof(data->ghcb_page));
667 panic("Can't map GHCBs unencrypted");
669 memset(&data->ghcb_page, 0, sizeof(data->ghcb_page));
671 data->ghcb_active = false;
672 data->backup_ghcb_active = false;
675 void __init sev_es_init_vc_handling(void)
679 BUILD_BUG_ON(offsetof(struct sev_es_runtime_data, ghcb_page) % PAGE_SIZE);
681 if (!sev_es_active())
684 if (!sev_es_check_cpu_features())
685 panic("SEV-ES CPU Features missing");
687 /* Enable SEV-ES special handling */
688 static_branch_enable(&sev_es_enable_key);
690 /* Initialize per-cpu GHCB pages */
691 for_each_possible_cpu(cpu) {
692 alloc_runtime_data(cpu);
694 setup_vc_stacks(cpu);
697 sev_es_setup_play_dead();
699 /* Secondary CPUs use the runtime #VC handler */
700 initial_vc_handler = (unsigned long)safe_stack_exc_vmm_communication;
703 static void __init vc_early_forward_exception(struct es_em_ctxt *ctxt)
705 int trapnr = ctxt->fi.vector;
707 if (trapnr == X86_TRAP_PF)
708 native_write_cr2(ctxt->fi.cr2);
710 ctxt->regs->orig_ax = ctxt->fi.error_code;
711 do_early_exception(ctxt->regs, trapnr);
714 static long *vc_insn_get_reg(struct es_em_ctxt *ctxt)
719 reg_array = (long *)ctxt->regs;
720 offset = insn_get_modrm_reg_off(&ctxt->insn, ctxt->regs);
725 offset /= sizeof(long);
727 return reg_array + offset;
730 static long *vc_insn_get_rm(struct es_em_ctxt *ctxt)
735 reg_array = (long *)ctxt->regs;
736 offset = insn_get_modrm_rm_off(&ctxt->insn, ctxt->regs);
741 offset /= sizeof(long);
743 return reg_array + offset;
745 static enum es_result vc_do_mmio(struct ghcb *ghcb, struct es_em_ctxt *ctxt,
746 unsigned int bytes, bool read)
748 u64 exit_code, exit_info_1, exit_info_2;
749 unsigned long ghcb_pa = __pa(ghcb);
754 ref = insn_get_addr_ref(&ctxt->insn, ctxt->regs);
755 if (ref == (void __user *)-1L)
756 return ES_UNSUPPORTED;
758 exit_code = read ? SVM_VMGEXIT_MMIO_READ : SVM_VMGEXIT_MMIO_WRITE;
760 res = vc_slow_virt_to_phys(ghcb, ctxt, (unsigned long)ref, &paddr);
762 if (res == ES_EXCEPTION && !read)
763 ctxt->fi.error_code |= X86_PF_WRITE;
769 /* Can never be greater than 8 */
772 ghcb_set_sw_scratch(ghcb, ghcb_pa + offsetof(struct ghcb, shared_buffer));
774 return sev_es_ghcb_hv_call(ghcb, ctxt, exit_code, exit_info_1, exit_info_2);
777 static enum es_result vc_handle_mmio_twobyte_ops(struct ghcb *ghcb,
778 struct es_em_ctxt *ctxt)
780 struct insn *insn = &ctxt->insn;
781 unsigned int bytes = 0;
786 switch (insn->opcode.bytes[1]) {
787 /* MMIO Read w/ zero-extension */
795 ret = vc_do_mmio(ghcb, ctxt, bytes, true);
799 /* Zero extend based on operand size */
800 reg_data = vc_insn_get_reg(ctxt);
802 return ES_DECODE_FAILED;
804 memset(reg_data, 0, insn->opnd_bytes);
806 memcpy(reg_data, ghcb->shared_buffer, bytes);
809 /* MMIO Read w/ sign-extension */
817 ret = vc_do_mmio(ghcb, ctxt, bytes, true);
821 /* Sign extend based on operand size */
822 reg_data = vc_insn_get_reg(ctxt);
824 return ES_DECODE_FAILED;
827 u8 *val = (u8 *)ghcb->shared_buffer;
829 sign_byte = (*val & 0x80) ? 0xff : 0x00;
831 u16 *val = (u16 *)ghcb->shared_buffer;
833 sign_byte = (*val & 0x8000) ? 0xff : 0x00;
835 memset(reg_data, sign_byte, insn->opnd_bytes);
837 memcpy(reg_data, ghcb->shared_buffer, bytes);
841 ret = ES_UNSUPPORTED;
848 * The MOVS instruction has two memory operands, which raises the
849 * problem that it is not known whether the access to the source or the
850 * destination caused the #VC exception (and hence whether an MMIO read
851 * or write operation needs to be emulated).
853 * Instead of playing games with walking page-tables and trying to guess
854 * whether the source or destination is an MMIO range, split the move
855 * into two operations, a read and a write with only one memory operand.
856 * This will cause a nested #VC exception on the MMIO address which can
859 * This implementation has the benefit that it also supports MOVS where
860 * source _and_ destination are MMIO regions.
862 * It will slow MOVS on MMIO down a lot, but in SEV-ES guests it is a
863 * rare operation. If it turns out to be a performance problem the split
864 * operations can be moved to memcpy_fromio() and memcpy_toio().
866 static enum es_result vc_handle_mmio_movs(struct es_em_ctxt *ctxt,
869 unsigned long ds_base, es_base;
870 unsigned char *src, *dst;
871 unsigned char buffer[8];
876 ds_base = insn_get_seg_base(ctxt->regs, INAT_SEG_REG_DS);
877 es_base = insn_get_seg_base(ctxt->regs, INAT_SEG_REG_ES);
879 if (ds_base == -1L || es_base == -1L) {
880 ctxt->fi.vector = X86_TRAP_GP;
881 ctxt->fi.error_code = 0;
885 src = ds_base + (unsigned char *)ctxt->regs->si;
886 dst = es_base + (unsigned char *)ctxt->regs->di;
888 ret = vc_read_mem(ctxt, src, buffer, bytes);
892 ret = vc_write_mem(ctxt, dst, buffer, bytes);
896 if (ctxt->regs->flags & X86_EFLAGS_DF)
901 ctxt->regs->si += off;
902 ctxt->regs->di += off;
904 rep = insn_has_rep_prefix(&ctxt->insn);
908 if (!rep || ctxt->regs->cx == 0)
914 static enum es_result vc_handle_mmio(struct ghcb *ghcb,
915 struct es_em_ctxt *ctxt)
917 struct insn *insn = &ctxt->insn;
918 unsigned int bytes = 0;
922 switch (insn->opcode.bytes[0]) {
929 bytes = insn->opnd_bytes;
931 reg_data = vc_insn_get_reg(ctxt);
933 return ES_DECODE_FAILED;
935 memcpy(ghcb->shared_buffer, reg_data, bytes);
937 ret = vc_do_mmio(ghcb, ctxt, bytes, false);
945 bytes = insn->opnd_bytes;
947 memcpy(ghcb->shared_buffer, insn->immediate1.bytes, bytes);
949 ret = vc_do_mmio(ghcb, ctxt, bytes, false);
958 bytes = insn->opnd_bytes;
960 ret = vc_do_mmio(ghcb, ctxt, bytes, true);
964 reg_data = vc_insn_get_reg(ctxt);
966 return ES_DECODE_FAILED;
968 /* Zero-extend for 32-bit operation */
972 memcpy(reg_data, ghcb->shared_buffer, bytes);
975 /* MOVS instruction */
981 bytes = insn->opnd_bytes;
983 ret = vc_handle_mmio_movs(ctxt, bytes);
985 /* Two-Byte Opcodes */
987 ret = vc_handle_mmio_twobyte_ops(ghcb, ctxt);
990 ret = ES_UNSUPPORTED;
996 static enum es_result vc_handle_dr7_write(struct ghcb *ghcb,
997 struct es_em_ctxt *ctxt)
999 struct sev_es_runtime_data *data = this_cpu_read(runtime_data);
1000 long val, *reg = vc_insn_get_rm(ctxt);
1004 return ES_DECODE_FAILED;
1008 /* Upper 32 bits must be written as zeroes */
1010 ctxt->fi.vector = X86_TRAP_GP;
1011 ctxt->fi.error_code = 0;
1012 return ES_EXCEPTION;
1015 /* Clear out other reserved bits and set bit 10 */
1016 val = (val & 0xffff23ffL) | BIT(10);
1018 /* Early non-zero writes to DR7 are not supported */
1019 if (!data && (val & ~DR7_RESET_VALUE))
1020 return ES_UNSUPPORTED;
1022 /* Using a value of 0 for ExitInfo1 means RAX holds the value */
1023 ghcb_set_rax(ghcb, val);
1024 ret = sev_es_ghcb_hv_call(ghcb, ctxt, SVM_EXIT_WRITE_DR7, 0, 0);
1034 static enum es_result vc_handle_dr7_read(struct ghcb *ghcb,
1035 struct es_em_ctxt *ctxt)
1037 struct sev_es_runtime_data *data = this_cpu_read(runtime_data);
1038 long *reg = vc_insn_get_rm(ctxt);
1041 return ES_DECODE_FAILED;
1046 *reg = DR7_RESET_VALUE;
1051 static enum es_result vc_handle_wbinvd(struct ghcb *ghcb,
1052 struct es_em_ctxt *ctxt)
1054 return sev_es_ghcb_hv_call(ghcb, ctxt, SVM_EXIT_WBINVD, 0, 0);
1057 static enum es_result vc_handle_rdpmc(struct ghcb *ghcb, struct es_em_ctxt *ctxt)
1061 ghcb_set_rcx(ghcb, ctxt->regs->cx);
1063 ret = sev_es_ghcb_hv_call(ghcb, ctxt, SVM_EXIT_RDPMC, 0, 0);
1067 if (!(ghcb_rax_is_valid(ghcb) && ghcb_rdx_is_valid(ghcb)))
1068 return ES_VMM_ERROR;
1070 ctxt->regs->ax = ghcb->save.rax;
1071 ctxt->regs->dx = ghcb->save.rdx;
1076 static enum es_result vc_handle_monitor(struct ghcb *ghcb,
1077 struct es_em_ctxt *ctxt)
1080 * Treat it as a NOP and do not leak a physical address to the
1086 static enum es_result vc_handle_mwait(struct ghcb *ghcb,
1087 struct es_em_ctxt *ctxt)
1089 /* Treat the same as MONITOR/MONITORX */
1093 static enum es_result vc_handle_vmmcall(struct ghcb *ghcb,
1094 struct es_em_ctxt *ctxt)
1098 ghcb_set_rax(ghcb, ctxt->regs->ax);
1099 ghcb_set_cpl(ghcb, user_mode(ctxt->regs) ? 3 : 0);
1101 if (x86_platform.hyper.sev_es_hcall_prepare)
1102 x86_platform.hyper.sev_es_hcall_prepare(ghcb, ctxt->regs);
1104 ret = sev_es_ghcb_hv_call(ghcb, ctxt, SVM_EXIT_VMMCALL, 0, 0);
1108 if (!ghcb_rax_is_valid(ghcb))
1109 return ES_VMM_ERROR;
1111 ctxt->regs->ax = ghcb->save.rax;
1114 * Call sev_es_hcall_finish() after regs->ax is already set.
1115 * This allows the hypervisor handler to overwrite it again if
1118 if (x86_platform.hyper.sev_es_hcall_finish &&
1119 !x86_platform.hyper.sev_es_hcall_finish(ghcb, ctxt->regs))
1120 return ES_VMM_ERROR;
1125 static enum es_result vc_handle_trap_ac(struct ghcb *ghcb,
1126 struct es_em_ctxt *ctxt)
1129 * Calling ecx_alignment_check() directly does not work, because it
1130 * enables IRQs and the GHCB is active. Forward the exception and call
1131 * it later from vc_forward_exception().
1133 ctxt->fi.vector = X86_TRAP_AC;
1134 ctxt->fi.error_code = 0;
1135 return ES_EXCEPTION;
1138 static __always_inline void vc_handle_trap_db(struct pt_regs *regs)
1140 if (user_mode(regs))
1141 noist_exc_debug(regs);
1146 static enum es_result vc_handle_exitcode(struct es_em_ctxt *ctxt,
1148 unsigned long exit_code)
1150 enum es_result result;
1152 switch (exit_code) {
1153 case SVM_EXIT_READ_DR7:
1154 result = vc_handle_dr7_read(ghcb, ctxt);
1156 case SVM_EXIT_WRITE_DR7:
1157 result = vc_handle_dr7_write(ghcb, ctxt);
1159 case SVM_EXIT_EXCP_BASE + X86_TRAP_AC:
1160 result = vc_handle_trap_ac(ghcb, ctxt);
1162 case SVM_EXIT_RDTSC:
1163 case SVM_EXIT_RDTSCP:
1164 result = vc_handle_rdtsc(ghcb, ctxt, exit_code);
1166 case SVM_EXIT_RDPMC:
1167 result = vc_handle_rdpmc(ghcb, ctxt);
1170 pr_err_ratelimited("#VC exception for INVD??? Seriously???\n");
1171 result = ES_UNSUPPORTED;
1173 case SVM_EXIT_CPUID:
1174 result = vc_handle_cpuid(ghcb, ctxt);
1177 result = vc_handle_ioio(ghcb, ctxt);
1180 result = vc_handle_msr(ghcb, ctxt);
1182 case SVM_EXIT_VMMCALL:
1183 result = vc_handle_vmmcall(ghcb, ctxt);
1185 case SVM_EXIT_WBINVD:
1186 result = vc_handle_wbinvd(ghcb, ctxt);
1188 case SVM_EXIT_MONITOR:
1189 result = vc_handle_monitor(ghcb, ctxt);
1191 case SVM_EXIT_MWAIT:
1192 result = vc_handle_mwait(ghcb, ctxt);
1195 result = vc_handle_mmio(ghcb, ctxt);
1199 * Unexpected #VC exception
1201 result = ES_UNSUPPORTED;
1207 static __always_inline void vc_forward_exception(struct es_em_ctxt *ctxt)
1209 long error_code = ctxt->fi.error_code;
1210 int trapnr = ctxt->fi.vector;
1212 ctxt->regs->orig_ax = ctxt->fi.error_code;
1216 exc_general_protection(ctxt->regs, error_code);
1219 exc_invalid_op(ctxt->regs);
1222 exc_alignment_check(ctxt->regs, error_code);
1225 pr_emerg("Unsupported exception in #VC instruction emulation - can't continue\n");
1230 static __always_inline bool on_vc_fallback_stack(struct pt_regs *regs)
1232 unsigned long sp = (unsigned long)regs;
1234 return (sp >= __this_cpu_ist_bottom_va(VC2) && sp < __this_cpu_ist_top_va(VC2));
1238 * Main #VC exception handler. It is called when the entry code was able to
1239 * switch off the IST to a safe kernel stack.
1241 * With the current implementation it is always possible to switch to a safe
1242 * stack because #VC exceptions only happen at known places, like intercepted
1243 * instructions or accesses to MMIO areas/IO ports. They can also happen with
1244 * code instrumentation when the hypervisor intercepts #DB, but the critical
1245 * paths are forbidden to be instrumented, so #DB exceptions currently also
1246 * only happen in safe places.
1248 DEFINE_IDTENTRY_VC_SAFE_STACK(exc_vmm_communication)
1250 struct sev_es_runtime_data *data = this_cpu_read(runtime_data);
1251 struct ghcb_state state;
1252 struct es_em_ctxt ctxt;
1253 enum es_result result;
1256 lockdep_assert_irqs_disabled();
1259 * Handle #DB before calling into !noinstr code to avoid recursive #DB.
1261 if (error_code == SVM_EXIT_EXCP_BASE + X86_TRAP_DB) {
1262 vc_handle_trap_db(regs);
1266 instrumentation_begin();
1269 * This is invoked through an interrupt gate, so IRQs are disabled. The
1270 * code below might walk page-tables for user or kernel addresses, so
1271 * keep the IRQs disabled to protect us against concurrent TLB flushes.
1274 ghcb = sev_es_get_ghcb(&state);
1277 * Mark GHCBs inactive so that panic() is able to print the
1280 data->ghcb_active = false;
1281 data->backup_ghcb_active = false;
1283 panic("Unable to handle #VC exception! GHCB and Backup GHCB are already in use");
1286 vc_ghcb_invalidate(ghcb);
1287 result = vc_init_em_ctxt(&ctxt, regs, error_code);
1289 if (result == ES_OK)
1290 result = vc_handle_exitcode(&ctxt, ghcb, error_code);
1292 sev_es_put_ghcb(&state);
1294 /* Done - now check the result */
1297 vc_finish_insn(&ctxt);
1299 case ES_UNSUPPORTED:
1300 pr_err_ratelimited("Unsupported exit-code 0x%02lx in early #VC exception (IP: 0x%lx)\n",
1301 error_code, regs->ip);
1304 pr_err_ratelimited("Failure in communication with VMM (exit-code 0x%02lx IP: 0x%lx)\n",
1305 error_code, regs->ip);
1307 case ES_DECODE_FAILED:
1308 pr_err_ratelimited("Failed to decode instruction (exit-code 0x%02lx IP: 0x%lx)\n",
1309 error_code, regs->ip);
1312 vc_forward_exception(&ctxt);
1318 pr_emerg("Unknown result in %s():%d\n", __func__, result);
1320 * Emulating the instruction which caused the #VC exception
1321 * failed - can't continue so print debug information
1327 instrumentation_end();
1332 if (user_mode(regs)) {
1334 * Do not kill the machine if user-space triggered the
1335 * exception. Send SIGBUS instead and let user-space deal with
1338 force_sig_fault(SIGBUS, BUS_OBJERR, (void __user *)0);
1340 pr_emerg("PANIC: Unhandled #VC exception in kernel space (result=%d)\n",
1343 /* Show some debug info */
1346 /* Ask hypervisor to sev_es_terminate */
1347 sev_es_terminate(GHCB_SEV_ES_REASON_GENERAL_REQUEST);
1349 /* If that fails and we get here - just panic */
1350 panic("Returned from Terminate-Request to Hypervisor\n");
1356 /* This handler runs on the #VC fall-back stack. It can cause further #VC exceptions */
1357 DEFINE_IDTENTRY_VC_IST(exc_vmm_communication)
1359 instrumentation_begin();
1360 panic("Can't handle #VC exception from unsupported context\n");
1361 instrumentation_end();
1364 DEFINE_IDTENTRY_VC(exc_vmm_communication)
1366 if (likely(!on_vc_fallback_stack(regs)))
1367 safe_stack_exc_vmm_communication(regs, error_code);
1369 ist_exc_vmm_communication(regs, error_code);
1372 bool __init handle_vc_boot_ghcb(struct pt_regs *regs)
1374 unsigned long exit_code = regs->orig_ax;
1375 struct es_em_ctxt ctxt;
1376 enum es_result result;
1378 /* Do initial setup or terminate the guest */
1379 if (unlikely(boot_ghcb == NULL && !sev_es_setup_ghcb()))
1380 sev_es_terminate(GHCB_SEV_ES_REASON_GENERAL_REQUEST);
1382 vc_ghcb_invalidate(boot_ghcb);
1384 result = vc_init_em_ctxt(&ctxt, regs, exit_code);
1385 if (result == ES_OK)
1386 result = vc_handle_exitcode(&ctxt, boot_ghcb, exit_code);
1388 /* Done - now check the result */
1391 vc_finish_insn(&ctxt);
1393 case ES_UNSUPPORTED:
1394 early_printk("PANIC: Unsupported exit-code 0x%02lx in early #VC exception (IP: 0x%lx)\n",
1395 exit_code, regs->ip);
1398 early_printk("PANIC: Failure in communication with VMM (exit-code 0x%02lx IP: 0x%lx)\n",
1399 exit_code, regs->ip);
1401 case ES_DECODE_FAILED:
1402 early_printk("PANIC: Failed to decode instruction (exit-code 0x%02lx IP: 0x%lx)\n",
1403 exit_code, regs->ip);
1406 vc_early_forward_exception(&ctxt);