xref: /linux/arch/x86/kvm/smm.c (revision 9e56ff53b4115875667760445b028357848b4748)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
3 
4 #include <linux/kvm_host.h>
5 #include "x86.h"
6 #include "kvm_cache_regs.h"
7 #include "kvm_emulate.h"
8 #include "smm.h"
9 #include "cpuid.h"
10 #include "trace.h"
11 
12 #define CHECK_SMRAM32_OFFSET(field, offset) \
13 	ASSERT_STRUCT_OFFSET(struct kvm_smram_state_32, field, offset - 0xFE00)
14 
15 #define CHECK_SMRAM64_OFFSET(field, offset) \
16 	ASSERT_STRUCT_OFFSET(struct kvm_smram_state_64, field, offset - 0xFE00)
17 
18 static void check_smram_offsets(void)
19 {
20 	/* 32 bit SMRAM image */
21 	CHECK_SMRAM32_OFFSET(reserved1,			0xFE00);
22 	CHECK_SMRAM32_OFFSET(smbase,			0xFEF8);
23 	CHECK_SMRAM32_OFFSET(smm_revision,		0xFEFC);
24 	CHECK_SMRAM32_OFFSET(io_inst_restart,		0xFF00);
25 	CHECK_SMRAM32_OFFSET(auto_hlt_restart,		0xFF02);
26 	CHECK_SMRAM32_OFFSET(io_restart_rdi,		0xFF04);
27 	CHECK_SMRAM32_OFFSET(io_restart_rcx,		0xFF08);
28 	CHECK_SMRAM32_OFFSET(io_restart_rsi,		0xFF0C);
29 	CHECK_SMRAM32_OFFSET(io_restart_rip,		0xFF10);
30 	CHECK_SMRAM32_OFFSET(cr4,			0xFF14);
31 	CHECK_SMRAM32_OFFSET(reserved2,			0xFF18);
32 	CHECK_SMRAM32_OFFSET(int_shadow,		0xFF1A);
33 	CHECK_SMRAM32_OFFSET(reserved3,			0xFF1B);
34 	CHECK_SMRAM32_OFFSET(ds,			0xFF2C);
35 	CHECK_SMRAM32_OFFSET(fs,			0xFF38);
36 	CHECK_SMRAM32_OFFSET(gs,			0xFF44);
37 	CHECK_SMRAM32_OFFSET(idtr,			0xFF50);
38 	CHECK_SMRAM32_OFFSET(tr,			0xFF5C);
39 	CHECK_SMRAM32_OFFSET(gdtr,			0xFF6C);
40 	CHECK_SMRAM32_OFFSET(ldtr,			0xFF78);
41 	CHECK_SMRAM32_OFFSET(es,			0xFF84);
42 	CHECK_SMRAM32_OFFSET(cs,			0xFF90);
43 	CHECK_SMRAM32_OFFSET(ss,			0xFF9C);
44 	CHECK_SMRAM32_OFFSET(es_sel,			0xFFA8);
45 	CHECK_SMRAM32_OFFSET(cs_sel,			0xFFAC);
46 	CHECK_SMRAM32_OFFSET(ss_sel,			0xFFB0);
47 	CHECK_SMRAM32_OFFSET(ds_sel,			0xFFB4);
48 	CHECK_SMRAM32_OFFSET(fs_sel,			0xFFB8);
49 	CHECK_SMRAM32_OFFSET(gs_sel,			0xFFBC);
50 	CHECK_SMRAM32_OFFSET(ldtr_sel,			0xFFC0);
51 	CHECK_SMRAM32_OFFSET(tr_sel,			0xFFC4);
52 	CHECK_SMRAM32_OFFSET(dr7,			0xFFC8);
53 	CHECK_SMRAM32_OFFSET(dr6,			0xFFCC);
54 	CHECK_SMRAM32_OFFSET(gprs,			0xFFD0);
55 	CHECK_SMRAM32_OFFSET(eip,			0xFFF0);
56 	CHECK_SMRAM32_OFFSET(eflags,			0xFFF4);
57 	CHECK_SMRAM32_OFFSET(cr3,			0xFFF8);
58 	CHECK_SMRAM32_OFFSET(cr0,			0xFFFC);
59 
60 	/* 64 bit SMRAM image */
61 	CHECK_SMRAM64_OFFSET(es,			0xFE00);
62 	CHECK_SMRAM64_OFFSET(cs,			0xFE10);
63 	CHECK_SMRAM64_OFFSET(ss,			0xFE20);
64 	CHECK_SMRAM64_OFFSET(ds,			0xFE30);
65 	CHECK_SMRAM64_OFFSET(fs,			0xFE40);
66 	CHECK_SMRAM64_OFFSET(gs,			0xFE50);
67 	CHECK_SMRAM64_OFFSET(gdtr,			0xFE60);
68 	CHECK_SMRAM64_OFFSET(ldtr,			0xFE70);
69 	CHECK_SMRAM64_OFFSET(idtr,			0xFE80);
70 	CHECK_SMRAM64_OFFSET(tr,			0xFE90);
71 	CHECK_SMRAM64_OFFSET(io_restart_rip,		0xFEA0);
72 	CHECK_SMRAM64_OFFSET(io_restart_rcx,		0xFEA8);
73 	CHECK_SMRAM64_OFFSET(io_restart_rsi,		0xFEB0);
74 	CHECK_SMRAM64_OFFSET(io_restart_rdi,		0xFEB8);
75 	CHECK_SMRAM64_OFFSET(io_restart_dword,		0xFEC0);
76 	CHECK_SMRAM64_OFFSET(reserved1,			0xFEC4);
77 	CHECK_SMRAM64_OFFSET(io_inst_restart,		0xFEC8);
78 	CHECK_SMRAM64_OFFSET(auto_hlt_restart,		0xFEC9);
79 	CHECK_SMRAM64_OFFSET(amd_nmi_mask,		0xFECA);
80 	CHECK_SMRAM64_OFFSET(int_shadow,		0xFECB);
81 	CHECK_SMRAM64_OFFSET(reserved2,			0xFECC);
82 	CHECK_SMRAM64_OFFSET(efer,			0xFED0);
83 	CHECK_SMRAM64_OFFSET(svm_guest_flag,		0xFED8);
84 	CHECK_SMRAM64_OFFSET(svm_guest_vmcb_gpa,	0xFEE0);
85 	CHECK_SMRAM64_OFFSET(svm_guest_virtual_int,	0xFEE8);
86 	CHECK_SMRAM64_OFFSET(reserved3,			0xFEF0);
87 	CHECK_SMRAM64_OFFSET(smm_revison,		0xFEFC);
88 	CHECK_SMRAM64_OFFSET(smbase,			0xFF00);
89 	CHECK_SMRAM64_OFFSET(reserved4,			0xFF04);
90 	CHECK_SMRAM64_OFFSET(ssp,			0xFF18);
91 	CHECK_SMRAM64_OFFSET(svm_guest_pat,		0xFF20);
92 	CHECK_SMRAM64_OFFSET(svm_host_efer,		0xFF28);
93 	CHECK_SMRAM64_OFFSET(svm_host_cr4,		0xFF30);
94 	CHECK_SMRAM64_OFFSET(svm_host_cr3,		0xFF38);
95 	CHECK_SMRAM64_OFFSET(svm_host_cr0,		0xFF40);
96 	CHECK_SMRAM64_OFFSET(cr4,			0xFF48);
97 	CHECK_SMRAM64_OFFSET(cr3,			0xFF50);
98 	CHECK_SMRAM64_OFFSET(cr0,			0xFF58);
99 	CHECK_SMRAM64_OFFSET(dr7,			0xFF60);
100 	CHECK_SMRAM64_OFFSET(dr6,			0xFF68);
101 	CHECK_SMRAM64_OFFSET(rflags,			0xFF70);
102 	CHECK_SMRAM64_OFFSET(rip,			0xFF78);
103 	CHECK_SMRAM64_OFFSET(gprs,			0xFF80);
104 
105 	BUILD_BUG_ON(sizeof(union kvm_smram) != 512);
106 }
107 
108 #undef CHECK_SMRAM64_OFFSET
109 #undef CHECK_SMRAM32_OFFSET
110 
111 
112 void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm)
113 {
114 	trace_kvm_smm_transition(vcpu->vcpu_id, vcpu->arch.smbase, entering_smm);
115 
116 	if (entering_smm) {
117 		vcpu->arch.hflags |= HF_SMM_MASK;
118 	} else {
119 		vcpu->arch.hflags &= ~(HF_SMM_MASK | HF_SMM_INSIDE_NMI_MASK);
120 
121 		/* Process a latched INIT or SMI, if any.  */
122 		kvm_make_request(KVM_REQ_EVENT, vcpu);
123 
124 		/*
125 		 * Even if KVM_SET_SREGS2 loaded PDPTRs out of band,
126 		 * on SMM exit we still need to reload them from
127 		 * guest memory
128 		 */
129 		vcpu->arch.pdptrs_from_userspace = false;
130 	}
131 
132 	kvm_mmu_reset_context(vcpu);
133 }
134 
135 void process_smi(struct kvm_vcpu *vcpu)
136 {
137 	vcpu->arch.smi_pending = true;
138 	kvm_make_request(KVM_REQ_EVENT, vcpu);
139 }
140 
141 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
142 {
143 	u32 flags = 0;
144 	flags |= seg->g       << 23;
145 	flags |= seg->db      << 22;
146 	flags |= seg->l       << 21;
147 	flags |= seg->avl     << 20;
148 	flags |= seg->present << 15;
149 	flags |= seg->dpl     << 13;
150 	flags |= seg->s       << 12;
151 	flags |= seg->type    << 8;
152 	return flags;
153 }
154 
155 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu,
156 				  struct kvm_smm_seg_state_32 *state,
157 				  u32 *selector, int n)
158 {
159 	struct kvm_segment seg;
160 
161 	kvm_get_segment(vcpu, &seg, n);
162 	*selector = seg.selector;
163 	state->base = seg.base;
164 	state->limit = seg.limit;
165 	state->flags = enter_smm_get_segment_flags(&seg);
166 }
167 
168 #ifdef CONFIG_X86_64
169 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu,
170 				  struct kvm_smm_seg_state_64 *state,
171 				  int n)
172 {
173 	struct kvm_segment seg;
174 
175 	kvm_get_segment(vcpu, &seg, n);
176 	state->selector = seg.selector;
177 	state->attributes = enter_smm_get_segment_flags(&seg) >> 8;
178 	state->limit = seg.limit;
179 	state->base = seg.base;
180 }
181 #endif
182 
183 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu,
184 				    struct kvm_smram_state_32 *smram)
185 {
186 	struct desc_ptr dt;
187 	unsigned long val;
188 	int i;
189 
190 	smram->cr0     = kvm_read_cr0(vcpu);
191 	smram->cr3     = kvm_read_cr3(vcpu);
192 	smram->eflags  = kvm_get_rflags(vcpu);
193 	smram->eip     = kvm_rip_read(vcpu);
194 
195 	for (i = 0; i < 8; i++)
196 		smram->gprs[i] = kvm_register_read_raw(vcpu, i);
197 
198 	kvm_get_dr(vcpu, 6, &val);
199 	smram->dr6     = (u32)val;
200 	kvm_get_dr(vcpu, 7, &val);
201 	smram->dr7     = (u32)val;
202 
203 	enter_smm_save_seg_32(vcpu, &smram->tr, &smram->tr_sel, VCPU_SREG_TR);
204 	enter_smm_save_seg_32(vcpu, &smram->ldtr, &smram->ldtr_sel, VCPU_SREG_LDTR);
205 
206 	static_call(kvm_x86_get_gdt)(vcpu, &dt);
207 	smram->gdtr.base = dt.address;
208 	smram->gdtr.limit = dt.size;
209 
210 	static_call(kvm_x86_get_idt)(vcpu, &dt);
211 	smram->idtr.base = dt.address;
212 	smram->idtr.limit = dt.size;
213 
214 	enter_smm_save_seg_32(vcpu, &smram->es, &smram->es_sel, VCPU_SREG_ES);
215 	enter_smm_save_seg_32(vcpu, &smram->cs, &smram->cs_sel, VCPU_SREG_CS);
216 	enter_smm_save_seg_32(vcpu, &smram->ss, &smram->ss_sel, VCPU_SREG_SS);
217 
218 	enter_smm_save_seg_32(vcpu, &smram->ds, &smram->ds_sel, VCPU_SREG_DS);
219 	enter_smm_save_seg_32(vcpu, &smram->fs, &smram->fs_sel, VCPU_SREG_FS);
220 	enter_smm_save_seg_32(vcpu, &smram->gs, &smram->gs_sel, VCPU_SREG_GS);
221 
222 	smram->cr4 = kvm_read_cr4(vcpu);
223 	smram->smm_revision = 0x00020000;
224 	smram->smbase = vcpu->arch.smbase;
225 
226 	smram->int_shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
227 }
228 
229 #ifdef CONFIG_X86_64
230 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu,
231 				    struct kvm_smram_state_64 *smram)
232 {
233 	struct desc_ptr dt;
234 	unsigned long val;
235 	int i;
236 
237 	for (i = 0; i < 16; i++)
238 		smram->gprs[15 - i] = kvm_register_read_raw(vcpu, i);
239 
240 	smram->rip    = kvm_rip_read(vcpu);
241 	smram->rflags = kvm_get_rflags(vcpu);
242 
243 
244 	kvm_get_dr(vcpu, 6, &val);
245 	smram->dr6 = val;
246 	kvm_get_dr(vcpu, 7, &val);
247 	smram->dr7 = val;
248 
249 	smram->cr0 = kvm_read_cr0(vcpu);
250 	smram->cr3 = kvm_read_cr3(vcpu);
251 	smram->cr4 = kvm_read_cr4(vcpu);
252 
253 	smram->smbase = vcpu->arch.smbase;
254 	smram->smm_revison = 0x00020064;
255 
256 	smram->efer = vcpu->arch.efer;
257 
258 	enter_smm_save_seg_64(vcpu, &smram->tr, VCPU_SREG_TR);
259 
260 	static_call(kvm_x86_get_idt)(vcpu, &dt);
261 	smram->idtr.limit = dt.size;
262 	smram->idtr.base = dt.address;
263 
264 	enter_smm_save_seg_64(vcpu, &smram->ldtr, VCPU_SREG_LDTR);
265 
266 	static_call(kvm_x86_get_gdt)(vcpu, &dt);
267 	smram->gdtr.limit = dt.size;
268 	smram->gdtr.base = dt.address;
269 
270 	enter_smm_save_seg_64(vcpu, &smram->es, VCPU_SREG_ES);
271 	enter_smm_save_seg_64(vcpu, &smram->cs, VCPU_SREG_CS);
272 	enter_smm_save_seg_64(vcpu, &smram->ss, VCPU_SREG_SS);
273 	enter_smm_save_seg_64(vcpu, &smram->ds, VCPU_SREG_DS);
274 	enter_smm_save_seg_64(vcpu, &smram->fs, VCPU_SREG_FS);
275 	enter_smm_save_seg_64(vcpu, &smram->gs, VCPU_SREG_GS);
276 
277 	smram->int_shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
278 }
279 #endif
280 
281 void enter_smm(struct kvm_vcpu *vcpu)
282 {
283 	struct kvm_segment cs, ds;
284 	struct desc_ptr dt;
285 	unsigned long cr0;
286 	union kvm_smram smram;
287 
288 	check_smram_offsets();
289 
290 	memset(smram.bytes, 0, sizeof(smram.bytes));
291 
292 #ifdef CONFIG_X86_64
293 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
294 		enter_smm_save_state_64(vcpu, &smram.smram64);
295 	else
296 #endif
297 		enter_smm_save_state_32(vcpu, &smram.smram32);
298 
299 	/*
300 	 * Give enter_smm() a chance to make ISA-specific changes to the vCPU
301 	 * state (e.g. leave guest mode) after we've saved the state into the
302 	 * SMM state-save area.
303 	 *
304 	 * Kill the VM in the unlikely case of failure, because the VM
305 	 * can be in undefined state in this case.
306 	 */
307 	if (static_call(kvm_x86_enter_smm)(vcpu, &smram))
308 		goto error;
309 
310 	kvm_smm_changed(vcpu, true);
311 
312 	if (kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, &smram, sizeof(smram)))
313 		goto error;
314 
315 	if (static_call(kvm_x86_get_nmi_mask)(vcpu))
316 		vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
317 	else
318 		static_call(kvm_x86_set_nmi_mask)(vcpu, true);
319 
320 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
321 	kvm_rip_write(vcpu, 0x8000);
322 
323 	static_call(kvm_x86_set_interrupt_shadow)(vcpu, 0);
324 
325 	cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
326 	static_call(kvm_x86_set_cr0)(vcpu, cr0);
327 
328 	static_call(kvm_x86_set_cr4)(vcpu, 0);
329 
330 	/* Undocumented: IDT limit is set to zero on entry to SMM.  */
331 	dt.address = dt.size = 0;
332 	static_call(kvm_x86_set_idt)(vcpu, &dt);
333 
334 	if (WARN_ON_ONCE(kvm_set_dr(vcpu, 7, DR7_FIXED_1)))
335 		goto error;
336 
337 	cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
338 	cs.base = vcpu->arch.smbase;
339 
340 	ds.selector = 0;
341 	ds.base = 0;
342 
343 	cs.limit    = ds.limit = 0xffffffff;
344 	cs.type     = ds.type = 0x3;
345 	cs.dpl      = ds.dpl = 0;
346 	cs.db       = ds.db = 0;
347 	cs.s        = ds.s = 1;
348 	cs.l        = ds.l = 0;
349 	cs.g        = ds.g = 1;
350 	cs.avl      = ds.avl = 0;
351 	cs.present  = ds.present = 1;
352 	cs.unusable = ds.unusable = 0;
353 	cs.padding  = ds.padding = 0;
354 
355 	kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
356 	kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
357 	kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
358 	kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
359 	kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
360 	kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
361 
362 #ifdef CONFIG_X86_64
363 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
364 		if (static_call(kvm_x86_set_efer)(vcpu, 0))
365 			goto error;
366 #endif
367 
368 	kvm_update_cpuid_runtime(vcpu);
369 	kvm_mmu_reset_context(vcpu);
370 	return;
371 error:
372 	kvm_vm_dead(vcpu->kvm);
373 }
374 
375 static void rsm_set_desc_flags(struct kvm_segment *desc, u32 flags)
376 {
377 	desc->g    = (flags >> 23) & 1;
378 	desc->db   = (flags >> 22) & 1;
379 	desc->l    = (flags >> 21) & 1;
380 	desc->avl  = (flags >> 20) & 1;
381 	desc->present = (flags >> 15) & 1;
382 	desc->dpl  = (flags >> 13) & 3;
383 	desc->s    = (flags >> 12) & 1;
384 	desc->type = (flags >>  8) & 15;
385 
386 	desc->unusable = !desc->present;
387 	desc->padding = 0;
388 }
389 
390 static int rsm_load_seg_32(struct kvm_vcpu *vcpu,
391 			   const struct kvm_smm_seg_state_32 *state,
392 			   u16 selector, int n)
393 {
394 	struct kvm_segment desc;
395 
396 	desc.selector =           selector;
397 	desc.base =               state->base;
398 	desc.limit =              state->limit;
399 	rsm_set_desc_flags(&desc, state->flags);
400 	kvm_set_segment(vcpu, &desc, n);
401 	return X86EMUL_CONTINUE;
402 }
403 
404 #ifdef CONFIG_X86_64
405 
406 static int rsm_load_seg_64(struct kvm_vcpu *vcpu,
407 			   const struct kvm_smm_seg_state_64 *state,
408 			   int n)
409 {
410 	struct kvm_segment desc;
411 
412 	desc.selector =           state->selector;
413 	rsm_set_desc_flags(&desc, state->attributes << 8);
414 	desc.limit =              state->limit;
415 	desc.base =               state->base;
416 	kvm_set_segment(vcpu, &desc, n);
417 	return X86EMUL_CONTINUE;
418 }
419 #endif
420 
421 static int rsm_enter_protected_mode(struct kvm_vcpu *vcpu,
422 				    u64 cr0, u64 cr3, u64 cr4)
423 {
424 	int bad;
425 	u64 pcid;
426 
427 	/* In order to later set CR4.PCIDE, CR3[11:0] must be zero.  */
428 	pcid = 0;
429 	if (cr4 & X86_CR4_PCIDE) {
430 		pcid = cr3 & 0xfff;
431 		cr3 &= ~0xfff;
432 	}
433 
434 	bad = kvm_set_cr3(vcpu, cr3);
435 	if (bad)
436 		return X86EMUL_UNHANDLEABLE;
437 
438 	/*
439 	 * First enable PAE, long mode needs it before CR0.PG = 1 is set.
440 	 * Then enable protected mode.	However, PCID cannot be enabled
441 	 * if EFER.LMA=0, so set it separately.
442 	 */
443 	bad = kvm_set_cr4(vcpu, cr4 & ~X86_CR4_PCIDE);
444 	if (bad)
445 		return X86EMUL_UNHANDLEABLE;
446 
447 	bad = kvm_set_cr0(vcpu, cr0);
448 	if (bad)
449 		return X86EMUL_UNHANDLEABLE;
450 
451 	if (cr4 & X86_CR4_PCIDE) {
452 		bad = kvm_set_cr4(vcpu, cr4);
453 		if (bad)
454 			return X86EMUL_UNHANDLEABLE;
455 		if (pcid) {
456 			bad = kvm_set_cr3(vcpu, cr3 | pcid);
457 			if (bad)
458 				return X86EMUL_UNHANDLEABLE;
459 		}
460 
461 	}
462 
463 	return X86EMUL_CONTINUE;
464 }
465 
466 static int rsm_load_state_32(struct x86_emulate_ctxt *ctxt,
467 			     const struct kvm_smram_state_32 *smstate)
468 {
469 	struct kvm_vcpu *vcpu = ctxt->vcpu;
470 	struct desc_ptr dt;
471 	int i, r;
472 
473 	ctxt->eflags =  smstate->eflags | X86_EFLAGS_FIXED;
474 	ctxt->_eip =  smstate->eip;
475 
476 	for (i = 0; i < 8; i++)
477 		*reg_write(ctxt, i) = smstate->gprs[i];
478 
479 	if (kvm_set_dr(vcpu, 6, smstate->dr6))
480 		return X86EMUL_UNHANDLEABLE;
481 	if (kvm_set_dr(vcpu, 7, smstate->dr7))
482 		return X86EMUL_UNHANDLEABLE;
483 
484 	rsm_load_seg_32(vcpu, &smstate->tr, smstate->tr_sel, VCPU_SREG_TR);
485 	rsm_load_seg_32(vcpu, &smstate->ldtr, smstate->ldtr_sel, VCPU_SREG_LDTR);
486 
487 	dt.address =               smstate->gdtr.base;
488 	dt.size =                  smstate->gdtr.limit;
489 	static_call(kvm_x86_set_gdt)(vcpu, &dt);
490 
491 	dt.address =               smstate->idtr.base;
492 	dt.size =                  smstate->idtr.limit;
493 	static_call(kvm_x86_set_idt)(vcpu, &dt);
494 
495 	rsm_load_seg_32(vcpu, &smstate->es, smstate->es_sel, VCPU_SREG_ES);
496 	rsm_load_seg_32(vcpu, &smstate->cs, smstate->cs_sel, VCPU_SREG_CS);
497 	rsm_load_seg_32(vcpu, &smstate->ss, smstate->ss_sel, VCPU_SREG_SS);
498 
499 	rsm_load_seg_32(vcpu, &smstate->ds, smstate->ds_sel, VCPU_SREG_DS);
500 	rsm_load_seg_32(vcpu, &smstate->fs, smstate->fs_sel, VCPU_SREG_FS);
501 	rsm_load_seg_32(vcpu, &smstate->gs, smstate->gs_sel, VCPU_SREG_GS);
502 
503 	vcpu->arch.smbase = smstate->smbase;
504 
505 	r = rsm_enter_protected_mode(vcpu, smstate->cr0,
506 					smstate->cr3, smstate->cr4);
507 
508 	if (r != X86EMUL_CONTINUE)
509 		return r;
510 
511 	static_call(kvm_x86_set_interrupt_shadow)(vcpu, 0);
512 	ctxt->interruptibility = (u8)smstate->int_shadow;
513 
514 	return r;
515 }
516 
517 #ifdef CONFIG_X86_64
518 static int rsm_load_state_64(struct x86_emulate_ctxt *ctxt,
519 			     const struct kvm_smram_state_64 *smstate)
520 {
521 	struct kvm_vcpu *vcpu = ctxt->vcpu;
522 	struct desc_ptr dt;
523 	int i, r;
524 
525 	for (i = 0; i < 16; i++)
526 		*reg_write(ctxt, i) = smstate->gprs[15 - i];
527 
528 	ctxt->_eip   = smstate->rip;
529 	ctxt->eflags = smstate->rflags | X86_EFLAGS_FIXED;
530 
531 	if (kvm_set_dr(vcpu, 6, smstate->dr6))
532 		return X86EMUL_UNHANDLEABLE;
533 	if (kvm_set_dr(vcpu, 7, smstate->dr7))
534 		return X86EMUL_UNHANDLEABLE;
535 
536 	vcpu->arch.smbase =         smstate->smbase;
537 
538 	if (kvm_set_msr(vcpu, MSR_EFER, smstate->efer & ~EFER_LMA))
539 		return X86EMUL_UNHANDLEABLE;
540 
541 	rsm_load_seg_64(vcpu, &smstate->tr, VCPU_SREG_TR);
542 
543 	dt.size =                   smstate->idtr.limit;
544 	dt.address =                smstate->idtr.base;
545 	static_call(kvm_x86_set_idt)(vcpu, &dt);
546 
547 	rsm_load_seg_64(vcpu, &smstate->ldtr, VCPU_SREG_LDTR);
548 
549 	dt.size =                   smstate->gdtr.limit;
550 	dt.address =                smstate->gdtr.base;
551 	static_call(kvm_x86_set_gdt)(vcpu, &dt);
552 
553 	r = rsm_enter_protected_mode(vcpu, smstate->cr0, smstate->cr3, smstate->cr4);
554 	if (r != X86EMUL_CONTINUE)
555 		return r;
556 
557 	rsm_load_seg_64(vcpu, &smstate->es, VCPU_SREG_ES);
558 	rsm_load_seg_64(vcpu, &smstate->cs, VCPU_SREG_CS);
559 	rsm_load_seg_64(vcpu, &smstate->ss, VCPU_SREG_SS);
560 	rsm_load_seg_64(vcpu, &smstate->ds, VCPU_SREG_DS);
561 	rsm_load_seg_64(vcpu, &smstate->fs, VCPU_SREG_FS);
562 	rsm_load_seg_64(vcpu, &smstate->gs, VCPU_SREG_GS);
563 
564 	static_call(kvm_x86_set_interrupt_shadow)(vcpu, 0);
565 	ctxt->interruptibility = (u8)smstate->int_shadow;
566 
567 	return X86EMUL_CONTINUE;
568 }
569 #endif
570 
571 int emulator_leave_smm(struct x86_emulate_ctxt *ctxt)
572 {
573 	struct kvm_vcpu *vcpu = ctxt->vcpu;
574 	unsigned long cr0;
575 	union kvm_smram smram;
576 	u64 smbase;
577 	int ret;
578 
579 	smbase = vcpu->arch.smbase;
580 
581 	ret = kvm_vcpu_read_guest(vcpu, smbase + 0xfe00, smram.bytes, sizeof(smram));
582 	if (ret < 0)
583 		return X86EMUL_UNHANDLEABLE;
584 
585 	if ((vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK) == 0)
586 		static_call(kvm_x86_set_nmi_mask)(vcpu, false);
587 
588 	kvm_smm_changed(vcpu, false);
589 
590 	/*
591 	 * Get back to real mode, to prepare a safe state in which to load
592 	 * CR0/CR3/CR4/EFER.  It's all a bit more complicated if the vCPU
593 	 * supports long mode.
594 	 */
595 #ifdef CONFIG_X86_64
596 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM)) {
597 		struct kvm_segment cs_desc;
598 		unsigned long cr4;
599 
600 		/* Zero CR4.PCIDE before CR0.PG.  */
601 		cr4 = kvm_read_cr4(vcpu);
602 		if (cr4 & X86_CR4_PCIDE)
603 			kvm_set_cr4(vcpu, cr4 & ~X86_CR4_PCIDE);
604 
605 		/* A 32-bit code segment is required to clear EFER.LMA.  */
606 		memset(&cs_desc, 0, sizeof(cs_desc));
607 		cs_desc.type = 0xb;
608 		cs_desc.s = cs_desc.g = cs_desc.present = 1;
609 		kvm_set_segment(vcpu, &cs_desc, VCPU_SREG_CS);
610 	}
611 #endif
612 
613 	/* For the 64-bit case, this will clear EFER.LMA.  */
614 	cr0 = kvm_read_cr0(vcpu);
615 	if (cr0 & X86_CR0_PE)
616 		kvm_set_cr0(vcpu, cr0 & ~(X86_CR0_PG | X86_CR0_PE));
617 
618 #ifdef CONFIG_X86_64
619 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM)) {
620 		unsigned long cr4, efer;
621 
622 		/* Clear CR4.PAE before clearing EFER.LME. */
623 		cr4 = kvm_read_cr4(vcpu);
624 		if (cr4 & X86_CR4_PAE)
625 			kvm_set_cr4(vcpu, cr4 & ~X86_CR4_PAE);
626 
627 		/* And finally go back to 32-bit mode.  */
628 		efer = 0;
629 		kvm_set_msr(vcpu, MSR_EFER, efer);
630 	}
631 #endif
632 
633 	/*
634 	 * Give leave_smm() a chance to make ISA-specific changes to the vCPU
635 	 * state (e.g. enter guest mode) before loading state from the SMM
636 	 * state-save area.
637 	 */
638 	if (static_call(kvm_x86_leave_smm)(vcpu, &smram))
639 		return X86EMUL_UNHANDLEABLE;
640 
641 #ifdef CONFIG_X86_64
642 	if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
643 		return rsm_load_state_64(ctxt, &smram.smram64);
644 	else
645 #endif
646 		return rsm_load_state_32(ctxt, &smram.smram32);
647 }
648