xref: /openbmc/linux/arch/x86/kvm/svm/svm.c (revision 41415b8a)
1 #define pr_fmt(fmt) "SVM: " fmt
2 
3 #include <linux/kvm_host.h>
4 
5 #include "irq.h"
6 #include "mmu.h"
7 #include "kvm_cache_regs.h"
8 #include "x86.h"
9 #include "cpuid.h"
10 #include "pmu.h"
11 
12 #include <linux/module.h>
13 #include <linux/mod_devicetable.h>
14 #include <linux/kernel.h>
15 #include <linux/vmalloc.h>
16 #include <linux/highmem.h>
17 #include <linux/amd-iommu.h>
18 #include <linux/sched.h>
19 #include <linux/trace_events.h>
20 #include <linux/slab.h>
21 #include <linux/hashtable.h>
22 #include <linux/objtool.h>
23 #include <linux/psp-sev.h>
24 #include <linux/file.h>
25 #include <linux/pagemap.h>
26 #include <linux/swap.h>
27 #include <linux/rwsem.h>
28 #include <linux/cc_platform.h>
29 
30 #include <asm/apic.h>
31 #include <asm/perf_event.h>
32 #include <asm/tlbflush.h>
33 #include <asm/desc.h>
34 #include <asm/debugreg.h>
35 #include <asm/kvm_para.h>
36 #include <asm/irq_remapping.h>
37 #include <asm/spec-ctrl.h>
38 #include <asm/cpu_device_id.h>
39 #include <asm/traps.h>
40 #include <asm/fpu/api.h>
41 
42 #include <asm/virtext.h>
43 #include "trace.h"
44 
45 #include "svm.h"
46 #include "svm_ops.h"
47 
48 #include "kvm_onhyperv.h"
49 #include "svm_onhyperv.h"
50 
51 MODULE_AUTHOR("Qumranet");
52 MODULE_LICENSE("GPL");
53 
54 #ifdef MODULE
55 static const struct x86_cpu_id svm_cpu_id[] = {
56 	X86_MATCH_FEATURE(X86_FEATURE_SVM, NULL),
57 	{}
58 };
59 MODULE_DEVICE_TABLE(x86cpu, svm_cpu_id);
60 #endif
61 
62 #define SEG_TYPE_LDT 2
63 #define SEG_TYPE_BUSY_TSS16 3
64 
65 #define SVM_FEATURE_LBRV           (1 <<  1)
66 #define SVM_FEATURE_SVML           (1 <<  2)
67 #define SVM_FEATURE_TSC_RATE       (1 <<  4)
68 #define SVM_FEATURE_VMCB_CLEAN     (1 <<  5)
69 #define SVM_FEATURE_FLUSH_ASID     (1 <<  6)
70 #define SVM_FEATURE_DECODE_ASSIST  (1 <<  7)
71 #define SVM_FEATURE_PAUSE_FILTER   (1 << 10)
72 
73 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
74 
75 #define TSC_RATIO_RSVD          0xffffff0000000000ULL
76 #define TSC_RATIO_MIN		0x0000000000000001ULL
77 #define TSC_RATIO_MAX		0x000000ffffffffffULL
78 
79 static bool erratum_383_found __read_mostly;
80 
81 u32 msrpm_offsets[MSRPM_OFFSETS] __read_mostly;
82 
83 /*
84  * Set osvw_len to higher value when updated Revision Guides
85  * are published and we know what the new status bits are
86  */
87 static uint64_t osvw_len = 4, osvw_status;
88 
89 static DEFINE_PER_CPU(u64, current_tsc_ratio);
90 #define TSC_RATIO_DEFAULT	0x0100000000ULL
91 
92 static const struct svm_direct_access_msrs {
93 	u32 index;   /* Index of the MSR */
94 	bool always; /* True if intercept is initially cleared */
95 } direct_access_msrs[MAX_DIRECT_ACCESS_MSRS] = {
96 	{ .index = MSR_STAR,				.always = true  },
97 	{ .index = MSR_IA32_SYSENTER_CS,		.always = true  },
98 	{ .index = MSR_IA32_SYSENTER_EIP,		.always = false },
99 	{ .index = MSR_IA32_SYSENTER_ESP,		.always = false },
100 #ifdef CONFIG_X86_64
101 	{ .index = MSR_GS_BASE,				.always = true  },
102 	{ .index = MSR_FS_BASE,				.always = true  },
103 	{ .index = MSR_KERNEL_GS_BASE,			.always = true  },
104 	{ .index = MSR_LSTAR,				.always = true  },
105 	{ .index = MSR_CSTAR,				.always = true  },
106 	{ .index = MSR_SYSCALL_MASK,			.always = true  },
107 #endif
108 	{ .index = MSR_IA32_SPEC_CTRL,			.always = false },
109 	{ .index = MSR_IA32_PRED_CMD,			.always = false },
110 	{ .index = MSR_IA32_LASTBRANCHFROMIP,		.always = false },
111 	{ .index = MSR_IA32_LASTBRANCHTOIP,		.always = false },
112 	{ .index = MSR_IA32_LASTINTFROMIP,		.always = false },
113 	{ .index = MSR_IA32_LASTINTTOIP,		.always = false },
114 	{ .index = MSR_EFER,				.always = false },
115 	{ .index = MSR_IA32_CR_PAT,			.always = false },
116 	{ .index = MSR_AMD64_SEV_ES_GHCB,		.always = true  },
117 	{ .index = MSR_INVALID,				.always = false },
118 };
119 
120 /*
121  * These 2 parameters are used to config the controls for Pause-Loop Exiting:
122  * pause_filter_count: On processors that support Pause filtering(indicated
123  *	by CPUID Fn8000_000A_EDX), the VMCB provides a 16 bit pause filter
124  *	count value. On VMRUN this value is loaded into an internal counter.
125  *	Each time a pause instruction is executed, this counter is decremented
126  *	until it reaches zero at which time a #VMEXIT is generated if pause
127  *	intercept is enabled. Refer to  AMD APM Vol 2 Section 15.14.4 Pause
128  *	Intercept Filtering for more details.
129  *	This also indicate if ple logic enabled.
130  *
131  * pause_filter_thresh: In addition, some processor families support advanced
132  *	pause filtering (indicated by CPUID Fn8000_000A_EDX) upper bound on
133  *	the amount of time a guest is allowed to execute in a pause loop.
134  *	In this mode, a 16-bit pause filter threshold field is added in the
135  *	VMCB. The threshold value is a cycle count that is used to reset the
136  *	pause counter. As with simple pause filtering, VMRUN loads the pause
137  *	count value from VMCB into an internal counter. Then, on each pause
138  *	instruction the hardware checks the elapsed number of cycles since
139  *	the most recent pause instruction against the pause filter threshold.
140  *	If the elapsed cycle count is greater than the pause filter threshold,
141  *	then the internal pause count is reloaded from the VMCB and execution
142  *	continues. If the elapsed cycle count is less than the pause filter
143  *	threshold, then the internal pause count is decremented. If the count
144  *	value is less than zero and PAUSE intercept is enabled, a #VMEXIT is
145  *	triggered. If advanced pause filtering is supported and pause filter
146  *	threshold field is set to zero, the filter will operate in the simpler,
147  *	count only mode.
148  */
149 
150 static unsigned short pause_filter_thresh = KVM_DEFAULT_PLE_GAP;
151 module_param(pause_filter_thresh, ushort, 0444);
152 
153 static unsigned short pause_filter_count = KVM_SVM_DEFAULT_PLE_WINDOW;
154 module_param(pause_filter_count, ushort, 0444);
155 
156 /* Default doubles per-vcpu window every exit. */
157 static unsigned short pause_filter_count_grow = KVM_DEFAULT_PLE_WINDOW_GROW;
158 module_param(pause_filter_count_grow, ushort, 0444);
159 
160 /* Default resets per-vcpu window every exit to pause_filter_count. */
161 static unsigned short pause_filter_count_shrink = KVM_DEFAULT_PLE_WINDOW_SHRINK;
162 module_param(pause_filter_count_shrink, ushort, 0444);
163 
164 /* Default is to compute the maximum so we can never overflow. */
165 static unsigned short pause_filter_count_max = KVM_SVM_DEFAULT_PLE_WINDOW_MAX;
166 module_param(pause_filter_count_max, ushort, 0444);
167 
168 /*
169  * Use nested page tables by default.  Note, NPT may get forced off by
170  * svm_hardware_setup() if it's unsupported by hardware or the host kernel.
171  */
172 bool npt_enabled = true;
173 module_param_named(npt, npt_enabled, bool, 0444);
174 
175 /* allow nested virtualization in KVM/SVM */
176 static int nested = true;
177 module_param(nested, int, S_IRUGO);
178 
179 /* enable/disable Next RIP Save */
180 static int nrips = true;
181 module_param(nrips, int, 0444);
182 
183 /* enable/disable Virtual VMLOAD VMSAVE */
184 static int vls = true;
185 module_param(vls, int, 0444);
186 
187 /* enable/disable Virtual GIF */
188 static int vgif = true;
189 module_param(vgif, int, 0444);
190 
191 /* enable/disable LBR virtualization */
192 static int lbrv = true;
193 module_param(lbrv, int, 0444);
194 
195 static int tsc_scaling = true;
196 module_param(tsc_scaling, int, 0444);
197 
198 /*
199  * enable / disable AVIC.  Because the defaults differ for APICv
200  * support between VMX and SVM we cannot use module_param_named.
201  */
202 static bool avic;
203 module_param(avic, bool, 0444);
204 
205 bool __read_mostly dump_invalid_vmcb;
206 module_param(dump_invalid_vmcb, bool, 0644);
207 
208 
209 bool intercept_smi = true;
210 module_param(intercept_smi, bool, 0444);
211 
212 
213 static bool svm_gp_erratum_intercept = true;
214 
215 static u8 rsm_ins_bytes[] = "\x0f\xaa";
216 
217 static unsigned long iopm_base;
218 
219 struct kvm_ldttss_desc {
220 	u16 limit0;
221 	u16 base0;
222 	unsigned base1:8, type:5, dpl:2, p:1;
223 	unsigned limit1:4, zero0:3, g:1, base2:8;
224 	u32 base3;
225 	u32 zero1;
226 } __attribute__((packed));
227 
228 DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
229 
230 /*
231  * Only MSR_TSC_AUX is switched via the user return hook.  EFER is switched via
232  * the VMCB, and the SYSCALL/SYSENTER MSRs are handled by VMLOAD/VMSAVE.
233  *
234  * RDTSCP and RDPID are not used in the kernel, specifically to allow KVM to
235  * defer the restoration of TSC_AUX until the CPU returns to userspace.
236  */
237 static int tsc_aux_uret_slot __read_mostly = -1;
238 
239 static const u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
240 
241 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
242 #define MSRS_RANGE_SIZE 2048
243 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
244 
245 u32 svm_msrpm_offset(u32 msr)
246 {
247 	u32 offset;
248 	int i;
249 
250 	for (i = 0; i < NUM_MSR_MAPS; i++) {
251 		if (msr < msrpm_ranges[i] ||
252 		    msr >= msrpm_ranges[i] + MSRS_IN_RANGE)
253 			continue;
254 
255 		offset  = (msr - msrpm_ranges[i]) / 4; /* 4 msrs per u8 */
256 		offset += (i * MSRS_RANGE_SIZE);       /* add range offset */
257 
258 		/* Now we have the u8 offset - but need the u32 offset */
259 		return offset / 4;
260 	}
261 
262 	/* MSR not in any range */
263 	return MSR_INVALID;
264 }
265 
266 #define MAX_INST_SIZE 15
267 
268 static int get_npt_level(void)
269 {
270 #ifdef CONFIG_X86_64
271 	return pgtable_l5_enabled() ? PT64_ROOT_5LEVEL : PT64_ROOT_4LEVEL;
272 #else
273 	return PT32E_ROOT_LEVEL;
274 #endif
275 }
276 
277 int svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
278 {
279 	struct vcpu_svm *svm = to_svm(vcpu);
280 	u64 old_efer = vcpu->arch.efer;
281 	vcpu->arch.efer = efer;
282 
283 	if (!npt_enabled) {
284 		/* Shadow paging assumes NX to be available.  */
285 		efer |= EFER_NX;
286 
287 		if (!(efer & EFER_LMA))
288 			efer &= ~EFER_LME;
289 	}
290 
291 	if ((old_efer & EFER_SVME) != (efer & EFER_SVME)) {
292 		if (!(efer & EFER_SVME)) {
293 			svm_leave_nested(vcpu);
294 			svm_set_gif(svm, true);
295 			/* #GP intercept is still needed for vmware backdoor */
296 			if (!enable_vmware_backdoor)
297 				clr_exception_intercept(svm, GP_VECTOR);
298 
299 			/*
300 			 * Free the nested guest state, unless we are in SMM.
301 			 * In this case we will return to the nested guest
302 			 * as soon as we leave SMM.
303 			 */
304 			if (!is_smm(vcpu))
305 				svm_free_nested(svm);
306 
307 		} else {
308 			int ret = svm_allocate_nested(svm);
309 
310 			if (ret) {
311 				vcpu->arch.efer = old_efer;
312 				return ret;
313 			}
314 
315 			/*
316 			 * Never intercept #GP for SEV guests, KVM can't
317 			 * decrypt guest memory to workaround the erratum.
318 			 */
319 			if (svm_gp_erratum_intercept && !sev_guest(vcpu->kvm))
320 				set_exception_intercept(svm, GP_VECTOR);
321 		}
322 	}
323 
324 	svm->vmcb->save.efer = efer | EFER_SVME;
325 	vmcb_mark_dirty(svm->vmcb, VMCB_CR);
326 	return 0;
327 }
328 
329 static int is_external_interrupt(u32 info)
330 {
331 	info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
332 	return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
333 }
334 
335 static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu)
336 {
337 	struct vcpu_svm *svm = to_svm(vcpu);
338 	u32 ret = 0;
339 
340 	if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)
341 		ret = KVM_X86_SHADOW_INT_STI | KVM_X86_SHADOW_INT_MOV_SS;
342 	return ret;
343 }
344 
345 static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
346 {
347 	struct vcpu_svm *svm = to_svm(vcpu);
348 
349 	if (mask == 0)
350 		svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
351 	else
352 		svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK;
353 
354 }
355 
356 static int skip_emulated_instruction(struct kvm_vcpu *vcpu)
357 {
358 	struct vcpu_svm *svm = to_svm(vcpu);
359 
360 	/*
361 	 * SEV-ES does not expose the next RIP. The RIP update is controlled by
362 	 * the type of exit and the #VC handler in the guest.
363 	 */
364 	if (sev_es_guest(vcpu->kvm))
365 		goto done;
366 
367 	if (nrips && svm->vmcb->control.next_rip != 0) {
368 		WARN_ON_ONCE(!static_cpu_has(X86_FEATURE_NRIPS));
369 		svm->next_rip = svm->vmcb->control.next_rip;
370 	}
371 
372 	if (!svm->next_rip) {
373 		if (!kvm_emulate_instruction(vcpu, EMULTYPE_SKIP))
374 			return 0;
375 	} else {
376 		kvm_rip_write(vcpu, svm->next_rip);
377 	}
378 
379 done:
380 	svm_set_interrupt_shadow(vcpu, 0);
381 
382 	return 1;
383 }
384 
385 static void svm_queue_exception(struct kvm_vcpu *vcpu)
386 {
387 	struct vcpu_svm *svm = to_svm(vcpu);
388 	unsigned nr = vcpu->arch.exception.nr;
389 	bool has_error_code = vcpu->arch.exception.has_error_code;
390 	u32 error_code = vcpu->arch.exception.error_code;
391 
392 	kvm_deliver_exception_payload(vcpu);
393 
394 	if (nr == BP_VECTOR && !nrips) {
395 		unsigned long rip, old_rip = kvm_rip_read(vcpu);
396 
397 		/*
398 		 * For guest debugging where we have to reinject #BP if some
399 		 * INT3 is guest-owned:
400 		 * Emulate nRIP by moving RIP forward. Will fail if injection
401 		 * raises a fault that is not intercepted. Still better than
402 		 * failing in all cases.
403 		 */
404 		(void)skip_emulated_instruction(vcpu);
405 		rip = kvm_rip_read(vcpu);
406 		svm->int3_rip = rip + svm->vmcb->save.cs.base;
407 		svm->int3_injected = rip - old_rip;
408 	}
409 
410 	svm->vmcb->control.event_inj = nr
411 		| SVM_EVTINJ_VALID
412 		| (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
413 		| SVM_EVTINJ_TYPE_EXEPT;
414 	svm->vmcb->control.event_inj_err = error_code;
415 }
416 
417 static void svm_init_erratum_383(void)
418 {
419 	u32 low, high;
420 	int err;
421 	u64 val;
422 
423 	if (!static_cpu_has_bug(X86_BUG_AMD_TLB_MMATCH))
424 		return;
425 
426 	/* Use _safe variants to not break nested virtualization */
427 	val = native_read_msr_safe(MSR_AMD64_DC_CFG, &err);
428 	if (err)
429 		return;
430 
431 	val |= (1ULL << 47);
432 
433 	low  = lower_32_bits(val);
434 	high = upper_32_bits(val);
435 
436 	native_write_msr_safe(MSR_AMD64_DC_CFG, low, high);
437 
438 	erratum_383_found = true;
439 }
440 
441 static void svm_init_osvw(struct kvm_vcpu *vcpu)
442 {
443 	/*
444 	 * Guests should see errata 400 and 415 as fixed (assuming that
445 	 * HLT and IO instructions are intercepted).
446 	 */
447 	vcpu->arch.osvw.length = (osvw_len >= 3) ? (osvw_len) : 3;
448 	vcpu->arch.osvw.status = osvw_status & ~(6ULL);
449 
450 	/*
451 	 * By increasing VCPU's osvw.length to 3 we are telling the guest that
452 	 * all osvw.status bits inside that length, including bit 0 (which is
453 	 * reserved for erratum 298), are valid. However, if host processor's
454 	 * osvw_len is 0 then osvw_status[0] carries no information. We need to
455 	 * be conservative here and therefore we tell the guest that erratum 298
456 	 * is present (because we really don't know).
457 	 */
458 	if (osvw_len == 0 && boot_cpu_data.x86 == 0x10)
459 		vcpu->arch.osvw.status |= 1;
460 }
461 
462 static int has_svm(void)
463 {
464 	const char *msg;
465 
466 	if (!cpu_has_svm(&msg)) {
467 		printk(KERN_INFO "has_svm: %s\n", msg);
468 		return 0;
469 	}
470 
471 	if (cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT)) {
472 		pr_info("KVM is unsupported when running as an SEV guest\n");
473 		return 0;
474 	}
475 
476 	return 1;
477 }
478 
479 static void svm_hardware_disable(void)
480 {
481 	/* Make sure we clean up behind us */
482 	if (tsc_scaling)
483 		wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
484 
485 	cpu_svm_disable();
486 
487 	amd_pmu_disable_virt();
488 }
489 
490 static int svm_hardware_enable(void)
491 {
492 
493 	struct svm_cpu_data *sd;
494 	uint64_t efer;
495 	struct desc_struct *gdt;
496 	int me = raw_smp_processor_id();
497 
498 	rdmsrl(MSR_EFER, efer);
499 	if (efer & EFER_SVME)
500 		return -EBUSY;
501 
502 	if (!has_svm()) {
503 		pr_err("%s: err EOPNOTSUPP on %d\n", __func__, me);
504 		return -EINVAL;
505 	}
506 	sd = per_cpu(svm_data, me);
507 	if (!sd) {
508 		pr_err("%s: svm_data is NULL on %d\n", __func__, me);
509 		return -EINVAL;
510 	}
511 
512 	sd->asid_generation = 1;
513 	sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
514 	sd->next_asid = sd->max_asid + 1;
515 	sd->min_asid = max_sev_asid + 1;
516 
517 	gdt = get_current_gdt_rw();
518 	sd->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
519 
520 	wrmsrl(MSR_EFER, efer | EFER_SVME);
521 
522 	wrmsrl(MSR_VM_HSAVE_PA, __sme_page_pa(sd->save_area));
523 
524 	if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
525 		/*
526 		 * Set the default value, even if we don't use TSC scaling
527 		 * to avoid having stale value in the msr
528 		 */
529 		wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
530 		__this_cpu_write(current_tsc_ratio, TSC_RATIO_DEFAULT);
531 	}
532 
533 
534 	/*
535 	 * Get OSVW bits.
536 	 *
537 	 * Note that it is possible to have a system with mixed processor
538 	 * revisions and therefore different OSVW bits. If bits are not the same
539 	 * on different processors then choose the worst case (i.e. if erratum
540 	 * is present on one processor and not on another then assume that the
541 	 * erratum is present everywhere).
542 	 */
543 	if (cpu_has(&boot_cpu_data, X86_FEATURE_OSVW)) {
544 		uint64_t len, status = 0;
545 		int err;
546 
547 		len = native_read_msr_safe(MSR_AMD64_OSVW_ID_LENGTH, &err);
548 		if (!err)
549 			status = native_read_msr_safe(MSR_AMD64_OSVW_STATUS,
550 						      &err);
551 
552 		if (err)
553 			osvw_status = osvw_len = 0;
554 		else {
555 			if (len < osvw_len)
556 				osvw_len = len;
557 			osvw_status |= status;
558 			osvw_status &= (1ULL << osvw_len) - 1;
559 		}
560 	} else
561 		osvw_status = osvw_len = 0;
562 
563 	svm_init_erratum_383();
564 
565 	amd_pmu_enable_virt();
566 
567 	return 0;
568 }
569 
570 static void svm_cpu_uninit(int cpu)
571 {
572 	struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
573 
574 	if (!sd)
575 		return;
576 
577 	per_cpu(svm_data, cpu) = NULL;
578 	kfree(sd->sev_vmcbs);
579 	__free_page(sd->save_area);
580 	kfree(sd);
581 }
582 
583 static int svm_cpu_init(int cpu)
584 {
585 	struct svm_cpu_data *sd;
586 	int ret = -ENOMEM;
587 
588 	sd = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
589 	if (!sd)
590 		return ret;
591 	sd->cpu = cpu;
592 	sd->save_area = alloc_page(GFP_KERNEL | __GFP_ZERO);
593 	if (!sd->save_area)
594 		goto free_cpu_data;
595 
596 	ret = sev_cpu_init(sd);
597 	if (ret)
598 		goto free_save_area;
599 
600 	per_cpu(svm_data, cpu) = sd;
601 
602 	return 0;
603 
604 free_save_area:
605 	__free_page(sd->save_area);
606 free_cpu_data:
607 	kfree(sd);
608 	return ret;
609 
610 }
611 
612 static int direct_access_msr_slot(u32 msr)
613 {
614 	u32 i;
615 
616 	for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++)
617 		if (direct_access_msrs[i].index == msr)
618 			return i;
619 
620 	return -ENOENT;
621 }
622 
623 static void set_shadow_msr_intercept(struct kvm_vcpu *vcpu, u32 msr, int read,
624 				     int write)
625 {
626 	struct vcpu_svm *svm = to_svm(vcpu);
627 	int slot = direct_access_msr_slot(msr);
628 
629 	if (slot == -ENOENT)
630 		return;
631 
632 	/* Set the shadow bitmaps to the desired intercept states */
633 	if (read)
634 		set_bit(slot, svm->shadow_msr_intercept.read);
635 	else
636 		clear_bit(slot, svm->shadow_msr_intercept.read);
637 
638 	if (write)
639 		set_bit(slot, svm->shadow_msr_intercept.write);
640 	else
641 		clear_bit(slot, svm->shadow_msr_intercept.write);
642 }
643 
644 static bool valid_msr_intercept(u32 index)
645 {
646 	return direct_access_msr_slot(index) != -ENOENT;
647 }
648 
649 static bool msr_write_intercepted(struct kvm_vcpu *vcpu, u32 msr)
650 {
651 	u8 bit_write;
652 	unsigned long tmp;
653 	u32 offset;
654 	u32 *msrpm;
655 
656 	msrpm = is_guest_mode(vcpu) ? to_svm(vcpu)->nested.msrpm:
657 				      to_svm(vcpu)->msrpm;
658 
659 	offset    = svm_msrpm_offset(msr);
660 	bit_write = 2 * (msr & 0x0f) + 1;
661 	tmp       = msrpm[offset];
662 
663 	BUG_ON(offset == MSR_INVALID);
664 
665 	return !!test_bit(bit_write,  &tmp);
666 }
667 
668 static void set_msr_interception_bitmap(struct kvm_vcpu *vcpu, u32 *msrpm,
669 					u32 msr, int read, int write)
670 {
671 	u8 bit_read, bit_write;
672 	unsigned long tmp;
673 	u32 offset;
674 
675 	/*
676 	 * If this warning triggers extend the direct_access_msrs list at the
677 	 * beginning of the file
678 	 */
679 	WARN_ON(!valid_msr_intercept(msr));
680 
681 	/* Enforce non allowed MSRs to trap */
682 	if (read && !kvm_msr_allowed(vcpu, msr, KVM_MSR_FILTER_READ))
683 		read = 0;
684 
685 	if (write && !kvm_msr_allowed(vcpu, msr, KVM_MSR_FILTER_WRITE))
686 		write = 0;
687 
688 	offset    = svm_msrpm_offset(msr);
689 	bit_read  = 2 * (msr & 0x0f);
690 	bit_write = 2 * (msr & 0x0f) + 1;
691 	tmp       = msrpm[offset];
692 
693 	BUG_ON(offset == MSR_INVALID);
694 
695 	read  ? clear_bit(bit_read,  &tmp) : set_bit(bit_read,  &tmp);
696 	write ? clear_bit(bit_write, &tmp) : set_bit(bit_write, &tmp);
697 
698 	msrpm[offset] = tmp;
699 
700 	svm_hv_vmcb_dirty_nested_enlightenments(vcpu);
701 
702 }
703 
704 void set_msr_interception(struct kvm_vcpu *vcpu, u32 *msrpm, u32 msr,
705 			  int read, int write)
706 {
707 	set_shadow_msr_intercept(vcpu, msr, read, write);
708 	set_msr_interception_bitmap(vcpu, msrpm, msr, read, write);
709 }
710 
711 u32 *svm_vcpu_alloc_msrpm(void)
712 {
713 	unsigned int order = get_order(MSRPM_SIZE);
714 	struct page *pages = alloc_pages(GFP_KERNEL_ACCOUNT, order);
715 	u32 *msrpm;
716 
717 	if (!pages)
718 		return NULL;
719 
720 	msrpm = page_address(pages);
721 	memset(msrpm, 0xff, PAGE_SIZE * (1 << order));
722 
723 	return msrpm;
724 }
725 
726 void svm_vcpu_init_msrpm(struct kvm_vcpu *vcpu, u32 *msrpm)
727 {
728 	int i;
729 
730 	for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
731 		if (!direct_access_msrs[i].always)
732 			continue;
733 		set_msr_interception(vcpu, msrpm, direct_access_msrs[i].index, 1, 1);
734 	}
735 }
736 
737 
738 void svm_vcpu_free_msrpm(u32 *msrpm)
739 {
740 	__free_pages(virt_to_page(msrpm), get_order(MSRPM_SIZE));
741 }
742 
743 static void svm_msr_filter_changed(struct kvm_vcpu *vcpu)
744 {
745 	struct vcpu_svm *svm = to_svm(vcpu);
746 	u32 i;
747 
748 	/*
749 	 * Set intercept permissions for all direct access MSRs again. They
750 	 * will automatically get filtered through the MSR filter, so we are
751 	 * back in sync after this.
752 	 */
753 	for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
754 		u32 msr = direct_access_msrs[i].index;
755 		u32 read = test_bit(i, svm->shadow_msr_intercept.read);
756 		u32 write = test_bit(i, svm->shadow_msr_intercept.write);
757 
758 		set_msr_interception_bitmap(vcpu, svm->msrpm, msr, read, write);
759 	}
760 }
761 
762 static void add_msr_offset(u32 offset)
763 {
764 	int i;
765 
766 	for (i = 0; i < MSRPM_OFFSETS; ++i) {
767 
768 		/* Offset already in list? */
769 		if (msrpm_offsets[i] == offset)
770 			return;
771 
772 		/* Slot used by another offset? */
773 		if (msrpm_offsets[i] != MSR_INVALID)
774 			continue;
775 
776 		/* Add offset to list */
777 		msrpm_offsets[i] = offset;
778 
779 		return;
780 	}
781 
782 	/*
783 	 * If this BUG triggers the msrpm_offsets table has an overflow. Just
784 	 * increase MSRPM_OFFSETS in this case.
785 	 */
786 	BUG();
787 }
788 
789 static void init_msrpm_offsets(void)
790 {
791 	int i;
792 
793 	memset(msrpm_offsets, 0xff, sizeof(msrpm_offsets));
794 
795 	for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
796 		u32 offset;
797 
798 		offset = svm_msrpm_offset(direct_access_msrs[i].index);
799 		BUG_ON(offset == MSR_INVALID);
800 
801 		add_msr_offset(offset);
802 	}
803 }
804 
805 static void svm_enable_lbrv(struct kvm_vcpu *vcpu)
806 {
807 	struct vcpu_svm *svm = to_svm(vcpu);
808 
809 	svm->vmcb->control.virt_ext |= LBR_CTL_ENABLE_MASK;
810 	set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
811 	set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
812 	set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
813 	set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
814 }
815 
816 static void svm_disable_lbrv(struct kvm_vcpu *vcpu)
817 {
818 	struct vcpu_svm *svm = to_svm(vcpu);
819 
820 	svm->vmcb->control.virt_ext &= ~LBR_CTL_ENABLE_MASK;
821 	set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
822 	set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
823 	set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
824 	set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
825 }
826 
827 void disable_nmi_singlestep(struct vcpu_svm *svm)
828 {
829 	svm->nmi_singlestep = false;
830 
831 	if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP)) {
832 		/* Clear our flags if they were not set by the guest */
833 		if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF))
834 			svm->vmcb->save.rflags &= ~X86_EFLAGS_TF;
835 		if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_RF))
836 			svm->vmcb->save.rflags &= ~X86_EFLAGS_RF;
837 	}
838 }
839 
840 static void grow_ple_window(struct kvm_vcpu *vcpu)
841 {
842 	struct vcpu_svm *svm = to_svm(vcpu);
843 	struct vmcb_control_area *control = &svm->vmcb->control;
844 	int old = control->pause_filter_count;
845 
846 	control->pause_filter_count = __grow_ple_window(old,
847 							pause_filter_count,
848 							pause_filter_count_grow,
849 							pause_filter_count_max);
850 
851 	if (control->pause_filter_count != old) {
852 		vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
853 		trace_kvm_ple_window_update(vcpu->vcpu_id,
854 					    control->pause_filter_count, old);
855 	}
856 }
857 
858 static void shrink_ple_window(struct kvm_vcpu *vcpu)
859 {
860 	struct vcpu_svm *svm = to_svm(vcpu);
861 	struct vmcb_control_area *control = &svm->vmcb->control;
862 	int old = control->pause_filter_count;
863 
864 	control->pause_filter_count =
865 				__shrink_ple_window(old,
866 						    pause_filter_count,
867 						    pause_filter_count_shrink,
868 						    pause_filter_count);
869 	if (control->pause_filter_count != old) {
870 		vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
871 		trace_kvm_ple_window_update(vcpu->vcpu_id,
872 					    control->pause_filter_count, old);
873 	}
874 }
875 
876 static void svm_hardware_teardown(void)
877 {
878 	int cpu;
879 
880 	sev_hardware_teardown();
881 
882 	for_each_possible_cpu(cpu)
883 		svm_cpu_uninit(cpu);
884 
885 	__free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT),
886 	get_order(IOPM_SIZE));
887 	iopm_base = 0;
888 }
889 
890 static void init_seg(struct vmcb_seg *seg)
891 {
892 	seg->selector = 0;
893 	seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
894 		      SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
895 	seg->limit = 0xffff;
896 	seg->base = 0;
897 }
898 
899 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
900 {
901 	seg->selector = 0;
902 	seg->attrib = SVM_SELECTOR_P_MASK | type;
903 	seg->limit = 0xffff;
904 	seg->base = 0;
905 }
906 
907 static u64 svm_get_l2_tsc_offset(struct kvm_vcpu *vcpu)
908 {
909 	struct vcpu_svm *svm = to_svm(vcpu);
910 
911 	return svm->nested.ctl.tsc_offset;
912 }
913 
914 static u64 svm_get_l2_tsc_multiplier(struct kvm_vcpu *vcpu)
915 {
916 	struct vcpu_svm *svm = to_svm(vcpu);
917 
918 	return svm->tsc_ratio_msr;
919 }
920 
921 static void svm_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
922 {
923 	struct vcpu_svm *svm = to_svm(vcpu);
924 
925 	svm->vmcb01.ptr->control.tsc_offset = vcpu->arch.l1_tsc_offset;
926 	svm->vmcb->control.tsc_offset = offset;
927 	vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
928 }
929 
930 void svm_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 multiplier)
931 {
932 	wrmsrl(MSR_AMD64_TSC_RATIO, multiplier);
933 }
934 
935 /* Evaluate instruction intercepts that depend on guest CPUID features. */
936 static void svm_recalc_instruction_intercepts(struct kvm_vcpu *vcpu,
937 					      struct vcpu_svm *svm)
938 {
939 	/*
940 	 * Intercept INVPCID if shadow paging is enabled to sync/free shadow
941 	 * roots, or if INVPCID is disabled in the guest to inject #UD.
942 	 */
943 	if (kvm_cpu_cap_has(X86_FEATURE_INVPCID)) {
944 		if (!npt_enabled ||
945 		    !guest_cpuid_has(&svm->vcpu, X86_FEATURE_INVPCID))
946 			svm_set_intercept(svm, INTERCEPT_INVPCID);
947 		else
948 			svm_clr_intercept(svm, INTERCEPT_INVPCID);
949 	}
950 
951 	if (kvm_cpu_cap_has(X86_FEATURE_RDTSCP)) {
952 		if (guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
953 			svm_clr_intercept(svm, INTERCEPT_RDTSCP);
954 		else
955 			svm_set_intercept(svm, INTERCEPT_RDTSCP);
956 	}
957 }
958 
959 static inline void init_vmcb_after_set_cpuid(struct kvm_vcpu *vcpu)
960 {
961 	struct vcpu_svm *svm = to_svm(vcpu);
962 
963 	if (guest_cpuid_is_intel(vcpu)) {
964 		/*
965 		 * We must intercept SYSENTER_EIP and SYSENTER_ESP
966 		 * accesses because the processor only stores 32 bits.
967 		 * For the same reason we cannot use virtual VMLOAD/VMSAVE.
968 		 */
969 		svm_set_intercept(svm, INTERCEPT_VMLOAD);
970 		svm_set_intercept(svm, INTERCEPT_VMSAVE);
971 		svm->vmcb->control.virt_ext &= ~VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK;
972 
973 		set_msr_interception(vcpu, svm->msrpm, MSR_IA32_SYSENTER_EIP, 0, 0);
974 		set_msr_interception(vcpu, svm->msrpm, MSR_IA32_SYSENTER_ESP, 0, 0);
975 	} else {
976 		/*
977 		 * If hardware supports Virtual VMLOAD VMSAVE then enable it
978 		 * in VMCB and clear intercepts to avoid #VMEXIT.
979 		 */
980 		if (vls) {
981 			svm_clr_intercept(svm, INTERCEPT_VMLOAD);
982 			svm_clr_intercept(svm, INTERCEPT_VMSAVE);
983 			svm->vmcb->control.virt_ext |= VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK;
984 		}
985 		/* No need to intercept these MSRs */
986 		set_msr_interception(vcpu, svm->msrpm, MSR_IA32_SYSENTER_EIP, 1, 1);
987 		set_msr_interception(vcpu, svm->msrpm, MSR_IA32_SYSENTER_ESP, 1, 1);
988 	}
989 }
990 
991 static void init_vmcb(struct kvm_vcpu *vcpu)
992 {
993 	struct vcpu_svm *svm = to_svm(vcpu);
994 	struct vmcb_control_area *control = &svm->vmcb->control;
995 	struct vmcb_save_area *save = &svm->vmcb->save;
996 
997 	svm_set_intercept(svm, INTERCEPT_CR0_READ);
998 	svm_set_intercept(svm, INTERCEPT_CR3_READ);
999 	svm_set_intercept(svm, INTERCEPT_CR4_READ);
1000 	svm_set_intercept(svm, INTERCEPT_CR0_WRITE);
1001 	svm_set_intercept(svm, INTERCEPT_CR3_WRITE);
1002 	svm_set_intercept(svm, INTERCEPT_CR4_WRITE);
1003 	if (!kvm_vcpu_apicv_active(vcpu))
1004 		svm_set_intercept(svm, INTERCEPT_CR8_WRITE);
1005 
1006 	set_dr_intercepts(svm);
1007 
1008 	set_exception_intercept(svm, PF_VECTOR);
1009 	set_exception_intercept(svm, UD_VECTOR);
1010 	set_exception_intercept(svm, MC_VECTOR);
1011 	set_exception_intercept(svm, AC_VECTOR);
1012 	set_exception_intercept(svm, DB_VECTOR);
1013 	/*
1014 	 * Guest access to VMware backdoor ports could legitimately
1015 	 * trigger #GP because of TSS I/O permission bitmap.
1016 	 * We intercept those #GP and allow access to them anyway
1017 	 * as VMware does.  Don't intercept #GP for SEV guests as KVM can't
1018 	 * decrypt guest memory to decode the faulting instruction.
1019 	 */
1020 	if (enable_vmware_backdoor && !sev_guest(vcpu->kvm))
1021 		set_exception_intercept(svm, GP_VECTOR);
1022 
1023 	svm_set_intercept(svm, INTERCEPT_INTR);
1024 	svm_set_intercept(svm, INTERCEPT_NMI);
1025 
1026 	if (intercept_smi)
1027 		svm_set_intercept(svm, INTERCEPT_SMI);
1028 
1029 	svm_set_intercept(svm, INTERCEPT_SELECTIVE_CR0);
1030 	svm_set_intercept(svm, INTERCEPT_RDPMC);
1031 	svm_set_intercept(svm, INTERCEPT_CPUID);
1032 	svm_set_intercept(svm, INTERCEPT_INVD);
1033 	svm_set_intercept(svm, INTERCEPT_INVLPG);
1034 	svm_set_intercept(svm, INTERCEPT_INVLPGA);
1035 	svm_set_intercept(svm, INTERCEPT_IOIO_PROT);
1036 	svm_set_intercept(svm, INTERCEPT_MSR_PROT);
1037 	svm_set_intercept(svm, INTERCEPT_TASK_SWITCH);
1038 	svm_set_intercept(svm, INTERCEPT_SHUTDOWN);
1039 	svm_set_intercept(svm, INTERCEPT_VMRUN);
1040 	svm_set_intercept(svm, INTERCEPT_VMMCALL);
1041 	svm_set_intercept(svm, INTERCEPT_VMLOAD);
1042 	svm_set_intercept(svm, INTERCEPT_VMSAVE);
1043 	svm_set_intercept(svm, INTERCEPT_STGI);
1044 	svm_set_intercept(svm, INTERCEPT_CLGI);
1045 	svm_set_intercept(svm, INTERCEPT_SKINIT);
1046 	svm_set_intercept(svm, INTERCEPT_WBINVD);
1047 	svm_set_intercept(svm, INTERCEPT_XSETBV);
1048 	svm_set_intercept(svm, INTERCEPT_RDPRU);
1049 	svm_set_intercept(svm, INTERCEPT_RSM);
1050 
1051 	if (!kvm_mwait_in_guest(vcpu->kvm)) {
1052 		svm_set_intercept(svm, INTERCEPT_MONITOR);
1053 		svm_set_intercept(svm, INTERCEPT_MWAIT);
1054 	}
1055 
1056 	if (!kvm_hlt_in_guest(vcpu->kvm))
1057 		svm_set_intercept(svm, INTERCEPT_HLT);
1058 
1059 	control->iopm_base_pa = __sme_set(iopm_base);
1060 	control->msrpm_base_pa = __sme_set(__pa(svm->msrpm));
1061 	control->int_ctl = V_INTR_MASKING_MASK;
1062 
1063 	init_seg(&save->es);
1064 	init_seg(&save->ss);
1065 	init_seg(&save->ds);
1066 	init_seg(&save->fs);
1067 	init_seg(&save->gs);
1068 
1069 	save->cs.selector = 0xf000;
1070 	save->cs.base = 0xffff0000;
1071 	/* Executable/Readable Code Segment */
1072 	save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
1073 		SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
1074 	save->cs.limit = 0xffff;
1075 
1076 	save->gdtr.base = 0;
1077 	save->gdtr.limit = 0xffff;
1078 	save->idtr.base = 0;
1079 	save->idtr.limit = 0xffff;
1080 
1081 	init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
1082 	init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
1083 
1084 	if (npt_enabled) {
1085 		/* Setup VMCB for Nested Paging */
1086 		control->nested_ctl |= SVM_NESTED_CTL_NP_ENABLE;
1087 		svm_clr_intercept(svm, INTERCEPT_INVLPG);
1088 		clr_exception_intercept(svm, PF_VECTOR);
1089 		svm_clr_intercept(svm, INTERCEPT_CR3_READ);
1090 		svm_clr_intercept(svm, INTERCEPT_CR3_WRITE);
1091 		save->g_pat = vcpu->arch.pat;
1092 		save->cr3 = 0;
1093 	}
1094 	svm->current_vmcb->asid_generation = 0;
1095 	svm->asid = 0;
1096 
1097 	svm->nested.vmcb12_gpa = INVALID_GPA;
1098 	svm->nested.last_vmcb12_gpa = INVALID_GPA;
1099 
1100 	if (!kvm_pause_in_guest(vcpu->kvm)) {
1101 		control->pause_filter_count = pause_filter_count;
1102 		if (pause_filter_thresh)
1103 			control->pause_filter_thresh = pause_filter_thresh;
1104 		svm_set_intercept(svm, INTERCEPT_PAUSE);
1105 	} else {
1106 		svm_clr_intercept(svm, INTERCEPT_PAUSE);
1107 	}
1108 
1109 	svm_recalc_instruction_intercepts(vcpu, svm);
1110 
1111 	/*
1112 	 * If the host supports V_SPEC_CTRL then disable the interception
1113 	 * of MSR_IA32_SPEC_CTRL.
1114 	 */
1115 	if (boot_cpu_has(X86_FEATURE_V_SPEC_CTRL))
1116 		set_msr_interception(vcpu, svm->msrpm, MSR_IA32_SPEC_CTRL, 1, 1);
1117 
1118 	if (kvm_vcpu_apicv_active(vcpu))
1119 		avic_init_vmcb(svm);
1120 
1121 	if (vgif) {
1122 		svm_clr_intercept(svm, INTERCEPT_STGI);
1123 		svm_clr_intercept(svm, INTERCEPT_CLGI);
1124 		svm->vmcb->control.int_ctl |= V_GIF_ENABLE_MASK;
1125 	}
1126 
1127 	if (sev_guest(vcpu->kvm)) {
1128 		svm->vmcb->control.nested_ctl |= SVM_NESTED_CTL_SEV_ENABLE;
1129 		clr_exception_intercept(svm, UD_VECTOR);
1130 
1131 		if (sev_es_guest(vcpu->kvm)) {
1132 			/* Perform SEV-ES specific VMCB updates */
1133 			sev_es_init_vmcb(svm);
1134 		}
1135 	}
1136 
1137 	svm_hv_init_vmcb(svm->vmcb);
1138 	init_vmcb_after_set_cpuid(vcpu);
1139 
1140 	vmcb_mark_all_dirty(svm->vmcb);
1141 
1142 	enable_gif(svm);
1143 }
1144 
1145 static void __svm_vcpu_reset(struct kvm_vcpu *vcpu)
1146 {
1147 	struct vcpu_svm *svm = to_svm(vcpu);
1148 
1149 	svm_vcpu_init_msrpm(vcpu, svm->msrpm);
1150 
1151 	svm_init_osvw(vcpu);
1152 	vcpu->arch.microcode_version = 0x01000065;
1153 	svm->tsc_ratio_msr = kvm_default_tsc_scaling_ratio;
1154 
1155 	if (sev_es_guest(vcpu->kvm))
1156 		sev_es_vcpu_reset(svm);
1157 }
1158 
1159 static void svm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
1160 {
1161 	struct vcpu_svm *svm = to_svm(vcpu);
1162 
1163 	svm->spec_ctrl = 0;
1164 	svm->virt_spec_ctrl = 0;
1165 
1166 	init_vmcb(vcpu);
1167 
1168 	if (!init_event)
1169 		__svm_vcpu_reset(vcpu);
1170 }
1171 
1172 void svm_switch_vmcb(struct vcpu_svm *svm, struct kvm_vmcb_info *target_vmcb)
1173 {
1174 	svm->current_vmcb = target_vmcb;
1175 	svm->vmcb = target_vmcb->ptr;
1176 }
1177 
1178 static int svm_create_vcpu(struct kvm_vcpu *vcpu)
1179 {
1180 	struct vcpu_svm *svm;
1181 	struct page *vmcb01_page;
1182 	struct page *vmsa_page = NULL;
1183 	int err;
1184 
1185 	BUILD_BUG_ON(offsetof(struct vcpu_svm, vcpu) != 0);
1186 	svm = to_svm(vcpu);
1187 
1188 	err = -ENOMEM;
1189 	vmcb01_page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
1190 	if (!vmcb01_page)
1191 		goto out;
1192 
1193 	if (sev_es_guest(vcpu->kvm)) {
1194 		/*
1195 		 * SEV-ES guests require a separate VMSA page used to contain
1196 		 * the encrypted register state of the guest.
1197 		 */
1198 		vmsa_page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
1199 		if (!vmsa_page)
1200 			goto error_free_vmcb_page;
1201 
1202 		/*
1203 		 * SEV-ES guests maintain an encrypted version of their FPU
1204 		 * state which is restored and saved on VMRUN and VMEXIT.
1205 		 * Mark vcpu->arch.guest_fpu->fpstate as scratch so it won't
1206 		 * do xsave/xrstor on it.
1207 		 */
1208 		fpstate_set_confidential(&vcpu->arch.guest_fpu);
1209 	}
1210 
1211 	err = avic_init_vcpu(svm);
1212 	if (err)
1213 		goto error_free_vmsa_page;
1214 
1215 	svm->msrpm = svm_vcpu_alloc_msrpm();
1216 	if (!svm->msrpm) {
1217 		err = -ENOMEM;
1218 		goto error_free_vmsa_page;
1219 	}
1220 
1221 	svm->vmcb01.ptr = page_address(vmcb01_page);
1222 	svm->vmcb01.pa = __sme_set(page_to_pfn(vmcb01_page) << PAGE_SHIFT);
1223 	svm_switch_vmcb(svm, &svm->vmcb01);
1224 
1225 	if (vmsa_page)
1226 		svm->sev_es.vmsa = page_address(vmsa_page);
1227 
1228 	svm->guest_state_loaded = false;
1229 
1230 	return 0;
1231 
1232 error_free_vmsa_page:
1233 	if (vmsa_page)
1234 		__free_page(vmsa_page);
1235 error_free_vmcb_page:
1236 	__free_page(vmcb01_page);
1237 out:
1238 	return err;
1239 }
1240 
1241 static void svm_clear_current_vmcb(struct vmcb *vmcb)
1242 {
1243 	int i;
1244 
1245 	for_each_online_cpu(i)
1246 		cmpxchg(&per_cpu(svm_data, i)->current_vmcb, vmcb, NULL);
1247 }
1248 
1249 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
1250 {
1251 	struct vcpu_svm *svm = to_svm(vcpu);
1252 
1253 	/*
1254 	 * The vmcb page can be recycled, causing a false negative in
1255 	 * svm_vcpu_load(). So, ensure that no logical CPU has this
1256 	 * vmcb page recorded as its current vmcb.
1257 	 */
1258 	svm_clear_current_vmcb(svm->vmcb);
1259 
1260 	svm_free_nested(svm);
1261 
1262 	sev_free_vcpu(vcpu);
1263 
1264 	__free_page(pfn_to_page(__sme_clr(svm->vmcb01.pa) >> PAGE_SHIFT));
1265 	__free_pages(virt_to_page(svm->msrpm), get_order(MSRPM_SIZE));
1266 }
1267 
1268 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
1269 {
1270 	struct vcpu_svm *svm = to_svm(vcpu);
1271 	struct svm_cpu_data *sd = per_cpu(svm_data, vcpu->cpu);
1272 
1273 	if (sev_es_guest(vcpu->kvm))
1274 		sev_es_unmap_ghcb(svm);
1275 
1276 	if (svm->guest_state_loaded)
1277 		return;
1278 
1279 	/*
1280 	 * Save additional host state that will be restored on VMEXIT (sev-es)
1281 	 * or subsequent vmload of host save area.
1282 	 */
1283 	if (sev_es_guest(vcpu->kvm)) {
1284 		sev_es_prepare_guest_switch(svm, vcpu->cpu);
1285 	} else {
1286 		vmsave(__sme_page_pa(sd->save_area));
1287 	}
1288 
1289 	if (tsc_scaling) {
1290 		u64 tsc_ratio = vcpu->arch.tsc_scaling_ratio;
1291 		if (tsc_ratio != __this_cpu_read(current_tsc_ratio)) {
1292 			__this_cpu_write(current_tsc_ratio, tsc_ratio);
1293 			wrmsrl(MSR_AMD64_TSC_RATIO, tsc_ratio);
1294 		}
1295 	}
1296 
1297 	if (likely(tsc_aux_uret_slot >= 0))
1298 		kvm_set_user_return_msr(tsc_aux_uret_slot, svm->tsc_aux, -1ull);
1299 
1300 	svm->guest_state_loaded = true;
1301 }
1302 
1303 static void svm_prepare_host_switch(struct kvm_vcpu *vcpu)
1304 {
1305 	to_svm(vcpu)->guest_state_loaded = false;
1306 }
1307 
1308 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1309 {
1310 	struct vcpu_svm *svm = to_svm(vcpu);
1311 	struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
1312 
1313 	if (sd->current_vmcb != svm->vmcb) {
1314 		sd->current_vmcb = svm->vmcb;
1315 		indirect_branch_prediction_barrier();
1316 	}
1317 	if (kvm_vcpu_apicv_active(vcpu))
1318 		avic_vcpu_load(vcpu, cpu);
1319 }
1320 
1321 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
1322 {
1323 	if (kvm_vcpu_apicv_active(vcpu))
1324 		avic_vcpu_put(vcpu);
1325 
1326 	svm_prepare_host_switch(vcpu);
1327 
1328 	++vcpu->stat.host_state_reload;
1329 }
1330 
1331 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
1332 {
1333 	struct vcpu_svm *svm = to_svm(vcpu);
1334 	unsigned long rflags = svm->vmcb->save.rflags;
1335 
1336 	if (svm->nmi_singlestep) {
1337 		/* Hide our flags if they were not set by the guest */
1338 		if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF))
1339 			rflags &= ~X86_EFLAGS_TF;
1340 		if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_RF))
1341 			rflags &= ~X86_EFLAGS_RF;
1342 	}
1343 	return rflags;
1344 }
1345 
1346 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1347 {
1348 	if (to_svm(vcpu)->nmi_singlestep)
1349 		rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
1350 
1351        /*
1352         * Any change of EFLAGS.VM is accompanied by a reload of SS
1353         * (caused by either a task switch or an inter-privilege IRET),
1354         * so we do not need to update the CPL here.
1355         */
1356 	to_svm(vcpu)->vmcb->save.rflags = rflags;
1357 }
1358 
1359 static bool svm_get_if_flag(struct kvm_vcpu *vcpu)
1360 {
1361 	struct vmcb *vmcb = to_svm(vcpu)->vmcb;
1362 
1363 	return sev_es_guest(vcpu->kvm)
1364 		? vmcb->control.int_state & SVM_GUEST_INTERRUPT_MASK
1365 		: kvm_get_rflags(vcpu) & X86_EFLAGS_IF;
1366 }
1367 
1368 static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1369 {
1370 	kvm_register_mark_available(vcpu, reg);
1371 
1372 	switch (reg) {
1373 	case VCPU_EXREG_PDPTR:
1374 		/*
1375 		 * When !npt_enabled, mmu->pdptrs[] is already available since
1376 		 * it is always updated per SDM when moving to CRs.
1377 		 */
1378 		if (npt_enabled)
1379 			load_pdptrs(vcpu, kvm_read_cr3(vcpu));
1380 		break;
1381 	default:
1382 		KVM_BUG_ON(1, vcpu->kvm);
1383 	}
1384 }
1385 
1386 static void svm_set_vintr(struct vcpu_svm *svm)
1387 {
1388 	struct vmcb_control_area *control;
1389 
1390 	/*
1391 	 * The following fields are ignored when AVIC is enabled
1392 	 */
1393 	WARN_ON(kvm_apicv_activated(svm->vcpu.kvm));
1394 
1395 	svm_set_intercept(svm, INTERCEPT_VINTR);
1396 
1397 	/*
1398 	 * This is just a dummy VINTR to actually cause a vmexit to happen.
1399 	 * Actual injection of virtual interrupts happens through EVENTINJ.
1400 	 */
1401 	control = &svm->vmcb->control;
1402 	control->int_vector = 0x0;
1403 	control->int_ctl &= ~V_INTR_PRIO_MASK;
1404 	control->int_ctl |= V_IRQ_MASK |
1405 		((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
1406 	vmcb_mark_dirty(svm->vmcb, VMCB_INTR);
1407 }
1408 
1409 static void svm_clear_vintr(struct vcpu_svm *svm)
1410 {
1411 	svm_clr_intercept(svm, INTERCEPT_VINTR);
1412 
1413 	/* Drop int_ctl fields related to VINTR injection.  */
1414 	svm->vmcb->control.int_ctl &= ~V_IRQ_INJECTION_BITS_MASK;
1415 	if (is_guest_mode(&svm->vcpu)) {
1416 		svm->vmcb01.ptr->control.int_ctl &= ~V_IRQ_INJECTION_BITS_MASK;
1417 
1418 		WARN_ON((svm->vmcb->control.int_ctl & V_TPR_MASK) !=
1419 			(svm->nested.ctl.int_ctl & V_TPR_MASK));
1420 
1421 		svm->vmcb->control.int_ctl |= svm->nested.ctl.int_ctl &
1422 			V_IRQ_INJECTION_BITS_MASK;
1423 
1424 		svm->vmcb->control.int_vector = svm->nested.ctl.int_vector;
1425 	}
1426 
1427 	vmcb_mark_dirty(svm->vmcb, VMCB_INTR);
1428 }
1429 
1430 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
1431 {
1432 	struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1433 	struct vmcb_save_area *save01 = &to_svm(vcpu)->vmcb01.ptr->save;
1434 
1435 	switch (seg) {
1436 	case VCPU_SREG_CS: return &save->cs;
1437 	case VCPU_SREG_DS: return &save->ds;
1438 	case VCPU_SREG_ES: return &save->es;
1439 	case VCPU_SREG_FS: return &save01->fs;
1440 	case VCPU_SREG_GS: return &save01->gs;
1441 	case VCPU_SREG_SS: return &save->ss;
1442 	case VCPU_SREG_TR: return &save01->tr;
1443 	case VCPU_SREG_LDTR: return &save01->ldtr;
1444 	}
1445 	BUG();
1446 	return NULL;
1447 }
1448 
1449 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1450 {
1451 	struct vmcb_seg *s = svm_seg(vcpu, seg);
1452 
1453 	return s->base;
1454 }
1455 
1456 static void svm_get_segment(struct kvm_vcpu *vcpu,
1457 			    struct kvm_segment *var, int seg)
1458 {
1459 	struct vmcb_seg *s = svm_seg(vcpu, seg);
1460 
1461 	var->base = s->base;
1462 	var->limit = s->limit;
1463 	var->selector = s->selector;
1464 	var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
1465 	var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
1466 	var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
1467 	var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
1468 	var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
1469 	var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
1470 	var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
1471 
1472 	/*
1473 	 * AMD CPUs circa 2014 track the G bit for all segments except CS.
1474 	 * However, the SVM spec states that the G bit is not observed by the
1475 	 * CPU, and some VMware virtual CPUs drop the G bit for all segments.
1476 	 * So let's synthesize a legal G bit for all segments, this helps
1477 	 * running KVM nested. It also helps cross-vendor migration, because
1478 	 * Intel's vmentry has a check on the 'G' bit.
1479 	 */
1480 	var->g = s->limit > 0xfffff;
1481 
1482 	/*
1483 	 * AMD's VMCB does not have an explicit unusable field, so emulate it
1484 	 * for cross vendor migration purposes by "not present"
1485 	 */
1486 	var->unusable = !var->present;
1487 
1488 	switch (seg) {
1489 	case VCPU_SREG_TR:
1490 		/*
1491 		 * Work around a bug where the busy flag in the tr selector
1492 		 * isn't exposed
1493 		 */
1494 		var->type |= 0x2;
1495 		break;
1496 	case VCPU_SREG_DS:
1497 	case VCPU_SREG_ES:
1498 	case VCPU_SREG_FS:
1499 	case VCPU_SREG_GS:
1500 		/*
1501 		 * The accessed bit must always be set in the segment
1502 		 * descriptor cache, although it can be cleared in the
1503 		 * descriptor, the cached bit always remains at 1. Since
1504 		 * Intel has a check on this, set it here to support
1505 		 * cross-vendor migration.
1506 		 */
1507 		if (!var->unusable)
1508 			var->type |= 0x1;
1509 		break;
1510 	case VCPU_SREG_SS:
1511 		/*
1512 		 * On AMD CPUs sometimes the DB bit in the segment
1513 		 * descriptor is left as 1, although the whole segment has
1514 		 * been made unusable. Clear it here to pass an Intel VMX
1515 		 * entry check when cross vendor migrating.
1516 		 */
1517 		if (var->unusable)
1518 			var->db = 0;
1519 		/* This is symmetric with svm_set_segment() */
1520 		var->dpl = to_svm(vcpu)->vmcb->save.cpl;
1521 		break;
1522 	}
1523 }
1524 
1525 static int svm_get_cpl(struct kvm_vcpu *vcpu)
1526 {
1527 	struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1528 
1529 	return save->cpl;
1530 }
1531 
1532 static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1533 {
1534 	struct vcpu_svm *svm = to_svm(vcpu);
1535 
1536 	dt->size = svm->vmcb->save.idtr.limit;
1537 	dt->address = svm->vmcb->save.idtr.base;
1538 }
1539 
1540 static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1541 {
1542 	struct vcpu_svm *svm = to_svm(vcpu);
1543 
1544 	svm->vmcb->save.idtr.limit = dt->size;
1545 	svm->vmcb->save.idtr.base = dt->address ;
1546 	vmcb_mark_dirty(svm->vmcb, VMCB_DT);
1547 }
1548 
1549 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1550 {
1551 	struct vcpu_svm *svm = to_svm(vcpu);
1552 
1553 	dt->size = svm->vmcb->save.gdtr.limit;
1554 	dt->address = svm->vmcb->save.gdtr.base;
1555 }
1556 
1557 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1558 {
1559 	struct vcpu_svm *svm = to_svm(vcpu);
1560 
1561 	svm->vmcb->save.gdtr.limit = dt->size;
1562 	svm->vmcb->save.gdtr.base = dt->address ;
1563 	vmcb_mark_dirty(svm->vmcb, VMCB_DT);
1564 }
1565 
1566 static void svm_post_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1567 {
1568 	struct vcpu_svm *svm = to_svm(vcpu);
1569 
1570 	/*
1571 	 * For guests that don't set guest_state_protected, the cr3 update is
1572 	 * handled via kvm_mmu_load() while entering the guest. For guests
1573 	 * that do (SEV-ES/SEV-SNP), the cr3 update needs to be written to
1574 	 * VMCB save area now, since the save area will become the initial
1575 	 * contents of the VMSA, and future VMCB save area updates won't be
1576 	 * seen.
1577 	 */
1578 	if (sev_es_guest(vcpu->kvm)) {
1579 		svm->vmcb->save.cr3 = cr3;
1580 		vmcb_mark_dirty(svm->vmcb, VMCB_CR);
1581 	}
1582 }
1583 
1584 void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1585 {
1586 	struct vcpu_svm *svm = to_svm(vcpu);
1587 	u64 hcr0 = cr0;
1588 	bool old_paging = is_paging(vcpu);
1589 
1590 #ifdef CONFIG_X86_64
1591 	if (vcpu->arch.efer & EFER_LME && !vcpu->arch.guest_state_protected) {
1592 		if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
1593 			vcpu->arch.efer |= EFER_LMA;
1594 			svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
1595 		}
1596 
1597 		if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
1598 			vcpu->arch.efer &= ~EFER_LMA;
1599 			svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
1600 		}
1601 	}
1602 #endif
1603 	vcpu->arch.cr0 = cr0;
1604 
1605 	if (!npt_enabled) {
1606 		hcr0 |= X86_CR0_PG | X86_CR0_WP;
1607 		if (old_paging != is_paging(vcpu))
1608 			svm_set_cr4(vcpu, kvm_read_cr4(vcpu));
1609 	}
1610 
1611 	/*
1612 	 * re-enable caching here because the QEMU bios
1613 	 * does not do it - this results in some delay at
1614 	 * reboot
1615 	 */
1616 	if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
1617 		hcr0 &= ~(X86_CR0_CD | X86_CR0_NW);
1618 
1619 	svm->vmcb->save.cr0 = hcr0;
1620 	vmcb_mark_dirty(svm->vmcb, VMCB_CR);
1621 
1622 	/*
1623 	 * SEV-ES guests must always keep the CR intercepts cleared. CR
1624 	 * tracking is done using the CR write traps.
1625 	 */
1626 	if (sev_es_guest(vcpu->kvm))
1627 		return;
1628 
1629 	if (hcr0 == cr0) {
1630 		/* Selective CR0 write remains on.  */
1631 		svm_clr_intercept(svm, INTERCEPT_CR0_READ);
1632 		svm_clr_intercept(svm, INTERCEPT_CR0_WRITE);
1633 	} else {
1634 		svm_set_intercept(svm, INTERCEPT_CR0_READ);
1635 		svm_set_intercept(svm, INTERCEPT_CR0_WRITE);
1636 	}
1637 }
1638 
1639 static bool svm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1640 {
1641 	return true;
1642 }
1643 
1644 void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1645 {
1646 	unsigned long host_cr4_mce = cr4_read_shadow() & X86_CR4_MCE;
1647 	unsigned long old_cr4 = vcpu->arch.cr4;
1648 
1649 	if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
1650 		svm_flush_tlb(vcpu);
1651 
1652 	vcpu->arch.cr4 = cr4;
1653 	if (!npt_enabled) {
1654 		cr4 |= X86_CR4_PAE;
1655 
1656 		if (!is_paging(vcpu))
1657 			cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
1658 	}
1659 	cr4 |= host_cr4_mce;
1660 	to_svm(vcpu)->vmcb->save.cr4 = cr4;
1661 	vmcb_mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
1662 
1663 	if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
1664 		kvm_update_cpuid_runtime(vcpu);
1665 }
1666 
1667 static void svm_set_segment(struct kvm_vcpu *vcpu,
1668 			    struct kvm_segment *var, int seg)
1669 {
1670 	struct vcpu_svm *svm = to_svm(vcpu);
1671 	struct vmcb_seg *s = svm_seg(vcpu, seg);
1672 
1673 	s->base = var->base;
1674 	s->limit = var->limit;
1675 	s->selector = var->selector;
1676 	s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
1677 	s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
1678 	s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
1679 	s->attrib |= ((var->present & 1) && !var->unusable) << SVM_SELECTOR_P_SHIFT;
1680 	s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
1681 	s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
1682 	s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
1683 	s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
1684 
1685 	/*
1686 	 * This is always accurate, except if SYSRET returned to a segment
1687 	 * with SS.DPL != 3.  Intel does not have this quirk, and always
1688 	 * forces SS.DPL to 3 on sysret, so we ignore that case; fixing it
1689 	 * would entail passing the CPL to userspace and back.
1690 	 */
1691 	if (seg == VCPU_SREG_SS)
1692 		/* This is symmetric with svm_get_segment() */
1693 		svm->vmcb->save.cpl = (var->dpl & 3);
1694 
1695 	vmcb_mark_dirty(svm->vmcb, VMCB_SEG);
1696 }
1697 
1698 static void svm_update_exception_bitmap(struct kvm_vcpu *vcpu)
1699 {
1700 	struct vcpu_svm *svm = to_svm(vcpu);
1701 
1702 	clr_exception_intercept(svm, BP_VECTOR);
1703 
1704 	if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
1705 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
1706 			set_exception_intercept(svm, BP_VECTOR);
1707 	}
1708 }
1709 
1710 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd)
1711 {
1712 	if (sd->next_asid > sd->max_asid) {
1713 		++sd->asid_generation;
1714 		sd->next_asid = sd->min_asid;
1715 		svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
1716 		vmcb_mark_dirty(svm->vmcb, VMCB_ASID);
1717 	}
1718 
1719 	svm->current_vmcb->asid_generation = sd->asid_generation;
1720 	svm->asid = sd->next_asid++;
1721 }
1722 
1723 static void svm_set_dr6(struct vcpu_svm *svm, unsigned long value)
1724 {
1725 	struct vmcb *vmcb = svm->vmcb;
1726 
1727 	if (svm->vcpu.arch.guest_state_protected)
1728 		return;
1729 
1730 	if (unlikely(value != vmcb->save.dr6)) {
1731 		vmcb->save.dr6 = value;
1732 		vmcb_mark_dirty(vmcb, VMCB_DR);
1733 	}
1734 }
1735 
1736 static void svm_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
1737 {
1738 	struct vcpu_svm *svm = to_svm(vcpu);
1739 
1740 	if (vcpu->arch.guest_state_protected)
1741 		return;
1742 
1743 	get_debugreg(vcpu->arch.db[0], 0);
1744 	get_debugreg(vcpu->arch.db[1], 1);
1745 	get_debugreg(vcpu->arch.db[2], 2);
1746 	get_debugreg(vcpu->arch.db[3], 3);
1747 	/*
1748 	 * We cannot reset svm->vmcb->save.dr6 to DR6_ACTIVE_LOW here,
1749 	 * because db_interception might need it.  We can do it before vmentry.
1750 	 */
1751 	vcpu->arch.dr6 = svm->vmcb->save.dr6;
1752 	vcpu->arch.dr7 = svm->vmcb->save.dr7;
1753 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
1754 	set_dr_intercepts(svm);
1755 }
1756 
1757 static void svm_set_dr7(struct kvm_vcpu *vcpu, unsigned long value)
1758 {
1759 	struct vcpu_svm *svm = to_svm(vcpu);
1760 
1761 	if (vcpu->arch.guest_state_protected)
1762 		return;
1763 
1764 	svm->vmcb->save.dr7 = value;
1765 	vmcb_mark_dirty(svm->vmcb, VMCB_DR);
1766 }
1767 
1768 static int pf_interception(struct kvm_vcpu *vcpu)
1769 {
1770 	struct vcpu_svm *svm = to_svm(vcpu);
1771 
1772 	u64 fault_address = svm->vmcb->control.exit_info_2;
1773 	u64 error_code = svm->vmcb->control.exit_info_1;
1774 
1775 	return kvm_handle_page_fault(vcpu, error_code, fault_address,
1776 			static_cpu_has(X86_FEATURE_DECODEASSISTS) ?
1777 			svm->vmcb->control.insn_bytes : NULL,
1778 			svm->vmcb->control.insn_len);
1779 }
1780 
1781 static int npf_interception(struct kvm_vcpu *vcpu)
1782 {
1783 	struct vcpu_svm *svm = to_svm(vcpu);
1784 
1785 	u64 fault_address = svm->vmcb->control.exit_info_2;
1786 	u64 error_code = svm->vmcb->control.exit_info_1;
1787 
1788 	trace_kvm_page_fault(fault_address, error_code);
1789 	return kvm_mmu_page_fault(vcpu, fault_address, error_code,
1790 			static_cpu_has(X86_FEATURE_DECODEASSISTS) ?
1791 			svm->vmcb->control.insn_bytes : NULL,
1792 			svm->vmcb->control.insn_len);
1793 }
1794 
1795 static int db_interception(struct kvm_vcpu *vcpu)
1796 {
1797 	struct kvm_run *kvm_run = vcpu->run;
1798 	struct vcpu_svm *svm = to_svm(vcpu);
1799 
1800 	if (!(vcpu->guest_debug &
1801 	      (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
1802 		!svm->nmi_singlestep) {
1803 		u32 payload = svm->vmcb->save.dr6 ^ DR6_ACTIVE_LOW;
1804 		kvm_queue_exception_p(vcpu, DB_VECTOR, payload);
1805 		return 1;
1806 	}
1807 
1808 	if (svm->nmi_singlestep) {
1809 		disable_nmi_singlestep(svm);
1810 		/* Make sure we check for pending NMIs upon entry */
1811 		kvm_make_request(KVM_REQ_EVENT, vcpu);
1812 	}
1813 
1814 	if (vcpu->guest_debug &
1815 	    (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) {
1816 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
1817 		kvm_run->debug.arch.dr6 = svm->vmcb->save.dr6;
1818 		kvm_run->debug.arch.dr7 = svm->vmcb->save.dr7;
1819 		kvm_run->debug.arch.pc =
1820 			svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1821 		kvm_run->debug.arch.exception = DB_VECTOR;
1822 		return 0;
1823 	}
1824 
1825 	return 1;
1826 }
1827 
1828 static int bp_interception(struct kvm_vcpu *vcpu)
1829 {
1830 	struct vcpu_svm *svm = to_svm(vcpu);
1831 	struct kvm_run *kvm_run = vcpu->run;
1832 
1833 	kvm_run->exit_reason = KVM_EXIT_DEBUG;
1834 	kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1835 	kvm_run->debug.arch.exception = BP_VECTOR;
1836 	return 0;
1837 }
1838 
1839 static int ud_interception(struct kvm_vcpu *vcpu)
1840 {
1841 	return handle_ud(vcpu);
1842 }
1843 
1844 static int ac_interception(struct kvm_vcpu *vcpu)
1845 {
1846 	kvm_queue_exception_e(vcpu, AC_VECTOR, 0);
1847 	return 1;
1848 }
1849 
1850 static bool is_erratum_383(void)
1851 {
1852 	int err, i;
1853 	u64 value;
1854 
1855 	if (!erratum_383_found)
1856 		return false;
1857 
1858 	value = native_read_msr_safe(MSR_IA32_MC0_STATUS, &err);
1859 	if (err)
1860 		return false;
1861 
1862 	/* Bit 62 may or may not be set for this mce */
1863 	value &= ~(1ULL << 62);
1864 
1865 	if (value != 0xb600000000010015ULL)
1866 		return false;
1867 
1868 	/* Clear MCi_STATUS registers */
1869 	for (i = 0; i < 6; ++i)
1870 		native_write_msr_safe(MSR_IA32_MCx_STATUS(i), 0, 0);
1871 
1872 	value = native_read_msr_safe(MSR_IA32_MCG_STATUS, &err);
1873 	if (!err) {
1874 		u32 low, high;
1875 
1876 		value &= ~(1ULL << 2);
1877 		low    = lower_32_bits(value);
1878 		high   = upper_32_bits(value);
1879 
1880 		native_write_msr_safe(MSR_IA32_MCG_STATUS, low, high);
1881 	}
1882 
1883 	/* Flush tlb to evict multi-match entries */
1884 	__flush_tlb_all();
1885 
1886 	return true;
1887 }
1888 
1889 static void svm_handle_mce(struct kvm_vcpu *vcpu)
1890 {
1891 	if (is_erratum_383()) {
1892 		/*
1893 		 * Erratum 383 triggered. Guest state is corrupt so kill the
1894 		 * guest.
1895 		 */
1896 		pr_err("KVM: Guest triggered AMD Erratum 383\n");
1897 
1898 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
1899 
1900 		return;
1901 	}
1902 
1903 	/*
1904 	 * On an #MC intercept the MCE handler is not called automatically in
1905 	 * the host. So do it by hand here.
1906 	 */
1907 	kvm_machine_check();
1908 }
1909 
1910 static int mc_interception(struct kvm_vcpu *vcpu)
1911 {
1912 	return 1;
1913 }
1914 
1915 static int shutdown_interception(struct kvm_vcpu *vcpu)
1916 {
1917 	struct kvm_run *kvm_run = vcpu->run;
1918 	struct vcpu_svm *svm = to_svm(vcpu);
1919 
1920 	/*
1921 	 * The VM save area has already been encrypted so it
1922 	 * cannot be reinitialized - just terminate.
1923 	 */
1924 	if (sev_es_guest(vcpu->kvm))
1925 		return -EINVAL;
1926 
1927 	/*
1928 	 * VMCB is undefined after a SHUTDOWN intercept.  INIT the vCPU to put
1929 	 * the VMCB in a known good state.  Unfortuately, KVM doesn't have
1930 	 * KVM_MP_STATE_SHUTDOWN and can't add it without potentially breaking
1931 	 * userspace.  At a platform view, INIT is acceptable behavior as
1932 	 * there exist bare metal platforms that automatically INIT the CPU
1933 	 * in response to shutdown.
1934 	 */
1935 	clear_page(svm->vmcb);
1936 	kvm_vcpu_reset(vcpu, true);
1937 
1938 	kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1939 	return 0;
1940 }
1941 
1942 static int io_interception(struct kvm_vcpu *vcpu)
1943 {
1944 	struct vcpu_svm *svm = to_svm(vcpu);
1945 	u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
1946 	int size, in, string;
1947 	unsigned port;
1948 
1949 	++vcpu->stat.io_exits;
1950 	string = (io_info & SVM_IOIO_STR_MASK) != 0;
1951 	in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
1952 	port = io_info >> 16;
1953 	size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
1954 
1955 	if (string) {
1956 		if (sev_es_guest(vcpu->kvm))
1957 			return sev_es_string_io(svm, size, port, in);
1958 		else
1959 			return kvm_emulate_instruction(vcpu, 0);
1960 	}
1961 
1962 	svm->next_rip = svm->vmcb->control.exit_info_2;
1963 
1964 	return kvm_fast_pio(vcpu, size, port, in);
1965 }
1966 
1967 static int nmi_interception(struct kvm_vcpu *vcpu)
1968 {
1969 	return 1;
1970 }
1971 
1972 static int smi_interception(struct kvm_vcpu *vcpu)
1973 {
1974 	return 1;
1975 }
1976 
1977 static int intr_interception(struct kvm_vcpu *vcpu)
1978 {
1979 	++vcpu->stat.irq_exits;
1980 	return 1;
1981 }
1982 
1983 static int vmload_vmsave_interception(struct kvm_vcpu *vcpu, bool vmload)
1984 {
1985 	struct vcpu_svm *svm = to_svm(vcpu);
1986 	struct vmcb *vmcb12;
1987 	struct kvm_host_map map;
1988 	int ret;
1989 
1990 	if (nested_svm_check_permissions(vcpu))
1991 		return 1;
1992 
1993 	ret = kvm_vcpu_map(vcpu, gpa_to_gfn(svm->vmcb->save.rax), &map);
1994 	if (ret) {
1995 		if (ret == -EINVAL)
1996 			kvm_inject_gp(vcpu, 0);
1997 		return 1;
1998 	}
1999 
2000 	vmcb12 = map.hva;
2001 
2002 	ret = kvm_skip_emulated_instruction(vcpu);
2003 
2004 	if (vmload) {
2005 		svm_copy_vmloadsave_state(svm->vmcb, vmcb12);
2006 		svm->sysenter_eip_hi = 0;
2007 		svm->sysenter_esp_hi = 0;
2008 	} else {
2009 		svm_copy_vmloadsave_state(vmcb12, svm->vmcb);
2010 	}
2011 
2012 	kvm_vcpu_unmap(vcpu, &map, true);
2013 
2014 	return ret;
2015 }
2016 
2017 static int vmload_interception(struct kvm_vcpu *vcpu)
2018 {
2019 	return vmload_vmsave_interception(vcpu, true);
2020 }
2021 
2022 static int vmsave_interception(struct kvm_vcpu *vcpu)
2023 {
2024 	return vmload_vmsave_interception(vcpu, false);
2025 }
2026 
2027 static int vmrun_interception(struct kvm_vcpu *vcpu)
2028 {
2029 	if (nested_svm_check_permissions(vcpu))
2030 		return 1;
2031 
2032 	return nested_svm_vmrun(vcpu);
2033 }
2034 
2035 enum {
2036 	NONE_SVM_INSTR,
2037 	SVM_INSTR_VMRUN,
2038 	SVM_INSTR_VMLOAD,
2039 	SVM_INSTR_VMSAVE,
2040 };
2041 
2042 /* Return NONE_SVM_INSTR if not SVM instrs, otherwise return decode result */
2043 static int svm_instr_opcode(struct kvm_vcpu *vcpu)
2044 {
2045 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
2046 
2047 	if (ctxt->b != 0x1 || ctxt->opcode_len != 2)
2048 		return NONE_SVM_INSTR;
2049 
2050 	switch (ctxt->modrm) {
2051 	case 0xd8: /* VMRUN */
2052 		return SVM_INSTR_VMRUN;
2053 	case 0xda: /* VMLOAD */
2054 		return SVM_INSTR_VMLOAD;
2055 	case 0xdb: /* VMSAVE */
2056 		return SVM_INSTR_VMSAVE;
2057 	default:
2058 		break;
2059 	}
2060 
2061 	return NONE_SVM_INSTR;
2062 }
2063 
2064 static int emulate_svm_instr(struct kvm_vcpu *vcpu, int opcode)
2065 {
2066 	const int guest_mode_exit_codes[] = {
2067 		[SVM_INSTR_VMRUN] = SVM_EXIT_VMRUN,
2068 		[SVM_INSTR_VMLOAD] = SVM_EXIT_VMLOAD,
2069 		[SVM_INSTR_VMSAVE] = SVM_EXIT_VMSAVE,
2070 	};
2071 	int (*const svm_instr_handlers[])(struct kvm_vcpu *vcpu) = {
2072 		[SVM_INSTR_VMRUN] = vmrun_interception,
2073 		[SVM_INSTR_VMLOAD] = vmload_interception,
2074 		[SVM_INSTR_VMSAVE] = vmsave_interception,
2075 	};
2076 	struct vcpu_svm *svm = to_svm(vcpu);
2077 	int ret;
2078 
2079 	if (is_guest_mode(vcpu)) {
2080 		/* Returns '1' or -errno on failure, '0' on success. */
2081 		ret = nested_svm_simple_vmexit(svm, guest_mode_exit_codes[opcode]);
2082 		if (ret)
2083 			return ret;
2084 		return 1;
2085 	}
2086 	return svm_instr_handlers[opcode](vcpu);
2087 }
2088 
2089 /*
2090  * #GP handling code. Note that #GP can be triggered under the following two
2091  * cases:
2092  *   1) SVM VM-related instructions (VMRUN/VMSAVE/VMLOAD) that trigger #GP on
2093  *      some AMD CPUs when EAX of these instructions are in the reserved memory
2094  *      regions (e.g. SMM memory on host).
2095  *   2) VMware backdoor
2096  */
2097 static int gp_interception(struct kvm_vcpu *vcpu)
2098 {
2099 	struct vcpu_svm *svm = to_svm(vcpu);
2100 	u32 error_code = svm->vmcb->control.exit_info_1;
2101 	int opcode;
2102 
2103 	/* Both #GP cases have zero error_code */
2104 	if (error_code)
2105 		goto reinject;
2106 
2107 	/* Decode the instruction for usage later */
2108 	if (x86_decode_emulated_instruction(vcpu, 0, NULL, 0) != EMULATION_OK)
2109 		goto reinject;
2110 
2111 	opcode = svm_instr_opcode(vcpu);
2112 
2113 	if (opcode == NONE_SVM_INSTR) {
2114 		if (!enable_vmware_backdoor)
2115 			goto reinject;
2116 
2117 		/*
2118 		 * VMware backdoor emulation on #GP interception only handles
2119 		 * IN{S}, OUT{S}, and RDPMC.
2120 		 */
2121 		if (!is_guest_mode(vcpu))
2122 			return kvm_emulate_instruction(vcpu,
2123 				EMULTYPE_VMWARE_GP | EMULTYPE_NO_DECODE);
2124 	} else {
2125 		/* All SVM instructions expect page aligned RAX */
2126 		if (svm->vmcb->save.rax & ~PAGE_MASK)
2127 			goto reinject;
2128 
2129 		return emulate_svm_instr(vcpu, opcode);
2130 	}
2131 
2132 reinject:
2133 	kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
2134 	return 1;
2135 }
2136 
2137 void svm_set_gif(struct vcpu_svm *svm, bool value)
2138 {
2139 	if (value) {
2140 		/*
2141 		 * If VGIF is enabled, the STGI intercept is only added to
2142 		 * detect the opening of the SMI/NMI window; remove it now.
2143 		 * Likewise, clear the VINTR intercept, we will set it
2144 		 * again while processing KVM_REQ_EVENT if needed.
2145 		 */
2146 		if (vgif_enabled(svm))
2147 			svm_clr_intercept(svm, INTERCEPT_STGI);
2148 		if (svm_is_intercept(svm, INTERCEPT_VINTR))
2149 			svm_clear_vintr(svm);
2150 
2151 		enable_gif(svm);
2152 		if (svm->vcpu.arch.smi_pending ||
2153 		    svm->vcpu.arch.nmi_pending ||
2154 		    kvm_cpu_has_injectable_intr(&svm->vcpu))
2155 			kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
2156 	} else {
2157 		disable_gif(svm);
2158 
2159 		/*
2160 		 * After a CLGI no interrupts should come.  But if vGIF is
2161 		 * in use, we still rely on the VINTR intercept (rather than
2162 		 * STGI) to detect an open interrupt window.
2163 		*/
2164 		if (!vgif_enabled(svm))
2165 			svm_clear_vintr(svm);
2166 	}
2167 }
2168 
2169 static int stgi_interception(struct kvm_vcpu *vcpu)
2170 {
2171 	int ret;
2172 
2173 	if (nested_svm_check_permissions(vcpu))
2174 		return 1;
2175 
2176 	ret = kvm_skip_emulated_instruction(vcpu);
2177 	svm_set_gif(to_svm(vcpu), true);
2178 	return ret;
2179 }
2180 
2181 static int clgi_interception(struct kvm_vcpu *vcpu)
2182 {
2183 	int ret;
2184 
2185 	if (nested_svm_check_permissions(vcpu))
2186 		return 1;
2187 
2188 	ret = kvm_skip_emulated_instruction(vcpu);
2189 	svm_set_gif(to_svm(vcpu), false);
2190 	return ret;
2191 }
2192 
2193 static int invlpga_interception(struct kvm_vcpu *vcpu)
2194 {
2195 	gva_t gva = kvm_rax_read(vcpu);
2196 	u32 asid = kvm_rcx_read(vcpu);
2197 
2198 	/* FIXME: Handle an address size prefix. */
2199 	if (!is_long_mode(vcpu))
2200 		gva = (u32)gva;
2201 
2202 	trace_kvm_invlpga(to_svm(vcpu)->vmcb->save.rip, asid, gva);
2203 
2204 	/* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
2205 	kvm_mmu_invlpg(vcpu, gva);
2206 
2207 	return kvm_skip_emulated_instruction(vcpu);
2208 }
2209 
2210 static int skinit_interception(struct kvm_vcpu *vcpu)
2211 {
2212 	trace_kvm_skinit(to_svm(vcpu)->vmcb->save.rip, kvm_rax_read(vcpu));
2213 
2214 	kvm_queue_exception(vcpu, UD_VECTOR);
2215 	return 1;
2216 }
2217 
2218 static int task_switch_interception(struct kvm_vcpu *vcpu)
2219 {
2220 	struct vcpu_svm *svm = to_svm(vcpu);
2221 	u16 tss_selector;
2222 	int reason;
2223 	int int_type = svm->vmcb->control.exit_int_info &
2224 		SVM_EXITINTINFO_TYPE_MASK;
2225 	int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
2226 	uint32_t type =
2227 		svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
2228 	uint32_t idt_v =
2229 		svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
2230 	bool has_error_code = false;
2231 	u32 error_code = 0;
2232 
2233 	tss_selector = (u16)svm->vmcb->control.exit_info_1;
2234 
2235 	if (svm->vmcb->control.exit_info_2 &
2236 	    (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
2237 		reason = TASK_SWITCH_IRET;
2238 	else if (svm->vmcb->control.exit_info_2 &
2239 		 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
2240 		reason = TASK_SWITCH_JMP;
2241 	else if (idt_v)
2242 		reason = TASK_SWITCH_GATE;
2243 	else
2244 		reason = TASK_SWITCH_CALL;
2245 
2246 	if (reason == TASK_SWITCH_GATE) {
2247 		switch (type) {
2248 		case SVM_EXITINTINFO_TYPE_NMI:
2249 			vcpu->arch.nmi_injected = false;
2250 			break;
2251 		case SVM_EXITINTINFO_TYPE_EXEPT:
2252 			if (svm->vmcb->control.exit_info_2 &
2253 			    (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE)) {
2254 				has_error_code = true;
2255 				error_code =
2256 					(u32)svm->vmcb->control.exit_info_2;
2257 			}
2258 			kvm_clear_exception_queue(vcpu);
2259 			break;
2260 		case SVM_EXITINTINFO_TYPE_INTR:
2261 			kvm_clear_interrupt_queue(vcpu);
2262 			break;
2263 		default:
2264 			break;
2265 		}
2266 	}
2267 
2268 	if (reason != TASK_SWITCH_GATE ||
2269 	    int_type == SVM_EXITINTINFO_TYPE_SOFT ||
2270 	    (int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
2271 	     (int_vec == OF_VECTOR || int_vec == BP_VECTOR))) {
2272 		if (!skip_emulated_instruction(vcpu))
2273 			return 0;
2274 	}
2275 
2276 	if (int_type != SVM_EXITINTINFO_TYPE_SOFT)
2277 		int_vec = -1;
2278 
2279 	return kvm_task_switch(vcpu, tss_selector, int_vec, reason,
2280 			       has_error_code, error_code);
2281 }
2282 
2283 static int iret_interception(struct kvm_vcpu *vcpu)
2284 {
2285 	struct vcpu_svm *svm = to_svm(vcpu);
2286 
2287 	++vcpu->stat.nmi_window_exits;
2288 	vcpu->arch.hflags |= HF_IRET_MASK;
2289 	if (!sev_es_guest(vcpu->kvm)) {
2290 		svm_clr_intercept(svm, INTERCEPT_IRET);
2291 		svm->nmi_iret_rip = kvm_rip_read(vcpu);
2292 	}
2293 	kvm_make_request(KVM_REQ_EVENT, vcpu);
2294 	return 1;
2295 }
2296 
2297 static int invlpg_interception(struct kvm_vcpu *vcpu)
2298 {
2299 	if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
2300 		return kvm_emulate_instruction(vcpu, 0);
2301 
2302 	kvm_mmu_invlpg(vcpu, to_svm(vcpu)->vmcb->control.exit_info_1);
2303 	return kvm_skip_emulated_instruction(vcpu);
2304 }
2305 
2306 static int emulate_on_interception(struct kvm_vcpu *vcpu)
2307 {
2308 	return kvm_emulate_instruction(vcpu, 0);
2309 }
2310 
2311 static int rsm_interception(struct kvm_vcpu *vcpu)
2312 {
2313 	return kvm_emulate_instruction_from_buffer(vcpu, rsm_ins_bytes, 2);
2314 }
2315 
2316 static bool check_selective_cr0_intercepted(struct kvm_vcpu *vcpu,
2317 					    unsigned long val)
2318 {
2319 	struct vcpu_svm *svm = to_svm(vcpu);
2320 	unsigned long cr0 = vcpu->arch.cr0;
2321 	bool ret = false;
2322 
2323 	if (!is_guest_mode(vcpu) ||
2324 	    (!(vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_SELECTIVE_CR0))))
2325 		return false;
2326 
2327 	cr0 &= ~SVM_CR0_SELECTIVE_MASK;
2328 	val &= ~SVM_CR0_SELECTIVE_MASK;
2329 
2330 	if (cr0 ^ val) {
2331 		svm->vmcb->control.exit_code = SVM_EXIT_CR0_SEL_WRITE;
2332 		ret = (nested_svm_exit_handled(svm) == NESTED_EXIT_DONE);
2333 	}
2334 
2335 	return ret;
2336 }
2337 
2338 #define CR_VALID (1ULL << 63)
2339 
2340 static int cr_interception(struct kvm_vcpu *vcpu)
2341 {
2342 	struct vcpu_svm *svm = to_svm(vcpu);
2343 	int reg, cr;
2344 	unsigned long val;
2345 	int err;
2346 
2347 	if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
2348 		return emulate_on_interception(vcpu);
2349 
2350 	if (unlikely((svm->vmcb->control.exit_info_1 & CR_VALID) == 0))
2351 		return emulate_on_interception(vcpu);
2352 
2353 	reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
2354 	if (svm->vmcb->control.exit_code == SVM_EXIT_CR0_SEL_WRITE)
2355 		cr = SVM_EXIT_WRITE_CR0 - SVM_EXIT_READ_CR0;
2356 	else
2357 		cr = svm->vmcb->control.exit_code - SVM_EXIT_READ_CR0;
2358 
2359 	err = 0;
2360 	if (cr >= 16) { /* mov to cr */
2361 		cr -= 16;
2362 		val = kvm_register_read(vcpu, reg);
2363 		trace_kvm_cr_write(cr, val);
2364 		switch (cr) {
2365 		case 0:
2366 			if (!check_selective_cr0_intercepted(vcpu, val))
2367 				err = kvm_set_cr0(vcpu, val);
2368 			else
2369 				return 1;
2370 
2371 			break;
2372 		case 3:
2373 			err = kvm_set_cr3(vcpu, val);
2374 			break;
2375 		case 4:
2376 			err = kvm_set_cr4(vcpu, val);
2377 			break;
2378 		case 8:
2379 			err = kvm_set_cr8(vcpu, val);
2380 			break;
2381 		default:
2382 			WARN(1, "unhandled write to CR%d", cr);
2383 			kvm_queue_exception(vcpu, UD_VECTOR);
2384 			return 1;
2385 		}
2386 	} else { /* mov from cr */
2387 		switch (cr) {
2388 		case 0:
2389 			val = kvm_read_cr0(vcpu);
2390 			break;
2391 		case 2:
2392 			val = vcpu->arch.cr2;
2393 			break;
2394 		case 3:
2395 			val = kvm_read_cr3(vcpu);
2396 			break;
2397 		case 4:
2398 			val = kvm_read_cr4(vcpu);
2399 			break;
2400 		case 8:
2401 			val = kvm_get_cr8(vcpu);
2402 			break;
2403 		default:
2404 			WARN(1, "unhandled read from CR%d", cr);
2405 			kvm_queue_exception(vcpu, UD_VECTOR);
2406 			return 1;
2407 		}
2408 		kvm_register_write(vcpu, reg, val);
2409 		trace_kvm_cr_read(cr, val);
2410 	}
2411 	return kvm_complete_insn_gp(vcpu, err);
2412 }
2413 
2414 static int cr_trap(struct kvm_vcpu *vcpu)
2415 {
2416 	struct vcpu_svm *svm = to_svm(vcpu);
2417 	unsigned long old_value, new_value;
2418 	unsigned int cr;
2419 	int ret = 0;
2420 
2421 	new_value = (unsigned long)svm->vmcb->control.exit_info_1;
2422 
2423 	cr = svm->vmcb->control.exit_code - SVM_EXIT_CR0_WRITE_TRAP;
2424 	switch (cr) {
2425 	case 0:
2426 		old_value = kvm_read_cr0(vcpu);
2427 		svm_set_cr0(vcpu, new_value);
2428 
2429 		kvm_post_set_cr0(vcpu, old_value, new_value);
2430 		break;
2431 	case 4:
2432 		old_value = kvm_read_cr4(vcpu);
2433 		svm_set_cr4(vcpu, new_value);
2434 
2435 		kvm_post_set_cr4(vcpu, old_value, new_value);
2436 		break;
2437 	case 8:
2438 		ret = kvm_set_cr8(vcpu, new_value);
2439 		break;
2440 	default:
2441 		WARN(1, "unhandled CR%d write trap", cr);
2442 		kvm_queue_exception(vcpu, UD_VECTOR);
2443 		return 1;
2444 	}
2445 
2446 	return kvm_complete_insn_gp(vcpu, ret);
2447 }
2448 
2449 static int dr_interception(struct kvm_vcpu *vcpu)
2450 {
2451 	struct vcpu_svm *svm = to_svm(vcpu);
2452 	int reg, dr;
2453 	unsigned long val;
2454 	int err = 0;
2455 
2456 	if (vcpu->guest_debug == 0) {
2457 		/*
2458 		 * No more DR vmexits; force a reload of the debug registers
2459 		 * and reenter on this instruction.  The next vmexit will
2460 		 * retrieve the full state of the debug registers.
2461 		 */
2462 		clr_dr_intercepts(svm);
2463 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
2464 		return 1;
2465 	}
2466 
2467 	if (!boot_cpu_has(X86_FEATURE_DECODEASSISTS))
2468 		return emulate_on_interception(vcpu);
2469 
2470 	reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
2471 	dr = svm->vmcb->control.exit_code - SVM_EXIT_READ_DR0;
2472 	if (dr >= 16) { /* mov to DRn  */
2473 		dr -= 16;
2474 		val = kvm_register_read(vcpu, reg);
2475 		err = kvm_set_dr(vcpu, dr, val);
2476 	} else {
2477 		kvm_get_dr(vcpu, dr, &val);
2478 		kvm_register_write(vcpu, reg, val);
2479 	}
2480 
2481 	return kvm_complete_insn_gp(vcpu, err);
2482 }
2483 
2484 static int cr8_write_interception(struct kvm_vcpu *vcpu)
2485 {
2486 	int r;
2487 
2488 	u8 cr8_prev = kvm_get_cr8(vcpu);
2489 	/* instruction emulation calls kvm_set_cr8() */
2490 	r = cr_interception(vcpu);
2491 	if (lapic_in_kernel(vcpu))
2492 		return r;
2493 	if (cr8_prev <= kvm_get_cr8(vcpu))
2494 		return r;
2495 	vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
2496 	return 0;
2497 }
2498 
2499 static int efer_trap(struct kvm_vcpu *vcpu)
2500 {
2501 	struct msr_data msr_info;
2502 	int ret;
2503 
2504 	/*
2505 	 * Clear the EFER_SVME bit from EFER. The SVM code always sets this
2506 	 * bit in svm_set_efer(), but __kvm_valid_efer() checks it against
2507 	 * whether the guest has X86_FEATURE_SVM - this avoids a failure if
2508 	 * the guest doesn't have X86_FEATURE_SVM.
2509 	 */
2510 	msr_info.host_initiated = false;
2511 	msr_info.index = MSR_EFER;
2512 	msr_info.data = to_svm(vcpu)->vmcb->control.exit_info_1 & ~EFER_SVME;
2513 	ret = kvm_set_msr_common(vcpu, &msr_info);
2514 
2515 	return kvm_complete_insn_gp(vcpu, ret);
2516 }
2517 
2518 static int svm_get_msr_feature(struct kvm_msr_entry *msr)
2519 {
2520 	msr->data = 0;
2521 
2522 	switch (msr->index) {
2523 	case MSR_F10H_DECFG:
2524 		if (boot_cpu_has(X86_FEATURE_LFENCE_RDTSC))
2525 			msr->data |= MSR_F10H_DECFG_LFENCE_SERIALIZE;
2526 		break;
2527 	case MSR_IA32_PERF_CAPABILITIES:
2528 		return 0;
2529 	default:
2530 		return KVM_MSR_RET_INVALID;
2531 	}
2532 
2533 	return 0;
2534 }
2535 
2536 static int svm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2537 {
2538 	struct vcpu_svm *svm = to_svm(vcpu);
2539 
2540 	switch (msr_info->index) {
2541 	case MSR_AMD64_TSC_RATIO:
2542 		if (!msr_info->host_initiated && !svm->tsc_scaling_enabled)
2543 			return 1;
2544 		msr_info->data = svm->tsc_ratio_msr;
2545 		break;
2546 	case MSR_STAR:
2547 		msr_info->data = svm->vmcb01.ptr->save.star;
2548 		break;
2549 #ifdef CONFIG_X86_64
2550 	case MSR_LSTAR:
2551 		msr_info->data = svm->vmcb01.ptr->save.lstar;
2552 		break;
2553 	case MSR_CSTAR:
2554 		msr_info->data = svm->vmcb01.ptr->save.cstar;
2555 		break;
2556 	case MSR_KERNEL_GS_BASE:
2557 		msr_info->data = svm->vmcb01.ptr->save.kernel_gs_base;
2558 		break;
2559 	case MSR_SYSCALL_MASK:
2560 		msr_info->data = svm->vmcb01.ptr->save.sfmask;
2561 		break;
2562 #endif
2563 	case MSR_IA32_SYSENTER_CS:
2564 		msr_info->data = svm->vmcb01.ptr->save.sysenter_cs;
2565 		break;
2566 	case MSR_IA32_SYSENTER_EIP:
2567 		msr_info->data = (u32)svm->vmcb01.ptr->save.sysenter_eip;
2568 		if (guest_cpuid_is_intel(vcpu))
2569 			msr_info->data |= (u64)svm->sysenter_eip_hi << 32;
2570 		break;
2571 	case MSR_IA32_SYSENTER_ESP:
2572 		msr_info->data = svm->vmcb01.ptr->save.sysenter_esp;
2573 		if (guest_cpuid_is_intel(vcpu))
2574 			msr_info->data |= (u64)svm->sysenter_esp_hi << 32;
2575 		break;
2576 	case MSR_TSC_AUX:
2577 		msr_info->data = svm->tsc_aux;
2578 		break;
2579 	/*
2580 	 * Nobody will change the following 5 values in the VMCB so we can
2581 	 * safely return them on rdmsr. They will always be 0 until LBRV is
2582 	 * implemented.
2583 	 */
2584 	case MSR_IA32_DEBUGCTLMSR:
2585 		msr_info->data = svm->vmcb->save.dbgctl;
2586 		break;
2587 	case MSR_IA32_LASTBRANCHFROMIP:
2588 		msr_info->data = svm->vmcb->save.br_from;
2589 		break;
2590 	case MSR_IA32_LASTBRANCHTOIP:
2591 		msr_info->data = svm->vmcb->save.br_to;
2592 		break;
2593 	case MSR_IA32_LASTINTFROMIP:
2594 		msr_info->data = svm->vmcb->save.last_excp_from;
2595 		break;
2596 	case MSR_IA32_LASTINTTOIP:
2597 		msr_info->data = svm->vmcb->save.last_excp_to;
2598 		break;
2599 	case MSR_VM_HSAVE_PA:
2600 		msr_info->data = svm->nested.hsave_msr;
2601 		break;
2602 	case MSR_VM_CR:
2603 		msr_info->data = svm->nested.vm_cr_msr;
2604 		break;
2605 	case MSR_IA32_SPEC_CTRL:
2606 		if (!msr_info->host_initiated &&
2607 		    !guest_has_spec_ctrl_msr(vcpu))
2608 			return 1;
2609 
2610 		if (boot_cpu_has(X86_FEATURE_V_SPEC_CTRL))
2611 			msr_info->data = svm->vmcb->save.spec_ctrl;
2612 		else
2613 			msr_info->data = svm->spec_ctrl;
2614 		break;
2615 	case MSR_AMD64_VIRT_SPEC_CTRL:
2616 		if (!msr_info->host_initiated &&
2617 		    !guest_cpuid_has(vcpu, X86_FEATURE_VIRT_SSBD))
2618 			return 1;
2619 
2620 		msr_info->data = svm->virt_spec_ctrl;
2621 		break;
2622 	case MSR_F15H_IC_CFG: {
2623 
2624 		int family, model;
2625 
2626 		family = guest_cpuid_family(vcpu);
2627 		model  = guest_cpuid_model(vcpu);
2628 
2629 		if (family < 0 || model < 0)
2630 			return kvm_get_msr_common(vcpu, msr_info);
2631 
2632 		msr_info->data = 0;
2633 
2634 		if (family == 0x15 &&
2635 		    (model >= 0x2 && model < 0x20))
2636 			msr_info->data = 0x1E;
2637 		}
2638 		break;
2639 	case MSR_F10H_DECFG:
2640 		msr_info->data = svm->msr_decfg;
2641 		break;
2642 	default:
2643 		return kvm_get_msr_common(vcpu, msr_info);
2644 	}
2645 	return 0;
2646 }
2647 
2648 static int svm_complete_emulated_msr(struct kvm_vcpu *vcpu, int err)
2649 {
2650 	struct vcpu_svm *svm = to_svm(vcpu);
2651 	if (!err || !sev_es_guest(vcpu->kvm) || WARN_ON_ONCE(!svm->sev_es.ghcb))
2652 		return kvm_complete_insn_gp(vcpu, err);
2653 
2654 	ghcb_set_sw_exit_info_1(svm->sev_es.ghcb, 1);
2655 	ghcb_set_sw_exit_info_2(svm->sev_es.ghcb,
2656 				X86_TRAP_GP |
2657 				SVM_EVTINJ_TYPE_EXEPT |
2658 				SVM_EVTINJ_VALID);
2659 	return 1;
2660 }
2661 
2662 static int svm_set_vm_cr(struct kvm_vcpu *vcpu, u64 data)
2663 {
2664 	struct vcpu_svm *svm = to_svm(vcpu);
2665 	int svm_dis, chg_mask;
2666 
2667 	if (data & ~SVM_VM_CR_VALID_MASK)
2668 		return 1;
2669 
2670 	chg_mask = SVM_VM_CR_VALID_MASK;
2671 
2672 	if (svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK)
2673 		chg_mask &= ~(SVM_VM_CR_SVM_LOCK_MASK | SVM_VM_CR_SVM_DIS_MASK);
2674 
2675 	svm->nested.vm_cr_msr &= ~chg_mask;
2676 	svm->nested.vm_cr_msr |= (data & chg_mask);
2677 
2678 	svm_dis = svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK;
2679 
2680 	/* check for svm_disable while efer.svme is set */
2681 	if (svm_dis && (vcpu->arch.efer & EFER_SVME))
2682 		return 1;
2683 
2684 	return 0;
2685 }
2686 
2687 static int svm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2688 {
2689 	struct vcpu_svm *svm = to_svm(vcpu);
2690 	int r;
2691 
2692 	u32 ecx = msr->index;
2693 	u64 data = msr->data;
2694 	switch (ecx) {
2695 	case MSR_AMD64_TSC_RATIO:
2696 		if (!msr->host_initiated && !svm->tsc_scaling_enabled)
2697 			return 1;
2698 
2699 		if (data & TSC_RATIO_RSVD)
2700 			return 1;
2701 
2702 		svm->tsc_ratio_msr = data;
2703 
2704 		if (svm->tsc_scaling_enabled && is_guest_mode(vcpu))
2705 			nested_svm_update_tsc_ratio_msr(vcpu);
2706 
2707 		break;
2708 	case MSR_IA32_CR_PAT:
2709 		if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
2710 			return 1;
2711 		vcpu->arch.pat = data;
2712 		svm->vmcb01.ptr->save.g_pat = data;
2713 		if (is_guest_mode(vcpu))
2714 			nested_vmcb02_compute_g_pat(svm);
2715 		vmcb_mark_dirty(svm->vmcb, VMCB_NPT);
2716 		break;
2717 	case MSR_IA32_SPEC_CTRL:
2718 		if (!msr->host_initiated &&
2719 		    !guest_has_spec_ctrl_msr(vcpu))
2720 			return 1;
2721 
2722 		if (kvm_spec_ctrl_test_value(data))
2723 			return 1;
2724 
2725 		if (boot_cpu_has(X86_FEATURE_V_SPEC_CTRL))
2726 			svm->vmcb->save.spec_ctrl = data;
2727 		else
2728 			svm->spec_ctrl = data;
2729 		if (!data)
2730 			break;
2731 
2732 		/*
2733 		 * For non-nested:
2734 		 * When it's written (to non-zero) for the first time, pass
2735 		 * it through.
2736 		 *
2737 		 * For nested:
2738 		 * The handling of the MSR bitmap for L2 guests is done in
2739 		 * nested_svm_vmrun_msrpm.
2740 		 * We update the L1 MSR bit as well since it will end up
2741 		 * touching the MSR anyway now.
2742 		 */
2743 		set_msr_interception(vcpu, svm->msrpm, MSR_IA32_SPEC_CTRL, 1, 1);
2744 		break;
2745 	case MSR_IA32_PRED_CMD:
2746 		if (!msr->host_initiated &&
2747 		    !guest_has_pred_cmd_msr(vcpu))
2748 			return 1;
2749 
2750 		if (data & ~PRED_CMD_IBPB)
2751 			return 1;
2752 		if (!boot_cpu_has(X86_FEATURE_IBPB))
2753 			return 1;
2754 		if (!data)
2755 			break;
2756 
2757 		wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
2758 		set_msr_interception(vcpu, svm->msrpm, MSR_IA32_PRED_CMD, 0, 1);
2759 		break;
2760 	case MSR_AMD64_VIRT_SPEC_CTRL:
2761 		if (!msr->host_initiated &&
2762 		    !guest_cpuid_has(vcpu, X86_FEATURE_VIRT_SSBD))
2763 			return 1;
2764 
2765 		if (data & ~SPEC_CTRL_SSBD)
2766 			return 1;
2767 
2768 		svm->virt_spec_ctrl = data;
2769 		break;
2770 	case MSR_STAR:
2771 		svm->vmcb01.ptr->save.star = data;
2772 		break;
2773 #ifdef CONFIG_X86_64
2774 	case MSR_LSTAR:
2775 		svm->vmcb01.ptr->save.lstar = data;
2776 		break;
2777 	case MSR_CSTAR:
2778 		svm->vmcb01.ptr->save.cstar = data;
2779 		break;
2780 	case MSR_KERNEL_GS_BASE:
2781 		svm->vmcb01.ptr->save.kernel_gs_base = data;
2782 		break;
2783 	case MSR_SYSCALL_MASK:
2784 		svm->vmcb01.ptr->save.sfmask = data;
2785 		break;
2786 #endif
2787 	case MSR_IA32_SYSENTER_CS:
2788 		svm->vmcb01.ptr->save.sysenter_cs = data;
2789 		break;
2790 	case MSR_IA32_SYSENTER_EIP:
2791 		svm->vmcb01.ptr->save.sysenter_eip = (u32)data;
2792 		/*
2793 		 * We only intercept the MSR_IA32_SYSENTER_{EIP|ESP} msrs
2794 		 * when we spoof an Intel vendor ID (for cross vendor migration).
2795 		 * In this case we use this intercept to track the high
2796 		 * 32 bit part of these msrs to support Intel's
2797 		 * implementation of SYSENTER/SYSEXIT.
2798 		 */
2799 		svm->sysenter_eip_hi = guest_cpuid_is_intel(vcpu) ? (data >> 32) : 0;
2800 		break;
2801 	case MSR_IA32_SYSENTER_ESP:
2802 		svm->vmcb01.ptr->save.sysenter_esp = (u32)data;
2803 		svm->sysenter_esp_hi = guest_cpuid_is_intel(vcpu) ? (data >> 32) : 0;
2804 		break;
2805 	case MSR_TSC_AUX:
2806 		/*
2807 		 * TSC_AUX is usually changed only during boot and never read
2808 		 * directly.  Intercept TSC_AUX instead of exposing it to the
2809 		 * guest via direct_access_msrs, and switch it via user return.
2810 		 */
2811 		preempt_disable();
2812 		r = kvm_set_user_return_msr(tsc_aux_uret_slot, data, -1ull);
2813 		preempt_enable();
2814 		if (r)
2815 			return 1;
2816 
2817 		svm->tsc_aux = data;
2818 		break;
2819 	case MSR_IA32_DEBUGCTLMSR:
2820 		if (!lbrv) {
2821 			vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
2822 				    __func__, data);
2823 			break;
2824 		}
2825 		if (data & DEBUGCTL_RESERVED_BITS)
2826 			return 1;
2827 
2828 		svm->vmcb->save.dbgctl = data;
2829 		vmcb_mark_dirty(svm->vmcb, VMCB_LBR);
2830 		if (data & (1ULL<<0))
2831 			svm_enable_lbrv(vcpu);
2832 		else
2833 			svm_disable_lbrv(vcpu);
2834 		break;
2835 	case MSR_VM_HSAVE_PA:
2836 		/*
2837 		 * Old kernels did not validate the value written to
2838 		 * MSR_VM_HSAVE_PA.  Allow KVM_SET_MSR to set an invalid
2839 		 * value to allow live migrating buggy or malicious guests
2840 		 * originating from those kernels.
2841 		 */
2842 		if (!msr->host_initiated && !page_address_valid(vcpu, data))
2843 			return 1;
2844 
2845 		svm->nested.hsave_msr = data & PAGE_MASK;
2846 		break;
2847 	case MSR_VM_CR:
2848 		return svm_set_vm_cr(vcpu, data);
2849 	case MSR_VM_IGNNE:
2850 		vcpu_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data);
2851 		break;
2852 	case MSR_F10H_DECFG: {
2853 		struct kvm_msr_entry msr_entry;
2854 
2855 		msr_entry.index = msr->index;
2856 		if (svm_get_msr_feature(&msr_entry))
2857 			return 1;
2858 
2859 		/* Check the supported bits */
2860 		if (data & ~msr_entry.data)
2861 			return 1;
2862 
2863 		/* Don't allow the guest to change a bit, #GP */
2864 		if (!msr->host_initiated && (data ^ msr_entry.data))
2865 			return 1;
2866 
2867 		svm->msr_decfg = data;
2868 		break;
2869 	}
2870 	default:
2871 		return kvm_set_msr_common(vcpu, msr);
2872 	}
2873 	return 0;
2874 }
2875 
2876 static int msr_interception(struct kvm_vcpu *vcpu)
2877 {
2878 	if (to_svm(vcpu)->vmcb->control.exit_info_1)
2879 		return kvm_emulate_wrmsr(vcpu);
2880 	else
2881 		return kvm_emulate_rdmsr(vcpu);
2882 }
2883 
2884 static int interrupt_window_interception(struct kvm_vcpu *vcpu)
2885 {
2886 	kvm_make_request(KVM_REQ_EVENT, vcpu);
2887 	svm_clear_vintr(to_svm(vcpu));
2888 
2889 	/*
2890 	 * For AVIC, the only reason to end up here is ExtINTs.
2891 	 * In this case AVIC was temporarily disabled for
2892 	 * requesting the IRQ window and we have to re-enable it.
2893 	 */
2894 	kvm_request_apicv_update(vcpu->kvm, true, APICV_INHIBIT_REASON_IRQWIN);
2895 
2896 	++vcpu->stat.irq_window_exits;
2897 	return 1;
2898 }
2899 
2900 static int pause_interception(struct kvm_vcpu *vcpu)
2901 {
2902 	bool in_kernel;
2903 
2904 	/*
2905 	 * CPL is not made available for an SEV-ES guest, therefore
2906 	 * vcpu->arch.preempted_in_kernel can never be true.  Just
2907 	 * set in_kernel to false as well.
2908 	 */
2909 	in_kernel = !sev_es_guest(vcpu->kvm) && svm_get_cpl(vcpu) == 0;
2910 
2911 	if (!kvm_pause_in_guest(vcpu->kvm))
2912 		grow_ple_window(vcpu);
2913 
2914 	kvm_vcpu_on_spin(vcpu, in_kernel);
2915 	return kvm_skip_emulated_instruction(vcpu);
2916 }
2917 
2918 static int invpcid_interception(struct kvm_vcpu *vcpu)
2919 {
2920 	struct vcpu_svm *svm = to_svm(vcpu);
2921 	unsigned long type;
2922 	gva_t gva;
2923 
2924 	if (!guest_cpuid_has(vcpu, X86_FEATURE_INVPCID)) {
2925 		kvm_queue_exception(vcpu, UD_VECTOR);
2926 		return 1;
2927 	}
2928 
2929 	/*
2930 	 * For an INVPCID intercept:
2931 	 * EXITINFO1 provides the linear address of the memory operand.
2932 	 * EXITINFO2 provides the contents of the register operand.
2933 	 */
2934 	type = svm->vmcb->control.exit_info_2;
2935 	gva = svm->vmcb->control.exit_info_1;
2936 
2937 	return kvm_handle_invpcid(vcpu, type, gva);
2938 }
2939 
2940 static int (*const svm_exit_handlers[])(struct kvm_vcpu *vcpu) = {
2941 	[SVM_EXIT_READ_CR0]			= cr_interception,
2942 	[SVM_EXIT_READ_CR3]			= cr_interception,
2943 	[SVM_EXIT_READ_CR4]			= cr_interception,
2944 	[SVM_EXIT_READ_CR8]			= cr_interception,
2945 	[SVM_EXIT_CR0_SEL_WRITE]		= cr_interception,
2946 	[SVM_EXIT_WRITE_CR0]			= cr_interception,
2947 	[SVM_EXIT_WRITE_CR3]			= cr_interception,
2948 	[SVM_EXIT_WRITE_CR4]			= cr_interception,
2949 	[SVM_EXIT_WRITE_CR8]			= cr8_write_interception,
2950 	[SVM_EXIT_READ_DR0]			= dr_interception,
2951 	[SVM_EXIT_READ_DR1]			= dr_interception,
2952 	[SVM_EXIT_READ_DR2]			= dr_interception,
2953 	[SVM_EXIT_READ_DR3]			= dr_interception,
2954 	[SVM_EXIT_READ_DR4]			= dr_interception,
2955 	[SVM_EXIT_READ_DR5]			= dr_interception,
2956 	[SVM_EXIT_READ_DR6]			= dr_interception,
2957 	[SVM_EXIT_READ_DR7]			= dr_interception,
2958 	[SVM_EXIT_WRITE_DR0]			= dr_interception,
2959 	[SVM_EXIT_WRITE_DR1]			= dr_interception,
2960 	[SVM_EXIT_WRITE_DR2]			= dr_interception,
2961 	[SVM_EXIT_WRITE_DR3]			= dr_interception,
2962 	[SVM_EXIT_WRITE_DR4]			= dr_interception,
2963 	[SVM_EXIT_WRITE_DR5]			= dr_interception,
2964 	[SVM_EXIT_WRITE_DR6]			= dr_interception,
2965 	[SVM_EXIT_WRITE_DR7]			= dr_interception,
2966 	[SVM_EXIT_EXCP_BASE + DB_VECTOR]	= db_interception,
2967 	[SVM_EXIT_EXCP_BASE + BP_VECTOR]	= bp_interception,
2968 	[SVM_EXIT_EXCP_BASE + UD_VECTOR]	= ud_interception,
2969 	[SVM_EXIT_EXCP_BASE + PF_VECTOR]	= pf_interception,
2970 	[SVM_EXIT_EXCP_BASE + MC_VECTOR]	= mc_interception,
2971 	[SVM_EXIT_EXCP_BASE + AC_VECTOR]	= ac_interception,
2972 	[SVM_EXIT_EXCP_BASE + GP_VECTOR]	= gp_interception,
2973 	[SVM_EXIT_INTR]				= intr_interception,
2974 	[SVM_EXIT_NMI]				= nmi_interception,
2975 	[SVM_EXIT_SMI]				= smi_interception,
2976 	[SVM_EXIT_VINTR]			= interrupt_window_interception,
2977 	[SVM_EXIT_RDPMC]			= kvm_emulate_rdpmc,
2978 	[SVM_EXIT_CPUID]			= kvm_emulate_cpuid,
2979 	[SVM_EXIT_IRET]                         = iret_interception,
2980 	[SVM_EXIT_INVD]                         = kvm_emulate_invd,
2981 	[SVM_EXIT_PAUSE]			= pause_interception,
2982 	[SVM_EXIT_HLT]				= kvm_emulate_halt,
2983 	[SVM_EXIT_INVLPG]			= invlpg_interception,
2984 	[SVM_EXIT_INVLPGA]			= invlpga_interception,
2985 	[SVM_EXIT_IOIO]				= io_interception,
2986 	[SVM_EXIT_MSR]				= msr_interception,
2987 	[SVM_EXIT_TASK_SWITCH]			= task_switch_interception,
2988 	[SVM_EXIT_SHUTDOWN]			= shutdown_interception,
2989 	[SVM_EXIT_VMRUN]			= vmrun_interception,
2990 	[SVM_EXIT_VMMCALL]			= kvm_emulate_hypercall,
2991 	[SVM_EXIT_VMLOAD]			= vmload_interception,
2992 	[SVM_EXIT_VMSAVE]			= vmsave_interception,
2993 	[SVM_EXIT_STGI]				= stgi_interception,
2994 	[SVM_EXIT_CLGI]				= clgi_interception,
2995 	[SVM_EXIT_SKINIT]			= skinit_interception,
2996 	[SVM_EXIT_RDTSCP]			= kvm_handle_invalid_op,
2997 	[SVM_EXIT_WBINVD]                       = kvm_emulate_wbinvd,
2998 	[SVM_EXIT_MONITOR]			= kvm_emulate_monitor,
2999 	[SVM_EXIT_MWAIT]			= kvm_emulate_mwait,
3000 	[SVM_EXIT_XSETBV]			= kvm_emulate_xsetbv,
3001 	[SVM_EXIT_RDPRU]			= kvm_handle_invalid_op,
3002 	[SVM_EXIT_EFER_WRITE_TRAP]		= efer_trap,
3003 	[SVM_EXIT_CR0_WRITE_TRAP]		= cr_trap,
3004 	[SVM_EXIT_CR4_WRITE_TRAP]		= cr_trap,
3005 	[SVM_EXIT_CR8_WRITE_TRAP]		= cr_trap,
3006 	[SVM_EXIT_INVPCID]                      = invpcid_interception,
3007 	[SVM_EXIT_NPF]				= npf_interception,
3008 	[SVM_EXIT_RSM]                          = rsm_interception,
3009 	[SVM_EXIT_AVIC_INCOMPLETE_IPI]		= avic_incomplete_ipi_interception,
3010 	[SVM_EXIT_AVIC_UNACCELERATED_ACCESS]	= avic_unaccelerated_access_interception,
3011 	[SVM_EXIT_VMGEXIT]			= sev_handle_vmgexit,
3012 };
3013 
3014 static void dump_vmcb(struct kvm_vcpu *vcpu)
3015 {
3016 	struct vcpu_svm *svm = to_svm(vcpu);
3017 	struct vmcb_control_area *control = &svm->vmcb->control;
3018 	struct vmcb_save_area *save = &svm->vmcb->save;
3019 	struct vmcb_save_area *save01 = &svm->vmcb01.ptr->save;
3020 
3021 	if (!dump_invalid_vmcb) {
3022 		pr_warn_ratelimited("set kvm_amd.dump_invalid_vmcb=1 to dump internal KVM state.\n");
3023 		return;
3024 	}
3025 
3026 	pr_err("VMCB %p, last attempted VMRUN on CPU %d\n",
3027 	       svm->current_vmcb->ptr, vcpu->arch.last_vmentry_cpu);
3028 	pr_err("VMCB Control Area:\n");
3029 	pr_err("%-20s%04x\n", "cr_read:", control->intercepts[INTERCEPT_CR] & 0xffff);
3030 	pr_err("%-20s%04x\n", "cr_write:", control->intercepts[INTERCEPT_CR] >> 16);
3031 	pr_err("%-20s%04x\n", "dr_read:", control->intercepts[INTERCEPT_DR] & 0xffff);
3032 	pr_err("%-20s%04x\n", "dr_write:", control->intercepts[INTERCEPT_DR] >> 16);
3033 	pr_err("%-20s%08x\n", "exceptions:", control->intercepts[INTERCEPT_EXCEPTION]);
3034 	pr_err("%-20s%08x %08x\n", "intercepts:",
3035               control->intercepts[INTERCEPT_WORD3],
3036 	       control->intercepts[INTERCEPT_WORD4]);
3037 	pr_err("%-20s%d\n", "pause filter count:", control->pause_filter_count);
3038 	pr_err("%-20s%d\n", "pause filter threshold:",
3039 	       control->pause_filter_thresh);
3040 	pr_err("%-20s%016llx\n", "iopm_base_pa:", control->iopm_base_pa);
3041 	pr_err("%-20s%016llx\n", "msrpm_base_pa:", control->msrpm_base_pa);
3042 	pr_err("%-20s%016llx\n", "tsc_offset:", control->tsc_offset);
3043 	pr_err("%-20s%d\n", "asid:", control->asid);
3044 	pr_err("%-20s%d\n", "tlb_ctl:", control->tlb_ctl);
3045 	pr_err("%-20s%08x\n", "int_ctl:", control->int_ctl);
3046 	pr_err("%-20s%08x\n", "int_vector:", control->int_vector);
3047 	pr_err("%-20s%08x\n", "int_state:", control->int_state);
3048 	pr_err("%-20s%08x\n", "exit_code:", control->exit_code);
3049 	pr_err("%-20s%016llx\n", "exit_info1:", control->exit_info_1);
3050 	pr_err("%-20s%016llx\n", "exit_info2:", control->exit_info_2);
3051 	pr_err("%-20s%08x\n", "exit_int_info:", control->exit_int_info);
3052 	pr_err("%-20s%08x\n", "exit_int_info_err:", control->exit_int_info_err);
3053 	pr_err("%-20s%lld\n", "nested_ctl:", control->nested_ctl);
3054 	pr_err("%-20s%016llx\n", "nested_cr3:", control->nested_cr3);
3055 	pr_err("%-20s%016llx\n", "avic_vapic_bar:", control->avic_vapic_bar);
3056 	pr_err("%-20s%016llx\n", "ghcb:", control->ghcb_gpa);
3057 	pr_err("%-20s%08x\n", "event_inj:", control->event_inj);
3058 	pr_err("%-20s%08x\n", "event_inj_err:", control->event_inj_err);
3059 	pr_err("%-20s%lld\n", "virt_ext:", control->virt_ext);
3060 	pr_err("%-20s%016llx\n", "next_rip:", control->next_rip);
3061 	pr_err("%-20s%016llx\n", "avic_backing_page:", control->avic_backing_page);
3062 	pr_err("%-20s%016llx\n", "avic_logical_id:", control->avic_logical_id);
3063 	pr_err("%-20s%016llx\n", "avic_physical_id:", control->avic_physical_id);
3064 	pr_err("%-20s%016llx\n", "vmsa_pa:", control->vmsa_pa);
3065 	pr_err("VMCB State Save Area:\n");
3066 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3067 	       "es:",
3068 	       save->es.selector, save->es.attrib,
3069 	       save->es.limit, save->es.base);
3070 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3071 	       "cs:",
3072 	       save->cs.selector, save->cs.attrib,
3073 	       save->cs.limit, save->cs.base);
3074 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3075 	       "ss:",
3076 	       save->ss.selector, save->ss.attrib,
3077 	       save->ss.limit, save->ss.base);
3078 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3079 	       "ds:",
3080 	       save->ds.selector, save->ds.attrib,
3081 	       save->ds.limit, save->ds.base);
3082 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3083 	       "fs:",
3084 	       save01->fs.selector, save01->fs.attrib,
3085 	       save01->fs.limit, save01->fs.base);
3086 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3087 	       "gs:",
3088 	       save01->gs.selector, save01->gs.attrib,
3089 	       save01->gs.limit, save01->gs.base);
3090 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3091 	       "gdtr:",
3092 	       save->gdtr.selector, save->gdtr.attrib,
3093 	       save->gdtr.limit, save->gdtr.base);
3094 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3095 	       "ldtr:",
3096 	       save01->ldtr.selector, save01->ldtr.attrib,
3097 	       save01->ldtr.limit, save01->ldtr.base);
3098 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3099 	       "idtr:",
3100 	       save->idtr.selector, save->idtr.attrib,
3101 	       save->idtr.limit, save->idtr.base);
3102 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3103 	       "tr:",
3104 	       save01->tr.selector, save01->tr.attrib,
3105 	       save01->tr.limit, save01->tr.base);
3106 	pr_err("cpl:            %d                efer:         %016llx\n",
3107 		save->cpl, save->efer);
3108 	pr_err("%-15s %016llx %-13s %016llx\n",
3109 	       "cr0:", save->cr0, "cr2:", save->cr2);
3110 	pr_err("%-15s %016llx %-13s %016llx\n",
3111 	       "cr3:", save->cr3, "cr4:", save->cr4);
3112 	pr_err("%-15s %016llx %-13s %016llx\n",
3113 	       "dr6:", save->dr6, "dr7:", save->dr7);
3114 	pr_err("%-15s %016llx %-13s %016llx\n",
3115 	       "rip:", save->rip, "rflags:", save->rflags);
3116 	pr_err("%-15s %016llx %-13s %016llx\n",
3117 	       "rsp:", save->rsp, "rax:", save->rax);
3118 	pr_err("%-15s %016llx %-13s %016llx\n",
3119 	       "star:", save01->star, "lstar:", save01->lstar);
3120 	pr_err("%-15s %016llx %-13s %016llx\n",
3121 	       "cstar:", save01->cstar, "sfmask:", save01->sfmask);
3122 	pr_err("%-15s %016llx %-13s %016llx\n",
3123 	       "kernel_gs_base:", save01->kernel_gs_base,
3124 	       "sysenter_cs:", save01->sysenter_cs);
3125 	pr_err("%-15s %016llx %-13s %016llx\n",
3126 	       "sysenter_esp:", save01->sysenter_esp,
3127 	       "sysenter_eip:", save01->sysenter_eip);
3128 	pr_err("%-15s %016llx %-13s %016llx\n",
3129 	       "gpat:", save->g_pat, "dbgctl:", save->dbgctl);
3130 	pr_err("%-15s %016llx %-13s %016llx\n",
3131 	       "br_from:", save->br_from, "br_to:", save->br_to);
3132 	pr_err("%-15s %016llx %-13s %016llx\n",
3133 	       "excp_from:", save->last_excp_from,
3134 	       "excp_to:", save->last_excp_to);
3135 }
3136 
3137 static bool svm_check_exit_valid(struct kvm_vcpu *vcpu, u64 exit_code)
3138 {
3139 	return (exit_code < ARRAY_SIZE(svm_exit_handlers) &&
3140 		svm_exit_handlers[exit_code]);
3141 }
3142 
3143 static int svm_handle_invalid_exit(struct kvm_vcpu *vcpu, u64 exit_code)
3144 {
3145 	vcpu_unimpl(vcpu, "svm: unexpected exit reason 0x%llx\n", exit_code);
3146 	dump_vmcb(vcpu);
3147 	vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3148 	vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON;
3149 	vcpu->run->internal.ndata = 2;
3150 	vcpu->run->internal.data[0] = exit_code;
3151 	vcpu->run->internal.data[1] = vcpu->arch.last_vmentry_cpu;
3152 	return 0;
3153 }
3154 
3155 int svm_invoke_exit_handler(struct kvm_vcpu *vcpu, u64 exit_code)
3156 {
3157 	if (!svm_check_exit_valid(vcpu, exit_code))
3158 		return svm_handle_invalid_exit(vcpu, exit_code);
3159 
3160 #ifdef CONFIG_RETPOLINE
3161 	if (exit_code == SVM_EXIT_MSR)
3162 		return msr_interception(vcpu);
3163 	else if (exit_code == SVM_EXIT_VINTR)
3164 		return interrupt_window_interception(vcpu);
3165 	else if (exit_code == SVM_EXIT_INTR)
3166 		return intr_interception(vcpu);
3167 	else if (exit_code == SVM_EXIT_HLT)
3168 		return kvm_emulate_halt(vcpu);
3169 	else if (exit_code == SVM_EXIT_NPF)
3170 		return npf_interception(vcpu);
3171 #endif
3172 	return svm_exit_handlers[exit_code](vcpu);
3173 }
3174 
3175 static void svm_get_exit_info(struct kvm_vcpu *vcpu, u32 *reason,
3176 			      u64 *info1, u64 *info2,
3177 			      u32 *intr_info, u32 *error_code)
3178 {
3179 	struct vmcb_control_area *control = &to_svm(vcpu)->vmcb->control;
3180 
3181 	*reason = control->exit_code;
3182 	*info1 = control->exit_info_1;
3183 	*info2 = control->exit_info_2;
3184 	*intr_info = control->exit_int_info;
3185 	if ((*intr_info & SVM_EXITINTINFO_VALID) &&
3186 	    (*intr_info & SVM_EXITINTINFO_VALID_ERR))
3187 		*error_code = control->exit_int_info_err;
3188 	else
3189 		*error_code = 0;
3190 }
3191 
3192 static int handle_exit(struct kvm_vcpu *vcpu, fastpath_t exit_fastpath)
3193 {
3194 	struct vcpu_svm *svm = to_svm(vcpu);
3195 	struct kvm_run *kvm_run = vcpu->run;
3196 	u32 exit_code = svm->vmcb->control.exit_code;
3197 
3198 	trace_kvm_exit(vcpu, KVM_ISA_SVM);
3199 
3200 	/* SEV-ES guests must use the CR write traps to track CR registers. */
3201 	if (!sev_es_guest(vcpu->kvm)) {
3202 		if (!svm_is_intercept(svm, INTERCEPT_CR0_WRITE))
3203 			vcpu->arch.cr0 = svm->vmcb->save.cr0;
3204 		if (npt_enabled)
3205 			vcpu->arch.cr3 = svm->vmcb->save.cr3;
3206 	}
3207 
3208 	if (is_guest_mode(vcpu)) {
3209 		int vmexit;
3210 
3211 		trace_kvm_nested_vmexit(vcpu, KVM_ISA_SVM);
3212 
3213 		vmexit = nested_svm_exit_special(svm);
3214 
3215 		if (vmexit == NESTED_EXIT_CONTINUE)
3216 			vmexit = nested_svm_exit_handled(svm);
3217 
3218 		if (vmexit == NESTED_EXIT_DONE)
3219 			return 1;
3220 	}
3221 
3222 	if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
3223 		kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3224 		kvm_run->fail_entry.hardware_entry_failure_reason
3225 			= svm->vmcb->control.exit_code;
3226 		kvm_run->fail_entry.cpu = vcpu->arch.last_vmentry_cpu;
3227 		dump_vmcb(vcpu);
3228 		return 0;
3229 	}
3230 
3231 	if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
3232 	    exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
3233 	    exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH &&
3234 	    exit_code != SVM_EXIT_INTR && exit_code != SVM_EXIT_NMI)
3235 		printk(KERN_ERR "%s: unexpected exit_int_info 0x%x "
3236 		       "exit_code 0x%x\n",
3237 		       __func__, svm->vmcb->control.exit_int_info,
3238 		       exit_code);
3239 
3240 	if (exit_fastpath != EXIT_FASTPATH_NONE)
3241 		return 1;
3242 
3243 	return svm_invoke_exit_handler(vcpu, exit_code);
3244 }
3245 
3246 static void reload_tss(struct kvm_vcpu *vcpu)
3247 {
3248 	struct svm_cpu_data *sd = per_cpu(svm_data, vcpu->cpu);
3249 
3250 	sd->tss_desc->type = 9; /* available 32/64-bit TSS */
3251 	load_TR_desc();
3252 }
3253 
3254 static void pre_svm_run(struct kvm_vcpu *vcpu)
3255 {
3256 	struct svm_cpu_data *sd = per_cpu(svm_data, vcpu->cpu);
3257 	struct vcpu_svm *svm = to_svm(vcpu);
3258 
3259 	/*
3260 	 * If the previous vmrun of the vmcb occurred on a different physical
3261 	 * cpu, then mark the vmcb dirty and assign a new asid.  Hardware's
3262 	 * vmcb clean bits are per logical CPU, as are KVM's asid assignments.
3263 	 */
3264 	if (unlikely(svm->current_vmcb->cpu != vcpu->cpu)) {
3265 		svm->current_vmcb->asid_generation = 0;
3266 		vmcb_mark_all_dirty(svm->vmcb);
3267 		svm->current_vmcb->cpu = vcpu->cpu;
3268         }
3269 
3270 	if (sev_guest(vcpu->kvm))
3271 		return pre_sev_run(svm, vcpu->cpu);
3272 
3273 	/* FIXME: handle wraparound of asid_generation */
3274 	if (svm->current_vmcb->asid_generation != sd->asid_generation)
3275 		new_asid(svm, sd);
3276 }
3277 
3278 static void svm_inject_nmi(struct kvm_vcpu *vcpu)
3279 {
3280 	struct vcpu_svm *svm = to_svm(vcpu);
3281 
3282 	svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
3283 	vcpu->arch.hflags |= HF_NMI_MASK;
3284 	if (!sev_es_guest(vcpu->kvm))
3285 		svm_set_intercept(svm, INTERCEPT_IRET);
3286 	++vcpu->stat.nmi_injections;
3287 }
3288 
3289 static void svm_set_irq(struct kvm_vcpu *vcpu)
3290 {
3291 	struct vcpu_svm *svm = to_svm(vcpu);
3292 
3293 	BUG_ON(!(gif_set(svm)));
3294 
3295 	trace_kvm_inj_virq(vcpu->arch.interrupt.nr);
3296 	++vcpu->stat.irq_injections;
3297 
3298 	svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr |
3299 		SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR;
3300 }
3301 
3302 void svm_complete_interrupt_delivery(struct kvm_vcpu *vcpu, int delivery_mode,
3303 				     int trig_mode, int vector)
3304 {
3305 	/*
3306 	 * vcpu->arch.apicv_active must be read after vcpu->mode.
3307 	 * Pairs with smp_store_release in vcpu_enter_guest.
3308 	 */
3309 	bool in_guest_mode = (smp_load_acquire(&vcpu->mode) == IN_GUEST_MODE);
3310 
3311 	if (!READ_ONCE(vcpu->arch.apicv_active)) {
3312 		/* Process the interrupt via inject_pending_event */
3313 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3314 		kvm_vcpu_kick(vcpu);
3315 		return;
3316 	}
3317 
3318 	trace_kvm_apicv_accept_irq(vcpu->vcpu_id, delivery_mode, trig_mode, vector);
3319 	if (in_guest_mode) {
3320 		/*
3321 		 * Signal the doorbell to tell hardware to inject the IRQ.  If
3322 		 * the vCPU exits the guest before the doorbell chimes, hardware
3323 		 * will automatically process AVIC interrupts at the next VMRUN.
3324 		 */
3325 		avic_ring_doorbell(vcpu);
3326 	} else {
3327 		/*
3328 		 * Wake the vCPU if it was blocking.  KVM will then detect the
3329 		 * pending IRQ when checking if the vCPU has a wake event.
3330 		 */
3331 		kvm_vcpu_wake_up(vcpu);
3332 	}
3333 }
3334 
3335 static void svm_deliver_interrupt(struct kvm_lapic *apic,  int delivery_mode,
3336 				  int trig_mode, int vector)
3337 {
3338 	kvm_lapic_set_irr(vector, apic);
3339 
3340 	/*
3341 	 * Pairs with the smp_mb_*() after setting vcpu->guest_mode in
3342 	 * vcpu_enter_guest() to ensure the write to the vIRR is ordered before
3343 	 * the read of guest_mode.  This guarantees that either VMRUN will see
3344 	 * and process the new vIRR entry, or that svm_complete_interrupt_delivery
3345 	 * will signal the doorbell if the CPU has already entered the guest.
3346 	 */
3347 	smp_mb__after_atomic();
3348 	svm_complete_interrupt_delivery(apic->vcpu, delivery_mode, trig_mode, vector);
3349 }
3350 
3351 static void svm_update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
3352 {
3353 	struct vcpu_svm *svm = to_svm(vcpu);
3354 
3355 	/*
3356 	 * SEV-ES guests must always keep the CR intercepts cleared. CR
3357 	 * tracking is done using the CR write traps.
3358 	 */
3359 	if (sev_es_guest(vcpu->kvm))
3360 		return;
3361 
3362 	if (nested_svm_virtualize_tpr(vcpu))
3363 		return;
3364 
3365 	svm_clr_intercept(svm, INTERCEPT_CR8_WRITE);
3366 
3367 	if (irr == -1)
3368 		return;
3369 
3370 	if (tpr >= irr)
3371 		svm_set_intercept(svm, INTERCEPT_CR8_WRITE);
3372 }
3373 
3374 bool svm_nmi_blocked(struct kvm_vcpu *vcpu)
3375 {
3376 	struct vcpu_svm *svm = to_svm(vcpu);
3377 	struct vmcb *vmcb = svm->vmcb;
3378 	bool ret;
3379 
3380 	if (!gif_set(svm))
3381 		return true;
3382 
3383 	if (is_guest_mode(vcpu) && nested_exit_on_nmi(svm))
3384 		return false;
3385 
3386 	ret = (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) ||
3387 	      (vcpu->arch.hflags & HF_NMI_MASK);
3388 
3389 	return ret;
3390 }
3391 
3392 static int svm_nmi_allowed(struct kvm_vcpu *vcpu, bool for_injection)
3393 {
3394 	struct vcpu_svm *svm = to_svm(vcpu);
3395 	if (svm->nested.nested_run_pending)
3396 		return -EBUSY;
3397 
3398 	if (svm_nmi_blocked(vcpu))
3399 		return 0;
3400 
3401 	/* An NMI must not be injected into L2 if it's supposed to VM-Exit.  */
3402 	if (for_injection && is_guest_mode(vcpu) && nested_exit_on_nmi(svm))
3403 		return -EBUSY;
3404 	return 1;
3405 }
3406 
3407 static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu)
3408 {
3409 	return !!(vcpu->arch.hflags & HF_NMI_MASK);
3410 }
3411 
3412 static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
3413 {
3414 	struct vcpu_svm *svm = to_svm(vcpu);
3415 
3416 	if (masked) {
3417 		vcpu->arch.hflags |= HF_NMI_MASK;
3418 		if (!sev_es_guest(vcpu->kvm))
3419 			svm_set_intercept(svm, INTERCEPT_IRET);
3420 	} else {
3421 		vcpu->arch.hflags &= ~HF_NMI_MASK;
3422 		if (!sev_es_guest(vcpu->kvm))
3423 			svm_clr_intercept(svm, INTERCEPT_IRET);
3424 	}
3425 }
3426 
3427 bool svm_interrupt_blocked(struct kvm_vcpu *vcpu)
3428 {
3429 	struct vcpu_svm *svm = to_svm(vcpu);
3430 	struct vmcb *vmcb = svm->vmcb;
3431 
3432 	if (!gif_set(svm))
3433 		return true;
3434 
3435 	if (is_guest_mode(vcpu)) {
3436 		/* As long as interrupts are being delivered...  */
3437 		if ((svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK)
3438 		    ? !(svm->vmcb01.ptr->save.rflags & X86_EFLAGS_IF)
3439 		    : !(kvm_get_rflags(vcpu) & X86_EFLAGS_IF))
3440 			return true;
3441 
3442 		/* ... vmexits aren't blocked by the interrupt shadow  */
3443 		if (nested_exit_on_intr(svm))
3444 			return false;
3445 	} else {
3446 		if (!svm_get_if_flag(vcpu))
3447 			return true;
3448 	}
3449 
3450 	return (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK);
3451 }
3452 
3453 static int svm_interrupt_allowed(struct kvm_vcpu *vcpu, bool for_injection)
3454 {
3455 	struct vcpu_svm *svm = to_svm(vcpu);
3456 
3457 	if (svm->nested.nested_run_pending)
3458 		return -EBUSY;
3459 
3460 	if (svm_interrupt_blocked(vcpu))
3461 		return 0;
3462 
3463 	/*
3464 	 * An IRQ must not be injected into L2 if it's supposed to VM-Exit,
3465 	 * e.g. if the IRQ arrived asynchronously after checking nested events.
3466 	 */
3467 	if (for_injection && is_guest_mode(vcpu) && nested_exit_on_intr(svm))
3468 		return -EBUSY;
3469 
3470 	return 1;
3471 }
3472 
3473 static void svm_enable_irq_window(struct kvm_vcpu *vcpu)
3474 {
3475 	struct vcpu_svm *svm = to_svm(vcpu);
3476 
3477 	/*
3478 	 * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
3479 	 * 1, because that's a separate STGI/VMRUN intercept.  The next time we
3480 	 * get that intercept, this function will be called again though and
3481 	 * we'll get the vintr intercept. However, if the vGIF feature is
3482 	 * enabled, the STGI interception will not occur. Enable the irq
3483 	 * window under the assumption that the hardware will set the GIF.
3484 	 */
3485 	if (vgif_enabled(svm) || gif_set(svm)) {
3486 		/*
3487 		 * IRQ window is not needed when AVIC is enabled,
3488 		 * unless we have pending ExtINT since it cannot be injected
3489 		 * via AVIC. In such case, we need to temporarily disable AVIC,
3490 		 * and fallback to injecting IRQ via V_IRQ.
3491 		 */
3492 		kvm_request_apicv_update(vcpu->kvm, false, APICV_INHIBIT_REASON_IRQWIN);
3493 		svm_set_vintr(svm);
3494 	}
3495 }
3496 
3497 static void svm_enable_nmi_window(struct kvm_vcpu *vcpu)
3498 {
3499 	struct vcpu_svm *svm = to_svm(vcpu);
3500 
3501 	if ((vcpu->arch.hflags & (HF_NMI_MASK | HF_IRET_MASK)) == HF_NMI_MASK)
3502 		return; /* IRET will cause a vm exit */
3503 
3504 	if (!gif_set(svm)) {
3505 		if (vgif_enabled(svm))
3506 			svm_set_intercept(svm, INTERCEPT_STGI);
3507 		return; /* STGI will cause a vm exit */
3508 	}
3509 
3510 	/*
3511 	 * Something prevents NMI from been injected. Single step over possible
3512 	 * problem (IRET or exception injection or interrupt shadow)
3513 	 */
3514 	svm->nmi_singlestep_guest_rflags = svm_get_rflags(vcpu);
3515 	svm->nmi_singlestep = true;
3516 	svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
3517 }
3518 
3519 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
3520 {
3521 	return 0;
3522 }
3523 
3524 static int svm_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
3525 {
3526 	return 0;
3527 }
3528 
3529 void svm_flush_tlb(struct kvm_vcpu *vcpu)
3530 {
3531 	struct vcpu_svm *svm = to_svm(vcpu);
3532 
3533 	/*
3534 	 * Flush only the current ASID even if the TLB flush was invoked via
3535 	 * kvm_flush_remote_tlbs().  Although flushing remote TLBs requires all
3536 	 * ASIDs to be flushed, KVM uses a single ASID for L1 and L2, and
3537 	 * unconditionally does a TLB flush on both nested VM-Enter and nested
3538 	 * VM-Exit (via kvm_mmu_reset_context()).
3539 	 */
3540 	if (static_cpu_has(X86_FEATURE_FLUSHBYASID))
3541 		svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
3542 	else
3543 		svm->current_vmcb->asid_generation--;
3544 }
3545 
3546 static void svm_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t gva)
3547 {
3548 	struct vcpu_svm *svm = to_svm(vcpu);
3549 
3550 	invlpga(gva, svm->vmcb->control.asid);
3551 }
3552 
3553 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
3554 {
3555 	struct vcpu_svm *svm = to_svm(vcpu);
3556 
3557 	if (nested_svm_virtualize_tpr(vcpu))
3558 		return;
3559 
3560 	if (!svm_is_intercept(svm, INTERCEPT_CR8_WRITE)) {
3561 		int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
3562 		kvm_set_cr8(vcpu, cr8);
3563 	}
3564 }
3565 
3566 static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
3567 {
3568 	struct vcpu_svm *svm = to_svm(vcpu);
3569 	u64 cr8;
3570 
3571 	if (nested_svm_virtualize_tpr(vcpu) ||
3572 	    kvm_vcpu_apicv_active(vcpu))
3573 		return;
3574 
3575 	cr8 = kvm_get_cr8(vcpu);
3576 	svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
3577 	svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
3578 }
3579 
3580 static void svm_complete_interrupts(struct kvm_vcpu *vcpu)
3581 {
3582 	struct vcpu_svm *svm = to_svm(vcpu);
3583 	u8 vector;
3584 	int type;
3585 	u32 exitintinfo = svm->vmcb->control.exit_int_info;
3586 	unsigned int3_injected = svm->int3_injected;
3587 
3588 	svm->int3_injected = 0;
3589 
3590 	/*
3591 	 * If we've made progress since setting HF_IRET_MASK, we've
3592 	 * executed an IRET and can allow NMI injection.
3593 	 */
3594 	if ((vcpu->arch.hflags & HF_IRET_MASK) &&
3595 	    (sev_es_guest(vcpu->kvm) ||
3596 	     kvm_rip_read(vcpu) != svm->nmi_iret_rip)) {
3597 		vcpu->arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK);
3598 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3599 	}
3600 
3601 	vcpu->arch.nmi_injected = false;
3602 	kvm_clear_exception_queue(vcpu);
3603 	kvm_clear_interrupt_queue(vcpu);
3604 
3605 	if (!(exitintinfo & SVM_EXITINTINFO_VALID))
3606 		return;
3607 
3608 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3609 
3610 	vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK;
3611 	type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK;
3612 
3613 	switch (type) {
3614 	case SVM_EXITINTINFO_TYPE_NMI:
3615 		vcpu->arch.nmi_injected = true;
3616 		break;
3617 	case SVM_EXITINTINFO_TYPE_EXEPT:
3618 		/*
3619 		 * Never re-inject a #VC exception.
3620 		 */
3621 		if (vector == X86_TRAP_VC)
3622 			break;
3623 
3624 		/*
3625 		 * In case of software exceptions, do not reinject the vector,
3626 		 * but re-execute the instruction instead. Rewind RIP first
3627 		 * if we emulated INT3 before.
3628 		 */
3629 		if (kvm_exception_is_soft(vector)) {
3630 			if (vector == BP_VECTOR && int3_injected &&
3631 			    kvm_is_linear_rip(vcpu, svm->int3_rip))
3632 				kvm_rip_write(vcpu,
3633 					      kvm_rip_read(vcpu) - int3_injected);
3634 			break;
3635 		}
3636 		if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) {
3637 			u32 err = svm->vmcb->control.exit_int_info_err;
3638 			kvm_requeue_exception_e(vcpu, vector, err);
3639 
3640 		} else
3641 			kvm_requeue_exception(vcpu, vector);
3642 		break;
3643 	case SVM_EXITINTINFO_TYPE_INTR:
3644 		kvm_queue_interrupt(vcpu, vector, false);
3645 		break;
3646 	default:
3647 		break;
3648 	}
3649 }
3650 
3651 static void svm_cancel_injection(struct kvm_vcpu *vcpu)
3652 {
3653 	struct vcpu_svm *svm = to_svm(vcpu);
3654 	struct vmcb_control_area *control = &svm->vmcb->control;
3655 
3656 	control->exit_int_info = control->event_inj;
3657 	control->exit_int_info_err = control->event_inj_err;
3658 	control->event_inj = 0;
3659 	svm_complete_interrupts(vcpu);
3660 }
3661 
3662 static int svm_vcpu_pre_run(struct kvm_vcpu *vcpu)
3663 {
3664 	return 1;
3665 }
3666 
3667 static fastpath_t svm_exit_handlers_fastpath(struct kvm_vcpu *vcpu)
3668 {
3669 	if (to_svm(vcpu)->vmcb->control.exit_code == SVM_EXIT_MSR &&
3670 	    to_svm(vcpu)->vmcb->control.exit_info_1)
3671 		return handle_fastpath_set_msr_irqoff(vcpu);
3672 
3673 	return EXIT_FASTPATH_NONE;
3674 }
3675 
3676 static noinstr void svm_vcpu_enter_exit(struct kvm_vcpu *vcpu)
3677 {
3678 	struct vcpu_svm *svm = to_svm(vcpu);
3679 	unsigned long vmcb_pa = svm->current_vmcb->pa;
3680 
3681 	guest_state_enter_irqoff();
3682 
3683 	if (sev_es_guest(vcpu->kvm)) {
3684 		__svm_sev_es_vcpu_run(vmcb_pa);
3685 	} else {
3686 		struct svm_cpu_data *sd = per_cpu(svm_data, vcpu->cpu);
3687 
3688 		/*
3689 		 * Use a single vmcb (vmcb01 because it's always valid) for
3690 		 * context switching guest state via VMLOAD/VMSAVE, that way
3691 		 * the state doesn't need to be copied between vmcb01 and
3692 		 * vmcb02 when switching vmcbs for nested virtualization.
3693 		 */
3694 		vmload(svm->vmcb01.pa);
3695 		__svm_vcpu_run(vmcb_pa, (unsigned long *)&vcpu->arch.regs);
3696 		vmsave(svm->vmcb01.pa);
3697 
3698 		vmload(__sme_page_pa(sd->save_area));
3699 	}
3700 
3701 	guest_state_exit_irqoff();
3702 }
3703 
3704 static __no_kcsan fastpath_t svm_vcpu_run(struct kvm_vcpu *vcpu)
3705 {
3706 	struct vcpu_svm *svm = to_svm(vcpu);
3707 
3708 	trace_kvm_entry(vcpu);
3709 
3710 	svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
3711 	svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
3712 	svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
3713 
3714 	/*
3715 	 * Disable singlestep if we're injecting an interrupt/exception.
3716 	 * We don't want our modified rflags to be pushed on the stack where
3717 	 * we might not be able to easily reset them if we disabled NMI
3718 	 * singlestep later.
3719 	 */
3720 	if (svm->nmi_singlestep && svm->vmcb->control.event_inj) {
3721 		/*
3722 		 * Event injection happens before external interrupts cause a
3723 		 * vmexit and interrupts are disabled here, so smp_send_reschedule
3724 		 * is enough to force an immediate vmexit.
3725 		 */
3726 		disable_nmi_singlestep(svm);
3727 		smp_send_reschedule(vcpu->cpu);
3728 	}
3729 
3730 	pre_svm_run(vcpu);
3731 
3732 	sync_lapic_to_cr8(vcpu);
3733 
3734 	if (unlikely(svm->asid != svm->vmcb->control.asid)) {
3735 		svm->vmcb->control.asid = svm->asid;
3736 		vmcb_mark_dirty(svm->vmcb, VMCB_ASID);
3737 	}
3738 	svm->vmcb->save.cr2 = vcpu->arch.cr2;
3739 
3740 	svm_hv_update_vp_id(svm->vmcb, vcpu);
3741 
3742 	/*
3743 	 * Run with all-zero DR6 unless needed, so that we can get the exact cause
3744 	 * of a #DB.
3745 	 */
3746 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT))
3747 		svm_set_dr6(svm, vcpu->arch.dr6);
3748 	else
3749 		svm_set_dr6(svm, DR6_ACTIVE_LOW);
3750 
3751 	clgi();
3752 	kvm_load_guest_xsave_state(vcpu);
3753 
3754 	kvm_wait_lapic_expire(vcpu);
3755 
3756 	/*
3757 	 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
3758 	 * it's non-zero. Since vmentry is serialising on affected CPUs, there
3759 	 * is no need to worry about the conditional branch over the wrmsr
3760 	 * being speculatively taken.
3761 	 */
3762 	if (!static_cpu_has(X86_FEATURE_V_SPEC_CTRL))
3763 		x86_spec_ctrl_set_guest(svm->spec_ctrl, svm->virt_spec_ctrl);
3764 
3765 	svm_vcpu_enter_exit(vcpu);
3766 
3767 	/*
3768 	 * We do not use IBRS in the kernel. If this vCPU has used the
3769 	 * SPEC_CTRL MSR it may have left it on; save the value and
3770 	 * turn it off. This is much more efficient than blindly adding
3771 	 * it to the atomic save/restore list. Especially as the former
3772 	 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
3773 	 *
3774 	 * For non-nested case:
3775 	 * If the L01 MSR bitmap does not intercept the MSR, then we need to
3776 	 * save it.
3777 	 *
3778 	 * For nested case:
3779 	 * If the L02 MSR bitmap does not intercept the MSR, then we need to
3780 	 * save it.
3781 	 */
3782 	if (!static_cpu_has(X86_FEATURE_V_SPEC_CTRL) &&
3783 	    unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
3784 		svm->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
3785 
3786 	if (!sev_es_guest(vcpu->kvm))
3787 		reload_tss(vcpu);
3788 
3789 	if (!static_cpu_has(X86_FEATURE_V_SPEC_CTRL))
3790 		x86_spec_ctrl_restore_host(svm->spec_ctrl, svm->virt_spec_ctrl);
3791 
3792 	if (!sev_es_guest(vcpu->kvm)) {
3793 		vcpu->arch.cr2 = svm->vmcb->save.cr2;
3794 		vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
3795 		vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
3796 		vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
3797 	}
3798 	vcpu->arch.regs_dirty = 0;
3799 
3800 	if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
3801 		kvm_before_interrupt(vcpu, KVM_HANDLING_NMI);
3802 
3803 	kvm_load_host_xsave_state(vcpu);
3804 	stgi();
3805 
3806 	/* Any pending NMI will happen here */
3807 
3808 	if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
3809 		kvm_after_interrupt(vcpu);
3810 
3811 	sync_cr8_to_lapic(vcpu);
3812 
3813 	svm->next_rip = 0;
3814 	if (is_guest_mode(vcpu)) {
3815 		nested_sync_control_from_vmcb02(svm);
3816 
3817 		/* Track VMRUNs that have made past consistency checking */
3818 		if (svm->nested.nested_run_pending &&
3819 		    svm->vmcb->control.exit_code != SVM_EXIT_ERR)
3820                         ++vcpu->stat.nested_run;
3821 
3822 		svm->nested.nested_run_pending = 0;
3823 	}
3824 
3825 	svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
3826 	vmcb_mark_all_clean(svm->vmcb);
3827 
3828 	/* if exit due to PF check for async PF */
3829 	if (svm->vmcb->control.exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR)
3830 		vcpu->arch.apf.host_apf_flags =
3831 			kvm_read_and_reset_apf_flags();
3832 
3833 	vcpu->arch.regs_avail &= ~SVM_REGS_LAZY_LOAD_SET;
3834 
3835 	/*
3836 	 * We need to handle MC intercepts here before the vcpu has a chance to
3837 	 * change the physical cpu
3838 	 */
3839 	if (unlikely(svm->vmcb->control.exit_code ==
3840 		     SVM_EXIT_EXCP_BASE + MC_VECTOR))
3841 		svm_handle_mce(vcpu);
3842 
3843 	svm_complete_interrupts(vcpu);
3844 
3845 	if (is_guest_mode(vcpu))
3846 		return EXIT_FASTPATH_NONE;
3847 
3848 	return svm_exit_handlers_fastpath(vcpu);
3849 }
3850 
3851 static void svm_load_mmu_pgd(struct kvm_vcpu *vcpu, hpa_t root_hpa,
3852 			     int root_level)
3853 {
3854 	struct vcpu_svm *svm = to_svm(vcpu);
3855 	unsigned long cr3;
3856 
3857 	if (npt_enabled) {
3858 		svm->vmcb->control.nested_cr3 = __sme_set(root_hpa);
3859 		vmcb_mark_dirty(svm->vmcb, VMCB_NPT);
3860 
3861 		hv_track_root_tdp(vcpu, root_hpa);
3862 
3863 		cr3 = vcpu->arch.cr3;
3864 	} else if (vcpu->arch.mmu->shadow_root_level >= PT64_ROOT_4LEVEL) {
3865 		cr3 = __sme_set(root_hpa) | kvm_get_active_pcid(vcpu);
3866 	} else {
3867 		/* PCID in the guest should be impossible with a 32-bit MMU. */
3868 		WARN_ON_ONCE(kvm_get_active_pcid(vcpu));
3869 		cr3 = root_hpa;
3870 	}
3871 
3872 	svm->vmcb->save.cr3 = cr3;
3873 	vmcb_mark_dirty(svm->vmcb, VMCB_CR);
3874 }
3875 
3876 static int is_disabled(void)
3877 {
3878 	u64 vm_cr;
3879 
3880 	rdmsrl(MSR_VM_CR, vm_cr);
3881 	if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
3882 		return 1;
3883 
3884 	return 0;
3885 }
3886 
3887 static void
3888 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
3889 {
3890 	/*
3891 	 * Patch in the VMMCALL instruction:
3892 	 */
3893 	hypercall[0] = 0x0f;
3894 	hypercall[1] = 0x01;
3895 	hypercall[2] = 0xd9;
3896 }
3897 
3898 static int __init svm_check_processor_compat(void)
3899 {
3900 	return 0;
3901 }
3902 
3903 static bool svm_cpu_has_accelerated_tpr(void)
3904 {
3905 	return false;
3906 }
3907 
3908 /*
3909  * The kvm parameter can be NULL (module initialization, or invocation before
3910  * VM creation). Be sure to check the kvm parameter before using it.
3911  */
3912 static bool svm_has_emulated_msr(struct kvm *kvm, u32 index)
3913 {
3914 	switch (index) {
3915 	case MSR_IA32_MCG_EXT_CTL:
3916 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
3917 		return false;
3918 	case MSR_IA32_SMBASE:
3919 		/* SEV-ES guests do not support SMM, so report false */
3920 		if (kvm && sev_es_guest(kvm))
3921 			return false;
3922 		break;
3923 	default:
3924 		break;
3925 	}
3926 
3927 	return true;
3928 }
3929 
3930 static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
3931 {
3932 	return 0;
3933 }
3934 
3935 static void svm_vcpu_after_set_cpuid(struct kvm_vcpu *vcpu)
3936 {
3937 	struct vcpu_svm *svm = to_svm(vcpu);
3938 	struct kvm_cpuid_entry2 *best;
3939 
3940 	vcpu->arch.xsaves_enabled = guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
3941 				    boot_cpu_has(X86_FEATURE_XSAVE) &&
3942 				    boot_cpu_has(X86_FEATURE_XSAVES);
3943 
3944 	/* Update nrips enabled cache */
3945 	svm->nrips_enabled = kvm_cpu_cap_has(X86_FEATURE_NRIPS) &&
3946 			     guest_cpuid_has(vcpu, X86_FEATURE_NRIPS);
3947 
3948 	svm->tsc_scaling_enabled = tsc_scaling && guest_cpuid_has(vcpu, X86_FEATURE_TSCRATEMSR);
3949 
3950 	svm_recalc_instruction_intercepts(vcpu, svm);
3951 
3952 	/* For sev guests, the memory encryption bit is not reserved in CR3.  */
3953 	if (sev_guest(vcpu->kvm)) {
3954 		best = kvm_find_cpuid_entry(vcpu, 0x8000001F, 0);
3955 		if (best)
3956 			vcpu->arch.reserved_gpa_bits &= ~(1UL << (best->ebx & 0x3f));
3957 	}
3958 
3959 	if (kvm_vcpu_apicv_active(vcpu)) {
3960 		/*
3961 		 * AVIC does not work with an x2APIC mode guest. If the X2APIC feature
3962 		 * is exposed to the guest, disable AVIC.
3963 		 */
3964 		if (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC))
3965 			kvm_request_apicv_update(vcpu->kvm, false,
3966 						 APICV_INHIBIT_REASON_X2APIC);
3967 
3968 		/*
3969 		 * Currently, AVIC does not work with nested virtualization.
3970 		 * So, we disable AVIC when cpuid for SVM is set in the L1 guest.
3971 		 */
3972 		if (nested && guest_cpuid_has(vcpu, X86_FEATURE_SVM))
3973 			kvm_request_apicv_update(vcpu->kvm, false,
3974 						 APICV_INHIBIT_REASON_NESTED);
3975 	}
3976 	init_vmcb_after_set_cpuid(vcpu);
3977 }
3978 
3979 static bool svm_has_wbinvd_exit(void)
3980 {
3981 	return true;
3982 }
3983 
3984 #define PRE_EX(exit)  { .exit_code = (exit), \
3985 			.stage = X86_ICPT_PRE_EXCEPT, }
3986 #define POST_EX(exit) { .exit_code = (exit), \
3987 			.stage = X86_ICPT_POST_EXCEPT, }
3988 #define POST_MEM(exit) { .exit_code = (exit), \
3989 			.stage = X86_ICPT_POST_MEMACCESS, }
3990 
3991 static const struct __x86_intercept {
3992 	u32 exit_code;
3993 	enum x86_intercept_stage stage;
3994 } x86_intercept_map[] = {
3995 	[x86_intercept_cr_read]		= POST_EX(SVM_EXIT_READ_CR0),
3996 	[x86_intercept_cr_write]	= POST_EX(SVM_EXIT_WRITE_CR0),
3997 	[x86_intercept_clts]		= POST_EX(SVM_EXIT_WRITE_CR0),
3998 	[x86_intercept_lmsw]		= POST_EX(SVM_EXIT_WRITE_CR0),
3999 	[x86_intercept_smsw]		= POST_EX(SVM_EXIT_READ_CR0),
4000 	[x86_intercept_dr_read]		= POST_EX(SVM_EXIT_READ_DR0),
4001 	[x86_intercept_dr_write]	= POST_EX(SVM_EXIT_WRITE_DR0),
4002 	[x86_intercept_sldt]		= POST_EX(SVM_EXIT_LDTR_READ),
4003 	[x86_intercept_str]		= POST_EX(SVM_EXIT_TR_READ),
4004 	[x86_intercept_lldt]		= POST_EX(SVM_EXIT_LDTR_WRITE),
4005 	[x86_intercept_ltr]		= POST_EX(SVM_EXIT_TR_WRITE),
4006 	[x86_intercept_sgdt]		= POST_EX(SVM_EXIT_GDTR_READ),
4007 	[x86_intercept_sidt]		= POST_EX(SVM_EXIT_IDTR_READ),
4008 	[x86_intercept_lgdt]		= POST_EX(SVM_EXIT_GDTR_WRITE),
4009 	[x86_intercept_lidt]		= POST_EX(SVM_EXIT_IDTR_WRITE),
4010 	[x86_intercept_vmrun]		= POST_EX(SVM_EXIT_VMRUN),
4011 	[x86_intercept_vmmcall]		= POST_EX(SVM_EXIT_VMMCALL),
4012 	[x86_intercept_vmload]		= POST_EX(SVM_EXIT_VMLOAD),
4013 	[x86_intercept_vmsave]		= POST_EX(SVM_EXIT_VMSAVE),
4014 	[x86_intercept_stgi]		= POST_EX(SVM_EXIT_STGI),
4015 	[x86_intercept_clgi]		= POST_EX(SVM_EXIT_CLGI),
4016 	[x86_intercept_skinit]		= POST_EX(SVM_EXIT_SKINIT),
4017 	[x86_intercept_invlpga]		= POST_EX(SVM_EXIT_INVLPGA),
4018 	[x86_intercept_rdtscp]		= POST_EX(SVM_EXIT_RDTSCP),
4019 	[x86_intercept_monitor]		= POST_MEM(SVM_EXIT_MONITOR),
4020 	[x86_intercept_mwait]		= POST_EX(SVM_EXIT_MWAIT),
4021 	[x86_intercept_invlpg]		= POST_EX(SVM_EXIT_INVLPG),
4022 	[x86_intercept_invd]		= POST_EX(SVM_EXIT_INVD),
4023 	[x86_intercept_wbinvd]		= POST_EX(SVM_EXIT_WBINVD),
4024 	[x86_intercept_wrmsr]		= POST_EX(SVM_EXIT_MSR),
4025 	[x86_intercept_rdtsc]		= POST_EX(SVM_EXIT_RDTSC),
4026 	[x86_intercept_rdmsr]		= POST_EX(SVM_EXIT_MSR),
4027 	[x86_intercept_rdpmc]		= POST_EX(SVM_EXIT_RDPMC),
4028 	[x86_intercept_cpuid]		= PRE_EX(SVM_EXIT_CPUID),
4029 	[x86_intercept_rsm]		= PRE_EX(SVM_EXIT_RSM),
4030 	[x86_intercept_pause]		= PRE_EX(SVM_EXIT_PAUSE),
4031 	[x86_intercept_pushf]		= PRE_EX(SVM_EXIT_PUSHF),
4032 	[x86_intercept_popf]		= PRE_EX(SVM_EXIT_POPF),
4033 	[x86_intercept_intn]		= PRE_EX(SVM_EXIT_SWINT),
4034 	[x86_intercept_iret]		= PRE_EX(SVM_EXIT_IRET),
4035 	[x86_intercept_icebp]		= PRE_EX(SVM_EXIT_ICEBP),
4036 	[x86_intercept_hlt]		= POST_EX(SVM_EXIT_HLT),
4037 	[x86_intercept_in]		= POST_EX(SVM_EXIT_IOIO),
4038 	[x86_intercept_ins]		= POST_EX(SVM_EXIT_IOIO),
4039 	[x86_intercept_out]		= POST_EX(SVM_EXIT_IOIO),
4040 	[x86_intercept_outs]		= POST_EX(SVM_EXIT_IOIO),
4041 	[x86_intercept_xsetbv]		= PRE_EX(SVM_EXIT_XSETBV),
4042 };
4043 
4044 #undef PRE_EX
4045 #undef POST_EX
4046 #undef POST_MEM
4047 
4048 static int svm_check_intercept(struct kvm_vcpu *vcpu,
4049 			       struct x86_instruction_info *info,
4050 			       enum x86_intercept_stage stage,
4051 			       struct x86_exception *exception)
4052 {
4053 	struct vcpu_svm *svm = to_svm(vcpu);
4054 	int vmexit, ret = X86EMUL_CONTINUE;
4055 	struct __x86_intercept icpt_info;
4056 	struct vmcb *vmcb = svm->vmcb;
4057 
4058 	if (info->intercept >= ARRAY_SIZE(x86_intercept_map))
4059 		goto out;
4060 
4061 	icpt_info = x86_intercept_map[info->intercept];
4062 
4063 	if (stage != icpt_info.stage)
4064 		goto out;
4065 
4066 	switch (icpt_info.exit_code) {
4067 	case SVM_EXIT_READ_CR0:
4068 		if (info->intercept == x86_intercept_cr_read)
4069 			icpt_info.exit_code += info->modrm_reg;
4070 		break;
4071 	case SVM_EXIT_WRITE_CR0: {
4072 		unsigned long cr0, val;
4073 
4074 		if (info->intercept == x86_intercept_cr_write)
4075 			icpt_info.exit_code += info->modrm_reg;
4076 
4077 		if (icpt_info.exit_code != SVM_EXIT_WRITE_CR0 ||
4078 		    info->intercept == x86_intercept_clts)
4079 			break;
4080 
4081 		if (!(vmcb12_is_intercept(&svm->nested.ctl,
4082 					INTERCEPT_SELECTIVE_CR0)))
4083 			break;
4084 
4085 		cr0 = vcpu->arch.cr0 & ~SVM_CR0_SELECTIVE_MASK;
4086 		val = info->src_val  & ~SVM_CR0_SELECTIVE_MASK;
4087 
4088 		if (info->intercept == x86_intercept_lmsw) {
4089 			cr0 &= 0xfUL;
4090 			val &= 0xfUL;
4091 			/* lmsw can't clear PE - catch this here */
4092 			if (cr0 & X86_CR0_PE)
4093 				val |= X86_CR0_PE;
4094 		}
4095 
4096 		if (cr0 ^ val)
4097 			icpt_info.exit_code = SVM_EXIT_CR0_SEL_WRITE;
4098 
4099 		break;
4100 	}
4101 	case SVM_EXIT_READ_DR0:
4102 	case SVM_EXIT_WRITE_DR0:
4103 		icpt_info.exit_code += info->modrm_reg;
4104 		break;
4105 	case SVM_EXIT_MSR:
4106 		if (info->intercept == x86_intercept_wrmsr)
4107 			vmcb->control.exit_info_1 = 1;
4108 		else
4109 			vmcb->control.exit_info_1 = 0;
4110 		break;
4111 	case SVM_EXIT_PAUSE:
4112 		/*
4113 		 * We get this for NOP only, but pause
4114 		 * is rep not, check this here
4115 		 */
4116 		if (info->rep_prefix != REPE_PREFIX)
4117 			goto out;
4118 		break;
4119 	case SVM_EXIT_IOIO: {
4120 		u64 exit_info;
4121 		u32 bytes;
4122 
4123 		if (info->intercept == x86_intercept_in ||
4124 		    info->intercept == x86_intercept_ins) {
4125 			exit_info = ((info->src_val & 0xffff) << 16) |
4126 				SVM_IOIO_TYPE_MASK;
4127 			bytes = info->dst_bytes;
4128 		} else {
4129 			exit_info = (info->dst_val & 0xffff) << 16;
4130 			bytes = info->src_bytes;
4131 		}
4132 
4133 		if (info->intercept == x86_intercept_outs ||
4134 		    info->intercept == x86_intercept_ins)
4135 			exit_info |= SVM_IOIO_STR_MASK;
4136 
4137 		if (info->rep_prefix)
4138 			exit_info |= SVM_IOIO_REP_MASK;
4139 
4140 		bytes = min(bytes, 4u);
4141 
4142 		exit_info |= bytes << SVM_IOIO_SIZE_SHIFT;
4143 
4144 		exit_info |= (u32)info->ad_bytes << (SVM_IOIO_ASIZE_SHIFT - 1);
4145 
4146 		vmcb->control.exit_info_1 = exit_info;
4147 		vmcb->control.exit_info_2 = info->next_rip;
4148 
4149 		break;
4150 	}
4151 	default:
4152 		break;
4153 	}
4154 
4155 	/* TODO: Advertise NRIPS to guest hypervisor unconditionally */
4156 	if (static_cpu_has(X86_FEATURE_NRIPS))
4157 		vmcb->control.next_rip  = info->next_rip;
4158 	vmcb->control.exit_code = icpt_info.exit_code;
4159 	vmexit = nested_svm_exit_handled(svm);
4160 
4161 	ret = (vmexit == NESTED_EXIT_DONE) ? X86EMUL_INTERCEPTED
4162 					   : X86EMUL_CONTINUE;
4163 
4164 out:
4165 	return ret;
4166 }
4167 
4168 static void svm_handle_exit_irqoff(struct kvm_vcpu *vcpu)
4169 {
4170 }
4171 
4172 static void svm_sched_in(struct kvm_vcpu *vcpu, int cpu)
4173 {
4174 	if (!kvm_pause_in_guest(vcpu->kvm))
4175 		shrink_ple_window(vcpu);
4176 }
4177 
4178 static void svm_setup_mce(struct kvm_vcpu *vcpu)
4179 {
4180 	/* [63:9] are reserved. */
4181 	vcpu->arch.mcg_cap &= 0x1ff;
4182 }
4183 
4184 bool svm_smi_blocked(struct kvm_vcpu *vcpu)
4185 {
4186 	struct vcpu_svm *svm = to_svm(vcpu);
4187 
4188 	/* Per APM Vol.2 15.22.2 "Response to SMI" */
4189 	if (!gif_set(svm))
4190 		return true;
4191 
4192 	return is_smm(vcpu);
4193 }
4194 
4195 static int svm_smi_allowed(struct kvm_vcpu *vcpu, bool for_injection)
4196 {
4197 	struct vcpu_svm *svm = to_svm(vcpu);
4198 	if (svm->nested.nested_run_pending)
4199 		return -EBUSY;
4200 
4201 	if (svm_smi_blocked(vcpu))
4202 		return 0;
4203 
4204 	/* An SMI must not be injected into L2 if it's supposed to VM-Exit.  */
4205 	if (for_injection && is_guest_mode(vcpu) && nested_exit_on_smi(svm))
4206 		return -EBUSY;
4207 
4208 	return 1;
4209 }
4210 
4211 static int svm_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
4212 {
4213 	struct vcpu_svm *svm = to_svm(vcpu);
4214 	struct kvm_host_map map_save;
4215 	int ret;
4216 
4217 	if (!is_guest_mode(vcpu))
4218 		return 0;
4219 
4220 	/* FED8h - SVM Guest */
4221 	put_smstate(u64, smstate, 0x7ed8, 1);
4222 	/* FEE0h - SVM Guest VMCB Physical Address */
4223 	put_smstate(u64, smstate, 0x7ee0, svm->nested.vmcb12_gpa);
4224 
4225 	svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
4226 	svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
4227 	svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
4228 
4229 	ret = nested_svm_vmexit(svm);
4230 	if (ret)
4231 		return ret;
4232 
4233 	/*
4234 	 * KVM uses VMCB01 to store L1 host state while L2 runs but
4235 	 * VMCB01 is going to be used during SMM and thus the state will
4236 	 * be lost. Temporary save non-VMLOAD/VMSAVE state to the host save
4237 	 * area pointed to by MSR_VM_HSAVE_PA. APM guarantees that the
4238 	 * format of the area is identical to guest save area offsetted
4239 	 * by 0x400 (matches the offset of 'struct vmcb_save_area'
4240 	 * within 'struct vmcb'). Note: HSAVE area may also be used by
4241 	 * L1 hypervisor to save additional host context (e.g. KVM does
4242 	 * that, see svm_prepare_guest_switch()) which must be
4243 	 * preserved.
4244 	 */
4245 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(svm->nested.hsave_msr),
4246 			 &map_save) == -EINVAL)
4247 		return 1;
4248 
4249 	BUILD_BUG_ON(offsetof(struct vmcb, save) != 0x400);
4250 
4251 	svm_copy_vmrun_state(map_save.hva + 0x400,
4252 			     &svm->vmcb01.ptr->save);
4253 
4254 	kvm_vcpu_unmap(vcpu, &map_save, true);
4255 	return 0;
4256 }
4257 
4258 static int svm_leave_smm(struct kvm_vcpu *vcpu, const char *smstate)
4259 {
4260 	struct vcpu_svm *svm = to_svm(vcpu);
4261 	struct kvm_host_map map, map_save;
4262 	u64 saved_efer, vmcb12_gpa;
4263 	struct vmcb *vmcb12;
4264 	int ret;
4265 
4266 	if (!guest_cpuid_has(vcpu, X86_FEATURE_LM))
4267 		return 0;
4268 
4269 	/* Non-zero if SMI arrived while vCPU was in guest mode. */
4270 	if (!GET_SMSTATE(u64, smstate, 0x7ed8))
4271 		return 0;
4272 
4273 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SVM))
4274 		return 1;
4275 
4276 	saved_efer = GET_SMSTATE(u64, smstate, 0x7ed0);
4277 	if (!(saved_efer & EFER_SVME))
4278 		return 1;
4279 
4280 	vmcb12_gpa = GET_SMSTATE(u64, smstate, 0x7ee0);
4281 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmcb12_gpa), &map) == -EINVAL)
4282 		return 1;
4283 
4284 	ret = 1;
4285 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(svm->nested.hsave_msr), &map_save) == -EINVAL)
4286 		goto unmap_map;
4287 
4288 	if (svm_allocate_nested(svm))
4289 		goto unmap_save;
4290 
4291 	/*
4292 	 * Restore L1 host state from L1 HSAVE area as VMCB01 was
4293 	 * used during SMM (see svm_enter_smm())
4294 	 */
4295 
4296 	svm_copy_vmrun_state(&svm->vmcb01.ptr->save, map_save.hva + 0x400);
4297 
4298 	/*
4299 	 * Enter the nested guest now
4300 	 */
4301 
4302 	vmcb_mark_all_dirty(svm->vmcb01.ptr);
4303 
4304 	vmcb12 = map.hva;
4305 	nested_copy_vmcb_control_to_cache(svm, &vmcb12->control);
4306 	nested_copy_vmcb_save_to_cache(svm, &vmcb12->save);
4307 	ret = enter_svm_guest_mode(vcpu, vmcb12_gpa, vmcb12, false);
4308 
4309 	if (ret)
4310 		goto unmap_save;
4311 
4312 	svm->nested.nested_run_pending = 1;
4313 
4314 unmap_save:
4315 	kvm_vcpu_unmap(vcpu, &map_save, true);
4316 unmap_map:
4317 	kvm_vcpu_unmap(vcpu, &map, true);
4318 	return ret;
4319 }
4320 
4321 static void svm_enable_smi_window(struct kvm_vcpu *vcpu)
4322 {
4323 	struct vcpu_svm *svm = to_svm(vcpu);
4324 
4325 	if (!gif_set(svm)) {
4326 		if (vgif_enabled(svm))
4327 			svm_set_intercept(svm, INTERCEPT_STGI);
4328 		/* STGI will cause a vm exit */
4329 	} else {
4330 		/* We must be in SMM; RSM will cause a vmexit anyway.  */
4331 	}
4332 }
4333 
4334 static bool svm_can_emulate_instruction(struct kvm_vcpu *vcpu, int emul_type,
4335 					void *insn, int insn_len)
4336 {
4337 	bool smep, smap, is_user;
4338 	unsigned long cr4;
4339 	u64 error_code;
4340 
4341 	/* Emulation is always possible when KVM has access to all guest state. */
4342 	if (!sev_guest(vcpu->kvm))
4343 		return true;
4344 
4345 	/* #UD and #GP should never be intercepted for SEV guests. */
4346 	WARN_ON_ONCE(emul_type & (EMULTYPE_TRAP_UD |
4347 				  EMULTYPE_TRAP_UD_FORCED |
4348 				  EMULTYPE_VMWARE_GP));
4349 
4350 	/*
4351 	 * Emulation is impossible for SEV-ES guests as KVM doesn't have access
4352 	 * to guest register state.
4353 	 */
4354 	if (sev_es_guest(vcpu->kvm))
4355 		return false;
4356 
4357 	/*
4358 	 * Emulation is possible if the instruction is already decoded, e.g.
4359 	 * when completing I/O after returning from userspace.
4360 	 */
4361 	if (emul_type & EMULTYPE_NO_DECODE)
4362 		return true;
4363 
4364 	/*
4365 	 * Emulation is possible for SEV guests if and only if a prefilled
4366 	 * buffer containing the bytes of the intercepted instruction is
4367 	 * available. SEV guest memory is encrypted with a guest specific key
4368 	 * and cannot be decrypted by KVM, i.e. KVM would read cyphertext and
4369 	 * decode garbage.
4370 	 *
4371 	 * Inject #UD if KVM reached this point without an instruction buffer.
4372 	 * In practice, this path should never be hit by a well-behaved guest,
4373 	 * e.g. KVM doesn't intercept #UD or #GP for SEV guests, but this path
4374 	 * is still theoretically reachable, e.g. via unaccelerated fault-like
4375 	 * AVIC access, and needs to be handled by KVM to avoid putting the
4376 	 * guest into an infinite loop.   Injecting #UD is somewhat arbitrary,
4377 	 * but its the least awful option given lack of insight into the guest.
4378 	 */
4379 	if (unlikely(!insn)) {
4380 		kvm_queue_exception(vcpu, UD_VECTOR);
4381 		return false;
4382 	}
4383 
4384 	/*
4385 	 * Emulate for SEV guests if the insn buffer is not empty.  The buffer
4386 	 * will be empty if the DecodeAssist microcode cannot fetch bytes for
4387 	 * the faulting instruction because the code fetch itself faulted, e.g.
4388 	 * the guest attempted to fetch from emulated MMIO or a guest page
4389 	 * table used to translate CS:RIP resides in emulated MMIO.
4390 	 */
4391 	if (likely(insn_len))
4392 		return true;
4393 
4394 	/*
4395 	 * Detect and workaround Errata 1096 Fam_17h_00_0Fh.
4396 	 *
4397 	 * Errata:
4398 	 * When CPU raises #NPF on guest data access and vCPU CR4.SMAP=1, it is
4399 	 * possible that CPU microcode implementing DecodeAssist will fail to
4400 	 * read guest memory at CS:RIP and vmcb.GuestIntrBytes will incorrectly
4401 	 * be '0'.  This happens because microcode reads CS:RIP using a _data_
4402 	 * loap uop with CPL=0 privileges.  If the load hits a SMAP #PF, ucode
4403 	 * gives up and does not fill the instruction bytes buffer.
4404 	 *
4405 	 * As above, KVM reaches this point iff the VM is an SEV guest, the CPU
4406 	 * supports DecodeAssist, a #NPF was raised, KVM's page fault handler
4407 	 * triggered emulation (e.g. for MMIO), and the CPU returned 0 in the
4408 	 * GuestIntrBytes field of the VMCB.
4409 	 *
4410 	 * This does _not_ mean that the erratum has been encountered, as the
4411 	 * DecodeAssist will also fail if the load for CS:RIP hits a legitimate
4412 	 * #PF, e.g. if the guest attempt to execute from emulated MMIO and
4413 	 * encountered a reserved/not-present #PF.
4414 	 *
4415 	 * To hit the erratum, the following conditions must be true:
4416 	 *    1. CR4.SMAP=1 (obviously).
4417 	 *    2. CR4.SMEP=0 || CPL=3.  If SMEP=1 and CPL<3, the erratum cannot
4418 	 *       have been hit as the guest would have encountered a SMEP
4419 	 *       violation #PF, not a #NPF.
4420 	 *    3. The #NPF is not due to a code fetch, in which case failure to
4421 	 *       retrieve the instruction bytes is legitimate (see abvoe).
4422 	 *
4423 	 * In addition, don't apply the erratum workaround if the #NPF occurred
4424 	 * while translating guest page tables (see below).
4425 	 */
4426 	error_code = to_svm(vcpu)->vmcb->control.exit_info_1;
4427 	if (error_code & (PFERR_GUEST_PAGE_MASK | PFERR_FETCH_MASK))
4428 		goto resume_guest;
4429 
4430 	cr4 = kvm_read_cr4(vcpu);
4431 	smep = cr4 & X86_CR4_SMEP;
4432 	smap = cr4 & X86_CR4_SMAP;
4433 	is_user = svm_get_cpl(vcpu) == 3;
4434 	if (smap && (!smep || is_user)) {
4435 		pr_err_ratelimited("KVM: SEV Guest triggered AMD Erratum 1096\n");
4436 
4437 		/*
4438 		 * If the fault occurred in userspace, arbitrarily inject #GP
4439 		 * to avoid killing the guest and to hopefully avoid confusing
4440 		 * the guest kernel too much, e.g. injecting #PF would not be
4441 		 * coherent with respect to the guest's page tables.  Request
4442 		 * triple fault if the fault occurred in the kernel as there's
4443 		 * no fault that KVM can inject without confusing the guest.
4444 		 * In practice, the triple fault is moot as no sane SEV kernel
4445 		 * will execute from user memory while also running with SMAP=1.
4446 		 */
4447 		if (is_user)
4448 			kvm_inject_gp(vcpu, 0);
4449 		else
4450 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4451 	}
4452 
4453 resume_guest:
4454 	/*
4455 	 * If the erratum was not hit, simply resume the guest and let it fault
4456 	 * again.  While awful, e.g. the vCPU may get stuck in an infinite loop
4457 	 * if the fault is at CPL=0, it's the lesser of all evils.  Exiting to
4458 	 * userspace will kill the guest, and letting the emulator read garbage
4459 	 * will yield random behavior and potentially corrupt the guest.
4460 	 *
4461 	 * Simply resuming the guest is technically not a violation of the SEV
4462 	 * architecture.  AMD's APM states that all code fetches and page table
4463 	 * accesses for SEV guest are encrypted, regardless of the C-Bit.  The
4464 	 * APM also states that encrypted accesses to MMIO are "ignored", but
4465 	 * doesn't explicitly define "ignored", i.e. doing nothing and letting
4466 	 * the guest spin is technically "ignoring" the access.
4467 	 */
4468 	return false;
4469 }
4470 
4471 static bool svm_apic_init_signal_blocked(struct kvm_vcpu *vcpu)
4472 {
4473 	struct vcpu_svm *svm = to_svm(vcpu);
4474 
4475 	/*
4476 	 * TODO: Last condition latch INIT signals on vCPU when
4477 	 * vCPU is in guest-mode and vmcb12 defines intercept on INIT.
4478 	 * To properly emulate the INIT intercept,
4479 	 * svm_check_nested_events() should call nested_svm_vmexit()
4480 	 * if an INIT signal is pending.
4481 	 */
4482 	return !gif_set(svm) ||
4483 		   (vmcb_is_intercept(&svm->vmcb->control, INTERCEPT_INIT));
4484 }
4485 
4486 static void svm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
4487 {
4488 	if (!sev_es_guest(vcpu->kvm))
4489 		return kvm_vcpu_deliver_sipi_vector(vcpu, vector);
4490 
4491 	sev_vcpu_deliver_sipi_vector(vcpu, vector);
4492 }
4493 
4494 static void svm_vm_destroy(struct kvm *kvm)
4495 {
4496 	avic_vm_destroy(kvm);
4497 	sev_vm_destroy(kvm);
4498 }
4499 
4500 static int svm_vm_init(struct kvm *kvm)
4501 {
4502 	if (!pause_filter_count || !pause_filter_thresh)
4503 		kvm->arch.pause_in_guest = true;
4504 
4505 	if (enable_apicv) {
4506 		int ret = avic_vm_init(kvm);
4507 		if (ret)
4508 			return ret;
4509 	}
4510 
4511 	return 0;
4512 }
4513 
4514 static struct kvm_x86_ops svm_x86_ops __initdata = {
4515 	.name = "kvm_amd",
4516 
4517 	.hardware_unsetup = svm_hardware_teardown,
4518 	.hardware_enable = svm_hardware_enable,
4519 	.hardware_disable = svm_hardware_disable,
4520 	.cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
4521 	.has_emulated_msr = svm_has_emulated_msr,
4522 
4523 	.vcpu_create = svm_create_vcpu,
4524 	.vcpu_free = svm_free_vcpu,
4525 	.vcpu_reset = svm_vcpu_reset,
4526 
4527 	.vm_size = sizeof(struct kvm_svm),
4528 	.vm_init = svm_vm_init,
4529 	.vm_destroy = svm_vm_destroy,
4530 
4531 	.prepare_guest_switch = svm_prepare_guest_switch,
4532 	.vcpu_load = svm_vcpu_load,
4533 	.vcpu_put = svm_vcpu_put,
4534 	.vcpu_blocking = avic_vcpu_blocking,
4535 	.vcpu_unblocking = avic_vcpu_unblocking,
4536 
4537 	.update_exception_bitmap = svm_update_exception_bitmap,
4538 	.get_msr_feature = svm_get_msr_feature,
4539 	.get_msr = svm_get_msr,
4540 	.set_msr = svm_set_msr,
4541 	.get_segment_base = svm_get_segment_base,
4542 	.get_segment = svm_get_segment,
4543 	.set_segment = svm_set_segment,
4544 	.get_cpl = svm_get_cpl,
4545 	.get_cs_db_l_bits = kvm_get_cs_db_l_bits,
4546 	.set_cr0 = svm_set_cr0,
4547 	.post_set_cr3 = svm_post_set_cr3,
4548 	.is_valid_cr4 = svm_is_valid_cr4,
4549 	.set_cr4 = svm_set_cr4,
4550 	.set_efer = svm_set_efer,
4551 	.get_idt = svm_get_idt,
4552 	.set_idt = svm_set_idt,
4553 	.get_gdt = svm_get_gdt,
4554 	.set_gdt = svm_set_gdt,
4555 	.set_dr7 = svm_set_dr7,
4556 	.sync_dirty_debug_regs = svm_sync_dirty_debug_regs,
4557 	.cache_reg = svm_cache_reg,
4558 	.get_rflags = svm_get_rflags,
4559 	.set_rflags = svm_set_rflags,
4560 	.get_if_flag = svm_get_if_flag,
4561 
4562 	.tlb_flush_all = svm_flush_tlb,
4563 	.tlb_flush_current = svm_flush_tlb,
4564 	.tlb_flush_gva = svm_flush_tlb_gva,
4565 	.tlb_flush_guest = svm_flush_tlb,
4566 
4567 	.vcpu_pre_run = svm_vcpu_pre_run,
4568 	.run = svm_vcpu_run,
4569 	.handle_exit = handle_exit,
4570 	.skip_emulated_instruction = skip_emulated_instruction,
4571 	.update_emulated_instruction = NULL,
4572 	.set_interrupt_shadow = svm_set_interrupt_shadow,
4573 	.get_interrupt_shadow = svm_get_interrupt_shadow,
4574 	.patch_hypercall = svm_patch_hypercall,
4575 	.set_irq = svm_set_irq,
4576 	.set_nmi = svm_inject_nmi,
4577 	.queue_exception = svm_queue_exception,
4578 	.cancel_injection = svm_cancel_injection,
4579 	.interrupt_allowed = svm_interrupt_allowed,
4580 	.nmi_allowed = svm_nmi_allowed,
4581 	.get_nmi_mask = svm_get_nmi_mask,
4582 	.set_nmi_mask = svm_set_nmi_mask,
4583 	.enable_nmi_window = svm_enable_nmi_window,
4584 	.enable_irq_window = svm_enable_irq_window,
4585 	.update_cr8_intercept = svm_update_cr8_intercept,
4586 	.set_virtual_apic_mode = svm_set_virtual_apic_mode,
4587 	.refresh_apicv_exec_ctrl = svm_refresh_apicv_exec_ctrl,
4588 	.check_apicv_inhibit_reasons = svm_check_apicv_inhibit_reasons,
4589 	.load_eoi_exitmap = svm_load_eoi_exitmap,
4590 	.hwapic_irr_update = svm_hwapic_irr_update,
4591 	.hwapic_isr_update = svm_hwapic_isr_update,
4592 	.apicv_post_state_restore = avic_post_state_restore,
4593 
4594 	.set_tss_addr = svm_set_tss_addr,
4595 	.set_identity_map_addr = svm_set_identity_map_addr,
4596 	.get_mt_mask = svm_get_mt_mask,
4597 
4598 	.get_exit_info = svm_get_exit_info,
4599 
4600 	.vcpu_after_set_cpuid = svm_vcpu_after_set_cpuid,
4601 
4602 	.has_wbinvd_exit = svm_has_wbinvd_exit,
4603 
4604 	.get_l2_tsc_offset = svm_get_l2_tsc_offset,
4605 	.get_l2_tsc_multiplier = svm_get_l2_tsc_multiplier,
4606 	.write_tsc_offset = svm_write_tsc_offset,
4607 	.write_tsc_multiplier = svm_write_tsc_multiplier,
4608 
4609 	.load_mmu_pgd = svm_load_mmu_pgd,
4610 
4611 	.check_intercept = svm_check_intercept,
4612 	.handle_exit_irqoff = svm_handle_exit_irqoff,
4613 
4614 	.request_immediate_exit = __kvm_request_immediate_exit,
4615 
4616 	.sched_in = svm_sched_in,
4617 
4618 	.pmu_ops = &amd_pmu_ops,
4619 	.nested_ops = &svm_nested_ops,
4620 
4621 	.deliver_interrupt = svm_deliver_interrupt,
4622 	.dy_apicv_has_pending_interrupt = svm_dy_apicv_has_pending_interrupt,
4623 	.update_pi_irte = svm_update_pi_irte,
4624 	.setup_mce = svm_setup_mce,
4625 
4626 	.smi_allowed = svm_smi_allowed,
4627 	.enter_smm = svm_enter_smm,
4628 	.leave_smm = svm_leave_smm,
4629 	.enable_smi_window = svm_enable_smi_window,
4630 
4631 	.mem_enc_op = svm_mem_enc_op,
4632 	.mem_enc_reg_region = svm_register_enc_region,
4633 	.mem_enc_unreg_region = svm_unregister_enc_region,
4634 
4635 	.vm_copy_enc_context_from = svm_vm_copy_asid_from,
4636 	.vm_move_enc_context_from = svm_vm_migrate_from,
4637 
4638 	.can_emulate_instruction = svm_can_emulate_instruction,
4639 
4640 	.apic_init_signal_blocked = svm_apic_init_signal_blocked,
4641 
4642 	.msr_filter_changed = svm_msr_filter_changed,
4643 	.complete_emulated_msr = svm_complete_emulated_msr,
4644 
4645 	.vcpu_deliver_sipi_vector = svm_vcpu_deliver_sipi_vector,
4646 };
4647 
4648 /*
4649  * The default MMIO mask is a single bit (excluding the present bit),
4650  * which could conflict with the memory encryption bit. Check for
4651  * memory encryption support and override the default MMIO mask if
4652  * memory encryption is enabled.
4653  */
4654 static __init void svm_adjust_mmio_mask(void)
4655 {
4656 	unsigned int enc_bit, mask_bit;
4657 	u64 msr, mask;
4658 
4659 	/* If there is no memory encryption support, use existing mask */
4660 	if (cpuid_eax(0x80000000) < 0x8000001f)
4661 		return;
4662 
4663 	/* If memory encryption is not enabled, use existing mask */
4664 	rdmsrl(MSR_AMD64_SYSCFG, msr);
4665 	if (!(msr & MSR_AMD64_SYSCFG_MEM_ENCRYPT))
4666 		return;
4667 
4668 	enc_bit = cpuid_ebx(0x8000001f) & 0x3f;
4669 	mask_bit = boot_cpu_data.x86_phys_bits;
4670 
4671 	/* Increment the mask bit if it is the same as the encryption bit */
4672 	if (enc_bit == mask_bit)
4673 		mask_bit++;
4674 
4675 	/*
4676 	 * If the mask bit location is below 52, then some bits above the
4677 	 * physical addressing limit will always be reserved, so use the
4678 	 * rsvd_bits() function to generate the mask. This mask, along with
4679 	 * the present bit, will be used to generate a page fault with
4680 	 * PFER.RSV = 1.
4681 	 *
4682 	 * If the mask bit location is 52 (or above), then clear the mask.
4683 	 */
4684 	mask = (mask_bit < 52) ? rsvd_bits(mask_bit, 51) | PT_PRESENT_MASK : 0;
4685 
4686 	kvm_mmu_set_mmio_spte_mask(mask, mask, PT_WRITABLE_MASK | PT_USER_MASK);
4687 }
4688 
4689 static __init void svm_set_cpu_caps(void)
4690 {
4691 	kvm_set_cpu_caps();
4692 
4693 	supported_xss = 0;
4694 
4695 	/* CPUID 0x80000001 and 0x8000000A (SVM features) */
4696 	if (nested) {
4697 		kvm_cpu_cap_set(X86_FEATURE_SVM);
4698 		kvm_cpu_cap_set(X86_FEATURE_VMCBCLEAN);
4699 
4700 		if (nrips)
4701 			kvm_cpu_cap_set(X86_FEATURE_NRIPS);
4702 
4703 		if (npt_enabled)
4704 			kvm_cpu_cap_set(X86_FEATURE_NPT);
4705 
4706 		if (tsc_scaling)
4707 			kvm_cpu_cap_set(X86_FEATURE_TSCRATEMSR);
4708 
4709 		/* Nested VM can receive #VMEXIT instead of triggering #GP */
4710 		kvm_cpu_cap_set(X86_FEATURE_SVME_ADDR_CHK);
4711 	}
4712 
4713 	/* CPUID 0x80000008 */
4714 	if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD) ||
4715 	    boot_cpu_has(X86_FEATURE_AMD_SSBD))
4716 		kvm_cpu_cap_set(X86_FEATURE_VIRT_SSBD);
4717 
4718 	/* AMD PMU PERFCTR_CORE CPUID */
4719 	if (enable_pmu && boot_cpu_has(X86_FEATURE_PERFCTR_CORE))
4720 		kvm_cpu_cap_set(X86_FEATURE_PERFCTR_CORE);
4721 
4722 	/* CPUID 0x8000001F (SME/SEV features) */
4723 	sev_set_cpu_caps();
4724 }
4725 
4726 static __init int svm_hardware_setup(void)
4727 {
4728 	int cpu;
4729 	struct page *iopm_pages;
4730 	void *iopm_va;
4731 	int r;
4732 	unsigned int order = get_order(IOPM_SIZE);
4733 
4734 	/*
4735 	 * NX is required for shadow paging and for NPT if the NX huge pages
4736 	 * mitigation is enabled.
4737 	 */
4738 	if (!boot_cpu_has(X86_FEATURE_NX)) {
4739 		pr_err_ratelimited("NX (Execute Disable) not supported\n");
4740 		return -EOPNOTSUPP;
4741 	}
4742 	kvm_enable_efer_bits(EFER_NX);
4743 
4744 	iopm_pages = alloc_pages(GFP_KERNEL, order);
4745 
4746 	if (!iopm_pages)
4747 		return -ENOMEM;
4748 
4749 	iopm_va = page_address(iopm_pages);
4750 	memset(iopm_va, 0xff, PAGE_SIZE * (1 << order));
4751 	iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
4752 
4753 	init_msrpm_offsets();
4754 
4755 	supported_xcr0 &= ~(XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR);
4756 
4757 	if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
4758 		kvm_enable_efer_bits(EFER_FFXSR);
4759 
4760 	if (tsc_scaling) {
4761 		if (!boot_cpu_has(X86_FEATURE_TSCRATEMSR)) {
4762 			tsc_scaling = false;
4763 		} else {
4764 			pr_info("TSC scaling supported\n");
4765 			kvm_has_tsc_control = true;
4766 			kvm_max_tsc_scaling_ratio = TSC_RATIO_MAX;
4767 			kvm_tsc_scaling_ratio_frac_bits = 32;
4768 		}
4769 	}
4770 
4771 	tsc_aux_uret_slot = kvm_add_user_return_msr(MSR_TSC_AUX);
4772 
4773 	/* Check for pause filtering support */
4774 	if (!boot_cpu_has(X86_FEATURE_PAUSEFILTER)) {
4775 		pause_filter_count = 0;
4776 		pause_filter_thresh = 0;
4777 	} else if (!boot_cpu_has(X86_FEATURE_PFTHRESHOLD)) {
4778 		pause_filter_thresh = 0;
4779 	}
4780 
4781 	if (nested) {
4782 		printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
4783 		kvm_enable_efer_bits(EFER_SVME | EFER_LMSLE);
4784 	}
4785 
4786 	/*
4787 	 * KVM's MMU doesn't support using 2-level paging for itself, and thus
4788 	 * NPT isn't supported if the host is using 2-level paging since host
4789 	 * CR4 is unchanged on VMRUN.
4790 	 */
4791 	if (!IS_ENABLED(CONFIG_X86_64) && !IS_ENABLED(CONFIG_X86_PAE))
4792 		npt_enabled = false;
4793 
4794 	if (!boot_cpu_has(X86_FEATURE_NPT))
4795 		npt_enabled = false;
4796 
4797 	/* Force VM NPT level equal to the host's paging level */
4798 	kvm_configure_mmu(npt_enabled, get_npt_level(),
4799 			  get_npt_level(), PG_LEVEL_1G);
4800 	pr_info("kvm: Nested Paging %sabled\n", npt_enabled ? "en" : "dis");
4801 
4802 	/* Note, SEV setup consumes npt_enabled. */
4803 	sev_hardware_setup();
4804 
4805 	svm_hv_hardware_setup();
4806 
4807 	svm_adjust_mmio_mask();
4808 
4809 	for_each_possible_cpu(cpu) {
4810 		r = svm_cpu_init(cpu);
4811 		if (r)
4812 			goto err;
4813 	}
4814 
4815 	if (nrips) {
4816 		if (!boot_cpu_has(X86_FEATURE_NRIPS))
4817 			nrips = false;
4818 	}
4819 
4820 	enable_apicv = avic = avic && npt_enabled && boot_cpu_has(X86_FEATURE_AVIC);
4821 
4822 	if (enable_apicv) {
4823 		pr_info("AVIC enabled\n");
4824 
4825 		amd_iommu_register_ga_log_notifier(&avic_ga_log_notifier);
4826 	} else {
4827 		svm_x86_ops.vcpu_blocking = NULL;
4828 		svm_x86_ops.vcpu_unblocking = NULL;
4829 	}
4830 
4831 	if (vls) {
4832 		if (!npt_enabled ||
4833 		    !boot_cpu_has(X86_FEATURE_V_VMSAVE_VMLOAD) ||
4834 		    !IS_ENABLED(CONFIG_X86_64)) {
4835 			vls = false;
4836 		} else {
4837 			pr_info("Virtual VMLOAD VMSAVE supported\n");
4838 		}
4839 	}
4840 
4841 	if (boot_cpu_has(X86_FEATURE_SVME_ADDR_CHK))
4842 		svm_gp_erratum_intercept = false;
4843 
4844 	if (vgif) {
4845 		if (!boot_cpu_has(X86_FEATURE_VGIF))
4846 			vgif = false;
4847 		else
4848 			pr_info("Virtual GIF supported\n");
4849 	}
4850 
4851 	if (lbrv) {
4852 		if (!boot_cpu_has(X86_FEATURE_LBRV))
4853 			lbrv = false;
4854 		else
4855 			pr_info("LBR virtualization supported\n");
4856 	}
4857 
4858 	if (!enable_pmu)
4859 		pr_info("PMU virtualization is disabled\n");
4860 
4861 	svm_set_cpu_caps();
4862 
4863 	/*
4864 	 * It seems that on AMD processors PTE's accessed bit is
4865 	 * being set by the CPU hardware before the NPF vmexit.
4866 	 * This is not expected behaviour and our tests fail because
4867 	 * of it.
4868 	 * A workaround here is to disable support for
4869 	 * GUEST_MAXPHYADDR < HOST_MAXPHYADDR if NPT is enabled.
4870 	 * In this case userspace can know if there is support using
4871 	 * KVM_CAP_SMALLER_MAXPHYADDR extension and decide how to handle
4872 	 * it
4873 	 * If future AMD CPU models change the behaviour described above,
4874 	 * this variable can be changed accordingly
4875 	 */
4876 	allow_smaller_maxphyaddr = !npt_enabled;
4877 
4878 	return 0;
4879 
4880 err:
4881 	svm_hardware_teardown();
4882 	return r;
4883 }
4884 
4885 
4886 static struct kvm_x86_init_ops svm_init_ops __initdata = {
4887 	.cpu_has_kvm_support = has_svm,
4888 	.disabled_by_bios = is_disabled,
4889 	.hardware_setup = svm_hardware_setup,
4890 	.check_processor_compatibility = svm_check_processor_compat,
4891 
4892 	.runtime_ops = &svm_x86_ops,
4893 };
4894 
4895 static int __init svm_init(void)
4896 {
4897 	__unused_size_checks();
4898 
4899 	return kvm_init(&svm_init_ops, sizeof(struct vcpu_svm),
4900 			__alignof__(struct vcpu_svm), THIS_MODULE);
4901 }
4902 
4903 static void __exit svm_exit(void)
4904 {
4905 	kvm_exit();
4906 }
4907 
4908 module_init(svm_init)
4909 module_exit(svm_exit)
4910