xref: /openbmc/linux/arch/x86/kvm/svm/svm.c (revision 10756dc5)
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(svm);
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 			if (svm_gp_erratum_intercept)
316 				set_exception_intercept(svm, GP_VECTOR);
317 		}
318 	}
319 
320 	svm->vmcb->save.efer = efer | EFER_SVME;
321 	vmcb_mark_dirty(svm->vmcb, VMCB_CR);
322 	return 0;
323 }
324 
325 static int is_external_interrupt(u32 info)
326 {
327 	info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
328 	return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
329 }
330 
331 static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu)
332 {
333 	struct vcpu_svm *svm = to_svm(vcpu);
334 	u32 ret = 0;
335 
336 	if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)
337 		ret = KVM_X86_SHADOW_INT_STI | KVM_X86_SHADOW_INT_MOV_SS;
338 	return ret;
339 }
340 
341 static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
342 {
343 	struct vcpu_svm *svm = to_svm(vcpu);
344 
345 	if (mask == 0)
346 		svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
347 	else
348 		svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK;
349 
350 }
351 
352 static int skip_emulated_instruction(struct kvm_vcpu *vcpu)
353 {
354 	struct vcpu_svm *svm = to_svm(vcpu);
355 
356 	/*
357 	 * SEV-ES does not expose the next RIP. The RIP update is controlled by
358 	 * the type of exit and the #VC handler in the guest.
359 	 */
360 	if (sev_es_guest(vcpu->kvm))
361 		goto done;
362 
363 	if (nrips && svm->vmcb->control.next_rip != 0) {
364 		WARN_ON_ONCE(!static_cpu_has(X86_FEATURE_NRIPS));
365 		svm->next_rip = svm->vmcb->control.next_rip;
366 	}
367 
368 	if (!svm->next_rip) {
369 		if (!kvm_emulate_instruction(vcpu, EMULTYPE_SKIP))
370 			return 0;
371 	} else {
372 		kvm_rip_write(vcpu, svm->next_rip);
373 	}
374 
375 done:
376 	svm_set_interrupt_shadow(vcpu, 0);
377 
378 	return 1;
379 }
380 
381 static void svm_queue_exception(struct kvm_vcpu *vcpu)
382 {
383 	struct vcpu_svm *svm = to_svm(vcpu);
384 	unsigned nr = vcpu->arch.exception.nr;
385 	bool has_error_code = vcpu->arch.exception.has_error_code;
386 	u32 error_code = vcpu->arch.exception.error_code;
387 
388 	kvm_deliver_exception_payload(vcpu);
389 
390 	if (nr == BP_VECTOR && !nrips) {
391 		unsigned long rip, old_rip = kvm_rip_read(vcpu);
392 
393 		/*
394 		 * For guest debugging where we have to reinject #BP if some
395 		 * INT3 is guest-owned:
396 		 * Emulate nRIP by moving RIP forward. Will fail if injection
397 		 * raises a fault that is not intercepted. Still better than
398 		 * failing in all cases.
399 		 */
400 		(void)skip_emulated_instruction(vcpu);
401 		rip = kvm_rip_read(vcpu);
402 		svm->int3_rip = rip + svm->vmcb->save.cs.base;
403 		svm->int3_injected = rip - old_rip;
404 	}
405 
406 	svm->vmcb->control.event_inj = nr
407 		| SVM_EVTINJ_VALID
408 		| (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
409 		| SVM_EVTINJ_TYPE_EXEPT;
410 	svm->vmcb->control.event_inj_err = error_code;
411 }
412 
413 static void svm_init_erratum_383(void)
414 {
415 	u32 low, high;
416 	int err;
417 	u64 val;
418 
419 	if (!static_cpu_has_bug(X86_BUG_AMD_TLB_MMATCH))
420 		return;
421 
422 	/* Use _safe variants to not break nested virtualization */
423 	val = native_read_msr_safe(MSR_AMD64_DC_CFG, &err);
424 	if (err)
425 		return;
426 
427 	val |= (1ULL << 47);
428 
429 	low  = lower_32_bits(val);
430 	high = upper_32_bits(val);
431 
432 	native_write_msr_safe(MSR_AMD64_DC_CFG, low, high);
433 
434 	erratum_383_found = true;
435 }
436 
437 static void svm_init_osvw(struct kvm_vcpu *vcpu)
438 {
439 	/*
440 	 * Guests should see errata 400 and 415 as fixed (assuming that
441 	 * HLT and IO instructions are intercepted).
442 	 */
443 	vcpu->arch.osvw.length = (osvw_len >= 3) ? (osvw_len) : 3;
444 	vcpu->arch.osvw.status = osvw_status & ~(6ULL);
445 
446 	/*
447 	 * By increasing VCPU's osvw.length to 3 we are telling the guest that
448 	 * all osvw.status bits inside that length, including bit 0 (which is
449 	 * reserved for erratum 298), are valid. However, if host processor's
450 	 * osvw_len is 0 then osvw_status[0] carries no information. We need to
451 	 * be conservative here and therefore we tell the guest that erratum 298
452 	 * is present (because we really don't know).
453 	 */
454 	if (osvw_len == 0 && boot_cpu_data.x86 == 0x10)
455 		vcpu->arch.osvw.status |= 1;
456 }
457 
458 static int has_svm(void)
459 {
460 	const char *msg;
461 
462 	if (!cpu_has_svm(&msg)) {
463 		printk(KERN_INFO "has_svm: %s\n", msg);
464 		return 0;
465 	}
466 
467 	if (cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT)) {
468 		pr_info("KVM is unsupported when running as an SEV guest\n");
469 		return 0;
470 	}
471 
472 	return 1;
473 }
474 
475 static void svm_hardware_disable(void)
476 {
477 	/* Make sure we clean up behind us */
478 	if (tsc_scaling)
479 		wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
480 
481 	cpu_svm_disable();
482 
483 	amd_pmu_disable_virt();
484 }
485 
486 static int svm_hardware_enable(void)
487 {
488 
489 	struct svm_cpu_data *sd;
490 	uint64_t efer;
491 	struct desc_struct *gdt;
492 	int me = raw_smp_processor_id();
493 
494 	rdmsrl(MSR_EFER, efer);
495 	if (efer & EFER_SVME)
496 		return -EBUSY;
497 
498 	if (!has_svm()) {
499 		pr_err("%s: err EOPNOTSUPP on %d\n", __func__, me);
500 		return -EINVAL;
501 	}
502 	sd = per_cpu(svm_data, me);
503 	if (!sd) {
504 		pr_err("%s: svm_data is NULL on %d\n", __func__, me);
505 		return -EINVAL;
506 	}
507 
508 	sd->asid_generation = 1;
509 	sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
510 	sd->next_asid = sd->max_asid + 1;
511 	sd->min_asid = max_sev_asid + 1;
512 
513 	gdt = get_current_gdt_rw();
514 	sd->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
515 
516 	wrmsrl(MSR_EFER, efer | EFER_SVME);
517 
518 	wrmsrl(MSR_VM_HSAVE_PA, __sme_page_pa(sd->save_area));
519 
520 	if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
521 		/*
522 		 * Set the default value, even if we don't use TSC scaling
523 		 * to avoid having stale value in the msr
524 		 */
525 		wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
526 		__this_cpu_write(current_tsc_ratio, TSC_RATIO_DEFAULT);
527 	}
528 
529 
530 	/*
531 	 * Get OSVW bits.
532 	 *
533 	 * Note that it is possible to have a system with mixed processor
534 	 * revisions and therefore different OSVW bits. If bits are not the same
535 	 * on different processors then choose the worst case (i.e. if erratum
536 	 * is present on one processor and not on another then assume that the
537 	 * erratum is present everywhere).
538 	 */
539 	if (cpu_has(&boot_cpu_data, X86_FEATURE_OSVW)) {
540 		uint64_t len, status = 0;
541 		int err;
542 
543 		len = native_read_msr_safe(MSR_AMD64_OSVW_ID_LENGTH, &err);
544 		if (!err)
545 			status = native_read_msr_safe(MSR_AMD64_OSVW_STATUS,
546 						      &err);
547 
548 		if (err)
549 			osvw_status = osvw_len = 0;
550 		else {
551 			if (len < osvw_len)
552 				osvw_len = len;
553 			osvw_status |= status;
554 			osvw_status &= (1ULL << osvw_len) - 1;
555 		}
556 	} else
557 		osvw_status = osvw_len = 0;
558 
559 	svm_init_erratum_383();
560 
561 	amd_pmu_enable_virt();
562 
563 	return 0;
564 }
565 
566 static void svm_cpu_uninit(int cpu)
567 {
568 	struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
569 
570 	if (!sd)
571 		return;
572 
573 	per_cpu(svm_data, cpu) = NULL;
574 	kfree(sd->sev_vmcbs);
575 	__free_page(sd->save_area);
576 	kfree(sd);
577 }
578 
579 static int svm_cpu_init(int cpu)
580 {
581 	struct svm_cpu_data *sd;
582 	int ret = -ENOMEM;
583 
584 	sd = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
585 	if (!sd)
586 		return ret;
587 	sd->cpu = cpu;
588 	sd->save_area = alloc_page(GFP_KERNEL | __GFP_ZERO);
589 	if (!sd->save_area)
590 		goto free_cpu_data;
591 
592 	ret = sev_cpu_init(sd);
593 	if (ret)
594 		goto free_save_area;
595 
596 	per_cpu(svm_data, cpu) = sd;
597 
598 	return 0;
599 
600 free_save_area:
601 	__free_page(sd->save_area);
602 free_cpu_data:
603 	kfree(sd);
604 	return ret;
605 
606 }
607 
608 static int direct_access_msr_slot(u32 msr)
609 {
610 	u32 i;
611 
612 	for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++)
613 		if (direct_access_msrs[i].index == msr)
614 			return i;
615 
616 	return -ENOENT;
617 }
618 
619 static void set_shadow_msr_intercept(struct kvm_vcpu *vcpu, u32 msr, int read,
620 				     int write)
621 {
622 	struct vcpu_svm *svm = to_svm(vcpu);
623 	int slot = direct_access_msr_slot(msr);
624 
625 	if (slot == -ENOENT)
626 		return;
627 
628 	/* Set the shadow bitmaps to the desired intercept states */
629 	if (read)
630 		set_bit(slot, svm->shadow_msr_intercept.read);
631 	else
632 		clear_bit(slot, svm->shadow_msr_intercept.read);
633 
634 	if (write)
635 		set_bit(slot, svm->shadow_msr_intercept.write);
636 	else
637 		clear_bit(slot, svm->shadow_msr_intercept.write);
638 }
639 
640 static bool valid_msr_intercept(u32 index)
641 {
642 	return direct_access_msr_slot(index) != -ENOENT;
643 }
644 
645 static bool msr_write_intercepted(struct kvm_vcpu *vcpu, u32 msr)
646 {
647 	u8 bit_write;
648 	unsigned long tmp;
649 	u32 offset;
650 	u32 *msrpm;
651 
652 	msrpm = is_guest_mode(vcpu) ? to_svm(vcpu)->nested.msrpm:
653 				      to_svm(vcpu)->msrpm;
654 
655 	offset    = svm_msrpm_offset(msr);
656 	bit_write = 2 * (msr & 0x0f) + 1;
657 	tmp       = msrpm[offset];
658 
659 	BUG_ON(offset == MSR_INVALID);
660 
661 	return !!test_bit(bit_write,  &tmp);
662 }
663 
664 static void set_msr_interception_bitmap(struct kvm_vcpu *vcpu, u32 *msrpm,
665 					u32 msr, int read, int write)
666 {
667 	u8 bit_read, bit_write;
668 	unsigned long tmp;
669 	u32 offset;
670 
671 	/*
672 	 * If this warning triggers extend the direct_access_msrs list at the
673 	 * beginning of the file
674 	 */
675 	WARN_ON(!valid_msr_intercept(msr));
676 
677 	/* Enforce non allowed MSRs to trap */
678 	if (read && !kvm_msr_allowed(vcpu, msr, KVM_MSR_FILTER_READ))
679 		read = 0;
680 
681 	if (write && !kvm_msr_allowed(vcpu, msr, KVM_MSR_FILTER_WRITE))
682 		write = 0;
683 
684 	offset    = svm_msrpm_offset(msr);
685 	bit_read  = 2 * (msr & 0x0f);
686 	bit_write = 2 * (msr & 0x0f) + 1;
687 	tmp       = msrpm[offset];
688 
689 	BUG_ON(offset == MSR_INVALID);
690 
691 	read  ? clear_bit(bit_read,  &tmp) : set_bit(bit_read,  &tmp);
692 	write ? clear_bit(bit_write, &tmp) : set_bit(bit_write, &tmp);
693 
694 	msrpm[offset] = tmp;
695 
696 	svm_hv_vmcb_dirty_nested_enlightenments(vcpu);
697 
698 }
699 
700 void set_msr_interception(struct kvm_vcpu *vcpu, u32 *msrpm, u32 msr,
701 			  int read, int write)
702 {
703 	set_shadow_msr_intercept(vcpu, msr, read, write);
704 	set_msr_interception_bitmap(vcpu, msrpm, msr, read, write);
705 }
706 
707 u32 *svm_vcpu_alloc_msrpm(void)
708 {
709 	unsigned int order = get_order(MSRPM_SIZE);
710 	struct page *pages = alloc_pages(GFP_KERNEL_ACCOUNT, order);
711 	u32 *msrpm;
712 
713 	if (!pages)
714 		return NULL;
715 
716 	msrpm = page_address(pages);
717 	memset(msrpm, 0xff, PAGE_SIZE * (1 << order));
718 
719 	return msrpm;
720 }
721 
722 void svm_vcpu_init_msrpm(struct kvm_vcpu *vcpu, u32 *msrpm)
723 {
724 	int i;
725 
726 	for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
727 		if (!direct_access_msrs[i].always)
728 			continue;
729 		set_msr_interception(vcpu, msrpm, direct_access_msrs[i].index, 1, 1);
730 	}
731 }
732 
733 
734 void svm_vcpu_free_msrpm(u32 *msrpm)
735 {
736 	__free_pages(virt_to_page(msrpm), get_order(MSRPM_SIZE));
737 }
738 
739 static void svm_msr_filter_changed(struct kvm_vcpu *vcpu)
740 {
741 	struct vcpu_svm *svm = to_svm(vcpu);
742 	u32 i;
743 
744 	/*
745 	 * Set intercept permissions for all direct access MSRs again. They
746 	 * will automatically get filtered through the MSR filter, so we are
747 	 * back in sync after this.
748 	 */
749 	for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
750 		u32 msr = direct_access_msrs[i].index;
751 		u32 read = test_bit(i, svm->shadow_msr_intercept.read);
752 		u32 write = test_bit(i, svm->shadow_msr_intercept.write);
753 
754 		set_msr_interception_bitmap(vcpu, svm->msrpm, msr, read, write);
755 	}
756 }
757 
758 static void add_msr_offset(u32 offset)
759 {
760 	int i;
761 
762 	for (i = 0; i < MSRPM_OFFSETS; ++i) {
763 
764 		/* Offset already in list? */
765 		if (msrpm_offsets[i] == offset)
766 			return;
767 
768 		/* Slot used by another offset? */
769 		if (msrpm_offsets[i] != MSR_INVALID)
770 			continue;
771 
772 		/* Add offset to list */
773 		msrpm_offsets[i] = offset;
774 
775 		return;
776 	}
777 
778 	/*
779 	 * If this BUG triggers the msrpm_offsets table has an overflow. Just
780 	 * increase MSRPM_OFFSETS in this case.
781 	 */
782 	BUG();
783 }
784 
785 static void init_msrpm_offsets(void)
786 {
787 	int i;
788 
789 	memset(msrpm_offsets, 0xff, sizeof(msrpm_offsets));
790 
791 	for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
792 		u32 offset;
793 
794 		offset = svm_msrpm_offset(direct_access_msrs[i].index);
795 		BUG_ON(offset == MSR_INVALID);
796 
797 		add_msr_offset(offset);
798 	}
799 }
800 
801 static void svm_enable_lbrv(struct kvm_vcpu *vcpu)
802 {
803 	struct vcpu_svm *svm = to_svm(vcpu);
804 
805 	svm->vmcb->control.virt_ext |= LBR_CTL_ENABLE_MASK;
806 	set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
807 	set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
808 	set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
809 	set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
810 }
811 
812 static void svm_disable_lbrv(struct kvm_vcpu *vcpu)
813 {
814 	struct vcpu_svm *svm = to_svm(vcpu);
815 
816 	svm->vmcb->control.virt_ext &= ~LBR_CTL_ENABLE_MASK;
817 	set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
818 	set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
819 	set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
820 	set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
821 }
822 
823 void disable_nmi_singlestep(struct vcpu_svm *svm)
824 {
825 	svm->nmi_singlestep = false;
826 
827 	if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP)) {
828 		/* Clear our flags if they were not set by the guest */
829 		if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF))
830 			svm->vmcb->save.rflags &= ~X86_EFLAGS_TF;
831 		if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_RF))
832 			svm->vmcb->save.rflags &= ~X86_EFLAGS_RF;
833 	}
834 }
835 
836 static void grow_ple_window(struct kvm_vcpu *vcpu)
837 {
838 	struct vcpu_svm *svm = to_svm(vcpu);
839 	struct vmcb_control_area *control = &svm->vmcb->control;
840 	int old = control->pause_filter_count;
841 
842 	control->pause_filter_count = __grow_ple_window(old,
843 							pause_filter_count,
844 							pause_filter_count_grow,
845 							pause_filter_count_max);
846 
847 	if (control->pause_filter_count != old) {
848 		vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
849 		trace_kvm_ple_window_update(vcpu->vcpu_id,
850 					    control->pause_filter_count, old);
851 	}
852 }
853 
854 static void shrink_ple_window(struct kvm_vcpu *vcpu)
855 {
856 	struct vcpu_svm *svm = to_svm(vcpu);
857 	struct vmcb_control_area *control = &svm->vmcb->control;
858 	int old = control->pause_filter_count;
859 
860 	control->pause_filter_count =
861 				__shrink_ple_window(old,
862 						    pause_filter_count,
863 						    pause_filter_count_shrink,
864 						    pause_filter_count);
865 	if (control->pause_filter_count != old) {
866 		vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
867 		trace_kvm_ple_window_update(vcpu->vcpu_id,
868 					    control->pause_filter_count, old);
869 	}
870 }
871 
872 static void svm_hardware_teardown(void)
873 {
874 	int cpu;
875 
876 	sev_hardware_teardown();
877 
878 	for_each_possible_cpu(cpu)
879 		svm_cpu_uninit(cpu);
880 
881 	__free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT),
882 	get_order(IOPM_SIZE));
883 	iopm_base = 0;
884 }
885 
886 static void init_seg(struct vmcb_seg *seg)
887 {
888 	seg->selector = 0;
889 	seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
890 		      SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
891 	seg->limit = 0xffff;
892 	seg->base = 0;
893 }
894 
895 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
896 {
897 	seg->selector = 0;
898 	seg->attrib = SVM_SELECTOR_P_MASK | type;
899 	seg->limit = 0xffff;
900 	seg->base = 0;
901 }
902 
903 static u64 svm_get_l2_tsc_offset(struct kvm_vcpu *vcpu)
904 {
905 	struct vcpu_svm *svm = to_svm(vcpu);
906 
907 	return svm->nested.ctl.tsc_offset;
908 }
909 
910 static u64 svm_get_l2_tsc_multiplier(struct kvm_vcpu *vcpu)
911 {
912 	struct vcpu_svm *svm = to_svm(vcpu);
913 
914 	return svm->tsc_ratio_msr;
915 }
916 
917 static void svm_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
918 {
919 	struct vcpu_svm *svm = to_svm(vcpu);
920 
921 	svm->vmcb01.ptr->control.tsc_offset = vcpu->arch.l1_tsc_offset;
922 	svm->vmcb->control.tsc_offset = offset;
923 	vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
924 }
925 
926 void svm_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 multiplier)
927 {
928 	wrmsrl(MSR_AMD64_TSC_RATIO, multiplier);
929 }
930 
931 /* Evaluate instruction intercepts that depend on guest CPUID features. */
932 static void svm_recalc_instruction_intercepts(struct kvm_vcpu *vcpu,
933 					      struct vcpu_svm *svm)
934 {
935 	/*
936 	 * Intercept INVPCID if shadow paging is enabled to sync/free shadow
937 	 * roots, or if INVPCID is disabled in the guest to inject #UD.
938 	 */
939 	if (kvm_cpu_cap_has(X86_FEATURE_INVPCID)) {
940 		if (!npt_enabled ||
941 		    !guest_cpuid_has(&svm->vcpu, X86_FEATURE_INVPCID))
942 			svm_set_intercept(svm, INTERCEPT_INVPCID);
943 		else
944 			svm_clr_intercept(svm, INTERCEPT_INVPCID);
945 	}
946 
947 	if (kvm_cpu_cap_has(X86_FEATURE_RDTSCP)) {
948 		if (guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
949 			svm_clr_intercept(svm, INTERCEPT_RDTSCP);
950 		else
951 			svm_set_intercept(svm, INTERCEPT_RDTSCP);
952 	}
953 }
954 
955 static inline void init_vmcb_after_set_cpuid(struct kvm_vcpu *vcpu)
956 {
957 	struct vcpu_svm *svm = to_svm(vcpu);
958 
959 	if (guest_cpuid_is_intel(vcpu)) {
960 		/*
961 		 * We must intercept SYSENTER_EIP and SYSENTER_ESP
962 		 * accesses because the processor only stores 32 bits.
963 		 * For the same reason we cannot use virtual VMLOAD/VMSAVE.
964 		 */
965 		svm_set_intercept(svm, INTERCEPT_VMLOAD);
966 		svm_set_intercept(svm, INTERCEPT_VMSAVE);
967 		svm->vmcb->control.virt_ext &= ~VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK;
968 
969 		set_msr_interception(vcpu, svm->msrpm, MSR_IA32_SYSENTER_EIP, 0, 0);
970 		set_msr_interception(vcpu, svm->msrpm, MSR_IA32_SYSENTER_ESP, 0, 0);
971 	} else {
972 		/*
973 		 * If hardware supports Virtual VMLOAD VMSAVE then enable it
974 		 * in VMCB and clear intercepts to avoid #VMEXIT.
975 		 */
976 		if (vls) {
977 			svm_clr_intercept(svm, INTERCEPT_VMLOAD);
978 			svm_clr_intercept(svm, INTERCEPT_VMSAVE);
979 			svm->vmcb->control.virt_ext |= VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK;
980 		}
981 		/* No need to intercept these MSRs */
982 		set_msr_interception(vcpu, svm->msrpm, MSR_IA32_SYSENTER_EIP, 1, 1);
983 		set_msr_interception(vcpu, svm->msrpm, MSR_IA32_SYSENTER_ESP, 1, 1);
984 	}
985 }
986 
987 static void init_vmcb(struct kvm_vcpu *vcpu)
988 {
989 	struct vcpu_svm *svm = to_svm(vcpu);
990 	struct vmcb_control_area *control = &svm->vmcb->control;
991 	struct vmcb_save_area *save = &svm->vmcb->save;
992 
993 	svm_set_intercept(svm, INTERCEPT_CR0_READ);
994 	svm_set_intercept(svm, INTERCEPT_CR3_READ);
995 	svm_set_intercept(svm, INTERCEPT_CR4_READ);
996 	svm_set_intercept(svm, INTERCEPT_CR0_WRITE);
997 	svm_set_intercept(svm, INTERCEPT_CR3_WRITE);
998 	svm_set_intercept(svm, INTERCEPT_CR4_WRITE);
999 	if (!kvm_vcpu_apicv_active(vcpu))
1000 		svm_set_intercept(svm, INTERCEPT_CR8_WRITE);
1001 
1002 	set_dr_intercepts(svm);
1003 
1004 	set_exception_intercept(svm, PF_VECTOR);
1005 	set_exception_intercept(svm, UD_VECTOR);
1006 	set_exception_intercept(svm, MC_VECTOR);
1007 	set_exception_intercept(svm, AC_VECTOR);
1008 	set_exception_intercept(svm, DB_VECTOR);
1009 	/*
1010 	 * Guest access to VMware backdoor ports could legitimately
1011 	 * trigger #GP because of TSS I/O permission bitmap.
1012 	 * We intercept those #GP and allow access to them anyway
1013 	 * as VMware does.
1014 	 */
1015 	if (enable_vmware_backdoor)
1016 		set_exception_intercept(svm, GP_VECTOR);
1017 
1018 	svm_set_intercept(svm, INTERCEPT_INTR);
1019 	svm_set_intercept(svm, INTERCEPT_NMI);
1020 
1021 	if (intercept_smi)
1022 		svm_set_intercept(svm, INTERCEPT_SMI);
1023 
1024 	svm_set_intercept(svm, INTERCEPT_SELECTIVE_CR0);
1025 	svm_set_intercept(svm, INTERCEPT_RDPMC);
1026 	svm_set_intercept(svm, INTERCEPT_CPUID);
1027 	svm_set_intercept(svm, INTERCEPT_INVD);
1028 	svm_set_intercept(svm, INTERCEPT_INVLPG);
1029 	svm_set_intercept(svm, INTERCEPT_INVLPGA);
1030 	svm_set_intercept(svm, INTERCEPT_IOIO_PROT);
1031 	svm_set_intercept(svm, INTERCEPT_MSR_PROT);
1032 	svm_set_intercept(svm, INTERCEPT_TASK_SWITCH);
1033 	svm_set_intercept(svm, INTERCEPT_SHUTDOWN);
1034 	svm_set_intercept(svm, INTERCEPT_VMRUN);
1035 	svm_set_intercept(svm, INTERCEPT_VMMCALL);
1036 	svm_set_intercept(svm, INTERCEPT_VMLOAD);
1037 	svm_set_intercept(svm, INTERCEPT_VMSAVE);
1038 	svm_set_intercept(svm, INTERCEPT_STGI);
1039 	svm_set_intercept(svm, INTERCEPT_CLGI);
1040 	svm_set_intercept(svm, INTERCEPT_SKINIT);
1041 	svm_set_intercept(svm, INTERCEPT_WBINVD);
1042 	svm_set_intercept(svm, INTERCEPT_XSETBV);
1043 	svm_set_intercept(svm, INTERCEPT_RDPRU);
1044 	svm_set_intercept(svm, INTERCEPT_RSM);
1045 
1046 	if (!kvm_mwait_in_guest(vcpu->kvm)) {
1047 		svm_set_intercept(svm, INTERCEPT_MONITOR);
1048 		svm_set_intercept(svm, INTERCEPT_MWAIT);
1049 	}
1050 
1051 	if (!kvm_hlt_in_guest(vcpu->kvm))
1052 		svm_set_intercept(svm, INTERCEPT_HLT);
1053 
1054 	control->iopm_base_pa = __sme_set(iopm_base);
1055 	control->msrpm_base_pa = __sme_set(__pa(svm->msrpm));
1056 	control->int_ctl = V_INTR_MASKING_MASK;
1057 
1058 	init_seg(&save->es);
1059 	init_seg(&save->ss);
1060 	init_seg(&save->ds);
1061 	init_seg(&save->fs);
1062 	init_seg(&save->gs);
1063 
1064 	save->cs.selector = 0xf000;
1065 	save->cs.base = 0xffff0000;
1066 	/* Executable/Readable Code Segment */
1067 	save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
1068 		SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
1069 	save->cs.limit = 0xffff;
1070 
1071 	save->gdtr.base = 0;
1072 	save->gdtr.limit = 0xffff;
1073 	save->idtr.base = 0;
1074 	save->idtr.limit = 0xffff;
1075 
1076 	init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
1077 	init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
1078 
1079 	if (npt_enabled) {
1080 		/* Setup VMCB for Nested Paging */
1081 		control->nested_ctl |= SVM_NESTED_CTL_NP_ENABLE;
1082 		svm_clr_intercept(svm, INTERCEPT_INVLPG);
1083 		clr_exception_intercept(svm, PF_VECTOR);
1084 		svm_clr_intercept(svm, INTERCEPT_CR3_READ);
1085 		svm_clr_intercept(svm, INTERCEPT_CR3_WRITE);
1086 		save->g_pat = vcpu->arch.pat;
1087 		save->cr3 = 0;
1088 	}
1089 	svm->current_vmcb->asid_generation = 0;
1090 	svm->asid = 0;
1091 
1092 	svm->nested.vmcb12_gpa = INVALID_GPA;
1093 	svm->nested.last_vmcb12_gpa = INVALID_GPA;
1094 
1095 	if (!kvm_pause_in_guest(vcpu->kvm)) {
1096 		control->pause_filter_count = pause_filter_count;
1097 		if (pause_filter_thresh)
1098 			control->pause_filter_thresh = pause_filter_thresh;
1099 		svm_set_intercept(svm, INTERCEPT_PAUSE);
1100 	} else {
1101 		svm_clr_intercept(svm, INTERCEPT_PAUSE);
1102 	}
1103 
1104 	svm_recalc_instruction_intercepts(vcpu, svm);
1105 
1106 	/*
1107 	 * If the host supports V_SPEC_CTRL then disable the interception
1108 	 * of MSR_IA32_SPEC_CTRL.
1109 	 */
1110 	if (boot_cpu_has(X86_FEATURE_V_SPEC_CTRL))
1111 		set_msr_interception(vcpu, svm->msrpm, MSR_IA32_SPEC_CTRL, 1, 1);
1112 
1113 	if (kvm_vcpu_apicv_active(vcpu))
1114 		avic_init_vmcb(svm);
1115 
1116 	if (vgif) {
1117 		svm_clr_intercept(svm, INTERCEPT_STGI);
1118 		svm_clr_intercept(svm, INTERCEPT_CLGI);
1119 		svm->vmcb->control.int_ctl |= V_GIF_ENABLE_MASK;
1120 	}
1121 
1122 	if (sev_guest(vcpu->kvm)) {
1123 		svm->vmcb->control.nested_ctl |= SVM_NESTED_CTL_SEV_ENABLE;
1124 		clr_exception_intercept(svm, UD_VECTOR);
1125 
1126 		if (sev_es_guest(vcpu->kvm)) {
1127 			/* Perform SEV-ES specific VMCB updates */
1128 			sev_es_init_vmcb(svm);
1129 		}
1130 	}
1131 
1132 	svm_hv_init_vmcb(svm->vmcb);
1133 	init_vmcb_after_set_cpuid(vcpu);
1134 
1135 	vmcb_mark_all_dirty(svm->vmcb);
1136 
1137 	enable_gif(svm);
1138 }
1139 
1140 static void __svm_vcpu_reset(struct kvm_vcpu *vcpu)
1141 {
1142 	struct vcpu_svm *svm = to_svm(vcpu);
1143 
1144 	svm_vcpu_init_msrpm(vcpu, svm->msrpm);
1145 
1146 	svm_init_osvw(vcpu);
1147 	vcpu->arch.microcode_version = 0x01000065;
1148 	svm->tsc_ratio_msr = kvm_default_tsc_scaling_ratio;
1149 
1150 	if (sev_es_guest(vcpu->kvm))
1151 		sev_es_vcpu_reset(svm);
1152 }
1153 
1154 static void svm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
1155 {
1156 	struct vcpu_svm *svm = to_svm(vcpu);
1157 
1158 	svm->spec_ctrl = 0;
1159 	svm->virt_spec_ctrl = 0;
1160 
1161 	init_vmcb(vcpu);
1162 
1163 	if (!init_event)
1164 		__svm_vcpu_reset(vcpu);
1165 }
1166 
1167 void svm_switch_vmcb(struct vcpu_svm *svm, struct kvm_vmcb_info *target_vmcb)
1168 {
1169 	svm->current_vmcb = target_vmcb;
1170 	svm->vmcb = target_vmcb->ptr;
1171 }
1172 
1173 static int svm_create_vcpu(struct kvm_vcpu *vcpu)
1174 {
1175 	struct vcpu_svm *svm;
1176 	struct page *vmcb01_page;
1177 	struct page *vmsa_page = NULL;
1178 	int err;
1179 
1180 	BUILD_BUG_ON(offsetof(struct vcpu_svm, vcpu) != 0);
1181 	svm = to_svm(vcpu);
1182 
1183 	err = -ENOMEM;
1184 	vmcb01_page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
1185 	if (!vmcb01_page)
1186 		goto out;
1187 
1188 	if (sev_es_guest(vcpu->kvm)) {
1189 		/*
1190 		 * SEV-ES guests require a separate VMSA page used to contain
1191 		 * the encrypted register state of the guest.
1192 		 */
1193 		vmsa_page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
1194 		if (!vmsa_page)
1195 			goto error_free_vmcb_page;
1196 
1197 		/*
1198 		 * SEV-ES guests maintain an encrypted version of their FPU
1199 		 * state which is restored and saved on VMRUN and VMEXIT.
1200 		 * Mark vcpu->arch.guest_fpu->fpstate as scratch so it won't
1201 		 * do xsave/xrstor on it.
1202 		 */
1203 		fpstate_set_confidential(&vcpu->arch.guest_fpu);
1204 	}
1205 
1206 	err = avic_init_vcpu(svm);
1207 	if (err)
1208 		goto error_free_vmsa_page;
1209 
1210 	svm->msrpm = svm_vcpu_alloc_msrpm();
1211 	if (!svm->msrpm) {
1212 		err = -ENOMEM;
1213 		goto error_free_vmsa_page;
1214 	}
1215 
1216 	svm->vmcb01.ptr = page_address(vmcb01_page);
1217 	svm->vmcb01.pa = __sme_set(page_to_pfn(vmcb01_page) << PAGE_SHIFT);
1218 	svm_switch_vmcb(svm, &svm->vmcb01);
1219 
1220 	if (vmsa_page)
1221 		svm->sev_es.vmsa = page_address(vmsa_page);
1222 
1223 	svm->guest_state_loaded = false;
1224 
1225 	return 0;
1226 
1227 error_free_vmsa_page:
1228 	if (vmsa_page)
1229 		__free_page(vmsa_page);
1230 error_free_vmcb_page:
1231 	__free_page(vmcb01_page);
1232 out:
1233 	return err;
1234 }
1235 
1236 static void svm_clear_current_vmcb(struct vmcb *vmcb)
1237 {
1238 	int i;
1239 
1240 	for_each_online_cpu(i)
1241 		cmpxchg(&per_cpu(svm_data, i)->current_vmcb, vmcb, NULL);
1242 }
1243 
1244 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
1245 {
1246 	struct vcpu_svm *svm = to_svm(vcpu);
1247 
1248 	/*
1249 	 * The vmcb page can be recycled, causing a false negative in
1250 	 * svm_vcpu_load(). So, ensure that no logical CPU has this
1251 	 * vmcb page recorded as its current vmcb.
1252 	 */
1253 	svm_clear_current_vmcb(svm->vmcb);
1254 
1255 	svm_free_nested(svm);
1256 
1257 	sev_free_vcpu(vcpu);
1258 
1259 	__free_page(pfn_to_page(__sme_clr(svm->vmcb01.pa) >> PAGE_SHIFT));
1260 	__free_pages(virt_to_page(svm->msrpm), get_order(MSRPM_SIZE));
1261 }
1262 
1263 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
1264 {
1265 	struct vcpu_svm *svm = to_svm(vcpu);
1266 	struct svm_cpu_data *sd = per_cpu(svm_data, vcpu->cpu);
1267 
1268 	if (sev_es_guest(vcpu->kvm))
1269 		sev_es_unmap_ghcb(svm);
1270 
1271 	if (svm->guest_state_loaded)
1272 		return;
1273 
1274 	/*
1275 	 * Save additional host state that will be restored on VMEXIT (sev-es)
1276 	 * or subsequent vmload of host save area.
1277 	 */
1278 	if (sev_es_guest(vcpu->kvm)) {
1279 		sev_es_prepare_guest_switch(svm, vcpu->cpu);
1280 	} else {
1281 		vmsave(__sme_page_pa(sd->save_area));
1282 	}
1283 
1284 	if (tsc_scaling) {
1285 		u64 tsc_ratio = vcpu->arch.tsc_scaling_ratio;
1286 		if (tsc_ratio != __this_cpu_read(current_tsc_ratio)) {
1287 			__this_cpu_write(current_tsc_ratio, tsc_ratio);
1288 			wrmsrl(MSR_AMD64_TSC_RATIO, tsc_ratio);
1289 		}
1290 	}
1291 
1292 	if (likely(tsc_aux_uret_slot >= 0))
1293 		kvm_set_user_return_msr(tsc_aux_uret_slot, svm->tsc_aux, -1ull);
1294 
1295 	svm->guest_state_loaded = true;
1296 }
1297 
1298 static void svm_prepare_host_switch(struct kvm_vcpu *vcpu)
1299 {
1300 	to_svm(vcpu)->guest_state_loaded = false;
1301 }
1302 
1303 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1304 {
1305 	struct vcpu_svm *svm = to_svm(vcpu);
1306 	struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
1307 
1308 	if (sd->current_vmcb != svm->vmcb) {
1309 		sd->current_vmcb = svm->vmcb;
1310 		indirect_branch_prediction_barrier();
1311 	}
1312 	if (kvm_vcpu_apicv_active(vcpu))
1313 		avic_vcpu_load(vcpu, cpu);
1314 }
1315 
1316 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
1317 {
1318 	if (kvm_vcpu_apicv_active(vcpu))
1319 		avic_vcpu_put(vcpu);
1320 
1321 	svm_prepare_host_switch(vcpu);
1322 
1323 	++vcpu->stat.host_state_reload;
1324 }
1325 
1326 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
1327 {
1328 	struct vcpu_svm *svm = to_svm(vcpu);
1329 	unsigned long rflags = svm->vmcb->save.rflags;
1330 
1331 	if (svm->nmi_singlestep) {
1332 		/* Hide our flags if they were not set by the guest */
1333 		if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF))
1334 			rflags &= ~X86_EFLAGS_TF;
1335 		if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_RF))
1336 			rflags &= ~X86_EFLAGS_RF;
1337 	}
1338 	return rflags;
1339 }
1340 
1341 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1342 {
1343 	if (to_svm(vcpu)->nmi_singlestep)
1344 		rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
1345 
1346        /*
1347         * Any change of EFLAGS.VM is accompanied by a reload of SS
1348         * (caused by either a task switch or an inter-privilege IRET),
1349         * so we do not need to update the CPL here.
1350         */
1351 	to_svm(vcpu)->vmcb->save.rflags = rflags;
1352 }
1353 
1354 static bool svm_get_if_flag(struct kvm_vcpu *vcpu)
1355 {
1356 	struct vmcb *vmcb = to_svm(vcpu)->vmcb;
1357 
1358 	return sev_es_guest(vcpu->kvm)
1359 		? vmcb->control.int_state & SVM_GUEST_INTERRUPT_MASK
1360 		: kvm_get_rflags(vcpu) & X86_EFLAGS_IF;
1361 }
1362 
1363 static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1364 {
1365 	kvm_register_mark_available(vcpu, reg);
1366 
1367 	switch (reg) {
1368 	case VCPU_EXREG_PDPTR:
1369 		/*
1370 		 * When !npt_enabled, mmu->pdptrs[] is already available since
1371 		 * it is always updated per SDM when moving to CRs.
1372 		 */
1373 		if (npt_enabled)
1374 			load_pdptrs(vcpu, kvm_read_cr3(vcpu));
1375 		break;
1376 	default:
1377 		KVM_BUG_ON(1, vcpu->kvm);
1378 	}
1379 }
1380 
1381 static void svm_set_vintr(struct vcpu_svm *svm)
1382 {
1383 	struct vmcb_control_area *control;
1384 
1385 	/*
1386 	 * The following fields are ignored when AVIC is enabled
1387 	 */
1388 	WARN_ON(kvm_apicv_activated(svm->vcpu.kvm));
1389 
1390 	svm_set_intercept(svm, INTERCEPT_VINTR);
1391 
1392 	/*
1393 	 * This is just a dummy VINTR to actually cause a vmexit to happen.
1394 	 * Actual injection of virtual interrupts happens through EVENTINJ.
1395 	 */
1396 	control = &svm->vmcb->control;
1397 	control->int_vector = 0x0;
1398 	control->int_ctl &= ~V_INTR_PRIO_MASK;
1399 	control->int_ctl |= V_IRQ_MASK |
1400 		((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
1401 	vmcb_mark_dirty(svm->vmcb, VMCB_INTR);
1402 }
1403 
1404 static void svm_clear_vintr(struct vcpu_svm *svm)
1405 {
1406 	svm_clr_intercept(svm, INTERCEPT_VINTR);
1407 
1408 	/* Drop int_ctl fields related to VINTR injection.  */
1409 	svm->vmcb->control.int_ctl &= ~V_IRQ_INJECTION_BITS_MASK;
1410 	if (is_guest_mode(&svm->vcpu)) {
1411 		svm->vmcb01.ptr->control.int_ctl &= ~V_IRQ_INJECTION_BITS_MASK;
1412 
1413 		WARN_ON((svm->vmcb->control.int_ctl & V_TPR_MASK) !=
1414 			(svm->nested.ctl.int_ctl & V_TPR_MASK));
1415 
1416 		svm->vmcb->control.int_ctl |= svm->nested.ctl.int_ctl &
1417 			V_IRQ_INJECTION_BITS_MASK;
1418 
1419 		svm->vmcb->control.int_vector = svm->nested.ctl.int_vector;
1420 	}
1421 
1422 	vmcb_mark_dirty(svm->vmcb, VMCB_INTR);
1423 }
1424 
1425 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
1426 {
1427 	struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1428 	struct vmcb_save_area *save01 = &to_svm(vcpu)->vmcb01.ptr->save;
1429 
1430 	switch (seg) {
1431 	case VCPU_SREG_CS: return &save->cs;
1432 	case VCPU_SREG_DS: return &save->ds;
1433 	case VCPU_SREG_ES: return &save->es;
1434 	case VCPU_SREG_FS: return &save01->fs;
1435 	case VCPU_SREG_GS: return &save01->gs;
1436 	case VCPU_SREG_SS: return &save->ss;
1437 	case VCPU_SREG_TR: return &save01->tr;
1438 	case VCPU_SREG_LDTR: return &save01->ldtr;
1439 	}
1440 	BUG();
1441 	return NULL;
1442 }
1443 
1444 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1445 {
1446 	struct vmcb_seg *s = svm_seg(vcpu, seg);
1447 
1448 	return s->base;
1449 }
1450 
1451 static void svm_get_segment(struct kvm_vcpu *vcpu,
1452 			    struct kvm_segment *var, int seg)
1453 {
1454 	struct vmcb_seg *s = svm_seg(vcpu, seg);
1455 
1456 	var->base = s->base;
1457 	var->limit = s->limit;
1458 	var->selector = s->selector;
1459 	var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
1460 	var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
1461 	var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
1462 	var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
1463 	var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
1464 	var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
1465 	var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
1466 
1467 	/*
1468 	 * AMD CPUs circa 2014 track the G bit for all segments except CS.
1469 	 * However, the SVM spec states that the G bit is not observed by the
1470 	 * CPU, and some VMware virtual CPUs drop the G bit for all segments.
1471 	 * So let's synthesize a legal G bit for all segments, this helps
1472 	 * running KVM nested. It also helps cross-vendor migration, because
1473 	 * Intel's vmentry has a check on the 'G' bit.
1474 	 */
1475 	var->g = s->limit > 0xfffff;
1476 
1477 	/*
1478 	 * AMD's VMCB does not have an explicit unusable field, so emulate it
1479 	 * for cross vendor migration purposes by "not present"
1480 	 */
1481 	var->unusable = !var->present;
1482 
1483 	switch (seg) {
1484 	case VCPU_SREG_TR:
1485 		/*
1486 		 * Work around a bug where the busy flag in the tr selector
1487 		 * isn't exposed
1488 		 */
1489 		var->type |= 0x2;
1490 		break;
1491 	case VCPU_SREG_DS:
1492 	case VCPU_SREG_ES:
1493 	case VCPU_SREG_FS:
1494 	case VCPU_SREG_GS:
1495 		/*
1496 		 * The accessed bit must always be set in the segment
1497 		 * descriptor cache, although it can be cleared in the
1498 		 * descriptor, the cached bit always remains at 1. Since
1499 		 * Intel has a check on this, set it here to support
1500 		 * cross-vendor migration.
1501 		 */
1502 		if (!var->unusable)
1503 			var->type |= 0x1;
1504 		break;
1505 	case VCPU_SREG_SS:
1506 		/*
1507 		 * On AMD CPUs sometimes the DB bit in the segment
1508 		 * descriptor is left as 1, although the whole segment has
1509 		 * been made unusable. Clear it here to pass an Intel VMX
1510 		 * entry check when cross vendor migrating.
1511 		 */
1512 		if (var->unusable)
1513 			var->db = 0;
1514 		/* This is symmetric with svm_set_segment() */
1515 		var->dpl = to_svm(vcpu)->vmcb->save.cpl;
1516 		break;
1517 	}
1518 }
1519 
1520 static int svm_get_cpl(struct kvm_vcpu *vcpu)
1521 {
1522 	struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1523 
1524 	return save->cpl;
1525 }
1526 
1527 static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1528 {
1529 	struct vcpu_svm *svm = to_svm(vcpu);
1530 
1531 	dt->size = svm->vmcb->save.idtr.limit;
1532 	dt->address = svm->vmcb->save.idtr.base;
1533 }
1534 
1535 static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1536 {
1537 	struct vcpu_svm *svm = to_svm(vcpu);
1538 
1539 	svm->vmcb->save.idtr.limit = dt->size;
1540 	svm->vmcb->save.idtr.base = dt->address ;
1541 	vmcb_mark_dirty(svm->vmcb, VMCB_DT);
1542 }
1543 
1544 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1545 {
1546 	struct vcpu_svm *svm = to_svm(vcpu);
1547 
1548 	dt->size = svm->vmcb->save.gdtr.limit;
1549 	dt->address = svm->vmcb->save.gdtr.base;
1550 }
1551 
1552 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1553 {
1554 	struct vcpu_svm *svm = to_svm(vcpu);
1555 
1556 	svm->vmcb->save.gdtr.limit = dt->size;
1557 	svm->vmcb->save.gdtr.base = dt->address ;
1558 	vmcb_mark_dirty(svm->vmcb, VMCB_DT);
1559 }
1560 
1561 static void svm_post_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1562 {
1563 	struct vcpu_svm *svm = to_svm(vcpu);
1564 
1565 	/*
1566 	 * For guests that don't set guest_state_protected, the cr3 update is
1567 	 * handled via kvm_mmu_load() while entering the guest. For guests
1568 	 * that do (SEV-ES/SEV-SNP), the cr3 update needs to be written to
1569 	 * VMCB save area now, since the save area will become the initial
1570 	 * contents of the VMSA, and future VMCB save area updates won't be
1571 	 * seen.
1572 	 */
1573 	if (sev_es_guest(vcpu->kvm)) {
1574 		svm->vmcb->save.cr3 = cr3;
1575 		vmcb_mark_dirty(svm->vmcb, VMCB_CR);
1576 	}
1577 }
1578 
1579 void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1580 {
1581 	struct vcpu_svm *svm = to_svm(vcpu);
1582 	u64 hcr0 = cr0;
1583 
1584 #ifdef CONFIG_X86_64
1585 	if (vcpu->arch.efer & EFER_LME && !vcpu->arch.guest_state_protected) {
1586 		if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
1587 			vcpu->arch.efer |= EFER_LMA;
1588 			svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
1589 		}
1590 
1591 		if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
1592 			vcpu->arch.efer &= ~EFER_LMA;
1593 			svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
1594 		}
1595 	}
1596 #endif
1597 	vcpu->arch.cr0 = cr0;
1598 
1599 	if (!npt_enabled)
1600 		hcr0 |= X86_CR0_PG | X86_CR0_WP;
1601 
1602 	/*
1603 	 * re-enable caching here because the QEMU bios
1604 	 * does not do it - this results in some delay at
1605 	 * reboot
1606 	 */
1607 	if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
1608 		hcr0 &= ~(X86_CR0_CD | X86_CR0_NW);
1609 
1610 	svm->vmcb->save.cr0 = hcr0;
1611 	vmcb_mark_dirty(svm->vmcb, VMCB_CR);
1612 
1613 	/*
1614 	 * SEV-ES guests must always keep the CR intercepts cleared. CR
1615 	 * tracking is done using the CR write traps.
1616 	 */
1617 	if (sev_es_guest(vcpu->kvm))
1618 		return;
1619 
1620 	if (hcr0 == cr0) {
1621 		/* Selective CR0 write remains on.  */
1622 		svm_clr_intercept(svm, INTERCEPT_CR0_READ);
1623 		svm_clr_intercept(svm, INTERCEPT_CR0_WRITE);
1624 	} else {
1625 		svm_set_intercept(svm, INTERCEPT_CR0_READ);
1626 		svm_set_intercept(svm, INTERCEPT_CR0_WRITE);
1627 	}
1628 }
1629 
1630 static bool svm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1631 {
1632 	return true;
1633 }
1634 
1635 void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1636 {
1637 	unsigned long host_cr4_mce = cr4_read_shadow() & X86_CR4_MCE;
1638 	unsigned long old_cr4 = vcpu->arch.cr4;
1639 
1640 	if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
1641 		svm_flush_tlb(vcpu);
1642 
1643 	vcpu->arch.cr4 = cr4;
1644 	if (!npt_enabled)
1645 		cr4 |= X86_CR4_PAE;
1646 	cr4 |= host_cr4_mce;
1647 	to_svm(vcpu)->vmcb->save.cr4 = cr4;
1648 	vmcb_mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
1649 
1650 	if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
1651 		kvm_update_cpuid_runtime(vcpu);
1652 }
1653 
1654 static void svm_set_segment(struct kvm_vcpu *vcpu,
1655 			    struct kvm_segment *var, int seg)
1656 {
1657 	struct vcpu_svm *svm = to_svm(vcpu);
1658 	struct vmcb_seg *s = svm_seg(vcpu, seg);
1659 
1660 	s->base = var->base;
1661 	s->limit = var->limit;
1662 	s->selector = var->selector;
1663 	s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
1664 	s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
1665 	s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
1666 	s->attrib |= ((var->present & 1) && !var->unusable) << SVM_SELECTOR_P_SHIFT;
1667 	s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
1668 	s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
1669 	s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
1670 	s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
1671 
1672 	/*
1673 	 * This is always accurate, except if SYSRET returned to a segment
1674 	 * with SS.DPL != 3.  Intel does not have this quirk, and always
1675 	 * forces SS.DPL to 3 on sysret, so we ignore that case; fixing it
1676 	 * would entail passing the CPL to userspace and back.
1677 	 */
1678 	if (seg == VCPU_SREG_SS)
1679 		/* This is symmetric with svm_get_segment() */
1680 		svm->vmcb->save.cpl = (var->dpl & 3);
1681 
1682 	vmcb_mark_dirty(svm->vmcb, VMCB_SEG);
1683 }
1684 
1685 static void svm_update_exception_bitmap(struct kvm_vcpu *vcpu)
1686 {
1687 	struct vcpu_svm *svm = to_svm(vcpu);
1688 
1689 	clr_exception_intercept(svm, BP_VECTOR);
1690 
1691 	if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
1692 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
1693 			set_exception_intercept(svm, BP_VECTOR);
1694 	}
1695 }
1696 
1697 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd)
1698 {
1699 	if (sd->next_asid > sd->max_asid) {
1700 		++sd->asid_generation;
1701 		sd->next_asid = sd->min_asid;
1702 		svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
1703 		vmcb_mark_dirty(svm->vmcb, VMCB_ASID);
1704 	}
1705 
1706 	svm->current_vmcb->asid_generation = sd->asid_generation;
1707 	svm->asid = sd->next_asid++;
1708 }
1709 
1710 static void svm_set_dr6(struct vcpu_svm *svm, unsigned long value)
1711 {
1712 	struct vmcb *vmcb = svm->vmcb;
1713 
1714 	if (svm->vcpu.arch.guest_state_protected)
1715 		return;
1716 
1717 	if (unlikely(value != vmcb->save.dr6)) {
1718 		vmcb->save.dr6 = value;
1719 		vmcb_mark_dirty(vmcb, VMCB_DR);
1720 	}
1721 }
1722 
1723 static void svm_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
1724 {
1725 	struct vcpu_svm *svm = to_svm(vcpu);
1726 
1727 	if (vcpu->arch.guest_state_protected)
1728 		return;
1729 
1730 	get_debugreg(vcpu->arch.db[0], 0);
1731 	get_debugreg(vcpu->arch.db[1], 1);
1732 	get_debugreg(vcpu->arch.db[2], 2);
1733 	get_debugreg(vcpu->arch.db[3], 3);
1734 	/*
1735 	 * We cannot reset svm->vmcb->save.dr6 to DR6_ACTIVE_LOW here,
1736 	 * because db_interception might need it.  We can do it before vmentry.
1737 	 */
1738 	vcpu->arch.dr6 = svm->vmcb->save.dr6;
1739 	vcpu->arch.dr7 = svm->vmcb->save.dr7;
1740 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
1741 	set_dr_intercepts(svm);
1742 }
1743 
1744 static void svm_set_dr7(struct kvm_vcpu *vcpu, unsigned long value)
1745 {
1746 	struct vcpu_svm *svm = to_svm(vcpu);
1747 
1748 	if (vcpu->arch.guest_state_protected)
1749 		return;
1750 
1751 	svm->vmcb->save.dr7 = value;
1752 	vmcb_mark_dirty(svm->vmcb, VMCB_DR);
1753 }
1754 
1755 static int pf_interception(struct kvm_vcpu *vcpu)
1756 {
1757 	struct vcpu_svm *svm = to_svm(vcpu);
1758 
1759 	u64 fault_address = svm->vmcb->control.exit_info_2;
1760 	u64 error_code = svm->vmcb->control.exit_info_1;
1761 
1762 	return kvm_handle_page_fault(vcpu, error_code, fault_address,
1763 			static_cpu_has(X86_FEATURE_DECODEASSISTS) ?
1764 			svm->vmcb->control.insn_bytes : NULL,
1765 			svm->vmcb->control.insn_len);
1766 }
1767 
1768 static int npf_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 	trace_kvm_page_fault(fault_address, error_code);
1776 	return kvm_mmu_page_fault(vcpu, fault_address, error_code,
1777 			static_cpu_has(X86_FEATURE_DECODEASSISTS) ?
1778 			svm->vmcb->control.insn_bytes : NULL,
1779 			svm->vmcb->control.insn_len);
1780 }
1781 
1782 static int db_interception(struct kvm_vcpu *vcpu)
1783 {
1784 	struct kvm_run *kvm_run = vcpu->run;
1785 	struct vcpu_svm *svm = to_svm(vcpu);
1786 
1787 	if (!(vcpu->guest_debug &
1788 	      (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
1789 		!svm->nmi_singlestep) {
1790 		u32 payload = svm->vmcb->save.dr6 ^ DR6_ACTIVE_LOW;
1791 		kvm_queue_exception_p(vcpu, DB_VECTOR, payload);
1792 		return 1;
1793 	}
1794 
1795 	if (svm->nmi_singlestep) {
1796 		disable_nmi_singlestep(svm);
1797 		/* Make sure we check for pending NMIs upon entry */
1798 		kvm_make_request(KVM_REQ_EVENT, vcpu);
1799 	}
1800 
1801 	if (vcpu->guest_debug &
1802 	    (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) {
1803 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
1804 		kvm_run->debug.arch.dr6 = svm->vmcb->save.dr6;
1805 		kvm_run->debug.arch.dr7 = svm->vmcb->save.dr7;
1806 		kvm_run->debug.arch.pc =
1807 			svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1808 		kvm_run->debug.arch.exception = DB_VECTOR;
1809 		return 0;
1810 	}
1811 
1812 	return 1;
1813 }
1814 
1815 static int bp_interception(struct kvm_vcpu *vcpu)
1816 {
1817 	struct vcpu_svm *svm = to_svm(vcpu);
1818 	struct kvm_run *kvm_run = vcpu->run;
1819 
1820 	kvm_run->exit_reason = KVM_EXIT_DEBUG;
1821 	kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1822 	kvm_run->debug.arch.exception = BP_VECTOR;
1823 	return 0;
1824 }
1825 
1826 static int ud_interception(struct kvm_vcpu *vcpu)
1827 {
1828 	return handle_ud(vcpu);
1829 }
1830 
1831 static int ac_interception(struct kvm_vcpu *vcpu)
1832 {
1833 	kvm_queue_exception_e(vcpu, AC_VECTOR, 0);
1834 	return 1;
1835 }
1836 
1837 static bool is_erratum_383(void)
1838 {
1839 	int err, i;
1840 	u64 value;
1841 
1842 	if (!erratum_383_found)
1843 		return false;
1844 
1845 	value = native_read_msr_safe(MSR_IA32_MC0_STATUS, &err);
1846 	if (err)
1847 		return false;
1848 
1849 	/* Bit 62 may or may not be set for this mce */
1850 	value &= ~(1ULL << 62);
1851 
1852 	if (value != 0xb600000000010015ULL)
1853 		return false;
1854 
1855 	/* Clear MCi_STATUS registers */
1856 	for (i = 0; i < 6; ++i)
1857 		native_write_msr_safe(MSR_IA32_MCx_STATUS(i), 0, 0);
1858 
1859 	value = native_read_msr_safe(MSR_IA32_MCG_STATUS, &err);
1860 	if (!err) {
1861 		u32 low, high;
1862 
1863 		value &= ~(1ULL << 2);
1864 		low    = lower_32_bits(value);
1865 		high   = upper_32_bits(value);
1866 
1867 		native_write_msr_safe(MSR_IA32_MCG_STATUS, low, high);
1868 	}
1869 
1870 	/* Flush tlb to evict multi-match entries */
1871 	__flush_tlb_all();
1872 
1873 	return true;
1874 }
1875 
1876 static void svm_handle_mce(struct kvm_vcpu *vcpu)
1877 {
1878 	if (is_erratum_383()) {
1879 		/*
1880 		 * Erratum 383 triggered. Guest state is corrupt so kill the
1881 		 * guest.
1882 		 */
1883 		pr_err("KVM: Guest triggered AMD Erratum 383\n");
1884 
1885 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
1886 
1887 		return;
1888 	}
1889 
1890 	/*
1891 	 * On an #MC intercept the MCE handler is not called automatically in
1892 	 * the host. So do it by hand here.
1893 	 */
1894 	kvm_machine_check();
1895 }
1896 
1897 static int mc_interception(struct kvm_vcpu *vcpu)
1898 {
1899 	return 1;
1900 }
1901 
1902 static int shutdown_interception(struct kvm_vcpu *vcpu)
1903 {
1904 	struct kvm_run *kvm_run = vcpu->run;
1905 	struct vcpu_svm *svm = to_svm(vcpu);
1906 
1907 	/*
1908 	 * The VM save area has already been encrypted so it
1909 	 * cannot be reinitialized - just terminate.
1910 	 */
1911 	if (sev_es_guest(vcpu->kvm))
1912 		return -EINVAL;
1913 
1914 	/*
1915 	 * VMCB is undefined after a SHUTDOWN intercept.  INIT the vCPU to put
1916 	 * the VMCB in a known good state.  Unfortuately, KVM doesn't have
1917 	 * KVM_MP_STATE_SHUTDOWN and can't add it without potentially breaking
1918 	 * userspace.  At a platform view, INIT is acceptable behavior as
1919 	 * there exist bare metal platforms that automatically INIT the CPU
1920 	 * in response to shutdown.
1921 	 */
1922 	clear_page(svm->vmcb);
1923 	kvm_vcpu_reset(vcpu, true);
1924 
1925 	kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1926 	return 0;
1927 }
1928 
1929 static int io_interception(struct kvm_vcpu *vcpu)
1930 {
1931 	struct vcpu_svm *svm = to_svm(vcpu);
1932 	u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
1933 	int size, in, string;
1934 	unsigned port;
1935 
1936 	++vcpu->stat.io_exits;
1937 	string = (io_info & SVM_IOIO_STR_MASK) != 0;
1938 	in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
1939 	port = io_info >> 16;
1940 	size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
1941 
1942 	if (string) {
1943 		if (sev_es_guest(vcpu->kvm))
1944 			return sev_es_string_io(svm, size, port, in);
1945 		else
1946 			return kvm_emulate_instruction(vcpu, 0);
1947 	}
1948 
1949 	svm->next_rip = svm->vmcb->control.exit_info_2;
1950 
1951 	return kvm_fast_pio(vcpu, size, port, in);
1952 }
1953 
1954 static int nmi_interception(struct kvm_vcpu *vcpu)
1955 {
1956 	return 1;
1957 }
1958 
1959 static int smi_interception(struct kvm_vcpu *vcpu)
1960 {
1961 	return 1;
1962 }
1963 
1964 static int intr_interception(struct kvm_vcpu *vcpu)
1965 {
1966 	++vcpu->stat.irq_exits;
1967 	return 1;
1968 }
1969 
1970 static int vmload_vmsave_interception(struct kvm_vcpu *vcpu, bool vmload)
1971 {
1972 	struct vcpu_svm *svm = to_svm(vcpu);
1973 	struct vmcb *vmcb12;
1974 	struct kvm_host_map map;
1975 	int ret;
1976 
1977 	if (nested_svm_check_permissions(vcpu))
1978 		return 1;
1979 
1980 	ret = kvm_vcpu_map(vcpu, gpa_to_gfn(svm->vmcb->save.rax), &map);
1981 	if (ret) {
1982 		if (ret == -EINVAL)
1983 			kvm_inject_gp(vcpu, 0);
1984 		return 1;
1985 	}
1986 
1987 	vmcb12 = map.hva;
1988 
1989 	ret = kvm_skip_emulated_instruction(vcpu);
1990 
1991 	if (vmload) {
1992 		svm_copy_vmloadsave_state(svm->vmcb, vmcb12);
1993 		svm->sysenter_eip_hi = 0;
1994 		svm->sysenter_esp_hi = 0;
1995 	} else {
1996 		svm_copy_vmloadsave_state(vmcb12, svm->vmcb);
1997 	}
1998 
1999 	kvm_vcpu_unmap(vcpu, &map, true);
2000 
2001 	return ret;
2002 }
2003 
2004 static int vmload_interception(struct kvm_vcpu *vcpu)
2005 {
2006 	return vmload_vmsave_interception(vcpu, true);
2007 }
2008 
2009 static int vmsave_interception(struct kvm_vcpu *vcpu)
2010 {
2011 	return vmload_vmsave_interception(vcpu, false);
2012 }
2013 
2014 static int vmrun_interception(struct kvm_vcpu *vcpu)
2015 {
2016 	if (nested_svm_check_permissions(vcpu))
2017 		return 1;
2018 
2019 	return nested_svm_vmrun(vcpu);
2020 }
2021 
2022 enum {
2023 	NONE_SVM_INSTR,
2024 	SVM_INSTR_VMRUN,
2025 	SVM_INSTR_VMLOAD,
2026 	SVM_INSTR_VMSAVE,
2027 };
2028 
2029 /* Return NONE_SVM_INSTR if not SVM instrs, otherwise return decode result */
2030 static int svm_instr_opcode(struct kvm_vcpu *vcpu)
2031 {
2032 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
2033 
2034 	if (ctxt->b != 0x1 || ctxt->opcode_len != 2)
2035 		return NONE_SVM_INSTR;
2036 
2037 	switch (ctxt->modrm) {
2038 	case 0xd8: /* VMRUN */
2039 		return SVM_INSTR_VMRUN;
2040 	case 0xda: /* VMLOAD */
2041 		return SVM_INSTR_VMLOAD;
2042 	case 0xdb: /* VMSAVE */
2043 		return SVM_INSTR_VMSAVE;
2044 	default:
2045 		break;
2046 	}
2047 
2048 	return NONE_SVM_INSTR;
2049 }
2050 
2051 static int emulate_svm_instr(struct kvm_vcpu *vcpu, int opcode)
2052 {
2053 	const int guest_mode_exit_codes[] = {
2054 		[SVM_INSTR_VMRUN] = SVM_EXIT_VMRUN,
2055 		[SVM_INSTR_VMLOAD] = SVM_EXIT_VMLOAD,
2056 		[SVM_INSTR_VMSAVE] = SVM_EXIT_VMSAVE,
2057 	};
2058 	int (*const svm_instr_handlers[])(struct kvm_vcpu *vcpu) = {
2059 		[SVM_INSTR_VMRUN] = vmrun_interception,
2060 		[SVM_INSTR_VMLOAD] = vmload_interception,
2061 		[SVM_INSTR_VMSAVE] = vmsave_interception,
2062 	};
2063 	struct vcpu_svm *svm = to_svm(vcpu);
2064 	int ret;
2065 
2066 	if (is_guest_mode(vcpu)) {
2067 		/* Returns '1' or -errno on failure, '0' on success. */
2068 		ret = nested_svm_simple_vmexit(svm, guest_mode_exit_codes[opcode]);
2069 		if (ret)
2070 			return ret;
2071 		return 1;
2072 	}
2073 	return svm_instr_handlers[opcode](vcpu);
2074 }
2075 
2076 /*
2077  * #GP handling code. Note that #GP can be triggered under the following two
2078  * cases:
2079  *   1) SVM VM-related instructions (VMRUN/VMSAVE/VMLOAD) that trigger #GP on
2080  *      some AMD CPUs when EAX of these instructions are in the reserved memory
2081  *      regions (e.g. SMM memory on host).
2082  *   2) VMware backdoor
2083  */
2084 static int gp_interception(struct kvm_vcpu *vcpu)
2085 {
2086 	struct vcpu_svm *svm = to_svm(vcpu);
2087 	u32 error_code = svm->vmcb->control.exit_info_1;
2088 	int opcode;
2089 
2090 	/* Both #GP cases have zero error_code */
2091 	if (error_code)
2092 		goto reinject;
2093 
2094 	/* All SVM instructions expect page aligned RAX */
2095 	if (svm->vmcb->save.rax & ~PAGE_MASK)
2096 		goto reinject;
2097 
2098 	/* Decode the instruction for usage later */
2099 	if (x86_decode_emulated_instruction(vcpu, 0, NULL, 0) != EMULATION_OK)
2100 		goto reinject;
2101 
2102 	opcode = svm_instr_opcode(vcpu);
2103 
2104 	if (opcode == NONE_SVM_INSTR) {
2105 		if (!enable_vmware_backdoor)
2106 			goto reinject;
2107 
2108 		/*
2109 		 * VMware backdoor emulation on #GP interception only handles
2110 		 * IN{S}, OUT{S}, and RDPMC.
2111 		 */
2112 		if (!is_guest_mode(vcpu))
2113 			return kvm_emulate_instruction(vcpu,
2114 				EMULTYPE_VMWARE_GP | EMULTYPE_NO_DECODE);
2115 	} else
2116 		return emulate_svm_instr(vcpu, opcode);
2117 
2118 reinject:
2119 	kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
2120 	return 1;
2121 }
2122 
2123 void svm_set_gif(struct vcpu_svm *svm, bool value)
2124 {
2125 	if (value) {
2126 		/*
2127 		 * If VGIF is enabled, the STGI intercept is only added to
2128 		 * detect the opening of the SMI/NMI window; remove it now.
2129 		 * Likewise, clear the VINTR intercept, we will set it
2130 		 * again while processing KVM_REQ_EVENT if needed.
2131 		 */
2132 		if (vgif_enabled(svm))
2133 			svm_clr_intercept(svm, INTERCEPT_STGI);
2134 		if (svm_is_intercept(svm, INTERCEPT_VINTR))
2135 			svm_clear_vintr(svm);
2136 
2137 		enable_gif(svm);
2138 		if (svm->vcpu.arch.smi_pending ||
2139 		    svm->vcpu.arch.nmi_pending ||
2140 		    kvm_cpu_has_injectable_intr(&svm->vcpu))
2141 			kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
2142 	} else {
2143 		disable_gif(svm);
2144 
2145 		/*
2146 		 * After a CLGI no interrupts should come.  But if vGIF is
2147 		 * in use, we still rely on the VINTR intercept (rather than
2148 		 * STGI) to detect an open interrupt window.
2149 		*/
2150 		if (!vgif_enabled(svm))
2151 			svm_clear_vintr(svm);
2152 	}
2153 }
2154 
2155 static int stgi_interception(struct kvm_vcpu *vcpu)
2156 {
2157 	int ret;
2158 
2159 	if (nested_svm_check_permissions(vcpu))
2160 		return 1;
2161 
2162 	ret = kvm_skip_emulated_instruction(vcpu);
2163 	svm_set_gif(to_svm(vcpu), true);
2164 	return ret;
2165 }
2166 
2167 static int clgi_interception(struct kvm_vcpu *vcpu)
2168 {
2169 	int ret;
2170 
2171 	if (nested_svm_check_permissions(vcpu))
2172 		return 1;
2173 
2174 	ret = kvm_skip_emulated_instruction(vcpu);
2175 	svm_set_gif(to_svm(vcpu), false);
2176 	return ret;
2177 }
2178 
2179 static int invlpga_interception(struct kvm_vcpu *vcpu)
2180 {
2181 	gva_t gva = kvm_rax_read(vcpu);
2182 	u32 asid = kvm_rcx_read(vcpu);
2183 
2184 	/* FIXME: Handle an address size prefix. */
2185 	if (!is_long_mode(vcpu))
2186 		gva = (u32)gva;
2187 
2188 	trace_kvm_invlpga(to_svm(vcpu)->vmcb->save.rip, asid, gva);
2189 
2190 	/* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
2191 	kvm_mmu_invlpg(vcpu, gva);
2192 
2193 	return kvm_skip_emulated_instruction(vcpu);
2194 }
2195 
2196 static int skinit_interception(struct kvm_vcpu *vcpu)
2197 {
2198 	trace_kvm_skinit(to_svm(vcpu)->vmcb->save.rip, kvm_rax_read(vcpu));
2199 
2200 	kvm_queue_exception(vcpu, UD_VECTOR);
2201 	return 1;
2202 }
2203 
2204 static int task_switch_interception(struct kvm_vcpu *vcpu)
2205 {
2206 	struct vcpu_svm *svm = to_svm(vcpu);
2207 	u16 tss_selector;
2208 	int reason;
2209 	int int_type = svm->vmcb->control.exit_int_info &
2210 		SVM_EXITINTINFO_TYPE_MASK;
2211 	int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
2212 	uint32_t type =
2213 		svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
2214 	uint32_t idt_v =
2215 		svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
2216 	bool has_error_code = false;
2217 	u32 error_code = 0;
2218 
2219 	tss_selector = (u16)svm->vmcb->control.exit_info_1;
2220 
2221 	if (svm->vmcb->control.exit_info_2 &
2222 	    (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
2223 		reason = TASK_SWITCH_IRET;
2224 	else if (svm->vmcb->control.exit_info_2 &
2225 		 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
2226 		reason = TASK_SWITCH_JMP;
2227 	else if (idt_v)
2228 		reason = TASK_SWITCH_GATE;
2229 	else
2230 		reason = TASK_SWITCH_CALL;
2231 
2232 	if (reason == TASK_SWITCH_GATE) {
2233 		switch (type) {
2234 		case SVM_EXITINTINFO_TYPE_NMI:
2235 			vcpu->arch.nmi_injected = false;
2236 			break;
2237 		case SVM_EXITINTINFO_TYPE_EXEPT:
2238 			if (svm->vmcb->control.exit_info_2 &
2239 			    (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE)) {
2240 				has_error_code = true;
2241 				error_code =
2242 					(u32)svm->vmcb->control.exit_info_2;
2243 			}
2244 			kvm_clear_exception_queue(vcpu);
2245 			break;
2246 		case SVM_EXITINTINFO_TYPE_INTR:
2247 			kvm_clear_interrupt_queue(vcpu);
2248 			break;
2249 		default:
2250 			break;
2251 		}
2252 	}
2253 
2254 	if (reason != TASK_SWITCH_GATE ||
2255 	    int_type == SVM_EXITINTINFO_TYPE_SOFT ||
2256 	    (int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
2257 	     (int_vec == OF_VECTOR || int_vec == BP_VECTOR))) {
2258 		if (!skip_emulated_instruction(vcpu))
2259 			return 0;
2260 	}
2261 
2262 	if (int_type != SVM_EXITINTINFO_TYPE_SOFT)
2263 		int_vec = -1;
2264 
2265 	return kvm_task_switch(vcpu, tss_selector, int_vec, reason,
2266 			       has_error_code, error_code);
2267 }
2268 
2269 static int iret_interception(struct kvm_vcpu *vcpu)
2270 {
2271 	struct vcpu_svm *svm = to_svm(vcpu);
2272 
2273 	++vcpu->stat.nmi_window_exits;
2274 	vcpu->arch.hflags |= HF_IRET_MASK;
2275 	if (!sev_es_guest(vcpu->kvm)) {
2276 		svm_clr_intercept(svm, INTERCEPT_IRET);
2277 		svm->nmi_iret_rip = kvm_rip_read(vcpu);
2278 	}
2279 	kvm_make_request(KVM_REQ_EVENT, vcpu);
2280 	return 1;
2281 }
2282 
2283 static int invlpg_interception(struct kvm_vcpu *vcpu)
2284 {
2285 	if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
2286 		return kvm_emulate_instruction(vcpu, 0);
2287 
2288 	kvm_mmu_invlpg(vcpu, to_svm(vcpu)->vmcb->control.exit_info_1);
2289 	return kvm_skip_emulated_instruction(vcpu);
2290 }
2291 
2292 static int emulate_on_interception(struct kvm_vcpu *vcpu)
2293 {
2294 	return kvm_emulate_instruction(vcpu, 0);
2295 }
2296 
2297 static int rsm_interception(struct kvm_vcpu *vcpu)
2298 {
2299 	return kvm_emulate_instruction_from_buffer(vcpu, rsm_ins_bytes, 2);
2300 }
2301 
2302 static bool check_selective_cr0_intercepted(struct kvm_vcpu *vcpu,
2303 					    unsigned long val)
2304 {
2305 	struct vcpu_svm *svm = to_svm(vcpu);
2306 	unsigned long cr0 = vcpu->arch.cr0;
2307 	bool ret = false;
2308 
2309 	if (!is_guest_mode(vcpu) ||
2310 	    (!(vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_SELECTIVE_CR0))))
2311 		return false;
2312 
2313 	cr0 &= ~SVM_CR0_SELECTIVE_MASK;
2314 	val &= ~SVM_CR0_SELECTIVE_MASK;
2315 
2316 	if (cr0 ^ val) {
2317 		svm->vmcb->control.exit_code = SVM_EXIT_CR0_SEL_WRITE;
2318 		ret = (nested_svm_exit_handled(svm) == NESTED_EXIT_DONE);
2319 	}
2320 
2321 	return ret;
2322 }
2323 
2324 #define CR_VALID (1ULL << 63)
2325 
2326 static int cr_interception(struct kvm_vcpu *vcpu)
2327 {
2328 	struct vcpu_svm *svm = to_svm(vcpu);
2329 	int reg, cr;
2330 	unsigned long val;
2331 	int err;
2332 
2333 	if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
2334 		return emulate_on_interception(vcpu);
2335 
2336 	if (unlikely((svm->vmcb->control.exit_info_1 & CR_VALID) == 0))
2337 		return emulate_on_interception(vcpu);
2338 
2339 	reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
2340 	if (svm->vmcb->control.exit_code == SVM_EXIT_CR0_SEL_WRITE)
2341 		cr = SVM_EXIT_WRITE_CR0 - SVM_EXIT_READ_CR0;
2342 	else
2343 		cr = svm->vmcb->control.exit_code - SVM_EXIT_READ_CR0;
2344 
2345 	err = 0;
2346 	if (cr >= 16) { /* mov to cr */
2347 		cr -= 16;
2348 		val = kvm_register_read(vcpu, reg);
2349 		trace_kvm_cr_write(cr, val);
2350 		switch (cr) {
2351 		case 0:
2352 			if (!check_selective_cr0_intercepted(vcpu, val))
2353 				err = kvm_set_cr0(vcpu, val);
2354 			else
2355 				return 1;
2356 
2357 			break;
2358 		case 3:
2359 			err = kvm_set_cr3(vcpu, val);
2360 			break;
2361 		case 4:
2362 			err = kvm_set_cr4(vcpu, val);
2363 			break;
2364 		case 8:
2365 			err = kvm_set_cr8(vcpu, val);
2366 			break;
2367 		default:
2368 			WARN(1, "unhandled write to CR%d", cr);
2369 			kvm_queue_exception(vcpu, UD_VECTOR);
2370 			return 1;
2371 		}
2372 	} else { /* mov from cr */
2373 		switch (cr) {
2374 		case 0:
2375 			val = kvm_read_cr0(vcpu);
2376 			break;
2377 		case 2:
2378 			val = vcpu->arch.cr2;
2379 			break;
2380 		case 3:
2381 			val = kvm_read_cr3(vcpu);
2382 			break;
2383 		case 4:
2384 			val = kvm_read_cr4(vcpu);
2385 			break;
2386 		case 8:
2387 			val = kvm_get_cr8(vcpu);
2388 			break;
2389 		default:
2390 			WARN(1, "unhandled read from CR%d", cr);
2391 			kvm_queue_exception(vcpu, UD_VECTOR);
2392 			return 1;
2393 		}
2394 		kvm_register_write(vcpu, reg, val);
2395 		trace_kvm_cr_read(cr, val);
2396 	}
2397 	return kvm_complete_insn_gp(vcpu, err);
2398 }
2399 
2400 static int cr_trap(struct kvm_vcpu *vcpu)
2401 {
2402 	struct vcpu_svm *svm = to_svm(vcpu);
2403 	unsigned long old_value, new_value;
2404 	unsigned int cr;
2405 	int ret = 0;
2406 
2407 	new_value = (unsigned long)svm->vmcb->control.exit_info_1;
2408 
2409 	cr = svm->vmcb->control.exit_code - SVM_EXIT_CR0_WRITE_TRAP;
2410 	switch (cr) {
2411 	case 0:
2412 		old_value = kvm_read_cr0(vcpu);
2413 		svm_set_cr0(vcpu, new_value);
2414 
2415 		kvm_post_set_cr0(vcpu, old_value, new_value);
2416 		break;
2417 	case 4:
2418 		old_value = kvm_read_cr4(vcpu);
2419 		svm_set_cr4(vcpu, new_value);
2420 
2421 		kvm_post_set_cr4(vcpu, old_value, new_value);
2422 		break;
2423 	case 8:
2424 		ret = kvm_set_cr8(vcpu, new_value);
2425 		break;
2426 	default:
2427 		WARN(1, "unhandled CR%d write trap", cr);
2428 		kvm_queue_exception(vcpu, UD_VECTOR);
2429 		return 1;
2430 	}
2431 
2432 	return kvm_complete_insn_gp(vcpu, ret);
2433 }
2434 
2435 static int dr_interception(struct kvm_vcpu *vcpu)
2436 {
2437 	struct vcpu_svm *svm = to_svm(vcpu);
2438 	int reg, dr;
2439 	unsigned long val;
2440 	int err = 0;
2441 
2442 	if (vcpu->guest_debug == 0) {
2443 		/*
2444 		 * No more DR vmexits; force a reload of the debug registers
2445 		 * and reenter on this instruction.  The next vmexit will
2446 		 * retrieve the full state of the debug registers.
2447 		 */
2448 		clr_dr_intercepts(svm);
2449 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
2450 		return 1;
2451 	}
2452 
2453 	if (!boot_cpu_has(X86_FEATURE_DECODEASSISTS))
2454 		return emulate_on_interception(vcpu);
2455 
2456 	reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
2457 	dr = svm->vmcb->control.exit_code - SVM_EXIT_READ_DR0;
2458 	if (dr >= 16) { /* mov to DRn  */
2459 		dr -= 16;
2460 		val = kvm_register_read(vcpu, reg);
2461 		err = kvm_set_dr(vcpu, dr, val);
2462 	} else {
2463 		kvm_get_dr(vcpu, dr, &val);
2464 		kvm_register_write(vcpu, reg, val);
2465 	}
2466 
2467 	return kvm_complete_insn_gp(vcpu, err);
2468 }
2469 
2470 static int cr8_write_interception(struct kvm_vcpu *vcpu)
2471 {
2472 	int r;
2473 
2474 	u8 cr8_prev = kvm_get_cr8(vcpu);
2475 	/* instruction emulation calls kvm_set_cr8() */
2476 	r = cr_interception(vcpu);
2477 	if (lapic_in_kernel(vcpu))
2478 		return r;
2479 	if (cr8_prev <= kvm_get_cr8(vcpu))
2480 		return r;
2481 	vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
2482 	return 0;
2483 }
2484 
2485 static int efer_trap(struct kvm_vcpu *vcpu)
2486 {
2487 	struct msr_data msr_info;
2488 	int ret;
2489 
2490 	/*
2491 	 * Clear the EFER_SVME bit from EFER. The SVM code always sets this
2492 	 * bit in svm_set_efer(), but __kvm_valid_efer() checks it against
2493 	 * whether the guest has X86_FEATURE_SVM - this avoids a failure if
2494 	 * the guest doesn't have X86_FEATURE_SVM.
2495 	 */
2496 	msr_info.host_initiated = false;
2497 	msr_info.index = MSR_EFER;
2498 	msr_info.data = to_svm(vcpu)->vmcb->control.exit_info_1 & ~EFER_SVME;
2499 	ret = kvm_set_msr_common(vcpu, &msr_info);
2500 
2501 	return kvm_complete_insn_gp(vcpu, ret);
2502 }
2503 
2504 static int svm_get_msr_feature(struct kvm_msr_entry *msr)
2505 {
2506 	msr->data = 0;
2507 
2508 	switch (msr->index) {
2509 	case MSR_F10H_DECFG:
2510 		if (boot_cpu_has(X86_FEATURE_LFENCE_RDTSC))
2511 			msr->data |= MSR_F10H_DECFG_LFENCE_SERIALIZE;
2512 		break;
2513 	case MSR_IA32_PERF_CAPABILITIES:
2514 		return 0;
2515 	default:
2516 		return KVM_MSR_RET_INVALID;
2517 	}
2518 
2519 	return 0;
2520 }
2521 
2522 static int svm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2523 {
2524 	struct vcpu_svm *svm = to_svm(vcpu);
2525 
2526 	switch (msr_info->index) {
2527 	case MSR_AMD64_TSC_RATIO:
2528 		if (!msr_info->host_initiated && !svm->tsc_scaling_enabled)
2529 			return 1;
2530 		msr_info->data = svm->tsc_ratio_msr;
2531 		break;
2532 	case MSR_STAR:
2533 		msr_info->data = svm->vmcb01.ptr->save.star;
2534 		break;
2535 #ifdef CONFIG_X86_64
2536 	case MSR_LSTAR:
2537 		msr_info->data = svm->vmcb01.ptr->save.lstar;
2538 		break;
2539 	case MSR_CSTAR:
2540 		msr_info->data = svm->vmcb01.ptr->save.cstar;
2541 		break;
2542 	case MSR_KERNEL_GS_BASE:
2543 		msr_info->data = svm->vmcb01.ptr->save.kernel_gs_base;
2544 		break;
2545 	case MSR_SYSCALL_MASK:
2546 		msr_info->data = svm->vmcb01.ptr->save.sfmask;
2547 		break;
2548 #endif
2549 	case MSR_IA32_SYSENTER_CS:
2550 		msr_info->data = svm->vmcb01.ptr->save.sysenter_cs;
2551 		break;
2552 	case MSR_IA32_SYSENTER_EIP:
2553 		msr_info->data = (u32)svm->vmcb01.ptr->save.sysenter_eip;
2554 		if (guest_cpuid_is_intel(vcpu))
2555 			msr_info->data |= (u64)svm->sysenter_eip_hi << 32;
2556 		break;
2557 	case MSR_IA32_SYSENTER_ESP:
2558 		msr_info->data = svm->vmcb01.ptr->save.sysenter_esp;
2559 		if (guest_cpuid_is_intel(vcpu))
2560 			msr_info->data |= (u64)svm->sysenter_esp_hi << 32;
2561 		break;
2562 	case MSR_TSC_AUX:
2563 		msr_info->data = svm->tsc_aux;
2564 		break;
2565 	/*
2566 	 * Nobody will change the following 5 values in the VMCB so we can
2567 	 * safely return them on rdmsr. They will always be 0 until LBRV is
2568 	 * implemented.
2569 	 */
2570 	case MSR_IA32_DEBUGCTLMSR:
2571 		msr_info->data = svm->vmcb->save.dbgctl;
2572 		break;
2573 	case MSR_IA32_LASTBRANCHFROMIP:
2574 		msr_info->data = svm->vmcb->save.br_from;
2575 		break;
2576 	case MSR_IA32_LASTBRANCHTOIP:
2577 		msr_info->data = svm->vmcb->save.br_to;
2578 		break;
2579 	case MSR_IA32_LASTINTFROMIP:
2580 		msr_info->data = svm->vmcb->save.last_excp_from;
2581 		break;
2582 	case MSR_IA32_LASTINTTOIP:
2583 		msr_info->data = svm->vmcb->save.last_excp_to;
2584 		break;
2585 	case MSR_VM_HSAVE_PA:
2586 		msr_info->data = svm->nested.hsave_msr;
2587 		break;
2588 	case MSR_VM_CR:
2589 		msr_info->data = svm->nested.vm_cr_msr;
2590 		break;
2591 	case MSR_IA32_SPEC_CTRL:
2592 		if (!msr_info->host_initiated &&
2593 		    !guest_has_spec_ctrl_msr(vcpu))
2594 			return 1;
2595 
2596 		if (boot_cpu_has(X86_FEATURE_V_SPEC_CTRL))
2597 			msr_info->data = svm->vmcb->save.spec_ctrl;
2598 		else
2599 			msr_info->data = svm->spec_ctrl;
2600 		break;
2601 	case MSR_AMD64_VIRT_SPEC_CTRL:
2602 		if (!msr_info->host_initiated &&
2603 		    !guest_cpuid_has(vcpu, X86_FEATURE_VIRT_SSBD))
2604 			return 1;
2605 
2606 		msr_info->data = svm->virt_spec_ctrl;
2607 		break;
2608 	case MSR_F15H_IC_CFG: {
2609 
2610 		int family, model;
2611 
2612 		family = guest_cpuid_family(vcpu);
2613 		model  = guest_cpuid_model(vcpu);
2614 
2615 		if (family < 0 || model < 0)
2616 			return kvm_get_msr_common(vcpu, msr_info);
2617 
2618 		msr_info->data = 0;
2619 
2620 		if (family == 0x15 &&
2621 		    (model >= 0x2 && model < 0x20))
2622 			msr_info->data = 0x1E;
2623 		}
2624 		break;
2625 	case MSR_F10H_DECFG:
2626 		msr_info->data = svm->msr_decfg;
2627 		break;
2628 	default:
2629 		return kvm_get_msr_common(vcpu, msr_info);
2630 	}
2631 	return 0;
2632 }
2633 
2634 static int svm_complete_emulated_msr(struct kvm_vcpu *vcpu, int err)
2635 {
2636 	struct vcpu_svm *svm = to_svm(vcpu);
2637 	if (!err || !sev_es_guest(vcpu->kvm) || WARN_ON_ONCE(!svm->sev_es.ghcb))
2638 		return kvm_complete_insn_gp(vcpu, err);
2639 
2640 	ghcb_set_sw_exit_info_1(svm->sev_es.ghcb, 1);
2641 	ghcb_set_sw_exit_info_2(svm->sev_es.ghcb,
2642 				X86_TRAP_GP |
2643 				SVM_EVTINJ_TYPE_EXEPT |
2644 				SVM_EVTINJ_VALID);
2645 	return 1;
2646 }
2647 
2648 static int svm_set_vm_cr(struct kvm_vcpu *vcpu, u64 data)
2649 {
2650 	struct vcpu_svm *svm = to_svm(vcpu);
2651 	int svm_dis, chg_mask;
2652 
2653 	if (data & ~SVM_VM_CR_VALID_MASK)
2654 		return 1;
2655 
2656 	chg_mask = SVM_VM_CR_VALID_MASK;
2657 
2658 	if (svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK)
2659 		chg_mask &= ~(SVM_VM_CR_SVM_LOCK_MASK | SVM_VM_CR_SVM_DIS_MASK);
2660 
2661 	svm->nested.vm_cr_msr &= ~chg_mask;
2662 	svm->nested.vm_cr_msr |= (data & chg_mask);
2663 
2664 	svm_dis = svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK;
2665 
2666 	/* check for svm_disable while efer.svme is set */
2667 	if (svm_dis && (vcpu->arch.efer & EFER_SVME))
2668 		return 1;
2669 
2670 	return 0;
2671 }
2672 
2673 static int svm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2674 {
2675 	struct vcpu_svm *svm = to_svm(vcpu);
2676 	int r;
2677 
2678 	u32 ecx = msr->index;
2679 	u64 data = msr->data;
2680 	switch (ecx) {
2681 	case MSR_AMD64_TSC_RATIO:
2682 		if (!msr->host_initiated && !svm->tsc_scaling_enabled)
2683 			return 1;
2684 
2685 		if (data & TSC_RATIO_RSVD)
2686 			return 1;
2687 
2688 		svm->tsc_ratio_msr = data;
2689 
2690 		if (svm->tsc_scaling_enabled && is_guest_mode(vcpu))
2691 			nested_svm_update_tsc_ratio_msr(vcpu);
2692 
2693 		break;
2694 	case MSR_IA32_CR_PAT:
2695 		if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
2696 			return 1;
2697 		vcpu->arch.pat = data;
2698 		svm->vmcb01.ptr->save.g_pat = data;
2699 		if (is_guest_mode(vcpu))
2700 			nested_vmcb02_compute_g_pat(svm);
2701 		vmcb_mark_dirty(svm->vmcb, VMCB_NPT);
2702 		break;
2703 	case MSR_IA32_SPEC_CTRL:
2704 		if (!msr->host_initiated &&
2705 		    !guest_has_spec_ctrl_msr(vcpu))
2706 			return 1;
2707 
2708 		if (kvm_spec_ctrl_test_value(data))
2709 			return 1;
2710 
2711 		if (boot_cpu_has(X86_FEATURE_V_SPEC_CTRL))
2712 			svm->vmcb->save.spec_ctrl = data;
2713 		else
2714 			svm->spec_ctrl = data;
2715 		if (!data)
2716 			break;
2717 
2718 		/*
2719 		 * For non-nested:
2720 		 * When it's written (to non-zero) for the first time, pass
2721 		 * it through.
2722 		 *
2723 		 * For nested:
2724 		 * The handling of the MSR bitmap for L2 guests is done in
2725 		 * nested_svm_vmrun_msrpm.
2726 		 * We update the L1 MSR bit as well since it will end up
2727 		 * touching the MSR anyway now.
2728 		 */
2729 		set_msr_interception(vcpu, svm->msrpm, MSR_IA32_SPEC_CTRL, 1, 1);
2730 		break;
2731 	case MSR_IA32_PRED_CMD:
2732 		if (!msr->host_initiated &&
2733 		    !guest_has_pred_cmd_msr(vcpu))
2734 			return 1;
2735 
2736 		if (data & ~PRED_CMD_IBPB)
2737 			return 1;
2738 		if (!boot_cpu_has(X86_FEATURE_IBPB))
2739 			return 1;
2740 		if (!data)
2741 			break;
2742 
2743 		wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
2744 		set_msr_interception(vcpu, svm->msrpm, MSR_IA32_PRED_CMD, 0, 1);
2745 		break;
2746 	case MSR_AMD64_VIRT_SPEC_CTRL:
2747 		if (!msr->host_initiated &&
2748 		    !guest_cpuid_has(vcpu, X86_FEATURE_VIRT_SSBD))
2749 			return 1;
2750 
2751 		if (data & ~SPEC_CTRL_SSBD)
2752 			return 1;
2753 
2754 		svm->virt_spec_ctrl = data;
2755 		break;
2756 	case MSR_STAR:
2757 		svm->vmcb01.ptr->save.star = data;
2758 		break;
2759 #ifdef CONFIG_X86_64
2760 	case MSR_LSTAR:
2761 		svm->vmcb01.ptr->save.lstar = data;
2762 		break;
2763 	case MSR_CSTAR:
2764 		svm->vmcb01.ptr->save.cstar = data;
2765 		break;
2766 	case MSR_KERNEL_GS_BASE:
2767 		svm->vmcb01.ptr->save.kernel_gs_base = data;
2768 		break;
2769 	case MSR_SYSCALL_MASK:
2770 		svm->vmcb01.ptr->save.sfmask = data;
2771 		break;
2772 #endif
2773 	case MSR_IA32_SYSENTER_CS:
2774 		svm->vmcb01.ptr->save.sysenter_cs = data;
2775 		break;
2776 	case MSR_IA32_SYSENTER_EIP:
2777 		svm->vmcb01.ptr->save.sysenter_eip = (u32)data;
2778 		/*
2779 		 * We only intercept the MSR_IA32_SYSENTER_{EIP|ESP} msrs
2780 		 * when we spoof an Intel vendor ID (for cross vendor migration).
2781 		 * In this case we use this intercept to track the high
2782 		 * 32 bit part of these msrs to support Intel's
2783 		 * implementation of SYSENTER/SYSEXIT.
2784 		 */
2785 		svm->sysenter_eip_hi = guest_cpuid_is_intel(vcpu) ? (data >> 32) : 0;
2786 		break;
2787 	case MSR_IA32_SYSENTER_ESP:
2788 		svm->vmcb01.ptr->save.sysenter_esp = (u32)data;
2789 		svm->sysenter_esp_hi = guest_cpuid_is_intel(vcpu) ? (data >> 32) : 0;
2790 		break;
2791 	case MSR_TSC_AUX:
2792 		/*
2793 		 * TSC_AUX is usually changed only during boot and never read
2794 		 * directly.  Intercept TSC_AUX instead of exposing it to the
2795 		 * guest via direct_access_msrs, and switch it via user return.
2796 		 */
2797 		preempt_disable();
2798 		r = kvm_set_user_return_msr(tsc_aux_uret_slot, data, -1ull);
2799 		preempt_enable();
2800 		if (r)
2801 			return 1;
2802 
2803 		svm->tsc_aux = data;
2804 		break;
2805 	case MSR_IA32_DEBUGCTLMSR:
2806 		if (!lbrv) {
2807 			vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
2808 				    __func__, data);
2809 			break;
2810 		}
2811 		if (data & DEBUGCTL_RESERVED_BITS)
2812 			return 1;
2813 
2814 		svm->vmcb->save.dbgctl = data;
2815 		vmcb_mark_dirty(svm->vmcb, VMCB_LBR);
2816 		if (data & (1ULL<<0))
2817 			svm_enable_lbrv(vcpu);
2818 		else
2819 			svm_disable_lbrv(vcpu);
2820 		break;
2821 	case MSR_VM_HSAVE_PA:
2822 		/*
2823 		 * Old kernels did not validate the value written to
2824 		 * MSR_VM_HSAVE_PA.  Allow KVM_SET_MSR to set an invalid
2825 		 * value to allow live migrating buggy or malicious guests
2826 		 * originating from those kernels.
2827 		 */
2828 		if (!msr->host_initiated && !page_address_valid(vcpu, data))
2829 			return 1;
2830 
2831 		svm->nested.hsave_msr = data & PAGE_MASK;
2832 		break;
2833 	case MSR_VM_CR:
2834 		return svm_set_vm_cr(vcpu, data);
2835 	case MSR_VM_IGNNE:
2836 		vcpu_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data);
2837 		break;
2838 	case MSR_F10H_DECFG: {
2839 		struct kvm_msr_entry msr_entry;
2840 
2841 		msr_entry.index = msr->index;
2842 		if (svm_get_msr_feature(&msr_entry))
2843 			return 1;
2844 
2845 		/* Check the supported bits */
2846 		if (data & ~msr_entry.data)
2847 			return 1;
2848 
2849 		/* Don't allow the guest to change a bit, #GP */
2850 		if (!msr->host_initiated && (data ^ msr_entry.data))
2851 			return 1;
2852 
2853 		svm->msr_decfg = data;
2854 		break;
2855 	}
2856 	default:
2857 		return kvm_set_msr_common(vcpu, msr);
2858 	}
2859 	return 0;
2860 }
2861 
2862 static int msr_interception(struct kvm_vcpu *vcpu)
2863 {
2864 	if (to_svm(vcpu)->vmcb->control.exit_info_1)
2865 		return kvm_emulate_wrmsr(vcpu);
2866 	else
2867 		return kvm_emulate_rdmsr(vcpu);
2868 }
2869 
2870 static int interrupt_window_interception(struct kvm_vcpu *vcpu)
2871 {
2872 	kvm_make_request(KVM_REQ_EVENT, vcpu);
2873 	svm_clear_vintr(to_svm(vcpu));
2874 
2875 	/*
2876 	 * For AVIC, the only reason to end up here is ExtINTs.
2877 	 * In this case AVIC was temporarily disabled for
2878 	 * requesting the IRQ window and we have to re-enable it.
2879 	 */
2880 	kvm_request_apicv_update(vcpu->kvm, true, APICV_INHIBIT_REASON_IRQWIN);
2881 
2882 	++vcpu->stat.irq_window_exits;
2883 	return 1;
2884 }
2885 
2886 static int pause_interception(struct kvm_vcpu *vcpu)
2887 {
2888 	bool in_kernel;
2889 
2890 	/*
2891 	 * CPL is not made available for an SEV-ES guest, therefore
2892 	 * vcpu->arch.preempted_in_kernel can never be true.  Just
2893 	 * set in_kernel to false as well.
2894 	 */
2895 	in_kernel = !sev_es_guest(vcpu->kvm) && svm_get_cpl(vcpu) == 0;
2896 
2897 	if (!kvm_pause_in_guest(vcpu->kvm))
2898 		grow_ple_window(vcpu);
2899 
2900 	kvm_vcpu_on_spin(vcpu, in_kernel);
2901 	return kvm_skip_emulated_instruction(vcpu);
2902 }
2903 
2904 static int invpcid_interception(struct kvm_vcpu *vcpu)
2905 {
2906 	struct vcpu_svm *svm = to_svm(vcpu);
2907 	unsigned long type;
2908 	gva_t gva;
2909 
2910 	if (!guest_cpuid_has(vcpu, X86_FEATURE_INVPCID)) {
2911 		kvm_queue_exception(vcpu, UD_VECTOR);
2912 		return 1;
2913 	}
2914 
2915 	/*
2916 	 * For an INVPCID intercept:
2917 	 * EXITINFO1 provides the linear address of the memory operand.
2918 	 * EXITINFO2 provides the contents of the register operand.
2919 	 */
2920 	type = svm->vmcb->control.exit_info_2;
2921 	gva = svm->vmcb->control.exit_info_1;
2922 
2923 	return kvm_handle_invpcid(vcpu, type, gva);
2924 }
2925 
2926 static int (*const svm_exit_handlers[])(struct kvm_vcpu *vcpu) = {
2927 	[SVM_EXIT_READ_CR0]			= cr_interception,
2928 	[SVM_EXIT_READ_CR3]			= cr_interception,
2929 	[SVM_EXIT_READ_CR4]			= cr_interception,
2930 	[SVM_EXIT_READ_CR8]			= cr_interception,
2931 	[SVM_EXIT_CR0_SEL_WRITE]		= cr_interception,
2932 	[SVM_EXIT_WRITE_CR0]			= cr_interception,
2933 	[SVM_EXIT_WRITE_CR3]			= cr_interception,
2934 	[SVM_EXIT_WRITE_CR4]			= cr_interception,
2935 	[SVM_EXIT_WRITE_CR8]			= cr8_write_interception,
2936 	[SVM_EXIT_READ_DR0]			= dr_interception,
2937 	[SVM_EXIT_READ_DR1]			= dr_interception,
2938 	[SVM_EXIT_READ_DR2]			= dr_interception,
2939 	[SVM_EXIT_READ_DR3]			= dr_interception,
2940 	[SVM_EXIT_READ_DR4]			= dr_interception,
2941 	[SVM_EXIT_READ_DR5]			= dr_interception,
2942 	[SVM_EXIT_READ_DR6]			= dr_interception,
2943 	[SVM_EXIT_READ_DR7]			= dr_interception,
2944 	[SVM_EXIT_WRITE_DR0]			= dr_interception,
2945 	[SVM_EXIT_WRITE_DR1]			= dr_interception,
2946 	[SVM_EXIT_WRITE_DR2]			= dr_interception,
2947 	[SVM_EXIT_WRITE_DR3]			= dr_interception,
2948 	[SVM_EXIT_WRITE_DR4]			= dr_interception,
2949 	[SVM_EXIT_WRITE_DR5]			= dr_interception,
2950 	[SVM_EXIT_WRITE_DR6]			= dr_interception,
2951 	[SVM_EXIT_WRITE_DR7]			= dr_interception,
2952 	[SVM_EXIT_EXCP_BASE + DB_VECTOR]	= db_interception,
2953 	[SVM_EXIT_EXCP_BASE + BP_VECTOR]	= bp_interception,
2954 	[SVM_EXIT_EXCP_BASE + UD_VECTOR]	= ud_interception,
2955 	[SVM_EXIT_EXCP_BASE + PF_VECTOR]	= pf_interception,
2956 	[SVM_EXIT_EXCP_BASE + MC_VECTOR]	= mc_interception,
2957 	[SVM_EXIT_EXCP_BASE + AC_VECTOR]	= ac_interception,
2958 	[SVM_EXIT_EXCP_BASE + GP_VECTOR]	= gp_interception,
2959 	[SVM_EXIT_INTR]				= intr_interception,
2960 	[SVM_EXIT_NMI]				= nmi_interception,
2961 	[SVM_EXIT_SMI]				= smi_interception,
2962 	[SVM_EXIT_VINTR]			= interrupt_window_interception,
2963 	[SVM_EXIT_RDPMC]			= kvm_emulate_rdpmc,
2964 	[SVM_EXIT_CPUID]			= kvm_emulate_cpuid,
2965 	[SVM_EXIT_IRET]                         = iret_interception,
2966 	[SVM_EXIT_INVD]                         = kvm_emulate_invd,
2967 	[SVM_EXIT_PAUSE]			= pause_interception,
2968 	[SVM_EXIT_HLT]				= kvm_emulate_halt,
2969 	[SVM_EXIT_INVLPG]			= invlpg_interception,
2970 	[SVM_EXIT_INVLPGA]			= invlpga_interception,
2971 	[SVM_EXIT_IOIO]				= io_interception,
2972 	[SVM_EXIT_MSR]				= msr_interception,
2973 	[SVM_EXIT_TASK_SWITCH]			= task_switch_interception,
2974 	[SVM_EXIT_SHUTDOWN]			= shutdown_interception,
2975 	[SVM_EXIT_VMRUN]			= vmrun_interception,
2976 	[SVM_EXIT_VMMCALL]			= kvm_emulate_hypercall,
2977 	[SVM_EXIT_VMLOAD]			= vmload_interception,
2978 	[SVM_EXIT_VMSAVE]			= vmsave_interception,
2979 	[SVM_EXIT_STGI]				= stgi_interception,
2980 	[SVM_EXIT_CLGI]				= clgi_interception,
2981 	[SVM_EXIT_SKINIT]			= skinit_interception,
2982 	[SVM_EXIT_RDTSCP]			= kvm_handle_invalid_op,
2983 	[SVM_EXIT_WBINVD]                       = kvm_emulate_wbinvd,
2984 	[SVM_EXIT_MONITOR]			= kvm_emulate_monitor,
2985 	[SVM_EXIT_MWAIT]			= kvm_emulate_mwait,
2986 	[SVM_EXIT_XSETBV]			= kvm_emulate_xsetbv,
2987 	[SVM_EXIT_RDPRU]			= kvm_handle_invalid_op,
2988 	[SVM_EXIT_EFER_WRITE_TRAP]		= efer_trap,
2989 	[SVM_EXIT_CR0_WRITE_TRAP]		= cr_trap,
2990 	[SVM_EXIT_CR4_WRITE_TRAP]		= cr_trap,
2991 	[SVM_EXIT_CR8_WRITE_TRAP]		= cr_trap,
2992 	[SVM_EXIT_INVPCID]                      = invpcid_interception,
2993 	[SVM_EXIT_NPF]				= npf_interception,
2994 	[SVM_EXIT_RSM]                          = rsm_interception,
2995 	[SVM_EXIT_AVIC_INCOMPLETE_IPI]		= avic_incomplete_ipi_interception,
2996 	[SVM_EXIT_AVIC_UNACCELERATED_ACCESS]	= avic_unaccelerated_access_interception,
2997 	[SVM_EXIT_VMGEXIT]			= sev_handle_vmgexit,
2998 };
2999 
3000 static void dump_vmcb(struct kvm_vcpu *vcpu)
3001 {
3002 	struct vcpu_svm *svm = to_svm(vcpu);
3003 	struct vmcb_control_area *control = &svm->vmcb->control;
3004 	struct vmcb_save_area *save = &svm->vmcb->save;
3005 	struct vmcb_save_area *save01 = &svm->vmcb01.ptr->save;
3006 
3007 	if (!dump_invalid_vmcb) {
3008 		pr_warn_ratelimited("set kvm_amd.dump_invalid_vmcb=1 to dump internal KVM state.\n");
3009 		return;
3010 	}
3011 
3012 	pr_err("VMCB %p, last attempted VMRUN on CPU %d\n",
3013 	       svm->current_vmcb->ptr, vcpu->arch.last_vmentry_cpu);
3014 	pr_err("VMCB Control Area:\n");
3015 	pr_err("%-20s%04x\n", "cr_read:", control->intercepts[INTERCEPT_CR] & 0xffff);
3016 	pr_err("%-20s%04x\n", "cr_write:", control->intercepts[INTERCEPT_CR] >> 16);
3017 	pr_err("%-20s%04x\n", "dr_read:", control->intercepts[INTERCEPT_DR] & 0xffff);
3018 	pr_err("%-20s%04x\n", "dr_write:", control->intercepts[INTERCEPT_DR] >> 16);
3019 	pr_err("%-20s%08x\n", "exceptions:", control->intercepts[INTERCEPT_EXCEPTION]);
3020 	pr_err("%-20s%08x %08x\n", "intercepts:",
3021               control->intercepts[INTERCEPT_WORD3],
3022 	       control->intercepts[INTERCEPT_WORD4]);
3023 	pr_err("%-20s%d\n", "pause filter count:", control->pause_filter_count);
3024 	pr_err("%-20s%d\n", "pause filter threshold:",
3025 	       control->pause_filter_thresh);
3026 	pr_err("%-20s%016llx\n", "iopm_base_pa:", control->iopm_base_pa);
3027 	pr_err("%-20s%016llx\n", "msrpm_base_pa:", control->msrpm_base_pa);
3028 	pr_err("%-20s%016llx\n", "tsc_offset:", control->tsc_offset);
3029 	pr_err("%-20s%d\n", "asid:", control->asid);
3030 	pr_err("%-20s%d\n", "tlb_ctl:", control->tlb_ctl);
3031 	pr_err("%-20s%08x\n", "int_ctl:", control->int_ctl);
3032 	pr_err("%-20s%08x\n", "int_vector:", control->int_vector);
3033 	pr_err("%-20s%08x\n", "int_state:", control->int_state);
3034 	pr_err("%-20s%08x\n", "exit_code:", control->exit_code);
3035 	pr_err("%-20s%016llx\n", "exit_info1:", control->exit_info_1);
3036 	pr_err("%-20s%016llx\n", "exit_info2:", control->exit_info_2);
3037 	pr_err("%-20s%08x\n", "exit_int_info:", control->exit_int_info);
3038 	pr_err("%-20s%08x\n", "exit_int_info_err:", control->exit_int_info_err);
3039 	pr_err("%-20s%lld\n", "nested_ctl:", control->nested_ctl);
3040 	pr_err("%-20s%016llx\n", "nested_cr3:", control->nested_cr3);
3041 	pr_err("%-20s%016llx\n", "avic_vapic_bar:", control->avic_vapic_bar);
3042 	pr_err("%-20s%016llx\n", "ghcb:", control->ghcb_gpa);
3043 	pr_err("%-20s%08x\n", "event_inj:", control->event_inj);
3044 	pr_err("%-20s%08x\n", "event_inj_err:", control->event_inj_err);
3045 	pr_err("%-20s%lld\n", "virt_ext:", control->virt_ext);
3046 	pr_err("%-20s%016llx\n", "next_rip:", control->next_rip);
3047 	pr_err("%-20s%016llx\n", "avic_backing_page:", control->avic_backing_page);
3048 	pr_err("%-20s%016llx\n", "avic_logical_id:", control->avic_logical_id);
3049 	pr_err("%-20s%016llx\n", "avic_physical_id:", control->avic_physical_id);
3050 	pr_err("%-20s%016llx\n", "vmsa_pa:", control->vmsa_pa);
3051 	pr_err("VMCB State Save Area:\n");
3052 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3053 	       "es:",
3054 	       save->es.selector, save->es.attrib,
3055 	       save->es.limit, save->es.base);
3056 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3057 	       "cs:",
3058 	       save->cs.selector, save->cs.attrib,
3059 	       save->cs.limit, save->cs.base);
3060 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3061 	       "ss:",
3062 	       save->ss.selector, save->ss.attrib,
3063 	       save->ss.limit, save->ss.base);
3064 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3065 	       "ds:",
3066 	       save->ds.selector, save->ds.attrib,
3067 	       save->ds.limit, save->ds.base);
3068 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3069 	       "fs:",
3070 	       save01->fs.selector, save01->fs.attrib,
3071 	       save01->fs.limit, save01->fs.base);
3072 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3073 	       "gs:",
3074 	       save01->gs.selector, save01->gs.attrib,
3075 	       save01->gs.limit, save01->gs.base);
3076 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3077 	       "gdtr:",
3078 	       save->gdtr.selector, save->gdtr.attrib,
3079 	       save->gdtr.limit, save->gdtr.base);
3080 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3081 	       "ldtr:",
3082 	       save01->ldtr.selector, save01->ldtr.attrib,
3083 	       save01->ldtr.limit, save01->ldtr.base);
3084 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3085 	       "idtr:",
3086 	       save->idtr.selector, save->idtr.attrib,
3087 	       save->idtr.limit, save->idtr.base);
3088 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3089 	       "tr:",
3090 	       save01->tr.selector, save01->tr.attrib,
3091 	       save01->tr.limit, save01->tr.base);
3092 	pr_err("cpl:            %d                efer:         %016llx\n",
3093 		save->cpl, save->efer);
3094 	pr_err("%-15s %016llx %-13s %016llx\n",
3095 	       "cr0:", save->cr0, "cr2:", save->cr2);
3096 	pr_err("%-15s %016llx %-13s %016llx\n",
3097 	       "cr3:", save->cr3, "cr4:", save->cr4);
3098 	pr_err("%-15s %016llx %-13s %016llx\n",
3099 	       "dr6:", save->dr6, "dr7:", save->dr7);
3100 	pr_err("%-15s %016llx %-13s %016llx\n",
3101 	       "rip:", save->rip, "rflags:", save->rflags);
3102 	pr_err("%-15s %016llx %-13s %016llx\n",
3103 	       "rsp:", save->rsp, "rax:", save->rax);
3104 	pr_err("%-15s %016llx %-13s %016llx\n",
3105 	       "star:", save01->star, "lstar:", save01->lstar);
3106 	pr_err("%-15s %016llx %-13s %016llx\n",
3107 	       "cstar:", save01->cstar, "sfmask:", save01->sfmask);
3108 	pr_err("%-15s %016llx %-13s %016llx\n",
3109 	       "kernel_gs_base:", save01->kernel_gs_base,
3110 	       "sysenter_cs:", save01->sysenter_cs);
3111 	pr_err("%-15s %016llx %-13s %016llx\n",
3112 	       "sysenter_esp:", save01->sysenter_esp,
3113 	       "sysenter_eip:", save01->sysenter_eip);
3114 	pr_err("%-15s %016llx %-13s %016llx\n",
3115 	       "gpat:", save->g_pat, "dbgctl:", save->dbgctl);
3116 	pr_err("%-15s %016llx %-13s %016llx\n",
3117 	       "br_from:", save->br_from, "br_to:", save->br_to);
3118 	pr_err("%-15s %016llx %-13s %016llx\n",
3119 	       "excp_from:", save->last_excp_from,
3120 	       "excp_to:", save->last_excp_to);
3121 }
3122 
3123 static bool svm_check_exit_valid(struct kvm_vcpu *vcpu, u64 exit_code)
3124 {
3125 	return (exit_code < ARRAY_SIZE(svm_exit_handlers) &&
3126 		svm_exit_handlers[exit_code]);
3127 }
3128 
3129 static int svm_handle_invalid_exit(struct kvm_vcpu *vcpu, u64 exit_code)
3130 {
3131 	vcpu_unimpl(vcpu, "svm: unexpected exit reason 0x%llx\n", exit_code);
3132 	dump_vmcb(vcpu);
3133 	vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3134 	vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON;
3135 	vcpu->run->internal.ndata = 2;
3136 	vcpu->run->internal.data[0] = exit_code;
3137 	vcpu->run->internal.data[1] = vcpu->arch.last_vmentry_cpu;
3138 	return 0;
3139 }
3140 
3141 int svm_invoke_exit_handler(struct kvm_vcpu *vcpu, u64 exit_code)
3142 {
3143 	if (!svm_check_exit_valid(vcpu, exit_code))
3144 		return svm_handle_invalid_exit(vcpu, exit_code);
3145 
3146 #ifdef CONFIG_RETPOLINE
3147 	if (exit_code == SVM_EXIT_MSR)
3148 		return msr_interception(vcpu);
3149 	else if (exit_code == SVM_EXIT_VINTR)
3150 		return interrupt_window_interception(vcpu);
3151 	else if (exit_code == SVM_EXIT_INTR)
3152 		return intr_interception(vcpu);
3153 	else if (exit_code == SVM_EXIT_HLT)
3154 		return kvm_emulate_halt(vcpu);
3155 	else if (exit_code == SVM_EXIT_NPF)
3156 		return npf_interception(vcpu);
3157 #endif
3158 	return svm_exit_handlers[exit_code](vcpu);
3159 }
3160 
3161 static void svm_get_exit_info(struct kvm_vcpu *vcpu, u32 *reason,
3162 			      u64 *info1, u64 *info2,
3163 			      u32 *intr_info, u32 *error_code)
3164 {
3165 	struct vmcb_control_area *control = &to_svm(vcpu)->vmcb->control;
3166 
3167 	*reason = control->exit_code;
3168 	*info1 = control->exit_info_1;
3169 	*info2 = control->exit_info_2;
3170 	*intr_info = control->exit_int_info;
3171 	if ((*intr_info & SVM_EXITINTINFO_VALID) &&
3172 	    (*intr_info & SVM_EXITINTINFO_VALID_ERR))
3173 		*error_code = control->exit_int_info_err;
3174 	else
3175 		*error_code = 0;
3176 }
3177 
3178 static int handle_exit(struct kvm_vcpu *vcpu, fastpath_t exit_fastpath)
3179 {
3180 	struct vcpu_svm *svm = to_svm(vcpu);
3181 	struct kvm_run *kvm_run = vcpu->run;
3182 	u32 exit_code = svm->vmcb->control.exit_code;
3183 
3184 	trace_kvm_exit(vcpu, KVM_ISA_SVM);
3185 
3186 	/* SEV-ES guests must use the CR write traps to track CR registers. */
3187 	if (!sev_es_guest(vcpu->kvm)) {
3188 		if (!svm_is_intercept(svm, INTERCEPT_CR0_WRITE))
3189 			vcpu->arch.cr0 = svm->vmcb->save.cr0;
3190 		if (npt_enabled)
3191 			vcpu->arch.cr3 = svm->vmcb->save.cr3;
3192 	}
3193 
3194 	if (is_guest_mode(vcpu)) {
3195 		int vmexit;
3196 
3197 		trace_kvm_nested_vmexit(vcpu, KVM_ISA_SVM);
3198 
3199 		vmexit = nested_svm_exit_special(svm);
3200 
3201 		if (vmexit == NESTED_EXIT_CONTINUE)
3202 			vmexit = nested_svm_exit_handled(svm);
3203 
3204 		if (vmexit == NESTED_EXIT_DONE)
3205 			return 1;
3206 	}
3207 
3208 	if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
3209 		kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3210 		kvm_run->fail_entry.hardware_entry_failure_reason
3211 			= svm->vmcb->control.exit_code;
3212 		kvm_run->fail_entry.cpu = vcpu->arch.last_vmentry_cpu;
3213 		dump_vmcb(vcpu);
3214 		return 0;
3215 	}
3216 
3217 	if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
3218 	    exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
3219 	    exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH &&
3220 	    exit_code != SVM_EXIT_INTR && exit_code != SVM_EXIT_NMI)
3221 		printk(KERN_ERR "%s: unexpected exit_int_info 0x%x "
3222 		       "exit_code 0x%x\n",
3223 		       __func__, svm->vmcb->control.exit_int_info,
3224 		       exit_code);
3225 
3226 	if (exit_fastpath != EXIT_FASTPATH_NONE)
3227 		return 1;
3228 
3229 	return svm_invoke_exit_handler(vcpu, exit_code);
3230 }
3231 
3232 static void reload_tss(struct kvm_vcpu *vcpu)
3233 {
3234 	struct svm_cpu_data *sd = per_cpu(svm_data, vcpu->cpu);
3235 
3236 	sd->tss_desc->type = 9; /* available 32/64-bit TSS */
3237 	load_TR_desc();
3238 }
3239 
3240 static void pre_svm_run(struct kvm_vcpu *vcpu)
3241 {
3242 	struct svm_cpu_data *sd = per_cpu(svm_data, vcpu->cpu);
3243 	struct vcpu_svm *svm = to_svm(vcpu);
3244 
3245 	/*
3246 	 * If the previous vmrun of the vmcb occurred on a different physical
3247 	 * cpu, then mark the vmcb dirty and assign a new asid.  Hardware's
3248 	 * vmcb clean bits are per logical CPU, as are KVM's asid assignments.
3249 	 */
3250 	if (unlikely(svm->current_vmcb->cpu != vcpu->cpu)) {
3251 		svm->current_vmcb->asid_generation = 0;
3252 		vmcb_mark_all_dirty(svm->vmcb);
3253 		svm->current_vmcb->cpu = vcpu->cpu;
3254         }
3255 
3256 	if (sev_guest(vcpu->kvm))
3257 		return pre_sev_run(svm, vcpu->cpu);
3258 
3259 	/* FIXME: handle wraparound of asid_generation */
3260 	if (svm->current_vmcb->asid_generation != sd->asid_generation)
3261 		new_asid(svm, sd);
3262 }
3263 
3264 static void svm_inject_nmi(struct kvm_vcpu *vcpu)
3265 {
3266 	struct vcpu_svm *svm = to_svm(vcpu);
3267 
3268 	svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
3269 	vcpu->arch.hflags |= HF_NMI_MASK;
3270 	if (!sev_es_guest(vcpu->kvm))
3271 		svm_set_intercept(svm, INTERCEPT_IRET);
3272 	++vcpu->stat.nmi_injections;
3273 }
3274 
3275 static void svm_set_irq(struct kvm_vcpu *vcpu)
3276 {
3277 	struct vcpu_svm *svm = to_svm(vcpu);
3278 
3279 	BUG_ON(!(gif_set(svm)));
3280 
3281 	trace_kvm_inj_virq(vcpu->arch.interrupt.nr);
3282 	++vcpu->stat.irq_injections;
3283 
3284 	svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr |
3285 		SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR;
3286 }
3287 
3288 static void svm_update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
3289 {
3290 	struct vcpu_svm *svm = to_svm(vcpu);
3291 
3292 	/*
3293 	 * SEV-ES guests must always keep the CR intercepts cleared. CR
3294 	 * tracking is done using the CR write traps.
3295 	 */
3296 	if (sev_es_guest(vcpu->kvm))
3297 		return;
3298 
3299 	if (nested_svm_virtualize_tpr(vcpu))
3300 		return;
3301 
3302 	svm_clr_intercept(svm, INTERCEPT_CR8_WRITE);
3303 
3304 	if (irr == -1)
3305 		return;
3306 
3307 	if (tpr >= irr)
3308 		svm_set_intercept(svm, INTERCEPT_CR8_WRITE);
3309 }
3310 
3311 bool svm_nmi_blocked(struct kvm_vcpu *vcpu)
3312 {
3313 	struct vcpu_svm *svm = to_svm(vcpu);
3314 	struct vmcb *vmcb = svm->vmcb;
3315 	bool ret;
3316 
3317 	if (!gif_set(svm))
3318 		return true;
3319 
3320 	if (is_guest_mode(vcpu) && nested_exit_on_nmi(svm))
3321 		return false;
3322 
3323 	ret = (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) ||
3324 	      (vcpu->arch.hflags & HF_NMI_MASK);
3325 
3326 	return ret;
3327 }
3328 
3329 static int svm_nmi_allowed(struct kvm_vcpu *vcpu, bool for_injection)
3330 {
3331 	struct vcpu_svm *svm = to_svm(vcpu);
3332 	if (svm->nested.nested_run_pending)
3333 		return -EBUSY;
3334 
3335 	/* An NMI must not be injected into L2 if it's supposed to VM-Exit.  */
3336 	if (for_injection && is_guest_mode(vcpu) && nested_exit_on_nmi(svm))
3337 		return -EBUSY;
3338 
3339 	return !svm_nmi_blocked(vcpu);
3340 }
3341 
3342 static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu)
3343 {
3344 	return !!(vcpu->arch.hflags & HF_NMI_MASK);
3345 }
3346 
3347 static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
3348 {
3349 	struct vcpu_svm *svm = to_svm(vcpu);
3350 
3351 	if (masked) {
3352 		vcpu->arch.hflags |= HF_NMI_MASK;
3353 		if (!sev_es_guest(vcpu->kvm))
3354 			svm_set_intercept(svm, INTERCEPT_IRET);
3355 	} else {
3356 		vcpu->arch.hflags &= ~HF_NMI_MASK;
3357 		if (!sev_es_guest(vcpu->kvm))
3358 			svm_clr_intercept(svm, INTERCEPT_IRET);
3359 	}
3360 }
3361 
3362 bool svm_interrupt_blocked(struct kvm_vcpu *vcpu)
3363 {
3364 	struct vcpu_svm *svm = to_svm(vcpu);
3365 	struct vmcb *vmcb = svm->vmcb;
3366 
3367 	if (!gif_set(svm))
3368 		return true;
3369 
3370 	if (is_guest_mode(vcpu)) {
3371 		/* As long as interrupts are being delivered...  */
3372 		if ((svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK)
3373 		    ? !(svm->vmcb01.ptr->save.rflags & X86_EFLAGS_IF)
3374 		    : !(kvm_get_rflags(vcpu) & X86_EFLAGS_IF))
3375 			return true;
3376 
3377 		/* ... vmexits aren't blocked by the interrupt shadow  */
3378 		if (nested_exit_on_intr(svm))
3379 			return false;
3380 	} else {
3381 		if (!svm_get_if_flag(vcpu))
3382 			return true;
3383 	}
3384 
3385 	return (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK);
3386 }
3387 
3388 static int svm_interrupt_allowed(struct kvm_vcpu *vcpu, bool for_injection)
3389 {
3390 	struct vcpu_svm *svm = to_svm(vcpu);
3391 	if (svm->nested.nested_run_pending)
3392 		return -EBUSY;
3393 
3394 	/*
3395 	 * An IRQ must not be injected into L2 if it's supposed to VM-Exit,
3396 	 * e.g. if the IRQ arrived asynchronously after checking nested events.
3397 	 */
3398 	if (for_injection && is_guest_mode(vcpu) && nested_exit_on_intr(svm))
3399 		return -EBUSY;
3400 
3401 	return !svm_interrupt_blocked(vcpu);
3402 }
3403 
3404 static void svm_enable_irq_window(struct kvm_vcpu *vcpu)
3405 {
3406 	struct vcpu_svm *svm = to_svm(vcpu);
3407 
3408 	/*
3409 	 * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
3410 	 * 1, because that's a separate STGI/VMRUN intercept.  The next time we
3411 	 * get that intercept, this function will be called again though and
3412 	 * we'll get the vintr intercept. However, if the vGIF feature is
3413 	 * enabled, the STGI interception will not occur. Enable the irq
3414 	 * window under the assumption that the hardware will set the GIF.
3415 	 */
3416 	if (vgif_enabled(svm) || gif_set(svm)) {
3417 		/*
3418 		 * IRQ window is not needed when AVIC is enabled,
3419 		 * unless we have pending ExtINT since it cannot be injected
3420 		 * via AVIC. In such case, we need to temporarily disable AVIC,
3421 		 * and fallback to injecting IRQ via V_IRQ.
3422 		 */
3423 		kvm_request_apicv_update(vcpu->kvm, false, APICV_INHIBIT_REASON_IRQWIN);
3424 		svm_set_vintr(svm);
3425 	}
3426 }
3427 
3428 static void svm_enable_nmi_window(struct kvm_vcpu *vcpu)
3429 {
3430 	struct vcpu_svm *svm = to_svm(vcpu);
3431 
3432 	if ((vcpu->arch.hflags & (HF_NMI_MASK | HF_IRET_MASK)) == HF_NMI_MASK)
3433 		return; /* IRET will cause a vm exit */
3434 
3435 	if (!gif_set(svm)) {
3436 		if (vgif_enabled(svm))
3437 			svm_set_intercept(svm, INTERCEPT_STGI);
3438 		return; /* STGI will cause a vm exit */
3439 	}
3440 
3441 	/*
3442 	 * Something prevents NMI from been injected. Single step over possible
3443 	 * problem (IRET or exception injection or interrupt shadow)
3444 	 */
3445 	svm->nmi_singlestep_guest_rflags = svm_get_rflags(vcpu);
3446 	svm->nmi_singlestep = true;
3447 	svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
3448 }
3449 
3450 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
3451 {
3452 	return 0;
3453 }
3454 
3455 static int svm_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
3456 {
3457 	return 0;
3458 }
3459 
3460 void svm_flush_tlb(struct kvm_vcpu *vcpu)
3461 {
3462 	struct vcpu_svm *svm = to_svm(vcpu);
3463 
3464 	/*
3465 	 * Flush only the current ASID even if the TLB flush was invoked via
3466 	 * kvm_flush_remote_tlbs().  Although flushing remote TLBs requires all
3467 	 * ASIDs to be flushed, KVM uses a single ASID for L1 and L2, and
3468 	 * unconditionally does a TLB flush on both nested VM-Enter and nested
3469 	 * VM-Exit (via kvm_mmu_reset_context()).
3470 	 */
3471 	if (static_cpu_has(X86_FEATURE_FLUSHBYASID))
3472 		svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
3473 	else
3474 		svm->current_vmcb->asid_generation--;
3475 }
3476 
3477 static void svm_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t gva)
3478 {
3479 	struct vcpu_svm *svm = to_svm(vcpu);
3480 
3481 	invlpga(gva, svm->vmcb->control.asid);
3482 }
3483 
3484 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
3485 {
3486 	struct vcpu_svm *svm = to_svm(vcpu);
3487 
3488 	if (nested_svm_virtualize_tpr(vcpu))
3489 		return;
3490 
3491 	if (!svm_is_intercept(svm, INTERCEPT_CR8_WRITE)) {
3492 		int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
3493 		kvm_set_cr8(vcpu, cr8);
3494 	}
3495 }
3496 
3497 static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
3498 {
3499 	struct vcpu_svm *svm = to_svm(vcpu);
3500 	u64 cr8;
3501 
3502 	if (nested_svm_virtualize_tpr(vcpu) ||
3503 	    kvm_vcpu_apicv_active(vcpu))
3504 		return;
3505 
3506 	cr8 = kvm_get_cr8(vcpu);
3507 	svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
3508 	svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
3509 }
3510 
3511 static void svm_complete_interrupts(struct kvm_vcpu *vcpu)
3512 {
3513 	struct vcpu_svm *svm = to_svm(vcpu);
3514 	u8 vector;
3515 	int type;
3516 	u32 exitintinfo = svm->vmcb->control.exit_int_info;
3517 	unsigned int3_injected = svm->int3_injected;
3518 
3519 	svm->int3_injected = 0;
3520 
3521 	/*
3522 	 * If we've made progress since setting HF_IRET_MASK, we've
3523 	 * executed an IRET and can allow NMI injection.
3524 	 */
3525 	if ((vcpu->arch.hflags & HF_IRET_MASK) &&
3526 	    (sev_es_guest(vcpu->kvm) ||
3527 	     kvm_rip_read(vcpu) != svm->nmi_iret_rip)) {
3528 		vcpu->arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK);
3529 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3530 	}
3531 
3532 	vcpu->arch.nmi_injected = false;
3533 	kvm_clear_exception_queue(vcpu);
3534 	kvm_clear_interrupt_queue(vcpu);
3535 
3536 	if (!(exitintinfo & SVM_EXITINTINFO_VALID))
3537 		return;
3538 
3539 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3540 
3541 	vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK;
3542 	type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK;
3543 
3544 	switch (type) {
3545 	case SVM_EXITINTINFO_TYPE_NMI:
3546 		vcpu->arch.nmi_injected = true;
3547 		break;
3548 	case SVM_EXITINTINFO_TYPE_EXEPT:
3549 		/*
3550 		 * Never re-inject a #VC exception.
3551 		 */
3552 		if (vector == X86_TRAP_VC)
3553 			break;
3554 
3555 		/*
3556 		 * In case of software exceptions, do not reinject the vector,
3557 		 * but re-execute the instruction instead. Rewind RIP first
3558 		 * if we emulated INT3 before.
3559 		 */
3560 		if (kvm_exception_is_soft(vector)) {
3561 			if (vector == BP_VECTOR && int3_injected &&
3562 			    kvm_is_linear_rip(vcpu, svm->int3_rip))
3563 				kvm_rip_write(vcpu,
3564 					      kvm_rip_read(vcpu) - int3_injected);
3565 			break;
3566 		}
3567 		if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) {
3568 			u32 err = svm->vmcb->control.exit_int_info_err;
3569 			kvm_requeue_exception_e(vcpu, vector, err);
3570 
3571 		} else
3572 			kvm_requeue_exception(vcpu, vector);
3573 		break;
3574 	case SVM_EXITINTINFO_TYPE_INTR:
3575 		kvm_queue_interrupt(vcpu, vector, false);
3576 		break;
3577 	default:
3578 		break;
3579 	}
3580 }
3581 
3582 static void svm_cancel_injection(struct kvm_vcpu *vcpu)
3583 {
3584 	struct vcpu_svm *svm = to_svm(vcpu);
3585 	struct vmcb_control_area *control = &svm->vmcb->control;
3586 
3587 	control->exit_int_info = control->event_inj;
3588 	control->exit_int_info_err = control->event_inj_err;
3589 	control->event_inj = 0;
3590 	svm_complete_interrupts(vcpu);
3591 }
3592 
3593 static int svm_vcpu_pre_run(struct kvm_vcpu *vcpu)
3594 {
3595 	return 1;
3596 }
3597 
3598 static fastpath_t svm_exit_handlers_fastpath(struct kvm_vcpu *vcpu)
3599 {
3600 	if (to_svm(vcpu)->vmcb->control.exit_code == SVM_EXIT_MSR &&
3601 	    to_svm(vcpu)->vmcb->control.exit_info_1)
3602 		return handle_fastpath_set_msr_irqoff(vcpu);
3603 
3604 	return EXIT_FASTPATH_NONE;
3605 }
3606 
3607 static noinstr void svm_vcpu_enter_exit(struct kvm_vcpu *vcpu)
3608 {
3609 	struct vcpu_svm *svm = to_svm(vcpu);
3610 	unsigned long vmcb_pa = svm->current_vmcb->pa;
3611 
3612 	kvm_guest_enter_irqoff();
3613 
3614 	if (sev_es_guest(vcpu->kvm)) {
3615 		__svm_sev_es_vcpu_run(vmcb_pa);
3616 	} else {
3617 		struct svm_cpu_data *sd = per_cpu(svm_data, vcpu->cpu);
3618 
3619 		/*
3620 		 * Use a single vmcb (vmcb01 because it's always valid) for
3621 		 * context switching guest state via VMLOAD/VMSAVE, that way
3622 		 * the state doesn't need to be copied between vmcb01 and
3623 		 * vmcb02 when switching vmcbs for nested virtualization.
3624 		 */
3625 		vmload(svm->vmcb01.pa);
3626 		__svm_vcpu_run(vmcb_pa, (unsigned long *)&vcpu->arch.regs);
3627 		vmsave(svm->vmcb01.pa);
3628 
3629 		vmload(__sme_page_pa(sd->save_area));
3630 	}
3631 
3632 	kvm_guest_exit_irqoff();
3633 }
3634 
3635 static __no_kcsan fastpath_t svm_vcpu_run(struct kvm_vcpu *vcpu)
3636 {
3637 	struct vcpu_svm *svm = to_svm(vcpu);
3638 
3639 	trace_kvm_entry(vcpu);
3640 
3641 	svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
3642 	svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
3643 	svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
3644 
3645 	/*
3646 	 * Disable singlestep if we're injecting an interrupt/exception.
3647 	 * We don't want our modified rflags to be pushed on the stack where
3648 	 * we might not be able to easily reset them if we disabled NMI
3649 	 * singlestep later.
3650 	 */
3651 	if (svm->nmi_singlestep && svm->vmcb->control.event_inj) {
3652 		/*
3653 		 * Event injection happens before external interrupts cause a
3654 		 * vmexit and interrupts are disabled here, so smp_send_reschedule
3655 		 * is enough to force an immediate vmexit.
3656 		 */
3657 		disable_nmi_singlestep(svm);
3658 		smp_send_reschedule(vcpu->cpu);
3659 	}
3660 
3661 	pre_svm_run(vcpu);
3662 
3663 	sync_lapic_to_cr8(vcpu);
3664 
3665 	if (unlikely(svm->asid != svm->vmcb->control.asid)) {
3666 		svm->vmcb->control.asid = svm->asid;
3667 		vmcb_mark_dirty(svm->vmcb, VMCB_ASID);
3668 	}
3669 	svm->vmcb->save.cr2 = vcpu->arch.cr2;
3670 
3671 	svm_hv_update_vp_id(svm->vmcb, vcpu);
3672 
3673 	/*
3674 	 * Run with all-zero DR6 unless needed, so that we can get the exact cause
3675 	 * of a #DB.
3676 	 */
3677 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT))
3678 		svm_set_dr6(svm, vcpu->arch.dr6);
3679 	else
3680 		svm_set_dr6(svm, DR6_ACTIVE_LOW);
3681 
3682 	clgi();
3683 	kvm_load_guest_xsave_state(vcpu);
3684 
3685 	kvm_wait_lapic_expire(vcpu);
3686 
3687 	/*
3688 	 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
3689 	 * it's non-zero. Since vmentry is serialising on affected CPUs, there
3690 	 * is no need to worry about the conditional branch over the wrmsr
3691 	 * being speculatively taken.
3692 	 */
3693 	if (!static_cpu_has(X86_FEATURE_V_SPEC_CTRL))
3694 		x86_spec_ctrl_set_guest(svm->spec_ctrl, svm->virt_spec_ctrl);
3695 
3696 	svm_vcpu_enter_exit(vcpu);
3697 
3698 	/*
3699 	 * We do not use IBRS in the kernel. If this vCPU has used the
3700 	 * SPEC_CTRL MSR it may have left it on; save the value and
3701 	 * turn it off. This is much more efficient than blindly adding
3702 	 * it to the atomic save/restore list. Especially as the former
3703 	 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
3704 	 *
3705 	 * For non-nested case:
3706 	 * If the L01 MSR bitmap does not intercept the MSR, then we need to
3707 	 * save it.
3708 	 *
3709 	 * For nested case:
3710 	 * If the L02 MSR bitmap does not intercept the MSR, then we need to
3711 	 * save it.
3712 	 */
3713 	if (!static_cpu_has(X86_FEATURE_V_SPEC_CTRL) &&
3714 	    unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
3715 		svm->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
3716 
3717 	if (!sev_es_guest(vcpu->kvm))
3718 		reload_tss(vcpu);
3719 
3720 	if (!static_cpu_has(X86_FEATURE_V_SPEC_CTRL))
3721 		x86_spec_ctrl_restore_host(svm->spec_ctrl, svm->virt_spec_ctrl);
3722 
3723 	if (!sev_es_guest(vcpu->kvm)) {
3724 		vcpu->arch.cr2 = svm->vmcb->save.cr2;
3725 		vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
3726 		vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
3727 		vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
3728 	}
3729 	vcpu->arch.regs_dirty = 0;
3730 
3731 	if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
3732 		kvm_before_interrupt(vcpu, KVM_HANDLING_NMI);
3733 
3734 	kvm_load_host_xsave_state(vcpu);
3735 	stgi();
3736 
3737 	/* Any pending NMI will happen here */
3738 
3739 	if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
3740 		kvm_after_interrupt(vcpu);
3741 
3742 	sync_cr8_to_lapic(vcpu);
3743 
3744 	svm->next_rip = 0;
3745 	if (is_guest_mode(vcpu)) {
3746 		nested_sync_control_from_vmcb02(svm);
3747 
3748 		/* Track VMRUNs that have made past consistency checking */
3749 		if (svm->nested.nested_run_pending &&
3750 		    svm->vmcb->control.exit_code != SVM_EXIT_ERR)
3751                         ++vcpu->stat.nested_run;
3752 
3753 		svm->nested.nested_run_pending = 0;
3754 	}
3755 
3756 	svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
3757 	vmcb_mark_all_clean(svm->vmcb);
3758 
3759 	/* if exit due to PF check for async PF */
3760 	if (svm->vmcb->control.exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR)
3761 		vcpu->arch.apf.host_apf_flags =
3762 			kvm_read_and_reset_apf_flags();
3763 
3764 	vcpu->arch.regs_avail &= ~SVM_REGS_LAZY_LOAD_SET;
3765 
3766 	/*
3767 	 * We need to handle MC intercepts here before the vcpu has a chance to
3768 	 * change the physical cpu
3769 	 */
3770 	if (unlikely(svm->vmcb->control.exit_code ==
3771 		     SVM_EXIT_EXCP_BASE + MC_VECTOR))
3772 		svm_handle_mce(vcpu);
3773 
3774 	svm_complete_interrupts(vcpu);
3775 
3776 	if (is_guest_mode(vcpu))
3777 		return EXIT_FASTPATH_NONE;
3778 
3779 	return svm_exit_handlers_fastpath(vcpu);
3780 }
3781 
3782 static void svm_load_mmu_pgd(struct kvm_vcpu *vcpu, hpa_t root_hpa,
3783 			     int root_level)
3784 {
3785 	struct vcpu_svm *svm = to_svm(vcpu);
3786 	unsigned long cr3;
3787 
3788 	if (npt_enabled) {
3789 		svm->vmcb->control.nested_cr3 = __sme_set(root_hpa);
3790 		vmcb_mark_dirty(svm->vmcb, VMCB_NPT);
3791 
3792 		hv_track_root_tdp(vcpu, root_hpa);
3793 
3794 		cr3 = vcpu->arch.cr3;
3795 	} else if (vcpu->arch.mmu->shadow_root_level >= PT64_ROOT_4LEVEL) {
3796 		cr3 = __sme_set(root_hpa) | kvm_get_active_pcid(vcpu);
3797 	} else {
3798 		/* PCID in the guest should be impossible with a 32-bit MMU. */
3799 		WARN_ON_ONCE(kvm_get_active_pcid(vcpu));
3800 		cr3 = root_hpa;
3801 	}
3802 
3803 	svm->vmcb->save.cr3 = cr3;
3804 	vmcb_mark_dirty(svm->vmcb, VMCB_CR);
3805 }
3806 
3807 static int is_disabled(void)
3808 {
3809 	u64 vm_cr;
3810 
3811 	rdmsrl(MSR_VM_CR, vm_cr);
3812 	if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
3813 		return 1;
3814 
3815 	return 0;
3816 }
3817 
3818 static void
3819 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
3820 {
3821 	/*
3822 	 * Patch in the VMMCALL instruction:
3823 	 */
3824 	hypercall[0] = 0x0f;
3825 	hypercall[1] = 0x01;
3826 	hypercall[2] = 0xd9;
3827 }
3828 
3829 static int __init svm_check_processor_compat(void)
3830 {
3831 	return 0;
3832 }
3833 
3834 static bool svm_cpu_has_accelerated_tpr(void)
3835 {
3836 	return false;
3837 }
3838 
3839 /*
3840  * The kvm parameter can be NULL (module initialization, or invocation before
3841  * VM creation). Be sure to check the kvm parameter before using it.
3842  */
3843 static bool svm_has_emulated_msr(struct kvm *kvm, u32 index)
3844 {
3845 	switch (index) {
3846 	case MSR_IA32_MCG_EXT_CTL:
3847 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
3848 		return false;
3849 	case MSR_IA32_SMBASE:
3850 		/* SEV-ES guests do not support SMM, so report false */
3851 		if (kvm && sev_es_guest(kvm))
3852 			return false;
3853 		break;
3854 	default:
3855 		break;
3856 	}
3857 
3858 	return true;
3859 }
3860 
3861 static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
3862 {
3863 	return 0;
3864 }
3865 
3866 static void svm_vcpu_after_set_cpuid(struct kvm_vcpu *vcpu)
3867 {
3868 	struct vcpu_svm *svm = to_svm(vcpu);
3869 	struct kvm_cpuid_entry2 *best;
3870 
3871 	vcpu->arch.xsaves_enabled = guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
3872 				    boot_cpu_has(X86_FEATURE_XSAVE) &&
3873 				    boot_cpu_has(X86_FEATURE_XSAVES);
3874 
3875 	/* Update nrips enabled cache */
3876 	svm->nrips_enabled = kvm_cpu_cap_has(X86_FEATURE_NRIPS) &&
3877 			     guest_cpuid_has(vcpu, X86_FEATURE_NRIPS);
3878 
3879 	svm->tsc_scaling_enabled = tsc_scaling && guest_cpuid_has(vcpu, X86_FEATURE_TSCRATEMSR);
3880 
3881 	svm_recalc_instruction_intercepts(vcpu, svm);
3882 
3883 	/* For sev guests, the memory encryption bit is not reserved in CR3.  */
3884 	if (sev_guest(vcpu->kvm)) {
3885 		best = kvm_find_cpuid_entry(vcpu, 0x8000001F, 0);
3886 		if (best)
3887 			vcpu->arch.reserved_gpa_bits &= ~(1UL << (best->ebx & 0x3f));
3888 	}
3889 
3890 	if (kvm_vcpu_apicv_active(vcpu)) {
3891 		/*
3892 		 * AVIC does not work with an x2APIC mode guest. If the X2APIC feature
3893 		 * is exposed to the guest, disable AVIC.
3894 		 */
3895 		if (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC))
3896 			kvm_request_apicv_update(vcpu->kvm, false,
3897 						 APICV_INHIBIT_REASON_X2APIC);
3898 
3899 		/*
3900 		 * Currently, AVIC does not work with nested virtualization.
3901 		 * So, we disable AVIC when cpuid for SVM is set in the L1 guest.
3902 		 */
3903 		if (nested && guest_cpuid_has(vcpu, X86_FEATURE_SVM))
3904 			kvm_request_apicv_update(vcpu->kvm, false,
3905 						 APICV_INHIBIT_REASON_NESTED);
3906 	}
3907 	init_vmcb_after_set_cpuid(vcpu);
3908 }
3909 
3910 static bool svm_has_wbinvd_exit(void)
3911 {
3912 	return true;
3913 }
3914 
3915 #define PRE_EX(exit)  { .exit_code = (exit), \
3916 			.stage = X86_ICPT_PRE_EXCEPT, }
3917 #define POST_EX(exit) { .exit_code = (exit), \
3918 			.stage = X86_ICPT_POST_EXCEPT, }
3919 #define POST_MEM(exit) { .exit_code = (exit), \
3920 			.stage = X86_ICPT_POST_MEMACCESS, }
3921 
3922 static const struct __x86_intercept {
3923 	u32 exit_code;
3924 	enum x86_intercept_stage stage;
3925 } x86_intercept_map[] = {
3926 	[x86_intercept_cr_read]		= POST_EX(SVM_EXIT_READ_CR0),
3927 	[x86_intercept_cr_write]	= POST_EX(SVM_EXIT_WRITE_CR0),
3928 	[x86_intercept_clts]		= POST_EX(SVM_EXIT_WRITE_CR0),
3929 	[x86_intercept_lmsw]		= POST_EX(SVM_EXIT_WRITE_CR0),
3930 	[x86_intercept_smsw]		= POST_EX(SVM_EXIT_READ_CR0),
3931 	[x86_intercept_dr_read]		= POST_EX(SVM_EXIT_READ_DR0),
3932 	[x86_intercept_dr_write]	= POST_EX(SVM_EXIT_WRITE_DR0),
3933 	[x86_intercept_sldt]		= POST_EX(SVM_EXIT_LDTR_READ),
3934 	[x86_intercept_str]		= POST_EX(SVM_EXIT_TR_READ),
3935 	[x86_intercept_lldt]		= POST_EX(SVM_EXIT_LDTR_WRITE),
3936 	[x86_intercept_ltr]		= POST_EX(SVM_EXIT_TR_WRITE),
3937 	[x86_intercept_sgdt]		= POST_EX(SVM_EXIT_GDTR_READ),
3938 	[x86_intercept_sidt]		= POST_EX(SVM_EXIT_IDTR_READ),
3939 	[x86_intercept_lgdt]		= POST_EX(SVM_EXIT_GDTR_WRITE),
3940 	[x86_intercept_lidt]		= POST_EX(SVM_EXIT_IDTR_WRITE),
3941 	[x86_intercept_vmrun]		= POST_EX(SVM_EXIT_VMRUN),
3942 	[x86_intercept_vmmcall]		= POST_EX(SVM_EXIT_VMMCALL),
3943 	[x86_intercept_vmload]		= POST_EX(SVM_EXIT_VMLOAD),
3944 	[x86_intercept_vmsave]		= POST_EX(SVM_EXIT_VMSAVE),
3945 	[x86_intercept_stgi]		= POST_EX(SVM_EXIT_STGI),
3946 	[x86_intercept_clgi]		= POST_EX(SVM_EXIT_CLGI),
3947 	[x86_intercept_skinit]		= POST_EX(SVM_EXIT_SKINIT),
3948 	[x86_intercept_invlpga]		= POST_EX(SVM_EXIT_INVLPGA),
3949 	[x86_intercept_rdtscp]		= POST_EX(SVM_EXIT_RDTSCP),
3950 	[x86_intercept_monitor]		= POST_MEM(SVM_EXIT_MONITOR),
3951 	[x86_intercept_mwait]		= POST_EX(SVM_EXIT_MWAIT),
3952 	[x86_intercept_invlpg]		= POST_EX(SVM_EXIT_INVLPG),
3953 	[x86_intercept_invd]		= POST_EX(SVM_EXIT_INVD),
3954 	[x86_intercept_wbinvd]		= POST_EX(SVM_EXIT_WBINVD),
3955 	[x86_intercept_wrmsr]		= POST_EX(SVM_EXIT_MSR),
3956 	[x86_intercept_rdtsc]		= POST_EX(SVM_EXIT_RDTSC),
3957 	[x86_intercept_rdmsr]		= POST_EX(SVM_EXIT_MSR),
3958 	[x86_intercept_rdpmc]		= POST_EX(SVM_EXIT_RDPMC),
3959 	[x86_intercept_cpuid]		= PRE_EX(SVM_EXIT_CPUID),
3960 	[x86_intercept_rsm]		= PRE_EX(SVM_EXIT_RSM),
3961 	[x86_intercept_pause]		= PRE_EX(SVM_EXIT_PAUSE),
3962 	[x86_intercept_pushf]		= PRE_EX(SVM_EXIT_PUSHF),
3963 	[x86_intercept_popf]		= PRE_EX(SVM_EXIT_POPF),
3964 	[x86_intercept_intn]		= PRE_EX(SVM_EXIT_SWINT),
3965 	[x86_intercept_iret]		= PRE_EX(SVM_EXIT_IRET),
3966 	[x86_intercept_icebp]		= PRE_EX(SVM_EXIT_ICEBP),
3967 	[x86_intercept_hlt]		= POST_EX(SVM_EXIT_HLT),
3968 	[x86_intercept_in]		= POST_EX(SVM_EXIT_IOIO),
3969 	[x86_intercept_ins]		= POST_EX(SVM_EXIT_IOIO),
3970 	[x86_intercept_out]		= POST_EX(SVM_EXIT_IOIO),
3971 	[x86_intercept_outs]		= POST_EX(SVM_EXIT_IOIO),
3972 	[x86_intercept_xsetbv]		= PRE_EX(SVM_EXIT_XSETBV),
3973 };
3974 
3975 #undef PRE_EX
3976 #undef POST_EX
3977 #undef POST_MEM
3978 
3979 static int svm_check_intercept(struct kvm_vcpu *vcpu,
3980 			       struct x86_instruction_info *info,
3981 			       enum x86_intercept_stage stage,
3982 			       struct x86_exception *exception)
3983 {
3984 	struct vcpu_svm *svm = to_svm(vcpu);
3985 	int vmexit, ret = X86EMUL_CONTINUE;
3986 	struct __x86_intercept icpt_info;
3987 	struct vmcb *vmcb = svm->vmcb;
3988 
3989 	if (info->intercept >= ARRAY_SIZE(x86_intercept_map))
3990 		goto out;
3991 
3992 	icpt_info = x86_intercept_map[info->intercept];
3993 
3994 	if (stage != icpt_info.stage)
3995 		goto out;
3996 
3997 	switch (icpt_info.exit_code) {
3998 	case SVM_EXIT_READ_CR0:
3999 		if (info->intercept == x86_intercept_cr_read)
4000 			icpt_info.exit_code += info->modrm_reg;
4001 		break;
4002 	case SVM_EXIT_WRITE_CR0: {
4003 		unsigned long cr0, val;
4004 
4005 		if (info->intercept == x86_intercept_cr_write)
4006 			icpt_info.exit_code += info->modrm_reg;
4007 
4008 		if (icpt_info.exit_code != SVM_EXIT_WRITE_CR0 ||
4009 		    info->intercept == x86_intercept_clts)
4010 			break;
4011 
4012 		if (!(vmcb12_is_intercept(&svm->nested.ctl,
4013 					INTERCEPT_SELECTIVE_CR0)))
4014 			break;
4015 
4016 		cr0 = vcpu->arch.cr0 & ~SVM_CR0_SELECTIVE_MASK;
4017 		val = info->src_val  & ~SVM_CR0_SELECTIVE_MASK;
4018 
4019 		if (info->intercept == x86_intercept_lmsw) {
4020 			cr0 &= 0xfUL;
4021 			val &= 0xfUL;
4022 			/* lmsw can't clear PE - catch this here */
4023 			if (cr0 & X86_CR0_PE)
4024 				val |= X86_CR0_PE;
4025 		}
4026 
4027 		if (cr0 ^ val)
4028 			icpt_info.exit_code = SVM_EXIT_CR0_SEL_WRITE;
4029 
4030 		break;
4031 	}
4032 	case SVM_EXIT_READ_DR0:
4033 	case SVM_EXIT_WRITE_DR0:
4034 		icpt_info.exit_code += info->modrm_reg;
4035 		break;
4036 	case SVM_EXIT_MSR:
4037 		if (info->intercept == x86_intercept_wrmsr)
4038 			vmcb->control.exit_info_1 = 1;
4039 		else
4040 			vmcb->control.exit_info_1 = 0;
4041 		break;
4042 	case SVM_EXIT_PAUSE:
4043 		/*
4044 		 * We get this for NOP only, but pause
4045 		 * is rep not, check this here
4046 		 */
4047 		if (info->rep_prefix != REPE_PREFIX)
4048 			goto out;
4049 		break;
4050 	case SVM_EXIT_IOIO: {
4051 		u64 exit_info;
4052 		u32 bytes;
4053 
4054 		if (info->intercept == x86_intercept_in ||
4055 		    info->intercept == x86_intercept_ins) {
4056 			exit_info = ((info->src_val & 0xffff) << 16) |
4057 				SVM_IOIO_TYPE_MASK;
4058 			bytes = info->dst_bytes;
4059 		} else {
4060 			exit_info = (info->dst_val & 0xffff) << 16;
4061 			bytes = info->src_bytes;
4062 		}
4063 
4064 		if (info->intercept == x86_intercept_outs ||
4065 		    info->intercept == x86_intercept_ins)
4066 			exit_info |= SVM_IOIO_STR_MASK;
4067 
4068 		if (info->rep_prefix)
4069 			exit_info |= SVM_IOIO_REP_MASK;
4070 
4071 		bytes = min(bytes, 4u);
4072 
4073 		exit_info |= bytes << SVM_IOIO_SIZE_SHIFT;
4074 
4075 		exit_info |= (u32)info->ad_bytes << (SVM_IOIO_ASIZE_SHIFT - 1);
4076 
4077 		vmcb->control.exit_info_1 = exit_info;
4078 		vmcb->control.exit_info_2 = info->next_rip;
4079 
4080 		break;
4081 	}
4082 	default:
4083 		break;
4084 	}
4085 
4086 	/* TODO: Advertise NRIPS to guest hypervisor unconditionally */
4087 	if (static_cpu_has(X86_FEATURE_NRIPS))
4088 		vmcb->control.next_rip  = info->next_rip;
4089 	vmcb->control.exit_code = icpt_info.exit_code;
4090 	vmexit = nested_svm_exit_handled(svm);
4091 
4092 	ret = (vmexit == NESTED_EXIT_DONE) ? X86EMUL_INTERCEPTED
4093 					   : X86EMUL_CONTINUE;
4094 
4095 out:
4096 	return ret;
4097 }
4098 
4099 static void svm_handle_exit_irqoff(struct kvm_vcpu *vcpu)
4100 {
4101 }
4102 
4103 static void svm_sched_in(struct kvm_vcpu *vcpu, int cpu)
4104 {
4105 	if (!kvm_pause_in_guest(vcpu->kvm))
4106 		shrink_ple_window(vcpu);
4107 }
4108 
4109 static void svm_setup_mce(struct kvm_vcpu *vcpu)
4110 {
4111 	/* [63:9] are reserved. */
4112 	vcpu->arch.mcg_cap &= 0x1ff;
4113 }
4114 
4115 bool svm_smi_blocked(struct kvm_vcpu *vcpu)
4116 {
4117 	struct vcpu_svm *svm = to_svm(vcpu);
4118 
4119 	/* Per APM Vol.2 15.22.2 "Response to SMI" */
4120 	if (!gif_set(svm))
4121 		return true;
4122 
4123 	return is_smm(vcpu);
4124 }
4125 
4126 static int svm_smi_allowed(struct kvm_vcpu *vcpu, bool for_injection)
4127 {
4128 	struct vcpu_svm *svm = to_svm(vcpu);
4129 	if (svm->nested.nested_run_pending)
4130 		return -EBUSY;
4131 
4132 	/* An SMI must not be injected into L2 if it's supposed to VM-Exit.  */
4133 	if (for_injection && is_guest_mode(vcpu) && nested_exit_on_smi(svm))
4134 		return -EBUSY;
4135 
4136 	return !svm_smi_blocked(vcpu);
4137 }
4138 
4139 static int svm_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
4140 {
4141 	struct vcpu_svm *svm = to_svm(vcpu);
4142 	struct kvm_host_map map_save;
4143 	int ret;
4144 
4145 	if (!is_guest_mode(vcpu))
4146 		return 0;
4147 
4148 	/* FED8h - SVM Guest */
4149 	put_smstate(u64, smstate, 0x7ed8, 1);
4150 	/* FEE0h - SVM Guest VMCB Physical Address */
4151 	put_smstate(u64, smstate, 0x7ee0, svm->nested.vmcb12_gpa);
4152 
4153 	svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
4154 	svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
4155 	svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
4156 
4157 	ret = nested_svm_vmexit(svm);
4158 	if (ret)
4159 		return ret;
4160 
4161 	/*
4162 	 * KVM uses VMCB01 to store L1 host state while L2 runs but
4163 	 * VMCB01 is going to be used during SMM and thus the state will
4164 	 * be lost. Temporary save non-VMLOAD/VMSAVE state to the host save
4165 	 * area pointed to by MSR_VM_HSAVE_PA. APM guarantees that the
4166 	 * format of the area is identical to guest save area offsetted
4167 	 * by 0x400 (matches the offset of 'struct vmcb_save_area'
4168 	 * within 'struct vmcb'). Note: HSAVE area may also be used by
4169 	 * L1 hypervisor to save additional host context (e.g. KVM does
4170 	 * that, see svm_prepare_guest_switch()) which must be
4171 	 * preserved.
4172 	 */
4173 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(svm->nested.hsave_msr),
4174 			 &map_save) == -EINVAL)
4175 		return 1;
4176 
4177 	BUILD_BUG_ON(offsetof(struct vmcb, save) != 0x400);
4178 
4179 	svm_copy_vmrun_state(map_save.hva + 0x400,
4180 			     &svm->vmcb01.ptr->save);
4181 
4182 	kvm_vcpu_unmap(vcpu, &map_save, true);
4183 	return 0;
4184 }
4185 
4186 static int svm_leave_smm(struct kvm_vcpu *vcpu, const char *smstate)
4187 {
4188 	struct vcpu_svm *svm = to_svm(vcpu);
4189 	struct kvm_host_map map, map_save;
4190 	u64 saved_efer, vmcb12_gpa;
4191 	struct vmcb *vmcb12;
4192 	int ret;
4193 
4194 	if (!guest_cpuid_has(vcpu, X86_FEATURE_LM))
4195 		return 0;
4196 
4197 	/* Non-zero if SMI arrived while vCPU was in guest mode. */
4198 	if (!GET_SMSTATE(u64, smstate, 0x7ed8))
4199 		return 0;
4200 
4201 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SVM))
4202 		return 1;
4203 
4204 	saved_efer = GET_SMSTATE(u64, smstate, 0x7ed0);
4205 	if (!(saved_efer & EFER_SVME))
4206 		return 1;
4207 
4208 	vmcb12_gpa = GET_SMSTATE(u64, smstate, 0x7ee0);
4209 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmcb12_gpa), &map) == -EINVAL)
4210 		return 1;
4211 
4212 	ret = 1;
4213 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(svm->nested.hsave_msr), &map_save) == -EINVAL)
4214 		goto unmap_map;
4215 
4216 	if (svm_allocate_nested(svm))
4217 		goto unmap_save;
4218 
4219 	/*
4220 	 * Restore L1 host state from L1 HSAVE area as VMCB01 was
4221 	 * used during SMM (see svm_enter_smm())
4222 	 */
4223 
4224 	svm_copy_vmrun_state(&svm->vmcb01.ptr->save, map_save.hva + 0x400);
4225 
4226 	/*
4227 	 * Enter the nested guest now
4228 	 */
4229 
4230 	vmcb12 = map.hva;
4231 	nested_copy_vmcb_control_to_cache(svm, &vmcb12->control);
4232 	nested_copy_vmcb_save_to_cache(svm, &vmcb12->save);
4233 	ret = enter_svm_guest_mode(vcpu, vmcb12_gpa, vmcb12, false);
4234 
4235 unmap_save:
4236 	kvm_vcpu_unmap(vcpu, &map_save, true);
4237 unmap_map:
4238 	kvm_vcpu_unmap(vcpu, &map, true);
4239 	return ret;
4240 }
4241 
4242 static void svm_enable_smi_window(struct kvm_vcpu *vcpu)
4243 {
4244 	struct vcpu_svm *svm = to_svm(vcpu);
4245 
4246 	if (!gif_set(svm)) {
4247 		if (vgif_enabled(svm))
4248 			svm_set_intercept(svm, INTERCEPT_STGI);
4249 		/* STGI will cause a vm exit */
4250 	} else {
4251 		/* We must be in SMM; RSM will cause a vmexit anyway.  */
4252 	}
4253 }
4254 
4255 static bool svm_can_emulate_instruction(struct kvm_vcpu *vcpu, void *insn, int insn_len)
4256 {
4257 	bool smep, smap, is_user;
4258 	unsigned long cr4;
4259 
4260 	/*
4261 	 * When the guest is an SEV-ES guest, emulation is not possible.
4262 	 */
4263 	if (sev_es_guest(vcpu->kvm))
4264 		return false;
4265 
4266 	/*
4267 	 * Detect and workaround Errata 1096 Fam_17h_00_0Fh.
4268 	 *
4269 	 * Errata:
4270 	 * When CPU raise #NPF on guest data access and vCPU CR4.SMAP=1, it is
4271 	 * possible that CPU microcode implementing DecodeAssist will fail
4272 	 * to read bytes of instruction which caused #NPF. In this case,
4273 	 * GuestIntrBytes field of the VMCB on a VMEXIT will incorrectly
4274 	 * return 0 instead of the correct guest instruction bytes.
4275 	 *
4276 	 * This happens because CPU microcode reading instruction bytes
4277 	 * uses a special opcode which attempts to read data using CPL=0
4278 	 * privileges. The microcode reads CS:RIP and if it hits a SMAP
4279 	 * fault, it gives up and returns no instruction bytes.
4280 	 *
4281 	 * Detection:
4282 	 * We reach here in case CPU supports DecodeAssist, raised #NPF and
4283 	 * returned 0 in GuestIntrBytes field of the VMCB.
4284 	 * First, errata can only be triggered in case vCPU CR4.SMAP=1.
4285 	 * Second, if vCPU CR4.SMEP=1, errata could only be triggered
4286 	 * in case vCPU CPL==3 (Because otherwise guest would have triggered
4287 	 * a SMEP fault instead of #NPF).
4288 	 * Otherwise, vCPU CR4.SMEP=0, errata could be triggered by any vCPU CPL.
4289 	 * As most guests enable SMAP if they have also enabled SMEP, use above
4290 	 * logic in order to attempt minimize false-positive of detecting errata
4291 	 * while still preserving all cases semantic correctness.
4292 	 *
4293 	 * Workaround:
4294 	 * To determine what instruction the guest was executing, the hypervisor
4295 	 * will have to decode the instruction at the instruction pointer.
4296 	 *
4297 	 * In non SEV guest, hypervisor will be able to read the guest
4298 	 * memory to decode the instruction pointer when insn_len is zero
4299 	 * so we return true to indicate that decoding is possible.
4300 	 *
4301 	 * But in the SEV guest, the guest memory is encrypted with the
4302 	 * guest specific key and hypervisor will not be able to decode the
4303 	 * instruction pointer so we will not able to workaround it. Lets
4304 	 * print the error and request to kill the guest.
4305 	 */
4306 	if (likely(!insn || insn_len))
4307 		return true;
4308 
4309 	/*
4310 	 * If RIP is invalid, go ahead with emulation which will cause an
4311 	 * internal error exit.
4312 	 */
4313 	if (!kvm_vcpu_gfn_to_memslot(vcpu, kvm_rip_read(vcpu) >> PAGE_SHIFT))
4314 		return true;
4315 
4316 	cr4 = kvm_read_cr4(vcpu);
4317 	smep = cr4 & X86_CR4_SMEP;
4318 	smap = cr4 & X86_CR4_SMAP;
4319 	is_user = svm_get_cpl(vcpu) == 3;
4320 	if (smap && (!smep || is_user)) {
4321 		if (!sev_guest(vcpu->kvm))
4322 			return true;
4323 
4324 		pr_err_ratelimited("KVM: SEV Guest triggered AMD Erratum 1096\n");
4325 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4326 	}
4327 
4328 	return false;
4329 }
4330 
4331 static bool svm_apic_init_signal_blocked(struct kvm_vcpu *vcpu)
4332 {
4333 	struct vcpu_svm *svm = to_svm(vcpu);
4334 
4335 	/*
4336 	 * TODO: Last condition latch INIT signals on vCPU when
4337 	 * vCPU is in guest-mode and vmcb12 defines intercept on INIT.
4338 	 * To properly emulate the INIT intercept,
4339 	 * svm_check_nested_events() should call nested_svm_vmexit()
4340 	 * if an INIT signal is pending.
4341 	 */
4342 	return !gif_set(svm) ||
4343 		   (vmcb_is_intercept(&svm->vmcb->control, INTERCEPT_INIT));
4344 }
4345 
4346 static void svm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
4347 {
4348 	if (!sev_es_guest(vcpu->kvm))
4349 		return kvm_vcpu_deliver_sipi_vector(vcpu, vector);
4350 
4351 	sev_vcpu_deliver_sipi_vector(vcpu, vector);
4352 }
4353 
4354 static void svm_vm_destroy(struct kvm *kvm)
4355 {
4356 	avic_vm_destroy(kvm);
4357 	sev_vm_destroy(kvm);
4358 }
4359 
4360 static int svm_vm_init(struct kvm *kvm)
4361 {
4362 	if (!pause_filter_count || !pause_filter_thresh)
4363 		kvm->arch.pause_in_guest = true;
4364 
4365 	if (enable_apicv) {
4366 		int ret = avic_vm_init(kvm);
4367 		if (ret)
4368 			return ret;
4369 	}
4370 
4371 	return 0;
4372 }
4373 
4374 static struct kvm_x86_ops svm_x86_ops __initdata = {
4375 	.name = "kvm_amd",
4376 
4377 	.hardware_unsetup = svm_hardware_teardown,
4378 	.hardware_enable = svm_hardware_enable,
4379 	.hardware_disable = svm_hardware_disable,
4380 	.cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
4381 	.has_emulated_msr = svm_has_emulated_msr,
4382 
4383 	.vcpu_create = svm_create_vcpu,
4384 	.vcpu_free = svm_free_vcpu,
4385 	.vcpu_reset = svm_vcpu_reset,
4386 
4387 	.vm_size = sizeof(struct kvm_svm),
4388 	.vm_init = svm_vm_init,
4389 	.vm_destroy = svm_vm_destroy,
4390 
4391 	.prepare_guest_switch = svm_prepare_guest_switch,
4392 	.vcpu_load = svm_vcpu_load,
4393 	.vcpu_put = svm_vcpu_put,
4394 	.vcpu_blocking = avic_vcpu_blocking,
4395 	.vcpu_unblocking = avic_vcpu_unblocking,
4396 
4397 	.update_exception_bitmap = svm_update_exception_bitmap,
4398 	.get_msr_feature = svm_get_msr_feature,
4399 	.get_msr = svm_get_msr,
4400 	.set_msr = svm_set_msr,
4401 	.get_segment_base = svm_get_segment_base,
4402 	.get_segment = svm_get_segment,
4403 	.set_segment = svm_set_segment,
4404 	.get_cpl = svm_get_cpl,
4405 	.get_cs_db_l_bits = kvm_get_cs_db_l_bits,
4406 	.set_cr0 = svm_set_cr0,
4407 	.post_set_cr3 = svm_post_set_cr3,
4408 	.is_valid_cr4 = svm_is_valid_cr4,
4409 	.set_cr4 = svm_set_cr4,
4410 	.set_efer = svm_set_efer,
4411 	.get_idt = svm_get_idt,
4412 	.set_idt = svm_set_idt,
4413 	.get_gdt = svm_get_gdt,
4414 	.set_gdt = svm_set_gdt,
4415 	.set_dr7 = svm_set_dr7,
4416 	.sync_dirty_debug_regs = svm_sync_dirty_debug_regs,
4417 	.cache_reg = svm_cache_reg,
4418 	.get_rflags = svm_get_rflags,
4419 	.set_rflags = svm_set_rflags,
4420 	.get_if_flag = svm_get_if_flag,
4421 
4422 	.tlb_flush_all = svm_flush_tlb,
4423 	.tlb_flush_current = svm_flush_tlb,
4424 	.tlb_flush_gva = svm_flush_tlb_gva,
4425 	.tlb_flush_guest = svm_flush_tlb,
4426 
4427 	.vcpu_pre_run = svm_vcpu_pre_run,
4428 	.run = svm_vcpu_run,
4429 	.handle_exit = handle_exit,
4430 	.skip_emulated_instruction = skip_emulated_instruction,
4431 	.update_emulated_instruction = NULL,
4432 	.set_interrupt_shadow = svm_set_interrupt_shadow,
4433 	.get_interrupt_shadow = svm_get_interrupt_shadow,
4434 	.patch_hypercall = svm_patch_hypercall,
4435 	.set_irq = svm_set_irq,
4436 	.set_nmi = svm_inject_nmi,
4437 	.queue_exception = svm_queue_exception,
4438 	.cancel_injection = svm_cancel_injection,
4439 	.interrupt_allowed = svm_interrupt_allowed,
4440 	.nmi_allowed = svm_nmi_allowed,
4441 	.get_nmi_mask = svm_get_nmi_mask,
4442 	.set_nmi_mask = svm_set_nmi_mask,
4443 	.enable_nmi_window = svm_enable_nmi_window,
4444 	.enable_irq_window = svm_enable_irq_window,
4445 	.update_cr8_intercept = svm_update_cr8_intercept,
4446 	.set_virtual_apic_mode = svm_set_virtual_apic_mode,
4447 	.refresh_apicv_exec_ctrl = svm_refresh_apicv_exec_ctrl,
4448 	.check_apicv_inhibit_reasons = svm_check_apicv_inhibit_reasons,
4449 	.load_eoi_exitmap = svm_load_eoi_exitmap,
4450 	.hwapic_irr_update = svm_hwapic_irr_update,
4451 	.hwapic_isr_update = svm_hwapic_isr_update,
4452 	.apicv_post_state_restore = avic_post_state_restore,
4453 
4454 	.set_tss_addr = svm_set_tss_addr,
4455 	.set_identity_map_addr = svm_set_identity_map_addr,
4456 	.get_mt_mask = svm_get_mt_mask,
4457 
4458 	.get_exit_info = svm_get_exit_info,
4459 
4460 	.vcpu_after_set_cpuid = svm_vcpu_after_set_cpuid,
4461 
4462 	.has_wbinvd_exit = svm_has_wbinvd_exit,
4463 
4464 	.get_l2_tsc_offset = svm_get_l2_tsc_offset,
4465 	.get_l2_tsc_multiplier = svm_get_l2_tsc_multiplier,
4466 	.write_tsc_offset = svm_write_tsc_offset,
4467 	.write_tsc_multiplier = svm_write_tsc_multiplier,
4468 
4469 	.load_mmu_pgd = svm_load_mmu_pgd,
4470 
4471 	.check_intercept = svm_check_intercept,
4472 	.handle_exit_irqoff = svm_handle_exit_irqoff,
4473 
4474 	.request_immediate_exit = __kvm_request_immediate_exit,
4475 
4476 	.sched_in = svm_sched_in,
4477 
4478 	.pmu_ops = &amd_pmu_ops,
4479 	.nested_ops = &svm_nested_ops,
4480 
4481 	.deliver_posted_interrupt = svm_deliver_avic_intr,
4482 	.dy_apicv_has_pending_interrupt = svm_dy_apicv_has_pending_interrupt,
4483 	.update_pi_irte = svm_update_pi_irte,
4484 	.setup_mce = svm_setup_mce,
4485 
4486 	.smi_allowed = svm_smi_allowed,
4487 	.enter_smm = svm_enter_smm,
4488 	.leave_smm = svm_leave_smm,
4489 	.enable_smi_window = svm_enable_smi_window,
4490 
4491 	.mem_enc_op = svm_mem_enc_op,
4492 	.mem_enc_reg_region = svm_register_enc_region,
4493 	.mem_enc_unreg_region = svm_unregister_enc_region,
4494 
4495 	.vm_copy_enc_context_from = svm_vm_copy_asid_from,
4496 	.vm_move_enc_context_from = svm_vm_migrate_from,
4497 
4498 	.can_emulate_instruction = svm_can_emulate_instruction,
4499 
4500 	.apic_init_signal_blocked = svm_apic_init_signal_blocked,
4501 
4502 	.msr_filter_changed = svm_msr_filter_changed,
4503 	.complete_emulated_msr = svm_complete_emulated_msr,
4504 
4505 	.vcpu_deliver_sipi_vector = svm_vcpu_deliver_sipi_vector,
4506 };
4507 
4508 /*
4509  * The default MMIO mask is a single bit (excluding the present bit),
4510  * which could conflict with the memory encryption bit. Check for
4511  * memory encryption support and override the default MMIO mask if
4512  * memory encryption is enabled.
4513  */
4514 static __init void svm_adjust_mmio_mask(void)
4515 {
4516 	unsigned int enc_bit, mask_bit;
4517 	u64 msr, mask;
4518 
4519 	/* If there is no memory encryption support, use existing mask */
4520 	if (cpuid_eax(0x80000000) < 0x8000001f)
4521 		return;
4522 
4523 	/* If memory encryption is not enabled, use existing mask */
4524 	rdmsrl(MSR_AMD64_SYSCFG, msr);
4525 	if (!(msr & MSR_AMD64_SYSCFG_MEM_ENCRYPT))
4526 		return;
4527 
4528 	enc_bit = cpuid_ebx(0x8000001f) & 0x3f;
4529 	mask_bit = boot_cpu_data.x86_phys_bits;
4530 
4531 	/* Increment the mask bit if it is the same as the encryption bit */
4532 	if (enc_bit == mask_bit)
4533 		mask_bit++;
4534 
4535 	/*
4536 	 * If the mask bit location is below 52, then some bits above the
4537 	 * physical addressing limit will always be reserved, so use the
4538 	 * rsvd_bits() function to generate the mask. This mask, along with
4539 	 * the present bit, will be used to generate a page fault with
4540 	 * PFER.RSV = 1.
4541 	 *
4542 	 * If the mask bit location is 52 (or above), then clear the mask.
4543 	 */
4544 	mask = (mask_bit < 52) ? rsvd_bits(mask_bit, 51) | PT_PRESENT_MASK : 0;
4545 
4546 	kvm_mmu_set_mmio_spte_mask(mask, mask, PT_WRITABLE_MASK | PT_USER_MASK);
4547 }
4548 
4549 static __init void svm_set_cpu_caps(void)
4550 {
4551 	kvm_set_cpu_caps();
4552 
4553 	supported_xss = 0;
4554 
4555 	/* CPUID 0x80000001 and 0x8000000A (SVM features) */
4556 	if (nested) {
4557 		kvm_cpu_cap_set(X86_FEATURE_SVM);
4558 
4559 		if (nrips)
4560 			kvm_cpu_cap_set(X86_FEATURE_NRIPS);
4561 
4562 		if (npt_enabled)
4563 			kvm_cpu_cap_set(X86_FEATURE_NPT);
4564 
4565 		if (tsc_scaling)
4566 			kvm_cpu_cap_set(X86_FEATURE_TSCRATEMSR);
4567 
4568 		/* Nested VM can receive #VMEXIT instead of triggering #GP */
4569 		kvm_cpu_cap_set(X86_FEATURE_SVME_ADDR_CHK);
4570 	}
4571 
4572 	/* CPUID 0x80000008 */
4573 	if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD) ||
4574 	    boot_cpu_has(X86_FEATURE_AMD_SSBD))
4575 		kvm_cpu_cap_set(X86_FEATURE_VIRT_SSBD);
4576 
4577 	/* AMD PMU PERFCTR_CORE CPUID */
4578 	if (enable_pmu && boot_cpu_has(X86_FEATURE_PERFCTR_CORE))
4579 		kvm_cpu_cap_set(X86_FEATURE_PERFCTR_CORE);
4580 
4581 	/* CPUID 0x8000001F (SME/SEV features) */
4582 	sev_set_cpu_caps();
4583 }
4584 
4585 static __init int svm_hardware_setup(void)
4586 {
4587 	int cpu;
4588 	struct page *iopm_pages;
4589 	void *iopm_va;
4590 	int r;
4591 	unsigned int order = get_order(IOPM_SIZE);
4592 
4593 	/*
4594 	 * NX is required for shadow paging and for NPT if the NX huge pages
4595 	 * mitigation is enabled.
4596 	 */
4597 	if (!boot_cpu_has(X86_FEATURE_NX)) {
4598 		pr_err_ratelimited("NX (Execute Disable) not supported\n");
4599 		return -EOPNOTSUPP;
4600 	}
4601 	kvm_enable_efer_bits(EFER_NX);
4602 
4603 	iopm_pages = alloc_pages(GFP_KERNEL, order);
4604 
4605 	if (!iopm_pages)
4606 		return -ENOMEM;
4607 
4608 	iopm_va = page_address(iopm_pages);
4609 	memset(iopm_va, 0xff, PAGE_SIZE * (1 << order));
4610 	iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
4611 
4612 	init_msrpm_offsets();
4613 
4614 	supported_xcr0 &= ~(XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR);
4615 
4616 	if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
4617 		kvm_enable_efer_bits(EFER_FFXSR);
4618 
4619 	if (tsc_scaling) {
4620 		if (!boot_cpu_has(X86_FEATURE_TSCRATEMSR)) {
4621 			tsc_scaling = false;
4622 		} else {
4623 			pr_info("TSC scaling supported\n");
4624 			kvm_has_tsc_control = true;
4625 			kvm_max_tsc_scaling_ratio = TSC_RATIO_MAX;
4626 			kvm_tsc_scaling_ratio_frac_bits = 32;
4627 		}
4628 	}
4629 
4630 	tsc_aux_uret_slot = kvm_add_user_return_msr(MSR_TSC_AUX);
4631 
4632 	/* Check for pause filtering support */
4633 	if (!boot_cpu_has(X86_FEATURE_PAUSEFILTER)) {
4634 		pause_filter_count = 0;
4635 		pause_filter_thresh = 0;
4636 	} else if (!boot_cpu_has(X86_FEATURE_PFTHRESHOLD)) {
4637 		pause_filter_thresh = 0;
4638 	}
4639 
4640 	if (nested) {
4641 		printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
4642 		kvm_enable_efer_bits(EFER_SVME | EFER_LMSLE);
4643 	}
4644 
4645 	/*
4646 	 * KVM's MMU doesn't support using 2-level paging for itself, and thus
4647 	 * NPT isn't supported if the host is using 2-level paging since host
4648 	 * CR4 is unchanged on VMRUN.
4649 	 */
4650 	if (!IS_ENABLED(CONFIG_X86_64) && !IS_ENABLED(CONFIG_X86_PAE))
4651 		npt_enabled = false;
4652 
4653 	if (!boot_cpu_has(X86_FEATURE_NPT))
4654 		npt_enabled = false;
4655 
4656 	/* Force VM NPT level equal to the host's paging level */
4657 	kvm_configure_mmu(npt_enabled, get_npt_level(),
4658 			  get_npt_level(), PG_LEVEL_1G);
4659 	pr_info("kvm: Nested Paging %sabled\n", npt_enabled ? "en" : "dis");
4660 
4661 	/* Note, SEV setup consumes npt_enabled. */
4662 	sev_hardware_setup();
4663 
4664 	svm_hv_hardware_setup();
4665 
4666 	svm_adjust_mmio_mask();
4667 
4668 	for_each_possible_cpu(cpu) {
4669 		r = svm_cpu_init(cpu);
4670 		if (r)
4671 			goto err;
4672 	}
4673 
4674 	if (nrips) {
4675 		if (!boot_cpu_has(X86_FEATURE_NRIPS))
4676 			nrips = false;
4677 	}
4678 
4679 	enable_apicv = avic = avic && npt_enabled && boot_cpu_has(X86_FEATURE_AVIC);
4680 
4681 	if (enable_apicv) {
4682 		pr_info("AVIC enabled\n");
4683 
4684 		amd_iommu_register_ga_log_notifier(&avic_ga_log_notifier);
4685 	} else {
4686 		svm_x86_ops.vcpu_blocking = NULL;
4687 		svm_x86_ops.vcpu_unblocking = NULL;
4688 	}
4689 
4690 	if (vls) {
4691 		if (!npt_enabled ||
4692 		    !boot_cpu_has(X86_FEATURE_V_VMSAVE_VMLOAD) ||
4693 		    !IS_ENABLED(CONFIG_X86_64)) {
4694 			vls = false;
4695 		} else {
4696 			pr_info("Virtual VMLOAD VMSAVE supported\n");
4697 		}
4698 	}
4699 
4700 	if (boot_cpu_has(X86_FEATURE_SVME_ADDR_CHK))
4701 		svm_gp_erratum_intercept = false;
4702 
4703 	if (vgif) {
4704 		if (!boot_cpu_has(X86_FEATURE_VGIF))
4705 			vgif = false;
4706 		else
4707 			pr_info("Virtual GIF supported\n");
4708 	}
4709 
4710 	if (lbrv) {
4711 		if (!boot_cpu_has(X86_FEATURE_LBRV))
4712 			lbrv = false;
4713 		else
4714 			pr_info("LBR virtualization supported\n");
4715 	}
4716 
4717 	if (!enable_pmu)
4718 		pr_info("PMU virtualization is disabled\n");
4719 
4720 	svm_set_cpu_caps();
4721 
4722 	/*
4723 	 * It seems that on AMD processors PTE's accessed bit is
4724 	 * being set by the CPU hardware before the NPF vmexit.
4725 	 * This is not expected behaviour and our tests fail because
4726 	 * of it.
4727 	 * A workaround here is to disable support for
4728 	 * GUEST_MAXPHYADDR < HOST_MAXPHYADDR if NPT is enabled.
4729 	 * In this case userspace can know if there is support using
4730 	 * KVM_CAP_SMALLER_MAXPHYADDR extension and decide how to handle
4731 	 * it
4732 	 * If future AMD CPU models change the behaviour described above,
4733 	 * this variable can be changed accordingly
4734 	 */
4735 	allow_smaller_maxphyaddr = !npt_enabled;
4736 
4737 	return 0;
4738 
4739 err:
4740 	svm_hardware_teardown();
4741 	return r;
4742 }
4743 
4744 
4745 static struct kvm_x86_init_ops svm_init_ops __initdata = {
4746 	.cpu_has_kvm_support = has_svm,
4747 	.disabled_by_bios = is_disabled,
4748 	.hardware_setup = svm_hardware_setup,
4749 	.check_processor_compatibility = svm_check_processor_compat,
4750 
4751 	.runtime_ops = &svm_x86_ops,
4752 };
4753 
4754 static int __init svm_init(void)
4755 {
4756 	__unused_size_checks();
4757 
4758 	return kvm_init(&svm_init_ops, sizeof(struct vcpu_svm),
4759 			__alignof__(struct vcpu_svm), THIS_MODULE);
4760 }
4761 
4762 static void __exit svm_exit(void)
4763 {
4764 	kvm_exit();
4765 }
4766 
4767 module_init(svm_init)
4768 module_exit(svm_exit)
4769