xref: /linux/arch/x86/kvm/svm/svm.c (revision 3a755ebcc2557e22b895b8976257f682c653db1d)
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 DEBUGCTL_RESERVED_BITS (~(0x3fULL))
66 
67 static bool erratum_383_found __read_mostly;
68 
69 u32 msrpm_offsets[MSRPM_OFFSETS] __read_mostly;
70 
71 /*
72  * Set osvw_len to higher value when updated Revision Guides
73  * are published and we know what the new status bits are
74  */
75 static uint64_t osvw_len = 4, osvw_status;
76 
77 static DEFINE_PER_CPU(u64, current_tsc_ratio);
78 
79 static const struct svm_direct_access_msrs {
80 	u32 index;   /* Index of the MSR */
81 	bool always; /* True if intercept is initially cleared */
82 } direct_access_msrs[MAX_DIRECT_ACCESS_MSRS] = {
83 	{ .index = MSR_STAR,				.always = true  },
84 	{ .index = MSR_IA32_SYSENTER_CS,		.always = true  },
85 	{ .index = MSR_IA32_SYSENTER_EIP,		.always = false },
86 	{ .index = MSR_IA32_SYSENTER_ESP,		.always = false },
87 #ifdef CONFIG_X86_64
88 	{ .index = MSR_GS_BASE,				.always = true  },
89 	{ .index = MSR_FS_BASE,				.always = true  },
90 	{ .index = MSR_KERNEL_GS_BASE,			.always = true  },
91 	{ .index = MSR_LSTAR,				.always = true  },
92 	{ .index = MSR_CSTAR,				.always = true  },
93 	{ .index = MSR_SYSCALL_MASK,			.always = true  },
94 #endif
95 	{ .index = MSR_IA32_SPEC_CTRL,			.always = false },
96 	{ .index = MSR_IA32_PRED_CMD,			.always = false },
97 	{ .index = MSR_IA32_LASTBRANCHFROMIP,		.always = false },
98 	{ .index = MSR_IA32_LASTBRANCHTOIP,		.always = false },
99 	{ .index = MSR_IA32_LASTINTFROMIP,		.always = false },
100 	{ .index = MSR_IA32_LASTINTTOIP,		.always = false },
101 	{ .index = MSR_EFER,				.always = false },
102 	{ .index = MSR_IA32_CR_PAT,			.always = false },
103 	{ .index = MSR_AMD64_SEV_ES_GHCB,		.always = true  },
104 	{ .index = MSR_INVALID,				.always = false },
105 };
106 
107 /*
108  * These 2 parameters are used to config the controls for Pause-Loop Exiting:
109  * pause_filter_count: On processors that support Pause filtering(indicated
110  *	by CPUID Fn8000_000A_EDX), the VMCB provides a 16 bit pause filter
111  *	count value. On VMRUN this value is loaded into an internal counter.
112  *	Each time a pause instruction is executed, this counter is decremented
113  *	until it reaches zero at which time a #VMEXIT is generated if pause
114  *	intercept is enabled. Refer to  AMD APM Vol 2 Section 15.14.4 Pause
115  *	Intercept Filtering for more details.
116  *	This also indicate if ple logic enabled.
117  *
118  * pause_filter_thresh: In addition, some processor families support advanced
119  *	pause filtering (indicated by CPUID Fn8000_000A_EDX) upper bound on
120  *	the amount of time a guest is allowed to execute in a pause loop.
121  *	In this mode, a 16-bit pause filter threshold field is added in the
122  *	VMCB. The threshold value is a cycle count that is used to reset the
123  *	pause counter. As with simple pause filtering, VMRUN loads the pause
124  *	count value from VMCB into an internal counter. Then, on each pause
125  *	instruction the hardware checks the elapsed number of cycles since
126  *	the most recent pause instruction against the pause filter threshold.
127  *	If the elapsed cycle count is greater than the pause filter threshold,
128  *	then the internal pause count is reloaded from the VMCB and execution
129  *	continues. If the elapsed cycle count is less than the pause filter
130  *	threshold, then the internal pause count is decremented. If the count
131  *	value is less than zero and PAUSE intercept is enabled, a #VMEXIT is
132  *	triggered. If advanced pause filtering is supported and pause filter
133  *	threshold field is set to zero, the filter will operate in the simpler,
134  *	count only mode.
135  */
136 
137 static unsigned short pause_filter_thresh = KVM_DEFAULT_PLE_GAP;
138 module_param(pause_filter_thresh, ushort, 0444);
139 
140 static unsigned short pause_filter_count = KVM_SVM_DEFAULT_PLE_WINDOW;
141 module_param(pause_filter_count, ushort, 0444);
142 
143 /* Default doubles per-vcpu window every exit. */
144 static unsigned short pause_filter_count_grow = KVM_DEFAULT_PLE_WINDOW_GROW;
145 module_param(pause_filter_count_grow, ushort, 0444);
146 
147 /* Default resets per-vcpu window every exit to pause_filter_count. */
148 static unsigned short pause_filter_count_shrink = KVM_DEFAULT_PLE_WINDOW_SHRINK;
149 module_param(pause_filter_count_shrink, ushort, 0444);
150 
151 /* Default is to compute the maximum so we can never overflow. */
152 static unsigned short pause_filter_count_max = KVM_SVM_DEFAULT_PLE_WINDOW_MAX;
153 module_param(pause_filter_count_max, ushort, 0444);
154 
155 /*
156  * Use nested page tables by default.  Note, NPT may get forced off by
157  * svm_hardware_setup() if it's unsupported by hardware or the host kernel.
158  */
159 bool npt_enabled = true;
160 module_param_named(npt, npt_enabled, bool, 0444);
161 
162 /* allow nested virtualization in KVM/SVM */
163 static int nested = true;
164 module_param(nested, int, S_IRUGO);
165 
166 /* enable/disable Next RIP Save */
167 static int nrips = true;
168 module_param(nrips, int, 0444);
169 
170 /* enable/disable Virtual VMLOAD VMSAVE */
171 static int vls = true;
172 module_param(vls, int, 0444);
173 
174 /* enable/disable Virtual GIF */
175 static int vgif = true;
176 module_param(vgif, int, 0444);
177 
178 /* enable/disable LBR virtualization */
179 static int lbrv = true;
180 module_param(lbrv, int, 0444);
181 
182 static int tsc_scaling = true;
183 module_param(tsc_scaling, int, 0444);
184 
185 /*
186  * enable / disable AVIC.  Because the defaults differ for APICv
187  * support between VMX and SVM we cannot use module_param_named.
188  */
189 static bool avic;
190 module_param(avic, bool, 0444);
191 
192 bool __read_mostly dump_invalid_vmcb;
193 module_param(dump_invalid_vmcb, bool, 0644);
194 
195 
196 bool intercept_smi = true;
197 module_param(intercept_smi, bool, 0444);
198 
199 
200 static bool svm_gp_erratum_intercept = true;
201 
202 static u8 rsm_ins_bytes[] = "\x0f\xaa";
203 
204 static unsigned long iopm_base;
205 
206 struct kvm_ldttss_desc {
207 	u16 limit0;
208 	u16 base0;
209 	unsigned base1:8, type:5, dpl:2, p:1;
210 	unsigned limit1:4, zero0:3, g:1, base2:8;
211 	u32 base3;
212 	u32 zero1;
213 } __attribute__((packed));
214 
215 DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
216 
217 /*
218  * Only MSR_TSC_AUX is switched via the user return hook.  EFER is switched via
219  * the VMCB, and the SYSCALL/SYSENTER MSRs are handled by VMLOAD/VMSAVE.
220  *
221  * RDTSCP and RDPID are not used in the kernel, specifically to allow KVM to
222  * defer the restoration of TSC_AUX until the CPU returns to userspace.
223  */
224 static int tsc_aux_uret_slot __read_mostly = -1;
225 
226 static const u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
227 
228 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
229 #define MSRS_RANGE_SIZE 2048
230 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
231 
232 u32 svm_msrpm_offset(u32 msr)
233 {
234 	u32 offset;
235 	int i;
236 
237 	for (i = 0; i < NUM_MSR_MAPS; i++) {
238 		if (msr < msrpm_ranges[i] ||
239 		    msr >= msrpm_ranges[i] + MSRS_IN_RANGE)
240 			continue;
241 
242 		offset  = (msr - msrpm_ranges[i]) / 4; /* 4 msrs per u8 */
243 		offset += (i * MSRS_RANGE_SIZE);       /* add range offset */
244 
245 		/* Now we have the u8 offset - but need the u32 offset */
246 		return offset / 4;
247 	}
248 
249 	/* MSR not in any range */
250 	return MSR_INVALID;
251 }
252 
253 static void svm_flush_tlb_current(struct kvm_vcpu *vcpu);
254 
255 static int get_npt_level(void)
256 {
257 #ifdef CONFIG_X86_64
258 	return pgtable_l5_enabled() ? PT64_ROOT_5LEVEL : PT64_ROOT_4LEVEL;
259 #else
260 	return PT32E_ROOT_LEVEL;
261 #endif
262 }
263 
264 int svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
265 {
266 	struct vcpu_svm *svm = to_svm(vcpu);
267 	u64 old_efer = vcpu->arch.efer;
268 	vcpu->arch.efer = efer;
269 
270 	if (!npt_enabled) {
271 		/* Shadow paging assumes NX to be available.  */
272 		efer |= EFER_NX;
273 
274 		if (!(efer & EFER_LMA))
275 			efer &= ~EFER_LME;
276 	}
277 
278 	if ((old_efer & EFER_SVME) != (efer & EFER_SVME)) {
279 		if (!(efer & EFER_SVME)) {
280 			svm_leave_nested(vcpu);
281 			svm_set_gif(svm, true);
282 			/* #GP intercept is still needed for vmware backdoor */
283 			if (!enable_vmware_backdoor)
284 				clr_exception_intercept(svm, GP_VECTOR);
285 
286 			/*
287 			 * Free the nested guest state, unless we are in SMM.
288 			 * In this case we will return to the nested guest
289 			 * as soon as we leave SMM.
290 			 */
291 			if (!is_smm(vcpu))
292 				svm_free_nested(svm);
293 
294 		} else {
295 			int ret = svm_allocate_nested(svm);
296 
297 			if (ret) {
298 				vcpu->arch.efer = old_efer;
299 				return ret;
300 			}
301 
302 			/*
303 			 * Never intercept #GP for SEV guests, KVM can't
304 			 * decrypt guest memory to workaround the erratum.
305 			 */
306 			if (svm_gp_erratum_intercept && !sev_guest(vcpu->kvm))
307 				set_exception_intercept(svm, GP_VECTOR);
308 		}
309 	}
310 
311 	svm->vmcb->save.efer = efer | EFER_SVME;
312 	vmcb_mark_dirty(svm->vmcb, VMCB_CR);
313 	return 0;
314 }
315 
316 static int is_external_interrupt(u32 info)
317 {
318 	info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
319 	return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
320 }
321 
322 static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu)
323 {
324 	struct vcpu_svm *svm = to_svm(vcpu);
325 	u32 ret = 0;
326 
327 	if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)
328 		ret = KVM_X86_SHADOW_INT_STI | KVM_X86_SHADOW_INT_MOV_SS;
329 	return ret;
330 }
331 
332 static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
333 {
334 	struct vcpu_svm *svm = to_svm(vcpu);
335 
336 	if (mask == 0)
337 		svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
338 	else
339 		svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK;
340 
341 }
342 
343 static int svm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
344 {
345 	struct vcpu_svm *svm = to_svm(vcpu);
346 
347 	/*
348 	 * SEV-ES does not expose the next RIP. The RIP update is controlled by
349 	 * the type of exit and the #VC handler in the guest.
350 	 */
351 	if (sev_es_guest(vcpu->kvm))
352 		goto done;
353 
354 	if (nrips && svm->vmcb->control.next_rip != 0) {
355 		WARN_ON_ONCE(!static_cpu_has(X86_FEATURE_NRIPS));
356 		svm->next_rip = svm->vmcb->control.next_rip;
357 	}
358 
359 	if (!svm->next_rip) {
360 		if (!kvm_emulate_instruction(vcpu, EMULTYPE_SKIP))
361 			return 0;
362 	} else {
363 		kvm_rip_write(vcpu, svm->next_rip);
364 	}
365 
366 done:
367 	svm_set_interrupt_shadow(vcpu, 0);
368 
369 	return 1;
370 }
371 
372 static void svm_queue_exception(struct kvm_vcpu *vcpu)
373 {
374 	struct vcpu_svm *svm = to_svm(vcpu);
375 	unsigned nr = vcpu->arch.exception.nr;
376 	bool has_error_code = vcpu->arch.exception.has_error_code;
377 	u32 error_code = vcpu->arch.exception.error_code;
378 
379 	kvm_deliver_exception_payload(vcpu);
380 
381 	if (nr == BP_VECTOR && !nrips) {
382 		unsigned long rip, old_rip = kvm_rip_read(vcpu);
383 
384 		/*
385 		 * For guest debugging where we have to reinject #BP if some
386 		 * INT3 is guest-owned:
387 		 * Emulate nRIP by moving RIP forward. Will fail if injection
388 		 * raises a fault that is not intercepted. Still better than
389 		 * failing in all cases.
390 		 */
391 		(void)svm_skip_emulated_instruction(vcpu);
392 		rip = kvm_rip_read(vcpu);
393 		svm->int3_rip = rip + svm->vmcb->save.cs.base;
394 		svm->int3_injected = rip - old_rip;
395 	}
396 
397 	svm->vmcb->control.event_inj = nr
398 		| SVM_EVTINJ_VALID
399 		| (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
400 		| SVM_EVTINJ_TYPE_EXEPT;
401 	svm->vmcb->control.event_inj_err = error_code;
402 }
403 
404 static void svm_init_erratum_383(void)
405 {
406 	u32 low, high;
407 	int err;
408 	u64 val;
409 
410 	if (!static_cpu_has_bug(X86_BUG_AMD_TLB_MMATCH))
411 		return;
412 
413 	/* Use _safe variants to not break nested virtualization */
414 	val = native_read_msr_safe(MSR_AMD64_DC_CFG, &err);
415 	if (err)
416 		return;
417 
418 	val |= (1ULL << 47);
419 
420 	low  = lower_32_bits(val);
421 	high = upper_32_bits(val);
422 
423 	native_write_msr_safe(MSR_AMD64_DC_CFG, low, high);
424 
425 	erratum_383_found = true;
426 }
427 
428 static void svm_init_osvw(struct kvm_vcpu *vcpu)
429 {
430 	/*
431 	 * Guests should see errata 400 and 415 as fixed (assuming that
432 	 * HLT and IO instructions are intercepted).
433 	 */
434 	vcpu->arch.osvw.length = (osvw_len >= 3) ? (osvw_len) : 3;
435 	vcpu->arch.osvw.status = osvw_status & ~(6ULL);
436 
437 	/*
438 	 * By increasing VCPU's osvw.length to 3 we are telling the guest that
439 	 * all osvw.status bits inside that length, including bit 0 (which is
440 	 * reserved for erratum 298), are valid. However, if host processor's
441 	 * osvw_len is 0 then osvw_status[0] carries no information. We need to
442 	 * be conservative here and therefore we tell the guest that erratum 298
443 	 * is present (because we really don't know).
444 	 */
445 	if (osvw_len == 0 && boot_cpu_data.x86 == 0x10)
446 		vcpu->arch.osvw.status |= 1;
447 }
448 
449 static int has_svm(void)
450 {
451 	const char *msg;
452 
453 	if (!cpu_has_svm(&msg)) {
454 		printk(KERN_INFO "has_svm: %s\n", msg);
455 		return 0;
456 	}
457 
458 	if (cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT)) {
459 		pr_info("KVM is unsupported when running as an SEV guest\n");
460 		return 0;
461 	}
462 
463 	return 1;
464 }
465 
466 static void svm_hardware_disable(void)
467 {
468 	/* Make sure we clean up behind us */
469 	if (tsc_scaling)
470 		wrmsrl(MSR_AMD64_TSC_RATIO, SVM_TSC_RATIO_DEFAULT);
471 
472 	cpu_svm_disable();
473 
474 	amd_pmu_disable_virt();
475 }
476 
477 static int svm_hardware_enable(void)
478 {
479 
480 	struct svm_cpu_data *sd;
481 	uint64_t efer;
482 	struct desc_struct *gdt;
483 	int me = raw_smp_processor_id();
484 
485 	rdmsrl(MSR_EFER, efer);
486 	if (efer & EFER_SVME)
487 		return -EBUSY;
488 
489 	if (!has_svm()) {
490 		pr_err("%s: err EOPNOTSUPP on %d\n", __func__, me);
491 		return -EINVAL;
492 	}
493 	sd = per_cpu(svm_data, me);
494 	if (!sd) {
495 		pr_err("%s: svm_data is NULL on %d\n", __func__, me);
496 		return -EINVAL;
497 	}
498 
499 	sd->asid_generation = 1;
500 	sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
501 	sd->next_asid = sd->max_asid + 1;
502 	sd->min_asid = max_sev_asid + 1;
503 
504 	gdt = get_current_gdt_rw();
505 	sd->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
506 
507 	wrmsrl(MSR_EFER, efer | EFER_SVME);
508 
509 	wrmsrl(MSR_VM_HSAVE_PA, __sme_page_pa(sd->save_area));
510 
511 	if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
512 		/*
513 		 * Set the default value, even if we don't use TSC scaling
514 		 * to avoid having stale value in the msr
515 		 */
516 		wrmsrl(MSR_AMD64_TSC_RATIO, SVM_TSC_RATIO_DEFAULT);
517 		__this_cpu_write(current_tsc_ratio, SVM_TSC_RATIO_DEFAULT);
518 	}
519 
520 
521 	/*
522 	 * Get OSVW bits.
523 	 *
524 	 * Note that it is possible to have a system with mixed processor
525 	 * revisions and therefore different OSVW bits. If bits are not the same
526 	 * on different processors then choose the worst case (i.e. if erratum
527 	 * is present on one processor and not on another then assume that the
528 	 * erratum is present everywhere).
529 	 */
530 	if (cpu_has(&boot_cpu_data, X86_FEATURE_OSVW)) {
531 		uint64_t len, status = 0;
532 		int err;
533 
534 		len = native_read_msr_safe(MSR_AMD64_OSVW_ID_LENGTH, &err);
535 		if (!err)
536 			status = native_read_msr_safe(MSR_AMD64_OSVW_STATUS,
537 						      &err);
538 
539 		if (err)
540 			osvw_status = osvw_len = 0;
541 		else {
542 			if (len < osvw_len)
543 				osvw_len = len;
544 			osvw_status |= status;
545 			osvw_status &= (1ULL << osvw_len) - 1;
546 		}
547 	} else
548 		osvw_status = osvw_len = 0;
549 
550 	svm_init_erratum_383();
551 
552 	amd_pmu_enable_virt();
553 
554 	return 0;
555 }
556 
557 static void svm_cpu_uninit(int cpu)
558 {
559 	struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
560 
561 	if (!sd)
562 		return;
563 
564 	per_cpu(svm_data, cpu) = NULL;
565 	kfree(sd->sev_vmcbs);
566 	__free_page(sd->save_area);
567 	kfree(sd);
568 }
569 
570 static int svm_cpu_init(int cpu)
571 {
572 	struct svm_cpu_data *sd;
573 	int ret = -ENOMEM;
574 
575 	sd = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
576 	if (!sd)
577 		return ret;
578 	sd->cpu = cpu;
579 	sd->save_area = alloc_page(GFP_KERNEL | __GFP_ZERO);
580 	if (!sd->save_area)
581 		goto free_cpu_data;
582 
583 	ret = sev_cpu_init(sd);
584 	if (ret)
585 		goto free_save_area;
586 
587 	per_cpu(svm_data, cpu) = sd;
588 
589 	return 0;
590 
591 free_save_area:
592 	__free_page(sd->save_area);
593 free_cpu_data:
594 	kfree(sd);
595 	return ret;
596 
597 }
598 
599 static int direct_access_msr_slot(u32 msr)
600 {
601 	u32 i;
602 
603 	for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++)
604 		if (direct_access_msrs[i].index == msr)
605 			return i;
606 
607 	return -ENOENT;
608 }
609 
610 static void set_shadow_msr_intercept(struct kvm_vcpu *vcpu, u32 msr, int read,
611 				     int write)
612 {
613 	struct vcpu_svm *svm = to_svm(vcpu);
614 	int slot = direct_access_msr_slot(msr);
615 
616 	if (slot == -ENOENT)
617 		return;
618 
619 	/* Set the shadow bitmaps to the desired intercept states */
620 	if (read)
621 		set_bit(slot, svm->shadow_msr_intercept.read);
622 	else
623 		clear_bit(slot, svm->shadow_msr_intercept.read);
624 
625 	if (write)
626 		set_bit(slot, svm->shadow_msr_intercept.write);
627 	else
628 		clear_bit(slot, svm->shadow_msr_intercept.write);
629 }
630 
631 static bool valid_msr_intercept(u32 index)
632 {
633 	return direct_access_msr_slot(index) != -ENOENT;
634 }
635 
636 static bool msr_write_intercepted(struct kvm_vcpu *vcpu, u32 msr)
637 {
638 	u8 bit_write;
639 	unsigned long tmp;
640 	u32 offset;
641 	u32 *msrpm;
642 
643 	msrpm = is_guest_mode(vcpu) ? to_svm(vcpu)->nested.msrpm:
644 				      to_svm(vcpu)->msrpm;
645 
646 	offset    = svm_msrpm_offset(msr);
647 	bit_write = 2 * (msr & 0x0f) + 1;
648 	tmp       = msrpm[offset];
649 
650 	BUG_ON(offset == MSR_INVALID);
651 
652 	return !!test_bit(bit_write,  &tmp);
653 }
654 
655 static void set_msr_interception_bitmap(struct kvm_vcpu *vcpu, u32 *msrpm,
656 					u32 msr, int read, int write)
657 {
658 	struct vcpu_svm *svm = to_svm(vcpu);
659 	u8 bit_read, bit_write;
660 	unsigned long tmp;
661 	u32 offset;
662 
663 	/*
664 	 * If this warning triggers extend the direct_access_msrs list at the
665 	 * beginning of the file
666 	 */
667 	WARN_ON(!valid_msr_intercept(msr));
668 
669 	/* Enforce non allowed MSRs to trap */
670 	if (read && !kvm_msr_allowed(vcpu, msr, KVM_MSR_FILTER_READ))
671 		read = 0;
672 
673 	if (write && !kvm_msr_allowed(vcpu, msr, KVM_MSR_FILTER_WRITE))
674 		write = 0;
675 
676 	offset    = svm_msrpm_offset(msr);
677 	bit_read  = 2 * (msr & 0x0f);
678 	bit_write = 2 * (msr & 0x0f) + 1;
679 	tmp       = msrpm[offset];
680 
681 	BUG_ON(offset == MSR_INVALID);
682 
683 	read  ? clear_bit(bit_read,  &tmp) : set_bit(bit_read,  &tmp);
684 	write ? clear_bit(bit_write, &tmp) : set_bit(bit_write, &tmp);
685 
686 	msrpm[offset] = tmp;
687 
688 	svm_hv_vmcb_dirty_nested_enlightenments(vcpu);
689 	svm->nested.force_msr_bitmap_recalc = true;
690 }
691 
692 void set_msr_interception(struct kvm_vcpu *vcpu, u32 *msrpm, u32 msr,
693 			  int read, int write)
694 {
695 	set_shadow_msr_intercept(vcpu, msr, read, write);
696 	set_msr_interception_bitmap(vcpu, msrpm, msr, read, write);
697 }
698 
699 u32 *svm_vcpu_alloc_msrpm(void)
700 {
701 	unsigned int order = get_order(MSRPM_SIZE);
702 	struct page *pages = alloc_pages(GFP_KERNEL_ACCOUNT, order);
703 	u32 *msrpm;
704 
705 	if (!pages)
706 		return NULL;
707 
708 	msrpm = page_address(pages);
709 	memset(msrpm, 0xff, PAGE_SIZE * (1 << order));
710 
711 	return msrpm;
712 }
713 
714 void svm_vcpu_init_msrpm(struct kvm_vcpu *vcpu, u32 *msrpm)
715 {
716 	int i;
717 
718 	for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
719 		if (!direct_access_msrs[i].always)
720 			continue;
721 		set_msr_interception(vcpu, msrpm, direct_access_msrs[i].index, 1, 1);
722 	}
723 }
724 
725 
726 void svm_vcpu_free_msrpm(u32 *msrpm)
727 {
728 	__free_pages(virt_to_page(msrpm), get_order(MSRPM_SIZE));
729 }
730 
731 static void svm_msr_filter_changed(struct kvm_vcpu *vcpu)
732 {
733 	struct vcpu_svm *svm = to_svm(vcpu);
734 	u32 i;
735 
736 	/*
737 	 * Set intercept permissions for all direct access MSRs again. They
738 	 * will automatically get filtered through the MSR filter, so we are
739 	 * back in sync after this.
740 	 */
741 	for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
742 		u32 msr = direct_access_msrs[i].index;
743 		u32 read = test_bit(i, svm->shadow_msr_intercept.read);
744 		u32 write = test_bit(i, svm->shadow_msr_intercept.write);
745 
746 		set_msr_interception_bitmap(vcpu, svm->msrpm, msr, read, write);
747 	}
748 }
749 
750 static void add_msr_offset(u32 offset)
751 {
752 	int i;
753 
754 	for (i = 0; i < MSRPM_OFFSETS; ++i) {
755 
756 		/* Offset already in list? */
757 		if (msrpm_offsets[i] == offset)
758 			return;
759 
760 		/* Slot used by another offset? */
761 		if (msrpm_offsets[i] != MSR_INVALID)
762 			continue;
763 
764 		/* Add offset to list */
765 		msrpm_offsets[i] = offset;
766 
767 		return;
768 	}
769 
770 	/*
771 	 * If this BUG triggers the msrpm_offsets table has an overflow. Just
772 	 * increase MSRPM_OFFSETS in this case.
773 	 */
774 	BUG();
775 }
776 
777 static void init_msrpm_offsets(void)
778 {
779 	int i;
780 
781 	memset(msrpm_offsets, 0xff, sizeof(msrpm_offsets));
782 
783 	for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
784 		u32 offset;
785 
786 		offset = svm_msrpm_offset(direct_access_msrs[i].index);
787 		BUG_ON(offset == MSR_INVALID);
788 
789 		add_msr_offset(offset);
790 	}
791 }
792 
793 static void svm_enable_lbrv(struct kvm_vcpu *vcpu)
794 {
795 	struct vcpu_svm *svm = to_svm(vcpu);
796 
797 	svm->vmcb->control.virt_ext |= LBR_CTL_ENABLE_MASK;
798 	set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
799 	set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
800 	set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
801 	set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
802 }
803 
804 static void svm_disable_lbrv(struct kvm_vcpu *vcpu)
805 {
806 	struct vcpu_svm *svm = to_svm(vcpu);
807 
808 	svm->vmcb->control.virt_ext &= ~LBR_CTL_ENABLE_MASK;
809 	set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
810 	set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
811 	set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
812 	set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
813 }
814 
815 void disable_nmi_singlestep(struct vcpu_svm *svm)
816 {
817 	svm->nmi_singlestep = false;
818 
819 	if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP)) {
820 		/* Clear our flags if they were not set by the guest */
821 		if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF))
822 			svm->vmcb->save.rflags &= ~X86_EFLAGS_TF;
823 		if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_RF))
824 			svm->vmcb->save.rflags &= ~X86_EFLAGS_RF;
825 	}
826 }
827 
828 static void grow_ple_window(struct kvm_vcpu *vcpu)
829 {
830 	struct vcpu_svm *svm = to_svm(vcpu);
831 	struct vmcb_control_area *control = &svm->vmcb->control;
832 	int old = control->pause_filter_count;
833 
834 	control->pause_filter_count = __grow_ple_window(old,
835 							pause_filter_count,
836 							pause_filter_count_grow,
837 							pause_filter_count_max);
838 
839 	if (control->pause_filter_count != old) {
840 		vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
841 		trace_kvm_ple_window_update(vcpu->vcpu_id,
842 					    control->pause_filter_count, old);
843 	}
844 }
845 
846 static void shrink_ple_window(struct kvm_vcpu *vcpu)
847 {
848 	struct vcpu_svm *svm = to_svm(vcpu);
849 	struct vmcb_control_area *control = &svm->vmcb->control;
850 	int old = control->pause_filter_count;
851 
852 	control->pause_filter_count =
853 				__shrink_ple_window(old,
854 						    pause_filter_count,
855 						    pause_filter_count_shrink,
856 						    pause_filter_count);
857 	if (control->pause_filter_count != old) {
858 		vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
859 		trace_kvm_ple_window_update(vcpu->vcpu_id,
860 					    control->pause_filter_count, old);
861 	}
862 }
863 
864 static void svm_hardware_unsetup(void)
865 {
866 	int cpu;
867 
868 	sev_hardware_unsetup();
869 
870 	for_each_possible_cpu(cpu)
871 		svm_cpu_uninit(cpu);
872 
873 	__free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT),
874 	get_order(IOPM_SIZE));
875 	iopm_base = 0;
876 }
877 
878 static void init_seg(struct vmcb_seg *seg)
879 {
880 	seg->selector = 0;
881 	seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
882 		      SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
883 	seg->limit = 0xffff;
884 	seg->base = 0;
885 }
886 
887 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
888 {
889 	seg->selector = 0;
890 	seg->attrib = SVM_SELECTOR_P_MASK | type;
891 	seg->limit = 0xffff;
892 	seg->base = 0;
893 }
894 
895 static u64 svm_get_l2_tsc_offset(struct kvm_vcpu *vcpu)
896 {
897 	struct vcpu_svm *svm = to_svm(vcpu);
898 
899 	return svm->nested.ctl.tsc_offset;
900 }
901 
902 static u64 svm_get_l2_tsc_multiplier(struct kvm_vcpu *vcpu)
903 {
904 	struct vcpu_svm *svm = to_svm(vcpu);
905 
906 	return svm->tsc_ratio_msr;
907 }
908 
909 static void svm_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
910 {
911 	struct vcpu_svm *svm = to_svm(vcpu);
912 
913 	svm->vmcb01.ptr->control.tsc_offset = vcpu->arch.l1_tsc_offset;
914 	svm->vmcb->control.tsc_offset = offset;
915 	vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
916 }
917 
918 void svm_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 multiplier)
919 {
920 	wrmsrl(MSR_AMD64_TSC_RATIO, multiplier);
921 }
922 
923 /* Evaluate instruction intercepts that depend on guest CPUID features. */
924 static void svm_recalc_instruction_intercepts(struct kvm_vcpu *vcpu,
925 					      struct vcpu_svm *svm)
926 {
927 	/*
928 	 * Intercept INVPCID if shadow paging is enabled to sync/free shadow
929 	 * roots, or if INVPCID is disabled in the guest to inject #UD.
930 	 */
931 	if (kvm_cpu_cap_has(X86_FEATURE_INVPCID)) {
932 		if (!npt_enabled ||
933 		    !guest_cpuid_has(&svm->vcpu, X86_FEATURE_INVPCID))
934 			svm_set_intercept(svm, INTERCEPT_INVPCID);
935 		else
936 			svm_clr_intercept(svm, INTERCEPT_INVPCID);
937 	}
938 
939 	if (kvm_cpu_cap_has(X86_FEATURE_RDTSCP)) {
940 		if (guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
941 			svm_clr_intercept(svm, INTERCEPT_RDTSCP);
942 		else
943 			svm_set_intercept(svm, INTERCEPT_RDTSCP);
944 	}
945 }
946 
947 static inline void init_vmcb_after_set_cpuid(struct kvm_vcpu *vcpu)
948 {
949 	struct vcpu_svm *svm = to_svm(vcpu);
950 
951 	if (guest_cpuid_is_intel(vcpu)) {
952 		/*
953 		 * We must intercept SYSENTER_EIP and SYSENTER_ESP
954 		 * accesses because the processor only stores 32 bits.
955 		 * For the same reason we cannot use virtual VMLOAD/VMSAVE.
956 		 */
957 		svm_set_intercept(svm, INTERCEPT_VMLOAD);
958 		svm_set_intercept(svm, INTERCEPT_VMSAVE);
959 		svm->vmcb->control.virt_ext &= ~VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK;
960 
961 		set_msr_interception(vcpu, svm->msrpm, MSR_IA32_SYSENTER_EIP, 0, 0);
962 		set_msr_interception(vcpu, svm->msrpm, MSR_IA32_SYSENTER_ESP, 0, 0);
963 	} else {
964 		/*
965 		 * If hardware supports Virtual VMLOAD VMSAVE then enable it
966 		 * in VMCB and clear intercepts to avoid #VMEXIT.
967 		 */
968 		if (vls) {
969 			svm_clr_intercept(svm, INTERCEPT_VMLOAD);
970 			svm_clr_intercept(svm, INTERCEPT_VMSAVE);
971 			svm->vmcb->control.virt_ext |= VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK;
972 		}
973 		/* No need to intercept these MSRs */
974 		set_msr_interception(vcpu, svm->msrpm, MSR_IA32_SYSENTER_EIP, 1, 1);
975 		set_msr_interception(vcpu, svm->msrpm, MSR_IA32_SYSENTER_ESP, 1, 1);
976 	}
977 }
978 
979 static void init_vmcb(struct kvm_vcpu *vcpu)
980 {
981 	struct vcpu_svm *svm = to_svm(vcpu);
982 	struct vmcb_control_area *control = &svm->vmcb->control;
983 	struct vmcb_save_area *save = &svm->vmcb->save;
984 
985 	svm_set_intercept(svm, INTERCEPT_CR0_READ);
986 	svm_set_intercept(svm, INTERCEPT_CR3_READ);
987 	svm_set_intercept(svm, INTERCEPT_CR4_READ);
988 	svm_set_intercept(svm, INTERCEPT_CR0_WRITE);
989 	svm_set_intercept(svm, INTERCEPT_CR3_WRITE);
990 	svm_set_intercept(svm, INTERCEPT_CR4_WRITE);
991 	if (!kvm_vcpu_apicv_active(vcpu))
992 		svm_set_intercept(svm, INTERCEPT_CR8_WRITE);
993 
994 	set_dr_intercepts(svm);
995 
996 	set_exception_intercept(svm, PF_VECTOR);
997 	set_exception_intercept(svm, UD_VECTOR);
998 	set_exception_intercept(svm, MC_VECTOR);
999 	set_exception_intercept(svm, AC_VECTOR);
1000 	set_exception_intercept(svm, DB_VECTOR);
1001 	/*
1002 	 * Guest access to VMware backdoor ports could legitimately
1003 	 * trigger #GP because of TSS I/O permission bitmap.
1004 	 * We intercept those #GP and allow access to them anyway
1005 	 * as VMware does.  Don't intercept #GP for SEV guests as KVM can't
1006 	 * decrypt guest memory to decode the faulting instruction.
1007 	 */
1008 	if (enable_vmware_backdoor && !sev_guest(vcpu->kvm))
1009 		set_exception_intercept(svm, GP_VECTOR);
1010 
1011 	svm_set_intercept(svm, INTERCEPT_INTR);
1012 	svm_set_intercept(svm, INTERCEPT_NMI);
1013 
1014 	if (intercept_smi)
1015 		svm_set_intercept(svm, INTERCEPT_SMI);
1016 
1017 	svm_set_intercept(svm, INTERCEPT_SELECTIVE_CR0);
1018 	svm_set_intercept(svm, INTERCEPT_RDPMC);
1019 	svm_set_intercept(svm, INTERCEPT_CPUID);
1020 	svm_set_intercept(svm, INTERCEPT_INVD);
1021 	svm_set_intercept(svm, INTERCEPT_INVLPG);
1022 	svm_set_intercept(svm, INTERCEPT_INVLPGA);
1023 	svm_set_intercept(svm, INTERCEPT_IOIO_PROT);
1024 	svm_set_intercept(svm, INTERCEPT_MSR_PROT);
1025 	svm_set_intercept(svm, INTERCEPT_TASK_SWITCH);
1026 	svm_set_intercept(svm, INTERCEPT_SHUTDOWN);
1027 	svm_set_intercept(svm, INTERCEPT_VMRUN);
1028 	svm_set_intercept(svm, INTERCEPT_VMMCALL);
1029 	svm_set_intercept(svm, INTERCEPT_VMLOAD);
1030 	svm_set_intercept(svm, INTERCEPT_VMSAVE);
1031 	svm_set_intercept(svm, INTERCEPT_STGI);
1032 	svm_set_intercept(svm, INTERCEPT_CLGI);
1033 	svm_set_intercept(svm, INTERCEPT_SKINIT);
1034 	svm_set_intercept(svm, INTERCEPT_WBINVD);
1035 	svm_set_intercept(svm, INTERCEPT_XSETBV);
1036 	svm_set_intercept(svm, INTERCEPT_RDPRU);
1037 	svm_set_intercept(svm, INTERCEPT_RSM);
1038 
1039 	if (!kvm_mwait_in_guest(vcpu->kvm)) {
1040 		svm_set_intercept(svm, INTERCEPT_MONITOR);
1041 		svm_set_intercept(svm, INTERCEPT_MWAIT);
1042 	}
1043 
1044 	if (!kvm_hlt_in_guest(vcpu->kvm))
1045 		svm_set_intercept(svm, INTERCEPT_HLT);
1046 
1047 	control->iopm_base_pa = __sme_set(iopm_base);
1048 	control->msrpm_base_pa = __sme_set(__pa(svm->msrpm));
1049 	control->int_ctl = V_INTR_MASKING_MASK;
1050 
1051 	init_seg(&save->es);
1052 	init_seg(&save->ss);
1053 	init_seg(&save->ds);
1054 	init_seg(&save->fs);
1055 	init_seg(&save->gs);
1056 
1057 	save->cs.selector = 0xf000;
1058 	save->cs.base = 0xffff0000;
1059 	/* Executable/Readable Code Segment */
1060 	save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
1061 		SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
1062 	save->cs.limit = 0xffff;
1063 
1064 	save->gdtr.base = 0;
1065 	save->gdtr.limit = 0xffff;
1066 	save->idtr.base = 0;
1067 	save->idtr.limit = 0xffff;
1068 
1069 	init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
1070 	init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
1071 
1072 	if (npt_enabled) {
1073 		/* Setup VMCB for Nested Paging */
1074 		control->nested_ctl |= SVM_NESTED_CTL_NP_ENABLE;
1075 		svm_clr_intercept(svm, INTERCEPT_INVLPG);
1076 		clr_exception_intercept(svm, PF_VECTOR);
1077 		svm_clr_intercept(svm, INTERCEPT_CR3_READ);
1078 		svm_clr_intercept(svm, INTERCEPT_CR3_WRITE);
1079 		save->g_pat = vcpu->arch.pat;
1080 		save->cr3 = 0;
1081 	}
1082 	svm->current_vmcb->asid_generation = 0;
1083 	svm->asid = 0;
1084 
1085 	svm->nested.vmcb12_gpa = INVALID_GPA;
1086 	svm->nested.last_vmcb12_gpa = INVALID_GPA;
1087 
1088 	if (!kvm_pause_in_guest(vcpu->kvm)) {
1089 		control->pause_filter_count = pause_filter_count;
1090 		if (pause_filter_thresh)
1091 			control->pause_filter_thresh = pause_filter_thresh;
1092 		svm_set_intercept(svm, INTERCEPT_PAUSE);
1093 	} else {
1094 		svm_clr_intercept(svm, INTERCEPT_PAUSE);
1095 	}
1096 
1097 	svm_recalc_instruction_intercepts(vcpu, svm);
1098 
1099 	/*
1100 	 * If the host supports V_SPEC_CTRL then disable the interception
1101 	 * of MSR_IA32_SPEC_CTRL.
1102 	 */
1103 	if (boot_cpu_has(X86_FEATURE_V_SPEC_CTRL))
1104 		set_msr_interception(vcpu, svm->msrpm, MSR_IA32_SPEC_CTRL, 1, 1);
1105 
1106 	if (kvm_vcpu_apicv_active(vcpu))
1107 		avic_init_vmcb(svm);
1108 
1109 	if (vgif) {
1110 		svm_clr_intercept(svm, INTERCEPT_STGI);
1111 		svm_clr_intercept(svm, INTERCEPT_CLGI);
1112 		svm->vmcb->control.int_ctl |= V_GIF_ENABLE_MASK;
1113 	}
1114 
1115 	if (sev_guest(vcpu->kvm)) {
1116 		svm->vmcb->control.nested_ctl |= SVM_NESTED_CTL_SEV_ENABLE;
1117 		clr_exception_intercept(svm, UD_VECTOR);
1118 
1119 		if (sev_es_guest(vcpu->kvm)) {
1120 			/* Perform SEV-ES specific VMCB updates */
1121 			sev_es_init_vmcb(svm);
1122 		}
1123 	}
1124 
1125 	svm_hv_init_vmcb(svm->vmcb);
1126 	init_vmcb_after_set_cpuid(vcpu);
1127 
1128 	vmcb_mark_all_dirty(svm->vmcb);
1129 
1130 	enable_gif(svm);
1131 }
1132 
1133 static void __svm_vcpu_reset(struct kvm_vcpu *vcpu)
1134 {
1135 	struct vcpu_svm *svm = to_svm(vcpu);
1136 
1137 	svm_vcpu_init_msrpm(vcpu, svm->msrpm);
1138 
1139 	svm_init_osvw(vcpu);
1140 	vcpu->arch.microcode_version = 0x01000065;
1141 	svm->tsc_ratio_msr = kvm_default_tsc_scaling_ratio;
1142 
1143 	if (sev_es_guest(vcpu->kvm))
1144 		sev_es_vcpu_reset(svm);
1145 }
1146 
1147 static void svm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
1148 {
1149 	struct vcpu_svm *svm = to_svm(vcpu);
1150 
1151 	svm->spec_ctrl = 0;
1152 	svm->virt_spec_ctrl = 0;
1153 
1154 	init_vmcb(vcpu);
1155 
1156 	if (!init_event)
1157 		__svm_vcpu_reset(vcpu);
1158 }
1159 
1160 void svm_switch_vmcb(struct vcpu_svm *svm, struct kvm_vmcb_info *target_vmcb)
1161 {
1162 	svm->current_vmcb = target_vmcb;
1163 	svm->vmcb = target_vmcb->ptr;
1164 }
1165 
1166 static int svm_vcpu_create(struct kvm_vcpu *vcpu)
1167 {
1168 	struct vcpu_svm *svm;
1169 	struct page *vmcb01_page;
1170 	struct page *vmsa_page = NULL;
1171 	int err;
1172 
1173 	BUILD_BUG_ON(offsetof(struct vcpu_svm, vcpu) != 0);
1174 	svm = to_svm(vcpu);
1175 
1176 	err = -ENOMEM;
1177 	vmcb01_page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
1178 	if (!vmcb01_page)
1179 		goto out;
1180 
1181 	if (sev_es_guest(vcpu->kvm)) {
1182 		/*
1183 		 * SEV-ES guests require a separate VMSA page used to contain
1184 		 * the encrypted register state of the guest.
1185 		 */
1186 		vmsa_page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
1187 		if (!vmsa_page)
1188 			goto error_free_vmcb_page;
1189 
1190 		/*
1191 		 * SEV-ES guests maintain an encrypted version of their FPU
1192 		 * state which is restored and saved on VMRUN and VMEXIT.
1193 		 * Mark vcpu->arch.guest_fpu->fpstate as scratch so it won't
1194 		 * do xsave/xrstor on it.
1195 		 */
1196 		fpstate_set_confidential(&vcpu->arch.guest_fpu);
1197 	}
1198 
1199 	err = avic_init_vcpu(svm);
1200 	if (err)
1201 		goto error_free_vmsa_page;
1202 
1203 	svm->msrpm = svm_vcpu_alloc_msrpm();
1204 	if (!svm->msrpm) {
1205 		err = -ENOMEM;
1206 		goto error_free_vmsa_page;
1207 	}
1208 
1209 	svm->vmcb01.ptr = page_address(vmcb01_page);
1210 	svm->vmcb01.pa = __sme_set(page_to_pfn(vmcb01_page) << PAGE_SHIFT);
1211 	svm_switch_vmcb(svm, &svm->vmcb01);
1212 
1213 	if (vmsa_page)
1214 		svm->sev_es.vmsa = page_address(vmsa_page);
1215 
1216 	svm->guest_state_loaded = false;
1217 
1218 	return 0;
1219 
1220 error_free_vmsa_page:
1221 	if (vmsa_page)
1222 		__free_page(vmsa_page);
1223 error_free_vmcb_page:
1224 	__free_page(vmcb01_page);
1225 out:
1226 	return err;
1227 }
1228 
1229 static void svm_clear_current_vmcb(struct vmcb *vmcb)
1230 {
1231 	int i;
1232 
1233 	for_each_online_cpu(i)
1234 		cmpxchg(&per_cpu(svm_data, i)->current_vmcb, vmcb, NULL);
1235 }
1236 
1237 static void svm_vcpu_free(struct kvm_vcpu *vcpu)
1238 {
1239 	struct vcpu_svm *svm = to_svm(vcpu);
1240 
1241 	/*
1242 	 * The vmcb page can be recycled, causing a false negative in
1243 	 * svm_vcpu_load(). So, ensure that no logical CPU has this
1244 	 * vmcb page recorded as its current vmcb.
1245 	 */
1246 	svm_clear_current_vmcb(svm->vmcb);
1247 
1248 	svm_free_nested(svm);
1249 
1250 	sev_free_vcpu(vcpu);
1251 
1252 	__free_page(pfn_to_page(__sme_clr(svm->vmcb01.pa) >> PAGE_SHIFT));
1253 	__free_pages(virt_to_page(svm->msrpm), get_order(MSRPM_SIZE));
1254 }
1255 
1256 static void svm_prepare_switch_to_guest(struct kvm_vcpu *vcpu)
1257 {
1258 	struct vcpu_svm *svm = to_svm(vcpu);
1259 	struct svm_cpu_data *sd = per_cpu(svm_data, vcpu->cpu);
1260 
1261 	if (sev_es_guest(vcpu->kvm))
1262 		sev_es_unmap_ghcb(svm);
1263 
1264 	if (svm->guest_state_loaded)
1265 		return;
1266 
1267 	/*
1268 	 * Save additional host state that will be restored on VMEXIT (sev-es)
1269 	 * or subsequent vmload of host save area.
1270 	 */
1271 	vmsave(__sme_page_pa(sd->save_area));
1272 	if (sev_es_guest(vcpu->kvm)) {
1273 		struct sev_es_save_area *hostsa;
1274 		hostsa = (struct sev_es_save_area *)(page_address(sd->save_area) + 0x400);
1275 
1276 		sev_es_prepare_switch_to_guest(hostsa);
1277 	}
1278 
1279 	if (tsc_scaling) {
1280 		u64 tsc_ratio = vcpu->arch.tsc_scaling_ratio;
1281 		if (tsc_ratio != __this_cpu_read(current_tsc_ratio)) {
1282 			__this_cpu_write(current_tsc_ratio, tsc_ratio);
1283 			wrmsrl(MSR_AMD64_TSC_RATIO, tsc_ratio);
1284 		}
1285 	}
1286 
1287 	if (likely(tsc_aux_uret_slot >= 0))
1288 		kvm_set_user_return_msr(tsc_aux_uret_slot, svm->tsc_aux, -1ull);
1289 
1290 	svm->guest_state_loaded = true;
1291 }
1292 
1293 static void svm_prepare_host_switch(struct kvm_vcpu *vcpu)
1294 {
1295 	to_svm(vcpu)->guest_state_loaded = false;
1296 }
1297 
1298 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1299 {
1300 	struct vcpu_svm *svm = to_svm(vcpu);
1301 	struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
1302 
1303 	if (sd->current_vmcb != svm->vmcb) {
1304 		sd->current_vmcb = svm->vmcb;
1305 		indirect_branch_prediction_barrier();
1306 	}
1307 	if (kvm_vcpu_apicv_active(vcpu))
1308 		__avic_vcpu_load(vcpu, cpu);
1309 }
1310 
1311 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
1312 {
1313 	if (kvm_vcpu_apicv_active(vcpu))
1314 		__avic_vcpu_put(vcpu);
1315 
1316 	svm_prepare_host_switch(vcpu);
1317 
1318 	++vcpu->stat.host_state_reload;
1319 }
1320 
1321 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
1322 {
1323 	struct vcpu_svm *svm = to_svm(vcpu);
1324 	unsigned long rflags = svm->vmcb->save.rflags;
1325 
1326 	if (svm->nmi_singlestep) {
1327 		/* Hide our flags if they were not set by the guest */
1328 		if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF))
1329 			rflags &= ~X86_EFLAGS_TF;
1330 		if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_RF))
1331 			rflags &= ~X86_EFLAGS_RF;
1332 	}
1333 	return rflags;
1334 }
1335 
1336 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1337 {
1338 	if (to_svm(vcpu)->nmi_singlestep)
1339 		rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
1340 
1341        /*
1342         * Any change of EFLAGS.VM is accompanied by a reload of SS
1343         * (caused by either a task switch or an inter-privilege IRET),
1344         * so we do not need to update the CPL here.
1345         */
1346 	to_svm(vcpu)->vmcb->save.rflags = rflags;
1347 }
1348 
1349 static bool svm_get_if_flag(struct kvm_vcpu *vcpu)
1350 {
1351 	struct vmcb *vmcb = to_svm(vcpu)->vmcb;
1352 
1353 	return sev_es_guest(vcpu->kvm)
1354 		? vmcb->control.int_state & SVM_GUEST_INTERRUPT_MASK
1355 		: kvm_get_rflags(vcpu) & X86_EFLAGS_IF;
1356 }
1357 
1358 static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1359 {
1360 	kvm_register_mark_available(vcpu, reg);
1361 
1362 	switch (reg) {
1363 	case VCPU_EXREG_PDPTR:
1364 		/*
1365 		 * When !npt_enabled, mmu->pdptrs[] is already available since
1366 		 * it is always updated per SDM when moving to CRs.
1367 		 */
1368 		if (npt_enabled)
1369 			load_pdptrs(vcpu, kvm_read_cr3(vcpu));
1370 		break;
1371 	default:
1372 		KVM_BUG_ON(1, vcpu->kvm);
1373 	}
1374 }
1375 
1376 static void svm_set_vintr(struct vcpu_svm *svm)
1377 {
1378 	struct vmcb_control_area *control;
1379 
1380 	/*
1381 	 * The following fields are ignored when AVIC is enabled
1382 	 */
1383 	WARN_ON(kvm_apicv_activated(svm->vcpu.kvm));
1384 
1385 	svm_set_intercept(svm, INTERCEPT_VINTR);
1386 
1387 	/*
1388 	 * This is just a dummy VINTR to actually cause a vmexit to happen.
1389 	 * Actual injection of virtual interrupts happens through EVENTINJ.
1390 	 */
1391 	control = &svm->vmcb->control;
1392 	control->int_vector = 0x0;
1393 	control->int_ctl &= ~V_INTR_PRIO_MASK;
1394 	control->int_ctl |= V_IRQ_MASK |
1395 		((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
1396 	vmcb_mark_dirty(svm->vmcb, VMCB_INTR);
1397 }
1398 
1399 static void svm_clear_vintr(struct vcpu_svm *svm)
1400 {
1401 	svm_clr_intercept(svm, INTERCEPT_VINTR);
1402 
1403 	/* Drop int_ctl fields related to VINTR injection.  */
1404 	svm->vmcb->control.int_ctl &= ~V_IRQ_INJECTION_BITS_MASK;
1405 	if (is_guest_mode(&svm->vcpu)) {
1406 		svm->vmcb01.ptr->control.int_ctl &= ~V_IRQ_INJECTION_BITS_MASK;
1407 
1408 		WARN_ON((svm->vmcb->control.int_ctl & V_TPR_MASK) !=
1409 			(svm->nested.ctl.int_ctl & V_TPR_MASK));
1410 
1411 		svm->vmcb->control.int_ctl |= svm->nested.ctl.int_ctl &
1412 			V_IRQ_INJECTION_BITS_MASK;
1413 
1414 		svm->vmcb->control.int_vector = svm->nested.ctl.int_vector;
1415 	}
1416 
1417 	vmcb_mark_dirty(svm->vmcb, VMCB_INTR);
1418 }
1419 
1420 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
1421 {
1422 	struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1423 	struct vmcb_save_area *save01 = &to_svm(vcpu)->vmcb01.ptr->save;
1424 
1425 	switch (seg) {
1426 	case VCPU_SREG_CS: return &save->cs;
1427 	case VCPU_SREG_DS: return &save->ds;
1428 	case VCPU_SREG_ES: return &save->es;
1429 	case VCPU_SREG_FS: return &save01->fs;
1430 	case VCPU_SREG_GS: return &save01->gs;
1431 	case VCPU_SREG_SS: return &save->ss;
1432 	case VCPU_SREG_TR: return &save01->tr;
1433 	case VCPU_SREG_LDTR: return &save01->ldtr;
1434 	}
1435 	BUG();
1436 	return NULL;
1437 }
1438 
1439 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1440 {
1441 	struct vmcb_seg *s = svm_seg(vcpu, seg);
1442 
1443 	return s->base;
1444 }
1445 
1446 static void svm_get_segment(struct kvm_vcpu *vcpu,
1447 			    struct kvm_segment *var, int seg)
1448 {
1449 	struct vmcb_seg *s = svm_seg(vcpu, seg);
1450 
1451 	var->base = s->base;
1452 	var->limit = s->limit;
1453 	var->selector = s->selector;
1454 	var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
1455 	var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
1456 	var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
1457 	var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
1458 	var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
1459 	var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
1460 	var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
1461 
1462 	/*
1463 	 * AMD CPUs circa 2014 track the G bit for all segments except CS.
1464 	 * However, the SVM spec states that the G bit is not observed by the
1465 	 * CPU, and some VMware virtual CPUs drop the G bit for all segments.
1466 	 * So let's synthesize a legal G bit for all segments, this helps
1467 	 * running KVM nested. It also helps cross-vendor migration, because
1468 	 * Intel's vmentry has a check on the 'G' bit.
1469 	 */
1470 	var->g = s->limit > 0xfffff;
1471 
1472 	/*
1473 	 * AMD's VMCB does not have an explicit unusable field, so emulate it
1474 	 * for cross vendor migration purposes by "not present"
1475 	 */
1476 	var->unusable = !var->present;
1477 
1478 	switch (seg) {
1479 	case VCPU_SREG_TR:
1480 		/*
1481 		 * Work around a bug where the busy flag in the tr selector
1482 		 * isn't exposed
1483 		 */
1484 		var->type |= 0x2;
1485 		break;
1486 	case VCPU_SREG_DS:
1487 	case VCPU_SREG_ES:
1488 	case VCPU_SREG_FS:
1489 	case VCPU_SREG_GS:
1490 		/*
1491 		 * The accessed bit must always be set in the segment
1492 		 * descriptor cache, although it can be cleared in the
1493 		 * descriptor, the cached bit always remains at 1. Since
1494 		 * Intel has a check on this, set it here to support
1495 		 * cross-vendor migration.
1496 		 */
1497 		if (!var->unusable)
1498 			var->type |= 0x1;
1499 		break;
1500 	case VCPU_SREG_SS:
1501 		/*
1502 		 * On AMD CPUs sometimes the DB bit in the segment
1503 		 * descriptor is left as 1, although the whole segment has
1504 		 * been made unusable. Clear it here to pass an Intel VMX
1505 		 * entry check when cross vendor migrating.
1506 		 */
1507 		if (var->unusable)
1508 			var->db = 0;
1509 		/* This is symmetric with svm_set_segment() */
1510 		var->dpl = to_svm(vcpu)->vmcb->save.cpl;
1511 		break;
1512 	}
1513 }
1514 
1515 static int svm_get_cpl(struct kvm_vcpu *vcpu)
1516 {
1517 	struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1518 
1519 	return save->cpl;
1520 }
1521 
1522 static void svm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
1523 {
1524 	struct kvm_segment cs;
1525 
1526 	svm_get_segment(vcpu, &cs, VCPU_SREG_CS);
1527 	*db = cs.db;
1528 	*l = cs.l;
1529 }
1530 
1531 static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1532 {
1533 	struct vcpu_svm *svm = to_svm(vcpu);
1534 
1535 	dt->size = svm->vmcb->save.idtr.limit;
1536 	dt->address = svm->vmcb->save.idtr.base;
1537 }
1538 
1539 static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1540 {
1541 	struct vcpu_svm *svm = to_svm(vcpu);
1542 
1543 	svm->vmcb->save.idtr.limit = dt->size;
1544 	svm->vmcb->save.idtr.base = dt->address ;
1545 	vmcb_mark_dirty(svm->vmcb, VMCB_DT);
1546 }
1547 
1548 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1549 {
1550 	struct vcpu_svm *svm = to_svm(vcpu);
1551 
1552 	dt->size = svm->vmcb->save.gdtr.limit;
1553 	dt->address = svm->vmcb->save.gdtr.base;
1554 }
1555 
1556 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1557 {
1558 	struct vcpu_svm *svm = to_svm(vcpu);
1559 
1560 	svm->vmcb->save.gdtr.limit = dt->size;
1561 	svm->vmcb->save.gdtr.base = dt->address ;
1562 	vmcb_mark_dirty(svm->vmcb, VMCB_DT);
1563 }
1564 
1565 static void sev_post_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1566 {
1567 	struct vcpu_svm *svm = to_svm(vcpu);
1568 
1569 	/*
1570 	 * For guests that don't set guest_state_protected, the cr3 update is
1571 	 * handled via kvm_mmu_load() while entering the guest. For guests
1572 	 * that do (SEV-ES/SEV-SNP), the cr3 update needs to be written to
1573 	 * VMCB save area now, since the save area will become the initial
1574 	 * contents of the VMSA, and future VMCB save area updates won't be
1575 	 * seen.
1576 	 */
1577 	if (sev_es_guest(vcpu->kvm)) {
1578 		svm->vmcb->save.cr3 = cr3;
1579 		vmcb_mark_dirty(svm->vmcb, VMCB_CR);
1580 	}
1581 }
1582 
1583 void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1584 {
1585 	struct vcpu_svm *svm = to_svm(vcpu);
1586 	u64 hcr0 = cr0;
1587 	bool old_paging = is_paging(vcpu);
1588 
1589 #ifdef CONFIG_X86_64
1590 	if (vcpu->arch.efer & EFER_LME && !vcpu->arch.guest_state_protected) {
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 		if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
1597 			vcpu->arch.efer &= ~EFER_LMA;
1598 			svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
1599 		}
1600 	}
1601 #endif
1602 	vcpu->arch.cr0 = cr0;
1603 
1604 	if (!npt_enabled) {
1605 		hcr0 |= X86_CR0_PG | X86_CR0_WP;
1606 		if (old_paging != is_paging(vcpu))
1607 			svm_set_cr4(vcpu, kvm_read_cr4(vcpu));
1608 	}
1609 
1610 	/*
1611 	 * re-enable caching here because the QEMU bios
1612 	 * does not do it - this results in some delay at
1613 	 * reboot
1614 	 */
1615 	if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
1616 		hcr0 &= ~(X86_CR0_CD | X86_CR0_NW);
1617 
1618 	svm->vmcb->save.cr0 = hcr0;
1619 	vmcb_mark_dirty(svm->vmcb, VMCB_CR);
1620 
1621 	/*
1622 	 * SEV-ES guests must always keep the CR intercepts cleared. CR
1623 	 * tracking is done using the CR write traps.
1624 	 */
1625 	if (sev_es_guest(vcpu->kvm))
1626 		return;
1627 
1628 	if (hcr0 == cr0) {
1629 		/* Selective CR0 write remains on.  */
1630 		svm_clr_intercept(svm, INTERCEPT_CR0_READ);
1631 		svm_clr_intercept(svm, INTERCEPT_CR0_WRITE);
1632 	} else {
1633 		svm_set_intercept(svm, INTERCEPT_CR0_READ);
1634 		svm_set_intercept(svm, INTERCEPT_CR0_WRITE);
1635 	}
1636 }
1637 
1638 static bool svm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1639 {
1640 	return true;
1641 }
1642 
1643 void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1644 {
1645 	unsigned long host_cr4_mce = cr4_read_shadow() & X86_CR4_MCE;
1646 	unsigned long old_cr4 = vcpu->arch.cr4;
1647 
1648 	if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
1649 		svm_flush_tlb_current(vcpu);
1650 
1651 	vcpu->arch.cr4 = cr4;
1652 	if (!npt_enabled) {
1653 		cr4 |= X86_CR4_PAE;
1654 
1655 		if (!is_paging(vcpu))
1656 			cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
1657 	}
1658 	cr4 |= host_cr4_mce;
1659 	to_svm(vcpu)->vmcb->save.cr4 = cr4;
1660 	vmcb_mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
1661 
1662 	if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
1663 		kvm_update_cpuid_runtime(vcpu);
1664 }
1665 
1666 static void svm_set_segment(struct kvm_vcpu *vcpu,
1667 			    struct kvm_segment *var, int seg)
1668 {
1669 	struct vcpu_svm *svm = to_svm(vcpu);
1670 	struct vmcb_seg *s = svm_seg(vcpu, seg);
1671 
1672 	s->base = var->base;
1673 	s->limit = var->limit;
1674 	s->selector = var->selector;
1675 	s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
1676 	s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
1677 	s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
1678 	s->attrib |= ((var->present & 1) && !var->unusable) << SVM_SELECTOR_P_SHIFT;
1679 	s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
1680 	s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
1681 	s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
1682 	s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
1683 
1684 	/*
1685 	 * This is always accurate, except if SYSRET returned to a segment
1686 	 * with SS.DPL != 3.  Intel does not have this quirk, and always
1687 	 * forces SS.DPL to 3 on sysret, so we ignore that case; fixing it
1688 	 * would entail passing the CPL to userspace and back.
1689 	 */
1690 	if (seg == VCPU_SREG_SS)
1691 		/* This is symmetric with svm_get_segment() */
1692 		svm->vmcb->save.cpl = (var->dpl & 3);
1693 
1694 	vmcb_mark_dirty(svm->vmcb, VMCB_SEG);
1695 }
1696 
1697 static void svm_update_exception_bitmap(struct kvm_vcpu *vcpu)
1698 {
1699 	struct vcpu_svm *svm = to_svm(vcpu);
1700 
1701 	clr_exception_intercept(svm, BP_VECTOR);
1702 
1703 	if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
1704 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
1705 			set_exception_intercept(svm, BP_VECTOR);
1706 	}
1707 }
1708 
1709 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd)
1710 {
1711 	if (sd->next_asid > sd->max_asid) {
1712 		++sd->asid_generation;
1713 		sd->next_asid = sd->min_asid;
1714 		svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
1715 		vmcb_mark_dirty(svm->vmcb, VMCB_ASID);
1716 	}
1717 
1718 	svm->current_vmcb->asid_generation = sd->asid_generation;
1719 	svm->asid = sd->next_asid++;
1720 }
1721 
1722 static void svm_set_dr6(struct vcpu_svm *svm, unsigned long value)
1723 {
1724 	struct vmcb *vmcb = svm->vmcb;
1725 
1726 	if (svm->vcpu.arch.guest_state_protected)
1727 		return;
1728 
1729 	if (unlikely(value != vmcb->save.dr6)) {
1730 		vmcb->save.dr6 = value;
1731 		vmcb_mark_dirty(vmcb, VMCB_DR);
1732 	}
1733 }
1734 
1735 static void svm_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
1736 {
1737 	struct vcpu_svm *svm = to_svm(vcpu);
1738 
1739 	if (vcpu->arch.guest_state_protected)
1740 		return;
1741 
1742 	get_debugreg(vcpu->arch.db[0], 0);
1743 	get_debugreg(vcpu->arch.db[1], 1);
1744 	get_debugreg(vcpu->arch.db[2], 2);
1745 	get_debugreg(vcpu->arch.db[3], 3);
1746 	/*
1747 	 * We cannot reset svm->vmcb->save.dr6 to DR6_ACTIVE_LOW here,
1748 	 * because db_interception might need it.  We can do it before vmentry.
1749 	 */
1750 	vcpu->arch.dr6 = svm->vmcb->save.dr6;
1751 	vcpu->arch.dr7 = svm->vmcb->save.dr7;
1752 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
1753 	set_dr_intercepts(svm);
1754 }
1755 
1756 static void svm_set_dr7(struct kvm_vcpu *vcpu, unsigned long value)
1757 {
1758 	struct vcpu_svm *svm = to_svm(vcpu);
1759 
1760 	if (vcpu->arch.guest_state_protected)
1761 		return;
1762 
1763 	svm->vmcb->save.dr7 = value;
1764 	vmcb_mark_dirty(svm->vmcb, VMCB_DR);
1765 }
1766 
1767 static int pf_interception(struct kvm_vcpu *vcpu)
1768 {
1769 	struct vcpu_svm *svm = to_svm(vcpu);
1770 
1771 	u64 fault_address = svm->vmcb->control.exit_info_2;
1772 	u64 error_code = svm->vmcb->control.exit_info_1;
1773 
1774 	return kvm_handle_page_fault(vcpu, error_code, fault_address,
1775 			static_cpu_has(X86_FEATURE_DECODEASSISTS) ?
1776 			svm->vmcb->control.insn_bytes : NULL,
1777 			svm->vmcb->control.insn_len);
1778 }
1779 
1780 static int npf_interception(struct kvm_vcpu *vcpu)
1781 {
1782 	struct vcpu_svm *svm = to_svm(vcpu);
1783 
1784 	u64 fault_address = svm->vmcb->control.exit_info_2;
1785 	u64 error_code = svm->vmcb->control.exit_info_1;
1786 
1787 	trace_kvm_page_fault(fault_address, error_code);
1788 	return kvm_mmu_page_fault(vcpu, fault_address, error_code,
1789 			static_cpu_has(X86_FEATURE_DECODEASSISTS) ?
1790 			svm->vmcb->control.insn_bytes : NULL,
1791 			svm->vmcb->control.insn_len);
1792 }
1793 
1794 static int db_interception(struct kvm_vcpu *vcpu)
1795 {
1796 	struct kvm_run *kvm_run = vcpu->run;
1797 	struct vcpu_svm *svm = to_svm(vcpu);
1798 
1799 	if (!(vcpu->guest_debug &
1800 	      (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
1801 		!svm->nmi_singlestep) {
1802 		u32 payload = svm->vmcb->save.dr6 ^ DR6_ACTIVE_LOW;
1803 		kvm_queue_exception_p(vcpu, DB_VECTOR, payload);
1804 		return 1;
1805 	}
1806 
1807 	if (svm->nmi_singlestep) {
1808 		disable_nmi_singlestep(svm);
1809 		/* Make sure we check for pending NMIs upon entry */
1810 		kvm_make_request(KVM_REQ_EVENT, vcpu);
1811 	}
1812 
1813 	if (vcpu->guest_debug &
1814 	    (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) {
1815 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
1816 		kvm_run->debug.arch.dr6 = svm->vmcb->save.dr6;
1817 		kvm_run->debug.arch.dr7 = svm->vmcb->save.dr7;
1818 		kvm_run->debug.arch.pc =
1819 			svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1820 		kvm_run->debug.arch.exception = DB_VECTOR;
1821 		return 0;
1822 	}
1823 
1824 	return 1;
1825 }
1826 
1827 static int bp_interception(struct kvm_vcpu *vcpu)
1828 {
1829 	struct vcpu_svm *svm = to_svm(vcpu);
1830 	struct kvm_run *kvm_run = vcpu->run;
1831 
1832 	kvm_run->exit_reason = KVM_EXIT_DEBUG;
1833 	kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1834 	kvm_run->debug.arch.exception = BP_VECTOR;
1835 	return 0;
1836 }
1837 
1838 static int ud_interception(struct kvm_vcpu *vcpu)
1839 {
1840 	return handle_ud(vcpu);
1841 }
1842 
1843 static int ac_interception(struct kvm_vcpu *vcpu)
1844 {
1845 	kvm_queue_exception_e(vcpu, AC_VECTOR, 0);
1846 	return 1;
1847 }
1848 
1849 static bool is_erratum_383(void)
1850 {
1851 	int err, i;
1852 	u64 value;
1853 
1854 	if (!erratum_383_found)
1855 		return false;
1856 
1857 	value = native_read_msr_safe(MSR_IA32_MC0_STATUS, &err);
1858 	if (err)
1859 		return false;
1860 
1861 	/* Bit 62 may or may not be set for this mce */
1862 	value &= ~(1ULL << 62);
1863 
1864 	if (value != 0xb600000000010015ULL)
1865 		return false;
1866 
1867 	/* Clear MCi_STATUS registers */
1868 	for (i = 0; i < 6; ++i)
1869 		native_write_msr_safe(MSR_IA32_MCx_STATUS(i), 0, 0);
1870 
1871 	value = native_read_msr_safe(MSR_IA32_MCG_STATUS, &err);
1872 	if (!err) {
1873 		u32 low, high;
1874 
1875 		value &= ~(1ULL << 2);
1876 		low    = lower_32_bits(value);
1877 		high   = upper_32_bits(value);
1878 
1879 		native_write_msr_safe(MSR_IA32_MCG_STATUS, low, high);
1880 	}
1881 
1882 	/* Flush tlb to evict multi-match entries */
1883 	__flush_tlb_all();
1884 
1885 	return true;
1886 }
1887 
1888 static void svm_handle_mce(struct kvm_vcpu *vcpu)
1889 {
1890 	if (is_erratum_383()) {
1891 		/*
1892 		 * Erratum 383 triggered. Guest state is corrupt so kill the
1893 		 * guest.
1894 		 */
1895 		pr_err("KVM: Guest triggered AMD Erratum 383\n");
1896 
1897 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
1898 
1899 		return;
1900 	}
1901 
1902 	/*
1903 	 * On an #MC intercept the MCE handler is not called automatically in
1904 	 * the host. So do it by hand here.
1905 	 */
1906 	kvm_machine_check();
1907 }
1908 
1909 static int mc_interception(struct kvm_vcpu *vcpu)
1910 {
1911 	return 1;
1912 }
1913 
1914 static int shutdown_interception(struct kvm_vcpu *vcpu)
1915 {
1916 	struct kvm_run *kvm_run = vcpu->run;
1917 	struct vcpu_svm *svm = to_svm(vcpu);
1918 
1919 	/*
1920 	 * The VM save area has already been encrypted so it
1921 	 * cannot be reinitialized - just terminate.
1922 	 */
1923 	if (sev_es_guest(vcpu->kvm))
1924 		return -EINVAL;
1925 
1926 	/*
1927 	 * VMCB is undefined after a SHUTDOWN intercept.  INIT the vCPU to put
1928 	 * the VMCB in a known good state.  Unfortuately, KVM doesn't have
1929 	 * KVM_MP_STATE_SHUTDOWN and can't add it without potentially breaking
1930 	 * userspace.  At a platform view, INIT is acceptable behavior as
1931 	 * there exist bare metal platforms that automatically INIT the CPU
1932 	 * in response to shutdown.
1933 	 */
1934 	clear_page(svm->vmcb);
1935 	kvm_vcpu_reset(vcpu, true);
1936 
1937 	kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1938 	return 0;
1939 }
1940 
1941 static int io_interception(struct kvm_vcpu *vcpu)
1942 {
1943 	struct vcpu_svm *svm = to_svm(vcpu);
1944 	u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
1945 	int size, in, string;
1946 	unsigned port;
1947 
1948 	++vcpu->stat.io_exits;
1949 	string = (io_info & SVM_IOIO_STR_MASK) != 0;
1950 	in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
1951 	port = io_info >> 16;
1952 	size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
1953 
1954 	if (string) {
1955 		if (sev_es_guest(vcpu->kvm))
1956 			return sev_es_string_io(svm, size, port, in);
1957 		else
1958 			return kvm_emulate_instruction(vcpu, 0);
1959 	}
1960 
1961 	svm->next_rip = svm->vmcb->control.exit_info_2;
1962 
1963 	return kvm_fast_pio(vcpu, size, port, in);
1964 }
1965 
1966 static int nmi_interception(struct kvm_vcpu *vcpu)
1967 {
1968 	return 1;
1969 }
1970 
1971 static int smi_interception(struct kvm_vcpu *vcpu)
1972 {
1973 	return 1;
1974 }
1975 
1976 static int intr_interception(struct kvm_vcpu *vcpu)
1977 {
1978 	++vcpu->stat.irq_exits;
1979 	return 1;
1980 }
1981 
1982 static int vmload_vmsave_interception(struct kvm_vcpu *vcpu, bool vmload)
1983 {
1984 	struct vcpu_svm *svm = to_svm(vcpu);
1985 	struct vmcb *vmcb12;
1986 	struct kvm_host_map map;
1987 	int ret;
1988 
1989 	if (nested_svm_check_permissions(vcpu))
1990 		return 1;
1991 
1992 	ret = kvm_vcpu_map(vcpu, gpa_to_gfn(svm->vmcb->save.rax), &map);
1993 	if (ret) {
1994 		if (ret == -EINVAL)
1995 			kvm_inject_gp(vcpu, 0);
1996 		return 1;
1997 	}
1998 
1999 	vmcb12 = map.hva;
2000 
2001 	ret = kvm_skip_emulated_instruction(vcpu);
2002 
2003 	if (vmload) {
2004 		svm_copy_vmloadsave_state(svm->vmcb, vmcb12);
2005 		svm->sysenter_eip_hi = 0;
2006 		svm->sysenter_esp_hi = 0;
2007 	} else {
2008 		svm_copy_vmloadsave_state(vmcb12, svm->vmcb);
2009 	}
2010 
2011 	kvm_vcpu_unmap(vcpu, &map, true);
2012 
2013 	return ret;
2014 }
2015 
2016 static int vmload_interception(struct kvm_vcpu *vcpu)
2017 {
2018 	return vmload_vmsave_interception(vcpu, true);
2019 }
2020 
2021 static int vmsave_interception(struct kvm_vcpu *vcpu)
2022 {
2023 	return vmload_vmsave_interception(vcpu, false);
2024 }
2025 
2026 static int vmrun_interception(struct kvm_vcpu *vcpu)
2027 {
2028 	if (nested_svm_check_permissions(vcpu))
2029 		return 1;
2030 
2031 	return nested_svm_vmrun(vcpu);
2032 }
2033 
2034 enum {
2035 	NONE_SVM_INSTR,
2036 	SVM_INSTR_VMRUN,
2037 	SVM_INSTR_VMLOAD,
2038 	SVM_INSTR_VMSAVE,
2039 };
2040 
2041 /* Return NONE_SVM_INSTR if not SVM instrs, otherwise return decode result */
2042 static int svm_instr_opcode(struct kvm_vcpu *vcpu)
2043 {
2044 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
2045 
2046 	if (ctxt->b != 0x1 || ctxt->opcode_len != 2)
2047 		return NONE_SVM_INSTR;
2048 
2049 	switch (ctxt->modrm) {
2050 	case 0xd8: /* VMRUN */
2051 		return SVM_INSTR_VMRUN;
2052 	case 0xda: /* VMLOAD */
2053 		return SVM_INSTR_VMLOAD;
2054 	case 0xdb: /* VMSAVE */
2055 		return SVM_INSTR_VMSAVE;
2056 	default:
2057 		break;
2058 	}
2059 
2060 	return NONE_SVM_INSTR;
2061 }
2062 
2063 static int emulate_svm_instr(struct kvm_vcpu *vcpu, int opcode)
2064 {
2065 	const int guest_mode_exit_codes[] = {
2066 		[SVM_INSTR_VMRUN] = SVM_EXIT_VMRUN,
2067 		[SVM_INSTR_VMLOAD] = SVM_EXIT_VMLOAD,
2068 		[SVM_INSTR_VMSAVE] = SVM_EXIT_VMSAVE,
2069 	};
2070 	int (*const svm_instr_handlers[])(struct kvm_vcpu *vcpu) = {
2071 		[SVM_INSTR_VMRUN] = vmrun_interception,
2072 		[SVM_INSTR_VMLOAD] = vmload_interception,
2073 		[SVM_INSTR_VMSAVE] = vmsave_interception,
2074 	};
2075 	struct vcpu_svm *svm = to_svm(vcpu);
2076 	int ret;
2077 
2078 	if (is_guest_mode(vcpu)) {
2079 		/* Returns '1' or -errno on failure, '0' on success. */
2080 		ret = nested_svm_simple_vmexit(svm, guest_mode_exit_codes[opcode]);
2081 		if (ret)
2082 			return ret;
2083 		return 1;
2084 	}
2085 	return svm_instr_handlers[opcode](vcpu);
2086 }
2087 
2088 /*
2089  * #GP handling code. Note that #GP can be triggered under the following two
2090  * cases:
2091  *   1) SVM VM-related instructions (VMRUN/VMSAVE/VMLOAD) that trigger #GP on
2092  *      some AMD CPUs when EAX of these instructions are in the reserved memory
2093  *      regions (e.g. SMM memory on host).
2094  *   2) VMware backdoor
2095  */
2096 static int gp_interception(struct kvm_vcpu *vcpu)
2097 {
2098 	struct vcpu_svm *svm = to_svm(vcpu);
2099 	u32 error_code = svm->vmcb->control.exit_info_1;
2100 	int opcode;
2101 
2102 	/* Both #GP cases have zero error_code */
2103 	if (error_code)
2104 		goto reinject;
2105 
2106 	/* Decode the instruction for usage later */
2107 	if (x86_decode_emulated_instruction(vcpu, 0, NULL, 0) != EMULATION_OK)
2108 		goto reinject;
2109 
2110 	opcode = svm_instr_opcode(vcpu);
2111 
2112 	if (opcode == NONE_SVM_INSTR) {
2113 		if (!enable_vmware_backdoor)
2114 			goto reinject;
2115 
2116 		/*
2117 		 * VMware backdoor emulation on #GP interception only handles
2118 		 * IN{S}, OUT{S}, and RDPMC.
2119 		 */
2120 		if (!is_guest_mode(vcpu))
2121 			return kvm_emulate_instruction(vcpu,
2122 				EMULTYPE_VMWARE_GP | EMULTYPE_NO_DECODE);
2123 	} else {
2124 		/* All SVM instructions expect page aligned RAX */
2125 		if (svm->vmcb->save.rax & ~PAGE_MASK)
2126 			goto reinject;
2127 
2128 		return emulate_svm_instr(vcpu, opcode);
2129 	}
2130 
2131 reinject:
2132 	kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
2133 	return 1;
2134 }
2135 
2136 void svm_set_gif(struct vcpu_svm *svm, bool value)
2137 {
2138 	if (value) {
2139 		/*
2140 		 * If VGIF is enabled, the STGI intercept is only added to
2141 		 * detect the opening of the SMI/NMI window; remove it now.
2142 		 * Likewise, clear the VINTR intercept, we will set it
2143 		 * again while processing KVM_REQ_EVENT if needed.
2144 		 */
2145 		if (vgif_enabled(svm))
2146 			svm_clr_intercept(svm, INTERCEPT_STGI);
2147 		if (svm_is_intercept(svm, INTERCEPT_VINTR))
2148 			svm_clear_vintr(svm);
2149 
2150 		enable_gif(svm);
2151 		if (svm->vcpu.arch.smi_pending ||
2152 		    svm->vcpu.arch.nmi_pending ||
2153 		    kvm_cpu_has_injectable_intr(&svm->vcpu))
2154 			kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
2155 	} else {
2156 		disable_gif(svm);
2157 
2158 		/*
2159 		 * After a CLGI no interrupts should come.  But if vGIF is
2160 		 * in use, we still rely on the VINTR intercept (rather than
2161 		 * STGI) to detect an open interrupt window.
2162 		*/
2163 		if (!vgif_enabled(svm))
2164 			svm_clear_vintr(svm);
2165 	}
2166 }
2167 
2168 static int stgi_interception(struct kvm_vcpu *vcpu)
2169 {
2170 	int ret;
2171 
2172 	if (nested_svm_check_permissions(vcpu))
2173 		return 1;
2174 
2175 	ret = kvm_skip_emulated_instruction(vcpu);
2176 	svm_set_gif(to_svm(vcpu), true);
2177 	return ret;
2178 }
2179 
2180 static int clgi_interception(struct kvm_vcpu *vcpu)
2181 {
2182 	int ret;
2183 
2184 	if (nested_svm_check_permissions(vcpu))
2185 		return 1;
2186 
2187 	ret = kvm_skip_emulated_instruction(vcpu);
2188 	svm_set_gif(to_svm(vcpu), false);
2189 	return ret;
2190 }
2191 
2192 static int invlpga_interception(struct kvm_vcpu *vcpu)
2193 {
2194 	gva_t gva = kvm_rax_read(vcpu);
2195 	u32 asid = kvm_rcx_read(vcpu);
2196 
2197 	/* FIXME: Handle an address size prefix. */
2198 	if (!is_long_mode(vcpu))
2199 		gva = (u32)gva;
2200 
2201 	trace_kvm_invlpga(to_svm(vcpu)->vmcb->save.rip, asid, gva);
2202 
2203 	/* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
2204 	kvm_mmu_invlpg(vcpu, gva);
2205 
2206 	return kvm_skip_emulated_instruction(vcpu);
2207 }
2208 
2209 static int skinit_interception(struct kvm_vcpu *vcpu)
2210 {
2211 	trace_kvm_skinit(to_svm(vcpu)->vmcb->save.rip, kvm_rax_read(vcpu));
2212 
2213 	kvm_queue_exception(vcpu, UD_VECTOR);
2214 	return 1;
2215 }
2216 
2217 static int task_switch_interception(struct kvm_vcpu *vcpu)
2218 {
2219 	struct vcpu_svm *svm = to_svm(vcpu);
2220 	u16 tss_selector;
2221 	int reason;
2222 	int int_type = svm->vmcb->control.exit_int_info &
2223 		SVM_EXITINTINFO_TYPE_MASK;
2224 	int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
2225 	uint32_t type =
2226 		svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
2227 	uint32_t idt_v =
2228 		svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
2229 	bool has_error_code = false;
2230 	u32 error_code = 0;
2231 
2232 	tss_selector = (u16)svm->vmcb->control.exit_info_1;
2233 
2234 	if (svm->vmcb->control.exit_info_2 &
2235 	    (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
2236 		reason = TASK_SWITCH_IRET;
2237 	else if (svm->vmcb->control.exit_info_2 &
2238 		 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
2239 		reason = TASK_SWITCH_JMP;
2240 	else if (idt_v)
2241 		reason = TASK_SWITCH_GATE;
2242 	else
2243 		reason = TASK_SWITCH_CALL;
2244 
2245 	if (reason == TASK_SWITCH_GATE) {
2246 		switch (type) {
2247 		case SVM_EXITINTINFO_TYPE_NMI:
2248 			vcpu->arch.nmi_injected = false;
2249 			break;
2250 		case SVM_EXITINTINFO_TYPE_EXEPT:
2251 			if (svm->vmcb->control.exit_info_2 &
2252 			    (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE)) {
2253 				has_error_code = true;
2254 				error_code =
2255 					(u32)svm->vmcb->control.exit_info_2;
2256 			}
2257 			kvm_clear_exception_queue(vcpu);
2258 			break;
2259 		case SVM_EXITINTINFO_TYPE_INTR:
2260 			kvm_clear_interrupt_queue(vcpu);
2261 			break;
2262 		default:
2263 			break;
2264 		}
2265 	}
2266 
2267 	if (reason != TASK_SWITCH_GATE ||
2268 	    int_type == SVM_EXITINTINFO_TYPE_SOFT ||
2269 	    (int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
2270 	     (int_vec == OF_VECTOR || int_vec == BP_VECTOR))) {
2271 		if (!svm_skip_emulated_instruction(vcpu))
2272 			return 0;
2273 	}
2274 
2275 	if (int_type != SVM_EXITINTINFO_TYPE_SOFT)
2276 		int_vec = -1;
2277 
2278 	return kvm_task_switch(vcpu, tss_selector, int_vec, reason,
2279 			       has_error_code, error_code);
2280 }
2281 
2282 static int iret_interception(struct kvm_vcpu *vcpu)
2283 {
2284 	struct vcpu_svm *svm = to_svm(vcpu);
2285 
2286 	++vcpu->stat.nmi_window_exits;
2287 	vcpu->arch.hflags |= HF_IRET_MASK;
2288 	if (!sev_es_guest(vcpu->kvm)) {
2289 		svm_clr_intercept(svm, INTERCEPT_IRET);
2290 		svm->nmi_iret_rip = kvm_rip_read(vcpu);
2291 	}
2292 	kvm_make_request(KVM_REQ_EVENT, vcpu);
2293 	return 1;
2294 }
2295 
2296 static int invlpg_interception(struct kvm_vcpu *vcpu)
2297 {
2298 	if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
2299 		return kvm_emulate_instruction(vcpu, 0);
2300 
2301 	kvm_mmu_invlpg(vcpu, to_svm(vcpu)->vmcb->control.exit_info_1);
2302 	return kvm_skip_emulated_instruction(vcpu);
2303 }
2304 
2305 static int emulate_on_interception(struct kvm_vcpu *vcpu)
2306 {
2307 	return kvm_emulate_instruction(vcpu, 0);
2308 }
2309 
2310 static int rsm_interception(struct kvm_vcpu *vcpu)
2311 {
2312 	return kvm_emulate_instruction_from_buffer(vcpu, rsm_ins_bytes, 2);
2313 }
2314 
2315 static bool check_selective_cr0_intercepted(struct kvm_vcpu *vcpu,
2316 					    unsigned long val)
2317 {
2318 	struct vcpu_svm *svm = to_svm(vcpu);
2319 	unsigned long cr0 = vcpu->arch.cr0;
2320 	bool ret = false;
2321 
2322 	if (!is_guest_mode(vcpu) ||
2323 	    (!(vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_SELECTIVE_CR0))))
2324 		return false;
2325 
2326 	cr0 &= ~SVM_CR0_SELECTIVE_MASK;
2327 	val &= ~SVM_CR0_SELECTIVE_MASK;
2328 
2329 	if (cr0 ^ val) {
2330 		svm->vmcb->control.exit_code = SVM_EXIT_CR0_SEL_WRITE;
2331 		ret = (nested_svm_exit_handled(svm) == NESTED_EXIT_DONE);
2332 	}
2333 
2334 	return ret;
2335 }
2336 
2337 #define CR_VALID (1ULL << 63)
2338 
2339 static int cr_interception(struct kvm_vcpu *vcpu)
2340 {
2341 	struct vcpu_svm *svm = to_svm(vcpu);
2342 	int reg, cr;
2343 	unsigned long val;
2344 	int err;
2345 
2346 	if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
2347 		return emulate_on_interception(vcpu);
2348 
2349 	if (unlikely((svm->vmcb->control.exit_info_1 & CR_VALID) == 0))
2350 		return emulate_on_interception(vcpu);
2351 
2352 	reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
2353 	if (svm->vmcb->control.exit_code == SVM_EXIT_CR0_SEL_WRITE)
2354 		cr = SVM_EXIT_WRITE_CR0 - SVM_EXIT_READ_CR0;
2355 	else
2356 		cr = svm->vmcb->control.exit_code - SVM_EXIT_READ_CR0;
2357 
2358 	err = 0;
2359 	if (cr >= 16) { /* mov to cr */
2360 		cr -= 16;
2361 		val = kvm_register_read(vcpu, reg);
2362 		trace_kvm_cr_write(cr, val);
2363 		switch (cr) {
2364 		case 0:
2365 			if (!check_selective_cr0_intercepted(vcpu, val))
2366 				err = kvm_set_cr0(vcpu, val);
2367 			else
2368 				return 1;
2369 
2370 			break;
2371 		case 3:
2372 			err = kvm_set_cr3(vcpu, val);
2373 			break;
2374 		case 4:
2375 			err = kvm_set_cr4(vcpu, val);
2376 			break;
2377 		case 8:
2378 			err = kvm_set_cr8(vcpu, val);
2379 			break;
2380 		default:
2381 			WARN(1, "unhandled write to CR%d", cr);
2382 			kvm_queue_exception(vcpu, UD_VECTOR);
2383 			return 1;
2384 		}
2385 	} else { /* mov from cr */
2386 		switch (cr) {
2387 		case 0:
2388 			val = kvm_read_cr0(vcpu);
2389 			break;
2390 		case 2:
2391 			val = vcpu->arch.cr2;
2392 			break;
2393 		case 3:
2394 			val = kvm_read_cr3(vcpu);
2395 			break;
2396 		case 4:
2397 			val = kvm_read_cr4(vcpu);
2398 			break;
2399 		case 8:
2400 			val = kvm_get_cr8(vcpu);
2401 			break;
2402 		default:
2403 			WARN(1, "unhandled read from CR%d", cr);
2404 			kvm_queue_exception(vcpu, UD_VECTOR);
2405 			return 1;
2406 		}
2407 		kvm_register_write(vcpu, reg, val);
2408 		trace_kvm_cr_read(cr, val);
2409 	}
2410 	return kvm_complete_insn_gp(vcpu, err);
2411 }
2412 
2413 static int cr_trap(struct kvm_vcpu *vcpu)
2414 {
2415 	struct vcpu_svm *svm = to_svm(vcpu);
2416 	unsigned long old_value, new_value;
2417 	unsigned int cr;
2418 	int ret = 0;
2419 
2420 	new_value = (unsigned long)svm->vmcb->control.exit_info_1;
2421 
2422 	cr = svm->vmcb->control.exit_code - SVM_EXIT_CR0_WRITE_TRAP;
2423 	switch (cr) {
2424 	case 0:
2425 		old_value = kvm_read_cr0(vcpu);
2426 		svm_set_cr0(vcpu, new_value);
2427 
2428 		kvm_post_set_cr0(vcpu, old_value, new_value);
2429 		break;
2430 	case 4:
2431 		old_value = kvm_read_cr4(vcpu);
2432 		svm_set_cr4(vcpu, new_value);
2433 
2434 		kvm_post_set_cr4(vcpu, old_value, new_value);
2435 		break;
2436 	case 8:
2437 		ret = kvm_set_cr8(vcpu, new_value);
2438 		break;
2439 	default:
2440 		WARN(1, "unhandled CR%d write trap", cr);
2441 		kvm_queue_exception(vcpu, UD_VECTOR);
2442 		return 1;
2443 	}
2444 
2445 	return kvm_complete_insn_gp(vcpu, ret);
2446 }
2447 
2448 static int dr_interception(struct kvm_vcpu *vcpu)
2449 {
2450 	struct vcpu_svm *svm = to_svm(vcpu);
2451 	int reg, dr;
2452 	unsigned long val;
2453 	int err = 0;
2454 
2455 	if (vcpu->guest_debug == 0) {
2456 		/*
2457 		 * No more DR vmexits; force a reload of the debug registers
2458 		 * and reenter on this instruction.  The next vmexit will
2459 		 * retrieve the full state of the debug registers.
2460 		 */
2461 		clr_dr_intercepts(svm);
2462 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
2463 		return 1;
2464 	}
2465 
2466 	if (!boot_cpu_has(X86_FEATURE_DECODEASSISTS))
2467 		return emulate_on_interception(vcpu);
2468 
2469 	reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
2470 	dr = svm->vmcb->control.exit_code - SVM_EXIT_READ_DR0;
2471 	if (dr >= 16) { /* mov to DRn  */
2472 		dr -= 16;
2473 		val = kvm_register_read(vcpu, reg);
2474 		err = kvm_set_dr(vcpu, dr, val);
2475 	} else {
2476 		kvm_get_dr(vcpu, dr, &val);
2477 		kvm_register_write(vcpu, reg, val);
2478 	}
2479 
2480 	return kvm_complete_insn_gp(vcpu, err);
2481 }
2482 
2483 static int cr8_write_interception(struct kvm_vcpu *vcpu)
2484 {
2485 	int r;
2486 
2487 	u8 cr8_prev = kvm_get_cr8(vcpu);
2488 	/* instruction emulation calls kvm_set_cr8() */
2489 	r = cr_interception(vcpu);
2490 	if (lapic_in_kernel(vcpu))
2491 		return r;
2492 	if (cr8_prev <= kvm_get_cr8(vcpu))
2493 		return r;
2494 	vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
2495 	return 0;
2496 }
2497 
2498 static int efer_trap(struct kvm_vcpu *vcpu)
2499 {
2500 	struct msr_data msr_info;
2501 	int ret;
2502 
2503 	/*
2504 	 * Clear the EFER_SVME bit from EFER. The SVM code always sets this
2505 	 * bit in svm_set_efer(), but __kvm_valid_efer() checks it against
2506 	 * whether the guest has X86_FEATURE_SVM - this avoids a failure if
2507 	 * the guest doesn't have X86_FEATURE_SVM.
2508 	 */
2509 	msr_info.host_initiated = false;
2510 	msr_info.index = MSR_EFER;
2511 	msr_info.data = to_svm(vcpu)->vmcb->control.exit_info_1 & ~EFER_SVME;
2512 	ret = kvm_set_msr_common(vcpu, &msr_info);
2513 
2514 	return kvm_complete_insn_gp(vcpu, ret);
2515 }
2516 
2517 static int svm_get_msr_feature(struct kvm_msr_entry *msr)
2518 {
2519 	msr->data = 0;
2520 
2521 	switch (msr->index) {
2522 	case MSR_F10H_DECFG:
2523 		if (boot_cpu_has(X86_FEATURE_LFENCE_RDTSC))
2524 			msr->data |= MSR_F10H_DECFG_LFENCE_SERIALIZE;
2525 		break;
2526 	case MSR_IA32_PERF_CAPABILITIES:
2527 		return 0;
2528 	default:
2529 		return KVM_MSR_RET_INVALID;
2530 	}
2531 
2532 	return 0;
2533 }
2534 
2535 static int svm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2536 {
2537 	struct vcpu_svm *svm = to_svm(vcpu);
2538 
2539 	switch (msr_info->index) {
2540 	case MSR_AMD64_TSC_RATIO:
2541 		if (!msr_info->host_initiated && !svm->tsc_scaling_enabled)
2542 			return 1;
2543 		msr_info->data = svm->tsc_ratio_msr;
2544 		break;
2545 	case MSR_STAR:
2546 		msr_info->data = svm->vmcb01.ptr->save.star;
2547 		break;
2548 #ifdef CONFIG_X86_64
2549 	case MSR_LSTAR:
2550 		msr_info->data = svm->vmcb01.ptr->save.lstar;
2551 		break;
2552 	case MSR_CSTAR:
2553 		msr_info->data = svm->vmcb01.ptr->save.cstar;
2554 		break;
2555 	case MSR_KERNEL_GS_BASE:
2556 		msr_info->data = svm->vmcb01.ptr->save.kernel_gs_base;
2557 		break;
2558 	case MSR_SYSCALL_MASK:
2559 		msr_info->data = svm->vmcb01.ptr->save.sfmask;
2560 		break;
2561 #endif
2562 	case MSR_IA32_SYSENTER_CS:
2563 		msr_info->data = svm->vmcb01.ptr->save.sysenter_cs;
2564 		break;
2565 	case MSR_IA32_SYSENTER_EIP:
2566 		msr_info->data = (u32)svm->vmcb01.ptr->save.sysenter_eip;
2567 		if (guest_cpuid_is_intel(vcpu))
2568 			msr_info->data |= (u64)svm->sysenter_eip_hi << 32;
2569 		break;
2570 	case MSR_IA32_SYSENTER_ESP:
2571 		msr_info->data = svm->vmcb01.ptr->save.sysenter_esp;
2572 		if (guest_cpuid_is_intel(vcpu))
2573 			msr_info->data |= (u64)svm->sysenter_esp_hi << 32;
2574 		break;
2575 	case MSR_TSC_AUX:
2576 		msr_info->data = svm->tsc_aux;
2577 		break;
2578 	/*
2579 	 * Nobody will change the following 5 values in the VMCB so we can
2580 	 * safely return them on rdmsr. They will always be 0 until LBRV is
2581 	 * implemented.
2582 	 */
2583 	case MSR_IA32_DEBUGCTLMSR:
2584 		msr_info->data = svm->vmcb->save.dbgctl;
2585 		break;
2586 	case MSR_IA32_LASTBRANCHFROMIP:
2587 		msr_info->data = svm->vmcb->save.br_from;
2588 		break;
2589 	case MSR_IA32_LASTBRANCHTOIP:
2590 		msr_info->data = svm->vmcb->save.br_to;
2591 		break;
2592 	case MSR_IA32_LASTINTFROMIP:
2593 		msr_info->data = svm->vmcb->save.last_excp_from;
2594 		break;
2595 	case MSR_IA32_LASTINTTOIP:
2596 		msr_info->data = svm->vmcb->save.last_excp_to;
2597 		break;
2598 	case MSR_VM_HSAVE_PA:
2599 		msr_info->data = svm->nested.hsave_msr;
2600 		break;
2601 	case MSR_VM_CR:
2602 		msr_info->data = svm->nested.vm_cr_msr;
2603 		break;
2604 	case MSR_IA32_SPEC_CTRL:
2605 		if (!msr_info->host_initiated &&
2606 		    !guest_has_spec_ctrl_msr(vcpu))
2607 			return 1;
2608 
2609 		if (boot_cpu_has(X86_FEATURE_V_SPEC_CTRL))
2610 			msr_info->data = svm->vmcb->save.spec_ctrl;
2611 		else
2612 			msr_info->data = svm->spec_ctrl;
2613 		break;
2614 	case MSR_AMD64_VIRT_SPEC_CTRL:
2615 		if (!msr_info->host_initiated &&
2616 		    !guest_cpuid_has(vcpu, X86_FEATURE_VIRT_SSBD))
2617 			return 1;
2618 
2619 		msr_info->data = svm->virt_spec_ctrl;
2620 		break;
2621 	case MSR_F15H_IC_CFG: {
2622 
2623 		int family, model;
2624 
2625 		family = guest_cpuid_family(vcpu);
2626 		model  = guest_cpuid_model(vcpu);
2627 
2628 		if (family < 0 || model < 0)
2629 			return kvm_get_msr_common(vcpu, msr_info);
2630 
2631 		msr_info->data = 0;
2632 
2633 		if (family == 0x15 &&
2634 		    (model >= 0x2 && model < 0x20))
2635 			msr_info->data = 0x1E;
2636 		}
2637 		break;
2638 	case MSR_F10H_DECFG:
2639 		msr_info->data = svm->msr_decfg;
2640 		break;
2641 	default:
2642 		return kvm_get_msr_common(vcpu, msr_info);
2643 	}
2644 	return 0;
2645 }
2646 
2647 static int svm_complete_emulated_msr(struct kvm_vcpu *vcpu, int err)
2648 {
2649 	struct vcpu_svm *svm = to_svm(vcpu);
2650 	if (!err || !sev_es_guest(vcpu->kvm) || WARN_ON_ONCE(!svm->sev_es.ghcb))
2651 		return kvm_complete_insn_gp(vcpu, err);
2652 
2653 	ghcb_set_sw_exit_info_1(svm->sev_es.ghcb, 1);
2654 	ghcb_set_sw_exit_info_2(svm->sev_es.ghcb,
2655 				X86_TRAP_GP |
2656 				SVM_EVTINJ_TYPE_EXEPT |
2657 				SVM_EVTINJ_VALID);
2658 	return 1;
2659 }
2660 
2661 static int svm_set_vm_cr(struct kvm_vcpu *vcpu, u64 data)
2662 {
2663 	struct vcpu_svm *svm = to_svm(vcpu);
2664 	int svm_dis, chg_mask;
2665 
2666 	if (data & ~SVM_VM_CR_VALID_MASK)
2667 		return 1;
2668 
2669 	chg_mask = SVM_VM_CR_VALID_MASK;
2670 
2671 	if (svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK)
2672 		chg_mask &= ~(SVM_VM_CR_SVM_LOCK_MASK | SVM_VM_CR_SVM_DIS_MASK);
2673 
2674 	svm->nested.vm_cr_msr &= ~chg_mask;
2675 	svm->nested.vm_cr_msr |= (data & chg_mask);
2676 
2677 	svm_dis = svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK;
2678 
2679 	/* check for svm_disable while efer.svme is set */
2680 	if (svm_dis && (vcpu->arch.efer & EFER_SVME))
2681 		return 1;
2682 
2683 	return 0;
2684 }
2685 
2686 static int svm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2687 {
2688 	struct vcpu_svm *svm = to_svm(vcpu);
2689 	int r;
2690 
2691 	u32 ecx = msr->index;
2692 	u64 data = msr->data;
2693 	switch (ecx) {
2694 	case MSR_AMD64_TSC_RATIO:
2695 
2696 		if (!svm->tsc_scaling_enabled) {
2697 
2698 			if (!msr->host_initiated)
2699 				return 1;
2700 			/*
2701 			 * In case TSC scaling is not enabled, always
2702 			 * leave this MSR at the default value.
2703 			 *
2704 			 * Due to bug in qemu 6.2.0, it would try to set
2705 			 * this msr to 0 if tsc scaling is not enabled.
2706 			 * Ignore this value as well.
2707 			 */
2708 			if (data != 0 && data != svm->tsc_ratio_msr)
2709 				return 1;
2710 			break;
2711 		}
2712 
2713 		if (data & SVM_TSC_RATIO_RSVD)
2714 			return 1;
2715 
2716 		svm->tsc_ratio_msr = data;
2717 
2718 		if (svm->tsc_scaling_enabled && is_guest_mode(vcpu))
2719 			nested_svm_update_tsc_ratio_msr(vcpu);
2720 
2721 		break;
2722 	case MSR_IA32_CR_PAT:
2723 		if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
2724 			return 1;
2725 		vcpu->arch.pat = data;
2726 		svm->vmcb01.ptr->save.g_pat = data;
2727 		if (is_guest_mode(vcpu))
2728 			nested_vmcb02_compute_g_pat(svm);
2729 		vmcb_mark_dirty(svm->vmcb, VMCB_NPT);
2730 		break;
2731 	case MSR_IA32_SPEC_CTRL:
2732 		if (!msr->host_initiated &&
2733 		    !guest_has_spec_ctrl_msr(vcpu))
2734 			return 1;
2735 
2736 		if (kvm_spec_ctrl_test_value(data))
2737 			return 1;
2738 
2739 		if (boot_cpu_has(X86_FEATURE_V_SPEC_CTRL))
2740 			svm->vmcb->save.spec_ctrl = data;
2741 		else
2742 			svm->spec_ctrl = data;
2743 		if (!data)
2744 			break;
2745 
2746 		/*
2747 		 * For non-nested:
2748 		 * When it's written (to non-zero) for the first time, pass
2749 		 * it through.
2750 		 *
2751 		 * For nested:
2752 		 * The handling of the MSR bitmap for L2 guests is done in
2753 		 * nested_svm_vmrun_msrpm.
2754 		 * We update the L1 MSR bit as well since it will end up
2755 		 * touching the MSR anyway now.
2756 		 */
2757 		set_msr_interception(vcpu, svm->msrpm, MSR_IA32_SPEC_CTRL, 1, 1);
2758 		break;
2759 	case MSR_IA32_PRED_CMD:
2760 		if (!msr->host_initiated &&
2761 		    !guest_has_pred_cmd_msr(vcpu))
2762 			return 1;
2763 
2764 		if (data & ~PRED_CMD_IBPB)
2765 			return 1;
2766 		if (!boot_cpu_has(X86_FEATURE_IBPB))
2767 			return 1;
2768 		if (!data)
2769 			break;
2770 
2771 		wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
2772 		set_msr_interception(vcpu, svm->msrpm, MSR_IA32_PRED_CMD, 0, 1);
2773 		break;
2774 	case MSR_AMD64_VIRT_SPEC_CTRL:
2775 		if (!msr->host_initiated &&
2776 		    !guest_cpuid_has(vcpu, X86_FEATURE_VIRT_SSBD))
2777 			return 1;
2778 
2779 		if (data & ~SPEC_CTRL_SSBD)
2780 			return 1;
2781 
2782 		svm->virt_spec_ctrl = data;
2783 		break;
2784 	case MSR_STAR:
2785 		svm->vmcb01.ptr->save.star = data;
2786 		break;
2787 #ifdef CONFIG_X86_64
2788 	case MSR_LSTAR:
2789 		svm->vmcb01.ptr->save.lstar = data;
2790 		break;
2791 	case MSR_CSTAR:
2792 		svm->vmcb01.ptr->save.cstar = data;
2793 		break;
2794 	case MSR_KERNEL_GS_BASE:
2795 		svm->vmcb01.ptr->save.kernel_gs_base = data;
2796 		break;
2797 	case MSR_SYSCALL_MASK:
2798 		svm->vmcb01.ptr->save.sfmask = data;
2799 		break;
2800 #endif
2801 	case MSR_IA32_SYSENTER_CS:
2802 		svm->vmcb01.ptr->save.sysenter_cs = data;
2803 		break;
2804 	case MSR_IA32_SYSENTER_EIP:
2805 		svm->vmcb01.ptr->save.sysenter_eip = (u32)data;
2806 		/*
2807 		 * We only intercept the MSR_IA32_SYSENTER_{EIP|ESP} msrs
2808 		 * when we spoof an Intel vendor ID (for cross vendor migration).
2809 		 * In this case we use this intercept to track the high
2810 		 * 32 bit part of these msrs to support Intel's
2811 		 * implementation of SYSENTER/SYSEXIT.
2812 		 */
2813 		svm->sysenter_eip_hi = guest_cpuid_is_intel(vcpu) ? (data >> 32) : 0;
2814 		break;
2815 	case MSR_IA32_SYSENTER_ESP:
2816 		svm->vmcb01.ptr->save.sysenter_esp = (u32)data;
2817 		svm->sysenter_esp_hi = guest_cpuid_is_intel(vcpu) ? (data >> 32) : 0;
2818 		break;
2819 	case MSR_TSC_AUX:
2820 		/*
2821 		 * TSC_AUX is usually changed only during boot and never read
2822 		 * directly.  Intercept TSC_AUX instead of exposing it to the
2823 		 * guest via direct_access_msrs, and switch it via user return.
2824 		 */
2825 		preempt_disable();
2826 		r = kvm_set_user_return_msr(tsc_aux_uret_slot, data, -1ull);
2827 		preempt_enable();
2828 		if (r)
2829 			return 1;
2830 
2831 		svm->tsc_aux = data;
2832 		break;
2833 	case MSR_IA32_DEBUGCTLMSR:
2834 		if (!lbrv) {
2835 			vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
2836 				    __func__, data);
2837 			break;
2838 		}
2839 		if (data & DEBUGCTL_RESERVED_BITS)
2840 			return 1;
2841 
2842 		svm->vmcb->save.dbgctl = data;
2843 		vmcb_mark_dirty(svm->vmcb, VMCB_LBR);
2844 		if (data & (1ULL<<0))
2845 			svm_enable_lbrv(vcpu);
2846 		else
2847 			svm_disable_lbrv(vcpu);
2848 		break;
2849 	case MSR_VM_HSAVE_PA:
2850 		/*
2851 		 * Old kernels did not validate the value written to
2852 		 * MSR_VM_HSAVE_PA.  Allow KVM_SET_MSR to set an invalid
2853 		 * value to allow live migrating buggy or malicious guests
2854 		 * originating from those kernels.
2855 		 */
2856 		if (!msr->host_initiated && !page_address_valid(vcpu, data))
2857 			return 1;
2858 
2859 		svm->nested.hsave_msr = data & PAGE_MASK;
2860 		break;
2861 	case MSR_VM_CR:
2862 		return svm_set_vm_cr(vcpu, data);
2863 	case MSR_VM_IGNNE:
2864 		vcpu_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data);
2865 		break;
2866 	case MSR_F10H_DECFG: {
2867 		struct kvm_msr_entry msr_entry;
2868 
2869 		msr_entry.index = msr->index;
2870 		if (svm_get_msr_feature(&msr_entry))
2871 			return 1;
2872 
2873 		/* Check the supported bits */
2874 		if (data & ~msr_entry.data)
2875 			return 1;
2876 
2877 		/* Don't allow the guest to change a bit, #GP */
2878 		if (!msr->host_initiated && (data ^ msr_entry.data))
2879 			return 1;
2880 
2881 		svm->msr_decfg = data;
2882 		break;
2883 	}
2884 	default:
2885 		return kvm_set_msr_common(vcpu, msr);
2886 	}
2887 	return 0;
2888 }
2889 
2890 static int msr_interception(struct kvm_vcpu *vcpu)
2891 {
2892 	if (to_svm(vcpu)->vmcb->control.exit_info_1)
2893 		return kvm_emulate_wrmsr(vcpu);
2894 	else
2895 		return kvm_emulate_rdmsr(vcpu);
2896 }
2897 
2898 static int interrupt_window_interception(struct kvm_vcpu *vcpu)
2899 {
2900 	kvm_make_request(KVM_REQ_EVENT, vcpu);
2901 	svm_clear_vintr(to_svm(vcpu));
2902 
2903 	/*
2904 	 * For AVIC, the only reason to end up here is ExtINTs.
2905 	 * In this case AVIC was temporarily disabled for
2906 	 * requesting the IRQ window and we have to re-enable it.
2907 	 */
2908 	kvm_clear_apicv_inhibit(vcpu->kvm, APICV_INHIBIT_REASON_IRQWIN);
2909 
2910 	++vcpu->stat.irq_window_exits;
2911 	return 1;
2912 }
2913 
2914 static int pause_interception(struct kvm_vcpu *vcpu)
2915 {
2916 	bool in_kernel;
2917 
2918 	/*
2919 	 * CPL is not made available for an SEV-ES guest, therefore
2920 	 * vcpu->arch.preempted_in_kernel can never be true.  Just
2921 	 * set in_kernel to false as well.
2922 	 */
2923 	in_kernel = !sev_es_guest(vcpu->kvm) && svm_get_cpl(vcpu) == 0;
2924 
2925 	if (!kvm_pause_in_guest(vcpu->kvm))
2926 		grow_ple_window(vcpu);
2927 
2928 	kvm_vcpu_on_spin(vcpu, in_kernel);
2929 	return kvm_skip_emulated_instruction(vcpu);
2930 }
2931 
2932 static int invpcid_interception(struct kvm_vcpu *vcpu)
2933 {
2934 	struct vcpu_svm *svm = to_svm(vcpu);
2935 	unsigned long type;
2936 	gva_t gva;
2937 
2938 	if (!guest_cpuid_has(vcpu, X86_FEATURE_INVPCID)) {
2939 		kvm_queue_exception(vcpu, UD_VECTOR);
2940 		return 1;
2941 	}
2942 
2943 	/*
2944 	 * For an INVPCID intercept:
2945 	 * EXITINFO1 provides the linear address of the memory operand.
2946 	 * EXITINFO2 provides the contents of the register operand.
2947 	 */
2948 	type = svm->vmcb->control.exit_info_2;
2949 	gva = svm->vmcb->control.exit_info_1;
2950 
2951 	return kvm_handle_invpcid(vcpu, type, gva);
2952 }
2953 
2954 static int (*const svm_exit_handlers[])(struct kvm_vcpu *vcpu) = {
2955 	[SVM_EXIT_READ_CR0]			= cr_interception,
2956 	[SVM_EXIT_READ_CR3]			= cr_interception,
2957 	[SVM_EXIT_READ_CR4]			= cr_interception,
2958 	[SVM_EXIT_READ_CR8]			= cr_interception,
2959 	[SVM_EXIT_CR0_SEL_WRITE]		= cr_interception,
2960 	[SVM_EXIT_WRITE_CR0]			= cr_interception,
2961 	[SVM_EXIT_WRITE_CR3]			= cr_interception,
2962 	[SVM_EXIT_WRITE_CR4]			= cr_interception,
2963 	[SVM_EXIT_WRITE_CR8]			= cr8_write_interception,
2964 	[SVM_EXIT_READ_DR0]			= dr_interception,
2965 	[SVM_EXIT_READ_DR1]			= dr_interception,
2966 	[SVM_EXIT_READ_DR2]			= dr_interception,
2967 	[SVM_EXIT_READ_DR3]			= dr_interception,
2968 	[SVM_EXIT_READ_DR4]			= dr_interception,
2969 	[SVM_EXIT_READ_DR5]			= dr_interception,
2970 	[SVM_EXIT_READ_DR6]			= dr_interception,
2971 	[SVM_EXIT_READ_DR7]			= dr_interception,
2972 	[SVM_EXIT_WRITE_DR0]			= dr_interception,
2973 	[SVM_EXIT_WRITE_DR1]			= dr_interception,
2974 	[SVM_EXIT_WRITE_DR2]			= dr_interception,
2975 	[SVM_EXIT_WRITE_DR3]			= dr_interception,
2976 	[SVM_EXIT_WRITE_DR4]			= dr_interception,
2977 	[SVM_EXIT_WRITE_DR5]			= dr_interception,
2978 	[SVM_EXIT_WRITE_DR6]			= dr_interception,
2979 	[SVM_EXIT_WRITE_DR7]			= dr_interception,
2980 	[SVM_EXIT_EXCP_BASE + DB_VECTOR]	= db_interception,
2981 	[SVM_EXIT_EXCP_BASE + BP_VECTOR]	= bp_interception,
2982 	[SVM_EXIT_EXCP_BASE + UD_VECTOR]	= ud_interception,
2983 	[SVM_EXIT_EXCP_BASE + PF_VECTOR]	= pf_interception,
2984 	[SVM_EXIT_EXCP_BASE + MC_VECTOR]	= mc_interception,
2985 	[SVM_EXIT_EXCP_BASE + AC_VECTOR]	= ac_interception,
2986 	[SVM_EXIT_EXCP_BASE + GP_VECTOR]	= gp_interception,
2987 	[SVM_EXIT_INTR]				= intr_interception,
2988 	[SVM_EXIT_NMI]				= nmi_interception,
2989 	[SVM_EXIT_SMI]				= smi_interception,
2990 	[SVM_EXIT_VINTR]			= interrupt_window_interception,
2991 	[SVM_EXIT_RDPMC]			= kvm_emulate_rdpmc,
2992 	[SVM_EXIT_CPUID]			= kvm_emulate_cpuid,
2993 	[SVM_EXIT_IRET]                         = iret_interception,
2994 	[SVM_EXIT_INVD]                         = kvm_emulate_invd,
2995 	[SVM_EXIT_PAUSE]			= pause_interception,
2996 	[SVM_EXIT_HLT]				= kvm_emulate_halt,
2997 	[SVM_EXIT_INVLPG]			= invlpg_interception,
2998 	[SVM_EXIT_INVLPGA]			= invlpga_interception,
2999 	[SVM_EXIT_IOIO]				= io_interception,
3000 	[SVM_EXIT_MSR]				= msr_interception,
3001 	[SVM_EXIT_TASK_SWITCH]			= task_switch_interception,
3002 	[SVM_EXIT_SHUTDOWN]			= shutdown_interception,
3003 	[SVM_EXIT_VMRUN]			= vmrun_interception,
3004 	[SVM_EXIT_VMMCALL]			= kvm_emulate_hypercall,
3005 	[SVM_EXIT_VMLOAD]			= vmload_interception,
3006 	[SVM_EXIT_VMSAVE]			= vmsave_interception,
3007 	[SVM_EXIT_STGI]				= stgi_interception,
3008 	[SVM_EXIT_CLGI]				= clgi_interception,
3009 	[SVM_EXIT_SKINIT]			= skinit_interception,
3010 	[SVM_EXIT_RDTSCP]			= kvm_handle_invalid_op,
3011 	[SVM_EXIT_WBINVD]                       = kvm_emulate_wbinvd,
3012 	[SVM_EXIT_MONITOR]			= kvm_emulate_monitor,
3013 	[SVM_EXIT_MWAIT]			= kvm_emulate_mwait,
3014 	[SVM_EXIT_XSETBV]			= kvm_emulate_xsetbv,
3015 	[SVM_EXIT_RDPRU]			= kvm_handle_invalid_op,
3016 	[SVM_EXIT_EFER_WRITE_TRAP]		= efer_trap,
3017 	[SVM_EXIT_CR0_WRITE_TRAP]		= cr_trap,
3018 	[SVM_EXIT_CR4_WRITE_TRAP]		= cr_trap,
3019 	[SVM_EXIT_CR8_WRITE_TRAP]		= cr_trap,
3020 	[SVM_EXIT_INVPCID]                      = invpcid_interception,
3021 	[SVM_EXIT_NPF]				= npf_interception,
3022 	[SVM_EXIT_RSM]                          = rsm_interception,
3023 	[SVM_EXIT_AVIC_INCOMPLETE_IPI]		= avic_incomplete_ipi_interception,
3024 	[SVM_EXIT_AVIC_UNACCELERATED_ACCESS]	= avic_unaccelerated_access_interception,
3025 	[SVM_EXIT_VMGEXIT]			= sev_handle_vmgexit,
3026 };
3027 
3028 static void dump_vmcb(struct kvm_vcpu *vcpu)
3029 {
3030 	struct vcpu_svm *svm = to_svm(vcpu);
3031 	struct vmcb_control_area *control = &svm->vmcb->control;
3032 	struct vmcb_save_area *save = &svm->vmcb->save;
3033 	struct vmcb_save_area *save01 = &svm->vmcb01.ptr->save;
3034 
3035 	if (!dump_invalid_vmcb) {
3036 		pr_warn_ratelimited("set kvm_amd.dump_invalid_vmcb=1 to dump internal KVM state.\n");
3037 		return;
3038 	}
3039 
3040 	pr_err("VMCB %p, last attempted VMRUN on CPU %d\n",
3041 	       svm->current_vmcb->ptr, vcpu->arch.last_vmentry_cpu);
3042 	pr_err("VMCB Control Area:\n");
3043 	pr_err("%-20s%04x\n", "cr_read:", control->intercepts[INTERCEPT_CR] & 0xffff);
3044 	pr_err("%-20s%04x\n", "cr_write:", control->intercepts[INTERCEPT_CR] >> 16);
3045 	pr_err("%-20s%04x\n", "dr_read:", control->intercepts[INTERCEPT_DR] & 0xffff);
3046 	pr_err("%-20s%04x\n", "dr_write:", control->intercepts[INTERCEPT_DR] >> 16);
3047 	pr_err("%-20s%08x\n", "exceptions:", control->intercepts[INTERCEPT_EXCEPTION]);
3048 	pr_err("%-20s%08x %08x\n", "intercepts:",
3049               control->intercepts[INTERCEPT_WORD3],
3050 	       control->intercepts[INTERCEPT_WORD4]);
3051 	pr_err("%-20s%d\n", "pause filter count:", control->pause_filter_count);
3052 	pr_err("%-20s%d\n", "pause filter threshold:",
3053 	       control->pause_filter_thresh);
3054 	pr_err("%-20s%016llx\n", "iopm_base_pa:", control->iopm_base_pa);
3055 	pr_err("%-20s%016llx\n", "msrpm_base_pa:", control->msrpm_base_pa);
3056 	pr_err("%-20s%016llx\n", "tsc_offset:", control->tsc_offset);
3057 	pr_err("%-20s%d\n", "asid:", control->asid);
3058 	pr_err("%-20s%d\n", "tlb_ctl:", control->tlb_ctl);
3059 	pr_err("%-20s%08x\n", "int_ctl:", control->int_ctl);
3060 	pr_err("%-20s%08x\n", "int_vector:", control->int_vector);
3061 	pr_err("%-20s%08x\n", "int_state:", control->int_state);
3062 	pr_err("%-20s%08x\n", "exit_code:", control->exit_code);
3063 	pr_err("%-20s%016llx\n", "exit_info1:", control->exit_info_1);
3064 	pr_err("%-20s%016llx\n", "exit_info2:", control->exit_info_2);
3065 	pr_err("%-20s%08x\n", "exit_int_info:", control->exit_int_info);
3066 	pr_err("%-20s%08x\n", "exit_int_info_err:", control->exit_int_info_err);
3067 	pr_err("%-20s%lld\n", "nested_ctl:", control->nested_ctl);
3068 	pr_err("%-20s%016llx\n", "nested_cr3:", control->nested_cr3);
3069 	pr_err("%-20s%016llx\n", "avic_vapic_bar:", control->avic_vapic_bar);
3070 	pr_err("%-20s%016llx\n", "ghcb:", control->ghcb_gpa);
3071 	pr_err("%-20s%08x\n", "event_inj:", control->event_inj);
3072 	pr_err("%-20s%08x\n", "event_inj_err:", control->event_inj_err);
3073 	pr_err("%-20s%lld\n", "virt_ext:", control->virt_ext);
3074 	pr_err("%-20s%016llx\n", "next_rip:", control->next_rip);
3075 	pr_err("%-20s%016llx\n", "avic_backing_page:", control->avic_backing_page);
3076 	pr_err("%-20s%016llx\n", "avic_logical_id:", control->avic_logical_id);
3077 	pr_err("%-20s%016llx\n", "avic_physical_id:", control->avic_physical_id);
3078 	pr_err("%-20s%016llx\n", "vmsa_pa:", control->vmsa_pa);
3079 	pr_err("VMCB State Save Area:\n");
3080 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3081 	       "es:",
3082 	       save->es.selector, save->es.attrib,
3083 	       save->es.limit, save->es.base);
3084 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3085 	       "cs:",
3086 	       save->cs.selector, save->cs.attrib,
3087 	       save->cs.limit, save->cs.base);
3088 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3089 	       "ss:",
3090 	       save->ss.selector, save->ss.attrib,
3091 	       save->ss.limit, save->ss.base);
3092 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3093 	       "ds:",
3094 	       save->ds.selector, save->ds.attrib,
3095 	       save->ds.limit, save->ds.base);
3096 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3097 	       "fs:",
3098 	       save01->fs.selector, save01->fs.attrib,
3099 	       save01->fs.limit, save01->fs.base);
3100 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3101 	       "gs:",
3102 	       save01->gs.selector, save01->gs.attrib,
3103 	       save01->gs.limit, save01->gs.base);
3104 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3105 	       "gdtr:",
3106 	       save->gdtr.selector, save->gdtr.attrib,
3107 	       save->gdtr.limit, save->gdtr.base);
3108 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3109 	       "ldtr:",
3110 	       save01->ldtr.selector, save01->ldtr.attrib,
3111 	       save01->ldtr.limit, save01->ldtr.base);
3112 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3113 	       "idtr:",
3114 	       save->idtr.selector, save->idtr.attrib,
3115 	       save->idtr.limit, save->idtr.base);
3116 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3117 	       "tr:",
3118 	       save01->tr.selector, save01->tr.attrib,
3119 	       save01->tr.limit, save01->tr.base);
3120 	pr_err("vmpl: %d   cpl:  %d               efer:          %016llx\n",
3121 	       save->vmpl, save->cpl, save->efer);
3122 	pr_err("%-15s %016llx %-13s %016llx\n",
3123 	       "cr0:", save->cr0, "cr2:", save->cr2);
3124 	pr_err("%-15s %016llx %-13s %016llx\n",
3125 	       "cr3:", save->cr3, "cr4:", save->cr4);
3126 	pr_err("%-15s %016llx %-13s %016llx\n",
3127 	       "dr6:", save->dr6, "dr7:", save->dr7);
3128 	pr_err("%-15s %016llx %-13s %016llx\n",
3129 	       "rip:", save->rip, "rflags:", save->rflags);
3130 	pr_err("%-15s %016llx %-13s %016llx\n",
3131 	       "rsp:", save->rsp, "rax:", save->rax);
3132 	pr_err("%-15s %016llx %-13s %016llx\n",
3133 	       "star:", save01->star, "lstar:", save01->lstar);
3134 	pr_err("%-15s %016llx %-13s %016llx\n",
3135 	       "cstar:", save01->cstar, "sfmask:", save01->sfmask);
3136 	pr_err("%-15s %016llx %-13s %016llx\n",
3137 	       "kernel_gs_base:", save01->kernel_gs_base,
3138 	       "sysenter_cs:", save01->sysenter_cs);
3139 	pr_err("%-15s %016llx %-13s %016llx\n",
3140 	       "sysenter_esp:", save01->sysenter_esp,
3141 	       "sysenter_eip:", save01->sysenter_eip);
3142 	pr_err("%-15s %016llx %-13s %016llx\n",
3143 	       "gpat:", save->g_pat, "dbgctl:", save->dbgctl);
3144 	pr_err("%-15s %016llx %-13s %016llx\n",
3145 	       "br_from:", save->br_from, "br_to:", save->br_to);
3146 	pr_err("%-15s %016llx %-13s %016llx\n",
3147 	       "excp_from:", save->last_excp_from,
3148 	       "excp_to:", save->last_excp_to);
3149 }
3150 
3151 static bool svm_check_exit_valid(u64 exit_code)
3152 {
3153 	return (exit_code < ARRAY_SIZE(svm_exit_handlers) &&
3154 		svm_exit_handlers[exit_code]);
3155 }
3156 
3157 static int svm_handle_invalid_exit(struct kvm_vcpu *vcpu, u64 exit_code)
3158 {
3159 	vcpu_unimpl(vcpu, "svm: unexpected exit reason 0x%llx\n", exit_code);
3160 	dump_vmcb(vcpu);
3161 	vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3162 	vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON;
3163 	vcpu->run->internal.ndata = 2;
3164 	vcpu->run->internal.data[0] = exit_code;
3165 	vcpu->run->internal.data[1] = vcpu->arch.last_vmentry_cpu;
3166 	return 0;
3167 }
3168 
3169 int svm_invoke_exit_handler(struct kvm_vcpu *vcpu, u64 exit_code)
3170 {
3171 	if (!svm_check_exit_valid(exit_code))
3172 		return svm_handle_invalid_exit(vcpu, exit_code);
3173 
3174 #ifdef CONFIG_RETPOLINE
3175 	if (exit_code == SVM_EXIT_MSR)
3176 		return msr_interception(vcpu);
3177 	else if (exit_code == SVM_EXIT_VINTR)
3178 		return interrupt_window_interception(vcpu);
3179 	else if (exit_code == SVM_EXIT_INTR)
3180 		return intr_interception(vcpu);
3181 	else if (exit_code == SVM_EXIT_HLT)
3182 		return kvm_emulate_halt(vcpu);
3183 	else if (exit_code == SVM_EXIT_NPF)
3184 		return npf_interception(vcpu);
3185 #endif
3186 	return svm_exit_handlers[exit_code](vcpu);
3187 }
3188 
3189 static void svm_get_exit_info(struct kvm_vcpu *vcpu, u32 *reason,
3190 			      u64 *info1, u64 *info2,
3191 			      u32 *intr_info, u32 *error_code)
3192 {
3193 	struct vmcb_control_area *control = &to_svm(vcpu)->vmcb->control;
3194 
3195 	*reason = control->exit_code;
3196 	*info1 = control->exit_info_1;
3197 	*info2 = control->exit_info_2;
3198 	*intr_info = control->exit_int_info;
3199 	if ((*intr_info & SVM_EXITINTINFO_VALID) &&
3200 	    (*intr_info & SVM_EXITINTINFO_VALID_ERR))
3201 		*error_code = control->exit_int_info_err;
3202 	else
3203 		*error_code = 0;
3204 }
3205 
3206 static int svm_handle_exit(struct kvm_vcpu *vcpu, fastpath_t exit_fastpath)
3207 {
3208 	struct vcpu_svm *svm = to_svm(vcpu);
3209 	struct kvm_run *kvm_run = vcpu->run;
3210 	u32 exit_code = svm->vmcb->control.exit_code;
3211 
3212 	trace_kvm_exit(vcpu, KVM_ISA_SVM);
3213 
3214 	/* SEV-ES guests must use the CR write traps to track CR registers. */
3215 	if (!sev_es_guest(vcpu->kvm)) {
3216 		if (!svm_is_intercept(svm, INTERCEPT_CR0_WRITE))
3217 			vcpu->arch.cr0 = svm->vmcb->save.cr0;
3218 		if (npt_enabled)
3219 			vcpu->arch.cr3 = svm->vmcb->save.cr3;
3220 	}
3221 
3222 	if (is_guest_mode(vcpu)) {
3223 		int vmexit;
3224 
3225 		trace_kvm_nested_vmexit(vcpu, KVM_ISA_SVM);
3226 
3227 		vmexit = nested_svm_exit_special(svm);
3228 
3229 		if (vmexit == NESTED_EXIT_CONTINUE)
3230 			vmexit = nested_svm_exit_handled(svm);
3231 
3232 		if (vmexit == NESTED_EXIT_DONE)
3233 			return 1;
3234 	}
3235 
3236 	if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
3237 		kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3238 		kvm_run->fail_entry.hardware_entry_failure_reason
3239 			= svm->vmcb->control.exit_code;
3240 		kvm_run->fail_entry.cpu = vcpu->arch.last_vmentry_cpu;
3241 		dump_vmcb(vcpu);
3242 		return 0;
3243 	}
3244 
3245 	if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
3246 	    exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
3247 	    exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH &&
3248 	    exit_code != SVM_EXIT_INTR && exit_code != SVM_EXIT_NMI)
3249 		printk(KERN_ERR "%s: unexpected exit_int_info 0x%x "
3250 		       "exit_code 0x%x\n",
3251 		       __func__, svm->vmcb->control.exit_int_info,
3252 		       exit_code);
3253 
3254 	if (exit_fastpath != EXIT_FASTPATH_NONE)
3255 		return 1;
3256 
3257 	return svm_invoke_exit_handler(vcpu, exit_code);
3258 }
3259 
3260 static void reload_tss(struct kvm_vcpu *vcpu)
3261 {
3262 	struct svm_cpu_data *sd = per_cpu(svm_data, vcpu->cpu);
3263 
3264 	sd->tss_desc->type = 9; /* available 32/64-bit TSS */
3265 	load_TR_desc();
3266 }
3267 
3268 static void pre_svm_run(struct kvm_vcpu *vcpu)
3269 {
3270 	struct svm_cpu_data *sd = per_cpu(svm_data, vcpu->cpu);
3271 	struct vcpu_svm *svm = to_svm(vcpu);
3272 
3273 	/*
3274 	 * If the previous vmrun of the vmcb occurred on a different physical
3275 	 * cpu, then mark the vmcb dirty and assign a new asid.  Hardware's
3276 	 * vmcb clean bits are per logical CPU, as are KVM's asid assignments.
3277 	 */
3278 	if (unlikely(svm->current_vmcb->cpu != vcpu->cpu)) {
3279 		svm->current_vmcb->asid_generation = 0;
3280 		vmcb_mark_all_dirty(svm->vmcb);
3281 		svm->current_vmcb->cpu = vcpu->cpu;
3282         }
3283 
3284 	if (sev_guest(vcpu->kvm))
3285 		return pre_sev_run(svm, vcpu->cpu);
3286 
3287 	/* FIXME: handle wraparound of asid_generation */
3288 	if (svm->current_vmcb->asid_generation != sd->asid_generation)
3289 		new_asid(svm, sd);
3290 }
3291 
3292 static void svm_inject_nmi(struct kvm_vcpu *vcpu)
3293 {
3294 	struct vcpu_svm *svm = to_svm(vcpu);
3295 
3296 	svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
3297 	vcpu->arch.hflags |= HF_NMI_MASK;
3298 	if (!sev_es_guest(vcpu->kvm))
3299 		svm_set_intercept(svm, INTERCEPT_IRET);
3300 	++vcpu->stat.nmi_injections;
3301 }
3302 
3303 static void svm_inject_irq(struct kvm_vcpu *vcpu)
3304 {
3305 	struct vcpu_svm *svm = to_svm(vcpu);
3306 
3307 	BUG_ON(!(gif_set(svm)));
3308 
3309 	trace_kvm_inj_virq(vcpu->arch.interrupt.nr);
3310 	++vcpu->stat.irq_injections;
3311 
3312 	svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr |
3313 		SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR;
3314 }
3315 
3316 void svm_complete_interrupt_delivery(struct kvm_vcpu *vcpu, int delivery_mode,
3317 				     int trig_mode, int vector)
3318 {
3319 	/*
3320 	 * vcpu->arch.apicv_active must be read after vcpu->mode.
3321 	 * Pairs with smp_store_release in vcpu_enter_guest.
3322 	 */
3323 	bool in_guest_mode = (smp_load_acquire(&vcpu->mode) == IN_GUEST_MODE);
3324 
3325 	if (!READ_ONCE(vcpu->arch.apicv_active)) {
3326 		/* Process the interrupt via inject_pending_event */
3327 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3328 		kvm_vcpu_kick(vcpu);
3329 		return;
3330 	}
3331 
3332 	trace_kvm_apicv_accept_irq(vcpu->vcpu_id, delivery_mode, trig_mode, vector);
3333 	if (in_guest_mode) {
3334 		/*
3335 		 * Signal the doorbell to tell hardware to inject the IRQ.  If
3336 		 * the vCPU exits the guest before the doorbell chimes, hardware
3337 		 * will automatically process AVIC interrupts at the next VMRUN.
3338 		 */
3339 		avic_ring_doorbell(vcpu);
3340 	} else {
3341 		/*
3342 		 * Wake the vCPU if it was blocking.  KVM will then detect the
3343 		 * pending IRQ when checking if the vCPU has a wake event.
3344 		 */
3345 		kvm_vcpu_wake_up(vcpu);
3346 	}
3347 }
3348 
3349 static void svm_deliver_interrupt(struct kvm_lapic *apic,  int delivery_mode,
3350 				  int trig_mode, int vector)
3351 {
3352 	kvm_lapic_set_irr(vector, apic);
3353 
3354 	/*
3355 	 * Pairs with the smp_mb_*() after setting vcpu->guest_mode in
3356 	 * vcpu_enter_guest() to ensure the write to the vIRR is ordered before
3357 	 * the read of guest_mode.  This guarantees that either VMRUN will see
3358 	 * and process the new vIRR entry, or that svm_complete_interrupt_delivery
3359 	 * will signal the doorbell if the CPU has already entered the guest.
3360 	 */
3361 	smp_mb__after_atomic();
3362 	svm_complete_interrupt_delivery(apic->vcpu, delivery_mode, trig_mode, vector);
3363 }
3364 
3365 static void svm_update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
3366 {
3367 	struct vcpu_svm *svm = to_svm(vcpu);
3368 
3369 	/*
3370 	 * SEV-ES guests must always keep the CR intercepts cleared. CR
3371 	 * tracking is done using the CR write traps.
3372 	 */
3373 	if (sev_es_guest(vcpu->kvm))
3374 		return;
3375 
3376 	if (nested_svm_virtualize_tpr(vcpu))
3377 		return;
3378 
3379 	svm_clr_intercept(svm, INTERCEPT_CR8_WRITE);
3380 
3381 	if (irr == -1)
3382 		return;
3383 
3384 	if (tpr >= irr)
3385 		svm_set_intercept(svm, INTERCEPT_CR8_WRITE);
3386 }
3387 
3388 bool svm_nmi_blocked(struct kvm_vcpu *vcpu)
3389 {
3390 	struct vcpu_svm *svm = to_svm(vcpu);
3391 	struct vmcb *vmcb = svm->vmcb;
3392 	bool ret;
3393 
3394 	if (!gif_set(svm))
3395 		return true;
3396 
3397 	if (is_guest_mode(vcpu) && nested_exit_on_nmi(svm))
3398 		return false;
3399 
3400 	ret = (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) ||
3401 	      (vcpu->arch.hflags & HF_NMI_MASK);
3402 
3403 	return ret;
3404 }
3405 
3406 static int svm_nmi_allowed(struct kvm_vcpu *vcpu, bool for_injection)
3407 {
3408 	struct vcpu_svm *svm = to_svm(vcpu);
3409 	if (svm->nested.nested_run_pending)
3410 		return -EBUSY;
3411 
3412 	if (svm_nmi_blocked(vcpu))
3413 		return 0;
3414 
3415 	/* An NMI must not be injected into L2 if it's supposed to VM-Exit.  */
3416 	if (for_injection && is_guest_mode(vcpu) && nested_exit_on_nmi(svm))
3417 		return -EBUSY;
3418 	return 1;
3419 }
3420 
3421 static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu)
3422 {
3423 	return !!(vcpu->arch.hflags & HF_NMI_MASK);
3424 }
3425 
3426 static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
3427 {
3428 	struct vcpu_svm *svm = to_svm(vcpu);
3429 
3430 	if (masked) {
3431 		vcpu->arch.hflags |= HF_NMI_MASK;
3432 		if (!sev_es_guest(vcpu->kvm))
3433 			svm_set_intercept(svm, INTERCEPT_IRET);
3434 	} else {
3435 		vcpu->arch.hflags &= ~HF_NMI_MASK;
3436 		if (!sev_es_guest(vcpu->kvm))
3437 			svm_clr_intercept(svm, INTERCEPT_IRET);
3438 	}
3439 }
3440 
3441 bool svm_interrupt_blocked(struct kvm_vcpu *vcpu)
3442 {
3443 	struct vcpu_svm *svm = to_svm(vcpu);
3444 	struct vmcb *vmcb = svm->vmcb;
3445 
3446 	if (!gif_set(svm))
3447 		return true;
3448 
3449 	if (is_guest_mode(vcpu)) {
3450 		/* As long as interrupts are being delivered...  */
3451 		if ((svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK)
3452 		    ? !(svm->vmcb01.ptr->save.rflags & X86_EFLAGS_IF)
3453 		    : !(kvm_get_rflags(vcpu) & X86_EFLAGS_IF))
3454 			return true;
3455 
3456 		/* ... vmexits aren't blocked by the interrupt shadow  */
3457 		if (nested_exit_on_intr(svm))
3458 			return false;
3459 	} else {
3460 		if (!svm_get_if_flag(vcpu))
3461 			return true;
3462 	}
3463 
3464 	return (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK);
3465 }
3466 
3467 static int svm_interrupt_allowed(struct kvm_vcpu *vcpu, bool for_injection)
3468 {
3469 	struct vcpu_svm *svm = to_svm(vcpu);
3470 
3471 	if (svm->nested.nested_run_pending)
3472 		return -EBUSY;
3473 
3474 	if (svm_interrupt_blocked(vcpu))
3475 		return 0;
3476 
3477 	/*
3478 	 * An IRQ must not be injected into L2 if it's supposed to VM-Exit,
3479 	 * e.g. if the IRQ arrived asynchronously after checking nested events.
3480 	 */
3481 	if (for_injection && is_guest_mode(vcpu) && nested_exit_on_intr(svm))
3482 		return -EBUSY;
3483 
3484 	return 1;
3485 }
3486 
3487 static void svm_enable_irq_window(struct kvm_vcpu *vcpu)
3488 {
3489 	struct vcpu_svm *svm = to_svm(vcpu);
3490 
3491 	/*
3492 	 * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
3493 	 * 1, because that's a separate STGI/VMRUN intercept.  The next time we
3494 	 * get that intercept, this function will be called again though and
3495 	 * we'll get the vintr intercept. However, if the vGIF feature is
3496 	 * enabled, the STGI interception will not occur. Enable the irq
3497 	 * window under the assumption that the hardware will set the GIF.
3498 	 */
3499 	if (vgif_enabled(svm) || gif_set(svm)) {
3500 		/*
3501 		 * IRQ window is not needed when AVIC is enabled,
3502 		 * unless we have pending ExtINT since it cannot be injected
3503 		 * via AVIC. In such case, we need to temporarily disable AVIC,
3504 		 * and fallback to injecting IRQ via V_IRQ.
3505 		 */
3506 		kvm_set_apicv_inhibit(vcpu->kvm, APICV_INHIBIT_REASON_IRQWIN);
3507 		svm_set_vintr(svm);
3508 	}
3509 }
3510 
3511 static void svm_enable_nmi_window(struct kvm_vcpu *vcpu)
3512 {
3513 	struct vcpu_svm *svm = to_svm(vcpu);
3514 
3515 	if ((vcpu->arch.hflags & (HF_NMI_MASK | HF_IRET_MASK)) == HF_NMI_MASK)
3516 		return; /* IRET will cause a vm exit */
3517 
3518 	if (!gif_set(svm)) {
3519 		if (vgif_enabled(svm))
3520 			svm_set_intercept(svm, INTERCEPT_STGI);
3521 		return; /* STGI will cause a vm exit */
3522 	}
3523 
3524 	/*
3525 	 * Something prevents NMI from been injected. Single step over possible
3526 	 * problem (IRET or exception injection or interrupt shadow)
3527 	 */
3528 	svm->nmi_singlestep_guest_rflags = svm_get_rflags(vcpu);
3529 	svm->nmi_singlestep = true;
3530 	svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
3531 }
3532 
3533 static void svm_flush_tlb_current(struct kvm_vcpu *vcpu)
3534 {
3535 	struct vcpu_svm *svm = to_svm(vcpu);
3536 
3537 	/*
3538 	 * Flush only the current ASID even if the TLB flush was invoked via
3539 	 * kvm_flush_remote_tlbs().  Although flushing remote TLBs requires all
3540 	 * ASIDs to be flushed, KVM uses a single ASID for L1 and L2, and
3541 	 * unconditionally does a TLB flush on both nested VM-Enter and nested
3542 	 * VM-Exit (via kvm_mmu_reset_context()).
3543 	 */
3544 	if (static_cpu_has(X86_FEATURE_FLUSHBYASID))
3545 		svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
3546 	else
3547 		svm->current_vmcb->asid_generation--;
3548 }
3549 
3550 static void svm_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t gva)
3551 {
3552 	struct vcpu_svm *svm = to_svm(vcpu);
3553 
3554 	invlpga(gva, svm->vmcb->control.asid);
3555 }
3556 
3557 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
3558 {
3559 	struct vcpu_svm *svm = to_svm(vcpu);
3560 
3561 	if (nested_svm_virtualize_tpr(vcpu))
3562 		return;
3563 
3564 	if (!svm_is_intercept(svm, INTERCEPT_CR8_WRITE)) {
3565 		int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
3566 		kvm_set_cr8(vcpu, cr8);
3567 	}
3568 }
3569 
3570 static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
3571 {
3572 	struct vcpu_svm *svm = to_svm(vcpu);
3573 	u64 cr8;
3574 
3575 	if (nested_svm_virtualize_tpr(vcpu) ||
3576 	    kvm_vcpu_apicv_active(vcpu))
3577 		return;
3578 
3579 	cr8 = kvm_get_cr8(vcpu);
3580 	svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
3581 	svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
3582 }
3583 
3584 static void svm_complete_interrupts(struct kvm_vcpu *vcpu)
3585 {
3586 	struct vcpu_svm *svm = to_svm(vcpu);
3587 	u8 vector;
3588 	int type;
3589 	u32 exitintinfo = svm->vmcb->control.exit_int_info;
3590 	unsigned int3_injected = svm->int3_injected;
3591 
3592 	svm->int3_injected = 0;
3593 
3594 	/*
3595 	 * If we've made progress since setting HF_IRET_MASK, we've
3596 	 * executed an IRET and can allow NMI injection.
3597 	 */
3598 	if ((vcpu->arch.hflags & HF_IRET_MASK) &&
3599 	    (sev_es_guest(vcpu->kvm) ||
3600 	     kvm_rip_read(vcpu) != svm->nmi_iret_rip)) {
3601 		vcpu->arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK);
3602 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3603 	}
3604 
3605 	vcpu->arch.nmi_injected = false;
3606 	kvm_clear_exception_queue(vcpu);
3607 	kvm_clear_interrupt_queue(vcpu);
3608 
3609 	if (!(exitintinfo & SVM_EXITINTINFO_VALID))
3610 		return;
3611 
3612 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3613 
3614 	vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK;
3615 	type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK;
3616 
3617 	switch (type) {
3618 	case SVM_EXITINTINFO_TYPE_NMI:
3619 		vcpu->arch.nmi_injected = true;
3620 		break;
3621 	case SVM_EXITINTINFO_TYPE_EXEPT:
3622 		/*
3623 		 * Never re-inject a #VC exception.
3624 		 */
3625 		if (vector == X86_TRAP_VC)
3626 			break;
3627 
3628 		/*
3629 		 * In case of software exceptions, do not reinject the vector,
3630 		 * but re-execute the instruction instead. Rewind RIP first
3631 		 * if we emulated INT3 before.
3632 		 */
3633 		if (kvm_exception_is_soft(vector)) {
3634 			if (vector == BP_VECTOR && int3_injected &&
3635 			    kvm_is_linear_rip(vcpu, svm->int3_rip))
3636 				kvm_rip_write(vcpu,
3637 					      kvm_rip_read(vcpu) - int3_injected);
3638 			break;
3639 		}
3640 		if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) {
3641 			u32 err = svm->vmcb->control.exit_int_info_err;
3642 			kvm_requeue_exception_e(vcpu, vector, err);
3643 
3644 		} else
3645 			kvm_requeue_exception(vcpu, vector);
3646 		break;
3647 	case SVM_EXITINTINFO_TYPE_INTR:
3648 		kvm_queue_interrupt(vcpu, vector, false);
3649 		break;
3650 	default:
3651 		break;
3652 	}
3653 }
3654 
3655 static void svm_cancel_injection(struct kvm_vcpu *vcpu)
3656 {
3657 	struct vcpu_svm *svm = to_svm(vcpu);
3658 	struct vmcb_control_area *control = &svm->vmcb->control;
3659 
3660 	control->exit_int_info = control->event_inj;
3661 	control->exit_int_info_err = control->event_inj_err;
3662 	control->event_inj = 0;
3663 	svm_complete_interrupts(vcpu);
3664 }
3665 
3666 static int svm_vcpu_pre_run(struct kvm_vcpu *vcpu)
3667 {
3668 	return 1;
3669 }
3670 
3671 static fastpath_t svm_exit_handlers_fastpath(struct kvm_vcpu *vcpu)
3672 {
3673 	if (to_svm(vcpu)->vmcb->control.exit_code == SVM_EXIT_MSR &&
3674 	    to_svm(vcpu)->vmcb->control.exit_info_1)
3675 		return handle_fastpath_set_msr_irqoff(vcpu);
3676 
3677 	return EXIT_FASTPATH_NONE;
3678 }
3679 
3680 static noinstr void svm_vcpu_enter_exit(struct kvm_vcpu *vcpu)
3681 {
3682 	struct vcpu_svm *svm = to_svm(vcpu);
3683 	unsigned long vmcb_pa = svm->current_vmcb->pa;
3684 
3685 	guest_state_enter_irqoff();
3686 
3687 	if (sev_es_guest(vcpu->kvm)) {
3688 		__svm_sev_es_vcpu_run(vmcb_pa);
3689 	} else {
3690 		struct svm_cpu_data *sd = per_cpu(svm_data, vcpu->cpu);
3691 
3692 		/*
3693 		 * Use a single vmcb (vmcb01 because it's always valid) for
3694 		 * context switching guest state via VMLOAD/VMSAVE, that way
3695 		 * the state doesn't need to be copied between vmcb01 and
3696 		 * vmcb02 when switching vmcbs for nested virtualization.
3697 		 */
3698 		vmload(svm->vmcb01.pa);
3699 		__svm_vcpu_run(vmcb_pa, (unsigned long *)&vcpu->arch.regs);
3700 		vmsave(svm->vmcb01.pa);
3701 
3702 		vmload(__sme_page_pa(sd->save_area));
3703 	}
3704 
3705 	guest_state_exit_irqoff();
3706 }
3707 
3708 static __no_kcsan fastpath_t svm_vcpu_run(struct kvm_vcpu *vcpu)
3709 {
3710 	struct vcpu_svm *svm = to_svm(vcpu);
3711 
3712 	trace_kvm_entry(vcpu);
3713 
3714 	svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
3715 	svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
3716 	svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
3717 
3718 	/*
3719 	 * Disable singlestep if we're injecting an interrupt/exception.
3720 	 * We don't want our modified rflags to be pushed on the stack where
3721 	 * we might not be able to easily reset them if we disabled NMI
3722 	 * singlestep later.
3723 	 */
3724 	if (svm->nmi_singlestep && svm->vmcb->control.event_inj) {
3725 		/*
3726 		 * Event injection happens before external interrupts cause a
3727 		 * vmexit and interrupts are disabled here, so smp_send_reschedule
3728 		 * is enough to force an immediate vmexit.
3729 		 */
3730 		disable_nmi_singlestep(svm);
3731 		smp_send_reschedule(vcpu->cpu);
3732 	}
3733 
3734 	pre_svm_run(vcpu);
3735 
3736 	sync_lapic_to_cr8(vcpu);
3737 
3738 	if (unlikely(svm->asid != svm->vmcb->control.asid)) {
3739 		svm->vmcb->control.asid = svm->asid;
3740 		vmcb_mark_dirty(svm->vmcb, VMCB_ASID);
3741 	}
3742 	svm->vmcb->save.cr2 = vcpu->arch.cr2;
3743 
3744 	svm_hv_update_vp_id(svm->vmcb, vcpu);
3745 
3746 	/*
3747 	 * Run with all-zero DR6 unless needed, so that we can get the exact cause
3748 	 * of a #DB.
3749 	 */
3750 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT))
3751 		svm_set_dr6(svm, vcpu->arch.dr6);
3752 	else
3753 		svm_set_dr6(svm, DR6_ACTIVE_LOW);
3754 
3755 	clgi();
3756 	kvm_load_guest_xsave_state(vcpu);
3757 
3758 	kvm_wait_lapic_expire(vcpu);
3759 
3760 	/*
3761 	 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
3762 	 * it's non-zero. Since vmentry is serialising on affected CPUs, there
3763 	 * is no need to worry about the conditional branch over the wrmsr
3764 	 * being speculatively taken.
3765 	 */
3766 	if (!static_cpu_has(X86_FEATURE_V_SPEC_CTRL))
3767 		x86_spec_ctrl_set_guest(svm->spec_ctrl, svm->virt_spec_ctrl);
3768 
3769 	svm_vcpu_enter_exit(vcpu);
3770 
3771 	/*
3772 	 * We do not use IBRS in the kernel. If this vCPU has used the
3773 	 * SPEC_CTRL MSR it may have left it on; save the value and
3774 	 * turn it off. This is much more efficient than blindly adding
3775 	 * it to the atomic save/restore list. Especially as the former
3776 	 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
3777 	 *
3778 	 * For non-nested case:
3779 	 * If the L01 MSR bitmap does not intercept the MSR, then we need to
3780 	 * save it.
3781 	 *
3782 	 * For nested case:
3783 	 * If the L02 MSR bitmap does not intercept the MSR, then we need to
3784 	 * save it.
3785 	 */
3786 	if (!static_cpu_has(X86_FEATURE_V_SPEC_CTRL) &&
3787 	    unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
3788 		svm->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
3789 
3790 	if (!sev_es_guest(vcpu->kvm))
3791 		reload_tss(vcpu);
3792 
3793 	if (!static_cpu_has(X86_FEATURE_V_SPEC_CTRL))
3794 		x86_spec_ctrl_restore_host(svm->spec_ctrl, svm->virt_spec_ctrl);
3795 
3796 	if (!sev_es_guest(vcpu->kvm)) {
3797 		vcpu->arch.cr2 = svm->vmcb->save.cr2;
3798 		vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
3799 		vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
3800 		vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
3801 	}
3802 	vcpu->arch.regs_dirty = 0;
3803 
3804 	if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
3805 		kvm_before_interrupt(vcpu, KVM_HANDLING_NMI);
3806 
3807 	kvm_load_host_xsave_state(vcpu);
3808 	stgi();
3809 
3810 	/* Any pending NMI will happen here */
3811 
3812 	if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
3813 		kvm_after_interrupt(vcpu);
3814 
3815 	sync_cr8_to_lapic(vcpu);
3816 
3817 	svm->next_rip = 0;
3818 	if (is_guest_mode(vcpu)) {
3819 		nested_sync_control_from_vmcb02(svm);
3820 
3821 		/* Track VMRUNs that have made past consistency checking */
3822 		if (svm->nested.nested_run_pending &&
3823 		    svm->vmcb->control.exit_code != SVM_EXIT_ERR)
3824                         ++vcpu->stat.nested_run;
3825 
3826 		svm->nested.nested_run_pending = 0;
3827 	}
3828 
3829 	svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
3830 	vmcb_mark_all_clean(svm->vmcb);
3831 
3832 	/* if exit due to PF check for async PF */
3833 	if (svm->vmcb->control.exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR)
3834 		vcpu->arch.apf.host_apf_flags =
3835 			kvm_read_and_reset_apf_flags();
3836 
3837 	vcpu->arch.regs_avail &= ~SVM_REGS_LAZY_LOAD_SET;
3838 
3839 	/*
3840 	 * We need to handle MC intercepts here before the vcpu has a chance to
3841 	 * change the physical cpu
3842 	 */
3843 	if (unlikely(svm->vmcb->control.exit_code ==
3844 		     SVM_EXIT_EXCP_BASE + MC_VECTOR))
3845 		svm_handle_mce(vcpu);
3846 
3847 	svm_complete_interrupts(vcpu);
3848 
3849 	if (is_guest_mode(vcpu))
3850 		return EXIT_FASTPATH_NONE;
3851 
3852 	return svm_exit_handlers_fastpath(vcpu);
3853 }
3854 
3855 static void svm_load_mmu_pgd(struct kvm_vcpu *vcpu, hpa_t root_hpa,
3856 			     int root_level)
3857 {
3858 	struct vcpu_svm *svm = to_svm(vcpu);
3859 	unsigned long cr3;
3860 
3861 	if (npt_enabled) {
3862 		svm->vmcb->control.nested_cr3 = __sme_set(root_hpa);
3863 		vmcb_mark_dirty(svm->vmcb, VMCB_NPT);
3864 
3865 		hv_track_root_tdp(vcpu, root_hpa);
3866 
3867 		cr3 = vcpu->arch.cr3;
3868 	} else if (vcpu->arch.mmu->shadow_root_level >= PT64_ROOT_4LEVEL) {
3869 		cr3 = __sme_set(root_hpa) | kvm_get_active_pcid(vcpu);
3870 	} else {
3871 		/* PCID in the guest should be impossible with a 32-bit MMU. */
3872 		WARN_ON_ONCE(kvm_get_active_pcid(vcpu));
3873 		cr3 = root_hpa;
3874 	}
3875 
3876 	svm->vmcb->save.cr3 = cr3;
3877 	vmcb_mark_dirty(svm->vmcb, VMCB_CR);
3878 }
3879 
3880 static int is_disabled(void)
3881 {
3882 	u64 vm_cr;
3883 
3884 	rdmsrl(MSR_VM_CR, vm_cr);
3885 	if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
3886 		return 1;
3887 
3888 	return 0;
3889 }
3890 
3891 static void
3892 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
3893 {
3894 	/*
3895 	 * Patch in the VMMCALL instruction:
3896 	 */
3897 	hypercall[0] = 0x0f;
3898 	hypercall[1] = 0x01;
3899 	hypercall[2] = 0xd9;
3900 }
3901 
3902 static int __init svm_check_processor_compat(void)
3903 {
3904 	return 0;
3905 }
3906 
3907 /*
3908  * The kvm parameter can be NULL (module initialization, or invocation before
3909  * VM creation). Be sure to check the kvm parameter before using it.
3910  */
3911 static bool svm_has_emulated_msr(struct kvm *kvm, u32 index)
3912 {
3913 	switch (index) {
3914 	case MSR_IA32_MCG_EXT_CTL:
3915 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
3916 		return false;
3917 	case MSR_IA32_SMBASE:
3918 		/* SEV-ES guests do not support SMM, so report false */
3919 		if (kvm && sev_es_guest(kvm))
3920 			return false;
3921 		break;
3922 	default:
3923 		break;
3924 	}
3925 
3926 	return true;
3927 }
3928 
3929 static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
3930 {
3931 	return 0;
3932 }
3933 
3934 static void svm_vcpu_after_set_cpuid(struct kvm_vcpu *vcpu)
3935 {
3936 	struct vcpu_svm *svm = to_svm(vcpu);
3937 	struct kvm_cpuid_entry2 *best;
3938 	struct kvm *kvm = vcpu->kvm;
3939 
3940 	vcpu->arch.xsaves_enabled = guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
3941 				    boot_cpu_has(X86_FEATURE_XSAVE) &&
3942 				    boot_cpu_has(X86_FEATURE_XSAVES);
3943 
3944 	/* Update nrips enabled cache */
3945 	svm->nrips_enabled = kvm_cpu_cap_has(X86_FEATURE_NRIPS) &&
3946 			     guest_cpuid_has(vcpu, X86_FEATURE_NRIPS);
3947 
3948 	svm->tsc_scaling_enabled = tsc_scaling && guest_cpuid_has(vcpu, X86_FEATURE_TSCRATEMSR);
3949 
3950 	svm_recalc_instruction_intercepts(vcpu, svm);
3951 
3952 	/* For sev guests, the memory encryption bit is not reserved in CR3.  */
3953 	if (sev_guest(vcpu->kvm)) {
3954 		best = kvm_find_cpuid_entry(vcpu, 0x8000001F, 0);
3955 		if (best)
3956 			vcpu->arch.reserved_gpa_bits &= ~(1UL << (best->ebx & 0x3f));
3957 	}
3958 
3959 	if (kvm_vcpu_apicv_active(vcpu)) {
3960 		/*
3961 		 * AVIC does not work with an x2APIC mode guest. If the X2APIC feature
3962 		 * is exposed to the guest, disable AVIC.
3963 		 */
3964 		if (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC))
3965 			kvm_set_apicv_inhibit(kvm, APICV_INHIBIT_REASON_X2APIC);
3966 
3967 		/*
3968 		 * Currently, AVIC does not work with nested virtualization.
3969 		 * So, we disable AVIC when cpuid for SVM is set in the L1 guest.
3970 		 */
3971 		if (nested && guest_cpuid_has(vcpu, X86_FEATURE_SVM))
3972 			kvm_set_apicv_inhibit(kvm, APICV_INHIBIT_REASON_NESTED);
3973 	}
3974 	init_vmcb_after_set_cpuid(vcpu);
3975 }
3976 
3977 static bool svm_has_wbinvd_exit(void)
3978 {
3979 	return true;
3980 }
3981 
3982 #define PRE_EX(exit)  { .exit_code = (exit), \
3983 			.stage = X86_ICPT_PRE_EXCEPT, }
3984 #define POST_EX(exit) { .exit_code = (exit), \
3985 			.stage = X86_ICPT_POST_EXCEPT, }
3986 #define POST_MEM(exit) { .exit_code = (exit), \
3987 			.stage = X86_ICPT_POST_MEMACCESS, }
3988 
3989 static const struct __x86_intercept {
3990 	u32 exit_code;
3991 	enum x86_intercept_stage stage;
3992 } x86_intercept_map[] = {
3993 	[x86_intercept_cr_read]		= POST_EX(SVM_EXIT_READ_CR0),
3994 	[x86_intercept_cr_write]	= POST_EX(SVM_EXIT_WRITE_CR0),
3995 	[x86_intercept_clts]		= POST_EX(SVM_EXIT_WRITE_CR0),
3996 	[x86_intercept_lmsw]		= POST_EX(SVM_EXIT_WRITE_CR0),
3997 	[x86_intercept_smsw]		= POST_EX(SVM_EXIT_READ_CR0),
3998 	[x86_intercept_dr_read]		= POST_EX(SVM_EXIT_READ_DR0),
3999 	[x86_intercept_dr_write]	= POST_EX(SVM_EXIT_WRITE_DR0),
4000 	[x86_intercept_sldt]		= POST_EX(SVM_EXIT_LDTR_READ),
4001 	[x86_intercept_str]		= POST_EX(SVM_EXIT_TR_READ),
4002 	[x86_intercept_lldt]		= POST_EX(SVM_EXIT_LDTR_WRITE),
4003 	[x86_intercept_ltr]		= POST_EX(SVM_EXIT_TR_WRITE),
4004 	[x86_intercept_sgdt]		= POST_EX(SVM_EXIT_GDTR_READ),
4005 	[x86_intercept_sidt]		= POST_EX(SVM_EXIT_IDTR_READ),
4006 	[x86_intercept_lgdt]		= POST_EX(SVM_EXIT_GDTR_WRITE),
4007 	[x86_intercept_lidt]		= POST_EX(SVM_EXIT_IDTR_WRITE),
4008 	[x86_intercept_vmrun]		= POST_EX(SVM_EXIT_VMRUN),
4009 	[x86_intercept_vmmcall]		= POST_EX(SVM_EXIT_VMMCALL),
4010 	[x86_intercept_vmload]		= POST_EX(SVM_EXIT_VMLOAD),
4011 	[x86_intercept_vmsave]		= POST_EX(SVM_EXIT_VMSAVE),
4012 	[x86_intercept_stgi]		= POST_EX(SVM_EXIT_STGI),
4013 	[x86_intercept_clgi]		= POST_EX(SVM_EXIT_CLGI),
4014 	[x86_intercept_skinit]		= POST_EX(SVM_EXIT_SKINIT),
4015 	[x86_intercept_invlpga]		= POST_EX(SVM_EXIT_INVLPGA),
4016 	[x86_intercept_rdtscp]		= POST_EX(SVM_EXIT_RDTSCP),
4017 	[x86_intercept_monitor]		= POST_MEM(SVM_EXIT_MONITOR),
4018 	[x86_intercept_mwait]		= POST_EX(SVM_EXIT_MWAIT),
4019 	[x86_intercept_invlpg]		= POST_EX(SVM_EXIT_INVLPG),
4020 	[x86_intercept_invd]		= POST_EX(SVM_EXIT_INVD),
4021 	[x86_intercept_wbinvd]		= POST_EX(SVM_EXIT_WBINVD),
4022 	[x86_intercept_wrmsr]		= POST_EX(SVM_EXIT_MSR),
4023 	[x86_intercept_rdtsc]		= POST_EX(SVM_EXIT_RDTSC),
4024 	[x86_intercept_rdmsr]		= POST_EX(SVM_EXIT_MSR),
4025 	[x86_intercept_rdpmc]		= POST_EX(SVM_EXIT_RDPMC),
4026 	[x86_intercept_cpuid]		= PRE_EX(SVM_EXIT_CPUID),
4027 	[x86_intercept_rsm]		= PRE_EX(SVM_EXIT_RSM),
4028 	[x86_intercept_pause]		= PRE_EX(SVM_EXIT_PAUSE),
4029 	[x86_intercept_pushf]		= PRE_EX(SVM_EXIT_PUSHF),
4030 	[x86_intercept_popf]		= PRE_EX(SVM_EXIT_POPF),
4031 	[x86_intercept_intn]		= PRE_EX(SVM_EXIT_SWINT),
4032 	[x86_intercept_iret]		= PRE_EX(SVM_EXIT_IRET),
4033 	[x86_intercept_icebp]		= PRE_EX(SVM_EXIT_ICEBP),
4034 	[x86_intercept_hlt]		= POST_EX(SVM_EXIT_HLT),
4035 	[x86_intercept_in]		= POST_EX(SVM_EXIT_IOIO),
4036 	[x86_intercept_ins]		= POST_EX(SVM_EXIT_IOIO),
4037 	[x86_intercept_out]		= POST_EX(SVM_EXIT_IOIO),
4038 	[x86_intercept_outs]		= POST_EX(SVM_EXIT_IOIO),
4039 	[x86_intercept_xsetbv]		= PRE_EX(SVM_EXIT_XSETBV),
4040 };
4041 
4042 #undef PRE_EX
4043 #undef POST_EX
4044 #undef POST_MEM
4045 
4046 static int svm_check_intercept(struct kvm_vcpu *vcpu,
4047 			       struct x86_instruction_info *info,
4048 			       enum x86_intercept_stage stage,
4049 			       struct x86_exception *exception)
4050 {
4051 	struct vcpu_svm *svm = to_svm(vcpu);
4052 	int vmexit, ret = X86EMUL_CONTINUE;
4053 	struct __x86_intercept icpt_info;
4054 	struct vmcb *vmcb = svm->vmcb;
4055 
4056 	if (info->intercept >= ARRAY_SIZE(x86_intercept_map))
4057 		goto out;
4058 
4059 	icpt_info = x86_intercept_map[info->intercept];
4060 
4061 	if (stage != icpt_info.stage)
4062 		goto out;
4063 
4064 	switch (icpt_info.exit_code) {
4065 	case SVM_EXIT_READ_CR0:
4066 		if (info->intercept == x86_intercept_cr_read)
4067 			icpt_info.exit_code += info->modrm_reg;
4068 		break;
4069 	case SVM_EXIT_WRITE_CR0: {
4070 		unsigned long cr0, val;
4071 
4072 		if (info->intercept == x86_intercept_cr_write)
4073 			icpt_info.exit_code += info->modrm_reg;
4074 
4075 		if (icpt_info.exit_code != SVM_EXIT_WRITE_CR0 ||
4076 		    info->intercept == x86_intercept_clts)
4077 			break;
4078 
4079 		if (!(vmcb12_is_intercept(&svm->nested.ctl,
4080 					INTERCEPT_SELECTIVE_CR0)))
4081 			break;
4082 
4083 		cr0 = vcpu->arch.cr0 & ~SVM_CR0_SELECTIVE_MASK;
4084 		val = info->src_val  & ~SVM_CR0_SELECTIVE_MASK;
4085 
4086 		if (info->intercept == x86_intercept_lmsw) {
4087 			cr0 &= 0xfUL;
4088 			val &= 0xfUL;
4089 			/* lmsw can't clear PE - catch this here */
4090 			if (cr0 & X86_CR0_PE)
4091 				val |= X86_CR0_PE;
4092 		}
4093 
4094 		if (cr0 ^ val)
4095 			icpt_info.exit_code = SVM_EXIT_CR0_SEL_WRITE;
4096 
4097 		break;
4098 	}
4099 	case SVM_EXIT_READ_DR0:
4100 	case SVM_EXIT_WRITE_DR0:
4101 		icpt_info.exit_code += info->modrm_reg;
4102 		break;
4103 	case SVM_EXIT_MSR:
4104 		if (info->intercept == x86_intercept_wrmsr)
4105 			vmcb->control.exit_info_1 = 1;
4106 		else
4107 			vmcb->control.exit_info_1 = 0;
4108 		break;
4109 	case SVM_EXIT_PAUSE:
4110 		/*
4111 		 * We get this for NOP only, but pause
4112 		 * is rep not, check this here
4113 		 */
4114 		if (info->rep_prefix != REPE_PREFIX)
4115 			goto out;
4116 		break;
4117 	case SVM_EXIT_IOIO: {
4118 		u64 exit_info;
4119 		u32 bytes;
4120 
4121 		if (info->intercept == x86_intercept_in ||
4122 		    info->intercept == x86_intercept_ins) {
4123 			exit_info = ((info->src_val & 0xffff) << 16) |
4124 				SVM_IOIO_TYPE_MASK;
4125 			bytes = info->dst_bytes;
4126 		} else {
4127 			exit_info = (info->dst_val & 0xffff) << 16;
4128 			bytes = info->src_bytes;
4129 		}
4130 
4131 		if (info->intercept == x86_intercept_outs ||
4132 		    info->intercept == x86_intercept_ins)
4133 			exit_info |= SVM_IOIO_STR_MASK;
4134 
4135 		if (info->rep_prefix)
4136 			exit_info |= SVM_IOIO_REP_MASK;
4137 
4138 		bytes = min(bytes, 4u);
4139 
4140 		exit_info |= bytes << SVM_IOIO_SIZE_SHIFT;
4141 
4142 		exit_info |= (u32)info->ad_bytes << (SVM_IOIO_ASIZE_SHIFT - 1);
4143 
4144 		vmcb->control.exit_info_1 = exit_info;
4145 		vmcb->control.exit_info_2 = info->next_rip;
4146 
4147 		break;
4148 	}
4149 	default:
4150 		break;
4151 	}
4152 
4153 	/* TODO: Advertise NRIPS to guest hypervisor unconditionally */
4154 	if (static_cpu_has(X86_FEATURE_NRIPS))
4155 		vmcb->control.next_rip  = info->next_rip;
4156 	vmcb->control.exit_code = icpt_info.exit_code;
4157 	vmexit = nested_svm_exit_handled(svm);
4158 
4159 	ret = (vmexit == NESTED_EXIT_DONE) ? X86EMUL_INTERCEPTED
4160 					   : X86EMUL_CONTINUE;
4161 
4162 out:
4163 	return ret;
4164 }
4165 
4166 static void svm_handle_exit_irqoff(struct kvm_vcpu *vcpu)
4167 {
4168 }
4169 
4170 static void svm_sched_in(struct kvm_vcpu *vcpu, int cpu)
4171 {
4172 	if (!kvm_pause_in_guest(vcpu->kvm))
4173 		shrink_ple_window(vcpu);
4174 }
4175 
4176 static void svm_setup_mce(struct kvm_vcpu *vcpu)
4177 {
4178 	/* [63:9] are reserved. */
4179 	vcpu->arch.mcg_cap &= 0x1ff;
4180 }
4181 
4182 bool svm_smi_blocked(struct kvm_vcpu *vcpu)
4183 {
4184 	struct vcpu_svm *svm = to_svm(vcpu);
4185 
4186 	/* Per APM Vol.2 15.22.2 "Response to SMI" */
4187 	if (!gif_set(svm))
4188 		return true;
4189 
4190 	return is_smm(vcpu);
4191 }
4192 
4193 static int svm_smi_allowed(struct kvm_vcpu *vcpu, bool for_injection)
4194 {
4195 	struct vcpu_svm *svm = to_svm(vcpu);
4196 	if (svm->nested.nested_run_pending)
4197 		return -EBUSY;
4198 
4199 	if (svm_smi_blocked(vcpu))
4200 		return 0;
4201 
4202 	/* An SMI must not be injected into L2 if it's supposed to VM-Exit.  */
4203 	if (for_injection && is_guest_mode(vcpu) && nested_exit_on_smi(svm))
4204 		return -EBUSY;
4205 
4206 	return 1;
4207 }
4208 
4209 static int svm_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
4210 {
4211 	struct vcpu_svm *svm = to_svm(vcpu);
4212 	struct kvm_host_map map_save;
4213 	int ret;
4214 
4215 	if (!is_guest_mode(vcpu))
4216 		return 0;
4217 
4218 	/* FED8h - SVM Guest */
4219 	put_smstate(u64, smstate, 0x7ed8, 1);
4220 	/* FEE0h - SVM Guest VMCB Physical Address */
4221 	put_smstate(u64, smstate, 0x7ee0, svm->nested.vmcb12_gpa);
4222 
4223 	svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
4224 	svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
4225 	svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
4226 
4227 	ret = nested_svm_vmexit(svm);
4228 	if (ret)
4229 		return ret;
4230 
4231 	/*
4232 	 * KVM uses VMCB01 to store L1 host state while L2 runs but
4233 	 * VMCB01 is going to be used during SMM and thus the state will
4234 	 * be lost. Temporary save non-VMLOAD/VMSAVE state to the host save
4235 	 * area pointed to by MSR_VM_HSAVE_PA. APM guarantees that the
4236 	 * format of the area is identical to guest save area offsetted
4237 	 * by 0x400 (matches the offset of 'struct vmcb_save_area'
4238 	 * within 'struct vmcb'). Note: HSAVE area may also be used by
4239 	 * L1 hypervisor to save additional host context (e.g. KVM does
4240 	 * that, see svm_prepare_switch_to_guest()) which must be
4241 	 * preserved.
4242 	 */
4243 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(svm->nested.hsave_msr),
4244 			 &map_save) == -EINVAL)
4245 		return 1;
4246 
4247 	BUILD_BUG_ON(offsetof(struct vmcb, save) != 0x400);
4248 
4249 	svm_copy_vmrun_state(map_save.hva + 0x400,
4250 			     &svm->vmcb01.ptr->save);
4251 
4252 	kvm_vcpu_unmap(vcpu, &map_save, true);
4253 	return 0;
4254 }
4255 
4256 static int svm_leave_smm(struct kvm_vcpu *vcpu, const char *smstate)
4257 {
4258 	struct vcpu_svm *svm = to_svm(vcpu);
4259 	struct kvm_host_map map, map_save;
4260 	u64 saved_efer, vmcb12_gpa;
4261 	struct vmcb *vmcb12;
4262 	int ret;
4263 
4264 	if (!guest_cpuid_has(vcpu, X86_FEATURE_LM))
4265 		return 0;
4266 
4267 	/* Non-zero if SMI arrived while vCPU was in guest mode. */
4268 	if (!GET_SMSTATE(u64, smstate, 0x7ed8))
4269 		return 0;
4270 
4271 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SVM))
4272 		return 1;
4273 
4274 	saved_efer = GET_SMSTATE(u64, smstate, 0x7ed0);
4275 	if (!(saved_efer & EFER_SVME))
4276 		return 1;
4277 
4278 	vmcb12_gpa = GET_SMSTATE(u64, smstate, 0x7ee0);
4279 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(vmcb12_gpa), &map) == -EINVAL)
4280 		return 1;
4281 
4282 	ret = 1;
4283 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(svm->nested.hsave_msr), &map_save) == -EINVAL)
4284 		goto unmap_map;
4285 
4286 	if (svm_allocate_nested(svm))
4287 		goto unmap_save;
4288 
4289 	/*
4290 	 * Restore L1 host state from L1 HSAVE area as VMCB01 was
4291 	 * used during SMM (see svm_enter_smm())
4292 	 */
4293 
4294 	svm_copy_vmrun_state(&svm->vmcb01.ptr->save, map_save.hva + 0x400);
4295 
4296 	/*
4297 	 * Enter the nested guest now
4298 	 */
4299 
4300 	vmcb_mark_all_dirty(svm->vmcb01.ptr);
4301 
4302 	vmcb12 = map.hva;
4303 	nested_copy_vmcb_control_to_cache(svm, &vmcb12->control);
4304 	nested_copy_vmcb_save_to_cache(svm, &vmcb12->save);
4305 	ret = enter_svm_guest_mode(vcpu, vmcb12_gpa, vmcb12, false);
4306 
4307 	if (ret)
4308 		goto unmap_save;
4309 
4310 	svm->nested.nested_run_pending = 1;
4311 
4312 unmap_save:
4313 	kvm_vcpu_unmap(vcpu, &map_save, true);
4314 unmap_map:
4315 	kvm_vcpu_unmap(vcpu, &map, true);
4316 	return ret;
4317 }
4318 
4319 static void svm_enable_smi_window(struct kvm_vcpu *vcpu)
4320 {
4321 	struct vcpu_svm *svm = to_svm(vcpu);
4322 
4323 	if (!gif_set(svm)) {
4324 		if (vgif_enabled(svm))
4325 			svm_set_intercept(svm, INTERCEPT_STGI);
4326 		/* STGI will cause a vm exit */
4327 	} else {
4328 		/* We must be in SMM; RSM will cause a vmexit anyway.  */
4329 	}
4330 }
4331 
4332 static bool svm_can_emulate_instruction(struct kvm_vcpu *vcpu, int emul_type,
4333 					void *insn, int insn_len)
4334 {
4335 	bool smep, smap, is_user;
4336 	unsigned long cr4;
4337 	u64 error_code;
4338 
4339 	/* Emulation is always possible when KVM has access to all guest state. */
4340 	if (!sev_guest(vcpu->kvm))
4341 		return true;
4342 
4343 	/* #UD and #GP should never be intercepted for SEV guests. */
4344 	WARN_ON_ONCE(emul_type & (EMULTYPE_TRAP_UD |
4345 				  EMULTYPE_TRAP_UD_FORCED |
4346 				  EMULTYPE_VMWARE_GP));
4347 
4348 	/*
4349 	 * Emulation is impossible for SEV-ES guests as KVM doesn't have access
4350 	 * to guest register state.
4351 	 */
4352 	if (sev_es_guest(vcpu->kvm))
4353 		return false;
4354 
4355 	/*
4356 	 * Emulation is possible if the instruction is already decoded, e.g.
4357 	 * when completing I/O after returning from userspace.
4358 	 */
4359 	if (emul_type & EMULTYPE_NO_DECODE)
4360 		return true;
4361 
4362 	/*
4363 	 * Emulation is possible for SEV guests if and only if a prefilled
4364 	 * buffer containing the bytes of the intercepted instruction is
4365 	 * available. SEV guest memory is encrypted with a guest specific key
4366 	 * and cannot be decrypted by KVM, i.e. KVM would read cyphertext and
4367 	 * decode garbage.
4368 	 *
4369 	 * Inject #UD if KVM reached this point without an instruction buffer.
4370 	 * In practice, this path should never be hit by a well-behaved guest,
4371 	 * e.g. KVM doesn't intercept #UD or #GP for SEV guests, but this path
4372 	 * is still theoretically reachable, e.g. via unaccelerated fault-like
4373 	 * AVIC access, and needs to be handled by KVM to avoid putting the
4374 	 * guest into an infinite loop.   Injecting #UD is somewhat arbitrary,
4375 	 * but its the least awful option given lack of insight into the guest.
4376 	 */
4377 	if (unlikely(!insn)) {
4378 		kvm_queue_exception(vcpu, UD_VECTOR);
4379 		return false;
4380 	}
4381 
4382 	/*
4383 	 * Emulate for SEV guests if the insn buffer is not empty.  The buffer
4384 	 * will be empty if the DecodeAssist microcode cannot fetch bytes for
4385 	 * the faulting instruction because the code fetch itself faulted, e.g.
4386 	 * the guest attempted to fetch from emulated MMIO or a guest page
4387 	 * table used to translate CS:RIP resides in emulated MMIO.
4388 	 */
4389 	if (likely(insn_len))
4390 		return true;
4391 
4392 	/*
4393 	 * Detect and workaround Errata 1096 Fam_17h_00_0Fh.
4394 	 *
4395 	 * Errata:
4396 	 * When CPU raises #NPF on guest data access and vCPU CR4.SMAP=1, it is
4397 	 * possible that CPU microcode implementing DecodeAssist will fail to
4398 	 * read guest memory at CS:RIP and vmcb.GuestIntrBytes will incorrectly
4399 	 * be '0'.  This happens because microcode reads CS:RIP using a _data_
4400 	 * loap uop with CPL=0 privileges.  If the load hits a SMAP #PF, ucode
4401 	 * gives up and does not fill the instruction bytes buffer.
4402 	 *
4403 	 * As above, KVM reaches this point iff the VM is an SEV guest, the CPU
4404 	 * supports DecodeAssist, a #NPF was raised, KVM's page fault handler
4405 	 * triggered emulation (e.g. for MMIO), and the CPU returned 0 in the
4406 	 * GuestIntrBytes field of the VMCB.
4407 	 *
4408 	 * This does _not_ mean that the erratum has been encountered, as the
4409 	 * DecodeAssist will also fail if the load for CS:RIP hits a legitimate
4410 	 * #PF, e.g. if the guest attempt to execute from emulated MMIO and
4411 	 * encountered a reserved/not-present #PF.
4412 	 *
4413 	 * To hit the erratum, the following conditions must be true:
4414 	 *    1. CR4.SMAP=1 (obviously).
4415 	 *    2. CR4.SMEP=0 || CPL=3.  If SMEP=1 and CPL<3, the erratum cannot
4416 	 *       have been hit as the guest would have encountered a SMEP
4417 	 *       violation #PF, not a #NPF.
4418 	 *    3. The #NPF is not due to a code fetch, in which case failure to
4419 	 *       retrieve the instruction bytes is legitimate (see abvoe).
4420 	 *
4421 	 * In addition, don't apply the erratum workaround if the #NPF occurred
4422 	 * while translating guest page tables (see below).
4423 	 */
4424 	error_code = to_svm(vcpu)->vmcb->control.exit_info_1;
4425 	if (error_code & (PFERR_GUEST_PAGE_MASK | PFERR_FETCH_MASK))
4426 		goto resume_guest;
4427 
4428 	cr4 = kvm_read_cr4(vcpu);
4429 	smep = cr4 & X86_CR4_SMEP;
4430 	smap = cr4 & X86_CR4_SMAP;
4431 	is_user = svm_get_cpl(vcpu) == 3;
4432 	if (smap && (!smep || is_user)) {
4433 		pr_err_ratelimited("KVM: SEV Guest triggered AMD Erratum 1096\n");
4434 
4435 		/*
4436 		 * If the fault occurred in userspace, arbitrarily inject #GP
4437 		 * to avoid killing the guest and to hopefully avoid confusing
4438 		 * the guest kernel too much, e.g. injecting #PF would not be
4439 		 * coherent with respect to the guest's page tables.  Request
4440 		 * triple fault if the fault occurred in the kernel as there's
4441 		 * no fault that KVM can inject without confusing the guest.
4442 		 * In practice, the triple fault is moot as no sane SEV kernel
4443 		 * will execute from user memory while also running with SMAP=1.
4444 		 */
4445 		if (is_user)
4446 			kvm_inject_gp(vcpu, 0);
4447 		else
4448 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4449 	}
4450 
4451 resume_guest:
4452 	/*
4453 	 * If the erratum was not hit, simply resume the guest and let it fault
4454 	 * again.  While awful, e.g. the vCPU may get stuck in an infinite loop
4455 	 * if the fault is at CPL=0, it's the lesser of all evils.  Exiting to
4456 	 * userspace will kill the guest, and letting the emulator read garbage
4457 	 * will yield random behavior and potentially corrupt the guest.
4458 	 *
4459 	 * Simply resuming the guest is technically not a violation of the SEV
4460 	 * architecture.  AMD's APM states that all code fetches and page table
4461 	 * accesses for SEV guest are encrypted, regardless of the C-Bit.  The
4462 	 * APM also states that encrypted accesses to MMIO are "ignored", but
4463 	 * doesn't explicitly define "ignored", i.e. doing nothing and letting
4464 	 * the guest spin is technically "ignoring" the access.
4465 	 */
4466 	return false;
4467 }
4468 
4469 static bool svm_apic_init_signal_blocked(struct kvm_vcpu *vcpu)
4470 {
4471 	struct vcpu_svm *svm = to_svm(vcpu);
4472 
4473 	/*
4474 	 * TODO: Last condition latch INIT signals on vCPU when
4475 	 * vCPU is in guest-mode and vmcb12 defines intercept on INIT.
4476 	 * To properly emulate the INIT intercept,
4477 	 * svm_check_nested_events() should call nested_svm_vmexit()
4478 	 * if an INIT signal is pending.
4479 	 */
4480 	return !gif_set(svm) ||
4481 		   (vmcb_is_intercept(&svm->vmcb->control, INTERCEPT_INIT));
4482 }
4483 
4484 static void svm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
4485 {
4486 	if (!sev_es_guest(vcpu->kvm))
4487 		return kvm_vcpu_deliver_sipi_vector(vcpu, vector);
4488 
4489 	sev_vcpu_deliver_sipi_vector(vcpu, vector);
4490 }
4491 
4492 static void svm_vm_destroy(struct kvm *kvm)
4493 {
4494 	avic_vm_destroy(kvm);
4495 	sev_vm_destroy(kvm);
4496 }
4497 
4498 static int svm_vm_init(struct kvm *kvm)
4499 {
4500 	if (!pause_filter_count || !pause_filter_thresh)
4501 		kvm->arch.pause_in_guest = true;
4502 
4503 	if (enable_apicv) {
4504 		int ret = avic_vm_init(kvm);
4505 		if (ret)
4506 			return ret;
4507 	}
4508 
4509 	return 0;
4510 }
4511 
4512 static struct kvm_x86_ops svm_x86_ops __initdata = {
4513 	.name = "kvm_amd",
4514 
4515 	.hardware_unsetup = svm_hardware_unsetup,
4516 	.hardware_enable = svm_hardware_enable,
4517 	.hardware_disable = svm_hardware_disable,
4518 	.has_emulated_msr = svm_has_emulated_msr,
4519 
4520 	.vcpu_create = svm_vcpu_create,
4521 	.vcpu_free = svm_vcpu_free,
4522 	.vcpu_reset = svm_vcpu_reset,
4523 
4524 	.vm_size = sizeof(struct kvm_svm),
4525 	.vm_init = svm_vm_init,
4526 	.vm_destroy = svm_vm_destroy,
4527 
4528 	.prepare_switch_to_guest = svm_prepare_switch_to_guest,
4529 	.vcpu_load = svm_vcpu_load,
4530 	.vcpu_put = svm_vcpu_put,
4531 	.vcpu_blocking = avic_vcpu_blocking,
4532 	.vcpu_unblocking = avic_vcpu_unblocking,
4533 
4534 	.update_exception_bitmap = svm_update_exception_bitmap,
4535 	.get_msr_feature = svm_get_msr_feature,
4536 	.get_msr = svm_get_msr,
4537 	.set_msr = svm_set_msr,
4538 	.get_segment_base = svm_get_segment_base,
4539 	.get_segment = svm_get_segment,
4540 	.set_segment = svm_set_segment,
4541 	.get_cpl = svm_get_cpl,
4542 	.get_cs_db_l_bits = svm_get_cs_db_l_bits,
4543 	.set_cr0 = svm_set_cr0,
4544 	.post_set_cr3 = sev_post_set_cr3,
4545 	.is_valid_cr4 = svm_is_valid_cr4,
4546 	.set_cr4 = svm_set_cr4,
4547 	.set_efer = svm_set_efer,
4548 	.get_idt = svm_get_idt,
4549 	.set_idt = svm_set_idt,
4550 	.get_gdt = svm_get_gdt,
4551 	.set_gdt = svm_set_gdt,
4552 	.set_dr7 = svm_set_dr7,
4553 	.sync_dirty_debug_regs = svm_sync_dirty_debug_regs,
4554 	.cache_reg = svm_cache_reg,
4555 	.get_rflags = svm_get_rflags,
4556 	.set_rflags = svm_set_rflags,
4557 	.get_if_flag = svm_get_if_flag,
4558 
4559 	.flush_tlb_all = svm_flush_tlb_current,
4560 	.flush_tlb_current = svm_flush_tlb_current,
4561 	.flush_tlb_gva = svm_flush_tlb_gva,
4562 	.flush_tlb_guest = svm_flush_tlb_current,
4563 
4564 	.vcpu_pre_run = svm_vcpu_pre_run,
4565 	.vcpu_run = svm_vcpu_run,
4566 	.handle_exit = svm_handle_exit,
4567 	.skip_emulated_instruction = svm_skip_emulated_instruction,
4568 	.update_emulated_instruction = NULL,
4569 	.set_interrupt_shadow = svm_set_interrupt_shadow,
4570 	.get_interrupt_shadow = svm_get_interrupt_shadow,
4571 	.patch_hypercall = svm_patch_hypercall,
4572 	.inject_irq = svm_inject_irq,
4573 	.inject_nmi = svm_inject_nmi,
4574 	.queue_exception = svm_queue_exception,
4575 	.cancel_injection = svm_cancel_injection,
4576 	.interrupt_allowed = svm_interrupt_allowed,
4577 	.nmi_allowed = svm_nmi_allowed,
4578 	.get_nmi_mask = svm_get_nmi_mask,
4579 	.set_nmi_mask = svm_set_nmi_mask,
4580 	.enable_nmi_window = svm_enable_nmi_window,
4581 	.enable_irq_window = svm_enable_irq_window,
4582 	.update_cr8_intercept = svm_update_cr8_intercept,
4583 	.refresh_apicv_exec_ctrl = avic_refresh_apicv_exec_ctrl,
4584 	.check_apicv_inhibit_reasons = avic_check_apicv_inhibit_reasons,
4585 	.apicv_post_state_restore = avic_apicv_post_state_restore,
4586 
4587 	.get_mt_mask = svm_get_mt_mask,
4588 	.get_exit_info = svm_get_exit_info,
4589 
4590 	.vcpu_after_set_cpuid = svm_vcpu_after_set_cpuid,
4591 
4592 	.has_wbinvd_exit = svm_has_wbinvd_exit,
4593 
4594 	.get_l2_tsc_offset = svm_get_l2_tsc_offset,
4595 	.get_l2_tsc_multiplier = svm_get_l2_tsc_multiplier,
4596 	.write_tsc_offset = svm_write_tsc_offset,
4597 	.write_tsc_multiplier = svm_write_tsc_multiplier,
4598 
4599 	.load_mmu_pgd = svm_load_mmu_pgd,
4600 
4601 	.check_intercept = svm_check_intercept,
4602 	.handle_exit_irqoff = svm_handle_exit_irqoff,
4603 
4604 	.request_immediate_exit = __kvm_request_immediate_exit,
4605 
4606 	.sched_in = svm_sched_in,
4607 
4608 	.pmu_ops = &amd_pmu_ops,
4609 	.nested_ops = &svm_nested_ops,
4610 
4611 	.deliver_interrupt = svm_deliver_interrupt,
4612 	.pi_update_irte = avic_pi_update_irte,
4613 	.setup_mce = svm_setup_mce,
4614 
4615 	.smi_allowed = svm_smi_allowed,
4616 	.enter_smm = svm_enter_smm,
4617 	.leave_smm = svm_leave_smm,
4618 	.enable_smi_window = svm_enable_smi_window,
4619 
4620 	.mem_enc_ioctl = sev_mem_enc_ioctl,
4621 	.mem_enc_register_region = sev_mem_enc_register_region,
4622 	.mem_enc_unregister_region = sev_mem_enc_unregister_region,
4623 	.guest_memory_reclaimed = sev_guest_memory_reclaimed,
4624 
4625 	.vm_copy_enc_context_from = sev_vm_copy_enc_context_from,
4626 	.vm_move_enc_context_from = sev_vm_move_enc_context_from,
4627 
4628 	.can_emulate_instruction = svm_can_emulate_instruction,
4629 
4630 	.apic_init_signal_blocked = svm_apic_init_signal_blocked,
4631 
4632 	.msr_filter_changed = svm_msr_filter_changed,
4633 	.complete_emulated_msr = svm_complete_emulated_msr,
4634 
4635 	.vcpu_deliver_sipi_vector = svm_vcpu_deliver_sipi_vector,
4636 };
4637 
4638 /*
4639  * The default MMIO mask is a single bit (excluding the present bit),
4640  * which could conflict with the memory encryption bit. Check for
4641  * memory encryption support and override the default MMIO mask if
4642  * memory encryption is enabled.
4643  */
4644 static __init void svm_adjust_mmio_mask(void)
4645 {
4646 	unsigned int enc_bit, mask_bit;
4647 	u64 msr, mask;
4648 
4649 	/* If there is no memory encryption support, use existing mask */
4650 	if (cpuid_eax(0x80000000) < 0x8000001f)
4651 		return;
4652 
4653 	/* If memory encryption is not enabled, use existing mask */
4654 	rdmsrl(MSR_AMD64_SYSCFG, msr);
4655 	if (!(msr & MSR_AMD64_SYSCFG_MEM_ENCRYPT))
4656 		return;
4657 
4658 	enc_bit = cpuid_ebx(0x8000001f) & 0x3f;
4659 	mask_bit = boot_cpu_data.x86_phys_bits;
4660 
4661 	/* Increment the mask bit if it is the same as the encryption bit */
4662 	if (enc_bit == mask_bit)
4663 		mask_bit++;
4664 
4665 	/*
4666 	 * If the mask bit location is below 52, then some bits above the
4667 	 * physical addressing limit will always be reserved, so use the
4668 	 * rsvd_bits() function to generate the mask. This mask, along with
4669 	 * the present bit, will be used to generate a page fault with
4670 	 * PFER.RSV = 1.
4671 	 *
4672 	 * If the mask bit location is 52 (or above), then clear the mask.
4673 	 */
4674 	mask = (mask_bit < 52) ? rsvd_bits(mask_bit, 51) | PT_PRESENT_MASK : 0;
4675 
4676 	kvm_mmu_set_mmio_spte_mask(mask, mask, PT_WRITABLE_MASK | PT_USER_MASK);
4677 }
4678 
4679 static __init void svm_set_cpu_caps(void)
4680 {
4681 	kvm_set_cpu_caps();
4682 
4683 	supported_xss = 0;
4684 
4685 	/* CPUID 0x80000001 and 0x8000000A (SVM features) */
4686 	if (nested) {
4687 		kvm_cpu_cap_set(X86_FEATURE_SVM);
4688 		kvm_cpu_cap_set(X86_FEATURE_VMCBCLEAN);
4689 
4690 		if (nrips)
4691 			kvm_cpu_cap_set(X86_FEATURE_NRIPS);
4692 
4693 		if (npt_enabled)
4694 			kvm_cpu_cap_set(X86_FEATURE_NPT);
4695 
4696 		if (tsc_scaling)
4697 			kvm_cpu_cap_set(X86_FEATURE_TSCRATEMSR);
4698 
4699 		/* Nested VM can receive #VMEXIT instead of triggering #GP */
4700 		kvm_cpu_cap_set(X86_FEATURE_SVME_ADDR_CHK);
4701 	}
4702 
4703 	/* CPUID 0x80000008 */
4704 	if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD) ||
4705 	    boot_cpu_has(X86_FEATURE_AMD_SSBD))
4706 		kvm_cpu_cap_set(X86_FEATURE_VIRT_SSBD);
4707 
4708 	/* AMD PMU PERFCTR_CORE CPUID */
4709 	if (enable_pmu && boot_cpu_has(X86_FEATURE_PERFCTR_CORE))
4710 		kvm_cpu_cap_set(X86_FEATURE_PERFCTR_CORE);
4711 
4712 	/* CPUID 0x8000001F (SME/SEV features) */
4713 	sev_set_cpu_caps();
4714 }
4715 
4716 static __init int svm_hardware_setup(void)
4717 {
4718 	int cpu;
4719 	struct page *iopm_pages;
4720 	void *iopm_va;
4721 	int r;
4722 	unsigned int order = get_order(IOPM_SIZE);
4723 
4724 	/*
4725 	 * NX is required for shadow paging and for NPT if the NX huge pages
4726 	 * mitigation is enabled.
4727 	 */
4728 	if (!boot_cpu_has(X86_FEATURE_NX)) {
4729 		pr_err_ratelimited("NX (Execute Disable) not supported\n");
4730 		return -EOPNOTSUPP;
4731 	}
4732 	kvm_enable_efer_bits(EFER_NX);
4733 
4734 	iopm_pages = alloc_pages(GFP_KERNEL, order);
4735 
4736 	if (!iopm_pages)
4737 		return -ENOMEM;
4738 
4739 	iopm_va = page_address(iopm_pages);
4740 	memset(iopm_va, 0xff, PAGE_SIZE * (1 << order));
4741 	iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
4742 
4743 	init_msrpm_offsets();
4744 
4745 	supported_xcr0 &= ~(XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR);
4746 
4747 	if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
4748 		kvm_enable_efer_bits(EFER_FFXSR);
4749 
4750 	if (tsc_scaling) {
4751 		if (!boot_cpu_has(X86_FEATURE_TSCRATEMSR)) {
4752 			tsc_scaling = false;
4753 		} else {
4754 			pr_info("TSC scaling supported\n");
4755 			kvm_has_tsc_control = true;
4756 		}
4757 	}
4758 	kvm_max_tsc_scaling_ratio = SVM_TSC_RATIO_MAX;
4759 	kvm_tsc_scaling_ratio_frac_bits = 32;
4760 
4761 	tsc_aux_uret_slot = kvm_add_user_return_msr(MSR_TSC_AUX);
4762 
4763 	/* Check for pause filtering support */
4764 	if (!boot_cpu_has(X86_FEATURE_PAUSEFILTER)) {
4765 		pause_filter_count = 0;
4766 		pause_filter_thresh = 0;
4767 	} else if (!boot_cpu_has(X86_FEATURE_PFTHRESHOLD)) {
4768 		pause_filter_thresh = 0;
4769 	}
4770 
4771 	if (nested) {
4772 		printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
4773 		kvm_enable_efer_bits(EFER_SVME | EFER_LMSLE);
4774 	}
4775 
4776 	/*
4777 	 * KVM's MMU doesn't support using 2-level paging for itself, and thus
4778 	 * NPT isn't supported if the host is using 2-level paging since host
4779 	 * CR4 is unchanged on VMRUN.
4780 	 */
4781 	if (!IS_ENABLED(CONFIG_X86_64) && !IS_ENABLED(CONFIG_X86_PAE))
4782 		npt_enabled = false;
4783 
4784 	if (!boot_cpu_has(X86_FEATURE_NPT))
4785 		npt_enabled = false;
4786 
4787 	/* Force VM NPT level equal to the host's paging level */
4788 	kvm_configure_mmu(npt_enabled, get_npt_level(),
4789 			  get_npt_level(), PG_LEVEL_1G);
4790 	pr_info("kvm: Nested Paging %sabled\n", npt_enabled ? "en" : "dis");
4791 
4792 	/* Note, SEV setup consumes npt_enabled. */
4793 	sev_hardware_setup();
4794 
4795 	svm_hv_hardware_setup();
4796 
4797 	svm_adjust_mmio_mask();
4798 
4799 	for_each_possible_cpu(cpu) {
4800 		r = svm_cpu_init(cpu);
4801 		if (r)
4802 			goto err;
4803 	}
4804 
4805 	if (nrips) {
4806 		if (!boot_cpu_has(X86_FEATURE_NRIPS))
4807 			nrips = false;
4808 	}
4809 
4810 	enable_apicv = avic = avic && npt_enabled && boot_cpu_has(X86_FEATURE_AVIC);
4811 
4812 	if (enable_apicv) {
4813 		pr_info("AVIC enabled\n");
4814 
4815 		amd_iommu_register_ga_log_notifier(&avic_ga_log_notifier);
4816 	} else {
4817 		svm_x86_ops.vcpu_blocking = NULL;
4818 		svm_x86_ops.vcpu_unblocking = NULL;
4819 	}
4820 
4821 	if (vls) {
4822 		if (!npt_enabled ||
4823 		    !boot_cpu_has(X86_FEATURE_V_VMSAVE_VMLOAD) ||
4824 		    !IS_ENABLED(CONFIG_X86_64)) {
4825 			vls = false;
4826 		} else {
4827 			pr_info("Virtual VMLOAD VMSAVE supported\n");
4828 		}
4829 	}
4830 
4831 	if (boot_cpu_has(X86_FEATURE_SVME_ADDR_CHK))
4832 		svm_gp_erratum_intercept = false;
4833 
4834 	if (vgif) {
4835 		if (!boot_cpu_has(X86_FEATURE_VGIF))
4836 			vgif = false;
4837 		else
4838 			pr_info("Virtual GIF supported\n");
4839 	}
4840 
4841 	if (lbrv) {
4842 		if (!boot_cpu_has(X86_FEATURE_LBRV))
4843 			lbrv = false;
4844 		else
4845 			pr_info("LBR virtualization supported\n");
4846 	}
4847 
4848 	if (!enable_pmu)
4849 		pr_info("PMU virtualization is disabled\n");
4850 
4851 	svm_set_cpu_caps();
4852 
4853 	/*
4854 	 * It seems that on AMD processors PTE's accessed bit is
4855 	 * being set by the CPU hardware before the NPF vmexit.
4856 	 * This is not expected behaviour and our tests fail because
4857 	 * of it.
4858 	 * A workaround here is to disable support for
4859 	 * GUEST_MAXPHYADDR < HOST_MAXPHYADDR if NPT is enabled.
4860 	 * In this case userspace can know if there is support using
4861 	 * KVM_CAP_SMALLER_MAXPHYADDR extension and decide how to handle
4862 	 * it
4863 	 * If future AMD CPU models change the behaviour described above,
4864 	 * this variable can be changed accordingly
4865 	 */
4866 	allow_smaller_maxphyaddr = !npt_enabled;
4867 
4868 	return 0;
4869 
4870 err:
4871 	svm_hardware_unsetup();
4872 	return r;
4873 }
4874 
4875 
4876 static struct kvm_x86_init_ops svm_init_ops __initdata = {
4877 	.cpu_has_kvm_support = has_svm,
4878 	.disabled_by_bios = is_disabled,
4879 	.hardware_setup = svm_hardware_setup,
4880 	.check_processor_compatibility = svm_check_processor_compat,
4881 
4882 	.runtime_ops = &svm_x86_ops,
4883 };
4884 
4885 static int __init svm_init(void)
4886 {
4887 	__unused_size_checks();
4888 
4889 	return kvm_init(&svm_init_ops, sizeof(struct vcpu_svm),
4890 			__alignof__(struct vcpu_svm), THIS_MODULE);
4891 }
4892 
4893 static void __exit svm_exit(void)
4894 {
4895 	kvm_exit();
4896 }
4897 
4898 module_init(svm_init)
4899 module_exit(svm_exit)
4900