xref: /openbmc/linux/arch/x86/kvm/svm/svm.c (revision 2f4503f9)
1 #define pr_fmt(fmt) KBUILD_MODNAME ": " 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 "smm.h"
10 #include "cpuid.h"
11 #include "pmu.h"
12 
13 #include <linux/module.h>
14 #include <linux/mod_devicetable.h>
15 #include <linux/kernel.h>
16 #include <linux/vmalloc.h>
17 #include <linux/highmem.h>
18 #include <linux/amd-iommu.h>
19 #include <linux/sched.h>
20 #include <linux/trace_events.h>
21 #include <linux/slab.h>
22 #include <linux/hashtable.h>
23 #include <linux/objtool.h>
24 #include <linux/psp-sev.h>
25 #include <linux/file.h>
26 #include <linux/pagemap.h>
27 #include <linux/swap.h>
28 #include <linux/rwsem.h>
29 #include <linux/cc_platform.h>
30 #include <linux/smp.h>
31 
32 #include <asm/apic.h>
33 #include <asm/perf_event.h>
34 #include <asm/tlbflush.h>
35 #include <asm/desc.h>
36 #include <asm/debugreg.h>
37 #include <asm/kvm_para.h>
38 #include <asm/irq_remapping.h>
39 #include <asm/spec-ctrl.h>
40 #include <asm/cpu_device_id.h>
41 #include <asm/traps.h>
42 #include <asm/fpu/api.h>
43 
44 #include <asm/virtext.h>
45 
46 #include <trace/events/ipi.h>
47 
48 #include "trace.h"
49 
50 #include "svm.h"
51 #include "svm_ops.h"
52 
53 #include "kvm_onhyperv.h"
54 #include "svm_onhyperv.h"
55 
56 MODULE_AUTHOR("Qumranet");
57 MODULE_LICENSE("GPL");
58 
59 #ifdef MODULE
60 static const struct x86_cpu_id svm_cpu_id[] = {
61 	X86_MATCH_FEATURE(X86_FEATURE_SVM, NULL),
62 	{}
63 };
64 MODULE_DEVICE_TABLE(x86cpu, svm_cpu_id);
65 #endif
66 
67 #define SEG_TYPE_LDT 2
68 #define SEG_TYPE_BUSY_TSS16 3
69 
70 static bool erratum_383_found __read_mostly;
71 
72 u32 msrpm_offsets[MSRPM_OFFSETS] __read_mostly;
73 
74 /*
75  * Set osvw_len to higher value when updated Revision Guides
76  * are published and we know what the new status bits are
77  */
78 static uint64_t osvw_len = 4, osvw_status;
79 
80 static DEFINE_PER_CPU(u64, current_tsc_ratio);
81 
82 #define X2APIC_MSR(x)	(APIC_BASE_MSR + (x >> 4))
83 
84 static const struct svm_direct_access_msrs {
85 	u32 index;   /* Index of the MSR */
86 	bool always; /* True if intercept is initially cleared */
87 } direct_access_msrs[MAX_DIRECT_ACCESS_MSRS] = {
88 	{ .index = MSR_STAR,				.always = true  },
89 	{ .index = MSR_IA32_SYSENTER_CS,		.always = true  },
90 	{ .index = MSR_IA32_SYSENTER_EIP,		.always = false },
91 	{ .index = MSR_IA32_SYSENTER_ESP,		.always = false },
92 #ifdef CONFIG_X86_64
93 	{ .index = MSR_GS_BASE,				.always = true  },
94 	{ .index = MSR_FS_BASE,				.always = true  },
95 	{ .index = MSR_KERNEL_GS_BASE,			.always = true  },
96 	{ .index = MSR_LSTAR,				.always = true  },
97 	{ .index = MSR_CSTAR,				.always = true  },
98 	{ .index = MSR_SYSCALL_MASK,			.always = true  },
99 #endif
100 	{ .index = MSR_IA32_SPEC_CTRL,			.always = false },
101 	{ .index = MSR_IA32_PRED_CMD,			.always = false },
102 	{ .index = MSR_IA32_FLUSH_CMD,			.always = false },
103 	{ .index = MSR_IA32_LASTBRANCHFROMIP,		.always = false },
104 	{ .index = MSR_IA32_LASTBRANCHTOIP,		.always = false },
105 	{ .index = MSR_IA32_LASTINTFROMIP,		.always = false },
106 	{ .index = MSR_IA32_LASTINTTOIP,		.always = false },
107 	{ .index = MSR_EFER,				.always = false },
108 	{ .index = MSR_IA32_CR_PAT,			.always = false },
109 	{ .index = MSR_AMD64_SEV_ES_GHCB,		.always = true  },
110 	{ .index = MSR_TSC_AUX,				.always = false },
111 	{ .index = X2APIC_MSR(APIC_ID),			.always = false },
112 	{ .index = X2APIC_MSR(APIC_LVR),		.always = false },
113 	{ .index = X2APIC_MSR(APIC_TASKPRI),		.always = false },
114 	{ .index = X2APIC_MSR(APIC_ARBPRI),		.always = false },
115 	{ .index = X2APIC_MSR(APIC_PROCPRI),		.always = false },
116 	{ .index = X2APIC_MSR(APIC_EOI),		.always = false },
117 	{ .index = X2APIC_MSR(APIC_RRR),		.always = false },
118 	{ .index = X2APIC_MSR(APIC_LDR),		.always = false },
119 	{ .index = X2APIC_MSR(APIC_DFR),		.always = false },
120 	{ .index = X2APIC_MSR(APIC_SPIV),		.always = false },
121 	{ .index = X2APIC_MSR(APIC_ISR),		.always = false },
122 	{ .index = X2APIC_MSR(APIC_TMR),		.always = false },
123 	{ .index = X2APIC_MSR(APIC_IRR),		.always = false },
124 	{ .index = X2APIC_MSR(APIC_ESR),		.always = false },
125 	{ .index = X2APIC_MSR(APIC_ICR),		.always = false },
126 	{ .index = X2APIC_MSR(APIC_ICR2),		.always = false },
127 
128 	/*
129 	 * Note:
130 	 * AMD does not virtualize APIC TSC-deadline timer mode, but it is
131 	 * emulated by KVM. When setting APIC LVTT (0x832) register bit 18,
132 	 * the AVIC hardware would generate GP fault. Therefore, always
133 	 * intercept the MSR 0x832, and do not setup direct_access_msr.
134 	 */
135 	{ .index = X2APIC_MSR(APIC_LVTTHMR),		.always = false },
136 	{ .index = X2APIC_MSR(APIC_LVTPC),		.always = false },
137 	{ .index = X2APIC_MSR(APIC_LVT0),		.always = false },
138 	{ .index = X2APIC_MSR(APIC_LVT1),		.always = false },
139 	{ .index = X2APIC_MSR(APIC_LVTERR),		.always = false },
140 	{ .index = X2APIC_MSR(APIC_TMICT),		.always = false },
141 	{ .index = X2APIC_MSR(APIC_TMCCT),		.always = false },
142 	{ .index = X2APIC_MSR(APIC_TDCR),		.always = false },
143 	{ .index = MSR_INVALID,				.always = false },
144 };
145 
146 /*
147  * These 2 parameters are used to config the controls for Pause-Loop Exiting:
148  * pause_filter_count: On processors that support Pause filtering(indicated
149  *	by CPUID Fn8000_000A_EDX), the VMCB provides a 16 bit pause filter
150  *	count value. On VMRUN this value is loaded into an internal counter.
151  *	Each time a pause instruction is executed, this counter is decremented
152  *	until it reaches zero at which time a #VMEXIT is generated if pause
153  *	intercept is enabled. Refer to  AMD APM Vol 2 Section 15.14.4 Pause
154  *	Intercept Filtering for more details.
155  *	This also indicate if ple logic enabled.
156  *
157  * pause_filter_thresh: In addition, some processor families support advanced
158  *	pause filtering (indicated by CPUID Fn8000_000A_EDX) upper bound on
159  *	the amount of time a guest is allowed to execute in a pause loop.
160  *	In this mode, a 16-bit pause filter threshold field is added in the
161  *	VMCB. The threshold value is a cycle count that is used to reset the
162  *	pause counter. As with simple pause filtering, VMRUN loads the pause
163  *	count value from VMCB into an internal counter. Then, on each pause
164  *	instruction the hardware checks the elapsed number of cycles since
165  *	the most recent pause instruction against the pause filter threshold.
166  *	If the elapsed cycle count is greater than the pause filter threshold,
167  *	then the internal pause count is reloaded from the VMCB and execution
168  *	continues. If the elapsed cycle count is less than the pause filter
169  *	threshold, then the internal pause count is decremented. If the count
170  *	value is less than zero and PAUSE intercept is enabled, a #VMEXIT is
171  *	triggered. If advanced pause filtering is supported and pause filter
172  *	threshold field is set to zero, the filter will operate in the simpler,
173  *	count only mode.
174  */
175 
176 static unsigned short pause_filter_thresh = KVM_DEFAULT_PLE_GAP;
177 module_param(pause_filter_thresh, ushort, 0444);
178 
179 static unsigned short pause_filter_count = KVM_SVM_DEFAULT_PLE_WINDOW;
180 module_param(pause_filter_count, ushort, 0444);
181 
182 /* Default doubles per-vcpu window every exit. */
183 static unsigned short pause_filter_count_grow = KVM_DEFAULT_PLE_WINDOW_GROW;
184 module_param(pause_filter_count_grow, ushort, 0444);
185 
186 /* Default resets per-vcpu window every exit to pause_filter_count. */
187 static unsigned short pause_filter_count_shrink = KVM_DEFAULT_PLE_WINDOW_SHRINK;
188 module_param(pause_filter_count_shrink, ushort, 0444);
189 
190 /* Default is to compute the maximum so we can never overflow. */
191 static unsigned short pause_filter_count_max = KVM_SVM_DEFAULT_PLE_WINDOW_MAX;
192 module_param(pause_filter_count_max, ushort, 0444);
193 
194 /*
195  * Use nested page tables by default.  Note, NPT may get forced off by
196  * svm_hardware_setup() if it's unsupported by hardware or the host kernel.
197  */
198 bool npt_enabled = true;
199 module_param_named(npt, npt_enabled, bool, 0444);
200 
201 /* allow nested virtualization in KVM/SVM */
202 static int nested = true;
203 module_param(nested, int, S_IRUGO);
204 
205 /* enable/disable Next RIP Save */
206 static int nrips = true;
207 module_param(nrips, int, 0444);
208 
209 /* enable/disable Virtual VMLOAD VMSAVE */
210 static int vls = true;
211 module_param(vls, int, 0444);
212 
213 /* enable/disable Virtual GIF */
214 int vgif = true;
215 module_param(vgif, int, 0444);
216 
217 /* enable/disable LBR virtualization */
218 static int lbrv = true;
219 module_param(lbrv, int, 0444);
220 
221 static int tsc_scaling = true;
222 module_param(tsc_scaling, int, 0444);
223 
224 /*
225  * enable / disable AVIC.  Because the defaults differ for APICv
226  * support between VMX and SVM we cannot use module_param_named.
227  */
228 static bool avic;
229 module_param(avic, bool, 0444);
230 
231 bool __read_mostly dump_invalid_vmcb;
232 module_param(dump_invalid_vmcb, bool, 0644);
233 
234 
235 bool intercept_smi = true;
236 module_param(intercept_smi, bool, 0444);
237 
238 bool vnmi = true;
239 module_param(vnmi, bool, 0444);
240 
241 static bool svm_gp_erratum_intercept = true;
242 
243 static u8 rsm_ins_bytes[] = "\x0f\xaa";
244 
245 static unsigned long iopm_base;
246 
247 DEFINE_PER_CPU(struct svm_cpu_data, svm_data);
248 
249 /*
250  * Only MSR_TSC_AUX is switched via the user return hook.  EFER is switched via
251  * the VMCB, and the SYSCALL/SYSENTER MSRs are handled by VMLOAD/VMSAVE.
252  *
253  * RDTSCP and RDPID are not used in the kernel, specifically to allow KVM to
254  * defer the restoration of TSC_AUX until the CPU returns to userspace.
255  */
256 static int tsc_aux_uret_slot __read_mostly = -1;
257 
258 static const u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
259 
260 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
261 #define MSRS_RANGE_SIZE 2048
262 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
263 
264 u32 svm_msrpm_offset(u32 msr)
265 {
266 	u32 offset;
267 	int i;
268 
269 	for (i = 0; i < NUM_MSR_MAPS; i++) {
270 		if (msr < msrpm_ranges[i] ||
271 		    msr >= msrpm_ranges[i] + MSRS_IN_RANGE)
272 			continue;
273 
274 		offset  = (msr - msrpm_ranges[i]) / 4; /* 4 msrs per u8 */
275 		offset += (i * MSRS_RANGE_SIZE);       /* add range offset */
276 
277 		/* Now we have the u8 offset - but need the u32 offset */
278 		return offset / 4;
279 	}
280 
281 	/* MSR not in any range */
282 	return MSR_INVALID;
283 }
284 
285 static void svm_flush_tlb_current(struct kvm_vcpu *vcpu);
286 
287 static int get_npt_level(void)
288 {
289 #ifdef CONFIG_X86_64
290 	return pgtable_l5_enabled() ? PT64_ROOT_5LEVEL : PT64_ROOT_4LEVEL;
291 #else
292 	return PT32E_ROOT_LEVEL;
293 #endif
294 }
295 
296 int svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
297 {
298 	struct vcpu_svm *svm = to_svm(vcpu);
299 	u64 old_efer = vcpu->arch.efer;
300 	vcpu->arch.efer = efer;
301 
302 	if (!npt_enabled) {
303 		/* Shadow paging assumes NX to be available.  */
304 		efer |= EFER_NX;
305 
306 		if (!(efer & EFER_LMA))
307 			efer &= ~EFER_LME;
308 	}
309 
310 	if ((old_efer & EFER_SVME) != (efer & EFER_SVME)) {
311 		if (!(efer & EFER_SVME)) {
312 			svm_leave_nested(vcpu);
313 			svm_set_gif(svm, true);
314 			/* #GP intercept is still needed for vmware backdoor */
315 			if (!enable_vmware_backdoor)
316 				clr_exception_intercept(svm, GP_VECTOR);
317 
318 			/*
319 			 * Free the nested guest state, unless we are in SMM.
320 			 * In this case we will return to the nested guest
321 			 * as soon as we leave SMM.
322 			 */
323 			if (!is_smm(vcpu))
324 				svm_free_nested(svm);
325 
326 		} else {
327 			int ret = svm_allocate_nested(svm);
328 
329 			if (ret) {
330 				vcpu->arch.efer = old_efer;
331 				return ret;
332 			}
333 
334 			/*
335 			 * Never intercept #GP for SEV guests, KVM can't
336 			 * decrypt guest memory to workaround the erratum.
337 			 */
338 			if (svm_gp_erratum_intercept && !sev_guest(vcpu->kvm))
339 				set_exception_intercept(svm, GP_VECTOR);
340 		}
341 	}
342 
343 	svm->vmcb->save.efer = efer | EFER_SVME;
344 	vmcb_mark_dirty(svm->vmcb, VMCB_CR);
345 	return 0;
346 }
347 
348 static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu)
349 {
350 	struct vcpu_svm *svm = to_svm(vcpu);
351 	u32 ret = 0;
352 
353 	if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)
354 		ret = KVM_X86_SHADOW_INT_STI | KVM_X86_SHADOW_INT_MOV_SS;
355 	return ret;
356 }
357 
358 static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
359 {
360 	struct vcpu_svm *svm = to_svm(vcpu);
361 
362 	if (mask == 0)
363 		svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
364 	else
365 		svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK;
366 
367 }
368 
369 static int __svm_skip_emulated_instruction(struct kvm_vcpu *vcpu,
370 					   bool commit_side_effects)
371 {
372 	struct vcpu_svm *svm = to_svm(vcpu);
373 	unsigned long old_rflags;
374 
375 	/*
376 	 * SEV-ES does not expose the next RIP. The RIP update is controlled by
377 	 * the type of exit and the #VC handler in the guest.
378 	 */
379 	if (sev_es_guest(vcpu->kvm))
380 		goto done;
381 
382 	if (nrips && svm->vmcb->control.next_rip != 0) {
383 		WARN_ON_ONCE(!static_cpu_has(X86_FEATURE_NRIPS));
384 		svm->next_rip = svm->vmcb->control.next_rip;
385 	}
386 
387 	if (!svm->next_rip) {
388 		if (unlikely(!commit_side_effects))
389 			old_rflags = svm->vmcb->save.rflags;
390 
391 		if (!kvm_emulate_instruction(vcpu, EMULTYPE_SKIP))
392 			return 0;
393 
394 		if (unlikely(!commit_side_effects))
395 			svm->vmcb->save.rflags = old_rflags;
396 	} else {
397 		kvm_rip_write(vcpu, svm->next_rip);
398 	}
399 
400 done:
401 	if (likely(commit_side_effects))
402 		svm_set_interrupt_shadow(vcpu, 0);
403 
404 	return 1;
405 }
406 
407 static int svm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
408 {
409 	return __svm_skip_emulated_instruction(vcpu, true);
410 }
411 
412 static int svm_update_soft_interrupt_rip(struct kvm_vcpu *vcpu)
413 {
414 	unsigned long rip, old_rip = kvm_rip_read(vcpu);
415 	struct vcpu_svm *svm = to_svm(vcpu);
416 
417 	/*
418 	 * Due to architectural shortcomings, the CPU doesn't always provide
419 	 * NextRIP, e.g. if KVM intercepted an exception that occurred while
420 	 * the CPU was vectoring an INTO/INT3 in the guest.  Temporarily skip
421 	 * the instruction even if NextRIP is supported to acquire the next
422 	 * RIP so that it can be shoved into the NextRIP field, otherwise
423 	 * hardware will fail to advance guest RIP during event injection.
424 	 * Drop the exception/interrupt if emulation fails and effectively
425 	 * retry the instruction, it's the least awful option.  If NRIPS is
426 	 * in use, the skip must not commit any side effects such as clearing
427 	 * the interrupt shadow or RFLAGS.RF.
428 	 */
429 	if (!__svm_skip_emulated_instruction(vcpu, !nrips))
430 		return -EIO;
431 
432 	rip = kvm_rip_read(vcpu);
433 
434 	/*
435 	 * Save the injection information, even when using next_rip, as the
436 	 * VMCB's next_rip will be lost (cleared on VM-Exit) if the injection
437 	 * doesn't complete due to a VM-Exit occurring while the CPU is
438 	 * vectoring the event.   Decoding the instruction isn't guaranteed to
439 	 * work as there may be no backing instruction, e.g. if the event is
440 	 * being injected by L1 for L2, or if the guest is patching INT3 into
441 	 * a different instruction.
442 	 */
443 	svm->soft_int_injected = true;
444 	svm->soft_int_csbase = svm->vmcb->save.cs.base;
445 	svm->soft_int_old_rip = old_rip;
446 	svm->soft_int_next_rip = rip;
447 
448 	if (nrips)
449 		kvm_rip_write(vcpu, old_rip);
450 
451 	if (static_cpu_has(X86_FEATURE_NRIPS))
452 		svm->vmcb->control.next_rip = rip;
453 
454 	return 0;
455 }
456 
457 static void svm_inject_exception(struct kvm_vcpu *vcpu)
458 {
459 	struct kvm_queued_exception *ex = &vcpu->arch.exception;
460 	struct vcpu_svm *svm = to_svm(vcpu);
461 
462 	kvm_deliver_exception_payload(vcpu, ex);
463 
464 	if (kvm_exception_is_soft(ex->vector) &&
465 	    svm_update_soft_interrupt_rip(vcpu))
466 		return;
467 
468 	svm->vmcb->control.event_inj = ex->vector
469 		| SVM_EVTINJ_VALID
470 		| (ex->has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
471 		| SVM_EVTINJ_TYPE_EXEPT;
472 	svm->vmcb->control.event_inj_err = ex->error_code;
473 }
474 
475 static void svm_init_erratum_383(void)
476 {
477 	u32 low, high;
478 	int err;
479 	u64 val;
480 
481 	if (!static_cpu_has_bug(X86_BUG_AMD_TLB_MMATCH))
482 		return;
483 
484 	/* Use _safe variants to not break nested virtualization */
485 	val = native_read_msr_safe(MSR_AMD64_DC_CFG, &err);
486 	if (err)
487 		return;
488 
489 	val |= (1ULL << 47);
490 
491 	low  = lower_32_bits(val);
492 	high = upper_32_bits(val);
493 
494 	native_write_msr_safe(MSR_AMD64_DC_CFG, low, high);
495 
496 	erratum_383_found = true;
497 }
498 
499 static void svm_init_osvw(struct kvm_vcpu *vcpu)
500 {
501 	/*
502 	 * Guests should see errata 400 and 415 as fixed (assuming that
503 	 * HLT and IO instructions are intercepted).
504 	 */
505 	vcpu->arch.osvw.length = (osvw_len >= 3) ? (osvw_len) : 3;
506 	vcpu->arch.osvw.status = osvw_status & ~(6ULL);
507 
508 	/*
509 	 * By increasing VCPU's osvw.length to 3 we are telling the guest that
510 	 * all osvw.status bits inside that length, including bit 0 (which is
511 	 * reserved for erratum 298), are valid. However, if host processor's
512 	 * osvw_len is 0 then osvw_status[0] carries no information. We need to
513 	 * be conservative here and therefore we tell the guest that erratum 298
514 	 * is present (because we really don't know).
515 	 */
516 	if (osvw_len == 0 && boot_cpu_data.x86 == 0x10)
517 		vcpu->arch.osvw.status |= 1;
518 }
519 
520 static bool kvm_is_svm_supported(void)
521 {
522 	int cpu = raw_smp_processor_id();
523 	const char *msg;
524 	u64 vm_cr;
525 
526 	if (!cpu_has_svm(&msg)) {
527 		pr_err("SVM not supported by CPU %d, %s\n", cpu, msg);
528 		return false;
529 	}
530 
531 	if (cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT)) {
532 		pr_info("KVM is unsupported when running as an SEV guest\n");
533 		return false;
534 	}
535 
536 	rdmsrl(MSR_VM_CR, vm_cr);
537 	if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE)) {
538 		pr_err("SVM disabled (by BIOS) in MSR_VM_CR on CPU %d\n", cpu);
539 		return false;
540 	}
541 
542 	return true;
543 }
544 
545 static int svm_check_processor_compat(void)
546 {
547 	if (!kvm_is_svm_supported())
548 		return -EIO;
549 
550 	return 0;
551 }
552 
553 void __svm_write_tsc_multiplier(u64 multiplier)
554 {
555 	preempt_disable();
556 
557 	if (multiplier == __this_cpu_read(current_tsc_ratio))
558 		goto out;
559 
560 	wrmsrl(MSR_AMD64_TSC_RATIO, multiplier);
561 	__this_cpu_write(current_tsc_ratio, multiplier);
562 out:
563 	preempt_enable();
564 }
565 
566 static void svm_hardware_disable(void)
567 {
568 	/* Make sure we clean up behind us */
569 	if (tsc_scaling)
570 		__svm_write_tsc_multiplier(SVM_TSC_RATIO_DEFAULT);
571 
572 	cpu_svm_disable();
573 
574 	amd_pmu_disable_virt();
575 }
576 
577 static int svm_hardware_enable(void)
578 {
579 
580 	struct svm_cpu_data *sd;
581 	uint64_t efer;
582 	int me = raw_smp_processor_id();
583 
584 	rdmsrl(MSR_EFER, efer);
585 	if (efer & EFER_SVME)
586 		return -EBUSY;
587 
588 	sd = per_cpu_ptr(&svm_data, me);
589 	sd->asid_generation = 1;
590 	sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
591 	sd->next_asid = sd->max_asid + 1;
592 	sd->min_asid = max_sev_asid + 1;
593 
594 	wrmsrl(MSR_EFER, efer | EFER_SVME);
595 
596 	wrmsrl(MSR_VM_HSAVE_PA, sd->save_area_pa);
597 
598 	if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
599 		/*
600 		 * Set the default value, even if we don't use TSC scaling
601 		 * to avoid having stale value in the msr
602 		 */
603 		__svm_write_tsc_multiplier(SVM_TSC_RATIO_DEFAULT);
604 	}
605 
606 
607 	/*
608 	 * Get OSVW bits.
609 	 *
610 	 * Note that it is possible to have a system with mixed processor
611 	 * revisions and therefore different OSVW bits. If bits are not the same
612 	 * on different processors then choose the worst case (i.e. if erratum
613 	 * is present on one processor and not on another then assume that the
614 	 * erratum is present everywhere).
615 	 */
616 	if (cpu_has(&boot_cpu_data, X86_FEATURE_OSVW)) {
617 		uint64_t len, status = 0;
618 		int err;
619 
620 		len = native_read_msr_safe(MSR_AMD64_OSVW_ID_LENGTH, &err);
621 		if (!err)
622 			status = native_read_msr_safe(MSR_AMD64_OSVW_STATUS,
623 						      &err);
624 
625 		if (err)
626 			osvw_status = osvw_len = 0;
627 		else {
628 			if (len < osvw_len)
629 				osvw_len = len;
630 			osvw_status |= status;
631 			osvw_status &= (1ULL << osvw_len) - 1;
632 		}
633 	} else
634 		osvw_status = osvw_len = 0;
635 
636 	svm_init_erratum_383();
637 
638 	amd_pmu_enable_virt();
639 
640 	return 0;
641 }
642 
643 static void svm_cpu_uninit(int cpu)
644 {
645 	struct svm_cpu_data *sd = per_cpu_ptr(&svm_data, cpu);
646 
647 	if (!sd->save_area)
648 		return;
649 
650 	kfree(sd->sev_vmcbs);
651 	__free_page(sd->save_area);
652 	sd->save_area_pa = 0;
653 	sd->save_area = NULL;
654 }
655 
656 static int svm_cpu_init(int cpu)
657 {
658 	struct svm_cpu_data *sd = per_cpu_ptr(&svm_data, cpu);
659 	int ret = -ENOMEM;
660 
661 	memset(sd, 0, sizeof(struct svm_cpu_data));
662 	sd->save_area = alloc_page(GFP_KERNEL | __GFP_ZERO);
663 	if (!sd->save_area)
664 		return ret;
665 
666 	ret = sev_cpu_init(sd);
667 	if (ret)
668 		goto free_save_area;
669 
670 	sd->save_area_pa = __sme_page_pa(sd->save_area);
671 	return 0;
672 
673 free_save_area:
674 	__free_page(sd->save_area);
675 	sd->save_area = NULL;
676 	return ret;
677 
678 }
679 
680 static int direct_access_msr_slot(u32 msr)
681 {
682 	u32 i;
683 
684 	for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++)
685 		if (direct_access_msrs[i].index == msr)
686 			return i;
687 
688 	return -ENOENT;
689 }
690 
691 static void set_shadow_msr_intercept(struct kvm_vcpu *vcpu, u32 msr, int read,
692 				     int write)
693 {
694 	struct vcpu_svm *svm = to_svm(vcpu);
695 	int slot = direct_access_msr_slot(msr);
696 
697 	if (slot == -ENOENT)
698 		return;
699 
700 	/* Set the shadow bitmaps to the desired intercept states */
701 	if (read)
702 		set_bit(slot, svm->shadow_msr_intercept.read);
703 	else
704 		clear_bit(slot, svm->shadow_msr_intercept.read);
705 
706 	if (write)
707 		set_bit(slot, svm->shadow_msr_intercept.write);
708 	else
709 		clear_bit(slot, svm->shadow_msr_intercept.write);
710 }
711 
712 static bool valid_msr_intercept(u32 index)
713 {
714 	return direct_access_msr_slot(index) != -ENOENT;
715 }
716 
717 static bool msr_write_intercepted(struct kvm_vcpu *vcpu, u32 msr)
718 {
719 	u8 bit_write;
720 	unsigned long tmp;
721 	u32 offset;
722 	u32 *msrpm;
723 
724 	/*
725 	 * For non-nested case:
726 	 * If the L01 MSR bitmap does not intercept the MSR, then we need to
727 	 * save it.
728 	 *
729 	 * For nested case:
730 	 * If the L02 MSR bitmap does not intercept the MSR, then we need to
731 	 * save it.
732 	 */
733 	msrpm = is_guest_mode(vcpu) ? to_svm(vcpu)->nested.msrpm:
734 				      to_svm(vcpu)->msrpm;
735 
736 	offset    = svm_msrpm_offset(msr);
737 	bit_write = 2 * (msr & 0x0f) + 1;
738 	tmp       = msrpm[offset];
739 
740 	BUG_ON(offset == MSR_INVALID);
741 
742 	return test_bit(bit_write, &tmp);
743 }
744 
745 static void set_msr_interception_bitmap(struct kvm_vcpu *vcpu, u32 *msrpm,
746 					u32 msr, int read, int write)
747 {
748 	struct vcpu_svm *svm = to_svm(vcpu);
749 	u8 bit_read, bit_write;
750 	unsigned long tmp;
751 	u32 offset;
752 
753 	/*
754 	 * If this warning triggers extend the direct_access_msrs list at the
755 	 * beginning of the file
756 	 */
757 	WARN_ON(!valid_msr_intercept(msr));
758 
759 	/* Enforce non allowed MSRs to trap */
760 	if (read && !kvm_msr_allowed(vcpu, msr, KVM_MSR_FILTER_READ))
761 		read = 0;
762 
763 	if (write && !kvm_msr_allowed(vcpu, msr, KVM_MSR_FILTER_WRITE))
764 		write = 0;
765 
766 	offset    = svm_msrpm_offset(msr);
767 	bit_read  = 2 * (msr & 0x0f);
768 	bit_write = 2 * (msr & 0x0f) + 1;
769 	tmp       = msrpm[offset];
770 
771 	BUG_ON(offset == MSR_INVALID);
772 
773 	read  ? clear_bit(bit_read,  &tmp) : set_bit(bit_read,  &tmp);
774 	write ? clear_bit(bit_write, &tmp) : set_bit(bit_write, &tmp);
775 
776 	msrpm[offset] = tmp;
777 
778 	svm_hv_vmcb_dirty_nested_enlightenments(vcpu);
779 	svm->nested.force_msr_bitmap_recalc = true;
780 }
781 
782 void set_msr_interception(struct kvm_vcpu *vcpu, u32 *msrpm, u32 msr,
783 			  int read, int write)
784 {
785 	set_shadow_msr_intercept(vcpu, msr, read, write);
786 	set_msr_interception_bitmap(vcpu, msrpm, msr, read, write);
787 }
788 
789 u32 *svm_vcpu_alloc_msrpm(void)
790 {
791 	unsigned int order = get_order(MSRPM_SIZE);
792 	struct page *pages = alloc_pages(GFP_KERNEL_ACCOUNT, order);
793 	u32 *msrpm;
794 
795 	if (!pages)
796 		return NULL;
797 
798 	msrpm = page_address(pages);
799 	memset(msrpm, 0xff, PAGE_SIZE * (1 << order));
800 
801 	return msrpm;
802 }
803 
804 void svm_vcpu_init_msrpm(struct kvm_vcpu *vcpu, u32 *msrpm)
805 {
806 	int i;
807 
808 	for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
809 		if (!direct_access_msrs[i].always)
810 			continue;
811 		set_msr_interception(vcpu, msrpm, direct_access_msrs[i].index, 1, 1);
812 	}
813 }
814 
815 void svm_set_x2apic_msr_interception(struct vcpu_svm *svm, bool intercept)
816 {
817 	int i;
818 
819 	if (intercept == svm->x2avic_msrs_intercepted)
820 		return;
821 
822 	if (!x2avic_enabled ||
823 	    !apic_x2apic_mode(svm->vcpu.arch.apic))
824 		return;
825 
826 	for (i = 0; i < MAX_DIRECT_ACCESS_MSRS; i++) {
827 		int index = direct_access_msrs[i].index;
828 
829 		if ((index < APIC_BASE_MSR) ||
830 		    (index > APIC_BASE_MSR + 0xff))
831 			continue;
832 		set_msr_interception(&svm->vcpu, svm->msrpm, index,
833 				     !intercept, !intercept);
834 	}
835 
836 	svm->x2avic_msrs_intercepted = intercept;
837 }
838 
839 void svm_vcpu_free_msrpm(u32 *msrpm)
840 {
841 	__free_pages(virt_to_page(msrpm), get_order(MSRPM_SIZE));
842 }
843 
844 static void svm_msr_filter_changed(struct kvm_vcpu *vcpu)
845 {
846 	struct vcpu_svm *svm = to_svm(vcpu);
847 	u32 i;
848 
849 	/*
850 	 * Set intercept permissions for all direct access MSRs again. They
851 	 * will automatically get filtered through the MSR filter, so we are
852 	 * back in sync after this.
853 	 */
854 	for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
855 		u32 msr = direct_access_msrs[i].index;
856 		u32 read = test_bit(i, svm->shadow_msr_intercept.read);
857 		u32 write = test_bit(i, svm->shadow_msr_intercept.write);
858 
859 		set_msr_interception_bitmap(vcpu, svm->msrpm, msr, read, write);
860 	}
861 }
862 
863 static void add_msr_offset(u32 offset)
864 {
865 	int i;
866 
867 	for (i = 0; i < MSRPM_OFFSETS; ++i) {
868 
869 		/* Offset already in list? */
870 		if (msrpm_offsets[i] == offset)
871 			return;
872 
873 		/* Slot used by another offset? */
874 		if (msrpm_offsets[i] != MSR_INVALID)
875 			continue;
876 
877 		/* Add offset to list */
878 		msrpm_offsets[i] = offset;
879 
880 		return;
881 	}
882 
883 	/*
884 	 * If this BUG triggers the msrpm_offsets table has an overflow. Just
885 	 * increase MSRPM_OFFSETS in this case.
886 	 */
887 	BUG();
888 }
889 
890 static void init_msrpm_offsets(void)
891 {
892 	int i;
893 
894 	memset(msrpm_offsets, 0xff, sizeof(msrpm_offsets));
895 
896 	for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
897 		u32 offset;
898 
899 		offset = svm_msrpm_offset(direct_access_msrs[i].index);
900 		BUG_ON(offset == MSR_INVALID);
901 
902 		add_msr_offset(offset);
903 	}
904 }
905 
906 void svm_copy_lbrs(struct vmcb *to_vmcb, struct vmcb *from_vmcb)
907 {
908 	to_vmcb->save.dbgctl		= from_vmcb->save.dbgctl;
909 	to_vmcb->save.br_from		= from_vmcb->save.br_from;
910 	to_vmcb->save.br_to		= from_vmcb->save.br_to;
911 	to_vmcb->save.last_excp_from	= from_vmcb->save.last_excp_from;
912 	to_vmcb->save.last_excp_to	= from_vmcb->save.last_excp_to;
913 
914 	vmcb_mark_dirty(to_vmcb, VMCB_LBR);
915 }
916 
917 static void svm_enable_lbrv(struct kvm_vcpu *vcpu)
918 {
919 	struct vcpu_svm *svm = to_svm(vcpu);
920 
921 	svm->vmcb->control.virt_ext |= LBR_CTL_ENABLE_MASK;
922 	set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
923 	set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
924 	set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
925 	set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
926 
927 	/* Move the LBR msrs to the vmcb02 so that the guest can see them. */
928 	if (is_guest_mode(vcpu))
929 		svm_copy_lbrs(svm->vmcb, svm->vmcb01.ptr);
930 }
931 
932 static void svm_disable_lbrv(struct kvm_vcpu *vcpu)
933 {
934 	struct vcpu_svm *svm = to_svm(vcpu);
935 
936 	svm->vmcb->control.virt_ext &= ~LBR_CTL_ENABLE_MASK;
937 	set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
938 	set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
939 	set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
940 	set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
941 
942 	/*
943 	 * Move the LBR msrs back to the vmcb01 to avoid copying them
944 	 * on nested guest entries.
945 	 */
946 	if (is_guest_mode(vcpu))
947 		svm_copy_lbrs(svm->vmcb01.ptr, svm->vmcb);
948 }
949 
950 static int svm_get_lbr_msr(struct vcpu_svm *svm, u32 index)
951 {
952 	/*
953 	 * If the LBR virtualization is disabled, the LBR msrs are always
954 	 * kept in the vmcb01 to avoid copying them on nested guest entries.
955 	 *
956 	 * If nested, and the LBR virtualization is enabled/disabled, the msrs
957 	 * are moved between the vmcb01 and vmcb02 as needed.
958 	 */
959 	struct vmcb *vmcb =
960 		(svm->vmcb->control.virt_ext & LBR_CTL_ENABLE_MASK) ?
961 			svm->vmcb : svm->vmcb01.ptr;
962 
963 	switch (index) {
964 	case MSR_IA32_DEBUGCTLMSR:
965 		return vmcb->save.dbgctl;
966 	case MSR_IA32_LASTBRANCHFROMIP:
967 		return vmcb->save.br_from;
968 	case MSR_IA32_LASTBRANCHTOIP:
969 		return vmcb->save.br_to;
970 	case MSR_IA32_LASTINTFROMIP:
971 		return vmcb->save.last_excp_from;
972 	case MSR_IA32_LASTINTTOIP:
973 		return vmcb->save.last_excp_to;
974 	default:
975 		KVM_BUG(false, svm->vcpu.kvm,
976 			"%s: Unknown MSR 0x%x", __func__, index);
977 		return 0;
978 	}
979 }
980 
981 void svm_update_lbrv(struct kvm_vcpu *vcpu)
982 {
983 	struct vcpu_svm *svm = to_svm(vcpu);
984 
985 	bool enable_lbrv = svm_get_lbr_msr(svm, MSR_IA32_DEBUGCTLMSR) &
986 					   DEBUGCTLMSR_LBR;
987 
988 	bool current_enable_lbrv = !!(svm->vmcb->control.virt_ext &
989 				      LBR_CTL_ENABLE_MASK);
990 
991 	if (unlikely(is_guest_mode(vcpu) && svm->lbrv_enabled))
992 		if (unlikely(svm->nested.ctl.virt_ext & LBR_CTL_ENABLE_MASK))
993 			enable_lbrv = true;
994 
995 	if (enable_lbrv == current_enable_lbrv)
996 		return;
997 
998 	if (enable_lbrv)
999 		svm_enable_lbrv(vcpu);
1000 	else
1001 		svm_disable_lbrv(vcpu);
1002 }
1003 
1004 void disable_nmi_singlestep(struct vcpu_svm *svm)
1005 {
1006 	svm->nmi_singlestep = false;
1007 
1008 	if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP)) {
1009 		/* Clear our flags if they were not set by the guest */
1010 		if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF))
1011 			svm->vmcb->save.rflags &= ~X86_EFLAGS_TF;
1012 		if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_RF))
1013 			svm->vmcb->save.rflags &= ~X86_EFLAGS_RF;
1014 	}
1015 }
1016 
1017 static void grow_ple_window(struct kvm_vcpu *vcpu)
1018 {
1019 	struct vcpu_svm *svm = to_svm(vcpu);
1020 	struct vmcb_control_area *control = &svm->vmcb->control;
1021 	int old = control->pause_filter_count;
1022 
1023 	if (kvm_pause_in_guest(vcpu->kvm))
1024 		return;
1025 
1026 	control->pause_filter_count = __grow_ple_window(old,
1027 							pause_filter_count,
1028 							pause_filter_count_grow,
1029 							pause_filter_count_max);
1030 
1031 	if (control->pause_filter_count != old) {
1032 		vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1033 		trace_kvm_ple_window_update(vcpu->vcpu_id,
1034 					    control->pause_filter_count, old);
1035 	}
1036 }
1037 
1038 static void shrink_ple_window(struct kvm_vcpu *vcpu)
1039 {
1040 	struct vcpu_svm *svm = to_svm(vcpu);
1041 	struct vmcb_control_area *control = &svm->vmcb->control;
1042 	int old = control->pause_filter_count;
1043 
1044 	if (kvm_pause_in_guest(vcpu->kvm))
1045 		return;
1046 
1047 	control->pause_filter_count =
1048 				__shrink_ple_window(old,
1049 						    pause_filter_count,
1050 						    pause_filter_count_shrink,
1051 						    pause_filter_count);
1052 	if (control->pause_filter_count != old) {
1053 		vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1054 		trace_kvm_ple_window_update(vcpu->vcpu_id,
1055 					    control->pause_filter_count, old);
1056 	}
1057 }
1058 
1059 static void svm_hardware_unsetup(void)
1060 {
1061 	int cpu;
1062 
1063 	sev_hardware_unsetup();
1064 
1065 	for_each_possible_cpu(cpu)
1066 		svm_cpu_uninit(cpu);
1067 
1068 	__free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT),
1069 	get_order(IOPM_SIZE));
1070 	iopm_base = 0;
1071 }
1072 
1073 static void init_seg(struct vmcb_seg *seg)
1074 {
1075 	seg->selector = 0;
1076 	seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
1077 		      SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
1078 	seg->limit = 0xffff;
1079 	seg->base = 0;
1080 }
1081 
1082 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
1083 {
1084 	seg->selector = 0;
1085 	seg->attrib = SVM_SELECTOR_P_MASK | type;
1086 	seg->limit = 0xffff;
1087 	seg->base = 0;
1088 }
1089 
1090 static u64 svm_get_l2_tsc_offset(struct kvm_vcpu *vcpu)
1091 {
1092 	struct vcpu_svm *svm = to_svm(vcpu);
1093 
1094 	return svm->nested.ctl.tsc_offset;
1095 }
1096 
1097 static u64 svm_get_l2_tsc_multiplier(struct kvm_vcpu *vcpu)
1098 {
1099 	struct vcpu_svm *svm = to_svm(vcpu);
1100 
1101 	return svm->tsc_ratio_msr;
1102 }
1103 
1104 static void svm_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1105 {
1106 	struct vcpu_svm *svm = to_svm(vcpu);
1107 
1108 	svm->vmcb01.ptr->control.tsc_offset = vcpu->arch.l1_tsc_offset;
1109 	svm->vmcb->control.tsc_offset = offset;
1110 	vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1111 }
1112 
1113 static void svm_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 multiplier)
1114 {
1115 	__svm_write_tsc_multiplier(multiplier);
1116 }
1117 
1118 
1119 /* Evaluate instruction intercepts that depend on guest CPUID features. */
1120 static void svm_recalc_instruction_intercepts(struct kvm_vcpu *vcpu,
1121 					      struct vcpu_svm *svm)
1122 {
1123 	/*
1124 	 * Intercept INVPCID if shadow paging is enabled to sync/free shadow
1125 	 * roots, or if INVPCID is disabled in the guest to inject #UD.
1126 	 */
1127 	if (kvm_cpu_cap_has(X86_FEATURE_INVPCID)) {
1128 		if (!npt_enabled ||
1129 		    !guest_cpuid_has(&svm->vcpu, X86_FEATURE_INVPCID))
1130 			svm_set_intercept(svm, INTERCEPT_INVPCID);
1131 		else
1132 			svm_clr_intercept(svm, INTERCEPT_INVPCID);
1133 	}
1134 
1135 	if (kvm_cpu_cap_has(X86_FEATURE_RDTSCP)) {
1136 		if (guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
1137 			svm_clr_intercept(svm, INTERCEPT_RDTSCP);
1138 		else
1139 			svm_set_intercept(svm, INTERCEPT_RDTSCP);
1140 	}
1141 }
1142 
1143 static inline void init_vmcb_after_set_cpuid(struct kvm_vcpu *vcpu)
1144 {
1145 	struct vcpu_svm *svm = to_svm(vcpu);
1146 
1147 	if (guest_cpuid_is_intel(vcpu)) {
1148 		/*
1149 		 * We must intercept SYSENTER_EIP and SYSENTER_ESP
1150 		 * accesses because the processor only stores 32 bits.
1151 		 * For the same reason we cannot use virtual VMLOAD/VMSAVE.
1152 		 */
1153 		svm_set_intercept(svm, INTERCEPT_VMLOAD);
1154 		svm_set_intercept(svm, INTERCEPT_VMSAVE);
1155 		svm->vmcb->control.virt_ext &= ~VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK;
1156 
1157 		set_msr_interception(vcpu, svm->msrpm, MSR_IA32_SYSENTER_EIP, 0, 0);
1158 		set_msr_interception(vcpu, svm->msrpm, MSR_IA32_SYSENTER_ESP, 0, 0);
1159 
1160 		svm->v_vmload_vmsave_enabled = false;
1161 	} else {
1162 		/*
1163 		 * If hardware supports Virtual VMLOAD VMSAVE then enable it
1164 		 * in VMCB and clear intercepts to avoid #VMEXIT.
1165 		 */
1166 		if (vls) {
1167 			svm_clr_intercept(svm, INTERCEPT_VMLOAD);
1168 			svm_clr_intercept(svm, INTERCEPT_VMSAVE);
1169 			svm->vmcb->control.virt_ext |= VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK;
1170 		}
1171 		/* No need to intercept these MSRs */
1172 		set_msr_interception(vcpu, svm->msrpm, MSR_IA32_SYSENTER_EIP, 1, 1);
1173 		set_msr_interception(vcpu, svm->msrpm, MSR_IA32_SYSENTER_ESP, 1, 1);
1174 	}
1175 }
1176 
1177 static void init_vmcb(struct kvm_vcpu *vcpu)
1178 {
1179 	struct vcpu_svm *svm = to_svm(vcpu);
1180 	struct vmcb *vmcb = svm->vmcb01.ptr;
1181 	struct vmcb_control_area *control = &vmcb->control;
1182 	struct vmcb_save_area *save = &vmcb->save;
1183 
1184 	svm_set_intercept(svm, INTERCEPT_CR0_READ);
1185 	svm_set_intercept(svm, INTERCEPT_CR3_READ);
1186 	svm_set_intercept(svm, INTERCEPT_CR4_READ);
1187 	svm_set_intercept(svm, INTERCEPT_CR0_WRITE);
1188 	svm_set_intercept(svm, INTERCEPT_CR3_WRITE);
1189 	svm_set_intercept(svm, INTERCEPT_CR4_WRITE);
1190 	if (!kvm_vcpu_apicv_active(vcpu))
1191 		svm_set_intercept(svm, INTERCEPT_CR8_WRITE);
1192 
1193 	set_dr_intercepts(svm);
1194 
1195 	set_exception_intercept(svm, PF_VECTOR);
1196 	set_exception_intercept(svm, UD_VECTOR);
1197 	set_exception_intercept(svm, MC_VECTOR);
1198 	set_exception_intercept(svm, AC_VECTOR);
1199 	set_exception_intercept(svm, DB_VECTOR);
1200 	/*
1201 	 * Guest access to VMware backdoor ports could legitimately
1202 	 * trigger #GP because of TSS I/O permission bitmap.
1203 	 * We intercept those #GP and allow access to them anyway
1204 	 * as VMware does.  Don't intercept #GP for SEV guests as KVM can't
1205 	 * decrypt guest memory to decode the faulting instruction.
1206 	 */
1207 	if (enable_vmware_backdoor && !sev_guest(vcpu->kvm))
1208 		set_exception_intercept(svm, GP_VECTOR);
1209 
1210 	svm_set_intercept(svm, INTERCEPT_INTR);
1211 	svm_set_intercept(svm, INTERCEPT_NMI);
1212 
1213 	if (intercept_smi)
1214 		svm_set_intercept(svm, INTERCEPT_SMI);
1215 
1216 	svm_set_intercept(svm, INTERCEPT_SELECTIVE_CR0);
1217 	svm_set_intercept(svm, INTERCEPT_RDPMC);
1218 	svm_set_intercept(svm, INTERCEPT_CPUID);
1219 	svm_set_intercept(svm, INTERCEPT_INVD);
1220 	svm_set_intercept(svm, INTERCEPT_INVLPG);
1221 	svm_set_intercept(svm, INTERCEPT_INVLPGA);
1222 	svm_set_intercept(svm, INTERCEPT_IOIO_PROT);
1223 	svm_set_intercept(svm, INTERCEPT_MSR_PROT);
1224 	svm_set_intercept(svm, INTERCEPT_TASK_SWITCH);
1225 	svm_set_intercept(svm, INTERCEPT_SHUTDOWN);
1226 	svm_set_intercept(svm, INTERCEPT_VMRUN);
1227 	svm_set_intercept(svm, INTERCEPT_VMMCALL);
1228 	svm_set_intercept(svm, INTERCEPT_VMLOAD);
1229 	svm_set_intercept(svm, INTERCEPT_VMSAVE);
1230 	svm_set_intercept(svm, INTERCEPT_STGI);
1231 	svm_set_intercept(svm, INTERCEPT_CLGI);
1232 	svm_set_intercept(svm, INTERCEPT_SKINIT);
1233 	svm_set_intercept(svm, INTERCEPT_WBINVD);
1234 	svm_set_intercept(svm, INTERCEPT_XSETBV);
1235 	svm_set_intercept(svm, INTERCEPT_RDPRU);
1236 	svm_set_intercept(svm, INTERCEPT_RSM);
1237 
1238 	if (!kvm_mwait_in_guest(vcpu->kvm)) {
1239 		svm_set_intercept(svm, INTERCEPT_MONITOR);
1240 		svm_set_intercept(svm, INTERCEPT_MWAIT);
1241 	}
1242 
1243 	if (!kvm_hlt_in_guest(vcpu->kvm))
1244 		svm_set_intercept(svm, INTERCEPT_HLT);
1245 
1246 	control->iopm_base_pa = __sme_set(iopm_base);
1247 	control->msrpm_base_pa = __sme_set(__pa(svm->msrpm));
1248 	control->int_ctl = V_INTR_MASKING_MASK;
1249 
1250 	init_seg(&save->es);
1251 	init_seg(&save->ss);
1252 	init_seg(&save->ds);
1253 	init_seg(&save->fs);
1254 	init_seg(&save->gs);
1255 
1256 	save->cs.selector = 0xf000;
1257 	save->cs.base = 0xffff0000;
1258 	/* Executable/Readable Code Segment */
1259 	save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
1260 		SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
1261 	save->cs.limit = 0xffff;
1262 
1263 	save->gdtr.base = 0;
1264 	save->gdtr.limit = 0xffff;
1265 	save->idtr.base = 0;
1266 	save->idtr.limit = 0xffff;
1267 
1268 	init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
1269 	init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
1270 
1271 	if (npt_enabled) {
1272 		/* Setup VMCB for Nested Paging */
1273 		control->nested_ctl |= SVM_NESTED_CTL_NP_ENABLE;
1274 		svm_clr_intercept(svm, INTERCEPT_INVLPG);
1275 		clr_exception_intercept(svm, PF_VECTOR);
1276 		svm_clr_intercept(svm, INTERCEPT_CR3_READ);
1277 		svm_clr_intercept(svm, INTERCEPT_CR3_WRITE);
1278 		save->g_pat = vcpu->arch.pat;
1279 		save->cr3 = 0;
1280 	}
1281 	svm->current_vmcb->asid_generation = 0;
1282 	svm->asid = 0;
1283 
1284 	svm->nested.vmcb12_gpa = INVALID_GPA;
1285 	svm->nested.last_vmcb12_gpa = INVALID_GPA;
1286 
1287 	if (!kvm_pause_in_guest(vcpu->kvm)) {
1288 		control->pause_filter_count = pause_filter_count;
1289 		if (pause_filter_thresh)
1290 			control->pause_filter_thresh = pause_filter_thresh;
1291 		svm_set_intercept(svm, INTERCEPT_PAUSE);
1292 	} else {
1293 		svm_clr_intercept(svm, INTERCEPT_PAUSE);
1294 	}
1295 
1296 	svm_recalc_instruction_intercepts(vcpu, svm);
1297 
1298 	/*
1299 	 * If the host supports V_SPEC_CTRL then disable the interception
1300 	 * of MSR_IA32_SPEC_CTRL.
1301 	 */
1302 	if (boot_cpu_has(X86_FEATURE_V_SPEC_CTRL))
1303 		set_msr_interception(vcpu, svm->msrpm, MSR_IA32_SPEC_CTRL, 1, 1);
1304 
1305 	if (kvm_vcpu_apicv_active(vcpu))
1306 		avic_init_vmcb(svm, vmcb);
1307 
1308 	if (vnmi)
1309 		svm->vmcb->control.int_ctl |= V_NMI_ENABLE_MASK;
1310 
1311 	if (vgif) {
1312 		svm_clr_intercept(svm, INTERCEPT_STGI);
1313 		svm_clr_intercept(svm, INTERCEPT_CLGI);
1314 		svm->vmcb->control.int_ctl |= V_GIF_ENABLE_MASK;
1315 	}
1316 
1317 	if (sev_guest(vcpu->kvm))
1318 		sev_init_vmcb(svm);
1319 
1320 	svm_hv_init_vmcb(vmcb);
1321 	init_vmcb_after_set_cpuid(vcpu);
1322 
1323 	vmcb_mark_all_dirty(vmcb);
1324 
1325 	enable_gif(svm);
1326 }
1327 
1328 static void __svm_vcpu_reset(struct kvm_vcpu *vcpu)
1329 {
1330 	struct vcpu_svm *svm = to_svm(vcpu);
1331 
1332 	svm_vcpu_init_msrpm(vcpu, svm->msrpm);
1333 
1334 	svm_init_osvw(vcpu);
1335 	vcpu->arch.microcode_version = 0x01000065;
1336 	svm->tsc_ratio_msr = kvm_caps.default_tsc_scaling_ratio;
1337 
1338 	svm->nmi_masked = false;
1339 	svm->awaiting_iret_completion = false;
1340 
1341 	if (sev_es_guest(vcpu->kvm))
1342 		sev_es_vcpu_reset(svm);
1343 }
1344 
1345 static void svm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
1346 {
1347 	struct vcpu_svm *svm = to_svm(vcpu);
1348 
1349 	svm->spec_ctrl = 0;
1350 	svm->virt_spec_ctrl = 0;
1351 
1352 	init_vmcb(vcpu);
1353 
1354 	if (!init_event)
1355 		__svm_vcpu_reset(vcpu);
1356 }
1357 
1358 void svm_switch_vmcb(struct vcpu_svm *svm, struct kvm_vmcb_info *target_vmcb)
1359 {
1360 	svm->current_vmcb = target_vmcb;
1361 	svm->vmcb = target_vmcb->ptr;
1362 }
1363 
1364 static int svm_vcpu_create(struct kvm_vcpu *vcpu)
1365 {
1366 	struct vcpu_svm *svm;
1367 	struct page *vmcb01_page;
1368 	struct page *vmsa_page = NULL;
1369 	int err;
1370 
1371 	BUILD_BUG_ON(offsetof(struct vcpu_svm, vcpu) != 0);
1372 	svm = to_svm(vcpu);
1373 
1374 	err = -ENOMEM;
1375 	vmcb01_page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
1376 	if (!vmcb01_page)
1377 		goto out;
1378 
1379 	if (sev_es_guest(vcpu->kvm)) {
1380 		/*
1381 		 * SEV-ES guests require a separate VMSA page used to contain
1382 		 * the encrypted register state of the guest.
1383 		 */
1384 		vmsa_page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
1385 		if (!vmsa_page)
1386 			goto error_free_vmcb_page;
1387 
1388 		/*
1389 		 * SEV-ES guests maintain an encrypted version of their FPU
1390 		 * state which is restored and saved on VMRUN and VMEXIT.
1391 		 * Mark vcpu->arch.guest_fpu->fpstate as scratch so it won't
1392 		 * do xsave/xrstor on it.
1393 		 */
1394 		fpstate_set_confidential(&vcpu->arch.guest_fpu);
1395 	}
1396 
1397 	err = avic_init_vcpu(svm);
1398 	if (err)
1399 		goto error_free_vmsa_page;
1400 
1401 	svm->msrpm = svm_vcpu_alloc_msrpm();
1402 	if (!svm->msrpm) {
1403 		err = -ENOMEM;
1404 		goto error_free_vmsa_page;
1405 	}
1406 
1407 	svm->x2avic_msrs_intercepted = true;
1408 
1409 	svm->vmcb01.ptr = page_address(vmcb01_page);
1410 	svm->vmcb01.pa = __sme_set(page_to_pfn(vmcb01_page) << PAGE_SHIFT);
1411 	svm_switch_vmcb(svm, &svm->vmcb01);
1412 
1413 	if (vmsa_page)
1414 		svm->sev_es.vmsa = page_address(vmsa_page);
1415 
1416 	svm->guest_state_loaded = false;
1417 
1418 	return 0;
1419 
1420 error_free_vmsa_page:
1421 	if (vmsa_page)
1422 		__free_page(vmsa_page);
1423 error_free_vmcb_page:
1424 	__free_page(vmcb01_page);
1425 out:
1426 	return err;
1427 }
1428 
1429 static void svm_clear_current_vmcb(struct vmcb *vmcb)
1430 {
1431 	int i;
1432 
1433 	for_each_online_cpu(i)
1434 		cmpxchg(per_cpu_ptr(&svm_data.current_vmcb, i), vmcb, NULL);
1435 }
1436 
1437 static void svm_vcpu_free(struct kvm_vcpu *vcpu)
1438 {
1439 	struct vcpu_svm *svm = to_svm(vcpu);
1440 
1441 	/*
1442 	 * The vmcb page can be recycled, causing a false negative in
1443 	 * svm_vcpu_load(). So, ensure that no logical CPU has this
1444 	 * vmcb page recorded as its current vmcb.
1445 	 */
1446 	svm_clear_current_vmcb(svm->vmcb);
1447 
1448 	svm_leave_nested(vcpu);
1449 	svm_free_nested(svm);
1450 
1451 	sev_free_vcpu(vcpu);
1452 
1453 	__free_page(pfn_to_page(__sme_clr(svm->vmcb01.pa) >> PAGE_SHIFT));
1454 	__free_pages(virt_to_page(svm->msrpm), get_order(MSRPM_SIZE));
1455 }
1456 
1457 static void svm_prepare_switch_to_guest(struct kvm_vcpu *vcpu)
1458 {
1459 	struct vcpu_svm *svm = to_svm(vcpu);
1460 	struct svm_cpu_data *sd = per_cpu_ptr(&svm_data, vcpu->cpu);
1461 
1462 	if (sev_es_guest(vcpu->kvm))
1463 		sev_es_unmap_ghcb(svm);
1464 
1465 	if (svm->guest_state_loaded)
1466 		return;
1467 
1468 	/*
1469 	 * Save additional host state that will be restored on VMEXIT (sev-es)
1470 	 * or subsequent vmload of host save area.
1471 	 */
1472 	vmsave(sd->save_area_pa);
1473 	if (sev_es_guest(vcpu->kvm)) {
1474 		struct sev_es_save_area *hostsa;
1475 		hostsa = (struct sev_es_save_area *)(page_address(sd->save_area) + 0x400);
1476 
1477 		sev_es_prepare_switch_to_guest(hostsa);
1478 	}
1479 
1480 	if (tsc_scaling)
1481 		__svm_write_tsc_multiplier(vcpu->arch.tsc_scaling_ratio);
1482 
1483 	if (likely(tsc_aux_uret_slot >= 0))
1484 		kvm_set_user_return_msr(tsc_aux_uret_slot, svm->tsc_aux, -1ull);
1485 
1486 	svm->guest_state_loaded = true;
1487 }
1488 
1489 static void svm_prepare_host_switch(struct kvm_vcpu *vcpu)
1490 {
1491 	to_svm(vcpu)->guest_state_loaded = false;
1492 }
1493 
1494 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1495 {
1496 	struct vcpu_svm *svm = to_svm(vcpu);
1497 	struct svm_cpu_data *sd = per_cpu_ptr(&svm_data, cpu);
1498 
1499 	if (sd->current_vmcb != svm->vmcb) {
1500 		sd->current_vmcb = svm->vmcb;
1501 
1502 		if (!cpu_feature_enabled(X86_FEATURE_IBPB_ON_VMEXIT))
1503 			indirect_branch_prediction_barrier();
1504 	}
1505 	if (kvm_vcpu_apicv_active(vcpu))
1506 		avic_vcpu_load(vcpu, cpu);
1507 }
1508 
1509 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
1510 {
1511 	if (kvm_vcpu_apicv_active(vcpu))
1512 		avic_vcpu_put(vcpu);
1513 
1514 	svm_prepare_host_switch(vcpu);
1515 
1516 	++vcpu->stat.host_state_reload;
1517 }
1518 
1519 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
1520 {
1521 	struct vcpu_svm *svm = to_svm(vcpu);
1522 	unsigned long rflags = svm->vmcb->save.rflags;
1523 
1524 	if (svm->nmi_singlestep) {
1525 		/* Hide our flags if they were not set by the guest */
1526 		if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF))
1527 			rflags &= ~X86_EFLAGS_TF;
1528 		if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_RF))
1529 			rflags &= ~X86_EFLAGS_RF;
1530 	}
1531 	return rflags;
1532 }
1533 
1534 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1535 {
1536 	if (to_svm(vcpu)->nmi_singlestep)
1537 		rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
1538 
1539        /*
1540         * Any change of EFLAGS.VM is accompanied by a reload of SS
1541         * (caused by either a task switch or an inter-privilege IRET),
1542         * so we do not need to update the CPL here.
1543         */
1544 	to_svm(vcpu)->vmcb->save.rflags = rflags;
1545 }
1546 
1547 static bool svm_get_if_flag(struct kvm_vcpu *vcpu)
1548 {
1549 	struct vmcb *vmcb = to_svm(vcpu)->vmcb;
1550 
1551 	return sev_es_guest(vcpu->kvm)
1552 		? vmcb->control.int_state & SVM_GUEST_INTERRUPT_MASK
1553 		: kvm_get_rflags(vcpu) & X86_EFLAGS_IF;
1554 }
1555 
1556 static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1557 {
1558 	kvm_register_mark_available(vcpu, reg);
1559 
1560 	switch (reg) {
1561 	case VCPU_EXREG_PDPTR:
1562 		/*
1563 		 * When !npt_enabled, mmu->pdptrs[] is already available since
1564 		 * it is always updated per SDM when moving to CRs.
1565 		 */
1566 		if (npt_enabled)
1567 			load_pdptrs(vcpu, kvm_read_cr3(vcpu));
1568 		break;
1569 	default:
1570 		KVM_BUG_ON(1, vcpu->kvm);
1571 	}
1572 }
1573 
1574 static void svm_set_vintr(struct vcpu_svm *svm)
1575 {
1576 	struct vmcb_control_area *control;
1577 
1578 	/*
1579 	 * The following fields are ignored when AVIC is enabled
1580 	 */
1581 	WARN_ON(kvm_vcpu_apicv_activated(&svm->vcpu));
1582 
1583 	svm_set_intercept(svm, INTERCEPT_VINTR);
1584 
1585 	/*
1586 	 * Recalculating intercepts may have cleared the VINTR intercept.  If
1587 	 * V_INTR_MASKING is enabled in vmcb12, then the effective RFLAGS.IF
1588 	 * for L1 physical interrupts is L1's RFLAGS.IF at the time of VMRUN.
1589 	 * Requesting an interrupt window if save.RFLAGS.IF=0 is pointless as
1590 	 * interrupts will never be unblocked while L2 is running.
1591 	 */
1592 	if (!svm_is_intercept(svm, INTERCEPT_VINTR))
1593 		return;
1594 
1595 	/*
1596 	 * This is just a dummy VINTR to actually cause a vmexit to happen.
1597 	 * Actual injection of virtual interrupts happens through EVENTINJ.
1598 	 */
1599 	control = &svm->vmcb->control;
1600 	control->int_vector = 0x0;
1601 	control->int_ctl &= ~V_INTR_PRIO_MASK;
1602 	control->int_ctl |= V_IRQ_MASK |
1603 		((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
1604 	vmcb_mark_dirty(svm->vmcb, VMCB_INTR);
1605 }
1606 
1607 static void svm_clear_vintr(struct vcpu_svm *svm)
1608 {
1609 	svm_clr_intercept(svm, INTERCEPT_VINTR);
1610 
1611 	/* Drop int_ctl fields related to VINTR injection.  */
1612 	svm->vmcb->control.int_ctl &= ~V_IRQ_INJECTION_BITS_MASK;
1613 	if (is_guest_mode(&svm->vcpu)) {
1614 		svm->vmcb01.ptr->control.int_ctl &= ~V_IRQ_INJECTION_BITS_MASK;
1615 
1616 		WARN_ON((svm->vmcb->control.int_ctl & V_TPR_MASK) !=
1617 			(svm->nested.ctl.int_ctl & V_TPR_MASK));
1618 
1619 		svm->vmcb->control.int_ctl |= svm->nested.ctl.int_ctl &
1620 			V_IRQ_INJECTION_BITS_MASK;
1621 
1622 		svm->vmcb->control.int_vector = svm->nested.ctl.int_vector;
1623 	}
1624 
1625 	vmcb_mark_dirty(svm->vmcb, VMCB_INTR);
1626 }
1627 
1628 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
1629 {
1630 	struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1631 	struct vmcb_save_area *save01 = &to_svm(vcpu)->vmcb01.ptr->save;
1632 
1633 	switch (seg) {
1634 	case VCPU_SREG_CS: return &save->cs;
1635 	case VCPU_SREG_DS: return &save->ds;
1636 	case VCPU_SREG_ES: return &save->es;
1637 	case VCPU_SREG_FS: return &save01->fs;
1638 	case VCPU_SREG_GS: return &save01->gs;
1639 	case VCPU_SREG_SS: return &save->ss;
1640 	case VCPU_SREG_TR: return &save01->tr;
1641 	case VCPU_SREG_LDTR: return &save01->ldtr;
1642 	}
1643 	BUG();
1644 	return NULL;
1645 }
1646 
1647 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1648 {
1649 	struct vmcb_seg *s = svm_seg(vcpu, seg);
1650 
1651 	return s->base;
1652 }
1653 
1654 static void svm_get_segment(struct kvm_vcpu *vcpu,
1655 			    struct kvm_segment *var, int seg)
1656 {
1657 	struct vmcb_seg *s = svm_seg(vcpu, seg);
1658 
1659 	var->base = s->base;
1660 	var->limit = s->limit;
1661 	var->selector = s->selector;
1662 	var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
1663 	var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
1664 	var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
1665 	var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
1666 	var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
1667 	var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
1668 	var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
1669 
1670 	/*
1671 	 * AMD CPUs circa 2014 track the G bit for all segments except CS.
1672 	 * However, the SVM spec states that the G bit is not observed by the
1673 	 * CPU, and some VMware virtual CPUs drop the G bit for all segments.
1674 	 * So let's synthesize a legal G bit for all segments, this helps
1675 	 * running KVM nested. It also helps cross-vendor migration, because
1676 	 * Intel's vmentry has a check on the 'G' bit.
1677 	 */
1678 	var->g = s->limit > 0xfffff;
1679 
1680 	/*
1681 	 * AMD's VMCB does not have an explicit unusable field, so emulate it
1682 	 * for cross vendor migration purposes by "not present"
1683 	 */
1684 	var->unusable = !var->present;
1685 
1686 	switch (seg) {
1687 	case VCPU_SREG_TR:
1688 		/*
1689 		 * Work around a bug where the busy flag in the tr selector
1690 		 * isn't exposed
1691 		 */
1692 		var->type |= 0x2;
1693 		break;
1694 	case VCPU_SREG_DS:
1695 	case VCPU_SREG_ES:
1696 	case VCPU_SREG_FS:
1697 	case VCPU_SREG_GS:
1698 		/*
1699 		 * The accessed bit must always be set in the segment
1700 		 * descriptor cache, although it can be cleared in the
1701 		 * descriptor, the cached bit always remains at 1. Since
1702 		 * Intel has a check on this, set it here to support
1703 		 * cross-vendor migration.
1704 		 */
1705 		if (!var->unusable)
1706 			var->type |= 0x1;
1707 		break;
1708 	case VCPU_SREG_SS:
1709 		/*
1710 		 * On AMD CPUs sometimes the DB bit in the segment
1711 		 * descriptor is left as 1, although the whole segment has
1712 		 * been made unusable. Clear it here to pass an Intel VMX
1713 		 * entry check when cross vendor migrating.
1714 		 */
1715 		if (var->unusable)
1716 			var->db = 0;
1717 		/* This is symmetric with svm_set_segment() */
1718 		var->dpl = to_svm(vcpu)->vmcb->save.cpl;
1719 		break;
1720 	}
1721 }
1722 
1723 static int svm_get_cpl(struct kvm_vcpu *vcpu)
1724 {
1725 	struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1726 
1727 	return save->cpl;
1728 }
1729 
1730 static void svm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
1731 {
1732 	struct kvm_segment cs;
1733 
1734 	svm_get_segment(vcpu, &cs, VCPU_SREG_CS);
1735 	*db = cs.db;
1736 	*l = cs.l;
1737 }
1738 
1739 static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1740 {
1741 	struct vcpu_svm *svm = to_svm(vcpu);
1742 
1743 	dt->size = svm->vmcb->save.idtr.limit;
1744 	dt->address = svm->vmcb->save.idtr.base;
1745 }
1746 
1747 static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1748 {
1749 	struct vcpu_svm *svm = to_svm(vcpu);
1750 
1751 	svm->vmcb->save.idtr.limit = dt->size;
1752 	svm->vmcb->save.idtr.base = dt->address ;
1753 	vmcb_mark_dirty(svm->vmcb, VMCB_DT);
1754 }
1755 
1756 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1757 {
1758 	struct vcpu_svm *svm = to_svm(vcpu);
1759 
1760 	dt->size = svm->vmcb->save.gdtr.limit;
1761 	dt->address = svm->vmcb->save.gdtr.base;
1762 }
1763 
1764 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1765 {
1766 	struct vcpu_svm *svm = to_svm(vcpu);
1767 
1768 	svm->vmcb->save.gdtr.limit = dt->size;
1769 	svm->vmcb->save.gdtr.base = dt->address ;
1770 	vmcb_mark_dirty(svm->vmcb, VMCB_DT);
1771 }
1772 
1773 static void sev_post_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1774 {
1775 	struct vcpu_svm *svm = to_svm(vcpu);
1776 
1777 	/*
1778 	 * For guests that don't set guest_state_protected, the cr3 update is
1779 	 * handled via kvm_mmu_load() while entering the guest. For guests
1780 	 * that do (SEV-ES/SEV-SNP), the cr3 update needs to be written to
1781 	 * VMCB save area now, since the save area will become the initial
1782 	 * contents of the VMSA, and future VMCB save area updates won't be
1783 	 * seen.
1784 	 */
1785 	if (sev_es_guest(vcpu->kvm)) {
1786 		svm->vmcb->save.cr3 = cr3;
1787 		vmcb_mark_dirty(svm->vmcb, VMCB_CR);
1788 	}
1789 }
1790 
1791 static bool svm_is_valid_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1792 {
1793 	return true;
1794 }
1795 
1796 void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1797 {
1798 	struct vcpu_svm *svm = to_svm(vcpu);
1799 	u64 hcr0 = cr0;
1800 	bool old_paging = is_paging(vcpu);
1801 
1802 #ifdef CONFIG_X86_64
1803 	if (vcpu->arch.efer & EFER_LME && !vcpu->arch.guest_state_protected) {
1804 		if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
1805 			vcpu->arch.efer |= EFER_LMA;
1806 			svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
1807 		}
1808 
1809 		if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
1810 			vcpu->arch.efer &= ~EFER_LMA;
1811 			svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
1812 		}
1813 	}
1814 #endif
1815 	vcpu->arch.cr0 = cr0;
1816 
1817 	if (!npt_enabled) {
1818 		hcr0 |= X86_CR0_PG | X86_CR0_WP;
1819 		if (old_paging != is_paging(vcpu))
1820 			svm_set_cr4(vcpu, kvm_read_cr4(vcpu));
1821 	}
1822 
1823 	/*
1824 	 * re-enable caching here because the QEMU bios
1825 	 * does not do it - this results in some delay at
1826 	 * reboot
1827 	 */
1828 	if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
1829 		hcr0 &= ~(X86_CR0_CD | X86_CR0_NW);
1830 
1831 	svm->vmcb->save.cr0 = hcr0;
1832 	vmcb_mark_dirty(svm->vmcb, VMCB_CR);
1833 
1834 	/*
1835 	 * SEV-ES guests must always keep the CR intercepts cleared. CR
1836 	 * tracking is done using the CR write traps.
1837 	 */
1838 	if (sev_es_guest(vcpu->kvm))
1839 		return;
1840 
1841 	if (hcr0 == cr0) {
1842 		/* Selective CR0 write remains on.  */
1843 		svm_clr_intercept(svm, INTERCEPT_CR0_READ);
1844 		svm_clr_intercept(svm, INTERCEPT_CR0_WRITE);
1845 	} else {
1846 		svm_set_intercept(svm, INTERCEPT_CR0_READ);
1847 		svm_set_intercept(svm, INTERCEPT_CR0_WRITE);
1848 	}
1849 }
1850 
1851 static bool svm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1852 {
1853 	return true;
1854 }
1855 
1856 void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1857 {
1858 	unsigned long host_cr4_mce = cr4_read_shadow() & X86_CR4_MCE;
1859 	unsigned long old_cr4 = vcpu->arch.cr4;
1860 
1861 	if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
1862 		svm_flush_tlb_current(vcpu);
1863 
1864 	vcpu->arch.cr4 = cr4;
1865 	if (!npt_enabled) {
1866 		cr4 |= X86_CR4_PAE;
1867 
1868 		if (!is_paging(vcpu))
1869 			cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
1870 	}
1871 	cr4 |= host_cr4_mce;
1872 	to_svm(vcpu)->vmcb->save.cr4 = cr4;
1873 	vmcb_mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
1874 
1875 	if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
1876 		kvm_update_cpuid_runtime(vcpu);
1877 }
1878 
1879 static void svm_set_segment(struct kvm_vcpu *vcpu,
1880 			    struct kvm_segment *var, int seg)
1881 {
1882 	struct vcpu_svm *svm = to_svm(vcpu);
1883 	struct vmcb_seg *s = svm_seg(vcpu, seg);
1884 
1885 	s->base = var->base;
1886 	s->limit = var->limit;
1887 	s->selector = var->selector;
1888 	s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
1889 	s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
1890 	s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
1891 	s->attrib |= ((var->present & 1) && !var->unusable) << SVM_SELECTOR_P_SHIFT;
1892 	s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
1893 	s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
1894 	s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
1895 	s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
1896 
1897 	/*
1898 	 * This is always accurate, except if SYSRET returned to a segment
1899 	 * with SS.DPL != 3.  Intel does not have this quirk, and always
1900 	 * forces SS.DPL to 3 on sysret, so we ignore that case; fixing it
1901 	 * would entail passing the CPL to userspace and back.
1902 	 */
1903 	if (seg == VCPU_SREG_SS)
1904 		/* This is symmetric with svm_get_segment() */
1905 		svm->vmcb->save.cpl = (var->dpl & 3);
1906 
1907 	vmcb_mark_dirty(svm->vmcb, VMCB_SEG);
1908 }
1909 
1910 static void svm_update_exception_bitmap(struct kvm_vcpu *vcpu)
1911 {
1912 	struct vcpu_svm *svm = to_svm(vcpu);
1913 
1914 	clr_exception_intercept(svm, BP_VECTOR);
1915 
1916 	if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
1917 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
1918 			set_exception_intercept(svm, BP_VECTOR);
1919 	}
1920 }
1921 
1922 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd)
1923 {
1924 	if (sd->next_asid > sd->max_asid) {
1925 		++sd->asid_generation;
1926 		sd->next_asid = sd->min_asid;
1927 		svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
1928 		vmcb_mark_dirty(svm->vmcb, VMCB_ASID);
1929 	}
1930 
1931 	svm->current_vmcb->asid_generation = sd->asid_generation;
1932 	svm->asid = sd->next_asid++;
1933 }
1934 
1935 static void svm_set_dr6(struct vcpu_svm *svm, unsigned long value)
1936 {
1937 	struct vmcb *vmcb = svm->vmcb;
1938 
1939 	if (svm->vcpu.arch.guest_state_protected)
1940 		return;
1941 
1942 	if (unlikely(value != vmcb->save.dr6)) {
1943 		vmcb->save.dr6 = value;
1944 		vmcb_mark_dirty(vmcb, VMCB_DR);
1945 	}
1946 }
1947 
1948 static void svm_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
1949 {
1950 	struct vcpu_svm *svm = to_svm(vcpu);
1951 
1952 	if (vcpu->arch.guest_state_protected)
1953 		return;
1954 
1955 	get_debugreg(vcpu->arch.db[0], 0);
1956 	get_debugreg(vcpu->arch.db[1], 1);
1957 	get_debugreg(vcpu->arch.db[2], 2);
1958 	get_debugreg(vcpu->arch.db[3], 3);
1959 	/*
1960 	 * We cannot reset svm->vmcb->save.dr6 to DR6_ACTIVE_LOW here,
1961 	 * because db_interception might need it.  We can do it before vmentry.
1962 	 */
1963 	vcpu->arch.dr6 = svm->vmcb->save.dr6;
1964 	vcpu->arch.dr7 = svm->vmcb->save.dr7;
1965 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
1966 	set_dr_intercepts(svm);
1967 }
1968 
1969 static void svm_set_dr7(struct kvm_vcpu *vcpu, unsigned long value)
1970 {
1971 	struct vcpu_svm *svm = to_svm(vcpu);
1972 
1973 	if (vcpu->arch.guest_state_protected)
1974 		return;
1975 
1976 	svm->vmcb->save.dr7 = value;
1977 	vmcb_mark_dirty(svm->vmcb, VMCB_DR);
1978 }
1979 
1980 static int pf_interception(struct kvm_vcpu *vcpu)
1981 {
1982 	struct vcpu_svm *svm = to_svm(vcpu);
1983 
1984 	u64 fault_address = svm->vmcb->control.exit_info_2;
1985 	u64 error_code = svm->vmcb->control.exit_info_1;
1986 
1987 	return kvm_handle_page_fault(vcpu, error_code, fault_address,
1988 			static_cpu_has(X86_FEATURE_DECODEASSISTS) ?
1989 			svm->vmcb->control.insn_bytes : NULL,
1990 			svm->vmcb->control.insn_len);
1991 }
1992 
1993 static int npf_interception(struct kvm_vcpu *vcpu)
1994 {
1995 	struct vcpu_svm *svm = to_svm(vcpu);
1996 
1997 	u64 fault_address = svm->vmcb->control.exit_info_2;
1998 	u64 error_code = svm->vmcb->control.exit_info_1;
1999 
2000 	trace_kvm_page_fault(vcpu, fault_address, error_code);
2001 	return kvm_mmu_page_fault(vcpu, fault_address, error_code,
2002 			static_cpu_has(X86_FEATURE_DECODEASSISTS) ?
2003 			svm->vmcb->control.insn_bytes : NULL,
2004 			svm->vmcb->control.insn_len);
2005 }
2006 
2007 static int db_interception(struct kvm_vcpu *vcpu)
2008 {
2009 	struct kvm_run *kvm_run = vcpu->run;
2010 	struct vcpu_svm *svm = to_svm(vcpu);
2011 
2012 	if (!(vcpu->guest_debug &
2013 	      (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
2014 		!svm->nmi_singlestep) {
2015 		u32 payload = svm->vmcb->save.dr6 ^ DR6_ACTIVE_LOW;
2016 		kvm_queue_exception_p(vcpu, DB_VECTOR, payload);
2017 		return 1;
2018 	}
2019 
2020 	if (svm->nmi_singlestep) {
2021 		disable_nmi_singlestep(svm);
2022 		/* Make sure we check for pending NMIs upon entry */
2023 		kvm_make_request(KVM_REQ_EVENT, vcpu);
2024 	}
2025 
2026 	if (vcpu->guest_debug &
2027 	    (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) {
2028 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
2029 		kvm_run->debug.arch.dr6 = svm->vmcb->save.dr6;
2030 		kvm_run->debug.arch.dr7 = svm->vmcb->save.dr7;
2031 		kvm_run->debug.arch.pc =
2032 			svm->vmcb->save.cs.base + svm->vmcb->save.rip;
2033 		kvm_run->debug.arch.exception = DB_VECTOR;
2034 		return 0;
2035 	}
2036 
2037 	return 1;
2038 }
2039 
2040 static int bp_interception(struct kvm_vcpu *vcpu)
2041 {
2042 	struct vcpu_svm *svm = to_svm(vcpu);
2043 	struct kvm_run *kvm_run = vcpu->run;
2044 
2045 	kvm_run->exit_reason = KVM_EXIT_DEBUG;
2046 	kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
2047 	kvm_run->debug.arch.exception = BP_VECTOR;
2048 	return 0;
2049 }
2050 
2051 static int ud_interception(struct kvm_vcpu *vcpu)
2052 {
2053 	return handle_ud(vcpu);
2054 }
2055 
2056 static int ac_interception(struct kvm_vcpu *vcpu)
2057 {
2058 	kvm_queue_exception_e(vcpu, AC_VECTOR, 0);
2059 	return 1;
2060 }
2061 
2062 static bool is_erratum_383(void)
2063 {
2064 	int err, i;
2065 	u64 value;
2066 
2067 	if (!erratum_383_found)
2068 		return false;
2069 
2070 	value = native_read_msr_safe(MSR_IA32_MC0_STATUS, &err);
2071 	if (err)
2072 		return false;
2073 
2074 	/* Bit 62 may or may not be set for this mce */
2075 	value &= ~(1ULL << 62);
2076 
2077 	if (value != 0xb600000000010015ULL)
2078 		return false;
2079 
2080 	/* Clear MCi_STATUS registers */
2081 	for (i = 0; i < 6; ++i)
2082 		native_write_msr_safe(MSR_IA32_MCx_STATUS(i), 0, 0);
2083 
2084 	value = native_read_msr_safe(MSR_IA32_MCG_STATUS, &err);
2085 	if (!err) {
2086 		u32 low, high;
2087 
2088 		value &= ~(1ULL << 2);
2089 		low    = lower_32_bits(value);
2090 		high   = upper_32_bits(value);
2091 
2092 		native_write_msr_safe(MSR_IA32_MCG_STATUS, low, high);
2093 	}
2094 
2095 	/* Flush tlb to evict multi-match entries */
2096 	__flush_tlb_all();
2097 
2098 	return true;
2099 }
2100 
2101 static void svm_handle_mce(struct kvm_vcpu *vcpu)
2102 {
2103 	if (is_erratum_383()) {
2104 		/*
2105 		 * Erratum 383 triggered. Guest state is corrupt so kill the
2106 		 * guest.
2107 		 */
2108 		pr_err("Guest triggered AMD Erratum 383\n");
2109 
2110 		kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2111 
2112 		return;
2113 	}
2114 
2115 	/*
2116 	 * On an #MC intercept the MCE handler is not called automatically in
2117 	 * the host. So do it by hand here.
2118 	 */
2119 	kvm_machine_check();
2120 }
2121 
2122 static int mc_interception(struct kvm_vcpu *vcpu)
2123 {
2124 	return 1;
2125 }
2126 
2127 static int shutdown_interception(struct kvm_vcpu *vcpu)
2128 {
2129 	struct kvm_run *kvm_run = vcpu->run;
2130 	struct vcpu_svm *svm = to_svm(vcpu);
2131 
2132 	/*
2133 	 * The VM save area has already been encrypted so it
2134 	 * cannot be reinitialized - just terminate.
2135 	 */
2136 	if (sev_es_guest(vcpu->kvm))
2137 		return -EINVAL;
2138 
2139 	/*
2140 	 * VMCB is undefined after a SHUTDOWN intercept.  INIT the vCPU to put
2141 	 * the VMCB in a known good state.  Unfortuately, KVM doesn't have
2142 	 * KVM_MP_STATE_SHUTDOWN and can't add it without potentially breaking
2143 	 * userspace.  At a platform view, INIT is acceptable behavior as
2144 	 * there exist bare metal platforms that automatically INIT the CPU
2145 	 * in response to shutdown.
2146 	 */
2147 	clear_page(svm->vmcb);
2148 	kvm_vcpu_reset(vcpu, true);
2149 
2150 	kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
2151 	return 0;
2152 }
2153 
2154 static int io_interception(struct kvm_vcpu *vcpu)
2155 {
2156 	struct vcpu_svm *svm = to_svm(vcpu);
2157 	u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
2158 	int size, in, string;
2159 	unsigned port;
2160 
2161 	++vcpu->stat.io_exits;
2162 	string = (io_info & SVM_IOIO_STR_MASK) != 0;
2163 	in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
2164 	port = io_info >> 16;
2165 	size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
2166 
2167 	if (string) {
2168 		if (sev_es_guest(vcpu->kvm))
2169 			return sev_es_string_io(svm, size, port, in);
2170 		else
2171 			return kvm_emulate_instruction(vcpu, 0);
2172 	}
2173 
2174 	svm->next_rip = svm->vmcb->control.exit_info_2;
2175 
2176 	return kvm_fast_pio(vcpu, size, port, in);
2177 }
2178 
2179 static int nmi_interception(struct kvm_vcpu *vcpu)
2180 {
2181 	return 1;
2182 }
2183 
2184 static int smi_interception(struct kvm_vcpu *vcpu)
2185 {
2186 	return 1;
2187 }
2188 
2189 static int intr_interception(struct kvm_vcpu *vcpu)
2190 {
2191 	++vcpu->stat.irq_exits;
2192 	return 1;
2193 }
2194 
2195 static int vmload_vmsave_interception(struct kvm_vcpu *vcpu, bool vmload)
2196 {
2197 	struct vcpu_svm *svm = to_svm(vcpu);
2198 	struct vmcb *vmcb12;
2199 	struct kvm_host_map map;
2200 	int ret;
2201 
2202 	if (nested_svm_check_permissions(vcpu))
2203 		return 1;
2204 
2205 	ret = kvm_vcpu_map(vcpu, gpa_to_gfn(svm->vmcb->save.rax), &map);
2206 	if (ret) {
2207 		if (ret == -EINVAL)
2208 			kvm_inject_gp(vcpu, 0);
2209 		return 1;
2210 	}
2211 
2212 	vmcb12 = map.hva;
2213 
2214 	ret = kvm_skip_emulated_instruction(vcpu);
2215 
2216 	if (vmload) {
2217 		svm_copy_vmloadsave_state(svm->vmcb, vmcb12);
2218 		svm->sysenter_eip_hi = 0;
2219 		svm->sysenter_esp_hi = 0;
2220 	} else {
2221 		svm_copy_vmloadsave_state(vmcb12, svm->vmcb);
2222 	}
2223 
2224 	kvm_vcpu_unmap(vcpu, &map, true);
2225 
2226 	return ret;
2227 }
2228 
2229 static int vmload_interception(struct kvm_vcpu *vcpu)
2230 {
2231 	return vmload_vmsave_interception(vcpu, true);
2232 }
2233 
2234 static int vmsave_interception(struct kvm_vcpu *vcpu)
2235 {
2236 	return vmload_vmsave_interception(vcpu, false);
2237 }
2238 
2239 static int vmrun_interception(struct kvm_vcpu *vcpu)
2240 {
2241 	if (nested_svm_check_permissions(vcpu))
2242 		return 1;
2243 
2244 	return nested_svm_vmrun(vcpu);
2245 }
2246 
2247 enum {
2248 	NONE_SVM_INSTR,
2249 	SVM_INSTR_VMRUN,
2250 	SVM_INSTR_VMLOAD,
2251 	SVM_INSTR_VMSAVE,
2252 };
2253 
2254 /* Return NONE_SVM_INSTR if not SVM instrs, otherwise return decode result */
2255 static int svm_instr_opcode(struct kvm_vcpu *vcpu)
2256 {
2257 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
2258 
2259 	if (ctxt->b != 0x1 || ctxt->opcode_len != 2)
2260 		return NONE_SVM_INSTR;
2261 
2262 	switch (ctxt->modrm) {
2263 	case 0xd8: /* VMRUN */
2264 		return SVM_INSTR_VMRUN;
2265 	case 0xda: /* VMLOAD */
2266 		return SVM_INSTR_VMLOAD;
2267 	case 0xdb: /* VMSAVE */
2268 		return SVM_INSTR_VMSAVE;
2269 	default:
2270 		break;
2271 	}
2272 
2273 	return NONE_SVM_INSTR;
2274 }
2275 
2276 static int emulate_svm_instr(struct kvm_vcpu *vcpu, int opcode)
2277 {
2278 	const int guest_mode_exit_codes[] = {
2279 		[SVM_INSTR_VMRUN] = SVM_EXIT_VMRUN,
2280 		[SVM_INSTR_VMLOAD] = SVM_EXIT_VMLOAD,
2281 		[SVM_INSTR_VMSAVE] = SVM_EXIT_VMSAVE,
2282 	};
2283 	int (*const svm_instr_handlers[])(struct kvm_vcpu *vcpu) = {
2284 		[SVM_INSTR_VMRUN] = vmrun_interception,
2285 		[SVM_INSTR_VMLOAD] = vmload_interception,
2286 		[SVM_INSTR_VMSAVE] = vmsave_interception,
2287 	};
2288 	struct vcpu_svm *svm = to_svm(vcpu);
2289 	int ret;
2290 
2291 	if (is_guest_mode(vcpu)) {
2292 		/* Returns '1' or -errno on failure, '0' on success. */
2293 		ret = nested_svm_simple_vmexit(svm, guest_mode_exit_codes[opcode]);
2294 		if (ret)
2295 			return ret;
2296 		return 1;
2297 	}
2298 	return svm_instr_handlers[opcode](vcpu);
2299 }
2300 
2301 /*
2302  * #GP handling code. Note that #GP can be triggered under the following two
2303  * cases:
2304  *   1) SVM VM-related instructions (VMRUN/VMSAVE/VMLOAD) that trigger #GP on
2305  *      some AMD CPUs when EAX of these instructions are in the reserved memory
2306  *      regions (e.g. SMM memory on host).
2307  *   2) VMware backdoor
2308  */
2309 static int gp_interception(struct kvm_vcpu *vcpu)
2310 {
2311 	struct vcpu_svm *svm = to_svm(vcpu);
2312 	u32 error_code = svm->vmcb->control.exit_info_1;
2313 	int opcode;
2314 
2315 	/* Both #GP cases have zero error_code */
2316 	if (error_code)
2317 		goto reinject;
2318 
2319 	/* Decode the instruction for usage later */
2320 	if (x86_decode_emulated_instruction(vcpu, 0, NULL, 0) != EMULATION_OK)
2321 		goto reinject;
2322 
2323 	opcode = svm_instr_opcode(vcpu);
2324 
2325 	if (opcode == NONE_SVM_INSTR) {
2326 		if (!enable_vmware_backdoor)
2327 			goto reinject;
2328 
2329 		/*
2330 		 * VMware backdoor emulation on #GP interception only handles
2331 		 * IN{S}, OUT{S}, and RDPMC.
2332 		 */
2333 		if (!is_guest_mode(vcpu))
2334 			return kvm_emulate_instruction(vcpu,
2335 				EMULTYPE_VMWARE_GP | EMULTYPE_NO_DECODE);
2336 	} else {
2337 		/* All SVM instructions expect page aligned RAX */
2338 		if (svm->vmcb->save.rax & ~PAGE_MASK)
2339 			goto reinject;
2340 
2341 		return emulate_svm_instr(vcpu, opcode);
2342 	}
2343 
2344 reinject:
2345 	kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
2346 	return 1;
2347 }
2348 
2349 void svm_set_gif(struct vcpu_svm *svm, bool value)
2350 {
2351 	if (value) {
2352 		/*
2353 		 * If VGIF is enabled, the STGI intercept is only added to
2354 		 * detect the opening of the SMI/NMI window; remove it now.
2355 		 * Likewise, clear the VINTR intercept, we will set it
2356 		 * again while processing KVM_REQ_EVENT if needed.
2357 		 */
2358 		if (vgif)
2359 			svm_clr_intercept(svm, INTERCEPT_STGI);
2360 		if (svm_is_intercept(svm, INTERCEPT_VINTR))
2361 			svm_clear_vintr(svm);
2362 
2363 		enable_gif(svm);
2364 		if (svm->vcpu.arch.smi_pending ||
2365 		    svm->vcpu.arch.nmi_pending ||
2366 		    kvm_cpu_has_injectable_intr(&svm->vcpu) ||
2367 		    kvm_apic_has_pending_init_or_sipi(&svm->vcpu))
2368 			kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
2369 	} else {
2370 		disable_gif(svm);
2371 
2372 		/*
2373 		 * After a CLGI no interrupts should come.  But if vGIF is
2374 		 * in use, we still rely on the VINTR intercept (rather than
2375 		 * STGI) to detect an open interrupt window.
2376 		*/
2377 		if (!vgif)
2378 			svm_clear_vintr(svm);
2379 	}
2380 }
2381 
2382 static int stgi_interception(struct kvm_vcpu *vcpu)
2383 {
2384 	int ret;
2385 
2386 	if (nested_svm_check_permissions(vcpu))
2387 		return 1;
2388 
2389 	ret = kvm_skip_emulated_instruction(vcpu);
2390 	svm_set_gif(to_svm(vcpu), true);
2391 	return ret;
2392 }
2393 
2394 static int clgi_interception(struct kvm_vcpu *vcpu)
2395 {
2396 	int ret;
2397 
2398 	if (nested_svm_check_permissions(vcpu))
2399 		return 1;
2400 
2401 	ret = kvm_skip_emulated_instruction(vcpu);
2402 	svm_set_gif(to_svm(vcpu), false);
2403 	return ret;
2404 }
2405 
2406 static int invlpga_interception(struct kvm_vcpu *vcpu)
2407 {
2408 	gva_t gva = kvm_rax_read(vcpu);
2409 	u32 asid = kvm_rcx_read(vcpu);
2410 
2411 	/* FIXME: Handle an address size prefix. */
2412 	if (!is_long_mode(vcpu))
2413 		gva = (u32)gva;
2414 
2415 	trace_kvm_invlpga(to_svm(vcpu)->vmcb->save.rip, asid, gva);
2416 
2417 	/* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
2418 	kvm_mmu_invlpg(vcpu, gva);
2419 
2420 	return kvm_skip_emulated_instruction(vcpu);
2421 }
2422 
2423 static int skinit_interception(struct kvm_vcpu *vcpu)
2424 {
2425 	trace_kvm_skinit(to_svm(vcpu)->vmcb->save.rip, kvm_rax_read(vcpu));
2426 
2427 	kvm_queue_exception(vcpu, UD_VECTOR);
2428 	return 1;
2429 }
2430 
2431 static int task_switch_interception(struct kvm_vcpu *vcpu)
2432 {
2433 	struct vcpu_svm *svm = to_svm(vcpu);
2434 	u16 tss_selector;
2435 	int reason;
2436 	int int_type = svm->vmcb->control.exit_int_info &
2437 		SVM_EXITINTINFO_TYPE_MASK;
2438 	int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
2439 	uint32_t type =
2440 		svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
2441 	uint32_t idt_v =
2442 		svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
2443 	bool has_error_code = false;
2444 	u32 error_code = 0;
2445 
2446 	tss_selector = (u16)svm->vmcb->control.exit_info_1;
2447 
2448 	if (svm->vmcb->control.exit_info_2 &
2449 	    (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
2450 		reason = TASK_SWITCH_IRET;
2451 	else if (svm->vmcb->control.exit_info_2 &
2452 		 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
2453 		reason = TASK_SWITCH_JMP;
2454 	else if (idt_v)
2455 		reason = TASK_SWITCH_GATE;
2456 	else
2457 		reason = TASK_SWITCH_CALL;
2458 
2459 	if (reason == TASK_SWITCH_GATE) {
2460 		switch (type) {
2461 		case SVM_EXITINTINFO_TYPE_NMI:
2462 			vcpu->arch.nmi_injected = false;
2463 			break;
2464 		case SVM_EXITINTINFO_TYPE_EXEPT:
2465 			if (svm->vmcb->control.exit_info_2 &
2466 			    (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE)) {
2467 				has_error_code = true;
2468 				error_code =
2469 					(u32)svm->vmcb->control.exit_info_2;
2470 			}
2471 			kvm_clear_exception_queue(vcpu);
2472 			break;
2473 		case SVM_EXITINTINFO_TYPE_INTR:
2474 		case SVM_EXITINTINFO_TYPE_SOFT:
2475 			kvm_clear_interrupt_queue(vcpu);
2476 			break;
2477 		default:
2478 			break;
2479 		}
2480 	}
2481 
2482 	if (reason != TASK_SWITCH_GATE ||
2483 	    int_type == SVM_EXITINTINFO_TYPE_SOFT ||
2484 	    (int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
2485 	     (int_vec == OF_VECTOR || int_vec == BP_VECTOR))) {
2486 		if (!svm_skip_emulated_instruction(vcpu))
2487 			return 0;
2488 	}
2489 
2490 	if (int_type != SVM_EXITINTINFO_TYPE_SOFT)
2491 		int_vec = -1;
2492 
2493 	return kvm_task_switch(vcpu, tss_selector, int_vec, reason,
2494 			       has_error_code, error_code);
2495 }
2496 
2497 static void svm_clr_iret_intercept(struct vcpu_svm *svm)
2498 {
2499 	if (!sev_es_guest(svm->vcpu.kvm))
2500 		svm_clr_intercept(svm, INTERCEPT_IRET);
2501 }
2502 
2503 static void svm_set_iret_intercept(struct vcpu_svm *svm)
2504 {
2505 	if (!sev_es_guest(svm->vcpu.kvm))
2506 		svm_set_intercept(svm, INTERCEPT_IRET);
2507 }
2508 
2509 static int iret_interception(struct kvm_vcpu *vcpu)
2510 {
2511 	struct vcpu_svm *svm = to_svm(vcpu);
2512 
2513 	++vcpu->stat.nmi_window_exits;
2514 	svm->awaiting_iret_completion = true;
2515 
2516 	svm_clr_iret_intercept(svm);
2517 	if (!sev_es_guest(vcpu->kvm))
2518 		svm->nmi_iret_rip = kvm_rip_read(vcpu);
2519 
2520 	kvm_make_request(KVM_REQ_EVENT, vcpu);
2521 	return 1;
2522 }
2523 
2524 static int invlpg_interception(struct kvm_vcpu *vcpu)
2525 {
2526 	if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
2527 		return kvm_emulate_instruction(vcpu, 0);
2528 
2529 	kvm_mmu_invlpg(vcpu, to_svm(vcpu)->vmcb->control.exit_info_1);
2530 	return kvm_skip_emulated_instruction(vcpu);
2531 }
2532 
2533 static int emulate_on_interception(struct kvm_vcpu *vcpu)
2534 {
2535 	return kvm_emulate_instruction(vcpu, 0);
2536 }
2537 
2538 static int rsm_interception(struct kvm_vcpu *vcpu)
2539 {
2540 	return kvm_emulate_instruction_from_buffer(vcpu, rsm_ins_bytes, 2);
2541 }
2542 
2543 static bool check_selective_cr0_intercepted(struct kvm_vcpu *vcpu,
2544 					    unsigned long val)
2545 {
2546 	struct vcpu_svm *svm = to_svm(vcpu);
2547 	unsigned long cr0 = vcpu->arch.cr0;
2548 	bool ret = false;
2549 
2550 	if (!is_guest_mode(vcpu) ||
2551 	    (!(vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_SELECTIVE_CR0))))
2552 		return false;
2553 
2554 	cr0 &= ~SVM_CR0_SELECTIVE_MASK;
2555 	val &= ~SVM_CR0_SELECTIVE_MASK;
2556 
2557 	if (cr0 ^ val) {
2558 		svm->vmcb->control.exit_code = SVM_EXIT_CR0_SEL_WRITE;
2559 		ret = (nested_svm_exit_handled(svm) == NESTED_EXIT_DONE);
2560 	}
2561 
2562 	return ret;
2563 }
2564 
2565 #define CR_VALID (1ULL << 63)
2566 
2567 static int cr_interception(struct kvm_vcpu *vcpu)
2568 {
2569 	struct vcpu_svm *svm = to_svm(vcpu);
2570 	int reg, cr;
2571 	unsigned long val;
2572 	int err;
2573 
2574 	if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
2575 		return emulate_on_interception(vcpu);
2576 
2577 	if (unlikely((svm->vmcb->control.exit_info_1 & CR_VALID) == 0))
2578 		return emulate_on_interception(vcpu);
2579 
2580 	reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
2581 	if (svm->vmcb->control.exit_code == SVM_EXIT_CR0_SEL_WRITE)
2582 		cr = SVM_EXIT_WRITE_CR0 - SVM_EXIT_READ_CR0;
2583 	else
2584 		cr = svm->vmcb->control.exit_code - SVM_EXIT_READ_CR0;
2585 
2586 	err = 0;
2587 	if (cr >= 16) { /* mov to cr */
2588 		cr -= 16;
2589 		val = kvm_register_read(vcpu, reg);
2590 		trace_kvm_cr_write(cr, val);
2591 		switch (cr) {
2592 		case 0:
2593 			if (!check_selective_cr0_intercepted(vcpu, val))
2594 				err = kvm_set_cr0(vcpu, val);
2595 			else
2596 				return 1;
2597 
2598 			break;
2599 		case 3:
2600 			err = kvm_set_cr3(vcpu, val);
2601 			break;
2602 		case 4:
2603 			err = kvm_set_cr4(vcpu, val);
2604 			break;
2605 		case 8:
2606 			err = kvm_set_cr8(vcpu, val);
2607 			break;
2608 		default:
2609 			WARN(1, "unhandled write to CR%d", cr);
2610 			kvm_queue_exception(vcpu, UD_VECTOR);
2611 			return 1;
2612 		}
2613 	} else { /* mov from cr */
2614 		switch (cr) {
2615 		case 0:
2616 			val = kvm_read_cr0(vcpu);
2617 			break;
2618 		case 2:
2619 			val = vcpu->arch.cr2;
2620 			break;
2621 		case 3:
2622 			val = kvm_read_cr3(vcpu);
2623 			break;
2624 		case 4:
2625 			val = kvm_read_cr4(vcpu);
2626 			break;
2627 		case 8:
2628 			val = kvm_get_cr8(vcpu);
2629 			break;
2630 		default:
2631 			WARN(1, "unhandled read from CR%d", cr);
2632 			kvm_queue_exception(vcpu, UD_VECTOR);
2633 			return 1;
2634 		}
2635 		kvm_register_write(vcpu, reg, val);
2636 		trace_kvm_cr_read(cr, val);
2637 	}
2638 	return kvm_complete_insn_gp(vcpu, err);
2639 }
2640 
2641 static int cr_trap(struct kvm_vcpu *vcpu)
2642 {
2643 	struct vcpu_svm *svm = to_svm(vcpu);
2644 	unsigned long old_value, new_value;
2645 	unsigned int cr;
2646 	int ret = 0;
2647 
2648 	new_value = (unsigned long)svm->vmcb->control.exit_info_1;
2649 
2650 	cr = svm->vmcb->control.exit_code - SVM_EXIT_CR0_WRITE_TRAP;
2651 	switch (cr) {
2652 	case 0:
2653 		old_value = kvm_read_cr0(vcpu);
2654 		svm_set_cr0(vcpu, new_value);
2655 
2656 		kvm_post_set_cr0(vcpu, old_value, new_value);
2657 		break;
2658 	case 4:
2659 		old_value = kvm_read_cr4(vcpu);
2660 		svm_set_cr4(vcpu, new_value);
2661 
2662 		kvm_post_set_cr4(vcpu, old_value, new_value);
2663 		break;
2664 	case 8:
2665 		ret = kvm_set_cr8(vcpu, new_value);
2666 		break;
2667 	default:
2668 		WARN(1, "unhandled CR%d write trap", cr);
2669 		kvm_queue_exception(vcpu, UD_VECTOR);
2670 		return 1;
2671 	}
2672 
2673 	return kvm_complete_insn_gp(vcpu, ret);
2674 }
2675 
2676 static int dr_interception(struct kvm_vcpu *vcpu)
2677 {
2678 	struct vcpu_svm *svm = to_svm(vcpu);
2679 	int reg, dr;
2680 	unsigned long val;
2681 	int err = 0;
2682 
2683 	if (vcpu->guest_debug == 0) {
2684 		/*
2685 		 * No more DR vmexits; force a reload of the debug registers
2686 		 * and reenter on this instruction.  The next vmexit will
2687 		 * retrieve the full state of the debug registers.
2688 		 */
2689 		clr_dr_intercepts(svm);
2690 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
2691 		return 1;
2692 	}
2693 
2694 	if (!boot_cpu_has(X86_FEATURE_DECODEASSISTS))
2695 		return emulate_on_interception(vcpu);
2696 
2697 	reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
2698 	dr = svm->vmcb->control.exit_code - SVM_EXIT_READ_DR0;
2699 	if (dr >= 16) { /* mov to DRn  */
2700 		dr -= 16;
2701 		val = kvm_register_read(vcpu, reg);
2702 		err = kvm_set_dr(vcpu, dr, val);
2703 	} else {
2704 		kvm_get_dr(vcpu, dr, &val);
2705 		kvm_register_write(vcpu, reg, val);
2706 	}
2707 
2708 	return kvm_complete_insn_gp(vcpu, err);
2709 }
2710 
2711 static int cr8_write_interception(struct kvm_vcpu *vcpu)
2712 {
2713 	int r;
2714 
2715 	u8 cr8_prev = kvm_get_cr8(vcpu);
2716 	/* instruction emulation calls kvm_set_cr8() */
2717 	r = cr_interception(vcpu);
2718 	if (lapic_in_kernel(vcpu))
2719 		return r;
2720 	if (cr8_prev <= kvm_get_cr8(vcpu))
2721 		return r;
2722 	vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
2723 	return 0;
2724 }
2725 
2726 static int efer_trap(struct kvm_vcpu *vcpu)
2727 {
2728 	struct msr_data msr_info;
2729 	int ret;
2730 
2731 	/*
2732 	 * Clear the EFER_SVME bit from EFER. The SVM code always sets this
2733 	 * bit in svm_set_efer(), but __kvm_valid_efer() checks it against
2734 	 * whether the guest has X86_FEATURE_SVM - this avoids a failure if
2735 	 * the guest doesn't have X86_FEATURE_SVM.
2736 	 */
2737 	msr_info.host_initiated = false;
2738 	msr_info.index = MSR_EFER;
2739 	msr_info.data = to_svm(vcpu)->vmcb->control.exit_info_1 & ~EFER_SVME;
2740 	ret = kvm_set_msr_common(vcpu, &msr_info);
2741 
2742 	return kvm_complete_insn_gp(vcpu, ret);
2743 }
2744 
2745 static int svm_get_msr_feature(struct kvm_msr_entry *msr)
2746 {
2747 	msr->data = 0;
2748 
2749 	switch (msr->index) {
2750 	case MSR_AMD64_DE_CFG:
2751 		if (cpu_feature_enabled(X86_FEATURE_LFENCE_RDTSC))
2752 			msr->data |= MSR_AMD64_DE_CFG_LFENCE_SERIALIZE;
2753 		break;
2754 	default:
2755 		return KVM_MSR_RET_INVALID;
2756 	}
2757 
2758 	return 0;
2759 }
2760 
2761 static int svm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2762 {
2763 	struct vcpu_svm *svm = to_svm(vcpu);
2764 
2765 	switch (msr_info->index) {
2766 	case MSR_AMD64_TSC_RATIO:
2767 		if (!msr_info->host_initiated && !svm->tsc_scaling_enabled)
2768 			return 1;
2769 		msr_info->data = svm->tsc_ratio_msr;
2770 		break;
2771 	case MSR_STAR:
2772 		msr_info->data = svm->vmcb01.ptr->save.star;
2773 		break;
2774 #ifdef CONFIG_X86_64
2775 	case MSR_LSTAR:
2776 		msr_info->data = svm->vmcb01.ptr->save.lstar;
2777 		break;
2778 	case MSR_CSTAR:
2779 		msr_info->data = svm->vmcb01.ptr->save.cstar;
2780 		break;
2781 	case MSR_KERNEL_GS_BASE:
2782 		msr_info->data = svm->vmcb01.ptr->save.kernel_gs_base;
2783 		break;
2784 	case MSR_SYSCALL_MASK:
2785 		msr_info->data = svm->vmcb01.ptr->save.sfmask;
2786 		break;
2787 #endif
2788 	case MSR_IA32_SYSENTER_CS:
2789 		msr_info->data = svm->vmcb01.ptr->save.sysenter_cs;
2790 		break;
2791 	case MSR_IA32_SYSENTER_EIP:
2792 		msr_info->data = (u32)svm->vmcb01.ptr->save.sysenter_eip;
2793 		if (guest_cpuid_is_intel(vcpu))
2794 			msr_info->data |= (u64)svm->sysenter_eip_hi << 32;
2795 		break;
2796 	case MSR_IA32_SYSENTER_ESP:
2797 		msr_info->data = svm->vmcb01.ptr->save.sysenter_esp;
2798 		if (guest_cpuid_is_intel(vcpu))
2799 			msr_info->data |= (u64)svm->sysenter_esp_hi << 32;
2800 		break;
2801 	case MSR_TSC_AUX:
2802 		msr_info->data = svm->tsc_aux;
2803 		break;
2804 	case MSR_IA32_DEBUGCTLMSR:
2805 	case MSR_IA32_LASTBRANCHFROMIP:
2806 	case MSR_IA32_LASTBRANCHTOIP:
2807 	case MSR_IA32_LASTINTFROMIP:
2808 	case MSR_IA32_LASTINTTOIP:
2809 		msr_info->data = svm_get_lbr_msr(svm, msr_info->index);
2810 		break;
2811 	case MSR_VM_HSAVE_PA:
2812 		msr_info->data = svm->nested.hsave_msr;
2813 		break;
2814 	case MSR_VM_CR:
2815 		msr_info->data = svm->nested.vm_cr_msr;
2816 		break;
2817 	case MSR_IA32_SPEC_CTRL:
2818 		if (!msr_info->host_initiated &&
2819 		    !guest_has_spec_ctrl_msr(vcpu))
2820 			return 1;
2821 
2822 		if (boot_cpu_has(X86_FEATURE_V_SPEC_CTRL))
2823 			msr_info->data = svm->vmcb->save.spec_ctrl;
2824 		else
2825 			msr_info->data = svm->spec_ctrl;
2826 		break;
2827 	case MSR_AMD64_VIRT_SPEC_CTRL:
2828 		if (!msr_info->host_initiated &&
2829 		    !guest_cpuid_has(vcpu, X86_FEATURE_VIRT_SSBD))
2830 			return 1;
2831 
2832 		msr_info->data = svm->virt_spec_ctrl;
2833 		break;
2834 	case MSR_F15H_IC_CFG: {
2835 
2836 		int family, model;
2837 
2838 		family = guest_cpuid_family(vcpu);
2839 		model  = guest_cpuid_model(vcpu);
2840 
2841 		if (family < 0 || model < 0)
2842 			return kvm_get_msr_common(vcpu, msr_info);
2843 
2844 		msr_info->data = 0;
2845 
2846 		if (family == 0x15 &&
2847 		    (model >= 0x2 && model < 0x20))
2848 			msr_info->data = 0x1E;
2849 		}
2850 		break;
2851 	case MSR_AMD64_DE_CFG:
2852 		msr_info->data = svm->msr_decfg;
2853 		break;
2854 	default:
2855 		return kvm_get_msr_common(vcpu, msr_info);
2856 	}
2857 	return 0;
2858 }
2859 
2860 static int svm_complete_emulated_msr(struct kvm_vcpu *vcpu, int err)
2861 {
2862 	struct vcpu_svm *svm = to_svm(vcpu);
2863 	if (!err || !sev_es_guest(vcpu->kvm) || WARN_ON_ONCE(!svm->sev_es.ghcb))
2864 		return kvm_complete_insn_gp(vcpu, err);
2865 
2866 	ghcb_set_sw_exit_info_1(svm->sev_es.ghcb, 1);
2867 	ghcb_set_sw_exit_info_2(svm->sev_es.ghcb,
2868 				X86_TRAP_GP |
2869 				SVM_EVTINJ_TYPE_EXEPT |
2870 				SVM_EVTINJ_VALID);
2871 	return 1;
2872 }
2873 
2874 static int svm_set_vm_cr(struct kvm_vcpu *vcpu, u64 data)
2875 {
2876 	struct vcpu_svm *svm = to_svm(vcpu);
2877 	int svm_dis, chg_mask;
2878 
2879 	if (data & ~SVM_VM_CR_VALID_MASK)
2880 		return 1;
2881 
2882 	chg_mask = SVM_VM_CR_VALID_MASK;
2883 
2884 	if (svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK)
2885 		chg_mask &= ~(SVM_VM_CR_SVM_LOCK_MASK | SVM_VM_CR_SVM_DIS_MASK);
2886 
2887 	svm->nested.vm_cr_msr &= ~chg_mask;
2888 	svm->nested.vm_cr_msr |= (data & chg_mask);
2889 
2890 	svm_dis = svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK;
2891 
2892 	/* check for svm_disable while efer.svme is set */
2893 	if (svm_dis && (vcpu->arch.efer & EFER_SVME))
2894 		return 1;
2895 
2896 	return 0;
2897 }
2898 
2899 static int svm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2900 {
2901 	struct vcpu_svm *svm = to_svm(vcpu);
2902 	int ret = 0;
2903 
2904 	u32 ecx = msr->index;
2905 	u64 data = msr->data;
2906 	switch (ecx) {
2907 	case MSR_AMD64_TSC_RATIO:
2908 
2909 		if (!svm->tsc_scaling_enabled) {
2910 
2911 			if (!msr->host_initiated)
2912 				return 1;
2913 			/*
2914 			 * In case TSC scaling is not enabled, always
2915 			 * leave this MSR at the default value.
2916 			 *
2917 			 * Due to bug in qemu 6.2.0, it would try to set
2918 			 * this msr to 0 if tsc scaling is not enabled.
2919 			 * Ignore this value as well.
2920 			 */
2921 			if (data != 0 && data != svm->tsc_ratio_msr)
2922 				return 1;
2923 			break;
2924 		}
2925 
2926 		if (data & SVM_TSC_RATIO_RSVD)
2927 			return 1;
2928 
2929 		svm->tsc_ratio_msr = data;
2930 
2931 		if (svm->tsc_scaling_enabled && is_guest_mode(vcpu))
2932 			nested_svm_update_tsc_ratio_msr(vcpu);
2933 
2934 		break;
2935 	case MSR_IA32_CR_PAT:
2936 		ret = kvm_set_msr_common(vcpu, msr);
2937 		if (ret)
2938 			break;
2939 
2940 		svm->vmcb01.ptr->save.g_pat = data;
2941 		if (is_guest_mode(vcpu))
2942 			nested_vmcb02_compute_g_pat(svm);
2943 		vmcb_mark_dirty(svm->vmcb, VMCB_NPT);
2944 		break;
2945 	case MSR_IA32_SPEC_CTRL:
2946 		if (!msr->host_initiated &&
2947 		    !guest_has_spec_ctrl_msr(vcpu))
2948 			return 1;
2949 
2950 		if (kvm_spec_ctrl_test_value(data))
2951 			return 1;
2952 
2953 		if (boot_cpu_has(X86_FEATURE_V_SPEC_CTRL))
2954 			svm->vmcb->save.spec_ctrl = data;
2955 		else
2956 			svm->spec_ctrl = data;
2957 		if (!data)
2958 			break;
2959 
2960 		/*
2961 		 * For non-nested:
2962 		 * When it's written (to non-zero) for the first time, pass
2963 		 * it through.
2964 		 *
2965 		 * For nested:
2966 		 * The handling of the MSR bitmap for L2 guests is done in
2967 		 * nested_svm_vmrun_msrpm.
2968 		 * We update the L1 MSR bit as well since it will end up
2969 		 * touching the MSR anyway now.
2970 		 */
2971 		set_msr_interception(vcpu, svm->msrpm, MSR_IA32_SPEC_CTRL, 1, 1);
2972 		break;
2973 	case MSR_AMD64_VIRT_SPEC_CTRL:
2974 		if (!msr->host_initiated &&
2975 		    !guest_cpuid_has(vcpu, X86_FEATURE_VIRT_SSBD))
2976 			return 1;
2977 
2978 		if (data & ~SPEC_CTRL_SSBD)
2979 			return 1;
2980 
2981 		svm->virt_spec_ctrl = data;
2982 		break;
2983 	case MSR_STAR:
2984 		svm->vmcb01.ptr->save.star = data;
2985 		break;
2986 #ifdef CONFIG_X86_64
2987 	case MSR_LSTAR:
2988 		svm->vmcb01.ptr->save.lstar = data;
2989 		break;
2990 	case MSR_CSTAR:
2991 		svm->vmcb01.ptr->save.cstar = data;
2992 		break;
2993 	case MSR_KERNEL_GS_BASE:
2994 		svm->vmcb01.ptr->save.kernel_gs_base = data;
2995 		break;
2996 	case MSR_SYSCALL_MASK:
2997 		svm->vmcb01.ptr->save.sfmask = data;
2998 		break;
2999 #endif
3000 	case MSR_IA32_SYSENTER_CS:
3001 		svm->vmcb01.ptr->save.sysenter_cs = data;
3002 		break;
3003 	case MSR_IA32_SYSENTER_EIP:
3004 		svm->vmcb01.ptr->save.sysenter_eip = (u32)data;
3005 		/*
3006 		 * We only intercept the MSR_IA32_SYSENTER_{EIP|ESP} msrs
3007 		 * when we spoof an Intel vendor ID (for cross vendor migration).
3008 		 * In this case we use this intercept to track the high
3009 		 * 32 bit part of these msrs to support Intel's
3010 		 * implementation of SYSENTER/SYSEXIT.
3011 		 */
3012 		svm->sysenter_eip_hi = guest_cpuid_is_intel(vcpu) ? (data >> 32) : 0;
3013 		break;
3014 	case MSR_IA32_SYSENTER_ESP:
3015 		svm->vmcb01.ptr->save.sysenter_esp = (u32)data;
3016 		svm->sysenter_esp_hi = guest_cpuid_is_intel(vcpu) ? (data >> 32) : 0;
3017 		break;
3018 	case MSR_TSC_AUX:
3019 		/*
3020 		 * TSC_AUX is usually changed only during boot and never read
3021 		 * directly.  Intercept TSC_AUX instead of exposing it to the
3022 		 * guest via direct_access_msrs, and switch it via user return.
3023 		 */
3024 		preempt_disable();
3025 		ret = kvm_set_user_return_msr(tsc_aux_uret_slot, data, -1ull);
3026 		preempt_enable();
3027 		if (ret)
3028 			break;
3029 
3030 		svm->tsc_aux = data;
3031 		break;
3032 	case MSR_IA32_DEBUGCTLMSR:
3033 		if (!lbrv) {
3034 			kvm_pr_unimpl_wrmsr(vcpu, ecx, data);
3035 			break;
3036 		}
3037 		if (data & DEBUGCTL_RESERVED_BITS)
3038 			return 1;
3039 
3040 		if (svm->vmcb->control.virt_ext & LBR_CTL_ENABLE_MASK)
3041 			svm->vmcb->save.dbgctl = data;
3042 		else
3043 			svm->vmcb01.ptr->save.dbgctl = data;
3044 
3045 		svm_update_lbrv(vcpu);
3046 
3047 		break;
3048 	case MSR_VM_HSAVE_PA:
3049 		/*
3050 		 * Old kernels did not validate the value written to
3051 		 * MSR_VM_HSAVE_PA.  Allow KVM_SET_MSR to set an invalid
3052 		 * value to allow live migrating buggy or malicious guests
3053 		 * originating from those kernels.
3054 		 */
3055 		if (!msr->host_initiated && !page_address_valid(vcpu, data))
3056 			return 1;
3057 
3058 		svm->nested.hsave_msr = data & PAGE_MASK;
3059 		break;
3060 	case MSR_VM_CR:
3061 		return svm_set_vm_cr(vcpu, data);
3062 	case MSR_VM_IGNNE:
3063 		kvm_pr_unimpl_wrmsr(vcpu, ecx, data);
3064 		break;
3065 	case MSR_AMD64_DE_CFG: {
3066 		struct kvm_msr_entry msr_entry;
3067 
3068 		msr_entry.index = msr->index;
3069 		if (svm_get_msr_feature(&msr_entry))
3070 			return 1;
3071 
3072 		/* Check the supported bits */
3073 		if (data & ~msr_entry.data)
3074 			return 1;
3075 
3076 		/* Don't allow the guest to change a bit, #GP */
3077 		if (!msr->host_initiated && (data ^ msr_entry.data))
3078 			return 1;
3079 
3080 		svm->msr_decfg = data;
3081 		break;
3082 	}
3083 	default:
3084 		return kvm_set_msr_common(vcpu, msr);
3085 	}
3086 	return ret;
3087 }
3088 
3089 static int msr_interception(struct kvm_vcpu *vcpu)
3090 {
3091 	if (to_svm(vcpu)->vmcb->control.exit_info_1)
3092 		return kvm_emulate_wrmsr(vcpu);
3093 	else
3094 		return kvm_emulate_rdmsr(vcpu);
3095 }
3096 
3097 static int interrupt_window_interception(struct kvm_vcpu *vcpu)
3098 {
3099 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3100 	svm_clear_vintr(to_svm(vcpu));
3101 
3102 	/*
3103 	 * If not running nested, for AVIC, the only reason to end up here is ExtINTs.
3104 	 * In this case AVIC was temporarily disabled for
3105 	 * requesting the IRQ window and we have to re-enable it.
3106 	 *
3107 	 * If running nested, still remove the VM wide AVIC inhibit to
3108 	 * support case in which the interrupt window was requested when the
3109 	 * vCPU was not running nested.
3110 
3111 	 * All vCPUs which run still run nested, will remain to have their
3112 	 * AVIC still inhibited due to per-cpu AVIC inhibition.
3113 	 */
3114 	kvm_clear_apicv_inhibit(vcpu->kvm, APICV_INHIBIT_REASON_IRQWIN);
3115 
3116 	++vcpu->stat.irq_window_exits;
3117 	return 1;
3118 }
3119 
3120 static int pause_interception(struct kvm_vcpu *vcpu)
3121 {
3122 	bool in_kernel;
3123 	/*
3124 	 * CPL is not made available for an SEV-ES guest, therefore
3125 	 * vcpu->arch.preempted_in_kernel can never be true.  Just
3126 	 * set in_kernel to false as well.
3127 	 */
3128 	in_kernel = !sev_es_guest(vcpu->kvm) && svm_get_cpl(vcpu) == 0;
3129 
3130 	grow_ple_window(vcpu);
3131 
3132 	kvm_vcpu_on_spin(vcpu, in_kernel);
3133 	return kvm_skip_emulated_instruction(vcpu);
3134 }
3135 
3136 static int invpcid_interception(struct kvm_vcpu *vcpu)
3137 {
3138 	struct vcpu_svm *svm = to_svm(vcpu);
3139 	unsigned long type;
3140 	gva_t gva;
3141 
3142 	if (!guest_cpuid_has(vcpu, X86_FEATURE_INVPCID)) {
3143 		kvm_queue_exception(vcpu, UD_VECTOR);
3144 		return 1;
3145 	}
3146 
3147 	/*
3148 	 * For an INVPCID intercept:
3149 	 * EXITINFO1 provides the linear address of the memory operand.
3150 	 * EXITINFO2 provides the contents of the register operand.
3151 	 */
3152 	type = svm->vmcb->control.exit_info_2;
3153 	gva = svm->vmcb->control.exit_info_1;
3154 
3155 	return kvm_handle_invpcid(vcpu, type, gva);
3156 }
3157 
3158 static int (*const svm_exit_handlers[])(struct kvm_vcpu *vcpu) = {
3159 	[SVM_EXIT_READ_CR0]			= cr_interception,
3160 	[SVM_EXIT_READ_CR3]			= cr_interception,
3161 	[SVM_EXIT_READ_CR4]			= cr_interception,
3162 	[SVM_EXIT_READ_CR8]			= cr_interception,
3163 	[SVM_EXIT_CR0_SEL_WRITE]		= cr_interception,
3164 	[SVM_EXIT_WRITE_CR0]			= cr_interception,
3165 	[SVM_EXIT_WRITE_CR3]			= cr_interception,
3166 	[SVM_EXIT_WRITE_CR4]			= cr_interception,
3167 	[SVM_EXIT_WRITE_CR8]			= cr8_write_interception,
3168 	[SVM_EXIT_READ_DR0]			= dr_interception,
3169 	[SVM_EXIT_READ_DR1]			= dr_interception,
3170 	[SVM_EXIT_READ_DR2]			= dr_interception,
3171 	[SVM_EXIT_READ_DR3]			= dr_interception,
3172 	[SVM_EXIT_READ_DR4]			= dr_interception,
3173 	[SVM_EXIT_READ_DR5]			= dr_interception,
3174 	[SVM_EXIT_READ_DR6]			= dr_interception,
3175 	[SVM_EXIT_READ_DR7]			= dr_interception,
3176 	[SVM_EXIT_WRITE_DR0]			= dr_interception,
3177 	[SVM_EXIT_WRITE_DR1]			= dr_interception,
3178 	[SVM_EXIT_WRITE_DR2]			= dr_interception,
3179 	[SVM_EXIT_WRITE_DR3]			= dr_interception,
3180 	[SVM_EXIT_WRITE_DR4]			= dr_interception,
3181 	[SVM_EXIT_WRITE_DR5]			= dr_interception,
3182 	[SVM_EXIT_WRITE_DR6]			= dr_interception,
3183 	[SVM_EXIT_WRITE_DR7]			= dr_interception,
3184 	[SVM_EXIT_EXCP_BASE + DB_VECTOR]	= db_interception,
3185 	[SVM_EXIT_EXCP_BASE + BP_VECTOR]	= bp_interception,
3186 	[SVM_EXIT_EXCP_BASE + UD_VECTOR]	= ud_interception,
3187 	[SVM_EXIT_EXCP_BASE + PF_VECTOR]	= pf_interception,
3188 	[SVM_EXIT_EXCP_BASE + MC_VECTOR]	= mc_interception,
3189 	[SVM_EXIT_EXCP_BASE + AC_VECTOR]	= ac_interception,
3190 	[SVM_EXIT_EXCP_BASE + GP_VECTOR]	= gp_interception,
3191 	[SVM_EXIT_INTR]				= intr_interception,
3192 	[SVM_EXIT_NMI]				= nmi_interception,
3193 	[SVM_EXIT_SMI]				= smi_interception,
3194 	[SVM_EXIT_VINTR]			= interrupt_window_interception,
3195 	[SVM_EXIT_RDPMC]			= kvm_emulate_rdpmc,
3196 	[SVM_EXIT_CPUID]			= kvm_emulate_cpuid,
3197 	[SVM_EXIT_IRET]                         = iret_interception,
3198 	[SVM_EXIT_INVD]                         = kvm_emulate_invd,
3199 	[SVM_EXIT_PAUSE]			= pause_interception,
3200 	[SVM_EXIT_HLT]				= kvm_emulate_halt,
3201 	[SVM_EXIT_INVLPG]			= invlpg_interception,
3202 	[SVM_EXIT_INVLPGA]			= invlpga_interception,
3203 	[SVM_EXIT_IOIO]				= io_interception,
3204 	[SVM_EXIT_MSR]				= msr_interception,
3205 	[SVM_EXIT_TASK_SWITCH]			= task_switch_interception,
3206 	[SVM_EXIT_SHUTDOWN]			= shutdown_interception,
3207 	[SVM_EXIT_VMRUN]			= vmrun_interception,
3208 	[SVM_EXIT_VMMCALL]			= kvm_emulate_hypercall,
3209 	[SVM_EXIT_VMLOAD]			= vmload_interception,
3210 	[SVM_EXIT_VMSAVE]			= vmsave_interception,
3211 	[SVM_EXIT_STGI]				= stgi_interception,
3212 	[SVM_EXIT_CLGI]				= clgi_interception,
3213 	[SVM_EXIT_SKINIT]			= skinit_interception,
3214 	[SVM_EXIT_RDTSCP]			= kvm_handle_invalid_op,
3215 	[SVM_EXIT_WBINVD]                       = kvm_emulate_wbinvd,
3216 	[SVM_EXIT_MONITOR]			= kvm_emulate_monitor,
3217 	[SVM_EXIT_MWAIT]			= kvm_emulate_mwait,
3218 	[SVM_EXIT_XSETBV]			= kvm_emulate_xsetbv,
3219 	[SVM_EXIT_RDPRU]			= kvm_handle_invalid_op,
3220 	[SVM_EXIT_EFER_WRITE_TRAP]		= efer_trap,
3221 	[SVM_EXIT_CR0_WRITE_TRAP]		= cr_trap,
3222 	[SVM_EXIT_CR4_WRITE_TRAP]		= cr_trap,
3223 	[SVM_EXIT_CR8_WRITE_TRAP]		= cr_trap,
3224 	[SVM_EXIT_INVPCID]                      = invpcid_interception,
3225 	[SVM_EXIT_NPF]				= npf_interception,
3226 	[SVM_EXIT_RSM]                          = rsm_interception,
3227 	[SVM_EXIT_AVIC_INCOMPLETE_IPI]		= avic_incomplete_ipi_interception,
3228 	[SVM_EXIT_AVIC_UNACCELERATED_ACCESS]	= avic_unaccelerated_access_interception,
3229 	[SVM_EXIT_VMGEXIT]			= sev_handle_vmgexit,
3230 };
3231 
3232 static void dump_vmcb(struct kvm_vcpu *vcpu)
3233 {
3234 	struct vcpu_svm *svm = to_svm(vcpu);
3235 	struct vmcb_control_area *control = &svm->vmcb->control;
3236 	struct vmcb_save_area *save = &svm->vmcb->save;
3237 	struct vmcb_save_area *save01 = &svm->vmcb01.ptr->save;
3238 
3239 	if (!dump_invalid_vmcb) {
3240 		pr_warn_ratelimited("set kvm_amd.dump_invalid_vmcb=1 to dump internal KVM state.\n");
3241 		return;
3242 	}
3243 
3244 	pr_err("VMCB %p, last attempted VMRUN on CPU %d\n",
3245 	       svm->current_vmcb->ptr, vcpu->arch.last_vmentry_cpu);
3246 	pr_err("VMCB Control Area:\n");
3247 	pr_err("%-20s%04x\n", "cr_read:", control->intercepts[INTERCEPT_CR] & 0xffff);
3248 	pr_err("%-20s%04x\n", "cr_write:", control->intercepts[INTERCEPT_CR] >> 16);
3249 	pr_err("%-20s%04x\n", "dr_read:", control->intercepts[INTERCEPT_DR] & 0xffff);
3250 	pr_err("%-20s%04x\n", "dr_write:", control->intercepts[INTERCEPT_DR] >> 16);
3251 	pr_err("%-20s%08x\n", "exceptions:", control->intercepts[INTERCEPT_EXCEPTION]);
3252 	pr_err("%-20s%08x %08x\n", "intercepts:",
3253               control->intercepts[INTERCEPT_WORD3],
3254 	       control->intercepts[INTERCEPT_WORD4]);
3255 	pr_err("%-20s%d\n", "pause filter count:", control->pause_filter_count);
3256 	pr_err("%-20s%d\n", "pause filter threshold:",
3257 	       control->pause_filter_thresh);
3258 	pr_err("%-20s%016llx\n", "iopm_base_pa:", control->iopm_base_pa);
3259 	pr_err("%-20s%016llx\n", "msrpm_base_pa:", control->msrpm_base_pa);
3260 	pr_err("%-20s%016llx\n", "tsc_offset:", control->tsc_offset);
3261 	pr_err("%-20s%d\n", "asid:", control->asid);
3262 	pr_err("%-20s%d\n", "tlb_ctl:", control->tlb_ctl);
3263 	pr_err("%-20s%08x\n", "int_ctl:", control->int_ctl);
3264 	pr_err("%-20s%08x\n", "int_vector:", control->int_vector);
3265 	pr_err("%-20s%08x\n", "int_state:", control->int_state);
3266 	pr_err("%-20s%08x\n", "exit_code:", control->exit_code);
3267 	pr_err("%-20s%016llx\n", "exit_info1:", control->exit_info_1);
3268 	pr_err("%-20s%016llx\n", "exit_info2:", control->exit_info_2);
3269 	pr_err("%-20s%08x\n", "exit_int_info:", control->exit_int_info);
3270 	pr_err("%-20s%08x\n", "exit_int_info_err:", control->exit_int_info_err);
3271 	pr_err("%-20s%lld\n", "nested_ctl:", control->nested_ctl);
3272 	pr_err("%-20s%016llx\n", "nested_cr3:", control->nested_cr3);
3273 	pr_err("%-20s%016llx\n", "avic_vapic_bar:", control->avic_vapic_bar);
3274 	pr_err("%-20s%016llx\n", "ghcb:", control->ghcb_gpa);
3275 	pr_err("%-20s%08x\n", "event_inj:", control->event_inj);
3276 	pr_err("%-20s%08x\n", "event_inj_err:", control->event_inj_err);
3277 	pr_err("%-20s%lld\n", "virt_ext:", control->virt_ext);
3278 	pr_err("%-20s%016llx\n", "next_rip:", control->next_rip);
3279 	pr_err("%-20s%016llx\n", "avic_backing_page:", control->avic_backing_page);
3280 	pr_err("%-20s%016llx\n", "avic_logical_id:", control->avic_logical_id);
3281 	pr_err("%-20s%016llx\n", "avic_physical_id:", control->avic_physical_id);
3282 	pr_err("%-20s%016llx\n", "vmsa_pa:", control->vmsa_pa);
3283 	pr_err("VMCB State Save Area:\n");
3284 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3285 	       "es:",
3286 	       save->es.selector, save->es.attrib,
3287 	       save->es.limit, save->es.base);
3288 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3289 	       "cs:",
3290 	       save->cs.selector, save->cs.attrib,
3291 	       save->cs.limit, save->cs.base);
3292 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3293 	       "ss:",
3294 	       save->ss.selector, save->ss.attrib,
3295 	       save->ss.limit, save->ss.base);
3296 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3297 	       "ds:",
3298 	       save->ds.selector, save->ds.attrib,
3299 	       save->ds.limit, save->ds.base);
3300 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3301 	       "fs:",
3302 	       save01->fs.selector, save01->fs.attrib,
3303 	       save01->fs.limit, save01->fs.base);
3304 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3305 	       "gs:",
3306 	       save01->gs.selector, save01->gs.attrib,
3307 	       save01->gs.limit, save01->gs.base);
3308 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3309 	       "gdtr:",
3310 	       save->gdtr.selector, save->gdtr.attrib,
3311 	       save->gdtr.limit, save->gdtr.base);
3312 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3313 	       "ldtr:",
3314 	       save01->ldtr.selector, save01->ldtr.attrib,
3315 	       save01->ldtr.limit, save01->ldtr.base);
3316 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3317 	       "idtr:",
3318 	       save->idtr.selector, save->idtr.attrib,
3319 	       save->idtr.limit, save->idtr.base);
3320 	pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3321 	       "tr:",
3322 	       save01->tr.selector, save01->tr.attrib,
3323 	       save01->tr.limit, save01->tr.base);
3324 	pr_err("vmpl: %d   cpl:  %d               efer:          %016llx\n",
3325 	       save->vmpl, save->cpl, save->efer);
3326 	pr_err("%-15s %016llx %-13s %016llx\n",
3327 	       "cr0:", save->cr0, "cr2:", save->cr2);
3328 	pr_err("%-15s %016llx %-13s %016llx\n",
3329 	       "cr3:", save->cr3, "cr4:", save->cr4);
3330 	pr_err("%-15s %016llx %-13s %016llx\n",
3331 	       "dr6:", save->dr6, "dr7:", save->dr7);
3332 	pr_err("%-15s %016llx %-13s %016llx\n",
3333 	       "rip:", save->rip, "rflags:", save->rflags);
3334 	pr_err("%-15s %016llx %-13s %016llx\n",
3335 	       "rsp:", save->rsp, "rax:", save->rax);
3336 	pr_err("%-15s %016llx %-13s %016llx\n",
3337 	       "star:", save01->star, "lstar:", save01->lstar);
3338 	pr_err("%-15s %016llx %-13s %016llx\n",
3339 	       "cstar:", save01->cstar, "sfmask:", save01->sfmask);
3340 	pr_err("%-15s %016llx %-13s %016llx\n",
3341 	       "kernel_gs_base:", save01->kernel_gs_base,
3342 	       "sysenter_cs:", save01->sysenter_cs);
3343 	pr_err("%-15s %016llx %-13s %016llx\n",
3344 	       "sysenter_esp:", save01->sysenter_esp,
3345 	       "sysenter_eip:", save01->sysenter_eip);
3346 	pr_err("%-15s %016llx %-13s %016llx\n",
3347 	       "gpat:", save->g_pat, "dbgctl:", save->dbgctl);
3348 	pr_err("%-15s %016llx %-13s %016llx\n",
3349 	       "br_from:", save->br_from, "br_to:", save->br_to);
3350 	pr_err("%-15s %016llx %-13s %016llx\n",
3351 	       "excp_from:", save->last_excp_from,
3352 	       "excp_to:", save->last_excp_to);
3353 }
3354 
3355 static bool svm_check_exit_valid(u64 exit_code)
3356 {
3357 	return (exit_code < ARRAY_SIZE(svm_exit_handlers) &&
3358 		svm_exit_handlers[exit_code]);
3359 }
3360 
3361 static int svm_handle_invalid_exit(struct kvm_vcpu *vcpu, u64 exit_code)
3362 {
3363 	vcpu_unimpl(vcpu, "svm: unexpected exit reason 0x%llx\n", exit_code);
3364 	dump_vmcb(vcpu);
3365 	vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3366 	vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON;
3367 	vcpu->run->internal.ndata = 2;
3368 	vcpu->run->internal.data[0] = exit_code;
3369 	vcpu->run->internal.data[1] = vcpu->arch.last_vmentry_cpu;
3370 	return 0;
3371 }
3372 
3373 int svm_invoke_exit_handler(struct kvm_vcpu *vcpu, u64 exit_code)
3374 {
3375 	if (!svm_check_exit_valid(exit_code))
3376 		return svm_handle_invalid_exit(vcpu, exit_code);
3377 
3378 #ifdef CONFIG_RETPOLINE
3379 	if (exit_code == SVM_EXIT_MSR)
3380 		return msr_interception(vcpu);
3381 	else if (exit_code == SVM_EXIT_VINTR)
3382 		return interrupt_window_interception(vcpu);
3383 	else if (exit_code == SVM_EXIT_INTR)
3384 		return intr_interception(vcpu);
3385 	else if (exit_code == SVM_EXIT_HLT)
3386 		return kvm_emulate_halt(vcpu);
3387 	else if (exit_code == SVM_EXIT_NPF)
3388 		return npf_interception(vcpu);
3389 #endif
3390 	return svm_exit_handlers[exit_code](vcpu);
3391 }
3392 
3393 static void svm_get_exit_info(struct kvm_vcpu *vcpu, u32 *reason,
3394 			      u64 *info1, u64 *info2,
3395 			      u32 *intr_info, u32 *error_code)
3396 {
3397 	struct vmcb_control_area *control = &to_svm(vcpu)->vmcb->control;
3398 
3399 	*reason = control->exit_code;
3400 	*info1 = control->exit_info_1;
3401 	*info2 = control->exit_info_2;
3402 	*intr_info = control->exit_int_info;
3403 	if ((*intr_info & SVM_EXITINTINFO_VALID) &&
3404 	    (*intr_info & SVM_EXITINTINFO_VALID_ERR))
3405 		*error_code = control->exit_int_info_err;
3406 	else
3407 		*error_code = 0;
3408 }
3409 
3410 static int svm_handle_exit(struct kvm_vcpu *vcpu, fastpath_t exit_fastpath)
3411 {
3412 	struct vcpu_svm *svm = to_svm(vcpu);
3413 	struct kvm_run *kvm_run = vcpu->run;
3414 	u32 exit_code = svm->vmcb->control.exit_code;
3415 
3416 	/* SEV-ES guests must use the CR write traps to track CR registers. */
3417 	if (!sev_es_guest(vcpu->kvm)) {
3418 		if (!svm_is_intercept(svm, INTERCEPT_CR0_WRITE))
3419 			vcpu->arch.cr0 = svm->vmcb->save.cr0;
3420 		if (npt_enabled)
3421 			vcpu->arch.cr3 = svm->vmcb->save.cr3;
3422 	}
3423 
3424 	if (is_guest_mode(vcpu)) {
3425 		int vmexit;
3426 
3427 		trace_kvm_nested_vmexit(vcpu, KVM_ISA_SVM);
3428 
3429 		vmexit = nested_svm_exit_special(svm);
3430 
3431 		if (vmexit == NESTED_EXIT_CONTINUE)
3432 			vmexit = nested_svm_exit_handled(svm);
3433 
3434 		if (vmexit == NESTED_EXIT_DONE)
3435 			return 1;
3436 	}
3437 
3438 	if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
3439 		kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3440 		kvm_run->fail_entry.hardware_entry_failure_reason
3441 			= svm->vmcb->control.exit_code;
3442 		kvm_run->fail_entry.cpu = vcpu->arch.last_vmentry_cpu;
3443 		dump_vmcb(vcpu);
3444 		return 0;
3445 	}
3446 
3447 	if (exit_fastpath != EXIT_FASTPATH_NONE)
3448 		return 1;
3449 
3450 	return svm_invoke_exit_handler(vcpu, exit_code);
3451 }
3452 
3453 static void pre_svm_run(struct kvm_vcpu *vcpu)
3454 {
3455 	struct svm_cpu_data *sd = per_cpu_ptr(&svm_data, vcpu->cpu);
3456 	struct vcpu_svm *svm = to_svm(vcpu);
3457 
3458 	/*
3459 	 * If the previous vmrun of the vmcb occurred on a different physical
3460 	 * cpu, then mark the vmcb dirty and assign a new asid.  Hardware's
3461 	 * vmcb clean bits are per logical CPU, as are KVM's asid assignments.
3462 	 */
3463 	if (unlikely(svm->current_vmcb->cpu != vcpu->cpu)) {
3464 		svm->current_vmcb->asid_generation = 0;
3465 		vmcb_mark_all_dirty(svm->vmcb);
3466 		svm->current_vmcb->cpu = vcpu->cpu;
3467         }
3468 
3469 	if (sev_guest(vcpu->kvm))
3470 		return pre_sev_run(svm, vcpu->cpu);
3471 
3472 	/* FIXME: handle wraparound of asid_generation */
3473 	if (svm->current_vmcb->asid_generation != sd->asid_generation)
3474 		new_asid(svm, sd);
3475 }
3476 
3477 static void svm_inject_nmi(struct kvm_vcpu *vcpu)
3478 {
3479 	struct vcpu_svm *svm = to_svm(vcpu);
3480 
3481 	svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
3482 
3483 	if (svm->nmi_l1_to_l2)
3484 		return;
3485 
3486 	svm->nmi_masked = true;
3487 	svm_set_iret_intercept(svm);
3488 	++vcpu->stat.nmi_injections;
3489 }
3490 
3491 static bool svm_is_vnmi_pending(struct kvm_vcpu *vcpu)
3492 {
3493 	struct vcpu_svm *svm = to_svm(vcpu);
3494 
3495 	if (!is_vnmi_enabled(svm))
3496 		return false;
3497 
3498 	return !!(svm->vmcb->control.int_ctl & V_NMI_PENDING_MASK);
3499 }
3500 
3501 static bool svm_set_vnmi_pending(struct kvm_vcpu *vcpu)
3502 {
3503 	struct vcpu_svm *svm = to_svm(vcpu);
3504 
3505 	if (!is_vnmi_enabled(svm))
3506 		return false;
3507 
3508 	if (svm->vmcb->control.int_ctl & V_NMI_PENDING_MASK)
3509 		return false;
3510 
3511 	svm->vmcb->control.int_ctl |= V_NMI_PENDING_MASK;
3512 	vmcb_mark_dirty(svm->vmcb, VMCB_INTR);
3513 
3514 	/*
3515 	 * Because the pending NMI is serviced by hardware, KVM can't know when
3516 	 * the NMI is "injected", but for all intents and purposes, passing the
3517 	 * NMI off to hardware counts as injection.
3518 	 */
3519 	++vcpu->stat.nmi_injections;
3520 
3521 	return true;
3522 }
3523 
3524 static void svm_inject_irq(struct kvm_vcpu *vcpu, bool reinjected)
3525 {
3526 	struct vcpu_svm *svm = to_svm(vcpu);
3527 	u32 type;
3528 
3529 	if (vcpu->arch.interrupt.soft) {
3530 		if (svm_update_soft_interrupt_rip(vcpu))
3531 			return;
3532 
3533 		type = SVM_EVTINJ_TYPE_SOFT;
3534 	} else {
3535 		type = SVM_EVTINJ_TYPE_INTR;
3536 	}
3537 
3538 	trace_kvm_inj_virq(vcpu->arch.interrupt.nr,
3539 			   vcpu->arch.interrupt.soft, reinjected);
3540 	++vcpu->stat.irq_injections;
3541 
3542 	svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr |
3543 				       SVM_EVTINJ_VALID | type;
3544 }
3545 
3546 void svm_complete_interrupt_delivery(struct kvm_vcpu *vcpu, int delivery_mode,
3547 				     int trig_mode, int vector)
3548 {
3549 	/*
3550 	 * apic->apicv_active must be read after vcpu->mode.
3551 	 * Pairs with smp_store_release in vcpu_enter_guest.
3552 	 */
3553 	bool in_guest_mode = (smp_load_acquire(&vcpu->mode) == IN_GUEST_MODE);
3554 
3555 	/* Note, this is called iff the local APIC is in-kernel. */
3556 	if (!READ_ONCE(vcpu->arch.apic->apicv_active)) {
3557 		/* Process the interrupt via kvm_check_and_inject_events(). */
3558 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3559 		kvm_vcpu_kick(vcpu);
3560 		return;
3561 	}
3562 
3563 	trace_kvm_apicv_accept_irq(vcpu->vcpu_id, delivery_mode, trig_mode, vector);
3564 	if (in_guest_mode) {
3565 		/*
3566 		 * Signal the doorbell to tell hardware to inject the IRQ.  If
3567 		 * the vCPU exits the guest before the doorbell chimes, hardware
3568 		 * will automatically process AVIC interrupts at the next VMRUN.
3569 		 */
3570 		avic_ring_doorbell(vcpu);
3571 	} else {
3572 		/*
3573 		 * Wake the vCPU if it was blocking.  KVM will then detect the
3574 		 * pending IRQ when checking if the vCPU has a wake event.
3575 		 */
3576 		kvm_vcpu_wake_up(vcpu);
3577 	}
3578 }
3579 
3580 static void svm_deliver_interrupt(struct kvm_lapic *apic,  int delivery_mode,
3581 				  int trig_mode, int vector)
3582 {
3583 	kvm_lapic_set_irr(vector, apic);
3584 
3585 	/*
3586 	 * Pairs with the smp_mb_*() after setting vcpu->guest_mode in
3587 	 * vcpu_enter_guest() to ensure the write to the vIRR is ordered before
3588 	 * the read of guest_mode.  This guarantees that either VMRUN will see
3589 	 * and process the new vIRR entry, or that svm_complete_interrupt_delivery
3590 	 * will signal the doorbell if the CPU has already entered the guest.
3591 	 */
3592 	smp_mb__after_atomic();
3593 	svm_complete_interrupt_delivery(apic->vcpu, delivery_mode, trig_mode, vector);
3594 }
3595 
3596 static void svm_update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
3597 {
3598 	struct vcpu_svm *svm = to_svm(vcpu);
3599 
3600 	/*
3601 	 * SEV-ES guests must always keep the CR intercepts cleared. CR
3602 	 * tracking is done using the CR write traps.
3603 	 */
3604 	if (sev_es_guest(vcpu->kvm))
3605 		return;
3606 
3607 	if (nested_svm_virtualize_tpr(vcpu))
3608 		return;
3609 
3610 	svm_clr_intercept(svm, INTERCEPT_CR8_WRITE);
3611 
3612 	if (irr == -1)
3613 		return;
3614 
3615 	if (tpr >= irr)
3616 		svm_set_intercept(svm, INTERCEPT_CR8_WRITE);
3617 }
3618 
3619 static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu)
3620 {
3621 	struct vcpu_svm *svm = to_svm(vcpu);
3622 
3623 	if (is_vnmi_enabled(svm))
3624 		return svm->vmcb->control.int_ctl & V_NMI_BLOCKING_MASK;
3625 	else
3626 		return svm->nmi_masked;
3627 }
3628 
3629 static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
3630 {
3631 	struct vcpu_svm *svm = to_svm(vcpu);
3632 
3633 	if (is_vnmi_enabled(svm)) {
3634 		if (masked)
3635 			svm->vmcb->control.int_ctl |= V_NMI_BLOCKING_MASK;
3636 		else
3637 			svm->vmcb->control.int_ctl &= ~V_NMI_BLOCKING_MASK;
3638 
3639 	} else {
3640 		svm->nmi_masked = masked;
3641 		if (masked)
3642 			svm_set_iret_intercept(svm);
3643 		else
3644 			svm_clr_iret_intercept(svm);
3645 	}
3646 }
3647 
3648 bool svm_nmi_blocked(struct kvm_vcpu *vcpu)
3649 {
3650 	struct vcpu_svm *svm = to_svm(vcpu);
3651 	struct vmcb *vmcb = svm->vmcb;
3652 
3653 	if (!gif_set(svm))
3654 		return true;
3655 
3656 	if (is_guest_mode(vcpu) && nested_exit_on_nmi(svm))
3657 		return false;
3658 
3659 	if (svm_get_nmi_mask(vcpu))
3660 		return true;
3661 
3662 	return vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK;
3663 }
3664 
3665 static int svm_nmi_allowed(struct kvm_vcpu *vcpu, bool for_injection)
3666 {
3667 	struct vcpu_svm *svm = to_svm(vcpu);
3668 	if (svm->nested.nested_run_pending)
3669 		return -EBUSY;
3670 
3671 	if (svm_nmi_blocked(vcpu))
3672 		return 0;
3673 
3674 	/* An NMI must not be injected into L2 if it's supposed to VM-Exit.  */
3675 	if (for_injection && is_guest_mode(vcpu) && nested_exit_on_nmi(svm))
3676 		return -EBUSY;
3677 	return 1;
3678 }
3679 
3680 bool svm_interrupt_blocked(struct kvm_vcpu *vcpu)
3681 {
3682 	struct vcpu_svm *svm = to_svm(vcpu);
3683 	struct vmcb *vmcb = svm->vmcb;
3684 
3685 	if (!gif_set(svm))
3686 		return true;
3687 
3688 	if (is_guest_mode(vcpu)) {
3689 		/* As long as interrupts are being delivered...  */
3690 		if ((svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK)
3691 		    ? !(svm->vmcb01.ptr->save.rflags & X86_EFLAGS_IF)
3692 		    : !(kvm_get_rflags(vcpu) & X86_EFLAGS_IF))
3693 			return true;
3694 
3695 		/* ... vmexits aren't blocked by the interrupt shadow  */
3696 		if (nested_exit_on_intr(svm))
3697 			return false;
3698 	} else {
3699 		if (!svm_get_if_flag(vcpu))
3700 			return true;
3701 	}
3702 
3703 	return (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK);
3704 }
3705 
3706 static int svm_interrupt_allowed(struct kvm_vcpu *vcpu, bool for_injection)
3707 {
3708 	struct vcpu_svm *svm = to_svm(vcpu);
3709 
3710 	if (svm->nested.nested_run_pending)
3711 		return -EBUSY;
3712 
3713 	if (svm_interrupt_blocked(vcpu))
3714 		return 0;
3715 
3716 	/*
3717 	 * An IRQ must not be injected into L2 if it's supposed to VM-Exit,
3718 	 * e.g. if the IRQ arrived asynchronously after checking nested events.
3719 	 */
3720 	if (for_injection && is_guest_mode(vcpu) && nested_exit_on_intr(svm))
3721 		return -EBUSY;
3722 
3723 	return 1;
3724 }
3725 
3726 static void svm_enable_irq_window(struct kvm_vcpu *vcpu)
3727 {
3728 	struct vcpu_svm *svm = to_svm(vcpu);
3729 
3730 	/*
3731 	 * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
3732 	 * 1, because that's a separate STGI/VMRUN intercept.  The next time we
3733 	 * get that intercept, this function will be called again though and
3734 	 * we'll get the vintr intercept. However, if the vGIF feature is
3735 	 * enabled, the STGI interception will not occur. Enable the irq
3736 	 * window under the assumption that the hardware will set the GIF.
3737 	 */
3738 	if (vgif || gif_set(svm)) {
3739 		/*
3740 		 * IRQ window is not needed when AVIC is enabled,
3741 		 * unless we have pending ExtINT since it cannot be injected
3742 		 * via AVIC. In such case, KVM needs to temporarily disable AVIC,
3743 		 * and fallback to injecting IRQ via V_IRQ.
3744 		 *
3745 		 * If running nested, AVIC is already locally inhibited
3746 		 * on this vCPU, therefore there is no need to request
3747 		 * the VM wide AVIC inhibition.
3748 		 */
3749 		if (!is_guest_mode(vcpu))
3750 			kvm_set_apicv_inhibit(vcpu->kvm, APICV_INHIBIT_REASON_IRQWIN);
3751 
3752 		svm_set_vintr(svm);
3753 	}
3754 }
3755 
3756 static void svm_enable_nmi_window(struct kvm_vcpu *vcpu)
3757 {
3758 	struct vcpu_svm *svm = to_svm(vcpu);
3759 
3760 	/*
3761 	 * KVM should never request an NMI window when vNMI is enabled, as KVM
3762 	 * allows at most one to-be-injected NMI and one pending NMI, i.e. if
3763 	 * two NMIs arrive simultaneously, KVM will inject one and set
3764 	 * V_NMI_PENDING for the other.  WARN, but continue with the standard
3765 	 * single-step approach to try and salvage the pending NMI.
3766 	 */
3767 	WARN_ON_ONCE(is_vnmi_enabled(svm));
3768 
3769 	if (svm_get_nmi_mask(vcpu) && !svm->awaiting_iret_completion)
3770 		return; /* IRET will cause a vm exit */
3771 
3772 	if (!gif_set(svm)) {
3773 		if (vgif)
3774 			svm_set_intercept(svm, INTERCEPT_STGI);
3775 		return; /* STGI will cause a vm exit */
3776 	}
3777 
3778 	/*
3779 	 * Something prevents NMI from been injected. Single step over possible
3780 	 * problem (IRET or exception injection or interrupt shadow)
3781 	 */
3782 	svm->nmi_singlestep_guest_rflags = svm_get_rflags(vcpu);
3783 	svm->nmi_singlestep = true;
3784 	svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
3785 }
3786 
3787 static void svm_flush_tlb_asid(struct kvm_vcpu *vcpu)
3788 {
3789 	struct vcpu_svm *svm = to_svm(vcpu);
3790 
3791 	/*
3792 	 * Unlike VMX, SVM doesn't provide a way to flush only NPT TLB entries.
3793 	 * A TLB flush for the current ASID flushes both "host" and "guest" TLB
3794 	 * entries, and thus is a superset of Hyper-V's fine grained flushing.
3795 	 */
3796 	kvm_hv_vcpu_purge_flush_tlb(vcpu);
3797 
3798 	/*
3799 	 * Flush only the current ASID even if the TLB flush was invoked via
3800 	 * kvm_flush_remote_tlbs().  Although flushing remote TLBs requires all
3801 	 * ASIDs to be flushed, KVM uses a single ASID for L1 and L2, and
3802 	 * unconditionally does a TLB flush on both nested VM-Enter and nested
3803 	 * VM-Exit (via kvm_mmu_reset_context()).
3804 	 */
3805 	if (static_cpu_has(X86_FEATURE_FLUSHBYASID))
3806 		svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
3807 	else
3808 		svm->current_vmcb->asid_generation--;
3809 }
3810 
3811 static void svm_flush_tlb_current(struct kvm_vcpu *vcpu)
3812 {
3813 	hpa_t root_tdp = vcpu->arch.mmu->root.hpa;
3814 
3815 	/*
3816 	 * When running on Hyper-V with EnlightenedNptTlb enabled, explicitly
3817 	 * flush the NPT mappings via hypercall as flushing the ASID only
3818 	 * affects virtual to physical mappings, it does not invalidate guest
3819 	 * physical to host physical mappings.
3820 	 */
3821 	if (svm_hv_is_enlightened_tlb_enabled(vcpu) && VALID_PAGE(root_tdp))
3822 		hyperv_flush_guest_mapping(root_tdp);
3823 
3824 	svm_flush_tlb_asid(vcpu);
3825 }
3826 
3827 static void svm_flush_tlb_all(struct kvm_vcpu *vcpu)
3828 {
3829 	/*
3830 	 * When running on Hyper-V with EnlightenedNptTlb enabled, remote TLB
3831 	 * flushes should be routed to hv_flush_remote_tlbs() without requesting
3832 	 * a "regular" remote flush.  Reaching this point means either there's
3833 	 * a KVM bug or a prior hv_flush_remote_tlbs() call failed, both of
3834 	 * which might be fatal to the guest.  Yell, but try to recover.
3835 	 */
3836 	if (WARN_ON_ONCE(svm_hv_is_enlightened_tlb_enabled(vcpu)))
3837 		hv_flush_remote_tlbs(vcpu->kvm);
3838 
3839 	svm_flush_tlb_asid(vcpu);
3840 }
3841 
3842 static void svm_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t gva)
3843 {
3844 	struct vcpu_svm *svm = to_svm(vcpu);
3845 
3846 	invlpga(gva, svm->vmcb->control.asid);
3847 }
3848 
3849 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
3850 {
3851 	struct vcpu_svm *svm = to_svm(vcpu);
3852 
3853 	if (nested_svm_virtualize_tpr(vcpu))
3854 		return;
3855 
3856 	if (!svm_is_intercept(svm, INTERCEPT_CR8_WRITE)) {
3857 		int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
3858 		kvm_set_cr8(vcpu, cr8);
3859 	}
3860 }
3861 
3862 static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
3863 {
3864 	struct vcpu_svm *svm = to_svm(vcpu);
3865 	u64 cr8;
3866 
3867 	if (nested_svm_virtualize_tpr(vcpu) ||
3868 	    kvm_vcpu_apicv_active(vcpu))
3869 		return;
3870 
3871 	cr8 = kvm_get_cr8(vcpu);
3872 	svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
3873 	svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
3874 }
3875 
3876 static void svm_complete_soft_interrupt(struct kvm_vcpu *vcpu, u8 vector,
3877 					int type)
3878 {
3879 	bool is_exception = (type == SVM_EXITINTINFO_TYPE_EXEPT);
3880 	bool is_soft = (type == SVM_EXITINTINFO_TYPE_SOFT);
3881 	struct vcpu_svm *svm = to_svm(vcpu);
3882 
3883 	/*
3884 	 * If NRIPS is enabled, KVM must snapshot the pre-VMRUN next_rip that's
3885 	 * associated with the original soft exception/interrupt.  next_rip is
3886 	 * cleared on all exits that can occur while vectoring an event, so KVM
3887 	 * needs to manually set next_rip for re-injection.  Unlike the !nrips
3888 	 * case below, this needs to be done if and only if KVM is re-injecting
3889 	 * the same event, i.e. if the event is a soft exception/interrupt,
3890 	 * otherwise next_rip is unused on VMRUN.
3891 	 */
3892 	if (nrips && (is_soft || (is_exception && kvm_exception_is_soft(vector))) &&
3893 	    kvm_is_linear_rip(vcpu, svm->soft_int_old_rip + svm->soft_int_csbase))
3894 		svm->vmcb->control.next_rip = svm->soft_int_next_rip;
3895 	/*
3896 	 * If NRIPS isn't enabled, KVM must manually advance RIP prior to
3897 	 * injecting the soft exception/interrupt.  That advancement needs to
3898 	 * be unwound if vectoring didn't complete.  Note, the new event may
3899 	 * not be the injected event, e.g. if KVM injected an INTn, the INTn
3900 	 * hit a #NP in the guest, and the #NP encountered a #PF, the #NP will
3901 	 * be the reported vectored event, but RIP still needs to be unwound.
3902 	 */
3903 	else if (!nrips && (is_soft || is_exception) &&
3904 		 kvm_is_linear_rip(vcpu, svm->soft_int_next_rip + svm->soft_int_csbase))
3905 		kvm_rip_write(vcpu, svm->soft_int_old_rip);
3906 }
3907 
3908 static void svm_complete_interrupts(struct kvm_vcpu *vcpu)
3909 {
3910 	struct vcpu_svm *svm = to_svm(vcpu);
3911 	u8 vector;
3912 	int type;
3913 	u32 exitintinfo = svm->vmcb->control.exit_int_info;
3914 	bool nmi_l1_to_l2 = svm->nmi_l1_to_l2;
3915 	bool soft_int_injected = svm->soft_int_injected;
3916 
3917 	svm->nmi_l1_to_l2 = false;
3918 	svm->soft_int_injected = false;
3919 
3920 	/*
3921 	 * If we've made progress since setting HF_IRET_MASK, we've
3922 	 * executed an IRET and can allow NMI injection.
3923 	 */
3924 	if (svm->awaiting_iret_completion &&
3925 	    (sev_es_guest(vcpu->kvm) ||
3926 	     kvm_rip_read(vcpu) != svm->nmi_iret_rip)) {
3927 		svm->awaiting_iret_completion = false;
3928 		svm->nmi_masked = false;
3929 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3930 	}
3931 
3932 	vcpu->arch.nmi_injected = false;
3933 	kvm_clear_exception_queue(vcpu);
3934 	kvm_clear_interrupt_queue(vcpu);
3935 
3936 	if (!(exitintinfo & SVM_EXITINTINFO_VALID))
3937 		return;
3938 
3939 	kvm_make_request(KVM_REQ_EVENT, vcpu);
3940 
3941 	vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK;
3942 	type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK;
3943 
3944 	if (soft_int_injected)
3945 		svm_complete_soft_interrupt(vcpu, vector, type);
3946 
3947 	switch (type) {
3948 	case SVM_EXITINTINFO_TYPE_NMI:
3949 		vcpu->arch.nmi_injected = true;
3950 		svm->nmi_l1_to_l2 = nmi_l1_to_l2;
3951 		break;
3952 	case SVM_EXITINTINFO_TYPE_EXEPT:
3953 		/*
3954 		 * Never re-inject a #VC exception.
3955 		 */
3956 		if (vector == X86_TRAP_VC)
3957 			break;
3958 
3959 		if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) {
3960 			u32 err = svm->vmcb->control.exit_int_info_err;
3961 			kvm_requeue_exception_e(vcpu, vector, err);
3962 
3963 		} else
3964 			kvm_requeue_exception(vcpu, vector);
3965 		break;
3966 	case SVM_EXITINTINFO_TYPE_INTR:
3967 		kvm_queue_interrupt(vcpu, vector, false);
3968 		break;
3969 	case SVM_EXITINTINFO_TYPE_SOFT:
3970 		kvm_queue_interrupt(vcpu, vector, true);
3971 		break;
3972 	default:
3973 		break;
3974 	}
3975 
3976 }
3977 
3978 static void svm_cancel_injection(struct kvm_vcpu *vcpu)
3979 {
3980 	struct vcpu_svm *svm = to_svm(vcpu);
3981 	struct vmcb_control_area *control = &svm->vmcb->control;
3982 
3983 	control->exit_int_info = control->event_inj;
3984 	control->exit_int_info_err = control->event_inj_err;
3985 	control->event_inj = 0;
3986 	svm_complete_interrupts(vcpu);
3987 }
3988 
3989 static int svm_vcpu_pre_run(struct kvm_vcpu *vcpu)
3990 {
3991 	return 1;
3992 }
3993 
3994 static fastpath_t svm_exit_handlers_fastpath(struct kvm_vcpu *vcpu)
3995 {
3996 	if (to_svm(vcpu)->vmcb->control.exit_code == SVM_EXIT_MSR &&
3997 	    to_svm(vcpu)->vmcb->control.exit_info_1)
3998 		return handle_fastpath_set_msr_irqoff(vcpu);
3999 
4000 	return EXIT_FASTPATH_NONE;
4001 }
4002 
4003 static noinstr void svm_vcpu_enter_exit(struct kvm_vcpu *vcpu, bool spec_ctrl_intercepted)
4004 {
4005 	struct vcpu_svm *svm = to_svm(vcpu);
4006 
4007 	guest_state_enter_irqoff();
4008 
4009 	if (sev_es_guest(vcpu->kvm))
4010 		__svm_sev_es_vcpu_run(svm, spec_ctrl_intercepted);
4011 	else
4012 		__svm_vcpu_run(svm, spec_ctrl_intercepted);
4013 
4014 	guest_state_exit_irqoff();
4015 }
4016 
4017 static __no_kcsan fastpath_t svm_vcpu_run(struct kvm_vcpu *vcpu)
4018 {
4019 	struct vcpu_svm *svm = to_svm(vcpu);
4020 	bool spec_ctrl_intercepted = msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL);
4021 
4022 	trace_kvm_entry(vcpu);
4023 
4024 	svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
4025 	svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
4026 	svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
4027 
4028 	/*
4029 	 * Disable singlestep if we're injecting an interrupt/exception.
4030 	 * We don't want our modified rflags to be pushed on the stack where
4031 	 * we might not be able to easily reset them if we disabled NMI
4032 	 * singlestep later.
4033 	 */
4034 	if (svm->nmi_singlestep && svm->vmcb->control.event_inj) {
4035 		/*
4036 		 * Event injection happens before external interrupts cause a
4037 		 * vmexit and interrupts are disabled here, so smp_send_reschedule
4038 		 * is enough to force an immediate vmexit.
4039 		 */
4040 		disable_nmi_singlestep(svm);
4041 		smp_send_reschedule(vcpu->cpu);
4042 	}
4043 
4044 	pre_svm_run(vcpu);
4045 
4046 	sync_lapic_to_cr8(vcpu);
4047 
4048 	if (unlikely(svm->asid != svm->vmcb->control.asid)) {
4049 		svm->vmcb->control.asid = svm->asid;
4050 		vmcb_mark_dirty(svm->vmcb, VMCB_ASID);
4051 	}
4052 	svm->vmcb->save.cr2 = vcpu->arch.cr2;
4053 
4054 	svm_hv_update_vp_id(svm->vmcb, vcpu);
4055 
4056 	/*
4057 	 * Run with all-zero DR6 unless needed, so that we can get the exact cause
4058 	 * of a #DB.
4059 	 */
4060 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT))
4061 		svm_set_dr6(svm, vcpu->arch.dr6);
4062 	else
4063 		svm_set_dr6(svm, DR6_ACTIVE_LOW);
4064 
4065 	clgi();
4066 	kvm_load_guest_xsave_state(vcpu);
4067 
4068 	kvm_wait_lapic_expire(vcpu);
4069 
4070 	/*
4071 	 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
4072 	 * it's non-zero. Since vmentry is serialising on affected CPUs, there
4073 	 * is no need to worry about the conditional branch over the wrmsr
4074 	 * being speculatively taken.
4075 	 */
4076 	if (!static_cpu_has(X86_FEATURE_V_SPEC_CTRL))
4077 		x86_spec_ctrl_set_guest(svm->virt_spec_ctrl);
4078 
4079 	svm_vcpu_enter_exit(vcpu, spec_ctrl_intercepted);
4080 
4081 	if (!static_cpu_has(X86_FEATURE_V_SPEC_CTRL))
4082 		x86_spec_ctrl_restore_host(svm->virt_spec_ctrl);
4083 
4084 	if (!sev_es_guest(vcpu->kvm)) {
4085 		vcpu->arch.cr2 = svm->vmcb->save.cr2;
4086 		vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
4087 		vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
4088 		vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
4089 	}
4090 	vcpu->arch.regs_dirty = 0;
4091 
4092 	if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
4093 		kvm_before_interrupt(vcpu, KVM_HANDLING_NMI);
4094 
4095 	kvm_load_host_xsave_state(vcpu);
4096 	stgi();
4097 
4098 	/* Any pending NMI will happen here */
4099 
4100 	if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
4101 		kvm_after_interrupt(vcpu);
4102 
4103 	sync_cr8_to_lapic(vcpu);
4104 
4105 	svm->next_rip = 0;
4106 	if (is_guest_mode(vcpu)) {
4107 		nested_sync_control_from_vmcb02(svm);
4108 
4109 		/* Track VMRUNs that have made past consistency checking */
4110 		if (svm->nested.nested_run_pending &&
4111 		    svm->vmcb->control.exit_code != SVM_EXIT_ERR)
4112                         ++vcpu->stat.nested_run;
4113 
4114 		svm->nested.nested_run_pending = 0;
4115 	}
4116 
4117 	svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
4118 	vmcb_mark_all_clean(svm->vmcb);
4119 
4120 	/* if exit due to PF check for async PF */
4121 	if (svm->vmcb->control.exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR)
4122 		vcpu->arch.apf.host_apf_flags =
4123 			kvm_read_and_reset_apf_flags();
4124 
4125 	vcpu->arch.regs_avail &= ~SVM_REGS_LAZY_LOAD_SET;
4126 
4127 	/*
4128 	 * We need to handle MC intercepts here before the vcpu has a chance to
4129 	 * change the physical cpu
4130 	 */
4131 	if (unlikely(svm->vmcb->control.exit_code ==
4132 		     SVM_EXIT_EXCP_BASE + MC_VECTOR))
4133 		svm_handle_mce(vcpu);
4134 
4135 	trace_kvm_exit(vcpu, KVM_ISA_SVM);
4136 
4137 	svm_complete_interrupts(vcpu);
4138 
4139 	if (is_guest_mode(vcpu))
4140 		return EXIT_FASTPATH_NONE;
4141 
4142 	return svm_exit_handlers_fastpath(vcpu);
4143 }
4144 
4145 static void svm_load_mmu_pgd(struct kvm_vcpu *vcpu, hpa_t root_hpa,
4146 			     int root_level)
4147 {
4148 	struct vcpu_svm *svm = to_svm(vcpu);
4149 	unsigned long cr3;
4150 
4151 	if (npt_enabled) {
4152 		svm->vmcb->control.nested_cr3 = __sme_set(root_hpa);
4153 		vmcb_mark_dirty(svm->vmcb, VMCB_NPT);
4154 
4155 		hv_track_root_tdp(vcpu, root_hpa);
4156 
4157 		cr3 = vcpu->arch.cr3;
4158 	} else if (root_level >= PT64_ROOT_4LEVEL) {
4159 		cr3 = __sme_set(root_hpa) | kvm_get_active_pcid(vcpu);
4160 	} else {
4161 		/* PCID in the guest should be impossible with a 32-bit MMU. */
4162 		WARN_ON_ONCE(kvm_get_active_pcid(vcpu));
4163 		cr3 = root_hpa;
4164 	}
4165 
4166 	svm->vmcb->save.cr3 = cr3;
4167 	vmcb_mark_dirty(svm->vmcb, VMCB_CR);
4168 }
4169 
4170 static void
4171 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4172 {
4173 	/*
4174 	 * Patch in the VMMCALL instruction:
4175 	 */
4176 	hypercall[0] = 0x0f;
4177 	hypercall[1] = 0x01;
4178 	hypercall[2] = 0xd9;
4179 }
4180 
4181 /*
4182  * The kvm parameter can be NULL (module initialization, or invocation before
4183  * VM creation). Be sure to check the kvm parameter before using it.
4184  */
4185 static bool svm_has_emulated_msr(struct kvm *kvm, u32 index)
4186 {
4187 	switch (index) {
4188 	case MSR_IA32_MCG_EXT_CTL:
4189 	case KVM_FIRST_EMULATED_VMX_MSR ... KVM_LAST_EMULATED_VMX_MSR:
4190 		return false;
4191 	case MSR_IA32_SMBASE:
4192 		if (!IS_ENABLED(CONFIG_KVM_SMM))
4193 			return false;
4194 		/* SEV-ES guests do not support SMM, so report false */
4195 		if (kvm && sev_es_guest(kvm))
4196 			return false;
4197 		break;
4198 	default:
4199 		break;
4200 	}
4201 
4202 	return true;
4203 }
4204 
4205 static void svm_vcpu_after_set_cpuid(struct kvm_vcpu *vcpu)
4206 {
4207 	struct vcpu_svm *svm = to_svm(vcpu);
4208 	struct kvm_cpuid_entry2 *best;
4209 
4210 	vcpu->arch.xsaves_enabled = guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
4211 				    boot_cpu_has(X86_FEATURE_XSAVE) &&
4212 				    boot_cpu_has(X86_FEATURE_XSAVES);
4213 
4214 	/* Update nrips enabled cache */
4215 	svm->nrips_enabled = kvm_cpu_cap_has(X86_FEATURE_NRIPS) &&
4216 			     guest_cpuid_has(vcpu, X86_FEATURE_NRIPS);
4217 
4218 	svm->tsc_scaling_enabled = tsc_scaling && guest_cpuid_has(vcpu, X86_FEATURE_TSCRATEMSR);
4219 	svm->lbrv_enabled = lbrv && guest_cpuid_has(vcpu, X86_FEATURE_LBRV);
4220 
4221 	svm->v_vmload_vmsave_enabled = vls && guest_cpuid_has(vcpu, X86_FEATURE_V_VMSAVE_VMLOAD);
4222 
4223 	svm->pause_filter_enabled = kvm_cpu_cap_has(X86_FEATURE_PAUSEFILTER) &&
4224 			guest_cpuid_has(vcpu, X86_FEATURE_PAUSEFILTER);
4225 
4226 	svm->pause_threshold_enabled = kvm_cpu_cap_has(X86_FEATURE_PFTHRESHOLD) &&
4227 			guest_cpuid_has(vcpu, X86_FEATURE_PFTHRESHOLD);
4228 
4229 	svm->vgif_enabled = vgif && guest_cpuid_has(vcpu, X86_FEATURE_VGIF);
4230 
4231 	svm->vnmi_enabled = vnmi && guest_cpuid_has(vcpu, X86_FEATURE_VNMI);
4232 
4233 	svm_recalc_instruction_intercepts(vcpu, svm);
4234 
4235 	if (boot_cpu_has(X86_FEATURE_IBPB))
4236 		set_msr_interception(vcpu, svm->msrpm, MSR_IA32_PRED_CMD, 0,
4237 				     !!guest_has_pred_cmd_msr(vcpu));
4238 
4239 	if (boot_cpu_has(X86_FEATURE_FLUSH_L1D))
4240 		set_msr_interception(vcpu, svm->msrpm, MSR_IA32_FLUSH_CMD, 0,
4241 				     !!guest_cpuid_has(vcpu, X86_FEATURE_FLUSH_L1D));
4242 
4243 	/* For sev guests, the memory encryption bit is not reserved in CR3.  */
4244 	if (sev_guest(vcpu->kvm)) {
4245 		best = kvm_find_cpuid_entry(vcpu, 0x8000001F);
4246 		if (best)
4247 			vcpu->arch.reserved_gpa_bits &= ~(1UL << (best->ebx & 0x3f));
4248 	}
4249 
4250 	init_vmcb_after_set_cpuid(vcpu);
4251 }
4252 
4253 static bool svm_has_wbinvd_exit(void)
4254 {
4255 	return true;
4256 }
4257 
4258 #define PRE_EX(exit)  { .exit_code = (exit), \
4259 			.stage = X86_ICPT_PRE_EXCEPT, }
4260 #define POST_EX(exit) { .exit_code = (exit), \
4261 			.stage = X86_ICPT_POST_EXCEPT, }
4262 #define POST_MEM(exit) { .exit_code = (exit), \
4263 			.stage = X86_ICPT_POST_MEMACCESS, }
4264 
4265 static const struct __x86_intercept {
4266 	u32 exit_code;
4267 	enum x86_intercept_stage stage;
4268 } x86_intercept_map[] = {
4269 	[x86_intercept_cr_read]		= POST_EX(SVM_EXIT_READ_CR0),
4270 	[x86_intercept_cr_write]	= POST_EX(SVM_EXIT_WRITE_CR0),
4271 	[x86_intercept_clts]		= POST_EX(SVM_EXIT_WRITE_CR0),
4272 	[x86_intercept_lmsw]		= POST_EX(SVM_EXIT_WRITE_CR0),
4273 	[x86_intercept_smsw]		= POST_EX(SVM_EXIT_READ_CR0),
4274 	[x86_intercept_dr_read]		= POST_EX(SVM_EXIT_READ_DR0),
4275 	[x86_intercept_dr_write]	= POST_EX(SVM_EXIT_WRITE_DR0),
4276 	[x86_intercept_sldt]		= POST_EX(SVM_EXIT_LDTR_READ),
4277 	[x86_intercept_str]		= POST_EX(SVM_EXIT_TR_READ),
4278 	[x86_intercept_lldt]		= POST_EX(SVM_EXIT_LDTR_WRITE),
4279 	[x86_intercept_ltr]		= POST_EX(SVM_EXIT_TR_WRITE),
4280 	[x86_intercept_sgdt]		= POST_EX(SVM_EXIT_GDTR_READ),
4281 	[x86_intercept_sidt]		= POST_EX(SVM_EXIT_IDTR_READ),
4282 	[x86_intercept_lgdt]		= POST_EX(SVM_EXIT_GDTR_WRITE),
4283 	[x86_intercept_lidt]		= POST_EX(SVM_EXIT_IDTR_WRITE),
4284 	[x86_intercept_vmrun]		= POST_EX(SVM_EXIT_VMRUN),
4285 	[x86_intercept_vmmcall]		= POST_EX(SVM_EXIT_VMMCALL),
4286 	[x86_intercept_vmload]		= POST_EX(SVM_EXIT_VMLOAD),
4287 	[x86_intercept_vmsave]		= POST_EX(SVM_EXIT_VMSAVE),
4288 	[x86_intercept_stgi]		= POST_EX(SVM_EXIT_STGI),
4289 	[x86_intercept_clgi]		= POST_EX(SVM_EXIT_CLGI),
4290 	[x86_intercept_skinit]		= POST_EX(SVM_EXIT_SKINIT),
4291 	[x86_intercept_invlpga]		= POST_EX(SVM_EXIT_INVLPGA),
4292 	[x86_intercept_rdtscp]		= POST_EX(SVM_EXIT_RDTSCP),
4293 	[x86_intercept_monitor]		= POST_MEM(SVM_EXIT_MONITOR),
4294 	[x86_intercept_mwait]		= POST_EX(SVM_EXIT_MWAIT),
4295 	[x86_intercept_invlpg]		= POST_EX(SVM_EXIT_INVLPG),
4296 	[x86_intercept_invd]		= POST_EX(SVM_EXIT_INVD),
4297 	[x86_intercept_wbinvd]		= POST_EX(SVM_EXIT_WBINVD),
4298 	[x86_intercept_wrmsr]		= POST_EX(SVM_EXIT_MSR),
4299 	[x86_intercept_rdtsc]		= POST_EX(SVM_EXIT_RDTSC),
4300 	[x86_intercept_rdmsr]		= POST_EX(SVM_EXIT_MSR),
4301 	[x86_intercept_rdpmc]		= POST_EX(SVM_EXIT_RDPMC),
4302 	[x86_intercept_cpuid]		= PRE_EX(SVM_EXIT_CPUID),
4303 	[x86_intercept_rsm]		= PRE_EX(SVM_EXIT_RSM),
4304 	[x86_intercept_pause]		= PRE_EX(SVM_EXIT_PAUSE),
4305 	[x86_intercept_pushf]		= PRE_EX(SVM_EXIT_PUSHF),
4306 	[x86_intercept_popf]		= PRE_EX(SVM_EXIT_POPF),
4307 	[x86_intercept_intn]		= PRE_EX(SVM_EXIT_SWINT),
4308 	[x86_intercept_iret]		= PRE_EX(SVM_EXIT_IRET),
4309 	[x86_intercept_icebp]		= PRE_EX(SVM_EXIT_ICEBP),
4310 	[x86_intercept_hlt]		= POST_EX(SVM_EXIT_HLT),
4311 	[x86_intercept_in]		= POST_EX(SVM_EXIT_IOIO),
4312 	[x86_intercept_ins]		= POST_EX(SVM_EXIT_IOIO),
4313 	[x86_intercept_out]		= POST_EX(SVM_EXIT_IOIO),
4314 	[x86_intercept_outs]		= POST_EX(SVM_EXIT_IOIO),
4315 	[x86_intercept_xsetbv]		= PRE_EX(SVM_EXIT_XSETBV),
4316 };
4317 
4318 #undef PRE_EX
4319 #undef POST_EX
4320 #undef POST_MEM
4321 
4322 static int svm_check_intercept(struct kvm_vcpu *vcpu,
4323 			       struct x86_instruction_info *info,
4324 			       enum x86_intercept_stage stage,
4325 			       struct x86_exception *exception)
4326 {
4327 	struct vcpu_svm *svm = to_svm(vcpu);
4328 	int vmexit, ret = X86EMUL_CONTINUE;
4329 	struct __x86_intercept icpt_info;
4330 	struct vmcb *vmcb = svm->vmcb;
4331 
4332 	if (info->intercept >= ARRAY_SIZE(x86_intercept_map))
4333 		goto out;
4334 
4335 	icpt_info = x86_intercept_map[info->intercept];
4336 
4337 	if (stage != icpt_info.stage)
4338 		goto out;
4339 
4340 	switch (icpt_info.exit_code) {
4341 	case SVM_EXIT_READ_CR0:
4342 		if (info->intercept == x86_intercept_cr_read)
4343 			icpt_info.exit_code += info->modrm_reg;
4344 		break;
4345 	case SVM_EXIT_WRITE_CR0: {
4346 		unsigned long cr0, val;
4347 
4348 		if (info->intercept == x86_intercept_cr_write)
4349 			icpt_info.exit_code += info->modrm_reg;
4350 
4351 		if (icpt_info.exit_code != SVM_EXIT_WRITE_CR0 ||
4352 		    info->intercept == x86_intercept_clts)
4353 			break;
4354 
4355 		if (!(vmcb12_is_intercept(&svm->nested.ctl,
4356 					INTERCEPT_SELECTIVE_CR0)))
4357 			break;
4358 
4359 		cr0 = vcpu->arch.cr0 & ~SVM_CR0_SELECTIVE_MASK;
4360 		val = info->src_val  & ~SVM_CR0_SELECTIVE_MASK;
4361 
4362 		if (info->intercept == x86_intercept_lmsw) {
4363 			cr0 &= 0xfUL;
4364 			val &= 0xfUL;
4365 			/* lmsw can't clear PE - catch this here */
4366 			if (cr0 & X86_CR0_PE)
4367 				val |= X86_CR0_PE;
4368 		}
4369 
4370 		if (cr0 ^ val)
4371 			icpt_info.exit_code = SVM_EXIT_CR0_SEL_WRITE;
4372 
4373 		break;
4374 	}
4375 	case SVM_EXIT_READ_DR0:
4376 	case SVM_EXIT_WRITE_DR0:
4377 		icpt_info.exit_code += info->modrm_reg;
4378 		break;
4379 	case SVM_EXIT_MSR:
4380 		if (info->intercept == x86_intercept_wrmsr)
4381 			vmcb->control.exit_info_1 = 1;
4382 		else
4383 			vmcb->control.exit_info_1 = 0;
4384 		break;
4385 	case SVM_EXIT_PAUSE:
4386 		/*
4387 		 * We get this for NOP only, but pause
4388 		 * is rep not, check this here
4389 		 */
4390 		if (info->rep_prefix != REPE_PREFIX)
4391 			goto out;
4392 		break;
4393 	case SVM_EXIT_IOIO: {
4394 		u64 exit_info;
4395 		u32 bytes;
4396 
4397 		if (info->intercept == x86_intercept_in ||
4398 		    info->intercept == x86_intercept_ins) {
4399 			exit_info = ((info->src_val & 0xffff) << 16) |
4400 				SVM_IOIO_TYPE_MASK;
4401 			bytes = info->dst_bytes;
4402 		} else {
4403 			exit_info = (info->dst_val & 0xffff) << 16;
4404 			bytes = info->src_bytes;
4405 		}
4406 
4407 		if (info->intercept == x86_intercept_outs ||
4408 		    info->intercept == x86_intercept_ins)
4409 			exit_info |= SVM_IOIO_STR_MASK;
4410 
4411 		if (info->rep_prefix)
4412 			exit_info |= SVM_IOIO_REP_MASK;
4413 
4414 		bytes = min(bytes, 4u);
4415 
4416 		exit_info |= bytes << SVM_IOIO_SIZE_SHIFT;
4417 
4418 		exit_info |= (u32)info->ad_bytes << (SVM_IOIO_ASIZE_SHIFT - 1);
4419 
4420 		vmcb->control.exit_info_1 = exit_info;
4421 		vmcb->control.exit_info_2 = info->next_rip;
4422 
4423 		break;
4424 	}
4425 	default:
4426 		break;
4427 	}
4428 
4429 	/* TODO: Advertise NRIPS to guest hypervisor unconditionally */
4430 	if (static_cpu_has(X86_FEATURE_NRIPS))
4431 		vmcb->control.next_rip  = info->next_rip;
4432 	vmcb->control.exit_code = icpt_info.exit_code;
4433 	vmexit = nested_svm_exit_handled(svm);
4434 
4435 	ret = (vmexit == NESTED_EXIT_DONE) ? X86EMUL_INTERCEPTED
4436 					   : X86EMUL_CONTINUE;
4437 
4438 out:
4439 	return ret;
4440 }
4441 
4442 static void svm_handle_exit_irqoff(struct kvm_vcpu *vcpu)
4443 {
4444 	if (to_svm(vcpu)->vmcb->control.exit_code == SVM_EXIT_INTR)
4445 		vcpu->arch.at_instruction_boundary = true;
4446 }
4447 
4448 static void svm_sched_in(struct kvm_vcpu *vcpu, int cpu)
4449 {
4450 	if (!kvm_pause_in_guest(vcpu->kvm))
4451 		shrink_ple_window(vcpu);
4452 }
4453 
4454 static void svm_setup_mce(struct kvm_vcpu *vcpu)
4455 {
4456 	/* [63:9] are reserved. */
4457 	vcpu->arch.mcg_cap &= 0x1ff;
4458 }
4459 
4460 #ifdef CONFIG_KVM_SMM
4461 bool svm_smi_blocked(struct kvm_vcpu *vcpu)
4462 {
4463 	struct vcpu_svm *svm = to_svm(vcpu);
4464 
4465 	/* Per APM Vol.2 15.22.2 "Response to SMI" */
4466 	if (!gif_set(svm))
4467 		return true;
4468 
4469 	return is_smm(vcpu);
4470 }
4471 
4472 static int svm_smi_allowed(struct kvm_vcpu *vcpu, bool for_injection)
4473 {
4474 	struct vcpu_svm *svm = to_svm(vcpu);
4475 	if (svm->nested.nested_run_pending)
4476 		return -EBUSY;
4477 
4478 	if (svm_smi_blocked(vcpu))
4479 		return 0;
4480 
4481 	/* An SMI must not be injected into L2 if it's supposed to VM-Exit.  */
4482 	if (for_injection && is_guest_mode(vcpu) && nested_exit_on_smi(svm))
4483 		return -EBUSY;
4484 
4485 	return 1;
4486 }
4487 
4488 static int svm_enter_smm(struct kvm_vcpu *vcpu, union kvm_smram *smram)
4489 {
4490 	struct vcpu_svm *svm = to_svm(vcpu);
4491 	struct kvm_host_map map_save;
4492 	int ret;
4493 
4494 	if (!is_guest_mode(vcpu))
4495 		return 0;
4496 
4497 	/*
4498 	 * 32-bit SMRAM format doesn't preserve EFER and SVM state.  Userspace is
4499 	 * responsible for ensuring nested SVM and SMIs are mutually exclusive.
4500 	 */
4501 
4502 	if (!guest_cpuid_has(vcpu, X86_FEATURE_LM))
4503 		return 1;
4504 
4505 	smram->smram64.svm_guest_flag = 1;
4506 	smram->smram64.svm_guest_vmcb_gpa = svm->nested.vmcb12_gpa;
4507 
4508 	svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
4509 	svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
4510 	svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
4511 
4512 	ret = nested_svm_simple_vmexit(svm, SVM_EXIT_SW);
4513 	if (ret)
4514 		return ret;
4515 
4516 	/*
4517 	 * KVM uses VMCB01 to store L1 host state while L2 runs but
4518 	 * VMCB01 is going to be used during SMM and thus the state will
4519 	 * be lost. Temporary save non-VMLOAD/VMSAVE state to the host save
4520 	 * area pointed to by MSR_VM_HSAVE_PA. APM guarantees that the
4521 	 * format of the area is identical to guest save area offsetted
4522 	 * by 0x400 (matches the offset of 'struct vmcb_save_area'
4523 	 * within 'struct vmcb'). Note: HSAVE area may also be used by
4524 	 * L1 hypervisor to save additional host context (e.g. KVM does
4525 	 * that, see svm_prepare_switch_to_guest()) which must be
4526 	 * preserved.
4527 	 */
4528 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(svm->nested.hsave_msr), &map_save))
4529 		return 1;
4530 
4531 	BUILD_BUG_ON(offsetof(struct vmcb, save) != 0x400);
4532 
4533 	svm_copy_vmrun_state(map_save.hva + 0x400,
4534 			     &svm->vmcb01.ptr->save);
4535 
4536 	kvm_vcpu_unmap(vcpu, &map_save, true);
4537 	return 0;
4538 }
4539 
4540 static int svm_leave_smm(struct kvm_vcpu *vcpu, const union kvm_smram *smram)
4541 {
4542 	struct vcpu_svm *svm = to_svm(vcpu);
4543 	struct kvm_host_map map, map_save;
4544 	struct vmcb *vmcb12;
4545 	int ret;
4546 
4547 	const struct kvm_smram_state_64 *smram64 = &smram->smram64;
4548 
4549 	if (!guest_cpuid_has(vcpu, X86_FEATURE_LM))
4550 		return 0;
4551 
4552 	/* Non-zero if SMI arrived while vCPU was in guest mode. */
4553 	if (!smram64->svm_guest_flag)
4554 		return 0;
4555 
4556 	if (!guest_cpuid_has(vcpu, X86_FEATURE_SVM))
4557 		return 1;
4558 
4559 	if (!(smram64->efer & EFER_SVME))
4560 		return 1;
4561 
4562 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(smram64->svm_guest_vmcb_gpa), &map))
4563 		return 1;
4564 
4565 	ret = 1;
4566 	if (kvm_vcpu_map(vcpu, gpa_to_gfn(svm->nested.hsave_msr), &map_save))
4567 		goto unmap_map;
4568 
4569 	if (svm_allocate_nested(svm))
4570 		goto unmap_save;
4571 
4572 	/*
4573 	 * Restore L1 host state from L1 HSAVE area as VMCB01 was
4574 	 * used during SMM (see svm_enter_smm())
4575 	 */
4576 
4577 	svm_copy_vmrun_state(&svm->vmcb01.ptr->save, map_save.hva + 0x400);
4578 
4579 	/*
4580 	 * Enter the nested guest now
4581 	 */
4582 
4583 	vmcb_mark_all_dirty(svm->vmcb01.ptr);
4584 
4585 	vmcb12 = map.hva;
4586 	nested_copy_vmcb_control_to_cache(svm, &vmcb12->control);
4587 	nested_copy_vmcb_save_to_cache(svm, &vmcb12->save);
4588 	ret = enter_svm_guest_mode(vcpu, smram64->svm_guest_vmcb_gpa, vmcb12, false);
4589 
4590 	if (ret)
4591 		goto unmap_save;
4592 
4593 	svm->nested.nested_run_pending = 1;
4594 
4595 unmap_save:
4596 	kvm_vcpu_unmap(vcpu, &map_save, true);
4597 unmap_map:
4598 	kvm_vcpu_unmap(vcpu, &map, true);
4599 	return ret;
4600 }
4601 
4602 static void svm_enable_smi_window(struct kvm_vcpu *vcpu)
4603 {
4604 	struct vcpu_svm *svm = to_svm(vcpu);
4605 
4606 	if (!gif_set(svm)) {
4607 		if (vgif)
4608 			svm_set_intercept(svm, INTERCEPT_STGI);
4609 		/* STGI will cause a vm exit */
4610 	} else {
4611 		/* We must be in SMM; RSM will cause a vmexit anyway.  */
4612 	}
4613 }
4614 #endif
4615 
4616 static bool svm_can_emulate_instruction(struct kvm_vcpu *vcpu, int emul_type,
4617 					void *insn, int insn_len)
4618 {
4619 	bool smep, smap, is_user;
4620 	u64 error_code;
4621 
4622 	/* Emulation is always possible when KVM has access to all guest state. */
4623 	if (!sev_guest(vcpu->kvm))
4624 		return true;
4625 
4626 	/* #UD and #GP should never be intercepted for SEV guests. */
4627 	WARN_ON_ONCE(emul_type & (EMULTYPE_TRAP_UD |
4628 				  EMULTYPE_TRAP_UD_FORCED |
4629 				  EMULTYPE_VMWARE_GP));
4630 
4631 	/*
4632 	 * Emulation is impossible for SEV-ES guests as KVM doesn't have access
4633 	 * to guest register state.
4634 	 */
4635 	if (sev_es_guest(vcpu->kvm))
4636 		return false;
4637 
4638 	/*
4639 	 * Emulation is possible if the instruction is already decoded, e.g.
4640 	 * when completing I/O after returning from userspace.
4641 	 */
4642 	if (emul_type & EMULTYPE_NO_DECODE)
4643 		return true;
4644 
4645 	/*
4646 	 * Emulation is possible for SEV guests if and only if a prefilled
4647 	 * buffer containing the bytes of the intercepted instruction is
4648 	 * available. SEV guest memory is encrypted with a guest specific key
4649 	 * and cannot be decrypted by KVM, i.e. KVM would read cyphertext and
4650 	 * decode garbage.
4651 	 *
4652 	 * Inject #UD if KVM reached this point without an instruction buffer.
4653 	 * In practice, this path should never be hit by a well-behaved guest,
4654 	 * e.g. KVM doesn't intercept #UD or #GP for SEV guests, but this path
4655 	 * is still theoretically reachable, e.g. via unaccelerated fault-like
4656 	 * AVIC access, and needs to be handled by KVM to avoid putting the
4657 	 * guest into an infinite loop.   Injecting #UD is somewhat arbitrary,
4658 	 * but its the least awful option given lack of insight into the guest.
4659 	 */
4660 	if (unlikely(!insn)) {
4661 		kvm_queue_exception(vcpu, UD_VECTOR);
4662 		return false;
4663 	}
4664 
4665 	/*
4666 	 * Emulate for SEV guests if the insn buffer is not empty.  The buffer
4667 	 * will be empty if the DecodeAssist microcode cannot fetch bytes for
4668 	 * the faulting instruction because the code fetch itself faulted, e.g.
4669 	 * the guest attempted to fetch from emulated MMIO or a guest page
4670 	 * table used to translate CS:RIP resides in emulated MMIO.
4671 	 */
4672 	if (likely(insn_len))
4673 		return true;
4674 
4675 	/*
4676 	 * Detect and workaround Errata 1096 Fam_17h_00_0Fh.
4677 	 *
4678 	 * Errata:
4679 	 * When CPU raises #NPF on guest data access and vCPU CR4.SMAP=1, it is
4680 	 * possible that CPU microcode implementing DecodeAssist will fail to
4681 	 * read guest memory at CS:RIP and vmcb.GuestIntrBytes will incorrectly
4682 	 * be '0'.  This happens because microcode reads CS:RIP using a _data_
4683 	 * loap uop with CPL=0 privileges.  If the load hits a SMAP #PF, ucode
4684 	 * gives up and does not fill the instruction bytes buffer.
4685 	 *
4686 	 * As above, KVM reaches this point iff the VM is an SEV guest, the CPU
4687 	 * supports DecodeAssist, a #NPF was raised, KVM's page fault handler
4688 	 * triggered emulation (e.g. for MMIO), and the CPU returned 0 in the
4689 	 * GuestIntrBytes field of the VMCB.
4690 	 *
4691 	 * This does _not_ mean that the erratum has been encountered, as the
4692 	 * DecodeAssist will also fail if the load for CS:RIP hits a legitimate
4693 	 * #PF, e.g. if the guest attempt to execute from emulated MMIO and
4694 	 * encountered a reserved/not-present #PF.
4695 	 *
4696 	 * To hit the erratum, the following conditions must be true:
4697 	 *    1. CR4.SMAP=1 (obviously).
4698 	 *    2. CR4.SMEP=0 || CPL=3.  If SMEP=1 and CPL<3, the erratum cannot
4699 	 *       have been hit as the guest would have encountered a SMEP
4700 	 *       violation #PF, not a #NPF.
4701 	 *    3. The #NPF is not due to a code fetch, in which case failure to
4702 	 *       retrieve the instruction bytes is legitimate (see abvoe).
4703 	 *
4704 	 * In addition, don't apply the erratum workaround if the #NPF occurred
4705 	 * while translating guest page tables (see below).
4706 	 */
4707 	error_code = to_svm(vcpu)->vmcb->control.exit_info_1;
4708 	if (error_code & (PFERR_GUEST_PAGE_MASK | PFERR_FETCH_MASK))
4709 		goto resume_guest;
4710 
4711 	smep = kvm_is_cr4_bit_set(vcpu, X86_CR4_SMEP);
4712 	smap = kvm_is_cr4_bit_set(vcpu, X86_CR4_SMAP);
4713 	is_user = svm_get_cpl(vcpu) == 3;
4714 	if (smap && (!smep || is_user)) {
4715 		pr_err_ratelimited("SEV Guest triggered AMD Erratum 1096\n");
4716 
4717 		/*
4718 		 * If the fault occurred in userspace, arbitrarily inject #GP
4719 		 * to avoid killing the guest and to hopefully avoid confusing
4720 		 * the guest kernel too much, e.g. injecting #PF would not be
4721 		 * coherent with respect to the guest's page tables.  Request
4722 		 * triple fault if the fault occurred in the kernel as there's
4723 		 * no fault that KVM can inject without confusing the guest.
4724 		 * In practice, the triple fault is moot as no sane SEV kernel
4725 		 * will execute from user memory while also running with SMAP=1.
4726 		 */
4727 		if (is_user)
4728 			kvm_inject_gp(vcpu, 0);
4729 		else
4730 			kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4731 	}
4732 
4733 resume_guest:
4734 	/*
4735 	 * If the erratum was not hit, simply resume the guest and let it fault
4736 	 * again.  While awful, e.g. the vCPU may get stuck in an infinite loop
4737 	 * if the fault is at CPL=0, it's the lesser of all evils.  Exiting to
4738 	 * userspace will kill the guest, and letting the emulator read garbage
4739 	 * will yield random behavior and potentially corrupt the guest.
4740 	 *
4741 	 * Simply resuming the guest is technically not a violation of the SEV
4742 	 * architecture.  AMD's APM states that all code fetches and page table
4743 	 * accesses for SEV guest are encrypted, regardless of the C-Bit.  The
4744 	 * APM also states that encrypted accesses to MMIO are "ignored", but
4745 	 * doesn't explicitly define "ignored", i.e. doing nothing and letting
4746 	 * the guest spin is technically "ignoring" the access.
4747 	 */
4748 	return false;
4749 }
4750 
4751 static bool svm_apic_init_signal_blocked(struct kvm_vcpu *vcpu)
4752 {
4753 	struct vcpu_svm *svm = to_svm(vcpu);
4754 
4755 	return !gif_set(svm);
4756 }
4757 
4758 static void svm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
4759 {
4760 	if (!sev_es_guest(vcpu->kvm))
4761 		return kvm_vcpu_deliver_sipi_vector(vcpu, vector);
4762 
4763 	sev_vcpu_deliver_sipi_vector(vcpu, vector);
4764 }
4765 
4766 static void svm_vm_destroy(struct kvm *kvm)
4767 {
4768 	avic_vm_destroy(kvm);
4769 	sev_vm_destroy(kvm);
4770 }
4771 
4772 static int svm_vm_init(struct kvm *kvm)
4773 {
4774 	if (!pause_filter_count || !pause_filter_thresh)
4775 		kvm->arch.pause_in_guest = true;
4776 
4777 	if (enable_apicv) {
4778 		int ret = avic_vm_init(kvm);
4779 		if (ret)
4780 			return ret;
4781 	}
4782 
4783 	return 0;
4784 }
4785 
4786 static struct kvm_x86_ops svm_x86_ops __initdata = {
4787 	.name = KBUILD_MODNAME,
4788 
4789 	.check_processor_compatibility = svm_check_processor_compat,
4790 
4791 	.hardware_unsetup = svm_hardware_unsetup,
4792 	.hardware_enable = svm_hardware_enable,
4793 	.hardware_disable = svm_hardware_disable,
4794 	.has_emulated_msr = svm_has_emulated_msr,
4795 
4796 	.vcpu_create = svm_vcpu_create,
4797 	.vcpu_free = svm_vcpu_free,
4798 	.vcpu_reset = svm_vcpu_reset,
4799 
4800 	.vm_size = sizeof(struct kvm_svm),
4801 	.vm_init = svm_vm_init,
4802 	.vm_destroy = svm_vm_destroy,
4803 
4804 	.prepare_switch_to_guest = svm_prepare_switch_to_guest,
4805 	.vcpu_load = svm_vcpu_load,
4806 	.vcpu_put = svm_vcpu_put,
4807 	.vcpu_blocking = avic_vcpu_blocking,
4808 	.vcpu_unblocking = avic_vcpu_unblocking,
4809 
4810 	.update_exception_bitmap = svm_update_exception_bitmap,
4811 	.get_msr_feature = svm_get_msr_feature,
4812 	.get_msr = svm_get_msr,
4813 	.set_msr = svm_set_msr,
4814 	.get_segment_base = svm_get_segment_base,
4815 	.get_segment = svm_get_segment,
4816 	.set_segment = svm_set_segment,
4817 	.get_cpl = svm_get_cpl,
4818 	.get_cs_db_l_bits = svm_get_cs_db_l_bits,
4819 	.is_valid_cr0 = svm_is_valid_cr0,
4820 	.set_cr0 = svm_set_cr0,
4821 	.post_set_cr3 = sev_post_set_cr3,
4822 	.is_valid_cr4 = svm_is_valid_cr4,
4823 	.set_cr4 = svm_set_cr4,
4824 	.set_efer = svm_set_efer,
4825 	.get_idt = svm_get_idt,
4826 	.set_idt = svm_set_idt,
4827 	.get_gdt = svm_get_gdt,
4828 	.set_gdt = svm_set_gdt,
4829 	.set_dr7 = svm_set_dr7,
4830 	.sync_dirty_debug_regs = svm_sync_dirty_debug_regs,
4831 	.cache_reg = svm_cache_reg,
4832 	.get_rflags = svm_get_rflags,
4833 	.set_rflags = svm_set_rflags,
4834 	.get_if_flag = svm_get_if_flag,
4835 
4836 	.flush_tlb_all = svm_flush_tlb_all,
4837 	.flush_tlb_current = svm_flush_tlb_current,
4838 	.flush_tlb_gva = svm_flush_tlb_gva,
4839 	.flush_tlb_guest = svm_flush_tlb_asid,
4840 
4841 	.vcpu_pre_run = svm_vcpu_pre_run,
4842 	.vcpu_run = svm_vcpu_run,
4843 	.handle_exit = svm_handle_exit,
4844 	.skip_emulated_instruction = svm_skip_emulated_instruction,
4845 	.update_emulated_instruction = NULL,
4846 	.set_interrupt_shadow = svm_set_interrupt_shadow,
4847 	.get_interrupt_shadow = svm_get_interrupt_shadow,
4848 	.patch_hypercall = svm_patch_hypercall,
4849 	.inject_irq = svm_inject_irq,
4850 	.inject_nmi = svm_inject_nmi,
4851 	.is_vnmi_pending = svm_is_vnmi_pending,
4852 	.set_vnmi_pending = svm_set_vnmi_pending,
4853 	.inject_exception = svm_inject_exception,
4854 	.cancel_injection = svm_cancel_injection,
4855 	.interrupt_allowed = svm_interrupt_allowed,
4856 	.nmi_allowed = svm_nmi_allowed,
4857 	.get_nmi_mask = svm_get_nmi_mask,
4858 	.set_nmi_mask = svm_set_nmi_mask,
4859 	.enable_nmi_window = svm_enable_nmi_window,
4860 	.enable_irq_window = svm_enable_irq_window,
4861 	.update_cr8_intercept = svm_update_cr8_intercept,
4862 	.set_virtual_apic_mode = avic_refresh_virtual_apic_mode,
4863 	.refresh_apicv_exec_ctrl = avic_refresh_apicv_exec_ctrl,
4864 	.apicv_post_state_restore = avic_apicv_post_state_restore,
4865 	.required_apicv_inhibits = AVIC_REQUIRED_APICV_INHIBITS,
4866 
4867 	.get_exit_info = svm_get_exit_info,
4868 
4869 	.vcpu_after_set_cpuid = svm_vcpu_after_set_cpuid,
4870 
4871 	.has_wbinvd_exit = svm_has_wbinvd_exit,
4872 
4873 	.get_l2_tsc_offset = svm_get_l2_tsc_offset,
4874 	.get_l2_tsc_multiplier = svm_get_l2_tsc_multiplier,
4875 	.write_tsc_offset = svm_write_tsc_offset,
4876 	.write_tsc_multiplier = svm_write_tsc_multiplier,
4877 
4878 	.load_mmu_pgd = svm_load_mmu_pgd,
4879 
4880 	.check_intercept = svm_check_intercept,
4881 	.handle_exit_irqoff = svm_handle_exit_irqoff,
4882 
4883 	.request_immediate_exit = __kvm_request_immediate_exit,
4884 
4885 	.sched_in = svm_sched_in,
4886 
4887 	.nested_ops = &svm_nested_ops,
4888 
4889 	.deliver_interrupt = svm_deliver_interrupt,
4890 	.pi_update_irte = avic_pi_update_irte,
4891 	.setup_mce = svm_setup_mce,
4892 
4893 #ifdef CONFIG_KVM_SMM
4894 	.smi_allowed = svm_smi_allowed,
4895 	.enter_smm = svm_enter_smm,
4896 	.leave_smm = svm_leave_smm,
4897 	.enable_smi_window = svm_enable_smi_window,
4898 #endif
4899 
4900 	.mem_enc_ioctl = sev_mem_enc_ioctl,
4901 	.mem_enc_register_region = sev_mem_enc_register_region,
4902 	.mem_enc_unregister_region = sev_mem_enc_unregister_region,
4903 	.guest_memory_reclaimed = sev_guest_memory_reclaimed,
4904 
4905 	.vm_copy_enc_context_from = sev_vm_copy_enc_context_from,
4906 	.vm_move_enc_context_from = sev_vm_move_enc_context_from,
4907 
4908 	.can_emulate_instruction = svm_can_emulate_instruction,
4909 
4910 	.apic_init_signal_blocked = svm_apic_init_signal_blocked,
4911 
4912 	.msr_filter_changed = svm_msr_filter_changed,
4913 	.complete_emulated_msr = svm_complete_emulated_msr,
4914 
4915 	.vcpu_deliver_sipi_vector = svm_vcpu_deliver_sipi_vector,
4916 	.vcpu_get_apicv_inhibit_reasons = avic_vcpu_get_apicv_inhibit_reasons,
4917 };
4918 
4919 /*
4920  * The default MMIO mask is a single bit (excluding the present bit),
4921  * which could conflict with the memory encryption bit. Check for
4922  * memory encryption support and override the default MMIO mask if
4923  * memory encryption is enabled.
4924  */
4925 static __init void svm_adjust_mmio_mask(void)
4926 {
4927 	unsigned int enc_bit, mask_bit;
4928 	u64 msr, mask;
4929 
4930 	/* If there is no memory encryption support, use existing mask */
4931 	if (cpuid_eax(0x80000000) < 0x8000001f)
4932 		return;
4933 
4934 	/* If memory encryption is not enabled, use existing mask */
4935 	rdmsrl(MSR_AMD64_SYSCFG, msr);
4936 	if (!(msr & MSR_AMD64_SYSCFG_MEM_ENCRYPT))
4937 		return;
4938 
4939 	enc_bit = cpuid_ebx(0x8000001f) & 0x3f;
4940 	mask_bit = boot_cpu_data.x86_phys_bits;
4941 
4942 	/* Increment the mask bit if it is the same as the encryption bit */
4943 	if (enc_bit == mask_bit)
4944 		mask_bit++;
4945 
4946 	/*
4947 	 * If the mask bit location is below 52, then some bits above the
4948 	 * physical addressing limit will always be reserved, so use the
4949 	 * rsvd_bits() function to generate the mask. This mask, along with
4950 	 * the present bit, will be used to generate a page fault with
4951 	 * PFER.RSV = 1.
4952 	 *
4953 	 * If the mask bit location is 52 (or above), then clear the mask.
4954 	 */
4955 	mask = (mask_bit < 52) ? rsvd_bits(mask_bit, 51) | PT_PRESENT_MASK : 0;
4956 
4957 	kvm_mmu_set_mmio_spte_mask(mask, mask, PT_WRITABLE_MASK | PT_USER_MASK);
4958 }
4959 
4960 static __init void svm_set_cpu_caps(void)
4961 {
4962 	kvm_set_cpu_caps();
4963 
4964 	kvm_caps.supported_perf_cap = 0;
4965 	kvm_caps.supported_xss = 0;
4966 
4967 	/* CPUID 0x80000001 and 0x8000000A (SVM features) */
4968 	if (nested) {
4969 		kvm_cpu_cap_set(X86_FEATURE_SVM);
4970 		kvm_cpu_cap_set(X86_FEATURE_VMCBCLEAN);
4971 
4972 		if (nrips)
4973 			kvm_cpu_cap_set(X86_FEATURE_NRIPS);
4974 
4975 		if (npt_enabled)
4976 			kvm_cpu_cap_set(X86_FEATURE_NPT);
4977 
4978 		if (tsc_scaling)
4979 			kvm_cpu_cap_set(X86_FEATURE_TSCRATEMSR);
4980 
4981 		if (vls)
4982 			kvm_cpu_cap_set(X86_FEATURE_V_VMSAVE_VMLOAD);
4983 		if (lbrv)
4984 			kvm_cpu_cap_set(X86_FEATURE_LBRV);
4985 
4986 		if (boot_cpu_has(X86_FEATURE_PAUSEFILTER))
4987 			kvm_cpu_cap_set(X86_FEATURE_PAUSEFILTER);
4988 
4989 		if (boot_cpu_has(X86_FEATURE_PFTHRESHOLD))
4990 			kvm_cpu_cap_set(X86_FEATURE_PFTHRESHOLD);
4991 
4992 		if (vgif)
4993 			kvm_cpu_cap_set(X86_FEATURE_VGIF);
4994 
4995 		if (vnmi)
4996 			kvm_cpu_cap_set(X86_FEATURE_VNMI);
4997 
4998 		/* Nested VM can receive #VMEXIT instead of triggering #GP */
4999 		kvm_cpu_cap_set(X86_FEATURE_SVME_ADDR_CHK);
5000 	}
5001 
5002 	/* CPUID 0x80000008 */
5003 	if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD) ||
5004 	    boot_cpu_has(X86_FEATURE_AMD_SSBD))
5005 		kvm_cpu_cap_set(X86_FEATURE_VIRT_SSBD);
5006 
5007 	if (enable_pmu) {
5008 		/*
5009 		 * Enumerate support for PERFCTR_CORE if and only if KVM has
5010 		 * access to enough counters to virtualize "core" support,
5011 		 * otherwise limit vPMU support to the legacy number of counters.
5012 		 */
5013 		if (kvm_pmu_cap.num_counters_gp < AMD64_NUM_COUNTERS_CORE)
5014 			kvm_pmu_cap.num_counters_gp = min(AMD64_NUM_COUNTERS,
5015 							  kvm_pmu_cap.num_counters_gp);
5016 		else
5017 			kvm_cpu_cap_check_and_set(X86_FEATURE_PERFCTR_CORE);
5018 
5019 		if (kvm_pmu_cap.version != 2 ||
5020 		    !kvm_cpu_cap_has(X86_FEATURE_PERFCTR_CORE))
5021 			kvm_cpu_cap_clear(X86_FEATURE_PERFMON_V2);
5022 	}
5023 
5024 	/* CPUID 0x8000001F (SME/SEV features) */
5025 	sev_set_cpu_caps();
5026 }
5027 
5028 static __init int svm_hardware_setup(void)
5029 {
5030 	int cpu;
5031 	struct page *iopm_pages;
5032 	void *iopm_va;
5033 	int r;
5034 	unsigned int order = get_order(IOPM_SIZE);
5035 
5036 	/*
5037 	 * NX is required for shadow paging and for NPT if the NX huge pages
5038 	 * mitigation is enabled.
5039 	 */
5040 	if (!boot_cpu_has(X86_FEATURE_NX)) {
5041 		pr_err_ratelimited("NX (Execute Disable) not supported\n");
5042 		return -EOPNOTSUPP;
5043 	}
5044 	kvm_enable_efer_bits(EFER_NX);
5045 
5046 	iopm_pages = alloc_pages(GFP_KERNEL, order);
5047 
5048 	if (!iopm_pages)
5049 		return -ENOMEM;
5050 
5051 	iopm_va = page_address(iopm_pages);
5052 	memset(iopm_va, 0xff, PAGE_SIZE * (1 << order));
5053 	iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
5054 
5055 	init_msrpm_offsets();
5056 
5057 	kvm_caps.supported_xcr0 &= ~(XFEATURE_MASK_BNDREGS |
5058 				     XFEATURE_MASK_BNDCSR);
5059 
5060 	if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
5061 		kvm_enable_efer_bits(EFER_FFXSR);
5062 
5063 	if (tsc_scaling) {
5064 		if (!boot_cpu_has(X86_FEATURE_TSCRATEMSR)) {
5065 			tsc_scaling = false;
5066 		} else {
5067 			pr_info("TSC scaling supported\n");
5068 			kvm_caps.has_tsc_control = true;
5069 		}
5070 	}
5071 	kvm_caps.max_tsc_scaling_ratio = SVM_TSC_RATIO_MAX;
5072 	kvm_caps.tsc_scaling_ratio_frac_bits = 32;
5073 
5074 	tsc_aux_uret_slot = kvm_add_user_return_msr(MSR_TSC_AUX);
5075 
5076 	if (boot_cpu_has(X86_FEATURE_AUTOIBRS))
5077 		kvm_enable_efer_bits(EFER_AUTOIBRS);
5078 
5079 	/* Check for pause filtering support */
5080 	if (!boot_cpu_has(X86_FEATURE_PAUSEFILTER)) {
5081 		pause_filter_count = 0;
5082 		pause_filter_thresh = 0;
5083 	} else if (!boot_cpu_has(X86_FEATURE_PFTHRESHOLD)) {
5084 		pause_filter_thresh = 0;
5085 	}
5086 
5087 	if (nested) {
5088 		pr_info("Nested Virtualization enabled\n");
5089 		kvm_enable_efer_bits(EFER_SVME | EFER_LMSLE);
5090 	}
5091 
5092 	/*
5093 	 * KVM's MMU doesn't support using 2-level paging for itself, and thus
5094 	 * NPT isn't supported if the host is using 2-level paging since host
5095 	 * CR4 is unchanged on VMRUN.
5096 	 */
5097 	if (!IS_ENABLED(CONFIG_X86_64) && !IS_ENABLED(CONFIG_X86_PAE))
5098 		npt_enabled = false;
5099 
5100 	if (!boot_cpu_has(X86_FEATURE_NPT))
5101 		npt_enabled = false;
5102 
5103 	/* Force VM NPT level equal to the host's paging level */
5104 	kvm_configure_mmu(npt_enabled, get_npt_level(),
5105 			  get_npt_level(), PG_LEVEL_1G);
5106 	pr_info("Nested Paging %sabled\n", npt_enabled ? "en" : "dis");
5107 
5108 	/* Setup shadow_me_value and shadow_me_mask */
5109 	kvm_mmu_set_me_spte_mask(sme_me_mask, sme_me_mask);
5110 
5111 	svm_adjust_mmio_mask();
5112 
5113 	/*
5114 	 * Note, SEV setup consumes npt_enabled and enable_mmio_caching (which
5115 	 * may be modified by svm_adjust_mmio_mask()).
5116 	 */
5117 	sev_hardware_setup();
5118 
5119 	svm_hv_hardware_setup();
5120 
5121 	for_each_possible_cpu(cpu) {
5122 		r = svm_cpu_init(cpu);
5123 		if (r)
5124 			goto err;
5125 	}
5126 
5127 	if (nrips) {
5128 		if (!boot_cpu_has(X86_FEATURE_NRIPS))
5129 			nrips = false;
5130 	}
5131 
5132 	enable_apicv = avic = avic && avic_hardware_setup();
5133 
5134 	if (!enable_apicv) {
5135 		svm_x86_ops.vcpu_blocking = NULL;
5136 		svm_x86_ops.vcpu_unblocking = NULL;
5137 		svm_x86_ops.vcpu_get_apicv_inhibit_reasons = NULL;
5138 	} else if (!x2avic_enabled) {
5139 		svm_x86_ops.allow_apicv_in_x2apic_without_x2apic_virtualization = true;
5140 	}
5141 
5142 	if (vls) {
5143 		if (!npt_enabled ||
5144 		    !boot_cpu_has(X86_FEATURE_V_VMSAVE_VMLOAD) ||
5145 		    !IS_ENABLED(CONFIG_X86_64)) {
5146 			vls = false;
5147 		} else {
5148 			pr_info("Virtual VMLOAD VMSAVE supported\n");
5149 		}
5150 	}
5151 
5152 	if (boot_cpu_has(X86_FEATURE_SVME_ADDR_CHK))
5153 		svm_gp_erratum_intercept = false;
5154 
5155 	if (vgif) {
5156 		if (!boot_cpu_has(X86_FEATURE_VGIF))
5157 			vgif = false;
5158 		else
5159 			pr_info("Virtual GIF supported\n");
5160 	}
5161 
5162 	vnmi = vgif && vnmi && boot_cpu_has(X86_FEATURE_VNMI);
5163 	if (vnmi)
5164 		pr_info("Virtual NMI enabled\n");
5165 
5166 	if (!vnmi) {
5167 		svm_x86_ops.is_vnmi_pending = NULL;
5168 		svm_x86_ops.set_vnmi_pending = NULL;
5169 	}
5170 
5171 
5172 	if (lbrv) {
5173 		if (!boot_cpu_has(X86_FEATURE_LBRV))
5174 			lbrv = false;
5175 		else
5176 			pr_info("LBR virtualization supported\n");
5177 	}
5178 
5179 	if (!enable_pmu)
5180 		pr_info("PMU virtualization is disabled\n");
5181 
5182 	svm_set_cpu_caps();
5183 
5184 	/*
5185 	 * It seems that on AMD processors PTE's accessed bit is
5186 	 * being set by the CPU hardware before the NPF vmexit.
5187 	 * This is not expected behaviour and our tests fail because
5188 	 * of it.
5189 	 * A workaround here is to disable support for
5190 	 * GUEST_MAXPHYADDR < HOST_MAXPHYADDR if NPT is enabled.
5191 	 * In this case userspace can know if there is support using
5192 	 * KVM_CAP_SMALLER_MAXPHYADDR extension and decide how to handle
5193 	 * it
5194 	 * If future AMD CPU models change the behaviour described above,
5195 	 * this variable can be changed accordingly
5196 	 */
5197 	allow_smaller_maxphyaddr = !npt_enabled;
5198 
5199 	return 0;
5200 
5201 err:
5202 	svm_hardware_unsetup();
5203 	return r;
5204 }
5205 
5206 
5207 static struct kvm_x86_init_ops svm_init_ops __initdata = {
5208 	.hardware_setup = svm_hardware_setup,
5209 
5210 	.runtime_ops = &svm_x86_ops,
5211 	.pmu_ops = &amd_pmu_ops,
5212 };
5213 
5214 static int __init svm_init(void)
5215 {
5216 	int r;
5217 
5218 	__unused_size_checks();
5219 
5220 	if (!kvm_is_svm_supported())
5221 		return -EOPNOTSUPP;
5222 
5223 	r = kvm_x86_vendor_init(&svm_init_ops);
5224 	if (r)
5225 		return r;
5226 
5227 	/*
5228 	 * Common KVM initialization _must_ come last, after this, /dev/kvm is
5229 	 * exposed to userspace!
5230 	 */
5231 	r = kvm_init(sizeof(struct vcpu_svm), __alignof__(struct vcpu_svm),
5232 		     THIS_MODULE);
5233 	if (r)
5234 		goto err_kvm_init;
5235 
5236 	return 0;
5237 
5238 err_kvm_init:
5239 	kvm_x86_vendor_exit();
5240 	return r;
5241 }
5242 
5243 static void __exit svm_exit(void)
5244 {
5245 	kvm_exit();
5246 	kvm_x86_vendor_exit();
5247 }
5248 
5249 module_init(svm_init)
5250 module_exit(svm_exit)
5251