xref: /linux/arch/x86/kvm/x86.h (revision 122333d6bd229af279cdb35d1b874b71b3b9ccfb)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef ARCH_X86_KVM_X86_H
3 #define ARCH_X86_KVM_X86_H
4 
5 #include <linux/kvm_host.h>
6 #include <asm/fpu/xstate.h>
7 #include <asm/mce.h>
8 #include <asm/pvclock.h>
9 #include "kvm_cache_regs.h"
10 #include "kvm_emulate.h"
11 
12 struct kvm_caps {
13 	/* control of guest tsc rate supported? */
14 	bool has_tsc_control;
15 	/* maximum supported tsc_khz for guests */
16 	u32  max_guest_tsc_khz;
17 	/* number of bits of the fractional part of the TSC scaling ratio */
18 	u8   tsc_scaling_ratio_frac_bits;
19 	/* maximum allowed value of TSC scaling ratio */
20 	u64  max_tsc_scaling_ratio;
21 	/* 1ull << kvm_caps.tsc_scaling_ratio_frac_bits */
22 	u64  default_tsc_scaling_ratio;
23 	/* bus lock detection supported? */
24 	bool has_bus_lock_exit;
25 	/* notify VM exit supported? */
26 	bool has_notify_vmexit;
27 
28 	u64 supported_mce_cap;
29 	u64 supported_xcr0;
30 	u64 supported_xss;
31 	u64 supported_perf_cap;
32 };
33 
34 void kvm_spurious_fault(void);
35 
36 #define KVM_NESTED_VMENTER_CONSISTENCY_CHECK(consistency_check)		\
37 ({									\
38 	bool failed = (consistency_check);				\
39 	if (failed)							\
40 		trace_kvm_nested_vmenter_failed(#consistency_check, 0);	\
41 	failed;								\
42 })
43 
44 /*
45  * The first...last VMX feature MSRs that are emulated by KVM.  This may or may
46  * not cover all known VMX MSRs, as KVM doesn't emulate an MSR until there's an
47  * associated feature that KVM supports for nested virtualization.
48  */
49 #define KVM_FIRST_EMULATED_VMX_MSR	MSR_IA32_VMX_BASIC
50 #define KVM_LAST_EMULATED_VMX_MSR	MSR_IA32_VMX_VMFUNC
51 
52 #define KVM_DEFAULT_PLE_GAP		128
53 #define KVM_VMX_DEFAULT_PLE_WINDOW	4096
54 #define KVM_DEFAULT_PLE_WINDOW_GROW	2
55 #define KVM_DEFAULT_PLE_WINDOW_SHRINK	0
56 #define KVM_VMX_DEFAULT_PLE_WINDOW_MAX	UINT_MAX
57 #define KVM_SVM_DEFAULT_PLE_WINDOW_MAX	USHRT_MAX
58 #define KVM_SVM_DEFAULT_PLE_WINDOW	3000
59 
60 static inline unsigned int __grow_ple_window(unsigned int val,
61 		unsigned int base, unsigned int modifier, unsigned int max)
62 {
63 	u64 ret = val;
64 
65 	if (modifier < 1)
66 		return base;
67 
68 	if (modifier < base)
69 		ret *= modifier;
70 	else
71 		ret += modifier;
72 
73 	return min(ret, (u64)max);
74 }
75 
76 static inline unsigned int __shrink_ple_window(unsigned int val,
77 		unsigned int base, unsigned int modifier, unsigned int min)
78 {
79 	if (modifier < 1)
80 		return base;
81 
82 	if (modifier < base)
83 		val /= modifier;
84 	else
85 		val -= modifier;
86 
87 	return max(val, min);
88 }
89 
90 #define MSR_IA32_CR_PAT_DEFAULT  0x0007040600070406ULL
91 
92 void kvm_service_local_tlb_flush_requests(struct kvm_vcpu *vcpu);
93 int kvm_check_nested_events(struct kvm_vcpu *vcpu);
94 
95 static inline bool kvm_vcpu_has_run(struct kvm_vcpu *vcpu)
96 {
97 	return vcpu->arch.last_vmentry_cpu != -1;
98 }
99 
100 static inline bool kvm_is_exception_pending(struct kvm_vcpu *vcpu)
101 {
102 	return vcpu->arch.exception.pending ||
103 	       vcpu->arch.exception_vmexit.pending ||
104 	       kvm_test_request(KVM_REQ_TRIPLE_FAULT, vcpu);
105 }
106 
107 static inline void kvm_clear_exception_queue(struct kvm_vcpu *vcpu)
108 {
109 	vcpu->arch.exception.pending = false;
110 	vcpu->arch.exception.injected = false;
111 	vcpu->arch.exception_vmexit.pending = false;
112 }
113 
114 static inline void kvm_queue_interrupt(struct kvm_vcpu *vcpu, u8 vector,
115 	bool soft)
116 {
117 	vcpu->arch.interrupt.injected = true;
118 	vcpu->arch.interrupt.soft = soft;
119 	vcpu->arch.interrupt.nr = vector;
120 }
121 
122 static inline void kvm_clear_interrupt_queue(struct kvm_vcpu *vcpu)
123 {
124 	vcpu->arch.interrupt.injected = false;
125 }
126 
127 static inline bool kvm_event_needs_reinjection(struct kvm_vcpu *vcpu)
128 {
129 	return vcpu->arch.exception.injected || vcpu->arch.interrupt.injected ||
130 		vcpu->arch.nmi_injected;
131 }
132 
133 static inline bool kvm_exception_is_soft(unsigned int nr)
134 {
135 	return (nr == BP_VECTOR) || (nr == OF_VECTOR);
136 }
137 
138 static inline bool is_protmode(struct kvm_vcpu *vcpu)
139 {
140 	return kvm_is_cr0_bit_set(vcpu, X86_CR0_PE);
141 }
142 
143 static inline bool is_long_mode(struct kvm_vcpu *vcpu)
144 {
145 #ifdef CONFIG_X86_64
146 	return !!(vcpu->arch.efer & EFER_LMA);
147 #else
148 	return false;
149 #endif
150 }
151 
152 static inline bool is_64_bit_mode(struct kvm_vcpu *vcpu)
153 {
154 	int cs_db, cs_l;
155 
156 	WARN_ON_ONCE(vcpu->arch.guest_state_protected);
157 
158 	if (!is_long_mode(vcpu))
159 		return false;
160 	static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
161 	return cs_l;
162 }
163 
164 static inline bool is_64_bit_hypercall(struct kvm_vcpu *vcpu)
165 {
166 	/*
167 	 * If running with protected guest state, the CS register is not
168 	 * accessible. The hypercall register values will have had to been
169 	 * provided in 64-bit mode, so assume the guest is in 64-bit.
170 	 */
171 	return vcpu->arch.guest_state_protected || is_64_bit_mode(vcpu);
172 }
173 
174 static inline bool x86_exception_has_error_code(unsigned int vector)
175 {
176 	static u32 exception_has_error_code = BIT(DF_VECTOR) | BIT(TS_VECTOR) |
177 			BIT(NP_VECTOR) | BIT(SS_VECTOR) | BIT(GP_VECTOR) |
178 			BIT(PF_VECTOR) | BIT(AC_VECTOR);
179 
180 	return (1U << vector) & exception_has_error_code;
181 }
182 
183 static inline bool mmu_is_nested(struct kvm_vcpu *vcpu)
184 {
185 	return vcpu->arch.walk_mmu == &vcpu->arch.nested_mmu;
186 }
187 
188 static inline bool is_pae(struct kvm_vcpu *vcpu)
189 {
190 	return kvm_is_cr4_bit_set(vcpu, X86_CR4_PAE);
191 }
192 
193 static inline bool is_pse(struct kvm_vcpu *vcpu)
194 {
195 	return kvm_is_cr4_bit_set(vcpu, X86_CR4_PSE);
196 }
197 
198 static inline bool is_paging(struct kvm_vcpu *vcpu)
199 {
200 	return likely(kvm_is_cr0_bit_set(vcpu, X86_CR0_PG));
201 }
202 
203 static inline bool is_pae_paging(struct kvm_vcpu *vcpu)
204 {
205 	return !is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu);
206 }
207 
208 static inline u8 vcpu_virt_addr_bits(struct kvm_vcpu *vcpu)
209 {
210 	return kvm_is_cr4_bit_set(vcpu, X86_CR4_LA57) ? 57 : 48;
211 }
212 
213 static inline bool is_noncanonical_address(u64 la, struct kvm_vcpu *vcpu)
214 {
215 	return !__is_canonical_address(la, vcpu_virt_addr_bits(vcpu));
216 }
217 
218 static inline void vcpu_cache_mmio_info(struct kvm_vcpu *vcpu,
219 					gva_t gva, gfn_t gfn, unsigned access)
220 {
221 	u64 gen = kvm_memslots(vcpu->kvm)->generation;
222 
223 	if (unlikely(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS))
224 		return;
225 
226 	/*
227 	 * If this is a shadow nested page table, the "GVA" is
228 	 * actually a nGPA.
229 	 */
230 	vcpu->arch.mmio_gva = mmu_is_nested(vcpu) ? 0 : gva & PAGE_MASK;
231 	vcpu->arch.mmio_access = access;
232 	vcpu->arch.mmio_gfn = gfn;
233 	vcpu->arch.mmio_gen = gen;
234 }
235 
236 static inline bool vcpu_match_mmio_gen(struct kvm_vcpu *vcpu)
237 {
238 	return vcpu->arch.mmio_gen == kvm_memslots(vcpu->kvm)->generation;
239 }
240 
241 /*
242  * Clear the mmio cache info for the given gva. If gva is MMIO_GVA_ANY, we
243  * clear all mmio cache info.
244  */
245 #define MMIO_GVA_ANY (~(gva_t)0)
246 
247 static inline void vcpu_clear_mmio_info(struct kvm_vcpu *vcpu, gva_t gva)
248 {
249 	if (gva != MMIO_GVA_ANY && vcpu->arch.mmio_gva != (gva & PAGE_MASK))
250 		return;
251 
252 	vcpu->arch.mmio_gva = 0;
253 }
254 
255 static inline bool vcpu_match_mmio_gva(struct kvm_vcpu *vcpu, unsigned long gva)
256 {
257 	if (vcpu_match_mmio_gen(vcpu) && vcpu->arch.mmio_gva &&
258 	      vcpu->arch.mmio_gva == (gva & PAGE_MASK))
259 		return true;
260 
261 	return false;
262 }
263 
264 static inline bool vcpu_match_mmio_gpa(struct kvm_vcpu *vcpu, gpa_t gpa)
265 {
266 	if (vcpu_match_mmio_gen(vcpu) && vcpu->arch.mmio_gfn &&
267 	      vcpu->arch.mmio_gfn == gpa >> PAGE_SHIFT)
268 		return true;
269 
270 	return false;
271 }
272 
273 static inline unsigned long kvm_register_read(struct kvm_vcpu *vcpu, int reg)
274 {
275 	unsigned long val = kvm_register_read_raw(vcpu, reg);
276 
277 	return is_64_bit_mode(vcpu) ? val : (u32)val;
278 }
279 
280 static inline void kvm_register_write(struct kvm_vcpu *vcpu,
281 				       int reg, unsigned long val)
282 {
283 	if (!is_64_bit_mode(vcpu))
284 		val = (u32)val;
285 	return kvm_register_write_raw(vcpu, reg, val);
286 }
287 
288 static inline bool kvm_check_has_quirk(struct kvm *kvm, u64 quirk)
289 {
290 	return !(kvm->arch.disabled_quirks & quirk);
291 }
292 
293 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip);
294 
295 u64 get_kvmclock_ns(struct kvm *kvm);
296 
297 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
298 	gva_t addr, void *val, unsigned int bytes,
299 	struct x86_exception *exception);
300 
301 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu,
302 	gva_t addr, void *val, unsigned int bytes,
303 	struct x86_exception *exception);
304 
305 int handle_ud(struct kvm_vcpu *vcpu);
306 
307 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu,
308 				   struct kvm_queued_exception *ex);
309 
310 void kvm_vcpu_mtrr_init(struct kvm_vcpu *vcpu);
311 u8 kvm_mtrr_get_guest_memory_type(struct kvm_vcpu *vcpu, gfn_t gfn);
312 bool kvm_mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data);
313 int kvm_mtrr_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data);
314 int kvm_mtrr_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata);
315 bool kvm_mtrr_check_gfn_range_consistency(struct kvm_vcpu *vcpu, gfn_t gfn,
316 					  int page_num);
317 bool kvm_vector_hashing_enabled(void);
318 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code);
319 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
320 				    void *insn, int insn_len);
321 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
322 			    int emulation_type, void *insn, int insn_len);
323 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu);
324 
325 extern u64 host_xcr0;
326 extern u64 host_xss;
327 
328 extern struct kvm_caps kvm_caps;
329 
330 extern bool enable_pmu;
331 
332 /*
333  * Get a filtered version of KVM's supported XCR0 that strips out dynamic
334  * features for which the current process doesn't (yet) have permission to use.
335  * This is intended to be used only when enumerating support to userspace,
336  * e.g. in KVM_GET_SUPPORTED_CPUID and KVM_CAP_XSAVE2, it does NOT need to be
337  * used to check/restrict guest behavior as KVM rejects KVM_SET_CPUID{2} if
338  * userspace attempts to enable unpermitted features.
339  */
340 static inline u64 kvm_get_filtered_xcr0(void)
341 {
342 	u64 permitted_xcr0 = kvm_caps.supported_xcr0;
343 
344 	BUILD_BUG_ON(XFEATURE_MASK_USER_DYNAMIC != XFEATURE_MASK_XTILE_DATA);
345 
346 	if (permitted_xcr0 & XFEATURE_MASK_USER_DYNAMIC) {
347 		permitted_xcr0 &= xstate_get_guest_group_perm();
348 
349 		/*
350 		 * Treat XTILE_CFG as unsupported if the current process isn't
351 		 * allowed to use XTILE_DATA, as attempting to set XTILE_CFG in
352 		 * XCR0 without setting XTILE_DATA is architecturally illegal.
353 		 */
354 		if (!(permitted_xcr0 & XFEATURE_MASK_XTILE_DATA))
355 			permitted_xcr0 &= ~XFEATURE_MASK_XTILE_CFG;
356 	}
357 	return permitted_xcr0;
358 }
359 
360 static inline bool kvm_mpx_supported(void)
361 {
362 	return (kvm_caps.supported_xcr0 & (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR))
363 		== (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR);
364 }
365 
366 extern unsigned int min_timer_period_us;
367 
368 extern bool enable_vmware_backdoor;
369 
370 extern int pi_inject_timer;
371 
372 extern bool report_ignored_msrs;
373 
374 extern bool eager_page_split;
375 
376 static inline void kvm_pr_unimpl_wrmsr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
377 {
378 	if (report_ignored_msrs)
379 		vcpu_unimpl(vcpu, "Unhandled WRMSR(0x%x) = 0x%llx\n", msr, data);
380 }
381 
382 static inline void kvm_pr_unimpl_rdmsr(struct kvm_vcpu *vcpu, u32 msr)
383 {
384 	if (report_ignored_msrs)
385 		vcpu_unimpl(vcpu, "Unhandled RDMSR(0x%x)\n", msr);
386 }
387 
388 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
389 {
390 	return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
391 				   vcpu->arch.virtual_tsc_shift);
392 }
393 
394 /* Same "calling convention" as do_div:
395  * - divide (n << 32) by base
396  * - put result in n
397  * - return remainder
398  */
399 #define do_shl32_div32(n, base)					\
400 	({							\
401 	    u32 __quot, __rem;					\
402 	    asm("divl %2" : "=a" (__quot), "=d" (__rem)		\
403 			: "rm" (base), "0" (0), "1" ((u32) n));	\
404 	    n = __quot;						\
405 	    __rem;						\
406 	 })
407 
408 static inline bool kvm_mwait_in_guest(struct kvm *kvm)
409 {
410 	return kvm->arch.mwait_in_guest;
411 }
412 
413 static inline bool kvm_hlt_in_guest(struct kvm *kvm)
414 {
415 	return kvm->arch.hlt_in_guest;
416 }
417 
418 static inline bool kvm_pause_in_guest(struct kvm *kvm)
419 {
420 	return kvm->arch.pause_in_guest;
421 }
422 
423 static inline bool kvm_cstate_in_guest(struct kvm *kvm)
424 {
425 	return kvm->arch.cstate_in_guest;
426 }
427 
428 static inline bool kvm_notify_vmexit_enabled(struct kvm *kvm)
429 {
430 	return kvm->arch.notify_vmexit_flags & KVM_X86_NOTIFY_VMEXIT_ENABLED;
431 }
432 
433 enum kvm_intr_type {
434 	/* Values are arbitrary, but must be non-zero. */
435 	KVM_HANDLING_IRQ = 1,
436 	KVM_HANDLING_NMI,
437 };
438 
439 static __always_inline void kvm_before_interrupt(struct kvm_vcpu *vcpu,
440 						 enum kvm_intr_type intr)
441 {
442 	WRITE_ONCE(vcpu->arch.handling_intr_from_guest, (u8)intr);
443 }
444 
445 static __always_inline void kvm_after_interrupt(struct kvm_vcpu *vcpu)
446 {
447 	WRITE_ONCE(vcpu->arch.handling_intr_from_guest, 0);
448 }
449 
450 static inline bool kvm_handling_nmi_from_guest(struct kvm_vcpu *vcpu)
451 {
452 	return vcpu->arch.handling_intr_from_guest == KVM_HANDLING_NMI;
453 }
454 
455 static inline bool kvm_pat_valid(u64 data)
456 {
457 	if (data & 0xF8F8F8F8F8F8F8F8ull)
458 		return false;
459 	/* 0, 1, 4, 5, 6, 7 are valid values.  */
460 	return (data | ((data & 0x0202020202020202ull) << 1)) == data;
461 }
462 
463 static inline bool kvm_dr7_valid(u64 data)
464 {
465 	/* Bits [63:32] are reserved */
466 	return !(data >> 32);
467 }
468 static inline bool kvm_dr6_valid(u64 data)
469 {
470 	/* Bits [63:32] are reserved */
471 	return !(data >> 32);
472 }
473 
474 /*
475  * Trigger machine check on the host. We assume all the MSRs are already set up
476  * by the CPU and that we still run on the same CPU as the MCE occurred on.
477  * We pass a fake environment to the machine check handler because we want
478  * the guest to be always treated like user space, no matter what context
479  * it used internally.
480  */
481 static inline void kvm_machine_check(void)
482 {
483 #if defined(CONFIG_X86_MCE)
484 	struct pt_regs regs = {
485 		.cs = 3, /* Fake ring 3 no matter what the guest ran on */
486 		.flags = X86_EFLAGS_IF,
487 	};
488 
489 	do_machine_check(&regs);
490 #endif
491 }
492 
493 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu);
494 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu);
495 int kvm_spec_ctrl_test_value(u64 value);
496 bool __kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
497 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
498 			      struct x86_exception *e);
499 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva);
500 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type);
501 
502 /*
503  * Internal error codes that are used to indicate that MSR emulation encountered
504  * an error that should result in #GP in the guest, unless userspace
505  * handles it.
506  */
507 #define  KVM_MSR_RET_INVALID	2	/* in-kernel MSR emulation #GP condition */
508 #define  KVM_MSR_RET_FILTERED	3	/* #GP due to userspace MSR filter */
509 
510 #define __cr4_reserved_bits(__cpu_has, __c)             \
511 ({                                                      \
512 	u64 __reserved_bits = CR4_RESERVED_BITS;        \
513                                                         \
514 	if (!__cpu_has(__c, X86_FEATURE_XSAVE))         \
515 		__reserved_bits |= X86_CR4_OSXSAVE;     \
516 	if (!__cpu_has(__c, X86_FEATURE_SMEP))          \
517 		__reserved_bits |= X86_CR4_SMEP;        \
518 	if (!__cpu_has(__c, X86_FEATURE_SMAP))          \
519 		__reserved_bits |= X86_CR4_SMAP;        \
520 	if (!__cpu_has(__c, X86_FEATURE_FSGSBASE))      \
521 		__reserved_bits |= X86_CR4_FSGSBASE;    \
522 	if (!__cpu_has(__c, X86_FEATURE_PKU))           \
523 		__reserved_bits |= X86_CR4_PKE;         \
524 	if (!__cpu_has(__c, X86_FEATURE_LA57))          \
525 		__reserved_bits |= X86_CR4_LA57;        \
526 	if (!__cpu_has(__c, X86_FEATURE_UMIP))          \
527 		__reserved_bits |= X86_CR4_UMIP;        \
528 	if (!__cpu_has(__c, X86_FEATURE_VMX))           \
529 		__reserved_bits |= X86_CR4_VMXE;        \
530 	if (!__cpu_has(__c, X86_FEATURE_PCID))          \
531 		__reserved_bits |= X86_CR4_PCIDE;       \
532 	__reserved_bits;                                \
533 })
534 
535 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t src, unsigned int bytes,
536 			  void *dst);
537 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t src, unsigned int bytes,
538 			 void *dst);
539 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
540 			 unsigned int port, void *data,  unsigned int count,
541 			 int in);
542 
543 #endif
544