xref: /linux/arch/x86/kvm/x86.h (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
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 	/* bit mask of VM types */
28 	u32 supported_vm_types;
29 
30 	u64 supported_mce_cap;
31 	u64 supported_xcr0;
32 	u64 supported_xss;
33 	u64 supported_perf_cap;
34 };
35 
36 struct kvm_host_values {
37 	/*
38 	 * The host's raw MAXPHYADDR, i.e. the number of non-reserved physical
39 	 * address bits irrespective of features that repurpose legal bits,
40 	 * e.g. MKTME.
41 	 */
42 	u8 maxphyaddr;
43 
44 	u64 efer;
45 	u64 xcr0;
46 	u64 xss;
47 	u64 arch_capabilities;
48 };
49 
50 void kvm_spurious_fault(void);
51 
52 #define KVM_NESTED_VMENTER_CONSISTENCY_CHECK(consistency_check)		\
53 ({									\
54 	bool failed = (consistency_check);				\
55 	if (failed)							\
56 		trace_kvm_nested_vmenter_failed(#consistency_check, 0);	\
57 	failed;								\
58 })
59 
60 /*
61  * The first...last VMX feature MSRs that are emulated by KVM.  This may or may
62  * not cover all known VMX MSRs, as KVM doesn't emulate an MSR until there's an
63  * associated feature that KVM supports for nested virtualization.
64  */
65 #define KVM_FIRST_EMULATED_VMX_MSR	MSR_IA32_VMX_BASIC
66 #define KVM_LAST_EMULATED_VMX_MSR	MSR_IA32_VMX_VMFUNC
67 
68 #define KVM_DEFAULT_PLE_GAP		128
69 #define KVM_VMX_DEFAULT_PLE_WINDOW	4096
70 #define KVM_DEFAULT_PLE_WINDOW_GROW	2
71 #define KVM_DEFAULT_PLE_WINDOW_SHRINK	0
72 #define KVM_VMX_DEFAULT_PLE_WINDOW_MAX	UINT_MAX
73 #define KVM_SVM_DEFAULT_PLE_WINDOW_MAX	USHRT_MAX
74 #define KVM_SVM_DEFAULT_PLE_WINDOW	3000
75 
76 static inline unsigned int __grow_ple_window(unsigned int val,
77 		unsigned int base, unsigned int modifier, unsigned int max)
78 {
79 	u64 ret = val;
80 
81 	if (modifier < 1)
82 		return base;
83 
84 	if (modifier < base)
85 		ret *= modifier;
86 	else
87 		ret += modifier;
88 
89 	return min(ret, (u64)max);
90 }
91 
92 static inline unsigned int __shrink_ple_window(unsigned int val,
93 		unsigned int base, unsigned int modifier, unsigned int min)
94 {
95 	if (modifier < 1)
96 		return base;
97 
98 	if (modifier < base)
99 		val /= modifier;
100 	else
101 		val -= modifier;
102 
103 	return max(val, min);
104 }
105 
106 #define MSR_IA32_CR_PAT_DEFAULT	\
107 	PAT_VALUE(WB, WT, UC_MINUS, UC, WB, WT, UC_MINUS, UC)
108 
109 void kvm_service_local_tlb_flush_requests(struct kvm_vcpu *vcpu);
110 int kvm_check_nested_events(struct kvm_vcpu *vcpu);
111 
112 /* Forcibly leave the nested mode in cases like a vCPU reset */
113 static inline void kvm_leave_nested(struct kvm_vcpu *vcpu)
114 {
115 	kvm_x86_ops.nested_ops->leave_nested(vcpu);
116 }
117 
118 static inline bool kvm_vcpu_has_run(struct kvm_vcpu *vcpu)
119 {
120 	return vcpu->arch.last_vmentry_cpu != -1;
121 }
122 
123 static inline bool kvm_is_exception_pending(struct kvm_vcpu *vcpu)
124 {
125 	return vcpu->arch.exception.pending ||
126 	       vcpu->arch.exception_vmexit.pending ||
127 	       kvm_test_request(KVM_REQ_TRIPLE_FAULT, vcpu);
128 }
129 
130 static inline void kvm_clear_exception_queue(struct kvm_vcpu *vcpu)
131 {
132 	vcpu->arch.exception.pending = false;
133 	vcpu->arch.exception.injected = false;
134 	vcpu->arch.exception_vmexit.pending = false;
135 }
136 
137 static inline void kvm_queue_interrupt(struct kvm_vcpu *vcpu, u8 vector,
138 	bool soft)
139 {
140 	vcpu->arch.interrupt.injected = true;
141 	vcpu->arch.interrupt.soft = soft;
142 	vcpu->arch.interrupt.nr = vector;
143 }
144 
145 static inline void kvm_clear_interrupt_queue(struct kvm_vcpu *vcpu)
146 {
147 	vcpu->arch.interrupt.injected = false;
148 }
149 
150 static inline bool kvm_event_needs_reinjection(struct kvm_vcpu *vcpu)
151 {
152 	return vcpu->arch.exception.injected || vcpu->arch.interrupt.injected ||
153 		vcpu->arch.nmi_injected;
154 }
155 
156 static inline bool kvm_exception_is_soft(unsigned int nr)
157 {
158 	return (nr == BP_VECTOR) || (nr == OF_VECTOR);
159 }
160 
161 static inline bool is_protmode(struct kvm_vcpu *vcpu)
162 {
163 	return kvm_is_cr0_bit_set(vcpu, X86_CR0_PE);
164 }
165 
166 static inline bool is_long_mode(struct kvm_vcpu *vcpu)
167 {
168 #ifdef CONFIG_X86_64
169 	return !!(vcpu->arch.efer & EFER_LMA);
170 #else
171 	return false;
172 #endif
173 }
174 
175 static inline bool is_64_bit_mode(struct kvm_vcpu *vcpu)
176 {
177 	int cs_db, cs_l;
178 
179 	WARN_ON_ONCE(vcpu->arch.guest_state_protected);
180 
181 	if (!is_long_mode(vcpu))
182 		return false;
183 	kvm_x86_call(get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
184 	return cs_l;
185 }
186 
187 static inline bool is_64_bit_hypercall(struct kvm_vcpu *vcpu)
188 {
189 	/*
190 	 * If running with protected guest state, the CS register is not
191 	 * accessible. The hypercall register values will have had to been
192 	 * provided in 64-bit mode, so assume the guest is in 64-bit.
193 	 */
194 	return vcpu->arch.guest_state_protected || is_64_bit_mode(vcpu);
195 }
196 
197 static inline bool x86_exception_has_error_code(unsigned int vector)
198 {
199 	static u32 exception_has_error_code = BIT(DF_VECTOR) | BIT(TS_VECTOR) |
200 			BIT(NP_VECTOR) | BIT(SS_VECTOR) | BIT(GP_VECTOR) |
201 			BIT(PF_VECTOR) | BIT(AC_VECTOR);
202 
203 	return (1U << vector) & exception_has_error_code;
204 }
205 
206 static inline bool mmu_is_nested(struct kvm_vcpu *vcpu)
207 {
208 	return vcpu->arch.walk_mmu == &vcpu->arch.nested_mmu;
209 }
210 
211 static inline bool is_pae(struct kvm_vcpu *vcpu)
212 {
213 	return kvm_is_cr4_bit_set(vcpu, X86_CR4_PAE);
214 }
215 
216 static inline bool is_pse(struct kvm_vcpu *vcpu)
217 {
218 	return kvm_is_cr4_bit_set(vcpu, X86_CR4_PSE);
219 }
220 
221 static inline bool is_paging(struct kvm_vcpu *vcpu)
222 {
223 	return likely(kvm_is_cr0_bit_set(vcpu, X86_CR0_PG));
224 }
225 
226 static inline bool is_pae_paging(struct kvm_vcpu *vcpu)
227 {
228 	return !is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu);
229 }
230 
231 static inline u8 vcpu_virt_addr_bits(struct kvm_vcpu *vcpu)
232 {
233 	return kvm_is_cr4_bit_set(vcpu, X86_CR4_LA57) ? 57 : 48;
234 }
235 
236 static inline bool is_noncanonical_address(u64 la, struct kvm_vcpu *vcpu)
237 {
238 	return !__is_canonical_address(la, vcpu_virt_addr_bits(vcpu));
239 }
240 
241 static inline void vcpu_cache_mmio_info(struct kvm_vcpu *vcpu,
242 					gva_t gva, gfn_t gfn, unsigned access)
243 {
244 	u64 gen = kvm_memslots(vcpu->kvm)->generation;
245 
246 	if (unlikely(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS))
247 		return;
248 
249 	/*
250 	 * If this is a shadow nested page table, the "GVA" is
251 	 * actually a nGPA.
252 	 */
253 	vcpu->arch.mmio_gva = mmu_is_nested(vcpu) ? 0 : gva & PAGE_MASK;
254 	vcpu->arch.mmio_access = access;
255 	vcpu->arch.mmio_gfn = gfn;
256 	vcpu->arch.mmio_gen = gen;
257 }
258 
259 static inline bool vcpu_match_mmio_gen(struct kvm_vcpu *vcpu)
260 {
261 	return vcpu->arch.mmio_gen == kvm_memslots(vcpu->kvm)->generation;
262 }
263 
264 /*
265  * Clear the mmio cache info for the given gva. If gva is MMIO_GVA_ANY, we
266  * clear all mmio cache info.
267  */
268 #define MMIO_GVA_ANY (~(gva_t)0)
269 
270 static inline void vcpu_clear_mmio_info(struct kvm_vcpu *vcpu, gva_t gva)
271 {
272 	if (gva != MMIO_GVA_ANY && vcpu->arch.mmio_gva != (gva & PAGE_MASK))
273 		return;
274 
275 	vcpu->arch.mmio_gva = 0;
276 }
277 
278 static inline bool vcpu_match_mmio_gva(struct kvm_vcpu *vcpu, unsigned long gva)
279 {
280 	if (vcpu_match_mmio_gen(vcpu) && vcpu->arch.mmio_gva &&
281 	      vcpu->arch.mmio_gva == (gva & PAGE_MASK))
282 		return true;
283 
284 	return false;
285 }
286 
287 static inline bool vcpu_match_mmio_gpa(struct kvm_vcpu *vcpu, gpa_t gpa)
288 {
289 	if (vcpu_match_mmio_gen(vcpu) && vcpu->arch.mmio_gfn &&
290 	      vcpu->arch.mmio_gfn == gpa >> PAGE_SHIFT)
291 		return true;
292 
293 	return false;
294 }
295 
296 static inline unsigned long kvm_register_read(struct kvm_vcpu *vcpu, int reg)
297 {
298 	unsigned long val = kvm_register_read_raw(vcpu, reg);
299 
300 	return is_64_bit_mode(vcpu) ? val : (u32)val;
301 }
302 
303 static inline void kvm_register_write(struct kvm_vcpu *vcpu,
304 				       int reg, unsigned long val)
305 {
306 	if (!is_64_bit_mode(vcpu))
307 		val = (u32)val;
308 	return kvm_register_write_raw(vcpu, reg, val);
309 }
310 
311 static inline bool kvm_check_has_quirk(struct kvm *kvm, u64 quirk)
312 {
313 	return !(kvm->arch.disabled_quirks & quirk);
314 }
315 
316 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip);
317 
318 u64 get_kvmclock_ns(struct kvm *kvm);
319 uint64_t kvm_get_wall_clock_epoch(struct kvm *kvm);
320 bool kvm_get_monotonic_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp);
321 
322 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
323 	gva_t addr, void *val, unsigned int bytes,
324 	struct x86_exception *exception);
325 
326 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu,
327 	gva_t addr, void *val, unsigned int bytes,
328 	struct x86_exception *exception);
329 
330 int handle_ud(struct kvm_vcpu *vcpu);
331 
332 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu,
333 				   struct kvm_queued_exception *ex);
334 
335 int kvm_mtrr_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data);
336 int kvm_mtrr_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata);
337 bool kvm_vector_hashing_enabled(void);
338 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code);
339 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
340 				    void *insn, int insn_len);
341 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
342 			    int emulation_type, void *insn, int insn_len);
343 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu);
344 fastpath_t handle_fastpath_hlt(struct kvm_vcpu *vcpu);
345 
346 extern struct kvm_caps kvm_caps;
347 extern struct kvm_host_values kvm_host;
348 
349 extern bool enable_pmu;
350 
351 /*
352  * Get a filtered version of KVM's supported XCR0 that strips out dynamic
353  * features for which the current process doesn't (yet) have permission to use.
354  * This is intended to be used only when enumerating support to userspace,
355  * e.g. in KVM_GET_SUPPORTED_CPUID and KVM_CAP_XSAVE2, it does NOT need to be
356  * used to check/restrict guest behavior as KVM rejects KVM_SET_CPUID{2} if
357  * userspace attempts to enable unpermitted features.
358  */
359 static inline u64 kvm_get_filtered_xcr0(void)
360 {
361 	u64 permitted_xcr0 = kvm_caps.supported_xcr0;
362 
363 	BUILD_BUG_ON(XFEATURE_MASK_USER_DYNAMIC != XFEATURE_MASK_XTILE_DATA);
364 
365 	if (permitted_xcr0 & XFEATURE_MASK_USER_DYNAMIC) {
366 		permitted_xcr0 &= xstate_get_guest_group_perm();
367 
368 		/*
369 		 * Treat XTILE_CFG as unsupported if the current process isn't
370 		 * allowed to use XTILE_DATA, as attempting to set XTILE_CFG in
371 		 * XCR0 without setting XTILE_DATA is architecturally illegal.
372 		 */
373 		if (!(permitted_xcr0 & XFEATURE_MASK_XTILE_DATA))
374 			permitted_xcr0 &= ~XFEATURE_MASK_XTILE_CFG;
375 	}
376 	return permitted_xcr0;
377 }
378 
379 static inline bool kvm_mpx_supported(void)
380 {
381 	return (kvm_caps.supported_xcr0 & (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR))
382 		== (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR);
383 }
384 
385 extern unsigned int min_timer_period_us;
386 
387 extern bool enable_vmware_backdoor;
388 
389 extern int pi_inject_timer;
390 
391 extern bool report_ignored_msrs;
392 
393 extern bool eager_page_split;
394 
395 static inline void kvm_pr_unimpl_wrmsr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
396 {
397 	if (report_ignored_msrs)
398 		vcpu_unimpl(vcpu, "Unhandled WRMSR(0x%x) = 0x%llx\n", msr, data);
399 }
400 
401 static inline void kvm_pr_unimpl_rdmsr(struct kvm_vcpu *vcpu, u32 msr)
402 {
403 	if (report_ignored_msrs)
404 		vcpu_unimpl(vcpu, "Unhandled RDMSR(0x%x)\n", msr);
405 }
406 
407 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
408 {
409 	return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
410 				   vcpu->arch.virtual_tsc_shift);
411 }
412 
413 /* Same "calling convention" as do_div:
414  * - divide (n << 32) by base
415  * - put result in n
416  * - return remainder
417  */
418 #define do_shl32_div32(n, base)					\
419 	({							\
420 	    u32 __quot, __rem;					\
421 	    asm("divl %2" : "=a" (__quot), "=d" (__rem)		\
422 			: "rm" (base), "0" (0), "1" ((u32) n));	\
423 	    n = __quot;						\
424 	    __rem;						\
425 	 })
426 
427 static inline bool kvm_mwait_in_guest(struct kvm *kvm)
428 {
429 	return kvm->arch.mwait_in_guest;
430 }
431 
432 static inline bool kvm_hlt_in_guest(struct kvm *kvm)
433 {
434 	return kvm->arch.hlt_in_guest;
435 }
436 
437 static inline bool kvm_pause_in_guest(struct kvm *kvm)
438 {
439 	return kvm->arch.pause_in_guest;
440 }
441 
442 static inline bool kvm_cstate_in_guest(struct kvm *kvm)
443 {
444 	return kvm->arch.cstate_in_guest;
445 }
446 
447 static inline bool kvm_notify_vmexit_enabled(struct kvm *kvm)
448 {
449 	return kvm->arch.notify_vmexit_flags & KVM_X86_NOTIFY_VMEXIT_ENABLED;
450 }
451 
452 static __always_inline void kvm_before_interrupt(struct kvm_vcpu *vcpu,
453 						 enum kvm_intr_type intr)
454 {
455 	WRITE_ONCE(vcpu->arch.handling_intr_from_guest, (u8)intr);
456 }
457 
458 static __always_inline void kvm_after_interrupt(struct kvm_vcpu *vcpu)
459 {
460 	WRITE_ONCE(vcpu->arch.handling_intr_from_guest, 0);
461 }
462 
463 static inline bool kvm_handling_nmi_from_guest(struct kvm_vcpu *vcpu)
464 {
465 	return vcpu->arch.handling_intr_from_guest == KVM_HANDLING_NMI;
466 }
467 
468 static inline bool kvm_pat_valid(u64 data)
469 {
470 	if (data & 0xF8F8F8F8F8F8F8F8ull)
471 		return false;
472 	/* 0, 1, 4, 5, 6, 7 are valid values.  */
473 	return (data | ((data & 0x0202020202020202ull) << 1)) == data;
474 }
475 
476 static inline bool kvm_dr7_valid(u64 data)
477 {
478 	/* Bits [63:32] are reserved */
479 	return !(data >> 32);
480 }
481 static inline bool kvm_dr6_valid(u64 data)
482 {
483 	/* Bits [63:32] are reserved */
484 	return !(data >> 32);
485 }
486 
487 /*
488  * Trigger machine check on the host. We assume all the MSRs are already set up
489  * by the CPU and that we still run on the same CPU as the MCE occurred on.
490  * We pass a fake environment to the machine check handler because we want
491  * the guest to be always treated like user space, no matter what context
492  * it used internally.
493  */
494 static inline void kvm_machine_check(void)
495 {
496 #if defined(CONFIG_X86_MCE)
497 	struct pt_regs regs = {
498 		.cs = 3, /* Fake ring 3 no matter what the guest ran on */
499 		.flags = X86_EFLAGS_IF,
500 	};
501 
502 	do_machine_check(&regs);
503 #endif
504 }
505 
506 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu);
507 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu);
508 int kvm_spec_ctrl_test_value(u64 value);
509 bool __kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
510 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
511 			      struct x86_exception *e);
512 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva);
513 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type);
514 
515 enum kvm_msr_access {
516 	MSR_TYPE_R	= BIT(0),
517 	MSR_TYPE_W	= BIT(1),
518 	MSR_TYPE_RW	= MSR_TYPE_R | MSR_TYPE_W,
519 };
520 
521 /*
522  * Internal error codes that are used to indicate that MSR emulation encountered
523  * an error that should result in #GP in the guest, unless userspace handles it.
524  * Note, '1', '0', and negative numbers are off limits, as they are used by KVM
525  * as part of KVM's lightly documented internal KVM_RUN return codes.
526  *
527  * UNSUPPORTED	- The MSR isn't supported, either because it is completely
528  *		  unknown to KVM, or because the MSR should not exist according
529  *		  to the vCPU model.
530  *
531  * FILTERED	- Access to the MSR is denied by a userspace MSR filter.
532  */
533 #define  KVM_MSR_RET_UNSUPPORTED	2
534 #define  KVM_MSR_RET_FILTERED		3
535 
536 #define __cr4_reserved_bits(__cpu_has, __c)             \
537 ({                                                      \
538 	u64 __reserved_bits = CR4_RESERVED_BITS;        \
539                                                         \
540 	if (!__cpu_has(__c, X86_FEATURE_XSAVE))         \
541 		__reserved_bits |= X86_CR4_OSXSAVE;     \
542 	if (!__cpu_has(__c, X86_FEATURE_SMEP))          \
543 		__reserved_bits |= X86_CR4_SMEP;        \
544 	if (!__cpu_has(__c, X86_FEATURE_SMAP))          \
545 		__reserved_bits |= X86_CR4_SMAP;        \
546 	if (!__cpu_has(__c, X86_FEATURE_FSGSBASE))      \
547 		__reserved_bits |= X86_CR4_FSGSBASE;    \
548 	if (!__cpu_has(__c, X86_FEATURE_PKU))           \
549 		__reserved_bits |= X86_CR4_PKE;         \
550 	if (!__cpu_has(__c, X86_FEATURE_LA57))          \
551 		__reserved_bits |= X86_CR4_LA57;        \
552 	if (!__cpu_has(__c, X86_FEATURE_UMIP))          \
553 		__reserved_bits |= X86_CR4_UMIP;        \
554 	if (!__cpu_has(__c, X86_FEATURE_VMX))           \
555 		__reserved_bits |= X86_CR4_VMXE;        \
556 	if (!__cpu_has(__c, X86_FEATURE_PCID))          \
557 		__reserved_bits |= X86_CR4_PCIDE;       \
558 	if (!__cpu_has(__c, X86_FEATURE_LAM))           \
559 		__reserved_bits |= X86_CR4_LAM_SUP;     \
560 	__reserved_bits;                                \
561 })
562 
563 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t src, unsigned int bytes,
564 			  void *dst);
565 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t src, unsigned int bytes,
566 			 void *dst);
567 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
568 			 unsigned int port, void *data,  unsigned int count,
569 			 int in);
570 
571 #endif
572