xref: /linux/arch/x86/kvm/x86.h (revision 4b132aacb0768ac1e652cf517097ea6f237214b9)
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  0x0007040600070406ULL
107 
108 void kvm_service_local_tlb_flush_requests(struct kvm_vcpu *vcpu);
109 int kvm_check_nested_events(struct kvm_vcpu *vcpu);
110 
111 static inline bool kvm_vcpu_has_run(struct kvm_vcpu *vcpu)
112 {
113 	return vcpu->arch.last_vmentry_cpu != -1;
114 }
115 
116 static inline bool kvm_is_exception_pending(struct kvm_vcpu *vcpu)
117 {
118 	return vcpu->arch.exception.pending ||
119 	       vcpu->arch.exception_vmexit.pending ||
120 	       kvm_test_request(KVM_REQ_TRIPLE_FAULT, vcpu);
121 }
122 
123 static inline void kvm_clear_exception_queue(struct kvm_vcpu *vcpu)
124 {
125 	vcpu->arch.exception.pending = false;
126 	vcpu->arch.exception.injected = false;
127 	vcpu->arch.exception_vmexit.pending = false;
128 }
129 
130 static inline void kvm_queue_interrupt(struct kvm_vcpu *vcpu, u8 vector,
131 	bool soft)
132 {
133 	vcpu->arch.interrupt.injected = true;
134 	vcpu->arch.interrupt.soft = soft;
135 	vcpu->arch.interrupt.nr = vector;
136 }
137 
138 static inline void kvm_clear_interrupt_queue(struct kvm_vcpu *vcpu)
139 {
140 	vcpu->arch.interrupt.injected = false;
141 }
142 
143 static inline bool kvm_event_needs_reinjection(struct kvm_vcpu *vcpu)
144 {
145 	return vcpu->arch.exception.injected || vcpu->arch.interrupt.injected ||
146 		vcpu->arch.nmi_injected;
147 }
148 
149 static inline bool kvm_exception_is_soft(unsigned int nr)
150 {
151 	return (nr == BP_VECTOR) || (nr == OF_VECTOR);
152 }
153 
154 static inline bool is_protmode(struct kvm_vcpu *vcpu)
155 {
156 	return kvm_is_cr0_bit_set(vcpu, X86_CR0_PE);
157 }
158 
159 static inline bool is_long_mode(struct kvm_vcpu *vcpu)
160 {
161 #ifdef CONFIG_X86_64
162 	return !!(vcpu->arch.efer & EFER_LMA);
163 #else
164 	return false;
165 #endif
166 }
167 
168 static inline bool is_64_bit_mode(struct kvm_vcpu *vcpu)
169 {
170 	int cs_db, cs_l;
171 
172 	WARN_ON_ONCE(vcpu->arch.guest_state_protected);
173 
174 	if (!is_long_mode(vcpu))
175 		return false;
176 	kvm_x86_call(get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
177 	return cs_l;
178 }
179 
180 static inline bool is_64_bit_hypercall(struct kvm_vcpu *vcpu)
181 {
182 	/*
183 	 * If running with protected guest state, the CS register is not
184 	 * accessible. The hypercall register values will have had to been
185 	 * provided in 64-bit mode, so assume the guest is in 64-bit.
186 	 */
187 	return vcpu->arch.guest_state_protected || is_64_bit_mode(vcpu);
188 }
189 
190 static inline bool x86_exception_has_error_code(unsigned int vector)
191 {
192 	static u32 exception_has_error_code = BIT(DF_VECTOR) | BIT(TS_VECTOR) |
193 			BIT(NP_VECTOR) | BIT(SS_VECTOR) | BIT(GP_VECTOR) |
194 			BIT(PF_VECTOR) | BIT(AC_VECTOR);
195 
196 	return (1U << vector) & exception_has_error_code;
197 }
198 
199 static inline bool mmu_is_nested(struct kvm_vcpu *vcpu)
200 {
201 	return vcpu->arch.walk_mmu == &vcpu->arch.nested_mmu;
202 }
203 
204 static inline bool is_pae(struct kvm_vcpu *vcpu)
205 {
206 	return kvm_is_cr4_bit_set(vcpu, X86_CR4_PAE);
207 }
208 
209 static inline bool is_pse(struct kvm_vcpu *vcpu)
210 {
211 	return kvm_is_cr4_bit_set(vcpu, X86_CR4_PSE);
212 }
213 
214 static inline bool is_paging(struct kvm_vcpu *vcpu)
215 {
216 	return likely(kvm_is_cr0_bit_set(vcpu, X86_CR0_PG));
217 }
218 
219 static inline bool is_pae_paging(struct kvm_vcpu *vcpu)
220 {
221 	return !is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu);
222 }
223 
224 static inline u8 vcpu_virt_addr_bits(struct kvm_vcpu *vcpu)
225 {
226 	return kvm_is_cr4_bit_set(vcpu, X86_CR4_LA57) ? 57 : 48;
227 }
228 
229 static inline bool is_noncanonical_address(u64 la, struct kvm_vcpu *vcpu)
230 {
231 	return !__is_canonical_address(la, vcpu_virt_addr_bits(vcpu));
232 }
233 
234 static inline void vcpu_cache_mmio_info(struct kvm_vcpu *vcpu,
235 					gva_t gva, gfn_t gfn, unsigned access)
236 {
237 	u64 gen = kvm_memslots(vcpu->kvm)->generation;
238 
239 	if (unlikely(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS))
240 		return;
241 
242 	/*
243 	 * If this is a shadow nested page table, the "GVA" is
244 	 * actually a nGPA.
245 	 */
246 	vcpu->arch.mmio_gva = mmu_is_nested(vcpu) ? 0 : gva & PAGE_MASK;
247 	vcpu->arch.mmio_access = access;
248 	vcpu->arch.mmio_gfn = gfn;
249 	vcpu->arch.mmio_gen = gen;
250 }
251 
252 static inline bool vcpu_match_mmio_gen(struct kvm_vcpu *vcpu)
253 {
254 	return vcpu->arch.mmio_gen == kvm_memslots(vcpu->kvm)->generation;
255 }
256 
257 /*
258  * Clear the mmio cache info for the given gva. If gva is MMIO_GVA_ANY, we
259  * clear all mmio cache info.
260  */
261 #define MMIO_GVA_ANY (~(gva_t)0)
262 
263 static inline void vcpu_clear_mmio_info(struct kvm_vcpu *vcpu, gva_t gva)
264 {
265 	if (gva != MMIO_GVA_ANY && vcpu->arch.mmio_gva != (gva & PAGE_MASK))
266 		return;
267 
268 	vcpu->arch.mmio_gva = 0;
269 }
270 
271 static inline bool vcpu_match_mmio_gva(struct kvm_vcpu *vcpu, unsigned long gva)
272 {
273 	if (vcpu_match_mmio_gen(vcpu) && vcpu->arch.mmio_gva &&
274 	      vcpu->arch.mmio_gva == (gva & PAGE_MASK))
275 		return true;
276 
277 	return false;
278 }
279 
280 static inline bool vcpu_match_mmio_gpa(struct kvm_vcpu *vcpu, gpa_t gpa)
281 {
282 	if (vcpu_match_mmio_gen(vcpu) && vcpu->arch.mmio_gfn &&
283 	      vcpu->arch.mmio_gfn == gpa >> PAGE_SHIFT)
284 		return true;
285 
286 	return false;
287 }
288 
289 static inline unsigned long kvm_register_read(struct kvm_vcpu *vcpu, int reg)
290 {
291 	unsigned long val = kvm_register_read_raw(vcpu, reg);
292 
293 	return is_64_bit_mode(vcpu) ? val : (u32)val;
294 }
295 
296 static inline void kvm_register_write(struct kvm_vcpu *vcpu,
297 				       int reg, unsigned long val)
298 {
299 	if (!is_64_bit_mode(vcpu))
300 		val = (u32)val;
301 	return kvm_register_write_raw(vcpu, reg, val);
302 }
303 
304 static inline bool kvm_check_has_quirk(struct kvm *kvm, u64 quirk)
305 {
306 	return !(kvm->arch.disabled_quirks & quirk);
307 }
308 
309 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip);
310 
311 u64 get_kvmclock_ns(struct kvm *kvm);
312 uint64_t kvm_get_wall_clock_epoch(struct kvm *kvm);
313 bool kvm_get_monotonic_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp);
314 
315 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
316 	gva_t addr, void *val, unsigned int bytes,
317 	struct x86_exception *exception);
318 
319 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu,
320 	gva_t addr, void *val, unsigned int bytes,
321 	struct x86_exception *exception);
322 
323 int handle_ud(struct kvm_vcpu *vcpu);
324 
325 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu,
326 				   struct kvm_queued_exception *ex);
327 
328 int kvm_mtrr_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data);
329 int kvm_mtrr_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata);
330 bool kvm_vector_hashing_enabled(void);
331 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code);
332 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
333 				    void *insn, int insn_len);
334 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
335 			    int emulation_type, void *insn, int insn_len);
336 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu);
337 
338 extern struct kvm_caps kvm_caps;
339 extern struct kvm_host_values kvm_host;
340 
341 extern bool enable_pmu;
342 
343 /*
344  * Get a filtered version of KVM's supported XCR0 that strips out dynamic
345  * features for which the current process doesn't (yet) have permission to use.
346  * This is intended to be used only when enumerating support to userspace,
347  * e.g. in KVM_GET_SUPPORTED_CPUID and KVM_CAP_XSAVE2, it does NOT need to be
348  * used to check/restrict guest behavior as KVM rejects KVM_SET_CPUID{2} if
349  * userspace attempts to enable unpermitted features.
350  */
351 static inline u64 kvm_get_filtered_xcr0(void)
352 {
353 	u64 permitted_xcr0 = kvm_caps.supported_xcr0;
354 
355 	BUILD_BUG_ON(XFEATURE_MASK_USER_DYNAMIC != XFEATURE_MASK_XTILE_DATA);
356 
357 	if (permitted_xcr0 & XFEATURE_MASK_USER_DYNAMIC) {
358 		permitted_xcr0 &= xstate_get_guest_group_perm();
359 
360 		/*
361 		 * Treat XTILE_CFG as unsupported if the current process isn't
362 		 * allowed to use XTILE_DATA, as attempting to set XTILE_CFG in
363 		 * XCR0 without setting XTILE_DATA is architecturally illegal.
364 		 */
365 		if (!(permitted_xcr0 & XFEATURE_MASK_XTILE_DATA))
366 			permitted_xcr0 &= ~XFEATURE_MASK_XTILE_CFG;
367 	}
368 	return permitted_xcr0;
369 }
370 
371 static inline bool kvm_mpx_supported(void)
372 {
373 	return (kvm_caps.supported_xcr0 & (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR))
374 		== (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR);
375 }
376 
377 extern unsigned int min_timer_period_us;
378 
379 extern bool enable_vmware_backdoor;
380 
381 extern int pi_inject_timer;
382 
383 extern bool report_ignored_msrs;
384 
385 extern bool eager_page_split;
386 
387 static inline void kvm_pr_unimpl_wrmsr(struct kvm_vcpu *vcpu, u32 msr, u64 data)
388 {
389 	if (report_ignored_msrs)
390 		vcpu_unimpl(vcpu, "Unhandled WRMSR(0x%x) = 0x%llx\n", msr, data);
391 }
392 
393 static inline void kvm_pr_unimpl_rdmsr(struct kvm_vcpu *vcpu, u32 msr)
394 {
395 	if (report_ignored_msrs)
396 		vcpu_unimpl(vcpu, "Unhandled RDMSR(0x%x)\n", msr);
397 }
398 
399 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
400 {
401 	return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
402 				   vcpu->arch.virtual_tsc_shift);
403 }
404 
405 /* Same "calling convention" as do_div:
406  * - divide (n << 32) by base
407  * - put result in n
408  * - return remainder
409  */
410 #define do_shl32_div32(n, base)					\
411 	({							\
412 	    u32 __quot, __rem;					\
413 	    asm("divl %2" : "=a" (__quot), "=d" (__rem)		\
414 			: "rm" (base), "0" (0), "1" ((u32) n));	\
415 	    n = __quot;						\
416 	    __rem;						\
417 	 })
418 
419 static inline bool kvm_mwait_in_guest(struct kvm *kvm)
420 {
421 	return kvm->arch.mwait_in_guest;
422 }
423 
424 static inline bool kvm_hlt_in_guest(struct kvm *kvm)
425 {
426 	return kvm->arch.hlt_in_guest;
427 }
428 
429 static inline bool kvm_pause_in_guest(struct kvm *kvm)
430 {
431 	return kvm->arch.pause_in_guest;
432 }
433 
434 static inline bool kvm_cstate_in_guest(struct kvm *kvm)
435 {
436 	return kvm->arch.cstate_in_guest;
437 }
438 
439 static inline bool kvm_notify_vmexit_enabled(struct kvm *kvm)
440 {
441 	return kvm->arch.notify_vmexit_flags & KVM_X86_NOTIFY_VMEXIT_ENABLED;
442 }
443 
444 static __always_inline void kvm_before_interrupt(struct kvm_vcpu *vcpu,
445 						 enum kvm_intr_type intr)
446 {
447 	WRITE_ONCE(vcpu->arch.handling_intr_from_guest, (u8)intr);
448 }
449 
450 static __always_inline void kvm_after_interrupt(struct kvm_vcpu *vcpu)
451 {
452 	WRITE_ONCE(vcpu->arch.handling_intr_from_guest, 0);
453 }
454 
455 static inline bool kvm_handling_nmi_from_guest(struct kvm_vcpu *vcpu)
456 {
457 	return vcpu->arch.handling_intr_from_guest == KVM_HANDLING_NMI;
458 }
459 
460 static inline bool kvm_pat_valid(u64 data)
461 {
462 	if (data & 0xF8F8F8F8F8F8F8F8ull)
463 		return false;
464 	/* 0, 1, 4, 5, 6, 7 are valid values.  */
465 	return (data | ((data & 0x0202020202020202ull) << 1)) == data;
466 }
467 
468 static inline bool kvm_dr7_valid(u64 data)
469 {
470 	/* Bits [63:32] are reserved */
471 	return !(data >> 32);
472 }
473 static inline bool kvm_dr6_valid(u64 data)
474 {
475 	/* Bits [63:32] are reserved */
476 	return !(data >> 32);
477 }
478 
479 /*
480  * Trigger machine check on the host. We assume all the MSRs are already set up
481  * by the CPU and that we still run on the same CPU as the MCE occurred on.
482  * We pass a fake environment to the machine check handler because we want
483  * the guest to be always treated like user space, no matter what context
484  * it used internally.
485  */
486 static inline void kvm_machine_check(void)
487 {
488 #if defined(CONFIG_X86_MCE)
489 	struct pt_regs regs = {
490 		.cs = 3, /* Fake ring 3 no matter what the guest ran on */
491 		.flags = X86_EFLAGS_IF,
492 	};
493 
494 	do_machine_check(&regs);
495 #endif
496 }
497 
498 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu);
499 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu);
500 int kvm_spec_ctrl_test_value(u64 value);
501 bool __kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
502 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
503 			      struct x86_exception *e);
504 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva);
505 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type);
506 
507 /*
508  * Internal error codes that are used to indicate that MSR emulation encountered
509  * an error that should result in #GP in the guest, unless userspace
510  * handles it.
511  */
512 #define  KVM_MSR_RET_INVALID	2	/* in-kernel MSR emulation #GP condition */
513 #define  KVM_MSR_RET_FILTERED	3	/* #GP due to userspace MSR filter */
514 
515 #define __cr4_reserved_bits(__cpu_has, __c)             \
516 ({                                                      \
517 	u64 __reserved_bits = CR4_RESERVED_BITS;        \
518                                                         \
519 	if (!__cpu_has(__c, X86_FEATURE_XSAVE))         \
520 		__reserved_bits |= X86_CR4_OSXSAVE;     \
521 	if (!__cpu_has(__c, X86_FEATURE_SMEP))          \
522 		__reserved_bits |= X86_CR4_SMEP;        \
523 	if (!__cpu_has(__c, X86_FEATURE_SMAP))          \
524 		__reserved_bits |= X86_CR4_SMAP;        \
525 	if (!__cpu_has(__c, X86_FEATURE_FSGSBASE))      \
526 		__reserved_bits |= X86_CR4_FSGSBASE;    \
527 	if (!__cpu_has(__c, X86_FEATURE_PKU))           \
528 		__reserved_bits |= X86_CR4_PKE;         \
529 	if (!__cpu_has(__c, X86_FEATURE_LA57))          \
530 		__reserved_bits |= X86_CR4_LA57;        \
531 	if (!__cpu_has(__c, X86_FEATURE_UMIP))          \
532 		__reserved_bits |= X86_CR4_UMIP;        \
533 	if (!__cpu_has(__c, X86_FEATURE_VMX))           \
534 		__reserved_bits |= X86_CR4_VMXE;        \
535 	if (!__cpu_has(__c, X86_FEATURE_PCID))          \
536 		__reserved_bits |= X86_CR4_PCIDE;       \
537 	if (!__cpu_has(__c, X86_FEATURE_LAM))           \
538 		__reserved_bits |= X86_CR4_LAM_SUP;     \
539 	__reserved_bits;                                \
540 })
541 
542 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t src, unsigned int bytes,
543 			  void *dst);
544 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t src, unsigned int bytes,
545 			 void *dst);
546 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
547 			 unsigned int port, void *data,  unsigned int count,
548 			 int in);
549 
550 #endif
551