xref: /openbmc/linux/arch/x86/kvm/x86.h (revision 6726d552)
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/mce.h>
7 #include <asm/pvclock.h>
8 #include "kvm_cache_regs.h"
9 #include "kvm_emulate.h"
10 
11 struct kvm_caps {
12 	/* control of guest tsc rate supported? */
13 	bool has_tsc_control;
14 	/* maximum supported tsc_khz for guests */
15 	u32  max_guest_tsc_khz;
16 	/* number of bits of the fractional part of the TSC scaling ratio */
17 	u8   tsc_scaling_ratio_frac_bits;
18 	/* maximum allowed value of TSC scaling ratio */
19 	u64  max_tsc_scaling_ratio;
20 	/* 1ull << kvm_caps.tsc_scaling_ratio_frac_bits */
21 	u64  default_tsc_scaling_ratio;
22 	/* bus lock detection supported? */
23 	bool has_bus_lock_exit;
24 	/* notify VM exit supported? */
25 	bool has_notify_vmexit;
26 
27 	u64 supported_mce_cap;
28 	u64 supported_xcr0;
29 	u64 supported_xss;
30 };
31 
32 void kvm_spurious_fault(void);
33 
34 #define KVM_NESTED_VMENTER_CONSISTENCY_CHECK(consistency_check)		\
35 ({									\
36 	bool failed = (consistency_check);				\
37 	if (failed)							\
38 		trace_kvm_nested_vmenter_failed(#consistency_check, 0);	\
39 	failed;								\
40 })
41 
42 #define KVM_DEFAULT_PLE_GAP		128
43 #define KVM_VMX_DEFAULT_PLE_WINDOW	4096
44 #define KVM_DEFAULT_PLE_WINDOW_GROW	2
45 #define KVM_DEFAULT_PLE_WINDOW_SHRINK	0
46 #define KVM_VMX_DEFAULT_PLE_WINDOW_MAX	UINT_MAX
47 #define KVM_SVM_DEFAULT_PLE_WINDOW_MAX	USHRT_MAX
48 #define KVM_SVM_DEFAULT_PLE_WINDOW	3000
49 
50 static inline unsigned int __grow_ple_window(unsigned int val,
51 		unsigned int base, unsigned int modifier, unsigned int max)
52 {
53 	u64 ret = val;
54 
55 	if (modifier < 1)
56 		return base;
57 
58 	if (modifier < base)
59 		ret *= modifier;
60 	else
61 		ret += modifier;
62 
63 	return min(ret, (u64)max);
64 }
65 
66 static inline unsigned int __shrink_ple_window(unsigned int val,
67 		unsigned int base, unsigned int modifier, unsigned int min)
68 {
69 	if (modifier < 1)
70 		return base;
71 
72 	if (modifier < base)
73 		val /= modifier;
74 	else
75 		val -= modifier;
76 
77 	return max(val, min);
78 }
79 
80 #define MSR_IA32_CR_PAT_DEFAULT  0x0007040600070406ULL
81 
82 void kvm_service_local_tlb_flush_requests(struct kvm_vcpu *vcpu);
83 int kvm_check_nested_events(struct kvm_vcpu *vcpu);
84 
85 static inline void kvm_clear_exception_queue(struct kvm_vcpu *vcpu)
86 {
87 	vcpu->arch.exception.pending = false;
88 	vcpu->arch.exception.injected = false;
89 }
90 
91 static inline void kvm_queue_interrupt(struct kvm_vcpu *vcpu, u8 vector,
92 	bool soft)
93 {
94 	vcpu->arch.interrupt.injected = true;
95 	vcpu->arch.interrupt.soft = soft;
96 	vcpu->arch.interrupt.nr = vector;
97 }
98 
99 static inline void kvm_clear_interrupt_queue(struct kvm_vcpu *vcpu)
100 {
101 	vcpu->arch.interrupt.injected = false;
102 }
103 
104 static inline bool kvm_event_needs_reinjection(struct kvm_vcpu *vcpu)
105 {
106 	return vcpu->arch.exception.injected || vcpu->arch.interrupt.injected ||
107 		vcpu->arch.nmi_injected;
108 }
109 
110 static inline bool kvm_exception_is_soft(unsigned int nr)
111 {
112 	return (nr == BP_VECTOR) || (nr == OF_VECTOR);
113 }
114 
115 static inline bool is_protmode(struct kvm_vcpu *vcpu)
116 {
117 	return kvm_read_cr0_bits(vcpu, X86_CR0_PE);
118 }
119 
120 static inline int is_long_mode(struct kvm_vcpu *vcpu)
121 {
122 #ifdef CONFIG_X86_64
123 	return vcpu->arch.efer & EFER_LMA;
124 #else
125 	return 0;
126 #endif
127 }
128 
129 static inline bool is_64_bit_mode(struct kvm_vcpu *vcpu)
130 {
131 	int cs_db, cs_l;
132 
133 	WARN_ON_ONCE(vcpu->arch.guest_state_protected);
134 
135 	if (!is_long_mode(vcpu))
136 		return false;
137 	static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
138 	return cs_l;
139 }
140 
141 static inline bool is_64_bit_hypercall(struct kvm_vcpu *vcpu)
142 {
143 	/*
144 	 * If running with protected guest state, the CS register is not
145 	 * accessible. The hypercall register values will have had to been
146 	 * provided in 64-bit mode, so assume the guest is in 64-bit.
147 	 */
148 	return vcpu->arch.guest_state_protected || is_64_bit_mode(vcpu);
149 }
150 
151 static inline bool x86_exception_has_error_code(unsigned int vector)
152 {
153 	static u32 exception_has_error_code = BIT(DF_VECTOR) | BIT(TS_VECTOR) |
154 			BIT(NP_VECTOR) | BIT(SS_VECTOR) | BIT(GP_VECTOR) |
155 			BIT(PF_VECTOR) | BIT(AC_VECTOR);
156 
157 	return (1U << vector) & exception_has_error_code;
158 }
159 
160 static inline bool mmu_is_nested(struct kvm_vcpu *vcpu)
161 {
162 	return vcpu->arch.walk_mmu == &vcpu->arch.nested_mmu;
163 }
164 
165 static inline int is_pae(struct kvm_vcpu *vcpu)
166 {
167 	return kvm_read_cr4_bits(vcpu, X86_CR4_PAE);
168 }
169 
170 static inline int is_pse(struct kvm_vcpu *vcpu)
171 {
172 	return kvm_read_cr4_bits(vcpu, X86_CR4_PSE);
173 }
174 
175 static inline int is_paging(struct kvm_vcpu *vcpu)
176 {
177 	return likely(kvm_read_cr0_bits(vcpu, X86_CR0_PG));
178 }
179 
180 static inline bool is_pae_paging(struct kvm_vcpu *vcpu)
181 {
182 	return !is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu);
183 }
184 
185 static inline u8 vcpu_virt_addr_bits(struct kvm_vcpu *vcpu)
186 {
187 	return kvm_read_cr4_bits(vcpu, X86_CR4_LA57) ? 57 : 48;
188 }
189 
190 static inline bool is_noncanonical_address(u64 la, struct kvm_vcpu *vcpu)
191 {
192 	return !__is_canonical_address(la, vcpu_virt_addr_bits(vcpu));
193 }
194 
195 static inline void vcpu_cache_mmio_info(struct kvm_vcpu *vcpu,
196 					gva_t gva, gfn_t gfn, unsigned access)
197 {
198 	u64 gen = kvm_memslots(vcpu->kvm)->generation;
199 
200 	if (unlikely(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS))
201 		return;
202 
203 	/*
204 	 * If this is a shadow nested page table, the "GVA" is
205 	 * actually a nGPA.
206 	 */
207 	vcpu->arch.mmio_gva = mmu_is_nested(vcpu) ? 0 : gva & PAGE_MASK;
208 	vcpu->arch.mmio_access = access;
209 	vcpu->arch.mmio_gfn = gfn;
210 	vcpu->arch.mmio_gen = gen;
211 }
212 
213 static inline bool vcpu_match_mmio_gen(struct kvm_vcpu *vcpu)
214 {
215 	return vcpu->arch.mmio_gen == kvm_memslots(vcpu->kvm)->generation;
216 }
217 
218 /*
219  * Clear the mmio cache info for the given gva. If gva is MMIO_GVA_ANY, we
220  * clear all mmio cache info.
221  */
222 #define MMIO_GVA_ANY (~(gva_t)0)
223 
224 static inline void vcpu_clear_mmio_info(struct kvm_vcpu *vcpu, gva_t gva)
225 {
226 	if (gva != MMIO_GVA_ANY && vcpu->arch.mmio_gva != (gva & PAGE_MASK))
227 		return;
228 
229 	vcpu->arch.mmio_gva = 0;
230 }
231 
232 static inline bool vcpu_match_mmio_gva(struct kvm_vcpu *vcpu, unsigned long gva)
233 {
234 	if (vcpu_match_mmio_gen(vcpu) && vcpu->arch.mmio_gva &&
235 	      vcpu->arch.mmio_gva == (gva & PAGE_MASK))
236 		return true;
237 
238 	return false;
239 }
240 
241 static inline bool vcpu_match_mmio_gpa(struct kvm_vcpu *vcpu, gpa_t gpa)
242 {
243 	if (vcpu_match_mmio_gen(vcpu) && vcpu->arch.mmio_gfn &&
244 	      vcpu->arch.mmio_gfn == gpa >> PAGE_SHIFT)
245 		return true;
246 
247 	return false;
248 }
249 
250 static inline unsigned long kvm_register_read(struct kvm_vcpu *vcpu, int reg)
251 {
252 	unsigned long val = kvm_register_read_raw(vcpu, reg);
253 
254 	return is_64_bit_mode(vcpu) ? val : (u32)val;
255 }
256 
257 static inline void kvm_register_write(struct kvm_vcpu *vcpu,
258 				       int reg, unsigned long val)
259 {
260 	if (!is_64_bit_mode(vcpu))
261 		val = (u32)val;
262 	return kvm_register_write_raw(vcpu, reg, val);
263 }
264 
265 static inline bool kvm_check_has_quirk(struct kvm *kvm, u64 quirk)
266 {
267 	return !(kvm->arch.disabled_quirks & quirk);
268 }
269 
270 static inline bool kvm_vcpu_latch_init(struct kvm_vcpu *vcpu)
271 {
272 	return is_smm(vcpu) || static_call(kvm_x86_apic_init_signal_blocked)(vcpu);
273 }
274 
275 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip);
276 
277 u64 get_kvmclock_ns(struct kvm *kvm);
278 
279 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
280 	gva_t addr, void *val, unsigned int bytes,
281 	struct x86_exception *exception);
282 
283 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu,
284 	gva_t addr, void *val, unsigned int bytes,
285 	struct x86_exception *exception);
286 
287 int handle_ud(struct kvm_vcpu *vcpu);
288 
289 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu);
290 
291 void kvm_vcpu_mtrr_init(struct kvm_vcpu *vcpu);
292 u8 kvm_mtrr_get_guest_memory_type(struct kvm_vcpu *vcpu, gfn_t gfn);
293 bool kvm_mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data);
294 int kvm_mtrr_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data);
295 int kvm_mtrr_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata);
296 bool kvm_mtrr_check_gfn_range_consistency(struct kvm_vcpu *vcpu, gfn_t gfn,
297 					  int page_num);
298 bool kvm_vector_hashing_enabled(void);
299 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code);
300 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
301 				    void *insn, int insn_len);
302 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
303 			    int emulation_type, void *insn, int insn_len);
304 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu);
305 
306 extern u64 host_xcr0;
307 extern u64 host_xss;
308 
309 extern struct kvm_caps kvm_caps;
310 
311 extern bool enable_pmu;
312 
313 static inline bool kvm_mpx_supported(void)
314 {
315 	return (kvm_caps.supported_xcr0 & (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR))
316 		== (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR);
317 }
318 
319 extern unsigned int min_timer_period_us;
320 
321 extern bool enable_vmware_backdoor;
322 
323 extern int pi_inject_timer;
324 
325 extern bool report_ignored_msrs;
326 
327 extern bool eager_page_split;
328 
329 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
330 {
331 	return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
332 				   vcpu->arch.virtual_tsc_shift);
333 }
334 
335 /* Same "calling convention" as do_div:
336  * - divide (n << 32) by base
337  * - put result in n
338  * - return remainder
339  */
340 #define do_shl32_div32(n, base)					\
341 	({							\
342 	    u32 __quot, __rem;					\
343 	    asm("divl %2" : "=a" (__quot), "=d" (__rem)		\
344 			: "rm" (base), "0" (0), "1" ((u32) n));	\
345 	    n = __quot;						\
346 	    __rem;						\
347 	 })
348 
349 static inline bool kvm_mwait_in_guest(struct kvm *kvm)
350 {
351 	return kvm->arch.mwait_in_guest;
352 }
353 
354 static inline bool kvm_hlt_in_guest(struct kvm *kvm)
355 {
356 	return kvm->arch.hlt_in_guest;
357 }
358 
359 static inline bool kvm_pause_in_guest(struct kvm *kvm)
360 {
361 	return kvm->arch.pause_in_guest;
362 }
363 
364 static inline bool kvm_cstate_in_guest(struct kvm *kvm)
365 {
366 	return kvm->arch.cstate_in_guest;
367 }
368 
369 static inline bool kvm_notify_vmexit_enabled(struct kvm *kvm)
370 {
371 	return kvm->arch.notify_vmexit_flags & KVM_X86_NOTIFY_VMEXIT_ENABLED;
372 }
373 
374 enum kvm_intr_type {
375 	/* Values are arbitrary, but must be non-zero. */
376 	KVM_HANDLING_IRQ = 1,
377 	KVM_HANDLING_NMI,
378 };
379 
380 static inline void kvm_before_interrupt(struct kvm_vcpu *vcpu,
381 					enum kvm_intr_type intr)
382 {
383 	WRITE_ONCE(vcpu->arch.handling_intr_from_guest, (u8)intr);
384 }
385 
386 static inline void kvm_after_interrupt(struct kvm_vcpu *vcpu)
387 {
388 	WRITE_ONCE(vcpu->arch.handling_intr_from_guest, 0);
389 }
390 
391 static inline bool kvm_handling_nmi_from_guest(struct kvm_vcpu *vcpu)
392 {
393 	return vcpu->arch.handling_intr_from_guest == KVM_HANDLING_NMI;
394 }
395 
396 static inline bool kvm_pat_valid(u64 data)
397 {
398 	if (data & 0xF8F8F8F8F8F8F8F8ull)
399 		return false;
400 	/* 0, 1, 4, 5, 6, 7 are valid values.  */
401 	return (data | ((data & 0x0202020202020202ull) << 1)) == data;
402 }
403 
404 static inline bool kvm_dr7_valid(u64 data)
405 {
406 	/* Bits [63:32] are reserved */
407 	return !(data >> 32);
408 }
409 static inline bool kvm_dr6_valid(u64 data)
410 {
411 	/* Bits [63:32] are reserved */
412 	return !(data >> 32);
413 }
414 
415 /*
416  * Trigger machine check on the host. We assume all the MSRs are already set up
417  * by the CPU and that we still run on the same CPU as the MCE occurred on.
418  * We pass a fake environment to the machine check handler because we want
419  * the guest to be always treated like user space, no matter what context
420  * it used internally.
421  */
422 static inline void kvm_machine_check(void)
423 {
424 #if defined(CONFIG_X86_MCE)
425 	struct pt_regs regs = {
426 		.cs = 3, /* Fake ring 3 no matter what the guest ran on */
427 		.flags = X86_EFLAGS_IF,
428 	};
429 
430 	do_machine_check(&regs);
431 #endif
432 }
433 
434 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu);
435 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu);
436 int kvm_spec_ctrl_test_value(u64 value);
437 bool __kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
438 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
439 			      struct x86_exception *e);
440 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva);
441 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type);
442 
443 /*
444  * Internal error codes that are used to indicate that MSR emulation encountered
445  * an error that should result in #GP in the guest, unless userspace
446  * handles it.
447  */
448 #define  KVM_MSR_RET_INVALID	2	/* in-kernel MSR emulation #GP condition */
449 #define  KVM_MSR_RET_FILTERED	3	/* #GP due to userspace MSR filter */
450 
451 #define __cr4_reserved_bits(__cpu_has, __c)             \
452 ({                                                      \
453 	u64 __reserved_bits = CR4_RESERVED_BITS;        \
454                                                         \
455 	if (!__cpu_has(__c, X86_FEATURE_XSAVE))         \
456 		__reserved_bits |= X86_CR4_OSXSAVE;     \
457 	if (!__cpu_has(__c, X86_FEATURE_SMEP))          \
458 		__reserved_bits |= X86_CR4_SMEP;        \
459 	if (!__cpu_has(__c, X86_FEATURE_SMAP))          \
460 		__reserved_bits |= X86_CR4_SMAP;        \
461 	if (!__cpu_has(__c, X86_FEATURE_FSGSBASE))      \
462 		__reserved_bits |= X86_CR4_FSGSBASE;    \
463 	if (!__cpu_has(__c, X86_FEATURE_PKU))           \
464 		__reserved_bits |= X86_CR4_PKE;         \
465 	if (!__cpu_has(__c, X86_FEATURE_LA57))          \
466 		__reserved_bits |= X86_CR4_LA57;        \
467 	if (!__cpu_has(__c, X86_FEATURE_UMIP))          \
468 		__reserved_bits |= X86_CR4_UMIP;        \
469 	if (!__cpu_has(__c, X86_FEATURE_VMX))           \
470 		__reserved_bits |= X86_CR4_VMXE;        \
471 	if (!__cpu_has(__c, X86_FEATURE_PCID))          \
472 		__reserved_bits |= X86_CR4_PCIDE;       \
473 	__reserved_bits;                                \
474 })
475 
476 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t src, unsigned int bytes,
477 			  void *dst);
478 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t src, unsigned int bytes,
479 			 void *dst);
480 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
481 			 unsigned int port, void *data,  unsigned int count,
482 			 int in);
483 
484 #endif
485