xref: /linux/arch/x86/kvm/x86.h (revision 9052e9c95d908d6c3d7570aadc8898e1d871c8bb)
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 void kvm_spurious_fault(void);
12 
13 static __always_inline void kvm_guest_enter_irqoff(void)
14 {
15 	/*
16 	 * VMENTER enables interrupts (host state), but the kernel state is
17 	 * interrupts disabled when this is invoked. Also tell RCU about
18 	 * it. This is the same logic as for exit_to_user_mode().
19 	 *
20 	 * This ensures that e.g. latency analysis on the host observes
21 	 * guest mode as interrupt enabled.
22 	 *
23 	 * guest_enter_irqoff() informs context tracking about the
24 	 * transition to guest mode and if enabled adjusts RCU state
25 	 * accordingly.
26 	 */
27 	instrumentation_begin();
28 	trace_hardirqs_on_prepare();
29 	lockdep_hardirqs_on_prepare(CALLER_ADDR0);
30 	instrumentation_end();
31 
32 	guest_enter_irqoff();
33 	lockdep_hardirqs_on(CALLER_ADDR0);
34 }
35 
36 static __always_inline void kvm_guest_exit_irqoff(void)
37 {
38 	/*
39 	 * VMEXIT disables interrupts (host state), but tracing and lockdep
40 	 * have them in state 'on' as recorded before entering guest mode.
41 	 * Same as enter_from_user_mode().
42 	 *
43 	 * context_tracking_guest_exit() restores host context and reinstates
44 	 * RCU if enabled and required.
45 	 *
46 	 * This needs to be done immediately after VM-Exit, before any code
47 	 * that might contain tracepoints or call out to the greater world,
48 	 * e.g. before x86_spec_ctrl_restore_host().
49 	 */
50 	lockdep_hardirqs_off(CALLER_ADDR0);
51 	context_tracking_guest_exit();
52 
53 	instrumentation_begin();
54 	trace_hardirqs_off_finish();
55 	instrumentation_end();
56 }
57 
58 #define KVM_NESTED_VMENTER_CONSISTENCY_CHECK(consistency_check)		\
59 ({									\
60 	bool failed = (consistency_check);				\
61 	if (failed)							\
62 		trace_kvm_nested_vmenter_failed(#consistency_check, 0);	\
63 	failed;								\
64 })
65 
66 #define KVM_DEFAULT_PLE_GAP		128
67 #define KVM_VMX_DEFAULT_PLE_WINDOW	4096
68 #define KVM_DEFAULT_PLE_WINDOW_GROW	2
69 #define KVM_DEFAULT_PLE_WINDOW_SHRINK	0
70 #define KVM_VMX_DEFAULT_PLE_WINDOW_MAX	UINT_MAX
71 #define KVM_SVM_DEFAULT_PLE_WINDOW_MAX	USHRT_MAX
72 #define KVM_SVM_DEFAULT_PLE_WINDOW	3000
73 
74 static inline unsigned int __grow_ple_window(unsigned int val,
75 		unsigned int base, unsigned int modifier, unsigned int max)
76 {
77 	u64 ret = val;
78 
79 	if (modifier < 1)
80 		return base;
81 
82 	if (modifier < base)
83 		ret *= modifier;
84 	else
85 		ret += modifier;
86 
87 	return min(ret, (u64)max);
88 }
89 
90 static inline unsigned int __shrink_ple_window(unsigned int val,
91 		unsigned int base, unsigned int modifier, unsigned int min)
92 {
93 	if (modifier < 1)
94 		return base;
95 
96 	if (modifier < base)
97 		val /= modifier;
98 	else
99 		val -= modifier;
100 
101 	return max(val, min);
102 }
103 
104 #define MSR_IA32_CR_PAT_DEFAULT  0x0007040600070406ULL
105 
106 int kvm_check_nested_events(struct kvm_vcpu *vcpu);
107 
108 static inline void kvm_clear_exception_queue(struct kvm_vcpu *vcpu)
109 {
110 	vcpu->arch.exception.pending = false;
111 	vcpu->arch.exception.injected = 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_read_cr0_bits(vcpu, X86_CR0_PE);
141 }
142 
143 static inline int is_long_mode(struct kvm_vcpu *vcpu)
144 {
145 #ifdef CONFIG_X86_64
146 	return vcpu->arch.efer & EFER_LMA;
147 #else
148 	return 0;
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 	if (!is_long_mode(vcpu))
157 		return false;
158 	static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
159 	return cs_l;
160 }
161 
162 static inline bool x86_exception_has_error_code(unsigned int vector)
163 {
164 	static u32 exception_has_error_code = BIT(DF_VECTOR) | BIT(TS_VECTOR) |
165 			BIT(NP_VECTOR) | BIT(SS_VECTOR) | BIT(GP_VECTOR) |
166 			BIT(PF_VECTOR) | BIT(AC_VECTOR);
167 
168 	return (1U << vector) & exception_has_error_code;
169 }
170 
171 static inline bool mmu_is_nested(struct kvm_vcpu *vcpu)
172 {
173 	return vcpu->arch.walk_mmu == &vcpu->arch.nested_mmu;
174 }
175 
176 static inline void kvm_vcpu_flush_tlb_current(struct kvm_vcpu *vcpu)
177 {
178 	++vcpu->stat.tlb_flush;
179 	static_call(kvm_x86_tlb_flush_current)(vcpu);
180 }
181 
182 static inline int is_pae(struct kvm_vcpu *vcpu)
183 {
184 	return kvm_read_cr4_bits(vcpu, X86_CR4_PAE);
185 }
186 
187 static inline int is_pse(struct kvm_vcpu *vcpu)
188 {
189 	return kvm_read_cr4_bits(vcpu, X86_CR4_PSE);
190 }
191 
192 static inline int is_paging(struct kvm_vcpu *vcpu)
193 {
194 	return likely(kvm_read_cr0_bits(vcpu, X86_CR0_PG));
195 }
196 
197 static inline bool is_pae_paging(struct kvm_vcpu *vcpu)
198 {
199 	return !is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu);
200 }
201 
202 static inline u8 vcpu_virt_addr_bits(struct kvm_vcpu *vcpu)
203 {
204 	return kvm_read_cr4_bits(vcpu, X86_CR4_LA57) ? 57 : 48;
205 }
206 
207 static inline u64 get_canonical(u64 la, u8 vaddr_bits)
208 {
209 	return ((int64_t)la << (64 - vaddr_bits)) >> (64 - vaddr_bits);
210 }
211 
212 static inline bool is_noncanonical_address(u64 la, struct kvm_vcpu *vcpu)
213 {
214 	return get_canonical(la, vcpu_virt_addr_bits(vcpu)) != la;
215 }
216 
217 static inline void vcpu_cache_mmio_info(struct kvm_vcpu *vcpu,
218 					gva_t gva, gfn_t gfn, unsigned access)
219 {
220 	u64 gen = kvm_memslots(vcpu->kvm)->generation;
221 
222 	if (unlikely(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS))
223 		return;
224 
225 	/*
226 	 * If this is a shadow nested page table, the "GVA" is
227 	 * actually a nGPA.
228 	 */
229 	vcpu->arch.mmio_gva = mmu_is_nested(vcpu) ? 0 : gva & PAGE_MASK;
230 	vcpu->arch.mmio_access = access;
231 	vcpu->arch.mmio_gfn = gfn;
232 	vcpu->arch.mmio_gen = gen;
233 }
234 
235 static inline bool vcpu_match_mmio_gen(struct kvm_vcpu *vcpu)
236 {
237 	return vcpu->arch.mmio_gen == kvm_memslots(vcpu->kvm)->generation;
238 }
239 
240 /*
241  * Clear the mmio cache info for the given gva. If gva is MMIO_GVA_ANY, we
242  * clear all mmio cache info.
243  */
244 #define MMIO_GVA_ANY (~(gva_t)0)
245 
246 static inline void vcpu_clear_mmio_info(struct kvm_vcpu *vcpu, gva_t gva)
247 {
248 	if (gva != MMIO_GVA_ANY && vcpu->arch.mmio_gva != (gva & PAGE_MASK))
249 		return;
250 
251 	vcpu->arch.mmio_gva = 0;
252 }
253 
254 static inline bool vcpu_match_mmio_gva(struct kvm_vcpu *vcpu, unsigned long gva)
255 {
256 	if (vcpu_match_mmio_gen(vcpu) && vcpu->arch.mmio_gva &&
257 	      vcpu->arch.mmio_gva == (gva & PAGE_MASK))
258 		return true;
259 
260 	return false;
261 }
262 
263 static inline bool vcpu_match_mmio_gpa(struct kvm_vcpu *vcpu, gpa_t gpa)
264 {
265 	if (vcpu_match_mmio_gen(vcpu) && vcpu->arch.mmio_gfn &&
266 	      vcpu->arch.mmio_gfn == gpa >> PAGE_SHIFT)
267 		return true;
268 
269 	return false;
270 }
271 
272 static inline unsigned long kvm_register_read(struct kvm_vcpu *vcpu, int reg)
273 {
274 	unsigned long val = kvm_register_read_raw(vcpu, reg);
275 
276 	return is_64_bit_mode(vcpu) ? val : (u32)val;
277 }
278 
279 static inline void kvm_register_write(struct kvm_vcpu *vcpu,
280 				       int reg, unsigned long val)
281 {
282 	if (!is_64_bit_mode(vcpu))
283 		val = (u32)val;
284 	return kvm_register_write_raw(vcpu, reg, val);
285 }
286 
287 static inline bool kvm_check_has_quirk(struct kvm *kvm, u64 quirk)
288 {
289 	return !(kvm->arch.disabled_quirks & quirk);
290 }
291 
292 static inline bool kvm_vcpu_latch_init(struct kvm_vcpu *vcpu)
293 {
294 	return is_smm(vcpu) || static_call(kvm_x86_apic_init_signal_blocked)(vcpu);
295 }
296 
297 void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock, int sec_hi_ofs);
298 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip);
299 
300 u64 get_kvmclock_ns(struct kvm *kvm);
301 
302 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
303 	gva_t addr, void *val, unsigned int bytes,
304 	struct x86_exception *exception);
305 
306 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu,
307 	gva_t addr, void *val, unsigned int bytes,
308 	struct x86_exception *exception);
309 
310 int handle_ud(struct kvm_vcpu *vcpu);
311 
312 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu);
313 
314 void kvm_vcpu_mtrr_init(struct kvm_vcpu *vcpu);
315 u8 kvm_mtrr_get_guest_memory_type(struct kvm_vcpu *vcpu, gfn_t gfn);
316 bool kvm_mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data);
317 int kvm_mtrr_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data);
318 int kvm_mtrr_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata);
319 bool kvm_mtrr_check_gfn_range_consistency(struct kvm_vcpu *vcpu, gfn_t gfn,
320 					  int page_num);
321 bool kvm_vector_hashing_enabled(void);
322 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code);
323 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
324 				    void *insn, int insn_len);
325 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
326 			    int emulation_type, void *insn, int insn_len);
327 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu);
328 
329 extern u64 host_xcr0;
330 extern u64 supported_xcr0;
331 extern u64 host_xss;
332 extern u64 supported_xss;
333 
334 static inline bool kvm_mpx_supported(void)
335 {
336 	return (supported_xcr0 & (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR))
337 		== (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR);
338 }
339 
340 extern unsigned int min_timer_period_us;
341 
342 extern bool enable_vmware_backdoor;
343 
344 extern int pi_inject_timer;
345 
346 extern struct static_key kvm_no_apic_vcpu;
347 
348 extern bool report_ignored_msrs;
349 
350 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
351 {
352 	return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
353 				   vcpu->arch.virtual_tsc_shift);
354 }
355 
356 /* Same "calling convention" as do_div:
357  * - divide (n << 32) by base
358  * - put result in n
359  * - return remainder
360  */
361 #define do_shl32_div32(n, base)					\
362 	({							\
363 	    u32 __quot, __rem;					\
364 	    asm("divl %2" : "=a" (__quot), "=d" (__rem)		\
365 			: "rm" (base), "0" (0), "1" ((u32) n));	\
366 	    n = __quot;						\
367 	    __rem;						\
368 	 })
369 
370 static inline bool kvm_mwait_in_guest(struct kvm *kvm)
371 {
372 	return kvm->arch.mwait_in_guest;
373 }
374 
375 static inline bool kvm_hlt_in_guest(struct kvm *kvm)
376 {
377 	return kvm->arch.hlt_in_guest;
378 }
379 
380 static inline bool kvm_pause_in_guest(struct kvm *kvm)
381 {
382 	return kvm->arch.pause_in_guest;
383 }
384 
385 static inline bool kvm_cstate_in_guest(struct kvm *kvm)
386 {
387 	return kvm->arch.cstate_in_guest;
388 }
389 
390 DECLARE_PER_CPU(struct kvm_vcpu *, current_vcpu);
391 
392 static inline void kvm_before_interrupt(struct kvm_vcpu *vcpu)
393 {
394 	__this_cpu_write(current_vcpu, vcpu);
395 }
396 
397 static inline void kvm_after_interrupt(struct kvm_vcpu *vcpu)
398 {
399 	__this_cpu_write(current_vcpu, NULL);
400 }
401 
402 
403 static inline bool kvm_pat_valid(u64 data)
404 {
405 	if (data & 0xF8F8F8F8F8F8F8F8ull)
406 		return false;
407 	/* 0, 1, 4, 5, 6, 7 are valid values.  */
408 	return (data | ((data & 0x0202020202020202ull) << 1)) == data;
409 }
410 
411 static inline bool kvm_dr7_valid(u64 data)
412 {
413 	/* Bits [63:32] are reserved */
414 	return !(data >> 32);
415 }
416 static inline bool kvm_dr6_valid(u64 data)
417 {
418 	/* Bits [63:32] are reserved */
419 	return !(data >> 32);
420 }
421 
422 /*
423  * Trigger machine check on the host. We assume all the MSRs are already set up
424  * by the CPU and that we still run on the same CPU as the MCE occurred on.
425  * We pass a fake environment to the machine check handler because we want
426  * the guest to be always treated like user space, no matter what context
427  * it used internally.
428  */
429 static inline void kvm_machine_check(void)
430 {
431 #if defined(CONFIG_X86_MCE)
432 	struct pt_regs regs = {
433 		.cs = 3, /* Fake ring 3 no matter what the guest ran on */
434 		.flags = X86_EFLAGS_IF,
435 	};
436 
437 	do_machine_check(&regs);
438 #endif
439 }
440 
441 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu);
442 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu);
443 int kvm_spec_ctrl_test_value(u64 value);
444 bool kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
445 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
446 			      struct x86_exception *e);
447 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva);
448 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type);
449 
450 /*
451  * Internal error codes that are used to indicate that MSR emulation encountered
452  * an error that should result in #GP in the guest, unless userspace
453  * handles it.
454  */
455 #define  KVM_MSR_RET_INVALID	2	/* in-kernel MSR emulation #GP condition */
456 #define  KVM_MSR_RET_FILTERED	3	/* #GP due to userspace MSR filter */
457 
458 #define __cr4_reserved_bits(__cpu_has, __c)             \
459 ({                                                      \
460 	u64 __reserved_bits = CR4_RESERVED_BITS;        \
461                                                         \
462 	if (!__cpu_has(__c, X86_FEATURE_XSAVE))         \
463 		__reserved_bits |= X86_CR4_OSXSAVE;     \
464 	if (!__cpu_has(__c, X86_FEATURE_SMEP))          \
465 		__reserved_bits |= X86_CR4_SMEP;        \
466 	if (!__cpu_has(__c, X86_FEATURE_SMAP))          \
467 		__reserved_bits |= X86_CR4_SMAP;        \
468 	if (!__cpu_has(__c, X86_FEATURE_FSGSBASE))      \
469 		__reserved_bits |= X86_CR4_FSGSBASE;    \
470 	if (!__cpu_has(__c, X86_FEATURE_PKU))           \
471 		__reserved_bits |= X86_CR4_PKE;         \
472 	if (!__cpu_has(__c, X86_FEATURE_LA57))          \
473 		__reserved_bits |= X86_CR4_LA57;        \
474 	if (!__cpu_has(__c, X86_FEATURE_UMIP))          \
475 		__reserved_bits |= X86_CR4_UMIP;        \
476 	if (!__cpu_has(__c, X86_FEATURE_VMX))           \
477 		__reserved_bits |= X86_CR4_VMXE;        \
478 	if (!__cpu_has(__c, X86_FEATURE_PCID))          \
479 		__reserved_bits |= X86_CR4_PCIDE;       \
480 	__reserved_bits;                                \
481 })
482 
483 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t src, unsigned int bytes,
484 			  void *dst);
485 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t src, unsigned int bytes,
486 			 void *dst);
487 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
488 			 unsigned int port, void *data,  unsigned int count,
489 			 int in);
490 
491 #endif
492