xref: /linux/arch/x86/kvm/x86.h (revision a8deba8547e39f26440101164a3bbc2899c5b305)
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 #define KVM_NESTED_VMENTER_CONSISTENCY_CHECK(consistency_check)		\
12 ({									\
13 	bool failed = (consistency_check);				\
14 	if (failed)							\
15 		trace_kvm_nested_vmenter_failed(#consistency_check, 0);	\
16 	failed;								\
17 })
18 
19 #define KVM_DEFAULT_PLE_GAP		128
20 #define KVM_VMX_DEFAULT_PLE_WINDOW	4096
21 #define KVM_DEFAULT_PLE_WINDOW_GROW	2
22 #define KVM_DEFAULT_PLE_WINDOW_SHRINK	0
23 #define KVM_VMX_DEFAULT_PLE_WINDOW_MAX	UINT_MAX
24 #define KVM_SVM_DEFAULT_PLE_WINDOW_MAX	USHRT_MAX
25 #define KVM_SVM_DEFAULT_PLE_WINDOW	3000
26 
27 static inline unsigned int __grow_ple_window(unsigned int val,
28 		unsigned int base, unsigned int modifier, unsigned int max)
29 {
30 	u64 ret = val;
31 
32 	if (modifier < 1)
33 		return base;
34 
35 	if (modifier < base)
36 		ret *= modifier;
37 	else
38 		ret += modifier;
39 
40 	return min(ret, (u64)max);
41 }
42 
43 static inline unsigned int __shrink_ple_window(unsigned int val,
44 		unsigned int base, unsigned int modifier, unsigned int min)
45 {
46 	if (modifier < 1)
47 		return base;
48 
49 	if (modifier < base)
50 		val /= modifier;
51 	else
52 		val -= modifier;
53 
54 	return max(val, min);
55 }
56 
57 #define MSR_IA32_CR_PAT_DEFAULT  0x0007040600070406ULL
58 
59 int kvm_check_nested_events(struct kvm_vcpu *vcpu);
60 
61 static inline void kvm_clear_exception_queue(struct kvm_vcpu *vcpu)
62 {
63 	vcpu->arch.exception.pending = false;
64 	vcpu->arch.exception.injected = false;
65 }
66 
67 static inline void kvm_queue_interrupt(struct kvm_vcpu *vcpu, u8 vector,
68 	bool soft)
69 {
70 	vcpu->arch.interrupt.injected = true;
71 	vcpu->arch.interrupt.soft = soft;
72 	vcpu->arch.interrupt.nr = vector;
73 }
74 
75 static inline void kvm_clear_interrupt_queue(struct kvm_vcpu *vcpu)
76 {
77 	vcpu->arch.interrupt.injected = false;
78 }
79 
80 static inline bool kvm_event_needs_reinjection(struct kvm_vcpu *vcpu)
81 {
82 	return vcpu->arch.exception.injected || vcpu->arch.interrupt.injected ||
83 		vcpu->arch.nmi_injected;
84 }
85 
86 static inline bool kvm_exception_is_soft(unsigned int nr)
87 {
88 	return (nr == BP_VECTOR) || (nr == OF_VECTOR);
89 }
90 
91 static inline bool is_protmode(struct kvm_vcpu *vcpu)
92 {
93 	return kvm_read_cr0_bits(vcpu, X86_CR0_PE);
94 }
95 
96 static inline int is_long_mode(struct kvm_vcpu *vcpu)
97 {
98 #ifdef CONFIG_X86_64
99 	return vcpu->arch.efer & EFER_LMA;
100 #else
101 	return 0;
102 #endif
103 }
104 
105 static inline bool is_64_bit_mode(struct kvm_vcpu *vcpu)
106 {
107 	int cs_db, cs_l;
108 
109 	if (!is_long_mode(vcpu))
110 		return false;
111 	static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
112 	return cs_l;
113 }
114 
115 static inline bool is_la57_mode(struct kvm_vcpu *vcpu)
116 {
117 #ifdef CONFIG_X86_64
118 	return (vcpu->arch.efer & EFER_LMA) &&
119 		 kvm_read_cr4_bits(vcpu, X86_CR4_LA57);
120 #else
121 	return 0;
122 #endif
123 }
124 
125 static inline bool x86_exception_has_error_code(unsigned int vector)
126 {
127 	static u32 exception_has_error_code = BIT(DF_VECTOR) | BIT(TS_VECTOR) |
128 			BIT(NP_VECTOR) | BIT(SS_VECTOR) | BIT(GP_VECTOR) |
129 			BIT(PF_VECTOR) | BIT(AC_VECTOR);
130 
131 	return (1U << vector) & exception_has_error_code;
132 }
133 
134 static inline bool mmu_is_nested(struct kvm_vcpu *vcpu)
135 {
136 	return vcpu->arch.walk_mmu == &vcpu->arch.nested_mmu;
137 }
138 
139 static inline void kvm_vcpu_flush_tlb_current(struct kvm_vcpu *vcpu)
140 {
141 	++vcpu->stat.tlb_flush;
142 	static_call(kvm_x86_tlb_flush_current)(vcpu);
143 }
144 
145 static inline int is_pae(struct kvm_vcpu *vcpu)
146 {
147 	return kvm_read_cr4_bits(vcpu, X86_CR4_PAE);
148 }
149 
150 static inline int is_pse(struct kvm_vcpu *vcpu)
151 {
152 	return kvm_read_cr4_bits(vcpu, X86_CR4_PSE);
153 }
154 
155 static inline int is_paging(struct kvm_vcpu *vcpu)
156 {
157 	return likely(kvm_read_cr0_bits(vcpu, X86_CR0_PG));
158 }
159 
160 static inline bool is_pae_paging(struct kvm_vcpu *vcpu)
161 {
162 	return !is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu);
163 }
164 
165 static inline u8 vcpu_virt_addr_bits(struct kvm_vcpu *vcpu)
166 {
167 	return kvm_read_cr4_bits(vcpu, X86_CR4_LA57) ? 57 : 48;
168 }
169 
170 static inline u64 get_canonical(u64 la, u8 vaddr_bits)
171 {
172 	return ((int64_t)la << (64 - vaddr_bits)) >> (64 - vaddr_bits);
173 }
174 
175 static inline bool is_noncanonical_address(u64 la, struct kvm_vcpu *vcpu)
176 {
177 	return get_canonical(la, vcpu_virt_addr_bits(vcpu)) != la;
178 }
179 
180 static inline void vcpu_cache_mmio_info(struct kvm_vcpu *vcpu,
181 					gva_t gva, gfn_t gfn, unsigned access)
182 {
183 	u64 gen = kvm_memslots(vcpu->kvm)->generation;
184 
185 	if (unlikely(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS))
186 		return;
187 
188 	/*
189 	 * If this is a shadow nested page table, the "GVA" is
190 	 * actually a nGPA.
191 	 */
192 	vcpu->arch.mmio_gva = mmu_is_nested(vcpu) ? 0 : gva & PAGE_MASK;
193 	vcpu->arch.mmio_access = access;
194 	vcpu->arch.mmio_gfn = gfn;
195 	vcpu->arch.mmio_gen = gen;
196 }
197 
198 static inline bool vcpu_match_mmio_gen(struct kvm_vcpu *vcpu)
199 {
200 	return vcpu->arch.mmio_gen == kvm_memslots(vcpu->kvm)->generation;
201 }
202 
203 /*
204  * Clear the mmio cache info for the given gva. If gva is MMIO_GVA_ANY, we
205  * clear all mmio cache info.
206  */
207 #define MMIO_GVA_ANY (~(gva_t)0)
208 
209 static inline void vcpu_clear_mmio_info(struct kvm_vcpu *vcpu, gva_t gva)
210 {
211 	if (gva != MMIO_GVA_ANY && vcpu->arch.mmio_gva != (gva & PAGE_MASK))
212 		return;
213 
214 	vcpu->arch.mmio_gva = 0;
215 }
216 
217 static inline bool vcpu_match_mmio_gva(struct kvm_vcpu *vcpu, unsigned long gva)
218 {
219 	if (vcpu_match_mmio_gen(vcpu) && vcpu->arch.mmio_gva &&
220 	      vcpu->arch.mmio_gva == (gva & PAGE_MASK))
221 		return true;
222 
223 	return false;
224 }
225 
226 static inline bool vcpu_match_mmio_gpa(struct kvm_vcpu *vcpu, gpa_t gpa)
227 {
228 	if (vcpu_match_mmio_gen(vcpu) && vcpu->arch.mmio_gfn &&
229 	      vcpu->arch.mmio_gfn == gpa >> PAGE_SHIFT)
230 		return true;
231 
232 	return false;
233 }
234 
235 static inline unsigned long kvm_register_read(struct kvm_vcpu *vcpu, int reg)
236 {
237 	unsigned long val = kvm_register_read_raw(vcpu, reg);
238 
239 	return is_64_bit_mode(vcpu) ? val : (u32)val;
240 }
241 
242 static inline void kvm_register_write(struct kvm_vcpu *vcpu,
243 				       int reg, unsigned long val)
244 {
245 	if (!is_64_bit_mode(vcpu))
246 		val = (u32)val;
247 	return kvm_register_write_raw(vcpu, reg, val);
248 }
249 
250 static inline bool kvm_check_has_quirk(struct kvm *kvm, u64 quirk)
251 {
252 	return !(kvm->arch.disabled_quirks & quirk);
253 }
254 
255 static inline bool kvm_vcpu_latch_init(struct kvm_vcpu *vcpu)
256 {
257 	return is_smm(vcpu) || static_call(kvm_x86_apic_init_signal_blocked)(vcpu);
258 }
259 
260 void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock, int sec_hi_ofs);
261 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip);
262 
263 u64 get_kvmclock_ns(struct kvm *kvm);
264 
265 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
266 	gva_t addr, void *val, unsigned int bytes,
267 	struct x86_exception *exception);
268 
269 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu,
270 	gva_t addr, void *val, unsigned int bytes,
271 	struct x86_exception *exception);
272 
273 int handle_ud(struct kvm_vcpu *vcpu);
274 
275 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu);
276 
277 void kvm_vcpu_mtrr_init(struct kvm_vcpu *vcpu);
278 u8 kvm_mtrr_get_guest_memory_type(struct kvm_vcpu *vcpu, gfn_t gfn);
279 bool kvm_mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data);
280 int kvm_mtrr_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data);
281 int kvm_mtrr_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata);
282 bool kvm_mtrr_check_gfn_range_consistency(struct kvm_vcpu *vcpu, gfn_t gfn,
283 					  int page_num);
284 bool kvm_vector_hashing_enabled(void);
285 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code);
286 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
287 				    void *insn, int insn_len);
288 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
289 			    int emulation_type, void *insn, int insn_len);
290 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu);
291 
292 extern u64 host_xcr0;
293 extern u64 supported_xcr0;
294 extern u64 host_xss;
295 extern u64 supported_xss;
296 
297 static inline bool kvm_mpx_supported(void)
298 {
299 	return (supported_xcr0 & (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR))
300 		== (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR);
301 }
302 
303 extern unsigned int min_timer_period_us;
304 
305 extern bool enable_vmware_backdoor;
306 
307 extern int pi_inject_timer;
308 
309 extern struct static_key kvm_no_apic_vcpu;
310 
311 extern bool report_ignored_msrs;
312 
313 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
314 {
315 	return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
316 				   vcpu->arch.virtual_tsc_shift);
317 }
318 
319 /* Same "calling convention" as do_div:
320  * - divide (n << 32) by base
321  * - put result in n
322  * - return remainder
323  */
324 #define do_shl32_div32(n, base)					\
325 	({							\
326 	    u32 __quot, __rem;					\
327 	    asm("divl %2" : "=a" (__quot), "=d" (__rem)		\
328 			: "rm" (base), "0" (0), "1" ((u32) n));	\
329 	    n = __quot;						\
330 	    __rem;						\
331 	 })
332 
333 static inline bool kvm_mwait_in_guest(struct kvm *kvm)
334 {
335 	return kvm->arch.mwait_in_guest;
336 }
337 
338 static inline bool kvm_hlt_in_guest(struct kvm *kvm)
339 {
340 	return kvm->arch.hlt_in_guest;
341 }
342 
343 static inline bool kvm_pause_in_guest(struct kvm *kvm)
344 {
345 	return kvm->arch.pause_in_guest;
346 }
347 
348 static inline bool kvm_cstate_in_guest(struct kvm *kvm)
349 {
350 	return kvm->arch.cstate_in_guest;
351 }
352 
353 DECLARE_PER_CPU(struct kvm_vcpu *, current_vcpu);
354 
355 static inline void kvm_before_interrupt(struct kvm_vcpu *vcpu)
356 {
357 	__this_cpu_write(current_vcpu, vcpu);
358 }
359 
360 static inline void kvm_after_interrupt(struct kvm_vcpu *vcpu)
361 {
362 	__this_cpu_write(current_vcpu, NULL);
363 }
364 
365 
366 static inline bool kvm_pat_valid(u64 data)
367 {
368 	if (data & 0xF8F8F8F8F8F8F8F8ull)
369 		return false;
370 	/* 0, 1, 4, 5, 6, 7 are valid values.  */
371 	return (data | ((data & 0x0202020202020202ull) << 1)) == data;
372 }
373 
374 static inline bool kvm_dr7_valid(u64 data)
375 {
376 	/* Bits [63:32] are reserved */
377 	return !(data >> 32);
378 }
379 static inline bool kvm_dr6_valid(u64 data)
380 {
381 	/* Bits [63:32] are reserved */
382 	return !(data >> 32);
383 }
384 
385 /*
386  * Trigger machine check on the host. We assume all the MSRs are already set up
387  * by the CPU and that we still run on the same CPU as the MCE occurred on.
388  * We pass a fake environment to the machine check handler because we want
389  * the guest to be always treated like user space, no matter what context
390  * it used internally.
391  */
392 static inline void kvm_machine_check(void)
393 {
394 #if defined(CONFIG_X86_MCE)
395 	struct pt_regs regs = {
396 		.cs = 3, /* Fake ring 3 no matter what the guest ran on */
397 		.flags = X86_EFLAGS_IF,
398 	};
399 
400 	do_machine_check(&regs);
401 #endif
402 }
403 
404 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu);
405 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu);
406 int kvm_spec_ctrl_test_value(u64 value);
407 bool kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
408 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
409 			      struct x86_exception *e);
410 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva);
411 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type);
412 
413 /*
414  * Internal error codes that are used to indicate that MSR emulation encountered
415  * an error that should result in #GP in the guest, unless userspace
416  * handles it.
417  */
418 #define  KVM_MSR_RET_INVALID	2	/* in-kernel MSR emulation #GP condition */
419 #define  KVM_MSR_RET_FILTERED	3	/* #GP due to userspace MSR filter */
420 
421 #define __cr4_reserved_bits(__cpu_has, __c)             \
422 ({                                                      \
423 	u64 __reserved_bits = CR4_RESERVED_BITS;        \
424                                                         \
425 	if (!__cpu_has(__c, X86_FEATURE_XSAVE))         \
426 		__reserved_bits |= X86_CR4_OSXSAVE;     \
427 	if (!__cpu_has(__c, X86_FEATURE_SMEP))          \
428 		__reserved_bits |= X86_CR4_SMEP;        \
429 	if (!__cpu_has(__c, X86_FEATURE_SMAP))          \
430 		__reserved_bits |= X86_CR4_SMAP;        \
431 	if (!__cpu_has(__c, X86_FEATURE_FSGSBASE))      \
432 		__reserved_bits |= X86_CR4_FSGSBASE;    \
433 	if (!__cpu_has(__c, X86_FEATURE_PKU))           \
434 		__reserved_bits |= X86_CR4_PKE;         \
435 	if (!__cpu_has(__c, X86_FEATURE_LA57))          \
436 		__reserved_bits |= X86_CR4_LA57;        \
437 	if (!__cpu_has(__c, X86_FEATURE_UMIP))          \
438 		__reserved_bits |= X86_CR4_UMIP;        \
439 	if (!__cpu_has(__c, X86_FEATURE_VMX))           \
440 		__reserved_bits |= X86_CR4_VMXE;        \
441 	if (!__cpu_has(__c, X86_FEATURE_PCID))          \
442 		__reserved_bits |= X86_CR4_PCIDE;       \
443 	__reserved_bits;                                \
444 })
445 
446 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t src, unsigned int bytes,
447 			  void *dst);
448 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t src, unsigned int bytes,
449 			 void *dst);
450 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
451 			 unsigned int port, void *data,  unsigned int count,
452 			 int in);
453 
454 #endif
455