xref: /linux/arch/x86/kvm/x86.h (revision fd639726bf15fca8ee1a00dce8e0096d0ad9bd18)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef ARCH_X86_KVM_X86_H
3 #define ARCH_X86_KVM_X86_H
4 
5 #include <asm/processor.h>
6 #include <asm/mwait.h>
7 #include <linux/kvm_host.h>
8 #include <asm/pvclock.h>
9 #include "kvm_cache_regs.h"
10 
11 #define MSR_IA32_CR_PAT_DEFAULT  0x0007040600070406ULL
12 
13 static inline void kvm_clear_exception_queue(struct kvm_vcpu *vcpu)
14 {
15 	vcpu->arch.exception.injected = false;
16 }
17 
18 static inline void kvm_queue_interrupt(struct kvm_vcpu *vcpu, u8 vector,
19 	bool soft)
20 {
21 	vcpu->arch.interrupt.pending = true;
22 	vcpu->arch.interrupt.soft = soft;
23 	vcpu->arch.interrupt.nr = vector;
24 }
25 
26 static inline void kvm_clear_interrupt_queue(struct kvm_vcpu *vcpu)
27 {
28 	vcpu->arch.interrupt.pending = false;
29 }
30 
31 static inline bool kvm_event_needs_reinjection(struct kvm_vcpu *vcpu)
32 {
33 	return vcpu->arch.exception.injected || vcpu->arch.interrupt.pending ||
34 		vcpu->arch.nmi_injected;
35 }
36 
37 static inline bool kvm_exception_is_soft(unsigned int nr)
38 {
39 	return (nr == BP_VECTOR) || (nr == OF_VECTOR);
40 }
41 
42 static inline bool is_protmode(struct kvm_vcpu *vcpu)
43 {
44 	return kvm_read_cr0_bits(vcpu, X86_CR0_PE);
45 }
46 
47 static inline int is_long_mode(struct kvm_vcpu *vcpu)
48 {
49 #ifdef CONFIG_X86_64
50 	return vcpu->arch.efer & EFER_LMA;
51 #else
52 	return 0;
53 #endif
54 }
55 
56 static inline bool is_64_bit_mode(struct kvm_vcpu *vcpu)
57 {
58 	int cs_db, cs_l;
59 
60 	if (!is_long_mode(vcpu))
61 		return false;
62 	kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
63 	return cs_l;
64 }
65 
66 static inline bool is_la57_mode(struct kvm_vcpu *vcpu)
67 {
68 #ifdef CONFIG_X86_64
69 	return (vcpu->arch.efer & EFER_LMA) &&
70 		 kvm_read_cr4_bits(vcpu, X86_CR4_LA57);
71 #else
72 	return 0;
73 #endif
74 }
75 
76 static inline bool mmu_is_nested(struct kvm_vcpu *vcpu)
77 {
78 	return vcpu->arch.walk_mmu == &vcpu->arch.nested_mmu;
79 }
80 
81 static inline int is_pae(struct kvm_vcpu *vcpu)
82 {
83 	return kvm_read_cr4_bits(vcpu, X86_CR4_PAE);
84 }
85 
86 static inline int is_pse(struct kvm_vcpu *vcpu)
87 {
88 	return kvm_read_cr4_bits(vcpu, X86_CR4_PSE);
89 }
90 
91 static inline int is_paging(struct kvm_vcpu *vcpu)
92 {
93 	return likely(kvm_read_cr0_bits(vcpu, X86_CR0_PG));
94 }
95 
96 static inline u32 bit(int bitno)
97 {
98 	return 1 << (bitno & 31);
99 }
100 
101 static inline u8 vcpu_virt_addr_bits(struct kvm_vcpu *vcpu)
102 {
103 	return kvm_read_cr4_bits(vcpu, X86_CR4_LA57) ? 57 : 48;
104 }
105 
106 static inline u8 ctxt_virt_addr_bits(struct x86_emulate_ctxt *ctxt)
107 {
108 	return (ctxt->ops->get_cr(ctxt, 4) & X86_CR4_LA57) ? 57 : 48;
109 }
110 
111 static inline u64 get_canonical(u64 la, u8 vaddr_bits)
112 {
113 	return ((int64_t)la << (64 - vaddr_bits)) >> (64 - vaddr_bits);
114 }
115 
116 static inline bool is_noncanonical_address(u64 la, struct kvm_vcpu *vcpu)
117 {
118 #ifdef CONFIG_X86_64
119 	return get_canonical(la, vcpu_virt_addr_bits(vcpu)) != la;
120 #else
121 	return false;
122 #endif
123 }
124 
125 static inline bool emul_is_noncanonical_address(u64 la,
126 						struct x86_emulate_ctxt *ctxt)
127 {
128 #ifdef CONFIG_X86_64
129 	return get_canonical(la, ctxt_virt_addr_bits(ctxt)) != la;
130 #else
131 	return false;
132 #endif
133 }
134 
135 static inline void vcpu_cache_mmio_info(struct kvm_vcpu *vcpu,
136 					gva_t gva, gfn_t gfn, unsigned access)
137 {
138 	/*
139 	 * If this is a shadow nested page table, the "GVA" is
140 	 * actually a nGPA.
141 	 */
142 	vcpu->arch.mmio_gva = mmu_is_nested(vcpu) ? 0 : gva & PAGE_MASK;
143 	vcpu->arch.access = access;
144 	vcpu->arch.mmio_gfn = gfn;
145 	vcpu->arch.mmio_gen = kvm_memslots(vcpu->kvm)->generation;
146 }
147 
148 static inline bool vcpu_match_mmio_gen(struct kvm_vcpu *vcpu)
149 {
150 	return vcpu->arch.mmio_gen == kvm_memslots(vcpu->kvm)->generation;
151 }
152 
153 /*
154  * Clear the mmio cache info for the given gva. If gva is MMIO_GVA_ANY, we
155  * clear all mmio cache info.
156  */
157 #define MMIO_GVA_ANY (~(gva_t)0)
158 
159 static inline void vcpu_clear_mmio_info(struct kvm_vcpu *vcpu, gva_t gva)
160 {
161 	if (gva != MMIO_GVA_ANY && vcpu->arch.mmio_gva != (gva & PAGE_MASK))
162 		return;
163 
164 	vcpu->arch.mmio_gva = 0;
165 }
166 
167 static inline bool vcpu_match_mmio_gva(struct kvm_vcpu *vcpu, unsigned long gva)
168 {
169 	if (vcpu_match_mmio_gen(vcpu) && vcpu->arch.mmio_gva &&
170 	      vcpu->arch.mmio_gva == (gva & PAGE_MASK))
171 		return true;
172 
173 	return false;
174 }
175 
176 static inline bool vcpu_match_mmio_gpa(struct kvm_vcpu *vcpu, gpa_t gpa)
177 {
178 	if (vcpu_match_mmio_gen(vcpu) && vcpu->arch.mmio_gfn &&
179 	      vcpu->arch.mmio_gfn == gpa >> PAGE_SHIFT)
180 		return true;
181 
182 	return false;
183 }
184 
185 static inline unsigned long kvm_register_readl(struct kvm_vcpu *vcpu,
186 					       enum kvm_reg reg)
187 {
188 	unsigned long val = kvm_register_read(vcpu, reg);
189 
190 	return is_64_bit_mode(vcpu) ? val : (u32)val;
191 }
192 
193 static inline void kvm_register_writel(struct kvm_vcpu *vcpu,
194 				       enum kvm_reg reg,
195 				       unsigned long val)
196 {
197 	if (!is_64_bit_mode(vcpu))
198 		val = (u32)val;
199 	return kvm_register_write(vcpu, reg, val);
200 }
201 
202 static inline bool kvm_check_has_quirk(struct kvm *kvm, u64 quirk)
203 {
204 	return !(kvm->arch.disabled_quirks & quirk);
205 }
206 
207 void kvm_before_handle_nmi(struct kvm_vcpu *vcpu);
208 void kvm_after_handle_nmi(struct kvm_vcpu *vcpu);
209 void kvm_set_pending_timer(struct kvm_vcpu *vcpu);
210 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip);
211 
212 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr);
213 u64 get_kvmclock_ns(struct kvm *kvm);
214 
215 int kvm_read_guest_virt(struct x86_emulate_ctxt *ctxt,
216 	gva_t addr, void *val, unsigned int bytes,
217 	struct x86_exception *exception);
218 
219 int kvm_write_guest_virt_system(struct x86_emulate_ctxt *ctxt,
220 	gva_t addr, void *val, unsigned int bytes,
221 	struct x86_exception *exception);
222 
223 void kvm_vcpu_mtrr_init(struct kvm_vcpu *vcpu);
224 u8 kvm_mtrr_get_guest_memory_type(struct kvm_vcpu *vcpu, gfn_t gfn);
225 bool kvm_mtrr_valid(struct kvm_vcpu *vcpu, u32 msr, u64 data);
226 int kvm_mtrr_set_msr(struct kvm_vcpu *vcpu, u32 msr, u64 data);
227 int kvm_mtrr_get_msr(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata);
228 bool kvm_mtrr_check_gfn_range_consistency(struct kvm_vcpu *vcpu, gfn_t gfn,
229 					  int page_num);
230 bool kvm_vector_hashing_enabled(void);
231 
232 #define KVM_SUPPORTED_XCR0     (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
233 				| XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
234 				| XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
235 				| XFEATURE_MASK_PKRU)
236 extern u64 host_xcr0;
237 
238 extern u64 kvm_supported_xcr0(void);
239 
240 extern unsigned int min_timer_period_us;
241 
242 extern unsigned int lapic_timer_advance_ns;
243 
244 extern struct static_key kvm_no_apic_vcpu;
245 
246 static inline u64 nsec_to_cycles(struct kvm_vcpu *vcpu, u64 nsec)
247 {
248 	return pvclock_scale_delta(nsec, vcpu->arch.virtual_tsc_mult,
249 				   vcpu->arch.virtual_tsc_shift);
250 }
251 
252 /* Same "calling convention" as do_div:
253  * - divide (n << 32) by base
254  * - put result in n
255  * - return remainder
256  */
257 #define do_shl32_div32(n, base)					\
258 	({							\
259 	    u32 __quot, __rem;					\
260 	    asm("divl %2" : "=a" (__quot), "=d" (__rem)		\
261 			: "rm" (base), "0" (0), "1" ((u32) n));	\
262 	    n = __quot;						\
263 	    __rem;						\
264 	 })
265 
266 static inline bool kvm_mwait_in_guest(void)
267 {
268 	unsigned int eax, ebx, ecx, edx;
269 
270 	if (!cpu_has(&boot_cpu_data, X86_FEATURE_MWAIT))
271 		return false;
272 
273 	switch (boot_cpu_data.x86_vendor) {
274 	case X86_VENDOR_AMD:
275 		/* All AMD CPUs have a working MWAIT implementation */
276 		return true;
277 	case X86_VENDOR_INTEL:
278 		/* Handle Intel below */
279 		break;
280 	default:
281 		return false;
282 	}
283 
284 	/*
285 	 * Intel CPUs without CPUID5_ECX_INTERRUPT_BREAK are problematic as
286 	 * they would allow guest to stop the CPU completely by disabling
287 	 * interrupts then invoking MWAIT.
288 	 */
289 	if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
290 		return false;
291 
292 	cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &edx);
293 
294 	if (!(ecx & CPUID5_ECX_INTERRUPT_BREAK))
295 		return false;
296 
297 	return true;
298 }
299 
300 #endif
301