xref: /linux/tools/testing/selftests/kvm/include/x86_64/processor.h (revision 24168c5e6dfbdd5b414f048f47f75d64533296ca)
1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /*
3  * tools/testing/selftests/kvm/include/x86_64/processor.h
4  *
5  * Copyright (C) 2018, Google LLC.
6  */
7 
8 #ifndef SELFTEST_KVM_PROCESSOR_H
9 #define SELFTEST_KVM_PROCESSOR_H
10 
11 #include <assert.h>
12 #include <stdint.h>
13 #include <syscall.h>
14 
15 #include <asm/msr-index.h>
16 #include <asm/prctl.h>
17 
18 #include <linux/kvm_para.h>
19 #include <linux/stringify.h>
20 
21 #include "kvm_util.h"
22 #include "ucall_common.h"
23 
24 extern bool host_cpu_is_intel;
25 extern bool host_cpu_is_amd;
26 
27 /* Forced emulation prefix, used to invoke the emulator unconditionally. */
28 #define KVM_FEP "ud2; .byte 'k', 'v', 'm';"
29 
30 #define NMI_VECTOR		0x02
31 
32 #define X86_EFLAGS_FIXED	 (1u << 1)
33 
34 #define X86_CR4_VME		(1ul << 0)
35 #define X86_CR4_PVI		(1ul << 1)
36 #define X86_CR4_TSD		(1ul << 2)
37 #define X86_CR4_DE		(1ul << 3)
38 #define X86_CR4_PSE		(1ul << 4)
39 #define X86_CR4_PAE		(1ul << 5)
40 #define X86_CR4_MCE		(1ul << 6)
41 #define X86_CR4_PGE		(1ul << 7)
42 #define X86_CR4_PCE		(1ul << 8)
43 #define X86_CR4_OSFXSR		(1ul << 9)
44 #define X86_CR4_OSXMMEXCPT	(1ul << 10)
45 #define X86_CR4_UMIP		(1ul << 11)
46 #define X86_CR4_LA57		(1ul << 12)
47 #define X86_CR4_VMXE		(1ul << 13)
48 #define X86_CR4_SMXE		(1ul << 14)
49 #define X86_CR4_FSGSBASE	(1ul << 16)
50 #define X86_CR4_PCIDE		(1ul << 17)
51 #define X86_CR4_OSXSAVE		(1ul << 18)
52 #define X86_CR4_SMEP		(1ul << 20)
53 #define X86_CR4_SMAP		(1ul << 21)
54 #define X86_CR4_PKE		(1ul << 22)
55 
56 struct xstate_header {
57 	u64				xstate_bv;
58 	u64				xcomp_bv;
59 	u64				reserved[6];
60 } __attribute__((packed));
61 
62 struct xstate {
63 	u8				i387[512];
64 	struct xstate_header		header;
65 	u8				extended_state_area[0];
66 } __attribute__ ((packed, aligned (64)));
67 
68 #define XFEATURE_MASK_FP		BIT_ULL(0)
69 #define XFEATURE_MASK_SSE		BIT_ULL(1)
70 #define XFEATURE_MASK_YMM		BIT_ULL(2)
71 #define XFEATURE_MASK_BNDREGS		BIT_ULL(3)
72 #define XFEATURE_MASK_BNDCSR		BIT_ULL(4)
73 #define XFEATURE_MASK_OPMASK		BIT_ULL(5)
74 #define XFEATURE_MASK_ZMM_Hi256		BIT_ULL(6)
75 #define XFEATURE_MASK_Hi16_ZMM		BIT_ULL(7)
76 #define XFEATURE_MASK_PT		BIT_ULL(8)
77 #define XFEATURE_MASK_PKRU		BIT_ULL(9)
78 #define XFEATURE_MASK_PASID		BIT_ULL(10)
79 #define XFEATURE_MASK_CET_USER		BIT_ULL(11)
80 #define XFEATURE_MASK_CET_KERNEL	BIT_ULL(12)
81 #define XFEATURE_MASK_LBR		BIT_ULL(15)
82 #define XFEATURE_MASK_XTILE_CFG		BIT_ULL(17)
83 #define XFEATURE_MASK_XTILE_DATA	BIT_ULL(18)
84 
85 #define XFEATURE_MASK_AVX512		(XFEATURE_MASK_OPMASK | \
86 					 XFEATURE_MASK_ZMM_Hi256 | \
87 					 XFEATURE_MASK_Hi16_ZMM)
88 #define XFEATURE_MASK_XTILE		(XFEATURE_MASK_XTILE_DATA | \
89 					 XFEATURE_MASK_XTILE_CFG)
90 
91 /* Note, these are ordered alphabetically to match kvm_cpuid_entry2.  Eww. */
92 enum cpuid_output_regs {
93 	KVM_CPUID_EAX,
94 	KVM_CPUID_EBX,
95 	KVM_CPUID_ECX,
96 	KVM_CPUID_EDX
97 };
98 
99 /*
100  * Pack the information into a 64-bit value so that each X86_FEATURE_XXX can be
101  * passed by value with no overhead.
102  */
103 struct kvm_x86_cpu_feature {
104 	u32	function;
105 	u16	index;
106 	u8	reg;
107 	u8	bit;
108 };
109 #define	KVM_X86_CPU_FEATURE(fn, idx, gpr, __bit)				\
110 ({										\
111 	struct kvm_x86_cpu_feature feature = {					\
112 		.function = fn,							\
113 		.index = idx,							\
114 		.reg = KVM_CPUID_##gpr,						\
115 		.bit = __bit,							\
116 	};									\
117 										\
118 	kvm_static_assert((fn & 0xc0000000) == 0 ||				\
119 			  (fn & 0xc0000000) == 0x40000000 ||			\
120 			  (fn & 0xc0000000) == 0x80000000 ||			\
121 			  (fn & 0xc0000000) == 0xc0000000);			\
122 	kvm_static_assert(idx < BIT(sizeof(feature.index) * BITS_PER_BYTE));	\
123 	feature;								\
124 })
125 
126 /*
127  * Basic Leafs, a.k.a. Intel defined
128  */
129 #define	X86_FEATURE_MWAIT		KVM_X86_CPU_FEATURE(0x1, 0, ECX, 3)
130 #define	X86_FEATURE_VMX			KVM_X86_CPU_FEATURE(0x1, 0, ECX, 5)
131 #define	X86_FEATURE_SMX			KVM_X86_CPU_FEATURE(0x1, 0, ECX, 6)
132 #define	X86_FEATURE_PDCM		KVM_X86_CPU_FEATURE(0x1, 0, ECX, 15)
133 #define	X86_FEATURE_PCID		KVM_X86_CPU_FEATURE(0x1, 0, ECX, 17)
134 #define X86_FEATURE_X2APIC		KVM_X86_CPU_FEATURE(0x1, 0, ECX, 21)
135 #define	X86_FEATURE_MOVBE		KVM_X86_CPU_FEATURE(0x1, 0, ECX, 22)
136 #define	X86_FEATURE_TSC_DEADLINE_TIMER	KVM_X86_CPU_FEATURE(0x1, 0, ECX, 24)
137 #define	X86_FEATURE_XSAVE		KVM_X86_CPU_FEATURE(0x1, 0, ECX, 26)
138 #define	X86_FEATURE_OSXSAVE		KVM_X86_CPU_FEATURE(0x1, 0, ECX, 27)
139 #define	X86_FEATURE_RDRAND		KVM_X86_CPU_FEATURE(0x1, 0, ECX, 30)
140 #define	X86_FEATURE_HYPERVISOR		KVM_X86_CPU_FEATURE(0x1, 0, ECX, 31)
141 #define X86_FEATURE_PAE			KVM_X86_CPU_FEATURE(0x1, 0, EDX, 6)
142 #define	X86_FEATURE_MCE			KVM_X86_CPU_FEATURE(0x1, 0, EDX, 7)
143 #define	X86_FEATURE_APIC		KVM_X86_CPU_FEATURE(0x1, 0, EDX, 9)
144 #define	X86_FEATURE_CLFLUSH		KVM_X86_CPU_FEATURE(0x1, 0, EDX, 19)
145 #define	X86_FEATURE_XMM			KVM_X86_CPU_FEATURE(0x1, 0, EDX, 25)
146 #define	X86_FEATURE_XMM2		KVM_X86_CPU_FEATURE(0x1, 0, EDX, 26)
147 #define	X86_FEATURE_FSGSBASE		KVM_X86_CPU_FEATURE(0x7, 0, EBX, 0)
148 #define	X86_FEATURE_TSC_ADJUST		KVM_X86_CPU_FEATURE(0x7, 0, EBX, 1)
149 #define	X86_FEATURE_SGX			KVM_X86_CPU_FEATURE(0x7, 0, EBX, 2)
150 #define	X86_FEATURE_HLE			KVM_X86_CPU_FEATURE(0x7, 0, EBX, 4)
151 #define	X86_FEATURE_SMEP	        KVM_X86_CPU_FEATURE(0x7, 0, EBX, 7)
152 #define	X86_FEATURE_INVPCID		KVM_X86_CPU_FEATURE(0x7, 0, EBX, 10)
153 #define	X86_FEATURE_RTM			KVM_X86_CPU_FEATURE(0x7, 0, EBX, 11)
154 #define	X86_FEATURE_MPX			KVM_X86_CPU_FEATURE(0x7, 0, EBX, 14)
155 #define	X86_FEATURE_SMAP		KVM_X86_CPU_FEATURE(0x7, 0, EBX, 20)
156 #define	X86_FEATURE_PCOMMIT		KVM_X86_CPU_FEATURE(0x7, 0, EBX, 22)
157 #define	X86_FEATURE_CLFLUSHOPT		KVM_X86_CPU_FEATURE(0x7, 0, EBX, 23)
158 #define	X86_FEATURE_CLWB		KVM_X86_CPU_FEATURE(0x7, 0, EBX, 24)
159 #define	X86_FEATURE_UMIP		KVM_X86_CPU_FEATURE(0x7, 0, ECX, 2)
160 #define	X86_FEATURE_PKU			KVM_X86_CPU_FEATURE(0x7, 0, ECX, 3)
161 #define	X86_FEATURE_OSPKE		KVM_X86_CPU_FEATURE(0x7, 0, ECX, 4)
162 #define	X86_FEATURE_LA57		KVM_X86_CPU_FEATURE(0x7, 0, ECX, 16)
163 #define	X86_FEATURE_RDPID		KVM_X86_CPU_FEATURE(0x7, 0, ECX, 22)
164 #define	X86_FEATURE_SGX_LC		KVM_X86_CPU_FEATURE(0x7, 0, ECX, 30)
165 #define	X86_FEATURE_SHSTK		KVM_X86_CPU_FEATURE(0x7, 0, ECX, 7)
166 #define	X86_FEATURE_IBT			KVM_X86_CPU_FEATURE(0x7, 0, EDX, 20)
167 #define	X86_FEATURE_AMX_TILE		KVM_X86_CPU_FEATURE(0x7, 0, EDX, 24)
168 #define	X86_FEATURE_SPEC_CTRL		KVM_X86_CPU_FEATURE(0x7, 0, EDX, 26)
169 #define	X86_FEATURE_ARCH_CAPABILITIES	KVM_X86_CPU_FEATURE(0x7, 0, EDX, 29)
170 #define	X86_FEATURE_PKS			KVM_X86_CPU_FEATURE(0x7, 0, ECX, 31)
171 #define	X86_FEATURE_XTILECFG		KVM_X86_CPU_FEATURE(0xD, 0, EAX, 17)
172 #define	X86_FEATURE_XTILEDATA		KVM_X86_CPU_FEATURE(0xD, 0, EAX, 18)
173 #define	X86_FEATURE_XSAVES		KVM_X86_CPU_FEATURE(0xD, 1, EAX, 3)
174 #define	X86_FEATURE_XFD			KVM_X86_CPU_FEATURE(0xD, 1, EAX, 4)
175 #define X86_FEATURE_XTILEDATA_XFD	KVM_X86_CPU_FEATURE(0xD, 18, ECX, 2)
176 
177 /*
178  * Extended Leafs, a.k.a. AMD defined
179  */
180 #define	X86_FEATURE_SVM			KVM_X86_CPU_FEATURE(0x80000001, 0, ECX, 2)
181 #define	X86_FEATURE_NX			KVM_X86_CPU_FEATURE(0x80000001, 0, EDX, 20)
182 #define	X86_FEATURE_GBPAGES		KVM_X86_CPU_FEATURE(0x80000001, 0, EDX, 26)
183 #define	X86_FEATURE_RDTSCP		KVM_X86_CPU_FEATURE(0x80000001, 0, EDX, 27)
184 #define	X86_FEATURE_LM			KVM_X86_CPU_FEATURE(0x80000001, 0, EDX, 29)
185 #define	X86_FEATURE_INVTSC		KVM_X86_CPU_FEATURE(0x80000007, 0, EDX, 8)
186 #define	X86_FEATURE_RDPRU		KVM_X86_CPU_FEATURE(0x80000008, 0, EBX, 4)
187 #define	X86_FEATURE_AMD_IBPB		KVM_X86_CPU_FEATURE(0x80000008, 0, EBX, 12)
188 #define	X86_FEATURE_NPT			KVM_X86_CPU_FEATURE(0x8000000A, 0, EDX, 0)
189 #define	X86_FEATURE_LBRV		KVM_X86_CPU_FEATURE(0x8000000A, 0, EDX, 1)
190 #define	X86_FEATURE_NRIPS		KVM_X86_CPU_FEATURE(0x8000000A, 0, EDX, 3)
191 #define X86_FEATURE_TSCRATEMSR          KVM_X86_CPU_FEATURE(0x8000000A, 0, EDX, 4)
192 #define X86_FEATURE_PAUSEFILTER         KVM_X86_CPU_FEATURE(0x8000000A, 0, EDX, 10)
193 #define X86_FEATURE_PFTHRESHOLD         KVM_X86_CPU_FEATURE(0x8000000A, 0, EDX, 12)
194 #define	X86_FEATURE_VGIF		KVM_X86_CPU_FEATURE(0x8000000A, 0, EDX, 16)
195 #define X86_FEATURE_SEV			KVM_X86_CPU_FEATURE(0x8000001F, 0, EAX, 1)
196 #define X86_FEATURE_SEV_ES		KVM_X86_CPU_FEATURE(0x8000001F, 0, EAX, 3)
197 
198 /*
199  * KVM defined paravirt features.
200  */
201 #define X86_FEATURE_KVM_CLOCKSOURCE	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 0)
202 #define X86_FEATURE_KVM_NOP_IO_DELAY	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 1)
203 #define X86_FEATURE_KVM_MMU_OP		KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 2)
204 #define X86_FEATURE_KVM_CLOCKSOURCE2	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 3)
205 #define X86_FEATURE_KVM_ASYNC_PF	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 4)
206 #define X86_FEATURE_KVM_STEAL_TIME	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 5)
207 #define X86_FEATURE_KVM_PV_EOI		KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 6)
208 #define X86_FEATURE_KVM_PV_UNHALT	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 7)
209 /* Bit 8 apparently isn't used?!?! */
210 #define X86_FEATURE_KVM_PV_TLB_FLUSH	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 9)
211 #define X86_FEATURE_KVM_ASYNC_PF_VMEXIT	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 10)
212 #define X86_FEATURE_KVM_PV_SEND_IPI	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 11)
213 #define X86_FEATURE_KVM_POLL_CONTROL	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 12)
214 #define X86_FEATURE_KVM_PV_SCHED_YIELD	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 13)
215 #define X86_FEATURE_KVM_ASYNC_PF_INT	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 14)
216 #define X86_FEATURE_KVM_MSI_EXT_DEST_ID	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 15)
217 #define X86_FEATURE_KVM_HC_MAP_GPA_RANGE	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 16)
218 #define X86_FEATURE_KVM_MIGRATION_CONTROL	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 17)
219 
220 /*
221  * Same idea as X86_FEATURE_XXX, but X86_PROPERTY_XXX retrieves a multi-bit
222  * value/property as opposed to a single-bit feature.  Again, pack the info
223  * into a 64-bit value to pass by value with no overhead.
224  */
225 struct kvm_x86_cpu_property {
226 	u32	function;
227 	u8	index;
228 	u8	reg;
229 	u8	lo_bit;
230 	u8	hi_bit;
231 };
232 #define	KVM_X86_CPU_PROPERTY(fn, idx, gpr, low_bit, high_bit)			\
233 ({										\
234 	struct kvm_x86_cpu_property property = {				\
235 		.function = fn,							\
236 		.index = idx,							\
237 		.reg = KVM_CPUID_##gpr,						\
238 		.lo_bit = low_bit,						\
239 		.hi_bit = high_bit,						\
240 	};									\
241 										\
242 	kvm_static_assert(low_bit < high_bit);					\
243 	kvm_static_assert((fn & 0xc0000000) == 0 ||				\
244 			  (fn & 0xc0000000) == 0x40000000 ||			\
245 			  (fn & 0xc0000000) == 0x80000000 ||			\
246 			  (fn & 0xc0000000) == 0xc0000000);			\
247 	kvm_static_assert(idx < BIT(sizeof(property.index) * BITS_PER_BYTE));	\
248 	property;								\
249 })
250 
251 #define X86_PROPERTY_MAX_BASIC_LEAF		KVM_X86_CPU_PROPERTY(0, 0, EAX, 0, 31)
252 #define X86_PROPERTY_PMU_VERSION		KVM_X86_CPU_PROPERTY(0xa, 0, EAX, 0, 7)
253 #define X86_PROPERTY_PMU_NR_GP_COUNTERS		KVM_X86_CPU_PROPERTY(0xa, 0, EAX, 8, 15)
254 #define X86_PROPERTY_PMU_GP_COUNTERS_BIT_WIDTH	KVM_X86_CPU_PROPERTY(0xa, 0, EAX, 16, 23)
255 #define X86_PROPERTY_PMU_EBX_BIT_VECTOR_LENGTH	KVM_X86_CPU_PROPERTY(0xa, 0, EAX, 24, 31)
256 #define X86_PROPERTY_PMU_EVENTS_MASK		KVM_X86_CPU_PROPERTY(0xa, 0, EBX, 0, 7)
257 #define X86_PROPERTY_PMU_FIXED_COUNTERS_BITMASK	KVM_X86_CPU_PROPERTY(0xa, 0, ECX, 0, 31)
258 #define X86_PROPERTY_PMU_NR_FIXED_COUNTERS	KVM_X86_CPU_PROPERTY(0xa, 0, EDX, 0, 4)
259 #define X86_PROPERTY_PMU_FIXED_COUNTERS_BIT_WIDTH	KVM_X86_CPU_PROPERTY(0xa, 0, EDX, 5, 12)
260 
261 #define X86_PROPERTY_SUPPORTED_XCR0_LO		KVM_X86_CPU_PROPERTY(0xd,  0, EAX,  0, 31)
262 #define X86_PROPERTY_XSTATE_MAX_SIZE_XCR0	KVM_X86_CPU_PROPERTY(0xd,  0, EBX,  0, 31)
263 #define X86_PROPERTY_XSTATE_MAX_SIZE		KVM_X86_CPU_PROPERTY(0xd,  0, ECX,  0, 31)
264 #define X86_PROPERTY_SUPPORTED_XCR0_HI		KVM_X86_CPU_PROPERTY(0xd,  0, EDX,  0, 31)
265 
266 #define X86_PROPERTY_XSTATE_TILE_SIZE		KVM_X86_CPU_PROPERTY(0xd, 18, EAX,  0, 31)
267 #define X86_PROPERTY_XSTATE_TILE_OFFSET		KVM_X86_CPU_PROPERTY(0xd, 18, EBX,  0, 31)
268 #define X86_PROPERTY_AMX_MAX_PALETTE_TABLES	KVM_X86_CPU_PROPERTY(0x1d, 0, EAX,  0, 31)
269 #define X86_PROPERTY_AMX_TOTAL_TILE_BYTES	KVM_X86_CPU_PROPERTY(0x1d, 1, EAX,  0, 15)
270 #define X86_PROPERTY_AMX_BYTES_PER_TILE		KVM_X86_CPU_PROPERTY(0x1d, 1, EAX, 16, 31)
271 #define X86_PROPERTY_AMX_BYTES_PER_ROW		KVM_X86_CPU_PROPERTY(0x1d, 1, EBX, 0,  15)
272 #define X86_PROPERTY_AMX_NR_TILE_REGS		KVM_X86_CPU_PROPERTY(0x1d, 1, EBX, 16, 31)
273 #define X86_PROPERTY_AMX_MAX_ROWS		KVM_X86_CPU_PROPERTY(0x1d, 1, ECX, 0,  15)
274 
275 #define X86_PROPERTY_MAX_KVM_LEAF		KVM_X86_CPU_PROPERTY(0x40000000, 0, EAX, 0, 31)
276 
277 #define X86_PROPERTY_MAX_EXT_LEAF		KVM_X86_CPU_PROPERTY(0x80000000, 0, EAX, 0, 31)
278 #define X86_PROPERTY_MAX_PHY_ADDR		KVM_X86_CPU_PROPERTY(0x80000008, 0, EAX, 0, 7)
279 #define X86_PROPERTY_MAX_VIRT_ADDR		KVM_X86_CPU_PROPERTY(0x80000008, 0, EAX, 8, 15)
280 #define X86_PROPERTY_SEV_C_BIT			KVM_X86_CPU_PROPERTY(0x8000001F, 0, EBX, 0, 5)
281 #define X86_PROPERTY_PHYS_ADDR_REDUCTION	KVM_X86_CPU_PROPERTY(0x8000001F, 0, EBX, 6, 11)
282 
283 #define X86_PROPERTY_MAX_CENTAUR_LEAF		KVM_X86_CPU_PROPERTY(0xC0000000, 0, EAX, 0, 31)
284 
285 /*
286  * Intel's architectural PMU events are bizarre.  They have a "feature" bit
287  * that indicates the feature is _not_ supported, and a property that states
288  * the length of the bit mask of unsupported features.  A feature is supported
289  * if the size of the bit mask is larger than the "unavailable" bit, and said
290  * bit is not set.  Fixed counters also bizarre enumeration, but inverted from
291  * arch events for general purpose counters.  Fixed counters are supported if a
292  * feature flag is set **OR** the total number of fixed counters is greater
293  * than index of the counter.
294  *
295  * Wrap the events for general purpose and fixed counters to simplify checking
296  * whether or not a given architectural event is supported.
297  */
298 struct kvm_x86_pmu_feature {
299 	struct kvm_x86_cpu_feature f;
300 };
301 #define	KVM_X86_PMU_FEATURE(__reg, __bit)				\
302 ({									\
303 	struct kvm_x86_pmu_feature feature = {				\
304 		.f = KVM_X86_CPU_FEATURE(0xa, 0, __reg, __bit),		\
305 	};								\
306 									\
307 	kvm_static_assert(KVM_CPUID_##__reg == KVM_CPUID_EBX ||		\
308 			  KVM_CPUID_##__reg == KVM_CPUID_ECX);		\
309 	feature;							\
310 })
311 
312 #define X86_PMU_FEATURE_CPU_CYCLES			KVM_X86_PMU_FEATURE(EBX, 0)
313 #define X86_PMU_FEATURE_INSNS_RETIRED			KVM_X86_PMU_FEATURE(EBX, 1)
314 #define X86_PMU_FEATURE_REFERENCE_CYCLES		KVM_X86_PMU_FEATURE(EBX, 2)
315 #define X86_PMU_FEATURE_LLC_REFERENCES			KVM_X86_PMU_FEATURE(EBX, 3)
316 #define X86_PMU_FEATURE_LLC_MISSES			KVM_X86_PMU_FEATURE(EBX, 4)
317 #define X86_PMU_FEATURE_BRANCH_INSNS_RETIRED		KVM_X86_PMU_FEATURE(EBX, 5)
318 #define X86_PMU_FEATURE_BRANCHES_MISPREDICTED		KVM_X86_PMU_FEATURE(EBX, 6)
319 #define X86_PMU_FEATURE_TOPDOWN_SLOTS			KVM_X86_PMU_FEATURE(EBX, 7)
320 
321 #define X86_PMU_FEATURE_INSNS_RETIRED_FIXED		KVM_X86_PMU_FEATURE(ECX, 0)
322 #define X86_PMU_FEATURE_CPU_CYCLES_FIXED		KVM_X86_PMU_FEATURE(ECX, 1)
323 #define X86_PMU_FEATURE_REFERENCE_TSC_CYCLES_FIXED	KVM_X86_PMU_FEATURE(ECX, 2)
324 #define X86_PMU_FEATURE_TOPDOWN_SLOTS_FIXED		KVM_X86_PMU_FEATURE(ECX, 3)
325 
326 static inline unsigned int x86_family(unsigned int eax)
327 {
328 	unsigned int x86;
329 
330 	x86 = (eax >> 8) & 0xf;
331 
332 	if (x86 == 0xf)
333 		x86 += (eax >> 20) & 0xff;
334 
335 	return x86;
336 }
337 
338 static inline unsigned int x86_model(unsigned int eax)
339 {
340 	return ((eax >> 12) & 0xf0) | ((eax >> 4) & 0x0f);
341 }
342 
343 /* Page table bitfield declarations */
344 #define PTE_PRESENT_MASK        BIT_ULL(0)
345 #define PTE_WRITABLE_MASK       BIT_ULL(1)
346 #define PTE_USER_MASK           BIT_ULL(2)
347 #define PTE_ACCESSED_MASK       BIT_ULL(5)
348 #define PTE_DIRTY_MASK          BIT_ULL(6)
349 #define PTE_LARGE_MASK          BIT_ULL(7)
350 #define PTE_GLOBAL_MASK         BIT_ULL(8)
351 #define PTE_NX_MASK             BIT_ULL(63)
352 
353 #define PHYSICAL_PAGE_MASK      GENMASK_ULL(51, 12)
354 
355 #define PAGE_SHIFT		12
356 #define PAGE_SIZE		(1ULL << PAGE_SHIFT)
357 #define PAGE_MASK		(~(PAGE_SIZE-1) & PHYSICAL_PAGE_MASK)
358 
359 #define HUGEPAGE_SHIFT(x)	(PAGE_SHIFT + (((x) - 1) * 9))
360 #define HUGEPAGE_SIZE(x)	(1UL << HUGEPAGE_SHIFT(x))
361 #define HUGEPAGE_MASK(x)	(~(HUGEPAGE_SIZE(x) - 1) & PHYSICAL_PAGE_MASK)
362 
363 #define PTE_GET_PA(pte)		((pte) & PHYSICAL_PAGE_MASK)
364 #define PTE_GET_PFN(pte)        (PTE_GET_PA(pte) >> PAGE_SHIFT)
365 
366 /* General Registers in 64-Bit Mode */
367 struct gpr64_regs {
368 	u64 rax;
369 	u64 rcx;
370 	u64 rdx;
371 	u64 rbx;
372 	u64 rsp;
373 	u64 rbp;
374 	u64 rsi;
375 	u64 rdi;
376 	u64 r8;
377 	u64 r9;
378 	u64 r10;
379 	u64 r11;
380 	u64 r12;
381 	u64 r13;
382 	u64 r14;
383 	u64 r15;
384 };
385 
386 struct desc64 {
387 	uint16_t limit0;
388 	uint16_t base0;
389 	unsigned base1:8, type:4, s:1, dpl:2, p:1;
390 	unsigned limit1:4, avl:1, l:1, db:1, g:1, base2:8;
391 	uint32_t base3;
392 	uint32_t zero1;
393 } __attribute__((packed));
394 
395 struct desc_ptr {
396 	uint16_t size;
397 	uint64_t address;
398 } __attribute__((packed));
399 
400 struct kvm_x86_state {
401 	struct kvm_xsave *xsave;
402 	struct kvm_vcpu_events events;
403 	struct kvm_mp_state mp_state;
404 	struct kvm_regs regs;
405 	struct kvm_xcrs xcrs;
406 	struct kvm_sregs sregs;
407 	struct kvm_debugregs debugregs;
408 	union {
409 		struct kvm_nested_state nested;
410 		char nested_[16384];
411 	};
412 	struct kvm_msrs msrs;
413 };
414 
415 static inline uint64_t get_desc64_base(const struct desc64 *desc)
416 {
417 	return ((uint64_t)desc->base3 << 32) |
418 		(desc->base0 | ((desc->base1) << 16) | ((desc->base2) << 24));
419 }
420 
421 static inline uint64_t rdtsc(void)
422 {
423 	uint32_t eax, edx;
424 	uint64_t tsc_val;
425 	/*
426 	 * The lfence is to wait (on Intel CPUs) until all previous
427 	 * instructions have been executed. If software requires RDTSC to be
428 	 * executed prior to execution of any subsequent instruction, it can
429 	 * execute LFENCE immediately after RDTSC
430 	 */
431 	__asm__ __volatile__("lfence; rdtsc; lfence" : "=a"(eax), "=d"(edx));
432 	tsc_val = ((uint64_t)edx) << 32 | eax;
433 	return tsc_val;
434 }
435 
436 static inline uint64_t rdtscp(uint32_t *aux)
437 {
438 	uint32_t eax, edx;
439 
440 	__asm__ __volatile__("rdtscp" : "=a"(eax), "=d"(edx), "=c"(*aux));
441 	return ((uint64_t)edx) << 32 | eax;
442 }
443 
444 static inline uint64_t rdmsr(uint32_t msr)
445 {
446 	uint32_t a, d;
447 
448 	__asm__ __volatile__("rdmsr" : "=a"(a), "=d"(d) : "c"(msr) : "memory");
449 
450 	return a | ((uint64_t) d << 32);
451 }
452 
453 static inline void wrmsr(uint32_t msr, uint64_t value)
454 {
455 	uint32_t a = value;
456 	uint32_t d = value >> 32;
457 
458 	__asm__ __volatile__("wrmsr" :: "a"(a), "d"(d), "c"(msr) : "memory");
459 }
460 
461 
462 static inline uint16_t inw(uint16_t port)
463 {
464 	uint16_t tmp;
465 
466 	__asm__ __volatile__("in %%dx, %%ax"
467 		: /* output */ "=a" (tmp)
468 		: /* input */ "d" (port));
469 
470 	return tmp;
471 }
472 
473 static inline uint16_t get_es(void)
474 {
475 	uint16_t es;
476 
477 	__asm__ __volatile__("mov %%es, %[es]"
478 			     : /* output */ [es]"=rm"(es));
479 	return es;
480 }
481 
482 static inline uint16_t get_cs(void)
483 {
484 	uint16_t cs;
485 
486 	__asm__ __volatile__("mov %%cs, %[cs]"
487 			     : /* output */ [cs]"=rm"(cs));
488 	return cs;
489 }
490 
491 static inline uint16_t get_ss(void)
492 {
493 	uint16_t ss;
494 
495 	__asm__ __volatile__("mov %%ss, %[ss]"
496 			     : /* output */ [ss]"=rm"(ss));
497 	return ss;
498 }
499 
500 static inline uint16_t get_ds(void)
501 {
502 	uint16_t ds;
503 
504 	__asm__ __volatile__("mov %%ds, %[ds]"
505 			     : /* output */ [ds]"=rm"(ds));
506 	return ds;
507 }
508 
509 static inline uint16_t get_fs(void)
510 {
511 	uint16_t fs;
512 
513 	__asm__ __volatile__("mov %%fs, %[fs]"
514 			     : /* output */ [fs]"=rm"(fs));
515 	return fs;
516 }
517 
518 static inline uint16_t get_gs(void)
519 {
520 	uint16_t gs;
521 
522 	__asm__ __volatile__("mov %%gs, %[gs]"
523 			     : /* output */ [gs]"=rm"(gs));
524 	return gs;
525 }
526 
527 static inline uint16_t get_tr(void)
528 {
529 	uint16_t tr;
530 
531 	__asm__ __volatile__("str %[tr]"
532 			     : /* output */ [tr]"=rm"(tr));
533 	return tr;
534 }
535 
536 static inline uint64_t get_cr0(void)
537 {
538 	uint64_t cr0;
539 
540 	__asm__ __volatile__("mov %%cr0, %[cr0]"
541 			     : /* output */ [cr0]"=r"(cr0));
542 	return cr0;
543 }
544 
545 static inline uint64_t get_cr3(void)
546 {
547 	uint64_t cr3;
548 
549 	__asm__ __volatile__("mov %%cr3, %[cr3]"
550 			     : /* output */ [cr3]"=r"(cr3));
551 	return cr3;
552 }
553 
554 static inline uint64_t get_cr4(void)
555 {
556 	uint64_t cr4;
557 
558 	__asm__ __volatile__("mov %%cr4, %[cr4]"
559 			     : /* output */ [cr4]"=r"(cr4));
560 	return cr4;
561 }
562 
563 static inline void set_cr4(uint64_t val)
564 {
565 	__asm__ __volatile__("mov %0, %%cr4" : : "r" (val) : "memory");
566 }
567 
568 static inline u64 xgetbv(u32 index)
569 {
570 	u32 eax, edx;
571 
572 	__asm__ __volatile__("xgetbv;"
573 		     : "=a" (eax), "=d" (edx)
574 		     : "c" (index));
575 	return eax | ((u64)edx << 32);
576 }
577 
578 static inline void xsetbv(u32 index, u64 value)
579 {
580 	u32 eax = value;
581 	u32 edx = value >> 32;
582 
583 	__asm__ __volatile__("xsetbv" :: "a" (eax), "d" (edx), "c" (index));
584 }
585 
586 static inline void wrpkru(u32 pkru)
587 {
588 	/* Note, ECX and EDX are architecturally required to be '0'. */
589 	asm volatile(".byte 0x0f,0x01,0xef\n\t"
590 		     : : "a" (pkru), "c"(0), "d"(0));
591 }
592 
593 static inline struct desc_ptr get_gdt(void)
594 {
595 	struct desc_ptr gdt;
596 	__asm__ __volatile__("sgdt %[gdt]"
597 			     : /* output */ [gdt]"=m"(gdt));
598 	return gdt;
599 }
600 
601 static inline struct desc_ptr get_idt(void)
602 {
603 	struct desc_ptr idt;
604 	__asm__ __volatile__("sidt %[idt]"
605 			     : /* output */ [idt]"=m"(idt));
606 	return idt;
607 }
608 
609 static inline void outl(uint16_t port, uint32_t value)
610 {
611 	__asm__ __volatile__("outl %%eax, %%dx" : : "d"(port), "a"(value));
612 }
613 
614 static inline void __cpuid(uint32_t function, uint32_t index,
615 			   uint32_t *eax, uint32_t *ebx,
616 			   uint32_t *ecx, uint32_t *edx)
617 {
618 	*eax = function;
619 	*ecx = index;
620 
621 	asm volatile("cpuid"
622 	    : "=a" (*eax),
623 	      "=b" (*ebx),
624 	      "=c" (*ecx),
625 	      "=d" (*edx)
626 	    : "0" (*eax), "2" (*ecx)
627 	    : "memory");
628 }
629 
630 static inline void cpuid(uint32_t function,
631 			 uint32_t *eax, uint32_t *ebx,
632 			 uint32_t *ecx, uint32_t *edx)
633 {
634 	return __cpuid(function, 0, eax, ebx, ecx, edx);
635 }
636 
637 static inline uint32_t this_cpu_fms(void)
638 {
639 	uint32_t eax, ebx, ecx, edx;
640 
641 	cpuid(1, &eax, &ebx, &ecx, &edx);
642 	return eax;
643 }
644 
645 static inline uint32_t this_cpu_family(void)
646 {
647 	return x86_family(this_cpu_fms());
648 }
649 
650 static inline uint32_t this_cpu_model(void)
651 {
652 	return x86_model(this_cpu_fms());
653 }
654 
655 static inline bool this_cpu_vendor_string_is(const char *vendor)
656 {
657 	const uint32_t *chunk = (const uint32_t *)vendor;
658 	uint32_t eax, ebx, ecx, edx;
659 
660 	cpuid(0, &eax, &ebx, &ecx, &edx);
661 	return (ebx == chunk[0] && edx == chunk[1] && ecx == chunk[2]);
662 }
663 
664 static inline bool this_cpu_is_intel(void)
665 {
666 	return this_cpu_vendor_string_is("GenuineIntel");
667 }
668 
669 /*
670  * Exclude early K5 samples with a vendor string of "AMDisbetter!"
671  */
672 static inline bool this_cpu_is_amd(void)
673 {
674 	return this_cpu_vendor_string_is("AuthenticAMD");
675 }
676 
677 static inline uint32_t __this_cpu_has(uint32_t function, uint32_t index,
678 				      uint8_t reg, uint8_t lo, uint8_t hi)
679 {
680 	uint32_t gprs[4];
681 
682 	__cpuid(function, index,
683 		&gprs[KVM_CPUID_EAX], &gprs[KVM_CPUID_EBX],
684 		&gprs[KVM_CPUID_ECX], &gprs[KVM_CPUID_EDX]);
685 
686 	return (gprs[reg] & GENMASK(hi, lo)) >> lo;
687 }
688 
689 static inline bool this_cpu_has(struct kvm_x86_cpu_feature feature)
690 {
691 	return __this_cpu_has(feature.function, feature.index,
692 			      feature.reg, feature.bit, feature.bit);
693 }
694 
695 static inline uint32_t this_cpu_property(struct kvm_x86_cpu_property property)
696 {
697 	return __this_cpu_has(property.function, property.index,
698 			      property.reg, property.lo_bit, property.hi_bit);
699 }
700 
701 static __always_inline bool this_cpu_has_p(struct kvm_x86_cpu_property property)
702 {
703 	uint32_t max_leaf;
704 
705 	switch (property.function & 0xc0000000) {
706 	case 0:
707 		max_leaf = this_cpu_property(X86_PROPERTY_MAX_BASIC_LEAF);
708 		break;
709 	case 0x40000000:
710 		max_leaf = this_cpu_property(X86_PROPERTY_MAX_KVM_LEAF);
711 		break;
712 	case 0x80000000:
713 		max_leaf = this_cpu_property(X86_PROPERTY_MAX_EXT_LEAF);
714 		break;
715 	case 0xc0000000:
716 		max_leaf = this_cpu_property(X86_PROPERTY_MAX_CENTAUR_LEAF);
717 	}
718 	return max_leaf >= property.function;
719 }
720 
721 static inline bool this_pmu_has(struct kvm_x86_pmu_feature feature)
722 {
723 	uint32_t nr_bits;
724 
725 	if (feature.f.reg == KVM_CPUID_EBX) {
726 		nr_bits = this_cpu_property(X86_PROPERTY_PMU_EBX_BIT_VECTOR_LENGTH);
727 		return nr_bits > feature.f.bit && !this_cpu_has(feature.f);
728 	}
729 
730 	GUEST_ASSERT(feature.f.reg == KVM_CPUID_ECX);
731 	nr_bits = this_cpu_property(X86_PROPERTY_PMU_NR_FIXED_COUNTERS);
732 	return nr_bits > feature.f.bit || this_cpu_has(feature.f);
733 }
734 
735 static __always_inline uint64_t this_cpu_supported_xcr0(void)
736 {
737 	if (!this_cpu_has_p(X86_PROPERTY_SUPPORTED_XCR0_LO))
738 		return 0;
739 
740 	return this_cpu_property(X86_PROPERTY_SUPPORTED_XCR0_LO) |
741 	       ((uint64_t)this_cpu_property(X86_PROPERTY_SUPPORTED_XCR0_HI) << 32);
742 }
743 
744 typedef u32		__attribute__((vector_size(16))) sse128_t;
745 #define __sse128_u	union { sse128_t vec; u64 as_u64[2]; u32 as_u32[4]; }
746 #define sse128_lo(x)	({ __sse128_u t; t.vec = x; t.as_u64[0]; })
747 #define sse128_hi(x)	({ __sse128_u t; t.vec = x; t.as_u64[1]; })
748 
749 static inline void read_sse_reg(int reg, sse128_t *data)
750 {
751 	switch (reg) {
752 	case 0:
753 		asm("movdqa %%xmm0, %0" : "=m"(*data));
754 		break;
755 	case 1:
756 		asm("movdqa %%xmm1, %0" : "=m"(*data));
757 		break;
758 	case 2:
759 		asm("movdqa %%xmm2, %0" : "=m"(*data));
760 		break;
761 	case 3:
762 		asm("movdqa %%xmm3, %0" : "=m"(*data));
763 		break;
764 	case 4:
765 		asm("movdqa %%xmm4, %0" : "=m"(*data));
766 		break;
767 	case 5:
768 		asm("movdqa %%xmm5, %0" : "=m"(*data));
769 		break;
770 	case 6:
771 		asm("movdqa %%xmm6, %0" : "=m"(*data));
772 		break;
773 	case 7:
774 		asm("movdqa %%xmm7, %0" : "=m"(*data));
775 		break;
776 	default:
777 		BUG();
778 	}
779 }
780 
781 static inline void write_sse_reg(int reg, const sse128_t *data)
782 {
783 	switch (reg) {
784 	case 0:
785 		asm("movdqa %0, %%xmm0" : : "m"(*data));
786 		break;
787 	case 1:
788 		asm("movdqa %0, %%xmm1" : : "m"(*data));
789 		break;
790 	case 2:
791 		asm("movdqa %0, %%xmm2" : : "m"(*data));
792 		break;
793 	case 3:
794 		asm("movdqa %0, %%xmm3" : : "m"(*data));
795 		break;
796 	case 4:
797 		asm("movdqa %0, %%xmm4" : : "m"(*data));
798 		break;
799 	case 5:
800 		asm("movdqa %0, %%xmm5" : : "m"(*data));
801 		break;
802 	case 6:
803 		asm("movdqa %0, %%xmm6" : : "m"(*data));
804 		break;
805 	case 7:
806 		asm("movdqa %0, %%xmm7" : : "m"(*data));
807 		break;
808 	default:
809 		BUG();
810 	}
811 }
812 
813 static inline void cpu_relax(void)
814 {
815 	asm volatile("rep; nop" ::: "memory");
816 }
817 
818 #define ud2()			\
819 	__asm__ __volatile__(	\
820 		"ud2\n"	\
821 		)
822 
823 #define hlt()			\
824 	__asm__ __volatile__(	\
825 		"hlt\n"	\
826 		)
827 
828 struct kvm_x86_state *vcpu_save_state(struct kvm_vcpu *vcpu);
829 void vcpu_load_state(struct kvm_vcpu *vcpu, struct kvm_x86_state *state);
830 void kvm_x86_state_cleanup(struct kvm_x86_state *state);
831 
832 const struct kvm_msr_list *kvm_get_msr_index_list(void);
833 const struct kvm_msr_list *kvm_get_feature_msr_index_list(void);
834 bool kvm_msr_is_in_save_restore_list(uint32_t msr_index);
835 uint64_t kvm_get_feature_msr(uint64_t msr_index);
836 
837 static inline void vcpu_msrs_get(struct kvm_vcpu *vcpu,
838 				 struct kvm_msrs *msrs)
839 {
840 	int r = __vcpu_ioctl(vcpu, KVM_GET_MSRS, msrs);
841 
842 	TEST_ASSERT(r == msrs->nmsrs,
843 		    "KVM_GET_MSRS failed, r: %i (failed on MSR %x)",
844 		    r, r < 0 || r >= msrs->nmsrs ? -1 : msrs->entries[r].index);
845 }
846 static inline void vcpu_msrs_set(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs)
847 {
848 	int r = __vcpu_ioctl(vcpu, KVM_SET_MSRS, msrs);
849 
850 	TEST_ASSERT(r == msrs->nmsrs,
851 		    "KVM_SET_MSRS failed, r: %i (failed on MSR %x)",
852 		    r, r < 0 || r >= msrs->nmsrs ? -1 : msrs->entries[r].index);
853 }
854 static inline void vcpu_debugregs_get(struct kvm_vcpu *vcpu,
855 				      struct kvm_debugregs *debugregs)
856 {
857 	vcpu_ioctl(vcpu, KVM_GET_DEBUGREGS, debugregs);
858 }
859 static inline void vcpu_debugregs_set(struct kvm_vcpu *vcpu,
860 				      struct kvm_debugregs *debugregs)
861 {
862 	vcpu_ioctl(vcpu, KVM_SET_DEBUGREGS, debugregs);
863 }
864 static inline void vcpu_xsave_get(struct kvm_vcpu *vcpu,
865 				  struct kvm_xsave *xsave)
866 {
867 	vcpu_ioctl(vcpu, KVM_GET_XSAVE, xsave);
868 }
869 static inline void vcpu_xsave2_get(struct kvm_vcpu *vcpu,
870 				   struct kvm_xsave *xsave)
871 {
872 	vcpu_ioctl(vcpu, KVM_GET_XSAVE2, xsave);
873 }
874 static inline void vcpu_xsave_set(struct kvm_vcpu *vcpu,
875 				  struct kvm_xsave *xsave)
876 {
877 	vcpu_ioctl(vcpu, KVM_SET_XSAVE, xsave);
878 }
879 static inline void vcpu_xcrs_get(struct kvm_vcpu *vcpu,
880 				 struct kvm_xcrs *xcrs)
881 {
882 	vcpu_ioctl(vcpu, KVM_GET_XCRS, xcrs);
883 }
884 static inline void vcpu_xcrs_set(struct kvm_vcpu *vcpu, struct kvm_xcrs *xcrs)
885 {
886 	vcpu_ioctl(vcpu, KVM_SET_XCRS, xcrs);
887 }
888 
889 const struct kvm_cpuid_entry2 *get_cpuid_entry(const struct kvm_cpuid2 *cpuid,
890 					       uint32_t function, uint32_t index);
891 const struct kvm_cpuid2 *kvm_get_supported_cpuid(void);
892 const struct kvm_cpuid2 *kvm_get_supported_hv_cpuid(void);
893 const struct kvm_cpuid2 *vcpu_get_supported_hv_cpuid(struct kvm_vcpu *vcpu);
894 
895 static inline uint32_t kvm_cpu_fms(void)
896 {
897 	return get_cpuid_entry(kvm_get_supported_cpuid(), 0x1, 0)->eax;
898 }
899 
900 static inline uint32_t kvm_cpu_family(void)
901 {
902 	return x86_family(kvm_cpu_fms());
903 }
904 
905 static inline uint32_t kvm_cpu_model(void)
906 {
907 	return x86_model(kvm_cpu_fms());
908 }
909 
910 bool kvm_cpuid_has(const struct kvm_cpuid2 *cpuid,
911 		   struct kvm_x86_cpu_feature feature);
912 
913 static inline bool kvm_cpu_has(struct kvm_x86_cpu_feature feature)
914 {
915 	return kvm_cpuid_has(kvm_get_supported_cpuid(), feature);
916 }
917 
918 uint32_t kvm_cpuid_property(const struct kvm_cpuid2 *cpuid,
919 			    struct kvm_x86_cpu_property property);
920 
921 static inline uint32_t kvm_cpu_property(struct kvm_x86_cpu_property property)
922 {
923 	return kvm_cpuid_property(kvm_get_supported_cpuid(), property);
924 }
925 
926 static __always_inline bool kvm_cpu_has_p(struct kvm_x86_cpu_property property)
927 {
928 	uint32_t max_leaf;
929 
930 	switch (property.function & 0xc0000000) {
931 	case 0:
932 		max_leaf = kvm_cpu_property(X86_PROPERTY_MAX_BASIC_LEAF);
933 		break;
934 	case 0x40000000:
935 		max_leaf = kvm_cpu_property(X86_PROPERTY_MAX_KVM_LEAF);
936 		break;
937 	case 0x80000000:
938 		max_leaf = kvm_cpu_property(X86_PROPERTY_MAX_EXT_LEAF);
939 		break;
940 	case 0xc0000000:
941 		max_leaf = kvm_cpu_property(X86_PROPERTY_MAX_CENTAUR_LEAF);
942 	}
943 	return max_leaf >= property.function;
944 }
945 
946 static inline bool kvm_pmu_has(struct kvm_x86_pmu_feature feature)
947 {
948 	uint32_t nr_bits;
949 
950 	if (feature.f.reg == KVM_CPUID_EBX) {
951 		nr_bits = kvm_cpu_property(X86_PROPERTY_PMU_EBX_BIT_VECTOR_LENGTH);
952 		return nr_bits > feature.f.bit && !kvm_cpu_has(feature.f);
953 	}
954 
955 	TEST_ASSERT_EQ(feature.f.reg, KVM_CPUID_ECX);
956 	nr_bits = kvm_cpu_property(X86_PROPERTY_PMU_NR_FIXED_COUNTERS);
957 	return nr_bits > feature.f.bit || kvm_cpu_has(feature.f);
958 }
959 
960 static __always_inline uint64_t kvm_cpu_supported_xcr0(void)
961 {
962 	if (!kvm_cpu_has_p(X86_PROPERTY_SUPPORTED_XCR0_LO))
963 		return 0;
964 
965 	return kvm_cpu_property(X86_PROPERTY_SUPPORTED_XCR0_LO) |
966 	       ((uint64_t)kvm_cpu_property(X86_PROPERTY_SUPPORTED_XCR0_HI) << 32);
967 }
968 
969 static inline size_t kvm_cpuid2_size(int nr_entries)
970 {
971 	return sizeof(struct kvm_cpuid2) +
972 	       sizeof(struct kvm_cpuid_entry2) * nr_entries;
973 }
974 
975 /*
976  * Allocate a "struct kvm_cpuid2* instance, with the 0-length arrary of
977  * entries sized to hold @nr_entries.  The caller is responsible for freeing
978  * the struct.
979  */
980 static inline struct kvm_cpuid2 *allocate_kvm_cpuid2(int nr_entries)
981 {
982 	struct kvm_cpuid2 *cpuid;
983 
984 	cpuid = malloc(kvm_cpuid2_size(nr_entries));
985 	TEST_ASSERT(cpuid, "-ENOMEM when allocating kvm_cpuid2");
986 
987 	cpuid->nent = nr_entries;
988 
989 	return cpuid;
990 }
991 
992 void vcpu_init_cpuid(struct kvm_vcpu *vcpu, const struct kvm_cpuid2 *cpuid);
993 void vcpu_set_hv_cpuid(struct kvm_vcpu *vcpu);
994 
995 static inline struct kvm_cpuid_entry2 *__vcpu_get_cpuid_entry(struct kvm_vcpu *vcpu,
996 							      uint32_t function,
997 							      uint32_t index)
998 {
999 	return (struct kvm_cpuid_entry2 *)get_cpuid_entry(vcpu->cpuid,
1000 							  function, index);
1001 }
1002 
1003 static inline struct kvm_cpuid_entry2 *vcpu_get_cpuid_entry(struct kvm_vcpu *vcpu,
1004 							    uint32_t function)
1005 {
1006 	return __vcpu_get_cpuid_entry(vcpu, function, 0);
1007 }
1008 
1009 static inline int __vcpu_set_cpuid(struct kvm_vcpu *vcpu)
1010 {
1011 	int r;
1012 
1013 	TEST_ASSERT(vcpu->cpuid, "Must do vcpu_init_cpuid() first");
1014 	r = __vcpu_ioctl(vcpu, KVM_SET_CPUID2, vcpu->cpuid);
1015 	if (r)
1016 		return r;
1017 
1018 	/* On success, refresh the cache to pick up adjustments made by KVM. */
1019 	vcpu_ioctl(vcpu, KVM_GET_CPUID2, vcpu->cpuid);
1020 	return 0;
1021 }
1022 
1023 static inline void vcpu_set_cpuid(struct kvm_vcpu *vcpu)
1024 {
1025 	TEST_ASSERT(vcpu->cpuid, "Must do vcpu_init_cpuid() first");
1026 	vcpu_ioctl(vcpu, KVM_SET_CPUID2, vcpu->cpuid);
1027 
1028 	/* Refresh the cache to pick up adjustments made by KVM. */
1029 	vcpu_ioctl(vcpu, KVM_GET_CPUID2, vcpu->cpuid);
1030 }
1031 
1032 void vcpu_set_cpuid_property(struct kvm_vcpu *vcpu,
1033 			     struct kvm_x86_cpu_property property,
1034 			     uint32_t value);
1035 void vcpu_set_cpuid_maxphyaddr(struct kvm_vcpu *vcpu, uint8_t maxphyaddr);
1036 
1037 void vcpu_clear_cpuid_entry(struct kvm_vcpu *vcpu, uint32_t function);
1038 
1039 static inline bool vcpu_cpuid_has(struct kvm_vcpu *vcpu,
1040 				  struct kvm_x86_cpu_feature feature)
1041 {
1042 	struct kvm_cpuid_entry2 *entry;
1043 
1044 	entry = __vcpu_get_cpuid_entry(vcpu, feature.function, feature.index);
1045 	return *((&entry->eax) + feature.reg) & BIT(feature.bit);
1046 }
1047 
1048 void vcpu_set_or_clear_cpuid_feature(struct kvm_vcpu *vcpu,
1049 				     struct kvm_x86_cpu_feature feature,
1050 				     bool set);
1051 
1052 static inline void vcpu_set_cpuid_feature(struct kvm_vcpu *vcpu,
1053 					  struct kvm_x86_cpu_feature feature)
1054 {
1055 	vcpu_set_or_clear_cpuid_feature(vcpu, feature, true);
1056 
1057 }
1058 
1059 static inline void vcpu_clear_cpuid_feature(struct kvm_vcpu *vcpu,
1060 					    struct kvm_x86_cpu_feature feature)
1061 {
1062 	vcpu_set_or_clear_cpuid_feature(vcpu, feature, false);
1063 }
1064 
1065 uint64_t vcpu_get_msr(struct kvm_vcpu *vcpu, uint64_t msr_index);
1066 int _vcpu_set_msr(struct kvm_vcpu *vcpu, uint64_t msr_index, uint64_t msr_value);
1067 
1068 /*
1069  * Assert on an MSR access(es) and pretty print the MSR name when possible.
1070  * Note, the caller provides the stringified name so that the name of macro is
1071  * printed, not the value the macro resolves to (due to macro expansion).
1072  */
1073 #define TEST_ASSERT_MSR(cond, fmt, msr, str, args...)				\
1074 do {										\
1075 	if (__builtin_constant_p(msr)) {					\
1076 		TEST_ASSERT(cond, fmt, str, args);				\
1077 	} else if (!(cond)) {							\
1078 		char buf[16];							\
1079 										\
1080 		snprintf(buf, sizeof(buf), "MSR 0x%x", msr);			\
1081 		TEST_ASSERT(cond, fmt, buf, args);				\
1082 	}									\
1083 } while (0)
1084 
1085 /*
1086  * Returns true if KVM should return the last written value when reading an MSR
1087  * from userspace, e.g. the MSR isn't a command MSR, doesn't emulate state that
1088  * is changing, etc.  This is NOT an exhaustive list!  The intent is to filter
1089  * out MSRs that are not durable _and_ that a selftest wants to write.
1090  */
1091 static inline bool is_durable_msr(uint32_t msr)
1092 {
1093 	return msr != MSR_IA32_TSC;
1094 }
1095 
1096 #define vcpu_set_msr(vcpu, msr, val)							\
1097 do {											\
1098 	uint64_t r, v = val;								\
1099 											\
1100 	TEST_ASSERT_MSR(_vcpu_set_msr(vcpu, msr, v) == 1,				\
1101 			"KVM_SET_MSRS failed on %s, value = 0x%lx", msr, #msr, v);	\
1102 	if (!is_durable_msr(msr))							\
1103 		break;									\
1104 	r = vcpu_get_msr(vcpu, msr);							\
1105 	TEST_ASSERT_MSR(r == v, "Set %s to '0x%lx', got back '0x%lx'", msr, #msr, v, r);\
1106 } while (0)
1107 
1108 void kvm_get_cpu_address_width(unsigned int *pa_bits, unsigned int *va_bits);
1109 void kvm_init_vm_address_properties(struct kvm_vm *vm);
1110 bool vm_is_unrestricted_guest(struct kvm_vm *vm);
1111 
1112 struct ex_regs {
1113 	uint64_t rax, rcx, rdx, rbx;
1114 	uint64_t rbp, rsi, rdi;
1115 	uint64_t r8, r9, r10, r11;
1116 	uint64_t r12, r13, r14, r15;
1117 	uint64_t vector;
1118 	uint64_t error_code;
1119 	uint64_t rip;
1120 	uint64_t cs;
1121 	uint64_t rflags;
1122 };
1123 
1124 struct idt_entry {
1125 	uint16_t offset0;
1126 	uint16_t selector;
1127 	uint16_t ist : 3;
1128 	uint16_t : 5;
1129 	uint16_t type : 4;
1130 	uint16_t : 1;
1131 	uint16_t dpl : 2;
1132 	uint16_t p : 1;
1133 	uint16_t offset1;
1134 	uint32_t offset2; uint32_t reserved;
1135 };
1136 
1137 void vm_install_exception_handler(struct kvm_vm *vm, int vector,
1138 			void (*handler)(struct ex_regs *));
1139 
1140 /* If a toddler were to say "abracadabra". */
1141 #define KVM_EXCEPTION_MAGIC 0xabacadabaULL
1142 
1143 /*
1144  * KVM selftest exception fixup uses registers to coordinate with the exception
1145  * handler, versus the kernel's in-memory tables and KVM-Unit-Tests's in-memory
1146  * per-CPU data.  Using only registers avoids having to map memory into the
1147  * guest, doesn't require a valid, stable GS.base, and reduces the risk of
1148  * for recursive faults when accessing memory in the handler.  The downside to
1149  * using registers is that it restricts what registers can be used by the actual
1150  * instruction.  But, selftests are 64-bit only, making register* pressure a
1151  * minor concern.  Use r9-r11 as they are volatile, i.e. don't need to be saved
1152  * by the callee, and except for r11 are not implicit parameters to any
1153  * instructions.  Ideally, fixup would use r8-r10 and thus avoid implicit
1154  * parameters entirely, but Hyper-V's hypercall ABI uses r8 and testing Hyper-V
1155  * is higher priority than testing non-faulting SYSCALL/SYSRET.
1156  *
1157  * Note, the fixup handler deliberately does not handle #DE, i.e. the vector
1158  * is guaranteed to be non-zero on fault.
1159  *
1160  * REGISTER INPUTS:
1161  * r9  = MAGIC
1162  * r10 = RIP
1163  * r11 = new RIP on fault
1164  *
1165  * REGISTER OUTPUTS:
1166  * r9  = exception vector (non-zero)
1167  * r10 = error code
1168  */
1169 #define __KVM_ASM_SAFE(insn, fep)				\
1170 	"mov $" __stringify(KVM_EXCEPTION_MAGIC) ", %%r9\n\t"	\
1171 	"lea 1f(%%rip), %%r10\n\t"				\
1172 	"lea 2f(%%rip), %%r11\n\t"				\
1173 	fep "1: " insn "\n\t"					\
1174 	"xor %%r9, %%r9\n\t"					\
1175 	"2:\n\t"						\
1176 	"mov  %%r9b, %[vector]\n\t"				\
1177 	"mov  %%r10, %[error_code]\n\t"
1178 
1179 #define KVM_ASM_SAFE(insn) __KVM_ASM_SAFE(insn, "")
1180 #define KVM_ASM_SAFE_FEP(insn) __KVM_ASM_SAFE(insn, KVM_FEP)
1181 
1182 #define KVM_ASM_SAFE_OUTPUTS(v, ec)	[vector] "=qm"(v), [error_code] "=rm"(ec)
1183 #define KVM_ASM_SAFE_CLOBBERS	"r9", "r10", "r11"
1184 
1185 #define kvm_asm_safe(insn, inputs...)					\
1186 ({									\
1187 	uint64_t ign_error_code;					\
1188 	uint8_t vector;							\
1189 									\
1190 	asm volatile(KVM_ASM_SAFE(insn)					\
1191 		     : KVM_ASM_SAFE_OUTPUTS(vector, ign_error_code)	\
1192 		     : inputs						\
1193 		     : KVM_ASM_SAFE_CLOBBERS);				\
1194 	vector;								\
1195 })
1196 
1197 #define kvm_asm_safe_ec(insn, error_code, inputs...)			\
1198 ({									\
1199 	uint8_t vector;							\
1200 									\
1201 	asm volatile(KVM_ASM_SAFE(insn)					\
1202 		     : KVM_ASM_SAFE_OUTPUTS(vector, error_code)		\
1203 		     : inputs						\
1204 		     : KVM_ASM_SAFE_CLOBBERS);				\
1205 	vector;								\
1206 })
1207 
1208 #define kvm_asm_safe_fep(insn, inputs...)				\
1209 ({									\
1210 	uint64_t ign_error_code;					\
1211 	uint8_t vector;							\
1212 									\
1213 	asm volatile(KVM_ASM_SAFE(insn)					\
1214 		     : KVM_ASM_SAFE_OUTPUTS(vector, ign_error_code)	\
1215 		     : inputs						\
1216 		     : KVM_ASM_SAFE_CLOBBERS);				\
1217 	vector;								\
1218 })
1219 
1220 #define kvm_asm_safe_ec_fep(insn, error_code, inputs...)		\
1221 ({									\
1222 	uint8_t vector;							\
1223 									\
1224 	asm volatile(KVM_ASM_SAFE_FEP(insn)				\
1225 		     : KVM_ASM_SAFE_OUTPUTS(vector, error_code)		\
1226 		     : inputs						\
1227 		     : KVM_ASM_SAFE_CLOBBERS);				\
1228 	vector;								\
1229 })
1230 
1231 #define BUILD_READ_U64_SAFE_HELPER(insn, _fep, _FEP)			\
1232 static inline uint8_t insn##_safe ##_fep(uint32_t idx, uint64_t *val)	\
1233 {									\
1234 	uint64_t error_code;						\
1235 	uint8_t vector;							\
1236 	uint32_t a, d;							\
1237 									\
1238 	asm volatile(KVM_ASM_SAFE##_FEP(#insn)				\
1239 		     : "=a"(a), "=d"(d),				\
1240 		       KVM_ASM_SAFE_OUTPUTS(vector, error_code)		\
1241 		     : "c"(idx)						\
1242 		     : KVM_ASM_SAFE_CLOBBERS);				\
1243 									\
1244 	*val = (uint64_t)a | ((uint64_t)d << 32);			\
1245 	return vector;							\
1246 }
1247 
1248 /*
1249  * Generate {insn}_safe() and {insn}_safe_fep() helpers for instructions that
1250  * use ECX as in input index, and EDX:EAX as a 64-bit output.
1251  */
1252 #define BUILD_READ_U64_SAFE_HELPERS(insn)				\
1253 	BUILD_READ_U64_SAFE_HELPER(insn, , )				\
1254 	BUILD_READ_U64_SAFE_HELPER(insn, _fep, _FEP)			\
1255 
1256 BUILD_READ_U64_SAFE_HELPERS(rdmsr)
1257 BUILD_READ_U64_SAFE_HELPERS(rdpmc)
1258 BUILD_READ_U64_SAFE_HELPERS(xgetbv)
1259 
1260 static inline uint8_t wrmsr_safe(uint32_t msr, uint64_t val)
1261 {
1262 	return kvm_asm_safe("wrmsr", "a"(val & -1u), "d"(val >> 32), "c"(msr));
1263 }
1264 
1265 static inline uint8_t xsetbv_safe(uint32_t index, uint64_t value)
1266 {
1267 	u32 eax = value;
1268 	u32 edx = value >> 32;
1269 
1270 	return kvm_asm_safe("xsetbv", "a" (eax), "d" (edx), "c" (index));
1271 }
1272 
1273 bool kvm_is_tdp_enabled(void);
1274 
1275 static inline bool kvm_is_pmu_enabled(void)
1276 {
1277 	return get_kvm_param_bool("enable_pmu");
1278 }
1279 
1280 static inline bool kvm_is_forced_emulation_enabled(void)
1281 {
1282 	return !!get_kvm_param_integer("force_emulation_prefix");
1283 }
1284 
1285 uint64_t *__vm_get_page_table_entry(struct kvm_vm *vm, uint64_t vaddr,
1286 				    int *level);
1287 uint64_t *vm_get_page_table_entry(struct kvm_vm *vm, uint64_t vaddr);
1288 
1289 uint64_t kvm_hypercall(uint64_t nr, uint64_t a0, uint64_t a1, uint64_t a2,
1290 		       uint64_t a3);
1291 uint64_t __xen_hypercall(uint64_t nr, uint64_t a0, void *a1);
1292 void xen_hypercall(uint64_t nr, uint64_t a0, void *a1);
1293 
1294 static inline uint64_t __kvm_hypercall_map_gpa_range(uint64_t gpa,
1295 						     uint64_t size, uint64_t flags)
1296 {
1297 	return kvm_hypercall(KVM_HC_MAP_GPA_RANGE, gpa, size >> PAGE_SHIFT, flags, 0);
1298 }
1299 
1300 static inline void kvm_hypercall_map_gpa_range(uint64_t gpa, uint64_t size,
1301 					       uint64_t flags)
1302 {
1303 	uint64_t ret = __kvm_hypercall_map_gpa_range(gpa, size, flags);
1304 
1305 	GUEST_ASSERT(!ret);
1306 }
1307 
1308 void __vm_xsave_require_permission(uint64_t xfeature, const char *name);
1309 
1310 #define vm_xsave_require_permission(xfeature)	\
1311 	__vm_xsave_require_permission(xfeature, #xfeature)
1312 
1313 enum pg_level {
1314 	PG_LEVEL_NONE,
1315 	PG_LEVEL_4K,
1316 	PG_LEVEL_2M,
1317 	PG_LEVEL_1G,
1318 	PG_LEVEL_512G,
1319 	PG_LEVEL_NUM
1320 };
1321 
1322 #define PG_LEVEL_SHIFT(_level) ((_level - 1) * 9 + 12)
1323 #define PG_LEVEL_SIZE(_level) (1ull << PG_LEVEL_SHIFT(_level))
1324 
1325 #define PG_SIZE_4K PG_LEVEL_SIZE(PG_LEVEL_4K)
1326 #define PG_SIZE_2M PG_LEVEL_SIZE(PG_LEVEL_2M)
1327 #define PG_SIZE_1G PG_LEVEL_SIZE(PG_LEVEL_1G)
1328 
1329 void __virt_pg_map(struct kvm_vm *vm, uint64_t vaddr, uint64_t paddr, int level);
1330 void virt_map_level(struct kvm_vm *vm, uint64_t vaddr, uint64_t paddr,
1331 		    uint64_t nr_bytes, int level);
1332 
1333 /*
1334  * Basic CPU control in CR0
1335  */
1336 #define X86_CR0_PE          (1UL<<0) /* Protection Enable */
1337 #define X86_CR0_MP          (1UL<<1) /* Monitor Coprocessor */
1338 #define X86_CR0_EM          (1UL<<2) /* Emulation */
1339 #define X86_CR0_TS          (1UL<<3) /* Task Switched */
1340 #define X86_CR0_ET          (1UL<<4) /* Extension Type */
1341 #define X86_CR0_NE          (1UL<<5) /* Numeric Error */
1342 #define X86_CR0_WP          (1UL<<16) /* Write Protect */
1343 #define X86_CR0_AM          (1UL<<18) /* Alignment Mask */
1344 #define X86_CR0_NW          (1UL<<29) /* Not Write-through */
1345 #define X86_CR0_CD          (1UL<<30) /* Cache Disable */
1346 #define X86_CR0_PG          (1UL<<31) /* Paging */
1347 
1348 #define PFERR_PRESENT_BIT 0
1349 #define PFERR_WRITE_BIT 1
1350 #define PFERR_USER_BIT 2
1351 #define PFERR_RSVD_BIT 3
1352 #define PFERR_FETCH_BIT 4
1353 #define PFERR_PK_BIT 5
1354 #define PFERR_SGX_BIT 15
1355 #define PFERR_GUEST_FINAL_BIT 32
1356 #define PFERR_GUEST_PAGE_BIT 33
1357 #define PFERR_IMPLICIT_ACCESS_BIT 48
1358 
1359 #define PFERR_PRESENT_MASK	BIT(PFERR_PRESENT_BIT)
1360 #define PFERR_WRITE_MASK	BIT(PFERR_WRITE_BIT)
1361 #define PFERR_USER_MASK		BIT(PFERR_USER_BIT)
1362 #define PFERR_RSVD_MASK		BIT(PFERR_RSVD_BIT)
1363 #define PFERR_FETCH_MASK	BIT(PFERR_FETCH_BIT)
1364 #define PFERR_PK_MASK		BIT(PFERR_PK_BIT)
1365 #define PFERR_SGX_MASK		BIT(PFERR_SGX_BIT)
1366 #define PFERR_GUEST_FINAL_MASK	BIT_ULL(PFERR_GUEST_FINAL_BIT)
1367 #define PFERR_GUEST_PAGE_MASK	BIT_ULL(PFERR_GUEST_PAGE_BIT)
1368 #define PFERR_IMPLICIT_ACCESS	BIT_ULL(PFERR_IMPLICIT_ACCESS_BIT)
1369 
1370 bool sys_clocksource_is_based_on_tsc(void);
1371 
1372 #endif /* SELFTEST_KVM_PROCESSOR_H */
1373