xref: /linux/tools/testing/selftests/kvm/include/x86/processor.h (revision c76a923828059ac9999e4415dcbd9706ac9540a0)
1 /* SPDX-License-Identifier: GPL-2.0-only */
2 /*
3  * Copyright (C) 2018, Google LLC.
4  */
5 
6 #ifndef SELFTEST_KVM_PROCESSOR_H
7 #define SELFTEST_KVM_PROCESSOR_H
8 
9 #include <assert.h>
10 #include <stdint.h>
11 #include <syscall.h>
12 
13 #include <asm/msr-index.h>
14 #include <asm/prctl.h>
15 
16 #include <linux/kvm_para.h>
17 #include <linux/stringify.h>
18 
19 #include "kvm_util.h"
20 #include "ucall_common.h"
21 
22 extern bool host_cpu_is_intel;
23 extern bool host_cpu_is_amd;
24 extern uint64_t guest_tsc_khz;
25 
26 #ifndef MAX_NR_CPUID_ENTRIES
27 #define MAX_NR_CPUID_ENTRIES 100
28 #endif
29 
30 /* Forced emulation prefix, used to invoke the emulator unconditionally. */
31 #define KVM_FEP "ud2; .byte 'k', 'v', 'm';"
32 
33 #define NMI_VECTOR		0x02
34 
35 #define X86_EFLAGS_FIXED	 (1u << 1)
36 
37 #define X86_CR4_VME		(1ul << 0)
38 #define X86_CR4_PVI		(1ul << 1)
39 #define X86_CR4_TSD		(1ul << 2)
40 #define X86_CR4_DE		(1ul << 3)
41 #define X86_CR4_PSE		(1ul << 4)
42 #define X86_CR4_PAE		(1ul << 5)
43 #define X86_CR4_MCE		(1ul << 6)
44 #define X86_CR4_PGE		(1ul << 7)
45 #define X86_CR4_PCE		(1ul << 8)
46 #define X86_CR4_OSFXSR		(1ul << 9)
47 #define X86_CR4_OSXMMEXCPT	(1ul << 10)
48 #define X86_CR4_UMIP		(1ul << 11)
49 #define X86_CR4_LA57		(1ul << 12)
50 #define X86_CR4_VMXE		(1ul << 13)
51 #define X86_CR4_SMXE		(1ul << 14)
52 #define X86_CR4_FSGSBASE	(1ul << 16)
53 #define X86_CR4_PCIDE		(1ul << 17)
54 #define X86_CR4_OSXSAVE		(1ul << 18)
55 #define X86_CR4_SMEP		(1ul << 20)
56 #define X86_CR4_SMAP		(1ul << 21)
57 #define X86_CR4_PKE		(1ul << 22)
58 
59 struct xstate_header {
60 	u64				xstate_bv;
61 	u64				xcomp_bv;
62 	u64				reserved[6];
63 } __attribute__((packed));
64 
65 struct xstate {
66 	u8				i387[512];
67 	struct xstate_header		header;
68 	u8				extended_state_area[0];
69 } __attribute__ ((packed, aligned (64)));
70 
71 #define XFEATURE_MASK_FP		BIT_ULL(0)
72 #define XFEATURE_MASK_SSE		BIT_ULL(1)
73 #define XFEATURE_MASK_YMM		BIT_ULL(2)
74 #define XFEATURE_MASK_BNDREGS		BIT_ULL(3)
75 #define XFEATURE_MASK_BNDCSR		BIT_ULL(4)
76 #define XFEATURE_MASK_OPMASK		BIT_ULL(5)
77 #define XFEATURE_MASK_ZMM_Hi256		BIT_ULL(6)
78 #define XFEATURE_MASK_Hi16_ZMM		BIT_ULL(7)
79 #define XFEATURE_MASK_PT		BIT_ULL(8)
80 #define XFEATURE_MASK_PKRU		BIT_ULL(9)
81 #define XFEATURE_MASK_PASID		BIT_ULL(10)
82 #define XFEATURE_MASK_CET_USER		BIT_ULL(11)
83 #define XFEATURE_MASK_CET_KERNEL	BIT_ULL(12)
84 #define XFEATURE_MASK_LBR		BIT_ULL(15)
85 #define XFEATURE_MASK_XTILE_CFG		BIT_ULL(17)
86 #define XFEATURE_MASK_XTILE_DATA	BIT_ULL(18)
87 
88 #define XFEATURE_MASK_AVX512		(XFEATURE_MASK_OPMASK | \
89 					 XFEATURE_MASK_ZMM_Hi256 | \
90 					 XFEATURE_MASK_Hi16_ZMM)
91 #define XFEATURE_MASK_XTILE		(XFEATURE_MASK_XTILE_DATA | \
92 					 XFEATURE_MASK_XTILE_CFG)
93 
94 /* Note, these are ordered alphabetically to match kvm_cpuid_entry2.  Eww. */
95 enum cpuid_output_regs {
96 	KVM_CPUID_EAX,
97 	KVM_CPUID_EBX,
98 	KVM_CPUID_ECX,
99 	KVM_CPUID_EDX
100 };
101 
102 /*
103  * Pack the information into a 64-bit value so that each X86_FEATURE_XXX can be
104  * passed by value with no overhead.
105  */
106 struct kvm_x86_cpu_feature {
107 	u32	function;
108 	u16	index;
109 	u8	reg;
110 	u8	bit;
111 };
112 #define	KVM_X86_CPU_FEATURE(fn, idx, gpr, __bit)				\
113 ({										\
114 	struct kvm_x86_cpu_feature feature = {					\
115 		.function = fn,							\
116 		.index = idx,							\
117 		.reg = KVM_CPUID_##gpr,						\
118 		.bit = __bit,							\
119 	};									\
120 										\
121 	kvm_static_assert((fn & 0xc0000000) == 0 ||				\
122 			  (fn & 0xc0000000) == 0x40000000 ||			\
123 			  (fn & 0xc0000000) == 0x80000000 ||			\
124 			  (fn & 0xc0000000) == 0xc0000000);			\
125 	kvm_static_assert(idx < BIT(sizeof(feature.index) * BITS_PER_BYTE));	\
126 	feature;								\
127 })
128 
129 /*
130  * Basic Leafs, a.k.a. Intel defined
131  */
132 #define	X86_FEATURE_MWAIT		KVM_X86_CPU_FEATURE(0x1, 0, ECX, 3)
133 #define	X86_FEATURE_VMX			KVM_X86_CPU_FEATURE(0x1, 0, ECX, 5)
134 #define	X86_FEATURE_SMX			KVM_X86_CPU_FEATURE(0x1, 0, ECX, 6)
135 #define	X86_FEATURE_PDCM		KVM_X86_CPU_FEATURE(0x1, 0, ECX, 15)
136 #define	X86_FEATURE_PCID		KVM_X86_CPU_FEATURE(0x1, 0, ECX, 17)
137 #define X86_FEATURE_X2APIC		KVM_X86_CPU_FEATURE(0x1, 0, ECX, 21)
138 #define	X86_FEATURE_MOVBE		KVM_X86_CPU_FEATURE(0x1, 0, ECX, 22)
139 #define	X86_FEATURE_TSC_DEADLINE_TIMER	KVM_X86_CPU_FEATURE(0x1, 0, ECX, 24)
140 #define	X86_FEATURE_XSAVE		KVM_X86_CPU_FEATURE(0x1, 0, ECX, 26)
141 #define	X86_FEATURE_OSXSAVE		KVM_X86_CPU_FEATURE(0x1, 0, ECX, 27)
142 #define	X86_FEATURE_RDRAND		KVM_X86_CPU_FEATURE(0x1, 0, ECX, 30)
143 #define	X86_FEATURE_HYPERVISOR		KVM_X86_CPU_FEATURE(0x1, 0, ECX, 31)
144 #define X86_FEATURE_PAE			KVM_X86_CPU_FEATURE(0x1, 0, EDX, 6)
145 #define	X86_FEATURE_MCE			KVM_X86_CPU_FEATURE(0x1, 0, EDX, 7)
146 #define	X86_FEATURE_APIC		KVM_X86_CPU_FEATURE(0x1, 0, EDX, 9)
147 #define	X86_FEATURE_CLFLUSH		KVM_X86_CPU_FEATURE(0x1, 0, EDX, 19)
148 #define	X86_FEATURE_XMM			KVM_X86_CPU_FEATURE(0x1, 0, EDX, 25)
149 #define	X86_FEATURE_XMM2		KVM_X86_CPU_FEATURE(0x1, 0, EDX, 26)
150 #define	X86_FEATURE_FSGSBASE		KVM_X86_CPU_FEATURE(0x7, 0, EBX, 0)
151 #define	X86_FEATURE_TSC_ADJUST		KVM_X86_CPU_FEATURE(0x7, 0, EBX, 1)
152 #define	X86_FEATURE_SGX			KVM_X86_CPU_FEATURE(0x7, 0, EBX, 2)
153 #define	X86_FEATURE_HLE			KVM_X86_CPU_FEATURE(0x7, 0, EBX, 4)
154 #define	X86_FEATURE_SMEP	        KVM_X86_CPU_FEATURE(0x7, 0, EBX, 7)
155 #define	X86_FEATURE_INVPCID		KVM_X86_CPU_FEATURE(0x7, 0, EBX, 10)
156 #define	X86_FEATURE_RTM			KVM_X86_CPU_FEATURE(0x7, 0, EBX, 11)
157 #define	X86_FEATURE_MPX			KVM_X86_CPU_FEATURE(0x7, 0, EBX, 14)
158 #define	X86_FEATURE_SMAP		KVM_X86_CPU_FEATURE(0x7, 0, EBX, 20)
159 #define	X86_FEATURE_PCOMMIT		KVM_X86_CPU_FEATURE(0x7, 0, EBX, 22)
160 #define	X86_FEATURE_CLFLUSHOPT		KVM_X86_CPU_FEATURE(0x7, 0, EBX, 23)
161 #define	X86_FEATURE_CLWB		KVM_X86_CPU_FEATURE(0x7, 0, EBX, 24)
162 #define	X86_FEATURE_UMIP		KVM_X86_CPU_FEATURE(0x7, 0, ECX, 2)
163 #define	X86_FEATURE_PKU			KVM_X86_CPU_FEATURE(0x7, 0, ECX, 3)
164 #define	X86_FEATURE_OSPKE		KVM_X86_CPU_FEATURE(0x7, 0, ECX, 4)
165 #define	X86_FEATURE_LA57		KVM_X86_CPU_FEATURE(0x7, 0, ECX, 16)
166 #define	X86_FEATURE_RDPID		KVM_X86_CPU_FEATURE(0x7, 0, ECX, 22)
167 #define	X86_FEATURE_SGX_LC		KVM_X86_CPU_FEATURE(0x7, 0, ECX, 30)
168 #define	X86_FEATURE_SHSTK		KVM_X86_CPU_FEATURE(0x7, 0, ECX, 7)
169 #define	X86_FEATURE_IBT			KVM_X86_CPU_FEATURE(0x7, 0, EDX, 20)
170 #define	X86_FEATURE_AMX_TILE		KVM_X86_CPU_FEATURE(0x7, 0, EDX, 24)
171 #define	X86_FEATURE_SPEC_CTRL		KVM_X86_CPU_FEATURE(0x7, 0, EDX, 26)
172 #define	X86_FEATURE_ARCH_CAPABILITIES	KVM_X86_CPU_FEATURE(0x7, 0, EDX, 29)
173 #define	X86_FEATURE_PKS			KVM_X86_CPU_FEATURE(0x7, 0, ECX, 31)
174 #define	X86_FEATURE_XTILECFG		KVM_X86_CPU_FEATURE(0xD, 0, EAX, 17)
175 #define	X86_FEATURE_XTILEDATA		KVM_X86_CPU_FEATURE(0xD, 0, EAX, 18)
176 #define	X86_FEATURE_XSAVES		KVM_X86_CPU_FEATURE(0xD, 1, EAX, 3)
177 #define	X86_FEATURE_XFD			KVM_X86_CPU_FEATURE(0xD, 1, EAX, 4)
178 #define X86_FEATURE_XTILEDATA_XFD	KVM_X86_CPU_FEATURE(0xD, 18, ECX, 2)
179 
180 /*
181  * Extended Leafs, a.k.a. AMD defined
182  */
183 #define	X86_FEATURE_SVM			KVM_X86_CPU_FEATURE(0x80000001, 0, ECX, 2)
184 #define	X86_FEATURE_PERFCTR_CORE	KVM_X86_CPU_FEATURE(0x80000001, 0, ECX, 23)
185 #define	X86_FEATURE_PERFCTR_NB		KVM_X86_CPU_FEATURE(0x80000001, 0, ECX, 24)
186 #define	X86_FEATURE_PERFCTR_LLC		KVM_X86_CPU_FEATURE(0x80000001, 0, ECX, 28)
187 #define	X86_FEATURE_NX			KVM_X86_CPU_FEATURE(0x80000001, 0, EDX, 20)
188 #define	X86_FEATURE_GBPAGES		KVM_X86_CPU_FEATURE(0x80000001, 0, EDX, 26)
189 #define	X86_FEATURE_RDTSCP		KVM_X86_CPU_FEATURE(0x80000001, 0, EDX, 27)
190 #define	X86_FEATURE_LM			KVM_X86_CPU_FEATURE(0x80000001, 0, EDX, 29)
191 #define	X86_FEATURE_INVTSC		KVM_X86_CPU_FEATURE(0x80000007, 0, EDX, 8)
192 #define	X86_FEATURE_RDPRU		KVM_X86_CPU_FEATURE(0x80000008, 0, EBX, 4)
193 #define	X86_FEATURE_AMD_IBPB		KVM_X86_CPU_FEATURE(0x80000008, 0, EBX, 12)
194 #define	X86_FEATURE_NPT			KVM_X86_CPU_FEATURE(0x8000000A, 0, EDX, 0)
195 #define	X86_FEATURE_LBRV		KVM_X86_CPU_FEATURE(0x8000000A, 0, EDX, 1)
196 #define	X86_FEATURE_NRIPS		KVM_X86_CPU_FEATURE(0x8000000A, 0, EDX, 3)
197 #define X86_FEATURE_TSCRATEMSR          KVM_X86_CPU_FEATURE(0x8000000A, 0, EDX, 4)
198 #define X86_FEATURE_PAUSEFILTER         KVM_X86_CPU_FEATURE(0x8000000A, 0, EDX, 10)
199 #define X86_FEATURE_PFTHRESHOLD         KVM_X86_CPU_FEATURE(0x8000000A, 0, EDX, 12)
200 #define	X86_FEATURE_VGIF		KVM_X86_CPU_FEATURE(0x8000000A, 0, EDX, 16)
201 #define X86_FEATURE_SEV			KVM_X86_CPU_FEATURE(0x8000001F, 0, EAX, 1)
202 #define X86_FEATURE_SEV_ES		KVM_X86_CPU_FEATURE(0x8000001F, 0, EAX, 3)
203 #define	X86_FEATURE_PERFMON_V2		KVM_X86_CPU_FEATURE(0x80000022, 0, EAX, 0)
204 #define	X86_FEATURE_LBR_PMC_FREEZE	KVM_X86_CPU_FEATURE(0x80000022, 0, EAX, 2)
205 
206 /*
207  * KVM defined paravirt features.
208  */
209 #define X86_FEATURE_KVM_CLOCKSOURCE	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 0)
210 #define X86_FEATURE_KVM_NOP_IO_DELAY	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 1)
211 #define X86_FEATURE_KVM_MMU_OP		KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 2)
212 #define X86_FEATURE_KVM_CLOCKSOURCE2	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 3)
213 #define X86_FEATURE_KVM_ASYNC_PF	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 4)
214 #define X86_FEATURE_KVM_STEAL_TIME	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 5)
215 #define X86_FEATURE_KVM_PV_EOI		KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 6)
216 #define X86_FEATURE_KVM_PV_UNHALT	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 7)
217 /* Bit 8 apparently isn't used?!?! */
218 #define X86_FEATURE_KVM_PV_TLB_FLUSH	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 9)
219 #define X86_FEATURE_KVM_ASYNC_PF_VMEXIT	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 10)
220 #define X86_FEATURE_KVM_PV_SEND_IPI	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 11)
221 #define X86_FEATURE_KVM_POLL_CONTROL	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 12)
222 #define X86_FEATURE_KVM_PV_SCHED_YIELD	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 13)
223 #define X86_FEATURE_KVM_ASYNC_PF_INT	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 14)
224 #define X86_FEATURE_KVM_MSI_EXT_DEST_ID	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 15)
225 #define X86_FEATURE_KVM_HC_MAP_GPA_RANGE	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 16)
226 #define X86_FEATURE_KVM_MIGRATION_CONTROL	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 17)
227 
228 /*
229  * Same idea as X86_FEATURE_XXX, but X86_PROPERTY_XXX retrieves a multi-bit
230  * value/property as opposed to a single-bit feature.  Again, pack the info
231  * into a 64-bit value to pass by value with no overhead.
232  */
233 struct kvm_x86_cpu_property {
234 	u32	function;
235 	u8	index;
236 	u8	reg;
237 	u8	lo_bit;
238 	u8	hi_bit;
239 };
240 #define	KVM_X86_CPU_PROPERTY(fn, idx, gpr, low_bit, high_bit)			\
241 ({										\
242 	struct kvm_x86_cpu_property property = {				\
243 		.function = fn,							\
244 		.index = idx,							\
245 		.reg = KVM_CPUID_##gpr,						\
246 		.lo_bit = low_bit,						\
247 		.hi_bit = high_bit,						\
248 	};									\
249 										\
250 	kvm_static_assert(low_bit < high_bit);					\
251 	kvm_static_assert((fn & 0xc0000000) == 0 ||				\
252 			  (fn & 0xc0000000) == 0x40000000 ||			\
253 			  (fn & 0xc0000000) == 0x80000000 ||			\
254 			  (fn & 0xc0000000) == 0xc0000000);			\
255 	kvm_static_assert(idx < BIT(sizeof(property.index) * BITS_PER_BYTE));	\
256 	property;								\
257 })
258 
259 #define X86_PROPERTY_MAX_BASIC_LEAF		KVM_X86_CPU_PROPERTY(0, 0, EAX, 0, 31)
260 #define X86_PROPERTY_PMU_VERSION		KVM_X86_CPU_PROPERTY(0xa, 0, EAX, 0, 7)
261 #define X86_PROPERTY_PMU_NR_GP_COUNTERS		KVM_X86_CPU_PROPERTY(0xa, 0, EAX, 8, 15)
262 #define X86_PROPERTY_PMU_GP_COUNTERS_BIT_WIDTH	KVM_X86_CPU_PROPERTY(0xa, 0, EAX, 16, 23)
263 #define X86_PROPERTY_PMU_EBX_BIT_VECTOR_LENGTH	KVM_X86_CPU_PROPERTY(0xa, 0, EAX, 24, 31)
264 #define X86_PROPERTY_PMU_EVENTS_MASK		KVM_X86_CPU_PROPERTY(0xa, 0, EBX, 0, 7)
265 #define X86_PROPERTY_PMU_FIXED_COUNTERS_BITMASK	KVM_X86_CPU_PROPERTY(0xa, 0, ECX, 0, 31)
266 #define X86_PROPERTY_PMU_NR_FIXED_COUNTERS	KVM_X86_CPU_PROPERTY(0xa, 0, EDX, 0, 4)
267 #define X86_PROPERTY_PMU_FIXED_COUNTERS_BIT_WIDTH	KVM_X86_CPU_PROPERTY(0xa, 0, EDX, 5, 12)
268 
269 #define X86_PROPERTY_SUPPORTED_XCR0_LO		KVM_X86_CPU_PROPERTY(0xd,  0, EAX,  0, 31)
270 #define X86_PROPERTY_XSTATE_MAX_SIZE_XCR0	KVM_X86_CPU_PROPERTY(0xd,  0, EBX,  0, 31)
271 #define X86_PROPERTY_XSTATE_MAX_SIZE		KVM_X86_CPU_PROPERTY(0xd,  0, ECX,  0, 31)
272 #define X86_PROPERTY_SUPPORTED_XCR0_HI		KVM_X86_CPU_PROPERTY(0xd,  0, EDX,  0, 31)
273 
274 #define X86_PROPERTY_XSTATE_TILE_SIZE		KVM_X86_CPU_PROPERTY(0xd, 18, EAX,  0, 31)
275 #define X86_PROPERTY_XSTATE_TILE_OFFSET		KVM_X86_CPU_PROPERTY(0xd, 18, EBX,  0, 31)
276 #define X86_PROPERTY_AMX_MAX_PALETTE_TABLES	KVM_X86_CPU_PROPERTY(0x1d, 0, EAX,  0, 31)
277 #define X86_PROPERTY_AMX_TOTAL_TILE_BYTES	KVM_X86_CPU_PROPERTY(0x1d, 1, EAX,  0, 15)
278 #define X86_PROPERTY_AMX_BYTES_PER_TILE		KVM_X86_CPU_PROPERTY(0x1d, 1, EAX, 16, 31)
279 #define X86_PROPERTY_AMX_BYTES_PER_ROW		KVM_X86_CPU_PROPERTY(0x1d, 1, EBX, 0,  15)
280 #define X86_PROPERTY_AMX_NR_TILE_REGS		KVM_X86_CPU_PROPERTY(0x1d, 1, EBX, 16, 31)
281 #define X86_PROPERTY_AMX_MAX_ROWS		KVM_X86_CPU_PROPERTY(0x1d, 1, ECX, 0,  15)
282 
283 #define X86_PROPERTY_MAX_KVM_LEAF		KVM_X86_CPU_PROPERTY(0x40000000, 0, EAX, 0, 31)
284 
285 #define X86_PROPERTY_MAX_EXT_LEAF		KVM_X86_CPU_PROPERTY(0x80000000, 0, EAX, 0, 31)
286 #define X86_PROPERTY_MAX_PHY_ADDR		KVM_X86_CPU_PROPERTY(0x80000008, 0, EAX, 0, 7)
287 #define X86_PROPERTY_MAX_VIRT_ADDR		KVM_X86_CPU_PROPERTY(0x80000008, 0, EAX, 8, 15)
288 #define X86_PROPERTY_GUEST_MAX_PHY_ADDR		KVM_X86_CPU_PROPERTY(0x80000008, 0, EAX, 16, 23)
289 #define X86_PROPERTY_SEV_C_BIT			KVM_X86_CPU_PROPERTY(0x8000001F, 0, EBX, 0, 5)
290 #define X86_PROPERTY_PHYS_ADDR_REDUCTION	KVM_X86_CPU_PROPERTY(0x8000001F, 0, EBX, 6, 11)
291 #define X86_PROPERTY_NR_PERFCTR_CORE		KVM_X86_CPU_PROPERTY(0x80000022, 0, EBX, 0, 3)
292 #define X86_PROPERTY_NR_PERFCTR_NB		KVM_X86_CPU_PROPERTY(0x80000022, 0, EBX, 10, 15)
293 
294 #define X86_PROPERTY_MAX_CENTAUR_LEAF		KVM_X86_CPU_PROPERTY(0xC0000000, 0, EAX, 0, 31)
295 
296 /*
297  * Intel's architectural PMU events are bizarre.  They have a "feature" bit
298  * that indicates the feature is _not_ supported, and a property that states
299  * the length of the bit mask of unsupported features.  A feature is supported
300  * if the size of the bit mask is larger than the "unavailable" bit, and said
301  * bit is not set.  Fixed counters also bizarre enumeration, but inverted from
302  * arch events for general purpose counters.  Fixed counters are supported if a
303  * feature flag is set **OR** the total number of fixed counters is greater
304  * than index of the counter.
305  *
306  * Wrap the events for general purpose and fixed counters to simplify checking
307  * whether or not a given architectural event is supported.
308  */
309 struct kvm_x86_pmu_feature {
310 	struct kvm_x86_cpu_feature f;
311 };
312 #define	KVM_X86_PMU_FEATURE(__reg, __bit)				\
313 ({									\
314 	struct kvm_x86_pmu_feature feature = {				\
315 		.f = KVM_X86_CPU_FEATURE(0xa, 0, __reg, __bit),		\
316 	};								\
317 									\
318 	kvm_static_assert(KVM_CPUID_##__reg == KVM_CPUID_EBX ||		\
319 			  KVM_CPUID_##__reg == KVM_CPUID_ECX);		\
320 	feature;							\
321 })
322 
323 #define X86_PMU_FEATURE_CPU_CYCLES			KVM_X86_PMU_FEATURE(EBX, 0)
324 #define X86_PMU_FEATURE_INSNS_RETIRED			KVM_X86_PMU_FEATURE(EBX, 1)
325 #define X86_PMU_FEATURE_REFERENCE_CYCLES		KVM_X86_PMU_FEATURE(EBX, 2)
326 #define X86_PMU_FEATURE_LLC_REFERENCES			KVM_X86_PMU_FEATURE(EBX, 3)
327 #define X86_PMU_FEATURE_LLC_MISSES			KVM_X86_PMU_FEATURE(EBX, 4)
328 #define X86_PMU_FEATURE_BRANCH_INSNS_RETIRED		KVM_X86_PMU_FEATURE(EBX, 5)
329 #define X86_PMU_FEATURE_BRANCHES_MISPREDICTED		KVM_X86_PMU_FEATURE(EBX, 6)
330 #define X86_PMU_FEATURE_TOPDOWN_SLOTS			KVM_X86_PMU_FEATURE(EBX, 7)
331 
332 #define X86_PMU_FEATURE_INSNS_RETIRED_FIXED		KVM_X86_PMU_FEATURE(ECX, 0)
333 #define X86_PMU_FEATURE_CPU_CYCLES_FIXED		KVM_X86_PMU_FEATURE(ECX, 1)
334 #define X86_PMU_FEATURE_REFERENCE_TSC_CYCLES_FIXED	KVM_X86_PMU_FEATURE(ECX, 2)
335 #define X86_PMU_FEATURE_TOPDOWN_SLOTS_FIXED		KVM_X86_PMU_FEATURE(ECX, 3)
336 
337 static inline unsigned int x86_family(unsigned int eax)
338 {
339 	unsigned int x86;
340 
341 	x86 = (eax >> 8) & 0xf;
342 
343 	if (x86 == 0xf)
344 		x86 += (eax >> 20) & 0xff;
345 
346 	return x86;
347 }
348 
349 static inline unsigned int x86_model(unsigned int eax)
350 {
351 	return ((eax >> 12) & 0xf0) | ((eax >> 4) & 0x0f);
352 }
353 
354 /* Page table bitfield declarations */
355 #define PTE_PRESENT_MASK        BIT_ULL(0)
356 #define PTE_WRITABLE_MASK       BIT_ULL(1)
357 #define PTE_USER_MASK           BIT_ULL(2)
358 #define PTE_ACCESSED_MASK       BIT_ULL(5)
359 #define PTE_DIRTY_MASK          BIT_ULL(6)
360 #define PTE_LARGE_MASK          BIT_ULL(7)
361 #define PTE_GLOBAL_MASK         BIT_ULL(8)
362 #define PTE_NX_MASK             BIT_ULL(63)
363 
364 #define PHYSICAL_PAGE_MASK      GENMASK_ULL(51, 12)
365 
366 #define PAGE_SHIFT		12
367 #define PAGE_SIZE		(1ULL << PAGE_SHIFT)
368 #define PAGE_MASK		(~(PAGE_SIZE-1) & PHYSICAL_PAGE_MASK)
369 
370 #define HUGEPAGE_SHIFT(x)	(PAGE_SHIFT + (((x) - 1) * 9))
371 #define HUGEPAGE_SIZE(x)	(1UL << HUGEPAGE_SHIFT(x))
372 #define HUGEPAGE_MASK(x)	(~(HUGEPAGE_SIZE(x) - 1) & PHYSICAL_PAGE_MASK)
373 
374 #define PTE_GET_PA(pte)		((pte) & PHYSICAL_PAGE_MASK)
375 #define PTE_GET_PFN(pte)        (PTE_GET_PA(pte) >> PAGE_SHIFT)
376 
377 /* General Registers in 64-Bit Mode */
378 struct gpr64_regs {
379 	u64 rax;
380 	u64 rcx;
381 	u64 rdx;
382 	u64 rbx;
383 	u64 rsp;
384 	u64 rbp;
385 	u64 rsi;
386 	u64 rdi;
387 	u64 r8;
388 	u64 r9;
389 	u64 r10;
390 	u64 r11;
391 	u64 r12;
392 	u64 r13;
393 	u64 r14;
394 	u64 r15;
395 };
396 
397 struct desc64 {
398 	uint16_t limit0;
399 	uint16_t base0;
400 	unsigned base1:8, type:4, s:1, dpl:2, p:1;
401 	unsigned limit1:4, avl:1, l:1, db:1, g:1, base2:8;
402 	uint32_t base3;
403 	uint32_t zero1;
404 } __attribute__((packed));
405 
406 struct desc_ptr {
407 	uint16_t size;
408 	uint64_t address;
409 } __attribute__((packed));
410 
411 struct kvm_x86_state {
412 	struct kvm_xsave *xsave;
413 	struct kvm_vcpu_events events;
414 	struct kvm_mp_state mp_state;
415 	struct kvm_regs regs;
416 	struct kvm_xcrs xcrs;
417 	struct kvm_sregs sregs;
418 	struct kvm_debugregs debugregs;
419 	union {
420 		struct kvm_nested_state nested;
421 		char nested_[16384];
422 	};
423 	struct kvm_msrs msrs;
424 };
425 
426 static inline uint64_t get_desc64_base(const struct desc64 *desc)
427 {
428 	return ((uint64_t)desc->base3 << 32) |
429 		(desc->base0 | ((desc->base1) << 16) | ((desc->base2) << 24));
430 }
431 
432 static inline uint64_t rdtsc(void)
433 {
434 	uint32_t eax, edx;
435 	uint64_t tsc_val;
436 	/*
437 	 * The lfence is to wait (on Intel CPUs) until all previous
438 	 * instructions have been executed. If software requires RDTSC to be
439 	 * executed prior to execution of any subsequent instruction, it can
440 	 * execute LFENCE immediately after RDTSC
441 	 */
442 	__asm__ __volatile__("lfence; rdtsc; lfence" : "=a"(eax), "=d"(edx));
443 	tsc_val = ((uint64_t)edx) << 32 | eax;
444 	return tsc_val;
445 }
446 
447 static inline uint64_t rdtscp(uint32_t *aux)
448 {
449 	uint32_t eax, edx;
450 
451 	__asm__ __volatile__("rdtscp" : "=a"(eax), "=d"(edx), "=c"(*aux));
452 	return ((uint64_t)edx) << 32 | eax;
453 }
454 
455 static inline uint64_t rdmsr(uint32_t msr)
456 {
457 	uint32_t a, d;
458 
459 	__asm__ __volatile__("rdmsr" : "=a"(a), "=d"(d) : "c"(msr) : "memory");
460 
461 	return a | ((uint64_t) d << 32);
462 }
463 
464 static inline void wrmsr(uint32_t msr, uint64_t value)
465 {
466 	uint32_t a = value;
467 	uint32_t d = value >> 32;
468 
469 	__asm__ __volatile__("wrmsr" :: "a"(a), "d"(d), "c"(msr) : "memory");
470 }
471 
472 
473 static inline uint16_t inw(uint16_t port)
474 {
475 	uint16_t tmp;
476 
477 	__asm__ __volatile__("in %%dx, %%ax"
478 		: /* output */ "=a" (tmp)
479 		: /* input */ "d" (port));
480 
481 	return tmp;
482 }
483 
484 static inline uint16_t get_es(void)
485 {
486 	uint16_t es;
487 
488 	__asm__ __volatile__("mov %%es, %[es]"
489 			     : /* output */ [es]"=rm"(es));
490 	return es;
491 }
492 
493 static inline uint16_t get_cs(void)
494 {
495 	uint16_t cs;
496 
497 	__asm__ __volatile__("mov %%cs, %[cs]"
498 			     : /* output */ [cs]"=rm"(cs));
499 	return cs;
500 }
501 
502 static inline uint16_t get_ss(void)
503 {
504 	uint16_t ss;
505 
506 	__asm__ __volatile__("mov %%ss, %[ss]"
507 			     : /* output */ [ss]"=rm"(ss));
508 	return ss;
509 }
510 
511 static inline uint16_t get_ds(void)
512 {
513 	uint16_t ds;
514 
515 	__asm__ __volatile__("mov %%ds, %[ds]"
516 			     : /* output */ [ds]"=rm"(ds));
517 	return ds;
518 }
519 
520 static inline uint16_t get_fs(void)
521 {
522 	uint16_t fs;
523 
524 	__asm__ __volatile__("mov %%fs, %[fs]"
525 			     : /* output */ [fs]"=rm"(fs));
526 	return fs;
527 }
528 
529 static inline uint16_t get_gs(void)
530 {
531 	uint16_t gs;
532 
533 	__asm__ __volatile__("mov %%gs, %[gs]"
534 			     : /* output */ [gs]"=rm"(gs));
535 	return gs;
536 }
537 
538 static inline uint16_t get_tr(void)
539 {
540 	uint16_t tr;
541 
542 	__asm__ __volatile__("str %[tr]"
543 			     : /* output */ [tr]"=rm"(tr));
544 	return tr;
545 }
546 
547 static inline uint64_t get_cr0(void)
548 {
549 	uint64_t cr0;
550 
551 	__asm__ __volatile__("mov %%cr0, %[cr0]"
552 			     : /* output */ [cr0]"=r"(cr0));
553 	return cr0;
554 }
555 
556 static inline uint64_t get_cr3(void)
557 {
558 	uint64_t cr3;
559 
560 	__asm__ __volatile__("mov %%cr3, %[cr3]"
561 			     : /* output */ [cr3]"=r"(cr3));
562 	return cr3;
563 }
564 
565 static inline uint64_t get_cr4(void)
566 {
567 	uint64_t cr4;
568 
569 	__asm__ __volatile__("mov %%cr4, %[cr4]"
570 			     : /* output */ [cr4]"=r"(cr4));
571 	return cr4;
572 }
573 
574 static inline void set_cr4(uint64_t val)
575 {
576 	__asm__ __volatile__("mov %0, %%cr4" : : "r" (val) : "memory");
577 }
578 
579 static inline u64 xgetbv(u32 index)
580 {
581 	u32 eax, edx;
582 
583 	__asm__ __volatile__("xgetbv;"
584 		     : "=a" (eax), "=d" (edx)
585 		     : "c" (index));
586 	return eax | ((u64)edx << 32);
587 }
588 
589 static inline void xsetbv(u32 index, u64 value)
590 {
591 	u32 eax = value;
592 	u32 edx = value >> 32;
593 
594 	__asm__ __volatile__("xsetbv" :: "a" (eax), "d" (edx), "c" (index));
595 }
596 
597 static inline void wrpkru(u32 pkru)
598 {
599 	/* Note, ECX and EDX are architecturally required to be '0'. */
600 	asm volatile(".byte 0x0f,0x01,0xef\n\t"
601 		     : : "a" (pkru), "c"(0), "d"(0));
602 }
603 
604 static inline struct desc_ptr get_gdt(void)
605 {
606 	struct desc_ptr gdt;
607 	__asm__ __volatile__("sgdt %[gdt]"
608 			     : /* output */ [gdt]"=m"(gdt));
609 	return gdt;
610 }
611 
612 static inline struct desc_ptr get_idt(void)
613 {
614 	struct desc_ptr idt;
615 	__asm__ __volatile__("sidt %[idt]"
616 			     : /* output */ [idt]"=m"(idt));
617 	return idt;
618 }
619 
620 static inline void outl(uint16_t port, uint32_t value)
621 {
622 	__asm__ __volatile__("outl %%eax, %%dx" : : "d"(port), "a"(value));
623 }
624 
625 static inline void __cpuid(uint32_t function, uint32_t index,
626 			   uint32_t *eax, uint32_t *ebx,
627 			   uint32_t *ecx, uint32_t *edx)
628 {
629 	*eax = function;
630 	*ecx = index;
631 
632 	asm volatile("cpuid"
633 	    : "=a" (*eax),
634 	      "=b" (*ebx),
635 	      "=c" (*ecx),
636 	      "=d" (*edx)
637 	    : "0" (*eax), "2" (*ecx)
638 	    : "memory");
639 }
640 
641 static inline void cpuid(uint32_t function,
642 			 uint32_t *eax, uint32_t *ebx,
643 			 uint32_t *ecx, uint32_t *edx)
644 {
645 	return __cpuid(function, 0, eax, ebx, ecx, edx);
646 }
647 
648 static inline uint32_t this_cpu_fms(void)
649 {
650 	uint32_t eax, ebx, ecx, edx;
651 
652 	cpuid(1, &eax, &ebx, &ecx, &edx);
653 	return eax;
654 }
655 
656 static inline uint32_t this_cpu_family(void)
657 {
658 	return x86_family(this_cpu_fms());
659 }
660 
661 static inline uint32_t this_cpu_model(void)
662 {
663 	return x86_model(this_cpu_fms());
664 }
665 
666 static inline bool this_cpu_vendor_string_is(const char *vendor)
667 {
668 	const uint32_t *chunk = (const uint32_t *)vendor;
669 	uint32_t eax, ebx, ecx, edx;
670 
671 	cpuid(0, &eax, &ebx, &ecx, &edx);
672 	return (ebx == chunk[0] && edx == chunk[1] && ecx == chunk[2]);
673 }
674 
675 static inline bool this_cpu_is_intel(void)
676 {
677 	return this_cpu_vendor_string_is("GenuineIntel");
678 }
679 
680 /*
681  * Exclude early K5 samples with a vendor string of "AMDisbetter!"
682  */
683 static inline bool this_cpu_is_amd(void)
684 {
685 	return this_cpu_vendor_string_is("AuthenticAMD");
686 }
687 
688 static inline uint32_t __this_cpu_has(uint32_t function, uint32_t index,
689 				      uint8_t reg, uint8_t lo, uint8_t hi)
690 {
691 	uint32_t gprs[4];
692 
693 	__cpuid(function, index,
694 		&gprs[KVM_CPUID_EAX], &gprs[KVM_CPUID_EBX],
695 		&gprs[KVM_CPUID_ECX], &gprs[KVM_CPUID_EDX]);
696 
697 	return (gprs[reg] & GENMASK(hi, lo)) >> lo;
698 }
699 
700 static inline bool this_cpu_has(struct kvm_x86_cpu_feature feature)
701 {
702 	return __this_cpu_has(feature.function, feature.index,
703 			      feature.reg, feature.bit, feature.bit);
704 }
705 
706 static inline uint32_t this_cpu_property(struct kvm_x86_cpu_property property)
707 {
708 	return __this_cpu_has(property.function, property.index,
709 			      property.reg, property.lo_bit, property.hi_bit);
710 }
711 
712 static __always_inline bool this_cpu_has_p(struct kvm_x86_cpu_property property)
713 {
714 	uint32_t max_leaf;
715 
716 	switch (property.function & 0xc0000000) {
717 	case 0:
718 		max_leaf = this_cpu_property(X86_PROPERTY_MAX_BASIC_LEAF);
719 		break;
720 	case 0x40000000:
721 		max_leaf = this_cpu_property(X86_PROPERTY_MAX_KVM_LEAF);
722 		break;
723 	case 0x80000000:
724 		max_leaf = this_cpu_property(X86_PROPERTY_MAX_EXT_LEAF);
725 		break;
726 	case 0xc0000000:
727 		max_leaf = this_cpu_property(X86_PROPERTY_MAX_CENTAUR_LEAF);
728 	}
729 	return max_leaf >= property.function;
730 }
731 
732 static inline bool this_pmu_has(struct kvm_x86_pmu_feature feature)
733 {
734 	uint32_t nr_bits;
735 
736 	if (feature.f.reg == KVM_CPUID_EBX) {
737 		nr_bits = this_cpu_property(X86_PROPERTY_PMU_EBX_BIT_VECTOR_LENGTH);
738 		return nr_bits > feature.f.bit && !this_cpu_has(feature.f);
739 	}
740 
741 	GUEST_ASSERT(feature.f.reg == KVM_CPUID_ECX);
742 	nr_bits = this_cpu_property(X86_PROPERTY_PMU_NR_FIXED_COUNTERS);
743 	return nr_bits > feature.f.bit || this_cpu_has(feature.f);
744 }
745 
746 static __always_inline uint64_t this_cpu_supported_xcr0(void)
747 {
748 	if (!this_cpu_has_p(X86_PROPERTY_SUPPORTED_XCR0_LO))
749 		return 0;
750 
751 	return this_cpu_property(X86_PROPERTY_SUPPORTED_XCR0_LO) |
752 	       ((uint64_t)this_cpu_property(X86_PROPERTY_SUPPORTED_XCR0_HI) << 32);
753 }
754 
755 typedef u32		__attribute__((vector_size(16))) sse128_t;
756 #define __sse128_u	union { sse128_t vec; u64 as_u64[2]; u32 as_u32[4]; }
757 #define sse128_lo(x)	({ __sse128_u t; t.vec = x; t.as_u64[0]; })
758 #define sse128_hi(x)	({ __sse128_u t; t.vec = x; t.as_u64[1]; })
759 
760 static inline void read_sse_reg(int reg, sse128_t *data)
761 {
762 	switch (reg) {
763 	case 0:
764 		asm("movdqa %%xmm0, %0" : "=m"(*data));
765 		break;
766 	case 1:
767 		asm("movdqa %%xmm1, %0" : "=m"(*data));
768 		break;
769 	case 2:
770 		asm("movdqa %%xmm2, %0" : "=m"(*data));
771 		break;
772 	case 3:
773 		asm("movdqa %%xmm3, %0" : "=m"(*data));
774 		break;
775 	case 4:
776 		asm("movdqa %%xmm4, %0" : "=m"(*data));
777 		break;
778 	case 5:
779 		asm("movdqa %%xmm5, %0" : "=m"(*data));
780 		break;
781 	case 6:
782 		asm("movdqa %%xmm6, %0" : "=m"(*data));
783 		break;
784 	case 7:
785 		asm("movdqa %%xmm7, %0" : "=m"(*data));
786 		break;
787 	default:
788 		BUG();
789 	}
790 }
791 
792 static inline void write_sse_reg(int reg, const sse128_t *data)
793 {
794 	switch (reg) {
795 	case 0:
796 		asm("movdqa %0, %%xmm0" : : "m"(*data));
797 		break;
798 	case 1:
799 		asm("movdqa %0, %%xmm1" : : "m"(*data));
800 		break;
801 	case 2:
802 		asm("movdqa %0, %%xmm2" : : "m"(*data));
803 		break;
804 	case 3:
805 		asm("movdqa %0, %%xmm3" : : "m"(*data));
806 		break;
807 	case 4:
808 		asm("movdqa %0, %%xmm4" : : "m"(*data));
809 		break;
810 	case 5:
811 		asm("movdqa %0, %%xmm5" : : "m"(*data));
812 		break;
813 	case 6:
814 		asm("movdqa %0, %%xmm6" : : "m"(*data));
815 		break;
816 	case 7:
817 		asm("movdqa %0, %%xmm7" : : "m"(*data));
818 		break;
819 	default:
820 		BUG();
821 	}
822 }
823 
824 static inline void cpu_relax(void)
825 {
826 	asm volatile("rep; nop" ::: "memory");
827 }
828 
829 static inline void udelay(unsigned long usec)
830 {
831 	uint64_t start, now, cycles;
832 
833 	GUEST_ASSERT(guest_tsc_khz);
834 	cycles = guest_tsc_khz / 1000 * usec;
835 
836 	/*
837 	 * Deliberately don't PAUSE, a.k.a. cpu_relax(), so that the delay is
838 	 * as accurate as possible, e.g. doesn't trigger PAUSE-Loop VM-Exits.
839 	 */
840 	start = rdtsc();
841 	do {
842 		now = rdtsc();
843 	} while (now - start < cycles);
844 }
845 
846 #define ud2()			\
847 	__asm__ __volatile__(	\
848 		"ud2\n"	\
849 		)
850 
851 #define hlt()			\
852 	__asm__ __volatile__(	\
853 		"hlt\n"	\
854 		)
855 
856 struct kvm_x86_state *vcpu_save_state(struct kvm_vcpu *vcpu);
857 void vcpu_load_state(struct kvm_vcpu *vcpu, struct kvm_x86_state *state);
858 void kvm_x86_state_cleanup(struct kvm_x86_state *state);
859 
860 const struct kvm_msr_list *kvm_get_msr_index_list(void);
861 const struct kvm_msr_list *kvm_get_feature_msr_index_list(void);
862 bool kvm_msr_is_in_save_restore_list(uint32_t msr_index);
863 uint64_t kvm_get_feature_msr(uint64_t msr_index);
864 
865 static inline void vcpu_msrs_get(struct kvm_vcpu *vcpu,
866 				 struct kvm_msrs *msrs)
867 {
868 	int r = __vcpu_ioctl(vcpu, KVM_GET_MSRS, msrs);
869 
870 	TEST_ASSERT(r == msrs->nmsrs,
871 		    "KVM_GET_MSRS failed, r: %i (failed on MSR %x)",
872 		    r, r < 0 || r >= msrs->nmsrs ? -1 : msrs->entries[r].index);
873 }
874 static inline void vcpu_msrs_set(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs)
875 {
876 	int r = __vcpu_ioctl(vcpu, KVM_SET_MSRS, msrs);
877 
878 	TEST_ASSERT(r == msrs->nmsrs,
879 		    "KVM_SET_MSRS failed, r: %i (failed on MSR %x)",
880 		    r, r < 0 || r >= msrs->nmsrs ? -1 : msrs->entries[r].index);
881 }
882 static inline void vcpu_debugregs_get(struct kvm_vcpu *vcpu,
883 				      struct kvm_debugregs *debugregs)
884 {
885 	vcpu_ioctl(vcpu, KVM_GET_DEBUGREGS, debugregs);
886 }
887 static inline void vcpu_debugregs_set(struct kvm_vcpu *vcpu,
888 				      struct kvm_debugregs *debugregs)
889 {
890 	vcpu_ioctl(vcpu, KVM_SET_DEBUGREGS, debugregs);
891 }
892 static inline void vcpu_xsave_get(struct kvm_vcpu *vcpu,
893 				  struct kvm_xsave *xsave)
894 {
895 	vcpu_ioctl(vcpu, KVM_GET_XSAVE, xsave);
896 }
897 static inline void vcpu_xsave2_get(struct kvm_vcpu *vcpu,
898 				   struct kvm_xsave *xsave)
899 {
900 	vcpu_ioctl(vcpu, KVM_GET_XSAVE2, xsave);
901 }
902 static inline void vcpu_xsave_set(struct kvm_vcpu *vcpu,
903 				  struct kvm_xsave *xsave)
904 {
905 	vcpu_ioctl(vcpu, KVM_SET_XSAVE, xsave);
906 }
907 static inline void vcpu_xcrs_get(struct kvm_vcpu *vcpu,
908 				 struct kvm_xcrs *xcrs)
909 {
910 	vcpu_ioctl(vcpu, KVM_GET_XCRS, xcrs);
911 }
912 static inline void vcpu_xcrs_set(struct kvm_vcpu *vcpu, struct kvm_xcrs *xcrs)
913 {
914 	vcpu_ioctl(vcpu, KVM_SET_XCRS, xcrs);
915 }
916 
917 const struct kvm_cpuid_entry2 *get_cpuid_entry(const struct kvm_cpuid2 *cpuid,
918 					       uint32_t function, uint32_t index);
919 const struct kvm_cpuid2 *kvm_get_supported_cpuid(void);
920 
921 static inline uint32_t kvm_cpu_fms(void)
922 {
923 	return get_cpuid_entry(kvm_get_supported_cpuid(), 0x1, 0)->eax;
924 }
925 
926 static inline uint32_t kvm_cpu_family(void)
927 {
928 	return x86_family(kvm_cpu_fms());
929 }
930 
931 static inline uint32_t kvm_cpu_model(void)
932 {
933 	return x86_model(kvm_cpu_fms());
934 }
935 
936 bool kvm_cpuid_has(const struct kvm_cpuid2 *cpuid,
937 		   struct kvm_x86_cpu_feature feature);
938 
939 static inline bool kvm_cpu_has(struct kvm_x86_cpu_feature feature)
940 {
941 	return kvm_cpuid_has(kvm_get_supported_cpuid(), feature);
942 }
943 
944 uint32_t kvm_cpuid_property(const struct kvm_cpuid2 *cpuid,
945 			    struct kvm_x86_cpu_property property);
946 
947 static inline uint32_t kvm_cpu_property(struct kvm_x86_cpu_property property)
948 {
949 	return kvm_cpuid_property(kvm_get_supported_cpuid(), property);
950 }
951 
952 static __always_inline bool kvm_cpu_has_p(struct kvm_x86_cpu_property property)
953 {
954 	uint32_t max_leaf;
955 
956 	switch (property.function & 0xc0000000) {
957 	case 0:
958 		max_leaf = kvm_cpu_property(X86_PROPERTY_MAX_BASIC_LEAF);
959 		break;
960 	case 0x40000000:
961 		max_leaf = kvm_cpu_property(X86_PROPERTY_MAX_KVM_LEAF);
962 		break;
963 	case 0x80000000:
964 		max_leaf = kvm_cpu_property(X86_PROPERTY_MAX_EXT_LEAF);
965 		break;
966 	case 0xc0000000:
967 		max_leaf = kvm_cpu_property(X86_PROPERTY_MAX_CENTAUR_LEAF);
968 	}
969 	return max_leaf >= property.function;
970 }
971 
972 static inline bool kvm_pmu_has(struct kvm_x86_pmu_feature feature)
973 {
974 	uint32_t nr_bits;
975 
976 	if (feature.f.reg == KVM_CPUID_EBX) {
977 		nr_bits = kvm_cpu_property(X86_PROPERTY_PMU_EBX_BIT_VECTOR_LENGTH);
978 		return nr_bits > feature.f.bit && !kvm_cpu_has(feature.f);
979 	}
980 
981 	TEST_ASSERT_EQ(feature.f.reg, KVM_CPUID_ECX);
982 	nr_bits = kvm_cpu_property(X86_PROPERTY_PMU_NR_FIXED_COUNTERS);
983 	return nr_bits > feature.f.bit || kvm_cpu_has(feature.f);
984 }
985 
986 static __always_inline uint64_t kvm_cpu_supported_xcr0(void)
987 {
988 	if (!kvm_cpu_has_p(X86_PROPERTY_SUPPORTED_XCR0_LO))
989 		return 0;
990 
991 	return kvm_cpu_property(X86_PROPERTY_SUPPORTED_XCR0_LO) |
992 	       ((uint64_t)kvm_cpu_property(X86_PROPERTY_SUPPORTED_XCR0_HI) << 32);
993 }
994 
995 static inline size_t kvm_cpuid2_size(int nr_entries)
996 {
997 	return sizeof(struct kvm_cpuid2) +
998 	       sizeof(struct kvm_cpuid_entry2) * nr_entries;
999 }
1000 
1001 /*
1002  * Allocate a "struct kvm_cpuid2* instance, with the 0-length arrary of
1003  * entries sized to hold @nr_entries.  The caller is responsible for freeing
1004  * the struct.
1005  */
1006 static inline struct kvm_cpuid2 *allocate_kvm_cpuid2(int nr_entries)
1007 {
1008 	struct kvm_cpuid2 *cpuid;
1009 
1010 	cpuid = malloc(kvm_cpuid2_size(nr_entries));
1011 	TEST_ASSERT(cpuid, "-ENOMEM when allocating kvm_cpuid2");
1012 
1013 	cpuid->nent = nr_entries;
1014 
1015 	return cpuid;
1016 }
1017 
1018 void vcpu_init_cpuid(struct kvm_vcpu *vcpu, const struct kvm_cpuid2 *cpuid);
1019 
1020 static inline struct kvm_cpuid_entry2 *__vcpu_get_cpuid_entry(struct kvm_vcpu *vcpu,
1021 							      uint32_t function,
1022 							      uint32_t index)
1023 {
1024 	return (struct kvm_cpuid_entry2 *)get_cpuid_entry(vcpu->cpuid,
1025 							  function, index);
1026 }
1027 
1028 static inline struct kvm_cpuid_entry2 *vcpu_get_cpuid_entry(struct kvm_vcpu *vcpu,
1029 							    uint32_t function)
1030 {
1031 	return __vcpu_get_cpuid_entry(vcpu, function, 0);
1032 }
1033 
1034 static inline int __vcpu_set_cpuid(struct kvm_vcpu *vcpu)
1035 {
1036 	int r;
1037 
1038 	TEST_ASSERT(vcpu->cpuid, "Must do vcpu_init_cpuid() first");
1039 	r = __vcpu_ioctl(vcpu, KVM_SET_CPUID2, vcpu->cpuid);
1040 	if (r)
1041 		return r;
1042 
1043 	/* On success, refresh the cache to pick up adjustments made by KVM. */
1044 	vcpu_ioctl(vcpu, KVM_GET_CPUID2, vcpu->cpuid);
1045 	return 0;
1046 }
1047 
1048 static inline void vcpu_set_cpuid(struct kvm_vcpu *vcpu)
1049 {
1050 	TEST_ASSERT(vcpu->cpuid, "Must do vcpu_init_cpuid() first");
1051 	vcpu_ioctl(vcpu, KVM_SET_CPUID2, vcpu->cpuid);
1052 
1053 	/* Refresh the cache to pick up adjustments made by KVM. */
1054 	vcpu_ioctl(vcpu, KVM_GET_CPUID2, vcpu->cpuid);
1055 }
1056 
1057 static inline void vcpu_get_cpuid(struct kvm_vcpu *vcpu)
1058 {
1059 	vcpu_ioctl(vcpu, KVM_GET_CPUID2, vcpu->cpuid);
1060 }
1061 
1062 void vcpu_set_cpuid_property(struct kvm_vcpu *vcpu,
1063 			     struct kvm_x86_cpu_property property,
1064 			     uint32_t value);
1065 void vcpu_set_cpuid_maxphyaddr(struct kvm_vcpu *vcpu, uint8_t maxphyaddr);
1066 
1067 void vcpu_clear_cpuid_entry(struct kvm_vcpu *vcpu, uint32_t function);
1068 
1069 static inline bool vcpu_cpuid_has(struct kvm_vcpu *vcpu,
1070 				  struct kvm_x86_cpu_feature feature)
1071 {
1072 	struct kvm_cpuid_entry2 *entry;
1073 
1074 	entry = __vcpu_get_cpuid_entry(vcpu, feature.function, feature.index);
1075 	return *((&entry->eax) + feature.reg) & BIT(feature.bit);
1076 }
1077 
1078 void vcpu_set_or_clear_cpuid_feature(struct kvm_vcpu *vcpu,
1079 				     struct kvm_x86_cpu_feature feature,
1080 				     bool set);
1081 
1082 static inline void vcpu_set_cpuid_feature(struct kvm_vcpu *vcpu,
1083 					  struct kvm_x86_cpu_feature feature)
1084 {
1085 	vcpu_set_or_clear_cpuid_feature(vcpu, feature, true);
1086 
1087 }
1088 
1089 static inline void vcpu_clear_cpuid_feature(struct kvm_vcpu *vcpu,
1090 					    struct kvm_x86_cpu_feature feature)
1091 {
1092 	vcpu_set_or_clear_cpuid_feature(vcpu, feature, false);
1093 }
1094 
1095 uint64_t vcpu_get_msr(struct kvm_vcpu *vcpu, uint64_t msr_index);
1096 int _vcpu_set_msr(struct kvm_vcpu *vcpu, uint64_t msr_index, uint64_t msr_value);
1097 
1098 /*
1099  * Assert on an MSR access(es) and pretty print the MSR name when possible.
1100  * Note, the caller provides the stringified name so that the name of macro is
1101  * printed, not the value the macro resolves to (due to macro expansion).
1102  */
1103 #define TEST_ASSERT_MSR(cond, fmt, msr, str, args...)				\
1104 do {										\
1105 	if (__builtin_constant_p(msr)) {					\
1106 		TEST_ASSERT(cond, fmt, str, args);				\
1107 	} else if (!(cond)) {							\
1108 		char buf[16];							\
1109 										\
1110 		snprintf(buf, sizeof(buf), "MSR 0x%x", msr);			\
1111 		TEST_ASSERT(cond, fmt, buf, args);				\
1112 	}									\
1113 } while (0)
1114 
1115 /*
1116  * Returns true if KVM should return the last written value when reading an MSR
1117  * from userspace, e.g. the MSR isn't a command MSR, doesn't emulate state that
1118  * is changing, etc.  This is NOT an exhaustive list!  The intent is to filter
1119  * out MSRs that are not durable _and_ that a selftest wants to write.
1120  */
1121 static inline bool is_durable_msr(uint32_t msr)
1122 {
1123 	return msr != MSR_IA32_TSC;
1124 }
1125 
1126 #define vcpu_set_msr(vcpu, msr, val)							\
1127 do {											\
1128 	uint64_t r, v = val;								\
1129 											\
1130 	TEST_ASSERT_MSR(_vcpu_set_msr(vcpu, msr, v) == 1,				\
1131 			"KVM_SET_MSRS failed on %s, value = 0x%lx", msr, #msr, v);	\
1132 	if (!is_durable_msr(msr))							\
1133 		break;									\
1134 	r = vcpu_get_msr(vcpu, msr);							\
1135 	TEST_ASSERT_MSR(r == v, "Set %s to '0x%lx', got back '0x%lx'", msr, #msr, v, r);\
1136 } while (0)
1137 
1138 void kvm_get_cpu_address_width(unsigned int *pa_bits, unsigned int *va_bits);
1139 void kvm_init_vm_address_properties(struct kvm_vm *vm);
1140 bool vm_is_unrestricted_guest(struct kvm_vm *vm);
1141 
1142 struct ex_regs {
1143 	uint64_t rax, rcx, rdx, rbx;
1144 	uint64_t rbp, rsi, rdi;
1145 	uint64_t r8, r9, r10, r11;
1146 	uint64_t r12, r13, r14, r15;
1147 	uint64_t vector;
1148 	uint64_t error_code;
1149 	uint64_t rip;
1150 	uint64_t cs;
1151 	uint64_t rflags;
1152 };
1153 
1154 struct idt_entry {
1155 	uint16_t offset0;
1156 	uint16_t selector;
1157 	uint16_t ist : 3;
1158 	uint16_t : 5;
1159 	uint16_t type : 4;
1160 	uint16_t : 1;
1161 	uint16_t dpl : 2;
1162 	uint16_t p : 1;
1163 	uint16_t offset1;
1164 	uint32_t offset2; uint32_t reserved;
1165 };
1166 
1167 void vm_install_exception_handler(struct kvm_vm *vm, int vector,
1168 			void (*handler)(struct ex_regs *));
1169 
1170 /* If a toddler were to say "abracadabra". */
1171 #define KVM_EXCEPTION_MAGIC 0xabacadabaULL
1172 
1173 /*
1174  * KVM selftest exception fixup uses registers to coordinate with the exception
1175  * handler, versus the kernel's in-memory tables and KVM-Unit-Tests's in-memory
1176  * per-CPU data.  Using only registers avoids having to map memory into the
1177  * guest, doesn't require a valid, stable GS.base, and reduces the risk of
1178  * for recursive faults when accessing memory in the handler.  The downside to
1179  * using registers is that it restricts what registers can be used by the actual
1180  * instruction.  But, selftests are 64-bit only, making register* pressure a
1181  * minor concern.  Use r9-r11 as they are volatile, i.e. don't need to be saved
1182  * by the callee, and except for r11 are not implicit parameters to any
1183  * instructions.  Ideally, fixup would use r8-r10 and thus avoid implicit
1184  * parameters entirely, but Hyper-V's hypercall ABI uses r8 and testing Hyper-V
1185  * is higher priority than testing non-faulting SYSCALL/SYSRET.
1186  *
1187  * Note, the fixup handler deliberately does not handle #DE, i.e. the vector
1188  * is guaranteed to be non-zero on fault.
1189  *
1190  * REGISTER INPUTS:
1191  * r9  = MAGIC
1192  * r10 = RIP
1193  * r11 = new RIP on fault
1194  *
1195  * REGISTER OUTPUTS:
1196  * r9  = exception vector (non-zero)
1197  * r10 = error code
1198  */
1199 #define __KVM_ASM_SAFE(insn, fep)				\
1200 	"mov $" __stringify(KVM_EXCEPTION_MAGIC) ", %%r9\n\t"	\
1201 	"lea 1f(%%rip), %%r10\n\t"				\
1202 	"lea 2f(%%rip), %%r11\n\t"				\
1203 	fep "1: " insn "\n\t"					\
1204 	"xor %%r9, %%r9\n\t"					\
1205 	"2:\n\t"						\
1206 	"mov  %%r9b, %[vector]\n\t"				\
1207 	"mov  %%r10, %[error_code]\n\t"
1208 
1209 #define KVM_ASM_SAFE(insn) __KVM_ASM_SAFE(insn, "")
1210 #define KVM_ASM_SAFE_FEP(insn) __KVM_ASM_SAFE(insn, KVM_FEP)
1211 
1212 #define KVM_ASM_SAFE_OUTPUTS(v, ec)	[vector] "=qm"(v), [error_code] "=rm"(ec)
1213 #define KVM_ASM_SAFE_CLOBBERS	"r9", "r10", "r11"
1214 
1215 #define kvm_asm_safe(insn, inputs...)					\
1216 ({									\
1217 	uint64_t ign_error_code;					\
1218 	uint8_t vector;							\
1219 									\
1220 	asm volatile(KVM_ASM_SAFE(insn)					\
1221 		     : KVM_ASM_SAFE_OUTPUTS(vector, ign_error_code)	\
1222 		     : inputs						\
1223 		     : KVM_ASM_SAFE_CLOBBERS);				\
1224 	vector;								\
1225 })
1226 
1227 #define kvm_asm_safe_ec(insn, error_code, inputs...)			\
1228 ({									\
1229 	uint8_t vector;							\
1230 									\
1231 	asm volatile(KVM_ASM_SAFE(insn)					\
1232 		     : KVM_ASM_SAFE_OUTPUTS(vector, error_code)		\
1233 		     : inputs						\
1234 		     : KVM_ASM_SAFE_CLOBBERS);				\
1235 	vector;								\
1236 })
1237 
1238 #define kvm_asm_safe_fep(insn, inputs...)				\
1239 ({									\
1240 	uint64_t ign_error_code;					\
1241 	uint8_t vector;							\
1242 									\
1243 	asm volatile(KVM_ASM_SAFE(insn)					\
1244 		     : KVM_ASM_SAFE_OUTPUTS(vector, ign_error_code)	\
1245 		     : inputs						\
1246 		     : KVM_ASM_SAFE_CLOBBERS);				\
1247 	vector;								\
1248 })
1249 
1250 #define kvm_asm_safe_ec_fep(insn, error_code, inputs...)		\
1251 ({									\
1252 	uint8_t vector;							\
1253 									\
1254 	asm volatile(KVM_ASM_SAFE_FEP(insn)				\
1255 		     : KVM_ASM_SAFE_OUTPUTS(vector, error_code)		\
1256 		     : inputs						\
1257 		     : KVM_ASM_SAFE_CLOBBERS);				\
1258 	vector;								\
1259 })
1260 
1261 #define BUILD_READ_U64_SAFE_HELPER(insn, _fep, _FEP)			\
1262 static inline uint8_t insn##_safe ##_fep(uint32_t idx, uint64_t *val)	\
1263 {									\
1264 	uint64_t error_code;						\
1265 	uint8_t vector;							\
1266 	uint32_t a, d;							\
1267 									\
1268 	asm volatile(KVM_ASM_SAFE##_FEP(#insn)				\
1269 		     : "=a"(a), "=d"(d),				\
1270 		       KVM_ASM_SAFE_OUTPUTS(vector, error_code)		\
1271 		     : "c"(idx)						\
1272 		     : KVM_ASM_SAFE_CLOBBERS);				\
1273 									\
1274 	*val = (uint64_t)a | ((uint64_t)d << 32);			\
1275 	return vector;							\
1276 }
1277 
1278 /*
1279  * Generate {insn}_safe() and {insn}_safe_fep() helpers for instructions that
1280  * use ECX as in input index, and EDX:EAX as a 64-bit output.
1281  */
1282 #define BUILD_READ_U64_SAFE_HELPERS(insn)				\
1283 	BUILD_READ_U64_SAFE_HELPER(insn, , )				\
1284 	BUILD_READ_U64_SAFE_HELPER(insn, _fep, _FEP)			\
1285 
1286 BUILD_READ_U64_SAFE_HELPERS(rdmsr)
1287 BUILD_READ_U64_SAFE_HELPERS(rdpmc)
1288 BUILD_READ_U64_SAFE_HELPERS(xgetbv)
1289 
1290 static inline uint8_t wrmsr_safe(uint32_t msr, uint64_t val)
1291 {
1292 	return kvm_asm_safe("wrmsr", "a"(val & -1u), "d"(val >> 32), "c"(msr));
1293 }
1294 
1295 static inline uint8_t xsetbv_safe(uint32_t index, uint64_t value)
1296 {
1297 	u32 eax = value;
1298 	u32 edx = value >> 32;
1299 
1300 	return kvm_asm_safe("xsetbv", "a" (eax), "d" (edx), "c" (index));
1301 }
1302 
1303 bool kvm_is_tdp_enabled(void);
1304 
1305 static inline bool kvm_is_pmu_enabled(void)
1306 {
1307 	return get_kvm_param_bool("enable_pmu");
1308 }
1309 
1310 static inline bool kvm_is_forced_emulation_enabled(void)
1311 {
1312 	return !!get_kvm_param_integer("force_emulation_prefix");
1313 }
1314 
1315 uint64_t *__vm_get_page_table_entry(struct kvm_vm *vm, uint64_t vaddr,
1316 				    int *level);
1317 uint64_t *vm_get_page_table_entry(struct kvm_vm *vm, uint64_t vaddr);
1318 
1319 uint64_t kvm_hypercall(uint64_t nr, uint64_t a0, uint64_t a1, uint64_t a2,
1320 		       uint64_t a3);
1321 uint64_t __xen_hypercall(uint64_t nr, uint64_t a0, void *a1);
1322 void xen_hypercall(uint64_t nr, uint64_t a0, void *a1);
1323 
1324 static inline uint64_t __kvm_hypercall_map_gpa_range(uint64_t gpa,
1325 						     uint64_t size, uint64_t flags)
1326 {
1327 	return kvm_hypercall(KVM_HC_MAP_GPA_RANGE, gpa, size >> PAGE_SHIFT, flags, 0);
1328 }
1329 
1330 static inline void kvm_hypercall_map_gpa_range(uint64_t gpa, uint64_t size,
1331 					       uint64_t flags)
1332 {
1333 	uint64_t ret = __kvm_hypercall_map_gpa_range(gpa, size, flags);
1334 
1335 	GUEST_ASSERT(!ret);
1336 }
1337 
1338 void __vm_xsave_require_permission(uint64_t xfeature, const char *name);
1339 
1340 #define vm_xsave_require_permission(xfeature)	\
1341 	__vm_xsave_require_permission(xfeature, #xfeature)
1342 
1343 enum pg_level {
1344 	PG_LEVEL_NONE,
1345 	PG_LEVEL_4K,
1346 	PG_LEVEL_2M,
1347 	PG_LEVEL_1G,
1348 	PG_LEVEL_512G,
1349 	PG_LEVEL_NUM
1350 };
1351 
1352 #define PG_LEVEL_SHIFT(_level) ((_level - 1) * 9 + 12)
1353 #define PG_LEVEL_SIZE(_level) (1ull << PG_LEVEL_SHIFT(_level))
1354 
1355 #define PG_SIZE_4K PG_LEVEL_SIZE(PG_LEVEL_4K)
1356 #define PG_SIZE_2M PG_LEVEL_SIZE(PG_LEVEL_2M)
1357 #define PG_SIZE_1G PG_LEVEL_SIZE(PG_LEVEL_1G)
1358 
1359 void __virt_pg_map(struct kvm_vm *vm, uint64_t vaddr, uint64_t paddr, int level);
1360 void virt_map_level(struct kvm_vm *vm, uint64_t vaddr, uint64_t paddr,
1361 		    uint64_t nr_bytes, int level);
1362 
1363 /*
1364  * Basic CPU control in CR0
1365  */
1366 #define X86_CR0_PE          (1UL<<0) /* Protection Enable */
1367 #define X86_CR0_MP          (1UL<<1) /* Monitor Coprocessor */
1368 #define X86_CR0_EM          (1UL<<2) /* Emulation */
1369 #define X86_CR0_TS          (1UL<<3) /* Task Switched */
1370 #define X86_CR0_ET          (1UL<<4) /* Extension Type */
1371 #define X86_CR0_NE          (1UL<<5) /* Numeric Error */
1372 #define X86_CR0_WP          (1UL<<16) /* Write Protect */
1373 #define X86_CR0_AM          (1UL<<18) /* Alignment Mask */
1374 #define X86_CR0_NW          (1UL<<29) /* Not Write-through */
1375 #define X86_CR0_CD          (1UL<<30) /* Cache Disable */
1376 #define X86_CR0_PG          (1UL<<31) /* Paging */
1377 
1378 #define PFERR_PRESENT_BIT 0
1379 #define PFERR_WRITE_BIT 1
1380 #define PFERR_USER_BIT 2
1381 #define PFERR_RSVD_BIT 3
1382 #define PFERR_FETCH_BIT 4
1383 #define PFERR_PK_BIT 5
1384 #define PFERR_SGX_BIT 15
1385 #define PFERR_GUEST_FINAL_BIT 32
1386 #define PFERR_GUEST_PAGE_BIT 33
1387 #define PFERR_IMPLICIT_ACCESS_BIT 48
1388 
1389 #define PFERR_PRESENT_MASK	BIT(PFERR_PRESENT_BIT)
1390 #define PFERR_WRITE_MASK	BIT(PFERR_WRITE_BIT)
1391 #define PFERR_USER_MASK		BIT(PFERR_USER_BIT)
1392 #define PFERR_RSVD_MASK		BIT(PFERR_RSVD_BIT)
1393 #define PFERR_FETCH_MASK	BIT(PFERR_FETCH_BIT)
1394 #define PFERR_PK_MASK		BIT(PFERR_PK_BIT)
1395 #define PFERR_SGX_MASK		BIT(PFERR_SGX_BIT)
1396 #define PFERR_GUEST_FINAL_MASK	BIT_ULL(PFERR_GUEST_FINAL_BIT)
1397 #define PFERR_GUEST_PAGE_MASK	BIT_ULL(PFERR_GUEST_PAGE_BIT)
1398 #define PFERR_IMPLICIT_ACCESS	BIT_ULL(PFERR_IMPLICIT_ACCESS_BIT)
1399 
1400 bool sys_clocksource_is_based_on_tsc(void);
1401 
1402 #endif /* SELFTEST_KVM_PROCESSOR_H */
1403