xref: /linux/tools/testing/selftests/kvm/include/x86_64/processor.h (revision eed4edda910fe34dfae8c6bfbcf57f4593a54295)
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 
23 extern bool host_cpu_is_intel;
24 extern bool host_cpu_is_amd;
25 
26 #define NMI_VECTOR		0x02
27 
28 #define X86_EFLAGS_FIXED	 (1u << 1)
29 
30 #define X86_CR4_VME		(1ul << 0)
31 #define X86_CR4_PVI		(1ul << 1)
32 #define X86_CR4_TSD		(1ul << 2)
33 #define X86_CR4_DE		(1ul << 3)
34 #define X86_CR4_PSE		(1ul << 4)
35 #define X86_CR4_PAE		(1ul << 5)
36 #define X86_CR4_MCE		(1ul << 6)
37 #define X86_CR4_PGE		(1ul << 7)
38 #define X86_CR4_PCE		(1ul << 8)
39 #define X86_CR4_OSFXSR		(1ul << 9)
40 #define X86_CR4_OSXMMEXCPT	(1ul << 10)
41 #define X86_CR4_UMIP		(1ul << 11)
42 #define X86_CR4_LA57		(1ul << 12)
43 #define X86_CR4_VMXE		(1ul << 13)
44 #define X86_CR4_SMXE		(1ul << 14)
45 #define X86_CR4_FSGSBASE	(1ul << 16)
46 #define X86_CR4_PCIDE		(1ul << 17)
47 #define X86_CR4_OSXSAVE		(1ul << 18)
48 #define X86_CR4_SMEP		(1ul << 20)
49 #define X86_CR4_SMAP		(1ul << 21)
50 #define X86_CR4_PKE		(1ul << 22)
51 
52 struct xstate_header {
53 	u64				xstate_bv;
54 	u64				xcomp_bv;
55 	u64				reserved[6];
56 } __attribute__((packed));
57 
58 struct xstate {
59 	u8				i387[512];
60 	struct xstate_header		header;
61 	u8				extended_state_area[0];
62 } __attribute__ ((packed, aligned (64)));
63 
64 #define XFEATURE_MASK_FP		BIT_ULL(0)
65 #define XFEATURE_MASK_SSE		BIT_ULL(1)
66 #define XFEATURE_MASK_YMM		BIT_ULL(2)
67 #define XFEATURE_MASK_BNDREGS		BIT_ULL(3)
68 #define XFEATURE_MASK_BNDCSR		BIT_ULL(4)
69 #define XFEATURE_MASK_OPMASK		BIT_ULL(5)
70 #define XFEATURE_MASK_ZMM_Hi256		BIT_ULL(6)
71 #define XFEATURE_MASK_Hi16_ZMM		BIT_ULL(7)
72 #define XFEATURE_MASK_PT		BIT_ULL(8)
73 #define XFEATURE_MASK_PKRU		BIT_ULL(9)
74 #define XFEATURE_MASK_PASID		BIT_ULL(10)
75 #define XFEATURE_MASK_CET_USER		BIT_ULL(11)
76 #define XFEATURE_MASK_CET_KERNEL	BIT_ULL(12)
77 #define XFEATURE_MASK_LBR		BIT_ULL(15)
78 #define XFEATURE_MASK_XTILE_CFG		BIT_ULL(17)
79 #define XFEATURE_MASK_XTILE_DATA	BIT_ULL(18)
80 
81 #define XFEATURE_MASK_AVX512		(XFEATURE_MASK_OPMASK | \
82 					 XFEATURE_MASK_ZMM_Hi256 | \
83 					 XFEATURE_MASK_Hi16_ZMM)
84 #define XFEATURE_MASK_XTILE		(XFEATURE_MASK_XTILE_DATA | \
85 					 XFEATURE_MASK_XTILE_CFG)
86 
87 /* Note, these are ordered alphabetically to match kvm_cpuid_entry2.  Eww. */
88 enum cpuid_output_regs {
89 	KVM_CPUID_EAX,
90 	KVM_CPUID_EBX,
91 	KVM_CPUID_ECX,
92 	KVM_CPUID_EDX
93 };
94 
95 /*
96  * Pack the information into a 64-bit value so that each X86_FEATURE_XXX can be
97  * passed by value with no overhead.
98  */
99 struct kvm_x86_cpu_feature {
100 	u32	function;
101 	u16	index;
102 	u8	reg;
103 	u8	bit;
104 };
105 #define	KVM_X86_CPU_FEATURE(fn, idx, gpr, __bit)				\
106 ({										\
107 	struct kvm_x86_cpu_feature feature = {					\
108 		.function = fn,							\
109 		.index = idx,							\
110 		.reg = KVM_CPUID_##gpr,						\
111 		.bit = __bit,							\
112 	};									\
113 										\
114 	kvm_static_assert((fn & 0xc0000000) == 0 ||				\
115 			  (fn & 0xc0000000) == 0x40000000 ||			\
116 			  (fn & 0xc0000000) == 0x80000000 ||			\
117 			  (fn & 0xc0000000) == 0xc0000000);			\
118 	kvm_static_assert(idx < BIT(sizeof(feature.index) * BITS_PER_BYTE));	\
119 	feature;								\
120 })
121 
122 /*
123  * Basic Leafs, a.k.a. Intel defined
124  */
125 #define	X86_FEATURE_MWAIT		KVM_X86_CPU_FEATURE(0x1, 0, ECX, 3)
126 #define	X86_FEATURE_VMX			KVM_X86_CPU_FEATURE(0x1, 0, ECX, 5)
127 #define	X86_FEATURE_SMX			KVM_X86_CPU_FEATURE(0x1, 0, ECX, 6)
128 #define	X86_FEATURE_PDCM		KVM_X86_CPU_FEATURE(0x1, 0, ECX, 15)
129 #define	X86_FEATURE_PCID		KVM_X86_CPU_FEATURE(0x1, 0, ECX, 17)
130 #define X86_FEATURE_X2APIC		KVM_X86_CPU_FEATURE(0x1, 0, ECX, 21)
131 #define	X86_FEATURE_MOVBE		KVM_X86_CPU_FEATURE(0x1, 0, ECX, 22)
132 #define	X86_FEATURE_TSC_DEADLINE_TIMER	KVM_X86_CPU_FEATURE(0x1, 0, ECX, 24)
133 #define	X86_FEATURE_XSAVE		KVM_X86_CPU_FEATURE(0x1, 0, ECX, 26)
134 #define	X86_FEATURE_OSXSAVE		KVM_X86_CPU_FEATURE(0x1, 0, ECX, 27)
135 #define	X86_FEATURE_RDRAND		KVM_X86_CPU_FEATURE(0x1, 0, ECX, 30)
136 #define	X86_FEATURE_HYPERVISOR		KVM_X86_CPU_FEATURE(0x1, 0, ECX, 31)
137 #define X86_FEATURE_PAE			KVM_X86_CPU_FEATURE(0x1, 0, EDX, 6)
138 #define	X86_FEATURE_MCE			KVM_X86_CPU_FEATURE(0x1, 0, EDX, 7)
139 #define	X86_FEATURE_APIC		KVM_X86_CPU_FEATURE(0x1, 0, EDX, 9)
140 #define	X86_FEATURE_CLFLUSH		KVM_X86_CPU_FEATURE(0x1, 0, EDX, 19)
141 #define	X86_FEATURE_XMM			KVM_X86_CPU_FEATURE(0x1, 0, EDX, 25)
142 #define	X86_FEATURE_XMM2		KVM_X86_CPU_FEATURE(0x1, 0, EDX, 26)
143 #define	X86_FEATURE_FSGSBASE		KVM_X86_CPU_FEATURE(0x7, 0, EBX, 0)
144 #define	X86_FEATURE_TSC_ADJUST		KVM_X86_CPU_FEATURE(0x7, 0, EBX, 1)
145 #define	X86_FEATURE_SGX			KVM_X86_CPU_FEATURE(0x7, 0, EBX, 2)
146 #define	X86_FEATURE_HLE			KVM_X86_CPU_FEATURE(0x7, 0, EBX, 4)
147 #define	X86_FEATURE_SMEP	        KVM_X86_CPU_FEATURE(0x7, 0, EBX, 7)
148 #define	X86_FEATURE_INVPCID		KVM_X86_CPU_FEATURE(0x7, 0, EBX, 10)
149 #define	X86_FEATURE_RTM			KVM_X86_CPU_FEATURE(0x7, 0, EBX, 11)
150 #define	X86_FEATURE_MPX			KVM_X86_CPU_FEATURE(0x7, 0, EBX, 14)
151 #define	X86_FEATURE_SMAP		KVM_X86_CPU_FEATURE(0x7, 0, EBX, 20)
152 #define	X86_FEATURE_PCOMMIT		KVM_X86_CPU_FEATURE(0x7, 0, EBX, 22)
153 #define	X86_FEATURE_CLFLUSHOPT		KVM_X86_CPU_FEATURE(0x7, 0, EBX, 23)
154 #define	X86_FEATURE_CLWB		KVM_X86_CPU_FEATURE(0x7, 0, EBX, 24)
155 #define	X86_FEATURE_UMIP		KVM_X86_CPU_FEATURE(0x7, 0, ECX, 2)
156 #define	X86_FEATURE_PKU			KVM_X86_CPU_FEATURE(0x7, 0, ECX, 3)
157 #define	X86_FEATURE_OSPKE		KVM_X86_CPU_FEATURE(0x7, 0, ECX, 4)
158 #define	X86_FEATURE_LA57		KVM_X86_CPU_FEATURE(0x7, 0, ECX, 16)
159 #define	X86_FEATURE_RDPID		KVM_X86_CPU_FEATURE(0x7, 0, ECX, 22)
160 #define	X86_FEATURE_SGX_LC		KVM_X86_CPU_FEATURE(0x7, 0, ECX, 30)
161 #define	X86_FEATURE_SHSTK		KVM_X86_CPU_FEATURE(0x7, 0, ECX, 7)
162 #define	X86_FEATURE_IBT			KVM_X86_CPU_FEATURE(0x7, 0, EDX, 20)
163 #define	X86_FEATURE_AMX_TILE		KVM_X86_CPU_FEATURE(0x7, 0, EDX, 24)
164 #define	X86_FEATURE_SPEC_CTRL		KVM_X86_CPU_FEATURE(0x7, 0, EDX, 26)
165 #define	X86_FEATURE_ARCH_CAPABILITIES	KVM_X86_CPU_FEATURE(0x7, 0, EDX, 29)
166 #define	X86_FEATURE_PKS			KVM_X86_CPU_FEATURE(0x7, 0, ECX, 31)
167 #define	X86_FEATURE_XTILECFG		KVM_X86_CPU_FEATURE(0xD, 0, EAX, 17)
168 #define	X86_FEATURE_XTILEDATA		KVM_X86_CPU_FEATURE(0xD, 0, EAX, 18)
169 #define	X86_FEATURE_XSAVES		KVM_X86_CPU_FEATURE(0xD, 1, EAX, 3)
170 #define	X86_FEATURE_XFD			KVM_X86_CPU_FEATURE(0xD, 1, EAX, 4)
171 #define X86_FEATURE_XTILEDATA_XFD	KVM_X86_CPU_FEATURE(0xD, 18, ECX, 2)
172 
173 /*
174  * Extended Leafs, a.k.a. AMD defined
175  */
176 #define	X86_FEATURE_SVM			KVM_X86_CPU_FEATURE(0x80000001, 0, ECX, 2)
177 #define	X86_FEATURE_NX			KVM_X86_CPU_FEATURE(0x80000001, 0, EDX, 20)
178 #define	X86_FEATURE_GBPAGES		KVM_X86_CPU_FEATURE(0x80000001, 0, EDX, 26)
179 #define	X86_FEATURE_RDTSCP		KVM_X86_CPU_FEATURE(0x80000001, 0, EDX, 27)
180 #define	X86_FEATURE_LM			KVM_X86_CPU_FEATURE(0x80000001, 0, EDX, 29)
181 #define	X86_FEATURE_INVTSC		KVM_X86_CPU_FEATURE(0x80000007, 0, EDX, 8)
182 #define	X86_FEATURE_RDPRU		KVM_X86_CPU_FEATURE(0x80000008, 0, EBX, 4)
183 #define	X86_FEATURE_AMD_IBPB		KVM_X86_CPU_FEATURE(0x80000008, 0, EBX, 12)
184 #define	X86_FEATURE_NPT			KVM_X86_CPU_FEATURE(0x8000000A, 0, EDX, 0)
185 #define	X86_FEATURE_LBRV		KVM_X86_CPU_FEATURE(0x8000000A, 0, EDX, 1)
186 #define	X86_FEATURE_NRIPS		KVM_X86_CPU_FEATURE(0x8000000A, 0, EDX, 3)
187 #define X86_FEATURE_TSCRATEMSR          KVM_X86_CPU_FEATURE(0x8000000A, 0, EDX, 4)
188 #define X86_FEATURE_PAUSEFILTER         KVM_X86_CPU_FEATURE(0x8000000A, 0, EDX, 10)
189 #define X86_FEATURE_PFTHRESHOLD         KVM_X86_CPU_FEATURE(0x8000000A, 0, EDX, 12)
190 #define	X86_FEATURE_VGIF		KVM_X86_CPU_FEATURE(0x8000000A, 0, EDX, 16)
191 #define X86_FEATURE_SEV			KVM_X86_CPU_FEATURE(0x8000001F, 0, EAX, 1)
192 #define X86_FEATURE_SEV_ES		KVM_X86_CPU_FEATURE(0x8000001F, 0, EAX, 3)
193 
194 /*
195  * KVM defined paravirt features.
196  */
197 #define X86_FEATURE_KVM_CLOCKSOURCE	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 0)
198 #define X86_FEATURE_KVM_NOP_IO_DELAY	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 1)
199 #define X86_FEATURE_KVM_MMU_OP		KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 2)
200 #define X86_FEATURE_KVM_CLOCKSOURCE2	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 3)
201 #define X86_FEATURE_KVM_ASYNC_PF	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 4)
202 #define X86_FEATURE_KVM_STEAL_TIME	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 5)
203 #define X86_FEATURE_KVM_PV_EOI		KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 6)
204 #define X86_FEATURE_KVM_PV_UNHALT	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 7)
205 /* Bit 8 apparently isn't used?!?! */
206 #define X86_FEATURE_KVM_PV_TLB_FLUSH	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 9)
207 #define X86_FEATURE_KVM_ASYNC_PF_VMEXIT	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 10)
208 #define X86_FEATURE_KVM_PV_SEND_IPI	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 11)
209 #define X86_FEATURE_KVM_POLL_CONTROL	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 12)
210 #define X86_FEATURE_KVM_PV_SCHED_YIELD	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 13)
211 #define X86_FEATURE_KVM_ASYNC_PF_INT	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 14)
212 #define X86_FEATURE_KVM_MSI_EXT_DEST_ID	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 15)
213 #define X86_FEATURE_KVM_HC_MAP_GPA_RANGE	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 16)
214 #define X86_FEATURE_KVM_MIGRATION_CONTROL	KVM_X86_CPU_FEATURE(0x40000001, 0, EAX, 17)
215 
216 /*
217  * Same idea as X86_FEATURE_XXX, but X86_PROPERTY_XXX retrieves a multi-bit
218  * value/property as opposed to a single-bit feature.  Again, pack the info
219  * into a 64-bit value to pass by value with no overhead.
220  */
221 struct kvm_x86_cpu_property {
222 	u32	function;
223 	u8	index;
224 	u8	reg;
225 	u8	lo_bit;
226 	u8	hi_bit;
227 };
228 #define	KVM_X86_CPU_PROPERTY(fn, idx, gpr, low_bit, high_bit)			\
229 ({										\
230 	struct kvm_x86_cpu_property property = {				\
231 		.function = fn,							\
232 		.index = idx,							\
233 		.reg = KVM_CPUID_##gpr,						\
234 		.lo_bit = low_bit,						\
235 		.hi_bit = high_bit,						\
236 	};									\
237 										\
238 	kvm_static_assert(low_bit < high_bit);					\
239 	kvm_static_assert((fn & 0xc0000000) == 0 ||				\
240 			  (fn & 0xc0000000) == 0x40000000 ||			\
241 			  (fn & 0xc0000000) == 0x80000000 ||			\
242 			  (fn & 0xc0000000) == 0xc0000000);			\
243 	kvm_static_assert(idx < BIT(sizeof(property.index) * BITS_PER_BYTE));	\
244 	property;								\
245 })
246 
247 #define X86_PROPERTY_MAX_BASIC_LEAF		KVM_X86_CPU_PROPERTY(0, 0, EAX, 0, 31)
248 #define X86_PROPERTY_PMU_VERSION		KVM_X86_CPU_PROPERTY(0xa, 0, EAX, 0, 7)
249 #define X86_PROPERTY_PMU_NR_GP_COUNTERS		KVM_X86_CPU_PROPERTY(0xa, 0, EAX, 8, 15)
250 #define X86_PROPERTY_PMU_GP_COUNTERS_BIT_WIDTH	KVM_X86_CPU_PROPERTY(0xa, 0, EAX, 16, 23)
251 #define X86_PROPERTY_PMU_EBX_BIT_VECTOR_LENGTH	KVM_X86_CPU_PROPERTY(0xa, 0, EAX, 24, 31)
252 #define X86_PROPERTY_PMU_EVENTS_MASK		KVM_X86_CPU_PROPERTY(0xa, 0, EBX, 0, 7)
253 #define X86_PROPERTY_PMU_FIXED_COUNTERS_BITMASK	KVM_X86_CPU_PROPERTY(0xa, 0, ECX, 0, 31)
254 #define X86_PROPERTY_PMU_NR_FIXED_COUNTERS	KVM_X86_CPU_PROPERTY(0xa, 0, EDX, 0, 4)
255 #define X86_PROPERTY_PMU_FIXED_COUNTERS_BIT_WIDTH	KVM_X86_CPU_PROPERTY(0xa, 0, EDX, 5, 12)
256 
257 #define X86_PROPERTY_SUPPORTED_XCR0_LO		KVM_X86_CPU_PROPERTY(0xd,  0, EAX,  0, 31)
258 #define X86_PROPERTY_XSTATE_MAX_SIZE_XCR0	KVM_X86_CPU_PROPERTY(0xd,  0, EBX,  0, 31)
259 #define X86_PROPERTY_XSTATE_MAX_SIZE		KVM_X86_CPU_PROPERTY(0xd,  0, ECX,  0, 31)
260 #define X86_PROPERTY_SUPPORTED_XCR0_HI		KVM_X86_CPU_PROPERTY(0xd,  0, EDX,  0, 31)
261 
262 #define X86_PROPERTY_XSTATE_TILE_SIZE		KVM_X86_CPU_PROPERTY(0xd, 18, EAX,  0, 31)
263 #define X86_PROPERTY_XSTATE_TILE_OFFSET		KVM_X86_CPU_PROPERTY(0xd, 18, EBX,  0, 31)
264 #define X86_PROPERTY_AMX_MAX_PALETTE_TABLES	KVM_X86_CPU_PROPERTY(0x1d, 0, EAX,  0, 31)
265 #define X86_PROPERTY_AMX_TOTAL_TILE_BYTES	KVM_X86_CPU_PROPERTY(0x1d, 1, EAX,  0, 15)
266 #define X86_PROPERTY_AMX_BYTES_PER_TILE		KVM_X86_CPU_PROPERTY(0x1d, 1, EAX, 16, 31)
267 #define X86_PROPERTY_AMX_BYTES_PER_ROW		KVM_X86_CPU_PROPERTY(0x1d, 1, EBX, 0,  15)
268 #define X86_PROPERTY_AMX_NR_TILE_REGS		KVM_X86_CPU_PROPERTY(0x1d, 1, EBX, 16, 31)
269 #define X86_PROPERTY_AMX_MAX_ROWS		KVM_X86_CPU_PROPERTY(0x1d, 1, ECX, 0,  15)
270 
271 #define X86_PROPERTY_MAX_KVM_LEAF		KVM_X86_CPU_PROPERTY(0x40000000, 0, EAX, 0, 31)
272 
273 #define X86_PROPERTY_MAX_EXT_LEAF		KVM_X86_CPU_PROPERTY(0x80000000, 0, EAX, 0, 31)
274 #define X86_PROPERTY_MAX_PHY_ADDR		KVM_X86_CPU_PROPERTY(0x80000008, 0, EAX, 0, 7)
275 #define X86_PROPERTY_MAX_VIRT_ADDR		KVM_X86_CPU_PROPERTY(0x80000008, 0, EAX, 8, 15)
276 #define X86_PROPERTY_PHYS_ADDR_REDUCTION	KVM_X86_CPU_PROPERTY(0x8000001F, 0, EBX, 6, 11)
277 
278 #define X86_PROPERTY_MAX_CENTAUR_LEAF		KVM_X86_CPU_PROPERTY(0xC0000000, 0, EAX, 0, 31)
279 
280 /*
281  * Intel's architectural PMU events are bizarre.  They have a "feature" bit
282  * that indicates the feature is _not_ supported, and a property that states
283  * the length of the bit mask of unsupported features.  A feature is supported
284  * if the size of the bit mask is larger than the "unavailable" bit, and said
285  * bit is not set.
286  *
287  * Wrap the "unavailable" feature to simplify checking whether or not a given
288  * architectural event is supported.
289  */
290 struct kvm_x86_pmu_feature {
291 	struct kvm_x86_cpu_feature anti_feature;
292 };
293 #define	KVM_X86_PMU_FEATURE(name, __bit)					\
294 ({										\
295 	struct kvm_x86_pmu_feature feature = {					\
296 		.anti_feature = KVM_X86_CPU_FEATURE(0xa, 0, EBX, __bit),	\
297 	};									\
298 										\
299 	feature;								\
300 })
301 
302 #define X86_PMU_FEATURE_BRANCH_INSNS_RETIRED	KVM_X86_PMU_FEATURE(BRANCH_INSNS_RETIRED, 5)
303 
304 static inline unsigned int x86_family(unsigned int eax)
305 {
306 	unsigned int x86;
307 
308 	x86 = (eax >> 8) & 0xf;
309 
310 	if (x86 == 0xf)
311 		x86 += (eax >> 20) & 0xff;
312 
313 	return x86;
314 }
315 
316 static inline unsigned int x86_model(unsigned int eax)
317 {
318 	return ((eax >> 12) & 0xf0) | ((eax >> 4) & 0x0f);
319 }
320 
321 /* Page table bitfield declarations */
322 #define PTE_PRESENT_MASK        BIT_ULL(0)
323 #define PTE_WRITABLE_MASK       BIT_ULL(1)
324 #define PTE_USER_MASK           BIT_ULL(2)
325 #define PTE_ACCESSED_MASK       BIT_ULL(5)
326 #define PTE_DIRTY_MASK          BIT_ULL(6)
327 #define PTE_LARGE_MASK          BIT_ULL(7)
328 #define PTE_GLOBAL_MASK         BIT_ULL(8)
329 #define PTE_NX_MASK             BIT_ULL(63)
330 
331 #define PHYSICAL_PAGE_MASK      GENMASK_ULL(51, 12)
332 
333 #define PAGE_SHIFT		12
334 #define PAGE_SIZE		(1ULL << PAGE_SHIFT)
335 #define PAGE_MASK		(~(PAGE_SIZE-1) & PHYSICAL_PAGE_MASK)
336 
337 #define HUGEPAGE_SHIFT(x)	(PAGE_SHIFT + (((x) - 1) * 9))
338 #define HUGEPAGE_SIZE(x)	(1UL << HUGEPAGE_SHIFT(x))
339 #define HUGEPAGE_MASK(x)	(~(HUGEPAGE_SIZE(x) - 1) & PHYSICAL_PAGE_MASK)
340 
341 #define PTE_GET_PA(pte)		((pte) & PHYSICAL_PAGE_MASK)
342 #define PTE_GET_PFN(pte)        (PTE_GET_PA(pte) >> PAGE_SHIFT)
343 
344 /* General Registers in 64-Bit Mode */
345 struct gpr64_regs {
346 	u64 rax;
347 	u64 rcx;
348 	u64 rdx;
349 	u64 rbx;
350 	u64 rsp;
351 	u64 rbp;
352 	u64 rsi;
353 	u64 rdi;
354 	u64 r8;
355 	u64 r9;
356 	u64 r10;
357 	u64 r11;
358 	u64 r12;
359 	u64 r13;
360 	u64 r14;
361 	u64 r15;
362 };
363 
364 struct desc64 {
365 	uint16_t limit0;
366 	uint16_t base0;
367 	unsigned base1:8, type:4, s:1, dpl:2, p:1;
368 	unsigned limit1:4, avl:1, l:1, db:1, g:1, base2:8;
369 	uint32_t base3;
370 	uint32_t zero1;
371 } __attribute__((packed));
372 
373 struct desc_ptr {
374 	uint16_t size;
375 	uint64_t address;
376 } __attribute__((packed));
377 
378 struct kvm_x86_state {
379 	struct kvm_xsave *xsave;
380 	struct kvm_vcpu_events events;
381 	struct kvm_mp_state mp_state;
382 	struct kvm_regs regs;
383 	struct kvm_xcrs xcrs;
384 	struct kvm_sregs sregs;
385 	struct kvm_debugregs debugregs;
386 	union {
387 		struct kvm_nested_state nested;
388 		char nested_[16384];
389 	};
390 	struct kvm_msrs msrs;
391 };
392 
393 static inline uint64_t get_desc64_base(const struct desc64 *desc)
394 {
395 	return ((uint64_t)desc->base3 << 32) |
396 		(desc->base0 | ((desc->base1) << 16) | ((desc->base2) << 24));
397 }
398 
399 static inline uint64_t rdtsc(void)
400 {
401 	uint32_t eax, edx;
402 	uint64_t tsc_val;
403 	/*
404 	 * The lfence is to wait (on Intel CPUs) until all previous
405 	 * instructions have been executed. If software requires RDTSC to be
406 	 * executed prior to execution of any subsequent instruction, it can
407 	 * execute LFENCE immediately after RDTSC
408 	 */
409 	__asm__ __volatile__("lfence; rdtsc; lfence" : "=a"(eax), "=d"(edx));
410 	tsc_val = ((uint64_t)edx) << 32 | eax;
411 	return tsc_val;
412 }
413 
414 static inline uint64_t rdtscp(uint32_t *aux)
415 {
416 	uint32_t eax, edx;
417 
418 	__asm__ __volatile__("rdtscp" : "=a"(eax), "=d"(edx), "=c"(*aux));
419 	return ((uint64_t)edx) << 32 | eax;
420 }
421 
422 static inline uint64_t rdmsr(uint32_t msr)
423 {
424 	uint32_t a, d;
425 
426 	__asm__ __volatile__("rdmsr" : "=a"(a), "=d"(d) : "c"(msr) : "memory");
427 
428 	return a | ((uint64_t) d << 32);
429 }
430 
431 static inline void wrmsr(uint32_t msr, uint64_t value)
432 {
433 	uint32_t a = value;
434 	uint32_t d = value >> 32;
435 
436 	__asm__ __volatile__("wrmsr" :: "a"(a), "d"(d), "c"(msr) : "memory");
437 }
438 
439 
440 static inline uint16_t inw(uint16_t port)
441 {
442 	uint16_t tmp;
443 
444 	__asm__ __volatile__("in %%dx, %%ax"
445 		: /* output */ "=a" (tmp)
446 		: /* input */ "d" (port));
447 
448 	return tmp;
449 }
450 
451 static inline uint16_t get_es(void)
452 {
453 	uint16_t es;
454 
455 	__asm__ __volatile__("mov %%es, %[es]"
456 			     : /* output */ [es]"=rm"(es));
457 	return es;
458 }
459 
460 static inline uint16_t get_cs(void)
461 {
462 	uint16_t cs;
463 
464 	__asm__ __volatile__("mov %%cs, %[cs]"
465 			     : /* output */ [cs]"=rm"(cs));
466 	return cs;
467 }
468 
469 static inline uint16_t get_ss(void)
470 {
471 	uint16_t ss;
472 
473 	__asm__ __volatile__("mov %%ss, %[ss]"
474 			     : /* output */ [ss]"=rm"(ss));
475 	return ss;
476 }
477 
478 static inline uint16_t get_ds(void)
479 {
480 	uint16_t ds;
481 
482 	__asm__ __volatile__("mov %%ds, %[ds]"
483 			     : /* output */ [ds]"=rm"(ds));
484 	return ds;
485 }
486 
487 static inline uint16_t get_fs(void)
488 {
489 	uint16_t fs;
490 
491 	__asm__ __volatile__("mov %%fs, %[fs]"
492 			     : /* output */ [fs]"=rm"(fs));
493 	return fs;
494 }
495 
496 static inline uint16_t get_gs(void)
497 {
498 	uint16_t gs;
499 
500 	__asm__ __volatile__("mov %%gs, %[gs]"
501 			     : /* output */ [gs]"=rm"(gs));
502 	return gs;
503 }
504 
505 static inline uint16_t get_tr(void)
506 {
507 	uint16_t tr;
508 
509 	__asm__ __volatile__("str %[tr]"
510 			     : /* output */ [tr]"=rm"(tr));
511 	return tr;
512 }
513 
514 static inline uint64_t get_cr0(void)
515 {
516 	uint64_t cr0;
517 
518 	__asm__ __volatile__("mov %%cr0, %[cr0]"
519 			     : /* output */ [cr0]"=r"(cr0));
520 	return cr0;
521 }
522 
523 static inline uint64_t get_cr3(void)
524 {
525 	uint64_t cr3;
526 
527 	__asm__ __volatile__("mov %%cr3, %[cr3]"
528 			     : /* output */ [cr3]"=r"(cr3));
529 	return cr3;
530 }
531 
532 static inline uint64_t get_cr4(void)
533 {
534 	uint64_t cr4;
535 
536 	__asm__ __volatile__("mov %%cr4, %[cr4]"
537 			     : /* output */ [cr4]"=r"(cr4));
538 	return cr4;
539 }
540 
541 static inline void set_cr4(uint64_t val)
542 {
543 	__asm__ __volatile__("mov %0, %%cr4" : : "r" (val) : "memory");
544 }
545 
546 static inline u64 xgetbv(u32 index)
547 {
548 	u32 eax, edx;
549 
550 	__asm__ __volatile__("xgetbv;"
551 		     : "=a" (eax), "=d" (edx)
552 		     : "c" (index));
553 	return eax | ((u64)edx << 32);
554 }
555 
556 static inline void xsetbv(u32 index, u64 value)
557 {
558 	u32 eax = value;
559 	u32 edx = value >> 32;
560 
561 	__asm__ __volatile__("xsetbv" :: "a" (eax), "d" (edx), "c" (index));
562 }
563 
564 static inline void wrpkru(u32 pkru)
565 {
566 	/* Note, ECX and EDX are architecturally required to be '0'. */
567 	asm volatile(".byte 0x0f,0x01,0xef\n\t"
568 		     : : "a" (pkru), "c"(0), "d"(0));
569 }
570 
571 static inline struct desc_ptr get_gdt(void)
572 {
573 	struct desc_ptr gdt;
574 	__asm__ __volatile__("sgdt %[gdt]"
575 			     : /* output */ [gdt]"=m"(gdt));
576 	return gdt;
577 }
578 
579 static inline struct desc_ptr get_idt(void)
580 {
581 	struct desc_ptr idt;
582 	__asm__ __volatile__("sidt %[idt]"
583 			     : /* output */ [idt]"=m"(idt));
584 	return idt;
585 }
586 
587 static inline void outl(uint16_t port, uint32_t value)
588 {
589 	__asm__ __volatile__("outl %%eax, %%dx" : : "d"(port), "a"(value));
590 }
591 
592 static inline void __cpuid(uint32_t function, uint32_t index,
593 			   uint32_t *eax, uint32_t *ebx,
594 			   uint32_t *ecx, uint32_t *edx)
595 {
596 	*eax = function;
597 	*ecx = index;
598 
599 	asm volatile("cpuid"
600 	    : "=a" (*eax),
601 	      "=b" (*ebx),
602 	      "=c" (*ecx),
603 	      "=d" (*edx)
604 	    : "0" (*eax), "2" (*ecx)
605 	    : "memory");
606 }
607 
608 static inline void cpuid(uint32_t function,
609 			 uint32_t *eax, uint32_t *ebx,
610 			 uint32_t *ecx, uint32_t *edx)
611 {
612 	return __cpuid(function, 0, eax, ebx, ecx, edx);
613 }
614 
615 static inline uint32_t this_cpu_fms(void)
616 {
617 	uint32_t eax, ebx, ecx, edx;
618 
619 	cpuid(1, &eax, &ebx, &ecx, &edx);
620 	return eax;
621 }
622 
623 static inline uint32_t this_cpu_family(void)
624 {
625 	return x86_family(this_cpu_fms());
626 }
627 
628 static inline uint32_t this_cpu_model(void)
629 {
630 	return x86_model(this_cpu_fms());
631 }
632 
633 static inline bool this_cpu_vendor_string_is(const char *vendor)
634 {
635 	const uint32_t *chunk = (const uint32_t *)vendor;
636 	uint32_t eax, ebx, ecx, edx;
637 
638 	cpuid(0, &eax, &ebx, &ecx, &edx);
639 	return (ebx == chunk[0] && edx == chunk[1] && ecx == chunk[2]);
640 }
641 
642 static inline bool this_cpu_is_intel(void)
643 {
644 	return this_cpu_vendor_string_is("GenuineIntel");
645 }
646 
647 /*
648  * Exclude early K5 samples with a vendor string of "AMDisbetter!"
649  */
650 static inline bool this_cpu_is_amd(void)
651 {
652 	return this_cpu_vendor_string_is("AuthenticAMD");
653 }
654 
655 static inline uint32_t __this_cpu_has(uint32_t function, uint32_t index,
656 				      uint8_t reg, uint8_t lo, uint8_t hi)
657 {
658 	uint32_t gprs[4];
659 
660 	__cpuid(function, index,
661 		&gprs[KVM_CPUID_EAX], &gprs[KVM_CPUID_EBX],
662 		&gprs[KVM_CPUID_ECX], &gprs[KVM_CPUID_EDX]);
663 
664 	return (gprs[reg] & GENMASK(hi, lo)) >> lo;
665 }
666 
667 static inline bool this_cpu_has(struct kvm_x86_cpu_feature feature)
668 {
669 	return __this_cpu_has(feature.function, feature.index,
670 			      feature.reg, feature.bit, feature.bit);
671 }
672 
673 static inline uint32_t this_cpu_property(struct kvm_x86_cpu_property property)
674 {
675 	return __this_cpu_has(property.function, property.index,
676 			      property.reg, property.lo_bit, property.hi_bit);
677 }
678 
679 static __always_inline bool this_cpu_has_p(struct kvm_x86_cpu_property property)
680 {
681 	uint32_t max_leaf;
682 
683 	switch (property.function & 0xc0000000) {
684 	case 0:
685 		max_leaf = this_cpu_property(X86_PROPERTY_MAX_BASIC_LEAF);
686 		break;
687 	case 0x40000000:
688 		max_leaf = this_cpu_property(X86_PROPERTY_MAX_KVM_LEAF);
689 		break;
690 	case 0x80000000:
691 		max_leaf = this_cpu_property(X86_PROPERTY_MAX_EXT_LEAF);
692 		break;
693 	case 0xc0000000:
694 		max_leaf = this_cpu_property(X86_PROPERTY_MAX_CENTAUR_LEAF);
695 	}
696 	return max_leaf >= property.function;
697 }
698 
699 static inline bool this_pmu_has(struct kvm_x86_pmu_feature feature)
700 {
701 	uint32_t nr_bits = this_cpu_property(X86_PROPERTY_PMU_EBX_BIT_VECTOR_LENGTH);
702 
703 	return nr_bits > feature.anti_feature.bit &&
704 	       !this_cpu_has(feature.anti_feature);
705 }
706 
707 static __always_inline uint64_t this_cpu_supported_xcr0(void)
708 {
709 	if (!this_cpu_has_p(X86_PROPERTY_SUPPORTED_XCR0_LO))
710 		return 0;
711 
712 	return this_cpu_property(X86_PROPERTY_SUPPORTED_XCR0_LO) |
713 	       ((uint64_t)this_cpu_property(X86_PROPERTY_SUPPORTED_XCR0_HI) << 32);
714 }
715 
716 typedef u32		__attribute__((vector_size(16))) sse128_t;
717 #define __sse128_u	union { sse128_t vec; u64 as_u64[2]; u32 as_u32[4]; }
718 #define sse128_lo(x)	({ __sse128_u t; t.vec = x; t.as_u64[0]; })
719 #define sse128_hi(x)	({ __sse128_u t; t.vec = x; t.as_u64[1]; })
720 
721 static inline void read_sse_reg(int reg, sse128_t *data)
722 {
723 	switch (reg) {
724 	case 0:
725 		asm("movdqa %%xmm0, %0" : "=m"(*data));
726 		break;
727 	case 1:
728 		asm("movdqa %%xmm1, %0" : "=m"(*data));
729 		break;
730 	case 2:
731 		asm("movdqa %%xmm2, %0" : "=m"(*data));
732 		break;
733 	case 3:
734 		asm("movdqa %%xmm3, %0" : "=m"(*data));
735 		break;
736 	case 4:
737 		asm("movdqa %%xmm4, %0" : "=m"(*data));
738 		break;
739 	case 5:
740 		asm("movdqa %%xmm5, %0" : "=m"(*data));
741 		break;
742 	case 6:
743 		asm("movdqa %%xmm6, %0" : "=m"(*data));
744 		break;
745 	case 7:
746 		asm("movdqa %%xmm7, %0" : "=m"(*data));
747 		break;
748 	default:
749 		BUG();
750 	}
751 }
752 
753 static inline void write_sse_reg(int reg, const sse128_t *data)
754 {
755 	switch (reg) {
756 	case 0:
757 		asm("movdqa %0, %%xmm0" : : "m"(*data));
758 		break;
759 	case 1:
760 		asm("movdqa %0, %%xmm1" : : "m"(*data));
761 		break;
762 	case 2:
763 		asm("movdqa %0, %%xmm2" : : "m"(*data));
764 		break;
765 	case 3:
766 		asm("movdqa %0, %%xmm3" : : "m"(*data));
767 		break;
768 	case 4:
769 		asm("movdqa %0, %%xmm4" : : "m"(*data));
770 		break;
771 	case 5:
772 		asm("movdqa %0, %%xmm5" : : "m"(*data));
773 		break;
774 	case 6:
775 		asm("movdqa %0, %%xmm6" : : "m"(*data));
776 		break;
777 	case 7:
778 		asm("movdqa %0, %%xmm7" : : "m"(*data));
779 		break;
780 	default:
781 		BUG();
782 	}
783 }
784 
785 static inline void cpu_relax(void)
786 {
787 	asm volatile("rep; nop" ::: "memory");
788 }
789 
790 #define ud2()			\
791 	__asm__ __volatile__(	\
792 		"ud2\n"	\
793 		)
794 
795 #define hlt()			\
796 	__asm__ __volatile__(	\
797 		"hlt\n"	\
798 		)
799 
800 struct kvm_x86_state *vcpu_save_state(struct kvm_vcpu *vcpu);
801 void vcpu_load_state(struct kvm_vcpu *vcpu, struct kvm_x86_state *state);
802 void kvm_x86_state_cleanup(struct kvm_x86_state *state);
803 
804 const struct kvm_msr_list *kvm_get_msr_index_list(void);
805 const struct kvm_msr_list *kvm_get_feature_msr_index_list(void);
806 bool kvm_msr_is_in_save_restore_list(uint32_t msr_index);
807 uint64_t kvm_get_feature_msr(uint64_t msr_index);
808 
809 static inline void vcpu_msrs_get(struct kvm_vcpu *vcpu,
810 				 struct kvm_msrs *msrs)
811 {
812 	int r = __vcpu_ioctl(vcpu, KVM_GET_MSRS, msrs);
813 
814 	TEST_ASSERT(r == msrs->nmsrs,
815 		    "KVM_GET_MSRS failed, r: %i (failed on MSR %x)",
816 		    r, r < 0 || r >= msrs->nmsrs ? -1 : msrs->entries[r].index);
817 }
818 static inline void vcpu_msrs_set(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs)
819 {
820 	int r = __vcpu_ioctl(vcpu, KVM_SET_MSRS, msrs);
821 
822 	TEST_ASSERT(r == msrs->nmsrs,
823 		    "KVM_SET_MSRS failed, r: %i (failed on MSR %x)",
824 		    r, r < 0 || r >= msrs->nmsrs ? -1 : msrs->entries[r].index);
825 }
826 static inline void vcpu_debugregs_get(struct kvm_vcpu *vcpu,
827 				      struct kvm_debugregs *debugregs)
828 {
829 	vcpu_ioctl(vcpu, KVM_GET_DEBUGREGS, debugregs);
830 }
831 static inline void vcpu_debugregs_set(struct kvm_vcpu *vcpu,
832 				      struct kvm_debugregs *debugregs)
833 {
834 	vcpu_ioctl(vcpu, KVM_SET_DEBUGREGS, debugregs);
835 }
836 static inline void vcpu_xsave_get(struct kvm_vcpu *vcpu,
837 				  struct kvm_xsave *xsave)
838 {
839 	vcpu_ioctl(vcpu, KVM_GET_XSAVE, xsave);
840 }
841 static inline void vcpu_xsave2_get(struct kvm_vcpu *vcpu,
842 				   struct kvm_xsave *xsave)
843 {
844 	vcpu_ioctl(vcpu, KVM_GET_XSAVE2, xsave);
845 }
846 static inline void vcpu_xsave_set(struct kvm_vcpu *vcpu,
847 				  struct kvm_xsave *xsave)
848 {
849 	vcpu_ioctl(vcpu, KVM_SET_XSAVE, xsave);
850 }
851 static inline void vcpu_xcrs_get(struct kvm_vcpu *vcpu,
852 				 struct kvm_xcrs *xcrs)
853 {
854 	vcpu_ioctl(vcpu, KVM_GET_XCRS, xcrs);
855 }
856 static inline void vcpu_xcrs_set(struct kvm_vcpu *vcpu, struct kvm_xcrs *xcrs)
857 {
858 	vcpu_ioctl(vcpu, KVM_SET_XCRS, xcrs);
859 }
860 
861 const struct kvm_cpuid_entry2 *get_cpuid_entry(const struct kvm_cpuid2 *cpuid,
862 					       uint32_t function, uint32_t index);
863 const struct kvm_cpuid2 *kvm_get_supported_cpuid(void);
864 const struct kvm_cpuid2 *kvm_get_supported_hv_cpuid(void);
865 const struct kvm_cpuid2 *vcpu_get_supported_hv_cpuid(struct kvm_vcpu *vcpu);
866 
867 static inline uint32_t kvm_cpu_fms(void)
868 {
869 	return get_cpuid_entry(kvm_get_supported_cpuid(), 0x1, 0)->eax;
870 }
871 
872 static inline uint32_t kvm_cpu_family(void)
873 {
874 	return x86_family(kvm_cpu_fms());
875 }
876 
877 static inline uint32_t kvm_cpu_model(void)
878 {
879 	return x86_model(kvm_cpu_fms());
880 }
881 
882 bool kvm_cpuid_has(const struct kvm_cpuid2 *cpuid,
883 		   struct kvm_x86_cpu_feature feature);
884 
885 static inline bool kvm_cpu_has(struct kvm_x86_cpu_feature feature)
886 {
887 	return kvm_cpuid_has(kvm_get_supported_cpuid(), feature);
888 }
889 
890 uint32_t kvm_cpuid_property(const struct kvm_cpuid2 *cpuid,
891 			    struct kvm_x86_cpu_property property);
892 
893 static inline uint32_t kvm_cpu_property(struct kvm_x86_cpu_property property)
894 {
895 	return kvm_cpuid_property(kvm_get_supported_cpuid(), property);
896 }
897 
898 static __always_inline bool kvm_cpu_has_p(struct kvm_x86_cpu_property property)
899 {
900 	uint32_t max_leaf;
901 
902 	switch (property.function & 0xc0000000) {
903 	case 0:
904 		max_leaf = kvm_cpu_property(X86_PROPERTY_MAX_BASIC_LEAF);
905 		break;
906 	case 0x40000000:
907 		max_leaf = kvm_cpu_property(X86_PROPERTY_MAX_KVM_LEAF);
908 		break;
909 	case 0x80000000:
910 		max_leaf = kvm_cpu_property(X86_PROPERTY_MAX_EXT_LEAF);
911 		break;
912 	case 0xc0000000:
913 		max_leaf = kvm_cpu_property(X86_PROPERTY_MAX_CENTAUR_LEAF);
914 	}
915 	return max_leaf >= property.function;
916 }
917 
918 static inline bool kvm_pmu_has(struct kvm_x86_pmu_feature feature)
919 {
920 	uint32_t nr_bits = kvm_cpu_property(X86_PROPERTY_PMU_EBX_BIT_VECTOR_LENGTH);
921 
922 	return nr_bits > feature.anti_feature.bit &&
923 	       !kvm_cpu_has(feature.anti_feature);
924 }
925 
926 static __always_inline uint64_t kvm_cpu_supported_xcr0(void)
927 {
928 	if (!kvm_cpu_has_p(X86_PROPERTY_SUPPORTED_XCR0_LO))
929 		return 0;
930 
931 	return kvm_cpu_property(X86_PROPERTY_SUPPORTED_XCR0_LO) |
932 	       ((uint64_t)kvm_cpu_property(X86_PROPERTY_SUPPORTED_XCR0_HI) << 32);
933 }
934 
935 static inline size_t kvm_cpuid2_size(int nr_entries)
936 {
937 	return sizeof(struct kvm_cpuid2) +
938 	       sizeof(struct kvm_cpuid_entry2) * nr_entries;
939 }
940 
941 /*
942  * Allocate a "struct kvm_cpuid2* instance, with the 0-length arrary of
943  * entries sized to hold @nr_entries.  The caller is responsible for freeing
944  * the struct.
945  */
946 static inline struct kvm_cpuid2 *allocate_kvm_cpuid2(int nr_entries)
947 {
948 	struct kvm_cpuid2 *cpuid;
949 
950 	cpuid = malloc(kvm_cpuid2_size(nr_entries));
951 	TEST_ASSERT(cpuid, "-ENOMEM when allocating kvm_cpuid2");
952 
953 	cpuid->nent = nr_entries;
954 
955 	return cpuid;
956 }
957 
958 void vcpu_init_cpuid(struct kvm_vcpu *vcpu, const struct kvm_cpuid2 *cpuid);
959 void vcpu_set_hv_cpuid(struct kvm_vcpu *vcpu);
960 
961 static inline struct kvm_cpuid_entry2 *__vcpu_get_cpuid_entry(struct kvm_vcpu *vcpu,
962 							      uint32_t function,
963 							      uint32_t index)
964 {
965 	return (struct kvm_cpuid_entry2 *)get_cpuid_entry(vcpu->cpuid,
966 							  function, index);
967 }
968 
969 static inline struct kvm_cpuid_entry2 *vcpu_get_cpuid_entry(struct kvm_vcpu *vcpu,
970 							    uint32_t function)
971 {
972 	return __vcpu_get_cpuid_entry(vcpu, function, 0);
973 }
974 
975 static inline int __vcpu_set_cpuid(struct kvm_vcpu *vcpu)
976 {
977 	int r;
978 
979 	TEST_ASSERT(vcpu->cpuid, "Must do vcpu_init_cpuid() first");
980 	r = __vcpu_ioctl(vcpu, KVM_SET_CPUID2, vcpu->cpuid);
981 	if (r)
982 		return r;
983 
984 	/* On success, refresh the cache to pick up adjustments made by KVM. */
985 	vcpu_ioctl(vcpu, KVM_GET_CPUID2, vcpu->cpuid);
986 	return 0;
987 }
988 
989 static inline void vcpu_set_cpuid(struct kvm_vcpu *vcpu)
990 {
991 	TEST_ASSERT(vcpu->cpuid, "Must do vcpu_init_cpuid() first");
992 	vcpu_ioctl(vcpu, KVM_SET_CPUID2, vcpu->cpuid);
993 
994 	/* Refresh the cache to pick up adjustments made by KVM. */
995 	vcpu_ioctl(vcpu, KVM_GET_CPUID2, vcpu->cpuid);
996 }
997 
998 void vcpu_set_cpuid_maxphyaddr(struct kvm_vcpu *vcpu, uint8_t maxphyaddr);
999 
1000 void vcpu_clear_cpuid_entry(struct kvm_vcpu *vcpu, uint32_t function);
1001 void vcpu_set_or_clear_cpuid_feature(struct kvm_vcpu *vcpu,
1002 				     struct kvm_x86_cpu_feature feature,
1003 				     bool set);
1004 
1005 static inline void vcpu_set_cpuid_feature(struct kvm_vcpu *vcpu,
1006 					  struct kvm_x86_cpu_feature feature)
1007 {
1008 	vcpu_set_or_clear_cpuid_feature(vcpu, feature, true);
1009 
1010 }
1011 
1012 static inline void vcpu_clear_cpuid_feature(struct kvm_vcpu *vcpu,
1013 					    struct kvm_x86_cpu_feature feature)
1014 {
1015 	vcpu_set_or_clear_cpuid_feature(vcpu, feature, false);
1016 }
1017 
1018 uint64_t vcpu_get_msr(struct kvm_vcpu *vcpu, uint64_t msr_index);
1019 int _vcpu_set_msr(struct kvm_vcpu *vcpu, uint64_t msr_index, uint64_t msr_value);
1020 
1021 /*
1022  * Assert on an MSR access(es) and pretty print the MSR name when possible.
1023  * Note, the caller provides the stringified name so that the name of macro is
1024  * printed, not the value the macro resolves to (due to macro expansion).
1025  */
1026 #define TEST_ASSERT_MSR(cond, fmt, msr, str, args...)				\
1027 do {										\
1028 	if (__builtin_constant_p(msr)) {					\
1029 		TEST_ASSERT(cond, fmt, str, args);				\
1030 	} else if (!(cond)) {							\
1031 		char buf[16];							\
1032 										\
1033 		snprintf(buf, sizeof(buf), "MSR 0x%x", msr);			\
1034 		TEST_ASSERT(cond, fmt, buf, args);				\
1035 	}									\
1036 } while (0)
1037 
1038 /*
1039  * Returns true if KVM should return the last written value when reading an MSR
1040  * from userspace, e.g. the MSR isn't a command MSR, doesn't emulate state that
1041  * is changing, etc.  This is NOT an exhaustive list!  The intent is to filter
1042  * out MSRs that are not durable _and_ that a selftest wants to write.
1043  */
1044 static inline bool is_durable_msr(uint32_t msr)
1045 {
1046 	return msr != MSR_IA32_TSC;
1047 }
1048 
1049 #define vcpu_set_msr(vcpu, msr, val)							\
1050 do {											\
1051 	uint64_t r, v = val;								\
1052 											\
1053 	TEST_ASSERT_MSR(_vcpu_set_msr(vcpu, msr, v) == 1,				\
1054 			"KVM_SET_MSRS failed on %s, value = 0x%lx", msr, #msr, v);	\
1055 	if (!is_durable_msr(msr))							\
1056 		break;									\
1057 	r = vcpu_get_msr(vcpu, msr);							\
1058 	TEST_ASSERT_MSR(r == v, "Set %s to '0x%lx', got back '0x%lx'", msr, #msr, v, r);\
1059 } while (0)
1060 
1061 void kvm_get_cpu_address_width(unsigned int *pa_bits, unsigned int *va_bits);
1062 bool vm_is_unrestricted_guest(struct kvm_vm *vm);
1063 
1064 struct ex_regs {
1065 	uint64_t rax, rcx, rdx, rbx;
1066 	uint64_t rbp, rsi, rdi;
1067 	uint64_t r8, r9, r10, r11;
1068 	uint64_t r12, r13, r14, r15;
1069 	uint64_t vector;
1070 	uint64_t error_code;
1071 	uint64_t rip;
1072 	uint64_t cs;
1073 	uint64_t rflags;
1074 };
1075 
1076 struct idt_entry {
1077 	uint16_t offset0;
1078 	uint16_t selector;
1079 	uint16_t ist : 3;
1080 	uint16_t : 5;
1081 	uint16_t type : 4;
1082 	uint16_t : 1;
1083 	uint16_t dpl : 2;
1084 	uint16_t p : 1;
1085 	uint16_t offset1;
1086 	uint32_t offset2; uint32_t reserved;
1087 };
1088 
1089 void vm_init_descriptor_tables(struct kvm_vm *vm);
1090 void vcpu_init_descriptor_tables(struct kvm_vcpu *vcpu);
1091 void vm_install_exception_handler(struct kvm_vm *vm, int vector,
1092 			void (*handler)(struct ex_regs *));
1093 
1094 /* If a toddler were to say "abracadabra". */
1095 #define KVM_EXCEPTION_MAGIC 0xabacadabaULL
1096 
1097 /*
1098  * KVM selftest exception fixup uses registers to coordinate with the exception
1099  * handler, versus the kernel's in-memory tables and KVM-Unit-Tests's in-memory
1100  * per-CPU data.  Using only registers avoids having to map memory into the
1101  * guest, doesn't require a valid, stable GS.base, and reduces the risk of
1102  * for recursive faults when accessing memory in the handler.  The downside to
1103  * using registers is that it restricts what registers can be used by the actual
1104  * instruction.  But, selftests are 64-bit only, making register* pressure a
1105  * minor concern.  Use r9-r11 as they are volatile, i.e. don't need to be saved
1106  * by the callee, and except for r11 are not implicit parameters to any
1107  * instructions.  Ideally, fixup would use r8-r10 and thus avoid implicit
1108  * parameters entirely, but Hyper-V's hypercall ABI uses r8 and testing Hyper-V
1109  * is higher priority than testing non-faulting SYSCALL/SYSRET.
1110  *
1111  * Note, the fixup handler deliberately does not handle #DE, i.e. the vector
1112  * is guaranteed to be non-zero on fault.
1113  *
1114  * REGISTER INPUTS:
1115  * r9  = MAGIC
1116  * r10 = RIP
1117  * r11 = new RIP on fault
1118  *
1119  * REGISTER OUTPUTS:
1120  * r9  = exception vector (non-zero)
1121  * r10 = error code
1122  */
1123 #define KVM_ASM_SAFE(insn)					\
1124 	"mov $" __stringify(KVM_EXCEPTION_MAGIC) ", %%r9\n\t"	\
1125 	"lea 1f(%%rip), %%r10\n\t"				\
1126 	"lea 2f(%%rip), %%r11\n\t"				\
1127 	"1: " insn "\n\t"					\
1128 	"xor %%r9, %%r9\n\t"					\
1129 	"2:\n\t"						\
1130 	"mov  %%r9b, %[vector]\n\t"				\
1131 	"mov  %%r10, %[error_code]\n\t"
1132 
1133 #define KVM_ASM_SAFE_OUTPUTS(v, ec)	[vector] "=qm"(v), [error_code] "=rm"(ec)
1134 #define KVM_ASM_SAFE_CLOBBERS	"r9", "r10", "r11"
1135 
1136 #define kvm_asm_safe(insn, inputs...)					\
1137 ({									\
1138 	uint64_t ign_error_code;					\
1139 	uint8_t vector;							\
1140 									\
1141 	asm volatile(KVM_ASM_SAFE(insn)					\
1142 		     : KVM_ASM_SAFE_OUTPUTS(vector, ign_error_code)	\
1143 		     : inputs						\
1144 		     : KVM_ASM_SAFE_CLOBBERS);				\
1145 	vector;								\
1146 })
1147 
1148 #define kvm_asm_safe_ec(insn, error_code, inputs...)			\
1149 ({									\
1150 	uint8_t vector;							\
1151 									\
1152 	asm volatile(KVM_ASM_SAFE(insn)					\
1153 		     : KVM_ASM_SAFE_OUTPUTS(vector, error_code)		\
1154 		     : inputs						\
1155 		     : KVM_ASM_SAFE_CLOBBERS);				\
1156 	vector;								\
1157 })
1158 
1159 static inline uint8_t rdmsr_safe(uint32_t msr, uint64_t *val)
1160 {
1161 	uint64_t error_code;
1162 	uint8_t vector;
1163 	uint32_t a, d;
1164 
1165 	asm volatile(KVM_ASM_SAFE("rdmsr")
1166 		     : "=a"(a), "=d"(d), KVM_ASM_SAFE_OUTPUTS(vector, error_code)
1167 		     : "c"(msr)
1168 		     : KVM_ASM_SAFE_CLOBBERS);
1169 
1170 	*val = (uint64_t)a | ((uint64_t)d << 32);
1171 	return vector;
1172 }
1173 
1174 static inline uint8_t wrmsr_safe(uint32_t msr, uint64_t val)
1175 {
1176 	return kvm_asm_safe("wrmsr", "a"(val & -1u), "d"(val >> 32), "c"(msr));
1177 }
1178 
1179 static inline uint8_t xsetbv_safe(uint32_t index, uint64_t value)
1180 {
1181 	u32 eax = value;
1182 	u32 edx = value >> 32;
1183 
1184 	return kvm_asm_safe("xsetbv", "a" (eax), "d" (edx), "c" (index));
1185 }
1186 
1187 bool kvm_is_tdp_enabled(void);
1188 
1189 uint64_t *__vm_get_page_table_entry(struct kvm_vm *vm, uint64_t vaddr,
1190 				    int *level);
1191 uint64_t *vm_get_page_table_entry(struct kvm_vm *vm, uint64_t vaddr);
1192 
1193 uint64_t kvm_hypercall(uint64_t nr, uint64_t a0, uint64_t a1, uint64_t a2,
1194 		       uint64_t a3);
1195 uint64_t __xen_hypercall(uint64_t nr, uint64_t a0, void *a1);
1196 void xen_hypercall(uint64_t nr, uint64_t a0, void *a1);
1197 
1198 static inline uint64_t __kvm_hypercall_map_gpa_range(uint64_t gpa,
1199 						     uint64_t size, uint64_t flags)
1200 {
1201 	return kvm_hypercall(KVM_HC_MAP_GPA_RANGE, gpa, size >> PAGE_SHIFT, flags, 0);
1202 }
1203 
1204 static inline void kvm_hypercall_map_gpa_range(uint64_t gpa, uint64_t size,
1205 					       uint64_t flags)
1206 {
1207 	uint64_t ret = __kvm_hypercall_map_gpa_range(gpa, size, flags);
1208 
1209 	GUEST_ASSERT(!ret);
1210 }
1211 
1212 void __vm_xsave_require_permission(uint64_t xfeature, const char *name);
1213 
1214 #define vm_xsave_require_permission(xfeature)	\
1215 	__vm_xsave_require_permission(xfeature, #xfeature)
1216 
1217 enum pg_level {
1218 	PG_LEVEL_NONE,
1219 	PG_LEVEL_4K,
1220 	PG_LEVEL_2M,
1221 	PG_LEVEL_1G,
1222 	PG_LEVEL_512G,
1223 	PG_LEVEL_NUM
1224 };
1225 
1226 #define PG_LEVEL_SHIFT(_level) ((_level - 1) * 9 + 12)
1227 #define PG_LEVEL_SIZE(_level) (1ull << PG_LEVEL_SHIFT(_level))
1228 
1229 #define PG_SIZE_4K PG_LEVEL_SIZE(PG_LEVEL_4K)
1230 #define PG_SIZE_2M PG_LEVEL_SIZE(PG_LEVEL_2M)
1231 #define PG_SIZE_1G PG_LEVEL_SIZE(PG_LEVEL_1G)
1232 
1233 void __virt_pg_map(struct kvm_vm *vm, uint64_t vaddr, uint64_t paddr, int level);
1234 void virt_map_level(struct kvm_vm *vm, uint64_t vaddr, uint64_t paddr,
1235 		    uint64_t nr_bytes, int level);
1236 
1237 /*
1238  * Basic CPU control in CR0
1239  */
1240 #define X86_CR0_PE          (1UL<<0) /* Protection Enable */
1241 #define X86_CR0_MP          (1UL<<1) /* Monitor Coprocessor */
1242 #define X86_CR0_EM          (1UL<<2) /* Emulation */
1243 #define X86_CR0_TS          (1UL<<3) /* Task Switched */
1244 #define X86_CR0_ET          (1UL<<4) /* Extension Type */
1245 #define X86_CR0_NE          (1UL<<5) /* Numeric Error */
1246 #define X86_CR0_WP          (1UL<<16) /* Write Protect */
1247 #define X86_CR0_AM          (1UL<<18) /* Alignment Mask */
1248 #define X86_CR0_NW          (1UL<<29) /* Not Write-through */
1249 #define X86_CR0_CD          (1UL<<30) /* Cache Disable */
1250 #define X86_CR0_PG          (1UL<<31) /* Paging */
1251 
1252 #define PFERR_PRESENT_BIT 0
1253 #define PFERR_WRITE_BIT 1
1254 #define PFERR_USER_BIT 2
1255 #define PFERR_RSVD_BIT 3
1256 #define PFERR_FETCH_BIT 4
1257 #define PFERR_PK_BIT 5
1258 #define PFERR_SGX_BIT 15
1259 #define PFERR_GUEST_FINAL_BIT 32
1260 #define PFERR_GUEST_PAGE_BIT 33
1261 #define PFERR_IMPLICIT_ACCESS_BIT 48
1262 
1263 #define PFERR_PRESENT_MASK	BIT(PFERR_PRESENT_BIT)
1264 #define PFERR_WRITE_MASK	BIT(PFERR_WRITE_BIT)
1265 #define PFERR_USER_MASK		BIT(PFERR_USER_BIT)
1266 #define PFERR_RSVD_MASK		BIT(PFERR_RSVD_BIT)
1267 #define PFERR_FETCH_MASK	BIT(PFERR_FETCH_BIT)
1268 #define PFERR_PK_MASK		BIT(PFERR_PK_BIT)
1269 #define PFERR_SGX_MASK		BIT(PFERR_SGX_BIT)
1270 #define PFERR_GUEST_FINAL_MASK	BIT_ULL(PFERR_GUEST_FINAL_BIT)
1271 #define PFERR_GUEST_PAGE_MASK	BIT_ULL(PFERR_GUEST_PAGE_BIT)
1272 #define PFERR_IMPLICIT_ACCESS	BIT_ULL(PFERR_IMPLICIT_ACCESS_BIT)
1273 
1274 bool sys_clocksource_is_based_on_tsc(void);
1275 
1276 #endif /* SELFTEST_KVM_PROCESSOR_H */
1277