1/* SPDX-License-Identifier: GPL-2.0 */ 2#ifndef _ASM_X86_MSR_H 3#define _ASM_X86_MSR_H 4 5#include "msr-index.h" 6 7#ifndef __ASSEMBLY__ 8 9#include <asm/asm.h> 10#include <asm/errno.h> 11#include <asm/cpumask.h> 12#include <uapi/asm/msr.h> 13 14struct msr { 15 union { 16 struct { 17 u32 l; 18 u32 h; 19 }; 20 u64 q; 21 }; 22}; 23 24struct msr_info { 25 u32 msr_no; 26 struct msr reg; 27 struct msr *msrs; 28 int err; 29}; 30 31struct msr_regs_info { 32 u32 *regs; 33 int err; 34}; 35 36struct saved_msr { 37 bool valid; 38 struct msr_info info; 39}; 40 41struct saved_msrs { 42 unsigned int num; 43 struct saved_msr *array; 44}; 45 46/* 47 * both i386 and x86_64 returns 64-bit value in edx:eax, but gcc's "A" 48 * constraint has different meanings. For i386, "A" means exactly 49 * edx:eax, while for x86_64 it doesn't mean rdx:rax or edx:eax. Instead, 50 * it means rax *or* rdx. 51 */ 52#ifdef CONFIG_X86_64 53/* Using 64-bit values saves one instruction clearing the high half of low */ 54#define DECLARE_ARGS(val, low, high) unsigned long low, high 55#define EAX_EDX_VAL(val, low, high) ((low) | (high) << 32) 56#define EAX_EDX_RET(val, low, high) "=a" (low), "=d" (high) 57#else 58#define DECLARE_ARGS(val, low, high) unsigned long long val 59#define EAX_EDX_VAL(val, low, high) (val) 60#define EAX_EDX_RET(val, low, high) "=A" (val) 61#endif 62 63/* 64 * Be very careful with includes. This header is prone to include loops. 65 */ 66#include <asm/atomic.h> 67#include <linux/tracepoint-defs.h> 68 69#ifdef CONFIG_TRACEPOINTS 70DECLARE_TRACEPOINT(read_msr); 71DECLARE_TRACEPOINT(write_msr); 72DECLARE_TRACEPOINT(rdpmc); 73extern void do_trace_write_msr(unsigned int msr, u64 val, int failed); 74extern void do_trace_read_msr(unsigned int msr, u64 val, int failed); 75extern void do_trace_rdpmc(unsigned int msr, u64 val, int failed); 76#else 77static inline void do_trace_write_msr(unsigned int msr, u64 val, int failed) {} 78static inline void do_trace_read_msr(unsigned int msr, u64 val, int failed) {} 79static inline void do_trace_rdpmc(unsigned int msr, u64 val, int failed) {} 80#endif 81 82/* 83 * __rdmsr() and __wrmsr() are the two primitives which are the bare minimum MSR 84 * accessors and should not have any tracing or other functionality piggybacking 85 * on them - those are *purely* for accessing MSRs and nothing more. So don't even 86 * think of extending them - you will be slapped with a stinking trout or a frozen 87 * shark will reach you, wherever you are! You've been warned. 88 */ 89static __always_inline unsigned long long __rdmsr(unsigned int msr) 90{ 91 DECLARE_ARGS(val, low, high); 92 93 asm volatile("1: rdmsr\n" 94 "2:\n" 95 _ASM_EXTABLE_HANDLE(1b, 2b, ex_handler_rdmsr_unsafe) 96 : EAX_EDX_RET(val, low, high) : "c" (msr)); 97 98 return EAX_EDX_VAL(val, low, high); 99} 100 101static __always_inline void __wrmsr(unsigned int msr, u32 low, u32 high) 102{ 103 asm volatile("1: wrmsr\n" 104 "2:\n" 105 _ASM_EXTABLE_HANDLE(1b, 2b, ex_handler_wrmsr_unsafe) 106 : : "c" (msr), "a"(low), "d" (high) : "memory"); 107} 108 109#define native_rdmsr(msr, val1, val2) \ 110do { \ 111 u64 __val = __rdmsr((msr)); \ 112 (void)((val1) = (u32)__val); \ 113 (void)((val2) = (u32)(__val >> 32)); \ 114} while (0) 115 116#define native_wrmsr(msr, low, high) \ 117 __wrmsr(msr, low, high) 118 119#define native_wrmsrl(msr, val) \ 120 __wrmsr((msr), (u32)((u64)(val)), \ 121 (u32)((u64)(val) >> 32)) 122 123static inline unsigned long long native_read_msr(unsigned int msr) 124{ 125 unsigned long long val; 126 127 val = __rdmsr(msr); 128 129 if (tracepoint_enabled(read_msr)) 130 do_trace_read_msr(msr, val, 0); 131 132 return val; 133} 134 135static inline unsigned long long native_read_msr_safe(unsigned int msr, 136 int *err) 137{ 138 DECLARE_ARGS(val, low, high); 139 140 asm volatile("2: rdmsr ; xor %[err],%[err]\n" 141 "1:\n\t" 142 ".section .fixup,\"ax\"\n\t" 143 "3: mov %[fault],%[err]\n\t" 144 "xorl %%eax, %%eax\n\t" 145 "xorl %%edx, %%edx\n\t" 146 "jmp 1b\n\t" 147 ".previous\n\t" 148 _ASM_EXTABLE(2b, 3b) 149 : [err] "=r" (*err), EAX_EDX_RET(val, low, high) 150 : "c" (msr), [fault] "i" (-EIO)); 151 if (tracepoint_enabled(read_msr)) 152 do_trace_read_msr(msr, EAX_EDX_VAL(val, low, high), *err); 153 return EAX_EDX_VAL(val, low, high); 154} 155 156/* Can be uninlined because referenced by paravirt */ 157static inline void notrace 158native_write_msr(unsigned int msr, u32 low, u32 high) 159{ 160 __wrmsr(msr, low, high); 161 162 if (tracepoint_enabled(write_msr)) 163 do_trace_write_msr(msr, ((u64)high << 32 | low), 0); 164} 165 166/* Can be uninlined because referenced by paravirt */ 167static inline int notrace 168native_write_msr_safe(unsigned int msr, u32 low, u32 high) 169{ 170 int err; 171 172 asm volatile("2: wrmsr ; xor %[err],%[err]\n" 173 "1:\n\t" 174 ".section .fixup,\"ax\"\n\t" 175 "3: mov %[fault],%[err] ; jmp 1b\n\t" 176 ".previous\n\t" 177 _ASM_EXTABLE(2b, 3b) 178 : [err] "=a" (err) 179 : "c" (msr), "0" (low), "d" (high), 180 [fault] "i" (-EIO) 181 : "memory"); 182 if (tracepoint_enabled(write_msr)) 183 do_trace_write_msr(msr, ((u64)high << 32 | low), err); 184 return err; 185} 186 187extern int rdmsr_safe_regs(u32 regs[8]); 188extern int wrmsr_safe_regs(u32 regs[8]); 189 190/** 191 * rdtsc() - returns the current TSC without ordering constraints 192 * 193 * rdtsc() returns the result of RDTSC as a 64-bit integer. The 194 * only ordering constraint it supplies is the ordering implied by 195 * "asm volatile": it will put the RDTSC in the place you expect. The 196 * CPU can and will speculatively execute that RDTSC, though, so the 197 * results can be non-monotonic if compared on different CPUs. 198 */ 199static __always_inline unsigned long long rdtsc(void) 200{ 201 DECLARE_ARGS(val, low, high); 202 203 asm volatile("rdtsc" : EAX_EDX_RET(val, low, high)); 204 205 return EAX_EDX_VAL(val, low, high); 206} 207 208/** 209 * rdtsc_ordered() - read the current TSC in program order 210 * 211 * rdtsc_ordered() returns the result of RDTSC as a 64-bit integer. 212 * It is ordered like a load to a global in-memory counter. It should 213 * be impossible to observe non-monotonic rdtsc_unordered() behavior 214 * across multiple CPUs as long as the TSC is synced. 215 */ 216static __always_inline unsigned long long rdtsc_ordered(void) 217{ 218 DECLARE_ARGS(val, low, high); 219 220 /* 221 * The RDTSC instruction is not ordered relative to memory 222 * access. The Intel SDM and the AMD APM are both vague on this 223 * point, but empirically an RDTSC instruction can be 224 * speculatively executed before prior loads. An RDTSC 225 * immediately after an appropriate barrier appears to be 226 * ordered as a normal load, that is, it provides the same 227 * ordering guarantees as reading from a global memory location 228 * that some other imaginary CPU is updating continuously with a 229 * time stamp. 230 * 231 * Thus, use the preferred barrier on the respective CPU, aiming for 232 * RDTSCP as the default. 233 */ 234 asm volatile(ALTERNATIVE_2("rdtsc", 235 "lfence; rdtsc", X86_FEATURE_LFENCE_RDTSC, 236 "rdtscp", X86_FEATURE_RDTSCP) 237 : EAX_EDX_RET(val, low, high) 238 /* RDTSCP clobbers ECX with MSR_TSC_AUX. */ 239 :: "ecx"); 240 241 return EAX_EDX_VAL(val, low, high); 242} 243 244static inline unsigned long long native_read_pmc(int counter) 245{ 246 DECLARE_ARGS(val, low, high); 247 248 asm volatile("rdpmc" : EAX_EDX_RET(val, low, high) : "c" (counter)); 249 if (tracepoint_enabled(rdpmc)) 250 do_trace_rdpmc(counter, EAX_EDX_VAL(val, low, high), 0); 251 return EAX_EDX_VAL(val, low, high); 252} 253 254#ifdef CONFIG_PARAVIRT_XXL 255#include <asm/paravirt.h> 256#else 257#include <linux/errno.h> 258/* 259 * Access to machine-specific registers (available on 586 and better only) 260 * Note: the rd* operations modify the parameters directly (without using 261 * pointer indirection), this allows gcc to optimize better 262 */ 263 264#define rdmsr(msr, low, high) \ 265do { \ 266 u64 __val = native_read_msr((msr)); \ 267 (void)((low) = (u32)__val); \ 268 (void)((high) = (u32)(__val >> 32)); \ 269} while (0) 270 271static inline void wrmsr(unsigned int msr, u32 low, u32 high) 272{ 273 native_write_msr(msr, low, high); 274} 275 276#define rdmsrl(msr, val) \ 277 ((val) = native_read_msr((msr))) 278 279static inline void wrmsrl(unsigned int msr, u64 val) 280{ 281 native_write_msr(msr, (u32)(val & 0xffffffffULL), (u32)(val >> 32)); 282} 283 284/* wrmsr with exception handling */ 285static inline int wrmsr_safe(unsigned int msr, u32 low, u32 high) 286{ 287 return native_write_msr_safe(msr, low, high); 288} 289 290/* rdmsr with exception handling */ 291#define rdmsr_safe(msr, low, high) \ 292({ \ 293 int __err; \ 294 u64 __val = native_read_msr_safe((msr), &__err); \ 295 (*low) = (u32)__val; \ 296 (*high) = (u32)(__val >> 32); \ 297 __err; \ 298}) 299 300static inline int rdmsrl_safe(unsigned int msr, unsigned long long *p) 301{ 302 int err; 303 304 *p = native_read_msr_safe(msr, &err); 305 return err; 306} 307 308#define rdpmc(counter, low, high) \ 309do { \ 310 u64 _l = native_read_pmc((counter)); \ 311 (low) = (u32)_l; \ 312 (high) = (u32)(_l >> 32); \ 313} while (0) 314 315#define rdpmcl(counter, val) ((val) = native_read_pmc(counter)) 316 317#endif /* !CONFIG_PARAVIRT_XXL */ 318 319/* 320 * 64-bit version of wrmsr_safe(): 321 */ 322static inline int wrmsrl_safe(u32 msr, u64 val) 323{ 324 return wrmsr_safe(msr, (u32)val, (u32)(val >> 32)); 325} 326 327#define write_tsc(low, high) wrmsr(MSR_IA32_TSC, (low), (high)) 328 329#define write_rdtscp_aux(val) wrmsr(MSR_TSC_AUX, (val), 0) 330 331struct msr *msrs_alloc(void); 332void msrs_free(struct msr *msrs); 333int msr_set_bit(u32 msr, u8 bit); 334int msr_clear_bit(u32 msr, u8 bit); 335 336#ifdef CONFIG_SMP 337int rdmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h); 338int wrmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h); 339int rdmsrl_on_cpu(unsigned int cpu, u32 msr_no, u64 *q); 340int wrmsrl_on_cpu(unsigned int cpu, u32 msr_no, u64 q); 341void rdmsr_on_cpus(const struct cpumask *mask, u32 msr_no, struct msr *msrs); 342void wrmsr_on_cpus(const struct cpumask *mask, u32 msr_no, struct msr *msrs); 343int rdmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h); 344int wrmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h); 345int rdmsrl_safe_on_cpu(unsigned int cpu, u32 msr_no, u64 *q); 346int wrmsrl_safe_on_cpu(unsigned int cpu, u32 msr_no, u64 q); 347int rdmsr_safe_regs_on_cpu(unsigned int cpu, u32 regs[8]); 348int wrmsr_safe_regs_on_cpu(unsigned int cpu, u32 regs[8]); 349#else /* CONFIG_SMP */ 350static inline int rdmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h) 351{ 352 rdmsr(msr_no, *l, *h); 353 return 0; 354} 355static inline int wrmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h) 356{ 357 wrmsr(msr_no, l, h); 358 return 0; 359} 360static inline int rdmsrl_on_cpu(unsigned int cpu, u32 msr_no, u64 *q) 361{ 362 rdmsrl(msr_no, *q); 363 return 0; 364} 365static inline int wrmsrl_on_cpu(unsigned int cpu, u32 msr_no, u64 q) 366{ 367 wrmsrl(msr_no, q); 368 return 0; 369} 370static inline void rdmsr_on_cpus(const struct cpumask *m, u32 msr_no, 371 struct msr *msrs) 372{ 373 rdmsr_on_cpu(0, msr_no, &(msrs[0].l), &(msrs[0].h)); 374} 375static inline void wrmsr_on_cpus(const struct cpumask *m, u32 msr_no, 376 struct msr *msrs) 377{ 378 wrmsr_on_cpu(0, msr_no, msrs[0].l, msrs[0].h); 379} 380static inline int rdmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, 381 u32 *l, u32 *h) 382{ 383 return rdmsr_safe(msr_no, l, h); 384} 385static inline int wrmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h) 386{ 387 return wrmsr_safe(msr_no, l, h); 388} 389static inline int rdmsrl_safe_on_cpu(unsigned int cpu, u32 msr_no, u64 *q) 390{ 391 return rdmsrl_safe(msr_no, q); 392} 393static inline int wrmsrl_safe_on_cpu(unsigned int cpu, u32 msr_no, u64 q) 394{ 395 return wrmsrl_safe(msr_no, q); 396} 397static inline int rdmsr_safe_regs_on_cpu(unsigned int cpu, u32 regs[8]) 398{ 399 return rdmsr_safe_regs(regs); 400} 401static inline int wrmsr_safe_regs_on_cpu(unsigned int cpu, u32 regs[8]) 402{ 403 return wrmsr_safe_regs(regs); 404} 405#endif /* CONFIG_SMP */ 406#endif /* __ASSEMBLY__ */ 407#endif /* _ASM_X86_MSR_H */ 408