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
14 struct msr {
15 union {
16 struct {
17 u32 l;
18 u32 h;
19 };
20 u64 q;
21 };
22 };
23
24 struct msr_info {
25 u32 msr_no;
26 struct msr reg;
27 struct msr *msrs;
28 int err;
29 };
30
31 struct msr_regs_info {
32 u32 *regs;
33 int err;
34 };
35
36 struct saved_msr {
37 bool valid;
38 struct msr_info info;
39 };
40
41 struct 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
70 DECLARE_TRACEPOINT(read_msr);
71 DECLARE_TRACEPOINT(write_msr);
72 DECLARE_TRACEPOINT(rdpmc);
73 extern void do_trace_write_msr(unsigned int msr, u64 val, int failed);
74 extern void do_trace_read_msr(unsigned int msr, u64 val, int failed);
75 extern void do_trace_rdpmc(unsigned int msr, u64 val, int failed);
76 #else
do_trace_write_msr(unsigned int msr,u64 val,int failed)77 static inline void do_trace_write_msr(unsigned int msr, u64 val, int failed) {}
do_trace_read_msr(unsigned int msr,u64 val,int failed)78 static inline void do_trace_read_msr(unsigned int msr, u64 val, int failed) {}
do_trace_rdpmc(unsigned int msr,u64 val,int failed)79 static 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 */
__rdmsr(unsigned int msr)89 static __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
__wrmsr(unsigned int msr,u32 low,u32 high)101 static __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) \
110 do { \
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
native_read_msr(unsigned int msr)123 static 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
native_read_msr_safe(unsigned int msr,int * err)135 static 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 */
157 static inline void notrace
native_write_msr(unsigned int msr,u32 low,u32 high)158 native_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 */
167 static inline int notrace
native_write_msr_safe(unsigned int msr,u32 low,u32 high)168 native_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
187 extern int rdmsr_safe_regs(u32 regs[8]);
188 extern 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 */
rdtsc(void)199 static __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 */
rdtsc_ordered(void)216 static __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
native_read_pmc(int counter)244 static 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) \
265 do { \
266 u64 __val = native_read_msr((msr)); \
267 (void)((low) = (u32)__val); \
268 (void)((high) = (u32)(__val >> 32)); \
269 } while (0)
270
wrmsr(unsigned int msr,u32 low,u32 high)271 static 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
wrmsrl(unsigned int msr,u64 val)279 static 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 */
wrmsr_safe(unsigned int msr,u32 low,u32 high)285 static 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
rdmsrl_safe(unsigned int msr,unsigned long long * p)300 static 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) \
309 do { \
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 */
wrmsrl_safe(u32 msr,u64 val)322 static 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
331 struct msr *msrs_alloc(void);
332 void msrs_free(struct msr *msrs);
333 int msr_set_bit(u32 msr, u8 bit);
334 int msr_clear_bit(u32 msr, u8 bit);
335
336 #ifdef CONFIG_SMP
337 int rdmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h);
338 int wrmsr_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h);
339 int rdmsrl_on_cpu(unsigned int cpu, u32 msr_no, u64 *q);
340 int wrmsrl_on_cpu(unsigned int cpu, u32 msr_no, u64 q);
341 void rdmsr_on_cpus(const struct cpumask *mask, u32 msr_no, struct msr *msrs);
342 void wrmsr_on_cpus(const struct cpumask *mask, u32 msr_no, struct msr *msrs);
343 int rdmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, u32 *l, u32 *h);
344 int wrmsr_safe_on_cpu(unsigned int cpu, u32 msr_no, u32 l, u32 h);
345 int rdmsrl_safe_on_cpu(unsigned int cpu, u32 msr_no, u64 *q);
346 int wrmsrl_safe_on_cpu(unsigned int cpu, u32 msr_no, u64 q);
347 int rdmsr_safe_regs_on_cpu(unsigned int cpu, u32 regs[8]);
348 int wrmsr_safe_regs_on_cpu(unsigned int cpu, u32 regs[8]);
349 #else /* CONFIG_SMP */
rdmsr_on_cpu(unsigned int cpu,u32 msr_no,u32 * l,u32 * h)350 static 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 }
wrmsr_on_cpu(unsigned int cpu,u32 msr_no,u32 l,u32 h)355 static 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 }
rdmsrl_on_cpu(unsigned int cpu,u32 msr_no,u64 * q)360 static inline int rdmsrl_on_cpu(unsigned int cpu, u32 msr_no, u64 *q)
361 {
362 rdmsrl(msr_no, *q);
363 return 0;
364 }
wrmsrl_on_cpu(unsigned int cpu,u32 msr_no,u64 q)365 static inline int wrmsrl_on_cpu(unsigned int cpu, u32 msr_no, u64 q)
366 {
367 wrmsrl(msr_no, q);
368 return 0;
369 }
rdmsr_on_cpus(const struct cpumask * m,u32 msr_no,struct msr * msrs)370 static 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 }
wrmsr_on_cpus(const struct cpumask * m,u32 msr_no,struct msr * msrs)375 static 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 }
rdmsr_safe_on_cpu(unsigned int cpu,u32 msr_no,u32 * l,u32 * h)380 static 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 }
wrmsr_safe_on_cpu(unsigned int cpu,u32 msr_no,u32 l,u32 h)385 static 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 }
rdmsrl_safe_on_cpu(unsigned int cpu,u32 msr_no,u64 * q)389 static inline int rdmsrl_safe_on_cpu(unsigned int cpu, u32 msr_no, u64 *q)
390 {
391 return rdmsrl_safe(msr_no, q);
392 }
wrmsrl_safe_on_cpu(unsigned int cpu,u32 msr_no,u64 q)393 static inline int wrmsrl_safe_on_cpu(unsigned int cpu, u32 msr_no, u64 q)
394 {
395 return wrmsrl_safe(msr_no, q);
396 }
rdmsr_safe_regs_on_cpu(unsigned int cpu,u32 regs[8])397 static inline int rdmsr_safe_regs_on_cpu(unsigned int cpu, u32 regs[8])
398 {
399 return rdmsr_safe_regs(regs);
400 }
wrmsr_safe_regs_on_cpu(unsigned int cpu,u32 regs[8])401 static 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