perf/x86: Only allow rdpmc if a perf_event is mapped
[deliverable/linux.git] / arch / x86 / kernel / i387.c
1 /*
2 * Copyright (C) 1994 Linus Torvalds
3 *
4 * Pentium III FXSR, SSE support
5 * General FPU state handling cleanups
6 * Gareth Hughes <gareth@valinux.com>, May 2000
7 */
8 #include <linux/module.h>
9 #include <linux/regset.h>
10 #include <linux/sched.h>
11 #include <linux/slab.h>
12
13 #include <asm/sigcontext.h>
14 #include <asm/processor.h>
15 #include <asm/math_emu.h>
16 #include <asm/tlbflush.h>
17 #include <asm/uaccess.h>
18 #include <asm/ptrace.h>
19 #include <asm/i387.h>
20 #include <asm/fpu-internal.h>
21 #include <asm/user.h>
22
23 /*
24 * Were we in an interrupt that interrupted kernel mode?
25 *
26 * On others, we can do a kernel_fpu_begin/end() pair *ONLY* if that
27 * pair does nothing at all: the thread must not have fpu (so
28 * that we don't try to save the FPU state), and TS must
29 * be set (so that the clts/stts pair does nothing that is
30 * visible in the interrupted kernel thread).
31 *
32 * Except for the eagerfpu case when we return 1 unless we've already
33 * been eager and saved the state in kernel_fpu_begin().
34 */
35 static inline bool interrupted_kernel_fpu_idle(void)
36 {
37 if (use_eager_fpu())
38 return __thread_has_fpu(current);
39
40 return !__thread_has_fpu(current) &&
41 (read_cr0() & X86_CR0_TS);
42 }
43
44 /*
45 * Were we in user mode (or vm86 mode) when we were
46 * interrupted?
47 *
48 * Doing kernel_fpu_begin/end() is ok if we are running
49 * in an interrupt context from user mode - we'll just
50 * save the FPU state as required.
51 */
52 static inline bool interrupted_user_mode(void)
53 {
54 struct pt_regs *regs = get_irq_regs();
55 return regs && user_mode_vm(regs);
56 }
57
58 /*
59 * Can we use the FPU in kernel mode with the
60 * whole "kernel_fpu_begin/end()" sequence?
61 *
62 * It's always ok in process context (ie "not interrupt")
63 * but it is sometimes ok even from an irq.
64 */
65 bool irq_fpu_usable(void)
66 {
67 return !in_interrupt() ||
68 interrupted_user_mode() ||
69 interrupted_kernel_fpu_idle();
70 }
71 EXPORT_SYMBOL(irq_fpu_usable);
72
73 void __kernel_fpu_begin(void)
74 {
75 struct task_struct *me = current;
76
77 if (__thread_has_fpu(me)) {
78 __thread_clear_has_fpu(me);
79 __save_init_fpu(me);
80 /* We do 'stts()' in __kernel_fpu_end() */
81 } else if (!use_eager_fpu()) {
82 this_cpu_write(fpu_owner_task, NULL);
83 clts();
84 }
85 }
86 EXPORT_SYMBOL(__kernel_fpu_begin);
87
88 void __kernel_fpu_end(void)
89 {
90 if (use_eager_fpu()) {
91 /*
92 * For eager fpu, most the time, tsk_used_math() is true.
93 * Restore the user math as we are done with the kernel usage.
94 * At few instances during thread exit, signal handling etc,
95 * tsk_used_math() is false. Those few places will take proper
96 * actions, so we don't need to restore the math here.
97 */
98 if (likely(tsk_used_math(current)))
99 math_state_restore();
100 } else {
101 stts();
102 }
103 }
104 EXPORT_SYMBOL(__kernel_fpu_end);
105
106 void unlazy_fpu(struct task_struct *tsk)
107 {
108 preempt_disable();
109 if (__thread_has_fpu(tsk)) {
110 __save_init_fpu(tsk);
111 __thread_fpu_end(tsk);
112 } else
113 tsk->thread.fpu_counter = 0;
114 preempt_enable();
115 }
116 EXPORT_SYMBOL(unlazy_fpu);
117
118 unsigned int mxcsr_feature_mask __read_mostly = 0xffffffffu;
119 unsigned int xstate_size;
120 EXPORT_SYMBOL_GPL(xstate_size);
121 static struct i387_fxsave_struct fx_scratch;
122
123 static void mxcsr_feature_mask_init(void)
124 {
125 unsigned long mask = 0;
126
127 if (cpu_has_fxsr) {
128 memset(&fx_scratch, 0, sizeof(struct i387_fxsave_struct));
129 asm volatile("fxsave %0" : "+m" (fx_scratch));
130 mask = fx_scratch.mxcsr_mask;
131 if (mask == 0)
132 mask = 0x0000ffbf;
133 }
134 mxcsr_feature_mask &= mask;
135 }
136
137 static void init_thread_xstate(void)
138 {
139 /*
140 * Note that xstate_size might be overwriten later during
141 * xsave_init().
142 */
143
144 if (!cpu_has_fpu) {
145 /*
146 * Disable xsave as we do not support it if i387
147 * emulation is enabled.
148 */
149 setup_clear_cpu_cap(X86_FEATURE_XSAVE);
150 setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT);
151 xstate_size = sizeof(struct i387_soft_struct);
152 return;
153 }
154
155 if (cpu_has_fxsr)
156 xstate_size = sizeof(struct i387_fxsave_struct);
157 else
158 xstate_size = sizeof(struct i387_fsave_struct);
159 }
160
161 /*
162 * Called at bootup to set up the initial FPU state that is later cloned
163 * into all processes.
164 */
165
166 void fpu_init(void)
167 {
168 unsigned long cr0;
169 unsigned long cr4_mask = 0;
170
171 #ifndef CONFIG_MATH_EMULATION
172 if (!cpu_has_fpu) {
173 pr_emerg("No FPU found and no math emulation present\n");
174 pr_emerg("Giving up\n");
175 for (;;)
176 asm volatile("hlt");
177 }
178 #endif
179 if (cpu_has_fxsr)
180 cr4_mask |= X86_CR4_OSFXSR;
181 if (cpu_has_xmm)
182 cr4_mask |= X86_CR4_OSXMMEXCPT;
183 if (cr4_mask)
184 cr4_set_bits(cr4_mask);
185
186 cr0 = read_cr0();
187 cr0 &= ~(X86_CR0_TS|X86_CR0_EM); /* clear TS and EM */
188 if (!cpu_has_fpu)
189 cr0 |= X86_CR0_EM;
190 write_cr0(cr0);
191
192 /*
193 * init_thread_xstate is only called once to avoid overriding
194 * xstate_size during boot time or during CPU hotplug.
195 */
196 if (xstate_size == 0)
197 init_thread_xstate();
198
199 mxcsr_feature_mask_init();
200 xsave_init();
201 eager_fpu_init();
202 }
203
204 void fpu_finit(struct fpu *fpu)
205 {
206 if (!cpu_has_fpu) {
207 finit_soft_fpu(&fpu->state->soft);
208 return;
209 }
210
211 if (cpu_has_fxsr) {
212 fx_finit(&fpu->state->fxsave);
213 } else {
214 struct i387_fsave_struct *fp = &fpu->state->fsave;
215 memset(fp, 0, xstate_size);
216 fp->cwd = 0xffff037fu;
217 fp->swd = 0xffff0000u;
218 fp->twd = 0xffffffffu;
219 fp->fos = 0xffff0000u;
220 }
221 }
222 EXPORT_SYMBOL_GPL(fpu_finit);
223
224 /*
225 * The _current_ task is using the FPU for the first time
226 * so initialize it and set the mxcsr to its default
227 * value at reset if we support XMM instructions and then
228 * remember the current task has used the FPU.
229 */
230 int init_fpu(struct task_struct *tsk)
231 {
232 int ret;
233
234 if (tsk_used_math(tsk)) {
235 if (cpu_has_fpu && tsk == current)
236 unlazy_fpu(tsk);
237 tsk->thread.fpu.last_cpu = ~0;
238 return 0;
239 }
240
241 /*
242 * Memory allocation at the first usage of the FPU and other state.
243 */
244 ret = fpu_alloc(&tsk->thread.fpu);
245 if (ret)
246 return ret;
247
248 fpu_finit(&tsk->thread.fpu);
249
250 set_stopped_child_used_math(tsk);
251 return 0;
252 }
253 EXPORT_SYMBOL_GPL(init_fpu);
254
255 /*
256 * The xstateregs_active() routine is the same as the fpregs_active() routine,
257 * as the "regset->n" for the xstate regset will be updated based on the feature
258 * capabilites supported by the xsave.
259 */
260 int fpregs_active(struct task_struct *target, const struct user_regset *regset)
261 {
262 return tsk_used_math(target) ? regset->n : 0;
263 }
264
265 int xfpregs_active(struct task_struct *target, const struct user_regset *regset)
266 {
267 return (cpu_has_fxsr && tsk_used_math(target)) ? regset->n : 0;
268 }
269
270 int xfpregs_get(struct task_struct *target, const struct user_regset *regset,
271 unsigned int pos, unsigned int count,
272 void *kbuf, void __user *ubuf)
273 {
274 int ret;
275
276 if (!cpu_has_fxsr)
277 return -ENODEV;
278
279 ret = init_fpu(target);
280 if (ret)
281 return ret;
282
283 sanitize_i387_state(target);
284
285 return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
286 &target->thread.fpu.state->fxsave, 0, -1);
287 }
288
289 int xfpregs_set(struct task_struct *target, const struct user_regset *regset,
290 unsigned int pos, unsigned int count,
291 const void *kbuf, const void __user *ubuf)
292 {
293 int ret;
294
295 if (!cpu_has_fxsr)
296 return -ENODEV;
297
298 ret = init_fpu(target);
299 if (ret)
300 return ret;
301
302 sanitize_i387_state(target);
303
304 ret = user_regset_copyin(&pos, &count, &kbuf, &ubuf,
305 &target->thread.fpu.state->fxsave, 0, -1);
306
307 /*
308 * mxcsr reserved bits must be masked to zero for security reasons.
309 */
310 target->thread.fpu.state->fxsave.mxcsr &= mxcsr_feature_mask;
311
312 /*
313 * update the header bits in the xsave header, indicating the
314 * presence of FP and SSE state.
315 */
316 if (cpu_has_xsave)
317 target->thread.fpu.state->xsave.xsave_hdr.xstate_bv |= XSTATE_FPSSE;
318
319 return ret;
320 }
321
322 int xstateregs_get(struct task_struct *target, const struct user_regset *regset,
323 unsigned int pos, unsigned int count,
324 void *kbuf, void __user *ubuf)
325 {
326 int ret;
327
328 if (!cpu_has_xsave)
329 return -ENODEV;
330
331 ret = init_fpu(target);
332 if (ret)
333 return ret;
334
335 /*
336 * Copy the 48bytes defined by the software first into the xstate
337 * memory layout in the thread struct, so that we can copy the entire
338 * xstateregs to the user using one user_regset_copyout().
339 */
340 memcpy(&target->thread.fpu.state->fxsave.sw_reserved,
341 xstate_fx_sw_bytes, sizeof(xstate_fx_sw_bytes));
342
343 /*
344 * Copy the xstate memory layout.
345 */
346 ret = user_regset_copyout(&pos, &count, &kbuf, &ubuf,
347 &target->thread.fpu.state->xsave, 0, -1);
348 return ret;
349 }
350
351 int xstateregs_set(struct task_struct *target, const struct user_regset *regset,
352 unsigned int pos, unsigned int count,
353 const void *kbuf, const void __user *ubuf)
354 {
355 int ret;
356 struct xsave_hdr_struct *xsave_hdr;
357
358 if (!cpu_has_xsave)
359 return -ENODEV;
360
361 ret = init_fpu(target);
362 if (ret)
363 return ret;
364
365 ret = user_regset_copyin(&pos, &count, &kbuf, &ubuf,
366 &target->thread.fpu.state->xsave, 0, -1);
367
368 /*
369 * mxcsr reserved bits must be masked to zero for security reasons.
370 */
371 target->thread.fpu.state->fxsave.mxcsr &= mxcsr_feature_mask;
372
373 xsave_hdr = &target->thread.fpu.state->xsave.xsave_hdr;
374
375 xsave_hdr->xstate_bv &= pcntxt_mask;
376 /*
377 * These bits must be zero.
378 */
379 memset(xsave_hdr->reserved, 0, 48);
380
381 return ret;
382 }
383
384 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
385
386 /*
387 * FPU tag word conversions.
388 */
389
390 static inline unsigned short twd_i387_to_fxsr(unsigned short twd)
391 {
392 unsigned int tmp; /* to avoid 16 bit prefixes in the code */
393
394 /* Transform each pair of bits into 01 (valid) or 00 (empty) */
395 tmp = ~twd;
396 tmp = (tmp | (tmp>>1)) & 0x5555; /* 0V0V0V0V0V0V0V0V */
397 /* and move the valid bits to the lower byte. */
398 tmp = (tmp | (tmp >> 1)) & 0x3333; /* 00VV00VV00VV00VV */
399 tmp = (tmp | (tmp >> 2)) & 0x0f0f; /* 0000VVVV0000VVVV */
400 tmp = (tmp | (tmp >> 4)) & 0x00ff; /* 00000000VVVVVVVV */
401
402 return tmp;
403 }
404
405 #define FPREG_ADDR(f, n) ((void *)&(f)->st_space + (n) * 16)
406 #define FP_EXP_TAG_VALID 0
407 #define FP_EXP_TAG_ZERO 1
408 #define FP_EXP_TAG_SPECIAL 2
409 #define FP_EXP_TAG_EMPTY 3
410
411 static inline u32 twd_fxsr_to_i387(struct i387_fxsave_struct *fxsave)
412 {
413 struct _fpxreg *st;
414 u32 tos = (fxsave->swd >> 11) & 7;
415 u32 twd = (unsigned long) fxsave->twd;
416 u32 tag;
417 u32 ret = 0xffff0000u;
418 int i;
419
420 for (i = 0; i < 8; i++, twd >>= 1) {
421 if (twd & 0x1) {
422 st = FPREG_ADDR(fxsave, (i - tos) & 7);
423
424 switch (st->exponent & 0x7fff) {
425 case 0x7fff:
426 tag = FP_EXP_TAG_SPECIAL;
427 break;
428 case 0x0000:
429 if (!st->significand[0] &&
430 !st->significand[1] &&
431 !st->significand[2] &&
432 !st->significand[3])
433 tag = FP_EXP_TAG_ZERO;
434 else
435 tag = FP_EXP_TAG_SPECIAL;
436 break;
437 default:
438 if (st->significand[3] & 0x8000)
439 tag = FP_EXP_TAG_VALID;
440 else
441 tag = FP_EXP_TAG_SPECIAL;
442 break;
443 }
444 } else {
445 tag = FP_EXP_TAG_EMPTY;
446 }
447 ret |= tag << (2 * i);
448 }
449 return ret;
450 }
451
452 /*
453 * FXSR floating point environment conversions.
454 */
455
456 void
457 convert_from_fxsr(struct user_i387_ia32_struct *env, struct task_struct *tsk)
458 {
459 struct i387_fxsave_struct *fxsave = &tsk->thread.fpu.state->fxsave;
460 struct _fpreg *to = (struct _fpreg *) &env->st_space[0];
461 struct _fpxreg *from = (struct _fpxreg *) &fxsave->st_space[0];
462 int i;
463
464 env->cwd = fxsave->cwd | 0xffff0000u;
465 env->swd = fxsave->swd | 0xffff0000u;
466 env->twd = twd_fxsr_to_i387(fxsave);
467
468 #ifdef CONFIG_X86_64
469 env->fip = fxsave->rip;
470 env->foo = fxsave->rdp;
471 /*
472 * should be actually ds/cs at fpu exception time, but
473 * that information is not available in 64bit mode.
474 */
475 env->fcs = task_pt_regs(tsk)->cs;
476 if (tsk == current) {
477 savesegment(ds, env->fos);
478 } else {
479 env->fos = tsk->thread.ds;
480 }
481 env->fos |= 0xffff0000;
482 #else
483 env->fip = fxsave->fip;
484 env->fcs = (u16) fxsave->fcs | ((u32) fxsave->fop << 16);
485 env->foo = fxsave->foo;
486 env->fos = fxsave->fos;
487 #endif
488
489 for (i = 0; i < 8; ++i)
490 memcpy(&to[i], &from[i], sizeof(to[0]));
491 }
492
493 void convert_to_fxsr(struct task_struct *tsk,
494 const struct user_i387_ia32_struct *env)
495
496 {
497 struct i387_fxsave_struct *fxsave = &tsk->thread.fpu.state->fxsave;
498 struct _fpreg *from = (struct _fpreg *) &env->st_space[0];
499 struct _fpxreg *to = (struct _fpxreg *) &fxsave->st_space[0];
500 int i;
501
502 fxsave->cwd = env->cwd;
503 fxsave->swd = env->swd;
504 fxsave->twd = twd_i387_to_fxsr(env->twd);
505 fxsave->fop = (u16) ((u32) env->fcs >> 16);
506 #ifdef CONFIG_X86_64
507 fxsave->rip = env->fip;
508 fxsave->rdp = env->foo;
509 /* cs and ds ignored */
510 #else
511 fxsave->fip = env->fip;
512 fxsave->fcs = (env->fcs & 0xffff);
513 fxsave->foo = env->foo;
514 fxsave->fos = env->fos;
515 #endif
516
517 for (i = 0; i < 8; ++i)
518 memcpy(&to[i], &from[i], sizeof(from[0]));
519 }
520
521 int fpregs_get(struct task_struct *target, const struct user_regset *regset,
522 unsigned int pos, unsigned int count,
523 void *kbuf, void __user *ubuf)
524 {
525 struct user_i387_ia32_struct env;
526 int ret;
527
528 ret = init_fpu(target);
529 if (ret)
530 return ret;
531
532 if (!static_cpu_has(X86_FEATURE_FPU))
533 return fpregs_soft_get(target, regset, pos, count, kbuf, ubuf);
534
535 if (!cpu_has_fxsr)
536 return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
537 &target->thread.fpu.state->fsave, 0,
538 -1);
539
540 sanitize_i387_state(target);
541
542 if (kbuf && pos == 0 && count == sizeof(env)) {
543 convert_from_fxsr(kbuf, target);
544 return 0;
545 }
546
547 convert_from_fxsr(&env, target);
548
549 return user_regset_copyout(&pos, &count, &kbuf, &ubuf, &env, 0, -1);
550 }
551
552 int fpregs_set(struct task_struct *target, const struct user_regset *regset,
553 unsigned int pos, unsigned int count,
554 const void *kbuf, const void __user *ubuf)
555 {
556 struct user_i387_ia32_struct env;
557 int ret;
558
559 ret = init_fpu(target);
560 if (ret)
561 return ret;
562
563 sanitize_i387_state(target);
564
565 if (!static_cpu_has(X86_FEATURE_FPU))
566 return fpregs_soft_set(target, regset, pos, count, kbuf, ubuf);
567
568 if (!cpu_has_fxsr)
569 return user_regset_copyin(&pos, &count, &kbuf, &ubuf,
570 &target->thread.fpu.state->fsave, 0,
571 -1);
572
573 if (pos > 0 || count < sizeof(env))
574 convert_from_fxsr(&env, target);
575
576 ret = user_regset_copyin(&pos, &count, &kbuf, &ubuf, &env, 0, -1);
577 if (!ret)
578 convert_to_fxsr(target, &env);
579
580 /*
581 * update the header bit in the xsave header, indicating the
582 * presence of FP.
583 */
584 if (cpu_has_xsave)
585 target->thread.fpu.state->xsave.xsave_hdr.xstate_bv |= XSTATE_FP;
586 return ret;
587 }
588
589 /*
590 * FPU state for core dumps.
591 * This is only used for a.out dumps now.
592 * It is declared generically using elf_fpregset_t (which is
593 * struct user_i387_struct) but is in fact only used for 32-bit
594 * dumps, so on 64-bit it is really struct user_i387_ia32_struct.
595 */
596 int dump_fpu(struct pt_regs *regs, struct user_i387_struct *fpu)
597 {
598 struct task_struct *tsk = current;
599 int fpvalid;
600
601 fpvalid = !!used_math();
602 if (fpvalid)
603 fpvalid = !fpregs_get(tsk, NULL,
604 0, sizeof(struct user_i387_ia32_struct),
605 fpu, NULL);
606
607 return fpvalid;
608 }
609 EXPORT_SYMBOL(dump_fpu);
610
611 #endif /* CONFIG_X86_32 || CONFIG_IA32_EMULATION */
612
613 static int __init no_387(char *s)
614 {
615 setup_clear_cpu_cap(X86_FEATURE_FPU);
616 return 1;
617 }
618
619 __setup("no387", no_387);
620
621 void fpu_detect(struct cpuinfo_x86 *c)
622 {
623 unsigned long cr0;
624 u16 fsw, fcw;
625
626 fsw = fcw = 0xffff;
627
628 cr0 = read_cr0();
629 cr0 &= ~(X86_CR0_TS | X86_CR0_EM);
630 write_cr0(cr0);
631
632 asm volatile("fninit ; fnstsw %0 ; fnstcw %1"
633 : "+m" (fsw), "+m" (fcw));
634
635 if (fsw == 0 && (fcw & 0x103f) == 0x003f)
636 set_cpu_cap(c, X86_FEATURE_FPU);
637 else
638 clear_cpu_cap(c, X86_FEATURE_FPU);
639
640 /* The final cr0 value is set in fpu_init() */
641 }
This page took 0.154758 seconds and 5 git commands to generate.