Merge branch 'next' of git://git.monstr.eu/linux-2.6-microblaze
[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/uaccess.h>
17 #include <asm/ptrace.h>
18 #include <asm/i387.h>
19 #include <asm/fpu-internal.h>
20 #include <asm/user.h>
21
22 /*
23 * Were we in an interrupt that interrupted kernel mode?
24 *
25 * For now, with eagerfpu we will return interrupted kernel FPU
26 * state as not-idle. TBD: Ideally we can change the return value
27 * to something like __thread_has_fpu(current). But we need to
28 * be careful of doing __thread_clear_has_fpu() before saving
29 * the FPU etc for supporting nested uses etc. For now, take
30 * the simple route!
31 *
32 * On others, we can do a kernel_fpu_begin/end() pair *ONLY* if that
33 * pair does nothing at all: the thread must not have fpu (so
34 * that we don't try to save the FPU state), and TS must
35 * be set (so that the clts/stts pair does nothing that is
36 * visible in the interrupted kernel thread).
37 */
38 static inline bool interrupted_kernel_fpu_idle(void)
39 {
40 if (use_eager_fpu())
41 return 0;
42
43 return !__thread_has_fpu(current) &&
44 (read_cr0() & X86_CR0_TS);
45 }
46
47 /*
48 * Were we in user mode (or vm86 mode) when we were
49 * interrupted?
50 *
51 * Doing kernel_fpu_begin/end() is ok if we are running
52 * in an interrupt context from user mode - we'll just
53 * save the FPU state as required.
54 */
55 static inline bool interrupted_user_mode(void)
56 {
57 struct pt_regs *regs = get_irq_regs();
58 return regs && user_mode_vm(regs);
59 }
60
61 /*
62 * Can we use the FPU in kernel mode with the
63 * whole "kernel_fpu_begin/end()" sequence?
64 *
65 * It's always ok in process context (ie "not interrupt")
66 * but it is sometimes ok even from an irq.
67 */
68 bool irq_fpu_usable(void)
69 {
70 return !in_interrupt() ||
71 interrupted_user_mode() ||
72 interrupted_kernel_fpu_idle();
73 }
74 EXPORT_SYMBOL(irq_fpu_usable);
75
76 void __kernel_fpu_begin(void)
77 {
78 struct task_struct *me = current;
79
80 if (__thread_has_fpu(me)) {
81 __save_init_fpu(me);
82 __thread_clear_has_fpu(me);
83 /* We do 'stts()' in __kernel_fpu_end() */
84 } else if (!use_eager_fpu()) {
85 this_cpu_write(fpu_owner_task, NULL);
86 clts();
87 }
88 }
89 EXPORT_SYMBOL(__kernel_fpu_begin);
90
91 void __kernel_fpu_end(void)
92 {
93 if (use_eager_fpu())
94 math_state_restore();
95 else
96 stts();
97 }
98 EXPORT_SYMBOL(__kernel_fpu_end);
99
100 void unlazy_fpu(struct task_struct *tsk)
101 {
102 preempt_disable();
103 if (__thread_has_fpu(tsk)) {
104 __save_init_fpu(tsk);
105 __thread_fpu_end(tsk);
106 } else
107 tsk->fpu_counter = 0;
108 preempt_enable();
109 }
110 EXPORT_SYMBOL(unlazy_fpu);
111
112 unsigned int mxcsr_feature_mask __read_mostly = 0xffffffffu;
113 unsigned int xstate_size;
114 EXPORT_SYMBOL_GPL(xstate_size);
115 static struct i387_fxsave_struct fx_scratch __cpuinitdata;
116
117 static void __cpuinit mxcsr_feature_mask_init(void)
118 {
119 unsigned long mask = 0;
120
121 if (cpu_has_fxsr) {
122 memset(&fx_scratch, 0, sizeof(struct i387_fxsave_struct));
123 asm volatile("fxsave %0" : : "m" (fx_scratch));
124 mask = fx_scratch.mxcsr_mask;
125 if (mask == 0)
126 mask = 0x0000ffbf;
127 }
128 mxcsr_feature_mask &= mask;
129 }
130
131 static void __cpuinit init_thread_xstate(void)
132 {
133 /*
134 * Note that xstate_size might be overwriten later during
135 * xsave_init().
136 */
137
138 if (!HAVE_HWFP) {
139 /*
140 * Disable xsave as we do not support it if i387
141 * emulation is enabled.
142 */
143 setup_clear_cpu_cap(X86_FEATURE_XSAVE);
144 setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT);
145 xstate_size = sizeof(struct i387_soft_struct);
146 return;
147 }
148
149 if (cpu_has_fxsr)
150 xstate_size = sizeof(struct i387_fxsave_struct);
151 else
152 xstate_size = sizeof(struct i387_fsave_struct);
153 }
154
155 /*
156 * Called at bootup to set up the initial FPU state that is later cloned
157 * into all processes.
158 */
159
160 void __cpuinit fpu_init(void)
161 {
162 unsigned long cr0;
163 unsigned long cr4_mask = 0;
164
165 if (cpu_has_fxsr)
166 cr4_mask |= X86_CR4_OSFXSR;
167 if (cpu_has_xmm)
168 cr4_mask |= X86_CR4_OSXMMEXCPT;
169 if (cr4_mask)
170 set_in_cr4(cr4_mask);
171
172 cr0 = read_cr0();
173 cr0 &= ~(X86_CR0_TS|X86_CR0_EM); /* clear TS and EM */
174 if (!HAVE_HWFP)
175 cr0 |= X86_CR0_EM;
176 write_cr0(cr0);
177
178 if (!smp_processor_id())
179 init_thread_xstate();
180
181 mxcsr_feature_mask_init();
182 xsave_init();
183 eager_fpu_init();
184 }
185
186 void fpu_finit(struct fpu *fpu)
187 {
188 if (!HAVE_HWFP) {
189 finit_soft_fpu(&fpu->state->soft);
190 return;
191 }
192
193 if (cpu_has_fxsr) {
194 fx_finit(&fpu->state->fxsave);
195 } else {
196 struct i387_fsave_struct *fp = &fpu->state->fsave;
197 memset(fp, 0, xstate_size);
198 fp->cwd = 0xffff037fu;
199 fp->swd = 0xffff0000u;
200 fp->twd = 0xffffffffu;
201 fp->fos = 0xffff0000u;
202 }
203 }
204 EXPORT_SYMBOL_GPL(fpu_finit);
205
206 /*
207 * The _current_ task is using the FPU for the first time
208 * so initialize it and set the mxcsr to its default
209 * value at reset if we support XMM instructions and then
210 * remember the current task has used the FPU.
211 */
212 int init_fpu(struct task_struct *tsk)
213 {
214 int ret;
215
216 if (tsk_used_math(tsk)) {
217 if (HAVE_HWFP && tsk == current)
218 unlazy_fpu(tsk);
219 tsk->thread.fpu.last_cpu = ~0;
220 return 0;
221 }
222
223 /*
224 * Memory allocation at the first usage of the FPU and other state.
225 */
226 ret = fpu_alloc(&tsk->thread.fpu);
227 if (ret)
228 return ret;
229
230 fpu_finit(&tsk->thread.fpu);
231
232 set_stopped_child_used_math(tsk);
233 return 0;
234 }
235 EXPORT_SYMBOL_GPL(init_fpu);
236
237 /*
238 * The xstateregs_active() routine is the same as the fpregs_active() routine,
239 * as the "regset->n" for the xstate regset will be updated based on the feature
240 * capabilites supported by the xsave.
241 */
242 int fpregs_active(struct task_struct *target, const struct user_regset *regset)
243 {
244 return tsk_used_math(target) ? regset->n : 0;
245 }
246
247 int xfpregs_active(struct task_struct *target, const struct user_regset *regset)
248 {
249 return (cpu_has_fxsr && tsk_used_math(target)) ? regset->n : 0;
250 }
251
252 int xfpregs_get(struct task_struct *target, const struct user_regset *regset,
253 unsigned int pos, unsigned int count,
254 void *kbuf, void __user *ubuf)
255 {
256 int ret;
257
258 if (!cpu_has_fxsr)
259 return -ENODEV;
260
261 ret = init_fpu(target);
262 if (ret)
263 return ret;
264
265 sanitize_i387_state(target);
266
267 return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
268 &target->thread.fpu.state->fxsave, 0, -1);
269 }
270
271 int xfpregs_set(struct task_struct *target, const struct user_regset *regset,
272 unsigned int pos, unsigned int count,
273 const void *kbuf, const void __user *ubuf)
274 {
275 int ret;
276
277 if (!cpu_has_fxsr)
278 return -ENODEV;
279
280 ret = init_fpu(target);
281 if (ret)
282 return ret;
283
284 sanitize_i387_state(target);
285
286 ret = user_regset_copyin(&pos, &count, &kbuf, &ubuf,
287 &target->thread.fpu.state->fxsave, 0, -1);
288
289 /*
290 * mxcsr reserved bits must be masked to zero for security reasons.
291 */
292 target->thread.fpu.state->fxsave.mxcsr &= mxcsr_feature_mask;
293
294 /*
295 * update the header bits in the xsave header, indicating the
296 * presence of FP and SSE state.
297 */
298 if (cpu_has_xsave)
299 target->thread.fpu.state->xsave.xsave_hdr.xstate_bv |= XSTATE_FPSSE;
300
301 return ret;
302 }
303
304 int xstateregs_get(struct task_struct *target, const struct user_regset *regset,
305 unsigned int pos, unsigned int count,
306 void *kbuf, void __user *ubuf)
307 {
308 int ret;
309
310 if (!cpu_has_xsave)
311 return -ENODEV;
312
313 ret = init_fpu(target);
314 if (ret)
315 return ret;
316
317 /*
318 * Copy the 48bytes defined by the software first into the xstate
319 * memory layout in the thread struct, so that we can copy the entire
320 * xstateregs to the user using one user_regset_copyout().
321 */
322 memcpy(&target->thread.fpu.state->fxsave.sw_reserved,
323 xstate_fx_sw_bytes, sizeof(xstate_fx_sw_bytes));
324
325 /*
326 * Copy the xstate memory layout.
327 */
328 ret = user_regset_copyout(&pos, &count, &kbuf, &ubuf,
329 &target->thread.fpu.state->xsave, 0, -1);
330 return ret;
331 }
332
333 int xstateregs_set(struct task_struct *target, const struct user_regset *regset,
334 unsigned int pos, unsigned int count,
335 const void *kbuf, const void __user *ubuf)
336 {
337 int ret;
338 struct xsave_hdr_struct *xsave_hdr;
339
340 if (!cpu_has_xsave)
341 return -ENODEV;
342
343 ret = init_fpu(target);
344 if (ret)
345 return ret;
346
347 ret = user_regset_copyin(&pos, &count, &kbuf, &ubuf,
348 &target->thread.fpu.state->xsave, 0, -1);
349
350 /*
351 * mxcsr reserved bits must be masked to zero for security reasons.
352 */
353 target->thread.fpu.state->fxsave.mxcsr &= mxcsr_feature_mask;
354
355 xsave_hdr = &target->thread.fpu.state->xsave.xsave_hdr;
356
357 xsave_hdr->xstate_bv &= pcntxt_mask;
358 /*
359 * These bits must be zero.
360 */
361 xsave_hdr->reserved1[0] = xsave_hdr->reserved1[1] = 0;
362
363 return ret;
364 }
365
366 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
367
368 /*
369 * FPU tag word conversions.
370 */
371
372 static inline unsigned short twd_i387_to_fxsr(unsigned short twd)
373 {
374 unsigned int tmp; /* to avoid 16 bit prefixes in the code */
375
376 /* Transform each pair of bits into 01 (valid) or 00 (empty) */
377 tmp = ~twd;
378 tmp = (tmp | (tmp>>1)) & 0x5555; /* 0V0V0V0V0V0V0V0V */
379 /* and move the valid bits to the lower byte. */
380 tmp = (tmp | (tmp >> 1)) & 0x3333; /* 00VV00VV00VV00VV */
381 tmp = (tmp | (tmp >> 2)) & 0x0f0f; /* 0000VVVV0000VVVV */
382 tmp = (tmp | (tmp >> 4)) & 0x00ff; /* 00000000VVVVVVVV */
383
384 return tmp;
385 }
386
387 #define FPREG_ADDR(f, n) ((void *)&(f)->st_space + (n) * 16)
388 #define FP_EXP_TAG_VALID 0
389 #define FP_EXP_TAG_ZERO 1
390 #define FP_EXP_TAG_SPECIAL 2
391 #define FP_EXP_TAG_EMPTY 3
392
393 static inline u32 twd_fxsr_to_i387(struct i387_fxsave_struct *fxsave)
394 {
395 struct _fpxreg *st;
396 u32 tos = (fxsave->swd >> 11) & 7;
397 u32 twd = (unsigned long) fxsave->twd;
398 u32 tag;
399 u32 ret = 0xffff0000u;
400 int i;
401
402 for (i = 0; i < 8; i++, twd >>= 1) {
403 if (twd & 0x1) {
404 st = FPREG_ADDR(fxsave, (i - tos) & 7);
405
406 switch (st->exponent & 0x7fff) {
407 case 0x7fff:
408 tag = FP_EXP_TAG_SPECIAL;
409 break;
410 case 0x0000:
411 if (!st->significand[0] &&
412 !st->significand[1] &&
413 !st->significand[2] &&
414 !st->significand[3])
415 tag = FP_EXP_TAG_ZERO;
416 else
417 tag = FP_EXP_TAG_SPECIAL;
418 break;
419 default:
420 if (st->significand[3] & 0x8000)
421 tag = FP_EXP_TAG_VALID;
422 else
423 tag = FP_EXP_TAG_SPECIAL;
424 break;
425 }
426 } else {
427 tag = FP_EXP_TAG_EMPTY;
428 }
429 ret |= tag << (2 * i);
430 }
431 return ret;
432 }
433
434 /*
435 * FXSR floating point environment conversions.
436 */
437
438 void
439 convert_from_fxsr(struct user_i387_ia32_struct *env, struct task_struct *tsk)
440 {
441 struct i387_fxsave_struct *fxsave = &tsk->thread.fpu.state->fxsave;
442 struct _fpreg *to = (struct _fpreg *) &env->st_space[0];
443 struct _fpxreg *from = (struct _fpxreg *) &fxsave->st_space[0];
444 int i;
445
446 env->cwd = fxsave->cwd | 0xffff0000u;
447 env->swd = fxsave->swd | 0xffff0000u;
448 env->twd = twd_fxsr_to_i387(fxsave);
449
450 #ifdef CONFIG_X86_64
451 env->fip = fxsave->rip;
452 env->foo = fxsave->rdp;
453 /*
454 * should be actually ds/cs at fpu exception time, but
455 * that information is not available in 64bit mode.
456 */
457 env->fcs = task_pt_regs(tsk)->cs;
458 if (tsk == current) {
459 savesegment(ds, env->fos);
460 } else {
461 env->fos = tsk->thread.ds;
462 }
463 env->fos |= 0xffff0000;
464 #else
465 env->fip = fxsave->fip;
466 env->fcs = (u16) fxsave->fcs | ((u32) fxsave->fop << 16);
467 env->foo = fxsave->foo;
468 env->fos = fxsave->fos;
469 #endif
470
471 for (i = 0; i < 8; ++i)
472 memcpy(&to[i], &from[i], sizeof(to[0]));
473 }
474
475 void convert_to_fxsr(struct task_struct *tsk,
476 const struct user_i387_ia32_struct *env)
477
478 {
479 struct i387_fxsave_struct *fxsave = &tsk->thread.fpu.state->fxsave;
480 struct _fpreg *from = (struct _fpreg *) &env->st_space[0];
481 struct _fpxreg *to = (struct _fpxreg *) &fxsave->st_space[0];
482 int i;
483
484 fxsave->cwd = env->cwd;
485 fxsave->swd = env->swd;
486 fxsave->twd = twd_i387_to_fxsr(env->twd);
487 fxsave->fop = (u16) ((u32) env->fcs >> 16);
488 #ifdef CONFIG_X86_64
489 fxsave->rip = env->fip;
490 fxsave->rdp = env->foo;
491 /* cs and ds ignored */
492 #else
493 fxsave->fip = env->fip;
494 fxsave->fcs = (env->fcs & 0xffff);
495 fxsave->foo = env->foo;
496 fxsave->fos = env->fos;
497 #endif
498
499 for (i = 0; i < 8; ++i)
500 memcpy(&to[i], &from[i], sizeof(from[0]));
501 }
502
503 int fpregs_get(struct task_struct *target, const struct user_regset *regset,
504 unsigned int pos, unsigned int count,
505 void *kbuf, void __user *ubuf)
506 {
507 struct user_i387_ia32_struct env;
508 int ret;
509
510 ret = init_fpu(target);
511 if (ret)
512 return ret;
513
514 if (!HAVE_HWFP)
515 return fpregs_soft_get(target, regset, pos, count, kbuf, ubuf);
516
517 if (!cpu_has_fxsr) {
518 return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
519 &target->thread.fpu.state->fsave, 0,
520 -1);
521 }
522
523 sanitize_i387_state(target);
524
525 if (kbuf && pos == 0 && count == sizeof(env)) {
526 convert_from_fxsr(kbuf, target);
527 return 0;
528 }
529
530 convert_from_fxsr(&env, target);
531
532 return user_regset_copyout(&pos, &count, &kbuf, &ubuf, &env, 0, -1);
533 }
534
535 int fpregs_set(struct task_struct *target, const struct user_regset *regset,
536 unsigned int pos, unsigned int count,
537 const void *kbuf, const void __user *ubuf)
538 {
539 struct user_i387_ia32_struct env;
540 int ret;
541
542 ret = init_fpu(target);
543 if (ret)
544 return ret;
545
546 sanitize_i387_state(target);
547
548 if (!HAVE_HWFP)
549 return fpregs_soft_set(target, regset, pos, count, kbuf, ubuf);
550
551 if (!cpu_has_fxsr) {
552 return user_regset_copyin(&pos, &count, &kbuf, &ubuf,
553 &target->thread.fpu.state->fsave, 0, -1);
554 }
555
556 if (pos > 0 || count < sizeof(env))
557 convert_from_fxsr(&env, target);
558
559 ret = user_regset_copyin(&pos, &count, &kbuf, &ubuf, &env, 0, -1);
560 if (!ret)
561 convert_to_fxsr(target, &env);
562
563 /*
564 * update the header bit in the xsave header, indicating the
565 * presence of FP.
566 */
567 if (cpu_has_xsave)
568 target->thread.fpu.state->xsave.xsave_hdr.xstate_bv |= XSTATE_FP;
569 return ret;
570 }
571
572 /*
573 * FPU state for core dumps.
574 * This is only used for a.out dumps now.
575 * It is declared generically using elf_fpregset_t (which is
576 * struct user_i387_struct) but is in fact only used for 32-bit
577 * dumps, so on 64-bit it is really struct user_i387_ia32_struct.
578 */
579 int dump_fpu(struct pt_regs *regs, struct user_i387_struct *fpu)
580 {
581 struct task_struct *tsk = current;
582 int fpvalid;
583
584 fpvalid = !!used_math();
585 if (fpvalid)
586 fpvalid = !fpregs_get(tsk, NULL,
587 0, sizeof(struct user_i387_ia32_struct),
588 fpu, NULL);
589
590 return fpvalid;
591 }
592 EXPORT_SYMBOL(dump_fpu);
593
594 #endif /* CONFIG_X86_32 || CONFIG_IA32_EMULATION */
This page took 0.111488 seconds and 6 git commands to generate.