x86/fpu: Clean up regset functions
[deliverable/linux.git] / arch / x86 / kernel / fpu / xsave.c
CommitLineData
dc1e35c6
SS
1/*
2 * xsave/xrstor support.
3 *
4 * Author: Suresh Siddha <suresh.b.siddha@intel.com>
5 */
6#include <linux/bootmem.h>
7#include <linux/compat.h>
7e7ce87f 8#include <linux/cpu.h>
df6b35f4 9#include <asm/fpu/api.h>
78f7f1e5 10#include <asm/fpu/internal.h>
72a671ce 11#include <asm/sigframe.h>
375074cc 12#include <asm/tlbflush.h>
6152e4b1 13#include <asm/xcr.h>
dc1e35c6
SS
14
15/*
614df7fb 16 * Mask of xstate features supported by the CPU and the kernel:
dc1e35c6 17 */
614df7fb 18u64 xfeatures_mask;
dc1e35c6 19
45c2d7f4
RR
20/*
21 * Represents init state for the supported extended state.
22 */
304bceda 23struct xsave_struct *init_xstate_buf;
45c2d7f4 24
72a671ce 25static struct _fpx_sw_bytes fx_sw_reserved, fx_sw_reserved_ia32;
7e7ce87f 26static unsigned int *xstate_offsets, *xstate_sizes;
614df7fb 27static unsigned int xstate_comp_offsets[sizeof(xfeatures_mask)*8];
84246fe4
IM
28
29/* The number of supported xfeatures in xfeatures_mask: */
30static unsigned int xfeatures_nr;
a1488f8b 31
29104e10
SS
32/*
33 * If a processor implementation discern that a processor state component is
34 * in its initialized state it may modify the corresponding bit in the
35 * xsave_hdr.xstate_bv as '0', with out modifying the corresponding memory
36 * layout in the case of xsaveopt. While presenting the xstate information to
37 * the user, we always ensure that the memory layout of a feature will be in
38 * the init state if the corresponding header bit is zero. This is to ensure
39 * that the user doesn't see some stale state in the memory layout during
40 * signal handling, debugging etc.
41 */
42void __sanitize_i387_state(struct task_struct *tsk)
43{
29104e10 44 struct i387_fxsave_struct *fx = &tsk->thread.fpu.state->fxsave;
72a671ce
SS
45 int feature_bit = 0x2;
46 u64 xstate_bv;
29104e10
SS
47
48 if (!fx)
49 return;
50
29104e10
SS
51 xstate_bv = tsk->thread.fpu.state->xsave.xsave_hdr.xstate_bv;
52
53 /*
54 * None of the feature bits are in init state. So nothing else
0d2eb44f 55 * to do for us, as the memory layout is up to date.
29104e10 56 */
614df7fb 57 if ((xstate_bv & xfeatures_mask) == xfeatures_mask)
29104e10
SS
58 return;
59
60 /*
61 * FP is in init state
62 */
63 if (!(xstate_bv & XSTATE_FP)) {
64 fx->cwd = 0x37f;
65 fx->swd = 0;
66 fx->twd = 0;
67 fx->fop = 0;
68 fx->rip = 0;
69 fx->rdp = 0;
70 memset(&fx->st_space[0], 0, 128);
71 }
72
73 /*
74 * SSE is in init state
75 */
76 if (!(xstate_bv & XSTATE_SSE))
77 memset(&fx->xmm_space[0], 0, 256);
78
614df7fb 79 xstate_bv = (xfeatures_mask & ~xstate_bv) >> 2;
29104e10
SS
80
81 /*
82 * Update all the other memory layouts for which the corresponding
83 * header bit is in the init state.
84 */
85 while (xstate_bv) {
86 if (xstate_bv & 0x1) {
87 int offset = xstate_offsets[feature_bit];
88 int size = xstate_sizes[feature_bit];
89
90 memcpy(((void *) fx) + offset,
91 ((void *) init_xstate_buf) + offset,
92 size);
93 }
94
95 xstate_bv >>= 1;
96 feature_bit++;
97 }
98}
99
c37b5efe
SS
100/*
101 * Check for the presence of extended state information in the
102 * user fpstate pointer in the sigcontext.
103 */
72a671ce
SS
104static inline int check_for_xstate(struct i387_fxsave_struct __user *buf,
105 void __user *fpstate,
106 struct _fpx_sw_bytes *fx_sw)
c37b5efe
SS
107{
108 int min_xstate_size = sizeof(struct i387_fxsave_struct) +
109 sizeof(struct xsave_hdr_struct);
110 unsigned int magic2;
c37b5efe 111
72a671ce
SS
112 if (__copy_from_user(fx_sw, &buf->sw_reserved[0], sizeof(*fx_sw)))
113 return -1;
c37b5efe 114
72a671ce
SS
115 /* Check for the first magic field and other error scenarios. */
116 if (fx_sw->magic1 != FP_XSTATE_MAGIC1 ||
117 fx_sw->xstate_size < min_xstate_size ||
118 fx_sw->xstate_size > xstate_size ||
119 fx_sw->xstate_size > fx_sw->extended_size)
120 return -1;
c37b5efe 121
c37b5efe
SS
122 /*
123 * Check for the presence of second magic word at the end of memory
124 * layout. This detects the case where the user just copied the legacy
125 * fpstate layout with out copying the extended state information
126 * in the memory layout.
127 */
72a671ce
SS
128 if (__get_user(magic2, (__u32 __user *)(fpstate + fx_sw->xstate_size))
129 || magic2 != FP_XSTATE_MAGIC2)
130 return -1;
c37b5efe
SS
131
132 return 0;
133}
134
ab513701
SS
135/*
136 * Signal frame handlers.
137 */
72a671ce
SS
138static inline int save_fsave_header(struct task_struct *tsk, void __user *buf)
139{
140 if (use_fxsr()) {
141 struct xsave_struct *xsave = &tsk->thread.fpu.state->xsave;
142 struct user_i387_ia32_struct env;
143 struct _fpstate_ia32 __user *fp = buf;
ab513701 144
72a671ce
SS
145 convert_from_fxsr(&env, tsk);
146
147 if (__copy_to_user(buf, &env, sizeof(env)) ||
148 __put_user(xsave->i387.swd, &fp->status) ||
149 __put_user(X86_FXSR_MAGIC, &fp->magic))
150 return -1;
151 } else {
152 struct i387_fsave_struct __user *fp = buf;
153 u32 swd;
154 if (__get_user(swd, &fp->swd) || __put_user(swd, &fp->status))
155 return -1;
156 }
157
158 return 0;
159}
160
161static inline int save_xstate_epilog(void __user *buf, int ia32_frame)
ab513701 162{
72a671ce
SS
163 struct xsave_struct __user *x = buf;
164 struct _fpx_sw_bytes *sw_bytes;
165 u32 xstate_bv;
166 int err;
ab513701 167
72a671ce
SS
168 /* Setup the bytes not touched by the [f]xsave and reserved for SW. */
169 sw_bytes = ia32_frame ? &fx_sw_reserved_ia32 : &fx_sw_reserved;
170 err = __copy_to_user(&x->i387.sw_reserved, sw_bytes, sizeof(*sw_bytes));
ab513701 171
72a671ce
SS
172 if (!use_xsave())
173 return err;
ab513701 174
72a671ce 175 err |= __put_user(FP_XSTATE_MAGIC2, (__u32 *)(buf + xstate_size));
ab513701 176
72a671ce
SS
177 /*
178 * Read the xstate_bv which we copied (directly from the cpu or
179 * from the state in task struct) to the user buffers.
180 */
181 err |= __get_user(xstate_bv, (__u32 *)&x->xsave_hdr.xstate_bv);
06c38d5e 182
72a671ce
SS
183 /*
184 * For legacy compatible, we always set FP/SSE bits in the bit
185 * vector while saving the state to the user context. This will
186 * enable us capturing any changes(during sigreturn) to
187 * the FP/SSE bits by the legacy applications which don't touch
188 * xstate_bv in the xsave header.
189 *
190 * xsave aware apps can change the xstate_bv in the xsave
191 * header as well as change any contents in the memory layout.
192 * xrestore as part of sigreturn will capture all the changes.
193 */
194 xstate_bv |= XSTATE_FPSSE;
c37b5efe 195
72a671ce
SS
196 err |= __put_user(xstate_bv, (__u32 *)&x->xsave_hdr.xstate_bv);
197
198 return err;
199}
200
201static inline int save_user_xstate(struct xsave_struct __user *buf)
202{
203 int err;
204
205 if (use_xsave())
206 err = xsave_user(buf);
207 else if (use_fxsr())
208 err = fxsave_user((struct i387_fxsave_struct __user *) buf);
209 else
210 err = fsave_user((struct i387_fsave_struct __user *) buf);
211
212 if (unlikely(err) && __clear_user(buf, xstate_size))
213 err = -EFAULT;
214 return err;
215}
216
217/*
218 * Save the fpu, extended register state to the user signal frame.
219 *
220 * 'buf_fx' is the 64-byte aligned pointer at which the [f|fx|x]save
221 * state is copied.
222 * 'buf' points to the 'buf_fx' or to the fsave header followed by 'buf_fx'.
223 *
224 * buf == buf_fx for 64-bit frames and 32-bit fsave frame.
225 * buf != buf_fx for 32-bit frames with fxstate.
226 *
227 * If the fpu, extended register state is live, save the state directly
228 * to the user frame pointed by the aligned pointer 'buf_fx'. Otherwise,
229 * copy the thread's fpu state to the user frame starting at 'buf_fx'.
230 *
231 * If this is a 32-bit frame with fxstate, put a fsave header before
232 * the aligned state at 'buf_fx'.
233 *
234 * For [f]xsave state, update the SW reserved fields in the [f]xsave frame
235 * indicating the absence/presence of the extended state to the user.
236 */
237int save_xstate_sig(void __user *buf, void __user *buf_fx, int size)
238{
239 struct xsave_struct *xsave = &current->thread.fpu.state->xsave;
240 struct task_struct *tsk = current;
241 int ia32_fxstate = (buf != buf_fx);
242
243 ia32_fxstate &= (config_enabled(CONFIG_X86_32) ||
244 config_enabled(CONFIG_IA32_EMULATION));
245
246 if (!access_ok(VERIFY_WRITE, buf, size))
247 return -EACCES;
248
60e019eb 249 if (!static_cpu_has(X86_FEATURE_FPU))
72a671ce
SS
250 return fpregs_soft_get(current, NULL, 0,
251 sizeof(struct user_i387_ia32_struct), NULL,
252 (struct _fpstate_ia32 __user *) buf) ? -1 : 1;
253
254 if (user_has_fpu()) {
255 /* Save the live register state to the user directly. */
256 if (save_user_xstate(buf_fx))
257 return -1;
258 /* Update the thread's fxstate to save the fsave header. */
259 if (ia32_fxstate)
260 fpu_fxsave(&tsk->thread.fpu);
ab513701 261 } else {
29104e10 262 sanitize_i387_state(tsk);
72a671ce 263 if (__copy_to_user(buf_fx, xsave, xstate_size))
ab513701
SS
264 return -1;
265 }
c37b5efe 266
72a671ce
SS
267 /* Save the fsave header for the 32-bit frames. */
268 if ((ia32_fxstate || !use_fxsr()) && save_fsave_header(tsk, buf))
269 return -1;
06c38d5e 270
72a671ce
SS
271 if (use_fxsr() && save_xstate_epilog(buf_fx, ia32_fxstate))
272 return -1;
273
72a671ce
SS
274 return 0;
275}
c37b5efe 276
72a671ce
SS
277static inline void
278sanitize_restored_xstate(struct task_struct *tsk,
279 struct user_i387_ia32_struct *ia32_env,
280 u64 xstate_bv, int fx_only)
281{
282 struct xsave_struct *xsave = &tsk->thread.fpu.state->xsave;
283 struct xsave_hdr_struct *xsave_hdr = &xsave->xsave_hdr;
c37b5efe 284
72a671ce
SS
285 if (use_xsave()) {
286 /* These bits must be zero. */
7e7ce87f 287 memset(xsave_hdr->reserved, 0, 48);
04944b79
SS
288
289 /*
72a671ce
SS
290 * Init the state that is not present in the memory
291 * layout and not enabled by the OS.
04944b79 292 */
72a671ce
SS
293 if (fx_only)
294 xsave_hdr->xstate_bv = XSTATE_FPSSE;
295 else
614df7fb 296 xsave_hdr->xstate_bv &= (xfeatures_mask & xstate_bv);
72a671ce 297 }
04944b79 298
72a671ce 299 if (use_fxsr()) {
04944b79 300 /*
72a671ce
SS
301 * mscsr reserved bits must be masked to zero for security
302 * reasons.
04944b79 303 */
72a671ce 304 xsave->i387.mxcsr &= mxcsr_feature_mask;
04944b79 305
72a671ce 306 convert_to_fxsr(tsk, ia32_env);
c37b5efe 307 }
ab513701
SS
308}
309
c37b5efe 310/*
72a671ce 311 * Restore the extended state if present. Otherwise, restore the FP/SSE state.
c37b5efe 312 */
72a671ce 313static inline int restore_user_xstate(void __user *buf, u64 xbv, int fx_only)
c37b5efe 314{
72a671ce
SS
315 if (use_xsave()) {
316 if ((unsigned long)buf % 64 || fx_only) {
614df7fb 317 u64 init_bv = xfeatures_mask & ~XSTATE_FPSSE;
72a671ce 318 xrstor_state(init_xstate_buf, init_bv);
e139e955 319 return fxrstor_user(buf);
72a671ce 320 } else {
614df7fb 321 u64 init_bv = xfeatures_mask & ~xbv;
72a671ce
SS
322 if (unlikely(init_bv))
323 xrstor_state(init_xstate_buf, init_bv);
324 return xrestore_user(buf, xbv);
325 }
326 } else if (use_fxsr()) {
e139e955 327 return fxrstor_user(buf);
72a671ce 328 } else
e139e955 329 return frstor_user(buf);
c37b5efe
SS
330}
331
72a671ce 332int __restore_xstate_sig(void __user *buf, void __user *buf_fx, int size)
ab513701 333{
72a671ce 334 int ia32_fxstate = (buf != buf_fx);
ab513701 335 struct task_struct *tsk = current;
c5bedc68 336 struct fpu *fpu = &tsk->thread.fpu;
72a671ce
SS
337 int state_size = xstate_size;
338 u64 xstate_bv = 0;
339 int fx_only = 0;
340
341 ia32_fxstate &= (config_enabled(CONFIG_X86_32) ||
342 config_enabled(CONFIG_IA32_EMULATION));
ab513701
SS
343
344 if (!buf) {
af2d94fd 345 fpu_reset_state(fpu);
ab513701 346 return 0;
72a671ce
SS
347 }
348
349 if (!access_ok(VERIFY_READ, buf, size))
350 return -EACCES;
351
db2b1d3a 352 if (!fpu->fpstate_active && fpstate_alloc_init(fpu))
72a671ce 353 return -1;
ab513701 354
60e019eb 355 if (!static_cpu_has(X86_FEATURE_FPU))
72a671ce
SS
356 return fpregs_soft_set(current, NULL,
357 0, sizeof(struct user_i387_ia32_struct),
358 NULL, buf) != 0;
ab513701 359
72a671ce
SS
360 if (use_xsave()) {
361 struct _fpx_sw_bytes fx_sw_user;
362 if (unlikely(check_for_xstate(buf_fx, buf_fx, &fx_sw_user))) {
363 /*
364 * Couldn't find the extended state information in the
365 * memory layout. Restore just the FP/SSE and init all
366 * the other extended state.
367 */
368 state_size = sizeof(struct i387_fxsave_struct);
369 fx_only = 1;
370 } else {
371 state_size = fx_sw_user.xstate_size;
372 xstate_bv = fx_sw_user.xstate_bv;
373 }
374 }
375
376 if (ia32_fxstate) {
377 /*
378 * For 32-bit frames with fxstate, copy the user state to the
379 * thread's fpu state, reconstruct fxstate from the fsave
380 * header. Sanitize the copied state etc.
381 */
a7c80ebc 382 struct fpu *fpu = &tsk->thread.fpu;
72a671ce 383 struct user_i387_ia32_struct env;
304bceda 384 int err = 0;
72a671ce 385
304bceda 386 /*
c5bedc68 387 * Drop the current fpu which clears fpu->fpstate_active. This ensures
304bceda
SS
388 * that any context-switch during the copy of the new state,
389 * avoids the intermediate state from getting restored/saved.
390 * Thus avoiding the new restored state from getting corrupted.
391 * We will be ready to restore/save the state only after
c5bedc68 392 * fpu->fpstate_active is again set.
304bceda 393 */
ca6787ba 394 drop_fpu(fpu);
72a671ce 395
a7c80ebc 396 if (__copy_from_user(&fpu->state->xsave, buf_fx, state_size) ||
304bceda 397 __copy_from_user(&env, buf, sizeof(env))) {
c0ee2cf6 398 fpstate_init(fpu);
304bceda
SS
399 err = -1;
400 } else {
401 sanitize_restored_xstate(tsk, &env, xstate_bv, fx_only);
304bceda 402 }
72a671ce 403
c5bedc68 404 fpu->fpstate_active = 1;
df24fb85
ON
405 if (use_eager_fpu()) {
406 preempt_disable();
3a0aee48 407 fpu__restore();
df24fb85
ON
408 preempt_enable();
409 }
304bceda
SS
410
411 return err;
72a671ce 412 } else {
ab513701 413 /*
72a671ce
SS
414 * For 64-bit frames and 32-bit fsave frames, restore the user
415 * state to the registers directly (with exceptions handled).
ab513701 416 */
72a671ce
SS
417 user_fpu_begin();
418 if (restore_user_xstate(buf_fx, xstate_bv, fx_only)) {
af2d94fd 419 fpu_reset_state(fpu);
72a671ce
SS
420 return -1;
421 }
ab513701 422 }
72a671ce
SS
423
424 return 0;
ab513701 425}
ab513701 426
c37b5efe
SS
427/*
428 * Prepare the SW reserved portion of the fxsave memory layout, indicating
429 * the presence of the extended state information in the memory layout
430 * pointed by the fpstate pointer in the sigcontext.
431 * This will be saved when ever the FP and extended state context is
432 * saved on the user stack during the signal handler delivery to the user.
433 */
8bcad30f 434static void prepare_fx_sw_frame(void)
c37b5efe 435{
72a671ce
SS
436 int fsave_header_size = sizeof(struct i387_fsave_struct);
437 int size = xstate_size + FP_XSTATE_MAGIC2_SIZE;
c37b5efe 438
72a671ce
SS
439 if (config_enabled(CONFIG_X86_32))
440 size += fsave_header_size;
c37b5efe
SS
441
442 fx_sw_reserved.magic1 = FP_XSTATE_MAGIC1;
72a671ce 443 fx_sw_reserved.extended_size = size;
614df7fb 444 fx_sw_reserved.xstate_bv = xfeatures_mask;
c37b5efe 445 fx_sw_reserved.xstate_size = xstate_size;
c37b5efe 446
72a671ce
SS
447 if (config_enabled(CONFIG_IA32_EMULATION)) {
448 fx_sw_reserved_ia32 = fx_sw_reserved;
449 fx_sw_reserved_ia32.extended_size += fsave_header_size;
450 }
451}
3c1c7f10 452
dc1e35c6
SS
453/*
454 * Enable the extended processor state save/restore feature
455 */
1cff92d8 456static inline void xstate_enable(void)
dc1e35c6 457{
375074cc 458 cr4_set_bits(X86_CR4_OSXSAVE);
614df7fb 459 xsetbv(XCR_XFEATURE_ENABLED_MASK, xfeatures_mask);
dc1e35c6
SS
460}
461
a1488f8b
SS
462/*
463 * Record the offsets and sizes of different state managed by the xsave
464 * memory layout.
465 */
4995b9db 466static void __init setup_xstate_features(void)
a1488f8b
SS
467{
468 int eax, ebx, ecx, edx, leaf = 0x2;
469
84246fe4
IM
470 xfeatures_nr = fls64(xfeatures_mask);
471 xstate_offsets = alloc_bootmem(xfeatures_nr * sizeof(int));
472 xstate_sizes = alloc_bootmem(xfeatures_nr * sizeof(int));
a1488f8b
SS
473
474 do {
ee813d53 475 cpuid_count(XSTATE_CPUID, leaf, &eax, &ebx, &ecx, &edx);
a1488f8b
SS
476
477 if (eax == 0)
478 break;
479
480 xstate_offsets[leaf] = ebx;
481 xstate_sizes[leaf] = eax;
482
483 leaf++;
484 } while (1);
485}
486
69496e10
IM
487static void print_xstate_feature(u64 xstate_mask, const char *desc)
488{
614df7fb 489 if (xfeatures_mask & xstate_mask) {
69496e10
IM
490 int xstate_feature = fls64(xstate_mask)-1;
491
492 pr_info("x86/fpu: Supporting XSAVE feature %2d: '%s'\n", xstate_feature, desc);
493 }
494}
495
496/*
497 * Print out all the supported xstate features:
498 */
499static void print_xstate_features(void)
500{
501 print_xstate_feature(XSTATE_FP, "x87 floating point registers");
502 print_xstate_feature(XSTATE_SSE, "SSE registers");
503 print_xstate_feature(XSTATE_YMM, "AVX registers");
504 print_xstate_feature(XSTATE_BNDREGS, "MPX bounds registers");
505 print_xstate_feature(XSTATE_BNDCSR, "MPX CSR");
506 print_xstate_feature(XSTATE_OPMASK, "AVX-512 opmask");
507 print_xstate_feature(XSTATE_ZMM_Hi256, "AVX-512 Hi256");
508 print_xstate_feature(XSTATE_Hi16_ZMM, "AVX-512 ZMM_Hi256");
509}
510
7496d645
FY
511/*
512 * This function sets up offsets and sizes of all extended states in
513 * xsave area. This supports both standard format and compacted format
514 * of the xsave aread.
515 *
516 * Input: void
517 * Output: void
518 */
519void setup_xstate_comp(void)
520{
614df7fb 521 unsigned int xstate_comp_sizes[sizeof(xfeatures_mask)*8];
7496d645
FY
522 int i;
523
8ff925e1
FY
524 /*
525 * The FP xstates and SSE xstates are legacy states. They are always
526 * in the fixed offsets in the xsave area in either compacted form
527 * or standard form.
528 */
529 xstate_comp_offsets[0] = 0;
530 xstate_comp_offsets[1] = offsetof(struct i387_fxsave_struct, xmm_space);
7496d645
FY
531
532 if (!cpu_has_xsaves) {
84246fe4 533 for (i = 2; i < xfeatures_nr; i++) {
614df7fb 534 if (test_bit(i, (unsigned long *)&xfeatures_mask)) {
7496d645
FY
535 xstate_comp_offsets[i] = xstate_offsets[i];
536 xstate_comp_sizes[i] = xstate_sizes[i];
537 }
538 }
539 return;
540 }
541
542 xstate_comp_offsets[2] = FXSAVE_SIZE + XSAVE_HDR_SIZE;
543
84246fe4 544 for (i = 2; i < xfeatures_nr; i++) {
614df7fb 545 if (test_bit(i, (unsigned long *)&xfeatures_mask))
7496d645
FY
546 xstate_comp_sizes[i] = xstate_sizes[i];
547 else
548 xstate_comp_sizes[i] = 0;
549
550 if (i > 2)
551 xstate_comp_offsets[i] = xstate_comp_offsets[i-1]
552 + xstate_comp_sizes[i-1];
553
554 }
555}
556
dc1e35c6
SS
557/*
558 * setup the xstate image representing the init state
559 */
5d2bd700 560static void __init setup_init_fpu_buf(void)
dc1e35c6 561{
29104e10
SS
562 /*
563 * Setup init_xstate_buf to represent the init state of
564 * all the features managed by the xsave
565 */
10340ae1
SS
566 init_xstate_buf = alloc_bootmem_align(xstate_size,
567 __alignof__(struct xsave_struct));
5d2bd700
SS
568 fx_finit(&init_xstate_buf->i387);
569
570 if (!cpu_has_xsave)
571 return;
572
573 setup_xstate_features();
69496e10 574 print_xstate_features();
a1488f8b 575
47c2f292
FY
576 if (cpu_has_xsaves) {
577 init_xstate_buf->xsave_hdr.xcomp_bv =
614df7fb
IM
578 (u64)1 << 63 | xfeatures_mask;
579 init_xstate_buf->xsave_hdr.xstate_bv = xfeatures_mask;
47c2f292
FY
580 }
581
29104e10
SS
582 /*
583 * Init all the features state with header_bv being 0x0
584 */
47c2f292 585 xrstor_state_booting(init_xstate_buf, -1);
3e261c14 586
29104e10
SS
587 /*
588 * Dump the init state again. This is to identify the init state
589 * of any feature which is not represented by all zero's.
590 */
3e261c14 591 xsave_state_booting(init_xstate_buf);
dc1e35c6
SS
592}
593
e0022981 594static enum { AUTO, ENABLE, DISABLE } eagerfpu = AUTO;
5d2bd700
SS
595static int __init eager_fpu_setup(char *s)
596{
597 if (!strcmp(s, "on"))
e0022981 598 eagerfpu = ENABLE;
5d2bd700 599 else if (!strcmp(s, "off"))
e0022981
SS
600 eagerfpu = DISABLE;
601 else if (!strcmp(s, "auto"))
602 eagerfpu = AUTO;
5d2bd700
SS
603 return 1;
604}
605__setup("eagerfpu=", eager_fpu_setup);
606
7e7ce87f
FY
607
608/*
614df7fb 609 * Calculate total size of enabled xstates in XCR0/xfeatures_mask.
7e7ce87f
FY
610 */
611static void __init init_xstate_size(void)
612{
613 unsigned int eax, ebx, ecx, edx;
614 int i;
615
616 if (!cpu_has_xsaves) {
617 cpuid_count(XSTATE_CPUID, 0, &eax, &ebx, &ecx, &edx);
618 xstate_size = ebx;
619 return;
620 }
621
622 xstate_size = FXSAVE_SIZE + XSAVE_HDR_SIZE;
623 for (i = 2; i < 64; i++) {
614df7fb 624 if (test_bit(i, (unsigned long *)&xfeatures_mask)) {
7e7ce87f
FY
625 cpuid_count(XSTATE_CPUID, i, &eax, &ebx, &ecx, &edx);
626 xstate_size += eax;
627 }
628 }
629}
630
dc1e35c6
SS
631/*
632 * Enable and initialize the xsave feature.
c0841e34
IM
633 *
634 * ( Not marked __init because of false positive section warnings
635 * generated by xsave_init(). )
dc1e35c6 636 */
c0841e34 637static void /* __init */ xstate_enable_boot_cpu(void)
dc1e35c6
SS
638{
639 unsigned int eax, ebx, ecx, edx;
640
ee813d53 641 if (boot_cpu_data.cpuid_level < XSTATE_CPUID) {
32d4d9cc 642 WARN(1, "x86/fpu: XSTATE_CPUID missing!\n");
ee813d53
RR
643 return;
644 }
645
646 cpuid_count(XSTATE_CPUID, 0, &eax, &ebx, &ecx, &edx);
614df7fb 647 xfeatures_mask = eax + ((u64)edx << 32);
dc1e35c6 648
614df7fb
IM
649 if ((xfeatures_mask & XSTATE_FPSSE) != XSTATE_FPSSE) {
650 pr_err("x86/fpu: FP/SSE not present amongst the CPU's xstate features: 0x%llx.\n", xfeatures_mask);
dc1e35c6
SS
651 BUG();
652 }
653
654 /*
a30469e7 655 * Support only the state known to OS.
dc1e35c6 656 */
614df7fb 657 xfeatures_mask = xfeatures_mask & XCNTXT_MASK;
97e80a70 658
1cff92d8 659 xstate_enable();
dc1e35c6
SS
660
661 /*
662 * Recompute the context size for enabled features
663 */
7e7ce87f 664 init_xstate_size();
dc1e35c6 665
614df7fb 666 update_regset_xstate_info(xstate_size, xfeatures_mask);
c37b5efe 667 prepare_fx_sw_frame();
5d2bd700 668 setup_init_fpu_buf();
dc1e35c6 669
e0022981
SS
670 /* Auto enable eagerfpu for xsaveopt */
671 if (cpu_has_xsaveopt && eagerfpu != DISABLE)
672 eagerfpu = ENABLE;
212b0212 673
614df7fb 674 if (xfeatures_mask & XSTATE_EAGER) {
e7d820a5 675 if (eagerfpu == DISABLE) {
32d4d9cc 676 pr_err("x86/fpu: eagerfpu switching disabled, disabling the following xstate features: 0x%llx.\n",
614df7fb
IM
677 xfeatures_mask & XSTATE_EAGER);
678 xfeatures_mask &= ~XSTATE_EAGER;
e7d820a5
QR
679 } else {
680 eagerfpu = ENABLE;
681 }
682 }
683
32d4d9cc 684 pr_info("x86/fpu: Enabled xstate features 0x%llx, context size is 0x%x bytes, using '%s' format.\n",
614df7fb 685 xfeatures_mask,
32d4d9cc
IM
686 xstate_size,
687 cpu_has_xsaves ? "compacted" : "standard");
dc1e35c6 688}
82d4150c 689
1cff92d8
PA
690/*
691 * For the very first instance, this calls xstate_enable_boot_cpu();
692 * for all subsequent instances, this calls xstate_enable().
1cff92d8 693 */
148f9bb8 694void xsave_init(void)
82d4150c 695{
c0841e34 696 static char on_boot_cpu = 1;
1cff92d8 697
32d4d9cc
IM
698 if (!cpu_has_xsave) {
699 if (on_boot_cpu) {
700 on_boot_cpu = 0;
701 pr_info("x86/fpu: Legacy x87 FPU detected.\n");
702 }
0e49bf66 703 return;
32d4d9cc 704 }
0e49bf66 705
c0841e34
IM
706 if (on_boot_cpu) {
707 on_boot_cpu = 0;
708 xstate_enable_boot_cpu();
709 } else {
710 xstate_enable();
711 }
82d4150c 712}
5d2bd700 713
7fc253e2
ON
714/*
715 * setup_init_fpu_buf() is __init and it is OK to call it here because
716 * init_xstate_buf will be unset only once during boot.
717 */
718void __init_refok eager_fpu_init(void)
5d2bd700 719{
c5bedc68 720 WARN_ON(current->thread.fpu.fpstate_active);
5d2bd700 721 current_thread_info()->status = 0;
e0022981
SS
722
723 if (eagerfpu == ENABLE)
724 setup_force_cpu_cap(X86_FEATURE_EAGER_FPU);
725
9a89b029
IM
726 printk_once(KERN_INFO "x86/fpu: Using '%s' FPU context switches.\n", eagerfpu == ENABLE ? "eager" : "lazy");
727
5d2bd700
SS
728 if (!cpu_has_eager_fpu) {
729 stts();
730 return;
731 }
732
7fc253e2
ON
733 if (!init_xstate_buf)
734 setup_init_fpu_buf();
5d2bd700 735}
7496d645 736
9254aaa0
IM
737/*
738 * Restore minimal FPU state after suspend:
739 */
740void fpu__resume_cpu(void)
741{
742 /*
743 * Restore XCR0 on xsave capable CPUs:
744 */
745 if (cpu_has_xsave)
746 xsetbv(XCR_XFEATURE_ENABLED_MASK, xfeatures_mask);
747}
748
7496d645
FY
749/*
750 * Given the xsave area and a state inside, this function returns the
751 * address of the state.
752 *
753 * This is the API that is called to get xstate address in either
754 * standard format or compacted format of xsave area.
755 *
756 * Inputs:
757 * xsave: base address of the xsave area;
758 * xstate: state which is defined in xsave.h (e.g. XSTATE_FP, XSTATE_SSE,
759 * etc.)
760 * Output:
761 * address of the state in the xsave area.
762 */
763void *get_xsave_addr(struct xsave_struct *xsave, int xstate)
764{
765 int feature = fls64(xstate) - 1;
614df7fb 766 if (!test_bit(feature, (unsigned long *)&xfeatures_mask))
7496d645
FY
767 return NULL;
768
769 return (void *)xsave + xstate_comp_offsets[feature];
770}
ba7b3920 771EXPORT_SYMBOL_GPL(get_xsave_addr);
This page took 0.443249 seconds and 5 git commands to generate.