Commit | Line | Data |
---|---|---|
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> | |
8 | #include <asm/i387.h> | |
c37b5efe SS |
9 | #ifdef CONFIG_IA32_EMULATION |
10 | #include <asm/sigcontext32.h> | |
11 | #endif | |
6152e4b1 | 12 | #include <asm/xcr.h> |
dc1e35c6 SS |
13 | |
14 | /* | |
15 | * Supported feature mask by the CPU and the kernel. | |
16 | */ | |
6152e4b1 | 17 | u64 pcntxt_mask; |
dc1e35c6 | 18 | |
45c2d7f4 RR |
19 | /* |
20 | * Represents init state for the supported extended state. | |
21 | */ | |
22 | static struct xsave_struct *init_xstate_buf; | |
23 | ||
c37b5efe SS |
24 | struct _fpx_sw_bytes fx_sw_reserved; |
25 | #ifdef CONFIG_IA32_EMULATION | |
26 | struct _fpx_sw_bytes fx_sw_reserved_ia32; | |
27 | #endif | |
28 | ||
a1488f8b SS |
29 | static unsigned int *xstate_offsets, *xstate_sizes, xstate_features; |
30 | ||
29104e10 SS |
31 | /* |
32 | * If a processor implementation discern that a processor state component is | |
33 | * in its initialized state it may modify the corresponding bit in the | |
34 | * xsave_hdr.xstate_bv as '0', with out modifying the corresponding memory | |
35 | * layout in the case of xsaveopt. While presenting the xstate information to | |
36 | * the user, we always ensure that the memory layout of a feature will be in | |
37 | * the init state if the corresponding header bit is zero. This is to ensure | |
38 | * that the user doesn't see some stale state in the memory layout during | |
39 | * signal handling, debugging etc. | |
40 | */ | |
41 | void __sanitize_i387_state(struct task_struct *tsk) | |
42 | { | |
43 | u64 xstate_bv; | |
44 | int feature_bit = 0x2; | |
45 | struct i387_fxsave_struct *fx = &tsk->thread.fpu.state->fxsave; | |
46 | ||
47 | if (!fx) | |
48 | return; | |
49 | ||
50 | BUG_ON(task_thread_info(tsk)->status & TS_USEDFPU); | |
51 | ||
52 | xstate_bv = tsk->thread.fpu.state->xsave.xsave_hdr.xstate_bv; | |
53 | ||
54 | /* | |
55 | * None of the feature bits are in init state. So nothing else | |
56 | * to do for us, as the memory layout is upto date. | |
57 | */ | |
58 | if ((xstate_bv & pcntxt_mask) == pcntxt_mask) | |
59 | return; | |
60 | ||
61 | /* | |
62 | * FP is in init state | |
63 | */ | |
64 | if (!(xstate_bv & XSTATE_FP)) { | |
65 | fx->cwd = 0x37f; | |
66 | fx->swd = 0; | |
67 | fx->twd = 0; | |
68 | fx->fop = 0; | |
69 | fx->rip = 0; | |
70 | fx->rdp = 0; | |
71 | memset(&fx->st_space[0], 0, 128); | |
72 | } | |
73 | ||
74 | /* | |
75 | * SSE is in init state | |
76 | */ | |
77 | if (!(xstate_bv & XSTATE_SSE)) | |
78 | memset(&fx->xmm_space[0], 0, 256); | |
79 | ||
80 | xstate_bv = (pcntxt_mask & ~xstate_bv) >> 2; | |
81 | ||
82 | /* | |
83 | * Update all the other memory layouts for which the corresponding | |
84 | * header bit is in the init state. | |
85 | */ | |
86 | while (xstate_bv) { | |
87 | if (xstate_bv & 0x1) { | |
88 | int offset = xstate_offsets[feature_bit]; | |
89 | int size = xstate_sizes[feature_bit]; | |
90 | ||
91 | memcpy(((void *) fx) + offset, | |
92 | ((void *) init_xstate_buf) + offset, | |
93 | size); | |
94 | } | |
95 | ||
96 | xstate_bv >>= 1; | |
97 | feature_bit++; | |
98 | } | |
99 | } | |
100 | ||
c37b5efe SS |
101 | /* |
102 | * Check for the presence of extended state information in the | |
103 | * user fpstate pointer in the sigcontext. | |
104 | */ | |
105 | int check_for_xstate(struct i387_fxsave_struct __user *buf, | |
106 | void __user *fpstate, | |
107 | struct _fpx_sw_bytes *fx_sw_user) | |
108 | { | |
109 | int min_xstate_size = sizeof(struct i387_fxsave_struct) + | |
110 | sizeof(struct xsave_hdr_struct); | |
111 | unsigned int magic2; | |
112 | int err; | |
113 | ||
114 | err = __copy_from_user(fx_sw_user, &buf->sw_reserved[0], | |
115 | sizeof(struct _fpx_sw_bytes)); | |
c37b5efe | 116 | if (err) |
d6d4d420 | 117 | return -EFAULT; |
c37b5efe SS |
118 | |
119 | /* | |
120 | * First Magic check failed. | |
121 | */ | |
122 | if (fx_sw_user->magic1 != FP_XSTATE_MAGIC1) | |
d6d4d420 | 123 | return -EINVAL; |
c37b5efe SS |
124 | |
125 | /* | |
126 | * Check for error scenarios. | |
127 | */ | |
128 | if (fx_sw_user->xstate_size < min_xstate_size || | |
129 | fx_sw_user->xstate_size > xstate_size || | |
130 | fx_sw_user->xstate_size > fx_sw_user->extended_size) | |
d6d4d420 | 131 | return -EINVAL; |
c37b5efe SS |
132 | |
133 | err = __get_user(magic2, (__u32 *) (((void *)fpstate) + | |
134 | fx_sw_user->extended_size - | |
135 | FP_XSTATE_MAGIC2_SIZE)); | |
d6d4d420 DC |
136 | if (err) |
137 | return err; | |
c37b5efe SS |
138 | /* |
139 | * Check for the presence of second magic word at the end of memory | |
140 | * layout. This detects the case where the user just copied the legacy | |
141 | * fpstate layout with out copying the extended state information | |
142 | * in the memory layout. | |
143 | */ | |
d6d4d420 DC |
144 | if (magic2 != FP_XSTATE_MAGIC2) |
145 | return -EFAULT; | |
c37b5efe SS |
146 | |
147 | return 0; | |
148 | } | |
149 | ||
ab513701 SS |
150 | #ifdef CONFIG_X86_64 |
151 | /* | |
152 | * Signal frame handlers. | |
153 | */ | |
154 | ||
155 | int save_i387_xstate(void __user *buf) | |
156 | { | |
157 | struct task_struct *tsk = current; | |
158 | int err = 0; | |
159 | ||
160 | if (!access_ok(VERIFY_WRITE, buf, sig_xstate_size)) | |
161 | return -EACCES; | |
162 | ||
f65bc214 | 163 | BUG_ON(sig_xstate_size < xstate_size); |
ab513701 | 164 | |
c37b5efe | 165 | if ((unsigned long)buf % 64) |
ab513701 SS |
166 | printk("save_i387_xstate: bad fpstate %p\n", buf); |
167 | ||
168 | if (!used_math()) | |
169 | return 0; | |
06c38d5e | 170 | |
ab513701 | 171 | if (task_thread_info(tsk)->status & TS_USEDFPU) { |
c9ad4882 | 172 | if (use_xsave()) |
c37b5efe SS |
173 | err = xsave_user(buf); |
174 | else | |
175 | err = fxsave_user(buf); | |
176 | ||
ab513701 SS |
177 | if (err) |
178 | return err; | |
179 | task_thread_info(tsk)->status &= ~TS_USEDFPU; | |
180 | stts(); | |
181 | } else { | |
29104e10 | 182 | sanitize_i387_state(tsk); |
86603283 | 183 | if (__copy_to_user(buf, &tsk->thread.fpu.state->fxsave, |
ab513701 SS |
184 | xstate_size)) |
185 | return -1; | |
186 | } | |
c37b5efe | 187 | |
06c38d5e SS |
188 | clear_used_math(); /* trigger finit */ |
189 | ||
c9ad4882 | 190 | if (use_xsave()) { |
c37b5efe | 191 | struct _fpstate __user *fx = buf; |
04944b79 SS |
192 | struct _xstate __user *x = buf; |
193 | u64 xstate_bv; | |
c37b5efe SS |
194 | |
195 | err = __copy_to_user(&fx->sw_reserved, &fx_sw_reserved, | |
196 | sizeof(struct _fpx_sw_bytes)); | |
197 | ||
198 | err |= __put_user(FP_XSTATE_MAGIC2, | |
199 | (__u32 __user *) (buf + sig_xstate_size | |
200 | - FP_XSTATE_MAGIC2_SIZE)); | |
04944b79 SS |
201 | |
202 | /* | |
203 | * Read the xstate_bv which we copied (directly from the cpu or | |
204 | * from the state in task struct) to the user buffers and | |
205 | * set the FP/SSE bits. | |
206 | */ | |
207 | err |= __get_user(xstate_bv, &x->xstate_hdr.xstate_bv); | |
208 | ||
209 | /* | |
210 | * For legacy compatible, we always set FP/SSE bits in the bit | |
211 | * vector while saving the state to the user context. This will | |
212 | * enable us capturing any changes(during sigreturn) to | |
213 | * the FP/SSE bits by the legacy applications which don't touch | |
214 | * xstate_bv in the xsave header. | |
215 | * | |
216 | * xsave aware apps can change the xstate_bv in the xsave | |
217 | * header as well as change any contents in the memory layout. | |
218 | * xrestore as part of sigreturn will capture all the changes. | |
219 | */ | |
220 | xstate_bv |= XSTATE_FPSSE; | |
221 | ||
222 | err |= __put_user(xstate_bv, &x->xstate_hdr.xstate_bv); | |
223 | ||
f364eada SS |
224 | if (err) |
225 | return err; | |
c37b5efe SS |
226 | } |
227 | ||
ab513701 SS |
228 | return 1; |
229 | } | |
230 | ||
c37b5efe SS |
231 | /* |
232 | * Restore the extended state if present. Otherwise, restore the FP/SSE | |
233 | * state. | |
234 | */ | |
7820b756 | 235 | static int restore_user_xstate(void __user *buf) |
c37b5efe SS |
236 | { |
237 | struct _fpx_sw_bytes fx_sw_user; | |
6152e4b1 | 238 | u64 mask; |
c37b5efe SS |
239 | int err; |
240 | ||
241 | if (((unsigned long)buf % 64) || | |
242 | check_for_xstate(buf, buf, &fx_sw_user)) | |
243 | goto fx_only; | |
244 | ||
6152e4b1 | 245 | mask = fx_sw_user.xstate_bv; |
c37b5efe SS |
246 | |
247 | /* | |
248 | * restore the state passed by the user. | |
249 | */ | |
6152e4b1 | 250 | err = xrestore_user(buf, mask); |
c37b5efe SS |
251 | if (err) |
252 | return err; | |
253 | ||
254 | /* | |
255 | * init the state skipped by the user. | |
256 | */ | |
6152e4b1 | 257 | mask = pcntxt_mask & ~mask; |
8e221b6d SS |
258 | if (unlikely(mask)) |
259 | xrstor_state(init_xstate_buf, mask); | |
c37b5efe SS |
260 | |
261 | return 0; | |
262 | ||
263 | fx_only: | |
264 | /* | |
265 | * couldn't find the extended state information in the | |
266 | * memory layout. Restore just the FP/SSE and init all | |
267 | * the other extended state. | |
268 | */ | |
6152e4b1 | 269 | xrstor_state(init_xstate_buf, pcntxt_mask & ~XSTATE_FPSSE); |
c37b5efe SS |
270 | return fxrstor_checking((__force struct i387_fxsave_struct *)buf); |
271 | } | |
272 | ||
ab513701 SS |
273 | /* |
274 | * This restores directly out of user space. Exceptions are handled. | |
275 | */ | |
276 | int restore_i387_xstate(void __user *buf) | |
277 | { | |
278 | struct task_struct *tsk = current; | |
c37b5efe | 279 | int err = 0; |
ab513701 SS |
280 | |
281 | if (!buf) { | |
c37b5efe SS |
282 | if (used_math()) |
283 | goto clear; | |
ab513701 SS |
284 | return 0; |
285 | } else | |
286 | if (!access_ok(VERIFY_READ, buf, sig_xstate_size)) | |
287 | return -EACCES; | |
288 | ||
289 | if (!used_math()) { | |
290 | err = init_fpu(tsk); | |
291 | if (err) | |
292 | return err; | |
293 | } | |
294 | ||
295 | if (!(task_thread_info(current)->status & TS_USEDFPU)) { | |
296 | clts(); | |
297 | task_thread_info(current)->status |= TS_USEDFPU; | |
298 | } | |
c9ad4882 | 299 | if (use_xsave()) |
c37b5efe SS |
300 | err = restore_user_xstate(buf); |
301 | else | |
302 | err = fxrstor_checking((__force struct i387_fxsave_struct *) | |
303 | buf); | |
ab513701 SS |
304 | if (unlikely(err)) { |
305 | /* | |
306 | * Encountered an error while doing the restore from the | |
307 | * user buffer, clear the fpu state. | |
308 | */ | |
c37b5efe | 309 | clear: |
ab513701 SS |
310 | clear_fpu(tsk); |
311 | clear_used_math(); | |
312 | } | |
313 | return err; | |
314 | } | |
315 | #endif | |
316 | ||
c37b5efe SS |
317 | /* |
318 | * Prepare the SW reserved portion of the fxsave memory layout, indicating | |
319 | * the presence of the extended state information in the memory layout | |
320 | * pointed by the fpstate pointer in the sigcontext. | |
321 | * This will be saved when ever the FP and extended state context is | |
322 | * saved on the user stack during the signal handler delivery to the user. | |
323 | */ | |
8bcad30f | 324 | static void prepare_fx_sw_frame(void) |
c37b5efe SS |
325 | { |
326 | int size_extended = (xstate_size - sizeof(struct i387_fxsave_struct)) + | |
327 | FP_XSTATE_MAGIC2_SIZE; | |
328 | ||
329 | sig_xstate_size = sizeof(struct _fpstate) + size_extended; | |
330 | ||
331 | #ifdef CONFIG_IA32_EMULATION | |
332 | sig_xstate_ia32_size = sizeof(struct _fpstate_ia32) + size_extended; | |
333 | #endif | |
334 | ||
335 | memset(&fx_sw_reserved, 0, sizeof(fx_sw_reserved)); | |
336 | ||
337 | fx_sw_reserved.magic1 = FP_XSTATE_MAGIC1; | |
338 | fx_sw_reserved.extended_size = sig_xstate_size; | |
6152e4b1 | 339 | fx_sw_reserved.xstate_bv = pcntxt_mask; |
c37b5efe SS |
340 | fx_sw_reserved.xstate_size = xstate_size; |
341 | #ifdef CONFIG_IA32_EMULATION | |
342 | memcpy(&fx_sw_reserved_ia32, &fx_sw_reserved, | |
343 | sizeof(struct _fpx_sw_bytes)); | |
344 | fx_sw_reserved_ia32.extended_size = sig_xstate_ia32_size; | |
345 | #endif | |
346 | } | |
347 | ||
3c1c7f10 SS |
348 | #ifdef CONFIG_X86_64 |
349 | unsigned int sig_xstate_size = sizeof(struct _fpstate); | |
350 | #endif | |
351 | ||
dc1e35c6 SS |
352 | /* |
353 | * Enable the extended processor state save/restore feature | |
354 | */ | |
1cff92d8 | 355 | static inline void xstate_enable(void) |
dc1e35c6 | 356 | { |
dc1e35c6 | 357 | set_in_cr4(X86_CR4_OSXSAVE); |
6152e4b1 | 358 | xsetbv(XCR_XFEATURE_ENABLED_MASK, pcntxt_mask); |
dc1e35c6 SS |
359 | } |
360 | ||
a1488f8b SS |
361 | /* |
362 | * Record the offsets and sizes of different state managed by the xsave | |
363 | * memory layout. | |
364 | */ | |
4995b9db | 365 | static void __init setup_xstate_features(void) |
a1488f8b SS |
366 | { |
367 | int eax, ebx, ecx, edx, leaf = 0x2; | |
368 | ||
369 | xstate_features = fls64(pcntxt_mask); | |
370 | xstate_offsets = alloc_bootmem(xstate_features * sizeof(int)); | |
371 | xstate_sizes = alloc_bootmem(xstate_features * sizeof(int)); | |
372 | ||
373 | do { | |
ee813d53 | 374 | cpuid_count(XSTATE_CPUID, leaf, &eax, &ebx, &ecx, &edx); |
a1488f8b SS |
375 | |
376 | if (eax == 0) | |
377 | break; | |
378 | ||
379 | xstate_offsets[leaf] = ebx; | |
380 | xstate_sizes[leaf] = eax; | |
381 | ||
382 | leaf++; | |
383 | } while (1); | |
384 | } | |
385 | ||
dc1e35c6 SS |
386 | /* |
387 | * setup the xstate image representing the init state | |
388 | */ | |
a19aac85 | 389 | static void __init setup_xstate_init(void) |
dc1e35c6 | 390 | { |
29104e10 SS |
391 | setup_xstate_features(); |
392 | ||
393 | /* | |
394 | * Setup init_xstate_buf to represent the init state of | |
395 | * all the features managed by the xsave | |
396 | */ | |
dc1e35c6 SS |
397 | init_xstate_buf = alloc_bootmem(xstate_size); |
398 | init_xstate_buf->i387.mxcsr = MXCSR_DEFAULT; | |
a1488f8b | 399 | |
29104e10 SS |
400 | clts(); |
401 | /* | |
402 | * Init all the features state with header_bv being 0x0 | |
403 | */ | |
404 | xrstor_state(init_xstate_buf, -1); | |
405 | /* | |
406 | * Dump the init state again. This is to identify the init state | |
407 | * of any feature which is not represented by all zero's. | |
408 | */ | |
409 | xsave_state(init_xstate_buf, -1); | |
410 | stts(); | |
dc1e35c6 SS |
411 | } |
412 | ||
413 | /* | |
414 | * Enable and initialize the xsave feature. | |
415 | */ | |
1cff92d8 | 416 | static void __init xstate_enable_boot_cpu(void) |
dc1e35c6 SS |
417 | { |
418 | unsigned int eax, ebx, ecx, edx; | |
419 | ||
ee813d53 RR |
420 | if (boot_cpu_data.cpuid_level < XSTATE_CPUID) { |
421 | WARN(1, KERN_ERR "XSTATE_CPUID missing\n"); | |
422 | return; | |
423 | } | |
424 | ||
425 | cpuid_count(XSTATE_CPUID, 0, &eax, &ebx, &ecx, &edx); | |
6152e4b1 | 426 | pcntxt_mask = eax + ((u64)edx << 32); |
dc1e35c6 | 427 | |
6152e4b1 PA |
428 | if ((pcntxt_mask & XSTATE_FPSSE) != XSTATE_FPSSE) { |
429 | printk(KERN_ERR "FP/SSE not shown under xsave features 0x%llx\n", | |
430 | pcntxt_mask); | |
dc1e35c6 SS |
431 | BUG(); |
432 | } | |
433 | ||
434 | /* | |
a30469e7 | 435 | * Support only the state known to OS. |
dc1e35c6 | 436 | */ |
6152e4b1 | 437 | pcntxt_mask = pcntxt_mask & XCNTXT_MASK; |
97e80a70 | 438 | |
1cff92d8 | 439 | xstate_enable(); |
dc1e35c6 SS |
440 | |
441 | /* | |
442 | * Recompute the context size for enabled features | |
443 | */ | |
ee813d53 | 444 | cpuid_count(XSTATE_CPUID, 0, &eax, &ebx, &ecx, &edx); |
dc1e35c6 SS |
445 | xstate_size = ebx; |
446 | ||
5b3efd50 | 447 | update_regset_xstate_info(xstate_size, pcntxt_mask); |
c37b5efe SS |
448 | prepare_fx_sw_frame(); |
449 | ||
dc1e35c6 SS |
450 | setup_xstate_init(); |
451 | ||
6152e4b1 | 452 | printk(KERN_INFO "xsave/xrstor: enabled xstate_bv 0x%llx, " |
dc1e35c6 | 453 | "cntxt size 0x%x\n", |
6152e4b1 | 454 | pcntxt_mask, xstate_size); |
dc1e35c6 | 455 | } |
82d4150c | 456 | |
1cff92d8 PA |
457 | /* |
458 | * For the very first instance, this calls xstate_enable_boot_cpu(); | |
459 | * for all subsequent instances, this calls xstate_enable(). | |
460 | * | |
461 | * This is somewhat obfuscated due to the lack of powerful enough | |
462 | * overrides for the section checks. | |
463 | */ | |
82d4150c RR |
464 | void __cpuinit xsave_init(void) |
465 | { | |
1cff92d8 PA |
466 | static __refdata void (*next_func)(void) = xstate_enable_boot_cpu; |
467 | void (*this_func)(void); | |
468 | ||
0e49bf66 RR |
469 | if (!cpu_has_xsave) |
470 | return; | |
471 | ||
1cff92d8 PA |
472 | this_func = next_func; |
473 | next_func = xstate_enable; | |
474 | this_func(); | |
82d4150c | 475 | } |