1 /* 2 * xsave/xrstor support. 3 * 4 * Author: Suresh Siddha <suresh.b.siddha@intel.com> 5 */ 6 #include <linux/compat.h> 7 #include <linux/cpu.h> 8 9 #include <asm/fpu/api.h> 10 #include <asm/fpu/internal.h> 11 #include <asm/fpu/signal.h> 12 #include <asm/fpu/regset.h> 13 14 #include <asm/tlbflush.h> 15 16 static const char *xfeature_names[] = 17 { 18 "x87 floating point registers" , 19 "SSE registers" , 20 "AVX registers" , 21 "MPX bounds registers" , 22 "MPX CSR" , 23 "AVX-512 opmask" , 24 "AVX-512 Hi256" , 25 "AVX-512 ZMM_Hi256" , 26 "unknown xstate feature" , 27 }; 28 29 /* 30 * Mask of xstate features supported by the CPU and the kernel: 31 */ 32 u64 xfeatures_mask __read_mostly; 33 34 static unsigned int xstate_offsets[XFEATURE_MAX] = { [ 0 ... XFEATURE_MAX - 1] = -1}; 35 static unsigned int xstate_sizes[XFEATURE_MAX] = { [ 0 ... XFEATURE_MAX - 1] = -1}; 36 static unsigned int xstate_comp_offsets[sizeof(xfeatures_mask)*8]; 37 38 /* 39 * Clear all of the X86_FEATURE_* bits that are unavailable 40 * when the CPU has no XSAVE support. 41 */ 42 void fpu__xstate_clear_all_cpu_caps(void) 43 { 44 setup_clear_cpu_cap(X86_FEATURE_XSAVE); 45 setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT); 46 setup_clear_cpu_cap(X86_FEATURE_XSAVEC); 47 setup_clear_cpu_cap(X86_FEATURE_XSAVES); 48 setup_clear_cpu_cap(X86_FEATURE_AVX); 49 setup_clear_cpu_cap(X86_FEATURE_AVX2); 50 setup_clear_cpu_cap(X86_FEATURE_AVX512F); 51 setup_clear_cpu_cap(X86_FEATURE_AVX512PF); 52 setup_clear_cpu_cap(X86_FEATURE_AVX512ER); 53 setup_clear_cpu_cap(X86_FEATURE_AVX512CD); 54 setup_clear_cpu_cap(X86_FEATURE_MPX); 55 setup_clear_cpu_cap(X86_FEATURE_XGETBV1); 56 } 57 58 /* 59 * Return whether the system supports a given xfeature. 60 * 61 * Also return the name of the (most advanced) feature that the caller requested: 62 */ 63 int cpu_has_xfeatures(u64 xfeatures_needed, const char **feature_name) 64 { 65 u64 xfeatures_missing = xfeatures_needed & ~xfeatures_mask; 66 67 if (unlikely(feature_name)) { 68 long xfeature_idx, max_idx; 69 u64 xfeatures_print; 70 /* 71 * So we use FLS here to be able to print the most advanced 72 * feature that was requested but is missing. So if a driver 73 * asks about "XFEATURE_MASK_SSE | XFEATURE_MASK_YMM" we'll print the 74 * missing AVX feature - this is the most informative message 75 * to users: 76 */ 77 if (xfeatures_missing) 78 xfeatures_print = xfeatures_missing; 79 else 80 xfeatures_print = xfeatures_needed; 81 82 xfeature_idx = fls64(xfeatures_print)-1; 83 max_idx = ARRAY_SIZE(xfeature_names)-1; 84 xfeature_idx = min(xfeature_idx, max_idx); 85 86 *feature_name = xfeature_names[xfeature_idx]; 87 } 88 89 if (xfeatures_missing) 90 return 0; 91 92 return 1; 93 } 94 EXPORT_SYMBOL_GPL(cpu_has_xfeatures); 95 96 /* 97 * When executing XSAVEOPT (or other optimized XSAVE instructions), if 98 * a processor implementation detects that an FPU state component is still 99 * (or is again) in its initialized state, it may clear the corresponding 100 * bit in the header.xfeatures field, and can skip the writeout of registers 101 * to the corresponding memory layout. 102 * 103 * This means that when the bit is zero, the state component might still contain 104 * some previous - non-initialized register state. 105 * 106 * Before writing xstate information to user-space we sanitize those components, 107 * to always ensure that the memory layout of a feature will be in the init state 108 * if the corresponding header bit is zero. This is to ensure that user-space doesn't 109 * see some stale state in the memory layout during signal handling, debugging etc. 110 */ 111 void fpstate_sanitize_xstate(struct fpu *fpu) 112 { 113 struct fxregs_state *fx = &fpu->state.fxsave; 114 int feature_bit; 115 u64 xfeatures; 116 117 if (!use_xsaveopt()) 118 return; 119 120 xfeatures = fpu->state.xsave.header.xfeatures; 121 122 /* 123 * None of the feature bits are in init state. So nothing else 124 * to do for us, as the memory layout is up to date. 125 */ 126 if ((xfeatures & xfeatures_mask) == xfeatures_mask) 127 return; 128 129 /* 130 * FP is in init state 131 */ 132 if (!(xfeatures & XFEATURE_MASK_FP)) { 133 fx->cwd = 0x37f; 134 fx->swd = 0; 135 fx->twd = 0; 136 fx->fop = 0; 137 fx->rip = 0; 138 fx->rdp = 0; 139 memset(&fx->st_space[0], 0, 128); 140 } 141 142 /* 143 * SSE is in init state 144 */ 145 if (!(xfeatures & XFEATURE_MASK_SSE)) 146 memset(&fx->xmm_space[0], 0, 256); 147 148 /* 149 * First two features are FPU and SSE, which above we handled 150 * in a special way already: 151 */ 152 feature_bit = 0x2; 153 xfeatures = (xfeatures_mask & ~xfeatures) >> 2; 154 155 /* 156 * Update all the remaining memory layouts according to their 157 * standard xstate layout, if their header bit is in the init 158 * state: 159 */ 160 while (xfeatures) { 161 if (xfeatures & 0x1) { 162 int offset = xstate_offsets[feature_bit]; 163 int size = xstate_sizes[feature_bit]; 164 165 memcpy((void *)fx + offset, 166 (void *)&init_fpstate.xsave + offset, 167 size); 168 } 169 170 xfeatures >>= 1; 171 feature_bit++; 172 } 173 } 174 175 /* 176 * Enable the extended processor state save/restore feature. 177 * Called once per CPU onlining. 178 */ 179 void fpu__init_cpu_xstate(void) 180 { 181 if (!cpu_has_xsave || !xfeatures_mask) 182 return; 183 184 cr4_set_bits(X86_CR4_OSXSAVE); 185 xsetbv(XCR_XFEATURE_ENABLED_MASK, xfeatures_mask); 186 } 187 188 /* 189 * Note that in the future we will likely need a pair of 190 * functions here: one for user xstates and the other for 191 * system xstates. For now, they are the same. 192 */ 193 static int xfeature_enabled(enum xfeature xfeature) 194 { 195 return !!(xfeatures_mask & (1UL << xfeature)); 196 } 197 198 /* 199 * Record the offsets and sizes of various xstates contained 200 * in the XSAVE state memory layout. 201 */ 202 static void __init setup_xstate_features(void) 203 { 204 u32 eax, ebx, ecx, edx, i; 205 /* start at the beginnning of the "extended state" */ 206 unsigned int last_good_offset = offsetof(struct xregs_state, 207 extended_state_area); 208 209 for (i = FIRST_EXTENDED_XFEATURE; i < XFEATURE_MAX; i++) { 210 if (!xfeature_enabled(i)) 211 continue; 212 213 cpuid_count(XSTATE_CPUID, i, &eax, &ebx, &ecx, &edx); 214 xstate_offsets[i] = ebx; 215 xstate_sizes[i] = eax; 216 /* 217 * In our xstate size checks, we assume that the 218 * highest-numbered xstate feature has the 219 * highest offset in the buffer. Ensure it does. 220 */ 221 WARN_ONCE(last_good_offset > xstate_offsets[i], 222 "x86/fpu: misordered xstate at %d\n", last_good_offset); 223 last_good_offset = xstate_offsets[i]; 224 225 printk(KERN_INFO "x86/fpu: xstate_offset[%d]: %4d, xstate_sizes[%d]: %4d\n", i, ebx, i, eax); 226 } 227 } 228 229 static void __init print_xstate_feature(u64 xstate_mask) 230 { 231 const char *feature_name; 232 233 if (cpu_has_xfeatures(xstate_mask, &feature_name)) 234 pr_info("x86/fpu: Supporting XSAVE feature 0x%02Lx: '%s'\n", xstate_mask, feature_name); 235 } 236 237 /* 238 * Print out all the supported xstate features: 239 */ 240 static void __init print_xstate_features(void) 241 { 242 print_xstate_feature(XFEATURE_MASK_FP); 243 print_xstate_feature(XFEATURE_MASK_SSE); 244 print_xstate_feature(XFEATURE_MASK_YMM); 245 print_xstate_feature(XFEATURE_MASK_BNDREGS); 246 print_xstate_feature(XFEATURE_MASK_BNDCSR); 247 print_xstate_feature(XFEATURE_MASK_OPMASK); 248 print_xstate_feature(XFEATURE_MASK_ZMM_Hi256); 249 print_xstate_feature(XFEATURE_MASK_Hi16_ZMM); 250 } 251 252 /* 253 * This function sets up offsets and sizes of all extended states in 254 * xsave area. This supports both standard format and compacted format 255 * of the xsave aread. 256 */ 257 static void __init setup_xstate_comp(void) 258 { 259 unsigned int xstate_comp_sizes[sizeof(xfeatures_mask)*8]; 260 int i; 261 262 /* 263 * The FP xstates and SSE xstates are legacy states. They are always 264 * in the fixed offsets in the xsave area in either compacted form 265 * or standard form. 266 */ 267 xstate_comp_offsets[0] = 0; 268 xstate_comp_offsets[1] = offsetof(struct fxregs_state, xmm_space); 269 270 if (!cpu_has_xsaves) { 271 for (i = FIRST_EXTENDED_XFEATURE; i < XFEATURE_MAX; i++) { 272 if (xfeature_enabled(i)) { 273 xstate_comp_offsets[i] = xstate_offsets[i]; 274 xstate_comp_sizes[i] = xstate_sizes[i]; 275 } 276 } 277 return; 278 } 279 280 xstate_comp_offsets[FIRST_EXTENDED_XFEATURE] = 281 FXSAVE_SIZE + XSAVE_HDR_SIZE; 282 283 for (i = FIRST_EXTENDED_XFEATURE; i < XFEATURE_MAX; i++) { 284 if (xfeature_enabled(i)) 285 xstate_comp_sizes[i] = xstate_sizes[i]; 286 else 287 xstate_comp_sizes[i] = 0; 288 289 if (i > FIRST_EXTENDED_XFEATURE) 290 xstate_comp_offsets[i] = xstate_comp_offsets[i-1] 291 + xstate_comp_sizes[i-1]; 292 293 } 294 } 295 296 /* 297 * setup the xstate image representing the init state 298 */ 299 static void __init setup_init_fpu_buf(void) 300 { 301 static int on_boot_cpu __initdata = 1; 302 303 WARN_ON_FPU(!on_boot_cpu); 304 on_boot_cpu = 0; 305 306 if (!cpu_has_xsave) 307 return; 308 309 setup_xstate_features(); 310 print_xstate_features(); 311 312 if (cpu_has_xsaves) { 313 init_fpstate.xsave.header.xcomp_bv = (u64)1 << 63 | xfeatures_mask; 314 init_fpstate.xsave.header.xfeatures = xfeatures_mask; 315 } 316 317 /* 318 * Init all the features state with header_bv being 0x0 319 */ 320 copy_kernel_to_xregs_booting(&init_fpstate.xsave); 321 322 /* 323 * Dump the init state again. This is to identify the init state 324 * of any feature which is not represented by all zero's. 325 */ 326 copy_xregs_to_kernel_booting(&init_fpstate.xsave); 327 } 328 329 static int xfeature_is_supervisor(int xfeature_nr) 330 { 331 /* 332 * We currently do not support supervisor states, but if 333 * we did, we could find out like this. 334 * 335 * SDM says: If state component i is a user state component, 336 * ECX[0] return 0; if state component i is a supervisor 337 * state component, ECX[0] returns 1. 338 u32 eax, ebx, ecx, edx; 339 cpuid_count(XSTATE_CPUID, xfeature_nr, &eax, &ebx, &ecx, &edx; 340 return !!(ecx & 1); 341 */ 342 return 0; 343 } 344 /* 345 static int xfeature_is_user(int xfeature_nr) 346 { 347 return !xfeature_is_supervisor(xfeature_nr); 348 } 349 */ 350 351 /* 352 * This check is important because it is easy to get XSTATE_* 353 * confused with XSTATE_BIT_*. 354 */ 355 #define CHECK_XFEATURE(nr) do { \ 356 WARN_ON(nr < FIRST_EXTENDED_XFEATURE); \ 357 WARN_ON(nr >= XFEATURE_MAX); \ 358 } while (0) 359 360 /* 361 * We could cache this like xstate_size[], but we only use 362 * it here, so it would be a waste of space. 363 */ 364 static int xfeature_is_aligned(int xfeature_nr) 365 { 366 u32 eax, ebx, ecx, edx; 367 368 CHECK_XFEATURE(xfeature_nr); 369 cpuid_count(XSTATE_CPUID, xfeature_nr, &eax, &ebx, &ecx, &edx); 370 /* 371 * The value returned by ECX[1] indicates the alignment 372 * of state component i when the compacted format 373 * of the extended region of an XSAVE area is used 374 */ 375 return !!(ecx & 2); 376 } 377 378 static int xfeature_uncompacted_offset(int xfeature_nr) 379 { 380 u32 eax, ebx, ecx, edx; 381 382 CHECK_XFEATURE(xfeature_nr); 383 cpuid_count(XSTATE_CPUID, xfeature_nr, &eax, &ebx, &ecx, &edx); 384 return ebx; 385 } 386 387 static int xfeature_size(int xfeature_nr) 388 { 389 u32 eax, ebx, ecx, edx; 390 391 CHECK_XFEATURE(xfeature_nr); 392 cpuid_count(XSTATE_CPUID, xfeature_nr, &eax, &ebx, &ecx, &edx); 393 return eax; 394 } 395 396 /* 397 * 'XSAVES' implies two different things: 398 * 1. saving of supervisor/system state 399 * 2. using the compacted format 400 * 401 * Use this function when dealing with the compacted format so 402 * that it is obvious which aspect of 'XSAVES' is being handled 403 * by the calling code. 404 */ 405 static int using_compacted_format(void) 406 { 407 return cpu_has_xsaves; 408 } 409 410 static void __xstate_dump_leaves(void) 411 { 412 int i; 413 u32 eax, ebx, ecx, edx; 414 static int should_dump = 1; 415 416 if (!should_dump) 417 return; 418 should_dump = 0; 419 /* 420 * Dump out a few leaves past the ones that we support 421 * just in case there are some goodies up there 422 */ 423 for (i = 0; i < XFEATURE_MAX + 10; i++) { 424 cpuid_count(XSTATE_CPUID, i, &eax, &ebx, &ecx, &edx); 425 pr_warn("CPUID[%02x, %02x]: eax=%08x ebx=%08x ecx=%08x edx=%08x\n", 426 XSTATE_CPUID, i, eax, ebx, ecx, edx); 427 } 428 } 429 430 #define XSTATE_WARN_ON(x) do { \ 431 if (WARN_ONCE(x, "XSAVE consistency problem, dumping leaves")) { \ 432 __xstate_dump_leaves(); \ 433 } \ 434 } while (0) 435 436 #define XCHECK_SZ(sz, nr, nr_macro, __struct) do { \ 437 if ((nr == nr_macro) && \ 438 WARN_ONCE(sz != sizeof(__struct), \ 439 "%s: struct is %zu bytes, cpu state %d bytes\n", \ 440 __stringify(nr_macro), sizeof(__struct), sz)) { \ 441 __xstate_dump_leaves(); \ 442 } \ 443 } while (0) 444 445 /* 446 * We have a C struct for each 'xstate'. We need to ensure 447 * that our software representation matches what the CPU 448 * tells us about the state's size. 449 */ 450 static void check_xstate_against_struct(int nr) 451 { 452 /* 453 * Ask the CPU for the size of the state. 454 */ 455 int sz = xfeature_size(nr); 456 /* 457 * Match each CPU state with the corresponding software 458 * structure. 459 */ 460 XCHECK_SZ(sz, nr, XFEATURE_YMM, struct ymmh_struct); 461 XCHECK_SZ(sz, nr, XFEATURE_BNDREGS, struct mpx_bndreg_state); 462 XCHECK_SZ(sz, nr, XFEATURE_BNDCSR, struct mpx_bndcsr_state); 463 XCHECK_SZ(sz, nr, XFEATURE_OPMASK, struct avx_512_opmask_state); 464 XCHECK_SZ(sz, nr, XFEATURE_ZMM_Hi256, struct avx_512_zmm_uppers_state); 465 XCHECK_SZ(sz, nr, XFEATURE_Hi16_ZMM, struct avx_512_hi16_state); 466 467 /* 468 * Make *SURE* to add any feature numbers in below if 469 * there are "holes" in the xsave state component 470 * numbers. 471 */ 472 if ((nr < XFEATURE_YMM) || 473 (nr >= XFEATURE_MAX)) { 474 WARN_ONCE(1, "no structure for xstate: %d\n", nr); 475 XSTATE_WARN_ON(1); 476 } 477 } 478 479 /* 480 * This essentially double-checks what the cpu told us about 481 * how large the XSAVE buffer needs to be. We are recalculating 482 * it to be safe. 483 */ 484 static void do_extra_xstate_size_checks(void) 485 { 486 int paranoid_xstate_size = FXSAVE_SIZE + XSAVE_HDR_SIZE; 487 int i; 488 489 for (i = FIRST_EXTENDED_XFEATURE; i < XFEATURE_MAX; i++) { 490 if (!xfeature_enabled(i)) 491 continue; 492 493 check_xstate_against_struct(i); 494 /* 495 * Supervisor state components can be managed only by 496 * XSAVES, which is compacted-format only. 497 */ 498 if (!using_compacted_format()) 499 XSTATE_WARN_ON(xfeature_is_supervisor(i)); 500 501 /* Align from the end of the previous feature */ 502 if (xfeature_is_aligned(i)) 503 paranoid_xstate_size = ALIGN(paranoid_xstate_size, 64); 504 /* 505 * The offset of a given state in the non-compacted 506 * format is given to us in a CPUID leaf. We check 507 * them for being ordered (increasing offsets) in 508 * setup_xstate_features(). 509 */ 510 if (!using_compacted_format()) 511 paranoid_xstate_size = xfeature_uncompacted_offset(i); 512 /* 513 * The compacted-format offset always depends on where 514 * the previous state ended. 515 */ 516 paranoid_xstate_size += xfeature_size(i); 517 } 518 XSTATE_WARN_ON(paranoid_xstate_size != xstate_size); 519 } 520 521 /* 522 * Calculate total size of enabled xstates in XCR0/xfeatures_mask. 523 * 524 * Note the SDM's wording here. "sub-function 0" only enumerates 525 * the size of the *user* states. If we use it to size a buffer 526 * that we use 'XSAVES' on, we could potentially overflow the 527 * buffer because 'XSAVES' saves system states too. 528 * 529 * Note that we do not currently set any bits on IA32_XSS so 530 * 'XCR0 | IA32_XSS == XCR0' for now. 531 */ 532 static unsigned int __init calculate_xstate_size(void) 533 { 534 unsigned int eax, ebx, ecx, edx; 535 unsigned int calculated_xstate_size; 536 537 if (!cpu_has_xsaves) { 538 /* 539 * - CPUID function 0DH, sub-function 0: 540 * EBX enumerates the size (in bytes) required by 541 * the XSAVE instruction for an XSAVE area 542 * containing all the *user* state components 543 * corresponding to bits currently set in XCR0. 544 */ 545 cpuid_count(XSTATE_CPUID, 0, &eax, &ebx, &ecx, &edx); 546 calculated_xstate_size = ebx; 547 } else { 548 /* 549 * - CPUID function 0DH, sub-function 1: 550 * EBX enumerates the size (in bytes) required by 551 * the XSAVES instruction for an XSAVE area 552 * containing all the state components 553 * corresponding to bits currently set in 554 * XCR0 | IA32_XSS. 555 */ 556 cpuid_count(XSTATE_CPUID, 1, &eax, &ebx, &ecx, &edx); 557 calculated_xstate_size = ebx; 558 } 559 return calculated_xstate_size; 560 } 561 562 /* 563 * Will the runtime-enumerated 'xstate_size' fit in the init 564 * task's statically-allocated buffer? 565 */ 566 static bool is_supported_xstate_size(unsigned int test_xstate_size) 567 { 568 if (test_xstate_size <= sizeof(union fpregs_state)) 569 return true; 570 571 pr_warn("x86/fpu: xstate buffer too small (%zu < %d), disabling xsave\n", 572 sizeof(union fpregs_state), test_xstate_size); 573 return false; 574 } 575 576 static int init_xstate_size(void) 577 { 578 /* Recompute the context size for enabled features: */ 579 unsigned int possible_xstate_size = calculate_xstate_size(); 580 581 /* Ensure we have the space to store all enabled: */ 582 if (!is_supported_xstate_size(possible_xstate_size)) 583 return -EINVAL; 584 585 /* 586 * The size is OK, we are definitely going to use xsave, 587 * make it known to the world that we need more space. 588 */ 589 xstate_size = possible_xstate_size; 590 do_extra_xstate_size_checks(); 591 return 0; 592 } 593 594 /* 595 * We enabled the XSAVE hardware, but something went wrong and 596 * we can not use it. Disable it. 597 */ 598 static void fpu__init_disable_system_xstate(void) 599 { 600 xfeatures_mask = 0; 601 cr4_clear_bits(X86_CR4_OSXSAVE); 602 fpu__xstate_clear_all_cpu_caps(); 603 } 604 605 /* 606 * Enable and initialize the xsave feature. 607 * Called once per system bootup. 608 */ 609 void __init fpu__init_system_xstate(void) 610 { 611 unsigned int eax, ebx, ecx, edx; 612 static int on_boot_cpu __initdata = 1; 613 int err; 614 615 WARN_ON_FPU(!on_boot_cpu); 616 on_boot_cpu = 0; 617 618 if (!cpu_has_xsave) { 619 pr_info("x86/fpu: Legacy x87 FPU detected.\n"); 620 return; 621 } 622 623 if (boot_cpu_data.cpuid_level < XSTATE_CPUID) { 624 WARN_ON_FPU(1); 625 return; 626 } 627 628 cpuid_count(XSTATE_CPUID, 0, &eax, &ebx, &ecx, &edx); 629 xfeatures_mask = eax + ((u64)edx << 32); 630 631 if ((xfeatures_mask & XFEATURE_MASK_FPSSE) != XFEATURE_MASK_FPSSE) { 632 pr_err("x86/fpu: FP/SSE not present amongst the CPU's xstate features: 0x%llx.\n", xfeatures_mask); 633 BUG(); 634 } 635 636 xfeatures_mask &= fpu__get_supported_xfeatures_mask(); 637 638 /* Enable xstate instructions to be able to continue with initialization: */ 639 fpu__init_cpu_xstate(); 640 err = init_xstate_size(); 641 if (err) { 642 /* something went wrong, boot without any XSAVE support */ 643 fpu__init_disable_system_xstate(); 644 return; 645 } 646 647 update_regset_xstate_info(xstate_size, xfeatures_mask); 648 fpu__init_prepare_fx_sw_frame(); 649 setup_init_fpu_buf(); 650 setup_xstate_comp(); 651 652 pr_info("x86/fpu: Enabled xstate features 0x%llx, context size is %d bytes, using '%s' format.\n", 653 xfeatures_mask, 654 xstate_size, 655 cpu_has_xsaves ? "compacted" : "standard"); 656 } 657 658 /* 659 * Restore minimal FPU state after suspend: 660 */ 661 void fpu__resume_cpu(void) 662 { 663 /* 664 * Restore XCR0 on xsave capable CPUs: 665 */ 666 if (cpu_has_xsave) 667 xsetbv(XCR_XFEATURE_ENABLED_MASK, xfeatures_mask); 668 } 669 670 /* 671 * Given the xsave area and a state inside, this function returns the 672 * address of the state. 673 * 674 * This is the API that is called to get xstate address in either 675 * standard format or compacted format of xsave area. 676 * 677 * Note that if there is no data for the field in the xsave buffer 678 * this will return NULL. 679 * 680 * Inputs: 681 * xstate: the thread's storage area for all FPU data 682 * xstate_feature: state which is defined in xsave.h (e.g. 683 * XFEATURE_MASK_FP, XFEATURE_MASK_SSE, etc...) 684 * Output: 685 * address of the state in the xsave area, or NULL if the 686 * field is not present in the xsave buffer. 687 */ 688 void *get_xsave_addr(struct xregs_state *xsave, int xstate_feature) 689 { 690 int feature_nr = fls64(xstate_feature) - 1; 691 /* 692 * Do we even *have* xsave state? 693 */ 694 if (!boot_cpu_has(X86_FEATURE_XSAVE)) 695 return NULL; 696 697 /* 698 * We should not ever be requesting features that we 699 * have not enabled. Remember that pcntxt_mask is 700 * what we write to the XCR0 register. 701 */ 702 WARN_ONCE(!(xfeatures_mask & xstate_feature), 703 "get of unsupported state"); 704 /* 705 * This assumes the last 'xsave*' instruction to 706 * have requested that 'xstate_feature' be saved. 707 * If it did not, we might be seeing and old value 708 * of the field in the buffer. 709 * 710 * This can happen because the last 'xsave' did not 711 * request that this feature be saved (unlikely) 712 * or because the "init optimization" caused it 713 * to not be saved. 714 */ 715 if (!(xsave->header.xfeatures & xstate_feature)) 716 return NULL; 717 718 return (void *)xsave + xstate_comp_offsets[feature_nr]; 719 } 720 EXPORT_SYMBOL_GPL(get_xsave_addr); 721 722 /* 723 * This wraps up the common operations that need to occur when retrieving 724 * data from xsave state. It first ensures that the current task was 725 * using the FPU and retrieves the data in to a buffer. It then calculates 726 * the offset of the requested field in the buffer. 727 * 728 * This function is safe to call whether the FPU is in use or not. 729 * 730 * Note that this only works on the current task. 731 * 732 * Inputs: 733 * @xsave_state: state which is defined in xsave.h (e.g. XFEATURE_MASK_FP, 734 * XFEATURE_MASK_SSE, etc...) 735 * Output: 736 * address of the state in the xsave area or NULL if the state 737 * is not present or is in its 'init state'. 738 */ 739 const void *get_xsave_field_ptr(int xsave_state) 740 { 741 struct fpu *fpu = ¤t->thread.fpu; 742 743 if (!fpu->fpstate_active) 744 return NULL; 745 /* 746 * fpu__save() takes the CPU's xstate registers 747 * and saves them off to the 'fpu memory buffer. 748 */ 749 fpu__save(fpu); 750 751 return get_xsave_addr(&fpu->state.xsave, xsave_state); 752 } 753