arch/x86/kernel/xsave.c

   1 /*
   2  * xsave/xrstor support.
   3  *
   4  * Author: Suresh Siddha <suresh.b.siddha@intel.com>
   5  */
   6
   7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   8
   9 #include <linux/bootmem.h>
  10 #include <linux/compat.h>
  11 #include <asm/i387.h>
  12 #include <asm/fpu-internal.h>
  13 #include <asm/sigframe.h>
  14 #include <asm/xcr.h>
  15
  16 /*
  17  * Supported feature mask by the CPU and the kernel.
  18  */
  19 u64 pcntxt_mask;
  20
  21 /*
  22  * Represents init state for the supported extended state.
  23  */
  24 struct xsave_struct *init_xstate_buf;
  25
  26 static struct _fpx_sw_bytes fx_sw_reserved, fx_sw_reserved_ia32;
  27 static unsigned int *xstate_offsets, *xstate_sizes, xstate_features;
  28
  29 /*
  30  * If a processor implementation discern that a processor state component is
  31  * in its initialized state it may modify the corresponding bit in the
  32  * xsave_hdr.xstate_bv as '0', with out modifying the corresponding memory
  33  * layout in the case of xsaveopt. While presenting the xstate information to
  34  * the user, we always ensure that the memory layout of a feature will be in
  35  * the init state if the corresponding header bit is zero. This is to ensure
  36  * that the user doesn't see some stale state in the memory layout during
  37  * signal handling, debugging etc.
  38  */
  39 void __sanitize_i387_state(struct task_struct *tsk)
  40 {
  41         struct i387_fxsave_struct *fx = &tsk->thread.fpu.state->fxsave;
  42         int feature_bit = 0x2;
  43         u64 xstate_bv;
  44
  45         if (!fx)
  46                 return;
  47
  48         xstate_bv = tsk->thread.fpu.state->xsave.xsave_hdr.xstate_bv;
  49
  50         /*
  51          * None of the feature bits are in init state. So nothing else
  52          * to do for us, as the memory layout is up to date.
  53          */
  54         if ((xstate_bv & pcntxt_mask) == pcntxt_mask)
  55                 return;
  56
  57         /*
  58          * FP is in init state
  59          */
  60         if (!(xstate_bv & XSTATE_FP)) {
  61                 fx->cwd = 0x37f;
  62                 fx->swd = 0;
  63                 fx->twd = 0;
  64                 fx->fop = 0;
  65                 fx->rip = 0;
  66                 fx->rdp = 0;
  67                 memset(&fx->st_space[0], 0, 128);
  68         }
  69
  70         /*
  71          * SSE is in init state
  72          */
  73         if (!(xstate_bv & XSTATE_SSE))
  74                 memset(&fx->xmm_space[0], 0, 256);
  75
  76         xstate_bv = (pcntxt_mask & ~xstate_bv) >> 2;
  77
  78         /*
  79          * Update all the other memory layouts for which the corresponding
  80          * header bit is in the init state.
  81          */
  82         while (xstate_bv) {
  83                 if (xstate_bv & 0x1) {
  84                         int offset = xstate_offsets[feature_bit];
  85                         int size = xstate_sizes[feature_bit];
  86
  87                         memcpy(((void *) fx) + offset,
  88                                ((void *) init_xstate_buf) + offset,
  89                                size);
  90                 }
  91
  92                 xstate_bv >>= 1;
  93                 feature_bit++;
  94         }
  95 }
  96
  97 /*
  98  * Check for the presence of extended state information in the
  99  * user fpstate pointer in the sigcontext.
 100  */
 101 static inline int check_for_xstate(struct i387_fxsave_struct __user *buf,
 102                                    void __user *fpstate,
 103                                    struct _fpx_sw_bytes *fx_sw)
 104 {
 105         int min_xstate_size = sizeof(struct i387_fxsave_struct) +
 106                               sizeof(struct xsave_hdr_struct);
 107         unsigned int magic2;
 108
 109         if (__copy_from_user(fx_sw, &buf->sw_reserved[0], sizeof(*fx_sw)))
 110                 return -1;
 111
 112         /* Check for the first magic field and other error scenarios. */
 113         if (fx_sw->magic1 != FP_XSTATE_MAGIC1 ||
 114             fx_sw->xstate_size < min_xstate_size ||
 115             fx_sw->xstate_size > xstate_size ||
 116             fx_sw->xstate_size > fx_sw->extended_size)
 117                 return -1;
 118
 119         /*
 120          * Check for the presence of second magic word at the end of memory
 121          * layout. This detects the case where the user just copied the legacy
 122          * fpstate layout with out copying the extended state information
 123          * in the memory layout.
 124          */
 125         if (__get_user(magic2, (__u32 __user *)(fpstate + fx_sw->xstate_size))
 126             || magic2 != FP_XSTATE_MAGIC2)
 127                 return -1;
 128
 129         return 0;
 130 }
 131
 132 /*
 133  * Signal frame handlers.
 134  */
 135 static inline int save_fsave_header(struct task_struct *tsk, void __user *buf)
 136 {
 137         if (use_fxsr()) {
 138                 struct xsave_struct *xsave = &tsk->thread.fpu.state->xsave;
 139                 struct user_i387_ia32_struct env;
 140                 struct _fpstate_ia32 __user *fp = buf;
 141
 142                 convert_from_fxsr(&env, tsk);
 143
 144                 if (__copy_to_user(buf, &env, sizeof(env)) ||
 145                     __put_user(xsave->i387.swd, &fp->status) ||
 146                     __put_user(X86_FXSR_MAGIC, &fp->magic))
 147                         return -1;
 148         } else {
 149                 struct i387_fsave_struct __user *fp = buf;
 150                 u32 swd;
 151                 if (__get_user(swd, &fp->swd) || __put_user(swd, &fp->status))
 152                         return -1;
 153         }
 154
 155         return 0;
 156 }
 157
 158 static inline int save_xstate_epilog(void __user *buf, int ia32_frame)
 159 {
 160         struct xsave_struct __user *x = buf;
 161         struct _fpx_sw_bytes *sw_bytes;
 162         u32 xstate_bv;
 163         int err;
 164
 165         /* Setup the bytes not touched by the [f]xsave and reserved for SW. */
 166         sw_bytes = ia32_frame ? &fx_sw_reserved_ia32 : &fx_sw_reserved;
 167         err = __copy_to_user(&x->i387.sw_reserved, sw_bytes, sizeof(*sw_bytes));
 168
 169         if (!use_xsave())
 170                 return err;
 171
 172         err |= __put_user(FP_XSTATE_MAGIC2, (__u32 *)(buf + xstate_size));
 173
 174         /*
 175          * Read the xstate_bv which we copied (directly from the cpu or
 176          * from the state in task struct) to the user buffers.
 177          */
 178         err |= __get_user(xstate_bv, (__u32 *)&x->xsave_hdr.xstate_bv);
 179
 180         /*
 181          * For legacy compatible, we always set FP/SSE bits in the bit
 182          * vector while saving the state to the user context. This will
 183          * enable us capturing any changes(during sigreturn) to
 184          * the FP/SSE bits by the legacy applications which don't touch
 185          * xstate_bv in the xsave header.
 186          *
 187          * xsave aware apps can change the xstate_bv in the xsave
 188          * header as well as change any contents in the memory layout.
 189          * xrestore as part of sigreturn will capture all the changes.
 190          */
 191         xstate_bv |= XSTATE_FPSSE;
 192
 193         err |= __put_user(xstate_bv, (__u32 *)&x->xsave_hdr.xstate_bv);
 194
 195         return err;
 196 }
 197
 198 static inline int save_user_xstate(struct xsave_struct __user *buf)
 199 {
 200         int err;
 201
 202         if (use_xsave())
 203                 err = xsave_user(buf);
 204         else if (use_fxsr())
 205                 err = fxsave_user((struct i387_fxsave_struct __user *) buf);
 206         else
 207                 err = fsave_user((struct i387_fsave_struct __user *) buf);
 208
 209         if (unlikely(err) && __clear_user(buf, xstate_size))
 210                 err = -EFAULT;
 211         return err;
 212 }
 213
 214 /*
 215  * Save the fpu, extended register state to the user signal frame.
 216  *
 217  * 'buf_fx' is the 64-byte aligned pointer at which the [f|fx|x]save
 218  *  state is copied.
 219  *  'buf' points to the 'buf_fx' or to the fsave header followed by 'buf_fx'.
 220  *
 221  *      buf == buf_fx for 64-bit frames and 32-bit fsave frame.
 222  *      buf != buf_fx for 32-bit frames with fxstate.
 223  *
 224  * If the fpu, extended register state is live, save the state directly
 225  * to the user frame pointed by the aligned pointer 'buf_fx'. Otherwise,
 226  * copy the thread's fpu state to the user frame starting at 'buf_fx'.
 227  *
 228  * If this is a 32-bit frame with fxstate, put a fsave header before
 229  * the aligned state at 'buf_fx'.
 230  *
 231  * For [f]xsave state, update the SW reserved fields in the [f]xsave frame
 232  * indicating the absence/presence of the extended state to the user.
 233  */
 234 int save_xstate_sig(void __user *buf, void __user *buf_fx, int size)
 235 {
 236         struct xsave_struct *xsave = &current->thread.fpu.state->xsave;
 237         struct task_struct *tsk = current;
 238         int ia32_fxstate = (buf != buf_fx);
 239
 240         ia32_fxstate &= (config_enabled(CONFIG_X86_32) ||
 241                          config_enabled(CONFIG_IA32_EMULATION));
 242
 243         if (!access_ok(VERIFY_WRITE, buf, size))
 244                 return -EACCES;
 245
 246         if (!static_cpu_has(X86_FEATURE_FPU))
 247                 return fpregs_soft_get(current, NULL, 0,
 248                         sizeof(struct user_i387_ia32_struct), NULL,
 249                         (struct _fpstate_ia32 __user *) buf) ? -1 : 1;
 250
 251         if (user_has_fpu()) {
 252                 /* Save the live register state to the user directly. */
 253                 if (save_user_xstate(buf_fx))
 254                         return -1;
 255                 /* Update the thread's fxstate to save the fsave header. */
 256                 if (ia32_fxstate)
 257                         fpu_fxsave(&tsk->thread.fpu);
 258         } else {
 259                 sanitize_i387_state(tsk);
 260                 if (__copy_to_user(buf_fx, xsave, xstate_size))
 261                         return -1;
 262         }
 263
 264         /* Save the fsave header for the 32-bit frames. */
 265         if ((ia32_fxstate || !use_fxsr()) && save_fsave_header(tsk, buf))
 266                 return -1;
 267
 268         if (use_fxsr() && save_xstate_epilog(buf_fx, ia32_fxstate))
 269                 return -1;
 270
 271         drop_init_fpu(tsk);     /* trigger finit */
 272
 273         return 0;
 274 }
 275
 276 static inline void
 277 sanitize_restored_xstate(struct task_struct *tsk,
 278                          struct user_i387_ia32_struct *ia32_env,
 279                          u64 xstate_bv, int fx_only)
 280 {
 281         struct xsave_struct *xsave = &tsk->thread.fpu.state->xsave;
 282         struct xsave_hdr_struct *xsave_hdr = &xsave->xsave_hdr;
 283
 284         if (use_xsave()) {
 285                 /* These bits must be zero. */
 286                 xsave_hdr->reserved1[0] = xsave_hdr->reserved1[1] = 0;
 287
 288                 /*
 289                  * Init the state that is not present in the memory
 290                  * layout and not enabled by the OS.
 291                  */
 292                 if (fx_only)
 293                         xsave_hdr->xstate_bv = XSTATE_FPSSE;
 294                 else
 295                         xsave_hdr->xstate_bv &= (pcntxt_mask & xstate_bv);
 296         }
 297
 298         if (use_fxsr()) {
 299                 /*
 300                  * mscsr reserved bits must be masked to zero for security
 301                  * reasons.
 302                  */
 303                 xsave->i387.mxcsr &= mxcsr_feature_mask;
 304
 305                 convert_to_fxsr(tsk, ia32_env);
 306         }
 307 }
 308
 309 /*
 310  * Restore the extended state if present. Otherwise, restore the FP/SSE state.
 311  */
 312 static inline int restore_user_xstate(void __user *buf, u64 xbv, int fx_only)
 313 {
 314         if (use_xsave()) {
 315                 if ((unsigned long)buf % 64 || fx_only) {
 316                         u64 init_bv = pcntxt_mask & ~XSTATE_FPSSE;
 317                         xrstor_state(init_xstate_buf, init_bv);
 318                         return fxrstor_user(buf);
 319                 } else {
 320                         u64 init_bv = pcntxt_mask & ~xbv;
 321                         if (unlikely(init_bv))
 322                                 xrstor_state(init_xstate_buf, init_bv);
 323                         return xrestore_user(buf, xbv);
 324                 }
 325         } else if (use_fxsr()) {
 326                 return fxrstor_user(buf);
 327         } else
 328                 return frstor_user(buf);
 329 }
 330
 331 int __restore_xstate_sig(void __user *buf, void __user *buf_fx, int size)
 332 {
 333         int ia32_fxstate = (buf != buf_fx);
 334         struct task_struct *tsk = current;
 335         int state_size = xstate_size;
 336         u64 xstate_bv = 0;
 337         int fx_only = 0;
 338
 339         ia32_fxstate &= (config_enabled(CONFIG_X86_32) ||
 340                          config_enabled(CONFIG_IA32_EMULATION));
 341
 342         if (!buf) {
 343                 drop_init_fpu(tsk);
 344                 return 0;
 345         }
 346
 347         if (!access_ok(VERIFY_READ, buf, size))
 348                 return -EACCES;
 349
 350         if (!used_math() && init_fpu(tsk))
 351                 return -1;
 352
 353         if (!static_cpu_has(X86_FEATURE_FPU))
 354                 return fpregs_soft_set(current, NULL,
 355                                        0, sizeof(struct user_i387_ia32_struct),
 356                                        NULL, buf) != 0;
 357
 358         if (use_xsave()) {
 359                 struct _fpx_sw_bytes fx_sw_user;
 360                 if (unlikely(check_for_xstate(buf_fx, buf_fx, &fx_sw_user))) {
 361                         /*
 362                          * Couldn't find the extended state information in the
 363                          * memory layout. Restore just the FP/SSE and init all
 364                          * the other extended state.
 365                          */
 366                         state_size = sizeof(struct i387_fxsave_struct);
 367                         fx_only = 1;
 368                 } else {
 369                         state_size = fx_sw_user.xstate_size;
 370                         xstate_bv = fx_sw_user.xstate_bv;
 371                 }
 372         }
 373
 374         if (ia32_fxstate) {
 375                 /*
 376                  * For 32-bit frames with fxstate, copy the user state to the
 377                  * thread's fpu state, reconstruct fxstate from the fsave
 378                  * header. Sanitize the copied state etc.
 379                  */
 380                 struct xsave_struct *xsave = &tsk->thread.fpu.state->xsave;
 381                 struct user_i387_ia32_struct env;
 382                 int err = 0;
 383
 384                 /*
 385                  * Drop the current fpu which clears used_math(). This ensures
 386                  * that any context-switch during the copy of the new state,
 387                  * avoids the intermediate state from getting restored/saved.
 388                  * Thus avoiding the new restored state from getting corrupted.
 389                  * We will be ready to restore/save the state only after
 390                  * set_used_math() is again set.
 391                  */
 392                 drop_fpu(tsk);
 393
 394                 if (__copy_from_user(xsave, buf_fx, state_size) ||
 395                     __copy_from_user(&env, buf, sizeof(env))) {
 396                         err = -1;
 397                 } else {
 398                         sanitize_restored_xstate(tsk, &env, xstate_bv, fx_only);
 399                         set_used_math();
 400                 }
 401
 402                 if (use_eager_fpu())
 403                         math_state_restore();
 404
 405                 return err;
 406         } else {
 407                 /*
 408                  * For 64-bit frames and 32-bit fsave frames, restore the user
 409                  * state to the registers directly (with exceptions handled).
 410                  */
 411                 user_fpu_begin();
 412                 if (restore_user_xstate(buf_fx, xstate_bv, fx_only)) {
 413                         drop_init_fpu(tsk);
 414                         return -1;
 415                 }
 416         }
 417
 418         return 0;
 419 }
 420
 421 /*
 422  * Prepare the SW reserved portion of the fxsave memory layout, indicating
 423  * the presence of the extended state information in the memory layout
 424  * pointed by the fpstate pointer in the sigcontext.
 425  * This will be saved when ever the FP and extended state context is
 426  * saved on the user stack during the signal handler delivery to the user.
 427  */
 428 static void prepare_fx_sw_frame(void)
 429 {
 430         int fsave_header_size = sizeof(struct i387_fsave_struct);
 431         int size = xstate_size + FP_XSTATE_MAGIC2_SIZE;
 432
 433         if (config_enabled(CONFIG_X86_32))
 434                 size += fsave_header_size;
 435
 436         fx_sw_reserved.magic1 = FP_XSTATE_MAGIC1;
 437         fx_sw_reserved.extended_size = size;
 438         fx_sw_reserved.xstate_bv = pcntxt_mask;
 439         fx_sw_reserved.xstate_size = xstate_size;
 440
 441         if (config_enabled(CONFIG_IA32_EMULATION)) {
 442                 fx_sw_reserved_ia32 = fx_sw_reserved;
 443                 fx_sw_reserved_ia32.extended_size += fsave_header_size;
 444         }
 445 }
 446
 447 /*
 448  * Enable the extended processor state save/restore feature
 449  */
 450 static inline void xstate_enable(void)
 451 {
 452         set_in_cr4(X86_CR4_OSXSAVE);
 453         xsetbv(XCR_XFEATURE_ENABLED_MASK, pcntxt_mask);
 454 }
 455
 456 /*
 457  * Record the offsets and sizes of different state managed by the xsave
 458  * memory layout.
 459  */
 460 static void __init setup_xstate_features(void)
 461 {
 462         int eax, ebx, ecx, edx, leaf = 0x2;
 463
 464         xstate_features = fls64(pcntxt_mask);
 465         xstate_offsets = alloc_bootmem(xstate_features * sizeof(int));
 466         xstate_sizes = alloc_bootmem(xstate_features * sizeof(int));
 467
 468         do {
 469                 cpuid_count(XSTATE_CPUID, leaf, &eax, &ebx, &ecx, &edx);
 470
 471                 if (eax == 0)
 472                         break;
 473
 474                 xstate_offsets[leaf] = ebx;
 475                 xstate_sizes[leaf] = eax;
 476
 477                 leaf++;
 478         } while (1);
 479 }
 480
 481 /*
 482  * setup the xstate image representing the init state
 483  */
 484 static void __init setup_init_fpu_buf(void)
 485 {
 486         /*
 487          * Setup init_xstate_buf to represent the init state of
 488          * all the features managed by the xsave
 489          */
 490         init_xstate_buf = alloc_bootmem_align(xstate_size,
 491                                               __alignof__(struct xsave_struct));
 492         fx_finit(&init_xstate_buf->i387);
 493
 494         if (!cpu_has_xsave)
 495                 return;
 496
 497         setup_xstate_features();
 498
 499         /*
 500          * Init all the features state with header_bv being 0x0
 501          */
 502         xrstor_state(init_xstate_buf, -1);
 503         /*
 504          * Dump the init state again. This is to identify the init state
 505          * of any feature which is not represented by all zero's.
 506          */
 507         xsave_state(init_xstate_buf, -1);
 508 }
 509
 510 static enum { AUTO, ENABLE, DISABLE } eagerfpu = AUTO;
 511 static int __init eager_fpu_setup(char *s)
 512 {
 513         if (!strcmp(s, "on"))
 514                 eagerfpu = ENABLE;
 515         else if (!strcmp(s, "off"))
 516                 eagerfpu = DISABLE;
 517         else if (!strcmp(s, "auto"))
 518                 eagerfpu = AUTO;
 519         return 1;
 520 }
 521 __setup("eagerfpu=", eager_fpu_setup);
 522
 523 /*
 524  * Enable and initialize the xsave feature.
 525  */
 526 static void __init xstate_enable_boot_cpu(void)
 527 {
 528         unsigned int eax, ebx, ecx, edx;
 529
 530         if (boot_cpu_data.cpuid_level < XSTATE_CPUID) {
 531                 WARN(1, KERN_ERR "XSTATE_CPUID missing\n");
 532                 return;
 533         }
 534
 535         cpuid_count(XSTATE_CPUID, 0, &eax, &ebx, &ecx, &edx);
 536         pcntxt_mask = eax + ((u64)edx << 32);
 537
 538         if ((pcntxt_mask & XSTATE_FPSSE) != XSTATE_FPSSE) {
 539                 pr_err("FP/SSE not shown under xsave features 0x%llx\n",
 540                        pcntxt_mask);
 541                 BUG();
 542         }
 543
 544         /*
 545          * Support only the state known to OS.
 546          */
 547         pcntxt_mask = pcntxt_mask & XCNTXT_MASK;
 548
 549         xstate_enable();
 550
 551         /*
 552          * Recompute the context size for enabled features
 553          */
 554         cpuid_count(XSTATE_CPUID, 0, &eax, &ebx, &ecx, &edx);
 555         xstate_size = ebx;
 556
 557         update_regset_xstate_info(xstate_size, pcntxt_mask);
 558         prepare_fx_sw_frame();
 559         setup_init_fpu_buf();
 560
 561         /* Auto enable eagerfpu for xsaveopt */
 562         if (cpu_has_xsaveopt && eagerfpu != DISABLE)
 563                 eagerfpu = ENABLE;
 564
 565         pr_info("enabled xstate_bv 0x%llx, cntxt size 0x%x\n",
 566                 pcntxt_mask, xstate_size);
 567 }
 568
 569 /*
 570  * For the very first instance, this calls xstate_enable_boot_cpu();
 571  * for all subsequent instances, this calls xstate_enable().
 572  *
 573  * This is somewhat obfuscated due to the lack of powerful enough
 574  * overrides for the section checks.
 575  */
 576 void xsave_init(void)
 577 {
 578         static __refdata void (*next_func)(void) = xstate_enable_boot_cpu;
 579         void (*this_func)(void);
 580
 581         if (!cpu_has_xsave)
 582                 return;
 583
 584         this_func = next_func;
 585         next_func = xstate_enable;
 586         this_func();
 587 }
 588
 589 static inline void __init eager_fpu_init_bp(void)
 590 {
 591         current->thread.fpu.state =
 592             alloc_bootmem_align(xstate_size, __alignof__(struct xsave_struct));
 593         if (!init_xstate_buf)
 594                 setup_init_fpu_buf();
 595 }
 596
 597 void eager_fpu_init(void)
 598 {
 599         static __refdata void (*boot_func)(void) = eager_fpu_init_bp;
 600
 601         clear_used_math();
 602         current_thread_info()->status = 0;
 603
 604         if (eagerfpu == ENABLE)
 605                 setup_force_cpu_cap(X86_FEATURE_EAGER_FPU);
 606
 607         if (!cpu_has_eager_fpu) {
 608                 stts();
 609                 return;
 610         }
 611
 612         if (boot_func) {
 613                 boot_func();
 614                 boot_func = NULL;
 615         }
 616
 617         /*
 618          * This is same as math_state_restore(). But use_xsave() is
 619          * not yet patched to use math_state_restore().
 620          */
 621         init_fpu(current);
 622         __thread_fpu_begin(current);
 623         if (cpu_has_xsave)
 624                 xrstor_state(init_xstate_buf, -1);
 625         else
 626                 fxrstor_checking(&init_xstate_buf->i387);
 627 }