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  * On others, we can do a kernel_fpu_begin/end() pair *ONLY* if that
  26  * pair does nothing at all: the thread must not have fpu (so
  27  * that we don't try to save the FPU state), and TS must
  28  * be set (so that the clts/stts pair does nothing that is
  29  * visible in the interrupted kernel thread).
  30  *
  31  * Except for the eagerfpu case when we return 1 unless we've already
  32  * been eager and saved the state in kernel_fpu_begin().
  33  */
  34 static inline bool interrupted_kernel_fpu_idle(void)
  35 {
  36         if (use_eager_fpu())
  37                 return __thread_has_fpu(current);
  38
  39         return !__thread_has_fpu(current) &&
  40                 (read_cr0() & X86_CR0_TS);
  41 }
  42
  43 /*
  44  * Were we in user mode (or vm86 mode) when we were
  45  * interrupted?
  46  *
  47  * Doing kernel_fpu_begin/end() is ok if we are running
  48  * in an interrupt context from user mode - we'll just
  49  * save the FPU state as required.
  50  */
  51 static inline bool interrupted_user_mode(void)
  52 {
  53         struct pt_regs *regs = get_irq_regs();
  54         return regs && user_mode_vm(regs);
  55 }
  56
  57 /*
  58  * Can we use the FPU in kernel mode with the
  59  * whole "kernel_fpu_begin/end()" sequence?
  60  *
  61  * It's always ok in process context (ie "not interrupt")
  62  * but it is sometimes ok even from an irq.
  63  */
  64 bool irq_fpu_usable(void)
  65 {
  66         return !in_interrupt() ||
  67                 interrupted_user_mode() ||
  68                 interrupted_kernel_fpu_idle();
  69 }
  70 EXPORT_SYMBOL(irq_fpu_usable);
  71
  72 void __kernel_fpu_begin(void)
  73 {
  74         struct task_struct *me = current;
  75
  76         if (__thread_has_fpu(me)) {
  77                 __thread_clear_has_fpu(me);
  78                 __save_init_fpu(me);
  79                 /* We do 'stts()' in __kernel_fpu_end() */
  80         } else if (!use_eager_fpu()) {
  81                 this_cpu_write(fpu_owner_task, NULL);
  82                 clts();
  83         }
  84 }
  85 EXPORT_SYMBOL(__kernel_fpu_begin);
  86
  87 void __kernel_fpu_end(void)
  88 {
  89         if (use_eager_fpu()) {
  90                 /*
  91                  * For eager fpu, most the time, tsk_used_math() is true.
  92                  * Restore the user math as we are done with the kernel usage.
  93                  * At few instances during thread exit, signal handling etc,
  94                  * tsk_used_math() is false. Those few places will take proper
  95                  * actions, so we don't need to restore the math here.
  96                  */
  97                 if (likely(tsk_used_math(current)))
  98                         math_state_restore();
  99         } else {
 100                 stts();
 101         }
 102 }
 103 EXPORT_SYMBOL(__kernel_fpu_end);
 104
 105 void unlazy_fpu(struct task_struct *tsk)
 106 {
 107         preempt_disable();
 108         if (__thread_has_fpu(tsk)) {
 109                 __save_init_fpu(tsk);
 110                 __thread_fpu_end(tsk);
 111         } else
 112                 tsk->thread.fpu_counter = 0;
 113         preempt_enable();
 114 }
 115 EXPORT_SYMBOL(unlazy_fpu);
 116
 117 unsigned int mxcsr_feature_mask __read_mostly = 0xffffffffu;
 118 unsigned int xstate_size;
 119 EXPORT_SYMBOL_GPL(xstate_size);
 120 static struct i387_fxsave_struct fx_scratch;
 121
 122 static void mxcsr_feature_mask_init(void)
 123 {
 124         unsigned long mask = 0;
 125
 126         if (cpu_has_fxsr) {
 127                 memset(&fx_scratch, 0, sizeof(struct i387_fxsave_struct));
 128                 asm volatile("fxsave %0" : "+m" (fx_scratch));
 129                 mask = fx_scratch.mxcsr_mask;
 130                 if (mask == 0)
 131                         mask = 0x0000ffbf;
 132         }
 133         mxcsr_feature_mask &= mask;
 134 }
 135
 136 static void init_thread_xstate(void)
 137 {
 138         /*
 139          * Note that xstate_size might be overwriten later during
 140          * xsave_init().
 141          */
 142
 143         if (!cpu_has_fpu) {
 144                 /*
 145                  * Disable xsave as we do not support it if i387
 146                  * emulation is enabled.
 147                  */
 148                 setup_clear_cpu_cap(X86_FEATURE_XSAVE);
 149                 setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT);
 150                 xstate_size = sizeof(struct i387_soft_struct);
 151                 return;
 152         }
 153
 154         if (cpu_has_fxsr)
 155                 xstate_size = sizeof(struct i387_fxsave_struct);
 156         else
 157                 xstate_size = sizeof(struct i387_fsave_struct);
 158 }
 159
 160 /*
 161  * Called at bootup to set up the initial FPU state that is later cloned
 162  * into all processes.
 163  */
 164
 165 void fpu_init(void)
 166 {
 167         unsigned long cr0;
 168         unsigned long cr4_mask = 0;
 169
 170 #ifndef CONFIG_MATH_EMULATION
 171         if (!cpu_has_fpu) {
 172                 pr_emerg("No FPU found and no math emulation present\n");
 173                 pr_emerg("Giving up\n");
 174                 for (;;)
 175                         asm volatile("hlt");
 176         }
 177 #endif
 178         if (cpu_has_fxsr)
 179                 cr4_mask |= X86_CR4_OSFXSR;
 180         if (cpu_has_xmm)
 181                 cr4_mask |= X86_CR4_OSXMMEXCPT;
 182         if (cr4_mask)
 183                 set_in_cr4(cr4_mask);
 184
 185         cr0 = read_cr0();
 186         cr0 &= ~(X86_CR0_TS|X86_CR0_EM); /* clear TS and EM */
 187         if (!cpu_has_fpu)
 188                 cr0 |= X86_CR0_EM;
 189         write_cr0(cr0);
 190
 191         /*
 192          * init_thread_xstate is only called once to avoid overriding
 193          * xstate_size during boot time or during CPU hotplug.
 194          */
 195         if (xstate_size == 0)
 196                 init_thread_xstate();
 197
 198         mxcsr_feature_mask_init();
 199         xsave_init();
 200         eager_fpu_init();
 201 }
 202
 203 void fpu_finit(struct fpu *fpu)
 204 {
 205         if (!cpu_has_fpu) {
 206                 finit_soft_fpu(&fpu->state->soft);
 207                 return;
 208         }
 209
 210         if (cpu_has_fxsr) {
 211                 fx_finit(&fpu->state->fxsave);
 212         } else {
 213                 struct i387_fsave_struct *fp = &fpu->state->fsave;
 214                 memset(fp, 0, xstate_size);
 215                 fp->cwd = 0xffff037fu;
 216                 fp->swd = 0xffff0000u;
 217                 fp->twd = 0xffffffffu;
 218                 fp->fos = 0xffff0000u;
 219         }
 220 }
 221 EXPORT_SYMBOL_GPL(fpu_finit);
 222
 223 /*
 224  * The _current_ task is using the FPU for the first time
 225  * so initialize it and set the mxcsr to its default
 226  * value at reset if we support XMM instructions and then
 227  * remember the current task has used the FPU.
 228  */
 229 int init_fpu(struct task_struct *tsk)
 230 {
 231         int ret;
 232
 233         if (tsk_used_math(tsk)) {
 234                 if (cpu_has_fpu && tsk == current)
 235                         unlazy_fpu(tsk);
 236                 tsk->thread.fpu.last_cpu = ~0;
 237                 return 0;
 238         }
 239
 240         /*
 241          * Memory allocation at the first usage of the FPU and other state.
 242          */
 243         ret = fpu_alloc(&tsk->thread.fpu);
 244         if (ret)
 245                 return ret;
 246
 247         fpu_finit(&tsk->thread.fpu);
 248
 249         set_stopped_child_used_math(tsk);
 250         return 0;
 251 }
 252 EXPORT_SYMBOL_GPL(init_fpu);
 253
 254 /*
 255  * The xstateregs_active() routine is the same as the fpregs_active() routine,
 256  * as the "regset->n" for the xstate regset will be updated based on the feature
 257  * capabilites supported by the xsave.
 258  */
 259 int fpregs_active(struct task_struct *target, const struct user_regset *regset)
 260 {
 261         return tsk_used_math(target) ? regset->n : 0;
 262 }
 263
 264 int xfpregs_active(struct task_struct *target, const struct user_regset *regset)
 265 {
 266         return (cpu_has_fxsr && tsk_used_math(target)) ? regset->n : 0;
 267 }
 268
 269 int xfpregs_get(struct task_struct *target, const struct user_regset *regset,
 270                 unsigned int pos, unsigned int count,
 271                 void *kbuf, void __user *ubuf)
 272 {
 273         int ret;
 274
 275         if (!cpu_has_fxsr)
 276                 return -ENODEV;
 277
 278         ret = init_fpu(target);
 279         if (ret)
 280                 return ret;
 281
 282         sanitize_i387_state(target);
 283
 284         return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
 285                                    &target->thread.fpu.state->fxsave, 0, -1);
 286 }
 287
 288 int xfpregs_set(struct task_struct *target, const struct user_regset *regset,
 289                 unsigned int pos, unsigned int count,
 290                 const void *kbuf, const void __user *ubuf)
 291 {
 292         int ret;
 293
 294         if (!cpu_has_fxsr)
 295                 return -ENODEV;
 296
 297         ret = init_fpu(target);
 298         if (ret)
 299                 return ret;
 300
 301         sanitize_i387_state(target);
 302
 303         ret = user_regset_copyin(&pos, &count, &kbuf, &ubuf,
 304                                  &target->thread.fpu.state->fxsave, 0, -1);
 305
 306         /*
 307          * mxcsr reserved bits must be masked to zero for security reasons.
 308          */
 309         target->thread.fpu.state->fxsave.mxcsr &= mxcsr_feature_mask;
 310
 311         /*
 312          * update the header bits in the xsave header, indicating the
 313          * presence of FP and SSE state.
 314          */
 315         if (cpu_has_xsave)
 316                 target->thread.fpu.state->xsave.xsave_hdr.xstate_bv |= XSTATE_FPSSE;
 317
 318         return ret;
 319 }
 320
 321 int xstateregs_get(struct task_struct *target, const struct user_regset *regset,
 322                 unsigned int pos, unsigned int count,
 323                 void *kbuf, void __user *ubuf)
 324 {
 325         int ret;
 326
 327         if (!cpu_has_xsave)
 328                 return -ENODEV;
 329
 330         ret = init_fpu(target);
 331         if (ret)
 332                 return ret;
 333
 334         /*
 335          * Copy the 48bytes defined by the software first into the xstate
 336          * memory layout in the thread struct, so that we can copy the entire
 337          * xstateregs to the user using one user_regset_copyout().
 338          */
 339         memcpy(&target->thread.fpu.state->fxsave.sw_reserved,
 340                xstate_fx_sw_bytes, sizeof(xstate_fx_sw_bytes));
 341
 342         /*
 343          * Copy the xstate memory layout.
 344          */
 345         ret = user_regset_copyout(&pos, &count, &kbuf, &ubuf,
 346                                   &target->thread.fpu.state->xsave, 0, -1);
 347         return ret;
 348 }
 349
 350 int xstateregs_set(struct task_struct *target, const struct user_regset *regset,
 351                   unsigned int pos, unsigned int count,
 352                   const void *kbuf, const void __user *ubuf)
 353 {
 354         int ret;
 355         struct xsave_hdr_struct *xsave_hdr;
 356
 357         if (!cpu_has_xsave)
 358                 return -ENODEV;
 359
 360         ret = init_fpu(target);
 361         if (ret)
 362                 return ret;
 363
 364         ret = user_regset_copyin(&pos, &count, &kbuf, &ubuf,
 365                                  &target->thread.fpu.state->xsave, 0, -1);
 366
 367         /*
 368          * mxcsr reserved bits must be masked to zero for security reasons.
 369          */
 370         target->thread.fpu.state->fxsave.mxcsr &= mxcsr_feature_mask;
 371
 372         xsave_hdr = &target->thread.fpu.state->xsave.xsave_hdr;
 373
 374         xsave_hdr->xstate_bv &= pcntxt_mask;
 375         /*
 376          * These bits must be zero.
 377          */
 378         memset(xsave_hdr->reserved, 0, 48);
 379
 380         return ret;
 381 }
 382
 383 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
 384
 385 /*
 386  * FPU tag word conversions.
 387  */
 388
 389 static inline unsigned short twd_i387_to_fxsr(unsigned short twd)
 390 {
 391         unsigned int tmp; /* to avoid 16 bit prefixes in the code */
 392
 393         /* Transform each pair of bits into 01 (valid) or 00 (empty) */
 394         tmp = ~twd;
 395         tmp = (tmp | (tmp>>1)) & 0x5555; /* 0V0V0V0V0V0V0V0V */
 396         /* and move the valid bits to the lower byte. */
 397         tmp = (tmp | (tmp >> 1)) & 0x3333; /* 00VV00VV00VV00VV */
 398         tmp = (tmp | (tmp >> 2)) & 0x0f0f; /* 0000VVVV0000VVVV */
 399         tmp = (tmp | (tmp >> 4)) & 0x00ff; /* 00000000VVVVVVVV */
 400
 401         return tmp;
 402 }
 403
 404 #define FPREG_ADDR(f, n)        ((void *)&(f)->st_space + (n) * 16)
 405 #define FP_EXP_TAG_VALID        0
 406 #define FP_EXP_TAG_ZERO         1
 407 #define FP_EXP_TAG_SPECIAL      2
 408 #define FP_EXP_TAG_EMPTY        3
 409
 410 static inline u32 twd_fxsr_to_i387(struct i387_fxsave_struct *fxsave)
 411 {
 412         struct _fpxreg *st;
 413         u32 tos = (fxsave->swd >> 11) & 7;
 414         u32 twd = (unsigned long) fxsave->twd;
 415         u32 tag;
 416         u32 ret = 0xffff0000u;
 417         int i;
 418
 419         for (i = 0; i < 8; i++, twd >>= 1) {
 420                 if (twd & 0x1) {
 421                         st = FPREG_ADDR(fxsave, (i - tos) & 7);
 422
 423                         switch (st->exponent & 0x7fff) {
 424                         case 0x7fff:
 425                                 tag = FP_EXP_TAG_SPECIAL;
 426                                 break;
 427                         case 0x0000:
 428                                 if (!st->significand[0] &&
 429                                     !st->significand[1] &&
 430                                     !st->significand[2] &&
 431                                     !st->significand[3])
 432                                         tag = FP_EXP_TAG_ZERO;
 433                                 else
 434                                         tag = FP_EXP_TAG_SPECIAL;
 435                                 break;
 436                         default:
 437                                 if (st->significand[3] & 0x8000)
 438                                         tag = FP_EXP_TAG_VALID;
 439                                 else
 440                                         tag = FP_EXP_TAG_SPECIAL;
 441                                 break;
 442                         }
 443                 } else {
 444                         tag = FP_EXP_TAG_EMPTY;
 445                 }
 446                 ret |= tag << (2 * i);
 447         }
 448         return ret;
 449 }
 450
 451 /*
 452  * FXSR floating point environment conversions.
 453  */
 454
 455 void
 456 convert_from_fxsr(struct user_i387_ia32_struct *env, struct task_struct *tsk)
 457 {
 458         struct i387_fxsave_struct *fxsave = &tsk->thread.fpu.state->fxsave;
 459         struct _fpreg *to = (struct _fpreg *) &env->st_space[0];
 460         struct _fpxreg *from = (struct _fpxreg *) &fxsave->st_space[0];
 461         int i;
 462
 463         env->cwd = fxsave->cwd | 0xffff0000u;
 464         env->swd = fxsave->swd | 0xffff0000u;
 465         env->twd = twd_fxsr_to_i387(fxsave);
 466
 467 #ifdef CONFIG_X86_64
 468         env->fip = fxsave->rip;
 469         env->foo = fxsave->rdp;
 470         /*
 471          * should be actually ds/cs at fpu exception time, but
 472          * that information is not available in 64bit mode.
 473          */
 474         env->fcs = task_pt_regs(tsk)->cs;
 475         if (tsk == current) {
 476                 savesegment(ds, env->fos);
 477         } else {
 478                 env->fos = tsk->thread.ds;
 479         }
 480         env->fos |= 0xffff0000;
 481 #else
 482         env->fip = fxsave->fip;
 483         env->fcs = (u16) fxsave->fcs | ((u32) fxsave->fop << 16);
 484         env->foo = fxsave->foo;
 485         env->fos = fxsave->fos;
 486 #endif
 487
 488         for (i = 0; i < 8; ++i)
 489                 memcpy(&to[i], &from[i], sizeof(to[0]));
 490 }
 491
 492 void convert_to_fxsr(struct task_struct *tsk,
 493                      const struct user_i387_ia32_struct *env)
 494
 495 {
 496         struct i387_fxsave_struct *fxsave = &tsk->thread.fpu.state->fxsave;
 497         struct _fpreg *from = (struct _fpreg *) &env->st_space[0];
 498         struct _fpxreg *to = (struct _fpxreg *) &fxsave->st_space[0];
 499         int i;
 500
 501         fxsave->cwd = env->cwd;
 502         fxsave->swd = env->swd;
 503         fxsave->twd = twd_i387_to_fxsr(env->twd);
 504         fxsave->fop = (u16) ((u32) env->fcs >> 16);
 505 #ifdef CONFIG_X86_64
 506         fxsave->rip = env->fip;
 507         fxsave->rdp = env->foo;
 508         /* cs and ds ignored */
 509 #else
 510         fxsave->fip = env->fip;
 511         fxsave->fcs = (env->fcs & 0xffff);
 512         fxsave->foo = env->foo;
 513         fxsave->fos = env->fos;
 514 #endif
 515
 516         for (i = 0; i < 8; ++i)
 517                 memcpy(&to[i], &from[i], sizeof(from[0]));
 518 }
 519
 520 int fpregs_get(struct task_struct *target, const struct user_regset *regset,
 521                unsigned int pos, unsigned int count,
 522                void *kbuf, void __user *ubuf)
 523 {
 524         struct user_i387_ia32_struct env;
 525         int ret;
 526
 527         ret = init_fpu(target);
 528         if (ret)
 529                 return ret;
 530
 531         if (!static_cpu_has(X86_FEATURE_FPU))
 532                 return fpregs_soft_get(target, regset, pos, count, kbuf, ubuf);
 533
 534         if (!cpu_has_fxsr)
 535                 return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
 536                                            &target->thread.fpu.state->fsave, 0,
 537                                            -1);
 538
 539         sanitize_i387_state(target);
 540
 541         if (kbuf && pos == 0 && count == sizeof(env)) {
 542                 convert_from_fxsr(kbuf, target);
 543                 return 0;
 544         }
 545
 546         convert_from_fxsr(&env, target);
 547
 548         return user_regset_copyout(&pos, &count, &kbuf, &ubuf, &env, 0, -1);
 549 }
 550
 551 int fpregs_set(struct task_struct *target, const struct user_regset *regset,
 552                unsigned int pos, unsigned int count,
 553                const void *kbuf, const void __user *ubuf)
 554 {
 555         struct user_i387_ia32_struct env;
 556         int ret;
 557
 558         ret = init_fpu(target);
 559         if (ret)
 560                 return ret;
 561
 562         sanitize_i387_state(target);
 563
 564         if (!static_cpu_has(X86_FEATURE_FPU))
 565                 return fpregs_soft_set(target, regset, pos, count, kbuf, ubuf);
 566
 567         if (!cpu_has_fxsr)
 568                 return user_regset_copyin(&pos, &count, &kbuf, &ubuf,
 569                                           &target->thread.fpu.state->fsave, 0,
 570                                           -1);
 571
 572         if (pos > 0 || count < sizeof(env))
 573                 convert_from_fxsr(&env, target);
 574
 575         ret = user_regset_copyin(&pos, &count, &kbuf, &ubuf, &env, 0, -1);
 576         if (!ret)
 577                 convert_to_fxsr(target, &env);
 578
 579         /*
 580          * update the header bit in the xsave header, indicating the
 581          * presence of FP.
 582          */
 583         if (cpu_has_xsave)
 584                 target->thread.fpu.state->xsave.xsave_hdr.xstate_bv |= XSTATE_FP;
 585         return ret;
 586 }
 587
 588 /*
 589  * FPU state for core dumps.
 590  * This is only used for a.out dumps now.
 591  * It is declared generically using elf_fpregset_t (which is
 592  * struct user_i387_struct) but is in fact only used for 32-bit
 593  * dumps, so on 64-bit it is really struct user_i387_ia32_struct.
 594  */
 595 int dump_fpu(struct pt_regs *regs, struct user_i387_struct *fpu)
 596 {
 597         struct task_struct *tsk = current;
 598         int fpvalid;
 599
 600         fpvalid = !!used_math();
 601         if (fpvalid)
 602                 fpvalid = !fpregs_get(tsk, NULL,
 603                                       0, sizeof(struct user_i387_ia32_struct),
 604                                       fpu, NULL);
 605
 606         return fpvalid;
 607 }
 608 EXPORT_SYMBOL(dump_fpu);
 609
 610 #endif  /* CONFIG_X86_32 || CONFIG_IA32_EMULATION */
 611
 612 static int __init no_387(char *s)
 613 {
 614         setup_clear_cpu_cap(X86_FEATURE_FPU);
 615         return 1;
 616 }
 617
 618 __setup("no387", no_387);
 619
 620 void fpu_detect(struct cpuinfo_x86 *c)
 621 {
 622         unsigned long cr0;
 623         u16 fsw, fcw;
 624
 625         fsw = fcw = 0xffff;
 626
 627         cr0 = read_cr0();
 628         cr0 &= ~(X86_CR0_TS | X86_CR0_EM);
 629         write_cr0(cr0);
 630
 631         asm volatile("fninit ; fnstsw %0 ; fnstcw %1"
 632                      : "+m" (fsw), "+m" (fcw));
 633
 634         if (fsw == 0 && (fcw & 0x103f) == 0x003f)
 635                 set_cpu_cap(c, X86_FEATURE_FPU);
 636         else
 637                 clear_cpu_cap(c, X86_FEATURE_FPU);
 638
 639         /* The final cr0 value is set in fpu_init() */
 640 }