arch/x86/kernel/ptrace.c

   1 /* By Ross Biro 1/23/92 */
   2 /*
   3  * Pentium III FXSR, SSE support
   4  *      Gareth Hughes <gareth@valinux.com>, May 2000
   5  *
   6  * BTS tracing
   7  *      Markus Metzger <markus.t.metzger@intel.com>, Dec 2007
   8  */
   9
  10 #include <linux/kernel.h>
  11 #include <linux/sched.h>
  12 #include <linux/mm.h>
  13 #include <linux/smp.h>
  14 #include <linux/errno.h>
  15 #include <linux/ptrace.h>
  16 #include <linux/regset.h>
  17 #include <linux/tracehook.h>
  18 #include <linux/user.h>
  19 #include <linux/elf.h>
  20 #include <linux/security.h>
  21 #include <linux/audit.h>
  22 #include <linux/seccomp.h>
  23 #include <linux/signal.h>
  24
  25 #include <asm/uaccess.h>
  26 #include <asm/pgtable.h>
  27 #include <asm/system.h>
  28 #include <asm/processor.h>
  29 #include <asm/i387.h>
  30 #include <asm/debugreg.h>
  31 #include <asm/ldt.h>
  32 #include <asm/desc.h>
  33 #include <asm/prctl.h>
  34 #include <asm/proto.h>
  35 #include <asm/ds.h>
  36
  37 #include "tls.h"
  38
  39 enum x86_regset {
  40         REGSET_GENERAL,
  41         REGSET_FP,
  42         REGSET_XFP,
  43         REGSET_IOPERM64 = REGSET_XFP,
  44         REGSET_TLS,
  45         REGSET_IOPERM32,
  46 };
  47
  48 /*
  49  * does not yet catch signals sent when the child dies.
  50  * in exit.c or in signal.c.
  51  */
  52
  53 /*
  54  * Determines which flags the user has access to [1 = access, 0 = no access].
  55  */
  56 #define FLAG_MASK_32            ((unsigned long)                        \
  57                                  (X86_EFLAGS_CF | X86_EFLAGS_PF |       \
  58                                   X86_EFLAGS_AF | X86_EFLAGS_ZF |       \
  59                                   X86_EFLAGS_SF | X86_EFLAGS_TF |       \
  60                                   X86_EFLAGS_DF | X86_EFLAGS_OF |       \
  61                                   X86_EFLAGS_RF | X86_EFLAGS_AC))
  62
  63 /*
  64  * Determines whether a value may be installed in a segment register.
  65  */
  66 static inline bool invalid_selector(u16 value)
  67 {
  68         return unlikely(value != 0 && (value & SEGMENT_RPL_MASK) != USER_RPL);
  69 }
  70
  71 #ifdef CONFIG_X86_32
  72
  73 #define FLAG_MASK               FLAG_MASK_32
  74
  75 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long regno)
  76 {
  77         BUILD_BUG_ON(offsetof(struct pt_regs, bx) != 0);
  78         regno >>= 2;
  79         if (regno > FS)
  80                 --regno;
  81         return &regs->bx + regno;
  82 }
  83
  84 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
  85 {
  86         /*
  87          * Returning the value truncates it to 16 bits.
  88          */
  89         unsigned int retval;
  90         if (offset != offsetof(struct user_regs_struct, gs))
  91                 retval = *pt_regs_access(task_pt_regs(task), offset);
  92         else {
  93                 retval = task->thread.gs;
  94                 if (task == current)
  95                         savesegment(gs, retval);
  96         }
  97         return retval;
  98 }
  99
 100 static int set_segment_reg(struct task_struct *task,
 101                            unsigned long offset, u16 value)
 102 {
 103         /*
 104          * The value argument was already truncated to 16 bits.
 105          */
 106         if (invalid_selector(value))
 107                 return -EIO;
 108
 109         /*
 110          * For %cs and %ss we cannot permit a null selector.
 111          * We can permit a bogus selector as long as it has USER_RPL.
 112          * Null selectors are fine for other segment registers, but
 113          * we will never get back to user mode with invalid %cs or %ss
 114          * and will take the trap in iret instead.  Much code relies
 115          * on user_mode() to distinguish a user trap frame (which can
 116          * safely use invalid selectors) from a kernel trap frame.
 117          */
 118         switch (offset) {
 119         case offsetof(struct user_regs_struct, cs):
 120         case offsetof(struct user_regs_struct, ss):
 121                 if (unlikely(value == 0))
 122                         return -EIO;
 123
 124         default:
 125                 *pt_regs_access(task_pt_regs(task), offset) = value;
 126                 break;
 127
 128         case offsetof(struct user_regs_struct, gs):
 129                 task->thread.gs = value;
 130                 if (task == current)
 131                         /*
 132                          * The user-mode %gs is not affected by
 133                          * kernel entry, so we must update the CPU.
 134                          */
 135                         loadsegment(gs, value);
 136         }
 137
 138         return 0;
 139 }
 140
 141 static unsigned long debugreg_addr_limit(struct task_struct *task)
 142 {
 143         return TASK_SIZE - 3;
 144 }
 145
 146 #else  /* CONFIG_X86_64 */
 147
 148 #define FLAG_MASK               (FLAG_MASK_32 | X86_EFLAGS_NT)
 149
 150 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long offset)
 151 {
 152         BUILD_BUG_ON(offsetof(struct pt_regs, r15) != 0);
 153         return &regs->r15 + (offset / sizeof(regs->r15));
 154 }
 155
 156 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
 157 {
 158         /*
 159          * Returning the value truncates it to 16 bits.
 160          */
 161         unsigned int seg;
 162
 163         switch (offset) {
 164         case offsetof(struct user_regs_struct, fs):
 165                 if (task == current) {
 166                         /* Older gas can't assemble movq %?s,%r?? */
 167                         asm("movl %%fs,%0" : "=r" (seg));
 168                         return seg;
 169                 }
 170                 return task->thread.fsindex;
 171         case offsetof(struct user_regs_struct, gs):
 172                 if (task == current) {
 173                         asm("movl %%gs,%0" : "=r" (seg));
 174                         return seg;
 175                 }
 176                 return task->thread.gsindex;
 177         case offsetof(struct user_regs_struct, ds):
 178                 if (task == current) {
 179                         asm("movl %%ds,%0" : "=r" (seg));
 180                         return seg;
 181                 }
 182                 return task->thread.ds;
 183         case offsetof(struct user_regs_struct, es):
 184                 if (task == current) {
 185                         asm("movl %%es,%0" : "=r" (seg));
 186                         return seg;
 187                 }
 188                 return task->thread.es;
 189
 190         case offsetof(struct user_regs_struct, cs):
 191         case offsetof(struct user_regs_struct, ss):
 192                 break;
 193         }
 194         return *pt_regs_access(task_pt_regs(task), offset);
 195 }
 196
 197 static int set_segment_reg(struct task_struct *task,
 198                            unsigned long offset, u16 value)
 199 {
 200         /*
 201          * The value argument was already truncated to 16 bits.
 202          */
 203         if (invalid_selector(value))
 204                 return -EIO;
 205
 206         switch (offset) {
 207         case offsetof(struct user_regs_struct,fs):
 208                 /*
 209                  * If this is setting fs as for normal 64-bit use but
 210                  * setting fs_base has implicitly changed it, leave it.
 211                  */
 212                 if ((value == FS_TLS_SEL && task->thread.fsindex == 0 &&
 213                      task->thread.fs != 0) ||
 214                     (value == 0 && task->thread.fsindex == FS_TLS_SEL &&
 215                      task->thread.fs == 0))
 216                         break;
 217                 task->thread.fsindex = value;
 218                 if (task == current)
 219                         loadsegment(fs, task->thread.fsindex);
 220                 break;
 221         case offsetof(struct user_regs_struct,gs):
 222                 /*
 223                  * If this is setting gs as for normal 64-bit use but
 224                  * setting gs_base has implicitly changed it, leave it.
 225                  */
 226                 if ((value == GS_TLS_SEL && task->thread.gsindex == 0 &&
 227                      task->thread.gs != 0) ||
 228                     (value == 0 && task->thread.gsindex == GS_TLS_SEL &&
 229                      task->thread.gs == 0))
 230                         break;
 231                 task->thread.gsindex = value;
 232                 if (task == current)
 233                         load_gs_index(task->thread.gsindex);
 234                 break;
 235         case offsetof(struct user_regs_struct,ds):
 236                 task->thread.ds = value;
 237                 if (task == current)
 238                         loadsegment(ds, task->thread.ds);
 239                 break;
 240         case offsetof(struct user_regs_struct,es):
 241                 task->thread.es = value;
 242                 if (task == current)
 243                         loadsegment(es, task->thread.es);
 244                 break;
 245
 246                 /*
 247                  * Can't actually change these in 64-bit mode.
 248                  */
 249         case offsetof(struct user_regs_struct,cs):
 250                 if (unlikely(value == 0))
 251                         return -EIO;
 252 #ifdef CONFIG_IA32_EMULATION
 253                 if (test_tsk_thread_flag(task, TIF_IA32))
 254                         task_pt_regs(task)->cs = value;
 255 #endif
 256                 break;
 257         case offsetof(struct user_regs_struct,ss):
 258                 if (unlikely(value == 0))
 259                         return -EIO;
 260 #ifdef CONFIG_IA32_EMULATION
 261                 if (test_tsk_thread_flag(task, TIF_IA32))
 262                         task_pt_regs(task)->ss = value;
 263 #endif
 264                 break;
 265         }
 266
 267         return 0;
 268 }
 269
 270 static unsigned long debugreg_addr_limit(struct task_struct *task)
 271 {
 272 #ifdef CONFIG_IA32_EMULATION
 273         if (test_tsk_thread_flag(task, TIF_IA32))
 274                 return IA32_PAGE_OFFSET - 3;
 275 #endif
 276         return TASK_SIZE64 - 7;
 277 }
 278
 279 #endif  /* CONFIG_X86_32 */
 280
 281 static unsigned long get_flags(struct task_struct *task)
 282 {
 283         unsigned long retval = task_pt_regs(task)->flags;
 284
 285         /*
 286          * If the debugger set TF, hide it from the readout.
 287          */
 288         if (test_tsk_thread_flag(task, TIF_FORCED_TF))
 289                 retval &= ~X86_EFLAGS_TF;
 290
 291         return retval;
 292 }
 293
 294 static int set_flags(struct task_struct *task, unsigned long value)
 295 {
 296         struct pt_regs *regs = task_pt_regs(task);
 297
 298         /*
 299          * If the user value contains TF, mark that
 300          * it was not "us" (the debugger) that set it.
 301          * If not, make sure it stays set if we had.
 302          */
 303         if (value & X86_EFLAGS_TF)
 304                 clear_tsk_thread_flag(task, TIF_FORCED_TF);
 305         else if (test_tsk_thread_flag(task, TIF_FORCED_TF))
 306                 value |= X86_EFLAGS_TF;
 307
 308         regs->flags = (regs->flags & ~FLAG_MASK) | (value & FLAG_MASK);
 309
 310         return 0;
 311 }
 312
 313 static int putreg(struct task_struct *child,
 314                   unsigned long offset, unsigned long value)
 315 {
 316         switch (offset) {
 317         case offsetof(struct user_regs_struct, cs):
 318         case offsetof(struct user_regs_struct, ds):
 319         case offsetof(struct user_regs_struct, es):
 320         case offsetof(struct user_regs_struct, fs):
 321         case offsetof(struct user_regs_struct, gs):
 322         case offsetof(struct user_regs_struct, ss):
 323                 return set_segment_reg(child, offset, value);
 324
 325         case offsetof(struct user_regs_struct, flags):
 326                 return set_flags(child, value);
 327
 328 #ifdef CONFIG_X86_64
 329         /*
 330          * Orig_ax is really just a flag with small positive and
 331          * negative values, so make sure to always sign-extend it
 332          * from 32 bits so that it works correctly regardless of
 333          * whether we come from a 32-bit environment or not.
 334          */
 335         case offsetof(struct user_regs_struct, orig_ax):
 336                 value = (long) (s32) value;
 337                 break;
 338
 339         case offsetof(struct user_regs_struct,fs_base):
 340                 if (value >= TASK_SIZE_OF(child))
 341                         return -EIO;
 342                 /*
 343                  * When changing the segment base, use do_arch_prctl
 344                  * to set either thread.fs or thread.fsindex and the
 345                  * corresponding GDT slot.
 346                  */
 347                 if (child->thread.fs != value)
 348                         return do_arch_prctl(child, ARCH_SET_FS, value);
 349                 return 0;
 350         case offsetof(struct user_regs_struct,gs_base):
 351                 /*
 352                  * Exactly the same here as the %fs handling above.
 353                  */
 354                 if (value >= TASK_SIZE_OF(child))
 355                         return -EIO;
 356                 if (child->thread.gs != value)
 357                         return do_arch_prctl(child, ARCH_SET_GS, value);
 358                 return 0;
 359 #endif
 360         }
 361
 362         *pt_regs_access(task_pt_regs(child), offset) = value;
 363         return 0;
 364 }
 365
 366 static unsigned long getreg(struct task_struct *task, unsigned long offset)
 367 {
 368         switch (offset) {
 369         case offsetof(struct user_regs_struct, cs):
 370         case offsetof(struct user_regs_struct, ds):
 371         case offsetof(struct user_regs_struct, es):
 372         case offsetof(struct user_regs_struct, fs):
 373         case offsetof(struct user_regs_struct, gs):
 374         case offsetof(struct user_regs_struct, ss):
 375                 return get_segment_reg(task, offset);
 376
 377         case offsetof(struct user_regs_struct, flags):
 378                 return get_flags(task);
 379
 380 #ifdef CONFIG_X86_64
 381         case offsetof(struct user_regs_struct, fs_base): {
 382                 /*
 383                  * do_arch_prctl may have used a GDT slot instead of
 384                  * the MSR.  To userland, it appears the same either
 385                  * way, except the %fs segment selector might not be 0.
 386                  */
 387                 unsigned int seg = task->thread.fsindex;
 388                 if (task->thread.fs != 0)
 389                         return task->thread.fs;
 390                 if (task == current)
 391                         asm("movl %%fs,%0" : "=r" (seg));
 392                 if (seg != FS_TLS_SEL)
 393                         return 0;
 394                 return get_desc_base(&task->thread.tls_array[FS_TLS]);
 395         }
 396         case offsetof(struct user_regs_struct, gs_base): {
 397                 /*
 398                  * Exactly the same here as the %fs handling above.
 399                  */
 400                 unsigned int seg = task->thread.gsindex;
 401                 if (task->thread.gs != 0)
 402                         return task->thread.gs;
 403                 if (task == current)
 404                         asm("movl %%gs,%0" : "=r" (seg));
 405                 if (seg != GS_TLS_SEL)
 406                         return 0;
 407                 return get_desc_base(&task->thread.tls_array[GS_TLS]);
 408         }
 409 #endif
 410         }
 411
 412         return *pt_regs_access(task_pt_regs(task), offset);
 413 }
 414
 415 static int genregs_get(struct task_struct *target,
 416                        const struct user_regset *regset,
 417                        unsigned int pos, unsigned int count,
 418                        void *kbuf, void __user *ubuf)
 419 {
 420         if (kbuf) {
 421                 unsigned long *k = kbuf;
 422                 while (count > 0) {
 423                         *k++ = getreg(target, pos);
 424                         count -= sizeof(*k);
 425                         pos += sizeof(*k);
 426                 }
 427         } else {
 428                 unsigned long __user *u = ubuf;
 429                 while (count > 0) {
 430                         if (__put_user(getreg(target, pos), u++))
 431                                 return -EFAULT;
 432                         count -= sizeof(*u);
 433                         pos += sizeof(*u);
 434                 }
 435         }
 436
 437         return 0;
 438 }
 439
 440 static int genregs_set(struct task_struct *target,
 441                        const struct user_regset *regset,
 442                        unsigned int pos, unsigned int count,
 443                        const void *kbuf, const void __user *ubuf)
 444 {
 445         int ret = 0;
 446         if (kbuf) {
 447                 const unsigned long *k = kbuf;
 448                 while (count > 0 && !ret) {
 449                         ret = putreg(target, pos, *k++);
 450                         count -= sizeof(*k);
 451                         pos += sizeof(*k);
 452                 }
 453         } else {
 454                 const unsigned long  __user *u = ubuf;
 455                 while (count > 0 && !ret) {
 456                         unsigned long word;
 457                         ret = __get_user(word, u++);
 458                         if (ret)
 459                                 break;
 460                         ret = putreg(target, pos, word);
 461                         count -= sizeof(*u);
 462                         pos += sizeof(*u);
 463                 }
 464         }
 465         return ret;
 466 }
 467
 468 /*
 469  * This function is trivial and will be inlined by the compiler.
 470  * Having it separates the implementation details of debug
 471  * registers from the interface details of ptrace.
 472  */
 473 static unsigned long ptrace_get_debugreg(struct task_struct *child, int n)
 474 {
 475         switch (n) {
 476         case 0:         return child->thread.debugreg0;
 477         case 1:         return child->thread.debugreg1;
 478         case 2:         return child->thread.debugreg2;
 479         case 3:         return child->thread.debugreg3;
 480         case 6:         return child->thread.debugreg6;
 481         case 7:         return child->thread.debugreg7;
 482         }
 483         return 0;
 484 }
 485
 486 static int ptrace_set_debugreg(struct task_struct *child,
 487                                int n, unsigned long data)
 488 {
 489         int i;
 490
 491         if (unlikely(n == 4 || n == 5))
 492                 return -EIO;
 493
 494         if (n < 4 && unlikely(data >= debugreg_addr_limit(child)))
 495                 return -EIO;
 496
 497         switch (n) {
 498         case 0:         child->thread.debugreg0 = data; break;
 499         case 1:         child->thread.debugreg1 = data; break;
 500         case 2:         child->thread.debugreg2 = data; break;
 501         case 3:         child->thread.debugreg3 = data; break;
 502
 503         case 6:
 504                 if ((data & ~0xffffffffUL) != 0)
 505                         return -EIO;
 506                 child->thread.debugreg6 = data;
 507                 break;
 508
 509         case 7:
 510                 /*
 511                  * Sanity-check data. Take one half-byte at once with
 512                  * check = (val >> (16 + 4*i)) & 0xf. It contains the
 513                  * R/Wi and LENi bits; bits 0 and 1 are R/Wi, and bits
 514                  * 2 and 3 are LENi. Given a list of invalid values,
 515                  * we do mask |= 1 << invalid_value, so that
 516                  * (mask >> check) & 1 is a correct test for invalid
 517                  * values.
 518                  *
 519                  * R/Wi contains the type of the breakpoint /
 520                  * watchpoint, LENi contains the length of the watched
 521                  * data in the watchpoint case.
 522                  *
 523                  * The invalid values are:
 524                  * - LENi == 0x10 (undefined), so mask |= 0x0f00.       [32-bit]
 525                  * - R/Wi == 0x10 (break on I/O reads or writes), so
 526                  *   mask |= 0x4444.
 527                  * - R/Wi == 0x00 && LENi != 0x00, so we have mask |=
 528                  *   0x1110.
 529                  *
 530                  * Finally, mask = 0x0f00 | 0x4444 | 0x1110 == 0x5f54.
 531                  *
 532                  * See the Intel Manual "System Programming Guide",
 533                  * 15.2.4
 534                  *
 535                  * Note that LENi == 0x10 is defined on x86_64 in long
 536                  * mode (i.e. even for 32-bit userspace software, but
 537                  * 64-bit kernel), so the x86_64 mask value is 0x5454.
 538                  * See the AMD manual no. 24593 (AMD64 System Programming)
 539                  */
 540 #ifdef CONFIG_X86_32
 541 #define DR7_MASK        0x5f54
 542 #else
 543 #define DR7_MASK        0x5554
 544 #endif
 545                 data &= ~DR_CONTROL_RESERVED;
 546                 for (i = 0; i < 4; i++)
 547                         if ((DR7_MASK >> ((data >> (16 + 4*i)) & 0xf)) & 1)
 548                                 return -EIO;
 549                 child->thread.debugreg7 = data;
 550                 if (data)
 551                         set_tsk_thread_flag(child, TIF_DEBUG);
 552                 else
 553                         clear_tsk_thread_flag(child, TIF_DEBUG);
 554                 break;
 555         }
 556
 557         return 0;
 558 }
 559
 560 /*
 561  * These access the current or another (stopped) task's io permission
 562  * bitmap for debugging or core dump.
 563  */
 564 static int ioperm_active(struct task_struct *target,
 565                          const struct user_regset *regset)
 566 {
 567         return target->thread.io_bitmap_max / regset->size;
 568 }
 569
 570 static int ioperm_get(struct task_struct *target,
 571                       const struct user_regset *regset,
 572                       unsigned int pos, unsigned int count,
 573                       void *kbuf, void __user *ubuf)
 574 {
 575         if (!target->thread.io_bitmap_ptr)
 576                 return -ENXIO;
 577
 578         return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
 579                                    target->thread.io_bitmap_ptr,
 580                                    0, IO_BITMAP_BYTES);
 581 }
 582
 583 #ifdef CONFIG_X86_PTRACE_BTS
 584 /*
 585  * The configuration for a particular BTS hardware implementation.
 586  */
 587 struct bts_configuration {
 588         /* the size of a BTS record in bytes; at most BTS_MAX_RECORD_SIZE */
 589         unsigned char  sizeof_bts;
 590         /* the size of a field in the BTS record in bytes */
 591         unsigned char  sizeof_field;
 592         /* a bitmask to enable/disable BTS in DEBUGCTL MSR */
 593         unsigned long debugctl_mask;
 594 };
 595 static struct bts_configuration bts_cfg;
 596
 597 #define BTS_MAX_RECORD_SIZE (8 * 3)
 598
 599
 600 /*
 601  * Branch Trace Store (BTS) uses the following format. Different
 602  * architectures vary in the size of those fields.
 603  * - source linear address
 604  * - destination linear address
 605  * - flags
 606  *
 607  * Later architectures use 64bit pointers throughout, whereas earlier
 608  * architectures use 32bit pointers in 32bit mode.
 609  *
 610  * We compute the base address for the first 8 fields based on:
 611  * - the field size stored in the DS configuration
 612  * - the relative field position
 613  *
 614  * In order to store additional information in the BTS buffer, we use
 615  * a special source address to indicate that the record requires
 616  * special interpretation.
 617  *
 618  * Netburst indicated via a bit in the flags field whether the branch
 619  * was predicted; this is ignored.
 620  */
 621
 622 enum bts_field {
 623         bts_from = 0,
 624         bts_to,
 625         bts_flags,
 626
 627         bts_escape = (unsigned long)-1,
 628         bts_qual = bts_to,
 629         bts_jiffies = bts_flags
 630 };
 631
 632 static inline unsigned long bts_get(const char *base, enum bts_field field)
 633 {
 634         base += (bts_cfg.sizeof_field * field);
 635         return *(unsigned long *)base;
 636 }
 637
 638 static inline void bts_set(char *base, enum bts_field field, unsigned long val)
 639 {
 640         base += (bts_cfg.sizeof_field * field);;
 641         (*(unsigned long *)base) = val;
 642 }
 643
 644 /*
 645  * Translate a BTS record from the raw format into the bts_struct format
 646  *
 647  * out (out): bts_struct interpretation
 648  * raw: raw BTS record
 649  */
 650 static void ptrace_bts_translate_record(struct bts_struct *out, const void *raw)
 651 {
 652         memset(out, 0, sizeof(*out));
 653         if (bts_get(raw, bts_from) == bts_escape) {
 654                 out->qualifier       = bts_get(raw, bts_qual);
 655                 out->variant.jiffies = bts_get(raw, bts_jiffies);
 656         } else {
 657                 out->qualifier = BTS_BRANCH;
 658                 out->variant.lbr.from_ip = bts_get(raw, bts_from);
 659                 out->variant.lbr.to_ip   = bts_get(raw, bts_to);
 660         }
 661 }
 662
 663 static int ptrace_bts_read_record(struct task_struct *child, size_t index,
 664                                   struct bts_struct __user *out)
 665 {
 666         struct bts_struct ret;
 667         const void *bts_record;
 668         size_t bts_index, bts_end;
 669         int error;
 670
 671         error = ds_get_bts_end(child, &bts_end);
 672         if (error < 0)
 673                 return error;
 674
 675         if (bts_end <= index)
 676                 return -EINVAL;
 677
 678         error = ds_get_bts_index(child, &bts_index);
 679         if (error < 0)
 680                 return error;
 681
 682         /* translate the ptrace bts index into the ds bts index */
 683         bts_index += bts_end - (index + 1);
 684         if (bts_end <= bts_index)
 685                 bts_index -= bts_end;
 686
 687         error = ds_access_bts(child, bts_index, &bts_record);
 688         if (error < 0)
 689                 return error;
 690
 691         ptrace_bts_translate_record(&ret, bts_record);
 692
 693         if (copy_to_user(out, &ret, sizeof(ret)))
 694                 return -EFAULT;
 695
 696         return sizeof(ret);
 697 }
 698
 699 static int ptrace_bts_drain(struct task_struct *child,
 700                             long size,
 701                             struct bts_struct __user *out)
 702 {
 703         struct bts_struct ret;
 704         const unsigned char *raw;
 705         size_t end, i;
 706         int error;
 707
 708         error = ds_get_bts_index(child, &end);
 709         if (error < 0)
 710                 return error;
 711
 712         if (size < (end * sizeof(struct bts_struct)))
 713                 return -EIO;
 714
 715         error = ds_access_bts(child, 0, (const void **)&raw);
 716         if (error < 0)
 717                 return error;
 718
 719         for (i = 0; i < end; i++, out++, raw += bts_cfg.sizeof_bts) {
 720                 ptrace_bts_translate_record(&ret, raw);
 721
 722                 if (copy_to_user(out, &ret, sizeof(ret)))
 723                         return -EFAULT;
 724         }
 725
 726         error = ds_clear_bts(child);
 727         if (error < 0)
 728                 return error;
 729
 730         return end;
 731 }
 732
 733 static void ptrace_bts_ovfl(struct task_struct *child)
 734 {
 735         send_sig(child->thread.bts_ovfl_signal, child, 0);
 736 }
 737
 738 static int ptrace_bts_config(struct task_struct *child,
 739                              long cfg_size,
 740                              const struct ptrace_bts_config __user *ucfg)
 741 {
 742         struct ptrace_bts_config cfg;
 743         int error = 0;
 744
 745         error = -EOPNOTSUPP;
 746         if (!bts_cfg.sizeof_bts)
 747                 goto errout;
 748
 749         error = -EIO;
 750         if (cfg_size < sizeof(cfg))
 751                 goto errout;
 752
 753         error = -EFAULT;
 754         if (copy_from_user(&cfg, ucfg, sizeof(cfg)))
 755                 goto errout;
 756
 757         error = -EINVAL;
 758         if ((cfg.flags & PTRACE_BTS_O_SIGNAL) &&
 759             !(cfg.flags & PTRACE_BTS_O_ALLOC))
 760                 goto errout;
 761
 762         if (cfg.flags & PTRACE_BTS_O_ALLOC) {
 763                 ds_ovfl_callback_t ovfl = NULL;
 764                 unsigned int sig = 0;
 765
 766                 /* we ignore the error in case we were not tracing child */
 767                 (void)ds_release_bts(child);
 768
 769                 if (cfg.flags & PTRACE_BTS_O_SIGNAL) {
 770                         if (!cfg.signal)
 771                                 goto errout;
 772
 773                         sig  = cfg.signal;
 774                         ovfl = ptrace_bts_ovfl;
 775                 }
 776
 777                 error = ds_request_bts(child, /* base = */ NULL, cfg.size, ovfl);
 778                 if (error < 0)
 779                         goto errout;
 780
 781                 child->thread.bts_ovfl_signal = sig;
 782         }
 783
 784         error = -EINVAL;
 785         if (!child->thread.ds_ctx && cfg.flags)
 786                 goto errout;
 787
 788         if (cfg.flags & PTRACE_BTS_O_TRACE)
 789                 child->thread.debugctlmsr |= bts_cfg.debugctl_mask;
 790         else
 791                 child->thread.debugctlmsr &= ~bts_cfg.debugctl_mask;
 792
 793         if (cfg.flags & PTRACE_BTS_O_SCHED)
 794                 set_tsk_thread_flag(child, TIF_BTS_TRACE_TS);
 795         else
 796                 clear_tsk_thread_flag(child, TIF_BTS_TRACE_TS);
 797
 798         error = sizeof(cfg);
 799
 800 out:
 801         if (child->thread.debugctlmsr)
 802                 set_tsk_thread_flag(child, TIF_DEBUGCTLMSR);
 803         else
 804                 clear_tsk_thread_flag(child, TIF_DEBUGCTLMSR);
 805
 806         return error;
 807
 808 errout:
 809         child->thread.debugctlmsr &= ~bts_cfg.debugctl_mask;
 810         clear_tsk_thread_flag(child, TIF_BTS_TRACE_TS);
 811         goto out;
 812 }
 813
 814 static int ptrace_bts_status(struct task_struct *child,
 815                              long cfg_size,
 816                              struct ptrace_bts_config __user *ucfg)
 817 {
 818         struct ptrace_bts_config cfg;
 819         size_t end;
 820         const void *base, *max;
 821         int error;
 822
 823         if (cfg_size < sizeof(cfg))
 824                 return -EIO;
 825
 826         error = ds_get_bts_end(child, &end);
 827         if (error < 0)
 828                 return error;
 829
 830         error = ds_access_bts(child, /* index = */ 0, &base);
 831         if (error < 0)
 832                 return error;
 833
 834         error = ds_access_bts(child, /* index = */ end, &max);
 835         if (error < 0)
 836                 return error;
 837
 838         memset(&cfg, 0, sizeof(cfg));
 839         cfg.size = (max - base);
 840         cfg.signal = child->thread.bts_ovfl_signal;
 841         cfg.bts_size = sizeof(struct bts_struct);
 842
 843         if (cfg.signal)
 844                 cfg.flags |= PTRACE_BTS_O_SIGNAL;
 845
 846         if (test_tsk_thread_flag(child, TIF_DEBUGCTLMSR) &&
 847             child->thread.debugctlmsr & bts_cfg.debugctl_mask)
 848                 cfg.flags |= PTRACE_BTS_O_TRACE;
 849
 850         if (test_tsk_thread_flag(child, TIF_BTS_TRACE_TS))
 851                 cfg.flags |= PTRACE_BTS_O_SCHED;
 852
 853         if (copy_to_user(ucfg, &cfg, sizeof(cfg)))
 854                 return -EFAULT;
 855
 856         return sizeof(cfg);
 857 }
 858
 859 static int ptrace_bts_write_record(struct task_struct *child,
 860                                    const struct bts_struct *in)
 861 {
 862         unsigned char bts_record[BTS_MAX_RECORD_SIZE];
 863
 864         BUG_ON(BTS_MAX_RECORD_SIZE < bts_cfg.sizeof_bts);
 865
 866         memset(bts_record, 0, bts_cfg.sizeof_bts);
 867         switch (in->qualifier) {
 868         case BTS_INVALID:
 869                 break;
 870
 871         case BTS_BRANCH:
 872                 bts_set(bts_record, bts_from, in->variant.lbr.from_ip);
 873                 bts_set(bts_record, bts_to,   in->variant.lbr.to_ip);
 874                 break;
 875
 876         case BTS_TASK_ARRIVES:
 877         case BTS_TASK_DEPARTS:
 878                 bts_set(bts_record, bts_from,    bts_escape);
 879                 bts_set(bts_record, bts_qual,    in->qualifier);
 880                 bts_set(bts_record, bts_jiffies, in->variant.jiffies);
 881                 break;
 882
 883         default:
 884                 return -EINVAL;
 885         }
 886
 887         /* The writing task will be the switched-to task on a context
 888          * switch. It needs to write into the switched-from task's BTS
 889          * buffer. */
 890         return ds_unchecked_write_bts(child, bts_record, bts_cfg.sizeof_bts);
 891 }
 892
 893 void ptrace_bts_take_timestamp(struct task_struct *tsk,
 894                                enum bts_qualifier qualifier)
 895 {
 896         struct bts_struct rec = {
 897                 .qualifier = qualifier,
 898                 .variant.jiffies = jiffies_64
 899         };
 900
 901         ptrace_bts_write_record(tsk, &rec);
 902 }
 903
 904 static const struct bts_configuration bts_cfg_netburst = {
 905         .sizeof_bts    = sizeof(long) * 3,
 906         .sizeof_field  = sizeof(long),
 907         .debugctl_mask = (1<<2)|(1<<3)|(1<<5)
 908 };
 909
 910 static const struct bts_configuration bts_cfg_pentium_m = {
 911         .sizeof_bts    = sizeof(long) * 3,
 912         .sizeof_field  = sizeof(long),
 913         .debugctl_mask = (1<<6)|(1<<7)
 914 };
 915
 916 static const struct bts_configuration bts_cfg_core2 = {
 917         .sizeof_bts    = 8 * 3,
 918         .sizeof_field  = 8,
 919         .debugctl_mask = (1<<6)|(1<<7)|(1<<9)
 920 };
 921
 922 static inline void bts_configure(const struct bts_configuration *cfg)
 923 {
 924         bts_cfg = *cfg;
 925 }
 926
 927 void __cpuinit ptrace_bts_init_intel(struct cpuinfo_x86 *c)
 928 {
 929         switch (c->x86) {
 930         case 0x6:
 931                 switch (c->x86_model) {
 932                 case 0xD:
 933                 case 0xE: /* Pentium M */
 934                         bts_configure(&bts_cfg_pentium_m);
 935                         break;
 936                 case 0xF: /* Core2 */
 937         case 0x1C: /* Atom */
 938                         bts_configure(&bts_cfg_core2);
 939                         break;
 940                 default:
 941                         /* sorry, don't know about them */
 942                         break;
 943                 }
 944                 break;
 945         case 0xF:
 946                 switch (c->x86_model) {
 947                 case 0x0:
 948                 case 0x1:
 949                 case 0x2: /* Netburst */
 950                         bts_configure(&bts_cfg_netburst);
 951                         break;
 952                 default:
 953                         /* sorry, don't know about them */
 954                         break;
 955                 }
 956                 break;
 957         default:
 958                 /* sorry, don't know about them */
 959                 break;
 960         }
 961 }
 962 #endif /* CONFIG_X86_PTRACE_BTS */
 963
 964 /*
 965  * Called by kernel/ptrace.c when detaching..
 966  *
 967  * Make sure the single step bit is not set.
 968  */
 969 void ptrace_disable(struct task_struct *child)
 970 {
 971         user_disable_single_step(child);
 972 #ifdef TIF_SYSCALL_EMU
 973         clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
 974 #endif
 975 #ifdef CONFIG_X86_PTRACE_BTS
 976         (void)ds_release_bts(child);
 977
 978         child->thread.debugctlmsr &= ~bts_cfg.debugctl_mask;
 979         if (!child->thread.debugctlmsr)
 980                 clear_tsk_thread_flag(child, TIF_DEBUGCTLMSR);
 981
 982         clear_tsk_thread_flag(child, TIF_BTS_TRACE_TS);
 983 #endif /* CONFIG_X86_PTRACE_BTS */
 984 }
 985
 986 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
 987 static const struct user_regset_view user_x86_32_view; /* Initialized below. */
 988 #endif
 989
 990 long arch_ptrace(struct task_struct *child, long request, long addr, long data)
 991 {
 992         int ret;
 993         unsigned long __user *datap = (unsigned long __user *)data;
 994
 995         switch (request) {
 996         /* read the word at location addr in the USER area. */
 997         case PTRACE_PEEKUSR: {
 998                 unsigned long tmp;
 999
1000                 ret = -EIO;
1001                 if ((addr & (sizeof(data) - 1)) || addr < 0 ||
1002                     addr >= sizeof(struct user))
1003                         break;
1004
1005                 tmp = 0;  /* Default return condition */
1006                 if (addr < sizeof(struct user_regs_struct))
1007                         tmp = getreg(child, addr);
1008                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
1009                          addr <= offsetof(struct user, u_debugreg[7])) {
1010                         addr -= offsetof(struct user, u_debugreg[0]);
1011                         tmp = ptrace_get_debugreg(child, addr / sizeof(data));
1012                 }
1013                 ret = put_user(tmp, datap);
1014                 break;
1015         }
1016
1017         case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
1018                 ret = -EIO;
1019                 if ((addr & (sizeof(data) - 1)) || addr < 0 ||
1020                     addr >= sizeof(struct user))
1021                         break;
1022
1023                 if (addr < sizeof(struct user_regs_struct))
1024                         ret = putreg(child, addr, data);
1025                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
1026                          addr <= offsetof(struct user, u_debugreg[7])) {
1027                         addr -= offsetof(struct user, u_debugreg[0]);
1028                         ret = ptrace_set_debugreg(child,
1029                                                   addr / sizeof(data), data);
1030                 }
1031                 break;
1032
1033         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
1034                 return copy_regset_to_user(child,
1035                                            task_user_regset_view(current),
1036                                            REGSET_GENERAL,
1037                                            0, sizeof(struct user_regs_struct),
1038                                            datap);
1039
1040         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
1041                 return copy_regset_from_user(child,
1042                                              task_user_regset_view(current),
1043                                              REGSET_GENERAL,
1044                                              0, sizeof(struct user_regs_struct),
1045                                              datap);
1046
1047         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
1048                 return copy_regset_to_user(child,
1049                                            task_user_regset_view(current),
1050                                            REGSET_FP,
1051                                            0, sizeof(struct user_i387_struct),
1052                                            datap);
1053
1054         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
1055                 return copy_regset_from_user(child,
1056                                              task_user_regset_view(current),
1057                                              REGSET_FP,
1058                                              0, sizeof(struct user_i387_struct),
1059                                              datap);
1060
1061 #ifdef CONFIG_X86_32
1062         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
1063                 return copy_regset_to_user(child, &user_x86_32_view,
1064                                            REGSET_XFP,
1065                                            0, sizeof(struct user_fxsr_struct),
1066                                            datap) ? -EIO : 0;
1067
1068         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
1069                 return copy_regset_from_user(child, &user_x86_32_view,
1070                                              REGSET_XFP,
1071                                              0, sizeof(struct user_fxsr_struct),
1072                                              datap) ? -EIO : 0;
1073 #endif
1074
1075 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1076         case PTRACE_GET_THREAD_AREA:
1077                 if (addr < 0)
1078                         return -EIO;
1079                 ret = do_get_thread_area(child, addr,
1080                                          (struct user_desc __user *) data);
1081                 break;
1082
1083         case PTRACE_SET_THREAD_AREA:
1084                 if (addr < 0)
1085                         return -EIO;
1086                 ret = do_set_thread_area(child, addr,
1087                                          (struct user_desc __user *) data, 0);
1088                 break;
1089 #endif
1090
1091 #ifdef CONFIG_X86_64
1092                 /* normal 64bit interface to access TLS data.
1093                    Works just like arch_prctl, except that the arguments
1094                    are reversed. */
1095         case PTRACE_ARCH_PRCTL:
1096                 ret = do_arch_prctl(child, data, addr);
1097                 break;
1098 #endif
1099
1100         /*
1101          * These bits need more cooking - not enabled yet:
1102          */
1103 #ifdef CONFIG_X86_PTRACE_BTS
1104         case PTRACE_BTS_CONFIG:
1105                 ret = ptrace_bts_config
1106                         (child, data, (struct ptrace_bts_config __user *)addr);
1107                 break;
1108
1109         case PTRACE_BTS_STATUS:
1110                 ret = ptrace_bts_status
1111                         (child, data, (struct ptrace_bts_config __user *)addr);
1112                 break;
1113
1114         case PTRACE_BTS_SIZE:
1115                 ret = ds_get_bts_index(child, /* pos = */ NULL);
1116                 break;
1117
1118         case PTRACE_BTS_GET:
1119                 ret = ptrace_bts_read_record
1120                         (child, data, (struct bts_struct __user *) addr);
1121                 break;
1122
1123         case PTRACE_BTS_CLEAR:
1124                 ret = ds_clear_bts(child);
1125                 break;
1126
1127         case PTRACE_BTS_DRAIN:
1128                 ret = ptrace_bts_drain
1129                         (child, data, (struct bts_struct __user *) addr);
1130                 break;
1131 #endif /* CONFIG_X86_PTRACE_BTS */
1132
1133         default:
1134                 ret = ptrace_request(child, request, addr, data);
1135                 break;
1136         }
1137
1138         return ret;
1139 }
1140
1141 #ifdef CONFIG_IA32_EMULATION
1142
1143 #include <linux/compat.h>
1144 #include <linux/syscalls.h>
1145 #include <asm/ia32.h>
1146 #include <asm/user32.h>
1147
1148 #define R32(l,q)                                                        \
1149         case offsetof(struct user32, regs.l):                           \
1150                 regs->q = value; break
1151
1152 #define SEG32(rs)                                                       \
1153         case offsetof(struct user32, regs.rs):                          \
1154                 return set_segment_reg(child,                           \
1155                                        offsetof(struct user_regs_struct, rs), \
1156                                        value);                          \
1157                 break
1158
1159 static int putreg32(struct task_struct *child, unsigned regno, u32 value)
1160 {
1161         struct pt_regs *regs = task_pt_regs(child);
1162
1163         switch (regno) {
1164
1165         SEG32(cs);
1166         SEG32(ds);
1167         SEG32(es);
1168         SEG32(fs);
1169         SEG32(gs);
1170         SEG32(ss);
1171
1172         R32(ebx, bx);
1173         R32(ecx, cx);
1174         R32(edx, dx);
1175         R32(edi, di);
1176         R32(esi, si);
1177         R32(ebp, bp);
1178         R32(eax, ax);
1179         R32(eip, ip);
1180         R32(esp, sp);
1181
1182         case offsetof(struct user32, regs.orig_eax):
1183                 /*
1184                  * Sign-extend the value so that orig_eax = -1
1185                  * causes (long)orig_ax < 0 tests to fire correctly.
1186                  */
1187                 regs->orig_ax = (long) (s32) value;
1188                 break;
1189
1190         case offsetof(struct user32, regs.eflags):
1191                 return set_flags(child, value);
1192
1193         case offsetof(struct user32, u_debugreg[0]) ...
1194                 offsetof(struct user32, u_debugreg[7]):
1195                 regno -= offsetof(struct user32, u_debugreg[0]);
1196                 return ptrace_set_debugreg(child, regno / 4, value);
1197
1198         default:
1199                 if (regno > sizeof(struct user32) || (regno & 3))
1200                         return -EIO;
1201
1202                 /*
1203                  * Other dummy fields in the virtual user structure
1204                  * are ignored
1205                  */
1206                 break;
1207         }
1208         return 0;
1209 }
1210
1211 #undef R32
1212 #undef SEG32
1213
1214 #define R32(l,q)                                                        \
1215         case offsetof(struct user32, regs.l):                           \
1216                 *val = regs->q; break
1217
1218 #define SEG32(rs)                                                       \
1219         case offsetof(struct user32, regs.rs):                          \
1220                 *val = get_segment_reg(child,                           \
1221                                        offsetof(struct user_regs_struct, rs)); \
1222                 break
1223
1224 static int getreg32(struct task_struct *child, unsigned regno, u32 *val)
1225 {
1226         struct pt_regs *regs = task_pt_regs(child);
1227
1228         switch (regno) {
1229
1230         SEG32(ds);
1231         SEG32(es);
1232         SEG32(fs);
1233         SEG32(gs);
1234
1235         R32(cs, cs);
1236         R32(ss, ss);
1237         R32(ebx, bx);
1238         R32(ecx, cx);
1239         R32(edx, dx);
1240         R32(edi, di);
1241         R32(esi, si);
1242         R32(ebp, bp);
1243         R32(eax, ax);
1244         R32(orig_eax, orig_ax);
1245         R32(eip, ip);
1246         R32(esp, sp);
1247
1248         case offsetof(struct user32, regs.eflags):
1249                 *val = get_flags(child);
1250                 break;
1251
1252         case offsetof(struct user32, u_debugreg[0]) ...
1253                 offsetof(struct user32, u_debugreg[7]):
1254                 regno -= offsetof(struct user32, u_debugreg[0]);
1255                 *val = ptrace_get_debugreg(child, regno / 4);
1256                 break;
1257
1258         default:
1259                 if (regno > sizeof(struct user32) || (regno & 3))
1260                         return -EIO;
1261
1262                 /*
1263                  * Other dummy fields in the virtual user structure
1264                  * are ignored
1265                  */
1266                 *val = 0;
1267                 break;
1268         }
1269         return 0;
1270 }
1271
1272 #undef R32
1273 #undef SEG32
1274
1275 static int genregs32_get(struct task_struct *target,
1276                          const struct user_regset *regset,
1277                          unsigned int pos, unsigned int count,
1278                          void *kbuf, void __user *ubuf)
1279 {
1280         if (kbuf) {
1281                 compat_ulong_t *k = kbuf;
1282                 while (count > 0) {
1283                         getreg32(target, pos, k++);
1284                         count -= sizeof(*k);
1285                         pos += sizeof(*k);
1286                 }
1287         } else {
1288                 compat_ulong_t __user *u = ubuf;
1289                 while (count > 0) {
1290                         compat_ulong_t word;
1291                         getreg32(target, pos, &word);
1292                         if (__put_user(word, u++))
1293                                 return -EFAULT;
1294                         count -= sizeof(*u);
1295                         pos += sizeof(*u);
1296                 }
1297         }
1298
1299         return 0;
1300 }
1301
1302 static int genregs32_set(struct task_struct *target,
1303                          const struct user_regset *regset,
1304                          unsigned int pos, unsigned int count,
1305                          const void *kbuf, const void __user *ubuf)
1306 {
1307         int ret = 0;
1308         if (kbuf) {
1309                 const compat_ulong_t *k = kbuf;
1310                 while (count > 0 && !ret) {
1311                         ret = putreg32(target, pos, *k++);
1312                         count -= sizeof(*k);
1313                         pos += sizeof(*k);
1314                 }
1315         } else {
1316                 const compat_ulong_t __user *u = ubuf;
1317                 while (count > 0 && !ret) {
1318                         compat_ulong_t word;
1319                         ret = __get_user(word, u++);
1320                         if (ret)
1321                                 break;
1322                         ret = putreg32(target, pos, word);
1323                         count -= sizeof(*u);
1324                         pos += sizeof(*u);
1325                 }
1326         }
1327         return ret;
1328 }
1329
1330 long compat_arch_ptrace(struct task_struct *child, compat_long_t request,
1331                         compat_ulong_t caddr, compat_ulong_t cdata)
1332 {
1333         unsigned long addr = caddr;
1334         unsigned long data = cdata;
1335         void __user *datap = compat_ptr(data);
1336         int ret;
1337         __u32 val;
1338
1339         switch (request) {
1340         case PTRACE_PEEKUSR:
1341                 ret = getreg32(child, addr, &val);
1342                 if (ret == 0)
1343                         ret = put_user(val, (__u32 __user *)datap);
1344                 break;
1345
1346         case PTRACE_POKEUSR:
1347                 ret = putreg32(child, addr, data);
1348                 break;
1349
1350         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
1351                 return copy_regset_to_user(child, &user_x86_32_view,
1352                                            REGSET_GENERAL,
1353                                            0, sizeof(struct user_regs_struct32),
1354                                            datap);
1355
1356         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
1357                 return copy_regset_from_user(child, &user_x86_32_view,
1358                                              REGSET_GENERAL, 0,
1359                                              sizeof(struct user_regs_struct32),
1360                                              datap);
1361
1362         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
1363                 return copy_regset_to_user(child, &user_x86_32_view,
1364                                            REGSET_FP, 0,
1365                                            sizeof(struct user_i387_ia32_struct),
1366                                            datap);
1367
1368         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
1369                 return copy_regset_from_user(
1370                         child, &user_x86_32_view, REGSET_FP,
1371                         0, sizeof(struct user_i387_ia32_struct), datap);
1372
1373         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
1374                 return copy_regset_to_user(child, &user_x86_32_view,
1375                                            REGSET_XFP, 0,
1376                                            sizeof(struct user32_fxsr_struct),
1377                                            datap);
1378
1379         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
1380                 return copy_regset_from_user(child, &user_x86_32_view,
1381                                              REGSET_XFP, 0,
1382                                              sizeof(struct user32_fxsr_struct),
1383                                              datap);
1384
1385         case PTRACE_GET_THREAD_AREA:
1386         case PTRACE_SET_THREAD_AREA:
1387                 return arch_ptrace(child, request, addr, data);
1388
1389         default:
1390                 return compat_ptrace_request(child, request, addr, data);
1391         }
1392
1393         return ret;
1394 }
1395
1396 #endif  /* CONFIG_IA32_EMULATION */
1397
1398 #ifdef CONFIG_X86_64
1399
1400 static const struct user_regset x86_64_regsets[] = {
1401         [REGSET_GENERAL] = {
1402                 .core_note_type = NT_PRSTATUS,
1403                 .n = sizeof(struct user_regs_struct) / sizeof(long),
1404                 .size = sizeof(long), .align = sizeof(long),
1405                 .get = genregs_get, .set = genregs_set
1406         },
1407         [REGSET_FP] = {
1408                 .core_note_type = NT_PRFPREG,
1409                 .n = sizeof(struct user_i387_struct) / sizeof(long),
1410                 .size = sizeof(long), .align = sizeof(long),
1411                 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1412         },
1413         [REGSET_IOPERM64] = {
1414                 .core_note_type = NT_386_IOPERM,
1415                 .n = IO_BITMAP_LONGS,
1416                 .size = sizeof(long), .align = sizeof(long),
1417                 .active = ioperm_active, .get = ioperm_get
1418         },
1419 };
1420
1421 static const struct user_regset_view user_x86_64_view = {
1422         .name = "x86_64", .e_machine = EM_X86_64,
1423         .regsets = x86_64_regsets, .n = ARRAY_SIZE(x86_64_regsets)
1424 };
1425
1426 #else  /* CONFIG_X86_32 */
1427
1428 #define user_regs_struct32      user_regs_struct
1429 #define genregs32_get           genregs_get
1430 #define genregs32_set           genregs_set
1431
1432 #define user_i387_ia32_struct   user_i387_struct
1433 #define user32_fxsr_struct      user_fxsr_struct
1434
1435 #endif  /* CONFIG_X86_64 */
1436
1437 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1438 static const struct user_regset x86_32_regsets[] = {
1439         [REGSET_GENERAL] = {
1440                 .core_note_type = NT_PRSTATUS,
1441                 .n = sizeof(struct user_regs_struct32) / sizeof(u32),
1442                 .size = sizeof(u32), .align = sizeof(u32),
1443                 .get = genregs32_get, .set = genregs32_set
1444         },
1445         [REGSET_FP] = {
1446                 .core_note_type = NT_PRFPREG,
1447                 .n = sizeof(struct user_i387_ia32_struct) / sizeof(u32),
1448                 .size = sizeof(u32), .align = sizeof(u32),
1449                 .active = fpregs_active, .get = fpregs_get, .set = fpregs_set
1450         },
1451         [REGSET_XFP] = {
1452                 .core_note_type = NT_PRXFPREG,
1453                 .n = sizeof(struct user32_fxsr_struct) / sizeof(u32),
1454                 .size = sizeof(u32), .align = sizeof(u32),
1455                 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1456         },
1457         [REGSET_TLS] = {
1458                 .core_note_type = NT_386_TLS,
1459                 .n = GDT_ENTRY_TLS_ENTRIES, .bias = GDT_ENTRY_TLS_MIN,
1460                 .size = sizeof(struct user_desc),
1461                 .align = sizeof(struct user_desc),
1462                 .active = regset_tls_active,
1463                 .get = regset_tls_get, .set = regset_tls_set
1464         },
1465         [REGSET_IOPERM32] = {
1466                 .core_note_type = NT_386_IOPERM,
1467                 .n = IO_BITMAP_BYTES / sizeof(u32),
1468                 .size = sizeof(u32), .align = sizeof(u32),
1469                 .active = ioperm_active, .get = ioperm_get
1470         },
1471 };
1472
1473 static const struct user_regset_view user_x86_32_view = {
1474         .name = "i386", .e_machine = EM_386,
1475         .regsets = x86_32_regsets, .n = ARRAY_SIZE(x86_32_regsets)
1476 };
1477 #endif
1478
1479 const struct user_regset_view *task_user_regset_view(struct task_struct *task)
1480 {
1481 #ifdef CONFIG_IA32_EMULATION
1482         if (test_tsk_thread_flag(task, TIF_IA32))
1483 #endif
1484 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1485                 return &user_x86_32_view;
1486 #endif
1487 #ifdef CONFIG_X86_64
1488         return &user_x86_64_view;
1489 #endif
1490 }
1491
1492 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs,
1493                                          int error_code, int si_code)
1494 {
1495         struct siginfo info;
1496
1497         tsk->thread.trap_no = 1;
1498         tsk->thread.error_code = error_code;
1499
1500         memset(&info, 0, sizeof(info));
1501         info.si_signo = SIGTRAP;
1502         info.si_code = si_code;
1503
1504         /* User-mode ip? */
1505         info.si_addr = user_mode_vm(regs) ? (void __user *) regs->ip : NULL;
1506
1507         /* Send us the fake SIGTRAP */
1508         force_sig_info(SIGTRAP, &info, tsk);
1509 }
1510
1511
1512 #ifdef CONFIG_X86_32
1513 # define IS_IA32        1
1514 #elif defined CONFIG_IA32_EMULATION
1515 # define IS_IA32        test_thread_flag(TIF_IA32)
1516 #else
1517 # define IS_IA32        0
1518 #endif
1519
1520 /*
1521  * We must return the syscall number to actually look up in the table.
1522  * This can be -1L to skip running any syscall at all.
1523  */
1524 asmregparm long syscall_trace_enter(struct pt_regs *regs)
1525 {
1526         long ret = 0;
1527
1528         /*
1529          * If we stepped into a sysenter/syscall insn, it trapped in
1530          * kernel mode; do_debug() cleared TF and set TIF_SINGLESTEP.
1531          * If user-mode had set TF itself, then it's still clear from
1532          * do_debug() and we need to set it again to restore the user
1533          * state.  If we entered on the slow path, TF was already set.
1534          */
1535         if (test_thread_flag(TIF_SINGLESTEP))
1536                 regs->flags |= X86_EFLAGS_TF;
1537
1538         /* do the secure computing check first */
1539         secure_computing(regs->orig_ax);
1540
1541         if (unlikely(test_thread_flag(TIF_SYSCALL_EMU)))
1542                 ret = -1L;
1543
1544         if ((ret || test_thread_flag(TIF_SYSCALL_TRACE)) &&
1545             tracehook_report_syscall_entry(regs))
1546                 ret = -1L;
1547
1548         if (unlikely(current->audit_context)) {
1549                 if (IS_IA32)
1550                         audit_syscall_entry(AUDIT_ARCH_I386,
1551                                             regs->orig_ax,
1552                                             regs->bx, regs->cx,
1553                                             regs->dx, regs->si);
1554 #ifdef CONFIG_X86_64
1555                 else
1556                         audit_syscall_entry(AUDIT_ARCH_X86_64,
1557                                             regs->orig_ax,
1558                                             regs->di, regs->si,
1559                                             regs->dx, regs->r10);
1560 #endif
1561         }
1562
1563         return ret ?: regs->orig_ax;
1564 }
1565
1566 asmregparm void syscall_trace_leave(struct pt_regs *regs)
1567 {
1568         if (unlikely(current->audit_context))
1569                 audit_syscall_exit(AUDITSC_RESULT(regs->ax), regs->ax);
1570
1571         if (test_thread_flag(TIF_SYSCALL_TRACE))
1572                 tracehook_report_syscall_exit(regs, 0);
1573
1574         /*
1575          * If TIF_SYSCALL_EMU is set, we only get here because of
1576          * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
1577          * We already reported this syscall instruction in
1578          * syscall_trace_enter(), so don't do any more now.
1579          */
1580         if (unlikely(test_thread_flag(TIF_SYSCALL_EMU)))
1581                 return;
1582
1583         /*
1584          * If we are single-stepping, synthesize a trap to follow the
1585          * system call instruction.
1586          */
1587         if (test_thread_flag(TIF_SINGLESTEP) &&
1588             tracehook_consider_fatal_signal(current, SIGTRAP, SIG_DFL))
1589                 send_sigtrap(current, regs, 0, TRAP_BRKPT);
1590 }