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