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