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/ftrace.h>
  25 #include <linux/workqueue.h>
  26
  27 #include <asm/uaccess.h>
  28 #include <asm/pgtable.h>
  29 #include <asm/system.h>
  30 #include <asm/processor.h>
  31 #include <asm/i387.h>
  32 #include <asm/debugreg.h>
  33 #include <asm/ldt.h>
  34 #include <asm/desc.h>
  35 #include <asm/prctl.h>
  36 #include <asm/proto.h>
  37 #include <asm/ds.h>
  38
  39 #include "tls.h"
  40
  41 enum x86_regset {
  42         REGSET_GENERAL,
  43         REGSET_FP,
  44         REGSET_XFP,
  45         REGSET_IOPERM64 = REGSET_XFP,
  46         REGSET_TLS,
  47         REGSET_IOPERM32,
  48 };
  49
  50 /*
  51  * does not yet catch signals sent when the child dies.
  52  * in exit.c or in signal.c.
  53  */
  54
  55 /*
  56  * Determines which flags the user has access to [1 = access, 0 = no access].
  57  */
  58 #define FLAG_MASK_32            ((unsigned long)                        \
  59                                  (X86_EFLAGS_CF | X86_EFLAGS_PF |       \
  60                                   X86_EFLAGS_AF | X86_EFLAGS_ZF |       \
  61                                   X86_EFLAGS_SF | X86_EFLAGS_TF |       \
  62                                   X86_EFLAGS_DF | X86_EFLAGS_OF |       \
  63                                   X86_EFLAGS_RF | X86_EFLAGS_AC))
  64
  65 /*
  66  * Determines whether a value may be installed in a segment register.
  67  */
  68 static inline bool invalid_selector(u16 value)
  69 {
  70         return unlikely(value != 0 && (value & SEGMENT_RPL_MASK) != USER_RPL);
  71 }
  72
  73 #ifdef CONFIG_X86_32
  74
  75 #define FLAG_MASK               FLAG_MASK_32
  76
  77 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long regno)
  78 {
  79         BUILD_BUG_ON(offsetof(struct pt_regs, bx) != 0);
  80         return &regs->bx + (regno >> 2);
  81 }
  82
  83 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
  84 {
  85         /*
  86          * Returning the value truncates it to 16 bits.
  87          */
  88         unsigned int retval;
  89         if (offset != offsetof(struct user_regs_struct, gs))
  90                 retval = *pt_regs_access(task_pt_regs(task), offset);
  91         else {
  92                 if (task == current)
  93                         retval = get_user_gs(task_pt_regs(task));
  94                 else
  95                         retval = task_user_gs(task);
  96         }
  97         return retval;
  98 }
  99
 100 static int set_segment_reg(struct task_struct *task,
 101                            unsigned long offset, u16 value)
 102 {
 103         /*
 104          * The value argument was already truncated to 16 bits.
 105          */
 106         if (invalid_selector(value))
 107                 return -EIO;
 108
 109         /*
 110          * For %cs and %ss we cannot permit a null selector.
 111          * We can permit a bogus selector as long as it has USER_RPL.
 112          * Null selectors are fine for other segment registers, but
 113          * we will never get back to user mode with invalid %cs or %ss
 114          * and will take the trap in iret instead.  Much code relies
 115          * on user_mode() to distinguish a user trap frame (which can
 116          * safely use invalid selectors) from a kernel trap frame.
 117          */
 118         switch (offset) {
 119         case offsetof(struct user_regs_struct, cs):
 120         case offsetof(struct user_regs_struct, ss):
 121                 if (unlikely(value == 0))
 122                         return -EIO;
 123
 124         default:
 125                 *pt_regs_access(task_pt_regs(task), offset) = value;
 126                 break;
 127
 128         case offsetof(struct user_regs_struct, gs):
 129                 if (task == current)
 130                         set_user_gs(task_pt_regs(task), value);
 131                 else
 132                         task_user_gs(task) = value;
 133         }
 134
 135         return 0;
 136 }
 137
 138 static unsigned long debugreg_addr_limit(struct task_struct *task)
 139 {
 140         return TASK_SIZE - 3;
 141 }
 142
 143 #else  /* CONFIG_X86_64 */
 144
 145 #define FLAG_MASK               (FLAG_MASK_32 | X86_EFLAGS_NT)
 146
 147 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long offset)
 148 {
 149         BUILD_BUG_ON(offsetof(struct pt_regs, r15) != 0);
 150         return &regs->r15 + (offset / sizeof(regs->r15));
 151 }
 152
 153 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
 154 {
 155         /*
 156          * Returning the value truncates it to 16 bits.
 157          */
 158         unsigned int seg;
 159
 160         switch (offset) {
 161         case offsetof(struct user_regs_struct, fs):
 162                 if (task == current) {
 163                         /* Older gas can't assemble movq %?s,%r?? */
 164                         asm("movl %%fs,%0" : "=r" (seg));
 165                         return seg;
 166                 }
 167                 return task->thread.fsindex;
 168         case offsetof(struct user_regs_struct, gs):
 169                 if (task == current) {
 170                         asm("movl %%gs,%0" : "=r" (seg));
 171                         return seg;
 172                 }
 173                 return task->thread.gsindex;
 174         case offsetof(struct user_regs_struct, ds):
 175                 if (task == current) {
 176                         asm("movl %%ds,%0" : "=r" (seg));
 177                         return seg;
 178                 }
 179                 return task->thread.ds;
 180         case offsetof(struct user_regs_struct, es):
 181                 if (task == current) {
 182                         asm("movl %%es,%0" : "=r" (seg));
 183                         return seg;
 184                 }
 185                 return task->thread.es;
 186
 187         case offsetof(struct user_regs_struct, cs):
 188         case offsetof(struct user_regs_struct, ss):
 189                 break;
 190         }
 191         return *pt_regs_access(task_pt_regs(task), offset);
 192 }
 193
 194 static int set_segment_reg(struct task_struct *task,
 195                            unsigned long offset, u16 value)
 196 {
 197         /*
 198          * The value argument was already truncated to 16 bits.
 199          */
 200         if (invalid_selector(value))
 201                 return -EIO;
 202
 203         switch (offset) {
 204         case offsetof(struct user_regs_struct,fs):
 205                 /*
 206                  * If this is setting fs as for normal 64-bit use but
 207                  * setting fs_base has implicitly changed it, leave it.
 208                  */
 209                 if ((value == FS_TLS_SEL && task->thread.fsindex == 0 &&
 210                      task->thread.fs != 0) ||
 211                     (value == 0 && task->thread.fsindex == FS_TLS_SEL &&
 212                      task->thread.fs == 0))
 213                         break;
 214                 task->thread.fsindex = value;
 215                 if (task == current)
 216                         loadsegment(fs, task->thread.fsindex);
 217                 break;
 218         case offsetof(struct user_regs_struct,gs):
 219                 /*
 220                  * If this is setting gs as for normal 64-bit use but
 221                  * setting gs_base has implicitly changed it, leave it.
 222                  */
 223                 if ((value == GS_TLS_SEL && task->thread.gsindex == 0 &&
 224                      task->thread.gs != 0) ||
 225                     (value == 0 && task->thread.gsindex == GS_TLS_SEL &&
 226                      task->thread.gs == 0))
 227                         break;
 228                 task->thread.gsindex = value;
 229                 if (task == current)
 230                         load_gs_index(task->thread.gsindex);
 231                 break;
 232         case offsetof(struct user_regs_struct,ds):
 233                 task->thread.ds = value;
 234                 if (task == current)
 235                         loadsegment(ds, task->thread.ds);
 236                 break;
 237         case offsetof(struct user_regs_struct,es):
 238                 task->thread.es = value;
 239                 if (task == current)
 240                         loadsegment(es, task->thread.es);
 241                 break;
 242
 243                 /*
 244                  * Can't actually change these in 64-bit mode.
 245                  */
 246         case offsetof(struct user_regs_struct,cs):
 247                 if (unlikely(value == 0))
 248                         return -EIO;
 249 #ifdef CONFIG_IA32_EMULATION
 250                 if (test_tsk_thread_flag(task, TIF_IA32))
 251                         task_pt_regs(task)->cs = value;
 252 #endif
 253                 break;
 254         case offsetof(struct user_regs_struct,ss):
 255                 if (unlikely(value == 0))
 256                         return -EIO;
 257 #ifdef CONFIG_IA32_EMULATION
 258                 if (test_tsk_thread_flag(task, TIF_IA32))
 259                         task_pt_regs(task)->ss = value;
 260 #endif
 261                 break;
 262         }
 263
 264         return 0;
 265 }
 266
 267 static unsigned long debugreg_addr_limit(struct task_struct *task)
 268 {
 269 #ifdef CONFIG_IA32_EMULATION
 270         if (test_tsk_thread_flag(task, TIF_IA32))
 271                 return IA32_PAGE_OFFSET - 3;
 272 #endif
 273         return TASK_SIZE_MAX - 7;
 274 }
 275
 276 #endif  /* CONFIG_X86_32 */
 277
 278 static unsigned long get_flags(struct task_struct *task)
 279 {
 280         unsigned long retval = task_pt_regs(task)->flags;
 281
 282         /*
 283          * If the debugger set TF, hide it from the readout.
 284          */
 285         if (test_tsk_thread_flag(task, TIF_FORCED_TF))
 286                 retval &= ~X86_EFLAGS_TF;
 287
 288         return retval;
 289 }
 290
 291 static int set_flags(struct task_struct *task, unsigned long value)
 292 {
 293         struct pt_regs *regs = task_pt_regs(task);
 294
 295         /*
 296          * If the user value contains TF, mark that
 297          * it was not "us" (the debugger) that set it.
 298          * If not, make sure it stays set if we had.
 299          */
 300         if (value & X86_EFLAGS_TF)
 301                 clear_tsk_thread_flag(task, TIF_FORCED_TF);
 302         else if (test_tsk_thread_flag(task, TIF_FORCED_TF))
 303                 value |= X86_EFLAGS_TF;
 304
 305         regs->flags = (regs->flags & ~FLAG_MASK) | (value & FLAG_MASK);
 306
 307         return 0;
 308 }
 309
 310 static int putreg(struct task_struct *child,
 311                   unsigned long offset, unsigned long value)
 312 {
 313         switch (offset) {
 314         case offsetof(struct user_regs_struct, cs):
 315         case offsetof(struct user_regs_struct, ds):
 316         case offsetof(struct user_regs_struct, es):
 317         case offsetof(struct user_regs_struct, fs):
 318         case offsetof(struct user_regs_struct, gs):
 319         case offsetof(struct user_regs_struct, ss):
 320                 return set_segment_reg(child, offset, value);
 321
 322         case offsetof(struct user_regs_struct, flags):
 323                 return set_flags(child, value);
 324
 325 #ifdef CONFIG_X86_64
 326         /*
 327          * Orig_ax is really just a flag with small positive and
 328          * negative values, so make sure to always sign-extend it
 329          * from 32 bits so that it works correctly regardless of
 330          * whether we come from a 32-bit environment or not.
 331          */
 332         case offsetof(struct user_regs_struct, orig_ax):
 333                 value = (long) (s32) value;
 334                 break;
 335
 336         case offsetof(struct user_regs_struct,fs_base):
 337                 if (value >= TASK_SIZE_OF(child))
 338                         return -EIO;
 339                 /*
 340                  * When changing the segment base, use do_arch_prctl
 341                  * to set either thread.fs or thread.fsindex and the
 342                  * corresponding GDT slot.
 343                  */
 344                 if (child->thread.fs != value)
 345                         return do_arch_prctl(child, ARCH_SET_FS, value);
 346                 return 0;
 347         case offsetof(struct user_regs_struct,gs_base):
 348                 /*
 349                  * Exactly the same here as the %fs handling above.
 350                  */
 351                 if (value >= TASK_SIZE_OF(child))
 352                         return -EIO;
 353                 if (child->thread.gs != value)
 354                         return do_arch_prctl(child, ARCH_SET_GS, value);
 355                 return 0;
 356 #endif
 357         }
 358
 359         *pt_regs_access(task_pt_regs(child), offset) = value;
 360         return 0;
 361 }
 362
 363 static unsigned long getreg(struct task_struct *task, unsigned long offset)
 364 {
 365         switch (offset) {
 366         case offsetof(struct user_regs_struct, cs):
 367         case offsetof(struct user_regs_struct, ds):
 368         case offsetof(struct user_regs_struct, es):
 369         case offsetof(struct user_regs_struct, fs):
 370         case offsetof(struct user_regs_struct, gs):
 371         case offsetof(struct user_regs_struct, ss):
 372                 return get_segment_reg(task, offset);
 373
 374         case offsetof(struct user_regs_struct, flags):
 375                 return get_flags(task);
 376
 377 #ifdef CONFIG_X86_64
 378         case offsetof(struct user_regs_struct, fs_base): {
 379                 /*
 380                  * do_arch_prctl may have used a GDT slot instead of
 381                  * the MSR.  To userland, it appears the same either
 382                  * way, except the %fs segment selector might not be 0.
 383                  */
 384                 unsigned int seg = task->thread.fsindex;
 385                 if (task->thread.fs != 0)
 386                         return task->thread.fs;
 387                 if (task == current)
 388                         asm("movl %%fs,%0" : "=r" (seg));
 389                 if (seg != FS_TLS_SEL)
 390                         return 0;
 391                 return get_desc_base(&task->thread.tls_array[FS_TLS]);
 392         }
 393         case offsetof(struct user_regs_struct, gs_base): {
 394                 /*
 395                  * Exactly the same here as the %fs handling above.
 396                  */
 397                 unsigned int seg = task->thread.gsindex;
 398                 if (task->thread.gs != 0)
 399                         return task->thread.gs;
 400                 if (task == current)
 401                         asm("movl %%gs,%0" : "=r" (seg));
 402                 if (seg != GS_TLS_SEL)
 403                         return 0;
 404                 return get_desc_base(&task->thread.tls_array[GS_TLS]);
 405         }
 406 #endif
 407         }
 408
 409         return *pt_regs_access(task_pt_regs(task), offset);
 410 }
 411
 412 static int genregs_get(struct task_struct *target,
 413                        const struct user_regset *regset,
 414                        unsigned int pos, unsigned int count,
 415                        void *kbuf, void __user *ubuf)
 416 {
 417         if (kbuf) {
 418                 unsigned long *k = kbuf;
 419                 while (count > 0) {
 420                         *k++ = getreg(target, pos);
 421                         count -= sizeof(*k);
 422                         pos += sizeof(*k);
 423                 }
 424         } else {
 425                 unsigned long __user *u = ubuf;
 426                 while (count > 0) {
 427                         if (__put_user(getreg(target, pos), u++))
 428                                 return -EFAULT;
 429                         count -= sizeof(*u);
 430                         pos += sizeof(*u);
 431                 }
 432         }
 433
 434         return 0;
 435 }
 436
 437 static int genregs_set(struct task_struct *target,
 438                        const struct user_regset *regset,
 439                        unsigned int pos, unsigned int count,
 440                        const void *kbuf, const void __user *ubuf)
 441 {
 442         int ret = 0;
 443         if (kbuf) {
 444                 const unsigned long *k = kbuf;
 445                 while (count > 0 && !ret) {
 446                         ret = putreg(target, pos, *k++);
 447                         count -= sizeof(*k);
 448                         pos += sizeof(*k);
 449                 }
 450         } else {
 451                 const unsigned long  __user *u = ubuf;
 452                 while (count > 0 && !ret) {
 453                         unsigned long word;
 454                         ret = __get_user(word, u++);
 455                         if (ret)
 456                                 break;
 457                         ret = putreg(target, pos, word);
 458                         count -= sizeof(*u);
 459                         pos += sizeof(*u);
 460                 }
 461         }
 462         return ret;
 463 }
 464
 465 /*
 466  * This function is trivial and will be inlined by the compiler.
 467  * Having it separates the implementation details of debug
 468  * registers from the interface details of ptrace.
 469  */
 470 static unsigned long ptrace_get_debugreg(struct task_struct *child, int n)
 471 {
 472         switch (n) {
 473         case 0:         return child->thread.debugreg0;
 474         case 1:         return child->thread.debugreg1;
 475         case 2:         return child->thread.debugreg2;
 476         case 3:         return child->thread.debugreg3;
 477         case 6:         return child->thread.debugreg6;
 478         case 7:         return child->thread.debugreg7;
 479         }
 480         return 0;
 481 }
 482
 483 static int ptrace_set_debugreg(struct task_struct *child,
 484                                int n, unsigned long data)
 485 {
 486         int i;
 487
 488         if (unlikely(n == 4 || n == 5))
 489                 return -EIO;
 490
 491         if (n < 4 && unlikely(data >= debugreg_addr_limit(child)))
 492                 return -EIO;
 493
 494         switch (n) {
 495         case 0:         child->thread.debugreg0 = data; break;
 496         case 1:         child->thread.debugreg1 = data; break;
 497         case 2:         child->thread.debugreg2 = data; break;
 498         case 3:         child->thread.debugreg3 = data; break;
 499
 500         case 6:
 501                 if ((data & ~0xffffffffUL) != 0)
 502                         return -EIO;
 503                 child->thread.debugreg6 = data;
 504                 break;
 505
 506         case 7:
 507                 /*
 508                  * Sanity-check data. Take one half-byte at once with
 509                  * check = (val >> (16 + 4*i)) & 0xf. It contains the
 510                  * R/Wi and LENi bits; bits 0 and 1 are R/Wi, and bits
 511                  * 2 and 3 are LENi. Given a list of invalid values,
 512                  * we do mask |= 1 << invalid_value, so that
 513                  * (mask >> check) & 1 is a correct test for invalid
 514                  * values.
 515                  *
 516                  * R/Wi contains the type of the breakpoint /
 517                  * watchpoint, LENi contains the length of the watched
 518                  * data in the watchpoint case.
 519                  *
 520                  * The invalid values are:
 521                  * - LENi == 0x10 (undefined), so mask |= 0x0f00.       [32-bit]
 522                  * - R/Wi == 0x10 (break on I/O reads or writes), so
 523                  *   mask |= 0x4444.
 524                  * - R/Wi == 0x00 && LENi != 0x00, so we have mask |=
 525                  *   0x1110.
 526                  *
 527                  * Finally, mask = 0x0f00 | 0x4444 | 0x1110 == 0x5f54.
 528                  *
 529                  * See the Intel Manual "System Programming Guide",
 530                  * 15.2.4
 531                  *
 532                  * Note that LENi == 0x10 is defined on x86_64 in long
 533                  * mode (i.e. even for 32-bit userspace software, but
 534                  * 64-bit kernel), so the x86_64 mask value is 0x5454.
 535                  * See the AMD manual no. 24593 (AMD64 System Programming)
 536                  */
 537 #ifdef CONFIG_X86_32
 538 #define DR7_MASK        0x5f54
 539 #else
 540 #define DR7_MASK        0x5554
 541 #endif
 542                 data &= ~DR_CONTROL_RESERVED;
 543                 for (i = 0; i < 4; i++)
 544                         if ((DR7_MASK >> ((data >> (16 + 4*i)) & 0xf)) & 1)
 545                                 return -EIO;
 546                 child->thread.debugreg7 = data;
 547                 if (data)
 548                         set_tsk_thread_flag(child, TIF_DEBUG);
 549                 else
 550                         clear_tsk_thread_flag(child, TIF_DEBUG);
 551                 break;
 552         }
 553
 554         return 0;
 555 }
 556
 557 /*
 558  * These access the current or another (stopped) task's io permission
 559  * bitmap for debugging or core dump.
 560  */
 561 static int ioperm_active(struct task_struct *target,
 562                          const struct user_regset *regset)
 563 {
 564         return target->thread.io_bitmap_max / regset->size;
 565 }
 566
 567 static int ioperm_get(struct task_struct *target,
 568                       const struct user_regset *regset,
 569                       unsigned int pos, unsigned int count,
 570                       void *kbuf, void __user *ubuf)
 571 {
 572         if (!target->thread.io_bitmap_ptr)
 573                 return -ENXIO;
 574
 575         return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
 576                                    target->thread.io_bitmap_ptr,
 577                                    0, IO_BITMAP_BYTES);
 578 }
 579
 580 #ifdef CONFIG_X86_PTRACE_BTS
 581 /*
 582  * A branch trace store context.
 583  *
 584  * Contexts may only be installed by ptrace_bts_config() and only for
 585  * ptraced tasks.
 586  *
 587  * Contexts are destroyed when the tracee is detached from the tracer.
 588  * The actual destruction work requires interrupts enabled, so the
 589  * work is deferred and will be scheduled during __ptrace_unlink().
 590  *
 591  * Contexts hold an additional task_struct reference on the traced
 592  * task, as well as a reference on the tracer's mm.
 593  *
 594  * Ptrace already holds a task_struct for the duration of ptrace operations,
 595  * but since destruction is deferred, it may be executed after both
 596  * tracer and tracee exited.
 597  */
 598 struct bts_context {
 599         /* The branch trace handle. */
 600         struct bts_tracer       *tracer;
 601
 602         /* The buffer used to store the branch trace and its size. */
 603         void                    *buffer;
 604         unsigned int            size;
 605
 606         /* The mm that paid for the above buffer. */
 607         struct mm_struct        *mm;
 608
 609         /* The task this context belongs to. */
 610         struct task_struct      *task;
 611
 612         /* The signal to send on a bts buffer overflow. */
 613         unsigned int            bts_ovfl_signal;
 614
 615         /* The work struct to destroy a context. */
 616         struct work_struct      work;
 617 };
 618
 619 static inline void alloc_bts_buffer(struct bts_context *context,
 620                                     unsigned int size)
 621 {
 622         void *buffer;
 623
 624         buffer = alloc_locked_buffer(size);
 625         if (buffer) {
 626                 context->buffer = buffer;
 627                 context->size = size;
 628                 context->mm = get_task_mm(current);
 629         }
 630 }
 631
 632 static inline void free_bts_buffer(struct bts_context *context)
 633 {
 634         if (!context->buffer)
 635                 return;
 636
 637         kfree(context->buffer);
 638         context->buffer = NULL;
 639
 640         refund_locked_buffer_memory(context->mm, context->size);
 641         context->size = 0;
 642
 643         mmput(context->mm);
 644         context->mm = NULL;
 645 }
 646
 647 static void free_bts_context_work(struct work_struct *w)
 648 {
 649         struct bts_context *context;
 650
 651         context = container_of(w, struct bts_context, work);
 652
 653         ds_release_bts(context->tracer);
 654         put_task_struct(context->task);
 655         free_bts_buffer(context);
 656         kfree(context);
 657 }
 658
 659 static inline void free_bts_context(struct bts_context *context)
 660 {
 661         INIT_WORK(&context->work, free_bts_context_work);
 662         schedule_work(&context->work);
 663 }
 664
 665 static inline struct bts_context *alloc_bts_context(struct task_struct *task)
 666 {
 667         struct bts_context *context = kzalloc(sizeof(*context), GFP_KERNEL);
 668         if (context) {
 669                 context->task = task;
 670                 task->bts = context;
 671
 672                 get_task_struct(task);
 673         }
 674
 675         return context;
 676 }
 677
 678 static int ptrace_bts_read_record(struct task_struct *child, size_t index,
 679                                   struct bts_struct __user *out)
 680 {
 681         struct bts_context *context;
 682         const struct bts_trace *trace;
 683         struct bts_struct bts;
 684         const unsigned char *at;
 685         int error;
 686
 687         context = child->bts;
 688         if (!context)
 689                 return -ESRCH;
 690
 691         trace = ds_read_bts(context->tracer);
 692         if (!trace)
 693                 return -ESRCH;
 694
 695         at = trace->ds.top - ((index + 1) * trace->ds.size);
 696         if ((void *)at < trace->ds.begin)
 697                 at += (trace->ds.n * trace->ds.size);
 698
 699         if (!trace->read)
 700                 return -EOPNOTSUPP;
 701
 702         error = trace->read(context->tracer, at, &bts);
 703         if (error < 0)
 704                 return error;
 705
 706         if (copy_to_user(out, &bts, sizeof(bts)))
 707                 return -EFAULT;
 708
 709         return sizeof(bts);
 710 }
 711
 712 static int ptrace_bts_drain(struct task_struct *child,
 713                             long size,
 714                             struct bts_struct __user *out)
 715 {
 716         struct bts_context *context;
 717         const struct bts_trace *trace;
 718         const unsigned char *at;
 719         int error, drained = 0;
 720
 721         context = child->bts;
 722         if (!context)
 723                 return -ESRCH;
 724
 725         trace = ds_read_bts(context->tracer);
 726         if (!trace)
 727                 return -ESRCH;
 728
 729         if (!trace->read)
 730                 return -EOPNOTSUPP;
 731
 732         if (size < (trace->ds.top - trace->ds.begin))
 733                 return -EIO;
 734
 735         for (at = trace->ds.begin; (void *)at < trace->ds.top;
 736              out++, drained++, at += trace->ds.size) {
 737                 struct bts_struct bts;
 738
 739                 error = trace->read(context->tracer, at, &bts);
 740                 if (error < 0)
 741                         return error;
 742
 743                 if (copy_to_user(out, &bts, sizeof(bts)))
 744                         return -EFAULT;
 745         }
 746
 747         memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size);
 748
 749         error = ds_reset_bts(context->tracer);
 750         if (error < 0)
 751                 return error;
 752
 753         return drained;
 754 }
 755
 756 static int ptrace_bts_config(struct task_struct *child,
 757                              long cfg_size,
 758                              const struct ptrace_bts_config __user *ucfg)
 759 {
 760         struct bts_context *context;
 761         struct ptrace_bts_config cfg;
 762         unsigned int flags = 0;
 763
 764         if (cfg_size < sizeof(cfg))
 765                 return -EIO;
 766
 767         if (copy_from_user(&cfg, ucfg, sizeof(cfg)))
 768                 return -EFAULT;
 769
 770         context = child->bts;
 771         if (!context)
 772                 context = alloc_bts_context(child);
 773         if (!context)
 774                 return -ENOMEM;
 775
 776         if (cfg.flags & PTRACE_BTS_O_SIGNAL) {
 777                 if (!cfg.signal)
 778                         return -EINVAL;
 779
 780                 return -EOPNOTSUPP;
 781                 context->bts_ovfl_signal = cfg.signal;
 782         }
 783
 784         ds_release_bts(context->tracer);
 785         context->tracer = NULL;
 786
 787         if ((cfg.flags & PTRACE_BTS_O_ALLOC) && (cfg.size != context->size)) {
 788                 free_bts_buffer(context);
 789                 if (!cfg.size)
 790                         return 0;
 791
 792                 alloc_bts_buffer(context, cfg.size);
 793                 if (!context->buffer)
 794                         return -ENOMEM;
 795         }
 796
 797         if (cfg.flags & PTRACE_BTS_O_TRACE)
 798                 flags |= BTS_USER;
 799
 800         if (cfg.flags & PTRACE_BTS_O_SCHED)
 801                 flags |= BTS_TIMESTAMPS;
 802
 803         context->tracer = ds_request_bts(child, context->buffer, context->size,
 804                                          NULL, (size_t)-1, flags);
 805         if (unlikely(IS_ERR(context->tracer))) {
 806                 int error = PTR_ERR(context->tracer);
 807
 808                 free_bts_buffer(context);
 809                 context->tracer = NULL;
 810                 return error;
 811         }
 812
 813         return sizeof(cfg);
 814 }
 815
 816 static int ptrace_bts_status(struct task_struct *child,
 817                              long cfg_size,
 818                              struct ptrace_bts_config __user *ucfg)
 819 {
 820         struct bts_context *context;
 821         const struct bts_trace *trace;
 822         struct ptrace_bts_config cfg;
 823
 824         context = child->bts;
 825         if (!context)
 826                 return -ESRCH;
 827
 828         if (cfg_size < sizeof(cfg))
 829                 return -EIO;
 830
 831         trace = ds_read_bts(context->tracer);
 832         if (!trace)
 833                 return -ESRCH;
 834
 835         memset(&cfg, 0, sizeof(cfg));
 836         cfg.size        = trace->ds.end - trace->ds.begin;
 837         cfg.signal      = context->bts_ovfl_signal;
 838         cfg.bts_size    = sizeof(struct bts_struct);
 839
 840         if (cfg.signal)
 841                 cfg.flags |= PTRACE_BTS_O_SIGNAL;
 842
 843         if (trace->ds.flags & BTS_USER)
 844                 cfg.flags |= PTRACE_BTS_O_TRACE;
 845
 846         if (trace->ds.flags & BTS_TIMESTAMPS)
 847                 cfg.flags |= PTRACE_BTS_O_SCHED;
 848
 849         if (copy_to_user(ucfg, &cfg, sizeof(cfg)))
 850                 return -EFAULT;
 851
 852         return sizeof(cfg);
 853 }
 854
 855 static int ptrace_bts_clear(struct task_struct *child)
 856 {
 857         struct bts_context *context;
 858         const struct bts_trace *trace;
 859
 860         context = child->bts;
 861         if (!context)
 862                 return -ESRCH;
 863
 864         trace = ds_read_bts(context->tracer);
 865         if (!trace)
 866                 return -ESRCH;
 867
 868         memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size);
 869
 870         return ds_reset_bts(context->tracer);
 871 }
 872
 873 static int ptrace_bts_size(struct task_struct *child)
 874 {
 875         struct bts_context *context;
 876         const struct bts_trace *trace;
 877
 878         context = child->bts;
 879         if (!context)
 880                 return -ESRCH;
 881
 882         trace = ds_read_bts(context->tracer);
 883         if (!trace)
 884                 return -ESRCH;
 885
 886         return (trace->ds.top - trace->ds.begin) / trace->ds.size;
 887 }
 888
 889 static inline void ptrace_bts_fork(struct task_struct *tsk)
 890 {
 891         tsk->bts = NULL;
 892 }
 893
 894 /*
 895  * Called from __ptrace_unlink() after the child has been moved back
 896  * to its original parent.
 897  */
 898 static inline void ptrace_bts_untrace(struct task_struct *child)
 899 {
 900         if (unlikely(child->bts)) {
 901                 free_bts_context(child->bts);
 902                 child->bts = NULL;
 903         }
 904 }
 905 #else
 906 static inline void ptrace_bts_fork(struct task_struct *tsk) {}
 907 static inline void ptrace_bts_untrace(struct task_struct *child) {}
 908 #endif /* CONFIG_X86_PTRACE_BTS */
 909
 910 void x86_ptrace_fork(struct task_struct *child, unsigned long clone_flags)
 911 {
 912         ptrace_bts_fork(child);
 913 }
 914
 915 void x86_ptrace_untrace(struct task_struct *child)
 916 {
 917         ptrace_bts_untrace(child);
 918 }
 919
 920 /*
 921  * Called by kernel/ptrace.c when detaching..
 922  *
 923  * Make sure the single step bit is not set.
 924  */
 925 void ptrace_disable(struct task_struct *child)
 926 {
 927         user_disable_single_step(child);
 928 #ifdef TIF_SYSCALL_EMU
 929         clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
 930 #endif
 931 }
 932
 933 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
 934 static const struct user_regset_view user_x86_32_view; /* Initialized below. */
 935 #endif
 936
 937 long arch_ptrace(struct task_struct *child, long request, long addr, long data)
 938 {
 939         int ret;
 940         unsigned long __user *datap = (unsigned long __user *)data;
 941
 942         switch (request) {
 943         /* read the word at location addr in the USER area. */
 944         case PTRACE_PEEKUSR: {
 945                 unsigned long tmp;
 946
 947                 ret = -EIO;
 948                 if ((addr & (sizeof(data) - 1)) || addr < 0 ||
 949                     addr >= sizeof(struct user))
 950                         break;
 951
 952                 tmp = 0;  /* Default return condition */
 953                 if (addr < sizeof(struct user_regs_struct))
 954                         tmp = getreg(child, addr);
 955                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
 956                          addr <= offsetof(struct user, u_debugreg[7])) {
 957                         addr -= offsetof(struct user, u_debugreg[0]);
 958                         tmp = ptrace_get_debugreg(child, addr / sizeof(data));
 959                 }
 960                 ret = put_user(tmp, datap);
 961                 break;
 962         }
 963
 964         case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
 965                 ret = -EIO;
 966                 if ((addr & (sizeof(data) - 1)) || addr < 0 ||
 967                     addr >= sizeof(struct user))
 968                         break;
 969
 970                 if (addr < sizeof(struct user_regs_struct))
 971                         ret = putreg(child, addr, data);
 972                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
 973                          addr <= offsetof(struct user, u_debugreg[7])) {
 974                         addr -= offsetof(struct user, u_debugreg[0]);
 975                         ret = ptrace_set_debugreg(child,
 976                                                   addr / sizeof(data), data);
 977                 }
 978                 break;
 979
 980         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
 981                 return copy_regset_to_user(child,
 982                                            task_user_regset_view(current),
 983                                            REGSET_GENERAL,
 984                                            0, sizeof(struct user_regs_struct),
 985                                            datap);
 986
 987         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
 988                 return copy_regset_from_user(child,
 989                                              task_user_regset_view(current),
 990                                              REGSET_GENERAL,
 991                                              0, sizeof(struct user_regs_struct),
 992                                              datap);
 993
 994         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
 995                 return copy_regset_to_user(child,
 996                                            task_user_regset_view(current),
 997                                            REGSET_FP,
 998                                            0, sizeof(struct user_i387_struct),
 999                                            datap);
1000
1001         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
1002                 return copy_regset_from_user(child,
1003                                              task_user_regset_view(current),
1004                                              REGSET_FP,
1005                                              0, sizeof(struct user_i387_struct),
1006                                              datap);
1007
1008 #ifdef CONFIG_X86_32
1009         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
1010                 return copy_regset_to_user(child, &user_x86_32_view,
1011                                            REGSET_XFP,
1012                                            0, sizeof(struct user_fxsr_struct),
1013                                            datap) ? -EIO : 0;
1014
1015         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
1016                 return copy_regset_from_user(child, &user_x86_32_view,
1017                                              REGSET_XFP,
1018                                              0, sizeof(struct user_fxsr_struct),
1019                                              datap) ? -EIO : 0;
1020 #endif
1021
1022 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1023         case PTRACE_GET_THREAD_AREA:
1024                 if (addr < 0)
1025                         return -EIO;
1026                 ret = do_get_thread_area(child, addr,
1027                                          (struct user_desc __user *) data);
1028                 break;
1029
1030         case PTRACE_SET_THREAD_AREA:
1031                 if (addr < 0)
1032                         return -EIO;
1033                 ret = do_set_thread_area(child, addr,
1034                                          (struct user_desc __user *) data, 0);
1035                 break;
1036 #endif
1037
1038 #ifdef CONFIG_X86_64
1039                 /* normal 64bit interface to access TLS data.
1040                    Works just like arch_prctl, except that the arguments
1041                    are reversed. */
1042         case PTRACE_ARCH_PRCTL:
1043                 ret = do_arch_prctl(child, data, addr);
1044                 break;
1045 #endif
1046
1047         /*
1048          * These bits need more cooking - not enabled yet:
1049          */
1050 #ifdef CONFIG_X86_PTRACE_BTS
1051         case PTRACE_BTS_CONFIG:
1052                 ret = ptrace_bts_config
1053                         (child, data, (struct ptrace_bts_config __user *)addr);
1054                 break;
1055
1056         case PTRACE_BTS_STATUS:
1057                 ret = ptrace_bts_status
1058                         (child, data, (struct ptrace_bts_config __user *)addr);
1059                 break;
1060
1061         case PTRACE_BTS_SIZE:
1062                 ret = ptrace_bts_size(child);
1063                 break;
1064
1065         case PTRACE_BTS_GET:
1066                 ret = ptrace_bts_read_record
1067                         (child, data, (struct bts_struct __user *) addr);
1068                 break;
1069
1070         case PTRACE_BTS_CLEAR:
1071                 ret = ptrace_bts_clear(child);
1072                 break;
1073
1074         case PTRACE_BTS_DRAIN:
1075                 ret = ptrace_bts_drain
1076                         (child, data, (struct bts_struct __user *) addr);
1077                 break;
1078 #endif /* CONFIG_X86_PTRACE_BTS */
1079
1080         default:
1081                 ret = ptrace_request(child, request, addr, data);
1082                 break;
1083         }
1084
1085         return ret;
1086 }
1087
1088 #ifdef CONFIG_IA32_EMULATION
1089
1090 #include <linux/compat.h>
1091 #include <linux/syscalls.h>
1092 #include <asm/ia32.h>
1093 #include <asm/user32.h>
1094
1095 #define R32(l,q)                                                        \
1096         case offsetof(struct user32, regs.l):                           \
1097                 regs->q = value; break
1098
1099 #define SEG32(rs)                                                       \
1100         case offsetof(struct user32, regs.rs):                          \
1101                 return set_segment_reg(child,                           \
1102                                        offsetof(struct user_regs_struct, rs), \
1103                                        value);                          \
1104                 break
1105
1106 static int putreg32(struct task_struct *child, unsigned regno, u32 value)
1107 {
1108         struct pt_regs *regs = task_pt_regs(child);
1109
1110         switch (regno) {
1111
1112         SEG32(cs);
1113         SEG32(ds);
1114         SEG32(es);
1115         SEG32(fs);
1116         SEG32(gs);
1117         SEG32(ss);
1118
1119         R32(ebx, bx);
1120         R32(ecx, cx);
1121         R32(edx, dx);
1122         R32(edi, di);
1123         R32(esi, si);
1124         R32(ebp, bp);
1125         R32(eax, ax);
1126         R32(eip, ip);
1127         R32(esp, sp);
1128
1129         case offsetof(struct user32, regs.orig_eax):
1130                 /*
1131                  * Sign-extend the value so that orig_eax = -1
1132                  * causes (long)orig_ax < 0 tests to fire correctly.
1133                  */
1134                 regs->orig_ax = (long) (s32) value;
1135                 break;
1136
1137         case offsetof(struct user32, regs.eflags):
1138                 return set_flags(child, value);
1139
1140         case offsetof(struct user32, u_debugreg[0]) ...
1141                 offsetof(struct user32, u_debugreg[7]):
1142                 regno -= offsetof(struct user32, u_debugreg[0]);
1143                 return ptrace_set_debugreg(child, regno / 4, value);
1144
1145         default:
1146                 if (regno > sizeof(struct user32) || (regno & 3))
1147                         return -EIO;
1148
1149                 /*
1150                  * Other dummy fields in the virtual user structure
1151                  * are ignored
1152                  */
1153                 break;
1154         }
1155         return 0;
1156 }
1157
1158 #undef R32
1159 #undef SEG32
1160
1161 #define R32(l,q)                                                        \
1162         case offsetof(struct user32, regs.l):                           \
1163                 *val = regs->q; break
1164
1165 #define SEG32(rs)                                                       \
1166         case offsetof(struct user32, regs.rs):                          \
1167                 *val = get_segment_reg(child,                           \
1168                                        offsetof(struct user_regs_struct, rs)); \
1169                 break
1170
1171 static int getreg32(struct task_struct *child, unsigned regno, u32 *val)
1172 {
1173         struct pt_regs *regs = task_pt_regs(child);
1174
1175         switch (regno) {
1176
1177         SEG32(ds);
1178         SEG32(es);
1179         SEG32(fs);
1180         SEG32(gs);
1181
1182         R32(cs, cs);
1183         R32(ss, ss);
1184         R32(ebx, bx);
1185         R32(ecx, cx);
1186         R32(edx, dx);
1187         R32(edi, di);
1188         R32(esi, si);
1189         R32(ebp, bp);
1190         R32(eax, ax);
1191         R32(orig_eax, orig_ax);
1192         R32(eip, ip);
1193         R32(esp, sp);
1194
1195         case offsetof(struct user32, regs.eflags):
1196                 *val = get_flags(child);
1197                 break;
1198
1199         case offsetof(struct user32, u_debugreg[0]) ...
1200                 offsetof(struct user32, u_debugreg[7]):
1201                 regno -= offsetof(struct user32, u_debugreg[0]);
1202                 *val = ptrace_get_debugreg(child, regno / 4);
1203                 break;
1204
1205         default:
1206                 if (regno > sizeof(struct user32) || (regno & 3))
1207                         return -EIO;
1208
1209                 /*
1210                  * Other dummy fields in the virtual user structure
1211                  * are ignored
1212                  */
1213                 *val = 0;
1214                 break;
1215         }
1216         return 0;
1217 }
1218
1219 #undef R32
1220 #undef SEG32
1221
1222 static int genregs32_get(struct task_struct *target,
1223                          const struct user_regset *regset,
1224                          unsigned int pos, unsigned int count,
1225                          void *kbuf, void __user *ubuf)
1226 {
1227         if (kbuf) {
1228                 compat_ulong_t *k = kbuf;
1229                 while (count > 0) {
1230                         getreg32(target, pos, k++);
1231                         count -= sizeof(*k);
1232                         pos += sizeof(*k);
1233                 }
1234         } else {
1235                 compat_ulong_t __user *u = ubuf;
1236                 while (count > 0) {
1237                         compat_ulong_t word;
1238                         getreg32(target, pos, &word);
1239                         if (__put_user(word, u++))
1240                                 return -EFAULT;
1241                         count -= sizeof(*u);
1242                         pos += sizeof(*u);
1243                 }
1244         }
1245
1246         return 0;
1247 }
1248
1249 static int genregs32_set(struct task_struct *target,
1250                          const struct user_regset *regset,
1251                          unsigned int pos, unsigned int count,
1252                          const void *kbuf, const void __user *ubuf)
1253 {
1254         int ret = 0;
1255         if (kbuf) {
1256                 const compat_ulong_t *k = kbuf;
1257                 while (count > 0 && !ret) {
1258                         ret = putreg32(target, pos, *k++);
1259                         count -= sizeof(*k);
1260                         pos += sizeof(*k);
1261                 }
1262         } else {
1263                 const compat_ulong_t __user *u = ubuf;
1264                 while (count > 0 && !ret) {
1265                         compat_ulong_t word;
1266                         ret = __get_user(word, u++);
1267                         if (ret)
1268                                 break;
1269                         ret = putreg32(target, pos, word);
1270                         count -= sizeof(*u);
1271                         pos += sizeof(*u);
1272                 }
1273         }
1274         return ret;
1275 }
1276
1277 long compat_arch_ptrace(struct task_struct *child, compat_long_t request,
1278                         compat_ulong_t caddr, compat_ulong_t cdata)
1279 {
1280         unsigned long addr = caddr;
1281         unsigned long data = cdata;
1282         void __user *datap = compat_ptr(data);
1283         int ret;
1284         __u32 val;
1285
1286         switch (request) {
1287         case PTRACE_PEEKUSR:
1288                 ret = getreg32(child, addr, &val);
1289                 if (ret == 0)
1290                         ret = put_user(val, (__u32 __user *)datap);
1291                 break;
1292
1293         case PTRACE_POKEUSR:
1294                 ret = putreg32(child, addr, data);
1295                 break;
1296
1297         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
1298                 return copy_regset_to_user(child, &user_x86_32_view,
1299                                            REGSET_GENERAL,
1300                                            0, sizeof(struct user_regs_struct32),
1301                                            datap);
1302
1303         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
1304                 return copy_regset_from_user(child, &user_x86_32_view,
1305                                              REGSET_GENERAL, 0,
1306                                              sizeof(struct user_regs_struct32),
1307                                              datap);
1308
1309         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
1310                 return copy_regset_to_user(child, &user_x86_32_view,
1311                                            REGSET_FP, 0,
1312                                            sizeof(struct user_i387_ia32_struct),
1313                                            datap);
1314
1315         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
1316                 return copy_regset_from_user(
1317                         child, &user_x86_32_view, REGSET_FP,
1318                         0, sizeof(struct user_i387_ia32_struct), datap);
1319
1320         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
1321                 return copy_regset_to_user(child, &user_x86_32_view,
1322                                            REGSET_XFP, 0,
1323                                            sizeof(struct user32_fxsr_struct),
1324                                            datap);
1325
1326         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
1327                 return copy_regset_from_user(child, &user_x86_32_view,
1328                                              REGSET_XFP, 0,
1329                                              sizeof(struct user32_fxsr_struct),
1330                                              datap);
1331
1332         case PTRACE_GET_THREAD_AREA:
1333         case PTRACE_SET_THREAD_AREA:
1334 #ifdef CONFIG_X86_PTRACE_BTS
1335         case PTRACE_BTS_CONFIG:
1336         case PTRACE_BTS_STATUS:
1337         case PTRACE_BTS_SIZE:
1338         case PTRACE_BTS_GET:
1339         case PTRACE_BTS_CLEAR:
1340         case PTRACE_BTS_DRAIN:
1341 #endif /* CONFIG_X86_PTRACE_BTS */
1342                 return arch_ptrace(child, request, addr, data);
1343
1344         default:
1345                 return compat_ptrace_request(child, request, addr, data);
1346         }
1347
1348         return ret;
1349 }
1350
1351 #endif  /* CONFIG_IA32_EMULATION */
1352
1353 #ifdef CONFIG_X86_64
1354
1355 static const struct user_regset x86_64_regsets[] = {
1356         [REGSET_GENERAL] = {
1357                 .core_note_type = NT_PRSTATUS,
1358                 .n = sizeof(struct user_regs_struct) / sizeof(long),
1359                 .size = sizeof(long), .align = sizeof(long),
1360                 .get = genregs_get, .set = genregs_set
1361         },
1362         [REGSET_FP] = {
1363                 .core_note_type = NT_PRFPREG,
1364                 .n = sizeof(struct user_i387_struct) / sizeof(long),
1365                 .size = sizeof(long), .align = sizeof(long),
1366                 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1367         },
1368         [REGSET_IOPERM64] = {
1369                 .core_note_type = NT_386_IOPERM,
1370                 .n = IO_BITMAP_LONGS,
1371                 .size = sizeof(long), .align = sizeof(long),
1372                 .active = ioperm_active, .get = ioperm_get
1373         },
1374 };
1375
1376 static const struct user_regset_view user_x86_64_view = {
1377         .name = "x86_64", .e_machine = EM_X86_64,
1378         .regsets = x86_64_regsets, .n = ARRAY_SIZE(x86_64_regsets)
1379 };
1380
1381 #else  /* CONFIG_X86_32 */
1382
1383 #define user_regs_struct32      user_regs_struct
1384 #define genregs32_get           genregs_get
1385 #define genregs32_set           genregs_set
1386
1387 #define user_i387_ia32_struct   user_i387_struct
1388 #define user32_fxsr_struct      user_fxsr_struct
1389
1390 #endif  /* CONFIG_X86_64 */
1391
1392 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1393 static const struct user_regset x86_32_regsets[] = {
1394         [REGSET_GENERAL] = {
1395                 .core_note_type = NT_PRSTATUS,
1396                 .n = sizeof(struct user_regs_struct32) / sizeof(u32),
1397                 .size = sizeof(u32), .align = sizeof(u32),
1398                 .get = genregs32_get, .set = genregs32_set
1399         },
1400         [REGSET_FP] = {
1401                 .core_note_type = NT_PRFPREG,
1402                 .n = sizeof(struct user_i387_ia32_struct) / sizeof(u32),
1403                 .size = sizeof(u32), .align = sizeof(u32),
1404                 .active = fpregs_active, .get = fpregs_get, .set = fpregs_set
1405         },
1406         [REGSET_XFP] = {
1407                 .core_note_type = NT_PRXFPREG,
1408                 .n = sizeof(struct user32_fxsr_struct) / sizeof(u32),
1409                 .size = sizeof(u32), .align = sizeof(u32),
1410                 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1411         },
1412         [REGSET_TLS] = {
1413                 .core_note_type = NT_386_TLS,
1414                 .n = GDT_ENTRY_TLS_ENTRIES, .bias = GDT_ENTRY_TLS_MIN,
1415                 .size = sizeof(struct user_desc),
1416                 .align = sizeof(struct user_desc),
1417                 .active = regset_tls_active,
1418                 .get = regset_tls_get, .set = regset_tls_set
1419         },
1420         [REGSET_IOPERM32] = {
1421                 .core_note_type = NT_386_IOPERM,
1422                 .n = IO_BITMAP_BYTES / sizeof(u32),
1423                 .size = sizeof(u32), .align = sizeof(u32),
1424                 .active = ioperm_active, .get = ioperm_get
1425         },
1426 };
1427
1428 static const struct user_regset_view user_x86_32_view = {
1429         .name = "i386", .e_machine = EM_386,
1430         .regsets = x86_32_regsets, .n = ARRAY_SIZE(x86_32_regsets)
1431 };
1432 #endif
1433
1434 const struct user_regset_view *task_user_regset_view(struct task_struct *task)
1435 {
1436 #ifdef CONFIG_IA32_EMULATION
1437         if (test_tsk_thread_flag(task, TIF_IA32))
1438 #endif
1439 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1440                 return &user_x86_32_view;
1441 #endif
1442 #ifdef CONFIG_X86_64
1443         return &user_x86_64_view;
1444 #endif
1445 }
1446
1447 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs,
1448                                          int error_code, int si_code)
1449 {
1450         struct siginfo info;
1451
1452         tsk->thread.trap_no = 1;
1453         tsk->thread.error_code = error_code;
1454
1455         memset(&info, 0, sizeof(info));
1456         info.si_signo = SIGTRAP;
1457         info.si_code = si_code;
1458
1459         /* User-mode ip? */
1460         info.si_addr = user_mode_vm(regs) ? (void __user *) regs->ip : NULL;
1461
1462         /* Send us the fake SIGTRAP */
1463         force_sig_info(SIGTRAP, &info, tsk);
1464 }
1465
1466
1467 #ifdef CONFIG_X86_32
1468 # define IS_IA32        1
1469 #elif defined CONFIG_IA32_EMULATION
1470 # define IS_IA32        is_compat_task()
1471 #else
1472 # define IS_IA32        0
1473 #endif
1474
1475 /*
1476  * We must return the syscall number to actually look up in the table.
1477  * This can be -1L to skip running any syscall at all.
1478  */
1479 asmregparm long syscall_trace_enter(struct pt_regs *regs)
1480 {
1481         long ret = 0;
1482
1483         /*
1484          * If we stepped into a sysenter/syscall insn, it trapped in
1485          * kernel mode; do_debug() cleared TF and set TIF_SINGLESTEP.
1486          * If user-mode had set TF itself, then it's still clear from
1487          * do_debug() and we need to set it again to restore the user
1488          * state.  If we entered on the slow path, TF was already set.
1489          */
1490         if (test_thread_flag(TIF_SINGLESTEP))
1491                 regs->flags |= X86_EFLAGS_TF;
1492
1493         /* do the secure computing check first */
1494         secure_computing(regs->orig_ax);
1495
1496         if (unlikely(test_thread_flag(TIF_SYSCALL_EMU)))
1497                 ret = -1L;
1498
1499         if ((ret || test_thread_flag(TIF_SYSCALL_TRACE)) &&
1500             tracehook_report_syscall_entry(regs))
1501                 ret = -1L;
1502
1503         if (unlikely(test_thread_flag(TIF_SYSCALL_FTRACE)))
1504                 ftrace_syscall_enter(regs);
1505
1506         if (unlikely(current->audit_context)) {
1507                 if (IS_IA32)
1508                         audit_syscall_entry(AUDIT_ARCH_I386,
1509                                             regs->orig_ax,
1510                                             regs->bx, regs->cx,
1511                                             regs->dx, regs->si);
1512 #ifdef CONFIG_X86_64
1513                 else
1514                         audit_syscall_entry(AUDIT_ARCH_X86_64,
1515                                             regs->orig_ax,
1516                                             regs->di, regs->si,
1517                                             regs->dx, regs->r10);
1518 #endif
1519         }
1520
1521         return ret ?: regs->orig_ax;
1522 }
1523
1524 asmregparm void syscall_trace_leave(struct pt_regs *regs)
1525 {
1526         if (unlikely(current->audit_context))
1527                 audit_syscall_exit(AUDITSC_RESULT(regs->ax), regs->ax);
1528
1529         if (unlikely(test_thread_flag(TIF_SYSCALL_FTRACE)))
1530                 ftrace_syscall_exit(regs);
1531
1532         if (test_thread_flag(TIF_SYSCALL_TRACE))
1533                 tracehook_report_syscall_exit(regs, 0);
1534
1535         /*
1536          * If TIF_SYSCALL_EMU is set, we only get here because of
1537          * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
1538          * We already reported this syscall instruction in
1539          * syscall_trace_enter(), so don't do any more now.
1540          */
1541         if (unlikely(test_thread_flag(TIF_SYSCALL_EMU)))
1542                 return;
1543
1544         /*
1545          * If we are single-stepping, synthesize a trap to follow the
1546          * system call instruction.
1547          */
1548         if (test_thread_flag(TIF_SINGLESTEP) &&
1549             tracehook_consider_fatal_signal(current, SIGTRAP))
1550                 send_sigtrap(current, regs, 0, TRAP_BRKPT);
1551 }