SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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/user.h>
  18 #include <linux/elf.h>
  19 #include <linux/security.h>
  20 #include <linux/audit.h>
  21 #include <linux/seccomp.h>
  22 #include <linux/signal.h>
  23
  24 #include <asm/uaccess.h>
  25 #include <asm/pgtable.h>
  26 #include <asm/system.h>
  27 #include <asm/processor.h>
  28 #include <asm/i387.h>
  29 #include <asm/debugreg.h>
  30 #include <asm/ldt.h>
  31 #include <asm/desc.h>
  32 #include <asm/prctl.h>
  33 #include <asm/proto.h>
  34 #include <asm/ds.h>
  35
  36 #include "tls.h"
  37
  38 enum x86_regset {
  39         REGSET_GENERAL,
  40         REGSET_FP,
  41         REGSET_XFP,
  42         REGSET_TLS,
  43 };
  44
  45 /*
  46  * does not yet catch signals sent when the child dies.
  47  * in exit.c or in signal.c.
  48  */
  49
  50 /*
  51  * Determines which flags the user has access to [1 = access, 0 = no access].
  52  */
  53 #define FLAG_MASK_32            ((unsigned long)                        \
  54                                  (X86_EFLAGS_CF | X86_EFLAGS_PF |       \
  55                                   X86_EFLAGS_AF | X86_EFLAGS_ZF |       \
  56                                   X86_EFLAGS_SF | X86_EFLAGS_TF |       \
  57                                   X86_EFLAGS_DF | X86_EFLAGS_OF |       \
  58                                   X86_EFLAGS_RF | X86_EFLAGS_AC))
  59
  60 /*
  61  * Determines whether a value may be installed in a segment register.
  62  */
  63 static inline bool invalid_selector(u16 value)
  64 {
  65         return unlikely(value != 0 && (value & SEGMENT_RPL_MASK) != USER_RPL);
  66 }
  67
  68 #ifdef CONFIG_X86_32
  69
  70 #define FLAG_MASK               FLAG_MASK_32
  71
  72 static long *pt_regs_access(struct pt_regs *regs, unsigned long regno)
  73 {
  74         BUILD_BUG_ON(offsetof(struct pt_regs, bx) != 0);
  75         regno >>= 2;
  76         if (regno > FS)
  77                 --regno;
  78         return &regs->bx + regno;
  79 }
  80
  81 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
  82 {
  83         /*
  84          * Returning the value truncates it to 16 bits.
  85          */
  86         unsigned int retval;
  87         if (offset != offsetof(struct user_regs_struct, gs))
  88                 retval = *pt_regs_access(task_pt_regs(task), offset);
  89         else {
  90                 retval = task->thread.gs;
  91                 if (task == current)
  92                         savesegment(gs, retval);
  93         }
  94         return retval;
  95 }
  96
  97 static int set_segment_reg(struct task_struct *task,
  98                            unsigned long offset, u16 value)
  99 {
 100         /*
 101          * The value argument was already truncated to 16 bits.
 102          */
 103         if (invalid_selector(value))
 104                 return -EIO;
 105
 106         /*
 107          * For %cs and %ss we cannot permit a null selector.
 108          * We can permit a bogus selector as long as it has USER_RPL.
 109          * Null selectors are fine for other segment registers, but
 110          * we will never get back to user mode with invalid %cs or %ss
 111          * and will take the trap in iret instead.  Much code relies
 112          * on user_mode() to distinguish a user trap frame (which can
 113          * safely use invalid selectors) from a kernel trap frame.
 114          */
 115         switch (offset) {
 116         case offsetof(struct user_regs_struct, cs):
 117         case offsetof(struct user_regs_struct, ss):
 118                 if (unlikely(value == 0))
 119                         return -EIO;
 120
 121         default:
 122                 *pt_regs_access(task_pt_regs(task), offset) = value;
 123                 break;
 124
 125         case offsetof(struct user_regs_struct, gs):
 126                 task->thread.gs = value;
 127                 if (task == current)
 128                         /*
 129                          * The user-mode %gs is not affected by
 130                          * kernel entry, so we must update the CPU.
 131                          */
 132                         loadsegment(gs, 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_SIZE64 - 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         case offsetof(struct user_regs_struct,fs_base):
 327                 if (value >= TASK_SIZE_OF(child))
 328                         return -EIO;
 329                 /*
 330                  * When changing the segment base, use do_arch_prctl
 331                  * to set either thread.fs or thread.fsindex and the
 332                  * corresponding GDT slot.
 333                  */
 334                 if (child->thread.fs != value)
 335                         return do_arch_prctl(child, ARCH_SET_FS, value);
 336                 return 0;
 337         case offsetof(struct user_regs_struct,gs_base):
 338                 /*
 339                  * Exactly the same here as the %fs handling above.
 340                  */
 341                 if (value >= TASK_SIZE_OF(child))
 342                         return -EIO;
 343                 if (child->thread.gs != value)
 344                         return do_arch_prctl(child, ARCH_SET_GS, value);
 345                 return 0;
 346 #endif
 347         }
 348
 349         *pt_regs_access(task_pt_regs(child), offset) = value;
 350         return 0;
 351 }
 352
 353 static unsigned long getreg(struct task_struct *task, unsigned long offset)
 354 {
 355         switch (offset) {
 356         case offsetof(struct user_regs_struct, cs):
 357         case offsetof(struct user_regs_struct, ds):
 358         case offsetof(struct user_regs_struct, es):
 359         case offsetof(struct user_regs_struct, fs):
 360         case offsetof(struct user_regs_struct, gs):
 361         case offsetof(struct user_regs_struct, ss):
 362                 return get_segment_reg(task, offset);
 363
 364         case offsetof(struct user_regs_struct, flags):
 365                 return get_flags(task);
 366
 367 #ifdef CONFIG_X86_64
 368         case offsetof(struct user_regs_struct, fs_base): {
 369                 /*
 370                  * do_arch_prctl may have used a GDT slot instead of
 371                  * the MSR.  To userland, it appears the same either
 372                  * way, except the %fs segment selector might not be 0.
 373                  */
 374                 unsigned int seg = task->thread.fsindex;
 375                 if (task->thread.fs != 0)
 376                         return task->thread.fs;
 377                 if (task == current)
 378                         asm("movl %%fs,%0" : "=r" (seg));
 379                 if (seg != FS_TLS_SEL)
 380                         return 0;
 381                 return get_desc_base(&task->thread.tls_array[FS_TLS]);
 382         }
 383         case offsetof(struct user_regs_struct, gs_base): {
 384                 /*
 385                  * Exactly the same here as the %fs handling above.
 386                  */
 387                 unsigned int seg = task->thread.gsindex;
 388                 if (task->thread.gs != 0)
 389                         return task->thread.gs;
 390                 if (task == current)
 391                         asm("movl %%gs,%0" : "=r" (seg));
 392                 if (seg != GS_TLS_SEL)
 393                         return 0;
 394                 return get_desc_base(&task->thread.tls_array[GS_TLS]);
 395         }
 396 #endif
 397         }
 398
 399         return *pt_regs_access(task_pt_regs(task), offset);
 400 }
 401
 402 static int genregs_get(struct task_struct *target,
 403                        const struct user_regset *regset,
 404                        unsigned int pos, unsigned int count,
 405                        void *kbuf, void __user *ubuf)
 406 {
 407         if (kbuf) {
 408                 unsigned long *k = kbuf;
 409                 while (count > 0) {
 410                         *k++ = getreg(target, pos);
 411                         count -= sizeof(*k);
 412                         pos += sizeof(*k);
 413                 }
 414         } else {
 415                 unsigned long __user *u = ubuf;
 416                 while (count > 0) {
 417                         if (__put_user(getreg(target, pos), u++))
 418                                 return -EFAULT;
 419                         count -= sizeof(*u);
 420                         pos += sizeof(*u);
 421                 }
 422         }
 423
 424         return 0;
 425 }
 426
 427 static int genregs_set(struct task_struct *target,
 428                        const struct user_regset *regset,
 429                        unsigned int pos, unsigned int count,
 430                        const void *kbuf, const void __user *ubuf)
 431 {
 432         int ret = 0;
 433         if (kbuf) {
 434                 const unsigned long *k = kbuf;
 435                 while (count > 0 && !ret) {
 436                         ret = putreg(target, pos, *k++);
 437                         count -= sizeof(*k);
 438                         pos += sizeof(*k);
 439                 }
 440         } else {
 441                 const unsigned long  __user *u = ubuf;
 442                 while (count > 0 && !ret) {
 443                         unsigned long word;
 444                         ret = __get_user(word, u++);
 445                         if (ret)
 446                                 break;
 447                         ret = putreg(target, pos, word);
 448                         count -= sizeof(*u);
 449                         pos += sizeof(*u);
 450                 }
 451         }
 452         return ret;
 453 }
 454
 455 /*
 456  * This function is trivial and will be inlined by the compiler.
 457  * Having it separates the implementation details of debug
 458  * registers from the interface details of ptrace.
 459  */
 460 static unsigned long ptrace_get_debugreg(struct task_struct *child, int n)
 461 {
 462         switch (n) {
 463         case 0:         return child->thread.debugreg0;
 464         case 1:         return child->thread.debugreg1;
 465         case 2:         return child->thread.debugreg2;
 466         case 3:         return child->thread.debugreg3;
 467         case 6:         return child->thread.debugreg6;
 468         case 7:         return child->thread.debugreg7;
 469         }
 470         return 0;
 471 }
 472
 473 static int ptrace_set_debugreg(struct task_struct *child,
 474                                int n, unsigned long data)
 475 {
 476         int i;
 477
 478         if (unlikely(n == 4 || n == 5))
 479                 return -EIO;
 480
 481         if (n < 4 && unlikely(data >= debugreg_addr_limit(child)))
 482                 return -EIO;
 483
 484         switch (n) {
 485         case 0:         child->thread.debugreg0 = data; break;
 486         case 1:         child->thread.debugreg1 = data; break;
 487         case 2:         child->thread.debugreg2 = data; break;
 488         case 3:         child->thread.debugreg3 = data; break;
 489
 490         case 6:
 491                 if ((data & ~0xffffffffUL) != 0)
 492                         return -EIO;
 493                 child->thread.debugreg6 = data;
 494                 break;
 495
 496         case 7:
 497                 /*
 498                  * Sanity-check data. Take one half-byte at once with
 499                  * check = (val >> (16 + 4*i)) & 0xf. It contains the
 500                  * R/Wi and LENi bits; bits 0 and 1 are R/Wi, and bits
 501                  * 2 and 3 are LENi. Given a list of invalid values,
 502                  * we do mask |= 1 << invalid_value, so that
 503                  * (mask >> check) & 1 is a correct test for invalid
 504                  * values.
 505                  *
 506                  * R/Wi contains the type of the breakpoint /
 507                  * watchpoint, LENi contains the length of the watched
 508                  * data in the watchpoint case.
 509                  *
 510                  * The invalid values are:
 511                  * - LENi == 0x10 (undefined), so mask |= 0x0f00.       [32-bit]
 512                  * - R/Wi == 0x10 (break on I/O reads or writes), so
 513                  *   mask |= 0x4444.
 514                  * - R/Wi == 0x00 && LENi != 0x00, so we have mask |=
 515                  *   0x1110.
 516                  *
 517                  * Finally, mask = 0x0f00 | 0x4444 | 0x1110 == 0x5f54.
 518                  *
 519                  * See the Intel Manual "System Programming Guide",
 520                  * 15.2.4
 521                  *
 522                  * Note that LENi == 0x10 is defined on x86_64 in long
 523                  * mode (i.e. even for 32-bit userspace software, but
 524                  * 64-bit kernel), so the x86_64 mask value is 0x5454.
 525                  * See the AMD manual no. 24593 (AMD64 System Programming)
 526                  */
 527 #ifdef CONFIG_X86_32
 528 #define DR7_MASK        0x5f54
 529 #else
 530 #define DR7_MASK        0x5554
 531 #endif
 532                 data &= ~DR_CONTROL_RESERVED;
 533                 for (i = 0; i < 4; i++)
 534                         if ((DR7_MASK >> ((data >> (16 + 4*i)) & 0xf)) & 1)
 535                                 return -EIO;
 536                 child->thread.debugreg7 = data;
 537                 if (data)
 538                         set_tsk_thread_flag(child, TIF_DEBUG);
 539                 else
 540                         clear_tsk_thread_flag(child, TIF_DEBUG);
 541                 break;
 542         }
 543
 544         return 0;
 545 }
 546
 547 #ifdef X86_BTS
 548
 549 static int ptrace_bts_get_size(struct task_struct *child)
 550 {
 551         if (!child->thread.ds_area_msr)
 552                 return -ENXIO;
 553
 554         return ds_get_bts_index((void *)child->thread.ds_area_msr);
 555 }
 556
 557 static int ptrace_bts_read_record(struct task_struct *child,
 558                                   long index,
 559                                   struct bts_struct __user *out)
 560 {
 561         struct bts_struct ret;
 562         int retval;
 563         int bts_end;
 564         int bts_index;
 565
 566         if (!child->thread.ds_area_msr)
 567                 return -ENXIO;
 568
 569         if (index < 0)
 570                 return -EINVAL;
 571
 572         bts_end = ds_get_bts_end((void *)child->thread.ds_area_msr);
 573         if (bts_end <= index)
 574                 return -EINVAL;
 575
 576         /* translate the ptrace bts index into the ds bts index */
 577         bts_index = ds_get_bts_index((void *)child->thread.ds_area_msr);
 578         bts_index -= (index + 1);
 579         if (bts_index < 0)
 580                 bts_index += bts_end;
 581
 582         retval = ds_read_bts((void *)child->thread.ds_area_msr,
 583                              bts_index, &ret);
 584         if (retval < 0)
 585                 return retval;
 586
 587         if (copy_to_user(out, &ret, sizeof(ret)))
 588                 return -EFAULT;
 589
 590         return sizeof(ret);
 591 }
 592
 593 static int ptrace_bts_write_record(struct task_struct *child,
 594                                    const struct bts_struct *in)
 595 {
 596         int retval;
 597
 598         if (!child->thread.ds_area_msr)
 599                 return -ENXIO;
 600
 601         retval = ds_write_bts((void *)child->thread.ds_area_msr, in);
 602         if (retval)
 603                 return retval;
 604
 605         return sizeof(*in);
 606 }
 607
 608 static int ptrace_bts_clear(struct task_struct *child)
 609 {
 610         if (!child->thread.ds_area_msr)
 611                 return -ENXIO;
 612
 613         return ds_clear((void *)child->thread.ds_area_msr);
 614 }
 615
 616 static int ptrace_bts_drain(struct task_struct *child,
 617                             long size,
 618                             struct bts_struct __user *out)
 619 {
 620         int end, i;
 621         void *ds = (void *)child->thread.ds_area_msr;
 622
 623         if (!ds)
 624                 return -ENXIO;
 625
 626         end = ds_get_bts_index(ds);
 627         if (end <= 0)
 628                 return end;
 629
 630         if (size < (end * sizeof(struct bts_struct)))
 631                 return -EIO;
 632
 633         for (i = 0; i < end; i++, out++) {
 634                 struct bts_struct ret;
 635                 int retval;
 636
 637                 retval = ds_read_bts(ds, i, &ret);
 638                 if (retval < 0)
 639                         return retval;
 640
 641                 if (copy_to_user(out, &ret, sizeof(ret)))
 642                         return -EFAULT;
 643         }
 644
 645         ds_clear(ds);
 646
 647         return end;
 648 }
 649
 650 static int ptrace_bts_realloc(struct task_struct *child,
 651                               int size, int reduce_size)
 652 {
 653         unsigned long rlim, vm;
 654         int ret, old_size;
 655
 656         if (size < 0)
 657                 return -EINVAL;
 658
 659         old_size = ds_get_bts_size((void *)child->thread.ds_area_msr);
 660         if (old_size < 0)
 661                 return old_size;
 662
 663         ret = ds_free((void **)&child->thread.ds_area_msr);
 664         if (ret < 0)
 665                 goto out;
 666
 667         size >>= PAGE_SHIFT;
 668         old_size >>= PAGE_SHIFT;
 669
 670         current->mm->total_vm  -= old_size;
 671         current->mm->locked_vm -= old_size;
 672
 673         if (size == 0)
 674                 goto out;
 675
 676         rlim = current->signal->rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT;
 677         vm = current->mm->total_vm  + size;
 678         if (rlim < vm) {
 679                 ret = -ENOMEM;
 680
 681                 if (!reduce_size)
 682                         goto out;
 683
 684                 size = rlim - current->mm->total_vm;
 685                 if (size <= 0)
 686                         goto out;
 687         }
 688
 689         rlim = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT;
 690         vm = current->mm->locked_vm  + size;
 691         if (rlim < vm) {
 692                 ret = -ENOMEM;
 693
 694                 if (!reduce_size)
 695                         goto out;
 696
 697                 size = rlim - current->mm->locked_vm;
 698                 if (size <= 0)
 699                         goto out;
 700         }
 701
 702         ret = ds_allocate((void **)&child->thread.ds_area_msr,
 703                           size << PAGE_SHIFT);
 704         if (ret < 0)
 705                 goto out;
 706
 707         current->mm->total_vm  += size;
 708         current->mm->locked_vm += size;
 709
 710 out:
 711         if (child->thread.ds_area_msr)
 712                 set_tsk_thread_flag(child, TIF_DS_AREA_MSR);
 713         else
 714                 clear_tsk_thread_flag(child, TIF_DS_AREA_MSR);
 715
 716         return ret;
 717 }
 718
 719 static int ptrace_bts_config(struct task_struct *child,
 720                              long cfg_size,
 721                              const struct ptrace_bts_config __user *ucfg)
 722 {
 723         struct ptrace_bts_config cfg;
 724         int bts_size, ret = 0;
 725         void *ds;
 726
 727         if (cfg_size < sizeof(cfg))
 728                 return -EIO;
 729
 730         if (copy_from_user(&cfg, ucfg, sizeof(cfg)))
 731                 return -EFAULT;
 732
 733         if ((int)cfg.size < 0)
 734                 return -EINVAL;
 735
 736         bts_size = 0;
 737         ds = (void *)child->thread.ds_area_msr;
 738         if (ds) {
 739                 bts_size = ds_get_bts_size(ds);
 740                 if (bts_size < 0)
 741                         return bts_size;
 742         }
 743         cfg.size = PAGE_ALIGN(cfg.size);
 744
 745         if (bts_size != cfg.size) {
 746                 ret = ptrace_bts_realloc(child, cfg.size,
 747                                          cfg.flags & PTRACE_BTS_O_CUT_SIZE);
 748                 if (ret < 0)
 749                         goto errout;
 750
 751                 ds = (void *)child->thread.ds_area_msr;
 752         }
 753
 754         if (cfg.flags & PTRACE_BTS_O_SIGNAL)
 755                 ret = ds_set_overflow(ds, DS_O_SIGNAL);
 756         else
 757                 ret = ds_set_overflow(ds, DS_O_WRAP);
 758         if (ret < 0)
 759                 goto errout;
 760
 761         if (cfg.flags & PTRACE_BTS_O_TRACE)
 762                 child->thread.debugctlmsr |= ds_debugctl_mask();
 763         else
 764                 child->thread.debugctlmsr &= ~ds_debugctl_mask();
 765
 766         if (cfg.flags & PTRACE_BTS_O_SCHED)
 767                 set_tsk_thread_flag(child, TIF_BTS_TRACE_TS);
 768         else
 769                 clear_tsk_thread_flag(child, TIF_BTS_TRACE_TS);
 770
 771         ret = sizeof(cfg);
 772
 773 out:
 774         if (child->thread.debugctlmsr)
 775                 set_tsk_thread_flag(child, TIF_DEBUGCTLMSR);
 776         else
 777                 clear_tsk_thread_flag(child, TIF_DEBUGCTLMSR);
 778
 779         return ret;
 780
 781 errout:
 782         child->thread.debugctlmsr &= ~ds_debugctl_mask();
 783         clear_tsk_thread_flag(child, TIF_BTS_TRACE_TS);
 784         goto out;
 785 }
 786
 787 static int ptrace_bts_status(struct task_struct *child,
 788                              long cfg_size,
 789                              struct ptrace_bts_config __user *ucfg)
 790 {
 791         void *ds = (void *)child->thread.ds_area_msr;
 792         struct ptrace_bts_config cfg;
 793
 794         if (cfg_size < sizeof(cfg))
 795                 return -EIO;
 796
 797         memset(&cfg, 0, sizeof(cfg));
 798
 799         if (ds) {
 800                 cfg.size = ds_get_bts_size(ds);
 801
 802                 if (ds_get_overflow(ds) == DS_O_SIGNAL)
 803                         cfg.flags |= PTRACE_BTS_O_SIGNAL;
 804
 805                 if (test_tsk_thread_flag(child, TIF_DEBUGCTLMSR) &&
 806                     child->thread.debugctlmsr & ds_debugctl_mask())
 807                         cfg.flags |= PTRACE_BTS_O_TRACE;
 808
 809                 if (test_tsk_thread_flag(child, TIF_BTS_TRACE_TS))
 810                         cfg.flags |= PTRACE_BTS_O_SCHED;
 811         }
 812
 813         cfg.bts_size = sizeof(struct bts_struct);
 814
 815         if (copy_to_user(ucfg, &cfg, sizeof(cfg)))
 816                 return -EFAULT;
 817
 818         return sizeof(cfg);
 819 }
 820
 821 void ptrace_bts_take_timestamp(struct task_struct *tsk,
 822                                enum bts_qualifier qualifier)
 823 {
 824         struct bts_struct rec = {
 825                 .qualifier = qualifier,
 826                 .variant.jiffies = jiffies_64
 827         };
 828
 829         ptrace_bts_write_record(tsk, &rec);
 830 }
 831 #endif /* X86_BTS */
 832
 833 /*
 834  * Called by kernel/ptrace.c when detaching..
 835  *
 836  * Make sure the single step bit is not set.
 837  */
 838 void ptrace_disable(struct task_struct *child)
 839 {
 840         user_disable_single_step(child);
 841 #ifdef TIF_SYSCALL_EMU
 842         clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
 843 #endif
 844         if (child->thread.ds_area_msr) {
 845 #ifdef X86_BTS
 846                 ptrace_bts_realloc(child, 0, 0);
 847 #endif
 848                 child->thread.debugctlmsr &= ~ds_debugctl_mask();
 849                 if (!child->thread.debugctlmsr)
 850                         clear_tsk_thread_flag(child, TIF_DEBUGCTLMSR);
 851                 clear_tsk_thread_flag(child, TIF_BTS_TRACE_TS);
 852         }
 853 }
 854
 855 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
 856 static const struct user_regset_view user_x86_32_view; /* Initialized below. */
 857 #endif
 858
 859 long arch_ptrace(struct task_struct *child, long request, long addr, long data)
 860 {
 861         int ret;
 862         unsigned long __user *datap = (unsigned long __user *)data;
 863
 864         switch (request) {
 865         /* read the word at location addr in the USER area. */
 866         case PTRACE_PEEKUSR: {
 867                 unsigned long tmp;
 868
 869                 ret = -EIO;
 870                 if ((addr & (sizeof(data) - 1)) || addr < 0 ||
 871                     addr >= sizeof(struct user))
 872                         break;
 873
 874                 tmp = 0;  /* Default return condition */
 875                 if (addr < sizeof(struct user_regs_struct))
 876                         tmp = getreg(child, addr);
 877                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
 878                          addr <= offsetof(struct user, u_debugreg[7])) {
 879                         addr -= offsetof(struct user, u_debugreg[0]);
 880                         tmp = ptrace_get_debugreg(child, addr / sizeof(data));
 881                 }
 882                 ret = put_user(tmp, datap);
 883                 break;
 884         }
 885
 886         case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
 887                 ret = -EIO;
 888                 if ((addr & (sizeof(data) - 1)) || addr < 0 ||
 889                     addr >= sizeof(struct user))
 890                         break;
 891
 892                 if (addr < sizeof(struct user_regs_struct))
 893                         ret = putreg(child, addr, data);
 894                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
 895                          addr <= offsetof(struct user, u_debugreg[7])) {
 896                         addr -= offsetof(struct user, u_debugreg[0]);
 897                         ret = ptrace_set_debugreg(child,
 898                                                   addr / sizeof(data), data);
 899                 }
 900                 break;
 901
 902         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
 903                 return copy_regset_to_user(child,
 904                                            task_user_regset_view(current),
 905                                            REGSET_GENERAL,
 906                                            0, sizeof(struct user_regs_struct),
 907                                            datap);
 908
 909         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
 910                 return copy_regset_from_user(child,
 911                                              task_user_regset_view(current),
 912                                              REGSET_GENERAL,
 913                                              0, sizeof(struct user_regs_struct),
 914                                              datap);
 915
 916         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
 917                 return copy_regset_to_user(child,
 918                                            task_user_regset_view(current),
 919                                            REGSET_FP,
 920                                            0, sizeof(struct user_i387_struct),
 921                                            datap);
 922
 923         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
 924                 return copy_regset_from_user(child,
 925                                              task_user_regset_view(current),
 926                                              REGSET_FP,
 927                                              0, sizeof(struct user_i387_struct),
 928                                              datap);
 929
 930 #ifdef CONFIG_X86_32
 931         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
 932                 return copy_regset_to_user(child, &user_x86_32_view,
 933                                            REGSET_XFP,
 934                                            0, sizeof(struct user_fxsr_struct),
 935                                            datap);
 936
 937         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
 938                 return copy_regset_from_user(child, &user_x86_32_view,
 939                                              REGSET_XFP,
 940                                              0, sizeof(struct user_fxsr_struct),
 941                                              datap);
 942 #endif
 943
 944 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
 945         case PTRACE_GET_THREAD_AREA:
 946                 if (addr < 0)
 947                         return -EIO;
 948                 ret = do_get_thread_area(child, addr,
 949                                          (struct user_desc __user *) data);
 950                 break;
 951
 952         case PTRACE_SET_THREAD_AREA:
 953                 if (addr < 0)
 954                         return -EIO;
 955                 ret = do_set_thread_area(child, addr,
 956                                          (struct user_desc __user *) data, 0);
 957                 break;
 958 #endif
 959
 960 #ifdef CONFIG_X86_64
 961                 /* normal 64bit interface to access TLS data.
 962                    Works just like arch_prctl, except that the arguments
 963                    are reversed. */
 964         case PTRACE_ARCH_PRCTL:
 965                 ret = do_arch_prctl(child, data, addr);
 966                 break;
 967 #endif
 968
 969         /*
 970          * These bits need more cooking - not enabled yet:
 971          */
 972 #ifdef X86_BTS
 973         case PTRACE_BTS_CONFIG:
 974                 ret = ptrace_bts_config
 975                         (child, data, (struct ptrace_bts_config __user *)addr);
 976                 break;
 977
 978         case PTRACE_BTS_STATUS:
 979                 ret = ptrace_bts_status
 980                         (child, data, (struct ptrace_bts_config __user *)addr);
 981                 break;
 982
 983         case PTRACE_BTS_SIZE:
 984                 ret = ptrace_bts_get_size(child);
 985                 break;
 986
 987         case PTRACE_BTS_GET:
 988                 ret = ptrace_bts_read_record
 989                         (child, data, (struct bts_struct __user *) addr);
 990                 break;
 991
 992         case PTRACE_BTS_CLEAR:
 993                 ret = ptrace_bts_clear(child);
 994                 break;
 995
 996         case PTRACE_BTS_DRAIN:
 997                 ret = ptrace_bts_drain
 998                         (child, data, (struct bts_struct __user *) addr);
 999                 break;
1000 #endif
1001
1002         default:
1003                 ret = ptrace_request(child, request, addr, data);
1004                 break;
1005         }
1006
1007         return ret;
1008 }
1009
1010 #ifdef CONFIG_IA32_EMULATION
1011
1012 #include <linux/compat.h>
1013 #include <linux/syscalls.h>
1014 #include <asm/ia32.h>
1015 #include <asm/user32.h>
1016
1017 #define R32(l,q)                                                        \
1018         case offsetof(struct user32, regs.l):                           \
1019                 regs->q = value; break
1020
1021 #define SEG32(rs)                                                       \
1022         case offsetof(struct user32, regs.rs):                          \
1023                 return set_segment_reg(child,                           \
1024                                        offsetof(struct user_regs_struct, rs), \
1025                                        value);                          \
1026                 break
1027
1028 static int putreg32(struct task_struct *child, unsigned regno, u32 value)
1029 {
1030         struct pt_regs *regs = task_pt_regs(child);
1031
1032         switch (regno) {
1033
1034         SEG32(cs);
1035         SEG32(ds);
1036         SEG32(es);
1037         SEG32(fs);
1038         SEG32(gs);
1039         SEG32(ss);
1040
1041         R32(ebx, bx);
1042         R32(ecx, cx);
1043         R32(edx, dx);
1044         R32(edi, di);
1045         R32(esi, si);
1046         R32(ebp, bp);
1047         R32(eax, ax);
1048         R32(orig_eax, orig_ax);
1049         R32(eip, ip);
1050         R32(esp, sp);
1051
1052         case offsetof(struct user32, regs.eflags):
1053                 return set_flags(child, value);
1054
1055         case offsetof(struct user32, u_debugreg[0]) ...
1056                 offsetof(struct user32, u_debugreg[7]):
1057                 regno -= offsetof(struct user32, u_debugreg[0]);
1058                 return ptrace_set_debugreg(child, regno / 4, value);
1059
1060         default:
1061                 if (regno > sizeof(struct user32) || (regno & 3))
1062                         return -EIO;
1063
1064                 /*
1065                  * Other dummy fields in the virtual user structure
1066                  * are ignored
1067                  */
1068                 break;
1069         }
1070         return 0;
1071 }
1072
1073 #undef R32
1074 #undef SEG32
1075
1076 #define R32(l,q)                                                        \
1077         case offsetof(struct user32, regs.l):                           \
1078                 *val = regs->q; break
1079
1080 #define SEG32(rs)                                                       \
1081         case offsetof(struct user32, regs.rs):                          \
1082                 *val = get_segment_reg(child,                           \
1083                                        offsetof(struct user_regs_struct, rs)); \
1084                 break
1085
1086 static int getreg32(struct task_struct *child, unsigned regno, u32 *val)
1087 {
1088         struct pt_regs *regs = task_pt_regs(child);
1089
1090         switch (regno) {
1091
1092         SEG32(ds);
1093         SEG32(es);
1094         SEG32(fs);
1095         SEG32(gs);
1096
1097         R32(cs, cs);
1098         R32(ss, ss);
1099         R32(ebx, bx);
1100         R32(ecx, cx);
1101         R32(edx, dx);
1102         R32(edi, di);
1103         R32(esi, si);
1104         R32(ebp, bp);
1105         R32(eax, ax);
1106         R32(orig_eax, orig_ax);
1107         R32(eip, ip);
1108         R32(esp, sp);
1109
1110         case offsetof(struct user32, regs.eflags):
1111                 *val = get_flags(child);
1112                 break;
1113
1114         case offsetof(struct user32, u_debugreg[0]) ...
1115                 offsetof(struct user32, u_debugreg[7]):
1116                 regno -= offsetof(struct user32, u_debugreg[0]);
1117                 *val = ptrace_get_debugreg(child, regno / 4);
1118                 break;
1119
1120         default:
1121                 if (regno > sizeof(struct user32) || (regno & 3))
1122                         return -EIO;
1123
1124                 /*
1125                  * Other dummy fields in the virtual user structure
1126                  * are ignored
1127                  */
1128                 *val = 0;
1129                 break;
1130         }
1131         return 0;
1132 }
1133
1134 #undef R32
1135 #undef SEG32
1136
1137 static int genregs32_get(struct task_struct *target,
1138                          const struct user_regset *regset,
1139                          unsigned int pos, unsigned int count,
1140                          void *kbuf, void __user *ubuf)
1141 {
1142         if (kbuf) {
1143                 compat_ulong_t *k = kbuf;
1144                 while (count > 0) {
1145                         getreg32(target, pos, k++);
1146                         count -= sizeof(*k);
1147                         pos += sizeof(*k);
1148                 }
1149         } else {
1150                 compat_ulong_t __user *u = ubuf;
1151                 while (count > 0) {
1152                         compat_ulong_t word;
1153                         getreg32(target, pos, &word);
1154                         if (__put_user(word, u++))
1155                                 return -EFAULT;
1156                         count -= sizeof(*u);
1157                         pos += sizeof(*u);
1158                 }
1159         }
1160
1161         return 0;
1162 }
1163
1164 static int genregs32_set(struct task_struct *target,
1165                          const struct user_regset *regset,
1166                          unsigned int pos, unsigned int count,
1167                          const void *kbuf, const void __user *ubuf)
1168 {
1169         int ret = 0;
1170         if (kbuf) {
1171                 const compat_ulong_t *k = kbuf;
1172                 while (count > 0 && !ret) {
1173                         ret = putreg32(target, pos, *k++);
1174                         count -= sizeof(*k);
1175                         pos += sizeof(*k);
1176                 }
1177         } else {
1178                 const compat_ulong_t __user *u = ubuf;
1179                 while (count > 0 && !ret) {
1180                         compat_ulong_t word;
1181                         ret = __get_user(word, u++);
1182                         if (ret)
1183                                 break;
1184                         ret = putreg32(target, pos, word);
1185                         count -= sizeof(*u);
1186                         pos += sizeof(*u);
1187                 }
1188         }
1189         return ret;
1190 }
1191
1192 static long ptrace32_siginfo(unsigned request, u32 pid, u32 addr, u32 data)
1193 {
1194         siginfo_t __user *si = compat_alloc_user_space(sizeof(siginfo_t));
1195         compat_siginfo_t __user *si32 = compat_ptr(data);
1196         siginfo_t ssi;
1197         int ret;
1198
1199         if (request == PTRACE_SETSIGINFO) {
1200                 memset(&ssi, 0, sizeof(siginfo_t));
1201                 ret = copy_siginfo_from_user32(&ssi, si32);
1202                 if (ret)
1203                         return ret;
1204                 if (copy_to_user(si, &ssi, sizeof(siginfo_t)))
1205                         return -EFAULT;
1206         }
1207         ret = sys_ptrace(request, pid, addr, (unsigned long)si);
1208         if (ret)
1209                 return ret;
1210         if (request == PTRACE_GETSIGINFO) {
1211                 if (copy_from_user(&ssi, si, sizeof(siginfo_t)))
1212                         return -EFAULT;
1213                 ret = copy_siginfo_to_user32(si32, &ssi);
1214         }
1215         return ret;
1216 }
1217
1218 asmlinkage long sys32_ptrace(long request, u32 pid, u32 addr, u32 data)
1219 {
1220         struct task_struct *child;
1221         struct pt_regs *childregs;
1222         void __user *datap = compat_ptr(data);
1223         int ret;
1224         __u32 val;
1225
1226         switch (request) {
1227         case PTRACE_TRACEME:
1228         case PTRACE_ATTACH:
1229         case PTRACE_KILL:
1230         case PTRACE_CONT:
1231         case PTRACE_SINGLESTEP:
1232         case PTRACE_SINGLEBLOCK:
1233         case PTRACE_DETACH:
1234         case PTRACE_SYSCALL:
1235         case PTRACE_OLDSETOPTIONS:
1236         case PTRACE_SETOPTIONS:
1237         case PTRACE_SET_THREAD_AREA:
1238         case PTRACE_GET_THREAD_AREA:
1239 #ifdef X86_BTS
1240         case PTRACE_BTS_CONFIG:
1241         case PTRACE_BTS_STATUS:
1242         case PTRACE_BTS_SIZE:
1243         case PTRACE_BTS_GET:
1244         case PTRACE_BTS_CLEAR:
1245         case PTRACE_BTS_DRAIN:
1246 #endif
1247                 return sys_ptrace(request, pid, addr, data);
1248
1249         default:
1250                 return -EINVAL;
1251
1252         case PTRACE_PEEKTEXT:
1253         case PTRACE_PEEKDATA:
1254         case PTRACE_POKEDATA:
1255         case PTRACE_POKETEXT:
1256         case PTRACE_POKEUSR:
1257         case PTRACE_PEEKUSR:
1258         case PTRACE_GETREGS:
1259         case PTRACE_SETREGS:
1260         case PTRACE_SETFPREGS:
1261         case PTRACE_GETFPREGS:
1262         case PTRACE_SETFPXREGS:
1263         case PTRACE_GETFPXREGS:
1264         case PTRACE_GETEVENTMSG:
1265                 break;
1266
1267         case PTRACE_SETSIGINFO:
1268         case PTRACE_GETSIGINFO:
1269                 return ptrace32_siginfo(request, pid, addr, data);
1270         }
1271
1272         child = ptrace_get_task_struct(pid);
1273         if (IS_ERR(child))
1274                 return PTR_ERR(child);
1275
1276         ret = ptrace_check_attach(child, request == PTRACE_KILL);
1277         if (ret < 0)
1278                 goto out;
1279
1280         childregs = task_pt_regs(child);
1281
1282         switch (request) {
1283         case PTRACE_PEEKUSR:
1284                 ret = getreg32(child, addr, &val);
1285                 if (ret == 0)
1286                         ret = put_user(val, (__u32 __user *)datap);
1287                 break;
1288
1289         case PTRACE_POKEUSR:
1290                 ret = putreg32(child, addr, data);
1291                 break;
1292
1293         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
1294                 return copy_regset_to_user(child, &user_x86_32_view,
1295                                            REGSET_GENERAL,
1296                                            0, sizeof(struct user_regs_struct32),
1297                                            datap);
1298
1299         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
1300                 return copy_regset_from_user(child, &user_x86_32_view,
1301                                              REGSET_GENERAL, 0,
1302                                              sizeof(struct user_regs_struct32),
1303                                              datap);
1304
1305         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
1306                 return copy_regset_to_user(child, &user_x86_32_view,
1307                                            REGSET_FP, 0,
1308                                            sizeof(struct user_i387_ia32_struct),
1309                                            datap);
1310
1311         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
1312                 return copy_regset_from_user(
1313                         child, &user_x86_32_view, REGSET_FP,
1314                         0, sizeof(struct user_i387_ia32_struct), datap);
1315
1316         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
1317                 return copy_regset_to_user(child, &user_x86_32_view,
1318                                            REGSET_XFP, 0,
1319                                            sizeof(struct user32_fxsr_struct),
1320                                            datap);
1321
1322         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
1323                 return copy_regset_from_user(child, &user_x86_32_view,
1324                                              REGSET_XFP, 0,
1325                                              sizeof(struct user32_fxsr_struct),
1326                                              datap);
1327
1328         default:
1329                 return compat_ptrace_request(child, request, addr, data);
1330         }
1331
1332  out:
1333         put_task_struct(child);
1334         return ret;
1335 }
1336
1337 #endif  /* CONFIG_IA32_EMULATION */
1338
1339 #ifdef CONFIG_X86_64
1340
1341 static const struct user_regset x86_64_regsets[] = {
1342         [REGSET_GENERAL] = {
1343                 .core_note_type = NT_PRSTATUS,
1344                 .n = sizeof(struct user_regs_struct) / sizeof(long),
1345                 .size = sizeof(long), .align = sizeof(long),
1346                 .get = genregs_get, .set = genregs_set
1347         },
1348         [REGSET_FP] = {
1349                 .core_note_type = NT_PRFPREG,
1350                 .n = sizeof(struct user_i387_struct) / sizeof(long),
1351                 .size = sizeof(long), .align = sizeof(long),
1352                 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1353         },
1354 };
1355
1356 static const struct user_regset_view user_x86_64_view = {
1357         .name = "x86_64", .e_machine = EM_X86_64,
1358         .regsets = x86_64_regsets, .n = ARRAY_SIZE(x86_64_regsets)
1359 };
1360
1361 #else  /* CONFIG_X86_32 */
1362
1363 #define user_regs_struct32      user_regs_struct
1364 #define genregs32_get           genregs_get
1365 #define genregs32_set           genregs_set
1366
1367 #endif  /* CONFIG_X86_64 */
1368
1369 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1370 static const struct user_regset x86_32_regsets[] = {
1371         [REGSET_GENERAL] = {
1372                 .core_note_type = NT_PRSTATUS,
1373                 .n = sizeof(struct user_regs_struct32) / sizeof(u32),
1374                 .size = sizeof(u32), .align = sizeof(u32),
1375                 .get = genregs32_get, .set = genregs32_set
1376         },
1377         [REGSET_FP] = {
1378                 .core_note_type = NT_PRFPREG,
1379                 .n = sizeof(struct user_i387_struct) / sizeof(u32),
1380                 .size = sizeof(u32), .align = sizeof(u32),
1381                 .active = fpregs_active, .get = fpregs_get, .set = fpregs_set
1382         },
1383         [REGSET_XFP] = {
1384                 .core_note_type = NT_PRXFPREG,
1385                 .n = sizeof(struct user_i387_struct) / sizeof(u32),
1386                 .size = sizeof(u32), .align = sizeof(u32),
1387                 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1388         },
1389         [REGSET_TLS] = {
1390                 .core_note_type = NT_386_TLS,
1391                 .n = GDT_ENTRY_TLS_ENTRIES, .bias = GDT_ENTRY_TLS_MIN,
1392                 .size = sizeof(struct user_desc),
1393                 .align = sizeof(struct user_desc),
1394                 .active = regset_tls_active,
1395                 .get = regset_tls_get, .set = regset_tls_set
1396         },
1397 };
1398
1399 static const struct user_regset_view user_x86_32_view = {
1400         .name = "i386", .e_machine = EM_386,
1401         .regsets = x86_32_regsets, .n = ARRAY_SIZE(x86_32_regsets)
1402 };
1403 #endif
1404
1405 const struct user_regset_view *task_user_regset_view(struct task_struct *task)
1406 {
1407 #ifdef CONFIG_IA32_EMULATION
1408         if (test_tsk_thread_flag(task, TIF_IA32))
1409 #endif
1410 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1411                 return &user_x86_32_view;
1412 #endif
1413 #ifdef CONFIG_X86_64
1414         return &user_x86_64_view;
1415 #endif
1416 }
1417
1418 #ifdef CONFIG_X86_32
1419
1420 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs, int error_code)
1421 {
1422         struct siginfo info;
1423
1424         tsk->thread.trap_no = 1;
1425         tsk->thread.error_code = error_code;
1426
1427         memset(&info, 0, sizeof(info));
1428         info.si_signo = SIGTRAP;
1429         info.si_code = TRAP_BRKPT;
1430
1431         /* User-mode ip? */
1432         info.si_addr = user_mode_vm(regs) ? (void __user *) regs->ip : NULL;
1433
1434         /* Send us the fake SIGTRAP */
1435         force_sig_info(SIGTRAP, &info, tsk);
1436 }
1437
1438 /* notification of system call entry/exit
1439  * - triggered by current->work.syscall_trace
1440  */
1441 __attribute__((regparm(3)))
1442 int do_syscall_trace(struct pt_regs *regs, int entryexit)
1443 {
1444         int is_sysemu = test_thread_flag(TIF_SYSCALL_EMU);
1445         /*
1446          * With TIF_SYSCALL_EMU set we want to ignore TIF_SINGLESTEP for syscall
1447          * interception
1448          */
1449         int is_singlestep = !is_sysemu && test_thread_flag(TIF_SINGLESTEP);
1450         int ret = 0;
1451
1452         /* do the secure computing check first */
1453         if (!entryexit)
1454                 secure_computing(regs->orig_ax);
1455
1456         if (unlikely(current->audit_context)) {
1457                 if (entryexit)
1458                         audit_syscall_exit(AUDITSC_RESULT(regs->ax),
1459                                                 regs->ax);
1460                 /* Debug traps, when using PTRACE_SINGLESTEP, must be sent only
1461                  * on the syscall exit path. Normally, when TIF_SYSCALL_AUDIT is
1462                  * not used, entry.S will call us only on syscall exit, not
1463                  * entry; so when TIF_SYSCALL_AUDIT is used we must avoid
1464                  * calling send_sigtrap() on syscall entry.
1465                  *
1466                  * Note that when PTRACE_SYSEMU_SINGLESTEP is used,
1467                  * is_singlestep is false, despite his name, so we will still do
1468                  * the correct thing.
1469                  */
1470                 else if (is_singlestep)
1471                         goto out;
1472         }
1473
1474         if (!(current->ptrace & PT_PTRACED))
1475                 goto out;
1476
1477         /* If a process stops on the 1st tracepoint with SYSCALL_TRACE
1478          * and then is resumed with SYSEMU_SINGLESTEP, it will come in
1479          * here. We have to check this and return */
1480         if (is_sysemu && entryexit)
1481                 return 0;
1482
1483         /* Fake a debug trap */
1484         if (is_singlestep)
1485                 send_sigtrap(current, regs, 0);
1486
1487         if (!test_thread_flag(TIF_SYSCALL_TRACE) && !is_sysemu)
1488                 goto out;
1489
1490         /* the 0x80 provides a way for the tracing parent to distinguish
1491            between a syscall stop and SIGTRAP delivery */
1492         /* Note that the debugger could change the result of test_thread_flag!*/
1493         ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD) ? 0x80:0));
1494
1495         /*
1496          * this isn't the same as continuing with a signal, but it will do
1497          * for normal use.  strace only continues with a signal if the
1498          * stopping signal is not SIGTRAP.  -brl
1499          */
1500         if (current->exit_code) {
1501                 send_sig(current->exit_code, current, 1);
1502                 current->exit_code = 0;
1503         }
1504         ret = is_sysemu;
1505 out:
1506         if (unlikely(current->audit_context) && !entryexit)
1507                 audit_syscall_entry(AUDIT_ARCH_I386, regs->orig_ax,
1508                                     regs->bx, regs->cx, regs->dx, regs->si);
1509         if (ret == 0)
1510                 return 0;
1511
1512         regs->orig_ax = -1; /* force skip of syscall restarting */
1513         if (unlikely(current->audit_context))
1514                 audit_syscall_exit(AUDITSC_RESULT(regs->ax), regs->ax);
1515         return 1;
1516 }
1517
1518 #else  /* CONFIG_X86_64 */
1519
1520 static void syscall_trace(struct pt_regs *regs)
1521 {
1522
1523 #if 0
1524         printk("trace %s ip %lx sp %lx ax %d origrax %d caller %lx tiflags %x ptrace %x\n",
1525                current->comm,
1526                regs->ip, regs->sp, regs->ax, regs->orig_ax, __builtin_return_address(0),
1527                current_thread_info()->flags, current->ptrace);
1528 #endif
1529
1530         ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD)
1531                                 ? 0x80 : 0));
1532         /*
1533          * this isn't the same as continuing with a signal, but it will do
1534          * for normal use.  strace only continues with a signal if the
1535          * stopping signal is not SIGTRAP.  -brl
1536          */
1537         if (current->exit_code) {
1538                 send_sig(current->exit_code, current, 1);
1539                 current->exit_code = 0;
1540         }
1541 }
1542
1543 asmlinkage void syscall_trace_enter(struct pt_regs *regs)
1544 {
1545         /* do the secure computing check first */
1546         secure_computing(regs->orig_ax);
1547
1548         if (test_thread_flag(TIF_SYSCALL_TRACE)
1549             && (current->ptrace & PT_PTRACED))
1550                 syscall_trace(regs);
1551
1552         if (unlikely(current->audit_context)) {
1553                 if (test_thread_flag(TIF_IA32)) {
1554                         audit_syscall_entry(AUDIT_ARCH_I386,
1555                                             regs->orig_ax,
1556                                             regs->bx, regs->cx,
1557                                             regs->dx, regs->si);
1558                 } else {
1559                         audit_syscall_entry(AUDIT_ARCH_X86_64,
1560                                             regs->orig_ax,
1561                                             regs->di, regs->si,
1562                                             regs->dx, regs->r10);
1563                 }
1564         }
1565 }
1566
1567 asmlinkage void syscall_trace_leave(struct pt_regs *regs)
1568 {
1569         if (unlikely(current->audit_context))
1570                 audit_syscall_exit(AUDITSC_RESULT(regs->ax), regs->ax);
1571
1572         if ((test_thread_flag(TIF_SYSCALL_TRACE)
1573              || test_thread_flag(TIF_SINGLESTEP))
1574             && (current->ptrace & PT_PTRACED))
1575                 syscall_trace(regs);
1576 }
1577
1578 #endif  /* CONFIG_X86_32 */