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
   7 #include <linux/kernel.h>
   8 #include <linux/sched.h>
   9 #include <linux/mm.h>
  10 #include <linux/smp.h>
  11 #include <linux/errno.h>
  12 #include <linux/slab.h>
  13 #include <linux/ptrace.h>
  14 #include <linux/regset.h>
  15 #include <linux/tracehook.h>
  16 #include <linux/user.h>
  17 #include <linux/elf.h>
  18 #include <linux/security.h>
  19 #include <linux/audit.h>
  20 #include <linux/seccomp.h>
  21 #include <linux/signal.h>
  22 #include <linux/perf_event.h>
  23 #include <linux/hw_breakpoint.h>
  24
  25 #include <asm/uaccess.h>
  26 #include <asm/pgtable.h>
  27 #include <asm/processor.h>
  28 #include <asm/i387.h>
  29 #include <asm/fpu-internal.h>
  30 #include <asm/debugreg.h>
  31 #include <asm/ldt.h>
  32 #include <asm/desc.h>
  33 #include <asm/prctl.h>
  34 #include <asm/proto.h>
  35 #include <asm/hw_breakpoint.h>
  36 #include <asm/traps.h>
  37
  38 #include "tls.h"
  39
  40 #define CREATE_TRACE_POINTS
  41 #include <trace/events/syscalls.h>
  42
  43 enum x86_regset {
  44         REGSET_GENERAL,
  45         REGSET_FP,
  46         REGSET_XFP,
  47         REGSET_IOPERM64 = REGSET_XFP,
  48         REGSET_XSTATE,
  49         REGSET_TLS,
  50         REGSET_IOPERM32,
  51 };
  52
  53 struct pt_regs_offset {
  54         const char *name;
  55         int offset;
  56 };
  57
  58 #define REG_OFFSET_NAME(r) {.name = #r, .offset = offsetof(struct pt_regs, r)}
  59 #define REG_OFFSET_END {.name = NULL, .offset = 0}
  60
  61 static const struct pt_regs_offset regoffset_table[] = {
  62 #ifdef CONFIG_X86_64
  63         REG_OFFSET_NAME(r15),
  64         REG_OFFSET_NAME(r14),
  65         REG_OFFSET_NAME(r13),
  66         REG_OFFSET_NAME(r12),
  67         REG_OFFSET_NAME(r11),
  68         REG_OFFSET_NAME(r10),
  69         REG_OFFSET_NAME(r9),
  70         REG_OFFSET_NAME(r8),
  71 #endif
  72         REG_OFFSET_NAME(bx),
  73         REG_OFFSET_NAME(cx),
  74         REG_OFFSET_NAME(dx),
  75         REG_OFFSET_NAME(si),
  76         REG_OFFSET_NAME(di),
  77         REG_OFFSET_NAME(bp),
  78         REG_OFFSET_NAME(ax),
  79 #ifdef CONFIG_X86_32
  80         REG_OFFSET_NAME(ds),
  81         REG_OFFSET_NAME(es),
  82         REG_OFFSET_NAME(fs),
  83         REG_OFFSET_NAME(gs),
  84 #endif
  85         REG_OFFSET_NAME(orig_ax),
  86         REG_OFFSET_NAME(ip),
  87         REG_OFFSET_NAME(cs),
  88         REG_OFFSET_NAME(flags),
  89         REG_OFFSET_NAME(sp),
  90         REG_OFFSET_NAME(ss),
  91         REG_OFFSET_END,
  92 };
  93
  94 /**
  95  * regs_query_register_offset() - query register offset from its name
  96  * @name:       the name of a register
  97  *
  98  * regs_query_register_offset() returns the offset of a register in struct
  99  * pt_regs from its name. If the name is invalid, this returns -EINVAL;
 100  */
 101 int regs_query_register_offset(const char *name)
 102 {
 103         const struct pt_regs_offset *roff;
 104         for (roff = regoffset_table; roff->name != NULL; roff++)
 105                 if (!strcmp(roff->name, name))
 106                         return roff->offset;
 107         return -EINVAL;
 108 }
 109
 110 /**
 111  * regs_query_register_name() - query register name from its offset
 112  * @offset:     the offset of a register in struct pt_regs.
 113  *
 114  * regs_query_register_name() returns the name of a register from its
 115  * offset in struct pt_regs. If the @offset is invalid, this returns NULL;
 116  */
 117 const char *regs_query_register_name(unsigned int offset)
 118 {
 119         const struct pt_regs_offset *roff;
 120         for (roff = regoffset_table; roff->name != NULL; roff++)
 121                 if (roff->offset == offset)
 122                         return roff->name;
 123         return NULL;
 124 }
 125
 126 static const int arg_offs_table[] = {
 127 #ifdef CONFIG_X86_32
 128         [0] = offsetof(struct pt_regs, ax),
 129         [1] = offsetof(struct pt_regs, dx),
 130         [2] = offsetof(struct pt_regs, cx)
 131 #else /* CONFIG_X86_64 */
 132         [0] = offsetof(struct pt_regs, di),
 133         [1] = offsetof(struct pt_regs, si),
 134         [2] = offsetof(struct pt_regs, dx),
 135         [3] = offsetof(struct pt_regs, cx),
 136         [4] = offsetof(struct pt_regs, r8),
 137         [5] = offsetof(struct pt_regs, r9)
 138 #endif
 139 };
 140
 141 /*
 142  * does not yet catch signals sent when the child dies.
 143  * in exit.c or in signal.c.
 144  */
 145
 146 /*
 147  * Determines which flags the user has access to [1 = access, 0 = no access].
 148  */
 149 #define FLAG_MASK_32            ((unsigned long)                        \
 150                                  (X86_EFLAGS_CF | X86_EFLAGS_PF |       \
 151                                   X86_EFLAGS_AF | X86_EFLAGS_ZF |       \
 152                                   X86_EFLAGS_SF | X86_EFLAGS_TF |       \
 153                                   X86_EFLAGS_DF | X86_EFLAGS_OF |       \
 154                                   X86_EFLAGS_RF | X86_EFLAGS_AC))
 155
 156 /*
 157  * Determines whether a value may be installed in a segment register.
 158  */
 159 static inline bool invalid_selector(u16 value)
 160 {
 161         return unlikely(value != 0 && (value & SEGMENT_RPL_MASK) != USER_RPL);
 162 }
 163
 164 #ifdef CONFIG_X86_32
 165
 166 #define FLAG_MASK               FLAG_MASK_32
 167
 168 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long regno)
 169 {
 170         BUILD_BUG_ON(offsetof(struct pt_regs, bx) != 0);
 171         return &regs->bx + (regno >> 2);
 172 }
 173
 174 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
 175 {
 176         /*
 177          * Returning the value truncates it to 16 bits.
 178          */
 179         unsigned int retval;
 180         if (offset != offsetof(struct user_regs_struct, gs))
 181                 retval = *pt_regs_access(task_pt_regs(task), offset);
 182         else {
 183                 if (task == current)
 184                         retval = get_user_gs(task_pt_regs(task));
 185                 else
 186                         retval = task_user_gs(task);
 187         }
 188         return retval;
 189 }
 190
 191 static int set_segment_reg(struct task_struct *task,
 192                            unsigned long offset, u16 value)
 193 {
 194         /*
 195          * The value argument was already truncated to 16 bits.
 196          */
 197         if (invalid_selector(value))
 198                 return -EIO;
 199
 200         /*
 201          * For %cs and %ss we cannot permit a null selector.
 202          * We can permit a bogus selector as long as it has USER_RPL.
 203          * Null selectors are fine for other segment registers, but
 204          * we will never get back to user mode with invalid %cs or %ss
 205          * and will take the trap in iret instead.  Much code relies
 206          * on user_mode() to distinguish a user trap frame (which can
 207          * safely use invalid selectors) from a kernel trap frame.
 208          */
 209         switch (offset) {
 210         case offsetof(struct user_regs_struct, cs):
 211         case offsetof(struct user_regs_struct, ss):
 212                 if (unlikely(value == 0))
 213                         return -EIO;
 214
 215         default:
 216                 *pt_regs_access(task_pt_regs(task), offset) = value;
 217                 break;
 218
 219         case offsetof(struct user_regs_struct, gs):
 220                 if (task == current)
 221                         set_user_gs(task_pt_regs(task), value);
 222                 else
 223                         task_user_gs(task) = value;
 224         }
 225
 226         return 0;
 227 }
 228
 229 #else  /* CONFIG_X86_64 */
 230
 231 #define FLAG_MASK               (FLAG_MASK_32 | X86_EFLAGS_NT)
 232
 233 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long offset)
 234 {
 235         BUILD_BUG_ON(offsetof(struct pt_regs, r15) != 0);
 236         return &regs->r15 + (offset / sizeof(regs->r15));
 237 }
 238
 239 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
 240 {
 241         /*
 242          * Returning the value truncates it to 16 bits.
 243          */
 244         unsigned int seg;
 245
 246         switch (offset) {
 247         case offsetof(struct user_regs_struct, fs):
 248                 if (task == current) {
 249                         /* Older gas can't assemble movq %?s,%r?? */
 250                         asm("movl %%fs,%0" : "=r" (seg));
 251                         return seg;
 252                 }
 253                 return task->thread.fsindex;
 254         case offsetof(struct user_regs_struct, gs):
 255                 if (task == current) {
 256                         asm("movl %%gs,%0" : "=r" (seg));
 257                         return seg;
 258                 }
 259                 return task->thread.gsindex;
 260         case offsetof(struct user_regs_struct, ds):
 261                 if (task == current) {
 262                         asm("movl %%ds,%0" : "=r" (seg));
 263                         return seg;
 264                 }
 265                 return task->thread.ds;
 266         case offsetof(struct user_regs_struct, es):
 267                 if (task == current) {
 268                         asm("movl %%es,%0" : "=r" (seg));
 269                         return seg;
 270                 }
 271                 return task->thread.es;
 272
 273         case offsetof(struct user_regs_struct, cs):
 274         case offsetof(struct user_regs_struct, ss):
 275                 break;
 276         }
 277         return *pt_regs_access(task_pt_regs(task), offset);
 278 }
 279
 280 static int set_segment_reg(struct task_struct *task,
 281                            unsigned long offset, u16 value)
 282 {
 283         /*
 284          * The value argument was already truncated to 16 bits.
 285          */
 286         if (invalid_selector(value))
 287                 return -EIO;
 288
 289         switch (offset) {
 290         case offsetof(struct user_regs_struct,fs):
 291                 /*
 292                  * If this is setting fs as for normal 64-bit use but
 293                  * setting fs_base has implicitly changed it, leave it.
 294                  */
 295                 if ((value == FS_TLS_SEL && task->thread.fsindex == 0 &&
 296                      task->thread.fs != 0) ||
 297                     (value == 0 && task->thread.fsindex == FS_TLS_SEL &&
 298                      task->thread.fs == 0))
 299                         break;
 300                 task->thread.fsindex = value;
 301                 if (task == current)
 302                         loadsegment(fs, task->thread.fsindex);
 303                 break;
 304         case offsetof(struct user_regs_struct,gs):
 305                 /*
 306                  * If this is setting gs as for normal 64-bit use but
 307                  * setting gs_base has implicitly changed it, leave it.
 308                  */
 309                 if ((value == GS_TLS_SEL && task->thread.gsindex == 0 &&
 310                      task->thread.gs != 0) ||
 311                     (value == 0 && task->thread.gsindex == GS_TLS_SEL &&
 312                      task->thread.gs == 0))
 313                         break;
 314                 task->thread.gsindex = value;
 315                 if (task == current)
 316                         load_gs_index(task->thread.gsindex);
 317                 break;
 318         case offsetof(struct user_regs_struct,ds):
 319                 task->thread.ds = value;
 320                 if (task == current)
 321                         loadsegment(ds, task->thread.ds);
 322                 break;
 323         case offsetof(struct user_regs_struct,es):
 324                 task->thread.es = value;
 325                 if (task == current)
 326                         loadsegment(es, task->thread.es);
 327                 break;
 328
 329                 /*
 330                  * Can't actually change these in 64-bit mode.
 331                  */
 332         case offsetof(struct user_regs_struct,cs):
 333                 if (unlikely(value == 0))
 334                         return -EIO;
 335 #ifdef CONFIG_IA32_EMULATION
 336                 if (test_tsk_thread_flag(task, TIF_IA32))
 337                         task_pt_regs(task)->cs = value;
 338 #endif
 339                 break;
 340         case offsetof(struct user_regs_struct,ss):
 341                 if (unlikely(value == 0))
 342                         return -EIO;
 343 #ifdef CONFIG_IA32_EMULATION
 344                 if (test_tsk_thread_flag(task, TIF_IA32))
 345                         task_pt_regs(task)->ss = value;
 346 #endif
 347                 break;
 348         }
 349
 350         return 0;
 351 }
 352
 353 #endif  /* CONFIG_X86_32 */
 354
 355 static unsigned long get_flags(struct task_struct *task)
 356 {
 357         unsigned long retval = task_pt_regs(task)->flags;
 358
 359         /*
 360          * If the debugger set TF, hide it from the readout.
 361          */
 362         if (test_tsk_thread_flag(task, TIF_FORCED_TF))
 363                 retval &= ~X86_EFLAGS_TF;
 364
 365         return retval;
 366 }
 367
 368 static int set_flags(struct task_struct *task, unsigned long value)
 369 {
 370         struct pt_regs *regs = task_pt_regs(task);
 371
 372         /*
 373          * If the user value contains TF, mark that
 374          * it was not "us" (the debugger) that set it.
 375          * If not, make sure it stays set if we had.
 376          */
 377         if (value & X86_EFLAGS_TF)
 378                 clear_tsk_thread_flag(task, TIF_FORCED_TF);
 379         else if (test_tsk_thread_flag(task, TIF_FORCED_TF))
 380                 value |= X86_EFLAGS_TF;
 381
 382         regs->flags = (regs->flags & ~FLAG_MASK) | (value & FLAG_MASK);
 383
 384         return 0;
 385 }
 386
 387 static int putreg(struct task_struct *child,
 388                   unsigned long offset, unsigned long value)
 389 {
 390         switch (offset) {
 391         case offsetof(struct user_regs_struct, cs):
 392         case offsetof(struct user_regs_struct, ds):
 393         case offsetof(struct user_regs_struct, es):
 394         case offsetof(struct user_regs_struct, fs):
 395         case offsetof(struct user_regs_struct, gs):
 396         case offsetof(struct user_regs_struct, ss):
 397                 return set_segment_reg(child, offset, value);
 398
 399         case offsetof(struct user_regs_struct, flags):
 400                 return set_flags(child, value);
 401
 402 #ifdef CONFIG_X86_64
 403         case offsetof(struct user_regs_struct,fs_base):
 404                 if (value >= TASK_SIZE_OF(child))
 405                         return -EIO;
 406                 /*
 407                  * When changing the segment base, use do_arch_prctl
 408                  * to set either thread.fs or thread.fsindex and the
 409                  * corresponding GDT slot.
 410                  */
 411                 if (child->thread.fs != value)
 412                         return do_arch_prctl(child, ARCH_SET_FS, value);
 413                 return 0;
 414         case offsetof(struct user_regs_struct,gs_base):
 415                 /*
 416                  * Exactly the same here as the %fs handling above.
 417                  */
 418                 if (value >= TASK_SIZE_OF(child))
 419                         return -EIO;
 420                 if (child->thread.gs != value)
 421                         return do_arch_prctl(child, ARCH_SET_GS, value);
 422                 return 0;
 423 #endif
 424         }
 425
 426         *pt_regs_access(task_pt_regs(child), offset) = value;
 427         return 0;
 428 }
 429
 430 static unsigned long getreg(struct task_struct *task, unsigned long offset)
 431 {
 432         switch (offset) {
 433         case offsetof(struct user_regs_struct, cs):
 434         case offsetof(struct user_regs_struct, ds):
 435         case offsetof(struct user_regs_struct, es):
 436         case offsetof(struct user_regs_struct, fs):
 437         case offsetof(struct user_regs_struct, gs):
 438         case offsetof(struct user_regs_struct, ss):
 439                 return get_segment_reg(task, offset);
 440
 441         case offsetof(struct user_regs_struct, flags):
 442                 return get_flags(task);
 443
 444 #ifdef CONFIG_X86_64
 445         case offsetof(struct user_regs_struct, fs_base): {
 446                 /*
 447                  * do_arch_prctl may have used a GDT slot instead of
 448                  * the MSR.  To userland, it appears the same either
 449                  * way, except the %fs segment selector might not be 0.
 450                  */
 451                 unsigned int seg = task->thread.fsindex;
 452                 if (task->thread.fs != 0)
 453                         return task->thread.fs;
 454                 if (task == current)
 455                         asm("movl %%fs,%0" : "=r" (seg));
 456                 if (seg != FS_TLS_SEL)
 457                         return 0;
 458                 return get_desc_base(&task->thread.tls_array[FS_TLS]);
 459         }
 460         case offsetof(struct user_regs_struct, gs_base): {
 461                 /*
 462                  * Exactly the same here as the %fs handling above.
 463                  */
 464                 unsigned int seg = task->thread.gsindex;
 465                 if (task->thread.gs != 0)
 466                         return task->thread.gs;
 467                 if (task == current)
 468                         asm("movl %%gs,%0" : "=r" (seg));
 469                 if (seg != GS_TLS_SEL)
 470                         return 0;
 471                 return get_desc_base(&task->thread.tls_array[GS_TLS]);
 472         }
 473 #endif
 474         }
 475
 476         return *pt_regs_access(task_pt_regs(task), offset);
 477 }
 478
 479 static int genregs_get(struct task_struct *target,
 480                        const struct user_regset *regset,
 481                        unsigned int pos, unsigned int count,
 482                        void *kbuf, void __user *ubuf)
 483 {
 484         if (kbuf) {
 485                 unsigned long *k = kbuf;
 486                 while (count >= sizeof(*k)) {
 487                         *k++ = getreg(target, pos);
 488                         count -= sizeof(*k);
 489                         pos += sizeof(*k);
 490                 }
 491         } else {
 492                 unsigned long __user *u = ubuf;
 493                 while (count >= sizeof(*u)) {
 494                         if (__put_user(getreg(target, pos), u++))
 495                                 return -EFAULT;
 496                         count -= sizeof(*u);
 497                         pos += sizeof(*u);
 498                 }
 499         }
 500
 501         return 0;
 502 }
 503
 504 static int genregs_set(struct task_struct *target,
 505                        const struct user_regset *regset,
 506                        unsigned int pos, unsigned int count,
 507                        const void *kbuf, const void __user *ubuf)
 508 {
 509         int ret = 0;
 510         if (kbuf) {
 511                 const unsigned long *k = kbuf;
 512                 while (count >= sizeof(*k) && !ret) {
 513                         ret = putreg(target, pos, *k++);
 514                         count -= sizeof(*k);
 515                         pos += sizeof(*k);
 516                 }
 517         } else {
 518                 const unsigned long  __user *u = ubuf;
 519                 while (count >= sizeof(*u) && !ret) {
 520                         unsigned long word;
 521                         ret = __get_user(word, u++);
 522                         if (ret)
 523                                 break;
 524                         ret = putreg(target, pos, word);
 525                         count -= sizeof(*u);
 526                         pos += sizeof(*u);
 527                 }
 528         }
 529         return ret;
 530 }
 531
 532 static void ptrace_triggered(struct perf_event *bp,
 533                              struct perf_sample_data *data,
 534                              struct pt_regs *regs)
 535 {
 536         int i;
 537         struct thread_struct *thread = &(current->thread);
 538
 539         /*
 540          * Store in the virtual DR6 register the fact that the breakpoint
 541          * was hit so the thread's debugger will see it.
 542          */
 543         for (i = 0; i < HBP_NUM; i++) {
 544                 if (thread->ptrace_bps[i] == bp)
 545                         break;
 546         }
 547
 548         thread->debugreg6 |= (DR_TRAP0 << i);
 549 }
 550
 551 /*
 552  * Walk through every ptrace breakpoints for this thread and
 553  * build the dr7 value on top of their attributes.
 554  *
 555  */
 556 static unsigned long ptrace_get_dr7(struct perf_event *bp[])
 557 {
 558         int i;
 559         int dr7 = 0;
 560         struct arch_hw_breakpoint *info;
 561
 562         for (i = 0; i < HBP_NUM; i++) {
 563                 if (bp[i] && !bp[i]->attr.disabled) {
 564                         info = counter_arch_bp(bp[i]);
 565                         dr7 |= encode_dr7(i, info->len, info->type);
 566                 }
 567         }
 568
 569         return dr7;
 570 }
 571
 572 static int
 573 ptrace_modify_breakpoint(struct perf_event *bp, int len, int type,
 574                          struct task_struct *tsk, int disabled)
 575 {
 576         int err;
 577         int gen_len, gen_type;
 578         struct perf_event_attr attr;
 579
 580         /*
 581          * We should have at least an inactive breakpoint at this
 582          * slot. It means the user is writing dr7 without having
 583          * written the address register first
 584          */
 585         if (!bp)
 586                 return -EINVAL;
 587
 588         err = arch_bp_generic_fields(len, type, &gen_len, &gen_type);
 589         if (err)
 590                 return err;
 591
 592         attr = bp->attr;
 593         attr.bp_len = gen_len;
 594         attr.bp_type = gen_type;
 595         attr.disabled = disabled;
 596
 597         return modify_user_hw_breakpoint(bp, &attr);
 598 }
 599
 600 /*
 601  * Handle ptrace writes to debug register 7.
 602  */
 603 static int ptrace_write_dr7(struct task_struct *tsk, unsigned long data)
 604 {
 605         struct thread_struct *thread = &(tsk->thread);
 606         unsigned long old_dr7;
 607         int i, orig_ret = 0, rc = 0;
 608         int enabled, second_pass = 0;
 609         unsigned len, type;
 610         struct perf_event *bp;
 611
 612         if (ptrace_get_breakpoints(tsk) < 0)
 613                 return -ESRCH;
 614
 615         data &= ~DR_CONTROL_RESERVED;
 616         old_dr7 = ptrace_get_dr7(thread->ptrace_bps);
 617 restore:
 618         /*
 619          * Loop through all the hardware breakpoints, making the
 620          * appropriate changes to each.
 621          */
 622         for (i = 0; i < HBP_NUM; i++) {
 623                 enabled = decode_dr7(data, i, &len, &type);
 624                 bp = thread->ptrace_bps[i];
 625
 626                 if (!enabled) {
 627                         if (bp) {
 628                                 /*
 629                                  * Don't unregister the breakpoints right-away,
 630                                  * unless all register_user_hw_breakpoint()
 631                                  * requests have succeeded. This prevents
 632                                  * any window of opportunity for debug
 633                                  * register grabbing by other users.
 634                                  */
 635                                 if (!second_pass)
 636                                         continue;
 637
 638                                 rc = ptrace_modify_breakpoint(bp, len, type,
 639                                                               tsk, 1);
 640                                 if (rc)
 641                                         break;
 642                         }
 643                         continue;
 644                 }
 645
 646                 rc = ptrace_modify_breakpoint(bp, len, type, tsk, 0);
 647                 if (rc)
 648                         break;
 649         }
 650         /*
 651          * Make a second pass to free the remaining unused breakpoints
 652          * or to restore the original breakpoints if an error occurred.
 653          */
 654         if (!second_pass) {
 655                 second_pass = 1;
 656                 if (rc < 0) {
 657                         orig_ret = rc;
 658                         data = old_dr7;
 659                 }
 660                 goto restore;
 661         }
 662
 663         ptrace_put_breakpoints(tsk);
 664
 665         return ((orig_ret < 0) ? orig_ret : rc);
 666 }
 667
 668 /*
 669  * Handle PTRACE_PEEKUSR calls for the debug register area.
 670  */
 671 static unsigned long ptrace_get_debugreg(struct task_struct *tsk, int n)
 672 {
 673         struct thread_struct *thread = &(tsk->thread);
 674         unsigned long val = 0;
 675
 676         if (n < HBP_NUM) {
 677                 struct perf_event *bp;
 678
 679                 if (ptrace_get_breakpoints(tsk) < 0)
 680                         return -ESRCH;
 681
 682                 bp = thread->ptrace_bps[n];
 683                 if (!bp)
 684                         val = 0;
 685                 else
 686                         val = bp->hw.info.address;
 687
 688                 ptrace_put_breakpoints(tsk);
 689         } else if (n == 6) {
 690                 val = thread->debugreg6;
 691          } else if (n == 7) {
 692                 val = thread->ptrace_dr7;
 693         }
 694         return val;
 695 }
 696
 697 static int ptrace_set_breakpoint_addr(struct task_struct *tsk, int nr,
 698                                       unsigned long addr)
 699 {
 700         struct perf_event *bp;
 701         struct thread_struct *t = &tsk->thread;
 702         struct perf_event_attr attr;
 703         int err = 0;
 704
 705         if (ptrace_get_breakpoints(tsk) < 0)
 706                 return -ESRCH;
 707
 708         if (!t->ptrace_bps[nr]) {
 709                 ptrace_breakpoint_init(&attr);
 710                 /*
 711                  * Put stub len and type to register (reserve) an inactive but
 712                  * correct bp
 713                  */
 714                 attr.bp_addr = addr;
 715                 attr.bp_len = HW_BREAKPOINT_LEN_1;
 716                 attr.bp_type = HW_BREAKPOINT_W;
 717                 attr.disabled = 1;
 718
 719                 bp = register_user_hw_breakpoint(&attr, ptrace_triggered,
 720                                                  NULL, tsk);
 721
 722                 /*
 723                  * CHECKME: the previous code returned -EIO if the addr wasn't
 724                  * a valid task virtual addr. The new one will return -EINVAL in
 725                  *  this case.
 726                  * -EINVAL may be what we want for in-kernel breakpoints users,
 727                  * but -EIO looks better for ptrace, since we refuse a register
 728                  * writing for the user. And anyway this is the previous
 729                  * behaviour.
 730                  */
 731                 if (IS_ERR(bp)) {
 732                         err = PTR_ERR(bp);
 733                         goto put;
 734                 }
 735
 736                 t->ptrace_bps[nr] = bp;
 737         } else {
 738                 bp = t->ptrace_bps[nr];
 739
 740                 attr = bp->attr;
 741                 attr.bp_addr = addr;
 742                 err = modify_user_hw_breakpoint(bp, &attr);
 743         }
 744
 745 put:
 746         ptrace_put_breakpoints(tsk);
 747         return err;
 748 }
 749
 750 /*
 751  * Handle PTRACE_POKEUSR calls for the debug register area.
 752  */
 753 static int ptrace_set_debugreg(struct task_struct *tsk, int n,
 754                                unsigned long val)
 755 {
 756         struct thread_struct *thread = &(tsk->thread);
 757         int rc = 0;
 758
 759         /* There are no DR4 or DR5 registers */
 760         if (n == 4 || n == 5)
 761                 return -EIO;
 762
 763         if (n == 6) {
 764                 thread->debugreg6 = val;
 765                 goto ret_path;
 766         }
 767         if (n < HBP_NUM) {
 768                 rc = ptrace_set_breakpoint_addr(tsk, n, val);
 769                 if (rc)
 770                         return rc;
 771         }
 772         /* All that's left is DR7 */
 773         if (n == 7) {
 774                 rc = ptrace_write_dr7(tsk, val);
 775                 if (!rc)
 776                         thread->ptrace_dr7 = val;
 777         }
 778
 779 ret_path:
 780         return rc;
 781 }
 782
 783 /*
 784  * These access the current or another (stopped) task's io permission
 785  * bitmap for debugging or core dump.
 786  */
 787 static int ioperm_active(struct task_struct *target,
 788                          const struct user_regset *regset)
 789 {
 790         return target->thread.io_bitmap_max / regset->size;
 791 }
 792
 793 static int ioperm_get(struct task_struct *target,
 794                       const struct user_regset *regset,
 795                       unsigned int pos, unsigned int count,
 796                       void *kbuf, void __user *ubuf)
 797 {
 798         if (!target->thread.io_bitmap_ptr)
 799                 return -ENXIO;
 800
 801         return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
 802                                    target->thread.io_bitmap_ptr,
 803                                    0, IO_BITMAP_BYTES);
 804 }
 805
 806 /*
 807  * Called by kernel/ptrace.c when detaching..
 808  *
 809  * Make sure the single step bit is not set.
 810  */
 811 void ptrace_disable(struct task_struct *child)
 812 {
 813         user_disable_single_step(child);
 814 #ifdef TIF_SYSCALL_EMU
 815         clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
 816 #endif
 817 }
 818
 819 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
 820 static const struct user_regset_view user_x86_32_view; /* Initialized below. */
 821 #endif
 822
 823 long arch_ptrace(struct task_struct *child, long request,
 824                  unsigned long addr, unsigned long data)
 825 {
 826         int ret;
 827         unsigned long __user *datap = (unsigned long __user *)data;
 828
 829         switch (request) {
 830         /* read the word at location addr in the USER area. */
 831         case PTRACE_PEEKUSR: {
 832                 unsigned long tmp;
 833
 834                 ret = -EIO;
 835                 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user))
 836                         break;
 837
 838                 tmp = 0;  /* Default return condition */
 839                 if (addr < sizeof(struct user_regs_struct))
 840                         tmp = getreg(child, addr);
 841                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
 842                          addr <= offsetof(struct user, u_debugreg[7])) {
 843                         addr -= offsetof(struct user, u_debugreg[0]);
 844                         tmp = ptrace_get_debugreg(child, addr / sizeof(data));
 845                 }
 846                 ret = put_user(tmp, datap);
 847                 break;
 848         }
 849
 850         case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
 851                 ret = -EIO;
 852                 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user))
 853                         break;
 854
 855                 if (addr < sizeof(struct user_regs_struct))
 856                         ret = putreg(child, addr, data);
 857                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
 858                          addr <= offsetof(struct user, u_debugreg[7])) {
 859                         addr -= offsetof(struct user, u_debugreg[0]);
 860                         ret = ptrace_set_debugreg(child,
 861                                                   addr / sizeof(data), data);
 862                 }
 863                 break;
 864
 865         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
 866                 return copy_regset_to_user(child,
 867                                            task_user_regset_view(current),
 868                                            REGSET_GENERAL,
 869                                            0, sizeof(struct user_regs_struct),
 870                                            datap);
 871
 872         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
 873                 return copy_regset_from_user(child,
 874                                              task_user_regset_view(current),
 875                                              REGSET_GENERAL,
 876                                              0, sizeof(struct user_regs_struct),
 877                                              datap);
 878
 879         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
 880                 return copy_regset_to_user(child,
 881                                            task_user_regset_view(current),
 882                                            REGSET_FP,
 883                                            0, sizeof(struct user_i387_struct),
 884                                            datap);
 885
 886         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
 887                 return copy_regset_from_user(child,
 888                                              task_user_regset_view(current),
 889                                              REGSET_FP,
 890                                              0, sizeof(struct user_i387_struct),
 891                                              datap);
 892
 893 #ifdef CONFIG_X86_32
 894         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
 895                 return copy_regset_to_user(child, &user_x86_32_view,
 896                                            REGSET_XFP,
 897                                            0, sizeof(struct user_fxsr_struct),
 898                                            datap) ? -EIO : 0;
 899
 900         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
 901                 return copy_regset_from_user(child, &user_x86_32_view,
 902                                              REGSET_XFP,
 903                                              0, sizeof(struct user_fxsr_struct),
 904                                              datap) ? -EIO : 0;
 905 #endif
 906
 907 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
 908         case PTRACE_GET_THREAD_AREA:
 909                 if ((int) addr < 0)
 910                         return -EIO;
 911                 ret = do_get_thread_area(child, addr,
 912                                         (struct user_desc __user *)data);
 913                 break;
 914
 915         case PTRACE_SET_THREAD_AREA:
 916                 if ((int) addr < 0)
 917                         return -EIO;
 918                 ret = do_set_thread_area(child, addr,
 919                                         (struct user_desc __user *)data, 0);
 920                 break;
 921 #endif
 922
 923 #ifdef CONFIG_X86_64
 924                 /* normal 64bit interface to access TLS data.
 925                    Works just like arch_prctl, except that the arguments
 926                    are reversed. */
 927         case PTRACE_ARCH_PRCTL:
 928                 ret = do_arch_prctl(child, data, addr);
 929                 break;
 930 #endif
 931
 932         default:
 933                 ret = ptrace_request(child, request, addr, data);
 934                 break;
 935         }
 936
 937         return ret;
 938 }
 939
 940 #ifdef CONFIG_IA32_EMULATION
 941
 942 #include <linux/compat.h>
 943 #include <linux/syscalls.h>
 944 #include <asm/ia32.h>
 945 #include <asm/user32.h>
 946
 947 #define R32(l,q)                                                        \
 948         case offsetof(struct user32, regs.l):                           \
 949                 regs->q = value; break
 950
 951 #define SEG32(rs)                                                       \
 952         case offsetof(struct user32, regs.rs):                          \
 953                 return set_segment_reg(child,                           \
 954                                        offsetof(struct user_regs_struct, rs), \
 955                                        value);                          \
 956                 break
 957
 958 static int putreg32(struct task_struct *child, unsigned regno, u32 value)
 959 {
 960         struct pt_regs *regs = task_pt_regs(child);
 961
 962         switch (regno) {
 963
 964         SEG32(cs);
 965         SEG32(ds);
 966         SEG32(es);
 967         SEG32(fs);
 968         SEG32(gs);
 969         SEG32(ss);
 970
 971         R32(ebx, bx);
 972         R32(ecx, cx);
 973         R32(edx, dx);
 974         R32(edi, di);
 975         R32(esi, si);
 976         R32(ebp, bp);
 977         R32(eax, ax);
 978         R32(eip, ip);
 979         R32(esp, sp);
 980
 981         case offsetof(struct user32, regs.orig_eax):
 982                 /*
 983                  * A 32-bit debugger setting orig_eax means to restore
 984                  * the state of the task restarting a 32-bit syscall.
 985                  * Make sure we interpret the -ERESTART* codes correctly
 986                  * in case the task is not actually still sitting at the
 987                  * exit from a 32-bit syscall with TS_COMPAT still set.
 988                  */
 989                 regs->orig_ax = value;
 990                 if (syscall_get_nr(child, regs) >= 0)
 991                         task_thread_info(child)->status |= TS_COMPAT;
 992                 break;
 993
 994         case offsetof(struct user32, regs.eflags):
 995                 return set_flags(child, value);
 996
 997         case offsetof(struct user32, u_debugreg[0]) ...
 998                 offsetof(struct user32, u_debugreg[7]):
 999                 regno -= offsetof(struct user32, u_debugreg[0]);
1000                 return ptrace_set_debugreg(child, regno / 4, value);
1001
1002         default:
1003                 if (regno > sizeof(struct user32) || (regno & 3))
1004                         return -EIO;
1005
1006                 /*
1007                  * Other dummy fields in the virtual user structure
1008                  * are ignored
1009                  */
1010                 break;
1011         }
1012         return 0;
1013 }
1014
1015 #undef R32
1016 #undef SEG32
1017
1018 #define R32(l,q)                                                        \
1019         case offsetof(struct user32, regs.l):                           \
1020                 *val = regs->q; break
1021
1022 #define SEG32(rs)                                                       \
1023         case offsetof(struct user32, regs.rs):                          \
1024                 *val = get_segment_reg(child,                           \
1025                                        offsetof(struct user_regs_struct, rs)); \
1026                 break
1027
1028 static int getreg32(struct task_struct *child, unsigned regno, u32 *val)
1029 {
1030         struct pt_regs *regs = task_pt_regs(child);
1031
1032         switch (regno) {
1033
1034         SEG32(ds);
1035         SEG32(es);
1036         SEG32(fs);
1037         SEG32(gs);
1038
1039         R32(cs, cs);
1040         R32(ss, ss);
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                 *val = get_flags(child);
1054                 break;
1055
1056         case offsetof(struct user32, u_debugreg[0]) ...
1057                 offsetof(struct user32, u_debugreg[7]):
1058                 regno -= offsetof(struct user32, u_debugreg[0]);
1059                 *val = ptrace_get_debugreg(child, regno / 4);
1060                 break;
1061
1062         default:
1063                 if (regno > sizeof(struct user32) || (regno & 3))
1064                         return -EIO;
1065
1066                 /*
1067                  * Other dummy fields in the virtual user structure
1068                  * are ignored
1069                  */
1070                 *val = 0;
1071                 break;
1072         }
1073         return 0;
1074 }
1075
1076 #undef R32
1077 #undef SEG32
1078
1079 static int genregs32_get(struct task_struct *target,
1080                          const struct user_regset *regset,
1081                          unsigned int pos, unsigned int count,
1082                          void *kbuf, void __user *ubuf)
1083 {
1084         if (kbuf) {
1085                 compat_ulong_t *k = kbuf;
1086                 while (count >= sizeof(*k)) {
1087                         getreg32(target, pos, k++);
1088                         count -= sizeof(*k);
1089                         pos += sizeof(*k);
1090                 }
1091         } else {
1092                 compat_ulong_t __user *u = ubuf;
1093                 while (count >= sizeof(*u)) {
1094                         compat_ulong_t word;
1095                         getreg32(target, pos, &word);
1096                         if (__put_user(word, u++))
1097                                 return -EFAULT;
1098                         count -= sizeof(*u);
1099                         pos += sizeof(*u);
1100                 }
1101         }
1102
1103         return 0;
1104 }
1105
1106 static int genregs32_set(struct task_struct *target,
1107                          const struct user_regset *regset,
1108                          unsigned int pos, unsigned int count,
1109                          const void *kbuf, const void __user *ubuf)
1110 {
1111         int ret = 0;
1112         if (kbuf) {
1113                 const compat_ulong_t *k = kbuf;
1114                 while (count >= sizeof(*k) && !ret) {
1115                         ret = putreg32(target, pos, *k++);
1116                         count -= sizeof(*k);
1117                         pos += sizeof(*k);
1118                 }
1119         } else {
1120                 const compat_ulong_t __user *u = ubuf;
1121                 while (count >= sizeof(*u) && !ret) {
1122                         compat_ulong_t word;
1123                         ret = __get_user(word, u++);
1124                         if (ret)
1125                                 break;
1126                         ret = putreg32(target, pos, word);
1127                         count -= sizeof(*u);
1128                         pos += sizeof(*u);
1129                 }
1130         }
1131         return ret;
1132 }
1133
1134 #ifdef CONFIG_X86_X32_ABI
1135 static long x32_arch_ptrace(struct task_struct *child,
1136                             compat_long_t request, compat_ulong_t caddr,
1137                             compat_ulong_t cdata)
1138 {
1139         unsigned long addr = caddr;
1140         unsigned long data = cdata;
1141         void __user *datap = compat_ptr(data);
1142         int ret;
1143
1144         switch (request) {
1145         /* Read 32bits at location addr in the USER area.  Only allow
1146            to return the lower 32bits of segment and debug registers.  */
1147         case PTRACE_PEEKUSR: {
1148                 u32 tmp;
1149
1150                 ret = -EIO;
1151                 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user) ||
1152                     addr < offsetof(struct user_regs_struct, cs))
1153                         break;
1154
1155                 tmp = 0;  /* Default return condition */
1156                 if (addr < sizeof(struct user_regs_struct))
1157                         tmp = getreg(child, addr);
1158                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
1159                          addr <= offsetof(struct user, u_debugreg[7])) {
1160                         addr -= offsetof(struct user, u_debugreg[0]);
1161                         tmp = ptrace_get_debugreg(child, addr / sizeof(data));
1162                 }
1163                 ret = put_user(tmp, (__u32 __user *)datap);
1164                 break;
1165         }
1166
1167         /* Write the word at location addr in the USER area.  Only allow
1168            to update segment and debug registers with the upper 32bits
1169            zero-extended. */
1170         case PTRACE_POKEUSR:
1171                 ret = -EIO;
1172                 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user) ||
1173                     addr < offsetof(struct user_regs_struct, cs))
1174                         break;
1175
1176                 if (addr < sizeof(struct user_regs_struct))
1177                         ret = putreg(child, addr, data);
1178                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
1179                          addr <= offsetof(struct user, u_debugreg[7])) {
1180                         addr -= offsetof(struct user, u_debugreg[0]);
1181                         ret = ptrace_set_debugreg(child,
1182                                                   addr / sizeof(data), data);
1183                 }
1184                 break;
1185
1186         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
1187                 return copy_regset_to_user(child,
1188                                            task_user_regset_view(current),
1189                                            REGSET_GENERAL,
1190                                            0, sizeof(struct user_regs_struct),
1191                                            datap);
1192
1193         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
1194                 return copy_regset_from_user(child,
1195                                              task_user_regset_view(current),
1196                                              REGSET_GENERAL,
1197                                              0, sizeof(struct user_regs_struct),
1198                                              datap);
1199
1200         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
1201                 return copy_regset_to_user(child,
1202                                            task_user_regset_view(current),
1203                                            REGSET_FP,
1204                                            0, sizeof(struct user_i387_struct),
1205                                            datap);
1206
1207         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
1208                 return copy_regset_from_user(child,
1209                                              task_user_regset_view(current),
1210                                              REGSET_FP,
1211                                              0, sizeof(struct user_i387_struct),
1212                                              datap);
1213
1214         default:
1215                 return compat_ptrace_request(child, request, addr, data);
1216         }
1217
1218         return ret;
1219 }
1220 #endif
1221
1222 long compat_arch_ptrace(struct task_struct *child, compat_long_t request,
1223                         compat_ulong_t caddr, compat_ulong_t cdata)
1224 {
1225         unsigned long addr = caddr;
1226         unsigned long data = cdata;
1227         void __user *datap = compat_ptr(data);
1228         int ret;
1229         __u32 val;
1230
1231 #ifdef CONFIG_X86_X32_ABI
1232         if (!is_ia32_task())
1233                 return x32_arch_ptrace(child, request, caddr, cdata);
1234 #endif
1235
1236         switch (request) {
1237         case PTRACE_PEEKUSR:
1238                 ret = getreg32(child, addr, &val);
1239                 if (ret == 0)
1240                         ret = put_user(val, (__u32 __user *)datap);
1241                 break;
1242
1243         case PTRACE_POKEUSR:
1244                 ret = putreg32(child, addr, data);
1245                 break;
1246
1247         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
1248                 return copy_regset_to_user(child, &user_x86_32_view,
1249                                            REGSET_GENERAL,
1250                                            0, sizeof(struct user_regs_struct32),
1251                                            datap);
1252
1253         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
1254                 return copy_regset_from_user(child, &user_x86_32_view,
1255                                              REGSET_GENERAL, 0,
1256                                              sizeof(struct user_regs_struct32),
1257                                              datap);
1258
1259         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
1260                 return copy_regset_to_user(child, &user_x86_32_view,
1261                                            REGSET_FP, 0,
1262                                            sizeof(struct user_i387_ia32_struct),
1263                                            datap);
1264
1265         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
1266                 return copy_regset_from_user(
1267                         child, &user_x86_32_view, REGSET_FP,
1268                         0, sizeof(struct user_i387_ia32_struct), datap);
1269
1270         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
1271                 return copy_regset_to_user(child, &user_x86_32_view,
1272                                            REGSET_XFP, 0,
1273                                            sizeof(struct user32_fxsr_struct),
1274                                            datap);
1275
1276         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
1277                 return copy_regset_from_user(child, &user_x86_32_view,
1278                                              REGSET_XFP, 0,
1279                                              sizeof(struct user32_fxsr_struct),
1280                                              datap);
1281
1282         case PTRACE_GET_THREAD_AREA:
1283         case PTRACE_SET_THREAD_AREA:
1284                 return arch_ptrace(child, request, addr, data);
1285
1286         default:
1287                 return compat_ptrace_request(child, request, addr, data);
1288         }
1289
1290         return ret;
1291 }
1292
1293 #endif  /* CONFIG_IA32_EMULATION */
1294
1295 #ifdef CONFIG_X86_64
1296
1297 static struct user_regset x86_64_regsets[] __read_mostly = {
1298         [REGSET_GENERAL] = {
1299                 .core_note_type = NT_PRSTATUS,
1300                 .n = sizeof(struct user_regs_struct) / sizeof(long),
1301                 .size = sizeof(long), .align = sizeof(long),
1302                 .get = genregs_get, .set = genregs_set
1303         },
1304         [REGSET_FP] = {
1305                 .core_note_type = NT_PRFPREG,
1306                 .n = sizeof(struct user_i387_struct) / sizeof(long),
1307                 .size = sizeof(long), .align = sizeof(long),
1308                 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1309         },
1310         [REGSET_XSTATE] = {
1311                 .core_note_type = NT_X86_XSTATE,
1312                 .size = sizeof(u64), .align = sizeof(u64),
1313                 .active = xstateregs_active, .get = xstateregs_get,
1314                 .set = xstateregs_set
1315         },
1316         [REGSET_IOPERM64] = {
1317                 .core_note_type = NT_386_IOPERM,
1318                 .n = IO_BITMAP_LONGS,
1319                 .size = sizeof(long), .align = sizeof(long),
1320                 .active = ioperm_active, .get = ioperm_get
1321         },
1322 };
1323
1324 static const struct user_regset_view user_x86_64_view = {
1325         .name = "x86_64", .e_machine = EM_X86_64,
1326         .regsets = x86_64_regsets, .n = ARRAY_SIZE(x86_64_regsets)
1327 };
1328
1329 #else  /* CONFIG_X86_32 */
1330
1331 #define user_regs_struct32      user_regs_struct
1332 #define genregs32_get           genregs_get
1333 #define genregs32_set           genregs_set
1334
1335 #define user_i387_ia32_struct   user_i387_struct
1336 #define user32_fxsr_struct      user_fxsr_struct
1337
1338 #endif  /* CONFIG_X86_64 */
1339
1340 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1341 static struct user_regset x86_32_regsets[] __read_mostly = {
1342         [REGSET_GENERAL] = {
1343                 .core_note_type = NT_PRSTATUS,
1344                 .n = sizeof(struct user_regs_struct32) / sizeof(u32),
1345                 .size = sizeof(u32), .align = sizeof(u32),
1346                 .get = genregs32_get, .set = genregs32_set
1347         },
1348         [REGSET_FP] = {
1349                 .core_note_type = NT_PRFPREG,
1350                 .n = sizeof(struct user_i387_ia32_struct) / sizeof(u32),
1351                 .size = sizeof(u32), .align = sizeof(u32),
1352                 .active = fpregs_active, .get = fpregs_get, .set = fpregs_set
1353         },
1354         [REGSET_XFP] = {
1355                 .core_note_type = NT_PRXFPREG,
1356                 .n = sizeof(struct user32_fxsr_struct) / sizeof(u32),
1357                 .size = sizeof(u32), .align = sizeof(u32),
1358                 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1359         },
1360         [REGSET_XSTATE] = {
1361                 .core_note_type = NT_X86_XSTATE,
1362                 .size = sizeof(u64), .align = sizeof(u64),
1363                 .active = xstateregs_active, .get = xstateregs_get,
1364                 .set = xstateregs_set
1365         },
1366         [REGSET_TLS] = {
1367                 .core_note_type = NT_386_TLS,
1368                 .n = GDT_ENTRY_TLS_ENTRIES, .bias = GDT_ENTRY_TLS_MIN,
1369                 .size = sizeof(struct user_desc),
1370                 .align = sizeof(struct user_desc),
1371                 .active = regset_tls_active,
1372                 .get = regset_tls_get, .set = regset_tls_set
1373         },
1374         [REGSET_IOPERM32] = {
1375                 .core_note_type = NT_386_IOPERM,
1376                 .n = IO_BITMAP_BYTES / sizeof(u32),
1377                 .size = sizeof(u32), .align = sizeof(u32),
1378                 .active = ioperm_active, .get = ioperm_get
1379         },
1380 };
1381
1382 static const struct user_regset_view user_x86_32_view = {
1383         .name = "i386", .e_machine = EM_386,
1384         .regsets = x86_32_regsets, .n = ARRAY_SIZE(x86_32_regsets)
1385 };
1386 #endif
1387
1388 /*
1389  * This represents bytes 464..511 in the memory layout exported through
1390  * the REGSET_XSTATE interface.
1391  */
1392 u64 xstate_fx_sw_bytes[USER_XSTATE_FX_SW_WORDS];
1393
1394 void update_regset_xstate_info(unsigned int size, u64 xstate_mask)
1395 {
1396 #ifdef CONFIG_X86_64
1397         x86_64_regsets[REGSET_XSTATE].n = size / sizeof(u64);
1398 #endif
1399 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1400         x86_32_regsets[REGSET_XSTATE].n = size / sizeof(u64);
1401 #endif
1402         xstate_fx_sw_bytes[USER_XSTATE_XCR0_WORD] = xstate_mask;
1403 }
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 static void fill_sigtrap_info(struct task_struct *tsk,
1419                                 struct pt_regs *regs,
1420                                 int error_code, int si_code,
1421                                 struct siginfo *info)
1422 {
1423         tsk->thread.trap_nr = X86_TRAP_DB;
1424         tsk->thread.error_code = error_code;
1425
1426         memset(info, 0, sizeof(*info));
1427         info->si_signo = SIGTRAP;
1428         info->si_code = si_code;
1429         info->si_addr = user_mode_vm(regs) ? (void __user *)regs->ip : NULL;
1430 }
1431
1432 void user_single_step_siginfo(struct task_struct *tsk,
1433                                 struct pt_regs *regs,
1434                                 struct siginfo *info)
1435 {
1436         fill_sigtrap_info(tsk, regs, 0, TRAP_BRKPT, info);
1437 }
1438
1439 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs,
1440                                          int error_code, int si_code)
1441 {
1442         struct siginfo info;
1443
1444         fill_sigtrap_info(tsk, regs, error_code, si_code, &info);
1445         /* Send us the fake SIGTRAP */
1446         force_sig_info(SIGTRAP, &info, tsk);
1447 }
1448
1449
1450 #ifdef CONFIG_X86_32
1451 # define IS_IA32        1
1452 #elif defined CONFIG_IA32_EMULATION
1453 # define IS_IA32        is_compat_task()
1454 #else
1455 # define IS_IA32        0
1456 #endif
1457
1458 /*
1459  * We must return the syscall number to actually look up in the table.
1460  * This can be -1L to skip running any syscall at all.
1461  */
1462 long syscall_trace_enter(struct pt_regs *regs)
1463 {
1464         long ret = 0;
1465
1466         /*
1467          * If we stepped into a sysenter/syscall insn, it trapped in
1468          * kernel mode; do_debug() cleared TF and set TIF_SINGLESTEP.
1469          * If user-mode had set TF itself, then it's still clear from
1470          * do_debug() and we need to set it again to restore the user
1471          * state.  If we entered on the slow path, TF was already set.
1472          */
1473         if (test_thread_flag(TIF_SINGLESTEP))
1474                 regs->flags |= X86_EFLAGS_TF;
1475
1476         /* do the secure computing check first */
1477         if (secure_computing(regs->orig_ax)) {
1478                 /* seccomp failures shouldn't expose any additional code. */
1479                 ret = -1L;
1480                 goto out;
1481         }
1482
1483         if (unlikely(test_thread_flag(TIF_SYSCALL_EMU)))
1484                 ret = -1L;
1485
1486         if ((ret || test_thread_flag(TIF_SYSCALL_TRACE)) &&
1487             tracehook_report_syscall_entry(regs))
1488                 ret = -1L;
1489
1490         if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
1491                 trace_sys_enter(regs, regs->orig_ax);
1492
1493         if (IS_IA32)
1494                 audit_syscall_entry(AUDIT_ARCH_I386,
1495                                     regs->orig_ax,
1496                                     regs->bx, regs->cx,
1497                                     regs->dx, regs->si);
1498 #ifdef CONFIG_X86_64
1499         else
1500                 audit_syscall_entry(AUDIT_ARCH_X86_64,
1501                                     regs->orig_ax,
1502                                     regs->di, regs->si,
1503                                     regs->dx, regs->r10);
1504 #endif
1505
1506 out:
1507         return ret ?: regs->orig_ax;
1508 }
1509
1510 void syscall_trace_leave(struct pt_regs *regs)
1511 {
1512         bool step;
1513
1514         audit_syscall_exit(regs);
1515
1516         if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
1517                 trace_sys_exit(regs, regs->ax);
1518
1519         /*
1520          * If TIF_SYSCALL_EMU is set, we only get here because of
1521          * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
1522          * We already reported this syscall instruction in
1523          * syscall_trace_enter().
1524          */
1525         step = unlikely(test_thread_flag(TIF_SINGLESTEP)) &&
1526                         !test_thread_flag(TIF_SYSCALL_EMU);
1527         if (step || test_thread_flag(TIF_SYSCALL_TRACE))
1528                 tracehook_report_syscall_exit(regs, step);
1529 }