Pileus Git - ~andy/linux/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
   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 #include <linux/rcupdate.h>
  25 #include <linux/module.h>
  26
  27 #include <asm/uaccess.h>
  28 #include <asm/pgtable.h>
  29 #include <asm/processor.h>
  30 #include <asm/i387.h>
  31 #include <asm/fpu-internal.h>
  32 #include <asm/debugreg.h>
  33 #include <asm/ldt.h>
  34 #include <asm/desc.h>
  35 #include <asm/prctl.h>
  36 #include <asm/proto.h>
  37 #include <asm/hw_breakpoint.h>
  38 #include <asm/traps.h>
  39
  40 #include "tls.h"
  41
  42 #define CREATE_TRACE_POINTS
  43 #include <trace/events/syscalls.h>
  44
  45 enum x86_regset {
  46         REGSET_GENERAL,
  47         REGSET_FP,
  48         REGSET_XFP,
  49         REGSET_IOPERM64 = REGSET_XFP,
  50         REGSET_XSTATE,
  51         REGSET_TLS,
  52         REGSET_IOPERM32,
  53 };
  54
  55 struct pt_regs_offset {
  56         const char *name;
  57         int offset;
  58 };
  59
  60 #define REG_OFFSET_NAME(r) {.name = #r, .offset = offsetof(struct pt_regs, r)}
  61 #define REG_OFFSET_END {.name = NULL, .offset = 0}
  62
  63 static const struct pt_regs_offset regoffset_table[] = {
  64 #ifdef CONFIG_X86_64
  65         REG_OFFSET_NAME(r15),
  66         REG_OFFSET_NAME(r14),
  67         REG_OFFSET_NAME(r13),
  68         REG_OFFSET_NAME(r12),
  69         REG_OFFSET_NAME(r11),
  70         REG_OFFSET_NAME(r10),
  71         REG_OFFSET_NAME(r9),
  72         REG_OFFSET_NAME(r8),
  73 #endif
  74         REG_OFFSET_NAME(bx),
  75         REG_OFFSET_NAME(cx),
  76         REG_OFFSET_NAME(dx),
  77         REG_OFFSET_NAME(si),
  78         REG_OFFSET_NAME(di),
  79         REG_OFFSET_NAME(bp),
  80         REG_OFFSET_NAME(ax),
  81 #ifdef CONFIG_X86_32
  82         REG_OFFSET_NAME(ds),
  83         REG_OFFSET_NAME(es),
  84         REG_OFFSET_NAME(fs),
  85         REG_OFFSET_NAME(gs),
  86 #endif
  87         REG_OFFSET_NAME(orig_ax),
  88         REG_OFFSET_NAME(ip),
  89         REG_OFFSET_NAME(cs),
  90         REG_OFFSET_NAME(flags),
  91         REG_OFFSET_NAME(sp),
  92         REG_OFFSET_NAME(ss),
  93         REG_OFFSET_END,
  94 };
  95
  96 /**
  97  * regs_query_register_offset() - query register offset from its name
  98  * @name:       the name of a register
  99  *
 100  * regs_query_register_offset() returns the offset of a register in struct
 101  * pt_regs from its name. If the name is invalid, this returns -EINVAL;
 102  */
 103 int regs_query_register_offset(const char *name)
 104 {
 105         const struct pt_regs_offset *roff;
 106         for (roff = regoffset_table; roff->name != NULL; roff++)
 107                 if (!strcmp(roff->name, name))
 108                         return roff->offset;
 109         return -EINVAL;
 110 }
 111
 112 /**
 113  * regs_query_register_name() - query register name from its offset
 114  * @offset:     the offset of a register in struct pt_regs.
 115  *
 116  * regs_query_register_name() returns the name of a register from its
 117  * offset in struct pt_regs. If the @offset is invalid, this returns NULL;
 118  */
 119 const char *regs_query_register_name(unsigned int offset)
 120 {
 121         const struct pt_regs_offset *roff;
 122         for (roff = regoffset_table; roff->name != NULL; roff++)
 123                 if (roff->offset == offset)
 124                         return roff->name;
 125         return NULL;
 126 }
 127
 128 static const int arg_offs_table[] = {
 129 #ifdef CONFIG_X86_32
 130         [0] = offsetof(struct pt_regs, ax),
 131         [1] = offsetof(struct pt_regs, dx),
 132         [2] = offsetof(struct pt_regs, cx)
 133 #else /* CONFIG_X86_64 */
 134         [0] = offsetof(struct pt_regs, di),
 135         [1] = offsetof(struct pt_regs, si),
 136         [2] = offsetof(struct pt_regs, dx),
 137         [3] = offsetof(struct pt_regs, cx),
 138         [4] = offsetof(struct pt_regs, r8),
 139         [5] = offsetof(struct pt_regs, r9)
 140 #endif
 141 };
 142
 143 /*
 144  * does not yet catch signals sent when the child dies.
 145  * in exit.c or in signal.c.
 146  */
 147
 148 /*
 149  * Determines which flags the user has access to [1 = access, 0 = no access].
 150  */
 151 #define FLAG_MASK_32            ((unsigned long)                        \
 152                                  (X86_EFLAGS_CF | X86_EFLAGS_PF |       \
 153                                   X86_EFLAGS_AF | X86_EFLAGS_ZF |       \
 154                                   X86_EFLAGS_SF | X86_EFLAGS_TF |       \
 155                                   X86_EFLAGS_DF | X86_EFLAGS_OF |       \
 156                                   X86_EFLAGS_RF | X86_EFLAGS_AC))
 157
 158 /*
 159  * Determines whether a value may be installed in a segment register.
 160  */
 161 static inline bool invalid_selector(u16 value)
 162 {
 163         return unlikely(value != 0 && (value & SEGMENT_RPL_MASK) != USER_RPL);
 164 }
 165
 166 #ifdef CONFIG_X86_32
 167
 168 #define FLAG_MASK               FLAG_MASK_32
 169
 170 /*
 171  * X86_32 CPUs don't save ss and esp if the CPU is already in kernel mode
 172  * when it traps.  The previous stack will be directly underneath the saved
 173  * registers, and 'sp/ss' won't even have been saved. Thus the '&regs->sp'.
 174  *
 175  * Now, if the stack is empty, '&regs->sp' is out of range. In this
 176  * case we try to take the previous stack. To always return a non-null
 177  * stack pointer we fall back to regs as stack if no previous stack
 178  * exists.
 179  *
 180  * This is valid only for kernel mode traps.
 181  */
 182 unsigned long kernel_stack_pointer(struct pt_regs *regs)
 183 {
 184         unsigned long context = (unsigned long)regs & ~(THREAD_SIZE - 1);
 185         unsigned long sp = (unsigned long)&regs->sp;
 186         struct thread_info *tinfo;
 187
 188         if (context == (sp & ~(THREAD_SIZE - 1)))
 189                 return sp;
 190
 191         tinfo = (struct thread_info *)context;
 192         if (tinfo->previous_esp)
 193                 return tinfo->previous_esp;
 194
 195         return (unsigned long)regs;
 196 }
 197 EXPORT_SYMBOL_GPL(kernel_stack_pointer);
 198
 199 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long regno)
 200 {
 201         BUILD_BUG_ON(offsetof(struct pt_regs, bx) != 0);
 202         return &regs->bx + (regno >> 2);
 203 }
 204
 205 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
 206 {
 207         /*
 208          * Returning the value truncates it to 16 bits.
 209          */
 210         unsigned int retval;
 211         if (offset != offsetof(struct user_regs_struct, gs))
 212                 retval = *pt_regs_access(task_pt_regs(task), offset);
 213         else {
 214                 if (task == current)
 215                         retval = get_user_gs(task_pt_regs(task));
 216                 else
 217                         retval = task_user_gs(task);
 218         }
 219         return retval;
 220 }
 221
 222 static int set_segment_reg(struct task_struct *task,
 223                            unsigned long offset, u16 value)
 224 {
 225         /*
 226          * The value argument was already truncated to 16 bits.
 227          */
 228         if (invalid_selector(value))
 229                 return -EIO;
 230
 231         /*
 232          * For %cs and %ss we cannot permit a null selector.
 233          * We can permit a bogus selector as long as it has USER_RPL.
 234          * Null selectors are fine for other segment registers, but
 235          * we will never get back to user mode with invalid %cs or %ss
 236          * and will take the trap in iret instead.  Much code relies
 237          * on user_mode() to distinguish a user trap frame (which can
 238          * safely use invalid selectors) from a kernel trap frame.
 239          */
 240         switch (offset) {
 241         case offsetof(struct user_regs_struct, cs):
 242         case offsetof(struct user_regs_struct, ss):
 243                 if (unlikely(value == 0))
 244                         return -EIO;
 245
 246         default:
 247                 *pt_regs_access(task_pt_regs(task), offset) = value;
 248                 break;
 249
 250         case offsetof(struct user_regs_struct, gs):
 251                 if (task == current)
 252                         set_user_gs(task_pt_regs(task), value);
 253                 else
 254                         task_user_gs(task) = value;
 255         }
 256
 257         return 0;
 258 }
 259
 260 #else  /* CONFIG_X86_64 */
 261
 262 #define FLAG_MASK               (FLAG_MASK_32 | X86_EFLAGS_NT)
 263
 264 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long offset)
 265 {
 266         BUILD_BUG_ON(offsetof(struct pt_regs, r15) != 0);
 267         return &regs->r15 + (offset / sizeof(regs->r15));
 268 }
 269
 270 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
 271 {
 272         /*
 273          * Returning the value truncates it to 16 bits.
 274          */
 275         unsigned int seg;
 276
 277         switch (offset) {
 278         case offsetof(struct user_regs_struct, fs):
 279                 if (task == current) {
 280                         /* Older gas can't assemble movq %?s,%r?? */
 281                         asm("movl %%fs,%0" : "=r" (seg));
 282                         return seg;
 283                 }
 284                 return task->thread.fsindex;
 285         case offsetof(struct user_regs_struct, gs):
 286                 if (task == current) {
 287                         asm("movl %%gs,%0" : "=r" (seg));
 288                         return seg;
 289                 }
 290                 return task->thread.gsindex;
 291         case offsetof(struct user_regs_struct, ds):
 292                 if (task == current) {
 293                         asm("movl %%ds,%0" : "=r" (seg));
 294                         return seg;
 295                 }
 296                 return task->thread.ds;
 297         case offsetof(struct user_regs_struct, es):
 298                 if (task == current) {
 299                         asm("movl %%es,%0" : "=r" (seg));
 300                         return seg;
 301                 }
 302                 return task->thread.es;
 303
 304         case offsetof(struct user_regs_struct, cs):
 305         case offsetof(struct user_regs_struct, ss):
 306                 break;
 307         }
 308         return *pt_regs_access(task_pt_regs(task), offset);
 309 }
 310
 311 static int set_segment_reg(struct task_struct *task,
 312                            unsigned long offset, u16 value)
 313 {
 314         /*
 315          * The value argument was already truncated to 16 bits.
 316          */
 317         if (invalid_selector(value))
 318                 return -EIO;
 319
 320         switch (offset) {
 321         case offsetof(struct user_regs_struct,fs):
 322                 /*
 323                  * If this is setting fs as for normal 64-bit use but
 324                  * setting fs_base has implicitly changed it, leave it.
 325                  */
 326                 if ((value == FS_TLS_SEL && task->thread.fsindex == 0 &&
 327                      task->thread.fs != 0) ||
 328                     (value == 0 && task->thread.fsindex == FS_TLS_SEL &&
 329                      task->thread.fs == 0))
 330                         break;
 331                 task->thread.fsindex = value;
 332                 if (task == current)
 333                         loadsegment(fs, task->thread.fsindex);
 334                 break;
 335         case offsetof(struct user_regs_struct,gs):
 336                 /*
 337                  * If this is setting gs as for normal 64-bit use but
 338                  * setting gs_base has implicitly changed it, leave it.
 339                  */
 340                 if ((value == GS_TLS_SEL && task->thread.gsindex == 0 &&
 341                      task->thread.gs != 0) ||
 342                     (value == 0 && task->thread.gsindex == GS_TLS_SEL &&
 343                      task->thread.gs == 0))
 344                         break;
 345                 task->thread.gsindex = value;
 346                 if (task == current)
 347                         load_gs_index(task->thread.gsindex);
 348                 break;
 349         case offsetof(struct user_regs_struct,ds):
 350                 task->thread.ds = value;
 351                 if (task == current)
 352                         loadsegment(ds, task->thread.ds);
 353                 break;
 354         case offsetof(struct user_regs_struct,es):
 355                 task->thread.es = value;
 356                 if (task == current)
 357                         loadsegment(es, task->thread.es);
 358                 break;
 359
 360                 /*
 361                  * Can't actually change these in 64-bit mode.
 362                  */
 363         case offsetof(struct user_regs_struct,cs):
 364                 if (unlikely(value == 0))
 365                         return -EIO;
 366 #ifdef CONFIG_IA32_EMULATION
 367                 if (test_tsk_thread_flag(task, TIF_IA32))
 368                         task_pt_regs(task)->cs = value;
 369 #endif
 370                 break;
 371         case offsetof(struct user_regs_struct,ss):
 372                 if (unlikely(value == 0))
 373                         return -EIO;
 374 #ifdef CONFIG_IA32_EMULATION
 375                 if (test_tsk_thread_flag(task, TIF_IA32))
 376                         task_pt_regs(task)->ss = value;
 377 #endif
 378                 break;
 379         }
 380
 381         return 0;
 382 }
 383
 384 #endif  /* CONFIG_X86_32 */
 385
 386 static unsigned long get_flags(struct task_struct *task)
 387 {
 388         unsigned long retval = task_pt_regs(task)->flags;
 389
 390         /*
 391          * If the debugger set TF, hide it from the readout.
 392          */
 393         if (test_tsk_thread_flag(task, TIF_FORCED_TF))
 394                 retval &= ~X86_EFLAGS_TF;
 395
 396         return retval;
 397 }
 398
 399 static int set_flags(struct task_struct *task, unsigned long value)
 400 {
 401         struct pt_regs *regs = task_pt_regs(task);
 402
 403         /*
 404          * If the user value contains TF, mark that
 405          * it was not "us" (the debugger) that set it.
 406          * If not, make sure it stays set if we had.
 407          */
 408         if (value & X86_EFLAGS_TF)
 409                 clear_tsk_thread_flag(task, TIF_FORCED_TF);
 410         else if (test_tsk_thread_flag(task, TIF_FORCED_TF))
 411                 value |= X86_EFLAGS_TF;
 412
 413         regs->flags = (regs->flags & ~FLAG_MASK) | (value & FLAG_MASK);
 414
 415         return 0;
 416 }
 417
 418 static int putreg(struct task_struct *child,
 419                   unsigned long offset, unsigned long value)
 420 {
 421         switch (offset) {
 422         case offsetof(struct user_regs_struct, cs):
 423         case offsetof(struct user_regs_struct, ds):
 424         case offsetof(struct user_regs_struct, es):
 425         case offsetof(struct user_regs_struct, fs):
 426         case offsetof(struct user_regs_struct, gs):
 427         case offsetof(struct user_regs_struct, ss):
 428                 return set_segment_reg(child, offset, value);
 429
 430         case offsetof(struct user_regs_struct, flags):
 431                 return set_flags(child, value);
 432
 433 #ifdef CONFIG_X86_64
 434         case offsetof(struct user_regs_struct,fs_base):
 435                 if (value >= TASK_SIZE_OF(child))
 436                         return -EIO;
 437                 /*
 438                  * When changing the segment base, use do_arch_prctl
 439                  * to set either thread.fs or thread.fsindex and the
 440                  * corresponding GDT slot.
 441                  */
 442                 if (child->thread.fs != value)
 443                         return do_arch_prctl(child, ARCH_SET_FS, value);
 444                 return 0;
 445         case offsetof(struct user_regs_struct,gs_base):
 446                 /*
 447                  * Exactly the same here as the %fs handling above.
 448                  */
 449                 if (value >= TASK_SIZE_OF(child))
 450                         return -EIO;
 451                 if (child->thread.gs != value)
 452                         return do_arch_prctl(child, ARCH_SET_GS, value);
 453                 return 0;
 454 #endif
 455         }
 456
 457         *pt_regs_access(task_pt_regs(child), offset) = value;
 458         return 0;
 459 }
 460
 461 static unsigned long getreg(struct task_struct *task, unsigned long offset)
 462 {
 463         switch (offset) {
 464         case offsetof(struct user_regs_struct, cs):
 465         case offsetof(struct user_regs_struct, ds):
 466         case offsetof(struct user_regs_struct, es):
 467         case offsetof(struct user_regs_struct, fs):
 468         case offsetof(struct user_regs_struct, gs):
 469         case offsetof(struct user_regs_struct, ss):
 470                 return get_segment_reg(task, offset);
 471
 472         case offsetof(struct user_regs_struct, flags):
 473                 return get_flags(task);
 474
 475 #ifdef CONFIG_X86_64
 476         case offsetof(struct user_regs_struct, fs_base): {
 477                 /*
 478                  * do_arch_prctl may have used a GDT slot instead of
 479                  * the MSR.  To userland, it appears the same either
 480                  * way, except the %fs segment selector might not be 0.
 481                  */
 482                 unsigned int seg = task->thread.fsindex;
 483                 if (task->thread.fs != 0)
 484                         return task->thread.fs;
 485                 if (task == current)
 486                         asm("movl %%fs,%0" : "=r" (seg));
 487                 if (seg != FS_TLS_SEL)
 488                         return 0;
 489                 return get_desc_base(&task->thread.tls_array[FS_TLS]);
 490         }
 491         case offsetof(struct user_regs_struct, gs_base): {
 492                 /*
 493                  * Exactly the same here as the %fs handling above.
 494                  */
 495                 unsigned int seg = task->thread.gsindex;
 496                 if (task->thread.gs != 0)
 497                         return task->thread.gs;
 498                 if (task == current)
 499                         asm("movl %%gs,%0" : "=r" (seg));
 500                 if (seg != GS_TLS_SEL)
 501                         return 0;
 502                 return get_desc_base(&task->thread.tls_array[GS_TLS]);
 503         }
 504 #endif
 505         }
 506
 507         return *pt_regs_access(task_pt_regs(task), offset);
 508 }
 509
 510 static int genregs_get(struct task_struct *target,
 511                        const struct user_regset *regset,
 512                        unsigned int pos, unsigned int count,
 513                        void *kbuf, void __user *ubuf)
 514 {
 515         if (kbuf) {
 516                 unsigned long *k = kbuf;
 517                 while (count >= sizeof(*k)) {
 518                         *k++ = getreg(target, pos);
 519                         count -= sizeof(*k);
 520                         pos += sizeof(*k);
 521                 }
 522         } else {
 523                 unsigned long __user *u = ubuf;
 524                 while (count >= sizeof(*u)) {
 525                         if (__put_user(getreg(target, pos), u++))
 526                                 return -EFAULT;
 527                         count -= sizeof(*u);
 528                         pos += sizeof(*u);
 529                 }
 530         }
 531
 532         return 0;
 533 }
 534
 535 static int genregs_set(struct task_struct *target,
 536                        const struct user_regset *regset,
 537                        unsigned int pos, unsigned int count,
 538                        const void *kbuf, const void __user *ubuf)
 539 {
 540         int ret = 0;
 541         if (kbuf) {
 542                 const unsigned long *k = kbuf;
 543                 while (count >= sizeof(*k) && !ret) {
 544                         ret = putreg(target, pos, *k++);
 545                         count -= sizeof(*k);
 546                         pos += sizeof(*k);
 547                 }
 548         } else {
 549                 const unsigned long  __user *u = ubuf;
 550                 while (count >= sizeof(*u) && !ret) {
 551                         unsigned long word;
 552                         ret = __get_user(word, u++);
 553                         if (ret)
 554                                 break;
 555                         ret = putreg(target, pos, word);
 556                         count -= sizeof(*u);
 557                         pos += sizeof(*u);
 558                 }
 559         }
 560         return ret;
 561 }
 562
 563 static void ptrace_triggered(struct perf_event *bp,
 564                              struct perf_sample_data *data,
 565                              struct pt_regs *regs)
 566 {
 567         int i;
 568         struct thread_struct *thread = &(current->thread);
 569
 570         /*
 571          * Store in the virtual DR6 register the fact that the breakpoint
 572          * was hit so the thread's debugger will see it.
 573          */
 574         for (i = 0; i < HBP_NUM; i++) {
 575                 if (thread->ptrace_bps[i] == bp)
 576                         break;
 577         }
 578
 579         thread->debugreg6 |= (DR_TRAP0 << i);
 580 }
 581
 582 /*
 583  * Walk through every ptrace breakpoints for this thread and
 584  * build the dr7 value on top of their attributes.
 585  *
 586  */
 587 static unsigned long ptrace_get_dr7(struct perf_event *bp[])
 588 {
 589         int i;
 590         int dr7 = 0;
 591         struct arch_hw_breakpoint *info;
 592
 593         for (i = 0; i < HBP_NUM; i++) {
 594                 if (bp[i] && !bp[i]->attr.disabled) {
 595                         info = counter_arch_bp(bp[i]);
 596                         dr7 |= encode_dr7(i, info->len, info->type);
 597                 }
 598         }
 599
 600         return dr7;
 601 }
 602
 603 static int
 604 ptrace_modify_breakpoint(struct perf_event *bp, int len, int type,
 605                          struct task_struct *tsk, int disabled)
 606 {
 607         int err;
 608         int gen_len, gen_type;
 609         struct perf_event_attr attr;
 610
 611         /*
 612          * We should have at least an inactive breakpoint at this
 613          * slot. It means the user is writing dr7 without having
 614          * written the address register first
 615          */
 616         if (!bp)
 617                 return -EINVAL;
 618
 619         err = arch_bp_generic_fields(len, type, &gen_len, &gen_type);
 620         if (err)
 621                 return err;
 622
 623         attr = bp->attr;
 624         attr.bp_len = gen_len;
 625         attr.bp_type = gen_type;
 626         attr.disabled = disabled;
 627
 628         return modify_user_hw_breakpoint(bp, &attr);
 629 }
 630
 631 /*
 632  * Handle ptrace writes to debug register 7.
 633  */
 634 static int ptrace_write_dr7(struct task_struct *tsk, unsigned long data)
 635 {
 636         struct thread_struct *thread = &(tsk->thread);
 637         unsigned long old_dr7;
 638         int i, orig_ret = 0, rc = 0;
 639         int enabled, second_pass = 0;
 640         unsigned len, type;
 641         struct perf_event *bp;
 642
 643         if (ptrace_get_breakpoints(tsk) < 0)
 644                 return -ESRCH;
 645
 646         data &= ~DR_CONTROL_RESERVED;
 647         old_dr7 = ptrace_get_dr7(thread->ptrace_bps);
 648 restore:
 649         /*
 650          * Loop through all the hardware breakpoints, making the
 651          * appropriate changes to each.
 652          */
 653         for (i = 0; i < HBP_NUM; i++) {
 654                 enabled = decode_dr7(data, i, &len, &type);
 655                 bp = thread->ptrace_bps[i];
 656
 657                 if (!enabled) {
 658                         if (bp) {
 659                                 /*
 660                                  * Don't unregister the breakpoints right-away,
 661                                  * unless all register_user_hw_breakpoint()
 662                                  * requests have succeeded. This prevents
 663                                  * any window of opportunity for debug
 664                                  * register grabbing by other users.
 665                                  */
 666                                 if (!second_pass)
 667                                         continue;
 668
 669                                 rc = ptrace_modify_breakpoint(bp, len, type,
 670                                                               tsk, 1);
 671                                 if (rc)
 672                                         break;
 673                         }
 674                         continue;
 675                 }
 676
 677                 rc = ptrace_modify_breakpoint(bp, len, type, tsk, 0);
 678                 if (rc)
 679                         break;
 680         }
 681         /*
 682          * Make a second pass to free the remaining unused breakpoints
 683          * or to restore the original breakpoints if an error occurred.
 684          */
 685         if (!second_pass) {
 686                 second_pass = 1;
 687                 if (rc < 0) {
 688                         orig_ret = rc;
 689                         data = old_dr7;
 690                 }
 691                 goto restore;
 692         }
 693
 694         ptrace_put_breakpoints(tsk);
 695
 696         return ((orig_ret < 0) ? orig_ret : rc);
 697 }
 698
 699 /*
 700  * Handle PTRACE_PEEKUSR calls for the debug register area.
 701  */
 702 static unsigned long ptrace_get_debugreg(struct task_struct *tsk, int n)
 703 {
 704         struct thread_struct *thread = &(tsk->thread);
 705         unsigned long val = 0;
 706
 707         if (n < HBP_NUM) {
 708                 struct perf_event *bp;
 709
 710                 if (ptrace_get_breakpoints(tsk) < 0)
 711                         return -ESRCH;
 712
 713                 bp = thread->ptrace_bps[n];
 714                 if (!bp)
 715                         val = 0;
 716                 else
 717                         val = bp->hw.info.address;
 718
 719                 ptrace_put_breakpoints(tsk);
 720         } else if (n == 6) {
 721                 val = thread->debugreg6;
 722          } else if (n == 7) {
 723                 val = thread->ptrace_dr7;
 724         }
 725         return val;
 726 }
 727
 728 static int ptrace_set_breakpoint_addr(struct task_struct *tsk, int nr,
 729                                       unsigned long addr)
 730 {
 731         struct perf_event *bp;
 732         struct thread_struct *t = &tsk->thread;
 733         struct perf_event_attr attr;
 734         int err = 0;
 735
 736         if (ptrace_get_breakpoints(tsk) < 0)
 737                 return -ESRCH;
 738
 739         if (!t->ptrace_bps[nr]) {
 740                 ptrace_breakpoint_init(&attr);
 741                 /*
 742                  * Put stub len and type to register (reserve) an inactive but
 743                  * correct bp
 744                  */
 745                 attr.bp_addr = addr;
 746                 attr.bp_len = HW_BREAKPOINT_LEN_1;
 747                 attr.bp_type = HW_BREAKPOINT_W;
 748                 attr.disabled = 1;
 749
 750                 bp = register_user_hw_breakpoint(&attr, ptrace_triggered,
 751                                                  NULL, tsk);
 752
 753                 /*
 754                  * CHECKME: the previous code returned -EIO if the addr wasn't
 755                  * a valid task virtual addr. The new one will return -EINVAL in
 756                  *  this case.
 757                  * -EINVAL may be what we want for in-kernel breakpoints users,
 758                  * but -EIO looks better for ptrace, since we refuse a register
 759                  * writing for the user. And anyway this is the previous
 760                  * behaviour.
 761                  */
 762                 if (IS_ERR(bp)) {
 763                         err = PTR_ERR(bp);
 764                         goto put;
 765                 }
 766
 767                 t->ptrace_bps[nr] = bp;
 768         } else {
 769                 bp = t->ptrace_bps[nr];
 770
 771                 attr = bp->attr;
 772                 attr.bp_addr = addr;
 773                 err = modify_user_hw_breakpoint(bp, &attr);
 774         }
 775
 776 put:
 777         ptrace_put_breakpoints(tsk);
 778         return err;
 779 }
 780
 781 /*
 782  * Handle PTRACE_POKEUSR calls for the debug register area.
 783  */
 784 static int ptrace_set_debugreg(struct task_struct *tsk, int n,
 785                                unsigned long val)
 786 {
 787         struct thread_struct *thread = &(tsk->thread);
 788         int rc = 0;
 789
 790         /* There are no DR4 or DR5 registers */
 791         if (n == 4 || n == 5)
 792                 return -EIO;
 793
 794         if (n == 6) {
 795                 thread->debugreg6 = val;
 796                 goto ret_path;
 797         }
 798         if (n < HBP_NUM) {
 799                 rc = ptrace_set_breakpoint_addr(tsk, n, val);
 800                 if (rc)
 801                         return rc;
 802         }
 803         /* All that's left is DR7 */
 804         if (n == 7) {
 805                 rc = ptrace_write_dr7(tsk, val);
 806                 if (!rc)
 807                         thread->ptrace_dr7 = val;
 808         }
 809
 810 ret_path:
 811         return rc;
 812 }
 813
 814 /*
 815  * These access the current or another (stopped) task's io permission
 816  * bitmap for debugging or core dump.
 817  */
 818 static int ioperm_active(struct task_struct *target,
 819                          const struct user_regset *regset)
 820 {
 821         return target->thread.io_bitmap_max / regset->size;
 822 }
 823
 824 static int ioperm_get(struct task_struct *target,
 825                       const struct user_regset *regset,
 826                       unsigned int pos, unsigned int count,
 827                       void *kbuf, void __user *ubuf)
 828 {
 829         if (!target->thread.io_bitmap_ptr)
 830                 return -ENXIO;
 831
 832         return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
 833                                    target->thread.io_bitmap_ptr,
 834                                    0, IO_BITMAP_BYTES);
 835 }
 836
 837 /*
 838  * Called by kernel/ptrace.c when detaching..
 839  *
 840  * Make sure the single step bit is not set.
 841  */
 842 void ptrace_disable(struct task_struct *child)
 843 {
 844         user_disable_single_step(child);
 845 #ifdef TIF_SYSCALL_EMU
 846         clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
 847 #endif
 848 }
 849
 850 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
 851 static const struct user_regset_view user_x86_32_view; /* Initialized below. */
 852 #endif
 853
 854 long arch_ptrace(struct task_struct *child, long request,
 855                  unsigned long addr, unsigned long data)
 856 {
 857         int ret;
 858         unsigned long __user *datap = (unsigned long __user *)data;
 859
 860         switch (request) {
 861         /* read the word at location addr in the USER area. */
 862         case PTRACE_PEEKUSR: {
 863                 unsigned long tmp;
 864
 865                 ret = -EIO;
 866                 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user))
 867                         break;
 868
 869                 tmp = 0;  /* Default return condition */
 870                 if (addr < sizeof(struct user_regs_struct))
 871                         tmp = getreg(child, addr);
 872                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
 873                          addr <= offsetof(struct user, u_debugreg[7])) {
 874                         addr -= offsetof(struct user, u_debugreg[0]);
 875                         tmp = ptrace_get_debugreg(child, addr / sizeof(data));
 876                 }
 877                 ret = put_user(tmp, datap);
 878                 break;
 879         }
 880
 881         case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
 882                 ret = -EIO;
 883                 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user))
 884                         break;
 885
 886                 if (addr < sizeof(struct user_regs_struct))
 887                         ret = putreg(child, addr, data);
 888                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
 889                          addr <= offsetof(struct user, u_debugreg[7])) {
 890                         addr -= offsetof(struct user, u_debugreg[0]);
 891                         ret = ptrace_set_debugreg(child,
 892                                                   addr / sizeof(data), data);
 893                 }
 894                 break;
 895
 896         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
 897                 return copy_regset_to_user(child,
 898                                            task_user_regset_view(current),
 899                                            REGSET_GENERAL,
 900                                            0, sizeof(struct user_regs_struct),
 901                                            datap);
 902
 903         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
 904                 return copy_regset_from_user(child,
 905                                              task_user_regset_view(current),
 906                                              REGSET_GENERAL,
 907                                              0, sizeof(struct user_regs_struct),
 908                                              datap);
 909
 910         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
 911                 return copy_regset_to_user(child,
 912                                            task_user_regset_view(current),
 913                                            REGSET_FP,
 914                                            0, sizeof(struct user_i387_struct),
 915                                            datap);
 916
 917         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
 918                 return copy_regset_from_user(child,
 919                                              task_user_regset_view(current),
 920                                              REGSET_FP,
 921                                              0, sizeof(struct user_i387_struct),
 922                                              datap);
 923
 924 #ifdef CONFIG_X86_32
 925         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
 926                 return copy_regset_to_user(child, &user_x86_32_view,
 927                                            REGSET_XFP,
 928                                            0, sizeof(struct user_fxsr_struct),
 929                                            datap) ? -EIO : 0;
 930
 931         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
 932                 return copy_regset_from_user(child, &user_x86_32_view,
 933                                              REGSET_XFP,
 934                                              0, sizeof(struct user_fxsr_struct),
 935                                              datap) ? -EIO : 0;
 936 #endif
 937
 938 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
 939         case PTRACE_GET_THREAD_AREA:
 940                 if ((int) addr < 0)
 941                         return -EIO;
 942                 ret = do_get_thread_area(child, addr,
 943                                         (struct user_desc __user *)data);
 944                 break;
 945
 946         case PTRACE_SET_THREAD_AREA:
 947                 if ((int) addr < 0)
 948                         return -EIO;
 949                 ret = do_set_thread_area(child, addr,
 950                                         (struct user_desc __user *)data, 0);
 951                 break;
 952 #endif
 953
 954 #ifdef CONFIG_X86_64
 955                 /* normal 64bit interface to access TLS data.
 956                    Works just like arch_prctl, except that the arguments
 957                    are reversed. */
 958         case PTRACE_ARCH_PRCTL:
 959                 ret = do_arch_prctl(child, data, addr);
 960                 break;
 961 #endif
 962
 963         default:
 964                 ret = ptrace_request(child, request, addr, data);
 965                 break;
 966         }
 967
 968         return ret;
 969 }
 970
 971 #ifdef CONFIG_IA32_EMULATION
 972
 973 #include <linux/compat.h>
 974 #include <linux/syscalls.h>
 975 #include <asm/ia32.h>
 976 #include <asm/user32.h>
 977
 978 #define R32(l,q)                                                        \
 979         case offsetof(struct user32, regs.l):                           \
 980                 regs->q = value; break
 981
 982 #define SEG32(rs)                                                       \
 983         case offsetof(struct user32, regs.rs):                          \
 984                 return set_segment_reg(child,                           \
 985                                        offsetof(struct user_regs_struct, rs), \
 986                                        value);                          \
 987                 break
 988
 989 static int putreg32(struct task_struct *child, unsigned regno, u32 value)
 990 {
 991         struct pt_regs *regs = task_pt_regs(child);
 992
 993         switch (regno) {
 994
 995         SEG32(cs);
 996         SEG32(ds);
 997         SEG32(es);
 998         SEG32(fs);
 999         SEG32(gs);
1000         SEG32(ss);
1001
1002         R32(ebx, bx);
1003         R32(ecx, cx);
1004         R32(edx, dx);
1005         R32(edi, di);
1006         R32(esi, si);
1007         R32(ebp, bp);
1008         R32(eax, ax);
1009         R32(eip, ip);
1010         R32(esp, sp);
1011
1012         case offsetof(struct user32, regs.orig_eax):
1013                 /*
1014                  * A 32-bit debugger setting orig_eax means to restore
1015                  * the state of the task restarting a 32-bit syscall.
1016                  * Make sure we interpret the -ERESTART* codes correctly
1017                  * in case the task is not actually still sitting at the
1018                  * exit from a 32-bit syscall with TS_COMPAT still set.
1019                  */
1020                 regs->orig_ax = value;
1021                 if (syscall_get_nr(child, regs) >= 0)
1022                         task_thread_info(child)->status |= TS_COMPAT;
1023                 break;
1024
1025         case offsetof(struct user32, regs.eflags):
1026                 return set_flags(child, value);
1027
1028         case offsetof(struct user32, u_debugreg[0]) ...
1029                 offsetof(struct user32, u_debugreg[7]):
1030                 regno -= offsetof(struct user32, u_debugreg[0]);
1031                 return ptrace_set_debugreg(child, regno / 4, value);
1032
1033         default:
1034                 if (regno > sizeof(struct user32) || (regno & 3))
1035                         return -EIO;
1036
1037                 /*
1038                  * Other dummy fields in the virtual user structure
1039                  * are ignored
1040                  */
1041                 break;
1042         }
1043         return 0;
1044 }
1045
1046 #undef R32
1047 #undef SEG32
1048
1049 #define R32(l,q)                                                        \
1050         case offsetof(struct user32, regs.l):                           \
1051                 *val = regs->q; break
1052
1053 #define SEG32(rs)                                                       \
1054         case offsetof(struct user32, regs.rs):                          \
1055                 *val = get_segment_reg(child,                           \
1056                                        offsetof(struct user_regs_struct, rs)); \
1057                 break
1058
1059 static int getreg32(struct task_struct *child, unsigned regno, u32 *val)
1060 {
1061         struct pt_regs *regs = task_pt_regs(child);
1062
1063         switch (regno) {
1064
1065         SEG32(ds);
1066         SEG32(es);
1067         SEG32(fs);
1068         SEG32(gs);
1069
1070         R32(cs, cs);
1071         R32(ss, ss);
1072         R32(ebx, bx);
1073         R32(ecx, cx);
1074         R32(edx, dx);
1075         R32(edi, di);
1076         R32(esi, si);
1077         R32(ebp, bp);
1078         R32(eax, ax);
1079         R32(orig_eax, orig_ax);
1080         R32(eip, ip);
1081         R32(esp, sp);
1082
1083         case offsetof(struct user32, regs.eflags):
1084                 *val = get_flags(child);
1085                 break;
1086
1087         case offsetof(struct user32, u_debugreg[0]) ...
1088                 offsetof(struct user32, u_debugreg[7]):
1089                 regno -= offsetof(struct user32, u_debugreg[0]);
1090                 *val = ptrace_get_debugreg(child, regno / 4);
1091                 break;
1092
1093         default:
1094                 if (regno > sizeof(struct user32) || (regno & 3))
1095                         return -EIO;
1096
1097                 /*
1098                  * Other dummy fields in the virtual user structure
1099                  * are ignored
1100                  */
1101                 *val = 0;
1102                 break;
1103         }
1104         return 0;
1105 }
1106
1107 #undef R32
1108 #undef SEG32
1109
1110 static int genregs32_get(struct task_struct *target,
1111                          const struct user_regset *regset,
1112                          unsigned int pos, unsigned int count,
1113                          void *kbuf, void __user *ubuf)
1114 {
1115         if (kbuf) {
1116                 compat_ulong_t *k = kbuf;
1117                 while (count >= sizeof(*k)) {
1118                         getreg32(target, pos, k++);
1119                         count -= sizeof(*k);
1120                         pos += sizeof(*k);
1121                 }
1122         } else {
1123                 compat_ulong_t __user *u = ubuf;
1124                 while (count >= sizeof(*u)) {
1125                         compat_ulong_t word;
1126                         getreg32(target, pos, &word);
1127                         if (__put_user(word, u++))
1128                                 return -EFAULT;
1129                         count -= sizeof(*u);
1130                         pos += sizeof(*u);
1131                 }
1132         }
1133
1134         return 0;
1135 }
1136
1137 static int genregs32_set(struct task_struct *target,
1138                          const struct user_regset *regset,
1139                          unsigned int pos, unsigned int count,
1140                          const void *kbuf, const void __user *ubuf)
1141 {
1142         int ret = 0;
1143         if (kbuf) {
1144                 const compat_ulong_t *k = kbuf;
1145                 while (count >= sizeof(*k) && !ret) {
1146                         ret = putreg32(target, pos, *k++);
1147                         count -= sizeof(*k);
1148                         pos += sizeof(*k);
1149                 }
1150         } else {
1151                 const compat_ulong_t __user *u = ubuf;
1152                 while (count >= sizeof(*u) && !ret) {
1153                         compat_ulong_t word;
1154                         ret = __get_user(word, u++);
1155                         if (ret)
1156                                 break;
1157                         ret = putreg32(target, pos, word);
1158                         count -= sizeof(*u);
1159                         pos += sizeof(*u);
1160                 }
1161         }
1162         return ret;
1163 }
1164
1165 #ifdef CONFIG_X86_X32_ABI
1166 static long x32_arch_ptrace(struct task_struct *child,
1167                             compat_long_t request, compat_ulong_t caddr,
1168                             compat_ulong_t cdata)
1169 {
1170         unsigned long addr = caddr;
1171         unsigned long data = cdata;
1172         void __user *datap = compat_ptr(data);
1173         int ret;
1174
1175         switch (request) {
1176         /* Read 32bits at location addr in the USER area.  Only allow
1177            to return the lower 32bits of segment and debug registers.  */
1178         case PTRACE_PEEKUSR: {
1179                 u32 tmp;
1180
1181                 ret = -EIO;
1182                 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user) ||
1183                     addr < offsetof(struct user_regs_struct, cs))
1184                         break;
1185
1186                 tmp = 0;  /* Default return condition */
1187                 if (addr < sizeof(struct user_regs_struct))
1188                         tmp = getreg(child, addr);
1189                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
1190                          addr <= offsetof(struct user, u_debugreg[7])) {
1191                         addr -= offsetof(struct user, u_debugreg[0]);
1192                         tmp = ptrace_get_debugreg(child, addr / sizeof(data));
1193                 }
1194                 ret = put_user(tmp, (__u32 __user *)datap);
1195                 break;
1196         }
1197
1198         /* Write the word at location addr in the USER area.  Only allow
1199            to update segment and debug registers with the upper 32bits
1200            zero-extended. */
1201         case PTRACE_POKEUSR:
1202                 ret = -EIO;
1203                 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user) ||
1204                     addr < offsetof(struct user_regs_struct, cs))
1205                         break;
1206
1207                 if (addr < sizeof(struct user_regs_struct))
1208                         ret = putreg(child, addr, data);
1209                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
1210                          addr <= offsetof(struct user, u_debugreg[7])) {
1211                         addr -= offsetof(struct user, u_debugreg[0]);
1212                         ret = ptrace_set_debugreg(child,
1213                                                   addr / sizeof(data), data);
1214                 }
1215                 break;
1216
1217         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
1218                 return copy_regset_to_user(child,
1219                                            task_user_regset_view(current),
1220                                            REGSET_GENERAL,
1221                                            0, sizeof(struct user_regs_struct),
1222                                            datap);
1223
1224         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
1225                 return copy_regset_from_user(child,
1226                                              task_user_regset_view(current),
1227                                              REGSET_GENERAL,
1228                                              0, sizeof(struct user_regs_struct),
1229                                              datap);
1230
1231         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
1232                 return copy_regset_to_user(child,
1233                                            task_user_regset_view(current),
1234                                            REGSET_FP,
1235                                            0, sizeof(struct user_i387_struct),
1236                                            datap);
1237
1238         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
1239                 return copy_regset_from_user(child,
1240                                              task_user_regset_view(current),
1241                                              REGSET_FP,
1242                                              0, sizeof(struct user_i387_struct),
1243                                              datap);
1244
1245         default:
1246                 return compat_ptrace_request(child, request, addr, data);
1247         }
1248
1249         return ret;
1250 }
1251 #endif
1252
1253 long compat_arch_ptrace(struct task_struct *child, compat_long_t request,
1254                         compat_ulong_t caddr, compat_ulong_t cdata)
1255 {
1256         unsigned long addr = caddr;
1257         unsigned long data = cdata;
1258         void __user *datap = compat_ptr(data);
1259         int ret;
1260         __u32 val;
1261
1262 #ifdef CONFIG_X86_X32_ABI
1263         if (!is_ia32_task())
1264                 return x32_arch_ptrace(child, request, caddr, cdata);
1265 #endif
1266
1267         switch (request) {
1268         case PTRACE_PEEKUSR:
1269                 ret = getreg32(child, addr, &val);
1270                 if (ret == 0)
1271                         ret = put_user(val, (__u32 __user *)datap);
1272                 break;
1273
1274         case PTRACE_POKEUSR:
1275                 ret = putreg32(child, addr, data);
1276                 break;
1277
1278         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
1279                 return copy_regset_to_user(child, &user_x86_32_view,
1280                                            REGSET_GENERAL,
1281                                            0, sizeof(struct user_regs_struct32),
1282                                            datap);
1283
1284         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
1285                 return copy_regset_from_user(child, &user_x86_32_view,
1286                                              REGSET_GENERAL, 0,
1287                                              sizeof(struct user_regs_struct32),
1288                                              datap);
1289
1290         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
1291                 return copy_regset_to_user(child, &user_x86_32_view,
1292                                            REGSET_FP, 0,
1293                                            sizeof(struct user_i387_ia32_struct),
1294                                            datap);
1295
1296         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
1297                 return copy_regset_from_user(
1298                         child, &user_x86_32_view, REGSET_FP,
1299                         0, sizeof(struct user_i387_ia32_struct), datap);
1300
1301         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
1302                 return copy_regset_to_user(child, &user_x86_32_view,
1303                                            REGSET_XFP, 0,
1304                                            sizeof(struct user32_fxsr_struct),
1305                                            datap);
1306
1307         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
1308                 return copy_regset_from_user(child, &user_x86_32_view,
1309                                              REGSET_XFP, 0,
1310                                              sizeof(struct user32_fxsr_struct),
1311                                              datap);
1312
1313         case PTRACE_GET_THREAD_AREA:
1314         case PTRACE_SET_THREAD_AREA:
1315                 return arch_ptrace(child, request, addr, data);
1316
1317         default:
1318                 return compat_ptrace_request(child, request, addr, data);
1319         }
1320
1321         return ret;
1322 }
1323
1324 #endif  /* CONFIG_IA32_EMULATION */
1325
1326 #ifdef CONFIG_X86_64
1327
1328 static struct user_regset x86_64_regsets[] __read_mostly = {
1329         [REGSET_GENERAL] = {
1330                 .core_note_type = NT_PRSTATUS,
1331                 .n = sizeof(struct user_regs_struct) / sizeof(long),
1332                 .size = sizeof(long), .align = sizeof(long),
1333                 .get = genregs_get, .set = genregs_set
1334         },
1335         [REGSET_FP] = {
1336                 .core_note_type = NT_PRFPREG,
1337                 .n = sizeof(struct user_i387_struct) / sizeof(long),
1338                 .size = sizeof(long), .align = sizeof(long),
1339                 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1340         },
1341         [REGSET_XSTATE] = {
1342                 .core_note_type = NT_X86_XSTATE,
1343                 .size = sizeof(u64), .align = sizeof(u64),
1344                 .active = xstateregs_active, .get = xstateregs_get,
1345                 .set = xstateregs_set
1346         },
1347         [REGSET_IOPERM64] = {
1348                 .core_note_type = NT_386_IOPERM,
1349                 .n = IO_BITMAP_LONGS,
1350                 .size = sizeof(long), .align = sizeof(long),
1351                 .active = ioperm_active, .get = ioperm_get
1352         },
1353 };
1354
1355 static const struct user_regset_view user_x86_64_view = {
1356         .name = "x86_64", .e_machine = EM_X86_64,
1357         .regsets = x86_64_regsets, .n = ARRAY_SIZE(x86_64_regsets)
1358 };
1359
1360 #else  /* CONFIG_X86_32 */
1361
1362 #define user_regs_struct32      user_regs_struct
1363 #define genregs32_get           genregs_get
1364 #define genregs32_set           genregs_set
1365
1366 #endif  /* CONFIG_X86_64 */
1367
1368 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1369 static struct user_regset x86_32_regsets[] __read_mostly = {
1370         [REGSET_GENERAL] = {
1371                 .core_note_type = NT_PRSTATUS,
1372                 .n = sizeof(struct user_regs_struct32) / sizeof(u32),
1373                 .size = sizeof(u32), .align = sizeof(u32),
1374                 .get = genregs32_get, .set = genregs32_set
1375         },
1376         [REGSET_FP] = {
1377                 .core_note_type = NT_PRFPREG,
1378                 .n = sizeof(struct user_i387_ia32_struct) / sizeof(u32),
1379                 .size = sizeof(u32), .align = sizeof(u32),
1380                 .active = fpregs_active, .get = fpregs_get, .set = fpregs_set
1381         },
1382         [REGSET_XFP] = {
1383                 .core_note_type = NT_PRXFPREG,
1384                 .n = sizeof(struct user32_fxsr_struct) / sizeof(u32),
1385                 .size = sizeof(u32), .align = sizeof(u32),
1386                 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1387         },
1388         [REGSET_XSTATE] = {
1389                 .core_note_type = NT_X86_XSTATE,
1390                 .size = sizeof(u64), .align = sizeof(u64),
1391                 .active = xstateregs_active, .get = xstateregs_get,
1392                 .set = xstateregs_set
1393         },
1394         [REGSET_TLS] = {
1395                 .core_note_type = NT_386_TLS,
1396                 .n = GDT_ENTRY_TLS_ENTRIES, .bias = GDT_ENTRY_TLS_MIN,
1397                 .size = sizeof(struct user_desc),
1398                 .align = sizeof(struct user_desc),
1399                 .active = regset_tls_active,
1400                 .get = regset_tls_get, .set = regset_tls_set
1401         },
1402         [REGSET_IOPERM32] = {
1403                 .core_note_type = NT_386_IOPERM,
1404                 .n = IO_BITMAP_BYTES / sizeof(u32),
1405                 .size = sizeof(u32), .align = sizeof(u32),
1406                 .active = ioperm_active, .get = ioperm_get
1407         },
1408 };
1409
1410 static const struct user_regset_view user_x86_32_view = {
1411         .name = "i386", .e_machine = EM_386,
1412         .regsets = x86_32_regsets, .n = ARRAY_SIZE(x86_32_regsets)
1413 };
1414 #endif
1415
1416 /*
1417  * This represents bytes 464..511 in the memory layout exported through
1418  * the REGSET_XSTATE interface.
1419  */
1420 u64 xstate_fx_sw_bytes[USER_XSTATE_FX_SW_WORDS];
1421
1422 void update_regset_xstate_info(unsigned int size, u64 xstate_mask)
1423 {
1424 #ifdef CONFIG_X86_64
1425         x86_64_regsets[REGSET_XSTATE].n = size / sizeof(u64);
1426 #endif
1427 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1428         x86_32_regsets[REGSET_XSTATE].n = size / sizeof(u64);
1429 #endif
1430         xstate_fx_sw_bytes[USER_XSTATE_XCR0_WORD] = xstate_mask;
1431 }
1432
1433 const struct user_regset_view *task_user_regset_view(struct task_struct *task)
1434 {
1435 #ifdef CONFIG_IA32_EMULATION
1436         if (test_tsk_thread_flag(task, TIF_IA32))
1437 #endif
1438 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1439                 return &user_x86_32_view;
1440 #endif
1441 #ifdef CONFIG_X86_64
1442         return &user_x86_64_view;
1443 #endif
1444 }
1445
1446 static void fill_sigtrap_info(struct task_struct *tsk,
1447                                 struct pt_regs *regs,
1448                                 int error_code, int si_code,
1449                                 struct siginfo *info)
1450 {
1451         tsk->thread.trap_nr = X86_TRAP_DB;
1452         tsk->thread.error_code = error_code;
1453
1454         memset(info, 0, sizeof(*info));
1455         info->si_signo = SIGTRAP;
1456         info->si_code = si_code;
1457         info->si_addr = user_mode_vm(regs) ? (void __user *)regs->ip : NULL;
1458 }
1459
1460 void user_single_step_siginfo(struct task_struct *tsk,
1461                                 struct pt_regs *regs,
1462                                 struct siginfo *info)
1463 {
1464         fill_sigtrap_info(tsk, regs, 0, TRAP_BRKPT, info);
1465 }
1466
1467 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs,
1468                                          int error_code, int si_code)
1469 {
1470         struct siginfo info;
1471
1472         fill_sigtrap_info(tsk, regs, error_code, si_code, &info);
1473         /* Send us the fake SIGTRAP */
1474         force_sig_info(SIGTRAP, &info, tsk);
1475 }
1476
1477
1478 #ifdef CONFIG_X86_32
1479 # define IS_IA32        1
1480 #elif defined CONFIG_IA32_EMULATION
1481 # define IS_IA32        is_compat_task()
1482 #else
1483 # define IS_IA32        0
1484 #endif
1485
1486 /*
1487  * We must return the syscall number to actually look up in the table.
1488  * This can be -1L to skip running any syscall at all.
1489  */
1490 long syscall_trace_enter(struct pt_regs *regs)
1491 {
1492         long ret = 0;
1493
1494         rcu_user_exit();
1495
1496         /*
1497          * If we stepped into a sysenter/syscall insn, it trapped in
1498          * kernel mode; do_debug() cleared TF and set TIF_SINGLESTEP.
1499          * If user-mode had set TF itself, then it's still clear from
1500          * do_debug() and we need to set it again to restore the user
1501          * state.  If we entered on the slow path, TF was already set.
1502          */
1503         if (test_thread_flag(TIF_SINGLESTEP))
1504                 regs->flags |= X86_EFLAGS_TF;
1505
1506         /* do the secure computing check first */
1507         if (secure_computing(regs->orig_ax)) {
1508                 /* seccomp failures shouldn't expose any additional code. */
1509                 ret = -1L;
1510                 goto out;
1511         }
1512
1513         if (unlikely(test_thread_flag(TIF_SYSCALL_EMU)))
1514                 ret = -1L;
1515
1516         if ((ret || test_thread_flag(TIF_SYSCALL_TRACE)) &&
1517             tracehook_report_syscall_entry(regs))
1518                 ret = -1L;
1519
1520         if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
1521                 trace_sys_enter(regs, regs->orig_ax);
1522
1523         if (IS_IA32)
1524                 audit_syscall_entry(AUDIT_ARCH_I386,
1525                                     regs->orig_ax,
1526                                     regs->bx, regs->cx,
1527                                     regs->dx, regs->si);
1528 #ifdef CONFIG_X86_64
1529         else
1530                 audit_syscall_entry(AUDIT_ARCH_X86_64,
1531                                     regs->orig_ax,
1532                                     regs->di, regs->si,
1533                                     regs->dx, regs->r10);
1534 #endif
1535
1536 out:
1537         return ret ?: regs->orig_ax;
1538 }
1539
1540 void syscall_trace_leave(struct pt_regs *regs)
1541 {
1542         bool step;
1543
1544         audit_syscall_exit(regs);
1545
1546         if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
1547                 trace_sys_exit(regs, regs->ax);
1548
1549         /*
1550          * If TIF_SYSCALL_EMU is set, we only get here because of
1551          * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
1552          * We already reported this syscall instruction in
1553          * syscall_trace_enter().
1554          */
1555         step = unlikely(test_thread_flag(TIF_SINGLESTEP)) &&
1556                         !test_thread_flag(TIF_SYSCALL_EMU);
1557         if (step || test_thread_flag(TIF_SYSCALL_TRACE))
1558                 tracehook_report_syscall_exit(regs, step);
1559
1560         rcu_user_enter();
1561 }