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