arch/ia64/kernel/ptrace.c

   1 /*
   2  * Kernel support for the ptrace() and syscall tracing interfaces.
   3  *
   4  * Copyright (C) 1999-2005 Hewlett-Packard Co
   5  *      David Mosberger-Tang <davidm@hpl.hp.com>
   6  *
   7  * Derived from the x86 and Alpha versions.
   8  */
   9 #include <linux/config.h>
  10 #include <linux/kernel.h>
  11 #include <linux/sched.h>
  12 #include <linux/slab.h>
  13 #include <linux/mm.h>
  14 #include <linux/errno.h>
  15 #include <linux/ptrace.h>
  16 #include <linux/smp_lock.h>
  17 #include <linux/user.h>
  18 #include <linux/security.h>
  19 #include <linux/audit.h>
  20
  21 #include <asm/pgtable.h>
  22 #include <asm/processor.h>
  23 #include <asm/ptrace_offsets.h>
  24 #include <asm/rse.h>
  25 #include <asm/system.h>
  26 #include <asm/uaccess.h>
  27 #include <asm/unwind.h>
  28 #ifdef CONFIG_PERFMON
  29 #include <asm/perfmon.h>
  30 #endif
  31
  32 #include "entry.h"
  33
  34 /*
  35  * Bits in the PSR that we allow ptrace() to change:
  36  *      be, up, ac, mfl, mfh (the user mask; five bits total)
  37  *      db (debug breakpoint fault; one bit)
  38  *      id (instruction debug fault disable; one bit)
  39  *      dd (data debug fault disable; one bit)
  40  *      ri (restart instruction; two bits)
  41  *      is (instruction set; one bit)
  42  */
  43 #define IPSR_MASK (IA64_PSR_UM | IA64_PSR_DB | IA64_PSR_IS      \
  44                    | IA64_PSR_ID | IA64_PSR_DD | IA64_PSR_RI)
  45
  46 #define MASK(nbits)     ((1UL << (nbits)) - 1)  /* mask with NBITS bits set */
  47 #define PFM_MASK        MASK(38)
  48
  49 #define PTRACE_DEBUG    0
  50
  51 #if PTRACE_DEBUG
  52 # define dprintk(format...)     printk(format)
  53 # define inline
  54 #else
  55 # define dprintk(format...)
  56 #endif
  57
  58 /* Return TRUE if PT was created due to kernel-entry via a system-call.  */
  59
  60 static inline int
  61 in_syscall (struct pt_regs *pt)
  62 {
  63         return (long) pt->cr_ifs >= 0;
  64 }
  65
  66 /*
  67  * Collect the NaT bits for r1-r31 from scratch_unat and return a NaT
  68  * bitset where bit i is set iff the NaT bit of register i is set.
  69  */
  70 unsigned long
  71 ia64_get_scratch_nat_bits (struct pt_regs *pt, unsigned long scratch_unat)
  72 {
  73 #       define GET_BITS(first, last, unat)                              \
  74         ({                                                              \
  75                 unsigned long bit = ia64_unat_pos(&pt->r##first);       \
  76                 unsigned long nbits = (last - first + 1);               \
  77                 unsigned long mask = MASK(nbits) << first;              \
  78                 unsigned long dist;                                     \
  79                 if (bit < first)                                        \
  80                         dist = 64 + bit - first;                        \
  81                 else                                                    \
  82                         dist = bit - first;                             \
  83                 ia64_rotr(unat, dist) & mask;                           \
  84         })
  85         unsigned long val;
  86
  87         /*
  88          * Registers that are stored consecutively in struct pt_regs
  89          * can be handled in parallel.  If the register order in
  90          * struct_pt_regs changes, this code MUST be updated.
  91          */
  92         val  = GET_BITS( 1,  1, scratch_unat);
  93         val |= GET_BITS( 2,  3, scratch_unat);
  94         val |= GET_BITS(12, 13, scratch_unat);
  95         val |= GET_BITS(14, 14, scratch_unat);
  96         val |= GET_BITS(15, 15, scratch_unat);
  97         val |= GET_BITS( 8, 11, scratch_unat);
  98         val |= GET_BITS(16, 31, scratch_unat);
  99         return val;
 100
 101 #       undef GET_BITS
 102 }
 103
 104 /*
 105  * Set the NaT bits for the scratch registers according to NAT and
 106  * return the resulting unat (assuming the scratch registers are
 107  * stored in PT).
 108  */
 109 unsigned long
 110 ia64_put_scratch_nat_bits (struct pt_regs *pt, unsigned long nat)
 111 {
 112 #       define PUT_BITS(first, last, nat)                               \
 113         ({                                                              \
 114                 unsigned long bit = ia64_unat_pos(&pt->r##first);       \
 115                 unsigned long nbits = (last - first + 1);               \
 116                 unsigned long mask = MASK(nbits) << first;              \
 117                 long dist;                                              \
 118                 if (bit < first)                                        \
 119                         dist = 64 + bit - first;                        \
 120                 else                                                    \
 121                         dist = bit - first;                             \
 122                 ia64_rotl(nat & mask, dist);                            \
 123         })
 124         unsigned long scratch_unat;
 125
 126         /*
 127          * Registers that are stored consecutively in struct pt_regs
 128          * can be handled in parallel.  If the register order in
 129          * struct_pt_regs changes, this code MUST be updated.
 130          */
 131         scratch_unat  = PUT_BITS( 1,  1, nat);
 132         scratch_unat |= PUT_BITS( 2,  3, nat);
 133         scratch_unat |= PUT_BITS(12, 13, nat);
 134         scratch_unat |= PUT_BITS(14, 14, nat);
 135         scratch_unat |= PUT_BITS(15, 15, nat);
 136         scratch_unat |= PUT_BITS( 8, 11, nat);
 137         scratch_unat |= PUT_BITS(16, 31, nat);
 138
 139         return scratch_unat;
 140
 141 #       undef PUT_BITS
 142 }
 143
 144 #define IA64_MLX_TEMPLATE       0x2
 145 #define IA64_MOVL_OPCODE        6
 146
 147 void
 148 ia64_increment_ip (struct pt_regs *regs)
 149 {
 150         unsigned long w0, ri = ia64_psr(regs)->ri + 1;
 151
 152         if (ri > 2) {
 153                 ri = 0;
 154                 regs->cr_iip += 16;
 155         } else if (ri == 2) {
 156                 get_user(w0, (char __user *) regs->cr_iip + 0);
 157                 if (((w0 >> 1) & 0xf) == IA64_MLX_TEMPLATE) {
 158                         /*
 159                          * rfi'ing to slot 2 of an MLX bundle causes
 160                          * an illegal operation fault.  We don't want
 161                          * that to happen...
 162                          */
 163                         ri = 0;
 164                         regs->cr_iip += 16;
 165                 }
 166         }
 167         ia64_psr(regs)->ri = ri;
 168 }
 169
 170 void
 171 ia64_decrement_ip (struct pt_regs *regs)
 172 {
 173         unsigned long w0, ri = ia64_psr(regs)->ri - 1;
 174
 175         if (ia64_psr(regs)->ri == 0) {
 176                 regs->cr_iip -= 16;
 177                 ri = 2;
 178                 get_user(w0, (char __user *) regs->cr_iip + 0);
 179                 if (((w0 >> 1) & 0xf) == IA64_MLX_TEMPLATE) {
 180                         /*
 181                          * rfi'ing to slot 2 of an MLX bundle causes
 182                          * an illegal operation fault.  We don't want
 183                          * that to happen...
 184                          */
 185                         ri = 1;
 186                 }
 187         }
 188         ia64_psr(regs)->ri = ri;
 189 }
 190
 191 /*
 192  * This routine is used to read an rnat bits that are stored on the
 193  * kernel backing store.  Since, in general, the alignment of the user
 194  * and kernel are different, this is not completely trivial.  In
 195  * essence, we need to construct the user RNAT based on up to two
 196  * kernel RNAT values and/or the RNAT value saved in the child's
 197  * pt_regs.
 198  *
 199  * user rbs
 200  *
 201  * +--------+ <-- lowest address
 202  * | slot62 |
 203  * +--------+
 204  * |  rnat  | 0x....1f8
 205  * +--------+
 206  * | slot00 | \
 207  * +--------+ |
 208  * | slot01 | > child_regs->ar_rnat
 209  * +--------+ |
 210  * | slot02 | /                         kernel rbs
 211  * +--------+                           +--------+
 212  *          <- child_regs->ar_bspstore  | slot61 | <-- krbs
 213  * +- - - - +                           +--------+
 214  *                                      | slot62 |
 215  * +- - - - +                           +--------+
 216  *                                      |  rnat  |
 217  * +- - - - +                           +--------+
 218  *   vrnat                              | slot00 |
 219  * +- - - - +                           +--------+
 220  *                                      =        =
 221  *                                      +--------+
 222  *                                      | slot00 | \
 223  *                                      +--------+ |
 224  *                                      | slot01 | > child_stack->ar_rnat
 225  *                                      +--------+ |
 226  *                                      | slot02 | /
 227  *                                      +--------+
 228  *                                                <--- child_stack->ar_bspstore
 229  *
 230  * The way to think of this code is as follows: bit 0 in the user rnat
 231  * corresponds to some bit N (0 <= N <= 62) in one of the kernel rnat
 232  * value.  The kernel rnat value holding this bit is stored in
 233  * variable rnat0.  rnat1 is loaded with the kernel rnat value that
 234  * form the upper bits of the user rnat value.
 235  *
 236  * Boundary cases:
 237  *
 238  * o when reading the rnat "below" the first rnat slot on the kernel
 239  *   backing store, rnat0/rnat1 are set to 0 and the low order bits are
 240  *   merged in from pt->ar_rnat.
 241  *
 242  * o when reading the rnat "above" the last rnat slot on the kernel
 243  *   backing store, rnat0/rnat1 gets its value from sw->ar_rnat.
 244  */
 245 static unsigned long
 246 get_rnat (struct task_struct *task, struct switch_stack *sw,
 247           unsigned long *krbs, unsigned long *urnat_addr,
 248           unsigned long *urbs_end)
 249 {
 250         unsigned long rnat0 = 0, rnat1 = 0, urnat = 0, *slot0_kaddr;
 251         unsigned long umask = 0, mask, m;
 252         unsigned long *kbsp, *ubspstore, *rnat0_kaddr, *rnat1_kaddr, shift;
 253         long num_regs, nbits;
 254         struct pt_regs *pt;
 255
 256         pt = ia64_task_regs(task);
 257         kbsp = (unsigned long *) sw->ar_bspstore;
 258         ubspstore = (unsigned long *) pt->ar_bspstore;
 259
 260         if (urbs_end < urnat_addr)
 261                 nbits = ia64_rse_num_regs(urnat_addr - 63, urbs_end);
 262         else
 263                 nbits = 63;
 264         mask = MASK(nbits);
 265         /*
 266          * First, figure out which bit number slot 0 in user-land maps
 267          * to in the kernel rnat.  Do this by figuring out how many
 268          * register slots we're beyond the user's backingstore and
 269          * then computing the equivalent address in kernel space.
 270          */
 271         num_regs = ia64_rse_num_regs(ubspstore, urnat_addr + 1);
 272         slot0_kaddr = ia64_rse_skip_regs(krbs, num_regs);
 273         shift = ia64_rse_slot_num(slot0_kaddr);
 274         rnat1_kaddr = ia64_rse_rnat_addr(slot0_kaddr);
 275         rnat0_kaddr = rnat1_kaddr - 64;
 276
 277         if (ubspstore + 63 > urnat_addr) {
 278                 /* some bits need to be merged in from pt->ar_rnat */
 279                 umask = MASK(ia64_rse_slot_num(ubspstore)) & mask;
 280                 urnat = (pt->ar_rnat & umask);
 281                 mask &= ~umask;
 282                 if (!mask)
 283                         return urnat;
 284         }
 285
 286         m = mask << shift;
 287         if (rnat0_kaddr >= kbsp)
 288                 rnat0 = sw->ar_rnat;
 289         else if (rnat0_kaddr > krbs)
 290                 rnat0 = *rnat0_kaddr;
 291         urnat |= (rnat0 & m) >> shift;
 292
 293         m = mask >> (63 - shift);
 294         if (rnat1_kaddr >= kbsp)
 295                 rnat1 = sw->ar_rnat;
 296         else if (rnat1_kaddr > krbs)
 297                 rnat1 = *rnat1_kaddr;
 298         urnat |= (rnat1 & m) << (63 - shift);
 299         return urnat;
 300 }
 301
 302 /*
 303  * The reverse of get_rnat.
 304  */
 305 static void
 306 put_rnat (struct task_struct *task, struct switch_stack *sw,
 307           unsigned long *krbs, unsigned long *urnat_addr, unsigned long urnat,
 308           unsigned long *urbs_end)
 309 {
 310         unsigned long rnat0 = 0, rnat1 = 0, *slot0_kaddr, umask = 0, mask, m;
 311         unsigned long *kbsp, *ubspstore, *rnat0_kaddr, *rnat1_kaddr, shift;
 312         long num_regs, nbits;
 313         struct pt_regs *pt;
 314         unsigned long cfm, *urbs_kargs;
 315
 316         pt = ia64_task_regs(task);
 317         kbsp = (unsigned long *) sw->ar_bspstore;
 318         ubspstore = (unsigned long *) pt->ar_bspstore;
 319
 320         urbs_kargs = urbs_end;
 321         if (in_syscall(pt)) {
 322                 /*
 323                  * If entered via syscall, don't allow user to set rnat bits
 324                  * for syscall args.
 325                  */
 326                 cfm = pt->cr_ifs;
 327                 urbs_kargs = ia64_rse_skip_regs(urbs_end, -(cfm & 0x7f));
 328         }
 329
 330         if (urbs_kargs >= urnat_addr)
 331                 nbits = 63;
 332         else {
 333                 if ((urnat_addr - 63) >= urbs_kargs)
 334                         return;
 335                 nbits = ia64_rse_num_regs(urnat_addr - 63, urbs_kargs);
 336         }
 337         mask = MASK(nbits);
 338
 339         /*
 340          * First, figure out which bit number slot 0 in user-land maps
 341          * to in the kernel rnat.  Do this by figuring out how many
 342          * register slots we're beyond the user's backingstore and
 343          * then computing the equivalent address in kernel space.
 344          */
 345         num_regs = ia64_rse_num_regs(ubspstore, urnat_addr + 1);
 346         slot0_kaddr = ia64_rse_skip_regs(krbs, num_regs);
 347         shift = ia64_rse_slot_num(slot0_kaddr);
 348         rnat1_kaddr = ia64_rse_rnat_addr(slot0_kaddr);
 349         rnat0_kaddr = rnat1_kaddr - 64;
 350
 351         if (ubspstore + 63 > urnat_addr) {
 352                 /* some bits need to be place in pt->ar_rnat: */
 353                 umask = MASK(ia64_rse_slot_num(ubspstore)) & mask;
 354                 pt->ar_rnat = (pt->ar_rnat & ~umask) | (urnat & umask);
 355                 mask &= ~umask;
 356                 if (!mask)
 357                         return;
 358         }
 359         /*
 360          * Note: Section 11.1 of the EAS guarantees that bit 63 of an
 361          * rnat slot is ignored. so we don't have to clear it here.
 362          */
 363         rnat0 = (urnat << shift);
 364         m = mask << shift;
 365         if (rnat0_kaddr >= kbsp)
 366                 sw->ar_rnat = (sw->ar_rnat & ~m) | (rnat0 & m);
 367         else if (rnat0_kaddr > krbs)
 368                 *rnat0_kaddr = ((*rnat0_kaddr & ~m) | (rnat0 & m));
 369
 370         rnat1 = (urnat >> (63 - shift));
 371         m = mask >> (63 - shift);
 372         if (rnat1_kaddr >= kbsp)
 373                 sw->ar_rnat = (sw->ar_rnat & ~m) | (rnat1 & m);
 374         else if (rnat1_kaddr > krbs)
 375                 *rnat1_kaddr = ((*rnat1_kaddr & ~m) | (rnat1 & m));
 376 }
 377
 378 static inline int
 379 on_kernel_rbs (unsigned long addr, unsigned long bspstore,
 380                unsigned long urbs_end)
 381 {
 382         unsigned long *rnat_addr = ia64_rse_rnat_addr((unsigned long *)
 383                                                       urbs_end);
 384         return (addr >= bspstore && addr <= (unsigned long) rnat_addr);
 385 }
 386
 387 /*
 388  * Read a word from the user-level backing store of task CHILD.  ADDR
 389  * is the user-level address to read the word from, VAL a pointer to
 390  * the return value, and USER_BSP gives the end of the user-level
 391  * backing store (i.e., it's the address that would be in ar.bsp after
 392  * the user executed a "cover" instruction).
 393  *
 394  * This routine takes care of accessing the kernel register backing
 395  * store for those registers that got spilled there.  It also takes
 396  * care of calculating the appropriate RNaT collection words.
 397  */
 398 long
 399 ia64_peek (struct task_struct *child, struct switch_stack *child_stack,
 400            unsigned long user_rbs_end, unsigned long addr, long *val)
 401 {
 402         unsigned long *bspstore, *krbs, regnum, *laddr, *urbs_end, *rnat_addr;
 403         struct pt_regs *child_regs;
 404         size_t copied;
 405         long ret;
 406
 407         urbs_end = (long *) user_rbs_end;
 408         laddr = (unsigned long *) addr;
 409         child_regs = ia64_task_regs(child);
 410         bspstore = (unsigned long *) child_regs->ar_bspstore;
 411         krbs = (unsigned long *) child + IA64_RBS_OFFSET/8;
 412         if (on_kernel_rbs(addr, (unsigned long) bspstore,
 413                           (unsigned long) urbs_end))
 414         {
 415                 /*
 416                  * Attempt to read the RBS in an area that's actually
 417                  * on the kernel RBS => read the corresponding bits in
 418                  * the kernel RBS.
 419                  */
 420                 rnat_addr = ia64_rse_rnat_addr(laddr);
 421                 ret = get_rnat(child, child_stack, krbs, rnat_addr, urbs_end);
 422
 423                 if (laddr == rnat_addr) {
 424                         /* return NaT collection word itself */
 425                         *val = ret;
 426                         return 0;
 427                 }
 428
 429                 if (((1UL << ia64_rse_slot_num(laddr)) & ret) != 0) {
 430                         /*
 431                          * It is implementation dependent whether the
 432                          * data portion of a NaT value gets saved on a
 433                          * st8.spill or RSE spill (e.g., see EAS 2.6,
 434                          * 4.4.4.6 Register Spill and Fill).  To get
 435                          * consistent behavior across all possible
 436                          * IA-64 implementations, we return zero in
 437                          * this case.
 438                          */
 439                         *val = 0;
 440                         return 0;
 441                 }
 442
 443                 if (laddr < urbs_end) {
 444                         /*
 445                          * The desired word is on the kernel RBS and
 446                          * is not a NaT.
 447                          */
 448                         regnum = ia64_rse_num_regs(bspstore, laddr);
 449                         *val = *ia64_rse_skip_regs(krbs, regnum);
 450                         return 0;
 451                 }
 452         }
 453         copied = access_process_vm(child, addr, &ret, sizeof(ret), 0);
 454         if (copied != sizeof(ret))
 455                 return -EIO;
 456         *val = ret;
 457         return 0;
 458 }
 459
 460 long
 461 ia64_poke (struct task_struct *child, struct switch_stack *child_stack,
 462            unsigned long user_rbs_end, unsigned long addr, long val)
 463 {
 464         unsigned long *bspstore, *krbs, regnum, *laddr;
 465         unsigned long *urbs_end = (long *) user_rbs_end;
 466         struct pt_regs *child_regs;
 467
 468         laddr = (unsigned long *) addr;
 469         child_regs = ia64_task_regs(child);
 470         bspstore = (unsigned long *) child_regs->ar_bspstore;
 471         krbs = (unsigned long *) child + IA64_RBS_OFFSET/8;
 472         if (on_kernel_rbs(addr, (unsigned long) bspstore,
 473                           (unsigned long) urbs_end))
 474         {
 475                 /*
 476                  * Attempt to write the RBS in an area that's actually
 477                  * on the kernel RBS => write the corresponding bits
 478                  * in the kernel RBS.
 479                  */
 480                 if (ia64_rse_is_rnat_slot(laddr))
 481                         put_rnat(child, child_stack, krbs, laddr, val,
 482                                  urbs_end);
 483                 else {
 484                         if (laddr < urbs_end) {
 485                                 regnum = ia64_rse_num_regs(bspstore, laddr);
 486                                 *ia64_rse_skip_regs(krbs, regnum) = val;
 487                         }
 488                 }
 489         } else if (access_process_vm(child, addr, &val, sizeof(val), 1)
 490                    != sizeof(val))
 491                 return -EIO;
 492         return 0;
 493 }
 494
 495 /*
 496  * Calculate the address of the end of the user-level register backing
 497  * store.  This is the address that would have been stored in ar.bsp
 498  * if the user had executed a "cover" instruction right before
 499  * entering the kernel.  If CFMP is not NULL, it is used to return the
 500  * "current frame mask" that was active at the time the kernel was
 501  * entered.
 502  */
 503 unsigned long
 504 ia64_get_user_rbs_end (struct task_struct *child, struct pt_regs *pt,
 505                        unsigned long *cfmp)
 506 {
 507         unsigned long *krbs, *bspstore, cfm = pt->cr_ifs;
 508         long ndirty;
 509
 510         krbs = (unsigned long *) child + IA64_RBS_OFFSET/8;
 511         bspstore = (unsigned long *) pt->ar_bspstore;
 512         ndirty = ia64_rse_num_regs(krbs, krbs + (pt->loadrs >> 19));
 513
 514         if (in_syscall(pt))
 515                 ndirty += (cfm & 0x7f);
 516         else
 517                 cfm &= ~(1UL << 63);    /* clear valid bit */
 518
 519         if (cfmp)
 520                 *cfmp = cfm;
 521         return (unsigned long) ia64_rse_skip_regs(bspstore, ndirty);
 522 }
 523
 524 /*
 525  * Synchronize (i.e, write) the RSE backing store living in kernel
 526  * space to the VM of the CHILD task.  SW and PT are the pointers to
 527  * the switch_stack and pt_regs structures, respectively.
 528  * USER_RBS_END is the user-level address at which the backing store
 529  * ends.
 530  */
 531 long
 532 ia64_sync_user_rbs (struct task_struct *child, struct switch_stack *sw,
 533                     unsigned long user_rbs_start, unsigned long user_rbs_end)
 534 {
 535         unsigned long addr, val;
 536         long ret;
 537
 538         /* now copy word for word from kernel rbs to user rbs: */
 539         for (addr = user_rbs_start; addr < user_rbs_end; addr += 8) {
 540                 ret = ia64_peek(child, sw, user_rbs_end, addr, &val);
 541                 if (ret < 0)
 542                         return ret;
 543                 if (access_process_vm(child, addr, &val, sizeof(val), 1)
 544                     != sizeof(val))
 545                         return -EIO;
 546         }
 547         return 0;
 548 }
 549
 550 static inline int
 551 thread_matches (struct task_struct *thread, unsigned long addr)
 552 {
 553         unsigned long thread_rbs_end;
 554         struct pt_regs *thread_regs;
 555
 556         if (ptrace_check_attach(thread, 0) < 0)
 557                 /*
 558                  * If the thread is not in an attachable state, we'll
 559                  * ignore it.  The net effect is that if ADDR happens
 560                  * to overlap with the portion of the thread's
 561                  * register backing store that is currently residing
 562                  * on the thread's kernel stack, then ptrace() may end
 563                  * up accessing a stale value.  But if the thread
 564                  * isn't stopped, that's a problem anyhow, so we're
 565                  * doing as well as we can...
 566                  */
 567                 return 0;
 568
 569         thread_regs = ia64_task_regs(thread);
 570         thread_rbs_end = ia64_get_user_rbs_end(thread, thread_regs, NULL);
 571         if (!on_kernel_rbs(addr, thread_regs->ar_bspstore, thread_rbs_end))
 572                 return 0;
 573
 574         return 1;       /* looks like we've got a winner */
 575 }
 576
 577 /*
 578  * GDB apparently wants to be able to read the register-backing store
 579  * of any thread when attached to a given process.  If we are peeking
 580  * or poking an address that happens to reside in the kernel-backing
 581  * store of another thread, we need to attach to that thread, because
 582  * otherwise we end up accessing stale data.
 583  *
 584  * task_list_lock must be read-locked before calling this routine!
 585  */
 586 static struct task_struct *
 587 find_thread_for_addr (struct task_struct *child, unsigned long addr)
 588 {
 589         struct task_struct *g, *p;
 590         struct mm_struct *mm;
 591         int mm_users;
 592
 593         if (!(mm = get_task_mm(child)))
 594                 return child;
 595
 596         /* -1 because of our get_task_mm(): */
 597         mm_users = atomic_read(&mm->mm_users) - 1;
 598         if (mm_users <= 1)
 599                 goto out;               /* not multi-threaded */
 600
 601         /*
 602          * First, traverse the child's thread-list.  Good for scalability with
 603          * NPTL-threads.
 604          */
 605         p = child;
 606         do {
 607                 if (thread_matches(p, addr)) {
 608                         child = p;
 609                         goto out;
 610                 }
 611                 if (mm_users-- <= 1)
 612                         goto out;
 613         } while ((p = next_thread(p)) != child);
 614
 615         do_each_thread(g, p) {
 616                 if (child->mm != mm)
 617                         continue;
 618
 619                 if (thread_matches(p, addr)) {
 620                         child = p;
 621                         goto out;
 622                 }
 623         } while_each_thread(g, p);
 624   out:
 625         mmput(mm);
 626         return child;
 627 }
 628
 629 /*
 630  * Write f32-f127 back to task->thread.fph if it has been modified.
 631  */
 632 inline void
 633 ia64_flush_fph (struct task_struct *task)
 634 {
 635         struct ia64_psr *psr = ia64_psr(ia64_task_regs(task));
 636
 637         if (ia64_is_local_fpu_owner(task) && psr->mfh) {
 638                 psr->mfh = 0;
 639                 task->thread.flags |= IA64_THREAD_FPH_VALID;
 640                 ia64_save_fpu(&task->thread.fph[0]);
 641         }
 642 }
 643
 644 /*
 645  * Sync the fph state of the task so that it can be manipulated
 646  * through thread.fph.  If necessary, f32-f127 are written back to
 647  * thread.fph or, if the fph state hasn't been used before, thread.fph
 648  * is cleared to zeroes.  Also, access to f32-f127 is disabled to
 649  * ensure that the task picks up the state from thread.fph when it
 650  * executes again.
 651  */
 652 void
 653 ia64_sync_fph (struct task_struct *task)
 654 {
 655         struct ia64_psr *psr = ia64_psr(ia64_task_regs(task));
 656
 657         ia64_flush_fph(task);
 658         if (!(task->thread.flags & IA64_THREAD_FPH_VALID)) {
 659                 task->thread.flags |= IA64_THREAD_FPH_VALID;
 660                 memset(&task->thread.fph, 0, sizeof(task->thread.fph));
 661         }
 662         ia64_drop_fpu(task);
 663         psr->dfh = 1;
 664 }
 665
 666 static int
 667 access_fr (struct unw_frame_info *info, int regnum, int hi,
 668            unsigned long *data, int write_access)
 669 {
 670         struct ia64_fpreg fpval;
 671         int ret;
 672
 673         ret = unw_get_fr(info, regnum, &fpval);
 674         if (ret < 0)
 675                 return ret;
 676
 677         if (write_access) {
 678                 fpval.u.bits[hi] = *data;
 679                 ret = unw_set_fr(info, regnum, fpval);
 680         } else
 681                 *data = fpval.u.bits[hi];
 682         return ret;
 683 }
 684
 685 /*
 686  * Change the machine-state of CHILD such that it will return via the normal
 687  * kernel exit-path, rather than the syscall-exit path.
 688  */
 689 static void
 690 convert_to_non_syscall (struct task_struct *child, struct pt_regs  *pt,
 691                         unsigned long cfm)
 692 {
 693         struct unw_frame_info info, prev_info;
 694         unsigned long ip, pr;
 695
 696         unw_init_from_blocked_task(&info, child);
 697         while (1) {
 698                 prev_info = info;
 699                 if (unw_unwind(&info) < 0)
 700                         return;
 701                 if (unw_get_rp(&info, &ip) < 0)
 702                         return;
 703                 if (ip < FIXADDR_USER_END)
 704                         break;
 705         }
 706
 707         unw_get_pr(&prev_info, &pr);
 708         pr &= ~(1UL << PRED_SYSCALL);
 709         pr |=  (1UL << PRED_NON_SYSCALL);
 710         unw_set_pr(&prev_info, pr);
 711
 712         pt->cr_ifs = (1UL << 63) | cfm;
 713 }
 714
 715 static int
 716 access_nat_bits (struct task_struct *child, struct pt_regs *pt,
 717                  struct unw_frame_info *info,
 718                  unsigned long *data, int write_access)
 719 {
 720         unsigned long regnum, nat_bits, scratch_unat, dummy = 0;
 721         char nat = 0;
 722
 723         if (write_access) {
 724                 nat_bits = *data;
 725                 scratch_unat = ia64_put_scratch_nat_bits(pt, nat_bits);
 726                 if (unw_set_ar(info, UNW_AR_UNAT, scratch_unat) < 0) {
 727                         dprintk("ptrace: failed to set ar.unat\n");
 728                         return -1;
 729                 }
 730                 for (regnum = 4; regnum <= 7; ++regnum) {
 731                         unw_get_gr(info, regnum, &dummy, &nat);
 732                         unw_set_gr(info, regnum, dummy,
 733                                    (nat_bits >> regnum) & 1);
 734                 }
 735         } else {
 736                 if (unw_get_ar(info, UNW_AR_UNAT, &scratch_unat) < 0) {
 737                         dprintk("ptrace: failed to read ar.unat\n");
 738                         return -1;
 739                 }
 740                 nat_bits = ia64_get_scratch_nat_bits(pt, scratch_unat);
 741                 for (regnum = 4; regnum <= 7; ++regnum) {
 742                         unw_get_gr(info, regnum, &dummy, &nat);
 743                         nat_bits |= (nat != 0) << regnum;
 744                 }
 745                 *data = nat_bits;
 746         }
 747         return 0;
 748 }
 749
 750 static int
 751 access_uarea (struct task_struct *child, unsigned long addr,
 752               unsigned long *data, int write_access)
 753 {
 754         unsigned long *ptr, regnum, urbs_end, rnat_addr, cfm;
 755         struct switch_stack *sw;
 756         struct pt_regs *pt;
 757 #       define pt_reg_addr(pt, reg)     ((void *)                           \
 758                                          ((unsigned long) (pt)              \
 759                                           + offsetof(struct pt_regs, reg)))
 760
 761
 762         pt = ia64_task_regs(child);
 763         sw = (struct switch_stack *) (child->thread.ksp + 16);
 764
 765         if ((addr & 0x7) != 0) {
 766                 dprintk("ptrace: unaligned register address 0x%lx\n", addr);
 767                 return -1;
 768         }
 769
 770         if (addr < PT_F127 + 16) {
 771                 /* accessing fph */
 772                 if (write_access)
 773                         ia64_sync_fph(child);
 774                 else
 775                         ia64_flush_fph(child);
 776                 ptr = (unsigned long *)
 777                         ((unsigned long) &child->thread.fph + addr);
 778         } else if ((addr >= PT_F10) && (addr < PT_F11 + 16)) {
 779                 /* scratch registers untouched by kernel (saved in pt_regs) */
 780                 ptr = pt_reg_addr(pt, f10) + (addr - PT_F10);
 781         } else if (addr >= PT_F12 && addr < PT_F15 + 16) {
 782                 /*
 783                  * Scratch registers untouched by kernel (saved in
 784                  * switch_stack).
 785                  */
 786                 ptr = (unsigned long *) ((long) sw
 787                                          + (addr - PT_NAT_BITS - 32));
 788         } else if (addr < PT_AR_LC + 8) {
 789                 /* preserved state: */
 790                 struct unw_frame_info info;
 791                 char nat = 0;
 792                 int ret;
 793
 794                 unw_init_from_blocked_task(&info, child);
 795                 if (unw_unwind_to_user(&info) < 0)
 796                         return -1;
 797
 798                 switch (addr) {
 799                       case PT_NAT_BITS:
 800                         return access_nat_bits(child, pt, &info,
 801                                                data, write_access);
 802
 803                       case PT_R4: case PT_R5: case PT_R6: case PT_R7:
 804                         if (write_access) {
 805                                 /* read NaT bit first: */
 806                                 unsigned long dummy;
 807
 808                                 ret = unw_get_gr(&info, (addr - PT_R4)/8 + 4,
 809                                                  &dummy, &nat);
 810                                 if (ret < 0)
 811                                         return ret;
 812                         }
 813                         return unw_access_gr(&info, (addr - PT_R4)/8 + 4, data,
 814                                              &nat, write_access);
 815
 816                       case PT_B1: case PT_B2: case PT_B3:
 817                       case PT_B4: case PT_B5:
 818                         return unw_access_br(&info, (addr - PT_B1)/8 + 1, data,
 819                                              write_access);
 820
 821                       case PT_AR_EC:
 822                         return unw_access_ar(&info, UNW_AR_EC, data,
 823                                              write_access);
 824
 825                       case PT_AR_LC:
 826                         return unw_access_ar(&info, UNW_AR_LC, data,
 827                                              write_access);
 828
 829                       default:
 830                         if (addr >= PT_F2 && addr < PT_F5 + 16)
 831                                 return access_fr(&info, (addr - PT_F2)/16 + 2,
 832                                                  (addr & 8) != 0, data,
 833                                                  write_access);
 834                         else if (addr >= PT_F16 && addr < PT_F31 + 16)
 835                                 return access_fr(&info,
 836                                                  (addr - PT_F16)/16 + 16,
 837                                                  (addr & 8) != 0,
 838                                                  data, write_access);
 839                         else {
 840                                 dprintk("ptrace: rejecting access to register "
 841                                         "address 0x%lx\n", addr);
 842                                 return -1;
 843                         }
 844                 }
 845         } else if (addr < PT_F9+16) {
 846                 /* scratch state */
 847                 switch (addr) {
 848                       case PT_AR_BSP:
 849                         /*
 850                          * By convention, we use PT_AR_BSP to refer to
 851                          * the end of the user-level backing store.
 852                          * Use ia64_rse_skip_regs(PT_AR_BSP, -CFM.sof)
 853                          * to get the real value of ar.bsp at the time
 854                          * the kernel was entered.
 855                          *
 856                          * Furthermore, when changing the contents of
 857                          * PT_AR_BSP (or PT_CFM) we MUST copy any
 858                          * users-level stacked registers that are
 859                          * stored on the kernel stack back to
 860                          * user-space because otherwise, we might end
 861                          * up clobbering kernel stacked registers.
 862                          * Also, if this happens while the task is
 863                          * blocked in a system call, which convert the
 864                          * state such that the non-system-call exit
 865                          * path is used.  This ensures that the proper
 866                          * state will be picked up when resuming
 867                          * execution.  However, it *also* means that
 868                          * once we write PT_AR_BSP/PT_CFM, it won't be
 869                          * possible to modify the syscall arguments of
 870                          * the pending system call any longer.  This
 871                          * shouldn't be an issue because modifying
 872                          * PT_AR_BSP/PT_CFM generally implies that
 873                          * we're either abandoning the pending system
 874                          * call or that we defer it's re-execution
 875                          * (e.g., due to GDB doing an inferior
 876                          * function call).
 877                          */
 878                         urbs_end = ia64_get_user_rbs_end(child, pt, &cfm);
 879                         if (write_access) {
 880                                 if (*data != urbs_end) {
 881                                         if (ia64_sync_user_rbs(child, sw,
 882                                                                pt->ar_bspstore,
 883                                                                urbs_end) < 0)
 884                                                 return -1;
 885                                         if (in_syscall(pt))
 886                                                 convert_to_non_syscall(child,
 887                                                                        pt,
 888                                                                        cfm);
 889                                         /*
 890                                          * Simulate user-level write
 891                                          * of ar.bsp:
 892                                          */
 893                                         pt->loadrs = 0;
 894                                         pt->ar_bspstore = *data;
 895                                 }
 896                         } else
 897                                 *data = urbs_end;
 898                         return 0;
 899
 900                       case PT_CFM:
 901                         urbs_end = ia64_get_user_rbs_end(child, pt, &cfm);
 902                         if (write_access) {
 903                                 if (((cfm ^ *data) & PFM_MASK) != 0) {
 904                                         if (ia64_sync_user_rbs(child, sw,
 905                                                                pt->ar_bspstore,
 906                                                                urbs_end) < 0)
 907                                                 return -1;
 908                                         if (in_syscall(pt))
 909                                                 convert_to_non_syscall(child,
 910                                                                        pt,
 911                                                                        cfm);
 912                                         pt->cr_ifs = ((pt->cr_ifs & ~PFM_MASK)
 913                                                       | (*data & PFM_MASK));
 914                                 }
 915                         } else
 916                                 *data = cfm;
 917                         return 0;
 918
 919                       case PT_CR_IPSR:
 920                         if (write_access)
 921                                 pt->cr_ipsr = ((*data & IPSR_MASK)
 922                                                | (pt->cr_ipsr & ~IPSR_MASK));
 923                         else
 924                                 *data = (pt->cr_ipsr & IPSR_MASK);
 925                         return 0;
 926
 927                       case PT_AR_RNAT:
 928                         urbs_end = ia64_get_user_rbs_end(child, pt, NULL);
 929                         rnat_addr = (long) ia64_rse_rnat_addr((long *)
 930                                                               urbs_end);
 931                         if (write_access)
 932                                 return ia64_poke(child, sw, urbs_end,
 933                                                  rnat_addr, *data);
 934                         else
 935                                 return ia64_peek(child, sw, urbs_end,
 936                                                  rnat_addr, data);
 937
 938                       case PT_R1:
 939                         ptr = pt_reg_addr(pt, r1);
 940                         break;
 941                       case PT_R2:  case PT_R3:
 942                         ptr = pt_reg_addr(pt, r2) + (addr - PT_R2);
 943                         break;
 944                       case PT_R8:  case PT_R9:  case PT_R10: case PT_R11:
 945                         ptr = pt_reg_addr(pt, r8) + (addr - PT_R8);
 946                         break;
 947                       case PT_R12: case PT_R13:
 948                         ptr = pt_reg_addr(pt, r12) + (addr - PT_R12);
 949                         break;
 950                       case PT_R14:
 951                         ptr = pt_reg_addr(pt, r14);
 952                         break;
 953                       case PT_R15:
 954                         ptr = pt_reg_addr(pt, r15);
 955                         break;
 956                       case PT_R16: case PT_R17: case PT_R18: case PT_R19:
 957                       case PT_R20: case PT_R21: case PT_R22: case PT_R23:
 958                       case PT_R24: case PT_R25: case PT_R26: case PT_R27:
 959                       case PT_R28: case PT_R29: case PT_R30: case PT_R31:
 960                         ptr = pt_reg_addr(pt, r16) + (addr - PT_R16);
 961                         break;
 962                       case PT_B0:
 963                         ptr = pt_reg_addr(pt, b0);
 964                         break;
 965                       case PT_B6:
 966                         ptr = pt_reg_addr(pt, b6);
 967                         break;
 968                       case PT_B7:
 969                         ptr = pt_reg_addr(pt, b7);
 970                         break;
 971                       case PT_F6:  case PT_F6+8: case PT_F7: case PT_F7+8:
 972                       case PT_F8:  case PT_F8+8: case PT_F9: case PT_F9+8:
 973                         ptr = pt_reg_addr(pt, f6) + (addr - PT_F6);
 974                         break;
 975                       case PT_AR_BSPSTORE:
 976                         ptr = pt_reg_addr(pt, ar_bspstore);
 977                         break;
 978                       case PT_AR_RSC:
 979                         ptr = pt_reg_addr(pt, ar_rsc);
 980                         break;
 981                       case PT_AR_UNAT:
 982                         ptr = pt_reg_addr(pt, ar_unat);
 983                         break;
 984                       case PT_AR_PFS:
 985                         ptr = pt_reg_addr(pt, ar_pfs);
 986                         break;
 987                       case PT_AR_CCV:
 988                         ptr = pt_reg_addr(pt, ar_ccv);
 989                         break;
 990                       case PT_AR_FPSR:
 991                         ptr = pt_reg_addr(pt, ar_fpsr);
 992                         break;
 993                       case PT_CR_IIP:
 994                         ptr = pt_reg_addr(pt, cr_iip);
 995                         break;
 996                       case PT_PR:
 997                         ptr = pt_reg_addr(pt, pr);
 998                         break;
 999                         /* scratch register */
1000
1001                       default:
1002                         /* disallow accessing anything else... */
1003                         dprintk("ptrace: rejecting access to register "
1004                                 "address 0x%lx\n", addr);
1005                         return -1;
1006                 }
1007         } else if (addr <= PT_AR_SSD) {
1008                 ptr = pt_reg_addr(pt, ar_csd) + (addr - PT_AR_CSD);
1009         } else {
1010                 /* access debug registers */
1011
1012                 if (addr >= PT_IBR) {
1013                         regnum = (addr - PT_IBR) >> 3;
1014                         ptr = &child->thread.ibr[0];
1015                 } else {
1016                         regnum = (addr - PT_DBR) >> 3;
1017                         ptr = &child->thread.dbr[0];
1018                 }
1019
1020                 if (regnum >= 8) {
1021                         dprintk("ptrace: rejecting access to register "
1022                                 "address 0x%lx\n", addr);
1023                         return -1;
1024                 }
1025 #ifdef CONFIG_PERFMON
1026                 /*
1027                  * Check if debug registers are used by perfmon. This
1028                  * test must be done once we know that we can do the
1029                  * operation, i.e. the arguments are all valid, but
1030                  * before we start modifying the state.
1031                  *
1032                  * Perfmon needs to keep a count of how many processes
1033                  * are trying to modify the debug registers for system
1034                  * wide monitoring sessions.
1035                  *
1036                  * We also include read access here, because they may
1037                  * cause the PMU-installed debug register state
1038                  * (dbr[], ibr[]) to be reset. The two arrays are also
1039                  * used by perfmon, but we do not use
1040                  * IA64_THREAD_DBG_VALID. The registers are restored
1041                  * by the PMU context switch code.
1042                  */
1043                 if (pfm_use_debug_registers(child)) return -1;
1044 #endif
1045
1046                 if (!(child->thread.flags & IA64_THREAD_DBG_VALID)) {
1047                         child->thread.flags |= IA64_THREAD_DBG_VALID;
1048                         memset(child->thread.dbr, 0,
1049                                sizeof(child->thread.dbr));
1050                         memset(child->thread.ibr, 0,
1051                                sizeof(child->thread.ibr));
1052                 }
1053
1054                 ptr += regnum;
1055
1056                 if ((regnum & 1) && write_access) {
1057                         /* don't let the user set kernel-level breakpoints: */
1058                         *ptr = *data & ~(7UL << 56);
1059                         return 0;
1060                 }
1061         }
1062         if (write_access)
1063                 *ptr = *data;
1064         else
1065                 *data = *ptr;
1066         return 0;
1067 }
1068
1069 static long
1070 ptrace_getregs (struct task_struct *child, struct pt_all_user_regs __user *ppr)
1071 {
1072         unsigned long psr, ec, lc, rnat, bsp, cfm, nat_bits, val;
1073         struct unw_frame_info info;
1074         struct ia64_fpreg fpval;
1075         struct switch_stack *sw;
1076         struct pt_regs *pt;
1077         long ret, retval = 0;
1078         char nat = 0;
1079         int i;
1080
1081         if (!access_ok(VERIFY_WRITE, ppr, sizeof(struct pt_all_user_regs)))
1082                 return -EIO;
1083
1084         pt = ia64_task_regs(child);
1085         sw = (struct switch_stack *) (child->thread.ksp + 16);
1086         unw_init_from_blocked_task(&info, child);
1087         if (unw_unwind_to_user(&info) < 0) {
1088                 return -EIO;
1089         }
1090
1091         if (((unsigned long) ppr & 0x7) != 0) {
1092                 dprintk("ptrace:unaligned register address %p\n", ppr);
1093                 return -EIO;
1094         }
1095
1096         if (access_uarea(child, PT_CR_IPSR, &psr, 0) < 0
1097             || access_uarea(child, PT_AR_EC, &ec, 0) < 0
1098             || access_uarea(child, PT_AR_LC, &lc, 0) < 0
1099             || access_uarea(child, PT_AR_RNAT, &rnat, 0) < 0
1100             || access_uarea(child, PT_AR_BSP, &bsp, 0) < 0
1101             || access_uarea(child, PT_CFM, &cfm, 0)
1102             || access_uarea(child, PT_NAT_BITS, &nat_bits, 0))
1103                 return -EIO;
1104
1105         /* control regs */
1106
1107         retval |= __put_user(pt->cr_iip, &ppr->cr_iip);
1108         retval |= __put_user(psr, &ppr->cr_ipsr);
1109
1110         /* app regs */
1111
1112         retval |= __put_user(pt->ar_pfs, &ppr->ar[PT_AUR_PFS]);
1113         retval |= __put_user(pt->ar_rsc, &ppr->ar[PT_AUR_RSC]);
1114         retval |= __put_user(pt->ar_bspstore, &ppr->ar[PT_AUR_BSPSTORE]);
1115         retval |= __put_user(pt->ar_unat, &ppr->ar[PT_AUR_UNAT]);
1116         retval |= __put_user(pt->ar_ccv, &ppr->ar[PT_AUR_CCV]);
1117         retval |= __put_user(pt->ar_fpsr, &ppr->ar[PT_AUR_FPSR]);
1118
1119         retval |= __put_user(ec, &ppr->ar[PT_AUR_EC]);
1120         retval |= __put_user(lc, &ppr->ar[PT_AUR_LC]);
1121         retval |= __put_user(rnat, &ppr->ar[PT_AUR_RNAT]);
1122         retval |= __put_user(bsp, &ppr->ar[PT_AUR_BSP]);
1123         retval |= __put_user(cfm, &ppr->cfm);
1124
1125         /* gr1-gr3 */
1126
1127         retval |= __copy_to_user(&ppr->gr[1], &pt->r1, sizeof(long));
1128         retval |= __copy_to_user(&ppr->gr[2], &pt->r2, sizeof(long) *2);
1129
1130         /* gr4-gr7 */
1131
1132         for (i = 4; i < 8; i++) {
1133                 if (unw_access_gr(&info, i, &val, &nat, 0) < 0)
1134                         return -EIO;
1135                 retval |= __put_user(val, &ppr->gr[i]);
1136         }
1137
1138         /* gr8-gr11 */
1139
1140         retval |= __copy_to_user(&ppr->gr[8], &pt->r8, sizeof(long) * 4);
1141
1142         /* gr12-gr15 */
1143
1144         retval |= __copy_to_user(&ppr->gr[12], &pt->r12, sizeof(long) * 2);
1145         retval |= __copy_to_user(&ppr->gr[14], &pt->r14, sizeof(long));
1146         retval |= __copy_to_user(&ppr->gr[15], &pt->r15, sizeof(long));
1147
1148         /* gr16-gr31 */
1149
1150         retval |= __copy_to_user(&ppr->gr[16], &pt->r16, sizeof(long) * 16);
1151
1152         /* b0 */
1153
1154         retval |= __put_user(pt->b0, &ppr->br[0]);
1155
1156         /* b1-b5 */
1157
1158         for (i = 1; i < 6; i++) {
1159                 if (unw_access_br(&info, i, &val, 0) < 0)
1160                         return -EIO;
1161                 __put_user(val, &ppr->br[i]);
1162         }
1163
1164         /* b6-b7 */
1165
1166         retval |= __put_user(pt->b6, &ppr->br[6]);
1167         retval |= __put_user(pt->b7, &ppr->br[7]);
1168
1169         /* fr2-fr5 */
1170
1171         for (i = 2; i < 6; i++) {
1172                 if (unw_get_fr(&info, i, &fpval) < 0)
1173                         return -EIO;
1174                 retval |= __copy_to_user(&ppr->fr[i], &fpval, sizeof (fpval));
1175         }
1176
1177         /* fr6-fr11 */
1178
1179         retval |= __copy_to_user(&ppr->fr[6], &pt->f6,
1180                                  sizeof(struct ia64_fpreg) * 6);
1181
1182         /* fp scratch regs(12-15) */
1183
1184         retval |= __copy_to_user(&ppr->fr[12], &sw->f12,
1185                                  sizeof(struct ia64_fpreg) * 4);
1186
1187         /* fr16-fr31 */
1188
1189         for (i = 16; i < 32; i++) {
1190                 if (unw_get_fr(&info, i, &fpval) < 0)
1191                         return -EIO;
1192                 retval |= __copy_to_user(&ppr->fr[i], &fpval, sizeof (fpval));
1193         }
1194
1195         /* fph */
1196
1197         ia64_flush_fph(child);
1198         retval |= __copy_to_user(&ppr->fr[32], &child->thread.fph,
1199                                  sizeof(ppr->fr[32]) * 96);
1200
1201         /*  preds */
1202
1203         retval |= __put_user(pt->pr, &ppr->pr);
1204
1205         /* nat bits */
1206
1207         retval |= __put_user(nat_bits, &ppr->nat);
1208
1209         ret = retval ? -EIO : 0;
1210         return ret;
1211 }
1212
1213 static long
1214 ptrace_setregs (struct task_struct *child, struct pt_all_user_regs __user *ppr)
1215 {
1216         unsigned long psr, ec, lc, rnat, bsp, cfm, nat_bits, val = 0;
1217         struct unw_frame_info info;
1218         struct switch_stack *sw;
1219         struct ia64_fpreg fpval;
1220         struct pt_regs *pt;
1221         long ret, retval = 0;
1222         int i;
1223
1224         memset(&fpval, 0, sizeof(fpval));
1225
1226         if (!access_ok(VERIFY_READ, ppr, sizeof(struct pt_all_user_regs)))
1227                 return -EIO;
1228
1229         pt = ia64_task_regs(child);
1230         sw = (struct switch_stack *) (child->thread.ksp + 16);
1231         unw_init_from_blocked_task(&info, child);
1232         if (unw_unwind_to_user(&info) < 0) {
1233                 return -EIO;
1234         }
1235
1236         if (((unsigned long) ppr & 0x7) != 0) {
1237                 dprintk("ptrace:unaligned register address %p\n", ppr);
1238                 return -EIO;
1239         }
1240
1241         /* control regs */
1242
1243         retval |= __get_user(pt->cr_iip, &ppr->cr_iip);
1244         retval |= __get_user(psr, &ppr->cr_ipsr);
1245
1246         /* app regs */
1247
1248         retval |= __get_user(pt->ar_pfs, &ppr->ar[PT_AUR_PFS]);
1249         retval |= __get_user(pt->ar_rsc, &ppr->ar[PT_AUR_RSC]);
1250         retval |= __get_user(pt->ar_bspstore, &ppr->ar[PT_AUR_BSPSTORE]);
1251         retval |= __get_user(pt->ar_unat, &ppr->ar[PT_AUR_UNAT]);
1252         retval |= __get_user(pt->ar_ccv, &ppr->ar[PT_AUR_CCV]);
1253         retval |= __get_user(pt->ar_fpsr, &ppr->ar[PT_AUR_FPSR]);
1254
1255         retval |= __get_user(ec, &ppr->ar[PT_AUR_EC]);
1256         retval |= __get_user(lc, &ppr->ar[PT_AUR_LC]);
1257         retval |= __get_user(rnat, &ppr->ar[PT_AUR_RNAT]);
1258         retval |= __get_user(bsp, &ppr->ar[PT_AUR_BSP]);
1259         retval |= __get_user(cfm, &ppr->cfm);
1260
1261         /* gr1-gr3 */
1262
1263         retval |= __copy_from_user(&pt->r1, &ppr->gr[1], sizeof(long));
1264         retval |= __copy_from_user(&pt->r2, &ppr->gr[2], sizeof(long) * 2);
1265
1266         /* gr4-gr7 */
1267
1268         for (i = 4; i < 8; i++) {
1269                 retval |= __get_user(val, &ppr->gr[i]);
1270                 /* NaT bit will be set via PT_NAT_BITS: */
1271                 if (unw_set_gr(&info, i, val, 0) < 0)
1272                         return -EIO;
1273         }
1274
1275         /* gr8-gr11 */
1276
1277         retval |= __copy_from_user(&pt->r8, &ppr->gr[8], sizeof(long) * 4);
1278
1279         /* gr12-gr15 */
1280
1281         retval |= __copy_from_user(&pt->r12, &ppr->gr[12], sizeof(long) * 2);
1282         retval |= __copy_from_user(&pt->r14, &ppr->gr[14], sizeof(long));
1283         retval |= __copy_from_user(&pt->r15, &ppr->gr[15], sizeof(long));
1284
1285         /* gr16-gr31 */
1286
1287         retval |= __copy_from_user(&pt->r16, &ppr->gr[16], sizeof(long) * 16);
1288
1289         /* b0 */
1290
1291         retval |= __get_user(pt->b0, &ppr->br[0]);
1292
1293         /* b1-b5 */
1294
1295         for (i = 1; i < 6; i++) {
1296                 retval |= __get_user(val, &ppr->br[i]);
1297                 unw_set_br(&info, i, val);
1298         }
1299
1300         /* b6-b7 */
1301
1302         retval |= __get_user(pt->b6, &ppr->br[6]);
1303         retval |= __get_user(pt->b7, &ppr->br[7]);
1304
1305         /* fr2-fr5 */
1306
1307         for (i = 2; i < 6; i++) {
1308                 retval |= __copy_from_user(&fpval, &ppr->fr[i], sizeof(fpval));
1309                 if (unw_set_fr(&info, i, fpval) < 0)
1310                         return -EIO;
1311         }
1312
1313         /* fr6-fr11 */
1314
1315         retval |= __copy_from_user(&pt->f6, &ppr->fr[6],
1316                                    sizeof(ppr->fr[6]) * 6);
1317
1318         /* fp scratch regs(12-15) */
1319
1320         retval |= __copy_from_user(&sw->f12, &ppr->fr[12],
1321                                    sizeof(ppr->fr[12]) * 4);
1322
1323         /* fr16-fr31 */
1324
1325         for (i = 16; i < 32; i++) {
1326                 retval |= __copy_from_user(&fpval, &ppr->fr[i],
1327                                            sizeof(fpval));
1328                 if (unw_set_fr(&info, i, fpval) < 0)
1329                         return -EIO;
1330         }
1331
1332         /* fph */
1333
1334         ia64_sync_fph(child);
1335         retval |= __copy_from_user(&child->thread.fph, &ppr->fr[32],
1336                                    sizeof(ppr->fr[32]) * 96);
1337
1338         /* preds */
1339
1340         retval |= __get_user(pt->pr, &ppr->pr);
1341
1342         /* nat bits */
1343
1344         retval |= __get_user(nat_bits, &ppr->nat);
1345
1346         retval |= access_uarea(child, PT_CR_IPSR, &psr, 1);
1347         retval |= access_uarea(child, PT_AR_EC, &ec, 1);
1348         retval |= access_uarea(child, PT_AR_LC, &lc, 1);
1349         retval |= access_uarea(child, PT_AR_RNAT, &rnat, 1);
1350         retval |= access_uarea(child, PT_AR_BSP, &bsp, 1);
1351         retval |= access_uarea(child, PT_CFM, &cfm, 1);
1352         retval |= access_uarea(child, PT_NAT_BITS, &nat_bits, 1);
1353
1354         ret = retval ? -EIO : 0;
1355         return ret;
1356 }
1357
1358 /*
1359  * Called by kernel/ptrace.c when detaching..
1360  *
1361  * Make sure the single step bit is not set.
1362  */
1363 void
1364 ptrace_disable (struct task_struct *child)
1365 {
1366         struct ia64_psr *child_psr = ia64_psr(ia64_task_regs(child));
1367
1368         /* make sure the single step/taken-branch trap bits are not set: */
1369         child_psr->ss = 0;
1370         child_psr->tb = 0;
1371 }
1372
1373 asmlinkage long
1374 sys_ptrace (long request, pid_t pid, unsigned long addr, unsigned long data)
1375 {
1376         struct pt_regs *pt;
1377         unsigned long urbs_end, peek_or_poke;
1378         struct task_struct *child;
1379         struct switch_stack *sw;
1380         long ret;
1381
1382         lock_kernel();
1383         ret = -EPERM;
1384         if (request == PTRACE_TRACEME) {
1385                 /* are we already being traced? */
1386                 if (current->ptrace & PT_PTRACED)
1387                         goto out;
1388                 ret = security_ptrace(current->parent, current);
1389                 if (ret)
1390                         goto out;
1391                 current->ptrace |= PT_PTRACED;
1392                 ret = 0;
1393                 goto out;
1394         }
1395
1396         peek_or_poke = (request == PTRACE_PEEKTEXT
1397                         || request == PTRACE_PEEKDATA
1398                         || request == PTRACE_POKETEXT
1399                         || request == PTRACE_POKEDATA);
1400         ret = -ESRCH;
1401         read_lock(&tasklist_lock);
1402         {
1403                 child = find_task_by_pid(pid);
1404                 if (child) {
1405                         if (peek_or_poke)
1406                                 child = find_thread_for_addr(child, addr);
1407                         get_task_struct(child);
1408                 }
1409         }
1410         read_unlock(&tasklist_lock);
1411         if (!child)
1412                 goto out;
1413         ret = -EPERM;
1414         if (pid == 1)           /* no messing around with init! */
1415                 goto out_tsk;
1416
1417         if (request == PTRACE_ATTACH) {
1418                 ret = ptrace_attach(child);
1419                 goto out_tsk;
1420         }
1421
1422         ret = ptrace_check_attach(child, request == PTRACE_KILL);
1423         if (ret < 0)
1424                 goto out_tsk;
1425
1426         pt = ia64_task_regs(child);
1427         sw = (struct switch_stack *) (child->thread.ksp + 16);
1428
1429         switch (request) {
1430               case PTRACE_PEEKTEXT:
1431               case PTRACE_PEEKDATA:
1432                 /* read word at location addr */
1433                 urbs_end = ia64_get_user_rbs_end(child, pt, NULL);
1434                 ret = ia64_peek(child, sw, urbs_end, addr, &data);
1435                 if (ret == 0) {
1436                         ret = data;
1437                         /* ensure "ret" is not mistaken as an error code: */
1438                         force_successful_syscall_return();
1439                 }
1440                 goto out_tsk;
1441
1442               case PTRACE_POKETEXT:
1443               case PTRACE_POKEDATA:
1444                 /* write the word at location addr */
1445                 urbs_end = ia64_get_user_rbs_end(child, pt, NULL);
1446                 ret = ia64_poke(child, sw, urbs_end, addr, data);
1447                 goto out_tsk;
1448
1449               case PTRACE_PEEKUSR:
1450                 /* read the word at addr in the USER area */
1451                 if (access_uarea(child, addr, &data, 0) < 0) {
1452                         ret = -EIO;
1453                         goto out_tsk;
1454                 }
1455                 ret = data;
1456                 /* ensure "ret" is not mistaken as an error code */
1457                 force_successful_syscall_return();
1458                 goto out_tsk;
1459
1460               case PTRACE_POKEUSR:
1461                 /* write the word at addr in the USER area */
1462                 if (access_uarea(child, addr, &data, 1) < 0) {
1463                         ret = -EIO;
1464                         goto out_tsk;
1465                 }
1466                 ret = 0;
1467                 goto out_tsk;
1468
1469               case PTRACE_OLD_GETSIGINFO:
1470                 /* for backwards-compatibility */
1471                 ret = ptrace_request(child, PTRACE_GETSIGINFO, addr, data);
1472                 goto out_tsk;
1473
1474               case PTRACE_OLD_SETSIGINFO:
1475                 /* for backwards-compatibility */
1476                 ret = ptrace_request(child, PTRACE_SETSIGINFO, addr, data);
1477                 goto out_tsk;
1478
1479               case PTRACE_SYSCALL:
1480                 /* continue and stop at next (return from) syscall */
1481               case PTRACE_CONT:
1482                 /* restart after signal. */
1483                 ret = -EIO;
1484                 if (data > _NSIG)
1485                         goto out_tsk;
1486                 if (request == PTRACE_SYSCALL)
1487                         set_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
1488                 else
1489                         clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
1490                 child->exit_code = data;
1491
1492                 /*
1493                  * Make sure the single step/taken-branch trap bits
1494                  * are not set:
1495                  */
1496                 ia64_psr(pt)->ss = 0;
1497                 ia64_psr(pt)->tb = 0;
1498
1499                 wake_up_process(child);
1500                 ret = 0;
1501                 goto out_tsk;
1502
1503               case PTRACE_KILL:
1504                 /*
1505                  * Make the child exit.  Best I can do is send it a
1506                  * sigkill.  Perhaps it should be put in the status
1507                  * that it wants to exit.
1508                  */
1509                 if (child->exit_state == EXIT_ZOMBIE)
1510                         /* already dead */
1511                         goto out_tsk;
1512                 child->exit_code = SIGKILL;
1513
1514                 ptrace_disable(child);
1515                 wake_up_process(child);
1516                 ret = 0;
1517                 goto out_tsk;
1518
1519               case PTRACE_SINGLESTEP:
1520                 /* let child execute for one instruction */
1521               case PTRACE_SINGLEBLOCK:
1522                 ret = -EIO;
1523                 if (data > _NSIG)
1524                         goto out_tsk;
1525
1526                 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
1527                 if (request == PTRACE_SINGLESTEP) {
1528                         ia64_psr(pt)->ss = 1;
1529                 } else {
1530                         ia64_psr(pt)->tb = 1;
1531                 }
1532                 child->exit_code = data;
1533
1534                 /* give it a chance to run. */
1535                 wake_up_process(child);
1536                 ret = 0;
1537                 goto out_tsk;
1538
1539               case PTRACE_DETACH:
1540                 /* detach a process that was attached. */
1541                 ret = ptrace_detach(child, data);
1542                 goto out_tsk;
1543
1544               case PTRACE_GETREGS:
1545                 ret = ptrace_getregs(child,
1546                                      (struct pt_all_user_regs __user *) data);
1547                 goto out_tsk;
1548
1549               case PTRACE_SETREGS:
1550                 ret = ptrace_setregs(child,
1551                                      (struct pt_all_user_regs __user *) data);
1552                 goto out_tsk;
1553
1554               default:
1555                 ret = ptrace_request(child, request, addr, data);
1556                 goto out_tsk;
1557         }
1558   out_tsk:
1559         put_task_struct(child);
1560   out:
1561         unlock_kernel();
1562         return ret;
1563 }
1564
1565
1566 void
1567 syscall_trace (void)
1568 {
1569         if (!test_thread_flag(TIF_SYSCALL_TRACE))
1570                 return;
1571         if (!(current->ptrace & PT_PTRACED))
1572                 return;
1573         /*
1574          * The 0x80 provides a way for the tracing parent to
1575          * distinguish between a syscall stop and SIGTRAP delivery.
1576          */
1577         ptrace_notify(SIGTRAP
1578                       | ((current->ptrace & PT_TRACESYSGOOD) ? 0x80 : 0));
1579
1580         /*
1581          * This isn't the same as continuing with a signal, but it
1582          * will do for normal use.  strace only continues with a
1583          * signal if the stopping signal is not SIGTRAP.  -brl
1584          */
1585         if (current->exit_code) {
1586                 send_sig(current->exit_code, current, 1);
1587                 current->exit_code = 0;
1588         }
1589 }
1590
1591 /* "asmlinkage" so the input arguments are preserved... */
1592
1593 asmlinkage void
1594 syscall_trace_enter (long arg0, long arg1, long arg2, long arg3,
1595                      long arg4, long arg5, long arg6, long arg7,
1596                      struct pt_regs regs)
1597 {
1598         long syscall;
1599
1600         if (unlikely(current->audit_context)) {
1601                 if (IS_IA32_PROCESS(&regs))
1602                         syscall = regs.r1;
1603                 else
1604                         syscall = regs.r15;
1605
1606                 audit_syscall_entry(current, syscall, arg0, arg1, arg2, arg3);
1607         }
1608
1609         if (test_thread_flag(TIF_SYSCALL_TRACE)
1610             && (current->ptrace & PT_PTRACED))
1611                 syscall_trace();
1612 }
1613
1614 /* "asmlinkage" so the input arguments are preserved... */
1615
1616 asmlinkage void
1617 syscall_trace_leave (long arg0, long arg1, long arg2, long arg3,
1618                      long arg4, long arg5, long arg6, long arg7,
1619                      struct pt_regs regs)
1620 {
1621         if (unlikely(current->audit_context))
1622                 audit_syscall_exit(current, regs.r8);
1623
1624         if (test_thread_flag(TIF_SYSCALL_TRACE)
1625             && (current->ptrace & PT_PTRACED))
1626                 syscall_trace();
1627 }