pid_ns: kill the now unused task_child_reaper()
[linux-2.6] / 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  * Copyright (C) 2006 Intel Co
7  *  2006-08-12  - IA64 Native Utrace implementation support added by
8  *      Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com>
9  *
10  * Derived from the x86 and Alpha versions.
11  */
12 #include <linux/kernel.h>
13 #include <linux/sched.h>
14 #include <linux/slab.h>
15 #include <linux/mm.h>
16 #include <linux/errno.h>
17 #include <linux/ptrace.h>
18 #include <linux/smp_lock.h>
19 #include <linux/user.h>
20 #include <linux/security.h>
21 #include <linux/audit.h>
22 #include <linux/signal.h>
23 #include <linux/regset.h>
24 #include <linux/elf.h>
25
26 #include <asm/pgtable.h>
27 #include <asm/processor.h>
28 #include <asm/ptrace_offsets.h>
29 #include <asm/rse.h>
30 #include <asm/system.h>
31 #include <asm/uaccess.h>
32 #include <asm/unwind.h>
33 #ifdef CONFIG_PERFMON
34 #include <asm/perfmon.h>
35 #endif
36
37 #include "entry.h"
38
39 /*
40  * Bits in the PSR that we allow ptrace() to change:
41  *      be, up, ac, mfl, mfh (the user mask; five bits total)
42  *      db (debug breakpoint fault; one bit)
43  *      id (instruction debug fault disable; one bit)
44  *      dd (data debug fault disable; one bit)
45  *      ri (restart instruction; two bits)
46  *      is (instruction set; one bit)
47  */
48 #define IPSR_MASK (IA64_PSR_UM | IA64_PSR_DB | IA64_PSR_IS      \
49                    | IA64_PSR_ID | IA64_PSR_DD | IA64_PSR_RI)
50
51 #define MASK(nbits)     ((1UL << (nbits)) - 1)  /* mask with NBITS bits set */
52 #define PFM_MASK        MASK(38)
53
54 #define PTRACE_DEBUG    0
55
56 #if PTRACE_DEBUG
57 # define dprintk(format...)     printk(format)
58 # define inline
59 #else
60 # define dprintk(format...)
61 #endif
62
63 /* Return TRUE if PT was created due to kernel-entry via a system-call.  */
64
65 static inline int
66 in_syscall (struct pt_regs *pt)
67 {
68         return (long) pt->cr_ifs >= 0;
69 }
70
71 /*
72  * Collect the NaT bits for r1-r31 from scratch_unat and return a NaT
73  * bitset where bit i is set iff the NaT bit of register i is set.
74  */
75 unsigned long
76 ia64_get_scratch_nat_bits (struct pt_regs *pt, unsigned long scratch_unat)
77 {
78 #       define GET_BITS(first, last, unat)                              \
79         ({                                                              \
80                 unsigned long bit = ia64_unat_pos(&pt->r##first);       \
81                 unsigned long nbits = (last - first + 1);               \
82                 unsigned long mask = MASK(nbits) << first;              \
83                 unsigned long dist;                                     \
84                 if (bit < first)                                        \
85                         dist = 64 + bit - first;                        \
86                 else                                                    \
87                         dist = bit - first;                             \
88                 ia64_rotr(unat, dist) & mask;                           \
89         })
90         unsigned long val;
91
92         /*
93          * Registers that are stored consecutively in struct pt_regs
94          * can be handled in parallel.  If the register order in
95          * struct_pt_regs changes, this code MUST be updated.
96          */
97         val  = GET_BITS( 1,  1, scratch_unat);
98         val |= GET_BITS( 2,  3, scratch_unat);
99         val |= GET_BITS(12, 13, scratch_unat);
100         val |= GET_BITS(14, 14, scratch_unat);
101         val |= GET_BITS(15, 15, scratch_unat);
102         val |= GET_BITS( 8, 11, scratch_unat);
103         val |= GET_BITS(16, 31, scratch_unat);
104         return val;
105
106 #       undef GET_BITS
107 }
108
109 /*
110  * Set the NaT bits for the scratch registers according to NAT and
111  * return the resulting unat (assuming the scratch registers are
112  * stored in PT).
113  */
114 unsigned long
115 ia64_put_scratch_nat_bits (struct pt_regs *pt, unsigned long nat)
116 {
117 #       define PUT_BITS(first, last, nat)                               \
118         ({                                                              \
119                 unsigned long bit = ia64_unat_pos(&pt->r##first);       \
120                 unsigned long nbits = (last - first + 1);               \
121                 unsigned long mask = MASK(nbits) << first;              \
122                 long dist;                                              \
123                 if (bit < first)                                        \
124                         dist = 64 + bit - first;                        \
125                 else                                                    \
126                         dist = bit - first;                             \
127                 ia64_rotl(nat & mask, dist);                            \
128         })
129         unsigned long scratch_unat;
130
131         /*
132          * Registers that are stored consecutively in struct pt_regs
133          * can be handled in parallel.  If the register order in
134          * struct_pt_regs changes, this code MUST be updated.
135          */
136         scratch_unat  = PUT_BITS( 1,  1, nat);
137         scratch_unat |= PUT_BITS( 2,  3, nat);
138         scratch_unat |= PUT_BITS(12, 13, nat);
139         scratch_unat |= PUT_BITS(14, 14, nat);
140         scratch_unat |= PUT_BITS(15, 15, nat);
141         scratch_unat |= PUT_BITS( 8, 11, nat);
142         scratch_unat |= PUT_BITS(16, 31, nat);
143
144         return scratch_unat;
145
146 #       undef PUT_BITS
147 }
148
149 #define IA64_MLX_TEMPLATE       0x2
150 #define IA64_MOVL_OPCODE        6
151
152 void
153 ia64_increment_ip (struct pt_regs *regs)
154 {
155         unsigned long w0, ri = ia64_psr(regs)->ri + 1;
156
157         if (ri > 2) {
158                 ri = 0;
159                 regs->cr_iip += 16;
160         } else if (ri == 2) {
161                 get_user(w0, (char __user *) regs->cr_iip + 0);
162                 if (((w0 >> 1) & 0xf) == IA64_MLX_TEMPLATE) {
163                         /*
164                          * rfi'ing to slot 2 of an MLX bundle causes
165                          * an illegal operation fault.  We don't want
166                          * that to happen...
167                          */
168                         ri = 0;
169                         regs->cr_iip += 16;
170                 }
171         }
172         ia64_psr(regs)->ri = ri;
173 }
174
175 void
176 ia64_decrement_ip (struct pt_regs *regs)
177 {
178         unsigned long w0, ri = ia64_psr(regs)->ri - 1;
179
180         if (ia64_psr(regs)->ri == 0) {
181                 regs->cr_iip -= 16;
182                 ri = 2;
183                 get_user(w0, (char __user *) regs->cr_iip + 0);
184                 if (((w0 >> 1) & 0xf) == IA64_MLX_TEMPLATE) {
185                         /*
186                          * rfi'ing to slot 2 of an MLX bundle causes
187                          * an illegal operation fault.  We don't want
188                          * that to happen...
189                          */
190                         ri = 1;
191                 }
192         }
193         ia64_psr(regs)->ri = ri;
194 }
195
196 /*
197  * This routine is used to read an rnat bits that are stored on the
198  * kernel backing store.  Since, in general, the alignment of the user
199  * and kernel are different, this is not completely trivial.  In
200  * essence, we need to construct the user RNAT based on up to two
201  * kernel RNAT values and/or the RNAT value saved in the child's
202  * pt_regs.
203  *
204  * user rbs
205  *
206  * +--------+ <-- lowest address
207  * | slot62 |
208  * +--------+
209  * |  rnat  | 0x....1f8
210  * +--------+
211  * | slot00 | \
212  * +--------+ |
213  * | slot01 | > child_regs->ar_rnat
214  * +--------+ |
215  * | slot02 | /                         kernel rbs
216  * +--------+                           +--------+
217  *          <- child_regs->ar_bspstore  | slot61 | <-- krbs
218  * +- - - - +                           +--------+
219  *                                      | slot62 |
220  * +- - - - +                           +--------+
221  *                                      |  rnat  |
222  * +- - - - +                           +--------+
223  *   vrnat                              | slot00 |
224  * +- - - - +                           +--------+
225  *                                      =        =
226  *                                      +--------+
227  *                                      | slot00 | \
228  *                                      +--------+ |
229  *                                      | slot01 | > child_stack->ar_rnat
230  *                                      +--------+ |
231  *                                      | slot02 | /
232  *                                      +--------+
233  *                                                <--- child_stack->ar_bspstore
234  *
235  * The way to think of this code is as follows: bit 0 in the user rnat
236  * corresponds to some bit N (0 <= N <= 62) in one of the kernel rnat
237  * value.  The kernel rnat value holding this bit is stored in
238  * variable rnat0.  rnat1 is loaded with the kernel rnat value that
239  * form the upper bits of the user rnat value.
240  *
241  * Boundary cases:
242  *
243  * o when reading the rnat "below" the first rnat slot on the kernel
244  *   backing store, rnat0/rnat1 are set to 0 and the low order bits are
245  *   merged in from pt->ar_rnat.
246  *
247  * o when reading the rnat "above" the last rnat slot on the kernel
248  *   backing store, rnat0/rnat1 gets its value from sw->ar_rnat.
249  */
250 static unsigned long
251 get_rnat (struct task_struct *task, struct switch_stack *sw,
252           unsigned long *krbs, unsigned long *urnat_addr,
253           unsigned long *urbs_end)
254 {
255         unsigned long rnat0 = 0, rnat1 = 0, urnat = 0, *slot0_kaddr;
256         unsigned long umask = 0, mask, m;
257         unsigned long *kbsp, *ubspstore, *rnat0_kaddr, *rnat1_kaddr, shift;
258         long num_regs, nbits;
259         struct pt_regs *pt;
260
261         pt = task_pt_regs(task);
262         kbsp = (unsigned long *) sw->ar_bspstore;
263         ubspstore = (unsigned long *) pt->ar_bspstore;
264
265         if (urbs_end < urnat_addr)
266                 nbits = ia64_rse_num_regs(urnat_addr - 63, urbs_end);
267         else
268                 nbits = 63;
269         mask = MASK(nbits);
270         /*
271          * First, figure out which bit number slot 0 in user-land maps
272          * to in the kernel rnat.  Do this by figuring out how many
273          * register slots we're beyond the user's backingstore and
274          * then computing the equivalent address in kernel space.
275          */
276         num_regs = ia64_rse_num_regs(ubspstore, urnat_addr + 1);
277         slot0_kaddr = ia64_rse_skip_regs(krbs, num_regs);
278         shift = ia64_rse_slot_num(slot0_kaddr);
279         rnat1_kaddr = ia64_rse_rnat_addr(slot0_kaddr);
280         rnat0_kaddr = rnat1_kaddr - 64;
281
282         if (ubspstore + 63 > urnat_addr) {
283                 /* some bits need to be merged in from pt->ar_rnat */
284                 umask = MASK(ia64_rse_slot_num(ubspstore)) & mask;
285                 urnat = (pt->ar_rnat & umask);
286                 mask &= ~umask;
287                 if (!mask)
288                         return urnat;
289         }
290
291         m = mask << shift;
292         if (rnat0_kaddr >= kbsp)
293                 rnat0 = sw->ar_rnat;
294         else if (rnat0_kaddr > krbs)
295                 rnat0 = *rnat0_kaddr;
296         urnat |= (rnat0 & m) >> shift;
297
298         m = mask >> (63 - shift);
299         if (rnat1_kaddr >= kbsp)
300                 rnat1 = sw->ar_rnat;
301         else if (rnat1_kaddr > krbs)
302                 rnat1 = *rnat1_kaddr;
303         urnat |= (rnat1 & m) << (63 - shift);
304         return urnat;
305 }
306
307 /*
308  * The reverse of get_rnat.
309  */
310 static void
311 put_rnat (struct task_struct *task, struct switch_stack *sw,
312           unsigned long *krbs, unsigned long *urnat_addr, unsigned long urnat,
313           unsigned long *urbs_end)
314 {
315         unsigned long rnat0 = 0, rnat1 = 0, *slot0_kaddr, umask = 0, mask, m;
316         unsigned long *kbsp, *ubspstore, *rnat0_kaddr, *rnat1_kaddr, shift;
317         long num_regs, nbits;
318         struct pt_regs *pt;
319         unsigned long cfm, *urbs_kargs;
320
321         pt = task_pt_regs(task);
322         kbsp = (unsigned long *) sw->ar_bspstore;
323         ubspstore = (unsigned long *) pt->ar_bspstore;
324
325         urbs_kargs = urbs_end;
326         if (in_syscall(pt)) {
327                 /*
328                  * If entered via syscall, don't allow user to set rnat bits
329                  * for syscall args.
330                  */
331                 cfm = pt->cr_ifs;
332                 urbs_kargs = ia64_rse_skip_regs(urbs_end, -(cfm & 0x7f));
333         }
334
335         if (urbs_kargs >= urnat_addr)
336                 nbits = 63;
337         else {
338                 if ((urnat_addr - 63) >= urbs_kargs)
339                         return;
340                 nbits = ia64_rse_num_regs(urnat_addr - 63, urbs_kargs);
341         }
342         mask = MASK(nbits);
343
344         /*
345          * First, figure out which bit number slot 0 in user-land maps
346          * to in the kernel rnat.  Do this by figuring out how many
347          * register slots we're beyond the user's backingstore and
348          * then computing the equivalent address in kernel space.
349          */
350         num_regs = ia64_rse_num_regs(ubspstore, urnat_addr + 1);
351         slot0_kaddr = ia64_rse_skip_regs(krbs, num_regs);
352         shift = ia64_rse_slot_num(slot0_kaddr);
353         rnat1_kaddr = ia64_rse_rnat_addr(slot0_kaddr);
354         rnat0_kaddr = rnat1_kaddr - 64;
355
356         if (ubspstore + 63 > urnat_addr) {
357                 /* some bits need to be place in pt->ar_rnat: */
358                 umask = MASK(ia64_rse_slot_num(ubspstore)) & mask;
359                 pt->ar_rnat = (pt->ar_rnat & ~umask) | (urnat & umask);
360                 mask &= ~umask;
361                 if (!mask)
362                         return;
363         }
364         /*
365          * Note: Section 11.1 of the EAS guarantees that bit 63 of an
366          * rnat slot is ignored. so we don't have to clear it here.
367          */
368         rnat0 = (urnat << shift);
369         m = mask << shift;
370         if (rnat0_kaddr >= kbsp)
371                 sw->ar_rnat = (sw->ar_rnat & ~m) | (rnat0 & m);
372         else if (rnat0_kaddr > krbs)
373                 *rnat0_kaddr = ((*rnat0_kaddr & ~m) | (rnat0 & m));
374
375         rnat1 = (urnat >> (63 - shift));
376         m = mask >> (63 - shift);
377         if (rnat1_kaddr >= kbsp)
378                 sw->ar_rnat = (sw->ar_rnat & ~m) | (rnat1 & m);
379         else if (rnat1_kaddr > krbs)
380                 *rnat1_kaddr = ((*rnat1_kaddr & ~m) | (rnat1 & m));
381 }
382
383 static inline int
384 on_kernel_rbs (unsigned long addr, unsigned long bspstore,
385                unsigned long urbs_end)
386 {
387         unsigned long *rnat_addr = ia64_rse_rnat_addr((unsigned long *)
388                                                       urbs_end);
389         return (addr >= bspstore && addr <= (unsigned long) rnat_addr);
390 }
391
392 /*
393  * Read a word from the user-level backing store of task CHILD.  ADDR
394  * is the user-level address to read the word from, VAL a pointer to
395  * the return value, and USER_BSP gives the end of the user-level
396  * backing store (i.e., it's the address that would be in ar.bsp after
397  * the user executed a "cover" instruction).
398  *
399  * This routine takes care of accessing the kernel register backing
400  * store for those registers that got spilled there.  It also takes
401  * care of calculating the appropriate RNaT collection words.
402  */
403 long
404 ia64_peek (struct task_struct *child, struct switch_stack *child_stack,
405            unsigned long user_rbs_end, unsigned long addr, long *val)
406 {
407         unsigned long *bspstore, *krbs, regnum, *laddr, *urbs_end, *rnat_addr;
408         struct pt_regs *child_regs;
409         size_t copied;
410         long ret;
411
412         urbs_end = (long *) user_rbs_end;
413         laddr = (unsigned long *) addr;
414         child_regs = task_pt_regs(child);
415         bspstore = (unsigned long *) child_regs->ar_bspstore;
416         krbs = (unsigned long *) child + IA64_RBS_OFFSET/8;
417         if (on_kernel_rbs(addr, (unsigned long) bspstore,
418                           (unsigned long) urbs_end))
419         {
420                 /*
421                  * Attempt to read the RBS in an area that's actually
422                  * on the kernel RBS => read the corresponding bits in
423                  * the kernel RBS.
424                  */
425                 rnat_addr = ia64_rse_rnat_addr(laddr);
426                 ret = get_rnat(child, child_stack, krbs, rnat_addr, urbs_end);
427
428                 if (laddr == rnat_addr) {
429                         /* return NaT collection word itself */
430                         *val = ret;
431                         return 0;
432                 }
433
434                 if (((1UL << ia64_rse_slot_num(laddr)) & ret) != 0) {
435                         /*
436                          * It is implementation dependent whether the
437                          * data portion of a NaT value gets saved on a
438                          * st8.spill or RSE spill (e.g., see EAS 2.6,
439                          * 4.4.4.6 Register Spill and Fill).  To get
440                          * consistent behavior across all possible
441                          * IA-64 implementations, we return zero in
442                          * this case.
443                          */
444                         *val = 0;
445                         return 0;
446                 }
447
448                 if (laddr < urbs_end) {
449                         /*
450                          * The desired word is on the kernel RBS and
451                          * is not a NaT.
452                          */
453                         regnum = ia64_rse_num_regs(bspstore, laddr);
454                         *val = *ia64_rse_skip_regs(krbs, regnum);
455                         return 0;
456                 }
457         }
458         copied = access_process_vm(child, addr, &ret, sizeof(ret), 0);
459         if (copied != sizeof(ret))
460                 return -EIO;
461         *val = ret;
462         return 0;
463 }
464
465 long
466 ia64_poke (struct task_struct *child, struct switch_stack *child_stack,
467            unsigned long user_rbs_end, unsigned long addr, long val)
468 {
469         unsigned long *bspstore, *krbs, regnum, *laddr;
470         unsigned long *urbs_end = (long *) user_rbs_end;
471         struct pt_regs *child_regs;
472
473         laddr = (unsigned long *) addr;
474         child_regs = task_pt_regs(child);
475         bspstore = (unsigned long *) child_regs->ar_bspstore;
476         krbs = (unsigned long *) child + IA64_RBS_OFFSET/8;
477         if (on_kernel_rbs(addr, (unsigned long) bspstore,
478                           (unsigned long) urbs_end))
479         {
480                 /*
481                  * Attempt to write the RBS in an area that's actually
482                  * on the kernel RBS => write the corresponding bits
483                  * in the kernel RBS.
484                  */
485                 if (ia64_rse_is_rnat_slot(laddr))
486                         put_rnat(child, child_stack, krbs, laddr, val,
487                                  urbs_end);
488                 else {
489                         if (laddr < urbs_end) {
490                                 regnum = ia64_rse_num_regs(bspstore, laddr);
491                                 *ia64_rse_skip_regs(krbs, regnum) = val;
492                         }
493                 }
494         } else if (access_process_vm(child, addr, &val, sizeof(val), 1)
495                    != sizeof(val))
496                 return -EIO;
497         return 0;
498 }
499
500 /*
501  * Calculate the address of the end of the user-level register backing
502  * store.  This is the address that would have been stored in ar.bsp
503  * if the user had executed a "cover" instruction right before
504  * entering the kernel.  If CFMP is not NULL, it is used to return the
505  * "current frame mask" that was active at the time the kernel was
506  * entered.
507  */
508 unsigned long
509 ia64_get_user_rbs_end (struct task_struct *child, struct pt_regs *pt,
510                        unsigned long *cfmp)
511 {
512         unsigned long *krbs, *bspstore, cfm = pt->cr_ifs;
513         long ndirty;
514
515         krbs = (unsigned long *) child + IA64_RBS_OFFSET/8;
516         bspstore = (unsigned long *) pt->ar_bspstore;
517         ndirty = ia64_rse_num_regs(krbs, krbs + (pt->loadrs >> 19));
518
519         if (in_syscall(pt))
520                 ndirty += (cfm & 0x7f);
521         else
522                 cfm &= ~(1UL << 63);    /* clear valid bit */
523
524         if (cfmp)
525                 *cfmp = cfm;
526         return (unsigned long) ia64_rse_skip_regs(bspstore, ndirty);
527 }
528
529 /*
530  * Synchronize (i.e, write) the RSE backing store living in kernel
531  * space to the VM of the CHILD task.  SW and PT are the pointers to
532  * the switch_stack and pt_regs structures, respectively.
533  * USER_RBS_END is the user-level address at which the backing store
534  * ends.
535  */
536 long
537 ia64_sync_user_rbs (struct task_struct *child, struct switch_stack *sw,
538                     unsigned long user_rbs_start, unsigned long user_rbs_end)
539 {
540         unsigned long addr, val;
541         long ret;
542
543         /* now copy word for word from kernel rbs to user rbs: */
544         for (addr = user_rbs_start; addr < user_rbs_end; addr += 8) {
545                 ret = ia64_peek(child, sw, user_rbs_end, addr, &val);
546                 if (ret < 0)
547                         return ret;
548                 if (access_process_vm(child, addr, &val, sizeof(val), 1)
549                     != sizeof(val))
550                         return -EIO;
551         }
552         return 0;
553 }
554
555 static long
556 ia64_sync_kernel_rbs (struct task_struct *child, struct switch_stack *sw,
557                 unsigned long user_rbs_start, unsigned long user_rbs_end)
558 {
559         unsigned long addr, val;
560         long ret;
561
562         /* now copy word for word from user rbs to kernel rbs: */
563         for (addr = user_rbs_start; addr < user_rbs_end; addr += 8) {
564                 if (access_process_vm(child, addr, &val, sizeof(val), 0)
565                                 != sizeof(val))
566                         return -EIO;
567
568                 ret = ia64_poke(child, sw, user_rbs_end, addr, val);
569                 if (ret < 0)
570                         return ret;
571         }
572         return 0;
573 }
574
575 typedef long (*syncfunc_t)(struct task_struct *, struct switch_stack *,
576                             unsigned long, unsigned long);
577
578 static void do_sync_rbs(struct unw_frame_info *info, void *arg)
579 {
580         struct pt_regs *pt;
581         unsigned long urbs_end;
582         syncfunc_t fn = arg;
583
584         if (unw_unwind_to_user(info) < 0)
585                 return;
586         pt = task_pt_regs(info->task);
587         urbs_end = ia64_get_user_rbs_end(info->task, pt, NULL);
588
589         fn(info->task, info->sw, pt->ar_bspstore, urbs_end);
590 }
591
592 /*
593  * when a thread is stopped (ptraced), debugger might change thread's user
594  * stack (change memory directly), and we must avoid the RSE stored in kernel
595  * to override user stack (user space's RSE is newer than kernel's in the
596  * case). To workaround the issue, we copy kernel RSE to user RSE before the
597  * task is stopped, so user RSE has updated data.  we then copy user RSE to
598  * kernel after the task is resummed from traced stop and kernel will use the
599  * newer RSE to return to user. TIF_RESTORE_RSE is the flag to indicate we need
600  * synchronize user RSE to kernel.
601  */
602 void ia64_ptrace_stop(void)
603 {
604         if (test_and_set_tsk_thread_flag(current, TIF_RESTORE_RSE))
605                 return;
606         tsk_set_notify_resume(current);
607         unw_init_running(do_sync_rbs, ia64_sync_user_rbs);
608 }
609
610 /*
611  * This is called to read back the register backing store.
612  */
613 void ia64_sync_krbs(void)
614 {
615         clear_tsk_thread_flag(current, TIF_RESTORE_RSE);
616         tsk_clear_notify_resume(current);
617
618         unw_init_running(do_sync_rbs, ia64_sync_kernel_rbs);
619 }
620
621 /*
622  * After PTRACE_ATTACH, a thread's register backing store area in user
623  * space is assumed to contain correct data whenever the thread is
624  * stopped.  arch_ptrace_stop takes care of this on tracing stops.
625  * But if the child was already stopped for job control when we attach
626  * to it, then it might not ever get into ptrace_stop by the time we
627  * want to examine the user memory containing the RBS.
628  */
629 void
630 ptrace_attach_sync_user_rbs (struct task_struct *child)
631 {
632         int stopped = 0;
633         struct unw_frame_info info;
634
635         /*
636          * If the child is in TASK_STOPPED, we need to change that to
637          * TASK_TRACED momentarily while we operate on it.  This ensures
638          * that the child won't be woken up and return to user mode while
639          * we are doing the sync.  (It can only be woken up for SIGKILL.)
640          */
641
642         read_lock(&tasklist_lock);
643         if (child->signal) {
644                 spin_lock_irq(&child->sighand->siglock);
645                 if (child->state == TASK_STOPPED &&
646                     !test_and_set_tsk_thread_flag(child, TIF_RESTORE_RSE)) {
647                         tsk_set_notify_resume(child);
648
649                         child->state = TASK_TRACED;
650                         stopped = 1;
651                 }
652                 spin_unlock_irq(&child->sighand->siglock);
653         }
654         read_unlock(&tasklist_lock);
655
656         if (!stopped)
657                 return;
658
659         unw_init_from_blocked_task(&info, child);
660         do_sync_rbs(&info, ia64_sync_user_rbs);
661
662         /*
663          * Now move the child back into TASK_STOPPED if it should be in a
664          * job control stop, so that SIGCONT can be used to wake it up.
665          */
666         read_lock(&tasklist_lock);
667         if (child->signal) {
668                 spin_lock_irq(&child->sighand->siglock);
669                 if (child->state == TASK_TRACED &&
670                     (child->signal->flags & SIGNAL_STOP_STOPPED)) {
671                         child->state = TASK_STOPPED;
672                 }
673                 spin_unlock_irq(&child->sighand->siglock);
674         }
675         read_unlock(&tasklist_lock);
676 }
677
678 static inline int
679 thread_matches (struct task_struct *thread, unsigned long addr)
680 {
681         unsigned long thread_rbs_end;
682         struct pt_regs *thread_regs;
683
684         if (ptrace_check_attach(thread, 0) < 0)
685                 /*
686                  * If the thread is not in an attachable state, we'll
687                  * ignore it.  The net effect is that if ADDR happens
688                  * to overlap with the portion of the thread's
689                  * register backing store that is currently residing
690                  * on the thread's kernel stack, then ptrace() may end
691                  * up accessing a stale value.  But if the thread
692                  * isn't stopped, that's a problem anyhow, so we're
693                  * doing as well as we can...
694                  */
695                 return 0;
696
697         thread_regs = task_pt_regs(thread);
698         thread_rbs_end = ia64_get_user_rbs_end(thread, thread_regs, NULL);
699         if (!on_kernel_rbs(addr, thread_regs->ar_bspstore, thread_rbs_end))
700                 return 0;
701
702         return 1;       /* looks like we've got a winner */
703 }
704
705 /*
706  * Write f32-f127 back to task->thread.fph if it has been modified.
707  */
708 inline void
709 ia64_flush_fph (struct task_struct *task)
710 {
711         struct ia64_psr *psr = ia64_psr(task_pt_regs(task));
712
713         /*
714          * Prevent migrating this task while
715          * we're fiddling with the FPU state
716          */
717         preempt_disable();
718         if (ia64_is_local_fpu_owner(task) && psr->mfh) {
719                 psr->mfh = 0;
720                 task->thread.flags |= IA64_THREAD_FPH_VALID;
721                 ia64_save_fpu(&task->thread.fph[0]);
722         }
723         preempt_enable();
724 }
725
726 /*
727  * Sync the fph state of the task so that it can be manipulated
728  * through thread.fph.  If necessary, f32-f127 are written back to
729  * thread.fph or, if the fph state hasn't been used before, thread.fph
730  * is cleared to zeroes.  Also, access to f32-f127 is disabled to
731  * ensure that the task picks up the state from thread.fph when it
732  * executes again.
733  */
734 void
735 ia64_sync_fph (struct task_struct *task)
736 {
737         struct ia64_psr *psr = ia64_psr(task_pt_regs(task));
738
739         ia64_flush_fph(task);
740         if (!(task->thread.flags & IA64_THREAD_FPH_VALID)) {
741                 task->thread.flags |= IA64_THREAD_FPH_VALID;
742                 memset(&task->thread.fph, 0, sizeof(task->thread.fph));
743         }
744         ia64_drop_fpu(task);
745         psr->dfh = 1;
746 }
747
748 /*
749  * Change the machine-state of CHILD such that it will return via the normal
750  * kernel exit-path, rather than the syscall-exit path.
751  */
752 static void
753 convert_to_non_syscall (struct task_struct *child, struct pt_regs  *pt,
754                         unsigned long cfm)
755 {
756         struct unw_frame_info info, prev_info;
757         unsigned long ip, sp, pr;
758
759         unw_init_from_blocked_task(&info, child);
760         while (1) {
761                 prev_info = info;
762                 if (unw_unwind(&info) < 0)
763                         return;
764
765                 unw_get_sp(&info, &sp);
766                 if ((long)((unsigned long)child + IA64_STK_OFFSET - sp)
767                     < IA64_PT_REGS_SIZE) {
768                         dprintk("ptrace.%s: ran off the top of the kernel "
769                                 "stack\n", __func__);
770                         return;
771                 }
772                 if (unw_get_pr (&prev_info, &pr) < 0) {
773                         unw_get_rp(&prev_info, &ip);
774                         dprintk("ptrace.%s: failed to read "
775                                 "predicate register (ip=0x%lx)\n",
776                                 __func__, ip);
777                         return;
778                 }
779                 if (unw_is_intr_frame(&info)
780                     && (pr & (1UL << PRED_USER_STACK)))
781                         break;
782         }
783
784         /*
785          * Note: at the time of this call, the target task is blocked
786          * in notify_resume_user() and by clearling PRED_LEAVE_SYSCALL
787          * (aka, "pLvSys") we redirect execution from
788          * .work_pending_syscall_end to .work_processed_kernel.
789          */
790         unw_get_pr(&prev_info, &pr);
791         pr &= ~((1UL << PRED_SYSCALL) | (1UL << PRED_LEAVE_SYSCALL));
792         pr |=  (1UL << PRED_NON_SYSCALL);
793         unw_set_pr(&prev_info, pr);
794
795         pt->cr_ifs = (1UL << 63) | cfm;
796         /*
797          * Clear the memory that is NOT written on syscall-entry to
798          * ensure we do not leak kernel-state to user when execution
799          * resumes.
800          */
801         pt->r2 = 0;
802         pt->r3 = 0;
803         pt->r14 = 0;
804         memset(&pt->r16, 0, 16*8);      /* clear r16-r31 */
805         memset(&pt->f6, 0, 6*16);       /* clear f6-f11 */
806         pt->b7 = 0;
807         pt->ar_ccv = 0;
808         pt->ar_csd = 0;
809         pt->ar_ssd = 0;
810 }
811
812 static int
813 access_nat_bits (struct task_struct *child, struct pt_regs *pt,
814                  struct unw_frame_info *info,
815                  unsigned long *data, int write_access)
816 {
817         unsigned long regnum, nat_bits, scratch_unat, dummy = 0;
818         char nat = 0;
819
820         if (write_access) {
821                 nat_bits = *data;
822                 scratch_unat = ia64_put_scratch_nat_bits(pt, nat_bits);
823                 if (unw_set_ar(info, UNW_AR_UNAT, scratch_unat) < 0) {
824                         dprintk("ptrace: failed to set ar.unat\n");
825                         return -1;
826                 }
827                 for (regnum = 4; regnum <= 7; ++regnum) {
828                         unw_get_gr(info, regnum, &dummy, &nat);
829                         unw_set_gr(info, regnum, dummy,
830                                    (nat_bits >> regnum) & 1);
831                 }
832         } else {
833                 if (unw_get_ar(info, UNW_AR_UNAT, &scratch_unat) < 0) {
834                         dprintk("ptrace: failed to read ar.unat\n");
835                         return -1;
836                 }
837                 nat_bits = ia64_get_scratch_nat_bits(pt, scratch_unat);
838                 for (regnum = 4; regnum <= 7; ++regnum) {
839                         unw_get_gr(info, regnum, &dummy, &nat);
840                         nat_bits |= (nat != 0) << regnum;
841                 }
842                 *data = nat_bits;
843         }
844         return 0;
845 }
846
847 static int
848 access_uarea (struct task_struct *child, unsigned long addr,
849               unsigned long *data, int write_access);
850
851 static long
852 ptrace_getregs (struct task_struct *child, struct pt_all_user_regs __user *ppr)
853 {
854         unsigned long psr, ec, lc, rnat, bsp, cfm, nat_bits, val;
855         struct unw_frame_info info;
856         struct ia64_fpreg fpval;
857         struct switch_stack *sw;
858         struct pt_regs *pt;
859         long ret, retval = 0;
860         char nat = 0;
861         int i;
862
863         if (!access_ok(VERIFY_WRITE, ppr, sizeof(struct pt_all_user_regs)))
864                 return -EIO;
865
866         pt = task_pt_regs(child);
867         sw = (struct switch_stack *) (child->thread.ksp + 16);
868         unw_init_from_blocked_task(&info, child);
869         if (unw_unwind_to_user(&info) < 0) {
870                 return -EIO;
871         }
872
873         if (((unsigned long) ppr & 0x7) != 0) {
874                 dprintk("ptrace:unaligned register address %p\n", ppr);
875                 return -EIO;
876         }
877
878         if (access_uarea(child, PT_CR_IPSR, &psr, 0) < 0
879             || access_uarea(child, PT_AR_EC, &ec, 0) < 0
880             || access_uarea(child, PT_AR_LC, &lc, 0) < 0
881             || access_uarea(child, PT_AR_RNAT, &rnat, 0) < 0
882             || access_uarea(child, PT_AR_BSP, &bsp, 0) < 0
883             || access_uarea(child, PT_CFM, &cfm, 0)
884             || access_uarea(child, PT_NAT_BITS, &nat_bits, 0))
885                 return -EIO;
886
887         /* control regs */
888
889         retval |= __put_user(pt->cr_iip, &ppr->cr_iip);
890         retval |= __put_user(psr, &ppr->cr_ipsr);
891
892         /* app regs */
893
894         retval |= __put_user(pt->ar_pfs, &ppr->ar[PT_AUR_PFS]);
895         retval |= __put_user(pt->ar_rsc, &ppr->ar[PT_AUR_RSC]);
896         retval |= __put_user(pt->ar_bspstore, &ppr->ar[PT_AUR_BSPSTORE]);
897         retval |= __put_user(pt->ar_unat, &ppr->ar[PT_AUR_UNAT]);
898         retval |= __put_user(pt->ar_ccv, &ppr->ar[PT_AUR_CCV]);
899         retval |= __put_user(pt->ar_fpsr, &ppr->ar[PT_AUR_FPSR]);
900
901         retval |= __put_user(ec, &ppr->ar[PT_AUR_EC]);
902         retval |= __put_user(lc, &ppr->ar[PT_AUR_LC]);
903         retval |= __put_user(rnat, &ppr->ar[PT_AUR_RNAT]);
904         retval |= __put_user(bsp, &ppr->ar[PT_AUR_BSP]);
905         retval |= __put_user(cfm, &ppr->cfm);
906
907         /* gr1-gr3 */
908
909         retval |= __copy_to_user(&ppr->gr[1], &pt->r1, sizeof(long));
910         retval |= __copy_to_user(&ppr->gr[2], &pt->r2, sizeof(long) *2);
911
912         /* gr4-gr7 */
913
914         for (i = 4; i < 8; i++) {
915                 if (unw_access_gr(&info, i, &val, &nat, 0) < 0)
916                         return -EIO;
917                 retval |= __put_user(val, &ppr->gr[i]);
918         }
919
920         /* gr8-gr11 */
921
922         retval |= __copy_to_user(&ppr->gr[8], &pt->r8, sizeof(long) * 4);
923
924         /* gr12-gr15 */
925
926         retval |= __copy_to_user(&ppr->gr[12], &pt->r12, sizeof(long) * 2);
927         retval |= __copy_to_user(&ppr->gr[14], &pt->r14, sizeof(long));
928         retval |= __copy_to_user(&ppr->gr[15], &pt->r15, sizeof(long));
929
930         /* gr16-gr31 */
931
932         retval |= __copy_to_user(&ppr->gr[16], &pt->r16, sizeof(long) * 16);
933
934         /* b0 */
935
936         retval |= __put_user(pt->b0, &ppr->br[0]);
937
938         /* b1-b5 */
939
940         for (i = 1; i < 6; i++) {
941                 if (unw_access_br(&info, i, &val, 0) < 0)
942                         return -EIO;
943                 __put_user(val, &ppr->br[i]);
944         }
945
946         /* b6-b7 */
947
948         retval |= __put_user(pt->b6, &ppr->br[6]);
949         retval |= __put_user(pt->b7, &ppr->br[7]);
950
951         /* fr2-fr5 */
952
953         for (i = 2; i < 6; i++) {
954                 if (unw_get_fr(&info, i, &fpval) < 0)
955                         return -EIO;
956                 retval |= __copy_to_user(&ppr->fr[i], &fpval, sizeof (fpval));
957         }
958
959         /* fr6-fr11 */
960
961         retval |= __copy_to_user(&ppr->fr[6], &pt->f6,
962                                  sizeof(struct ia64_fpreg) * 6);
963
964         /* fp scratch regs(12-15) */
965
966         retval |= __copy_to_user(&ppr->fr[12], &sw->f12,
967                                  sizeof(struct ia64_fpreg) * 4);
968
969         /* fr16-fr31 */
970
971         for (i = 16; i < 32; i++) {
972                 if (unw_get_fr(&info, i, &fpval) < 0)
973                         return -EIO;
974                 retval |= __copy_to_user(&ppr->fr[i], &fpval, sizeof (fpval));
975         }
976
977         /* fph */
978
979         ia64_flush_fph(child);
980         retval |= __copy_to_user(&ppr->fr[32], &child->thread.fph,
981                                  sizeof(ppr->fr[32]) * 96);
982
983         /*  preds */
984
985         retval |= __put_user(pt->pr, &ppr->pr);
986
987         /* nat bits */
988
989         retval |= __put_user(nat_bits, &ppr->nat);
990
991         ret = retval ? -EIO : 0;
992         return ret;
993 }
994
995 static long
996 ptrace_setregs (struct task_struct *child, struct pt_all_user_regs __user *ppr)
997 {
998         unsigned long psr, rsc, ec, lc, rnat, bsp, cfm, nat_bits, val = 0;
999         struct unw_frame_info info;
1000         struct switch_stack *sw;
1001         struct ia64_fpreg fpval;
1002         struct pt_regs *pt;
1003         long ret, retval = 0;
1004         int i;
1005
1006         memset(&fpval, 0, sizeof(fpval));
1007
1008         if (!access_ok(VERIFY_READ, ppr, sizeof(struct pt_all_user_regs)))
1009                 return -EIO;
1010
1011         pt = task_pt_regs(child);
1012         sw = (struct switch_stack *) (child->thread.ksp + 16);
1013         unw_init_from_blocked_task(&info, child);
1014         if (unw_unwind_to_user(&info) < 0) {
1015                 return -EIO;
1016         }
1017
1018         if (((unsigned long) ppr & 0x7) != 0) {
1019                 dprintk("ptrace:unaligned register address %p\n", ppr);
1020                 return -EIO;
1021         }
1022
1023         /* control regs */
1024
1025         retval |= __get_user(pt->cr_iip, &ppr->cr_iip);
1026         retval |= __get_user(psr, &ppr->cr_ipsr);
1027
1028         /* app regs */
1029
1030         retval |= __get_user(pt->ar_pfs, &ppr->ar[PT_AUR_PFS]);
1031         retval |= __get_user(rsc, &ppr->ar[PT_AUR_RSC]);
1032         retval |= __get_user(pt->ar_bspstore, &ppr->ar[PT_AUR_BSPSTORE]);
1033         retval |= __get_user(pt->ar_unat, &ppr->ar[PT_AUR_UNAT]);
1034         retval |= __get_user(pt->ar_ccv, &ppr->ar[PT_AUR_CCV]);
1035         retval |= __get_user(pt->ar_fpsr, &ppr->ar[PT_AUR_FPSR]);
1036
1037         retval |= __get_user(ec, &ppr->ar[PT_AUR_EC]);
1038         retval |= __get_user(lc, &ppr->ar[PT_AUR_LC]);
1039         retval |= __get_user(rnat, &ppr->ar[PT_AUR_RNAT]);
1040         retval |= __get_user(bsp, &ppr->ar[PT_AUR_BSP]);
1041         retval |= __get_user(cfm, &ppr->cfm);
1042
1043         /* gr1-gr3 */
1044
1045         retval |= __copy_from_user(&pt->r1, &ppr->gr[1], sizeof(long));
1046         retval |= __copy_from_user(&pt->r2, &ppr->gr[2], sizeof(long) * 2);
1047
1048         /* gr4-gr7 */
1049
1050         for (i = 4; i < 8; i++) {
1051                 retval |= __get_user(val, &ppr->gr[i]);
1052                 /* NaT bit will be set via PT_NAT_BITS: */
1053                 if (unw_set_gr(&info, i, val, 0) < 0)
1054                         return -EIO;
1055         }
1056
1057         /* gr8-gr11 */
1058
1059         retval |= __copy_from_user(&pt->r8, &ppr->gr[8], sizeof(long) * 4);
1060
1061         /* gr12-gr15 */
1062
1063         retval |= __copy_from_user(&pt->r12, &ppr->gr[12], sizeof(long) * 2);
1064         retval |= __copy_from_user(&pt->r14, &ppr->gr[14], sizeof(long));
1065         retval |= __copy_from_user(&pt->r15, &ppr->gr[15], sizeof(long));
1066
1067         /* gr16-gr31 */
1068
1069         retval |= __copy_from_user(&pt->r16, &ppr->gr[16], sizeof(long) * 16);
1070
1071         /* b0 */
1072
1073         retval |= __get_user(pt->b0, &ppr->br[0]);
1074
1075         /* b1-b5 */
1076
1077         for (i = 1; i < 6; i++) {
1078                 retval |= __get_user(val, &ppr->br[i]);
1079                 unw_set_br(&info, i, val);
1080         }
1081
1082         /* b6-b7 */
1083
1084         retval |= __get_user(pt->b6, &ppr->br[6]);
1085         retval |= __get_user(pt->b7, &ppr->br[7]);
1086
1087         /* fr2-fr5 */
1088
1089         for (i = 2; i < 6; i++) {
1090                 retval |= __copy_from_user(&fpval, &ppr->fr[i], sizeof(fpval));
1091                 if (unw_set_fr(&info, i, fpval) < 0)
1092                         return -EIO;
1093         }
1094
1095         /* fr6-fr11 */
1096
1097         retval |= __copy_from_user(&pt->f6, &ppr->fr[6],
1098                                    sizeof(ppr->fr[6]) * 6);
1099
1100         /* fp scratch regs(12-15) */
1101
1102         retval |= __copy_from_user(&sw->f12, &ppr->fr[12],
1103                                    sizeof(ppr->fr[12]) * 4);
1104
1105         /* fr16-fr31 */
1106
1107         for (i = 16; i < 32; i++) {
1108                 retval |= __copy_from_user(&fpval, &ppr->fr[i],
1109                                            sizeof(fpval));
1110                 if (unw_set_fr(&info, i, fpval) < 0)
1111                         return -EIO;
1112         }
1113
1114         /* fph */
1115
1116         ia64_sync_fph(child);
1117         retval |= __copy_from_user(&child->thread.fph, &ppr->fr[32],
1118                                    sizeof(ppr->fr[32]) * 96);
1119
1120         /* preds */
1121
1122         retval |= __get_user(pt->pr, &ppr->pr);
1123
1124         /* nat bits */
1125
1126         retval |= __get_user(nat_bits, &ppr->nat);
1127
1128         retval |= access_uarea(child, PT_CR_IPSR, &psr, 1);
1129         retval |= access_uarea(child, PT_AR_RSC, &rsc, 1);
1130         retval |= access_uarea(child, PT_AR_EC, &ec, 1);
1131         retval |= access_uarea(child, PT_AR_LC, &lc, 1);
1132         retval |= access_uarea(child, PT_AR_RNAT, &rnat, 1);
1133         retval |= access_uarea(child, PT_AR_BSP, &bsp, 1);
1134         retval |= access_uarea(child, PT_CFM, &cfm, 1);
1135         retval |= access_uarea(child, PT_NAT_BITS, &nat_bits, 1);
1136
1137         ret = retval ? -EIO : 0;
1138         return ret;
1139 }
1140
1141 void
1142 user_enable_single_step (struct task_struct *child)
1143 {
1144         struct ia64_psr *child_psr = ia64_psr(task_pt_regs(child));
1145
1146         set_tsk_thread_flag(child, TIF_SINGLESTEP);
1147         child_psr->ss = 1;
1148 }
1149
1150 void
1151 user_enable_block_step (struct task_struct *child)
1152 {
1153         struct ia64_psr *child_psr = ia64_psr(task_pt_regs(child));
1154
1155         set_tsk_thread_flag(child, TIF_SINGLESTEP);
1156         child_psr->tb = 1;
1157 }
1158
1159 void
1160 user_disable_single_step (struct task_struct *child)
1161 {
1162         struct ia64_psr *child_psr = ia64_psr(task_pt_regs(child));
1163
1164         /* make sure the single step/taken-branch trap bits are not set: */
1165         clear_tsk_thread_flag(child, TIF_SINGLESTEP);
1166         child_psr->ss = 0;
1167         child_psr->tb = 0;
1168 }
1169
1170 /*
1171  * Called by kernel/ptrace.c when detaching..
1172  *
1173  * Make sure the single step bit is not set.
1174  */
1175 void
1176 ptrace_disable (struct task_struct *child)
1177 {
1178         user_disable_single_step(child);
1179 }
1180
1181 long
1182 arch_ptrace (struct task_struct *child, long request, long addr, long data)
1183 {
1184         switch (request) {
1185         case PTRACE_PEEKTEXT:
1186         case PTRACE_PEEKDATA:
1187                 /* read word at location addr */
1188                 if (access_process_vm(child, addr, &data, sizeof(data), 0)
1189                     != sizeof(data))
1190                         return -EIO;
1191                 /* ensure return value is not mistaken for error code */
1192                 force_successful_syscall_return();
1193                 return data;
1194
1195         /* PTRACE_POKETEXT and PTRACE_POKEDATA is handled
1196          * by the generic ptrace_request().
1197          */
1198
1199         case PTRACE_PEEKUSR:
1200                 /* read the word at addr in the USER area */
1201                 if (access_uarea(child, addr, &data, 0) < 0)
1202                         return -EIO;
1203                 /* ensure return value is not mistaken for error code */
1204                 force_successful_syscall_return();
1205                 return data;
1206
1207         case PTRACE_POKEUSR:
1208                 /* write the word at addr in the USER area */
1209                 if (access_uarea(child, addr, &data, 1) < 0)
1210                         return -EIO;
1211                 return 0;
1212
1213         case PTRACE_OLD_GETSIGINFO:
1214                 /* for backwards-compatibility */
1215                 return ptrace_request(child, PTRACE_GETSIGINFO, addr, data);
1216
1217         case PTRACE_OLD_SETSIGINFO:
1218                 /* for backwards-compatibility */
1219                 return ptrace_request(child, PTRACE_SETSIGINFO, addr, data);
1220
1221         case PTRACE_GETREGS:
1222                 return ptrace_getregs(child,
1223                                       (struct pt_all_user_regs __user *) data);
1224
1225         case PTRACE_SETREGS:
1226                 return ptrace_setregs(child,
1227                                       (struct pt_all_user_regs __user *) data);
1228
1229         default:
1230                 return ptrace_request(child, request, addr, data);
1231         }
1232 }
1233
1234
1235 static void
1236 syscall_trace (void)
1237 {
1238         /*
1239          * The 0x80 provides a way for the tracing parent to
1240          * distinguish between a syscall stop and SIGTRAP delivery.
1241          */
1242         ptrace_notify(SIGTRAP
1243                       | ((current->ptrace & PT_TRACESYSGOOD) ? 0x80 : 0));
1244
1245         /*
1246          * This isn't the same as continuing with a signal, but it
1247          * will do for normal use.  strace only continues with a
1248          * signal if the stopping signal is not SIGTRAP.  -brl
1249          */
1250         if (current->exit_code) {
1251                 send_sig(current->exit_code, current, 1);
1252                 current->exit_code = 0;
1253         }
1254 }
1255
1256 /* "asmlinkage" so the input arguments are preserved... */
1257
1258 asmlinkage void
1259 syscall_trace_enter (long arg0, long arg1, long arg2, long arg3,
1260                      long arg4, long arg5, long arg6, long arg7,
1261                      struct pt_regs regs)
1262 {
1263         if (test_thread_flag(TIF_SYSCALL_TRACE) 
1264             && (current->ptrace & PT_PTRACED))
1265                 syscall_trace();
1266
1267         /* copy user rbs to kernel rbs */
1268         if (test_thread_flag(TIF_RESTORE_RSE))
1269                 ia64_sync_krbs();
1270
1271         if (unlikely(current->audit_context)) {
1272                 long syscall;
1273                 int arch;
1274
1275                 if (IS_IA32_PROCESS(&regs)) {
1276                         syscall = regs.r1;
1277                         arch = AUDIT_ARCH_I386;
1278                 } else {
1279                         syscall = regs.r15;
1280                         arch = AUDIT_ARCH_IA64;
1281                 }
1282
1283                 audit_syscall_entry(arch, syscall, arg0, arg1, arg2, arg3);
1284         }
1285
1286 }
1287
1288 /* "asmlinkage" so the input arguments are preserved... */
1289
1290 asmlinkage void
1291 syscall_trace_leave (long arg0, long arg1, long arg2, long arg3,
1292                      long arg4, long arg5, long arg6, long arg7,
1293                      struct pt_regs regs)
1294 {
1295         if (unlikely(current->audit_context)) {
1296                 int success = AUDITSC_RESULT(regs.r10);
1297                 long result = regs.r8;
1298
1299                 if (success != AUDITSC_SUCCESS)
1300                         result = -result;
1301                 audit_syscall_exit(success, result);
1302         }
1303
1304         if ((test_thread_flag(TIF_SYSCALL_TRACE)
1305             || test_thread_flag(TIF_SINGLESTEP))
1306             && (current->ptrace & PT_PTRACED))
1307                 syscall_trace();
1308
1309         /* copy user rbs to kernel rbs */
1310         if (test_thread_flag(TIF_RESTORE_RSE))
1311                 ia64_sync_krbs();
1312 }
1313
1314 /* Utrace implementation starts here */
1315 struct regset_get {
1316         void *kbuf;
1317         void __user *ubuf;
1318 };
1319
1320 struct regset_set {
1321         const void *kbuf;
1322         const void __user *ubuf;
1323 };
1324
1325 struct regset_getset {
1326         struct task_struct *target;
1327         const struct user_regset *regset;
1328         union {
1329                 struct regset_get get;
1330                 struct regset_set set;
1331         } u;
1332         unsigned int pos;
1333         unsigned int count;
1334         int ret;
1335 };
1336
1337 static int
1338 access_elf_gpreg(struct task_struct *target, struct unw_frame_info *info,
1339                 unsigned long addr, unsigned long *data, int write_access)
1340 {
1341         struct pt_regs *pt;
1342         unsigned long *ptr = NULL;
1343         int ret;
1344         char nat = 0;
1345
1346         pt = task_pt_regs(target);
1347         switch (addr) {
1348         case ELF_GR_OFFSET(1):
1349                 ptr = &pt->r1;
1350                 break;
1351         case ELF_GR_OFFSET(2):
1352         case ELF_GR_OFFSET(3):
1353                 ptr = (void *)&pt->r2 + (addr - ELF_GR_OFFSET(2));
1354                 break;
1355         case ELF_GR_OFFSET(4) ... ELF_GR_OFFSET(7):
1356                 if (write_access) {
1357                         /* read NaT bit first: */
1358                         unsigned long dummy;
1359
1360                         ret = unw_get_gr(info, addr/8, &dummy, &nat);
1361                         if (ret < 0)
1362                                 return ret;
1363                 }
1364                 return unw_access_gr(info, addr/8, data, &nat, write_access);
1365         case ELF_GR_OFFSET(8) ... ELF_GR_OFFSET(11):
1366                 ptr = (void *)&pt->r8 + addr - ELF_GR_OFFSET(8);
1367                 break;
1368         case ELF_GR_OFFSET(12):
1369         case ELF_GR_OFFSET(13):
1370                 ptr = (void *)&pt->r12 + addr - ELF_GR_OFFSET(12);
1371                 break;
1372         case ELF_GR_OFFSET(14):
1373                 ptr = &pt->r14;
1374                 break;
1375         case ELF_GR_OFFSET(15):
1376                 ptr = &pt->r15;
1377         }
1378         if (write_access)
1379                 *ptr = *data;
1380         else
1381                 *data = *ptr;
1382         return 0;
1383 }
1384
1385 static int
1386 access_elf_breg(struct task_struct *target, struct unw_frame_info *info,
1387                 unsigned long addr, unsigned long *data, int write_access)
1388 {
1389         struct pt_regs *pt;
1390         unsigned long *ptr = NULL;
1391
1392         pt = task_pt_regs(target);
1393         switch (addr) {
1394         case ELF_BR_OFFSET(0):
1395                 ptr = &pt->b0;
1396                 break;
1397         case ELF_BR_OFFSET(1) ... ELF_BR_OFFSET(5):
1398                 return unw_access_br(info, (addr - ELF_BR_OFFSET(0))/8,
1399                                      data, write_access);
1400         case ELF_BR_OFFSET(6):
1401                 ptr = &pt->b6;
1402                 break;
1403         case ELF_BR_OFFSET(7):
1404                 ptr = &pt->b7;
1405         }
1406         if (write_access)
1407                 *ptr = *data;
1408         else
1409                 *data = *ptr;
1410         return 0;
1411 }
1412
1413 static int
1414 access_elf_areg(struct task_struct *target, struct unw_frame_info *info,
1415                 unsigned long addr, unsigned long *data, int write_access)
1416 {
1417         struct pt_regs *pt;
1418         unsigned long cfm, urbs_end;
1419         unsigned long *ptr = NULL;
1420
1421         pt = task_pt_regs(target);
1422         if (addr >= ELF_AR_RSC_OFFSET && addr <= ELF_AR_SSD_OFFSET) {
1423                 switch (addr) {
1424                 case ELF_AR_RSC_OFFSET:
1425                         /* force PL3 */
1426                         if (write_access)
1427                                 pt->ar_rsc = *data | (3 << 2);
1428                         else
1429                                 *data = pt->ar_rsc;
1430                         return 0;
1431                 case ELF_AR_BSP_OFFSET:
1432                         /*
1433                          * By convention, we use PT_AR_BSP to refer to
1434                          * the end of the user-level backing store.
1435                          * Use ia64_rse_skip_regs(PT_AR_BSP, -CFM.sof)
1436                          * to get the real value of ar.bsp at the time
1437                          * the kernel was entered.
1438                          *
1439                          * Furthermore, when changing the contents of
1440                          * PT_AR_BSP (or PT_CFM) while the task is
1441                          * blocked in a system call, convert the state
1442                          * so that the non-system-call exit
1443                          * path is used.  This ensures that the proper
1444                          * state will be picked up when resuming
1445                          * execution.  However, it *also* means that
1446                          * once we write PT_AR_BSP/PT_CFM, it won't be
1447                          * possible to modify the syscall arguments of
1448                          * the pending system call any longer.  This
1449                          * shouldn't be an issue because modifying
1450                          * PT_AR_BSP/PT_CFM generally implies that
1451                          * we're either abandoning the pending system
1452                          * call or that we defer it's re-execution
1453                          * (e.g., due to GDB doing an inferior
1454                          * function call).
1455                          */
1456                         urbs_end = ia64_get_user_rbs_end(target, pt, &cfm);
1457                         if (write_access) {
1458                                 if (*data != urbs_end) {
1459                                         if (in_syscall(pt))
1460                                                 convert_to_non_syscall(target,
1461                                                                        pt,
1462                                                                        cfm);
1463                                         /*
1464                                          * Simulate user-level write
1465                                          * of ar.bsp:
1466                                          */
1467                                         pt->loadrs = 0;
1468                                         pt->ar_bspstore = *data;
1469                                 }
1470                         } else
1471                                 *data = urbs_end;
1472                         return 0;
1473                 case ELF_AR_BSPSTORE_OFFSET:
1474                         ptr = &pt->ar_bspstore;
1475                         break;
1476                 case ELF_AR_RNAT_OFFSET:
1477                         ptr = &pt->ar_rnat;
1478                         break;
1479                 case ELF_AR_CCV_OFFSET:
1480                         ptr = &pt->ar_ccv;
1481                         break;
1482                 case ELF_AR_UNAT_OFFSET:
1483                         ptr = &pt->ar_unat;
1484                         break;
1485                 case ELF_AR_FPSR_OFFSET:
1486                         ptr = &pt->ar_fpsr;
1487                         break;
1488                 case ELF_AR_PFS_OFFSET:
1489                         ptr = &pt->ar_pfs;
1490                         break;
1491                 case ELF_AR_LC_OFFSET:
1492                         return unw_access_ar(info, UNW_AR_LC, data,
1493                                              write_access);
1494                 case ELF_AR_EC_OFFSET:
1495                         return unw_access_ar(info, UNW_AR_EC, data,
1496                                              write_access);
1497                 case ELF_AR_CSD_OFFSET:
1498                         ptr = &pt->ar_csd;
1499                         break;
1500                 case ELF_AR_SSD_OFFSET:
1501                         ptr = &pt->ar_ssd;
1502                 }
1503         } else if (addr >= ELF_CR_IIP_OFFSET && addr <= ELF_CR_IPSR_OFFSET) {
1504                 switch (addr) {
1505                 case ELF_CR_IIP_OFFSET:
1506                         ptr = &pt->cr_iip;
1507                         break;
1508                 case ELF_CFM_OFFSET:
1509                         urbs_end = ia64_get_user_rbs_end(target, pt, &cfm);
1510                         if (write_access) {
1511                                 if (((cfm ^ *data) & PFM_MASK) != 0) {
1512                                         if (in_syscall(pt))
1513                                                 convert_to_non_syscall(target,
1514                                                                        pt,
1515                                                                        cfm);
1516                                         pt->cr_ifs = ((pt->cr_ifs & ~PFM_MASK)
1517                                                       | (*data & PFM_MASK));
1518                                 }
1519                         } else
1520                                 *data = cfm;
1521                         return 0;
1522                 case ELF_CR_IPSR_OFFSET:
1523                         if (write_access) {
1524                                 unsigned long tmp = *data;
1525                                 /* psr.ri==3 is a reserved value: SDM 2:25 */
1526                                 if ((tmp & IA64_PSR_RI) == IA64_PSR_RI)
1527                                         tmp &= ~IA64_PSR_RI;
1528                                 pt->cr_ipsr = ((tmp & IPSR_MASK)
1529                                                | (pt->cr_ipsr & ~IPSR_MASK));
1530                         } else
1531                                 *data = (pt->cr_ipsr & IPSR_MASK);
1532                         return 0;
1533                 }
1534         } else if (addr == ELF_NAT_OFFSET)
1535                 return access_nat_bits(target, pt, info,
1536                                        data, write_access);
1537         else if (addr == ELF_PR_OFFSET)
1538                 ptr = &pt->pr;
1539         else
1540                 return -1;
1541
1542         if (write_access)
1543                 *ptr = *data;
1544         else
1545                 *data = *ptr;
1546
1547         return 0;
1548 }
1549
1550 static int
1551 access_elf_reg(struct task_struct *target, struct unw_frame_info *info,
1552                 unsigned long addr, unsigned long *data, int write_access)
1553 {
1554         if (addr >= ELF_GR_OFFSET(1) && addr <= ELF_GR_OFFSET(15))
1555                 return access_elf_gpreg(target, info, addr, data, write_access);
1556         else if (addr >= ELF_BR_OFFSET(0) && addr <= ELF_BR_OFFSET(7))
1557                 return access_elf_breg(target, info, addr, data, write_access);
1558         else
1559                 return access_elf_areg(target, info, addr, data, write_access);
1560 }
1561
1562 void do_gpregs_get(struct unw_frame_info *info, void *arg)
1563 {
1564         struct pt_regs *pt;
1565         struct regset_getset *dst = arg;
1566         elf_greg_t tmp[16];
1567         unsigned int i, index, min_copy;
1568
1569         if (unw_unwind_to_user(info) < 0)
1570                 return;
1571
1572         /*
1573          * coredump format:
1574          *      r0-r31
1575          *      NaT bits (for r0-r31; bit N == 1 iff rN is a NaT)
1576          *      predicate registers (p0-p63)
1577          *      b0-b7
1578          *      ip cfm user-mask
1579          *      ar.rsc ar.bsp ar.bspstore ar.rnat
1580          *      ar.ccv ar.unat ar.fpsr ar.pfs ar.lc ar.ec
1581          */
1582
1583
1584         /* Skip r0 */
1585         if (dst->count > 0 && dst->pos < ELF_GR_OFFSET(1)) {
1586                 dst->ret = user_regset_copyout_zero(&dst->pos, &dst->count,
1587                                                       &dst->u.get.kbuf,
1588                                                       &dst->u.get.ubuf,
1589                                                       0, ELF_GR_OFFSET(1));
1590                 if (dst->ret || dst->count == 0)
1591                         return;
1592         }
1593
1594         /* gr1 - gr15 */
1595         if (dst->count > 0 && dst->pos < ELF_GR_OFFSET(16)) {
1596                 index = (dst->pos - ELF_GR_OFFSET(1)) / sizeof(elf_greg_t);
1597                 min_copy = ELF_GR_OFFSET(16) > (dst->pos + dst->count) ?
1598                          (dst->pos + dst->count) : ELF_GR_OFFSET(16);
1599                 for (i = dst->pos; i < min_copy; i += sizeof(elf_greg_t),
1600                                 index++)
1601                         if (access_elf_reg(dst->target, info, i,
1602                                                 &tmp[index], 0) < 0) {
1603                                 dst->ret = -EIO;
1604                                 return;
1605                         }
1606                 dst->ret = user_regset_copyout(&dst->pos, &dst->count,
1607                                 &dst->u.get.kbuf, &dst->u.get.ubuf, tmp,
1608                                 ELF_GR_OFFSET(1), ELF_GR_OFFSET(16));
1609                 if (dst->ret || dst->count == 0)
1610                         return;
1611         }
1612
1613         /* r16-r31 */
1614         if (dst->count > 0 && dst->pos < ELF_NAT_OFFSET) {
1615                 pt = task_pt_regs(dst->target);
1616                 dst->ret = user_regset_copyout(&dst->pos, &dst->count,
1617                                 &dst->u.get.kbuf, &dst->u.get.ubuf, &pt->r16,
1618                                 ELF_GR_OFFSET(16), ELF_NAT_OFFSET);
1619                 if (dst->ret || dst->count == 0)
1620                         return;
1621         }
1622
1623         /* nat, pr, b0 - b7 */
1624         if (dst->count > 0 && dst->pos < ELF_CR_IIP_OFFSET) {
1625                 index = (dst->pos - ELF_NAT_OFFSET) / sizeof(elf_greg_t);
1626                 min_copy = ELF_CR_IIP_OFFSET > (dst->pos + dst->count) ?
1627                          (dst->pos + dst->count) : ELF_CR_IIP_OFFSET;
1628                 for (i = dst->pos; i < min_copy; i += sizeof(elf_greg_t),
1629                                 index++)
1630                         if (access_elf_reg(dst->target, info, i,
1631                                                 &tmp[index], 0) < 0) {
1632                                 dst->ret = -EIO;
1633                                 return;
1634                         }
1635                 dst->ret = user_regset_copyout(&dst->pos, &dst->count,
1636                                 &dst->u.get.kbuf, &dst->u.get.ubuf, tmp,
1637                                 ELF_NAT_OFFSET, ELF_CR_IIP_OFFSET);
1638                 if (dst->ret || dst->count == 0)
1639                         return;
1640         }
1641
1642         /* ip cfm psr ar.rsc ar.bsp ar.bspstore ar.rnat
1643          * ar.ccv ar.unat ar.fpsr ar.pfs ar.lc ar.ec ar.csd ar.ssd
1644          */
1645         if (dst->count > 0 && dst->pos < (ELF_AR_END_OFFSET)) {
1646                 index = (dst->pos - ELF_CR_IIP_OFFSET) / sizeof(elf_greg_t);
1647                 min_copy = ELF_AR_END_OFFSET > (dst->pos + dst->count) ?
1648                          (dst->pos + dst->count) : ELF_AR_END_OFFSET;
1649                 for (i = dst->pos; i < min_copy; i += sizeof(elf_greg_t),
1650                                 index++)
1651                         if (access_elf_reg(dst->target, info, i,
1652                                                 &tmp[index], 0) < 0) {
1653                                 dst->ret = -EIO;
1654                                 return;
1655                         }
1656                 dst->ret = user_regset_copyout(&dst->pos, &dst->count,
1657                                 &dst->u.get.kbuf, &dst->u.get.ubuf, tmp,
1658                                 ELF_CR_IIP_OFFSET, ELF_AR_END_OFFSET);
1659         }
1660 }
1661
1662 void do_gpregs_set(struct unw_frame_info *info, void *arg)
1663 {
1664         struct pt_regs *pt;
1665         struct regset_getset *dst = arg;
1666         elf_greg_t tmp[16];
1667         unsigned int i, index;
1668
1669         if (unw_unwind_to_user(info) < 0)
1670                 return;
1671
1672         /* Skip r0 */
1673         if (dst->count > 0 && dst->pos < ELF_GR_OFFSET(1)) {
1674                 dst->ret = user_regset_copyin_ignore(&dst->pos, &dst->count,
1675                                                        &dst->u.set.kbuf,
1676                                                        &dst->u.set.ubuf,
1677                                                        0, ELF_GR_OFFSET(1));
1678                 if (dst->ret || dst->count == 0)
1679                         return;
1680         }
1681
1682         /* gr1-gr15 */
1683         if (dst->count > 0 && dst->pos < ELF_GR_OFFSET(16)) {
1684                 i = dst->pos;
1685                 index = (dst->pos - ELF_GR_OFFSET(1)) / sizeof(elf_greg_t);
1686                 dst->ret = user_regset_copyin(&dst->pos, &dst->count,
1687                                 &dst->u.set.kbuf, &dst->u.set.ubuf, tmp,
1688                                 ELF_GR_OFFSET(1), ELF_GR_OFFSET(16));
1689                 if (dst->ret)
1690                         return;
1691                 for ( ; i < dst->pos; i += sizeof(elf_greg_t), index++)
1692                         if (access_elf_reg(dst->target, info, i,
1693                                                 &tmp[index], 1) < 0) {
1694                                 dst->ret = -EIO;
1695                                 return;
1696                         }
1697                 if (dst->count == 0)
1698                         return;
1699         }
1700
1701         /* gr16-gr31 */
1702         if (dst->count > 0 && dst->pos < ELF_NAT_OFFSET) {
1703                 pt = task_pt_regs(dst->target);
1704                 dst->ret = user_regset_copyin(&dst->pos, &dst->count,
1705                                 &dst->u.set.kbuf, &dst->u.set.ubuf, &pt->r16,
1706                                 ELF_GR_OFFSET(16), ELF_NAT_OFFSET);
1707                 if (dst->ret || dst->count == 0)
1708                         return;
1709         }
1710
1711         /* nat, pr, b0 - b7 */
1712         if (dst->count > 0 && dst->pos < ELF_CR_IIP_OFFSET) {
1713                 i = dst->pos;
1714                 index = (dst->pos - ELF_NAT_OFFSET) / sizeof(elf_greg_t);
1715                 dst->ret = user_regset_copyin(&dst->pos, &dst->count,
1716                                 &dst->u.set.kbuf, &dst->u.set.ubuf, tmp,
1717                                 ELF_NAT_OFFSET, ELF_CR_IIP_OFFSET);
1718                 if (dst->ret)
1719                         return;
1720                 for (; i < dst->pos; i += sizeof(elf_greg_t), index++)
1721                         if (access_elf_reg(dst->target, info, i,
1722                                                 &tmp[index], 1) < 0) {
1723                                 dst->ret = -EIO;
1724                                 return;
1725                         }
1726                 if (dst->count == 0)
1727                         return;
1728         }
1729
1730         /* ip cfm psr ar.rsc ar.bsp ar.bspstore ar.rnat
1731          * ar.ccv ar.unat ar.fpsr ar.pfs ar.lc ar.ec ar.csd ar.ssd
1732          */
1733         if (dst->count > 0 && dst->pos < (ELF_AR_END_OFFSET)) {
1734                 i = dst->pos;
1735                 index = (dst->pos - ELF_CR_IIP_OFFSET) / sizeof(elf_greg_t);
1736                 dst->ret = user_regset_copyin(&dst->pos, &dst->count,
1737                                 &dst->u.set.kbuf, &dst->u.set.ubuf, tmp,
1738                                 ELF_CR_IIP_OFFSET, ELF_AR_END_OFFSET);
1739                 if (dst->ret)
1740                         return;
1741                 for ( ; i < dst->pos; i += sizeof(elf_greg_t), index++)
1742                         if (access_elf_reg(dst->target, info, i,
1743                                                 &tmp[index], 1) < 0) {
1744                                 dst->ret = -EIO;
1745                                 return;
1746                         }
1747         }
1748 }
1749
1750 #define ELF_FP_OFFSET(i)        (i * sizeof(elf_fpreg_t))
1751
1752 void do_fpregs_get(struct unw_frame_info *info, void *arg)
1753 {
1754         struct regset_getset *dst = arg;
1755         struct task_struct *task = dst->target;
1756         elf_fpreg_t tmp[30];
1757         int index, min_copy, i;
1758
1759         if (unw_unwind_to_user(info) < 0)
1760                 return;
1761
1762         /* Skip pos 0 and 1 */
1763         if (dst->count > 0 && dst->pos < ELF_FP_OFFSET(2)) {
1764                 dst->ret = user_regset_copyout_zero(&dst->pos, &dst->count,
1765                                                       &dst->u.get.kbuf,
1766                                                       &dst->u.get.ubuf,
1767                                                       0, ELF_FP_OFFSET(2));
1768                 if (dst->count == 0 || dst->ret)
1769                         return;
1770         }
1771
1772         /* fr2-fr31 */
1773         if (dst->count > 0 && dst->pos < ELF_FP_OFFSET(32)) {
1774                 index = (dst->pos - ELF_FP_OFFSET(2)) / sizeof(elf_fpreg_t);
1775
1776                 min_copy = min(((unsigned int)ELF_FP_OFFSET(32)),
1777                                 dst->pos + dst->count);
1778                 for (i = dst->pos; i < min_copy; i += sizeof(elf_fpreg_t),
1779                                 index++)
1780                         if (unw_get_fr(info, i / sizeof(elf_fpreg_t),
1781                                          &tmp[index])) {
1782                                 dst->ret = -EIO;
1783                                 return;
1784                         }
1785                 dst->ret = user_regset_copyout(&dst->pos, &dst->count,
1786                                 &dst->u.get.kbuf, &dst->u.get.ubuf, tmp,
1787                                 ELF_FP_OFFSET(2), ELF_FP_OFFSET(32));
1788                 if (dst->count == 0 || dst->ret)
1789                         return;
1790         }
1791
1792         /* fph */
1793         if (dst->count > 0) {
1794                 ia64_flush_fph(dst->target);
1795                 if (task->thread.flags & IA64_THREAD_FPH_VALID)
1796                         dst->ret = user_regset_copyout(
1797                                 &dst->pos, &dst->count,
1798                                 &dst->u.get.kbuf, &dst->u.get.ubuf,
1799                                 &dst->target->thread.fph,
1800                                 ELF_FP_OFFSET(32), -1);
1801                 else
1802                         /* Zero fill instead.  */
1803                         dst->ret = user_regset_copyout_zero(
1804                                 &dst->pos, &dst->count,
1805                                 &dst->u.get.kbuf, &dst->u.get.ubuf,
1806                                 ELF_FP_OFFSET(32), -1);
1807         }
1808 }
1809
1810 void do_fpregs_set(struct unw_frame_info *info, void *arg)
1811 {
1812         struct regset_getset *dst = arg;
1813         elf_fpreg_t fpreg, tmp[30];
1814         int index, start, end;
1815
1816         if (unw_unwind_to_user(info) < 0)
1817                 return;
1818
1819         /* Skip pos 0 and 1 */
1820         if (dst->count > 0 && dst->pos < ELF_FP_OFFSET(2)) {
1821                 dst->ret = user_regset_copyin_ignore(&dst->pos, &dst->count,
1822                                                        &dst->u.set.kbuf,
1823                                                        &dst->u.set.ubuf,
1824                                                        0, ELF_FP_OFFSET(2));
1825                 if (dst->count == 0 || dst->ret)
1826                         return;
1827         }
1828
1829         /* fr2-fr31 */
1830         if (dst->count > 0 && dst->pos < ELF_FP_OFFSET(32)) {
1831                 start = dst->pos;
1832                 end = min(((unsigned int)ELF_FP_OFFSET(32)),
1833                          dst->pos + dst->count);
1834                 dst->ret = user_regset_copyin(&dst->pos, &dst->count,
1835                                 &dst->u.set.kbuf, &dst->u.set.ubuf, tmp,
1836                                 ELF_FP_OFFSET(2), ELF_FP_OFFSET(32));
1837                 if (dst->ret)
1838                         return;
1839
1840                 if (start & 0xF) { /* only write high part */
1841                         if (unw_get_fr(info, start / sizeof(elf_fpreg_t),
1842                                          &fpreg)) {
1843                                 dst->ret = -EIO;
1844                                 return;
1845                         }
1846                         tmp[start / sizeof(elf_fpreg_t) - 2].u.bits[0]
1847                                 = fpreg.u.bits[0];
1848                         start &= ~0xFUL;
1849                 }
1850                 if (end & 0xF) { /* only write low part */
1851                         if (unw_get_fr(info, end / sizeof(elf_fpreg_t),
1852                                         &fpreg)) {
1853                                 dst->ret = -EIO;
1854                                 return;
1855                         }
1856                         tmp[end / sizeof(elf_fpreg_t) - 2].u.bits[1]
1857                                 = fpreg.u.bits[1];
1858                         end = (end + 0xF) & ~0xFUL;
1859                 }
1860
1861                 for ( ; start < end ; start += sizeof(elf_fpreg_t)) {
1862                         index = start / sizeof(elf_fpreg_t);
1863                         if (unw_set_fr(info, index, tmp[index - 2])) {
1864                                 dst->ret = -EIO;
1865                                 return;
1866                         }
1867                 }
1868                 if (dst->ret || dst->count == 0)
1869                         return;
1870         }
1871
1872         /* fph */
1873         if (dst->count > 0 && dst->pos < ELF_FP_OFFSET(128)) {
1874                 ia64_sync_fph(dst->target);
1875                 dst->ret = user_regset_copyin(&dst->pos, &dst->count,
1876                                                 &dst->u.set.kbuf,
1877                                                 &dst->u.set.ubuf,
1878                                                 &dst->target->thread.fph,
1879                                                 ELF_FP_OFFSET(32), -1);
1880         }
1881 }
1882
1883 static int
1884 do_regset_call(void (*call)(struct unw_frame_info *, void *),
1885                struct task_struct *target,
1886                const struct user_regset *regset,
1887                unsigned int pos, unsigned int count,
1888                const void *kbuf, const void __user *ubuf)
1889 {
1890         struct regset_getset info = { .target = target, .regset = regset,
1891                                  .pos = pos, .count = count,
1892                                  .u.set = { .kbuf = kbuf, .ubuf = ubuf },
1893                                  .ret = 0 };
1894
1895         if (target == current)
1896                 unw_init_running(call, &info);
1897         else {
1898                 struct unw_frame_info ufi;
1899                 memset(&ufi, 0, sizeof(ufi));
1900                 unw_init_from_blocked_task(&ufi, target);
1901                 (*call)(&ufi, &info);
1902         }
1903
1904         return info.ret;
1905 }
1906
1907 static int
1908 gpregs_get(struct task_struct *target,
1909            const struct user_regset *regset,
1910            unsigned int pos, unsigned int count,
1911            void *kbuf, void __user *ubuf)
1912 {
1913         return do_regset_call(do_gpregs_get, target, regset, pos, count,
1914                 kbuf, ubuf);
1915 }
1916
1917 static int gpregs_set(struct task_struct *target,
1918                 const struct user_regset *regset,
1919                 unsigned int pos, unsigned int count,
1920                 const void *kbuf, const void __user *ubuf)
1921 {
1922         return do_regset_call(do_gpregs_set, target, regset, pos, count,
1923                 kbuf, ubuf);
1924 }
1925
1926 static void do_gpregs_writeback(struct unw_frame_info *info, void *arg)
1927 {
1928         do_sync_rbs(info, ia64_sync_user_rbs);
1929 }
1930
1931 /*
1932  * This is called to write back the register backing store.
1933  * ptrace does this before it stops, so that a tracer reading the user
1934  * memory after the thread stops will get the current register data.
1935  */
1936 static int
1937 gpregs_writeback(struct task_struct *target,
1938                  const struct user_regset *regset,
1939                  int now)
1940 {
1941         if (test_and_set_tsk_thread_flag(target, TIF_RESTORE_RSE))
1942                 return 0;
1943         tsk_set_notify_resume(target);
1944         return do_regset_call(do_gpregs_writeback, target, regset, 0, 0,
1945                 NULL, NULL);
1946 }
1947
1948 static int
1949 fpregs_active(struct task_struct *target, const struct user_regset *regset)
1950 {
1951         return (target->thread.flags & IA64_THREAD_FPH_VALID) ? 128 : 32;
1952 }
1953
1954 static int fpregs_get(struct task_struct *target,
1955                 const struct user_regset *regset,
1956                 unsigned int pos, unsigned int count,
1957                 void *kbuf, void __user *ubuf)
1958 {
1959         return do_regset_call(do_fpregs_get, target, regset, pos, count,
1960                 kbuf, ubuf);
1961 }
1962
1963 static int fpregs_set(struct task_struct *target,
1964                 const struct user_regset *regset,
1965                 unsigned int pos, unsigned int count,
1966                 const void *kbuf, const void __user *ubuf)
1967 {
1968         return do_regset_call(do_fpregs_set, target, regset, pos, count,
1969                 kbuf, ubuf);
1970 }
1971
1972 static int
1973 access_uarea(struct task_struct *child, unsigned long addr,
1974               unsigned long *data, int write_access)
1975 {
1976         unsigned int pos = -1; /* an invalid value */
1977         int ret;
1978         unsigned long *ptr, regnum;
1979
1980         if ((addr & 0x7) != 0) {
1981                 dprintk("ptrace: unaligned register address 0x%lx\n", addr);
1982                 return -1;
1983         }
1984         if ((addr >= PT_NAT_BITS + 8 && addr < PT_F2) ||
1985                 (addr >= PT_R7 + 8 && addr < PT_B1) ||
1986                 (addr >= PT_AR_LC + 8 && addr < PT_CR_IPSR) ||
1987                 (addr >= PT_AR_SSD + 8 && addr < PT_DBR)) {
1988                 dprintk("ptrace: rejecting access to register "
1989                                         "address 0x%lx\n", addr);
1990                 return -1;
1991         }
1992
1993         switch (addr) {
1994         case PT_F32 ... (PT_F127 + 15):
1995                 pos = addr - PT_F32 + ELF_FP_OFFSET(32);
1996                 break;
1997         case PT_F2 ... (PT_F5 + 15):
1998                 pos = addr - PT_F2 + ELF_FP_OFFSET(2);
1999                 break;
2000         case PT_F10 ... (PT_F31 + 15):
2001                 pos = addr - PT_F10 + ELF_FP_OFFSET(10);
2002                 break;
2003         case PT_F6 ... (PT_F9 + 15):
2004                 pos = addr - PT_F6 + ELF_FP_OFFSET(6);
2005                 break;
2006         }
2007
2008         if (pos != -1) {
2009                 if (write_access)
2010                         ret = fpregs_set(child, NULL, pos,
2011                                 sizeof(unsigned long), data, NULL);
2012                 else
2013                         ret = fpregs_get(child, NULL, pos,
2014                                 sizeof(unsigned long), data, NULL);
2015                 if (ret != 0)
2016                         return -1;
2017                 return 0;
2018         }
2019
2020         switch (addr) {
2021         case PT_NAT_BITS:
2022                 pos = ELF_NAT_OFFSET;
2023                 break;
2024         case PT_R4 ... PT_R7:
2025                 pos = addr - PT_R4 + ELF_GR_OFFSET(4);
2026                 break;
2027         case PT_B1 ... PT_B5:
2028                 pos = addr - PT_B1 + ELF_BR_OFFSET(1);
2029                 break;
2030         case PT_AR_EC:
2031                 pos = ELF_AR_EC_OFFSET;
2032                 break;
2033         case PT_AR_LC:
2034                 pos = ELF_AR_LC_OFFSET;
2035                 break;
2036         case PT_CR_IPSR:
2037                 pos = ELF_CR_IPSR_OFFSET;
2038                 break;
2039         case PT_CR_IIP:
2040                 pos = ELF_CR_IIP_OFFSET;
2041                 break;
2042         case PT_CFM:
2043                 pos = ELF_CFM_OFFSET;
2044                 break;
2045         case PT_AR_UNAT:
2046                 pos = ELF_AR_UNAT_OFFSET;
2047                 break;
2048         case PT_AR_PFS:
2049                 pos = ELF_AR_PFS_OFFSET;
2050                 break;
2051         case PT_AR_RSC:
2052                 pos = ELF_AR_RSC_OFFSET;
2053                 break;
2054         case PT_AR_RNAT:
2055                 pos = ELF_AR_RNAT_OFFSET;
2056                 break;
2057         case PT_AR_BSPSTORE:
2058                 pos = ELF_AR_BSPSTORE_OFFSET;
2059                 break;
2060         case PT_PR:
2061                 pos = ELF_PR_OFFSET;
2062                 break;
2063         case PT_B6:
2064                 pos = ELF_BR_OFFSET(6);
2065                 break;
2066         case PT_AR_BSP:
2067                 pos = ELF_AR_BSP_OFFSET;
2068                 break;
2069         case PT_R1 ... PT_R3:
2070                 pos = addr - PT_R1 + ELF_GR_OFFSET(1);
2071                 break;
2072         case PT_R12 ... PT_R15:
2073                 pos = addr - PT_R12 + ELF_GR_OFFSET(12);
2074                 break;
2075         case PT_R8 ... PT_R11:
2076                 pos = addr - PT_R8 + ELF_GR_OFFSET(8);
2077                 break;
2078         case PT_R16 ... PT_R31:
2079                 pos = addr - PT_R16 + ELF_GR_OFFSET(16);
2080                 break;
2081         case PT_AR_CCV:
2082                 pos = ELF_AR_CCV_OFFSET;
2083                 break;
2084         case PT_AR_FPSR:
2085                 pos = ELF_AR_FPSR_OFFSET;
2086                 break;
2087         case PT_B0:
2088                 pos = ELF_BR_OFFSET(0);
2089                 break;
2090         case PT_B7:
2091                 pos = ELF_BR_OFFSET(7);
2092                 break;
2093         case PT_AR_CSD:
2094                 pos = ELF_AR_CSD_OFFSET;
2095                 break;
2096         case PT_AR_SSD:
2097                 pos = ELF_AR_SSD_OFFSET;
2098                 break;
2099         }
2100
2101         if (pos != -1) {
2102                 if (write_access)
2103                         ret = gpregs_set(child, NULL, pos,
2104                                 sizeof(unsigned long), data, NULL);
2105                 else
2106                         ret = gpregs_get(child, NULL, pos,
2107                                 sizeof(unsigned long), data, NULL);
2108                 if (ret != 0)
2109                         return -1;
2110                 return 0;
2111         }
2112
2113         /* access debug registers */
2114         if (addr >= PT_IBR) {
2115                 regnum = (addr - PT_IBR) >> 3;
2116                 ptr = &child->thread.ibr[0];
2117         } else {
2118                 regnum = (addr - PT_DBR) >> 3;
2119                 ptr = &child->thread.dbr[0];
2120         }
2121
2122         if (regnum >= 8) {
2123                 dprintk("ptrace: rejecting access to register "
2124                                 "address 0x%lx\n", addr);
2125                 return -1;
2126         }
2127 #ifdef CONFIG_PERFMON
2128         /*
2129          * Check if debug registers are used by perfmon. This
2130          * test must be done once we know that we can do the
2131          * operation, i.e. the arguments are all valid, but
2132          * before we start modifying the state.
2133          *
2134          * Perfmon needs to keep a count of how many processes
2135          * are trying to modify the debug registers for system
2136          * wide monitoring sessions.
2137          *
2138          * We also include read access here, because they may
2139          * cause the PMU-installed debug register state
2140          * (dbr[], ibr[]) to be reset. The two arrays are also
2141          * used by perfmon, but we do not use
2142          * IA64_THREAD_DBG_VALID. The registers are restored
2143          * by the PMU context switch code.
2144          */
2145         if (pfm_use_debug_registers(child))
2146                 return -1;
2147 #endif
2148
2149         if (!(child->thread.flags & IA64_THREAD_DBG_VALID)) {
2150                 child->thread.flags |= IA64_THREAD_DBG_VALID;
2151                 memset(child->thread.dbr, 0,
2152                                 sizeof(child->thread.dbr));
2153                 memset(child->thread.ibr, 0,
2154                                 sizeof(child->thread.ibr));
2155         }
2156
2157         ptr += regnum;
2158
2159         if ((regnum & 1) && write_access) {
2160                 /* don't let the user set kernel-level breakpoints: */
2161                 *ptr = *data & ~(7UL << 56);
2162                 return 0;
2163         }
2164         if (write_access)
2165                 *ptr = *data;
2166         else
2167                 *data = *ptr;
2168         return 0;
2169 }
2170
2171 static const struct user_regset native_regsets[] = {
2172         {
2173                 .core_note_type = NT_PRSTATUS,
2174                 .n = ELF_NGREG,
2175                 .size = sizeof(elf_greg_t), .align = sizeof(elf_greg_t),
2176                 .get = gpregs_get, .set = gpregs_set,
2177                 .writeback = gpregs_writeback
2178         },
2179         {
2180                 .core_note_type = NT_PRFPREG,
2181                 .n = ELF_NFPREG,
2182                 .size = sizeof(elf_fpreg_t), .align = sizeof(elf_fpreg_t),
2183                 .get = fpregs_get, .set = fpregs_set, .active = fpregs_active
2184         },
2185 };
2186
2187 static const struct user_regset_view user_ia64_view = {
2188         .name = "ia64",
2189         .e_machine = EM_IA_64,
2190         .regsets = native_regsets, .n = ARRAY_SIZE(native_regsets)
2191 };
2192
2193 const struct user_regset_view *task_user_regset_view(struct task_struct *tsk)
2194 {
2195 #ifdef CONFIG_IA32_SUPPORT
2196         extern const struct user_regset_view user_ia32_view;
2197         if (IS_IA32_PROCESS(task_pt_regs(tsk)))
2198                 return &user_ia32_view;
2199 #endif
2200         return &user_ia64_view;
2201 }