Pull ec into release branch
[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  *
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 #include <linux/signal.h>
21
22 #include <asm/pgtable.h>
23 #include <asm/processor.h>
24 #include <asm/ptrace_offsets.h>
25 #include <asm/rse.h>
26 #include <asm/system.h>
27 #include <asm/uaccess.h>
28 #include <asm/unwind.h>
29 #ifdef CONFIG_PERFMON
30 #include <asm/perfmon.h>
31 #endif
32
33 #include "entry.h"
34
35 /*
36  * Bits in the PSR that we allow ptrace() to change:
37  *      be, up, ac, mfl, mfh (the user mask; five bits total)
38  *      db (debug breakpoint fault; one bit)
39  *      id (instruction debug fault disable; one bit)
40  *      dd (data debug fault disable; one bit)
41  *      ri (restart instruction; two bits)
42  *      is (instruction set; one bit)
43  */
44 #define IPSR_MASK (IA64_PSR_UM | IA64_PSR_DB | IA64_PSR_IS      \
45                    | IA64_PSR_ID | IA64_PSR_DD | IA64_PSR_RI)
46
47 #define MASK(nbits)     ((1UL << (nbits)) - 1)  /* mask with NBITS bits set */
48 #define PFM_MASK        MASK(38)
49
50 #define PTRACE_DEBUG    0
51
52 #if PTRACE_DEBUG
53 # define dprintk(format...)     printk(format)
54 # define inline
55 #else
56 # define dprintk(format...)
57 #endif
58
59 /* Return TRUE if PT was created due to kernel-entry via a system-call.  */
60
61 static inline int
62 in_syscall (struct pt_regs *pt)
63 {
64         return (long) pt->cr_ifs >= 0;
65 }
66
67 /*
68  * Collect the NaT bits for r1-r31 from scratch_unat and return a NaT
69  * bitset where bit i is set iff the NaT bit of register i is set.
70  */
71 unsigned long
72 ia64_get_scratch_nat_bits (struct pt_regs *pt, unsigned long scratch_unat)
73 {
74 #       define GET_BITS(first, last, unat)                              \
75         ({                                                              \
76                 unsigned long bit = ia64_unat_pos(&pt->r##first);       \
77                 unsigned long nbits = (last - first + 1);               \
78                 unsigned long mask = MASK(nbits) << first;              \
79                 unsigned long dist;                                     \
80                 if (bit < first)                                        \
81                         dist = 64 + bit - first;                        \
82                 else                                                    \
83                         dist = bit - first;                             \
84                 ia64_rotr(unat, dist) & mask;                           \
85         })
86         unsigned long val;
87
88         /*
89          * Registers that are stored consecutively in struct pt_regs
90          * can be handled in parallel.  If the register order in
91          * struct_pt_regs changes, this code MUST be updated.
92          */
93         val  = GET_BITS( 1,  1, scratch_unat);
94         val |= GET_BITS( 2,  3, scratch_unat);
95         val |= GET_BITS(12, 13, scratch_unat);
96         val |= GET_BITS(14, 14, scratch_unat);
97         val |= GET_BITS(15, 15, scratch_unat);
98         val |= GET_BITS( 8, 11, scratch_unat);
99         val |= GET_BITS(16, 31, scratch_unat);
100         return val;
101
102 #       undef GET_BITS
103 }
104
105 /*
106  * Set the NaT bits for the scratch registers according to NAT and
107  * return the resulting unat (assuming the scratch registers are
108  * stored in PT).
109  */
110 unsigned long
111 ia64_put_scratch_nat_bits (struct pt_regs *pt, unsigned long nat)
112 {
113 #       define PUT_BITS(first, last, nat)                               \
114         ({                                                              \
115                 unsigned long bit = ia64_unat_pos(&pt->r##first);       \
116                 unsigned long nbits = (last - first + 1);               \
117                 unsigned long mask = MASK(nbits) << first;              \
118                 long dist;                                              \
119                 if (bit < first)                                        \
120                         dist = 64 + bit - first;                        \
121                 else                                                    \
122                         dist = bit - first;                             \
123                 ia64_rotl(nat & mask, dist);                            \
124         })
125         unsigned long scratch_unat;
126
127         /*
128          * Registers that are stored consecutively in struct pt_regs
129          * can be handled in parallel.  If the register order in
130          * struct_pt_regs changes, this code MUST be updated.
131          */
132         scratch_unat  = PUT_BITS( 1,  1, nat);
133         scratch_unat |= PUT_BITS( 2,  3, nat);
134         scratch_unat |= PUT_BITS(12, 13, nat);
135         scratch_unat |= PUT_BITS(14, 14, nat);
136         scratch_unat |= PUT_BITS(15, 15, nat);
137         scratch_unat |= PUT_BITS( 8, 11, nat);
138         scratch_unat |= PUT_BITS(16, 31, nat);
139
140         return scratch_unat;
141
142 #       undef PUT_BITS
143 }
144
145 #define IA64_MLX_TEMPLATE       0x2
146 #define IA64_MOVL_OPCODE        6
147
148 void
149 ia64_increment_ip (struct pt_regs *regs)
150 {
151         unsigned long w0, ri = ia64_psr(regs)->ri + 1;
152
153         if (ri > 2) {
154                 ri = 0;
155                 regs->cr_iip += 16;
156         } else if (ri == 2) {
157                 get_user(w0, (char __user *) regs->cr_iip + 0);
158                 if (((w0 >> 1) & 0xf) == IA64_MLX_TEMPLATE) {
159                         /*
160                          * rfi'ing to slot 2 of an MLX bundle causes
161                          * an illegal operation fault.  We don't want
162                          * that to happen...
163                          */
164                         ri = 0;
165                         regs->cr_iip += 16;
166                 }
167         }
168         ia64_psr(regs)->ri = ri;
169 }
170
171 void
172 ia64_decrement_ip (struct pt_regs *regs)
173 {
174         unsigned long w0, ri = ia64_psr(regs)->ri - 1;
175
176         if (ia64_psr(regs)->ri == 0) {
177                 regs->cr_iip -= 16;
178                 ri = 2;
179                 get_user(w0, (char __user *) regs->cr_iip + 0);
180                 if (((w0 >> 1) & 0xf) == IA64_MLX_TEMPLATE) {
181                         /*
182                          * rfi'ing to slot 2 of an MLX bundle causes
183                          * an illegal operation fault.  We don't want
184                          * that to happen...
185                          */
186                         ri = 1;
187                 }
188         }
189         ia64_psr(regs)->ri = ri;
190 }
191
192 /*
193  * This routine is used to read an rnat bits that are stored on the
194  * kernel backing store.  Since, in general, the alignment of the user
195  * and kernel are different, this is not completely trivial.  In
196  * essence, we need to construct the user RNAT based on up to two
197  * kernel RNAT values and/or the RNAT value saved in the child's
198  * pt_regs.
199  *
200  * user rbs
201  *
202  * +--------+ <-- lowest address
203  * | slot62 |
204  * +--------+
205  * |  rnat  | 0x....1f8
206  * +--------+
207  * | slot00 | \
208  * +--------+ |
209  * | slot01 | > child_regs->ar_rnat
210  * +--------+ |
211  * | slot02 | /                         kernel rbs
212  * +--------+                           +--------+
213  *          <- child_regs->ar_bspstore  | slot61 | <-- krbs
214  * +- - - - +                           +--------+
215  *                                      | slot62 |
216  * +- - - - +                           +--------+
217  *                                      |  rnat  |
218  * +- - - - +                           +--------+
219  *   vrnat                              | slot00 |
220  * +- - - - +                           +--------+
221  *                                      =        =
222  *                                      +--------+
223  *                                      | slot00 | \
224  *                                      +--------+ |
225  *                                      | slot01 | > child_stack->ar_rnat
226  *                                      +--------+ |
227  *                                      | slot02 | /
228  *                                      +--------+
229  *                                                <--- child_stack->ar_bspstore
230  *
231  * The way to think of this code is as follows: bit 0 in the user rnat
232  * corresponds to some bit N (0 <= N <= 62) in one of the kernel rnat
233  * value.  The kernel rnat value holding this bit is stored in
234  * variable rnat0.  rnat1 is loaded with the kernel rnat value that
235  * form the upper bits of the user rnat value.
236  *
237  * Boundary cases:
238  *
239  * o when reading the rnat "below" the first rnat slot on the kernel
240  *   backing store, rnat0/rnat1 are set to 0 and the low order bits are
241  *   merged in from pt->ar_rnat.
242  *
243  * o when reading the rnat "above" the last rnat slot on the kernel
244  *   backing store, rnat0/rnat1 gets its value from sw->ar_rnat.
245  */
246 static unsigned long
247 get_rnat (struct task_struct *task, struct switch_stack *sw,
248           unsigned long *krbs, unsigned long *urnat_addr,
249           unsigned long *urbs_end)
250 {
251         unsigned long rnat0 = 0, rnat1 = 0, urnat = 0, *slot0_kaddr;
252         unsigned long umask = 0, mask, m;
253         unsigned long *kbsp, *ubspstore, *rnat0_kaddr, *rnat1_kaddr, shift;
254         long num_regs, nbits;
255         struct pt_regs *pt;
256
257         pt = task_pt_regs(task);
258         kbsp = (unsigned long *) sw->ar_bspstore;
259         ubspstore = (unsigned long *) pt->ar_bspstore;
260
261         if (urbs_end < urnat_addr)
262                 nbits = ia64_rse_num_regs(urnat_addr - 63, urbs_end);
263         else
264                 nbits = 63;
265         mask = MASK(nbits);
266         /*
267          * First, figure out which bit number slot 0 in user-land maps
268          * to in the kernel rnat.  Do this by figuring out how many
269          * register slots we're beyond the user's backingstore and
270          * then computing the equivalent address in kernel space.
271          */
272         num_regs = ia64_rse_num_regs(ubspstore, urnat_addr + 1);
273         slot0_kaddr = ia64_rse_skip_regs(krbs, num_regs);
274         shift = ia64_rse_slot_num(slot0_kaddr);
275         rnat1_kaddr = ia64_rse_rnat_addr(slot0_kaddr);
276         rnat0_kaddr = rnat1_kaddr - 64;
277
278         if (ubspstore + 63 > urnat_addr) {
279                 /* some bits need to be merged in from pt->ar_rnat */
280                 umask = MASK(ia64_rse_slot_num(ubspstore)) & mask;
281                 urnat = (pt->ar_rnat & umask);
282                 mask &= ~umask;
283                 if (!mask)
284                         return urnat;
285         }
286
287         m = mask << shift;
288         if (rnat0_kaddr >= kbsp)
289                 rnat0 = sw->ar_rnat;
290         else if (rnat0_kaddr > krbs)
291                 rnat0 = *rnat0_kaddr;
292         urnat |= (rnat0 & m) >> shift;
293
294         m = mask >> (63 - shift);
295         if (rnat1_kaddr >= kbsp)
296                 rnat1 = sw->ar_rnat;
297         else if (rnat1_kaddr > krbs)
298                 rnat1 = *rnat1_kaddr;
299         urnat |= (rnat1 & m) << (63 - shift);
300         return urnat;
301 }
302
303 /*
304  * The reverse of get_rnat.
305  */
306 static void
307 put_rnat (struct task_struct *task, struct switch_stack *sw,
308           unsigned long *krbs, unsigned long *urnat_addr, unsigned long urnat,
309           unsigned long *urbs_end)
310 {
311         unsigned long rnat0 = 0, rnat1 = 0, *slot0_kaddr, umask = 0, mask, m;
312         unsigned long *kbsp, *ubspstore, *rnat0_kaddr, *rnat1_kaddr, shift;
313         long num_regs, nbits;
314         struct pt_regs *pt;
315         unsigned long cfm, *urbs_kargs;
316
317         pt = task_pt_regs(task);
318         kbsp = (unsigned long *) sw->ar_bspstore;
319         ubspstore = (unsigned long *) pt->ar_bspstore;
320
321         urbs_kargs = urbs_end;
322         if (in_syscall(pt)) {
323                 /*
324                  * If entered via syscall, don't allow user to set rnat bits
325                  * for syscall args.
326                  */
327                 cfm = pt->cr_ifs;
328                 urbs_kargs = ia64_rse_skip_regs(urbs_end, -(cfm & 0x7f));
329         }
330
331         if (urbs_kargs >= urnat_addr)
332                 nbits = 63;
333         else {
334                 if ((urnat_addr - 63) >= urbs_kargs)
335                         return;
336                 nbits = ia64_rse_num_regs(urnat_addr - 63, urbs_kargs);
337         }
338         mask = MASK(nbits);
339
340         /*
341          * First, figure out which bit number slot 0 in user-land maps
342          * to in the kernel rnat.  Do this by figuring out how many
343          * register slots we're beyond the user's backingstore and
344          * then computing the equivalent address in kernel space.
345          */
346         num_regs = ia64_rse_num_regs(ubspstore, urnat_addr + 1);
347         slot0_kaddr = ia64_rse_skip_regs(krbs, num_regs);
348         shift = ia64_rse_slot_num(slot0_kaddr);
349         rnat1_kaddr = ia64_rse_rnat_addr(slot0_kaddr);
350         rnat0_kaddr = rnat1_kaddr - 64;
351
352         if (ubspstore + 63 > urnat_addr) {
353                 /* some bits need to be place in pt->ar_rnat: */
354                 umask = MASK(ia64_rse_slot_num(ubspstore)) & mask;
355                 pt->ar_rnat = (pt->ar_rnat & ~umask) | (urnat & umask);
356                 mask &= ~umask;
357                 if (!mask)
358                         return;
359         }
360         /*
361          * Note: Section 11.1 of the EAS guarantees that bit 63 of an
362          * rnat slot is ignored. so we don't have to clear it here.
363          */
364         rnat0 = (urnat << shift);
365         m = mask << shift;
366         if (rnat0_kaddr >= kbsp)
367                 sw->ar_rnat = (sw->ar_rnat & ~m) | (rnat0 & m);
368         else if (rnat0_kaddr > krbs)
369                 *rnat0_kaddr = ((*rnat0_kaddr & ~m) | (rnat0 & m));
370
371         rnat1 = (urnat >> (63 - shift));
372         m = mask >> (63 - shift);
373         if (rnat1_kaddr >= kbsp)
374                 sw->ar_rnat = (sw->ar_rnat & ~m) | (rnat1 & m);
375         else if (rnat1_kaddr > krbs)
376                 *rnat1_kaddr = ((*rnat1_kaddr & ~m) | (rnat1 & m));
377 }
378
379 static inline int
380 on_kernel_rbs (unsigned long addr, unsigned long bspstore,
381                unsigned long urbs_end)
382 {
383         unsigned long *rnat_addr = ia64_rse_rnat_addr((unsigned long *)
384                                                       urbs_end);
385         return (addr >= bspstore && addr <= (unsigned long) rnat_addr);
386 }
387
388 /*
389  * Read a word from the user-level backing store of task CHILD.  ADDR
390  * is the user-level address to read the word from, VAL a pointer to
391  * the return value, and USER_BSP gives the end of the user-level
392  * backing store (i.e., it's the address that would be in ar.bsp after
393  * the user executed a "cover" instruction).
394  *
395  * This routine takes care of accessing the kernel register backing
396  * store for those registers that got spilled there.  It also takes
397  * care of calculating the appropriate RNaT collection words.
398  */
399 long
400 ia64_peek (struct task_struct *child, struct switch_stack *child_stack,
401            unsigned long user_rbs_end, unsigned long addr, long *val)
402 {
403         unsigned long *bspstore, *krbs, regnum, *laddr, *urbs_end, *rnat_addr;
404         struct pt_regs *child_regs;
405         size_t copied;
406         long ret;
407
408         urbs_end = (long *) user_rbs_end;
409         laddr = (unsigned long *) addr;
410         child_regs = task_pt_regs(child);
411         bspstore = (unsigned long *) child_regs->ar_bspstore;
412         krbs = (unsigned long *) child + IA64_RBS_OFFSET/8;
413         if (on_kernel_rbs(addr, (unsigned long) bspstore,
414                           (unsigned long) urbs_end))
415         {
416                 /*
417                  * Attempt to read the RBS in an area that's actually
418                  * on the kernel RBS => read the corresponding bits in
419                  * the kernel RBS.
420                  */
421                 rnat_addr = ia64_rse_rnat_addr(laddr);
422                 ret = get_rnat(child, child_stack, krbs, rnat_addr, urbs_end);
423
424                 if (laddr == rnat_addr) {
425                         /* return NaT collection word itself */
426                         *val = ret;
427                         return 0;
428                 }
429
430                 if (((1UL << ia64_rse_slot_num(laddr)) & ret) != 0) {
431                         /*
432                          * It is implementation dependent whether the
433                          * data portion of a NaT value gets saved on a
434                          * st8.spill or RSE spill (e.g., see EAS 2.6,
435                          * 4.4.4.6 Register Spill and Fill).  To get
436                          * consistent behavior across all possible
437                          * IA-64 implementations, we return zero in
438                          * this case.
439                          */
440                         *val = 0;
441                         return 0;
442                 }
443
444                 if (laddr < urbs_end) {
445                         /*
446                          * The desired word is on the kernel RBS and
447                          * is not a NaT.
448                          */
449                         regnum = ia64_rse_num_regs(bspstore, laddr);
450                         *val = *ia64_rse_skip_regs(krbs, regnum);
451                         return 0;
452                 }
453         }
454         copied = access_process_vm(child, addr, &ret, sizeof(ret), 0);
455         if (copied != sizeof(ret))
456                 return -EIO;
457         *val = ret;
458         return 0;
459 }
460
461 long
462 ia64_poke (struct task_struct *child, struct switch_stack *child_stack,
463            unsigned long user_rbs_end, unsigned long addr, long val)
464 {
465         unsigned long *bspstore, *krbs, regnum, *laddr;
466         unsigned long *urbs_end = (long *) user_rbs_end;
467         struct pt_regs *child_regs;
468
469         laddr = (unsigned long *) addr;
470         child_regs = task_pt_regs(child);
471         bspstore = (unsigned long *) child_regs->ar_bspstore;
472         krbs = (unsigned long *) child + IA64_RBS_OFFSET/8;
473         if (on_kernel_rbs(addr, (unsigned long) bspstore,
474                           (unsigned long) urbs_end))
475         {
476                 /*
477                  * Attempt to write the RBS in an area that's actually
478                  * on the kernel RBS => write the corresponding bits
479                  * in the kernel RBS.
480                  */
481                 if (ia64_rse_is_rnat_slot(laddr))
482                         put_rnat(child, child_stack, krbs, laddr, val,
483                                  urbs_end);
484                 else {
485                         if (laddr < urbs_end) {
486                                 regnum = ia64_rse_num_regs(bspstore, laddr);
487                                 *ia64_rse_skip_regs(krbs, regnum) = val;
488                         }
489                 }
490         } else if (access_process_vm(child, addr, &val, sizeof(val), 1)
491                    != sizeof(val))
492                 return -EIO;
493         return 0;
494 }
495
496 /*
497  * Calculate the address of the end of the user-level register backing
498  * store.  This is the address that would have been stored in ar.bsp
499  * if the user had executed a "cover" instruction right before
500  * entering the kernel.  If CFMP is not NULL, it is used to return the
501  * "current frame mask" that was active at the time the kernel was
502  * entered.
503  */
504 unsigned long
505 ia64_get_user_rbs_end (struct task_struct *child, struct pt_regs *pt,
506                        unsigned long *cfmp)
507 {
508         unsigned long *krbs, *bspstore, cfm = pt->cr_ifs;
509         long ndirty;
510
511         krbs = (unsigned long *) child + IA64_RBS_OFFSET/8;
512         bspstore = (unsigned long *) pt->ar_bspstore;
513         ndirty = ia64_rse_num_regs(krbs, krbs + (pt->loadrs >> 19));
514
515         if (in_syscall(pt))
516                 ndirty += (cfm & 0x7f);
517         else
518                 cfm &= ~(1UL << 63);    /* clear valid bit */
519
520         if (cfmp)
521                 *cfmp = cfm;
522         return (unsigned long) ia64_rse_skip_regs(bspstore, ndirty);
523 }
524
525 /*
526  * Synchronize (i.e, write) the RSE backing store living in kernel
527  * space to the VM of the CHILD task.  SW and PT are the pointers to
528  * the switch_stack and pt_regs structures, respectively.
529  * USER_RBS_END is the user-level address at which the backing store
530  * ends.
531  */
532 long
533 ia64_sync_user_rbs (struct task_struct *child, struct switch_stack *sw,
534                     unsigned long user_rbs_start, unsigned long user_rbs_end)
535 {
536         unsigned long addr, val;
537         long ret;
538
539         /* now copy word for word from kernel rbs to user rbs: */
540         for (addr = user_rbs_start; addr < user_rbs_end; addr += 8) {
541                 ret = ia64_peek(child, sw, user_rbs_end, addr, &val);
542                 if (ret < 0)
543                         return ret;
544                 if (access_process_vm(child, addr, &val, sizeof(val), 1)
545                     != sizeof(val))
546                         return -EIO;
547         }
548         return 0;
549 }
550
551 static inline int
552 thread_matches (struct task_struct *thread, unsigned long addr)
553 {
554         unsigned long thread_rbs_end;
555         struct pt_regs *thread_regs;
556
557         if (ptrace_check_attach(thread, 0) < 0)
558                 /*
559                  * If the thread is not in an attachable state, we'll
560                  * ignore it.  The net effect is that if ADDR happens
561                  * to overlap with the portion of the thread's
562                  * register backing store that is currently residing
563                  * on the thread's kernel stack, then ptrace() may end
564                  * up accessing a stale value.  But if the thread
565                  * isn't stopped, that's a problem anyhow, so we're
566                  * doing as well as we can...
567                  */
568                 return 0;
569
570         thread_regs = task_pt_regs(thread);
571         thread_rbs_end = ia64_get_user_rbs_end(thread, thread_regs, NULL);
572         if (!on_kernel_rbs(addr, thread_regs->ar_bspstore, thread_rbs_end))
573                 return 0;
574
575         return 1;       /* looks like we've got a winner */
576 }
577
578 /*
579  * GDB apparently wants to be able to read the register-backing store
580  * of any thread when attached to a given process.  If we are peeking
581  * or poking an address that happens to reside in the kernel-backing
582  * store of another thread, we need to attach to that thread, because
583  * otherwise we end up accessing stale data.
584  *
585  * task_list_lock must be read-locked before calling this routine!
586  */
587 static struct task_struct *
588 find_thread_for_addr (struct task_struct *child, unsigned long addr)
589 {
590         struct task_struct *p;
591         struct mm_struct *mm;
592         struct list_head *this, *next;
593         int mm_users;
594
595         if (!(mm = get_task_mm(child)))
596                 return child;
597
598         /* -1 because of our get_task_mm(): */
599         mm_users = atomic_read(&mm->mm_users) - 1;
600         if (mm_users <= 1)
601                 goto out;               /* not multi-threaded */
602
603         /*
604          * Traverse the current process' children list.  Every task that
605          * one attaches to becomes a child.  And it is only attached children
606          * of the debugger that are of interest (ptrace_check_attach checks
607          * for this).
608          */
609         list_for_each_safe(this, next, &current->children) {
610                 p = list_entry(this, struct task_struct, sibling);
611                 if (p->mm != mm)
612                         continue;
613                 if (thread_matches(p, addr)) {
614                         child = p;
615                         goto out;
616                 }
617         }
618
619   out:
620         mmput(mm);
621         return child;
622 }
623
624 /*
625  * Write f32-f127 back to task->thread.fph if it has been modified.
626  */
627 inline void
628 ia64_flush_fph (struct task_struct *task)
629 {
630         struct ia64_psr *psr = ia64_psr(task_pt_regs(task));
631
632         /*
633          * Prevent migrating this task while
634          * we're fiddling with the FPU state
635          */
636         preempt_disable();
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         preempt_enable();
643 }
644
645 /*
646  * Sync the fph state of the task so that it can be manipulated
647  * through thread.fph.  If necessary, f32-f127 are written back to
648  * thread.fph or, if the fph state hasn't been used before, thread.fph
649  * is cleared to zeroes.  Also, access to f32-f127 is disabled to
650  * ensure that the task picks up the state from thread.fph when it
651  * executes again.
652  */
653 void
654 ia64_sync_fph (struct task_struct *task)
655 {
656         struct ia64_psr *psr = ia64_psr(task_pt_regs(task));
657
658         ia64_flush_fph(task);
659         if (!(task->thread.flags & IA64_THREAD_FPH_VALID)) {
660                 task->thread.flags |= IA64_THREAD_FPH_VALID;
661                 memset(&task->thread.fph, 0, sizeof(task->thread.fph));
662         }
663         ia64_drop_fpu(task);
664         psr->dfh = 1;
665 }
666
667 static int
668 access_fr (struct unw_frame_info *info, int regnum, int hi,
669            unsigned long *data, int write_access)
670 {
671         struct ia64_fpreg fpval;
672         int ret;
673
674         ret = unw_get_fr(info, regnum, &fpval);
675         if (ret < 0)
676                 return ret;
677
678         if (write_access) {
679                 fpval.u.bits[hi] = *data;
680                 ret = unw_set_fr(info, regnum, fpval);
681         } else
682                 *data = fpval.u.bits[hi];
683         return ret;
684 }
685
686 /*
687  * Change the machine-state of CHILD such that it will return via the normal
688  * kernel exit-path, rather than the syscall-exit path.
689  */
690 static void
691 convert_to_non_syscall (struct task_struct *child, struct pt_regs  *pt,
692                         unsigned long cfm)
693 {
694         struct unw_frame_info info, prev_info;
695         unsigned long ip, sp, pr;
696
697         unw_init_from_blocked_task(&info, child);
698         while (1) {
699                 prev_info = info;
700                 if (unw_unwind(&info) < 0)
701                         return;
702
703                 unw_get_sp(&info, &sp);
704                 if ((long)((unsigned long)child + IA64_STK_OFFSET - sp)
705                     < IA64_PT_REGS_SIZE) {
706                         dprintk("ptrace.%s: ran off the top of the kernel "
707                                 "stack\n", __FUNCTION__);
708                         return;
709                 }
710                 if (unw_get_pr (&prev_info, &pr) < 0) {
711                         unw_get_rp(&prev_info, &ip);
712                         dprintk("ptrace.%s: failed to read "
713                                 "predicate register (ip=0x%lx)\n",
714                                 __FUNCTION__, ip);
715                         return;
716                 }
717                 if (unw_is_intr_frame(&info)
718                     && (pr & (1UL << PRED_USER_STACK)))
719                         break;
720         }
721
722         /*
723          * Note: at the time of this call, the target task is blocked
724          * in notify_resume_user() and by clearling PRED_LEAVE_SYSCALL
725          * (aka, "pLvSys") we redirect execution from
726          * .work_pending_syscall_end to .work_processed_kernel.
727          */
728         unw_get_pr(&prev_info, &pr);
729         pr &= ~((1UL << PRED_SYSCALL) | (1UL << PRED_LEAVE_SYSCALL));
730         pr |=  (1UL << PRED_NON_SYSCALL);
731         unw_set_pr(&prev_info, pr);
732
733         pt->cr_ifs = (1UL << 63) | cfm;
734         /*
735          * Clear the memory that is NOT written on syscall-entry to
736          * ensure we do not leak kernel-state to user when execution
737          * resumes.
738          */
739         pt->r2 = 0;
740         pt->r3 = 0;
741         pt->r14 = 0;
742         memset(&pt->r16, 0, 16*8);      /* clear r16-r31 */
743         memset(&pt->f6, 0, 6*16);       /* clear f6-f11 */
744         pt->b7 = 0;
745         pt->ar_ccv = 0;
746         pt->ar_csd = 0;
747         pt->ar_ssd = 0;
748 }
749
750 static int
751 access_nat_bits (struct task_struct *child, struct pt_regs *pt,
752                  struct unw_frame_info *info,
753                  unsigned long *data, int write_access)
754 {
755         unsigned long regnum, nat_bits, scratch_unat, dummy = 0;
756         char nat = 0;
757
758         if (write_access) {
759                 nat_bits = *data;
760                 scratch_unat = ia64_put_scratch_nat_bits(pt, nat_bits);
761                 if (unw_set_ar(info, UNW_AR_UNAT, scratch_unat) < 0) {
762                         dprintk("ptrace: failed to set ar.unat\n");
763                         return -1;
764                 }
765                 for (regnum = 4; regnum <= 7; ++regnum) {
766                         unw_get_gr(info, regnum, &dummy, &nat);
767                         unw_set_gr(info, regnum, dummy,
768                                    (nat_bits >> regnum) & 1);
769                 }
770         } else {
771                 if (unw_get_ar(info, UNW_AR_UNAT, &scratch_unat) < 0) {
772                         dprintk("ptrace: failed to read ar.unat\n");
773                         return -1;
774                 }
775                 nat_bits = ia64_get_scratch_nat_bits(pt, scratch_unat);
776                 for (regnum = 4; regnum <= 7; ++regnum) {
777                         unw_get_gr(info, regnum, &dummy, &nat);
778                         nat_bits |= (nat != 0) << regnum;
779                 }
780                 *data = nat_bits;
781         }
782         return 0;
783 }
784
785 static int
786 access_uarea (struct task_struct *child, unsigned long addr,
787               unsigned long *data, int write_access)
788 {
789         unsigned long *ptr, regnum, urbs_end, rnat_addr, cfm;
790         struct switch_stack *sw;
791         struct pt_regs *pt;
792 #       define pt_reg_addr(pt, reg)     ((void *)                           \
793                                          ((unsigned long) (pt)              \
794                                           + offsetof(struct pt_regs, reg)))
795
796
797         pt = task_pt_regs(child);
798         sw = (struct switch_stack *) (child->thread.ksp + 16);
799
800         if ((addr & 0x7) != 0) {
801                 dprintk("ptrace: unaligned register address 0x%lx\n", addr);
802                 return -1;
803         }
804
805         if (addr < PT_F127 + 16) {
806                 /* accessing fph */
807                 if (write_access)
808                         ia64_sync_fph(child);
809                 else
810                         ia64_flush_fph(child);
811                 ptr = (unsigned long *)
812                         ((unsigned long) &child->thread.fph + addr);
813         } else if ((addr >= PT_F10) && (addr < PT_F11 + 16)) {
814                 /* scratch registers untouched by kernel (saved in pt_regs) */
815                 ptr = pt_reg_addr(pt, f10) + (addr - PT_F10);
816         } else if (addr >= PT_F12 && addr < PT_F15 + 16) {
817                 /*
818                  * Scratch registers untouched by kernel (saved in
819                  * switch_stack).
820                  */
821                 ptr = (unsigned long *) ((long) sw
822                                          + (addr - PT_NAT_BITS - 32));
823         } else if (addr < PT_AR_LC + 8) {
824                 /* preserved state: */
825                 struct unw_frame_info info;
826                 char nat = 0;
827                 int ret;
828
829                 unw_init_from_blocked_task(&info, child);
830                 if (unw_unwind_to_user(&info) < 0)
831                         return -1;
832
833                 switch (addr) {
834                       case PT_NAT_BITS:
835                         return access_nat_bits(child, pt, &info,
836                                                data, write_access);
837
838                       case PT_R4: case PT_R5: case PT_R6: case PT_R7:
839                         if (write_access) {
840                                 /* read NaT bit first: */
841                                 unsigned long dummy;
842
843                                 ret = unw_get_gr(&info, (addr - PT_R4)/8 + 4,
844                                                  &dummy, &nat);
845                                 if (ret < 0)
846                                         return ret;
847                         }
848                         return unw_access_gr(&info, (addr - PT_R4)/8 + 4, data,
849                                              &nat, write_access);
850
851                       case PT_B1: case PT_B2: case PT_B3:
852                       case PT_B4: case PT_B5:
853                         return unw_access_br(&info, (addr - PT_B1)/8 + 1, data,
854                                              write_access);
855
856                       case PT_AR_EC:
857                         return unw_access_ar(&info, UNW_AR_EC, data,
858                                              write_access);
859
860                       case PT_AR_LC:
861                         return unw_access_ar(&info, UNW_AR_LC, data,
862                                              write_access);
863
864                       default:
865                         if (addr >= PT_F2 && addr < PT_F5 + 16)
866                                 return access_fr(&info, (addr - PT_F2)/16 + 2,
867                                                  (addr & 8) != 0, data,
868                                                  write_access);
869                         else if (addr >= PT_F16 && addr < PT_F31 + 16)
870                                 return access_fr(&info,
871                                                  (addr - PT_F16)/16 + 16,
872                                                  (addr & 8) != 0,
873                                                  data, write_access);
874                         else {
875                                 dprintk("ptrace: rejecting access to register "
876                                         "address 0x%lx\n", addr);
877                                 return -1;
878                         }
879                 }
880         } else if (addr < PT_F9+16) {
881                 /* scratch state */
882                 switch (addr) {
883                       case PT_AR_BSP:
884                         /*
885                          * By convention, we use PT_AR_BSP to refer to
886                          * the end of the user-level backing store.
887                          * Use ia64_rse_skip_regs(PT_AR_BSP, -CFM.sof)
888                          * to get the real value of ar.bsp at the time
889                          * the kernel was entered.
890                          *
891                          * Furthermore, when changing the contents of
892                          * PT_AR_BSP (or PT_CFM) we MUST copy any
893                          * users-level stacked registers that are
894                          * stored on the kernel stack back to
895                          * user-space because otherwise, we might end
896                          * up clobbering kernel stacked registers.
897                          * Also, if this happens while the task is
898                          * blocked in a system call, which convert the
899                          * state such that the non-system-call exit
900                          * path is used.  This ensures that the proper
901                          * state will be picked up when resuming
902                          * execution.  However, it *also* means that
903                          * once we write PT_AR_BSP/PT_CFM, it won't be
904                          * possible to modify the syscall arguments of
905                          * the pending system call any longer.  This
906                          * shouldn't be an issue because modifying
907                          * PT_AR_BSP/PT_CFM generally implies that
908                          * we're either abandoning the pending system
909                          * call or that we defer it's re-execution
910                          * (e.g., due to GDB doing an inferior
911                          * function call).
912                          */
913                         urbs_end = ia64_get_user_rbs_end(child, pt, &cfm);
914                         if (write_access) {
915                                 if (*data != urbs_end) {
916                                         if (ia64_sync_user_rbs(child, sw,
917                                                                pt->ar_bspstore,
918                                                                urbs_end) < 0)
919                                                 return -1;
920                                         if (in_syscall(pt))
921                                                 convert_to_non_syscall(child,
922                                                                        pt,
923                                                                        cfm);
924                                         /*
925                                          * Simulate user-level write
926                                          * of ar.bsp:
927                                          */
928                                         pt->loadrs = 0;
929                                         pt->ar_bspstore = *data;
930                                 }
931                         } else
932                                 *data = urbs_end;
933                         return 0;
934
935                       case PT_CFM:
936                         urbs_end = ia64_get_user_rbs_end(child, pt, &cfm);
937                         if (write_access) {
938                                 if (((cfm ^ *data) & PFM_MASK) != 0) {
939                                         if (ia64_sync_user_rbs(child, sw,
940                                                                pt->ar_bspstore,
941                                                                urbs_end) < 0)
942                                                 return -1;
943                                         if (in_syscall(pt))
944                                                 convert_to_non_syscall(child,
945                                                                        pt,
946                                                                        cfm);
947                                         pt->cr_ifs = ((pt->cr_ifs & ~PFM_MASK)
948                                                       | (*data & PFM_MASK));
949                                 }
950                         } else
951                                 *data = cfm;
952                         return 0;
953
954                       case PT_CR_IPSR:
955                         if (write_access)
956                                 pt->cr_ipsr = ((*data & IPSR_MASK)
957                                                | (pt->cr_ipsr & ~IPSR_MASK));
958                         else
959                                 *data = (pt->cr_ipsr & IPSR_MASK);
960                         return 0;
961
962                       case PT_AR_RSC:
963                         if (write_access)
964                                 pt->ar_rsc = *data | (3 << 2); /* force PL3 */
965                         else
966                                 *data = pt->ar_rsc;
967                         return 0;
968
969                       case PT_AR_RNAT:
970                         urbs_end = ia64_get_user_rbs_end(child, pt, NULL);
971                         rnat_addr = (long) ia64_rse_rnat_addr((long *)
972                                                               urbs_end);
973                         if (write_access)
974                                 return ia64_poke(child, sw, urbs_end,
975                                                  rnat_addr, *data);
976                         else
977                                 return ia64_peek(child, sw, urbs_end,
978                                                  rnat_addr, data);
979
980                       case PT_R1:
981                         ptr = pt_reg_addr(pt, r1);
982                         break;
983                       case PT_R2:  case PT_R3:
984                         ptr = pt_reg_addr(pt, r2) + (addr - PT_R2);
985                         break;
986                       case PT_R8:  case PT_R9:  case PT_R10: case PT_R11:
987                         ptr = pt_reg_addr(pt, r8) + (addr - PT_R8);
988                         break;
989                       case PT_R12: case PT_R13:
990                         ptr = pt_reg_addr(pt, r12) + (addr - PT_R12);
991                         break;
992                       case PT_R14:
993                         ptr = pt_reg_addr(pt, r14);
994                         break;
995                       case PT_R15:
996                         ptr = pt_reg_addr(pt, r15);
997                         break;
998                       case PT_R16: case PT_R17: case PT_R18: case PT_R19:
999                       case PT_R20: case PT_R21: case PT_R22: case PT_R23:
1000                       case PT_R24: case PT_R25: case PT_R26: case PT_R27:
1001                       case PT_R28: case PT_R29: case PT_R30: case PT_R31:
1002                         ptr = pt_reg_addr(pt, r16) + (addr - PT_R16);
1003                         break;
1004                       case PT_B0:
1005                         ptr = pt_reg_addr(pt, b0);
1006                         break;
1007                       case PT_B6:
1008                         ptr = pt_reg_addr(pt, b6);
1009                         break;
1010                       case PT_B7:
1011                         ptr = pt_reg_addr(pt, b7);
1012                         break;
1013                       case PT_F6:  case PT_F6+8: case PT_F7: case PT_F7+8:
1014                       case PT_F8:  case PT_F8+8: case PT_F9: case PT_F9+8:
1015                         ptr = pt_reg_addr(pt, f6) + (addr - PT_F6);
1016                         break;
1017                       case PT_AR_BSPSTORE:
1018                         ptr = pt_reg_addr(pt, ar_bspstore);
1019                         break;
1020                       case PT_AR_UNAT:
1021                         ptr = pt_reg_addr(pt, ar_unat);
1022                         break;
1023                       case PT_AR_PFS:
1024                         ptr = pt_reg_addr(pt, ar_pfs);
1025                         break;
1026                       case PT_AR_CCV:
1027                         ptr = pt_reg_addr(pt, ar_ccv);
1028                         break;
1029                       case PT_AR_FPSR:
1030                         ptr = pt_reg_addr(pt, ar_fpsr);
1031                         break;
1032                       case PT_CR_IIP:
1033                         ptr = pt_reg_addr(pt, cr_iip);
1034                         break;
1035                       case PT_PR:
1036                         ptr = pt_reg_addr(pt, pr);
1037                         break;
1038                         /* scratch register */
1039
1040                       default:
1041                         /* disallow accessing anything else... */
1042                         dprintk("ptrace: rejecting access to register "
1043                                 "address 0x%lx\n", addr);
1044                         return -1;
1045                 }
1046         } else if (addr <= PT_AR_SSD) {
1047                 ptr = pt_reg_addr(pt, ar_csd) + (addr - PT_AR_CSD);
1048         } else {
1049                 /* access debug registers */
1050
1051                 if (addr >= PT_IBR) {
1052                         regnum = (addr - PT_IBR) >> 3;
1053                         ptr = &child->thread.ibr[0];
1054                 } else {
1055                         regnum = (addr - PT_DBR) >> 3;
1056                         ptr = &child->thread.dbr[0];
1057                 }
1058
1059                 if (regnum >= 8) {
1060                         dprintk("ptrace: rejecting access to register "
1061                                 "address 0x%lx\n", addr);
1062                         return -1;
1063                 }
1064 #ifdef CONFIG_PERFMON
1065                 /*
1066                  * Check if debug registers are used by perfmon. This
1067                  * test must be done once we know that we can do the
1068                  * operation, i.e. the arguments are all valid, but
1069                  * before we start modifying the state.
1070                  *
1071                  * Perfmon needs to keep a count of how many processes
1072                  * are trying to modify the debug registers for system
1073                  * wide monitoring sessions.
1074                  *
1075                  * We also include read access here, because they may
1076                  * cause the PMU-installed debug register state
1077                  * (dbr[], ibr[]) to be reset. The two arrays are also
1078                  * used by perfmon, but we do not use
1079                  * IA64_THREAD_DBG_VALID. The registers are restored
1080                  * by the PMU context switch code.
1081                  */
1082                 if (pfm_use_debug_registers(child)) return -1;
1083 #endif
1084
1085                 if (!(child->thread.flags & IA64_THREAD_DBG_VALID)) {
1086                         child->thread.flags |= IA64_THREAD_DBG_VALID;
1087                         memset(child->thread.dbr, 0,
1088                                sizeof(child->thread.dbr));
1089                         memset(child->thread.ibr, 0,
1090                                sizeof(child->thread.ibr));
1091                 }
1092
1093                 ptr += regnum;
1094
1095                 if ((regnum & 1) && write_access) {
1096                         /* don't let the user set kernel-level breakpoints: */
1097                         *ptr = *data & ~(7UL << 56);
1098                         return 0;
1099                 }
1100         }
1101         if (write_access)
1102                 *ptr = *data;
1103         else
1104                 *data = *ptr;
1105         return 0;
1106 }
1107
1108 static long
1109 ptrace_getregs (struct task_struct *child, struct pt_all_user_regs __user *ppr)
1110 {
1111         unsigned long psr, ec, lc, rnat, bsp, cfm, nat_bits, val;
1112         struct unw_frame_info info;
1113         struct ia64_fpreg fpval;
1114         struct switch_stack *sw;
1115         struct pt_regs *pt;
1116         long ret, retval = 0;
1117         char nat = 0;
1118         int i;
1119
1120         if (!access_ok(VERIFY_WRITE, ppr, sizeof(struct pt_all_user_regs)))
1121                 return -EIO;
1122
1123         pt = task_pt_regs(child);
1124         sw = (struct switch_stack *) (child->thread.ksp + 16);
1125         unw_init_from_blocked_task(&info, child);
1126         if (unw_unwind_to_user(&info) < 0) {
1127                 return -EIO;
1128         }
1129
1130         if (((unsigned long) ppr & 0x7) != 0) {
1131                 dprintk("ptrace:unaligned register address %p\n", ppr);
1132                 return -EIO;
1133         }
1134
1135         if (access_uarea(child, PT_CR_IPSR, &psr, 0) < 0
1136             || access_uarea(child, PT_AR_EC, &ec, 0) < 0
1137             || access_uarea(child, PT_AR_LC, &lc, 0) < 0
1138             || access_uarea(child, PT_AR_RNAT, &rnat, 0) < 0
1139             || access_uarea(child, PT_AR_BSP, &bsp, 0) < 0
1140             || access_uarea(child, PT_CFM, &cfm, 0)
1141             || access_uarea(child, PT_NAT_BITS, &nat_bits, 0))
1142                 return -EIO;
1143
1144         /* control regs */
1145
1146         retval |= __put_user(pt->cr_iip, &ppr->cr_iip);
1147         retval |= __put_user(psr, &ppr->cr_ipsr);
1148
1149         /* app regs */
1150
1151         retval |= __put_user(pt->ar_pfs, &ppr->ar[PT_AUR_PFS]);
1152         retval |= __put_user(pt->ar_rsc, &ppr->ar[PT_AUR_RSC]);
1153         retval |= __put_user(pt->ar_bspstore, &ppr->ar[PT_AUR_BSPSTORE]);
1154         retval |= __put_user(pt->ar_unat, &ppr->ar[PT_AUR_UNAT]);
1155         retval |= __put_user(pt->ar_ccv, &ppr->ar[PT_AUR_CCV]);
1156         retval |= __put_user(pt->ar_fpsr, &ppr->ar[PT_AUR_FPSR]);
1157
1158         retval |= __put_user(ec, &ppr->ar[PT_AUR_EC]);
1159         retval |= __put_user(lc, &ppr->ar[PT_AUR_LC]);
1160         retval |= __put_user(rnat, &ppr->ar[PT_AUR_RNAT]);
1161         retval |= __put_user(bsp, &ppr->ar[PT_AUR_BSP]);
1162         retval |= __put_user(cfm, &ppr->cfm);
1163
1164         /* gr1-gr3 */
1165
1166         retval |= __copy_to_user(&ppr->gr[1], &pt->r1, sizeof(long));
1167         retval |= __copy_to_user(&ppr->gr[2], &pt->r2, sizeof(long) *2);
1168
1169         /* gr4-gr7 */
1170
1171         for (i = 4; i < 8; i++) {
1172                 if (unw_access_gr(&info, i, &val, &nat, 0) < 0)
1173                         return -EIO;
1174                 retval |= __put_user(val, &ppr->gr[i]);
1175         }
1176
1177         /* gr8-gr11 */
1178
1179         retval |= __copy_to_user(&ppr->gr[8], &pt->r8, sizeof(long) * 4);
1180
1181         /* gr12-gr15 */
1182
1183         retval |= __copy_to_user(&ppr->gr[12], &pt->r12, sizeof(long) * 2);
1184         retval |= __copy_to_user(&ppr->gr[14], &pt->r14, sizeof(long));
1185         retval |= __copy_to_user(&ppr->gr[15], &pt->r15, sizeof(long));
1186
1187         /* gr16-gr31 */
1188
1189         retval |= __copy_to_user(&ppr->gr[16], &pt->r16, sizeof(long) * 16);
1190
1191         /* b0 */
1192
1193         retval |= __put_user(pt->b0, &ppr->br[0]);
1194
1195         /* b1-b5 */
1196
1197         for (i = 1; i < 6; i++) {
1198                 if (unw_access_br(&info, i, &val, 0) < 0)
1199                         return -EIO;
1200                 __put_user(val, &ppr->br[i]);
1201         }
1202
1203         /* b6-b7 */
1204
1205         retval |= __put_user(pt->b6, &ppr->br[6]);
1206         retval |= __put_user(pt->b7, &ppr->br[7]);
1207
1208         /* fr2-fr5 */
1209
1210         for (i = 2; i < 6; i++) {
1211                 if (unw_get_fr(&info, i, &fpval) < 0)
1212                         return -EIO;
1213                 retval |= __copy_to_user(&ppr->fr[i], &fpval, sizeof (fpval));
1214         }
1215
1216         /* fr6-fr11 */
1217
1218         retval |= __copy_to_user(&ppr->fr[6], &pt->f6,
1219                                  sizeof(struct ia64_fpreg) * 6);
1220
1221         /* fp scratch regs(12-15) */
1222
1223         retval |= __copy_to_user(&ppr->fr[12], &sw->f12,
1224                                  sizeof(struct ia64_fpreg) * 4);
1225
1226         /* fr16-fr31 */
1227
1228         for (i = 16; i < 32; i++) {
1229                 if (unw_get_fr(&info, i, &fpval) < 0)
1230                         return -EIO;
1231                 retval |= __copy_to_user(&ppr->fr[i], &fpval, sizeof (fpval));
1232         }
1233
1234         /* fph */
1235
1236         ia64_flush_fph(child);
1237         retval |= __copy_to_user(&ppr->fr[32], &child->thread.fph,
1238                                  sizeof(ppr->fr[32]) * 96);
1239
1240         /*  preds */
1241
1242         retval |= __put_user(pt->pr, &ppr->pr);
1243
1244         /* nat bits */
1245
1246         retval |= __put_user(nat_bits, &ppr->nat);
1247
1248         ret = retval ? -EIO : 0;
1249         return ret;
1250 }
1251
1252 static long
1253 ptrace_setregs (struct task_struct *child, struct pt_all_user_regs __user *ppr)
1254 {
1255         unsigned long psr, rsc, ec, lc, rnat, bsp, cfm, nat_bits, val = 0;
1256         struct unw_frame_info info;
1257         struct switch_stack *sw;
1258         struct ia64_fpreg fpval;
1259         struct pt_regs *pt;
1260         long ret, retval = 0;
1261         int i;
1262
1263         memset(&fpval, 0, sizeof(fpval));
1264
1265         if (!access_ok(VERIFY_READ, ppr, sizeof(struct pt_all_user_regs)))
1266                 return -EIO;
1267
1268         pt = task_pt_regs(child);
1269         sw = (struct switch_stack *) (child->thread.ksp + 16);
1270         unw_init_from_blocked_task(&info, child);
1271         if (unw_unwind_to_user(&info) < 0) {
1272                 return -EIO;
1273         }
1274
1275         if (((unsigned long) ppr & 0x7) != 0) {
1276                 dprintk("ptrace:unaligned register address %p\n", ppr);
1277                 return -EIO;
1278         }
1279
1280         /* control regs */
1281
1282         retval |= __get_user(pt->cr_iip, &ppr->cr_iip);
1283         retval |= __get_user(psr, &ppr->cr_ipsr);
1284
1285         /* app regs */
1286
1287         retval |= __get_user(pt->ar_pfs, &ppr->ar[PT_AUR_PFS]);
1288         retval |= __get_user(rsc, &ppr->ar[PT_AUR_RSC]);
1289         retval |= __get_user(pt->ar_bspstore, &ppr->ar[PT_AUR_BSPSTORE]);
1290         retval |= __get_user(pt->ar_unat, &ppr->ar[PT_AUR_UNAT]);
1291         retval |= __get_user(pt->ar_ccv, &ppr->ar[PT_AUR_CCV]);
1292         retval |= __get_user(pt->ar_fpsr, &ppr->ar[PT_AUR_FPSR]);
1293
1294         retval |= __get_user(ec, &ppr->ar[PT_AUR_EC]);
1295         retval |= __get_user(lc, &ppr->ar[PT_AUR_LC]);
1296         retval |= __get_user(rnat, &ppr->ar[PT_AUR_RNAT]);
1297         retval |= __get_user(bsp, &ppr->ar[PT_AUR_BSP]);
1298         retval |= __get_user(cfm, &ppr->cfm);
1299
1300         /* gr1-gr3 */
1301
1302         retval |= __copy_from_user(&pt->r1, &ppr->gr[1], sizeof(long));
1303         retval |= __copy_from_user(&pt->r2, &ppr->gr[2], sizeof(long) * 2);
1304
1305         /* gr4-gr7 */
1306
1307         for (i = 4; i < 8; i++) {
1308                 retval |= __get_user(val, &ppr->gr[i]);
1309                 /* NaT bit will be set via PT_NAT_BITS: */
1310                 if (unw_set_gr(&info, i, val, 0) < 0)
1311                         return -EIO;
1312         }
1313
1314         /* gr8-gr11 */
1315
1316         retval |= __copy_from_user(&pt->r8, &ppr->gr[8], sizeof(long) * 4);
1317
1318         /* gr12-gr15 */
1319
1320         retval |= __copy_from_user(&pt->r12, &ppr->gr[12], sizeof(long) * 2);
1321         retval |= __copy_from_user(&pt->r14, &ppr->gr[14], sizeof(long));
1322         retval |= __copy_from_user(&pt->r15, &ppr->gr[15], sizeof(long));
1323
1324         /* gr16-gr31 */
1325
1326         retval |= __copy_from_user(&pt->r16, &ppr->gr[16], sizeof(long) * 16);
1327
1328         /* b0 */
1329
1330         retval |= __get_user(pt->b0, &ppr->br[0]);
1331
1332         /* b1-b5 */
1333
1334         for (i = 1; i < 6; i++) {
1335                 retval |= __get_user(val, &ppr->br[i]);
1336                 unw_set_br(&info, i, val);
1337         }
1338
1339         /* b6-b7 */
1340
1341         retval |= __get_user(pt->b6, &ppr->br[6]);
1342         retval |= __get_user(pt->b7, &ppr->br[7]);
1343
1344         /* fr2-fr5 */
1345
1346         for (i = 2; i < 6; i++) {
1347                 retval |= __copy_from_user(&fpval, &ppr->fr[i], sizeof(fpval));
1348                 if (unw_set_fr(&info, i, fpval) < 0)
1349                         return -EIO;
1350         }
1351
1352         /* fr6-fr11 */
1353
1354         retval |= __copy_from_user(&pt->f6, &ppr->fr[6],
1355                                    sizeof(ppr->fr[6]) * 6);
1356
1357         /* fp scratch regs(12-15) */
1358
1359         retval |= __copy_from_user(&sw->f12, &ppr->fr[12],
1360                                    sizeof(ppr->fr[12]) * 4);
1361
1362         /* fr16-fr31 */
1363
1364         for (i = 16; i < 32; i++) {
1365                 retval |= __copy_from_user(&fpval, &ppr->fr[i],
1366                                            sizeof(fpval));
1367                 if (unw_set_fr(&info, i, fpval) < 0)
1368                         return -EIO;
1369         }
1370
1371         /* fph */
1372
1373         ia64_sync_fph(child);
1374         retval |= __copy_from_user(&child->thread.fph, &ppr->fr[32],
1375                                    sizeof(ppr->fr[32]) * 96);
1376
1377         /* preds */
1378
1379         retval |= __get_user(pt->pr, &ppr->pr);
1380
1381         /* nat bits */
1382
1383         retval |= __get_user(nat_bits, &ppr->nat);
1384
1385         retval |= access_uarea(child, PT_CR_IPSR, &psr, 1);
1386         retval |= access_uarea(child, PT_AR_RSC, &rsc, 1);
1387         retval |= access_uarea(child, PT_AR_EC, &ec, 1);
1388         retval |= access_uarea(child, PT_AR_LC, &lc, 1);
1389         retval |= access_uarea(child, PT_AR_RNAT, &rnat, 1);
1390         retval |= access_uarea(child, PT_AR_BSP, &bsp, 1);
1391         retval |= access_uarea(child, PT_CFM, &cfm, 1);
1392         retval |= access_uarea(child, PT_NAT_BITS, &nat_bits, 1);
1393
1394         ret = retval ? -EIO : 0;
1395         return ret;
1396 }
1397
1398 /*
1399  * Called by kernel/ptrace.c when detaching..
1400  *
1401  * Make sure the single step bit is not set.
1402  */
1403 void
1404 ptrace_disable (struct task_struct *child)
1405 {
1406         struct ia64_psr *child_psr = ia64_psr(task_pt_regs(child));
1407
1408         /* make sure the single step/taken-branch trap bits are not set: */
1409         child_psr->ss = 0;
1410         child_psr->tb = 0;
1411 }
1412
1413 asmlinkage long
1414 sys_ptrace (long request, pid_t pid, unsigned long addr, unsigned long data)
1415 {
1416         struct pt_regs *pt;
1417         unsigned long urbs_end, peek_or_poke;
1418         struct task_struct *child;
1419         struct switch_stack *sw;
1420         long ret;
1421
1422         lock_kernel();
1423         ret = -EPERM;
1424         if (request == PTRACE_TRACEME) {
1425                 ret = ptrace_traceme();
1426                 goto out;
1427         }
1428
1429         peek_or_poke = (request == PTRACE_PEEKTEXT
1430                         || request == PTRACE_PEEKDATA
1431                         || request == PTRACE_POKETEXT
1432                         || request == PTRACE_POKEDATA);
1433         ret = -ESRCH;
1434         read_lock(&tasklist_lock);
1435         {
1436                 child = find_task_by_pid(pid);
1437                 if (child) {
1438                         if (peek_or_poke)
1439                                 child = find_thread_for_addr(child, addr);
1440                         get_task_struct(child);
1441                 }
1442         }
1443         read_unlock(&tasklist_lock);
1444         if (!child)
1445                 goto out;
1446         ret = -EPERM;
1447         if (pid == 1)           /* no messing around with init! */
1448                 goto out_tsk;
1449
1450         if (request == PTRACE_ATTACH) {
1451                 ret = ptrace_attach(child);
1452                 goto out_tsk;
1453         }
1454
1455         ret = ptrace_check_attach(child, request == PTRACE_KILL);
1456         if (ret < 0)
1457                 goto out_tsk;
1458
1459         pt = task_pt_regs(child);
1460         sw = (struct switch_stack *) (child->thread.ksp + 16);
1461
1462         switch (request) {
1463               case PTRACE_PEEKTEXT:
1464               case PTRACE_PEEKDATA:
1465                 /* read word at location addr */
1466                 urbs_end = ia64_get_user_rbs_end(child, pt, NULL);
1467                 ret = ia64_peek(child, sw, urbs_end, addr, &data);
1468                 if (ret == 0) {
1469                         ret = data;
1470                         /* ensure "ret" is not mistaken as an error code: */
1471                         force_successful_syscall_return();
1472                 }
1473                 goto out_tsk;
1474
1475               case PTRACE_POKETEXT:
1476               case PTRACE_POKEDATA:
1477                 /* write the word at location addr */
1478                 urbs_end = ia64_get_user_rbs_end(child, pt, NULL);
1479                 ret = ia64_poke(child, sw, urbs_end, addr, data);
1480                 goto out_tsk;
1481
1482               case PTRACE_PEEKUSR:
1483                 /* read the word at addr in the USER area */
1484                 if (access_uarea(child, addr, &data, 0) < 0) {
1485                         ret = -EIO;
1486                         goto out_tsk;
1487                 }
1488                 ret = data;
1489                 /* ensure "ret" is not mistaken as an error code */
1490                 force_successful_syscall_return();
1491                 goto out_tsk;
1492
1493               case PTRACE_POKEUSR:
1494                 /* write the word at addr in the USER area */
1495                 if (access_uarea(child, addr, &data, 1) < 0) {
1496                         ret = -EIO;
1497                         goto out_tsk;
1498                 }
1499                 ret = 0;
1500                 goto out_tsk;
1501
1502               case PTRACE_OLD_GETSIGINFO:
1503                 /* for backwards-compatibility */
1504                 ret = ptrace_request(child, PTRACE_GETSIGINFO, addr, data);
1505                 goto out_tsk;
1506
1507               case PTRACE_OLD_SETSIGINFO:
1508                 /* for backwards-compatibility */
1509                 ret = ptrace_request(child, PTRACE_SETSIGINFO, addr, data);
1510                 goto out_tsk;
1511
1512               case PTRACE_SYSCALL:
1513                 /* continue and stop at next (return from) syscall */
1514               case PTRACE_CONT:
1515                 /* restart after signal. */
1516                 ret = -EIO;
1517                 if (!valid_signal(data))
1518                         goto out_tsk;
1519                 if (request == PTRACE_SYSCALL)
1520                         set_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
1521                 else
1522                         clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
1523                 child->exit_code = data;
1524
1525                 /*
1526                  * Make sure the single step/taken-branch trap bits
1527                  * are not set:
1528                  */
1529                 ia64_psr(pt)->ss = 0;
1530                 ia64_psr(pt)->tb = 0;
1531
1532                 wake_up_process(child);
1533                 ret = 0;
1534                 goto out_tsk;
1535
1536               case PTRACE_KILL:
1537                 /*
1538                  * Make the child exit.  Best I can do is send it a
1539                  * sigkill.  Perhaps it should be put in the status
1540                  * that it wants to exit.
1541                  */
1542                 if (child->exit_state == EXIT_ZOMBIE)
1543                         /* already dead */
1544                         goto out_tsk;
1545                 child->exit_code = SIGKILL;
1546
1547                 ptrace_disable(child);
1548                 wake_up_process(child);
1549                 ret = 0;
1550                 goto out_tsk;
1551
1552               case PTRACE_SINGLESTEP:
1553                 /* let child execute for one instruction */
1554               case PTRACE_SINGLEBLOCK:
1555                 ret = -EIO;
1556                 if (!valid_signal(data))
1557                         goto out_tsk;
1558
1559                 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
1560                 if (request == PTRACE_SINGLESTEP) {
1561                         ia64_psr(pt)->ss = 1;
1562                 } else {
1563                         ia64_psr(pt)->tb = 1;
1564                 }
1565                 child->exit_code = data;
1566
1567                 /* give it a chance to run. */
1568                 wake_up_process(child);
1569                 ret = 0;
1570                 goto out_tsk;
1571
1572               case PTRACE_DETACH:
1573                 /* detach a process that was attached. */
1574                 ret = ptrace_detach(child, data);
1575                 goto out_tsk;
1576
1577               case PTRACE_GETREGS:
1578                 ret = ptrace_getregs(child,
1579                                      (struct pt_all_user_regs __user *) data);
1580                 goto out_tsk;
1581
1582               case PTRACE_SETREGS:
1583                 ret = ptrace_setregs(child,
1584                                      (struct pt_all_user_regs __user *) data);
1585                 goto out_tsk;
1586
1587               default:
1588                 ret = ptrace_request(child, request, addr, data);
1589                 goto out_tsk;
1590         }
1591   out_tsk:
1592         put_task_struct(child);
1593   out:
1594         unlock_kernel();
1595         return ret;
1596 }
1597
1598
1599 void
1600 syscall_trace (void)
1601 {
1602         if (!test_thread_flag(TIF_SYSCALL_TRACE))
1603                 return;
1604         if (!(current->ptrace & PT_PTRACED))
1605                 return;
1606         /*
1607          * The 0x80 provides a way for the tracing parent to
1608          * distinguish between a syscall stop and SIGTRAP delivery.
1609          */
1610         ptrace_notify(SIGTRAP
1611                       | ((current->ptrace & PT_TRACESYSGOOD) ? 0x80 : 0));
1612
1613         /*
1614          * This isn't the same as continuing with a signal, but it
1615          * will do for normal use.  strace only continues with a
1616          * signal if the stopping signal is not SIGTRAP.  -brl
1617          */
1618         if (current->exit_code) {
1619                 send_sig(current->exit_code, current, 1);
1620                 current->exit_code = 0;
1621         }
1622 }
1623
1624 /* "asmlinkage" so the input arguments are preserved... */
1625
1626 asmlinkage void
1627 syscall_trace_enter (long arg0, long arg1, long arg2, long arg3,
1628                      long arg4, long arg5, long arg6, long arg7,
1629                      struct pt_regs regs)
1630 {
1631         if (test_thread_flag(TIF_SYSCALL_TRACE) 
1632             && (current->ptrace & PT_PTRACED))
1633                 syscall_trace();
1634
1635         if (unlikely(current->audit_context)) {
1636                 long syscall;
1637                 int arch;
1638
1639                 if (IS_IA32_PROCESS(&regs)) {
1640                         syscall = regs.r1;
1641                         arch = AUDIT_ARCH_I386;
1642                 } else {
1643                         syscall = regs.r15;
1644                         arch = AUDIT_ARCH_IA64;
1645                 }
1646
1647                 audit_syscall_entry(arch, syscall, arg0, arg1, arg2, arg3);
1648         }
1649
1650 }
1651
1652 /* "asmlinkage" so the input arguments are preserved... */
1653
1654 asmlinkage void
1655 syscall_trace_leave (long arg0, long arg1, long arg2, long arg3,
1656                      long arg4, long arg5, long arg6, long arg7,
1657                      struct pt_regs regs)
1658 {
1659         if (unlikely(current->audit_context)) {
1660                 int success = AUDITSC_RESULT(regs.r10);
1661                 long result = regs.r8;
1662
1663                 if (success != AUDITSC_SUCCESS)
1664                         result = -result;
1665                 audit_syscall_exit(success, result);
1666         }
1667
1668         if (test_thread_flag(TIF_SYSCALL_TRACE)
1669             && (current->ptrace & PT_PTRACED))
1670                 syscall_trace();
1671 }