Merge branch 'next' of git://git.kernel.org/pub/scm/linux/kernel/git/djbw/async_tx
[linux-2.6.git] / 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/mm.h>
15 #include <linux/errno.h>
16 #include <linux/ptrace.h>
17 #include <linux/user.h>
18 #include <linux/security.h>
19 #include <linux/audit.h>
20 #include <linux/signal.h>
21 #include <linux/regset.h>
22 #include <linux/elf.h>
23 #include <linux/tracehook.h>
24
25 #include <asm/pgtable.h>
26 #include <asm/processor.h>
27 #include <asm/ptrace_offsets.h>
28 #include <asm/rse.h>
29 #include <asm/system.h>
30 #include <asm/uaccess.h>
31 #include <asm/unwind.h>
32 #ifdef CONFIG_PERFMON
33 #include <asm/perfmon.h>
34 #endif
35
36 #include "entry.h"
37
38 /*
39  * Bits in the PSR that we allow ptrace() to change:
40  *      be, up, ac, mfl, mfh (the user mask; five bits total)
41  *      db (debug breakpoint fault; one bit)
42  *      id (instruction debug fault disable; one bit)
43  *      dd (data debug fault disable; one bit)
44  *      ri (restart instruction; two bits)
45  *      is (instruction set; one bit)
46  */
47 #define IPSR_MASK (IA64_PSR_UM | IA64_PSR_DB | IA64_PSR_IS      \
48                    | IA64_PSR_ID | IA64_PSR_DD | IA64_PSR_RI)
49
50 #define MASK(nbits)     ((1UL << (nbits)) - 1)  /* mask with NBITS bits set */
51 #define PFM_MASK        MASK(38)
52
53 #define PTRACE_DEBUG    0
54
55 #if PTRACE_DEBUG
56 # define dprintk(format...)     printk(format)
57 # define inline
58 #else
59 # define dprintk(format...)
60 #endif
61
62 /* Return TRUE if PT was created due to kernel-entry via a system-call.  */
63
64 static inline int
65 in_syscall (struct pt_regs *pt)
66 {
67         return (long) pt->cr_ifs >= 0;
68 }
69
70 /*
71  * Collect the NaT bits for r1-r31 from scratch_unat and return a NaT
72  * bitset where bit i is set iff the NaT bit of register i is set.
73  */
74 unsigned long
75 ia64_get_scratch_nat_bits (struct pt_regs *pt, unsigned long scratch_unat)
76 {
77 #       define GET_BITS(first, last, unat)                              \
78         ({                                                              \
79                 unsigned long bit = ia64_unat_pos(&pt->r##first);       \
80                 unsigned long nbits = (last - first + 1);               \
81                 unsigned long mask = MASK(nbits) << first;              \
82                 unsigned long dist;                                     \
83                 if (bit < first)                                        \
84                         dist = 64 + bit - first;                        \
85                 else                                                    \
86                         dist = bit - first;                             \
87                 ia64_rotr(unat, dist) & mask;                           \
88         })
89         unsigned long val;
90
91         /*
92          * Registers that are stored consecutively in struct pt_regs
93          * can be handled in parallel.  If the register order in
94          * struct_pt_regs changes, this code MUST be updated.
95          */
96         val  = GET_BITS( 1,  1, scratch_unat);
97         val |= GET_BITS( 2,  3, scratch_unat);
98         val |= GET_BITS(12, 13, scratch_unat);
99         val |= GET_BITS(14, 14, scratch_unat);
100         val |= GET_BITS(15, 15, scratch_unat);
101         val |= GET_BITS( 8, 11, scratch_unat);
102         val |= GET_BITS(16, 31, scratch_unat);
103         return val;
104
105 #       undef GET_BITS
106 }
107
108 /*
109  * Set the NaT bits for the scratch registers according to NAT and
110  * return the resulting unat (assuming the scratch registers are
111  * stored in PT).
112  */
113 unsigned long
114 ia64_put_scratch_nat_bits (struct pt_regs *pt, unsigned long nat)
115 {
116 #       define PUT_BITS(first, last, nat)                               \
117         ({                                                              \
118                 unsigned long bit = ia64_unat_pos(&pt->r##first);       \
119                 unsigned long nbits = (last - first + 1);               \
120                 unsigned long mask = MASK(nbits) << first;              \
121                 long dist;                                              \
122                 if (bit < first)                                        \
123                         dist = 64 + bit - first;                        \
124                 else                                                    \
125                         dist = bit - first;                             \
126                 ia64_rotl(nat & mask, dist);                            \
127         })
128         unsigned long scratch_unat;
129
130         /*
131          * Registers that are stored consecutively in struct pt_regs
132          * can be handled in parallel.  If the register order in
133          * struct_pt_regs changes, this code MUST be updated.
134          */
135         scratch_unat  = PUT_BITS( 1,  1, nat);
136         scratch_unat |= PUT_BITS( 2,  3, nat);
137         scratch_unat |= PUT_BITS(12, 13, nat);
138         scratch_unat |= PUT_BITS(14, 14, nat);
139         scratch_unat |= PUT_BITS(15, 15, nat);
140         scratch_unat |= PUT_BITS( 8, 11, nat);
141         scratch_unat |= PUT_BITS(16, 31, nat);
142
143         return scratch_unat;
144
145 #       undef PUT_BITS
146 }
147
148 #define IA64_MLX_TEMPLATE       0x2
149 #define IA64_MOVL_OPCODE        6
150
151 void
152 ia64_increment_ip (struct pt_regs *regs)
153 {
154         unsigned long w0, ri = ia64_psr(regs)->ri + 1;
155
156         if (ri > 2) {
157                 ri = 0;
158                 regs->cr_iip += 16;
159         } else if (ri == 2) {
160                 get_user(w0, (char __user *) regs->cr_iip + 0);
161                 if (((w0 >> 1) & 0xf) == IA64_MLX_TEMPLATE) {
162                         /*
163                          * rfi'ing to slot 2 of an MLX bundle causes
164                          * an illegal operation fault.  We don't want
165                          * that to happen...
166                          */
167                         ri = 0;
168                         regs->cr_iip += 16;
169                 }
170         }
171         ia64_psr(regs)->ri = ri;
172 }
173
174 void
175 ia64_decrement_ip (struct pt_regs *regs)
176 {
177         unsigned long w0, ri = ia64_psr(regs)->ri - 1;
178
179         if (ia64_psr(regs)->ri == 0) {
180                 regs->cr_iip -= 16;
181                 ri = 2;
182                 get_user(w0, (char __user *) regs->cr_iip + 0);
183                 if (((w0 >> 1) & 0xf) == IA64_MLX_TEMPLATE) {
184                         /*
185                          * rfi'ing to slot 2 of an MLX bundle causes
186                          * an illegal operation fault.  We don't want
187                          * that to happen...
188                          */
189                         ri = 1;
190                 }
191         }
192         ia64_psr(regs)->ri = ri;
193 }
194
195 /*
196  * This routine is used to read an rnat bits that are stored on the
197  * kernel backing store.  Since, in general, the alignment of the user
198  * and kernel are different, this is not completely trivial.  In
199  * essence, we need to construct the user RNAT based on up to two
200  * kernel RNAT values and/or the RNAT value saved in the child's
201  * pt_regs.
202  *
203  * user rbs
204  *
205  * +--------+ <-- lowest address
206  * | slot62 |
207  * +--------+
208  * |  rnat  | 0x....1f8
209  * +--------+
210  * | slot00 | \
211  * +--------+ |
212  * | slot01 | > child_regs->ar_rnat
213  * +--------+ |
214  * | slot02 | /                         kernel rbs
215  * +--------+                           +--------+
216  *          <- child_regs->ar_bspstore  | slot61 | <-- krbs
217  * +- - - - +                           +--------+
218  *                                      | slot62 |
219  * +- - - - +                           +--------+
220  *                                      |  rnat  |
221  * +- - - - +                           +--------+
222  *   vrnat                              | slot00 |
223  * +- - - - +                           +--------+
224  *                                      =        =
225  *                                      +--------+
226  *                                      | slot00 | \
227  *                                      +--------+ |
228  *                                      | slot01 | > child_stack->ar_rnat
229  *                                      +--------+ |
230  *                                      | slot02 | /
231  *                                      +--------+
232  *                                                <--- child_stack->ar_bspstore
233  *
234  * The way to think of this code is as follows: bit 0 in the user rnat
235  * corresponds to some bit N (0 <= N <= 62) in one of the kernel rnat
236  * value.  The kernel rnat value holding this bit is stored in
237  * variable rnat0.  rnat1 is loaded with the kernel rnat value that
238  * form the upper bits of the user rnat value.
239  *
240  * Boundary cases:
241  *
242  * o when reading the rnat "below" the first rnat slot on the kernel
243  *   backing store, rnat0/rnat1 are set to 0 and the low order bits are
244  *   merged in from pt->ar_rnat.
245  *
246  * o when reading the rnat "above" the last rnat slot on the kernel
247  *   backing store, rnat0/rnat1 gets its value from sw->ar_rnat.
248  */
249 static unsigned long
250 get_rnat (struct task_struct *task, struct switch_stack *sw,
251           unsigned long *krbs, unsigned long *urnat_addr,
252           unsigned long *urbs_end)
253 {
254         unsigned long rnat0 = 0, rnat1 = 0, urnat = 0, *slot0_kaddr;
255         unsigned long umask = 0, mask, m;
256         unsigned long *kbsp, *ubspstore, *rnat0_kaddr, *rnat1_kaddr, shift;
257         long num_regs, nbits;
258         struct pt_regs *pt;
259
260         pt = task_pt_regs(task);
261         kbsp = (unsigned long *) sw->ar_bspstore;
262         ubspstore = (unsigned long *) pt->ar_bspstore;
263
264         if (urbs_end < urnat_addr)
265                 nbits = ia64_rse_num_regs(urnat_addr - 63, urbs_end);
266         else
267                 nbits = 63;
268         mask = MASK(nbits);
269         /*
270          * First, figure out which bit number slot 0 in user-land maps
271          * to in the kernel rnat.  Do this by figuring out how many
272          * register slots we're beyond the user's backingstore and
273          * then computing the equivalent address in kernel space.
274          */
275         num_regs = ia64_rse_num_regs(ubspstore, urnat_addr + 1);
276         slot0_kaddr = ia64_rse_skip_regs(krbs, num_regs);
277         shift = ia64_rse_slot_num(slot0_kaddr);
278         rnat1_kaddr = ia64_rse_rnat_addr(slot0_kaddr);
279         rnat0_kaddr = rnat1_kaddr - 64;
280
281         if (ubspstore + 63 > urnat_addr) {
282                 /* some bits need to be merged in from pt->ar_rnat */
283                 umask = MASK(ia64_rse_slot_num(ubspstore)) & mask;
284                 urnat = (pt->ar_rnat & umask);
285                 mask &= ~umask;
286                 if (!mask)
287                         return urnat;
288         }
289
290         m = mask << shift;
291         if (rnat0_kaddr >= kbsp)
292                 rnat0 = sw->ar_rnat;
293         else if (rnat0_kaddr > krbs)
294                 rnat0 = *rnat0_kaddr;
295         urnat |= (rnat0 & m) >> shift;
296
297         m = mask >> (63 - shift);
298         if (rnat1_kaddr >= kbsp)
299                 rnat1 = sw->ar_rnat;
300         else if (rnat1_kaddr > krbs)
301                 rnat1 = *rnat1_kaddr;
302         urnat |= (rnat1 & m) << (63 - shift);
303         return urnat;
304 }
305
306 /*
307  * The reverse of get_rnat.
308  */
309 static void
310 put_rnat (struct task_struct *task, struct switch_stack *sw,
311           unsigned long *krbs, unsigned long *urnat_addr, unsigned long urnat,
312           unsigned long *urbs_end)
313 {
314         unsigned long rnat0 = 0, rnat1 = 0, *slot0_kaddr, umask = 0, mask, m;
315         unsigned long *kbsp, *ubspstore, *rnat0_kaddr, *rnat1_kaddr, shift;
316         long num_regs, nbits;
317         struct pt_regs *pt;
318         unsigned long cfm, *urbs_kargs;
319
320         pt = task_pt_regs(task);
321         kbsp = (unsigned long *) sw->ar_bspstore;
322         ubspstore = (unsigned long *) pt->ar_bspstore;
323
324         urbs_kargs = urbs_end;
325         if (in_syscall(pt)) {
326                 /*
327                  * If entered via syscall, don't allow user to set rnat bits
328                  * for syscall args.
329                  */
330                 cfm = pt->cr_ifs;
331                 urbs_kargs = ia64_rse_skip_regs(urbs_end, -(cfm & 0x7f));
332         }
333
334         if (urbs_kargs >= urnat_addr)
335                 nbits = 63;
336         else {
337                 if ((urnat_addr - 63) >= urbs_kargs)
338                         return;
339                 nbits = ia64_rse_num_regs(urnat_addr - 63, urbs_kargs);
340         }
341         mask = MASK(nbits);
342
343         /*
344          * First, figure out which bit number slot 0 in user-land maps
345          * to in the kernel rnat.  Do this by figuring out how many
346          * register slots we're beyond the user's backingstore and
347          * then computing the equivalent address in kernel space.
348          */
349         num_regs = ia64_rse_num_regs(ubspstore, urnat_addr + 1);
350         slot0_kaddr = ia64_rse_skip_regs(krbs, num_regs);
351         shift = ia64_rse_slot_num(slot0_kaddr);
352         rnat1_kaddr = ia64_rse_rnat_addr(slot0_kaddr);
353         rnat0_kaddr = rnat1_kaddr - 64;
354
355         if (ubspstore + 63 > urnat_addr) {
356                 /* some bits need to be place in pt->ar_rnat: */
357                 umask = MASK(ia64_rse_slot_num(ubspstore)) & mask;
358                 pt->ar_rnat = (pt->ar_rnat & ~umask) | (urnat & umask);
359                 mask &= ~umask;
360                 if (!mask)
361                         return;
362         }
363         /*
364          * Note: Section 11.1 of the EAS guarantees that bit 63 of an
365          * rnat slot is ignored. so we don't have to clear it here.
366          */
367         rnat0 = (urnat << shift);
368         m = mask << shift;
369         if (rnat0_kaddr >= kbsp)
370                 sw->ar_rnat = (sw->ar_rnat & ~m) | (rnat0 & m);
371         else if (rnat0_kaddr > krbs)
372                 *rnat0_kaddr = ((*rnat0_kaddr & ~m) | (rnat0 & m));
373
374         rnat1 = (urnat >> (63 - shift));
375         m = mask >> (63 - shift);
376         if (rnat1_kaddr >= kbsp)
377                 sw->ar_rnat = (sw->ar_rnat & ~m) | (rnat1 & m);
378         else if (rnat1_kaddr > krbs)
379                 *rnat1_kaddr = ((*rnat1_kaddr & ~m) | (rnat1 & m));
380 }
381
382 static inline int
383 on_kernel_rbs (unsigned long addr, unsigned long bspstore,
384                unsigned long urbs_end)
385 {
386         unsigned long *rnat_addr = ia64_rse_rnat_addr((unsigned long *)
387                                                       urbs_end);
388         return (addr >= bspstore && addr <= (unsigned long) rnat_addr);
389 }
390
391 /*
392  * Read a word from the user-level backing store of task CHILD.  ADDR
393  * is the user-level address to read the word from, VAL a pointer to
394  * the return value, and USER_BSP gives the end of the user-level
395  * backing store (i.e., it's the address that would be in ar.bsp after
396  * the user executed a "cover" instruction).
397  *
398  * This routine takes care of accessing the kernel register backing
399  * store for those registers that got spilled there.  It also takes
400  * care of calculating the appropriate RNaT collection words.
401  */
402 long
403 ia64_peek (struct task_struct *child, struct switch_stack *child_stack,
404            unsigned long user_rbs_end, unsigned long addr, long *val)
405 {
406         unsigned long *bspstore, *krbs, regnum, *laddr, *urbs_end, *rnat_addr;
407         struct pt_regs *child_regs;
408         size_t copied;
409         long ret;
410
411         urbs_end = (long *) user_rbs_end;
412         laddr = (unsigned long *) addr;
413         child_regs = task_pt_regs(child);
414         bspstore = (unsigned long *) child_regs->ar_bspstore;
415         krbs = (unsigned long *) child + IA64_RBS_OFFSET/8;
416         if (on_kernel_rbs(addr, (unsigned long) bspstore,
417                           (unsigned long) urbs_end))
418         {
419                 /*
420                  * Attempt to read the RBS in an area that's actually
421                  * on the kernel RBS => read the corresponding bits in
422                  * the kernel RBS.
423                  */
424                 rnat_addr = ia64_rse_rnat_addr(laddr);
425                 ret = get_rnat(child, child_stack, krbs, rnat_addr, urbs_end);
426
427                 if (laddr == rnat_addr) {
428                         /* return NaT collection word itself */
429                         *val = ret;
430                         return 0;
431                 }
432
433                 if (((1UL << ia64_rse_slot_num(laddr)) & ret) != 0) {
434                         /*
435                          * It is implementation dependent whether the
436                          * data portion of a NaT value gets saved on a
437                          * st8.spill or RSE spill (e.g., see EAS 2.6,
438                          * 4.4.4.6 Register Spill and Fill).  To get
439                          * consistent behavior across all possible
440                          * IA-64 implementations, we return zero in
441                          * this case.
442                          */
443                         *val = 0;
444                         return 0;
445                 }
446
447                 if (laddr < urbs_end) {
448                         /*
449                          * The desired word is on the kernel RBS and
450                          * is not a NaT.
451                          */
452                         regnum = ia64_rse_num_regs(bspstore, laddr);
453                         *val = *ia64_rse_skip_regs(krbs, regnum);
454                         return 0;
455                 }
456         }
457         copied = access_process_vm(child, addr, &ret, sizeof(ret), 0);
458         if (copied != sizeof(ret))
459                 return -EIO;
460         *val = ret;
461         return 0;
462 }
463
464 long
465 ia64_poke (struct task_struct *child, struct switch_stack *child_stack,
466            unsigned long user_rbs_end, unsigned long addr, long val)
467 {
468         unsigned long *bspstore, *krbs, regnum, *laddr;
469         unsigned long *urbs_end = (long *) user_rbs_end;
470         struct pt_regs *child_regs;
471
472         laddr = (unsigned long *) addr;
473         child_regs = task_pt_regs(child);
474         bspstore = (unsigned long *) child_regs->ar_bspstore;
475         krbs = (unsigned long *) child + IA64_RBS_OFFSET/8;
476         if (on_kernel_rbs(addr, (unsigned long) bspstore,
477                           (unsigned long) urbs_end))
478         {
479                 /*
480                  * Attempt to write the RBS in an area that's actually
481                  * on the kernel RBS => write the corresponding bits
482                  * in the kernel RBS.
483                  */
484                 if (ia64_rse_is_rnat_slot(laddr))
485                         put_rnat(child, child_stack, krbs, laddr, val,
486                                  urbs_end);
487                 else {
488                         if (laddr < urbs_end) {
489                                 regnum = ia64_rse_num_regs(bspstore, laddr);
490                                 *ia64_rse_skip_regs(krbs, regnum) = val;
491                         }
492                 }
493         } else if (access_process_vm(child, addr, &val, sizeof(val), 1)
494                    != sizeof(val))
495                 return -EIO;
496         return 0;
497 }
498
499 /*
500  * Calculate the address of the end of the user-level register backing
501  * store.  This is the address that would have been stored in ar.bsp
502  * if the user had executed a "cover" instruction right before
503  * entering the kernel.  If CFMP is not NULL, it is used to return the
504  * "current frame mask" that was active at the time the kernel was
505  * entered.
506  */
507 unsigned long
508 ia64_get_user_rbs_end (struct task_struct *child, struct pt_regs *pt,
509                        unsigned long *cfmp)
510 {
511         unsigned long *krbs, *bspstore, cfm = pt->cr_ifs;
512         long ndirty;
513
514         krbs = (unsigned long *) child + IA64_RBS_OFFSET/8;
515         bspstore = (unsigned long *) pt->ar_bspstore;
516         ndirty = ia64_rse_num_regs(krbs, krbs + (pt->loadrs >> 19));
517
518         if (in_syscall(pt))
519                 ndirty += (cfm & 0x7f);
520         else
521                 cfm &= ~(1UL << 63);    /* clear valid bit */
522
523         if (cfmp)
524                 *cfmp = cfm;
525         return (unsigned long) ia64_rse_skip_regs(bspstore, ndirty);
526 }
527
528 /*
529  * Synchronize (i.e, write) the RSE backing store living in kernel
530  * space to the VM of the CHILD task.  SW and PT are the pointers to
531  * the switch_stack and pt_regs structures, respectively.
532  * USER_RBS_END is the user-level address at which the backing store
533  * ends.
534  */
535 long
536 ia64_sync_user_rbs (struct task_struct *child, struct switch_stack *sw,
537                     unsigned long user_rbs_start, unsigned long user_rbs_end)
538 {
539         unsigned long addr, val;
540         long ret;
541
542         /* now copy word for word from kernel rbs to user rbs: */
543         for (addr = user_rbs_start; addr < user_rbs_end; addr += 8) {
544                 ret = ia64_peek(child, sw, user_rbs_end, addr, &val);
545                 if (ret < 0)
546                         return ret;
547                 if (access_process_vm(child, addr, &val, sizeof(val), 1)
548                     != sizeof(val))
549                         return -EIO;
550         }
551         return 0;
552 }
553
554 static long
555 ia64_sync_kernel_rbs (struct task_struct *child, struct switch_stack *sw,
556                 unsigned long user_rbs_start, unsigned long user_rbs_end)
557 {
558         unsigned long addr, val;
559         long ret;
560
561         /* now copy word for word from user rbs to kernel rbs: */
562         for (addr = user_rbs_start; addr < user_rbs_end; addr += 8) {
563                 if (access_process_vm(child, addr, &val, sizeof(val), 0)
564                                 != sizeof(val))
565                         return -EIO;
566
567                 ret = ia64_poke(child, sw, user_rbs_end, addr, val);
568                 if (ret < 0)
569                         return ret;
570         }
571         return 0;
572 }
573
574 typedef long (*syncfunc_t)(struct task_struct *, struct switch_stack *,
575                             unsigned long, unsigned long);
576
577 static void do_sync_rbs(struct unw_frame_info *info, void *arg)
578 {
579         struct pt_regs *pt;
580         unsigned long urbs_end;
581         syncfunc_t fn = arg;
582
583         if (unw_unwind_to_user(info) < 0)
584                 return;
585         pt = task_pt_regs(info->task);
586         urbs_end = ia64_get_user_rbs_end(info->task, pt, NULL);
587
588         fn(info->task, info->sw, pt->ar_bspstore, urbs_end);
589 }
590
591 /*
592  * when a thread is stopped (ptraced), debugger might change thread's user
593  * stack (change memory directly), and we must avoid the RSE stored in kernel
594  * to override user stack (user space's RSE is newer than kernel's in the
595  * case). To workaround the issue, we copy kernel RSE to user RSE before the
596  * task is stopped, so user RSE has updated data.  we then copy user RSE to
597  * kernel after the task is resummed from traced stop and kernel will use the
598  * newer RSE to return to user. TIF_RESTORE_RSE is the flag to indicate we need
599  * synchronize user RSE to kernel.
600  */
601 void ia64_ptrace_stop(void)
602 {
603         if (test_and_set_tsk_thread_flag(current, TIF_RESTORE_RSE))
604                 return;
605         set_notify_resume(current);
606         unw_init_running(do_sync_rbs, ia64_sync_user_rbs);
607 }
608
609 /*
610  * This is called to read back the register backing store.
611  */
612 void ia64_sync_krbs(void)
613 {
614         clear_tsk_thread_flag(current, TIF_RESTORE_RSE);
615
616         unw_init_running(do_sync_rbs, ia64_sync_kernel_rbs);
617 }
618
619 /*
620  * After PTRACE_ATTACH, a thread's register backing store area in user
621  * space is assumed to contain correct data whenever the thread is
622  * stopped.  arch_ptrace_stop takes care of this on tracing stops.
623  * But if the child was already stopped for job control when we attach
624  * to it, then it might not ever get into ptrace_stop by the time we
625  * want to examine the user memory containing the RBS.
626  */
627 void
628 ptrace_attach_sync_user_rbs (struct task_struct *child)
629 {
630         int stopped = 0;
631         struct unw_frame_info info;
632
633         /*
634          * If the child is in TASK_STOPPED, we need to change that to
635          * TASK_TRACED momentarily while we operate on it.  This ensures
636          * that the child won't be woken up and return to user mode while
637          * we are doing the sync.  (It can only be woken up for SIGKILL.)
638          */
639
640         read_lock(&tasklist_lock);
641         if (child->sighand) {
642                 spin_lock_irq(&child->sighand->siglock);
643                 if (child->state == TASK_STOPPED &&
644                     !test_and_set_tsk_thread_flag(child, TIF_RESTORE_RSE)) {
645                         set_notify_resume(child);
646
647                         child->state = TASK_TRACED;
648                         stopped = 1;
649                 }
650                 spin_unlock_irq(&child->sighand->siglock);
651         }
652         read_unlock(&tasklist_lock);
653
654         if (!stopped)
655                 return;
656
657         unw_init_from_blocked_task(&info, child);
658         do_sync_rbs(&info, ia64_sync_user_rbs);
659
660         /*
661          * Now move the child back into TASK_STOPPED if it should be in a
662          * job control stop, so that SIGCONT can be used to wake it up.
663          */
664         read_lock(&tasklist_lock);
665         if (child->sighand) {
666                 spin_lock_irq(&child->sighand->siglock);
667                 if (child->state == TASK_TRACED &&
668                     (child->signal->flags & SIGNAL_STOP_STOPPED)) {
669                         child->state = TASK_STOPPED;
670                 }
671                 spin_unlock_irq(&child->sighand->siglock);
672         }
673         read_unlock(&tasklist_lock);
674 }
675
676 static inline int
677 thread_matches (struct task_struct *thread, unsigned long addr)
678 {
679         unsigned long thread_rbs_end;
680         struct pt_regs *thread_regs;
681
682         if (ptrace_check_attach(thread, 0) < 0)
683                 /*
684                  * If the thread is not in an attachable state, we'll
685                  * ignore it.  The net effect is that if ADDR happens
686                  * to overlap with the portion of the thread's
687                  * register backing store that is currently residing
688                  * on the thread's kernel stack, then ptrace() may end
689                  * up accessing a stale value.  But if the thread
690                  * isn't stopped, that's a problem anyhow, so we're
691                  * doing as well as we can...
692                  */
693                 return 0;
694
695         thread_regs = task_pt_regs(thread);
696         thread_rbs_end = ia64_get_user_rbs_end(thread, thread_regs, NULL);
697         if (!on_kernel_rbs(addr, thread_regs->ar_bspstore, thread_rbs_end))
698                 return 0;
699
700         return 1;       /* looks like we've got a winner */
701 }
702
703 /*
704  * Write f32-f127 back to task->thread.fph if it has been modified.
705  */
706 inline void
707 ia64_flush_fph (struct task_struct *task)
708 {
709         struct ia64_psr *psr = ia64_psr(task_pt_regs(task));
710
711         /*
712          * Prevent migrating this task while
713          * we're fiddling with the FPU state
714          */
715         preempt_disable();
716         if (ia64_is_local_fpu_owner(task) && psr->mfh) {
717                 psr->mfh = 0;
718                 task->thread.flags |= IA64_THREAD_FPH_VALID;
719                 ia64_save_fpu(&task->thread.fph[0]);
720         }
721         preempt_enable();
722 }
723
724 /*
725  * Sync the fph state of the task so that it can be manipulated
726  * through thread.fph.  If necessary, f32-f127 are written back to
727  * thread.fph or, if the fph state hasn't been used before, thread.fph
728  * is cleared to zeroes.  Also, access to f32-f127 is disabled to
729  * ensure that the task picks up the state from thread.fph when it
730  * executes again.
731  */
732 void
733 ia64_sync_fph (struct task_struct *task)
734 {
735         struct ia64_psr *psr = ia64_psr(task_pt_regs(task));
736
737         ia64_flush_fph(task);
738         if (!(task->thread.flags & IA64_THREAD_FPH_VALID)) {
739                 task->thread.flags |= IA64_THREAD_FPH_VALID;
740                 memset(&task->thread.fph, 0, sizeof(task->thread.fph));
741         }
742         ia64_drop_fpu(task);
743         psr->dfh = 1;
744 }
745
746 /*
747  * Change the machine-state of CHILD such that it will return via the normal
748  * kernel exit-path, rather than the syscall-exit path.
749  */
750 static void
751 convert_to_non_syscall (struct task_struct *child, struct pt_regs  *pt,
752                         unsigned long cfm)
753 {
754         struct unw_frame_info info, prev_info;
755         unsigned long ip, sp, pr;
756
757         unw_init_from_blocked_task(&info, child);
758         while (1) {
759                 prev_info = info;
760                 if (unw_unwind(&info) < 0)
761                         return;
762
763                 unw_get_sp(&info, &sp);
764                 if ((long)((unsigned long)child + IA64_STK_OFFSET - sp)
765                     < IA64_PT_REGS_SIZE) {
766                         dprintk("ptrace.%s: ran off the top of the kernel "
767                                 "stack\n", __func__);
768                         return;
769                 }
770                 if (unw_get_pr (&prev_info, &pr) < 0) {
771                         unw_get_rp(&prev_info, &ip);
772                         dprintk("ptrace.%s: failed to read "
773                                 "predicate register (ip=0x%lx)\n",
774                                 __func__, ip);
775                         return;
776                 }
777                 if (unw_is_intr_frame(&info)
778                     && (pr & (1UL << PRED_USER_STACK)))
779                         break;
780         }
781
782         /*
783          * Note: at the time of this call, the target task is blocked
784          * in notify_resume_user() and by clearling PRED_LEAVE_SYSCALL
785          * (aka, "pLvSys") we redirect execution from
786          * .work_pending_syscall_end to .work_processed_kernel.
787          */
788         unw_get_pr(&prev_info, &pr);
789         pr &= ~((1UL << PRED_SYSCALL) | (1UL << PRED_LEAVE_SYSCALL));
790         pr |=  (1UL << PRED_NON_SYSCALL);
791         unw_set_pr(&prev_info, pr);
792
793         pt->cr_ifs = (1UL << 63) | cfm;
794         /*
795          * Clear the memory that is NOT written on syscall-entry to
796          * ensure we do not leak kernel-state to user when execution
797          * resumes.
798          */
799         pt->r2 = 0;
800         pt->r3 = 0;
801         pt->r14 = 0;
802         memset(&pt->r16, 0, 16*8);      /* clear r16-r31 */
803         memset(&pt->f6, 0, 6*16);       /* clear f6-f11 */
804         pt->b7 = 0;
805         pt->ar_ccv = 0;
806         pt->ar_csd = 0;
807         pt->ar_ssd = 0;
808 }
809
810 static int
811 access_nat_bits (struct task_struct *child, struct pt_regs *pt,
812                  struct unw_frame_info *info,
813                  unsigned long *data, int write_access)
814 {
815         unsigned long regnum, nat_bits, scratch_unat, dummy = 0;
816         char nat = 0;
817
818         if (write_access) {
819                 nat_bits = *data;
820                 scratch_unat = ia64_put_scratch_nat_bits(pt, nat_bits);
821                 if (unw_set_ar(info, UNW_AR_UNAT, scratch_unat) < 0) {
822                         dprintk("ptrace: failed to set ar.unat\n");
823                         return -1;
824                 }
825                 for (regnum = 4; regnum <= 7; ++regnum) {
826                         unw_get_gr(info, regnum, &dummy, &nat);
827                         unw_set_gr(info, regnum, dummy,
828                                    (nat_bits >> regnum) & 1);
829                 }
830         } else {
831                 if (unw_get_ar(info, UNW_AR_UNAT, &scratch_unat) < 0) {
832                         dprintk("ptrace: failed to read ar.unat\n");
833                         return -1;
834                 }
835                 nat_bits = ia64_get_scratch_nat_bits(pt, scratch_unat);
836                 for (regnum = 4; regnum <= 7; ++regnum) {
837                         unw_get_gr(info, regnum, &dummy, &nat);
838                         nat_bits |= (nat != 0) << regnum;
839                 }
840                 *data = nat_bits;
841         }
842         return 0;
843 }
844
845 static int
846 access_uarea (struct task_struct *child, unsigned long addr,
847               unsigned long *data, int write_access);
848
849 static long
850 ptrace_getregs (struct task_struct *child, struct pt_all_user_regs __user *ppr)
851 {
852         unsigned long psr, ec, lc, rnat, bsp, cfm, nat_bits, val;
853         struct unw_frame_info info;
854         struct ia64_fpreg fpval;
855         struct switch_stack *sw;
856         struct pt_regs *pt;
857         long ret, retval = 0;
858         char nat = 0;
859         int i;
860
861         if (!access_ok(VERIFY_WRITE, ppr, sizeof(struct pt_all_user_regs)))
862                 return -EIO;
863
864         pt = task_pt_regs(child);
865         sw = (struct switch_stack *) (child->thread.ksp + 16);
866         unw_init_from_blocked_task(&info, child);
867         if (unw_unwind_to_user(&info) < 0) {
868                 return -EIO;
869         }
870
871         if (((unsigned long) ppr & 0x7) != 0) {
872                 dprintk("ptrace:unaligned register address %p\n", ppr);
873                 return -EIO;
874         }
875
876         if (access_uarea(child, PT_CR_IPSR, &psr, 0) < 0
877             || access_uarea(child, PT_AR_EC, &ec, 0) < 0
878             || access_uarea(child, PT_AR_LC, &lc, 0) < 0
879             || access_uarea(child, PT_AR_RNAT, &rnat, 0) < 0
880             || access_uarea(child, PT_AR_BSP, &bsp, 0) < 0
881             || access_uarea(child, PT_CFM, &cfm, 0)
882             || access_uarea(child, PT_NAT_BITS, &nat_bits, 0))
883                 return -EIO;
884
885         /* control regs */
886
887         retval |= __put_user(pt->cr_iip, &ppr->cr_iip);
888         retval |= __put_user(psr, &ppr->cr_ipsr);
889
890         /* app regs */
891
892         retval |= __put_user(pt->ar_pfs, &ppr->ar[PT_AUR_PFS]);
893         retval |= __put_user(pt->ar_rsc, &ppr->ar[PT_AUR_RSC]);
894         retval |= __put_user(pt->ar_bspstore, &ppr->ar[PT_AUR_BSPSTORE]);
895         retval |= __put_user(pt->ar_unat, &ppr->ar[PT_AUR_UNAT]);
896         retval |= __put_user(pt->ar_ccv, &ppr->ar[PT_AUR_CCV]);
897         retval |= __put_user(pt->ar_fpsr, &ppr->ar[PT_AUR_FPSR]);
898
899         retval |= __put_user(ec, &ppr->ar[PT_AUR_EC]);
900         retval |= __put_user(lc, &ppr->ar[PT_AUR_LC]);
901         retval |= __put_user(rnat, &ppr->ar[PT_AUR_RNAT]);
902         retval |= __put_user(bsp, &ppr->ar[PT_AUR_BSP]);
903         retval |= __put_user(cfm, &ppr->cfm);
904
905         /* gr1-gr3 */
906
907         retval |= __copy_to_user(&ppr->gr[1], &pt->r1, sizeof(long));
908         retval |= __copy_to_user(&ppr->gr[2], &pt->r2, sizeof(long) *2);
909
910         /* gr4-gr7 */
911
912         for (i = 4; i < 8; i++) {
913                 if (unw_access_gr(&info, i, &val, &nat, 0) < 0)
914                         return -EIO;
915                 retval |= __put_user(val, &ppr->gr[i]);
916         }
917
918         /* gr8-gr11 */
919
920         retval |= __copy_to_user(&ppr->gr[8], &pt->r8, sizeof(long) * 4);
921
922         /* gr12-gr15 */
923
924         retval |= __copy_to_user(&ppr->gr[12], &pt->r12, sizeof(long) * 2);
925         retval |= __copy_to_user(&ppr->gr[14], &pt->r14, sizeof(long));
926         retval |= __copy_to_user(&ppr->gr[15], &pt->r15, sizeof(long));
927
928         /* gr16-gr31 */
929
930         retval |= __copy_to_user(&ppr->gr[16], &pt->r16, sizeof(long) * 16);
931
932         /* b0 */
933
934         retval |= __put_user(pt->b0, &ppr->br[0]);
935
936         /* b1-b5 */
937
938         for (i = 1; i < 6; i++) {
939                 if (unw_access_br(&info, i, &val, 0) < 0)
940                         return -EIO;
941                 __put_user(val, &ppr->br[i]);
942         }
943
944         /* b6-b7 */
945
946         retval |= __put_user(pt->b6, &ppr->br[6]);
947         retval |= __put_user(pt->b7, &ppr->br[7]);
948
949         /* fr2-fr5 */
950
951         for (i = 2; i < 6; i++) {
952                 if (unw_get_fr(&info, i, &fpval) < 0)
953                         return -EIO;
954                 retval |= __copy_to_user(&ppr->fr[i], &fpval, sizeof (fpval));
955         }
956
957         /* fr6-fr11 */
958
959         retval |= __copy_to_user(&ppr->fr[6], &pt->f6,
960                                  sizeof(struct ia64_fpreg) * 6);
961
962         /* fp scratch regs(12-15) */
963
964         retval |= __copy_to_user(&ppr->fr[12], &sw->f12,
965                                  sizeof(struct ia64_fpreg) * 4);
966
967         /* fr16-fr31 */
968
969         for (i = 16; i < 32; i++) {
970                 if (unw_get_fr(&info, i, &fpval) < 0)
971                         return -EIO;
972                 retval |= __copy_to_user(&ppr->fr[i], &fpval, sizeof (fpval));
973         }
974
975         /* fph */
976
977         ia64_flush_fph(child);
978         retval |= __copy_to_user(&ppr->fr[32], &child->thread.fph,
979                                  sizeof(ppr->fr[32]) * 96);
980
981         /*  preds */
982
983         retval |= __put_user(pt->pr, &ppr->pr);
984
985         /* nat bits */
986
987         retval |= __put_user(nat_bits, &ppr->nat);
988
989         ret = retval ? -EIO : 0;
990         return ret;
991 }
992
993 static long
994 ptrace_setregs (struct task_struct *child, struct pt_all_user_regs __user *ppr)
995 {
996         unsigned long psr, rsc, ec, lc, rnat, bsp, cfm, nat_bits, val = 0;
997         struct unw_frame_info info;
998         struct switch_stack *sw;
999         struct ia64_fpreg fpval;
1000         struct pt_regs *pt;
1001         long ret, retval = 0;
1002         int i;
1003
1004         memset(&fpval, 0, sizeof(fpval));
1005
1006         if (!access_ok(VERIFY_READ, ppr, sizeof(struct pt_all_user_regs)))
1007                 return -EIO;
1008
1009         pt = task_pt_regs(child);
1010         sw = (struct switch_stack *) (child->thread.ksp + 16);
1011         unw_init_from_blocked_task(&info, child);
1012         if (unw_unwind_to_user(&info) < 0) {
1013                 return -EIO;
1014         }
1015
1016         if (((unsigned long) ppr & 0x7) != 0) {
1017                 dprintk("ptrace:unaligned register address %p\n", ppr);
1018                 return -EIO;
1019         }
1020
1021         /* control regs */
1022
1023         retval |= __get_user(pt->cr_iip, &ppr->cr_iip);
1024         retval |= __get_user(psr, &ppr->cr_ipsr);
1025
1026         /* app regs */
1027
1028         retval |= __get_user(pt->ar_pfs, &ppr->ar[PT_AUR_PFS]);
1029         retval |= __get_user(rsc, &ppr->ar[PT_AUR_RSC]);
1030         retval |= __get_user(pt->ar_bspstore, &ppr->ar[PT_AUR_BSPSTORE]);
1031         retval |= __get_user(pt->ar_unat, &ppr->ar[PT_AUR_UNAT]);
1032         retval |= __get_user(pt->ar_ccv, &ppr->ar[PT_AUR_CCV]);
1033         retval |= __get_user(pt->ar_fpsr, &ppr->ar[PT_AUR_FPSR]);
1034
1035         retval |= __get_user(ec, &ppr->ar[PT_AUR_EC]);
1036         retval |= __get_user(lc, &ppr->ar[PT_AUR_LC]);
1037         retval |= __get_user(rnat, &ppr->ar[PT_AUR_RNAT]);
1038         retval |= __get_user(bsp, &ppr->ar[PT_AUR_BSP]);
1039         retval |= __get_user(cfm, &ppr->cfm);
1040
1041         /* gr1-gr3 */
1042
1043         retval |= __copy_from_user(&pt->r1, &ppr->gr[1], sizeof(long));
1044         retval |= __copy_from_user(&pt->r2, &ppr->gr[2], sizeof(long) * 2);
1045
1046         /* gr4-gr7 */
1047
1048         for (i = 4; i < 8; i++) {
1049                 retval |= __get_user(val, &ppr->gr[i]);
1050                 /* NaT bit will be set via PT_NAT_BITS: */
1051                 if (unw_set_gr(&info, i, val, 0) < 0)
1052                         return -EIO;
1053         }
1054
1055         /* gr8-gr11 */
1056
1057         retval |= __copy_from_user(&pt->r8, &ppr->gr[8], sizeof(long) * 4);
1058
1059         /* gr12-gr15 */
1060
1061         retval |= __copy_from_user(&pt->r12, &ppr->gr[12], sizeof(long) * 2);
1062         retval |= __copy_from_user(&pt->r14, &ppr->gr[14], sizeof(long));
1063         retval |= __copy_from_user(&pt->r15, &ppr->gr[15], sizeof(long));
1064
1065         /* gr16-gr31 */
1066
1067         retval |= __copy_from_user(&pt->r16, &ppr->gr[16], sizeof(long) * 16);
1068
1069         /* b0 */
1070
1071         retval |= __get_user(pt->b0, &ppr->br[0]);
1072
1073         /* b1-b5 */
1074
1075         for (i = 1; i < 6; i++) {
1076                 retval |= __get_user(val, &ppr->br[i]);
1077                 unw_set_br(&info, i, val);
1078         }
1079
1080         /* b6-b7 */
1081
1082         retval |= __get_user(pt->b6, &ppr->br[6]);
1083         retval |= __get_user(pt->b7, &ppr->br[7]);
1084
1085         /* fr2-fr5 */
1086
1087         for (i = 2; i < 6; i++) {
1088                 retval |= __copy_from_user(&fpval, &ppr->fr[i], sizeof(fpval));
1089                 if (unw_set_fr(&info, i, fpval) < 0)
1090                         return -EIO;
1091         }
1092
1093         /* fr6-fr11 */
1094
1095         retval |= __copy_from_user(&pt->f6, &ppr->fr[6],
1096                                    sizeof(ppr->fr[6]) * 6);
1097
1098         /* fp scratch regs(12-15) */
1099
1100         retval |= __copy_from_user(&sw->f12, &ppr->fr[12],
1101                                    sizeof(ppr->fr[12]) * 4);
1102
1103         /* fr16-fr31 */
1104
1105         for (i = 16; i < 32; i++) {
1106                 retval |= __copy_from_user(&fpval, &ppr->fr[i],
1107                                            sizeof(fpval));
1108                 if (unw_set_fr(&info, i, fpval) < 0)
1109                         return -EIO;
1110         }
1111
1112         /* fph */
1113
1114         ia64_sync_fph(child);
1115         retval |= __copy_from_user(&child->thread.fph, &ppr->fr[32],
1116                                    sizeof(ppr->fr[32]) * 96);
1117
1118         /* preds */
1119
1120         retval |= __get_user(pt->pr, &ppr->pr);
1121
1122         /* nat bits */
1123
1124         retval |= __get_user(nat_bits, &ppr->nat);
1125
1126         retval |= access_uarea(child, PT_CR_IPSR, &psr, 1);
1127         retval |= access_uarea(child, PT_AR_RSC, &rsc, 1);
1128         retval |= access_uarea(child, PT_AR_EC, &ec, 1);
1129         retval |= access_uarea(child, PT_AR_LC, &lc, 1);
1130         retval |= access_uarea(child, PT_AR_RNAT, &rnat, 1);
1131         retval |= access_uarea(child, PT_AR_BSP, &bsp, 1);
1132         retval |= access_uarea(child, PT_CFM, &cfm, 1);
1133         retval |= access_uarea(child, PT_NAT_BITS, &nat_bits, 1);
1134
1135         ret = retval ? -EIO : 0;
1136         return ret;
1137 }
1138
1139 void
1140 user_enable_single_step (struct task_struct *child)
1141 {
1142         struct ia64_psr *child_psr = ia64_psr(task_pt_regs(child));
1143
1144         set_tsk_thread_flag(child, TIF_SINGLESTEP);
1145         child_psr->ss = 1;
1146 }
1147
1148 void
1149 user_enable_block_step (struct task_struct *child)
1150 {
1151         struct ia64_psr *child_psr = ia64_psr(task_pt_regs(child));
1152
1153         set_tsk_thread_flag(child, TIF_SINGLESTEP);
1154         child_psr->tb = 1;
1155 }
1156
1157 void
1158 user_disable_single_step (struct task_struct *child)
1159 {
1160         struct ia64_psr *child_psr = ia64_psr(task_pt_regs(child));
1161
1162         /* make sure the single step/taken-branch trap bits are not set: */
1163         clear_tsk_thread_flag(child, TIF_SINGLESTEP);
1164         child_psr->ss = 0;
1165         child_psr->tb = 0;
1166 }
1167
1168 /*
1169  * Called by kernel/ptrace.c when detaching..
1170  *
1171  * Make sure the single step bit is not set.
1172  */
1173 void
1174 ptrace_disable (struct task_struct *child)
1175 {
1176         user_disable_single_step(child);
1177 }
1178
1179 long
1180 arch_ptrace (struct task_struct *child, long request,
1181              unsigned long addr, unsigned long data)
1182 {
1183         switch (request) {
1184         case PTRACE_PEEKTEXT:
1185         case PTRACE_PEEKDATA:
1186                 /* read word at location addr */
1187                 if (access_process_vm(child, addr, &data, sizeof(data), 0)
1188                     != sizeof(data))
1189                         return -EIO;
1190                 /* ensure return value is not mistaken for error code */
1191                 force_successful_syscall_return();
1192                 return data;
1193
1194         /* PTRACE_POKETEXT and PTRACE_POKEDATA is handled
1195          * by the generic ptrace_request().
1196          */
1197
1198         case PTRACE_PEEKUSR:
1199                 /* read the word at addr in the USER area */
1200                 if (access_uarea(child, addr, &data, 0) < 0)
1201                         return -EIO;
1202                 /* ensure return value is not mistaken for error code */
1203                 force_successful_syscall_return();
1204                 return data;
1205
1206         case PTRACE_POKEUSR:
1207                 /* write the word at addr in the USER area */
1208                 if (access_uarea(child, addr, &data, 1) < 0)
1209                         return -EIO;
1210                 return 0;
1211
1212         case PTRACE_OLD_GETSIGINFO:
1213                 /* for backwards-compatibility */
1214                 return ptrace_request(child, PTRACE_GETSIGINFO, addr, data);
1215
1216         case PTRACE_OLD_SETSIGINFO:
1217                 /* for backwards-compatibility */
1218                 return ptrace_request(child, PTRACE_SETSIGINFO, addr, data);
1219
1220         case PTRACE_GETREGS:
1221                 return ptrace_getregs(child,
1222                                       (struct pt_all_user_regs __user *) data);
1223
1224         case PTRACE_SETREGS:
1225                 return ptrace_setregs(child,
1226                                       (struct pt_all_user_regs __user *) data);
1227
1228         default:
1229                 return ptrace_request(child, request, addr, data);
1230         }
1231 }
1232
1233
1234 /* "asmlinkage" so the input arguments are preserved... */
1235
1236 asmlinkage long
1237 syscall_trace_enter (long arg0, long arg1, long arg2, long arg3,
1238                      long arg4, long arg5, long arg6, long arg7,
1239                      struct pt_regs regs)
1240 {
1241         if (test_thread_flag(TIF_SYSCALL_TRACE))
1242                 if (tracehook_report_syscall_entry(&regs))
1243                         return -ENOSYS;
1244
1245         /* copy user rbs to kernel rbs */
1246         if (test_thread_flag(TIF_RESTORE_RSE))
1247                 ia64_sync_krbs();
1248
1249         if (unlikely(current->audit_context)) {
1250                 long syscall;
1251                 int arch;
1252
1253                 syscall = regs.r15;
1254                 arch = AUDIT_ARCH_IA64;
1255
1256                 audit_syscall_entry(arch, syscall, arg0, arg1, arg2, arg3);
1257         }
1258
1259         return 0;
1260 }
1261
1262 /* "asmlinkage" so the input arguments are preserved... */
1263
1264 asmlinkage void
1265 syscall_trace_leave (long arg0, long arg1, long arg2, long arg3,
1266                      long arg4, long arg5, long arg6, long arg7,
1267                      struct pt_regs regs)
1268 {
1269         int step;
1270
1271         if (unlikely(current->audit_context)) {
1272                 int success = AUDITSC_RESULT(regs.r10);
1273                 long result = regs.r8;
1274
1275                 if (success != AUDITSC_SUCCESS)
1276                         result = -result;
1277                 audit_syscall_exit(success, result);
1278         }
1279
1280         step = test_thread_flag(TIF_SINGLESTEP);
1281         if (step || test_thread_flag(TIF_SYSCALL_TRACE))
1282                 tracehook_report_syscall_exit(&regs, step);
1283
1284         /* copy user rbs to kernel rbs */
1285         if (test_thread_flag(TIF_RESTORE_RSE))
1286                 ia64_sync_krbs();
1287 }
1288
1289 /* Utrace implementation starts here */
1290 struct regset_get {
1291         void *kbuf;
1292         void __user *ubuf;
1293 };
1294
1295 struct regset_set {
1296         const void *kbuf;
1297         const void __user *ubuf;
1298 };
1299
1300 struct regset_getset {
1301         struct task_struct *target;
1302         const struct user_regset *regset;
1303         union {
1304                 struct regset_get get;
1305                 struct regset_set set;
1306         } u;
1307         unsigned int pos;
1308         unsigned int count;
1309         int ret;
1310 };
1311
1312 static int
1313 access_elf_gpreg(struct task_struct *target, struct unw_frame_info *info,
1314                 unsigned long addr, unsigned long *data, int write_access)
1315 {
1316         struct pt_regs *pt;
1317         unsigned long *ptr = NULL;
1318         int ret;
1319         char nat = 0;
1320
1321         pt = task_pt_regs(target);
1322         switch (addr) {
1323         case ELF_GR_OFFSET(1):
1324                 ptr = &pt->r1;
1325                 break;
1326         case ELF_GR_OFFSET(2):
1327         case ELF_GR_OFFSET(3):
1328                 ptr = (void *)&pt->r2 + (addr - ELF_GR_OFFSET(2));
1329                 break;
1330         case ELF_GR_OFFSET(4) ... ELF_GR_OFFSET(7):
1331                 if (write_access) {
1332                         /* read NaT bit first: */
1333                         unsigned long dummy;
1334
1335                         ret = unw_get_gr(info, addr/8, &dummy, &nat);
1336                         if (ret < 0)
1337                                 return ret;
1338                 }
1339                 return unw_access_gr(info, addr/8, data, &nat, write_access);
1340         case ELF_GR_OFFSET(8) ... ELF_GR_OFFSET(11):
1341                 ptr = (void *)&pt->r8 + addr - ELF_GR_OFFSET(8);
1342                 break;
1343         case ELF_GR_OFFSET(12):
1344         case ELF_GR_OFFSET(13):
1345                 ptr = (void *)&pt->r12 + addr - ELF_GR_OFFSET(12);
1346                 break;
1347         case ELF_GR_OFFSET(14):
1348                 ptr = &pt->r14;
1349                 break;
1350         case ELF_GR_OFFSET(15):
1351                 ptr = &pt->r15;
1352         }
1353         if (write_access)
1354                 *ptr = *data;
1355         else
1356                 *data = *ptr;
1357         return 0;
1358 }
1359
1360 static int
1361 access_elf_breg(struct task_struct *target, struct unw_frame_info *info,
1362                 unsigned long addr, unsigned long *data, int write_access)
1363 {
1364         struct pt_regs *pt;
1365         unsigned long *ptr = NULL;
1366
1367         pt = task_pt_regs(target);
1368         switch (addr) {
1369         case ELF_BR_OFFSET(0):
1370                 ptr = &pt->b0;
1371                 break;
1372         case ELF_BR_OFFSET(1) ... ELF_BR_OFFSET(5):
1373                 return unw_access_br(info, (addr - ELF_BR_OFFSET(0))/8,
1374                                      data, write_access);
1375         case ELF_BR_OFFSET(6):
1376                 ptr = &pt->b6;
1377                 break;
1378         case ELF_BR_OFFSET(7):
1379                 ptr = &pt->b7;
1380         }
1381         if (write_access)
1382                 *ptr = *data;
1383         else
1384                 *data = *ptr;
1385         return 0;
1386 }
1387
1388 static int
1389 access_elf_areg(struct task_struct *target, struct unw_frame_info *info,
1390                 unsigned long addr, unsigned long *data, int write_access)
1391 {
1392         struct pt_regs *pt;
1393         unsigned long cfm, urbs_end;
1394         unsigned long *ptr = NULL;
1395
1396         pt = task_pt_regs(target);
1397         if (addr >= ELF_AR_RSC_OFFSET && addr <= ELF_AR_SSD_OFFSET) {
1398                 switch (addr) {
1399                 case ELF_AR_RSC_OFFSET:
1400                         /* force PL3 */
1401                         if (write_access)
1402                                 pt->ar_rsc = *data | (3 << 2);
1403                         else
1404                                 *data = pt->ar_rsc;
1405                         return 0;
1406                 case ELF_AR_BSP_OFFSET:
1407                         /*
1408                          * By convention, we use PT_AR_BSP to refer to
1409                          * the end of the user-level backing store.
1410                          * Use ia64_rse_skip_regs(PT_AR_BSP, -CFM.sof)
1411                          * to get the real value of ar.bsp at the time
1412                          * the kernel was entered.
1413                          *
1414                          * Furthermore, when changing the contents of
1415                          * PT_AR_BSP (or PT_CFM) while the task is
1416                          * blocked in a system call, convert the state
1417                          * so that the non-system-call exit
1418                          * path is used.  This ensures that the proper
1419                          * state will be picked up when resuming
1420                          * execution.  However, it *also* means that
1421                          * once we write PT_AR_BSP/PT_CFM, it won't be
1422                          * possible to modify the syscall arguments of
1423                          * the pending system call any longer.  This
1424                          * shouldn't be an issue because modifying
1425                          * PT_AR_BSP/PT_CFM generally implies that
1426                          * we're either abandoning the pending system
1427                          * call or that we defer it's re-execution
1428                          * (e.g., due to GDB doing an inferior
1429                          * function call).
1430                          */
1431                         urbs_end = ia64_get_user_rbs_end(target, pt, &cfm);
1432                         if (write_access) {
1433                                 if (*data != urbs_end) {
1434                                         if (in_syscall(pt))
1435                                                 convert_to_non_syscall(target,
1436                                                                        pt,
1437                                                                        cfm);
1438                                         /*
1439                                          * Simulate user-level write
1440                                          * of ar.bsp:
1441                                          */
1442                                         pt->loadrs = 0;
1443                                         pt->ar_bspstore = *data;
1444                                 }
1445                         } else
1446                                 *data = urbs_end;
1447                         return 0;
1448                 case ELF_AR_BSPSTORE_OFFSET:
1449                         ptr = &pt->ar_bspstore;
1450                         break;
1451                 case ELF_AR_RNAT_OFFSET:
1452                         ptr = &pt->ar_rnat;
1453                         break;
1454                 case ELF_AR_CCV_OFFSET:
1455                         ptr = &pt->ar_ccv;
1456                         break;
1457                 case ELF_AR_UNAT_OFFSET:
1458                         ptr = &pt->ar_unat;
1459                         break;
1460                 case ELF_AR_FPSR_OFFSET:
1461                         ptr = &pt->ar_fpsr;
1462                         break;
1463                 case ELF_AR_PFS_OFFSET:
1464                         ptr = &pt->ar_pfs;
1465                         break;
1466                 case ELF_AR_LC_OFFSET:
1467                         return unw_access_ar(info, UNW_AR_LC, data,
1468                                              write_access);
1469                 case ELF_AR_EC_OFFSET:
1470                         return unw_access_ar(info, UNW_AR_EC, data,
1471                                              write_access);
1472                 case ELF_AR_CSD_OFFSET:
1473                         ptr = &pt->ar_csd;
1474                         break;
1475                 case ELF_AR_SSD_OFFSET:
1476                         ptr = &pt->ar_ssd;
1477                 }
1478         } else if (addr >= ELF_CR_IIP_OFFSET && addr <= ELF_CR_IPSR_OFFSET) {
1479                 switch (addr) {
1480                 case ELF_CR_IIP_OFFSET:
1481                         ptr = &pt->cr_iip;
1482                         break;
1483                 case ELF_CFM_OFFSET:
1484                         urbs_end = ia64_get_user_rbs_end(target, pt, &cfm);
1485                         if (write_access) {
1486                                 if (((cfm ^ *data) & PFM_MASK) != 0) {
1487                                         if (in_syscall(pt))
1488                                                 convert_to_non_syscall(target,
1489                                                                        pt,
1490                                                                        cfm);
1491                                         pt->cr_ifs = ((pt->cr_ifs & ~PFM_MASK)
1492                                                       | (*data & PFM_MASK));
1493                                 }
1494                         } else
1495                                 *data = cfm;
1496                         return 0;
1497                 case ELF_CR_IPSR_OFFSET:
1498                         if (write_access) {
1499                                 unsigned long tmp = *data;
1500                                 /* psr.ri==3 is a reserved value: SDM 2:25 */
1501                                 if ((tmp & IA64_PSR_RI) == IA64_PSR_RI)
1502                                         tmp &= ~IA64_PSR_RI;
1503                                 pt->cr_ipsr = ((tmp & IPSR_MASK)
1504                                                | (pt->cr_ipsr & ~IPSR_MASK));
1505                         } else
1506                                 *data = (pt->cr_ipsr & IPSR_MASK);
1507                         return 0;
1508                 }
1509         } else if (addr == ELF_NAT_OFFSET)
1510                 return access_nat_bits(target, pt, info,
1511                                        data, write_access);
1512         else if (addr == ELF_PR_OFFSET)
1513                 ptr = &pt->pr;
1514         else
1515                 return -1;
1516
1517         if (write_access)
1518                 *ptr = *data;
1519         else
1520                 *data = *ptr;
1521
1522         return 0;
1523 }
1524
1525 static int
1526 access_elf_reg(struct task_struct *target, struct unw_frame_info *info,
1527                 unsigned long addr, unsigned long *data, int write_access)
1528 {
1529         if (addr >= ELF_GR_OFFSET(1) && addr <= ELF_GR_OFFSET(15))
1530                 return access_elf_gpreg(target, info, addr, data, write_access);
1531         else if (addr >= ELF_BR_OFFSET(0) && addr <= ELF_BR_OFFSET(7))
1532                 return access_elf_breg(target, info, addr, data, write_access);
1533         else
1534                 return access_elf_areg(target, info, addr, data, write_access);
1535 }
1536
1537 void do_gpregs_get(struct unw_frame_info *info, void *arg)
1538 {
1539         struct pt_regs *pt;
1540         struct regset_getset *dst = arg;
1541         elf_greg_t tmp[16];
1542         unsigned int i, index, min_copy;
1543
1544         if (unw_unwind_to_user(info) < 0)
1545                 return;
1546
1547         /*
1548          * coredump format:
1549          *      r0-r31
1550          *      NaT bits (for r0-r31; bit N == 1 iff rN is a NaT)
1551          *      predicate registers (p0-p63)
1552          *      b0-b7
1553          *      ip cfm user-mask
1554          *      ar.rsc ar.bsp ar.bspstore ar.rnat
1555          *      ar.ccv ar.unat ar.fpsr ar.pfs ar.lc ar.ec
1556          */
1557
1558
1559         /* Skip r0 */
1560         if (dst->count > 0 && dst->pos < ELF_GR_OFFSET(1)) {
1561                 dst->ret = user_regset_copyout_zero(&dst->pos, &dst->count,
1562                                                       &dst->u.get.kbuf,
1563                                                       &dst->u.get.ubuf,
1564                                                       0, ELF_GR_OFFSET(1));
1565                 if (dst->ret || dst->count == 0)
1566                         return;
1567         }
1568
1569         /* gr1 - gr15 */
1570         if (dst->count > 0 && dst->pos < ELF_GR_OFFSET(16)) {
1571                 index = (dst->pos - ELF_GR_OFFSET(1)) / sizeof(elf_greg_t);
1572                 min_copy = ELF_GR_OFFSET(16) > (dst->pos + dst->count) ?
1573                          (dst->pos + dst->count) : ELF_GR_OFFSET(16);
1574                 for (i = dst->pos; i < min_copy; i += sizeof(elf_greg_t),
1575                                 index++)
1576                         if (access_elf_reg(dst->target, info, i,
1577                                                 &tmp[index], 0) < 0) {
1578                                 dst->ret = -EIO;
1579                                 return;
1580                         }
1581                 dst->ret = user_regset_copyout(&dst->pos, &dst->count,
1582                                 &dst->u.get.kbuf, &dst->u.get.ubuf, tmp,
1583                                 ELF_GR_OFFSET(1), ELF_GR_OFFSET(16));
1584                 if (dst->ret || dst->count == 0)
1585                         return;
1586         }
1587
1588         /* r16-r31 */
1589         if (dst->count > 0 && dst->pos < ELF_NAT_OFFSET) {
1590                 pt = task_pt_regs(dst->target);
1591                 dst->ret = user_regset_copyout(&dst->pos, &dst->count,
1592                                 &dst->u.get.kbuf, &dst->u.get.ubuf, &pt->r16,
1593                                 ELF_GR_OFFSET(16), ELF_NAT_OFFSET);
1594                 if (dst->ret || dst->count == 0)
1595                         return;
1596         }
1597
1598         /* nat, pr, b0 - b7 */
1599         if (dst->count > 0 && dst->pos < ELF_CR_IIP_OFFSET) {
1600                 index = (dst->pos - ELF_NAT_OFFSET) / sizeof(elf_greg_t);
1601                 min_copy = ELF_CR_IIP_OFFSET > (dst->pos + dst->count) ?
1602                          (dst->pos + dst->count) : ELF_CR_IIP_OFFSET;
1603                 for (i = dst->pos; i < min_copy; i += sizeof(elf_greg_t),
1604                                 index++)
1605                         if (access_elf_reg(dst->target, info, i,
1606                                                 &tmp[index], 0) < 0) {
1607                                 dst->ret = -EIO;
1608                                 return;
1609                         }
1610                 dst->ret = user_regset_copyout(&dst->pos, &dst->count,
1611                                 &dst->u.get.kbuf, &dst->u.get.ubuf, tmp,
1612                                 ELF_NAT_OFFSET, ELF_CR_IIP_OFFSET);
1613                 if (dst->ret || dst->count == 0)
1614                         return;
1615         }
1616
1617         /* ip cfm psr ar.rsc ar.bsp ar.bspstore ar.rnat
1618          * ar.ccv ar.unat ar.fpsr ar.pfs ar.lc ar.ec ar.csd ar.ssd
1619          */
1620         if (dst->count > 0 && dst->pos < (ELF_AR_END_OFFSET)) {
1621                 index = (dst->pos - ELF_CR_IIP_OFFSET) / sizeof(elf_greg_t);
1622                 min_copy = ELF_AR_END_OFFSET > (dst->pos + dst->count) ?
1623                          (dst->pos + dst->count) : ELF_AR_END_OFFSET;
1624                 for (i = dst->pos; i < min_copy; i += sizeof(elf_greg_t),
1625                                 index++)
1626                         if (access_elf_reg(dst->target, info, i,
1627                                                 &tmp[index], 0) < 0) {
1628                                 dst->ret = -EIO;
1629                                 return;
1630                         }
1631                 dst->ret = user_regset_copyout(&dst->pos, &dst->count,
1632                                 &dst->u.get.kbuf, &dst->u.get.ubuf, tmp,
1633                                 ELF_CR_IIP_OFFSET, ELF_AR_END_OFFSET);
1634         }
1635 }
1636
1637 void do_gpregs_set(struct unw_frame_info *info, void *arg)
1638 {
1639         struct pt_regs *pt;
1640         struct regset_getset *dst = arg;
1641         elf_greg_t tmp[16];
1642         unsigned int i, index;
1643
1644         if (unw_unwind_to_user(info) < 0)
1645                 return;
1646
1647         /* Skip r0 */
1648         if (dst->count > 0 && dst->pos < ELF_GR_OFFSET(1)) {
1649                 dst->ret = user_regset_copyin_ignore(&dst->pos, &dst->count,
1650                                                        &dst->u.set.kbuf,
1651                                                        &dst->u.set.ubuf,
1652                                                        0, ELF_GR_OFFSET(1));
1653                 if (dst->ret || dst->count == 0)
1654                         return;
1655         }
1656
1657         /* gr1-gr15 */
1658         if (dst->count > 0 && dst->pos < ELF_GR_OFFSET(16)) {
1659                 i = dst->pos;
1660                 index = (dst->pos - ELF_GR_OFFSET(1)) / sizeof(elf_greg_t);
1661                 dst->ret = user_regset_copyin(&dst->pos, &dst->count,
1662                                 &dst->u.set.kbuf, &dst->u.set.ubuf, tmp,
1663                                 ELF_GR_OFFSET(1), ELF_GR_OFFSET(16));
1664                 if (dst->ret)
1665                         return;
1666                 for ( ; i < dst->pos; i += sizeof(elf_greg_t), index++)
1667                         if (access_elf_reg(dst->target, info, i,
1668                                                 &tmp[index], 1) < 0) {
1669                                 dst->ret = -EIO;
1670                                 return;
1671                         }
1672                 if (dst->count == 0)
1673                         return;
1674         }
1675
1676         /* gr16-gr31 */
1677         if (dst->count > 0 && dst->pos < ELF_NAT_OFFSET) {
1678                 pt = task_pt_regs(dst->target);
1679                 dst->ret = user_regset_copyin(&dst->pos, &dst->count,
1680                                 &dst->u.set.kbuf, &dst->u.set.ubuf, &pt->r16,
1681                                 ELF_GR_OFFSET(16), ELF_NAT_OFFSET);
1682                 if (dst->ret || dst->count == 0)
1683                         return;
1684         }
1685
1686         /* nat, pr, b0 - b7 */
1687         if (dst->count > 0 && dst->pos < ELF_CR_IIP_OFFSET) {
1688                 i = dst->pos;
1689                 index = (dst->pos - ELF_NAT_OFFSET) / sizeof(elf_greg_t);
1690                 dst->ret = user_regset_copyin(&dst->pos, &dst->count,
1691                                 &dst->u.set.kbuf, &dst->u.set.ubuf, tmp,
1692                                 ELF_NAT_OFFSET, ELF_CR_IIP_OFFSET);
1693                 if (dst->ret)
1694                         return;
1695                 for (; i < dst->pos; i += sizeof(elf_greg_t), index++)
1696                         if (access_elf_reg(dst->target, info, i,
1697                                                 &tmp[index], 1) < 0) {
1698                                 dst->ret = -EIO;
1699                                 return;
1700                         }
1701                 if (dst->count == 0)
1702                         return;
1703         }
1704
1705         /* ip cfm psr ar.rsc ar.bsp ar.bspstore ar.rnat
1706          * ar.ccv ar.unat ar.fpsr ar.pfs ar.lc ar.ec ar.csd ar.ssd
1707          */
1708         if (dst->count > 0 && dst->pos < (ELF_AR_END_OFFSET)) {
1709                 i = dst->pos;
1710                 index = (dst->pos - ELF_CR_IIP_OFFSET) / sizeof(elf_greg_t);
1711                 dst->ret = user_regset_copyin(&dst->pos, &dst->count,
1712                                 &dst->u.set.kbuf, &dst->u.set.ubuf, tmp,
1713                                 ELF_CR_IIP_OFFSET, ELF_AR_END_OFFSET);
1714                 if (dst->ret)
1715                         return;
1716                 for ( ; i < dst->pos; i += sizeof(elf_greg_t), index++)
1717                         if (access_elf_reg(dst->target, info, i,
1718                                                 &tmp[index], 1) < 0) {
1719                                 dst->ret = -EIO;
1720                                 return;
1721                         }
1722         }
1723 }
1724
1725 #define ELF_FP_OFFSET(i)        (i * sizeof(elf_fpreg_t))
1726
1727 void do_fpregs_get(struct unw_frame_info *info, void *arg)
1728 {
1729         struct regset_getset *dst = arg;
1730         struct task_struct *task = dst->target;
1731         elf_fpreg_t tmp[30];
1732         int index, min_copy, i;
1733
1734         if (unw_unwind_to_user(info) < 0)
1735                 return;
1736
1737         /* Skip pos 0 and 1 */
1738         if (dst->count > 0 && dst->pos < ELF_FP_OFFSET(2)) {
1739                 dst->ret = user_regset_copyout_zero(&dst->pos, &dst->count,
1740                                                       &dst->u.get.kbuf,
1741                                                       &dst->u.get.ubuf,
1742                                                       0, ELF_FP_OFFSET(2));
1743                 if (dst->count == 0 || dst->ret)
1744                         return;
1745         }
1746
1747         /* fr2-fr31 */
1748         if (dst->count > 0 && dst->pos < ELF_FP_OFFSET(32)) {
1749                 index = (dst->pos - ELF_FP_OFFSET(2)) / sizeof(elf_fpreg_t);
1750
1751                 min_copy = min(((unsigned int)ELF_FP_OFFSET(32)),
1752                                 dst->pos + dst->count);
1753                 for (i = dst->pos; i < min_copy; i += sizeof(elf_fpreg_t),
1754                                 index++)
1755                         if (unw_get_fr(info, i / sizeof(elf_fpreg_t),
1756                                          &tmp[index])) {
1757                                 dst->ret = -EIO;
1758                                 return;
1759                         }
1760                 dst->ret = user_regset_copyout(&dst->pos, &dst->count,
1761                                 &dst->u.get.kbuf, &dst->u.get.ubuf, tmp,
1762                                 ELF_FP_OFFSET(2), ELF_FP_OFFSET(32));
1763                 if (dst->count == 0 || dst->ret)
1764                         return;
1765         }
1766
1767         /* fph */
1768         if (dst->count > 0) {
1769                 ia64_flush_fph(dst->target);
1770                 if (task->thread.flags & IA64_THREAD_FPH_VALID)
1771                         dst->ret = user_regset_copyout(
1772                                 &dst->pos, &dst->count,
1773                                 &dst->u.get.kbuf, &dst->u.get.ubuf,
1774                                 &dst->target->thread.fph,
1775                                 ELF_FP_OFFSET(32), -1);
1776                 else
1777                         /* Zero fill instead.  */
1778                         dst->ret = user_regset_copyout_zero(
1779                                 &dst->pos, &dst->count,
1780                                 &dst->u.get.kbuf, &dst->u.get.ubuf,
1781                                 ELF_FP_OFFSET(32), -1);
1782         }
1783 }
1784
1785 void do_fpregs_set(struct unw_frame_info *info, void *arg)
1786 {
1787         struct regset_getset *dst = arg;
1788         elf_fpreg_t fpreg, tmp[30];
1789         int index, start, end;
1790
1791         if (unw_unwind_to_user(info) < 0)
1792                 return;
1793
1794         /* Skip pos 0 and 1 */
1795         if (dst->count > 0 && dst->pos < ELF_FP_OFFSET(2)) {
1796                 dst->ret = user_regset_copyin_ignore(&dst->pos, &dst->count,
1797                                                        &dst->u.set.kbuf,
1798                                                        &dst->u.set.ubuf,
1799                                                        0, ELF_FP_OFFSET(2));
1800                 if (dst->count == 0 || dst->ret)
1801                         return;
1802         }
1803
1804         /* fr2-fr31 */
1805         if (dst->count > 0 && dst->pos < ELF_FP_OFFSET(32)) {
1806                 start = dst->pos;
1807                 end = min(((unsigned int)ELF_FP_OFFSET(32)),
1808                          dst->pos + dst->count);
1809                 dst->ret = user_regset_copyin(&dst->pos, &dst->count,
1810                                 &dst->u.set.kbuf, &dst->u.set.ubuf, tmp,
1811                                 ELF_FP_OFFSET(2), ELF_FP_OFFSET(32));
1812                 if (dst->ret)
1813                         return;
1814
1815                 if (start & 0xF) { /* only write high part */
1816                         if (unw_get_fr(info, start / sizeof(elf_fpreg_t),
1817                                          &fpreg)) {
1818                                 dst->ret = -EIO;
1819                                 return;
1820                         }
1821                         tmp[start / sizeof(elf_fpreg_t) - 2].u.bits[0]
1822                                 = fpreg.u.bits[0];
1823                         start &= ~0xFUL;
1824                 }
1825                 if (end & 0xF) { /* only write low part */
1826                         if (unw_get_fr(info, end / sizeof(elf_fpreg_t),
1827                                         &fpreg)) {
1828                                 dst->ret = -EIO;
1829                                 return;
1830                         }
1831                         tmp[end / sizeof(elf_fpreg_t) - 2].u.bits[1]
1832                                 = fpreg.u.bits[1];
1833                         end = (end + 0xF) & ~0xFUL;
1834                 }
1835
1836                 for ( ; start < end ; start += sizeof(elf_fpreg_t)) {
1837                         index = start / sizeof(elf_fpreg_t);
1838                         if (unw_set_fr(info, index, tmp[index - 2])) {
1839                                 dst->ret = -EIO;
1840                                 return;
1841                         }
1842                 }
1843                 if (dst->ret || dst->count == 0)
1844                         return;
1845         }
1846
1847         /* fph */
1848         if (dst->count > 0 && dst->pos < ELF_FP_OFFSET(128)) {
1849                 ia64_sync_fph(dst->target);
1850                 dst->ret = user_regset_copyin(&dst->pos, &dst->count,
1851                                                 &dst->u.set.kbuf,
1852                                                 &dst->u.set.ubuf,
1853                                                 &dst->target->thread.fph,
1854                                                 ELF_FP_OFFSET(32), -1);
1855         }
1856 }
1857
1858 static int
1859 do_regset_call(void (*call)(struct unw_frame_info *, void *),
1860                struct task_struct *target,
1861                const struct user_regset *regset,
1862                unsigned int pos, unsigned int count,
1863                const void *kbuf, const void __user *ubuf)
1864 {
1865         struct regset_getset info = { .target = target, .regset = regset,
1866                                  .pos = pos, .count = count,
1867                                  .u.set = { .kbuf = kbuf, .ubuf = ubuf },
1868                                  .ret = 0 };
1869
1870         if (target == current)
1871                 unw_init_running(call, &info);
1872         else {
1873                 struct unw_frame_info ufi;
1874                 memset(&ufi, 0, sizeof(ufi));
1875                 unw_init_from_blocked_task(&ufi, target);
1876                 (*call)(&ufi, &info);
1877         }
1878
1879         return info.ret;
1880 }
1881
1882 static int
1883 gpregs_get(struct task_struct *target,
1884            const struct user_regset *regset,
1885            unsigned int pos, unsigned int count,
1886            void *kbuf, void __user *ubuf)
1887 {
1888         return do_regset_call(do_gpregs_get, target, regset, pos, count,
1889                 kbuf, ubuf);
1890 }
1891
1892 static int gpregs_set(struct task_struct *target,
1893                 const struct user_regset *regset,
1894                 unsigned int pos, unsigned int count,
1895                 const void *kbuf, const void __user *ubuf)
1896 {
1897         return do_regset_call(do_gpregs_set, target, regset, pos, count,
1898                 kbuf, ubuf);
1899 }
1900
1901 static void do_gpregs_writeback(struct unw_frame_info *info, void *arg)
1902 {
1903         do_sync_rbs(info, ia64_sync_user_rbs);
1904 }
1905
1906 /*
1907  * This is called to write back the register backing store.
1908  * ptrace does this before it stops, so that a tracer reading the user
1909  * memory after the thread stops will get the current register data.
1910  */
1911 static int
1912 gpregs_writeback(struct task_struct *target,
1913                  const struct user_regset *regset,
1914                  int now)
1915 {
1916         if (test_and_set_tsk_thread_flag(target, TIF_RESTORE_RSE))
1917                 return 0;
1918         set_notify_resume(target);
1919         return do_regset_call(do_gpregs_writeback, target, regset, 0, 0,
1920                 NULL, NULL);
1921 }
1922
1923 static int
1924 fpregs_active(struct task_struct *target, const struct user_regset *regset)
1925 {
1926         return (target->thread.flags & IA64_THREAD_FPH_VALID) ? 128 : 32;
1927 }
1928
1929 static int fpregs_get(struct task_struct *target,
1930                 const struct user_regset *regset,
1931                 unsigned int pos, unsigned int count,
1932                 void *kbuf, void __user *ubuf)
1933 {
1934         return do_regset_call(do_fpregs_get, target, regset, pos, count,
1935                 kbuf, ubuf);
1936 }
1937
1938 static int fpregs_set(struct task_struct *target,
1939                 const struct user_regset *regset,
1940                 unsigned int pos, unsigned int count,
1941                 const void *kbuf, const void __user *ubuf)
1942 {
1943         return do_regset_call(do_fpregs_set, target, regset, pos, count,
1944                 kbuf, ubuf);
1945 }
1946
1947 static int
1948 access_uarea(struct task_struct *child, unsigned long addr,
1949               unsigned long *data, int write_access)
1950 {
1951         unsigned int pos = -1; /* an invalid value */
1952         int ret;
1953         unsigned long *ptr, regnum;
1954
1955         if ((addr & 0x7) != 0) {
1956                 dprintk("ptrace: unaligned register address 0x%lx\n", addr);
1957                 return -1;
1958         }
1959         if ((addr >= PT_NAT_BITS + 8 && addr < PT_F2) ||
1960                 (addr >= PT_R7 + 8 && addr < PT_B1) ||
1961                 (addr >= PT_AR_LC + 8 && addr < PT_CR_IPSR) ||
1962                 (addr >= PT_AR_SSD + 8 && addr < PT_DBR)) {
1963                 dprintk("ptrace: rejecting access to register "
1964                                         "address 0x%lx\n", addr);
1965                 return -1;
1966         }
1967
1968         switch (addr) {
1969         case PT_F32 ... (PT_F127 + 15):
1970                 pos = addr - PT_F32 + ELF_FP_OFFSET(32);
1971                 break;
1972         case PT_F2 ... (PT_F5 + 15):
1973                 pos = addr - PT_F2 + ELF_FP_OFFSET(2);
1974                 break;
1975         case PT_F10 ... (PT_F31 + 15):
1976                 pos = addr - PT_F10 + ELF_FP_OFFSET(10);
1977                 break;
1978         case PT_F6 ... (PT_F9 + 15):
1979                 pos = addr - PT_F6 + ELF_FP_OFFSET(6);
1980                 break;
1981         }
1982
1983         if (pos != -1) {
1984                 if (write_access)
1985                         ret = fpregs_set(child, NULL, pos,
1986                                 sizeof(unsigned long), data, NULL);
1987                 else
1988                         ret = fpregs_get(child, NULL, pos,
1989                                 sizeof(unsigned long), data, NULL);
1990                 if (ret != 0)
1991                         return -1;
1992                 return 0;
1993         }
1994
1995         switch (addr) {
1996         case PT_NAT_BITS:
1997                 pos = ELF_NAT_OFFSET;
1998                 break;
1999         case PT_R4 ... PT_R7:
2000                 pos = addr - PT_R4 + ELF_GR_OFFSET(4);
2001                 break;
2002         case PT_B1 ... PT_B5:
2003                 pos = addr - PT_B1 + ELF_BR_OFFSET(1);
2004                 break;
2005         case PT_AR_EC:
2006                 pos = ELF_AR_EC_OFFSET;
2007                 break;
2008         case PT_AR_LC:
2009                 pos = ELF_AR_LC_OFFSET;
2010                 break;
2011         case PT_CR_IPSR:
2012                 pos = ELF_CR_IPSR_OFFSET;
2013                 break;
2014         case PT_CR_IIP:
2015                 pos = ELF_CR_IIP_OFFSET;
2016                 break;
2017         case PT_CFM:
2018                 pos = ELF_CFM_OFFSET;
2019                 break;
2020         case PT_AR_UNAT:
2021                 pos = ELF_AR_UNAT_OFFSET;
2022                 break;
2023         case PT_AR_PFS:
2024                 pos = ELF_AR_PFS_OFFSET;
2025                 break;
2026         case PT_AR_RSC:
2027                 pos = ELF_AR_RSC_OFFSET;
2028                 break;
2029         case PT_AR_RNAT:
2030                 pos = ELF_AR_RNAT_OFFSET;
2031                 break;
2032         case PT_AR_BSPSTORE:
2033                 pos = ELF_AR_BSPSTORE_OFFSET;
2034                 break;
2035         case PT_PR:
2036                 pos = ELF_PR_OFFSET;
2037                 break;
2038         case PT_B6:
2039                 pos = ELF_BR_OFFSET(6);
2040                 break;
2041         case PT_AR_BSP:
2042                 pos = ELF_AR_BSP_OFFSET;
2043                 break;
2044         case PT_R1 ... PT_R3:
2045                 pos = addr - PT_R1 + ELF_GR_OFFSET(1);
2046                 break;
2047         case PT_R12 ... PT_R15:
2048                 pos = addr - PT_R12 + ELF_GR_OFFSET(12);
2049                 break;
2050         case PT_R8 ... PT_R11:
2051                 pos = addr - PT_R8 + ELF_GR_OFFSET(8);
2052                 break;
2053         case PT_R16 ... PT_R31:
2054                 pos = addr - PT_R16 + ELF_GR_OFFSET(16);
2055                 break;
2056         case PT_AR_CCV:
2057                 pos = ELF_AR_CCV_OFFSET;
2058                 break;
2059         case PT_AR_FPSR:
2060                 pos = ELF_AR_FPSR_OFFSET;
2061                 break;
2062         case PT_B0:
2063                 pos = ELF_BR_OFFSET(0);
2064                 break;
2065         case PT_B7:
2066                 pos = ELF_BR_OFFSET(7);
2067                 break;
2068         case PT_AR_CSD:
2069                 pos = ELF_AR_CSD_OFFSET;
2070                 break;
2071         case PT_AR_SSD:
2072                 pos = ELF_AR_SSD_OFFSET;
2073                 break;
2074         }
2075
2076         if (pos != -1) {
2077                 if (write_access)
2078                         ret = gpregs_set(child, NULL, pos,
2079                                 sizeof(unsigned long), data, NULL);
2080                 else
2081                         ret = gpregs_get(child, NULL, pos,
2082                                 sizeof(unsigned long), data, NULL);
2083                 if (ret != 0)
2084                         return -1;
2085                 return 0;
2086         }
2087
2088         /* access debug registers */
2089         if (addr >= PT_IBR) {
2090                 regnum = (addr - PT_IBR) >> 3;
2091                 ptr = &child->thread.ibr[0];
2092         } else {
2093                 regnum = (addr - PT_DBR) >> 3;
2094                 ptr = &child->thread.dbr[0];
2095         }
2096
2097         if (regnum >= 8) {
2098                 dprintk("ptrace: rejecting access to register "
2099                                 "address 0x%lx\n", addr);
2100                 return -1;
2101         }
2102 #ifdef CONFIG_PERFMON
2103         /*
2104          * Check if debug registers are used by perfmon. This
2105          * test must be done once we know that we can do the
2106          * operation, i.e. the arguments are all valid, but
2107          * before we start modifying the state.
2108          *
2109          * Perfmon needs to keep a count of how many processes
2110          * are trying to modify the debug registers for system
2111          * wide monitoring sessions.
2112          *
2113          * We also include read access here, because they may
2114          * cause the PMU-installed debug register state
2115          * (dbr[], ibr[]) to be reset. The two arrays are also
2116          * used by perfmon, but we do not use
2117          * IA64_THREAD_DBG_VALID. The registers are restored
2118          * by the PMU context switch code.
2119          */
2120         if (pfm_use_debug_registers(child))
2121                 return -1;
2122 #endif
2123
2124         if (!(child->thread.flags & IA64_THREAD_DBG_VALID)) {
2125                 child->thread.flags |= IA64_THREAD_DBG_VALID;
2126                 memset(child->thread.dbr, 0,
2127                                 sizeof(child->thread.dbr));
2128                 memset(child->thread.ibr, 0,
2129                                 sizeof(child->thread.ibr));
2130         }
2131
2132         ptr += regnum;
2133
2134         if ((regnum & 1) && write_access) {
2135                 /* don't let the user set kernel-level breakpoints: */
2136                 *ptr = *data & ~(7UL << 56);
2137                 return 0;
2138         }
2139         if (write_access)
2140                 *ptr = *data;
2141         else
2142                 *data = *ptr;
2143         return 0;
2144 }
2145
2146 static const struct user_regset native_regsets[] = {
2147         {
2148                 .core_note_type = NT_PRSTATUS,
2149                 .n = ELF_NGREG,
2150                 .size = sizeof(elf_greg_t), .align = sizeof(elf_greg_t),
2151                 .get = gpregs_get, .set = gpregs_set,
2152                 .writeback = gpregs_writeback
2153         },
2154         {
2155                 .core_note_type = NT_PRFPREG,
2156                 .n = ELF_NFPREG,
2157                 .size = sizeof(elf_fpreg_t), .align = sizeof(elf_fpreg_t),
2158                 .get = fpregs_get, .set = fpregs_set, .active = fpregs_active
2159         },
2160 };
2161
2162 static const struct user_regset_view user_ia64_view = {
2163         .name = "ia64",
2164         .e_machine = EM_IA_64,
2165         .regsets = native_regsets, .n = ARRAY_SIZE(native_regsets)
2166 };
2167
2168 const struct user_regset_view *task_user_regset_view(struct task_struct *tsk)
2169 {
2170         return &user_ia64_view;
2171 }
2172
2173 struct syscall_get_set_args {
2174         unsigned int i;
2175         unsigned int n;
2176         unsigned long *args;
2177         struct pt_regs *regs;
2178         int rw;
2179 };
2180
2181 static void syscall_get_set_args_cb(struct unw_frame_info *info, void *data)
2182 {
2183         struct syscall_get_set_args *args = data;
2184         struct pt_regs *pt = args->regs;
2185         unsigned long *krbs, cfm, ndirty;
2186         int i, count;
2187
2188         if (unw_unwind_to_user(info) < 0)
2189                 return;
2190
2191         cfm = pt->cr_ifs;
2192         krbs = (unsigned long *)info->task + IA64_RBS_OFFSET/8;
2193         ndirty = ia64_rse_num_regs(krbs, krbs + (pt->loadrs >> 19));
2194
2195         count = 0;
2196         if (in_syscall(pt))
2197                 count = min_t(int, args->n, cfm & 0x7f);
2198
2199         for (i = 0; i < count; i++) {
2200                 if (args->rw)
2201                         *ia64_rse_skip_regs(krbs, ndirty + i + args->i) =
2202                                 args->args[i];
2203                 else
2204                         args->args[i] = *ia64_rse_skip_regs(krbs,
2205                                 ndirty + i + args->i);
2206         }
2207
2208         if (!args->rw) {
2209                 while (i < args->n) {
2210                         args->args[i] = 0;
2211                         i++;
2212                 }
2213         }
2214 }
2215
2216 void ia64_syscall_get_set_arguments(struct task_struct *task,
2217         struct pt_regs *regs, unsigned int i, unsigned int n,
2218         unsigned long *args, int rw)
2219 {
2220         struct syscall_get_set_args data = {
2221                 .i = i,
2222                 .n = n,
2223                 .args = args,
2224                 .regs = regs,
2225                 .rw = rw,
2226         };
2227
2228         if (task == current)
2229                 unw_init_running(syscall_get_set_args_cb, &data);
2230         else {
2231                 struct unw_frame_info ufi;
2232                 memset(&ufi, 0, sizeof(ufi));
2233                 unw_init_from_blocked_task(&ufi, task);
2234                 syscall_get_set_args_cb(&ufi, &data);
2235         }
2236 }