aboutsummaryrefslogblamecommitdiffstats
path: root/arch/ia64/kernel/ptrace.c
blob: 6d57aebad485fd35db55c8ffbc9497fd6aba34b9 (plain) (tree)
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637


















                                                                  
                         








































































































































































































































































































































































































































































































































































































































                                                                               




                                            




                                                            
                         



















































                                                                            
                                 





                                                 












                                                                           
                               


                                                       






























































































































































































































                                                                               






                                                                              


















































                                                                          










































































































































































































































                                                                               
                                                                          































                                                                          
                                                        
































































































                                                                               
                                                          








































































































































                                                                            
                                        





































                                                                        
                                        









































































                                                                              


                                                

                                               



                                             
                                          

                                               
                                           

                                               
 
                                                                                    

         









                                                                
                                                                               




                                               
/*
 * Kernel support for the ptrace() and syscall tracing interfaces.
 *
 * Copyright (C) 1999-2005 Hewlett-Packard Co
 *	David Mosberger-Tang <davidm@hpl.hp.com>
 *
 * Derived from the x86 and Alpha versions.
 */
#include <linux/config.h>
#include <linux/kernel.h>
#include <linux/sched.h>
#include <linux/slab.h>
#include <linux/mm.h>
#include <linux/errno.h>
#include <linux/ptrace.h>
#include <linux/smp_lock.h>
#include <linux/user.h>
#include <linux/security.h>
#include <linux/audit.h>
#include <linux/signal.h>

#include <asm/pgtable.h>
#include <asm/processor.h>
#include <asm/ptrace_offsets.h>
#include <asm/rse.h>
#include <asm/system.h>
#include <asm/uaccess.h>
#include <asm/unwind.h>
#ifdef CONFIG_PERFMON
#include <asm/perfmon.h>
#endif

#include "entry.h"

/*
 * Bits in the PSR that we allow ptrace() to change:
 *	be, up, ac, mfl, mfh (the user mask; five bits total)
 *	db (debug breakpoint fault; one bit)
 *	id (instruction debug fault disable; one bit)
 *	dd (data debug fault disable; one bit)
 *	ri (restart instruction; two bits)
 *	is (instruction set; one bit)
 */
#define IPSR_MASK (IA64_PSR_UM | IA64_PSR_DB | IA64_PSR_IS	\
		   | IA64_PSR_ID | IA64_PSR_DD | IA64_PSR_RI)

#define MASK(nbits)	((1UL << (nbits)) - 1)	/* mask with NBITS bits set */
#define PFM_MASK	MASK(38)

#define PTRACE_DEBUG	0

#if PTRACE_DEBUG
# define dprintk(format...)	printk(format)
# define inline
#else
# define dprintk(format...)
#endif

/* Return TRUE if PT was created due to kernel-entry via a system-call.  */

static inline int
in_syscall (struct pt_regs *pt)
{
	return (long) pt->cr_ifs >= 0;
}

/*
 * Collect the NaT bits for r1-r31 from scratch_unat and return a NaT
 * bitset where bit i is set iff the NaT bit of register i is set.
 */
unsigned long
ia64_get_scratch_nat_bits (struct pt_regs *pt, unsigned long scratch_unat)
{
#	define GET_BITS(first, last, unat)				\
	({								\
		unsigned long bit = ia64_unat_pos(&pt->r##first);	\
		unsigned long nbits = (last - first + 1);		\
		unsigned long mask = MASK(nbits) << first;		\
		unsigned long dist;					\
		if (bit < first)					\
			dist = 64 + bit - first;			\
		else							\
			dist = bit - first;				\
		ia64_rotr(unat, dist) & mask;				\
	})
	unsigned long val;

	/*
	 * Registers that are stored consecutively in struct pt_regs
	 * can be handled in parallel.  If the register order in
	 * struct_pt_regs changes, this code MUST be updated.
	 */
	val  = GET_BITS( 1,  1, scratch_unat);
	val |= GET_BITS( 2,  3, scratch_unat);
	val |= GET_BITS(12, 13, scratch_unat);
	val |= GET_BITS(14, 14, scratch_unat);
	val |= GET_BITS(15, 15, scratch_unat);
	val |= GET_BITS( 8, 11, scratch_unat);
	val |= GET_BITS(16, 31, scratch_unat);
	return val;

#	undef GET_BITS
}

/*
 * Set the NaT bits for the scratch registers according to NAT and
 * return the resulting unat (assuming the scratch registers are
 * stored in PT).
 */
unsigned long
ia64_put_scratch_nat_bits (struct pt_regs *pt, unsigned long nat)
{
#	define PUT_BITS(first, last, nat)				\
	({								\
		unsigned long bit = ia64_unat_pos(&pt->r##first);	\
		unsigned long nbits = (last - first + 1);		\
		unsigned long mask = MASK(nbits) << first;		\
		long dist;						\
		if (bit < first)					\
			dist = 64 + bit - first;			\
		else							\
			dist = bit - first;				\
		ia64_rotl(nat & mask, dist);				\
	})
	unsigned long scratch_unat;

	/*
	 * Registers that are stored consecutively in struct pt_regs
	 * can be handled in parallel.  If the register order in
	 * struct_pt_regs changes, this code MUST be updated.
	 */
	scratch_unat  = PUT_BITS( 1,  1, nat);
	scratch_unat |= PUT_BITS( 2,  3, nat);
	scratch_unat |= PUT_BITS(12, 13, nat);
	scratch_unat |= PUT_BITS(14, 14, nat);
	scratch_unat |= PUT_BITS(15, 15, nat);
	scratch_unat |= PUT_BITS( 8, 11, nat);
	scratch_unat |= PUT_BITS(16, 31, nat);

	return scratch_unat;

#	undef PUT_BITS
}

#define IA64_MLX_TEMPLATE	0x2
#define IA64_MOVL_OPCODE	6

void
ia64_increment_ip (struct pt_regs *regs)
{
	unsigned long w0, ri = ia64_psr(regs)->ri + 1;

	if (ri > 2) {
		ri = 0;
		regs->cr_iip += 16;
	} else if (ri == 2) {
		get_user(w0, (char __user *) regs->cr_iip + 0);
		if (((w0 >> 1) & 0xf) == IA64_MLX_TEMPLATE) {
			/*
			 * rfi'ing to slot 2 of an MLX bundle causes
			 * an illegal operation fault.  We don't want
			 * that to happen...
			 */
			ri = 0;
			regs->cr_iip += 16;
		}
	}
	ia64_psr(regs)->ri = ri;
}

void
ia64_decrement_ip (struct pt_regs *regs)
{
	unsigned long w0, ri = ia64_psr(regs)->ri - 1;

	if (ia64_psr(regs)->ri == 0) {
		regs->cr_iip -= 16;
		ri = 2;
		get_user(w0, (char __user *) regs->cr_iip + 0);
		if (((w0 >> 1) & 0xf) == IA64_MLX_TEMPLATE) {
			/*
			 * rfi'ing to slot 2 of an MLX bundle causes
			 * an illegal operation fault.  We don't want
			 * that to happen...
			 */
			ri = 1;
		}
	}
	ia64_psr(regs)->ri = ri;
}

/*
 * This routine is used to read an rnat bits that are stored on the
 * kernel backing store.  Since, in general, the alignment of the user
 * and kernel are different, this is not completely trivial.  In
 * essence, we need to construct the user RNAT based on up to two
 * kernel RNAT values and/or the RNAT value saved in the child's
 * pt_regs.
 *
 * user rbs
 *
 * +--------+ <-- lowest address
 * | slot62 |
 * +--------+
 * |  rnat  | 0x....1f8
 * +--------+
 * | slot00 | \
 * +--------+ |
 * | slot01 | > child_regs->ar_rnat
 * +--------+ |
 * | slot02 | /				kernel rbs
 * +--------+				+--------+
 *	    <- child_regs->ar_bspstore	| slot61 | <-- krbs
 * +- - - - +				+--------+
 *					| slot62 |
 * +- - - - +				+--------+
 *					|  rnat	 |
 * +- - - - +				+--------+
 *   vrnat				| slot00 |
 * +- - - - +				+--------+
 *					=	 =
 *					+--------+
 *					| slot00 | \
 *					+--------+ |
 *					| slot01 | > child_stack->ar_rnat
 *					+--------+ |
 *					| slot02 | /
 *					+--------+
 *						  <--- child_stack->ar_bspstore
 *
 * The way to think of this code is as follows: bit 0 in the user rnat
 * corresponds to some bit N (0 <= N <= 62) in one of the kernel rnat
 * value.  The kernel rnat value holding this bit is stored in
 * variable rnat0.  rnat1 is loaded with the kernel rnat value that
 * form the upper bits of the user rnat value.
 *
 * Boundary cases:
 *
 * o when reading the rnat "below" the first rnat slot on the kernel
 *   backing store, rnat0/rnat1 are set to 0 and the low order bits are
 *   merged in from pt->ar_rnat.
 *
 * o when reading the rnat "above" the last rnat slot on the kernel
 *   backing store, rnat0/rnat1 gets its value from sw->ar_rnat.
 */
static unsigned long
get_rnat (struct task_struct *task, struct switch_stack *sw,
	  unsigned long *krbs, unsigned long *urnat_addr,
	  unsigned long *urbs_end)
{
	unsigned long rnat0 = 0, rnat1 = 0, urnat = 0, *slot0_kaddr;
	unsigned long umask = 0, mask, m;
	unsigned long *kbsp, *ubspstore, *rnat0_kaddr, *rnat1_kaddr, shift;
	long num_regs, nbits;
	struct pt_regs *pt;

	pt = ia64_task_regs(task);
	kbsp = (unsigned long *) sw->ar_bspstore;
	ubspstore = (unsigned long *) pt->ar_bspstore;

	if (urbs_end < urnat_addr)
		nbits = ia64_rse_num_regs(urnat_addr - 63, urbs_end);
	else
		nbits = 63;
	mask = MASK(nbits);
	/*
	 * First, figure out which bit number slot 0 in user-land maps
	 * to in the kernel rnat.  Do this by figuring out how many
	 * register slots we're beyond the user's backingstore and
	 * then computing the equivalent address in kernel space.
	 */
	num_regs = ia64_rse_num_regs(ubspstore, urnat_addr + 1);
	slot0_kaddr = ia64_rse_skip_regs(krbs, num_regs);
	shift = ia64_rse_slot_num(slot0_kaddr);
	rnat1_kaddr = ia64_rse_rnat_addr(slot0_kaddr);
	rnat0_kaddr = rnat1_kaddr - 64;

	if (ubspstore + 63 > urnat_addr) {
		/* some bits need to be merged in from pt->ar_rnat */
		umask = MASK(ia64_rse_slot_num(ubspstore)) & mask;
		urnat = (pt->ar_rnat & umask);
		mask &= ~umask;
		if (!mask)
			return urnat;
	}

	m = mask << shift;
	if (rnat0_kaddr >= kbsp)
		rnat0 = sw->ar_rnat;
	else if (rnat0_kaddr > krbs)
		rnat0 = *rnat0_kaddr;
	urnat |= (rnat0 & m) >> shift;

	m = mask >> (63 - shift);
	if (rnat1_kaddr >= kbsp)
		rnat1 = sw->ar_rnat;
	else if (rnat1_kaddr > krbs)
		rnat1 = *rnat1_kaddr;
	urnat |= (rnat1 & m) << (63 - shift);
	return urnat;
}

/*
 * The reverse of get_rnat.
 */
static void
put_rnat (struct task_struct *task, struct switch_stack *sw,
	  unsigned long *krbs, unsigned long *urnat_addr, unsigned long urnat,
	  unsigned long *urbs_end)
{
	unsigned long rnat0 = 0, rnat1 = 0, *slot0_kaddr, umask = 0, mask, m;
	unsigned long *kbsp, *ubspstore, *rnat0_kaddr, *rnat1_kaddr, shift;
	long num_regs, nbits;
	struct pt_regs *pt;
	unsigned long cfm, *urbs_kargs;

	pt = ia64_task_regs(task);
	kbsp = (unsigned long *) sw->ar_bspstore;
	ubspstore = (unsigned long *) pt->ar_bspstore;

	urbs_kargs = urbs_end;
	if (in_syscall(pt)) {
		/*
		 * If entered via syscall, don't allow user to set rnat bits
		 * for syscall args.
		 */
		cfm = pt->cr_ifs;
		urbs_kargs = ia64_rse_skip_regs(urbs_end, -(cfm & 0x7f));
	}

	if (urbs_kargs >= urnat_addr)
		nbits = 63;
	else {
		if ((urnat_addr - 63) >= urbs_kargs)
			return;
		nbits = ia64_rse_num_regs(urnat_addr - 63, urbs_kargs);
	}
	mask = MASK(nbits);

	/*
	 * First, figure out which bit number slot 0 in user-land maps
	 * to in the kernel rnat.  Do this by figuring out how many
	 * register slots we're beyond the user's backingstore and
	 * then computing the equivalent address in kernel space.
	 */
	num_regs = ia64_rse_num_regs(ubspstore, urnat_addr + 1);
	slot0_kaddr = ia64_rse_skip_regs(krbs, num_regs);
	shift = ia64_rse_slot_num(slot0_kaddr);
	rnat1_kaddr = ia64_rse_rnat_addr(slot0_kaddr);
	rnat0_kaddr = rnat1_kaddr - 64;

	if (ubspstore + 63 > urnat_addr) {
		/* some bits need to be place in pt->ar_rnat: */
		umask = MASK(ia64_rse_slot_num(ubspstore)) & mask;
		pt->ar_rnat = (pt->ar_rnat & ~umask) | (urnat & umask);
		mask &= ~umask;
		if (!mask)
			return;
	}
	/*
	 * Note: Section 11.1 of the EAS guarantees that bit 63 of an
	 * rnat slot is ignored. so we don't have to clear it here.
	 */
	rnat0 = (urnat << shift);
	m = mask << shift;
	if (rnat0_kaddr >= kbsp)
		sw->ar_rnat = (sw->ar_rnat & ~m) | (rnat0 & m);
	else if (rnat0_kaddr > krbs)
		*rnat0_kaddr = ((*rnat0_kaddr & ~m) | (rnat0 & m));

	rnat1 = (urnat >> (63 - shift));
	m = mask >> (63 - shift);
	if (rnat1_kaddr >= kbsp)
		sw->ar_rnat = (sw->ar_rnat & ~m) | (rnat1 & m);
	else if (rnat1_kaddr > krbs)
		*rnat1_kaddr = ((*rnat1_kaddr & ~m) | (rnat1 & m));
}

static inline int
on_kernel_rbs (unsigned long addr, unsigned long bspstore,
	       unsigned long urbs_end)
{
	unsigned long *rnat_addr = ia64_rse_rnat_addr((unsigned long *)
						      urbs_end);
	return (addr >= bspstore && addr <= (unsigned long) rnat_addr);
}

/*
 * Read a word from the user-level backing store of task CHILD.  ADDR
 * is the user-level address to read the word from, VAL a pointer to
 * the return value, and USER_BSP gives the end of the user-level
 * backing store (i.e., it's the address that would be in ar.bsp after
 * the user executed a "cover" instruction).
 *
 * This routine takes care of accessing the kernel register backing
 * store for those registers that got spilled there.  It also takes
 * care of calculating the appropriate RNaT collection words.
 */
long
ia64_peek (struct task_struct *child, struct switch_stack *child_stack,
	   unsigned long user_rbs_end, unsigned long addr, long *val)
{
	unsigned long *bspstore, *krbs, regnum, *laddr, *urbs_end, *rnat_addr;
	struct pt_regs *child_regs;
	size_t copied;
	long ret;

	urbs_end = (long *) user_rbs_end;
	laddr = (unsigned long *) addr;
	child_regs = ia64_task_regs(child);
	bspstore = (unsigned long *) child_regs->ar_bspstore;
	krbs = (unsigned long *) child + IA64_RBS_OFFSET/8;
	if (on_kernel_rbs(addr, (unsigned long) bspstore,
			  (unsigned long) urbs_end))
	{
		/*
		 * Attempt to read the RBS in an area that's actually
		 * on the kernel RBS => read the corresponding bits in
		 * the kernel RBS.
		 */
		rnat_addr = ia64_rse_rnat_addr(laddr);
		ret = get_rnat(child, child_stack, krbs, rnat_addr, urbs_end);

		if (laddr == rnat_addr) {
			/* return NaT collection word itself */
			*val = ret;
			return 0;
		}

		if (((1UL << ia64_rse_slot_num(laddr)) & ret) != 0) {
			/*
			 * It is implementation dependent whether the
			 * data portion of a NaT value gets saved on a
			 * st8.spill or RSE spill (e.g., see EAS 2.6,
			 * 4.4.4.6 Register Spill and Fill).  To get
			 * consistent behavior across all possible
			 * IA-64 implementations, we return zero in
			 * this case.
			 */
			*val = 0;
			return 0;
		}

		if (laddr < urbs_end) {
			/*
			 * The desired word is on the kernel RBS and
			 * is not a NaT.
			 */
			regnum = ia64_rse_num_regs(bspstore, laddr);
			*val = *ia64_rse_skip_regs(krbs, regnum);
			return 0;
		}
	}
	copied = access_process_vm(child, addr, &ret, sizeof(ret), 0);
	if (copied != sizeof(ret))
		return -EIO;
	*val = ret;
	return 0;
}

long
ia64_poke (struct task_struct *child, struct switch_stack *child_stack,
	   unsigned long user_rbs_end, unsigned long addr, long val)
{
	unsigned long *bspstore, *krbs, regnum, *laddr;
	unsigned long *urbs_end = (long *) user_rbs_end;
	struct pt_regs *child_regs;

	laddr = (unsigned long *) addr;
	child_regs = ia64_task_regs(child);
	bspstore = (unsigned long *) child_regs->ar_bspstore;
	krbs = (unsigned long *) child + IA64_RBS_OFFSET/8;
	if (on_kernel_rbs(addr, (unsigned long) bspstore,
			  (unsigned long) urbs_end))
	{
		/*
		 * Attempt to write the RBS in an area that's actually
		 * on the kernel RBS => write the corresponding bits
		 * in the kernel RBS.
		 */
		if (ia64_rse_is_rnat_slot(laddr))
			put_rnat(child, child_stack, krbs, laddr, val,
				 urbs_end);
		else {
			if (laddr < urbs_end) {
				regnum = ia64_rse_num_regs(bspstore, laddr);
				*ia64_rse_skip_regs(krbs, regnum) = val;
			}
		}
	} else if (access_process_vm(child, addr, &val, sizeof(val), 1)
		   != sizeof(val))
		return -EIO;
	return 0;
}

/*
 * Calculate the address of the end of the user-level register backing
 * store.  This is the address that would have been stored in ar.bsp
 * if the user had executed a "cover" instruction right before
 * entering the kernel.  If CFMP is not NULL, it is used to return the
 * "current frame mask" that was active at the time the kernel was
 * entered.
 */
unsigned long
ia64_get_user_rbs_end (struct task_struct *child, struct pt_regs *pt,
		       unsigned long *cfmp)
{
	unsigned long *krbs, *bspstore, cfm = pt->cr_ifs;
	long ndirty;

	krbs = (unsigned long *) child + IA64_RBS_OFFSET/8;
	bspstore = (unsigned long *) pt->ar_bspstore;
	ndirty = ia64_rse_num_regs(krbs, krbs + (pt->loadrs >> 19));

	if (in_syscall(pt))
		ndirty += (cfm & 0x7f);
	else
		cfm &= ~(1UL << 63);	/* clear valid bit */

	if (cfmp)
		*cfmp = cfm;
	return (unsigned long) ia64_rse_skip_regs(bspstore, ndirty);
}

/*
 * Synchronize (i.e, write) the RSE backing store living in kernel
 * space to the VM of the CHILD task.  SW and PT are the pointers to
 * the switch_stack and pt_regs structures, respectively.
 * USER_RBS_END is the user-level address at which the backing store
 * ends.
 */
long
ia64_sync_user_rbs (struct task_struct *child, struct switch_stack *sw,
		    unsigned long user_rbs_start, unsigned long user_rbs_end)
{
	unsigned long addr, val;
	long ret;

	/* now copy word for word from kernel rbs to user rbs: */
	for (addr = user_rbs_start; addr < user_rbs_end; addr += 8) {
		ret = ia64_peek(child, sw, user_rbs_end, addr, &val);
		if (ret < 0)
			return ret;
		if (access_process_vm(child, addr, &val, sizeof(val), 1)
		    != sizeof(val))
			return -EIO;
	}
	return 0;
}

static inline int
thread_matches (struct task_struct *thread, unsigned long addr)
{
	unsigned long thread_rbs_end;
	struct pt_regs *thread_regs;

	if (ptrace_check_attach(thread, 0) < 0)
		/*
		 * If the thread is not in an attachable state, we'll
		 * ignore it.  The net effect is that if ADDR happens
		 * to overlap with the portion of the thread's
		 * register backing store that is currently residing
		 * on the thread's kernel stack, then ptrace() may end
		 * up accessing a stale value.  But if the thread
		 * isn't stopped, that's a problem anyhow, so we're
		 * doing as well as we can...
		 */
		return 0;

	thread_regs = ia64_task_regs(thread);
	thread_rbs_end = ia64_get_user_rbs_end(thread, thread_regs, NULL);
	if (!on_kernel_rbs(addr, thread_regs->ar_bspstore, thread_rbs_end))
		return 0;

	return 1;	/* looks like we've got a winner */
}

/*
 * GDB apparently wants to be able to read the register-backing store
 * of any thread when attached to a given process.  If we are peeking
 * or poking an address that happens to reside in the kernel-backing
 * store of another thread, we need to attach to that thread, because
 * otherwise we end up accessing stale data.
 *
 * task_list_lock must be read-locked before calling this routine!
 */
static struct task_struct *
find_thread_for_addr (struct task_struct *child, unsigned long addr)
{
	struct task_struct *g, *p;
	struct mm_struct *mm;
	int mm_users;

	if (!(mm = get_task_mm(child)))
		return child;

	/* -1 because of our get_task_mm(): */
	mm_users = atomic_read(&mm->mm_users) - 1;
	if (mm_users <= 1)
		goto out;		/* not multi-threaded */

	/*
	 * First, traverse the child's thread-list.  Good for scalability with
	 * NPTL-threads.
	 */
	p = child;
	do {
		if (thread_matches(p, addr)) {
			child = p;
			goto out;
		}
		if (mm_users-- <= 1)
			goto out;
	} while ((p = next_thread(p)) != child);

	do_each_thread(g, p) {
		if (child->mm != mm)
			continue;

		if (thread_matches(p, addr)) {
			child = p;
			goto out;
		}
	} while_each_thread(g, p);
  out:
	mmput(mm);
	return child;
}

/*
 * Write f32-f127 back to task->thread.fph if it has been modified.
 */
inline void
ia64_flush_fph (struct task_struct *task)
{
	struct ia64_psr *psr = ia64_psr(ia64_task_regs(task));

	/*
	 * Prevent migrating this task while
	 * we're fiddling with the FPU state
	 */
	preempt_disable();
	if (ia64_is_local_fpu_owner(task) && psr->mfh) {
		psr->mfh = 0;
		task->thread.flags |= IA64_THREAD_FPH_VALID;
		ia64_save_fpu(&task->thread.fph[0]);
	}
	preempt_enable();
}

/*
 * Sync the fph state of the task so that it can be manipulated
 * through thread.fph.  If necessary, f32-f127 are written back to
 * thread.fph or, if the fph state hasn't been used before, thread.fph
 * is cleared to zeroes.  Also, access to f32-f127 is disabled to
 * ensure that the task picks up the state from thread.fph when it
 * executes again.
 */
void
ia64_sync_fph (struct task_struct *task)
{
	struct ia64_psr *psr = ia64_psr(ia64_task_regs(task));

	ia64_flush_fph(task);
	if (!(task->thread.flags & IA64_THREAD_FPH_VALID)) {
		task->thread.flags |= IA64_THREAD_FPH_VALID;
		memset(&task->thread.fph, 0, sizeof(task->thread.fph));
	}
	ia64_drop_fpu(task);
	psr->dfh = 1;
}

static int
access_fr (struct unw_frame_info *info, int regnum, int hi,
	   unsigned long *data, int write_access)
{
	struct ia64_fpreg fpval;
	int ret;

	ret = unw_get_fr(info, regnum, &fpval);
	if (ret < 0)
		return ret;

	if (write_access) {
		fpval.u.bits[hi] = *data;
		ret = unw_set_fr(info, regnum, fpval);
	} else
		*data = fpval.u.bits[hi];
	return ret;
}

/*
 * Change the machine-state of CHILD such that it will return via the normal
 * kernel exit-path, rather than the syscall-exit path.
 */
static void
convert_to_non_syscall (struct task_struct *child, struct pt_regs  *pt,
			unsigned long cfm)
{
	struct unw_frame_info info, prev_info;
	unsigned long ip, sp, pr;

	unw_init_from_blocked_task(&info, child);
	while (1) {
		prev_info = info;
		if (unw_unwind(&info) < 0)
			return;

		unw_get_sp(&info, &sp);
		if ((long)((unsigned long)child + IA64_STK_OFFSET - sp)
		    < IA64_PT_REGS_SIZE) {
			dprintk("ptrace.%s: ran off the top of the kernel "
				"stack\n", __FUNCTION__);
			return;
		}
		if (unw_get_pr (&prev_info, &pr) < 0) {
			unw_get_rp(&prev_info, &ip);
			dprintk("ptrace.%s: failed to read "
				"predicate register (ip=0x%lx)\n",
				__FUNCTION__, ip);
			return;
		}
		if (unw_is_intr_frame(&info)
		    && (pr & (1UL << PRED_USER_STACK)))
			break;
	}

	unw_get_pr(&prev_info, &pr);
	pr &= ~(1UL << PRED_SYSCALL);
	pr |=  (1UL << PRED_NON_SYSCALL);
	unw_set_pr(&prev_info, pr);

	pt->cr_ifs = (1UL << 63) | cfm;
}

static int
access_nat_bits (struct task_struct *child, struct pt_regs *pt,
		 struct unw_frame_info *info,
		 unsigned long *data, int write_access)
{
	unsigned long regnum, nat_bits, scratch_unat, dummy = 0;
	char nat = 0;

	if (write_access) {
		nat_bits = *data;
		scratch_unat = ia64_put_scratch_nat_bits(pt, nat_bits);
		if (unw_set_ar(info, UNW_AR_UNAT, scratch_unat) < 0) {
			dprintk("ptrace: failed to set ar.unat\n");
			return -1;
		}
		for (regnum = 4; regnum <= 7; ++regnum) {
			unw_get_gr(info, regnum, &dummy, &nat);
			unw_set_gr(info, regnum, dummy,
				   (nat_bits >> regnum) & 1);
		}
	} else {
		if (unw_get_ar(info, UNW_AR_UNAT, &scratch_unat) < 0) {
			dprintk("ptrace: failed to read ar.unat\n");
			return -1;
		}
		nat_bits = ia64_get_scratch_nat_bits(pt, scratch_unat);
		for (regnum = 4; regnum <= 7; ++regnum) {
			unw_get_gr(info, regnum, &dummy, &nat);
			nat_bits |= (nat != 0) << regnum;
		}
		*data = nat_bits;
	}
	return 0;
}

static int
access_uarea (struct task_struct *child, unsigned long addr,
	      unsigned long *data, int write_access)
{
	unsigned long *ptr, regnum, urbs_end, rnat_addr, cfm;
	struct switch_stack *sw;
	struct pt_regs *pt;
#	define pt_reg_addr(pt, reg)	((void *)			    \
					 ((unsigned long) (pt)		    \
					  + offsetof(struct pt_regs, reg)))


	pt = ia64_task_regs(child);
	sw = (struct switch_stack *) (child->thread.ksp + 16);

	if ((addr & 0x7) != 0) {
		dprintk("ptrace: unaligned register address 0x%lx\n", addr);
		return -1;
	}

	if (addr < PT_F127 + 16) {
		/* accessing fph */
		if (write_access)
			ia64_sync_fph(child);
		else
			ia64_flush_fph(child);
		ptr = (unsigned long *)
			((unsigned long) &child->thread.fph + addr);
	} else if ((addr >= PT_F10) && (addr < PT_F11 + 16)) {
		/* scratch registers untouched by kernel (saved in pt_regs) */
		ptr = pt_reg_addr(pt, f10) + (addr - PT_F10);
	} else if (addr >= PT_F12 && addr < PT_F15 + 16) {
		/*
		 * Scratch registers untouched by kernel (saved in
		 * switch_stack).
		 */
		ptr = (unsigned long *) ((long) sw
					 + (addr - PT_NAT_BITS - 32));
	} else if (addr < PT_AR_LC + 8) {
		/* preserved state: */
		struct unw_frame_info info;
		char nat = 0;
		int ret;

		unw_init_from_blocked_task(&info, child);
		if (unw_unwind_to_user(&info) < 0)
			return -1;

		switch (addr) {
		      case PT_NAT_BITS:
			return access_nat_bits(child, pt, &info,
					       data, write_access);

		      case PT_R4: case PT_R5: case PT_R6: case PT_R7:
			if (write_access) {
				/* read NaT bit first: */
				unsigned long dummy;

				ret = unw_get_gr(&info, (addr - PT_R4)/8 + 4,
						 &dummy, &nat);
				if (ret < 0)
					return ret;
			}
			return unw_access_gr(&info, (addr - PT_R4)/8 + 4, data,
					     &nat, write_access);

		      case PT_B1: case PT_B2: case PT_B3:
		      case PT_B4: case PT_B5:
			return unw_access_br(&info, (addr - PT_B1)/8 + 1, data,
					     write_access);

		      case PT_AR_EC:
			return unw_access_ar(&info, UNW_AR_EC, data,
					     write_access);

		      case PT_AR_LC:
			return unw_access_ar(&info, UNW_AR_LC, data,
					     write_access);

		      default:
			if (addr >= PT_F2 && addr < PT_F5 + 16)
				return access_fr(&info, (addr - PT_F2)/16 + 2,
						 (addr & 8) != 0, data,
						 write_access);
			else if (addr >= PT_F16 && addr < PT_F31 + 16)
				return access_fr(&info,
						 (addr - PT_F16)/16 + 16,
						 (addr & 8) != 0,
						 data, write_access);
			else {
				dprintk("ptrace: rejecting access to register "
					"address 0x%lx\n", addr);
				return -1;
			}
		}
	} else if (addr < PT_F9+16) {
		/* scratch state */
		switch (addr) {
		      case PT_AR_BSP:
			/*
			 * By convention, we use PT_AR_BSP to refer to
			 * the end of the user-level backing store.
			 * Use ia64_rse_skip_regs(PT_AR_BSP, -CFM.sof)
			 * to get the real value of ar.bsp at the time
			 * the kernel was entered.
			 *
			 * Furthermore, when changing the contents of
			 * PT_AR_BSP (or PT_CFM) we MUST copy any
			 * users-level stacked registers that are
			 * stored on the kernel stack back to
			 * user-space because otherwise, we might end
			 * up clobbering kernel stacked registers.
			 * Also, if this happens while the task is
			 * blocked in a system call, which convert the
			 * state such that the non-system-call exit
			 * path is used.  This ensures that the proper
			 * state will be picked up when resuming
			 * execution.  However, it *also* means that
			 * once we write PT_AR_BSP/PT_CFM, it won't be
			 * possible to modify the syscall arguments of
			 * the pending system call any longer.  This
			 * shouldn't be an issue because modifying
			 * PT_AR_BSP/PT_CFM generally implies that
			 * we're either abandoning the pending system
			 * call or that we defer it's re-execution
			 * (e.g., due to GDB doing an inferior
			 * function call).
			 */
			urbs_end = ia64_get_user_rbs_end(child, pt, &cfm);
			if (write_access) {
				if (*data != urbs_end) {
					if (ia64_sync_user_rbs(child, sw,
							       pt->ar_bspstore,
							       urbs_end) < 0)
						return -1;
					if (in_syscall(pt))
						convert_to_non_syscall(child,
								       pt,
								       cfm);
					/*
					 * Simulate user-level write
					 * of ar.bsp:
					 */
					pt->loadrs = 0;
					pt->ar_bspstore = *data;
				}
			} else
				*data = urbs_end;
			return 0;

		      case PT_CFM:
			urbs_end = ia64_get_user_rbs_end(child, pt, &cfm);
			if (write_access) {
				if (((cfm ^ *data) & PFM_MASK) != 0) {
					if (ia64_sync_user_rbs(child, sw,
							       pt->ar_bspstore,
							       urbs_end) < 0)
						return -1;
					if (in_syscall(pt))
						convert_to_non_syscall(child,
								       pt,
								       cfm);
					pt->cr_ifs = ((pt->cr_ifs & ~PFM_MASK)
						      | (*data & PFM_MASK));
				}
			} else
				*data = cfm;
			return 0;

		      case PT_CR_IPSR:
			if (write_access)
				pt->cr_ipsr = ((*data & IPSR_MASK)
					       | (pt->cr_ipsr & ~IPSR_MASK));
			else
				*data = (pt->cr_ipsr & IPSR_MASK);
			return 0;

		      case PT_AR_RSC:
			if (write_access)
				pt->ar_rsc = *data | (3 << 2); /* force PL3 */
			else
				*data = pt->ar_rsc;
			return 0;

		      case PT_AR_RNAT:
			urbs_end = ia64_get_user_rbs_end(child, pt, NULL);
			rnat_addr = (long) ia64_rse_rnat_addr((long *)
							      urbs_end);
			if (write_access)
				return ia64_poke(child, sw, urbs_end,
						 rnat_addr, *data);
			else
				return ia64_peek(child, sw, urbs_end,
						 rnat_addr, data);

		      case PT_R1:
			ptr = pt_reg_addr(pt, r1);
			break;
		      case PT_R2:  case PT_R3:
			ptr = pt_reg_addr(pt, r2) + (addr - PT_R2);
			break;
		      case PT_R8:  case PT_R9:  case PT_R10: case PT_R11:
			ptr = pt_reg_addr(pt, r8) + (addr - PT_R8);
			break;
		      case PT_R12: case PT_R13:
			ptr = pt_reg_addr(pt, r12) + (addr - PT_R12);
			break;
		      case PT_R14:
			ptr = pt_reg_addr(pt, r14);
			break;
		      case PT_R15:
			ptr = pt_reg_addr(pt, r15);
			break;
		      case PT_R16: case PT_R17: case PT_R18: case PT_R19:
		      case PT_R20: case PT_R21: case PT_R22: case PT_R23:
		      case PT_R24: case PT_R25: case PT_R26: case PT_R27:
		      case PT_R28: case PT_R29: case PT_R30: case PT_R31:
			ptr = pt_reg_addr(pt, r16) + (addr - PT_R16);
			break;
		      case PT_B0:
			ptr = pt_reg_addr(pt, b0);
			break;
		      case PT_B6:
			ptr = pt_reg_addr(pt, b6);
			break;
		      case PT_B7:
			ptr = pt_reg_addr(pt, b7);
			break;
		      case PT_F6:  case PT_F6+8: case PT_F7: case PT_F7+8:
		      case PT_F8:  case PT_F8+8: case PT_F9: case PT_F9+8:
			ptr = pt_reg_addr(pt, f6) + (addr - PT_F6);
			break;
		      case PT_AR_BSPSTORE:
			ptr = pt_reg_addr(pt, ar_bspstore);
			break;
		      case PT_AR_UNAT:
			ptr = pt_reg_addr(pt, ar_unat);
			break;
		      case PT_AR_PFS:
			ptr = pt_reg_addr(pt, ar_pfs);
			break;
		      case PT_AR_CCV:
			ptr = pt_reg_addr(pt, ar_ccv);
			break;
		      case PT_AR_FPSR:
			ptr = pt_reg_addr(pt, ar_fpsr);
			break;
		      case PT_CR_IIP:
			ptr = pt_reg_addr(pt, cr_iip);
			break;
		      case PT_PR:
			ptr = pt_reg_addr(pt, pr);
			break;
			/* scratch register */

		      default:
			/* disallow accessing anything else... */
			dprintk("ptrace: rejecting access to register "
				"address 0x%lx\n", addr);
			return -1;
		}
	} else if (addr <= PT_AR_SSD) {
		ptr = pt_reg_addr(pt, ar_csd) + (addr - PT_AR_CSD);
	} else {
		/* access debug registers */

		if (addr >= PT_IBR) {
			regnum = (addr - PT_IBR) >> 3;
			ptr = &child->thread.ibr[0];
		} else {
			regnum = (addr - PT_DBR) >> 3;
			ptr = &child->thread.dbr[0];
		}

		if (regnum >= 8) {
			dprintk("ptrace: rejecting access to register "
				"address 0x%lx\n", addr);
			return -1;
		}
#ifdef CONFIG_PERFMON
		/*
		 * Check if debug registers are used by perfmon. This
		 * test must be done once we know that we can do the
		 * operation, i.e. the arguments are all valid, but
		 * before we start modifying the state.
		 *
		 * Perfmon needs to keep a count of how many processes
		 * are trying to modify the debug registers for system
		 * wide monitoring sessions.
		 *
		 * We also include read access here, because they may
		 * cause the PMU-installed debug register state
		 * (dbr[], ibr[]) to be reset. The two arrays are also
		 * used by perfmon, but we do not use
		 * IA64_THREAD_DBG_VALID. The registers are restored
		 * by the PMU context switch code.
		 */
		if (pfm_use_debug_registers(child)) return -1;
#endif

		if (!(child->thread.flags & IA64_THREAD_DBG_VALID)) {
			child->thread.flags |= IA64_THREAD_DBG_VALID;
			memset(child->thread.dbr, 0,
			       sizeof(child->thread.dbr));
			memset(child->thread.ibr, 0,
			       sizeof(child->thread.ibr));
		}

		ptr += regnum;

		if ((regnum & 1) && write_access) {
			/* don't let the user set kernel-level breakpoints: */
			*ptr = *data & ~(7UL << 56);
			return 0;
		}
	}
	if (write_access)
		*ptr = *data;
	else
		*data = *ptr;
	return 0;
}

static long
ptrace_getregs (struct task_struct *child, struct pt_all_user_regs __user *ppr)
{
	unsigned long psr, ec, lc, rnat, bsp, cfm, nat_bits, val;
	struct unw_frame_info info;
	struct ia64_fpreg fpval;
	struct switch_stack *sw;
	struct pt_regs *pt;
	long ret, retval = 0;
	char nat = 0;
	int i;

	if (!access_ok(VERIFY_WRITE, ppr, sizeof(struct pt_all_user_regs)))
		return -EIO;

	pt = ia64_task_regs(child);
	sw = (struct switch_stack *) (child->thread.ksp + 16);
	unw_init_from_blocked_task(&info, child);
	if (unw_unwind_to_user(&info) < 0) {
		return -EIO;
	}

	if (((unsigned long) ppr & 0x7) != 0) {
		dprintk("ptrace:unaligned register address %p\n", ppr);
		return -EIO;
	}

	if (access_uarea(child, PT_CR_IPSR, &psr, 0) < 0
	    || access_uarea(child, PT_AR_EC, &ec, 0) < 0
	    || access_uarea(child, PT_AR_LC, &lc, 0) < 0
	    || access_uarea(child, PT_AR_RNAT, &rnat, 0) < 0
	    || access_uarea(child, PT_AR_BSP, &bsp, 0) < 0
	    || access_uarea(child, PT_CFM, &cfm, 0)
	    || access_uarea(child, PT_NAT_BITS, &nat_bits, 0))
		return -EIO;

	/* control regs */

	retval |= __put_user(pt->cr_iip, &ppr->cr_iip);
	retval |= __put_user(psr, &ppr->cr_ipsr);

	/* app regs */

	retval |= __put_user(pt->ar_pfs, &ppr->ar[PT_AUR_PFS]);
	retval |= __put_user(pt->ar_rsc, &ppr->ar[PT_AUR_RSC]);
	retval |= __put_user(pt->ar_bspstore, &ppr->ar[PT_AUR_BSPSTORE]);
	retval |= __put_user(pt->ar_unat, &ppr->ar[PT_AUR_UNAT]);
	retval |= __put_user(pt->ar_ccv, &ppr->ar[PT_AUR_CCV]);
	retval |= __put_user(pt->ar_fpsr, &ppr->ar[PT_AUR_FPSR]);

	retval |= __put_user(ec, &ppr->ar[PT_AUR_EC]);
	retval |= __put_user(lc, &ppr->ar[PT_AUR_LC]);
	retval |= __put_user(rnat, &ppr->ar[PT_AUR_RNAT]);
	retval |= __put_user(bsp, &ppr->ar[PT_AUR_BSP]);
	retval |= __put_user(cfm, &ppr->cfm);

	/* gr1-gr3 */

	retval |= __copy_to_user(&ppr->gr[1], &pt->r1, sizeof(long));
	retval |= __copy_to_user(&ppr->gr[2], &pt->r2, sizeof(long) *2);

	/* gr4-gr7 */

	for (i = 4; i < 8; i++) {
		if (unw_access_gr(&info, i, &val, &nat, 0) < 0)
			return -EIO;
		retval |= __put_user(val, &ppr->gr[i]);
	}

	/* gr8-gr11 */

	retval |= __copy_to_user(&ppr->gr[8], &pt->r8, sizeof(long) * 4);

	/* gr12-gr15 */

	retval |= __copy_to_user(&ppr->gr[12], &pt->r12, sizeof(long) * 2);
	retval |= __copy_to_user(&ppr->gr[14], &pt->r14, sizeof(long));
	retval |= __copy_to_user(&ppr->gr[15], &pt->r15, sizeof(long));

	/* gr16-gr31 */

	retval |= __copy_to_user(&ppr->gr[16], &pt->r16, sizeof(long) * 16);

	/* b0 */

	retval |= __put_user(pt->b0, &ppr->br[0]);

	/* b1-b5 */

	for (i = 1; i < 6; i++) {
		if (unw_access_br(&info, i, &val, 0) < 0)
			return -EIO;
		__put_user(val, &ppr->br[i]);
	}

	/* b6-b7 */

	retval |= __put_user(pt->b6, &ppr->br[6]);
	retval |= __put_user(pt->b7, &ppr->br[7]);

	/* fr2-fr5 */

	for (i = 2; i < 6; i++) {
		if (unw_get_fr(&info, i, &fpval) < 0)
			return -EIO;
		retval |= __copy_to_user(&ppr->fr[i], &fpval, sizeof (fpval));
	}

	/* fr6-fr11 */

	retval |= __copy_to_user(&ppr->fr[6], &pt->f6,
				 sizeof(struct ia64_fpreg) * 6);

	/* fp scratch regs(12-15) */

	retval |= __copy_to_user(&ppr->fr[12], &sw->f12,
				 sizeof(struct ia64_fpreg) * 4);

	/* fr16-fr31 */

	for (i = 16; i < 32; i++) {
		if (unw_get_fr(&info, i, &fpval) < 0)
			return -EIO;
		retval |= __copy_to_user(&ppr->fr[i], &fpval, sizeof (fpval));
	}

	/* fph */

	ia64_flush_fph(child);
	retval |= __copy_to_user(&ppr->fr[32], &child->thread.fph,
				 sizeof(ppr->fr[32]) * 96);

	/*  preds */

	retval |= __put_user(pt->pr, &ppr->pr);

	/* nat bits */

	retval |= __put_user(nat_bits, &ppr->nat);

	ret = retval ? -EIO : 0;
	return ret;
}

static long
ptrace_setregs (struct task_struct *child, struct pt_all_user_regs __user *ppr)
{
	unsigned long psr, rsc, ec, lc, rnat, bsp, cfm, nat_bits, val = 0;
	struct unw_frame_info info;
	struct switch_stack *sw;
	struct ia64_fpreg fpval;
	struct pt_regs *pt;
	long ret, retval = 0;
	int i;

	memset(&fpval, 0, sizeof(fpval));

	if (!access_ok(VERIFY_READ, ppr, sizeof(struct pt_all_user_regs)))
		return -EIO;

	pt = ia64_task_regs(child);
	sw = (struct switch_stack *) (child->thread.ksp + 16);
	unw_init_from_blocked_task(&info, child);
	if (unw_unwind_to_user(&info) < 0) {
		return -EIO;
	}

	if (((unsigned long) ppr & 0x7) != 0) {
		dprintk("ptrace:unaligned register address %p\n", ppr);
		return -EIO;
	}

	/* control regs */

	retval |= __get_user(pt->cr_iip, &ppr->cr_iip);
	retval |= __get_user(psr, &ppr->cr_ipsr);

	/* app regs */

	retval |= __get_user(pt->ar_pfs, &ppr->ar[PT_AUR_PFS]);
	retval |= __get_user(rsc, &ppr->ar[PT_AUR_RSC]);
	retval |= __get_user(pt->ar_bspstore, &ppr->ar[PT_AUR_BSPSTORE]);
	retval |= __get_user(pt->ar_unat, &ppr->ar[PT_AUR_UNAT]);
	retval |= __get_user(pt->ar_ccv, &ppr->ar[PT_AUR_CCV]);
	retval |= __get_user(pt->ar_fpsr, &ppr->ar[PT_AUR_FPSR]);

	retval |= __get_user(ec, &ppr->ar[PT_AUR_EC]);
	retval |= __get_user(lc, &ppr->ar[PT_AUR_LC]);
	retval |= __get_user(rnat, &ppr->ar[PT_AUR_RNAT]);
	retval |= __get_user(bsp, &ppr->ar[PT_AUR_BSP]);
	retval |= __get_user(cfm, &ppr->cfm);

	/* gr1-gr3 */

	retval |= __copy_from_user(&pt->r1, &ppr->gr[1], sizeof(long));
	retval |= __copy_from_user(&pt->r2, &ppr->gr[2], sizeof(long) * 2);

	/* gr4-gr7 */

	for (i = 4; i < 8; i++) {
		retval |= __get_user(val, &ppr->gr[i]);
		/* NaT bit will be set via PT_NAT_BITS: */
		if (unw_set_gr(&info, i, val, 0) < 0)
			return -EIO;
	}

	/* gr8-gr11 */

	retval |= __copy_from_user(&pt->r8, &ppr->gr[8], sizeof(long) * 4);

	/* gr12-gr15 */

	retval |= __copy_from_user(&pt->r12, &ppr->gr[12], sizeof(long) * 2);
	retval |= __copy_from_user(&pt->r14, &ppr->gr[14], sizeof(long));
	retval |= __copy_from_user(&pt->r15, &ppr->gr[15], sizeof(long));

	/* gr16-gr31 */

	retval |= __copy_from_user(&pt->r16, &ppr->gr[16], sizeof(long) * 16);

	/* b0 */

	retval |= __get_user(pt->b0, &ppr->br[0]);

	/* b1-b5 */

	for (i = 1; i < 6; i++) {
		retval |= __get_user(val, &ppr->br[i]);
		unw_set_br(&info, i, val);
	}

	/* b6-b7 */

	retval |= __get_user(pt->b6, &ppr->br[6]);
	retval |= __get_user(pt->b7, &ppr->br[7]);

	/* fr2-fr5 */

	for (i = 2; i < 6; i++) {
		retval |= __copy_from_user(&fpval, &ppr->fr[i], sizeof(fpval));
		if (unw_set_fr(&info, i, fpval) < 0)
			return -EIO;
	}

	/* fr6-fr11 */

	retval |= __copy_from_user(&pt->f6, &ppr->fr[6],
				   sizeof(ppr->fr[6]) * 6);

	/* fp scratch regs(12-15) */

	retval |= __copy_from_user(&sw->f12, &ppr->fr[12],
				   sizeof(ppr->fr[12]) * 4);

	/* fr16-fr31 */

	for (i = 16; i < 32; i++) {
		retval |= __copy_from_user(&fpval, &ppr->fr[i],
					   sizeof(fpval));
		if (unw_set_fr(&info, i, fpval) < 0)
			return -EIO;
	}

	/* fph */

	ia64_sync_fph(child);
	retval |= __copy_from_user(&child->thread.fph, &ppr->fr[32],
				   sizeof(ppr->fr[32]) * 96);

	/* preds */

	retval |= __get_user(pt->pr, &ppr->pr);

	/* nat bits */

	retval |= __get_user(nat_bits, &ppr->nat);

	retval |= access_uarea(child, PT_CR_IPSR, &psr, 1);
	retval |= access_uarea(child, PT_AR_RSC, &rsc, 1);
	retval |= access_uarea(child, PT_AR_EC, &ec, 1);
	retval |= access_uarea(child, PT_AR_LC, &lc, 1);
	retval |= access_uarea(child, PT_AR_RNAT, &rnat, 1);
	retval |= access_uarea(child, PT_AR_BSP, &bsp, 1);
	retval |= access_uarea(child, PT_CFM, &cfm, 1);
	retval |= access_uarea(child, PT_NAT_BITS, &nat_bits, 1);

	ret = retval ? -EIO : 0;
	return ret;
}

/*
 * Called by kernel/ptrace.c when detaching..
 *
 * Make sure the single step bit is not set.
 */
void
ptrace_disable (struct task_struct *child)
{
	struct ia64_psr *child_psr = ia64_psr(ia64_task_regs(child));

	/* make sure the single step/taken-branch trap bits are not set: */
	child_psr->ss = 0;
	child_psr->tb = 0;
}

asmlinkage long
sys_ptrace (long request, pid_t pid, unsigned long addr, unsigned long data)
{
	struct pt_regs *pt;
	unsigned long urbs_end, peek_or_poke;
	struct task_struct *child;
	struct switch_stack *sw;
	long ret;

	lock_kernel();
	ret = -EPERM;
	if (request == PTRACE_TRACEME) {
		/* are we already being traced? */
		if (current->ptrace & PT_PTRACED)
			goto out;
		ret = security_ptrace(current->parent, current);
		if (ret)
			goto out;
		current->ptrace |= PT_PTRACED;
		ret = 0;
		goto out;
	}

	peek_or_poke = (request == PTRACE_PEEKTEXT
			|| request == PTRACE_PEEKDATA
			|| request == PTRACE_POKETEXT
			|| request == PTRACE_POKEDATA);
	ret = -ESRCH;
	read_lock(&tasklist_lock);
	{
		child = find_task_by_pid(pid);
		if (child) {
			if (peek_or_poke)
				child = find_thread_for_addr(child, addr);
			get_task_struct(child);
		}
	}
	read_unlock(&tasklist_lock);
	if (!child)
		goto out;
	ret = -EPERM;
	if (pid == 1)		/* no messing around with init! */
		goto out_tsk;

	if (request == PTRACE_ATTACH) {
		ret = ptrace_attach(child);
		goto out_tsk;
	}

	ret = ptrace_check_attach(child, request == PTRACE_KILL);
	if (ret < 0)
		goto out_tsk;

	pt = ia64_task_regs(child);
	sw = (struct switch_stack *) (child->thread.ksp + 16);

	switch (request) {
	      case PTRACE_PEEKTEXT:
	      case PTRACE_PEEKDATA:
		/* read word at location addr */
		urbs_end = ia64_get_user_rbs_end(child, pt, NULL);
		ret = ia64_peek(child, sw, urbs_end, addr, &data);
		if (ret == 0) {
			ret = data;
			/* ensure "ret" is not mistaken as an error code: */
			force_successful_syscall_return();
		}
		goto out_tsk;

	      case PTRACE_POKETEXT:
	      case PTRACE_POKEDATA:
		/* write the word at location addr */
		urbs_end = ia64_get_user_rbs_end(child, pt, NULL);
		ret = ia64_poke(child, sw, urbs_end, addr, data);
		goto out_tsk;

	      case PTRACE_PEEKUSR:
		/* read the word at addr in the USER area */
		if (access_uarea(child, addr, &data, 0) < 0) {
			ret = -EIO;
			goto out_tsk;
		}
		ret = data;
		/* ensure "ret" is not mistaken as an error code */
		force_successful_syscall_return();
		goto out_tsk;

	      case PTRACE_POKEUSR:
		/* write the word at addr in the USER area */
		if (access_uarea(child, addr, &data, 1) < 0) {
			ret = -EIO;
			goto out_tsk;
		}
		ret = 0;
		goto out_tsk;

	      case PTRACE_OLD_GETSIGINFO:
		/* for backwards-compatibility */
		ret = ptrace_request(child, PTRACE_GETSIGINFO, addr, data);
		goto out_tsk;

	      case PTRACE_OLD_SETSIGINFO:
		/* for backwards-compatibility */
		ret = ptrace_request(child, PTRACE_SETSIGINFO, addr, data);
		goto out_tsk;

	      case PTRACE_SYSCALL:
		/* continue and stop at next (return from) syscall */
	      case PTRACE_CONT:
		/* restart after signal. */
		ret = -EIO;
		if (!valid_signal(data))
			goto out_tsk;
		if (request == PTRACE_SYSCALL)
			set_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
		else
			clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
		child->exit_code = data;

		/*
		 * Make sure the single step/taken-branch trap bits
		 * are not set:
		 */
		ia64_psr(pt)->ss = 0;
		ia64_psr(pt)->tb = 0;

		wake_up_process(child);
		ret = 0;
		goto out_tsk;

	      case PTRACE_KILL:
		/*
		 * Make the child exit.  Best I can do is send it a
		 * sigkill.  Perhaps it should be put in the status
		 * that it wants to exit.
		 */
		if (child->exit_state == EXIT_ZOMBIE)
			/* already dead */
			goto out_tsk;
		child->exit_code = SIGKILL;

		ptrace_disable(child);
		wake_up_process(child);
		ret = 0;
		goto out_tsk;

	      case PTRACE_SINGLESTEP:
		/* let child execute for one instruction */
	      case PTRACE_SINGLEBLOCK:
		ret = -EIO;
		if (!valid_signal(data))
			goto out_tsk;

		clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
		if (request == PTRACE_SINGLESTEP) {
			ia64_psr(pt)->ss = 1;
		} else {
			ia64_psr(pt)->tb = 1;
		}
		child->exit_code = data;

		/* give it a chance to run. */
		wake_up_process(child);
		ret = 0;
		goto out_tsk;

	      case PTRACE_DETACH:
		/* detach a process that was attached. */
		ret = ptrace_detach(child, data);
		goto out_tsk;

	      case PTRACE_GETREGS:
		ret = ptrace_getregs(child,
				     (struct pt_all_user_regs __user *) data);
		goto out_tsk;

	      case PTRACE_SETREGS:
		ret = ptrace_setregs(child,
				     (struct pt_all_user_regs __user *) data);
		goto out_tsk;

	      default:
		ret = ptrace_request(child, request, addr, data);
		goto out_tsk;
	}
  out_tsk:
	put_task_struct(child);
  out:
	unlock_kernel();
	return ret;
}


void
syscall_trace (void)
{
	if (!test_thread_flag(TIF_SYSCALL_TRACE))
		return;
	if (!(current->ptrace & PT_PTRACED))
		return;
	/*
	 * The 0x80 provides a way for the tracing parent to
	 * distinguish between a syscall stop and SIGTRAP delivery.
	 */
	ptrace_notify(SIGTRAP
		      | ((current->ptrace & PT_TRACESYSGOOD) ? 0x80 : 0));

	/*
	 * This isn't the same as continuing with a signal, but it
	 * will do for normal use.  strace only continues with a
	 * signal if the stopping signal is not SIGTRAP.  -brl
	 */
	if (current->exit_code) {
		send_sig(current->exit_code, current, 1);
		current->exit_code = 0;
	}
}

/* "asmlinkage" so the input arguments are preserved... */

asmlinkage void
syscall_trace_enter (long arg0, long arg1, long arg2, long arg3,
		     long arg4, long arg5, long arg6, long arg7,
		     struct pt_regs regs)
{
	if (test_thread_flag(TIF_SYSCALL_TRACE) 
	    && (current->ptrace & PT_PTRACED))
		syscall_trace();

	if (unlikely(current->audit_context)) {
		long syscall;
		int arch;

		if (IS_IA32_PROCESS(&regs)) {
			syscall = regs.r1;
			arch = AUDIT_ARCH_I386;
		} else {
			syscall = regs.r15;
			arch = AUDIT_ARCH_IA64;
		}

		audit_syscall_entry(current, arch, syscall, arg0, arg1, arg2, arg3);
	}

}

/* "asmlinkage" so the input arguments are preserved... */

asmlinkage void
syscall_trace_leave (long arg0, long arg1, long arg2, long arg3,
		     long arg4, long arg5, long arg6, long arg7,
		     struct pt_regs regs)
{
	if (unlikely(current->audit_context))
		audit_syscall_exit(current, AUDITSC_RESULT(regs.r10), regs.r8);

	if (test_thread_flag(TIF_SYSCALL_TRACE)
	    && (current->ptrace & PT_PTRACED))
		syscall_trace();
}