xref: /openbmc/linux/arch/ia64/kernel/ptrace.c (revision 2c684d89)
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/uaccess.h>
30 #include <asm/unwind.h>
31 #ifdef CONFIG_PERFMON
32 #include <asm/perfmon.h>
33 #endif
34 
35 #include "entry.h"
36 
37 /*
38  * Bits in the PSR that we allow ptrace() to change:
39  *	be, up, ac, mfl, mfh (the user mask; five bits total)
40  *	db (debug breakpoint fault; one bit)
41  *	id (instruction debug fault disable; one bit)
42  *	dd (data debug fault disable; one bit)
43  *	ri (restart instruction; two bits)
44  *	is (instruction set; one bit)
45  */
46 #define IPSR_MASK (IA64_PSR_UM | IA64_PSR_DB | IA64_PSR_IS	\
47 		   | IA64_PSR_ID | IA64_PSR_DD | IA64_PSR_RI)
48 
49 #define MASK(nbits)	((1UL << (nbits)) - 1)	/* mask with NBITS bits set */
50 #define PFM_MASK	MASK(38)
51 
52 #define PTRACE_DEBUG	0
53 
54 #if PTRACE_DEBUG
55 # define dprintk(format...)	printk(format)
56 # define inline
57 #else
58 # define dprintk(format...)
59 #endif
60 
61 /* Return TRUE if PT was created due to kernel-entry via a system-call.  */
62 
63 static inline int
64 in_syscall (struct pt_regs *pt)
65 {
66 	return (long) pt->cr_ifs >= 0;
67 }
68 
69 /*
70  * Collect the NaT bits for r1-r31 from scratch_unat and return a NaT
71  * bitset where bit i is set iff the NaT bit of register i is set.
72  */
73 unsigned long
74 ia64_get_scratch_nat_bits (struct pt_regs *pt, unsigned long scratch_unat)
75 {
76 #	define GET_BITS(first, last, unat)				\
77 	({								\
78 		unsigned long bit = ia64_unat_pos(&pt->r##first);	\
79 		unsigned long nbits = (last - first + 1);		\
80 		unsigned long mask = MASK(nbits) << first;		\
81 		unsigned long dist;					\
82 		if (bit < first)					\
83 			dist = 64 + bit - first;			\
84 		else							\
85 			dist = bit - first;				\
86 		ia64_rotr(unat, dist) & mask;				\
87 	})
88 	unsigned long val;
89 
90 	/*
91 	 * Registers that are stored consecutively in struct pt_regs
92 	 * can be handled in parallel.  If the register order in
93 	 * struct_pt_regs changes, this code MUST be updated.
94 	 */
95 	val  = GET_BITS( 1,  1, scratch_unat);
96 	val |= GET_BITS( 2,  3, scratch_unat);
97 	val |= GET_BITS(12, 13, scratch_unat);
98 	val |= GET_BITS(14, 14, scratch_unat);
99 	val |= GET_BITS(15, 15, scratch_unat);
100 	val |= GET_BITS( 8, 11, scratch_unat);
101 	val |= GET_BITS(16, 31, scratch_unat);
102 	return val;
103 
104 #	undef GET_BITS
105 }
106 
107 /*
108  * Set the NaT bits for the scratch registers according to NAT and
109  * return the resulting unat (assuming the scratch registers are
110  * stored in PT).
111  */
112 unsigned long
113 ia64_put_scratch_nat_bits (struct pt_regs *pt, unsigned long nat)
114 {
115 #	define PUT_BITS(first, last, nat)				\
116 	({								\
117 		unsigned long bit = ia64_unat_pos(&pt->r##first);	\
118 		unsigned long nbits = (last - first + 1);		\
119 		unsigned long mask = MASK(nbits) << first;		\
120 		long dist;						\
121 		if (bit < first)					\
122 			dist = 64 + bit - first;			\
123 		else							\
124 			dist = bit - first;				\
125 		ia64_rotl(nat & mask, dist);				\
126 	})
127 	unsigned long scratch_unat;
128 
129 	/*
130 	 * Registers that are stored consecutively in struct pt_regs
131 	 * can be handled in parallel.  If the register order in
132 	 * struct_pt_regs changes, this code MUST be updated.
133 	 */
134 	scratch_unat  = PUT_BITS( 1,  1, nat);
135 	scratch_unat |= PUT_BITS( 2,  3, nat);
136 	scratch_unat |= PUT_BITS(12, 13, nat);
137 	scratch_unat |= PUT_BITS(14, 14, nat);
138 	scratch_unat |= PUT_BITS(15, 15, nat);
139 	scratch_unat |= PUT_BITS( 8, 11, nat);
140 	scratch_unat |= PUT_BITS(16, 31, nat);
141 
142 	return scratch_unat;
143 
144 #	undef PUT_BITS
145 }
146 
147 #define IA64_MLX_TEMPLATE	0x2
148 #define IA64_MOVL_OPCODE	6
149 
150 void
151 ia64_increment_ip (struct pt_regs *regs)
152 {
153 	unsigned long w0, ri = ia64_psr(regs)->ri + 1;
154 
155 	if (ri > 2) {
156 		ri = 0;
157 		regs->cr_iip += 16;
158 	} else if (ri == 2) {
159 		get_user(w0, (char __user *) regs->cr_iip + 0);
160 		if (((w0 >> 1) & 0xf) == IA64_MLX_TEMPLATE) {
161 			/*
162 			 * rfi'ing to slot 2 of an MLX bundle causes
163 			 * an illegal operation fault.  We don't want
164 			 * that to happen...
165 			 */
166 			ri = 0;
167 			regs->cr_iip += 16;
168 		}
169 	}
170 	ia64_psr(regs)->ri = ri;
171 }
172 
173 void
174 ia64_decrement_ip (struct pt_regs *regs)
175 {
176 	unsigned long w0, ri = ia64_psr(regs)->ri - 1;
177 
178 	if (ia64_psr(regs)->ri == 0) {
179 		regs->cr_iip -= 16;
180 		ri = 2;
181 		get_user(w0, (char __user *) regs->cr_iip + 0);
182 		if (((w0 >> 1) & 0xf) == IA64_MLX_TEMPLATE) {
183 			/*
184 			 * rfi'ing to slot 2 of an MLX bundle causes
185 			 * an illegal operation fault.  We don't want
186 			 * that to happen...
187 			 */
188 			ri = 1;
189 		}
190 	}
191 	ia64_psr(regs)->ri = ri;
192 }
193 
194 /*
195  * This routine is used to read an rnat bits that are stored on the
196  * kernel backing store.  Since, in general, the alignment of the user
197  * and kernel are different, this is not completely trivial.  In
198  * essence, we need to construct the user RNAT based on up to two
199  * kernel RNAT values and/or the RNAT value saved in the child's
200  * pt_regs.
201  *
202  * user rbs
203  *
204  * +--------+ <-- lowest address
205  * | slot62 |
206  * +--------+
207  * |  rnat  | 0x....1f8
208  * +--------+
209  * | slot00 | \
210  * +--------+ |
211  * | slot01 | > child_regs->ar_rnat
212  * +--------+ |
213  * | slot02 | /				kernel rbs
214  * +--------+				+--------+
215  *	    <- child_regs->ar_bspstore	| slot61 | <-- krbs
216  * +- - - - +				+--------+
217  *					| slot62 |
218  * +- - - - +				+--------+
219  *					|  rnat	 |
220  * +- - - - +				+--------+
221  *   vrnat				| slot00 |
222  * +- - - - +				+--------+
223  *					=	 =
224  *					+--------+
225  *					| slot00 | \
226  *					+--------+ |
227  *					| slot01 | > child_stack->ar_rnat
228  *					+--------+ |
229  *					| slot02 | /
230  *					+--------+
231  *						  <--- child_stack->ar_bspstore
232  *
233  * The way to think of this code is as follows: bit 0 in the user rnat
234  * corresponds to some bit N (0 <= N <= 62) in one of the kernel rnat
235  * value.  The kernel rnat value holding this bit is stored in
236  * variable rnat0.  rnat1 is loaded with the kernel rnat value that
237  * form the upper bits of the user rnat value.
238  *
239  * Boundary cases:
240  *
241  * o when reading the rnat "below" the first rnat slot on the kernel
242  *   backing store, rnat0/rnat1 are set to 0 and the low order bits are
243  *   merged in from pt->ar_rnat.
244  *
245  * o when reading the rnat "above" the last rnat slot on the kernel
246  *   backing store, rnat0/rnat1 gets its value from sw->ar_rnat.
247  */
248 static unsigned long
249 get_rnat (struct task_struct *task, struct switch_stack *sw,
250 	  unsigned long *krbs, unsigned long *urnat_addr,
251 	  unsigned long *urbs_end)
252 {
253 	unsigned long rnat0 = 0, rnat1 = 0, urnat = 0, *slot0_kaddr;
254 	unsigned long umask = 0, mask, m;
255 	unsigned long *kbsp, *ubspstore, *rnat0_kaddr, *rnat1_kaddr, shift;
256 	long num_regs, nbits;
257 	struct pt_regs *pt;
258 
259 	pt = task_pt_regs(task);
260 	kbsp = (unsigned long *) sw->ar_bspstore;
261 	ubspstore = (unsigned long *) pt->ar_bspstore;
262 
263 	if (urbs_end < urnat_addr)
264 		nbits = ia64_rse_num_regs(urnat_addr - 63, urbs_end);
265 	else
266 		nbits = 63;
267 	mask = MASK(nbits);
268 	/*
269 	 * First, figure out which bit number slot 0 in user-land maps
270 	 * to in the kernel rnat.  Do this by figuring out how many
271 	 * register slots we're beyond the user's backingstore and
272 	 * then computing the equivalent address in kernel space.
273 	 */
274 	num_regs = ia64_rse_num_regs(ubspstore, urnat_addr + 1);
275 	slot0_kaddr = ia64_rse_skip_regs(krbs, num_regs);
276 	shift = ia64_rse_slot_num(slot0_kaddr);
277 	rnat1_kaddr = ia64_rse_rnat_addr(slot0_kaddr);
278 	rnat0_kaddr = rnat1_kaddr - 64;
279 
280 	if (ubspstore + 63 > urnat_addr) {
281 		/* some bits need to be merged in from pt->ar_rnat */
282 		umask = MASK(ia64_rse_slot_num(ubspstore)) & mask;
283 		urnat = (pt->ar_rnat & umask);
284 		mask &= ~umask;
285 		if (!mask)
286 			return urnat;
287 	}
288 
289 	m = mask << shift;
290 	if (rnat0_kaddr >= kbsp)
291 		rnat0 = sw->ar_rnat;
292 	else if (rnat0_kaddr > krbs)
293 		rnat0 = *rnat0_kaddr;
294 	urnat |= (rnat0 & m) >> shift;
295 
296 	m = mask >> (63 - shift);
297 	if (rnat1_kaddr >= kbsp)
298 		rnat1 = sw->ar_rnat;
299 	else if (rnat1_kaddr > krbs)
300 		rnat1 = *rnat1_kaddr;
301 	urnat |= (rnat1 & m) << (63 - shift);
302 	return urnat;
303 }
304 
305 /*
306  * The reverse of get_rnat.
307  */
308 static void
309 put_rnat (struct task_struct *task, struct switch_stack *sw,
310 	  unsigned long *krbs, unsigned long *urnat_addr, unsigned long urnat,
311 	  unsigned long *urbs_end)
312 {
313 	unsigned long rnat0 = 0, rnat1 = 0, *slot0_kaddr, umask = 0, mask, m;
314 	unsigned long *kbsp, *ubspstore, *rnat0_kaddr, *rnat1_kaddr, shift;
315 	long num_regs, nbits;
316 	struct pt_regs *pt;
317 	unsigned long cfm, *urbs_kargs;
318 
319 	pt = task_pt_regs(task);
320 	kbsp = (unsigned long *) sw->ar_bspstore;
321 	ubspstore = (unsigned long *) pt->ar_bspstore;
322 
323 	urbs_kargs = urbs_end;
324 	if (in_syscall(pt)) {
325 		/*
326 		 * If entered via syscall, don't allow user to set rnat bits
327 		 * for syscall args.
328 		 */
329 		cfm = pt->cr_ifs;
330 		urbs_kargs = ia64_rse_skip_regs(urbs_end, -(cfm & 0x7f));
331 	}
332 
333 	if (urbs_kargs >= urnat_addr)
334 		nbits = 63;
335 	else {
336 		if ((urnat_addr - 63) >= urbs_kargs)
337 			return;
338 		nbits = ia64_rse_num_regs(urnat_addr - 63, urbs_kargs);
339 	}
340 	mask = MASK(nbits);
341 
342 	/*
343 	 * First, figure out which bit number slot 0 in user-land maps
344 	 * to in the kernel rnat.  Do this by figuring out how many
345 	 * register slots we're beyond the user's backingstore and
346 	 * then computing the equivalent address in kernel space.
347 	 */
348 	num_regs = ia64_rse_num_regs(ubspstore, urnat_addr + 1);
349 	slot0_kaddr = ia64_rse_skip_regs(krbs, num_regs);
350 	shift = ia64_rse_slot_num(slot0_kaddr);
351 	rnat1_kaddr = ia64_rse_rnat_addr(slot0_kaddr);
352 	rnat0_kaddr = rnat1_kaddr - 64;
353 
354 	if (ubspstore + 63 > urnat_addr) {
355 		/* some bits need to be place in pt->ar_rnat: */
356 		umask = MASK(ia64_rse_slot_num(ubspstore)) & mask;
357 		pt->ar_rnat = (pt->ar_rnat & ~umask) | (urnat & umask);
358 		mask &= ~umask;
359 		if (!mask)
360 			return;
361 	}
362 	/*
363 	 * Note: Section 11.1 of the EAS guarantees that bit 63 of an
364 	 * rnat slot is ignored. so we don't have to clear it here.
365 	 */
366 	rnat0 = (urnat << shift);
367 	m = mask << shift;
368 	if (rnat0_kaddr >= kbsp)
369 		sw->ar_rnat = (sw->ar_rnat & ~m) | (rnat0 & m);
370 	else if (rnat0_kaddr > krbs)
371 		*rnat0_kaddr = ((*rnat0_kaddr & ~m) | (rnat0 & m));
372 
373 	rnat1 = (urnat >> (63 - shift));
374 	m = mask >> (63 - shift);
375 	if (rnat1_kaddr >= kbsp)
376 		sw->ar_rnat = (sw->ar_rnat & ~m) | (rnat1 & m);
377 	else if (rnat1_kaddr > krbs)
378 		*rnat1_kaddr = ((*rnat1_kaddr & ~m) | (rnat1 & m));
379 }
380 
381 static inline int
382 on_kernel_rbs (unsigned long addr, unsigned long bspstore,
383 	       unsigned long urbs_end)
384 {
385 	unsigned long *rnat_addr = ia64_rse_rnat_addr((unsigned long *)
386 						      urbs_end);
387 	return (addr >= bspstore && addr <= (unsigned long) rnat_addr);
388 }
389 
390 /*
391  * Read a word from the user-level backing store of task CHILD.  ADDR
392  * is the user-level address to read the word from, VAL a pointer to
393  * the return value, and USER_BSP gives the end of the user-level
394  * backing store (i.e., it's the address that would be in ar.bsp after
395  * the user executed a "cover" instruction).
396  *
397  * This routine takes care of accessing the kernel register backing
398  * store for those registers that got spilled there.  It also takes
399  * care of calculating the appropriate RNaT collection words.
400  */
401 long
402 ia64_peek (struct task_struct *child, struct switch_stack *child_stack,
403 	   unsigned long user_rbs_end, unsigned long addr, long *val)
404 {
405 	unsigned long *bspstore, *krbs, regnum, *laddr, *urbs_end, *rnat_addr;
406 	struct pt_regs *child_regs;
407 	size_t copied;
408 	long ret;
409 
410 	urbs_end = (long *) user_rbs_end;
411 	laddr = (unsigned long *) addr;
412 	child_regs = task_pt_regs(child);
413 	bspstore = (unsigned long *) child_regs->ar_bspstore;
414 	krbs = (unsigned long *) child + IA64_RBS_OFFSET/8;
415 	if (on_kernel_rbs(addr, (unsigned long) bspstore,
416 			  (unsigned long) urbs_end))
417 	{
418 		/*
419 		 * Attempt to read the RBS in an area that's actually
420 		 * on the kernel RBS => read the corresponding bits in
421 		 * the kernel RBS.
422 		 */
423 		rnat_addr = ia64_rse_rnat_addr(laddr);
424 		ret = get_rnat(child, child_stack, krbs, rnat_addr, urbs_end);
425 
426 		if (laddr == rnat_addr) {
427 			/* return NaT collection word itself */
428 			*val = ret;
429 			return 0;
430 		}
431 
432 		if (((1UL << ia64_rse_slot_num(laddr)) & ret) != 0) {
433 			/*
434 			 * It is implementation dependent whether the
435 			 * data portion of a NaT value gets saved on a
436 			 * st8.spill or RSE spill (e.g., see EAS 2.6,
437 			 * 4.4.4.6 Register Spill and Fill).  To get
438 			 * consistent behavior across all possible
439 			 * IA-64 implementations, we return zero in
440 			 * this case.
441 			 */
442 			*val = 0;
443 			return 0;
444 		}
445 
446 		if (laddr < urbs_end) {
447 			/*
448 			 * The desired word is on the kernel RBS and
449 			 * is not a NaT.
450 			 */
451 			regnum = ia64_rse_num_regs(bspstore, laddr);
452 			*val = *ia64_rse_skip_regs(krbs, regnum);
453 			return 0;
454 		}
455 	}
456 	copied = access_process_vm(child, addr, &ret, sizeof(ret), 0);
457 	if (copied != sizeof(ret))
458 		return -EIO;
459 	*val = ret;
460 	return 0;
461 }
462 
463 long
464 ia64_poke (struct task_struct *child, struct switch_stack *child_stack,
465 	   unsigned long user_rbs_end, unsigned long addr, long val)
466 {
467 	unsigned long *bspstore, *krbs, regnum, *laddr;
468 	unsigned long *urbs_end = (long *) user_rbs_end;
469 	struct pt_regs *child_regs;
470 
471 	laddr = (unsigned long *) addr;
472 	child_regs = task_pt_regs(child);
473 	bspstore = (unsigned long *) child_regs->ar_bspstore;
474 	krbs = (unsigned long *) child + IA64_RBS_OFFSET/8;
475 	if (on_kernel_rbs(addr, (unsigned long) bspstore,
476 			  (unsigned long) urbs_end))
477 	{
478 		/*
479 		 * Attempt to write the RBS in an area that's actually
480 		 * on the kernel RBS => write the corresponding bits
481 		 * in the kernel RBS.
482 		 */
483 		if (ia64_rse_is_rnat_slot(laddr))
484 			put_rnat(child, child_stack, krbs, laddr, val,
485 				 urbs_end);
486 		else {
487 			if (laddr < urbs_end) {
488 				regnum = ia64_rse_num_regs(bspstore, laddr);
489 				*ia64_rse_skip_regs(krbs, regnum) = val;
490 			}
491 		}
492 	} else if (access_process_vm(child, addr, &val, sizeof(val), 1)
493 		   != sizeof(val))
494 		return -EIO;
495 	return 0;
496 }
497 
498 /*
499  * Calculate the address of the end of the user-level register backing
500  * store.  This is the address that would have been stored in ar.bsp
501  * if the user had executed a "cover" instruction right before
502  * entering the kernel.  If CFMP is not NULL, it is used to return the
503  * "current frame mask" that was active at the time the kernel was
504  * entered.
505  */
506 unsigned long
507 ia64_get_user_rbs_end (struct task_struct *child, struct pt_regs *pt,
508 		       unsigned long *cfmp)
509 {
510 	unsigned long *krbs, *bspstore, cfm = pt->cr_ifs;
511 	long ndirty;
512 
513 	krbs = (unsigned long *) child + IA64_RBS_OFFSET/8;
514 	bspstore = (unsigned long *) pt->ar_bspstore;
515 	ndirty = ia64_rse_num_regs(krbs, krbs + (pt->loadrs >> 19));
516 
517 	if (in_syscall(pt))
518 		ndirty += (cfm & 0x7f);
519 	else
520 		cfm &= ~(1UL << 63);	/* clear valid bit */
521 
522 	if (cfmp)
523 		*cfmp = cfm;
524 	return (unsigned long) ia64_rse_skip_regs(bspstore, ndirty);
525 }
526 
527 /*
528  * Synchronize (i.e, write) the RSE backing store living in kernel
529  * space to the VM of the CHILD task.  SW and PT are the pointers to
530  * the switch_stack and pt_regs structures, respectively.
531  * USER_RBS_END is the user-level address at which the backing store
532  * ends.
533  */
534 long
535 ia64_sync_user_rbs (struct task_struct *child, struct switch_stack *sw,
536 		    unsigned long user_rbs_start, unsigned long user_rbs_end)
537 {
538 	unsigned long addr, val;
539 	long ret;
540 
541 	/* now copy word for word from kernel rbs to user rbs: */
542 	for (addr = user_rbs_start; addr < user_rbs_end; addr += 8) {
543 		ret = ia64_peek(child, sw, user_rbs_end, addr, &val);
544 		if (ret < 0)
545 			return ret;
546 		if (access_process_vm(child, addr, &val, sizeof(val), 1)
547 		    != sizeof(val))
548 			return -EIO;
549 	}
550 	return 0;
551 }
552 
553 static long
554 ia64_sync_kernel_rbs (struct task_struct *child, struct switch_stack *sw,
555 		unsigned long user_rbs_start, unsigned long user_rbs_end)
556 {
557 	unsigned long addr, val;
558 	long ret;
559 
560 	/* now copy word for word from user rbs to kernel rbs: */
561 	for (addr = user_rbs_start; addr < user_rbs_end; addr += 8) {
562 		if (access_process_vm(child, addr, &val, sizeof(val), 0)
563 				!= sizeof(val))
564 			return -EIO;
565 
566 		ret = ia64_poke(child, sw, user_rbs_end, addr, val);
567 		if (ret < 0)
568 			return ret;
569 	}
570 	return 0;
571 }
572 
573 typedef long (*syncfunc_t)(struct task_struct *, struct switch_stack *,
574 			    unsigned long, unsigned long);
575 
576 static void do_sync_rbs(struct unw_frame_info *info, void *arg)
577 {
578 	struct pt_regs *pt;
579 	unsigned long urbs_end;
580 	syncfunc_t fn = arg;
581 
582 	if (unw_unwind_to_user(info) < 0)
583 		return;
584 	pt = task_pt_regs(info->task);
585 	urbs_end = ia64_get_user_rbs_end(info->task, pt, NULL);
586 
587 	fn(info->task, info->sw, pt->ar_bspstore, urbs_end);
588 }
589 
590 /*
591  * when a thread is stopped (ptraced), debugger might change thread's user
592  * stack (change memory directly), and we must avoid the RSE stored in kernel
593  * to override user stack (user space's RSE is newer than kernel's in the
594  * case). To workaround the issue, we copy kernel RSE to user RSE before the
595  * task is stopped, so user RSE has updated data.  we then copy user RSE to
596  * kernel after the task is resummed from traced stop and kernel will use the
597  * newer RSE to return to user. TIF_RESTORE_RSE is the flag to indicate we need
598  * synchronize user RSE to kernel.
599  */
600 void ia64_ptrace_stop(void)
601 {
602 	if (test_and_set_tsk_thread_flag(current, TIF_RESTORE_RSE))
603 		return;
604 	set_notify_resume(current);
605 	unw_init_running(do_sync_rbs, ia64_sync_user_rbs);
606 }
607 
608 /*
609  * This is called to read back the register backing store.
610  */
611 void ia64_sync_krbs(void)
612 {
613 	clear_tsk_thread_flag(current, TIF_RESTORE_RSE);
614 
615 	unw_init_running(do_sync_rbs, ia64_sync_kernel_rbs);
616 }
617 
618 /*
619  * After PTRACE_ATTACH, a thread's register backing store area in user
620  * space is assumed to contain correct data whenever the thread is
621  * stopped.  arch_ptrace_stop takes care of this on tracing stops.
622  * But if the child was already stopped for job control when we attach
623  * to it, then it might not ever get into ptrace_stop by the time we
624  * want to examine the user memory containing the RBS.
625  */
626 void
627 ptrace_attach_sync_user_rbs (struct task_struct *child)
628 {
629 	int stopped = 0;
630 	struct unw_frame_info info;
631 
632 	/*
633 	 * If the child is in TASK_STOPPED, we need to change that to
634 	 * TASK_TRACED momentarily while we operate on it.  This ensures
635 	 * that the child won't be woken up and return to user mode while
636 	 * we are doing the sync.  (It can only be woken up for SIGKILL.)
637 	 */
638 
639 	read_lock(&tasklist_lock);
640 	if (child->sighand) {
641 		spin_lock_irq(&child->sighand->siglock);
642 		if (child->state == TASK_STOPPED &&
643 		    !test_and_set_tsk_thread_flag(child, TIF_RESTORE_RSE)) {
644 			set_notify_resume(child);
645 
646 			child->state = TASK_TRACED;
647 			stopped = 1;
648 		}
649 		spin_unlock_irq(&child->sighand->siglock);
650 	}
651 	read_unlock(&tasklist_lock);
652 
653 	if (!stopped)
654 		return;
655 
656 	unw_init_from_blocked_task(&info, child);
657 	do_sync_rbs(&info, ia64_sync_user_rbs);
658 
659 	/*
660 	 * Now move the child back into TASK_STOPPED if it should be in a
661 	 * job control stop, so that SIGCONT can be used to wake it up.
662 	 */
663 	read_lock(&tasklist_lock);
664 	if (child->sighand) {
665 		spin_lock_irq(&child->sighand->siglock);
666 		if (child->state == TASK_TRACED &&
667 		    (child->signal->flags & SIGNAL_STOP_STOPPED)) {
668 			child->state = TASK_STOPPED;
669 		}
670 		spin_unlock_irq(&child->sighand->siglock);
671 	}
672 	read_unlock(&tasklist_lock);
673 }
674 
675 /*
676  * Write f32-f127 back to task->thread.fph if it has been modified.
677  */
678 inline void
679 ia64_flush_fph (struct task_struct *task)
680 {
681 	struct ia64_psr *psr = ia64_psr(task_pt_regs(task));
682 
683 	/*
684 	 * Prevent migrating this task while
685 	 * we're fiddling with the FPU state
686 	 */
687 	preempt_disable();
688 	if (ia64_is_local_fpu_owner(task) && psr->mfh) {
689 		psr->mfh = 0;
690 		task->thread.flags |= IA64_THREAD_FPH_VALID;
691 		ia64_save_fpu(&task->thread.fph[0]);
692 	}
693 	preempt_enable();
694 }
695 
696 /*
697  * Sync the fph state of the task so that it can be manipulated
698  * through thread.fph.  If necessary, f32-f127 are written back to
699  * thread.fph or, if the fph state hasn't been used before, thread.fph
700  * is cleared to zeroes.  Also, access to f32-f127 is disabled to
701  * ensure that the task picks up the state from thread.fph when it
702  * executes again.
703  */
704 void
705 ia64_sync_fph (struct task_struct *task)
706 {
707 	struct ia64_psr *psr = ia64_psr(task_pt_regs(task));
708 
709 	ia64_flush_fph(task);
710 	if (!(task->thread.flags & IA64_THREAD_FPH_VALID)) {
711 		task->thread.flags |= IA64_THREAD_FPH_VALID;
712 		memset(&task->thread.fph, 0, sizeof(task->thread.fph));
713 	}
714 	ia64_drop_fpu(task);
715 	psr->dfh = 1;
716 }
717 
718 /*
719  * Change the machine-state of CHILD such that it will return via the normal
720  * kernel exit-path, rather than the syscall-exit path.
721  */
722 static void
723 convert_to_non_syscall (struct task_struct *child, struct pt_regs  *pt,
724 			unsigned long cfm)
725 {
726 	struct unw_frame_info info, prev_info;
727 	unsigned long ip, sp, pr;
728 
729 	unw_init_from_blocked_task(&info, child);
730 	while (1) {
731 		prev_info = info;
732 		if (unw_unwind(&info) < 0)
733 			return;
734 
735 		unw_get_sp(&info, &sp);
736 		if ((long)((unsigned long)child + IA64_STK_OFFSET - sp)
737 		    < IA64_PT_REGS_SIZE) {
738 			dprintk("ptrace.%s: ran off the top of the kernel "
739 				"stack\n", __func__);
740 			return;
741 		}
742 		if (unw_get_pr (&prev_info, &pr) < 0) {
743 			unw_get_rp(&prev_info, &ip);
744 			dprintk("ptrace.%s: failed to read "
745 				"predicate register (ip=0x%lx)\n",
746 				__func__, ip);
747 			return;
748 		}
749 		if (unw_is_intr_frame(&info)
750 		    && (pr & (1UL << PRED_USER_STACK)))
751 			break;
752 	}
753 
754 	/*
755 	 * Note: at the time of this call, the target task is blocked
756 	 * in notify_resume_user() and by clearling PRED_LEAVE_SYSCALL
757 	 * (aka, "pLvSys") we redirect execution from
758 	 * .work_pending_syscall_end to .work_processed_kernel.
759 	 */
760 	unw_get_pr(&prev_info, &pr);
761 	pr &= ~((1UL << PRED_SYSCALL) | (1UL << PRED_LEAVE_SYSCALL));
762 	pr |=  (1UL << PRED_NON_SYSCALL);
763 	unw_set_pr(&prev_info, pr);
764 
765 	pt->cr_ifs = (1UL << 63) | cfm;
766 	/*
767 	 * Clear the memory that is NOT written on syscall-entry to
768 	 * ensure we do not leak kernel-state to user when execution
769 	 * resumes.
770 	 */
771 	pt->r2 = 0;
772 	pt->r3 = 0;
773 	pt->r14 = 0;
774 	memset(&pt->r16, 0, 16*8);	/* clear r16-r31 */
775 	memset(&pt->f6, 0, 6*16);	/* clear f6-f11 */
776 	pt->b7 = 0;
777 	pt->ar_ccv = 0;
778 	pt->ar_csd = 0;
779 	pt->ar_ssd = 0;
780 }
781 
782 static int
783 access_nat_bits (struct task_struct *child, struct pt_regs *pt,
784 		 struct unw_frame_info *info,
785 		 unsigned long *data, int write_access)
786 {
787 	unsigned long regnum, nat_bits, scratch_unat, dummy = 0;
788 	char nat = 0;
789 
790 	if (write_access) {
791 		nat_bits = *data;
792 		scratch_unat = ia64_put_scratch_nat_bits(pt, nat_bits);
793 		if (unw_set_ar(info, UNW_AR_UNAT, scratch_unat) < 0) {
794 			dprintk("ptrace: failed to set ar.unat\n");
795 			return -1;
796 		}
797 		for (regnum = 4; regnum <= 7; ++regnum) {
798 			unw_get_gr(info, regnum, &dummy, &nat);
799 			unw_set_gr(info, regnum, dummy,
800 				   (nat_bits >> regnum) & 1);
801 		}
802 	} else {
803 		if (unw_get_ar(info, UNW_AR_UNAT, &scratch_unat) < 0) {
804 			dprintk("ptrace: failed to read ar.unat\n");
805 			return -1;
806 		}
807 		nat_bits = ia64_get_scratch_nat_bits(pt, scratch_unat);
808 		for (regnum = 4; regnum <= 7; ++regnum) {
809 			unw_get_gr(info, regnum, &dummy, &nat);
810 			nat_bits |= (nat != 0) << regnum;
811 		}
812 		*data = nat_bits;
813 	}
814 	return 0;
815 }
816 
817 static int
818 access_uarea (struct task_struct *child, unsigned long addr,
819 	      unsigned long *data, int write_access);
820 
821 static long
822 ptrace_getregs (struct task_struct *child, struct pt_all_user_regs __user *ppr)
823 {
824 	unsigned long psr, ec, lc, rnat, bsp, cfm, nat_bits, val;
825 	struct unw_frame_info info;
826 	struct ia64_fpreg fpval;
827 	struct switch_stack *sw;
828 	struct pt_regs *pt;
829 	long ret, retval = 0;
830 	char nat = 0;
831 	int i;
832 
833 	if (!access_ok(VERIFY_WRITE, ppr, sizeof(struct pt_all_user_regs)))
834 		return -EIO;
835 
836 	pt = task_pt_regs(child);
837 	sw = (struct switch_stack *) (child->thread.ksp + 16);
838 	unw_init_from_blocked_task(&info, child);
839 	if (unw_unwind_to_user(&info) < 0) {
840 		return -EIO;
841 	}
842 
843 	if (((unsigned long) ppr & 0x7) != 0) {
844 		dprintk("ptrace:unaligned register address %p\n", ppr);
845 		return -EIO;
846 	}
847 
848 	if (access_uarea(child, PT_CR_IPSR, &psr, 0) < 0
849 	    || access_uarea(child, PT_AR_EC, &ec, 0) < 0
850 	    || access_uarea(child, PT_AR_LC, &lc, 0) < 0
851 	    || access_uarea(child, PT_AR_RNAT, &rnat, 0) < 0
852 	    || access_uarea(child, PT_AR_BSP, &bsp, 0) < 0
853 	    || access_uarea(child, PT_CFM, &cfm, 0)
854 	    || access_uarea(child, PT_NAT_BITS, &nat_bits, 0))
855 		return -EIO;
856 
857 	/* control regs */
858 
859 	retval |= __put_user(pt->cr_iip, &ppr->cr_iip);
860 	retval |= __put_user(psr, &ppr->cr_ipsr);
861 
862 	/* app regs */
863 
864 	retval |= __put_user(pt->ar_pfs, &ppr->ar[PT_AUR_PFS]);
865 	retval |= __put_user(pt->ar_rsc, &ppr->ar[PT_AUR_RSC]);
866 	retval |= __put_user(pt->ar_bspstore, &ppr->ar[PT_AUR_BSPSTORE]);
867 	retval |= __put_user(pt->ar_unat, &ppr->ar[PT_AUR_UNAT]);
868 	retval |= __put_user(pt->ar_ccv, &ppr->ar[PT_AUR_CCV]);
869 	retval |= __put_user(pt->ar_fpsr, &ppr->ar[PT_AUR_FPSR]);
870 
871 	retval |= __put_user(ec, &ppr->ar[PT_AUR_EC]);
872 	retval |= __put_user(lc, &ppr->ar[PT_AUR_LC]);
873 	retval |= __put_user(rnat, &ppr->ar[PT_AUR_RNAT]);
874 	retval |= __put_user(bsp, &ppr->ar[PT_AUR_BSP]);
875 	retval |= __put_user(cfm, &ppr->cfm);
876 
877 	/* gr1-gr3 */
878 
879 	retval |= __copy_to_user(&ppr->gr[1], &pt->r1, sizeof(long));
880 	retval |= __copy_to_user(&ppr->gr[2], &pt->r2, sizeof(long) *2);
881 
882 	/* gr4-gr7 */
883 
884 	for (i = 4; i < 8; i++) {
885 		if (unw_access_gr(&info, i, &val, &nat, 0) < 0)
886 			return -EIO;
887 		retval |= __put_user(val, &ppr->gr[i]);
888 	}
889 
890 	/* gr8-gr11 */
891 
892 	retval |= __copy_to_user(&ppr->gr[8], &pt->r8, sizeof(long) * 4);
893 
894 	/* gr12-gr15 */
895 
896 	retval |= __copy_to_user(&ppr->gr[12], &pt->r12, sizeof(long) * 2);
897 	retval |= __copy_to_user(&ppr->gr[14], &pt->r14, sizeof(long));
898 	retval |= __copy_to_user(&ppr->gr[15], &pt->r15, sizeof(long));
899 
900 	/* gr16-gr31 */
901 
902 	retval |= __copy_to_user(&ppr->gr[16], &pt->r16, sizeof(long) * 16);
903 
904 	/* b0 */
905 
906 	retval |= __put_user(pt->b0, &ppr->br[0]);
907 
908 	/* b1-b5 */
909 
910 	for (i = 1; i < 6; i++) {
911 		if (unw_access_br(&info, i, &val, 0) < 0)
912 			return -EIO;
913 		__put_user(val, &ppr->br[i]);
914 	}
915 
916 	/* b6-b7 */
917 
918 	retval |= __put_user(pt->b6, &ppr->br[6]);
919 	retval |= __put_user(pt->b7, &ppr->br[7]);
920 
921 	/* fr2-fr5 */
922 
923 	for (i = 2; i < 6; i++) {
924 		if (unw_get_fr(&info, i, &fpval) < 0)
925 			return -EIO;
926 		retval |= __copy_to_user(&ppr->fr[i], &fpval, sizeof (fpval));
927 	}
928 
929 	/* fr6-fr11 */
930 
931 	retval |= __copy_to_user(&ppr->fr[6], &pt->f6,
932 				 sizeof(struct ia64_fpreg) * 6);
933 
934 	/* fp scratch regs(12-15) */
935 
936 	retval |= __copy_to_user(&ppr->fr[12], &sw->f12,
937 				 sizeof(struct ia64_fpreg) * 4);
938 
939 	/* fr16-fr31 */
940 
941 	for (i = 16; i < 32; i++) {
942 		if (unw_get_fr(&info, i, &fpval) < 0)
943 			return -EIO;
944 		retval |= __copy_to_user(&ppr->fr[i], &fpval, sizeof (fpval));
945 	}
946 
947 	/* fph */
948 
949 	ia64_flush_fph(child);
950 	retval |= __copy_to_user(&ppr->fr[32], &child->thread.fph,
951 				 sizeof(ppr->fr[32]) * 96);
952 
953 	/*  preds */
954 
955 	retval |= __put_user(pt->pr, &ppr->pr);
956 
957 	/* nat bits */
958 
959 	retval |= __put_user(nat_bits, &ppr->nat);
960 
961 	ret = retval ? -EIO : 0;
962 	return ret;
963 }
964 
965 static long
966 ptrace_setregs (struct task_struct *child, struct pt_all_user_regs __user *ppr)
967 {
968 	unsigned long psr, rsc, ec, lc, rnat, bsp, cfm, nat_bits, val = 0;
969 	struct unw_frame_info info;
970 	struct switch_stack *sw;
971 	struct ia64_fpreg fpval;
972 	struct pt_regs *pt;
973 	long ret, retval = 0;
974 	int i;
975 
976 	memset(&fpval, 0, sizeof(fpval));
977 
978 	if (!access_ok(VERIFY_READ, ppr, sizeof(struct pt_all_user_regs)))
979 		return -EIO;
980 
981 	pt = task_pt_regs(child);
982 	sw = (struct switch_stack *) (child->thread.ksp + 16);
983 	unw_init_from_blocked_task(&info, child);
984 	if (unw_unwind_to_user(&info) < 0) {
985 		return -EIO;
986 	}
987 
988 	if (((unsigned long) ppr & 0x7) != 0) {
989 		dprintk("ptrace:unaligned register address %p\n", ppr);
990 		return -EIO;
991 	}
992 
993 	/* control regs */
994 
995 	retval |= __get_user(pt->cr_iip, &ppr->cr_iip);
996 	retval |= __get_user(psr, &ppr->cr_ipsr);
997 
998 	/* app regs */
999 
1000 	retval |= __get_user(pt->ar_pfs, &ppr->ar[PT_AUR_PFS]);
1001 	retval |= __get_user(rsc, &ppr->ar[PT_AUR_RSC]);
1002 	retval |= __get_user(pt->ar_bspstore, &ppr->ar[PT_AUR_BSPSTORE]);
1003 	retval |= __get_user(pt->ar_unat, &ppr->ar[PT_AUR_UNAT]);
1004 	retval |= __get_user(pt->ar_ccv, &ppr->ar[PT_AUR_CCV]);
1005 	retval |= __get_user(pt->ar_fpsr, &ppr->ar[PT_AUR_FPSR]);
1006 
1007 	retval |= __get_user(ec, &ppr->ar[PT_AUR_EC]);
1008 	retval |= __get_user(lc, &ppr->ar[PT_AUR_LC]);
1009 	retval |= __get_user(rnat, &ppr->ar[PT_AUR_RNAT]);
1010 	retval |= __get_user(bsp, &ppr->ar[PT_AUR_BSP]);
1011 	retval |= __get_user(cfm, &ppr->cfm);
1012 
1013 	/* gr1-gr3 */
1014 
1015 	retval |= __copy_from_user(&pt->r1, &ppr->gr[1], sizeof(long));
1016 	retval |= __copy_from_user(&pt->r2, &ppr->gr[2], sizeof(long) * 2);
1017 
1018 	/* gr4-gr7 */
1019 
1020 	for (i = 4; i < 8; i++) {
1021 		retval |= __get_user(val, &ppr->gr[i]);
1022 		/* NaT bit will be set via PT_NAT_BITS: */
1023 		if (unw_set_gr(&info, i, val, 0) < 0)
1024 			return -EIO;
1025 	}
1026 
1027 	/* gr8-gr11 */
1028 
1029 	retval |= __copy_from_user(&pt->r8, &ppr->gr[8], sizeof(long) * 4);
1030 
1031 	/* gr12-gr15 */
1032 
1033 	retval |= __copy_from_user(&pt->r12, &ppr->gr[12], sizeof(long) * 2);
1034 	retval |= __copy_from_user(&pt->r14, &ppr->gr[14], sizeof(long));
1035 	retval |= __copy_from_user(&pt->r15, &ppr->gr[15], sizeof(long));
1036 
1037 	/* gr16-gr31 */
1038 
1039 	retval |= __copy_from_user(&pt->r16, &ppr->gr[16], sizeof(long) * 16);
1040 
1041 	/* b0 */
1042 
1043 	retval |= __get_user(pt->b0, &ppr->br[0]);
1044 
1045 	/* b1-b5 */
1046 
1047 	for (i = 1; i < 6; i++) {
1048 		retval |= __get_user(val, &ppr->br[i]);
1049 		unw_set_br(&info, i, val);
1050 	}
1051 
1052 	/* b6-b7 */
1053 
1054 	retval |= __get_user(pt->b6, &ppr->br[6]);
1055 	retval |= __get_user(pt->b7, &ppr->br[7]);
1056 
1057 	/* fr2-fr5 */
1058 
1059 	for (i = 2; i < 6; i++) {
1060 		retval |= __copy_from_user(&fpval, &ppr->fr[i], sizeof(fpval));
1061 		if (unw_set_fr(&info, i, fpval) < 0)
1062 			return -EIO;
1063 	}
1064 
1065 	/* fr6-fr11 */
1066 
1067 	retval |= __copy_from_user(&pt->f6, &ppr->fr[6],
1068 				   sizeof(ppr->fr[6]) * 6);
1069 
1070 	/* fp scratch regs(12-15) */
1071 
1072 	retval |= __copy_from_user(&sw->f12, &ppr->fr[12],
1073 				   sizeof(ppr->fr[12]) * 4);
1074 
1075 	/* fr16-fr31 */
1076 
1077 	for (i = 16; i < 32; i++) {
1078 		retval |= __copy_from_user(&fpval, &ppr->fr[i],
1079 					   sizeof(fpval));
1080 		if (unw_set_fr(&info, i, fpval) < 0)
1081 			return -EIO;
1082 	}
1083 
1084 	/* fph */
1085 
1086 	ia64_sync_fph(child);
1087 	retval |= __copy_from_user(&child->thread.fph, &ppr->fr[32],
1088 				   sizeof(ppr->fr[32]) * 96);
1089 
1090 	/* preds */
1091 
1092 	retval |= __get_user(pt->pr, &ppr->pr);
1093 
1094 	/* nat bits */
1095 
1096 	retval |= __get_user(nat_bits, &ppr->nat);
1097 
1098 	retval |= access_uarea(child, PT_CR_IPSR, &psr, 1);
1099 	retval |= access_uarea(child, PT_AR_RSC, &rsc, 1);
1100 	retval |= access_uarea(child, PT_AR_EC, &ec, 1);
1101 	retval |= access_uarea(child, PT_AR_LC, &lc, 1);
1102 	retval |= access_uarea(child, PT_AR_RNAT, &rnat, 1);
1103 	retval |= access_uarea(child, PT_AR_BSP, &bsp, 1);
1104 	retval |= access_uarea(child, PT_CFM, &cfm, 1);
1105 	retval |= access_uarea(child, PT_NAT_BITS, &nat_bits, 1);
1106 
1107 	ret = retval ? -EIO : 0;
1108 	return ret;
1109 }
1110 
1111 void
1112 user_enable_single_step (struct task_struct *child)
1113 {
1114 	struct ia64_psr *child_psr = ia64_psr(task_pt_regs(child));
1115 
1116 	set_tsk_thread_flag(child, TIF_SINGLESTEP);
1117 	child_psr->ss = 1;
1118 }
1119 
1120 void
1121 user_enable_block_step (struct task_struct *child)
1122 {
1123 	struct ia64_psr *child_psr = ia64_psr(task_pt_regs(child));
1124 
1125 	set_tsk_thread_flag(child, TIF_SINGLESTEP);
1126 	child_psr->tb = 1;
1127 }
1128 
1129 void
1130 user_disable_single_step (struct task_struct *child)
1131 {
1132 	struct ia64_psr *child_psr = ia64_psr(task_pt_regs(child));
1133 
1134 	/* make sure the single step/taken-branch trap bits are not set: */
1135 	clear_tsk_thread_flag(child, TIF_SINGLESTEP);
1136 	child_psr->ss = 0;
1137 	child_psr->tb = 0;
1138 }
1139 
1140 /*
1141  * Called by kernel/ptrace.c when detaching..
1142  *
1143  * Make sure the single step bit is not set.
1144  */
1145 void
1146 ptrace_disable (struct task_struct *child)
1147 {
1148 	user_disable_single_step(child);
1149 }
1150 
1151 long
1152 arch_ptrace (struct task_struct *child, long request,
1153 	     unsigned long addr, unsigned long data)
1154 {
1155 	switch (request) {
1156 	case PTRACE_PEEKTEXT:
1157 	case PTRACE_PEEKDATA:
1158 		/* read word at location addr */
1159 		if (access_process_vm(child, addr, &data, sizeof(data), 0)
1160 		    != sizeof(data))
1161 			return -EIO;
1162 		/* ensure return value is not mistaken for error code */
1163 		force_successful_syscall_return();
1164 		return data;
1165 
1166 	/* PTRACE_POKETEXT and PTRACE_POKEDATA is handled
1167 	 * by the generic ptrace_request().
1168 	 */
1169 
1170 	case PTRACE_PEEKUSR:
1171 		/* read the word at addr in the USER area */
1172 		if (access_uarea(child, addr, &data, 0) < 0)
1173 			return -EIO;
1174 		/* ensure return value is not mistaken for error code */
1175 		force_successful_syscall_return();
1176 		return data;
1177 
1178 	case PTRACE_POKEUSR:
1179 		/* write the word at addr in the USER area */
1180 		if (access_uarea(child, addr, &data, 1) < 0)
1181 			return -EIO;
1182 		return 0;
1183 
1184 	case PTRACE_OLD_GETSIGINFO:
1185 		/* for backwards-compatibility */
1186 		return ptrace_request(child, PTRACE_GETSIGINFO, addr, data);
1187 
1188 	case PTRACE_OLD_SETSIGINFO:
1189 		/* for backwards-compatibility */
1190 		return ptrace_request(child, PTRACE_SETSIGINFO, addr, data);
1191 
1192 	case PTRACE_GETREGS:
1193 		return ptrace_getregs(child,
1194 				      (struct pt_all_user_regs __user *) data);
1195 
1196 	case PTRACE_SETREGS:
1197 		return ptrace_setregs(child,
1198 				      (struct pt_all_user_regs __user *) data);
1199 
1200 	default:
1201 		return ptrace_request(child, request, addr, data);
1202 	}
1203 }
1204 
1205 
1206 /* "asmlinkage" so the input arguments are preserved... */
1207 
1208 asmlinkage long
1209 syscall_trace_enter (long arg0, long arg1, long arg2, long arg3,
1210 		     long arg4, long arg5, long arg6, long arg7,
1211 		     struct pt_regs regs)
1212 {
1213 	if (test_thread_flag(TIF_SYSCALL_TRACE))
1214 		if (tracehook_report_syscall_entry(&regs))
1215 			return -ENOSYS;
1216 
1217 	/* copy user rbs to kernel rbs */
1218 	if (test_thread_flag(TIF_RESTORE_RSE))
1219 		ia64_sync_krbs();
1220 
1221 
1222 	audit_syscall_entry(regs.r15, arg0, arg1, arg2, arg3);
1223 
1224 	return 0;
1225 }
1226 
1227 /* "asmlinkage" so the input arguments are preserved... */
1228 
1229 asmlinkage void
1230 syscall_trace_leave (long arg0, long arg1, long arg2, long arg3,
1231 		     long arg4, long arg5, long arg6, long arg7,
1232 		     struct pt_regs regs)
1233 {
1234 	int step;
1235 
1236 	audit_syscall_exit(&regs);
1237 
1238 	step = test_thread_flag(TIF_SINGLESTEP);
1239 	if (step || test_thread_flag(TIF_SYSCALL_TRACE))
1240 		tracehook_report_syscall_exit(&regs, step);
1241 
1242 	/* copy user rbs to kernel rbs */
1243 	if (test_thread_flag(TIF_RESTORE_RSE))
1244 		ia64_sync_krbs();
1245 }
1246 
1247 /* Utrace implementation starts here */
1248 struct regset_get {
1249 	void *kbuf;
1250 	void __user *ubuf;
1251 };
1252 
1253 struct regset_set {
1254 	const void *kbuf;
1255 	const void __user *ubuf;
1256 };
1257 
1258 struct regset_getset {
1259 	struct task_struct *target;
1260 	const struct user_regset *regset;
1261 	union {
1262 		struct regset_get get;
1263 		struct regset_set set;
1264 	} u;
1265 	unsigned int pos;
1266 	unsigned int count;
1267 	int ret;
1268 };
1269 
1270 static int
1271 access_elf_gpreg(struct task_struct *target, struct unw_frame_info *info,
1272 		unsigned long addr, unsigned long *data, int write_access)
1273 {
1274 	struct pt_regs *pt;
1275 	unsigned long *ptr = NULL;
1276 	int ret;
1277 	char nat = 0;
1278 
1279 	pt = task_pt_regs(target);
1280 	switch (addr) {
1281 	case ELF_GR_OFFSET(1):
1282 		ptr = &pt->r1;
1283 		break;
1284 	case ELF_GR_OFFSET(2):
1285 	case ELF_GR_OFFSET(3):
1286 		ptr = (void *)&pt->r2 + (addr - ELF_GR_OFFSET(2));
1287 		break;
1288 	case ELF_GR_OFFSET(4) ... ELF_GR_OFFSET(7):
1289 		if (write_access) {
1290 			/* read NaT bit first: */
1291 			unsigned long dummy;
1292 
1293 			ret = unw_get_gr(info, addr/8, &dummy, &nat);
1294 			if (ret < 0)
1295 				return ret;
1296 		}
1297 		return unw_access_gr(info, addr/8, data, &nat, write_access);
1298 	case ELF_GR_OFFSET(8) ... ELF_GR_OFFSET(11):
1299 		ptr = (void *)&pt->r8 + addr - ELF_GR_OFFSET(8);
1300 		break;
1301 	case ELF_GR_OFFSET(12):
1302 	case ELF_GR_OFFSET(13):
1303 		ptr = (void *)&pt->r12 + addr - ELF_GR_OFFSET(12);
1304 		break;
1305 	case ELF_GR_OFFSET(14):
1306 		ptr = &pt->r14;
1307 		break;
1308 	case ELF_GR_OFFSET(15):
1309 		ptr = &pt->r15;
1310 	}
1311 	if (write_access)
1312 		*ptr = *data;
1313 	else
1314 		*data = *ptr;
1315 	return 0;
1316 }
1317 
1318 static int
1319 access_elf_breg(struct task_struct *target, struct unw_frame_info *info,
1320 		unsigned long addr, unsigned long *data, int write_access)
1321 {
1322 	struct pt_regs *pt;
1323 	unsigned long *ptr = NULL;
1324 
1325 	pt = task_pt_regs(target);
1326 	switch (addr) {
1327 	case ELF_BR_OFFSET(0):
1328 		ptr = &pt->b0;
1329 		break;
1330 	case ELF_BR_OFFSET(1) ... ELF_BR_OFFSET(5):
1331 		return unw_access_br(info, (addr - ELF_BR_OFFSET(0))/8,
1332 				     data, write_access);
1333 	case ELF_BR_OFFSET(6):
1334 		ptr = &pt->b6;
1335 		break;
1336 	case ELF_BR_OFFSET(7):
1337 		ptr = &pt->b7;
1338 	}
1339 	if (write_access)
1340 		*ptr = *data;
1341 	else
1342 		*data = *ptr;
1343 	return 0;
1344 }
1345 
1346 static int
1347 access_elf_areg(struct task_struct *target, struct unw_frame_info *info,
1348 		unsigned long addr, unsigned long *data, int write_access)
1349 {
1350 	struct pt_regs *pt;
1351 	unsigned long cfm, urbs_end;
1352 	unsigned long *ptr = NULL;
1353 
1354 	pt = task_pt_regs(target);
1355 	if (addr >= ELF_AR_RSC_OFFSET && addr <= ELF_AR_SSD_OFFSET) {
1356 		switch (addr) {
1357 		case ELF_AR_RSC_OFFSET:
1358 			/* force PL3 */
1359 			if (write_access)
1360 				pt->ar_rsc = *data | (3 << 2);
1361 			else
1362 				*data = pt->ar_rsc;
1363 			return 0;
1364 		case ELF_AR_BSP_OFFSET:
1365 			/*
1366 			 * By convention, we use PT_AR_BSP to refer to
1367 			 * the end of the user-level backing store.
1368 			 * Use ia64_rse_skip_regs(PT_AR_BSP, -CFM.sof)
1369 			 * to get the real value of ar.bsp at the time
1370 			 * the kernel was entered.
1371 			 *
1372 			 * Furthermore, when changing the contents of
1373 			 * PT_AR_BSP (or PT_CFM) while the task is
1374 			 * blocked in a system call, convert the state
1375 			 * so that the non-system-call exit
1376 			 * path is used.  This ensures that the proper
1377 			 * state will be picked up when resuming
1378 			 * execution.  However, it *also* means that
1379 			 * once we write PT_AR_BSP/PT_CFM, it won't be
1380 			 * possible to modify the syscall arguments of
1381 			 * the pending system call any longer.  This
1382 			 * shouldn't be an issue because modifying
1383 			 * PT_AR_BSP/PT_CFM generally implies that
1384 			 * we're either abandoning the pending system
1385 			 * call or that we defer it's re-execution
1386 			 * (e.g., due to GDB doing an inferior
1387 			 * function call).
1388 			 */
1389 			urbs_end = ia64_get_user_rbs_end(target, pt, &cfm);
1390 			if (write_access) {
1391 				if (*data != urbs_end) {
1392 					if (in_syscall(pt))
1393 						convert_to_non_syscall(target,
1394 								       pt,
1395 								       cfm);
1396 					/*
1397 					 * Simulate user-level write
1398 					 * of ar.bsp:
1399 					 */
1400 					pt->loadrs = 0;
1401 					pt->ar_bspstore = *data;
1402 				}
1403 			} else
1404 				*data = urbs_end;
1405 			return 0;
1406 		case ELF_AR_BSPSTORE_OFFSET:
1407 			ptr = &pt->ar_bspstore;
1408 			break;
1409 		case ELF_AR_RNAT_OFFSET:
1410 			ptr = &pt->ar_rnat;
1411 			break;
1412 		case ELF_AR_CCV_OFFSET:
1413 			ptr = &pt->ar_ccv;
1414 			break;
1415 		case ELF_AR_UNAT_OFFSET:
1416 			ptr = &pt->ar_unat;
1417 			break;
1418 		case ELF_AR_FPSR_OFFSET:
1419 			ptr = &pt->ar_fpsr;
1420 			break;
1421 		case ELF_AR_PFS_OFFSET:
1422 			ptr = &pt->ar_pfs;
1423 			break;
1424 		case ELF_AR_LC_OFFSET:
1425 			return unw_access_ar(info, UNW_AR_LC, data,
1426 					     write_access);
1427 		case ELF_AR_EC_OFFSET:
1428 			return unw_access_ar(info, UNW_AR_EC, data,
1429 					     write_access);
1430 		case ELF_AR_CSD_OFFSET:
1431 			ptr = &pt->ar_csd;
1432 			break;
1433 		case ELF_AR_SSD_OFFSET:
1434 			ptr = &pt->ar_ssd;
1435 		}
1436 	} else if (addr >= ELF_CR_IIP_OFFSET && addr <= ELF_CR_IPSR_OFFSET) {
1437 		switch (addr) {
1438 		case ELF_CR_IIP_OFFSET:
1439 			ptr = &pt->cr_iip;
1440 			break;
1441 		case ELF_CFM_OFFSET:
1442 			urbs_end = ia64_get_user_rbs_end(target, pt, &cfm);
1443 			if (write_access) {
1444 				if (((cfm ^ *data) & PFM_MASK) != 0) {
1445 					if (in_syscall(pt))
1446 						convert_to_non_syscall(target,
1447 								       pt,
1448 								       cfm);
1449 					pt->cr_ifs = ((pt->cr_ifs & ~PFM_MASK)
1450 						      | (*data & PFM_MASK));
1451 				}
1452 			} else
1453 				*data = cfm;
1454 			return 0;
1455 		case ELF_CR_IPSR_OFFSET:
1456 			if (write_access) {
1457 				unsigned long tmp = *data;
1458 				/* psr.ri==3 is a reserved value: SDM 2:25 */
1459 				if ((tmp & IA64_PSR_RI) == IA64_PSR_RI)
1460 					tmp &= ~IA64_PSR_RI;
1461 				pt->cr_ipsr = ((tmp & IPSR_MASK)
1462 					       | (pt->cr_ipsr & ~IPSR_MASK));
1463 			} else
1464 				*data = (pt->cr_ipsr & IPSR_MASK);
1465 			return 0;
1466 		}
1467 	} else if (addr == ELF_NAT_OFFSET)
1468 		return access_nat_bits(target, pt, info,
1469 				       data, write_access);
1470 	else if (addr == ELF_PR_OFFSET)
1471 		ptr = &pt->pr;
1472 	else
1473 		return -1;
1474 
1475 	if (write_access)
1476 		*ptr = *data;
1477 	else
1478 		*data = *ptr;
1479 
1480 	return 0;
1481 }
1482 
1483 static int
1484 access_elf_reg(struct task_struct *target, struct unw_frame_info *info,
1485 		unsigned long addr, unsigned long *data, int write_access)
1486 {
1487 	if (addr >= ELF_GR_OFFSET(1) && addr <= ELF_GR_OFFSET(15))
1488 		return access_elf_gpreg(target, info, addr, data, write_access);
1489 	else if (addr >= ELF_BR_OFFSET(0) && addr <= ELF_BR_OFFSET(7))
1490 		return access_elf_breg(target, info, addr, data, write_access);
1491 	else
1492 		return access_elf_areg(target, info, addr, data, write_access);
1493 }
1494 
1495 void do_gpregs_get(struct unw_frame_info *info, void *arg)
1496 {
1497 	struct pt_regs *pt;
1498 	struct regset_getset *dst = arg;
1499 	elf_greg_t tmp[16];
1500 	unsigned int i, index, min_copy;
1501 
1502 	if (unw_unwind_to_user(info) < 0)
1503 		return;
1504 
1505 	/*
1506 	 * coredump format:
1507 	 *      r0-r31
1508 	 *      NaT bits (for r0-r31; bit N == 1 iff rN is a NaT)
1509 	 *      predicate registers (p0-p63)
1510 	 *      b0-b7
1511 	 *      ip cfm user-mask
1512 	 *      ar.rsc ar.bsp ar.bspstore ar.rnat
1513 	 *      ar.ccv ar.unat ar.fpsr ar.pfs ar.lc ar.ec
1514 	 */
1515 
1516 
1517 	/* Skip r0 */
1518 	if (dst->count > 0 && dst->pos < ELF_GR_OFFSET(1)) {
1519 		dst->ret = user_regset_copyout_zero(&dst->pos, &dst->count,
1520 						      &dst->u.get.kbuf,
1521 						      &dst->u.get.ubuf,
1522 						      0, ELF_GR_OFFSET(1));
1523 		if (dst->ret || dst->count == 0)
1524 			return;
1525 	}
1526 
1527 	/* gr1 - gr15 */
1528 	if (dst->count > 0 && dst->pos < ELF_GR_OFFSET(16)) {
1529 		index = (dst->pos - ELF_GR_OFFSET(1)) / sizeof(elf_greg_t);
1530 		min_copy = ELF_GR_OFFSET(16) > (dst->pos + dst->count) ?
1531 			 (dst->pos + dst->count) : ELF_GR_OFFSET(16);
1532 		for (i = dst->pos; i < min_copy; i += sizeof(elf_greg_t),
1533 				index++)
1534 			if (access_elf_reg(dst->target, info, i,
1535 						&tmp[index], 0) < 0) {
1536 				dst->ret = -EIO;
1537 				return;
1538 			}
1539 		dst->ret = user_regset_copyout(&dst->pos, &dst->count,
1540 				&dst->u.get.kbuf, &dst->u.get.ubuf, tmp,
1541 				ELF_GR_OFFSET(1), ELF_GR_OFFSET(16));
1542 		if (dst->ret || dst->count == 0)
1543 			return;
1544 	}
1545 
1546 	/* r16-r31 */
1547 	if (dst->count > 0 && dst->pos < ELF_NAT_OFFSET) {
1548 		pt = task_pt_regs(dst->target);
1549 		dst->ret = user_regset_copyout(&dst->pos, &dst->count,
1550 				&dst->u.get.kbuf, &dst->u.get.ubuf, &pt->r16,
1551 				ELF_GR_OFFSET(16), ELF_NAT_OFFSET);
1552 		if (dst->ret || dst->count == 0)
1553 			return;
1554 	}
1555 
1556 	/* nat, pr, b0 - b7 */
1557 	if (dst->count > 0 && dst->pos < ELF_CR_IIP_OFFSET) {
1558 		index = (dst->pos - ELF_NAT_OFFSET) / sizeof(elf_greg_t);
1559 		min_copy = ELF_CR_IIP_OFFSET > (dst->pos + dst->count) ?
1560 			 (dst->pos + dst->count) : ELF_CR_IIP_OFFSET;
1561 		for (i = dst->pos; i < min_copy; i += sizeof(elf_greg_t),
1562 				index++)
1563 			if (access_elf_reg(dst->target, info, i,
1564 						&tmp[index], 0) < 0) {
1565 				dst->ret = -EIO;
1566 				return;
1567 			}
1568 		dst->ret = user_regset_copyout(&dst->pos, &dst->count,
1569 				&dst->u.get.kbuf, &dst->u.get.ubuf, tmp,
1570 				ELF_NAT_OFFSET, ELF_CR_IIP_OFFSET);
1571 		if (dst->ret || dst->count == 0)
1572 			return;
1573 	}
1574 
1575 	/* ip cfm psr ar.rsc ar.bsp ar.bspstore ar.rnat
1576 	 * ar.ccv ar.unat ar.fpsr ar.pfs ar.lc ar.ec ar.csd ar.ssd
1577 	 */
1578 	if (dst->count > 0 && dst->pos < (ELF_AR_END_OFFSET)) {
1579 		index = (dst->pos - ELF_CR_IIP_OFFSET) / sizeof(elf_greg_t);
1580 		min_copy = ELF_AR_END_OFFSET > (dst->pos + dst->count) ?
1581 			 (dst->pos + dst->count) : ELF_AR_END_OFFSET;
1582 		for (i = dst->pos; i < min_copy; i += sizeof(elf_greg_t),
1583 				index++)
1584 			if (access_elf_reg(dst->target, info, i,
1585 						&tmp[index], 0) < 0) {
1586 				dst->ret = -EIO;
1587 				return;
1588 			}
1589 		dst->ret = user_regset_copyout(&dst->pos, &dst->count,
1590 				&dst->u.get.kbuf, &dst->u.get.ubuf, tmp,
1591 				ELF_CR_IIP_OFFSET, ELF_AR_END_OFFSET);
1592 	}
1593 }
1594 
1595 void do_gpregs_set(struct unw_frame_info *info, void *arg)
1596 {
1597 	struct pt_regs *pt;
1598 	struct regset_getset *dst = arg;
1599 	elf_greg_t tmp[16];
1600 	unsigned int i, index;
1601 
1602 	if (unw_unwind_to_user(info) < 0)
1603 		return;
1604 
1605 	/* Skip r0 */
1606 	if (dst->count > 0 && dst->pos < ELF_GR_OFFSET(1)) {
1607 		dst->ret = user_regset_copyin_ignore(&dst->pos, &dst->count,
1608 						       &dst->u.set.kbuf,
1609 						       &dst->u.set.ubuf,
1610 						       0, ELF_GR_OFFSET(1));
1611 		if (dst->ret || dst->count == 0)
1612 			return;
1613 	}
1614 
1615 	/* gr1-gr15 */
1616 	if (dst->count > 0 && dst->pos < ELF_GR_OFFSET(16)) {
1617 		i = dst->pos;
1618 		index = (dst->pos - ELF_GR_OFFSET(1)) / sizeof(elf_greg_t);
1619 		dst->ret = user_regset_copyin(&dst->pos, &dst->count,
1620 				&dst->u.set.kbuf, &dst->u.set.ubuf, tmp,
1621 				ELF_GR_OFFSET(1), ELF_GR_OFFSET(16));
1622 		if (dst->ret)
1623 			return;
1624 		for ( ; i < dst->pos; i += sizeof(elf_greg_t), index++)
1625 			if (access_elf_reg(dst->target, info, i,
1626 						&tmp[index], 1) < 0) {
1627 				dst->ret = -EIO;
1628 				return;
1629 			}
1630 		if (dst->count == 0)
1631 			return;
1632 	}
1633 
1634 	/* gr16-gr31 */
1635 	if (dst->count > 0 && dst->pos < ELF_NAT_OFFSET) {
1636 		pt = task_pt_regs(dst->target);
1637 		dst->ret = user_regset_copyin(&dst->pos, &dst->count,
1638 				&dst->u.set.kbuf, &dst->u.set.ubuf, &pt->r16,
1639 				ELF_GR_OFFSET(16), ELF_NAT_OFFSET);
1640 		if (dst->ret || dst->count == 0)
1641 			return;
1642 	}
1643 
1644 	/* nat, pr, b0 - b7 */
1645 	if (dst->count > 0 && dst->pos < ELF_CR_IIP_OFFSET) {
1646 		i = dst->pos;
1647 		index = (dst->pos - ELF_NAT_OFFSET) / sizeof(elf_greg_t);
1648 		dst->ret = user_regset_copyin(&dst->pos, &dst->count,
1649 				&dst->u.set.kbuf, &dst->u.set.ubuf, tmp,
1650 				ELF_NAT_OFFSET, ELF_CR_IIP_OFFSET);
1651 		if (dst->ret)
1652 			return;
1653 		for (; i < dst->pos; i += sizeof(elf_greg_t), index++)
1654 			if (access_elf_reg(dst->target, info, i,
1655 						&tmp[index], 1) < 0) {
1656 				dst->ret = -EIO;
1657 				return;
1658 			}
1659 		if (dst->count == 0)
1660 			return;
1661 	}
1662 
1663 	/* ip cfm psr ar.rsc ar.bsp ar.bspstore ar.rnat
1664 	 * ar.ccv ar.unat ar.fpsr ar.pfs ar.lc ar.ec ar.csd ar.ssd
1665 	 */
1666 	if (dst->count > 0 && dst->pos < (ELF_AR_END_OFFSET)) {
1667 		i = dst->pos;
1668 		index = (dst->pos - ELF_CR_IIP_OFFSET) / sizeof(elf_greg_t);
1669 		dst->ret = user_regset_copyin(&dst->pos, &dst->count,
1670 				&dst->u.set.kbuf, &dst->u.set.ubuf, tmp,
1671 				ELF_CR_IIP_OFFSET, ELF_AR_END_OFFSET);
1672 		if (dst->ret)
1673 			return;
1674 		for ( ; i < dst->pos; i += sizeof(elf_greg_t), index++)
1675 			if (access_elf_reg(dst->target, info, i,
1676 						&tmp[index], 1) < 0) {
1677 				dst->ret = -EIO;
1678 				return;
1679 			}
1680 	}
1681 }
1682 
1683 #define ELF_FP_OFFSET(i)	(i * sizeof(elf_fpreg_t))
1684 
1685 void do_fpregs_get(struct unw_frame_info *info, void *arg)
1686 {
1687 	struct regset_getset *dst = arg;
1688 	struct task_struct *task = dst->target;
1689 	elf_fpreg_t tmp[30];
1690 	int index, min_copy, i;
1691 
1692 	if (unw_unwind_to_user(info) < 0)
1693 		return;
1694 
1695 	/* Skip pos 0 and 1 */
1696 	if (dst->count > 0 && dst->pos < ELF_FP_OFFSET(2)) {
1697 		dst->ret = user_regset_copyout_zero(&dst->pos, &dst->count,
1698 						      &dst->u.get.kbuf,
1699 						      &dst->u.get.ubuf,
1700 						      0, ELF_FP_OFFSET(2));
1701 		if (dst->count == 0 || dst->ret)
1702 			return;
1703 	}
1704 
1705 	/* fr2-fr31 */
1706 	if (dst->count > 0 && dst->pos < ELF_FP_OFFSET(32)) {
1707 		index = (dst->pos - ELF_FP_OFFSET(2)) / sizeof(elf_fpreg_t);
1708 
1709 		min_copy = min(((unsigned int)ELF_FP_OFFSET(32)),
1710 				dst->pos + dst->count);
1711 		for (i = dst->pos; i < min_copy; i += sizeof(elf_fpreg_t),
1712 				index++)
1713 			if (unw_get_fr(info, i / sizeof(elf_fpreg_t),
1714 					 &tmp[index])) {
1715 				dst->ret = -EIO;
1716 				return;
1717 			}
1718 		dst->ret = user_regset_copyout(&dst->pos, &dst->count,
1719 				&dst->u.get.kbuf, &dst->u.get.ubuf, tmp,
1720 				ELF_FP_OFFSET(2), ELF_FP_OFFSET(32));
1721 		if (dst->count == 0 || dst->ret)
1722 			return;
1723 	}
1724 
1725 	/* fph */
1726 	if (dst->count > 0) {
1727 		ia64_flush_fph(dst->target);
1728 		if (task->thread.flags & IA64_THREAD_FPH_VALID)
1729 			dst->ret = user_regset_copyout(
1730 				&dst->pos, &dst->count,
1731 				&dst->u.get.kbuf, &dst->u.get.ubuf,
1732 				&dst->target->thread.fph,
1733 				ELF_FP_OFFSET(32), -1);
1734 		else
1735 			/* Zero fill instead.  */
1736 			dst->ret = user_regset_copyout_zero(
1737 				&dst->pos, &dst->count,
1738 				&dst->u.get.kbuf, &dst->u.get.ubuf,
1739 				ELF_FP_OFFSET(32), -1);
1740 	}
1741 }
1742 
1743 void do_fpregs_set(struct unw_frame_info *info, void *arg)
1744 {
1745 	struct regset_getset *dst = arg;
1746 	elf_fpreg_t fpreg, tmp[30];
1747 	int index, start, end;
1748 
1749 	if (unw_unwind_to_user(info) < 0)
1750 		return;
1751 
1752 	/* Skip pos 0 and 1 */
1753 	if (dst->count > 0 && dst->pos < ELF_FP_OFFSET(2)) {
1754 		dst->ret = user_regset_copyin_ignore(&dst->pos, &dst->count,
1755 						       &dst->u.set.kbuf,
1756 						       &dst->u.set.ubuf,
1757 						       0, ELF_FP_OFFSET(2));
1758 		if (dst->count == 0 || dst->ret)
1759 			return;
1760 	}
1761 
1762 	/* fr2-fr31 */
1763 	if (dst->count > 0 && dst->pos < ELF_FP_OFFSET(32)) {
1764 		start = dst->pos;
1765 		end = min(((unsigned int)ELF_FP_OFFSET(32)),
1766 			 dst->pos + dst->count);
1767 		dst->ret = user_regset_copyin(&dst->pos, &dst->count,
1768 				&dst->u.set.kbuf, &dst->u.set.ubuf, tmp,
1769 				ELF_FP_OFFSET(2), ELF_FP_OFFSET(32));
1770 		if (dst->ret)
1771 			return;
1772 
1773 		if (start & 0xF) { /* only write high part */
1774 			if (unw_get_fr(info, start / sizeof(elf_fpreg_t),
1775 					 &fpreg)) {
1776 				dst->ret = -EIO;
1777 				return;
1778 			}
1779 			tmp[start / sizeof(elf_fpreg_t) - 2].u.bits[0]
1780 				= fpreg.u.bits[0];
1781 			start &= ~0xFUL;
1782 		}
1783 		if (end & 0xF) { /* only write low part */
1784 			if (unw_get_fr(info, end / sizeof(elf_fpreg_t),
1785 					&fpreg)) {
1786 				dst->ret = -EIO;
1787 				return;
1788 			}
1789 			tmp[end / sizeof(elf_fpreg_t) - 2].u.bits[1]
1790 				= fpreg.u.bits[1];
1791 			end = (end + 0xF) & ~0xFUL;
1792 		}
1793 
1794 		for ( ;	start < end ; start += sizeof(elf_fpreg_t)) {
1795 			index = start / sizeof(elf_fpreg_t);
1796 			if (unw_set_fr(info, index, tmp[index - 2])) {
1797 				dst->ret = -EIO;
1798 				return;
1799 			}
1800 		}
1801 		if (dst->ret || dst->count == 0)
1802 			return;
1803 	}
1804 
1805 	/* fph */
1806 	if (dst->count > 0 && dst->pos < ELF_FP_OFFSET(128)) {
1807 		ia64_sync_fph(dst->target);
1808 		dst->ret = user_regset_copyin(&dst->pos, &dst->count,
1809 						&dst->u.set.kbuf,
1810 						&dst->u.set.ubuf,
1811 						&dst->target->thread.fph,
1812 						ELF_FP_OFFSET(32), -1);
1813 	}
1814 }
1815 
1816 static int
1817 do_regset_call(void (*call)(struct unw_frame_info *, void *),
1818 	       struct task_struct *target,
1819 	       const struct user_regset *regset,
1820 	       unsigned int pos, unsigned int count,
1821 	       const void *kbuf, const void __user *ubuf)
1822 {
1823 	struct regset_getset info = { .target = target, .regset = regset,
1824 				 .pos = pos, .count = count,
1825 				 .u.set = { .kbuf = kbuf, .ubuf = ubuf },
1826 				 .ret = 0 };
1827 
1828 	if (target == current)
1829 		unw_init_running(call, &info);
1830 	else {
1831 		struct unw_frame_info ufi;
1832 		memset(&ufi, 0, sizeof(ufi));
1833 		unw_init_from_blocked_task(&ufi, target);
1834 		(*call)(&ufi, &info);
1835 	}
1836 
1837 	return info.ret;
1838 }
1839 
1840 static int
1841 gpregs_get(struct task_struct *target,
1842 	   const struct user_regset *regset,
1843 	   unsigned int pos, unsigned int count,
1844 	   void *kbuf, void __user *ubuf)
1845 {
1846 	return do_regset_call(do_gpregs_get, target, regset, pos, count,
1847 		kbuf, ubuf);
1848 }
1849 
1850 static int gpregs_set(struct task_struct *target,
1851 		const struct user_regset *regset,
1852 		unsigned int pos, unsigned int count,
1853 		const void *kbuf, const void __user *ubuf)
1854 {
1855 	return do_regset_call(do_gpregs_set, target, regset, pos, count,
1856 		kbuf, ubuf);
1857 }
1858 
1859 static void do_gpregs_writeback(struct unw_frame_info *info, void *arg)
1860 {
1861 	do_sync_rbs(info, ia64_sync_user_rbs);
1862 }
1863 
1864 /*
1865  * This is called to write back the register backing store.
1866  * ptrace does this before it stops, so that a tracer reading the user
1867  * memory after the thread stops will get the current register data.
1868  */
1869 static int
1870 gpregs_writeback(struct task_struct *target,
1871 		 const struct user_regset *regset,
1872 		 int now)
1873 {
1874 	if (test_and_set_tsk_thread_flag(target, TIF_RESTORE_RSE))
1875 		return 0;
1876 	set_notify_resume(target);
1877 	return do_regset_call(do_gpregs_writeback, target, regset, 0, 0,
1878 		NULL, NULL);
1879 }
1880 
1881 static int
1882 fpregs_active(struct task_struct *target, const struct user_regset *regset)
1883 {
1884 	return (target->thread.flags & IA64_THREAD_FPH_VALID) ? 128 : 32;
1885 }
1886 
1887 static int fpregs_get(struct task_struct *target,
1888 		const struct user_regset *regset,
1889 		unsigned int pos, unsigned int count,
1890 		void *kbuf, void __user *ubuf)
1891 {
1892 	return do_regset_call(do_fpregs_get, target, regset, pos, count,
1893 		kbuf, ubuf);
1894 }
1895 
1896 static int fpregs_set(struct task_struct *target,
1897 		const struct user_regset *regset,
1898 		unsigned int pos, unsigned int count,
1899 		const void *kbuf, const void __user *ubuf)
1900 {
1901 	return do_regset_call(do_fpregs_set, target, regset, pos, count,
1902 		kbuf, ubuf);
1903 }
1904 
1905 static int
1906 access_uarea(struct task_struct *child, unsigned long addr,
1907 	      unsigned long *data, int write_access)
1908 {
1909 	unsigned int pos = -1; /* an invalid value */
1910 	int ret;
1911 	unsigned long *ptr, regnum;
1912 
1913 	if ((addr & 0x7) != 0) {
1914 		dprintk("ptrace: unaligned register address 0x%lx\n", addr);
1915 		return -1;
1916 	}
1917 	if ((addr >= PT_NAT_BITS + 8 && addr < PT_F2) ||
1918 		(addr >= PT_R7 + 8 && addr < PT_B1) ||
1919 		(addr >= PT_AR_LC + 8 && addr < PT_CR_IPSR) ||
1920 		(addr >= PT_AR_SSD + 8 && addr < PT_DBR)) {
1921 		dprintk("ptrace: rejecting access to register "
1922 					"address 0x%lx\n", addr);
1923 		return -1;
1924 	}
1925 
1926 	switch (addr) {
1927 	case PT_F32 ... (PT_F127 + 15):
1928 		pos = addr - PT_F32 + ELF_FP_OFFSET(32);
1929 		break;
1930 	case PT_F2 ... (PT_F5 + 15):
1931 		pos = addr - PT_F2 + ELF_FP_OFFSET(2);
1932 		break;
1933 	case PT_F10 ... (PT_F31 + 15):
1934 		pos = addr - PT_F10 + ELF_FP_OFFSET(10);
1935 		break;
1936 	case PT_F6 ... (PT_F9 + 15):
1937 		pos = addr - PT_F6 + ELF_FP_OFFSET(6);
1938 		break;
1939 	}
1940 
1941 	if (pos != -1) {
1942 		if (write_access)
1943 			ret = fpregs_set(child, NULL, pos,
1944 				sizeof(unsigned long), data, NULL);
1945 		else
1946 			ret = fpregs_get(child, NULL, pos,
1947 				sizeof(unsigned long), data, NULL);
1948 		if (ret != 0)
1949 			return -1;
1950 		return 0;
1951 	}
1952 
1953 	switch (addr) {
1954 	case PT_NAT_BITS:
1955 		pos = ELF_NAT_OFFSET;
1956 		break;
1957 	case PT_R4 ... PT_R7:
1958 		pos = addr - PT_R4 + ELF_GR_OFFSET(4);
1959 		break;
1960 	case PT_B1 ... PT_B5:
1961 		pos = addr - PT_B1 + ELF_BR_OFFSET(1);
1962 		break;
1963 	case PT_AR_EC:
1964 		pos = ELF_AR_EC_OFFSET;
1965 		break;
1966 	case PT_AR_LC:
1967 		pos = ELF_AR_LC_OFFSET;
1968 		break;
1969 	case PT_CR_IPSR:
1970 		pos = ELF_CR_IPSR_OFFSET;
1971 		break;
1972 	case PT_CR_IIP:
1973 		pos = ELF_CR_IIP_OFFSET;
1974 		break;
1975 	case PT_CFM:
1976 		pos = ELF_CFM_OFFSET;
1977 		break;
1978 	case PT_AR_UNAT:
1979 		pos = ELF_AR_UNAT_OFFSET;
1980 		break;
1981 	case PT_AR_PFS:
1982 		pos = ELF_AR_PFS_OFFSET;
1983 		break;
1984 	case PT_AR_RSC:
1985 		pos = ELF_AR_RSC_OFFSET;
1986 		break;
1987 	case PT_AR_RNAT:
1988 		pos = ELF_AR_RNAT_OFFSET;
1989 		break;
1990 	case PT_AR_BSPSTORE:
1991 		pos = ELF_AR_BSPSTORE_OFFSET;
1992 		break;
1993 	case PT_PR:
1994 		pos = ELF_PR_OFFSET;
1995 		break;
1996 	case PT_B6:
1997 		pos = ELF_BR_OFFSET(6);
1998 		break;
1999 	case PT_AR_BSP:
2000 		pos = ELF_AR_BSP_OFFSET;
2001 		break;
2002 	case PT_R1 ... PT_R3:
2003 		pos = addr - PT_R1 + ELF_GR_OFFSET(1);
2004 		break;
2005 	case PT_R12 ... PT_R15:
2006 		pos = addr - PT_R12 + ELF_GR_OFFSET(12);
2007 		break;
2008 	case PT_R8 ... PT_R11:
2009 		pos = addr - PT_R8 + ELF_GR_OFFSET(8);
2010 		break;
2011 	case PT_R16 ... PT_R31:
2012 		pos = addr - PT_R16 + ELF_GR_OFFSET(16);
2013 		break;
2014 	case PT_AR_CCV:
2015 		pos = ELF_AR_CCV_OFFSET;
2016 		break;
2017 	case PT_AR_FPSR:
2018 		pos = ELF_AR_FPSR_OFFSET;
2019 		break;
2020 	case PT_B0:
2021 		pos = ELF_BR_OFFSET(0);
2022 		break;
2023 	case PT_B7:
2024 		pos = ELF_BR_OFFSET(7);
2025 		break;
2026 	case PT_AR_CSD:
2027 		pos = ELF_AR_CSD_OFFSET;
2028 		break;
2029 	case PT_AR_SSD:
2030 		pos = ELF_AR_SSD_OFFSET;
2031 		break;
2032 	}
2033 
2034 	if (pos != -1) {
2035 		if (write_access)
2036 			ret = gpregs_set(child, NULL, pos,
2037 				sizeof(unsigned long), data, NULL);
2038 		else
2039 			ret = gpregs_get(child, NULL, pos,
2040 				sizeof(unsigned long), data, NULL);
2041 		if (ret != 0)
2042 			return -1;
2043 		return 0;
2044 	}
2045 
2046 	/* access debug registers */
2047 	if (addr >= PT_IBR) {
2048 		regnum = (addr - PT_IBR) >> 3;
2049 		ptr = &child->thread.ibr[0];
2050 	} else {
2051 		regnum = (addr - PT_DBR) >> 3;
2052 		ptr = &child->thread.dbr[0];
2053 	}
2054 
2055 	if (regnum >= 8) {
2056 		dprintk("ptrace: rejecting access to register "
2057 				"address 0x%lx\n", addr);
2058 		return -1;
2059 	}
2060 #ifdef CONFIG_PERFMON
2061 	/*
2062 	 * Check if debug registers are used by perfmon. This
2063 	 * test must be done once we know that we can do the
2064 	 * operation, i.e. the arguments are all valid, but
2065 	 * before we start modifying the state.
2066 	 *
2067 	 * Perfmon needs to keep a count of how many processes
2068 	 * are trying to modify the debug registers for system
2069 	 * wide monitoring sessions.
2070 	 *
2071 	 * We also include read access here, because they may
2072 	 * cause the PMU-installed debug register state
2073 	 * (dbr[], ibr[]) to be reset. The two arrays are also
2074 	 * used by perfmon, but we do not use
2075 	 * IA64_THREAD_DBG_VALID. The registers are restored
2076 	 * by the PMU context switch code.
2077 	 */
2078 	if (pfm_use_debug_registers(child))
2079 		return -1;
2080 #endif
2081 
2082 	if (!(child->thread.flags & IA64_THREAD_DBG_VALID)) {
2083 		child->thread.flags |= IA64_THREAD_DBG_VALID;
2084 		memset(child->thread.dbr, 0,
2085 				sizeof(child->thread.dbr));
2086 		memset(child->thread.ibr, 0,
2087 				sizeof(child->thread.ibr));
2088 	}
2089 
2090 	ptr += regnum;
2091 
2092 	if ((regnum & 1) && write_access) {
2093 		/* don't let the user set kernel-level breakpoints: */
2094 		*ptr = *data & ~(7UL << 56);
2095 		return 0;
2096 	}
2097 	if (write_access)
2098 		*ptr = *data;
2099 	else
2100 		*data = *ptr;
2101 	return 0;
2102 }
2103 
2104 static const struct user_regset native_regsets[] = {
2105 	{
2106 		.core_note_type = NT_PRSTATUS,
2107 		.n = ELF_NGREG,
2108 		.size = sizeof(elf_greg_t), .align = sizeof(elf_greg_t),
2109 		.get = gpregs_get, .set = gpregs_set,
2110 		.writeback = gpregs_writeback
2111 	},
2112 	{
2113 		.core_note_type = NT_PRFPREG,
2114 		.n = ELF_NFPREG,
2115 		.size = sizeof(elf_fpreg_t), .align = sizeof(elf_fpreg_t),
2116 		.get = fpregs_get, .set = fpregs_set, .active = fpregs_active
2117 	},
2118 };
2119 
2120 static const struct user_regset_view user_ia64_view = {
2121 	.name = "ia64",
2122 	.e_machine = EM_IA_64,
2123 	.regsets = native_regsets, .n = ARRAY_SIZE(native_regsets)
2124 };
2125 
2126 const struct user_regset_view *task_user_regset_view(struct task_struct *tsk)
2127 {
2128 	return &user_ia64_view;
2129 }
2130 
2131 struct syscall_get_set_args {
2132 	unsigned int i;
2133 	unsigned int n;
2134 	unsigned long *args;
2135 	struct pt_regs *regs;
2136 	int rw;
2137 };
2138 
2139 static void syscall_get_set_args_cb(struct unw_frame_info *info, void *data)
2140 {
2141 	struct syscall_get_set_args *args = data;
2142 	struct pt_regs *pt = args->regs;
2143 	unsigned long *krbs, cfm, ndirty;
2144 	int i, count;
2145 
2146 	if (unw_unwind_to_user(info) < 0)
2147 		return;
2148 
2149 	cfm = pt->cr_ifs;
2150 	krbs = (unsigned long *)info->task + IA64_RBS_OFFSET/8;
2151 	ndirty = ia64_rse_num_regs(krbs, krbs + (pt->loadrs >> 19));
2152 
2153 	count = 0;
2154 	if (in_syscall(pt))
2155 		count = min_t(int, args->n, cfm & 0x7f);
2156 
2157 	for (i = 0; i < count; i++) {
2158 		if (args->rw)
2159 			*ia64_rse_skip_regs(krbs, ndirty + i + args->i) =
2160 				args->args[i];
2161 		else
2162 			args->args[i] = *ia64_rse_skip_regs(krbs,
2163 				ndirty + i + args->i);
2164 	}
2165 
2166 	if (!args->rw) {
2167 		while (i < args->n) {
2168 			args->args[i] = 0;
2169 			i++;
2170 		}
2171 	}
2172 }
2173 
2174 void ia64_syscall_get_set_arguments(struct task_struct *task,
2175 	struct pt_regs *regs, unsigned int i, unsigned int n,
2176 	unsigned long *args, int rw)
2177 {
2178 	struct syscall_get_set_args data = {
2179 		.i = i,
2180 		.n = n,
2181 		.args = args,
2182 		.regs = regs,
2183 		.rw = rw,
2184 	};
2185 
2186 	if (task == current)
2187 		unw_init_running(syscall_get_set_args_cb, &data);
2188 	else {
2189 		struct unw_frame_info ufi;
2190 		memset(&ufi, 0, sizeof(ufi));
2191 		unw_init_from_blocked_task(&ufi, task);
2192 		syscall_get_set_args_cb(&ufi, &data);
2193 	}
2194 }
2195