xref: /linux/arch/x86/kernel/ptrace.c (revision f7a904dffe30a02636053d8022498ced7e44d31c)
1 /* By Ross Biro 1/23/92 */
2 /*
3  * Pentium III FXSR, SSE support
4  *	Gareth Hughes <gareth@valinux.com>, May 2000
5  *
6  * BTS tracing
7  *	Markus Metzger <markus.t.metzger@intel.com>, Dec 2007
8  */
9 
10 #include <linux/kernel.h>
11 #include <linux/sched.h>
12 #include <linux/mm.h>
13 #include <linux/smp.h>
14 #include <linux/errno.h>
15 #include <linux/ptrace.h>
16 #include <linux/regset.h>
17 #include <linux/tracehook.h>
18 #include <linux/user.h>
19 #include <linux/elf.h>
20 #include <linux/security.h>
21 #include <linux/audit.h>
22 #include <linux/seccomp.h>
23 #include <linux/signal.h>
24 #include <linux/workqueue.h>
25 #include <linux/perf_event.h>
26 #include <linux/hw_breakpoint.h>
27 
28 #include <asm/uaccess.h>
29 #include <asm/pgtable.h>
30 #include <asm/system.h>
31 #include <asm/processor.h>
32 #include <asm/i387.h>
33 #include <asm/debugreg.h>
34 #include <asm/ldt.h>
35 #include <asm/desc.h>
36 #include <asm/prctl.h>
37 #include <asm/proto.h>
38 #include <asm/ds.h>
39 #include <asm/hw_breakpoint.h>
40 
41 #include "tls.h"
42 
43 #define CREATE_TRACE_POINTS
44 #include <trace/events/syscalls.h>
45 
46 enum x86_regset {
47 	REGSET_GENERAL,
48 	REGSET_FP,
49 	REGSET_XFP,
50 	REGSET_IOPERM64 = REGSET_XFP,
51 	REGSET_TLS,
52 	REGSET_IOPERM32,
53 };
54 
55 struct pt_regs_offset {
56 	const char *name;
57 	int offset;
58 };
59 
60 #define REG_OFFSET_NAME(r) {.name = #r, .offset = offsetof(struct pt_regs, r)}
61 #define REG_OFFSET_END {.name = NULL, .offset = 0}
62 
63 static const struct pt_regs_offset regoffset_table[] = {
64 #ifdef CONFIG_X86_64
65 	REG_OFFSET_NAME(r15),
66 	REG_OFFSET_NAME(r14),
67 	REG_OFFSET_NAME(r13),
68 	REG_OFFSET_NAME(r12),
69 	REG_OFFSET_NAME(r11),
70 	REG_OFFSET_NAME(r10),
71 	REG_OFFSET_NAME(r9),
72 	REG_OFFSET_NAME(r8),
73 #endif
74 	REG_OFFSET_NAME(bx),
75 	REG_OFFSET_NAME(cx),
76 	REG_OFFSET_NAME(dx),
77 	REG_OFFSET_NAME(si),
78 	REG_OFFSET_NAME(di),
79 	REG_OFFSET_NAME(bp),
80 	REG_OFFSET_NAME(ax),
81 #ifdef CONFIG_X86_32
82 	REG_OFFSET_NAME(ds),
83 	REG_OFFSET_NAME(es),
84 	REG_OFFSET_NAME(fs),
85 	REG_OFFSET_NAME(gs),
86 #endif
87 	REG_OFFSET_NAME(orig_ax),
88 	REG_OFFSET_NAME(ip),
89 	REG_OFFSET_NAME(cs),
90 	REG_OFFSET_NAME(flags),
91 	REG_OFFSET_NAME(sp),
92 	REG_OFFSET_NAME(ss),
93 	REG_OFFSET_END,
94 };
95 
96 /**
97  * regs_query_register_offset() - query register offset from its name
98  * @name:	the name of a register
99  *
100  * regs_query_register_offset() returns the offset of a register in struct
101  * pt_regs from its name. If the name is invalid, this returns -EINVAL;
102  */
103 int regs_query_register_offset(const char *name)
104 {
105 	const struct pt_regs_offset *roff;
106 	for (roff = regoffset_table; roff->name != NULL; roff++)
107 		if (!strcmp(roff->name, name))
108 			return roff->offset;
109 	return -EINVAL;
110 }
111 
112 /**
113  * regs_query_register_name() - query register name from its offset
114  * @offset:	the offset of a register in struct pt_regs.
115  *
116  * regs_query_register_name() returns the name of a register from its
117  * offset in struct pt_regs. If the @offset is invalid, this returns NULL;
118  */
119 const char *regs_query_register_name(unsigned int offset)
120 {
121 	const struct pt_regs_offset *roff;
122 	for (roff = regoffset_table; roff->name != NULL; roff++)
123 		if (roff->offset == offset)
124 			return roff->name;
125 	return NULL;
126 }
127 
128 static const int arg_offs_table[] = {
129 #ifdef CONFIG_X86_32
130 	[0] = offsetof(struct pt_regs, ax),
131 	[1] = offsetof(struct pt_regs, dx),
132 	[2] = offsetof(struct pt_regs, cx)
133 #else /* CONFIG_X86_64 */
134 	[0] = offsetof(struct pt_regs, di),
135 	[1] = offsetof(struct pt_regs, si),
136 	[2] = offsetof(struct pt_regs, dx),
137 	[3] = offsetof(struct pt_regs, cx),
138 	[4] = offsetof(struct pt_regs, r8),
139 	[5] = offsetof(struct pt_regs, r9)
140 #endif
141 };
142 
143 /**
144  * regs_get_argument_nth() - get Nth argument at function call
145  * @regs:	pt_regs which contains registers at function entry.
146  * @n:		argument number.
147  *
148  * regs_get_argument_nth() returns @n th argument of a function call.
149  * Since usually the kernel stack will be changed right after function entry,
150  * you must use this at function entry. If the @n th entry is NOT in the
151  * kernel stack or pt_regs, this returns 0.
152  */
153 unsigned long regs_get_argument_nth(struct pt_regs *regs, unsigned int n)
154 {
155 	if (n < ARRAY_SIZE(arg_offs_table))
156 		return *(unsigned long *)((char *)regs + arg_offs_table[n]);
157 	else {
158 		/*
159 		 * The typical case: arg n is on the stack.
160 		 * (Note: stack[0] = return address, so skip it)
161 		 */
162 		n -= ARRAY_SIZE(arg_offs_table);
163 		return regs_get_kernel_stack_nth(regs, 1 + n);
164 	}
165 }
166 
167 /*
168  * does not yet catch signals sent when the child dies.
169  * in exit.c or in signal.c.
170  */
171 
172 /*
173  * Determines which flags the user has access to [1 = access, 0 = no access].
174  */
175 #define FLAG_MASK_32		((unsigned long)			\
176 				 (X86_EFLAGS_CF | X86_EFLAGS_PF |	\
177 				  X86_EFLAGS_AF | X86_EFLAGS_ZF |	\
178 				  X86_EFLAGS_SF | X86_EFLAGS_TF |	\
179 				  X86_EFLAGS_DF | X86_EFLAGS_OF |	\
180 				  X86_EFLAGS_RF | X86_EFLAGS_AC))
181 
182 /*
183  * Determines whether a value may be installed in a segment register.
184  */
185 static inline bool invalid_selector(u16 value)
186 {
187 	return unlikely(value != 0 && (value & SEGMENT_RPL_MASK) != USER_RPL);
188 }
189 
190 #ifdef CONFIG_X86_32
191 
192 #define FLAG_MASK		FLAG_MASK_32
193 
194 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long regno)
195 {
196 	BUILD_BUG_ON(offsetof(struct pt_regs, bx) != 0);
197 	return &regs->bx + (regno >> 2);
198 }
199 
200 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
201 {
202 	/*
203 	 * Returning the value truncates it to 16 bits.
204 	 */
205 	unsigned int retval;
206 	if (offset != offsetof(struct user_regs_struct, gs))
207 		retval = *pt_regs_access(task_pt_regs(task), offset);
208 	else {
209 		if (task == current)
210 			retval = get_user_gs(task_pt_regs(task));
211 		else
212 			retval = task_user_gs(task);
213 	}
214 	return retval;
215 }
216 
217 static int set_segment_reg(struct task_struct *task,
218 			   unsigned long offset, u16 value)
219 {
220 	/*
221 	 * The value argument was already truncated to 16 bits.
222 	 */
223 	if (invalid_selector(value))
224 		return -EIO;
225 
226 	/*
227 	 * For %cs and %ss we cannot permit a null selector.
228 	 * We can permit a bogus selector as long as it has USER_RPL.
229 	 * Null selectors are fine for other segment registers, but
230 	 * we will never get back to user mode with invalid %cs or %ss
231 	 * and will take the trap in iret instead.  Much code relies
232 	 * on user_mode() to distinguish a user trap frame (which can
233 	 * safely use invalid selectors) from a kernel trap frame.
234 	 */
235 	switch (offset) {
236 	case offsetof(struct user_regs_struct, cs):
237 	case offsetof(struct user_regs_struct, ss):
238 		if (unlikely(value == 0))
239 			return -EIO;
240 
241 	default:
242 		*pt_regs_access(task_pt_regs(task), offset) = value;
243 		break;
244 
245 	case offsetof(struct user_regs_struct, gs):
246 		if (task == current)
247 			set_user_gs(task_pt_regs(task), value);
248 		else
249 			task_user_gs(task) = value;
250 	}
251 
252 	return 0;
253 }
254 
255 #else  /* CONFIG_X86_64 */
256 
257 #define FLAG_MASK		(FLAG_MASK_32 | X86_EFLAGS_NT)
258 
259 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long offset)
260 {
261 	BUILD_BUG_ON(offsetof(struct pt_regs, r15) != 0);
262 	return &regs->r15 + (offset / sizeof(regs->r15));
263 }
264 
265 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
266 {
267 	/*
268 	 * Returning the value truncates it to 16 bits.
269 	 */
270 	unsigned int seg;
271 
272 	switch (offset) {
273 	case offsetof(struct user_regs_struct, fs):
274 		if (task == current) {
275 			/* Older gas can't assemble movq %?s,%r?? */
276 			asm("movl %%fs,%0" : "=r" (seg));
277 			return seg;
278 		}
279 		return task->thread.fsindex;
280 	case offsetof(struct user_regs_struct, gs):
281 		if (task == current) {
282 			asm("movl %%gs,%0" : "=r" (seg));
283 			return seg;
284 		}
285 		return task->thread.gsindex;
286 	case offsetof(struct user_regs_struct, ds):
287 		if (task == current) {
288 			asm("movl %%ds,%0" : "=r" (seg));
289 			return seg;
290 		}
291 		return task->thread.ds;
292 	case offsetof(struct user_regs_struct, es):
293 		if (task == current) {
294 			asm("movl %%es,%0" : "=r" (seg));
295 			return seg;
296 		}
297 		return task->thread.es;
298 
299 	case offsetof(struct user_regs_struct, cs):
300 	case offsetof(struct user_regs_struct, ss):
301 		break;
302 	}
303 	return *pt_regs_access(task_pt_regs(task), offset);
304 }
305 
306 static int set_segment_reg(struct task_struct *task,
307 			   unsigned long offset, u16 value)
308 {
309 	/*
310 	 * The value argument was already truncated to 16 bits.
311 	 */
312 	if (invalid_selector(value))
313 		return -EIO;
314 
315 	switch (offset) {
316 	case offsetof(struct user_regs_struct,fs):
317 		/*
318 		 * If this is setting fs as for normal 64-bit use but
319 		 * setting fs_base has implicitly changed it, leave it.
320 		 */
321 		if ((value == FS_TLS_SEL && task->thread.fsindex == 0 &&
322 		     task->thread.fs != 0) ||
323 		    (value == 0 && task->thread.fsindex == FS_TLS_SEL &&
324 		     task->thread.fs == 0))
325 			break;
326 		task->thread.fsindex = value;
327 		if (task == current)
328 			loadsegment(fs, task->thread.fsindex);
329 		break;
330 	case offsetof(struct user_regs_struct,gs):
331 		/*
332 		 * If this is setting gs as for normal 64-bit use but
333 		 * setting gs_base has implicitly changed it, leave it.
334 		 */
335 		if ((value == GS_TLS_SEL && task->thread.gsindex == 0 &&
336 		     task->thread.gs != 0) ||
337 		    (value == 0 && task->thread.gsindex == GS_TLS_SEL &&
338 		     task->thread.gs == 0))
339 			break;
340 		task->thread.gsindex = value;
341 		if (task == current)
342 			load_gs_index(task->thread.gsindex);
343 		break;
344 	case offsetof(struct user_regs_struct,ds):
345 		task->thread.ds = value;
346 		if (task == current)
347 			loadsegment(ds, task->thread.ds);
348 		break;
349 	case offsetof(struct user_regs_struct,es):
350 		task->thread.es = value;
351 		if (task == current)
352 			loadsegment(es, task->thread.es);
353 		break;
354 
355 		/*
356 		 * Can't actually change these in 64-bit mode.
357 		 */
358 	case offsetof(struct user_regs_struct,cs):
359 		if (unlikely(value == 0))
360 			return -EIO;
361 #ifdef CONFIG_IA32_EMULATION
362 		if (test_tsk_thread_flag(task, TIF_IA32))
363 			task_pt_regs(task)->cs = value;
364 #endif
365 		break;
366 	case offsetof(struct user_regs_struct,ss):
367 		if (unlikely(value == 0))
368 			return -EIO;
369 #ifdef CONFIG_IA32_EMULATION
370 		if (test_tsk_thread_flag(task, TIF_IA32))
371 			task_pt_regs(task)->ss = value;
372 #endif
373 		break;
374 	}
375 
376 	return 0;
377 }
378 
379 #endif	/* CONFIG_X86_32 */
380 
381 static unsigned long get_flags(struct task_struct *task)
382 {
383 	unsigned long retval = task_pt_regs(task)->flags;
384 
385 	/*
386 	 * If the debugger set TF, hide it from the readout.
387 	 */
388 	if (test_tsk_thread_flag(task, TIF_FORCED_TF))
389 		retval &= ~X86_EFLAGS_TF;
390 
391 	return retval;
392 }
393 
394 static int set_flags(struct task_struct *task, unsigned long value)
395 {
396 	struct pt_regs *regs = task_pt_regs(task);
397 
398 	/*
399 	 * If the user value contains TF, mark that
400 	 * it was not "us" (the debugger) that set it.
401 	 * If not, make sure it stays set if we had.
402 	 */
403 	if (value & X86_EFLAGS_TF)
404 		clear_tsk_thread_flag(task, TIF_FORCED_TF);
405 	else if (test_tsk_thread_flag(task, TIF_FORCED_TF))
406 		value |= X86_EFLAGS_TF;
407 
408 	regs->flags = (regs->flags & ~FLAG_MASK) | (value & FLAG_MASK);
409 
410 	return 0;
411 }
412 
413 static int putreg(struct task_struct *child,
414 		  unsigned long offset, unsigned long value)
415 {
416 	switch (offset) {
417 	case offsetof(struct user_regs_struct, cs):
418 	case offsetof(struct user_regs_struct, ds):
419 	case offsetof(struct user_regs_struct, es):
420 	case offsetof(struct user_regs_struct, fs):
421 	case offsetof(struct user_regs_struct, gs):
422 	case offsetof(struct user_regs_struct, ss):
423 		return set_segment_reg(child, offset, value);
424 
425 	case offsetof(struct user_regs_struct, flags):
426 		return set_flags(child, value);
427 
428 #ifdef CONFIG_X86_64
429 	case offsetof(struct user_regs_struct,fs_base):
430 		if (value >= TASK_SIZE_OF(child))
431 			return -EIO;
432 		/*
433 		 * When changing the segment base, use do_arch_prctl
434 		 * to set either thread.fs or thread.fsindex and the
435 		 * corresponding GDT slot.
436 		 */
437 		if (child->thread.fs != value)
438 			return do_arch_prctl(child, ARCH_SET_FS, value);
439 		return 0;
440 	case offsetof(struct user_regs_struct,gs_base):
441 		/*
442 		 * Exactly the same here as the %fs handling above.
443 		 */
444 		if (value >= TASK_SIZE_OF(child))
445 			return -EIO;
446 		if (child->thread.gs != value)
447 			return do_arch_prctl(child, ARCH_SET_GS, value);
448 		return 0;
449 #endif
450 	}
451 
452 	*pt_regs_access(task_pt_regs(child), offset) = value;
453 	return 0;
454 }
455 
456 static unsigned long getreg(struct task_struct *task, unsigned long offset)
457 {
458 	switch (offset) {
459 	case offsetof(struct user_regs_struct, cs):
460 	case offsetof(struct user_regs_struct, ds):
461 	case offsetof(struct user_regs_struct, es):
462 	case offsetof(struct user_regs_struct, fs):
463 	case offsetof(struct user_regs_struct, gs):
464 	case offsetof(struct user_regs_struct, ss):
465 		return get_segment_reg(task, offset);
466 
467 	case offsetof(struct user_regs_struct, flags):
468 		return get_flags(task);
469 
470 #ifdef CONFIG_X86_64
471 	case offsetof(struct user_regs_struct, fs_base): {
472 		/*
473 		 * do_arch_prctl may have used a GDT slot instead of
474 		 * the MSR.  To userland, it appears the same either
475 		 * way, except the %fs segment selector might not be 0.
476 		 */
477 		unsigned int seg = task->thread.fsindex;
478 		if (task->thread.fs != 0)
479 			return task->thread.fs;
480 		if (task == current)
481 			asm("movl %%fs,%0" : "=r" (seg));
482 		if (seg != FS_TLS_SEL)
483 			return 0;
484 		return get_desc_base(&task->thread.tls_array[FS_TLS]);
485 	}
486 	case offsetof(struct user_regs_struct, gs_base): {
487 		/*
488 		 * Exactly the same here as the %fs handling above.
489 		 */
490 		unsigned int seg = task->thread.gsindex;
491 		if (task->thread.gs != 0)
492 			return task->thread.gs;
493 		if (task == current)
494 			asm("movl %%gs,%0" : "=r" (seg));
495 		if (seg != GS_TLS_SEL)
496 			return 0;
497 		return get_desc_base(&task->thread.tls_array[GS_TLS]);
498 	}
499 #endif
500 	}
501 
502 	return *pt_regs_access(task_pt_regs(task), offset);
503 }
504 
505 static int genregs_get(struct task_struct *target,
506 		       const struct user_regset *regset,
507 		       unsigned int pos, unsigned int count,
508 		       void *kbuf, void __user *ubuf)
509 {
510 	if (kbuf) {
511 		unsigned long *k = kbuf;
512 		while (count >= sizeof(*k)) {
513 			*k++ = getreg(target, pos);
514 			count -= sizeof(*k);
515 			pos += sizeof(*k);
516 		}
517 	} else {
518 		unsigned long __user *u = ubuf;
519 		while (count >= sizeof(*u)) {
520 			if (__put_user(getreg(target, pos), u++))
521 				return -EFAULT;
522 			count -= sizeof(*u);
523 			pos += sizeof(*u);
524 		}
525 	}
526 
527 	return 0;
528 }
529 
530 static int genregs_set(struct task_struct *target,
531 		       const struct user_regset *regset,
532 		       unsigned int pos, unsigned int count,
533 		       const void *kbuf, const void __user *ubuf)
534 {
535 	int ret = 0;
536 	if (kbuf) {
537 		const unsigned long *k = kbuf;
538 		while (count >= sizeof(*k) && !ret) {
539 			ret = putreg(target, pos, *k++);
540 			count -= sizeof(*k);
541 			pos += sizeof(*k);
542 		}
543 	} else {
544 		const unsigned long  __user *u = ubuf;
545 		while (count >= sizeof(*u) && !ret) {
546 			unsigned long word;
547 			ret = __get_user(word, u++);
548 			if (ret)
549 				break;
550 			ret = putreg(target, pos, word);
551 			count -= sizeof(*u);
552 			pos += sizeof(*u);
553 		}
554 	}
555 	return ret;
556 }
557 
558 static void ptrace_triggered(struct perf_event *bp, int nmi,
559 			     struct perf_sample_data *data,
560 			     struct pt_regs *regs)
561 {
562 	int i;
563 	struct thread_struct *thread = &(current->thread);
564 
565 	/*
566 	 * Store in the virtual DR6 register the fact that the breakpoint
567 	 * was hit so the thread's debugger will see it.
568 	 */
569 	for (i = 0; i < HBP_NUM; i++) {
570 		if (thread->ptrace_bps[i] == bp)
571 			break;
572 	}
573 
574 	thread->debugreg6 |= (DR_TRAP0 << i);
575 }
576 
577 /*
578  * Walk through every ptrace breakpoints for this thread and
579  * build the dr7 value on top of their attributes.
580  *
581  */
582 static unsigned long ptrace_get_dr7(struct perf_event *bp[])
583 {
584 	int i;
585 	int dr7 = 0;
586 	struct arch_hw_breakpoint *info;
587 
588 	for (i = 0; i < HBP_NUM; i++) {
589 		if (bp[i] && !bp[i]->attr.disabled) {
590 			info = counter_arch_bp(bp[i]);
591 			dr7 |= encode_dr7(i, info->len, info->type);
592 		}
593 	}
594 
595 	return dr7;
596 }
597 
598 static int
599 ptrace_modify_breakpoint(struct perf_event *bp, int len, int type,
600 			 struct task_struct *tsk, int disabled)
601 {
602 	int err;
603 	int gen_len, gen_type;
604 	struct perf_event_attr attr;
605 
606 	/*
607 	 * We shoud have at least an inactive breakpoint at this
608 	 * slot. It means the user is writing dr7 without having
609 	 * written the address register first
610 	 */
611 	if (!bp)
612 		return -EINVAL;
613 
614 	err = arch_bp_generic_fields(len, type, &gen_len, &gen_type);
615 	if (err)
616 		return err;
617 
618 	attr = bp->attr;
619 	attr.bp_len = gen_len;
620 	attr.bp_type = gen_type;
621 	attr.disabled = disabled;
622 
623 	return modify_user_hw_breakpoint(bp, &attr);
624 }
625 
626 /*
627  * Handle ptrace writes to debug register 7.
628  */
629 static int ptrace_write_dr7(struct task_struct *tsk, unsigned long data)
630 {
631 	struct thread_struct *thread = &(tsk->thread);
632 	unsigned long old_dr7;
633 	int i, orig_ret = 0, rc = 0;
634 	int enabled, second_pass = 0;
635 	unsigned len, type;
636 	struct perf_event *bp;
637 
638 	data &= ~DR_CONTROL_RESERVED;
639 	old_dr7 = ptrace_get_dr7(thread->ptrace_bps);
640 restore:
641 	/*
642 	 * Loop through all the hardware breakpoints, making the
643 	 * appropriate changes to each.
644 	 */
645 	for (i = 0; i < HBP_NUM; i++) {
646 		enabled = decode_dr7(data, i, &len, &type);
647 		bp = thread->ptrace_bps[i];
648 
649 		if (!enabled) {
650 			if (bp) {
651 				/*
652 				 * Don't unregister the breakpoints right-away,
653 				 * unless all register_user_hw_breakpoint()
654 				 * requests have succeeded. This prevents
655 				 * any window of opportunity for debug
656 				 * register grabbing by other users.
657 				 */
658 				if (!second_pass)
659 					continue;
660 
661 				rc = ptrace_modify_breakpoint(bp, len, type,
662 							      tsk, 1);
663 				if (rc)
664 					break;
665 			}
666 			continue;
667 		}
668 
669 		rc = ptrace_modify_breakpoint(bp, len, type, tsk, 0);
670 		if (rc)
671 			break;
672 	}
673 	/*
674 	 * Make a second pass to free the remaining unused breakpoints
675 	 * or to restore the original breakpoints if an error occurred.
676 	 */
677 	if (!second_pass) {
678 		second_pass = 1;
679 		if (rc < 0) {
680 			orig_ret = rc;
681 			data = old_dr7;
682 		}
683 		goto restore;
684 	}
685 	return ((orig_ret < 0) ? orig_ret : rc);
686 }
687 
688 /*
689  * Handle PTRACE_PEEKUSR calls for the debug register area.
690  */
691 static unsigned long ptrace_get_debugreg(struct task_struct *tsk, int n)
692 {
693 	struct thread_struct *thread = &(tsk->thread);
694 	unsigned long val = 0;
695 
696 	if (n < HBP_NUM) {
697 		struct perf_event *bp;
698 		bp = thread->ptrace_bps[n];
699 		if (!bp)
700 			return 0;
701 		val = bp->hw.info.address;
702 	} else if (n == 6) {
703 		val = thread->debugreg6;
704 	 } else if (n == 7) {
705 		val = thread->ptrace_dr7;
706 	}
707 	return val;
708 }
709 
710 static int ptrace_set_breakpoint_addr(struct task_struct *tsk, int nr,
711 				      unsigned long addr)
712 {
713 	struct perf_event *bp;
714 	struct thread_struct *t = &tsk->thread;
715 	struct perf_event_attr attr;
716 
717 	if (!t->ptrace_bps[nr]) {
718 		hw_breakpoint_init(&attr);
719 		/*
720 		 * Put stub len and type to register (reserve) an inactive but
721 		 * correct bp
722 		 */
723 		attr.bp_addr = addr;
724 		attr.bp_len = HW_BREAKPOINT_LEN_1;
725 		attr.bp_type = HW_BREAKPOINT_W;
726 		attr.disabled = 1;
727 
728 		bp = register_user_hw_breakpoint(&attr, ptrace_triggered, tsk);
729 
730 		/*
731 		 * CHECKME: the previous code returned -EIO if the addr wasn't
732 		 * a valid task virtual addr. The new one will return -EINVAL in
733 		 *  this case.
734 		 * -EINVAL may be what we want for in-kernel breakpoints users,
735 		 * but -EIO looks better for ptrace, since we refuse a register
736 		 * writing for the user. And anyway this is the previous
737 		 * behaviour.
738 		 */
739 		if (IS_ERR(bp))
740 			return PTR_ERR(bp);
741 
742 		t->ptrace_bps[nr] = bp;
743 	} else {
744 		int err;
745 
746 		bp = t->ptrace_bps[nr];
747 
748 		attr = bp->attr;
749 		attr.bp_addr = addr;
750 		err = modify_user_hw_breakpoint(bp, &attr);
751 		if (err)
752 			return err;
753 	}
754 
755 
756 	return 0;
757 }
758 
759 /*
760  * Handle PTRACE_POKEUSR calls for the debug register area.
761  */
762 int ptrace_set_debugreg(struct task_struct *tsk, int n, unsigned long val)
763 {
764 	struct thread_struct *thread = &(tsk->thread);
765 	int rc = 0;
766 
767 	/* There are no DR4 or DR5 registers */
768 	if (n == 4 || n == 5)
769 		return -EIO;
770 
771 	if (n == 6) {
772 		thread->debugreg6 = val;
773 		goto ret_path;
774 	}
775 	if (n < HBP_NUM) {
776 		rc = ptrace_set_breakpoint_addr(tsk, n, val);
777 		if (rc)
778 			return rc;
779 	}
780 	/* All that's left is DR7 */
781 	if (n == 7) {
782 		rc = ptrace_write_dr7(tsk, val);
783 		if (!rc)
784 			thread->ptrace_dr7 = val;
785 	}
786 
787 ret_path:
788 	return rc;
789 }
790 
791 /*
792  * These access the current or another (stopped) task's io permission
793  * bitmap for debugging or core dump.
794  */
795 static int ioperm_active(struct task_struct *target,
796 			 const struct user_regset *regset)
797 {
798 	return target->thread.io_bitmap_max / regset->size;
799 }
800 
801 static int ioperm_get(struct task_struct *target,
802 		      const struct user_regset *regset,
803 		      unsigned int pos, unsigned int count,
804 		      void *kbuf, void __user *ubuf)
805 {
806 	if (!target->thread.io_bitmap_ptr)
807 		return -ENXIO;
808 
809 	return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
810 				   target->thread.io_bitmap_ptr,
811 				   0, IO_BITMAP_BYTES);
812 }
813 
814 #ifdef CONFIG_X86_PTRACE_BTS
815 /*
816  * A branch trace store context.
817  *
818  * Contexts may only be installed by ptrace_bts_config() and only for
819  * ptraced tasks.
820  *
821  * Contexts are destroyed when the tracee is detached from the tracer.
822  * The actual destruction work requires interrupts enabled, so the
823  * work is deferred and will be scheduled during __ptrace_unlink().
824  *
825  * Contexts hold an additional task_struct reference on the traced
826  * task, as well as a reference on the tracer's mm.
827  *
828  * Ptrace already holds a task_struct for the duration of ptrace operations,
829  * but since destruction is deferred, it may be executed after both
830  * tracer and tracee exited.
831  */
832 struct bts_context {
833 	/* The branch trace handle. */
834 	struct bts_tracer	*tracer;
835 
836 	/* The buffer used to store the branch trace and its size. */
837 	void			*buffer;
838 	unsigned int		size;
839 
840 	/* The mm that paid for the above buffer. */
841 	struct mm_struct	*mm;
842 
843 	/* The task this context belongs to. */
844 	struct task_struct	*task;
845 
846 	/* The signal to send on a bts buffer overflow. */
847 	unsigned int		bts_ovfl_signal;
848 
849 	/* The work struct to destroy a context. */
850 	struct work_struct	work;
851 };
852 
853 static int alloc_bts_buffer(struct bts_context *context, unsigned int size)
854 {
855 	void *buffer = NULL;
856 	int err = -ENOMEM;
857 
858 	err = account_locked_memory(current->mm, current->signal->rlim, size);
859 	if (err < 0)
860 		return err;
861 
862 	buffer = kzalloc(size, GFP_KERNEL);
863 	if (!buffer)
864 		goto out_refund;
865 
866 	context->buffer = buffer;
867 	context->size = size;
868 	context->mm = get_task_mm(current);
869 
870 	return 0;
871 
872  out_refund:
873 	refund_locked_memory(current->mm, size);
874 	return err;
875 }
876 
877 static inline void free_bts_buffer(struct bts_context *context)
878 {
879 	if (!context->buffer)
880 		return;
881 
882 	kfree(context->buffer);
883 	context->buffer = NULL;
884 
885 	refund_locked_memory(context->mm, context->size);
886 	context->size = 0;
887 
888 	mmput(context->mm);
889 	context->mm = NULL;
890 }
891 
892 static void free_bts_context_work(struct work_struct *w)
893 {
894 	struct bts_context *context;
895 
896 	context = container_of(w, struct bts_context, work);
897 
898 	ds_release_bts(context->tracer);
899 	put_task_struct(context->task);
900 	free_bts_buffer(context);
901 	kfree(context);
902 }
903 
904 static inline void free_bts_context(struct bts_context *context)
905 {
906 	INIT_WORK(&context->work, free_bts_context_work);
907 	schedule_work(&context->work);
908 }
909 
910 static inline struct bts_context *alloc_bts_context(struct task_struct *task)
911 {
912 	struct bts_context *context = kzalloc(sizeof(*context), GFP_KERNEL);
913 	if (context) {
914 		context->task = task;
915 		task->bts = context;
916 
917 		get_task_struct(task);
918 	}
919 
920 	return context;
921 }
922 
923 static int ptrace_bts_read_record(struct task_struct *child, size_t index,
924 				  struct bts_struct __user *out)
925 {
926 	struct bts_context *context;
927 	const struct bts_trace *trace;
928 	struct bts_struct bts;
929 	const unsigned char *at;
930 	int error;
931 
932 	context = child->bts;
933 	if (!context)
934 		return -ESRCH;
935 
936 	trace = ds_read_bts(context->tracer);
937 	if (!trace)
938 		return -ESRCH;
939 
940 	at = trace->ds.top - ((index + 1) * trace->ds.size);
941 	if ((void *)at < trace->ds.begin)
942 		at += (trace->ds.n * trace->ds.size);
943 
944 	if (!trace->read)
945 		return -EOPNOTSUPP;
946 
947 	error = trace->read(context->tracer, at, &bts);
948 	if (error < 0)
949 		return error;
950 
951 	if (copy_to_user(out, &bts, sizeof(bts)))
952 		return -EFAULT;
953 
954 	return sizeof(bts);
955 }
956 
957 static int ptrace_bts_drain(struct task_struct *child,
958 			    long size,
959 			    struct bts_struct __user *out)
960 {
961 	struct bts_context *context;
962 	const struct bts_trace *trace;
963 	const unsigned char *at;
964 	int error, drained = 0;
965 
966 	context = child->bts;
967 	if (!context)
968 		return -ESRCH;
969 
970 	trace = ds_read_bts(context->tracer);
971 	if (!trace)
972 		return -ESRCH;
973 
974 	if (!trace->read)
975 		return -EOPNOTSUPP;
976 
977 	if (size < (trace->ds.top - trace->ds.begin))
978 		return -EIO;
979 
980 	for (at = trace->ds.begin; (void *)at < trace->ds.top;
981 	     out++, drained++, at += trace->ds.size) {
982 		struct bts_struct bts;
983 
984 		error = trace->read(context->tracer, at, &bts);
985 		if (error < 0)
986 			return error;
987 
988 		if (copy_to_user(out, &bts, sizeof(bts)))
989 			return -EFAULT;
990 	}
991 
992 	memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size);
993 
994 	error = ds_reset_bts(context->tracer);
995 	if (error < 0)
996 		return error;
997 
998 	return drained;
999 }
1000 
1001 static int ptrace_bts_config(struct task_struct *child,
1002 			     long cfg_size,
1003 			     const struct ptrace_bts_config __user *ucfg)
1004 {
1005 	struct bts_context *context;
1006 	struct ptrace_bts_config cfg;
1007 	unsigned int flags = 0;
1008 
1009 	if (cfg_size < sizeof(cfg))
1010 		return -EIO;
1011 
1012 	if (copy_from_user(&cfg, ucfg, sizeof(cfg)))
1013 		return -EFAULT;
1014 
1015 	context = child->bts;
1016 	if (!context)
1017 		context = alloc_bts_context(child);
1018 	if (!context)
1019 		return -ENOMEM;
1020 
1021 	if (cfg.flags & PTRACE_BTS_O_SIGNAL) {
1022 		if (!cfg.signal)
1023 			return -EINVAL;
1024 
1025 		return -EOPNOTSUPP;
1026 		context->bts_ovfl_signal = cfg.signal;
1027 	}
1028 
1029 	ds_release_bts(context->tracer);
1030 	context->tracer = NULL;
1031 
1032 	if ((cfg.flags & PTRACE_BTS_O_ALLOC) && (cfg.size != context->size)) {
1033 		int err;
1034 
1035 		free_bts_buffer(context);
1036 		if (!cfg.size)
1037 			return 0;
1038 
1039 		err = alloc_bts_buffer(context, cfg.size);
1040 		if (err < 0)
1041 			return err;
1042 	}
1043 
1044 	if (cfg.flags & PTRACE_BTS_O_TRACE)
1045 		flags |= BTS_USER;
1046 
1047 	if (cfg.flags & PTRACE_BTS_O_SCHED)
1048 		flags |= BTS_TIMESTAMPS;
1049 
1050 	context->tracer =
1051 		ds_request_bts_task(child, context->buffer, context->size,
1052 				    NULL, (size_t)-1, flags);
1053 	if (unlikely(IS_ERR(context->tracer))) {
1054 		int error = PTR_ERR(context->tracer);
1055 
1056 		free_bts_buffer(context);
1057 		context->tracer = NULL;
1058 		return error;
1059 	}
1060 
1061 	return sizeof(cfg);
1062 }
1063 
1064 static int ptrace_bts_status(struct task_struct *child,
1065 			     long cfg_size,
1066 			     struct ptrace_bts_config __user *ucfg)
1067 {
1068 	struct bts_context *context;
1069 	const struct bts_trace *trace;
1070 	struct ptrace_bts_config cfg;
1071 
1072 	context = child->bts;
1073 	if (!context)
1074 		return -ESRCH;
1075 
1076 	if (cfg_size < sizeof(cfg))
1077 		return -EIO;
1078 
1079 	trace = ds_read_bts(context->tracer);
1080 	if (!trace)
1081 		return -ESRCH;
1082 
1083 	memset(&cfg, 0, sizeof(cfg));
1084 	cfg.size	= trace->ds.end - trace->ds.begin;
1085 	cfg.signal	= context->bts_ovfl_signal;
1086 	cfg.bts_size	= sizeof(struct bts_struct);
1087 
1088 	if (cfg.signal)
1089 		cfg.flags |= PTRACE_BTS_O_SIGNAL;
1090 
1091 	if (trace->ds.flags & BTS_USER)
1092 		cfg.flags |= PTRACE_BTS_O_TRACE;
1093 
1094 	if (trace->ds.flags & BTS_TIMESTAMPS)
1095 		cfg.flags |= PTRACE_BTS_O_SCHED;
1096 
1097 	if (copy_to_user(ucfg, &cfg, sizeof(cfg)))
1098 		return -EFAULT;
1099 
1100 	return sizeof(cfg);
1101 }
1102 
1103 static int ptrace_bts_clear(struct task_struct *child)
1104 {
1105 	struct bts_context *context;
1106 	const struct bts_trace *trace;
1107 
1108 	context = child->bts;
1109 	if (!context)
1110 		return -ESRCH;
1111 
1112 	trace = ds_read_bts(context->tracer);
1113 	if (!trace)
1114 		return -ESRCH;
1115 
1116 	memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size);
1117 
1118 	return ds_reset_bts(context->tracer);
1119 }
1120 
1121 static int ptrace_bts_size(struct task_struct *child)
1122 {
1123 	struct bts_context *context;
1124 	const struct bts_trace *trace;
1125 
1126 	context = child->bts;
1127 	if (!context)
1128 		return -ESRCH;
1129 
1130 	trace = ds_read_bts(context->tracer);
1131 	if (!trace)
1132 		return -ESRCH;
1133 
1134 	return (trace->ds.top - trace->ds.begin) / trace->ds.size;
1135 }
1136 
1137 /*
1138  * Called from __ptrace_unlink() after the child has been moved back
1139  * to its original parent.
1140  */
1141 void ptrace_bts_untrace(struct task_struct *child)
1142 {
1143 	if (unlikely(child->bts)) {
1144 		free_bts_context(child->bts);
1145 		child->bts = NULL;
1146 	}
1147 }
1148 #endif /* CONFIG_X86_PTRACE_BTS */
1149 
1150 /*
1151  * Called by kernel/ptrace.c when detaching..
1152  *
1153  * Make sure the single step bit is not set.
1154  */
1155 void ptrace_disable(struct task_struct *child)
1156 {
1157 	user_disable_single_step(child);
1158 #ifdef TIF_SYSCALL_EMU
1159 	clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
1160 #endif
1161 }
1162 
1163 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1164 static const struct user_regset_view user_x86_32_view; /* Initialized below. */
1165 #endif
1166 
1167 long arch_ptrace(struct task_struct *child, long request, long addr, long data)
1168 {
1169 	int ret;
1170 	unsigned long __user *datap = (unsigned long __user *)data;
1171 
1172 	switch (request) {
1173 	/* read the word at location addr in the USER area. */
1174 	case PTRACE_PEEKUSR: {
1175 		unsigned long tmp;
1176 
1177 		ret = -EIO;
1178 		if ((addr & (sizeof(data) - 1)) || addr < 0 ||
1179 		    addr >= sizeof(struct user))
1180 			break;
1181 
1182 		tmp = 0;  /* Default return condition */
1183 		if (addr < sizeof(struct user_regs_struct))
1184 			tmp = getreg(child, addr);
1185 		else if (addr >= offsetof(struct user, u_debugreg[0]) &&
1186 			 addr <= offsetof(struct user, u_debugreg[7])) {
1187 			addr -= offsetof(struct user, u_debugreg[0]);
1188 			tmp = ptrace_get_debugreg(child, addr / sizeof(data));
1189 		}
1190 		ret = put_user(tmp, datap);
1191 		break;
1192 	}
1193 
1194 	case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
1195 		ret = -EIO;
1196 		if ((addr & (sizeof(data) - 1)) || addr < 0 ||
1197 		    addr >= sizeof(struct user))
1198 			break;
1199 
1200 		if (addr < sizeof(struct user_regs_struct))
1201 			ret = putreg(child, addr, data);
1202 		else if (addr >= offsetof(struct user, u_debugreg[0]) &&
1203 			 addr <= offsetof(struct user, u_debugreg[7])) {
1204 			addr -= offsetof(struct user, u_debugreg[0]);
1205 			ret = ptrace_set_debugreg(child,
1206 						  addr / sizeof(data), data);
1207 		}
1208 		break;
1209 
1210 	case PTRACE_GETREGS:	/* Get all gp regs from the child. */
1211 		return copy_regset_to_user(child,
1212 					   task_user_regset_view(current),
1213 					   REGSET_GENERAL,
1214 					   0, sizeof(struct user_regs_struct),
1215 					   datap);
1216 
1217 	case PTRACE_SETREGS:	/* Set all gp regs in the child. */
1218 		return copy_regset_from_user(child,
1219 					     task_user_regset_view(current),
1220 					     REGSET_GENERAL,
1221 					     0, sizeof(struct user_regs_struct),
1222 					     datap);
1223 
1224 	case PTRACE_GETFPREGS:	/* Get the child FPU state. */
1225 		return copy_regset_to_user(child,
1226 					   task_user_regset_view(current),
1227 					   REGSET_FP,
1228 					   0, sizeof(struct user_i387_struct),
1229 					   datap);
1230 
1231 	case PTRACE_SETFPREGS:	/* Set the child FPU state. */
1232 		return copy_regset_from_user(child,
1233 					     task_user_regset_view(current),
1234 					     REGSET_FP,
1235 					     0, sizeof(struct user_i387_struct),
1236 					     datap);
1237 
1238 #ifdef CONFIG_X86_32
1239 	case PTRACE_GETFPXREGS:	/* Get the child extended FPU state. */
1240 		return copy_regset_to_user(child, &user_x86_32_view,
1241 					   REGSET_XFP,
1242 					   0, sizeof(struct user_fxsr_struct),
1243 					   datap) ? -EIO : 0;
1244 
1245 	case PTRACE_SETFPXREGS:	/* Set the child extended FPU state. */
1246 		return copy_regset_from_user(child, &user_x86_32_view,
1247 					     REGSET_XFP,
1248 					     0, sizeof(struct user_fxsr_struct),
1249 					     datap) ? -EIO : 0;
1250 #endif
1251 
1252 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1253 	case PTRACE_GET_THREAD_AREA:
1254 		if (addr < 0)
1255 			return -EIO;
1256 		ret = do_get_thread_area(child, addr,
1257 					 (struct user_desc __user *) data);
1258 		break;
1259 
1260 	case PTRACE_SET_THREAD_AREA:
1261 		if (addr < 0)
1262 			return -EIO;
1263 		ret = do_set_thread_area(child, addr,
1264 					 (struct user_desc __user *) data, 0);
1265 		break;
1266 #endif
1267 
1268 #ifdef CONFIG_X86_64
1269 		/* normal 64bit interface to access TLS data.
1270 		   Works just like arch_prctl, except that the arguments
1271 		   are reversed. */
1272 	case PTRACE_ARCH_PRCTL:
1273 		ret = do_arch_prctl(child, data, addr);
1274 		break;
1275 #endif
1276 
1277 	/*
1278 	 * These bits need more cooking - not enabled yet:
1279 	 */
1280 #ifdef CONFIG_X86_PTRACE_BTS
1281 	case PTRACE_BTS_CONFIG:
1282 		ret = ptrace_bts_config
1283 			(child, data, (struct ptrace_bts_config __user *)addr);
1284 		break;
1285 
1286 	case PTRACE_BTS_STATUS:
1287 		ret = ptrace_bts_status
1288 			(child, data, (struct ptrace_bts_config __user *)addr);
1289 		break;
1290 
1291 	case PTRACE_BTS_SIZE:
1292 		ret = ptrace_bts_size(child);
1293 		break;
1294 
1295 	case PTRACE_BTS_GET:
1296 		ret = ptrace_bts_read_record
1297 			(child, data, (struct bts_struct __user *) addr);
1298 		break;
1299 
1300 	case PTRACE_BTS_CLEAR:
1301 		ret = ptrace_bts_clear(child);
1302 		break;
1303 
1304 	case PTRACE_BTS_DRAIN:
1305 		ret = ptrace_bts_drain
1306 			(child, data, (struct bts_struct __user *) addr);
1307 		break;
1308 #endif /* CONFIG_X86_PTRACE_BTS */
1309 
1310 	default:
1311 		ret = ptrace_request(child, request, addr, data);
1312 		break;
1313 	}
1314 
1315 	return ret;
1316 }
1317 
1318 #ifdef CONFIG_IA32_EMULATION
1319 
1320 #include <linux/compat.h>
1321 #include <linux/syscalls.h>
1322 #include <asm/ia32.h>
1323 #include <asm/user32.h>
1324 
1325 #define R32(l,q)							\
1326 	case offsetof(struct user32, regs.l):				\
1327 		regs->q = value; break
1328 
1329 #define SEG32(rs)							\
1330 	case offsetof(struct user32, regs.rs):				\
1331 		return set_segment_reg(child,				\
1332 				       offsetof(struct user_regs_struct, rs), \
1333 				       value);				\
1334 		break
1335 
1336 static int putreg32(struct task_struct *child, unsigned regno, u32 value)
1337 {
1338 	struct pt_regs *regs = task_pt_regs(child);
1339 
1340 	switch (regno) {
1341 
1342 	SEG32(cs);
1343 	SEG32(ds);
1344 	SEG32(es);
1345 	SEG32(fs);
1346 	SEG32(gs);
1347 	SEG32(ss);
1348 
1349 	R32(ebx, bx);
1350 	R32(ecx, cx);
1351 	R32(edx, dx);
1352 	R32(edi, di);
1353 	R32(esi, si);
1354 	R32(ebp, bp);
1355 	R32(eax, ax);
1356 	R32(eip, ip);
1357 	R32(esp, sp);
1358 
1359 	case offsetof(struct user32, regs.orig_eax):
1360 		/*
1361 		 * A 32-bit debugger setting orig_eax means to restore
1362 		 * the state of the task restarting a 32-bit syscall.
1363 		 * Make sure we interpret the -ERESTART* codes correctly
1364 		 * in case the task is not actually still sitting at the
1365 		 * exit from a 32-bit syscall with TS_COMPAT still set.
1366 		 */
1367 		regs->orig_ax = value;
1368 		if (syscall_get_nr(child, regs) >= 0)
1369 			task_thread_info(child)->status |= TS_COMPAT;
1370 		break;
1371 
1372 	case offsetof(struct user32, regs.eflags):
1373 		return set_flags(child, value);
1374 
1375 	case offsetof(struct user32, u_debugreg[0]) ...
1376 		offsetof(struct user32, u_debugreg[7]):
1377 		regno -= offsetof(struct user32, u_debugreg[0]);
1378 		return ptrace_set_debugreg(child, regno / 4, value);
1379 
1380 	default:
1381 		if (regno > sizeof(struct user32) || (regno & 3))
1382 			return -EIO;
1383 
1384 		/*
1385 		 * Other dummy fields in the virtual user structure
1386 		 * are ignored
1387 		 */
1388 		break;
1389 	}
1390 	return 0;
1391 }
1392 
1393 #undef R32
1394 #undef SEG32
1395 
1396 #define R32(l,q)							\
1397 	case offsetof(struct user32, regs.l):				\
1398 		*val = regs->q; break
1399 
1400 #define SEG32(rs)							\
1401 	case offsetof(struct user32, regs.rs):				\
1402 		*val = get_segment_reg(child,				\
1403 				       offsetof(struct user_regs_struct, rs)); \
1404 		break
1405 
1406 static int getreg32(struct task_struct *child, unsigned regno, u32 *val)
1407 {
1408 	struct pt_regs *regs = task_pt_regs(child);
1409 
1410 	switch (regno) {
1411 
1412 	SEG32(ds);
1413 	SEG32(es);
1414 	SEG32(fs);
1415 	SEG32(gs);
1416 
1417 	R32(cs, cs);
1418 	R32(ss, ss);
1419 	R32(ebx, bx);
1420 	R32(ecx, cx);
1421 	R32(edx, dx);
1422 	R32(edi, di);
1423 	R32(esi, si);
1424 	R32(ebp, bp);
1425 	R32(eax, ax);
1426 	R32(orig_eax, orig_ax);
1427 	R32(eip, ip);
1428 	R32(esp, sp);
1429 
1430 	case offsetof(struct user32, regs.eflags):
1431 		*val = get_flags(child);
1432 		break;
1433 
1434 	case offsetof(struct user32, u_debugreg[0]) ...
1435 		offsetof(struct user32, u_debugreg[7]):
1436 		regno -= offsetof(struct user32, u_debugreg[0]);
1437 		*val = ptrace_get_debugreg(child, regno / 4);
1438 		break;
1439 
1440 	default:
1441 		if (regno > sizeof(struct user32) || (regno & 3))
1442 			return -EIO;
1443 
1444 		/*
1445 		 * Other dummy fields in the virtual user structure
1446 		 * are ignored
1447 		 */
1448 		*val = 0;
1449 		break;
1450 	}
1451 	return 0;
1452 }
1453 
1454 #undef R32
1455 #undef SEG32
1456 
1457 static int genregs32_get(struct task_struct *target,
1458 			 const struct user_regset *regset,
1459 			 unsigned int pos, unsigned int count,
1460 			 void *kbuf, void __user *ubuf)
1461 {
1462 	if (kbuf) {
1463 		compat_ulong_t *k = kbuf;
1464 		while (count >= sizeof(*k)) {
1465 			getreg32(target, pos, k++);
1466 			count -= sizeof(*k);
1467 			pos += sizeof(*k);
1468 		}
1469 	} else {
1470 		compat_ulong_t __user *u = ubuf;
1471 		while (count >= sizeof(*u)) {
1472 			compat_ulong_t word;
1473 			getreg32(target, pos, &word);
1474 			if (__put_user(word, u++))
1475 				return -EFAULT;
1476 			count -= sizeof(*u);
1477 			pos += sizeof(*u);
1478 		}
1479 	}
1480 
1481 	return 0;
1482 }
1483 
1484 static int genregs32_set(struct task_struct *target,
1485 			 const struct user_regset *regset,
1486 			 unsigned int pos, unsigned int count,
1487 			 const void *kbuf, const void __user *ubuf)
1488 {
1489 	int ret = 0;
1490 	if (kbuf) {
1491 		const compat_ulong_t *k = kbuf;
1492 		while (count >= sizeof(*k) && !ret) {
1493 			ret = putreg32(target, pos, *k++);
1494 			count -= sizeof(*k);
1495 			pos += sizeof(*k);
1496 		}
1497 	} else {
1498 		const compat_ulong_t __user *u = ubuf;
1499 		while (count >= sizeof(*u) && !ret) {
1500 			compat_ulong_t word;
1501 			ret = __get_user(word, u++);
1502 			if (ret)
1503 				break;
1504 			ret = putreg32(target, pos, word);
1505 			count -= sizeof(*u);
1506 			pos += sizeof(*u);
1507 		}
1508 	}
1509 	return ret;
1510 }
1511 
1512 long compat_arch_ptrace(struct task_struct *child, compat_long_t request,
1513 			compat_ulong_t caddr, compat_ulong_t cdata)
1514 {
1515 	unsigned long addr = caddr;
1516 	unsigned long data = cdata;
1517 	void __user *datap = compat_ptr(data);
1518 	int ret;
1519 	__u32 val;
1520 
1521 	switch (request) {
1522 	case PTRACE_PEEKUSR:
1523 		ret = getreg32(child, addr, &val);
1524 		if (ret == 0)
1525 			ret = put_user(val, (__u32 __user *)datap);
1526 		break;
1527 
1528 	case PTRACE_POKEUSR:
1529 		ret = putreg32(child, addr, data);
1530 		break;
1531 
1532 	case PTRACE_GETREGS:	/* Get all gp regs from the child. */
1533 		return copy_regset_to_user(child, &user_x86_32_view,
1534 					   REGSET_GENERAL,
1535 					   0, sizeof(struct user_regs_struct32),
1536 					   datap);
1537 
1538 	case PTRACE_SETREGS:	/* Set all gp regs in the child. */
1539 		return copy_regset_from_user(child, &user_x86_32_view,
1540 					     REGSET_GENERAL, 0,
1541 					     sizeof(struct user_regs_struct32),
1542 					     datap);
1543 
1544 	case PTRACE_GETFPREGS:	/* Get the child FPU state. */
1545 		return copy_regset_to_user(child, &user_x86_32_view,
1546 					   REGSET_FP, 0,
1547 					   sizeof(struct user_i387_ia32_struct),
1548 					   datap);
1549 
1550 	case PTRACE_SETFPREGS:	/* Set the child FPU state. */
1551 		return copy_regset_from_user(
1552 			child, &user_x86_32_view, REGSET_FP,
1553 			0, sizeof(struct user_i387_ia32_struct), datap);
1554 
1555 	case PTRACE_GETFPXREGS:	/* Get the child extended FPU state. */
1556 		return copy_regset_to_user(child, &user_x86_32_view,
1557 					   REGSET_XFP, 0,
1558 					   sizeof(struct user32_fxsr_struct),
1559 					   datap);
1560 
1561 	case PTRACE_SETFPXREGS:	/* Set the child extended FPU state. */
1562 		return copy_regset_from_user(child, &user_x86_32_view,
1563 					     REGSET_XFP, 0,
1564 					     sizeof(struct user32_fxsr_struct),
1565 					     datap);
1566 
1567 	case PTRACE_GET_THREAD_AREA:
1568 	case PTRACE_SET_THREAD_AREA:
1569 #ifdef CONFIG_X86_PTRACE_BTS
1570 	case PTRACE_BTS_CONFIG:
1571 	case PTRACE_BTS_STATUS:
1572 	case PTRACE_BTS_SIZE:
1573 	case PTRACE_BTS_GET:
1574 	case PTRACE_BTS_CLEAR:
1575 	case PTRACE_BTS_DRAIN:
1576 #endif /* CONFIG_X86_PTRACE_BTS */
1577 		return arch_ptrace(child, request, addr, data);
1578 
1579 	default:
1580 		return compat_ptrace_request(child, request, addr, data);
1581 	}
1582 
1583 	return ret;
1584 }
1585 
1586 #endif	/* CONFIG_IA32_EMULATION */
1587 
1588 #ifdef CONFIG_X86_64
1589 
1590 static const struct user_regset x86_64_regsets[] = {
1591 	[REGSET_GENERAL] = {
1592 		.core_note_type = NT_PRSTATUS,
1593 		.n = sizeof(struct user_regs_struct) / sizeof(long),
1594 		.size = sizeof(long), .align = sizeof(long),
1595 		.get = genregs_get, .set = genregs_set
1596 	},
1597 	[REGSET_FP] = {
1598 		.core_note_type = NT_PRFPREG,
1599 		.n = sizeof(struct user_i387_struct) / sizeof(long),
1600 		.size = sizeof(long), .align = sizeof(long),
1601 		.active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1602 	},
1603 	[REGSET_IOPERM64] = {
1604 		.core_note_type = NT_386_IOPERM,
1605 		.n = IO_BITMAP_LONGS,
1606 		.size = sizeof(long), .align = sizeof(long),
1607 		.active = ioperm_active, .get = ioperm_get
1608 	},
1609 };
1610 
1611 static const struct user_regset_view user_x86_64_view = {
1612 	.name = "x86_64", .e_machine = EM_X86_64,
1613 	.regsets = x86_64_regsets, .n = ARRAY_SIZE(x86_64_regsets)
1614 };
1615 
1616 #else  /* CONFIG_X86_32 */
1617 
1618 #define user_regs_struct32	user_regs_struct
1619 #define genregs32_get		genregs_get
1620 #define genregs32_set		genregs_set
1621 
1622 #define user_i387_ia32_struct	user_i387_struct
1623 #define user32_fxsr_struct	user_fxsr_struct
1624 
1625 #endif	/* CONFIG_X86_64 */
1626 
1627 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1628 static const struct user_regset x86_32_regsets[] = {
1629 	[REGSET_GENERAL] = {
1630 		.core_note_type = NT_PRSTATUS,
1631 		.n = sizeof(struct user_regs_struct32) / sizeof(u32),
1632 		.size = sizeof(u32), .align = sizeof(u32),
1633 		.get = genregs32_get, .set = genregs32_set
1634 	},
1635 	[REGSET_FP] = {
1636 		.core_note_type = NT_PRFPREG,
1637 		.n = sizeof(struct user_i387_ia32_struct) / sizeof(u32),
1638 		.size = sizeof(u32), .align = sizeof(u32),
1639 		.active = fpregs_active, .get = fpregs_get, .set = fpregs_set
1640 	},
1641 	[REGSET_XFP] = {
1642 		.core_note_type = NT_PRXFPREG,
1643 		.n = sizeof(struct user32_fxsr_struct) / sizeof(u32),
1644 		.size = sizeof(u32), .align = sizeof(u32),
1645 		.active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1646 	},
1647 	[REGSET_TLS] = {
1648 		.core_note_type = NT_386_TLS,
1649 		.n = GDT_ENTRY_TLS_ENTRIES, .bias = GDT_ENTRY_TLS_MIN,
1650 		.size = sizeof(struct user_desc),
1651 		.align = sizeof(struct user_desc),
1652 		.active = regset_tls_active,
1653 		.get = regset_tls_get, .set = regset_tls_set
1654 	},
1655 	[REGSET_IOPERM32] = {
1656 		.core_note_type = NT_386_IOPERM,
1657 		.n = IO_BITMAP_BYTES / sizeof(u32),
1658 		.size = sizeof(u32), .align = sizeof(u32),
1659 		.active = ioperm_active, .get = ioperm_get
1660 	},
1661 };
1662 
1663 static const struct user_regset_view user_x86_32_view = {
1664 	.name = "i386", .e_machine = EM_386,
1665 	.regsets = x86_32_regsets, .n = ARRAY_SIZE(x86_32_regsets)
1666 };
1667 #endif
1668 
1669 const struct user_regset_view *task_user_regset_view(struct task_struct *task)
1670 {
1671 #ifdef CONFIG_IA32_EMULATION
1672 	if (test_tsk_thread_flag(task, TIF_IA32))
1673 #endif
1674 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1675 		return &user_x86_32_view;
1676 #endif
1677 #ifdef CONFIG_X86_64
1678 	return &user_x86_64_view;
1679 #endif
1680 }
1681 
1682 static void fill_sigtrap_info(struct task_struct *tsk,
1683 				struct pt_regs *regs,
1684 				int error_code, int si_code,
1685 				struct siginfo *info)
1686 {
1687 	tsk->thread.trap_no = 1;
1688 	tsk->thread.error_code = error_code;
1689 
1690 	memset(info, 0, sizeof(*info));
1691 	info->si_signo = SIGTRAP;
1692 	info->si_code = si_code;
1693 	info->si_addr = user_mode_vm(regs) ? (void __user *)regs->ip : NULL;
1694 }
1695 
1696 void user_single_step_siginfo(struct task_struct *tsk,
1697 				struct pt_regs *regs,
1698 				struct siginfo *info)
1699 {
1700 	fill_sigtrap_info(tsk, regs, 0, TRAP_BRKPT, info);
1701 }
1702 
1703 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs,
1704 					 int error_code, int si_code)
1705 {
1706 	struct siginfo info;
1707 
1708 	fill_sigtrap_info(tsk, regs, error_code, si_code, &info);
1709 	/* Send us the fake SIGTRAP */
1710 	force_sig_info(SIGTRAP, &info, tsk);
1711 }
1712 
1713 
1714 #ifdef CONFIG_X86_32
1715 # define IS_IA32	1
1716 #elif defined CONFIG_IA32_EMULATION
1717 # define IS_IA32	is_compat_task()
1718 #else
1719 # define IS_IA32	0
1720 #endif
1721 
1722 /*
1723  * We must return the syscall number to actually look up in the table.
1724  * This can be -1L to skip running any syscall at all.
1725  */
1726 asmregparm long syscall_trace_enter(struct pt_regs *regs)
1727 {
1728 	long ret = 0;
1729 
1730 	/*
1731 	 * If we stepped into a sysenter/syscall insn, it trapped in
1732 	 * kernel mode; do_debug() cleared TF and set TIF_SINGLESTEP.
1733 	 * If user-mode had set TF itself, then it's still clear from
1734 	 * do_debug() and we need to set it again to restore the user
1735 	 * state.  If we entered on the slow path, TF was already set.
1736 	 */
1737 	if (test_thread_flag(TIF_SINGLESTEP))
1738 		regs->flags |= X86_EFLAGS_TF;
1739 
1740 	/* do the secure computing check first */
1741 	secure_computing(regs->orig_ax);
1742 
1743 	if (unlikely(test_thread_flag(TIF_SYSCALL_EMU)))
1744 		ret = -1L;
1745 
1746 	if ((ret || test_thread_flag(TIF_SYSCALL_TRACE)) &&
1747 	    tracehook_report_syscall_entry(regs))
1748 		ret = -1L;
1749 
1750 	if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
1751 		trace_sys_enter(regs, regs->orig_ax);
1752 
1753 	if (unlikely(current->audit_context)) {
1754 		if (IS_IA32)
1755 			audit_syscall_entry(AUDIT_ARCH_I386,
1756 					    regs->orig_ax,
1757 					    regs->bx, regs->cx,
1758 					    regs->dx, regs->si);
1759 #ifdef CONFIG_X86_64
1760 		else
1761 			audit_syscall_entry(AUDIT_ARCH_X86_64,
1762 					    regs->orig_ax,
1763 					    regs->di, regs->si,
1764 					    regs->dx, regs->r10);
1765 #endif
1766 	}
1767 
1768 	return ret ?: regs->orig_ax;
1769 }
1770 
1771 asmregparm void syscall_trace_leave(struct pt_regs *regs)
1772 {
1773 	bool step;
1774 
1775 	if (unlikely(current->audit_context))
1776 		audit_syscall_exit(AUDITSC_RESULT(regs->ax), regs->ax);
1777 
1778 	if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
1779 		trace_sys_exit(regs, regs->ax);
1780 
1781 	/*
1782 	 * If TIF_SYSCALL_EMU is set, we only get here because of
1783 	 * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
1784 	 * We already reported this syscall instruction in
1785 	 * syscall_trace_enter().
1786 	 */
1787 	step = unlikely(test_thread_flag(TIF_SINGLESTEP)) &&
1788 			!test_thread_flag(TIF_SYSCALL_EMU);
1789 	if (step || test_thread_flag(TIF_SYSCALL_TRACE))
1790 		tracehook_report_syscall_exit(regs, step);
1791 }
1792