xref: /linux/arch/x86/kernel/ptrace.c (revision 27258e448eb301cf89e351df87aa8cb916653bf2)
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 
26 #include <asm/uaccess.h>
27 #include <asm/pgtable.h>
28 #include <asm/system.h>
29 #include <asm/processor.h>
30 #include <asm/i387.h>
31 #include <asm/debugreg.h>
32 #include <asm/ldt.h>
33 #include <asm/desc.h>
34 #include <asm/prctl.h>
35 #include <asm/proto.h>
36 #include <asm/ds.h>
37 
38 #include "tls.h"
39 
40 #define CREATE_TRACE_POINTS
41 #include <trace/events/syscalls.h>
42 
43 enum x86_regset {
44 	REGSET_GENERAL,
45 	REGSET_FP,
46 	REGSET_XFP,
47 	REGSET_IOPERM64 = REGSET_XFP,
48 	REGSET_TLS,
49 	REGSET_IOPERM32,
50 };
51 
52 /*
53  * does not yet catch signals sent when the child dies.
54  * in exit.c or in signal.c.
55  */
56 
57 /*
58  * Determines which flags the user has access to [1 = access, 0 = no access].
59  */
60 #define FLAG_MASK_32		((unsigned long)			\
61 				 (X86_EFLAGS_CF | X86_EFLAGS_PF |	\
62 				  X86_EFLAGS_AF | X86_EFLAGS_ZF |	\
63 				  X86_EFLAGS_SF | X86_EFLAGS_TF |	\
64 				  X86_EFLAGS_DF | X86_EFLAGS_OF |	\
65 				  X86_EFLAGS_RF | X86_EFLAGS_AC))
66 
67 /*
68  * Determines whether a value may be installed in a segment register.
69  */
70 static inline bool invalid_selector(u16 value)
71 {
72 	return unlikely(value != 0 && (value & SEGMENT_RPL_MASK) != USER_RPL);
73 }
74 
75 #ifdef CONFIG_X86_32
76 
77 #define FLAG_MASK		FLAG_MASK_32
78 
79 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long regno)
80 {
81 	BUILD_BUG_ON(offsetof(struct pt_regs, bx) != 0);
82 	return &regs->bx + (regno >> 2);
83 }
84 
85 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
86 {
87 	/*
88 	 * Returning the value truncates it to 16 bits.
89 	 */
90 	unsigned int retval;
91 	if (offset != offsetof(struct user_regs_struct, gs))
92 		retval = *pt_regs_access(task_pt_regs(task), offset);
93 	else {
94 		if (task == current)
95 			retval = get_user_gs(task_pt_regs(task));
96 		else
97 			retval = task_user_gs(task);
98 	}
99 	return retval;
100 }
101 
102 static int set_segment_reg(struct task_struct *task,
103 			   unsigned long offset, u16 value)
104 {
105 	/*
106 	 * The value argument was already truncated to 16 bits.
107 	 */
108 	if (invalid_selector(value))
109 		return -EIO;
110 
111 	/*
112 	 * For %cs and %ss we cannot permit a null selector.
113 	 * We can permit a bogus selector as long as it has USER_RPL.
114 	 * Null selectors are fine for other segment registers, but
115 	 * we will never get back to user mode with invalid %cs or %ss
116 	 * and will take the trap in iret instead.  Much code relies
117 	 * on user_mode() to distinguish a user trap frame (which can
118 	 * safely use invalid selectors) from a kernel trap frame.
119 	 */
120 	switch (offset) {
121 	case offsetof(struct user_regs_struct, cs):
122 	case offsetof(struct user_regs_struct, ss):
123 		if (unlikely(value == 0))
124 			return -EIO;
125 
126 	default:
127 		*pt_regs_access(task_pt_regs(task), offset) = value;
128 		break;
129 
130 	case offsetof(struct user_regs_struct, gs):
131 		if (task == current)
132 			set_user_gs(task_pt_regs(task), value);
133 		else
134 			task_user_gs(task) = value;
135 	}
136 
137 	return 0;
138 }
139 
140 static unsigned long debugreg_addr_limit(struct task_struct *task)
141 {
142 	return TASK_SIZE - 3;
143 }
144 
145 #else  /* CONFIG_X86_64 */
146 
147 #define FLAG_MASK		(FLAG_MASK_32 | X86_EFLAGS_NT)
148 
149 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long offset)
150 {
151 	BUILD_BUG_ON(offsetof(struct pt_regs, r15) != 0);
152 	return &regs->r15 + (offset / sizeof(regs->r15));
153 }
154 
155 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
156 {
157 	/*
158 	 * Returning the value truncates it to 16 bits.
159 	 */
160 	unsigned int seg;
161 
162 	switch (offset) {
163 	case offsetof(struct user_regs_struct, fs):
164 		if (task == current) {
165 			/* Older gas can't assemble movq %?s,%r?? */
166 			asm("movl %%fs,%0" : "=r" (seg));
167 			return seg;
168 		}
169 		return task->thread.fsindex;
170 	case offsetof(struct user_regs_struct, gs):
171 		if (task == current) {
172 			asm("movl %%gs,%0" : "=r" (seg));
173 			return seg;
174 		}
175 		return task->thread.gsindex;
176 	case offsetof(struct user_regs_struct, ds):
177 		if (task == current) {
178 			asm("movl %%ds,%0" : "=r" (seg));
179 			return seg;
180 		}
181 		return task->thread.ds;
182 	case offsetof(struct user_regs_struct, es):
183 		if (task == current) {
184 			asm("movl %%es,%0" : "=r" (seg));
185 			return seg;
186 		}
187 		return task->thread.es;
188 
189 	case offsetof(struct user_regs_struct, cs):
190 	case offsetof(struct user_regs_struct, ss):
191 		break;
192 	}
193 	return *pt_regs_access(task_pt_regs(task), offset);
194 }
195 
196 static int set_segment_reg(struct task_struct *task,
197 			   unsigned long offset, u16 value)
198 {
199 	/*
200 	 * The value argument was already truncated to 16 bits.
201 	 */
202 	if (invalid_selector(value))
203 		return -EIO;
204 
205 	switch (offset) {
206 	case offsetof(struct user_regs_struct,fs):
207 		/*
208 		 * If this is setting fs as for normal 64-bit use but
209 		 * setting fs_base has implicitly changed it, leave it.
210 		 */
211 		if ((value == FS_TLS_SEL && task->thread.fsindex == 0 &&
212 		     task->thread.fs != 0) ||
213 		    (value == 0 && task->thread.fsindex == FS_TLS_SEL &&
214 		     task->thread.fs == 0))
215 			break;
216 		task->thread.fsindex = value;
217 		if (task == current)
218 			loadsegment(fs, task->thread.fsindex);
219 		break;
220 	case offsetof(struct user_regs_struct,gs):
221 		/*
222 		 * If this is setting gs as for normal 64-bit use but
223 		 * setting gs_base has implicitly changed it, leave it.
224 		 */
225 		if ((value == GS_TLS_SEL && task->thread.gsindex == 0 &&
226 		     task->thread.gs != 0) ||
227 		    (value == 0 && task->thread.gsindex == GS_TLS_SEL &&
228 		     task->thread.gs == 0))
229 			break;
230 		task->thread.gsindex = value;
231 		if (task == current)
232 			load_gs_index(task->thread.gsindex);
233 		break;
234 	case offsetof(struct user_regs_struct,ds):
235 		task->thread.ds = value;
236 		if (task == current)
237 			loadsegment(ds, task->thread.ds);
238 		break;
239 	case offsetof(struct user_regs_struct,es):
240 		task->thread.es = value;
241 		if (task == current)
242 			loadsegment(es, task->thread.es);
243 		break;
244 
245 		/*
246 		 * Can't actually change these in 64-bit mode.
247 		 */
248 	case offsetof(struct user_regs_struct,cs):
249 		if (unlikely(value == 0))
250 			return -EIO;
251 #ifdef CONFIG_IA32_EMULATION
252 		if (test_tsk_thread_flag(task, TIF_IA32))
253 			task_pt_regs(task)->cs = value;
254 #endif
255 		break;
256 	case offsetof(struct user_regs_struct,ss):
257 		if (unlikely(value == 0))
258 			return -EIO;
259 #ifdef CONFIG_IA32_EMULATION
260 		if (test_tsk_thread_flag(task, TIF_IA32))
261 			task_pt_regs(task)->ss = value;
262 #endif
263 		break;
264 	}
265 
266 	return 0;
267 }
268 
269 static unsigned long debugreg_addr_limit(struct task_struct *task)
270 {
271 #ifdef CONFIG_IA32_EMULATION
272 	if (test_tsk_thread_flag(task, TIF_IA32))
273 		return IA32_PAGE_OFFSET - 3;
274 #endif
275 	return TASK_SIZE_MAX - 7;
276 }
277 
278 #endif	/* CONFIG_X86_32 */
279 
280 static unsigned long get_flags(struct task_struct *task)
281 {
282 	unsigned long retval = task_pt_regs(task)->flags;
283 
284 	/*
285 	 * If the debugger set TF, hide it from the readout.
286 	 */
287 	if (test_tsk_thread_flag(task, TIF_FORCED_TF))
288 		retval &= ~X86_EFLAGS_TF;
289 
290 	return retval;
291 }
292 
293 static int set_flags(struct task_struct *task, unsigned long value)
294 {
295 	struct pt_regs *regs = task_pt_regs(task);
296 
297 	/*
298 	 * If the user value contains TF, mark that
299 	 * it was not "us" (the debugger) that set it.
300 	 * If not, make sure it stays set if we had.
301 	 */
302 	if (value & X86_EFLAGS_TF)
303 		clear_tsk_thread_flag(task, TIF_FORCED_TF);
304 	else if (test_tsk_thread_flag(task, TIF_FORCED_TF))
305 		value |= X86_EFLAGS_TF;
306 
307 	regs->flags = (regs->flags & ~FLAG_MASK) | (value & FLAG_MASK);
308 
309 	return 0;
310 }
311 
312 static int putreg(struct task_struct *child,
313 		  unsigned long offset, unsigned long value)
314 {
315 	switch (offset) {
316 	case offsetof(struct user_regs_struct, cs):
317 	case offsetof(struct user_regs_struct, ds):
318 	case offsetof(struct user_regs_struct, es):
319 	case offsetof(struct user_regs_struct, fs):
320 	case offsetof(struct user_regs_struct, gs):
321 	case offsetof(struct user_regs_struct, ss):
322 		return set_segment_reg(child, offset, value);
323 
324 	case offsetof(struct user_regs_struct, flags):
325 		return set_flags(child, value);
326 
327 #ifdef CONFIG_X86_64
328 	/*
329 	 * Orig_ax is really just a flag with small positive and
330 	 * negative values, so make sure to always sign-extend it
331 	 * from 32 bits so that it works correctly regardless of
332 	 * whether we come from a 32-bit environment or not.
333 	 */
334 	case offsetof(struct user_regs_struct, orig_ax):
335 		value = (long) (s32) value;
336 		break;
337 
338 	case offsetof(struct user_regs_struct,fs_base):
339 		if (value >= TASK_SIZE_OF(child))
340 			return -EIO;
341 		/*
342 		 * When changing the segment base, use do_arch_prctl
343 		 * to set either thread.fs or thread.fsindex and the
344 		 * corresponding GDT slot.
345 		 */
346 		if (child->thread.fs != value)
347 			return do_arch_prctl(child, ARCH_SET_FS, value);
348 		return 0;
349 	case offsetof(struct user_regs_struct,gs_base):
350 		/*
351 		 * Exactly the same here as the %fs handling above.
352 		 */
353 		if (value >= TASK_SIZE_OF(child))
354 			return -EIO;
355 		if (child->thread.gs != value)
356 			return do_arch_prctl(child, ARCH_SET_GS, value);
357 		return 0;
358 #endif
359 	}
360 
361 	*pt_regs_access(task_pt_regs(child), offset) = value;
362 	return 0;
363 }
364 
365 static unsigned long getreg(struct task_struct *task, unsigned long offset)
366 {
367 	switch (offset) {
368 	case offsetof(struct user_regs_struct, cs):
369 	case offsetof(struct user_regs_struct, ds):
370 	case offsetof(struct user_regs_struct, es):
371 	case offsetof(struct user_regs_struct, fs):
372 	case offsetof(struct user_regs_struct, gs):
373 	case offsetof(struct user_regs_struct, ss):
374 		return get_segment_reg(task, offset);
375 
376 	case offsetof(struct user_regs_struct, flags):
377 		return get_flags(task);
378 
379 #ifdef CONFIG_X86_64
380 	case offsetof(struct user_regs_struct, fs_base): {
381 		/*
382 		 * do_arch_prctl may have used a GDT slot instead of
383 		 * the MSR.  To userland, it appears the same either
384 		 * way, except the %fs segment selector might not be 0.
385 		 */
386 		unsigned int seg = task->thread.fsindex;
387 		if (task->thread.fs != 0)
388 			return task->thread.fs;
389 		if (task == current)
390 			asm("movl %%fs,%0" : "=r" (seg));
391 		if (seg != FS_TLS_SEL)
392 			return 0;
393 		return get_desc_base(&task->thread.tls_array[FS_TLS]);
394 	}
395 	case offsetof(struct user_regs_struct, gs_base): {
396 		/*
397 		 * Exactly the same here as the %fs handling above.
398 		 */
399 		unsigned int seg = task->thread.gsindex;
400 		if (task->thread.gs != 0)
401 			return task->thread.gs;
402 		if (task == current)
403 			asm("movl %%gs,%0" : "=r" (seg));
404 		if (seg != GS_TLS_SEL)
405 			return 0;
406 		return get_desc_base(&task->thread.tls_array[GS_TLS]);
407 	}
408 #endif
409 	}
410 
411 	return *pt_regs_access(task_pt_regs(task), offset);
412 }
413 
414 static int genregs_get(struct task_struct *target,
415 		       const struct user_regset *regset,
416 		       unsigned int pos, unsigned int count,
417 		       void *kbuf, void __user *ubuf)
418 {
419 	if (kbuf) {
420 		unsigned long *k = kbuf;
421 		while (count > 0) {
422 			*k++ = getreg(target, pos);
423 			count -= sizeof(*k);
424 			pos += sizeof(*k);
425 		}
426 	} else {
427 		unsigned long __user *u = ubuf;
428 		while (count > 0) {
429 			if (__put_user(getreg(target, pos), u++))
430 				return -EFAULT;
431 			count -= sizeof(*u);
432 			pos += sizeof(*u);
433 		}
434 	}
435 
436 	return 0;
437 }
438 
439 static int genregs_set(struct task_struct *target,
440 		       const struct user_regset *regset,
441 		       unsigned int pos, unsigned int count,
442 		       const void *kbuf, const void __user *ubuf)
443 {
444 	int ret = 0;
445 	if (kbuf) {
446 		const unsigned long *k = kbuf;
447 		while (count > 0 && !ret) {
448 			ret = putreg(target, pos, *k++);
449 			count -= sizeof(*k);
450 			pos += sizeof(*k);
451 		}
452 	} else {
453 		const unsigned long  __user *u = ubuf;
454 		while (count > 0 && !ret) {
455 			unsigned long word;
456 			ret = __get_user(word, u++);
457 			if (ret)
458 				break;
459 			ret = putreg(target, pos, word);
460 			count -= sizeof(*u);
461 			pos += sizeof(*u);
462 		}
463 	}
464 	return ret;
465 }
466 
467 /*
468  * This function is trivial and will be inlined by the compiler.
469  * Having it separates the implementation details of debug
470  * registers from the interface details of ptrace.
471  */
472 static unsigned long ptrace_get_debugreg(struct task_struct *child, int n)
473 {
474 	switch (n) {
475 	case 0:		return child->thread.debugreg0;
476 	case 1:		return child->thread.debugreg1;
477 	case 2:		return child->thread.debugreg2;
478 	case 3:		return child->thread.debugreg3;
479 	case 6:		return child->thread.debugreg6;
480 	case 7:		return child->thread.debugreg7;
481 	}
482 	return 0;
483 }
484 
485 static int ptrace_set_debugreg(struct task_struct *child,
486 			       int n, unsigned long data)
487 {
488 	int i;
489 
490 	if (unlikely(n == 4 || n == 5))
491 		return -EIO;
492 
493 	if (n < 4 && unlikely(data >= debugreg_addr_limit(child)))
494 		return -EIO;
495 
496 	switch (n) {
497 	case 0:		child->thread.debugreg0 = data; break;
498 	case 1:		child->thread.debugreg1 = data; break;
499 	case 2:		child->thread.debugreg2 = data; break;
500 	case 3:		child->thread.debugreg3 = data; break;
501 
502 	case 6:
503 		if ((data & ~0xffffffffUL) != 0)
504 			return -EIO;
505 		child->thread.debugreg6 = data;
506 		break;
507 
508 	case 7:
509 		/*
510 		 * Sanity-check data. Take one half-byte at once with
511 		 * check = (val >> (16 + 4*i)) & 0xf. It contains the
512 		 * R/Wi and LENi bits; bits 0 and 1 are R/Wi, and bits
513 		 * 2 and 3 are LENi. Given a list of invalid values,
514 		 * we do mask |= 1 << invalid_value, so that
515 		 * (mask >> check) & 1 is a correct test for invalid
516 		 * values.
517 		 *
518 		 * R/Wi contains the type of the breakpoint /
519 		 * watchpoint, LENi contains the length of the watched
520 		 * data in the watchpoint case.
521 		 *
522 		 * The invalid values are:
523 		 * - LENi == 0x10 (undefined), so mask |= 0x0f00.	[32-bit]
524 		 * - R/Wi == 0x10 (break on I/O reads or writes), so
525 		 *   mask |= 0x4444.
526 		 * - R/Wi == 0x00 && LENi != 0x00, so we have mask |=
527 		 *   0x1110.
528 		 *
529 		 * Finally, mask = 0x0f00 | 0x4444 | 0x1110 == 0x5f54.
530 		 *
531 		 * See the Intel Manual "System Programming Guide",
532 		 * 15.2.4
533 		 *
534 		 * Note that LENi == 0x10 is defined on x86_64 in long
535 		 * mode (i.e. even for 32-bit userspace software, but
536 		 * 64-bit kernel), so the x86_64 mask value is 0x5454.
537 		 * See the AMD manual no. 24593 (AMD64 System Programming)
538 		 */
539 #ifdef CONFIG_X86_32
540 #define	DR7_MASK	0x5f54
541 #else
542 #define	DR7_MASK	0x5554
543 #endif
544 		data &= ~DR_CONTROL_RESERVED;
545 		for (i = 0; i < 4; i++)
546 			if ((DR7_MASK >> ((data >> (16 + 4*i)) & 0xf)) & 1)
547 				return -EIO;
548 		child->thread.debugreg7 = data;
549 		if (data)
550 			set_tsk_thread_flag(child, TIF_DEBUG);
551 		else
552 			clear_tsk_thread_flag(child, TIF_DEBUG);
553 		break;
554 	}
555 
556 	return 0;
557 }
558 
559 /*
560  * These access the current or another (stopped) task's io permission
561  * bitmap for debugging or core dump.
562  */
563 static int ioperm_active(struct task_struct *target,
564 			 const struct user_regset *regset)
565 {
566 	return target->thread.io_bitmap_max / regset->size;
567 }
568 
569 static int ioperm_get(struct task_struct *target,
570 		      const struct user_regset *regset,
571 		      unsigned int pos, unsigned int count,
572 		      void *kbuf, void __user *ubuf)
573 {
574 	if (!target->thread.io_bitmap_ptr)
575 		return -ENXIO;
576 
577 	return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
578 				   target->thread.io_bitmap_ptr,
579 				   0, IO_BITMAP_BYTES);
580 }
581 
582 #ifdef CONFIG_X86_PTRACE_BTS
583 /*
584  * A branch trace store context.
585  *
586  * Contexts may only be installed by ptrace_bts_config() and only for
587  * ptraced tasks.
588  *
589  * Contexts are destroyed when the tracee is detached from the tracer.
590  * The actual destruction work requires interrupts enabled, so the
591  * work is deferred and will be scheduled during __ptrace_unlink().
592  *
593  * Contexts hold an additional task_struct reference on the traced
594  * task, as well as a reference on the tracer's mm.
595  *
596  * Ptrace already holds a task_struct for the duration of ptrace operations,
597  * but since destruction is deferred, it may be executed after both
598  * tracer and tracee exited.
599  */
600 struct bts_context {
601 	/* The branch trace handle. */
602 	struct bts_tracer	*tracer;
603 
604 	/* The buffer used to store the branch trace and its size. */
605 	void			*buffer;
606 	unsigned int		size;
607 
608 	/* The mm that paid for the above buffer. */
609 	struct mm_struct	*mm;
610 
611 	/* The task this context belongs to. */
612 	struct task_struct	*task;
613 
614 	/* The signal to send on a bts buffer overflow. */
615 	unsigned int		bts_ovfl_signal;
616 
617 	/* The work struct to destroy a context. */
618 	struct work_struct	work;
619 };
620 
621 static int alloc_bts_buffer(struct bts_context *context, unsigned int size)
622 {
623 	void *buffer = NULL;
624 	int err = -ENOMEM;
625 
626 	err = account_locked_memory(current->mm, current->signal->rlim, size);
627 	if (err < 0)
628 		return err;
629 
630 	buffer = kzalloc(size, GFP_KERNEL);
631 	if (!buffer)
632 		goto out_refund;
633 
634 	context->buffer = buffer;
635 	context->size = size;
636 	context->mm = get_task_mm(current);
637 
638 	return 0;
639 
640  out_refund:
641 	refund_locked_memory(current->mm, size);
642 	return err;
643 }
644 
645 static inline void free_bts_buffer(struct bts_context *context)
646 {
647 	if (!context->buffer)
648 		return;
649 
650 	kfree(context->buffer);
651 	context->buffer = NULL;
652 
653 	refund_locked_memory(context->mm, context->size);
654 	context->size = 0;
655 
656 	mmput(context->mm);
657 	context->mm = NULL;
658 }
659 
660 static void free_bts_context_work(struct work_struct *w)
661 {
662 	struct bts_context *context;
663 
664 	context = container_of(w, struct bts_context, work);
665 
666 	ds_release_bts(context->tracer);
667 	put_task_struct(context->task);
668 	free_bts_buffer(context);
669 	kfree(context);
670 }
671 
672 static inline void free_bts_context(struct bts_context *context)
673 {
674 	INIT_WORK(&context->work, free_bts_context_work);
675 	schedule_work(&context->work);
676 }
677 
678 static inline struct bts_context *alloc_bts_context(struct task_struct *task)
679 {
680 	struct bts_context *context = kzalloc(sizeof(*context), GFP_KERNEL);
681 	if (context) {
682 		context->task = task;
683 		task->bts = context;
684 
685 		get_task_struct(task);
686 	}
687 
688 	return context;
689 }
690 
691 static int ptrace_bts_read_record(struct task_struct *child, size_t index,
692 				  struct bts_struct __user *out)
693 {
694 	struct bts_context *context;
695 	const struct bts_trace *trace;
696 	struct bts_struct bts;
697 	const unsigned char *at;
698 	int error;
699 
700 	context = child->bts;
701 	if (!context)
702 		return -ESRCH;
703 
704 	trace = ds_read_bts(context->tracer);
705 	if (!trace)
706 		return -ESRCH;
707 
708 	at = trace->ds.top - ((index + 1) * trace->ds.size);
709 	if ((void *)at < trace->ds.begin)
710 		at += (trace->ds.n * trace->ds.size);
711 
712 	if (!trace->read)
713 		return -EOPNOTSUPP;
714 
715 	error = trace->read(context->tracer, at, &bts);
716 	if (error < 0)
717 		return error;
718 
719 	if (copy_to_user(out, &bts, sizeof(bts)))
720 		return -EFAULT;
721 
722 	return sizeof(bts);
723 }
724 
725 static int ptrace_bts_drain(struct task_struct *child,
726 			    long size,
727 			    struct bts_struct __user *out)
728 {
729 	struct bts_context *context;
730 	const struct bts_trace *trace;
731 	const unsigned char *at;
732 	int error, drained = 0;
733 
734 	context = child->bts;
735 	if (!context)
736 		return -ESRCH;
737 
738 	trace = ds_read_bts(context->tracer);
739 	if (!trace)
740 		return -ESRCH;
741 
742 	if (!trace->read)
743 		return -EOPNOTSUPP;
744 
745 	if (size < (trace->ds.top - trace->ds.begin))
746 		return -EIO;
747 
748 	for (at = trace->ds.begin; (void *)at < trace->ds.top;
749 	     out++, drained++, at += trace->ds.size) {
750 		struct bts_struct bts;
751 
752 		error = trace->read(context->tracer, at, &bts);
753 		if (error < 0)
754 			return error;
755 
756 		if (copy_to_user(out, &bts, sizeof(bts)))
757 			return -EFAULT;
758 	}
759 
760 	memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size);
761 
762 	error = ds_reset_bts(context->tracer);
763 	if (error < 0)
764 		return error;
765 
766 	return drained;
767 }
768 
769 static int ptrace_bts_config(struct task_struct *child,
770 			     long cfg_size,
771 			     const struct ptrace_bts_config __user *ucfg)
772 {
773 	struct bts_context *context;
774 	struct ptrace_bts_config cfg;
775 	unsigned int flags = 0;
776 
777 	if (cfg_size < sizeof(cfg))
778 		return -EIO;
779 
780 	if (copy_from_user(&cfg, ucfg, sizeof(cfg)))
781 		return -EFAULT;
782 
783 	context = child->bts;
784 	if (!context)
785 		context = alloc_bts_context(child);
786 	if (!context)
787 		return -ENOMEM;
788 
789 	if (cfg.flags & PTRACE_BTS_O_SIGNAL) {
790 		if (!cfg.signal)
791 			return -EINVAL;
792 
793 		return -EOPNOTSUPP;
794 		context->bts_ovfl_signal = cfg.signal;
795 	}
796 
797 	ds_release_bts(context->tracer);
798 	context->tracer = NULL;
799 
800 	if ((cfg.flags & PTRACE_BTS_O_ALLOC) && (cfg.size != context->size)) {
801 		int err;
802 
803 		free_bts_buffer(context);
804 		if (!cfg.size)
805 			return 0;
806 
807 		err = alloc_bts_buffer(context, cfg.size);
808 		if (err < 0)
809 			return err;
810 	}
811 
812 	if (cfg.flags & PTRACE_BTS_O_TRACE)
813 		flags |= BTS_USER;
814 
815 	if (cfg.flags & PTRACE_BTS_O_SCHED)
816 		flags |= BTS_TIMESTAMPS;
817 
818 	context->tracer =
819 		ds_request_bts_task(child, context->buffer, context->size,
820 				    NULL, (size_t)-1, flags);
821 	if (unlikely(IS_ERR(context->tracer))) {
822 		int error = PTR_ERR(context->tracer);
823 
824 		free_bts_buffer(context);
825 		context->tracer = NULL;
826 		return error;
827 	}
828 
829 	return sizeof(cfg);
830 }
831 
832 static int ptrace_bts_status(struct task_struct *child,
833 			     long cfg_size,
834 			     struct ptrace_bts_config __user *ucfg)
835 {
836 	struct bts_context *context;
837 	const struct bts_trace *trace;
838 	struct ptrace_bts_config cfg;
839 
840 	context = child->bts;
841 	if (!context)
842 		return -ESRCH;
843 
844 	if (cfg_size < sizeof(cfg))
845 		return -EIO;
846 
847 	trace = ds_read_bts(context->tracer);
848 	if (!trace)
849 		return -ESRCH;
850 
851 	memset(&cfg, 0, sizeof(cfg));
852 	cfg.size	= trace->ds.end - trace->ds.begin;
853 	cfg.signal	= context->bts_ovfl_signal;
854 	cfg.bts_size	= sizeof(struct bts_struct);
855 
856 	if (cfg.signal)
857 		cfg.flags |= PTRACE_BTS_O_SIGNAL;
858 
859 	if (trace->ds.flags & BTS_USER)
860 		cfg.flags |= PTRACE_BTS_O_TRACE;
861 
862 	if (trace->ds.flags & BTS_TIMESTAMPS)
863 		cfg.flags |= PTRACE_BTS_O_SCHED;
864 
865 	if (copy_to_user(ucfg, &cfg, sizeof(cfg)))
866 		return -EFAULT;
867 
868 	return sizeof(cfg);
869 }
870 
871 static int ptrace_bts_clear(struct task_struct *child)
872 {
873 	struct bts_context *context;
874 	const struct bts_trace *trace;
875 
876 	context = child->bts;
877 	if (!context)
878 		return -ESRCH;
879 
880 	trace = ds_read_bts(context->tracer);
881 	if (!trace)
882 		return -ESRCH;
883 
884 	memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size);
885 
886 	return ds_reset_bts(context->tracer);
887 }
888 
889 static int ptrace_bts_size(struct task_struct *child)
890 {
891 	struct bts_context *context;
892 	const struct bts_trace *trace;
893 
894 	context = child->bts;
895 	if (!context)
896 		return -ESRCH;
897 
898 	trace = ds_read_bts(context->tracer);
899 	if (!trace)
900 		return -ESRCH;
901 
902 	return (trace->ds.top - trace->ds.begin) / trace->ds.size;
903 }
904 
905 /*
906  * Called from __ptrace_unlink() after the child has been moved back
907  * to its original parent.
908  */
909 void ptrace_bts_untrace(struct task_struct *child)
910 {
911 	if (unlikely(child->bts)) {
912 		free_bts_context(child->bts);
913 		child->bts = NULL;
914 	}
915 }
916 #endif /* CONFIG_X86_PTRACE_BTS */
917 
918 /*
919  * Called by kernel/ptrace.c when detaching..
920  *
921  * Make sure the single step bit is not set.
922  */
923 void ptrace_disable(struct task_struct *child)
924 {
925 	user_disable_single_step(child);
926 #ifdef TIF_SYSCALL_EMU
927 	clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
928 #endif
929 }
930 
931 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
932 static const struct user_regset_view user_x86_32_view; /* Initialized below. */
933 #endif
934 
935 long arch_ptrace(struct task_struct *child, long request, long addr, long data)
936 {
937 	int ret;
938 	unsigned long __user *datap = (unsigned long __user *)data;
939 
940 	switch (request) {
941 	/* read the word at location addr in the USER area. */
942 	case PTRACE_PEEKUSR: {
943 		unsigned long tmp;
944 
945 		ret = -EIO;
946 		if ((addr & (sizeof(data) - 1)) || addr < 0 ||
947 		    addr >= sizeof(struct user))
948 			break;
949 
950 		tmp = 0;  /* Default return condition */
951 		if (addr < sizeof(struct user_regs_struct))
952 			tmp = getreg(child, addr);
953 		else if (addr >= offsetof(struct user, u_debugreg[0]) &&
954 			 addr <= offsetof(struct user, u_debugreg[7])) {
955 			addr -= offsetof(struct user, u_debugreg[0]);
956 			tmp = ptrace_get_debugreg(child, addr / sizeof(data));
957 		}
958 		ret = put_user(tmp, datap);
959 		break;
960 	}
961 
962 	case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
963 		ret = -EIO;
964 		if ((addr & (sizeof(data) - 1)) || addr < 0 ||
965 		    addr >= sizeof(struct user))
966 			break;
967 
968 		if (addr < sizeof(struct user_regs_struct))
969 			ret = putreg(child, addr, data);
970 		else if (addr >= offsetof(struct user, u_debugreg[0]) &&
971 			 addr <= offsetof(struct user, u_debugreg[7])) {
972 			addr -= offsetof(struct user, u_debugreg[0]);
973 			ret = ptrace_set_debugreg(child,
974 						  addr / sizeof(data), data);
975 		}
976 		break;
977 
978 	case PTRACE_GETREGS:	/* Get all gp regs from the child. */
979 		return copy_regset_to_user(child,
980 					   task_user_regset_view(current),
981 					   REGSET_GENERAL,
982 					   0, sizeof(struct user_regs_struct),
983 					   datap);
984 
985 	case PTRACE_SETREGS:	/* Set all gp regs in the child. */
986 		return copy_regset_from_user(child,
987 					     task_user_regset_view(current),
988 					     REGSET_GENERAL,
989 					     0, sizeof(struct user_regs_struct),
990 					     datap);
991 
992 	case PTRACE_GETFPREGS:	/* Get the child FPU state. */
993 		return copy_regset_to_user(child,
994 					   task_user_regset_view(current),
995 					   REGSET_FP,
996 					   0, sizeof(struct user_i387_struct),
997 					   datap);
998 
999 	case PTRACE_SETFPREGS:	/* Set the child FPU state. */
1000 		return copy_regset_from_user(child,
1001 					     task_user_regset_view(current),
1002 					     REGSET_FP,
1003 					     0, sizeof(struct user_i387_struct),
1004 					     datap);
1005 
1006 #ifdef CONFIG_X86_32
1007 	case PTRACE_GETFPXREGS:	/* Get the child extended FPU state. */
1008 		return copy_regset_to_user(child, &user_x86_32_view,
1009 					   REGSET_XFP,
1010 					   0, sizeof(struct user_fxsr_struct),
1011 					   datap) ? -EIO : 0;
1012 
1013 	case PTRACE_SETFPXREGS:	/* Set the child extended FPU state. */
1014 		return copy_regset_from_user(child, &user_x86_32_view,
1015 					     REGSET_XFP,
1016 					     0, sizeof(struct user_fxsr_struct),
1017 					     datap) ? -EIO : 0;
1018 #endif
1019 
1020 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1021 	case PTRACE_GET_THREAD_AREA:
1022 		if (addr < 0)
1023 			return -EIO;
1024 		ret = do_get_thread_area(child, addr,
1025 					 (struct user_desc __user *) data);
1026 		break;
1027 
1028 	case PTRACE_SET_THREAD_AREA:
1029 		if (addr < 0)
1030 			return -EIO;
1031 		ret = do_set_thread_area(child, addr,
1032 					 (struct user_desc __user *) data, 0);
1033 		break;
1034 #endif
1035 
1036 #ifdef CONFIG_X86_64
1037 		/* normal 64bit interface to access TLS data.
1038 		   Works just like arch_prctl, except that the arguments
1039 		   are reversed. */
1040 	case PTRACE_ARCH_PRCTL:
1041 		ret = do_arch_prctl(child, data, addr);
1042 		break;
1043 #endif
1044 
1045 	/*
1046 	 * These bits need more cooking - not enabled yet:
1047 	 */
1048 #ifdef CONFIG_X86_PTRACE_BTS
1049 	case PTRACE_BTS_CONFIG:
1050 		ret = ptrace_bts_config
1051 			(child, data, (struct ptrace_bts_config __user *)addr);
1052 		break;
1053 
1054 	case PTRACE_BTS_STATUS:
1055 		ret = ptrace_bts_status
1056 			(child, data, (struct ptrace_bts_config __user *)addr);
1057 		break;
1058 
1059 	case PTRACE_BTS_SIZE:
1060 		ret = ptrace_bts_size(child);
1061 		break;
1062 
1063 	case PTRACE_BTS_GET:
1064 		ret = ptrace_bts_read_record
1065 			(child, data, (struct bts_struct __user *) addr);
1066 		break;
1067 
1068 	case PTRACE_BTS_CLEAR:
1069 		ret = ptrace_bts_clear(child);
1070 		break;
1071 
1072 	case PTRACE_BTS_DRAIN:
1073 		ret = ptrace_bts_drain
1074 			(child, data, (struct bts_struct __user *) addr);
1075 		break;
1076 #endif /* CONFIG_X86_PTRACE_BTS */
1077 
1078 	default:
1079 		ret = ptrace_request(child, request, addr, data);
1080 		break;
1081 	}
1082 
1083 	return ret;
1084 }
1085 
1086 #ifdef CONFIG_IA32_EMULATION
1087 
1088 #include <linux/compat.h>
1089 #include <linux/syscalls.h>
1090 #include <asm/ia32.h>
1091 #include <asm/user32.h>
1092 
1093 #define R32(l,q)							\
1094 	case offsetof(struct user32, regs.l):				\
1095 		regs->q = value; break
1096 
1097 #define SEG32(rs)							\
1098 	case offsetof(struct user32, regs.rs):				\
1099 		return set_segment_reg(child,				\
1100 				       offsetof(struct user_regs_struct, rs), \
1101 				       value);				\
1102 		break
1103 
1104 static int putreg32(struct task_struct *child, unsigned regno, u32 value)
1105 {
1106 	struct pt_regs *regs = task_pt_regs(child);
1107 
1108 	switch (regno) {
1109 
1110 	SEG32(cs);
1111 	SEG32(ds);
1112 	SEG32(es);
1113 	SEG32(fs);
1114 	SEG32(gs);
1115 	SEG32(ss);
1116 
1117 	R32(ebx, bx);
1118 	R32(ecx, cx);
1119 	R32(edx, dx);
1120 	R32(edi, di);
1121 	R32(esi, si);
1122 	R32(ebp, bp);
1123 	R32(eax, ax);
1124 	R32(eip, ip);
1125 	R32(esp, sp);
1126 
1127 	case offsetof(struct user32, regs.orig_eax):
1128 		/*
1129 		 * Sign-extend the value so that orig_eax = -1
1130 		 * causes (long)orig_ax < 0 tests to fire correctly.
1131 		 */
1132 		regs->orig_ax = (long) (s32) value;
1133 		break;
1134 
1135 	case offsetof(struct user32, regs.eflags):
1136 		return set_flags(child, value);
1137 
1138 	case offsetof(struct user32, u_debugreg[0]) ...
1139 		offsetof(struct user32, u_debugreg[7]):
1140 		regno -= offsetof(struct user32, u_debugreg[0]);
1141 		return ptrace_set_debugreg(child, regno / 4, value);
1142 
1143 	default:
1144 		if (regno > sizeof(struct user32) || (regno & 3))
1145 			return -EIO;
1146 
1147 		/*
1148 		 * Other dummy fields in the virtual user structure
1149 		 * are ignored
1150 		 */
1151 		break;
1152 	}
1153 	return 0;
1154 }
1155 
1156 #undef R32
1157 #undef SEG32
1158 
1159 #define R32(l,q)							\
1160 	case offsetof(struct user32, regs.l):				\
1161 		*val = regs->q; break
1162 
1163 #define SEG32(rs)							\
1164 	case offsetof(struct user32, regs.rs):				\
1165 		*val = get_segment_reg(child,				\
1166 				       offsetof(struct user_regs_struct, rs)); \
1167 		break
1168 
1169 static int getreg32(struct task_struct *child, unsigned regno, u32 *val)
1170 {
1171 	struct pt_regs *regs = task_pt_regs(child);
1172 
1173 	switch (regno) {
1174 
1175 	SEG32(ds);
1176 	SEG32(es);
1177 	SEG32(fs);
1178 	SEG32(gs);
1179 
1180 	R32(cs, cs);
1181 	R32(ss, ss);
1182 	R32(ebx, bx);
1183 	R32(ecx, cx);
1184 	R32(edx, dx);
1185 	R32(edi, di);
1186 	R32(esi, si);
1187 	R32(ebp, bp);
1188 	R32(eax, ax);
1189 	R32(orig_eax, orig_ax);
1190 	R32(eip, ip);
1191 	R32(esp, sp);
1192 
1193 	case offsetof(struct user32, regs.eflags):
1194 		*val = get_flags(child);
1195 		break;
1196 
1197 	case offsetof(struct user32, u_debugreg[0]) ...
1198 		offsetof(struct user32, u_debugreg[7]):
1199 		regno -= offsetof(struct user32, u_debugreg[0]);
1200 		*val = ptrace_get_debugreg(child, regno / 4);
1201 		break;
1202 
1203 	default:
1204 		if (regno > sizeof(struct user32) || (regno & 3))
1205 			return -EIO;
1206 
1207 		/*
1208 		 * Other dummy fields in the virtual user structure
1209 		 * are ignored
1210 		 */
1211 		*val = 0;
1212 		break;
1213 	}
1214 	return 0;
1215 }
1216 
1217 #undef R32
1218 #undef SEG32
1219 
1220 static int genregs32_get(struct task_struct *target,
1221 			 const struct user_regset *regset,
1222 			 unsigned int pos, unsigned int count,
1223 			 void *kbuf, void __user *ubuf)
1224 {
1225 	if (kbuf) {
1226 		compat_ulong_t *k = kbuf;
1227 		while (count > 0) {
1228 			getreg32(target, pos, k++);
1229 			count -= sizeof(*k);
1230 			pos += sizeof(*k);
1231 		}
1232 	} else {
1233 		compat_ulong_t __user *u = ubuf;
1234 		while (count > 0) {
1235 			compat_ulong_t word;
1236 			getreg32(target, pos, &word);
1237 			if (__put_user(word, u++))
1238 				return -EFAULT;
1239 			count -= sizeof(*u);
1240 			pos += sizeof(*u);
1241 		}
1242 	}
1243 
1244 	return 0;
1245 }
1246 
1247 static int genregs32_set(struct task_struct *target,
1248 			 const struct user_regset *regset,
1249 			 unsigned int pos, unsigned int count,
1250 			 const void *kbuf, const void __user *ubuf)
1251 {
1252 	int ret = 0;
1253 	if (kbuf) {
1254 		const compat_ulong_t *k = kbuf;
1255 		while (count > 0 && !ret) {
1256 			ret = putreg32(target, pos, *k++);
1257 			count -= sizeof(*k);
1258 			pos += sizeof(*k);
1259 		}
1260 	} else {
1261 		const compat_ulong_t __user *u = ubuf;
1262 		while (count > 0 && !ret) {
1263 			compat_ulong_t word;
1264 			ret = __get_user(word, u++);
1265 			if (ret)
1266 				break;
1267 			ret = putreg32(target, pos, word);
1268 			count -= sizeof(*u);
1269 			pos += sizeof(*u);
1270 		}
1271 	}
1272 	return ret;
1273 }
1274 
1275 long compat_arch_ptrace(struct task_struct *child, compat_long_t request,
1276 			compat_ulong_t caddr, compat_ulong_t cdata)
1277 {
1278 	unsigned long addr = caddr;
1279 	unsigned long data = cdata;
1280 	void __user *datap = compat_ptr(data);
1281 	int ret;
1282 	__u32 val;
1283 
1284 	switch (request) {
1285 	case PTRACE_PEEKUSR:
1286 		ret = getreg32(child, addr, &val);
1287 		if (ret == 0)
1288 			ret = put_user(val, (__u32 __user *)datap);
1289 		break;
1290 
1291 	case PTRACE_POKEUSR:
1292 		ret = putreg32(child, addr, data);
1293 		break;
1294 
1295 	case PTRACE_GETREGS:	/* Get all gp regs from the child. */
1296 		return copy_regset_to_user(child, &user_x86_32_view,
1297 					   REGSET_GENERAL,
1298 					   0, sizeof(struct user_regs_struct32),
1299 					   datap);
1300 
1301 	case PTRACE_SETREGS:	/* Set all gp regs in the child. */
1302 		return copy_regset_from_user(child, &user_x86_32_view,
1303 					     REGSET_GENERAL, 0,
1304 					     sizeof(struct user_regs_struct32),
1305 					     datap);
1306 
1307 	case PTRACE_GETFPREGS:	/* Get the child FPU state. */
1308 		return copy_regset_to_user(child, &user_x86_32_view,
1309 					   REGSET_FP, 0,
1310 					   sizeof(struct user_i387_ia32_struct),
1311 					   datap);
1312 
1313 	case PTRACE_SETFPREGS:	/* Set the child FPU state. */
1314 		return copy_regset_from_user(
1315 			child, &user_x86_32_view, REGSET_FP,
1316 			0, sizeof(struct user_i387_ia32_struct), datap);
1317 
1318 	case PTRACE_GETFPXREGS:	/* Get the child extended FPU state. */
1319 		return copy_regset_to_user(child, &user_x86_32_view,
1320 					   REGSET_XFP, 0,
1321 					   sizeof(struct user32_fxsr_struct),
1322 					   datap);
1323 
1324 	case PTRACE_SETFPXREGS:	/* Set the child extended FPU state. */
1325 		return copy_regset_from_user(child, &user_x86_32_view,
1326 					     REGSET_XFP, 0,
1327 					     sizeof(struct user32_fxsr_struct),
1328 					     datap);
1329 
1330 	case PTRACE_GET_THREAD_AREA:
1331 	case PTRACE_SET_THREAD_AREA:
1332 #ifdef CONFIG_X86_PTRACE_BTS
1333 	case PTRACE_BTS_CONFIG:
1334 	case PTRACE_BTS_STATUS:
1335 	case PTRACE_BTS_SIZE:
1336 	case PTRACE_BTS_GET:
1337 	case PTRACE_BTS_CLEAR:
1338 	case PTRACE_BTS_DRAIN:
1339 #endif /* CONFIG_X86_PTRACE_BTS */
1340 		return arch_ptrace(child, request, addr, data);
1341 
1342 	default:
1343 		return compat_ptrace_request(child, request, addr, data);
1344 	}
1345 
1346 	return ret;
1347 }
1348 
1349 #endif	/* CONFIG_IA32_EMULATION */
1350 
1351 #ifdef CONFIG_X86_64
1352 
1353 static const struct user_regset x86_64_regsets[] = {
1354 	[REGSET_GENERAL] = {
1355 		.core_note_type = NT_PRSTATUS,
1356 		.n = sizeof(struct user_regs_struct) / sizeof(long),
1357 		.size = sizeof(long), .align = sizeof(long),
1358 		.get = genregs_get, .set = genregs_set
1359 	},
1360 	[REGSET_FP] = {
1361 		.core_note_type = NT_PRFPREG,
1362 		.n = sizeof(struct user_i387_struct) / sizeof(long),
1363 		.size = sizeof(long), .align = sizeof(long),
1364 		.active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1365 	},
1366 	[REGSET_IOPERM64] = {
1367 		.core_note_type = NT_386_IOPERM,
1368 		.n = IO_BITMAP_LONGS,
1369 		.size = sizeof(long), .align = sizeof(long),
1370 		.active = ioperm_active, .get = ioperm_get
1371 	},
1372 };
1373 
1374 static const struct user_regset_view user_x86_64_view = {
1375 	.name = "x86_64", .e_machine = EM_X86_64,
1376 	.regsets = x86_64_regsets, .n = ARRAY_SIZE(x86_64_regsets)
1377 };
1378 
1379 #else  /* CONFIG_X86_32 */
1380 
1381 #define user_regs_struct32	user_regs_struct
1382 #define genregs32_get		genregs_get
1383 #define genregs32_set		genregs_set
1384 
1385 #define user_i387_ia32_struct	user_i387_struct
1386 #define user32_fxsr_struct	user_fxsr_struct
1387 
1388 #endif	/* CONFIG_X86_64 */
1389 
1390 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1391 static const struct user_regset x86_32_regsets[] = {
1392 	[REGSET_GENERAL] = {
1393 		.core_note_type = NT_PRSTATUS,
1394 		.n = sizeof(struct user_regs_struct32) / sizeof(u32),
1395 		.size = sizeof(u32), .align = sizeof(u32),
1396 		.get = genregs32_get, .set = genregs32_set
1397 	},
1398 	[REGSET_FP] = {
1399 		.core_note_type = NT_PRFPREG,
1400 		.n = sizeof(struct user_i387_ia32_struct) / sizeof(u32),
1401 		.size = sizeof(u32), .align = sizeof(u32),
1402 		.active = fpregs_active, .get = fpregs_get, .set = fpregs_set
1403 	},
1404 	[REGSET_XFP] = {
1405 		.core_note_type = NT_PRXFPREG,
1406 		.n = sizeof(struct user32_fxsr_struct) / sizeof(u32),
1407 		.size = sizeof(u32), .align = sizeof(u32),
1408 		.active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1409 	},
1410 	[REGSET_TLS] = {
1411 		.core_note_type = NT_386_TLS,
1412 		.n = GDT_ENTRY_TLS_ENTRIES, .bias = GDT_ENTRY_TLS_MIN,
1413 		.size = sizeof(struct user_desc),
1414 		.align = sizeof(struct user_desc),
1415 		.active = regset_tls_active,
1416 		.get = regset_tls_get, .set = regset_tls_set
1417 	},
1418 	[REGSET_IOPERM32] = {
1419 		.core_note_type = NT_386_IOPERM,
1420 		.n = IO_BITMAP_BYTES / sizeof(u32),
1421 		.size = sizeof(u32), .align = sizeof(u32),
1422 		.active = ioperm_active, .get = ioperm_get
1423 	},
1424 };
1425 
1426 static const struct user_regset_view user_x86_32_view = {
1427 	.name = "i386", .e_machine = EM_386,
1428 	.regsets = x86_32_regsets, .n = ARRAY_SIZE(x86_32_regsets)
1429 };
1430 #endif
1431 
1432 const struct user_regset_view *task_user_regset_view(struct task_struct *task)
1433 {
1434 #ifdef CONFIG_IA32_EMULATION
1435 	if (test_tsk_thread_flag(task, TIF_IA32))
1436 #endif
1437 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1438 		return &user_x86_32_view;
1439 #endif
1440 #ifdef CONFIG_X86_64
1441 	return &user_x86_64_view;
1442 #endif
1443 }
1444 
1445 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs,
1446 					 int error_code, int si_code)
1447 {
1448 	struct siginfo info;
1449 
1450 	tsk->thread.trap_no = 1;
1451 	tsk->thread.error_code = error_code;
1452 
1453 	memset(&info, 0, sizeof(info));
1454 	info.si_signo = SIGTRAP;
1455 	info.si_code = si_code;
1456 
1457 	/* User-mode ip? */
1458 	info.si_addr = user_mode_vm(regs) ? (void __user *) regs->ip : NULL;
1459 
1460 	/* Send us the fake SIGTRAP */
1461 	force_sig_info(SIGTRAP, &info, tsk);
1462 }
1463 
1464 
1465 #ifdef CONFIG_X86_32
1466 # define IS_IA32	1
1467 #elif defined CONFIG_IA32_EMULATION
1468 # define IS_IA32	is_compat_task()
1469 #else
1470 # define IS_IA32	0
1471 #endif
1472 
1473 /*
1474  * We must return the syscall number to actually look up in the table.
1475  * This can be -1L to skip running any syscall at all.
1476  */
1477 asmregparm long syscall_trace_enter(struct pt_regs *regs)
1478 {
1479 	long ret = 0;
1480 
1481 	/*
1482 	 * If we stepped into a sysenter/syscall insn, it trapped in
1483 	 * kernel mode; do_debug() cleared TF and set TIF_SINGLESTEP.
1484 	 * If user-mode had set TF itself, then it's still clear from
1485 	 * do_debug() and we need to set it again to restore the user
1486 	 * state.  If we entered on the slow path, TF was already set.
1487 	 */
1488 	if (test_thread_flag(TIF_SINGLESTEP))
1489 		regs->flags |= X86_EFLAGS_TF;
1490 
1491 	/* do the secure computing check first */
1492 	secure_computing(regs->orig_ax);
1493 
1494 	if (unlikely(test_thread_flag(TIF_SYSCALL_EMU)))
1495 		ret = -1L;
1496 
1497 	if ((ret || test_thread_flag(TIF_SYSCALL_TRACE)) &&
1498 	    tracehook_report_syscall_entry(regs))
1499 		ret = -1L;
1500 
1501 	if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
1502 		trace_sys_enter(regs, regs->orig_ax);
1503 
1504 	if (unlikely(current->audit_context)) {
1505 		if (IS_IA32)
1506 			audit_syscall_entry(AUDIT_ARCH_I386,
1507 					    regs->orig_ax,
1508 					    regs->bx, regs->cx,
1509 					    regs->dx, regs->si);
1510 #ifdef CONFIG_X86_64
1511 		else
1512 			audit_syscall_entry(AUDIT_ARCH_X86_64,
1513 					    regs->orig_ax,
1514 					    regs->di, regs->si,
1515 					    regs->dx, regs->r10);
1516 #endif
1517 	}
1518 
1519 	return ret ?: regs->orig_ax;
1520 }
1521 
1522 asmregparm void syscall_trace_leave(struct pt_regs *regs)
1523 {
1524 	if (unlikely(current->audit_context))
1525 		audit_syscall_exit(AUDITSC_RESULT(regs->ax), regs->ax);
1526 
1527 	if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
1528 		trace_sys_exit(regs, regs->ax);
1529 
1530 	if (test_thread_flag(TIF_SYSCALL_TRACE))
1531 		tracehook_report_syscall_exit(regs, 0);
1532 
1533 	/*
1534 	 * If TIF_SYSCALL_EMU is set, we only get here because of
1535 	 * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
1536 	 * We already reported this syscall instruction in
1537 	 * syscall_trace_enter(), so don't do any more now.
1538 	 */
1539 	if (unlikely(test_thread_flag(TIF_SYSCALL_EMU)))
1540 		return;
1541 
1542 	/*
1543 	 * If we are single-stepping, synthesize a trap to follow the
1544 	 * system call instruction.
1545 	 */
1546 	if (test_thread_flag(TIF_SINGLESTEP) &&
1547 	    tracehook_consider_fatal_signal(current, SIGTRAP))
1548 		send_sigtrap(current, regs, 0, TRAP_BRKPT);
1549 }
1550