x86, ptrace: rlimit BTS buffer allocation
[safe/jmp/linux-2.6] / arch / x86 / kernel / ptrace.c
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/user.h>
18 #include <linux/elf.h>
19 #include <linux/security.h>
20 #include <linux/audit.h>
21 #include <linux/seccomp.h>
22 #include <linux/signal.h>
23
24 #include <asm/uaccess.h>
25 #include <asm/pgtable.h>
26 #include <asm/system.h>
27 #include <asm/processor.h>
28 #include <asm/i387.h>
29 #include <asm/debugreg.h>
30 #include <asm/ldt.h>
31 #include <asm/desc.h>
32 #include <asm/prctl.h>
33 #include <asm/proto.h>
34 #include <asm/ds.h>
35
36 #include "tls.h"
37
38 enum x86_regset {
39         REGSET_GENERAL,
40         REGSET_FP,
41         REGSET_XFP,
42         REGSET_TLS,
43 };
44
45 /*
46  * does not yet catch signals sent when the child dies.
47  * in exit.c or in signal.c.
48  */
49
50 /*
51  * Determines which flags the user has access to [1 = access, 0 = no access].
52  */
53 #define FLAG_MASK_32            ((unsigned long)                        \
54                                  (X86_EFLAGS_CF | X86_EFLAGS_PF |       \
55                                   X86_EFLAGS_AF | X86_EFLAGS_ZF |       \
56                                   X86_EFLAGS_SF | X86_EFLAGS_TF |       \
57                                   X86_EFLAGS_DF | X86_EFLAGS_OF |       \
58                                   X86_EFLAGS_RF | X86_EFLAGS_AC))
59
60 /*
61  * Determines whether a value may be installed in a segment register.
62  */
63 static inline bool invalid_selector(u16 value)
64 {
65         return unlikely(value != 0 && (value & SEGMENT_RPL_MASK) != USER_RPL);
66 }
67
68 #ifdef CONFIG_X86_32
69
70 #define FLAG_MASK               FLAG_MASK_32
71
72 static long *pt_regs_access(struct pt_regs *regs, unsigned long regno)
73 {
74         BUILD_BUG_ON(offsetof(struct pt_regs, bx) != 0);
75         regno >>= 2;
76         if (regno > FS)
77                 --regno;
78         return &regs->bx + regno;
79 }
80
81 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
82 {
83         /*
84          * Returning the value truncates it to 16 bits.
85          */
86         unsigned int retval;
87         if (offset != offsetof(struct user_regs_struct, gs))
88                 retval = *pt_regs_access(task_pt_regs(task), offset);
89         else {
90                 retval = task->thread.gs;
91                 if (task == current)
92                         savesegment(gs, retval);
93         }
94         return retval;
95 }
96
97 static int set_segment_reg(struct task_struct *task,
98                            unsigned long offset, u16 value)
99 {
100         /*
101          * The value argument was already truncated to 16 bits.
102          */
103         if (invalid_selector(value))
104                 return -EIO;
105
106         if (offset != offsetof(struct user_regs_struct, gs))
107                 *pt_regs_access(task_pt_regs(task), offset) = value;
108         else {
109                 task->thread.gs = value;
110                 if (task == current)
111                         /*
112                          * The user-mode %gs is not affected by
113                          * kernel entry, so we must update the CPU.
114                          */
115                         loadsegment(gs, value);
116         }
117
118         return 0;
119 }
120
121 static unsigned long debugreg_addr_limit(struct task_struct *task)
122 {
123         return TASK_SIZE - 3;
124 }
125
126 #else  /* CONFIG_X86_64 */
127
128 #define FLAG_MASK               (FLAG_MASK_32 | X86_EFLAGS_NT)
129
130 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long offset)
131 {
132         BUILD_BUG_ON(offsetof(struct pt_regs, r15) != 0);
133         return &regs->r15 + (offset / sizeof(regs->r15));
134 }
135
136 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
137 {
138         /*
139          * Returning the value truncates it to 16 bits.
140          */
141         unsigned int seg;
142
143         switch (offset) {
144         case offsetof(struct user_regs_struct, fs):
145                 if (task == current) {
146                         /* Older gas can't assemble movq %?s,%r?? */
147                         asm("movl %%fs,%0" : "=r" (seg));
148                         return seg;
149                 }
150                 return task->thread.fsindex;
151         case offsetof(struct user_regs_struct, gs):
152                 if (task == current) {
153                         asm("movl %%gs,%0" : "=r" (seg));
154                         return seg;
155                 }
156                 return task->thread.gsindex;
157         case offsetof(struct user_regs_struct, ds):
158                 if (task == current) {
159                         asm("movl %%ds,%0" : "=r" (seg));
160                         return seg;
161                 }
162                 return task->thread.ds;
163         case offsetof(struct user_regs_struct, es):
164                 if (task == current) {
165                         asm("movl %%es,%0" : "=r" (seg));
166                         return seg;
167                 }
168                 return task->thread.es;
169
170         case offsetof(struct user_regs_struct, cs):
171         case offsetof(struct user_regs_struct, ss):
172                 break;
173         }
174         return *pt_regs_access(task_pt_regs(task), offset);
175 }
176
177 static int set_segment_reg(struct task_struct *task,
178                            unsigned long offset, u16 value)
179 {
180         /*
181          * The value argument was already truncated to 16 bits.
182          */
183         if (invalid_selector(value))
184                 return -EIO;
185
186         switch (offset) {
187         case offsetof(struct user_regs_struct,fs):
188                 /*
189                  * If this is setting fs as for normal 64-bit use but
190                  * setting fs_base has implicitly changed it, leave it.
191                  */
192                 if ((value == FS_TLS_SEL && task->thread.fsindex == 0 &&
193                      task->thread.fs != 0) ||
194                     (value == 0 && task->thread.fsindex == FS_TLS_SEL &&
195                      task->thread.fs == 0))
196                         break;
197                 task->thread.fsindex = value;
198                 if (task == current)
199                         loadsegment(fs, task->thread.fsindex);
200                 break;
201         case offsetof(struct user_regs_struct,gs):
202                 /*
203                  * If this is setting gs as for normal 64-bit use but
204                  * setting gs_base has implicitly changed it, leave it.
205                  */
206                 if ((value == GS_TLS_SEL && task->thread.gsindex == 0 &&
207                      task->thread.gs != 0) ||
208                     (value == 0 && task->thread.gsindex == GS_TLS_SEL &&
209                      task->thread.gs == 0))
210                         break;
211                 task->thread.gsindex = value;
212                 if (task == current)
213                         load_gs_index(task->thread.gsindex);
214                 break;
215         case offsetof(struct user_regs_struct,ds):
216                 task->thread.ds = value;
217                 if (task == current)
218                         loadsegment(ds, task->thread.ds);
219                 break;
220         case offsetof(struct user_regs_struct,es):
221                 task->thread.es = value;
222                 if (task == current)
223                         loadsegment(es, task->thread.es);
224                 break;
225
226                 /*
227                  * Can't actually change these in 64-bit mode.
228                  */
229         case offsetof(struct user_regs_struct,cs):
230 #ifdef CONFIG_IA32_EMULATION
231                 if (test_tsk_thread_flag(task, TIF_IA32))
232                         task_pt_regs(task)->cs = value;
233 #endif
234                 break;
235         case offsetof(struct user_regs_struct,ss):
236 #ifdef CONFIG_IA32_EMULATION
237                 if (test_tsk_thread_flag(task, TIF_IA32))
238                         task_pt_regs(task)->ss = value;
239 #endif
240                 break;
241         }
242
243         return 0;
244 }
245
246 static unsigned long debugreg_addr_limit(struct task_struct *task)
247 {
248 #ifdef CONFIG_IA32_EMULATION
249         if (test_tsk_thread_flag(task, TIF_IA32))
250                 return IA32_PAGE_OFFSET - 3;
251 #endif
252         return TASK_SIZE64 - 7;
253 }
254
255 #endif  /* CONFIG_X86_32 */
256
257 static unsigned long get_flags(struct task_struct *task)
258 {
259         unsigned long retval = task_pt_regs(task)->flags;
260
261         /*
262          * If the debugger set TF, hide it from the readout.
263          */
264         if (test_tsk_thread_flag(task, TIF_FORCED_TF))
265                 retval &= ~X86_EFLAGS_TF;
266
267         return retval;
268 }
269
270 static int set_flags(struct task_struct *task, unsigned long value)
271 {
272         struct pt_regs *regs = task_pt_regs(task);
273
274         /*
275          * If the user value contains TF, mark that
276          * it was not "us" (the debugger) that set it.
277          * If not, make sure it stays set if we had.
278          */
279         if (value & X86_EFLAGS_TF)
280                 clear_tsk_thread_flag(task, TIF_FORCED_TF);
281         else if (test_tsk_thread_flag(task, TIF_FORCED_TF))
282                 value |= X86_EFLAGS_TF;
283
284         regs->flags = (regs->flags & ~FLAG_MASK) | (value & FLAG_MASK);
285
286         return 0;
287 }
288
289 static int putreg(struct task_struct *child,
290                   unsigned long offset, unsigned long value)
291 {
292         switch (offset) {
293         case offsetof(struct user_regs_struct, cs):
294         case offsetof(struct user_regs_struct, ds):
295         case offsetof(struct user_regs_struct, es):
296         case offsetof(struct user_regs_struct, fs):
297         case offsetof(struct user_regs_struct, gs):
298         case offsetof(struct user_regs_struct, ss):
299                 return set_segment_reg(child, offset, value);
300
301         case offsetof(struct user_regs_struct, flags):
302                 return set_flags(child, value);
303
304 #ifdef CONFIG_X86_64
305         case offsetof(struct user_regs_struct,fs_base):
306                 if (value >= TASK_SIZE_OF(child))
307                         return -EIO;
308                 /*
309                  * When changing the segment base, use do_arch_prctl
310                  * to set either thread.fs or thread.fsindex and the
311                  * corresponding GDT slot.
312                  */
313                 if (child->thread.fs != value)
314                         return do_arch_prctl(child, ARCH_SET_FS, value);
315                 return 0;
316         case offsetof(struct user_regs_struct,gs_base):
317                 /*
318                  * Exactly the same here as the %fs handling above.
319                  */
320                 if (value >= TASK_SIZE_OF(child))
321                         return -EIO;
322                 if (child->thread.gs != value)
323                         return do_arch_prctl(child, ARCH_SET_GS, value);
324                 return 0;
325 #endif
326         }
327
328         *pt_regs_access(task_pt_regs(child), offset) = value;
329         return 0;
330 }
331
332 static unsigned long getreg(struct task_struct *task, unsigned long offset)
333 {
334         switch (offset) {
335         case offsetof(struct user_regs_struct, cs):
336         case offsetof(struct user_regs_struct, ds):
337         case offsetof(struct user_regs_struct, es):
338         case offsetof(struct user_regs_struct, fs):
339         case offsetof(struct user_regs_struct, gs):
340         case offsetof(struct user_regs_struct, ss):
341                 return get_segment_reg(task, offset);
342
343         case offsetof(struct user_regs_struct, flags):
344                 return get_flags(task);
345
346 #ifdef CONFIG_X86_64
347         case offsetof(struct user_regs_struct, fs_base): {
348                 /*
349                  * do_arch_prctl may have used a GDT slot instead of
350                  * the MSR.  To userland, it appears the same either
351                  * way, except the %fs segment selector might not be 0.
352                  */
353                 unsigned int seg = task->thread.fsindex;
354                 if (task->thread.fs != 0)
355                         return task->thread.fs;
356                 if (task == current)
357                         asm("movl %%fs,%0" : "=r" (seg));
358                 if (seg != FS_TLS_SEL)
359                         return 0;
360                 return get_desc_base(&task->thread.tls_array[FS_TLS]);
361         }
362         case offsetof(struct user_regs_struct, gs_base): {
363                 /*
364                  * Exactly the same here as the %fs handling above.
365                  */
366                 unsigned int seg = task->thread.gsindex;
367                 if (task->thread.gs != 0)
368                         return task->thread.gs;
369                 if (task == current)
370                         asm("movl %%gs,%0" : "=r" (seg));
371                 if (seg != GS_TLS_SEL)
372                         return 0;
373                 return get_desc_base(&task->thread.tls_array[GS_TLS]);
374         }
375 #endif
376         }
377
378         return *pt_regs_access(task_pt_regs(task), offset);
379 }
380
381 static int genregs_get(struct task_struct *target,
382                        const struct user_regset *regset,
383                        unsigned int pos, unsigned int count,
384                        void *kbuf, void __user *ubuf)
385 {
386         if (kbuf) {
387                 unsigned long *k = kbuf;
388                 while (count > 0) {
389                         *k++ = getreg(target, pos);
390                         count -= sizeof(*k);
391                         pos += sizeof(*k);
392                 }
393         } else {
394                 unsigned long __user *u = ubuf;
395                 while (count > 0) {
396                         if (__put_user(getreg(target, pos), u++))
397                                 return -EFAULT;
398                         count -= sizeof(*u);
399                         pos += sizeof(*u);
400                 }
401         }
402
403         return 0;
404 }
405
406 static int genregs_set(struct task_struct *target,
407                        const struct user_regset *regset,
408                        unsigned int pos, unsigned int count,
409                        const void *kbuf, const void __user *ubuf)
410 {
411         int ret = 0;
412         if (kbuf) {
413                 const unsigned long *k = kbuf;
414                 while (count > 0 && !ret) {
415                         ret = putreg(target, pos, *k++);
416                         count -= sizeof(*k);
417                         pos += sizeof(*k);
418                 }
419         } else {
420                 const unsigned long  __user *u = ubuf;
421                 while (count > 0 && !ret) {
422                         unsigned long word;
423                         ret = __get_user(word, u++);
424                         if (ret)
425                                 break;
426                         ret = putreg(target, pos, word);
427                         count -= sizeof(*u);
428                         pos += sizeof(*u);
429                 }
430         }
431         return ret;
432 }
433
434 /*
435  * This function is trivial and will be inlined by the compiler.
436  * Having it separates the implementation details of debug
437  * registers from the interface details of ptrace.
438  */
439 static unsigned long ptrace_get_debugreg(struct task_struct *child, int n)
440 {
441         switch (n) {
442         case 0:         return child->thread.debugreg0;
443         case 1:         return child->thread.debugreg1;
444         case 2:         return child->thread.debugreg2;
445         case 3:         return child->thread.debugreg3;
446         case 6:         return child->thread.debugreg6;
447         case 7:         return child->thread.debugreg7;
448         }
449         return 0;
450 }
451
452 static int ptrace_set_debugreg(struct task_struct *child,
453                                int n, unsigned long data)
454 {
455         int i;
456
457         if (unlikely(n == 4 || n == 5))
458                 return -EIO;
459
460         if (n < 4 && unlikely(data >= debugreg_addr_limit(child)))
461                 return -EIO;
462
463         switch (n) {
464         case 0:         child->thread.debugreg0 = data; break;
465         case 1:         child->thread.debugreg1 = data; break;
466         case 2:         child->thread.debugreg2 = data; break;
467         case 3:         child->thread.debugreg3 = data; break;
468
469         case 6:
470                 if ((data & ~0xffffffffUL) != 0)
471                         return -EIO;
472                 child->thread.debugreg6 = data;
473                 break;
474
475         case 7:
476                 /*
477                  * Sanity-check data. Take one half-byte at once with
478                  * check = (val >> (16 + 4*i)) & 0xf. It contains the
479                  * R/Wi and LENi bits; bits 0 and 1 are R/Wi, and bits
480                  * 2 and 3 are LENi. Given a list of invalid values,
481                  * we do mask |= 1 << invalid_value, so that
482                  * (mask >> check) & 1 is a correct test for invalid
483                  * values.
484                  *
485                  * R/Wi contains the type of the breakpoint /
486                  * watchpoint, LENi contains the length of the watched
487                  * data in the watchpoint case.
488                  *
489                  * The invalid values are:
490                  * - LENi == 0x10 (undefined), so mask |= 0x0f00.       [32-bit]
491                  * - R/Wi == 0x10 (break on I/O reads or writes), so
492                  *   mask |= 0x4444.
493                  * - R/Wi == 0x00 && LENi != 0x00, so we have mask |=
494                  *   0x1110.
495                  *
496                  * Finally, mask = 0x0f00 | 0x4444 | 0x1110 == 0x5f54.
497                  *
498                  * See the Intel Manual "System Programming Guide",
499                  * 15.2.4
500                  *
501                  * Note that LENi == 0x10 is defined on x86_64 in long
502                  * mode (i.e. even for 32-bit userspace software, but
503                  * 64-bit kernel), so the x86_64 mask value is 0x5454.
504                  * See the AMD manual no. 24593 (AMD64 System Programming)
505                  */
506 #ifdef CONFIG_X86_32
507 #define DR7_MASK        0x5f54
508 #else
509 #define DR7_MASK        0x5554
510 #endif
511                 data &= ~DR_CONTROL_RESERVED;
512                 for (i = 0; i < 4; i++)
513                         if ((DR7_MASK >> ((data >> (16 + 4*i)) & 0xf)) & 1)
514                                 return -EIO;
515                 child->thread.debugreg7 = data;
516                 if (data)
517                         set_tsk_thread_flag(child, TIF_DEBUG);
518                 else
519                         clear_tsk_thread_flag(child, TIF_DEBUG);
520                 break;
521         }
522
523         return 0;
524 }
525
526 static int ptrace_bts_get_size(struct task_struct *child)
527 {
528         if (!child->thread.ds_area_msr)
529                 return -ENXIO;
530
531         return ds_get_bts_index((void *)child->thread.ds_area_msr);
532 }
533
534 static int ptrace_bts_read_record(struct task_struct *child,
535                                   long index,
536                                   struct bts_struct __user *out)
537 {
538         struct bts_struct ret;
539         int retval;
540         int bts_end;
541         int bts_index;
542
543         if (!child->thread.ds_area_msr)
544                 return -ENXIO;
545
546         if (index < 0)
547                 return -EINVAL;
548
549         bts_end = ds_get_bts_end((void *)child->thread.ds_area_msr);
550         if (bts_end <= index)
551                 return -EINVAL;
552
553         /* translate the ptrace bts index into the ds bts index */
554         bts_index = ds_get_bts_index((void *)child->thread.ds_area_msr);
555         bts_index -= (index + 1);
556         if (bts_index < 0)
557                 bts_index += bts_end;
558
559         retval = ds_read_bts((void *)child->thread.ds_area_msr,
560                              bts_index, &ret);
561         if (retval)
562                 return retval;
563
564         if (copy_to_user(out, &ret, sizeof(ret)))
565                 return -EFAULT;
566
567         return sizeof(ret);
568 }
569
570 static int ptrace_bts_write_record(struct task_struct *child,
571                                    const struct bts_struct *in)
572 {
573         int retval;
574
575         if (!child->thread.ds_area_msr)
576                 return -ENXIO;
577
578         retval = ds_write_bts((void *)child->thread.ds_area_msr, in);
579         if (retval)
580                 return retval;
581
582         return sizeof(*in);
583 }
584
585 static int ptrace_bts_clear(struct task_struct *child)
586 {
587         if (!child->thread.ds_area_msr)
588                 return -ENXIO;
589
590         return ds_clear((void *)child->thread.ds_area_msr);
591 }
592
593 static int ptrace_bts_drain(struct task_struct *child,
594                             struct bts_struct __user *out)
595 {
596         int end, i;
597         void *ds = (void *)child->thread.ds_area_msr;
598
599         if (!ds)
600                 return -ENXIO;
601
602         end = ds_get_bts_index(ds);
603         if (end <= 0)
604                 return end;
605
606         for (i = 0; i < end; i++, out++) {
607                 struct bts_struct ret;
608                 int retval;
609
610                 retval = ds_read_bts(ds, i, &ret);
611                 if (retval < 0)
612                         return retval;
613
614                 if (copy_to_user(out, &ret, sizeof(ret)))
615                         return -EFAULT;
616         }
617
618         ds_clear(ds);
619
620         return i;
621 }
622
623 static int ptrace_bts_realloc(struct task_struct *child,
624                               int size, int reduce_size)
625 {
626         unsigned long rlim, vm;
627         int ret, old_size;
628
629         if (size < 0)
630                 return -EINVAL;
631
632         old_size = ds_get_bts_size((void *)child->thread.ds_area_msr);
633         if (old_size < 0)
634                 return old_size;
635
636         ret = ds_free((void **)&child->thread.ds_area_msr);
637         if (ret < 0)
638                 goto out;
639
640         size >>= PAGE_SHIFT;
641         old_size >>= PAGE_SHIFT;
642
643         current->mm->total_vm  -= old_size;
644         current->mm->locked_vm -= old_size;
645
646         if (size == 0)
647                 goto out;
648
649         rlim = current->signal->rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT;
650         vm = current->mm->total_vm  + size;
651         if (rlim < vm) {
652                 ret = -ENOMEM;
653
654                 if (!reduce_size)
655                         goto out;
656
657                 size = rlim - current->mm->total_vm;
658                 if (size <= 0)
659                         goto out;
660         }
661
662         rlim = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT;
663         vm = current->mm->locked_vm  + size;
664         if (rlim < vm) {
665                 ret = -ENOMEM;
666
667                 if (!reduce_size)
668                         goto out;
669
670                 size = rlim - current->mm->locked_vm;
671                 if (size <= 0)
672                         goto out;
673         }
674
675         ret = ds_allocate((void **)&child->thread.ds_area_msr,
676                           size << PAGE_SHIFT);
677         if (ret < 0)
678                 goto out;
679
680         current->mm->total_vm  += size;
681         current->mm->locked_vm += size;
682
683 out:
684         if (child->thread.ds_area_msr)
685                 set_tsk_thread_flag(child, TIF_DS_AREA_MSR);
686         else
687                 clear_tsk_thread_flag(child, TIF_DS_AREA_MSR);
688
689         return ret;
690 }
691
692 static int ptrace_bts_config(struct task_struct *child,
693                              const struct ptrace_bts_config __user *ucfg)
694 {
695         struct ptrace_bts_config cfg;
696         int bts_size, ret = 0;
697         void *ds;
698
699         if (copy_from_user(&cfg, ucfg, sizeof(cfg)))
700                 return -EFAULT;
701
702         bts_size = 0;
703         ds = (void *)child->thread.ds_area_msr;
704         if (ds) {
705                 bts_size = ds_get_bts_size(ds);
706                 if (bts_size < 0)
707                         return bts_size;
708         }
709         cfg.size = PAGE_ALIGN(cfg.size);
710
711         if (bts_size != cfg.size) {
712                 ret = ptrace_bts_realloc(child, cfg.size,
713                                          cfg.flags & PTRACE_BTS_O_CUT_SIZE);
714                 if (ret < 0)
715                         goto errout;
716
717                 ds = (void *)child->thread.ds_area_msr;
718         }
719
720         if (cfg.flags & PTRACE_BTS_O_SIGNAL)
721                 ret = ds_set_overflow(ds, DS_O_SIGNAL);
722         else
723                 ret = ds_set_overflow(ds, DS_O_WRAP);
724         if (ret < 0)
725                 goto errout;
726
727         if (cfg.flags & PTRACE_BTS_O_TRACE)
728                 child->thread.debugctlmsr |= ds_debugctl_mask();
729         else
730                 child->thread.debugctlmsr &= ~ds_debugctl_mask();
731
732         if (cfg.flags & PTRACE_BTS_O_SCHED)
733                 set_tsk_thread_flag(child, TIF_BTS_TRACE_TS);
734         else
735                 clear_tsk_thread_flag(child, TIF_BTS_TRACE_TS);
736
737 out:
738         if (child->thread.debugctlmsr)
739                 set_tsk_thread_flag(child, TIF_DEBUGCTLMSR);
740         else
741                 clear_tsk_thread_flag(child, TIF_DEBUGCTLMSR);
742
743         return ret;
744
745 errout:
746         child->thread.debugctlmsr &= ~ds_debugctl_mask();
747         clear_tsk_thread_flag(child, TIF_BTS_TRACE_TS);
748         goto out;
749 }
750
751 static int ptrace_bts_status(struct task_struct *child,
752                              struct ptrace_bts_config __user *ucfg)
753 {
754         void *ds = (void *)child->thread.ds_area_msr;
755         struct ptrace_bts_config cfg;
756
757         memset(&cfg, 0, sizeof(cfg));
758
759         if (ds) {
760                 cfg.size = ds_get_bts_size(ds);
761
762                 if (ds_get_overflow(ds) == DS_O_SIGNAL)
763                         cfg.flags |= PTRACE_BTS_O_SIGNAL;
764
765                 if (test_tsk_thread_flag(child, TIF_DEBUGCTLMSR) &&
766                     child->thread.debugctlmsr & ds_debugctl_mask())
767                         cfg.flags |= PTRACE_BTS_O_TRACE;
768
769                 if (test_tsk_thread_flag(child, TIF_BTS_TRACE_TS))
770                         cfg.flags |= PTRACE_BTS_O_SCHED;
771         }
772
773         if (copy_to_user(ucfg, &cfg, sizeof(cfg)))
774                 return -EFAULT;
775
776         return sizeof(cfg);
777 }
778
779 void ptrace_bts_take_timestamp(struct task_struct *tsk,
780                                enum bts_qualifier qualifier)
781 {
782         struct bts_struct rec = {
783                 .qualifier = qualifier,
784                 .variant.jiffies = jiffies_64
785         };
786
787         ptrace_bts_write_record(tsk, &rec);
788 }
789
790 /*
791  * Called by kernel/ptrace.c when detaching..
792  *
793  * Make sure the single step bit is not set.
794  */
795 void ptrace_disable(struct task_struct *child)
796 {
797         user_disable_single_step(child);
798 #ifdef TIF_SYSCALL_EMU
799         clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
800 #endif
801         if (child->thread.ds_area_msr) {
802                 ptrace_bts_realloc(child, 0, 0);
803                 child->thread.debugctlmsr &= ~ds_debugctl_mask();
804                 if (!child->thread.debugctlmsr)
805                         clear_tsk_thread_flag(child, TIF_DEBUGCTLMSR);
806                 clear_tsk_thread_flag(child, TIF_BTS_TRACE_TS);
807         }
808 }
809
810 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
811 static const struct user_regset_view user_x86_32_view; /* Initialized below. */
812 #endif
813
814 long arch_ptrace(struct task_struct *child, long request, long addr, long data)
815 {
816         int ret;
817         unsigned long __user *datap = (unsigned long __user *)data;
818
819         switch (request) {
820         /* read the word at location addr in the USER area. */
821         case PTRACE_PEEKUSR: {
822                 unsigned long tmp;
823
824                 ret = -EIO;
825                 if ((addr & (sizeof(data) - 1)) || addr < 0 ||
826                     addr >= sizeof(struct user))
827                         break;
828
829                 tmp = 0;  /* Default return condition */
830                 if (addr < sizeof(struct user_regs_struct))
831                         tmp = getreg(child, addr);
832                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
833                          addr <= offsetof(struct user, u_debugreg[7])) {
834                         addr -= offsetof(struct user, u_debugreg[0]);
835                         tmp = ptrace_get_debugreg(child, addr / sizeof(data));
836                 }
837                 ret = put_user(tmp, datap);
838                 break;
839         }
840
841         case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
842                 ret = -EIO;
843                 if ((addr & (sizeof(data) - 1)) || addr < 0 ||
844                     addr >= sizeof(struct user))
845                         break;
846
847                 if (addr < sizeof(struct user_regs_struct))
848                         ret = putreg(child, addr, data);
849                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
850                          addr <= offsetof(struct user, u_debugreg[7])) {
851                         addr -= offsetof(struct user, u_debugreg[0]);
852                         ret = ptrace_set_debugreg(child,
853                                                   addr / sizeof(data), data);
854                 }
855                 break;
856
857         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
858                 return copy_regset_to_user(child,
859                                            task_user_regset_view(current),
860                                            REGSET_GENERAL,
861                                            0, sizeof(struct user_regs_struct),
862                                            datap);
863
864         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
865                 return copy_regset_from_user(child,
866                                              task_user_regset_view(current),
867                                              REGSET_GENERAL,
868                                              0, sizeof(struct user_regs_struct),
869                                              datap);
870
871         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
872                 return copy_regset_to_user(child,
873                                            task_user_regset_view(current),
874                                            REGSET_FP,
875                                            0, sizeof(struct user_i387_struct),
876                                            datap);
877
878         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
879                 return copy_regset_from_user(child,
880                                              task_user_regset_view(current),
881                                              REGSET_FP,
882                                              0, sizeof(struct user_i387_struct),
883                                              datap);
884
885 #ifdef CONFIG_X86_32
886         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
887                 return copy_regset_to_user(child, &user_x86_32_view,
888                                            REGSET_XFP,
889                                            0, sizeof(struct user_fxsr_struct),
890                                            datap);
891
892         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
893                 return copy_regset_from_user(child, &user_x86_32_view,
894                                              REGSET_XFP,
895                                              0, sizeof(struct user_fxsr_struct),
896                                              datap);
897 #endif
898
899 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
900         case PTRACE_GET_THREAD_AREA:
901                 if (addr < 0)
902                         return -EIO;
903                 ret = do_get_thread_area(child, addr,
904                                          (struct user_desc __user *) data);
905                 break;
906
907         case PTRACE_SET_THREAD_AREA:
908                 if (addr < 0)
909                         return -EIO;
910                 ret = do_set_thread_area(child, addr,
911                                          (struct user_desc __user *) data, 0);
912                 break;
913 #endif
914
915 #ifdef CONFIG_X86_64
916                 /* normal 64bit interface to access TLS data.
917                    Works just like arch_prctl, except that the arguments
918                    are reversed. */
919         case PTRACE_ARCH_PRCTL:
920                 ret = do_arch_prctl(child, data, addr);
921                 break;
922 #endif
923
924         case PTRACE_BTS_CONFIG:
925                 ret = ptrace_bts_config
926                         (child, (struct ptrace_bts_config __user *)addr);
927                 break;
928
929         case PTRACE_BTS_STATUS:
930                 ret = ptrace_bts_status
931                         (child, (struct ptrace_bts_config __user *)addr);
932                 break;
933
934         case PTRACE_BTS_SIZE:
935                 ret = ptrace_bts_get_size(child);
936                 break;
937
938         case PTRACE_BTS_GET:
939                 ret = ptrace_bts_read_record
940                         (child, data, (struct bts_struct __user *) addr);
941                 break;
942
943         case PTRACE_BTS_CLEAR:
944                 ret = ptrace_bts_clear(child);
945                 break;
946
947         case PTRACE_BTS_DRAIN:
948                 ret = ptrace_bts_drain
949                         (child, (struct bts_struct __user *) addr);
950                 break;
951
952         default:
953                 ret = ptrace_request(child, request, addr, data);
954                 break;
955         }
956
957         return ret;
958 }
959
960 #ifdef CONFIG_IA32_EMULATION
961
962 #include <linux/compat.h>
963 #include <linux/syscalls.h>
964 #include <asm/ia32.h>
965 #include <asm/user32.h>
966
967 #define R32(l,q)                                                        \
968         case offsetof(struct user32, regs.l):                           \
969                 regs->q = value; break
970
971 #define SEG32(rs)                                                       \
972         case offsetof(struct user32, regs.rs):                          \
973                 return set_segment_reg(child,                           \
974                                        offsetof(struct user_regs_struct, rs), \
975                                        value);                          \
976                 break
977
978 static int putreg32(struct task_struct *child, unsigned regno, u32 value)
979 {
980         struct pt_regs *regs = task_pt_regs(child);
981
982         switch (regno) {
983
984         SEG32(cs);
985         SEG32(ds);
986         SEG32(es);
987         SEG32(fs);
988         SEG32(gs);
989         SEG32(ss);
990
991         R32(ebx, bx);
992         R32(ecx, cx);
993         R32(edx, dx);
994         R32(edi, di);
995         R32(esi, si);
996         R32(ebp, bp);
997         R32(eax, ax);
998         R32(orig_eax, orig_ax);
999         R32(eip, ip);
1000         R32(esp, sp);
1001
1002         case offsetof(struct user32, regs.eflags):
1003                 return set_flags(child, value);
1004
1005         case offsetof(struct user32, u_debugreg[0]) ...
1006                 offsetof(struct user32, u_debugreg[7]):
1007                 regno -= offsetof(struct user32, u_debugreg[0]);
1008                 return ptrace_set_debugreg(child, regno / 4, value);
1009
1010         default:
1011                 if (regno > sizeof(struct user32) || (regno & 3))
1012                         return -EIO;
1013
1014                 /*
1015                  * Other dummy fields in the virtual user structure
1016                  * are ignored
1017                  */
1018                 break;
1019         }
1020         return 0;
1021 }
1022
1023 #undef R32
1024 #undef SEG32
1025
1026 #define R32(l,q)                                                        \
1027         case offsetof(struct user32, regs.l):                           \
1028                 *val = regs->q; break
1029
1030 #define SEG32(rs)                                                       \
1031         case offsetof(struct user32, regs.rs):                          \
1032                 *val = get_segment_reg(child,                           \
1033                                        offsetof(struct user_regs_struct, rs)); \
1034                 break
1035
1036 static int getreg32(struct task_struct *child, unsigned regno, u32 *val)
1037 {
1038         struct pt_regs *regs = task_pt_regs(child);
1039
1040         switch (regno) {
1041
1042         SEG32(ds);
1043         SEG32(es);
1044         SEG32(fs);
1045         SEG32(gs);
1046
1047         R32(cs, cs);
1048         R32(ss, ss);
1049         R32(ebx, bx);
1050         R32(ecx, cx);
1051         R32(edx, dx);
1052         R32(edi, di);
1053         R32(esi, si);
1054         R32(ebp, bp);
1055         R32(eax, ax);
1056         R32(orig_eax, orig_ax);
1057         R32(eip, ip);
1058         R32(esp, sp);
1059
1060         case offsetof(struct user32, regs.eflags):
1061                 *val = get_flags(child);
1062                 break;
1063
1064         case offsetof(struct user32, u_debugreg[0]) ...
1065                 offsetof(struct user32, u_debugreg[7]):
1066                 regno -= offsetof(struct user32, u_debugreg[0]);
1067                 *val = ptrace_get_debugreg(child, regno / 4);
1068                 break;
1069
1070         default:
1071                 if (regno > sizeof(struct user32) || (regno & 3))
1072                         return -EIO;
1073
1074                 /*
1075                  * Other dummy fields in the virtual user structure
1076                  * are ignored
1077                  */
1078                 *val = 0;
1079                 break;
1080         }
1081         return 0;
1082 }
1083
1084 #undef R32
1085 #undef SEG32
1086
1087 static int genregs32_get(struct task_struct *target,
1088                          const struct user_regset *regset,
1089                          unsigned int pos, unsigned int count,
1090                          void *kbuf, void __user *ubuf)
1091 {
1092         if (kbuf) {
1093                 compat_ulong_t *k = kbuf;
1094                 while (count > 0) {
1095                         getreg32(target, pos, k++);
1096                         count -= sizeof(*k);
1097                         pos += sizeof(*k);
1098                 }
1099         } else {
1100                 compat_ulong_t __user *u = ubuf;
1101                 while (count > 0) {
1102                         compat_ulong_t word;
1103                         getreg32(target, pos, &word);
1104                         if (__put_user(word, u++))
1105                                 return -EFAULT;
1106                         count -= sizeof(*u);
1107                         pos += sizeof(*u);
1108                 }
1109         }
1110
1111         return 0;
1112 }
1113
1114 static int genregs32_set(struct task_struct *target,
1115                          const struct user_regset *regset,
1116                          unsigned int pos, unsigned int count,
1117                          const void *kbuf, const void __user *ubuf)
1118 {
1119         int ret = 0;
1120         if (kbuf) {
1121                 const compat_ulong_t *k = kbuf;
1122                 while (count > 0 && !ret) {
1123                         ret = putreg(target, pos, *k++);
1124                         count -= sizeof(*k);
1125                         pos += sizeof(*k);
1126                 }
1127         } else {
1128                 const compat_ulong_t __user *u = ubuf;
1129                 while (count > 0 && !ret) {
1130                         compat_ulong_t word;
1131                         ret = __get_user(word, u++);
1132                         if (ret)
1133                                 break;
1134                         ret = putreg(target, pos, word);
1135                         count -= sizeof(*u);
1136                         pos += sizeof(*u);
1137                 }
1138         }
1139         return ret;
1140 }
1141
1142 static long ptrace32_siginfo(unsigned request, u32 pid, u32 addr, u32 data)
1143 {
1144         siginfo_t __user *si = compat_alloc_user_space(sizeof(siginfo_t));
1145         compat_siginfo_t __user *si32 = compat_ptr(data);
1146         siginfo_t ssi;
1147         int ret;
1148
1149         if (request == PTRACE_SETSIGINFO) {
1150                 memset(&ssi, 0, sizeof(siginfo_t));
1151                 ret = copy_siginfo_from_user32(&ssi, si32);
1152                 if (ret)
1153                         return ret;
1154                 if (copy_to_user(si, &ssi, sizeof(siginfo_t)))
1155                         return -EFAULT;
1156         }
1157         ret = sys_ptrace(request, pid, addr, (unsigned long)si);
1158         if (ret)
1159                 return ret;
1160         if (request == PTRACE_GETSIGINFO) {
1161                 if (copy_from_user(&ssi, si, sizeof(siginfo_t)))
1162                         return -EFAULT;
1163                 ret = copy_siginfo_to_user32(si32, &ssi);
1164         }
1165         return ret;
1166 }
1167
1168 asmlinkage long sys32_ptrace(long request, u32 pid, u32 addr, u32 data)
1169 {
1170         struct task_struct *child;
1171         struct pt_regs *childregs;
1172         void __user *datap = compat_ptr(data);
1173         int ret;
1174         __u32 val;
1175
1176         switch (request) {
1177         case PTRACE_TRACEME:
1178         case PTRACE_ATTACH:
1179         case PTRACE_KILL:
1180         case PTRACE_CONT:
1181         case PTRACE_SINGLESTEP:
1182         case PTRACE_SINGLEBLOCK:
1183         case PTRACE_DETACH:
1184         case PTRACE_SYSCALL:
1185         case PTRACE_OLDSETOPTIONS:
1186         case PTRACE_SETOPTIONS:
1187         case PTRACE_SET_THREAD_AREA:
1188         case PTRACE_GET_THREAD_AREA:
1189         case PTRACE_BTS_CONFIG:
1190         case PTRACE_BTS_STATUS:
1191         case PTRACE_BTS_SIZE:
1192         case PTRACE_BTS_GET:
1193         case PTRACE_BTS_CLEAR:
1194         case PTRACE_BTS_DRAIN:
1195                 return sys_ptrace(request, pid, addr, data);
1196
1197         default:
1198                 return -EINVAL;
1199
1200         case PTRACE_PEEKTEXT:
1201         case PTRACE_PEEKDATA:
1202         case PTRACE_POKEDATA:
1203         case PTRACE_POKETEXT:
1204         case PTRACE_POKEUSR:
1205         case PTRACE_PEEKUSR:
1206         case PTRACE_GETREGS:
1207         case PTRACE_SETREGS:
1208         case PTRACE_SETFPREGS:
1209         case PTRACE_GETFPREGS:
1210         case PTRACE_SETFPXREGS:
1211         case PTRACE_GETFPXREGS:
1212         case PTRACE_GETEVENTMSG:
1213                 break;
1214
1215         case PTRACE_SETSIGINFO:
1216         case PTRACE_GETSIGINFO:
1217                 return ptrace32_siginfo(request, pid, addr, data);
1218         }
1219
1220         child = ptrace_get_task_struct(pid);
1221         if (IS_ERR(child))
1222                 return PTR_ERR(child);
1223
1224         ret = ptrace_check_attach(child, request == PTRACE_KILL);
1225         if (ret < 0)
1226                 goto out;
1227
1228         childregs = task_pt_regs(child);
1229
1230         switch (request) {
1231         case PTRACE_PEEKUSR:
1232                 ret = getreg32(child, addr, &val);
1233                 if (ret == 0)
1234                         ret = put_user(val, (__u32 __user *)datap);
1235                 break;
1236
1237         case PTRACE_POKEUSR:
1238                 ret = putreg32(child, addr, data);
1239                 break;
1240
1241         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
1242                 return copy_regset_to_user(child, &user_x86_32_view,
1243                                            REGSET_GENERAL,
1244                                            0, sizeof(struct user_regs_struct32),
1245                                            datap);
1246
1247         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
1248                 return copy_regset_from_user(child, &user_x86_32_view,
1249                                              REGSET_GENERAL, 0,
1250                                              sizeof(struct user_regs_struct32),
1251                                              datap);
1252
1253         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
1254                 return copy_regset_to_user(child, &user_x86_32_view,
1255                                            REGSET_FP, 0,
1256                                            sizeof(struct user_i387_ia32_struct),
1257                                            datap);
1258
1259         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
1260                 return copy_regset_from_user(
1261                         child, &user_x86_32_view, REGSET_FP,
1262                         0, sizeof(struct user_i387_ia32_struct), datap);
1263
1264         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
1265                 return copy_regset_to_user(child, &user_x86_32_view,
1266                                            REGSET_XFP, 0,
1267                                            sizeof(struct user32_fxsr_struct),
1268                                            datap);
1269
1270         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
1271                 return copy_regset_from_user(child, &user_x86_32_view,
1272                                              REGSET_XFP, 0,
1273                                              sizeof(struct user32_fxsr_struct),
1274                                              datap);
1275
1276         default:
1277                 return compat_ptrace_request(child, request, addr, data);
1278         }
1279
1280  out:
1281         put_task_struct(child);
1282         return ret;
1283 }
1284
1285 #endif  /* CONFIG_IA32_EMULATION */
1286
1287 #ifdef CONFIG_X86_64
1288
1289 static const struct user_regset x86_64_regsets[] = {
1290         [REGSET_GENERAL] = {
1291                 .core_note_type = NT_PRSTATUS,
1292                 .n = sizeof(struct user_regs_struct) / sizeof(long),
1293                 .size = sizeof(long), .align = sizeof(long),
1294                 .get = genregs_get, .set = genregs_set
1295         },
1296         [REGSET_FP] = {
1297                 .core_note_type = NT_PRFPREG,
1298                 .n = sizeof(struct user_i387_struct) / sizeof(long),
1299                 .size = sizeof(long), .align = sizeof(long),
1300                 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1301         },
1302 };
1303
1304 static const struct user_regset_view user_x86_64_view = {
1305         .name = "x86_64", .e_machine = EM_X86_64,
1306         .regsets = x86_64_regsets, .n = ARRAY_SIZE(x86_64_regsets)
1307 };
1308
1309 #else  /* CONFIG_X86_32 */
1310
1311 #define user_regs_struct32      user_regs_struct
1312 #define genregs32_get           genregs_get
1313 #define genregs32_set           genregs_set
1314
1315 #endif  /* CONFIG_X86_64 */
1316
1317 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1318 static const struct user_regset x86_32_regsets[] = {
1319         [REGSET_GENERAL] = {
1320                 .core_note_type = NT_PRSTATUS,
1321                 .n = sizeof(struct user_regs_struct32) / sizeof(u32),
1322                 .size = sizeof(u32), .align = sizeof(u32),
1323                 .get = genregs32_get, .set = genregs32_set
1324         },
1325         [REGSET_FP] = {
1326                 .core_note_type = NT_PRFPREG,
1327                 .n = sizeof(struct user_i387_struct) / sizeof(u32),
1328                 .size = sizeof(u32), .align = sizeof(u32),
1329                 .active = fpregs_active, .get = fpregs_get, .set = fpregs_set
1330         },
1331         [REGSET_XFP] = {
1332                 .core_note_type = NT_PRXFPREG,
1333                 .n = sizeof(struct user_i387_struct) / sizeof(u32),
1334                 .size = sizeof(u32), .align = sizeof(u32),
1335                 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1336         },
1337         [REGSET_TLS] = {
1338                 .core_note_type = NT_386_TLS,
1339                 .n = GDT_ENTRY_TLS_ENTRIES, .bias = GDT_ENTRY_TLS_MIN,
1340                 .size = sizeof(struct user_desc),
1341                 .align = sizeof(struct user_desc),
1342                 .active = regset_tls_active,
1343                 .get = regset_tls_get, .set = regset_tls_set
1344         },
1345 };
1346
1347 static const struct user_regset_view user_x86_32_view = {
1348         .name = "i386", .e_machine = EM_386,
1349         .regsets = x86_32_regsets, .n = ARRAY_SIZE(x86_32_regsets)
1350 };
1351 #endif
1352
1353 const struct user_regset_view *task_user_regset_view(struct task_struct *task)
1354 {
1355 #ifdef CONFIG_IA32_EMULATION
1356         if (test_tsk_thread_flag(task, TIF_IA32))
1357 #endif
1358 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1359                 return &user_x86_32_view;
1360 #endif
1361 #ifdef CONFIG_X86_64
1362         return &user_x86_64_view;
1363 #endif
1364 }
1365
1366 #ifdef CONFIG_X86_32
1367
1368 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs, int error_code)
1369 {
1370         struct siginfo info;
1371
1372         tsk->thread.trap_no = 1;
1373         tsk->thread.error_code = error_code;
1374
1375         memset(&info, 0, sizeof(info));
1376         info.si_signo = SIGTRAP;
1377         info.si_code = TRAP_BRKPT;
1378
1379         /* User-mode ip? */
1380         info.si_addr = user_mode_vm(regs) ? (void __user *) regs->ip : NULL;
1381
1382         /* Send us the fake SIGTRAP */
1383         force_sig_info(SIGTRAP, &info, tsk);
1384 }
1385
1386 /* notification of system call entry/exit
1387  * - triggered by current->work.syscall_trace
1388  */
1389 __attribute__((regparm(3)))
1390 int do_syscall_trace(struct pt_regs *regs, int entryexit)
1391 {
1392         int is_sysemu = test_thread_flag(TIF_SYSCALL_EMU);
1393         /*
1394          * With TIF_SYSCALL_EMU set we want to ignore TIF_SINGLESTEP for syscall
1395          * interception
1396          */
1397         int is_singlestep = !is_sysemu && test_thread_flag(TIF_SINGLESTEP);
1398         int ret = 0;
1399
1400         /* do the secure computing check first */
1401         if (!entryexit)
1402                 secure_computing(regs->orig_ax);
1403
1404         if (unlikely(current->audit_context)) {
1405                 if (entryexit)
1406                         audit_syscall_exit(AUDITSC_RESULT(regs->ax),
1407                                                 regs->ax);
1408                 /* Debug traps, when using PTRACE_SINGLESTEP, must be sent only
1409                  * on the syscall exit path. Normally, when TIF_SYSCALL_AUDIT is
1410                  * not used, entry.S will call us only on syscall exit, not
1411                  * entry; so when TIF_SYSCALL_AUDIT is used we must avoid
1412                  * calling send_sigtrap() on syscall entry.
1413                  *
1414                  * Note that when PTRACE_SYSEMU_SINGLESTEP is used,
1415                  * is_singlestep is false, despite his name, so we will still do
1416                  * the correct thing.
1417                  */
1418                 else if (is_singlestep)
1419                         goto out;
1420         }
1421
1422         if (!(current->ptrace & PT_PTRACED))
1423                 goto out;
1424
1425         /* If a process stops on the 1st tracepoint with SYSCALL_TRACE
1426          * and then is resumed with SYSEMU_SINGLESTEP, it will come in
1427          * here. We have to check this and return */
1428         if (is_sysemu && entryexit)
1429                 return 0;
1430
1431         /* Fake a debug trap */
1432         if (is_singlestep)
1433                 send_sigtrap(current, regs, 0);
1434
1435         if (!test_thread_flag(TIF_SYSCALL_TRACE) && !is_sysemu)
1436                 goto out;
1437
1438         /* the 0x80 provides a way for the tracing parent to distinguish
1439            between a syscall stop and SIGTRAP delivery */
1440         /* Note that the debugger could change the result of test_thread_flag!*/
1441         ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD) ? 0x80:0));
1442
1443         /*
1444          * this isn't the same as continuing with a signal, but it will do
1445          * for normal use.  strace only continues with a signal if the
1446          * stopping signal is not SIGTRAP.  -brl
1447          */
1448         if (current->exit_code) {
1449                 send_sig(current->exit_code, current, 1);
1450                 current->exit_code = 0;
1451         }
1452         ret = is_sysemu;
1453 out:
1454         if (unlikely(current->audit_context) && !entryexit)
1455                 audit_syscall_entry(AUDIT_ARCH_I386, regs->orig_ax,
1456                                     regs->bx, regs->cx, regs->dx, regs->si);
1457         if (ret == 0)
1458                 return 0;
1459
1460         regs->orig_ax = -1; /* force skip of syscall restarting */
1461         if (unlikely(current->audit_context))
1462                 audit_syscall_exit(AUDITSC_RESULT(regs->ax), regs->ax);
1463         return 1;
1464 }
1465
1466 #else  /* CONFIG_X86_64 */
1467
1468 static void syscall_trace(struct pt_regs *regs)
1469 {
1470
1471 #if 0
1472         printk("trace %s ip %lx sp %lx ax %d origrax %d caller %lx tiflags %x ptrace %x\n",
1473                current->comm,
1474                regs->ip, regs->sp, regs->ax, regs->orig_ax, __builtin_return_address(0),
1475                current_thread_info()->flags, current->ptrace);
1476 #endif
1477
1478         ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD)
1479                                 ? 0x80 : 0));
1480         /*
1481          * this isn't the same as continuing with a signal, but it will do
1482          * for normal use.  strace only continues with a signal if the
1483          * stopping signal is not SIGTRAP.  -brl
1484          */
1485         if (current->exit_code) {
1486                 send_sig(current->exit_code, current, 1);
1487                 current->exit_code = 0;
1488         }
1489 }
1490
1491 asmlinkage void syscall_trace_enter(struct pt_regs *regs)
1492 {
1493         /* do the secure computing check first */
1494         secure_computing(regs->orig_ax);
1495
1496         if (test_thread_flag(TIF_SYSCALL_TRACE)
1497             && (current->ptrace & PT_PTRACED))
1498                 syscall_trace(regs);
1499
1500         if (unlikely(current->audit_context)) {
1501                 if (test_thread_flag(TIF_IA32)) {
1502                         audit_syscall_entry(AUDIT_ARCH_I386,
1503                                             regs->orig_ax,
1504                                             regs->bx, regs->cx,
1505                                             regs->dx, regs->si);
1506                 } else {
1507                         audit_syscall_entry(AUDIT_ARCH_X86_64,
1508                                             regs->orig_ax,
1509                                             regs->di, regs->si,
1510                                             regs->dx, regs->r10);
1511                 }
1512         }
1513 }
1514
1515 asmlinkage void syscall_trace_leave(struct pt_regs *regs)
1516 {
1517         if (unlikely(current->audit_context))
1518                 audit_syscall_exit(AUDITSC_RESULT(regs->ax), regs->ax);
1519
1520         if ((test_thread_flag(TIF_SYSCALL_TRACE)
1521              || test_thread_flag(TIF_SINGLESTEP))
1522             && (current->ptrace & PT_PTRACED))
1523                 syscall_trace(regs);
1524 }
1525
1526 #endif  /* CONFIG_X86_32 */