1 /* By Ross Biro 1/23/92 */
3 * Pentium III FXSR, SSE support
4 * Gareth Hughes <gareth@valinux.com>, May 2000
7 * Markus Metzger <markus.t.metzger@intel.com>, Dec 2007
10 #include <linux/kernel.h>
11 #include <linux/sched.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>
25 #include <asm/uaccess.h>
26 #include <asm/pgtable.h>
27 #include <asm/system.h>
28 #include <asm/processor.h>
30 #include <asm/debugreg.h>
33 #include <asm/prctl.h>
34 #include <asm/proto.h>
43 REGSET_IOPERM64
= REGSET_XFP
,
49 * does not yet catch signals sent when the child dies.
50 * in exit.c or in signal.c.
54 * Determines which flags the user has access to [1 = access, 0 = no access].
56 #define FLAG_MASK_32 ((unsigned long) \
57 (X86_EFLAGS_CF | X86_EFLAGS_PF | \
58 X86_EFLAGS_AF | X86_EFLAGS_ZF | \
59 X86_EFLAGS_SF | X86_EFLAGS_TF | \
60 X86_EFLAGS_DF | X86_EFLAGS_OF | \
61 X86_EFLAGS_RF | X86_EFLAGS_AC))
64 * Determines whether a value may be installed in a segment register.
66 static inline bool invalid_selector(u16 value
)
68 return unlikely(value
!= 0 && (value
& SEGMENT_RPL_MASK
) != USER_RPL
);
73 #define FLAG_MASK FLAG_MASK_32
75 static unsigned long *pt_regs_access(struct pt_regs
*regs
, unsigned long regno
)
77 BUILD_BUG_ON(offsetof(struct pt_regs
, bx
) != 0);
81 return ®s
->bx
+ regno
;
84 static u16
get_segment_reg(struct task_struct
*task
, unsigned long offset
)
87 * Returning the value truncates it to 16 bits.
90 if (offset
!= offsetof(struct user_regs_struct
, gs
))
91 retval
= *pt_regs_access(task_pt_regs(task
), offset
);
93 retval
= task
->thread
.gs
;
95 savesegment(gs
, retval
);
100 static int set_segment_reg(struct task_struct
*task
,
101 unsigned long offset
, u16 value
)
104 * The value argument was already truncated to 16 bits.
106 if (invalid_selector(value
))
110 * For %cs and %ss we cannot permit a null selector.
111 * We can permit a bogus selector as long as it has USER_RPL.
112 * Null selectors are fine for other segment registers, but
113 * we will never get back to user mode with invalid %cs or %ss
114 * and will take the trap in iret instead. Much code relies
115 * on user_mode() to distinguish a user trap frame (which can
116 * safely use invalid selectors) from a kernel trap frame.
119 case offsetof(struct user_regs_struct
, cs
):
120 case offsetof(struct user_regs_struct
, ss
):
121 if (unlikely(value
== 0))
125 *pt_regs_access(task_pt_regs(task
), offset
) = value
;
128 case offsetof(struct user_regs_struct
, gs
):
129 task
->thread
.gs
= value
;
132 * The user-mode %gs is not affected by
133 * kernel entry, so we must update the CPU.
135 loadsegment(gs
, value
);
141 static unsigned long debugreg_addr_limit(struct task_struct
*task
)
143 return TASK_SIZE
- 3;
146 #else /* CONFIG_X86_64 */
148 #define FLAG_MASK (FLAG_MASK_32 | X86_EFLAGS_NT)
150 static unsigned long *pt_regs_access(struct pt_regs
*regs
, unsigned long offset
)
152 BUILD_BUG_ON(offsetof(struct pt_regs
, r15
) != 0);
153 return ®s
->r15
+ (offset
/ sizeof(regs
->r15
));
156 static u16
get_segment_reg(struct task_struct
*task
, unsigned long offset
)
159 * Returning the value truncates it to 16 bits.
164 case offsetof(struct user_regs_struct
, fs
):
165 if (task
== current
) {
166 /* Older gas can't assemble movq %?s,%r?? */
167 asm("movl %%fs,%0" : "=r" (seg
));
170 return task
->thread
.fsindex
;
171 case offsetof(struct user_regs_struct
, gs
):
172 if (task
== current
) {
173 asm("movl %%gs,%0" : "=r" (seg
));
176 return task
->thread
.gsindex
;
177 case offsetof(struct user_regs_struct
, ds
):
178 if (task
== current
) {
179 asm("movl %%ds,%0" : "=r" (seg
));
182 return task
->thread
.ds
;
183 case offsetof(struct user_regs_struct
, es
):
184 if (task
== current
) {
185 asm("movl %%es,%0" : "=r" (seg
));
188 return task
->thread
.es
;
190 case offsetof(struct user_regs_struct
, cs
):
191 case offsetof(struct user_regs_struct
, ss
):
194 return *pt_regs_access(task_pt_regs(task
), offset
);
197 static int set_segment_reg(struct task_struct
*task
,
198 unsigned long offset
, u16 value
)
201 * The value argument was already truncated to 16 bits.
203 if (invalid_selector(value
))
207 case offsetof(struct user_regs_struct
,fs
):
209 * If this is setting fs as for normal 64-bit use but
210 * setting fs_base has implicitly changed it, leave it.
212 if ((value
== FS_TLS_SEL
&& task
->thread
.fsindex
== 0 &&
213 task
->thread
.fs
!= 0) ||
214 (value
== 0 && task
->thread
.fsindex
== FS_TLS_SEL
&&
215 task
->thread
.fs
== 0))
217 task
->thread
.fsindex
= value
;
219 loadsegment(fs
, task
->thread
.fsindex
);
221 case offsetof(struct user_regs_struct
,gs
):
223 * If this is setting gs as for normal 64-bit use but
224 * setting gs_base has implicitly changed it, leave it.
226 if ((value
== GS_TLS_SEL
&& task
->thread
.gsindex
== 0 &&
227 task
->thread
.gs
!= 0) ||
228 (value
== 0 && task
->thread
.gsindex
== GS_TLS_SEL
&&
229 task
->thread
.gs
== 0))
231 task
->thread
.gsindex
= value
;
233 load_gs_index(task
->thread
.gsindex
);
235 case offsetof(struct user_regs_struct
,ds
):
236 task
->thread
.ds
= value
;
238 loadsegment(ds
, task
->thread
.ds
);
240 case offsetof(struct user_regs_struct
,es
):
241 task
->thread
.es
= value
;
243 loadsegment(es
, task
->thread
.es
);
247 * Can't actually change these in 64-bit mode.
249 case offsetof(struct user_regs_struct
,cs
):
250 if (unlikely(value
== 0))
252 #ifdef CONFIG_IA32_EMULATION
253 if (test_tsk_thread_flag(task
, TIF_IA32
))
254 task_pt_regs(task
)->cs
= value
;
257 case offsetof(struct user_regs_struct
,ss
):
258 if (unlikely(value
== 0))
260 #ifdef CONFIG_IA32_EMULATION
261 if (test_tsk_thread_flag(task
, TIF_IA32
))
262 task_pt_regs(task
)->ss
= value
;
270 static unsigned long debugreg_addr_limit(struct task_struct
*task
)
272 #ifdef CONFIG_IA32_EMULATION
273 if (test_tsk_thread_flag(task
, TIF_IA32
))
274 return IA32_PAGE_OFFSET
- 3;
276 return TASK_SIZE64
- 7;
279 #endif /* CONFIG_X86_32 */
281 static unsigned long get_flags(struct task_struct
*task
)
283 unsigned long retval
= task_pt_regs(task
)->flags
;
286 * If the debugger set TF, hide it from the readout.
288 if (test_tsk_thread_flag(task
, TIF_FORCED_TF
))
289 retval
&= ~X86_EFLAGS_TF
;
294 static int set_flags(struct task_struct
*task
, unsigned long value
)
296 struct pt_regs
*regs
= task_pt_regs(task
);
299 * If the user value contains TF, mark that
300 * it was not "us" (the debugger) that set it.
301 * If not, make sure it stays set if we had.
303 if (value
& X86_EFLAGS_TF
)
304 clear_tsk_thread_flag(task
, TIF_FORCED_TF
);
305 else if (test_tsk_thread_flag(task
, TIF_FORCED_TF
))
306 value
|= X86_EFLAGS_TF
;
308 regs
->flags
= (regs
->flags
& ~FLAG_MASK
) | (value
& FLAG_MASK
);
313 static int putreg(struct task_struct
*child
,
314 unsigned long offset
, unsigned long value
)
317 case offsetof(struct user_regs_struct
, cs
):
318 case offsetof(struct user_regs_struct
, ds
):
319 case offsetof(struct user_regs_struct
, es
):
320 case offsetof(struct user_regs_struct
, fs
):
321 case offsetof(struct user_regs_struct
, gs
):
322 case offsetof(struct user_regs_struct
, ss
):
323 return set_segment_reg(child
, offset
, value
);
325 case offsetof(struct user_regs_struct
, flags
):
326 return set_flags(child
, value
);
330 * Orig_ax is really just a flag with small positive and
331 * negative values, so make sure to always sign-extend it
332 * from 32 bits so that it works correctly regardless of
333 * whether we come from a 32-bit environment or not.
335 case offsetof(struct user_regs_struct
, orig_ax
):
336 value
= (long) (s32
) value
;
339 case offsetof(struct user_regs_struct
,fs_base
):
340 if (value
>= TASK_SIZE_OF(child
))
343 * When changing the segment base, use do_arch_prctl
344 * to set either thread.fs or thread.fsindex and the
345 * corresponding GDT slot.
347 if (child
->thread
.fs
!= value
)
348 return do_arch_prctl(child
, ARCH_SET_FS
, value
);
350 case offsetof(struct user_regs_struct
,gs_base
):
352 * Exactly the same here as the %fs handling above.
354 if (value
>= TASK_SIZE_OF(child
))
356 if (child
->thread
.gs
!= value
)
357 return do_arch_prctl(child
, ARCH_SET_GS
, value
);
362 *pt_regs_access(task_pt_regs(child
), offset
) = value
;
366 static unsigned long getreg(struct task_struct
*task
, unsigned long offset
)
369 case offsetof(struct user_regs_struct
, cs
):
370 case offsetof(struct user_regs_struct
, ds
):
371 case offsetof(struct user_regs_struct
, es
):
372 case offsetof(struct user_regs_struct
, fs
):
373 case offsetof(struct user_regs_struct
, gs
):
374 case offsetof(struct user_regs_struct
, ss
):
375 return get_segment_reg(task
, offset
);
377 case offsetof(struct user_regs_struct
, flags
):
378 return get_flags(task
);
381 case offsetof(struct user_regs_struct
, fs_base
): {
383 * do_arch_prctl may have used a GDT slot instead of
384 * the MSR. To userland, it appears the same either
385 * way, except the %fs segment selector might not be 0.
387 unsigned int seg
= task
->thread
.fsindex
;
388 if (task
->thread
.fs
!= 0)
389 return task
->thread
.fs
;
391 asm("movl %%fs,%0" : "=r" (seg
));
392 if (seg
!= FS_TLS_SEL
)
394 return get_desc_base(&task
->thread
.tls_array
[FS_TLS
]);
396 case offsetof(struct user_regs_struct
, gs_base
): {
398 * Exactly the same here as the %fs handling above.
400 unsigned int seg
= task
->thread
.gsindex
;
401 if (task
->thread
.gs
!= 0)
402 return task
->thread
.gs
;
404 asm("movl %%gs,%0" : "=r" (seg
));
405 if (seg
!= GS_TLS_SEL
)
407 return get_desc_base(&task
->thread
.tls_array
[GS_TLS
]);
412 return *pt_regs_access(task_pt_regs(task
), offset
);
415 static int genregs_get(struct task_struct
*target
,
416 const struct user_regset
*regset
,
417 unsigned int pos
, unsigned int count
,
418 void *kbuf
, void __user
*ubuf
)
421 unsigned long *k
= kbuf
;
423 *k
++ = getreg(target
, pos
);
428 unsigned long __user
*u
= ubuf
;
430 if (__put_user(getreg(target
, pos
), u
++))
440 static int genregs_set(struct task_struct
*target
,
441 const struct user_regset
*regset
,
442 unsigned int pos
, unsigned int count
,
443 const void *kbuf
, const void __user
*ubuf
)
447 const unsigned long *k
= kbuf
;
448 while (count
> 0 && !ret
) {
449 ret
= putreg(target
, pos
, *k
++);
454 const unsigned long __user
*u
= ubuf
;
455 while (count
> 0 && !ret
) {
457 ret
= __get_user(word
, u
++);
460 ret
= putreg(target
, pos
, word
);
469 * This function is trivial and will be inlined by the compiler.
470 * Having it separates the implementation details of debug
471 * registers from the interface details of ptrace.
473 static unsigned long ptrace_get_debugreg(struct task_struct
*child
, int n
)
476 case 0: return child
->thread
.debugreg0
;
477 case 1: return child
->thread
.debugreg1
;
478 case 2: return child
->thread
.debugreg2
;
479 case 3: return child
->thread
.debugreg3
;
480 case 6: return child
->thread
.debugreg6
;
481 case 7: return child
->thread
.debugreg7
;
486 static int ptrace_set_debugreg(struct task_struct
*child
,
487 int n
, unsigned long data
)
491 if (unlikely(n
== 4 || n
== 5))
494 if (n
< 4 && unlikely(data
>= debugreg_addr_limit(child
)))
498 case 0: child
->thread
.debugreg0
= data
; break;
499 case 1: child
->thread
.debugreg1
= data
; break;
500 case 2: child
->thread
.debugreg2
= data
; break;
501 case 3: child
->thread
.debugreg3
= data
; break;
504 if ((data
& ~0xffffffffUL
) != 0)
506 child
->thread
.debugreg6
= data
;
511 * Sanity-check data. Take one half-byte at once with
512 * check = (val >> (16 + 4*i)) & 0xf. It contains the
513 * R/Wi and LENi bits; bits 0 and 1 are R/Wi, and bits
514 * 2 and 3 are LENi. Given a list of invalid values,
515 * we do mask |= 1 << invalid_value, so that
516 * (mask >> check) & 1 is a correct test for invalid
519 * R/Wi contains the type of the breakpoint /
520 * watchpoint, LENi contains the length of the watched
521 * data in the watchpoint case.
523 * The invalid values are:
524 * - LENi == 0x10 (undefined), so mask |= 0x0f00. [32-bit]
525 * - R/Wi == 0x10 (break on I/O reads or writes), so
527 * - R/Wi == 0x00 && LENi != 0x00, so we have mask |=
530 * Finally, mask = 0x0f00 | 0x4444 | 0x1110 == 0x5f54.
532 * See the Intel Manual "System Programming Guide",
535 * Note that LENi == 0x10 is defined on x86_64 in long
536 * mode (i.e. even for 32-bit userspace software, but
537 * 64-bit kernel), so the x86_64 mask value is 0x5454.
538 * See the AMD manual no. 24593 (AMD64 System Programming)
541 #define DR7_MASK 0x5f54
543 #define DR7_MASK 0x5554
545 data
&= ~DR_CONTROL_RESERVED
;
546 for (i
= 0; i
< 4; i
++)
547 if ((DR7_MASK
>> ((data
>> (16 + 4*i
)) & 0xf)) & 1)
549 child
->thread
.debugreg7
= data
;
551 set_tsk_thread_flag(child
, TIF_DEBUG
);
553 clear_tsk_thread_flag(child
, TIF_DEBUG
);
561 * These access the current or another (stopped) task's io permission
562 * bitmap for debugging or core dump.
564 static int ioperm_active(struct task_struct
*target
,
565 const struct user_regset
*regset
)
567 return target
->thread
.io_bitmap_max
/ regset
->size
;
570 static int ioperm_get(struct task_struct
*target
,
571 const struct user_regset
*regset
,
572 unsigned int pos
, unsigned int count
,
573 void *kbuf
, void __user
*ubuf
)
575 if (!target
->thread
.io_bitmap_ptr
)
578 return user_regset_copyout(&pos
, &count
, &kbuf
, &ubuf
,
579 target
->thread
.io_bitmap_ptr
,
583 #ifdef CONFIG_X86_PTRACE_BTS
585 * The configuration for a particular BTS hardware implementation.
587 struct bts_configuration
{
588 /* the size of a BTS record in bytes; at most BTS_MAX_RECORD_SIZE */
589 unsigned char sizeof_bts
;
590 /* the size of a field in the BTS record in bytes */
591 unsigned char sizeof_field
;
592 /* a bitmask to enable/disable BTS in DEBUGCTL MSR */
593 unsigned long debugctl_mask
;
595 static struct bts_configuration bts_cfg
;
597 #define BTS_MAX_RECORD_SIZE (8 * 3)
601 * Branch Trace Store (BTS) uses the following format. Different
602 * architectures vary in the size of those fields.
603 * - source linear address
604 * - destination linear address
607 * Later architectures use 64bit pointers throughout, whereas earlier
608 * architectures use 32bit pointers in 32bit mode.
610 * We compute the base address for the first 8 fields based on:
611 * - the field size stored in the DS configuration
612 * - the relative field position
614 * In order to store additional information in the BTS buffer, we use
615 * a special source address to indicate that the record requires
616 * special interpretation.
618 * Netburst indicated via a bit in the flags field whether the branch
619 * was predicted; this is ignored.
627 bts_escape
= (unsigned long)-1,
629 bts_jiffies
= bts_flags
632 static inline unsigned long bts_get(const char *base
, enum bts_field field
)
634 base
+= (bts_cfg
.sizeof_field
* field
);
635 return *(unsigned long *)base
;
638 static inline void bts_set(char *base
, enum bts_field field
, unsigned long val
)
640 base
+= (bts_cfg
.sizeof_field
* field
);;
641 (*(unsigned long *)base
) = val
;
645 * Translate a BTS record from the raw format into the bts_struct format
647 * out (out): bts_struct interpretation
648 * raw: raw BTS record
650 static void ptrace_bts_translate_record(struct bts_struct
*out
, const void *raw
)
652 memset(out
, 0, sizeof(*out
));
653 if (bts_get(raw
, bts_from
) == bts_escape
) {
654 out
->qualifier
= bts_get(raw
, bts_qual
);
655 out
->variant
.jiffies
= bts_get(raw
, bts_jiffies
);
657 out
->qualifier
= BTS_BRANCH
;
658 out
->variant
.lbr
.from_ip
= bts_get(raw
, bts_from
);
659 out
->variant
.lbr
.to_ip
= bts_get(raw
, bts_to
);
663 static int ptrace_bts_read_record(struct task_struct
*child
, size_t index
,
664 struct bts_struct __user
*out
)
666 struct bts_struct ret
;
667 const void *bts_record
;
668 size_t bts_index
, bts_end
;
671 error
= ds_get_bts_end(child
, &bts_end
);
675 if (bts_end
<= index
)
678 error
= ds_get_bts_index(child
, &bts_index
);
682 /* translate the ptrace bts index into the ds bts index */
683 bts_index
+= bts_end
- (index
+ 1);
684 if (bts_end
<= bts_index
)
685 bts_index
-= bts_end
;
687 error
= ds_access_bts(child
, bts_index
, &bts_record
);
691 ptrace_bts_translate_record(&ret
, bts_record
);
693 if (copy_to_user(out
, &ret
, sizeof(ret
)))
699 static int ptrace_bts_drain(struct task_struct
*child
,
701 struct bts_struct __user
*out
)
703 struct bts_struct ret
;
704 const unsigned char *raw
;
708 error
= ds_get_bts_index(child
, &end
);
712 if (size
< (end
* sizeof(struct bts_struct
)))
715 error
= ds_access_bts(child
, 0, (const void **)&raw
);
719 for (i
= 0; i
< end
; i
++, out
++, raw
+= bts_cfg
.sizeof_bts
) {
720 ptrace_bts_translate_record(&ret
, raw
);
722 if (copy_to_user(out
, &ret
, sizeof(ret
)))
726 error
= ds_clear_bts(child
);
733 static void ptrace_bts_ovfl(struct task_struct
*child
)
735 send_sig(child
->thread
.bts_ovfl_signal
, child
, 0);
738 static int ptrace_bts_config(struct task_struct
*child
,
740 const struct ptrace_bts_config __user
*ucfg
)
742 struct ptrace_bts_config cfg
;
746 if (!bts_cfg
.sizeof_bts
)
750 if (cfg_size
< sizeof(cfg
))
754 if (copy_from_user(&cfg
, ucfg
, sizeof(cfg
)))
758 if ((cfg
.flags
& PTRACE_BTS_O_SIGNAL
) &&
759 !(cfg
.flags
& PTRACE_BTS_O_ALLOC
))
762 if (cfg
.flags
& PTRACE_BTS_O_ALLOC
) {
763 ds_ovfl_callback_t ovfl
= NULL
;
764 unsigned int sig
= 0;
766 /* we ignore the error in case we were not tracing child */
767 (void)ds_release_bts(child
);
769 if (cfg
.flags
& PTRACE_BTS_O_SIGNAL
) {
774 ovfl
= ptrace_bts_ovfl
;
777 error
= ds_request_bts(child
, /* base = */ NULL
, cfg
.size
, ovfl
);
781 child
->thread
.bts_ovfl_signal
= sig
;
785 if (!child
->thread
.ds_ctx
&& cfg
.flags
)
788 if (cfg
.flags
& PTRACE_BTS_O_TRACE
)
789 child
->thread
.debugctlmsr
|= bts_cfg
.debugctl_mask
;
791 child
->thread
.debugctlmsr
&= ~bts_cfg
.debugctl_mask
;
793 if (cfg
.flags
& PTRACE_BTS_O_SCHED
)
794 set_tsk_thread_flag(child
, TIF_BTS_TRACE_TS
);
796 clear_tsk_thread_flag(child
, TIF_BTS_TRACE_TS
);
801 if (child
->thread
.debugctlmsr
)
802 set_tsk_thread_flag(child
, TIF_DEBUGCTLMSR
);
804 clear_tsk_thread_flag(child
, TIF_DEBUGCTLMSR
);
809 child
->thread
.debugctlmsr
&= ~bts_cfg
.debugctl_mask
;
810 clear_tsk_thread_flag(child
, TIF_BTS_TRACE_TS
);
814 static int ptrace_bts_status(struct task_struct
*child
,
816 struct ptrace_bts_config __user
*ucfg
)
818 struct ptrace_bts_config cfg
;
820 const void *base
, *max
;
823 if (cfg_size
< sizeof(cfg
))
826 error
= ds_get_bts_end(child
, &end
);
830 error
= ds_access_bts(child
, /* index = */ 0, &base
);
834 error
= ds_access_bts(child
, /* index = */ end
, &max
);
838 memset(&cfg
, 0, sizeof(cfg
));
839 cfg
.size
= (max
- base
);
840 cfg
.signal
= child
->thread
.bts_ovfl_signal
;
841 cfg
.bts_size
= sizeof(struct bts_struct
);
844 cfg
.flags
|= PTRACE_BTS_O_SIGNAL
;
846 if (test_tsk_thread_flag(child
, TIF_DEBUGCTLMSR
) &&
847 child
->thread
.debugctlmsr
& bts_cfg
.debugctl_mask
)
848 cfg
.flags
|= PTRACE_BTS_O_TRACE
;
850 if (test_tsk_thread_flag(child
, TIF_BTS_TRACE_TS
))
851 cfg
.flags
|= PTRACE_BTS_O_SCHED
;
853 if (copy_to_user(ucfg
, &cfg
, sizeof(cfg
)))
859 static int ptrace_bts_write_record(struct task_struct
*child
,
860 const struct bts_struct
*in
)
862 unsigned char bts_record
[BTS_MAX_RECORD_SIZE
];
864 BUG_ON(BTS_MAX_RECORD_SIZE
< bts_cfg
.sizeof_bts
);
866 memset(bts_record
, 0, bts_cfg
.sizeof_bts
);
867 switch (in
->qualifier
) {
872 bts_set(bts_record
, bts_from
, in
->variant
.lbr
.from_ip
);
873 bts_set(bts_record
, bts_to
, in
->variant
.lbr
.to_ip
);
876 case BTS_TASK_ARRIVES
:
877 case BTS_TASK_DEPARTS
:
878 bts_set(bts_record
, bts_from
, bts_escape
);
879 bts_set(bts_record
, bts_qual
, in
->qualifier
);
880 bts_set(bts_record
, bts_jiffies
, in
->variant
.jiffies
);
887 /* The writing task will be the switched-to task on a context
888 * switch. It needs to write into the switched-from task's BTS
890 return ds_unchecked_write_bts(child
, bts_record
, bts_cfg
.sizeof_bts
);
893 void ptrace_bts_take_timestamp(struct task_struct
*tsk
,
894 enum bts_qualifier qualifier
)
896 struct bts_struct rec
= {
897 .qualifier
= qualifier
,
898 .variant
.jiffies
= jiffies_64
901 ptrace_bts_write_record(tsk
, &rec
);
904 static const struct bts_configuration bts_cfg_netburst
= {
905 .sizeof_bts
= sizeof(long) * 3,
906 .sizeof_field
= sizeof(long),
907 .debugctl_mask
= (1<<2)|(1<<3)|(1<<5)
910 static const struct bts_configuration bts_cfg_pentium_m
= {
911 .sizeof_bts
= sizeof(long) * 3,
912 .sizeof_field
= sizeof(long),
913 .debugctl_mask
= (1<<6)|(1<<7)
916 static const struct bts_configuration bts_cfg_core2
= {
919 .debugctl_mask
= (1<<6)|(1<<7)|(1<<9)
922 static inline void bts_configure(const struct bts_configuration
*cfg
)
927 void __cpuinit
ptrace_bts_init_intel(struct cpuinfo_x86
*c
)
931 switch (c
->x86_model
) {
933 case 0xE: /* Pentium M */
934 bts_configure(&bts_cfg_pentium_m
);
936 case 0xF: /* Core2 */
937 case 0x1C: /* Atom */
938 bts_configure(&bts_cfg_core2
);
941 /* sorry, don't know about them */
946 switch (c
->x86_model
) {
949 case 0x2: /* Netburst */
950 bts_configure(&bts_cfg_netburst
);
953 /* sorry, don't know about them */
958 /* sorry, don't know about them */
962 #endif /* CONFIG_X86_PTRACE_BTS */
965 * Called by kernel/ptrace.c when detaching..
967 * Make sure the single step bit is not set.
969 void ptrace_disable(struct task_struct
*child
)
971 user_disable_single_step(child
);
972 #ifdef TIF_SYSCALL_EMU
973 clear_tsk_thread_flag(child
, TIF_SYSCALL_EMU
);
975 #ifdef CONFIG_X86_PTRACE_BTS
976 (void)ds_release_bts(child
);
978 child
->thread
.debugctlmsr
&= ~bts_cfg
.debugctl_mask
;
979 if (!child
->thread
.debugctlmsr
)
980 clear_tsk_thread_flag(child
, TIF_DEBUGCTLMSR
);
982 clear_tsk_thread_flag(child
, TIF_BTS_TRACE_TS
);
983 #endif /* CONFIG_X86_PTRACE_BTS */
986 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
987 static const struct user_regset_view user_x86_32_view
; /* Initialized below. */
990 long arch_ptrace(struct task_struct
*child
, long request
, long addr
, long data
)
993 unsigned long __user
*datap
= (unsigned long __user
*)data
;
996 /* read the word at location addr in the USER area. */
997 case PTRACE_PEEKUSR
: {
1001 if ((addr
& (sizeof(data
) - 1)) || addr
< 0 ||
1002 addr
>= sizeof(struct user
))
1005 tmp
= 0; /* Default return condition */
1006 if (addr
< sizeof(struct user_regs_struct
))
1007 tmp
= getreg(child
, addr
);
1008 else if (addr
>= offsetof(struct user
, u_debugreg
[0]) &&
1009 addr
<= offsetof(struct user
, u_debugreg
[7])) {
1010 addr
-= offsetof(struct user
, u_debugreg
[0]);
1011 tmp
= ptrace_get_debugreg(child
, addr
/ sizeof(data
));
1013 ret
= put_user(tmp
, datap
);
1017 case PTRACE_POKEUSR
: /* write the word at location addr in the USER area */
1019 if ((addr
& (sizeof(data
) - 1)) || addr
< 0 ||
1020 addr
>= sizeof(struct user
))
1023 if (addr
< sizeof(struct user_regs_struct
))
1024 ret
= putreg(child
, addr
, data
);
1025 else if (addr
>= offsetof(struct user
, u_debugreg
[0]) &&
1026 addr
<= offsetof(struct user
, u_debugreg
[7])) {
1027 addr
-= offsetof(struct user
, u_debugreg
[0]);
1028 ret
= ptrace_set_debugreg(child
,
1029 addr
/ sizeof(data
), data
);
1033 case PTRACE_GETREGS
: /* Get all gp regs from the child. */
1034 return copy_regset_to_user(child
,
1035 task_user_regset_view(current
),
1037 0, sizeof(struct user_regs_struct
),
1040 case PTRACE_SETREGS
: /* Set all gp regs in the child. */
1041 return copy_regset_from_user(child
,
1042 task_user_regset_view(current
),
1044 0, sizeof(struct user_regs_struct
),
1047 case PTRACE_GETFPREGS
: /* Get the child FPU state. */
1048 return copy_regset_to_user(child
,
1049 task_user_regset_view(current
),
1051 0, sizeof(struct user_i387_struct
),
1054 case PTRACE_SETFPREGS
: /* Set the child FPU state. */
1055 return copy_regset_from_user(child
,
1056 task_user_regset_view(current
),
1058 0, sizeof(struct user_i387_struct
),
1061 #ifdef CONFIG_X86_32
1062 case PTRACE_GETFPXREGS
: /* Get the child extended FPU state. */
1063 return copy_regset_to_user(child
, &user_x86_32_view
,
1065 0, sizeof(struct user_fxsr_struct
),
1068 case PTRACE_SETFPXREGS
: /* Set the child extended FPU state. */
1069 return copy_regset_from_user(child
, &user_x86_32_view
,
1071 0, sizeof(struct user_fxsr_struct
),
1075 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1076 case PTRACE_GET_THREAD_AREA
:
1079 ret
= do_get_thread_area(child
, addr
,
1080 (struct user_desc __user
*) data
);
1083 case PTRACE_SET_THREAD_AREA
:
1086 ret
= do_set_thread_area(child
, addr
,
1087 (struct user_desc __user
*) data
, 0);
1091 #ifdef CONFIG_X86_64
1092 /* normal 64bit interface to access TLS data.
1093 Works just like arch_prctl, except that the arguments
1095 case PTRACE_ARCH_PRCTL
:
1096 ret
= do_arch_prctl(child
, data
, addr
);
1101 * These bits need more cooking - not enabled yet:
1103 #ifdef CONFIG_X86_PTRACE_BTS
1104 case PTRACE_BTS_CONFIG
:
1105 ret
= ptrace_bts_config
1106 (child
, data
, (struct ptrace_bts_config __user
*)addr
);
1109 case PTRACE_BTS_STATUS
:
1110 ret
= ptrace_bts_status
1111 (child
, data
, (struct ptrace_bts_config __user
*)addr
);
1114 case PTRACE_BTS_SIZE
:
1115 ret
= ds_get_bts_index(child
, /* pos = */ NULL
);
1118 case PTRACE_BTS_GET
:
1119 ret
= ptrace_bts_read_record
1120 (child
, data
, (struct bts_struct __user
*) addr
);
1123 case PTRACE_BTS_CLEAR
:
1124 ret
= ds_clear_bts(child
);
1127 case PTRACE_BTS_DRAIN
:
1128 ret
= ptrace_bts_drain
1129 (child
, data
, (struct bts_struct __user
*) addr
);
1131 #endif /* CONFIG_X86_PTRACE_BTS */
1134 ret
= ptrace_request(child
, request
, addr
, data
);
1141 #ifdef CONFIG_IA32_EMULATION
1143 #include <linux/compat.h>
1144 #include <linux/syscalls.h>
1145 #include <asm/ia32.h>
1146 #include <asm/user32.h>
1149 case offsetof(struct user32, regs.l): \
1150 regs->q = value; break
1153 case offsetof(struct user32, regs.rs): \
1154 return set_segment_reg(child, \
1155 offsetof(struct user_regs_struct, rs), \
1159 static int putreg32(struct task_struct
*child
, unsigned regno
, u32 value
)
1161 struct pt_regs
*regs
= task_pt_regs(child
);
1182 case offsetof(struct user32
, regs
.orig_eax
):
1184 * Sign-extend the value so that orig_eax = -1
1185 * causes (long)orig_ax < 0 tests to fire correctly.
1187 regs
->orig_ax
= (long) (s32
) value
;
1190 case offsetof(struct user32
, regs
.eflags
):
1191 return set_flags(child
, value
);
1193 case offsetof(struct user32
, u_debugreg
[0]) ...
1194 offsetof(struct user32
, u_debugreg
[7]):
1195 regno
-= offsetof(struct user32
, u_debugreg
[0]);
1196 return ptrace_set_debugreg(child
, regno
/ 4, value
);
1199 if (regno
> sizeof(struct user32
) || (regno
& 3))
1203 * Other dummy fields in the virtual user structure
1215 case offsetof(struct user32, regs.l): \
1216 *val = regs->q; break
1219 case offsetof(struct user32, regs.rs): \
1220 *val = get_segment_reg(child, \
1221 offsetof(struct user_regs_struct, rs)); \
1224 static int getreg32(struct task_struct
*child
, unsigned regno
, u32
*val
)
1226 struct pt_regs
*regs
= task_pt_regs(child
);
1244 R32(orig_eax
, orig_ax
);
1248 case offsetof(struct user32
, regs
.eflags
):
1249 *val
= get_flags(child
);
1252 case offsetof(struct user32
, u_debugreg
[0]) ...
1253 offsetof(struct user32
, u_debugreg
[7]):
1254 regno
-= offsetof(struct user32
, u_debugreg
[0]);
1255 *val
= ptrace_get_debugreg(child
, regno
/ 4);
1259 if (regno
> sizeof(struct user32
) || (regno
& 3))
1263 * Other dummy fields in the virtual user structure
1275 static int genregs32_get(struct task_struct
*target
,
1276 const struct user_regset
*regset
,
1277 unsigned int pos
, unsigned int count
,
1278 void *kbuf
, void __user
*ubuf
)
1281 compat_ulong_t
*k
= kbuf
;
1283 getreg32(target
, pos
, k
++);
1284 count
-= sizeof(*k
);
1288 compat_ulong_t __user
*u
= ubuf
;
1290 compat_ulong_t word
;
1291 getreg32(target
, pos
, &word
);
1292 if (__put_user(word
, u
++))
1294 count
-= sizeof(*u
);
1302 static int genregs32_set(struct task_struct
*target
,
1303 const struct user_regset
*regset
,
1304 unsigned int pos
, unsigned int count
,
1305 const void *kbuf
, const void __user
*ubuf
)
1309 const compat_ulong_t
*k
= kbuf
;
1310 while (count
> 0 && !ret
) {
1311 ret
= putreg32(target
, pos
, *k
++);
1312 count
-= sizeof(*k
);
1316 const compat_ulong_t __user
*u
= ubuf
;
1317 while (count
> 0 && !ret
) {
1318 compat_ulong_t word
;
1319 ret
= __get_user(word
, u
++);
1322 ret
= putreg32(target
, pos
, word
);
1323 count
-= sizeof(*u
);
1330 long compat_arch_ptrace(struct task_struct
*child
, compat_long_t request
,
1331 compat_ulong_t caddr
, compat_ulong_t cdata
)
1333 unsigned long addr
= caddr
;
1334 unsigned long data
= cdata
;
1335 void __user
*datap
= compat_ptr(data
);
1340 case PTRACE_PEEKUSR
:
1341 ret
= getreg32(child
, addr
, &val
);
1343 ret
= put_user(val
, (__u32 __user
*)datap
);
1346 case PTRACE_POKEUSR
:
1347 ret
= putreg32(child
, addr
, data
);
1350 case PTRACE_GETREGS
: /* Get all gp regs from the child. */
1351 return copy_regset_to_user(child
, &user_x86_32_view
,
1353 0, sizeof(struct user_regs_struct32
),
1356 case PTRACE_SETREGS
: /* Set all gp regs in the child. */
1357 return copy_regset_from_user(child
, &user_x86_32_view
,
1359 sizeof(struct user_regs_struct32
),
1362 case PTRACE_GETFPREGS
: /* Get the child FPU state. */
1363 return copy_regset_to_user(child
, &user_x86_32_view
,
1365 sizeof(struct user_i387_ia32_struct
),
1368 case PTRACE_SETFPREGS
: /* Set the child FPU state. */
1369 return copy_regset_from_user(
1370 child
, &user_x86_32_view
, REGSET_FP
,
1371 0, sizeof(struct user_i387_ia32_struct
), datap
);
1373 case PTRACE_GETFPXREGS
: /* Get the child extended FPU state. */
1374 return copy_regset_to_user(child
, &user_x86_32_view
,
1376 sizeof(struct user32_fxsr_struct
),
1379 case PTRACE_SETFPXREGS
: /* Set the child extended FPU state. */
1380 return copy_regset_from_user(child
, &user_x86_32_view
,
1382 sizeof(struct user32_fxsr_struct
),
1385 case PTRACE_GET_THREAD_AREA
:
1386 case PTRACE_SET_THREAD_AREA
:
1387 return arch_ptrace(child
, request
, addr
, data
);
1390 return compat_ptrace_request(child
, request
, addr
, data
);
1396 #endif /* CONFIG_IA32_EMULATION */
1398 #ifdef CONFIG_X86_64
1400 static const struct user_regset x86_64_regsets
[] = {
1401 [REGSET_GENERAL
] = {
1402 .core_note_type
= NT_PRSTATUS
,
1403 .n
= sizeof(struct user_regs_struct
) / sizeof(long),
1404 .size
= sizeof(long), .align
= sizeof(long),
1405 .get
= genregs_get
, .set
= genregs_set
1408 .core_note_type
= NT_PRFPREG
,
1409 .n
= sizeof(struct user_i387_struct
) / sizeof(long),
1410 .size
= sizeof(long), .align
= sizeof(long),
1411 .active
= xfpregs_active
, .get
= xfpregs_get
, .set
= xfpregs_set
1413 [REGSET_IOPERM64
] = {
1414 .core_note_type
= NT_386_IOPERM
,
1415 .n
= IO_BITMAP_LONGS
,
1416 .size
= sizeof(long), .align
= sizeof(long),
1417 .active
= ioperm_active
, .get
= ioperm_get
1421 static const struct user_regset_view user_x86_64_view
= {
1422 .name
= "x86_64", .e_machine
= EM_X86_64
,
1423 .regsets
= x86_64_regsets
, .n
= ARRAY_SIZE(x86_64_regsets
)
1426 #else /* CONFIG_X86_32 */
1428 #define user_regs_struct32 user_regs_struct
1429 #define genregs32_get genregs_get
1430 #define genregs32_set genregs_set
1432 #define user_i387_ia32_struct user_i387_struct
1433 #define user32_fxsr_struct user_fxsr_struct
1435 #endif /* CONFIG_X86_64 */
1437 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1438 static const struct user_regset x86_32_regsets
[] = {
1439 [REGSET_GENERAL
] = {
1440 .core_note_type
= NT_PRSTATUS
,
1441 .n
= sizeof(struct user_regs_struct32
) / sizeof(u32
),
1442 .size
= sizeof(u32
), .align
= sizeof(u32
),
1443 .get
= genregs32_get
, .set
= genregs32_set
1446 .core_note_type
= NT_PRFPREG
,
1447 .n
= sizeof(struct user_i387_ia32_struct
) / sizeof(u32
),
1448 .size
= sizeof(u32
), .align
= sizeof(u32
),
1449 .active
= fpregs_active
, .get
= fpregs_get
, .set
= fpregs_set
1452 .core_note_type
= NT_PRXFPREG
,
1453 .n
= sizeof(struct user32_fxsr_struct
) / sizeof(u32
),
1454 .size
= sizeof(u32
), .align
= sizeof(u32
),
1455 .active
= xfpregs_active
, .get
= xfpregs_get
, .set
= xfpregs_set
1458 .core_note_type
= NT_386_TLS
,
1459 .n
= GDT_ENTRY_TLS_ENTRIES
, .bias
= GDT_ENTRY_TLS_MIN
,
1460 .size
= sizeof(struct user_desc
),
1461 .align
= sizeof(struct user_desc
),
1462 .active
= regset_tls_active
,
1463 .get
= regset_tls_get
, .set
= regset_tls_set
1465 [REGSET_IOPERM32
] = {
1466 .core_note_type
= NT_386_IOPERM
,
1467 .n
= IO_BITMAP_BYTES
/ sizeof(u32
),
1468 .size
= sizeof(u32
), .align
= sizeof(u32
),
1469 .active
= ioperm_active
, .get
= ioperm_get
1473 static const struct user_regset_view user_x86_32_view
= {
1474 .name
= "i386", .e_machine
= EM_386
,
1475 .regsets
= x86_32_regsets
, .n
= ARRAY_SIZE(x86_32_regsets
)
1479 const struct user_regset_view
*task_user_regset_view(struct task_struct
*task
)
1481 #ifdef CONFIG_IA32_EMULATION
1482 if (test_tsk_thread_flag(task
, TIF_IA32
))
1484 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1485 return &user_x86_32_view
;
1487 #ifdef CONFIG_X86_64
1488 return &user_x86_64_view
;
1492 void send_sigtrap(struct task_struct
*tsk
, struct pt_regs
*regs
,
1493 int error_code
, int si_code
)
1495 struct siginfo info
;
1497 tsk
->thread
.trap_no
= 1;
1498 tsk
->thread
.error_code
= error_code
;
1500 memset(&info
, 0, sizeof(info
));
1501 info
.si_signo
= SIGTRAP
;
1502 info
.si_code
= si_code
;
1505 info
.si_addr
= user_mode_vm(regs
) ? (void __user
*) regs
->ip
: NULL
;
1507 /* Send us the fake SIGTRAP */
1508 force_sig_info(SIGTRAP
, &info
, tsk
);
1512 #ifdef CONFIG_X86_32
1514 #elif defined CONFIG_IA32_EMULATION
1515 # define IS_IA32 test_thread_flag(TIF_IA32)
1521 * We must return the syscall number to actually look up in the table.
1522 * This can be -1L to skip running any syscall at all.
1524 asmregparm
long syscall_trace_enter(struct pt_regs
*regs
)
1529 * If we stepped into a sysenter/syscall insn, it trapped in
1530 * kernel mode; do_debug() cleared TF and set TIF_SINGLESTEP.
1531 * If user-mode had set TF itself, then it's still clear from
1532 * do_debug() and we need to set it again to restore the user
1533 * state. If we entered on the slow path, TF was already set.
1535 if (test_thread_flag(TIF_SINGLESTEP
))
1536 regs
->flags
|= X86_EFLAGS_TF
;
1538 /* do the secure computing check first */
1539 secure_computing(regs
->orig_ax
);
1541 if (unlikely(test_thread_flag(TIF_SYSCALL_EMU
)))
1544 if ((ret
|| test_thread_flag(TIF_SYSCALL_TRACE
)) &&
1545 tracehook_report_syscall_entry(regs
))
1548 if (unlikely(current
->audit_context
)) {
1550 audit_syscall_entry(AUDIT_ARCH_I386
,
1553 regs
->dx
, regs
->si
);
1554 #ifdef CONFIG_X86_64
1556 audit_syscall_entry(AUDIT_ARCH_X86_64
,
1559 regs
->dx
, regs
->r10
);
1563 return ret
?: regs
->orig_ax
;
1566 asmregparm
void syscall_trace_leave(struct pt_regs
*regs
)
1568 if (unlikely(current
->audit_context
))
1569 audit_syscall_exit(AUDITSC_RESULT(regs
->ax
), regs
->ax
);
1571 if (test_thread_flag(TIF_SYSCALL_TRACE
))
1572 tracehook_report_syscall_exit(regs
, 0);
1575 * If TIF_SYSCALL_EMU is set, we only get here because of
1576 * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
1577 * We already reported this syscall instruction in
1578 * syscall_trace_enter(), so don't do any more now.
1580 if (unlikely(test_thread_flag(TIF_SYSCALL_EMU
)))
1584 * If we are single-stepping, synthesize a trap to follow the
1585 * system call instruction.
1587 if (test_thread_flag(TIF_SINGLESTEP
) &&
1588 tracehook_consider_fatal_signal(current
, SIGTRAP
, SIG_DFL
))
1589 send_sigtrap(current
, regs
, 0, TRAP_BRKPT
);