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>
24 #include <linux/workqueue.h>
26 #include <asm/uaccess.h>
27 #include <asm/pgtable.h>
28 #include <asm/system.h>
29 #include <asm/processor.h>
31 #include <asm/debugreg.h>
34 #include <asm/prctl.h>
35 #include <asm/proto.h>
40 #define CREATE_TRACE_POINTS
41 #include <trace/events/syscalls.h>
47 REGSET_IOPERM64 = REGSET_XFP,
53 * does not yet catch signals sent when the child dies.
54 * in exit.c or in signal.c.
58 * Determines which flags the user has access to [1 = access, 0 = no access].
60 #define FLAG_MASK_32 ((unsigned long) \
61 (X86_EFLAGS_CF | X86_EFLAGS_PF | \
62 X86_EFLAGS_AF | X86_EFLAGS_ZF | \
63 X86_EFLAGS_SF | X86_EFLAGS_TF | \
64 X86_EFLAGS_DF | X86_EFLAGS_OF | \
65 X86_EFLAGS_RF | X86_EFLAGS_AC))
68 * Determines whether a value may be installed in a segment register.
70 static inline bool invalid_selector(u16 value)
72 return unlikely(value != 0 && (value & SEGMENT_RPL_MASK) != USER_RPL);
77 #define FLAG_MASK FLAG_MASK_32
79 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long regno)
81 BUILD_BUG_ON(offsetof(struct pt_regs, bx) != 0);
82 return ®s->bx + (regno >> 2);
85 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
88 * Returning the value truncates it to 16 bits.
91 if (offset != offsetof(struct user_regs_struct, gs))
92 retval = *pt_regs_access(task_pt_regs(task), offset);
95 retval = get_user_gs(task_pt_regs(task));
97 retval = task_user_gs(task);
102 static int set_segment_reg(struct task_struct *task,
103 unsigned long offset, u16 value)
106 * The value argument was already truncated to 16 bits.
108 if (invalid_selector(value))
112 * For %cs and %ss we cannot permit a null selector.
113 * We can permit a bogus selector as long as it has USER_RPL.
114 * Null selectors are fine for other segment registers, but
115 * we will never get back to user mode with invalid %cs or %ss
116 * and will take the trap in iret instead. Much code relies
117 * on user_mode() to distinguish a user trap frame (which can
118 * safely use invalid selectors) from a kernel trap frame.
121 case offsetof(struct user_regs_struct, cs):
122 case offsetof(struct user_regs_struct, ss):
123 if (unlikely(value == 0))
127 *pt_regs_access(task_pt_regs(task), offset) = value;
130 case offsetof(struct user_regs_struct, gs):
132 set_user_gs(task_pt_regs(task), value);
134 task_user_gs(task) = value;
140 static unsigned long debugreg_addr_limit(struct task_struct *task)
142 return TASK_SIZE - 3;
145 #else /* CONFIG_X86_64 */
147 #define FLAG_MASK (FLAG_MASK_32 | X86_EFLAGS_NT)
149 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long offset)
151 BUILD_BUG_ON(offsetof(struct pt_regs, r15) != 0);
152 return ®s->r15 + (offset / sizeof(regs->r15));
155 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
158 * Returning the value truncates it to 16 bits.
163 case offsetof(struct user_regs_struct, fs):
164 if (task == current) {
165 /* Older gas can't assemble movq %?s,%r?? */
166 asm("movl %%fs,%0" : "=r" (seg));
169 return task->thread.fsindex;
170 case offsetof(struct user_regs_struct, gs):
171 if (task == current) {
172 asm("movl %%gs,%0" : "=r" (seg));
175 return task->thread.gsindex;
176 case offsetof(struct user_regs_struct, ds):
177 if (task == current) {
178 asm("movl %%ds,%0" : "=r" (seg));
181 return task->thread.ds;
182 case offsetof(struct user_regs_struct, es):
183 if (task == current) {
184 asm("movl %%es,%0" : "=r" (seg));
187 return task->thread.es;
189 case offsetof(struct user_regs_struct, cs):
190 case offsetof(struct user_regs_struct, ss):
193 return *pt_regs_access(task_pt_regs(task), offset);
196 static int set_segment_reg(struct task_struct *task,
197 unsigned long offset, u16 value)
200 * The value argument was already truncated to 16 bits.
202 if (invalid_selector(value))
206 case offsetof(struct user_regs_struct,fs):
208 * If this is setting fs as for normal 64-bit use but
209 * setting fs_base has implicitly changed it, leave it.
211 if ((value == FS_TLS_SEL && task->thread.fsindex == 0 &&
212 task->thread.fs != 0) ||
213 (value == 0 && task->thread.fsindex == FS_TLS_SEL &&
214 task->thread.fs == 0))
216 task->thread.fsindex = value;
218 loadsegment(fs, task->thread.fsindex);
220 case offsetof(struct user_regs_struct,gs):
222 * If this is setting gs as for normal 64-bit use but
223 * setting gs_base has implicitly changed it, leave it.
225 if ((value == GS_TLS_SEL && task->thread.gsindex == 0 &&
226 task->thread.gs != 0) ||
227 (value == 0 && task->thread.gsindex == GS_TLS_SEL &&
228 task->thread.gs == 0))
230 task->thread.gsindex = value;
232 load_gs_index(task->thread.gsindex);
234 case offsetof(struct user_regs_struct,ds):
235 task->thread.ds = value;
237 loadsegment(ds, task->thread.ds);
239 case offsetof(struct user_regs_struct,es):
240 task->thread.es = value;
242 loadsegment(es, task->thread.es);
246 * Can't actually change these in 64-bit mode.
248 case offsetof(struct user_regs_struct,cs):
249 if (unlikely(value == 0))
251 #ifdef CONFIG_IA32_EMULATION
252 if (test_tsk_thread_flag(task, TIF_IA32))
253 task_pt_regs(task)->cs = value;
256 case offsetof(struct user_regs_struct,ss):
257 if (unlikely(value == 0))
259 #ifdef CONFIG_IA32_EMULATION
260 if (test_tsk_thread_flag(task, TIF_IA32))
261 task_pt_regs(task)->ss = value;
269 static unsigned long debugreg_addr_limit(struct task_struct *task)
271 #ifdef CONFIG_IA32_EMULATION
272 if (test_tsk_thread_flag(task, TIF_IA32))
273 return IA32_PAGE_OFFSET - 3;
275 return TASK_SIZE_MAX - 7;
278 #endif /* CONFIG_X86_32 */
280 static unsigned long get_flags(struct task_struct *task)
282 unsigned long retval = task_pt_regs(task)->flags;
285 * If the debugger set TF, hide it from the readout.
287 if (test_tsk_thread_flag(task, TIF_FORCED_TF))
288 retval &= ~X86_EFLAGS_TF;
293 static int set_flags(struct task_struct *task, unsigned long value)
295 struct pt_regs *regs = task_pt_regs(task);
298 * If the user value contains TF, mark that
299 * it was not "us" (the debugger) that set it.
300 * If not, make sure it stays set if we had.
302 if (value & X86_EFLAGS_TF)
303 clear_tsk_thread_flag(task, TIF_FORCED_TF);
304 else if (test_tsk_thread_flag(task, TIF_FORCED_TF))
305 value |= X86_EFLAGS_TF;
307 regs->flags = (regs->flags & ~FLAG_MASK) | (value & FLAG_MASK);
312 static int putreg(struct task_struct *child,
313 unsigned long offset, unsigned long value)
316 case offsetof(struct user_regs_struct, cs):
317 case offsetof(struct user_regs_struct, ds):
318 case offsetof(struct user_regs_struct, es):
319 case offsetof(struct user_regs_struct, fs):
320 case offsetof(struct user_regs_struct, gs):
321 case offsetof(struct user_regs_struct, ss):
322 return set_segment_reg(child, offset, value);
324 case offsetof(struct user_regs_struct, flags):
325 return set_flags(child, value);
328 case offsetof(struct user_regs_struct,fs_base):
329 if (value >= TASK_SIZE_OF(child))
332 * When changing the segment base, use do_arch_prctl
333 * to set either thread.fs or thread.fsindex and the
334 * corresponding GDT slot.
336 if (child->thread.fs != value)
337 return do_arch_prctl(child, ARCH_SET_FS, value);
339 case offsetof(struct user_regs_struct,gs_base):
341 * Exactly the same here as the %fs handling above.
343 if (value >= TASK_SIZE_OF(child))
345 if (child->thread.gs != value)
346 return do_arch_prctl(child, ARCH_SET_GS, value);
351 *pt_regs_access(task_pt_regs(child), offset) = value;
355 static unsigned long getreg(struct task_struct *task, unsigned long offset)
358 case offsetof(struct user_regs_struct, cs):
359 case offsetof(struct user_regs_struct, ds):
360 case offsetof(struct user_regs_struct, es):
361 case offsetof(struct user_regs_struct, fs):
362 case offsetof(struct user_regs_struct, gs):
363 case offsetof(struct user_regs_struct, ss):
364 return get_segment_reg(task, offset);
366 case offsetof(struct user_regs_struct, flags):
367 return get_flags(task);
370 case offsetof(struct user_regs_struct, fs_base): {
372 * do_arch_prctl may have used a GDT slot instead of
373 * the MSR. To userland, it appears the same either
374 * way, except the %fs segment selector might not be 0.
376 unsigned int seg = task->thread.fsindex;
377 if (task->thread.fs != 0)
378 return task->thread.fs;
380 asm("movl %%fs,%0" : "=r" (seg));
381 if (seg != FS_TLS_SEL)
383 return get_desc_base(&task->thread.tls_array[FS_TLS]);
385 case offsetof(struct user_regs_struct, gs_base): {
387 * Exactly the same here as the %fs handling above.
389 unsigned int seg = task->thread.gsindex;
390 if (task->thread.gs != 0)
391 return task->thread.gs;
393 asm("movl %%gs,%0" : "=r" (seg));
394 if (seg != GS_TLS_SEL)
396 return get_desc_base(&task->thread.tls_array[GS_TLS]);
401 return *pt_regs_access(task_pt_regs(task), offset);
404 static int genregs_get(struct task_struct *target,
405 const struct user_regset *regset,
406 unsigned int pos, unsigned int count,
407 void *kbuf, void __user *ubuf)
410 unsigned long *k = kbuf;
412 *k++ = getreg(target, pos);
417 unsigned long __user *u = ubuf;
419 if (__put_user(getreg(target, pos), u++))
429 static int genregs_set(struct task_struct *target,
430 const struct user_regset *regset,
431 unsigned int pos, unsigned int count,
432 const void *kbuf, const void __user *ubuf)
436 const unsigned long *k = kbuf;
437 while (count > 0 && !ret) {
438 ret = putreg(target, pos, *k++);
443 const unsigned long __user *u = ubuf;
444 while (count > 0 && !ret) {
446 ret = __get_user(word, u++);
449 ret = putreg(target, pos, word);
458 * This function is trivial and will be inlined by the compiler.
459 * Having it separates the implementation details of debug
460 * registers from the interface details of ptrace.
462 static unsigned long ptrace_get_debugreg(struct task_struct *child, int n)
465 case 0: return child->thread.debugreg0;
466 case 1: return child->thread.debugreg1;
467 case 2: return child->thread.debugreg2;
468 case 3: return child->thread.debugreg3;
469 case 6: return child->thread.debugreg6;
470 case 7: return child->thread.debugreg7;
475 static int ptrace_set_debugreg(struct task_struct *child,
476 int n, unsigned long data)
480 if (unlikely(n == 4 || n == 5))
483 if (n < 4 && unlikely(data >= debugreg_addr_limit(child)))
487 case 0: child->thread.debugreg0 = data; break;
488 case 1: child->thread.debugreg1 = data; break;
489 case 2: child->thread.debugreg2 = data; break;
490 case 3: child->thread.debugreg3 = data; break;
493 if ((data & ~0xffffffffUL) != 0)
495 child->thread.debugreg6 = data;
500 * Sanity-check data. Take one half-byte at once with
501 * check = (val >> (16 + 4*i)) & 0xf. It contains the
502 * R/Wi and LENi bits; bits 0 and 1 are R/Wi, and bits
503 * 2 and 3 are LENi. Given a list of invalid values,
504 * we do mask |= 1 << invalid_value, so that
505 * (mask >> check) & 1 is a correct test for invalid
508 * R/Wi contains the type of the breakpoint /
509 * watchpoint, LENi contains the length of the watched
510 * data in the watchpoint case.
512 * The invalid values are:
513 * - LENi == 0x10 (undefined), so mask |= 0x0f00. [32-bit]
514 * - R/Wi == 0x10 (break on I/O reads or writes), so
516 * - R/Wi == 0x00 && LENi != 0x00, so we have mask |=
519 * Finally, mask = 0x0f00 | 0x4444 | 0x1110 == 0x5f54.
521 * See the Intel Manual "System Programming Guide",
524 * Note that LENi == 0x10 is defined on x86_64 in long
525 * mode (i.e. even for 32-bit userspace software, but
526 * 64-bit kernel), so the x86_64 mask value is 0x5454.
527 * See the AMD manual no. 24593 (AMD64 System Programming)
530 #define DR7_MASK 0x5f54
532 #define DR7_MASK 0x5554
534 data &= ~DR_CONTROL_RESERVED;
535 for (i = 0; i < 4; i++)
536 if ((DR7_MASK >> ((data >> (16 + 4*i)) & 0xf)) & 1)
538 child->thread.debugreg7 = data;
540 set_tsk_thread_flag(child, TIF_DEBUG);
542 clear_tsk_thread_flag(child, TIF_DEBUG);
550 * These access the current or another (stopped) task's io permission
551 * bitmap for debugging or core dump.
553 static int ioperm_active(struct task_struct *target,
554 const struct user_regset *regset)
556 return target->thread.io_bitmap_max / regset->size;
559 static int ioperm_get(struct task_struct *target,
560 const struct user_regset *regset,
561 unsigned int pos, unsigned int count,
562 void *kbuf, void __user *ubuf)
564 if (!target->thread.io_bitmap_ptr)
567 return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
568 target->thread.io_bitmap_ptr,
572 #ifdef CONFIG_X86_PTRACE_BTS
574 * A branch trace store context.
576 * Contexts may only be installed by ptrace_bts_config() and only for
579 * Contexts are destroyed when the tracee is detached from the tracer.
580 * The actual destruction work requires interrupts enabled, so the
581 * work is deferred and will be scheduled during __ptrace_unlink().
583 * Contexts hold an additional task_struct reference on the traced
584 * task, as well as a reference on the tracer's mm.
586 * Ptrace already holds a task_struct for the duration of ptrace operations,
587 * but since destruction is deferred, it may be executed after both
588 * tracer and tracee exited.
591 /* The branch trace handle. */
592 struct bts_tracer *tracer;
594 /* The buffer used to store the branch trace and its size. */
598 /* The mm that paid for the above buffer. */
599 struct mm_struct *mm;
601 /* The task this context belongs to. */
602 struct task_struct *task;
604 /* The signal to send on a bts buffer overflow. */
605 unsigned int bts_ovfl_signal;
607 /* The work struct to destroy a context. */
608 struct work_struct work;
611 static int alloc_bts_buffer(struct bts_context *context, unsigned int size)
616 err = account_locked_memory(current->mm, current->signal->rlim, size);
620 buffer = kzalloc(size, GFP_KERNEL);
624 context->buffer = buffer;
625 context->size = size;
626 context->mm = get_task_mm(current);
631 refund_locked_memory(current->mm, size);
635 static inline void free_bts_buffer(struct bts_context *context)
637 if (!context->buffer)
640 kfree(context->buffer);
641 context->buffer = NULL;
643 refund_locked_memory(context->mm, context->size);
650 static void free_bts_context_work(struct work_struct *w)
652 struct bts_context *context;
654 context = container_of(w, struct bts_context, work);
656 ds_release_bts(context->tracer);
657 put_task_struct(context->task);
658 free_bts_buffer(context);
662 static inline void free_bts_context(struct bts_context *context)
664 INIT_WORK(&context->work, free_bts_context_work);
665 schedule_work(&context->work);
668 static inline struct bts_context *alloc_bts_context(struct task_struct *task)
670 struct bts_context *context = kzalloc(sizeof(*context), GFP_KERNEL);
672 context->task = task;
675 get_task_struct(task);
681 static int ptrace_bts_read_record(struct task_struct *child, size_t index,
682 struct bts_struct __user *out)
684 struct bts_context *context;
685 const struct bts_trace *trace;
686 struct bts_struct bts;
687 const unsigned char *at;
690 context = child->bts;
694 trace = ds_read_bts(context->tracer);
698 at = trace->ds.top - ((index + 1) * trace->ds.size);
699 if ((void *)at < trace->ds.begin)
700 at += (trace->ds.n * trace->ds.size);
705 error = trace->read(context->tracer, at, &bts);
709 if (copy_to_user(out, &bts, sizeof(bts)))
715 static int ptrace_bts_drain(struct task_struct *child,
717 struct bts_struct __user *out)
719 struct bts_context *context;
720 const struct bts_trace *trace;
721 const unsigned char *at;
722 int error, drained = 0;
724 context = child->bts;
728 trace = ds_read_bts(context->tracer);
735 if (size < (trace->ds.top - trace->ds.begin))
738 for (at = trace->ds.begin; (void *)at < trace->ds.top;
739 out++, drained++, at += trace->ds.size) {
740 struct bts_struct bts;
742 error = trace->read(context->tracer, at, &bts);
746 if (copy_to_user(out, &bts, sizeof(bts)))
750 memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size);
752 error = ds_reset_bts(context->tracer);
759 static int ptrace_bts_config(struct task_struct *child,
761 const struct ptrace_bts_config __user *ucfg)
763 struct bts_context *context;
764 struct ptrace_bts_config cfg;
765 unsigned int flags = 0;
767 if (cfg_size < sizeof(cfg))
770 if (copy_from_user(&cfg, ucfg, sizeof(cfg)))
773 context = child->bts;
775 context = alloc_bts_context(child);
779 if (cfg.flags & PTRACE_BTS_O_SIGNAL) {
784 context->bts_ovfl_signal = cfg.signal;
787 ds_release_bts(context->tracer);
788 context->tracer = NULL;
790 if ((cfg.flags & PTRACE_BTS_O_ALLOC) && (cfg.size != context->size)) {
793 free_bts_buffer(context);
797 err = alloc_bts_buffer(context, cfg.size);
802 if (cfg.flags & PTRACE_BTS_O_TRACE)
805 if (cfg.flags & PTRACE_BTS_O_SCHED)
806 flags |= BTS_TIMESTAMPS;
809 ds_request_bts_task(child, context->buffer, context->size,
810 NULL, (size_t)-1, flags);
811 if (unlikely(IS_ERR(context->tracer))) {
812 int error = PTR_ERR(context->tracer);
814 free_bts_buffer(context);
815 context->tracer = NULL;
822 static int ptrace_bts_status(struct task_struct *child,
824 struct ptrace_bts_config __user *ucfg)
826 struct bts_context *context;
827 const struct bts_trace *trace;
828 struct ptrace_bts_config cfg;
830 context = child->bts;
834 if (cfg_size < sizeof(cfg))
837 trace = ds_read_bts(context->tracer);
841 memset(&cfg, 0, sizeof(cfg));
842 cfg.size = trace->ds.end - trace->ds.begin;
843 cfg.signal = context->bts_ovfl_signal;
844 cfg.bts_size = sizeof(struct bts_struct);
847 cfg.flags |= PTRACE_BTS_O_SIGNAL;
849 if (trace->ds.flags & BTS_USER)
850 cfg.flags |= PTRACE_BTS_O_TRACE;
852 if (trace->ds.flags & BTS_TIMESTAMPS)
853 cfg.flags |= PTRACE_BTS_O_SCHED;
855 if (copy_to_user(ucfg, &cfg, sizeof(cfg)))
861 static int ptrace_bts_clear(struct task_struct *child)
863 struct bts_context *context;
864 const struct bts_trace *trace;
866 context = child->bts;
870 trace = ds_read_bts(context->tracer);
874 memset(trace->ds.begin, 0, trace->ds.n * trace->ds.size);
876 return ds_reset_bts(context->tracer);
879 static int ptrace_bts_size(struct task_struct *child)
881 struct bts_context *context;
882 const struct bts_trace *trace;
884 context = child->bts;
888 trace = ds_read_bts(context->tracer);
892 return (trace->ds.top - trace->ds.begin) / trace->ds.size;
896 * Called from __ptrace_unlink() after the child has been moved back
897 * to its original parent.
899 void ptrace_bts_untrace(struct task_struct *child)
901 if (unlikely(child->bts)) {
902 free_bts_context(child->bts);
906 #endif /* CONFIG_X86_PTRACE_BTS */
909 * Called by kernel/ptrace.c when detaching..
911 * Make sure the single step bit is not set.
913 void ptrace_disable(struct task_struct *child)
915 user_disable_single_step(child);
916 #ifdef TIF_SYSCALL_EMU
917 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
921 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
922 static const struct user_regset_view user_x86_32_view; /* Initialized below. */
925 long arch_ptrace(struct task_struct *child, long request, long addr, long data)
928 unsigned long __user *datap = (unsigned long __user *)data;
931 /* read the word at location addr in the USER area. */
932 case PTRACE_PEEKUSR: {
936 if ((addr & (sizeof(data) - 1)) || addr < 0 ||
937 addr >= sizeof(struct user))
940 tmp = 0; /* Default return condition */
941 if (addr < sizeof(struct user_regs_struct))
942 tmp = getreg(child, addr);
943 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
944 addr <= offsetof(struct user, u_debugreg[7])) {
945 addr -= offsetof(struct user, u_debugreg[0]);
946 tmp = ptrace_get_debugreg(child, addr / sizeof(data));
948 ret = put_user(tmp, datap);
952 case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
954 if ((addr & (sizeof(data) - 1)) || addr < 0 ||
955 addr >= sizeof(struct user))
958 if (addr < sizeof(struct user_regs_struct))
959 ret = putreg(child, addr, data);
960 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
961 addr <= offsetof(struct user, u_debugreg[7])) {
962 addr -= offsetof(struct user, u_debugreg[0]);
963 ret = ptrace_set_debugreg(child,
964 addr / sizeof(data), data);
968 case PTRACE_GETREGS: /* Get all gp regs from the child. */
969 return copy_regset_to_user(child,
970 task_user_regset_view(current),
972 0, sizeof(struct user_regs_struct),
975 case PTRACE_SETREGS: /* Set all gp regs in the child. */
976 return copy_regset_from_user(child,
977 task_user_regset_view(current),
979 0, sizeof(struct user_regs_struct),
982 case PTRACE_GETFPREGS: /* Get the child FPU state. */
983 return copy_regset_to_user(child,
984 task_user_regset_view(current),
986 0, sizeof(struct user_i387_struct),
989 case PTRACE_SETFPREGS: /* Set the child FPU state. */
990 return copy_regset_from_user(child,
991 task_user_regset_view(current),
993 0, sizeof(struct user_i387_struct),
997 case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
998 return copy_regset_to_user(child, &user_x86_32_view,
1000 0, sizeof(struct user_fxsr_struct),
1003 case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
1004 return copy_regset_from_user(child, &user_x86_32_view,
1006 0, sizeof(struct user_fxsr_struct),
1010 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1011 case PTRACE_GET_THREAD_AREA:
1014 ret = do_get_thread_area(child, addr,
1015 (struct user_desc __user *) data);
1018 case PTRACE_SET_THREAD_AREA:
1021 ret = do_set_thread_area(child, addr,
1022 (struct user_desc __user *) data, 0);
1026 #ifdef CONFIG_X86_64
1027 /* normal 64bit interface to access TLS data.
1028 Works just like arch_prctl, except that the arguments
1030 case PTRACE_ARCH_PRCTL:
1031 ret = do_arch_prctl(child, data, addr);
1036 * These bits need more cooking - not enabled yet:
1038 #ifdef CONFIG_X86_PTRACE_BTS
1039 case PTRACE_BTS_CONFIG:
1040 ret = ptrace_bts_config
1041 (child, data, (struct ptrace_bts_config __user *)addr);
1044 case PTRACE_BTS_STATUS:
1045 ret = ptrace_bts_status
1046 (child, data, (struct ptrace_bts_config __user *)addr);
1049 case PTRACE_BTS_SIZE:
1050 ret = ptrace_bts_size(child);
1053 case PTRACE_BTS_GET:
1054 ret = ptrace_bts_read_record
1055 (child, data, (struct bts_struct __user *) addr);
1058 case PTRACE_BTS_CLEAR:
1059 ret = ptrace_bts_clear(child);
1062 case PTRACE_BTS_DRAIN:
1063 ret = ptrace_bts_drain
1064 (child, data, (struct bts_struct __user *) addr);
1066 #endif /* CONFIG_X86_PTRACE_BTS */
1069 ret = ptrace_request(child, request, addr, data);
1076 #ifdef CONFIG_IA32_EMULATION
1078 #include <linux/compat.h>
1079 #include <linux/syscalls.h>
1080 #include <asm/ia32.h>
1081 #include <asm/user32.h>
1084 case offsetof(struct user32, regs.l): \
1085 regs->q = value; break
1088 case offsetof(struct user32, regs.rs): \
1089 return set_segment_reg(child, \
1090 offsetof(struct user_regs_struct, rs), \
1094 static int putreg32(struct task_struct *child, unsigned regno, u32 value)
1096 struct pt_regs *regs = task_pt_regs(child);
1117 case offsetof(struct user32, regs.orig_eax):
1119 * A 32-bit debugger setting orig_eax means to restore
1120 * the state of the task restarting a 32-bit syscall.
1121 * Make sure we interpret the -ERESTART* codes correctly
1122 * in case the task is not actually still sitting at the
1123 * exit from a 32-bit syscall with TS_COMPAT still set.
1125 regs->orig_ax = value;
1126 if (syscall_get_nr(child, regs) >= 0)
1127 task_thread_info(child)->status |= TS_COMPAT;
1130 case offsetof(struct user32, regs.eflags):
1131 return set_flags(child, value);
1133 case offsetof(struct user32, u_debugreg[0]) ...
1134 offsetof(struct user32, u_debugreg[7]):
1135 regno -= offsetof(struct user32, u_debugreg[0]);
1136 return ptrace_set_debugreg(child, regno / 4, value);
1139 if (regno > sizeof(struct user32) || (regno & 3))
1143 * Other dummy fields in the virtual user structure
1155 case offsetof(struct user32, regs.l): \
1156 *val = regs->q; break
1159 case offsetof(struct user32, regs.rs): \
1160 *val = get_segment_reg(child, \
1161 offsetof(struct user_regs_struct, rs)); \
1164 static int getreg32(struct task_struct *child, unsigned regno, u32 *val)
1166 struct pt_regs *regs = task_pt_regs(child);
1184 R32(orig_eax, orig_ax);
1188 case offsetof(struct user32, regs.eflags):
1189 *val = get_flags(child);
1192 case offsetof(struct user32, u_debugreg[0]) ...
1193 offsetof(struct user32, u_debugreg[7]):
1194 regno -= offsetof(struct user32, u_debugreg[0]);
1195 *val = ptrace_get_debugreg(child, regno / 4);
1199 if (regno > sizeof(struct user32) || (regno & 3))
1203 * Other dummy fields in the virtual user structure
1215 static int genregs32_get(struct task_struct *target,
1216 const struct user_regset *regset,
1217 unsigned int pos, unsigned int count,
1218 void *kbuf, void __user *ubuf)
1221 compat_ulong_t *k = kbuf;
1223 getreg32(target, pos, k++);
1224 count -= sizeof(*k);
1228 compat_ulong_t __user *u = ubuf;
1230 compat_ulong_t word;
1231 getreg32(target, pos, &word);
1232 if (__put_user(word, u++))
1234 count -= sizeof(*u);
1242 static int genregs32_set(struct task_struct *target,
1243 const struct user_regset *regset,
1244 unsigned int pos, unsigned int count,
1245 const void *kbuf, const void __user *ubuf)
1249 const compat_ulong_t *k = kbuf;
1250 while (count > 0 && !ret) {
1251 ret = putreg32(target, pos, *k++);
1252 count -= sizeof(*k);
1256 const compat_ulong_t __user *u = ubuf;
1257 while (count > 0 && !ret) {
1258 compat_ulong_t word;
1259 ret = __get_user(word, u++);
1262 ret = putreg32(target, pos, word);
1263 count -= sizeof(*u);
1270 long compat_arch_ptrace(struct task_struct *child, compat_long_t request,
1271 compat_ulong_t caddr, compat_ulong_t cdata)
1273 unsigned long addr = caddr;
1274 unsigned long data = cdata;
1275 void __user *datap = compat_ptr(data);
1280 case PTRACE_PEEKUSR:
1281 ret = getreg32(child, addr, &val);
1283 ret = put_user(val, (__u32 __user *)datap);
1286 case PTRACE_POKEUSR:
1287 ret = putreg32(child, addr, data);
1290 case PTRACE_GETREGS: /* Get all gp regs from the child. */
1291 return copy_regset_to_user(child, &user_x86_32_view,
1293 0, sizeof(struct user_regs_struct32),
1296 case PTRACE_SETREGS: /* Set all gp regs in the child. */
1297 return copy_regset_from_user(child, &user_x86_32_view,
1299 sizeof(struct user_regs_struct32),
1302 case PTRACE_GETFPREGS: /* Get the child FPU state. */
1303 return copy_regset_to_user(child, &user_x86_32_view,
1305 sizeof(struct user_i387_ia32_struct),
1308 case PTRACE_SETFPREGS: /* Set the child FPU state. */
1309 return copy_regset_from_user(
1310 child, &user_x86_32_view, REGSET_FP,
1311 0, sizeof(struct user_i387_ia32_struct), datap);
1313 case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
1314 return copy_regset_to_user(child, &user_x86_32_view,
1316 sizeof(struct user32_fxsr_struct),
1319 case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
1320 return copy_regset_from_user(child, &user_x86_32_view,
1322 sizeof(struct user32_fxsr_struct),
1325 case PTRACE_GET_THREAD_AREA:
1326 case PTRACE_SET_THREAD_AREA:
1327 #ifdef CONFIG_X86_PTRACE_BTS
1328 case PTRACE_BTS_CONFIG:
1329 case PTRACE_BTS_STATUS:
1330 case PTRACE_BTS_SIZE:
1331 case PTRACE_BTS_GET:
1332 case PTRACE_BTS_CLEAR:
1333 case PTRACE_BTS_DRAIN:
1334 #endif /* CONFIG_X86_PTRACE_BTS */
1335 return arch_ptrace(child, request, addr, data);
1338 return compat_ptrace_request(child, request, addr, data);
1344 #endif /* CONFIG_IA32_EMULATION */
1346 #ifdef CONFIG_X86_64
1348 static const struct user_regset x86_64_regsets[] = {
1349 [REGSET_GENERAL] = {
1350 .core_note_type = NT_PRSTATUS,
1351 .n = sizeof(struct user_regs_struct) / sizeof(long),
1352 .size = sizeof(long), .align = sizeof(long),
1353 .get = genregs_get, .set = genregs_set
1356 .core_note_type = NT_PRFPREG,
1357 .n = sizeof(struct user_i387_struct) / sizeof(long),
1358 .size = sizeof(long), .align = sizeof(long),
1359 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1361 [REGSET_IOPERM64] = {
1362 .core_note_type = NT_386_IOPERM,
1363 .n = IO_BITMAP_LONGS,
1364 .size = sizeof(long), .align = sizeof(long),
1365 .active = ioperm_active, .get = ioperm_get
1369 static const struct user_regset_view user_x86_64_view = {
1370 .name = "x86_64", .e_machine = EM_X86_64,
1371 .regsets = x86_64_regsets, .n = ARRAY_SIZE(x86_64_regsets)
1374 #else /* CONFIG_X86_32 */
1376 #define user_regs_struct32 user_regs_struct
1377 #define genregs32_get genregs_get
1378 #define genregs32_set genregs_set
1380 #define user_i387_ia32_struct user_i387_struct
1381 #define user32_fxsr_struct user_fxsr_struct
1383 #endif /* CONFIG_X86_64 */
1385 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1386 static const struct user_regset x86_32_regsets[] = {
1387 [REGSET_GENERAL] = {
1388 .core_note_type = NT_PRSTATUS,
1389 .n = sizeof(struct user_regs_struct32) / sizeof(u32),
1390 .size = sizeof(u32), .align = sizeof(u32),
1391 .get = genregs32_get, .set = genregs32_set
1394 .core_note_type = NT_PRFPREG,
1395 .n = sizeof(struct user_i387_ia32_struct) / sizeof(u32),
1396 .size = sizeof(u32), .align = sizeof(u32),
1397 .active = fpregs_active, .get = fpregs_get, .set = fpregs_set
1400 .core_note_type = NT_PRXFPREG,
1401 .n = sizeof(struct user32_fxsr_struct) / sizeof(u32),
1402 .size = sizeof(u32), .align = sizeof(u32),
1403 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1406 .core_note_type = NT_386_TLS,
1407 .n = GDT_ENTRY_TLS_ENTRIES, .bias = GDT_ENTRY_TLS_MIN,
1408 .size = sizeof(struct user_desc),
1409 .align = sizeof(struct user_desc),
1410 .active = regset_tls_active,
1411 .get = regset_tls_get, .set = regset_tls_set
1413 [REGSET_IOPERM32] = {
1414 .core_note_type = NT_386_IOPERM,
1415 .n = IO_BITMAP_BYTES / sizeof(u32),
1416 .size = sizeof(u32), .align = sizeof(u32),
1417 .active = ioperm_active, .get = ioperm_get
1421 static const struct user_regset_view user_x86_32_view = {
1422 .name = "i386", .e_machine = EM_386,
1423 .regsets = x86_32_regsets, .n = ARRAY_SIZE(x86_32_regsets)
1427 const struct user_regset_view *task_user_regset_view(struct task_struct *task)
1429 #ifdef CONFIG_IA32_EMULATION
1430 if (test_tsk_thread_flag(task, TIF_IA32))
1432 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1433 return &user_x86_32_view;
1435 #ifdef CONFIG_X86_64
1436 return &user_x86_64_view;
1440 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs,
1441 int error_code, int si_code)
1443 struct siginfo info;
1445 tsk->thread.trap_no = 1;
1446 tsk->thread.error_code = error_code;
1448 memset(&info, 0, sizeof(info));
1449 info.si_signo = SIGTRAP;
1450 info.si_code = si_code;
1453 info.si_addr = user_mode_vm(regs) ? (void __user *) regs->ip : NULL;
1455 /* Send us the fake SIGTRAP */
1456 force_sig_info(SIGTRAP, &info, tsk);
1460 #ifdef CONFIG_X86_32
1462 #elif defined CONFIG_IA32_EMULATION
1463 # define IS_IA32 is_compat_task()
1469 * We must return the syscall number to actually look up in the table.
1470 * This can be -1L to skip running any syscall at all.
1472 asmregparm long syscall_trace_enter(struct pt_regs *regs)
1477 * If we stepped into a sysenter/syscall insn, it trapped in
1478 * kernel mode; do_debug() cleared TF and set TIF_SINGLESTEP.
1479 * If user-mode had set TF itself, then it's still clear from
1480 * do_debug() and we need to set it again to restore the user
1481 * state. If we entered on the slow path, TF was already set.
1483 if (test_thread_flag(TIF_SINGLESTEP))
1484 regs->flags |= X86_EFLAGS_TF;
1486 /* do the secure computing check first */
1487 secure_computing(regs->orig_ax);
1489 if (unlikely(test_thread_flag(TIF_SYSCALL_EMU)))
1492 if ((ret || test_thread_flag(TIF_SYSCALL_TRACE)) &&
1493 tracehook_report_syscall_entry(regs))
1496 if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
1497 trace_sys_enter(regs, regs->orig_ax);
1499 if (unlikely(current->audit_context)) {
1501 audit_syscall_entry(AUDIT_ARCH_I386,
1504 regs->dx, regs->si);
1505 #ifdef CONFIG_X86_64
1507 audit_syscall_entry(AUDIT_ARCH_X86_64,
1510 regs->dx, regs->r10);
1514 return ret ?: regs->orig_ax;
1517 asmregparm void syscall_trace_leave(struct pt_regs *regs)
1519 if (unlikely(current->audit_context))
1520 audit_syscall_exit(AUDITSC_RESULT(regs->ax), regs->ax);
1522 if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
1523 trace_sys_exit(regs, regs->ax);
1525 if (test_thread_flag(TIF_SYSCALL_TRACE))
1526 tracehook_report_syscall_exit(regs, 0);
1529 * If TIF_SYSCALL_EMU is set, we only get here because of
1530 * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
1531 * We already reported this syscall instruction in
1532 * syscall_trace_enter(), so don't do any more now.
1534 if (unlikely(test_thread_flag(TIF_SYSCALL_EMU)))
1538 * If we are single-stepping, synthesize a trap to follow the
1539 * system call instruction.
1541 if (test_thread_flag(TIF_SINGLESTEP) &&
1542 tracehook_consider_fatal_signal(current, SIGTRAP))
1543 send_sigtrap(current, regs, 0, TRAP_BRKPT);