1 /* By Ross Biro 1/23/92 */
3 * Pentium III FXSR, SSE support
4 * Gareth Hughes <gareth@valinux.com>, May 2000
7 #include <linux/kernel.h>
8 #include <linux/sched.h>
10 #include <linux/smp.h>
11 #include <linux/errno.h>
12 #include <linux/slab.h>
13 #include <linux/ptrace.h>
14 #include <linux/regset.h>
15 #include <linux/tracehook.h>
16 #include <linux/user.h>
17 #include <linux/elf.h>
18 #include <linux/security.h>
19 #include <linux/audit.h>
20 #include <linux/seccomp.h>
21 #include <linux/signal.h>
22 #include <linux/perf_event.h>
23 #include <linux/hw_breakpoint.h>
24 #include <linux/rcupdate.h>
25 #include <linux/module.h>
27 #include <asm/uaccess.h>
28 #include <asm/pgtable.h>
29 #include <asm/processor.h>
31 #include <asm/fpu-internal.h>
32 #include <asm/debugreg.h>
35 #include <asm/prctl.h>
36 #include <asm/proto.h>
37 #include <asm/hw_breakpoint.h>
38 #include <asm/traps.h>
42 #define CREATE_TRACE_POINTS
43 #include <trace/events/syscalls.h>
49 REGSET_IOPERM64 = REGSET_XFP,
55 struct pt_regs_offset {
60 #define REG_OFFSET_NAME(r) {.name = #r, .offset = offsetof(struct pt_regs, r)}
61 #define REG_OFFSET_END {.name = NULL, .offset = 0}
63 static const struct pt_regs_offset regoffset_table[] = {
87 REG_OFFSET_NAME(orig_ax),
90 REG_OFFSET_NAME(flags),
97 * regs_query_register_offset() - query register offset from its name
98 * @name: the name of a register
100 * regs_query_register_offset() returns the offset of a register in struct
101 * pt_regs from its name. If the name is invalid, this returns -EINVAL;
103 int regs_query_register_offset(const char *name)
105 const struct pt_regs_offset *roff;
106 for (roff = regoffset_table; roff->name != NULL; roff++)
107 if (!strcmp(roff->name, name))
113 * regs_query_register_name() - query register name from its offset
114 * @offset: the offset of a register in struct pt_regs.
116 * regs_query_register_name() returns the name of a register from its
117 * offset in struct pt_regs. If the @offset is invalid, this returns NULL;
119 const char *regs_query_register_name(unsigned int offset)
121 const struct pt_regs_offset *roff;
122 for (roff = regoffset_table; roff->name != NULL; roff++)
123 if (roff->offset == offset)
128 static const int arg_offs_table[] = {
130 [0] = offsetof(struct pt_regs, ax),
131 [1] = offsetof(struct pt_regs, dx),
132 [2] = offsetof(struct pt_regs, cx)
133 #else /* CONFIG_X86_64 */
134 [0] = offsetof(struct pt_regs, di),
135 [1] = offsetof(struct pt_regs, si),
136 [2] = offsetof(struct pt_regs, dx),
137 [3] = offsetof(struct pt_regs, cx),
138 [4] = offsetof(struct pt_regs, r8),
139 [5] = offsetof(struct pt_regs, r9)
144 * does not yet catch signals sent when the child dies.
145 * in exit.c or in signal.c.
149 * Determines which flags the user has access to [1 = access, 0 = no access].
151 #define FLAG_MASK_32 ((unsigned long) \
152 (X86_EFLAGS_CF | X86_EFLAGS_PF | \
153 X86_EFLAGS_AF | X86_EFLAGS_ZF | \
154 X86_EFLAGS_SF | X86_EFLAGS_TF | \
155 X86_EFLAGS_DF | X86_EFLAGS_OF | \
156 X86_EFLAGS_RF | X86_EFLAGS_AC))
159 * Determines whether a value may be installed in a segment register.
161 static inline bool invalid_selector(u16 value)
163 return unlikely(value != 0 && (value & SEGMENT_RPL_MASK) != USER_RPL);
168 #define FLAG_MASK FLAG_MASK_32
171 * X86_32 CPUs don't save ss and esp if the CPU is already in kernel mode
172 * when it traps. The previous stack will be directly underneath the saved
173 * registers, and 'sp/ss' won't even have been saved. Thus the '®s->sp'.
175 * Now, if the stack is empty, '®s->sp' is out of range. In this
176 * case we try to take the previous stack. To always return a non-null
177 * stack pointer we fall back to regs as stack if no previous stack
180 * This is valid only for kernel mode traps.
182 unsigned long kernel_stack_pointer(struct pt_regs *regs)
184 unsigned long context = (unsigned long)regs & ~(THREAD_SIZE - 1);
185 unsigned long sp = (unsigned long)®s->sp;
186 struct thread_info *tinfo;
188 if (context == (sp & ~(THREAD_SIZE - 1)))
191 tinfo = (struct thread_info *)context;
192 if (tinfo->previous_esp)
193 return tinfo->previous_esp;
195 return (unsigned long)regs;
197 EXPORT_SYMBOL_GPL(kernel_stack_pointer);
199 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long regno)
201 BUILD_BUG_ON(offsetof(struct pt_regs, bx) != 0);
202 return ®s->bx + (regno >> 2);
205 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
208 * Returning the value truncates it to 16 bits.
211 if (offset != offsetof(struct user_regs_struct, gs))
212 retval = *pt_regs_access(task_pt_regs(task), offset);
215 retval = get_user_gs(task_pt_regs(task));
217 retval = task_user_gs(task);
222 static int set_segment_reg(struct task_struct *task,
223 unsigned long offset, u16 value)
226 * The value argument was already truncated to 16 bits.
228 if (invalid_selector(value))
232 * For %cs and %ss we cannot permit a null selector.
233 * We can permit a bogus selector as long as it has USER_RPL.
234 * Null selectors are fine for other segment registers, but
235 * we will never get back to user mode with invalid %cs or %ss
236 * and will take the trap in iret instead. Much code relies
237 * on user_mode() to distinguish a user trap frame (which can
238 * safely use invalid selectors) from a kernel trap frame.
241 case offsetof(struct user_regs_struct, cs):
242 case offsetof(struct user_regs_struct, ss):
243 if (unlikely(value == 0))
247 *pt_regs_access(task_pt_regs(task), offset) = value;
250 case offsetof(struct user_regs_struct, gs):
252 set_user_gs(task_pt_regs(task), value);
254 task_user_gs(task) = value;
260 #else /* CONFIG_X86_64 */
262 #define FLAG_MASK (FLAG_MASK_32 | X86_EFLAGS_NT)
264 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long offset)
266 BUILD_BUG_ON(offsetof(struct pt_regs, r15) != 0);
267 return ®s->r15 + (offset / sizeof(regs->r15));
270 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
273 * Returning the value truncates it to 16 bits.
278 case offsetof(struct user_regs_struct, fs):
279 if (task == current) {
280 /* Older gas can't assemble movq %?s,%r?? */
281 asm("movl %%fs,%0" : "=r" (seg));
284 return task->thread.fsindex;
285 case offsetof(struct user_regs_struct, gs):
286 if (task == current) {
287 asm("movl %%gs,%0" : "=r" (seg));
290 return task->thread.gsindex;
291 case offsetof(struct user_regs_struct, ds):
292 if (task == current) {
293 asm("movl %%ds,%0" : "=r" (seg));
296 return task->thread.ds;
297 case offsetof(struct user_regs_struct, es):
298 if (task == current) {
299 asm("movl %%es,%0" : "=r" (seg));
302 return task->thread.es;
304 case offsetof(struct user_regs_struct, cs):
305 case offsetof(struct user_regs_struct, ss):
308 return *pt_regs_access(task_pt_regs(task), offset);
311 static int set_segment_reg(struct task_struct *task,
312 unsigned long offset, u16 value)
315 * The value argument was already truncated to 16 bits.
317 if (invalid_selector(value))
321 case offsetof(struct user_regs_struct,fs):
323 * If this is setting fs as for normal 64-bit use but
324 * setting fs_base has implicitly changed it, leave it.
326 if ((value == FS_TLS_SEL && task->thread.fsindex == 0 &&
327 task->thread.fs != 0) ||
328 (value == 0 && task->thread.fsindex == FS_TLS_SEL &&
329 task->thread.fs == 0))
331 task->thread.fsindex = value;
333 loadsegment(fs, task->thread.fsindex);
335 case offsetof(struct user_regs_struct,gs):
337 * If this is setting gs as for normal 64-bit use but
338 * setting gs_base has implicitly changed it, leave it.
340 if ((value == GS_TLS_SEL && task->thread.gsindex == 0 &&
341 task->thread.gs != 0) ||
342 (value == 0 && task->thread.gsindex == GS_TLS_SEL &&
343 task->thread.gs == 0))
345 task->thread.gsindex = value;
347 load_gs_index(task->thread.gsindex);
349 case offsetof(struct user_regs_struct,ds):
350 task->thread.ds = value;
352 loadsegment(ds, task->thread.ds);
354 case offsetof(struct user_regs_struct,es):
355 task->thread.es = value;
357 loadsegment(es, task->thread.es);
361 * Can't actually change these in 64-bit mode.
363 case offsetof(struct user_regs_struct,cs):
364 if (unlikely(value == 0))
366 #ifdef CONFIG_IA32_EMULATION
367 if (test_tsk_thread_flag(task, TIF_IA32))
368 task_pt_regs(task)->cs = value;
371 case offsetof(struct user_regs_struct,ss):
372 if (unlikely(value == 0))
374 #ifdef CONFIG_IA32_EMULATION
375 if (test_tsk_thread_flag(task, TIF_IA32))
376 task_pt_regs(task)->ss = value;
384 #endif /* CONFIG_X86_32 */
386 static unsigned long get_flags(struct task_struct *task)
388 unsigned long retval = task_pt_regs(task)->flags;
391 * If the debugger set TF, hide it from the readout.
393 if (test_tsk_thread_flag(task, TIF_FORCED_TF))
394 retval &= ~X86_EFLAGS_TF;
399 static int set_flags(struct task_struct *task, unsigned long value)
401 struct pt_regs *regs = task_pt_regs(task);
404 * If the user value contains TF, mark that
405 * it was not "us" (the debugger) that set it.
406 * If not, make sure it stays set if we had.
408 if (value & X86_EFLAGS_TF)
409 clear_tsk_thread_flag(task, TIF_FORCED_TF);
410 else if (test_tsk_thread_flag(task, TIF_FORCED_TF))
411 value |= X86_EFLAGS_TF;
413 regs->flags = (regs->flags & ~FLAG_MASK) | (value & FLAG_MASK);
418 static int putreg(struct task_struct *child,
419 unsigned long offset, unsigned long value)
422 case offsetof(struct user_regs_struct, cs):
423 case offsetof(struct user_regs_struct, ds):
424 case offsetof(struct user_regs_struct, es):
425 case offsetof(struct user_regs_struct, fs):
426 case offsetof(struct user_regs_struct, gs):
427 case offsetof(struct user_regs_struct, ss):
428 return set_segment_reg(child, offset, value);
430 case offsetof(struct user_regs_struct, flags):
431 return set_flags(child, value);
434 case offsetof(struct user_regs_struct,fs_base):
435 if (value >= TASK_SIZE_OF(child))
438 * When changing the segment base, use do_arch_prctl
439 * to set either thread.fs or thread.fsindex and the
440 * corresponding GDT slot.
442 if (child->thread.fs != value)
443 return do_arch_prctl(child, ARCH_SET_FS, value);
445 case offsetof(struct user_regs_struct,gs_base):
447 * Exactly the same here as the %fs handling above.
449 if (value >= TASK_SIZE_OF(child))
451 if (child->thread.gs != value)
452 return do_arch_prctl(child, ARCH_SET_GS, value);
457 *pt_regs_access(task_pt_regs(child), offset) = value;
461 static unsigned long getreg(struct task_struct *task, unsigned long offset)
464 case offsetof(struct user_regs_struct, cs):
465 case offsetof(struct user_regs_struct, ds):
466 case offsetof(struct user_regs_struct, es):
467 case offsetof(struct user_regs_struct, fs):
468 case offsetof(struct user_regs_struct, gs):
469 case offsetof(struct user_regs_struct, ss):
470 return get_segment_reg(task, offset);
472 case offsetof(struct user_regs_struct, flags):
473 return get_flags(task);
476 case offsetof(struct user_regs_struct, fs_base): {
478 * do_arch_prctl may have used a GDT slot instead of
479 * the MSR. To userland, it appears the same either
480 * way, except the %fs segment selector might not be 0.
482 unsigned int seg = task->thread.fsindex;
483 if (task->thread.fs != 0)
484 return task->thread.fs;
486 asm("movl %%fs,%0" : "=r" (seg));
487 if (seg != FS_TLS_SEL)
489 return get_desc_base(&task->thread.tls_array[FS_TLS]);
491 case offsetof(struct user_regs_struct, gs_base): {
493 * Exactly the same here as the %fs handling above.
495 unsigned int seg = task->thread.gsindex;
496 if (task->thread.gs != 0)
497 return task->thread.gs;
499 asm("movl %%gs,%0" : "=r" (seg));
500 if (seg != GS_TLS_SEL)
502 return get_desc_base(&task->thread.tls_array[GS_TLS]);
507 return *pt_regs_access(task_pt_regs(task), offset);
510 static int genregs_get(struct task_struct *target,
511 const struct user_regset *regset,
512 unsigned int pos, unsigned int count,
513 void *kbuf, void __user *ubuf)
516 unsigned long *k = kbuf;
517 while (count >= sizeof(*k)) {
518 *k++ = getreg(target, pos);
523 unsigned long __user *u = ubuf;
524 while (count >= sizeof(*u)) {
525 if (__put_user(getreg(target, pos), u++))
535 static int genregs_set(struct task_struct *target,
536 const struct user_regset *regset,
537 unsigned int pos, unsigned int count,
538 const void *kbuf, const void __user *ubuf)
542 const unsigned long *k = kbuf;
543 while (count >= sizeof(*k) && !ret) {
544 ret = putreg(target, pos, *k++);
549 const unsigned long __user *u = ubuf;
550 while (count >= sizeof(*u) && !ret) {
552 ret = __get_user(word, u++);
555 ret = putreg(target, pos, word);
563 static void ptrace_triggered(struct perf_event *bp,
564 struct perf_sample_data *data,
565 struct pt_regs *regs)
568 struct thread_struct *thread = &(current->thread);
571 * Store in the virtual DR6 register the fact that the breakpoint
572 * was hit so the thread's debugger will see it.
574 for (i = 0; i < HBP_NUM; i++) {
575 if (thread->ptrace_bps[i] == bp)
579 thread->debugreg6 |= (DR_TRAP0 << i);
583 * Walk through every ptrace breakpoints for this thread and
584 * build the dr7 value on top of their attributes.
587 static unsigned long ptrace_get_dr7(struct perf_event *bp[])
591 struct arch_hw_breakpoint *info;
593 for (i = 0; i < HBP_NUM; i++) {
594 if (bp[i] && !bp[i]->attr.disabled) {
595 info = counter_arch_bp(bp[i]);
596 dr7 |= encode_dr7(i, info->len, info->type);
604 ptrace_modify_breakpoint(struct perf_event *bp, int len, int type,
605 struct task_struct *tsk, int disabled)
608 int gen_len, gen_type;
609 struct perf_event_attr attr;
612 * We should have at least an inactive breakpoint at this
613 * slot. It means the user is writing dr7 without having
614 * written the address register first
619 err = arch_bp_generic_fields(len, type, &gen_len, &gen_type);
624 attr.bp_len = gen_len;
625 attr.bp_type = gen_type;
626 attr.disabled = disabled;
628 return modify_user_hw_breakpoint(bp, &attr);
632 * Handle ptrace writes to debug register 7.
634 static int ptrace_write_dr7(struct task_struct *tsk, unsigned long data)
636 struct thread_struct *thread = &(tsk->thread);
637 unsigned long old_dr7;
638 int i, orig_ret = 0, rc = 0;
639 int enabled, second_pass = 0;
641 struct perf_event *bp;
643 if (ptrace_get_breakpoints(tsk) < 0)
646 data &= ~DR_CONTROL_RESERVED;
647 old_dr7 = ptrace_get_dr7(thread->ptrace_bps);
650 * Loop through all the hardware breakpoints, making the
651 * appropriate changes to each.
653 for (i = 0; i < HBP_NUM; i++) {
654 enabled = decode_dr7(data, i, &len, &type);
655 bp = thread->ptrace_bps[i];
660 * Don't unregister the breakpoints right-away,
661 * unless all register_user_hw_breakpoint()
662 * requests have succeeded. This prevents
663 * any window of opportunity for debug
664 * register grabbing by other users.
669 rc = ptrace_modify_breakpoint(bp, len, type,
677 rc = ptrace_modify_breakpoint(bp, len, type, tsk, 0);
682 * Make a second pass to free the remaining unused breakpoints
683 * or to restore the original breakpoints if an error occurred.
694 ptrace_put_breakpoints(tsk);
696 return ((orig_ret < 0) ? orig_ret : rc);
700 * Handle PTRACE_PEEKUSR calls for the debug register area.
702 static unsigned long ptrace_get_debugreg(struct task_struct *tsk, int n)
704 struct thread_struct *thread = &(tsk->thread);
705 unsigned long val = 0;
708 struct perf_event *bp;
710 if (ptrace_get_breakpoints(tsk) < 0)
713 bp = thread->ptrace_bps[n];
717 val = bp->hw.info.address;
719 ptrace_put_breakpoints(tsk);
721 val = thread->debugreg6;
723 val = thread->ptrace_dr7;
728 static int ptrace_set_breakpoint_addr(struct task_struct *tsk, int nr,
731 struct perf_event *bp;
732 struct thread_struct *t = &tsk->thread;
733 struct perf_event_attr attr;
736 if (ptrace_get_breakpoints(tsk) < 0)
739 if (!t->ptrace_bps[nr]) {
740 ptrace_breakpoint_init(&attr);
742 * Put stub len and type to register (reserve) an inactive but
746 attr.bp_len = HW_BREAKPOINT_LEN_1;
747 attr.bp_type = HW_BREAKPOINT_W;
750 bp = register_user_hw_breakpoint(&attr, ptrace_triggered,
754 * CHECKME: the previous code returned -EIO if the addr wasn't
755 * a valid task virtual addr. The new one will return -EINVAL in
757 * -EINVAL may be what we want for in-kernel breakpoints users,
758 * but -EIO looks better for ptrace, since we refuse a register
759 * writing for the user. And anyway this is the previous
767 t->ptrace_bps[nr] = bp;
769 bp = t->ptrace_bps[nr];
773 err = modify_user_hw_breakpoint(bp, &attr);
777 ptrace_put_breakpoints(tsk);
782 * Handle PTRACE_POKEUSR calls for the debug register area.
784 static int ptrace_set_debugreg(struct task_struct *tsk, int n,
787 struct thread_struct *thread = &(tsk->thread);
790 /* There are no DR4 or DR5 registers */
791 if (n == 4 || n == 5)
795 thread->debugreg6 = val;
799 rc = ptrace_set_breakpoint_addr(tsk, n, val);
803 /* All that's left is DR7 */
805 rc = ptrace_write_dr7(tsk, val);
807 thread->ptrace_dr7 = val;
815 * These access the current or another (stopped) task's io permission
816 * bitmap for debugging or core dump.
818 static int ioperm_active(struct task_struct *target,
819 const struct user_regset *regset)
821 return target->thread.io_bitmap_max / regset->size;
824 static int ioperm_get(struct task_struct *target,
825 const struct user_regset *regset,
826 unsigned int pos, unsigned int count,
827 void *kbuf, void __user *ubuf)
829 if (!target->thread.io_bitmap_ptr)
832 return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
833 target->thread.io_bitmap_ptr,
838 * Called by kernel/ptrace.c when detaching..
840 * Make sure the single step bit is not set.
842 void ptrace_disable(struct task_struct *child)
844 user_disable_single_step(child);
845 #ifdef TIF_SYSCALL_EMU
846 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
850 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
851 static const struct user_regset_view user_x86_32_view; /* Initialized below. */
854 long arch_ptrace(struct task_struct *child, long request,
855 unsigned long addr, unsigned long data)
858 unsigned long __user *datap = (unsigned long __user *)data;
861 /* read the word at location addr in the USER area. */
862 case PTRACE_PEEKUSR: {
866 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user))
869 tmp = 0; /* Default return condition */
870 if (addr < sizeof(struct user_regs_struct))
871 tmp = getreg(child, addr);
872 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
873 addr <= offsetof(struct user, u_debugreg[7])) {
874 addr -= offsetof(struct user, u_debugreg[0]);
875 tmp = ptrace_get_debugreg(child, addr / sizeof(data));
877 ret = put_user(tmp, datap);
881 case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
883 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user))
886 if (addr < sizeof(struct user_regs_struct))
887 ret = putreg(child, addr, data);
888 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
889 addr <= offsetof(struct user, u_debugreg[7])) {
890 addr -= offsetof(struct user, u_debugreg[0]);
891 ret = ptrace_set_debugreg(child,
892 addr / sizeof(data), data);
896 case PTRACE_GETREGS: /* Get all gp regs from the child. */
897 return copy_regset_to_user(child,
898 task_user_regset_view(current),
900 0, sizeof(struct user_regs_struct),
903 case PTRACE_SETREGS: /* Set all gp regs in the child. */
904 return copy_regset_from_user(child,
905 task_user_regset_view(current),
907 0, sizeof(struct user_regs_struct),
910 case PTRACE_GETFPREGS: /* Get the child FPU state. */
911 return copy_regset_to_user(child,
912 task_user_regset_view(current),
914 0, sizeof(struct user_i387_struct),
917 case PTRACE_SETFPREGS: /* Set the child FPU state. */
918 return copy_regset_from_user(child,
919 task_user_regset_view(current),
921 0, sizeof(struct user_i387_struct),
925 case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
926 return copy_regset_to_user(child, &user_x86_32_view,
928 0, sizeof(struct user_fxsr_struct),
931 case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
932 return copy_regset_from_user(child, &user_x86_32_view,
934 0, sizeof(struct user_fxsr_struct),
938 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
939 case PTRACE_GET_THREAD_AREA:
942 ret = do_get_thread_area(child, addr,
943 (struct user_desc __user *)data);
946 case PTRACE_SET_THREAD_AREA:
949 ret = do_set_thread_area(child, addr,
950 (struct user_desc __user *)data, 0);
955 /* normal 64bit interface to access TLS data.
956 Works just like arch_prctl, except that the arguments
958 case PTRACE_ARCH_PRCTL:
959 ret = do_arch_prctl(child, data, addr);
964 ret = ptrace_request(child, request, addr, data);
971 #ifdef CONFIG_IA32_EMULATION
973 #include <linux/compat.h>
974 #include <linux/syscalls.h>
975 #include <asm/ia32.h>
976 #include <asm/user32.h>
979 case offsetof(struct user32, regs.l): \
980 regs->q = value; break
983 case offsetof(struct user32, regs.rs): \
984 return set_segment_reg(child, \
985 offsetof(struct user_regs_struct, rs), \
989 static int putreg32(struct task_struct *child, unsigned regno, u32 value)
991 struct pt_regs *regs = task_pt_regs(child);
1012 case offsetof(struct user32, regs.orig_eax):
1014 * A 32-bit debugger setting orig_eax means to restore
1015 * the state of the task restarting a 32-bit syscall.
1016 * Make sure we interpret the -ERESTART* codes correctly
1017 * in case the task is not actually still sitting at the
1018 * exit from a 32-bit syscall with TS_COMPAT still set.
1020 regs->orig_ax = value;
1021 if (syscall_get_nr(child, regs) >= 0)
1022 task_thread_info(child)->status |= TS_COMPAT;
1025 case offsetof(struct user32, regs.eflags):
1026 return set_flags(child, value);
1028 case offsetof(struct user32, u_debugreg[0]) ...
1029 offsetof(struct user32, u_debugreg[7]):
1030 regno -= offsetof(struct user32, u_debugreg[0]);
1031 return ptrace_set_debugreg(child, regno / 4, value);
1034 if (regno > sizeof(struct user32) || (regno & 3))
1038 * Other dummy fields in the virtual user structure
1050 case offsetof(struct user32, regs.l): \
1051 *val = regs->q; break
1054 case offsetof(struct user32, regs.rs): \
1055 *val = get_segment_reg(child, \
1056 offsetof(struct user_regs_struct, rs)); \
1059 static int getreg32(struct task_struct *child, unsigned regno, u32 *val)
1061 struct pt_regs *regs = task_pt_regs(child);
1079 R32(orig_eax, orig_ax);
1083 case offsetof(struct user32, regs.eflags):
1084 *val = get_flags(child);
1087 case offsetof(struct user32, u_debugreg[0]) ...
1088 offsetof(struct user32, u_debugreg[7]):
1089 regno -= offsetof(struct user32, u_debugreg[0]);
1090 *val = ptrace_get_debugreg(child, regno / 4);
1094 if (regno > sizeof(struct user32) || (regno & 3))
1098 * Other dummy fields in the virtual user structure
1110 static int genregs32_get(struct task_struct *target,
1111 const struct user_regset *regset,
1112 unsigned int pos, unsigned int count,
1113 void *kbuf, void __user *ubuf)
1116 compat_ulong_t *k = kbuf;
1117 while (count >= sizeof(*k)) {
1118 getreg32(target, pos, k++);
1119 count -= sizeof(*k);
1123 compat_ulong_t __user *u = ubuf;
1124 while (count >= sizeof(*u)) {
1125 compat_ulong_t word;
1126 getreg32(target, pos, &word);
1127 if (__put_user(word, u++))
1129 count -= sizeof(*u);
1137 static int genregs32_set(struct task_struct *target,
1138 const struct user_regset *regset,
1139 unsigned int pos, unsigned int count,
1140 const void *kbuf, const void __user *ubuf)
1144 const compat_ulong_t *k = kbuf;
1145 while (count >= sizeof(*k) && !ret) {
1146 ret = putreg32(target, pos, *k++);
1147 count -= sizeof(*k);
1151 const compat_ulong_t __user *u = ubuf;
1152 while (count >= sizeof(*u) && !ret) {
1153 compat_ulong_t word;
1154 ret = __get_user(word, u++);
1157 ret = putreg32(target, pos, word);
1158 count -= sizeof(*u);
1165 #ifdef CONFIG_X86_X32_ABI
1166 static long x32_arch_ptrace(struct task_struct *child,
1167 compat_long_t request, compat_ulong_t caddr,
1168 compat_ulong_t cdata)
1170 unsigned long addr = caddr;
1171 unsigned long data = cdata;
1172 void __user *datap = compat_ptr(data);
1176 /* Read 32bits at location addr in the USER area. Only allow
1177 to return the lower 32bits of segment and debug registers. */
1178 case PTRACE_PEEKUSR: {
1182 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user) ||
1183 addr < offsetof(struct user_regs_struct, cs))
1186 tmp = 0; /* Default return condition */
1187 if (addr < sizeof(struct user_regs_struct))
1188 tmp = getreg(child, addr);
1189 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
1190 addr <= offsetof(struct user, u_debugreg[7])) {
1191 addr -= offsetof(struct user, u_debugreg[0]);
1192 tmp = ptrace_get_debugreg(child, addr / sizeof(data));
1194 ret = put_user(tmp, (__u32 __user *)datap);
1198 /* Write the word at location addr in the USER area. Only allow
1199 to update segment and debug registers with the upper 32bits
1201 case PTRACE_POKEUSR:
1203 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user) ||
1204 addr < offsetof(struct user_regs_struct, cs))
1207 if (addr < sizeof(struct user_regs_struct))
1208 ret = putreg(child, addr, data);
1209 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
1210 addr <= offsetof(struct user, u_debugreg[7])) {
1211 addr -= offsetof(struct user, u_debugreg[0]);
1212 ret = ptrace_set_debugreg(child,
1213 addr / sizeof(data), data);
1217 case PTRACE_GETREGS: /* Get all gp regs from the child. */
1218 return copy_regset_to_user(child,
1219 task_user_regset_view(current),
1221 0, sizeof(struct user_regs_struct),
1224 case PTRACE_SETREGS: /* Set all gp regs in the child. */
1225 return copy_regset_from_user(child,
1226 task_user_regset_view(current),
1228 0, sizeof(struct user_regs_struct),
1231 case PTRACE_GETFPREGS: /* Get the child FPU state. */
1232 return copy_regset_to_user(child,
1233 task_user_regset_view(current),
1235 0, sizeof(struct user_i387_struct),
1238 case PTRACE_SETFPREGS: /* Set the child FPU state. */
1239 return copy_regset_from_user(child,
1240 task_user_regset_view(current),
1242 0, sizeof(struct user_i387_struct),
1246 return compat_ptrace_request(child, request, addr, data);
1253 long compat_arch_ptrace(struct task_struct *child, compat_long_t request,
1254 compat_ulong_t caddr, compat_ulong_t cdata)
1256 unsigned long addr = caddr;
1257 unsigned long data = cdata;
1258 void __user *datap = compat_ptr(data);
1262 #ifdef CONFIG_X86_X32_ABI
1263 if (!is_ia32_task())
1264 return x32_arch_ptrace(child, request, caddr, cdata);
1268 case PTRACE_PEEKUSR:
1269 ret = getreg32(child, addr, &val);
1271 ret = put_user(val, (__u32 __user *)datap);
1274 case PTRACE_POKEUSR:
1275 ret = putreg32(child, addr, data);
1278 case PTRACE_GETREGS: /* Get all gp regs from the child. */
1279 return copy_regset_to_user(child, &user_x86_32_view,
1281 0, sizeof(struct user_regs_struct32),
1284 case PTRACE_SETREGS: /* Set all gp regs in the child. */
1285 return copy_regset_from_user(child, &user_x86_32_view,
1287 sizeof(struct user_regs_struct32),
1290 case PTRACE_GETFPREGS: /* Get the child FPU state. */
1291 return copy_regset_to_user(child, &user_x86_32_view,
1293 sizeof(struct user_i387_ia32_struct),
1296 case PTRACE_SETFPREGS: /* Set the child FPU state. */
1297 return copy_regset_from_user(
1298 child, &user_x86_32_view, REGSET_FP,
1299 0, sizeof(struct user_i387_ia32_struct), datap);
1301 case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
1302 return copy_regset_to_user(child, &user_x86_32_view,
1304 sizeof(struct user32_fxsr_struct),
1307 case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
1308 return copy_regset_from_user(child, &user_x86_32_view,
1310 sizeof(struct user32_fxsr_struct),
1313 case PTRACE_GET_THREAD_AREA:
1314 case PTRACE_SET_THREAD_AREA:
1315 return arch_ptrace(child, request, addr, data);
1318 return compat_ptrace_request(child, request, addr, data);
1324 #endif /* CONFIG_IA32_EMULATION */
1326 #ifdef CONFIG_X86_64
1328 static struct user_regset x86_64_regsets[] __read_mostly = {
1329 [REGSET_GENERAL] = {
1330 .core_note_type = NT_PRSTATUS,
1331 .n = sizeof(struct user_regs_struct) / sizeof(long),
1332 .size = sizeof(long), .align = sizeof(long),
1333 .get = genregs_get, .set = genregs_set
1336 .core_note_type = NT_PRFPREG,
1337 .n = sizeof(struct user_i387_struct) / sizeof(long),
1338 .size = sizeof(long), .align = sizeof(long),
1339 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1342 .core_note_type = NT_X86_XSTATE,
1343 .size = sizeof(u64), .align = sizeof(u64),
1344 .active = xstateregs_active, .get = xstateregs_get,
1345 .set = xstateregs_set
1347 [REGSET_IOPERM64] = {
1348 .core_note_type = NT_386_IOPERM,
1349 .n = IO_BITMAP_LONGS,
1350 .size = sizeof(long), .align = sizeof(long),
1351 .active = ioperm_active, .get = ioperm_get
1355 static const struct user_regset_view user_x86_64_view = {
1356 .name = "x86_64", .e_machine = EM_X86_64,
1357 .regsets = x86_64_regsets, .n = ARRAY_SIZE(x86_64_regsets)
1360 #else /* CONFIG_X86_32 */
1362 #define user_regs_struct32 user_regs_struct
1363 #define genregs32_get genregs_get
1364 #define genregs32_set genregs_set
1366 #endif /* CONFIG_X86_64 */
1368 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1369 static struct user_regset x86_32_regsets[] __read_mostly = {
1370 [REGSET_GENERAL] = {
1371 .core_note_type = NT_PRSTATUS,
1372 .n = sizeof(struct user_regs_struct32) / sizeof(u32),
1373 .size = sizeof(u32), .align = sizeof(u32),
1374 .get = genregs32_get, .set = genregs32_set
1377 .core_note_type = NT_PRFPREG,
1378 .n = sizeof(struct user_i387_ia32_struct) / sizeof(u32),
1379 .size = sizeof(u32), .align = sizeof(u32),
1380 .active = fpregs_active, .get = fpregs_get, .set = fpregs_set
1383 .core_note_type = NT_PRXFPREG,
1384 .n = sizeof(struct user32_fxsr_struct) / sizeof(u32),
1385 .size = sizeof(u32), .align = sizeof(u32),
1386 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1389 .core_note_type = NT_X86_XSTATE,
1390 .size = sizeof(u64), .align = sizeof(u64),
1391 .active = xstateregs_active, .get = xstateregs_get,
1392 .set = xstateregs_set
1395 .core_note_type = NT_386_TLS,
1396 .n = GDT_ENTRY_TLS_ENTRIES, .bias = GDT_ENTRY_TLS_MIN,
1397 .size = sizeof(struct user_desc),
1398 .align = sizeof(struct user_desc),
1399 .active = regset_tls_active,
1400 .get = regset_tls_get, .set = regset_tls_set
1402 [REGSET_IOPERM32] = {
1403 .core_note_type = NT_386_IOPERM,
1404 .n = IO_BITMAP_BYTES / sizeof(u32),
1405 .size = sizeof(u32), .align = sizeof(u32),
1406 .active = ioperm_active, .get = ioperm_get
1410 static const struct user_regset_view user_x86_32_view = {
1411 .name = "i386", .e_machine = EM_386,
1412 .regsets = x86_32_regsets, .n = ARRAY_SIZE(x86_32_regsets)
1417 * This represents bytes 464..511 in the memory layout exported through
1418 * the REGSET_XSTATE interface.
1420 u64 xstate_fx_sw_bytes[USER_XSTATE_FX_SW_WORDS];
1422 void update_regset_xstate_info(unsigned int size, u64 xstate_mask)
1424 #ifdef CONFIG_X86_64
1425 x86_64_regsets[REGSET_XSTATE].n = size / sizeof(u64);
1427 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1428 x86_32_regsets[REGSET_XSTATE].n = size / sizeof(u64);
1430 xstate_fx_sw_bytes[USER_XSTATE_XCR0_WORD] = xstate_mask;
1433 const struct user_regset_view *task_user_regset_view(struct task_struct *task)
1435 #ifdef CONFIG_IA32_EMULATION
1436 if (test_tsk_thread_flag(task, TIF_IA32))
1438 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1439 return &user_x86_32_view;
1441 #ifdef CONFIG_X86_64
1442 return &user_x86_64_view;
1446 static void fill_sigtrap_info(struct task_struct *tsk,
1447 struct pt_regs *regs,
1448 int error_code, int si_code,
1449 struct siginfo *info)
1451 tsk->thread.trap_nr = X86_TRAP_DB;
1452 tsk->thread.error_code = error_code;
1454 memset(info, 0, sizeof(*info));
1455 info->si_signo = SIGTRAP;
1456 info->si_code = si_code;
1457 info->si_addr = user_mode_vm(regs) ? (void __user *)regs->ip : NULL;
1460 void user_single_step_siginfo(struct task_struct *tsk,
1461 struct pt_regs *regs,
1462 struct siginfo *info)
1464 fill_sigtrap_info(tsk, regs, 0, TRAP_BRKPT, info);
1467 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs,
1468 int error_code, int si_code)
1470 struct siginfo info;
1472 fill_sigtrap_info(tsk, regs, error_code, si_code, &info);
1473 /* Send us the fake SIGTRAP */
1474 force_sig_info(SIGTRAP, &info, tsk);
1478 #ifdef CONFIG_X86_32
1480 #elif defined CONFIG_IA32_EMULATION
1481 # define IS_IA32 is_compat_task()
1487 * We must return the syscall number to actually look up in the table.
1488 * This can be -1L to skip running any syscall at all.
1490 long syscall_trace_enter(struct pt_regs *regs)
1497 * If we stepped into a sysenter/syscall insn, it trapped in
1498 * kernel mode; do_debug() cleared TF and set TIF_SINGLESTEP.
1499 * If user-mode had set TF itself, then it's still clear from
1500 * do_debug() and we need to set it again to restore the user
1501 * state. If we entered on the slow path, TF was already set.
1503 if (test_thread_flag(TIF_SINGLESTEP))
1504 regs->flags |= X86_EFLAGS_TF;
1506 /* do the secure computing check first */
1507 if (secure_computing(regs->orig_ax)) {
1508 /* seccomp failures shouldn't expose any additional code. */
1513 if (unlikely(test_thread_flag(TIF_SYSCALL_EMU)))
1516 if ((ret || test_thread_flag(TIF_SYSCALL_TRACE)) &&
1517 tracehook_report_syscall_entry(regs))
1520 if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
1521 trace_sys_enter(regs, regs->orig_ax);
1524 audit_syscall_entry(AUDIT_ARCH_I386,
1527 regs->dx, regs->si);
1528 #ifdef CONFIG_X86_64
1530 audit_syscall_entry(AUDIT_ARCH_X86_64,
1533 regs->dx, regs->r10);
1537 return ret ?: regs->orig_ax;
1540 void syscall_trace_leave(struct pt_regs *regs)
1544 audit_syscall_exit(regs);
1546 if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
1547 trace_sys_exit(regs, regs->ax);
1550 * If TIF_SYSCALL_EMU is set, we only get here because of
1551 * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
1552 * We already reported this syscall instruction in
1553 * syscall_trace_enter().
1555 step = unlikely(test_thread_flag(TIF_SINGLESTEP)) &&
1556 !test_thread_flag(TIF_SYSCALL_EMU);
1557 if (step || test_thread_flag(TIF_SYSCALL_TRACE))
1558 tracehook_report_syscall_exit(regs, step);