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/export.h>
26 #include <linux/context_tracking.h>
28 #include <asm/uaccess.h>
29 #include <asm/pgtable.h>
30 #include <asm/processor.h>
32 #include <asm/fpu-internal.h>
33 #include <asm/debugreg.h>
36 #include <asm/prctl.h>
37 #include <asm/proto.h>
38 #include <asm/hw_breakpoint.h>
39 #include <asm/traps.h>
43 #define CREATE_TRACE_POINTS
44 #include <trace/events/syscalls.h>
50 REGSET_IOPERM64 = REGSET_XFP,
56 struct pt_regs_offset {
61 #define REG_OFFSET_NAME(r) {.name = #r, .offset = offsetof(struct pt_regs, r)}
62 #define REG_OFFSET_END {.name = NULL, .offset = 0}
64 static const struct pt_regs_offset regoffset_table[] = {
88 REG_OFFSET_NAME(orig_ax),
91 REG_OFFSET_NAME(flags),
98 * regs_query_register_offset() - query register offset from its name
99 * @name: the name of a register
101 * regs_query_register_offset() returns the offset of a register in struct
102 * pt_regs from its name. If the name is invalid, this returns -EINVAL;
104 int regs_query_register_offset(const char *name)
106 const struct pt_regs_offset *roff;
107 for (roff = regoffset_table; roff->name != NULL; roff++)
108 if (!strcmp(roff->name, name))
114 * regs_query_register_name() - query register name from its offset
115 * @offset: the offset of a register in struct pt_regs.
117 * regs_query_register_name() returns the name of a register from its
118 * offset in struct pt_regs. If the @offset is invalid, this returns NULL;
120 const char *regs_query_register_name(unsigned int offset)
122 const struct pt_regs_offset *roff;
123 for (roff = regoffset_table; roff->name != NULL; roff++)
124 if (roff->offset == offset)
129 static const int arg_offs_table[] = {
131 [0] = offsetof(struct pt_regs, ax),
132 [1] = offsetof(struct pt_regs, dx),
133 [2] = offsetof(struct pt_regs, cx)
134 #else /* CONFIG_X86_64 */
135 [0] = offsetof(struct pt_regs, di),
136 [1] = offsetof(struct pt_regs, si),
137 [2] = offsetof(struct pt_regs, dx),
138 [3] = offsetof(struct pt_regs, cx),
139 [4] = offsetof(struct pt_regs, r8),
140 [5] = offsetof(struct pt_regs, r9)
145 * does not yet catch signals sent when the child dies.
146 * in exit.c or in signal.c.
150 * Determines which flags the user has access to [1 = access, 0 = no access].
152 #define FLAG_MASK_32 ((unsigned long) \
153 (X86_EFLAGS_CF | X86_EFLAGS_PF | \
154 X86_EFLAGS_AF | X86_EFLAGS_ZF | \
155 X86_EFLAGS_SF | X86_EFLAGS_TF | \
156 X86_EFLAGS_DF | X86_EFLAGS_OF | \
157 X86_EFLAGS_RF | X86_EFLAGS_AC))
160 * Determines whether a value may be installed in a segment register.
162 static inline bool invalid_selector(u16 value)
164 return unlikely(value != 0 && (value & SEGMENT_RPL_MASK) != USER_RPL);
169 #define FLAG_MASK FLAG_MASK_32
172 * X86_32 CPUs don't save ss and esp if the CPU is already in kernel mode
173 * when it traps. The previous stack will be directly underneath the saved
174 * registers, and 'sp/ss' won't even have been saved. Thus the '®s->sp'.
176 * Now, if the stack is empty, '®s->sp' is out of range. In this
177 * case we try to take the previous stack. To always return a non-null
178 * stack pointer we fall back to regs as stack if no previous stack
181 * This is valid only for kernel mode traps.
183 unsigned long kernel_stack_pointer(struct pt_regs *regs)
185 unsigned long context = (unsigned long)regs & ~(THREAD_SIZE - 1);
186 unsigned long sp = (unsigned long)®s->sp;
187 struct thread_info *tinfo;
189 if (context == (sp & ~(THREAD_SIZE - 1)))
192 tinfo = (struct thread_info *)context;
193 if (tinfo->previous_esp)
194 return tinfo->previous_esp;
196 return (unsigned long)regs;
198 EXPORT_SYMBOL_GPL(kernel_stack_pointer);
200 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long regno)
202 BUILD_BUG_ON(offsetof(struct pt_regs, bx) != 0);
203 return ®s->bx + (regno >> 2);
206 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
209 * Returning the value truncates it to 16 bits.
212 if (offset != offsetof(struct user_regs_struct, gs))
213 retval = *pt_regs_access(task_pt_regs(task), offset);
216 retval = get_user_gs(task_pt_regs(task));
218 retval = task_user_gs(task);
223 static int set_segment_reg(struct task_struct *task,
224 unsigned long offset, u16 value)
227 * The value argument was already truncated to 16 bits.
229 if (invalid_selector(value))
233 * For %cs and %ss we cannot permit a null selector.
234 * We can permit a bogus selector as long as it has USER_RPL.
235 * Null selectors are fine for other segment registers, but
236 * we will never get back to user mode with invalid %cs or %ss
237 * and will take the trap in iret instead. Much code relies
238 * on user_mode() to distinguish a user trap frame (which can
239 * safely use invalid selectors) from a kernel trap frame.
242 case offsetof(struct user_regs_struct, cs):
243 case offsetof(struct user_regs_struct, ss):
244 if (unlikely(value == 0))
248 *pt_regs_access(task_pt_regs(task), offset) = value;
251 case offsetof(struct user_regs_struct, gs):
253 set_user_gs(task_pt_regs(task), value);
255 task_user_gs(task) = value;
261 #else /* CONFIG_X86_64 */
263 #define FLAG_MASK (FLAG_MASK_32 | X86_EFLAGS_NT)
265 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long offset)
267 BUILD_BUG_ON(offsetof(struct pt_regs, r15) != 0);
268 return ®s->r15 + (offset / sizeof(regs->r15));
271 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
274 * Returning the value truncates it to 16 bits.
279 case offsetof(struct user_regs_struct, fs):
280 if (task == current) {
281 /* Older gas can't assemble movq %?s,%r?? */
282 asm("movl %%fs,%0" : "=r" (seg));
285 return task->thread.fsindex;
286 case offsetof(struct user_regs_struct, gs):
287 if (task == current) {
288 asm("movl %%gs,%0" : "=r" (seg));
291 return task->thread.gsindex;
292 case offsetof(struct user_regs_struct, ds):
293 if (task == current) {
294 asm("movl %%ds,%0" : "=r" (seg));
297 return task->thread.ds;
298 case offsetof(struct user_regs_struct, es):
299 if (task == current) {
300 asm("movl %%es,%0" : "=r" (seg));
303 return task->thread.es;
305 case offsetof(struct user_regs_struct, cs):
306 case offsetof(struct user_regs_struct, ss):
309 return *pt_regs_access(task_pt_regs(task), offset);
312 static int set_segment_reg(struct task_struct *task,
313 unsigned long offset, u16 value)
316 * The value argument was already truncated to 16 bits.
318 if (invalid_selector(value))
322 case offsetof(struct user_regs_struct,fs):
324 * If this is setting fs as for normal 64-bit use but
325 * setting fs_base has implicitly changed it, leave it.
327 if ((value == FS_TLS_SEL && task->thread.fsindex == 0 &&
328 task->thread.fs != 0) ||
329 (value == 0 && task->thread.fsindex == FS_TLS_SEL &&
330 task->thread.fs == 0))
332 task->thread.fsindex = value;
334 loadsegment(fs, task->thread.fsindex);
336 case offsetof(struct user_regs_struct,gs):
338 * If this is setting gs as for normal 64-bit use but
339 * setting gs_base has implicitly changed it, leave it.
341 if ((value == GS_TLS_SEL && task->thread.gsindex == 0 &&
342 task->thread.gs != 0) ||
343 (value == 0 && task->thread.gsindex == GS_TLS_SEL &&
344 task->thread.gs == 0))
346 task->thread.gsindex = value;
348 load_gs_index(task->thread.gsindex);
350 case offsetof(struct user_regs_struct,ds):
351 task->thread.ds = value;
353 loadsegment(ds, task->thread.ds);
355 case offsetof(struct user_regs_struct,es):
356 task->thread.es = value;
358 loadsegment(es, task->thread.es);
362 * Can't actually change these in 64-bit mode.
364 case offsetof(struct user_regs_struct,cs):
365 if (unlikely(value == 0))
367 #ifdef CONFIG_IA32_EMULATION
368 if (test_tsk_thread_flag(task, TIF_IA32))
369 task_pt_regs(task)->cs = value;
372 case offsetof(struct user_regs_struct,ss):
373 if (unlikely(value == 0))
375 #ifdef CONFIG_IA32_EMULATION
376 if (test_tsk_thread_flag(task, TIF_IA32))
377 task_pt_regs(task)->ss = value;
385 #endif /* CONFIG_X86_32 */
387 static unsigned long get_flags(struct task_struct *task)
389 unsigned long retval = task_pt_regs(task)->flags;
392 * If the debugger set TF, hide it from the readout.
394 if (test_tsk_thread_flag(task, TIF_FORCED_TF))
395 retval &= ~X86_EFLAGS_TF;
400 static int set_flags(struct task_struct *task, unsigned long value)
402 struct pt_regs *regs = task_pt_regs(task);
405 * If the user value contains TF, mark that
406 * it was not "us" (the debugger) that set it.
407 * If not, make sure it stays set if we had.
409 if (value & X86_EFLAGS_TF)
410 clear_tsk_thread_flag(task, TIF_FORCED_TF);
411 else if (test_tsk_thread_flag(task, TIF_FORCED_TF))
412 value |= X86_EFLAGS_TF;
414 regs->flags = (regs->flags & ~FLAG_MASK) | (value & FLAG_MASK);
419 static int putreg(struct task_struct *child,
420 unsigned long offset, unsigned long value)
423 case offsetof(struct user_regs_struct, cs):
424 case offsetof(struct user_regs_struct, ds):
425 case offsetof(struct user_regs_struct, es):
426 case offsetof(struct user_regs_struct, fs):
427 case offsetof(struct user_regs_struct, gs):
428 case offsetof(struct user_regs_struct, ss):
429 return set_segment_reg(child, offset, value);
431 case offsetof(struct user_regs_struct, flags):
432 return set_flags(child, value);
435 case offsetof(struct user_regs_struct,fs_base):
436 if (value >= TASK_SIZE_OF(child))
439 * When changing the segment base, use do_arch_prctl
440 * to set either thread.fs or thread.fsindex and the
441 * corresponding GDT slot.
443 if (child->thread.fs != value)
444 return do_arch_prctl(child, ARCH_SET_FS, value);
446 case offsetof(struct user_regs_struct,gs_base):
448 * Exactly the same here as the %fs handling above.
450 if (value >= TASK_SIZE_OF(child))
452 if (child->thread.gs != value)
453 return do_arch_prctl(child, ARCH_SET_GS, value);
458 *pt_regs_access(task_pt_regs(child), offset) = value;
462 static unsigned long getreg(struct task_struct *task, unsigned long offset)
465 case offsetof(struct user_regs_struct, cs):
466 case offsetof(struct user_regs_struct, ds):
467 case offsetof(struct user_regs_struct, es):
468 case offsetof(struct user_regs_struct, fs):
469 case offsetof(struct user_regs_struct, gs):
470 case offsetof(struct user_regs_struct, ss):
471 return get_segment_reg(task, offset);
473 case offsetof(struct user_regs_struct, flags):
474 return get_flags(task);
477 case offsetof(struct user_regs_struct, fs_base): {
479 * do_arch_prctl may have used a GDT slot instead of
480 * the MSR. To userland, it appears the same either
481 * way, except the %fs segment selector might not be 0.
483 unsigned int seg = task->thread.fsindex;
484 if (task->thread.fs != 0)
485 return task->thread.fs;
487 asm("movl %%fs,%0" : "=r" (seg));
488 if (seg != FS_TLS_SEL)
490 return get_desc_base(&task->thread.tls_array[FS_TLS]);
492 case offsetof(struct user_regs_struct, gs_base): {
494 * Exactly the same here as the %fs handling above.
496 unsigned int seg = task->thread.gsindex;
497 if (task->thread.gs != 0)
498 return task->thread.gs;
500 asm("movl %%gs,%0" : "=r" (seg));
501 if (seg != GS_TLS_SEL)
503 return get_desc_base(&task->thread.tls_array[GS_TLS]);
508 return *pt_regs_access(task_pt_regs(task), offset);
511 static int genregs_get(struct task_struct *target,
512 const struct user_regset *regset,
513 unsigned int pos, unsigned int count,
514 void *kbuf, void __user *ubuf)
517 unsigned long *k = kbuf;
518 while (count >= sizeof(*k)) {
519 *k++ = getreg(target, pos);
524 unsigned long __user *u = ubuf;
525 while (count >= sizeof(*u)) {
526 if (__put_user(getreg(target, pos), u++))
536 static int genregs_set(struct task_struct *target,
537 const struct user_regset *regset,
538 unsigned int pos, unsigned int count,
539 const void *kbuf, const void __user *ubuf)
543 const unsigned long *k = kbuf;
544 while (count >= sizeof(*k) && !ret) {
545 ret = putreg(target, pos, *k++);
550 const unsigned long __user *u = ubuf;
551 while (count >= sizeof(*u) && !ret) {
553 ret = __get_user(word, u++);
556 ret = putreg(target, pos, word);
564 static void ptrace_triggered(struct perf_event *bp,
565 struct perf_sample_data *data,
566 struct pt_regs *regs)
569 struct thread_struct *thread = &(current->thread);
572 * Store in the virtual DR6 register the fact that the breakpoint
573 * was hit so the thread's debugger will see it.
575 for (i = 0; i < HBP_NUM; i++) {
576 if (thread->ptrace_bps[i] == bp)
580 thread->debugreg6 |= (DR_TRAP0 << i);
584 * Walk through every ptrace breakpoints for this thread and
585 * build the dr7 value on top of their attributes.
588 static unsigned long ptrace_get_dr7(struct perf_event *bp[])
592 struct arch_hw_breakpoint *info;
594 for (i = 0; i < HBP_NUM; i++) {
595 if (bp[i] && !bp[i]->attr.disabled) {
596 info = counter_arch_bp(bp[i]);
597 dr7 |= encode_dr7(i, info->len, info->type);
604 static int ptrace_fill_bp_fields(struct perf_event_attr *attr,
605 int len, int type, bool disabled)
607 int err, bp_len, bp_type;
609 err = arch_bp_generic_fields(len, type, &bp_len, &bp_type);
611 attr->bp_len = bp_len;
612 attr->bp_type = bp_type;
613 attr->disabled = disabled;
619 static struct perf_event *
620 ptrace_register_breakpoint(struct task_struct *tsk, int len, int type,
621 unsigned long addr, bool disabled)
623 struct perf_event_attr attr;
626 ptrace_breakpoint_init(&attr);
629 err = ptrace_fill_bp_fields(&attr, len, type, disabled);
633 return register_user_hw_breakpoint(&attr, ptrace_triggered,
637 static int ptrace_modify_breakpoint(struct perf_event *bp, int len, int type,
640 struct perf_event_attr attr = bp->attr;
643 err = ptrace_fill_bp_fields(&attr, len, type, disabled);
647 return modify_user_hw_breakpoint(bp, &attr);
651 * Handle ptrace writes to debug register 7.
653 static int ptrace_write_dr7(struct task_struct *tsk, unsigned long data)
655 struct thread_struct *thread = &tsk->thread;
656 unsigned long old_dr7;
657 bool second_pass = false;
660 data &= ~DR_CONTROL_RESERVED;
661 old_dr7 = ptrace_get_dr7(thread->ptrace_bps);
665 for (i = 0; i < HBP_NUM; i++) {
667 bool disabled = !decode_dr7(data, i, &len, &type);
668 struct perf_event *bp = thread->ptrace_bps[i];
674 * We should have at least an inactive breakpoint at
675 * this slot. It means the user is writing dr7 without
676 * having written the address register first.
682 rc = ptrace_modify_breakpoint(bp, len, type, disabled);
687 /* Restore if the first pass failed, second_pass shouldn't fail. */
688 if (rc && !WARN_ON(second_pass)) {
699 * Handle PTRACE_PEEKUSR calls for the debug register area.
701 static unsigned long ptrace_get_debugreg(struct task_struct *tsk, int n)
703 struct thread_struct *thread = &(tsk->thread);
704 unsigned long val = 0;
707 struct perf_event *bp = thread->ptrace_bps[n];
710 val = bp->hw.info.address;
712 val = thread->debugreg6;
714 val = thread->ptrace_dr7;
719 static int ptrace_set_breakpoint_addr(struct task_struct *tsk, int nr,
722 struct thread_struct *t = &tsk->thread;
723 struct perf_event *bp = t->ptrace_bps[nr];
728 * Put stub len and type to create an inactive but correct bp.
730 * CHECKME: the previous code returned -EIO if the addr wasn't
731 * a valid task virtual addr. The new one will return -EINVAL in
733 * -EINVAL may be what we want for in-kernel breakpoints users,
734 * but -EIO looks better for ptrace, since we refuse a register
735 * writing for the user. And anyway this is the previous
738 bp = ptrace_register_breakpoint(tsk,
739 X86_BREAKPOINT_LEN_1, X86_BREAKPOINT_WRITE,
744 t->ptrace_bps[nr] = bp;
746 struct perf_event_attr attr = bp->attr;
749 err = modify_user_hw_breakpoint(bp, &attr);
756 * Handle PTRACE_POKEUSR calls for the debug register area.
758 static int ptrace_set_debugreg(struct task_struct *tsk, int n,
761 struct thread_struct *thread = &(tsk->thread);
764 /* There are no DR4 or DR5 registers */
765 if (n == 4 || n == 5)
769 thread->debugreg6 = val;
773 rc = ptrace_set_breakpoint_addr(tsk, n, val);
777 /* All that's left is DR7 */
779 rc = ptrace_write_dr7(tsk, val);
781 thread->ptrace_dr7 = val;
789 * These access the current or another (stopped) task's io permission
790 * bitmap for debugging or core dump.
792 static int ioperm_active(struct task_struct *target,
793 const struct user_regset *regset)
795 return target->thread.io_bitmap_max / regset->size;
798 static int ioperm_get(struct task_struct *target,
799 const struct user_regset *regset,
800 unsigned int pos, unsigned int count,
801 void *kbuf, void __user *ubuf)
803 if (!target->thread.io_bitmap_ptr)
806 return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
807 target->thread.io_bitmap_ptr,
812 * Called by kernel/ptrace.c when detaching..
814 * Make sure the single step bit is not set.
816 void ptrace_disable(struct task_struct *child)
818 user_disable_single_step(child);
819 #ifdef TIF_SYSCALL_EMU
820 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
824 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
825 static const struct user_regset_view user_x86_32_view; /* Initialized below. */
828 long arch_ptrace(struct task_struct *child, long request,
829 unsigned long addr, unsigned long data)
832 unsigned long __user *datap = (unsigned long __user *)data;
835 /* read the word at location addr in the USER area. */
836 case PTRACE_PEEKUSR: {
840 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user))
843 tmp = 0; /* Default return condition */
844 if (addr < sizeof(struct user_regs_struct))
845 tmp = getreg(child, addr);
846 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
847 addr <= offsetof(struct user, u_debugreg[7])) {
848 addr -= offsetof(struct user, u_debugreg[0]);
849 tmp = ptrace_get_debugreg(child, addr / sizeof(data));
851 ret = put_user(tmp, datap);
855 case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
857 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user))
860 if (addr < sizeof(struct user_regs_struct))
861 ret = putreg(child, addr, data);
862 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
863 addr <= offsetof(struct user, u_debugreg[7])) {
864 addr -= offsetof(struct user, u_debugreg[0]);
865 ret = ptrace_set_debugreg(child,
866 addr / sizeof(data), data);
870 case PTRACE_GETREGS: /* Get all gp regs from the child. */
871 return copy_regset_to_user(child,
872 task_user_regset_view(current),
874 0, sizeof(struct user_regs_struct),
877 case PTRACE_SETREGS: /* Set all gp regs in the child. */
878 return copy_regset_from_user(child,
879 task_user_regset_view(current),
881 0, sizeof(struct user_regs_struct),
884 case PTRACE_GETFPREGS: /* Get the child FPU state. */
885 return copy_regset_to_user(child,
886 task_user_regset_view(current),
888 0, sizeof(struct user_i387_struct),
891 case PTRACE_SETFPREGS: /* Set the child FPU state. */
892 return copy_regset_from_user(child,
893 task_user_regset_view(current),
895 0, sizeof(struct user_i387_struct),
899 case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
900 return copy_regset_to_user(child, &user_x86_32_view,
902 0, sizeof(struct user_fxsr_struct),
905 case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
906 return copy_regset_from_user(child, &user_x86_32_view,
908 0, sizeof(struct user_fxsr_struct),
912 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
913 case PTRACE_GET_THREAD_AREA:
916 ret = do_get_thread_area(child, addr,
917 (struct user_desc __user *)data);
920 case PTRACE_SET_THREAD_AREA:
923 ret = do_set_thread_area(child, addr,
924 (struct user_desc __user *)data, 0);
929 /* normal 64bit interface to access TLS data.
930 Works just like arch_prctl, except that the arguments
932 case PTRACE_ARCH_PRCTL:
933 ret = do_arch_prctl(child, data, addr);
938 ret = ptrace_request(child, request, addr, data);
945 #ifdef CONFIG_IA32_EMULATION
947 #include <linux/compat.h>
948 #include <linux/syscalls.h>
949 #include <asm/ia32.h>
950 #include <asm/user32.h>
953 case offsetof(struct user32, regs.l): \
954 regs->q = value; break
957 case offsetof(struct user32, regs.rs): \
958 return set_segment_reg(child, \
959 offsetof(struct user_regs_struct, rs), \
963 static int putreg32(struct task_struct *child, unsigned regno, u32 value)
965 struct pt_regs *regs = task_pt_regs(child);
986 case offsetof(struct user32, regs.orig_eax):
988 * A 32-bit debugger setting orig_eax means to restore
989 * the state of the task restarting a 32-bit syscall.
990 * Make sure we interpret the -ERESTART* codes correctly
991 * in case the task is not actually still sitting at the
992 * exit from a 32-bit syscall with TS_COMPAT still set.
994 regs->orig_ax = value;
995 if (syscall_get_nr(child, regs) >= 0)
996 task_thread_info(child)->status |= TS_COMPAT;
999 case offsetof(struct user32, regs.eflags):
1000 return set_flags(child, value);
1002 case offsetof(struct user32, u_debugreg[0]) ...
1003 offsetof(struct user32, u_debugreg[7]):
1004 regno -= offsetof(struct user32, u_debugreg[0]);
1005 return ptrace_set_debugreg(child, regno / 4, value);
1008 if (regno > sizeof(struct user32) || (regno & 3))
1012 * Other dummy fields in the virtual user structure
1024 case offsetof(struct user32, regs.l): \
1025 *val = regs->q; break
1028 case offsetof(struct user32, regs.rs): \
1029 *val = get_segment_reg(child, \
1030 offsetof(struct user_regs_struct, rs)); \
1033 static int getreg32(struct task_struct *child, unsigned regno, u32 *val)
1035 struct pt_regs *regs = task_pt_regs(child);
1053 R32(orig_eax, orig_ax);
1057 case offsetof(struct user32, regs.eflags):
1058 *val = get_flags(child);
1061 case offsetof(struct user32, u_debugreg[0]) ...
1062 offsetof(struct user32, u_debugreg[7]):
1063 regno -= offsetof(struct user32, u_debugreg[0]);
1064 *val = ptrace_get_debugreg(child, regno / 4);
1068 if (regno > sizeof(struct user32) || (regno & 3))
1072 * Other dummy fields in the virtual user structure
1084 static int genregs32_get(struct task_struct *target,
1085 const struct user_regset *regset,
1086 unsigned int pos, unsigned int count,
1087 void *kbuf, void __user *ubuf)
1090 compat_ulong_t *k = kbuf;
1091 while (count >= sizeof(*k)) {
1092 getreg32(target, pos, k++);
1093 count -= sizeof(*k);
1097 compat_ulong_t __user *u = ubuf;
1098 while (count >= sizeof(*u)) {
1099 compat_ulong_t word;
1100 getreg32(target, pos, &word);
1101 if (__put_user(word, u++))
1103 count -= sizeof(*u);
1111 static int genregs32_set(struct task_struct *target,
1112 const struct user_regset *regset,
1113 unsigned int pos, unsigned int count,
1114 const void *kbuf, const void __user *ubuf)
1118 const compat_ulong_t *k = kbuf;
1119 while (count >= sizeof(*k) && !ret) {
1120 ret = putreg32(target, pos, *k++);
1121 count -= sizeof(*k);
1125 const compat_ulong_t __user *u = ubuf;
1126 while (count >= sizeof(*u) && !ret) {
1127 compat_ulong_t word;
1128 ret = __get_user(word, u++);
1131 ret = putreg32(target, pos, word);
1132 count -= sizeof(*u);
1139 #ifdef CONFIG_X86_X32_ABI
1140 static long x32_arch_ptrace(struct task_struct *child,
1141 compat_long_t request, compat_ulong_t caddr,
1142 compat_ulong_t cdata)
1144 unsigned long addr = caddr;
1145 unsigned long data = cdata;
1146 void __user *datap = compat_ptr(data);
1150 /* Read 32bits at location addr in the USER area. Only allow
1151 to return the lower 32bits of segment and debug registers. */
1152 case PTRACE_PEEKUSR: {
1156 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user) ||
1157 addr < offsetof(struct user_regs_struct, cs))
1160 tmp = 0; /* Default return condition */
1161 if (addr < sizeof(struct user_regs_struct))
1162 tmp = getreg(child, addr);
1163 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
1164 addr <= offsetof(struct user, u_debugreg[7])) {
1165 addr -= offsetof(struct user, u_debugreg[0]);
1166 tmp = ptrace_get_debugreg(child, addr / sizeof(data));
1168 ret = put_user(tmp, (__u32 __user *)datap);
1172 /* Write the word at location addr in the USER area. Only allow
1173 to update segment and debug registers with the upper 32bits
1175 case PTRACE_POKEUSR:
1177 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user) ||
1178 addr < offsetof(struct user_regs_struct, cs))
1181 if (addr < sizeof(struct user_regs_struct))
1182 ret = putreg(child, addr, data);
1183 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
1184 addr <= offsetof(struct user, u_debugreg[7])) {
1185 addr -= offsetof(struct user, u_debugreg[0]);
1186 ret = ptrace_set_debugreg(child,
1187 addr / sizeof(data), data);
1191 case PTRACE_GETREGS: /* Get all gp regs from the child. */
1192 return copy_regset_to_user(child,
1193 task_user_regset_view(current),
1195 0, sizeof(struct user_regs_struct),
1198 case PTRACE_SETREGS: /* Set all gp regs in the child. */
1199 return copy_regset_from_user(child,
1200 task_user_regset_view(current),
1202 0, sizeof(struct user_regs_struct),
1205 case PTRACE_GETFPREGS: /* Get the child FPU state. */
1206 return copy_regset_to_user(child,
1207 task_user_regset_view(current),
1209 0, sizeof(struct user_i387_struct),
1212 case PTRACE_SETFPREGS: /* Set the child FPU state. */
1213 return copy_regset_from_user(child,
1214 task_user_regset_view(current),
1216 0, sizeof(struct user_i387_struct),
1220 return compat_ptrace_request(child, request, addr, data);
1227 long compat_arch_ptrace(struct task_struct *child, compat_long_t request,
1228 compat_ulong_t caddr, compat_ulong_t cdata)
1230 unsigned long addr = caddr;
1231 unsigned long data = cdata;
1232 void __user *datap = compat_ptr(data);
1236 #ifdef CONFIG_X86_X32_ABI
1237 if (!is_ia32_task())
1238 return x32_arch_ptrace(child, request, caddr, cdata);
1242 case PTRACE_PEEKUSR:
1243 ret = getreg32(child, addr, &val);
1245 ret = put_user(val, (__u32 __user *)datap);
1248 case PTRACE_POKEUSR:
1249 ret = putreg32(child, addr, data);
1252 case PTRACE_GETREGS: /* Get all gp regs from the child. */
1253 return copy_regset_to_user(child, &user_x86_32_view,
1255 0, sizeof(struct user_regs_struct32),
1258 case PTRACE_SETREGS: /* Set all gp regs in the child. */
1259 return copy_regset_from_user(child, &user_x86_32_view,
1261 sizeof(struct user_regs_struct32),
1264 case PTRACE_GETFPREGS: /* Get the child FPU state. */
1265 return copy_regset_to_user(child, &user_x86_32_view,
1267 sizeof(struct user_i387_ia32_struct),
1270 case PTRACE_SETFPREGS: /* Set the child FPU state. */
1271 return copy_regset_from_user(
1272 child, &user_x86_32_view, REGSET_FP,
1273 0, sizeof(struct user_i387_ia32_struct), datap);
1275 case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
1276 return copy_regset_to_user(child, &user_x86_32_view,
1278 sizeof(struct user32_fxsr_struct),
1281 case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
1282 return copy_regset_from_user(child, &user_x86_32_view,
1284 sizeof(struct user32_fxsr_struct),
1287 case PTRACE_GET_THREAD_AREA:
1288 case PTRACE_SET_THREAD_AREA:
1289 return arch_ptrace(child, request, addr, data);
1292 return compat_ptrace_request(child, request, addr, data);
1298 #endif /* CONFIG_IA32_EMULATION */
1300 #ifdef CONFIG_X86_64
1302 static struct user_regset x86_64_regsets[] __read_mostly = {
1303 [REGSET_GENERAL] = {
1304 .core_note_type = NT_PRSTATUS,
1305 .n = sizeof(struct user_regs_struct) / sizeof(long),
1306 .size = sizeof(long), .align = sizeof(long),
1307 .get = genregs_get, .set = genregs_set
1310 .core_note_type = NT_PRFPREG,
1311 .n = sizeof(struct user_i387_struct) / sizeof(long),
1312 .size = sizeof(long), .align = sizeof(long),
1313 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1316 .core_note_type = NT_X86_XSTATE,
1317 .size = sizeof(u64), .align = sizeof(u64),
1318 .active = xstateregs_active, .get = xstateregs_get,
1319 .set = xstateregs_set
1321 [REGSET_IOPERM64] = {
1322 .core_note_type = NT_386_IOPERM,
1323 .n = IO_BITMAP_LONGS,
1324 .size = sizeof(long), .align = sizeof(long),
1325 .active = ioperm_active, .get = ioperm_get
1329 static const struct user_regset_view user_x86_64_view = {
1330 .name = "x86_64", .e_machine = EM_X86_64,
1331 .regsets = x86_64_regsets, .n = ARRAY_SIZE(x86_64_regsets)
1334 #else /* CONFIG_X86_32 */
1336 #define user_regs_struct32 user_regs_struct
1337 #define genregs32_get genregs_get
1338 #define genregs32_set genregs_set
1340 #endif /* CONFIG_X86_64 */
1342 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1343 static struct user_regset x86_32_regsets[] __read_mostly = {
1344 [REGSET_GENERAL] = {
1345 .core_note_type = NT_PRSTATUS,
1346 .n = sizeof(struct user_regs_struct32) / sizeof(u32),
1347 .size = sizeof(u32), .align = sizeof(u32),
1348 .get = genregs32_get, .set = genregs32_set
1351 .core_note_type = NT_PRFPREG,
1352 .n = sizeof(struct user_i387_ia32_struct) / sizeof(u32),
1353 .size = sizeof(u32), .align = sizeof(u32),
1354 .active = fpregs_active, .get = fpregs_get, .set = fpregs_set
1357 .core_note_type = NT_PRXFPREG,
1358 .n = sizeof(struct user32_fxsr_struct) / sizeof(u32),
1359 .size = sizeof(u32), .align = sizeof(u32),
1360 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1363 .core_note_type = NT_X86_XSTATE,
1364 .size = sizeof(u64), .align = sizeof(u64),
1365 .active = xstateregs_active, .get = xstateregs_get,
1366 .set = xstateregs_set
1369 .core_note_type = NT_386_TLS,
1370 .n = GDT_ENTRY_TLS_ENTRIES, .bias = GDT_ENTRY_TLS_MIN,
1371 .size = sizeof(struct user_desc),
1372 .align = sizeof(struct user_desc),
1373 .active = regset_tls_active,
1374 .get = regset_tls_get, .set = regset_tls_set
1376 [REGSET_IOPERM32] = {
1377 .core_note_type = NT_386_IOPERM,
1378 .n = IO_BITMAP_BYTES / sizeof(u32),
1379 .size = sizeof(u32), .align = sizeof(u32),
1380 .active = ioperm_active, .get = ioperm_get
1384 static const struct user_regset_view user_x86_32_view = {
1385 .name = "i386", .e_machine = EM_386,
1386 .regsets = x86_32_regsets, .n = ARRAY_SIZE(x86_32_regsets)
1391 * This represents bytes 464..511 in the memory layout exported through
1392 * the REGSET_XSTATE interface.
1394 u64 xstate_fx_sw_bytes[USER_XSTATE_FX_SW_WORDS];
1396 void update_regset_xstate_info(unsigned int size, u64 xstate_mask)
1398 #ifdef CONFIG_X86_64
1399 x86_64_regsets[REGSET_XSTATE].n = size / sizeof(u64);
1401 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1402 x86_32_regsets[REGSET_XSTATE].n = size / sizeof(u64);
1404 xstate_fx_sw_bytes[USER_XSTATE_XCR0_WORD] = xstate_mask;
1407 const struct user_regset_view *task_user_regset_view(struct task_struct *task)
1409 #ifdef CONFIG_IA32_EMULATION
1410 if (test_tsk_thread_flag(task, TIF_IA32))
1412 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1413 return &user_x86_32_view;
1415 #ifdef CONFIG_X86_64
1416 return &user_x86_64_view;
1420 static void fill_sigtrap_info(struct task_struct *tsk,
1421 struct pt_regs *regs,
1422 int error_code, int si_code,
1423 struct siginfo *info)
1425 tsk->thread.trap_nr = X86_TRAP_DB;
1426 tsk->thread.error_code = error_code;
1428 memset(info, 0, sizeof(*info));
1429 info->si_signo = SIGTRAP;
1430 info->si_code = si_code;
1431 info->si_addr = user_mode_vm(regs) ? (void __user *)regs->ip : NULL;
1434 void user_single_step_siginfo(struct task_struct *tsk,
1435 struct pt_regs *regs,
1436 struct siginfo *info)
1438 fill_sigtrap_info(tsk, regs, 0, TRAP_BRKPT, info);
1441 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs,
1442 int error_code, int si_code)
1444 struct siginfo info;
1446 fill_sigtrap_info(tsk, regs, error_code, si_code, &info);
1447 /* Send us the fake SIGTRAP */
1448 force_sig_info(SIGTRAP, &info, tsk);
1452 #ifdef CONFIG_X86_32
1454 #elif defined CONFIG_IA32_EMULATION
1455 # define IS_IA32 is_compat_task()
1461 * We must return the syscall number to actually look up in the table.
1462 * This can be -1L to skip running any syscall at all.
1464 long syscall_trace_enter(struct pt_regs *regs)
1471 * If we stepped into a sysenter/syscall insn, it trapped in
1472 * kernel mode; do_debug() cleared TF and set TIF_SINGLESTEP.
1473 * If user-mode had set TF itself, then it's still clear from
1474 * do_debug() and we need to set it again to restore the user
1475 * state. If we entered on the slow path, TF was already set.
1477 if (test_thread_flag(TIF_SINGLESTEP))
1478 regs->flags |= X86_EFLAGS_TF;
1480 /* do the secure computing check first */
1481 if (secure_computing(regs->orig_ax)) {
1482 /* seccomp failures shouldn't expose any additional code. */
1487 if (unlikely(test_thread_flag(TIF_SYSCALL_EMU)))
1490 if ((ret || test_thread_flag(TIF_SYSCALL_TRACE)) &&
1491 tracehook_report_syscall_entry(regs))
1494 if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
1495 trace_sys_enter(regs, regs->orig_ax);
1498 audit_syscall_entry(AUDIT_ARCH_I386,
1501 regs->dx, regs->si);
1502 #ifdef CONFIG_X86_64
1504 audit_syscall_entry(AUDIT_ARCH_X86_64,
1507 regs->dx, regs->r10);
1511 return ret ?: regs->orig_ax;
1514 void syscall_trace_leave(struct pt_regs *regs)
1519 * We may come here right after calling schedule_user()
1520 * or do_notify_resume(), in which case we can be in RCU
1525 audit_syscall_exit(regs);
1527 if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
1528 trace_sys_exit(regs, regs->ax);
1531 * If TIF_SYSCALL_EMU is set, we only get here because of
1532 * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
1533 * We already reported this syscall instruction in
1534 * syscall_trace_enter().
1536 step = unlikely(test_thread_flag(TIF_SINGLESTEP)) &&
1537 !test_thread_flag(TIF_SYSCALL_EMU);
1538 if (step || test_thread_flag(TIF_SYSCALL_TRACE))
1539 tracehook_report_syscall_exit(regs, step);