2 /* By Ross Biro 1/23/92 */
4 * Pentium III FXSR, SSE support
5 * Gareth Hughes <gareth@valinux.com>, May 2000
8 #include <linux/kernel.h>
9 #include <linux/sched.h>
11 #include <linux/smp.h>
12 #include <linux/errno.h>
13 #include <linux/ptrace.h>
14 #include <linux/user.h>
15 #include <linux/security.h>
16 #include <linux/audit.h>
17 #include <linux/seccomp.h>
18 #include <linux/signal.h>
20 #include <asm/uaccess.h>
21 #include <asm/pgtable.h>
22 #include <asm/system.h>
23 #include <asm/processor.h>
25 #include <asm/debugreg.h>
30 * does not yet catch signals sent when the child dies.
31 * in exit.c or in signal.c.
35 * Determines which flags the user has access to [1 = access, 0 = no access].
36 * Prohibits changing ID(21), VIP(20), VIF(19), VM(17), NT(14), IOPL(12-13), IF(9).
37 * Also masks reserved bits (31-22, 15, 5, 3, 1).
39 #define FLAG_MASK 0x00050dd5
41 /* set's the trap flag. */
42 #define TRAP_FLAG 0x100
45 * Offset of eflags on child stack..
47 #define EFL_OFFSET offsetof(struct pt_regs, eflags)
49 static inline struct pt_regs *get_child_regs(struct task_struct *task)
51 void *stack_top = (void *)task->thread.esp0;
52 return stack_top - sizeof(struct pt_regs);
56 * This routine will get a word off of the processes privileged stack.
57 * the offset is bytes into the pt_regs structure on the stack.
58 * This routine assumes that all the privileged stacks are in our
61 static inline int get_stack_long(struct task_struct *task, int offset)
65 stack = (unsigned char *)task->thread.esp0 - sizeof(struct pt_regs);
67 return (*((int *)stack));
71 * This routine will put a word on the processes privileged stack.
72 * the offset is bytes into the pt_regs structure on the stack.
73 * This routine assumes that all the privileged stacks are in our
76 static inline int put_stack_long(struct task_struct *task, int offset,
79 unsigned char * stack;
81 stack = (unsigned char *)task->thread.esp0 - sizeof(struct pt_regs);
83 *(unsigned long *) stack = data;
87 static int putreg(struct task_struct *child,
88 unsigned long regno, unsigned long value)
92 if (value && (value & 3) != 3)
94 child->thread.gs = value;
99 if (value && (value & 3) != 3)
105 if ((value & 3) != 3)
111 value |= get_stack_long(child, EFL_OFFSET) & ~FLAG_MASK;
116 put_stack_long(child, regno, value);
120 static unsigned long getreg(struct task_struct *child,
123 unsigned long retval = ~0UL;
125 switch (regno >> 2) {
127 retval = child->thread.gs;
139 retval &= get_stack_long(child, regno);
144 #define LDT_SEGMENT 4
146 static unsigned long convert_eip_to_linear(struct task_struct *child, struct pt_regs *regs)
148 unsigned long addr, seg;
151 seg = regs->xcs & 0xffff;
152 if (regs->eflags & VM_MASK) {
153 addr = (addr & 0xffff) + (seg << 4);
158 * We'll assume that the code segments in the GDT
159 * are all zero-based. That is largely true: the
160 * TLS segments are used for data, and the PNPBIOS
161 * and APM bios ones we just ignore here.
163 if (seg & LDT_SEGMENT) {
169 down(&child->mm->context.sem);
170 if (unlikely((seg >> 3) >= child->mm->context.size))
171 addr = -1L; /* bogus selector, access would fault */
173 desc = child->mm->context.ldt + seg;
174 base = ((desc[0] >> 16) |
175 ((desc[1] & 0xff) << 16) |
176 (desc[1] & 0xff000000));
178 /* 16-bit code segment? */
179 if (!((desc[1] >> 22) & 1))
183 up(&child->mm->context.sem);
188 static inline int is_setting_trap_flag(struct task_struct *child, struct pt_regs *regs)
191 unsigned char opcode[15];
192 unsigned long addr = convert_eip_to_linear(child, regs);
194 copied = access_process_vm(child, addr, opcode, sizeof(opcode), 0);
195 for (i = 0; i < copied; i++) {
198 case 0x9d: case 0xcf:
200 /* opcode and address size prefixes */
201 case 0x66: case 0x67:
203 /* irrelevant prefixes (segment overrides and repeats) */
204 case 0x26: case 0x2e:
205 case 0x36: case 0x3e:
206 case 0x64: case 0x65:
207 case 0xf0: case 0xf2: case 0xf3:
211 * pushf: NOTE! We should probably not let
212 * the user see the TF bit being set. But
213 * it's more pain than it's worth to avoid
214 * it, and a debugger could emulate this
215 * all in user space if it _really_ cares.
225 static void set_singlestep(struct task_struct *child)
227 struct pt_regs *regs = get_child_regs(child);
230 * Always set TIF_SINGLESTEP - this guarantees that
231 * we single-step system calls etc.. This will also
232 * cause us to set TF when returning to user mode.
234 set_tsk_thread_flag(child, TIF_SINGLESTEP);
237 * If TF was already set, don't do anything else
239 if (regs->eflags & TRAP_FLAG)
242 /* Set TF on the kernel stack.. */
243 regs->eflags |= TRAP_FLAG;
246 * ..but if TF is changed by the instruction we will trace,
247 * don't mark it as being "us" that set it, so that we
248 * won't clear it by hand later.
250 if (is_setting_trap_flag(child, regs))
253 child->ptrace |= PT_DTRACE;
256 static void clear_singlestep(struct task_struct *child)
258 /* Always clear TIF_SINGLESTEP... */
259 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
261 /* But touch TF only if it was set by us.. */
262 if (child->ptrace & PT_DTRACE) {
263 struct pt_regs *regs = get_child_regs(child);
264 regs->eflags &= ~TRAP_FLAG;
265 child->ptrace &= ~PT_DTRACE;
270 * Called by kernel/ptrace.c when detaching..
272 * Make sure the single step bit is not set.
274 void ptrace_disable(struct task_struct *child)
276 clear_singlestep(child);
277 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
281 * Perform get_thread_area on behalf of the traced child.
284 ptrace_get_thread_area(struct task_struct *child,
285 int idx, struct user_desc __user *user_desc)
287 struct user_desc info;
288 struct desc_struct *desc;
291 * Get the current Thread-Local Storage area:
294 #define GET_BASE(desc) ( \
295 (((desc)->a >> 16) & 0x0000ffff) | \
296 (((desc)->b << 16) & 0x00ff0000) | \
297 ( (desc)->b & 0xff000000) )
299 #define GET_LIMIT(desc) ( \
300 ((desc)->a & 0x0ffff) | \
301 ((desc)->b & 0xf0000) )
303 #define GET_32BIT(desc) (((desc)->b >> 22) & 1)
304 #define GET_CONTENTS(desc) (((desc)->b >> 10) & 3)
305 #define GET_WRITABLE(desc) (((desc)->b >> 9) & 1)
306 #define GET_LIMIT_PAGES(desc) (((desc)->b >> 23) & 1)
307 #define GET_PRESENT(desc) (((desc)->b >> 15) & 1)
308 #define GET_USEABLE(desc) (((desc)->b >> 20) & 1)
310 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
313 desc = child->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
315 info.entry_number = idx;
316 info.base_addr = GET_BASE(desc);
317 info.limit = GET_LIMIT(desc);
318 info.seg_32bit = GET_32BIT(desc);
319 info.contents = GET_CONTENTS(desc);
320 info.read_exec_only = !GET_WRITABLE(desc);
321 info.limit_in_pages = GET_LIMIT_PAGES(desc);
322 info.seg_not_present = !GET_PRESENT(desc);
323 info.useable = GET_USEABLE(desc);
325 if (copy_to_user(user_desc, &info, sizeof(info)))
332 * Perform set_thread_area on behalf of the traced child.
335 ptrace_set_thread_area(struct task_struct *child,
336 int idx, struct user_desc __user *user_desc)
338 struct user_desc info;
339 struct desc_struct *desc;
341 if (copy_from_user(&info, user_desc, sizeof(info)))
344 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
347 desc = child->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
348 if (LDT_empty(&info)) {
352 desc->a = LDT_entry_a(&info);
353 desc->b = LDT_entry_b(&info);
359 long arch_ptrace(struct task_struct *child, long request, long addr, long data)
361 struct user * dummy = NULL;
363 unsigned long __user *datap = (unsigned long __user *)data;
366 /* when I and D space are separate, these will need to be fixed. */
367 case PTRACE_PEEKTEXT: /* read word at location addr. */
368 case PTRACE_PEEKDATA:
369 ret = generic_ptrace_peekdata(child, addr, data);
372 /* read the word at location addr in the USER area. */
373 case PTRACE_PEEKUSR: {
377 if ((addr & 3) || addr < 0 ||
378 addr > sizeof(struct user) - 3)
381 tmp = 0; /* Default return condition */
382 if(addr < FRAME_SIZE*sizeof(long))
383 tmp = getreg(child, addr);
384 if(addr >= (long) &dummy->u_debugreg[0] &&
385 addr <= (long) &dummy->u_debugreg[7]){
386 addr -= (long) &dummy->u_debugreg[0];
388 tmp = child->thread.debugreg[addr];
390 ret = put_user(tmp, datap);
394 /* when I and D space are separate, this will have to be fixed. */
395 case PTRACE_POKETEXT: /* write the word at location addr. */
396 case PTRACE_POKEDATA:
397 ret = generic_ptrace_pokedata(child, addr, data);
400 case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
402 if ((addr & 3) || addr < 0 ||
403 addr > sizeof(struct user) - 3)
406 if (addr < FRAME_SIZE*sizeof(long)) {
407 ret = putreg(child, addr, data);
410 /* We need to be very careful here. We implicitly
411 want to modify a portion of the task_struct, and we
412 have to be selective about what portions we allow someone
416 if(addr >= (long) &dummy->u_debugreg[0] &&
417 addr <= (long) &dummy->u_debugreg[7]){
419 if(addr == (long) &dummy->u_debugreg[4]) break;
420 if(addr == (long) &dummy->u_debugreg[5]) break;
421 if(addr < (long) &dummy->u_debugreg[4] &&
422 ((unsigned long) data) >= TASK_SIZE-3) break;
424 /* Sanity-check data. Take one half-byte at once with
425 * check = (val >> (16 + 4*i)) & 0xf. It contains the
426 * R/Wi and LENi bits; bits 0 and 1 are R/Wi, and bits
427 * 2 and 3 are LENi. Given a list of invalid values,
428 * we do mask |= 1 << invalid_value, so that
429 * (mask >> check) & 1 is a correct test for invalid
432 * R/Wi contains the type of the breakpoint /
433 * watchpoint, LENi contains the length of the watched
434 * data in the watchpoint case.
436 * The invalid values are:
437 * - LENi == 0x10 (undefined), so mask |= 0x0f00.
438 * - R/Wi == 0x10 (break on I/O reads or writes), so
440 * - R/Wi == 0x00 && LENi != 0x00, so we have mask |=
443 * Finally, mask = 0x0f00 | 0x4444 | 0x1110 == 0x5f54.
445 * See the Intel Manual "System Programming Guide",
448 * Note that LENi == 0x10 is defined on x86_64 in long
449 * mode (i.e. even for 32-bit userspace software, but
450 * 64-bit kernel), so the x86_64 mask value is 0x5454.
451 * See the AMD manual no. 24593 (AMD64 System
454 if(addr == (long) &dummy->u_debugreg[7]) {
455 data &= ~DR_CONTROL_RESERVED;
457 if ((0x5f54 >> ((data >> (16 + 4*i)) & 0xf)) & 1)
460 set_tsk_thread_flag(child, TIF_DEBUG);
462 clear_tsk_thread_flag(child, TIF_DEBUG);
464 addr -= (long) &dummy->u_debugreg;
466 child->thread.debugreg[addr] = data;
471 case PTRACE_SYSEMU: /* continue and stop at next syscall, which will not be executed */
472 case PTRACE_SYSCALL: /* continue and stop at next (return from) syscall */
473 case PTRACE_CONT: /* restart after signal. */
475 if (!valid_signal(data))
477 if (request == PTRACE_SYSEMU) {
478 set_tsk_thread_flag(child, TIF_SYSCALL_EMU);
479 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
480 } else if (request == PTRACE_SYSCALL) {
481 set_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
482 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
484 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
485 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
487 child->exit_code = data;
488 /* make sure the single step bit is not set. */
489 clear_singlestep(child);
490 wake_up_process(child);
495 * make the child exit. Best I can do is send it a sigkill.
496 * perhaps it should be put in the status that it wants to
501 if (child->exit_state == EXIT_ZOMBIE) /* already dead */
503 child->exit_code = SIGKILL;
504 /* make sure the single step bit is not set. */
505 clear_singlestep(child);
506 wake_up_process(child);
509 case PTRACE_SYSEMU_SINGLESTEP: /* Same as SYSEMU, but singlestep if not syscall */
510 case PTRACE_SINGLESTEP: /* set the trap flag. */
512 if (!valid_signal(data))
515 if (request == PTRACE_SYSEMU_SINGLESTEP)
516 set_tsk_thread_flag(child, TIF_SYSCALL_EMU);
518 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
520 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
521 set_singlestep(child);
522 child->exit_code = data;
523 /* give it a chance to run. */
524 wake_up_process(child);
529 /* detach a process that was attached. */
530 ret = ptrace_detach(child, data);
533 case PTRACE_GETREGS: { /* Get all gp regs from the child. */
534 if (!access_ok(VERIFY_WRITE, datap, FRAME_SIZE*sizeof(long))) {
538 for ( i = 0; i < FRAME_SIZE*sizeof(long); i += sizeof(long) ) {
539 __put_user(getreg(child, i), datap);
546 case PTRACE_SETREGS: { /* Set all gp regs in the child. */
548 if (!access_ok(VERIFY_READ, datap, FRAME_SIZE*sizeof(long))) {
552 for ( i = 0; i < FRAME_SIZE*sizeof(long); i += sizeof(long) ) {
553 __get_user(tmp, datap);
554 putreg(child, i, tmp);
561 case PTRACE_GETFPREGS: { /* Get the child FPU state. */
562 if (!access_ok(VERIFY_WRITE, datap,
563 sizeof(struct user_i387_struct))) {
568 if (!tsk_used_math(child))
570 get_fpregs((struct user_i387_struct __user *)data, child);
574 case PTRACE_SETFPREGS: { /* Set the child FPU state. */
575 if (!access_ok(VERIFY_READ, datap,
576 sizeof(struct user_i387_struct))) {
580 set_stopped_child_used_math(child);
581 set_fpregs(child, (struct user_i387_struct __user *)data);
586 case PTRACE_GETFPXREGS: { /* Get the child extended FPU state. */
587 if (!access_ok(VERIFY_WRITE, datap,
588 sizeof(struct user_fxsr_struct))) {
592 if (!tsk_used_math(child))
594 ret = get_fpxregs((struct user_fxsr_struct __user *)data, child);
598 case PTRACE_SETFPXREGS: { /* Set the child extended FPU state. */
599 if (!access_ok(VERIFY_READ, datap,
600 sizeof(struct user_fxsr_struct))) {
604 set_stopped_child_used_math(child);
605 ret = set_fpxregs(child, (struct user_fxsr_struct __user *)data);
609 case PTRACE_GET_THREAD_AREA:
610 ret = ptrace_get_thread_area(child, addr,
611 (struct user_desc __user *) data);
614 case PTRACE_SET_THREAD_AREA:
615 ret = ptrace_set_thread_area(child, addr,
616 (struct user_desc __user *) data);
620 ret = ptrace_request(child, request, addr, data);
627 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs, int error_code)
631 tsk->thread.trap_no = 1;
632 tsk->thread.error_code = error_code;
634 memset(&info, 0, sizeof(info));
635 info.si_signo = SIGTRAP;
636 info.si_code = TRAP_BRKPT;
639 info.si_addr = user_mode_vm(regs) ? (void __user *) regs->eip : NULL;
641 /* Send us the fakey SIGTRAP */
642 force_sig_info(SIGTRAP, &info, tsk);
645 /* notification of system call entry/exit
646 * - triggered by current->work.syscall_trace
648 __attribute__((regparm(3)))
649 int do_syscall_trace(struct pt_regs *regs, int entryexit)
651 int is_sysemu = test_thread_flag(TIF_SYSCALL_EMU);
653 * With TIF_SYSCALL_EMU set we want to ignore TIF_SINGLESTEP for syscall
656 int is_singlestep = !is_sysemu && test_thread_flag(TIF_SINGLESTEP);
659 /* do the secure computing check first */
661 secure_computing(regs->orig_eax);
663 if (unlikely(current->audit_context)) {
665 audit_syscall_exit(AUDITSC_RESULT(regs->eax),
667 /* Debug traps, when using PTRACE_SINGLESTEP, must be sent only
668 * on the syscall exit path. Normally, when TIF_SYSCALL_AUDIT is
669 * not used, entry.S will call us only on syscall exit, not
670 * entry; so when TIF_SYSCALL_AUDIT is used we must avoid
671 * calling send_sigtrap() on syscall entry.
673 * Note that when PTRACE_SYSEMU_SINGLESTEP is used,
674 * is_singlestep is false, despite his name, so we will still do
677 else if (is_singlestep)
681 if (!(current->ptrace & PT_PTRACED))
684 /* If a process stops on the 1st tracepoint with SYSCALL_TRACE
685 * and then is resumed with SYSEMU_SINGLESTEP, it will come in
686 * here. We have to check this and return */
687 if (is_sysemu && entryexit)
690 /* Fake a debug trap */
692 send_sigtrap(current, regs, 0);
694 if (!test_thread_flag(TIF_SYSCALL_TRACE) && !is_sysemu)
697 /* the 0x80 provides a way for the tracing parent to distinguish
698 between a syscall stop and SIGTRAP delivery */
699 /* Note that the debugger could change the result of test_thread_flag!*/
700 ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD) ? 0x80:0));
703 * this isn't the same as continuing with a signal, but it will do
704 * for normal use. strace only continues with a signal if the
705 * stopping signal is not SIGTRAP. -brl
707 if (current->exit_code) {
708 send_sig(current->exit_code, current, 1);
709 current->exit_code = 0;
713 if (unlikely(current->audit_context) && !entryexit)
714 audit_syscall_entry(AUDIT_ARCH_I386, regs->orig_eax,
715 regs->ebx, regs->ecx, regs->edx, regs->esi);
719 regs->orig_eax = -1; /* force skip of syscall restarting */
720 if (unlikely(current->audit_context))
721 audit_syscall_exit(AUDITSC_RESULT(regs->eax), regs->eax);