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/ptrace.h>
13 #include <linux/user.h>
14 #include <linux/security.h>
15 #include <linux/audit.h>
16 #include <linux/seccomp.h>
17 #include <linux/signal.h>
19 #include <asm/uaccess.h>
20 #include <asm/pgtable.h>
21 #include <asm/system.h>
22 #include <asm/processor.h>
24 #include <asm/debugreg.h>
29 * does not yet catch signals sent when the child dies.
30 * in exit.c or in signal.c.
34 * Determines which flags the user has access to [1 = access, 0 = no access].
35 * Prohibits changing ID(21), VIP(20), VIF(19), VM(17), NT(14), IOPL(12-13), IF(9).
36 * Also masks reserved bits (31-22, 15, 5, 3, 1).
38 #define FLAG_MASK 0x00050dd5
41 * Offset of eflags on child stack..
43 #define EFL_OFFSET offsetof(struct pt_regs, eflags)
45 static inline struct pt_regs *get_child_regs(struct task_struct *task)
47 void *stack_top = (void *)task->thread.esp0;
48 return stack_top - sizeof(struct pt_regs);
52 * This routine will get a word off of the processes privileged stack.
53 * the offset is bytes into the pt_regs structure on the stack.
54 * This routine assumes that all the privileged stacks are in our
57 static inline int get_stack_long(struct task_struct *task, int offset)
61 stack = (unsigned char *)task->thread.esp0 - sizeof(struct pt_regs);
63 return (*((int *)stack));
67 * This routine will put a word on the processes privileged stack.
68 * the offset is bytes into the pt_regs structure on the stack.
69 * This routine assumes that all the privileged stacks are in our
72 static inline int put_stack_long(struct task_struct *task, int offset,
75 unsigned char * stack;
77 stack = (unsigned char *)task->thread.esp0 - sizeof(struct pt_regs);
79 *(unsigned long *) stack = data;
83 static int putreg(struct task_struct *child,
84 unsigned long regno, unsigned long value)
88 if (value && (value & 3) != 3)
90 child->thread.gs = value;
95 if (value && (value & 3) != 3)
101 if ((value & 3) != 3)
107 value |= get_stack_long(child, EFL_OFFSET) & ~FLAG_MASK;
112 put_stack_long(child, regno, value);
116 static unsigned long getreg(struct task_struct *child,
119 unsigned long retval = ~0UL;
121 switch (regno >> 2) {
123 retval = child->thread.gs;
135 retval &= get_stack_long(child, regno);
140 #define LDT_SEGMENT 4
142 static unsigned long convert_eip_to_linear(struct task_struct *child, struct pt_regs *regs)
144 unsigned long addr, seg;
147 seg = regs->xcs & 0xffff;
148 if (regs->eflags & VM_MASK) {
149 addr = (addr & 0xffff) + (seg << 4);
154 * We'll assume that the code segments in the GDT
155 * are all zero-based. That is largely true: the
156 * TLS segments are used for data, and the PNPBIOS
157 * and APM bios ones we just ignore here.
159 if (seg & LDT_SEGMENT) {
165 mutex_lock(&child->mm->context.lock);
166 if (unlikely((seg >> 3) >= child->mm->context.size))
167 addr = -1L; /* bogus selector, access would fault */
169 desc = child->mm->context.ldt + seg;
170 base = ((desc[0] >> 16) |
171 ((desc[1] & 0xff) << 16) |
172 (desc[1] & 0xff000000));
174 /* 16-bit code segment? */
175 if (!((desc[1] >> 22) & 1))
179 mutex_unlock(&child->mm->context.lock);
184 static inline int is_setting_trap_flag(struct task_struct *child, struct pt_regs *regs)
187 unsigned char opcode[15];
188 unsigned long addr = convert_eip_to_linear(child, regs);
190 copied = access_process_vm(child, addr, opcode, sizeof(opcode), 0);
191 for (i = 0; i < copied; i++) {
194 case 0x9d: case 0xcf:
196 /* opcode and address size prefixes */
197 case 0x66: case 0x67:
199 /* irrelevant prefixes (segment overrides and repeats) */
200 case 0x26: case 0x2e:
201 case 0x36: case 0x3e:
202 case 0x64: case 0x65:
203 case 0xf0: case 0xf2: case 0xf3:
207 * pushf: NOTE! We should probably not let
208 * the user see the TF bit being set. But
209 * it's more pain than it's worth to avoid
210 * it, and a debugger could emulate this
211 * all in user space if it _really_ cares.
221 static void set_singlestep(struct task_struct *child)
223 struct pt_regs *regs = get_child_regs(child);
226 * Always set TIF_SINGLESTEP - this guarantees that
227 * we single-step system calls etc.. This will also
228 * cause us to set TF when returning to user mode.
230 set_tsk_thread_flag(child, TIF_SINGLESTEP);
233 * If TF was already set, don't do anything else
235 if (regs->eflags & X86_EFLAGS_TF)
238 /* Set TF on the kernel stack.. */
239 regs->eflags |= X86_EFLAGS_TF;
242 * ..but if TF is changed by the instruction we will trace,
243 * don't mark it as being "us" that set it, so that we
244 * won't clear it by hand later.
246 if (is_setting_trap_flag(child, regs))
249 child->ptrace |= PT_DTRACE;
252 static void clear_singlestep(struct task_struct *child)
254 /* Always clear TIF_SINGLESTEP... */
255 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
257 /* But touch TF only if it was set by us.. */
258 if (child->ptrace & PT_DTRACE) {
259 struct pt_regs *regs = get_child_regs(child);
260 regs->eflags &= ~X86_EFLAGS_TF;
261 child->ptrace &= ~PT_DTRACE;
266 * Called by kernel/ptrace.c when detaching..
268 * Make sure the single step bit is not set.
270 void ptrace_disable(struct task_struct *child)
272 clear_singlestep(child);
273 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
276 long arch_ptrace(struct task_struct *child, long request, long addr, long data)
278 struct user * dummy = NULL;
280 unsigned long __user *datap = (unsigned long __user *)data;
283 /* when I and D space are separate, these will need to be fixed. */
284 case PTRACE_PEEKTEXT: /* read word at location addr. */
285 case PTRACE_PEEKDATA:
286 ret = generic_ptrace_peekdata(child, addr, data);
289 /* read the word at location addr in the USER area. */
290 case PTRACE_PEEKUSR: {
294 if ((addr & 3) || addr < 0 ||
295 addr > sizeof(struct user) - 3)
298 tmp = 0; /* Default return condition */
299 if(addr < FRAME_SIZE*sizeof(long))
300 tmp = getreg(child, addr);
301 if(addr >= (long) &dummy->u_debugreg[0] &&
302 addr <= (long) &dummy->u_debugreg[7]){
303 addr -= (long) &dummy->u_debugreg[0];
305 tmp = child->thread.debugreg[addr];
307 ret = put_user(tmp, datap);
311 /* when I and D space are separate, this will have to be fixed. */
312 case PTRACE_POKETEXT: /* write the word at location addr. */
313 case PTRACE_POKEDATA:
314 ret = generic_ptrace_pokedata(child, addr, data);
317 case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
319 if ((addr & 3) || addr < 0 ||
320 addr > sizeof(struct user) - 3)
323 if (addr < FRAME_SIZE*sizeof(long)) {
324 ret = putreg(child, addr, data);
327 /* We need to be very careful here. We implicitly
328 want to modify a portion of the task_struct, and we
329 have to be selective about what portions we allow someone
333 if(addr >= (long) &dummy->u_debugreg[0] &&
334 addr <= (long) &dummy->u_debugreg[7]){
336 if(addr == (long) &dummy->u_debugreg[4]) break;
337 if(addr == (long) &dummy->u_debugreg[5]) break;
338 if(addr < (long) &dummy->u_debugreg[4] &&
339 ((unsigned long) data) >= TASK_SIZE-3) break;
341 /* Sanity-check data. Take one half-byte at once with
342 * check = (val >> (16 + 4*i)) & 0xf. It contains the
343 * R/Wi and LENi bits; bits 0 and 1 are R/Wi, and bits
344 * 2 and 3 are LENi. Given a list of invalid values,
345 * we do mask |= 1 << invalid_value, so that
346 * (mask >> check) & 1 is a correct test for invalid
349 * R/Wi contains the type of the breakpoint /
350 * watchpoint, LENi contains the length of the watched
351 * data in the watchpoint case.
353 * The invalid values are:
354 * - LENi == 0x10 (undefined), so mask |= 0x0f00.
355 * - R/Wi == 0x10 (break on I/O reads or writes), so
357 * - R/Wi == 0x00 && LENi != 0x00, so we have mask |=
360 * Finally, mask = 0x0f00 | 0x4444 | 0x1110 == 0x5f54.
362 * See the Intel Manual "System Programming Guide",
365 * Note that LENi == 0x10 is defined on x86_64 in long
366 * mode (i.e. even for 32-bit userspace software, but
367 * 64-bit kernel), so the x86_64 mask value is 0x5454.
368 * See the AMD manual no. 24593 (AMD64 System
371 if(addr == (long) &dummy->u_debugreg[7]) {
372 data &= ~DR_CONTROL_RESERVED;
374 if ((0x5f54 >> ((data >> (16 + 4*i)) & 0xf)) & 1)
377 set_tsk_thread_flag(child, TIF_DEBUG);
379 clear_tsk_thread_flag(child, TIF_DEBUG);
381 addr -= (long) &dummy->u_debugreg;
383 child->thread.debugreg[addr] = data;
388 case PTRACE_SYSEMU: /* continue and stop at next syscall, which will not be executed */
389 case PTRACE_SYSCALL: /* continue and stop at next (return from) syscall */
390 case PTRACE_CONT: /* restart after signal. */
392 if (!valid_signal(data))
394 if (request == PTRACE_SYSEMU) {
395 set_tsk_thread_flag(child, TIF_SYSCALL_EMU);
396 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
397 } else if (request == PTRACE_SYSCALL) {
398 set_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
399 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
401 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
402 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
404 child->exit_code = data;
405 /* make sure the single step bit is not set. */
406 clear_singlestep(child);
407 wake_up_process(child);
412 * make the child exit. Best I can do is send it a sigkill.
413 * perhaps it should be put in the status that it wants to
418 if (child->exit_state == EXIT_ZOMBIE) /* already dead */
420 child->exit_code = SIGKILL;
421 /* make sure the single step bit is not set. */
422 clear_singlestep(child);
423 wake_up_process(child);
426 case PTRACE_SYSEMU_SINGLESTEP: /* Same as SYSEMU, but singlestep if not syscall */
427 case PTRACE_SINGLESTEP: /* set the trap flag. */
429 if (!valid_signal(data))
432 if (request == PTRACE_SYSEMU_SINGLESTEP)
433 set_tsk_thread_flag(child, TIF_SYSCALL_EMU);
435 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
437 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
438 set_singlestep(child);
439 child->exit_code = data;
440 /* give it a chance to run. */
441 wake_up_process(child);
445 case PTRACE_GETREGS: { /* Get all gp regs from the child. */
446 if (!access_ok(VERIFY_WRITE, datap, FRAME_SIZE*sizeof(long))) {
450 for ( i = 0; i < FRAME_SIZE*sizeof(long); i += sizeof(long) ) {
451 __put_user(getreg(child, i), datap);
458 case PTRACE_SETREGS: { /* Set all gp regs in the child. */
460 if (!access_ok(VERIFY_READ, datap, FRAME_SIZE*sizeof(long))) {
464 for ( i = 0; i < FRAME_SIZE*sizeof(long); i += sizeof(long) ) {
465 __get_user(tmp, datap);
466 putreg(child, i, tmp);
473 case PTRACE_GETFPREGS: { /* Get the child FPU state. */
474 if (!access_ok(VERIFY_WRITE, datap,
475 sizeof(struct user_i387_struct))) {
480 if (!tsk_used_math(child))
482 get_fpregs((struct user_i387_struct __user *)data, child);
486 case PTRACE_SETFPREGS: { /* Set the child FPU state. */
487 if (!access_ok(VERIFY_READ, datap,
488 sizeof(struct user_i387_struct))) {
492 set_stopped_child_used_math(child);
493 set_fpregs(child, (struct user_i387_struct __user *)data);
498 case PTRACE_GETFPXREGS: { /* Get the child extended FPU state. */
499 if (!access_ok(VERIFY_WRITE, datap,
500 sizeof(struct user_fxsr_struct))) {
504 if (!tsk_used_math(child))
506 ret = get_fpxregs((struct user_fxsr_struct __user *)data, child);
510 case PTRACE_SETFPXREGS: { /* Set the child extended FPU state. */
511 if (!access_ok(VERIFY_READ, datap,
512 sizeof(struct user_fxsr_struct))) {
516 set_stopped_child_used_math(child);
517 ret = set_fpxregs(child, (struct user_fxsr_struct __user *)data);
521 case PTRACE_GET_THREAD_AREA:
524 ret = do_get_thread_area(child, addr,
525 (struct user_desc __user *) data);
528 case PTRACE_SET_THREAD_AREA:
531 ret = do_set_thread_area(child, addr,
532 (struct user_desc __user *) data, 0);
536 ret = ptrace_request(child, request, addr, data);
543 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs, int error_code)
547 tsk->thread.trap_no = 1;
548 tsk->thread.error_code = error_code;
550 memset(&info, 0, sizeof(info));
551 info.si_signo = SIGTRAP;
552 info.si_code = TRAP_BRKPT;
555 info.si_addr = user_mode_vm(regs) ? (void __user *) regs->eip : NULL;
557 /* Send us the fake SIGTRAP */
558 force_sig_info(SIGTRAP, &info, tsk);
561 /* notification of system call entry/exit
562 * - triggered by current->work.syscall_trace
564 __attribute__((regparm(3)))
565 int do_syscall_trace(struct pt_regs *regs, int entryexit)
567 int is_sysemu = test_thread_flag(TIF_SYSCALL_EMU);
569 * With TIF_SYSCALL_EMU set we want to ignore TIF_SINGLESTEP for syscall
572 int is_singlestep = !is_sysemu && test_thread_flag(TIF_SINGLESTEP);
575 /* do the secure computing check first */
577 secure_computing(regs->orig_eax);
579 if (unlikely(current->audit_context)) {
581 audit_syscall_exit(AUDITSC_RESULT(regs->eax),
583 /* Debug traps, when using PTRACE_SINGLESTEP, must be sent only
584 * on the syscall exit path. Normally, when TIF_SYSCALL_AUDIT is
585 * not used, entry.S will call us only on syscall exit, not
586 * entry; so when TIF_SYSCALL_AUDIT is used we must avoid
587 * calling send_sigtrap() on syscall entry.
589 * Note that when PTRACE_SYSEMU_SINGLESTEP is used,
590 * is_singlestep is false, despite his name, so we will still do
593 else if (is_singlestep)
597 if (!(current->ptrace & PT_PTRACED))
600 /* If a process stops on the 1st tracepoint with SYSCALL_TRACE
601 * and then is resumed with SYSEMU_SINGLESTEP, it will come in
602 * here. We have to check this and return */
603 if (is_sysemu && entryexit)
606 /* Fake a debug trap */
608 send_sigtrap(current, regs, 0);
610 if (!test_thread_flag(TIF_SYSCALL_TRACE) && !is_sysemu)
613 /* the 0x80 provides a way for the tracing parent to distinguish
614 between a syscall stop and SIGTRAP delivery */
615 /* Note that the debugger could change the result of test_thread_flag!*/
616 ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD) ? 0x80:0));
619 * this isn't the same as continuing with a signal, but it will do
620 * for normal use. strace only continues with a signal if the
621 * stopping signal is not SIGTRAP. -brl
623 if (current->exit_code) {
624 send_sig(current->exit_code, current, 1);
625 current->exit_code = 0;
629 if (unlikely(current->audit_context) && !entryexit)
630 audit_syscall_entry(AUDIT_ARCH_I386, regs->orig_eax,
631 regs->ebx, regs->ecx, regs->edx, regs->esi);
635 regs->orig_eax = -1; /* force skip of syscall restarting */
636 if (unlikely(current->audit_context))
637 audit_syscall_exit(AUDITSC_RESULT(regs->eax), regs->eax);