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) {
167 down(&child->mm->context.sem);
168 desc = child->mm->context.ldt + (seg & ~7);
169 base = (desc[0] >> 16) | ((desc[1] & 0xff) << 16) | (desc[1] & 0xff000000);
171 /* 16-bit code segment? */
172 if (!((desc[1] >> 22) & 1))
175 up(&child->mm->context.sem);
180 static inline int is_setting_trap_flag(struct task_struct *child, struct pt_regs *regs)
183 unsigned char opcode[15];
184 unsigned long addr = convert_eip_to_linear(child, regs);
186 copied = access_process_vm(child, addr, opcode, sizeof(opcode), 0);
187 for (i = 0; i < copied; i++) {
190 case 0x9d: case 0xcf:
192 /* opcode and address size prefixes */
193 case 0x66: case 0x67:
195 /* irrelevant prefixes (segment overrides and repeats) */
196 case 0x26: case 0x2e:
197 case 0x36: case 0x3e:
198 case 0x64: case 0x65:
199 case 0xf0: case 0xf2: case 0xf3:
203 * pushf: NOTE! We should probably not let
204 * the user see the TF bit being set. But
205 * it's more pain than it's worth to avoid
206 * it, and a debugger could emulate this
207 * all in user space if it _really_ cares.
217 static void set_singlestep(struct task_struct *child)
219 struct pt_regs *regs = get_child_regs(child);
222 * Always set TIF_SINGLESTEP - this guarantees that
223 * we single-step system calls etc.. This will also
224 * cause us to set TF when returning to user mode.
226 set_tsk_thread_flag(child, TIF_SINGLESTEP);
229 * If TF was already set, don't do anything else
231 if (regs->eflags & TRAP_FLAG)
234 /* Set TF on the kernel stack.. */
235 regs->eflags |= TRAP_FLAG;
238 * ..but if TF is changed by the instruction we will trace,
239 * don't mark it as being "us" that set it, so that we
240 * won't clear it by hand later.
242 if (is_setting_trap_flag(child, regs))
245 child->ptrace |= PT_DTRACE;
248 static void clear_singlestep(struct task_struct *child)
250 /* Always clear TIF_SINGLESTEP... */
251 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
253 /* But touch TF only if it was set by us.. */
254 if (child->ptrace & PT_DTRACE) {
255 struct pt_regs *regs = get_child_regs(child);
256 regs->eflags &= ~TRAP_FLAG;
257 child->ptrace &= ~PT_DTRACE;
262 * Called by kernel/ptrace.c when detaching..
264 * Make sure the single step bit is not set.
266 void ptrace_disable(struct task_struct *child)
268 clear_singlestep(child);
269 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
270 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
274 * Perform get_thread_area on behalf of the traced child.
277 ptrace_get_thread_area(struct task_struct *child,
278 int idx, struct user_desc __user *user_desc)
280 struct user_desc info;
281 struct desc_struct *desc;
284 * Get the current Thread-Local Storage area:
287 #define GET_BASE(desc) ( \
288 (((desc)->a >> 16) & 0x0000ffff) | \
289 (((desc)->b << 16) & 0x00ff0000) | \
290 ( (desc)->b & 0xff000000) )
292 #define GET_LIMIT(desc) ( \
293 ((desc)->a & 0x0ffff) | \
294 ((desc)->b & 0xf0000) )
296 #define GET_32BIT(desc) (((desc)->b >> 22) & 1)
297 #define GET_CONTENTS(desc) (((desc)->b >> 10) & 3)
298 #define GET_WRITABLE(desc) (((desc)->b >> 9) & 1)
299 #define GET_LIMIT_PAGES(desc) (((desc)->b >> 23) & 1)
300 #define GET_PRESENT(desc) (((desc)->b >> 15) & 1)
301 #define GET_USEABLE(desc) (((desc)->b >> 20) & 1)
303 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
306 desc = child->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
308 info.entry_number = idx;
309 info.base_addr = GET_BASE(desc);
310 info.limit = GET_LIMIT(desc);
311 info.seg_32bit = GET_32BIT(desc);
312 info.contents = GET_CONTENTS(desc);
313 info.read_exec_only = !GET_WRITABLE(desc);
314 info.limit_in_pages = GET_LIMIT_PAGES(desc);
315 info.seg_not_present = !GET_PRESENT(desc);
316 info.useable = GET_USEABLE(desc);
318 if (copy_to_user(user_desc, &info, sizeof(info)))
325 * Perform set_thread_area on behalf of the traced child.
328 ptrace_set_thread_area(struct task_struct *child,
329 int idx, struct user_desc __user *user_desc)
331 struct user_desc info;
332 struct desc_struct *desc;
334 if (copy_from_user(&info, user_desc, sizeof(info)))
337 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
340 desc = child->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
341 if (LDT_empty(&info)) {
345 desc->a = LDT_entry_a(&info);
346 desc->b = LDT_entry_b(&info);
352 long arch_ptrace(struct task_struct *child, long request, long addr, long data)
354 struct user * dummy = NULL;
356 unsigned long __user *datap = (unsigned long __user *)data;
359 /* when I and D space are separate, these will need to be fixed. */
360 case PTRACE_PEEKTEXT: /* read word at location addr. */
361 case PTRACE_PEEKDATA: {
365 copied = access_process_vm(child, addr, &tmp, sizeof(tmp), 0);
367 if (copied != sizeof(tmp))
369 ret = put_user(tmp, datap);
373 /* read the word at location addr in the USER area. */
374 case PTRACE_PEEKUSR: {
378 if ((addr & 3) || addr < 0 ||
379 addr > sizeof(struct user) - 3)
382 tmp = 0; /* Default return condition */
383 if(addr < FRAME_SIZE*sizeof(long))
384 tmp = getreg(child, addr);
385 if(addr >= (long) &dummy->u_debugreg[0] &&
386 addr <= (long) &dummy->u_debugreg[7]){
387 addr -= (long) &dummy->u_debugreg[0];
389 tmp = child->thread.debugreg[addr];
391 ret = put_user(tmp, datap);
395 /* when I and D space are separate, this will have to be fixed. */
396 case PTRACE_POKETEXT: /* write the word at location addr. */
397 case PTRACE_POKEDATA:
399 if (access_process_vm(child, addr, &data, sizeof(data), 1) == sizeof(data))
404 case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
406 if ((addr & 3) || addr < 0 ||
407 addr > sizeof(struct user) - 3)
410 if (addr < FRAME_SIZE*sizeof(long)) {
411 ret = putreg(child, addr, data);
414 /* We need to be very careful here. We implicitly
415 want to modify a portion of the task_struct, and we
416 have to be selective about what portions we allow someone
420 if(addr >= (long) &dummy->u_debugreg[0] &&
421 addr <= (long) &dummy->u_debugreg[7]){
423 if(addr == (long) &dummy->u_debugreg[4]) break;
424 if(addr == (long) &dummy->u_debugreg[5]) break;
425 if(addr < (long) &dummy->u_debugreg[4] &&
426 ((unsigned long) data) >= TASK_SIZE-3) break;
428 /* Sanity-check data. Take one half-byte at once with
429 * check = (val >> (16 + 4*i)) & 0xf. It contains the
430 * R/Wi and LENi bits; bits 0 and 1 are R/Wi, and bits
431 * 2 and 3 are LENi. Given a list of invalid values,
432 * we do mask |= 1 << invalid_value, so that
433 * (mask >> check) & 1 is a correct test for invalid
436 * R/Wi contains the type of the breakpoint /
437 * watchpoint, LENi contains the length of the watched
438 * data in the watchpoint case.
440 * The invalid values are:
441 * - LENi == 0x10 (undefined), so mask |= 0x0f00.
442 * - R/Wi == 0x10 (break on I/O reads or writes), so
444 * - R/Wi == 0x00 && LENi != 0x00, so we have mask |=
447 * Finally, mask = 0x0f00 | 0x4444 | 0x1110 == 0x5f54.
449 * See the Intel Manual "System Programming Guide",
452 * Note that LENi == 0x10 is defined on x86_64 in long
453 * mode (i.e. even for 32-bit userspace software, but
454 * 64-bit kernel), so the x86_64 mask value is 0x5454.
455 * See the AMD manual no. 24593 (AMD64 System
458 if(addr == (long) &dummy->u_debugreg[7]) {
459 data &= ~DR_CONTROL_RESERVED;
461 if ((0x5f54 >> ((data >> (16 + 4*i)) & 0xf)) & 1)
464 set_tsk_thread_flag(child, TIF_DEBUG);
466 clear_tsk_thread_flag(child, TIF_DEBUG);
468 addr -= (long) &dummy->u_debugreg;
470 child->thread.debugreg[addr] = data;
475 case PTRACE_SYSEMU: /* continue and stop at next syscall, which will not be executed */
476 case PTRACE_SYSCALL: /* continue and stop at next (return from) syscall */
477 case PTRACE_CONT: /* restart after signal. */
479 if (!valid_signal(data))
481 if (request == PTRACE_SYSEMU) {
482 set_tsk_thread_flag(child, TIF_SYSCALL_EMU);
483 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
484 } else if (request == PTRACE_SYSCALL) {
485 set_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
486 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
488 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
489 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
491 child->exit_code = data;
492 /* make sure the single step bit is not set. */
493 clear_singlestep(child);
494 wake_up_process(child);
499 * make the child exit. Best I can do is send it a sigkill.
500 * perhaps it should be put in the status that it wants to
505 if (child->exit_state == EXIT_ZOMBIE) /* already dead */
507 child->exit_code = SIGKILL;
508 /* make sure the single step bit is not set. */
509 clear_singlestep(child);
510 wake_up_process(child);
513 case PTRACE_SYSEMU_SINGLESTEP: /* Same as SYSEMU, but singlestep if not syscall */
514 case PTRACE_SINGLESTEP: /* set the trap flag. */
516 if (!valid_signal(data))
519 if (request == PTRACE_SYSEMU_SINGLESTEP)
520 set_tsk_thread_flag(child, TIF_SYSCALL_EMU);
522 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
524 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
525 set_singlestep(child);
526 child->exit_code = data;
527 /* give it a chance to run. */
528 wake_up_process(child);
533 /* detach a process that was attached. */
534 ret = ptrace_detach(child, data);
537 case PTRACE_GETREGS: { /* Get all gp regs from the child. */
538 if (!access_ok(VERIFY_WRITE, datap, FRAME_SIZE*sizeof(long))) {
542 for ( i = 0; i < FRAME_SIZE*sizeof(long); i += sizeof(long) ) {
543 __put_user(getreg(child, i), datap);
550 case PTRACE_SETREGS: { /* Set all gp regs in the child. */
552 if (!access_ok(VERIFY_READ, datap, FRAME_SIZE*sizeof(long))) {
556 for ( i = 0; i < FRAME_SIZE*sizeof(long); i += sizeof(long) ) {
557 __get_user(tmp, datap);
558 putreg(child, i, tmp);
565 case PTRACE_GETFPREGS: { /* Get the child FPU state. */
566 if (!access_ok(VERIFY_WRITE, datap,
567 sizeof(struct user_i387_struct))) {
572 if (!tsk_used_math(child))
574 get_fpregs((struct user_i387_struct __user *)data, child);
578 case PTRACE_SETFPREGS: { /* Set the child FPU state. */
579 if (!access_ok(VERIFY_READ, datap,
580 sizeof(struct user_i387_struct))) {
584 set_stopped_child_used_math(child);
585 set_fpregs(child, (struct user_i387_struct __user *)data);
590 case PTRACE_GETFPXREGS: { /* Get the child extended FPU state. */
591 if (!access_ok(VERIFY_WRITE, datap,
592 sizeof(struct user_fxsr_struct))) {
596 if (!tsk_used_math(child))
598 ret = get_fpxregs((struct user_fxsr_struct __user *)data, child);
602 case PTRACE_SETFPXREGS: { /* Set the child extended FPU state. */
603 if (!access_ok(VERIFY_READ, datap,
604 sizeof(struct user_fxsr_struct))) {
608 set_stopped_child_used_math(child);
609 ret = set_fpxregs(child, (struct user_fxsr_struct __user *)data);
613 case PTRACE_GET_THREAD_AREA:
614 ret = ptrace_get_thread_area(child, addr,
615 (struct user_desc __user *) data);
618 case PTRACE_SET_THREAD_AREA:
619 ret = ptrace_set_thread_area(child, addr,
620 (struct user_desc __user *) data);
624 ret = ptrace_request(child, request, addr, data);
631 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs, int error_code)
635 tsk->thread.trap_no = 1;
636 tsk->thread.error_code = error_code;
638 memset(&info, 0, sizeof(info));
639 info.si_signo = SIGTRAP;
640 info.si_code = TRAP_BRKPT;
643 info.si_addr = user_mode_vm(regs) ? (void __user *) regs->eip : NULL;
645 /* Send us the fakey SIGTRAP */
646 force_sig_info(SIGTRAP, &info, tsk);
649 /* notification of system call entry/exit
650 * - triggered by current->work.syscall_trace
652 __attribute__((regparm(3)))
653 int do_syscall_trace(struct pt_regs *regs, int entryexit)
655 int is_sysemu = test_thread_flag(TIF_SYSCALL_EMU);
657 * With TIF_SYSCALL_EMU set we want to ignore TIF_SINGLESTEP for syscall
660 int is_singlestep = !is_sysemu && test_thread_flag(TIF_SINGLESTEP);
663 /* do the secure computing check first */
665 secure_computing(regs->orig_eax);
667 if (unlikely(current->audit_context)) {
669 audit_syscall_exit(AUDITSC_RESULT(regs->eax),
671 /* Debug traps, when using PTRACE_SINGLESTEP, must be sent only
672 * on the syscall exit path. Normally, when TIF_SYSCALL_AUDIT is
673 * not used, entry.S will call us only on syscall exit, not
674 * entry; so when TIF_SYSCALL_AUDIT is used we must avoid
675 * calling send_sigtrap() on syscall entry.
677 * Note that when PTRACE_SYSEMU_SINGLESTEP is used,
678 * is_singlestep is false, despite his name, so we will still do
681 else if (is_singlestep)
685 if (!(current->ptrace & PT_PTRACED))
688 /* If a process stops on the 1st tracepoint with SYSCALL_TRACE
689 * and then is resumed with SYSEMU_SINGLESTEP, it will come in
690 * here. We have to check this and return */
691 if (is_sysemu && entryexit)
694 /* Fake a debug trap */
696 send_sigtrap(current, regs, 0);
698 if (!test_thread_flag(TIF_SYSCALL_TRACE) && !is_sysemu)
701 /* the 0x80 provides a way for the tracing parent to distinguish
702 between a syscall stop and SIGTRAP delivery */
703 /* Note that the debugger could change the result of test_thread_flag!*/
704 ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD) ? 0x80:0));
707 * this isn't the same as continuing with a signal, but it will do
708 * for normal use. strace only continues with a signal if the
709 * stopping signal is not SIGTRAP. -brl
711 if (current->exit_code) {
712 send_sig(current->exit_code, current, 1);
713 current->exit_code = 0;
717 if (unlikely(current->audit_context) && !entryexit)
718 audit_syscall_entry(AUDIT_ARCH_I386, regs->orig_eax,
719 regs->ebx, regs->ecx, regs->edx, regs->esi);
723 regs->orig_eax = -1; /* force skip of syscall restarting */
724 if (unlikely(current->audit_context))
725 audit_syscall_exit(AUDITSC_RESULT(regs->eax), regs->eax);