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/smp_lock.h>
13 #include <linux/errno.h>
14 #include <linux/ptrace.h>
15 #include <linux/user.h>
16 #include <linux/security.h>
17 #include <linux/audit.h>
18 #include <linux/seccomp.h>
19 #include <linux/signal.h>
21 #include <asm/uaccess.h>
22 #include <asm/pgtable.h>
23 #include <asm/system.h>
24 #include <asm/processor.h>
26 #include <asm/debugreg.h>
31 * does not yet catch signals sent when the child dies.
32 * in exit.c or in signal.c.
36 * Determines which flags the user has access to [1 = access, 0 = no access].
37 * Prohibits changing ID(21), VIP(20), VIF(19), VM(17), NT(14), IOPL(12-13), IF(9).
38 * Also masks reserved bits (31-22, 15, 5, 3, 1).
40 #define FLAG_MASK 0x00050dd5
42 /* set's the trap flag. */
43 #define TRAP_FLAG 0x100
46 * Offset of eflags on child stack..
48 #define EFL_OFFSET ((EFL-2)*4-sizeof(struct pt_regs))
50 static inline struct pt_regs *get_child_regs(struct task_struct *task)
52 void *stack_top = (void *)task->thread.esp0;
53 return stack_top - sizeof(struct pt_regs);
57 * this routine will get a word off of the processes privileged stack.
58 * the offset is how far from the base addr as stored in the TSS.
59 * this routine assumes that all the privileged stacks are in our
62 static inline int get_stack_long(struct task_struct *task, int offset)
66 stack = (unsigned char *)task->thread.esp0;
68 return (*((int *)stack));
72 * this routine will put a word on the processes privileged stack.
73 * the offset is how far from the base addr as stored in the TSS.
74 * this routine assumes that all the privileged stacks are in our
77 static inline int put_stack_long(struct task_struct *task, int offset,
80 unsigned char * stack;
82 stack = (unsigned char *) task->thread.esp0;
84 *(unsigned long *) stack = data;
88 static int putreg(struct task_struct *child,
89 unsigned long regno, unsigned long value)
93 if (value && (value & 3) != 3)
95 child->thread.fs = value;
98 if (value && (value & 3) != 3)
100 child->thread.gs = value;
104 if (value && (value & 3) != 3)
110 if ((value & 3) != 3)
116 value |= get_stack_long(child, EFL_OFFSET) & ~FLAG_MASK;
121 put_stack_long(child, regno - sizeof(struct pt_regs), value);
125 static unsigned long getreg(struct task_struct *child,
128 unsigned long retval = ~0UL;
130 switch (regno >> 2) {
132 retval = child->thread.fs;
135 retval = child->thread.gs;
146 regno = regno - sizeof(struct pt_regs);
147 retval &= get_stack_long(child, regno);
152 #define LDT_SEGMENT 4
154 static unsigned long convert_eip_to_linear(struct task_struct *child, struct pt_regs *regs)
156 unsigned long addr, seg;
159 seg = regs->xcs & 0xffff;
160 if (regs->eflags & VM_MASK) {
161 addr = (addr & 0xffff) + (seg << 4);
166 * We'll assume that the code segments in the GDT
167 * are all zero-based. That is largely true: the
168 * TLS segments are used for data, and the PNPBIOS
169 * and APM bios ones we just ignore here.
171 if (seg & LDT_SEGMENT) {
175 down(&child->mm->context.sem);
176 desc = child->mm->context.ldt + (seg & ~7);
177 base = (desc[0] >> 16) | ((desc[1] & 0xff) << 16) | (desc[1] & 0xff000000);
179 /* 16-bit code segment? */
180 if (!((desc[1] >> 22) & 1))
183 up(&child->mm->context.sem);
188 static inline int is_at_popf(struct task_struct *child, struct pt_regs *regs)
191 unsigned char opcode[16];
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++) {
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_at_popf(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_TRACE);
278 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
282 * Perform get_thread_area on behalf of the traced child.
285 ptrace_get_thread_area(struct task_struct *child,
286 int idx, struct user_desc __user *user_desc)
288 struct user_desc info;
289 struct desc_struct *desc;
292 * Get the current Thread-Local Storage area:
295 #define GET_BASE(desc) ( \
296 (((desc)->a >> 16) & 0x0000ffff) | \
297 (((desc)->b << 16) & 0x00ff0000) | \
298 ( (desc)->b & 0xff000000) )
300 #define GET_LIMIT(desc) ( \
301 ((desc)->a & 0x0ffff) | \
302 ((desc)->b & 0xf0000) )
304 #define GET_32BIT(desc) (((desc)->b >> 22) & 1)
305 #define GET_CONTENTS(desc) (((desc)->b >> 10) & 3)
306 #define GET_WRITABLE(desc) (((desc)->b >> 9) & 1)
307 #define GET_LIMIT_PAGES(desc) (((desc)->b >> 23) & 1)
308 #define GET_PRESENT(desc) (((desc)->b >> 15) & 1)
309 #define GET_USEABLE(desc) (((desc)->b >> 20) & 1)
311 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
314 desc = child->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
316 info.entry_number = idx;
317 info.base_addr = GET_BASE(desc);
318 info.limit = GET_LIMIT(desc);
319 info.seg_32bit = GET_32BIT(desc);
320 info.contents = GET_CONTENTS(desc);
321 info.read_exec_only = !GET_WRITABLE(desc);
322 info.limit_in_pages = GET_LIMIT_PAGES(desc);
323 info.seg_not_present = !GET_PRESENT(desc);
324 info.useable = GET_USEABLE(desc);
326 if (copy_to_user(user_desc, &info, sizeof(info)))
333 * Perform set_thread_area on behalf of the traced child.
336 ptrace_set_thread_area(struct task_struct *child,
337 int idx, struct user_desc __user *user_desc)
339 struct user_desc info;
340 struct desc_struct *desc;
342 if (copy_from_user(&info, user_desc, sizeof(info)))
345 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
348 desc = child->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
349 if (LDT_empty(&info)) {
353 desc->a = LDT_entry_a(&info);
354 desc->b = LDT_entry_b(&info);
360 long arch_ptrace(struct task_struct *child, long request, long addr, long data)
362 struct user * dummy = NULL;
364 unsigned long __user *datap = (unsigned long __user *)data;
367 /* when I and D space are separate, these will need to be fixed. */
368 case PTRACE_PEEKTEXT: /* read word at location addr. */
369 case PTRACE_PEEKDATA: {
373 copied = access_process_vm(child, addr, &tmp, sizeof(tmp), 0);
375 if (copied != sizeof(tmp))
377 ret = put_user(tmp, datap);
381 /* read the word at location addr in the USER area. */
382 case PTRACE_PEEKUSR: {
386 if ((addr & 3) || addr < 0 ||
387 addr > sizeof(struct user) - 3)
390 tmp = 0; /* Default return condition */
391 if(addr < FRAME_SIZE*sizeof(long))
392 tmp = getreg(child, addr);
393 if(addr >= (long) &dummy->u_debugreg[0] &&
394 addr <= (long) &dummy->u_debugreg[7]){
395 addr -= (long) &dummy->u_debugreg[0];
397 tmp = child->thread.debugreg[addr];
399 ret = put_user(tmp, datap);
403 /* when I and D space are separate, this will have to be fixed. */
404 case PTRACE_POKETEXT: /* write the word at location addr. */
405 case PTRACE_POKEDATA:
407 if (access_process_vm(child, addr, &data, sizeof(data), 1) == sizeof(data))
412 case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
414 if ((addr & 3) || addr < 0 ||
415 addr > sizeof(struct user) - 3)
418 if (addr < FRAME_SIZE*sizeof(long)) {
419 ret = putreg(child, addr, data);
422 /* We need to be very careful here. We implicitly
423 want to modify a portion of the task_struct, and we
424 have to be selective about what portions we allow someone
428 if(addr >= (long) &dummy->u_debugreg[0] &&
429 addr <= (long) &dummy->u_debugreg[7]){
431 if(addr == (long) &dummy->u_debugreg[4]) break;
432 if(addr == (long) &dummy->u_debugreg[5]) break;
433 if(addr < (long) &dummy->u_debugreg[4] &&
434 ((unsigned long) data) >= TASK_SIZE-3) break;
436 /* Sanity-check data. Take one half-byte at once with
437 * check = (val >> (16 + 4*i)) & 0xf. It contains the
438 * R/Wi and LENi bits; bits 0 and 1 are R/Wi, and bits
439 * 2 and 3 are LENi. Given a list of invalid values,
440 * we do mask |= 1 << invalid_value, so that
441 * (mask >> check) & 1 is a correct test for invalid
444 * R/Wi contains the type of the breakpoint /
445 * watchpoint, LENi contains the length of the watched
446 * data in the watchpoint case.
448 * The invalid values are:
449 * - LENi == 0x10 (undefined), so mask |= 0x0f00.
450 * - R/Wi == 0x10 (break on I/O reads or writes), so
452 * - R/Wi == 0x00 && LENi != 0x00, so we have mask |=
455 * Finally, mask = 0x0f00 | 0x4444 | 0x1110 == 0x5f54.
457 * See the Intel Manual "System Programming Guide",
460 * Note that LENi == 0x10 is defined on x86_64 in long
461 * mode (i.e. even for 32-bit userspace software, but
462 * 64-bit kernel), so the x86_64 mask value is 0x5454.
463 * See the AMD manual no. 24593 (AMD64 System
466 if(addr == (long) &dummy->u_debugreg[7]) {
467 data &= ~DR_CONTROL_RESERVED;
469 if ((0x5f54 >> ((data >> (16 + 4*i)) & 0xf)) & 1)
473 addr -= (long) &dummy->u_debugreg;
475 child->thread.debugreg[addr] = data;
480 case PTRACE_SYSEMU: /* continue and stop at next syscall, which will not be executed */
481 case PTRACE_SYSCALL: /* continue and stop at next (return from) syscall */
482 case PTRACE_CONT: /* restart after signal. */
484 if (!valid_signal(data))
486 if (request == PTRACE_SYSEMU) {
487 set_tsk_thread_flag(child, TIF_SYSCALL_EMU);
488 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
489 } else if (request == PTRACE_SYSCALL) {
490 set_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
491 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
493 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
494 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
496 child->exit_code = data;
497 /* make sure the single step bit is not set. */
498 clear_singlestep(child);
499 wake_up_process(child);
504 * make the child exit. Best I can do is send it a sigkill.
505 * perhaps it should be put in the status that it wants to
510 if (child->exit_state == EXIT_ZOMBIE) /* already dead */
512 child->exit_code = SIGKILL;
513 /* make sure the single step bit is not set. */
514 clear_singlestep(child);
515 wake_up_process(child);
518 case PTRACE_SYSEMU_SINGLESTEP: /* Same as SYSEMU, but singlestep if not syscall */
519 case PTRACE_SINGLESTEP: /* set the trap flag. */
521 if (!valid_signal(data))
524 if (request == PTRACE_SYSEMU_SINGLESTEP)
525 set_tsk_thread_flag(child, TIF_SYSCALL_EMU);
527 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
529 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
530 set_singlestep(child);
531 child->exit_code = data;
532 /* give it a chance to run. */
533 wake_up_process(child);
538 /* detach a process that was attached. */
539 ret = ptrace_detach(child, data);
542 case PTRACE_GETREGS: { /* Get all gp regs from the child. */
543 if (!access_ok(VERIFY_WRITE, datap, FRAME_SIZE*sizeof(long))) {
547 for ( i = 0; i < FRAME_SIZE*sizeof(long); i += sizeof(long) ) {
548 __put_user(getreg(child, i), datap);
555 case PTRACE_SETREGS: { /* Set all gp regs in the child. */
557 if (!access_ok(VERIFY_READ, datap, FRAME_SIZE*sizeof(long))) {
561 for ( i = 0; i < FRAME_SIZE*sizeof(long); i += sizeof(long) ) {
562 __get_user(tmp, datap);
563 putreg(child, i, tmp);
570 case PTRACE_GETFPREGS: { /* Get the child FPU state. */
571 if (!access_ok(VERIFY_WRITE, datap,
572 sizeof(struct user_i387_struct))) {
577 if (!tsk_used_math(child))
579 get_fpregs((struct user_i387_struct __user *)data, child);
583 case PTRACE_SETFPREGS: { /* Set the child FPU state. */
584 if (!access_ok(VERIFY_READ, datap,
585 sizeof(struct user_i387_struct))) {
589 set_stopped_child_used_math(child);
590 set_fpregs(child, (struct user_i387_struct __user *)data);
595 case PTRACE_GETFPXREGS: { /* Get the child extended FPU state. */
596 if (!access_ok(VERIFY_WRITE, datap,
597 sizeof(struct user_fxsr_struct))) {
601 if (!tsk_used_math(child))
603 ret = get_fpxregs((struct user_fxsr_struct __user *)data, child);
607 case PTRACE_SETFPXREGS: { /* Set the child extended FPU state. */
608 if (!access_ok(VERIFY_READ, datap,
609 sizeof(struct user_fxsr_struct))) {
613 set_stopped_child_used_math(child);
614 ret = set_fpxregs(child, (struct user_fxsr_struct __user *)data);
618 case PTRACE_GET_THREAD_AREA:
619 ret = ptrace_get_thread_area(child, addr,
620 (struct user_desc __user *) data);
623 case PTRACE_SET_THREAD_AREA:
624 ret = ptrace_set_thread_area(child, addr,
625 (struct user_desc __user *) data);
629 ret = ptrace_request(child, request, addr, data);
636 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs, int error_code)
640 tsk->thread.trap_no = 1;
641 tsk->thread.error_code = error_code;
643 memset(&info, 0, sizeof(info));
644 info.si_signo = SIGTRAP;
645 info.si_code = TRAP_BRKPT;
648 info.si_addr = user_mode_vm(regs) ? (void __user *) regs->eip : NULL;
650 /* Send us the fakey SIGTRAP */
651 force_sig_info(SIGTRAP, &info, tsk);
654 /* notification of system call entry/exit
655 * - triggered by current->work.syscall_trace
657 __attribute__((regparm(3)))
658 int do_syscall_trace(struct pt_regs *regs, int entryexit)
660 int is_sysemu = test_thread_flag(TIF_SYSCALL_EMU);
662 * With TIF_SYSCALL_EMU set we want to ignore TIF_SINGLESTEP for syscall
665 int is_singlestep = !is_sysemu && test_thread_flag(TIF_SINGLESTEP);
668 /* do the secure computing check first */
670 secure_computing(regs->orig_eax);
672 if (unlikely(current->audit_context)) {
674 audit_syscall_exit(current, AUDITSC_RESULT(regs->eax),
676 /* Debug traps, when using PTRACE_SINGLESTEP, must be sent only
677 * on the syscall exit path. Normally, when TIF_SYSCALL_AUDIT is
678 * not used, entry.S will call us only on syscall exit, not
679 * entry; so when TIF_SYSCALL_AUDIT is used we must avoid
680 * calling send_sigtrap() on syscall entry.
682 * Note that when PTRACE_SYSEMU_SINGLESTEP is used,
683 * is_singlestep is false, despite his name, so we will still do
686 else if (is_singlestep)
690 if (!(current->ptrace & PT_PTRACED))
693 /* If a process stops on the 1st tracepoint with SYSCALL_TRACE
694 * and then is resumed with SYSEMU_SINGLESTEP, it will come in
695 * here. We have to check this and return */
696 if (is_sysemu && entryexit)
699 /* Fake a debug trap */
701 send_sigtrap(current, regs, 0);
703 if (!test_thread_flag(TIF_SYSCALL_TRACE) && !is_sysemu)
706 /* the 0x80 provides a way for the tracing parent to distinguish
707 between a syscall stop and SIGTRAP delivery */
708 /* Note that the debugger could change the result of test_thread_flag!*/
709 ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD) ? 0x80:0));
712 * this isn't the same as continuing with a signal, but it will do
713 * for normal use. strace only continues with a signal if the
714 * stopping signal is not SIGTRAP. -brl
716 if (current->exit_code) {
717 send_sig(current->exit_code, current, 1);
718 current->exit_code = 0;
722 if (unlikely(current->audit_context) && !entryexit)
723 audit_syscall_entry(current, AUDIT_ARCH_I386, regs->orig_eax,
724 regs->ebx, regs->ecx, regs->edx, regs->esi);
728 regs->orig_eax = -1; /* force skip of syscall restarting */
729 if (unlikely(current->audit_context))
730 audit_syscall_exit(current, AUDITSC_RESULT(regs->eax),