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
40 /* set's the trap flag. */
41 #define TRAP_FLAG 0x100
44 * Offset of eflags on child stack..
46 #define EFL_OFFSET offsetof(struct pt_regs, eflags)
48 static inline struct pt_regs *get_child_regs(struct task_struct *task)
50 void *stack_top = (void *)task->thread.esp0;
51 return stack_top - sizeof(struct pt_regs);
55 * This routine will get a word off of the processes privileged stack.
56 * the offset is bytes into the pt_regs structure on the stack.
57 * This routine assumes that all the privileged stacks are in our
60 static inline int get_stack_long(struct task_struct *task, int offset)
64 stack = (unsigned char *)task->thread.esp0 - sizeof(struct pt_regs);
66 return (*((int *)stack));
70 * This routine will put a word on the processes privileged stack.
71 * the offset is bytes into the pt_regs structure on the stack.
72 * This routine assumes that all the privileged stacks are in our
75 static inline int put_stack_long(struct task_struct *task, int offset,
78 unsigned char * stack;
80 stack = (unsigned char *)task->thread.esp0 - sizeof(struct pt_regs);
82 *(unsigned long *) stack = data;
86 static int putreg(struct task_struct *child,
87 unsigned long regno, unsigned long value)
91 if (value && (value & 3) != 3)
93 child->thread.gs = value;
98 if (value && (value & 3) != 3)
104 if ((value & 3) != 3)
110 value |= get_stack_long(child, EFL_OFFSET) & ~FLAG_MASK;
115 put_stack_long(child, regno, value);
119 static unsigned long getreg(struct task_struct *child,
122 unsigned long retval = ~0UL;
124 switch (regno >> 2) {
126 retval = child->thread.gs;
138 retval &= get_stack_long(child, regno);
143 #define LDT_SEGMENT 4
145 static unsigned long convert_eip_to_linear(struct task_struct *child, struct pt_regs *regs)
147 unsigned long addr, seg;
150 seg = regs->xcs & 0xffff;
151 if (regs->eflags & VM_MASK) {
152 addr = (addr & 0xffff) + (seg << 4);
157 * We'll assume that the code segments in the GDT
158 * are all zero-based. That is largely true: the
159 * TLS segments are used for data, and the PNPBIOS
160 * and APM bios ones we just ignore here.
162 if (seg & LDT_SEGMENT) {
168 mutex_lock(&child->mm->context.lock);
169 if (unlikely((seg >> 3) >= child->mm->context.size))
170 addr = -1L; /* bogus selector, access would fault */
172 desc = child->mm->context.ldt + seg;
173 base = ((desc[0] >> 16) |
174 ((desc[1] & 0xff) << 16) |
175 (desc[1] & 0xff000000));
177 /* 16-bit code segment? */
178 if (!((desc[1] >> 22) & 1))
182 mutex_unlock(&child->mm->context.lock);
187 static inline int is_setting_trap_flag(struct task_struct *child, struct pt_regs *regs)
190 unsigned char opcode[15];
191 unsigned long addr = convert_eip_to_linear(child, regs);
193 copied = access_process_vm(child, addr, opcode, sizeof(opcode), 0);
194 for (i = 0; i < copied; i++) {
197 case 0x9d: case 0xcf:
199 /* opcode and address size prefixes */
200 case 0x66: case 0x67:
202 /* irrelevant prefixes (segment overrides and repeats) */
203 case 0x26: case 0x2e:
204 case 0x36: case 0x3e:
205 case 0x64: case 0x65:
206 case 0xf0: case 0xf2: case 0xf3:
210 * pushf: NOTE! We should probably not let
211 * the user see the TF bit being set. But
212 * it's more pain than it's worth to avoid
213 * it, and a debugger could emulate this
214 * all in user space if it _really_ cares.
224 static void set_singlestep(struct task_struct *child)
226 struct pt_regs *regs = get_child_regs(child);
229 * Always set TIF_SINGLESTEP - this guarantees that
230 * we single-step system calls etc.. This will also
231 * cause us to set TF when returning to user mode.
233 set_tsk_thread_flag(child, TIF_SINGLESTEP);
236 * If TF was already set, don't do anything else
238 if (regs->eflags & TRAP_FLAG)
241 /* Set TF on the kernel stack.. */
242 regs->eflags |= TRAP_FLAG;
245 * ..but if TF is changed by the instruction we will trace,
246 * don't mark it as being "us" that set it, so that we
247 * won't clear it by hand later.
249 if (is_setting_trap_flag(child, regs))
252 child->ptrace |= PT_DTRACE;
255 static void clear_singlestep(struct task_struct *child)
257 /* Always clear TIF_SINGLESTEP... */
258 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
260 /* But touch TF only if it was set by us.. */
261 if (child->ptrace & PT_DTRACE) {
262 struct pt_regs *regs = get_child_regs(child);
263 regs->eflags &= ~TRAP_FLAG;
264 child->ptrace &= ~PT_DTRACE;
269 * Called by kernel/ptrace.c when detaching..
271 * Make sure the single step bit is not set.
273 void ptrace_disable(struct task_struct *child)
275 clear_singlestep(child);
276 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
280 * Perform get_thread_area on behalf of the traced child.
283 ptrace_get_thread_area(struct task_struct *child,
284 int idx, struct user_desc __user *user_desc)
286 struct user_desc info;
287 struct desc_struct *desc;
290 * Get the current Thread-Local Storage area:
293 #define GET_BASE(desc) ( \
294 (((desc)->a >> 16) & 0x0000ffff) | \
295 (((desc)->b << 16) & 0x00ff0000) | \
296 ( (desc)->b & 0xff000000) )
298 #define GET_LIMIT(desc) ( \
299 ((desc)->a & 0x0ffff) | \
300 ((desc)->b & 0xf0000) )
302 #define GET_32BIT(desc) (((desc)->b >> 22) & 1)
303 #define GET_CONTENTS(desc) (((desc)->b >> 10) & 3)
304 #define GET_WRITABLE(desc) (((desc)->b >> 9) & 1)
305 #define GET_LIMIT_PAGES(desc) (((desc)->b >> 23) & 1)
306 #define GET_PRESENT(desc) (((desc)->b >> 15) & 1)
307 #define GET_USEABLE(desc) (((desc)->b >> 20) & 1)
309 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
312 desc = child->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
314 info.entry_number = idx;
315 info.base_addr = GET_BASE(desc);
316 info.limit = GET_LIMIT(desc);
317 info.seg_32bit = GET_32BIT(desc);
318 info.contents = GET_CONTENTS(desc);
319 info.read_exec_only = !GET_WRITABLE(desc);
320 info.limit_in_pages = GET_LIMIT_PAGES(desc);
321 info.seg_not_present = !GET_PRESENT(desc);
322 info.useable = GET_USEABLE(desc);
324 if (copy_to_user(user_desc, &info, sizeof(info)))
331 * Perform set_thread_area on behalf of the traced child.
334 ptrace_set_thread_area(struct task_struct *child,
335 int idx, struct user_desc __user *user_desc)
337 struct user_desc info;
338 struct desc_struct *desc;
340 if (copy_from_user(&info, user_desc, sizeof(info)))
343 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
346 desc = child->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
347 if (LDT_empty(&info)) {
351 desc->a = LDT_entry_a(&info);
352 desc->b = LDT_entry_b(&info);
358 long arch_ptrace(struct task_struct *child, long request, long addr, long data)
360 struct user * dummy = NULL;
362 unsigned long __user *datap = (unsigned long __user *)data;
365 /* when I and D space are separate, these will need to be fixed. */
366 case PTRACE_PEEKTEXT: /* read word at location addr. */
367 case PTRACE_PEEKDATA:
368 ret = generic_ptrace_peekdata(child, addr, data);
371 /* read the word at location addr in the USER area. */
372 case PTRACE_PEEKUSR: {
376 if ((addr & 3) || addr < 0 ||
377 addr > sizeof(struct user) - 3)
380 tmp = 0; /* Default return condition */
381 if(addr < FRAME_SIZE*sizeof(long))
382 tmp = getreg(child, addr);
383 if(addr >= (long) &dummy->u_debugreg[0] &&
384 addr <= (long) &dummy->u_debugreg[7]){
385 addr -= (long) &dummy->u_debugreg[0];
387 tmp = child->thread.debugreg[addr];
389 ret = put_user(tmp, datap);
393 /* when I and D space are separate, this will have to be fixed. */
394 case PTRACE_POKETEXT: /* write the word at location addr. */
395 case PTRACE_POKEDATA:
396 ret = generic_ptrace_pokedata(child, addr, data);
399 case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
401 if ((addr & 3) || addr < 0 ||
402 addr > sizeof(struct user) - 3)
405 if (addr < FRAME_SIZE*sizeof(long)) {
406 ret = putreg(child, addr, data);
409 /* We need to be very careful here. We implicitly
410 want to modify a portion of the task_struct, and we
411 have to be selective about what portions we allow someone
415 if(addr >= (long) &dummy->u_debugreg[0] &&
416 addr <= (long) &dummy->u_debugreg[7]){
418 if(addr == (long) &dummy->u_debugreg[4]) break;
419 if(addr == (long) &dummy->u_debugreg[5]) break;
420 if(addr < (long) &dummy->u_debugreg[4] &&
421 ((unsigned long) data) >= TASK_SIZE-3) break;
423 /* Sanity-check data. Take one half-byte at once with
424 * check = (val >> (16 + 4*i)) & 0xf. It contains the
425 * R/Wi and LENi bits; bits 0 and 1 are R/Wi, and bits
426 * 2 and 3 are LENi. Given a list of invalid values,
427 * we do mask |= 1 << invalid_value, so that
428 * (mask >> check) & 1 is a correct test for invalid
431 * R/Wi contains the type of the breakpoint /
432 * watchpoint, LENi contains the length of the watched
433 * data in the watchpoint case.
435 * The invalid values are:
436 * - LENi == 0x10 (undefined), so mask |= 0x0f00.
437 * - R/Wi == 0x10 (break on I/O reads or writes), so
439 * - R/Wi == 0x00 && LENi != 0x00, so we have mask |=
442 * Finally, mask = 0x0f00 | 0x4444 | 0x1110 == 0x5f54.
444 * See the Intel Manual "System Programming Guide",
447 * Note that LENi == 0x10 is defined on x86_64 in long
448 * mode (i.e. even for 32-bit userspace software, but
449 * 64-bit kernel), so the x86_64 mask value is 0x5454.
450 * See the AMD manual no. 24593 (AMD64 System
453 if(addr == (long) &dummy->u_debugreg[7]) {
454 data &= ~DR_CONTROL_RESERVED;
456 if ((0x5f54 >> ((data >> (16 + 4*i)) & 0xf)) & 1)
459 set_tsk_thread_flag(child, TIF_DEBUG);
461 clear_tsk_thread_flag(child, TIF_DEBUG);
463 addr -= (long) &dummy->u_debugreg;
465 child->thread.debugreg[addr] = data;
470 case PTRACE_SYSEMU: /* continue and stop at next syscall, which will not be executed */
471 case PTRACE_SYSCALL: /* continue and stop at next (return from) syscall */
472 case PTRACE_CONT: /* restart after signal. */
474 if (!valid_signal(data))
476 if (request == PTRACE_SYSEMU) {
477 set_tsk_thread_flag(child, TIF_SYSCALL_EMU);
478 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
479 } else if (request == PTRACE_SYSCALL) {
480 set_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
481 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
483 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
484 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
486 child->exit_code = data;
487 /* make sure the single step bit is not set. */
488 clear_singlestep(child);
489 wake_up_process(child);
494 * make the child exit. Best I can do is send it a sigkill.
495 * perhaps it should be put in the status that it wants to
500 if (child->exit_state == EXIT_ZOMBIE) /* already dead */
502 child->exit_code = SIGKILL;
503 /* make sure the single step bit is not set. */
504 clear_singlestep(child);
505 wake_up_process(child);
508 case PTRACE_SYSEMU_SINGLESTEP: /* Same as SYSEMU, but singlestep if not syscall */
509 case PTRACE_SINGLESTEP: /* set the trap flag. */
511 if (!valid_signal(data))
514 if (request == PTRACE_SYSEMU_SINGLESTEP)
515 set_tsk_thread_flag(child, TIF_SYSCALL_EMU);
517 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
519 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
520 set_singlestep(child);
521 child->exit_code = data;
522 /* give it a chance to run. */
523 wake_up_process(child);
527 case PTRACE_GETREGS: { /* Get all gp regs from the child. */
528 if (!access_ok(VERIFY_WRITE, datap, FRAME_SIZE*sizeof(long))) {
532 for ( i = 0; i < FRAME_SIZE*sizeof(long); i += sizeof(long) ) {
533 __put_user(getreg(child, i), datap);
540 case PTRACE_SETREGS: { /* Set all gp regs in the child. */
542 if (!access_ok(VERIFY_READ, datap, FRAME_SIZE*sizeof(long))) {
546 for ( i = 0; i < FRAME_SIZE*sizeof(long); i += sizeof(long) ) {
547 __get_user(tmp, datap);
548 putreg(child, i, tmp);
555 case PTRACE_GETFPREGS: { /* Get the child FPU state. */
556 if (!access_ok(VERIFY_WRITE, datap,
557 sizeof(struct user_i387_struct))) {
562 if (!tsk_used_math(child))
564 get_fpregs((struct user_i387_struct __user *)data, child);
568 case PTRACE_SETFPREGS: { /* Set the child FPU state. */
569 if (!access_ok(VERIFY_READ, datap,
570 sizeof(struct user_i387_struct))) {
574 set_stopped_child_used_math(child);
575 set_fpregs(child, (struct user_i387_struct __user *)data);
580 case PTRACE_GETFPXREGS: { /* Get the child extended FPU state. */
581 if (!access_ok(VERIFY_WRITE, datap,
582 sizeof(struct user_fxsr_struct))) {
586 if (!tsk_used_math(child))
588 ret = get_fpxregs((struct user_fxsr_struct __user *)data, child);
592 case PTRACE_SETFPXREGS: { /* Set the child extended FPU state. */
593 if (!access_ok(VERIFY_READ, datap,
594 sizeof(struct user_fxsr_struct))) {
598 set_stopped_child_used_math(child);
599 ret = set_fpxregs(child, (struct user_fxsr_struct __user *)data);
603 case PTRACE_GET_THREAD_AREA:
604 ret = ptrace_get_thread_area(child, addr,
605 (struct user_desc __user *) data);
608 case PTRACE_SET_THREAD_AREA:
609 ret = ptrace_set_thread_area(child, addr,
610 (struct user_desc __user *) data);
614 ret = ptrace_request(child, request, addr, data);
621 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs, int error_code)
625 tsk->thread.trap_no = 1;
626 tsk->thread.error_code = error_code;
628 memset(&info, 0, sizeof(info));
629 info.si_signo = SIGTRAP;
630 info.si_code = TRAP_BRKPT;
633 info.si_addr = user_mode_vm(regs) ? (void __user *) regs->eip : NULL;
635 /* Send us the fake SIGTRAP */
636 force_sig_info(SIGTRAP, &info, tsk);
639 /* notification of system call entry/exit
640 * - triggered by current->work.syscall_trace
642 __attribute__((regparm(3)))
643 int do_syscall_trace(struct pt_regs *regs, int entryexit)
645 int is_sysemu = test_thread_flag(TIF_SYSCALL_EMU);
647 * With TIF_SYSCALL_EMU set we want to ignore TIF_SINGLESTEP for syscall
650 int is_singlestep = !is_sysemu && test_thread_flag(TIF_SINGLESTEP);
653 /* do the secure computing check first */
655 secure_computing(regs->orig_eax);
657 if (unlikely(current->audit_context)) {
659 audit_syscall_exit(AUDITSC_RESULT(regs->eax),
661 /* Debug traps, when using PTRACE_SINGLESTEP, must be sent only
662 * on the syscall exit path. Normally, when TIF_SYSCALL_AUDIT is
663 * not used, entry.S will call us only on syscall exit, not
664 * entry; so when TIF_SYSCALL_AUDIT is used we must avoid
665 * calling send_sigtrap() on syscall entry.
667 * Note that when PTRACE_SYSEMU_SINGLESTEP is used,
668 * is_singlestep is false, despite his name, so we will still do
671 else if (is_singlestep)
675 if (!(current->ptrace & PT_PTRACED))
678 /* If a process stops on the 1st tracepoint with SYSCALL_TRACE
679 * and then is resumed with SYSEMU_SINGLESTEP, it will come in
680 * here. We have to check this and return */
681 if (is_sysemu && entryexit)
684 /* Fake a debug trap */
686 send_sigtrap(current, regs, 0);
688 if (!test_thread_flag(TIF_SYSCALL_TRACE) && !is_sysemu)
691 /* the 0x80 provides a way for the tracing parent to distinguish
692 between a syscall stop and SIGTRAP delivery */
693 /* Note that the debugger could change the result of test_thread_flag!*/
694 ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD) ? 0x80:0));
697 * this isn't the same as continuing with a signal, but it will do
698 * for normal use. strace only continues with a signal if the
699 * stopping signal is not SIGTRAP. -brl
701 if (current->exit_code) {
702 send_sig(current->exit_code, current, 1);
703 current->exit_code = 0;
707 if (unlikely(current->audit_context) && !entryexit)
708 audit_syscall_entry(AUDIT_ARCH_I386, regs->orig_eax,
709 regs->ebx, regs->ecx, regs->edx, regs->esi);
713 regs->orig_eax = -1; /* force skip of syscall restarting */
714 if (unlikely(current->audit_context))
715 audit_syscall_exit(AUDITSC_RESULT(regs->eax), regs->eax);