2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
5 * Pentium III FXSR, SSE support
6 * Gareth Hughes <gareth@valinux.com>, May 2000
10 * 'Traps.c' handles hardware traps and faults after we have saved some
13 #include <linux/moduleparam.h>
14 #include <linux/interrupt.h>
15 #include <linux/kallsyms.h>
16 #include <linux/spinlock.h>
17 #include <linux/kprobes.h>
18 #include <linux/uaccess.h>
19 #include <linux/utsname.h>
20 #include <linux/kdebug.h>
21 #include <linux/kernel.h>
22 #include <linux/module.h>
23 #include <linux/ptrace.h>
24 #include <linux/string.h>
25 #include <linux/unwind.h>
26 #include <linux/delay.h>
27 #include <linux/errno.h>
28 #include <linux/kexec.h>
29 #include <linux/sched.h>
30 #include <linux/timer.h>
31 #include <linux/init.h>
32 #include <linux/bug.h>
33 #include <linux/nmi.h>
35 #include <linux/smp.h>
38 #if defined(CONFIG_EDAC)
39 #include <linux/edac.h>
42 #include <asm/stacktrace.h>
43 #include <asm/processor.h>
44 #include <asm/debugreg.h>
45 #include <asm/atomic.h>
46 #include <asm/system.h>
47 #include <asm/unwind.h>
50 #include <asm/pgalloc.h>
51 #include <asm/proto.h>
53 #include <asm/traps.h>
55 #include <mach_traps.h>
57 int panic_on_unrecovered_nmi;
58 int kstack_depth_to_print = 12;
59 static unsigned int code_bytes = 64;
60 static int ignore_nmis;
61 static int die_counter;
63 static inline void conditional_sti(struct pt_regs *regs)
65 if (regs->flags & X86_EFLAGS_IF)
69 static inline void preempt_conditional_sti(struct pt_regs *regs)
72 if (regs->flags & X86_EFLAGS_IF)
76 static inline void preempt_conditional_cli(struct pt_regs *regs)
78 if (regs->flags & X86_EFLAGS_IF)
80 /* Make sure to not schedule here because we could be running
81 on an exception stack. */
85 void printk_address(unsigned long address, int reliable)
87 printk(" [<%016lx>] %s%pS\n", address, reliable ?
88 "" : "? ", (void *) address);
91 static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
92 unsigned *usedp, char **idp)
94 static char ids[][8] = {
95 [DEBUG_STACK - 1] = "#DB",
96 [NMI_STACK - 1] = "NMI",
97 [DOUBLEFAULT_STACK - 1] = "#DF",
98 [STACKFAULT_STACK - 1] = "#SS",
99 [MCE_STACK - 1] = "#MC",
100 #if DEBUG_STKSZ > EXCEPTION_STKSZ
101 [N_EXCEPTION_STACKS ... N_EXCEPTION_STACKS + DEBUG_STKSZ /
102 EXCEPTION_STKSZ - 2] = "#DB[?]"
108 * Iterate over all exception stacks, and figure out whether
109 * 'stack' is in one of them:
111 for (k = 0; k < N_EXCEPTION_STACKS; k++) {
112 unsigned long end = per_cpu(orig_ist, cpu).ist[k];
114 * Is 'stack' above this exception frame's end?
115 * If yes then skip to the next frame.
120 * Is 'stack' above this exception frame's start address?
121 * If yes then we found the right frame.
123 if (stack >= end - EXCEPTION_STKSZ) {
125 * Make sure we only iterate through an exception
126 * stack once. If it comes up for the second time
127 * then there's something wrong going on - just
128 * break out and return NULL:
130 if (*usedp & (1U << k))
134 return (unsigned long *)end;
137 * If this is a debug stack, and if it has a larger size than
138 * the usual exception stacks, then 'stack' might still
139 * be within the lower portion of the debug stack:
141 #if DEBUG_STKSZ > EXCEPTION_STKSZ
142 if (k == DEBUG_STACK - 1 && stack >= end - DEBUG_STKSZ) {
143 unsigned j = N_EXCEPTION_STACKS - 1;
146 * Black magic. A large debug stack is composed of
147 * multiple exception stack entries, which we
148 * iterate through now. Dont look:
152 end -= EXCEPTION_STKSZ;
153 ids[j][4] = '1' + (j - N_EXCEPTION_STACKS);
154 } while (stack < end - EXCEPTION_STKSZ);
155 if (*usedp & (1U << j))
159 return (unsigned long *)end;
167 * x86-64 can have up to three kernel stacks:
170 * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
173 static inline int valid_stack_ptr(struct thread_info *tinfo,
174 void *p, unsigned int size, void *end)
178 if (p < end && p >= (end-THREAD_SIZE))
183 return p > t && p < t + THREAD_SIZE - size;
186 /* The form of the top of the frame on the stack */
188 struct stack_frame *next_frame;
189 unsigned long return_address;
192 static inline unsigned long
193 print_context_stack(struct thread_info *tinfo,
194 unsigned long *stack, unsigned long bp,
195 const struct stacktrace_ops *ops, void *data,
198 struct stack_frame *frame = (struct stack_frame *)bp;
200 while (valid_stack_ptr(tinfo, stack, sizeof(*stack), end)) {
204 if (__kernel_text_address(addr)) {
205 if ((unsigned long) stack == bp + 8) {
206 ops->address(data, addr, 1);
207 frame = frame->next_frame;
208 bp = (unsigned long) frame;
210 ops->address(data, addr, bp == 0);
218 void dump_trace(struct task_struct *task, struct pt_regs *regs,
219 unsigned long *stack, unsigned long bp,
220 const struct stacktrace_ops *ops, void *data)
222 const unsigned cpu = get_cpu();
223 unsigned long *irqstack_end = (unsigned long *)cpu_pda(cpu)->irqstackptr;
225 struct thread_info *tinfo;
233 if (task && task != current)
234 stack = (unsigned long *)task->thread.sp;
237 #ifdef CONFIG_FRAME_POINTER
239 if (task == current) {
240 /* Grab bp right from our regs */
241 asm("movq %%rbp, %0" : "=r" (bp) : );
243 /* bp is the last reg pushed by switch_to */
244 bp = *(unsigned long *) task->thread.sp;
250 * Print function call entries in all stacks, starting at the
251 * current stack address. If the stacks consist of nested
254 tinfo = task_thread_info(task);
257 unsigned long *estack_end;
258 estack_end = in_exception_stack(cpu, (unsigned long)stack,
262 if (ops->stack(data, id) < 0)
265 bp = print_context_stack(tinfo, stack, bp, ops,
267 ops->stack(data, "<EOE>");
269 * We link to the next stack via the
270 * second-to-last pointer (index -2 to end) in the
273 stack = (unsigned long *) estack_end[-2];
277 unsigned long *irqstack;
278 irqstack = irqstack_end -
279 (IRQSTACKSIZE - 64) / sizeof(*irqstack);
281 if (stack >= irqstack && stack < irqstack_end) {
282 if (ops->stack(data, "IRQ") < 0)
284 bp = print_context_stack(tinfo, stack, bp,
285 ops, data, irqstack_end);
287 * We link to the next stack (which would be
288 * the process stack normally) the last
289 * pointer (index -1 to end) in the IRQ stack:
291 stack = (unsigned long *) (irqstack_end[-1]);
293 ops->stack(data, "EOI");
301 * This handles the process stack:
303 bp = print_context_stack(tinfo, stack, bp, ops, data, NULL);
306 EXPORT_SYMBOL(dump_trace);
309 print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
311 print_symbol(msg, symbol);
315 static void print_trace_warning(void *data, char *msg)
320 static int print_trace_stack(void *data, char *name)
322 printk(" <%s> ", name);
326 static void print_trace_address(void *data, unsigned long addr, int reliable)
328 touch_nmi_watchdog();
329 printk_address(addr, reliable);
332 static const struct stacktrace_ops print_trace_ops = {
333 .warning = print_trace_warning,
334 .warning_symbol = print_trace_warning_symbol,
335 .stack = print_trace_stack,
336 .address = print_trace_address,
340 show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
341 unsigned long *stack, unsigned long bp, char *log_lvl)
343 printk("\nCall Trace:\n");
344 dump_trace(task, regs, stack, bp, &print_trace_ops, log_lvl);
348 void show_trace(struct task_struct *task, struct pt_regs *regs,
349 unsigned long *stack, unsigned long bp)
351 show_trace_log_lvl(task, regs, stack, bp, "");
355 show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
356 unsigned long *sp, unsigned long bp, char *log_lvl)
358 unsigned long *stack;
360 const int cpu = smp_processor_id();
361 unsigned long *irqstack_end =
362 (unsigned long *) (cpu_pda(cpu)->irqstackptr);
363 unsigned long *irqstack =
364 (unsigned long *) (cpu_pda(cpu)->irqstackptr - IRQSTACKSIZE);
366 /* debugging aid: "show_stack(NULL, NULL);" prints the
367 back trace for this cpu. */
371 sp = (unsigned long *)task->thread.sp;
373 sp = (unsigned long *)&sp;
377 for (i = 0; i < kstack_depth_to_print; i++) {
378 if (stack >= irqstack && stack <= irqstack_end) {
379 if (stack == irqstack_end) {
380 stack = (unsigned long *) (irqstack_end[-1]);
384 if (((long) stack & (THREAD_SIZE-1)) == 0)
387 if (i && ((i % 4) == 0))
389 printk(" %016lx", *stack++);
390 touch_nmi_watchdog();
392 show_trace_log_lvl(task, regs, sp, bp, log_lvl);
395 void show_stack(struct task_struct *task, unsigned long *sp)
397 show_stack_log_lvl(task, NULL, sp, 0, "");
401 * The architecture-independent dump_stack generator
403 void dump_stack(void)
405 unsigned long bp = 0;
408 #ifdef CONFIG_FRAME_POINTER
410 asm("movq %%rbp, %0" : "=r" (bp) : );
413 printk("Pid: %d, comm: %.20s %s %s %.*s\n",
414 current->pid, current->comm, print_tainted(),
415 init_utsname()->release,
416 (int)strcspn(init_utsname()->version, " "),
417 init_utsname()->version);
418 show_trace(NULL, NULL, &stack, bp);
420 EXPORT_SYMBOL(dump_stack);
422 void show_registers(struct pt_regs *regs)
426 const int cpu = smp_processor_id();
427 struct task_struct *cur = cpu_pda(cpu)->pcurrent;
430 printk("CPU %d ", cpu);
432 printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
433 cur->comm, cur->pid, task_thread_info(cur), cur);
436 * When in-kernel, we also print out the stack and code at the
437 * time of the fault..
439 if (!user_mode(regs)) {
440 unsigned int code_prologue = code_bytes * 43 / 64;
441 unsigned int code_len = code_bytes;
446 show_stack_log_lvl(NULL, regs, (unsigned long *)sp,
450 printk(KERN_EMERG "Code: ");
452 ip = (u8 *)regs->ip - code_prologue;
453 if (ip < (u8 *)PAGE_OFFSET || probe_kernel_address(ip, c)) {
454 /* try starting at RIP */
456 code_len = code_len - code_prologue + 1;
458 for (i = 0; i < code_len; i++, ip++) {
459 if (ip < (u8 *)PAGE_OFFSET ||
460 probe_kernel_address(ip, c)) {
461 printk(" Bad RIP value.");
464 if (ip == (u8 *)regs->ip)
465 printk("<%02x> ", c);
473 int is_valid_bugaddr(unsigned long ip)
477 if (__copy_from_user(&ud2, (const void __user *) ip, sizeof(ud2)))
480 return ud2 == 0x0b0f;
483 static raw_spinlock_t die_lock = __RAW_SPIN_LOCK_UNLOCKED;
484 static int die_owner = -1;
485 static unsigned int die_nest_count;
487 unsigned __kprobes long oops_begin(void)
494 /* racy, but better than risking deadlock. */
495 raw_local_irq_save(flags);
496 cpu = smp_processor_id();
497 if (!__raw_spin_trylock(&die_lock)) {
498 if (cpu == die_owner)
499 /* nested oops. should stop eventually */;
501 __raw_spin_lock(&die_lock);
510 void __kprobes oops_end(unsigned long flags, struct pt_regs *regs, int signr)
516 /* Nest count reaches zero, release the lock. */
517 __raw_spin_unlock(&die_lock);
518 raw_local_irq_restore(flags);
524 panic("Fatal exception");
529 int __kprobes __die(const char *str, struct pt_regs *regs, long err)
531 printk(KERN_EMERG "%s: %04lx [%u] ", str, err & 0xffff, ++die_counter);
532 #ifdef CONFIG_PREEMPT
538 #ifdef CONFIG_DEBUG_PAGEALLOC
539 printk("DEBUG_PAGEALLOC");
542 if (notify_die(DIE_OOPS, str, regs, err,
543 current->thread.trap_no, SIGSEGV) == NOTIFY_STOP)
546 show_registers(regs);
547 add_taint(TAINT_DIE);
548 /* Executive summary in case the oops scrolled away */
549 printk(KERN_ALERT "RIP ");
550 printk_address(regs->ip, 1);
551 printk(" RSP <%016lx>\n", regs->sp);
552 if (kexec_should_crash(current))
557 void die(const char *str, struct pt_regs *regs, long err)
559 unsigned long flags = oops_begin();
561 if (!user_mode(regs))
562 report_bug(regs->ip, regs);
564 if (__die(str, regs, err))
566 oops_end(flags, regs, SIGSEGV);
569 notrace __kprobes void
570 die_nmi(char *str, struct pt_regs *regs, int do_panic)
574 if (notify_die(DIE_NMIWATCHDOG, str, regs, 0, 2, SIGINT) == NOTIFY_STOP)
577 flags = oops_begin();
579 * We are in trouble anyway, lets at least try
580 * to get a message out.
582 printk(KERN_EMERG "%s", str);
583 printk(" on CPU%d, ip %08lx, registers:\n",
584 smp_processor_id(), regs->ip);
585 show_registers(regs);
586 if (kexec_should_crash(current))
588 if (do_panic || panic_on_oops)
589 panic("Non maskable interrupt");
590 oops_end(flags, NULL, SIGBUS);
596 static void __kprobes
597 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
598 long error_code, siginfo_t *info)
600 struct task_struct *tsk = current;
602 if (!user_mode(regs))
606 * We want error_code and trap_no set for userspace faults and
607 * kernelspace faults which result in die(), but not
608 * kernelspace faults which are fixed up. die() gives the
609 * process no chance to handle the signal and notice the
610 * kernel fault information, so that won't result in polluting
611 * the information about previously queued, but not yet
612 * delivered, faults. See also do_general_protection below.
614 tsk->thread.error_code = error_code;
615 tsk->thread.trap_no = trapnr;
617 if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
618 printk_ratelimit()) {
620 "%s[%d] trap %s ip:%lx sp:%lx error:%lx",
621 tsk->comm, tsk->pid, str,
622 regs->ip, regs->sp, error_code);
623 print_vma_addr(" in ", regs->ip);
628 force_sig_info(signr, info, tsk);
630 force_sig(signr, tsk);
634 if (!fixup_exception(regs)) {
635 tsk->thread.error_code = error_code;
636 tsk->thread.trap_no = trapnr;
637 die(str, regs, error_code);
642 #define DO_ERROR(trapnr, signr, str, name) \
643 asmlinkage void do_##name(struct pt_regs *regs, long error_code) \
645 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
648 conditional_sti(regs); \
649 do_trap(trapnr, signr, str, regs, error_code, NULL); \
652 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
653 asmlinkage void do_##name(struct pt_regs *regs, long error_code) \
656 info.si_signo = signr; \
658 info.si_code = sicode; \
659 info.si_addr = (void __user *)siaddr; \
660 trace_hardirqs_fixup(); \
661 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
664 conditional_sti(regs); \
665 do_trap(trapnr, signr, str, regs, error_code, &info); \
668 DO_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
669 DO_ERROR(4, SIGSEGV, "overflow", overflow)
670 DO_ERROR(5, SIGSEGV, "bounds", bounds)
671 DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip)
672 DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
673 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
674 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
675 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
677 /* Runs on IST stack */
678 asmlinkage void do_stack_segment(struct pt_regs *regs, long error_code)
680 if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
681 12, SIGBUS) == NOTIFY_STOP)
683 preempt_conditional_sti(regs);
684 do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
685 preempt_conditional_cli(regs);
688 asmlinkage void do_double_fault(struct pt_regs *regs, long error_code)
690 static const char str[] = "double fault";
691 struct task_struct *tsk = current;
693 /* Return not checked because double check cannot be ignored */
694 notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
696 tsk->thread.error_code = error_code;
697 tsk->thread.trap_no = 8;
699 /* This is always a kernel trap and never fixable (and thus must
702 die(str, regs, error_code);
705 asmlinkage void __kprobes
706 do_general_protection(struct pt_regs *regs, long error_code)
708 struct task_struct *tsk;
710 conditional_sti(regs);
713 if (!user_mode(regs))
716 tsk->thread.error_code = error_code;
717 tsk->thread.trap_no = 13;
719 if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
720 printk_ratelimit()) {
722 "%s[%d] general protection ip:%lx sp:%lx error:%lx",
724 regs->ip, regs->sp, error_code);
725 print_vma_addr(" in ", regs->ip);
729 force_sig(SIGSEGV, tsk);
733 if (fixup_exception(regs))
736 tsk->thread.error_code = error_code;
737 tsk->thread.trap_no = 13;
738 if (notify_die(DIE_GPF, "general protection fault", regs,
739 error_code, 13, SIGSEGV) == NOTIFY_STOP)
741 die("general protection fault", regs, error_code);
744 static notrace __kprobes void
745 mem_parity_error(unsigned char reason, struct pt_regs *regs)
747 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x.\n",
749 printk(KERN_EMERG "You have some hardware problem, likely on the PCI bus.\n");
751 #if defined(CONFIG_EDAC)
752 if (edac_handler_set()) {
753 edac_atomic_assert_error();
758 if (panic_on_unrecovered_nmi)
759 panic("NMI: Not continuing");
761 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
763 /* Clear and disable the memory parity error line. */
764 reason = (reason & 0xf) | 4;
768 static notrace __kprobes void
769 io_check_error(unsigned char reason, struct pt_regs *regs)
771 printk("NMI: IOCK error (debug interrupt?)\n");
772 show_registers(regs);
774 /* Re-enable the IOCK line, wait for a few seconds */
775 reason = (reason & 0xf) | 8;
782 static notrace __kprobes void
783 unknown_nmi_error(unsigned char reason, struct pt_regs *regs)
785 if (notify_die(DIE_NMIUNKNOWN, "nmi", regs, reason, 2, SIGINT) ==
788 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x.\n",
790 printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
792 if (panic_on_unrecovered_nmi)
793 panic("NMI: Not continuing");
795 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
798 /* Runs on IST stack. This code must keep interrupts off all the time.
799 Nested NMIs are prevented by the CPU. */
800 asmlinkage notrace __kprobes void default_do_nmi(struct pt_regs *regs)
802 unsigned char reason = 0;
805 cpu = smp_processor_id();
807 /* Only the BSP gets external NMIs from the system. */
809 reason = get_nmi_reason();
811 if (!(reason & 0xc0)) {
812 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
816 * Ok, so this is none of the documented NMI sources,
817 * so it must be the NMI watchdog.
819 if (nmi_watchdog_tick(regs, reason))
821 if (!do_nmi_callback(regs, cpu))
822 unknown_nmi_error(reason, regs);
826 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
829 /* AK: following checks seem to be broken on modern chipsets. FIXME */
831 mem_parity_error(reason, regs);
833 io_check_error(reason, regs);
836 asmlinkage notrace __kprobes void
837 do_nmi(struct pt_regs *regs, long error_code)
841 add_pda(__nmi_count, 1);
844 default_do_nmi(regs);
855 void restart_nmi(void)
861 /* runs on IST stack. */
862 asmlinkage void __kprobes do_int3(struct pt_regs *regs, long error_code)
864 trace_hardirqs_fixup();
866 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
870 preempt_conditional_sti(regs);
871 do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
872 preempt_conditional_cli(regs);
875 /* Help handler running on IST stack to switch back to user stack
876 for scheduling or signal handling. The actual stack switch is done in
878 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
880 struct pt_regs *regs = eregs;
881 /* Did already sync */
882 if (eregs == (struct pt_regs *)eregs->sp)
884 /* Exception from user space */
885 else if (user_mode(eregs))
886 regs = task_pt_regs(current);
887 /* Exception from kernel and interrupts are enabled. Move to
888 kernel process stack. */
889 else if (eregs->flags & X86_EFLAGS_IF)
890 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
896 /* runs on IST stack. */
897 asmlinkage void __kprobes do_debug(struct pt_regs *regs,
898 unsigned long error_code)
900 struct task_struct *tsk = current;
901 unsigned long condition;
904 trace_hardirqs_fixup();
906 get_debugreg(condition, 6);
909 * The processor cleared BTF, so don't mark that we need it set.
911 clear_tsk_thread_flag(tsk, TIF_DEBUGCTLMSR);
912 tsk->thread.debugctlmsr = 0;
914 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
915 SIGTRAP) == NOTIFY_STOP)
918 preempt_conditional_sti(regs);
920 /* Mask out spurious debug traps due to lazy DR7 setting */
921 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
922 if (!tsk->thread.debugreg7)
926 tsk->thread.debugreg6 = condition;
929 * Single-stepping through TF: make sure we ignore any events in
930 * kernel space (but re-enable TF when returning to user mode).
932 if (condition & DR_STEP) {
933 if (!user_mode(regs))
934 goto clear_TF_reenable;
937 /* Ok, finally something we can handle */
938 tsk->thread.trap_no = 1;
939 tsk->thread.error_code = error_code;
940 info.si_signo = SIGTRAP;
942 info.si_code = TRAP_BRKPT;
943 info.si_addr = user_mode(regs) ? (void __user *)regs->ip : NULL;
944 force_sig_info(SIGTRAP, &info, tsk);
948 preempt_conditional_cli(regs);
952 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
953 regs->flags &= ~X86_EFLAGS_TF;
954 preempt_conditional_cli(regs);
958 static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
960 if (fixup_exception(regs))
963 notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE);
964 /* Illegal floating point operation in the kernel */
965 current->thread.trap_no = trapnr;
971 * Note that we play around with the 'TS' bit in an attempt to get
972 * the correct behaviour even in the presence of the asynchronous
975 asmlinkage void do_coprocessor_error(struct pt_regs *regs)
977 void __user *ip = (void __user *)(regs->ip);
978 struct task_struct *task;
980 unsigned short cwd, swd;
982 conditional_sti(regs);
983 if (!user_mode(regs) &&
984 kernel_math_error(regs, "kernel x87 math error", 16))
988 * Save the info for the exception handler and clear the error.
992 task->thread.trap_no = 16;
993 task->thread.error_code = 0;
994 info.si_signo = SIGFPE;
996 info.si_code = __SI_FAULT;
999 * (~cwd & swd) will mask out exceptions that are not set to unmasked
1000 * status. 0x3f is the exception bits in these regs, 0x200 is the
1001 * C1 reg you need in case of a stack fault, 0x040 is the stack
1002 * fault bit. We should only be taking one exception at a time,
1003 * so if this combination doesn't produce any single exception,
1004 * then we have a bad program that isn't synchronizing its FPU usage
1005 * and it will suffer the consequences since we won't be able to
1006 * fully reproduce the context of the exception
1008 cwd = get_fpu_cwd(task);
1009 swd = get_fpu_swd(task);
1010 switch (swd & ~cwd & 0x3f) {
1011 case 0x000: /* No unmasked exception */
1012 default: /* Multiple exceptions */
1014 case 0x001: /* Invalid Op */
1016 * swd & 0x240 == 0x040: Stack Underflow
1017 * swd & 0x240 == 0x240: Stack Overflow
1018 * User must clear the SF bit (0x40) if set
1020 info.si_code = FPE_FLTINV;
1022 case 0x002: /* Denormalize */
1023 case 0x010: /* Underflow */
1024 info.si_code = FPE_FLTUND;
1026 case 0x004: /* Zero Divide */
1027 info.si_code = FPE_FLTDIV;
1029 case 0x008: /* Overflow */
1030 info.si_code = FPE_FLTOVF;
1032 case 0x020: /* Precision */
1033 info.si_code = FPE_FLTRES;
1036 force_sig_info(SIGFPE, &info, task);
1039 asmlinkage void bad_intr(void)
1041 printk("bad interrupt");
1044 asmlinkage void do_simd_coprocessor_error(struct pt_regs *regs)
1046 void __user *ip = (void __user *)(regs->ip);
1047 struct task_struct *task;
1049 unsigned short mxcsr;
1051 conditional_sti(regs);
1052 if (!user_mode(regs) &&
1053 kernel_math_error(regs, "kernel simd math error", 19))
1057 * Save the info for the exception handler and clear the error.
1060 save_init_fpu(task);
1061 task->thread.trap_no = 19;
1062 task->thread.error_code = 0;
1063 info.si_signo = SIGFPE;
1065 info.si_code = __SI_FAULT;
1068 * The SIMD FPU exceptions are handled a little differently, as there
1069 * is only a single status/control register. Thus, to determine which
1070 * unmasked exception was caught we must mask the exception mask bits
1071 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1073 mxcsr = get_fpu_mxcsr(task);
1074 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
1078 case 0x001: /* Invalid Op */
1079 info.si_code = FPE_FLTINV;
1081 case 0x002: /* Denormalize */
1082 case 0x010: /* Underflow */
1083 info.si_code = FPE_FLTUND;
1085 case 0x004: /* Zero Divide */
1086 info.si_code = FPE_FLTDIV;
1088 case 0x008: /* Overflow */
1089 info.si_code = FPE_FLTOVF;
1091 case 0x020: /* Precision */
1092 info.si_code = FPE_FLTRES;
1095 force_sig_info(SIGFPE, &info, task);
1098 asmlinkage void do_spurious_interrupt_bug(struct pt_regs *regs)
1102 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
1106 asmlinkage void __attribute__((weak)) mce_threshold_interrupt(void)
1111 * 'math_state_restore()' saves the current math information in the
1112 * old math state array, and gets the new ones from the current task
1114 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1115 * Don't touch unless you *really* know how it works.
1117 asmlinkage void math_state_restore(void)
1119 struct task_struct *me = current;
1124 * does a slab alloc which can sleep
1128 * ran out of memory!
1130 do_group_exit(SIGKILL);
1133 local_irq_disable();
1136 clts(); /* Allow maths ops (or we recurse) */
1137 restore_fpu_checking(&me->thread.xstate->fxsave);
1138 task_thread_info(me)->status |= TS_USEDFPU;
1141 EXPORT_SYMBOL_GPL(math_state_restore);
1143 void __init trap_init(void)
1145 set_intr_gate(0, ÷_error);
1146 set_intr_gate_ist(1, &debug, DEBUG_STACK);
1147 set_intr_gate_ist(2, &nmi, NMI_STACK);
1148 /* int3 can be called from all */
1149 set_system_gate_ist(3, &int3, DEBUG_STACK);
1150 /* int4 can be called from all */
1151 set_system_gate(4, &overflow);
1152 set_intr_gate(5, &bounds);
1153 set_intr_gate(6, &invalid_op);
1154 set_intr_gate(7, &device_not_available);
1155 set_intr_gate_ist(8, &double_fault, DOUBLEFAULT_STACK);
1156 set_intr_gate(9, &coprocessor_segment_overrun);
1157 set_intr_gate(10, &invalid_TSS);
1158 set_intr_gate(11, &segment_not_present);
1159 set_intr_gate_ist(12, &stack_segment, STACKFAULT_STACK);
1160 set_intr_gate(13, &general_protection);
1161 set_intr_gate(14, &page_fault);
1162 set_intr_gate(15, &spurious_interrupt_bug);
1163 set_intr_gate(16, &coprocessor_error);
1164 set_intr_gate(17, &alignment_check);
1165 #ifdef CONFIG_X86_MCE
1166 set_intr_gate_ist(18, &machine_check, MCE_STACK);
1168 set_intr_gate(19, &simd_coprocessor_error);
1170 #ifdef CONFIG_IA32_EMULATION
1171 set_system_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
1174 * initialize the per thread extended state:
1176 init_thread_xstate();
1178 * Should be a barrier for any external CPU state:
1183 static int __init oops_setup(char *s)
1187 if (!strcmp(s, "panic"))
1191 early_param("oops", oops_setup);
1193 static int __init kstack_setup(char *s)
1197 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1200 early_param("kstack", kstack_setup);
1202 static int __init code_bytes_setup(char *s)
1204 code_bytes = simple_strtoul(s, NULL, 0);
1205 if (code_bytes > 8192)
1210 __setup("code_bytes=", code_bytes_setup);