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>
36 #if defined(CONFIG_EDAC)
37 #include <linux/edac.h>
40 #include <asm/stacktrace.h>
41 #include <asm/processor.h>
42 #include <asm/debugreg.h>
43 #include <asm/atomic.h>
44 #include <asm/system.h>
45 #include <asm/unwind.h>
51 #include <asm/pgalloc.h>
52 #include <asm/proto.h>
55 #include <mach_traps.h>
57 asmlinkage void divide_error(void);
58 asmlinkage void debug(void);
59 asmlinkage void nmi(void);
60 asmlinkage void int3(void);
61 asmlinkage void overflow(void);
62 asmlinkage void bounds(void);
63 asmlinkage void invalid_op(void);
64 asmlinkage void device_not_available(void);
65 asmlinkage void double_fault(void);
66 asmlinkage void coprocessor_segment_overrun(void);
67 asmlinkage void invalid_TSS(void);
68 asmlinkage void segment_not_present(void);
69 asmlinkage void stack_segment(void);
70 asmlinkage void general_protection(void);
71 asmlinkage void page_fault(void);
72 asmlinkage void coprocessor_error(void);
73 asmlinkage void simd_coprocessor_error(void);
74 asmlinkage void alignment_check(void);
75 asmlinkage void spurious_interrupt_bug(void);
76 asmlinkage void machine_check(void);
78 int panic_on_unrecovered_nmi;
79 int kstack_depth_to_print = 12;
80 static unsigned int code_bytes = 64;
81 static int ignore_nmis;
82 static int die_counter;
84 static inline void conditional_sti(struct pt_regs *regs)
86 if (regs->flags & X86_EFLAGS_IF)
90 static inline void preempt_conditional_sti(struct pt_regs *regs)
93 if (regs->flags & X86_EFLAGS_IF)
97 static inline void preempt_conditional_cli(struct pt_regs *regs)
99 if (regs->flags & X86_EFLAGS_IF)
101 /* Make sure to not schedule here because we could be running
102 on an exception stack. */
106 void printk_address(unsigned long address, int reliable)
108 printk(" [<%016lx>] %s%pS\n", address, reliable ? "": "? ", (void *) address);
111 static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
112 unsigned *usedp, char **idp)
114 static char ids[][8] = {
115 [DEBUG_STACK - 1] = "#DB",
116 [NMI_STACK - 1] = "NMI",
117 [DOUBLEFAULT_STACK - 1] = "#DF",
118 [STACKFAULT_STACK - 1] = "#SS",
119 [MCE_STACK - 1] = "#MC",
120 #if DEBUG_STKSZ > EXCEPTION_STKSZ
121 [N_EXCEPTION_STACKS ... N_EXCEPTION_STACKS + DEBUG_STKSZ / EXCEPTION_STKSZ - 2] = "#DB[?]"
127 * Iterate over all exception stacks, and figure out whether
128 * 'stack' is in one of them:
130 for (k = 0; k < N_EXCEPTION_STACKS; k++) {
131 unsigned long end = per_cpu(orig_ist, cpu).ist[k];
133 * Is 'stack' above this exception frame's end?
134 * If yes then skip to the next frame.
139 * Is 'stack' above this exception frame's start address?
140 * If yes then we found the right frame.
142 if (stack >= end - EXCEPTION_STKSZ) {
144 * Make sure we only iterate through an exception
145 * stack once. If it comes up for the second time
146 * then there's something wrong going on - just
147 * break out and return NULL:
149 if (*usedp & (1U << k))
153 return (unsigned long *)end;
156 * If this is a debug stack, and if it has a larger size than
157 * the usual exception stacks, then 'stack' might still
158 * be within the lower portion of the debug stack:
160 #if DEBUG_STKSZ > EXCEPTION_STKSZ
161 if (k == DEBUG_STACK - 1 && stack >= end - DEBUG_STKSZ) {
162 unsigned j = N_EXCEPTION_STACKS - 1;
165 * Black magic. A large debug stack is composed of
166 * multiple exception stack entries, which we
167 * iterate through now. Dont look:
171 end -= EXCEPTION_STKSZ;
172 ids[j][4] = '1' + (j - N_EXCEPTION_STACKS);
173 } while (stack < end - EXCEPTION_STKSZ);
174 if (*usedp & (1U << j))
178 return (unsigned long *)end;
186 * x86-64 can have up to three kernel stacks:
189 * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
192 static inline int valid_stack_ptr(struct thread_info *tinfo,
193 void *p, unsigned int size, void *end)
197 if (p < end && p >= (end-THREAD_SIZE))
202 return p > t && p < t + THREAD_SIZE - size;
205 /* The form of the top of the frame on the stack */
207 struct stack_frame *next_frame;
208 unsigned long return_address;
211 static inline unsigned long
212 print_context_stack(struct thread_info *tinfo,
213 unsigned long *stack, unsigned long bp,
214 const struct stacktrace_ops *ops, void *data,
217 struct stack_frame *frame = (struct stack_frame *)bp;
219 while (valid_stack_ptr(tinfo, stack, sizeof(*stack), end)) {
223 if (__kernel_text_address(addr)) {
224 if ((unsigned long) stack == bp + 8) {
225 ops->address(data, addr, 1);
226 frame = frame->next_frame;
227 bp = (unsigned long) frame;
229 ops->address(data, addr, bp == 0);
237 void dump_trace(struct task_struct *task, struct pt_regs *regs,
238 unsigned long *stack, unsigned long bp,
239 const struct stacktrace_ops *ops, void *data)
241 const unsigned cpu = get_cpu();
242 unsigned long *irqstack_end = (unsigned long*)cpu_pda(cpu)->irqstackptr;
244 struct thread_info *tinfo;
252 if (task && task != current)
253 stack = (unsigned long *)task->thread.sp;
256 #ifdef CONFIG_FRAME_POINTER
258 if (task == current) {
259 /* Grab bp right from our regs */
260 asm("movq %%rbp, %0" : "=r" (bp) :);
262 /* bp is the last reg pushed by switch_to */
263 bp = *(unsigned long *) task->thread.sp;
269 * Print function call entries in all stacks, starting at the
270 * current stack address. If the stacks consist of nested
273 tinfo = task_thread_info(task);
276 unsigned long *estack_end;
277 estack_end = in_exception_stack(cpu, (unsigned long)stack,
281 if (ops->stack(data, id) < 0)
284 bp = print_context_stack(tinfo, stack, bp, ops,
286 ops->stack(data, "<EOE>");
288 * We link to the next stack via the
289 * second-to-last pointer (index -2 to end) in the
292 stack = (unsigned long *) estack_end[-2];
296 unsigned long *irqstack;
297 irqstack = irqstack_end -
298 (IRQSTACKSIZE - 64) / sizeof(*irqstack);
300 if (stack >= irqstack && stack < irqstack_end) {
301 if (ops->stack(data, "IRQ") < 0)
303 bp = print_context_stack(tinfo, stack, bp,
304 ops, data, irqstack_end);
306 * We link to the next stack (which would be
307 * the process stack normally) the last
308 * pointer (index -1 to end) in the IRQ stack:
310 stack = (unsigned long *) (irqstack_end[-1]);
312 ops->stack(data, "EOI");
320 * This handles the process stack:
322 bp = print_context_stack(tinfo, stack, bp, ops, data, NULL);
325 EXPORT_SYMBOL(dump_trace);
328 print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
330 print_symbol(msg, symbol);
334 static void print_trace_warning(void *data, char *msg)
339 static int print_trace_stack(void *data, char *name)
341 printk(" <%s> ", name);
345 static void print_trace_address(void *data, unsigned long addr, int reliable)
347 touch_nmi_watchdog();
348 printk_address(addr, reliable);
351 static const struct stacktrace_ops print_trace_ops = {
352 .warning = print_trace_warning,
353 .warning_symbol = print_trace_warning_symbol,
354 .stack = print_trace_stack,
355 .address = print_trace_address,
358 void show_trace(struct task_struct *task, struct pt_regs *regs,
359 unsigned long *stack, unsigned long bp)
361 printk("\nCall Trace:\n");
362 dump_trace(task, regs, stack, bp, &print_trace_ops, NULL);
367 _show_stack(struct task_struct *task, struct pt_regs *regs,
368 unsigned long *sp, unsigned long bp)
370 unsigned long *stack;
372 const int cpu = smp_processor_id();
373 unsigned long *irqstack_end = (unsigned long *) (cpu_pda(cpu)->irqstackptr);
374 unsigned long *irqstack = (unsigned long *) (cpu_pda(cpu)->irqstackptr - IRQSTACKSIZE);
376 // debugging aid: "show_stack(NULL, NULL);" prints the
377 // back trace for this cpu.
381 sp = (unsigned long *)task->thread.sp;
383 sp = (unsigned long *)&sp;
387 for (i = 0; i < kstack_depth_to_print; i++) {
388 if (stack >= irqstack && stack <= irqstack_end) {
389 if (stack == irqstack_end) {
390 stack = (unsigned long *) (irqstack_end[-1]);
394 if (((long) stack & (THREAD_SIZE-1)) == 0)
397 if (i && ((i % 4) == 0))
399 printk(" %016lx", *stack++);
400 touch_nmi_watchdog();
402 show_trace(task, regs, sp, bp);
405 void show_stack(struct task_struct *task, unsigned long *sp)
407 _show_stack(task, NULL, sp, 0);
411 * The architecture-independent dump_stack generator
413 void dump_stack(void)
415 unsigned long bp = 0;
418 #ifdef CONFIG_FRAME_POINTER
420 asm("movq %%rbp, %0" : "=r" (bp):);
423 printk("Pid: %d, comm: %.20s %s %s %.*s\n",
424 current->pid, current->comm, print_tainted(),
425 init_utsname()->release,
426 (int)strcspn(init_utsname()->version, " "),
427 init_utsname()->version);
428 show_trace(NULL, NULL, &stack, bp);
431 EXPORT_SYMBOL(dump_stack);
433 void show_registers(struct pt_regs *regs)
437 const int cpu = smp_processor_id();
438 struct task_struct *cur = cpu_pda(cpu)->pcurrent;
441 printk("CPU %d ", cpu);
443 printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
444 cur->comm, cur->pid, task_thread_info(cur), cur);
447 * When in-kernel, we also print out the stack and code at the
448 * time of the fault..
450 if (!user_mode(regs)) {
451 unsigned int code_prologue = code_bytes * 43 / 64;
452 unsigned int code_len = code_bytes;
457 _show_stack(NULL, regs, (unsigned long *)sp, regs->bp);
460 printk(KERN_EMERG "Code: ");
462 ip = (u8 *)regs->ip - code_prologue;
463 if (ip < (u8 *)PAGE_OFFSET || probe_kernel_address(ip, c)) {
464 /* try starting at RIP */
466 code_len = code_len - code_prologue + 1;
468 for (i = 0; i < code_len; i++, ip++) {
469 if (ip < (u8 *)PAGE_OFFSET ||
470 probe_kernel_address(ip, c)) {
471 printk(" Bad RIP value.");
474 if (ip == (u8 *)regs->ip)
475 printk("<%02x> ", c);
483 int is_valid_bugaddr(unsigned long ip)
487 if (__copy_from_user(&ud2, (const void __user *) ip, sizeof(ud2)))
490 return ud2 == 0x0b0f;
493 static raw_spinlock_t die_lock = __RAW_SPIN_LOCK_UNLOCKED;
494 static int die_owner = -1;
495 static unsigned int die_nest_count;
497 unsigned __kprobes long oops_begin(void)
504 /* racy, but better than risking deadlock. */
505 raw_local_irq_save(flags);
506 cpu = smp_processor_id();
507 if (!__raw_spin_trylock(&die_lock)) {
508 if (cpu == die_owner)
509 /* nested oops. should stop eventually */;
511 __raw_spin_lock(&die_lock);
520 void __kprobes oops_end(unsigned long flags, struct pt_regs *regs, int signr)
526 /* Nest count reaches zero, release the lock. */
527 __raw_spin_unlock(&die_lock);
528 raw_local_irq_restore(flags);
534 panic("Fatal exception");
539 int __kprobes __die(const char *str, struct pt_regs *regs, long err)
541 printk(KERN_EMERG "%s: %04lx [%u] ", str, err & 0xffff, ++die_counter);
542 #ifdef CONFIG_PREEMPT
548 #ifdef CONFIG_DEBUG_PAGEALLOC
549 printk("DEBUG_PAGEALLOC");
552 if (notify_die(DIE_OOPS, str, regs, err,
553 current->thread.trap_no, SIGSEGV) == NOTIFY_STOP)
556 show_registers(regs);
557 add_taint(TAINT_DIE);
558 /* Executive summary in case the oops scrolled away */
559 printk(KERN_ALERT "RIP ");
560 printk_address(regs->ip, 1);
561 printk(" RSP <%016lx>\n", regs->sp);
562 if (kexec_should_crash(current))
567 void die(const char *str, struct pt_regs *regs, long err)
569 unsigned long flags = oops_begin();
571 if (!user_mode(regs))
572 report_bug(regs->ip, regs);
574 if (__die(str, regs, err))
576 oops_end(flags, regs, SIGSEGV);
579 notrace __kprobes void
580 die_nmi(char *str, struct pt_regs *regs, int do_panic)
584 if (notify_die(DIE_NMIWATCHDOG, str, regs, 0, 2, SIGINT) == NOTIFY_STOP)
587 flags = oops_begin();
589 * We are in trouble anyway, lets at least try
590 * to get a message out.
592 printk(KERN_EMERG "%s", str);
593 printk(" on CPU%d, ip %08lx, registers:\n",
594 smp_processor_id(), regs->ip);
595 show_registers(regs);
596 if (kexec_should_crash(current))
598 if (do_panic || panic_on_oops)
599 panic("Non maskable interrupt");
600 oops_end(flags, NULL, SIGBUS);
606 static void __kprobes
607 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
608 long error_code, siginfo_t *info)
610 struct task_struct *tsk = current;
612 if (!user_mode(regs))
616 * We want error_code and trap_no set for userspace faults and
617 * kernelspace faults which result in die(), but not
618 * kernelspace faults which are fixed up. die() gives the
619 * process no chance to handle the signal and notice the
620 * kernel fault information, so that won't result in polluting
621 * the information about previously queued, but not yet
622 * delivered, faults. See also do_general_protection below.
624 tsk->thread.error_code = error_code;
625 tsk->thread.trap_no = trapnr;
627 if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
628 printk_ratelimit()) {
630 "%s[%d] trap %s ip:%lx sp:%lx error:%lx",
631 tsk->comm, tsk->pid, str,
632 regs->ip, regs->sp, error_code);
633 print_vma_addr(" in ", regs->ip);
638 force_sig_info(signr, info, tsk);
640 force_sig(signr, tsk);
644 if (!fixup_exception(regs)) {
645 tsk->thread.error_code = error_code;
646 tsk->thread.trap_no = trapnr;
647 die(str, regs, error_code);
652 #define DO_ERROR(trapnr, signr, str, name) \
653 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
655 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
658 conditional_sti(regs); \
659 do_trap(trapnr, signr, str, regs, error_code, NULL); \
662 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
663 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
666 info.si_signo = signr; \
668 info.si_code = sicode; \
669 info.si_addr = (void __user *)siaddr; \
670 trace_hardirqs_fixup(); \
671 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
674 conditional_sti(regs); \
675 do_trap(trapnr, signr, str, regs, error_code, &info); \
678 DO_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
679 DO_ERROR(4, SIGSEGV, "overflow", overflow)
680 DO_ERROR(5, SIGSEGV, "bounds", bounds)
681 DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip)
682 DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
683 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
684 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
685 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
687 /* Runs on IST stack */
688 asmlinkage void do_stack_segment(struct pt_regs *regs, long error_code)
690 if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
691 12, SIGBUS) == NOTIFY_STOP)
693 preempt_conditional_sti(regs);
694 do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
695 preempt_conditional_cli(regs);
698 asmlinkage void do_double_fault(struct pt_regs * regs, long error_code)
700 static const char str[] = "double fault";
701 struct task_struct *tsk = current;
703 /* Return not checked because double check cannot be ignored */
704 notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
706 tsk->thread.error_code = error_code;
707 tsk->thread.trap_no = 8;
709 /* This is always a kernel trap and never fixable (and thus must
712 die(str, regs, error_code);
715 asmlinkage void __kprobes
716 do_general_protection(struct pt_regs *regs, long error_code)
718 struct task_struct *tsk;
720 conditional_sti(regs);
723 if (!user_mode(regs))
726 tsk->thread.error_code = error_code;
727 tsk->thread.trap_no = 13;
729 if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
730 printk_ratelimit()) {
732 "%s[%d] general protection ip:%lx sp:%lx error:%lx",
734 regs->ip, regs->sp, error_code);
735 print_vma_addr(" in ", regs->ip);
739 force_sig(SIGSEGV, tsk);
743 if (fixup_exception(regs))
746 tsk->thread.error_code = error_code;
747 tsk->thread.trap_no = 13;
748 if (notify_die(DIE_GPF, "general protection fault", regs,
749 error_code, 13, SIGSEGV) == NOTIFY_STOP)
751 die("general protection fault", regs, error_code);
754 static notrace __kprobes void
755 mem_parity_error(unsigned char reason, struct pt_regs *regs)
757 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x.\n",
759 printk(KERN_EMERG "You have some hardware problem, likely on the PCI bus.\n");
761 #if defined(CONFIG_EDAC)
762 if (edac_handler_set()) {
763 edac_atomic_assert_error();
768 if (panic_on_unrecovered_nmi)
769 panic("NMI: Not continuing");
771 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
773 /* Clear and disable the memory parity error line. */
774 reason = (reason & 0xf) | 4;
778 static notrace __kprobes void
779 io_check_error(unsigned char reason, struct pt_regs *regs)
781 printk("NMI: IOCK error (debug interrupt?)\n");
782 show_registers(regs);
784 /* Re-enable the IOCK line, wait for a few seconds */
785 reason = (reason & 0xf) | 8;
792 static notrace __kprobes void
793 unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
795 if (notify_die(DIE_NMIUNKNOWN, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
797 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x.\n",
799 printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
801 if (panic_on_unrecovered_nmi)
802 panic("NMI: Not continuing");
804 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
807 /* Runs on IST stack. This code must keep interrupts off all the time.
808 Nested NMIs are prevented by the CPU. */
809 asmlinkage notrace __kprobes void default_do_nmi(struct pt_regs *regs)
811 unsigned char reason = 0;
814 cpu = smp_processor_id();
816 /* Only the BSP gets external NMIs from the system. */
818 reason = get_nmi_reason();
820 if (!(reason & 0xc0)) {
821 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
825 * Ok, so this is none of the documented NMI sources,
826 * so it must be the NMI watchdog.
828 if (nmi_watchdog_tick(regs, reason))
830 if (!do_nmi_callback(regs, cpu))
831 unknown_nmi_error(reason, regs);
835 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
838 /* AK: following checks seem to be broken on modern chipsets. FIXME */
840 mem_parity_error(reason, regs);
842 io_check_error(reason, regs);
845 asmlinkage notrace __kprobes void
846 do_nmi(struct pt_regs *regs, long error_code)
850 add_pda(__nmi_count, 1);
853 default_do_nmi(regs);
864 void restart_nmi(void)
870 /* runs on IST stack. */
871 asmlinkage void __kprobes do_int3(struct pt_regs *regs, long error_code)
873 trace_hardirqs_fixup();
875 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
879 preempt_conditional_sti(regs);
880 do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
881 preempt_conditional_cli(regs);
884 /* Help handler running on IST stack to switch back to user stack
885 for scheduling or signal handling. The actual stack switch is done in
887 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
889 struct pt_regs *regs = eregs;
890 /* Did already sync */
891 if (eregs == (struct pt_regs *)eregs->sp)
893 /* Exception from user space */
894 else if (user_mode(eregs))
895 regs = task_pt_regs(current);
896 /* Exception from kernel and interrupts are enabled. Move to
897 kernel process stack. */
898 else if (eregs->flags & X86_EFLAGS_IF)
899 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
905 /* runs on IST stack. */
906 asmlinkage void __kprobes do_debug(struct pt_regs * regs,
907 unsigned long error_code)
909 struct task_struct *tsk = current;
910 unsigned long condition;
913 trace_hardirqs_fixup();
915 get_debugreg(condition, 6);
918 * The processor cleared BTF, so don't mark that we need it set.
920 clear_tsk_thread_flag(tsk, TIF_DEBUGCTLMSR);
921 tsk->thread.debugctlmsr = 0;
923 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
924 SIGTRAP) == NOTIFY_STOP)
927 preempt_conditional_sti(regs);
929 /* Mask out spurious debug traps due to lazy DR7 setting */
930 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
931 if (!tsk->thread.debugreg7)
935 tsk->thread.debugreg6 = condition;
938 * Single-stepping through TF: make sure we ignore any events in
939 * kernel space (but re-enable TF when returning to user mode).
941 if (condition & DR_STEP) {
942 if (!user_mode(regs))
943 goto clear_TF_reenable;
946 /* Ok, finally something we can handle */
947 tsk->thread.trap_no = 1;
948 tsk->thread.error_code = error_code;
949 info.si_signo = SIGTRAP;
951 info.si_code = TRAP_BRKPT;
952 info.si_addr = user_mode(regs) ? (void __user *)regs->ip : NULL;
953 force_sig_info(SIGTRAP, &info, tsk);
957 preempt_conditional_cli(regs);
961 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
962 regs->flags &= ~X86_EFLAGS_TF;
963 preempt_conditional_cli(regs);
967 static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
969 if (fixup_exception(regs))
972 notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE);
973 /* Illegal floating point operation in the kernel */
974 current->thread.trap_no = trapnr;
980 * Note that we play around with the 'TS' bit in an attempt to get
981 * the correct behaviour even in the presence of the asynchronous
984 asmlinkage void do_coprocessor_error(struct pt_regs *regs)
986 void __user *ip = (void __user *)(regs->ip);
987 struct task_struct *task;
989 unsigned short cwd, swd;
991 conditional_sti(regs);
992 if (!user_mode(regs) &&
993 kernel_math_error(regs, "kernel x87 math error", 16))
997 * Save the info for the exception handler and clear the error.
1000 save_init_fpu(task);
1001 task->thread.trap_no = 16;
1002 task->thread.error_code = 0;
1003 info.si_signo = SIGFPE;
1005 info.si_code = __SI_FAULT;
1008 * (~cwd & swd) will mask out exceptions that are not set to unmasked
1009 * status. 0x3f is the exception bits in these regs, 0x200 is the
1010 * C1 reg you need in case of a stack fault, 0x040 is the stack
1011 * fault bit. We should only be taking one exception at a time,
1012 * so if this combination doesn't produce any single exception,
1013 * then we have a bad program that isn't synchronizing its FPU usage
1014 * and it will suffer the consequences since we won't be able to
1015 * fully reproduce the context of the exception
1017 cwd = get_fpu_cwd(task);
1018 swd = get_fpu_swd(task);
1019 switch (swd & ~cwd & 0x3f) {
1020 case 0x000: /* No unmasked exception */
1021 default: /* Multiple exceptions */
1023 case 0x001: /* Invalid Op */
1025 * swd & 0x240 == 0x040: Stack Underflow
1026 * swd & 0x240 == 0x240: Stack Overflow
1027 * User must clear the SF bit (0x40) if set
1029 info.si_code = FPE_FLTINV;
1031 case 0x002: /* Denormalize */
1032 case 0x010: /* Underflow */
1033 info.si_code = FPE_FLTUND;
1035 case 0x004: /* Zero Divide */
1036 info.si_code = FPE_FLTDIV;
1038 case 0x008: /* Overflow */
1039 info.si_code = FPE_FLTOVF;
1041 case 0x020: /* Precision */
1042 info.si_code = FPE_FLTRES;
1045 force_sig_info(SIGFPE, &info, task);
1048 asmlinkage void bad_intr(void)
1050 printk("bad interrupt");
1053 asmlinkage void do_simd_coprocessor_error(struct pt_regs *regs)
1055 void __user *ip = (void __user *)(regs->ip);
1056 struct task_struct *task;
1058 unsigned short mxcsr;
1060 conditional_sti(regs);
1061 if (!user_mode(regs) &&
1062 kernel_math_error(regs, "kernel simd math error", 19))
1066 * Save the info for the exception handler and clear the error.
1069 save_init_fpu(task);
1070 task->thread.trap_no = 19;
1071 task->thread.error_code = 0;
1072 info.si_signo = SIGFPE;
1074 info.si_code = __SI_FAULT;
1077 * The SIMD FPU exceptions are handled a little differently, as there
1078 * is only a single status/control register. Thus, to determine which
1079 * unmasked exception was caught we must mask the exception mask bits
1080 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1082 mxcsr = get_fpu_mxcsr(task);
1083 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
1087 case 0x001: /* Invalid Op */
1088 info.si_code = FPE_FLTINV;
1090 case 0x002: /* Denormalize */
1091 case 0x010: /* Underflow */
1092 info.si_code = FPE_FLTUND;
1094 case 0x004: /* Zero Divide */
1095 info.si_code = FPE_FLTDIV;
1097 case 0x008: /* Overflow */
1098 info.si_code = FPE_FLTOVF;
1100 case 0x020: /* Precision */
1101 info.si_code = FPE_FLTRES;
1104 force_sig_info(SIGFPE, &info, task);
1107 asmlinkage void do_spurious_interrupt_bug(struct pt_regs * regs)
1111 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
1115 asmlinkage void __attribute__((weak)) mce_threshold_interrupt(void)
1120 * 'math_state_restore()' saves the current math information in the
1121 * old math state array, and gets the new ones from the current task
1123 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1124 * Don't touch unless you *really* know how it works.
1126 asmlinkage void math_state_restore(void)
1128 struct task_struct *me = current;
1133 * does a slab alloc which can sleep
1137 * ran out of memory!
1139 do_group_exit(SIGKILL);
1142 local_irq_disable();
1145 clts(); /* Allow maths ops (or we recurse) */
1146 restore_fpu_checking(&me->thread.xstate->fxsave);
1147 task_thread_info(me)->status |= TS_USEDFPU;
1150 EXPORT_SYMBOL_GPL(math_state_restore);
1152 void __init trap_init(void)
1154 set_intr_gate(0, ÷_error);
1155 set_intr_gate_ist(1, &debug, DEBUG_STACK);
1156 set_intr_gate_ist(2, &nmi, NMI_STACK);
1157 set_system_gate_ist(3, &int3, DEBUG_STACK); /* int3 can be called from all */
1158 set_system_gate(4, &overflow); /* int4 can be called from all */
1159 set_intr_gate(5, &bounds);
1160 set_intr_gate(6, &invalid_op);
1161 set_intr_gate(7, &device_not_available);
1162 set_intr_gate_ist(8, &double_fault, DOUBLEFAULT_STACK);
1163 set_intr_gate(9, &coprocessor_segment_overrun);
1164 set_intr_gate(10, &invalid_TSS);
1165 set_intr_gate(11, &segment_not_present);
1166 set_intr_gate_ist(12, &stack_segment, STACKFAULT_STACK);
1167 set_intr_gate(13, &general_protection);
1168 set_intr_gate(14, &page_fault);
1169 set_intr_gate(15, &spurious_interrupt_bug);
1170 set_intr_gate(16, &coprocessor_error);
1171 set_intr_gate(17, &alignment_check);
1172 #ifdef CONFIG_X86_MCE
1173 set_intr_gate_ist(18, &machine_check, MCE_STACK);
1175 set_intr_gate(19, &simd_coprocessor_error);
1177 #ifdef CONFIG_IA32_EMULATION
1178 set_system_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
1181 * initialize the per thread extended state:
1183 init_thread_xstate();
1185 * Should be a barrier for any external CPU state:
1190 static int __init oops_setup(char *s)
1194 if (!strcmp(s, "panic"))
1198 early_param("oops", oops_setup);
1200 static int __init kstack_setup(char *s)
1204 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1207 early_param("kstack", kstack_setup);
1209 static int __init code_bytes_setup(char *s)
1211 code_bytes = simple_strtoul(s, NULL, 0);
1212 if (code_bytes > 8192)
1217 __setup("code_bytes=", code_bytes_setup);