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 * Handle hardware traps and faults.
12 #include <linux/interrupt.h>
13 #include <linux/kallsyms.h>
14 #include <linux/spinlock.h>
15 #include <linux/kprobes.h>
16 #include <linux/uaccess.h>
17 #include <linux/kdebug.h>
18 #include <linux/kgdb.h>
19 #include <linux/kernel.h>
20 #include <linux/module.h>
21 #include <linux/ptrace.h>
22 #include <linux/string.h>
23 #include <linux/delay.h>
24 #include <linux/errno.h>
25 #include <linux/kexec.h>
26 #include <linux/sched.h>
27 #include <linux/timer.h>
28 #include <linux/init.h>
29 #include <linux/bug.h>
30 #include <linux/nmi.h>
32 #include <linux/smp.h>
36 #include <linux/ioport.h>
37 #include <linux/eisa.h>
41 #include <linux/mca.h>
44 #if defined(CONFIG_EDAC)
45 #include <linux/edac.h>
48 #include <asm/kmemcheck.h>
49 #include <asm/stacktrace.h>
50 #include <asm/processor.h>
51 #include <asm/debugreg.h>
52 #include <linux/atomic.h>
53 #include <asm/system.h>
54 #include <asm/traps.h>
59 #include <asm/mach_traps.h>
62 #include <asm/x86_init.h>
63 #include <asm/pgalloc.h>
64 #include <asm/proto.h>
66 #include <asm/processor-flags.h>
67 #include <asm/setup.h>
69 asmlinkage int system_call(void);
71 /* Do we ignore FPU interrupts ? */
75 * The IDT has to be page-aligned to simplify the Pentium
76 * F0 0F bug workaround.
78 gate_desc idt_table[NR_VECTORS] __page_aligned_data = { { { { 0, 0 } } }, };
81 DECLARE_BITMAP(used_vectors, NR_VECTORS);
82 EXPORT_SYMBOL_GPL(used_vectors);
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 conditional_cli(struct pt_regs *regs)
99 if (regs->flags & X86_EFLAGS_IF)
103 static inline void preempt_conditional_cli(struct pt_regs *regs)
105 if (regs->flags & X86_EFLAGS_IF)
110 static void __kprobes
111 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
112 long error_code, siginfo_t *info)
114 struct task_struct *tsk = current;
117 if (regs->flags & X86_VM_MASK) {
119 * traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
120 * On nmi (interrupt 2), do_trap should not be called.
122 if (trapnr < X86_TRAP_UD)
128 if (!user_mode(regs))
135 * We want error_code and trap_no set for userspace faults and
136 * kernelspace faults which result in die(), but not
137 * kernelspace faults which are fixed up. die() gives the
138 * process no chance to handle the signal and notice the
139 * kernel fault information, so that won't result in polluting
140 * the information about previously queued, but not yet
141 * delivered, faults. See also do_general_protection below.
143 tsk->thread.error_code = error_code;
144 tsk->thread.trap_no = trapnr;
147 if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
148 printk_ratelimit()) {
150 "%s[%d] trap %s ip:%lx sp:%lx error:%lx",
151 tsk->comm, tsk->pid, str,
152 regs->ip, regs->sp, error_code);
153 print_vma_addr(" in ", regs->ip);
159 force_sig_info(signr, info, tsk);
161 force_sig(signr, tsk);
165 if (!fixup_exception(regs)) {
166 tsk->thread.error_code = error_code;
167 tsk->thread.trap_no = trapnr;
168 die(str, regs, error_code);
174 if (handle_vm86_trap((struct kernel_vm86_regs *) regs,
181 #define DO_ERROR(trapnr, signr, str, name) \
182 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
184 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
187 conditional_sti(regs); \
188 do_trap(trapnr, signr, str, regs, error_code, NULL); \
191 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
192 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
195 info.si_signo = signr; \
197 info.si_code = sicode; \
198 info.si_addr = (void __user *)siaddr; \
199 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
202 conditional_sti(regs); \
203 do_trap(trapnr, signr, str, regs, error_code, &info); \
206 DO_ERROR_INFO(X86_TRAP_DE, SIGFPE, "divide error", divide_error, FPE_INTDIV,
208 DO_ERROR(X86_TRAP_OF, SIGSEGV, "overflow", overflow)
209 DO_ERROR(X86_TRAP_BR, SIGSEGV, "bounds", bounds)
210 DO_ERROR_INFO(X86_TRAP_UD, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN,
212 DO_ERROR(X86_TRAP_OLD_MF, SIGFPE, "coprocessor segment overrun",
213 coprocessor_segment_overrun)
214 DO_ERROR(X86_TRAP_TS, SIGSEGV, "invalid TSS", invalid_TSS)
215 DO_ERROR(X86_TRAP_NP, SIGBUS, "segment not present", segment_not_present)
216 DO_ERROR(X86_TRAP_SS, SIGBUS, "stack segment", stack_segment)
217 DO_ERROR_INFO(X86_TRAP_AC, SIGBUS, "alignment check", alignment_check,
221 /* Runs on IST stack */
222 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
224 static const char str[] = "double fault";
225 struct task_struct *tsk = current;
227 /* Return not checked because double check cannot be ignored */
228 notify_die(DIE_TRAP, str, regs, error_code, X86_TRAP_DF, SIGSEGV);
230 tsk->thread.error_code = error_code;
231 tsk->thread.trap_no = X86_TRAP_DF;
234 * This is always a kernel trap and never fixable (and thus must
238 die(str, regs, error_code);
242 dotraplinkage void __kprobes
243 do_general_protection(struct pt_regs *regs, long error_code)
245 struct task_struct *tsk;
247 conditional_sti(regs);
250 if (regs->flags & X86_VM_MASK)
255 if (!user_mode(regs))
258 tsk->thread.error_code = error_code;
259 tsk->thread.trap_no = X86_TRAP_GP;
261 if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
262 printk_ratelimit()) {
264 "%s[%d] general protection ip:%lx sp:%lx error:%lx",
265 tsk->comm, task_pid_nr(tsk),
266 regs->ip, regs->sp, error_code);
267 print_vma_addr(" in ", regs->ip);
271 force_sig(SIGSEGV, tsk);
277 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
282 if (fixup_exception(regs))
285 tsk->thread.error_code = error_code;
286 tsk->thread.trap_no = X86_TRAP_GP;
287 if (notify_die(DIE_GPF, "general protection fault", regs, error_code,
288 X86_TRAP_GP, SIGSEGV) == NOTIFY_STOP)
290 die("general protection fault", regs, error_code);
293 /* May run on IST stack. */
294 dotraplinkage void __kprobes do_int3(struct pt_regs *regs, long error_code)
296 #ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
297 if (kgdb_ll_trap(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
298 SIGTRAP) == NOTIFY_STOP)
300 #endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
301 #ifdef CONFIG_KPROBES
303 if (notify_die(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
304 SIGTRAP) == NOTIFY_STOP)
307 if (notify_die(DIE_TRAP, "int3", regs, error_code, 3, SIGTRAP)
312 preempt_conditional_sti(regs);
313 do_trap(X86_TRAP_BP, SIGTRAP, "int3", regs, error_code, NULL);
314 preempt_conditional_cli(regs);
319 * Help handler running on IST stack to switch back to user stack
320 * for scheduling or signal handling. The actual stack switch is done in
323 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
325 struct pt_regs *regs = eregs;
326 /* Did already sync */
327 if (eregs == (struct pt_regs *)eregs->sp)
329 /* Exception from user space */
330 else if (user_mode(eregs))
331 regs = task_pt_regs(current);
333 * Exception from kernel and interrupts are enabled. Move to
334 * kernel process stack.
336 else if (eregs->flags & X86_EFLAGS_IF)
337 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
345 * Our handling of the processor debug registers is non-trivial.
346 * We do not clear them on entry and exit from the kernel. Therefore
347 * it is possible to get a watchpoint trap here from inside the kernel.
348 * However, the code in ./ptrace.c has ensured that the user can
349 * only set watchpoints on userspace addresses. Therefore the in-kernel
350 * watchpoint trap can only occur in code which is reading/writing
351 * from user space. Such code must not hold kernel locks (since it
352 * can equally take a page fault), therefore it is safe to call
353 * force_sig_info even though that claims and releases locks.
355 * Code in ./signal.c ensures that the debug control register
356 * is restored before we deliver any signal, and therefore that
357 * user code runs with the correct debug control register even though
360 * Being careful here means that we don't have to be as careful in a
361 * lot of more complicated places (task switching can be a bit lazy
362 * about restoring all the debug state, and ptrace doesn't have to
363 * find every occurrence of the TF bit that could be saved away even
366 * May run on IST stack.
368 dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
370 struct task_struct *tsk = current;
375 get_debugreg(dr6, 6);
377 /* Filter out all the reserved bits which are preset to 1 */
378 dr6 &= ~DR6_RESERVED;
381 * If dr6 has no reason to give us about the origin of this trap,
382 * then it's very likely the result of an icebp/int01 trap.
383 * User wants a sigtrap for that.
385 if (!dr6 && user_mode(regs))
388 /* Catch kmemcheck conditions first of all! */
389 if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
392 /* DR6 may or may not be cleared by the CPU */
396 * The processor cleared BTF, so don't mark that we need it set.
398 clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
400 /* Store the virtualized DR6 value */
401 tsk->thread.debugreg6 = dr6;
403 if (notify_die(DIE_DEBUG, "debug", regs, PTR_ERR(&dr6), error_code,
404 SIGTRAP) == NOTIFY_STOP)
407 /* It's safe to allow irq's after DR6 has been saved */
408 preempt_conditional_sti(regs);
410 if (regs->flags & X86_VM_MASK) {
411 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code,
413 preempt_conditional_cli(regs);
418 * Single-stepping through system calls: ignore any exceptions in
419 * kernel space, but re-enable TF when returning to user mode.
421 * We already checked v86 mode above, so we can check for kernel mode
422 * by just checking the CPL of CS.
424 if ((dr6 & DR_STEP) && !user_mode(regs)) {
425 tsk->thread.debugreg6 &= ~DR_STEP;
426 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
427 regs->flags &= ~X86_EFLAGS_TF;
429 si_code = get_si_code(tsk->thread.debugreg6);
430 if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS) || user_icebp)
431 send_sigtrap(tsk, regs, error_code, si_code);
432 preempt_conditional_cli(regs);
438 * Note that we play around with the 'TS' bit in an attempt to get
439 * the correct behaviour even in the presence of the asynchronous
442 void math_error(struct pt_regs *regs, int error_code, int trapnr)
444 struct task_struct *task = current;
447 char *str = (trapnr == X86_TRAP_MF) ? "fpu exception" :
450 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, SIGFPE) == NOTIFY_STOP)
452 conditional_sti(regs);
454 if (!user_mode_vm(regs))
456 if (!fixup_exception(regs)) {
457 task->thread.error_code = error_code;
458 task->thread.trap_no = trapnr;
459 die(str, regs, error_code);
465 * Save the info for the exception handler and clear the error.
468 task->thread.trap_no = trapnr;
469 task->thread.error_code = error_code;
470 info.si_signo = SIGFPE;
472 info.si_addr = (void __user *)regs->ip;
473 if (trapnr == X86_TRAP_MF) {
474 unsigned short cwd, swd;
476 * (~cwd & swd) will mask out exceptions that are not set to unmasked
477 * status. 0x3f is the exception bits in these regs, 0x200 is the
478 * C1 reg you need in case of a stack fault, 0x040 is the stack
479 * fault bit. We should only be taking one exception at a time,
480 * so if this combination doesn't produce any single exception,
481 * then we have a bad program that isn't synchronizing its FPU usage
482 * and it will suffer the consequences since we won't be able to
483 * fully reproduce the context of the exception
485 cwd = get_fpu_cwd(task);
486 swd = get_fpu_swd(task);
491 * The SIMD FPU exceptions are handled a little differently, as there
492 * is only a single status/control register. Thus, to determine which
493 * unmasked exception was caught we must mask the exception mask bits
494 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
496 unsigned short mxcsr = get_fpu_mxcsr(task);
497 err = ~(mxcsr >> 7) & mxcsr;
500 if (err & 0x001) { /* Invalid op */
502 * swd & 0x240 == 0x040: Stack Underflow
503 * swd & 0x240 == 0x240: Stack Overflow
504 * User must clear the SF bit (0x40) if set
506 info.si_code = FPE_FLTINV;
507 } else if (err & 0x004) { /* Divide by Zero */
508 info.si_code = FPE_FLTDIV;
509 } else if (err & 0x008) { /* Overflow */
510 info.si_code = FPE_FLTOVF;
511 } else if (err & 0x012) { /* Denormal, Underflow */
512 info.si_code = FPE_FLTUND;
513 } else if (err & 0x020) { /* Precision */
514 info.si_code = FPE_FLTRES;
517 * If we're using IRQ 13, or supposedly even some trap
518 * X86_TRAP_MF implementations, it's possible
519 * we get a spurious trap, which is not an error.
523 force_sig_info(SIGFPE, &info, task);
526 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
532 math_error(regs, error_code, X86_TRAP_MF);
536 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
538 math_error(regs, error_code, X86_TRAP_XF);
542 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
544 conditional_sti(regs);
546 /* No need to warn about this any longer. */
547 printk(KERN_INFO "Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
551 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
555 asmlinkage void __attribute__((weak)) smp_threshold_interrupt(void)
560 * This gets called with the process already owning the
561 * FPU state, and with CR0.TS cleared. It just needs to
562 * restore the FPU register state.
564 void __math_state_restore(struct task_struct *tsk)
566 /* We need a safe address that is cheap to find and that is already
567 in L1. We've just brought in "tsk->thread.has_fpu", so use that */
568 #define safe_address (tsk->thread.has_fpu)
570 /* AMD K7/K8 CPUs don't save/restore FDP/FIP/FOP unless an exception
571 is pending. Clear the x87 state here by setting it to fixed
572 values. safe_address is a random variable that should be in L1 */
573 if (unlikely(static_cpu_has(X86_FEATURE_FXSAVE_LEAK))) {
577 "fildl %P[addr]" /* set F?P to defined value */
578 : : [addr] "m" (safe_address));
582 * Paranoid restore. send a SIGSEGV if we fail to restore the state.
584 if (unlikely(restore_fpu_checking(tsk))) {
585 __thread_fpu_end(tsk);
586 force_sig(SIGSEGV, tsk);
592 * 'math_state_restore()' saves the current math information in the
593 * old math state array, and gets the new ones from the current task
595 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
596 * Don't touch unless you *really* know how it works.
598 * Must be called with kernel preemption disabled (eg with local
599 * local interrupts as in the case of do_device_not_available).
601 void math_state_restore(void)
603 struct task_struct *tsk = current;
605 if (!tsk_used_math(tsk)) {
608 * does a slab alloc which can sleep
614 do_group_exit(SIGKILL);
620 __thread_fpu_begin(tsk);
621 __math_state_restore(tsk);
625 EXPORT_SYMBOL_GPL(math_state_restore);
627 dotraplinkage void __kprobes
628 do_device_not_available(struct pt_regs *regs, long error_code)
630 #ifdef CONFIG_MATH_EMULATION
631 if (read_cr0() & X86_CR0_EM) {
632 struct math_emu_info info = { };
634 conditional_sti(regs);
641 math_state_restore(); /* interrupts still off */
643 conditional_sti(regs);
648 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
653 info.si_signo = SIGILL;
655 info.si_code = ILL_BADSTK;
657 if (notify_die(DIE_TRAP, "iret exception", regs, error_code,
658 X86_TRAP_IRET, SIGILL) == NOTIFY_STOP)
660 do_trap(X86_TRAP_IRET, SIGILL, "iret exception", regs, error_code,
665 /* Set of traps needed for early debugging. */
666 void __init early_trap_init(void)
668 set_intr_gate_ist(X86_TRAP_DB, &debug, DEBUG_STACK);
669 /* int3 can be called from all */
670 set_system_intr_gate_ist(X86_TRAP_BP, &int3, DEBUG_STACK);
671 set_intr_gate(X86_TRAP_PF, &page_fault);
672 load_idt(&idt_descr);
675 void __init trap_init(void)
680 void __iomem *p = early_ioremap(0x0FFFD9, 4);
682 if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
687 set_intr_gate(X86_TRAP_DE, ÷_error);
688 set_intr_gate_ist(X86_TRAP_NMI, &nmi, NMI_STACK);
689 /* int4 can be called from all */
690 set_system_intr_gate(X86_TRAP_OF, &overflow);
691 set_intr_gate(X86_TRAP_BR, &bounds);
692 set_intr_gate(X86_TRAP_UD, &invalid_op);
693 set_intr_gate(X86_TRAP_NM, &device_not_available);
695 set_task_gate(X86_TRAP_DF, GDT_ENTRY_DOUBLEFAULT_TSS);
697 set_intr_gate_ist(X86_TRAP_DF, &double_fault, DOUBLEFAULT_STACK);
699 set_intr_gate(X86_TRAP_OLD_MF, &coprocessor_segment_overrun);
700 set_intr_gate(X86_TRAP_TS, &invalid_TSS);
701 set_intr_gate(X86_TRAP_NP, &segment_not_present);
702 set_intr_gate(X86_TRAP_SS, stack_segment);
703 set_intr_gate(X86_TRAP_GP, &general_protection);
704 set_intr_gate(X86_TRAP_SPURIOUS, &spurious_interrupt_bug);
705 set_intr_gate(X86_TRAP_MF, &coprocessor_error);
706 set_intr_gate(X86_TRAP_AC, &alignment_check);
707 #ifdef CONFIG_X86_MCE
708 set_intr_gate_ist(X86_TRAP_MC, &machine_check, MCE_STACK);
710 set_intr_gate(X86_TRAP_XF, &simd_coprocessor_error);
712 /* Reserve all the builtin and the syscall vector: */
713 for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
714 set_bit(i, used_vectors);
716 #ifdef CONFIG_IA32_EMULATION
717 set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
718 set_bit(IA32_SYSCALL_VECTOR, used_vectors);
722 set_system_trap_gate(SYSCALL_VECTOR, &system_call);
723 set_bit(SYSCALL_VECTOR, used_vectors);
727 * Should be a barrier for any external CPU state:
731 x86_init.irqs.trap_init();