71c2629997b3332258e51737cef3d8f3eff18009
[deliverable/linux.git] / arch / x86 / kernel / traps_64.c
1 /*
2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
4 *
5 * Pentium III FXSR, SSE support
6 * Gareth Hughes <gareth@valinux.com>, May 2000
7 */
8
9 /*
10 * 'Traps.c' handles hardware traps and faults after we have saved some
11 * state in 'entry.S'.
12 */
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>
34 #include <linux/mm.h>
35 #include <linux/smp.h>
36 #include <linux/io.h>
37
38 #if defined(CONFIG_EDAC)
39 #include <linux/edac.h>
40 #endif
41
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>
48 #include <asm/desc.h>
49 #include <asm/i387.h>
50 #include <asm/pgalloc.h>
51 #include <asm/proto.h>
52 #include <asm/pda.h>
53 #include <asm/traps.h>
54
55 #include <mach_traps.h>
56
57 static int ignore_nmis;
58
59 static inline void conditional_sti(struct pt_regs *regs)
60 {
61 if (regs->flags & X86_EFLAGS_IF)
62 local_irq_enable();
63 }
64
65 static inline void preempt_conditional_sti(struct pt_regs *regs)
66 {
67 inc_preempt_count();
68 if (regs->flags & X86_EFLAGS_IF)
69 local_irq_enable();
70 }
71
72 static inline void preempt_conditional_cli(struct pt_regs *regs)
73 {
74 if (regs->flags & X86_EFLAGS_IF)
75 local_irq_disable();
76 /* Make sure to not schedule here because we could be running
77 on an exception stack. */
78 dec_preempt_count();
79 }
80
81 static void __kprobes
82 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
83 long error_code, siginfo_t *info)
84 {
85 struct task_struct *tsk = current;
86
87 if (!user_mode(regs))
88 goto kernel_trap;
89
90 /*
91 * We want error_code and trap_no set for userspace faults and
92 * kernelspace faults which result in die(), but not
93 * kernelspace faults which are fixed up. die() gives the
94 * process no chance to handle the signal and notice the
95 * kernel fault information, so that won't result in polluting
96 * the information about previously queued, but not yet
97 * delivered, faults. See also do_general_protection below.
98 */
99 tsk->thread.error_code = error_code;
100 tsk->thread.trap_no = trapnr;
101
102 if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
103 printk_ratelimit()) {
104 printk(KERN_INFO
105 "%s[%d] trap %s ip:%lx sp:%lx error:%lx",
106 tsk->comm, tsk->pid, str,
107 regs->ip, regs->sp, error_code);
108 print_vma_addr(" in ", regs->ip);
109 printk("\n");
110 }
111
112 if (info)
113 force_sig_info(signr, info, tsk);
114 else
115 force_sig(signr, tsk);
116 return;
117
118 kernel_trap:
119 if (!fixup_exception(regs)) {
120 tsk->thread.error_code = error_code;
121 tsk->thread.trap_no = trapnr;
122 die(str, regs, error_code);
123 }
124 return;
125 }
126
127 #define DO_ERROR(trapnr, signr, str, name) \
128 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
129 { \
130 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
131 == NOTIFY_STOP) \
132 return; \
133 conditional_sti(regs); \
134 do_trap(trapnr, signr, str, regs, error_code, NULL); \
135 }
136
137 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
138 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
139 { \
140 siginfo_t info; \
141 info.si_signo = signr; \
142 info.si_errno = 0; \
143 info.si_code = sicode; \
144 info.si_addr = (void __user *)siaddr; \
145 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
146 == NOTIFY_STOP) \
147 return; \
148 conditional_sti(regs); \
149 do_trap(trapnr, signr, str, regs, error_code, &info); \
150 }
151
152 DO_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
153 DO_ERROR(4, SIGSEGV, "overflow", overflow)
154 DO_ERROR(5, SIGSEGV, "bounds", bounds)
155 DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip)
156 DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
157 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
158 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
159 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
160
161 /* Runs on IST stack */
162 dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
163 {
164 if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
165 12, SIGBUS) == NOTIFY_STOP)
166 return;
167 preempt_conditional_sti(regs);
168 do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
169 preempt_conditional_cli(regs);
170 }
171
172 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
173 {
174 static const char str[] = "double fault";
175 struct task_struct *tsk = current;
176
177 /* Return not checked because double check cannot be ignored */
178 notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
179
180 tsk->thread.error_code = error_code;
181 tsk->thread.trap_no = 8;
182
183 /* This is always a kernel trap and never fixable (and thus must
184 never return). */
185 for (;;)
186 die(str, regs, error_code);
187 }
188
189 dotraplinkage void __kprobes
190 do_general_protection(struct pt_regs *regs, long error_code)
191 {
192 struct task_struct *tsk;
193
194 conditional_sti(regs);
195
196 tsk = current;
197 if (!user_mode(regs))
198 goto gp_in_kernel;
199
200 tsk->thread.error_code = error_code;
201 tsk->thread.trap_no = 13;
202
203 if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
204 printk_ratelimit()) {
205 printk(KERN_INFO
206 "%s[%d] general protection ip:%lx sp:%lx error:%lx",
207 tsk->comm, task_pid_nr(tsk),
208 regs->ip, regs->sp, error_code);
209 print_vma_addr(" in ", regs->ip);
210 printk("\n");
211 }
212
213 force_sig(SIGSEGV, tsk);
214 return;
215
216 gp_in_kernel:
217 if (fixup_exception(regs))
218 return;
219
220 tsk->thread.error_code = error_code;
221 tsk->thread.trap_no = 13;
222 if (notify_die(DIE_GPF, "general protection fault", regs,
223 error_code, 13, SIGSEGV) == NOTIFY_STOP)
224 return;
225 die("general protection fault", regs, error_code);
226 }
227
228 static notrace __kprobes void
229 mem_parity_error(unsigned char reason, struct pt_regs *regs)
230 {
231 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x.\n",
232 reason);
233 printk(KERN_EMERG "You have some hardware problem, likely on the PCI bus.\n");
234
235 #if defined(CONFIG_EDAC)
236 if (edac_handler_set()) {
237 edac_atomic_assert_error();
238 return;
239 }
240 #endif
241
242 if (panic_on_unrecovered_nmi)
243 panic("NMI: Not continuing");
244
245 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
246
247 /* Clear and disable the memory parity error line. */
248 reason = (reason & 0xf) | 4;
249 outb(reason, 0x61);
250 }
251
252 static notrace __kprobes void
253 io_check_error(unsigned char reason, struct pt_regs *regs)
254 {
255 unsigned long i;
256
257 printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
258 show_registers(regs);
259
260 /* Re-enable the IOCK line, wait for a few seconds */
261 reason = (reason & 0xf) | 8;
262 outb(reason, 0x61);
263
264 i = 2000;
265 while (--i)
266 udelay(1000);
267
268 reason &= ~8;
269 outb(reason, 0x61);
270 }
271
272 static notrace __kprobes void
273 unknown_nmi_error(unsigned char reason, struct pt_regs *regs)
274 {
275 if (notify_die(DIE_NMIUNKNOWN, "nmi", regs, reason, 2, SIGINT) ==
276 NOTIFY_STOP)
277 return;
278 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x.\n",
279 reason);
280 printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
281
282 if (panic_on_unrecovered_nmi)
283 panic("NMI: Not continuing");
284
285 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
286 }
287
288 /* Runs on IST stack. This code must keep interrupts off all the time.
289 Nested NMIs are prevented by the CPU. */
290 asmlinkage notrace __kprobes void default_do_nmi(struct pt_regs *regs)
291 {
292 unsigned char reason = 0;
293 int cpu;
294
295 cpu = smp_processor_id();
296
297 /* Only the BSP gets external NMIs from the system. */
298 if (!cpu)
299 reason = get_nmi_reason();
300
301 if (!(reason & 0xc0)) {
302 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
303 == NOTIFY_STOP)
304 return;
305 /*
306 * Ok, so this is none of the documented NMI sources,
307 * so it must be the NMI watchdog.
308 */
309 if (nmi_watchdog_tick(regs, reason))
310 return;
311 if (!do_nmi_callback(regs, cpu))
312 unknown_nmi_error(reason, regs);
313
314 return;
315 }
316 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
317 return;
318
319 /* AK: following checks seem to be broken on modern chipsets. FIXME */
320 if (reason & 0x80)
321 mem_parity_error(reason, regs);
322 if (reason & 0x40)
323 io_check_error(reason, regs);
324 }
325
326 dotraplinkage notrace __kprobes void
327 do_nmi(struct pt_regs *regs, long error_code)
328 {
329 nmi_enter();
330
331 add_pda(__nmi_count, 1);
332
333 if (!ignore_nmis)
334 default_do_nmi(regs);
335
336 nmi_exit();
337 }
338
339 void stop_nmi(void)
340 {
341 acpi_nmi_disable();
342 ignore_nmis++;
343 }
344
345 void restart_nmi(void)
346 {
347 ignore_nmis--;
348 acpi_nmi_enable();
349 }
350
351 /* runs on IST stack. */
352 dotraplinkage void __kprobes do_int3(struct pt_regs *regs, long error_code)
353 {
354 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
355 == NOTIFY_STOP)
356 return;
357
358 preempt_conditional_sti(regs);
359 do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
360 preempt_conditional_cli(regs);
361 }
362
363 /* Help handler running on IST stack to switch back to user stack
364 for scheduling or signal handling. The actual stack switch is done in
365 entry.S */
366 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
367 {
368 struct pt_regs *regs = eregs;
369 /* Did already sync */
370 if (eregs == (struct pt_regs *)eregs->sp)
371 ;
372 /* Exception from user space */
373 else if (user_mode(eregs))
374 regs = task_pt_regs(current);
375 /* Exception from kernel and interrupts are enabled. Move to
376 kernel process stack. */
377 else if (eregs->flags & X86_EFLAGS_IF)
378 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
379 if (eregs != regs)
380 *regs = *eregs;
381 return regs;
382 }
383
384 /* runs on IST stack. */
385 dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
386 {
387 struct task_struct *tsk = current;
388 unsigned long condition;
389 int si_code;
390
391 get_debugreg(condition, 6);
392
393 /*
394 * The processor cleared BTF, so don't mark that we need it set.
395 */
396 clear_tsk_thread_flag(tsk, TIF_DEBUGCTLMSR);
397 tsk->thread.debugctlmsr = 0;
398
399 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
400 SIGTRAP) == NOTIFY_STOP)
401 return;
402
403 /* It's safe to allow irq's after DR6 has been saved */
404 preempt_conditional_sti(regs);
405
406 /* Mask out spurious debug traps due to lazy DR7 setting */
407 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
408 if (!tsk->thread.debugreg7)
409 goto clear_dr7;
410 }
411
412 /* Save debug status register where ptrace can see it */
413 tsk->thread.debugreg6 = condition;
414
415 /*
416 * Single-stepping through TF: make sure we ignore any events in
417 * kernel space (but re-enable TF when returning to user mode).
418 */
419 if (condition & DR_STEP) {
420 if (!user_mode(regs))
421 goto clear_TF_reenable;
422 }
423
424 si_code = get_si_code(condition);
425 /* Ok, finally something we can handle */
426 send_sigtrap(tsk, regs, error_code, si_code);
427
428 /*
429 * Disable additional traps. They'll be re-enabled when
430 * the signal is delivered.
431 */
432 clear_dr7:
433 set_debugreg(0, 7);
434 preempt_conditional_cli(regs);
435 return;
436
437 clear_TF_reenable:
438 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
439 regs->flags &= ~X86_EFLAGS_TF;
440 preempt_conditional_cli(regs);
441 return;
442 }
443
444 static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
445 {
446 if (fixup_exception(regs))
447 return 1;
448
449 notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE);
450 /* Illegal floating point operation in the kernel */
451 current->thread.trap_no = trapnr;
452 die(str, regs, 0);
453 return 0;
454 }
455
456 /*
457 * Note that we play around with the 'TS' bit in an attempt to get
458 * the correct behaviour even in the presence of the asynchronous
459 * IRQ13 behaviour
460 */
461 void math_error(void __user *ip)
462 {
463 struct task_struct *task;
464 siginfo_t info;
465 unsigned short cwd, swd;
466
467 /*
468 * Save the info for the exception handler and clear the error.
469 */
470 task = current;
471 save_init_fpu(task);
472 task->thread.trap_no = 16;
473 task->thread.error_code = 0;
474 info.si_signo = SIGFPE;
475 info.si_errno = 0;
476 info.si_code = __SI_FAULT;
477 info.si_addr = ip;
478 /*
479 * (~cwd & swd) will mask out exceptions that are not set to unmasked
480 * status. 0x3f is the exception bits in these regs, 0x200 is the
481 * C1 reg you need in case of a stack fault, 0x040 is the stack
482 * fault bit. We should only be taking one exception at a time,
483 * so if this combination doesn't produce any single exception,
484 * then we have a bad program that isn't synchronizing its FPU usage
485 * and it will suffer the consequences since we won't be able to
486 * fully reproduce the context of the exception
487 */
488 cwd = get_fpu_cwd(task);
489 swd = get_fpu_swd(task);
490 switch (swd & ~cwd & 0x3f) {
491 case 0x000: /* No unmasked exception */
492 default: /* Multiple exceptions */
493 break;
494 case 0x001: /* Invalid Op */
495 /*
496 * swd & 0x240 == 0x040: Stack Underflow
497 * swd & 0x240 == 0x240: Stack Overflow
498 * User must clear the SF bit (0x40) if set
499 */
500 info.si_code = FPE_FLTINV;
501 break;
502 case 0x002: /* Denormalize */
503 case 0x010: /* Underflow */
504 info.si_code = FPE_FLTUND;
505 break;
506 case 0x004: /* Zero Divide */
507 info.si_code = FPE_FLTDIV;
508 break;
509 case 0x008: /* Overflow */
510 info.si_code = FPE_FLTOVF;
511 break;
512 case 0x020: /* Precision */
513 info.si_code = FPE_FLTRES;
514 break;
515 }
516 force_sig_info(SIGFPE, &info, task);
517 }
518
519 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
520 {
521 conditional_sti(regs);
522 if (!user_mode(regs) &&
523 kernel_math_error(regs, "kernel x87 math error", 16))
524 return;
525 math_error((void __user *)regs->ip);
526 }
527
528 asmlinkage void bad_intr(void)
529 {
530 printk("bad interrupt");
531 }
532
533 static void simd_math_error(void __user *ip)
534 {
535 struct task_struct *task;
536 siginfo_t info;
537 unsigned short mxcsr;
538
539 /*
540 * Save the info for the exception handler and clear the error.
541 */
542 task = current;
543 save_init_fpu(task);
544 task->thread.trap_no = 19;
545 task->thread.error_code = 0;
546 info.si_signo = SIGFPE;
547 info.si_errno = 0;
548 info.si_code = __SI_FAULT;
549 info.si_addr = ip;
550 /*
551 * The SIMD FPU exceptions are handled a little differently, as there
552 * is only a single status/control register. Thus, to determine which
553 * unmasked exception was caught we must mask the exception mask bits
554 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
555 */
556 mxcsr = get_fpu_mxcsr(task);
557 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
558 case 0x000:
559 default:
560 break;
561 case 0x001: /* Invalid Op */
562 info.si_code = FPE_FLTINV;
563 break;
564 case 0x002: /* Denormalize */
565 case 0x010: /* Underflow */
566 info.si_code = FPE_FLTUND;
567 break;
568 case 0x004: /* Zero Divide */
569 info.si_code = FPE_FLTDIV;
570 break;
571 case 0x008: /* Overflow */
572 info.si_code = FPE_FLTOVF;
573 break;
574 case 0x020: /* Precision */
575 info.si_code = FPE_FLTRES;
576 break;
577 }
578 force_sig_info(SIGFPE, &info, task);
579 }
580
581 dotraplinkage void
582 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
583 {
584 conditional_sti(regs);
585 if (!user_mode(regs) &&
586 kernel_math_error(regs, "kernel simd math error", 19))
587 return;
588 simd_math_error((void __user *)regs->ip);
589 }
590
591 dotraplinkage void
592 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
593 {
594 }
595
596 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
597 {
598 }
599
600 asmlinkage void __attribute__((weak)) mce_threshold_interrupt(void)
601 {
602 }
603
604 /*
605 * 'math_state_restore()' saves the current math information in the
606 * old math state array, and gets the new ones from the current task
607 *
608 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
609 * Don't touch unless you *really* know how it works.
610 */
611 asmlinkage void math_state_restore(void)
612 {
613 struct thread_info *thread = current_thread_info();
614 struct task_struct *tsk = thread->task;
615
616 if (!tsk_used_math(tsk)) {
617 local_irq_enable();
618 /*
619 * does a slab alloc which can sleep
620 */
621 if (init_fpu(tsk)) {
622 /*
623 * ran out of memory!
624 */
625 do_group_exit(SIGKILL);
626 return;
627 }
628 local_irq_disable();
629 }
630
631 clts(); /* Allow maths ops (or we recurse) */
632 /*
633 * Paranoid restore. send a SIGSEGV if we fail to restore the state.
634 */
635 if (unlikely(restore_fpu_checking(tsk))) {
636 stts();
637 force_sig(SIGSEGV, tsk);
638 return;
639 }
640 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
641 tsk->fpu_counter++;
642 }
643 EXPORT_SYMBOL_GPL(math_state_restore);
644
645 dotraplinkage void __kprobes
646 do_device_not_available(struct pt_regs *regs, long error)
647 {
648 math_state_restore();
649 }
650
651 void __init trap_init(void)
652 {
653 set_intr_gate(0, &divide_error);
654 set_intr_gate_ist(1, &debug, DEBUG_STACK);
655 set_intr_gate_ist(2, &nmi, NMI_STACK);
656 /* int3 can be called from all */
657 set_system_intr_gate_ist(3, &int3, DEBUG_STACK);
658 /* int4 can be called from all */
659 set_system_intr_gate(4, &overflow);
660 set_intr_gate(5, &bounds);
661 set_intr_gate(6, &invalid_op);
662 set_intr_gate(7, &device_not_available);
663 set_intr_gate_ist(8, &double_fault, DOUBLEFAULT_STACK);
664 set_intr_gate(9, &coprocessor_segment_overrun);
665 set_intr_gate(10, &invalid_TSS);
666 set_intr_gate(11, &segment_not_present);
667 set_intr_gate_ist(12, &stack_segment, STACKFAULT_STACK);
668 set_intr_gate(13, &general_protection);
669 set_intr_gate(14, &page_fault);
670 set_intr_gate(15, &spurious_interrupt_bug);
671 set_intr_gate(16, &coprocessor_error);
672 set_intr_gate(17, &alignment_check);
673 #ifdef CONFIG_X86_MCE
674 set_intr_gate_ist(18, &machine_check, MCE_STACK);
675 #endif
676 set_intr_gate(19, &simd_coprocessor_error);
677
678 #ifdef CONFIG_IA32_EMULATION
679 set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
680 #endif
681 /*
682 * Should be a barrier for any external CPU state:
683 */
684 cpu_init();
685 }
This page took 0.053058 seconds and 4 git commands to generate.