Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/arch/x86-64/traps.c | |
3 | * | |
4 | * Copyright (C) 1991, 1992 Linus Torvalds | |
5 | * Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs | |
6 | * | |
7 | * Pentium III FXSR, SSE support | |
8 | * Gareth Hughes <gareth@valinux.com>, May 2000 | |
9 | * | |
10 | * $Id: traps.c,v 1.36 2002/03/24 11:09:10 ak Exp $ | |
11 | */ | |
12 | ||
13 | /* | |
14 | * 'Traps.c' handles hardware traps and faults after we have saved some | |
15 | * state in 'entry.S'. | |
16 | */ | |
17 | #include <linux/config.h> | |
18 | #include <linux/sched.h> | |
19 | #include <linux/kernel.h> | |
20 | #include <linux/string.h> | |
21 | #include <linux/errno.h> | |
22 | #include <linux/ptrace.h> | |
23 | #include <linux/timer.h> | |
24 | #include <linux/mm.h> | |
25 | #include <linux/init.h> | |
26 | #include <linux/delay.h> | |
27 | #include <linux/spinlock.h> | |
28 | #include <linux/interrupt.h> | |
29 | #include <linux/module.h> | |
30 | #include <linux/moduleparam.h> | |
35faa714 | 31 | #include <linux/nmi.h> |
1da177e4 LT |
32 | |
33 | #include <asm/system.h> | |
34 | #include <asm/uaccess.h> | |
35 | #include <asm/io.h> | |
36 | #include <asm/atomic.h> | |
37 | #include <asm/debugreg.h> | |
38 | #include <asm/desc.h> | |
39 | #include <asm/i387.h> | |
40 | #include <asm/kdebug.h> | |
41 | #include <asm/processor.h> | |
42 | ||
43 | #include <asm/smp.h> | |
44 | #include <asm/pgalloc.h> | |
45 | #include <asm/pda.h> | |
46 | #include <asm/proto.h> | |
47 | #include <asm/nmi.h> | |
48 | ||
49 | #include <linux/irq.h> | |
50 | ||
51 | ||
52 | extern struct gate_struct idt_table[256]; | |
53 | ||
54 | asmlinkage void divide_error(void); | |
55 | asmlinkage void debug(void); | |
56 | asmlinkage void nmi(void); | |
57 | asmlinkage void int3(void); | |
58 | asmlinkage void overflow(void); | |
59 | asmlinkage void bounds(void); | |
60 | asmlinkage void invalid_op(void); | |
61 | asmlinkage void device_not_available(void); | |
62 | asmlinkage void double_fault(void); | |
63 | asmlinkage void coprocessor_segment_overrun(void); | |
64 | asmlinkage void invalid_TSS(void); | |
65 | asmlinkage void segment_not_present(void); | |
66 | asmlinkage void stack_segment(void); | |
67 | asmlinkage void general_protection(void); | |
68 | asmlinkage void page_fault(void); | |
69 | asmlinkage void coprocessor_error(void); | |
70 | asmlinkage void simd_coprocessor_error(void); | |
71 | asmlinkage void reserved(void); | |
72 | asmlinkage void alignment_check(void); | |
73 | asmlinkage void machine_check(void); | |
74 | asmlinkage void spurious_interrupt_bug(void); | |
75 | asmlinkage void call_debug(void); | |
76 | ||
77 | struct notifier_block *die_chain; | |
78 | static DEFINE_SPINLOCK(die_notifier_lock); | |
79 | ||
80 | int register_die_notifier(struct notifier_block *nb) | |
81 | { | |
82 | int err = 0; | |
83 | unsigned long flags; | |
84 | spin_lock_irqsave(&die_notifier_lock, flags); | |
85 | err = notifier_chain_register(&die_chain, nb); | |
86 | spin_unlock_irqrestore(&die_notifier_lock, flags); | |
87 | return err; | |
88 | } | |
89 | ||
90 | static inline void conditional_sti(struct pt_regs *regs) | |
91 | { | |
92 | if (regs->eflags & X86_EFLAGS_IF) | |
93 | local_irq_enable(); | |
94 | } | |
95 | ||
96 | static int kstack_depth_to_print = 10; | |
97 | ||
98 | #ifdef CONFIG_KALLSYMS | |
99 | #include <linux/kallsyms.h> | |
100 | int printk_address(unsigned long address) | |
101 | { | |
102 | unsigned long offset = 0, symsize; | |
103 | const char *symname; | |
104 | char *modname; | |
105 | char *delim = ":"; | |
106 | char namebuf[128]; | |
107 | ||
108 | symname = kallsyms_lookup(address, &symsize, &offset, &modname, namebuf); | |
109 | if (!symname) | |
110 | return printk("[<%016lx>]", address); | |
111 | if (!modname) | |
112 | modname = delim = ""; | |
113 | return printk("<%016lx>{%s%s%s%s%+ld}", | |
114 | address,delim,modname,delim,symname,offset); | |
115 | } | |
116 | #else | |
117 | int printk_address(unsigned long address) | |
118 | { | |
119 | return printk("[<%016lx>]", address); | |
120 | } | |
121 | #endif | |
122 | ||
123 | unsigned long *in_exception_stack(int cpu, unsigned long stack) | |
124 | { | |
125 | int k; | |
126 | for (k = 0; k < N_EXCEPTION_STACKS; k++) { | |
127 | struct tss_struct *tss = &per_cpu(init_tss, cpu); | |
128 | unsigned long start = tss->ist[k] - EXCEPTION_STKSZ; | |
129 | ||
130 | if (stack >= start && stack < tss->ist[k]) | |
131 | return (unsigned long *)tss->ist[k]; | |
132 | } | |
133 | return NULL; | |
134 | } | |
135 | ||
136 | /* | |
137 | * x86-64 can have upto three kernel stacks: | |
138 | * process stack | |
139 | * interrupt stack | |
140 | * severe exception (double fault, nmi, stack fault) hardware stack | |
141 | * Check and process them in order. | |
142 | */ | |
143 | ||
144 | void show_trace(unsigned long *stack) | |
145 | { | |
146 | unsigned long addr; | |
147 | unsigned long *irqstack, *irqstack_end, *estack_end; | |
148 | const int cpu = safe_smp_processor_id(); | |
149 | int i; | |
150 | ||
151 | printk("\nCall Trace:"); | |
152 | i = 0; | |
153 | ||
154 | estack_end = in_exception_stack(cpu, (unsigned long)stack); | |
155 | if (estack_end) { | |
156 | while (stack < estack_end) { | |
157 | addr = *stack++; | |
158 | if (__kernel_text_address(addr)) { | |
159 | i += printk_address(addr); | |
160 | i += printk(" "); | |
161 | if (i > 50) { | |
162 | printk("\n"); | |
163 | i = 0; | |
164 | } | |
165 | } | |
166 | } | |
167 | i += printk(" <EOE> "); | |
168 | i += 7; | |
169 | stack = (unsigned long *) estack_end[-2]; | |
170 | } | |
171 | ||
172 | irqstack_end = (unsigned long *) (cpu_pda[cpu].irqstackptr); | |
173 | irqstack = (unsigned long *) (cpu_pda[cpu].irqstackptr - IRQSTACKSIZE + 64); | |
174 | ||
175 | if (stack >= irqstack && stack < irqstack_end) { | |
176 | printk("<IRQ> "); | |
177 | while (stack < irqstack_end) { | |
178 | addr = *stack++; | |
179 | /* | |
180 | * If the address is either in the text segment of the | |
181 | * kernel, or in the region which contains vmalloc'ed | |
182 | * memory, it *may* be the address of a calling | |
183 | * routine; if so, print it so that someone tracing | |
184 | * down the cause of the crash will be able to figure | |
185 | * out the call path that was taken. | |
186 | */ | |
187 | if (__kernel_text_address(addr)) { | |
188 | i += printk_address(addr); | |
189 | i += printk(" "); | |
190 | if (i > 50) { | |
191 | printk("\n "); | |
192 | i = 0; | |
193 | } | |
194 | } | |
195 | } | |
196 | stack = (unsigned long *) (irqstack_end[-1]); | |
197 | printk(" <EOI> "); | |
198 | i += 7; | |
199 | } | |
200 | ||
201 | while (((long) stack & (THREAD_SIZE-1)) != 0) { | |
202 | addr = *stack++; | |
203 | if (__kernel_text_address(addr)) { | |
204 | i += printk_address(addr); | |
205 | i += printk(" "); | |
206 | if (i > 50) { | |
207 | printk("\n "); | |
208 | i = 0; | |
209 | } | |
210 | } | |
211 | } | |
212 | printk("\n"); | |
213 | } | |
214 | ||
215 | void show_stack(struct task_struct *tsk, unsigned long * rsp) | |
216 | { | |
217 | unsigned long *stack; | |
218 | int i; | |
219 | const int cpu = safe_smp_processor_id(); | |
220 | unsigned long *irqstack_end = (unsigned long *) (cpu_pda[cpu].irqstackptr); | |
221 | unsigned long *irqstack = (unsigned long *) (cpu_pda[cpu].irqstackptr - IRQSTACKSIZE); | |
222 | ||
223 | // debugging aid: "show_stack(NULL, NULL);" prints the | |
224 | // back trace for this cpu. | |
225 | ||
226 | if (rsp == NULL) { | |
227 | if (tsk) | |
228 | rsp = (unsigned long *)tsk->thread.rsp; | |
229 | else | |
230 | rsp = (unsigned long *)&rsp; | |
231 | } | |
232 | ||
233 | stack = rsp; | |
234 | for(i=0; i < kstack_depth_to_print; i++) { | |
235 | if (stack >= irqstack && stack <= irqstack_end) { | |
236 | if (stack == irqstack_end) { | |
237 | stack = (unsigned long *) (irqstack_end[-1]); | |
238 | printk(" <EOI> "); | |
239 | } | |
240 | } else { | |
241 | if (((long) stack & (THREAD_SIZE-1)) == 0) | |
242 | break; | |
243 | } | |
244 | if (i && ((i % 4) == 0)) | |
245 | printk("\n "); | |
246 | printk("%016lx ", *stack++); | |
35faa714 | 247 | touch_nmi_watchdog(); |
1da177e4 LT |
248 | } |
249 | show_trace((unsigned long *)rsp); | |
250 | } | |
251 | ||
252 | /* | |
253 | * The architecture-independent dump_stack generator | |
254 | */ | |
255 | void dump_stack(void) | |
256 | { | |
257 | unsigned long dummy; | |
258 | show_trace(&dummy); | |
259 | } | |
260 | ||
261 | EXPORT_SYMBOL(dump_stack); | |
262 | ||
263 | void show_registers(struct pt_regs *regs) | |
264 | { | |
265 | int i; | |
266 | int in_kernel = (regs->cs & 3) == 0; | |
267 | unsigned long rsp; | |
268 | const int cpu = safe_smp_processor_id(); | |
269 | struct task_struct *cur = cpu_pda[cpu].pcurrent; | |
270 | ||
271 | rsp = regs->rsp; | |
272 | ||
273 | printk("CPU %d ", cpu); | |
274 | __show_regs(regs); | |
275 | printk("Process %s (pid: %d, threadinfo %p, task %p)\n", | |
276 | cur->comm, cur->pid, cur->thread_info, cur); | |
277 | ||
278 | /* | |
279 | * When in-kernel, we also print out the stack and code at the | |
280 | * time of the fault.. | |
281 | */ | |
282 | if (in_kernel) { | |
283 | ||
284 | printk("Stack: "); | |
285 | show_stack(NULL, (unsigned long*)rsp); | |
286 | ||
287 | printk("\nCode: "); | |
288 | if(regs->rip < PAGE_OFFSET) | |
289 | goto bad; | |
290 | ||
291 | for(i=0;i<20;i++) | |
292 | { | |
293 | unsigned char c; | |
294 | if(__get_user(c, &((unsigned char*)regs->rip)[i])) { | |
295 | bad: | |
296 | printk(" Bad RIP value."); | |
297 | break; | |
298 | } | |
299 | printk("%02x ", c); | |
300 | } | |
301 | } | |
302 | printk("\n"); | |
303 | } | |
304 | ||
305 | void handle_BUG(struct pt_regs *regs) | |
306 | { | |
307 | struct bug_frame f; | |
308 | char tmp; | |
309 | ||
310 | if (regs->cs & 3) | |
311 | return; | |
312 | if (__copy_from_user(&f, (struct bug_frame *) regs->rip, | |
313 | sizeof(struct bug_frame))) | |
314 | return; | |
315 | if ((unsigned long)f.filename < __PAGE_OFFSET || | |
316 | f.ud2[0] != 0x0f || f.ud2[1] != 0x0b) | |
317 | return; | |
318 | if (__get_user(tmp, f.filename)) | |
319 | f.filename = "unmapped filename"; | |
320 | printk("----------- [cut here ] --------- [please bite here ] ---------\n"); | |
321 | printk(KERN_ALERT "Kernel BUG at %.50s:%d\n", f.filename, f.line); | |
322 | } | |
323 | ||
324 | void out_of_line_bug(void) | |
325 | { | |
326 | BUG(); | |
327 | } | |
328 | ||
329 | static DEFINE_SPINLOCK(die_lock); | |
330 | static int die_owner = -1; | |
331 | ||
332 | void oops_begin(void) | |
333 | { | |
334 | int cpu = safe_smp_processor_id(); | |
335 | /* racy, but better than risking deadlock. */ | |
336 | local_irq_disable(); | |
337 | if (!spin_trylock(&die_lock)) { | |
338 | if (cpu == die_owner) | |
339 | /* nested oops. should stop eventually */; | |
340 | else | |
341 | spin_lock(&die_lock); | |
342 | } | |
343 | die_owner = cpu; | |
344 | console_verbose(); | |
345 | bust_spinlocks(1); | |
346 | } | |
347 | ||
348 | void oops_end(void) | |
349 | { | |
350 | die_owner = -1; | |
351 | bust_spinlocks(0); | |
352 | spin_unlock(&die_lock); | |
353 | if (panic_on_oops) | |
354 | panic("Oops"); | |
355 | } | |
356 | ||
357 | void __die(const char * str, struct pt_regs * regs, long err) | |
358 | { | |
359 | static int die_counter; | |
360 | printk(KERN_EMERG "%s: %04lx [%u] ", str, err & 0xffff,++die_counter); | |
361 | #ifdef CONFIG_PREEMPT | |
362 | printk("PREEMPT "); | |
363 | #endif | |
364 | #ifdef CONFIG_SMP | |
365 | printk("SMP "); | |
366 | #endif | |
367 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
368 | printk("DEBUG_PAGEALLOC"); | |
369 | #endif | |
370 | printk("\n"); | |
371 | notify_die(DIE_OOPS, (char *)str, regs, err, 255, SIGSEGV); | |
372 | show_registers(regs); | |
373 | /* Executive summary in case the oops scrolled away */ | |
374 | printk(KERN_ALERT "RIP "); | |
375 | printk_address(regs->rip); | |
376 | printk(" RSP <%016lx>\n", regs->rsp); | |
377 | } | |
378 | ||
379 | void die(const char * str, struct pt_regs * regs, long err) | |
380 | { | |
381 | oops_begin(); | |
382 | handle_BUG(regs); | |
383 | __die(str, regs, err); | |
384 | oops_end(); | |
385 | do_exit(SIGSEGV); | |
386 | } | |
387 | static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err) | |
388 | { | |
389 | if (!(regs->eflags & VM_MASK) && (regs->cs == __KERNEL_CS)) | |
390 | die(str, regs, err); | |
391 | } | |
392 | ||
393 | void die_nmi(char *str, struct pt_regs *regs) | |
394 | { | |
395 | oops_begin(); | |
396 | /* | |
397 | * We are in trouble anyway, lets at least try | |
398 | * to get a message out. | |
399 | */ | |
400 | printk(str, safe_smp_processor_id()); | |
401 | show_registers(regs); | |
402 | if (panic_on_timeout || panic_on_oops) | |
403 | panic("nmi watchdog"); | |
404 | printk("console shuts up ...\n"); | |
405 | oops_end(); | |
406 | do_exit(SIGSEGV); | |
407 | } | |
408 | ||
409 | static void do_trap(int trapnr, int signr, char *str, | |
410 | struct pt_regs * regs, long error_code, siginfo_t *info) | |
411 | { | |
412 | conditional_sti(regs); | |
413 | ||
414 | #ifdef CONFIG_CHECKING | |
415 | { | |
416 | unsigned long gs; | |
417 | struct x8664_pda *pda = cpu_pda + safe_smp_processor_id(); | |
418 | rdmsrl(MSR_GS_BASE, gs); | |
419 | if (gs != (unsigned long)pda) { | |
420 | wrmsrl(MSR_GS_BASE, pda); | |
421 | printk("%s: wrong gs %lx expected %p rip %lx\n", str, gs, pda, | |
422 | regs->rip); | |
423 | } | |
424 | } | |
425 | #endif | |
426 | ||
427 | if ((regs->cs & 3) != 0) { | |
428 | struct task_struct *tsk = current; | |
429 | ||
430 | if (exception_trace && unhandled_signal(tsk, signr)) | |
431 | printk(KERN_INFO | |
432 | "%s[%d] trap %s rip:%lx rsp:%lx error:%lx\n", | |
433 | tsk->comm, tsk->pid, str, | |
434 | regs->rip,regs->rsp,error_code); | |
435 | ||
436 | tsk->thread.error_code = error_code; | |
437 | tsk->thread.trap_no = trapnr; | |
438 | if (info) | |
439 | force_sig_info(signr, info, tsk); | |
440 | else | |
441 | force_sig(signr, tsk); | |
442 | return; | |
443 | } | |
444 | ||
445 | ||
446 | /* kernel trap */ | |
447 | { | |
448 | const struct exception_table_entry *fixup; | |
449 | fixup = search_exception_tables(regs->rip); | |
450 | if (fixup) { | |
451 | regs->rip = fixup->fixup; | |
452 | } else | |
453 | die(str, regs, error_code); | |
454 | return; | |
455 | } | |
456 | } | |
457 | ||
458 | #define DO_ERROR(trapnr, signr, str, name) \ | |
459 | asmlinkage void do_##name(struct pt_regs * regs, long error_code) \ | |
460 | { \ | |
461 | if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \ | |
462 | == NOTIFY_STOP) \ | |
463 | return; \ | |
464 | do_trap(trapnr, signr, str, regs, error_code, NULL); \ | |
465 | } | |
466 | ||
467 | #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \ | |
468 | asmlinkage void do_##name(struct pt_regs * regs, long error_code) \ | |
469 | { \ | |
470 | siginfo_t info; \ | |
471 | info.si_signo = signr; \ | |
472 | info.si_errno = 0; \ | |
473 | info.si_code = sicode; \ | |
474 | info.si_addr = (void __user *)siaddr; \ | |
475 | if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \ | |
476 | == NOTIFY_STOP) \ | |
477 | return; \ | |
478 | do_trap(trapnr, signr, str, regs, error_code, &info); \ | |
479 | } | |
480 | ||
481 | DO_ERROR_INFO( 0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->rip) | |
482 | DO_ERROR( 4, SIGSEGV, "overflow", overflow) | |
483 | DO_ERROR( 5, SIGSEGV, "bounds", bounds) | |
484 | DO_ERROR_INFO( 6, SIGILL, "invalid operand", invalid_op, ILL_ILLOPN, regs->rip) | |
485 | DO_ERROR( 7, SIGSEGV, "device not available", device_not_available) | |
486 | DO_ERROR( 9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun) | |
487 | DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS) | |
488 | DO_ERROR(11, SIGBUS, "segment not present", segment_not_present) | |
489 | DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0) | |
490 | DO_ERROR(18, SIGSEGV, "reserved", reserved) | |
6fefb0d1 AK |
491 | DO_ERROR(12, SIGBUS, "stack segment", stack_segment) |
492 | DO_ERROR( 8, SIGSEGV, "double fault", double_fault) | |
1da177e4 LT |
493 | |
494 | asmlinkage void do_general_protection(struct pt_regs * regs, long error_code) | |
495 | { | |
496 | conditional_sti(regs); | |
497 | ||
498 | #ifdef CONFIG_CHECKING | |
499 | { | |
500 | unsigned long gs; | |
501 | struct x8664_pda *pda = cpu_pda + safe_smp_processor_id(); | |
502 | rdmsrl(MSR_GS_BASE, gs); | |
503 | if (gs != (unsigned long)pda) { | |
504 | wrmsrl(MSR_GS_BASE, pda); | |
505 | oops_in_progress++; | |
506 | printk("general protection handler: wrong gs %lx expected %p\n", gs, pda); | |
507 | oops_in_progress--; | |
508 | } | |
509 | } | |
510 | #endif | |
511 | ||
512 | if ((regs->cs & 3)!=0) { | |
513 | struct task_struct *tsk = current; | |
514 | ||
515 | if (exception_trace && unhandled_signal(tsk, SIGSEGV)) | |
516 | printk(KERN_INFO | |
517 | "%s[%d] general protection rip:%lx rsp:%lx error:%lx\n", | |
518 | tsk->comm, tsk->pid, | |
519 | regs->rip,regs->rsp,error_code); | |
520 | ||
521 | tsk->thread.error_code = error_code; | |
522 | tsk->thread.trap_no = 13; | |
523 | force_sig(SIGSEGV, tsk); | |
524 | return; | |
525 | } | |
526 | ||
527 | /* kernel gp */ | |
528 | { | |
529 | const struct exception_table_entry *fixup; | |
530 | fixup = search_exception_tables(regs->rip); | |
531 | if (fixup) { | |
532 | regs->rip = fixup->fixup; | |
533 | return; | |
534 | } | |
535 | if (notify_die(DIE_GPF, "general protection fault", regs, | |
536 | error_code, 13, SIGSEGV) == NOTIFY_STOP) | |
537 | return; | |
538 | die("general protection fault", regs, error_code); | |
539 | } | |
540 | } | |
541 | ||
542 | static void mem_parity_error(unsigned char reason, struct pt_regs * regs) | |
543 | { | |
544 | printk("Uhhuh. NMI received. Dazed and confused, but trying to continue\n"); | |
545 | printk("You probably have a hardware problem with your RAM chips\n"); | |
546 | ||
547 | /* Clear and disable the memory parity error line. */ | |
548 | reason = (reason & 0xf) | 4; | |
549 | outb(reason, 0x61); | |
550 | } | |
551 | ||
552 | static void io_check_error(unsigned char reason, struct pt_regs * regs) | |
553 | { | |
554 | printk("NMI: IOCK error (debug interrupt?)\n"); | |
555 | show_registers(regs); | |
556 | ||
557 | /* Re-enable the IOCK line, wait for a few seconds */ | |
558 | reason = (reason & 0xf) | 8; | |
559 | outb(reason, 0x61); | |
560 | mdelay(2000); | |
561 | reason &= ~8; | |
562 | outb(reason, 0x61); | |
563 | } | |
564 | ||
565 | static void unknown_nmi_error(unsigned char reason, struct pt_regs * regs) | |
566 | { printk("Uhhuh. NMI received for unknown reason %02x.\n", reason); | |
567 | printk("Dazed and confused, but trying to continue\n"); | |
568 | printk("Do you have a strange power saving mode enabled?\n"); | |
569 | } | |
570 | ||
6fefb0d1 AK |
571 | /* Runs on IST stack. This code must keep interrupts off all the time. |
572 | Nested NMIs are prevented by the CPU. */ | |
1da177e4 LT |
573 | asmlinkage void default_do_nmi(struct pt_regs *regs) |
574 | { | |
575 | unsigned char reason = 0; | |
576 | ||
577 | /* Only the BSP gets external NMIs from the system. */ | |
578 | if (!smp_processor_id()) | |
579 | reason = get_nmi_reason(); | |
580 | ||
581 | if (!(reason & 0xc0)) { | |
582 | if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 0, SIGINT) | |
583 | == NOTIFY_STOP) | |
584 | return; | |
585 | #ifdef CONFIG_X86_LOCAL_APIC | |
586 | /* | |
587 | * Ok, so this is none of the documented NMI sources, | |
588 | * so it must be the NMI watchdog. | |
589 | */ | |
590 | if (nmi_watchdog > 0) { | |
591 | nmi_watchdog_tick(regs,reason); | |
592 | return; | |
593 | } | |
594 | #endif | |
595 | unknown_nmi_error(reason, regs); | |
596 | return; | |
597 | } | |
598 | if (notify_die(DIE_NMI, "nmi", regs, reason, 0, SIGINT) == NOTIFY_STOP) | |
599 | return; | |
600 | ||
601 | /* AK: following checks seem to be broken on modern chipsets. FIXME */ | |
602 | ||
603 | if (reason & 0x80) | |
604 | mem_parity_error(reason, regs); | |
605 | if (reason & 0x40) | |
606 | io_check_error(reason, regs); | |
607 | } | |
608 | ||
609 | asmlinkage void do_int3(struct pt_regs * regs, long error_code) | |
610 | { | |
611 | if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP) == NOTIFY_STOP) { | |
612 | return; | |
613 | } | |
614 | do_trap(3, SIGTRAP, "int3", regs, error_code, NULL); | |
615 | return; | |
616 | } | |
617 | ||
6fefb0d1 AK |
618 | /* Help handler running on IST stack to switch back to user stack |
619 | for scheduling or signal handling. The actual stack switch is done in | |
620 | entry.S */ | |
621 | asmlinkage struct pt_regs *sync_regs(struct pt_regs *eregs) | |
622 | { | |
623 | struct pt_regs *regs = eregs; | |
624 | /* Did already sync */ | |
625 | if (eregs == (struct pt_regs *)eregs->rsp) | |
626 | ; | |
627 | /* Exception from user space */ | |
628 | else if (eregs->cs & 3) | |
629 | regs = ((struct pt_regs *)current->thread.rsp0) - 1; | |
630 | /* Exception from kernel and interrupts are enabled. Move to | |
631 | kernel process stack. */ | |
632 | else if (eregs->eflags & X86_EFLAGS_IF) | |
633 | regs = (struct pt_regs *)(eregs->rsp -= sizeof(struct pt_regs)); | |
634 | if (eregs != regs) | |
635 | *regs = *eregs; | |
636 | return regs; | |
637 | } | |
638 | ||
1da177e4 | 639 | /* runs on IST stack. */ |
6fefb0d1 | 640 | asmlinkage void do_debug(struct pt_regs * regs, unsigned long error_code) |
1da177e4 | 641 | { |
1da177e4 LT |
642 | unsigned long condition; |
643 | struct task_struct *tsk = current; | |
644 | siginfo_t info; | |
645 | ||
1da177e4 LT |
646 | #ifdef CONFIG_CHECKING |
647 | { | |
648 | /* RED-PEN interaction with debugger - could destroy gs */ | |
649 | unsigned long gs; | |
650 | struct x8664_pda *pda = cpu_pda + safe_smp_processor_id(); | |
651 | rdmsrl(MSR_GS_BASE, gs); | |
652 | if (gs != (unsigned long)pda) { | |
653 | wrmsrl(MSR_GS_BASE, pda); | |
654 | printk("debug handler: wrong gs %lx expected %p\n", gs, pda); | |
655 | } | |
656 | } | |
657 | #endif | |
658 | ||
659 | asm("movq %%db6,%0" : "=r" (condition)); | |
660 | ||
661 | if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code, | |
662 | SIGTRAP) == NOTIFY_STOP) { | |
6fefb0d1 | 663 | return; |
1da177e4 LT |
664 | } |
665 | conditional_sti(regs); | |
666 | ||
667 | /* Mask out spurious debug traps due to lazy DR7 setting */ | |
668 | if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) { | |
669 | if (!tsk->thread.debugreg7) { | |
670 | goto clear_dr7; | |
671 | } | |
672 | } | |
673 | ||
674 | tsk->thread.debugreg6 = condition; | |
675 | ||
676 | /* Mask out spurious TF errors due to lazy TF clearing */ | |
677 | if ((condition & DR_STEP) && | |
678 | (notify_die(DIE_DEBUGSTEP, "debugstep", regs, condition, | |
679 | 1, SIGTRAP) != NOTIFY_STOP)) { | |
680 | /* | |
681 | * The TF error should be masked out only if the current | |
682 | * process is not traced and if the TRAP flag has been set | |
683 | * previously by a tracing process (condition detected by | |
684 | * the PT_DTRACE flag); remember that the i386 TRAP flag | |
685 | * can be modified by the process itself in user mode, | |
686 | * allowing programs to debug themselves without the ptrace() | |
687 | * interface. | |
688 | */ | |
689 | if ((regs->cs & 3) == 0) | |
690 | goto clear_TF_reenable; | |
be61bff7 AK |
691 | /* |
692 | * Was the TF flag set by a debugger? If so, clear it now, | |
693 | * so that register information is correct. | |
694 | */ | |
695 | if (tsk->ptrace & PT_DTRACE) { | |
696 | regs->eflags &= ~TF_MASK; | |
697 | tsk->ptrace &= ~PT_DTRACE; | |
698 | } | |
1da177e4 LT |
699 | } |
700 | ||
701 | /* Ok, finally something we can handle */ | |
702 | tsk->thread.trap_no = 1; | |
703 | tsk->thread.error_code = error_code; | |
704 | info.si_signo = SIGTRAP; | |
705 | info.si_errno = 0; | |
706 | info.si_code = TRAP_BRKPT; | |
707 | if ((regs->cs & 3) == 0) | |
708 | goto clear_dr7; | |
709 | ||
710 | info.si_addr = (void __user *)regs->rip; | |
711 | force_sig_info(SIGTRAP, &info, tsk); | |
712 | clear_dr7: | |
713 | asm volatile("movq %0,%%db7"::"r"(0UL)); | |
714 | notify_die(DIE_DEBUG, "debug", regs, condition, 1, SIGTRAP); | |
6fefb0d1 | 715 | return; |
1da177e4 LT |
716 | |
717 | clear_TF_reenable: | |
718 | set_tsk_thread_flag(tsk, TIF_SINGLESTEP); | |
719 | ||
720 | clear_TF: | |
721 | /* RED-PEN could cause spurious errors */ | |
722 | if (notify_die(DIE_DEBUG, "debug2", regs, condition, 1, SIGTRAP) | |
723 | != NOTIFY_STOP) | |
724 | regs->eflags &= ~TF_MASK; | |
1da177e4 LT |
725 | } |
726 | ||
727 | static int kernel_math_error(struct pt_regs *regs, char *str) | |
728 | { | |
729 | const struct exception_table_entry *fixup; | |
730 | fixup = search_exception_tables(regs->rip); | |
731 | if (fixup) { | |
732 | regs->rip = fixup->fixup; | |
733 | return 1; | |
734 | } | |
735 | notify_die(DIE_GPF, str, regs, 0, 16, SIGFPE); | |
3a848f63 | 736 | /* Illegal floating point operation in the kernel */ |
1da177e4 | 737 | die(str, regs, 0); |
1da177e4 LT |
738 | return 0; |
739 | } | |
740 | ||
741 | /* | |
742 | * Note that we play around with the 'TS' bit in an attempt to get | |
743 | * the correct behaviour even in the presence of the asynchronous | |
744 | * IRQ13 behaviour | |
745 | */ | |
746 | asmlinkage void do_coprocessor_error(struct pt_regs *regs) | |
747 | { | |
748 | void __user *rip = (void __user *)(regs->rip); | |
749 | struct task_struct * task; | |
750 | siginfo_t info; | |
751 | unsigned short cwd, swd; | |
752 | ||
753 | conditional_sti(regs); | |
754 | if ((regs->cs & 3) == 0 && | |
755 | kernel_math_error(regs, "kernel x87 math error")) | |
756 | return; | |
757 | ||
758 | /* | |
759 | * Save the info for the exception handler and clear the error. | |
760 | */ | |
761 | task = current; | |
762 | save_init_fpu(task); | |
763 | task->thread.trap_no = 16; | |
764 | task->thread.error_code = 0; | |
765 | info.si_signo = SIGFPE; | |
766 | info.si_errno = 0; | |
767 | info.si_code = __SI_FAULT; | |
768 | info.si_addr = rip; | |
769 | /* | |
770 | * (~cwd & swd) will mask out exceptions that are not set to unmasked | |
771 | * status. 0x3f is the exception bits in these regs, 0x200 is the | |
772 | * C1 reg you need in case of a stack fault, 0x040 is the stack | |
773 | * fault bit. We should only be taking one exception at a time, | |
774 | * so if this combination doesn't produce any single exception, | |
775 | * then we have a bad program that isn't synchronizing its FPU usage | |
776 | * and it will suffer the consequences since we won't be able to | |
777 | * fully reproduce the context of the exception | |
778 | */ | |
779 | cwd = get_fpu_cwd(task); | |
780 | swd = get_fpu_swd(task); | |
781 | switch (((~cwd) & swd & 0x3f) | (swd & 0x240)) { | |
782 | case 0x000: | |
783 | default: | |
784 | break; | |
785 | case 0x001: /* Invalid Op */ | |
786 | case 0x041: /* Stack Fault */ | |
787 | case 0x241: /* Stack Fault | Direction */ | |
788 | info.si_code = FPE_FLTINV; | |
789 | break; | |
790 | case 0x002: /* Denormalize */ | |
791 | case 0x010: /* Underflow */ | |
792 | info.si_code = FPE_FLTUND; | |
793 | break; | |
794 | case 0x004: /* Zero Divide */ | |
795 | info.si_code = FPE_FLTDIV; | |
796 | break; | |
797 | case 0x008: /* Overflow */ | |
798 | info.si_code = FPE_FLTOVF; | |
799 | break; | |
800 | case 0x020: /* Precision */ | |
801 | info.si_code = FPE_FLTRES; | |
802 | break; | |
803 | } | |
804 | force_sig_info(SIGFPE, &info, task); | |
805 | } | |
806 | ||
807 | asmlinkage void bad_intr(void) | |
808 | { | |
809 | printk("bad interrupt"); | |
810 | } | |
811 | ||
812 | asmlinkage void do_simd_coprocessor_error(struct pt_regs *regs) | |
813 | { | |
814 | void __user *rip = (void __user *)(regs->rip); | |
815 | struct task_struct * task; | |
816 | siginfo_t info; | |
817 | unsigned short mxcsr; | |
818 | ||
819 | conditional_sti(regs); | |
820 | if ((regs->cs & 3) == 0 && | |
3a848f63 | 821 | kernel_math_error(regs, "kernel simd math error")) |
1da177e4 LT |
822 | return; |
823 | ||
824 | /* | |
825 | * Save the info for the exception handler and clear the error. | |
826 | */ | |
827 | task = current; | |
828 | save_init_fpu(task); | |
829 | task->thread.trap_no = 19; | |
830 | task->thread.error_code = 0; | |
831 | info.si_signo = SIGFPE; | |
832 | info.si_errno = 0; | |
833 | info.si_code = __SI_FAULT; | |
834 | info.si_addr = rip; | |
835 | /* | |
836 | * The SIMD FPU exceptions are handled a little differently, as there | |
837 | * is only a single status/control register. Thus, to determine which | |
838 | * unmasked exception was caught we must mask the exception mask bits | |
839 | * at 0x1f80, and then use these to mask the exception bits at 0x3f. | |
840 | */ | |
841 | mxcsr = get_fpu_mxcsr(task); | |
842 | switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) { | |
843 | case 0x000: | |
844 | default: | |
845 | break; | |
846 | case 0x001: /* Invalid Op */ | |
847 | info.si_code = FPE_FLTINV; | |
848 | break; | |
849 | case 0x002: /* Denormalize */ | |
850 | case 0x010: /* Underflow */ | |
851 | info.si_code = FPE_FLTUND; | |
852 | break; | |
853 | case 0x004: /* Zero Divide */ | |
854 | info.si_code = FPE_FLTDIV; | |
855 | break; | |
856 | case 0x008: /* Overflow */ | |
857 | info.si_code = FPE_FLTOVF; | |
858 | break; | |
859 | case 0x020: /* Precision */ | |
860 | info.si_code = FPE_FLTRES; | |
861 | break; | |
862 | } | |
863 | force_sig_info(SIGFPE, &info, task); | |
864 | } | |
865 | ||
866 | asmlinkage void do_spurious_interrupt_bug(struct pt_regs * regs) | |
867 | { | |
868 | } | |
869 | ||
870 | asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void) | |
871 | { | |
872 | } | |
873 | ||
874 | /* | |
875 | * 'math_state_restore()' saves the current math information in the | |
876 | * old math state array, and gets the new ones from the current task | |
877 | * | |
878 | * Careful.. There are problems with IBM-designed IRQ13 behaviour. | |
879 | * Don't touch unless you *really* know how it works. | |
880 | */ | |
881 | asmlinkage void math_state_restore(void) | |
882 | { | |
883 | struct task_struct *me = current; | |
884 | clts(); /* Allow maths ops (or we recurse) */ | |
885 | ||
886 | if (!used_math()) | |
887 | init_fpu(me); | |
888 | restore_fpu_checking(&me->thread.i387.fxsave); | |
889 | me->thread_info->status |= TS_USEDFPU; | |
890 | } | |
891 | ||
892 | void do_call_debug(struct pt_regs *regs) | |
893 | { | |
894 | notify_die(DIE_CALL, "debug call", regs, 0, 255, SIGINT); | |
895 | } | |
896 | ||
897 | void __init trap_init(void) | |
898 | { | |
899 | set_intr_gate(0,÷_error); | |
900 | set_intr_gate_ist(1,&debug,DEBUG_STACK); | |
901 | set_intr_gate_ist(2,&nmi,NMI_STACK); | |
902 | set_system_gate(3,&int3); | |
903 | set_system_gate(4,&overflow); /* int4-5 can be called from all */ | |
904 | set_system_gate(5,&bounds); | |
905 | set_intr_gate(6,&invalid_op); | |
906 | set_intr_gate(7,&device_not_available); | |
907 | set_intr_gate_ist(8,&double_fault, DOUBLEFAULT_STACK); | |
908 | set_intr_gate(9,&coprocessor_segment_overrun); | |
909 | set_intr_gate(10,&invalid_TSS); | |
910 | set_intr_gate(11,&segment_not_present); | |
911 | set_intr_gate_ist(12,&stack_segment,STACKFAULT_STACK); | |
912 | set_intr_gate(13,&general_protection); | |
913 | set_intr_gate(14,&page_fault); | |
914 | set_intr_gate(15,&spurious_interrupt_bug); | |
915 | set_intr_gate(16,&coprocessor_error); | |
916 | set_intr_gate(17,&alignment_check); | |
917 | #ifdef CONFIG_X86_MCE | |
918 | set_intr_gate_ist(18,&machine_check, MCE_STACK); | |
919 | #endif | |
920 | set_intr_gate(19,&simd_coprocessor_error); | |
921 | ||
922 | #ifdef CONFIG_IA32_EMULATION | |
923 | set_system_gate(IA32_SYSCALL_VECTOR, ia32_syscall); | |
924 | #endif | |
925 | ||
926 | set_intr_gate(KDB_VECTOR, call_debug); | |
927 | ||
928 | /* | |
929 | * Should be a barrier for any external CPU state. | |
930 | */ | |
931 | cpu_init(); | |
932 | } | |
933 | ||
934 | ||
935 | /* Actual parsing is done early in setup.c. */ | |
936 | static int __init oops_dummy(char *s) | |
937 | { | |
938 | panic_on_oops = 1; | |
939 | return -1; | |
940 | } | |
941 | __setup("oops=", oops_dummy); | |
942 | ||
943 | static int __init kstack_setup(char *s) | |
944 | { | |
945 | kstack_depth_to_print = simple_strtoul(s,NULL,0); | |
946 | return 0; | |
947 | } | |
948 | __setup("kstack=", kstack_setup); | |
949 |