]>
Commit | Line | Data |
---|---|---|
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> | |
31 | #include <linux/nmi.h> | |
32 | #include <linux/kprobes.h> | |
33 | ||
34 | #include <asm/system.h> | |
35 | #include <asm/uaccess.h> | |
36 | #include <asm/io.h> | |
37 | #include <asm/atomic.h> | |
38 | #include <asm/debugreg.h> | |
39 | #include <asm/desc.h> | |
40 | #include <asm/i387.h> | |
41 | #include <asm/kdebug.h> | |
42 | #include <asm/processor.h> | |
43 | ||
44 | #include <asm/smp.h> | |
45 | #include <asm/pgalloc.h> | |
46 | #include <asm/pda.h> | |
47 | #include <asm/proto.h> | |
48 | #include <asm/nmi.h> | |
49 | ||
50 | extern struct gate_struct idt_table[256]; | |
51 | ||
52 | asmlinkage void divide_error(void); | |
53 | asmlinkage void debug(void); | |
54 | asmlinkage void nmi(void); | |
55 | asmlinkage void int3(void); | |
56 | asmlinkage void overflow(void); | |
57 | asmlinkage void bounds(void); | |
58 | asmlinkage void invalid_op(void); | |
59 | asmlinkage void device_not_available(void); | |
60 | asmlinkage void double_fault(void); | |
61 | asmlinkage void coprocessor_segment_overrun(void); | |
62 | asmlinkage void invalid_TSS(void); | |
63 | asmlinkage void segment_not_present(void); | |
64 | asmlinkage void stack_segment(void); | |
65 | asmlinkage void general_protection(void); | |
66 | asmlinkage void page_fault(void); | |
67 | asmlinkage void coprocessor_error(void); | |
68 | asmlinkage void simd_coprocessor_error(void); | |
69 | asmlinkage void reserved(void); | |
70 | asmlinkage void alignment_check(void); | |
71 | asmlinkage void machine_check(void); | |
72 | asmlinkage void spurious_interrupt_bug(void); | |
73 | ||
74 | struct notifier_block *die_chain; | |
75 | static DEFINE_SPINLOCK(die_notifier_lock); | |
76 | ||
77 | int register_die_notifier(struct notifier_block *nb) | |
78 | { | |
79 | int err = 0; | |
80 | unsigned long flags; | |
81 | spin_lock_irqsave(&die_notifier_lock, flags); | |
82 | err = notifier_chain_register(&die_chain, nb); | |
83 | spin_unlock_irqrestore(&die_notifier_lock, flags); | |
84 | return err; | |
85 | } | |
86 | ||
87 | static inline void conditional_sti(struct pt_regs *regs) | |
88 | { | |
89 | if (regs->eflags & X86_EFLAGS_IF) | |
90 | local_irq_enable(); | |
91 | } | |
92 | ||
93 | static inline void preempt_conditional_sti(struct pt_regs *regs) | |
94 | { | |
95 | preempt_disable(); | |
96 | if (regs->eflags & X86_EFLAGS_IF) | |
97 | local_irq_enable(); | |
98 | } | |
99 | ||
100 | static inline void preempt_conditional_cli(struct pt_regs *regs) | |
101 | { | |
102 | if (regs->eflags & X86_EFLAGS_IF) | |
103 | local_irq_disable(); | |
104 | preempt_enable_no_resched(); | |
105 | } | |
106 | ||
107 | static int kstack_depth_to_print = 10; | |
108 | ||
109 | #ifdef CONFIG_KALLSYMS | |
110 | #include <linux/kallsyms.h> | |
111 | int printk_address(unsigned long address) | |
112 | { | |
113 | unsigned long offset = 0, symsize; | |
114 | const char *symname; | |
115 | char *modname; | |
116 | char *delim = ":"; | |
117 | char namebuf[128]; | |
118 | ||
119 | symname = kallsyms_lookup(address, &symsize, &offset, &modname, namebuf); | |
120 | if (!symname) | |
121 | return printk("[<%016lx>]", address); | |
122 | if (!modname) | |
123 | modname = delim = ""; | |
124 | return printk("<%016lx>{%s%s%s%s%+ld}", | |
125 | address,delim,modname,delim,symname,offset); | |
126 | } | |
127 | #else | |
128 | int printk_address(unsigned long address) | |
129 | { | |
130 | return printk("[<%016lx>]", address); | |
131 | } | |
132 | #endif | |
133 | ||
134 | static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack, | |
135 | unsigned *usedp, const char **idp) | |
136 | { | |
137 | static char ids[][8] = { | |
138 | [DEBUG_STACK - 1] = "#DB", | |
139 | [NMI_STACK - 1] = "NMI", | |
140 | [DOUBLEFAULT_STACK - 1] = "#DF", | |
141 | [STACKFAULT_STACK - 1] = "#SS", | |
142 | [MCE_STACK - 1] = "#MC", | |
143 | #if DEBUG_STKSZ > EXCEPTION_STKSZ | |
144 | [N_EXCEPTION_STACKS ... N_EXCEPTION_STACKS + DEBUG_STKSZ / EXCEPTION_STKSZ - 2] = "#DB[?]" | |
145 | #endif | |
146 | }; | |
147 | unsigned k; | |
148 | ||
149 | for (k = 0; k < N_EXCEPTION_STACKS; k++) { | |
150 | unsigned long end; | |
151 | ||
152 | switch (k + 1) { | |
153 | #if DEBUG_STKSZ > EXCEPTION_STKSZ | |
154 | case DEBUG_STACK: | |
155 | end = cpu_pda(cpu)->debugstack + DEBUG_STKSZ; | |
156 | break; | |
157 | #endif | |
158 | default: | |
159 | end = per_cpu(init_tss, cpu).ist[k]; | |
160 | break; | |
161 | } | |
162 | if (stack >= end) | |
163 | continue; | |
164 | if (stack >= end - EXCEPTION_STKSZ) { | |
165 | if (*usedp & (1U << k)) | |
166 | break; | |
167 | *usedp |= 1U << k; | |
168 | *idp = ids[k]; | |
169 | return (unsigned long *)end; | |
170 | } | |
171 | #if DEBUG_STKSZ > EXCEPTION_STKSZ | |
172 | if (k == DEBUG_STACK - 1 && stack >= end - DEBUG_STKSZ) { | |
173 | unsigned j = N_EXCEPTION_STACKS - 1; | |
174 | ||
175 | do { | |
176 | ++j; | |
177 | end -= EXCEPTION_STKSZ; | |
178 | ids[j][4] = '1' + (j - N_EXCEPTION_STACKS); | |
179 | } while (stack < end - EXCEPTION_STKSZ); | |
180 | if (*usedp & (1U << j)) | |
181 | break; | |
182 | *usedp |= 1U << j; | |
183 | *idp = ids[j]; | |
184 | return (unsigned long *)end; | |
185 | } | |
186 | #endif | |
187 | } | |
188 | return NULL; | |
189 | } | |
190 | ||
191 | /* | |
192 | * x86-64 can have upto three kernel stacks: | |
193 | * process stack | |
194 | * interrupt stack | |
195 | * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack | |
196 | */ | |
197 | ||
198 | void show_trace(unsigned long *stack) | |
199 | { | |
200 | const unsigned cpu = safe_smp_processor_id(); | |
201 | unsigned long *irqstack_end = (unsigned long *)cpu_pda(cpu)->irqstackptr; | |
202 | int i; | |
203 | unsigned used = 0; | |
204 | ||
205 | printk("\nCall Trace:"); | |
206 | ||
207 | #define HANDLE_STACK(cond) \ | |
208 | do while (cond) { \ | |
209 | unsigned long addr = *stack++; \ | |
210 | if (kernel_text_address(addr)) { \ | |
211 | if (i > 50) { \ | |
212 | printk("\n "); \ | |
213 | i = 0; \ | |
214 | } \ | |
215 | else \ | |
216 | i += printk(" "); \ | |
217 | /* \ | |
218 | * If the address is either in the text segment of the \ | |
219 | * kernel, or in the region which contains vmalloc'ed \ | |
220 | * memory, it *may* be the address of a calling \ | |
221 | * routine; if so, print it so that someone tracing \ | |
222 | * down the cause of the crash will be able to figure \ | |
223 | * out the call path that was taken. \ | |
224 | */ \ | |
225 | i += printk_address(addr); \ | |
226 | } \ | |
227 | } while (0) | |
228 | ||
229 | for(i = 11; ; ) { | |
230 | const char *id; | |
231 | unsigned long *estack_end; | |
232 | estack_end = in_exception_stack(cpu, (unsigned long)stack, | |
233 | &used, &id); | |
234 | ||
235 | if (estack_end) { | |
236 | i += printk(" <%s>", id); | |
237 | HANDLE_STACK (stack < estack_end); | |
238 | i += printk(" <EOE>"); | |
239 | stack = (unsigned long *) estack_end[-2]; | |
240 | continue; | |
241 | } | |
242 | if (irqstack_end) { | |
243 | unsigned long *irqstack; | |
244 | irqstack = irqstack_end - | |
245 | (IRQSTACKSIZE - 64) / sizeof(*irqstack); | |
246 | ||
247 | if (stack >= irqstack && stack < irqstack_end) { | |
248 | i += printk(" <IRQ>"); | |
249 | HANDLE_STACK (stack < irqstack_end); | |
250 | stack = (unsigned long *) (irqstack_end[-1]); | |
251 | irqstack_end = NULL; | |
252 | i += printk(" <EOI>"); | |
253 | continue; | |
254 | } | |
255 | } | |
256 | break; | |
257 | } | |
258 | ||
259 | HANDLE_STACK (((long) stack & (THREAD_SIZE-1)) != 0); | |
260 | #undef HANDLE_STACK | |
261 | printk("\n"); | |
262 | } | |
263 | ||
264 | void show_stack(struct task_struct *tsk, unsigned long * rsp) | |
265 | { | |
266 | unsigned long *stack; | |
267 | int i; | |
268 | const int cpu = safe_smp_processor_id(); | |
269 | unsigned long *irqstack_end = (unsigned long *) (cpu_pda(cpu)->irqstackptr); | |
270 | unsigned long *irqstack = (unsigned long *) (cpu_pda(cpu)->irqstackptr - IRQSTACKSIZE); | |
271 | ||
272 | // debugging aid: "show_stack(NULL, NULL);" prints the | |
273 | // back trace for this cpu. | |
274 | ||
275 | if (rsp == NULL) { | |
276 | if (tsk) | |
277 | rsp = (unsigned long *)tsk->thread.rsp; | |
278 | else | |
279 | rsp = (unsigned long *)&rsp; | |
280 | } | |
281 | ||
282 | stack = rsp; | |
283 | for(i=0; i < kstack_depth_to_print; i++) { | |
284 | if (stack >= irqstack && stack <= irqstack_end) { | |
285 | if (stack == irqstack_end) { | |
286 | stack = (unsigned long *) (irqstack_end[-1]); | |
287 | printk(" <EOI> "); | |
288 | } | |
289 | } else { | |
290 | if (((long) stack & (THREAD_SIZE-1)) == 0) | |
291 | break; | |
292 | } | |
293 | if (i && ((i % 4) == 0)) | |
294 | printk("\n "); | |
295 | printk("%016lx ", *stack++); | |
296 | touch_nmi_watchdog(); | |
297 | } | |
298 | show_trace((unsigned long *)rsp); | |
299 | } | |
300 | ||
301 | /* | |
302 | * The architecture-independent dump_stack generator | |
303 | */ | |
304 | void dump_stack(void) | |
305 | { | |
306 | unsigned long dummy; | |
307 | show_trace(&dummy); | |
308 | } | |
309 | ||
310 | EXPORT_SYMBOL(dump_stack); | |
311 | ||
312 | void show_registers(struct pt_regs *regs) | |
313 | { | |
314 | int i; | |
315 | int in_kernel = !user_mode(regs); | |
316 | unsigned long rsp; | |
317 | const int cpu = safe_smp_processor_id(); | |
318 | struct task_struct *cur = cpu_pda(cpu)->pcurrent; | |
319 | ||
320 | rsp = regs->rsp; | |
321 | ||
322 | printk("CPU %d ", cpu); | |
323 | __show_regs(regs); | |
324 | printk("Process %s (pid: %d, threadinfo %p, task %p)\n", | |
325 | cur->comm, cur->pid, task_thread_info(cur), cur); | |
326 | ||
327 | /* | |
328 | * When in-kernel, we also print out the stack and code at the | |
329 | * time of the fault.. | |
330 | */ | |
331 | if (in_kernel) { | |
332 | ||
333 | printk("Stack: "); | |
334 | show_stack(NULL, (unsigned long*)rsp); | |
335 | ||
336 | printk("\nCode: "); | |
337 | if(regs->rip < PAGE_OFFSET) | |
338 | goto bad; | |
339 | ||
340 | for(i=0;i<20;i++) | |
341 | { | |
342 | unsigned char c; | |
343 | if(__get_user(c, &((unsigned char*)regs->rip)[i])) { | |
344 | bad: | |
345 | printk(" Bad RIP value."); | |
346 | break; | |
347 | } | |
348 | printk("%02x ", c); | |
349 | } | |
350 | } | |
351 | printk("\n"); | |
352 | } | |
353 | ||
354 | void handle_BUG(struct pt_regs *regs) | |
355 | { | |
356 | struct bug_frame f; | |
357 | long len; | |
358 | const char *prefix = ""; | |
359 | ||
360 | if (user_mode(regs)) | |
361 | return; | |
362 | if (__copy_from_user(&f, (const void __user *) regs->rip, | |
363 | sizeof(struct bug_frame))) | |
364 | return; | |
365 | if (f.filename >= 0 || | |
366 | f.ud2[0] != 0x0f || f.ud2[1] != 0x0b) | |
367 | return; | |
368 | len = __strnlen_user((char *)(long)f.filename, PATH_MAX) - 1; | |
369 | if (len < 0 || len >= PATH_MAX) | |
370 | f.filename = (int)(long)"unmapped filename"; | |
371 | else if (len > 50) { | |
372 | f.filename += len - 50; | |
373 | prefix = "..."; | |
374 | } | |
375 | printk("----------- [cut here ] --------- [please bite here ] ---------\n"); | |
376 | printk(KERN_ALERT "Kernel BUG at %s%.50s:%d\n", prefix, (char *)(long)f.filename, f.line); | |
377 | } | |
378 | ||
379 | #ifdef CONFIG_BUG | |
380 | void out_of_line_bug(void) | |
381 | { | |
382 | BUG(); | |
383 | } | |
384 | #endif | |
385 | ||
386 | static DEFINE_SPINLOCK(die_lock); | |
387 | static int die_owner = -1; | |
388 | ||
389 | unsigned __kprobes long oops_begin(void) | |
390 | { | |
391 | int cpu = safe_smp_processor_id(); | |
392 | unsigned long flags; | |
393 | ||
394 | /* racy, but better than risking deadlock. */ | |
395 | local_irq_save(flags); | |
396 | if (!spin_trylock(&die_lock)) { | |
397 | if (cpu == die_owner) | |
398 | /* nested oops. should stop eventually */; | |
399 | else | |
400 | spin_lock(&die_lock); | |
401 | } | |
402 | die_owner = cpu; | |
403 | console_verbose(); | |
404 | bust_spinlocks(1); | |
405 | return flags; | |
406 | } | |
407 | ||
408 | void __kprobes oops_end(unsigned long flags) | |
409 | { | |
410 | die_owner = -1; | |
411 | bust_spinlocks(0); | |
412 | spin_unlock_irqrestore(&die_lock, flags); | |
413 | if (panic_on_oops) | |
414 | panic("Oops"); | |
415 | } | |
416 | ||
417 | void __kprobes __die(const char * str, struct pt_regs * regs, long err) | |
418 | { | |
419 | static int die_counter; | |
420 | printk(KERN_EMERG "%s: %04lx [%u] ", str, err & 0xffff,++die_counter); | |
421 | #ifdef CONFIG_PREEMPT | |
422 | printk("PREEMPT "); | |
423 | #endif | |
424 | #ifdef CONFIG_SMP | |
425 | printk("SMP "); | |
426 | #endif | |
427 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
428 | printk("DEBUG_PAGEALLOC"); | |
429 | #endif | |
430 | printk("\n"); | |
431 | notify_die(DIE_OOPS, str, regs, err, current->thread.trap_no, SIGSEGV); | |
432 | show_registers(regs); | |
433 | /* Executive summary in case the oops scrolled away */ | |
434 | printk(KERN_ALERT "RIP "); | |
435 | printk_address(regs->rip); | |
436 | printk(" RSP <%016lx>\n", regs->rsp); | |
437 | } | |
438 | ||
439 | void die(const char * str, struct pt_regs * regs, long err) | |
440 | { | |
441 | unsigned long flags = oops_begin(); | |
442 | ||
443 | handle_BUG(regs); | |
444 | __die(str, regs, err); | |
445 | oops_end(flags); | |
446 | do_exit(SIGSEGV); | |
447 | } | |
448 | ||
449 | void __kprobes die_nmi(char *str, struct pt_regs *regs) | |
450 | { | |
451 | unsigned long flags = oops_begin(); | |
452 | ||
453 | /* | |
454 | * We are in trouble anyway, lets at least try | |
455 | * to get a message out. | |
456 | */ | |
457 | printk(str, safe_smp_processor_id()); | |
458 | show_registers(regs); | |
459 | if (panic_on_timeout || panic_on_oops) | |
460 | panic("nmi watchdog"); | |
461 | printk("console shuts up ...\n"); | |
462 | oops_end(flags); | |
463 | do_exit(SIGSEGV); | |
464 | } | |
465 | ||
466 | static void __kprobes do_trap(int trapnr, int signr, char *str, | |
467 | struct pt_regs * regs, long error_code, | |
468 | siginfo_t *info) | |
469 | { | |
470 | struct task_struct *tsk = current; | |
471 | ||
472 | conditional_sti(regs); | |
473 | ||
474 | tsk->thread.error_code = error_code; | |
475 | tsk->thread.trap_no = trapnr; | |
476 | ||
477 | if (user_mode(regs)) { | |
478 | if (exception_trace && unhandled_signal(tsk, signr)) | |
479 | printk(KERN_INFO | |
480 | "%s[%d] trap %s rip:%lx rsp:%lx error:%lx\n", | |
481 | tsk->comm, tsk->pid, str, | |
482 | regs->rip,regs->rsp,error_code); | |
483 | ||
484 | if (info) | |
485 | force_sig_info(signr, info, tsk); | |
486 | else | |
487 | force_sig(signr, tsk); | |
488 | return; | |
489 | } | |
490 | ||
491 | ||
492 | /* kernel trap */ | |
493 | { | |
494 | const struct exception_table_entry *fixup; | |
495 | fixup = search_exception_tables(regs->rip); | |
496 | if (fixup) { | |
497 | regs->rip = fixup->fixup; | |
498 | } else | |
499 | die(str, regs, error_code); | |
500 | return; | |
501 | } | |
502 | } | |
503 | ||
504 | #define DO_ERROR(trapnr, signr, str, name) \ | |
505 | asmlinkage void do_##name(struct pt_regs * regs, long error_code) \ | |
506 | { \ | |
507 | if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \ | |
508 | == NOTIFY_STOP) \ | |
509 | return; \ | |
510 | do_trap(trapnr, signr, str, regs, error_code, NULL); \ | |
511 | } | |
512 | ||
513 | #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \ | |
514 | asmlinkage void do_##name(struct pt_regs * regs, long error_code) \ | |
515 | { \ | |
516 | siginfo_t info; \ | |
517 | info.si_signo = signr; \ | |
518 | info.si_errno = 0; \ | |
519 | info.si_code = sicode; \ | |
520 | info.si_addr = (void __user *)siaddr; \ | |
521 | if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \ | |
522 | == NOTIFY_STOP) \ | |
523 | return; \ | |
524 | do_trap(trapnr, signr, str, regs, error_code, &info); \ | |
525 | } | |
526 | ||
527 | DO_ERROR_INFO( 0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->rip) | |
528 | DO_ERROR( 4, SIGSEGV, "overflow", overflow) | |
529 | DO_ERROR( 5, SIGSEGV, "bounds", bounds) | |
530 | DO_ERROR_INFO( 6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->rip) | |
531 | DO_ERROR( 7, SIGSEGV, "device not available", device_not_available) | |
532 | DO_ERROR( 9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun) | |
533 | DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS) | |
534 | DO_ERROR(11, SIGBUS, "segment not present", segment_not_present) | |
535 | DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0) | |
536 | DO_ERROR(18, SIGSEGV, "reserved", reserved) | |
537 | DO_ERROR(12, SIGBUS, "stack segment", stack_segment) | |
538 | ||
539 | asmlinkage void do_double_fault(struct pt_regs * regs, long error_code) | |
540 | { | |
541 | static const char str[] = "double fault"; | |
542 | struct task_struct *tsk = current; | |
543 | ||
544 | /* Return not checked because double check cannot be ignored */ | |
545 | notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV); | |
546 | ||
547 | tsk->thread.error_code = error_code; | |
548 | tsk->thread.trap_no = 8; | |
549 | ||
550 | /* This is always a kernel trap and never fixable (and thus must | |
551 | never return). */ | |
552 | for (;;) | |
553 | die(str, regs, error_code); | |
554 | } | |
555 | ||
556 | asmlinkage void __kprobes do_general_protection(struct pt_regs * regs, | |
557 | long error_code) | |
558 | { | |
559 | struct task_struct *tsk = current; | |
560 | ||
561 | conditional_sti(regs); | |
562 | ||
563 | tsk->thread.error_code = error_code; | |
564 | tsk->thread.trap_no = 13; | |
565 | ||
566 | if (user_mode(regs)) { | |
567 | if (exception_trace && unhandled_signal(tsk, SIGSEGV)) | |
568 | printk(KERN_INFO | |
569 | "%s[%d] general protection rip:%lx rsp:%lx error:%lx\n", | |
570 | tsk->comm, tsk->pid, | |
571 | regs->rip,regs->rsp,error_code); | |
572 | ||
573 | force_sig(SIGSEGV, tsk); | |
574 | return; | |
575 | } | |
576 | ||
577 | /* kernel gp */ | |
578 | { | |
579 | const struct exception_table_entry *fixup; | |
580 | fixup = search_exception_tables(regs->rip); | |
581 | if (fixup) { | |
582 | regs->rip = fixup->fixup; | |
583 | return; | |
584 | } | |
585 | if (notify_die(DIE_GPF, "general protection fault", regs, | |
586 | error_code, 13, SIGSEGV) == NOTIFY_STOP) | |
587 | return; | |
588 | die("general protection fault", regs, error_code); | |
589 | } | |
590 | } | |
591 | ||
592 | static __kprobes void | |
593 | mem_parity_error(unsigned char reason, struct pt_regs * regs) | |
594 | { | |
595 | printk("Uhhuh. NMI received. Dazed and confused, but trying to continue\n"); | |
596 | printk("You probably have a hardware problem with your RAM chips\n"); | |
597 | ||
598 | /* Clear and disable the memory parity error line. */ | |
599 | reason = (reason & 0xf) | 4; | |
600 | outb(reason, 0x61); | |
601 | } | |
602 | ||
603 | static __kprobes void | |
604 | io_check_error(unsigned char reason, struct pt_regs * regs) | |
605 | { | |
606 | printk("NMI: IOCK error (debug interrupt?)\n"); | |
607 | show_registers(regs); | |
608 | ||
609 | /* Re-enable the IOCK line, wait for a few seconds */ | |
610 | reason = (reason & 0xf) | 8; | |
611 | outb(reason, 0x61); | |
612 | mdelay(2000); | |
613 | reason &= ~8; | |
614 | outb(reason, 0x61); | |
615 | } | |
616 | ||
617 | static __kprobes void | |
618 | unknown_nmi_error(unsigned char reason, struct pt_regs * regs) | |
619 | { printk("Uhhuh. NMI received for unknown reason %02x.\n", reason); | |
620 | printk("Dazed and confused, but trying to continue\n"); | |
621 | printk("Do you have a strange power saving mode enabled?\n"); | |
622 | } | |
623 | ||
624 | /* Runs on IST stack. This code must keep interrupts off all the time. | |
625 | Nested NMIs are prevented by the CPU. */ | |
626 | asmlinkage __kprobes void default_do_nmi(struct pt_regs *regs) | |
627 | { | |
628 | unsigned char reason = 0; | |
629 | int cpu; | |
630 | ||
631 | cpu = smp_processor_id(); | |
632 | ||
633 | /* Only the BSP gets external NMIs from the system. */ | |
634 | if (!cpu) | |
635 | reason = get_nmi_reason(); | |
636 | ||
637 | if (!(reason & 0xc0)) { | |
638 | if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT) | |
639 | == NOTIFY_STOP) | |
640 | return; | |
641 | #ifdef CONFIG_X86_LOCAL_APIC | |
642 | /* | |
643 | * Ok, so this is none of the documented NMI sources, | |
644 | * so it must be the NMI watchdog. | |
645 | */ | |
646 | if (nmi_watchdog > 0) { | |
647 | nmi_watchdog_tick(regs,reason); | |
648 | return; | |
649 | } | |
650 | #endif | |
651 | unknown_nmi_error(reason, regs); | |
652 | return; | |
653 | } | |
654 | if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP) | |
655 | return; | |
656 | ||
657 | /* AK: following checks seem to be broken on modern chipsets. FIXME */ | |
658 | ||
659 | if (reason & 0x80) | |
660 | mem_parity_error(reason, regs); | |
661 | if (reason & 0x40) | |
662 | io_check_error(reason, regs); | |
663 | } | |
664 | ||
665 | /* runs on IST stack. */ | |
666 | asmlinkage void __kprobes do_int3(struct pt_regs * regs, long error_code) | |
667 | { | |
668 | if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP) == NOTIFY_STOP) { | |
669 | return; | |
670 | } | |
671 | do_trap(3, SIGTRAP, "int3", regs, error_code, NULL); | |
672 | return; | |
673 | } | |
674 | ||
675 | /* Help handler running on IST stack to switch back to user stack | |
676 | for scheduling or signal handling. The actual stack switch is done in | |
677 | entry.S */ | |
678 | asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs) | |
679 | { | |
680 | struct pt_regs *regs = eregs; | |
681 | /* Did already sync */ | |
682 | if (eregs == (struct pt_regs *)eregs->rsp) | |
683 | ; | |
684 | /* Exception from user space */ | |
685 | else if (user_mode(eregs)) | |
686 | regs = task_pt_regs(current); | |
687 | /* Exception from kernel and interrupts are enabled. Move to | |
688 | kernel process stack. */ | |
689 | else if (eregs->eflags & X86_EFLAGS_IF) | |
690 | regs = (struct pt_regs *)(eregs->rsp -= sizeof(struct pt_regs)); | |
691 | if (eregs != regs) | |
692 | *regs = *eregs; | |
693 | return regs; | |
694 | } | |
695 | ||
696 | /* runs on IST stack. */ | |
697 | asmlinkage void __kprobes do_debug(struct pt_regs * regs, | |
698 | unsigned long error_code) | |
699 | { | |
700 | unsigned long condition; | |
701 | struct task_struct *tsk = current; | |
702 | siginfo_t info; | |
703 | ||
704 | get_debugreg(condition, 6); | |
705 | ||
706 | if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code, | |
707 | SIGTRAP) == NOTIFY_STOP) | |
708 | return; | |
709 | ||
710 | preempt_conditional_sti(regs); | |
711 | ||
712 | /* Mask out spurious debug traps due to lazy DR7 setting */ | |
713 | if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) { | |
714 | if (!tsk->thread.debugreg7) { | |
715 | goto clear_dr7; | |
716 | } | |
717 | } | |
718 | ||
719 | tsk->thread.debugreg6 = condition; | |
720 | ||
721 | /* Mask out spurious TF errors due to lazy TF clearing */ | |
722 | if (condition & DR_STEP) { | |
723 | /* | |
724 | * The TF error should be masked out only if the current | |
725 | * process is not traced and if the TRAP flag has been set | |
726 | * previously by a tracing process (condition detected by | |
727 | * the PT_DTRACE flag); remember that the i386 TRAP flag | |
728 | * can be modified by the process itself in user mode, | |
729 | * allowing programs to debug themselves without the ptrace() | |
730 | * interface. | |
731 | */ | |
732 | if (!user_mode(regs)) | |
733 | goto clear_TF_reenable; | |
734 | /* | |
735 | * Was the TF flag set by a debugger? If so, clear it now, | |
736 | * so that register information is correct. | |
737 | */ | |
738 | if (tsk->ptrace & PT_DTRACE) { | |
739 | regs->eflags &= ~TF_MASK; | |
740 | tsk->ptrace &= ~PT_DTRACE; | |
741 | } | |
742 | } | |
743 | ||
744 | /* Ok, finally something we can handle */ | |
745 | tsk->thread.trap_no = 1; | |
746 | tsk->thread.error_code = error_code; | |
747 | info.si_signo = SIGTRAP; | |
748 | info.si_errno = 0; | |
749 | info.si_code = TRAP_BRKPT; | |
750 | info.si_addr = user_mode(regs) ? (void __user *)regs->rip : NULL; | |
751 | force_sig_info(SIGTRAP, &info, tsk); | |
752 | ||
753 | clear_dr7: | |
754 | set_debugreg(0UL, 7); | |
755 | preempt_conditional_cli(regs); | |
756 | return; | |
757 | ||
758 | clear_TF_reenable: | |
759 | set_tsk_thread_flag(tsk, TIF_SINGLESTEP); | |
760 | regs->eflags &= ~TF_MASK; | |
761 | preempt_conditional_cli(regs); | |
762 | } | |
763 | ||
764 | static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr) | |
765 | { | |
766 | const struct exception_table_entry *fixup; | |
767 | fixup = search_exception_tables(regs->rip); | |
768 | if (fixup) { | |
769 | regs->rip = fixup->fixup; | |
770 | return 1; | |
771 | } | |
772 | notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE); | |
773 | /* Illegal floating point operation in the kernel */ | |
774 | current->thread.trap_no = trapnr; | |
775 | die(str, regs, 0); | |
776 | return 0; | |
777 | } | |
778 | ||
779 | /* | |
780 | * Note that we play around with the 'TS' bit in an attempt to get | |
781 | * the correct behaviour even in the presence of the asynchronous | |
782 | * IRQ13 behaviour | |
783 | */ | |
784 | asmlinkage void do_coprocessor_error(struct pt_regs *regs) | |
785 | { | |
786 | void __user *rip = (void __user *)(regs->rip); | |
787 | struct task_struct * task; | |
788 | siginfo_t info; | |
789 | unsigned short cwd, swd; | |
790 | ||
791 | conditional_sti(regs); | |
792 | if (!user_mode(regs) && | |
793 | kernel_math_error(regs, "kernel x87 math error", 16)) | |
794 | return; | |
795 | ||
796 | /* | |
797 | * Save the info for the exception handler and clear the error. | |
798 | */ | |
799 | task = current; | |
800 | save_init_fpu(task); | |
801 | task->thread.trap_no = 16; | |
802 | task->thread.error_code = 0; | |
803 | info.si_signo = SIGFPE; | |
804 | info.si_errno = 0; | |
805 | info.si_code = __SI_FAULT; | |
806 | info.si_addr = rip; | |
807 | /* | |
808 | * (~cwd & swd) will mask out exceptions that are not set to unmasked | |
809 | * status. 0x3f is the exception bits in these regs, 0x200 is the | |
810 | * C1 reg you need in case of a stack fault, 0x040 is the stack | |
811 | * fault bit. We should only be taking one exception at a time, | |
812 | * so if this combination doesn't produce any single exception, | |
813 | * then we have a bad program that isn't synchronizing its FPU usage | |
814 | * and it will suffer the consequences since we won't be able to | |
815 | * fully reproduce the context of the exception | |
816 | */ | |
817 | cwd = get_fpu_cwd(task); | |
818 | swd = get_fpu_swd(task); | |
819 | switch (swd & ~cwd & 0x3f) { | |
820 | case 0x000: | |
821 | default: | |
822 | break; | |
823 | case 0x001: /* Invalid Op */ | |
824 | /* | |
825 | * swd & 0x240 == 0x040: Stack Underflow | |
826 | * swd & 0x240 == 0x240: Stack Overflow | |
827 | * User must clear the SF bit (0x40) if set | |
828 | */ | |
829 | info.si_code = FPE_FLTINV; | |
830 | break; | |
831 | case 0x002: /* Denormalize */ | |
832 | case 0x010: /* Underflow */ | |
833 | info.si_code = FPE_FLTUND; | |
834 | break; | |
835 | case 0x004: /* Zero Divide */ | |
836 | info.si_code = FPE_FLTDIV; | |
837 | break; | |
838 | case 0x008: /* Overflow */ | |
839 | info.si_code = FPE_FLTOVF; | |
840 | break; | |
841 | case 0x020: /* Precision */ | |
842 | info.si_code = FPE_FLTRES; | |
843 | break; | |
844 | } | |
845 | force_sig_info(SIGFPE, &info, task); | |
846 | } | |
847 | ||
848 | asmlinkage void bad_intr(void) | |
849 | { | |
850 | printk("bad interrupt"); | |
851 | } | |
852 | ||
853 | asmlinkage void do_simd_coprocessor_error(struct pt_regs *regs) | |
854 | { | |
855 | void __user *rip = (void __user *)(regs->rip); | |
856 | struct task_struct * task; | |
857 | siginfo_t info; | |
858 | unsigned short mxcsr; | |
859 | ||
860 | conditional_sti(regs); | |
861 | if (!user_mode(regs) && | |
862 | kernel_math_error(regs, "kernel simd math error", 19)) | |
863 | return; | |
864 | ||
865 | /* | |
866 | * Save the info for the exception handler and clear the error. | |
867 | */ | |
868 | task = current; | |
869 | save_init_fpu(task); | |
870 | task->thread.trap_no = 19; | |
871 | task->thread.error_code = 0; | |
872 | info.si_signo = SIGFPE; | |
873 | info.si_errno = 0; | |
874 | info.si_code = __SI_FAULT; | |
875 | info.si_addr = rip; | |
876 | /* | |
877 | * The SIMD FPU exceptions are handled a little differently, as there | |
878 | * is only a single status/control register. Thus, to determine which | |
879 | * unmasked exception was caught we must mask the exception mask bits | |
880 | * at 0x1f80, and then use these to mask the exception bits at 0x3f. | |
881 | */ | |
882 | mxcsr = get_fpu_mxcsr(task); | |
883 | switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) { | |
884 | case 0x000: | |
885 | default: | |
886 | break; | |
887 | case 0x001: /* Invalid Op */ | |
888 | info.si_code = FPE_FLTINV; | |
889 | break; | |
890 | case 0x002: /* Denormalize */ | |
891 | case 0x010: /* Underflow */ | |
892 | info.si_code = FPE_FLTUND; | |
893 | break; | |
894 | case 0x004: /* Zero Divide */ | |
895 | info.si_code = FPE_FLTDIV; | |
896 | break; | |
897 | case 0x008: /* Overflow */ | |
898 | info.si_code = FPE_FLTOVF; | |
899 | break; | |
900 | case 0x020: /* Precision */ | |
901 | info.si_code = FPE_FLTRES; | |
902 | break; | |
903 | } | |
904 | force_sig_info(SIGFPE, &info, task); | |
905 | } | |
906 | ||
907 | asmlinkage void do_spurious_interrupt_bug(struct pt_regs * regs) | |
908 | { | |
909 | } | |
910 | ||
911 | asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void) | |
912 | { | |
913 | } | |
914 | ||
915 | asmlinkage void __attribute__((weak)) mce_threshold_interrupt(void) | |
916 | { | |
917 | } | |
918 | ||
919 | /* | |
920 | * 'math_state_restore()' saves the current math information in the | |
921 | * old math state array, and gets the new ones from the current task | |
922 | * | |
923 | * Careful.. There are problems with IBM-designed IRQ13 behaviour. | |
924 | * Don't touch unless you *really* know how it works. | |
925 | */ | |
926 | asmlinkage void math_state_restore(void) | |
927 | { | |
928 | struct task_struct *me = current; | |
929 | clts(); /* Allow maths ops (or we recurse) */ | |
930 | ||
931 | if (!used_math()) | |
932 | init_fpu(me); | |
933 | restore_fpu_checking(&me->thread.i387.fxsave); | |
934 | task_thread_info(me)->status |= TS_USEDFPU; | |
935 | } | |
936 | ||
937 | void __init trap_init(void) | |
938 | { | |
939 | set_intr_gate(0,÷_error); | |
940 | set_intr_gate_ist(1,&debug,DEBUG_STACK); | |
941 | set_intr_gate_ist(2,&nmi,NMI_STACK); | |
942 | set_system_gate_ist(3,&int3,DEBUG_STACK); /* int3 can be called from all */ | |
943 | set_system_gate(4,&overflow); /* int4 can be called from all */ | |
944 | set_intr_gate(5,&bounds); | |
945 | set_intr_gate(6,&invalid_op); | |
946 | set_intr_gate(7,&device_not_available); | |
947 | set_intr_gate_ist(8,&double_fault, DOUBLEFAULT_STACK); | |
948 | set_intr_gate(9,&coprocessor_segment_overrun); | |
949 | set_intr_gate(10,&invalid_TSS); | |
950 | set_intr_gate(11,&segment_not_present); | |
951 | set_intr_gate_ist(12,&stack_segment,STACKFAULT_STACK); | |
952 | set_intr_gate(13,&general_protection); | |
953 | set_intr_gate(14,&page_fault); | |
954 | set_intr_gate(15,&spurious_interrupt_bug); | |
955 | set_intr_gate(16,&coprocessor_error); | |
956 | set_intr_gate(17,&alignment_check); | |
957 | #ifdef CONFIG_X86_MCE | |
958 | set_intr_gate_ist(18,&machine_check, MCE_STACK); | |
959 | #endif | |
960 | set_intr_gate(19,&simd_coprocessor_error); | |
961 | ||
962 | #ifdef CONFIG_IA32_EMULATION | |
963 | set_system_gate(IA32_SYSCALL_VECTOR, ia32_syscall); | |
964 | #endif | |
965 | ||
966 | /* | |
967 | * Should be a barrier for any external CPU state. | |
968 | */ | |
969 | cpu_init(); | |
970 | } | |
971 | ||
972 | ||
973 | /* Actual parsing is done early in setup.c. */ | |
974 | static int __init oops_dummy(char *s) | |
975 | { | |
976 | panic_on_oops = 1; | |
977 | return -1; | |
978 | } | |
979 | __setup("oops=", oops_dummy); | |
980 | ||
981 | static int __init kstack_setup(char *s) | |
982 | { | |
983 | kstack_depth_to_print = simple_strtoul(s,NULL,0); | |
984 | return 0; | |
985 | } | |
986 | __setup("kstack=", kstack_setup); | |
987 |