]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 | 2 | * Copyright (C) 1991, 1992 Linus Torvalds |
a8c1be9d | 3 | * Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs |
1da177e4 LT |
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 'asm.s'. | |
12 | */ | |
b5964405 IM |
13 | #include <linux/interrupt.h> |
14 | #include <linux/kallsyms.h> | |
15 | #include <linux/spinlock.h> | |
16 | #include <linux/highmem.h> | |
17 | #include <linux/kprobes.h> | |
18 | #include <linux/uaccess.h> | |
19 | #include <linux/utsname.h> | |
20 | #include <linux/kdebug.h> | |
1da177e4 | 21 | #include <linux/kernel.h> |
b5964405 IM |
22 | #include <linux/module.h> |
23 | #include <linux/ptrace.h> | |
1da177e4 | 24 | #include <linux/string.h> |
b5964405 IM |
25 | #include <linux/unwind.h> |
26 | #include <linux/delay.h> | |
1da177e4 | 27 | #include <linux/errno.h> |
b5964405 IM |
28 | #include <linux/kexec.h> |
29 | #include <linux/sched.h> | |
1da177e4 | 30 | #include <linux/timer.h> |
1da177e4 | 31 | #include <linux/init.h> |
91768d6c | 32 | #include <linux/bug.h> |
b5964405 IM |
33 | #include <linux/nmi.h> |
34 | #include <linux/mm.h> | |
1da177e4 LT |
35 | |
36 | #ifdef CONFIG_EISA | |
37 | #include <linux/ioport.h> | |
38 | #include <linux/eisa.h> | |
39 | #endif | |
40 | ||
41 | #ifdef CONFIG_MCA | |
42 | #include <linux/mca.h> | |
43 | #endif | |
44 | ||
c0d12172 DJ |
45 | #if defined(CONFIG_EDAC) |
46 | #include <linux/edac.h> | |
47 | #endif | |
48 | ||
7643e9b9 | 49 | #include <asm/processor-flags.h> |
b5964405 IM |
50 | #include <asm/arch_hooks.h> |
51 | #include <asm/stacktrace.h> | |
1da177e4 | 52 | #include <asm/processor.h> |
1da177e4 | 53 | #include <asm/debugreg.h> |
b5964405 IM |
54 | #include <asm/atomic.h> |
55 | #include <asm/system.h> | |
56 | #include <asm/unwind.h> | |
1da177e4 LT |
57 | #include <asm/desc.h> |
58 | #include <asm/i387.h> | |
59 | #include <asm/nmi.h> | |
1da177e4 | 60 | #include <asm/smp.h> |
b5964405 | 61 | #include <asm/io.h> |
6ac8d51f | 62 | #include <asm/traps.h> |
1da177e4 LT |
63 | |
64 | #include "mach_traps.h" | |
65 | ||
dbeb2be2 RR |
66 | DECLARE_BITMAP(used_vectors, NR_VECTORS); |
67 | EXPORT_SYMBOL_GPL(used_vectors); | |
68 | ||
1da177e4 LT |
69 | asmlinkage int system_call(void); |
70 | ||
1da177e4 | 71 | /* Do we ignore FPU interrupts ? */ |
b5964405 | 72 | char ignore_fpu_irq; |
1da177e4 LT |
73 | |
74 | /* | |
75 | * The IDT has to be page-aligned to simplify the Pentium | |
76 | * F0 0F bug workaround.. We have a special link segment | |
77 | * for this. | |
78 | */ | |
010d4f82 | 79 | gate_desc idt_table[256] |
6842ef0e | 80 | __attribute__((__section__(".data.idt"))) = { { { { 0, 0 } } }, }; |
1da177e4 | 81 | |
badc7652 | 82 | int panic_on_unrecovered_nmi; |
0741f4d2 | 83 | int kstack_depth_to_print = 24; |
86c41837 | 84 | static unsigned int code_bytes = 64; |
badc7652 AH |
85 | static int ignore_nmis; |
86 | static int die_counter; | |
e041c683 | 87 | |
762db434 AH |
88 | static inline void conditional_sti(struct pt_regs *regs) |
89 | { | |
90 | if (regs->flags & X86_EFLAGS_IF) | |
91 | local_irq_enable(); | |
92 | } | |
93 | ||
a5ff677c HH |
94 | void printk_address(unsigned long address, int reliable) |
95 | { | |
96 | #ifdef CONFIG_KALLSYMS | |
b5964405 IM |
97 | unsigned long offset = 0; |
98 | unsigned long symsize; | |
a5ff677c | 99 | const char *symname; |
b5964405 | 100 | char *modname; |
7b4fd4bb AH |
101 | char *delim = ":"; |
102 | char namebuf[KSYM_NAME_LEN]; | |
103 | char reliab[4] = ""; | |
a5ff677c HH |
104 | |
105 | symname = kallsyms_lookup(address, &symsize, &offset, | |
106 | &modname, namebuf); | |
107 | if (!symname) { | |
108 | printk(" [<%08lx>]\n", address); | |
109 | return; | |
110 | } | |
111 | if (!reliable) | |
112 | strcpy(reliab, "? "); | |
113 | ||
114 | if (!modname) | |
115 | modname = delim = ""; | |
116 | printk(" [<%08lx>] %s%s%s%s%s+0x%lx/0x%lx\n", | |
117 | address, reliab, delim, modname, delim, symname, offset, symsize); | |
118 | #else | |
119 | printk(" [<%08lx>]\n", address); | |
120 | #endif | |
121 | } | |
122 | ||
a8c1be9d AH |
123 | static inline int valid_stack_ptr(struct thread_info *tinfo, |
124 | void *p, unsigned int size) | |
1da177e4 | 125 | { |
7b4fd4bb AH |
126 | void *t = tinfo; |
127 | return p > t && p <= t + THREAD_SIZE - size; | |
1da177e4 LT |
128 | } |
129 | ||
36ad4885 LT |
130 | /* The form of the top of the frame on the stack */ |
131 | struct stack_frame { | |
a8c1be9d AH |
132 | struct stack_frame *next_frame; |
133 | unsigned long return_address; | |
36ad4885 LT |
134 | }; |
135 | ||
b5964405 IM |
136 | static inline unsigned long |
137 | print_context_stack(struct thread_info *tinfo, | |
a8c1be9d AH |
138 | unsigned long *stack, unsigned long bp, |
139 | const struct stacktrace_ops *ops, void *data) | |
1da177e4 | 140 | { |
65ea5b03 | 141 | struct stack_frame *frame = (struct stack_frame *)bp; |
36ad4885 | 142 | |
36ad4885 LT |
143 | while (valid_stack_ptr(tinfo, stack, sizeof(*stack))) { |
144 | unsigned long addr; | |
145 | ||
e9d4efdd AV |
146 | addr = *stack; |
147 | if (__kernel_text_address(addr)) { | |
148 | if ((unsigned long) stack == bp + 4) { | |
149 | ops->address(data, addr, 1); | |
150 | frame = frame->next_frame; | |
151 | bp = (unsigned long) frame; | |
152 | } else { | |
5bc27dc2 | 153 | ops->address(data, addr, bp == 0); |
e9d4efdd AV |
154 | } |
155 | } | |
156 | stack++; | |
1da177e4 | 157 | } |
65ea5b03 | 158 | return bp; |
1da177e4 LT |
159 | } |
160 | ||
2b14a78c | 161 | void dump_trace(struct task_struct *task, struct pt_regs *regs, |
5bc27dc2 | 162 | unsigned long *stack, unsigned long bp, |
9689ba8a | 163 | const struct stacktrace_ops *ops, void *data) |
1da177e4 | 164 | { |
1da177e4 LT |
165 | if (!task) |
166 | task = current; | |
167 | ||
a32cf397 | 168 | if (!stack) { |
2b14a78c AK |
169 | unsigned long dummy; |
170 | stack = &dummy; | |
028a690a | 171 | if (task != current) |
faca6227 | 172 | stack = (unsigned long *)task->thread.sp; |
176a2718 JB |
173 | } |
174 | ||
a32cf397 | 175 | #ifdef CONFIG_FRAME_POINTER |
65ea5b03 | 176 | if (!bp) { |
a32cf397 | 177 | if (task == current) { |
65ea5b03 | 178 | /* Grab bp right from our regs */ |
b5964405 | 179 | asm("movl %%ebp, %0" : "=r" (bp) :); |
a32cf397 | 180 | } else { |
65ea5b03 | 181 | /* bp is the last reg pushed by switch_to */ |
faca6227 | 182 | bp = *(unsigned long *) task->thread.sp; |
a32cf397 | 183 | } |
1da177e4 | 184 | } |
a32cf397 | 185 | #endif |
1da177e4 | 186 | |
a8c1be9d | 187 | for (;;) { |
1da177e4 | 188 | struct thread_info *context; |
b5964405 | 189 | |
1da177e4 LT |
190 | context = (struct thread_info *) |
191 | ((unsigned long)stack & (~(THREAD_SIZE - 1))); | |
65ea5b03 | 192 | bp = print_context_stack(context, stack, bp, ops, data); |
b5964405 IM |
193 | /* |
194 | * Should be after the line below, but somewhere | |
195 | * in early boot context comes out corrupted and we | |
196 | * can't reference it: | |
197 | */ | |
2b14a78c AK |
198 | if (ops->stack(data, "IRQ") < 0) |
199 | break; | |
b5964405 | 200 | stack = (unsigned long *)context->previous_esp; |
1da177e4 LT |
201 | if (!stack) |
202 | break; | |
a36df98a | 203 | touch_nmi_watchdog(); |
1da177e4 LT |
204 | } |
205 | } | |
2b14a78c AK |
206 | EXPORT_SYMBOL(dump_trace); |
207 | ||
208 | static void | |
209 | print_trace_warning_symbol(void *data, char *msg, unsigned long symbol) | |
210 | { | |
211 | printk(data); | |
212 | print_symbol(msg, symbol); | |
213 | printk("\n"); | |
214 | } | |
215 | ||
216 | static void print_trace_warning(void *data, char *msg) | |
217 | { | |
218 | printk("%s%s\n", (char *)data, msg); | |
219 | } | |
220 | ||
221 | static int print_trace_stack(void *data, char *name) | |
222 | { | |
223 | return 0; | |
224 | } | |
225 | ||
226 | /* | |
227 | * Print one address/symbol entries per line. | |
228 | */ | |
bc850d6b | 229 | static void print_trace_address(void *data, unsigned long addr, int reliable) |
2b14a78c AK |
230 | { |
231 | printk("%s [<%08lx>] ", (char *)data, addr); | |
bc850d6b AV |
232 | if (!reliable) |
233 | printk("? "); | |
2b14a78c | 234 | print_symbol("%s\n", addr); |
601e6255 | 235 | touch_nmi_watchdog(); |
2b14a78c AK |
236 | } |
237 | ||
9689ba8a | 238 | static const struct stacktrace_ops print_trace_ops = { |
a8c1be9d AH |
239 | .warning = print_trace_warning, |
240 | .warning_symbol = print_trace_warning_symbol, | |
241 | .stack = print_trace_stack, | |
242 | .address = print_trace_address, | |
2b14a78c AK |
243 | }; |
244 | ||
245 | static void | |
246 | show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs, | |
7dedcee3 | 247 | unsigned long *stack, unsigned long bp, char *log_lvl) |
2b14a78c | 248 | { |
5bc27dc2 | 249 | dump_trace(task, regs, stack, bp, &print_trace_ops, log_lvl); |
2b14a78c AK |
250 | printk("%s =======================\n", log_lvl); |
251 | } | |
1da177e4 | 252 | |
2b14a78c | 253 | void show_trace(struct task_struct *task, struct pt_regs *regs, |
5bc27dc2 | 254 | unsigned long *stack, unsigned long bp) |
7aa89746 | 255 | { |
5bc27dc2 | 256 | show_trace_log_lvl(task, regs, stack, bp, ""); |
7aa89746 CE |
257 | } |
258 | ||
b5964405 IM |
259 | static void |
260 | show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs, | |
261 | unsigned long *sp, unsigned long bp, char *log_lvl) | |
1da177e4 LT |
262 | { |
263 | unsigned long *stack; | |
264 | int i; | |
265 | ||
65ea5b03 | 266 | if (sp == NULL) { |
1da177e4 | 267 | if (task) |
b5964405 | 268 | sp = (unsigned long *)task->thread.sp; |
1da177e4 | 269 | else |
65ea5b03 | 270 | sp = (unsigned long *)&sp; |
1da177e4 LT |
271 | } |
272 | ||
65ea5b03 | 273 | stack = sp; |
b5964405 | 274 | for (i = 0; i < kstack_depth_to_print; i++) { |
1da177e4 LT |
275 | if (kstack_end(stack)) |
276 | break; | |
75874d5c CE |
277 | if (i && ((i % 8) == 0)) |
278 | printk("\n%s ", log_lvl); | |
1da177e4 LT |
279 | printk("%08lx ", *stack++); |
280 | } | |
75874d5c | 281 | printk("\n%sCall Trace:\n", log_lvl); |
b5964405 | 282 | |
5bc27dc2 | 283 | show_trace_log_lvl(task, regs, sp, bp, log_lvl); |
7aa89746 CE |
284 | } |
285 | ||
65ea5b03 | 286 | void show_stack(struct task_struct *task, unsigned long *sp) |
7aa89746 | 287 | { |
75874d5c | 288 | printk(" "); |
5bc27dc2 | 289 | show_stack_log_lvl(task, NULL, sp, 0, ""); |
1da177e4 LT |
290 | } |
291 | ||
292 | /* | |
293 | * The architecture-independent dump_stack generator | |
294 | */ | |
295 | void dump_stack(void) | |
296 | { | |
5bc27dc2 | 297 | unsigned long bp = 0; |
b5964405 | 298 | unsigned long stack; |
5bc27dc2 AV |
299 | |
300 | #ifdef CONFIG_FRAME_POINTER | |
301 | if (!bp) | |
302 | asm("movl %%ebp, %0" : "=r" (bp):); | |
303 | #endif | |
1da177e4 | 304 | |
57c351de AV |
305 | printk("Pid: %d, comm: %.20s %s %s %.*s\n", |
306 | current->pid, current->comm, print_tainted(), | |
307 | init_utsname()->release, | |
308 | (int)strcspn(init_utsname()->version, " "), | |
309 | init_utsname()->version); | |
b5964405 | 310 | |
5bc27dc2 | 311 | show_trace(current, NULL, &stack, bp); |
1da177e4 LT |
312 | } |
313 | ||
314 | EXPORT_SYMBOL(dump_stack); | |
315 | ||
316 | void show_registers(struct pt_regs *regs) | |
317 | { | |
318 | int i; | |
9d975ebd | 319 | |
1da177e4 | 320 | print_modules(); |
9d975ebd | 321 | __show_registers(regs, 0); |
b5964405 | 322 | |
7e04a118 | 323 | printk(KERN_EMERG "Process %.*s (pid: %d, ti=%p task=%p task.ti=%p)", |
19c5870c | 324 | TASK_COMM_LEN, current->comm, task_pid_nr(current), |
c9f4f06d | 325 | current_thread_info(), current, task_thread_info(current)); |
1da177e4 LT |
326 | /* |
327 | * When in-kernel, we also print out the stack and code at the | |
328 | * time of the fault.. | |
329 | */ | |
9d975ebd | 330 | if (!user_mode_vm(regs)) { |
86c41837 CE |
331 | unsigned int code_prologue = code_bytes * 43 / 64; |
332 | unsigned int code_len = code_bytes; | |
99325326 | 333 | unsigned char c; |
b5964405 | 334 | u8 *ip; |
1da177e4 | 335 | |
9c107805 | 336 | printk("\n" KERN_EMERG "Stack: "); |
5bc27dc2 | 337 | show_stack_log_lvl(NULL, regs, ®s->sp, 0, KERN_EMERG); |
1da177e4 | 338 | |
9c107805 | 339 | printk(KERN_EMERG "Code: "); |
1da177e4 | 340 | |
65ea5b03 | 341 | ip = (u8 *)regs->ip - code_prologue; |
a8c1be9d | 342 | if (ip < (u8 *)PAGE_OFFSET || probe_kernel_address(ip, c)) { |
99325326 | 343 | /* try starting at EIP */ |
65ea5b03 | 344 | ip = (u8 *)regs->ip; |
86c41837 | 345 | code_len = code_len - code_prologue + 1; |
99325326 | 346 | } |
65ea5b03 PA |
347 | for (i = 0; i < code_len; i++, ip++) { |
348 | if (ip < (u8 *)PAGE_OFFSET || | |
a8c1be9d | 349 | probe_kernel_address(ip, c)) { |
1da177e4 LT |
350 | printk(" Bad EIP value."); |
351 | break; | |
352 | } | |
65ea5b03 | 353 | if (ip == (u8 *)regs->ip) |
1da177e4 LT |
354 | printk("<%02x> ", c); |
355 | else | |
356 | printk("%02x ", c); | |
357 | } | |
358 | } | |
359 | printk("\n"); | |
b5964405 | 360 | } |
1da177e4 | 361 | |
65ea5b03 | 362 | int is_valid_bugaddr(unsigned long ip) |
1da177e4 LT |
363 | { |
364 | unsigned short ud2; | |
1da177e4 | 365 | |
65ea5b03 | 366 | if (ip < PAGE_OFFSET) |
91768d6c | 367 | return 0; |
65ea5b03 | 368 | if (probe_kernel_address((unsigned short *)ip, ud2)) |
91768d6c | 369 | return 0; |
1da177e4 | 370 | |
91768d6c | 371 | return ud2 == 0x0b0f; |
1da177e4 LT |
372 | } |
373 | ||
78cbac65 AH |
374 | static raw_spinlock_t die_lock = __RAW_SPIN_LOCK_UNLOCKED; |
375 | static int die_owner = -1; | |
376 | static unsigned int die_nest_count; | |
377 | ||
378 | unsigned __kprobes long oops_begin(void) | |
379 | { | |
380 | unsigned long flags; | |
381 | ||
382 | oops_enter(); | |
383 | ||
384 | if (die_owner != raw_smp_processor_id()) { | |
385 | console_verbose(); | |
386 | raw_local_irq_save(flags); | |
387 | __raw_spin_lock(&die_lock); | |
388 | die_owner = smp_processor_id(); | |
389 | die_nest_count = 0; | |
390 | bust_spinlocks(1); | |
391 | } else { | |
392 | raw_local_irq_save(flags); | |
393 | } | |
394 | die_nest_count++; | |
395 | return flags; | |
396 | } | |
397 | ||
398 | void __kprobes oops_end(unsigned long flags, struct pt_regs *regs, int signr) | |
399 | { | |
400 | bust_spinlocks(0); | |
401 | die_owner = -1; | |
402 | add_taint(TAINT_DIE); | |
403 | __raw_spin_unlock(&die_lock); | |
404 | raw_local_irq_restore(flags); | |
405 | ||
406 | if (!regs) | |
407 | return; | |
408 | ||
409 | if (kexec_should_crash(current)) | |
410 | crash_kexec(regs); | |
411 | ||
412 | if (in_interrupt()) | |
413 | panic("Fatal exception in interrupt"); | |
414 | ||
415 | if (panic_on_oops) | |
416 | panic("Fatal exception"); | |
417 | ||
418 | oops_exit(); | |
419 | do_exit(signr); | |
420 | } | |
421 | ||
b5964405 | 422 | int __kprobes __die(const char *str, struct pt_regs *regs, long err) |
a604b380 | 423 | { |
a604b380 | 424 | unsigned short ss; |
b5964405 | 425 | unsigned long sp; |
a604b380 HH |
426 | |
427 | printk(KERN_EMERG "%s: %04lx [#%d] ", str, err & 0xffff, ++die_counter); | |
428 | #ifdef CONFIG_PREEMPT | |
429 | printk("PREEMPT "); | |
430 | #endif | |
431 | #ifdef CONFIG_SMP | |
432 | printk("SMP "); | |
433 | #endif | |
434 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
435 | printk("DEBUG_PAGEALLOC"); | |
436 | #endif | |
437 | printk("\n"); | |
a604b380 | 438 | if (notify_die(DIE_OOPS, str, regs, err, |
e423f49f AH |
439 | current->thread.trap_no, SIGSEGV) == NOTIFY_STOP) |
440 | return 1; | |
441 | ||
442 | show_registers(regs); | |
443 | /* Executive summary in case the oops scrolled away */ | |
444 | sp = (unsigned long) (®s->sp); | |
445 | savesegment(ss, ss); | |
446 | if (user_mode(regs)) { | |
447 | sp = regs->sp; | |
448 | ss = regs->ss & 0xffff; | |
a604b380 | 449 | } |
e423f49f AH |
450 | printk(KERN_EMERG "EIP: [<%08lx>] ", regs->ip); |
451 | print_symbol("%s", regs->ip); | |
452 | printk(" SS:ESP %04x:%08lx\n", ss, sp); | |
453 | return 0; | |
a604b380 HH |
454 | } |
455 | ||
91768d6c | 456 | /* |
b5964405 IM |
457 | * This is gone through when something in the kernel has done something bad |
458 | * and is about to be terminated: | |
91768d6c | 459 | */ |
b5964405 | 460 | void die(const char *str, struct pt_regs *regs, long err) |
1da177e4 | 461 | { |
78cbac65 | 462 | unsigned long flags = oops_begin(); |
1da177e4 | 463 | |
78cbac65 | 464 | if (die_nest_count < 3) { |
65ea5b03 | 465 | report_bug(regs->ip, regs); |
91768d6c | 466 | |
a604b380 | 467 | if (__die(str, regs, err)) |
20c0d2d4 | 468 | regs = NULL; |
a604b380 | 469 | } else { |
9c107805 | 470 | printk(KERN_EMERG "Recursive die() failure, output suppressed\n"); |
a604b380 | 471 | } |
1da177e4 | 472 | |
78cbac65 | 473 | oops_end(flags, regs, SIGSEGV); |
1da177e4 LT |
474 | } |
475 | ||
b5964405 IM |
476 | static inline void |
477 | die_if_kernel(const char *str, struct pt_regs *regs, long err) | |
1da177e4 | 478 | { |
717b594a | 479 | if (!user_mode_vm(regs)) |
1da177e4 LT |
480 | die(str, regs, err); |
481 | } | |
482 | ||
b5964405 IM |
483 | static void __kprobes |
484 | do_trap(int trapnr, int signr, char *str, int vm86, struct pt_regs *regs, | |
485 | long error_code, siginfo_t *info) | |
1da177e4 | 486 | { |
4f339ecb | 487 | struct task_struct *tsk = current; |
4f339ecb | 488 | |
6b6891f9 | 489 | if (regs->flags & X86_VM_MASK) { |
1da177e4 LT |
490 | if (vm86) |
491 | goto vm86_trap; | |
492 | goto trap_signal; | |
493 | } | |
494 | ||
717b594a | 495 | if (!user_mode(regs)) |
1da177e4 LT |
496 | goto kernel_trap; |
497 | ||
b5964405 IM |
498 | trap_signal: |
499 | /* | |
500 | * We want error_code and trap_no set for userspace faults and | |
501 | * kernelspace faults which result in die(), but not | |
502 | * kernelspace faults which are fixed up. die() gives the | |
503 | * process no chance to handle the signal and notice the | |
504 | * kernel fault information, so that won't result in polluting | |
505 | * the information about previously queued, but not yet | |
506 | * delivered, faults. See also do_general_protection below. | |
507 | */ | |
508 | tsk->thread.error_code = error_code; | |
509 | tsk->thread.trap_no = trapnr; | |
d1895183 | 510 | |
b5964405 IM |
511 | if (info) |
512 | force_sig_info(signr, info, tsk); | |
513 | else | |
514 | force_sig(signr, tsk); | |
515 | return; | |
1da177e4 | 516 | |
b5964405 IM |
517 | kernel_trap: |
518 | if (!fixup_exception(regs)) { | |
519 | tsk->thread.error_code = error_code; | |
520 | tsk->thread.trap_no = trapnr; | |
521 | die(str, regs, error_code); | |
1da177e4 | 522 | } |
b5964405 | 523 | return; |
1da177e4 | 524 | |
b5964405 IM |
525 | vm86_trap: |
526 | if (handle_vm86_trap((struct kernel_vm86_regs *) regs, | |
527 | error_code, trapnr)) | |
528 | goto trap_signal; | |
529 | return; | |
1da177e4 LT |
530 | } |
531 | ||
61aef7d2 AH |
532 | #define DO_TRAP(trapnr, signr, str, name) \ |
533 | void do_##name(struct pt_regs *regs, long error_code) \ | |
534 | { \ | |
535 | trace_hardirqs_fixup(); \ | |
536 | if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \ | |
537 | == NOTIFY_STOP) \ | |
538 | return; \ | |
539 | do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \ | |
540 | } | |
541 | ||
542 | #define DO_TRAP_INFO(trapnr, signr, str, name, sicode, siaddr, irq) \ | |
543 | void do_##name(struct pt_regs *regs, long error_code) \ | |
544 | { \ | |
545 | siginfo_t info; \ | |
546 | if (irq) \ | |
547 | local_irq_enable(); \ | |
548 | info.si_signo = signr; \ | |
549 | info.si_errno = 0; \ | |
550 | info.si_code = sicode; \ | |
551 | info.si_addr = (void __user *)siaddr; \ | |
552 | if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \ | |
553 | == NOTIFY_STOP) \ | |
554 | return; \ | |
555 | do_trap(trapnr, signr, str, 0, regs, error_code, &info); \ | |
556 | } | |
557 | ||
558 | #define DO_VM86_TRAP(trapnr, signr, str, name) \ | |
559 | void do_##name(struct pt_regs *regs, long error_code) \ | |
560 | { \ | |
561 | if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \ | |
562 | == NOTIFY_STOP) \ | |
563 | return; \ | |
564 | do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \ | |
565 | } | |
566 | ||
567 | #define DO_VM86_TRAP_INFO(trapnr, signr, str, name, sicode, siaddr) \ | |
568 | void do_##name(struct pt_regs *regs, long error_code) \ | |
569 | { \ | |
570 | siginfo_t info; \ | |
571 | info.si_signo = signr; \ | |
572 | info.si_errno = 0; \ | |
573 | info.si_code = sicode; \ | |
574 | info.si_addr = (void __user *)siaddr; \ | |
575 | trace_hardirqs_fixup(); \ | |
576 | if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \ | |
577 | == NOTIFY_STOP) \ | |
578 | return; \ | |
579 | do_trap(trapnr, signr, str, 1, regs, error_code, &info); \ | |
580 | } | |
581 | ||
b5964405 IM |
582 | #define DO_ERROR(trapnr, signr, str, name) \ |
583 | void do_##name(struct pt_regs *regs, long error_code) \ | |
584 | { \ | |
4461145e | 585 | trace_hardirqs_fixup(); \ |
b5964405 | 586 | if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \ |
a8c1be9d | 587 | == NOTIFY_STOP) \ |
b5964405 | 588 | return; \ |
61aef7d2 | 589 | conditional_sti(regs); \ |
b5964405 | 590 | do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \ |
1da177e4 LT |
591 | } |
592 | ||
b5964405 IM |
593 | #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr, irq) \ |
594 | void do_##name(struct pt_regs *regs, long error_code) \ | |
595 | { \ | |
596 | siginfo_t info; \ | |
597 | if (irq) \ | |
598 | local_irq_enable(); \ | |
599 | info.si_signo = signr; \ | |
600 | info.si_errno = 0; \ | |
601 | info.si_code = sicode; \ | |
602 | info.si_addr = (void __user *)siaddr; \ | |
603 | if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \ | |
a8c1be9d | 604 | == NOTIFY_STOP) \ |
b5964405 | 605 | return; \ |
61aef7d2 | 606 | conditional_sti(regs); \ |
b5964405 | 607 | do_trap(trapnr, signr, str, 0, regs, error_code, &info); \ |
1da177e4 LT |
608 | } |
609 | ||
b5964405 IM |
610 | #define DO_VM86_ERROR(trapnr, signr, str, name) \ |
611 | void do_##name(struct pt_regs *regs, long error_code) \ | |
612 | { \ | |
613 | if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \ | |
a8c1be9d | 614 | == NOTIFY_STOP) \ |
b5964405 | 615 | return; \ |
61aef7d2 | 616 | conditional_sti(regs); \ |
b5964405 | 617 | do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \ |
1da177e4 LT |
618 | } |
619 | ||
b5964405 IM |
620 | #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \ |
621 | void do_##name(struct pt_regs *regs, long error_code) \ | |
622 | { \ | |
623 | siginfo_t info; \ | |
624 | info.si_signo = signr; \ | |
625 | info.si_errno = 0; \ | |
626 | info.si_code = sicode; \ | |
627 | info.si_addr = (void __user *)siaddr; \ | |
628 | trace_hardirqs_fixup(); \ | |
629 | if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \ | |
a8c1be9d | 630 | == NOTIFY_STOP) \ |
b5964405 | 631 | return; \ |
61aef7d2 | 632 | conditional_sti(regs); \ |
b5964405 | 633 | do_trap(trapnr, signr, str, 1, regs, error_code, &info); \ |
1da177e4 LT |
634 | } |
635 | ||
976382dc | 636 | DO_VM86_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip) |
8d6f9d69 | 637 | DO_VM86_ERROR(4, SIGSEGV, "overflow", overflow) |
64f644c0 | 638 | DO_VM86_ERROR(5, SIGSEGV, "bounds", bounds) |
12394cf5 | 639 | DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip, 0) |
51bc1ed6 | 640 | DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun) |
6bf77bf9 | 641 | DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS) |
36d936c7 | 642 | DO_ERROR(11, SIGBUS, "segment not present", segment_not_present) |
f5ca8187 | 643 | DO_ERROR(12, SIGBUS, "stack segment", stack_segment) |
61aef7d2 AH |
644 | DO_TRAP_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0, 0) |
645 | DO_TRAP_INFO(32, SIGILL, "iret exception", iret_error, ILL_BADSTK, 0, 1) | |
1da177e4 | 646 | |
13485ab5 AH |
647 | void __kprobes |
648 | do_general_protection(struct pt_regs *regs, long error_code) | |
1da177e4 | 649 | { |
13485ab5 | 650 | struct task_struct *tsk; |
b5964405 IM |
651 | struct thread_struct *thread; |
652 | struct tss_struct *tss; | |
653 | int cpu; | |
654 | ||
c6df0d71 AH |
655 | conditional_sti(regs); |
656 | ||
b5964405 IM |
657 | cpu = get_cpu(); |
658 | tss = &per_cpu(init_tss, cpu); | |
659 | thread = ¤t->thread; | |
1da177e4 LT |
660 | |
661 | /* | |
662 | * Perform the lazy TSS's I/O bitmap copy. If the TSS has an | |
663 | * invalid offset set (the LAZY one) and the faulting thread has | |
664 | * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS | |
665 | * and we set the offset field correctly. Then we let the CPU to | |
666 | * restart the faulting instruction. | |
667 | */ | |
a75c54f9 | 668 | if (tss->x86_tss.io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY && |
1da177e4 LT |
669 | thread->io_bitmap_ptr) { |
670 | memcpy(tss->io_bitmap, thread->io_bitmap_ptr, | |
671 | thread->io_bitmap_max); | |
672 | /* | |
673 | * If the previously set map was extending to higher ports | |
674 | * than the current one, pad extra space with 0xff (no access). | |
675 | */ | |
b5964405 | 676 | if (thread->io_bitmap_max < tss->io_bitmap_max) { |
1da177e4 LT |
677 | memset((char *) tss->io_bitmap + |
678 | thread->io_bitmap_max, 0xff, | |
679 | tss->io_bitmap_max - thread->io_bitmap_max); | |
b5964405 | 680 | } |
1da177e4 | 681 | tss->io_bitmap_max = thread->io_bitmap_max; |
a75c54f9 | 682 | tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET; |
d5cd4aad | 683 | tss->io_bitmap_owner = thread; |
1da177e4 | 684 | put_cpu(); |
b5964405 | 685 | |
1da177e4 LT |
686 | return; |
687 | } | |
688 | put_cpu(); | |
689 | ||
6b6891f9 | 690 | if (regs->flags & X86_VM_MASK) |
1da177e4 LT |
691 | goto gp_in_vm86; |
692 | ||
13485ab5 | 693 | tsk = current; |
717b594a | 694 | if (!user_mode(regs)) |
1da177e4 LT |
695 | goto gp_in_kernel; |
696 | ||
13485ab5 AH |
697 | tsk->thread.error_code = error_code; |
698 | tsk->thread.trap_no = 13; | |
b5964405 | 699 | |
13485ab5 AH |
700 | if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) && |
701 | printk_ratelimit()) { | |
abd4f750 | 702 | printk(KERN_INFO |
13485ab5 AH |
703 | "%s[%d] general protection ip:%lx sp:%lx error:%lx", |
704 | tsk->comm, task_pid_nr(tsk), | |
705 | regs->ip, regs->sp, error_code); | |
03252919 AK |
706 | print_vma_addr(" in ", regs->ip); |
707 | printk("\n"); | |
708 | } | |
abd4f750 | 709 | |
13485ab5 | 710 | force_sig(SIGSEGV, tsk); |
1da177e4 LT |
711 | return; |
712 | ||
713 | gp_in_vm86: | |
714 | local_irq_enable(); | |
715 | handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code); | |
716 | return; | |
717 | ||
718 | gp_in_kernel: | |
13485ab5 AH |
719 | if (fixup_exception(regs)) |
720 | return; | |
721 | ||
722 | tsk->thread.error_code = error_code; | |
723 | tsk->thread.trap_no = 13; | |
724 | if (notify_die(DIE_GPF, "general protection fault", regs, | |
1da177e4 | 725 | error_code, 13, SIGSEGV) == NOTIFY_STOP) |
13485ab5 AH |
726 | return; |
727 | die("general protection fault", regs, error_code); | |
1da177e4 LT |
728 | } |
729 | ||
5deb45e3 | 730 | static notrace __kprobes void |
b5964405 | 731 | mem_parity_error(unsigned char reason, struct pt_regs *regs) |
1da177e4 | 732 | { |
b5964405 IM |
733 | printk(KERN_EMERG |
734 | "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n", | |
735 | reason, smp_processor_id()); | |
736 | ||
737 | printk(KERN_EMERG | |
738 | "You have some hardware problem, likely on the PCI bus.\n"); | |
c0d12172 DJ |
739 | |
740 | #if defined(CONFIG_EDAC) | |
b5964405 | 741 | if (edac_handler_set()) { |
c0d12172 DJ |
742 | edac_atomic_assert_error(); |
743 | return; | |
744 | } | |
745 | #endif | |
746 | ||
8da5adda | 747 | if (panic_on_unrecovered_nmi) |
b5964405 | 748 | panic("NMI: Not continuing"); |
1da177e4 | 749 | |
c41c5cd3 | 750 | printk(KERN_EMERG "Dazed and confused, but trying to continue\n"); |
1da177e4 LT |
751 | |
752 | /* Clear and disable the memory parity error line. */ | |
753 | clear_mem_error(reason); | |
754 | } | |
755 | ||
5deb45e3 | 756 | static notrace __kprobes void |
b5964405 | 757 | io_check_error(unsigned char reason, struct pt_regs *regs) |
1da177e4 LT |
758 | { |
759 | unsigned long i; | |
760 | ||
9c107805 | 761 | printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n"); |
1da177e4 LT |
762 | show_registers(regs); |
763 | ||
764 | /* Re-enable the IOCK line, wait for a few seconds */ | |
765 | reason = (reason & 0xf) | 8; | |
766 | outb(reason, 0x61); | |
b5964405 | 767 | |
1da177e4 | 768 | i = 2000; |
b5964405 IM |
769 | while (--i) |
770 | udelay(1000); | |
771 | ||
1da177e4 LT |
772 | reason &= ~8; |
773 | outb(reason, 0x61); | |
774 | } | |
775 | ||
5deb45e3 | 776 | static notrace __kprobes void |
b5964405 | 777 | unknown_nmi_error(unsigned char reason, struct pt_regs *regs) |
1da177e4 | 778 | { |
d3597524 JW |
779 | if (notify_die(DIE_NMIUNKNOWN, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP) |
780 | return; | |
1da177e4 | 781 | #ifdef CONFIG_MCA |
b5964405 IM |
782 | /* |
783 | * Might actually be able to figure out what the guilty party | |
784 | * is: | |
785 | */ | |
786 | if (MCA_bus) { | |
1da177e4 LT |
787 | mca_handle_nmi(); |
788 | return; | |
789 | } | |
790 | #endif | |
b5964405 IM |
791 | printk(KERN_EMERG |
792 | "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n", | |
793 | reason, smp_processor_id()); | |
794 | ||
c41c5cd3 | 795 | printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n"); |
8da5adda | 796 | if (panic_on_unrecovered_nmi) |
b5964405 | 797 | panic("NMI: Not continuing"); |
8da5adda | 798 | |
c41c5cd3 | 799 | printk(KERN_EMERG "Dazed and confused, but trying to continue\n"); |
1da177e4 LT |
800 | } |
801 | ||
802 | static DEFINE_SPINLOCK(nmi_print_lock); | |
803 | ||
ddca03c9 | 804 | void notrace __kprobes die_nmi(char *str, struct pt_regs *regs, int do_panic) |
1da177e4 | 805 | { |
ddca03c9 | 806 | if (notify_die(DIE_NMIWATCHDOG, str, regs, 0, 2, SIGINT) == NOTIFY_STOP) |
748f2edb GA |
807 | return; |
808 | ||
1da177e4 LT |
809 | spin_lock(&nmi_print_lock); |
810 | /* | |
811 | * We are in trouble anyway, lets at least try | |
b5964405 | 812 | * to get a message out: |
1da177e4 LT |
813 | */ |
814 | bust_spinlocks(1); | |
ddca03c9 | 815 | printk(KERN_EMERG "%s", str); |
65ea5b03 PA |
816 | printk(" on CPU%d, ip %08lx, registers:\n", |
817 | smp_processor_id(), regs->ip); | |
1da177e4 | 818 | show_registers(regs); |
ddca03c9 CG |
819 | if (do_panic) |
820 | panic("Non maskable interrupt"); | |
1da177e4 LT |
821 | console_silent(); |
822 | spin_unlock(&nmi_print_lock); | |
823 | bust_spinlocks(0); | |
6e274d14 | 824 | |
b5964405 IM |
825 | /* |
826 | * If we are in kernel we are probably nested up pretty bad | |
827 | * and might aswell get out now while we still can: | |
828 | */ | |
db753bdf | 829 | if (!user_mode_vm(regs)) { |
6e274d14 AN |
830 | current->thread.trap_no = 2; |
831 | crash_kexec(regs); | |
832 | } | |
833 | ||
1da177e4 LT |
834 | do_exit(SIGSEGV); |
835 | } | |
836 | ||
5deb45e3 | 837 | static notrace __kprobes void default_do_nmi(struct pt_regs *regs) |
1da177e4 LT |
838 | { |
839 | unsigned char reason = 0; | |
abd34807 AH |
840 | int cpu; |
841 | ||
842 | cpu = smp_processor_id(); | |
1da177e4 | 843 | |
abd34807 AH |
844 | /* Only the BSP gets external NMIs from the system. */ |
845 | if (!cpu) | |
1da177e4 | 846 | reason = get_nmi_reason(); |
b5964405 | 847 | |
1da177e4 | 848 | if (!(reason & 0xc0)) { |
20c0d2d4 | 849 | if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT) |
a8c1be9d | 850 | == NOTIFY_STOP) |
1da177e4 LT |
851 | return; |
852 | #ifdef CONFIG_X86_LOCAL_APIC | |
853 | /* | |
854 | * Ok, so this is none of the documented NMI sources, | |
855 | * so it must be the NMI watchdog. | |
856 | */ | |
3adbbcce | 857 | if (nmi_watchdog_tick(regs, reason)) |
1da177e4 | 858 | return; |
abd34807 | 859 | if (!do_nmi_callback(regs, cpu)) |
3adbbcce | 860 | unknown_nmi_error(reason, regs); |
b5964405 IM |
861 | #else |
862 | unknown_nmi_error(reason, regs); | |
863 | #endif | |
2fbe7b25 | 864 | |
1da177e4 LT |
865 | return; |
866 | } | |
20c0d2d4 | 867 | if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP) |
1da177e4 | 868 | return; |
a8c1be9d AH |
869 | |
870 | /* AK: following checks seem to be broken on modern chipsets. FIXME */ | |
1da177e4 LT |
871 | if (reason & 0x80) |
872 | mem_parity_error(reason, regs); | |
873 | if (reason & 0x40) | |
874 | io_check_error(reason, regs); | |
875 | /* | |
876 | * Reassert NMI in case it became active meanwhile | |
b5964405 | 877 | * as it's edge-triggered: |
1da177e4 LT |
878 | */ |
879 | reassert_nmi(); | |
880 | } | |
881 | ||
5deb45e3 | 882 | notrace __kprobes void do_nmi(struct pt_regs *regs, long error_code) |
1da177e4 LT |
883 | { |
884 | int cpu; | |
885 | ||
886 | nmi_enter(); | |
887 | ||
888 | cpu = smp_processor_id(); | |
f3705136 | 889 | |
1da177e4 LT |
890 | ++nmi_count(cpu); |
891 | ||
8f4e956b AK |
892 | if (!ignore_nmis) |
893 | default_do_nmi(regs); | |
1da177e4 LT |
894 | |
895 | nmi_exit(); | |
896 | } | |
897 | ||
8f4e956b AK |
898 | void stop_nmi(void) |
899 | { | |
900 | acpi_nmi_disable(); | |
901 | ignore_nmis++; | |
902 | } | |
903 | ||
904 | void restart_nmi(void) | |
905 | { | |
906 | ignore_nmis--; | |
907 | acpi_nmi_enable(); | |
908 | } | |
909 | ||
75604d7f | 910 | void __kprobes do_int3(struct pt_regs *regs, long error_code) |
1da177e4 | 911 | { |
b94da1e4 | 912 | #ifdef CONFIG_KPROBES |
143a5d32 PZ |
913 | trace_hardirqs_fixup(); |
914 | ||
1da177e4 LT |
915 | if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP) |
916 | == NOTIFY_STOP) | |
48c88211 | 917 | return; |
b5964405 IM |
918 | /* |
919 | * This is an interrupt gate, because kprobes wants interrupts | |
920 | * disabled. Normal trap handlers don't. | |
921 | */ | |
762db434 | 922 | conditional_sti(regs); |
b94da1e4 AH |
923 | #else |
924 | if (notify_die(DIE_TRAP, "int3", regs, error_code, 3, SIGTRAP) | |
925 | == NOTIFY_STOP) | |
926 | return; | |
927 | #endif | |
b5964405 | 928 | |
1da177e4 | 929 | do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL); |
1da177e4 | 930 | } |
1da177e4 LT |
931 | |
932 | /* | |
933 | * Our handling of the processor debug registers is non-trivial. | |
934 | * We do not clear them on entry and exit from the kernel. Therefore | |
935 | * it is possible to get a watchpoint trap here from inside the kernel. | |
936 | * However, the code in ./ptrace.c has ensured that the user can | |
937 | * only set watchpoints on userspace addresses. Therefore the in-kernel | |
938 | * watchpoint trap can only occur in code which is reading/writing | |
939 | * from user space. Such code must not hold kernel locks (since it | |
940 | * can equally take a page fault), therefore it is safe to call | |
941 | * force_sig_info even though that claims and releases locks. | |
b5964405 | 942 | * |
1da177e4 LT |
943 | * Code in ./signal.c ensures that the debug control register |
944 | * is restored before we deliver any signal, and therefore that | |
945 | * user code runs with the correct debug control register even though | |
946 | * we clear it here. | |
947 | * | |
948 | * Being careful here means that we don't have to be as careful in a | |
949 | * lot of more complicated places (task switching can be a bit lazy | |
950 | * about restoring all the debug state, and ptrace doesn't have to | |
951 | * find every occurrence of the TF bit that could be saved away even | |
952 | * by user code) | |
953 | */ | |
b5964405 | 954 | void __kprobes do_debug(struct pt_regs *regs, long error_code) |
1da177e4 | 955 | { |
1da177e4 | 956 | struct task_struct *tsk = current; |
b5964405 | 957 | unsigned int condition; |
da654b74 | 958 | int si_code; |
1da177e4 | 959 | |
000f4a9e PZ |
960 | trace_hardirqs_fixup(); |
961 | ||
1cc6f12e | 962 | get_debugreg(condition, 6); |
1da177e4 | 963 | |
10faa81e RM |
964 | /* |
965 | * The processor cleared BTF, so don't mark that we need it set. | |
966 | */ | |
967 | clear_tsk_thread_flag(tsk, TIF_DEBUGCTLMSR); | |
968 | tsk->thread.debugctlmsr = 0; | |
969 | ||
1da177e4 | 970 | if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code, |
a8c1be9d | 971 | SIGTRAP) == NOTIFY_STOP) |
1da177e4 LT |
972 | return; |
973 | /* It's safe to allow irq's after DR6 has been saved */ | |
65ea5b03 | 974 | if (regs->flags & X86_EFLAGS_IF) |
1da177e4 LT |
975 | local_irq_enable(); |
976 | ||
977 | /* Mask out spurious debug traps due to lazy DR7 setting */ | |
978 | if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) { | |
0f534093 | 979 | if (!tsk->thread.debugreg7) |
1da177e4 LT |
980 | goto clear_dr7; |
981 | } | |
982 | ||
6b6891f9 | 983 | if (regs->flags & X86_VM_MASK) |
1da177e4 LT |
984 | goto debug_vm86; |
985 | ||
986 | /* Save debug status register where ptrace can see it */ | |
0f534093 | 987 | tsk->thread.debugreg6 = condition; |
1da177e4 LT |
988 | |
989 | /* | |
990 | * Single-stepping through TF: make sure we ignore any events in | |
991 | * kernel space (but re-enable TF when returning to user mode). | |
992 | */ | |
993 | if (condition & DR_STEP) { | |
994 | /* | |
995 | * We already checked v86 mode above, so we can | |
996 | * check for kernel mode by just checking the CPL | |
997 | * of CS. | |
998 | */ | |
717b594a | 999 | if (!user_mode(regs)) |
1da177e4 LT |
1000 | goto clear_TF_reenable; |
1001 | } | |
1002 | ||
da654b74 | 1003 | si_code = get_si_code((unsigned long)condition); |
1da177e4 | 1004 | /* Ok, finally something we can handle */ |
da654b74 | 1005 | send_sigtrap(tsk, regs, error_code, si_code); |
1da177e4 | 1006 | |
b5964405 IM |
1007 | /* |
1008 | * Disable additional traps. They'll be re-enabled when | |
1da177e4 LT |
1009 | * the signal is delivered. |
1010 | */ | |
1011 | clear_dr7: | |
1cc6f12e | 1012 | set_debugreg(0, 7); |
1da177e4 LT |
1013 | return; |
1014 | ||
1015 | debug_vm86: | |
1016 | handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1); | |
1017 | return; | |
1018 | ||
1019 | clear_TF_reenable: | |
1020 | set_tsk_thread_flag(tsk, TIF_SINGLESTEP); | |
6093015d | 1021 | regs->flags &= ~X86_EFLAGS_TF; |
1da177e4 LT |
1022 | return; |
1023 | } | |
1024 | ||
1025 | /* | |
1026 | * Note that we play around with the 'TS' bit in an attempt to get | |
1027 | * the correct behaviour even in the presence of the asynchronous | |
1028 | * IRQ13 behaviour | |
1029 | */ | |
65ea5b03 | 1030 | void math_error(void __user *ip) |
1da177e4 | 1031 | { |
b5964405 | 1032 | struct task_struct *task; |
1da177e4 | 1033 | siginfo_t info; |
7b4fd4bb | 1034 | unsigned short cwd, swd; |
1da177e4 LT |
1035 | |
1036 | /* | |
1037 | * Save the info for the exception handler and clear the error. | |
1038 | */ | |
1039 | task = current; | |
1040 | save_init_fpu(task); | |
1041 | task->thread.trap_no = 16; | |
1042 | task->thread.error_code = 0; | |
1043 | info.si_signo = SIGFPE; | |
1044 | info.si_errno = 0; | |
1045 | info.si_code = __SI_FAULT; | |
65ea5b03 | 1046 | info.si_addr = ip; |
1da177e4 LT |
1047 | /* |
1048 | * (~cwd & swd) will mask out exceptions that are not set to unmasked | |
1049 | * status. 0x3f is the exception bits in these regs, 0x200 is the | |
1050 | * C1 reg you need in case of a stack fault, 0x040 is the stack | |
1051 | * fault bit. We should only be taking one exception at a time, | |
1052 | * so if this combination doesn't produce any single exception, | |
a8c1be9d | 1053 | * then we have a bad program that isn't synchronizing its FPU usage |
1da177e4 LT |
1054 | * and it will suffer the consequences since we won't be able to |
1055 | * fully reproduce the context of the exception | |
1056 | */ | |
1057 | cwd = get_fpu_cwd(task); | |
1058 | swd = get_fpu_swd(task); | |
b1daec30 | 1059 | switch (swd & ~cwd & 0x3f) { |
b5964405 IM |
1060 | case 0x000: /* No unmasked exception */ |
1061 | return; | |
a8c1be9d | 1062 | default: /* Multiple exceptions */ |
b5964405 IM |
1063 | break; |
1064 | case 0x001: /* Invalid Op */ | |
1065 | /* | |
1066 | * swd & 0x240 == 0x040: Stack Underflow | |
1067 | * swd & 0x240 == 0x240: Stack Overflow | |
1068 | * User must clear the SF bit (0x40) if set | |
1069 | */ | |
1070 | info.si_code = FPE_FLTINV; | |
1071 | break; | |
1072 | case 0x002: /* Denormalize */ | |
1073 | case 0x010: /* Underflow */ | |
1074 | info.si_code = FPE_FLTUND; | |
1075 | break; | |
1076 | case 0x004: /* Zero Divide */ | |
1077 | info.si_code = FPE_FLTDIV; | |
1078 | break; | |
1079 | case 0x008: /* Overflow */ | |
1080 | info.si_code = FPE_FLTOVF; | |
1081 | break; | |
1082 | case 0x020: /* Precision */ | |
1083 | info.si_code = FPE_FLTRES; | |
1084 | break; | |
1da177e4 LT |
1085 | } |
1086 | force_sig_info(SIGFPE, &info, task); | |
1087 | } | |
1088 | ||
b5964405 | 1089 | void do_coprocessor_error(struct pt_regs *regs, long error_code) |
1da177e4 LT |
1090 | { |
1091 | ignore_fpu_irq = 1; | |
65ea5b03 | 1092 | math_error((void __user *)regs->ip); |
1da177e4 LT |
1093 | } |
1094 | ||
65ea5b03 | 1095 | static void simd_math_error(void __user *ip) |
1da177e4 | 1096 | { |
b5964405 | 1097 | struct task_struct *task; |
b5964405 | 1098 | siginfo_t info; |
7b4fd4bb | 1099 | unsigned short mxcsr; |
1da177e4 LT |
1100 | |
1101 | /* | |
1102 | * Save the info for the exception handler and clear the error. | |
1103 | */ | |
1104 | task = current; | |
1105 | save_init_fpu(task); | |
1106 | task->thread.trap_no = 19; | |
1107 | task->thread.error_code = 0; | |
1108 | info.si_signo = SIGFPE; | |
1109 | info.si_errno = 0; | |
1110 | info.si_code = __SI_FAULT; | |
65ea5b03 | 1111 | info.si_addr = ip; |
1da177e4 LT |
1112 | /* |
1113 | * The SIMD FPU exceptions are handled a little differently, as there | |
1114 | * is only a single status/control register. Thus, to determine which | |
1115 | * unmasked exception was caught we must mask the exception mask bits | |
1116 | * at 0x1f80, and then use these to mask the exception bits at 0x3f. | |
1117 | */ | |
1118 | mxcsr = get_fpu_mxcsr(task); | |
1119 | switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) { | |
b5964405 IM |
1120 | case 0x000: |
1121 | default: | |
1122 | break; | |
1123 | case 0x001: /* Invalid Op */ | |
1124 | info.si_code = FPE_FLTINV; | |
1125 | break; | |
1126 | case 0x002: /* Denormalize */ | |
1127 | case 0x010: /* Underflow */ | |
1128 | info.si_code = FPE_FLTUND; | |
1129 | break; | |
1130 | case 0x004: /* Zero Divide */ | |
1131 | info.si_code = FPE_FLTDIV; | |
1132 | break; | |
1133 | case 0x008: /* Overflow */ | |
1134 | info.si_code = FPE_FLTOVF; | |
1135 | break; | |
1136 | case 0x020: /* Precision */ | |
1137 | info.si_code = FPE_FLTRES; | |
1138 | break; | |
1da177e4 LT |
1139 | } |
1140 | force_sig_info(SIGFPE, &info, task); | |
1141 | } | |
1142 | ||
b5964405 | 1143 | void do_simd_coprocessor_error(struct pt_regs *regs, long error_code) |
1da177e4 LT |
1144 | { |
1145 | if (cpu_has_xmm) { | |
1146 | /* Handle SIMD FPU exceptions on PIII+ processors. */ | |
1147 | ignore_fpu_irq = 1; | |
65ea5b03 | 1148 | simd_math_error((void __user *)regs->ip); |
b5964405 IM |
1149 | return; |
1150 | } | |
1151 | /* | |
1152 | * Handle strange cache flush from user space exception | |
1153 | * in all other cases. This is undocumented behaviour. | |
1154 | */ | |
6b6891f9 | 1155 | if (regs->flags & X86_VM_MASK) { |
b5964405 IM |
1156 | handle_vm86_fault((struct kernel_vm86_regs *)regs, error_code); |
1157 | return; | |
1da177e4 | 1158 | } |
b5964405 IM |
1159 | current->thread.trap_no = 19; |
1160 | current->thread.error_code = error_code; | |
1161 | die_if_kernel("cache flush denied", regs, error_code); | |
1162 | force_sig(SIGSEGV, current); | |
1da177e4 LT |
1163 | } |
1164 | ||
b5964405 | 1165 | void do_spurious_interrupt_bug(struct pt_regs *regs, long error_code) |
1da177e4 | 1166 | { |
cf81978d | 1167 | conditional_sti(regs); |
1da177e4 LT |
1168 | #if 0 |
1169 | /* No need to warn about this any longer. */ | |
b5964405 | 1170 | printk(KERN_INFO "Ignoring P6 Local APIC Spurious Interrupt Bug...\n"); |
1da177e4 LT |
1171 | #endif |
1172 | } | |
1173 | ||
b5964405 | 1174 | unsigned long patch_espfix_desc(unsigned long uesp, unsigned long kesp) |
1da177e4 | 1175 | { |
736f12bf | 1176 | struct desc_struct *gdt = get_cpu_gdt_table(smp_processor_id()); |
be44d2aa SS |
1177 | unsigned long base = (kesp - uesp) & -THREAD_SIZE; |
1178 | unsigned long new_kesp = kesp - base; | |
1179 | unsigned long lim_pages = (new_kesp | (THREAD_SIZE - 1)) >> PAGE_SHIFT; | |
1180 | __u64 desc = *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS]; | |
b5964405 | 1181 | |
be44d2aa | 1182 | /* Set up base for espfix segment */ |
b5964405 IM |
1183 | desc &= 0x00f0ff0000000000ULL; |
1184 | desc |= ((((__u64)base) << 16) & 0x000000ffffff0000ULL) | | |
be44d2aa SS |
1185 | ((((__u64)base) << 32) & 0xff00000000000000ULL) | |
1186 | ((((__u64)lim_pages) << 32) & 0x000f000000000000ULL) | | |
1187 | (lim_pages & 0xffff); | |
1188 | *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS] = desc; | |
b5964405 | 1189 | |
be44d2aa | 1190 | return new_kesp; |
1da177e4 LT |
1191 | } |
1192 | ||
1193 | /* | |
b5964405 | 1194 | * 'math_state_restore()' saves the current math information in the |
1da177e4 LT |
1195 | * old math state array, and gets the new ones from the current task |
1196 | * | |
1197 | * Careful.. There are problems with IBM-designed IRQ13 behaviour. | |
1198 | * Don't touch unless you *really* know how it works. | |
1199 | * | |
1200 | * Must be called with kernel preemption disabled (in this case, | |
1201 | * local interrupts are disabled at the call-site in entry.S). | |
1202 | */ | |
acc20761 | 1203 | asmlinkage void math_state_restore(void) |
1da177e4 LT |
1204 | { |
1205 | struct thread_info *thread = current_thread_info(); | |
1206 | struct task_struct *tsk = thread->task; | |
1207 | ||
aa283f49 SS |
1208 | if (!tsk_used_math(tsk)) { |
1209 | local_irq_enable(); | |
1210 | /* | |
1211 | * does a slab alloc which can sleep | |
1212 | */ | |
1213 | if (init_fpu(tsk)) { | |
1214 | /* | |
1215 | * ran out of memory! | |
1216 | */ | |
1217 | do_group_exit(SIGKILL); | |
1218 | return; | |
1219 | } | |
1220 | local_irq_disable(); | |
1221 | } | |
1222 | ||
b5964405 | 1223 | clts(); /* Allow maths ops (or we recurse) */ |
1da177e4 LT |
1224 | restore_fpu(tsk); |
1225 | thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */ | |
acc20761 | 1226 | tsk->fpu_counter++; |
1da177e4 | 1227 | } |
5992b6da | 1228 | EXPORT_SYMBOL_GPL(math_state_restore); |
1da177e4 LT |
1229 | |
1230 | #ifndef CONFIG_MATH_EMULATION | |
1231 | ||
1232 | asmlinkage void math_emulate(long arg) | |
1233 | { | |
b5964405 IM |
1234 | printk(KERN_EMERG |
1235 | "math-emulation not enabled and no coprocessor found.\n"); | |
1236 | printk(KERN_EMERG "killing %s.\n", current->comm); | |
1237 | force_sig(SIGFPE, current); | |
1da177e4 LT |
1238 | schedule(); |
1239 | } | |
1240 | ||
1241 | #endif /* CONFIG_MATH_EMULATION */ | |
1242 | ||
7643e9b9 AH |
1243 | void __kprobes do_device_not_available(struct pt_regs *regs, long error) |
1244 | { | |
1245 | if (read_cr0() & X86_CR0_EM) { | |
1246 | conditional_sti(regs); | |
1247 | math_emulate(0); | |
1248 | } else { | |
1249 | math_state_restore(); /* interrupts still off */ | |
1250 | conditional_sti(regs); | |
1251 | } | |
1252 | } | |
1253 | ||
1da177e4 LT |
1254 | void __init trap_init(void) |
1255 | { | |
dbeb2be2 RR |
1256 | int i; |
1257 | ||
1da177e4 | 1258 | #ifdef CONFIG_EISA |
927222b1 | 1259 | void __iomem *p = early_ioremap(0x0FFFD9, 4); |
b5964405 IM |
1260 | |
1261 | if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24)) | |
1da177e4 | 1262 | EISA_bus = 1; |
927222b1 | 1263 | early_iounmap(p, 4); |
1da177e4 LT |
1264 | #endif |
1265 | ||
976382dc | 1266 | set_intr_gate(0, ÷_error); |
a8c1be9d AH |
1267 | set_intr_gate(1, &debug); |
1268 | set_intr_gate(2, &nmi); | |
1269 | set_system_intr_gate(3, &int3); /* int3 can be called from all */ | |
8d6f9d69 | 1270 | set_system_intr_gate(4, &overflow); /* int4 can be called from all */ |
64f644c0 | 1271 | set_intr_gate(5, &bounds); |
12394cf5 | 1272 | set_intr_gate(6, &invalid_op); |
7643e9b9 | 1273 | set_intr_gate(7, &device_not_available); |
a8c1be9d | 1274 | set_task_gate(8, GDT_ENTRY_DOUBLEFAULT_TSS); |
51bc1ed6 | 1275 | set_intr_gate(9, &coprocessor_segment_overrun); |
6bf77bf9 | 1276 | set_intr_gate(10, &invalid_TSS); |
36d936c7 | 1277 | set_intr_gate(11, &segment_not_present); |
f5ca8187 | 1278 | set_intr_gate(12, &stack_segment); |
c6df0d71 | 1279 | set_intr_gate(13, &general_protection); |
b5964405 | 1280 | set_intr_gate(14, &page_fault); |
cf81978d | 1281 | set_intr_gate(15, &spurious_interrupt_bug); |
b5964405 IM |
1282 | set_trap_gate(16, &coprocessor_error); |
1283 | set_trap_gate(17, &alignment_check); | |
1da177e4 | 1284 | #ifdef CONFIG_X86_MCE |
b5964405 | 1285 | set_trap_gate(18, &machine_check); |
1da177e4 | 1286 | #endif |
b5964405 | 1287 | set_trap_gate(19, &simd_coprocessor_error); |
1da177e4 | 1288 | |
d43c6e80 | 1289 | if (cpu_has_fxsr) { |
d43c6e80 JB |
1290 | printk(KERN_INFO "Enabling fast FPU save and restore... "); |
1291 | set_in_cr4(X86_CR4_OSFXSR); | |
1292 | printk("done.\n"); | |
1293 | } | |
1294 | if (cpu_has_xmm) { | |
b5964405 IM |
1295 | printk(KERN_INFO |
1296 | "Enabling unmasked SIMD FPU exception support... "); | |
d43c6e80 JB |
1297 | set_in_cr4(X86_CR4_OSXMMEXCPT); |
1298 | printk("done.\n"); | |
1299 | } | |
1300 | ||
b5964405 | 1301 | set_system_gate(SYSCALL_VECTOR, &system_call); |
1da177e4 | 1302 | |
b5964405 | 1303 | /* Reserve all the builtin and the syscall vector: */ |
dbeb2be2 RR |
1304 | for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++) |
1305 | set_bit(i, used_vectors); | |
b5964405 | 1306 | |
dbeb2be2 RR |
1307 | set_bit(SYSCALL_VECTOR, used_vectors); |
1308 | ||
1da177e4 | 1309 | /* |
b5964405 | 1310 | * Should be a barrier for any external CPU state: |
1da177e4 LT |
1311 | */ |
1312 | cpu_init(); | |
1313 | ||
1314 | trap_init_hook(); | |
1315 | } | |
1316 | ||
1317 | static int __init kstack_setup(char *s) | |
1318 | { | |
1319 | kstack_depth_to_print = simple_strtoul(s, NULL, 0); | |
b5964405 | 1320 | |
9b41046c | 1321 | return 1; |
1da177e4 LT |
1322 | } |
1323 | __setup("kstack=", kstack_setup); | |
86c41837 CE |
1324 | |
1325 | static int __init code_bytes_setup(char *s) | |
1326 | { | |
1327 | code_bytes = simple_strtoul(s, NULL, 0); | |
1328 | if (code_bytes > 8192) | |
1329 | code_bytes = 8192; | |
1330 | ||
1331 | return 1; | |
1332 | } | |
1333 | __setup("code_bytes=", code_bytes_setup); |