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