]>
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> |
1da177e4 LT |
61 | |
62 | #include "mach_traps.h" | |
63 | ||
29cbc78b AK |
64 | int panic_on_unrecovered_nmi; |
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 LT |
81 | |
82 | asmlinkage void divide_error(void); | |
83 | asmlinkage void debug(void); | |
84 | asmlinkage void nmi(void); | |
85 | asmlinkage void int3(void); | |
86 | asmlinkage void overflow(void); | |
87 | asmlinkage void bounds(void); | |
88 | asmlinkage void invalid_op(void); | |
89 | asmlinkage void device_not_available(void); | |
90 | asmlinkage void coprocessor_segment_overrun(void); | |
91 | asmlinkage void invalid_TSS(void); | |
92 | asmlinkage void segment_not_present(void); | |
93 | asmlinkage void stack_segment(void); | |
94 | asmlinkage void general_protection(void); | |
95 | asmlinkage void page_fault(void); | |
96 | asmlinkage void coprocessor_error(void); | |
97 | asmlinkage void simd_coprocessor_error(void); | |
98 | asmlinkage void alignment_check(void); | |
99 | asmlinkage void spurious_interrupt_bug(void); | |
100 | asmlinkage void machine_check(void); | |
101 | ||
0741f4d2 | 102 | int kstack_depth_to_print = 24; |
86c41837 | 103 | static unsigned int code_bytes = 64; |
e041c683 | 104 | |
a5ff677c HH |
105 | void printk_address(unsigned long address, int reliable) |
106 | { | |
107 | #ifdef CONFIG_KALLSYMS | |
b5964405 IM |
108 | char namebuf[KSYM_NAME_LEN]; |
109 | unsigned long offset = 0; | |
110 | unsigned long symsize; | |
a5ff677c | 111 | const char *symname; |
a5ff677c | 112 | char reliab[4] = ""; |
b5964405 IM |
113 | char *delim = ":"; |
114 | char *modname; | |
a5ff677c HH |
115 | |
116 | symname = kallsyms_lookup(address, &symsize, &offset, | |
117 | &modname, namebuf); | |
118 | if (!symname) { | |
119 | printk(" [<%08lx>]\n", address); | |
120 | return; | |
121 | } | |
122 | if (!reliable) | |
123 | strcpy(reliab, "? "); | |
124 | ||
125 | if (!modname) | |
126 | modname = delim = ""; | |
127 | printk(" [<%08lx>] %s%s%s%s%s+0x%lx/0x%lx\n", | |
128 | address, reliab, delim, modname, delim, symname, offset, symsize); | |
129 | #else | |
130 | printk(" [<%08lx>]\n", address); | |
131 | #endif | |
132 | } | |
133 | ||
a8c1be9d AH |
134 | static inline int valid_stack_ptr(struct thread_info *tinfo, |
135 | void *p, unsigned int size) | |
1da177e4 LT |
136 | { |
137 | return p > (void *)tinfo && | |
36ad4885 | 138 | p <= (void *)tinfo + THREAD_SIZE - size; |
1da177e4 LT |
139 | } |
140 | ||
36ad4885 LT |
141 | /* The form of the top of the frame on the stack */ |
142 | struct stack_frame { | |
a8c1be9d AH |
143 | struct stack_frame *next_frame; |
144 | unsigned long return_address; | |
36ad4885 LT |
145 | }; |
146 | ||
b5964405 IM |
147 | static inline unsigned long |
148 | print_context_stack(struct thread_info *tinfo, | |
a8c1be9d AH |
149 | unsigned long *stack, unsigned long bp, |
150 | const struct stacktrace_ops *ops, void *data) | |
1da177e4 | 151 | { |
65ea5b03 | 152 | struct stack_frame *frame = (struct stack_frame *)bp; |
36ad4885 | 153 | |
36ad4885 LT |
154 | while (valid_stack_ptr(tinfo, stack, sizeof(*stack))) { |
155 | unsigned long addr; | |
156 | ||
e9d4efdd AV |
157 | addr = *stack; |
158 | if (__kernel_text_address(addr)) { | |
159 | if ((unsigned long) stack == bp + 4) { | |
160 | ops->address(data, addr, 1); | |
161 | frame = frame->next_frame; | |
162 | bp = (unsigned long) frame; | |
163 | } else { | |
5bc27dc2 | 164 | ops->address(data, addr, bp == 0); |
e9d4efdd AV |
165 | } |
166 | } | |
167 | stack++; | |
1da177e4 | 168 | } |
65ea5b03 | 169 | return bp; |
1da177e4 LT |
170 | } |
171 | ||
2b14a78c | 172 | void dump_trace(struct task_struct *task, struct pt_regs *regs, |
5bc27dc2 | 173 | unsigned long *stack, unsigned long bp, |
9689ba8a | 174 | const struct stacktrace_ops *ops, void *data) |
1da177e4 | 175 | { |
1da177e4 LT |
176 | if (!task) |
177 | task = current; | |
178 | ||
a32cf397 | 179 | if (!stack) { |
2b14a78c AK |
180 | unsigned long dummy; |
181 | stack = &dummy; | |
028a690a | 182 | if (task != current) |
faca6227 | 183 | stack = (unsigned long *)task->thread.sp; |
176a2718 JB |
184 | } |
185 | ||
a32cf397 | 186 | #ifdef CONFIG_FRAME_POINTER |
65ea5b03 | 187 | if (!bp) { |
a32cf397 | 188 | if (task == current) { |
65ea5b03 | 189 | /* Grab bp right from our regs */ |
b5964405 | 190 | asm("movl %%ebp, %0" : "=r" (bp) :); |
a32cf397 | 191 | } else { |
65ea5b03 | 192 | /* bp is the last reg pushed by switch_to */ |
faca6227 | 193 | bp = *(unsigned long *) task->thread.sp; |
a32cf397 | 194 | } |
1da177e4 | 195 | } |
a32cf397 | 196 | #endif |
1da177e4 | 197 | |
a8c1be9d | 198 | for (;;) { |
1da177e4 | 199 | struct thread_info *context; |
b5964405 | 200 | |
1da177e4 LT |
201 | context = (struct thread_info *) |
202 | ((unsigned long)stack & (~(THREAD_SIZE - 1))); | |
65ea5b03 | 203 | bp = print_context_stack(context, stack, bp, ops, data); |
b5964405 IM |
204 | /* |
205 | * Should be after the line below, but somewhere | |
206 | * in early boot context comes out corrupted and we | |
207 | * can't reference it: | |
208 | */ | |
2b14a78c AK |
209 | if (ops->stack(data, "IRQ") < 0) |
210 | break; | |
b5964405 | 211 | stack = (unsigned long *)context->previous_esp; |
1da177e4 LT |
212 | if (!stack) |
213 | break; | |
a36df98a | 214 | touch_nmi_watchdog(); |
1da177e4 LT |
215 | } |
216 | } | |
2b14a78c AK |
217 | EXPORT_SYMBOL(dump_trace); |
218 | ||
219 | static void | |
220 | print_trace_warning_symbol(void *data, char *msg, unsigned long symbol) | |
221 | { | |
222 | printk(data); | |
223 | print_symbol(msg, symbol); | |
224 | printk("\n"); | |
225 | } | |
226 | ||
227 | static void print_trace_warning(void *data, char *msg) | |
228 | { | |
229 | printk("%s%s\n", (char *)data, msg); | |
230 | } | |
231 | ||
232 | static int print_trace_stack(void *data, char *name) | |
233 | { | |
234 | return 0; | |
235 | } | |
236 | ||
237 | /* | |
238 | * Print one address/symbol entries per line. | |
239 | */ | |
bc850d6b | 240 | static void print_trace_address(void *data, unsigned long addr, int reliable) |
2b14a78c AK |
241 | { |
242 | printk("%s [<%08lx>] ", (char *)data, addr); | |
bc850d6b AV |
243 | if (!reliable) |
244 | printk("? "); | |
2b14a78c | 245 | print_symbol("%s\n", addr); |
601e6255 | 246 | touch_nmi_watchdog(); |
2b14a78c AK |
247 | } |
248 | ||
9689ba8a | 249 | static const struct stacktrace_ops print_trace_ops = { |
a8c1be9d AH |
250 | .warning = print_trace_warning, |
251 | .warning_symbol = print_trace_warning_symbol, | |
252 | .stack = print_trace_stack, | |
253 | .address = print_trace_address, | |
2b14a78c AK |
254 | }; |
255 | ||
256 | static void | |
257 | show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs, | |
b5964405 | 258 | unsigned long *stack, unsigned long bp, char *log_lvl) |
2b14a78c | 259 | { |
5bc27dc2 | 260 | dump_trace(task, regs, stack, bp, &print_trace_ops, log_lvl); |
2b14a78c AK |
261 | printk("%s =======================\n", log_lvl); |
262 | } | |
1da177e4 | 263 | |
2b14a78c | 264 | void show_trace(struct task_struct *task, struct pt_regs *regs, |
5bc27dc2 | 265 | unsigned long *stack, unsigned long bp) |
7aa89746 | 266 | { |
5bc27dc2 | 267 | show_trace_log_lvl(task, regs, stack, bp, ""); |
7aa89746 CE |
268 | } |
269 | ||
b5964405 IM |
270 | static void |
271 | show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs, | |
272 | unsigned long *sp, unsigned long bp, char *log_lvl) | |
1da177e4 LT |
273 | { |
274 | unsigned long *stack; | |
275 | int i; | |
276 | ||
65ea5b03 | 277 | if (sp == NULL) { |
1da177e4 | 278 | if (task) |
b5964405 | 279 | sp = (unsigned long *)task->thread.sp; |
1da177e4 | 280 | else |
65ea5b03 | 281 | sp = (unsigned long *)&sp; |
1da177e4 LT |
282 | } |
283 | ||
65ea5b03 | 284 | stack = sp; |
b5964405 | 285 | for (i = 0; i < kstack_depth_to_print; i++) { |
1da177e4 LT |
286 | if (kstack_end(stack)) |
287 | break; | |
75874d5c CE |
288 | if (i && ((i % 8) == 0)) |
289 | printk("\n%s ", log_lvl); | |
1da177e4 LT |
290 | printk("%08lx ", *stack++); |
291 | } | |
75874d5c | 292 | printk("\n%sCall Trace:\n", log_lvl); |
b5964405 | 293 | |
5bc27dc2 | 294 | show_trace_log_lvl(task, regs, sp, bp, log_lvl); |
7aa89746 CE |
295 | } |
296 | ||
65ea5b03 | 297 | void show_stack(struct task_struct *task, unsigned long *sp) |
7aa89746 | 298 | { |
75874d5c | 299 | printk(" "); |
5bc27dc2 | 300 | show_stack_log_lvl(task, NULL, sp, 0, ""); |
1da177e4 LT |
301 | } |
302 | ||
303 | /* | |
304 | * The architecture-independent dump_stack generator | |
305 | */ | |
306 | void dump_stack(void) | |
307 | { | |
5bc27dc2 | 308 | unsigned long bp = 0; |
b5964405 | 309 | unsigned long stack; |
5bc27dc2 AV |
310 | |
311 | #ifdef CONFIG_FRAME_POINTER | |
312 | if (!bp) | |
313 | asm("movl %%ebp, %0" : "=r" (bp):); | |
314 | #endif | |
1da177e4 | 315 | |
57c351de AV |
316 | printk("Pid: %d, comm: %.20s %s %s %.*s\n", |
317 | current->pid, current->comm, print_tainted(), | |
318 | init_utsname()->release, | |
319 | (int)strcspn(init_utsname()->version, " "), | |
320 | init_utsname()->version); | |
b5964405 | 321 | |
5bc27dc2 | 322 | show_trace(current, NULL, &stack, bp); |
1da177e4 LT |
323 | } |
324 | ||
325 | EXPORT_SYMBOL(dump_stack); | |
326 | ||
327 | void show_registers(struct pt_regs *regs) | |
328 | { | |
329 | int i; | |
9d975ebd | 330 | |
1da177e4 | 331 | print_modules(); |
9d975ebd | 332 | __show_registers(regs, 0); |
b5964405 | 333 | |
7e04a118 | 334 | printk(KERN_EMERG "Process %.*s (pid: %d, ti=%p task=%p task.ti=%p)", |
19c5870c | 335 | TASK_COMM_LEN, current->comm, task_pid_nr(current), |
c9f4f06d | 336 | current_thread_info(), current, task_thread_info(current)); |
1da177e4 LT |
337 | /* |
338 | * When in-kernel, we also print out the stack and code at the | |
339 | * time of the fault.. | |
340 | */ | |
9d975ebd | 341 | if (!user_mode_vm(regs)) { |
86c41837 CE |
342 | unsigned int code_prologue = code_bytes * 43 / 64; |
343 | unsigned int code_len = code_bytes; | |
99325326 | 344 | unsigned char c; |
b5964405 | 345 | u8 *ip; |
1da177e4 | 346 | |
9c107805 | 347 | printk("\n" KERN_EMERG "Stack: "); |
5bc27dc2 | 348 | show_stack_log_lvl(NULL, regs, ®s->sp, 0, KERN_EMERG); |
1da177e4 | 349 | |
9c107805 | 350 | printk(KERN_EMERG "Code: "); |
1da177e4 | 351 | |
65ea5b03 | 352 | ip = (u8 *)regs->ip - code_prologue; |
a8c1be9d | 353 | if (ip < (u8 *)PAGE_OFFSET || probe_kernel_address(ip, c)) { |
99325326 | 354 | /* try starting at EIP */ |
65ea5b03 | 355 | ip = (u8 *)regs->ip; |
86c41837 | 356 | code_len = code_len - code_prologue + 1; |
99325326 | 357 | } |
65ea5b03 PA |
358 | for (i = 0; i < code_len; i++, ip++) { |
359 | if (ip < (u8 *)PAGE_OFFSET || | |
a8c1be9d | 360 | probe_kernel_address(ip, c)) { |
1da177e4 LT |
361 | printk(" Bad EIP value."); |
362 | break; | |
363 | } | |
65ea5b03 | 364 | if (ip == (u8 *)regs->ip) |
1da177e4 LT |
365 | printk("<%02x> ", c); |
366 | else | |
367 | printk("%02x ", c); | |
368 | } | |
369 | } | |
370 | printk("\n"); | |
b5964405 | 371 | } |
1da177e4 | 372 | |
65ea5b03 | 373 | int is_valid_bugaddr(unsigned long ip) |
1da177e4 LT |
374 | { |
375 | unsigned short ud2; | |
1da177e4 | 376 | |
65ea5b03 | 377 | if (ip < PAGE_OFFSET) |
91768d6c | 378 | return 0; |
65ea5b03 | 379 | if (probe_kernel_address((unsigned short *)ip, ud2)) |
91768d6c | 380 | return 0; |
1da177e4 | 381 | |
91768d6c | 382 | return ud2 == 0x0b0f; |
1da177e4 LT |
383 | } |
384 | ||
a604b380 HH |
385 | static int die_counter; |
386 | ||
b5964405 | 387 | int __kprobes __die(const char *str, struct pt_regs *regs, long err) |
a604b380 | 388 | { |
a604b380 | 389 | unsigned short ss; |
b5964405 | 390 | unsigned long sp; |
a604b380 HH |
391 | |
392 | printk(KERN_EMERG "%s: %04lx [#%d] ", str, err & 0xffff, ++die_counter); | |
393 | #ifdef CONFIG_PREEMPT | |
394 | printk("PREEMPT "); | |
395 | #endif | |
396 | #ifdef CONFIG_SMP | |
397 | printk("SMP "); | |
398 | #endif | |
399 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
400 | printk("DEBUG_PAGEALLOC"); | |
401 | #endif | |
402 | printk("\n"); | |
403 | ||
404 | if (notify_die(DIE_OOPS, str, regs, err, | |
b5964405 IM |
405 | current->thread.trap_no, SIGSEGV) != NOTIFY_STOP) { |
406 | ||
a604b380 HH |
407 | show_registers(regs); |
408 | /* Executive summary in case the oops scrolled away */ | |
409 | sp = (unsigned long) (®s->sp); | |
410 | savesegment(ss, ss); | |
411 | if (user_mode(regs)) { | |
412 | sp = regs->sp; | |
413 | ss = regs->ss & 0xffff; | |
414 | } | |
415 | printk(KERN_EMERG "EIP: [<%08lx>] ", regs->ip); | |
416 | print_symbol("%s", regs->ip); | |
417 | printk(" SS:ESP %04x:%08lx\n", ss, sp); | |
b5964405 | 418 | |
a604b380 | 419 | return 0; |
a604b380 | 420 | } |
b5964405 IM |
421 | |
422 | return 1; | |
a604b380 HH |
423 | } |
424 | ||
91768d6c | 425 | /* |
b5964405 IM |
426 | * This is gone through when something in the kernel has done something bad |
427 | * and is about to be terminated: | |
91768d6c | 428 | */ |
b5964405 | 429 | void die(const char *str, struct pt_regs *regs, long err) |
1da177e4 LT |
430 | { |
431 | static struct { | |
39743c9e | 432 | raw_spinlock_t lock; |
1da177e4 LT |
433 | u32 lock_owner; |
434 | int lock_owner_depth; | |
435 | } die = { | |
39743c9e | 436 | .lock = __RAW_SPIN_LOCK_UNLOCKED, |
1da177e4 LT |
437 | .lock_owner = -1, |
438 | .lock_owner_depth = 0 | |
439 | }; | |
e43d674f | 440 | unsigned long flags; |
1da177e4 | 441 | |
dd287796 AM |
442 | oops_enter(); |
443 | ||
39c715b7 | 444 | if (die.lock_owner != raw_smp_processor_id()) { |
1da177e4 | 445 | console_verbose(); |
c0a698b7 | 446 | raw_local_irq_save(flags); |
39743c9e | 447 | __raw_spin_lock(&die.lock); |
1da177e4 LT |
448 | die.lock_owner = smp_processor_id(); |
449 | die.lock_owner_depth = 0; | |
450 | bust_spinlocks(1); | |
b5964405 | 451 | } else { |
c0a698b7 | 452 | raw_local_irq_save(flags); |
b5964405 | 453 | } |
1da177e4 LT |
454 | |
455 | if (++die.lock_owner_depth < 3) { | |
65ea5b03 | 456 | report_bug(regs->ip, regs); |
91768d6c | 457 | |
a604b380 | 458 | if (__die(str, regs, err)) |
20c0d2d4 | 459 | regs = NULL; |
a604b380 | 460 | } else { |
9c107805 | 461 | printk(KERN_EMERG "Recursive die() failure, output suppressed\n"); |
a604b380 | 462 | } |
1da177e4 LT |
463 | |
464 | bust_spinlocks(0); | |
465 | die.lock_owner = -1; | |
bcdcd8e7 | 466 | add_taint(TAINT_DIE); |
39743c9e AK |
467 | __raw_spin_unlock(&die.lock); |
468 | raw_local_irq_restore(flags); | |
6e274d14 | 469 | |
20c0d2d4 JB |
470 | if (!regs) |
471 | return; | |
472 | ||
6e274d14 AN |
473 | if (kexec_should_crash(current)) |
474 | crash_kexec(regs); | |
475 | ||
1da177e4 LT |
476 | if (in_interrupt()) |
477 | panic("Fatal exception in interrupt"); | |
478 | ||
cea6a4ba | 479 | if (panic_on_oops) |
012c437d | 480 | panic("Fatal exception"); |
cea6a4ba | 481 | |
dd287796 | 482 | oops_exit(); |
1da177e4 LT |
483 | do_exit(SIGSEGV); |
484 | } | |
485 | ||
b5964405 IM |
486 | static inline void |
487 | die_if_kernel(const char *str, struct pt_regs *regs, long err) | |
1da177e4 | 488 | { |
717b594a | 489 | if (!user_mode_vm(regs)) |
1da177e4 LT |
490 | die(str, regs, err); |
491 | } | |
492 | ||
b5964405 IM |
493 | static void __kprobes |
494 | do_trap(int trapnr, int signr, char *str, int vm86, struct pt_regs *regs, | |
495 | long error_code, siginfo_t *info) | |
1da177e4 | 496 | { |
4f339ecb | 497 | struct task_struct *tsk = current; |
4f339ecb | 498 | |
6b6891f9 | 499 | if (regs->flags & X86_VM_MASK) { |
1da177e4 LT |
500 | if (vm86) |
501 | goto vm86_trap; | |
502 | goto trap_signal; | |
503 | } | |
504 | ||
717b594a | 505 | if (!user_mode(regs)) |
1da177e4 LT |
506 | goto kernel_trap; |
507 | ||
b5964405 IM |
508 | trap_signal: |
509 | /* | |
510 | * We want error_code and trap_no set for userspace faults and | |
511 | * kernelspace faults which result in die(), but not | |
512 | * kernelspace faults which are fixed up. die() gives the | |
513 | * process no chance to handle the signal and notice the | |
514 | * kernel fault information, so that won't result in polluting | |
515 | * the information about previously queued, but not yet | |
516 | * delivered, faults. See also do_general_protection below. | |
517 | */ | |
518 | tsk->thread.error_code = error_code; | |
519 | tsk->thread.trap_no = trapnr; | |
d1895183 | 520 | |
b5964405 IM |
521 | if (info) |
522 | force_sig_info(signr, info, tsk); | |
523 | else | |
524 | force_sig(signr, tsk); | |
525 | return; | |
1da177e4 | 526 | |
b5964405 IM |
527 | kernel_trap: |
528 | if (!fixup_exception(regs)) { | |
529 | tsk->thread.error_code = error_code; | |
530 | tsk->thread.trap_no = trapnr; | |
531 | die(str, regs, error_code); | |
1da177e4 | 532 | } |
b5964405 | 533 | return; |
1da177e4 | 534 | |
b5964405 IM |
535 | vm86_trap: |
536 | if (handle_vm86_trap((struct kernel_vm86_regs *) regs, | |
537 | error_code, trapnr)) | |
538 | goto trap_signal; | |
539 | return; | |
1da177e4 LT |
540 | } |
541 | ||
b5964405 IM |
542 | #define DO_ERROR(trapnr, signr, str, name) \ |
543 | void do_##name(struct pt_regs *regs, long error_code) \ | |
544 | { \ | |
4461145e | 545 | trace_hardirqs_fixup(); \ |
b5964405 | 546 | if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \ |
a8c1be9d | 547 | == NOTIFY_STOP) \ |
b5964405 IM |
548 | return; \ |
549 | do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \ | |
1da177e4 LT |
550 | } |
551 | ||
b5964405 IM |
552 | #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr, irq) \ |
553 | void do_##name(struct pt_regs *regs, long error_code) \ | |
554 | { \ | |
555 | siginfo_t info; \ | |
556 | if (irq) \ | |
557 | local_irq_enable(); \ | |
558 | info.si_signo = signr; \ | |
559 | info.si_errno = 0; \ | |
560 | info.si_code = sicode; \ | |
561 | info.si_addr = (void __user *)siaddr; \ | |
562 | if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \ | |
a8c1be9d | 563 | == NOTIFY_STOP) \ |
b5964405 IM |
564 | return; \ |
565 | do_trap(trapnr, signr, str, 0, regs, error_code, &info); \ | |
1da177e4 LT |
566 | } |
567 | ||
b5964405 IM |
568 | #define DO_VM86_ERROR(trapnr, signr, str, name) \ |
569 | void do_##name(struct pt_regs *regs, long error_code) \ | |
570 | { \ | |
571 | if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \ | |
a8c1be9d | 572 | == NOTIFY_STOP) \ |
b5964405 IM |
573 | return; \ |
574 | do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \ | |
1da177e4 LT |
575 | } |
576 | ||
b5964405 IM |
577 | #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \ |
578 | void do_##name(struct pt_regs *regs, long error_code) \ | |
579 | { \ | |
580 | siginfo_t info; \ | |
581 | info.si_signo = signr; \ | |
582 | info.si_errno = 0; \ | |
583 | info.si_code = sicode; \ | |
584 | info.si_addr = (void __user *)siaddr; \ | |
585 | trace_hardirqs_fixup(); \ | |
586 | if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \ | |
a8c1be9d | 587 | == NOTIFY_STOP) \ |
b5964405 IM |
588 | return; \ |
589 | do_trap(trapnr, signr, str, 1, regs, error_code, &info); \ | |
1da177e4 LT |
590 | } |
591 | ||
a8c1be9d | 592 | DO_VM86_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip) |
1da177e4 | 593 | #ifndef CONFIG_KPROBES |
b5964405 | 594 | DO_VM86_ERROR(3, SIGTRAP, "int3", int3) |
1da177e4 | 595 | #endif |
b5964405 IM |
596 | DO_VM86_ERROR(4, SIGSEGV, "overflow", overflow) |
597 | DO_VM86_ERROR(5, SIGSEGV, "bounds", bounds) | |
a8c1be9d AH |
598 | DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip, 0) |
599 | DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun) | |
1da177e4 | 600 | DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS) |
a8c1be9d AH |
601 | DO_ERROR(11, SIGBUS, "segment not present", segment_not_present) |
602 | DO_ERROR(12, SIGBUS, "stack segment", stack_segment) | |
a10d9a71 | 603 | DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0, 0) |
79bf0e03 | 604 | DO_ERROR_INFO(32, SIGILL, "iret exception", iret_error, ILL_BADSTK, 0, 1) |
1da177e4 | 605 | |
b5964405 | 606 | void __kprobes do_general_protection(struct pt_regs *regs, long error_code) |
1da177e4 | 607 | { |
b5964405 IM |
608 | struct thread_struct *thread; |
609 | struct tss_struct *tss; | |
610 | int cpu; | |
611 | ||
612 | cpu = get_cpu(); | |
613 | tss = &per_cpu(init_tss, cpu); | |
614 | thread = ¤t->thread; | |
1da177e4 LT |
615 | |
616 | /* | |
617 | * Perform the lazy TSS's I/O bitmap copy. If the TSS has an | |
618 | * invalid offset set (the LAZY one) and the faulting thread has | |
619 | * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS | |
620 | * and we set the offset field correctly. Then we let the CPU to | |
621 | * restart the faulting instruction. | |
622 | */ | |
a75c54f9 | 623 | if (tss->x86_tss.io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY && |
1da177e4 LT |
624 | thread->io_bitmap_ptr) { |
625 | memcpy(tss->io_bitmap, thread->io_bitmap_ptr, | |
626 | thread->io_bitmap_max); | |
627 | /* | |
628 | * If the previously set map was extending to higher ports | |
629 | * than the current one, pad extra space with 0xff (no access). | |
630 | */ | |
b5964405 | 631 | if (thread->io_bitmap_max < tss->io_bitmap_max) { |
1da177e4 LT |
632 | memset((char *) tss->io_bitmap + |
633 | thread->io_bitmap_max, 0xff, | |
634 | tss->io_bitmap_max - thread->io_bitmap_max); | |
b5964405 | 635 | } |
1da177e4 | 636 | tss->io_bitmap_max = thread->io_bitmap_max; |
a75c54f9 | 637 | tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET; |
d5cd4aad | 638 | tss->io_bitmap_owner = thread; |
1da177e4 | 639 | put_cpu(); |
b5964405 | 640 | |
1da177e4 LT |
641 | return; |
642 | } | |
643 | put_cpu(); | |
644 | ||
6b6891f9 | 645 | if (regs->flags & X86_VM_MASK) |
1da177e4 LT |
646 | goto gp_in_vm86; |
647 | ||
717b594a | 648 | if (!user_mode(regs)) |
1da177e4 LT |
649 | goto gp_in_kernel; |
650 | ||
651 | current->thread.error_code = error_code; | |
652 | current->thread.trap_no = 13; | |
b5964405 | 653 | |
abd4f750 | 654 | if (show_unhandled_signals && unhandled_signal(current, SIGSEGV) && |
03252919 | 655 | printk_ratelimit()) { |
abd4f750 | 656 | printk(KERN_INFO |
03252919 | 657 | "%s[%d] general protection ip:%lx sp:%lx error:%lx", |
19c5870c | 658 | current->comm, task_pid_nr(current), |
65ea5b03 | 659 | regs->ip, regs->sp, error_code); |
03252919 AK |
660 | print_vma_addr(" in ", regs->ip); |
661 | printk("\n"); | |
662 | } | |
abd4f750 | 663 | |
1da177e4 LT |
664 | force_sig(SIGSEGV, current); |
665 | return; | |
666 | ||
667 | gp_in_vm86: | |
668 | local_irq_enable(); | |
669 | handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code); | |
670 | return; | |
671 | ||
672 | gp_in_kernel: | |
673 | if (!fixup_exception(regs)) { | |
d1895183 AK |
674 | current->thread.error_code = error_code; |
675 | current->thread.trap_no = 13; | |
1da177e4 LT |
676 | if (notify_die(DIE_GPF, "general protection fault", regs, |
677 | error_code, 13, SIGSEGV) == NOTIFY_STOP) | |
678 | return; | |
679 | die("general protection fault", regs, error_code); | |
680 | } | |
681 | } | |
682 | ||
5deb45e3 | 683 | static notrace __kprobes void |
b5964405 | 684 | mem_parity_error(unsigned char reason, struct pt_regs *regs) |
1da177e4 | 685 | { |
b5964405 IM |
686 | printk(KERN_EMERG |
687 | "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n", | |
688 | reason, smp_processor_id()); | |
689 | ||
690 | printk(KERN_EMERG | |
691 | "You have some hardware problem, likely on the PCI bus.\n"); | |
c0d12172 DJ |
692 | |
693 | #if defined(CONFIG_EDAC) | |
b5964405 | 694 | if (edac_handler_set()) { |
c0d12172 DJ |
695 | edac_atomic_assert_error(); |
696 | return; | |
697 | } | |
698 | #endif | |
699 | ||
8da5adda | 700 | if (panic_on_unrecovered_nmi) |
b5964405 | 701 | panic("NMI: Not continuing"); |
1da177e4 | 702 | |
c41c5cd3 | 703 | printk(KERN_EMERG "Dazed and confused, but trying to continue\n"); |
1da177e4 LT |
704 | |
705 | /* Clear and disable the memory parity error line. */ | |
706 | clear_mem_error(reason); | |
707 | } | |
708 | ||
5deb45e3 | 709 | static notrace __kprobes void |
b5964405 | 710 | io_check_error(unsigned char reason, struct pt_regs *regs) |
1da177e4 LT |
711 | { |
712 | unsigned long i; | |
713 | ||
9c107805 | 714 | printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n"); |
1da177e4 LT |
715 | show_registers(regs); |
716 | ||
717 | /* Re-enable the IOCK line, wait for a few seconds */ | |
718 | reason = (reason & 0xf) | 8; | |
719 | outb(reason, 0x61); | |
b5964405 | 720 | |
1da177e4 | 721 | i = 2000; |
b5964405 IM |
722 | while (--i) |
723 | udelay(1000); | |
724 | ||
1da177e4 LT |
725 | reason &= ~8; |
726 | outb(reason, 0x61); | |
727 | } | |
728 | ||
5deb45e3 | 729 | static notrace __kprobes void |
b5964405 | 730 | unknown_nmi_error(unsigned char reason, struct pt_regs *regs) |
1da177e4 | 731 | { |
d3597524 JW |
732 | if (notify_die(DIE_NMIUNKNOWN, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP) |
733 | return; | |
1da177e4 | 734 | #ifdef CONFIG_MCA |
b5964405 IM |
735 | /* |
736 | * Might actually be able to figure out what the guilty party | |
737 | * is: | |
738 | */ | |
739 | if (MCA_bus) { | |
1da177e4 LT |
740 | mca_handle_nmi(); |
741 | return; | |
742 | } | |
743 | #endif | |
b5964405 IM |
744 | printk(KERN_EMERG |
745 | "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n", | |
746 | reason, smp_processor_id()); | |
747 | ||
c41c5cd3 | 748 | printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n"); |
8da5adda | 749 | if (panic_on_unrecovered_nmi) |
b5964405 | 750 | panic("NMI: Not continuing"); |
8da5adda | 751 | |
c41c5cd3 | 752 | printk(KERN_EMERG "Dazed and confused, but trying to continue\n"); |
1da177e4 LT |
753 | } |
754 | ||
755 | static DEFINE_SPINLOCK(nmi_print_lock); | |
756 | ||
ddca03c9 | 757 | void notrace __kprobes die_nmi(char *str, struct pt_regs *regs, int do_panic) |
1da177e4 | 758 | { |
ddca03c9 | 759 | if (notify_die(DIE_NMIWATCHDOG, str, regs, 0, 2, SIGINT) == NOTIFY_STOP) |
748f2edb GA |
760 | return; |
761 | ||
1da177e4 LT |
762 | spin_lock(&nmi_print_lock); |
763 | /* | |
764 | * We are in trouble anyway, lets at least try | |
b5964405 | 765 | * to get a message out: |
1da177e4 LT |
766 | */ |
767 | bust_spinlocks(1); | |
ddca03c9 | 768 | printk(KERN_EMERG "%s", str); |
65ea5b03 PA |
769 | printk(" on CPU%d, ip %08lx, registers:\n", |
770 | smp_processor_id(), regs->ip); | |
1da177e4 | 771 | show_registers(regs); |
ddca03c9 CG |
772 | if (do_panic) |
773 | panic("Non maskable interrupt"); | |
1da177e4 LT |
774 | console_silent(); |
775 | spin_unlock(&nmi_print_lock); | |
776 | bust_spinlocks(0); | |
6e274d14 | 777 | |
b5964405 IM |
778 | /* |
779 | * If we are in kernel we are probably nested up pretty bad | |
780 | * and might aswell get out now while we still can: | |
781 | */ | |
db753bdf | 782 | if (!user_mode_vm(regs)) { |
6e274d14 AN |
783 | current->thread.trap_no = 2; |
784 | crash_kexec(regs); | |
785 | } | |
786 | ||
1da177e4 LT |
787 | do_exit(SIGSEGV); |
788 | } | |
789 | ||
5deb45e3 | 790 | static notrace __kprobes void default_do_nmi(struct pt_regs *regs) |
1da177e4 LT |
791 | { |
792 | unsigned char reason = 0; | |
793 | ||
b5964405 | 794 | /* Only the BSP gets external NMIs from the system: */ |
1da177e4 LT |
795 | if (!smp_processor_id()) |
796 | reason = get_nmi_reason(); | |
b5964405 | 797 | |
1da177e4 | 798 | if (!(reason & 0xc0)) { |
20c0d2d4 | 799 | if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT) |
a8c1be9d | 800 | == NOTIFY_STOP) |
1da177e4 LT |
801 | return; |
802 | #ifdef CONFIG_X86_LOCAL_APIC | |
803 | /* | |
804 | * Ok, so this is none of the documented NMI sources, | |
805 | * so it must be the NMI watchdog. | |
806 | */ | |
3adbbcce | 807 | if (nmi_watchdog_tick(regs, reason)) |
1da177e4 | 808 | return; |
2fbe7b25 | 809 | if (!do_nmi_callback(regs, smp_processor_id())) |
3adbbcce | 810 | unknown_nmi_error(reason, regs); |
b5964405 IM |
811 | #else |
812 | unknown_nmi_error(reason, regs); | |
813 | #endif | |
2fbe7b25 | 814 | |
1da177e4 LT |
815 | return; |
816 | } | |
20c0d2d4 | 817 | if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP) |
1da177e4 | 818 | return; |
a8c1be9d AH |
819 | |
820 | /* AK: following checks seem to be broken on modern chipsets. FIXME */ | |
1da177e4 LT |
821 | if (reason & 0x80) |
822 | mem_parity_error(reason, regs); | |
823 | if (reason & 0x40) | |
824 | io_check_error(reason, regs); | |
825 | /* | |
826 | * Reassert NMI in case it became active meanwhile | |
b5964405 | 827 | * as it's edge-triggered: |
1da177e4 LT |
828 | */ |
829 | reassert_nmi(); | |
830 | } | |
831 | ||
8f4e956b AK |
832 | static int ignore_nmis; |
833 | ||
5deb45e3 | 834 | notrace __kprobes void do_nmi(struct pt_regs *regs, long error_code) |
1da177e4 LT |
835 | { |
836 | int cpu; | |
837 | ||
838 | nmi_enter(); | |
839 | ||
840 | cpu = smp_processor_id(); | |
f3705136 | 841 | |
1da177e4 LT |
842 | ++nmi_count(cpu); |
843 | ||
8f4e956b AK |
844 | if (!ignore_nmis) |
845 | default_do_nmi(regs); | |
1da177e4 LT |
846 | |
847 | nmi_exit(); | |
848 | } | |
849 | ||
8f4e956b AK |
850 | void stop_nmi(void) |
851 | { | |
852 | acpi_nmi_disable(); | |
853 | ignore_nmis++; | |
854 | } | |
855 | ||
856 | void restart_nmi(void) | |
857 | { | |
858 | ignore_nmis--; | |
859 | acpi_nmi_enable(); | |
860 | } | |
861 | ||
1da177e4 | 862 | #ifdef CONFIG_KPROBES |
75604d7f | 863 | void __kprobes do_int3(struct pt_regs *regs, long error_code) |
1da177e4 | 864 | { |
143a5d32 PZ |
865 | trace_hardirqs_fixup(); |
866 | ||
1da177e4 LT |
867 | if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP) |
868 | == NOTIFY_STOP) | |
48c88211 | 869 | return; |
b5964405 IM |
870 | /* |
871 | * This is an interrupt gate, because kprobes wants interrupts | |
872 | * disabled. Normal trap handlers don't. | |
873 | */ | |
1da177e4 | 874 | restore_interrupts(regs); |
b5964405 | 875 | |
1da177e4 | 876 | do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL); |
1da177e4 LT |
877 | } |
878 | #endif | |
879 | ||
880 | /* | |
881 | * Our handling of the processor debug registers is non-trivial. | |
882 | * We do not clear them on entry and exit from the kernel. Therefore | |
883 | * it is possible to get a watchpoint trap here from inside the kernel. | |
884 | * However, the code in ./ptrace.c has ensured that the user can | |
885 | * only set watchpoints on userspace addresses. Therefore the in-kernel | |
886 | * watchpoint trap can only occur in code which is reading/writing | |
887 | * from user space. Such code must not hold kernel locks (since it | |
888 | * can equally take a page fault), therefore it is safe to call | |
889 | * force_sig_info even though that claims and releases locks. | |
b5964405 | 890 | * |
1da177e4 LT |
891 | * Code in ./signal.c ensures that the debug control register |
892 | * is restored before we deliver any signal, and therefore that | |
893 | * user code runs with the correct debug control register even though | |
894 | * we clear it here. | |
895 | * | |
896 | * Being careful here means that we don't have to be as careful in a | |
897 | * lot of more complicated places (task switching can be a bit lazy | |
898 | * about restoring all the debug state, and ptrace doesn't have to | |
899 | * find every occurrence of the TF bit that could be saved away even | |
900 | * by user code) | |
901 | */ | |
b5964405 | 902 | void __kprobes do_debug(struct pt_regs *regs, long error_code) |
1da177e4 | 903 | { |
1da177e4 | 904 | struct task_struct *tsk = current; |
b5964405 | 905 | unsigned int condition; |
1da177e4 | 906 | |
000f4a9e PZ |
907 | trace_hardirqs_fixup(); |
908 | ||
1cc6f12e | 909 | get_debugreg(condition, 6); |
1da177e4 | 910 | |
10faa81e RM |
911 | /* |
912 | * The processor cleared BTF, so don't mark that we need it set. | |
913 | */ | |
914 | clear_tsk_thread_flag(tsk, TIF_DEBUGCTLMSR); | |
915 | tsk->thread.debugctlmsr = 0; | |
916 | ||
1da177e4 | 917 | if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code, |
a8c1be9d | 918 | SIGTRAP) == NOTIFY_STOP) |
1da177e4 LT |
919 | return; |
920 | /* It's safe to allow irq's after DR6 has been saved */ | |
65ea5b03 | 921 | if (regs->flags & X86_EFLAGS_IF) |
1da177e4 LT |
922 | local_irq_enable(); |
923 | ||
924 | /* Mask out spurious debug traps due to lazy DR7 setting */ | |
925 | if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) { | |
0f534093 | 926 | if (!tsk->thread.debugreg7) |
1da177e4 LT |
927 | goto clear_dr7; |
928 | } | |
929 | ||
6b6891f9 | 930 | if (regs->flags & X86_VM_MASK) |
1da177e4 LT |
931 | goto debug_vm86; |
932 | ||
933 | /* Save debug status register where ptrace can see it */ | |
0f534093 | 934 | tsk->thread.debugreg6 = condition; |
1da177e4 LT |
935 | |
936 | /* | |
937 | * Single-stepping through TF: make sure we ignore any events in | |
938 | * kernel space (but re-enable TF when returning to user mode). | |
939 | */ | |
940 | if (condition & DR_STEP) { | |
941 | /* | |
942 | * We already checked v86 mode above, so we can | |
943 | * check for kernel mode by just checking the CPL | |
944 | * of CS. | |
945 | */ | |
717b594a | 946 | if (!user_mode(regs)) |
1da177e4 LT |
947 | goto clear_TF_reenable; |
948 | } | |
949 | ||
950 | /* Ok, finally something we can handle */ | |
951 | send_sigtrap(tsk, regs, error_code); | |
952 | ||
b5964405 IM |
953 | /* |
954 | * Disable additional traps. They'll be re-enabled when | |
1da177e4 LT |
955 | * the signal is delivered. |
956 | */ | |
957 | clear_dr7: | |
1cc6f12e | 958 | set_debugreg(0, 7); |
1da177e4 LT |
959 | return; |
960 | ||
961 | debug_vm86: | |
962 | handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1); | |
963 | return; | |
964 | ||
965 | clear_TF_reenable: | |
966 | set_tsk_thread_flag(tsk, TIF_SINGLESTEP); | |
6093015d | 967 | regs->flags &= ~X86_EFLAGS_TF; |
1da177e4 LT |
968 | return; |
969 | } | |
970 | ||
971 | /* | |
972 | * Note that we play around with the 'TS' bit in an attempt to get | |
973 | * the correct behaviour even in the presence of the asynchronous | |
974 | * IRQ13 behaviour | |
975 | */ | |
65ea5b03 | 976 | void math_error(void __user *ip) |
1da177e4 | 977 | { |
b5964405 IM |
978 | struct task_struct *task; |
979 | unsigned short cwd; | |
980 | unsigned short swd; | |
1da177e4 | 981 | siginfo_t info; |
1da177e4 LT |
982 | |
983 | /* | |
984 | * Save the info for the exception handler and clear the error. | |
985 | */ | |
986 | task = current; | |
987 | save_init_fpu(task); | |
988 | task->thread.trap_no = 16; | |
989 | task->thread.error_code = 0; | |
990 | info.si_signo = SIGFPE; | |
991 | info.si_errno = 0; | |
992 | info.si_code = __SI_FAULT; | |
65ea5b03 | 993 | info.si_addr = ip; |
1da177e4 LT |
994 | /* |
995 | * (~cwd & swd) will mask out exceptions that are not set to unmasked | |
996 | * status. 0x3f is the exception bits in these regs, 0x200 is the | |
997 | * C1 reg you need in case of a stack fault, 0x040 is the stack | |
998 | * fault bit. We should only be taking one exception at a time, | |
999 | * so if this combination doesn't produce any single exception, | |
a8c1be9d | 1000 | * then we have a bad program that isn't synchronizing its FPU usage |
1da177e4 LT |
1001 | * and it will suffer the consequences since we won't be able to |
1002 | * fully reproduce the context of the exception | |
1003 | */ | |
1004 | cwd = get_fpu_cwd(task); | |
1005 | swd = get_fpu_swd(task); | |
b1daec30 | 1006 | switch (swd & ~cwd & 0x3f) { |
b5964405 IM |
1007 | case 0x000: /* No unmasked exception */ |
1008 | return; | |
a8c1be9d | 1009 | default: /* Multiple exceptions */ |
b5964405 IM |
1010 | break; |
1011 | case 0x001: /* Invalid Op */ | |
1012 | /* | |
1013 | * swd & 0x240 == 0x040: Stack Underflow | |
1014 | * swd & 0x240 == 0x240: Stack Overflow | |
1015 | * User must clear the SF bit (0x40) if set | |
1016 | */ | |
1017 | info.si_code = FPE_FLTINV; | |
1018 | break; | |
1019 | case 0x002: /* Denormalize */ | |
1020 | case 0x010: /* Underflow */ | |
1021 | info.si_code = FPE_FLTUND; | |
1022 | break; | |
1023 | case 0x004: /* Zero Divide */ | |
1024 | info.si_code = FPE_FLTDIV; | |
1025 | break; | |
1026 | case 0x008: /* Overflow */ | |
1027 | info.si_code = FPE_FLTOVF; | |
1028 | break; | |
1029 | case 0x020: /* Precision */ | |
1030 | info.si_code = FPE_FLTRES; | |
1031 | break; | |
1da177e4 LT |
1032 | } |
1033 | force_sig_info(SIGFPE, &info, task); | |
1034 | } | |
1035 | ||
b5964405 | 1036 | void do_coprocessor_error(struct pt_regs *regs, long error_code) |
1da177e4 LT |
1037 | { |
1038 | ignore_fpu_irq = 1; | |
65ea5b03 | 1039 | math_error((void __user *)regs->ip); |
1da177e4 LT |
1040 | } |
1041 | ||
65ea5b03 | 1042 | static void simd_math_error(void __user *ip) |
1da177e4 | 1043 | { |
b5964405 | 1044 | struct task_struct *task; |
1da177e4 | 1045 | unsigned short mxcsr; |
b5964405 | 1046 | siginfo_t info; |
1da177e4 LT |
1047 | |
1048 | /* | |
1049 | * Save the info for the exception handler and clear the error. | |
1050 | */ | |
1051 | task = current; | |
1052 | save_init_fpu(task); | |
1053 | task->thread.trap_no = 19; | |
1054 | task->thread.error_code = 0; | |
1055 | info.si_signo = SIGFPE; | |
1056 | info.si_errno = 0; | |
1057 | info.si_code = __SI_FAULT; | |
65ea5b03 | 1058 | info.si_addr = ip; |
1da177e4 LT |
1059 | /* |
1060 | * The SIMD FPU exceptions are handled a little differently, as there | |
1061 | * is only a single status/control register. Thus, to determine which | |
1062 | * unmasked exception was caught we must mask the exception mask bits | |
1063 | * at 0x1f80, and then use these to mask the exception bits at 0x3f. | |
1064 | */ | |
1065 | mxcsr = get_fpu_mxcsr(task); | |
1066 | switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) { | |
b5964405 IM |
1067 | case 0x000: |
1068 | default: | |
1069 | break; | |
1070 | case 0x001: /* Invalid Op */ | |
1071 | info.si_code = FPE_FLTINV; | |
1072 | break; | |
1073 | case 0x002: /* Denormalize */ | |
1074 | case 0x010: /* Underflow */ | |
1075 | info.si_code = FPE_FLTUND; | |
1076 | break; | |
1077 | case 0x004: /* Zero Divide */ | |
1078 | info.si_code = FPE_FLTDIV; | |
1079 | break; | |
1080 | case 0x008: /* Overflow */ | |
1081 | info.si_code = FPE_FLTOVF; | |
1082 | break; | |
1083 | case 0x020: /* Precision */ | |
1084 | info.si_code = FPE_FLTRES; | |
1085 | break; | |
1da177e4 LT |
1086 | } |
1087 | force_sig_info(SIGFPE, &info, task); | |
1088 | } | |
1089 | ||
b5964405 | 1090 | void do_simd_coprocessor_error(struct pt_regs *regs, long error_code) |
1da177e4 LT |
1091 | { |
1092 | if (cpu_has_xmm) { | |
1093 | /* Handle SIMD FPU exceptions on PIII+ processors. */ | |
1094 | ignore_fpu_irq = 1; | |
65ea5b03 | 1095 | simd_math_error((void __user *)regs->ip); |
b5964405 IM |
1096 | return; |
1097 | } | |
1098 | /* | |
1099 | * Handle strange cache flush from user space exception | |
1100 | * in all other cases. This is undocumented behaviour. | |
1101 | */ | |
6b6891f9 | 1102 | if (regs->flags & X86_VM_MASK) { |
b5964405 IM |
1103 | handle_vm86_fault((struct kernel_vm86_regs *)regs, error_code); |
1104 | return; | |
1da177e4 | 1105 | } |
b5964405 IM |
1106 | current->thread.trap_no = 19; |
1107 | current->thread.error_code = error_code; | |
1108 | die_if_kernel("cache flush denied", regs, error_code); | |
1109 | force_sig(SIGSEGV, current); | |
1da177e4 LT |
1110 | } |
1111 | ||
b5964405 | 1112 | void do_spurious_interrupt_bug(struct pt_regs *regs, long error_code) |
1da177e4 LT |
1113 | { |
1114 | #if 0 | |
1115 | /* No need to warn about this any longer. */ | |
b5964405 | 1116 | printk(KERN_INFO "Ignoring P6 Local APIC Spurious Interrupt Bug...\n"); |
1da177e4 LT |
1117 | #endif |
1118 | } | |
1119 | ||
b5964405 | 1120 | unsigned long patch_espfix_desc(unsigned long uesp, unsigned long kesp) |
1da177e4 | 1121 | { |
736f12bf | 1122 | struct desc_struct *gdt = get_cpu_gdt_table(smp_processor_id()); |
be44d2aa SS |
1123 | unsigned long base = (kesp - uesp) & -THREAD_SIZE; |
1124 | unsigned long new_kesp = kesp - base; | |
1125 | unsigned long lim_pages = (new_kesp | (THREAD_SIZE - 1)) >> PAGE_SHIFT; | |
1126 | __u64 desc = *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS]; | |
b5964405 | 1127 | |
be44d2aa | 1128 | /* Set up base for espfix segment */ |
b5964405 IM |
1129 | desc &= 0x00f0ff0000000000ULL; |
1130 | desc |= ((((__u64)base) << 16) & 0x000000ffffff0000ULL) | | |
be44d2aa SS |
1131 | ((((__u64)base) << 32) & 0xff00000000000000ULL) | |
1132 | ((((__u64)lim_pages) << 32) & 0x000f000000000000ULL) | | |
1133 | (lim_pages & 0xffff); | |
1134 | *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS] = desc; | |
b5964405 | 1135 | |
be44d2aa | 1136 | return new_kesp; |
1da177e4 LT |
1137 | } |
1138 | ||
1139 | /* | |
b5964405 | 1140 | * 'math_state_restore()' saves the current math information in the |
1da177e4 LT |
1141 | * old math state array, and gets the new ones from the current task |
1142 | * | |
1143 | * Careful.. There are problems with IBM-designed IRQ13 behaviour. | |
1144 | * Don't touch unless you *really* know how it works. | |
1145 | * | |
1146 | * Must be called with kernel preemption disabled (in this case, | |
1147 | * local interrupts are disabled at the call-site in entry.S). | |
1148 | */ | |
acc20761 | 1149 | asmlinkage void math_state_restore(void) |
1da177e4 LT |
1150 | { |
1151 | struct thread_info *thread = current_thread_info(); | |
1152 | struct task_struct *tsk = thread->task; | |
1153 | ||
aa283f49 SS |
1154 | if (!tsk_used_math(tsk)) { |
1155 | local_irq_enable(); | |
1156 | /* | |
1157 | * does a slab alloc which can sleep | |
1158 | */ | |
1159 | if (init_fpu(tsk)) { | |
1160 | /* | |
1161 | * ran out of memory! | |
1162 | */ | |
1163 | do_group_exit(SIGKILL); | |
1164 | return; | |
1165 | } | |
1166 | local_irq_disable(); | |
1167 | } | |
1168 | ||
b5964405 | 1169 | clts(); /* Allow maths ops (or we recurse) */ |
1da177e4 LT |
1170 | restore_fpu(tsk); |
1171 | thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */ | |
acc20761 | 1172 | tsk->fpu_counter++; |
1da177e4 | 1173 | } |
5992b6da | 1174 | EXPORT_SYMBOL_GPL(math_state_restore); |
1da177e4 LT |
1175 | |
1176 | #ifndef CONFIG_MATH_EMULATION | |
1177 | ||
1178 | asmlinkage void math_emulate(long arg) | |
1179 | { | |
b5964405 IM |
1180 | printk(KERN_EMERG |
1181 | "math-emulation not enabled and no coprocessor found.\n"); | |
1182 | printk(KERN_EMERG "killing %s.\n", current->comm); | |
1183 | force_sig(SIGFPE, current); | |
1da177e4 LT |
1184 | schedule(); |
1185 | } | |
1186 | ||
1187 | #endif /* CONFIG_MATH_EMULATION */ | |
1188 | ||
1da177e4 LT |
1189 | void __init trap_init(void) |
1190 | { | |
dbeb2be2 RR |
1191 | int i; |
1192 | ||
1da177e4 | 1193 | #ifdef CONFIG_EISA |
927222b1 | 1194 | void __iomem *p = early_ioremap(0x0FFFD9, 4); |
b5964405 IM |
1195 | |
1196 | if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24)) | |
1da177e4 | 1197 | EISA_bus = 1; |
927222b1 | 1198 | early_iounmap(p, 4); |
1da177e4 LT |
1199 | #endif |
1200 | ||
a8c1be9d AH |
1201 | set_trap_gate(0, ÷_error); |
1202 | set_intr_gate(1, &debug); | |
1203 | set_intr_gate(2, &nmi); | |
1204 | set_system_intr_gate(3, &int3); /* int3 can be called from all */ | |
1205 | set_system_gate(4, &overflow); /* int4 can be called from all */ | |
1206 | set_trap_gate(5, &bounds); | |
1207 | set_trap_gate(6, &invalid_op); | |
1208 | set_trap_gate(7, &device_not_available); | |
1209 | set_task_gate(8, GDT_ENTRY_DOUBLEFAULT_TSS); | |
1210 | set_trap_gate(9, &coprocessor_segment_overrun); | |
b5964405 IM |
1211 | set_trap_gate(10, &invalid_TSS); |
1212 | set_trap_gate(11, &segment_not_present); | |
1213 | set_trap_gate(12, &stack_segment); | |
1214 | set_trap_gate(13, &general_protection); | |
1215 | set_intr_gate(14, &page_fault); | |
1216 | set_trap_gate(15, &spurious_interrupt_bug); | |
1217 | set_trap_gate(16, &coprocessor_error); | |
1218 | set_trap_gate(17, &alignment_check); | |
1da177e4 | 1219 | #ifdef CONFIG_X86_MCE |
b5964405 | 1220 | set_trap_gate(18, &machine_check); |
1da177e4 | 1221 | #endif |
b5964405 | 1222 | set_trap_gate(19, &simd_coprocessor_error); |
1da177e4 | 1223 | |
d43c6e80 | 1224 | if (cpu_has_fxsr) { |
d43c6e80 JB |
1225 | printk(KERN_INFO "Enabling fast FPU save and restore... "); |
1226 | set_in_cr4(X86_CR4_OSFXSR); | |
1227 | printk("done.\n"); | |
1228 | } | |
1229 | if (cpu_has_xmm) { | |
b5964405 IM |
1230 | printk(KERN_INFO |
1231 | "Enabling unmasked SIMD FPU exception support... "); | |
d43c6e80 JB |
1232 | set_in_cr4(X86_CR4_OSXMMEXCPT); |
1233 | printk("done.\n"); | |
1234 | } | |
1235 | ||
b5964405 | 1236 | set_system_gate(SYSCALL_VECTOR, &system_call); |
1da177e4 | 1237 | |
b5964405 | 1238 | /* Reserve all the builtin and the syscall vector: */ |
dbeb2be2 RR |
1239 | for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++) |
1240 | set_bit(i, used_vectors); | |
b5964405 | 1241 | |
dbeb2be2 RR |
1242 | set_bit(SYSCALL_VECTOR, used_vectors); |
1243 | ||
61c4628b | 1244 | init_thread_xstate(); |
1da177e4 | 1245 | /* |
b5964405 | 1246 | * Should be a barrier for any external CPU state: |
1da177e4 LT |
1247 | */ |
1248 | cpu_init(); | |
1249 | ||
1250 | trap_init_hook(); | |
1251 | } | |
1252 | ||
1253 | static int __init kstack_setup(char *s) | |
1254 | { | |
1255 | kstack_depth_to_print = simple_strtoul(s, NULL, 0); | |
b5964405 | 1256 | |
9b41046c | 1257 | return 1; |
1da177e4 LT |
1258 | } |
1259 | __setup("kstack=", kstack_setup); | |
86c41837 CE |
1260 | |
1261 | static int __init code_bytes_setup(char *s) | |
1262 | { | |
1263 | code_bytes = simple_strtoul(s, NULL, 0); | |
1264 | if (code_bytes > 8192) | |
1265 | code_bytes = 8192; | |
1266 | ||
1267 | return 1; | |
1268 | } | |
1269 | __setup("code_bytes=", code_bytes_setup); |