projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
[PATCH] Notifier chain update: API changes
[linux-2.6-block.git] / arch / i386 / kernel / traps.c
CommitLineData
1da177e4
LT		 1/*
2 * linux/arch/i386/traps.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 *
6 * Pentium III FXSR, SSE support
7 * Gareth Hughes <gareth@valinux.com>, May 2000
8 */
9
10/*
11 * 'Traps.c' handles hardware traps and faults after we have saved some
12 * state in 'asm.s'.
13 */
14#include <linux/config.h>
15#include <linux/sched.h>
16#include <linux/kernel.h>
17#include <linux/string.h>
18#include <linux/errno.h>
19#include <linux/timer.h>
20#include <linux/mm.h>
21#include <linux/init.h>
22#include <linux/delay.h>
23#include <linux/spinlock.h>
24#include <linux/interrupt.h>
25#include <linux/highmem.h>
26#include <linux/kallsyms.h>
27#include <linux/ptrace.h>
28#include <linux/utsname.h>
29#include <linux/kprobes.h>
6e274d14 30#include <linux/kexec.h>
1da177e4
LT		 31
32#ifdef CONFIG_EISA
33#include <linux/ioport.h>
34#include <linux/eisa.h>
35#endif
36
37#ifdef CONFIG_MCA
38#include <linux/mca.h>
39#endif
40
41#include <asm/processor.h>
42#include <asm/system.h>
43#include <asm/uaccess.h>
44#include <asm/io.h>
45#include <asm/atomic.h>
46#include <asm/debugreg.h>
47#include <asm/desc.h>
48#include <asm/i387.h>
49#include <asm/nmi.h>
50
51#include <asm/smp.h>
52#include <asm/arch_hooks.h>
53#include <asm/kdebug.h>
54
1da177e4
LT		 55#include <linux/module.h>
56
57#include "mach_traps.h"
58
59asmlinkage int system_call(void);
60
61struct desc_struct default_ldt[] = { { 0, 0 }, { 0, 0 }, { 0, 0 },
62 { 0, 0 }, { 0, 0 } };
63
64/* Do we ignore FPU interrupts ? */
65char ignore_fpu_irq = 0;
66
67/*
68 * The IDT has to be page-aligned to simplify the Pentium
69 * F0 0F bug workaround.. We have a special link segment
70 * for this.
71 */
72struct desc_struct idt_table[256] __attribute__((__section__(".data.idt"))) = { {0, 0}, };
73
74asmlinkage void divide_error(void);
75asmlinkage void debug(void);
76asmlinkage void nmi(void);
77asmlinkage void int3(void);
78asmlinkage void overflow(void);
79asmlinkage void bounds(void);
80asmlinkage void invalid_op(void);
81asmlinkage void device_not_available(void);
82asmlinkage void coprocessor_segment_overrun(void);
83asmlinkage void invalid_TSS(void);
84asmlinkage void segment_not_present(void);
85asmlinkage void stack_segment(void);
86asmlinkage void general_protection(void);
87asmlinkage void page_fault(void);
88asmlinkage void coprocessor_error(void);
89asmlinkage void simd_coprocessor_error(void);
90asmlinkage void alignment_check(void);
91asmlinkage void spurious_interrupt_bug(void);
92asmlinkage void machine_check(void);
93
94static int kstack_depth_to_print = 24;
e041c683 95ATOMIC_NOTIFIER_HEAD(i386die_chain);
1da177e4
LT		 96
97int register_die_notifier(struct notifier_block *nb)
98{
101f12af 99 vmalloc_sync_all();
e041c683 100 return atomic_notifier_chain_register(&i386die_chain, nb);
1da177e4 101}
129f6946 102EXPORT_SYMBOL(register_die_notifier);
1da177e4 103
e041c683
AS		 104int unregister_die_notifier(struct notifier_block *nb)
105{
106 return atomic_notifier_chain_unregister(&i386die_chain, nb);
107}
108EXPORT_SYMBOL(unregister_die_notifier);
109
1da177e4
LT		 110static inline int valid_stack_ptr(struct thread_info *tinfo, void *p)
111{
112 return p > (void *)tinfo &&
113 p < (void *)tinfo + THREAD_SIZE - 3;
114}
115
4d7d8c82
CE		 116/*
117 * Print CONFIG_STACK_BACKTRACE_COLS address/symbol entries per line.
118 */
119static inline int print_addr_and_symbol(unsigned long addr, char *log_lvl,
120 int printed)
7aa89746 121{
4d7d8c82
CE		 122 if (!printed)
123 printk(log_lvl);
124
125#if CONFIG_STACK_BACKTRACE_COLS == 1
7aa89746 126 printk(" [<%08lx>] ", addr);
4d7d8c82
CE		 127#else
128 printk(" <%08lx> ", addr);
129#endif
7aa89746 130 print_symbol("%s", addr);
4d7d8c82
CE		 131
132 printed = (printed + 1) % CONFIG_STACK_BACKTRACE_COLS;
133
134 if (printed)
135 printk(" ");
136 else
137 printk("\n");
138
139 return printed;
7aa89746
CE		 140}
141
1da177e4 142static inline unsigned long print_context_stack(struct thread_info *tinfo,
7aa89746
CE		 143 unsigned long *stack, unsigned long ebp,
144 char *log_lvl)
1da177e4
LT		 145{
146 unsigned long addr;
4d7d8c82 147 int printed = 0; /* nr of entries already printed on current line */
1da177e4
LT		 148
149#ifdef CONFIG_FRAME_POINTER
150 while (valid_stack_ptr(tinfo, (void *)ebp)) {
151 addr = *(unsigned long *)(ebp + 4);
4d7d8c82 152 printed = print_addr_and_symbol(addr, log_lvl, printed);
1da177e4
LT		 153 ebp = *(unsigned long *)ebp;
154 }
155#else
156 while (valid_stack_ptr(tinfo, stack)) {
157 addr = *stack++;
7aa89746 158 if (__kernel_text_address(addr))
4d7d8c82 159 printed = print_addr_and_symbol(addr, log_lvl, printed);
1da177e4
LT		 160 }
161#endif
4d7d8c82
CE		 162 if (printed)
163 printk("\n");
164
1da177e4
LT		 165 return ebp;
166}
167
7aa89746
CE		 168static void show_trace_log_lvl(struct task_struct *task,
169 unsigned long *stack, char *log_lvl)
1da177e4
LT		 170{
171 unsigned long ebp;
172
173 if (!task)
174 task = current;
175
176 if (task == current) {
177 /* Grab ebp right from our regs */
178 asm ("movl %%ebp, %0" : "=r" (ebp) : );
179 } else {
180 /* ebp is the last reg pushed by switch_to */
181 ebp = *(unsigned long *) task->thread.esp;
182 }
183
184 while (1) {
185 struct thread_info *context;
186 context = (struct thread_info *)
187 ((unsigned long)stack & (~(THREAD_SIZE - 1)));
7aa89746 188 ebp = print_context_stack(context, stack, ebp, log_lvl);
1da177e4
LT		 189 stack = (unsigned long*)context->previous_esp;
190 if (!stack)
191 break;
cc04ee9c 192 printk("%s =======================\n", log_lvl);
1da177e4
LT		 193 }
194}
195
7aa89746
CE		 196void show_trace(struct task_struct *task, unsigned long * stack)
197{
198 show_trace_log_lvl(task, stack, "");
199}
200
201static void show_stack_log_lvl(struct task_struct *task, unsigned long *esp,
202 char *log_lvl)
1da177e4
LT		 203{
204 unsigned long *stack;
205 int i;
206
207 if (esp == NULL) {
208 if (task)
209 esp = (unsigned long*)task->thread.esp;
210 else
211 esp = (unsigned long *)&esp;
212 }
213
214 stack = esp;
7aa89746 215 printk(log_lvl);
1da177e4
LT		 216 for(i = 0; i < kstack_depth_to_print; i++) {
217 if (kstack_end(stack))
218 break;
75874d5c
CE		 219 if (i && ((i % 8) == 0))
220 printk("\n%s ", log_lvl);
1da177e4
LT		 221 printk("%08lx ", *stack++);
222 }
75874d5c 223 printk("\n%sCall Trace:\n", log_lvl);
7aa89746
CE		 224 show_trace_log_lvl(task, esp, log_lvl);
225}
226
227void show_stack(struct task_struct *task, unsigned long *esp)
228{
75874d5c 229 printk(" ");
7aa89746 230 show_stack_log_lvl(task, esp, "");
1da177e4
LT		 231}
232
233/*
234 * The architecture-independent dump_stack generator
235 */
236void dump_stack(void)
237{
238 unsigned long stack;
239
240 show_trace(current, &stack);
241}
242
243EXPORT_SYMBOL(dump_stack);
244
245void show_registers(struct pt_regs *regs)
246{
247 int i;
248 int in_kernel = 1;
249 unsigned long esp;
250 unsigned short ss;
251
252 esp = (unsigned long) (&regs->esp);
0998e422 253 savesegment(ss, ss);
db753bdf 254 if (user_mode_vm(regs)) {
1da177e4
LT		 255 in_kernel = 0;
256 esp = regs->esp;
257 ss = regs->xss & 0xffff;
258 }
259 print_modules();
9c107805 260 printk(KERN_EMERG "CPU: %d\nEIP: %04x:[<%08lx>] %s VLI\n"
b53e8f68 261 "EFLAGS: %08lx (%s %.*s) \n",
1da177e4 262 smp_processor_id(), 0xffff & regs->xcs, regs->eip,
b53e8f68
CE		 263 print_tainted(), regs->eflags, system_utsname.release,
264 (int)strcspn(system_utsname.version, " "),
265 system_utsname.version);
9c107805
DJ		 266 print_symbol(KERN_EMERG "EIP is at %s\n", regs->eip);
267 printk(KERN_EMERG "eax: %08lx ebx: %08lx ecx: %08lx edx: %08lx\n",
1da177e4 268 regs->eax, regs->ebx, regs->ecx, regs->edx);
9c107805 269 printk(KERN_EMERG "esi: %08lx edi: %08lx ebp: %08lx esp: %08lx\n",
1da177e4 270 regs->esi, regs->edi, regs->ebp, esp);
9c107805 271 printk(KERN_EMERG "ds: %04x es: %04x ss: %04x\n",
1da177e4 272 regs->xds & 0xffff, regs->xes & 0xffff, ss);
9c107805 273 printk(KERN_EMERG "Process %s (pid: %d, threadinfo=%p task=%p)",
1da177e4
LT		 274 current->comm, current->pid, current_thread_info(), current);
275 /*
276 * When in-kernel, we also print out the stack and code at the
277 * time of the fault..
278 */
279 if (in_kernel) {
3f3ae347 280 u8 __user *eip;
1da177e4 281
9c107805 282 printk("\n" KERN_EMERG "Stack: ");
7aa89746 283 show_stack_log_lvl(NULL, (unsigned long *)esp, KERN_EMERG);
1da177e4 284
9c107805 285 printk(KERN_EMERG "Code: ");
1da177e4 286
3f3ae347 287 eip = (u8 __user *)regs->eip - 43;
1da177e4
LT		 288 for (i = 0; i < 64; i++, eip++) {
289 unsigned char c;
290
3f3ae347 291 if (eip < (u8 __user *)PAGE_OFFSET || __get_user(c, eip)) {
1da177e4
LT		 292 printk(" Bad EIP value.");
293 break;
294 }
3f3ae347 295 if (eip == (u8 __user *)regs->eip)
1da177e4
LT		 296 printk("<%02x> ", c);
297 else
298 printk("%02x ", c);
299 }
300 }
301 printk("\n");
302}
303
304static void handle_BUG(struct pt_regs *regs)
305{
306 unsigned short ud2;
307 unsigned short line;
308 char *file;
309 char c;
310 unsigned long eip;
311
1da177e4
LT		 312 eip = regs->eip;
313
314 if (eip < PAGE_OFFSET)
315 goto no_bug;
3f3ae347 316 if (__get_user(ud2, (unsigned short __user *)eip))
1da177e4
LT		 317 goto no_bug;
318 if (ud2 != 0x0b0f)
319 goto no_bug;
3f3ae347 320 if (__get_user(line, (unsigned short __user *)(eip + 2)))
1da177e4 321 goto bug;
3f3ae347 322 if (__get_user(file, (char * __user *)(eip + 4)) ||
1da177e4
LT		 323 (unsigned long)file < PAGE_OFFSET || __get_user(c, file))
324 file = "<bad filename>";
325
9c107805
DJ		 326 printk(KERN_EMERG "------------[ cut here ]------------\n");
327 printk(KERN_EMERG "kernel BUG at %s:%d!\n", file, line);
1da177e4
LT		 328
329no_bug:
330 return;
331
332 /* Here we know it was a BUG but file-n-line is unavailable */
333bug:
9c107805 334 printk(KERN_EMERG "Kernel BUG\n");
1da177e4
LT		 335}
336
6e274d14
AN		 337/* This is gone through when something in the kernel
338 * has done something bad and is about to be terminated.
339*/
1da177e4
LT		 340void die(const char * str, struct pt_regs * regs, long err)
341{
342 static struct {
343 spinlock_t lock;
344 u32 lock_owner;
345 int lock_owner_depth;
346 } die = {
347 .lock = SPIN_LOCK_UNLOCKED,
348 .lock_owner = -1,
349 .lock_owner_depth = 0
350 };
351 static int die_counter;
e43d674f 352 unsigned long flags;
1da177e4 353
dd287796
AM		 354 oops_enter();
355
39c715b7 356 if (die.lock_owner != raw_smp_processor_id()) {
1da177e4 357 console_verbose();
e43d674f 358 spin_lock_irqsave(&die.lock, flags);
1da177e4
LT		 359 die.lock_owner = smp_processor_id();
360 die.lock_owner_depth = 0;
361 bust_spinlocks(1);
362 }
e43d674f
JB		 363 else
364 local_save_flags(flags);
1da177e4
LT		 365
366 if (++die.lock_owner_depth < 3) {
367 int nl = 0;
368 handle_BUG(regs);
9c107805 369 printk(KERN_EMERG "%s: %04lx [#%d]\n", str, err & 0xffff, ++die_counter);
1da177e4 370#ifdef CONFIG_PREEMPT
9c107805 371 printk(KERN_EMERG "PREEMPT ");
1da177e4
LT		 372 nl = 1;
373#endif
374#ifdef CONFIG_SMP
9c107805
DJ		 375 if (!nl)
376 printk(KERN_EMERG);
1da177e4
LT		 377 printk("SMP ");
378 nl = 1;
379#endif
380#ifdef CONFIG_DEBUG_PAGEALLOC
9c107805
DJ		 381 if (!nl)
382 printk(KERN_EMERG);
1da177e4
LT		 383 printk("DEBUG_PAGEALLOC");
384 nl = 1;
385#endif
386 if (nl)
387 printk("\n");
20c0d2d4
JB		 388 if (notify_die(DIE_OOPS, str, regs, err,
389 current->thread.trap_no, SIGSEGV) !=
390 NOTIFY_STOP)
391 show_registers(regs);
392 else
393 regs = NULL;
1da177e4 394 } else
9c107805 395 printk(KERN_EMERG "Recursive die() failure, output suppressed\n");
1da177e4
LT		 396
397 bust_spinlocks(0);
398 die.lock_owner = -1;
e43d674f 399 spin_unlock_irqrestore(&die.lock, flags);
6e274d14 400
20c0d2d4
JB		 401 if (!regs)
402 return;
403
6e274d14
AN		 404 if (kexec_should_crash(current))
405 crash_kexec(regs);
406
1da177e4
LT		 407 if (in_interrupt())
408 panic("Fatal exception in interrupt");
409
410 if (panic_on_oops) {
411 printk(KERN_EMERG "Fatal exception: panic in 5 seconds\n");
412 ssleep(5);
413 panic("Fatal exception");
414 }
dd287796 415 oops_exit();
1da177e4
LT		 416 do_exit(SIGSEGV);
417}
418
419static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
420{
717b594a 421 if (!user_mode_vm(regs))
1da177e4
LT		 422 die(str, regs, err);
423}
424
3d97ae5b
PP		 425static void __kprobes do_trap(int trapnr, int signr, char *str, int vm86,
426 struct pt_regs * regs, long error_code,
427 siginfo_t *info)
1da177e4 428{
4f339ecb
AN		 429 struct task_struct *tsk = current;
430 tsk->thread.error_code = error_code;
431 tsk->thread.trap_no = trapnr;
432
1da177e4
LT		 433 if (regs->eflags & VM_MASK) {
434 if (vm86)
435 goto vm86_trap;
436 goto trap_signal;
437 }
438
717b594a 439 if (!user_mode(regs))
1da177e4
LT		 440 goto kernel_trap;
441
442 trap_signal: {
1da177e4
LT		 443 if (info)
444 force_sig_info(signr, info, tsk);
445 else
446 force_sig(signr, tsk);
447 return;
448 }
449
450 kernel_trap: {
451 if (!fixup_exception(regs))
452 die(str, regs, error_code);
453 return;
454 }
455
456 vm86_trap: {
457 int ret = handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, trapnr);
458 if (ret) goto trap_signal;
459 return;
460 }
461}
462
463#define DO_ERROR(trapnr, signr, str, name) \
464fastcall void do_##name(struct pt_regs * regs, long error_code) \
465{ \
466 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
467 == NOTIFY_STOP) \
468 return; \
469 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
470}
471
472#define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
473fastcall void do_##name(struct pt_regs * regs, long error_code) \
474{ \
475 siginfo_t info; \
476 info.si_signo = signr; \
477 info.si_errno = 0; \
478 info.si_code = sicode; \
479 info.si_addr = (void __user *)siaddr; \
480 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
481 == NOTIFY_STOP) \
482 return; \
483 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
484}
485
486#define DO_VM86_ERROR(trapnr, signr, str, name) \
487fastcall void do_##name(struct pt_regs * regs, long error_code) \
488{ \
489 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
490 == NOTIFY_STOP) \
491 return; \
492 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
493}
494
495#define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
496fastcall void do_##name(struct pt_regs * regs, long error_code) \
497{ \
498 siginfo_t info; \
499 info.si_signo = signr; \
500 info.si_errno = 0; \
501 info.si_code = sicode; \
502 info.si_addr = (void __user *)siaddr; \
503 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
504 == NOTIFY_STOP) \
505 return; \
506 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
507}
508
509DO_VM86_ERROR_INFO( 0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->eip)
510#ifndef CONFIG_KPROBES
511DO_VM86_ERROR( 3, SIGTRAP, "int3", int3)
512#endif
513DO_VM86_ERROR( 4, SIGSEGV, "overflow", overflow)
514DO_VM86_ERROR( 5, SIGSEGV, "bounds", bounds)
631b0347 515DO_ERROR_INFO( 6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->eip)
1da177e4
LT		 516DO_ERROR( 9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
517DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
518DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
519DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
520DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
a879cbbb 521DO_ERROR_INFO(32, SIGSEGV, "iret exception", iret_error, ILL_BADSTK, 0)
1da177e4 522
3d97ae5b
PP		 523fastcall void __kprobes do_general_protection(struct pt_regs * regs,
524 long error_code)
1da177e4
LT		 525{
526 int cpu = get_cpu();
527 struct tss_struct *tss = &per_cpu(init_tss, cpu);
528 struct thread_struct *thread = &current->thread;
529
530 /*
531 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
532 * invalid offset set (the LAZY one) and the faulting thread has
533 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
534 * and we set the offset field correctly. Then we let the CPU to
535 * restart the faulting instruction.
536 */
537 if (tss->io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
538 thread->io_bitmap_ptr) {
539 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
540 thread->io_bitmap_max);
541 /*
542 * If the previously set map was extending to higher ports
543 * than the current one, pad extra space with 0xff (no access).
544 */
545 if (thread->io_bitmap_max < tss->io_bitmap_max)
546 memset((char *) tss->io_bitmap +
547 thread->io_bitmap_max, 0xff,
548 tss->io_bitmap_max - thread->io_bitmap_max);
549 tss->io_bitmap_max = thread->io_bitmap_max;
550 tss->io_bitmap_base = IO_BITMAP_OFFSET;
d5cd4aad 551 tss->io_bitmap_owner = thread;
1da177e4
LT		 552 put_cpu();
553 return;
554 }
555 put_cpu();
556
4f339ecb
AN		 557 current->thread.error_code = error_code;
558 current->thread.trap_no = 13;
559
1da177e4
LT		 560 if (regs->eflags & VM_MASK)
561 goto gp_in_vm86;
562
717b594a 563 if (!user_mode(regs))
1da177e4
LT		 564 goto gp_in_kernel;
565
566 current->thread.error_code = error_code;
567 current->thread.trap_no = 13;
568 force_sig(SIGSEGV, current);
569 return;
570
571gp_in_vm86:
572 local_irq_enable();
573 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
574 return;
575
576gp_in_kernel:
577 if (!fixup_exception(regs)) {
578 if (notify_die(DIE_GPF, "general protection fault", regs,
579 error_code, 13, SIGSEGV) == NOTIFY_STOP)
580 return;
581 die("general protection fault", regs, error_code);
582 }
583}
584
585static void mem_parity_error(unsigned char reason, struct pt_regs * regs)
586{
9c107805
DJ		 587 printk(KERN_EMERG "Uhhuh. NMI received. Dazed and confused, but trying "
588 "to continue\n");
589 printk(KERN_EMERG "You probably have a hardware problem with your RAM "
590 "chips\n");
1da177e4
LT		 591
592 /* Clear and disable the memory parity error line. */
593 clear_mem_error(reason);
594}
595
596static void io_check_error(unsigned char reason, struct pt_regs * regs)
597{
598 unsigned long i;
599
9c107805 600 printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
1da177e4
LT		 601 show_registers(regs);
602
603 /* Re-enable the IOCK line, wait for a few seconds */
604 reason = (reason & 0xf) | 8;
605 outb(reason, 0x61);
606 i = 2000;
607 while (--i) udelay(1000);
608 reason &= ~8;
609 outb(reason, 0x61);
610}
611
612static void unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
613{
614#ifdef CONFIG_MCA
615 /* Might actually be able to figure out what the guilty party
616 * is. */
617 if( MCA_bus ) {
618 mca_handle_nmi();
619 return;
620 }
621#endif
622 printk("Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
623 reason, smp_processor_id());
624 printk("Dazed and confused, but trying to continue\n");
625 printk("Do you have a strange power saving mode enabled?\n");
626}
627
628static DEFINE_SPINLOCK(nmi_print_lock);
629
630void die_nmi (struct pt_regs *regs, const char *msg)
631{
20c0d2d4 632 if (notify_die(DIE_NMIWATCHDOG, msg, regs, 0, 2, SIGINT) ==
748f2edb
GA		 633 NOTIFY_STOP)
634 return;
635
1da177e4
LT		 636 spin_lock(&nmi_print_lock);
637 /*
638 * We are in trouble anyway, lets at least try
639 * to get a message out.
640 */
641 bust_spinlocks(1);
9c107805 642 printk(KERN_EMERG "%s", msg);
1da177e4
LT		 643 printk(" on CPU%d, eip %08lx, registers:\n",
644 smp_processor_id(), regs->eip);
645 show_registers(regs);
9c107805 646 printk(KERN_EMERG "console shuts up ...\n");
1da177e4
LT		 647 console_silent();
648 spin_unlock(&nmi_print_lock);
649 bust_spinlocks(0);
6e274d14
AN		 650
651 /* If we are in kernel we are probably nested up pretty bad
652 * and might aswell get out now while we still can.
653 */
db753bdf 654 if (!user_mode_vm(regs)) {
6e274d14
AN		 655 current->thread.trap_no = 2;
656 crash_kexec(regs);
657 }
658
1da177e4
LT		 659 do_exit(SIGSEGV);
660}
661
662static void default_do_nmi(struct pt_regs * regs)
663{
664 unsigned char reason = 0;
665
666 /* Only the BSP gets external NMIs from the system. */
667 if (!smp_processor_id())
668 reason = get_nmi_reason();
669
670 if (!(reason & 0xc0)) {
20c0d2d4 671 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
1da177e4
LT		 672 == NOTIFY_STOP)
673 return;
674#ifdef CONFIG_X86_LOCAL_APIC
675 /*
676 * Ok, so this is none of the documented NMI sources,
677 * so it must be the NMI watchdog.
678 */
679 if (nmi_watchdog) {
680 nmi_watchdog_tick(regs);
681 return;
682 }
683#endif
684 unknown_nmi_error(reason, regs);
685 return;
686 }
20c0d2d4 687 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
1da177e4
LT		 688 return;
689 if (reason & 0x80)
690 mem_parity_error(reason, regs);
691 if (reason & 0x40)
692 io_check_error(reason, regs);
693 /*
694 * Reassert NMI in case it became active meanwhile
695 * as it's edge-triggered.
696 */
697 reassert_nmi();
698}
699
700static int dummy_nmi_callback(struct pt_regs * regs, int cpu)
701{
702 return 0;
703}
704
705static nmi_callback_t nmi_callback = dummy_nmi_callback;
706
707fastcall void do_nmi(struct pt_regs * regs, long error_code)
708{
709 int cpu;
710
711 nmi_enter();
712
713 cpu = smp_processor_id();
f3705136 714
1da177e4
LT		 715 ++nmi_count(cpu);
716
19306059 717 if (!rcu_dereference(nmi_callback)(regs, cpu))
1da177e4
LT		 718 default_do_nmi(regs);
719
720 nmi_exit();
721}
722
723void set_nmi_callback(nmi_callback_t callback)
724{
101f12af 725 vmalloc_sync_all();
19306059 726 rcu_assign_pointer(nmi_callback, callback);
1da177e4 727}
129f6946 728EXPORT_SYMBOL_GPL(set_nmi_callback);
1da177e4
LT		 729
730void unset_nmi_callback(void)
731{
732 nmi_callback = dummy_nmi_callback;
733}
129f6946 734EXPORT_SYMBOL_GPL(unset_nmi_callback);
1da177e4
LT		 735
736#ifdef CONFIG_KPROBES
3d97ae5b 737fastcall void __kprobes do_int3(struct pt_regs *regs, long error_code)
1da177e4
LT		 738{
739 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
740 == NOTIFY_STOP)
48c88211 741 return;
1da177e4
LT		 742 /* This is an interrupt gate, because kprobes wants interrupts
743 disabled. Normal trap handlers don't. */
744 restore_interrupts(regs);
745 do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL);
1da177e4
LT		 746}
747#endif
748
749/*
750 * Our handling of the processor debug registers is non-trivial.
751 * We do not clear them on entry and exit from the kernel. Therefore
752 * it is possible to get a watchpoint trap here from inside the kernel.
753 * However, the code in ./ptrace.c has ensured that the user can
754 * only set watchpoints on userspace addresses. Therefore the in-kernel
755 * watchpoint trap can only occur in code which is reading/writing
756 * from user space. Such code must not hold kernel locks (since it
757 * can equally take a page fault), therefore it is safe to call
758 * force_sig_info even though that claims and releases locks.
759 *
760 * Code in ./signal.c ensures that the debug control register
761 * is restored before we deliver any signal, and therefore that
762 * user code runs with the correct debug control register even though
763 * we clear it here.
764 *
765 * Being careful here means that we don't have to be as careful in a
766 * lot of more complicated places (task switching can be a bit lazy
767 * about restoring all the debug state, and ptrace doesn't have to
768 * find every occurrence of the TF bit that could be saved away even
769 * by user code)
770 */
3d97ae5b 771fastcall void __kprobes do_debug(struct pt_regs * regs, long error_code)
1da177e4
LT		 772{
773 unsigned int condition;
774 struct task_struct *tsk = current;
775
1cc6f12e 776 get_debugreg(condition, 6);
1da177e4
LT		 777
778 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
779 SIGTRAP) == NOTIFY_STOP)
780 return;
781 /* It's safe to allow irq's after DR6 has been saved */
782 if (regs->eflags & X86_EFLAGS_IF)
783 local_irq_enable();
784
785 /* Mask out spurious debug traps due to lazy DR7 setting */
786 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
787 if (!tsk->thread.debugreg[7])
788 goto clear_dr7;
789 }
790
791 if (regs->eflags & VM_MASK)
792 goto debug_vm86;
793
794 /* Save debug status register where ptrace can see it */
795 tsk->thread.debugreg[6] = condition;
796
797 /*
798 * Single-stepping through TF: make sure we ignore any events in
799 * kernel space (but re-enable TF when returning to user mode).
800 */
801 if (condition & DR_STEP) {
802 /*
803 * We already checked v86 mode above, so we can
804 * check for kernel mode by just checking the CPL
805 * of CS.
806 */
717b594a 807 if (!user_mode(regs))
1da177e4
LT		 808 goto clear_TF_reenable;
809 }
810
811 /* Ok, finally something we can handle */
812 send_sigtrap(tsk, regs, error_code);
813
814 /* Disable additional traps. They'll be re-enabled when
815 * the signal is delivered.
816 */
817clear_dr7:
1cc6f12e 818 set_debugreg(0, 7);
1da177e4
LT		 819 return;
820
821debug_vm86:
822 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
823 return;
824
825clear_TF_reenable:
826 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
827 regs->eflags &= ~TF_MASK;
828 return;
829}
830
831/*
832 * Note that we play around with the 'TS' bit in an attempt to get
833 * the correct behaviour even in the presence of the asynchronous
834 * IRQ13 behaviour
835 */
836void math_error(void __user *eip)
837{
838 struct task_struct * task;
839 siginfo_t info;
840 unsigned short cwd, swd;
841
842 /*
843 * Save the info for the exception handler and clear the error.
844 */
845 task = current;
846 save_init_fpu(task);
847 task->thread.trap_no = 16;
848 task->thread.error_code = 0;
849 info.si_signo = SIGFPE;
850 info.si_errno = 0;
851 info.si_code = __SI_FAULT;
852 info.si_addr = eip;
853 /*
854 * (~cwd & swd) will mask out exceptions that are not set to unmasked
855 * status. 0x3f is the exception bits in these regs, 0x200 is the
856 * C1 reg you need in case of a stack fault, 0x040 is the stack
857 * fault bit. We should only be taking one exception at a time,
858 * so if this combination doesn't produce any single exception,
859 * then we have a bad program that isn't syncronizing its FPU usage
860 * and it will suffer the consequences since we won't be able to
861 * fully reproduce the context of the exception
862 */
863 cwd = get_fpu_cwd(task);
864 swd = get_fpu_swd(task);
b1daec30 865 switch (swd & ~cwd & 0x3f) {
33333373
CE		 866 case 0x000: /* No unmasked exception */
867 return;
868 default: /* Multiple exceptions */
1da177e4
LT		 869 break;
870 case 0x001: /* Invalid Op */
b1daec30
CE		 871 /*
872 * swd & 0x240 == 0x040: Stack Underflow
873 * swd & 0x240 == 0x240: Stack Overflow
874 * User must clear the SF bit (0x40) if set
875 */
1da177e4 876 info.si_code = FPE_FLTINV;
1da177e4
LT		 877 break;
878 case 0x002: /* Denormalize */
879 case 0x010: /* Underflow */
880 info.si_code = FPE_FLTUND;
881 break;
882 case 0x004: /* Zero Divide */
883 info.si_code = FPE_FLTDIV;
884 break;
885 case 0x008: /* Overflow */
886 info.si_code = FPE_FLTOVF;
887 break;
888 case 0x020: /* Precision */
889 info.si_code = FPE_FLTRES;
890 break;
891 }
892 force_sig_info(SIGFPE, &info, task);
893}
894
895fastcall void do_coprocessor_error(struct pt_regs * regs, long error_code)
896{
897 ignore_fpu_irq = 1;
898 math_error((void __user *)regs->eip);
899}
900
901static void simd_math_error(void __user *eip)
902{
903 struct task_struct * task;
904 siginfo_t info;
905 unsigned short mxcsr;
906
907 /*
908 * Save the info for the exception handler and clear the error.
909 */
910 task = current;
911 save_init_fpu(task);
912 task->thread.trap_no = 19;
913 task->thread.error_code = 0;
914 info.si_signo = SIGFPE;
915 info.si_errno = 0;
916 info.si_code = __SI_FAULT;
917 info.si_addr = eip;
918 /*
919 * The SIMD FPU exceptions are handled a little differently, as there
920 * is only a single status/control register. Thus, to determine which
921 * unmasked exception was caught we must mask the exception mask bits
922 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
923 */
924 mxcsr = get_fpu_mxcsr(task);
925 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
926 case 0x000:
927 default:
928 break;
929 case 0x001: /* Invalid Op */
930 info.si_code = FPE_FLTINV;
931 break;
932 case 0x002: /* Denormalize */
933 case 0x010: /* Underflow */
934 info.si_code = FPE_FLTUND;
935 break;
936 case 0x004: /* Zero Divide */
937 info.si_code = FPE_FLTDIV;
938 break;
939 case 0x008: /* Overflow */
940 info.si_code = FPE_FLTOVF;
941 break;
942 case 0x020: /* Precision */
943 info.si_code = FPE_FLTRES;
944 break;
945 }
946 force_sig_info(SIGFPE, &info, task);
947}
948
949fastcall void do_simd_coprocessor_error(struct pt_regs * regs,
950 long error_code)
951{
952 if (cpu_has_xmm) {
953 /* Handle SIMD FPU exceptions on PIII+ processors. */
954 ignore_fpu_irq = 1;
955 simd_math_error((void __user *)regs->eip);
956 } else {
957 /*
958 * Handle strange cache flush from user space exception
959 * in all other cases. This is undocumented behaviour.
960 */
961 if (regs->eflags & VM_MASK) {
962 handle_vm86_fault((struct kernel_vm86_regs *)regs,
963 error_code);
964 return;
965 }
1da177e4
LT		 966 current->thread.trap_no = 19;
967 current->thread.error_code = error_code;
4f339ecb 968 die_if_kernel("cache flush denied", regs, error_code);
1da177e4
LT		 969 force_sig(SIGSEGV, current);
970 }
971}
972
973fastcall void do_spurious_interrupt_bug(struct pt_regs * regs,
974 long error_code)
975{
976#if 0
977 /* No need to warn about this any longer. */
978 printk("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
979#endif
980}
981
982fastcall void setup_x86_bogus_stack(unsigned char * stk)
983{
984 unsigned long *switch16_ptr, *switch32_ptr;
985 struct pt_regs *regs;
986 unsigned long stack_top, stack_bot;
987 unsigned short iret_frame16_off;
988 int cpu = smp_processor_id();
989 /* reserve the space on 32bit stack for the magic switch16 pointer */
990 memmove(stk, stk + 8, sizeof(struct pt_regs));
991 switch16_ptr = (unsigned long *)(stk + sizeof(struct pt_regs));
992 regs = (struct pt_regs *)stk;
993 /* now the switch32 on 16bit stack */
994 stack_bot = (unsigned long)&per_cpu(cpu_16bit_stack, cpu);
995 stack_top = stack_bot + CPU_16BIT_STACK_SIZE;
996 switch32_ptr = (unsigned long *)(stack_top - 8);
997 iret_frame16_off = CPU_16BIT_STACK_SIZE - 8 - 20;
998 /* copy iret frame on 16bit stack */
999 memcpy((void *)(stack_bot + iret_frame16_off), &regs->eip, 20);
1000 /* fill in the switch pointers */
1001 switch16_ptr[0] = (regs->esp & 0xffff0000) | iret_frame16_off;
1002 switch16_ptr[1] = __ESPFIX_SS;
1003 switch32_ptr[0] = (unsigned long)stk + sizeof(struct pt_regs) +
1004 8 - CPU_16BIT_STACK_SIZE;
1005 switch32_ptr[1] = __KERNEL_DS;
1006}
1007
1008fastcall unsigned char * fixup_x86_bogus_stack(unsigned short sp)
1009{
1010 unsigned long *switch32_ptr;
1011 unsigned char *stack16, *stack32;
1012 unsigned long stack_top, stack_bot;
1013 int len;
1014 int cpu = smp_processor_id();
1015 stack_bot = (unsigned long)&per_cpu(cpu_16bit_stack, cpu);
1016 stack_top = stack_bot + CPU_16BIT_STACK_SIZE;
1017 switch32_ptr = (unsigned long *)(stack_top - 8);
1018 /* copy the data from 16bit stack to 32bit stack */
1019 len = CPU_16BIT_STACK_SIZE - 8 - sp;
1020 stack16 = (unsigned char *)(stack_bot + sp);
1021 stack32 = (unsigned char *)
1022 (switch32_ptr[0] + CPU_16BIT_STACK_SIZE - 8 - len);
1023 memcpy(stack32, stack16, len);
1024 return stack32;
1025}
1026
1027/*
1028 * 'math_state_restore()' saves the current math information in the
1029 * old math state array, and gets the new ones from the current task
1030 *
1031 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1032 * Don't touch unless you *really* know how it works.
1033 *
1034 * Must be called with kernel preemption disabled (in this case,
1035 * local interrupts are disabled at the call-site in entry.S).
1036 */
1037asmlinkage void math_state_restore(struct pt_regs regs)
1038{
1039 struct thread_info *thread = current_thread_info();
1040 struct task_struct *tsk = thread->task;
1041
1042 clts(); /* Allow maths ops (or we recurse) */
1043 if (!tsk_used_math(tsk))
1044 init_fpu(tsk);
1045 restore_fpu(tsk);
1046 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
1047}
1048
1049#ifndef CONFIG_MATH_EMULATION
1050
1051asmlinkage void math_emulate(long arg)
1052{
9c107805
DJ		 1053 printk(KERN_EMERG "math-emulation not enabled and no coprocessor found.\n");
1054 printk(KERN_EMERG "killing %s.\n",current->comm);
1da177e4
LT		 1055 force_sig(SIGFPE,current);
1056 schedule();
1057}
1058
1059#endif /* CONFIG_MATH_EMULATION */
1060
1061#ifdef CONFIG_X86_F00F_BUG
1062void __init trap_init_f00f_bug(void)
1063{
1064 __set_fixmap(FIX_F00F_IDT, __pa(&idt_table), PAGE_KERNEL_RO);
1065
1066 /*
1067 * Update the IDT descriptor and reload the IDT so that
1068 * it uses the read-only mapped virtual address.
1069 */
1070 idt_descr.address = fix_to_virt(FIX_F00F_IDT);
4d37e7e3 1071 load_idt(&idt_descr);
1da177e4
LT		 1072}
1073#endif
1074
1075#define _set_gate(gate_addr,type,dpl,addr,seg) \
1076do { \
1077 int __d0, __d1; \
1078 __asm__ __volatile__ ("movw %%dx,%%ax\n\t" \
1079 "movw %4,%%dx\n\t" \
1080 "movl %%eax,%0\n\t" \
1081 "movl %%edx,%1" \
1082 :"=m" (*((long *) (gate_addr))), \
1083 "=m" (*(1+(long *) (gate_addr))), "=&a" (__d0), "=&d" (__d1) \
1084 :"i" ((short) (0x8000+(dpl<<13)+(type<<8))), \
1085 "3" ((char *) (addr)),"2" ((seg) << 16)); \
1086} while (0)
1087
1088
1089/*
1090 * This needs to use 'idt_table' rather than 'idt', and
1091 * thus use the _nonmapped_ version of the IDT, as the
1092 * Pentium F0 0F bugfix can have resulted in the mapped
1093 * IDT being write-protected.
1094 */
1095void set_intr_gate(unsigned int n, void *addr)
1096{
1097 _set_gate(idt_table+n,14,0,addr,__KERNEL_CS);
1098}
1099
1100/*
1101 * This routine sets up an interrupt gate at directory privilege level 3.
1102 */
1103static inline void set_system_intr_gate(unsigned int n, void *addr)
1104{
1105 _set_gate(idt_table+n, 14, 3, addr, __KERNEL_CS);
1106}
1107
1108static void __init set_trap_gate(unsigned int n, void *addr)
1109{
1110 _set_gate(idt_table+n,15,0,addr,__KERNEL_CS);
1111}
1112
1113static void __init set_system_gate(unsigned int n, void *addr)
1114{
1115 _set_gate(idt_table+n,15,3,addr,__KERNEL_CS);
1116}
1117
1118static void __init set_task_gate(unsigned int n, unsigned int gdt_entry)
1119{
1120 _set_gate(idt_table+n,5,0,0,(gdt_entry<<3));
1121}
1122
1123
1124void __init trap_init(void)
1125{
1126#ifdef CONFIG_EISA
1127 void __iomem *p = ioremap(0x0FFFD9, 4);
1128 if (readl(p) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1129 EISA_bus = 1;
1130 }
1131 iounmap(p);
1132#endif
1133
1134#ifdef CONFIG_X86_LOCAL_APIC
1135 init_apic_mappings();
1136#endif
1137
1138 set_trap_gate(0,&divide_error);
1139 set_intr_gate(1,&debug);
1140 set_intr_gate(2,&nmi);
eb05c324 1141 set_system_intr_gate(3, &int3); /* int3/4 can be called from all */
1da177e4 1142 set_system_gate(4,&overflow);
eb05c324 1143 set_trap_gate(5,&bounds);
1da177e4
LT		 1144 set_trap_gate(6,&invalid_op);
1145 set_trap_gate(7,&device_not_available);
1146 set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS);
1147 set_trap_gate(9,&coprocessor_segment_overrun);
1148 set_trap_gate(10,&invalid_TSS);
1149 set_trap_gate(11,&segment_not_present);
1150 set_trap_gate(12,&stack_segment);
1151 set_trap_gate(13,&general_protection);
1152 set_intr_gate(14,&page_fault);
1153 set_trap_gate(15,&spurious_interrupt_bug);
1154 set_trap_gate(16,&coprocessor_error);
1155 set_trap_gate(17,&alignment_check);
1156#ifdef CONFIG_X86_MCE
1157 set_trap_gate(18,&machine_check);
1158#endif
1159 set_trap_gate(19,&simd_coprocessor_error);
1160
d43c6e80
JB		 1161 if (cpu_has_fxsr) {
1162 /*
1163 * Verify that the FXSAVE/FXRSTOR data will be 16-byte aligned.
1164 * Generates a compile-time "error: zero width for bit-field" if
1165 * the alignment is wrong.
1166 */
1167 struct fxsrAlignAssert {
1168 int _:!(offsetof(struct task_struct,
1169 thread.i387.fxsave) & 15);
1170 };
1171
1172 printk(KERN_INFO "Enabling fast FPU save and restore... ");
1173 set_in_cr4(X86_CR4_OSFXSR);
1174 printk("done.\n");
1175 }
1176 if (cpu_has_xmm) {
1177 printk(KERN_INFO "Enabling unmasked SIMD FPU exception "
1178 "support... ");
1179 set_in_cr4(X86_CR4_OSXMMEXCPT);
1180 printk("done.\n");
1181 }
1182
1da177e4
LT		 1183 set_system_gate(SYSCALL_VECTOR,&system_call);
1184
1185 /*
1186 * Should be a barrier for any external CPU state.
1187 */
1188 cpu_init();
1189
1190 trap_init_hook();
1191}
1192
1193static int __init kstack_setup(char *s)
1194{
1195 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1196 return 0;
1197}
1198__setup("kstack=", kstack_setup);
Linux 6.x block layer and io_uring tree(s)