aboutsummaryrefslogtreecommitdiffstats
path: root/arch/i386/kernel/traps.c
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@ppc970.osdl.org>2005-04-16 18:20:36 -0400
committerLinus Torvalds <torvalds@ppc970.osdl.org>2005-04-16 18:20:36 -0400
commit1da177e4c3f41524e886b7f1b8a0c1fc7321cac2 (patch)
tree0bba044c4ce775e45a88a51686b5d9f90697ea9d /arch/i386/kernel/traps.c
Linux-2.6.12-rc2v2.6.12-rc2
Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
Diffstat (limited to 'arch/i386/kernel/traps.c')
-rw-r--r--arch/i386/kernel/traps.c1084
1 files changed, 1084 insertions, 0 deletions
diff --git a/arch/i386/kernel/traps.c b/arch/i386/kernel/traps.c
new file mode 100644
index 000000000000..6c0e383915b6
--- /dev/null
+++ b/arch/i386/kernel/traps.c
@@ -0,0 +1,1084 @@
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>
30
31#ifdef CONFIG_EISA
32#include <linux/ioport.h>
33#include <linux/eisa.h>
34#endif
35
36#ifdef CONFIG_MCA
37#include <linux/mca.h>
38#endif
39
40#include <asm/processor.h>
41#include <asm/system.h>
42#include <asm/uaccess.h>
43#include <asm/io.h>
44#include <asm/atomic.h>
45#include <asm/debugreg.h>
46#include <asm/desc.h>
47#include <asm/i387.h>
48#include <asm/nmi.h>
49
50#include <asm/smp.h>
51#include <asm/arch_hooks.h>
52#include <asm/kdebug.h>
53
54#include <linux/irq.h>
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;
95struct notifier_block *i386die_chain;
96static DEFINE_SPINLOCK(die_notifier_lock);
97
98int register_die_notifier(struct notifier_block *nb)
99{
100 int err = 0;
101 unsigned long flags;
102 spin_lock_irqsave(&die_notifier_lock, flags);
103 err = notifier_chain_register(&i386die_chain, nb);
104 spin_unlock_irqrestore(&die_notifier_lock, flags);
105 return err;
106}
107
108static inline int valid_stack_ptr(struct thread_info *tinfo, void *p)
109{
110 return p > (void *)tinfo &&
111 p < (void *)tinfo + THREAD_SIZE - 3;
112}
113
114static inline unsigned long print_context_stack(struct thread_info *tinfo,
115 unsigned long *stack, unsigned long ebp)
116{
117 unsigned long addr;
118
119#ifdef CONFIG_FRAME_POINTER
120 while (valid_stack_ptr(tinfo, (void *)ebp)) {
121 addr = *(unsigned long *)(ebp + 4);
122 printk(" [<%08lx>] ", addr);
123 print_symbol("%s", addr);
124 printk("\n");
125 ebp = *(unsigned long *)ebp;
126 }
127#else
128 while (valid_stack_ptr(tinfo, stack)) {
129 addr = *stack++;
130 if (__kernel_text_address(addr)) {
131 printk(" [<%08lx>]", addr);
132 print_symbol(" %s", addr);
133 printk("\n");
134 }
135 }
136#endif
137 return ebp;
138}
139
140void show_trace(struct task_struct *task, unsigned long * stack)
141{
142 unsigned long ebp;
143
144 if (!task)
145 task = current;
146
147 if (task == current) {
148 /* Grab ebp right from our regs */
149 asm ("movl %%ebp, %0" : "=r" (ebp) : );
150 } else {
151 /* ebp is the last reg pushed by switch_to */
152 ebp = *(unsigned long *) task->thread.esp;
153 }
154
155 while (1) {
156 struct thread_info *context;
157 context = (struct thread_info *)
158 ((unsigned long)stack & (~(THREAD_SIZE - 1)));
159 ebp = print_context_stack(context, stack, ebp);
160 stack = (unsigned long*)context->previous_esp;
161 if (!stack)
162 break;
163 printk(" =======================\n");
164 }
165}
166
167void show_stack(struct task_struct *task, unsigned long *esp)
168{
169 unsigned long *stack;
170 int i;
171
172 if (esp == NULL) {
173 if (task)
174 esp = (unsigned long*)task->thread.esp;
175 else
176 esp = (unsigned long *)&esp;
177 }
178
179 stack = esp;
180 for(i = 0; i < kstack_depth_to_print; i++) {
181 if (kstack_end(stack))
182 break;
183 if (i && ((i % 8) == 0))
184 printk("\n ");
185 printk("%08lx ", *stack++);
186 }
187 printk("\nCall Trace:\n");
188 show_trace(task, esp);
189}
190
191/*
192 * The architecture-independent dump_stack generator
193 */
194void dump_stack(void)
195{
196 unsigned long stack;
197
198 show_trace(current, &stack);
199}
200
201EXPORT_SYMBOL(dump_stack);
202
203void show_registers(struct pt_regs *regs)
204{
205 int i;
206 int in_kernel = 1;
207 unsigned long esp;
208 unsigned short ss;
209
210 esp = (unsigned long) (&regs->esp);
211 ss = __KERNEL_DS;
212 if (regs->xcs & 3) {
213 in_kernel = 0;
214 esp = regs->esp;
215 ss = regs->xss & 0xffff;
216 }
217 print_modules();
218 printk("CPU: %d\nEIP: %04x:[<%08lx>] %s VLI\nEFLAGS: %08lx"
219 " (%s) \n",
220 smp_processor_id(), 0xffff & regs->xcs, regs->eip,
221 print_tainted(), regs->eflags, system_utsname.release);
222 print_symbol("EIP is at %s\n", regs->eip);
223 printk("eax: %08lx ebx: %08lx ecx: %08lx edx: %08lx\n",
224 regs->eax, regs->ebx, regs->ecx, regs->edx);
225 printk("esi: %08lx edi: %08lx ebp: %08lx esp: %08lx\n",
226 regs->esi, regs->edi, regs->ebp, esp);
227 printk("ds: %04x es: %04x ss: %04x\n",
228 regs->xds & 0xffff, regs->xes & 0xffff, ss);
229 printk("Process %s (pid: %d, threadinfo=%p task=%p)",
230 current->comm, current->pid, current_thread_info(), current);
231 /*
232 * When in-kernel, we also print out the stack and code at the
233 * time of the fault..
234 */
235 if (in_kernel) {
236 u8 *eip;
237
238 printk("\nStack: ");
239 show_stack(NULL, (unsigned long*)esp);
240
241 printk("Code: ");
242
243 eip = (u8 *)regs->eip - 43;
244 for (i = 0; i < 64; i++, eip++) {
245 unsigned char c;
246
247 if (eip < (u8 *)PAGE_OFFSET || __get_user(c, eip)) {
248 printk(" Bad EIP value.");
249 break;
250 }
251 if (eip == (u8 *)regs->eip)
252 printk("<%02x> ", c);
253 else
254 printk("%02x ", c);
255 }
256 }
257 printk("\n");
258}
259
260static void handle_BUG(struct pt_regs *regs)
261{
262 unsigned short ud2;
263 unsigned short line;
264 char *file;
265 char c;
266 unsigned long eip;
267
268 if (regs->xcs & 3)
269 goto no_bug; /* Not in kernel */
270
271 eip = regs->eip;
272
273 if (eip < PAGE_OFFSET)
274 goto no_bug;
275 if (__get_user(ud2, (unsigned short *)eip))
276 goto no_bug;
277 if (ud2 != 0x0b0f)
278 goto no_bug;
279 if (__get_user(line, (unsigned short *)(eip + 2)))
280 goto bug;
281 if (__get_user(file, (char **)(eip + 4)) ||
282 (unsigned long)file < PAGE_OFFSET || __get_user(c, file))
283 file = "<bad filename>";
284
285 printk("------------[ cut here ]------------\n");
286 printk(KERN_ALERT "kernel BUG at %s:%d!\n", file, line);
287
288no_bug:
289 return;
290
291 /* Here we know it was a BUG but file-n-line is unavailable */
292bug:
293 printk("Kernel BUG\n");
294}
295
296void die(const char * str, struct pt_regs * regs, long err)
297{
298 static struct {
299 spinlock_t lock;
300 u32 lock_owner;
301 int lock_owner_depth;
302 } die = {
303 .lock = SPIN_LOCK_UNLOCKED,
304 .lock_owner = -1,
305 .lock_owner_depth = 0
306 };
307 static int die_counter;
308
309 if (die.lock_owner != _smp_processor_id()) {
310 console_verbose();
311 spin_lock_irq(&die.lock);
312 die.lock_owner = smp_processor_id();
313 die.lock_owner_depth = 0;
314 bust_spinlocks(1);
315 }
316
317 if (++die.lock_owner_depth < 3) {
318 int nl = 0;
319 handle_BUG(regs);
320 printk(KERN_ALERT "%s: %04lx [#%d]\n", str, err & 0xffff, ++die_counter);
321#ifdef CONFIG_PREEMPT
322 printk("PREEMPT ");
323 nl = 1;
324#endif
325#ifdef CONFIG_SMP
326 printk("SMP ");
327 nl = 1;
328#endif
329#ifdef CONFIG_DEBUG_PAGEALLOC
330 printk("DEBUG_PAGEALLOC");
331 nl = 1;
332#endif
333 if (nl)
334 printk("\n");
335 notify_die(DIE_OOPS, (char *)str, regs, err, 255, SIGSEGV);
336 show_registers(regs);
337 } else
338 printk(KERN_ERR "Recursive die() failure, output suppressed\n");
339
340 bust_spinlocks(0);
341 die.lock_owner = -1;
342 spin_unlock_irq(&die.lock);
343 if (in_interrupt())
344 panic("Fatal exception in interrupt");
345
346 if (panic_on_oops) {
347 printk(KERN_EMERG "Fatal exception: panic in 5 seconds\n");
348 ssleep(5);
349 panic("Fatal exception");
350 }
351 do_exit(SIGSEGV);
352}
353
354static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
355{
356 if (!(regs->eflags & VM_MASK) && !(3 & regs->xcs))
357 die(str, regs, err);
358}
359
360static void do_trap(int trapnr, int signr, char *str, int vm86,
361 struct pt_regs * regs, long error_code, siginfo_t *info)
362{
363 if (regs->eflags & VM_MASK) {
364 if (vm86)
365 goto vm86_trap;
366 goto trap_signal;
367 }
368
369 if (!(regs->xcs & 3))
370 goto kernel_trap;
371
372 trap_signal: {
373 struct task_struct *tsk = current;
374 tsk->thread.error_code = error_code;
375 tsk->thread.trap_no = trapnr;
376 if (info)
377 force_sig_info(signr, info, tsk);
378 else
379 force_sig(signr, tsk);
380 return;
381 }
382
383 kernel_trap: {
384 if (!fixup_exception(regs))
385 die(str, regs, error_code);
386 return;
387 }
388
389 vm86_trap: {
390 int ret = handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, trapnr);
391 if (ret) goto trap_signal;
392 return;
393 }
394}
395
396#define DO_ERROR(trapnr, signr, str, name) \
397fastcall void do_##name(struct pt_regs * regs, long error_code) \
398{ \
399 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
400 == NOTIFY_STOP) \
401 return; \
402 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
403}
404
405#define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
406fastcall void do_##name(struct pt_regs * regs, long error_code) \
407{ \
408 siginfo_t info; \
409 info.si_signo = signr; \
410 info.si_errno = 0; \
411 info.si_code = sicode; \
412 info.si_addr = (void __user *)siaddr; \
413 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
414 == NOTIFY_STOP) \
415 return; \
416 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
417}
418
419#define DO_VM86_ERROR(trapnr, signr, str, name) \
420fastcall void do_##name(struct pt_regs * regs, long error_code) \
421{ \
422 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
423 == NOTIFY_STOP) \
424 return; \
425 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
426}
427
428#define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
429fastcall void do_##name(struct pt_regs * regs, long error_code) \
430{ \
431 siginfo_t info; \
432 info.si_signo = signr; \
433 info.si_errno = 0; \
434 info.si_code = sicode; \
435 info.si_addr = (void __user *)siaddr; \
436 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
437 == NOTIFY_STOP) \
438 return; \
439 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
440}
441
442DO_VM86_ERROR_INFO( 0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->eip)
443#ifndef CONFIG_KPROBES
444DO_VM86_ERROR( 3, SIGTRAP, "int3", int3)
445#endif
446DO_VM86_ERROR( 4, SIGSEGV, "overflow", overflow)
447DO_VM86_ERROR( 5, SIGSEGV, "bounds", bounds)
448DO_ERROR_INFO( 6, SIGILL, "invalid operand", invalid_op, ILL_ILLOPN, regs->eip)
449DO_ERROR( 9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
450DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
451DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
452DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
453DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
454
455fastcall void do_general_protection(struct pt_regs * regs, long error_code)
456{
457 int cpu = get_cpu();
458 struct tss_struct *tss = &per_cpu(init_tss, cpu);
459 struct thread_struct *thread = &current->thread;
460
461 /*
462 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
463 * invalid offset set (the LAZY one) and the faulting thread has
464 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
465 * and we set the offset field correctly. Then we let the CPU to
466 * restart the faulting instruction.
467 */
468 if (tss->io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
469 thread->io_bitmap_ptr) {
470 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
471 thread->io_bitmap_max);
472 /*
473 * If the previously set map was extending to higher ports
474 * than the current one, pad extra space with 0xff (no access).
475 */
476 if (thread->io_bitmap_max < tss->io_bitmap_max)
477 memset((char *) tss->io_bitmap +
478 thread->io_bitmap_max, 0xff,
479 tss->io_bitmap_max - thread->io_bitmap_max);
480 tss->io_bitmap_max = thread->io_bitmap_max;
481 tss->io_bitmap_base = IO_BITMAP_OFFSET;
482 put_cpu();
483 return;
484 }
485 put_cpu();
486
487 if (regs->eflags & VM_MASK)
488 goto gp_in_vm86;
489
490 if (!(regs->xcs & 3))
491 goto gp_in_kernel;
492
493 current->thread.error_code = error_code;
494 current->thread.trap_no = 13;
495 force_sig(SIGSEGV, current);
496 return;
497
498gp_in_vm86:
499 local_irq_enable();
500 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
501 return;
502
503gp_in_kernel:
504 if (!fixup_exception(regs)) {
505 if (notify_die(DIE_GPF, "general protection fault", regs,
506 error_code, 13, SIGSEGV) == NOTIFY_STOP)
507 return;
508 die("general protection fault", regs, error_code);
509 }
510}
511
512static void mem_parity_error(unsigned char reason, struct pt_regs * regs)
513{
514 printk("Uhhuh. NMI received. Dazed and confused, but trying to continue\n");
515 printk("You probably have a hardware problem with your RAM chips\n");
516
517 /* Clear and disable the memory parity error line. */
518 clear_mem_error(reason);
519}
520
521static void io_check_error(unsigned char reason, struct pt_regs * regs)
522{
523 unsigned long i;
524
525 printk("NMI: IOCK error (debug interrupt?)\n");
526 show_registers(regs);
527
528 /* Re-enable the IOCK line, wait for a few seconds */
529 reason = (reason & 0xf) | 8;
530 outb(reason, 0x61);
531 i = 2000;
532 while (--i) udelay(1000);
533 reason &= ~8;
534 outb(reason, 0x61);
535}
536
537static void unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
538{
539#ifdef CONFIG_MCA
540 /* Might actually be able to figure out what the guilty party
541 * is. */
542 if( MCA_bus ) {
543 mca_handle_nmi();
544 return;
545 }
546#endif
547 printk("Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
548 reason, smp_processor_id());
549 printk("Dazed and confused, but trying to continue\n");
550 printk("Do you have a strange power saving mode enabled?\n");
551}
552
553static DEFINE_SPINLOCK(nmi_print_lock);
554
555void die_nmi (struct pt_regs *regs, const char *msg)
556{
557 spin_lock(&nmi_print_lock);
558 /*
559 * We are in trouble anyway, lets at least try
560 * to get a message out.
561 */
562 bust_spinlocks(1);
563 printk(msg);
564 printk(" on CPU%d, eip %08lx, registers:\n",
565 smp_processor_id(), regs->eip);
566 show_registers(regs);
567 printk("console shuts up ...\n");
568 console_silent();
569 spin_unlock(&nmi_print_lock);
570 bust_spinlocks(0);
571 do_exit(SIGSEGV);
572}
573
574static void default_do_nmi(struct pt_regs * regs)
575{
576 unsigned char reason = 0;
577
578 /* Only the BSP gets external NMIs from the system. */
579 if (!smp_processor_id())
580 reason = get_nmi_reason();
581
582 if (!(reason & 0xc0)) {
583 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 0, SIGINT)
584 == NOTIFY_STOP)
585 return;
586#ifdef CONFIG_X86_LOCAL_APIC
587 /*
588 * Ok, so this is none of the documented NMI sources,
589 * so it must be the NMI watchdog.
590 */
591 if (nmi_watchdog) {
592 nmi_watchdog_tick(regs);
593 return;
594 }
595#endif
596 unknown_nmi_error(reason, regs);
597 return;
598 }
599 if (notify_die(DIE_NMI, "nmi", regs, reason, 0, SIGINT) == NOTIFY_STOP)
600 return;
601 if (reason & 0x80)
602 mem_parity_error(reason, regs);
603 if (reason & 0x40)
604 io_check_error(reason, regs);
605 /*
606 * Reassert NMI in case it became active meanwhile
607 * as it's edge-triggered.
608 */
609 reassert_nmi();
610}
611
612static int dummy_nmi_callback(struct pt_regs * regs, int cpu)
613{
614 return 0;
615}
616
617static nmi_callback_t nmi_callback = dummy_nmi_callback;
618
619fastcall void do_nmi(struct pt_regs * regs, long error_code)
620{
621 int cpu;
622
623 nmi_enter();
624
625 cpu = smp_processor_id();
626 ++nmi_count(cpu);
627
628 if (!nmi_callback(regs, cpu))
629 default_do_nmi(regs);
630
631 nmi_exit();
632}
633
634void set_nmi_callback(nmi_callback_t callback)
635{
636 nmi_callback = callback;
637}
638
639void unset_nmi_callback(void)
640{
641 nmi_callback = dummy_nmi_callback;
642}
643
644#ifdef CONFIG_KPROBES
645fastcall int do_int3(struct pt_regs *regs, long error_code)
646{
647 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
648 == NOTIFY_STOP)
649 return 1;
650 /* This is an interrupt gate, because kprobes wants interrupts
651 disabled. Normal trap handlers don't. */
652 restore_interrupts(regs);
653 do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL);
654 return 0;
655}
656#endif
657
658/*
659 * Our handling of the processor debug registers is non-trivial.
660 * We do not clear them on entry and exit from the kernel. Therefore
661 * it is possible to get a watchpoint trap here from inside the kernel.
662 * However, the code in ./ptrace.c has ensured that the user can
663 * only set watchpoints on userspace addresses. Therefore the in-kernel
664 * watchpoint trap can only occur in code which is reading/writing
665 * from user space. Such code must not hold kernel locks (since it
666 * can equally take a page fault), therefore it is safe to call
667 * force_sig_info even though that claims and releases locks.
668 *
669 * Code in ./signal.c ensures that the debug control register
670 * is restored before we deliver any signal, and therefore that
671 * user code runs with the correct debug control register even though
672 * we clear it here.
673 *
674 * Being careful here means that we don't have to be as careful in a
675 * lot of more complicated places (task switching can be a bit lazy
676 * about restoring all the debug state, and ptrace doesn't have to
677 * find every occurrence of the TF bit that could be saved away even
678 * by user code)
679 */
680fastcall void do_debug(struct pt_regs * regs, long error_code)
681{
682 unsigned int condition;
683 struct task_struct *tsk = current;
684
685 __asm__ __volatile__("movl %%db6,%0" : "=r" (condition));
686
687 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
688 SIGTRAP) == NOTIFY_STOP)
689 return;
690 /* It's safe to allow irq's after DR6 has been saved */
691 if (regs->eflags & X86_EFLAGS_IF)
692 local_irq_enable();
693
694 /* Mask out spurious debug traps due to lazy DR7 setting */
695 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
696 if (!tsk->thread.debugreg[7])
697 goto clear_dr7;
698 }
699
700 if (regs->eflags & VM_MASK)
701 goto debug_vm86;
702
703 /* Save debug status register where ptrace can see it */
704 tsk->thread.debugreg[6] = condition;
705
706 /*
707 * Single-stepping through TF: make sure we ignore any events in
708 * kernel space (but re-enable TF when returning to user mode).
709 */
710 if (condition & DR_STEP) {
711 /*
712 * We already checked v86 mode above, so we can
713 * check for kernel mode by just checking the CPL
714 * of CS.
715 */
716 if ((regs->xcs & 3) == 0)
717 goto clear_TF_reenable;
718 }
719
720 /* Ok, finally something we can handle */
721 send_sigtrap(tsk, regs, error_code);
722
723 /* Disable additional traps. They'll be re-enabled when
724 * the signal is delivered.
725 */
726clear_dr7:
727 __asm__("movl %0,%%db7"
728 : /* no output */
729 : "r" (0));
730 return;
731
732debug_vm86:
733 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
734 return;
735
736clear_TF_reenable:
737 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
738 regs->eflags &= ~TF_MASK;
739 return;
740}
741
742/*
743 * Note that we play around with the 'TS' bit in an attempt to get
744 * the correct behaviour even in the presence of the asynchronous
745 * IRQ13 behaviour
746 */
747void math_error(void __user *eip)
748{
749 struct task_struct * task;
750 siginfo_t info;
751 unsigned short cwd, swd;
752
753 /*
754 * Save the info for the exception handler and clear the error.
755 */
756 task = current;
757 save_init_fpu(task);
758 task->thread.trap_no = 16;
759 task->thread.error_code = 0;
760 info.si_signo = SIGFPE;
761 info.si_errno = 0;
762 info.si_code = __SI_FAULT;
763 info.si_addr = eip;
764 /*
765 * (~cwd & swd) will mask out exceptions that are not set to unmasked
766 * status. 0x3f is the exception bits in these regs, 0x200 is the
767 * C1 reg you need in case of a stack fault, 0x040 is the stack
768 * fault bit. We should only be taking one exception at a time,
769 * so if this combination doesn't produce any single exception,
770 * then we have a bad program that isn't syncronizing its FPU usage
771 * and it will suffer the consequences since we won't be able to
772 * fully reproduce the context of the exception
773 */
774 cwd = get_fpu_cwd(task);
775 swd = get_fpu_swd(task);
776 switch (((~cwd) & swd & 0x3f) | (swd & 0x240)) {
777 case 0x000:
778 default:
779 break;
780 case 0x001: /* Invalid Op */
781 case 0x041: /* Stack Fault */
782 case 0x241: /* Stack Fault | Direction */
783 info.si_code = FPE_FLTINV;
784 /* Should we clear the SF or let user space do it ???? */
785 break;
786 case 0x002: /* Denormalize */
787 case 0x010: /* Underflow */
788 info.si_code = FPE_FLTUND;
789 break;
790 case 0x004: /* Zero Divide */
791 info.si_code = FPE_FLTDIV;
792 break;
793 case 0x008: /* Overflow */
794 info.si_code = FPE_FLTOVF;
795 break;
796 case 0x020: /* Precision */
797 info.si_code = FPE_FLTRES;
798 break;
799 }
800 force_sig_info(SIGFPE, &info, task);
801}
802
803fastcall void do_coprocessor_error(struct pt_regs * regs, long error_code)
804{
805 ignore_fpu_irq = 1;
806 math_error((void __user *)regs->eip);
807}
808
809static void simd_math_error(void __user *eip)
810{
811 struct task_struct * task;
812 siginfo_t info;
813 unsigned short mxcsr;
814
815 /*
816 * Save the info for the exception handler and clear the error.
817 */
818 task = current;
819 save_init_fpu(task);
820 task->thread.trap_no = 19;
821 task->thread.error_code = 0;
822 info.si_signo = SIGFPE;
823 info.si_errno = 0;
824 info.si_code = __SI_FAULT;
825 info.si_addr = eip;
826 /*
827 * The SIMD FPU exceptions are handled a little differently, as there
828 * is only a single status/control register. Thus, to determine which
829 * unmasked exception was caught we must mask the exception mask bits
830 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
831 */
832 mxcsr = get_fpu_mxcsr(task);
833 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
834 case 0x000:
835 default:
836 break;
837 case 0x001: /* Invalid Op */
838 info.si_code = FPE_FLTINV;
839 break;
840 case 0x002: /* Denormalize */
841 case 0x010: /* Underflow */
842 info.si_code = FPE_FLTUND;
843 break;
844 case 0x004: /* Zero Divide */
845 info.si_code = FPE_FLTDIV;
846 break;
847 case 0x008: /* Overflow */
848 info.si_code = FPE_FLTOVF;
849 break;
850 case 0x020: /* Precision */
851 info.si_code = FPE_FLTRES;
852 break;
853 }
854 force_sig_info(SIGFPE, &info, task);
855}
856
857fastcall void do_simd_coprocessor_error(struct pt_regs * regs,
858 long error_code)
859{
860 if (cpu_has_xmm) {
861 /* Handle SIMD FPU exceptions on PIII+ processors. */
862 ignore_fpu_irq = 1;
863 simd_math_error((void __user *)regs->eip);
864 } else {
865 /*
866 * Handle strange cache flush from user space exception
867 * in all other cases. This is undocumented behaviour.
868 */
869 if (regs->eflags & VM_MASK) {
870 handle_vm86_fault((struct kernel_vm86_regs *)regs,
871 error_code);
872 return;
873 }
874 die_if_kernel("cache flush denied", regs, error_code);
875 current->thread.trap_no = 19;
876 current->thread.error_code = error_code;
877 force_sig(SIGSEGV, current);
878 }
879}
880
881fastcall void do_spurious_interrupt_bug(struct pt_regs * regs,
882 long error_code)
883{
884#if 0
885 /* No need to warn about this any longer. */
886 printk("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
887#endif
888}
889
890fastcall void setup_x86_bogus_stack(unsigned char * stk)
891{
892 unsigned long *switch16_ptr, *switch32_ptr;
893 struct pt_regs *regs;
894 unsigned long stack_top, stack_bot;
895 unsigned short iret_frame16_off;
896 int cpu = smp_processor_id();
897 /* reserve the space on 32bit stack for the magic switch16 pointer */
898 memmove(stk, stk + 8, sizeof(struct pt_regs));
899 switch16_ptr = (unsigned long *)(stk + sizeof(struct pt_regs));
900 regs = (struct pt_regs *)stk;
901 /* now the switch32 on 16bit stack */
902 stack_bot = (unsigned long)&per_cpu(cpu_16bit_stack, cpu);
903 stack_top = stack_bot + CPU_16BIT_STACK_SIZE;
904 switch32_ptr = (unsigned long *)(stack_top - 8);
905 iret_frame16_off = CPU_16BIT_STACK_SIZE - 8 - 20;
906 /* copy iret frame on 16bit stack */
907 memcpy((void *)(stack_bot + iret_frame16_off), &regs->eip, 20);
908 /* fill in the switch pointers */
909 switch16_ptr[0] = (regs->esp & 0xffff0000) | iret_frame16_off;
910 switch16_ptr[1] = __ESPFIX_SS;
911 switch32_ptr[0] = (unsigned long)stk + sizeof(struct pt_regs) +
912 8 - CPU_16BIT_STACK_SIZE;
913 switch32_ptr[1] = __KERNEL_DS;
914}
915
916fastcall unsigned char * fixup_x86_bogus_stack(unsigned short sp)
917{
918 unsigned long *switch32_ptr;
919 unsigned char *stack16, *stack32;
920 unsigned long stack_top, stack_bot;
921 int len;
922 int cpu = smp_processor_id();
923 stack_bot = (unsigned long)&per_cpu(cpu_16bit_stack, cpu);
924 stack_top = stack_bot + CPU_16BIT_STACK_SIZE;
925 switch32_ptr = (unsigned long *)(stack_top - 8);
926 /* copy the data from 16bit stack to 32bit stack */
927 len = CPU_16BIT_STACK_SIZE - 8 - sp;
928 stack16 = (unsigned char *)(stack_bot + sp);
929 stack32 = (unsigned char *)
930 (switch32_ptr[0] + CPU_16BIT_STACK_SIZE - 8 - len);
931 memcpy(stack32, stack16, len);
932 return stack32;
933}
934
935/*
936 * 'math_state_restore()' saves the current math information in the
937 * old math state array, and gets the new ones from the current task
938 *
939 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
940 * Don't touch unless you *really* know how it works.
941 *
942 * Must be called with kernel preemption disabled (in this case,
943 * local interrupts are disabled at the call-site in entry.S).
944 */
945asmlinkage void math_state_restore(struct pt_regs regs)
946{
947 struct thread_info *thread = current_thread_info();
948 struct task_struct *tsk = thread->task;
949
950 clts(); /* Allow maths ops (or we recurse) */
951 if (!tsk_used_math(tsk))
952 init_fpu(tsk);
953 restore_fpu(tsk);
954 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
955}
956
957#ifndef CONFIG_MATH_EMULATION
958
959asmlinkage void math_emulate(long arg)
960{
961 printk("math-emulation not enabled and no coprocessor found.\n");
962 printk("killing %s.\n",current->comm);
963 force_sig(SIGFPE,current);
964 schedule();
965}
966
967#endif /* CONFIG_MATH_EMULATION */
968
969#ifdef CONFIG_X86_F00F_BUG
970void __init trap_init_f00f_bug(void)
971{
972 __set_fixmap(FIX_F00F_IDT, __pa(&idt_table), PAGE_KERNEL_RO);
973
974 /*
975 * Update the IDT descriptor and reload the IDT so that
976 * it uses the read-only mapped virtual address.
977 */
978 idt_descr.address = fix_to_virt(FIX_F00F_IDT);
979 __asm__ __volatile__("lidt %0" : : "m" (idt_descr));
980}
981#endif
982
983#define _set_gate(gate_addr,type,dpl,addr,seg) \
984do { \
985 int __d0, __d1; \
986 __asm__ __volatile__ ("movw %%dx,%%ax\n\t" \
987 "movw %4,%%dx\n\t" \
988 "movl %%eax,%0\n\t" \
989 "movl %%edx,%1" \
990 :"=m" (*((long *) (gate_addr))), \
991 "=m" (*(1+(long *) (gate_addr))), "=&a" (__d0), "=&d" (__d1) \
992 :"i" ((short) (0x8000+(dpl<<13)+(type<<8))), \
993 "3" ((char *) (addr)),"2" ((seg) << 16)); \
994} while (0)
995
996
997/*
998 * This needs to use 'idt_table' rather than 'idt', and
999 * thus use the _nonmapped_ version of the IDT, as the
1000 * Pentium F0 0F bugfix can have resulted in the mapped
1001 * IDT being write-protected.
1002 */
1003void set_intr_gate(unsigned int n, void *addr)
1004{
1005 _set_gate(idt_table+n,14,0,addr,__KERNEL_CS);
1006}
1007
1008/*
1009 * This routine sets up an interrupt gate at directory privilege level 3.
1010 */
1011static inline void set_system_intr_gate(unsigned int n, void *addr)
1012{
1013 _set_gate(idt_table+n, 14, 3, addr, __KERNEL_CS);
1014}
1015
1016static void __init set_trap_gate(unsigned int n, void *addr)
1017{
1018 _set_gate(idt_table+n,15,0,addr,__KERNEL_CS);
1019}
1020
1021static void __init set_system_gate(unsigned int n, void *addr)
1022{
1023 _set_gate(idt_table+n,15,3,addr,__KERNEL_CS);
1024}
1025
1026static void __init set_task_gate(unsigned int n, unsigned int gdt_entry)
1027{
1028 _set_gate(idt_table+n,5,0,0,(gdt_entry<<3));
1029}
1030
1031
1032void __init trap_init(void)
1033{
1034#ifdef CONFIG_EISA
1035 void __iomem *p = ioremap(0x0FFFD9, 4);
1036 if (readl(p) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1037 EISA_bus = 1;
1038 }
1039 iounmap(p);
1040#endif
1041
1042#ifdef CONFIG_X86_LOCAL_APIC
1043 init_apic_mappings();
1044#endif
1045
1046 set_trap_gate(0,&divide_error);
1047 set_intr_gate(1,&debug);
1048 set_intr_gate(2,&nmi);
1049 set_system_intr_gate(3, &int3); /* int3-5 can be called from all */
1050 set_system_gate(4,&overflow);
1051 set_system_gate(5,&bounds);
1052 set_trap_gate(6,&invalid_op);
1053 set_trap_gate(7,&device_not_available);
1054 set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS);
1055 set_trap_gate(9,&coprocessor_segment_overrun);
1056 set_trap_gate(10,&invalid_TSS);
1057 set_trap_gate(11,&segment_not_present);
1058 set_trap_gate(12,&stack_segment);
1059 set_trap_gate(13,&general_protection);
1060 set_intr_gate(14,&page_fault);
1061 set_trap_gate(15,&spurious_interrupt_bug);
1062 set_trap_gate(16,&coprocessor_error);
1063 set_trap_gate(17,&alignment_check);
1064#ifdef CONFIG_X86_MCE
1065 set_trap_gate(18,&machine_check);
1066#endif
1067 set_trap_gate(19,&simd_coprocessor_error);
1068
1069 set_system_gate(SYSCALL_VECTOR,&system_call);
1070
1071 /*
1072 * Should be a barrier for any external CPU state.
1073 */
1074 cpu_init();
1075
1076 trap_init_hook();
1077}
1078
1079static int __init kstack_setup(char *s)
1080{
1081 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1082 return 0;
1083}
1084__setup("kstack=", kstack_setup);