Linux kernel & device driver programming

Cross-Referenced Linux and Device Driver Code

[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ]
Version: [ 2.6.11.8 ] [ 2.6.25 ] [ 2.6.25.8 ] [ 2.6.31.13 ] Architecture: [ i386 ]
  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 
 59 asmlinkage int system_call(void);
 60 
 61 struct desc_struct default_ldt[] = { { 0, 0 }, { 0, 0 }, { 0, 0 },
 62                 { 0, 0 }, { 0, 0 } };
 63 
 64 /* Do we ignore FPU interrupts ? */
 65 char 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  */
 72 struct desc_struct idt_table[256] __attribute__((__section__(".data.idt"))) = { {0, 0}, };
 73 
 74 asmlinkage void divide_error(void);
 75 asmlinkage void debug(void);
 76 asmlinkage void nmi(void);
 77 asmlinkage void int3(void);
 78 asmlinkage void overflow(void);
 79 asmlinkage void bounds(void);
 80 asmlinkage void invalid_op(void);
 81 asmlinkage void device_not_available(void);
 82 asmlinkage void coprocessor_segment_overrun(void);
 83 asmlinkage void invalid_TSS(void);
 84 asmlinkage void segment_not_present(void);
 85 asmlinkage void stack_segment(void);
 86 asmlinkage void general_protection(void);
 87 asmlinkage void page_fault(void);
 88 asmlinkage void coprocessor_error(void);
 89 asmlinkage void simd_coprocessor_error(void);
 90 asmlinkage void alignment_check(void);
 91 asmlinkage void spurious_interrupt_bug(void);
 92 asmlinkage void machine_check(void);
 93 
 94 static int kstack_depth_to_print = 24;
 95 struct notifier_block *i386die_chain;
 96 static DEFINE_SPINLOCK(die_notifier_lock);
 97 
 98 int 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 
108 static 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 
114 static 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 
140 void 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 
167 void 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  */
194 void dump_stack(void)
195 {
196         unsigned long stack;
197 
198         show_trace(current, &stack);
199 }
200 
201 EXPORT_SYMBOL(dump_stack);
202 
203 void 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 
260 static 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 
288 no_bug:
289         return;
290 
291         /* Here we know it was a BUG but file-n-line is unavailable */
292 bug:
293         printk("Kernel BUG\n");
294 }
295 
296 void 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                 set_current_state(TASK_UNINTERRUPTIBLE);
349                 schedule_timeout(5 * HZ);
350                 panic("Fatal exception");
351         }
352         do_exit(SIGSEGV);
353 }
354 
355 static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
356 {
357         if (!(regs->eflags & VM_MASK) && !(3 & regs->xcs))
358                 die(str, regs, err);
359 }
360 
361 static void do_trap(int trapnr, int signr, char *str, int vm86,
362                            struct pt_regs * regs, long error_code, siginfo_t *info)
363 {
364         if (regs->eflags & VM_MASK) {
365                 if (vm86)
366                         goto vm86_trap;
367                 goto trap_signal;
368         }
369 
370         if (!(regs->xcs & 3))
371                 goto kernel_trap;
372 
373         trap_signal: {
374                 struct task_struct *tsk = current;
375                 tsk->thread.error_code = error_code;
376                 tsk->thread.trap_no = trapnr;
377                 if (info)
378                         force_sig_info(signr, info, tsk);
379                 else
380                         force_sig(signr, tsk);
381                 return;
382         }
383 
384         kernel_trap: {
385                 if (!fixup_exception(regs))
386                         die(str, regs, error_code);
387                 return;
388         }
389 
390         vm86_trap: {
391                 int ret = handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, trapnr);
392                 if (ret) goto trap_signal;
393                 return;
394         }
395 }
396 
397 #define DO_ERROR(trapnr, signr, str, name) \
398 fastcall void do_##name(struct pt_regs * regs, long error_code) \
399 { \
400         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
401                                                 == NOTIFY_STOP) \
402                 return; \
403         do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
404 }
405 
406 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
407 fastcall void do_##name(struct pt_regs * regs, long error_code) \
408 { \
409         siginfo_t info; \
410         info.si_signo = signr; \
411         info.si_errno = 0; \
412         info.si_code = sicode; \
413         info.si_addr = (void __user *)siaddr; \
414         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
415                                                 == NOTIFY_STOP) \
416                 return; \
417         do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
418 }
419 
420 #define DO_VM86_ERROR(trapnr, signr, str, name) \
421 fastcall void do_##name(struct pt_regs * regs, long error_code) \
422 { \
423         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
424                                                 == NOTIFY_STOP) \
425                 return; \
426         do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
427 }
428 
429 #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
430 fastcall void do_##name(struct pt_regs * regs, long error_code) \
431 { \
432         siginfo_t info; \
433         info.si_signo = signr; \
434         info.si_errno = 0; \
435         info.si_code = sicode; \
436         info.si_addr = (void __user *)siaddr; \
437         if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
438                                                 == NOTIFY_STOP) \
439                 return; \
440         do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
441 }
442 
443 DO_VM86_ERROR_INFO( 0, SIGFPE,  "divide error", divide_error, FPE_INTDIV, regs->eip)
444 #ifndef CONFIG_KPROBES
445 DO_VM86_ERROR( 3, SIGTRAP, "int3", int3)
446 #endif
447 DO_VM86_ERROR( 4, SIGSEGV, "overflow", overflow)
448 DO_VM86_ERROR( 5, SIGSEGV, "bounds", bounds)
449 DO_ERROR_INFO( 6, SIGILL,  "invalid operand", invalid_op, ILL_ILLOPN, regs->eip)
450 DO_ERROR( 9, SIGFPE,  "coprocessor segment overrun", coprocessor_segment_overrun)
451 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
452 DO_ERROR(11, SIGBUS,  "segment not present", segment_not_present)
453 DO_ERROR(12, SIGBUS,  "stack segment", stack_segment)
454 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
455 
456 fastcall void do_general_protection(struct pt_regs * regs, long error_code)
457 {
458         int cpu = get_cpu();
459         struct tss_struct *tss = &per_cpu(init_tss, cpu);
460         struct thread_struct *thread = &current->thread;
461 
462         /*
463          * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
464          * invalid offset set (the LAZY one) and the faulting thread has
465          * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
466          * and we set the offset field correctly. Then we let the CPU to
467          * restart the faulting instruction.
468          */
469         if (tss->io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
470             thread->io_bitmap_ptr) {
471                 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
472                        thread->io_bitmap_max);
473                 /*
474                  * If the previously set map was extending to higher ports
475                  * than the current one, pad extra space with 0xff (no access).
476                  */
477                 if (thread->io_bitmap_max < tss->io_bitmap_max)
478                         memset((char *) tss->io_bitmap +
479                                 thread->io_bitmap_max, 0xff,
480                                 tss->io_bitmap_max - thread->io_bitmap_max);
481                 tss->io_bitmap_max = thread->io_bitmap_max;
482                 tss->io_bitmap_base = IO_BITMAP_OFFSET;
483                 put_cpu();
484                 return;
485         }
486         put_cpu();
487 
488         if (regs->eflags & VM_MASK)
489                 goto gp_in_vm86;
490 
491         if (!(regs->xcs & 3))
492                 goto gp_in_kernel;
493 
494         current->thread.error_code = error_code;
495         current->thread.trap_no = 13;
496         force_sig(SIGSEGV, current);
497         return;
498 
499 gp_in_vm86:
500         local_irq_enable();
501         handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
502         return;
503 
504 gp_in_kernel:
505         if (!fixup_exception(regs)) {
506                 if (notify_die(DIE_GPF, "general protection fault", regs,
507                                 error_code, 13, SIGSEGV) == NOTIFY_STOP)
508                         return;
509                 die("general protection fault", regs, error_code);
510         }
511 }
512 
513 static void mem_parity_error(unsigned char reason, struct pt_regs * regs)
514 {
515         printk("Uhhuh. NMI received. Dazed and confused, but trying to continue\n");
516         printk("You probably have a hardware problem with your RAM chips\n");
517 
518         /* Clear and disable the memory parity error line. */
519         clear_mem_error(reason);
520 }
521 
522 static void io_check_error(unsigned char reason, struct pt_regs * regs)
523 {
524         unsigned long i;
525 
526         printk("NMI: IOCK error (debug interrupt?)\n");
527         show_registers(regs);
528 
529         /* Re-enable the IOCK line, wait for a few seconds */
530         reason = (reason & 0xf) | 8;
531         outb(reason, 0x61);
532         i = 2000;
533         while (--i) udelay(1000);
534         reason &= ~8;
535         outb(reason, 0x61);
536 }
537 
538 static void unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
539 {
540 #ifdef CONFIG_MCA
541         /* Might actually be able to figure out what the guilty party
542         * is. */
543         if( MCA_bus ) {
544                 mca_handle_nmi();
545                 return;
546         }
547 #endif
548         printk("Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
549                 reason, smp_processor_id());
550         printk("Dazed and confused, but trying to continue\n");
551         printk("Do you have a strange power saving mode enabled?\n");
552 }
553 
554 static DEFINE_SPINLOCK(nmi_print_lock);
555 
556 void die_nmi (struct pt_regs *regs, const char *msg)
557 {
558         spin_lock(&nmi_print_lock);
559         /*
560         * We are in trouble anyway, lets at least try
561         * to get a message out.
562         */
563         bust_spinlocks(1);
564         printk(msg);
565         printk(" on CPU%d, eip %08lx, registers:\n",
566                 smp_processor_id(), regs->eip);
567         show_registers(regs);
568         printk("console shuts up ...\n");
569         console_silent();
570         spin_unlock(&nmi_print_lock);
571         bust_spinlocks(0);
572         do_exit(SIGSEGV);
573 }
574 
575 static void default_do_nmi(struct pt_regs * regs)
576 {
577         unsigned char reason = 0;
578 
579         /* Only the BSP gets external NMIs from the system.  */
580         if (!smp_processor_id())
581                 reason = get_nmi_reason();
582  
583         if (!(reason & 0xc0)) {
584                 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 0, SIGINT)
585                                                         == NOTIFY_STOP)
586                         return;
587 #ifdef CONFIG_X86_LOCAL_APIC
588                 /*
589                  * Ok, so this is none of the documented NMI sources,
590                  * so it must be the NMI watchdog.
591                  */
592                 if (nmi_watchdog) {
593                         nmi_watchdog_tick(regs);
594                         return;
595                 }
596 #endif
597                 unknown_nmi_error(reason, regs);
598                 return;
599         }
600         if (notify_die(DIE_NMI, "nmi", regs, reason, 0, SIGINT) == NOTIFY_STOP)
601                 return;
602         if (reason & 0x80)
603                 mem_parity_error(reason, regs);
604         if (reason & 0x40)
605                 io_check_error(reason, regs);
606         /*
607          * Reassert NMI in case it became active meanwhile
608          * as it's edge-triggered.
609          */
610         reassert_nmi();
611 }
612 
613 static int dummy_nmi_callback(struct pt_regs * regs, int cpu)
614 {
615         return 0;
616 }
617  
618 static nmi_callback_t nmi_callback = dummy_nmi_callback;
619  
620 fastcall void do_nmi(struct pt_regs * regs, long error_code)
621 {
622         int cpu;
623 
624         nmi_enter();
625 
626         cpu = smp_processor_id();
627         ++nmi_count(cpu);
628 
629         if (!nmi_callback(regs, cpu))
630                 default_do_nmi(regs);
631 
632         nmi_exit();
633 }
634 
635 void set_nmi_callback(nmi_callback_t callback)
636 {
637         nmi_callback = callback;
638 }
639 
640 void unset_nmi_callback(void)
641 {
642         nmi_callback = dummy_nmi_callback;
643 }
644 
645 #ifdef CONFIG_KPROBES
646 fastcall int do_int3(struct pt_regs *regs, long error_code)
647 {
648         if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
649                         == NOTIFY_STOP)
650                 return 1;
651         /* This is an interrupt gate, because kprobes wants interrupts
652         disabled.  Normal trap handlers don't. */
653         restore_interrupts(regs);
654         do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL);
655         return 0;
656 }
657 #endif
658 
659 /*
660  * Our handling of the processor debug registers is non-trivial.
661  * We do not clear them on entry and exit from the kernel. Therefore
662  * it is possible to get a watchpoint trap here from inside the kernel.
663  * However, the code in ./ptrace.c has ensured that the user can
664  * only set watchpoints on userspace addresses. Therefore the in-kernel
665  * watchpoint trap can only occur in code which is reading/writing
666  * from user space. Such code must not hold kernel locks (since it
667  * can equally take a page fault), therefore it is safe to call
668  * force_sig_info even though that claims and releases locks.
669  * 
670  * Code in ./signal.c ensures that the debug control register
671  * is restored before we deliver any signal, and therefore that
672  * user code runs with the correct debug control register even though
673  * we clear it here.
674  *
675  * Being careful here means that we don't have to be as careful in a
676  * lot of more complicated places (task switching can be a bit lazy
677  * about restoring all the debug state, and ptrace doesn't have to
678  * find every occurrence of the TF bit that could be saved away even
679  * by user code)
680  */
681 fastcall void do_debug(struct pt_regs * regs, long error_code)
682 {
683         unsigned int condition;
684         struct task_struct *tsk = current;
685 
686         __asm__ __volatile__("movl %%db6,%0" : "=r" (condition));
687 
688         if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
689                                         SIGTRAP) == NOTIFY_STOP)
690                 return;
691         /* It's safe to allow irq's after DR6 has been saved */
692         if (regs->eflags & X86_EFLAGS_IF)
693                 local_irq_enable();
694 
695         /* Mask out spurious debug traps due to lazy DR7 setting */
696         if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
697                 if (!tsk->thread.debugreg[7])
698                         goto clear_dr7;
699         }
700 
701         if (regs->eflags & VM_MASK)
702                 goto debug_vm86;
703 
704         /* Save debug status register where ptrace can see it */
705         tsk->thread.debugreg[6] = condition;
706 
707         /*
708          * Single-stepping through TF: make sure we ignore any events in
709          * kernel space (but re-enable TF when returning to user mode).
710          * And if the event was due to a debugger (PT_DTRACE), clear the
711          * TF flag so that register information is correct.
712          */
713         if (condition & DR_STEP) {
714                 /*
715                  * We already checked v86 mode above, so we can
716                  * check for kernel mode by just checking the CPL
717                  * of CS.
718                  */
719                 if ((regs->xcs & 3) == 0)
720                         goto clear_TF_reenable;
721 
722                 if (likely(tsk->ptrace & PT_DTRACE)) {
723                         tsk->ptrace &= ~PT_DTRACE;
724                         regs->eflags &= ~TF_MASK;
725                 }
726         }
727 
728         /* Ok, finally something we can handle */
729         send_sigtrap(tsk, regs, error_code);
730 
731         /* Disable additional traps. They'll be re-enabled when
732          * the signal is delivered.
733          */
734 clear_dr7:
735         __asm__("movl %0,%%db7"
736                 : /* no output */
737                 : "r" (0));
738         return;
739 
740 debug_vm86:
741         handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
742         return;
743 
744 clear_TF_reenable:
745         set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
746         regs->eflags &= ~TF_MASK;
747         return;
748 }
749 
750 /*
751  * Note that we play around with the 'TS' bit in an attempt to get
752  * the correct behaviour even in the presence of the asynchronous
753  * IRQ13 behaviour
754  */
755 void math_error(void __user *eip)
756 {
757         struct task_struct * task;
758         siginfo_t info;
759         unsigned short cwd, swd;
760 
761         /*
762          * Save the info for the exception handler and clear the error.
763          */
764         task = current;
765         save_init_fpu(task);
766         task->thread.trap_no = 16;
767         task->thread.error_code = 0;
768         info.si_signo = SIGFPE;
769         info.si_errno = 0;
770         info.si_code = __SI_FAULT;
771         info.si_addr = eip;
772         /*
773          * (~cwd & swd) will mask out exceptions that are not set to unmasked
774          * status.  0x3f is the exception bits in these regs, 0x200 is the
775          * C1 reg you need in case of a stack fault, 0x040 is the stack
776          * fault bit.  We should only be taking one exception at a time,
777          * so if this combination doesn't produce any single exception,
778          * then we have a bad program that isn't syncronizing its FPU usage
779          * and it will suffer the consequences since we won't be able to
780          * fully reproduce the context of the exception
781          */
782         cwd = get_fpu_cwd(task);
783         swd = get_fpu_swd(task);
784         switch (((~cwd) & swd & 0x3f) | (swd & 0x240)) {
785                 case 0x000:
786                 default:
787                         break;
788                 case 0x001: /* Invalid Op */
789                 case 0x041: /* Stack Fault */
790                 case 0x241: /* Stack Fault | Direction */
791                         info.si_code = FPE_FLTINV;
792                         /* Should we clear the SF or let user space do it ???? */
793                         break;
794                 case 0x002: /* Denormalize */
795                 case 0x010: /* Underflow */
796                         info.si_code = FPE_FLTUND;
797                         break;
798                 case 0x004: /* Zero Divide */
799                         info.si_code = FPE_FLTDIV;
800                         break;
801                 case 0x008: /* Overflow */
802                         info.si_code = FPE_FLTOVF;
803                         break;
804                 case 0x020: /* Precision */
805                         info.si_code = FPE_FLTRES;
806                         break;
807         }
808         force_sig_info(SIGFPE, &info, task);
809 }
810 
811 fastcall void do_coprocessor_error(struct pt_regs * regs, long error_code)
812 {
813         ignore_fpu_irq = 1;
814         math_error((void __user *)regs->eip);
815 }
816 
817 void simd_math_error(void __user *eip)
818 {
819         struct task_struct * task;
820         siginfo_t info;
821         unsigned short mxcsr;
822 
823         /*
824          * Save the info for the exception handler and clear the error.
825          */
826         task = current;
827         save_init_fpu(task);
828         task->thread.trap_no = 19;
829         task->thread.error_code = 0;
830         info.si_signo = SIGFPE;
831         info.si_errno = 0;
832         info.si_code = __SI_FAULT;
833         info.si_addr = eip;
834         /*
835          * The SIMD FPU exceptions are handled a little differently, as there
836          * is only a single status/control register.  Thus, to determine which
837          * unmasked exception was caught we must mask the exception mask bits
838          * at 0x1f80, and then use these to mask the exception bits at 0x3f.
839          */
840         mxcsr = get_fpu_mxcsr(task);
841         switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
842                 case 0x000:
843                 default:
844                         break;
845                 case 0x001: /* Invalid Op */
846                         info.si_code = FPE_FLTINV;
847                         break;
848                 case 0x002: /* Denormalize */
849                 case 0x010: /* Underflow */
850                         info.si_code = FPE_FLTUND;
851                         break;
852                 case 0x004: /* Zero Divide */
853                         info.si_code = FPE_FLTDIV;
854                         break;
855                 case 0x008: /* Overflow */
856                         info.si_code = FPE_FLTOVF;
857                         break;
858                 case 0x020: /* Precision */
859                         info.si_code = FPE_FLTRES;
860                         break;
861         }
862         force_sig_info(SIGFPE, &info, task);
863 }
864 
865 fastcall void do_simd_coprocessor_error(struct pt_regs * regs,
866                                           long error_code)
867 {
868         if (cpu_has_xmm) {
869                 /* Handle SIMD FPU exceptions on PIII+ processors. */
870                 ignore_fpu_irq = 1;
871                 simd_math_error((void __user *)regs->eip);
872         } else {
873                 /*
874                  * Handle strange cache flush from user space exception
875                  * in all other cases.  This is undocumented behaviour.
876                  */
877                 if (regs->eflags & VM_MASK) {
878                         handle_vm86_fault((struct kernel_vm86_regs *)regs,
879                                           error_code);
880                         return;
881                 }
882                 die_if_kernel("cache flush denied", regs, error_code);
883                 current->thread.trap_no = 19;
884                 current->thread.error_code = error_code;
885                 force_sig(SIGSEGV, current);
886         }
887 }
888 
889 fastcall void do_spurious_interrupt_bug(struct pt_regs * regs,
890                                           long error_code)
891 {
892 #if 0
893         /* No need to warn about this any longer. */
894         printk("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
895 #endif
896 }
897 
898 /*
899  *  'math_state_restore()' saves the current math information in the
900  * old math state array, and gets the new ones from the current task
901  *
902  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
903  * Don't touch unless you *really* know how it works.
904  *
905  * Must be called with kernel preemption disabled (in this case,
906  * local interrupts are disabled at the call-site in entry.S).
907  */
908 asmlinkage void math_state_restore(struct pt_regs regs)
909 {
910         struct thread_info *thread = current_thread_info();
911         struct task_struct *tsk = thread->task;
912 
913         clts();         /* Allow maths ops (or we recurse) */
914         if (!tsk_used_math(tsk))
915                 init_fpu(tsk);
916         restore_fpu(tsk);
917         thread->status |= TS_USEDFPU;   /* So we fnsave on switch_to() */
918 }
919 
920 #ifndef CONFIG_MATH_EMULATION
921 
922 asmlinkage void math_emulate(long arg)
923 {
924         printk("math-emulation not enabled and no coprocessor found.\n");
925         printk("killing %s.\n",current->comm);
926         force_sig(SIGFPE,current);
927         schedule();
928 }
929 
930 #endif /* CONFIG_MATH_EMULATION */
931 
932 #ifdef CONFIG_X86_F00F_BUG
933 void __init trap_init_f00f_bug(void)
934 {
935         __set_fixmap(FIX_F00F_IDT, __pa(&idt_table), PAGE_KERNEL_RO);
936 
937         /*
938          * Update the IDT descriptor and reload the IDT so that
939          * it uses the read-only mapped virtual address.
940          */
941         idt_descr.address = fix_to_virt(FIX_F00F_IDT);
942         __asm__ __volatile__("lidt %0" : : "m" (idt_descr));
943 }
944 #endif
945 
946 #define _set_gate(gate_addr,type,dpl,addr,seg) \
947 do { \
948   int __d0, __d1; \
949   __asm__ __volatile__ ("movw %%dx,%%ax\n\t" \
950         "movw %4,%%dx\n\t" \
951         "movl %%eax,%0\n\t" \
952         "movl %%edx,%1" \
953         :"=m" (*((long *) (gate_addr))), \
954          "=m" (*(1+(long *) (gate_addr))), "=&a" (__d0), "=&d" (__d1) \
955         :"i" ((short) (0x8000+(dpl<<13)+(type<<8))), \
956          "3" ((char *) (addr)),"2" ((seg) << 16)); \
957 } while (0)
958 
959 
960 /*
961  * This needs to use 'idt_table' rather than 'idt', and
962  * thus use the _nonmapped_ version of the IDT, as the
963  * Pentium F0 0F bugfix can have resulted in the mapped
964  * IDT being write-protected.
965  */
966 void set_intr_gate(unsigned int n, void *addr)
967 {
968         _set_gate(idt_table+n,14,0,addr,__KERNEL_CS);
969 }
970 
971 /*
972  * This routine sets up an interrupt gate at directory privilege level 3.
973  */
974 static inline void set_system_intr_gate(unsigned int n, void *addr)
975 {
976         _set_gate(idt_table+n, 14, 3, addr, __KERNEL_CS);
977 }
978 
979 static void __init set_trap_gate(unsigned int n, void *addr)
980 {
981         _set_gate(idt_table+n,15,0,addr,__KERNEL_CS);
982 }
983 
984 static void __init set_system_gate(unsigned int n, void *addr)
985 {
986         _set_gate(idt_table+n,15,3,addr,__KERNEL_CS);
987 }
988 
989 static void __init set_task_gate(unsigned int n, unsigned int gdt_entry)
990 {
991         _set_gate(idt_table+n,5,0,0,(gdt_entry<<3));
992 }
993 
994 
995 void __init trap_init(void)
996 {
997 #ifdef CONFIG_EISA
998         void __iomem *p = ioremap(0x0FFFD9, 4);
999         if (readl(p) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1000                 EISA_bus = 1;
1001         }
1002         iounmap(p);
1003 #endif
1004 
1005 #ifdef CONFIG_X86_LOCAL_APIC
1006         init_apic_mappings();
1007 #endif
1008 
1009         set_trap_gate(0,&divide_error);
1010         set_intr_gate(1,&debug);
1011         set_intr_gate(2,&nmi);
1012         set_system_intr_gate(3, &int3); /* int3-5 can be called from all */
1013         set_system_gate(4,&overflow);
1014         set_system_gate(5,&bounds);
1015         set_trap_gate(6,&invalid_op);
1016         set_trap_gate(7,&device_not_available);
1017         set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS);
1018         set_trap_gate(9,&coprocessor_segment_overrun);
1019         set_trap_gate(10,&invalid_TSS);
1020         set_trap_gate(11,&segment_not_present);
1021         set_trap_gate(12,&stack_segment);
1022         set_trap_gate(13,&general_protection);
1023         set_intr_gate(14,&page_fault);
1024         set_trap_gate(15,&spurious_interrupt_bug);
1025         set_trap_gate(16,&coprocessor_error);
1026         set_trap_gate(17,&alignment_check);
1027 #ifdef CONFIG_X86_MCE
1028         set_trap_gate(18,&machine_check);
1029 #endif
1030         set_trap_gate(19,&simd_coprocessor_error);
1031 
1032         set_system_gate(SYSCALL_VECTOR,&system_call);
1033 
1034         /*
1035          * Should be a barrier for any external CPU state.
1036          */
1037         cpu_init();
1038 
1039         trap_init_hook();
1040 }
1041 
  This page was automatically generated by the LXR engine.