]> CyberLeo.Net >> Repos - FreeBSD/releng/9.2.git/blob - sys/i386/i386/trap.c
- Copy stable/9 to releng/9.2 as part of the 9.2-RELEASE cycle.
[FreeBSD/releng/9.2.git] / sys / i386 / i386 / trap.c
1 /*-
2  * Copyright (C) 1994, David Greenman
3  * Copyright (c) 1990, 1993
4  *      The Regents of the University of California.  All rights reserved.
5  *
6  * This code is derived from software contributed to Berkeley by
7  * the University of Utah, and William Jolitz.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 3. All advertising materials mentioning features or use of this software
18  *    must display the following acknowledgement:
19  *      This product includes software developed by the University of
20  *      California, Berkeley and its contributors.
21  * 4. Neither the name of the University nor the names of its contributors
22  *    may be used to endorse or promote products derived from this software
23  *    without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  *
37  *      from: @(#)trap.c        7.4 (Berkeley) 5/13/91
38  */
39
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
42
43 /*
44  * 386 Trap and System call handling
45  */
46
47 #include "opt_clock.h"
48 #include "opt_cpu.h"
49 #include "opt_hwpmc_hooks.h"
50 #include "opt_isa.h"
51 #include "opt_kdb.h"
52 #include "opt_kdtrace.h"
53 #include "opt_npx.h"
54 #include "opt_trap.h"
55
56 #include <sys/param.h>
57 #include <sys/bus.h>
58 #include <sys/systm.h>
59 #include <sys/proc.h>
60 #include <sys/pioctl.h>
61 #include <sys/ptrace.h>
62 #include <sys/kdb.h>
63 #include <sys/kernel.h>
64 #include <sys/ktr.h>
65 #include <sys/lock.h>
66 #include <sys/mutex.h>
67 #include <sys/resourcevar.h>
68 #include <sys/signalvar.h>
69 #include <sys/syscall.h>
70 #include <sys/sysctl.h>
71 #include <sys/sysent.h>
72 #include <sys/uio.h>
73 #include <sys/vmmeter.h>
74 #ifdef HWPMC_HOOKS
75 #include <sys/pmckern.h>
76 PMC_SOFT_DEFINE( , , page_fault, all);
77 PMC_SOFT_DEFINE( , , page_fault, read);
78 PMC_SOFT_DEFINE( , , page_fault, write);
79 #endif
80 #include <security/audit/audit.h>
81
82 #include <vm/vm.h>
83 #include <vm/vm_param.h>
84 #include <vm/pmap.h>
85 #include <vm/vm_kern.h>
86 #include <vm/vm_map.h>
87 #include <vm/vm_page.h>
88 #include <vm/vm_extern.h>
89
90 #include <machine/cpu.h>
91 #include <machine/intr_machdep.h>
92 #include <x86/mca.h>
93 #include <machine/md_var.h>
94 #include <machine/pcb.h>
95 #ifdef SMP
96 #include <machine/smp.h>
97 #endif
98 #include <machine/tss.h>
99 #include <machine/vm86.h>
100
101 #ifdef POWERFAIL_NMI
102 #include <sys/syslog.h>
103 #include <machine/clock.h>
104 #endif
105
106 #ifdef KDTRACE_HOOKS
107 #include <sys/dtrace_bsd.h>
108
109 /*
110  * This is a hook which is initialised by the dtrace module
111  * to handle traps which might occur during DTrace probe
112  * execution.
113  */
114 dtrace_trap_func_t      dtrace_trap_func;
115
116 dtrace_doubletrap_func_t        dtrace_doubletrap_func;
117
118 /*
119  * This is a hook which is initialised by the systrace module
120  * when it is loaded. This keeps the DTrace syscall provider
121  * implementation opaque. 
122  */
123 systrace_probe_func_t   systrace_probe_func;
124
125 /*
126  * These hooks are necessary for the pid, usdt and fasttrap providers.
127  */
128 dtrace_fasttrap_probe_ptr_t     dtrace_fasttrap_probe_ptr;
129 dtrace_pid_probe_ptr_t          dtrace_pid_probe_ptr;
130 dtrace_return_probe_ptr_t       dtrace_return_probe_ptr;
131 #endif
132
133 extern void trap(struct trapframe *frame);
134 extern void syscall(struct trapframe *frame);
135
136 static int trap_pfault(struct trapframe *, int, vm_offset_t);
137 static void trap_fatal(struct trapframe *, vm_offset_t);
138 void dblfault_handler(void);
139
140 extern inthand_t IDTVEC(lcall_syscall);
141
142 #define MAX_TRAP_MSG            33
143 static char *trap_msg[] = {
144         "",                                     /*  0 unused */
145         "privileged instruction fault",         /*  1 T_PRIVINFLT */
146         "",                                     /*  2 unused */
147         "breakpoint instruction fault",         /*  3 T_BPTFLT */
148         "",                                     /*  4 unused */
149         "",                                     /*  5 unused */
150         "arithmetic trap",                      /*  6 T_ARITHTRAP */
151         "",                                     /*  7 unused */
152         "",                                     /*  8 unused */
153         "general protection fault",             /*  9 T_PROTFLT */
154         "trace trap",                           /* 10 T_TRCTRAP */
155         "",                                     /* 11 unused */
156         "page fault",                           /* 12 T_PAGEFLT */
157         "",                                     /* 13 unused */
158         "alignment fault",                      /* 14 T_ALIGNFLT */
159         "",                                     /* 15 unused */
160         "",                                     /* 16 unused */
161         "",                                     /* 17 unused */
162         "integer divide fault",                 /* 18 T_DIVIDE */
163         "non-maskable interrupt trap",          /* 19 T_NMI */
164         "overflow trap",                        /* 20 T_OFLOW */
165         "FPU bounds check fault",               /* 21 T_BOUND */
166         "FPU device not available",             /* 22 T_DNA */
167         "double fault",                         /* 23 T_DOUBLEFLT */
168         "FPU operand fetch fault",              /* 24 T_FPOPFLT */
169         "invalid TSS fault",                    /* 25 T_TSSFLT */
170         "segment not present fault",            /* 26 T_SEGNPFLT */
171         "stack fault",                          /* 27 T_STKFLT */
172         "machine check trap",                   /* 28 T_MCHK */
173         "SIMD floating-point exception",        /* 29 T_XMMFLT */
174         "reserved (unknown) fault",             /* 30 T_RESERVED */
175         "",                                     /* 31 unused (reserved) */
176         "DTrace pid return trap",               /* 32 T_DTRACE_RET */
177         "DTrace fasttrap probe trap",           /* 33 T_DTRACE_PROBE */
178
179 };
180
181 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
182 extern int has_f00f_bug;
183 #endif
184
185 #ifdef KDB
186 static int kdb_on_nmi = 1;
187 SYSCTL_INT(_machdep, OID_AUTO, kdb_on_nmi, CTLFLAG_RW,
188         &kdb_on_nmi, 0, "Go to KDB on NMI");
189 TUNABLE_INT("machdep.kdb_on_nmi", &kdb_on_nmi);
190 #endif
191 static int panic_on_nmi = 1;
192 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
193         &panic_on_nmi, 0, "Panic on NMI");
194 TUNABLE_INT("machdep.panic_on_nmi", &panic_on_nmi);
195 static int prot_fault_translation = 0;
196 SYSCTL_INT(_machdep, OID_AUTO, prot_fault_translation, CTLFLAG_RW,
197         &prot_fault_translation, 0, "Select signal to deliver on protection fault");
198 static int uprintf_signal;
199 SYSCTL_INT(_machdep, OID_AUTO, uprintf_signal, CTLFLAG_RW,
200     &uprintf_signal, 0,
201     "Print debugging information on trap signal to ctty");
202
203 /*
204  * Exception, fault, and trap interface to the FreeBSD kernel.
205  * This common code is called from assembly language IDT gate entry
206  * routines that prepare a suitable stack frame, and restore this
207  * frame after the exception has been processed.
208  */
209
210 void
211 trap(struct trapframe *frame)
212 {
213         struct thread *td = curthread;
214         struct proc *p = td->td_proc;
215         int i = 0, ucode = 0, code;
216         u_int type;
217         register_t addr = 0;
218         vm_offset_t eva;
219         ksiginfo_t ksi;
220 #ifdef POWERFAIL_NMI
221         static int lastalert = 0;
222 #endif
223
224         PCPU_INC(cnt.v_trap);
225         type = frame->tf_trapno;
226
227 #ifdef SMP
228         /* Handler for NMI IPIs used for stopping CPUs. */
229         if (type == T_NMI) {
230                  if (ipi_nmi_handler() == 0)
231                            goto out;
232         }
233 #endif /* SMP */
234
235 #ifdef KDB
236         if (kdb_active) {
237                 kdb_reenter();
238                 goto out;
239         }
240 #endif
241
242         if (type == T_RESERVED) {
243                 trap_fatal(frame, 0);
244                 goto out;
245         }
246
247 #ifdef  HWPMC_HOOKS
248         /*
249          * CPU PMCs interrupt using an NMI so we check for that first.
250          * If the HWPMC module is active, 'pmc_hook' will point to
251          * the function to be called.  A return value of '1' from the
252          * hook means that the NMI was handled by it and that we can
253          * return immediately.
254          */
255         if (type == T_NMI && pmc_intr &&
256             (*pmc_intr)(PCPU_GET(cpuid), frame))
257             goto out;
258 #endif
259
260         if (type == T_MCHK) {
261                 mca_intr();
262                 goto out;
263         }
264
265 #ifdef KDTRACE_HOOKS
266         /*
267          * A trap can occur while DTrace executes a probe. Before
268          * executing the probe, DTrace blocks re-scheduling and sets
269          * a flag in it's per-cpu flags to indicate that it doesn't
270          * want to fault. On returning from the probe, the no-fault
271          * flag is cleared and finally re-scheduling is enabled.
272          *
273          * If the DTrace kernel module has registered a trap handler,
274          * call it and if it returns non-zero, assume that it has
275          * handled the trap and modified the trap frame so that this
276          * function can return normally.
277          */
278         if (type == T_DTRACE_PROBE || type == T_DTRACE_RET ||
279             type == T_BPTFLT) {
280                 struct reg regs;
281
282                 fill_frame_regs(frame, &regs);
283                 if (type == T_DTRACE_PROBE &&
284                     dtrace_fasttrap_probe_ptr != NULL &&
285                     dtrace_fasttrap_probe_ptr(&regs) == 0)
286                         goto out;
287                 if (type == T_BPTFLT &&
288                     dtrace_pid_probe_ptr != NULL &&
289                     dtrace_pid_probe_ptr(&regs) == 0)
290                         goto out;
291                 if (type == T_DTRACE_RET &&
292                     dtrace_return_probe_ptr != NULL &&
293                     dtrace_return_probe_ptr(&regs) == 0)
294                         goto out;
295         }
296         if ((type == T_PROTFLT || type == T_PAGEFLT) &&
297             dtrace_trap_func != NULL && (*dtrace_trap_func)(frame, type))
298                 goto out;
299 #endif
300
301         if ((frame->tf_eflags & PSL_I) == 0) {
302                 /*
303                  * Buggy application or kernel code has disabled
304                  * interrupts and then trapped.  Enabling interrupts
305                  * now is wrong, but it is better than running with
306                  * interrupts disabled until they are accidentally
307                  * enabled later.
308                  */
309                 if (ISPL(frame->tf_cs) == SEL_UPL || (frame->tf_eflags & PSL_VM))
310                         uprintf(
311                             "pid %ld (%s): trap %d with interrupts disabled\n",
312                             (long)curproc->p_pid, curthread->td_name, type);
313                 else if (type != T_BPTFLT && type != T_TRCTRAP &&
314                          frame->tf_eip != (int)cpu_switch_load_gs) {
315                         /*
316                          * XXX not quite right, since this may be for a
317                          * multiple fault in user mode.
318                          */
319                         printf("kernel trap %d with interrupts disabled\n",
320                             type);
321                         /*
322                          * Page faults need interrupts disabled until later,
323                          * and we shouldn't enable interrupts while holding
324                          * a spin lock or if servicing an NMI.
325                          */
326                         if (type != T_NMI && type != T_PAGEFLT &&
327                             td->td_md.md_spinlock_count == 0)
328                                 enable_intr();
329                 }
330         }
331         eva = 0;
332         code = frame->tf_err;
333         if (type == T_PAGEFLT) {
334                 /*
335                  * For some Cyrix CPUs, %cr2 is clobbered by
336                  * interrupts.  This problem is worked around by using
337                  * an interrupt gate for the pagefault handler.  We
338                  * are finally ready to read %cr2 and conditionally
339                  * reenable interrupts.  If we hold a spin lock, then
340                  * we must not reenable interrupts.  This might be a
341                  * spurious page fault.
342                  */
343                 eva = rcr2();
344                 if (td->td_md.md_spinlock_count == 0)
345                         enable_intr();
346         }
347
348         if ((ISPL(frame->tf_cs) == SEL_UPL) ||
349             ((frame->tf_eflags & PSL_VM) && 
350                 !(curpcb->pcb_flags & PCB_VM86CALL))) {
351                 /* user trap */
352
353                 td->td_pticks = 0;
354                 td->td_frame = frame;
355                 addr = frame->tf_eip;
356                 if (td->td_ucred != p->p_ucred) 
357                         cred_update_thread(td);
358
359                 switch (type) {
360                 case T_PRIVINFLT:       /* privileged instruction fault */
361                         i = SIGILL;
362                         ucode = ILL_PRVOPC;
363                         break;
364
365                 case T_BPTFLT:          /* bpt instruction fault */
366                 case T_TRCTRAP:         /* trace trap */
367                         enable_intr();
368                         frame->tf_eflags &= ~PSL_T;
369                         i = SIGTRAP;
370                         ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
371                         break;
372
373                 case T_ARITHTRAP:       /* arithmetic trap */
374 #ifdef DEV_NPX
375                         ucode = npxtrap_x87();
376                         if (ucode == -1)
377                                 goto userout;
378 #else
379                         ucode = 0;
380 #endif
381                         i = SIGFPE;
382                         break;
383
384                         /*
385                          * The following two traps can happen in
386                          * vm86 mode, and, if so, we want to handle
387                          * them specially.
388                          */
389                 case T_PROTFLT:         /* general protection fault */
390                 case T_STKFLT:          /* stack fault */
391                         if (frame->tf_eflags & PSL_VM) {
392                                 i = vm86_emulate((struct vm86frame *)frame);
393                                 if (i == 0)
394                                         goto user;
395                                 break;
396                         }
397                         i = SIGBUS;
398                         ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR;
399                         break;
400                 case T_SEGNPFLT:        /* segment not present fault */
401                         i = SIGBUS;
402                         ucode = BUS_ADRERR;
403                         break;
404                 case T_TSSFLT:          /* invalid TSS fault */
405                         i = SIGBUS;
406                         ucode = BUS_OBJERR;
407                         break;
408                 case T_DOUBLEFLT:       /* double fault */
409                 default:
410                         i = SIGBUS;
411                         ucode = BUS_OBJERR;
412                         break;
413
414                 case T_PAGEFLT:         /* page fault */
415
416                         i = trap_pfault(frame, TRUE, eva);
417 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
418                         if (i == -2) {
419                                 /*
420                                  * The f00f hack workaround has triggered, so
421                                  * treat the fault as an illegal instruction 
422                                  * (T_PRIVINFLT) instead of a page fault.
423                                  */
424                                 type = frame->tf_trapno = T_PRIVINFLT;
425
426                                 /* Proceed as in that case. */
427                                 ucode = ILL_PRVOPC;
428                                 i = SIGILL;
429                                 break;
430                         }
431 #endif
432                         if (i == -1)
433                                 goto userout;
434                         if (i == 0)
435                                 goto user;
436
437                         if (i == SIGSEGV)
438                                 ucode = SEGV_MAPERR;
439                         else {
440                                 if (prot_fault_translation == 0) {
441                                         /*
442                                          * Autodetect.
443                                          * This check also covers the images
444                                          * without the ABI-tag ELF note.
445                                          */
446                                         if (SV_CURPROC_ABI() == SV_ABI_FREEBSD
447                                             && p->p_osrel >= P_OSREL_SIGSEGV) {
448                                                 i = SIGSEGV;
449                                                 ucode = SEGV_ACCERR;
450                                         } else {
451                                                 i = SIGBUS;
452                                                 ucode = BUS_PAGE_FAULT;
453                                         }
454                                 } else if (prot_fault_translation == 1) {
455                                         /*
456                                          * Always compat mode.
457                                          */
458                                         i = SIGBUS;
459                                         ucode = BUS_PAGE_FAULT;
460                                 } else {
461                                         /*
462                                          * Always SIGSEGV mode.
463                                          */
464                                         i = SIGSEGV;
465                                         ucode = SEGV_ACCERR;
466                                 }
467                         }
468                         addr = eva;
469                         break;
470
471                 case T_DIVIDE:          /* integer divide fault */
472                         ucode = FPE_INTDIV;
473                         i = SIGFPE;
474                         break;
475
476 #ifdef DEV_ISA
477                 case T_NMI:
478 #ifdef POWERFAIL_NMI
479 #ifndef TIMER_FREQ
480 #  define TIMER_FREQ 1193182
481 #endif
482                         if (time_second - lastalert > 10) {
483                                 log(LOG_WARNING, "NMI: power fail\n");
484                                 sysbeep(880, hz);
485                                 lastalert = time_second;
486                         }
487                         goto userout;
488 #else /* !POWERFAIL_NMI */
489                         /* machine/parity/power fail/"kitchen sink" faults */
490                         if (isa_nmi(code) == 0) {
491 #ifdef KDB
492                                 /*
493                                  * NMI can be hooked up to a pushbutton
494                                  * for debugging.
495                                  */
496                                 if (kdb_on_nmi) {
497                                         printf ("NMI ... going to debugger\n");
498                                         kdb_trap(type, 0, frame);
499                                 }
500 #endif /* KDB */
501                                 goto userout;
502                         } else if (panic_on_nmi)
503                                 panic("NMI indicates hardware failure");
504                         break;
505 #endif /* POWERFAIL_NMI */
506 #endif /* DEV_ISA */
507
508                 case T_OFLOW:           /* integer overflow fault */
509                         ucode = FPE_INTOVF;
510                         i = SIGFPE;
511                         break;
512
513                 case T_BOUND:           /* bounds check fault */
514                         ucode = FPE_FLTSUB;
515                         i = SIGFPE;
516                         break;
517
518                 case T_DNA:
519 #ifdef DEV_NPX
520                         KASSERT(PCB_USER_FPU(td->td_pcb),
521                             ("kernel FPU ctx has leaked"));
522                         /* transparent fault (due to context switch "late") */
523                         if (npxdna())
524                                 goto userout;
525 #endif
526                         uprintf("pid %d killed due to lack of floating point\n",
527                                 p->p_pid);
528                         i = SIGKILL;
529                         ucode = 0;
530                         break;
531
532                 case T_FPOPFLT:         /* FPU operand fetch fault */
533                         ucode = ILL_COPROC;
534                         i = SIGILL;
535                         break;
536
537                 case T_XMMFLT:          /* SIMD floating-point exception */
538 #if defined(DEV_NPX) && !defined(CPU_DISABLE_SSE) && defined(I686_CPU)
539                         ucode = npxtrap_sse();
540                         if (ucode == -1)
541                                 goto userout;
542 #else
543                         ucode = 0;
544 #endif
545                         i = SIGFPE;
546                         break;
547                 }
548         } else {
549                 /* kernel trap */
550
551                 KASSERT(cold || td->td_ucred != NULL,
552                     ("kernel trap doesn't have ucred"));
553                 switch (type) {
554                 case T_PAGEFLT:                 /* page fault */
555                         (void) trap_pfault(frame, FALSE, eva);
556                         goto out;
557
558                 case T_DNA:
559 #ifdef DEV_NPX
560                         KASSERT(!PCB_USER_FPU(td->td_pcb),
561                             ("Unregistered use of FPU in kernel"));
562                         if (npxdna())
563                                 goto out;
564 #endif
565                         break;
566
567                 case T_ARITHTRAP:       /* arithmetic trap */
568                 case T_XMMFLT:          /* SIMD floating-point exception */
569                 case T_FPOPFLT:         /* FPU operand fetch fault */
570                         /*
571                          * XXXKIB for now disable any FPU traps in kernel
572                          * handler registration seems to be overkill
573                          */
574                         trap_fatal(frame, 0);
575                         goto out;
576
577                         /*
578                          * The following two traps can happen in
579                          * vm86 mode, and, if so, we want to handle
580                          * them specially.
581                          */
582                 case T_PROTFLT:         /* general protection fault */
583                 case T_STKFLT:          /* stack fault */
584                         if (frame->tf_eflags & PSL_VM) {
585                                 i = vm86_emulate((struct vm86frame *)frame);
586                                 if (i != 0)
587                                         /*
588                                          * returns to original process
589                                          */
590                                         vm86_trap((struct vm86frame *)frame);
591                                 goto out;
592                         }
593                         if (type == T_STKFLT)
594                                 break;
595
596                         /* FALL THROUGH */
597
598                 case T_SEGNPFLT:        /* segment not present fault */
599                         if (curpcb->pcb_flags & PCB_VM86CALL)
600                                 break;
601
602                         /*
603                          * Invalid %fs's and %gs's can be created using
604                          * procfs or PT_SETREGS or by invalidating the
605                          * underlying LDT entry.  This causes a fault
606                          * in kernel mode when the kernel attempts to
607                          * switch contexts.  Lose the bad context
608                          * (XXX) so that we can continue, and generate
609                          * a signal.
610                          */
611                         if (frame->tf_eip == (int)cpu_switch_load_gs) {
612                                 curpcb->pcb_gs = 0;
613 #if 0                           
614                                 PROC_LOCK(p);
615                                 kern_psignal(p, SIGBUS);
616                                 PROC_UNLOCK(p);
617 #endif                          
618                                 goto out;
619                         }
620
621                         if (td->td_intr_nesting_level != 0)
622                                 break;
623
624                         /*
625                          * Invalid segment selectors and out of bounds
626                          * %eip's and %esp's can be set up in user mode.
627                          * This causes a fault in kernel mode when the
628                          * kernel tries to return to user mode.  We want
629                          * to get this fault so that we can fix the
630                          * problem here and not have to check all the
631                          * selectors and pointers when the user changes
632                          * them.
633                          */
634                         if (frame->tf_eip == (int)doreti_iret) {
635                                 frame->tf_eip = (int)doreti_iret_fault;
636                                 goto out;
637                         }
638                         if (frame->tf_eip == (int)doreti_popl_ds) {
639                                 frame->tf_eip = (int)doreti_popl_ds_fault;
640                                 goto out;
641                         }
642                         if (frame->tf_eip == (int)doreti_popl_es) {
643                                 frame->tf_eip = (int)doreti_popl_es_fault;
644                                 goto out;
645                         }
646                         if (frame->tf_eip == (int)doreti_popl_fs) {
647                                 frame->tf_eip = (int)doreti_popl_fs_fault;
648                                 goto out;
649                         }
650                         if (curpcb->pcb_onfault != NULL) {
651                                 frame->tf_eip =
652                                     (int)curpcb->pcb_onfault;
653                                 goto out;
654                         }
655                         break;
656
657                 case T_TSSFLT:
658                         /*
659                          * PSL_NT can be set in user mode and isn't cleared
660                          * automatically when the kernel is entered.  This
661                          * causes a TSS fault when the kernel attempts to
662                          * `iret' because the TSS link is uninitialized.  We
663                          * want to get this fault so that we can fix the
664                          * problem here and not every time the kernel is
665                          * entered.
666                          */
667                         if (frame->tf_eflags & PSL_NT) {
668                                 frame->tf_eflags &= ~PSL_NT;
669                                 goto out;
670                         }
671                         break;
672
673                 case T_TRCTRAP:  /* trace trap */
674                         if (frame->tf_eip == (int)IDTVEC(lcall_syscall)) {
675                                 /*
676                                  * We've just entered system mode via the
677                                  * syscall lcall.  Continue single stepping
678                                  * silently until the syscall handler has
679                                  * saved the flags.
680                                  */
681                                 goto out;
682                         }
683                         if (frame->tf_eip == (int)IDTVEC(lcall_syscall) + 1) {
684                                 /*
685                                  * The syscall handler has now saved the
686                                  * flags.  Stop single stepping it.
687                                  */
688                                 frame->tf_eflags &= ~PSL_T;
689                                 goto out;
690                         }
691                         /*
692                          * Ignore debug register trace traps due to
693                          * accesses in the user's address space, which
694                          * can happen under several conditions such as
695                          * if a user sets a watchpoint on a buffer and
696                          * then passes that buffer to a system call.
697                          * We still want to get TRCTRAPS for addresses
698                          * in kernel space because that is useful when
699                          * debugging the kernel.
700                          */
701                         if (user_dbreg_trap() && 
702                            !(curpcb->pcb_flags & PCB_VM86CALL)) {
703                                 /*
704                                  * Reset breakpoint bits because the
705                                  * processor doesn't
706                                  */
707                                 load_dr6(rdr6() & 0xfffffff0);
708                                 goto out;
709                         }
710                         /*
711                          * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
712                          */
713                 case T_BPTFLT:
714                         /*
715                          * If KDB is enabled, let it handle the debugger trap.
716                          * Otherwise, debugger traps "can't happen".
717                          */
718 #ifdef KDB
719                         if (kdb_trap(type, 0, frame))
720                                 goto out;
721 #endif
722                         break;
723
724 #ifdef DEV_ISA
725                 case T_NMI:
726 #ifdef POWERFAIL_NMI
727                         if (time_second - lastalert > 10) {
728                                 log(LOG_WARNING, "NMI: power fail\n");
729                                 sysbeep(880, hz);
730                                 lastalert = time_second;
731                         }
732                         goto out;
733 #else /* !POWERFAIL_NMI */
734                         /* machine/parity/power fail/"kitchen sink" faults */
735                         if (isa_nmi(code) == 0) {
736 #ifdef KDB
737                                 /*
738                                  * NMI can be hooked up to a pushbutton
739                                  * for debugging.
740                                  */
741                                 if (kdb_on_nmi) {
742                                         printf ("NMI ... going to debugger\n");
743                                         kdb_trap(type, 0, frame);
744                                 }
745 #endif /* KDB */
746                                 goto out;
747                         } else if (panic_on_nmi == 0)
748                                 goto out;
749                         /* FALLTHROUGH */
750 #endif /* POWERFAIL_NMI */
751 #endif /* DEV_ISA */
752                 }
753
754                 trap_fatal(frame, eva);
755                 goto out;
756         }
757
758         /* Translate fault for emulators (e.g. Linux) */
759         if (*p->p_sysent->sv_transtrap)
760                 i = (*p->p_sysent->sv_transtrap)(i, type);
761
762         ksiginfo_init_trap(&ksi);
763         ksi.ksi_signo = i;
764         ksi.ksi_code = ucode;
765         ksi.ksi_addr = (void *)addr;
766         ksi.ksi_trapno = type;
767         if (uprintf_signal) {
768                 uprintf("pid %d comm %s: signal %d err %x code %d type %d "
769                     "addr 0x%x esp 0x%08x eip 0x%08x "
770                     "<%02x %02x %02x %02x %02x %02x %02x %02x>\n",
771                     p->p_pid, p->p_comm, i, frame->tf_err, ucode, type, addr,
772                     frame->tf_esp, frame->tf_eip,
773                     fubyte((void *)(frame->tf_eip + 0)),
774                     fubyte((void *)(frame->tf_eip + 1)),
775                     fubyte((void *)(frame->tf_eip + 2)),
776                     fubyte((void *)(frame->tf_eip + 3)),
777                     fubyte((void *)(frame->tf_eip + 4)),
778                     fubyte((void *)(frame->tf_eip + 5)),
779                     fubyte((void *)(frame->tf_eip + 6)),
780                     fubyte((void *)(frame->tf_eip + 7)));
781         }
782         KASSERT((read_eflags() & PSL_I) != 0, ("interrupts disabled"));
783         trapsignal(td, &ksi);
784
785 #ifdef DEBUG
786         if (type <= MAX_TRAP_MSG) {
787                 uprintf("fatal process exception: %s",
788                         trap_msg[type]);
789                 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
790                         uprintf(", fault VA = 0x%lx", (u_long)eva);
791                 uprintf("\n");
792         }
793 #endif
794
795 user:
796         userret(td, frame);
797         mtx_assert(&Giant, MA_NOTOWNED);
798         KASSERT(PCB_USER_FPU(td->td_pcb),
799             ("Return from trap with kernel FPU ctx leaked"));
800 userout:
801 out:
802         return;
803 }
804
805 static int
806 trap_pfault(frame, usermode, eva)
807         struct trapframe *frame;
808         int usermode;
809         vm_offset_t eva;
810 {
811         vm_offset_t va;
812         struct vmspace *vm = NULL;
813         vm_map_t map;
814         int rv = 0;
815         vm_prot_t ftype;
816         struct thread *td = curthread;
817         struct proc *p = td->td_proc;
818
819         if (__predict_false((td->td_pflags & TDP_NOFAULTING) != 0)) {
820                 /*
821                  * Due to both processor errata and lazy TLB invalidation when
822                  * access restrictions are removed from virtual pages, memory
823                  * accesses that are allowed by the physical mapping layer may
824                  * nonetheless cause one spurious page fault per virtual page. 
825                  * When the thread is executing a "no faulting" section that
826                  * is bracketed by vm_fault_{disable,enable}_pagefaults(),
827                  * every page fault is treated as a spurious page fault,
828                  * unless it accesses the same virtual address as the most
829                  * recent page fault within the same "no faulting" section.
830                  */
831                 if (td->td_md.md_spurflt_addr != eva ||
832                     (td->td_pflags & TDP_RESETSPUR) != 0) {
833                         /*
834                          * Do nothing to the TLB.  A stale TLB entry is
835                          * flushed automatically by a page fault.
836                          */
837                         td->td_md.md_spurflt_addr = eva;
838                         td->td_pflags &= ~TDP_RESETSPUR;
839                         return (0);
840                 }
841         } else {
842                 /*
843                  * If we get a page fault while in a critical section, then
844                  * it is most likely a fatal kernel page fault.  The kernel
845                  * is already going to panic trying to get a sleep lock to
846                  * do the VM lookup, so just consider it a fatal trap so the
847                  * kernel can print out a useful trap message and even get
848                  * to the debugger.
849                  *
850                  * If we get a page fault while holding a non-sleepable
851                  * lock, then it is most likely a fatal kernel page fault.
852                  * If WITNESS is enabled, then it's going to whine about
853                  * bogus LORs with various VM locks, so just skip to the
854                  * fatal trap handling directly.
855                  */
856                 if (td->td_critnest != 0 ||
857                     WITNESS_CHECK(WARN_SLEEPOK | WARN_GIANTOK, NULL,
858                     "Kernel page fault") != 0) {
859                         trap_fatal(frame, eva);
860                         return (-1);
861                 }
862         }
863         va = trunc_page(eva);
864         if (va >= KERNBASE) {
865                 /*
866                  * Don't allow user-mode faults in kernel address space.
867                  * An exception:  if the faulting address is the invalid
868                  * instruction entry in the IDT, then the Intel Pentium
869                  * F00F bug workaround was triggered, and we need to
870                  * treat it is as an illegal instruction, and not a page
871                  * fault.
872                  */
873 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
874                 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug)
875                         return -2;
876 #endif
877                 if (usermode)
878                         goto nogo;
879
880                 map = kernel_map;
881         } else {
882                 /*
883                  * This is a fault on non-kernel virtual memory.
884                  * vm is initialized above to NULL. If curproc is NULL
885                  * or curproc->p_vmspace is NULL the fault is fatal.
886                  */
887                 if (p != NULL)
888                         vm = p->p_vmspace;
889
890                 if (vm == NULL)
891                         goto nogo;
892
893                 map = &vm->vm_map;
894                 if (!usermode && (td->td_intr_nesting_level != 0 ||
895                     curpcb->pcb_onfault == NULL)) {
896                         trap_fatal(frame, eva);
897                         return (-1);
898                 }
899         }
900
901         /*
902          * PGEX_I is defined only if the execute disable bit capability is
903          * supported and enabled.
904          */
905         if (frame->tf_err & PGEX_W)
906                 ftype = VM_PROT_WRITE;
907 #ifdef PAE
908         else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
909                 ftype = VM_PROT_EXECUTE;
910 #endif
911         else
912                 ftype = VM_PROT_READ;
913
914         if (map != kernel_map) {
915                 /*
916                  * Keep swapout from messing with us during this
917                  *      critical time.
918                  */
919                 PROC_LOCK(p);
920                 ++p->p_lock;
921                 PROC_UNLOCK(p);
922
923                 /* Fault in the user page: */
924                 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
925
926                 PROC_LOCK(p);
927                 --p->p_lock;
928                 PROC_UNLOCK(p);
929         } else {
930                 /*
931                  * Don't have to worry about process locking or stacks in the
932                  * kernel.
933                  */
934                 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
935         }
936         if (rv == KERN_SUCCESS) {
937 #ifdef HWPMC_HOOKS
938                 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
939                         PMC_SOFT_CALL_TF( , , page_fault, all, frame);
940                         if (ftype == VM_PROT_READ)
941                                 PMC_SOFT_CALL_TF( , , page_fault, read,
942                                     frame);
943                         else
944                                 PMC_SOFT_CALL_TF( , , page_fault, write,
945                                     frame);
946                 }
947 #endif
948                 return (0);
949         }
950 nogo:
951         if (!usermode) {
952                 if (td->td_intr_nesting_level == 0 &&
953                     curpcb->pcb_onfault != NULL) {
954                         frame->tf_eip = (int)curpcb->pcb_onfault;
955                         return (0);
956                 }
957                 trap_fatal(frame, eva);
958                 return (-1);
959         }
960
961         return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
962 }
963
964 static void
965 trap_fatal(frame, eva)
966         struct trapframe *frame;
967         vm_offset_t eva;
968 {
969         int code, ss, esp;
970         u_int type;
971         struct soft_segment_descriptor softseg;
972         char *msg;
973
974         code = frame->tf_err;
975         type = frame->tf_trapno;
976         sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
977
978         if (type <= MAX_TRAP_MSG)
979                 msg = trap_msg[type];
980         else
981                 msg = "UNKNOWN";
982         printf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
983             frame->tf_eflags & PSL_VM ? "vm86" :
984             ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
985 #ifdef SMP
986         /* two separate prints in case of a trap on an unmapped page */
987         printf("cpuid = %d; ", PCPU_GET(cpuid));
988         printf("apic id = %02x\n", PCPU_GET(apic_id));
989 #endif
990         if (type == T_PAGEFLT) {
991                 printf("fault virtual address   = 0x%x\n", eva);
992                 printf("fault code              = %s %s, %s\n",
993                         code & PGEX_U ? "user" : "supervisor",
994                         code & PGEX_W ? "write" : "read",
995                         code & PGEX_P ? "protection violation" : "page not present");
996         }
997         printf("instruction pointer     = 0x%x:0x%x\n",
998                frame->tf_cs & 0xffff, frame->tf_eip);
999         if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
1000                 ss = frame->tf_ss & 0xffff;
1001                 esp = frame->tf_esp;
1002         } else {
1003                 ss = GSEL(GDATA_SEL, SEL_KPL);
1004                 esp = (int)&frame->tf_esp;
1005         }
1006         printf("stack pointer           = 0x%x:0x%x\n", ss, esp);
1007         printf("frame pointer           = 0x%x:0x%x\n", ss, frame->tf_ebp);
1008         printf("code segment            = base 0x%x, limit 0x%x, type 0x%x\n",
1009                softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
1010         printf("                        = DPL %d, pres %d, def32 %d, gran %d\n",
1011                softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
1012                softseg.ssd_gran);
1013         printf("processor eflags        = ");
1014         if (frame->tf_eflags & PSL_T)
1015                 printf("trace trap, ");
1016         if (frame->tf_eflags & PSL_I)
1017                 printf("interrupt enabled, ");
1018         if (frame->tf_eflags & PSL_NT)
1019                 printf("nested task, ");
1020         if (frame->tf_eflags & PSL_RF)
1021                 printf("resume, ");
1022         if (frame->tf_eflags & PSL_VM)
1023                 printf("vm86, ");
1024         printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
1025         printf("current process         = ");
1026         if (curproc) {
1027                 printf("%lu (%s)\n", (u_long)curproc->p_pid, curthread->td_name);
1028         } else {
1029                 printf("Idle\n");
1030         }
1031
1032 #ifdef KDB
1033         if (debugger_on_panic || kdb_active) {
1034                 frame->tf_err = eva;    /* smuggle fault address to ddb */
1035                 if (kdb_trap(type, 0, frame)) {
1036                         frame->tf_err = code;   /* restore error code */
1037                         return;
1038                 }
1039                 frame->tf_err = code;           /* restore error code */
1040         }
1041 #endif
1042         printf("trap number             = %d\n", type);
1043         if (type <= MAX_TRAP_MSG)
1044                 panic("%s", trap_msg[type]);
1045         else
1046                 panic("unknown/reserved trap");
1047 }
1048
1049 /*
1050  * Double fault handler. Called when a fault occurs while writing
1051  * a frame for a trap/exception onto the stack. This usually occurs
1052  * when the stack overflows (such is the case with infinite recursion,
1053  * for example).
1054  *
1055  * XXX Note that the current PTD gets replaced by IdlePTD when the
1056  * task switch occurs. This means that the stack that was active at
1057  * the time of the double fault is not available at <kstack> unless
1058  * the machine was idle when the double fault occurred. The downside
1059  * of this is that "trace <ebp>" in ddb won't work.
1060  */
1061 void
1062 dblfault_handler()
1063 {
1064 #ifdef KDTRACE_HOOKS
1065         if (dtrace_doubletrap_func != NULL)
1066                 (*dtrace_doubletrap_func)();
1067 #endif
1068         printf("\nFatal double fault:\n");
1069         printf("eip = 0x%x\n", PCPU_GET(common_tss.tss_eip));
1070         printf("esp = 0x%x\n", PCPU_GET(common_tss.tss_esp));
1071         printf("ebp = 0x%x\n", PCPU_GET(common_tss.tss_ebp));
1072 #ifdef SMP
1073         /* two separate prints in case of a trap on an unmapped page */
1074         printf("cpuid = %d; ", PCPU_GET(cpuid));
1075         printf("apic id = %02x\n", PCPU_GET(apic_id));
1076 #endif
1077         panic("double fault");
1078 }
1079
1080 int
1081 cpu_fetch_syscall_args(struct thread *td, struct syscall_args *sa)
1082 {
1083         struct proc *p;
1084         struct trapframe *frame;
1085         caddr_t params;
1086         int error;
1087
1088         p = td->td_proc;
1089         frame = td->td_frame;
1090
1091         params = (caddr_t)frame->tf_esp + sizeof(int);
1092         sa->code = frame->tf_eax;
1093
1094         /*
1095          * Need to check if this is a 32 bit or 64 bit syscall.
1096          */
1097         if (sa->code == SYS_syscall) {
1098                 /*
1099                  * Code is first argument, followed by actual args.
1100                  */
1101                 sa->code = fuword(params);
1102                 params += sizeof(int);
1103         } else if (sa->code == SYS___syscall) {
1104                 /*
1105                  * Like syscall, but code is a quad, so as to maintain
1106                  * quad alignment for the rest of the arguments.
1107                  */
1108                 sa->code = fuword(params);
1109                 params += sizeof(quad_t);
1110         }
1111
1112         if (p->p_sysent->sv_mask)
1113                 sa->code &= p->p_sysent->sv_mask;
1114         if (sa->code >= p->p_sysent->sv_size)
1115                 sa->callp = &p->p_sysent->sv_table[0];
1116         else
1117                 sa->callp = &p->p_sysent->sv_table[sa->code];
1118         sa->narg = sa->callp->sy_narg;
1119
1120         if (params != NULL && sa->narg != 0)
1121                 error = copyin(params, (caddr_t)sa->args,
1122                     (u_int)(sa->narg * sizeof(int)));
1123         else
1124                 error = 0;
1125
1126         if (error == 0) {
1127                 td->td_retval[0] = 0;
1128                 td->td_retval[1] = frame->tf_edx;
1129         }
1130                 
1131         return (error);
1132 }
1133
1134 #include "../../kern/subr_syscall.c"
1135
1136 /*
1137  * syscall - system call request C handler.  A system call is
1138  * essentially treated as a trap by reusing the frame layout.
1139  */
1140 void
1141 syscall(struct trapframe *frame)
1142 {
1143         struct thread *td;
1144         struct syscall_args sa;
1145         register_t orig_tf_eflags;
1146         int error;
1147         ksiginfo_t ksi;
1148
1149 #ifdef DIAGNOSTIC
1150         if (ISPL(frame->tf_cs) != SEL_UPL) {
1151                 panic("syscall");
1152                 /* NOT REACHED */
1153         }
1154 #endif
1155         orig_tf_eflags = frame->tf_eflags;
1156
1157         td = curthread;
1158         td->td_frame = frame;
1159
1160         error = syscallenter(td, &sa);
1161
1162         /*
1163          * Traced syscall.
1164          */
1165         if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1166                 frame->tf_eflags &= ~PSL_T;
1167                 ksiginfo_init_trap(&ksi);
1168                 ksi.ksi_signo = SIGTRAP;
1169                 ksi.ksi_code = TRAP_TRACE;
1170                 ksi.ksi_addr = (void *)frame->tf_eip;
1171                 trapsignal(td, &ksi);
1172         }
1173
1174         KASSERT(PCB_USER_FPU(td->td_pcb),
1175             ("System call %s returning with kernel FPU ctx leaked",
1176              syscallname(td->td_proc, sa.code)));
1177         KASSERT(td->td_pcb->pcb_save == &td->td_pcb->pcb_user_save,
1178             ("System call %s returning with mangled pcb_save",
1179              syscallname(td->td_proc, sa.code)));
1180
1181         syscallret(td, error, &sa);
1182 }