]> CyberLeo.Net >> Repos - FreeBSD/stable/8.git/blob - sys/amd64/amd64/trap.c
MFC r207463:
[FreeBSD/stable/8.git] / sys / amd64 / amd64 / 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  * AMD64 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_ktrace.h"
54
55 #include <sys/param.h>
56 #include <sys/bus.h>
57 #include <sys/systm.h>
58 #include <sys/proc.h>
59 #include <sys/pioctl.h>
60 #include <sys/ptrace.h>
61 #include <sys/kdb.h>
62 #include <sys/kernel.h>
63 #include <sys/ktr.h>
64 #include <sys/lock.h>
65 #include <sys/mutex.h>
66 #include <sys/resourcevar.h>
67 #include <sys/signalvar.h>
68 #include <sys/syscall.h>
69 #include <sys/sysctl.h>
70 #include <sys/sysent.h>
71 #include <sys/uio.h>
72 #include <sys/vmmeter.h>
73 #ifdef KTRACE
74 #include <sys/ktrace.h>
75 #endif
76 #ifdef HWPMC_HOOKS
77 #include <sys/pmckern.h>
78 #endif
79 #include <security/audit/audit.h>
80
81 #include <vm/vm.h>
82 #include <vm/vm_param.h>
83 #include <vm/pmap.h>
84 #include <vm/vm_kern.h>
85 #include <vm/vm_map.h>
86 #include <vm/vm_page.h>
87 #include <vm/vm_extern.h>
88
89 #include <machine/cpu.h>
90 #include <machine/intr_machdep.h>
91 #include <machine/mca.h>
92 #include <machine/md_var.h>
93 #include <machine/pcb.h>
94 #ifdef SMP
95 #include <machine/smp.h>
96 #endif
97 #include <machine/tss.h>
98
99 #ifdef KDTRACE_HOOKS
100 #include <sys/dtrace_bsd.h>
101
102 /*
103  * This is a hook which is initialised by the dtrace module
104  * to handle traps which might occur during DTrace probe
105  * execution.
106  */
107 dtrace_trap_func_t      dtrace_trap_func;
108
109 dtrace_doubletrap_func_t        dtrace_doubletrap_func;
110
111 /*
112  * This is a hook which is initialised by the systrace module
113  * when it is loaded. This keeps the DTrace syscall provider
114  * implementation opaque. 
115  */
116 systrace_probe_func_t   systrace_probe_func;
117 #endif
118
119 extern void trap(struct trapframe *frame);
120 extern void syscall(struct trapframe *frame);
121 void dblfault_handler(struct trapframe *frame);
122
123 static int trap_pfault(struct trapframe *, int);
124 static void trap_fatal(struct trapframe *, vm_offset_t);
125
126 #define MAX_TRAP_MSG            30
127 static char *trap_msg[] = {
128         "",                                     /*  0 unused */
129         "privileged instruction fault",         /*  1 T_PRIVINFLT */
130         "",                                     /*  2 unused */
131         "breakpoint instruction fault",         /*  3 T_BPTFLT */
132         "",                                     /*  4 unused */
133         "",                                     /*  5 unused */
134         "arithmetic trap",                      /*  6 T_ARITHTRAP */
135         "",                                     /*  7 unused */
136         "",                                     /*  8 unused */
137         "general protection fault",             /*  9 T_PROTFLT */
138         "trace trap",                           /* 10 T_TRCTRAP */
139         "",                                     /* 11 unused */
140         "page fault",                           /* 12 T_PAGEFLT */
141         "",                                     /* 13 unused */
142         "alignment fault",                      /* 14 T_ALIGNFLT */
143         "",                                     /* 15 unused */
144         "",                                     /* 16 unused */
145         "",                                     /* 17 unused */
146         "integer divide fault",                 /* 18 T_DIVIDE */
147         "non-maskable interrupt trap",          /* 19 T_NMI */
148         "overflow trap",                        /* 20 T_OFLOW */
149         "FPU bounds check fault",               /* 21 T_BOUND */
150         "FPU device not available",             /* 22 T_DNA */
151         "double fault",                         /* 23 T_DOUBLEFLT */
152         "FPU operand fetch fault",              /* 24 T_FPOPFLT */
153         "invalid TSS fault",                    /* 25 T_TSSFLT */
154         "segment not present fault",            /* 26 T_SEGNPFLT */
155         "stack fault",                          /* 27 T_STKFLT */
156         "machine check trap",                   /* 28 T_MCHK */
157         "SIMD floating-point exception",        /* 29 T_XMMFLT */
158         "reserved (unknown) fault",             /* 30 T_RESERVED */
159 };
160
161 #ifdef KDB
162 static int kdb_on_nmi = 1;
163 SYSCTL_INT(_machdep, OID_AUTO, kdb_on_nmi, CTLFLAG_RW,
164         &kdb_on_nmi, 0, "Go to KDB on NMI");
165 #endif
166 static int panic_on_nmi = 1;
167 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
168         &panic_on_nmi, 0, "Panic on NMI");
169 static int prot_fault_translation = 0;
170 SYSCTL_INT(_machdep, OID_AUTO, prot_fault_translation, CTLFLAG_RW,
171         &prot_fault_translation, 0, "Select signal to deliver on protection fault");
172
173 extern char *syscallnames[];
174
175 /*
176  * Exception, fault, and trap interface to the FreeBSD kernel.
177  * This common code is called from assembly language IDT gate entry
178  * routines that prepare a suitable stack frame, and restore this
179  * frame after the exception has been processed.
180  */
181
182 void
183 trap(struct trapframe *frame)
184 {
185         struct thread *td = curthread;
186         struct proc *p = td->td_proc;
187         int i = 0, ucode = 0, code;
188         u_int type;
189         register_t addr = 0;
190         ksiginfo_t ksi;
191
192         PCPU_INC(cnt.v_trap);
193         type = frame->tf_trapno;
194
195 #ifdef SMP
196         /* Handler for NMI IPIs used for stopping CPUs. */
197         if (type == T_NMI) {
198                  if (ipi_nmi_handler() == 0)
199                            goto out;
200         }
201 #endif /* SMP */
202
203 #ifdef KDB
204         if (kdb_active) {
205                 kdb_reenter();
206                 goto out;
207         }
208 #endif
209
210         if (type == T_RESERVED) {
211                 trap_fatal(frame, 0);
212                 goto out;
213         }
214
215 #ifdef  HWPMC_HOOKS
216         /*
217          * CPU PMCs interrupt using an NMI.  If the PMC module is
218          * active, pass the 'rip' value to the PMC module's interrupt
219          * handler.  A return value of '1' from the handler means that
220          * the NMI was handled by it and we can return immediately.
221          */
222         if (type == T_NMI && pmc_intr &&
223             (*pmc_intr)(PCPU_GET(cpuid), frame))
224                 goto out;
225 #endif
226
227         if (type == T_MCHK) {
228                 if (!mca_intr())
229                         trap_fatal(frame, 0);
230                 goto out;
231         }
232
233 #ifdef KDTRACE_HOOKS
234         /*
235          * A trap can occur while DTrace executes a probe. Before
236          * executing the probe, DTrace blocks re-scheduling and sets
237          * a flag in it's per-cpu flags to indicate that it doesn't
238          * want to fault. On returning from the the probe, the no-fault
239          * flag is cleared and finally re-scheduling is enabled.
240          *
241          * If the DTrace kernel module has registered a trap handler,
242          * call it and if it returns non-zero, assume that it has
243          * handled the trap and modified the trap frame so that this
244          * function can return normally.
245          */
246         if (dtrace_trap_func != NULL)
247                 if ((*dtrace_trap_func)(frame, type))
248                         goto out;
249 #endif
250
251         if ((frame->tf_rflags & PSL_I) == 0) {
252                 /*
253                  * Buggy application or kernel code has disabled
254                  * interrupts and then trapped.  Enabling interrupts
255                  * now is wrong, but it is better than running with
256                  * interrupts disabled until they are accidentally
257                  * enabled later.
258                  */
259                 if (ISPL(frame->tf_cs) == SEL_UPL)
260                         uprintf(
261                             "pid %ld (%s): trap %d with interrupts disabled\n",
262                             (long)curproc->p_pid, curthread->td_name, type);
263                 else if (type != T_NMI && type != T_BPTFLT &&
264                     type != T_TRCTRAP) {
265                         /*
266                          * XXX not quite right, since this may be for a
267                          * multiple fault in user mode.
268                          */
269                         printf("kernel trap %d with interrupts disabled\n",
270                             type);
271
272                         /*
273                          * We shouldn't enable interrupts while holding a
274                          * spin lock or servicing an NMI.
275                          */
276                         if (type != T_NMI && td->td_md.md_spinlock_count == 0)
277                                 enable_intr();
278                 }
279         }
280
281         code = frame->tf_err;
282         if (type == T_PAGEFLT) {
283                 /*
284                  * If we get a page fault while in a critical section, then
285                  * it is most likely a fatal kernel page fault.  The kernel
286                  * is already going to panic trying to get a sleep lock to
287                  * do the VM lookup, so just consider it a fatal trap so the
288                  * kernel can print out a useful trap message and even get
289                  * to the debugger.
290                  *
291                  * If we get a page fault while holding a non-sleepable
292                  * lock, then it is most likely a fatal kernel page fault.
293                  * If WITNESS is enabled, then it's going to whine about
294                  * bogus LORs with various VM locks, so just skip to the
295                  * fatal trap handling directly.
296                  */
297                 if (td->td_critnest != 0 ||
298                     WITNESS_CHECK(WARN_SLEEPOK | WARN_GIANTOK, NULL,
299                     "Kernel page fault") != 0)
300                         trap_fatal(frame, frame->tf_addr);
301         }
302
303         if (ISPL(frame->tf_cs) == SEL_UPL) {
304                 /* user trap */
305
306                 td->td_pticks = 0;
307                 td->td_frame = frame;
308                 addr = frame->tf_rip;
309                 if (td->td_ucred != p->p_ucred) 
310                         cred_update_thread(td);
311
312                 switch (type) {
313                 case T_PRIVINFLT:       /* privileged instruction fault */
314                         i = SIGILL;
315                         ucode = ILL_PRVOPC;
316                         break;
317
318                 case T_BPTFLT:          /* bpt instruction fault */
319                 case T_TRCTRAP:         /* trace trap */
320                         enable_intr();
321                         frame->tf_rflags &= ~PSL_T;
322                         i = SIGTRAP;
323                         ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
324                         break;
325
326                 case T_ARITHTRAP:       /* arithmetic trap */
327                         ucode = fputrap();
328                         if (ucode == -1)
329                                 goto userout;
330                         i = SIGFPE;
331                         break;
332
333                 case T_PROTFLT:         /* general protection fault */
334                         i = SIGBUS;
335                         ucode = BUS_OBJERR;
336                         break;
337                 case T_STKFLT:          /* stack fault */
338                 case T_SEGNPFLT:        /* segment not present fault */
339                         i = SIGBUS;
340                         ucode = BUS_ADRERR;
341                         break;
342                 case T_TSSFLT:          /* invalid TSS fault */
343                         i = SIGBUS;
344                         ucode = BUS_OBJERR;
345                         break;
346                 case T_DOUBLEFLT:       /* double fault */
347                 default:
348                         i = SIGBUS;
349                         ucode = BUS_OBJERR;
350                         break;
351
352                 case T_PAGEFLT:         /* page fault */
353                         addr = frame->tf_addr;
354                         i = trap_pfault(frame, TRUE);
355                         if (i == -1)
356                                 goto userout;
357                         if (i == 0)
358                                 goto user;
359
360                         if (i == SIGSEGV)
361                                 ucode = SEGV_MAPERR;
362                         else {
363                                 if (prot_fault_translation == 0) {
364                                         /*
365                                          * Autodetect.
366                                          * This check also covers the images
367                                          * without the ABI-tag ELF note.
368                                          */
369                                         if (SV_CURPROC_ABI() ==
370                                             SV_ABI_FREEBSD &&
371                                             p->p_osrel >= 700004) {
372                                                 i = SIGSEGV;
373                                                 ucode = SEGV_ACCERR;
374                                         } else {
375                                                 i = SIGBUS;
376                                                 ucode = BUS_PAGE_FAULT;
377                                         }
378                                 } else if (prot_fault_translation == 1) {
379                                         /*
380                                          * Always compat mode.
381                                          */
382                                         i = SIGBUS;
383                                         ucode = BUS_PAGE_FAULT;
384                                 } else {
385                                         /*
386                                          * Always SIGSEGV mode.
387                                          */
388                                         i = SIGSEGV;
389                                         ucode = SEGV_ACCERR;
390                                 }
391                         }
392                         break;
393
394                 case T_DIVIDE:          /* integer divide fault */
395                         ucode = FPE_INTDIV;
396                         i = SIGFPE;
397                         break;
398
399 #ifdef DEV_ISA
400                 case T_NMI:
401                         /* machine/parity/power fail/"kitchen sink" faults */
402                         if (isa_nmi(code) == 0) {
403 #ifdef KDB
404                                 /*
405                                  * NMI can be hooked up to a pushbutton
406                                  * for debugging.
407                                  */
408                                 if (kdb_on_nmi) {
409                                         printf ("NMI ... going to debugger\n");
410                                         kdb_trap(type, 0, frame);
411                                 }
412 #endif /* KDB */
413                                 goto userout;
414                         } else if (panic_on_nmi)
415                                 panic("NMI indicates hardware failure");
416                         break;
417 #endif /* DEV_ISA */
418
419                 case T_OFLOW:           /* integer overflow fault */
420                         ucode = FPE_INTOVF;
421                         i = SIGFPE;
422                         break;
423
424                 case T_BOUND:           /* bounds check fault */
425                         ucode = FPE_FLTSUB;
426                         i = SIGFPE;
427                         break;
428
429                 case T_DNA:
430                         /* transparent fault (due to context switch "late") */
431                         fpudna();
432                         goto userout;
433
434                 case T_FPOPFLT:         /* FPU operand fetch fault */
435                         ucode = ILL_COPROC;
436                         i = SIGILL;
437                         break;
438
439                 case T_XMMFLT:          /* SIMD floating-point exception */
440                         ucode = 0; /* XXX */
441                         i = SIGFPE;
442                         break;
443                 }
444         } else {
445                 /* kernel trap */
446
447                 KASSERT(cold || td->td_ucred != NULL,
448                     ("kernel trap doesn't have ucred"));
449                 switch (type) {
450                 case T_PAGEFLT:                 /* page fault */
451                         (void) trap_pfault(frame, FALSE);
452                         goto out;
453
454                 case T_DNA:
455                         /*
456                          * The kernel is apparently using fpu for copying.
457                          * XXX this should be fatal unless the kernel has
458                          * registered such use.
459                          */
460                         fpudna();
461                         printf("fpudna in kernel mode!\n");
462                         goto out;
463
464                 case T_STKFLT:          /* stack fault */
465                         break;
466
467                 case T_PROTFLT:         /* general protection fault */
468                 case T_SEGNPFLT:        /* segment not present fault */
469                         if (td->td_intr_nesting_level != 0)
470                                 break;
471
472                         /*
473                          * Invalid segment selectors and out of bounds
474                          * %rip's and %rsp's can be set up in user mode.
475                          * This causes a fault in kernel mode when the
476                          * kernel tries to return to user mode.  We want
477                          * to get this fault so that we can fix the
478                          * problem here and not have to check all the
479                          * selectors and pointers when the user changes
480                          * them.
481                          */
482                         if (frame->tf_rip == (long)doreti_iret) {
483                                 frame->tf_rip = (long)doreti_iret_fault;
484                                 goto out;
485                         }
486                         if (frame->tf_rip == (long)ld_ds) {
487                                 frame->tf_rip = (long)ds_load_fault;
488                                 frame->tf_ds = _udatasel;
489                                 goto out;
490                         }
491                         if (frame->tf_rip == (long)ld_es) {
492                                 frame->tf_rip = (long)es_load_fault;
493                                 frame->tf_es = _udatasel;
494                                 goto out;
495                         }
496                         if (frame->tf_rip == (long)ld_fs) {
497                                 frame->tf_rip = (long)fs_load_fault;
498                                 frame->tf_fs = _ufssel;
499                                 goto out;
500                         }
501                         if (frame->tf_rip == (long)ld_gs) {
502                                 frame->tf_rip = (long)gs_load_fault;
503                                 frame->tf_gs = _ugssel;
504                                 goto out;
505                         }
506                         if (frame->tf_rip == (long)ld_gsbase) {
507                                 frame->tf_rip = (long)gsbase_load_fault;
508                                 goto out;
509                         }
510                         if (frame->tf_rip == (long)ld_fsbase) {
511                                 frame->tf_rip = (long)fsbase_load_fault;
512                                 goto out;
513                         }
514                         if (PCPU_GET(curpcb)->pcb_onfault != NULL) {
515                                 frame->tf_rip =
516                                     (long)PCPU_GET(curpcb)->pcb_onfault;
517                                 goto out;
518                         }
519                         break;
520
521                 case T_TSSFLT:
522                         /*
523                          * PSL_NT can be set in user mode and isn't cleared
524                          * automatically when the kernel is entered.  This
525                          * causes a TSS fault when the kernel attempts to
526                          * `iret' because the TSS link is uninitialized.  We
527                          * want to get this fault so that we can fix the
528                          * problem here and not every time the kernel is
529                          * entered.
530                          */
531                         if (frame->tf_rflags & PSL_NT) {
532                                 frame->tf_rflags &= ~PSL_NT;
533                                 goto out;
534                         }
535                         break;
536
537                 case T_TRCTRAP:  /* trace trap */
538                         /*
539                          * Ignore debug register trace traps due to
540                          * accesses in the user's address space, which
541                          * can happen under several conditions such as
542                          * if a user sets a watchpoint on a buffer and
543                          * then passes that buffer to a system call.
544                          * We still want to get TRCTRAPS for addresses
545                          * in kernel space because that is useful when
546                          * debugging the kernel.
547                          */
548                         if (user_dbreg_trap()) {
549                                 /*
550                                  * Reset breakpoint bits because the
551                                  * processor doesn't
552                                  */
553                                 /* XXX check upper bits here */
554                                 load_dr6(rdr6() & 0xfffffff0);
555                                 goto out;
556                         }
557                         /*
558                          * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
559                          */
560                 case T_BPTFLT:
561                         /*
562                          * If KDB is enabled, let it handle the debugger trap.
563                          * Otherwise, debugger traps "can't happen".
564                          */
565 #ifdef KDB
566                         if (kdb_trap(type, 0, frame))
567                                 goto out;
568 #endif
569                         break;
570
571 #ifdef DEV_ISA
572                 case T_NMI:
573                         /* machine/parity/power fail/"kitchen sink" faults */
574                         if (isa_nmi(code) == 0) {
575 #ifdef KDB
576                                 /*
577                                  * NMI can be hooked up to a pushbutton
578                                  * for debugging.
579                                  */
580                                 if (kdb_on_nmi) {
581                                         printf ("NMI ... going to debugger\n");
582                                         kdb_trap(type, 0, frame);
583                                 }
584 #endif /* KDB */
585                                 goto out;
586                         } else if (panic_on_nmi == 0)
587                                 goto out;
588                         /* FALLTHROUGH */
589 #endif /* DEV_ISA */
590                 }
591
592                 trap_fatal(frame, 0);
593                 goto out;
594         }
595
596         /* Translate fault for emulators (e.g. Linux) */
597         if (*p->p_sysent->sv_transtrap)
598                 i = (*p->p_sysent->sv_transtrap)(i, type);
599
600         ksiginfo_init_trap(&ksi);
601         ksi.ksi_signo = i;
602         ksi.ksi_code = ucode;
603         ksi.ksi_trapno = type;
604         ksi.ksi_addr = (void *)addr;
605         trapsignal(td, &ksi);
606
607 user:
608         userret(td, frame);
609         mtx_assert(&Giant, MA_NOTOWNED);
610 userout:
611 out:
612         return;
613 }
614
615 static int
616 trap_pfault(frame, usermode)
617         struct trapframe *frame;
618         int usermode;
619 {
620         vm_offset_t va;
621         struct vmspace *vm = NULL;
622         vm_map_t map;
623         int rv = 0;
624         vm_prot_t ftype;
625         struct thread *td = curthread;
626         struct proc *p = td->td_proc;
627         vm_offset_t eva = frame->tf_addr;
628
629         va = trunc_page(eva);
630         if (va >= VM_MIN_KERNEL_ADDRESS) {
631                 /*
632                  * Don't allow user-mode faults in kernel address space.
633                  */
634                 if (usermode)
635                         goto nogo;
636
637                 map = kernel_map;
638         } else {
639                 /*
640                  * This is a fault on non-kernel virtual memory.
641                  * vm is initialized above to NULL. If curproc is NULL
642                  * or curproc->p_vmspace is NULL the fault is fatal.
643                  */
644                 if (p != NULL)
645                         vm = p->p_vmspace;
646
647                 if (vm == NULL)
648                         goto nogo;
649
650                 map = &vm->vm_map;
651         }
652
653         /*
654          * PGEX_I is defined only if the execute disable bit capability is
655          * supported and enabled.
656          */
657         if (frame->tf_err & PGEX_W)
658                 ftype = VM_PROT_WRITE;
659         else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
660                 ftype = VM_PROT_EXECUTE;
661         else
662                 ftype = VM_PROT_READ;
663
664         if (map != kernel_map) {
665                 /*
666                  * Keep swapout from messing with us during this
667                  *      critical time.
668                  */
669                 PROC_LOCK(p);
670                 ++p->p_lock;
671                 PROC_UNLOCK(p);
672
673                 /* Fault in the user page: */
674                 rv = vm_fault(map, va, ftype,
675                               (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
676                                                       : VM_FAULT_NORMAL);
677
678                 PROC_LOCK(p);
679                 --p->p_lock;
680                 PROC_UNLOCK(p);
681         } else {
682                 /*
683                  * Don't have to worry about process locking or stacks in the
684                  * kernel.
685                  */
686                 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
687         }
688         if (rv == KERN_SUCCESS)
689                 return (0);
690 nogo:
691         if (!usermode) {
692                 if (td->td_intr_nesting_level == 0 &&
693                     PCPU_GET(curpcb)->pcb_onfault != NULL) {
694                         frame->tf_rip = (long)PCPU_GET(curpcb)->pcb_onfault;
695                         return (0);
696                 }
697                 trap_fatal(frame, eva);
698                 return (-1);
699         }
700
701         return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
702 }
703
704 static void
705 trap_fatal(frame, eva)
706         struct trapframe *frame;
707         vm_offset_t eva;
708 {
709         int code, ss;
710         u_int type;
711         long esp;
712         struct soft_segment_descriptor softseg;
713         char *msg;
714
715         code = frame->tf_err;
716         type = frame->tf_trapno;
717         sdtossd(&gdt[NGDT * PCPU_GET(cpuid) + IDXSEL(frame->tf_cs & 0xffff)],
718             &softseg);
719
720         if (type <= MAX_TRAP_MSG)
721                 msg = trap_msg[type];
722         else
723                 msg = "UNKNOWN";
724         printf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
725             ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
726 #ifdef SMP
727         /* two separate prints in case of a trap on an unmapped page */
728         printf("cpuid = %d; ", PCPU_GET(cpuid));
729         printf("apic id = %02x\n", PCPU_GET(apic_id));
730 #endif
731         if (type == T_PAGEFLT) {
732                 printf("fault virtual address   = 0x%lx\n", eva);
733                 printf("fault code              = %s %s %s, %s\n",
734                         code & PGEX_U ? "user" : "supervisor",
735                         code & PGEX_W ? "write" : "read",
736                         code & PGEX_I ? "instruction" : "data",
737                         code & PGEX_P ? "protection violation" : "page not present");
738         }
739         printf("instruction pointer     = 0x%lx:0x%lx\n",
740                frame->tf_cs & 0xffff, frame->tf_rip);
741         if (ISPL(frame->tf_cs) == SEL_UPL) {
742                 ss = frame->tf_ss & 0xffff;
743                 esp = frame->tf_rsp;
744         } else {
745                 ss = GSEL(GDATA_SEL, SEL_KPL);
746                 esp = (long)&frame->tf_rsp;
747         }
748         printf("stack pointer           = 0x%x:0x%lx\n", ss, esp);
749         printf("frame pointer           = 0x%x:0x%lx\n", ss, frame->tf_rbp);
750         printf("code segment            = base 0x%lx, limit 0x%lx, type 0x%x\n",
751                softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
752         printf("                        = DPL %d, pres %d, long %d, def32 %d, gran %d\n",
753                softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32,
754                softseg.ssd_gran);
755         printf("processor eflags        = ");
756         if (frame->tf_rflags & PSL_T)
757                 printf("trace trap, ");
758         if (frame->tf_rflags & PSL_I)
759                 printf("interrupt enabled, ");
760         if (frame->tf_rflags & PSL_NT)
761                 printf("nested task, ");
762         if (frame->tf_rflags & PSL_RF)
763                 printf("resume, ");
764         printf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12);
765         printf("current process         = ");
766         if (curproc) {
767                 printf("%lu (%s)\n",
768                     (u_long)curproc->p_pid, curthread->td_name ?
769                     curthread->td_name : "");
770         } else {
771                 printf("Idle\n");
772         }
773
774 #ifdef KDB
775         if (debugger_on_panic || kdb_active)
776                 if (kdb_trap(type, 0, frame))
777                         return;
778 #endif
779         printf("trap number             = %d\n", type);
780         if (type <= MAX_TRAP_MSG)
781                 panic("%s", trap_msg[type]);
782         else
783                 panic("unknown/reserved trap");
784 }
785
786 /*
787  * Double fault handler. Called when a fault occurs while writing
788  * a frame for a trap/exception onto the stack. This usually occurs
789  * when the stack overflows (such is the case with infinite recursion,
790  * for example).
791  */
792 void
793 dblfault_handler(struct trapframe *frame)
794 {
795 #ifdef KDTRACE_HOOKS
796         if (dtrace_doubletrap_func != NULL)
797                 (*dtrace_doubletrap_func)();
798 #endif
799         printf("\nFatal double fault\n");
800         printf("rip = 0x%lx\n", frame->tf_rip);
801         printf("rsp = 0x%lx\n", frame->tf_rsp);
802         printf("rbp = 0x%lx\n", frame->tf_rbp);
803 #ifdef SMP
804         /* two separate prints in case of a trap on an unmapped page */
805         printf("cpuid = %d; ", PCPU_GET(cpuid));
806         printf("apic id = %02x\n", PCPU_GET(apic_id));
807 #endif
808         panic("double fault");
809 }
810
811 struct syscall_args {
812         u_int code;
813         struct sysent *callp;
814         register_t args[8];
815         register_t *argp;
816         int narg;
817 };
818
819 static int
820 fetch_syscall_args(struct thread *td, struct syscall_args *sa)
821 {
822         struct proc *p;
823         struct trapframe *frame;
824         caddr_t params;
825         int reg, regcnt, error;
826
827         p = td->td_proc;
828         frame = td->td_frame;
829         reg = 0;
830         regcnt = 6;
831
832         params = (caddr_t)frame->tf_rsp + sizeof(register_t);
833         sa->code = frame->tf_rax;
834
835         if (p->p_sysent->sv_prepsyscall) {
836                 (*p->p_sysent->sv_prepsyscall)(frame, (int *)sa->args,
837                     &sa->code, &params);
838         } else {
839                 if (sa->code == SYS_syscall || sa->code == SYS___syscall) {
840                         sa->code = frame->tf_rdi;
841                         reg++;
842                         regcnt--;
843                 }
844         }
845         if (p->p_sysent->sv_mask)
846                 sa->code &= p->p_sysent->sv_mask;
847
848         if (sa->code >= p->p_sysent->sv_size)
849                 sa->callp = &p->p_sysent->sv_table[0];
850         else
851                 sa->callp = &p->p_sysent->sv_table[sa->code];
852
853         sa->narg = sa->callp->sy_narg;
854         KASSERT(sa->narg <= sizeof(sa->args) / sizeof(sa->args[0]),
855             ("Too many syscall arguments!"));
856         error = 0;
857         sa->argp = &frame->tf_rdi;
858         sa->argp += reg;
859         bcopy(sa->argp, sa->args, sizeof(sa->args[0]) * regcnt);
860         if (sa->narg > regcnt) {
861                 KASSERT(params != NULL, ("copyin args with no params!"));
862                 error = copyin(params, &sa->args[regcnt],
863                     (sa->narg - regcnt) * sizeof(sa->args[0]));
864         }
865         sa->argp = &sa->args[0];
866
867         /*
868          * This may result in two records if debugger modified
869          * registers or memory during sleep at stop/ptrace point.
870          */
871 #ifdef KTRACE
872         if (KTRPOINT(td, KTR_SYSCALL))
873                 ktrsyscall(sa->code, sa->narg, sa->argp);
874 #endif
875         return (error);
876 }
877
878 /*
879  *      syscall -       system call request C handler
880  *
881  *      A system call is essentially treated as a trap.
882  */
883 void
884 syscall(struct trapframe *frame)
885 {
886         struct thread *td;
887         struct proc *p;
888         struct syscall_args sa;
889         register_t orig_tf_rflags;
890         int error;
891         ksiginfo_t ksi;
892
893         PCPU_INC(cnt.v_syscall);
894         td = curthread;
895         p = td->td_proc;
896         td->td_syscalls++;
897
898 #ifdef DIAGNOSTIC
899         if (ISPL(frame->tf_cs) != SEL_UPL) {
900                 panic("syscall");
901                 /* NOT REACHED */
902         }
903 #endif
904
905         td->td_pticks = 0;
906         td->td_frame = frame;
907         if (td->td_ucred != p->p_ucred) 
908                 cred_update_thread(td);
909         orig_tf_rflags = frame->tf_rflags;
910         if (p->p_flag & P_TRACED) {
911                 PROC_LOCK(p);
912                 td->td_dbgflags &= ~TDB_USERWR;
913                 PROC_UNLOCK(p);
914         }
915         error = fetch_syscall_args(td, &sa);
916
917         CTR4(KTR_SYSC, "syscall enter thread %p pid %d proc %s code %d", td,
918             td->td_proc->p_pid, td->td_name, sa.code);
919
920         if (error == 0) {
921                 td->td_retval[0] = 0;
922                 td->td_retval[1] = frame->tf_rdx;
923
924                 STOPEVENT(p, S_SCE, sa.narg);
925                 PTRACESTOP_SC(p, td, S_PT_SCE);
926                 if (td->td_dbgflags & TDB_USERWR) {
927                         /*
928                          * Reread syscall number and arguments if
929                          * debugger modified registers or memory.
930                          */
931                         error = fetch_syscall_args(td, &sa);
932                         if (error != 0)
933                                 goto retval;
934                         td->td_retval[1] = frame->tf_rdx;
935                 }
936
937 #ifdef KDTRACE_HOOKS
938                 /*
939                  * If the systrace module has registered it's probe
940                  * callback and if there is a probe active for the
941                  * syscall 'entry', process the probe.
942                  */
943                 if (systrace_probe_func != NULL && sa.callp->sy_entry != 0)
944                         (*systrace_probe_func)(sa.callp->sy_entry, sa.code,
945                             sa.callp, sa.args);
946 #endif
947
948                 AUDIT_SYSCALL_ENTER(sa.code, td);
949                 error = (*sa.callp->sy_call)(td, sa.argp);
950                 AUDIT_SYSCALL_EXIT(error, td);
951
952                 /* Save the latest error return value. */
953                 td->td_errno = error;
954
955 #ifdef KDTRACE_HOOKS
956                 /*
957                  * If the systrace module has registered it's probe
958                  * callback and if there is a probe active for the
959                  * syscall 'return', process the probe.
960                  */
961                 if (systrace_probe_func != NULL && sa.callp->sy_return != 0)
962                         (*systrace_probe_func)(sa.callp->sy_return, sa.code,
963                             sa.callp, sa.args);
964 #endif
965         }
966  retval:
967         cpu_set_syscall_retval(td, error);
968
969         /*
970          * Traced syscall.
971          */
972         if (orig_tf_rflags & PSL_T) {
973                 frame->tf_rflags &= ~PSL_T;
974                 ksiginfo_init_trap(&ksi);
975                 ksi.ksi_signo = SIGTRAP;
976                 ksi.ksi_code = TRAP_TRACE;
977                 ksi.ksi_addr = (void *)frame->tf_rip;
978                 trapsignal(td, &ksi);
979         }
980
981         /*
982          * Check for misbehavior.
983          */
984         WITNESS_WARN(WARN_PANIC, NULL, "System call %s returning",
985             (sa.code >= 0 && sa.code < SYS_MAXSYSCALL) ?
986              syscallnames[sa.code] : "???");
987         KASSERT(td->td_critnest == 0,
988             ("System call %s returning in a critical section",
989             (sa.code >= 0 && sa.code < SYS_MAXSYSCALL) ?
990              syscallnames[sa.code] : "???"));
991         KASSERT(td->td_locks == 0,
992             ("System call %s returning with %d locks held",
993             (sa.code >= 0 && sa.code < SYS_MAXSYSCALL) ?
994              syscallnames[sa.code] : "???", td->td_locks));
995
996         /*
997          * Handle reschedule and other end-of-syscall issues
998          */
999         userret(td, frame);
1000
1001         CTR4(KTR_SYSC, "syscall exit thread %p pid %d proc %s code %d", td,
1002             td->td_proc->p_pid, td->td_name, sa.code);
1003
1004 #ifdef KTRACE
1005         if (KTRPOINT(td, KTR_SYSRET))
1006                 ktrsysret(sa.code, error, td->td_retval[0]);
1007 #endif
1008
1009         /*
1010          * This works because errno is findable through the
1011          * register set.  If we ever support an emulation where this
1012          * is not the case, this code will need to be revisited.
1013          */
1014         STOPEVENT(p, S_SCX, sa.code);
1015
1016         PTRACESTOP_SC(p, td, S_PT_SCX);
1017 }