FreeBSD/Linux Kernel Cross Reference
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: releng/10.0/sys/i386/i386/trap.c 251324 2013-06-03 17:40:05Z kib $");
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, ®s);
283 if (type == T_DTRACE_PROBE &&
284 dtrace_fasttrap_probe_ptr != NULL &&
285 dtrace_fasttrap_probe_ptr(®s) == 0)
286 goto out;
287 if (type == T_BPTFLT &&
288 dtrace_pid_probe_ptr != NULL &&
289 dtrace_pid_probe_ptr(®s) == 0)
290 goto out;
291 if (type == T_DTRACE_RET &&
292 dtrace_return_probe_ptr != NULL &&
293 dtrace_return_probe_ptr(®s) == 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_NMI && type != T_BPTFLT &&
314 type != T_TRCTRAP &&
315 frame->tf_eip != (int)cpu_switch_load_gs) {
316 /*
317 * XXX not quite right, since this may be for a
318 * multiple fault in user mode.
319 */
320 printf("kernel trap %d with interrupts disabled\n",
321 type);
322 /*
323 * Page faults need interrupts disabled until later,
324 * and we shouldn't enable interrupts while holding
325 * a spin lock.
326 */
327 if (type != T_PAGEFLT &&
328 td->td_md.md_spinlock_count == 0)
329 enable_intr();
330 }
331 }
332 eva = 0;
333 code = frame->tf_err;
334 if (type == T_PAGEFLT) {
335 /*
336 * For some Cyrix CPUs, %cr2 is clobbered by
337 * interrupts. This problem is worked around by using
338 * an interrupt gate for the pagefault handler. We
339 * are finally ready to read %cr2 and conditionally
340 * reenable interrupts. If we hold a spin lock, then
341 * we must not reenable interrupts. This might be a
342 * spurious page fault.
343 */
344 eva = rcr2();
345 if (td->td_md.md_spinlock_count == 0)
346 enable_intr();
347 }
348
349 if ((ISPL(frame->tf_cs) == SEL_UPL) ||
350 ((frame->tf_eflags & PSL_VM) &&
351 !(curpcb->pcb_flags & PCB_VM86CALL))) {
352 /* user trap */
353
354 td->td_pticks = 0;
355 td->td_frame = frame;
356 addr = frame->tf_eip;
357 if (td->td_ucred != p->p_ucred)
358 cred_update_thread(td);
359
360 switch (type) {
361 case T_PRIVINFLT: /* privileged instruction fault */
362 i = SIGILL;
363 ucode = ILL_PRVOPC;
364 break;
365
366 case T_BPTFLT: /* bpt instruction fault */
367 case T_TRCTRAP: /* trace trap */
368 enable_intr();
369 frame->tf_eflags &= ~PSL_T;
370 i = SIGTRAP;
371 ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
372 break;
373
374 case T_ARITHTRAP: /* arithmetic trap */
375 #ifdef DEV_NPX
376 ucode = npxtrap_x87();
377 if (ucode == -1)
378 goto userout;
379 #else
380 ucode = 0;
381 #endif
382 i = SIGFPE;
383 break;
384
385 /*
386 * The following two traps can happen in
387 * vm86 mode, and, if so, we want to handle
388 * them specially.
389 */
390 case T_PROTFLT: /* general protection fault */
391 case T_STKFLT: /* stack fault */
392 if (frame->tf_eflags & PSL_VM) {
393 i = vm86_emulate((struct vm86frame *)frame);
394 if (i == 0)
395 goto user;
396 break;
397 }
398 i = SIGBUS;
399 ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR;
400 break;
401 case T_SEGNPFLT: /* segment not present fault */
402 i = SIGBUS;
403 ucode = BUS_ADRERR;
404 break;
405 case T_TSSFLT: /* invalid TSS fault */
406 i = SIGBUS;
407 ucode = BUS_OBJERR;
408 break;
409 case T_DOUBLEFLT: /* double fault */
410 default:
411 i = SIGBUS;
412 ucode = BUS_OBJERR;
413 break;
414
415 case T_PAGEFLT: /* page fault */
416
417 i = trap_pfault(frame, TRUE, eva);
418 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
419 if (i == -2) {
420 /*
421 * The f00f hack workaround has triggered, so
422 * treat the fault as an illegal instruction
423 * (T_PRIVINFLT) instead of a page fault.
424 */
425 type = frame->tf_trapno = T_PRIVINFLT;
426
427 /* Proceed as in that case. */
428 ucode = ILL_PRVOPC;
429 i = SIGILL;
430 break;
431 }
432 #endif
433 if (i == -1)
434 goto userout;
435 if (i == 0)
436 goto user;
437
438 if (i == SIGSEGV)
439 ucode = SEGV_MAPERR;
440 else {
441 if (prot_fault_translation == 0) {
442 /*
443 * Autodetect.
444 * This check also covers the images
445 * without the ABI-tag ELF note.
446 */
447 if (SV_CURPROC_ABI() == SV_ABI_FREEBSD
448 && p->p_osrel >= P_OSREL_SIGSEGV) {
449 i = SIGSEGV;
450 ucode = SEGV_ACCERR;
451 } else {
452 i = SIGBUS;
453 ucode = BUS_PAGE_FAULT;
454 }
455 } else if (prot_fault_translation == 1) {
456 /*
457 * Always compat mode.
458 */
459 i = SIGBUS;
460 ucode = BUS_PAGE_FAULT;
461 } else {
462 /*
463 * Always SIGSEGV mode.
464 */
465 i = SIGSEGV;
466 ucode = SEGV_ACCERR;
467 }
468 }
469 addr = eva;
470 break;
471
472 case T_DIVIDE: /* integer divide fault */
473 ucode = FPE_INTDIV;
474 i = SIGFPE;
475 break;
476
477 #ifdef DEV_ISA
478 case T_NMI:
479 #ifdef POWERFAIL_NMI
480 #ifndef TIMER_FREQ
481 # define TIMER_FREQ 1193182
482 #endif
483 if (time_second - lastalert > 10) {
484 log(LOG_WARNING, "NMI: power fail\n");
485 sysbeep(880, hz);
486 lastalert = time_second;
487 }
488 goto userout;
489 #else /* !POWERFAIL_NMI */
490 /* machine/parity/power fail/"kitchen sink" faults */
491 if (isa_nmi(code) == 0) {
492 #ifdef KDB
493 /*
494 * NMI can be hooked up to a pushbutton
495 * for debugging.
496 */
497 if (kdb_on_nmi) {
498 printf ("NMI ... going to debugger\n");
499 kdb_trap(type, 0, frame);
500 }
501 #endif /* KDB */
502 goto userout;
503 } else if (panic_on_nmi)
504 panic("NMI indicates hardware failure");
505 break;
506 #endif /* POWERFAIL_NMI */
507 #endif /* DEV_ISA */
508
509 case T_OFLOW: /* integer overflow fault */
510 ucode = FPE_INTOVF;
511 i = SIGFPE;
512 break;
513
514 case T_BOUND: /* bounds check fault */
515 ucode = FPE_FLTSUB;
516 i = SIGFPE;
517 break;
518
519 case T_DNA:
520 #ifdef DEV_NPX
521 KASSERT(PCB_USER_FPU(td->td_pcb),
522 ("kernel FPU ctx has leaked"));
523 /* transparent fault (due to context switch "late") */
524 if (npxdna())
525 goto userout;
526 #endif
527 uprintf("pid %d killed due to lack of floating point\n",
528 p->p_pid);
529 i = SIGKILL;
530 ucode = 0;
531 break;
532
533 case T_FPOPFLT: /* FPU operand fetch fault */
534 ucode = ILL_COPROC;
535 i = SIGILL;
536 break;
537
538 case T_XMMFLT: /* SIMD floating-point exception */
539 #if defined(DEV_NPX) && !defined(CPU_DISABLE_SSE) && defined(I686_CPU)
540 ucode = npxtrap_sse();
541 if (ucode == -1)
542 goto userout;
543 #else
544 ucode = 0;
545 #endif
546 i = SIGFPE;
547 break;
548 }
549 } else {
550 /* kernel trap */
551
552 KASSERT(cold || td->td_ucred != NULL,
553 ("kernel trap doesn't have ucred"));
554 switch (type) {
555 case T_PAGEFLT: /* page fault */
556 (void) trap_pfault(frame, FALSE, eva);
557 goto out;
558
559 case T_DNA:
560 #ifdef DEV_NPX
561 KASSERT(!PCB_USER_FPU(td->td_pcb),
562 ("Unregistered use of FPU in kernel"));
563 if (npxdna())
564 goto out;
565 #endif
566 break;
567
568 case T_ARITHTRAP: /* arithmetic trap */
569 case T_XMMFLT: /* SIMD floating-point exception */
570 case T_FPOPFLT: /* FPU operand fetch fault */
571 /*
572 * XXXKIB for now disable any FPU traps in kernel
573 * handler registration seems to be overkill
574 */
575 trap_fatal(frame, 0);
576 goto out;
577
578 /*
579 * The following two traps can happen in
580 * vm86 mode, and, if so, we want to handle
581 * them specially.
582 */
583 case T_PROTFLT: /* general protection fault */
584 case T_STKFLT: /* stack fault */
585 if (frame->tf_eflags & PSL_VM) {
586 i = vm86_emulate((struct vm86frame *)frame);
587 if (i != 0)
588 /*
589 * returns to original process
590 */
591 vm86_trap((struct vm86frame *)frame);
592 goto out;
593 }
594 if (type == T_STKFLT)
595 break;
596
597 /* FALL THROUGH */
598
599 case T_SEGNPFLT: /* segment not present fault */
600 if (curpcb->pcb_flags & PCB_VM86CALL)
601 break;
602
603 /*
604 * Invalid %fs's and %gs's can be created using
605 * procfs or PT_SETREGS or by invalidating the
606 * underlying LDT entry. This causes a fault
607 * in kernel mode when the kernel attempts to
608 * switch contexts. Lose the bad context
609 * (XXX) so that we can continue, and generate
610 * a signal.
611 */
612 if (frame->tf_eip == (int)cpu_switch_load_gs) {
613 curpcb->pcb_gs = 0;
614 #if 0
615 PROC_LOCK(p);
616 kern_psignal(p, SIGBUS);
617 PROC_UNLOCK(p);
618 #endif
619 goto out;
620 }
621
622 if (td->td_intr_nesting_level != 0)
623 break;
624
625 /*
626 * Invalid segment selectors and out of bounds
627 * %eip's and %esp's can be set up in user mode.
628 * This causes a fault in kernel mode when the
629 * kernel tries to return to user mode. We want
630 * to get this fault so that we can fix the
631 * problem here and not have to check all the
632 * selectors and pointers when the user changes
633 * them.
634 */
635 if (frame->tf_eip == (int)doreti_iret) {
636 frame->tf_eip = (int)doreti_iret_fault;
637 goto out;
638 }
639 if (frame->tf_eip == (int)doreti_popl_ds) {
640 frame->tf_eip = (int)doreti_popl_ds_fault;
641 goto out;
642 }
643 if (frame->tf_eip == (int)doreti_popl_es) {
644 frame->tf_eip = (int)doreti_popl_es_fault;
645 goto out;
646 }
647 if (frame->tf_eip == (int)doreti_popl_fs) {
648 frame->tf_eip = (int)doreti_popl_fs_fault;
649 goto out;
650 }
651 if (curpcb->pcb_onfault != NULL) {
652 frame->tf_eip =
653 (int)curpcb->pcb_onfault;
654 goto out;
655 }
656 break;
657
658 case T_TSSFLT:
659 /*
660 * PSL_NT can be set in user mode and isn't cleared
661 * automatically when the kernel is entered. This
662 * causes a TSS fault when the kernel attempts to
663 * `iret' because the TSS link is uninitialized. We
664 * want to get this fault so that we can fix the
665 * problem here and not every time the kernel is
666 * entered.
667 */
668 if (frame->tf_eflags & PSL_NT) {
669 frame->tf_eflags &= ~PSL_NT;
670 goto out;
671 }
672 break;
673
674 case T_TRCTRAP: /* trace trap */
675 if (frame->tf_eip == (int)IDTVEC(lcall_syscall)) {
676 /*
677 * We've just entered system mode via the
678 * syscall lcall. Continue single stepping
679 * silently until the syscall handler has
680 * saved the flags.
681 */
682 goto out;
683 }
684 if (frame->tf_eip == (int)IDTVEC(lcall_syscall) + 1) {
685 /*
686 * The syscall handler has now saved the
687 * flags. Stop single stepping it.
688 */
689 frame->tf_eflags &= ~PSL_T;
690 goto out;
691 }
692 /*
693 * Ignore debug register trace traps due to
694 * accesses in the user's address space, which
695 * can happen under several conditions such as
696 * if a user sets a watchpoint on a buffer and
697 * then passes that buffer to a system call.
698 * We still want to get TRCTRAPS for addresses
699 * in kernel space because that is useful when
700 * debugging the kernel.
701 */
702 if (user_dbreg_trap() &&
703 !(curpcb->pcb_flags & PCB_VM86CALL)) {
704 /*
705 * Reset breakpoint bits because the
706 * processor doesn't
707 */
708 load_dr6(rdr6() & 0xfffffff0);
709 goto out;
710 }
711 /*
712 * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
713 */
714 case T_BPTFLT:
715 /*
716 * If KDB is enabled, let it handle the debugger trap.
717 * Otherwise, debugger traps "can't happen".
718 */
719 #ifdef KDB
720 if (kdb_trap(type, 0, frame))
721 goto out;
722 #endif
723 break;
724
725 #ifdef DEV_ISA
726 case T_NMI:
727 #ifdef POWERFAIL_NMI
728 if (time_second - lastalert > 10) {
729 log(LOG_WARNING, "NMI: power fail\n");
730 sysbeep(880, hz);
731 lastalert = time_second;
732 }
733 goto out;
734 #else /* !POWERFAIL_NMI */
735 /* machine/parity/power fail/"kitchen sink" faults */
736 if (isa_nmi(code) == 0) {
737 #ifdef KDB
738 /*
739 * NMI can be hooked up to a pushbutton
740 * for debugging.
741 */
742 if (kdb_on_nmi) {
743 printf ("NMI ... going to debugger\n");
744 kdb_trap(type, 0, frame);
745 }
746 #endif /* KDB */
747 goto out;
748 } else if (panic_on_nmi == 0)
749 goto out;
750 /* FALLTHROUGH */
751 #endif /* POWERFAIL_NMI */
752 #endif /* DEV_ISA */
753 }
754
755 trap_fatal(frame, eva);
756 goto out;
757 }
758
759 /* Translate fault for emulators (e.g. Linux) */
760 if (*p->p_sysent->sv_transtrap)
761 i = (*p->p_sysent->sv_transtrap)(i, type);
762
763 ksiginfo_init_trap(&ksi);
764 ksi.ksi_signo = i;
765 ksi.ksi_code = ucode;
766 ksi.ksi_addr = (void *)addr;
767 ksi.ksi_trapno = type;
768 if (uprintf_signal) {
769 uprintf("pid %d comm %s: signal %d err %x code %d type %d "
770 "addr 0x%x esp 0x%08x eip 0x%08x "
771 "<%02x %02x %02x %02x %02x %02x %02x %02x>\n",
772 p->p_pid, p->p_comm, i, frame->tf_err, ucode, type, addr,
773 frame->tf_esp, frame->tf_eip,
774 fubyte((void *)(frame->tf_eip + 0)),
775 fubyte((void *)(frame->tf_eip + 1)),
776 fubyte((void *)(frame->tf_eip + 2)),
777 fubyte((void *)(frame->tf_eip + 3)),
778 fubyte((void *)(frame->tf_eip + 4)),
779 fubyte((void *)(frame->tf_eip + 5)),
780 fubyte((void *)(frame->tf_eip + 6)),
781 fubyte((void *)(frame->tf_eip + 7)));
782 }
783 KASSERT((read_eflags() & PSL_I) != 0, ("interrupts disabled"));
784 trapsignal(td, &ksi);
785
786 #ifdef DEBUG
787 if (type <= MAX_TRAP_MSG) {
788 uprintf("fatal process exception: %s",
789 trap_msg[type]);
790 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
791 uprintf(", fault VA = 0x%lx", (u_long)eva);
792 uprintf("\n");
793 }
794 #endif
795
796 user:
797 userret(td, frame);
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;
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. If either
884 * p or p->p_vmspace is NULL, then the fault is fatal.
885 */
886 if (p == NULL || (vm = p->p_vmspace) == NULL)
887 goto nogo;
888
889 map = &vm->vm_map;
890
891 /*
892 * When accessing a user-space address, kernel must be
893 * ready to accept the page fault, and provide a
894 * handling routine. Since accessing the address
895 * without the handler is a bug, do not try to handle
896 * it normally, and panic immediately.
897 */
898 if (!usermode && (td->td_intr_nesting_level != 0 ||
899 curpcb->pcb_onfault == NULL)) {
900 trap_fatal(frame, eva);
901 return (-1);
902 }
903 }
904
905 /*
906 * PGEX_I is defined only if the execute disable bit capability is
907 * supported and enabled.
908 */
909 if (frame->tf_err & PGEX_W)
910 ftype = VM_PROT_WRITE;
911 #ifdef PAE
912 else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
913 ftype = VM_PROT_EXECUTE;
914 #endif
915 else
916 ftype = VM_PROT_READ;
917
918 if (map != kernel_map) {
919 /*
920 * Keep swapout from messing with us during this
921 * critical time.
922 */
923 PROC_LOCK(p);
924 ++p->p_lock;
925 PROC_UNLOCK(p);
926
927 /* Fault in the user page: */
928 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
929
930 PROC_LOCK(p);
931 --p->p_lock;
932 PROC_UNLOCK(p);
933 } else {
934 /*
935 * Don't have to worry about process locking or stacks in the
936 * kernel.
937 */
938 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
939 }
940 if (rv == KERN_SUCCESS) {
941 #ifdef HWPMC_HOOKS
942 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
943 PMC_SOFT_CALL_TF( , , page_fault, all, frame);
944 if (ftype == VM_PROT_READ)
945 PMC_SOFT_CALL_TF( , , page_fault, read,
946 frame);
947 else
948 PMC_SOFT_CALL_TF( , , page_fault, write,
949 frame);
950 }
951 #endif
952 return (0);
953 }
954 nogo:
955 if (!usermode) {
956 if (td->td_intr_nesting_level == 0 &&
957 curpcb->pcb_onfault != NULL) {
958 frame->tf_eip = (int)curpcb->pcb_onfault;
959 return (0);
960 }
961 trap_fatal(frame, eva);
962 return (-1);
963 }
964 return ((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
965 }
966
967 static void
968 trap_fatal(frame, eva)
969 struct trapframe *frame;
970 vm_offset_t eva;
971 {
972 int code, ss, esp;
973 u_int type;
974 struct soft_segment_descriptor softseg;
975 char *msg;
976
977 code = frame->tf_err;
978 type = frame->tf_trapno;
979 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
980
981 if (type <= MAX_TRAP_MSG)
982 msg = trap_msg[type];
983 else
984 msg = "UNKNOWN";
985 printf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
986 frame->tf_eflags & PSL_VM ? "vm86" :
987 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
988 #ifdef SMP
989 /* two separate prints in case of a trap on an unmapped page */
990 printf("cpuid = %d; ", PCPU_GET(cpuid));
991 printf("apic id = %02x\n", PCPU_GET(apic_id));
992 #endif
993 if (type == T_PAGEFLT) {
994 printf("fault virtual address = 0x%x\n", eva);
995 printf("fault code = %s %s, %s\n",
996 code & PGEX_U ? "user" : "supervisor",
997 code & PGEX_W ? "write" : "read",
998 code & PGEX_P ? "protection violation" : "page not present");
999 }
1000 printf("instruction pointer = 0x%x:0x%x\n",
1001 frame->tf_cs & 0xffff, frame->tf_eip);
1002 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
1003 ss = frame->tf_ss & 0xffff;
1004 esp = frame->tf_esp;
1005 } else {
1006 ss = GSEL(GDATA_SEL, SEL_KPL);
1007 esp = (int)&frame->tf_esp;
1008 }
1009 printf("stack pointer = 0x%x:0x%x\n", ss, esp);
1010 printf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
1011 printf("code segment = base 0x%x, limit 0x%x, type 0x%x\n",
1012 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
1013 printf(" = DPL %d, pres %d, def32 %d, gran %d\n",
1014 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
1015 softseg.ssd_gran);
1016 printf("processor eflags = ");
1017 if (frame->tf_eflags & PSL_T)
1018 printf("trace trap, ");
1019 if (frame->tf_eflags & PSL_I)
1020 printf("interrupt enabled, ");
1021 if (frame->tf_eflags & PSL_NT)
1022 printf("nested task, ");
1023 if (frame->tf_eflags & PSL_RF)
1024 printf("resume, ");
1025 if (frame->tf_eflags & PSL_VM)
1026 printf("vm86, ");
1027 printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
1028 printf("current process = ");
1029 if (curproc) {
1030 printf("%lu (%s)\n", (u_long)curproc->p_pid, curthread->td_name);
1031 } else {
1032 printf("Idle\n");
1033 }
1034
1035 #ifdef KDB
1036 if (debugger_on_panic || kdb_active) {
1037 frame->tf_err = eva; /* smuggle fault address to ddb */
1038 if (kdb_trap(type, 0, frame)) {
1039 frame->tf_err = code; /* restore error code */
1040 return;
1041 }
1042 frame->tf_err = code; /* restore error code */
1043 }
1044 #endif
1045 printf("trap number = %d\n", type);
1046 if (type <= MAX_TRAP_MSG)
1047 panic("%s", trap_msg[type]);
1048 else
1049 panic("unknown/reserved trap");
1050 }
1051
1052 /*
1053 * Double fault handler. Called when a fault occurs while writing
1054 * a frame for a trap/exception onto the stack. This usually occurs
1055 * when the stack overflows (such is the case with infinite recursion,
1056 * for example).
1057 *
1058 * XXX Note that the current PTD gets replaced by IdlePTD when the
1059 * task switch occurs. This means that the stack that was active at
1060 * the time of the double fault is not available at <kstack> unless
1061 * the machine was idle when the double fault occurred. The downside
1062 * of this is that "trace <ebp>" in ddb won't work.
1063 */
1064 void
1065 dblfault_handler()
1066 {
1067 #ifdef KDTRACE_HOOKS
1068 if (dtrace_doubletrap_func != NULL)
1069 (*dtrace_doubletrap_func)();
1070 #endif
1071 printf("\nFatal double fault:\n");
1072 printf("eip = 0x%x\n", PCPU_GET(common_tss.tss_eip));
1073 printf("esp = 0x%x\n", PCPU_GET(common_tss.tss_esp));
1074 printf("ebp = 0x%x\n", PCPU_GET(common_tss.tss_ebp));
1075 #ifdef SMP
1076 /* two separate prints in case of a trap on an unmapped page */
1077 printf("cpuid = %d; ", PCPU_GET(cpuid));
1078 printf("apic id = %02x\n", PCPU_GET(apic_id));
1079 #endif
1080 panic("double fault");
1081 }
1082
1083 int
1084 cpu_fetch_syscall_args(struct thread *td, struct syscall_args *sa)
1085 {
1086 struct proc *p;
1087 struct trapframe *frame;
1088 caddr_t params;
1089 int error;
1090
1091 p = td->td_proc;
1092 frame = td->td_frame;
1093
1094 params = (caddr_t)frame->tf_esp + sizeof(int);
1095 sa->code = frame->tf_eax;
1096
1097 /*
1098 * Need to check if this is a 32 bit or 64 bit syscall.
1099 */
1100 if (sa->code == SYS_syscall) {
1101 /*
1102 * Code is first argument, followed by actual args.
1103 */
1104 sa->code = fuword(params);
1105 params += sizeof(int);
1106 } else if (sa->code == SYS___syscall) {
1107 /*
1108 * Like syscall, but code is a quad, so as to maintain
1109 * quad alignment for the rest of the arguments.
1110 */
1111 sa->code = fuword(params);
1112 params += sizeof(quad_t);
1113 }
1114
1115 if (p->p_sysent->sv_mask)
1116 sa->code &= p->p_sysent->sv_mask;
1117 if (sa->code >= p->p_sysent->sv_size)
1118 sa->callp = &p->p_sysent->sv_table[0];
1119 else
1120 sa->callp = &p->p_sysent->sv_table[sa->code];
1121 sa->narg = sa->callp->sy_narg;
1122
1123 if (params != NULL && sa->narg != 0)
1124 error = copyin(params, (caddr_t)sa->args,
1125 (u_int)(sa->narg * sizeof(int)));
1126 else
1127 error = 0;
1128
1129 if (error == 0) {
1130 td->td_retval[0] = 0;
1131 td->td_retval[1] = frame->tf_edx;
1132 }
1133
1134 return (error);
1135 }
1136
1137 #include "../../kern/subr_syscall.c"
1138
1139 /*
1140 * syscall - system call request C handler. A system call is
1141 * essentially treated as a trap by reusing the frame layout.
1142 */
1143 void
1144 syscall(struct trapframe *frame)
1145 {
1146 struct thread *td;
1147 struct syscall_args sa;
1148 register_t orig_tf_eflags;
1149 int error;
1150 ksiginfo_t ksi;
1151
1152 #ifdef DIAGNOSTIC
1153 if (ISPL(frame->tf_cs) != SEL_UPL) {
1154 panic("syscall");
1155 /* NOT REACHED */
1156 }
1157 #endif
1158 orig_tf_eflags = frame->tf_eflags;
1159
1160 td = curthread;
1161 td->td_frame = frame;
1162
1163 error = syscallenter(td, &sa);
1164
1165 /*
1166 * Traced syscall.
1167 */
1168 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1169 frame->tf_eflags &= ~PSL_T;
1170 ksiginfo_init_trap(&ksi);
1171 ksi.ksi_signo = SIGTRAP;
1172 ksi.ksi_code = TRAP_TRACE;
1173 ksi.ksi_addr = (void *)frame->tf_eip;
1174 trapsignal(td, &ksi);
1175 }
1176
1177 KASSERT(PCB_USER_FPU(td->td_pcb),
1178 ("System call %s returning with kernel FPU ctx leaked",
1179 syscallname(td->td_proc, sa.code)));
1180 KASSERT(td->td_pcb->pcb_save == &td->td_pcb->pcb_user_save,
1181 ("System call %s returning with mangled pcb_save",
1182 syscallname(td->td_proc, sa.code)));
1183
1184 syscallret(td, error, &sa);
1185 }
Cache object: 5b20a79401ad8b467086fda385ba050c
|