FreeBSD/Linux Kernel Cross Reference
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
80 #include <vm/vm.h>
81 #include <vm/vm_param.h>
82 #include <vm/pmap.h>
83 #include <vm/vm_kern.h>
84 #include <vm/vm_map.h>
85 #include <vm/vm_page.h>
86 #include <vm/vm_extern.h>
87
88 #include <machine/cpu.h>
89 #include <machine/intr_machdep.h>
90 #include <machine/mca.h>
91 #include <machine/md_var.h>
92 #include <machine/pcb.h>
93 #ifdef SMP
94 #include <machine/smp.h>
95 #endif
96 #include <machine/tss.h>
97
98 #ifdef KDTRACE_HOOKS
99 #include <sys/dtrace_bsd.h>
100
101 /*
102 * This is a hook which is initialised by the dtrace module
103 * to handle traps which might occur during DTrace probe
104 * execution.
105 */
106 dtrace_trap_func_t dtrace_trap_func;
107
108 dtrace_doubletrap_func_t dtrace_doubletrap_func;
109
110 /*
111 * This is a hook which is initialised by the systrace module
112 * when it is loaded. This keeps the DTrace syscall provider
113 * implementation opaque.
114 */
115 systrace_probe_func_t systrace_probe_func;
116
117 /*
118 * These hooks are necessary for the pid, usdt and fasttrap providers.
119 */
120 dtrace_fasttrap_probe_ptr_t dtrace_fasttrap_probe_ptr;
121 dtrace_pid_probe_ptr_t dtrace_pid_probe_ptr;
122 dtrace_return_probe_ptr_t dtrace_return_probe_ptr;
123 #endif
124
125 extern void trap(struct trapframe *frame);
126 extern void syscall(struct trapframe *frame);
127 void dblfault_handler(struct trapframe *frame);
128
129 static int trap_pfault(struct trapframe *, int);
130 static void trap_fatal(struct trapframe *, vm_offset_t);
131
132 #define MAX_TRAP_MSG 33
133 static char *trap_msg[] = {
134 "", /* 0 unused */
135 "privileged instruction fault", /* 1 T_PRIVINFLT */
136 "", /* 2 unused */
137 "breakpoint instruction fault", /* 3 T_BPTFLT */
138 "", /* 4 unused */
139 "", /* 5 unused */
140 "arithmetic trap", /* 6 T_ARITHTRAP */
141 "", /* 7 unused */
142 "", /* 8 unused */
143 "general protection fault", /* 9 T_PROTFLT */
144 "trace trap", /* 10 T_TRCTRAP */
145 "", /* 11 unused */
146 "page fault", /* 12 T_PAGEFLT */
147 "", /* 13 unused */
148 "alignment fault", /* 14 T_ALIGNFLT */
149 "", /* 15 unused */
150 "", /* 16 unused */
151 "", /* 17 unused */
152 "integer divide fault", /* 18 T_DIVIDE */
153 "non-maskable interrupt trap", /* 19 T_NMI */
154 "overflow trap", /* 20 T_OFLOW */
155 "FPU bounds check fault", /* 21 T_BOUND */
156 "FPU device not available", /* 22 T_DNA */
157 "double fault", /* 23 T_DOUBLEFLT */
158 "FPU operand fetch fault", /* 24 T_FPOPFLT */
159 "invalid TSS fault", /* 25 T_TSSFLT */
160 "segment not present fault", /* 26 T_SEGNPFLT */
161 "stack fault", /* 27 T_STKFLT */
162 "machine check trap", /* 28 T_MCHK */
163 "SIMD floating-point exception", /* 29 T_XMMFLT */
164 "reserved (unknown) fault", /* 30 T_RESERVED */
165 "", /* 31 unused (reserved) */
166 "DTrace pid return trap", /* 32 T_DTRACE_RET */
167 "DTrace fasttrap probe trap", /* 33 T_DTRACE_PROBE */
168 };
169
170 #ifdef KDB
171 static int kdb_on_nmi = 1;
172 SYSCTL_INT(_machdep, OID_AUTO, kdb_on_nmi, CTLFLAG_RW,
173 &kdb_on_nmi, 0, "Go to KDB on NMI");
174 TUNABLE_INT("machdep.kdb_on_nmi", &kdb_on_nmi);
175 #endif
176 static int panic_on_nmi = 1;
177 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
178 &panic_on_nmi, 0, "Panic on NMI");
179 TUNABLE_INT("machdep.panic_on_nmi", &panic_on_nmi);
180 static int prot_fault_translation = 0;
181 SYSCTL_INT(_machdep, OID_AUTO, prot_fault_translation, CTLFLAG_RW,
182 &prot_fault_translation, 0, "Select signal to deliver on protection fault");
183
184 /*
185 * Exception, fault, and trap interface to the FreeBSD kernel.
186 * This common code is called from assembly language IDT gate entry
187 * routines that prepare a suitable stack frame, and restore this
188 * frame after the exception has been processed.
189 */
190
191 void
192 trap(struct trapframe *frame)
193 {
194 struct thread *td = curthread;
195 struct proc *p = td->td_proc;
196 int i = 0, ucode = 0, code;
197 u_int type;
198 register_t addr = 0;
199 ksiginfo_t ksi;
200
201 PCPU_INC(cnt.v_trap);
202 type = frame->tf_trapno;
203
204 #ifdef SMP
205 /* Handler for NMI IPIs used for stopping CPUs. */
206 if (type == T_NMI) {
207 if (ipi_nmi_handler() == 0)
208 goto out;
209 }
210 #endif /* SMP */
211
212 #ifdef KDB
213 if (kdb_active) {
214 kdb_reenter();
215 goto out;
216 }
217 #endif
218
219 if (type == T_RESERVED) {
220 trap_fatal(frame, 0);
221 goto out;
222 }
223
224 #ifdef HWPMC_HOOKS
225 /*
226 * CPU PMCs interrupt using an NMI. If the PMC module is
227 * active, pass the 'rip' value to the PMC module's interrupt
228 * handler. A return value of '1' from the handler means that
229 * the NMI was handled by it and we can return immediately.
230 */
231 if (type == T_NMI && pmc_intr &&
232 (*pmc_intr)(PCPU_GET(cpuid), frame))
233 goto out;
234 #endif
235
236 if (type == T_MCHK) {
237 mca_intr();
238 goto out;
239 }
240
241 #ifdef KDTRACE_HOOKS
242 /*
243 * A trap can occur while DTrace executes a probe. Before
244 * executing the probe, DTrace blocks re-scheduling and sets
245 * a flag in it's per-cpu flags to indicate that it doesn't
246 * want to fault. On returning from the probe, the no-fault
247 * flag is cleared and finally re-scheduling is enabled.
248 *
249 * If the DTrace kernel module has registered a trap handler,
250 * call it and if it returns non-zero, assume that it has
251 * handled the trap and modified the trap frame so that this
252 * function can return normally.
253 */
254 if (type == T_DTRACE_PROBE || type == T_DTRACE_RET ||
255 type == T_BPTFLT) {
256 struct reg regs;
257
258 fill_frame_regs(frame, ®s);
259 if (type == T_DTRACE_PROBE &&
260 dtrace_fasttrap_probe_ptr != NULL &&
261 dtrace_fasttrap_probe_ptr(®s) == 0)
262 goto out;
263 else if (type == T_BPTFLT &&
264 dtrace_pid_probe_ptr != NULL &&
265 dtrace_pid_probe_ptr(®s) == 0)
266 goto out;
267 else if (type == T_DTRACE_RET &&
268 dtrace_return_probe_ptr != NULL &&
269 dtrace_return_probe_ptr(®s) == 0)
270 goto out;
271 }
272 if (dtrace_trap_func != NULL && (*dtrace_trap_func)(frame, type))
273 goto out;
274 #endif
275
276 if ((frame->tf_rflags & PSL_I) == 0) {
277 /*
278 * Buggy application or kernel code has disabled
279 * interrupts and then trapped. Enabling interrupts
280 * now is wrong, but it is better than running with
281 * interrupts disabled until they are accidentally
282 * enabled later.
283 */
284 if (ISPL(frame->tf_cs) == SEL_UPL)
285 uprintf(
286 "pid %ld (%s): trap %d with interrupts disabled\n",
287 (long)curproc->p_pid, curthread->td_name, type);
288 else if (type != T_NMI && type != T_BPTFLT &&
289 type != T_TRCTRAP) {
290 /*
291 * XXX not quite right, since this may be for a
292 * multiple fault in user mode.
293 */
294 printf("kernel trap %d with interrupts disabled\n",
295 type);
296
297 /*
298 * We shouldn't enable interrupts while holding a
299 * spin lock.
300 */
301 if (td->td_md.md_spinlock_count == 0)
302 enable_intr();
303 }
304 }
305
306 code = frame->tf_err;
307
308 if (ISPL(frame->tf_cs) == SEL_UPL) {
309 /* user trap */
310
311 td->td_pticks = 0;
312 td->td_frame = frame;
313 addr = frame->tf_rip;
314 if (td->td_ucred != p->p_ucred)
315 cred_update_thread(td);
316
317 switch (type) {
318 case T_PRIVINFLT: /* privileged instruction fault */
319 i = SIGILL;
320 ucode = ILL_PRVOPC;
321 break;
322
323 case T_BPTFLT: /* bpt instruction fault */
324 case T_TRCTRAP: /* trace trap */
325 enable_intr();
326 frame->tf_rflags &= ~PSL_T;
327 i = SIGTRAP;
328 ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
329 break;
330
331 case T_ARITHTRAP: /* arithmetic trap */
332 ucode = fputrap();
333 if (ucode == -1)
334 goto userout;
335 i = SIGFPE;
336 break;
337
338 case T_PROTFLT: /* general protection fault */
339 i = SIGBUS;
340 ucode = BUS_OBJERR;
341 break;
342 case T_STKFLT: /* stack fault */
343 case T_SEGNPFLT: /* segment not present fault */
344 i = SIGBUS;
345 ucode = BUS_ADRERR;
346 break;
347 case T_TSSFLT: /* invalid TSS fault */
348 i = SIGBUS;
349 ucode = BUS_OBJERR;
350 break;
351 case T_DOUBLEFLT: /* double fault */
352 default:
353 i = SIGBUS;
354 ucode = BUS_OBJERR;
355 break;
356
357 case T_PAGEFLT: /* page fault */
358 addr = frame->tf_addr;
359 i = trap_pfault(frame, TRUE);
360 if (i == -1)
361 goto userout;
362 if (i == 0)
363 goto user;
364
365 if (i == SIGSEGV)
366 ucode = SEGV_MAPERR;
367 else {
368 if (prot_fault_translation == 0) {
369 /*
370 * Autodetect.
371 * This check also covers the images
372 * without the ABI-tag ELF note.
373 */
374 if (SV_CURPROC_ABI() == SV_ABI_FREEBSD
375 && p->p_osrel >= P_OSREL_SIGSEGV) {
376 i = SIGSEGV;
377 ucode = SEGV_ACCERR;
378 } else {
379 i = SIGBUS;
380 ucode = BUS_PAGE_FAULT;
381 }
382 } else if (prot_fault_translation == 1) {
383 /*
384 * Always compat mode.
385 */
386 i = SIGBUS;
387 ucode = BUS_PAGE_FAULT;
388 } else {
389 /*
390 * Always SIGSEGV mode.
391 */
392 i = SIGSEGV;
393 ucode = SEGV_ACCERR;
394 }
395 }
396 break;
397
398 case T_DIVIDE: /* integer divide fault */
399 ucode = FPE_INTDIV;
400 i = SIGFPE;
401 break;
402
403 #ifdef DEV_ISA
404 case T_NMI:
405 /* machine/parity/power fail/"kitchen sink" faults */
406 if (isa_nmi(code) == 0) {
407 #ifdef KDB
408 /*
409 * NMI can be hooked up to a pushbutton
410 * for debugging.
411 */
412 if (kdb_on_nmi) {
413 printf ("NMI ... going to debugger\n");
414 kdb_trap(type, 0, frame);
415 }
416 #endif /* KDB */
417 goto userout;
418 } else if (panic_on_nmi)
419 panic("NMI indicates hardware failure");
420 break;
421 #endif /* DEV_ISA */
422
423 case T_OFLOW: /* integer overflow fault */
424 ucode = FPE_INTOVF;
425 i = SIGFPE;
426 break;
427
428 case T_BOUND: /* bounds check fault */
429 ucode = FPE_FLTSUB;
430 i = SIGFPE;
431 break;
432
433 case T_DNA:
434 /* transparent fault (due to context switch "late") */
435 KASSERT(PCB_USER_FPU(td->td_pcb),
436 ("kernel FPU ctx has leaked"));
437 fpudna();
438 goto userout;
439
440 case T_FPOPFLT: /* FPU operand fetch fault */
441 ucode = ILL_COPROC;
442 i = SIGILL;
443 break;
444
445 case T_XMMFLT: /* SIMD floating-point exception */
446 ucode = 0; /* XXX */
447 i = SIGFPE;
448 break;
449 }
450 } else {
451 /* kernel trap */
452
453 KASSERT(cold || td->td_ucred != NULL,
454 ("kernel trap doesn't have ucred"));
455 switch (type) {
456 case T_PAGEFLT: /* page fault */
457 (void) trap_pfault(frame, FALSE);
458 goto out;
459
460 case T_DNA:
461 KASSERT(!PCB_USER_FPU(td->td_pcb),
462 ("Unregistered use of FPU in kernel"));
463 fpudna();
464 goto out;
465
466 case T_ARITHTRAP: /* arithmetic trap */
467 case T_XMMFLT: /* SIMD floating-point exception */
468 case T_FPOPFLT: /* FPU operand fetch fault */
469 /*
470 * XXXKIB for now disable any FPU traps in kernel
471 * handler registration seems to be overkill
472 */
473 trap_fatal(frame, 0);
474 goto out;
475
476 case T_STKFLT: /* stack fault */
477 break;
478
479 case T_PROTFLT: /* general protection fault */
480 case T_SEGNPFLT: /* segment not present fault */
481 if (td->td_intr_nesting_level != 0)
482 break;
483
484 /*
485 * Invalid segment selectors and out of bounds
486 * %rip's and %rsp's can be set up in user mode.
487 * This causes a fault in kernel mode when the
488 * kernel tries to return to user mode. We want
489 * to get this fault so that we can fix the
490 * problem here and not have to check all the
491 * selectors and pointers when the user changes
492 * them.
493 */
494 if (frame->tf_rip == (long)doreti_iret) {
495 frame->tf_rip = (long)doreti_iret_fault;
496 goto out;
497 }
498 if (frame->tf_rip == (long)ld_ds) {
499 frame->tf_rip = (long)ds_load_fault;
500 goto out;
501 }
502 if (frame->tf_rip == (long)ld_es) {
503 frame->tf_rip = (long)es_load_fault;
504 goto out;
505 }
506 if (frame->tf_rip == (long)ld_fs) {
507 frame->tf_rip = (long)fs_load_fault;
508 goto out;
509 }
510 if (frame->tf_rip == (long)ld_gs) {
511 frame->tf_rip = (long)gs_load_fault;
512 goto out;
513 }
514 if (frame->tf_rip == (long)ld_gsbase) {
515 frame->tf_rip = (long)gsbase_load_fault;
516 goto out;
517 }
518 if (frame->tf_rip == (long)ld_fsbase) {
519 frame->tf_rip = (long)fsbase_load_fault;
520 goto out;
521 }
522 if (PCPU_GET(curpcb)->pcb_onfault != NULL) {
523 frame->tf_rip =
524 (long)PCPU_GET(curpcb)->pcb_onfault;
525 goto out;
526 }
527 break;
528
529 case T_TSSFLT:
530 /*
531 * PSL_NT can be set in user mode and isn't cleared
532 * automatically when the kernel is entered. This
533 * causes a TSS fault when the kernel attempts to
534 * `iret' because the TSS link is uninitialized. We
535 * want to get this fault so that we can fix the
536 * problem here and not every time the kernel is
537 * entered.
538 */
539 if (frame->tf_rflags & PSL_NT) {
540 frame->tf_rflags &= ~PSL_NT;
541 goto out;
542 }
543 break;
544
545 case T_TRCTRAP: /* trace trap */
546 /*
547 * Ignore debug register trace traps due to
548 * accesses in the user's address space, which
549 * can happen under several conditions such as
550 * if a user sets a watchpoint on a buffer and
551 * then passes that buffer to a system call.
552 * We still want to get TRCTRAPS for addresses
553 * in kernel space because that is useful when
554 * debugging the kernel.
555 */
556 if (user_dbreg_trap()) {
557 /*
558 * Reset breakpoint bits because the
559 * processor doesn't
560 */
561 /* XXX check upper bits here */
562 load_dr6(rdr6() & 0xfffffff0);
563 goto out;
564 }
565 /*
566 * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
567 */
568 case T_BPTFLT:
569 /*
570 * If KDB is enabled, let it handle the debugger trap.
571 * Otherwise, debugger traps "can't happen".
572 */
573 #ifdef KDB
574 if (kdb_trap(type, 0, frame))
575 goto out;
576 #endif
577 break;
578
579 #ifdef DEV_ISA
580 case T_NMI:
581 /* machine/parity/power fail/"kitchen sink" faults */
582 if (isa_nmi(code) == 0) {
583 #ifdef KDB
584 /*
585 * NMI can be hooked up to a pushbutton
586 * for debugging.
587 */
588 if (kdb_on_nmi) {
589 printf ("NMI ... going to debugger\n");
590 kdb_trap(type, 0, frame);
591 }
592 #endif /* KDB */
593 goto out;
594 } else if (panic_on_nmi == 0)
595 goto out;
596 /* FALLTHROUGH */
597 #endif /* DEV_ISA */
598 }
599
600 trap_fatal(frame, 0);
601 goto out;
602 }
603
604 /* Translate fault for emulators (e.g. Linux) */
605 if (*p->p_sysent->sv_transtrap)
606 i = (*p->p_sysent->sv_transtrap)(i, type);
607
608 ksiginfo_init_trap(&ksi);
609 ksi.ksi_signo = i;
610 ksi.ksi_code = ucode;
611 ksi.ksi_trapno = type;
612 ksi.ksi_addr = (void *)addr;
613 KASSERT((read_rflags() & PSL_I) != 0, ("interrupts disabled"));
614 trapsignal(td, &ksi);
615
616 user:
617 userret(td, frame);
618 mtx_assert(&Giant, MA_NOTOWNED);
619 KASSERT(PCB_USER_FPU(td->td_pcb),
620 ("Return from trap with kernel FPU ctx leaked"));
621 userout:
622 out:
623 return;
624 }
625
626 static int
627 trap_pfault(frame, usermode)
628 struct trapframe *frame;
629 int usermode;
630 {
631 vm_offset_t va;
632 struct vmspace *vm = NULL;
633 vm_map_t map;
634 int rv = 0;
635 vm_prot_t ftype;
636 struct thread *td = curthread;
637 struct proc *p = td->td_proc;
638 vm_offset_t eva = frame->tf_addr;
639
640 if (__predict_false((td->td_pflags & TDP_NOFAULTING) != 0)) {
641 /*
642 * Due to both processor errata and lazy TLB invalidation when
643 * access restrictions are removed from virtual pages, memory
644 * accesses that are allowed by the physical mapping layer may
645 * nonetheless cause one spurious page fault per virtual page.
646 * When the thread is executing a "no faulting" section that
647 * is bracketed by vm_fault_{disable,enable}_pagefaults(),
648 * every page fault is treated as a spurious page fault,
649 * unless it accesses the same virtual address as the most
650 * recent page fault within the same "no faulting" section.
651 */
652 if (td->td_md.md_spurflt_addr != eva ||
653 (td->td_pflags & TDP_RESETSPUR) != 0) {
654 /*
655 * Do nothing to the TLB. A stale TLB entry is
656 * flushed automatically by a page fault.
657 */
658 td->td_md.md_spurflt_addr = eva;
659 td->td_pflags &= ~TDP_RESETSPUR;
660 return (0);
661 }
662 } else {
663 /*
664 * If we get a page fault while in a critical section, then
665 * it is most likely a fatal kernel page fault. The kernel
666 * is already going to panic trying to get a sleep lock to
667 * do the VM lookup, so just consider it a fatal trap so the
668 * kernel can print out a useful trap message and even get
669 * to the debugger.
670 *
671 * If we get a page fault while holding a non-sleepable
672 * lock, then it is most likely a fatal kernel page fault.
673 * If WITNESS is enabled, then it's going to whine about
674 * bogus LORs with various VM locks, so just skip to the
675 * fatal trap handling directly.
676 */
677 if (td->td_critnest != 0 ||
678 WITNESS_CHECK(WARN_SLEEPOK | WARN_GIANTOK, NULL,
679 "Kernel page fault") != 0) {
680 trap_fatal(frame, eva);
681 return (-1);
682 }
683 }
684 va = trunc_page(eva);
685 if (va >= VM_MIN_KERNEL_ADDRESS) {
686 /*
687 * Don't allow user-mode faults in kernel address space.
688 */
689 if (usermode)
690 goto nogo;
691
692 map = kernel_map;
693 } else {
694 /*
695 * This is a fault on non-kernel virtual memory.
696 * vm is initialized above to NULL. If curproc is NULL
697 * or curproc->p_vmspace is NULL the fault is fatal.
698 */
699 if (p != NULL)
700 vm = p->p_vmspace;
701
702 if (vm == NULL)
703 goto nogo;
704
705 map = &vm->vm_map;
706
707 /*
708 * When accessing a usermode address, kernel must be
709 * ready to accept the page fault, and provide a
710 * handling routine. Since accessing the address
711 * without the handler is a bug, do not try to handle
712 * it normally, and panic immediately.
713 */
714 if (!usermode && (td->td_intr_nesting_level != 0 ||
715 PCPU_GET(curpcb)->pcb_onfault == NULL)) {
716 trap_fatal(frame, eva);
717 return (-1);
718 }
719 }
720
721 /*
722 * PGEX_I is defined only if the execute disable bit capability is
723 * supported and enabled.
724 */
725 if (frame->tf_err & PGEX_W)
726 ftype = VM_PROT_WRITE;
727 else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
728 ftype = VM_PROT_EXECUTE;
729 else
730 ftype = VM_PROT_READ;
731
732 if (map != kernel_map) {
733 /*
734 * Keep swapout from messing with us during this
735 * critical time.
736 */
737 PROC_LOCK(p);
738 ++p->p_lock;
739 PROC_UNLOCK(p);
740
741 /* Fault in the user page: */
742 rv = vm_fault(map, va, ftype,
743 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
744 : VM_FAULT_NORMAL);
745
746 PROC_LOCK(p);
747 --p->p_lock;
748 PROC_UNLOCK(p);
749 } else {
750 /*
751 * Don't have to worry about process locking or stacks in the
752 * kernel.
753 */
754 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
755 }
756 if (rv == KERN_SUCCESS)
757 return (0);
758 nogo:
759 if (!usermode) {
760 if (td->td_intr_nesting_level == 0 &&
761 PCPU_GET(curpcb)->pcb_onfault != NULL) {
762 frame->tf_rip = (long)PCPU_GET(curpcb)->pcb_onfault;
763 return (0);
764 }
765 trap_fatal(frame, eva);
766 return (-1);
767 }
768
769 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
770 }
771
772 static void
773 trap_fatal(frame, eva)
774 struct trapframe *frame;
775 vm_offset_t eva;
776 {
777 int code, ss;
778 u_int type;
779 long esp;
780 struct soft_segment_descriptor softseg;
781 char *msg;
782
783 code = frame->tf_err;
784 type = frame->tf_trapno;
785 sdtossd(&gdt[NGDT * PCPU_GET(cpuid) + IDXSEL(frame->tf_cs & 0xffff)],
786 &softseg);
787
788 if (type <= MAX_TRAP_MSG)
789 msg = trap_msg[type];
790 else
791 msg = "UNKNOWN";
792 printf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
793 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
794 #ifdef SMP
795 /* two separate prints in case of a trap on an unmapped page */
796 printf("cpuid = %d; ", PCPU_GET(cpuid));
797 printf("apic id = %02x\n", PCPU_GET(apic_id));
798 #endif
799 if (type == T_PAGEFLT) {
800 printf("fault virtual address = 0x%lx\n", eva);
801 printf("fault code = %s %s %s, %s\n",
802 code & PGEX_U ? "user" : "supervisor",
803 code & PGEX_W ? "write" : "read",
804 code & PGEX_I ? "instruction" : "data",
805 code & PGEX_P ? "protection violation" : "page not present");
806 }
807 printf("instruction pointer = 0x%lx:0x%lx\n",
808 frame->tf_cs & 0xffff, frame->tf_rip);
809 if (ISPL(frame->tf_cs) == SEL_UPL) {
810 ss = frame->tf_ss & 0xffff;
811 esp = frame->tf_rsp;
812 } else {
813 ss = GSEL(GDATA_SEL, SEL_KPL);
814 esp = (long)&frame->tf_rsp;
815 }
816 printf("stack pointer = 0x%x:0x%lx\n", ss, esp);
817 printf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp);
818 printf("code segment = base 0x%lx, limit 0x%lx, type 0x%x\n",
819 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
820 printf(" = DPL %d, pres %d, long %d, def32 %d, gran %d\n",
821 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32,
822 softseg.ssd_gran);
823 printf("processor eflags = ");
824 if (frame->tf_rflags & PSL_T)
825 printf("trace trap, ");
826 if (frame->tf_rflags & PSL_I)
827 printf("interrupt enabled, ");
828 if (frame->tf_rflags & PSL_NT)
829 printf("nested task, ");
830 if (frame->tf_rflags & PSL_RF)
831 printf("resume, ");
832 printf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12);
833 printf("current process = ");
834 if (curproc) {
835 printf("%lu (%s)\n",
836 (u_long)curproc->p_pid, curthread->td_name ?
837 curthread->td_name : "");
838 } else {
839 printf("Idle\n");
840 }
841
842 #ifdef KDB
843 if (debugger_on_panic || kdb_active)
844 if (kdb_trap(type, 0, frame))
845 return;
846 #endif
847 printf("trap number = %d\n", type);
848 if (type <= MAX_TRAP_MSG)
849 panic("%s", trap_msg[type]);
850 else
851 panic("unknown/reserved trap");
852 }
853
854 /*
855 * Double fault handler. Called when a fault occurs while writing
856 * a frame for a trap/exception onto the stack. This usually occurs
857 * when the stack overflows (such is the case with infinite recursion,
858 * for example).
859 */
860 void
861 dblfault_handler(struct trapframe *frame)
862 {
863 #ifdef KDTRACE_HOOKS
864 if (dtrace_doubletrap_func != NULL)
865 (*dtrace_doubletrap_func)();
866 #endif
867 printf("\nFatal double fault\n");
868 printf("rip = 0x%lx\n", frame->tf_rip);
869 printf("rsp = 0x%lx\n", frame->tf_rsp);
870 printf("rbp = 0x%lx\n", frame->tf_rbp);
871 #ifdef SMP
872 /* two separate prints in case of a trap on an unmapped page */
873 printf("cpuid = %d; ", PCPU_GET(cpuid));
874 printf("apic id = %02x\n", PCPU_GET(apic_id));
875 #endif
876 panic("double fault");
877 }
878
879 int
880 cpu_fetch_syscall_args(struct thread *td, struct syscall_args *sa)
881 {
882 struct proc *p;
883 struct trapframe *frame;
884 register_t *argp;
885 caddr_t params;
886 int reg, regcnt, error;
887
888 p = td->td_proc;
889 frame = td->td_frame;
890 reg = 0;
891 regcnt = 6;
892
893 params = (caddr_t)frame->tf_rsp + sizeof(register_t);
894 sa->code = frame->tf_rax;
895
896 if (sa->code == SYS_syscall || sa->code == SYS___syscall) {
897 sa->code = frame->tf_rdi;
898 reg++;
899 regcnt--;
900 }
901 if (p->p_sysent->sv_mask)
902 sa->code &= p->p_sysent->sv_mask;
903
904 if (sa->code >= p->p_sysent->sv_size)
905 sa->callp = &p->p_sysent->sv_table[0];
906 else
907 sa->callp = &p->p_sysent->sv_table[sa->code];
908
909 sa->narg = sa->callp->sy_narg;
910 KASSERT(sa->narg <= sizeof(sa->args) / sizeof(sa->args[0]),
911 ("Too many syscall arguments!"));
912 error = 0;
913 argp = &frame->tf_rdi;
914 argp += reg;
915 bcopy(argp, sa->args, sizeof(sa->args[0]) * regcnt);
916 if (sa->narg > regcnt) {
917 KASSERT(params != NULL, ("copyin args with no params!"));
918 error = copyin(params, &sa->args[regcnt],
919 (sa->narg - regcnt) * sizeof(sa->args[0]));
920 }
921
922 if (error == 0) {
923 td->td_retval[0] = 0;
924 td->td_retval[1] = frame->tf_rdx;
925 }
926
927 return (error);
928 }
929
930 #include "../../kern/subr_syscall.c"
931
932 /*
933 * syscall - system call request C handler
934 *
935 * A system call is essentially treated as a trap.
936 */
937 void
938 amd64_syscall(struct thread *td, int traced)
939 {
940 struct syscall_args sa;
941 int error;
942 ksiginfo_t ksi;
943
944 #ifdef DIAGNOSTIC
945 if (ISPL(td->td_frame->tf_cs) != SEL_UPL) {
946 panic("syscall");
947 /* NOT REACHED */
948 }
949 #endif
950 error = syscallenter(td, &sa);
951
952 /*
953 * Traced syscall.
954 */
955 if (__predict_false(traced)) {
956 td->td_frame->tf_rflags &= ~PSL_T;
957 ksiginfo_init_trap(&ksi);
958 ksi.ksi_signo = SIGTRAP;
959 ksi.ksi_code = TRAP_TRACE;
960 ksi.ksi_addr = (void *)td->td_frame->tf_rip;
961 trapsignal(td, &ksi);
962 }
963
964 KASSERT(PCB_USER_FPU(td->td_pcb),
965 ("System call %s returing with kernel FPU ctx leaked",
966 syscallname(td->td_proc, sa.code)));
967 KASSERT(td->td_pcb->pcb_save == get_pcb_user_save_td(td),
968 ("System call %s returning with mangled pcb_save",
969 syscallname(td->td_proc, sa.code)));
970
971 syscallret(td, error, &sa);
972
973 /*
974 * If the user-supplied value of %rip is not a canonical
975 * address, then some CPUs will trigger a ring 0 #GP during
976 * the sysret instruction. However, the fault handler would
977 * execute with the user's %gs and %rsp in ring 0 which would
978 * not be safe. Instead, preemptively kill the thread with a
979 * SIGBUS.
980 */
981 if (td->td_frame->tf_rip >= VM_MAXUSER_ADDRESS) {
982 ksiginfo_init_trap(&ksi);
983 ksi.ksi_signo = SIGBUS;
984 ksi.ksi_code = BUS_OBJERR;
985 ksi.ksi_trapno = T_PROTFLT;
986 ksi.ksi_addr = (void *)td->td_frame->tf_rip;
987 trapsignal(td, &ksi);
988 }
989 }
Cache object: 7d12723301b886261422b1fb176bf2fc
|