2 * Copyright (C) 1994, David Greenman
3 * Copyright (c) 1990, 1993
4 * The Regents of the University of California. All rights reserved.
6 * This code is derived from software contributed to Berkeley by
7 * the University of Utah, and William Jolitz.
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
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.
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
37 * from: @(#)trap.c 7.4 (Berkeley) 5/13/91
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
44 * 386 Trap and System call handling
47 #include "opt_clock.h"
49 #include "opt_hwpmc_hooks.h"
52 #include "opt_kdtrace.h"
56 #include <sys/param.h>
58 #include <sys/systm.h>
60 #include <sys/pioctl.h>
61 #include <sys/ptrace.h>
63 #include <sys/kernel.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>
73 #include <sys/vmmeter.h>
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);
80 #include <security/audit/audit.h>
83 #include <vm/vm_param.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>
90 #include <machine/cpu.h>
91 #include <machine/intr_machdep.h>
93 #include <machine/md_var.h>
94 #include <machine/pcb.h>
96 #include <machine/smp.h>
98 #include <machine/tss.h>
99 #include <machine/vm86.h>
102 #include <sys/syslog.h>
103 #include <machine/clock.h>
107 #include <sys/dtrace_bsd.h>
110 * This is a hook which is initialised by the dtrace module
111 * to handle traps which might occur during DTrace probe
114 dtrace_trap_func_t dtrace_trap_func;
116 dtrace_doubletrap_func_t dtrace_doubletrap_func;
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.
123 systrace_probe_func_t systrace_probe_func;
126 * These hooks are necessary for the pid and usdt providers.
128 dtrace_pid_probe_ptr_t dtrace_pid_probe_ptr;
129 dtrace_return_probe_ptr_t dtrace_return_probe_ptr;
132 extern void trap(struct trapframe *frame);
133 extern void syscall(struct trapframe *frame);
135 static int trap_pfault(struct trapframe *, int, vm_offset_t);
136 static void trap_fatal(struct trapframe *, vm_offset_t);
137 void dblfault_handler(void);
139 extern inthand_t IDTVEC(lcall_syscall);
141 #define MAX_TRAP_MSG 32
142 static char *trap_msg[] = {
144 "privileged instruction fault", /* 1 T_PRIVINFLT */
146 "breakpoint instruction fault", /* 3 T_BPTFLT */
149 "arithmetic trap", /* 6 T_ARITHTRAP */
152 "general protection fault", /* 9 T_PROTFLT */
153 "trace trap", /* 10 T_TRCTRAP */
155 "page fault", /* 12 T_PAGEFLT */
157 "alignment fault", /* 14 T_ALIGNFLT */
161 "integer divide fault", /* 18 T_DIVIDE */
162 "non-maskable interrupt trap", /* 19 T_NMI */
163 "overflow trap", /* 20 T_OFLOW */
164 "FPU bounds check fault", /* 21 T_BOUND */
165 "FPU device not available", /* 22 T_DNA */
166 "double fault", /* 23 T_DOUBLEFLT */
167 "FPU operand fetch fault", /* 24 T_FPOPFLT */
168 "invalid TSS fault", /* 25 T_TSSFLT */
169 "segment not present fault", /* 26 T_SEGNPFLT */
170 "stack fault", /* 27 T_STKFLT */
171 "machine check trap", /* 28 T_MCHK */
172 "SIMD floating-point exception", /* 29 T_XMMFLT */
173 "reserved (unknown) fault", /* 30 T_RESERVED */
174 "", /* 31 unused (reserved) */
175 "DTrace pid return trap", /* 32 T_DTRACE_RET */
178 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
179 extern int has_f00f_bug;
183 static int kdb_on_nmi = 1;
184 SYSCTL_INT(_machdep, OID_AUTO, kdb_on_nmi, CTLFLAG_RW,
185 &kdb_on_nmi, 0, "Go to KDB on NMI");
186 TUNABLE_INT("machdep.kdb_on_nmi", &kdb_on_nmi);
188 static int panic_on_nmi = 1;
189 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
190 &panic_on_nmi, 0, "Panic on NMI");
191 TUNABLE_INT("machdep.panic_on_nmi", &panic_on_nmi);
192 static int prot_fault_translation = 0;
193 SYSCTL_INT(_machdep, OID_AUTO, prot_fault_translation, CTLFLAG_RW,
194 &prot_fault_translation, 0, "Select signal to deliver on protection fault");
195 static int uprintf_signal;
196 SYSCTL_INT(_machdep, OID_AUTO, uprintf_signal, CTLFLAG_RW,
198 "Print debugging information on trap signal to ctty");
201 * Exception, fault, and trap interface to the FreeBSD kernel.
202 * This common code is called from assembly language IDT gate entry
203 * routines that prepare a suitable stack frame, and restore this
204 * frame after the exception has been processed.
208 trap(struct trapframe *frame)
210 struct thread *td = curthread;
211 struct proc *p = td->td_proc;
212 int i = 0, ucode = 0, code;
218 static int lastalert = 0;
221 PCPU_INC(cnt.v_trap);
222 type = frame->tf_trapno;
225 /* Handler for NMI IPIs used for stopping CPUs. */
227 if (ipi_nmi_handler() == 0)
239 if (type == T_RESERVED) {
240 trap_fatal(frame, 0);
246 * CPU PMCs interrupt using an NMI so we check for that first.
247 * If the HWPMC module is active, 'pmc_hook' will point to
248 * the function to be called. A return value of '1' from the
249 * hook means that the NMI was handled by it and that we can
250 * return immediately.
252 if (type == T_NMI && pmc_intr &&
253 (*pmc_intr)(PCPU_GET(cpuid), frame))
257 if (type == T_MCHK) {
264 * A trap can occur while DTrace executes a probe. Before
265 * executing the probe, DTrace blocks re-scheduling and sets
266 * a flag in it's per-cpu flags to indicate that it doesn't
267 * want to fault. On returning from the probe, the no-fault
268 * flag is cleared and finally re-scheduling is enabled.
270 * If the DTrace kernel module has registered a trap handler,
271 * call it and if it returns non-zero, assume that it has
272 * handled the trap and modified the trap frame so that this
273 * function can return normally.
275 if (type == T_DTRACE_RET || type == T_BPTFLT) {
278 fill_frame_regs(frame, ®s);
279 if (type == T_BPTFLT &&
280 dtrace_pid_probe_ptr != NULL &&
281 dtrace_pid_probe_ptr(®s) == 0)
283 if (type == T_DTRACE_RET &&
284 dtrace_return_probe_ptr != NULL &&
285 dtrace_return_probe_ptr(®s) == 0)
288 if ((type == T_PROTFLT || type == T_PAGEFLT) &&
289 dtrace_trap_func != NULL && (*dtrace_trap_func)(frame, type))
293 if ((frame->tf_eflags & PSL_I) == 0) {
295 * Buggy application or kernel code has disabled
296 * interrupts and then trapped. Enabling interrupts
297 * now is wrong, but it is better than running with
298 * interrupts disabled until they are accidentally
301 if (ISPL(frame->tf_cs) == SEL_UPL || (frame->tf_eflags & PSL_VM))
303 "pid %ld (%s): trap %d with interrupts disabled\n",
304 (long)curproc->p_pid, curthread->td_name, type);
305 else if (type != T_NMI && type != T_BPTFLT &&
307 frame->tf_eip != (int)cpu_switch_load_gs) {
309 * XXX not quite right, since this may be for a
310 * multiple fault in user mode.
312 printf("kernel trap %d with interrupts disabled\n",
315 * Page faults need interrupts disabled until later,
316 * and we shouldn't enable interrupts while holding
319 if (type != T_PAGEFLT &&
320 td->td_md.md_spinlock_count == 0)
325 code = frame->tf_err;
326 if (type == T_PAGEFLT) {
328 * For some Cyrix CPUs, %cr2 is clobbered by
329 * interrupts. This problem is worked around by using
330 * an interrupt gate for the pagefault handler. We
331 * are finally ready to read %cr2 and conditionally
332 * reenable interrupts. If we hold a spin lock, then
333 * we must not reenable interrupts. This might be a
334 * spurious page fault.
337 if (td->td_md.md_spinlock_count == 0)
341 if ((ISPL(frame->tf_cs) == SEL_UPL) ||
342 ((frame->tf_eflags & PSL_VM) &&
343 !(curpcb->pcb_flags & PCB_VM86CALL))) {
347 td->td_frame = frame;
348 addr = frame->tf_eip;
349 if (td->td_ucred != p->p_ucred)
350 cred_update_thread(td);
353 case T_PRIVINFLT: /* privileged instruction fault */
358 case T_BPTFLT: /* bpt instruction fault */
359 case T_TRCTRAP: /* trace trap */
361 frame->tf_eflags &= ~PSL_T;
363 ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
366 case T_ARITHTRAP: /* arithmetic trap */
368 ucode = npxtrap_x87();
378 * The following two traps can happen in
379 * vm86 mode, and, if so, we want to handle
382 case T_PROTFLT: /* general protection fault */
383 case T_STKFLT: /* stack fault */
384 if (frame->tf_eflags & PSL_VM) {
385 i = vm86_emulate((struct vm86frame *)frame);
391 ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR;
393 case T_SEGNPFLT: /* segment not present fault */
397 case T_TSSFLT: /* invalid TSS fault */
405 case T_DOUBLEFLT: /* double fault */
411 case T_PAGEFLT: /* page fault */
413 i = trap_pfault(frame, TRUE, eva);
414 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
417 * The f00f hack workaround has triggered, so
418 * treat the fault as an illegal instruction
419 * (T_PRIVINFLT) instead of a page fault.
421 type = frame->tf_trapno = T_PRIVINFLT;
423 /* Proceed as in that case. */
437 if (prot_fault_translation == 0) {
440 * This check also covers the images
441 * without the ABI-tag ELF note.
443 if (SV_CURPROC_ABI() == SV_ABI_FREEBSD
444 && p->p_osrel >= P_OSREL_SIGSEGV) {
449 ucode = BUS_PAGE_FAULT;
451 } else if (prot_fault_translation == 1) {
453 * Always compat mode.
456 ucode = BUS_PAGE_FAULT;
459 * Always SIGSEGV mode.
468 case T_DIVIDE: /* integer divide fault */
477 # define TIMER_FREQ 1193182
479 if (time_second - lastalert > 10) {
480 log(LOG_WARNING, "NMI: power fail\n");
482 lastalert = time_second;
485 #else /* !POWERFAIL_NMI */
486 /* machine/parity/power fail/"kitchen sink" faults */
487 if (isa_nmi(code) == 0) {
490 * NMI can be hooked up to a pushbutton
494 printf ("NMI ... going to debugger\n");
495 kdb_trap(type, 0, frame);
499 } else if (panic_on_nmi)
500 panic("NMI indicates hardware failure");
502 #endif /* POWERFAIL_NMI */
505 case T_OFLOW: /* integer overflow fault */
510 case T_BOUND: /* bounds check fault */
517 KASSERT(PCB_USER_FPU(td->td_pcb),
518 ("kernel FPU ctx has leaked"));
519 /* transparent fault (due to context switch "late") */
523 uprintf("pid %d killed due to lack of floating point\n",
529 case T_FPOPFLT: /* FPU operand fetch fault */
534 case T_XMMFLT: /* SIMD floating-point exception */
535 #if defined(DEV_NPX) && !defined(CPU_DISABLE_SSE) && defined(I686_CPU)
536 ucode = npxtrap_sse();
548 KASSERT(cold || td->td_ucred != NULL,
549 ("kernel trap doesn't have ucred"));
551 case T_PAGEFLT: /* page fault */
552 (void) trap_pfault(frame, FALSE, eva);
557 KASSERT(!PCB_USER_FPU(td->td_pcb),
558 ("Unregistered use of FPU in kernel"));
564 case T_ARITHTRAP: /* arithmetic trap */
565 case T_XMMFLT: /* SIMD floating-point exception */
566 case T_FPOPFLT: /* FPU operand fetch fault */
568 * XXXKIB for now disable any FPU traps in kernel
569 * handler registration seems to be overkill
571 trap_fatal(frame, 0);
575 * The following two traps can happen in
576 * vm86 mode, and, if so, we want to handle
579 case T_PROTFLT: /* general protection fault */
580 case T_STKFLT: /* stack fault */
581 if (frame->tf_eflags & PSL_VM) {
582 i = vm86_emulate((struct vm86frame *)frame);
585 * returns to original process
587 vm86_trap((struct vm86frame *)frame);
590 if (type == T_STKFLT)
595 case T_SEGNPFLT: /* segment not present fault */
596 if (curpcb->pcb_flags & PCB_VM86CALL)
600 * Invalid %fs's and %gs's can be created using
601 * procfs or PT_SETREGS or by invalidating the
602 * underlying LDT entry. This causes a fault
603 * in kernel mode when the kernel attempts to
604 * switch contexts. Lose the bad context
605 * (XXX) so that we can continue, and generate
608 if (frame->tf_eip == (int)cpu_switch_load_gs) {
612 kern_psignal(p, SIGBUS);
618 if (td->td_intr_nesting_level != 0)
622 * Invalid segment selectors and out of bounds
623 * %eip's and %esp's can be set up in user mode.
624 * This causes a fault in kernel mode when the
625 * kernel tries to return to user mode. We want
626 * to get this fault so that we can fix the
627 * problem here and not have to check all the
628 * selectors and pointers when the user changes
631 if (frame->tf_eip == (int)doreti_iret) {
632 frame->tf_eip = (int)doreti_iret_fault;
635 if (frame->tf_eip == (int)doreti_popl_ds) {
636 frame->tf_eip = (int)doreti_popl_ds_fault;
639 if (frame->tf_eip == (int)doreti_popl_es) {
640 frame->tf_eip = (int)doreti_popl_es_fault;
643 if (frame->tf_eip == (int)doreti_popl_fs) {
644 frame->tf_eip = (int)doreti_popl_fs_fault;
647 if (curpcb->pcb_onfault != NULL) {
649 (int)curpcb->pcb_onfault;
656 * PSL_NT can be set in user mode and isn't cleared
657 * automatically when the kernel is entered. This
658 * causes a TSS fault when the kernel attempts to
659 * `iret' because the TSS link is uninitialized. We
660 * want to get this fault so that we can fix the
661 * problem here and not every time the kernel is
664 if (frame->tf_eflags & PSL_NT) {
665 frame->tf_eflags &= ~PSL_NT;
670 case T_TRCTRAP: /* trace trap */
671 if (frame->tf_eip == (int)IDTVEC(lcall_syscall)) {
673 * We've just entered system mode via the
674 * syscall lcall. Continue single stepping
675 * silently until the syscall handler has
680 if (frame->tf_eip == (int)IDTVEC(lcall_syscall) + 1) {
682 * The syscall handler has now saved the
683 * flags. Stop single stepping it.
685 frame->tf_eflags &= ~PSL_T;
689 * Ignore debug register trace traps due to
690 * accesses in the user's address space, which
691 * can happen under several conditions such as
692 * if a user sets a watchpoint on a buffer and
693 * then passes that buffer to a system call.
694 * We still want to get TRCTRAPS for addresses
695 * in kernel space because that is useful when
696 * debugging the kernel.
698 if (user_dbreg_trap() &&
699 !(curpcb->pcb_flags & PCB_VM86CALL)) {
701 * Reset breakpoint bits because the
704 load_dr6(rdr6() & 0xfffffff0);
708 * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
712 * If KDB is enabled, let it handle the debugger trap.
713 * Otherwise, debugger traps "can't happen".
716 if (kdb_trap(type, 0, frame))
724 if (time_second - lastalert > 10) {
725 log(LOG_WARNING, "NMI: power fail\n");
727 lastalert = time_second;
730 #else /* !POWERFAIL_NMI */
731 /* machine/parity/power fail/"kitchen sink" faults */
732 if (isa_nmi(code) == 0) {
735 * NMI can be hooked up to a pushbutton
739 printf ("NMI ... going to debugger\n");
740 kdb_trap(type, 0, frame);
744 } else if (panic_on_nmi == 0)
747 #endif /* POWERFAIL_NMI */
751 trap_fatal(frame, eva);
755 /* Translate fault for emulators (e.g. Linux) */
756 if (*p->p_sysent->sv_transtrap)
757 i = (*p->p_sysent->sv_transtrap)(i, type);
759 ksiginfo_init_trap(&ksi);
761 ksi.ksi_code = ucode;
762 ksi.ksi_addr = (void *)addr;
763 ksi.ksi_trapno = type;
764 if (uprintf_signal) {
765 uprintf("pid %d comm %s: signal %d err %x code %d type %d "
766 "addr 0x%x esp 0x%08x eip 0x%08x "
767 "<%02x %02x %02x %02x %02x %02x %02x %02x>\n",
768 p->p_pid, p->p_comm, i, frame->tf_err, ucode, type, addr,
769 frame->tf_esp, frame->tf_eip,
770 fubyte((void *)(frame->tf_eip + 0)),
771 fubyte((void *)(frame->tf_eip + 1)),
772 fubyte((void *)(frame->tf_eip + 2)),
773 fubyte((void *)(frame->tf_eip + 3)),
774 fubyte((void *)(frame->tf_eip + 4)),
775 fubyte((void *)(frame->tf_eip + 5)),
776 fubyte((void *)(frame->tf_eip + 6)),
777 fubyte((void *)(frame->tf_eip + 7)));
779 KASSERT((read_eflags() & PSL_I) != 0, ("interrupts disabled"));
780 trapsignal(td, &ksi);
783 if (type <= MAX_TRAP_MSG) {
784 uprintf("fatal process exception: %s",
786 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
787 uprintf(", fault VA = 0x%lx", (u_long)eva);
794 KASSERT(PCB_USER_FPU(td->td_pcb),
795 ("Return from trap with kernel FPU ctx leaked"));
802 trap_pfault(frame, usermode, eva)
803 struct trapframe *frame;
812 struct thread *td = curthread;
813 struct proc *p = td->td_proc;
815 if (__predict_false((td->td_pflags & TDP_NOFAULTING) != 0)) {
817 * Due to both processor errata and lazy TLB invalidation when
818 * access restrictions are removed from virtual pages, memory
819 * accesses that are allowed by the physical mapping layer may
820 * nonetheless cause one spurious page fault per virtual page.
821 * When the thread is executing a "no faulting" section that
822 * is bracketed by vm_fault_{disable,enable}_pagefaults(),
823 * every page fault is treated as a spurious page fault,
824 * unless it accesses the same virtual address as the most
825 * recent page fault within the same "no faulting" section.
827 if (td->td_md.md_spurflt_addr != eva ||
828 (td->td_pflags & TDP_RESETSPUR) != 0) {
830 * Do nothing to the TLB. A stale TLB entry is
831 * flushed automatically by a page fault.
833 td->td_md.md_spurflt_addr = eva;
834 td->td_pflags &= ~TDP_RESETSPUR;
839 * If we get a page fault while in a critical section, then
840 * it is most likely a fatal kernel page fault. The kernel
841 * is already going to panic trying to get a sleep lock to
842 * do the VM lookup, so just consider it a fatal trap so the
843 * kernel can print out a useful trap message and even get
846 * If we get a page fault while holding a non-sleepable
847 * lock, then it is most likely a fatal kernel page fault.
848 * If WITNESS is enabled, then it's going to whine about
849 * bogus LORs with various VM locks, so just skip to the
850 * fatal trap handling directly.
852 if (td->td_critnest != 0 ||
853 WITNESS_CHECK(WARN_SLEEPOK | WARN_GIANTOK, NULL,
854 "Kernel page fault") != 0) {
855 trap_fatal(frame, eva);
859 va = trunc_page(eva);
860 if (va >= KERNBASE) {
862 * Don't allow user-mode faults in kernel address space.
863 * An exception: if the faulting address is the invalid
864 * instruction entry in the IDT, then the Intel Pentium
865 * F00F bug workaround was triggered, and we need to
866 * treat it is as an illegal instruction, and not a page
869 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
870 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug)
879 * This is a fault on non-kernel virtual memory. If either
880 * p or p->p_vmspace is NULL, then the fault is fatal.
882 if (p == NULL || (vm = p->p_vmspace) == NULL)
888 * When accessing a user-space address, kernel must be
889 * ready to accept the page fault, and provide a
890 * handling routine. Since accessing the address
891 * without the handler is a bug, do not try to handle
892 * it normally, and panic immediately.
894 if (!usermode && (td->td_intr_nesting_level != 0 ||
895 curpcb->pcb_onfault == NULL)) {
896 trap_fatal(frame, eva);
902 * PGEX_I is defined only if the execute disable bit capability is
903 * supported and enabled.
905 if (frame->tf_err & PGEX_W)
906 ftype = VM_PROT_WRITE;
908 else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
909 ftype = VM_PROT_EXECUTE;
912 ftype = VM_PROT_READ;
914 if (map != kernel_map) {
916 * Keep swapout from messing with us during this
923 /* Fault in the user page: */
924 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
931 * Don't have to worry about process locking or stacks in the
934 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
936 if (rv == KERN_SUCCESS) {
938 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
939 PMC_SOFT_CALL_TF( , , page_fault, all, frame);
940 if (ftype == VM_PROT_READ)
941 PMC_SOFT_CALL_TF( , , page_fault, read,
944 PMC_SOFT_CALL_TF( , , page_fault, write,
952 if (td->td_intr_nesting_level == 0 &&
953 curpcb->pcb_onfault != NULL) {
954 frame->tf_eip = (int)curpcb->pcb_onfault;
957 trap_fatal(frame, eva);
960 return ((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
964 trap_fatal(frame, eva)
965 struct trapframe *frame;
970 struct soft_segment_descriptor softseg;
973 code = frame->tf_err;
974 type = frame->tf_trapno;
975 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
977 if (type <= MAX_TRAP_MSG)
978 msg = trap_msg[type];
981 printf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
982 frame->tf_eflags & PSL_VM ? "vm86" :
983 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
985 /* two separate prints in case of a trap on an unmapped page */
986 printf("cpuid = %d; ", PCPU_GET(cpuid));
987 printf("apic id = %02x\n", PCPU_GET(apic_id));
989 if (type == T_PAGEFLT) {
990 printf("fault virtual address = 0x%x\n", eva);
991 printf("fault code = %s %s, %s\n",
992 code & PGEX_U ? "user" : "supervisor",
993 code & PGEX_W ? "write" : "read",
994 code & PGEX_P ? "protection violation" : "page not present");
996 printf("instruction pointer = 0x%x:0x%x\n",
997 frame->tf_cs & 0xffff, frame->tf_eip);
998 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
999 ss = frame->tf_ss & 0xffff;
1000 esp = frame->tf_esp;
1002 ss = GSEL(GDATA_SEL, SEL_KPL);
1003 esp = (int)&frame->tf_esp;
1005 printf("stack pointer = 0x%x:0x%x\n", ss, esp);
1006 printf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
1007 printf("code segment = base 0x%x, limit 0x%x, type 0x%x\n",
1008 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
1009 printf(" = DPL %d, pres %d, def32 %d, gran %d\n",
1010 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
1012 printf("processor eflags = ");
1013 if (frame->tf_eflags & PSL_T)
1014 printf("trace trap, ");
1015 if (frame->tf_eflags & PSL_I)
1016 printf("interrupt enabled, ");
1017 if (frame->tf_eflags & PSL_NT)
1018 printf("nested task, ");
1019 if (frame->tf_eflags & PSL_RF)
1021 if (frame->tf_eflags & PSL_VM)
1023 printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
1024 printf("current process = ");
1026 printf("%lu (%s)\n", (u_long)curproc->p_pid, curthread->td_name);
1032 if (debugger_on_panic || kdb_active) {
1033 frame->tf_err = eva; /* smuggle fault address to ddb */
1034 if (kdb_trap(type, 0, frame)) {
1035 frame->tf_err = code; /* restore error code */
1038 frame->tf_err = code; /* restore error code */
1041 printf("trap number = %d\n", type);
1042 if (type <= MAX_TRAP_MSG)
1043 panic("%s", trap_msg[type]);
1045 panic("unknown/reserved trap");
1049 * Double fault handler. Called when a fault occurs while writing
1050 * a frame for a trap/exception onto the stack. This usually occurs
1051 * when the stack overflows (such is the case with infinite recursion,
1054 * XXX Note that the current PTD gets replaced by IdlePTD when the
1055 * task switch occurs. This means that the stack that was active at
1056 * the time of the double fault is not available at <kstack> unless
1057 * the machine was idle when the double fault occurred. The downside
1058 * of this is that "trace <ebp>" in ddb won't work.
1063 #ifdef KDTRACE_HOOKS
1064 if (dtrace_doubletrap_func != NULL)
1065 (*dtrace_doubletrap_func)();
1067 printf("\nFatal double fault:\n");
1068 printf("eip = 0x%x\n", PCPU_GET(common_tss.tss_eip));
1069 printf("esp = 0x%x\n", PCPU_GET(common_tss.tss_esp));
1070 printf("ebp = 0x%x\n", PCPU_GET(common_tss.tss_ebp));
1072 /* two separate prints in case of a trap on an unmapped page */
1073 printf("cpuid = %d; ", PCPU_GET(cpuid));
1074 printf("apic id = %02x\n", PCPU_GET(apic_id));
1076 panic("double fault");
1080 cpu_fetch_syscall_args(struct thread *td, struct syscall_args *sa)
1083 struct trapframe *frame;
1088 frame = td->td_frame;
1090 params = (caddr_t)frame->tf_esp + sizeof(int);
1091 sa->code = frame->tf_eax;
1094 * Need to check if this is a 32 bit or 64 bit syscall.
1096 if (sa->code == SYS_syscall) {
1098 * Code is first argument, followed by actual args.
1100 sa->code = fuword(params);
1101 params += sizeof(int);
1102 } else if (sa->code == SYS___syscall) {
1104 * Like syscall, but code is a quad, so as to maintain
1105 * quad alignment for the rest of the arguments.
1107 sa->code = fuword(params);
1108 params += sizeof(quad_t);
1111 if (p->p_sysent->sv_mask)
1112 sa->code &= p->p_sysent->sv_mask;
1113 if (sa->code >= p->p_sysent->sv_size)
1114 sa->callp = &p->p_sysent->sv_table[0];
1116 sa->callp = &p->p_sysent->sv_table[sa->code];
1117 sa->narg = sa->callp->sy_narg;
1119 if (params != NULL && sa->narg != 0)
1120 error = copyin(params, (caddr_t)sa->args,
1121 (u_int)(sa->narg * sizeof(int)));
1126 td->td_retval[0] = 0;
1127 td->td_retval[1] = frame->tf_edx;
1133 #include "../../kern/subr_syscall.c"
1136 * syscall - system call request C handler. A system call is
1137 * essentially treated as a trap by reusing the frame layout.
1140 syscall(struct trapframe *frame)
1143 struct syscall_args sa;
1144 register_t orig_tf_eflags;
1149 if (ISPL(frame->tf_cs) != SEL_UPL) {
1154 orig_tf_eflags = frame->tf_eflags;
1157 td->td_frame = frame;
1159 error = syscallenter(td, &sa);
1164 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1165 frame->tf_eflags &= ~PSL_T;
1166 ksiginfo_init_trap(&ksi);
1167 ksi.ksi_signo = SIGTRAP;
1168 ksi.ksi_code = TRAP_TRACE;
1169 ksi.ksi_addr = (void *)frame->tf_eip;
1170 trapsignal(td, &ksi);
1173 KASSERT(PCB_USER_FPU(td->td_pcb),
1174 ("System call %s returning with kernel FPU ctx leaked",
1175 syscallname(td->td_proc, sa.code)));
1176 KASSERT(td->td_pcb->pcb_save == &td->td_pcb->pcb_user_save,
1177 ("System call %s returning with mangled pcb_save",
1178 syscallname(td->td_proc, sa.code)));
1180 syscallret(td, error, &sa);