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
38 * $FreeBSD: src/sys/i386/i386/trap.c,v 1.147.2.11 2003/02/27 19:09:59 luoqi Exp $
39 * $DragonFly: src/sys/i386/i386/Attic/trap.c,v 1.13 2003/07/01 04:37:44 dillon Exp $
43 * 386 Trap and System call handling
48 #include "opt_ktrace.h"
49 #include "opt_clock.h"
52 #include <sys/param.h>
53 #include <sys/systm.h>
55 #include <sys/pioctl.h>
56 #include <sys/kernel.h>
57 #include <sys/resourcevar.h>
58 #include <sys/signalvar.h>
59 #include <sys/syscall.h>
60 #include <sys/sysctl.h>
61 #include <sys/sysent.h>
63 #include <sys/vmmeter.h>
65 #include <sys/ktrace.h>
69 #include <vm/vm_param.h>
72 #include <vm/vm_kern.h>
73 #include <vm/vm_map.h>
74 #include <vm/vm_page.h>
75 #include <vm/vm_extern.h>
77 #include <machine/cpu.h>
78 #include <machine/ipl.h>
79 #include <machine/md_var.h>
80 #include <machine/pcb.h>
82 #include <machine/smp.h>
84 #include <machine/tss.h>
85 #include <machine/globaldata.h>
87 #include <i386/isa/intr_machdep.h>
90 #include <sys/syslog.h>
91 #include <machine/clock.h>
94 #include <machine/vm86.h>
97 #include <sys/thread2.h>
102 int (*pmath_emulate) __P((struct trapframe *));
104 extern void trap __P((struct trapframe frame));
105 extern int trapwrite __P((unsigned addr));
106 extern void syscall2 __P((struct trapframe frame));
108 static int trap_pfault __P((struct trapframe *, int, vm_offset_t));
109 static void trap_fatal __P((struct trapframe *, vm_offset_t));
110 void dblfault_handler __P((void));
112 extern inthand_t IDTVEC(syscall);
114 #define MAX_TRAP_MSG 28
115 static char *trap_msg[] = {
117 "privileged instruction fault", /* 1 T_PRIVINFLT */
119 "breakpoint instruction fault", /* 3 T_BPTFLT */
122 "arithmetic trap", /* 6 T_ARITHTRAP */
123 "system forced exception", /* 7 T_ASTFLT */
125 "general protection fault", /* 9 T_PROTFLT */
126 "trace trap", /* 10 T_TRCTRAP */
128 "page fault", /* 12 T_PAGEFLT */
130 "alignment fault", /* 14 T_ALIGNFLT */
134 "integer divide fault", /* 18 T_DIVIDE */
135 "non-maskable interrupt trap", /* 19 T_NMI */
136 "overflow trap", /* 20 T_OFLOW */
137 "FPU bounds check fault", /* 21 T_BOUND */
138 "FPU device not available", /* 22 T_DNA */
139 "double fault", /* 23 T_DOUBLEFLT */
140 "FPU operand fetch fault", /* 24 T_FPOPFLT */
141 "invalid TSS fault", /* 25 T_TSSFLT */
142 "segment not present fault", /* 26 T_SEGNPFLT */
143 "stack fault", /* 27 T_STKFLT */
144 "machine check trap", /* 28 T_MCHK */
147 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
148 extern int has_f00f_bug;
152 static int ddb_on_nmi = 1;
153 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
154 &ddb_on_nmi, 0, "Go to DDB on NMI");
156 static int panic_on_nmi = 1;
157 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
158 &panic_on_nmi, 0, "Panic on NMI");
161 * USER->KERNEL transition. Do not transition us out of userland from the
162 * point of view of the userland scheduler unless we actually have to
165 * usertdsw is called from within a critical section.
168 usertdsw(struct thread *ntd)
170 struct thread *td = curthread;
172 td->td_switch = cpu_heavy_switch;
173 lwkt_setpri_self(TDPRI_KERN_USER);
176 * This is where we might want to catch the P_CURPROC designation
177 * and fix it for *any* switchout rather then just an mi_switch()
178 * switchout (move from mi_switch()?) YYY
180 if (p->p_flag & P_CURPROC) {
188 * Note that userenter() may be re-entered several times due to AST
194 struct thread *td = curthread;
196 KKASSERT(td->td_switch == cpu_heavy_switch ||
197 td->td_switch == usertdsw);
198 td->td_switch = usertdsw;
202 userret(struct proc *p, struct trapframe *frame,
203 u_quad_t oticks, int have_mplock)
206 struct thread *td = curthread;
209 * Post any pending signals
212 while ((sig = CURSIG(p)) != 0) {
213 if (have_mplock == 0) {
223 * Set our priority properly and restore our switch function
225 if (td->td_switch == cpu_heavy_switch) {
226 switch(p->p_rtprio.type) {
228 lwkt_setpri_self(TDPRI_USER_IDLE);
230 case RTP_PRIO_REALTIME:
232 lwkt_setpri_self(TDPRI_USER_REAL);
235 lwkt_setpri_self(TDPRI_USER_NORM);
239 KKASSERT(td->td_switch == usertdsw);
240 td->td_switch = cpu_heavy_switch;
245 * If a reschedule has been requested we call chooseproc() to locate
246 * the next runnable process. When we wakeup from that we check
247 * for pending signals again.
249 if (resched_wanted()) {
251 if (have_mplock == 0) {
255 while ((sig = CURSIG(p)) != 0)
260 * Charge system time if profiling.
262 if (p->p_flag & P_PROFIL) {
263 if (have_mplock == 0) {
267 addupc_task(p, frame->tf_eip,
268 (u_int)(curthread->td_sticks - oticks) * psratio);
272 * In order to return to userland we need to be the designated
273 * current (user) process on this cpu. We have to wait for
274 * the userland scheduler to schedule as P_CURPROC.
277 while ((p->p_flag & P_CURPROC) == 0) {
278 p->p_stats->p_ru.ru_nivcsw++;
279 lwkt_deschedule_self();
283 KKASSERT(mycpu->gd_uprocscheduled == 1);
288 #ifdef DEVICE_POLLING
289 extern u_int32_t poll_in_trap;
290 extern int ether_poll __P((int count));
291 #endif /* DEVICE_POLLING */
294 * Exception, fault, and trap interface to the FreeBSD kernel.
295 * This common code is called from assembly language IDT gate entry
296 * routines that prepare a suitable stack frame, and restore this
297 * frame after the exception has been processed.
302 struct trapframe frame;
304 struct proc *p = curproc;
306 int i = 0, ucode = 0, type, code;
311 eva = (frame.tf_trapno == T_PAGEFLT ? rcr2() : 0);
312 trap_fatal(&frame, eva);
317 if (!(frame.tf_eflags & PSL_I)) {
319 * Buggy application or kernel code has disabled interrupts
320 * and then trapped. Enabling interrupts now is wrong, but
321 * it is better than running with interrupts disabled until
322 * they are accidentally enabled later.
324 type = frame.tf_trapno;
325 if (ISPL(frame.tf_cs) == SEL_UPL || (frame.tf_eflags & PSL_VM))
327 "pid %ld (%s): trap %d with interrupts disabled\n",
328 (long)curproc->p_pid, curproc->p_comm, type);
329 else if (type != T_BPTFLT && type != T_TRCTRAP)
331 * XXX not quite right, since this may be for a
332 * multiple fault in user mode.
334 printf("kernel trap %d with interrupts disabled\n",
340 if (frame.tf_trapno == T_PAGEFLT) {
342 * For some Cyrix CPUs, %cr2 is clobbered by interrupts.
343 * This problem is worked around by using an interrupt
344 * gate for the pagefault handler. We are finally ready
345 * to read %cr2 and then must reenable interrupts.
347 * XXX this should be in the switch statement, but the
348 * NO_FOOF_HACK and VM86 goto and ifdefs obfuscate the
349 * flow of control too much for this to be obviously
356 #ifdef DEVICE_POLLING
358 ether_poll(poll_in_trap);
359 #endif /* DEVICE_POLLING */
361 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
364 type = frame.tf_trapno;
368 if (frame.tf_eflags & PSL_VM &&
369 (type == T_PROTFLT || type == T_STKFLT)) {
370 i = vm86_emulate((struct vm86frame *)&frame);
373 * returns to original process
375 vm86_trap((struct vm86frame *)&frame);
380 * these traps want either a process context, or
381 * assume a normal userspace trap.
385 trap_fatal(&frame, eva);
388 type = T_BPTFLT; /* kernel breakpoint */
391 goto kernel_trap; /* normal kernel trap handling */
394 if ((ISPL(frame.tf_cs) == SEL_UPL) || (frame.tf_eflags & PSL_VM)) {
399 sticks = curthread->td_sticks;
400 p->p_md.md_regs = &frame;
403 case T_PRIVINFLT: /* privileged instruction fault */
408 case T_BPTFLT: /* bpt instruction fault */
409 case T_TRCTRAP: /* trace trap */
410 frame.tf_eflags &= ~PSL_T;
414 case T_ARITHTRAP: /* arithmetic trap */
419 case T_ASTFLT: /* Allow process switch */
422 if (p->p_flag & P_OWEUPC) {
423 p->p_flag &= ~P_OWEUPC;
424 addupc_task(p, p->p_stats->p_prof.pr_addr,
425 p->p_stats->p_prof.pr_ticks);
430 * The following two traps can happen in
431 * vm86 mode, and, if so, we want to handle
434 case T_PROTFLT: /* general protection fault */
435 case T_STKFLT: /* stack fault */
436 if (frame.tf_eflags & PSL_VM) {
437 i = vm86_emulate((struct vm86frame *)&frame);
444 case T_SEGNPFLT: /* segment not present fault */
445 case T_TSSFLT: /* invalid TSS fault */
446 case T_DOUBLEFLT: /* double fault */
448 ucode = code + BUS_SEGM_FAULT ;
452 case T_PAGEFLT: /* page fault */
453 i = trap_pfault(&frame, TRUE, eva);
456 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
466 case T_DIVIDE: /* integer divide fault */
474 goto handle_powerfail;
475 #else /* !POWERFAIL_NMI */
476 /* machine/parity/power fail/"kitchen sink" faults */
477 if (isa_nmi(code) == 0) {
480 * NMI can be hooked up to a pushbutton
484 printf ("NMI ... going to debugger\n");
485 kdb_trap (type, 0, &frame);
489 } else if (panic_on_nmi)
490 panic("NMI indicates hardware failure");
492 #endif /* POWERFAIL_NMI */
493 #endif /* NISA > 0 */
495 case T_OFLOW: /* integer overflow fault */
500 case T_BOUND: /* bounds check fault */
507 /* if a transparent fault (due to context switch "late") */
511 if (!pmath_emulate) {
513 ucode = FPE_FPU_NP_TRAP;
516 i = (*pmath_emulate)(&frame);
518 if (!(frame.tf_eflags & PSL_T))
520 frame.tf_eflags &= ~PSL_T;
523 /* else ucode = emulator_only_knows() XXX */
526 case T_FPOPFLT: /* FPU operand fetch fault */
531 case T_XMMFLT: /* SIMD floating-point exception */
541 case T_PAGEFLT: /* page fault */
542 (void) trap_pfault(&frame, FALSE, eva);
548 * The kernel is apparently using npx for copying.
549 * XXX this should be fatal unless the kernel has
550 * registered such use.
557 case T_PROTFLT: /* general protection fault */
558 case T_SEGNPFLT: /* segment not present fault */
560 * Invalid segment selectors and out of bounds
561 * %eip's and %esp's can be set up in user mode.
562 * This causes a fault in kernel mode when the
563 * kernel tries to return to user mode. We want
564 * to get this fault so that we can fix the
565 * problem here and not have to check all the
566 * selectors and pointers when the user changes
569 #define MAYBE_DORETI_FAULT(where, whereto) \
571 if (frame.tf_eip == (int)where) { \
572 frame.tf_eip = (int)whereto; \
577 if (mycpu->gd_intr_nesting_level == 0) {
579 * Invalid %fs's and %gs's can be created using
580 * procfs or PT_SETREGS or by invalidating the
581 * underlying LDT entry. This causes a fault
582 * in kernel mode when the kernel attempts to
583 * switch contexts. Lose the bad context
584 * (XXX) so that we can continue, and generate
587 if (frame.tf_eip == (int)cpu_switch_load_gs) {
588 curthread->td_pcb->pcb_gs = 0;
592 MAYBE_DORETI_FAULT(doreti_iret,
594 MAYBE_DORETI_FAULT(doreti_popl_ds,
595 doreti_popl_ds_fault);
596 MAYBE_DORETI_FAULT(doreti_popl_es,
597 doreti_popl_es_fault);
598 MAYBE_DORETI_FAULT(doreti_popl_fs,
599 doreti_popl_fs_fault);
600 if (curthread->td_pcb->pcb_onfault) {
601 frame.tf_eip = (int)curthread->td_pcb->pcb_onfault;
609 * PSL_NT can be set in user mode and isn't cleared
610 * automatically when the kernel is entered. This
611 * causes a TSS fault when the kernel attempts to
612 * `iret' because the TSS link is uninitialized. We
613 * want to get this fault so that we can fix the
614 * problem here and not every time the kernel is
617 if (frame.tf_eflags & PSL_NT) {
618 frame.tf_eflags &= ~PSL_NT;
623 case T_TRCTRAP: /* trace trap */
624 if (frame.tf_eip == (int)IDTVEC(syscall)) {
626 * We've just entered system mode via the
627 * syscall lcall. Continue single stepping
628 * silently until the syscall handler has
633 if (frame.tf_eip == (int)IDTVEC(syscall) + 1) {
635 * The syscall handler has now saved the
636 * flags. Stop single stepping it.
638 frame.tf_eflags &= ~PSL_T;
642 * Ignore debug register trace traps due to
643 * accesses in the user's address space, which
644 * can happen under several conditions such as
645 * if a user sets a watchpoint on a buffer and
646 * then passes that buffer to a system call.
647 * We still want to get TRCTRAPS for addresses
648 * in kernel space because that is useful when
649 * debugging the kernel.
651 if (user_dbreg_trap()) {
653 * Reset breakpoint bits because the
656 load_dr6(rdr6() & 0xfffffff0);
660 * Fall through (TRCTRAP kernel mode, kernel address)
664 * If DDB is enabled, let it handle the debugger trap.
665 * Otherwise, debugger traps "can't happen".
668 if (kdb_trap (type, 0, &frame))
677 # define TIMER_FREQ 1193182
681 static unsigned lastalert = 0;
683 if(time_second - lastalert > 10)
685 log(LOG_WARNING, "NMI: power fail\n");
686 sysbeep(TIMER_FREQ/880, hz);
687 lastalert = time_second;
691 #else /* !POWERFAIL_NMI */
692 /* machine/parity/power fail/"kitchen sink" faults */
693 if (isa_nmi(code) == 0) {
696 * NMI can be hooked up to a pushbutton
700 printf ("NMI ... going to debugger\n");
701 kdb_trap (type, 0, &frame);
705 } else if (panic_on_nmi == 0)
708 #endif /* POWERFAIL_NMI */
709 #endif /* NISA > 0 */
712 trap_fatal(&frame, eva);
716 /* Translate fault for emulators (e.g. Linux) */
717 if (*p->p_sysent->sv_transtrap)
718 i = (*p->p_sysent->sv_transtrap)(i, type);
720 trapsignal(p, i, ucode);
723 if (type <= MAX_TRAP_MSG) {
724 uprintf("fatal process exception: %s",
726 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
727 uprintf(", fault VA = 0x%lx", (u_long)eva);
733 userret(p, &frame, sticks, 1);
738 * This version doesn't allow a page fault to user space while
739 * in the kernel. The rest of the kernel needs to be made "safe"
740 * before this can be used. I think the only things remaining
741 * to be made safe are the iBCS2 code and the process tracing/
745 trap_pfault(frame, usermode, eva)
746 struct trapframe *frame;
751 struct vmspace *vm = NULL;
755 struct proc *p = curproc;
757 if (frame->tf_err & PGEX_W)
758 ftype = VM_PROT_WRITE;
760 ftype = VM_PROT_READ;
762 va = trunc_page(eva);
763 if (va < VM_MIN_KERNEL_ADDRESS) {
768 (!usermode && va < VM_MAXUSER_ADDRESS &&
769 (mycpu->gd_intr_nesting_level != 0 ||
770 curthread->td_pcb->pcb_onfault == NULL))) {
771 trap_fatal(frame, eva);
776 * This is a fault on non-kernel virtual memory.
777 * vm is initialized above to NULL. If curproc is NULL
778 * or curproc->p_vmspace is NULL the fault is fatal.
787 * Keep swapout from messing with us during this
793 * Grow the stack if necessary
795 /* grow_stack returns false only if va falls into
796 * a growable stack region and the stack growth
797 * fails. It returns true if va was not within
798 * a growable stack region, or if the stack
801 if (!grow_stack (p, va)) {
807 /* Fault in the user page: */
808 rv = vm_fault(map, va, ftype,
809 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
815 * Don't allow user-mode faults in kernel address space.
821 * Since we know that kernel virtual address addresses
822 * always have pte pages mapped, we just have to fault
825 rv = vm_fault(kernel_map, va, ftype, VM_FAULT_NORMAL);
828 if (rv == KERN_SUCCESS)
832 if (mycpu->gd_intr_nesting_level == 0 && curthread->td_pcb->pcb_onfault) {
833 frame->tf_eip = (int)curthread->td_pcb->pcb_onfault;
836 trap_fatal(frame, eva);
840 /* kludge to pass faulting virtual address to sendsig */
843 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
848 trap_pfault(frame, usermode, eva)
849 struct trapframe *frame;
854 struct vmspace *vm = NULL;
858 struct proc *p = curproc;
860 va = trunc_page(eva);
861 if (va >= KERNBASE) {
863 * Don't allow user-mode faults in kernel address space.
864 * An exception: if the faulting address is the invalid
865 * instruction entry in the IDT, then the Intel Pentium
866 * F00F bug workaround was triggered, and we need to
867 * treat it is as an illegal instruction, and not a page
870 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
871 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) {
872 frame->tf_trapno = T_PRIVINFLT;
882 * This is a fault on non-kernel virtual memory.
883 * vm is initialized above to NULL. If curproc is NULL
884 * or curproc->p_vmspace is NULL the fault is fatal.
895 if (frame->tf_err & PGEX_W)
896 ftype = VM_PROT_WRITE;
898 ftype = VM_PROT_READ;
900 if (map != kernel_map) {
902 * Keep swapout from messing with us during this
908 * Grow the stack if necessary
910 /* grow_stack returns false only if va falls into
911 * a growable stack region and the stack growth
912 * fails. It returns true if va was not within
913 * a growable stack region, or if the stack
916 if (!grow_stack (p, va)) {
922 /* Fault in the user page: */
923 rv = vm_fault(map, va, ftype,
924 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
930 * Don't have to worry about process locking or stacks in the kernel.
932 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
935 if (rv == KERN_SUCCESS)
939 if (mycpu->gd_intr_nesting_level == 0 && curthread->td_pcb->pcb_onfault) {
940 frame->tf_eip = (int)curthread->td_pcb->pcb_onfault;
943 trap_fatal(frame, eva);
947 /* kludge to pass faulting virtual address to sendsig */
950 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
954 trap_fatal(frame, eva)
955 struct trapframe *frame;
958 int code, type, ss, esp;
959 struct soft_segment_descriptor softseg;
961 code = frame->tf_err;
962 type = frame->tf_trapno;
963 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
965 if (type <= MAX_TRAP_MSG)
966 printf("\n\nFatal trap %d: %s while in %s mode\n",
967 type, trap_msg[type],
968 frame->tf_eflags & PSL_VM ? "vm86" :
969 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
971 /* three seperate prints in case of a trap on an unmapped page */
972 printf("mp_lock = %08x; ", mp_lock);
973 printf("cpuid = %d; ", cpuid);
974 printf("lapic.id = %08x\n", lapic.id);
976 if (type == T_PAGEFLT) {
977 printf("fault virtual address = 0x%x\n", eva);
978 printf("fault code = %s %s, %s\n",
979 code & PGEX_U ? "user" : "supervisor",
980 code & PGEX_W ? "write" : "read",
981 code & PGEX_P ? "protection violation" : "page not present");
983 printf("instruction pointer = 0x%x:0x%x\n",
984 frame->tf_cs & 0xffff, frame->tf_eip);
985 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
986 ss = frame->tf_ss & 0xffff;
989 ss = GSEL(GDATA_SEL, SEL_KPL);
990 esp = (int)&frame->tf_esp;
992 printf("stack pointer = 0x%x:0x%x\n", ss, esp);
993 printf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
994 printf("code segment = base 0x%x, limit 0x%x, type 0x%x\n",
995 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
996 printf(" = DPL %d, pres %d, def32 %d, gran %d\n",
997 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
999 printf("processor eflags = ");
1000 if (frame->tf_eflags & PSL_T)
1001 printf("trace trap, ");
1002 if (frame->tf_eflags & PSL_I)
1003 printf("interrupt enabled, ");
1004 if (frame->tf_eflags & PSL_NT)
1005 printf("nested task, ");
1006 if (frame->tf_eflags & PSL_RF)
1008 if (frame->tf_eflags & PSL_VM)
1010 printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
1011 printf("current process = ");
1013 printf("%lu (%s)\n",
1014 (u_long)curproc->p_pid, curproc->p_comm ?
1015 curproc->p_comm : "");
1019 printf("current thread = pri %d ", curthread->td_pri);
1020 if (curthread->td_pri >= TDPRI_CRIT)
1023 printf("interrupt mask = ");
1024 if ((curthread->td_cpl & net_imask) == net_imask)
1026 if ((curthread->td_cpl & tty_imask) == tty_imask)
1028 if ((curthread->td_cpl & bio_imask) == bio_imask)
1030 if ((curthread->td_cpl & cam_imask) == cam_imask)
1032 if (curthread->td_cpl == 0)
1037 * we probably SHOULD have stopped the other CPUs before now!
1038 * another CPU COULD have been touching cpl at this moment...
1040 printf(" <- SMP: XXX");
1049 if ((debugger_on_panic || db_active) && kdb_trap(type, 0, frame))
1052 printf("trap number = %d\n", type);
1053 if (type <= MAX_TRAP_MSG)
1054 panic("%s", trap_msg[type]);
1056 panic("unknown/reserved trap");
1060 * Double fault handler. Called when a fault occurs while writing
1061 * a frame for a trap/exception onto the stack. This usually occurs
1062 * when the stack overflows (such is the case with infinite recursion,
1065 * XXX Note that the current PTD gets replaced by IdlePTD when the
1066 * task switch occurs. This means that the stack that was active at
1067 * the time of the double fault is not available at <kstack> unless
1068 * the machine was idle when the double fault occurred. The downside
1069 * of this is that "trace <ebp>" in ddb won't work.
1074 struct mdglobaldata *gd = mdcpu;
1076 printf("\nFatal double fault:\n");
1077 printf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1078 printf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1079 printf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
1081 /* three seperate prints in case of a trap on an unmapped page */
1082 printf("mp_lock = %08x; ", mp_lock);
1083 printf("cpuid = %d; ", cpuid);
1084 printf("lapic.id = %08x\n", lapic.id);
1086 panic("double fault");
1090 * Compensate for 386 brain damage (missing URKR).
1091 * This is a little simpler than the pagefault handler in trap() because
1092 * it the page tables have already been faulted in and high addresses
1093 * are thrown out early for other reasons.
1103 va = trunc_page((vm_offset_t)addr);
1105 * XXX - MAX is END. Changed > to >= for temp. fix.
1107 if (va >= VM_MAXUSER_ADDRESS)
1115 if (!grow_stack (p, va)) {
1121 * fault the data page
1123 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY);
1127 if (rv != KERN_SUCCESS)
1134 * syscall2 - MP aware system call request C handler
1136 * A system call is essentially treated as a trap except that the
1137 * MP lock is not held on entry or return. We are responsible for
1138 * obtaining the MP lock if necessary and for handling ASTs
1139 * (e.g. a task switch) prior to return.
1141 * In general, only simple access and manipulation of curproc and
1142 * the current stack is allowed without having to hold MP lock.
1146 struct trapframe frame;
1148 struct thread *td = curthread;
1149 struct proc *p = td->td_proc;
1152 struct sysent *callp;
1153 register_t orig_tf_eflags;
1158 int have_mplock = 0;
1162 if (ISPL(frame.tf_cs) != SEL_UPL) {
1170 * access non-atomic field from critical section. p_sticks is
1171 * updated by the clock interrupt. Also use this opportunity
1172 * to raise our LWKT priority.
1176 sticks = curthread->td_sticks;
1179 p->p_md.md_regs = &frame;
1180 params = (caddr_t)frame.tf_esp + sizeof(int);
1181 code = frame.tf_eax;
1182 orig_tf_eflags = frame.tf_eflags;
1184 if (p->p_sysent->sv_prepsyscall) {
1186 * The prep code is not MP aware.
1189 (*p->p_sysent->sv_prepsyscall)(&frame, args, &code, ¶ms);
1193 * Need to check if this is a 32 bit or 64 bit syscall.
1194 * fuword is MP aware.
1196 if (code == SYS_syscall) {
1198 * Code is first argument, followed by actual args.
1200 code = fuword(params);
1201 params += sizeof(int);
1202 } else if (code == SYS___syscall) {
1204 * Like syscall, but code is a quad, so as to maintain
1205 * quad alignment for the rest of the arguments.
1207 code = fuword(params);
1208 params += sizeof(quad_t);
1212 if (p->p_sysent->sv_mask)
1213 code &= p->p_sysent->sv_mask;
1215 if (code >= p->p_sysent->sv_size)
1216 callp = &p->p_sysent->sv_table[0];
1218 callp = &p->p_sysent->sv_table[code];
1220 narg = callp->sy_narg & SYF_ARGMASK;
1223 * copyin is MP aware, but the tracing code is not
1225 if (params && (i = narg * sizeof(int)) &&
1226 (error = copyin(params, (caddr_t)args, (u_int)i))) {
1230 if (KTRPOINT(td, KTR_SYSCALL))
1231 ktrsyscall(p->p_tracep, code, narg, args);
1237 * Try to run the syscall without the MP lock if the syscall
1238 * is MP safe. We have to obtain the MP lock no matter what if
1241 if ((callp->sy_narg & SYF_MPSAFE) == 0) {
1247 if (KTRPOINT(td, KTR_SYSCALL)) {
1248 if (have_mplock == 0) {
1252 ktrsyscall(p->p_tracep, code, narg, args);
1256 p->p_retval[1] = frame.tf_edx;
1258 STOPEVENT(p, S_SCE, narg); /* MP aware */
1260 error = (*callp->sy_call)(args);
1263 * MP SAFE (we may or may not have the MP lock at this point)
1268 * Reinitialize proc pointer `p' as it may be different
1269 * if this is a child returning from fork syscall.
1272 frame.tf_eax = p->p_retval[0];
1273 frame.tf_edx = p->p_retval[1];
1274 frame.tf_eflags &= ~PSL_C;
1279 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1280 * int 0x80 is 2 bytes. We saved this in tf_err.
1282 frame.tf_eip -= frame.tf_err;
1290 if (p->p_sysent->sv_errsize) {
1291 if (error >= p->p_sysent->sv_errsize)
1292 error = -1; /* XXX */
1294 error = p->p_sysent->sv_errtbl[error];
1296 frame.tf_eax = error;
1297 frame.tf_eflags |= PSL_C;
1302 * Traced syscall. trapsignal() is not MP aware.
1304 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1305 if (have_mplock == 0) {
1309 frame.tf_eflags &= ~PSL_T;
1310 trapsignal(p, SIGTRAP, 0);
1314 * Handle reschedule and other end-of-syscall issues
1316 have_mplock = userret(p, &frame, sticks, have_mplock);
1319 if (KTRPOINT(td, KTR_SYSRET)) {
1320 if (have_mplock == 0) {
1324 ktrsysret(p->p_tracep, code, error, p->p_retval[0]);
1329 * This works because errno is findable through the
1330 * register set. If we ever support an emulation where this
1331 * is not the case, this code will need to be revisited.
1333 STOPEVENT(p, S_SCX, code);
1336 * Release the MP lock if we had to get it
1343 * Simplified back end of syscall(), used when returning from fork()
1344 * directly into user mode. MP lock is held on entry and should be
1348 fork_return(p, frame)
1350 struct trapframe frame;
1352 frame.tf_eax = 0; /* Child returns zero */
1353 frame.tf_eflags &= ~PSL_C; /* success */
1356 userret(p, &frame, 0, 1);
1358 if (KTRPOINT(p->p_thread, KTR_SYSRET))
1359 ktrsysret(p->p_tracep, SYS_fork, 0, 0);