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 1.25 2003/07/20 07:14:23 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");
159 static int fast_release;
160 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
161 &fast_release, 0, "Passive Release was optimal");
162 static int slow_release;
163 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
164 &slow_release, 0, "Passive Release was nonoptimal");
167 * USER->KERNEL transition. Do not transition us out of userland from the
168 * point of view of the userland scheduler unless we actually have to
171 * usertdsw is called from within a critical section and the BGL will still
172 * be held. This function is NOT called for preemptions, only for switchouts.
175 passive_release(struct thread *td)
177 struct proc *p = td->td_proc;
179 td->td_release = NULL;
180 lwkt_setpri_self(TDPRI_KERN_USER);
181 if (p->p_flag & P_CURPROC) {
187 * userenter() passively intercepts the thread switch function to increase
188 * the thread priority from a user priority to a kernel priority, reducing
189 * syscall and trap overhead for the case where no switch occurs.
195 struct thread *td = curthread;
197 td->td_release = passive_release;
201 userexit(struct proc *p)
203 struct thread *td = p->p_thread;
206 * If we did not have to release we should already be P_CURPROC. If
207 * we did have to release we must acquire P_CURPROC again and then
208 * restore our priority for user return.
210 * Lowering our priority may make other higher priority threads
211 * runnable. lwkt_setpri_self() does not switch away, so call
212 * lwkt_maybe_switch() to deal with it.
214 if (td->td_release) {
216 td->td_release = NULL;
217 KKASSERT(p->p_flag & P_CURPROC);
221 switch(p->p_rtprio.type) {
223 lwkt_setpri_self(TDPRI_USER_IDLE);
225 case RTP_PRIO_REALTIME:
227 lwkt_setpri_self(TDPRI_USER_REAL);
230 lwkt_setpri_self(TDPRI_USER_NORM);
239 userret(struct proc *p, struct trapframe *frame, u_quad_t oticks)
244 * Post any pending signals
246 while ((sig = CURSIG(p)) != 0) {
251 * If a reschedule has been requested then the easiest solution
252 * is to run our passive release function which will possibly
253 * shift our P_CURPROC designation to another user process.
254 * We don't actually switch here because that would be a waste
255 * of cycles (the newly scheduled user process would just switch
256 * back to us since we might be running at a kernel priority).
257 * Instead we fall through and will switch away when we attempt
258 * to reacquire our P_CURPROC designation.
260 if (resched_wanted()) {
261 if (curthread->td_release)
262 passive_release(curthread);
266 * Charge system time if profiling. Note: times are in microseconds.
268 if (p->p_flag & P_PROFIL) {
269 addupc_task(p, frame->tf_eip,
270 (u_int)(curthread->td_sticks - oticks));
274 * Post any pending signals XXX
276 while ((sig = CURSIG(p)) != 0)
280 #ifdef DEVICE_POLLING
281 extern u_int32_t poll_in_trap;
282 extern int ether_poll __P((int count));
283 #endif /* DEVICE_POLLING */
286 * Exception, fault, and trap interface to the FreeBSD kernel.
287 * This common code is called from assembly language IDT gate entry
288 * routines that prepare a suitable stack frame, and restore this
289 * frame after the exception has been processed.
291 * This function is also called from doreti in an interlock to handle ASTs.
292 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
294 * NOTE! We have to retrieve the fault address prior to obtaining the
295 * MP lock because get_mplock() may switch out. YYY cr2 really ought
296 * to be retrieved by the assembly code, not here.
300 struct trapframe frame;
302 struct proc *p = curproc;
304 int i = 0, ucode = 0, type, code;
309 eva = (frame.tf_trapno == T_PAGEFLT ? rcr2() : 0);
311 trap_fatal(&frame, eva);
317 if (frame.tf_trapno == T_PAGEFLT) {
319 * For some Cyrix CPUs, %cr2 is clobbered by interrupts.
320 * This problem is worked around by using an interrupt
321 * gate for the pagefault handler. We are finally ready
322 * to read %cr2 and then must reenable interrupts.
324 * XXX this should be in the switch statement, but the
325 * NO_FOOF_HACK and VM86 goto and ifdefs obfuscate the
326 * flow of control too much for this to be obviously
336 * MP lock is held at this point
339 if (!(frame.tf_eflags & PSL_I)) {
341 * Buggy application or kernel code has disabled interrupts
342 * and then trapped. Enabling interrupts now is wrong, but
343 * it is better than running with interrupts disabled until
344 * they are accidentally enabled later.
346 type = frame.tf_trapno;
347 if (ISPL(frame.tf_cs)==SEL_UPL || (frame.tf_eflags & PSL_VM)) {
349 "pid %ld (%s): trap %d with interrupts disabled\n",
350 (long)curproc->p_pid, curproc->p_comm, type);
351 } else if (type != T_BPTFLT && type != T_TRCTRAP) {
353 * XXX not quite right, since this may be for a
354 * multiple fault in user mode.
356 printf("kernel trap %d with interrupts disabled\n",
363 #ifdef DEVICE_POLLING
365 ether_poll(poll_in_trap);
366 #endif /* DEVICE_POLLING */
368 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
371 type = frame.tf_trapno;
375 if (frame.tf_eflags & PSL_VM &&
376 (type == T_PROTFLT || type == T_STKFLT)) {
378 KKASSERT(curthread->td_mpcount > 0);
380 i = vm86_emulate((struct vm86frame *)&frame);
382 KKASSERT(curthread->td_mpcount > 0);
386 * returns to original process
388 vm86_trap((struct vm86frame *)&frame);
395 * these traps want either a process context, or
396 * assume a normal userspace trap.
400 trap_fatal(&frame, eva);
403 type = T_BPTFLT; /* kernel breakpoint */
406 goto kernel_trap; /* normal kernel trap handling */
409 if ((ISPL(frame.tf_cs) == SEL_UPL) || (frame.tf_eflags & PSL_VM)) {
414 sticks = curthread->td_sticks;
415 p->p_md.md_regs = &frame;
418 case T_PRIVINFLT: /* privileged instruction fault */
423 case T_BPTFLT: /* bpt instruction fault */
424 case T_TRCTRAP: /* trace trap */
425 frame.tf_eflags &= ~PSL_T;
429 case T_ARITHTRAP: /* arithmetic trap */
434 case T_ASTFLT: /* Allow process switch */
435 mycpu->gd_cnt.v_soft++;
436 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
437 atomic_clear_int_nonlocked(&mycpu->gd_reqflags,
439 addupc_task(p, p->p_stats->p_prof.pr_addr,
440 p->p_stats->p_prof.pr_ticks);
445 * The following two traps can happen in
446 * vm86 mode, and, if so, we want to handle
449 case T_PROTFLT: /* general protection fault */
450 case T_STKFLT: /* stack fault */
451 if (frame.tf_eflags & PSL_VM) {
452 i = vm86_emulate((struct vm86frame *)&frame);
459 case T_SEGNPFLT: /* segment not present fault */
460 case T_TSSFLT: /* invalid TSS fault */
461 case T_DOUBLEFLT: /* double fault */
463 ucode = code + BUS_SEGM_FAULT ;
467 case T_PAGEFLT: /* page fault */
468 i = trap_pfault(&frame, TRUE, eva);
471 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
481 case T_DIVIDE: /* integer divide fault */
489 goto handle_powerfail;
490 #else /* !POWERFAIL_NMI */
491 /* machine/parity/power fail/"kitchen sink" faults */
492 if (isa_nmi(code) == 0) {
495 * NMI can be hooked up to a pushbutton
499 printf ("NMI ... going to debugger\n");
500 kdb_trap (type, 0, &frame);
504 } else if (panic_on_nmi)
505 panic("NMI indicates hardware failure");
507 #endif /* POWERFAIL_NMI */
508 #endif /* NISA > 0 */
510 case T_OFLOW: /* integer overflow fault */
515 case T_BOUND: /* bounds check fault */
522 /* if a transparent fault (due to context switch "late") */
526 if (!pmath_emulate) {
528 ucode = FPE_FPU_NP_TRAP;
531 i = (*pmath_emulate)(&frame);
533 if (!(frame.tf_eflags & PSL_T))
535 frame.tf_eflags &= ~PSL_T;
538 /* else ucode = emulator_only_knows() XXX */
541 case T_FPOPFLT: /* FPU operand fetch fault */
546 case T_XMMFLT: /* SIMD floating-point exception */
556 case T_PAGEFLT: /* page fault */
557 (void) trap_pfault(&frame, FALSE, eva);
563 * The kernel is apparently using npx for copying.
564 * XXX this should be fatal unless the kernel has
565 * registered such use.
572 case T_PROTFLT: /* general protection fault */
573 case T_SEGNPFLT: /* segment not present fault */
575 * Invalid segment selectors and out of bounds
576 * %eip's and %esp's can be set up in user mode.
577 * This causes a fault in kernel mode when the
578 * kernel tries to return to user mode. We want
579 * to get this fault so that we can fix the
580 * problem here and not have to check all the
581 * selectors and pointers when the user changes
584 #define MAYBE_DORETI_FAULT(where, whereto) \
586 if (frame.tf_eip == (int)where) { \
587 frame.tf_eip = (int)whereto; \
592 if (mycpu->gd_intr_nesting_level == 0) {
594 * Invalid %fs's and %gs's can be created using
595 * procfs or PT_SETREGS or by invalidating the
596 * underlying LDT entry. This causes a fault
597 * in kernel mode when the kernel attempts to
598 * switch contexts. Lose the bad context
599 * (XXX) so that we can continue, and generate
602 if (frame.tf_eip == (int)cpu_switch_load_gs) {
603 curthread->td_pcb->pcb_gs = 0;
607 MAYBE_DORETI_FAULT(doreti_iret,
609 MAYBE_DORETI_FAULT(doreti_popl_ds,
610 doreti_popl_ds_fault);
611 MAYBE_DORETI_FAULT(doreti_popl_es,
612 doreti_popl_es_fault);
613 MAYBE_DORETI_FAULT(doreti_popl_fs,
614 doreti_popl_fs_fault);
615 if (curthread->td_pcb->pcb_onfault) {
616 frame.tf_eip = (int)curthread->td_pcb->pcb_onfault;
624 * PSL_NT can be set in user mode and isn't cleared
625 * automatically when the kernel is entered. This
626 * causes a TSS fault when the kernel attempts to
627 * `iret' because the TSS link is uninitialized. We
628 * want to get this fault so that we can fix the
629 * problem here and not every time the kernel is
632 if (frame.tf_eflags & PSL_NT) {
633 frame.tf_eflags &= ~PSL_NT;
638 case T_TRCTRAP: /* trace trap */
639 if (frame.tf_eip == (int)IDTVEC(syscall)) {
641 * We've just entered system mode via the
642 * syscall lcall. Continue single stepping
643 * silently until the syscall handler has
648 if (frame.tf_eip == (int)IDTVEC(syscall) + 1) {
650 * The syscall handler has now saved the
651 * flags. Stop single stepping it.
653 frame.tf_eflags &= ~PSL_T;
657 * Ignore debug register trace traps due to
658 * accesses in the user's address space, which
659 * can happen under several conditions such as
660 * if a user sets a watchpoint on a buffer and
661 * then passes that buffer to a system call.
662 * We still want to get TRCTRAPS for addresses
663 * in kernel space because that is useful when
664 * debugging the kernel.
666 if (user_dbreg_trap()) {
668 * Reset breakpoint bits because the
671 load_dr6(rdr6() & 0xfffffff0);
675 * Fall through (TRCTRAP kernel mode, kernel address)
679 * If DDB is enabled, let it handle the debugger trap.
680 * Otherwise, debugger traps "can't happen".
683 if (kdb_trap (type, 0, &frame))
692 # define TIMER_FREQ 1193182
696 static unsigned lastalert = 0;
698 if(time_second - lastalert > 10)
700 log(LOG_WARNING, "NMI: power fail\n");
701 sysbeep(TIMER_FREQ/880, hz);
702 lastalert = time_second;
707 #else /* !POWERFAIL_NMI */
708 /* machine/parity/power fail/"kitchen sink" faults */
709 if (isa_nmi(code) == 0) {
712 * NMI can be hooked up to a pushbutton
716 printf ("NMI ... going to debugger\n");
717 kdb_trap (type, 0, &frame);
721 } else if (panic_on_nmi == 0)
724 #endif /* POWERFAIL_NMI */
725 #endif /* NISA > 0 */
728 trap_fatal(&frame, eva);
732 /* Translate fault for emulators (e.g. Linux) */
733 if (*p->p_sysent->sv_transtrap)
734 i = (*p->p_sysent->sv_transtrap)(i, type);
736 trapsignal(p, i, ucode);
739 if (type <= MAX_TRAP_MSG) {
740 uprintf("fatal process exception: %s",
742 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
743 uprintf(", fault VA = 0x%lx", (u_long)eva);
750 if (ISPL(frame.tf_cs) == SEL_UPL)
751 KASSERT(curthread->td_mpcount == 1, ("badmpcount trap from %p", (void *)frame.tf_eip));
753 userret(p, &frame, sticks);
757 KKASSERT(curthread->td_mpcount > 0);
764 * This version doesn't allow a page fault to user space while
765 * in the kernel. The rest of the kernel needs to be made "safe"
766 * before this can be used. I think the only things remaining
767 * to be made safe are the iBCS2 code and the process tracing/
771 trap_pfault(frame, usermode, eva)
772 struct trapframe *frame;
777 struct vmspace *vm = NULL;
781 struct proc *p = curproc;
783 if (frame->tf_err & PGEX_W)
784 ftype = VM_PROT_WRITE;
786 ftype = VM_PROT_READ;
788 va = trunc_page(eva);
789 if (va < VM_MIN_KERNEL_ADDRESS) {
794 (!usermode && va < VM_MAXUSER_ADDRESS &&
795 (mycpu->gd_intr_nesting_level != 0 ||
796 curthread->td_pcb->pcb_onfault == NULL))) {
797 trap_fatal(frame, eva);
802 * This is a fault on non-kernel virtual memory.
803 * vm is initialized above to NULL. If curproc is NULL
804 * or curproc->p_vmspace is NULL the fault is fatal.
813 * Keep swapout from messing with us during this
819 * Grow the stack if necessary
821 /* grow_stack returns false only if va falls into
822 * a growable stack region and the stack growth
823 * fails. It returns true if va was not within
824 * a growable stack region, or if the stack
827 if (!grow_stack (p, va)) {
833 /* Fault in the user page: */
834 rv = vm_fault(map, va, ftype,
835 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
841 * Don't allow user-mode faults in kernel address space.
847 * Since we know that kernel virtual address addresses
848 * always have pte pages mapped, we just have to fault
851 rv = vm_fault(kernel_map, va, ftype, VM_FAULT_NORMAL);
854 if (rv == KERN_SUCCESS)
858 if (mycpu->gd_intr_nesting_level == 0 && curthread->td_pcb->pcb_onfault) {
859 frame->tf_eip = (int)curthread->td_pcb->pcb_onfault;
862 trap_fatal(frame, eva);
866 /* kludge to pass faulting virtual address to sendsig */
869 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
874 trap_pfault(frame, usermode, eva)
875 struct trapframe *frame;
880 struct vmspace *vm = NULL;
884 struct proc *p = curproc;
886 va = trunc_page(eva);
887 if (va >= KERNBASE) {
889 * Don't allow user-mode faults in kernel address space.
890 * An exception: if the faulting address is the invalid
891 * instruction entry in the IDT, then the Intel Pentium
892 * F00F bug workaround was triggered, and we need to
893 * treat it is as an illegal instruction, and not a page
896 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
897 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) {
898 frame->tf_trapno = T_PRIVINFLT;
908 * This is a fault on non-kernel virtual memory.
909 * vm is initialized above to NULL. If curproc is NULL
910 * or curproc->p_vmspace is NULL the fault is fatal.
921 if (frame->tf_err & PGEX_W)
922 ftype = VM_PROT_WRITE;
924 ftype = VM_PROT_READ;
926 if (map != kernel_map) {
928 * Keep swapout from messing with us during this
934 * Grow the stack if necessary
936 /* grow_stack returns false only if va falls into
937 * a growable stack region and the stack growth
938 * fails. It returns true if va was not within
939 * a growable stack region, or if the stack
942 if (!grow_stack (p, va)) {
948 /* Fault in the user page: */
949 rv = vm_fault(map, va, ftype,
950 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
956 * Don't have to worry about process locking or stacks in the kernel.
958 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
961 if (rv == KERN_SUCCESS)
965 if (mycpu->gd_intr_nesting_level == 0 && curthread->td_pcb->pcb_onfault) {
966 frame->tf_eip = (int)curthread->td_pcb->pcb_onfault;
969 trap_fatal(frame, eva);
973 /* kludge to pass faulting virtual address to sendsig */
976 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
980 trap_fatal(frame, eva)
981 struct trapframe *frame;
984 int code, type, ss, esp;
985 struct soft_segment_descriptor softseg;
987 code = frame->tf_err;
988 type = frame->tf_trapno;
989 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
991 if (type <= MAX_TRAP_MSG)
992 printf("\n\nFatal trap %d: %s while in %s mode\n",
993 type, trap_msg[type],
994 frame->tf_eflags & PSL_VM ? "vm86" :
995 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
997 /* three seperate prints in case of a trap on an unmapped page */
998 printf("mp_lock = %08x; ", mp_lock);
999 printf("cpuid = %d; ", mycpu->gd_cpuid);
1000 printf("lapic.id = %08x\n", lapic.id);
1002 if (type == T_PAGEFLT) {
1003 printf("fault virtual address = 0x%x\n", eva);
1004 printf("fault code = %s %s, %s\n",
1005 code & PGEX_U ? "user" : "supervisor",
1006 code & PGEX_W ? "write" : "read",
1007 code & PGEX_P ? "protection violation" : "page not present");
1009 printf("instruction pointer = 0x%x:0x%x\n",
1010 frame->tf_cs & 0xffff, frame->tf_eip);
1011 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
1012 ss = frame->tf_ss & 0xffff;
1013 esp = frame->tf_esp;
1015 ss = GSEL(GDATA_SEL, SEL_KPL);
1016 esp = (int)&frame->tf_esp;
1018 printf("stack pointer = 0x%x:0x%x\n", ss, esp);
1019 printf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
1020 printf("code segment = base 0x%x, limit 0x%x, type 0x%x\n",
1021 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
1022 printf(" = DPL %d, pres %d, def32 %d, gran %d\n",
1023 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
1025 printf("processor eflags = ");
1026 if (frame->tf_eflags & PSL_T)
1027 printf("trace trap, ");
1028 if (frame->tf_eflags & PSL_I)
1029 printf("interrupt enabled, ");
1030 if (frame->tf_eflags & PSL_NT)
1031 printf("nested task, ");
1032 if (frame->tf_eflags & PSL_RF)
1034 if (frame->tf_eflags & PSL_VM)
1036 printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
1037 printf("current process = ");
1039 printf("%lu (%s)\n",
1040 (u_long)curproc->p_pid, curproc->p_comm ?
1041 curproc->p_comm : "");
1045 printf("current thread = pri %d ", curthread->td_pri);
1046 if (curthread->td_pri >= TDPRI_CRIT)
1049 printf("interrupt mask = ");
1050 if ((curthread->td_cpl & net_imask) == net_imask)
1052 if ((curthread->td_cpl & tty_imask) == tty_imask)
1054 if ((curthread->td_cpl & bio_imask) == bio_imask)
1056 if ((curthread->td_cpl & cam_imask) == cam_imask)
1058 if (curthread->td_cpl == 0)
1063 * we probably SHOULD have stopped the other CPUs before now!
1064 * another CPU COULD have been touching cpl at this moment...
1066 printf(" <- SMP: XXX");
1075 if ((debugger_on_panic || db_active) && kdb_trap(type, 0, frame))
1078 printf("trap number = %d\n", type);
1079 if (type <= MAX_TRAP_MSG)
1080 panic("%s", trap_msg[type]);
1082 panic("unknown/reserved trap");
1086 * Double fault handler. Called when a fault occurs while writing
1087 * a frame for a trap/exception onto the stack. This usually occurs
1088 * when the stack overflows (such is the case with infinite recursion,
1091 * XXX Note that the current PTD gets replaced by IdlePTD when the
1092 * task switch occurs. This means that the stack that was active at
1093 * the time of the double fault is not available at <kstack> unless
1094 * the machine was idle when the double fault occurred. The downside
1095 * of this is that "trace <ebp>" in ddb won't work.
1100 struct mdglobaldata *gd = mdcpu;
1102 printf("\nFatal double fault:\n");
1103 printf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1104 printf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1105 printf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
1107 /* three seperate prints in case of a trap on an unmapped page */
1108 printf("mp_lock = %08x; ", mp_lock);
1109 printf("cpuid = %d; ", mycpu->gd_cpuid);
1110 printf("lapic.id = %08x\n", lapic.id);
1112 panic("double fault");
1116 * Compensate for 386 brain damage (missing URKR).
1117 * This is a little simpler than the pagefault handler in trap() because
1118 * it the page tables have already been faulted in and high addresses
1119 * are thrown out early for other reasons.
1129 va = trunc_page((vm_offset_t)addr);
1131 * XXX - MAX is END. Changed > to >= for temp. fix.
1133 if (va >= VM_MAXUSER_ADDRESS)
1141 if (!grow_stack (p, va)) {
1147 * fault the data page
1149 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY);
1153 if (rv != KERN_SUCCESS)
1160 * syscall2 - MP aware system call request C handler
1162 * A system call is essentially treated as a trap except that the
1163 * MP lock is not held on entry or return. We are responsible for
1164 * obtaining the MP lock if necessary and for handling ASTs
1165 * (e.g. a task switch) prior to return.
1167 * In general, only simple access and manipulation of curproc and
1168 * the current stack is allowed without having to hold MP lock.
1172 struct trapframe frame;
1174 struct thread *td = curthread;
1175 struct proc *p = td->td_proc;
1178 struct sysent *callp;
1179 register_t orig_tf_eflags;
1187 if (ISPL(frame.tf_cs) != SEL_UPL) {
1195 KASSERT(curthread->td_mpcount == 0, ("badmpcount syscall from %p", (void *)frame.tf_eip));
1199 * access non-atomic field from critical section. p_sticks is
1200 * updated by the clock interrupt. Also use this opportunity
1201 * to lazy-raise our LWKT priority.
1205 sticks = curthread->td_sticks;
1208 p->p_md.md_regs = &frame;
1209 params = (caddr_t)frame.tf_esp + sizeof(int);
1210 code = frame.tf_eax;
1211 orig_tf_eflags = frame.tf_eflags;
1213 if (p->p_sysent->sv_prepsyscall) {
1215 * The prep code is not MP aware.
1217 (*p->p_sysent->sv_prepsyscall)(&frame, args, &code, ¶ms);
1220 * Need to check if this is a 32 bit or 64 bit syscall.
1221 * fuword is MP aware.
1223 if (code == SYS_syscall) {
1225 * Code is first argument, followed by actual args.
1227 code = fuword(params);
1228 params += sizeof(int);
1229 } else if (code == SYS___syscall) {
1231 * Like syscall, but code is a quad, so as to maintain
1232 * quad alignment for the rest of the arguments.
1234 code = fuword(params);
1235 params += sizeof(quad_t);
1239 if (p->p_sysent->sv_mask)
1240 code &= p->p_sysent->sv_mask;
1242 if (code >= p->p_sysent->sv_size)
1243 callp = &p->p_sysent->sv_table[0];
1245 callp = &p->p_sysent->sv_table[code];
1247 narg = callp->sy_narg & SYF_ARGMASK;
1250 * copyin is MP aware, but the tracing code is not
1252 if (params && (i = narg * sizeof(int)) &&
1253 (error = copyin(params, (caddr_t)args, (u_int)i))) {
1255 if (KTRPOINT(td, KTR_SYSCALL))
1256 ktrsyscall(p->p_tracep, code, narg, args);
1263 * Try to run the syscall without the MP lock if the syscall
1264 * is MP safe. We have to obtain the MP lock no matter what if
1267 if ((callp->sy_narg & SYF_MPSAFE) == 0) {
1274 if (KTRPOINT(td, KTR_SYSCALL)) {
1275 ktrsyscall(p->p_tracep, code, narg, args);
1279 p->p_retval[1] = frame.tf_edx;
1281 STOPEVENT(p, S_SCE, narg); /* MP aware */
1283 error = (*callp->sy_call)(args);
1286 * MP SAFE (we may or may not have the MP lock at this point)
1291 * Reinitialize proc pointer `p' as it may be different
1292 * if this is a child returning from fork syscall.
1295 frame.tf_eax = p->p_retval[0];
1296 frame.tf_edx = p->p_retval[1];
1297 frame.tf_eflags &= ~PSL_C;
1302 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1303 * int 0x80 is 2 bytes. We saved this in tf_err.
1305 frame.tf_eip -= frame.tf_err;
1313 if (p->p_sysent->sv_errsize) {
1314 if (error >= p->p_sysent->sv_errsize)
1315 error = -1; /* XXX */
1317 error = p->p_sysent->sv_errtbl[error];
1319 frame.tf_eax = error;
1320 frame.tf_eflags |= PSL_C;
1325 * Traced syscall. trapsignal() is not MP aware.
1327 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1328 frame.tf_eflags &= ~PSL_T;
1329 trapsignal(p, SIGTRAP, 0);
1333 * Handle reschedule and other end-of-syscall issues
1335 userret(p, &frame, sticks);
1338 if (KTRPOINT(td, KTR_SYSRET)) {
1339 ktrsysret(p->p_tracep, code, error, p->p_retval[0]);
1344 * This works because errno is findable through the
1345 * register set. If we ever support an emulation where this
1346 * is not the case, this code will need to be revisited.
1348 STOPEVENT(p, S_SCX, code);
1353 * Release the MP lock if we had to get it
1355 KASSERT(curthread->td_mpcount == 1, ("badmpcount syscall from %p", (void *)frame.tf_eip));
1361 * Simplified back end of syscall(), used when returning from fork()
1362 * directly into user mode. MP lock is held on entry and should be
1363 * released on return. This code will return back into the fork
1364 * trampoline code which then runs doreti.
1367 fork_return(p, frame)
1369 struct trapframe frame;
1371 frame.tf_eax = 0; /* Child returns zero */
1372 frame.tf_eflags &= ~PSL_C; /* success */
1375 userret(p, &frame, 0);
1377 if (KTRPOINT(p->p_thread, KTR_SYSRET))
1378 ktrsysret(p->p_tracep, SYS_fork, 0, 0);
1380 p->p_flag |= P_PASSIVE_ACQ;
1382 p->p_flag &= ~P_PASSIVE_ACQ;
1384 KKASSERT(curthread->td_mpcount == 1);