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 1.23 2003/07/11 23:26:15 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 shift our
253 * P_CURPROC designation to another user process. We don't actually
254 * switch here because that would be a waste of cycles (the newly
255 * scheduled user process would just switch back to us since we are
256 * running at a kernel priority). Instead we fall through and will
257 * switch away when we attempt to reacquire our P_CURPROC designation.
259 if (resched_wanted()) {
260 if (curthread->td_release)
261 passive_release(curthread);
265 * Charge system time if profiling. Note: times are in microseconds.
267 if (p->p_flag & P_PROFIL) {
268 addupc_task(p, frame->tf_eip,
269 (u_int)(curthread->td_sticks - oticks));
273 * Post any pending signals XXX
275 while ((sig = CURSIG(p)) != 0)
279 #ifdef DEVICE_POLLING
280 extern u_int32_t poll_in_trap;
281 extern int ether_poll __P((int count));
282 #endif /* DEVICE_POLLING */
285 * Exception, fault, and trap interface to the FreeBSD kernel.
286 * This common code is called from assembly language IDT gate entry
287 * routines that prepare a suitable stack frame, and restore this
288 * frame after the exception has been processed.
290 * This function is also called from doreti in an interlock to handle ASTs.
291 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
293 * NOTE! We have to retrieve the fault address prior to obtaining the
294 * MP lock because get_mplock() may switch out. YYY cr2 really ought
295 * to be retrieved by the assembly code, not here.
299 struct trapframe frame;
301 struct proc *p = curproc;
303 int i = 0, ucode = 0, type, code;
308 eva = (frame.tf_trapno == T_PAGEFLT ? rcr2() : 0);
310 trap_fatal(&frame, eva);
316 if (frame.tf_trapno == T_PAGEFLT) {
318 * For some Cyrix CPUs, %cr2 is clobbered by interrupts.
319 * This problem is worked around by using an interrupt
320 * gate for the pagefault handler. We are finally ready
321 * to read %cr2 and then must reenable interrupts.
323 * XXX this should be in the switch statement, but the
324 * NO_FOOF_HACK and VM86 goto and ifdefs obfuscate the
325 * flow of control too much for this to be obviously
335 * MP lock is held at this point
338 if (!(frame.tf_eflags & PSL_I)) {
340 * Buggy application or kernel code has disabled interrupts
341 * and then trapped. Enabling interrupts now is wrong, but
342 * it is better than running with interrupts disabled until
343 * they are accidentally enabled later.
345 type = frame.tf_trapno;
346 if (ISPL(frame.tf_cs)==SEL_UPL || (frame.tf_eflags & PSL_VM)) {
348 "pid %ld (%s): trap %d with interrupts disabled\n",
349 (long)curproc->p_pid, curproc->p_comm, type);
350 } else if (type != T_BPTFLT && type != T_TRCTRAP) {
352 * XXX not quite right, since this may be for a
353 * multiple fault in user mode.
355 printf("kernel trap %d with interrupts disabled\n",
362 #ifdef DEVICE_POLLING
364 ether_poll(poll_in_trap);
365 #endif /* DEVICE_POLLING */
367 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
370 type = frame.tf_trapno;
374 if (frame.tf_eflags & PSL_VM &&
375 (type == T_PROTFLT || type == T_STKFLT)) {
377 KKASSERT(curthread->td_mpcount > 0);
379 i = vm86_emulate((struct vm86frame *)&frame);
381 KKASSERT(curthread->td_mpcount > 0);
385 * returns to original process
387 vm86_trap((struct vm86frame *)&frame);
394 * these traps want either a process context, or
395 * assume a normal userspace trap.
399 trap_fatal(&frame, eva);
402 type = T_BPTFLT; /* kernel breakpoint */
405 goto kernel_trap; /* normal kernel trap handling */
408 if ((ISPL(frame.tf_cs) == SEL_UPL) || (frame.tf_eflags & PSL_VM)) {
413 sticks = curthread->td_sticks;
414 p->p_md.md_regs = &frame;
417 case T_PRIVINFLT: /* privileged instruction fault */
422 case T_BPTFLT: /* bpt instruction fault */
423 case T_TRCTRAP: /* trace trap */
424 frame.tf_eflags &= ~PSL_T;
428 case T_ARITHTRAP: /* arithmetic trap */
433 case T_ASTFLT: /* Allow process switch */
435 mycpu->gd_cnt.v_soft++;
436 if (p->p_flag & P_OWEUPC) {
437 p->p_flag &= ~P_OWEUPC;
438 addupc_task(p, p->p_stats->p_prof.pr_addr,
439 p->p_stats->p_prof.pr_ticks);
444 * The following two traps can happen in
445 * vm86 mode, and, if so, we want to handle
448 case T_PROTFLT: /* general protection fault */
449 case T_STKFLT: /* stack fault */
450 if (frame.tf_eflags & PSL_VM) {
451 i = vm86_emulate((struct vm86frame *)&frame);
458 case T_SEGNPFLT: /* segment not present fault */
459 case T_TSSFLT: /* invalid TSS fault */
460 case T_DOUBLEFLT: /* double fault */
462 ucode = code + BUS_SEGM_FAULT ;
466 case T_PAGEFLT: /* page fault */
467 i = trap_pfault(&frame, TRUE, eva);
470 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
480 case T_DIVIDE: /* integer divide fault */
488 goto handle_powerfail;
489 #else /* !POWERFAIL_NMI */
490 /* machine/parity/power fail/"kitchen sink" faults */
491 if (isa_nmi(code) == 0) {
494 * NMI can be hooked up to a pushbutton
498 printf ("NMI ... going to debugger\n");
499 kdb_trap (type, 0, &frame);
503 } else if (panic_on_nmi)
504 panic("NMI indicates hardware failure");
506 #endif /* POWERFAIL_NMI */
507 #endif /* NISA > 0 */
509 case T_OFLOW: /* integer overflow fault */
514 case T_BOUND: /* bounds check fault */
521 /* if a transparent fault (due to context switch "late") */
525 if (!pmath_emulate) {
527 ucode = FPE_FPU_NP_TRAP;
530 i = (*pmath_emulate)(&frame);
532 if (!(frame.tf_eflags & PSL_T))
534 frame.tf_eflags &= ~PSL_T;
537 /* else ucode = emulator_only_knows() XXX */
540 case T_FPOPFLT: /* FPU operand fetch fault */
545 case T_XMMFLT: /* SIMD floating-point exception */
555 case T_PAGEFLT: /* page fault */
556 (void) trap_pfault(&frame, FALSE, eva);
562 * The kernel is apparently using npx for copying.
563 * XXX this should be fatal unless the kernel has
564 * registered such use.
571 case T_PROTFLT: /* general protection fault */
572 case T_SEGNPFLT: /* segment not present fault */
574 * Invalid segment selectors and out of bounds
575 * %eip's and %esp's can be set up in user mode.
576 * This causes a fault in kernel mode when the
577 * kernel tries to return to user mode. We want
578 * to get this fault so that we can fix the
579 * problem here and not have to check all the
580 * selectors and pointers when the user changes
583 #define MAYBE_DORETI_FAULT(where, whereto) \
585 if (frame.tf_eip == (int)where) { \
586 frame.tf_eip = (int)whereto; \
591 if (mycpu->gd_intr_nesting_level == 0) {
593 * Invalid %fs's and %gs's can be created using
594 * procfs or PT_SETREGS or by invalidating the
595 * underlying LDT entry. This causes a fault
596 * in kernel mode when the kernel attempts to
597 * switch contexts. Lose the bad context
598 * (XXX) so that we can continue, and generate
601 if (frame.tf_eip == (int)cpu_switch_load_gs) {
602 curthread->td_pcb->pcb_gs = 0;
606 MAYBE_DORETI_FAULT(doreti_iret,
608 MAYBE_DORETI_FAULT(doreti_popl_ds,
609 doreti_popl_ds_fault);
610 MAYBE_DORETI_FAULT(doreti_popl_es,
611 doreti_popl_es_fault);
612 MAYBE_DORETI_FAULT(doreti_popl_fs,
613 doreti_popl_fs_fault);
614 if (curthread->td_pcb->pcb_onfault) {
615 frame.tf_eip = (int)curthread->td_pcb->pcb_onfault;
623 * PSL_NT can be set in user mode and isn't cleared
624 * automatically when the kernel is entered. This
625 * causes a TSS fault when the kernel attempts to
626 * `iret' because the TSS link is uninitialized. We
627 * want to get this fault so that we can fix the
628 * problem here and not every time the kernel is
631 if (frame.tf_eflags & PSL_NT) {
632 frame.tf_eflags &= ~PSL_NT;
637 case T_TRCTRAP: /* trace trap */
638 if (frame.tf_eip == (int)IDTVEC(syscall)) {
640 * We've just entered system mode via the
641 * syscall lcall. Continue single stepping
642 * silently until the syscall handler has
647 if (frame.tf_eip == (int)IDTVEC(syscall) + 1) {
649 * The syscall handler has now saved the
650 * flags. Stop single stepping it.
652 frame.tf_eflags &= ~PSL_T;
656 * Ignore debug register trace traps due to
657 * accesses in the user's address space, which
658 * can happen under several conditions such as
659 * if a user sets a watchpoint on a buffer and
660 * then passes that buffer to a system call.
661 * We still want to get TRCTRAPS for addresses
662 * in kernel space because that is useful when
663 * debugging the kernel.
665 if (user_dbreg_trap()) {
667 * Reset breakpoint bits because the
670 load_dr6(rdr6() & 0xfffffff0);
674 * Fall through (TRCTRAP kernel mode, kernel address)
678 * If DDB is enabled, let it handle the debugger trap.
679 * Otherwise, debugger traps "can't happen".
682 if (kdb_trap (type, 0, &frame))
691 # define TIMER_FREQ 1193182
695 static unsigned lastalert = 0;
697 if(time_second - lastalert > 10)
699 log(LOG_WARNING, "NMI: power fail\n");
700 sysbeep(TIMER_FREQ/880, hz);
701 lastalert = time_second;
706 #else /* !POWERFAIL_NMI */
707 /* machine/parity/power fail/"kitchen sink" faults */
708 if (isa_nmi(code) == 0) {
711 * NMI can be hooked up to a pushbutton
715 printf ("NMI ... going to debugger\n");
716 kdb_trap (type, 0, &frame);
720 } else if (panic_on_nmi == 0)
723 #endif /* POWERFAIL_NMI */
724 #endif /* NISA > 0 */
727 trap_fatal(&frame, eva);
731 /* Translate fault for emulators (e.g. Linux) */
732 if (*p->p_sysent->sv_transtrap)
733 i = (*p->p_sysent->sv_transtrap)(i, type);
735 trapsignal(p, i, ucode);
738 if (type <= MAX_TRAP_MSG) {
739 uprintf("fatal process exception: %s",
741 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
742 uprintf(", fault VA = 0x%lx", (u_long)eva);
749 if (ISPL(frame.tf_cs) == SEL_UPL)
750 KASSERT(curthread->td_mpcount == 1, ("badmpcount trap from %p", (void *)frame.tf_eip));
752 userret(p, &frame, sticks);
756 KKASSERT(curthread->td_mpcount > 0);
763 * This version doesn't allow a page fault to user space while
764 * in the kernel. The rest of the kernel needs to be made "safe"
765 * before this can be used. I think the only things remaining
766 * to be made safe are the iBCS2 code and the process tracing/
770 trap_pfault(frame, usermode, eva)
771 struct trapframe *frame;
776 struct vmspace *vm = NULL;
780 struct proc *p = curproc;
782 if (frame->tf_err & PGEX_W)
783 ftype = VM_PROT_WRITE;
785 ftype = VM_PROT_READ;
787 va = trunc_page(eva);
788 if (va < VM_MIN_KERNEL_ADDRESS) {
793 (!usermode && va < VM_MAXUSER_ADDRESS &&
794 (mycpu->gd_intr_nesting_level != 0 ||
795 curthread->td_pcb->pcb_onfault == NULL))) {
796 trap_fatal(frame, eva);
801 * This is a fault on non-kernel virtual memory.
802 * vm is initialized above to NULL. If curproc is NULL
803 * or curproc->p_vmspace is NULL the fault is fatal.
812 * Keep swapout from messing with us during this
818 * Grow the stack if necessary
820 /* grow_stack returns false only if va falls into
821 * a growable stack region and the stack growth
822 * fails. It returns true if va was not within
823 * a growable stack region, or if the stack
826 if (!grow_stack (p, va)) {
832 /* Fault in the user page: */
833 rv = vm_fault(map, va, ftype,
834 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
840 * Don't allow user-mode faults in kernel address space.
846 * Since we know that kernel virtual address addresses
847 * always have pte pages mapped, we just have to fault
850 rv = vm_fault(kernel_map, va, ftype, VM_FAULT_NORMAL);
853 if (rv == KERN_SUCCESS)
857 if (mycpu->gd_intr_nesting_level == 0 && curthread->td_pcb->pcb_onfault) {
858 frame->tf_eip = (int)curthread->td_pcb->pcb_onfault;
861 trap_fatal(frame, eva);
865 /* kludge to pass faulting virtual address to sendsig */
868 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
873 trap_pfault(frame, usermode, eva)
874 struct trapframe *frame;
879 struct vmspace *vm = NULL;
883 struct proc *p = curproc;
885 va = trunc_page(eva);
886 if (va >= KERNBASE) {
888 * Don't allow user-mode faults in kernel address space.
889 * An exception: if the faulting address is the invalid
890 * instruction entry in the IDT, then the Intel Pentium
891 * F00F bug workaround was triggered, and we need to
892 * treat it is as an illegal instruction, and not a page
895 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
896 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) {
897 frame->tf_trapno = T_PRIVINFLT;
907 * This is a fault on non-kernel virtual memory.
908 * vm is initialized above to NULL. If curproc is NULL
909 * or curproc->p_vmspace is NULL the fault is fatal.
920 if (frame->tf_err & PGEX_W)
921 ftype = VM_PROT_WRITE;
923 ftype = VM_PROT_READ;
925 if (map != kernel_map) {
927 * Keep swapout from messing with us during this
933 * Grow the stack if necessary
935 /* grow_stack returns false only if va falls into
936 * a growable stack region and the stack growth
937 * fails. It returns true if va was not within
938 * a growable stack region, or if the stack
941 if (!grow_stack (p, va)) {
947 /* Fault in the user page: */
948 rv = vm_fault(map, va, ftype,
949 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
955 * Don't have to worry about process locking or stacks in the kernel.
957 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
960 if (rv == KERN_SUCCESS)
964 if (mycpu->gd_intr_nesting_level == 0 && curthread->td_pcb->pcb_onfault) {
965 frame->tf_eip = (int)curthread->td_pcb->pcb_onfault;
968 trap_fatal(frame, eva);
972 /* kludge to pass faulting virtual address to sendsig */
975 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
979 trap_fatal(frame, eva)
980 struct trapframe *frame;
983 int code, type, ss, esp;
984 struct soft_segment_descriptor softseg;
986 code = frame->tf_err;
987 type = frame->tf_trapno;
988 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
990 if (type <= MAX_TRAP_MSG)
991 printf("\n\nFatal trap %d: %s while in %s mode\n",
992 type, trap_msg[type],
993 frame->tf_eflags & PSL_VM ? "vm86" :
994 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
996 /* three seperate prints in case of a trap on an unmapped page */
997 printf("mp_lock = %08x; ", mp_lock);
998 printf("cpuid = %d; ", mycpu->gd_cpuid);
999 printf("lapic.id = %08x\n", lapic.id);
1001 if (type == T_PAGEFLT) {
1002 printf("fault virtual address = 0x%x\n", eva);
1003 printf("fault code = %s %s, %s\n",
1004 code & PGEX_U ? "user" : "supervisor",
1005 code & PGEX_W ? "write" : "read",
1006 code & PGEX_P ? "protection violation" : "page not present");
1008 printf("instruction pointer = 0x%x:0x%x\n",
1009 frame->tf_cs & 0xffff, frame->tf_eip);
1010 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
1011 ss = frame->tf_ss & 0xffff;
1012 esp = frame->tf_esp;
1014 ss = GSEL(GDATA_SEL, SEL_KPL);
1015 esp = (int)&frame->tf_esp;
1017 printf("stack pointer = 0x%x:0x%x\n", ss, esp);
1018 printf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
1019 printf("code segment = base 0x%x, limit 0x%x, type 0x%x\n",
1020 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
1021 printf(" = DPL %d, pres %d, def32 %d, gran %d\n",
1022 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
1024 printf("processor eflags = ");
1025 if (frame->tf_eflags & PSL_T)
1026 printf("trace trap, ");
1027 if (frame->tf_eflags & PSL_I)
1028 printf("interrupt enabled, ");
1029 if (frame->tf_eflags & PSL_NT)
1030 printf("nested task, ");
1031 if (frame->tf_eflags & PSL_RF)
1033 if (frame->tf_eflags & PSL_VM)
1035 printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
1036 printf("current process = ");
1038 printf("%lu (%s)\n",
1039 (u_long)curproc->p_pid, curproc->p_comm ?
1040 curproc->p_comm : "");
1044 printf("current thread = pri %d ", curthread->td_pri);
1045 if (curthread->td_pri >= TDPRI_CRIT)
1048 printf("interrupt mask = ");
1049 if ((curthread->td_cpl & net_imask) == net_imask)
1051 if ((curthread->td_cpl & tty_imask) == tty_imask)
1053 if ((curthread->td_cpl & bio_imask) == bio_imask)
1055 if ((curthread->td_cpl & cam_imask) == cam_imask)
1057 if (curthread->td_cpl == 0)
1062 * we probably SHOULD have stopped the other CPUs before now!
1063 * another CPU COULD have been touching cpl at this moment...
1065 printf(" <- SMP: XXX");
1074 if ((debugger_on_panic || db_active) && kdb_trap(type, 0, frame))
1077 printf("trap number = %d\n", type);
1078 if (type <= MAX_TRAP_MSG)
1079 panic("%s", trap_msg[type]);
1081 panic("unknown/reserved trap");
1085 * Double fault handler. Called when a fault occurs while writing
1086 * a frame for a trap/exception onto the stack. This usually occurs
1087 * when the stack overflows (such is the case with infinite recursion,
1090 * XXX Note that the current PTD gets replaced by IdlePTD when the
1091 * task switch occurs. This means that the stack that was active at
1092 * the time of the double fault is not available at <kstack> unless
1093 * the machine was idle when the double fault occurred. The downside
1094 * of this is that "trace <ebp>" in ddb won't work.
1099 struct mdglobaldata *gd = mdcpu;
1101 printf("\nFatal double fault:\n");
1102 printf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1103 printf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1104 printf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
1106 /* three seperate prints in case of a trap on an unmapped page */
1107 printf("mp_lock = %08x; ", mp_lock);
1108 printf("cpuid = %d; ", mycpu->gd_cpuid);
1109 printf("lapic.id = %08x\n", lapic.id);
1111 panic("double fault");
1115 * Compensate for 386 brain damage (missing URKR).
1116 * This is a little simpler than the pagefault handler in trap() because
1117 * it the page tables have already been faulted in and high addresses
1118 * are thrown out early for other reasons.
1128 va = trunc_page((vm_offset_t)addr);
1130 * XXX - MAX is END. Changed > to >= for temp. fix.
1132 if (va >= VM_MAXUSER_ADDRESS)
1140 if (!grow_stack (p, va)) {
1146 * fault the data page
1148 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY);
1152 if (rv != KERN_SUCCESS)
1159 * syscall2 - MP aware system call request C handler
1161 * A system call is essentially treated as a trap except that the
1162 * MP lock is not held on entry or return. We are responsible for
1163 * obtaining the MP lock if necessary and for handling ASTs
1164 * (e.g. a task switch) prior to return.
1166 * In general, only simple access and manipulation of curproc and
1167 * the current stack is allowed without having to hold MP lock.
1171 struct trapframe frame;
1173 struct thread *td = curthread;
1174 struct proc *p = td->td_proc;
1177 struct sysent *callp;
1178 register_t orig_tf_eflags;
1186 if (ISPL(frame.tf_cs) != SEL_UPL) {
1194 KASSERT(curthread->td_mpcount == 0, ("badmpcount syscall from %p", (void *)frame.tf_eip));
1198 * access non-atomic field from critical section. p_sticks is
1199 * updated by the clock interrupt. Also use this opportunity
1200 * to lazy-raise our LWKT priority.
1204 sticks = curthread->td_sticks;
1207 p->p_md.md_regs = &frame;
1208 params = (caddr_t)frame.tf_esp + sizeof(int);
1209 code = frame.tf_eax;
1210 orig_tf_eflags = frame.tf_eflags;
1212 if (p->p_sysent->sv_prepsyscall) {
1214 * The prep code is not MP aware.
1216 (*p->p_sysent->sv_prepsyscall)(&frame, args, &code, ¶ms);
1219 * Need to check if this is a 32 bit or 64 bit syscall.
1220 * fuword is MP aware.
1222 if (code == SYS_syscall) {
1224 * Code is first argument, followed by actual args.
1226 code = fuword(params);
1227 params += sizeof(int);
1228 } else if (code == SYS___syscall) {
1230 * Like syscall, but code is a quad, so as to maintain
1231 * quad alignment for the rest of the arguments.
1233 code = fuword(params);
1234 params += sizeof(quad_t);
1238 if (p->p_sysent->sv_mask)
1239 code &= p->p_sysent->sv_mask;
1241 if (code >= p->p_sysent->sv_size)
1242 callp = &p->p_sysent->sv_table[0];
1244 callp = &p->p_sysent->sv_table[code];
1246 narg = callp->sy_narg & SYF_ARGMASK;
1249 * copyin is MP aware, but the tracing code is not
1251 if (params && (i = narg * sizeof(int)) &&
1252 (error = copyin(params, (caddr_t)args, (u_int)i))) {
1254 if (KTRPOINT(td, KTR_SYSCALL))
1255 ktrsyscall(p->p_tracep, code, narg, args);
1262 * Try to run the syscall without the MP lock if the syscall
1263 * is MP safe. We have to obtain the MP lock no matter what if
1266 if ((callp->sy_narg & SYF_MPSAFE) == 0) {
1273 if (KTRPOINT(td, KTR_SYSCALL)) {
1274 ktrsyscall(p->p_tracep, code, narg, args);
1278 p->p_retval[1] = frame.tf_edx;
1280 STOPEVENT(p, S_SCE, narg); /* MP aware */
1282 error = (*callp->sy_call)(args);
1285 * MP SAFE (we may or may not have the MP lock at this point)
1290 * Reinitialize proc pointer `p' as it may be different
1291 * if this is a child returning from fork syscall.
1294 frame.tf_eax = p->p_retval[0];
1295 frame.tf_edx = p->p_retval[1];
1296 frame.tf_eflags &= ~PSL_C;
1301 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1302 * int 0x80 is 2 bytes. We saved this in tf_err.
1304 frame.tf_eip -= frame.tf_err;
1312 if (p->p_sysent->sv_errsize) {
1313 if (error >= p->p_sysent->sv_errsize)
1314 error = -1; /* XXX */
1316 error = p->p_sysent->sv_errtbl[error];
1318 frame.tf_eax = error;
1319 frame.tf_eflags |= PSL_C;
1324 * Traced syscall. trapsignal() is not MP aware.
1326 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1327 frame.tf_eflags &= ~PSL_T;
1328 trapsignal(p, SIGTRAP, 0);
1332 * Handle reschedule and other end-of-syscall issues
1334 userret(p, &frame, sticks);
1337 if (KTRPOINT(td, KTR_SYSRET)) {
1338 ktrsysret(p->p_tracep, code, error, p->p_retval[0]);
1343 * This works because errno is findable through the
1344 * register set. If we ever support an emulation where this
1345 * is not the case, this code will need to be revisited.
1347 STOPEVENT(p, S_SCX, code);
1352 * Release the MP lock if we had to get it
1354 KASSERT(curthread->td_mpcount == 1, ("badmpcount syscall from %p", (void *)frame.tf_eip));
1360 * Simplified back end of syscall(), used when returning from fork()
1361 * directly into user mode. MP lock is held on entry and should be
1362 * released on return. This code will return back into the fork
1363 * trampoline code which then runs doreti.
1366 fork_return(p, frame)
1368 struct trapframe frame;
1370 frame.tf_eax = 0; /* Child returns zero */
1371 frame.tf_eflags &= ~PSL_C; /* success */
1374 userret(p, &frame, 0);
1376 if (KTRPOINT(p->p_thread, KTR_SYSRET))
1377 ktrsysret(p->p_tracep, SYS_fork, 0, 0);
1379 p->p_flag |= P_PASSIVE_ACQ;
1381 p->p_flag &= ~P_PASSIVE_ACQ;
1383 KKASSERT(curthread->td_mpcount == 1);