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 $
42 * 386 Trap and System call handling
50 #include "opt_ktrace.h"
51 #include "opt_clock.h"
54 #include <sys/param.h>
55 #include <sys/systm.h>
57 #include <sys/pioctl.h>
58 #include <sys/kernel.h>
59 #include <sys/kerneldump.h>
60 #include <sys/resourcevar.h>
61 #include <sys/signalvar.h>
62 #include <sys/signal2.h>
63 #include <sys/syscall.h>
64 #include <sys/sysctl.h>
65 #include <sys/sysent.h>
67 #include <sys/vmmeter.h>
68 #include <sys/malloc.h>
70 #include <sys/ktrace.h>
73 #include <sys/vkernel.h>
74 #include <sys/sysproto.h>
75 #include <sys/sysunion.h>
78 #include <vm/vm_param.h>
81 #include <vm/vm_kern.h>
82 #include <vm/vm_map.h>
83 #include <vm/vm_page.h>
84 #include <vm/vm_extern.h>
86 #include <machine/cpu.h>
87 #include <machine/md_var.h>
88 #include <machine/pcb.h>
89 #include <machine/smp.h>
90 #include <machine/tss.h>
91 #include <machine/specialreg.h>
92 #include <machine/globaldata.h>
93 #include <machine/intr_machdep.h>
95 #include <machine_base/isa/isa_intr.h>
96 #include <machine_base/apic/lapic.h>
99 #include <sys/syslog.h>
100 #include <machine/clock.h>
103 #include <machine/vm86.h>
107 #include <sys/msgport2.h>
108 #include <sys/thread2.h>
109 #include <sys/mplock2.h>
111 #define MAKEMPSAFE(have_mplock) \
112 if (have_mplock == 0) { \
117 int (*pmath_emulate) (struct trapframe *);
119 extern void trap (struct trapframe *frame);
120 extern void syscall2 (struct trapframe *frame);
122 static int trap_pfault (struct trapframe *, int, vm_offset_t);
123 static void trap_fatal (struct trapframe *, vm_offset_t);
124 void dblfault_handler (void);
126 extern inthand_t IDTVEC(syscall);
128 #define MAX_TRAP_MSG 28
129 static char *trap_msg[] = {
131 "privileged instruction fault", /* 1 T_PRIVINFLT */
133 "breakpoint instruction fault", /* 3 T_BPTFLT */
136 "arithmetic trap", /* 6 T_ARITHTRAP */
137 "system forced exception", /* 7 T_ASTFLT */
139 "general protection fault", /* 9 T_PROTFLT */
140 "trace trap", /* 10 T_TRCTRAP */
142 "page fault", /* 12 T_PAGEFLT */
144 "alignment fault", /* 14 T_ALIGNFLT */
148 "integer divide fault", /* 18 T_DIVIDE */
149 "non-maskable interrupt trap", /* 19 T_NMI */
150 "overflow trap", /* 20 T_OFLOW */
151 "FPU bounds check fault", /* 21 T_BOUND */
152 "FPU device not available", /* 22 T_DNA */
153 "double fault", /* 23 T_DOUBLEFLT */
154 "FPU operand fetch fault", /* 24 T_FPOPFLT */
155 "invalid TSS fault", /* 25 T_TSSFLT */
156 "segment not present fault", /* 26 T_SEGNPFLT */
157 "stack fault", /* 27 T_STKFLT */
158 "machine check trap", /* 28 T_MCHK */
161 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
162 extern int has_f00f_bug;
166 static int ddb_on_nmi = 1;
167 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
168 &ddb_on_nmi, 0, "Go to DDB on NMI");
170 static int panic_on_nmi = 1;
171 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
172 &panic_on_nmi, 0, "Panic on NMI");
173 static int fast_release;
174 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
175 &fast_release, 0, "Passive Release was optimal");
176 static int slow_release;
177 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
178 &slow_release, 0, "Passive Release was nonoptimal");
181 * Passively intercepts the thread switch function to increase the thread
182 * priority from a user priority to a kernel priority, reducing
183 * syscall and trap overhead for the case where no switch occurs.
185 * Synchronizes td_ucred with p_ucred. This is used by system calls,
186 * signal handling, faults, AST traps, and anything else that enters the
187 * kernel from userland and provides the kernel with a stable read-only
188 * copy of the process ucred.
191 userenter(struct thread *curtd, struct proc *curp)
196 curtd->td_release = lwkt_passive_release;
198 if (curtd->td_ucred != curp->p_ucred) {
199 ncred = crhold(curp->p_ucred);
200 ocred = curtd->td_ucred;
201 curtd->td_ucred = ncred;
209 * Handle signals, profiling, and other AST's and/or tasks that
210 * must be completed before we can return to or try to return to userland.
212 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
213 * arithmatic on the delta calculation so the absolute tick values are
214 * truncated to an integer.
217 userret(struct lwp *lp, struct trapframe *frame, int sticks)
219 struct proc *p = lp->lwp_proc;
223 if (p->p_userret != NULL) {
230 * Charge system time if profiling. Note: times are in microseconds.
231 * This may do a copyout and block, so do it first even though it
232 * means some system time will be charged as user time.
234 if (p->p_flags & P_PROFIL) {
235 addupc_task(p, frame->tf_eip,
236 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
241 * Specific on-return-to-usermode checks (LWP_MP_WEXIT,
242 * LWP_MP_VNLRU, etc).
244 if (lp->lwp_mpflags & LWP_MP_URETMASK)
248 * Block here if we are in a stopped state.
250 if (p->p_stat == SSTOP || dump_stop_usertds) {
251 lwkt_gettoken(&p->p_token);
253 lwkt_reltoken(&p->p_token);
258 * Post any pending upcalls. If running a virtual kernel be sure
259 * to restore the virtual kernel's vmspace before posting the upcall.
261 if (p->p_flags & (P_SIGVTALRM | P_SIGPROF)) {
262 lwkt_gettoken(&p->p_token);
263 if (p->p_flags & P_SIGVTALRM) {
264 p->p_flags &= ~P_SIGVTALRM;
265 ksignal(p, SIGVTALRM);
267 if (p->p_flags & P_SIGPROF) {
268 p->p_flags &= ~P_SIGPROF;
271 lwkt_reltoken(&p->p_token);
276 * Post any pending signals. If running a virtual kernel be sure
277 * to restore the virtual kernel's vmspace before posting the signal.
279 * WARNING! postsig() can exit and not return.
281 if ((sig = CURSIG_TRACE(lp)) != 0) {
282 lwkt_gettoken(&p->p_token);
284 lwkt_reltoken(&p->p_token);
289 * block here if we are swapped out, but still process signals
290 * (such as SIGKILL). proc0 (the swapin scheduler) is already
291 * aware of our situation, we do not have to wake it up.
293 if (p->p_flags & P_SWAPPEDOUT) {
294 lwkt_gettoken(&p->p_token);
296 p->p_flags |= P_SWAPWAIT;
298 if (p->p_flags & P_SWAPWAIT)
299 tsleep(p, PCATCH, "SWOUT", 0);
300 p->p_flags &= ~P_SWAPWAIT;
302 lwkt_reltoken(&p->p_token);
307 * In a multi-threaded program it is possible for a thread to change
308 * signal state during a system call which temporarily changes the
309 * signal mask. In this case postsig() might not be run and we
310 * have to restore the mask ourselves.
312 if (lp->lwp_flags & LWP_OLDMASK) {
313 lp->lwp_flags &= ~LWP_OLDMASK;
314 lp->lwp_sigmask = lp->lwp_oldsigmask;
320 * Cleanup from userenter and any passive release that might have occured.
321 * We must reclaim the current-process designation before we can return
322 * to usermode. We also handle both LWKT and USER reschedule requests.
325 userexit(struct lwp *lp)
327 struct thread *td = lp->lwp_thread;
328 /* globaldata_t gd = td->td_gd; */
331 * Handle stop requests at kernel priority. Any requests queued
332 * after this loop will generate another AST.
334 while (lp->lwp_proc->p_stat == SSTOP) {
335 lwkt_gettoken(&lp->lwp_proc->p_token);
337 lwkt_reltoken(&lp->lwp_proc->p_token);
341 * Become the current user scheduled process if we aren't already,
342 * and deal with reschedule requests and other factors.
344 lp->lwp_proc->p_usched->acquire_curproc(lp);
345 /* WARNING: we may have migrated cpu's */
346 /* gd = td->td_gd; */
349 * Reduce our priority in preparation for a return to userland. If
350 * our passive release function was still in place, our priority was
351 * never raised and does not need to be reduced.
353 lwkt_passive_recover(td);
356 #if !defined(KTR_KERNENTRY)
357 #define KTR_KERNENTRY KTR_ALL
359 KTR_INFO_MASTER(kernentry);
360 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0,
361 "TRAP(pid %d, tid %d, trapno %d, eva %lu)",
362 pid_t pid, lwpid_t tid, register_t trapno, vm_offset_t eva);
363 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "TRAP_RET(pid %d, tid %d)",
364 pid_t pid, lwpid_t tid);
365 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "SYSC(pid %d, tid %d, nr %d)",
366 pid_t pid, lwpid_t tid, register_t trapno);
367 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "SYSRET(pid %d, tid %d, err %d)",
368 pid_t pid, lwpid_t tid, int err);
369 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "FORKRET(pid %d, tid %d)",
370 pid_t pid, lwpid_t tid);
373 * Exception, fault, and trap interface to the kernel.
374 * This common code is called from assembly language IDT gate entry
375 * routines that prepare a suitable stack frame, and restore this
376 * frame after the exception has been processed.
378 * This function is also called from doreti in an interlock to handle ASTs.
379 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
381 * NOTE! We have to retrieve the fault address prior to obtaining the
382 * MP lock because get_mplock() may switch out. YYY cr2 really ought
383 * to be retrieved by the assembly code, not here.
385 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
386 * if an attempt is made to switch from a fast interrupt or IPI. This is
387 * necessary to properly take fatal kernel traps on SMP machines if
388 * get_mplock() has to block.
392 trap(struct trapframe *frame)
394 struct globaldata *gd = mycpu;
395 struct thread *td = gd->gd_curthread;
396 struct lwp *lp = td->td_lwp;
399 int i = 0, ucode = 0, type, code;
402 int crit_count = td->td_critcount;
403 lwkt_tokref_t curstop = td->td_toks_stop;
410 * We need to allow T_DNA faults when the debugger is active since
411 * some dumping paths do large bcopy() which use the floating
412 * point registers for faster copying.
414 if (db_active && frame->tf_trapno != T_DNA) {
415 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
416 ++gd->gd_trap_nesting_level;
417 MAKEMPSAFE(have_mplock);
418 trap_fatal(frame, eva);
419 --gd->gd_trap_nesting_level;
425 ++gd->gd_trap_nesting_level;
426 if (frame->tf_trapno == T_PAGEFLT) {
428 * For some Cyrix CPUs, %cr2 is clobbered by interrupts.
429 * This problem is worked around by using an interrupt
430 * gate for the pagefault handler. We are finally ready
431 * to read %cr2 and then must reenable interrupts.
433 * XXX this should be in the switch statement, but the
434 * NO_FOOF_HACK and VM86 goto and ifdefs obfuscate the
435 * flow of control too much for this to be obviously
442 --gd->gd_trap_nesting_level;
444 if (!(frame->tf_eflags & PSL_I)) {
446 * Buggy application or kernel code has disabled interrupts
447 * and then trapped. Enabling interrupts now is wrong, but
448 * it is better than running with interrupts disabled until
449 * they are accidentally enabled later.
451 type = frame->tf_trapno;
452 if (ISPL(frame->tf_cs)==SEL_UPL || (frame->tf_eflags & PSL_VM)) {
453 MAKEMPSAFE(have_mplock);
455 "pid %ld (%s): trap %d with interrupts disabled\n",
456 (long)curproc->p_pid, curproc->p_comm, type);
457 } else if (type != T_BPTFLT && type != T_TRCTRAP) {
459 * XXX not quite right, since this may be for a
460 * multiple fault in user mode.
462 MAKEMPSAFE(have_mplock);
463 kprintf("kernel trap %d with interrupts disabled\n",
469 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
472 type = frame->tf_trapno;
473 code = frame->tf_err;
476 if (frame->tf_eflags & PSL_VM &&
477 (type == T_PROTFLT || type == T_STKFLT)) {
478 KKASSERT(get_mplock_count(curthread) > 0);
479 i = vm86_emulate((struct vm86frame *)frame);
480 KKASSERT(get_mplock_count(curthread) > 0);
483 * returns to original process
485 vm86_trap((struct vm86frame *)frame,
487 KKASSERT(0); /* NOT REACHED */
493 * these traps want either a process context, or
494 * assume a normal userspace trap.
498 trap_fatal(frame, eva);
501 type = T_BPTFLT; /* kernel breakpoint */
504 goto kernel_trap; /* normal kernel trap handling */
507 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
510 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid,
511 frame->tf_trapno, eva);
515 sticks = (int)td->td_sticks;
516 lp->lwp_md.md_regs = frame;
519 case T_PRIVINFLT: /* privileged instruction fault */
524 case T_BPTFLT: /* bpt instruction fault */
525 case T_TRCTRAP: /* trace trap */
526 frame->tf_eflags &= ~PSL_T;
528 ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
531 case T_ARITHTRAP: /* arithmetic trap */
536 case T_ASTFLT: /* Allow process switch */
537 mycpu->gd_cnt.v_soft++;
538 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
539 atomic_clear_int(&mycpu->gd_reqflags,
541 addupc_task(p, p->p_prof.pr_addr,
547 * The following two traps can happen in
548 * vm86 mode, and, if so, we want to handle
551 case T_PROTFLT: /* general protection fault */
552 case T_STKFLT: /* stack fault */
553 if (frame->tf_eflags & PSL_VM) {
554 i = vm86_emulate((struct vm86frame *)frame);
560 ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR;
562 case T_SEGNPFLT: /* segment not present fault */
566 case T_TSSFLT: /* invalid TSS fault */
567 case T_DOUBLEFLT: /* double fault */
573 case T_PAGEFLT: /* page fault */
574 i = trap_pfault(frame, TRUE, eva);
577 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
592 case T_DIVIDE: /* integer divide fault */
599 MAKEMPSAFE(have_mplock);
601 goto handle_powerfail;
602 #else /* !POWERFAIL_NMI */
603 /* machine/parity/power fail/"kitchen sink" faults */
604 if (isa_nmi(code) == 0) {
607 * NMI can be hooked up to a pushbutton
611 kprintf ("NMI ... going to debugger\n");
612 kdb_trap (type, 0, frame);
616 } else if (panic_on_nmi)
617 panic("NMI indicates hardware failure");
619 #endif /* POWERFAIL_NMI */
620 #endif /* NISA > 0 */
622 case T_OFLOW: /* integer overflow fault */
627 case T_BOUND: /* bounds check fault */
634 * Virtual kernel intercept - pass the DNA exception
635 * to the virtual kernel if it asked to handle it.
636 * This occurs when the virtual kernel is holding
637 * onto the FP context for a different emulated
638 * process then the one currently running.
640 * We must still call npxdna() since we may have
641 * saved FP state that the virtual kernel needs
642 * to hand over to a different emulated process.
644 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
645 (td->td_pcb->pcb_flags & FP_VIRTFP)
653 * The kernel may have switched out the FP unit's
654 * state, causing the user process to take a fault
655 * when it tries to use the FP unit. Restore the
661 if (!pmath_emulate) {
663 ucode = FPE_FPU_NP_TRAP;
666 i = (*pmath_emulate)(frame);
668 if (!(frame->tf_eflags & PSL_T))
670 frame->tf_eflags &= ~PSL_T;
673 /* else ucode = emulator_only_knows() XXX */
676 case T_FPOPFLT: /* FPU operand fetch fault */
681 case T_XMMFLT: /* SIMD floating-point exception */
691 case T_PAGEFLT: /* page fault */
692 trap_pfault(frame, FALSE, eva);
698 * The kernel may be using npx for copying or other
706 case T_PROTFLT: /* general protection fault */
707 case T_SEGNPFLT: /* segment not present fault */
709 * Invalid segment selectors and out of bounds
710 * %eip's and %esp's can be set up in user mode.
711 * This causes a fault in kernel mode when the
712 * kernel tries to return to user mode. We want
713 * to get this fault so that we can fix the
714 * problem here and not have to check all the
715 * selectors and pointers when the user changes
718 #define MAYBE_DORETI_FAULT(where, whereto) \
720 if (frame->tf_eip == (int)where) { \
721 frame->tf_eip = (int)whereto; \
725 if (mycpu->gd_intr_nesting_level == 0) {
727 * Invalid %fs's and %gs's can be created using
728 * procfs or PT_SETREGS or by invalidating the
729 * underlying LDT entry. This causes a fault
730 * in kernel mode when the kernel attempts to
731 * switch contexts. Lose the bad context
732 * (XXX) so that we can continue, and generate
735 MAYBE_DORETI_FAULT(doreti_iret,
737 MAYBE_DORETI_FAULT(doreti_popl_ds,
738 doreti_popl_ds_fault);
739 MAYBE_DORETI_FAULT(doreti_popl_es,
740 doreti_popl_es_fault);
741 MAYBE_DORETI_FAULT(doreti_popl_fs,
742 doreti_popl_fs_fault);
743 MAYBE_DORETI_FAULT(doreti_popl_gs,
744 doreti_popl_gs_fault);
747 * NOTE: cpu doesn't push esp on kernel trap
749 if (td->td_pcb->pcb_onfault &&
750 td->td_pcb->pcb_onfault_sp ==
751 (int)&frame->tf_esp) {
753 (register_t)td->td_pcb->pcb_onfault;
761 * PSL_NT can be set in user mode and isn't cleared
762 * automatically when the kernel is entered. This
763 * causes a TSS fault when the kernel attempts to
764 * `iret' because the TSS link is uninitialized. We
765 * want to get this fault so that we can fix the
766 * problem here and not every time the kernel is
769 if (frame->tf_eflags & PSL_NT) {
770 frame->tf_eflags &= ~PSL_NT;
775 case T_TRCTRAP: /* trace trap */
776 if (frame->tf_eip == (int)IDTVEC(syscall)) {
778 * We've just entered system mode via the
779 * syscall lcall. Continue single stepping
780 * silently until the syscall handler has
785 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
787 * The syscall handler has now saved the
788 * flags. Stop single stepping it.
790 frame->tf_eflags &= ~PSL_T;
794 * Ignore debug register trace traps due to
795 * accesses in the user's address space, which
796 * can happen under several conditions such as
797 * if a user sets a watchpoint on a buffer and
798 * then passes that buffer to a system call.
799 * We still want to get TRCTRAPS for addresses
800 * in kernel space because that is useful when
801 * debugging the kernel.
803 if (user_dbreg_trap()) {
805 * Reset breakpoint bits because the
808 load_dr6(rdr6() & 0xfffffff0);
812 * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
816 * If DDB is enabled, let it handle the debugger trap.
817 * Otherwise, debugger traps "can't happen".
821 MAKEMPSAFE(have_mplock);
822 if (kdb_trap (type, 0, frame))
829 MAKEMPSAFE(have_mplock);
832 # define TIMER_FREQ 1193182
836 static unsigned lastalert = 0;
838 if (time_uptime - lastalert > 10) {
839 log(LOG_WARNING, "NMI: power fail\n");
840 sysbeep(TIMER_FREQ/880, hz);
841 lastalert = time_uptime;
846 #else /* !POWERFAIL_NMI */
847 /* machine/parity/power fail/"kitchen sink" faults */
848 if (isa_nmi(code) == 0) {
851 * NMI can be hooked up to a pushbutton
855 kprintf ("NMI ... going to debugger\n");
856 kdb_trap (type, 0, frame);
860 } else if (panic_on_nmi == 0)
863 #endif /* POWERFAIL_NMI */
864 #endif /* NISA > 0 */
867 MAKEMPSAFE(have_mplock);
868 trap_fatal(frame, eva);
873 * Virtual kernel intercept - if the fault is directly related to a
874 * VM context managed by a virtual kernel then let the virtual kernel
877 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
878 vkernel_trap(lp, frame);
882 /* Translate fault for emulators (e.g. Linux) */
883 if (*p->p_sysent->sv_transtrap)
884 i = (*p->p_sysent->sv_transtrap)(i, type);
886 MAKEMPSAFE(have_mplock);
887 trapsignal(lp, i, ucode);
890 if (type <= MAX_TRAP_MSG) {
891 uprintf("fatal process exception: %s",
893 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
894 uprintf(", fault VA = 0x%lx", (u_long)eva);
900 userret(lp, frame, sticks);
905 if (p != NULL && lp != NULL)
906 KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid);
908 KASSERT(crit_count == td->td_critcount,
909 ("trap: critical section count mismatch! %d/%d",
910 crit_count, td->td_pri));
911 KASSERT(curstop == td->td_toks_stop,
912 ("trap: extra tokens held after trap! %zd/%zd",
913 curstop - &td->td_toks_base,
914 td->td_toks_stop - &td->td_toks_base));
919 trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
922 struct vmspace *vm = NULL;
927 thread_t td = curthread;
928 struct lwp *lp = td->td_lwp;
930 va = trunc_page(eva);
931 if (va >= KERNBASE) {
933 * Don't allow user-mode faults in kernel address space.
934 * An exception: if the faulting address is the invalid
935 * instruction entry in the IDT, then the Intel Pentium
936 * F00F bug workaround was triggered, and we need to
937 * treat it is as an illegal instruction, and not a page
940 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
941 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) {
942 frame->tf_trapno = T_PRIVINFLT;
952 * This is a fault on non-kernel virtual memory.
953 * vm is initialized above to NULL. If curproc is NULL
954 * or curproc->p_vmspace is NULL the fault is fatal.
957 vm = lp->lwp_vmspace;
965 if (frame->tf_err & PGEX_W)
966 ftype = VM_PROT_WRITE;
968 ftype = VM_PROT_READ;
970 if (map != &kernel_map) {
972 * Keep swapout from messing with us during this
982 fault_flags |= VM_FAULT_BURST;
983 if (ftype & VM_PROT_WRITE)
984 fault_flags |= VM_FAULT_DIRTY;
986 fault_flags |= VM_FAULT_NORMAL;
987 rv = vm_fault(map, va, ftype, fault_flags);
991 * Don't have to worry about process locking or stacks in the
994 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
996 if (rv == KERN_SUCCESS)
1001 * NOTE: cpu doesn't push esp on kernel trap
1003 if (td->td_gd->gd_intr_nesting_level == 0 &&
1004 td->td_pcb->pcb_onfault &&
1005 td->td_pcb->pcb_onfault_sp == (int)&frame->tf_esp) {
1006 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
1009 if (td->td_gd->gd_intr_nesting_level == 0 &&
1010 td->td_pcb->pcb_onfault) {
1011 kprintf("ESP mismatch %p %08x\n",
1012 &frame->tf_esp, td->td_pcb->pcb_onfault_sp);
1014 trap_fatal(frame, eva);
1018 /* kludge to pass faulting virtual address to sendsig */
1019 frame->tf_xflags = frame->tf_err;
1020 frame->tf_err = eva;
1022 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
1026 trap_fatal(struct trapframe *frame, vm_offset_t eva)
1028 int code, type, ss, esp;
1029 struct soft_segment_descriptor softseg;
1031 code = frame->tf_err;
1032 type = frame->tf_trapno;
1033 sdtossd(&gdt[mycpu->gd_cpuid * NGDT + IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
1035 if (type <= MAX_TRAP_MSG)
1036 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
1037 type, trap_msg[type],
1038 frame->tf_eflags & PSL_VM ? "vm86" :
1039 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
1040 /* three separate prints in case of a trap on an unmapped page */
1041 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
1042 kprintf("lapic.id = %08x\n", lapic->id);
1043 if (type == T_PAGEFLT) {
1044 kprintf("fault virtual address = %p\n", (void *)eva);
1045 kprintf("fault code = %s %s, %s\n",
1046 code & PGEX_U ? "user" : "supervisor",
1047 code & PGEX_W ? "write" : "read",
1048 code & PGEX_P ? "protection violation" : "page not present");
1050 kprintf("instruction pointer = 0x%x:0x%x\n",
1051 frame->tf_cs & 0xffff, frame->tf_eip);
1052 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
1053 ss = frame->tf_ss & 0xffff;
1054 esp = frame->tf_esp;
1056 ss = GSEL(GDATA_SEL, SEL_KPL);
1057 esp = (int)&frame->tf_esp;
1059 kprintf("stack pointer = 0x%x:0x%x\n", ss, esp);
1060 kprintf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
1061 kprintf("code segment = base 0x%x, limit 0x%x, type 0x%x\n",
1062 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
1063 kprintf(" = DPL %d, pres %d, def32 %d, gran %d\n",
1064 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
1066 kprintf("processor eflags = ");
1067 if (frame->tf_eflags & PSL_T)
1068 kprintf("trace trap, ");
1069 if (frame->tf_eflags & PSL_I)
1070 kprintf("interrupt enabled, ");
1071 if (frame->tf_eflags & PSL_NT)
1072 kprintf("nested task, ");
1073 if (frame->tf_eflags & PSL_RF)
1074 kprintf("resume, ");
1075 if (frame->tf_eflags & PSL_VM)
1077 kprintf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
1078 kprintf("current process = ");
1080 kprintf("%lu (%s)\n",
1081 (u_long)curproc->p_pid, curproc->p_comm ?
1082 curproc->p_comm : "");
1086 kprintf("current thread = pri %d ", curthread->td_pri);
1087 if (curthread->td_critcount)
1092 * we probably SHOULD have stopped the other CPUs before now!
1093 * another CPU COULD have been touching cpl at this moment...
1095 kprintf(" <- SMP: XXX");
1103 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1106 kprintf("trap number = %d\n", type);
1107 if (type <= MAX_TRAP_MSG)
1108 panic("%s", trap_msg[type]);
1110 panic("unknown/reserved trap");
1114 * Double fault handler. Called when a fault occurs while writing
1115 * a frame for a trap/exception onto the stack. This usually occurs
1116 * when the stack overflows (such is the case with infinite recursion,
1119 * XXX Note that the current PTD gets replaced by IdlePTD when the
1120 * task switch occurs. This means that the stack that was active at
1121 * the time of the double fault is not available at <kstack> unless
1122 * the machine was idle when the double fault occurred. The downside
1123 * of this is that "trace <ebp>" in ddb won't work.
1127 in_kstack_guard(register_t rptr)
1129 thread_t td = curthread;
1131 if ((char *)rptr >= td->td_kstack &&
1132 (char *)rptr < td->td_kstack + PAGE_SIZE) {
1139 dblfault_handler(void)
1141 struct mdglobaldata *gd = mdcpu;
1143 if (in_kstack_guard(gd->gd_common_tss.tss_esp) ||
1144 in_kstack_guard(gd->gd_common_tss.tss_ebp)) {
1145 kprintf("DOUBLE FAULT - KERNEL STACK GUARD HIT!\n");
1147 kprintf("DOUBLE FAULT:\n");
1149 kprintf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1150 kprintf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1151 kprintf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
1152 /* three separate prints in case of a trap on an unmapped page */
1153 kprintf("cpuid = %d; ", gd->mi.gd_cpuid);
1154 kprintf("lapic.id = %08x\n", lapic->id);
1155 panic("double fault");
1159 * syscall2 - MP aware system call request C handler
1161 * A system call is essentially treated as a trap. The MP lock is not
1162 * held on entry or return. We are responsible for handling ASTs
1163 * (e.g. a task switch) prior to return.
1168 syscall2(struct trapframe *frame)
1170 struct thread *td = curthread;
1171 struct proc *p = td->td_proc;
1172 struct lwp *lp = td->td_lwp;
1174 struct sysent *callp;
1175 register_t orig_tf_eflags;
1180 int crit_count = td->td_critcount;
1182 int have_mplock = 0;
1184 union sysunion args;
1187 if (ISPL(frame->tf_cs) != SEL_UPL) {
1194 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid,
1197 userenter(td, p); /* lazy raise our priority */
1202 sticks = (int)td->td_sticks;
1203 orig_tf_eflags = frame->tf_eflags;
1206 * Virtual kernel intercept - if a VM context managed by a virtual
1207 * kernel issues a system call the virtual kernel handles it, not us.
1208 * Restore the virtual kernel context and return from its system
1209 * call. The current frame is copied out to the virtual kernel.
1211 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1212 vkernel_trap(lp, frame);
1213 error = EJUSTRETURN;
1219 * Get the system call parameters and account for time
1221 lp->lwp_md.md_regs = frame;
1222 params = (caddr_t)frame->tf_esp + sizeof(int);
1223 code = frame->tf_eax;
1225 if (p->p_sysent->sv_prepsyscall) {
1226 (*p->p_sysent->sv_prepsyscall)(
1227 frame, (int *)(&args.nosys.sysmsg + 1),
1231 * Need to check if this is a 32 bit or 64 bit syscall.
1232 * fuword is MP aware.
1234 if (code == SYS_syscall) {
1236 * Code is first argument, followed by actual args.
1238 code = fuword(params);
1239 params += sizeof(int);
1240 } else if (code == SYS___syscall) {
1242 * Like syscall, but code is a quad, so as to maintain
1243 * quad alignment for the rest of the arguments.
1245 code = fuword(params);
1246 params += sizeof(quad_t);
1250 code &= p->p_sysent->sv_mask;
1252 if (code >= p->p_sysent->sv_size)
1253 callp = &p->p_sysent->sv_table[0];
1255 callp = &p->p_sysent->sv_table[code];
1257 narg = callp->sy_narg & SYF_ARGMASK;
1260 if (p->p_sysent->sv_name[0] == 'L')
1261 kprintf("Linux syscall, code = %d\n", code);
1265 * copyin is MP aware, but the tracing code is not
1267 if (narg && params) {
1268 error = copyin(params, (caddr_t)(&args.nosys.sysmsg + 1),
1269 narg * sizeof(register_t));
1272 if (KTRPOINT(td, KTR_SYSCALL)) {
1273 MAKEMPSAFE(have_mplock);
1275 ktrsyscall(lp, code, narg,
1276 (void *)(&args.nosys.sysmsg + 1));
1284 if (KTRPOINT(td, KTR_SYSCALL)) {
1285 MAKEMPSAFE(have_mplock);
1286 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1291 * For traditional syscall code edx is left untouched when 32 bit
1292 * results are returned. Since edx is loaded from fds[1] when the
1293 * system call returns we pre-set it here.
1295 args.sysmsg_fds[0] = 0;
1296 args.sysmsg_fds[1] = frame->tf_edx;
1299 * The syscall might manipulate the trap frame. If it does it
1300 * will probably return EJUSTRETURN.
1302 args.sysmsg_frame = frame;
1304 STOPEVENT(p, S_SCE, narg); /* MP aware */
1307 * NOTE: All system calls run MPSAFE now. The system call itself
1308 * is responsible for getting the MP lock.
1310 error = (*callp->sy_call)(&args);
1314 * MP SAFE (we may or may not have the MP lock at this point)
1319 * Reinitialize proc pointer `p' as it may be different
1320 * if this is a child returning from fork syscall.
1323 lp = curthread->td_lwp;
1324 frame->tf_eax = args.sysmsg_fds[0];
1325 frame->tf_edx = args.sysmsg_fds[1];
1326 frame->tf_eflags &= ~PSL_C;
1330 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1331 * int 0x80 is 2 bytes. We saved this in tf_err.
1333 frame->tf_eip -= frame->tf_err;
1338 panic("Unexpected EASYNC return value (for now)");
1341 if (p->p_sysent->sv_errsize) {
1342 if (error >= p->p_sysent->sv_errsize)
1343 error = -1; /* XXX */
1345 error = p->p_sysent->sv_errtbl[error];
1347 frame->tf_eax = error;
1348 frame->tf_eflags |= PSL_C;
1353 * Traced syscall. trapsignal() is not MP aware.
1355 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1356 MAKEMPSAFE(have_mplock);
1357 frame->tf_eflags &= ~PSL_T;
1358 trapsignal(lp, SIGTRAP, TRAP_TRACE);
1362 * Handle reschedule and other end-of-syscall issues
1364 userret(lp, frame, sticks);
1367 if (KTRPOINT(td, KTR_SYSRET)) {
1368 MAKEMPSAFE(have_mplock);
1369 ktrsysret(lp, code, error, args.sysmsg_result);
1374 * This works because errno is findable through the
1375 * register set. If we ever support an emulation where this
1376 * is not the case, this code will need to be revisited.
1378 STOPEVENT(p, S_SCX, code);
1382 * Release the MP lock if we had to get it
1386 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error);
1388 KASSERT(crit_count == td->td_critcount,
1389 ("syscall: critical section count mismatch! %d/%d",
1390 crit_count, td->td_pri));
1391 KASSERT(&td->td_toks_base == td->td_toks_stop,
1392 ("syscall: extra tokens held after trap! %zd",
1393 td->td_toks_stop - &td->td_toks_base));
1398 * NOTE: MP lock not held at any point.
1401 fork_return(struct lwp *lp, struct trapframe *frame)
1403 frame->tf_eax = 0; /* Child returns zero */
1404 frame->tf_eflags &= ~PSL_C; /* success */
1407 generic_lwp_return(lp, frame);
1408 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1412 * Simplified back end of syscall(), used when returning from fork()
1413 * directly into user mode.
1415 * This code will return back into the fork trampoline code which then
1418 * NOTE: The mplock is not held at any point.
1421 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1423 struct proc *p = lp->lwp_proc;
1426 * Newly forked processes are given a kernel priority. We have to
1427 * adjust the priority to a normal user priority and fake entry
1428 * into the kernel (call userenter()) to install a passive release
1429 * function just in case userret() decides to stop the process. This
1430 * can occur when ^Z races a fork. If we do not install the passive
1431 * release function the current process designation will not be
1432 * released when the thread goes to sleep.
1434 lwkt_setpri_self(TDPRI_USER_NORM);
1435 userenter(lp->lwp_thread, p);
1436 userret(lp, frame, 0);
1438 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1439 ktrsysret(lp, SYS_fork, 0, 0);
1441 lp->lwp_flags |= LWP_PASSIVE_ACQ;
1443 lp->lwp_flags &= ~LWP_PASSIVE_ACQ;
1447 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1448 * fault (which is then passed back to the virtual kernel) if an attempt is
1449 * made to use the FP unit.
1451 * XXX this is a fairly big hack.
1454 set_vkernel_fp(struct trapframe *frame)
1456 struct thread *td = curthread;
1458 if (frame->tf_xflags & PGEX_FPFAULT) {
1459 td->td_pcb->pcb_flags |= FP_VIRTFP;
1460 if (mdcpu->gd_npxthread == td)
1463 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1468 * Called from vkernel_trap() to fixup the vkernel's syscall
1469 * frame for vmspace_ctl() return.
1472 cpu_vkernel_trap(struct trapframe *frame, int error)
1474 frame->tf_eax = error;
1476 frame->tf_eflags |= PSL_C;
1478 frame->tf_eflags &= ~PSL_C;