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/upcall.h>
74 #include <sys/vkernel.h>
75 #include <sys/sysproto.h>
76 #include <sys/sysunion.h>
79 #include <vm/vm_param.h>
82 #include <vm/vm_kern.h>
83 #include <vm/vm_map.h>
84 #include <vm/vm_page.h>
85 #include <vm/vm_extern.h>
87 #include <machine/cpu.h>
88 #include <machine/md_var.h>
89 #include <machine/pcb.h>
90 #include <machine/smp.h>
91 #include <machine/tss.h>
92 #include <machine/specialreg.h>
93 #include <machine/globaldata.h>
94 #include <machine/intr_machdep.h>
96 #include <machine_base/isa/isa_intr.h>
97 #include <machine_base/apic/lapic.h>
100 #include <sys/syslog.h>
101 #include <machine/clock.h>
104 #include <machine/vm86.h>
108 #include <sys/msgport2.h>
109 #include <sys/thread2.h>
110 #include <sys/mplock2.h>
112 #define MAKEMPSAFE(have_mplock) \
113 if (have_mplock == 0) { \
118 int (*pmath_emulate) (struct trapframe *);
120 extern void trap (struct trapframe *frame);
121 extern void syscall2 (struct trapframe *frame);
123 static int trap_pfault (struct trapframe *, int, vm_offset_t);
124 static void trap_fatal (struct trapframe *, vm_offset_t);
125 void dblfault_handler (void);
127 extern inthand_t IDTVEC(syscall);
129 #define MAX_TRAP_MSG 28
130 static char *trap_msg[] = {
132 "privileged instruction fault", /* 1 T_PRIVINFLT */
134 "breakpoint instruction fault", /* 3 T_BPTFLT */
137 "arithmetic trap", /* 6 T_ARITHTRAP */
138 "system forced exception", /* 7 T_ASTFLT */
140 "general protection fault", /* 9 T_PROTFLT */
141 "trace trap", /* 10 T_TRCTRAP */
143 "page fault", /* 12 T_PAGEFLT */
145 "alignment fault", /* 14 T_ALIGNFLT */
149 "integer divide fault", /* 18 T_DIVIDE */
150 "non-maskable interrupt trap", /* 19 T_NMI */
151 "overflow trap", /* 20 T_OFLOW */
152 "FPU bounds check fault", /* 21 T_BOUND */
153 "FPU device not available", /* 22 T_DNA */
154 "double fault", /* 23 T_DOUBLEFLT */
155 "FPU operand fetch fault", /* 24 T_FPOPFLT */
156 "invalid TSS fault", /* 25 T_TSSFLT */
157 "segment not present fault", /* 26 T_SEGNPFLT */
158 "stack fault", /* 27 T_STKFLT */
159 "machine check trap", /* 28 T_MCHK */
162 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
163 extern int has_f00f_bug;
167 static int ddb_on_nmi = 1;
168 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
169 &ddb_on_nmi, 0, "Go to DDB on NMI");
171 static int panic_on_nmi = 1;
172 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
173 &panic_on_nmi, 0, "Panic on NMI");
174 static int fast_release;
175 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
176 &fast_release, 0, "Passive Release was optimal");
177 static int slow_release;
178 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
179 &slow_release, 0, "Passive Release was nonoptimal");
181 MALLOC_DEFINE(M_SYSMSG, "sysmsg", "sysmsg structure");
182 extern int max_sysmsg;
185 * Passively intercepts the thread switch function to increase the thread
186 * priority from a user priority to a kernel priority, reducing
187 * syscall and trap overhead for the case where no switch occurs.
189 * Synchronizes td_ucred with p_ucred. This is used by system calls,
190 * signal handling, faults, AST traps, and anything else that enters the
191 * kernel from userland and provides the kernel with a stable read-only
192 * copy of the process ucred.
195 userenter(struct thread *curtd, struct proc *curp)
200 curtd->td_release = lwkt_passive_release;
202 if (curtd->td_ucred != curp->p_ucred) {
203 ncred = crhold(curp->p_ucred);
204 ocred = curtd->td_ucred;
205 curtd->td_ucred = ncred;
213 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
214 * must be completed before we can return to or try to return to userland.
216 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
217 * arithmatic on the delta calculation so the absolute tick values are
218 * truncated to an integer.
221 userret(struct lwp *lp, struct trapframe *frame, int sticks)
223 struct proc *p = lp->lwp_proc;
227 if (p->p_userret != NULL) {
234 * Charge system time if profiling. Note: times are in microseconds.
235 * This may do a copyout and block, so do it first even though it
236 * means some system time will be charged as user time.
238 if (p->p_flags & P_PROFIL) {
239 addupc_task(p, frame->tf_eip,
240 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
245 * If the jungle wants us dead, so be it.
247 if (lp->lwp_mpflags & LWP_MP_WEXIT) {
248 lwkt_gettoken(&p->p_token);
250 lwkt_reltoken(&p->p_token); /* NOT REACHED */
254 * Block here if we are in a stopped state.
256 if (p->p_stat == SSTOP || dump_stop_usertds) {
257 lwkt_gettoken(&p->p_token);
259 lwkt_reltoken(&p->p_token);
264 * Post any pending upcalls. If running a virtual kernel be sure
265 * to restore the virtual kernel's vmspace before posting the upcall.
267 if (p->p_flags & (P_SIGVTALRM | P_SIGPROF | P_UPCALLPEND)) {
268 lwkt_gettoken(&p->p_token);
269 if (p->p_flags & P_SIGVTALRM) {
270 p->p_flags &= ~P_SIGVTALRM;
271 ksignal(p, SIGVTALRM);
273 if (p->p_flags & P_SIGPROF) {
274 p->p_flags &= ~P_SIGPROF;
277 if (p->p_flags & P_UPCALLPEND) {
278 p->p_flags &= ~P_UPCALLPEND;
281 lwkt_reltoken(&p->p_token);
286 * Post any pending signals. If running a virtual kernel be sure
287 * to restore the virtual kernel's vmspace before posting the signal.
289 * WARNING! postsig() can exit and not return.
291 if ((sig = CURSIG_TRACE(lp)) != 0) {
292 lwkt_gettoken(&p->p_token);
294 lwkt_reltoken(&p->p_token);
299 * block here if we are swapped out, but still process signals
300 * (such as SIGKILL). proc0 (the swapin scheduler) is already
301 * aware of our situation, we do not have to wake it up.
303 if (p->p_flags & P_SWAPPEDOUT) {
304 lwkt_gettoken(&p->p_token);
306 p->p_flags |= P_SWAPWAIT;
308 if (p->p_flags & P_SWAPWAIT)
309 tsleep(p, PCATCH, "SWOUT", 0);
310 p->p_flags &= ~P_SWAPWAIT;
312 lwkt_reltoken(&p->p_token);
317 * In a multi-threaded program it is possible for a thread to change
318 * signal state during a system call which temporarily changes the
319 * signal mask. In this case postsig() might not be run and we
320 * have to restore the mask ourselves.
322 if (lp->lwp_flags & LWP_OLDMASK) {
323 lp->lwp_flags &= ~LWP_OLDMASK;
324 lp->lwp_sigmask = lp->lwp_oldsigmask;
330 * Cleanup from userenter and any passive release that might have occured.
331 * We must reclaim the current-process designation before we can return
332 * to usermode. We also handle both LWKT and USER reschedule requests.
335 userexit(struct lwp *lp)
337 struct thread *td = lp->lwp_thread;
338 /* globaldata_t gd = td->td_gd; */
341 * Handle stop requests at kernel priority. Any requests queued
342 * after this loop will generate another AST.
344 while (lp->lwp_proc->p_stat == SSTOP) {
345 lwkt_gettoken(&lp->lwp_proc->p_token);
347 lwkt_reltoken(&lp->lwp_proc->p_token);
351 * Become the current user scheduled process if we aren't already,
352 * and deal with reschedule requests and other factors.
354 lp->lwp_proc->p_usched->acquire_curproc(lp);
355 /* WARNING: we may have migrated cpu's */
356 /* gd = td->td_gd; */
359 * Reduce our priority in preparation for a return to userland. If
360 * our passive release function was still in place, our priority was
361 * never raised and does not need to be reduced.
363 lwkt_passive_recover(td);
366 #if !defined(KTR_KERNENTRY)
367 #define KTR_KERNENTRY KTR_ALL
369 KTR_INFO_MASTER(kernentry);
370 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0,
371 "TRAP(pid %d, tid %d, trapno %d, eva %lu)",
372 pid_t pid, lwpid_t tid, register_t trapno, vm_offset_t eva);
373 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "TRAP_RET(pid %d, tid %d)",
374 pid_t pid, lwpid_t tid);
375 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "SYSC(pid %d, tid %d, nr %d)",
376 pid_t pid, lwpid_t tid, register_t trapno);
377 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "SYSRET(pid %d, tid %d, err %d)",
378 pid_t pid, lwpid_t tid, int err);
379 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "FORKRET(pid %d, tid %d)",
380 pid_t pid, lwpid_t tid);
383 * Exception, fault, and trap interface to the kernel.
384 * This common code is called from assembly language IDT gate entry
385 * routines that prepare a suitable stack frame, and restore this
386 * frame after the exception has been processed.
388 * This function is also called from doreti in an interlock to handle ASTs.
389 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
391 * NOTE! We have to retrieve the fault address prior to obtaining the
392 * MP lock because get_mplock() may switch out. YYY cr2 really ought
393 * to be retrieved by the assembly code, not here.
395 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
396 * if an attempt is made to switch from a fast interrupt or IPI. This is
397 * necessary to properly take fatal kernel traps on SMP machines if
398 * get_mplock() has to block.
402 trap(struct trapframe *frame)
404 struct globaldata *gd = mycpu;
405 struct thread *td = gd->gd_curthread;
406 struct lwp *lp = td->td_lwp;
409 int i = 0, ucode = 0, type, code;
412 int crit_count = td->td_critcount;
413 lwkt_tokref_t curstop = td->td_toks_stop;
420 * We need to allow T_DNA faults when the debugger is active since
421 * some dumping paths do large bcopy() which use the floating
422 * point registers for faster copying.
424 if (db_active && frame->tf_trapno != T_DNA) {
425 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
426 ++gd->gd_trap_nesting_level;
427 MAKEMPSAFE(have_mplock);
428 trap_fatal(frame, eva);
429 --gd->gd_trap_nesting_level;
435 ++gd->gd_trap_nesting_level;
436 if (frame->tf_trapno == T_PAGEFLT) {
438 * For some Cyrix CPUs, %cr2 is clobbered by interrupts.
439 * This problem is worked around by using an interrupt
440 * gate for the pagefault handler. We are finally ready
441 * to read %cr2 and then must reenable interrupts.
443 * XXX this should be in the switch statement, but the
444 * NO_FOOF_HACK and VM86 goto and ifdefs obfuscate the
445 * flow of control too much for this to be obviously
452 --gd->gd_trap_nesting_level;
454 if (!(frame->tf_eflags & PSL_I)) {
456 * Buggy application or kernel code has disabled interrupts
457 * and then trapped. Enabling interrupts now is wrong, but
458 * it is better than running with interrupts disabled until
459 * they are accidentally enabled later.
461 type = frame->tf_trapno;
462 if (ISPL(frame->tf_cs)==SEL_UPL || (frame->tf_eflags & PSL_VM)) {
463 MAKEMPSAFE(have_mplock);
465 "pid %ld (%s): trap %d with interrupts disabled\n",
466 (long)curproc->p_pid, curproc->p_comm, type);
467 } else if (type != T_BPTFLT && type != T_TRCTRAP) {
469 * XXX not quite right, since this may be for a
470 * multiple fault in user mode.
472 MAKEMPSAFE(have_mplock);
473 kprintf("kernel trap %d with interrupts disabled\n",
479 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
482 type = frame->tf_trapno;
483 code = frame->tf_err;
486 if (frame->tf_eflags & PSL_VM &&
487 (type == T_PROTFLT || type == T_STKFLT)) {
488 KKASSERT(get_mplock_count(curthread) > 0);
489 i = vm86_emulate((struct vm86frame *)frame);
490 KKASSERT(get_mplock_count(curthread) > 0);
493 * returns to original process
495 vm86_trap((struct vm86frame *)frame,
497 KKASSERT(0); /* NOT REACHED */
503 * these traps want either a process context, or
504 * assume a normal userspace trap.
508 trap_fatal(frame, eva);
511 type = T_BPTFLT; /* kernel breakpoint */
514 goto kernel_trap; /* normal kernel trap handling */
517 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
520 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid,
521 frame->tf_trapno, eva);
525 sticks = (int)td->td_sticks;
526 lp->lwp_md.md_regs = frame;
529 case T_PRIVINFLT: /* privileged instruction fault */
534 case T_BPTFLT: /* bpt instruction fault */
535 case T_TRCTRAP: /* trace trap */
536 frame->tf_eflags &= ~PSL_T;
538 ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
541 case T_ARITHTRAP: /* arithmetic trap */
546 case T_ASTFLT: /* Allow process switch */
547 mycpu->gd_cnt.v_soft++;
548 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
549 atomic_clear_int(&mycpu->gd_reqflags,
551 addupc_task(p, p->p_prof.pr_addr,
557 * The following two traps can happen in
558 * vm86 mode, and, if so, we want to handle
561 case T_PROTFLT: /* general protection fault */
562 case T_STKFLT: /* stack fault */
563 if (frame->tf_eflags & PSL_VM) {
564 i = vm86_emulate((struct vm86frame *)frame);
570 ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR;
572 case T_SEGNPFLT: /* segment not present fault */
576 case T_TSSFLT: /* invalid TSS fault */
577 case T_DOUBLEFLT: /* double fault */
583 case T_PAGEFLT: /* page fault */
584 i = trap_pfault(frame, TRUE, eva);
587 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
602 case T_DIVIDE: /* integer divide fault */
609 MAKEMPSAFE(have_mplock);
611 goto handle_powerfail;
612 #else /* !POWERFAIL_NMI */
613 /* machine/parity/power fail/"kitchen sink" faults */
614 if (isa_nmi(code) == 0) {
617 * NMI can be hooked up to a pushbutton
621 kprintf ("NMI ... going to debugger\n");
622 kdb_trap (type, 0, frame);
626 } else if (panic_on_nmi)
627 panic("NMI indicates hardware failure");
629 #endif /* POWERFAIL_NMI */
630 #endif /* NISA > 0 */
632 case T_OFLOW: /* integer overflow fault */
637 case T_BOUND: /* bounds check fault */
644 * Virtual kernel intercept - pass the DNA exception
645 * to the virtual kernel if it asked to handle it.
646 * This occurs when the virtual kernel is holding
647 * onto the FP context for a different emulated
648 * process then the one currently running.
650 * We must still call npxdna() since we may have
651 * saved FP state that the virtual kernel needs
652 * to hand over to a different emulated process.
654 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
655 (td->td_pcb->pcb_flags & FP_VIRTFP)
663 * The kernel may have switched out the FP unit's
664 * state, causing the user process to take a fault
665 * when it tries to use the FP unit. Restore the
671 if (!pmath_emulate) {
673 ucode = FPE_FPU_NP_TRAP;
676 i = (*pmath_emulate)(frame);
678 if (!(frame->tf_eflags & PSL_T))
680 frame->tf_eflags &= ~PSL_T;
683 /* else ucode = emulator_only_knows() XXX */
686 case T_FPOPFLT: /* FPU operand fetch fault */
691 case T_XMMFLT: /* SIMD floating-point exception */
701 case T_PAGEFLT: /* page fault */
702 trap_pfault(frame, FALSE, eva);
708 * The kernel may be using npx for copying or other
716 case T_PROTFLT: /* general protection fault */
717 case T_SEGNPFLT: /* segment not present fault */
719 * Invalid segment selectors and out of bounds
720 * %eip's and %esp's can be set up in user mode.
721 * This causes a fault in kernel mode when the
722 * kernel tries to return to user mode. We want
723 * to get this fault so that we can fix the
724 * problem here and not have to check all the
725 * selectors and pointers when the user changes
728 #define MAYBE_DORETI_FAULT(where, whereto) \
730 if (frame->tf_eip == (int)where) { \
731 frame->tf_eip = (int)whereto; \
735 if (mycpu->gd_intr_nesting_level == 0) {
737 * Invalid %fs's and %gs's can be created using
738 * procfs or PT_SETREGS or by invalidating the
739 * underlying LDT entry. This causes a fault
740 * in kernel mode when the kernel attempts to
741 * switch contexts. Lose the bad context
742 * (XXX) so that we can continue, and generate
745 MAYBE_DORETI_FAULT(doreti_iret,
747 MAYBE_DORETI_FAULT(doreti_popl_ds,
748 doreti_popl_ds_fault);
749 MAYBE_DORETI_FAULT(doreti_popl_es,
750 doreti_popl_es_fault);
751 MAYBE_DORETI_FAULT(doreti_popl_fs,
752 doreti_popl_fs_fault);
753 MAYBE_DORETI_FAULT(doreti_popl_gs,
754 doreti_popl_gs_fault);
757 * NOTE: cpu doesn't push esp on kernel trap
759 if (td->td_pcb->pcb_onfault &&
760 td->td_pcb->pcb_onfault_sp ==
761 (int)&frame->tf_esp) {
763 (register_t)td->td_pcb->pcb_onfault;
771 * PSL_NT can be set in user mode and isn't cleared
772 * automatically when the kernel is entered. This
773 * causes a TSS fault when the kernel attempts to
774 * `iret' because the TSS link is uninitialized. We
775 * want to get this fault so that we can fix the
776 * problem here and not every time the kernel is
779 if (frame->tf_eflags & PSL_NT) {
780 frame->tf_eflags &= ~PSL_NT;
785 case T_TRCTRAP: /* trace trap */
786 if (frame->tf_eip == (int)IDTVEC(syscall)) {
788 * We've just entered system mode via the
789 * syscall lcall. Continue single stepping
790 * silently until the syscall handler has
795 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
797 * The syscall handler has now saved the
798 * flags. Stop single stepping it.
800 frame->tf_eflags &= ~PSL_T;
804 * Ignore debug register trace traps due to
805 * accesses in the user's address space, which
806 * can happen under several conditions such as
807 * if a user sets a watchpoint on a buffer and
808 * then passes that buffer to a system call.
809 * We still want to get TRCTRAPS for addresses
810 * in kernel space because that is useful when
811 * debugging the kernel.
813 if (user_dbreg_trap()) {
815 * Reset breakpoint bits because the
818 load_dr6(rdr6() & 0xfffffff0);
822 * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
826 * If DDB is enabled, let it handle the debugger trap.
827 * Otherwise, debugger traps "can't happen".
831 MAKEMPSAFE(have_mplock);
832 if (kdb_trap (type, 0, frame))
839 MAKEMPSAFE(have_mplock);
842 # define TIMER_FREQ 1193182
846 static unsigned lastalert = 0;
848 if(time_second - lastalert > 10)
850 log(LOG_WARNING, "NMI: power fail\n");
851 sysbeep(TIMER_FREQ/880, hz);
852 lastalert = time_second;
857 #else /* !POWERFAIL_NMI */
858 /* machine/parity/power fail/"kitchen sink" faults */
859 if (isa_nmi(code) == 0) {
862 * NMI can be hooked up to a pushbutton
866 kprintf ("NMI ... going to debugger\n");
867 kdb_trap (type, 0, frame);
871 } else if (panic_on_nmi == 0)
874 #endif /* POWERFAIL_NMI */
875 #endif /* NISA > 0 */
878 MAKEMPSAFE(have_mplock);
879 trap_fatal(frame, eva);
884 * Virtual kernel intercept - if the fault is directly related to a
885 * VM context managed by a virtual kernel then let the virtual kernel
888 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
889 vkernel_trap(lp, frame);
893 /* Translate fault for emulators (e.g. Linux) */
894 if (*p->p_sysent->sv_transtrap)
895 i = (*p->p_sysent->sv_transtrap)(i, type);
897 MAKEMPSAFE(have_mplock);
898 trapsignal(lp, i, ucode);
901 if (type <= MAX_TRAP_MSG) {
902 uprintf("fatal process exception: %s",
904 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
905 uprintf(", fault VA = 0x%lx", (u_long)eva);
911 userret(lp, frame, sticks);
916 if (p != NULL && lp != NULL)
917 KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid);
919 KASSERT(crit_count == td->td_critcount,
920 ("trap: critical section count mismatch! %d/%d",
921 crit_count, td->td_pri));
922 KASSERT(curstop == td->td_toks_stop,
923 ("trap: extra tokens held after trap! %zd/%zd",
924 curstop - &td->td_toks_base,
925 td->td_toks_stop - &td->td_toks_base));
930 trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
933 struct vmspace *vm = NULL;
938 thread_t td = curthread;
939 struct lwp *lp = td->td_lwp;
941 va = trunc_page(eva);
942 if (va >= KERNBASE) {
944 * Don't allow user-mode faults in kernel address space.
945 * An exception: if the faulting address is the invalid
946 * instruction entry in the IDT, then the Intel Pentium
947 * F00F bug workaround was triggered, and we need to
948 * treat it is as an illegal instruction, and not a page
951 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
952 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) {
953 frame->tf_trapno = T_PRIVINFLT;
963 * This is a fault on non-kernel virtual memory.
964 * vm is initialized above to NULL. If curproc is NULL
965 * or curproc->p_vmspace is NULL the fault is fatal.
968 vm = lp->lwp_vmspace;
976 if (frame->tf_err & PGEX_W)
977 ftype = VM_PROT_WRITE;
979 ftype = VM_PROT_READ;
981 if (map != &kernel_map) {
983 * Keep swapout from messing with us during this
993 fault_flags |= VM_FAULT_BURST;
994 if (ftype & VM_PROT_WRITE)
995 fault_flags |= VM_FAULT_DIRTY;
997 fault_flags |= VM_FAULT_NORMAL;
998 rv = vm_fault(map, va, ftype, fault_flags);
1002 * Don't have to worry about process locking or stacks in the
1005 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
1007 if (rv == KERN_SUCCESS)
1012 * NOTE: cpu doesn't push esp on kernel trap
1014 if (td->td_gd->gd_intr_nesting_level == 0 &&
1015 td->td_pcb->pcb_onfault &&
1016 td->td_pcb->pcb_onfault_sp == (int)&frame->tf_esp) {
1017 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
1020 if (td->td_gd->gd_intr_nesting_level == 0 &&
1021 td->td_pcb->pcb_onfault) {
1022 kprintf("ESP mismatch %p %08x\n",
1023 &frame->tf_esp, td->td_pcb->pcb_onfault_sp);
1025 trap_fatal(frame, eva);
1029 /* kludge to pass faulting virtual address to sendsig */
1030 frame->tf_xflags = frame->tf_err;
1031 frame->tf_err = eva;
1033 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
1037 trap_fatal(struct trapframe *frame, vm_offset_t eva)
1039 int code, type, ss, esp;
1040 struct soft_segment_descriptor softseg;
1042 code = frame->tf_err;
1043 type = frame->tf_trapno;
1044 sdtossd(&gdt[mycpu->gd_cpuid * NGDT + IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
1046 if (type <= MAX_TRAP_MSG)
1047 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
1048 type, trap_msg[type],
1049 frame->tf_eflags & PSL_VM ? "vm86" :
1050 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
1051 /* three separate prints in case of a trap on an unmapped page */
1052 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
1053 kprintf("lapic.id = %08x\n", lapic->id);
1054 if (type == T_PAGEFLT) {
1055 kprintf("fault virtual address = %p\n", (void *)eva);
1056 kprintf("fault code = %s %s, %s\n",
1057 code & PGEX_U ? "user" : "supervisor",
1058 code & PGEX_W ? "write" : "read",
1059 code & PGEX_P ? "protection violation" : "page not present");
1061 kprintf("instruction pointer = 0x%x:0x%x\n",
1062 frame->tf_cs & 0xffff, frame->tf_eip);
1063 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
1064 ss = frame->tf_ss & 0xffff;
1065 esp = frame->tf_esp;
1067 ss = GSEL(GDATA_SEL, SEL_KPL);
1068 esp = (int)&frame->tf_esp;
1070 kprintf("stack pointer = 0x%x:0x%x\n", ss, esp);
1071 kprintf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
1072 kprintf("code segment = base 0x%x, limit 0x%x, type 0x%x\n",
1073 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
1074 kprintf(" = DPL %d, pres %d, def32 %d, gran %d\n",
1075 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
1077 kprintf("processor eflags = ");
1078 if (frame->tf_eflags & PSL_T)
1079 kprintf("trace trap, ");
1080 if (frame->tf_eflags & PSL_I)
1081 kprintf("interrupt enabled, ");
1082 if (frame->tf_eflags & PSL_NT)
1083 kprintf("nested task, ");
1084 if (frame->tf_eflags & PSL_RF)
1085 kprintf("resume, ");
1086 if (frame->tf_eflags & PSL_VM)
1088 kprintf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
1089 kprintf("current process = ");
1091 kprintf("%lu (%s)\n",
1092 (u_long)curproc->p_pid, curproc->p_comm ?
1093 curproc->p_comm : "");
1097 kprintf("current thread = pri %d ", curthread->td_pri);
1098 if (curthread->td_critcount)
1103 * we probably SHOULD have stopped the other CPUs before now!
1104 * another CPU COULD have been touching cpl at this moment...
1106 kprintf(" <- SMP: XXX");
1114 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1117 kprintf("trap number = %d\n", type);
1118 if (type <= MAX_TRAP_MSG)
1119 panic("%s", trap_msg[type]);
1121 panic("unknown/reserved trap");
1125 * Double fault handler. Called when a fault occurs while writing
1126 * a frame for a trap/exception onto the stack. This usually occurs
1127 * when the stack overflows (such is the case with infinite recursion,
1130 * XXX Note that the current PTD gets replaced by IdlePTD when the
1131 * task switch occurs. This means that the stack that was active at
1132 * the time of the double fault is not available at <kstack> unless
1133 * the machine was idle when the double fault occurred. The downside
1134 * of this is that "trace <ebp>" in ddb won't work.
1138 in_kstack_guard(register_t rptr)
1140 thread_t td = curthread;
1142 if ((char *)rptr >= td->td_kstack &&
1143 (char *)rptr < td->td_kstack + PAGE_SIZE) {
1150 dblfault_handler(void)
1152 struct mdglobaldata *gd = mdcpu;
1154 if (in_kstack_guard(gd->gd_common_tss.tss_esp) ||
1155 in_kstack_guard(gd->gd_common_tss.tss_ebp)) {
1156 kprintf("DOUBLE FAULT - KERNEL STACK GUARD HIT!\n");
1158 kprintf("DOUBLE FAULT:\n");
1160 kprintf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1161 kprintf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1162 kprintf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
1163 /* three separate prints in case of a trap on an unmapped page */
1164 kprintf("cpuid = %d; ", gd->mi.gd_cpuid);
1165 kprintf("lapic.id = %08x\n", lapic->id);
1166 panic("double fault");
1170 * syscall2 - MP aware system call request C handler
1172 * A system call is essentially treated as a trap. The MP lock is not
1173 * held on entry or return. We are responsible for handling ASTs
1174 * (e.g. a task switch) prior to return.
1179 syscall2(struct trapframe *frame)
1181 struct thread *td = curthread;
1182 struct proc *p = td->td_proc;
1183 struct lwp *lp = td->td_lwp;
1185 struct sysent *callp;
1186 register_t orig_tf_eflags;
1191 int crit_count = td->td_critcount;
1193 int have_mplock = 0;
1195 union sysunion args;
1198 if (ISPL(frame->tf_cs) != SEL_UPL) {
1205 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid,
1208 userenter(td, p); /* lazy raise our priority */
1213 sticks = (int)td->td_sticks;
1214 orig_tf_eflags = frame->tf_eflags;
1217 * Virtual kernel intercept - if a VM context managed by a virtual
1218 * kernel issues a system call the virtual kernel handles it, not us.
1219 * Restore the virtual kernel context and return from its system
1220 * call. The current frame is copied out to the virtual kernel.
1222 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1223 vkernel_trap(lp, frame);
1224 error = EJUSTRETURN;
1230 * Get the system call parameters and account for time
1232 lp->lwp_md.md_regs = frame;
1233 params = (caddr_t)frame->tf_esp + sizeof(int);
1234 code = frame->tf_eax;
1236 if (p->p_sysent->sv_prepsyscall) {
1237 (*p->p_sysent->sv_prepsyscall)(
1238 frame, (int *)(&args.nosys.sysmsg + 1),
1242 * Need to check if this is a 32 bit or 64 bit syscall.
1243 * fuword is MP aware.
1245 if (code == SYS_syscall) {
1247 * Code is first argument, followed by actual args.
1249 code = fuword(params);
1250 params += sizeof(int);
1251 } else if (code == SYS___syscall) {
1253 * Like syscall, but code is a quad, so as to maintain
1254 * quad alignment for the rest of the arguments.
1256 code = fuword(params);
1257 params += sizeof(quad_t);
1261 code &= p->p_sysent->sv_mask;
1263 if (code >= p->p_sysent->sv_size)
1264 callp = &p->p_sysent->sv_table[0];
1266 callp = &p->p_sysent->sv_table[code];
1268 narg = callp->sy_narg & SYF_ARGMASK;
1271 if (p->p_sysent->sv_name[0] == 'L')
1272 kprintf("Linux syscall, code = %d\n", code);
1276 * copyin is MP aware, but the tracing code is not
1278 if (narg && params) {
1279 error = copyin(params, (caddr_t)(&args.nosys.sysmsg + 1),
1280 narg * sizeof(register_t));
1283 if (KTRPOINT(td, KTR_SYSCALL)) {
1284 MAKEMPSAFE(have_mplock);
1286 ktrsyscall(lp, code, narg,
1287 (void *)(&args.nosys.sysmsg + 1));
1295 if (KTRPOINT(td, KTR_SYSCALL)) {
1296 MAKEMPSAFE(have_mplock);
1297 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1302 * For traditional syscall code edx is left untouched when 32 bit
1303 * results are returned. Since edx is loaded from fds[1] when the
1304 * system call returns we pre-set it here.
1306 args.sysmsg_fds[0] = 0;
1307 args.sysmsg_fds[1] = frame->tf_edx;
1310 * The syscall might manipulate the trap frame. If it does it
1311 * will probably return EJUSTRETURN.
1313 args.sysmsg_frame = frame;
1315 STOPEVENT(p, S_SCE, narg); /* MP aware */
1318 * NOTE: All system calls run MPSAFE now. The system call itself
1319 * is responsible for getting the MP lock.
1321 error = (*callp->sy_call)(&args);
1325 * MP SAFE (we may or may not have the MP lock at this point)
1330 * Reinitialize proc pointer `p' as it may be different
1331 * if this is a child returning from fork syscall.
1334 lp = curthread->td_lwp;
1335 frame->tf_eax = args.sysmsg_fds[0];
1336 frame->tf_edx = args.sysmsg_fds[1];
1337 frame->tf_eflags &= ~PSL_C;
1341 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1342 * int 0x80 is 2 bytes. We saved this in tf_err.
1344 frame->tf_eip -= frame->tf_err;
1349 panic("Unexpected EASYNC return value (for now)");
1352 if (p->p_sysent->sv_errsize) {
1353 if (error >= p->p_sysent->sv_errsize)
1354 error = -1; /* XXX */
1356 error = p->p_sysent->sv_errtbl[error];
1358 frame->tf_eax = error;
1359 frame->tf_eflags |= PSL_C;
1364 * Traced syscall. trapsignal() is not MP aware.
1366 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1367 MAKEMPSAFE(have_mplock);
1368 frame->tf_eflags &= ~PSL_T;
1369 trapsignal(lp, SIGTRAP, TRAP_TRACE);
1373 * Handle reschedule and other end-of-syscall issues
1375 userret(lp, frame, sticks);
1378 if (KTRPOINT(td, KTR_SYSRET)) {
1379 MAKEMPSAFE(have_mplock);
1380 ktrsysret(lp, code, error, args.sysmsg_result);
1385 * This works because errno is findable through the
1386 * register set. If we ever support an emulation where this
1387 * is not the case, this code will need to be revisited.
1389 STOPEVENT(p, S_SCX, code);
1393 * Release the MP lock if we had to get it
1397 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error);
1399 KASSERT(crit_count == td->td_critcount,
1400 ("syscall: critical section count mismatch! %d/%d",
1401 crit_count, td->td_pri));
1402 KASSERT(&td->td_toks_base == td->td_toks_stop,
1403 ("syscall: extra tokens held after trap! %zd",
1404 td->td_toks_stop - &td->td_toks_base));
1409 * NOTE: MP lock not held at any point.
1412 fork_return(struct lwp *lp, struct trapframe *frame)
1414 frame->tf_eax = 0; /* Child returns zero */
1415 frame->tf_eflags &= ~PSL_C; /* success */
1418 generic_lwp_return(lp, frame);
1419 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1423 * Simplified back end of syscall(), used when returning from fork()
1424 * directly into user mode.
1426 * This code will return back into the fork trampoline code which then
1429 * NOTE: The mplock is not held at any point.
1432 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1434 struct proc *p = lp->lwp_proc;
1437 * Newly forked processes are given a kernel priority. We have to
1438 * adjust the priority to a normal user priority and fake entry
1439 * into the kernel (call userenter()) to install a passive release
1440 * function just in case userret() decides to stop the process. This
1441 * can occur when ^Z races a fork. If we do not install the passive
1442 * release function the current process designation will not be
1443 * released when the thread goes to sleep.
1445 lwkt_setpri_self(TDPRI_USER_NORM);
1446 userenter(lp->lwp_thread, p);
1447 userret(lp, frame, 0);
1449 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1450 ktrsysret(lp, SYS_fork, 0, 0);
1452 lp->lwp_flags |= LWP_PASSIVE_ACQ;
1454 lp->lwp_flags &= ~LWP_PASSIVE_ACQ;
1458 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1459 * fault (which is then passed back to the virtual kernel) if an attempt is
1460 * made to use the FP unit.
1462 * XXX this is a fairly big hack.
1465 set_vkernel_fp(struct trapframe *frame)
1467 struct thread *td = curthread;
1469 if (frame->tf_xflags & PGEX_FPFAULT) {
1470 td->td_pcb->pcb_flags |= FP_VIRTFP;
1471 if (mdcpu->gd_npxthread == td)
1474 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1479 * Called from vkernel_trap() to fixup the vkernel's syscall
1480 * frame for vmspace_ctl() return.
1483 cpu_vkernel_trap(struct trapframe *frame, int error)
1485 frame->tf_eax = error;
1487 frame->tf_eflags |= PSL_C;
1489 frame->tf_eflags &= ~PSL_C;