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>
114 #define MAKEMPSAFE(have_mplock) \
115 if (have_mplock == 0) { \
122 #define MAKEMPSAFE(have_mplock)
126 int (*pmath_emulate) (struct trapframe *);
128 extern void trap (struct trapframe *frame);
129 extern void syscall2 (struct trapframe *frame);
131 static int trap_pfault (struct trapframe *, int, vm_offset_t);
132 static void trap_fatal (struct trapframe *, vm_offset_t);
133 void dblfault_handler (void);
135 extern inthand_t IDTVEC(syscall);
137 #define MAX_TRAP_MSG 28
138 static char *trap_msg[] = {
140 "privileged instruction fault", /* 1 T_PRIVINFLT */
142 "breakpoint instruction fault", /* 3 T_BPTFLT */
145 "arithmetic trap", /* 6 T_ARITHTRAP */
146 "system forced exception", /* 7 T_ASTFLT */
148 "general protection fault", /* 9 T_PROTFLT */
149 "trace trap", /* 10 T_TRCTRAP */
151 "page fault", /* 12 T_PAGEFLT */
153 "alignment fault", /* 14 T_ALIGNFLT */
157 "integer divide fault", /* 18 T_DIVIDE */
158 "non-maskable interrupt trap", /* 19 T_NMI */
159 "overflow trap", /* 20 T_OFLOW */
160 "FPU bounds check fault", /* 21 T_BOUND */
161 "FPU device not available", /* 22 T_DNA */
162 "double fault", /* 23 T_DOUBLEFLT */
163 "FPU operand fetch fault", /* 24 T_FPOPFLT */
164 "invalid TSS fault", /* 25 T_TSSFLT */
165 "segment not present fault", /* 26 T_SEGNPFLT */
166 "stack fault", /* 27 T_STKFLT */
167 "machine check trap", /* 28 T_MCHK */
170 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
171 extern int has_f00f_bug;
175 static int ddb_on_nmi = 1;
176 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
177 &ddb_on_nmi, 0, "Go to DDB on NMI");
179 static int panic_on_nmi = 1;
180 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
181 &panic_on_nmi, 0, "Panic on NMI");
182 static int fast_release;
183 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
184 &fast_release, 0, "Passive Release was optimal");
185 static int slow_release;
186 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
187 &slow_release, 0, "Passive Release was nonoptimal");
189 MALLOC_DEFINE(M_SYSMSG, "sysmsg", "sysmsg structure");
190 extern int max_sysmsg;
193 * Passively intercepts the thread switch function to increase the thread
194 * priority from a user priority to a kernel priority, reducing
195 * syscall and trap overhead for the case where no switch occurs.
197 * Synchronizes td_ucred with p_ucred. This is used by system calls,
198 * signal handling, faults, AST traps, and anything else that enters the
199 * kernel from userland and provides the kernel with a stable read-only
200 * copy of the process ucred.
203 userenter(struct thread *curtd, struct proc *curp)
208 curtd->td_release = lwkt_passive_release;
210 if (curtd->td_ucred != curp->p_ucred) {
211 ncred = crhold(curp->p_ucred);
212 ocred = curtd->td_ucred;
213 curtd->td_ucred = ncred;
221 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
222 * must be completed before we can return to or try to return to userland.
224 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
225 * arithmatic on the delta calculation so the absolute tick values are
226 * truncated to an integer.
229 userret(struct lwp *lp, struct trapframe *frame, int sticks)
231 struct proc *p = lp->lwp_proc;
235 if (p->p_userret != NULL) {
242 * Charge system time if profiling. Note: times are in microseconds.
243 * This may do a copyout and block, so do it first even though it
244 * means some system time will be charged as user time.
246 if (p->p_flags & P_PROFIL) {
247 addupc_task(p, frame->tf_eip,
248 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
253 * If the jungle wants us dead, so be it.
255 if (lp->lwp_mpflags & LWP_MP_WEXIT) {
256 lwkt_gettoken(&p->p_token);
258 lwkt_reltoken(&p->p_token); /* NOT REACHED */
262 * Block here if we are in a stopped state.
264 if (p->p_stat == SSTOP || dump_stop_usertds) {
265 lwkt_gettoken(&p->p_token);
267 lwkt_reltoken(&p->p_token);
272 * Post any pending upcalls. If running a virtual kernel be sure
273 * to restore the virtual kernel's vmspace before posting the upcall.
275 if (p->p_flags & (P_SIGVTALRM | P_SIGPROF | P_UPCALLPEND)) {
276 lwkt_gettoken(&p->p_token);
277 if (p->p_flags & P_SIGVTALRM) {
278 p->p_flags &= ~P_SIGVTALRM;
279 ksignal(p, SIGVTALRM);
281 if (p->p_flags & P_SIGPROF) {
282 p->p_flags &= ~P_SIGPROF;
285 if (p->p_flags & P_UPCALLPEND) {
286 p->p_flags &= ~P_UPCALLPEND;
289 lwkt_reltoken(&p->p_token);
294 * Post any pending signals. If running a virtual kernel be sure
295 * to restore the virtual kernel's vmspace before posting the signal.
297 * WARNING! postsig() can exit and not return.
299 if ((sig = CURSIG_TRACE(lp)) != 0) {
300 lwkt_gettoken(&p->p_token);
302 lwkt_reltoken(&p->p_token);
307 * block here if we are swapped out, but still process signals
308 * (such as SIGKILL). proc0 (the swapin scheduler) is already
309 * aware of our situation, we do not have to wake it up.
311 if (p->p_flags & P_SWAPPEDOUT) {
312 lwkt_gettoken(&p->p_token);
314 p->p_flags |= P_SWAPWAIT;
316 if (p->p_flags & P_SWAPWAIT)
317 tsleep(p, PCATCH, "SWOUT", 0);
318 p->p_flags &= ~P_SWAPWAIT;
320 lwkt_reltoken(&p->p_token);
325 * Make sure postsig() handled request to restore old signal mask after
326 * running signal handler.
328 KKASSERT((lp->lwp_flags & LWP_OLDMASK) == 0);
332 * Cleanup from userenter and any passive release that might have occured.
333 * We must reclaim the current-process designation before we can return
334 * to usermode. We also handle both LWKT and USER reschedule requests.
337 userexit(struct lwp *lp)
339 struct thread *td = lp->lwp_thread;
340 /* globaldata_t gd = td->td_gd; */
343 * Handle stop requests at kernel priority. Any requests queued
344 * after this loop will generate another AST.
346 while (lp->lwp_proc->p_stat == SSTOP) {
347 lwkt_gettoken(&lp->lwp_proc->p_token);
349 lwkt_reltoken(&lp->lwp_proc->p_token);
353 * Reduce our priority in preparation for a return to userland. If
354 * our passive release function was still in place, our priority was
355 * never raised and does not need to be reduced.
357 lwkt_passive_recover(td);
360 * Become the current user scheduled process if we aren't already,
361 * and deal with reschedule requests and other factors.
363 lp->lwp_proc->p_usched->acquire_curproc(lp);
364 /* WARNING: we may have migrated cpu's */
365 /* gd = td->td_gd; */
368 #if !defined(KTR_KERNENTRY)
369 #define KTR_KERNENTRY KTR_ALL
371 KTR_INFO_MASTER(kernentry);
372 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0,
373 "TRAP(pid %d, tid %d, trapno %d, eva %lu)",
374 pid_t pid, lwpid_t tid, register_t trapno, vm_offset_t eva);
375 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "TRAP_RET(pid %d, tid %d)",
376 pid_t pid, lwpid_t tid);
377 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "SYSC(pid %d, tid %d, nr %d)",
378 pid_t pid, lwpid_t tid, register_t trapno);
379 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "SYSRET(pid %d, tid %d, err %d)",
380 pid_t pid, lwpid_t tid, int err);
381 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "FORKRET(pid %d, tid %d)",
382 pid_t pid, lwpid_t tid);
385 * Exception, fault, and trap interface to the kernel.
386 * This common code is called from assembly language IDT gate entry
387 * routines that prepare a suitable stack frame, and restore this
388 * frame after the exception has been processed.
390 * This function is also called from doreti in an interlock to handle ASTs.
391 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
393 * NOTE! We have to retrieve the fault address prior to obtaining the
394 * MP lock because get_mplock() may switch out. YYY cr2 really ought
395 * to be retrieved by the assembly code, not here.
397 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
398 * if an attempt is made to switch from a fast interrupt or IPI. This is
399 * necessary to properly take fatal kernel traps on SMP machines if
400 * get_mplock() has to block.
404 trap(struct trapframe *frame)
406 struct globaldata *gd = mycpu;
407 struct thread *td = gd->gd_curthread;
408 struct lwp *lp = td->td_lwp;
411 int i = 0, ucode = 0, type, code;
416 int crit_count = td->td_critcount;
417 lwkt_tokref_t curstop = td->td_toks_stop;
424 * We need to allow T_DNA faults when the debugger is active since
425 * some dumping paths do large bcopy() which use the floating
426 * point registers for faster copying.
428 if (db_active && frame->tf_trapno != T_DNA) {
429 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
430 ++gd->gd_trap_nesting_level;
431 MAKEMPSAFE(have_mplock);
432 trap_fatal(frame, eva);
433 --gd->gd_trap_nesting_level;
439 ++gd->gd_trap_nesting_level;
440 if (frame->tf_trapno == T_PAGEFLT) {
442 * For some Cyrix CPUs, %cr2 is clobbered by interrupts.
443 * This problem is worked around by using an interrupt
444 * gate for the pagefault handler. We are finally ready
445 * to read %cr2 and then must reenable interrupts.
447 * XXX this should be in the switch statement, but the
448 * NO_FOOF_HACK and VM86 goto and ifdefs obfuscate the
449 * flow of control too much for this to be obviously
456 --gd->gd_trap_nesting_level;
458 if (!(frame->tf_eflags & PSL_I)) {
460 * Buggy application or kernel code has disabled interrupts
461 * and then trapped. Enabling interrupts now is wrong, but
462 * it is better than running with interrupts disabled until
463 * they are accidentally enabled later.
465 type = frame->tf_trapno;
466 if (ISPL(frame->tf_cs)==SEL_UPL || (frame->tf_eflags & PSL_VM)) {
467 MAKEMPSAFE(have_mplock);
469 "pid %ld (%s): trap %d with interrupts disabled\n",
470 (long)curproc->p_pid, curproc->p_comm, type);
471 } else if (type != T_BPTFLT && type != T_TRCTRAP) {
473 * XXX not quite right, since this may be for a
474 * multiple fault in user mode.
476 MAKEMPSAFE(have_mplock);
477 kprintf("kernel trap %d with interrupts disabled\n",
483 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
486 type = frame->tf_trapno;
487 code = frame->tf_err;
490 if (frame->tf_eflags & PSL_VM &&
491 (type == T_PROTFLT || type == T_STKFLT)) {
493 KKASSERT(get_mplock_count(curthread) > 0);
495 i = vm86_emulate((struct vm86frame *)frame);
497 KKASSERT(get_mplock_count(curthread) > 0);
501 * returns to original process
504 vm86_trap((struct vm86frame *)frame,
507 vm86_trap((struct vm86frame *)frame, 0);
509 KKASSERT(0); /* NOT REACHED */
515 * these traps want either a process context, or
516 * assume a normal userspace trap.
520 trap_fatal(frame, eva);
523 type = T_BPTFLT; /* kernel breakpoint */
526 goto kernel_trap; /* normal kernel trap handling */
529 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
532 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid,
533 frame->tf_trapno, eva);
537 sticks = (int)td->td_sticks;
538 lp->lwp_md.md_regs = frame;
541 case T_PRIVINFLT: /* privileged instruction fault */
546 case T_BPTFLT: /* bpt instruction fault */
547 case T_TRCTRAP: /* trace trap */
548 frame->tf_eflags &= ~PSL_T;
550 ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
553 case T_ARITHTRAP: /* arithmetic trap */
558 case T_ASTFLT: /* Allow process switch */
559 mycpu->gd_cnt.v_soft++;
560 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
561 atomic_clear_int(&mycpu->gd_reqflags,
563 addupc_task(p, p->p_prof.pr_addr,
569 * The following two traps can happen in
570 * vm86 mode, and, if so, we want to handle
573 case T_PROTFLT: /* general protection fault */
574 case T_STKFLT: /* stack fault */
575 if (frame->tf_eflags & PSL_VM) {
576 i = vm86_emulate((struct vm86frame *)frame);
582 ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR;
584 case T_SEGNPFLT: /* segment not present fault */
588 case T_TSSFLT: /* invalid TSS fault */
589 case T_DOUBLEFLT: /* double fault */
595 case T_PAGEFLT: /* page fault */
596 i = trap_pfault(frame, TRUE, eva);
599 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
614 case T_DIVIDE: /* integer divide fault */
621 MAKEMPSAFE(have_mplock);
623 goto handle_powerfail;
624 #else /* !POWERFAIL_NMI */
625 /* machine/parity/power fail/"kitchen sink" faults */
626 if (isa_nmi(code) == 0) {
629 * NMI can be hooked up to a pushbutton
633 kprintf ("NMI ... going to debugger\n");
634 kdb_trap (type, 0, frame);
638 } else if (panic_on_nmi)
639 panic("NMI indicates hardware failure");
641 #endif /* POWERFAIL_NMI */
642 #endif /* NISA > 0 */
644 case T_OFLOW: /* integer overflow fault */
649 case T_BOUND: /* bounds check fault */
656 * Virtual kernel intercept - pass the DNA exception
657 * to the virtual kernel if it asked to handle it.
658 * This occurs when the virtual kernel is holding
659 * onto the FP context for a different emulated
660 * process then the one currently running.
662 * We must still call npxdna() since we may have
663 * saved FP state that the virtual kernel needs
664 * to hand over to a different emulated process.
666 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
667 (td->td_pcb->pcb_flags & FP_VIRTFP)
675 * The kernel may have switched out the FP unit's
676 * state, causing the user process to take a fault
677 * when it tries to use the FP unit. Restore the
683 if (!pmath_emulate) {
685 ucode = FPE_FPU_NP_TRAP;
688 i = (*pmath_emulate)(frame);
690 if (!(frame->tf_eflags & PSL_T))
692 frame->tf_eflags &= ~PSL_T;
695 /* else ucode = emulator_only_knows() XXX */
698 case T_FPOPFLT: /* FPU operand fetch fault */
703 case T_XMMFLT: /* SIMD floating-point exception */
713 case T_PAGEFLT: /* page fault */
714 trap_pfault(frame, FALSE, eva);
720 * The kernel may be using npx for copying or other
728 case T_PROTFLT: /* general protection fault */
729 case T_SEGNPFLT: /* segment not present fault */
731 * Invalid segment selectors and out of bounds
732 * %eip's and %esp's can be set up in user mode.
733 * This causes a fault in kernel mode when the
734 * kernel tries to return to user mode. We want
735 * to get this fault so that we can fix the
736 * problem here and not have to check all the
737 * selectors and pointers when the user changes
740 #define MAYBE_DORETI_FAULT(where, whereto) \
742 if (frame->tf_eip == (int)where) { \
743 frame->tf_eip = (int)whereto; \
747 if (mycpu->gd_intr_nesting_level == 0) {
749 * Invalid %fs's and %gs's can be created using
750 * procfs or PT_SETREGS or by invalidating the
751 * underlying LDT entry. This causes a fault
752 * in kernel mode when the kernel attempts to
753 * switch contexts. Lose the bad context
754 * (XXX) so that we can continue, and generate
757 MAYBE_DORETI_FAULT(doreti_iret,
759 MAYBE_DORETI_FAULT(doreti_popl_ds,
760 doreti_popl_ds_fault);
761 MAYBE_DORETI_FAULT(doreti_popl_es,
762 doreti_popl_es_fault);
763 MAYBE_DORETI_FAULT(doreti_popl_fs,
764 doreti_popl_fs_fault);
765 MAYBE_DORETI_FAULT(doreti_popl_gs,
766 doreti_popl_gs_fault);
767 if (td->td_pcb->pcb_onfault) {
769 (register_t)td->td_pcb->pcb_onfault;
777 * PSL_NT can be set in user mode and isn't cleared
778 * automatically when the kernel is entered. This
779 * causes a TSS fault when the kernel attempts to
780 * `iret' because the TSS link is uninitialized. We
781 * want to get this fault so that we can fix the
782 * problem here and not every time the kernel is
785 if (frame->tf_eflags & PSL_NT) {
786 frame->tf_eflags &= ~PSL_NT;
791 case T_TRCTRAP: /* trace trap */
792 if (frame->tf_eip == (int)IDTVEC(syscall)) {
794 * We've just entered system mode via the
795 * syscall lcall. Continue single stepping
796 * silently until the syscall handler has
801 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
803 * The syscall handler has now saved the
804 * flags. Stop single stepping it.
806 frame->tf_eflags &= ~PSL_T;
810 * Ignore debug register trace traps due to
811 * accesses in the user's address space, which
812 * can happen under several conditions such as
813 * if a user sets a watchpoint on a buffer and
814 * then passes that buffer to a system call.
815 * We still want to get TRCTRAPS for addresses
816 * in kernel space because that is useful when
817 * debugging the kernel.
819 if (user_dbreg_trap()) {
821 * Reset breakpoint bits because the
824 load_dr6(rdr6() & 0xfffffff0);
828 * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
832 * If DDB is enabled, let it handle the debugger trap.
833 * Otherwise, debugger traps "can't happen".
837 MAKEMPSAFE(have_mplock);
838 if (kdb_trap (type, 0, frame))
845 MAKEMPSAFE(have_mplock);
848 # define TIMER_FREQ 1193182
852 static unsigned lastalert = 0;
854 if(time_second - lastalert > 10)
856 log(LOG_WARNING, "NMI: power fail\n");
857 sysbeep(TIMER_FREQ/880, hz);
858 lastalert = time_second;
863 #else /* !POWERFAIL_NMI */
864 /* machine/parity/power fail/"kitchen sink" faults */
865 if (isa_nmi(code) == 0) {
868 * NMI can be hooked up to a pushbutton
872 kprintf ("NMI ... going to debugger\n");
873 kdb_trap (type, 0, frame);
877 } else if (panic_on_nmi == 0)
880 #endif /* POWERFAIL_NMI */
881 #endif /* NISA > 0 */
884 MAKEMPSAFE(have_mplock);
885 trap_fatal(frame, eva);
890 * Virtual kernel intercept - if the fault is directly related to a
891 * VM context managed by a virtual kernel then let the virtual kernel
894 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
895 vkernel_trap(lp, frame);
899 /* Translate fault for emulators (e.g. Linux) */
900 if (*p->p_sysent->sv_transtrap)
901 i = (*p->p_sysent->sv_transtrap)(i, type);
903 MAKEMPSAFE(have_mplock);
904 trapsignal(lp, i, ucode);
907 if (type <= MAX_TRAP_MSG) {
908 uprintf("fatal process exception: %s",
910 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
911 uprintf(", fault VA = 0x%lx", (u_long)eva);
917 userret(lp, frame, sticks);
924 if (p != NULL && lp != NULL)
925 KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid);
927 KASSERT(crit_count == td->td_critcount,
928 ("trap: critical section count mismatch! %d/%d",
929 crit_count, td->td_pri));
930 KASSERT(curstop == td->td_toks_stop,
931 ("trap: extra tokens held after trap! %zd/%zd",
932 curstop - &td->td_toks_base,
933 td->td_toks_stop - &td->td_toks_base));
938 trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
941 struct vmspace *vm = NULL;
946 thread_t td = curthread;
947 struct lwp *lp = td->td_lwp;
949 va = trunc_page(eva);
950 if (va >= KERNBASE) {
952 * Don't allow user-mode faults in kernel address space.
953 * An exception: if the faulting address is the invalid
954 * instruction entry in the IDT, then the Intel Pentium
955 * F00F bug workaround was triggered, and we need to
956 * treat it is as an illegal instruction, and not a page
959 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
960 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) {
961 frame->tf_trapno = T_PRIVINFLT;
971 * This is a fault on non-kernel virtual memory.
972 * vm is initialized above to NULL. If curproc is NULL
973 * or curproc->p_vmspace is NULL the fault is fatal.
976 vm = lp->lwp_vmspace;
984 if (frame->tf_err & PGEX_W)
985 ftype = VM_PROT_WRITE;
987 ftype = VM_PROT_READ;
989 if (map != &kernel_map) {
991 * Keep swapout from messing with us during this
1001 fault_flags |= VM_FAULT_BURST;
1002 if (ftype & VM_PROT_WRITE)
1003 fault_flags |= VM_FAULT_DIRTY;
1005 fault_flags |= VM_FAULT_NORMAL;
1006 rv = vm_fault(map, va, ftype, fault_flags);
1007 PRELE(lp->lwp_proc);
1010 * Don't have to worry about process locking or stacks in the
1013 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
1015 if (rv == KERN_SUCCESS)
1019 if (td->td_gd->gd_intr_nesting_level == 0 &&
1020 td->td_pcb->pcb_onfault) {
1021 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
1024 trap_fatal(frame, eva);
1028 /* kludge to pass faulting virtual address to sendsig */
1029 frame->tf_xflags = frame->tf_err;
1030 frame->tf_err = eva;
1032 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
1036 trap_fatal(struct trapframe *frame, vm_offset_t eva)
1038 int code, type, ss, esp;
1039 struct soft_segment_descriptor softseg;
1041 code = frame->tf_err;
1042 type = frame->tf_trapno;
1043 sdtossd(&gdt[mycpu->gd_cpuid * NGDT + IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
1045 if (type <= MAX_TRAP_MSG)
1046 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
1047 type, trap_msg[type],
1048 frame->tf_eflags & PSL_VM ? "vm86" :
1049 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);
1055 if (type == T_PAGEFLT) {
1056 kprintf("fault virtual address = %p\n", (void *)eva);
1057 kprintf("fault code = %s %s, %s\n",
1058 code & PGEX_U ? "user" : "supervisor",
1059 code & PGEX_W ? "write" : "read",
1060 code & PGEX_P ? "protection violation" : "page not present");
1062 kprintf("instruction pointer = 0x%x:0x%x\n",
1063 frame->tf_cs & 0xffff, frame->tf_eip);
1064 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
1065 ss = frame->tf_ss & 0xffff;
1066 esp = frame->tf_esp;
1068 ss = GSEL(GDATA_SEL, SEL_KPL);
1069 esp = (int)&frame->tf_esp;
1071 kprintf("stack pointer = 0x%x:0x%x\n", ss, esp);
1072 kprintf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
1073 kprintf("code segment = base 0x%x, limit 0x%x, type 0x%x\n",
1074 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
1075 kprintf(" = DPL %d, pres %d, def32 %d, gran %d\n",
1076 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
1078 kprintf("processor eflags = ");
1079 if (frame->tf_eflags & PSL_T)
1080 kprintf("trace trap, ");
1081 if (frame->tf_eflags & PSL_I)
1082 kprintf("interrupt enabled, ");
1083 if (frame->tf_eflags & PSL_NT)
1084 kprintf("nested task, ");
1085 if (frame->tf_eflags & PSL_RF)
1086 kprintf("resume, ");
1087 if (frame->tf_eflags & PSL_VM)
1089 kprintf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
1090 kprintf("current process = ");
1092 kprintf("%lu (%s)\n",
1093 (u_long)curproc->p_pid, curproc->p_comm ?
1094 curproc->p_comm : "");
1098 kprintf("current thread = pri %d ", curthread->td_pri);
1099 if (curthread->td_critcount)
1105 * we probably SHOULD have stopped the other CPUs before now!
1106 * another CPU COULD have been touching cpl at this moment...
1108 kprintf(" <- SMP: XXX");
1117 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1120 kprintf("trap number = %d\n", type);
1121 if (type <= MAX_TRAP_MSG)
1122 panic("%s", trap_msg[type]);
1124 panic("unknown/reserved trap");
1128 * Double fault handler. Called when a fault occurs while writing
1129 * a frame for a trap/exception onto the stack. This usually occurs
1130 * when the stack overflows (such is the case with infinite recursion,
1133 * XXX Note that the current PTD gets replaced by IdlePTD when the
1134 * task switch occurs. This means that the stack that was active at
1135 * the time of the double fault is not available at <kstack> unless
1136 * the machine was idle when the double fault occurred. The downside
1137 * of this is that "trace <ebp>" in ddb won't work.
1141 in_kstack_guard(register_t rptr)
1143 thread_t td = curthread;
1145 if ((char *)rptr >= td->td_kstack &&
1146 (char *)rptr < td->td_kstack + PAGE_SIZE) {
1153 dblfault_handler(void)
1155 struct mdglobaldata *gd = mdcpu;
1157 if (in_kstack_guard(gd->gd_common_tss.tss_esp) ||
1158 in_kstack_guard(gd->gd_common_tss.tss_ebp)) {
1159 kprintf("DOUBLE FAULT - KERNEL STACK GUARD HIT!\n");
1161 kprintf("DOUBLE FAULT:\n");
1163 kprintf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1164 kprintf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1165 kprintf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
1167 /* three separate prints in case of a trap on an unmapped page */
1168 kprintf("cpuid = %d; ", gd->mi.gd_cpuid);
1169 kprintf("lapic.id = %08x\n", lapic->id);
1171 panic("double fault");
1175 * syscall2 - MP aware system call request C handler
1177 * A system call is essentially treated as a trap. The MP lock is not
1178 * held on entry or return. We are responsible for handling ASTs
1179 * (e.g. a task switch) prior to return.
1184 syscall2(struct trapframe *frame)
1186 struct thread *td = curthread;
1187 struct proc *p = td->td_proc;
1188 struct lwp *lp = td->td_lwp;
1190 struct sysent *callp;
1191 register_t orig_tf_eflags;
1196 int crit_count = td->td_critcount;
1199 int have_mplock = 0;
1202 union sysunion args;
1205 if (ISPL(frame->tf_cs) != SEL_UPL) {
1212 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid,
1215 userenter(td, p); /* lazy raise our priority */
1220 sticks = (int)td->td_sticks;
1221 orig_tf_eflags = frame->tf_eflags;
1224 * Virtual kernel intercept - if a VM context managed by a virtual
1225 * kernel issues a system call the virtual kernel handles it, not us.
1226 * Restore the virtual kernel context and return from its system
1227 * call. The current frame is copied out to the virtual kernel.
1229 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1230 vkernel_trap(lp, frame);
1231 error = EJUSTRETURN;
1237 * Get the system call parameters and account for time
1239 lp->lwp_md.md_regs = frame;
1240 params = (caddr_t)frame->tf_esp + sizeof(int);
1241 code = frame->tf_eax;
1243 if (p->p_sysent->sv_prepsyscall) {
1244 (*p->p_sysent->sv_prepsyscall)(
1245 frame, (int *)(&args.nosys.sysmsg + 1),
1249 * Need to check if this is a 32 bit or 64 bit syscall.
1250 * fuword is MP aware.
1252 if (code == SYS_syscall) {
1254 * Code is first argument, followed by actual args.
1256 code = fuword(params);
1257 params += sizeof(int);
1258 } else if (code == SYS___syscall) {
1260 * Like syscall, but code is a quad, so as to maintain
1261 * quad alignment for the rest of the arguments.
1263 code = fuword(params);
1264 params += sizeof(quad_t);
1268 code &= p->p_sysent->sv_mask;
1270 if (code >= p->p_sysent->sv_size)
1271 callp = &p->p_sysent->sv_table[0];
1273 callp = &p->p_sysent->sv_table[code];
1275 narg = callp->sy_narg & SYF_ARGMASK;
1278 if (p->p_sysent->sv_name[0] == 'L')
1279 kprintf("Linux syscall, code = %d\n", code);
1283 * copyin is MP aware, but the tracing code is not
1285 if (narg && params) {
1286 error = copyin(params, (caddr_t)(&args.nosys.sysmsg + 1),
1287 narg * sizeof(register_t));
1290 if (KTRPOINT(td, KTR_SYSCALL)) {
1291 MAKEMPSAFE(have_mplock);
1293 ktrsyscall(lp, code, narg,
1294 (void *)(&args.nosys.sysmsg + 1));
1302 if (KTRPOINT(td, KTR_SYSCALL)) {
1303 MAKEMPSAFE(have_mplock);
1304 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1309 * For traditional syscall code edx is left untouched when 32 bit
1310 * results are returned. Since edx is loaded from fds[1] when the
1311 * system call returns we pre-set it here.
1313 args.sysmsg_fds[0] = 0;
1314 args.sysmsg_fds[1] = frame->tf_edx;
1317 * The syscall might manipulate the trap frame. If it does it
1318 * will probably return EJUSTRETURN.
1320 args.sysmsg_frame = frame;
1322 STOPEVENT(p, S_SCE, narg); /* MP aware */
1325 * NOTE: All system calls run MPSAFE now. The system call itself
1326 * is responsible for getting the MP lock.
1328 error = (*callp->sy_call)(&args);
1332 * MP SAFE (we may or may not have the MP lock at this point)
1337 * Reinitialize proc pointer `p' as it may be different
1338 * if this is a child returning from fork syscall.
1341 lp = curthread->td_lwp;
1342 frame->tf_eax = args.sysmsg_fds[0];
1343 frame->tf_edx = args.sysmsg_fds[1];
1344 frame->tf_eflags &= ~PSL_C;
1348 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1349 * int 0x80 is 2 bytes. We saved this in tf_err.
1351 frame->tf_eip -= frame->tf_err;
1356 panic("Unexpected EASYNC return value (for now)");
1359 if (p->p_sysent->sv_errsize) {
1360 if (error >= p->p_sysent->sv_errsize)
1361 error = -1; /* XXX */
1363 error = p->p_sysent->sv_errtbl[error];
1365 frame->tf_eax = error;
1366 frame->tf_eflags |= PSL_C;
1371 * Traced syscall. trapsignal() is not MP aware.
1373 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1374 MAKEMPSAFE(have_mplock);
1375 frame->tf_eflags &= ~PSL_T;
1376 trapsignal(lp, SIGTRAP, TRAP_TRACE);
1380 * Handle reschedule and other end-of-syscall issues
1382 userret(lp, frame, sticks);
1385 if (KTRPOINT(td, KTR_SYSRET)) {
1386 MAKEMPSAFE(have_mplock);
1387 ktrsysret(lp, code, error, args.sysmsg_result);
1392 * This works because errno is findable through the
1393 * register set. If we ever support an emulation where this
1394 * is not the case, this code will need to be revisited.
1396 STOPEVENT(p, S_SCX, code);
1401 * Release the MP lock if we had to get it
1406 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error);
1408 KASSERT(crit_count == td->td_critcount,
1409 ("syscall: critical section count mismatch! %d/%d",
1410 crit_count, td->td_pri));
1411 KASSERT(&td->td_toks_base == td->td_toks_stop,
1412 ("syscall: extra tokens held after trap! %zd",
1413 td->td_toks_stop - &td->td_toks_base));
1418 * NOTE: MP lock not held at any point.
1421 fork_return(struct lwp *lp, struct trapframe *frame)
1423 frame->tf_eax = 0; /* Child returns zero */
1424 frame->tf_eflags &= ~PSL_C; /* success */
1427 generic_lwp_return(lp, frame);
1428 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1432 * Simplified back end of syscall(), used when returning from fork()
1433 * directly into user mode.
1435 * This code will return back into the fork trampoline code which then
1438 * NOTE: The mplock is not held at any point.
1441 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1443 struct proc *p = lp->lwp_proc;
1446 * Newly forked processes are given a kernel priority. We have to
1447 * adjust the priority to a normal user priority and fake entry
1448 * into the kernel (call userenter()) to install a passive release
1449 * function just in case userret() decides to stop the process. This
1450 * can occur when ^Z races a fork. If we do not install the passive
1451 * release function the current process designation will not be
1452 * released when the thread goes to sleep.
1454 lwkt_setpri_self(TDPRI_USER_NORM);
1455 userenter(lp->lwp_thread, p);
1456 userret(lp, frame, 0);
1458 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1459 ktrsysret(lp, SYS_fork, 0, 0);
1461 lp->lwp_flags |= LWP_PASSIVE_ACQ;
1463 lp->lwp_flags &= ~LWP_PASSIVE_ACQ;
1467 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1468 * fault (which is then passed back to the virtual kernel) if an attempt is
1469 * made to use the FP unit.
1471 * XXX this is a fairly big hack.
1474 set_vkernel_fp(struct trapframe *frame)
1476 struct thread *td = curthread;
1478 if (frame->tf_xflags & PGEX_FPFAULT) {
1479 td->td_pcb->pcb_flags |= FP_VIRTFP;
1480 if (mdcpu->gd_npxthread == td)
1483 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1488 * Called from vkernel_trap() to fixup the vkernel's syscall
1489 * frame for vmspace_ctl() return.
1492 cpu_vkernel_trap(struct trapframe *frame, int error)
1494 frame->tf_eax = error;
1496 frame->tf_eflags |= PSL_C;
1498 frame->tf_eflags &= ~PSL_C;