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>
110 #include <sys/spinlock2.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");
182 * Passively intercepts the thread switch function to increase the thread
183 * priority from a user priority to a kernel priority, reducing
184 * syscall and trap overhead for the case where no switch occurs.
186 * Synchronizes td_ucred with p_ucred. This is used by system calls,
187 * signal handling, faults, AST traps, and anything else that enters the
188 * kernel from userland and provides the kernel with a stable read-only
189 * copy of the process ucred.
192 userenter(struct thread *curtd, struct proc *curp)
197 curtd->td_release = lwkt_passive_release;
199 if (curtd->td_ucred != curp->p_ucred) {
200 spin_lock(&curp->p_spin);
201 ncred = crhold(curp->p_ucred);
202 spin_unlock(&curp->p_spin);
203 ocred = curtd->td_ucred;
204 curtd->td_ucred = ncred;
212 * Handle signals, profiling, and other AST's and/or tasks that
213 * must be completed before we can return to or try to return to userland.
215 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
216 * arithmatic on the delta calculation so the absolute tick values are
217 * truncated to an integer.
220 userret(struct lwp *lp, struct trapframe *frame, int sticks)
222 struct proc *p = lp->lwp_proc;
226 if (p->p_userret != NULL) {
233 * Charge system time if profiling. Note: times are in microseconds.
234 * This may do a copyout and block, so do it first even though it
235 * means some system time will be charged as user time.
237 if (p->p_flags & P_PROFIL) {
238 addupc_task(p, frame->tf_eip,
239 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
244 * Specific on-return-to-usermode checks (LWP_MP_WEXIT,
245 * LWP_MP_VNLRU, etc).
247 if (lp->lwp_mpflags & LWP_MP_URETMASK)
251 * Block here if we are in a stopped state.
253 if (p->p_stat == SSTOP || dump_stop_usertds) {
254 lwkt_gettoken(&p->p_token);
256 lwkt_reltoken(&p->p_token);
261 * Post any pending upcalls. If running a virtual kernel be sure
262 * to restore the virtual kernel's vmspace before posting the upcall.
264 if (p->p_flags & (P_SIGVTALRM | P_SIGPROF)) {
265 lwkt_gettoken(&p->p_token);
266 if (p->p_flags & P_SIGVTALRM) {
267 p->p_flags &= ~P_SIGVTALRM;
268 ksignal(p, SIGVTALRM);
270 if (p->p_flags & P_SIGPROF) {
271 p->p_flags &= ~P_SIGPROF;
274 lwkt_reltoken(&p->p_token);
279 * Post any pending signals. If running a virtual kernel be sure
280 * to restore the virtual kernel's vmspace before posting the signal.
282 * WARNING! postsig() can exit and not return.
284 if ((sig = CURSIG_TRACE(lp)) != 0) {
285 lwkt_gettoken(&p->p_token);
287 lwkt_reltoken(&p->p_token);
292 * block here if we are swapped out, but still process signals
293 * (such as SIGKILL). proc0 (the swapin scheduler) is already
294 * aware of our situation, we do not have to wake it up.
296 if (p->p_flags & P_SWAPPEDOUT) {
297 lwkt_gettoken(&p->p_token);
299 p->p_flags |= P_SWAPWAIT;
301 if (p->p_flags & P_SWAPWAIT)
302 tsleep(p, PCATCH, "SWOUT", 0);
303 p->p_flags &= ~P_SWAPWAIT;
305 lwkt_reltoken(&p->p_token);
310 * In a multi-threaded program it is possible for a thread to change
311 * signal state during a system call which temporarily changes the
312 * signal mask. In this case postsig() might not be run and we
313 * have to restore the mask ourselves.
315 if (lp->lwp_flags & LWP_OLDMASK) {
316 lp->lwp_flags &= ~LWP_OLDMASK;
317 lp->lwp_sigmask = lp->lwp_oldsigmask;
323 * Cleanup from userenter and any passive release that might have occured.
324 * We must reclaim the current-process designation before we can return
325 * to usermode. We also handle both LWKT and USER reschedule requests.
328 userexit(struct lwp *lp)
330 struct thread *td = lp->lwp_thread;
331 /* globaldata_t gd = td->td_gd; */
334 * Handle stop requests at kernel priority. Any requests queued
335 * after this loop will generate another AST.
337 while (lp->lwp_proc->p_stat == SSTOP) {
338 lwkt_gettoken(&lp->lwp_proc->p_token);
340 lwkt_reltoken(&lp->lwp_proc->p_token);
344 * Become the current user scheduled process if we aren't already,
345 * and deal with reschedule requests and other factors.
347 lp->lwp_proc->p_usched->acquire_curproc(lp);
348 /* WARNING: we may have migrated cpu's */
349 /* gd = td->td_gd; */
352 * Reduce our priority in preparation for a return to userland. If
353 * our passive release function was still in place, our priority was
354 * never raised and does not need to be reduced.
356 lwkt_passive_recover(td);
359 #if !defined(KTR_KERNENTRY)
360 #define KTR_KERNENTRY KTR_ALL
362 KTR_INFO_MASTER(kernentry);
363 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0,
364 "TRAP(pid %d, tid %d, trapno %d, eva %lu)",
365 pid_t pid, lwpid_t tid, register_t trapno, vm_offset_t eva);
366 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "TRAP_RET(pid %d, tid %d)",
367 pid_t pid, lwpid_t tid);
368 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "SYSC(pid %d, tid %d, nr %d)",
369 pid_t pid, lwpid_t tid, register_t trapno);
370 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "SYSRET(pid %d, tid %d, err %d)",
371 pid_t pid, lwpid_t tid, int err);
372 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "FORKRET(pid %d, tid %d)",
373 pid_t pid, lwpid_t tid);
376 * Exception, fault, and trap interface to the kernel.
377 * This common code is called from assembly language IDT gate entry
378 * routines that prepare a suitable stack frame, and restore this
379 * frame after the exception has been processed.
381 * This function is also called from doreti in an interlock to handle ASTs.
382 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
384 * NOTE! We have to retrieve the fault address prior to obtaining the
385 * MP lock because get_mplock() may switch out. YYY cr2 really ought
386 * to be retrieved by the assembly code, not here.
388 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
389 * if an attempt is made to switch from a fast interrupt or IPI. This is
390 * necessary to properly take fatal kernel traps on SMP machines if
391 * get_mplock() has to block.
395 trap(struct trapframe *frame)
397 struct globaldata *gd = mycpu;
398 struct thread *td = gd->gd_curthread;
399 struct lwp *lp = td->td_lwp;
402 int i = 0, ucode = 0, type, code;
405 int crit_count = td->td_critcount;
406 lwkt_tokref_t curstop = td->td_toks_stop;
413 * We need to allow T_DNA faults when the debugger is active since
414 * some dumping paths do large bcopy() which use the floating
415 * point registers for faster copying.
417 if (db_active && frame->tf_trapno != T_DNA) {
418 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
419 ++gd->gd_trap_nesting_level;
420 MAKEMPSAFE(have_mplock);
421 trap_fatal(frame, eva);
422 --gd->gd_trap_nesting_level;
428 ++gd->gd_trap_nesting_level;
429 if (frame->tf_trapno == T_PAGEFLT) {
431 * For some Cyrix CPUs, %cr2 is clobbered by interrupts.
432 * This problem is worked around by using an interrupt
433 * gate for the pagefault handler. We are finally ready
434 * to read %cr2 and then must reenable interrupts.
436 * XXX this should be in the switch statement, but the
437 * NO_FOOF_HACK and VM86 goto and ifdefs obfuscate the
438 * flow of control too much for this to be obviously
445 --gd->gd_trap_nesting_level;
447 if (!(frame->tf_eflags & PSL_I)) {
449 * Buggy application or kernel code has disabled interrupts
450 * and then trapped. Enabling interrupts now is wrong, but
451 * it is better than running with interrupts disabled until
452 * they are accidentally enabled later.
454 type = frame->tf_trapno;
455 if (ISPL(frame->tf_cs)==SEL_UPL || (frame->tf_eflags & PSL_VM)) {
456 MAKEMPSAFE(have_mplock);
458 "pid %ld (%s): trap %d with interrupts disabled\n",
459 (long)curproc->p_pid, curproc->p_comm, type);
460 } else if (type != T_BPTFLT && type != T_TRCTRAP) {
462 * XXX not quite right, since this may be for a
463 * multiple fault in user mode.
465 MAKEMPSAFE(have_mplock);
466 kprintf("kernel trap %d with interrupts disabled\n",
472 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
475 type = frame->tf_trapno;
476 code = frame->tf_err;
479 if (frame->tf_eflags & PSL_VM &&
480 (type == T_PROTFLT || type == T_STKFLT)) {
481 KKASSERT(get_mplock_count(curthread) > 0);
482 i = vm86_emulate((struct vm86frame *)frame);
483 KKASSERT(get_mplock_count(curthread) > 0);
486 * returns to original process
488 vm86_trap((struct vm86frame *)frame,
490 KKASSERT(0); /* NOT REACHED */
496 * these traps want either a process context, or
497 * assume a normal userspace trap.
501 trap_fatal(frame, eva);
504 type = T_BPTFLT; /* kernel breakpoint */
507 goto kernel_trap; /* normal kernel trap handling */
510 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
513 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid,
514 frame->tf_trapno, eva);
518 sticks = (int)td->td_sticks;
519 lp->lwp_md.md_regs = frame;
522 case T_PRIVINFLT: /* privileged instruction fault */
527 case T_BPTFLT: /* bpt instruction fault */
528 case T_TRCTRAP: /* trace trap */
529 frame->tf_eflags &= ~PSL_T;
531 ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
534 case T_ARITHTRAP: /* arithmetic trap */
539 case T_ASTFLT: /* Allow process switch */
540 mycpu->gd_cnt.v_soft++;
541 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
542 atomic_clear_int(&mycpu->gd_reqflags,
544 addupc_task(p, p->p_prof.pr_addr,
550 * The following two traps can happen in
551 * vm86 mode, and, if so, we want to handle
554 case T_PROTFLT: /* general protection fault */
555 case T_STKFLT: /* stack fault */
556 if (frame->tf_eflags & PSL_VM) {
557 i = vm86_emulate((struct vm86frame *)frame);
563 ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR;
565 case T_SEGNPFLT: /* segment not present fault */
569 case T_TSSFLT: /* invalid TSS fault */
570 case T_DOUBLEFLT: /* double fault */
576 case T_PAGEFLT: /* page fault */
577 i = trap_pfault(frame, TRUE, eva);
580 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
595 case T_DIVIDE: /* integer divide fault */
602 MAKEMPSAFE(have_mplock);
604 goto handle_powerfail;
605 #else /* !POWERFAIL_NMI */
606 /* machine/parity/power fail/"kitchen sink" faults */
607 if (isa_nmi(code) == 0) {
610 * NMI can be hooked up to a pushbutton
614 kprintf ("NMI ... going to debugger\n");
615 kdb_trap (type, 0, frame);
619 } else if (panic_on_nmi)
620 panic("NMI indicates hardware failure");
622 #endif /* POWERFAIL_NMI */
623 #endif /* NISA > 0 */
625 case T_OFLOW: /* integer overflow fault */
630 case T_BOUND: /* bounds check fault */
637 * Virtual kernel intercept - pass the DNA exception
638 * to the virtual kernel if it asked to handle it.
639 * This occurs when the virtual kernel is holding
640 * onto the FP context for a different emulated
641 * process then the one currently running.
643 * We must still call npxdna() since we may have
644 * saved FP state that the virtual kernel needs
645 * to hand over to a different emulated process.
647 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
648 (td->td_pcb->pcb_flags & FP_VIRTFP)
656 * The kernel may have switched out the FP unit's
657 * state, causing the user process to take a fault
658 * when it tries to use the FP unit. Restore the
664 if (!pmath_emulate) {
666 ucode = FPE_FPU_NP_TRAP;
669 i = (*pmath_emulate)(frame);
671 if (!(frame->tf_eflags & PSL_T))
673 frame->tf_eflags &= ~PSL_T;
676 /* else ucode = emulator_only_knows() XXX */
679 case T_FPOPFLT: /* FPU operand fetch fault */
684 case T_XMMFLT: /* SIMD floating-point exception */
694 case T_PAGEFLT: /* page fault */
695 trap_pfault(frame, FALSE, eva);
701 * The kernel may be using npx for copying or other
709 case T_PROTFLT: /* general protection fault */
710 case T_SEGNPFLT: /* segment not present fault */
712 * Invalid segment selectors and out of bounds
713 * %eip's and %esp's can be set up in user mode.
714 * This causes a fault in kernel mode when the
715 * kernel tries to return to user mode. We want
716 * to get this fault so that we can fix the
717 * problem here and not have to check all the
718 * selectors and pointers when the user changes
721 #define MAYBE_DORETI_FAULT(where, whereto) \
723 if (frame->tf_eip == (int)where) { \
724 frame->tf_eip = (int)whereto; \
728 if (mycpu->gd_intr_nesting_level == 0) {
730 * Invalid %fs's and %gs's can be created using
731 * procfs or PT_SETREGS or by invalidating the
732 * underlying LDT entry. This causes a fault
733 * in kernel mode when the kernel attempts to
734 * switch contexts. Lose the bad context
735 * (XXX) so that we can continue, and generate
738 MAYBE_DORETI_FAULT(doreti_iret,
740 MAYBE_DORETI_FAULT(doreti_popl_ds,
741 doreti_popl_ds_fault);
742 MAYBE_DORETI_FAULT(doreti_popl_es,
743 doreti_popl_es_fault);
744 MAYBE_DORETI_FAULT(doreti_popl_fs,
745 doreti_popl_fs_fault);
746 MAYBE_DORETI_FAULT(doreti_popl_gs,
747 doreti_popl_gs_fault);
750 * NOTE: cpu doesn't push esp on kernel trap
752 if (td->td_pcb->pcb_onfault &&
753 td->td_pcb->pcb_onfault_sp ==
754 (int)&frame->tf_esp) {
756 (register_t)td->td_pcb->pcb_onfault;
764 * PSL_NT can be set in user mode and isn't cleared
765 * automatically when the kernel is entered. This
766 * causes a TSS fault when the kernel attempts to
767 * `iret' because the TSS link is uninitialized. We
768 * want to get this fault so that we can fix the
769 * problem here and not every time the kernel is
772 if (frame->tf_eflags & PSL_NT) {
773 frame->tf_eflags &= ~PSL_NT;
778 case T_TRCTRAP: /* trace trap */
779 if (frame->tf_eip == (int)IDTVEC(syscall)) {
781 * We've just entered system mode via the
782 * syscall lcall. Continue single stepping
783 * silently until the syscall handler has
788 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
790 * The syscall handler has now saved the
791 * flags. Stop single stepping it.
793 frame->tf_eflags &= ~PSL_T;
797 * Ignore debug register trace traps due to
798 * accesses in the user's address space, which
799 * can happen under several conditions such as
800 * if a user sets a watchpoint on a buffer and
801 * then passes that buffer to a system call.
802 * We still want to get TRCTRAPS for addresses
803 * in kernel space because that is useful when
804 * debugging the kernel.
806 if (user_dbreg_trap()) {
808 * Reset breakpoint bits because the
811 load_dr6(rdr6() & 0xfffffff0);
815 * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
819 * If DDB is enabled, let it handle the debugger trap.
820 * Otherwise, debugger traps "can't happen".
824 MAKEMPSAFE(have_mplock);
825 if (kdb_trap (type, 0, frame))
832 MAKEMPSAFE(have_mplock);
835 # define TIMER_FREQ 1193182
839 static unsigned lastalert = 0;
841 if (time_uptime - lastalert > 10) {
842 log(LOG_WARNING, "NMI: power fail\n");
843 sysbeep(TIMER_FREQ/880, hz);
844 lastalert = time_uptime;
849 #else /* !POWERFAIL_NMI */
850 /* machine/parity/power fail/"kitchen sink" faults */
851 if (isa_nmi(code) == 0) {
854 * NMI can be hooked up to a pushbutton
858 kprintf ("NMI ... going to debugger\n");
859 kdb_trap (type, 0, frame);
863 } else if (panic_on_nmi == 0)
866 #endif /* POWERFAIL_NMI */
867 #endif /* NISA > 0 */
870 MAKEMPSAFE(have_mplock);
871 trap_fatal(frame, eva);
876 * Virtual kernel intercept - if the fault is directly related to a
877 * VM context managed by a virtual kernel then let the virtual kernel
880 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
881 vkernel_trap(lp, frame);
885 /* Translate fault for emulators (e.g. Linux) */
886 if (*p->p_sysent->sv_transtrap)
887 i = (*p->p_sysent->sv_transtrap)(i, type);
889 MAKEMPSAFE(have_mplock);
890 trapsignal(lp, i, ucode);
893 if (type <= MAX_TRAP_MSG) {
894 uprintf("fatal process exception: %s",
896 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
897 uprintf(", fault VA = 0x%lx", (u_long)eva);
903 userret(lp, frame, sticks);
908 if (p != NULL && lp != NULL)
909 KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid);
911 KASSERT(crit_count == td->td_critcount,
912 ("trap: critical section count mismatch! %d/%d",
913 crit_count, td->td_pri));
914 KASSERT(curstop == td->td_toks_stop,
915 ("trap: extra tokens held after trap! %zd/%zd",
916 curstop - &td->td_toks_base,
917 td->td_toks_stop - &td->td_toks_base));
922 trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
925 struct vmspace *vm = NULL;
930 thread_t td = curthread;
931 struct lwp *lp = td->td_lwp;
933 va = trunc_page(eva);
934 if (va >= KERNBASE) {
936 * Don't allow user-mode faults in kernel address space.
937 * An exception: if the faulting address is the invalid
938 * instruction entry in the IDT, then the Intel Pentium
939 * F00F bug workaround was triggered, and we need to
940 * treat it is as an illegal instruction, and not a page
943 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
944 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) {
945 frame->tf_trapno = T_PRIVINFLT;
955 * This is a fault on non-kernel virtual memory.
956 * vm is initialized above to NULL. If curproc is NULL
957 * or curproc->p_vmspace is NULL the fault is fatal.
960 vm = lp->lwp_vmspace;
968 if (frame->tf_err & PGEX_W)
969 ftype = VM_PROT_WRITE;
971 ftype = VM_PROT_READ;
973 if (map != &kernel_map) {
975 * Keep swapout from messing with us during this
985 fault_flags |= VM_FAULT_BURST;
986 if (ftype & VM_PROT_WRITE)
987 fault_flags |= VM_FAULT_DIRTY;
989 fault_flags |= VM_FAULT_NORMAL;
990 rv = vm_fault(map, va, ftype, fault_flags);
994 * Don't have to worry about process locking or stacks in the
997 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
999 if (rv == KERN_SUCCESS)
1004 * NOTE: cpu doesn't push esp on kernel trap
1006 if (td->td_gd->gd_intr_nesting_level == 0 &&
1007 td->td_pcb->pcb_onfault &&
1008 td->td_pcb->pcb_onfault_sp == (int)&frame->tf_esp) {
1009 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
1012 if (td->td_gd->gd_intr_nesting_level == 0 &&
1013 td->td_pcb->pcb_onfault) {
1014 kprintf("ESP mismatch %p %08x\n",
1015 &frame->tf_esp, td->td_pcb->pcb_onfault_sp);
1017 trap_fatal(frame, eva);
1021 /* kludge to pass faulting virtual address to sendsig */
1022 frame->tf_xflags = frame->tf_err;
1023 frame->tf_err = eva;
1025 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
1029 trap_fatal(struct trapframe *frame, vm_offset_t eva)
1031 int code, type, ss, esp;
1032 struct soft_segment_descriptor softseg;
1034 code = frame->tf_err;
1035 type = frame->tf_trapno;
1036 sdtossd(&gdt[mycpu->gd_cpuid * NGDT + IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
1038 if (type <= MAX_TRAP_MSG)
1039 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
1040 type, trap_msg[type],
1041 frame->tf_eflags & PSL_VM ? "vm86" :
1042 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
1043 /* three separate prints in case of a trap on an unmapped page */
1044 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
1045 kprintf("lapic.id = %08x\n", lapic->id);
1046 if (type == T_PAGEFLT) {
1047 kprintf("fault virtual address = %p\n", (void *)eva);
1048 kprintf("fault code = %s %s, %s\n",
1049 code & PGEX_U ? "user" : "supervisor",
1050 code & PGEX_W ? "write" : "read",
1051 code & PGEX_P ? "protection violation" : "page not present");
1053 kprintf("instruction pointer = 0x%x:0x%x\n",
1054 frame->tf_cs & 0xffff, frame->tf_eip);
1055 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
1056 ss = frame->tf_ss & 0xffff;
1057 esp = frame->tf_esp;
1059 ss = GSEL(GDATA_SEL, SEL_KPL);
1060 esp = (int)&frame->tf_esp;
1062 kprintf("stack pointer = 0x%x:0x%x\n", ss, esp);
1063 kprintf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
1064 kprintf("code segment = base 0x%x, limit 0x%x, type 0x%x\n",
1065 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
1066 kprintf(" = DPL %d, pres %d, def32 %d, gran %d\n",
1067 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
1069 kprintf("processor eflags = ");
1070 if (frame->tf_eflags & PSL_T)
1071 kprintf("trace trap, ");
1072 if (frame->tf_eflags & PSL_I)
1073 kprintf("interrupt enabled, ");
1074 if (frame->tf_eflags & PSL_NT)
1075 kprintf("nested task, ");
1076 if (frame->tf_eflags & PSL_RF)
1077 kprintf("resume, ");
1078 if (frame->tf_eflags & PSL_VM)
1080 kprintf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
1081 kprintf("current process = ");
1083 kprintf("%lu (%s)\n",
1084 (u_long)curproc->p_pid, curproc->p_comm ?
1085 curproc->p_comm : "");
1089 kprintf("current thread = pri %d ", curthread->td_pri);
1090 if (curthread->td_critcount)
1095 * we probably SHOULD have stopped the other CPUs before now!
1096 * another CPU COULD have been touching cpl at this moment...
1098 kprintf(" <- SMP: XXX");
1106 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1109 kprintf("trap number = %d\n", type);
1110 if (type <= MAX_TRAP_MSG)
1111 panic("%s", trap_msg[type]);
1113 panic("unknown/reserved trap");
1117 * Double fault handler. Called when a fault occurs while writing
1118 * a frame for a trap/exception onto the stack. This usually occurs
1119 * when the stack overflows (such is the case with infinite recursion,
1122 * XXX Note that the current PTD gets replaced by IdlePTD when the
1123 * task switch occurs. This means that the stack that was active at
1124 * the time of the double fault is not available at <kstack> unless
1125 * the machine was idle when the double fault occurred. The downside
1126 * of this is that "trace <ebp>" in ddb won't work.
1130 in_kstack_guard(register_t rptr)
1132 thread_t td = curthread;
1134 if ((char *)rptr >= td->td_kstack &&
1135 (char *)rptr < td->td_kstack + PAGE_SIZE) {
1142 dblfault_handler(void)
1144 struct mdglobaldata *gd = mdcpu;
1146 if (in_kstack_guard(gd->gd_common_tss.tss_esp) ||
1147 in_kstack_guard(gd->gd_common_tss.tss_ebp)) {
1148 kprintf("DOUBLE FAULT - KERNEL STACK GUARD HIT!\n");
1150 kprintf("DOUBLE FAULT:\n");
1152 kprintf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1153 kprintf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1154 kprintf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
1155 /* three separate prints in case of a trap on an unmapped page */
1156 kprintf("cpuid = %d; ", gd->mi.gd_cpuid);
1157 kprintf("lapic.id = %08x\n", lapic->id);
1158 panic("double fault");
1162 * syscall2 - MP aware system call request C handler
1164 * A system call is essentially treated as a trap. The MP lock is not
1165 * held on entry or return. We are responsible for handling ASTs
1166 * (e.g. a task switch) prior to return.
1171 syscall2(struct trapframe *frame)
1173 struct thread *td = curthread;
1174 struct proc *p = td->td_proc;
1175 struct lwp *lp = td->td_lwp;
1177 struct sysent *callp;
1178 register_t orig_tf_eflags;
1183 int crit_count = td->td_critcount;
1185 int have_mplock = 0;
1187 union sysunion args;
1190 if (ISPL(frame->tf_cs) != SEL_UPL) {
1197 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid,
1200 userenter(td, p); /* lazy raise our priority */
1205 sticks = (int)td->td_sticks;
1206 orig_tf_eflags = frame->tf_eflags;
1209 * Virtual kernel intercept - if a VM context managed by a virtual
1210 * kernel issues a system call the virtual kernel handles it, not us.
1211 * Restore the virtual kernel context and return from its system
1212 * call. The current frame is copied out to the virtual kernel.
1214 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1215 vkernel_trap(lp, frame);
1216 error = EJUSTRETURN;
1222 * Get the system call parameters and account for time
1224 lp->lwp_md.md_regs = frame;
1225 params = (caddr_t)frame->tf_esp + sizeof(int);
1226 code = frame->tf_eax;
1228 if (p->p_sysent->sv_prepsyscall) {
1229 (*p->p_sysent->sv_prepsyscall)(
1230 frame, (int *)(&args.nosys.sysmsg + 1),
1234 * Need to check if this is a 32 bit or 64 bit syscall.
1235 * fuword is MP aware.
1237 if (code == SYS_syscall) {
1239 * Code is first argument, followed by actual args.
1241 code = fuword(params);
1242 params += sizeof(int);
1243 } else if (code == SYS___syscall) {
1245 * Like syscall, but code is a quad, so as to maintain
1246 * quad alignment for the rest of the arguments.
1248 code = fuword(params);
1249 params += sizeof(quad_t);
1253 code &= p->p_sysent->sv_mask;
1255 if (code >= p->p_sysent->sv_size)
1256 callp = &p->p_sysent->sv_table[0];
1258 callp = &p->p_sysent->sv_table[code];
1260 narg = callp->sy_narg & SYF_ARGMASK;
1263 if (p->p_sysent->sv_name[0] == 'L')
1264 kprintf("Linux syscall, code = %d\n", code);
1268 * copyin is MP aware, but the tracing code is not
1270 if (narg && params) {
1271 error = copyin(params, (caddr_t)(&args.nosys.sysmsg + 1),
1272 narg * sizeof(register_t));
1275 if (KTRPOINT(td, KTR_SYSCALL)) {
1276 MAKEMPSAFE(have_mplock);
1278 ktrsyscall(lp, code, narg,
1279 (void *)(&args.nosys.sysmsg + 1));
1287 if (KTRPOINT(td, KTR_SYSCALL)) {
1288 MAKEMPSAFE(have_mplock);
1289 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1294 * For traditional syscall code edx is left untouched when 32 bit
1295 * results are returned. Since edx is loaded from fds[1] when the
1296 * system call returns we pre-set it here.
1298 args.sysmsg_fds[0] = 0;
1299 args.sysmsg_fds[1] = frame->tf_edx;
1302 * The syscall might manipulate the trap frame. If it does it
1303 * will probably return EJUSTRETURN.
1305 args.sysmsg_frame = frame;
1307 STOPEVENT(p, S_SCE, narg); /* MP aware */
1310 * NOTE: All system calls run MPSAFE now. The system call itself
1311 * is responsible for getting the MP lock.
1313 error = (*callp->sy_call)(&args);
1317 * MP SAFE (we may or may not have the MP lock at this point)
1322 * Reinitialize proc pointer `p' as it may be different
1323 * if this is a child returning from fork syscall.
1326 lp = curthread->td_lwp;
1327 frame->tf_eax = args.sysmsg_fds[0];
1328 frame->tf_edx = args.sysmsg_fds[1];
1329 frame->tf_eflags &= ~PSL_C;
1333 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1334 * int 0x80 is 2 bytes. We saved this in tf_err.
1336 frame->tf_eip -= frame->tf_err;
1341 panic("Unexpected EASYNC return value (for now)");
1344 if (p->p_sysent->sv_errsize) {
1345 if (error >= p->p_sysent->sv_errsize)
1346 error = -1; /* XXX */
1348 error = p->p_sysent->sv_errtbl[error];
1350 frame->tf_eax = error;
1351 frame->tf_eflags |= PSL_C;
1356 * Traced syscall. trapsignal() is not MP aware.
1358 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1359 MAKEMPSAFE(have_mplock);
1360 frame->tf_eflags &= ~PSL_T;
1361 trapsignal(lp, SIGTRAP, TRAP_TRACE);
1365 * Handle reschedule and other end-of-syscall issues
1367 userret(lp, frame, sticks);
1370 if (KTRPOINT(td, KTR_SYSRET)) {
1371 MAKEMPSAFE(have_mplock);
1372 ktrsysret(lp, code, error, args.sysmsg_result);
1377 * This works because errno is findable through the
1378 * register set. If we ever support an emulation where this
1379 * is not the case, this code will need to be revisited.
1381 STOPEVENT(p, S_SCX, code);
1385 * Release the MP lock if we had to get it
1389 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error);
1391 KASSERT(crit_count == td->td_critcount,
1392 ("syscall: critical section count mismatch! %d/%d",
1393 crit_count, td->td_pri));
1394 KASSERT(&td->td_toks_base == td->td_toks_stop,
1395 ("syscall: extra tokens held after trap! %zd",
1396 td->td_toks_stop - &td->td_toks_base));
1401 * NOTE: MP lock not held at any point.
1404 fork_return(struct lwp *lp, struct trapframe *frame)
1406 frame->tf_eax = 0; /* Child returns zero */
1407 frame->tf_eflags &= ~PSL_C; /* success */
1410 generic_lwp_return(lp, frame);
1411 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1415 * Simplified back end of syscall(), used when returning from fork()
1416 * directly into user mode.
1418 * This code will return back into the fork trampoline code which then
1421 * NOTE: The mplock is not held at any point.
1424 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1426 struct proc *p = lp->lwp_proc;
1429 * Newly forked processes are given a kernel priority. We have to
1430 * adjust the priority to a normal user priority and fake entry
1431 * into the kernel (call userenter()) to install a passive release
1432 * function just in case userret() decides to stop the process. This
1433 * can occur when ^Z races a fork. If we do not install the passive
1434 * release function the current process designation will not be
1435 * released when the thread goes to sleep.
1437 lwkt_setpri_self(TDPRI_USER_NORM);
1438 userenter(lp->lwp_thread, p);
1439 userret(lp, frame, 0);
1441 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1442 ktrsysret(lp, SYS_fork, 0, 0);
1444 lp->lwp_flags |= LWP_PASSIVE_ACQ;
1446 lp->lwp_flags &= ~LWP_PASSIVE_ACQ;
1450 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1451 * fault (which is then passed back to the virtual kernel) if an attempt is
1452 * made to use the FP unit.
1454 * XXX this is a fairly big hack.
1457 set_vkernel_fp(struct trapframe *frame)
1459 struct thread *td = curthread;
1461 if (frame->tf_xflags & PGEX_FPFAULT) {
1462 td->td_pcb->pcb_flags |= FP_VIRTFP;
1463 if (mdcpu->gd_npxthread == td)
1466 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1471 * Called from vkernel_trap() to fixup the vkernel's syscall
1472 * frame for vmspace_ctl() return.
1475 cpu_vkernel_trap(struct trapframe *frame, int error)
1477 frame->tf_eax = error;
1479 frame->tf_eflags |= PSL_C;
1481 frame->tf_eflags &= ~PSL_C;