2 * Copyright (C) 1994, David Greenman
3 * Copyright (c) 1990, 1993
4 * The Regents of the University of California. All rights reserved.
6 * This code is derived from software contributed to Berkeley by
7 * the University of Utah, and William Jolitz.
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. All advertising materials mentioning features or use of this software
18 * must display the following acknowledgement:
19 * This product includes software developed by the University of
20 * California, Berkeley and its contributors.
21 * 4. Neither the name of the University nor the names of its contributors
22 * may be used to endorse or promote products derived from this software
23 * without specific prior written permission.
25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 * from: @(#)trap.c 7.4 (Berkeley) 5/13/91
38 * $FreeBSD: src/sys/i386/i386/trap.c,v 1.147.2.11 2003/02/27 19:09:59 luoqi Exp $
39 * $DragonFly: src/sys/platform/pc32/i386/trap.c,v 1.115 2008/09/09 04:06:17 dillon Exp $
43 * 386 Trap and System call handling
51 #include "opt_ktrace.h"
52 #include "opt_clock.h"
55 #include <sys/param.h>
56 #include <sys/systm.h>
58 #include <sys/pioctl.h>
59 #include <sys/kernel.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>
95 #include <machine_base/isa/intr_machdep.h>
98 #include <sys/syslog.h>
99 #include <machine/clock.h>
102 #include <machine/vm86.h>
105 #include <sys/msgport2.h>
106 #include <sys/thread2.h>
110 #define MAKEMPSAFE(have_mplock) \
111 if (have_mplock == 0) { \
118 #define MAKEMPSAFE(have_mplock)
122 int (*pmath_emulate) (struct trapframe *);
124 extern void trap (struct trapframe *frame);
125 extern int trapwrite (unsigned addr);
126 extern void syscall2 (struct trapframe *frame);
128 static int trap_pfault (struct trapframe *, int, vm_offset_t);
129 static void trap_fatal (struct trapframe *, vm_offset_t);
130 void dblfault_handler (void);
132 extern inthand_t IDTVEC(syscall);
134 #define MAX_TRAP_MSG 28
135 static char *trap_msg[] = {
137 "privileged instruction fault", /* 1 T_PRIVINFLT */
139 "breakpoint instruction fault", /* 3 T_BPTFLT */
142 "arithmetic trap", /* 6 T_ARITHTRAP */
143 "system forced exception", /* 7 T_ASTFLT */
145 "general protection fault", /* 9 T_PROTFLT */
146 "trace trap", /* 10 T_TRCTRAP */
148 "page fault", /* 12 T_PAGEFLT */
150 "alignment fault", /* 14 T_ALIGNFLT */
154 "integer divide fault", /* 18 T_DIVIDE */
155 "non-maskable interrupt trap", /* 19 T_NMI */
156 "overflow trap", /* 20 T_OFLOW */
157 "FPU bounds check fault", /* 21 T_BOUND */
158 "FPU device not available", /* 22 T_DNA */
159 "double fault", /* 23 T_DOUBLEFLT */
160 "FPU operand fetch fault", /* 24 T_FPOPFLT */
161 "invalid TSS fault", /* 25 T_TSSFLT */
162 "segment not present fault", /* 26 T_SEGNPFLT */
163 "stack fault", /* 27 T_STKFLT */
164 "machine check trap", /* 28 T_MCHK */
167 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
168 extern int has_f00f_bug;
172 static int ddb_on_nmi = 1;
173 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
174 &ddb_on_nmi, 0, "Go to DDB on NMI");
176 static int panic_on_nmi = 1;
177 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
178 &panic_on_nmi, 0, "Panic on NMI");
179 static int fast_release;
180 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
181 &fast_release, 0, "Passive Release was optimal");
182 static int slow_release;
183 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
184 &slow_release, 0, "Passive Release was nonoptimal");
186 static int syscall_mpsafe = 1;
187 SYSCTL_INT(_kern, OID_AUTO, syscall_mpsafe, CTLFLAG_RW,
188 &syscall_mpsafe, 0, "Allow MPSAFE marked syscalls to run without BGL");
189 TUNABLE_INT("kern.syscall_mpsafe", &syscall_mpsafe);
190 static int trap_mpsafe = 1;
191 SYSCTL_INT(_kern, OID_AUTO, trap_mpsafe, CTLFLAG_RW,
192 &trap_mpsafe, 0, "Allow traps to mostly run without the BGL");
193 TUNABLE_INT("kern.trap_mpsafe", &trap_mpsafe);
196 MALLOC_DEFINE(M_SYSMSG, "sysmsg", "sysmsg structure");
197 extern int max_sysmsg;
200 * Passive USER->KERNEL transition. This only occurs if we block in the
201 * kernel while still holding our userland priority. We have to fixup our
202 * priority in order to avoid potential deadlocks before we allow the system
203 * to switch us to another thread.
206 passive_release(struct thread *td)
208 struct lwp *lp = td->td_lwp;
210 td->td_release = NULL;
211 lwkt_setpri_self(TDPRI_KERN_USER);
212 lp->lwp_proc->p_usched->release_curproc(lp);
216 * userenter() passively intercepts the thread switch function to increase
217 * the thread priority from a user priority to a kernel priority, reducing
218 * syscall and trap overhead for the case where no switch occurs.
222 userenter(struct thread *curtd)
224 curtd->td_release = passive_release;
228 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
229 * must be completed before we can return to or try to return to userland.
231 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
232 * arithmatic on the delta calculation so the absolute tick values are
233 * truncated to an integer.
236 userret(struct lwp *lp, struct trapframe *frame, int sticks)
238 struct proc *p = lp->lwp_proc;
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_flag & 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_flag & LWP_WEXIT) {
258 rel_mplock(); /* NOT REACHED */
262 * Block here if we are in a stopped state.
264 if (p->p_stat == SSTOP) {
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_flag & P_UPCALLPEND) {
276 p->p_flag &= ~P_UPCALLPEND;
284 * Post any pending signals. If running a virtual kernel be sure
285 * to restore the virtual kernel's vmspace before posting the signal.
287 if ((sig = CURSIG(lp)) != 0) {
295 * block here if we are swapped out, but still process signals
296 * (such as SIGKILL). proc0 (the swapin scheduler) is already
297 * aware of our situation, we do not have to wake it up.
299 if (p->p_flag & P_SWAPPEDOUT) {
301 p->p_flag |= P_SWAPWAIT;
303 if (p->p_flag & P_SWAPWAIT)
304 tsleep(p, PCATCH, "SWOUT", 0);
305 p->p_flag &= ~P_SWAPWAIT;
311 * Make sure postsig() handled request to restore old signal mask after
312 * running signal handler.
314 KKASSERT((lp->lwp_flag & LWP_OLDMASK) == 0);
318 * Cleanup from userenter and any passive release that might have occured.
319 * We must reclaim the current-process designation before we can return
320 * to usermode. We also handle both LWKT and USER reschedule requests.
323 userexit(struct lwp *lp)
325 struct thread *td = lp->lwp_thread;
326 globaldata_t gd = td->td_gd;
330 * If a user reschedule is requested force a new process to be
331 * chosen by releasing the current process. Our process will only
332 * be chosen again if it has a considerably better priority.
334 if (user_resched_wanted())
335 lp->lwp_proc->p_usched->release_curproc(lp);
339 * Handle a LWKT reschedule request first. Since our passive release
340 * is still in place we do not have to do anything special. This
341 * will also allow other kernel threads running on behalf of user
342 * mode to force us to release and acquire the current process before
345 while (lwkt_resched_wanted()) {
349 * The thread that preempted us may have stopped our process.
351 while (lp->lwp_proc->p_stat == SSTOP) {
359 * Acquire the current process designation for this user scheduler
360 * on this cpu. This will also handle any user-reschedule requests.
362 * If we never blocked we never released and this function is
363 * typically a nop. However, if a user reschedule was requested
364 * this function may allow another user process to run before
367 lp->lwp_proc->p_usched->acquire_curproc(lp);
368 /* We may have switched cpus on acquisition */
372 * Reduce our priority in preparation for a return to userland. If
373 * our passive release function was still in place, our priority was
374 * never raised and does not need to be reduced.
376 * Note that at this point there may be other LWKT thread at
377 * TDPRI_KERN_USER (aka higher then our current priority). We
378 * do NOT want to run these threads yet.
380 if (td->td_release == NULL)
381 lwkt_setpri_self(TDPRI_USER_NORM);
382 td->td_release = NULL;
385 #if !defined(KTR_KERNENTRY)
386 #define KTR_KERNENTRY KTR_ALL
388 KTR_INFO_MASTER(kernentry);
389 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, "pid=%d, tid=%d, trapno=%d, eva=%p",
390 sizeof(int) + sizeof(int) + sizeof(int) + sizeof(vm_offset_t));
391 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "pid=%d, tid=%d",
392 sizeof(int) + sizeof(int));
393 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "pid=%d, tid=%d, call=%d",
394 sizeof(int) + sizeof(int) + sizeof(int));
395 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "pid=%d, tid=%d, err=%d",
396 sizeof(int) + sizeof(int) + sizeof(int));
397 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "pid=%d, tid=%d",
398 sizeof(int) + sizeof(int));
401 * Exception, fault, and trap interface to the kernel.
402 * This common code is called from assembly language IDT gate entry
403 * routines that prepare a suitable stack frame, and restore this
404 * frame after the exception has been processed.
406 * This function is also called from doreti in an interlock to handle ASTs.
407 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
409 * NOTE! We have to retrieve the fault address prior to obtaining the
410 * MP lock because get_mplock() may switch out. YYY cr2 really ought
411 * to be retrieved by the assembly code, not here.
413 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
414 * if an attempt is made to switch from a fast interrupt or IPI. This is
415 * necessary to properly take fatal kernel traps on SMP machines if
416 * get_mplock() has to block.
420 trap(struct trapframe *frame)
422 struct globaldata *gd = mycpu;
423 struct thread *td = gd->gd_curthread;
424 struct lwp *lp = td->td_lwp;
427 int i = 0, ucode = 0, type, code;
432 int crit_count = td->td_pri & ~TDPRI_MASK;
439 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
440 ++gd->gd_trap_nesting_level;
441 MAKEMPSAFE(have_mplock);
442 trap_fatal(frame, eva);
443 --gd->gd_trap_nesting_level;
449 ++gd->gd_trap_nesting_level;
450 if (frame->tf_trapno == T_PAGEFLT) {
452 * For some Cyrix CPUs, %cr2 is clobbered by interrupts.
453 * This problem is worked around by using an interrupt
454 * gate for the pagefault handler. We are finally ready
455 * to read %cr2 and then must reenable interrupts.
457 * XXX this should be in the switch statement, but the
458 * NO_FOOF_HACK and VM86 goto and ifdefs obfuscate the
459 * flow of control too much for this to be obviously
467 if (trap_mpsafe == 0)
468 MAKEMPSAFE(have_mplock);
471 --gd->gd_trap_nesting_level;
473 if (!(frame->tf_eflags & PSL_I)) {
475 * Buggy application or kernel code has disabled interrupts
476 * and then trapped. Enabling interrupts now is wrong, but
477 * it is better than running with interrupts disabled until
478 * they are accidentally enabled later.
480 type = frame->tf_trapno;
481 if (ISPL(frame->tf_cs)==SEL_UPL || (frame->tf_eflags & PSL_VM)) {
482 MAKEMPSAFE(have_mplock);
484 "pid %ld (%s): trap %d with interrupts disabled\n",
485 (long)curproc->p_pid, curproc->p_comm, type);
486 } else if (type != T_BPTFLT && type != T_TRCTRAP) {
488 * XXX not quite right, since this may be for a
489 * multiple fault in user mode.
491 MAKEMPSAFE(have_mplock);
492 kprintf("kernel trap %d with interrupts disabled\n",
498 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
501 type = frame->tf_trapno;
502 code = frame->tf_err;
505 if (frame->tf_eflags & PSL_VM &&
506 (type == T_PROTFLT || type == T_STKFLT)) {
508 KKASSERT(td->td_mpcount > 0);
510 i = vm86_emulate((struct vm86frame *)frame);
512 KKASSERT(td->td_mpcount > 0);
516 * returns to original process
519 vm86_trap((struct vm86frame *)frame,
522 vm86_trap((struct vm86frame *)frame, 0);
524 KKASSERT(0); /* NOT REACHED */
530 * these traps want either a process context, or
531 * assume a normal userspace trap.
535 trap_fatal(frame, eva);
538 type = T_BPTFLT; /* kernel breakpoint */
541 goto kernel_trap; /* normal kernel trap handling */
544 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
547 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid,
548 frame->tf_trapno, eva);
552 sticks = (int)td->td_sticks;
553 lp->lwp_md.md_regs = frame;
556 case T_PRIVINFLT: /* privileged instruction fault */
561 case T_BPTFLT: /* bpt instruction fault */
562 case T_TRCTRAP: /* trace trap */
563 frame->tf_eflags &= ~PSL_T;
567 case T_ARITHTRAP: /* arithmetic trap */
572 case T_ASTFLT: /* Allow process switch */
573 mycpu->gd_cnt.v_soft++;
574 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
575 atomic_clear_int_nonlocked(&mycpu->gd_reqflags,
577 addupc_task(p, p->p_prof.pr_addr,
583 * The following two traps can happen in
584 * vm86 mode, and, if so, we want to handle
587 case T_PROTFLT: /* general protection fault */
588 case T_STKFLT: /* stack fault */
589 if (frame->tf_eflags & PSL_VM) {
590 i = vm86_emulate((struct vm86frame *)frame);
597 case T_SEGNPFLT: /* segment not present fault */
598 case T_TSSFLT: /* invalid TSS fault */
599 case T_DOUBLEFLT: /* double fault */
601 ucode = code + BUS_SEGM_FAULT ;
605 case T_PAGEFLT: /* page fault */
606 MAKEMPSAFE(have_mplock);
607 i = trap_pfault(frame, TRUE, eva);
610 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
620 case T_DIVIDE: /* integer divide fault */
627 MAKEMPSAFE(have_mplock);
629 goto handle_powerfail;
630 #else /* !POWERFAIL_NMI */
631 /* machine/parity/power fail/"kitchen sink" faults */
632 if (isa_nmi(code) == 0) {
635 * NMI can be hooked up to a pushbutton
639 kprintf ("NMI ... going to debugger\n");
640 kdb_trap (type, 0, frame);
644 } else if (panic_on_nmi)
645 panic("NMI indicates hardware failure");
647 #endif /* POWERFAIL_NMI */
648 #endif /* NISA > 0 */
650 case T_OFLOW: /* integer overflow fault */
655 case T_BOUND: /* bounds check fault */
662 * Virtual kernel intercept - pass the DNA exception
663 * to the virtual kernel if it asked to handle it.
664 * This occurs when the virtual kernel is holding
665 * onto the FP context for a different emulated
666 * process then the one currently running.
668 * We must still call npxdna() since we may have
669 * saved FP state that the virtual kernel needs
670 * to hand over to a different emulated process.
672 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
673 (td->td_pcb->pcb_flags & FP_VIRTFP)
681 * The kernel may have switched out the FP unit's
682 * state, causing the user process to take a fault
683 * when it tries to use the FP unit. Restore the
689 if (!pmath_emulate) {
691 ucode = FPE_FPU_NP_TRAP;
694 i = (*pmath_emulate)(frame);
696 if (!(frame->tf_eflags & PSL_T))
698 frame->tf_eflags &= ~PSL_T;
701 /* else ucode = emulator_only_knows() XXX */
704 case T_FPOPFLT: /* FPU operand fetch fault */
709 case T_XMMFLT: /* SIMD floating-point exception */
719 case T_PAGEFLT: /* page fault */
720 MAKEMPSAFE(have_mplock);
721 trap_pfault(frame, FALSE, eva);
727 * The kernel may be using npx for copying or other
735 case T_PROTFLT: /* general protection fault */
736 case T_SEGNPFLT: /* segment not present fault */
738 * Invalid segment selectors and out of bounds
739 * %eip's and %esp's can be set up in user mode.
740 * This causes a fault in kernel mode when the
741 * kernel tries to return to user mode. We want
742 * to get this fault so that we can fix the
743 * problem here and not have to check all the
744 * selectors and pointers when the user changes
747 #define MAYBE_DORETI_FAULT(where, whereto) \
749 if (frame->tf_eip == (int)where) { \
750 frame->tf_eip = (int)whereto; \
754 if (mycpu->gd_intr_nesting_level == 0) {
756 * Invalid %fs's and %gs's can be created using
757 * procfs or PT_SETREGS or by invalidating the
758 * underlying LDT entry. This causes a fault
759 * in kernel mode when the kernel attempts to
760 * switch contexts. Lose the bad context
761 * (XXX) so that we can continue, and generate
764 MAYBE_DORETI_FAULT(doreti_iret,
766 MAYBE_DORETI_FAULT(doreti_popl_ds,
767 doreti_popl_ds_fault);
768 MAYBE_DORETI_FAULT(doreti_popl_es,
769 doreti_popl_es_fault);
770 MAYBE_DORETI_FAULT(doreti_popl_fs,
771 doreti_popl_fs_fault);
772 MAYBE_DORETI_FAULT(doreti_popl_gs,
773 doreti_popl_gs_fault);
774 if (td->td_pcb->pcb_onfault) {
776 (register_t)td->td_pcb->pcb_onfault;
784 * PSL_NT can be set in user mode and isn't cleared
785 * automatically when the kernel is entered. This
786 * causes a TSS fault when the kernel attempts to
787 * `iret' because the TSS link is uninitialized. We
788 * want to get this fault so that we can fix the
789 * problem here and not every time the kernel is
792 if (frame->tf_eflags & PSL_NT) {
793 frame->tf_eflags &= ~PSL_NT;
798 case T_TRCTRAP: /* trace trap */
799 if (frame->tf_eip == (int)IDTVEC(syscall)) {
801 * We've just entered system mode via the
802 * syscall lcall. Continue single stepping
803 * silently until the syscall handler has
808 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
810 * The syscall handler has now saved the
811 * flags. Stop single stepping it.
813 frame->tf_eflags &= ~PSL_T;
817 * Ignore debug register trace traps due to
818 * accesses in the user's address space, which
819 * can happen under several conditions such as
820 * if a user sets a watchpoint on a buffer and
821 * then passes that buffer to a system call.
822 * We still want to get TRCTRAPS for addresses
823 * in kernel space because that is useful when
824 * debugging the kernel.
826 if (user_dbreg_trap()) {
828 * Reset breakpoint bits because the
831 load_dr6(rdr6() & 0xfffffff0);
835 * Fall through (TRCTRAP kernel mode, kernel address)
839 * If DDB is enabled, let it handle the debugger trap.
840 * Otherwise, debugger traps "can't happen".
843 MAKEMPSAFE(have_mplock);
844 if (kdb_trap (type, 0, frame))
851 MAKEMPSAFE(have_mplock);
854 # define TIMER_FREQ 1193182
858 static unsigned lastalert = 0;
860 if(time_second - lastalert > 10)
862 log(LOG_WARNING, "NMI: power fail\n");
863 sysbeep(TIMER_FREQ/880, hz);
864 lastalert = time_second;
869 #else /* !POWERFAIL_NMI */
870 /* machine/parity/power fail/"kitchen sink" faults */
871 if (isa_nmi(code) == 0) {
874 * NMI can be hooked up to a pushbutton
878 kprintf ("NMI ... going to debugger\n");
879 kdb_trap (type, 0, frame);
883 } else if (panic_on_nmi == 0)
886 #endif /* POWERFAIL_NMI */
887 #endif /* NISA > 0 */
890 MAKEMPSAFE(have_mplock);
891 trap_fatal(frame, eva);
896 * Virtual kernel intercept - if the fault is directly related to a
897 * VM context managed by a virtual kernel then let the virtual kernel
900 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
901 vkernel_trap(lp, frame);
906 * Translate fault for emulators (e.g. Linux)
908 if (*p->p_sysent->sv_transtrap)
909 i = (*p->p_sysent->sv_transtrap)(i, type);
911 MAKEMPSAFE(have_mplock);
912 trapsignal(lp, i, ucode);
915 if (type <= MAX_TRAP_MSG) {
916 uprintf("fatal process exception: %s",
918 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
919 uprintf(", fault VA = 0x%lx", (u_long)eva);
926 if (ISPL(frame->tf_cs) == SEL_UPL)
927 KASSERT(td->td_mpcount == have_mplock, ("badmpcount trap/end from %p", (void *)frame->tf_eip));
929 userret(lp, frame, sticks);
936 if (p != NULL && lp != NULL)
937 KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid);
939 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
940 ("syscall: critical section count mismatch! %d/%d",
941 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
946 trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
949 struct vmspace *vm = NULL;
953 thread_t td = curthread;
954 struct lwp *lp = td->td_lwp;
956 va = trunc_page(eva);
957 if (va >= KERNBASE) {
959 * Don't allow user-mode faults in kernel address space.
960 * An exception: if the faulting address is the invalid
961 * instruction entry in the IDT, then the Intel Pentium
962 * F00F bug workaround was triggered, and we need to
963 * treat it is as an illegal instruction, and not a page
966 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
967 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) {
968 frame->tf_trapno = T_PRIVINFLT;
978 * This is a fault on non-kernel virtual memory.
979 * vm is initialized above to NULL. If curproc is NULL
980 * or curproc->p_vmspace is NULL the fault is fatal.
983 vm = lp->lwp_vmspace;
991 if (frame->tf_err & PGEX_W)
992 ftype = VM_PROT_WRITE;
994 ftype = VM_PROT_READ;
996 if (map != &kernel_map) {
998 * Keep swapout from messing with us during this
1001 PHOLD(lp->lwp_proc);
1004 * Grow the stack if necessary
1006 /* grow_stack returns false only if va falls into
1007 * a growable stack region and the stack growth
1008 * fails. It returns true if va was not within
1009 * a growable stack region, or if the stack
1012 if (!grow_stack(lp->lwp_proc, va)) {
1014 PRELE(lp->lwp_proc);
1018 /* Fault in the user page: */
1019 rv = vm_fault(map, va, ftype,
1020 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
1023 PRELE(lp->lwp_proc);
1026 * Don't have to worry about process locking or stacks
1029 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
1032 if (rv == KERN_SUCCESS)
1036 if (td->td_gd->gd_intr_nesting_level == 0 &&
1037 td->td_pcb->pcb_onfault) {
1038 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
1041 trap_fatal(frame, eva);
1045 /* kludge to pass faulting virtual address to sendsig */
1046 frame->tf_xflags = frame->tf_err;
1047 frame->tf_err = eva;
1049 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
1053 trap_fatal(struct trapframe *frame, vm_offset_t eva)
1055 int code, type, ss, esp;
1056 struct soft_segment_descriptor softseg;
1058 code = frame->tf_err;
1059 type = frame->tf_trapno;
1060 sdtossd(&gdt[mycpu->gd_cpuid * NGDT + IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
1062 if (type <= MAX_TRAP_MSG)
1063 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
1064 type, trap_msg[type],
1065 frame->tf_eflags & PSL_VM ? "vm86" :
1066 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
1068 /* three separate prints in case of a trap on an unmapped page */
1069 kprintf("mp_lock = %08x; ", mp_lock);
1070 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
1071 kprintf("lapic.id = %08x\n", lapic.id);
1073 if (type == T_PAGEFLT) {
1074 kprintf("fault virtual address = 0x%x\n", eva);
1075 kprintf("fault code = %s %s, %s\n",
1076 code & PGEX_U ? "user" : "supervisor",
1077 code & PGEX_W ? "write" : "read",
1078 code & PGEX_P ? "protection violation" : "page not present");
1080 kprintf("instruction pointer = 0x%x:0x%x\n",
1081 frame->tf_cs & 0xffff, frame->tf_eip);
1082 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
1083 ss = frame->tf_ss & 0xffff;
1084 esp = frame->tf_esp;
1086 ss = GSEL(GDATA_SEL, SEL_KPL);
1087 esp = (int)&frame->tf_esp;
1089 kprintf("stack pointer = 0x%x:0x%x\n", ss, esp);
1090 kprintf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
1091 kprintf("code segment = base 0x%x, limit 0x%x, type 0x%x\n",
1092 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
1093 kprintf(" = DPL %d, pres %d, def32 %d, gran %d\n",
1094 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
1096 kprintf("processor eflags = ");
1097 if (frame->tf_eflags & PSL_T)
1098 kprintf("trace trap, ");
1099 if (frame->tf_eflags & PSL_I)
1100 kprintf("interrupt enabled, ");
1101 if (frame->tf_eflags & PSL_NT)
1102 kprintf("nested task, ");
1103 if (frame->tf_eflags & PSL_RF)
1104 kprintf("resume, ");
1105 if (frame->tf_eflags & PSL_VM)
1107 kprintf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
1108 kprintf("current process = ");
1110 kprintf("%lu (%s)\n",
1111 (u_long)curproc->p_pid, curproc->p_comm ?
1112 curproc->p_comm : "");
1116 kprintf("current thread = pri %d ", curthread->td_pri);
1117 if (curthread->td_pri >= TDPRI_CRIT)
1123 * we probably SHOULD have stopped the other CPUs before now!
1124 * another CPU COULD have been touching cpl at this moment...
1126 kprintf(" <- SMP: XXX");
1135 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1138 kprintf("trap number = %d\n", type);
1139 if (type <= MAX_TRAP_MSG)
1140 panic("%s", trap_msg[type]);
1142 panic("unknown/reserved trap");
1146 * Double fault handler. Called when a fault occurs while writing
1147 * a frame for a trap/exception onto the stack. This usually occurs
1148 * when the stack overflows (such is the case with infinite recursion,
1151 * XXX Note that the current PTD gets replaced by IdlePTD when the
1152 * task switch occurs. This means that the stack that was active at
1153 * the time of the double fault is not available at <kstack> unless
1154 * the machine was idle when the double fault occurred. The downside
1155 * of this is that "trace <ebp>" in ddb won't work.
1158 dblfault_handler(void)
1160 struct mdglobaldata *gd = mdcpu;
1162 kprintf("\nFatal 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("mp_lock = %08x; ", mp_lock);
1169 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
1170 kprintf("lapic.id = %08x\n", lapic.id);
1172 panic("double fault");
1176 * Compensate for 386 brain damage (missing URKR).
1177 * This is a little simpler than the pagefault handler in trap() because
1178 * it the page tables have already been faulted in and high addresses
1179 * are thrown out early for other reasons.
1182 trapwrite(unsigned addr)
1189 va = trunc_page((vm_offset_t)addr);
1191 * XXX - MAX is END. Changed > to >= for temp. fix.
1193 if (va >= VM_MAX_USER_ADDRESS)
1196 lp = curthread->td_lwp;
1197 vm = lp->lwp_vmspace;
1199 PHOLD(lp->lwp_proc);
1201 if (!grow_stack(lp->lwp_proc, va)) {
1202 PRELE(lp->lwp_proc);
1207 * fault the data page
1209 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY);
1211 PRELE(lp->lwp_proc);
1213 if (rv != KERN_SUCCESS)
1220 * syscall2 - MP aware system call request C handler
1222 * A system call is essentially treated as a trap except that the
1223 * MP lock is not held on entry or return. We are responsible for
1224 * obtaining the MP lock if necessary and for handling ASTs
1225 * (e.g. a task switch) prior to return.
1227 * In general, only simple access and manipulation of curproc and
1228 * the current stack is allowed without having to hold MP lock.
1230 * MPSAFE - note that large sections of this routine are run without
1235 syscall2(struct trapframe *frame)
1237 struct thread *td = curthread;
1238 struct proc *p = td->td_proc;
1239 struct lwp *lp = td->td_lwp;
1241 struct sysent *callp;
1242 register_t orig_tf_eflags;
1247 int crit_count = td->td_pri & ~TDPRI_MASK;
1250 int have_mplock = 0;
1253 union sysunion args;
1256 if (ISPL(frame->tf_cs) != SEL_UPL) {
1263 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid,
1267 KASSERT(td->td_mpcount == 0, ("badmpcount syscall2 from %p", (void *)frame->tf_eip));
1268 if (syscall_mpsafe == 0)
1269 MAKEMPSAFE(have_mplock);
1271 userenter(td); /* lazy raise our priority */
1276 sticks = (int)td->td_sticks;
1277 orig_tf_eflags = frame->tf_eflags;
1280 * Virtual kernel intercept - if a VM context managed by a virtual
1281 * kernel issues a system call the virtual kernel handles it, not us.
1282 * Restore the virtual kernel context and return from its system
1283 * call. The current frame is copied out to the virtual kernel.
1285 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1286 error = vkernel_trap(lp, frame);
1287 frame->tf_eax = error;
1289 frame->tf_eflags |= PSL_C;
1290 error = EJUSTRETURN;
1295 * Get the system call parameters and account for time
1297 lp->lwp_md.md_regs = frame;
1298 params = (caddr_t)frame->tf_esp + sizeof(int);
1299 code = frame->tf_eax;
1301 if (p->p_sysent->sv_prepsyscall) {
1302 (*p->p_sysent->sv_prepsyscall)(
1303 frame, (int *)(&args.nosys.sysmsg + 1),
1307 * Need to check if this is a 32 bit or 64 bit syscall.
1308 * fuword is MP aware.
1310 if (code == SYS_syscall) {
1312 * Code is first argument, followed by actual args.
1314 code = fuword(params);
1315 params += sizeof(int);
1316 } else if (code == SYS___syscall) {
1318 * Like syscall, but code is a quad, so as to maintain
1319 * quad alignment for the rest of the arguments.
1321 code = fuword(params);
1322 params += sizeof(quad_t);
1326 code &= p->p_sysent->sv_mask;
1327 if (code >= p->p_sysent->sv_size)
1328 callp = &p->p_sysent->sv_table[0];
1330 callp = &p->p_sysent->sv_table[code];
1332 narg = callp->sy_narg & SYF_ARGMASK;
1335 * copyin is MP aware, but the tracing code is not
1337 if (narg && params) {
1338 error = copyin(params, (caddr_t)(&args.nosys.sysmsg + 1),
1339 narg * sizeof(register_t));
1342 if (KTRPOINT(td, KTR_SYSCALL)) {
1343 MAKEMPSAFE(have_mplock);
1345 ktrsyscall(lp, code, narg,
1346 (void *)(&args.nosys.sysmsg + 1));
1354 if (KTRPOINT(td, KTR_SYSCALL)) {
1355 MAKEMPSAFE(have_mplock);
1356 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1361 * For traditional syscall code edx is left untouched when 32 bit
1362 * results are returned. Since edx is loaded from fds[1] when the
1363 * system call returns we pre-set it here.
1365 args.sysmsg_fds[0] = 0;
1366 args.sysmsg_fds[1] = frame->tf_edx;
1369 * The syscall might manipulate the trap frame. If it does it
1370 * will probably return EJUSTRETURN.
1372 args.sysmsg_frame = frame;
1374 STOPEVENT(p, S_SCE, narg); /* MP aware */
1378 * Try to run the syscall without the MP lock if the syscall
1379 * is MP safe. We have to obtain the MP lock no matter what if
1382 if ((callp->sy_narg & SYF_MPSAFE) == 0)
1383 MAKEMPSAFE(have_mplock);
1386 error = (*callp->sy_call)(&args);
1390 * MP SAFE (we may or may not have the MP lock at this point)
1395 * Reinitialize proc pointer `p' as it may be different
1396 * if this is a child returning from fork syscall.
1399 lp = curthread->td_lwp;
1400 frame->tf_eax = args.sysmsg_fds[0];
1401 frame->tf_edx = args.sysmsg_fds[1];
1402 frame->tf_eflags &= ~PSL_C;
1406 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1407 * int 0x80 is 2 bytes. We saved this in tf_err.
1409 frame->tf_eip -= frame->tf_err;
1414 panic("Unexpected EASYNC return value (for now)");
1417 if (p->p_sysent->sv_errsize) {
1418 if (error >= p->p_sysent->sv_errsize)
1419 error = -1; /* XXX */
1421 error = p->p_sysent->sv_errtbl[error];
1423 frame->tf_eax = error;
1424 frame->tf_eflags |= PSL_C;
1429 * Traced syscall. trapsignal() is not MP aware.
1431 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1432 MAKEMPSAFE(have_mplock);
1433 frame->tf_eflags &= ~PSL_T;
1434 trapsignal(lp, SIGTRAP, 0);
1438 * Handle reschedule and other end-of-syscall issues
1440 userret(lp, frame, sticks);
1443 if (KTRPOINT(td, KTR_SYSRET)) {
1444 MAKEMPSAFE(have_mplock);
1445 ktrsysret(lp, code, error, args.sysmsg_result);
1450 * This works because errno is findable through the
1451 * register set. If we ever support an emulation where this
1452 * is not the case, this code will need to be revisited.
1454 STOPEVENT(p, S_SCX, code);
1459 * Release the MP lock if we had to get it
1461 KASSERT(td->td_mpcount == have_mplock,
1462 ("badmpcount syscall2/end from %p", (void *)frame->tf_eip));
1466 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error);
1468 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
1469 ("syscall: critical section count mismatch! %d/%d",
1470 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
1475 fork_return(struct lwp *lp, struct trapframe *frame)
1477 frame->tf_eax = 0; /* Child returns zero */
1478 frame->tf_eflags &= ~PSL_C; /* success */
1481 generic_lwp_return(lp, frame);
1482 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1486 * Simplified back end of syscall(), used when returning from fork()
1487 * directly into user mode. MP lock is held on entry and should be
1488 * released on return. This code will return back into the fork
1489 * trampoline code which then runs doreti.
1492 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1494 struct proc *p = lp->lwp_proc;
1497 * Newly forked processes are given a kernel priority. We have to
1498 * adjust the priority to a normal user priority and fake entry
1499 * into the kernel (call userenter()) to install a passive release
1500 * function just in case userret() decides to stop the process. This
1501 * can occur when ^Z races a fork. If we do not install the passive
1502 * release function the current process designation will not be
1503 * released when the thread goes to sleep.
1505 lwkt_setpri_self(TDPRI_USER_NORM);
1506 userenter(lp->lwp_thread);
1507 userret(lp, frame, 0);
1509 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1510 ktrsysret(lp, SYS_fork, 0, 0);
1512 p->p_flag |= P_PASSIVE_ACQ;
1514 p->p_flag &= ~P_PASSIVE_ACQ;
1516 KKASSERT(lp->lwp_thread->td_mpcount == 1);
1522 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1523 * fault (which is then passed back to the virtual kernel) if an attempt is
1524 * made to use the FP unit.
1526 * XXX this is a fairly big hack.
1529 set_vkernel_fp(struct trapframe *frame)
1531 struct thread *td = curthread;
1533 if (frame->tf_xflags & PGEX_FPFAULT) {
1534 td->td_pcb->pcb_flags |= FP_VIRTFP;
1535 if (mdcpu->gd_npxthread == td)
1538 td->td_pcb->pcb_flags &= ~FP_VIRTFP;