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 1.114 2008/07/13 10:28:51 nth 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.
342 while (lwkt_resched_wanted()) {
346 * The thread that preempted us may have stopped our process.
348 while (lp->lwp_proc->p_stat == SSTOP) {
356 * Acquire the current process designation for this user scheduler
357 * on this cpu. This will also handle any user-reschedule requests.
359 lp->lwp_proc->p_usched->acquire_curproc(lp);
360 /* We may have switched cpus on acquisition */
364 * Reduce our priority in preparation for a return to userland. If
365 * our passive release function was still in place, our priority was
366 * never raised and does not need to be reduced.
368 if (td->td_release == NULL)
369 lwkt_setpri_self(TDPRI_USER_NORM);
370 td->td_release = NULL;
373 * After reducing our priority there might be other kernel-level
374 * LWKTs that now have a greater priority. Run them as necessary.
375 * We don't have to worry about losing cpu to userland because
376 * we still control the current-process designation and we no longer
377 * have a passive release function installed.
379 if (lwkt_checkpri_self())
383 #if !defined(KTR_KERNENTRY)
384 #define KTR_KERNENTRY KTR_ALL
386 KTR_INFO_MASTER(kernentry);
387 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, "pid=%d, tid=%d, trapno=%d, eva=%p",
388 sizeof(int) + sizeof(int) + sizeof(int) + sizeof(vm_offset_t));
389 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "pid=%d, tid=%d",
390 sizeof(int) + sizeof(int));
391 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "pid=%d, tid=%d, call=%d",
392 sizeof(int) + sizeof(int) + sizeof(int));
393 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "pid=%d, tid=%d, err=%d",
394 sizeof(int) + sizeof(int) + sizeof(int));
395 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "pid=%d, tid=%d",
396 sizeof(int) + sizeof(int));
399 * Exception, fault, and trap interface to the kernel.
400 * This common code is called from assembly language IDT gate entry
401 * routines that prepare a suitable stack frame, and restore this
402 * frame after the exception has been processed.
404 * This function is also called from doreti in an interlock to handle ASTs.
405 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
407 * NOTE! We have to retrieve the fault address prior to obtaining the
408 * MP lock because get_mplock() may switch out. YYY cr2 really ought
409 * to be retrieved by the assembly code, not here.
411 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
412 * if an attempt is made to switch from a fast interrupt or IPI. This is
413 * necessary to properly take fatal kernel traps on SMP machines if
414 * get_mplock() has to block.
418 trap(struct trapframe *frame)
420 struct globaldata *gd = mycpu;
421 struct thread *td = gd->gd_curthread;
422 struct lwp *lp = td->td_lwp;
425 int i = 0, ucode = 0, type, code;
430 int crit_count = td->td_pri & ~TDPRI_MASK;
437 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
438 ++gd->gd_trap_nesting_level;
439 MAKEMPSAFE(have_mplock);
440 trap_fatal(frame, eva);
441 --gd->gd_trap_nesting_level;
447 ++gd->gd_trap_nesting_level;
448 if (frame->tf_trapno == T_PAGEFLT) {
450 * For some Cyrix CPUs, %cr2 is clobbered by interrupts.
451 * This problem is worked around by using an interrupt
452 * gate for the pagefault handler. We are finally ready
453 * to read %cr2 and then must reenable interrupts.
455 * XXX this should be in the switch statement, but the
456 * NO_FOOF_HACK and VM86 goto and ifdefs obfuscate the
457 * flow of control too much for this to be obviously
465 if (trap_mpsafe == 0)
466 MAKEMPSAFE(have_mplock);
469 --gd->gd_trap_nesting_level;
471 if (!(frame->tf_eflags & PSL_I)) {
473 * Buggy application or kernel code has disabled interrupts
474 * and then trapped. Enabling interrupts now is wrong, but
475 * it is better than running with interrupts disabled until
476 * they are accidentally enabled later.
478 type = frame->tf_trapno;
479 if (ISPL(frame->tf_cs)==SEL_UPL || (frame->tf_eflags & PSL_VM)) {
480 MAKEMPSAFE(have_mplock);
482 "pid %ld (%s): trap %d with interrupts disabled\n",
483 (long)curproc->p_pid, curproc->p_comm, type);
484 } else if (type != T_BPTFLT && type != T_TRCTRAP) {
486 * XXX not quite right, since this may be for a
487 * multiple fault in user mode.
489 MAKEMPSAFE(have_mplock);
490 kprintf("kernel trap %d with interrupts disabled\n",
496 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
499 type = frame->tf_trapno;
500 code = frame->tf_err;
503 if (frame->tf_eflags & PSL_VM &&
504 (type == T_PROTFLT || type == T_STKFLT)) {
506 KKASSERT(td->td_mpcount > 0);
508 i = vm86_emulate((struct vm86frame *)frame);
510 KKASSERT(td->td_mpcount > 0);
514 * returns to original process
517 vm86_trap((struct vm86frame *)frame,
520 vm86_trap((struct vm86frame *)frame, 0);
522 KKASSERT(0); /* NOT REACHED */
528 * these traps want either a process context, or
529 * assume a normal userspace trap.
533 trap_fatal(frame, eva);
536 type = T_BPTFLT; /* kernel breakpoint */
539 goto kernel_trap; /* normal kernel trap handling */
542 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
545 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid,
546 frame->tf_trapno, eva);
550 sticks = (int)td->td_sticks;
551 lp->lwp_md.md_regs = frame;
554 case T_PRIVINFLT: /* privileged instruction fault */
559 case T_BPTFLT: /* bpt instruction fault */
560 case T_TRCTRAP: /* trace trap */
561 frame->tf_eflags &= ~PSL_T;
565 case T_ARITHTRAP: /* arithmetic trap */
570 case T_ASTFLT: /* Allow process switch */
571 mycpu->gd_cnt.v_soft++;
572 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
573 atomic_clear_int_nonlocked(&mycpu->gd_reqflags,
575 addupc_task(p, p->p_prof.pr_addr,
581 * The following two traps can happen in
582 * vm86 mode, and, if so, we want to handle
585 case T_PROTFLT: /* general protection fault */
586 case T_STKFLT: /* stack fault */
587 if (frame->tf_eflags & PSL_VM) {
588 i = vm86_emulate((struct vm86frame *)frame);
595 case T_SEGNPFLT: /* segment not present fault */
596 case T_TSSFLT: /* invalid TSS fault */
597 case T_DOUBLEFLT: /* double fault */
599 ucode = code + BUS_SEGM_FAULT ;
603 case T_PAGEFLT: /* page fault */
604 MAKEMPSAFE(have_mplock);
605 i = trap_pfault(frame, TRUE, eva);
608 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
618 case T_DIVIDE: /* integer divide fault */
625 MAKEMPSAFE(have_mplock);
627 goto handle_powerfail;
628 #else /* !POWERFAIL_NMI */
629 /* machine/parity/power fail/"kitchen sink" faults */
630 if (isa_nmi(code) == 0) {
633 * NMI can be hooked up to a pushbutton
637 kprintf ("NMI ... going to debugger\n");
638 kdb_trap (type, 0, frame);
642 } else if (panic_on_nmi)
643 panic("NMI indicates hardware failure");
645 #endif /* POWERFAIL_NMI */
646 #endif /* NISA > 0 */
648 case T_OFLOW: /* integer overflow fault */
653 case T_BOUND: /* bounds check fault */
660 * Virtual kernel intercept - pass the DNA exception
661 * to the virtual kernel if it asked to handle it.
662 * This occurs when the virtual kernel is holding
663 * onto the FP context for a different emulated
664 * process then the one currently running.
666 * We must still call npxdna() since we may have
667 * saved FP state that the virtual kernel needs
668 * to hand over to a different emulated process.
670 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
671 (td->td_pcb->pcb_flags & FP_VIRTFP)
679 * The kernel may have switched out the FP unit's
680 * state, causing the user process to take a fault
681 * when it tries to use the FP unit. Restore the
687 if (!pmath_emulate) {
689 ucode = FPE_FPU_NP_TRAP;
692 i = (*pmath_emulate)(frame);
694 if (!(frame->tf_eflags & PSL_T))
696 frame->tf_eflags &= ~PSL_T;
699 /* else ucode = emulator_only_knows() XXX */
702 case T_FPOPFLT: /* FPU operand fetch fault */
707 case T_XMMFLT: /* SIMD floating-point exception */
717 case T_PAGEFLT: /* page fault */
718 MAKEMPSAFE(have_mplock);
719 trap_pfault(frame, FALSE, eva);
725 * The kernel may be using npx for copying or other
733 case T_PROTFLT: /* general protection fault */
734 case T_SEGNPFLT: /* segment not present fault */
736 * Invalid segment selectors and out of bounds
737 * %eip's and %esp's can be set up in user mode.
738 * This causes a fault in kernel mode when the
739 * kernel tries to return to user mode. We want
740 * to get this fault so that we can fix the
741 * problem here and not have to check all the
742 * selectors and pointers when the user changes
745 #define MAYBE_DORETI_FAULT(where, whereto) \
747 if (frame->tf_eip == (int)where) { \
748 frame->tf_eip = (int)whereto; \
752 if (mycpu->gd_intr_nesting_level == 0) {
754 * Invalid %fs's and %gs's can be created using
755 * procfs or PT_SETREGS or by invalidating the
756 * underlying LDT entry. This causes a fault
757 * in kernel mode when the kernel attempts to
758 * switch contexts. Lose the bad context
759 * (XXX) so that we can continue, and generate
762 MAYBE_DORETI_FAULT(doreti_iret,
764 MAYBE_DORETI_FAULT(doreti_popl_ds,
765 doreti_popl_ds_fault);
766 MAYBE_DORETI_FAULT(doreti_popl_es,
767 doreti_popl_es_fault);
768 MAYBE_DORETI_FAULT(doreti_popl_fs,
769 doreti_popl_fs_fault);
770 MAYBE_DORETI_FAULT(doreti_popl_gs,
771 doreti_popl_gs_fault);
772 if (td->td_pcb->pcb_onfault) {
774 (register_t)td->td_pcb->pcb_onfault;
782 * PSL_NT can be set in user mode and isn't cleared
783 * automatically when the kernel is entered. This
784 * causes a TSS fault when the kernel attempts to
785 * `iret' because the TSS link is uninitialized. We
786 * want to get this fault so that we can fix the
787 * problem here and not every time the kernel is
790 if (frame->tf_eflags & PSL_NT) {
791 frame->tf_eflags &= ~PSL_NT;
796 case T_TRCTRAP: /* trace trap */
797 if (frame->tf_eip == (int)IDTVEC(syscall)) {
799 * We've just entered system mode via the
800 * syscall lcall. Continue single stepping
801 * silently until the syscall handler has
806 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
808 * The syscall handler has now saved the
809 * flags. Stop single stepping it.
811 frame->tf_eflags &= ~PSL_T;
815 * Ignore debug register trace traps due to
816 * accesses in the user's address space, which
817 * can happen under several conditions such as
818 * if a user sets a watchpoint on a buffer and
819 * then passes that buffer to a system call.
820 * We still want to get TRCTRAPS for addresses
821 * in kernel space because that is useful when
822 * debugging the kernel.
824 if (user_dbreg_trap()) {
826 * Reset breakpoint bits because the
829 load_dr6(rdr6() & 0xfffffff0);
833 * Fall through (TRCTRAP kernel mode, kernel address)
837 * If DDB is enabled, let it handle the debugger trap.
838 * Otherwise, debugger traps "can't happen".
841 MAKEMPSAFE(have_mplock);
842 if (kdb_trap (type, 0, frame))
849 MAKEMPSAFE(have_mplock);
852 # define TIMER_FREQ 1193182
856 static unsigned lastalert = 0;
858 if(time_second - lastalert > 10)
860 log(LOG_WARNING, "NMI: power fail\n");
861 sysbeep(TIMER_FREQ/880, hz);
862 lastalert = time_second;
867 #else /* !POWERFAIL_NMI */
868 /* machine/parity/power fail/"kitchen sink" faults */
869 if (isa_nmi(code) == 0) {
872 * NMI can be hooked up to a pushbutton
876 kprintf ("NMI ... going to debugger\n");
877 kdb_trap (type, 0, frame);
881 } else if (panic_on_nmi == 0)
884 #endif /* POWERFAIL_NMI */
885 #endif /* NISA > 0 */
888 MAKEMPSAFE(have_mplock);
889 trap_fatal(frame, eva);
894 * Virtual kernel intercept - if the fault is directly related to a
895 * VM context managed by a virtual kernel then let the virtual kernel
898 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
899 vkernel_trap(lp, frame);
904 * Translate fault for emulators (e.g. Linux)
906 if (*p->p_sysent->sv_transtrap)
907 i = (*p->p_sysent->sv_transtrap)(i, type);
909 MAKEMPSAFE(have_mplock);
910 trapsignal(lp, i, ucode);
913 if (type <= MAX_TRAP_MSG) {
914 uprintf("fatal process exception: %s",
916 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
917 uprintf(", fault VA = 0x%lx", (u_long)eva);
924 if (ISPL(frame->tf_cs) == SEL_UPL)
925 KASSERT(td->td_mpcount == have_mplock, ("badmpcount trap/end from %p", (void *)frame->tf_eip));
927 userret(lp, frame, sticks);
934 if (p != NULL && lp != NULL)
935 KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid);
937 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
938 ("syscall: critical section count mismatch! %d/%d",
939 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
944 trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
947 struct vmspace *vm = NULL;
951 thread_t td = curthread;
952 struct lwp *lp = td->td_lwp;
954 va = trunc_page(eva);
955 if (va >= KERNBASE) {
957 * Don't allow user-mode faults in kernel address space.
958 * An exception: if the faulting address is the invalid
959 * instruction entry in the IDT, then the Intel Pentium
960 * F00F bug workaround was triggered, and we need to
961 * treat it is as an illegal instruction, and not a page
964 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
965 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) {
966 frame->tf_trapno = T_PRIVINFLT;
976 * This is a fault on non-kernel virtual memory.
977 * vm is initialized above to NULL. If curproc is NULL
978 * or curproc->p_vmspace is NULL the fault is fatal.
981 vm = lp->lwp_vmspace;
989 if (frame->tf_err & PGEX_W)
990 ftype = VM_PROT_WRITE;
992 ftype = VM_PROT_READ;
994 if (map != &kernel_map) {
996 * Keep swapout from messing with us during this
1002 * Grow the stack if necessary
1004 /* grow_stack returns false only if va falls into
1005 * a growable stack region and the stack growth
1006 * fails. It returns true if va was not within
1007 * a growable stack region, or if the stack
1010 if (!grow_stack(lp->lwp_proc, va)) {
1012 PRELE(lp->lwp_proc);
1016 /* Fault in the user page: */
1017 rv = vm_fault(map, va, ftype,
1018 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
1021 PRELE(lp->lwp_proc);
1024 * Don't have to worry about process locking or stacks
1027 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
1030 if (rv == KERN_SUCCESS)
1034 if (td->td_gd->gd_intr_nesting_level == 0 &&
1035 td->td_pcb->pcb_onfault) {
1036 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
1039 trap_fatal(frame, eva);
1043 /* kludge to pass faulting virtual address to sendsig */
1044 frame->tf_xflags = frame->tf_err;
1045 frame->tf_err = eva;
1047 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
1051 trap_fatal(struct trapframe *frame, vm_offset_t eva)
1053 int code, type, ss, esp;
1054 struct soft_segment_descriptor softseg;
1056 code = frame->tf_err;
1057 type = frame->tf_trapno;
1058 sdtossd(&gdt[mycpu->gd_cpuid * NGDT + IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
1060 if (type <= MAX_TRAP_MSG)
1061 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
1062 type, trap_msg[type],
1063 frame->tf_eflags & PSL_VM ? "vm86" :
1064 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
1066 /* three separate prints in case of a trap on an unmapped page */
1067 kprintf("mp_lock = %08x; ", mp_lock);
1068 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
1069 kprintf("lapic.id = %08x\n", lapic.id);
1071 if (type == T_PAGEFLT) {
1072 kprintf("fault virtual address = 0x%x\n", eva);
1073 kprintf("fault code = %s %s, %s\n",
1074 code & PGEX_U ? "user" : "supervisor",
1075 code & PGEX_W ? "write" : "read",
1076 code & PGEX_P ? "protection violation" : "page not present");
1078 kprintf("instruction pointer = 0x%x:0x%x\n",
1079 frame->tf_cs & 0xffff, frame->tf_eip);
1080 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
1081 ss = frame->tf_ss & 0xffff;
1082 esp = frame->tf_esp;
1084 ss = GSEL(GDATA_SEL, SEL_KPL);
1085 esp = (int)&frame->tf_esp;
1087 kprintf("stack pointer = 0x%x:0x%x\n", ss, esp);
1088 kprintf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
1089 kprintf("code segment = base 0x%x, limit 0x%x, type 0x%x\n",
1090 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
1091 kprintf(" = DPL %d, pres %d, def32 %d, gran %d\n",
1092 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
1094 kprintf("processor eflags = ");
1095 if (frame->tf_eflags & PSL_T)
1096 kprintf("trace trap, ");
1097 if (frame->tf_eflags & PSL_I)
1098 kprintf("interrupt enabled, ");
1099 if (frame->tf_eflags & PSL_NT)
1100 kprintf("nested task, ");
1101 if (frame->tf_eflags & PSL_RF)
1102 kprintf("resume, ");
1103 if (frame->tf_eflags & PSL_VM)
1105 kprintf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
1106 kprintf("current process = ");
1108 kprintf("%lu (%s)\n",
1109 (u_long)curproc->p_pid, curproc->p_comm ?
1110 curproc->p_comm : "");
1114 kprintf("current thread = pri %d ", curthread->td_pri);
1115 if (curthread->td_pri >= TDPRI_CRIT)
1121 * we probably SHOULD have stopped the other CPUs before now!
1122 * another CPU COULD have been touching cpl at this moment...
1124 kprintf(" <- SMP: XXX");
1133 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1136 kprintf("trap number = %d\n", type);
1137 if (type <= MAX_TRAP_MSG)
1138 panic("%s", trap_msg[type]);
1140 panic("unknown/reserved trap");
1144 * Double fault handler. Called when a fault occurs while writing
1145 * a frame for a trap/exception onto the stack. This usually occurs
1146 * when the stack overflows (such is the case with infinite recursion,
1149 * XXX Note that the current PTD gets replaced by IdlePTD when the
1150 * task switch occurs. This means that the stack that was active at
1151 * the time of the double fault is not available at <kstack> unless
1152 * the machine was idle when the double fault occurred. The downside
1153 * of this is that "trace <ebp>" in ddb won't work.
1156 dblfault_handler(void)
1158 struct mdglobaldata *gd = mdcpu;
1160 kprintf("\nFatal double fault:\n");
1161 kprintf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1162 kprintf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1163 kprintf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
1165 /* three separate prints in case of a trap on an unmapped page */
1166 kprintf("mp_lock = %08x; ", mp_lock);
1167 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
1168 kprintf("lapic.id = %08x\n", lapic.id);
1170 panic("double fault");
1174 * Compensate for 386 brain damage (missing URKR).
1175 * This is a little simpler than the pagefault handler in trap() because
1176 * it the page tables have already been faulted in and high addresses
1177 * are thrown out early for other reasons.
1180 trapwrite(unsigned addr)
1187 va = trunc_page((vm_offset_t)addr);
1189 * XXX - MAX is END. Changed > to >= for temp. fix.
1191 if (va >= VM_MAX_USER_ADDRESS)
1194 lp = curthread->td_lwp;
1195 vm = lp->lwp_vmspace;
1197 PHOLD(lp->lwp_proc);
1199 if (!grow_stack(lp->lwp_proc, va)) {
1200 PRELE(lp->lwp_proc);
1205 * fault the data page
1207 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY);
1209 PRELE(lp->lwp_proc);
1211 if (rv != KERN_SUCCESS)
1218 * syscall2 - MP aware system call request C handler
1220 * A system call is essentially treated as a trap except that the
1221 * MP lock is not held on entry or return. We are responsible for
1222 * obtaining the MP lock if necessary and for handling ASTs
1223 * (e.g. a task switch) prior to return.
1225 * In general, only simple access and manipulation of curproc and
1226 * the current stack is allowed without having to hold MP lock.
1228 * MPSAFE - note that large sections of this routine are run without
1233 syscall2(struct trapframe *frame)
1235 struct thread *td = curthread;
1236 struct proc *p = td->td_proc;
1237 struct lwp *lp = td->td_lwp;
1239 struct sysent *callp;
1240 register_t orig_tf_eflags;
1245 int crit_count = td->td_pri & ~TDPRI_MASK;
1248 int have_mplock = 0;
1251 union sysunion args;
1254 if (ISPL(frame->tf_cs) != SEL_UPL) {
1261 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid,
1265 KASSERT(td->td_mpcount == 0, ("badmpcount syscall2 from %p", (void *)frame->tf_eip));
1266 if (syscall_mpsafe == 0)
1267 MAKEMPSAFE(have_mplock);
1269 userenter(td); /* lazy raise our priority */
1274 sticks = (int)td->td_sticks;
1275 orig_tf_eflags = frame->tf_eflags;
1278 * Virtual kernel intercept - if a VM context managed by a virtual
1279 * kernel issues a system call the virtual kernel handles it, not us.
1280 * Restore the virtual kernel context and return from its system
1281 * call. The current frame is copied out to the virtual kernel.
1283 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1284 error = vkernel_trap(lp, frame);
1285 frame->tf_eax = error;
1287 frame->tf_eflags |= PSL_C;
1288 error = EJUSTRETURN;
1293 * Get the system call parameters and account for time
1295 lp->lwp_md.md_regs = frame;
1296 params = (caddr_t)frame->tf_esp + sizeof(int);
1297 code = frame->tf_eax;
1299 if (p->p_sysent->sv_prepsyscall) {
1300 (*p->p_sysent->sv_prepsyscall)(
1301 frame, (int *)(&args.nosys.sysmsg + 1),
1305 * Need to check if this is a 32 bit or 64 bit syscall.
1306 * fuword is MP aware.
1308 if (code == SYS_syscall) {
1310 * Code is first argument, followed by actual args.
1312 code = fuword(params);
1313 params += sizeof(int);
1314 } else if (code == SYS___syscall) {
1316 * Like syscall, but code is a quad, so as to maintain
1317 * quad alignment for the rest of the arguments.
1319 code = fuword(params);
1320 params += sizeof(quad_t);
1324 code &= p->p_sysent->sv_mask;
1325 if (code >= p->p_sysent->sv_size)
1326 callp = &p->p_sysent->sv_table[0];
1328 callp = &p->p_sysent->sv_table[code];
1330 narg = callp->sy_narg & SYF_ARGMASK;
1333 * copyin is MP aware, but the tracing code is not
1335 if (narg && params) {
1336 error = copyin(params, (caddr_t)(&args.nosys.sysmsg + 1),
1337 narg * sizeof(register_t));
1340 if (KTRPOINT(td, KTR_SYSCALL)) {
1341 MAKEMPSAFE(have_mplock);
1343 ktrsyscall(lp, code, narg,
1344 (void *)(&args.nosys.sysmsg + 1));
1352 if (KTRPOINT(td, KTR_SYSCALL)) {
1353 MAKEMPSAFE(have_mplock);
1354 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1359 * For traditional syscall code edx is left untouched when 32 bit
1360 * results are returned. Since edx is loaded from fds[1] when the
1361 * system call returns we pre-set it here.
1363 args.sysmsg_fds[0] = 0;
1364 args.sysmsg_fds[1] = frame->tf_edx;
1367 * The syscall might manipulate the trap frame. If it does it
1368 * will probably return EJUSTRETURN.
1370 args.sysmsg_frame = frame;
1372 STOPEVENT(p, S_SCE, narg); /* MP aware */
1376 * Try to run the syscall without the MP lock if the syscall
1377 * is MP safe. We have to obtain the MP lock no matter what if
1380 if ((callp->sy_narg & SYF_MPSAFE) == 0)
1381 MAKEMPSAFE(have_mplock);
1384 error = (*callp->sy_call)(&args);
1388 * MP SAFE (we may or may not have the MP lock at this point)
1393 * Reinitialize proc pointer `p' as it may be different
1394 * if this is a child returning from fork syscall.
1397 lp = curthread->td_lwp;
1398 frame->tf_eax = args.sysmsg_fds[0];
1399 frame->tf_edx = args.sysmsg_fds[1];
1400 frame->tf_eflags &= ~PSL_C;
1404 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1405 * int 0x80 is 2 bytes. We saved this in tf_err.
1407 frame->tf_eip -= frame->tf_err;
1412 panic("Unexpected EASYNC return value (for now)");
1415 if (p->p_sysent->sv_errsize) {
1416 if (error >= p->p_sysent->sv_errsize)
1417 error = -1; /* XXX */
1419 error = p->p_sysent->sv_errtbl[error];
1421 frame->tf_eax = error;
1422 frame->tf_eflags |= PSL_C;
1427 * Traced syscall. trapsignal() is not MP aware.
1429 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1430 MAKEMPSAFE(have_mplock);
1431 frame->tf_eflags &= ~PSL_T;
1432 trapsignal(lp, SIGTRAP, 0);
1436 * Handle reschedule and other end-of-syscall issues
1438 userret(lp, frame, sticks);
1441 if (KTRPOINT(td, KTR_SYSRET)) {
1442 MAKEMPSAFE(have_mplock);
1443 ktrsysret(lp, code, error, args.sysmsg_result);
1448 * This works because errno is findable through the
1449 * register set. If we ever support an emulation where this
1450 * is not the case, this code will need to be revisited.
1452 STOPEVENT(p, S_SCX, code);
1457 * Release the MP lock if we had to get it
1459 KASSERT(td->td_mpcount == have_mplock,
1460 ("badmpcount syscall2/end from %p", (void *)frame->tf_eip));
1464 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error);
1466 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
1467 ("syscall: critical section count mismatch! %d/%d",
1468 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
1473 fork_return(struct lwp *lp, struct trapframe *frame)
1475 frame->tf_eax = 0; /* Child returns zero */
1476 frame->tf_eflags &= ~PSL_C; /* success */
1479 generic_lwp_return(lp, frame);
1480 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1484 * Simplified back end of syscall(), used when returning from fork()
1485 * directly into user mode. MP lock is held on entry and should be
1486 * released on return. This code will return back into the fork
1487 * trampoline code which then runs doreti.
1490 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1492 struct proc *p = lp->lwp_proc;
1495 * Newly forked processes are given a kernel priority. We have to
1496 * adjust the priority to a normal user priority and fake entry
1497 * into the kernel (call userenter()) to install a passive release
1498 * function just in case userret() decides to stop the process. This
1499 * can occur when ^Z races a fork. If we do not install the passive
1500 * release function the current process designation will not be
1501 * released when the thread goes to sleep.
1503 lwkt_setpri_self(TDPRI_USER_NORM);
1504 userenter(lp->lwp_thread);
1505 userret(lp, frame, 0);
1507 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1508 ktrsysret(lp, SYS_fork, 0, 0);
1510 p->p_flag |= P_PASSIVE_ACQ;
1512 p->p_flag &= ~P_PASSIVE_ACQ;
1514 KKASSERT(lp->lwp_thread->td_mpcount == 1);
1520 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1521 * fault (which is then passed back to the virtual kernel) if an attempt is
1522 * made to use the FP unit.
1524 * XXX this is a fairly big hack.
1527 set_vkernel_fp(struct trapframe *frame)
1529 struct thread *td = curthread;
1531 if (frame->tf_xflags & PGEX_FPFAULT) {
1532 td->td_pcb->pcb_flags |= FP_VIRTFP;
1533 if (mdcpu->gd_npxthread == td)
1536 td->td_pcb->pcb_flags &= ~FP_VIRTFP;