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/vkernel/i386/trap.c,v 1.35 2008/09/09 04:06:19 dillon Exp $
43 * 386 Trap and System call handling
50 #include "opt_ktrace.h"
52 #include <sys/param.h>
53 #include <sys/systm.h>
55 #include <sys/pioctl.h>
56 #include <sys/kernel.h>
57 #include <sys/resourcevar.h>
58 #include <sys/signalvar.h>
59 #include <sys/signal2.h>
60 #include <sys/syscall.h>
61 #include <sys/sysctl.h>
62 #include <sys/sysent.h>
64 #include <sys/vmmeter.h>
65 #include <sys/malloc.h>
67 #include <sys/ktrace.h>
70 #include <sys/upcall.h>
71 #include <sys/vkernel.h>
72 #include <sys/sysproto.h>
73 #include <sys/sysunion.h>
74 #include <sys/vmspace.h>
77 #include <vm/vm_param.h>
80 #include <vm/vm_kern.h>
81 #include <vm/vm_map.h>
82 #include <vm/vm_page.h>
83 #include <vm/vm_extern.h>
85 #include <machine/cpu.h>
86 #include <machine/md_var.h>
87 #include <machine/pcb.h>
88 #include <machine/smp.h>
89 #include <machine/tss.h>
90 #include <machine/globaldata.h>
92 #include <machine/vm86.h>
95 #include <sys/msgport2.h>
96 #include <sys/thread2.h>
100 #define MAKEMPSAFE(have_mplock) \
101 if (have_mplock == 0) { \
108 #define MAKEMPSAFE(have_mplock)
112 int (*pmath_emulate) (struct trapframe *);
114 extern int trapwrite (unsigned addr);
116 static int trap_pfault (struct trapframe *, int, vm_offset_t);
117 static void trap_fatal (struct trapframe *, int, vm_offset_t);
118 void dblfault_handler (void);
121 extern inthand_t IDTVEC(syscall);
124 #define MAX_TRAP_MSG 28
125 static char *trap_msg[] = {
127 "privileged instruction fault", /* 1 T_PRIVINFLT */
129 "breakpoint instruction fault", /* 3 T_BPTFLT */
132 "arithmetic trap", /* 6 T_ARITHTRAP */
133 "system forced exception", /* 7 T_ASTFLT */
135 "general protection fault", /* 9 T_PROTFLT */
136 "trace trap", /* 10 T_TRCTRAP */
138 "page fault", /* 12 T_PAGEFLT */
140 "alignment fault", /* 14 T_ALIGNFLT */
144 "integer divide fault", /* 18 T_DIVIDE */
145 "non-maskable interrupt trap", /* 19 T_NMI */
146 "overflow trap", /* 20 T_OFLOW */
147 "FPU bounds check fault", /* 21 T_BOUND */
148 "FPU device not available", /* 22 T_DNA */
149 "double fault", /* 23 T_DOUBLEFLT */
150 "FPU operand fetch fault", /* 24 T_FPOPFLT */
151 "invalid TSS fault", /* 25 T_TSSFLT */
152 "segment not present fault", /* 26 T_SEGNPFLT */
153 "stack fault", /* 27 T_STKFLT */
154 "machine check trap", /* 28 T_MCHK */
158 static int ddb_on_nmi = 1;
159 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
160 &ddb_on_nmi, 0, "Go to DDB on NMI");
162 static int panic_on_nmi = 1;
163 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
164 &panic_on_nmi, 0, "Panic on NMI");
165 static int fast_release;
166 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
167 &fast_release, 0, "Passive Release was optimal");
168 static int slow_release;
169 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
170 &slow_release, 0, "Passive Release was nonoptimal");
172 static int syscall_mpsafe = 1;
173 SYSCTL_INT(_kern, OID_AUTO, syscall_mpsafe, CTLFLAG_RW,
174 &syscall_mpsafe, 0, "Allow MPSAFE marked syscalls to run without BGL");
175 TUNABLE_INT("kern.syscall_mpsafe", &syscall_mpsafe);
176 static int trap_mpsafe = 1;
177 SYSCTL_INT(_kern, OID_AUTO, trap_mpsafe, CTLFLAG_RW,
178 &trap_mpsafe, 0, "Allow traps to mostly run without the BGL");
179 TUNABLE_INT("kern.trap_mpsafe", &trap_mpsafe);
182 MALLOC_DEFINE(M_SYSMSG, "sysmsg", "sysmsg structure");
183 extern int max_sysmsg;
186 * Passive USER->KERNEL transition. This only occurs if we block in the
187 * kernel while still holding our userland priority. We have to fixup our
188 * priority in order to avoid potential deadlocks before we allow the system
189 * to switch us to another thread.
192 passive_release(struct thread *td)
194 struct lwp *lp = td->td_lwp;
196 td->td_release = NULL;
197 lwkt_setpri_self(TDPRI_KERN_USER);
198 lp->lwp_proc->p_usched->release_curproc(lp);
202 * userenter() passively intercepts the thread switch function to increase
203 * the thread priority from a user priority to a kernel priority, reducing
204 * syscall and trap overhead for the case where no switch occurs.
208 userenter(struct thread *curtd)
210 curtd->td_release = passive_release;
214 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
215 * must be completed before we can return to or try to return to userland.
217 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
218 * arithmatic on the delta calculation so the absolute tick values are
219 * truncated to an integer.
222 userret(struct lwp *lp, struct trapframe *frame, int sticks)
224 struct proc *p = lp->lwp_proc;
228 * Charge system time if profiling. Note: times are in microseconds.
229 * This may do a copyout and block, so do it first even though it
230 * means some system time will be charged as user time.
232 if (p->p_flag & P_PROFIL) {
233 addupc_task(p, frame->tf_eip,
234 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
239 * If the jungle wants us dead, so be it.
241 if (lp->lwp_flag & LWP_WEXIT) {
244 rel_mplock(); /* NOT REACHED */
248 * Block here if we are in a stopped state.
250 if (p->p_stat == SSTOP) {
258 * Post any pending upcalls
260 if (p->p_flag & P_UPCALLPEND) {
262 p->p_flag &= ~P_UPCALLPEND;
269 * Post any pending signals
271 if ((sig = CURSIG(lp)) != 0) {
279 * block here if we are swapped out, but still process signals
280 * (such as SIGKILL). proc0 (the swapin scheduler) is already
281 * aware of our situation, we do not have to wake it up.
283 if (p->p_flag & P_SWAPPEDOUT) {
285 p->p_flag |= P_SWAPWAIT;
287 if (p->p_flag & P_SWAPWAIT)
288 tsleep(p, PCATCH, "SWOUT", 0);
289 p->p_flag &= ~P_SWAPWAIT;
295 * Make sure postsig() handled request to restore old signal mask after
296 * running signal handler.
298 KKASSERT((lp->lwp_flag & LWP_OLDMASK) == 0);
302 * Cleanup from userenter and any passive release that might have occured.
303 * We must reclaim the current-process designation before we can return
304 * to usermode. We also handle both LWKT and USER reschedule requests.
307 userexit(struct lwp *lp)
309 struct thread *td = lp->lwp_thread;
310 globaldata_t gd = td->td_gd;
314 * If a user reschedule is requested force a new process to be
315 * chosen by releasing the current process. Our process will only
316 * be chosen again if it has a considerably better priority.
318 if (user_resched_wanted())
319 lp->lwp_proc->p_usched->release_curproc(lp);
323 * Handle a LWKT reschedule request first. Since our passive release
324 * is still in place we do not have to do anything special.
326 while (lwkt_resched_wanted()) {
330 * The thread that preempted us may have stopped our process.
332 while (lp->lwp_proc->p_stat == SSTOP) {
340 * Acquire the current process designation for this user scheduler
341 * on this cpu. This will also handle any user-reschedule requests.
343 lp->lwp_proc->p_usched->acquire_curproc(lp);
344 /* We may have switched cpus on acquisition */
348 * Reduce our priority in preparation for a return to userland. If
349 * our passive release function was still in place, our priority was
350 * never raised and does not need to be reduced.
352 * Note that at this point there may be other LWKT thread at
353 * TDPRI_KERN_USER (aka higher then our currenet priority). We
354 * do NOT want to run these threads yet.
356 if (td->td_release == NULL)
357 lwkt_setpri_self(TDPRI_USER_NORM);
358 td->td_release = NULL;
362 #if !defined(KTR_KERNENTRY)
363 #define KTR_KERNENTRY KTR_ALL
365 KTR_INFO_MASTER(kernentry);
366 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, "pid=%d, tid=%d, trapno=%d, eva=%p",
367 sizeof(int) + sizeof(int) + sizeof(int) + sizeof(vm_offset_t));
368 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "pid=%d, tid=%d",
369 sizeof(int) + sizeof(int));
370 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "pid=%d, tid=%d, call=%d",
371 sizeof(int) + sizeof(int) + sizeof(int));
372 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "pid=%d, tid=%d, err=%d",
373 sizeof(int) + sizeof(int) + sizeof(int));
374 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "pid=%d, tid=%d",
375 sizeof(int) + sizeof(int));
378 * Exception, fault, and trap interface to the kernel.
379 * This common code is called from assembly language IDT gate entry
380 * routines that prepare a suitable stack frame, and restore this
381 * frame after the exception has been processed.
383 * This function is also called from doreti in an interlock to handle ASTs.
384 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
386 * NOTE! We have to retrieve the fault address prior to obtaining the
387 * MP lock because get_mplock() may switch out. YYY cr2 really ought
388 * to be retrieved by the assembly code, not here.
390 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
391 * if an attempt is made to switch from a fast interrupt or IPI. This is
392 * necessary to properly take fatal kernel traps on SMP machines if
393 * get_mplock() has to block.
397 user_trap(struct trapframe *frame)
399 struct globaldata *gd = mycpu;
400 struct thread *td = gd->gd_curthread;
401 struct lwp *lp = td->td_lwp;
404 int i = 0, ucode = 0, type, code;
409 int crit_count = td->td_pri & ~TDPRI_MASK;
416 * This is a bad kludge to avoid changing the various trapframe
417 * structures. Because we are enabled as a virtual kernel,
418 * the original tf_err field will be passed to us shifted 16
419 * over in the tf_trapno field for T_PAGEFLT.
421 if (frame->tf_trapno == T_PAGEFLT)
426 kprintf("USER_TRAP AT %08x xflags %d trapno %d eva %08x\n",
427 frame->tf_eip, frame->tf_xflags, frame->tf_trapno, eva);
431 * Everything coming from user mode runs through user_trap,
432 * including system calls.
434 if (frame->tf_trapno == T_SYSCALL80) {
439 KTR_LOG(kernentry_trap, lp->lwp_proc->p_pid, lp->lwp_tid,
440 frame->tf_trapno, eva);
444 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
445 ++gd->gd_trap_nesting_level;
446 MAKEMPSAFE(have_mplock);
447 trap_fatal(frame, TRUE, eva);
448 --gd->gd_trap_nesting_level;
453 ++gd->gd_trap_nesting_level;
455 if (trap_mpsafe == 0)
456 MAKEMPSAFE(have_mplock);
459 --gd->gd_trap_nesting_level;
461 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
464 type = frame->tf_trapno;
465 code = frame->tf_err;
469 sticks = (int)td->td_sticks;
470 lp->lwp_md.md_regs = frame;
473 case T_PRIVINFLT: /* privileged instruction fault */
478 case T_BPTFLT: /* bpt instruction fault */
479 case T_TRCTRAP: /* trace trap */
480 frame->tf_eflags &= ~PSL_T;
484 case T_ARITHTRAP: /* arithmetic trap */
489 case T_ASTFLT: /* Allow process switch */
490 mycpu->gd_cnt.v_soft++;
491 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
492 atomic_clear_int_nonlocked(&mycpu->gd_reqflags,
494 addupc_task(p, p->p_prof.pr_addr,
500 * The following two traps can happen in
501 * vm86 mode, and, if so, we want to handle
504 case T_PROTFLT: /* general protection fault */
505 case T_STKFLT: /* stack fault */
507 if (frame->tf_eflags & PSL_VM) {
508 i = vm86_emulate((struct vm86frame *)frame);
516 case T_SEGNPFLT: /* segment not present fault */
517 case T_TSSFLT: /* invalid TSS fault */
518 case T_DOUBLEFLT: /* double fault */
520 ucode = code + BUS_SEGM_FAULT ;
524 case T_PAGEFLT: /* page fault */
525 MAKEMPSAFE(have_mplock);
526 i = trap_pfault(frame, TRUE, eva);
529 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
539 case T_DIVIDE: /* integer divide fault */
546 MAKEMPSAFE(have_mplock);
547 /* machine/parity/power fail/"kitchen sink" faults */
548 if (isa_nmi(code) == 0) {
551 * NMI can be hooked up to a pushbutton
555 kprintf ("NMI ... going to debugger\n");
556 kdb_trap (type, 0, frame);
560 } else if (panic_on_nmi)
561 panic("NMI indicates hardware failure");
563 #endif /* NISA > 0 */
565 case T_OFLOW: /* integer overflow fault */
570 case T_BOUND: /* bounds check fault */
577 * Virtual kernel intercept - pass the DNA exception
578 * to the (emulated) virtual kernel if it asked to handle
579 * it. This occurs when the virtual kernel is holding
580 * onto the FP context for a different emulated
581 * process then the one currently running.
583 * We must still call npxdna() since we may have
584 * saved FP state that the (emulated) virtual kernel
585 * needs to hand over to a different emulated process.
587 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
588 (td->td_pcb->pcb_flags & FP_VIRTFP)
595 * The kernel may have switched out the FP unit's
596 * state, causing the user process to take a fault
597 * when it tries to use the FP unit. Restore the
603 if (!pmath_emulate) {
605 ucode = FPE_FPU_NP_TRAP;
608 i = (*pmath_emulate)(frame);
610 if (!(frame->tf_eflags & PSL_T))
612 frame->tf_eflags &= ~PSL_T;
615 /* else ucode = emulator_only_knows() XXX */
618 case T_FPOPFLT: /* FPU operand fetch fault */
623 case T_XMMFLT: /* SIMD floating-point exception */
630 * Virtual kernel intercept - if the fault is directly related to a
631 * VM context managed by a virtual kernel then let the virtual kernel
634 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
635 vkernel_trap(lp, frame);
640 * Translate fault for emulators (e.g. Linux)
642 if (*p->p_sysent->sv_transtrap)
643 i = (*p->p_sysent->sv_transtrap)(i, type);
645 MAKEMPSAFE(have_mplock);
646 trapsignal(lp, i, ucode);
649 if (type <= MAX_TRAP_MSG) {
650 uprintf("fatal process exception: %s",
652 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
653 uprintf(", fault VA = 0x%lx", (u_long)eva);
660 KASSERT(td->td_mpcount == have_mplock, ("badmpcount trap/end from %p", (void *)frame->tf_eip));
662 userret(lp, frame, sticks);
669 KTR_LOG(kernentry_trap_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
671 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
672 ("syscall: critical section count mismatch! %d/%d",
673 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
678 kern_trap(struct trapframe *frame)
680 struct globaldata *gd = mycpu;
681 struct thread *td = gd->gd_curthread;
684 int i = 0, ucode = 0, type, code;
689 int crit_count = td->td_pri & ~TDPRI_MASK;
696 if (frame->tf_trapno == T_PAGEFLT)
703 ++gd->gd_trap_nesting_level;
704 MAKEMPSAFE(have_mplock);
705 trap_fatal(frame, FALSE, eva);
706 --gd->gd_trap_nesting_level;
711 ++gd->gd_trap_nesting_level;
714 if (trap_mpsafe == 0)
715 MAKEMPSAFE(have_mplock);
718 --gd->gd_trap_nesting_level;
720 type = frame->tf_trapno;
721 code = frame->tf_err;
729 case T_PAGEFLT: /* page fault */
730 MAKEMPSAFE(have_mplock);
731 trap_pfault(frame, FALSE, eva);
737 * The kernel may be using npx for copying or other
740 panic("kernel NPX should not happen");
746 case T_PROTFLT: /* general protection fault */
747 case T_SEGNPFLT: /* segment not present fault */
749 * Invalid segment selectors and out of bounds
750 * %eip's and %esp's can be set up in user mode.
751 * This causes a fault in kernel mode when the
752 * kernel tries to return to user mode. We want
753 * to get this fault so that we can fix the
754 * problem here and not have to check all the
755 * selectors and pointers when the user changes
758 if (mycpu->gd_intr_nesting_level == 0) {
759 if (td->td_pcb->pcb_onfault) {
761 (register_t)td->td_pcb->pcb_onfault;
769 * PSL_NT can be set in user mode and isn't cleared
770 * automatically when the kernel is entered. This
771 * causes a TSS fault when the kernel attempts to
772 * `iret' because the TSS link is uninitialized. We
773 * want to get this fault so that we can fix the
774 * problem here and not every time the kernel is
777 if (frame->tf_eflags & PSL_NT) {
778 frame->tf_eflags &= ~PSL_NT;
783 case T_TRCTRAP: /* trace trap */
785 if (frame->tf_eip == (int)IDTVEC(syscall)) {
787 * We've just entered system mode via the
788 * syscall lcall. Continue single stepping
789 * silently until the syscall handler has
794 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
796 * The syscall handler has now saved the
797 * flags. Stop single stepping it.
799 frame->tf_eflags &= ~PSL_T;
805 * Ignore debug register trace traps due to
806 * accesses in the user's address space, which
807 * can happen under several conditions such as
808 * if a user sets a watchpoint on a buffer and
809 * then passes that buffer to a system call.
810 * We still want to get TRCTRAPS for addresses
811 * in kernel space because that is useful when
812 * debugging the kernel.
814 if (user_dbreg_trap()) {
816 * Reset breakpoint bits because the
819 load_dr6(rdr6() & 0xfffffff0);
824 * Fall through (TRCTRAP kernel mode, kernel address)
828 * If DDB is enabled, let it handle the debugger trap.
829 * Otherwise, debugger traps "can't happen".
832 MAKEMPSAFE(have_mplock);
833 if (kdb_trap (type, 0, frame))
838 MAKEMPSAFE(have_mplock);
839 trap_fatal(frame, FALSE, eva);
842 MAKEMPSAFE(have_mplock);
843 trap_fatal(frame, FALSE, eva);
847 * Ignore this trap generated from a spurious SIGTRAP.
849 * single stepping in / syscalls leads to spurious / SIGTRAP
852 * Haiku (c) 2007 Simon 'corecode' Schubert
858 * Translate fault for emulators (e.g. Linux)
860 if (*p->p_sysent->sv_transtrap)
861 i = (*p->p_sysent->sv_transtrap)(i, type);
863 MAKEMPSAFE(have_mplock);
864 trapsignal(lp, i, ucode);
867 if (type <= MAX_TRAP_MSG) {
868 uprintf("fatal process exception: %s",
870 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
871 uprintf(", fault VA = 0x%lx", (u_long)eva);
883 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
884 ("syscall: critical section count mismatch! %d/%d",
885 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
890 trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
893 struct vmspace *vm = NULL;
897 thread_t td = curthread;
898 struct lwp *lp = td->td_lwp;
900 va = trunc_page(eva);
901 if (usermode == FALSE) {
903 * This is a fault on kernel virtual memory.
908 * This is a fault on non-kernel virtual memory.
909 * vm is initialized above to NULL. If curproc is NULL
910 * or curproc->p_vmspace is NULL the fault is fatal.
913 vm = lp->lwp_vmspace;
921 if (frame->tf_xflags & PGEX_W)
922 ftype = VM_PROT_READ | VM_PROT_WRITE;
924 ftype = VM_PROT_READ;
926 if (map != &kernel_map) {
928 * Keep swapout from messing with us during this
934 * Grow the stack if necessary
936 /* grow_stack returns false only if va falls into
937 * a growable stack region and the stack growth
938 * fails. It returns true if va was not within
939 * a growable stack region, or if the stack
942 if (!grow_stack (lp->lwp_proc, va)) {
948 /* Fault in the user page: */
949 rv = vm_fault(map, va, ftype,
950 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
956 * Don't have to worry about process locking or stacks in the kernel.
958 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
961 if (rv == KERN_SUCCESS)
965 if (td->td_gd->gd_intr_nesting_level == 0 &&
966 td->td_pcb->pcb_onfault) {
967 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
970 trap_fatal(frame, usermode, eva);
973 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
977 trap_fatal(struct trapframe *frame, int usermode, vm_offset_t eva)
979 int code, type, ss, esp;
981 code = frame->tf_xflags;
982 type = frame->tf_trapno;
984 if (type <= MAX_TRAP_MSG) {
985 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
986 type, trap_msg[type],
987 (usermode ? "user" : "kernel"));
990 /* two separate prints in case of a trap on an unmapped page */
991 kprintf("mp_lock = %08x; ", mp_lock);
992 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
994 if (type == T_PAGEFLT) {
995 kprintf("fault virtual address = 0x%x\n", eva);
996 kprintf("fault code = %s %s, %s\n",
997 usermode ? "user" : "supervisor",
998 code & PGEX_W ? "write" : "read",
999 code & PGEX_P ? "protection violation" : "page not present");
1001 kprintf("instruction pointer = 0x%x:0x%x\n",
1002 frame->tf_cs & 0xffff, frame->tf_eip);
1004 ss = frame->tf_ss & 0xffff;
1005 esp = frame->tf_esp;
1007 ss = GSEL(GDATA_SEL, SEL_KPL);
1008 esp = (int)&frame->tf_esp;
1010 kprintf("stack pointer = 0x%x:0x%x\n", ss, esp);
1011 kprintf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
1012 kprintf("processor eflags = ");
1013 if (frame->tf_eflags & PSL_T)
1014 kprintf("trace trap, ");
1015 if (frame->tf_eflags & PSL_I)
1016 kprintf("interrupt enabled, ");
1017 if (frame->tf_eflags & PSL_NT)
1018 kprintf("nested task, ");
1019 if (frame->tf_eflags & PSL_RF)
1020 kprintf("resume, ");
1022 if (frame->tf_eflags & PSL_VM)
1025 kprintf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
1026 kprintf("current process = ");
1028 kprintf("%lu (%s)\n",
1029 (u_long)curproc->p_pid, curproc->p_comm ?
1030 curproc->p_comm : "");
1034 kprintf("current thread = pri %d ", curthread->td_pri);
1035 if (curthread->td_pri >= TDPRI_CRIT)
1041 * we probably SHOULD have stopped the other CPUs before now!
1042 * another CPU COULD have been touching cpl at this moment...
1044 kprintf(" <- SMP: XXX");
1053 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1056 kprintf("trap number = %d\n", type);
1057 if (type <= MAX_TRAP_MSG)
1058 panic("%s", trap_msg[type]);
1060 panic("unknown/reserved trap");
1064 * Double fault handler. Called when a fault occurs while writing
1065 * a frame for a trap/exception onto the stack. This usually occurs
1066 * when the stack overflows (such is the case with infinite recursion,
1069 * XXX Note that the current PTD gets replaced by IdlePTD when the
1070 * task switch occurs. This means that the stack that was active at
1071 * the time of the double fault is not available at <kstack> unless
1072 * the machine was idle when the double fault occurred. The downside
1073 * of this is that "trace <ebp>" in ddb won't work.
1076 dblfault_handler(void)
1078 struct mdglobaldata *gd = mdcpu;
1080 kprintf("\nFatal double fault:\n");
1081 kprintf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1082 kprintf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1083 kprintf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
1085 /* two separate prints in case of a trap on an unmapped page */
1086 kprintf("mp_lock = %08x; ", mp_lock);
1087 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
1089 panic("double fault");
1093 * Compensate for 386 brain damage (missing URKR).
1094 * This is a little simpler than the pagefault handler in trap() because
1095 * it the page tables have already been faulted in and high addresses
1096 * are thrown out early for other reasons.
1099 trapwrite(unsigned addr)
1106 va = trunc_page((vm_offset_t)addr);
1108 * XXX - MAX is END. Changed > to >= for temp. fix.
1110 if (va >= VM_MAX_USER_ADDRESS)
1113 lp = curthread->td_lwp;
1114 vm = lp->lwp_vmspace;
1116 PHOLD(lp->lwp_proc);
1118 if (!grow_stack (lp->lwp_proc, va)) {
1119 PRELE(lp->lwp_proc);
1124 * fault the data page
1126 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY);
1128 PRELE(lp->lwp_proc);
1130 if (rv != KERN_SUCCESS)
1137 * syscall2 - MP aware system call request C handler
1139 * A system call is essentially treated as a trap except that the
1140 * MP lock is not held on entry or return. We are responsible for
1141 * obtaining the MP lock if necessary and for handling ASTs
1142 * (e.g. a task switch) prior to return.
1144 * In general, only simple access and manipulation of curproc and
1145 * the current stack is allowed without having to hold MP lock.
1147 * MPSAFE - note that large sections of this routine are run without
1152 syscall2(struct trapframe *frame)
1154 struct thread *td = curthread;
1155 struct proc *p = td->td_proc;
1156 struct lwp *lp = td->td_lwp;
1158 struct sysent *callp;
1159 register_t orig_tf_eflags;
1164 int crit_count = td->td_pri & ~TDPRI_MASK;
1167 int have_mplock = 0;
1170 union sysunion args;
1172 KTR_LOG(kernentry_syscall, lp->lwp_proc->p_pid, lp->lwp_tid,
1176 KASSERT(td->td_mpcount == 0, ("badmpcount syscall2 from %p", (void *)frame->tf_eip));
1177 if (syscall_mpsafe == 0)
1178 MAKEMPSAFE(have_mplock);
1180 userenter(td); /* lazy raise our priority */
1185 sticks = (int)td->td_sticks;
1186 orig_tf_eflags = frame->tf_eflags;
1189 * Virtual kernel intercept - if a VM context managed by a virtual
1190 * kernel issues a system call the virtual kernel handles it, not us.
1191 * Restore the virtual kernel context and return from its system
1192 * call. The current frame is copied out to the virtual kernel.
1194 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1195 error = vkernel_trap(lp, frame);
1196 frame->tf_eax = error;
1198 frame->tf_eflags |= PSL_C;
1199 error = EJUSTRETURN;
1204 * Get the system call parameters and account for time
1206 lp->lwp_md.md_regs = frame;
1207 params = (caddr_t)frame->tf_esp + sizeof(int);
1208 code = frame->tf_eax;
1210 if (p->p_sysent->sv_prepsyscall) {
1211 (*p->p_sysent->sv_prepsyscall)(
1212 frame, (int *)(&args.nosys.sysmsg + 1),
1216 * Need to check if this is a 32 bit or 64 bit syscall.
1217 * fuword is MP aware.
1219 if (code == SYS_syscall) {
1221 * Code is first argument, followed by actual args.
1223 code = fuword(params);
1224 params += sizeof(int);
1225 } else if (code == SYS___syscall) {
1227 * Like syscall, but code is a quad, so as to maintain
1228 * quad alignment for the rest of the arguments.
1230 code = fuword(params);
1231 params += sizeof(quad_t);
1235 code &= p->p_sysent->sv_mask;
1236 if (code >= p->p_sysent->sv_size)
1237 callp = &p->p_sysent->sv_table[0];
1239 callp = &p->p_sysent->sv_table[code];
1241 narg = callp->sy_narg & SYF_ARGMASK;
1244 * copyin is MP aware, but the tracing code is not
1246 if (narg && params) {
1247 error = copyin(params, (caddr_t)(&args.nosys.sysmsg + 1),
1248 narg * sizeof(register_t));
1251 if (KTRPOINT(td, KTR_SYSCALL)) {
1252 MAKEMPSAFE(have_mplock);
1254 ktrsyscall(lp, code, narg,
1255 (void *)(&args.nosys.sysmsg + 1));
1263 if (KTRPOINT(td, KTR_SYSCALL)) {
1264 MAKEMPSAFE(have_mplock);
1265 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1270 * For traditional syscall code edx is left untouched when 32 bit
1271 * results are returned. Since edx is loaded from fds[1] when the
1272 * system call returns we pre-set it here.
1274 args.sysmsg_fds[0] = 0;
1275 args.sysmsg_fds[1] = frame->tf_edx;
1278 * The syscall might manipulate the trap frame. If it does it
1279 * will probably return EJUSTRETURN.
1281 args.sysmsg_frame = frame;
1283 STOPEVENT(p, S_SCE, narg); /* MP aware */
1287 * Try to run the syscall without the MP lock if the syscall
1288 * is MP safe. We have to obtain the MP lock no matter what if
1291 if ((callp->sy_narg & SYF_MPSAFE) == 0)
1292 MAKEMPSAFE(have_mplock);
1295 error = (*callp->sy_call)(&args);
1298 kprintf("system call %d returned %d\n", code, error);
1303 * MP SAFE (we may or may not have the MP lock at this point)
1308 * Reinitialize proc pointer `p' as it may be different
1309 * if this is a child returning from fork syscall.
1312 lp = curthread->td_lwp;
1313 frame->tf_eax = args.sysmsg_fds[0];
1314 frame->tf_edx = args.sysmsg_fds[1];
1315 frame->tf_eflags &= ~PSL_C;
1319 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1320 * int 0x80 is 2 bytes. We saved this in tf_err.
1322 frame->tf_eip -= frame->tf_err;
1327 panic("Unexpected EASYNC return value (for now)");
1330 if (p->p_sysent->sv_errsize) {
1331 if (error >= p->p_sysent->sv_errsize)
1332 error = -1; /* XXX */
1334 error = p->p_sysent->sv_errtbl[error];
1336 frame->tf_eax = error;
1337 frame->tf_eflags |= PSL_C;
1342 * Traced syscall. trapsignal() is not MP aware.
1344 if ((orig_tf_eflags & PSL_T) /*&& !(orig_tf_eflags & PSL_VM)*/) {
1345 MAKEMPSAFE(have_mplock);
1346 frame->tf_eflags &= ~PSL_T;
1347 trapsignal(lp, SIGTRAP, 0);
1351 * Handle reschedule and other end-of-syscall issues
1353 userret(lp, frame, sticks);
1356 if (KTRPOINT(td, KTR_SYSRET)) {
1357 MAKEMPSAFE(have_mplock);
1358 ktrsysret(lp, code, error, args.sysmsg_result);
1363 * This works because errno is findable through the
1364 * register set. If we ever support an emulation where this
1365 * is not the case, this code will need to be revisited.
1367 STOPEVENT(p, S_SCX, code);
1372 * Release the MP lock if we had to get it
1374 KASSERT(td->td_mpcount == have_mplock,
1375 ("badmpcount syscall2/end from %p", (void *)frame->tf_eip));
1379 KTR_LOG(kernentry_syscall_ret, lp->lwp_proc->p_pid, lp->lwp_tid, error);
1381 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
1382 ("syscall: critical section count mismatch! %d/%d",
1383 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
1388 fork_return(struct lwp *lp, struct trapframe *frame)
1390 frame->tf_eax = 0; /* Child returns zero */
1391 frame->tf_eflags &= ~PSL_C; /* success */
1394 generic_lwp_return(lp, frame);
1395 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1399 * Simplified back end of syscall(), used when returning from fork()
1400 * or lwp_create() directly into user mode. MP lock is held on entry and
1401 * should be released on return. This code will return back into the fork
1402 * trampoline code which then runs doreti.
1405 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1407 struct proc *p = lp->lwp_proc;
1410 * Newly forked processes are given a kernel priority. We have to
1411 * adjust the priority to a normal user priority and fake entry
1412 * into the kernel (call userenter()) to install a passive release
1413 * function just in case userret() decides to stop the process. This
1414 * can occur when ^Z races a fork. If we do not install the passive
1415 * release function the current process designation will not be
1416 * released when the thread goes to sleep.
1418 lwkt_setpri_self(TDPRI_USER_NORM);
1419 userenter(lp->lwp_thread);
1420 userret(lp, frame, 0);
1422 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1423 ktrsysret(lp, SYS_fork, 0, 0);
1425 p->p_flag |= P_PASSIVE_ACQ;
1427 p->p_flag &= ~P_PASSIVE_ACQ;
1429 KKASSERT(lp->lwp_thread->td_mpcount == 1);
1435 * doreti has turned into this. The frame is directly on the stack. We
1436 * pull everything else we need (fpu and tls context) from the current
1439 * Note on fpu interactions: In a virtual kernel, the fpu context for
1440 * an emulated user mode process is not shared with the virtual kernel's
1441 * fpu context, so we only have to 'stack' fpu contexts within the virtual
1442 * kernel itself, and not even then since the signal() contexts that we care
1443 * about save and restore the FPU state (I think anyhow).
1445 * vmspace_ctl() returns an error only if it had problems instaling the
1446 * context we supplied or problems copying data to/from our VM space.
1449 go_user(struct intrframe *frame)
1451 struct trapframe *tf = (void *)&frame->if_gs;
1455 * Interrupts may be disabled on entry, make sure all signals
1456 * can be received before beginning our loop.
1461 * Switch to the current simulated user process, then call
1462 * user_trap() when we break out of it (usually due to a signal).
1466 * Tell the real kernel whether it is ok to use the FP
1469 if (mdcpu->gd_npxthread == curthread) {
1470 tf->tf_xflags &= ~PGEX_FPFAULT;
1472 tf->tf_xflags |= PGEX_FPFAULT;
1476 * Run emulated user process context. This call interlocks
1477 * with new mailbox signals.
1479 * Set PGEX_U unconditionally, indicating a user frame (the
1480 * bit is normally set only by T_PAGEFLT).
1482 r = vmspace_ctl(&curproc->p_vmspace->vm_pmap, VMSPACE_CTL_RUN,
1483 tf, &curthread->td_savevext);
1484 frame->if_xflags |= PGEX_U;
1486 kprintf("GO USER %d trap %d EVA %08x EIP %08x ESP %08x XFLAGS %02x/%02x\n",
1487 r, tf->tf_trapno, tf->tf_err, tf->tf_eip, tf->tf_esp,
1488 tf->tf_xflags, frame->if_xflags);
1492 panic("vmspace_ctl failed");
1494 if (tf->tf_trapno) {
1496 } else if (mycpu->gd_reqflags & RQF_AST_MASK) {
1497 tf->tf_trapno = T_ASTFLT;
1506 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1507 * fault (which is then passed back to the virtual kernel) if an attempt is
1508 * made to use the FP unit.
1510 * XXX this is a fairly big hack.
1513 set_vkernel_fp(struct trapframe *frame)
1515 struct thread *td = curthread;
1517 if (frame->tf_xflags & PGEX_FPFAULT) {
1518 td->td_pcb->pcb_flags |= FP_VIRTFP;
1519 if (mdcpu->gd_npxthread == td)
1522 td->td_pcb->pcb_flags &= ~FP_VIRTFP;