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 1.33 2008/05/19 10:28:06 corecode 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 if (lwkt_resched_wanted())
330 * Acquire the current process designation for this user scheduler
331 * on this cpu. This will also handle any user-reschedule requests.
333 lp->lwp_proc->p_usched->acquire_curproc(lp);
334 /* We may have switched cpus on acquisition */
338 * Reduce our priority in preparation for a return to userland. If
339 * our passive release function was still in place, our priority was
340 * never raised and does not need to be reduced.
342 if (td->td_release == NULL)
343 lwkt_setpri_self(TDPRI_USER_NORM);
344 td->td_release = NULL;
347 * After reducing our priority there might be other kernel-level
348 * LWKTs that now have a greater priority. Run them as necessary.
349 * We don't have to worry about losing cpu to userland because
350 * we still control the current-process designation and we no longer
351 * have a passive release function installed.
353 if (lwkt_checkpri_self())
358 #if !defined(KTR_KERNENTRY)
359 #define KTR_KERNENTRY KTR_ALL
361 KTR_INFO_MASTER(kernentry);
362 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, "pid=%d, tid=%d, trapno=%d, eva=%p",
363 sizeof(int) + sizeof(int) + sizeof(int) + sizeof(vm_offset_t));
364 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "pid=%d, tid=%d",
365 sizeof(int) + sizeof(int));
366 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "pid=%d, tid=%d, call=%d",
367 sizeof(int) + sizeof(int) + sizeof(int));
368 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "pid=%d, tid=%d, err=%d",
369 sizeof(int) + sizeof(int) + sizeof(int));
370 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "pid=%d, tid=%d",
371 sizeof(int) + sizeof(int));
374 * Exception, fault, and trap interface to the kernel.
375 * This common code is called from assembly language IDT gate entry
376 * routines that prepare a suitable stack frame, and restore this
377 * frame after the exception has been processed.
379 * This function is also called from doreti in an interlock to handle ASTs.
380 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
382 * NOTE! We have to retrieve the fault address prior to obtaining the
383 * MP lock because get_mplock() may switch out. YYY cr2 really ought
384 * to be retrieved by the assembly code, not here.
386 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
387 * if an attempt is made to switch from a fast interrupt or IPI. This is
388 * necessary to properly take fatal kernel traps on SMP machines if
389 * get_mplock() has to block.
393 user_trap(struct trapframe *frame)
395 struct globaldata *gd = mycpu;
396 struct thread *td = gd->gd_curthread;
397 struct lwp *lp = td->td_lwp;
400 int i = 0, ucode = 0, type, code;
405 int crit_count = td->td_pri & ~TDPRI_MASK;
412 * This is a bad kludge to avoid changing the various trapframe
413 * structures. Because we are enabled as a virtual kernel,
414 * the original tf_err field will be passed to us shifted 16
415 * over in the tf_trapno field for T_PAGEFLT.
417 if (frame->tf_trapno == T_PAGEFLT)
422 kprintf("USER_TRAP AT %08x xflags %d trapno %d eva %08x\n",
423 frame->tf_eip, frame->tf_xflags, frame->tf_trapno, eva);
427 * Everything coming from user mode runs through user_trap,
428 * including system calls.
430 if (frame->tf_trapno == T_SYSCALL80) {
435 KTR_LOG(kernentry_trap, lp->lwp_proc->p_pid, lp->lwp_tid,
436 frame->tf_trapno, eva);
440 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
441 ++gd->gd_trap_nesting_level;
442 MAKEMPSAFE(have_mplock);
443 trap_fatal(frame, TRUE, eva);
444 --gd->gd_trap_nesting_level;
449 ++gd->gd_trap_nesting_level;
451 if (trap_mpsafe == 0)
452 MAKEMPSAFE(have_mplock);
455 --gd->gd_trap_nesting_level;
457 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
460 type = frame->tf_trapno;
461 code = frame->tf_err;
465 sticks = (int)td->td_sticks;
466 lp->lwp_md.md_regs = frame;
469 case T_PRIVINFLT: /* privileged instruction fault */
474 case T_BPTFLT: /* bpt instruction fault */
475 case T_TRCTRAP: /* trace trap */
476 frame->tf_eflags &= ~PSL_T;
480 case T_ARITHTRAP: /* arithmetic trap */
485 case T_ASTFLT: /* Allow process switch */
486 mycpu->gd_cnt.v_soft++;
487 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
488 atomic_clear_int_nonlocked(&mycpu->gd_reqflags,
490 addupc_task(p, p->p_prof.pr_addr,
496 * The following two traps can happen in
497 * vm86 mode, and, if so, we want to handle
500 case T_PROTFLT: /* general protection fault */
501 case T_STKFLT: /* stack fault */
503 if (frame->tf_eflags & PSL_VM) {
504 i = vm86_emulate((struct vm86frame *)frame);
512 case T_SEGNPFLT: /* segment not present fault */
513 case T_TSSFLT: /* invalid TSS fault */
514 case T_DOUBLEFLT: /* double fault */
516 ucode = code + BUS_SEGM_FAULT ;
520 case T_PAGEFLT: /* page fault */
521 MAKEMPSAFE(have_mplock);
522 i = trap_pfault(frame, TRUE, eva);
525 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
535 case T_DIVIDE: /* integer divide fault */
542 MAKEMPSAFE(have_mplock);
543 /* machine/parity/power fail/"kitchen sink" faults */
544 if (isa_nmi(code) == 0) {
547 * NMI can be hooked up to a pushbutton
551 kprintf ("NMI ... going to debugger\n");
552 kdb_trap (type, 0, frame);
556 } else if (panic_on_nmi)
557 panic("NMI indicates hardware failure");
559 #endif /* NISA > 0 */
561 case T_OFLOW: /* integer overflow fault */
566 case T_BOUND: /* bounds check fault */
573 * Virtual kernel intercept - pass the DNA exception
574 * to the (emulated) virtual kernel if it asked to handle
575 * it. This occurs when the virtual kernel is holding
576 * onto the FP context for a different emulated
577 * process then the one currently running.
579 * We must still call npxdna() since we may have
580 * saved FP state that the (emulated) virtual kernel
581 * needs to hand over to a different emulated process.
583 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
584 (td->td_pcb->pcb_flags & FP_VIRTFP)
591 * The kernel may have switched out the FP unit's
592 * state, causing the user process to take a fault
593 * when it tries to use the FP unit. Restore the
599 if (!pmath_emulate) {
601 ucode = FPE_FPU_NP_TRAP;
604 i = (*pmath_emulate)(frame);
606 if (!(frame->tf_eflags & PSL_T))
608 frame->tf_eflags &= ~PSL_T;
611 /* else ucode = emulator_only_knows() XXX */
614 case T_FPOPFLT: /* FPU operand fetch fault */
619 case T_XMMFLT: /* SIMD floating-point exception */
626 * Virtual kernel intercept - if the fault is directly related to a
627 * VM context managed by a virtual kernel then let the virtual kernel
630 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
631 vkernel_trap(lp, frame);
636 * Translate fault for emulators (e.g. Linux)
638 if (*p->p_sysent->sv_transtrap)
639 i = (*p->p_sysent->sv_transtrap)(i, type);
641 MAKEMPSAFE(have_mplock);
642 trapsignal(lp, i, ucode);
645 if (type <= MAX_TRAP_MSG) {
646 uprintf("fatal process exception: %s",
648 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
649 uprintf(", fault VA = 0x%lx", (u_long)eva);
656 KASSERT(td->td_mpcount == have_mplock, ("badmpcount trap/end from %p", (void *)frame->tf_eip));
658 userret(lp, frame, sticks);
665 KTR_LOG(kernentry_trap_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
667 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
668 ("syscall: critical section count mismatch! %d/%d",
669 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
674 kern_trap(struct trapframe *frame)
676 struct globaldata *gd = mycpu;
677 struct thread *td = gd->gd_curthread;
680 int i = 0, ucode = 0, type, code;
685 int crit_count = td->td_pri & ~TDPRI_MASK;
692 if (frame->tf_trapno == T_PAGEFLT)
699 ++gd->gd_trap_nesting_level;
700 MAKEMPSAFE(have_mplock);
701 trap_fatal(frame, FALSE, eva);
702 --gd->gd_trap_nesting_level;
707 ++gd->gd_trap_nesting_level;
710 if (trap_mpsafe == 0)
711 MAKEMPSAFE(have_mplock);
714 --gd->gd_trap_nesting_level;
716 type = frame->tf_trapno;
717 code = frame->tf_err;
725 case T_PAGEFLT: /* page fault */
726 MAKEMPSAFE(have_mplock);
727 trap_pfault(frame, FALSE, eva);
733 * The kernel may be using npx for copying or other
736 panic("kernel NPX should not happen");
742 case T_PROTFLT: /* general protection fault */
743 case T_SEGNPFLT: /* segment not present fault */
745 * Invalid segment selectors and out of bounds
746 * %eip's and %esp's can be set up in user mode.
747 * This causes a fault in kernel mode when the
748 * kernel tries to return to user mode. We want
749 * to get this fault so that we can fix the
750 * problem here and not have to check all the
751 * selectors and pointers when the user changes
754 if (mycpu->gd_intr_nesting_level == 0) {
755 if (td->td_pcb->pcb_onfault) {
757 (register_t)td->td_pcb->pcb_onfault;
765 * PSL_NT can be set in user mode and isn't cleared
766 * automatically when the kernel is entered. This
767 * causes a TSS fault when the kernel attempts to
768 * `iret' because the TSS link is uninitialized. We
769 * want to get this fault so that we can fix the
770 * problem here and not every time the kernel is
773 if (frame->tf_eflags & PSL_NT) {
774 frame->tf_eflags &= ~PSL_NT;
779 case T_TRCTRAP: /* trace trap */
781 if (frame->tf_eip == (int)IDTVEC(syscall)) {
783 * We've just entered system mode via the
784 * syscall lcall. Continue single stepping
785 * silently until the syscall handler has
790 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
792 * The syscall handler has now saved the
793 * flags. Stop single stepping it.
795 frame->tf_eflags &= ~PSL_T;
801 * Ignore debug register trace traps due to
802 * accesses in the user's address space, which
803 * can happen under several conditions such as
804 * if a user sets a watchpoint on a buffer and
805 * then passes that buffer to a system call.
806 * We still want to get TRCTRAPS for addresses
807 * in kernel space because that is useful when
808 * debugging the kernel.
810 if (user_dbreg_trap()) {
812 * Reset breakpoint bits because the
815 load_dr6(rdr6() & 0xfffffff0);
820 * Fall through (TRCTRAP kernel mode, kernel address)
824 * If DDB is enabled, let it handle the debugger trap.
825 * Otherwise, debugger traps "can't happen".
828 MAKEMPSAFE(have_mplock);
829 if (kdb_trap (type, 0, frame))
834 MAKEMPSAFE(have_mplock);
835 trap_fatal(frame, FALSE, eva);
838 MAKEMPSAFE(have_mplock);
839 trap_fatal(frame, FALSE, eva);
843 * Ignore this trap generated from a spurious SIGTRAP.
845 * single stepping in / syscalls leads to spurious / SIGTRAP
848 * Haiku (c) 2007 Simon 'corecode' Schubert
854 * Translate fault for emulators (e.g. Linux)
856 if (*p->p_sysent->sv_transtrap)
857 i = (*p->p_sysent->sv_transtrap)(i, type);
859 MAKEMPSAFE(have_mplock);
860 trapsignal(lp, i, ucode);
863 if (type <= MAX_TRAP_MSG) {
864 uprintf("fatal process exception: %s",
866 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
867 uprintf(", fault VA = 0x%lx", (u_long)eva);
879 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
880 ("syscall: critical section count mismatch! %d/%d",
881 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
886 trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
889 struct vmspace *vm = NULL;
893 thread_t td = curthread;
894 struct lwp *lp = td->td_lwp;
896 va = trunc_page(eva);
897 if (usermode == FALSE) {
899 * This is a fault on kernel virtual memory.
904 * This is a fault on non-kernel virtual memory.
905 * vm is initialized above to NULL. If curproc is NULL
906 * or curproc->p_vmspace is NULL the fault is fatal.
909 vm = lp->lwp_vmspace;
917 if (frame->tf_xflags & PGEX_W)
918 ftype = VM_PROT_READ | VM_PROT_WRITE;
920 ftype = VM_PROT_READ;
922 if (map != &kernel_map) {
924 * Keep swapout from messing with us during this
930 * Grow the stack if necessary
932 /* grow_stack returns false only if va falls into
933 * a growable stack region and the stack growth
934 * fails. It returns true if va was not within
935 * a growable stack region, or if the stack
938 if (!grow_stack (lp->lwp_proc, va)) {
944 /* Fault in the user page: */
945 rv = vm_fault(map, va, ftype,
946 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
952 * Don't have to worry about process locking or stacks in the kernel.
954 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
957 if (rv == KERN_SUCCESS)
961 if (td->td_gd->gd_intr_nesting_level == 0 &&
962 td->td_pcb->pcb_onfault) {
963 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
966 trap_fatal(frame, usermode, eva);
969 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
973 trap_fatal(struct trapframe *frame, int usermode, vm_offset_t eva)
975 int code, type, ss, esp;
977 code = frame->tf_xflags;
978 type = frame->tf_trapno;
980 if (type <= MAX_TRAP_MSG) {
981 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
982 type, trap_msg[type],
983 (usermode ? "user" : "kernel"));
986 /* two separate prints in case of a trap on an unmapped page */
987 kprintf("mp_lock = %08x; ", mp_lock);
988 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
990 if (type == T_PAGEFLT) {
991 kprintf("fault virtual address = 0x%x\n", eva);
992 kprintf("fault code = %s %s, %s\n",
993 usermode ? "user" : "supervisor",
994 code & PGEX_W ? "write" : "read",
995 code & PGEX_P ? "protection violation" : "page not present");
997 kprintf("instruction pointer = 0x%x:0x%x\n",
998 frame->tf_cs & 0xffff, frame->tf_eip);
1000 ss = frame->tf_ss & 0xffff;
1001 esp = frame->tf_esp;
1003 ss = GSEL(GDATA_SEL, SEL_KPL);
1004 esp = (int)&frame->tf_esp;
1006 kprintf("stack pointer = 0x%x:0x%x\n", ss, esp);
1007 kprintf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
1008 kprintf("processor eflags = ");
1009 if (frame->tf_eflags & PSL_T)
1010 kprintf("trace trap, ");
1011 if (frame->tf_eflags & PSL_I)
1012 kprintf("interrupt enabled, ");
1013 if (frame->tf_eflags & PSL_NT)
1014 kprintf("nested task, ");
1015 if (frame->tf_eflags & PSL_RF)
1016 kprintf("resume, ");
1018 if (frame->tf_eflags & PSL_VM)
1021 kprintf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
1022 kprintf("current process = ");
1024 kprintf("%lu (%s)\n",
1025 (u_long)curproc->p_pid, curproc->p_comm ?
1026 curproc->p_comm : "");
1030 kprintf("current thread = pri %d ", curthread->td_pri);
1031 if (curthread->td_pri >= TDPRI_CRIT)
1037 * we probably SHOULD have stopped the other CPUs before now!
1038 * another CPU COULD have been touching cpl at this moment...
1040 kprintf(" <- SMP: XXX");
1049 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1052 kprintf("trap number = %d\n", type);
1053 if (type <= MAX_TRAP_MSG)
1054 panic("%s", trap_msg[type]);
1056 panic("unknown/reserved trap");
1060 * Double fault handler. Called when a fault occurs while writing
1061 * a frame for a trap/exception onto the stack. This usually occurs
1062 * when the stack overflows (such is the case with infinite recursion,
1065 * XXX Note that the current PTD gets replaced by IdlePTD when the
1066 * task switch occurs. This means that the stack that was active at
1067 * the time of the double fault is not available at <kstack> unless
1068 * the machine was idle when the double fault occurred. The downside
1069 * of this is that "trace <ebp>" in ddb won't work.
1072 dblfault_handler(void)
1074 struct mdglobaldata *gd = mdcpu;
1076 kprintf("\nFatal double fault:\n");
1077 kprintf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1078 kprintf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1079 kprintf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
1081 /* two separate prints in case of a trap on an unmapped page */
1082 kprintf("mp_lock = %08x; ", mp_lock);
1083 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
1085 panic("double fault");
1089 * Compensate for 386 brain damage (missing URKR).
1090 * This is a little simpler than the pagefault handler in trap() because
1091 * it the page tables have already been faulted in and high addresses
1092 * are thrown out early for other reasons.
1095 trapwrite(unsigned addr)
1102 va = trunc_page((vm_offset_t)addr);
1104 * XXX - MAX is END. Changed > to >= for temp. fix.
1106 if (va >= VM_MAX_USER_ADDRESS)
1109 lp = curthread->td_lwp;
1110 vm = lp->lwp_vmspace;
1112 PHOLD(lp->lwp_proc);
1114 if (!grow_stack (lp->lwp_proc, va)) {
1115 PRELE(lp->lwp_proc);
1120 * fault the data page
1122 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY);
1124 PRELE(lp->lwp_proc);
1126 if (rv != KERN_SUCCESS)
1133 * syscall2 - MP aware system call request C handler
1135 * A system call is essentially treated as a trap except that the
1136 * MP lock is not held on entry or return. We are responsible for
1137 * obtaining the MP lock if necessary and for handling ASTs
1138 * (e.g. a task switch) prior to return.
1140 * In general, only simple access and manipulation of curproc and
1141 * the current stack is allowed without having to hold MP lock.
1143 * MPSAFE - note that large sections of this routine are run without
1148 syscall2(struct trapframe *frame)
1150 struct thread *td = curthread;
1151 struct proc *p = td->td_proc;
1152 struct lwp *lp = td->td_lwp;
1154 struct sysent *callp;
1155 register_t orig_tf_eflags;
1160 int crit_count = td->td_pri & ~TDPRI_MASK;
1163 int have_mplock = 0;
1166 union sysunion args;
1168 KTR_LOG(kernentry_syscall, lp->lwp_proc->p_pid, lp->lwp_tid,
1172 KASSERT(td->td_mpcount == 0, ("badmpcount syscall2 from %p", (void *)frame->tf_eip));
1173 if (syscall_mpsafe == 0)
1174 MAKEMPSAFE(have_mplock);
1176 userenter(td); /* lazy raise our priority */
1181 sticks = (int)td->td_sticks;
1182 orig_tf_eflags = frame->tf_eflags;
1185 * Virtual kernel intercept - if a VM context managed by a virtual
1186 * kernel issues a system call the virtual kernel handles it, not us.
1187 * Restore the virtual kernel context and return from its system
1188 * call. The current frame is copied out to the virtual kernel.
1190 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1191 error = vkernel_trap(lp, frame);
1192 frame->tf_eax = error;
1194 frame->tf_eflags |= PSL_C;
1195 error = EJUSTRETURN;
1200 * Get the system call parameters and account for time
1202 lp->lwp_md.md_regs = frame;
1203 params = (caddr_t)frame->tf_esp + sizeof(int);
1204 code = frame->tf_eax;
1206 if (p->p_sysent->sv_prepsyscall) {
1207 (*p->p_sysent->sv_prepsyscall)(
1208 frame, (int *)(&args.nosys.sysmsg + 1),
1212 * Need to check if this is a 32 bit or 64 bit syscall.
1213 * fuword is MP aware.
1215 if (code == SYS_syscall) {
1217 * Code is first argument, followed by actual args.
1219 code = fuword(params);
1220 params += sizeof(int);
1221 } else if (code == SYS___syscall) {
1223 * Like syscall, but code is a quad, so as to maintain
1224 * quad alignment for the rest of the arguments.
1226 code = fuword(params);
1227 params += sizeof(quad_t);
1231 code &= p->p_sysent->sv_mask;
1232 if (code >= p->p_sysent->sv_size)
1233 callp = &p->p_sysent->sv_table[0];
1235 callp = &p->p_sysent->sv_table[code];
1237 narg = callp->sy_narg & SYF_ARGMASK;
1240 * copyin is MP aware, but the tracing code is not
1242 if (narg && params) {
1243 error = copyin(params, (caddr_t)(&args.nosys.sysmsg + 1),
1244 narg * sizeof(register_t));
1247 if (KTRPOINT(td, KTR_SYSCALL)) {
1248 MAKEMPSAFE(have_mplock);
1250 ktrsyscall(lp, code, narg,
1251 (void *)(&args.nosys.sysmsg + 1));
1259 if (KTRPOINT(td, KTR_SYSCALL)) {
1260 MAKEMPSAFE(have_mplock);
1261 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1266 * For traditional syscall code edx is left untouched when 32 bit
1267 * results are returned. Since edx is loaded from fds[1] when the
1268 * system call returns we pre-set it here.
1270 args.sysmsg_fds[0] = 0;
1271 args.sysmsg_fds[1] = frame->tf_edx;
1274 * The syscall might manipulate the trap frame. If it does it
1275 * will probably return EJUSTRETURN.
1277 args.sysmsg_frame = frame;
1279 STOPEVENT(p, S_SCE, narg); /* MP aware */
1283 * Try to run the syscall without the MP lock if the syscall
1284 * is MP safe. We have to obtain the MP lock no matter what if
1287 if ((callp->sy_narg & SYF_MPSAFE) == 0)
1288 MAKEMPSAFE(have_mplock);
1291 error = (*callp->sy_call)(&args);
1294 kprintf("system call %d returned %d\n", code, error);
1299 * MP SAFE (we may or may not have the MP lock at this point)
1304 * Reinitialize proc pointer `p' as it may be different
1305 * if this is a child returning from fork syscall.
1308 lp = curthread->td_lwp;
1309 frame->tf_eax = args.sysmsg_fds[0];
1310 frame->tf_edx = args.sysmsg_fds[1];
1311 frame->tf_eflags &= ~PSL_C;
1315 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1316 * int 0x80 is 2 bytes. We saved this in tf_err.
1318 frame->tf_eip -= frame->tf_err;
1323 panic("Unexpected EASYNC return value (for now)");
1326 if (p->p_sysent->sv_errsize) {
1327 if (error >= p->p_sysent->sv_errsize)
1328 error = -1; /* XXX */
1330 error = p->p_sysent->sv_errtbl[error];
1332 frame->tf_eax = error;
1333 frame->tf_eflags |= PSL_C;
1338 * Traced syscall. trapsignal() is not MP aware.
1340 if ((orig_tf_eflags & PSL_T) /*&& !(orig_tf_eflags & PSL_VM)*/) {
1341 MAKEMPSAFE(have_mplock);
1342 frame->tf_eflags &= ~PSL_T;
1343 trapsignal(lp, SIGTRAP, 0);
1347 * Handle reschedule and other end-of-syscall issues
1349 userret(lp, frame, sticks);
1352 if (KTRPOINT(td, KTR_SYSRET)) {
1353 MAKEMPSAFE(have_mplock);
1354 ktrsysret(lp, code, error, args.sysmsg_result);
1359 * This works because errno is findable through the
1360 * register set. If we ever support an emulation where this
1361 * is not the case, this code will need to be revisited.
1363 STOPEVENT(p, S_SCX, code);
1368 * Release the MP lock if we had to get it
1370 KASSERT(td->td_mpcount == have_mplock,
1371 ("badmpcount syscall2/end from %p", (void *)frame->tf_eip));
1375 KTR_LOG(kernentry_syscall_ret, lp->lwp_proc->p_pid, lp->lwp_tid, error);
1377 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
1378 ("syscall: critical section count mismatch! %d/%d",
1379 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
1384 fork_return(struct lwp *lp, struct trapframe *frame)
1386 frame->tf_eax = 0; /* Child returns zero */
1387 frame->tf_eflags &= ~PSL_C; /* success */
1390 generic_lwp_return(lp, frame);
1391 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1395 * Simplified back end of syscall(), used when returning from fork()
1396 * or lwp_create() directly into user mode. MP lock is held on entry and
1397 * should be released on return. This code will return back into the fork
1398 * trampoline code which then runs doreti.
1401 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1403 struct proc *p = lp->lwp_proc;
1406 * Newly forked processes are given a kernel priority. We have to
1407 * adjust the priority to a normal user priority and fake entry
1408 * into the kernel (call userenter()) to install a passive release
1409 * function just in case userret() decides to stop the process. This
1410 * can occur when ^Z races a fork. If we do not install the passive
1411 * release function the current process designation will not be
1412 * released when the thread goes to sleep.
1414 lwkt_setpri_self(TDPRI_USER_NORM);
1415 userenter(lp->lwp_thread);
1416 userret(lp, frame, 0);
1418 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1419 ktrsysret(lp, SYS_fork, 0, 0);
1421 p->p_flag |= P_PASSIVE_ACQ;
1423 p->p_flag &= ~P_PASSIVE_ACQ;
1425 KKASSERT(lp->lwp_thread->td_mpcount == 1);
1431 * doreti has turned into this. The frame is directly on the stack. We
1432 * pull everything else we need (fpu and tls context) from the current
1435 * Note on fpu interactions: In a virtual kernel, the fpu context for
1436 * an emulated user mode process is not shared with the virtual kernel's
1437 * fpu context, so we only have to 'stack' fpu contexts within the virtual
1438 * kernel itself, and not even then since the signal() contexts that we care
1439 * about save and restore the FPU state (I think anyhow).
1441 * vmspace_ctl() returns an error only if it had problems instaling the
1442 * context we supplied or problems copying data to/from our VM space.
1445 go_user(struct intrframe *frame)
1447 struct trapframe *tf = (void *)&frame->if_gs;
1451 * Interrupts may be disabled on entry, make sure all signals
1452 * can be received before beginning our loop.
1457 * Switch to the current simulated user process, then call
1458 * user_trap() when we break out of it (usually due to a signal).
1462 * Tell the real kernel whether it is ok to use the FP
1465 if (mdcpu->gd_npxthread == curthread) {
1466 tf->tf_xflags &= ~PGEX_FPFAULT;
1468 tf->tf_xflags |= PGEX_FPFAULT;
1472 * Run emulated user process context. This call interlocks
1473 * with new mailbox signals.
1475 * Set PGEX_U unconditionally, indicating a user frame (the
1476 * bit is normally set only by T_PAGEFLT).
1478 r = vmspace_ctl(&curproc->p_vmspace->vm_pmap, VMSPACE_CTL_RUN,
1479 tf, &curthread->td_savevext);
1480 frame->if_xflags |= PGEX_U;
1482 kprintf("GO USER %d trap %d EVA %08x EIP %08x ESP %08x XFLAGS %02x/%02x\n",
1483 r, tf->tf_trapno, tf->tf_err, tf->tf_eip, tf->tf_esp,
1484 tf->tf_xflags, frame->if_xflags);
1488 panic("vmspace_ctl failed");
1490 if (tf->tf_trapno) {
1492 } else if (mycpu->gd_reqflags & RQF_AST_MASK) {
1493 tf->tf_trapno = T_ASTFLT;
1502 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1503 * fault (which is then passed back to the virtual kernel) if an attempt is
1504 * made to use the FP unit.
1506 * XXX this is a fairly big hack.
1509 set_vkernel_fp(struct trapframe *frame)
1511 struct thread *td = curthread;
1513 if (frame->tf_xflags & PGEX_FPFAULT) {
1514 td->td_pcb->pcb_flags |= FP_VIRTFP;
1515 if (mdcpu->gd_npxthread == td)
1518 td->td_pcb->pcb_flags &= ~FP_VIRTFP;