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 1.108 2007/09/07 19:31:30 dillon Exp $
43 * 386 Trap and System call handling
51 #include "opt_ktrace.h"
52 #include "opt_clock.h"
55 #include <sys/param.h>
56 #include <sys/systm.h>
58 #include <sys/pioctl.h>
59 #include <sys/kernel.h>
60 #include <sys/resourcevar.h>
61 #include <sys/signalvar.h>
62 #include <sys/signal2.h>
63 #include <sys/syscall.h>
64 #include <sys/sysctl.h>
65 #include <sys/sysent.h>
67 #include <sys/vmmeter.h>
68 #include <sys/malloc.h>
70 #include <sys/ktrace.h>
72 #include <sys/upcall.h>
73 #include <sys/vkernel.h>
74 #include <sys/sysproto.h>
75 #include <sys/sysunion.h>
78 #include <vm/vm_param.h>
81 #include <vm/vm_kern.h>
82 #include <vm/vm_map.h>
83 #include <vm/vm_page.h>
84 #include <vm/vm_extern.h>
86 #include <machine/cpu.h>
87 #include <machine/md_var.h>
88 #include <machine/pcb.h>
89 #include <machine/smp.h>
90 #include <machine/tss.h>
91 #include <machine/specialreg.h>
92 #include <machine/globaldata.h>
94 #include <machine_base/isa/intr_machdep.h>
97 #include <sys/syslog.h>
98 #include <machine/clock.h>
101 #include <machine/vm86.h>
104 #include <sys/msgport2.h>
105 #include <sys/thread2.h>
109 #define MAKEMPSAFE(have_mplock) \
110 if (have_mplock == 0) { \
117 #define MAKEMPSAFE(have_mplock)
121 int (*pmath_emulate) (struct trapframe *);
123 extern void trap (struct trapframe *frame);
124 extern int trapwrite (unsigned addr);
125 extern void syscall2 (struct trapframe *frame);
127 static int trap_pfault (struct trapframe *, int, vm_offset_t);
128 static void trap_fatal (struct trapframe *, vm_offset_t);
129 void dblfault_handler (void);
131 extern inthand_t IDTVEC(syscall);
133 #define MAX_TRAP_MSG 28
134 static char *trap_msg[] = {
136 "privileged instruction fault", /* 1 T_PRIVINFLT */
138 "breakpoint instruction fault", /* 3 T_BPTFLT */
141 "arithmetic trap", /* 6 T_ARITHTRAP */
142 "system forced exception", /* 7 T_ASTFLT */
144 "general protection fault", /* 9 T_PROTFLT */
145 "trace trap", /* 10 T_TRCTRAP */
147 "page fault", /* 12 T_PAGEFLT */
149 "alignment fault", /* 14 T_ALIGNFLT */
153 "integer divide fault", /* 18 T_DIVIDE */
154 "non-maskable interrupt trap", /* 19 T_NMI */
155 "overflow trap", /* 20 T_OFLOW */
156 "FPU bounds check fault", /* 21 T_BOUND */
157 "FPU device not available", /* 22 T_DNA */
158 "double fault", /* 23 T_DOUBLEFLT */
159 "FPU operand fetch fault", /* 24 T_FPOPFLT */
160 "invalid TSS fault", /* 25 T_TSSFLT */
161 "segment not present fault", /* 26 T_SEGNPFLT */
162 "stack fault", /* 27 T_STKFLT */
163 "machine check trap", /* 28 T_MCHK */
166 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
167 extern int has_f00f_bug;
171 static int ddb_on_nmi = 1;
172 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
173 &ddb_on_nmi, 0, "Go to DDB on NMI");
175 static int panic_on_nmi = 1;
176 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
177 &panic_on_nmi, 0, "Panic on NMI");
178 static int fast_release;
179 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
180 &fast_release, 0, "Passive Release was optimal");
181 static int slow_release;
182 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
183 &slow_release, 0, "Passive Release was nonoptimal");
185 static int syscall_mpsafe = 1;
186 SYSCTL_INT(_kern, OID_AUTO, syscall_mpsafe, CTLFLAG_RW,
187 &syscall_mpsafe, 0, "Allow MPSAFE marked syscalls to run without BGL");
188 TUNABLE_INT("kern.syscall_mpsafe", &syscall_mpsafe);
189 static int trap_mpsafe = 1;
190 SYSCTL_INT(_kern, OID_AUTO, trap_mpsafe, CTLFLAG_RW,
191 &trap_mpsafe, 0, "Allow traps to mostly run without the BGL");
192 TUNABLE_INT("kern.trap_mpsafe", &trap_mpsafe);
195 MALLOC_DEFINE(M_SYSMSG, "sysmsg", "sysmsg structure");
196 extern int max_sysmsg;
199 * Passive USER->KERNEL transition. This only occurs if we block in the
200 * kernel while still holding our userland priority. We have to fixup our
201 * priority in order to avoid potential deadlocks before we allow the system
202 * to switch us to another thread.
205 passive_release(struct thread *td)
207 struct lwp *lp = td->td_lwp;
209 td->td_release = NULL;
210 lwkt_setpri_self(TDPRI_KERN_USER);
211 lp->lwp_proc->p_usched->release_curproc(lp);
215 * userenter() passively intercepts the thread switch function to increase
216 * the thread priority from a user priority to a kernel priority, reducing
217 * syscall and trap overhead for the case where no switch occurs.
221 userenter(struct thread *curtd)
223 curtd->td_release = passive_release;
227 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
228 * must be completed before we can return to or try to return to userland.
230 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
231 * arithmatic on the delta calculation so the absolute tick values are
232 * truncated to an integer.
235 userret(struct lwp *lp, struct trapframe *frame, int sticks)
237 struct proc *p = lp->lwp_proc;
241 * Charge system time if profiling. Note: times are in microseconds.
242 * This may do a copyout and block, so do it first even though it
243 * means some system time will be charged as user time.
245 if (p->p_flag & P_PROFIL) {
246 addupc_task(p, frame->tf_eip,
247 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
252 * If the jungle wants us dead, so be it.
254 if (lp->lwp_flag & LWP_WEXIT)
258 * Block here if we are in a stopped state.
260 if (p->p_stat == SSTOP) {
268 * Post any pending upcalls. If running a virtual kernel be sure
269 * to restore the virtual kernel's vmspace before posting the upcall.
271 if (p->p_flag & P_UPCALLPEND) {
272 p->p_flag &= ~P_UPCALLPEND;
280 * Post any pending signals. If running a virtual kernel be sure
281 * to restore the virtual kernel's vmspace before posting the signal.
283 if ((sig = CURSIG(lp)) != 0) {
291 * block here if we are swapped out, but still process signals
292 * (such as SIGKILL). proc0 (the swapin scheduler) is already
293 * aware of our situation, we do not have to wake it up.
295 if (p->p_flag & P_SWAPPEDOUT) {
297 p->p_flag |= P_SWAPWAIT;
299 if (p->p_flag & P_SWAPWAIT)
300 tsleep(p, PCATCH, "SWOUT", 0);
301 p->p_flag &= ~P_SWAPWAIT;
308 * Cleanup from userenter and any passive release that might have occured.
309 * We must reclaim the current-process designation before we can return
310 * to usermode. We also handle both LWKT and USER reschedule requests.
313 userexit(struct lwp *lp)
315 struct thread *td = lp->lwp_thread;
316 globaldata_t gd = td->td_gd;
320 * If a user reschedule is requested force a new process to be
321 * chosen by releasing the current process. Our process will only
322 * be chosen again if it has a considerably better priority.
324 if (user_resched_wanted())
325 lp->lwp_proc->p_usched->release_curproc(lp);
329 * Handle a LWKT reschedule request first. Since our passive release
330 * is still in place we do not have to do anything special.
332 if (lwkt_resched_wanted())
336 * Acquire the current process designation for this user scheduler
337 * on this cpu. This will also handle any user-reschedule requests.
339 lp->lwp_proc->p_usched->acquire_curproc(lp);
340 /* We may have switched cpus on acquisition */
344 * Reduce our priority in preparation for a return to userland. If
345 * our passive release function was still in place, our priority was
346 * never raised and does not need to be reduced.
348 if (td->td_release == NULL)
349 lwkt_setpri_self(TDPRI_USER_NORM);
350 td->td_release = NULL;
353 * After reducing our priority there might be other kernel-level
354 * LWKTs that now have a greater priority. Run them as necessary.
355 * We don't have to worry about losing cpu to userland because
356 * we still control the current-process designation and we no longer
357 * have a passive release function installed.
359 if (lwkt_checkpri_self())
364 * Exception, fault, and trap interface to the kernel.
365 * This common code is called from assembly language IDT gate entry
366 * routines that prepare a suitable stack frame, and restore this
367 * frame after the exception has been processed.
369 * This function is also called from doreti in an interlock to handle ASTs.
370 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
372 * NOTE! We have to retrieve the fault address prior to obtaining the
373 * MP lock because get_mplock() may switch out. YYY cr2 really ought
374 * to be retrieved by the assembly code, not here.
376 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
377 * if an attempt is made to switch from a fast interrupt or IPI. This is
378 * necessary to properly take fatal kernel traps on SMP machines if
379 * get_mplock() has to block.
383 trap(struct trapframe *frame)
385 struct globaldata *gd = mycpu;
386 struct thread *td = gd->gd_curthread;
387 struct lwp *lp = td->td_lwp;
390 int i = 0, ucode = 0, type, code;
395 int crit_count = td->td_pri & ~TDPRI_MASK;
402 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
403 ++gd->gd_trap_nesting_level;
404 MAKEMPSAFE(have_mplock);
405 trap_fatal(frame, eva);
406 --gd->gd_trap_nesting_level;
412 ++gd->gd_trap_nesting_level;
413 if (frame->tf_trapno == T_PAGEFLT) {
415 * For some Cyrix CPUs, %cr2 is clobbered by interrupts.
416 * This problem is worked around by using an interrupt
417 * gate for the pagefault handler. We are finally ready
418 * to read %cr2 and then must reenable interrupts.
420 * XXX this should be in the switch statement, but the
421 * NO_FOOF_HACK and VM86 goto and ifdefs obfuscate the
422 * flow of control too much for this to be obviously
429 if (trap_mpsafe == 0)
430 MAKEMPSAFE(have_mplock);
433 --gd->gd_trap_nesting_level;
435 if (!(frame->tf_eflags & PSL_I)) {
437 * Buggy application or kernel code has disabled interrupts
438 * and then trapped. Enabling interrupts now is wrong, but
439 * it is better than running with interrupts disabled until
440 * they are accidentally enabled later.
442 type = frame->tf_trapno;
443 if (ISPL(frame->tf_cs)==SEL_UPL || (frame->tf_eflags & PSL_VM)) {
444 MAKEMPSAFE(have_mplock);
446 "pid %ld (%s): trap %d with interrupts disabled\n",
447 (long)curproc->p_pid, curproc->p_comm, type);
448 } else if (type != T_BPTFLT && type != T_TRCTRAP) {
450 * XXX not quite right, since this may be for a
451 * multiple fault in user mode.
453 MAKEMPSAFE(have_mplock);
454 kprintf("kernel trap %d with interrupts disabled\n",
460 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
463 type = frame->tf_trapno;
464 code = frame->tf_err;
467 if (frame->tf_eflags & PSL_VM &&
468 (type == T_PROTFLT || type == T_STKFLT)) {
470 KKASSERT(td->td_mpcount > 0);
472 i = vm86_emulate((struct vm86frame *)frame);
474 KKASSERT(td->td_mpcount > 0);
478 * returns to original process
481 vm86_trap((struct vm86frame *)frame,
484 vm86_trap((struct vm86frame *)frame, 0);
486 KKASSERT(0); /* NOT REACHED */
492 * these traps want either a process context, or
493 * assume a normal userspace trap.
497 trap_fatal(frame, eva);
500 type = T_BPTFLT; /* kernel breakpoint */
503 goto kernel_trap; /* normal kernel trap handling */
506 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
511 sticks = (int)td->td_sticks;
512 lp->lwp_md.md_regs = frame;
515 case T_PRIVINFLT: /* privileged instruction fault */
520 case T_BPTFLT: /* bpt instruction fault */
521 case T_TRCTRAP: /* trace trap */
522 frame->tf_eflags &= ~PSL_T;
526 case T_ARITHTRAP: /* arithmetic trap */
531 case T_ASTFLT: /* Allow process switch */
532 mycpu->gd_cnt.v_soft++;
533 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
534 atomic_clear_int_nonlocked(&mycpu->gd_reqflags,
536 addupc_task(p, p->p_prof.pr_addr,
542 * The following two traps can happen in
543 * vm86 mode, and, if so, we want to handle
546 case T_PROTFLT: /* general protection fault */
547 case T_STKFLT: /* stack fault */
548 if (frame->tf_eflags & PSL_VM) {
549 i = vm86_emulate((struct vm86frame *)frame);
556 case T_SEGNPFLT: /* segment not present fault */
557 case T_TSSFLT: /* invalid TSS fault */
558 case T_DOUBLEFLT: /* double fault */
560 ucode = code + BUS_SEGM_FAULT ;
564 case T_PAGEFLT: /* page fault */
565 MAKEMPSAFE(have_mplock);
566 i = trap_pfault(frame, TRUE, eva);
569 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
579 case T_DIVIDE: /* integer divide fault */
586 MAKEMPSAFE(have_mplock);
588 goto handle_powerfail;
589 #else /* !POWERFAIL_NMI */
590 /* machine/parity/power fail/"kitchen sink" faults */
591 if (isa_nmi(code) == 0) {
594 * NMI can be hooked up to a pushbutton
598 kprintf ("NMI ... going to debugger\n");
599 kdb_trap (type, 0, frame);
603 } else if (panic_on_nmi)
604 panic("NMI indicates hardware failure");
606 #endif /* POWERFAIL_NMI */
607 #endif /* NISA > 0 */
609 case T_OFLOW: /* integer overflow fault */
614 case T_BOUND: /* bounds check fault */
621 * Virtual kernel intercept - pass the DNA exception
622 * to the virtual kernel if it asked to handle it.
623 * This occurs when the virtual kernel is holding
624 * onto the FP context for a different emulated
625 * process then the one currently running.
627 * We must still call npxdna() since we may have
628 * saved FP state that the virtual kernel needs
629 * to hand over to a different emulated process.
631 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
632 (td->td_pcb->pcb_flags & FP_VIRTFP)
640 * The kernel may have switched out the FP unit's
641 * state, causing the user process to take a fault
642 * when it tries to use the FP unit. Restore the
648 if (!pmath_emulate) {
650 ucode = FPE_FPU_NP_TRAP;
653 i = (*pmath_emulate)(frame);
655 if (!(frame->tf_eflags & PSL_T))
657 frame->tf_eflags &= ~PSL_T;
660 /* else ucode = emulator_only_knows() XXX */
663 case T_FPOPFLT: /* FPU operand fetch fault */
668 case T_XMMFLT: /* SIMD floating-point exception */
678 case T_PAGEFLT: /* page fault */
679 MAKEMPSAFE(have_mplock);
680 trap_pfault(frame, FALSE, eva);
686 * The kernel may be using npx for copying or other
694 case T_PROTFLT: /* general protection fault */
695 case T_SEGNPFLT: /* segment not present fault */
697 * Invalid segment selectors and out of bounds
698 * %eip's and %esp's can be set up in user mode.
699 * This causes a fault in kernel mode when the
700 * kernel tries to return to user mode. We want
701 * to get this fault so that we can fix the
702 * problem here and not have to check all the
703 * selectors and pointers when the user changes
706 #define MAYBE_DORETI_FAULT(where, whereto) \
708 if (frame->tf_eip == (int)where) { \
709 frame->tf_eip = (int)whereto; \
713 if (mycpu->gd_intr_nesting_level == 0) {
715 * Invalid %fs's and %gs's can be created using
716 * procfs or PT_SETREGS or by invalidating the
717 * underlying LDT entry. This causes a fault
718 * in kernel mode when the kernel attempts to
719 * switch contexts. Lose the bad context
720 * (XXX) so that we can continue, and generate
723 MAYBE_DORETI_FAULT(doreti_iret,
725 MAYBE_DORETI_FAULT(doreti_popl_ds,
726 doreti_popl_ds_fault);
727 MAYBE_DORETI_FAULT(doreti_popl_es,
728 doreti_popl_es_fault);
729 MAYBE_DORETI_FAULT(doreti_popl_fs,
730 doreti_popl_fs_fault);
731 MAYBE_DORETI_FAULT(doreti_popl_gs,
732 doreti_popl_gs_fault);
733 if (td->td_pcb->pcb_onfault) {
735 (register_t)td->td_pcb->pcb_onfault;
743 * PSL_NT can be set in user mode and isn't cleared
744 * automatically when the kernel is entered. This
745 * causes a TSS fault when the kernel attempts to
746 * `iret' because the TSS link is uninitialized. We
747 * want to get this fault so that we can fix the
748 * problem here and not every time the kernel is
751 if (frame->tf_eflags & PSL_NT) {
752 frame->tf_eflags &= ~PSL_NT;
757 case T_TRCTRAP: /* trace trap */
758 if (frame->tf_eip == (int)IDTVEC(syscall)) {
760 * We've just entered system mode via the
761 * syscall lcall. Continue single stepping
762 * silently until the syscall handler has
767 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
769 * The syscall handler has now saved the
770 * flags. Stop single stepping it.
772 frame->tf_eflags &= ~PSL_T;
776 * Ignore debug register trace traps due to
777 * accesses in the user's address space, which
778 * can happen under several conditions such as
779 * if a user sets a watchpoint on a buffer and
780 * then passes that buffer to a system call.
781 * We still want to get TRCTRAPS for addresses
782 * in kernel space because that is useful when
783 * debugging the kernel.
785 if (user_dbreg_trap()) {
787 * Reset breakpoint bits because the
790 load_dr6(rdr6() & 0xfffffff0);
794 * Fall through (TRCTRAP kernel mode, kernel address)
798 * If DDB is enabled, let it handle the debugger trap.
799 * Otherwise, debugger traps "can't happen".
802 MAKEMPSAFE(have_mplock);
803 if (kdb_trap (type, 0, frame))
810 MAKEMPSAFE(have_mplock);
813 # define TIMER_FREQ 1193182
817 static unsigned lastalert = 0;
819 if(time_second - lastalert > 10)
821 log(LOG_WARNING, "NMI: power fail\n");
822 sysbeep(TIMER_FREQ/880, hz);
823 lastalert = time_second;
828 #else /* !POWERFAIL_NMI */
829 /* machine/parity/power fail/"kitchen sink" faults */
830 if (isa_nmi(code) == 0) {
833 * NMI can be hooked up to a pushbutton
837 kprintf ("NMI ... going to debugger\n");
838 kdb_trap (type, 0, frame);
842 } else if (panic_on_nmi == 0)
845 #endif /* POWERFAIL_NMI */
846 #endif /* NISA > 0 */
849 MAKEMPSAFE(have_mplock);
850 trap_fatal(frame, eva);
855 * Virtual kernel intercept - if the fault is directly related to a
856 * VM context managed by a virtual kernel then let the virtual kernel
859 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
860 vkernel_trap(lp, frame);
865 * Translate fault for emulators (e.g. Linux)
867 if (*p->p_sysent->sv_transtrap)
868 i = (*p->p_sysent->sv_transtrap)(i, type);
870 MAKEMPSAFE(have_mplock);
871 trapsignal(lp, i, ucode);
874 if (type <= MAX_TRAP_MSG) {
875 uprintf("fatal process exception: %s",
877 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
878 uprintf(", fault VA = 0x%lx", (u_long)eva);
885 if (ISPL(frame->tf_cs) == SEL_UPL)
886 KASSERT(td->td_mpcount == have_mplock, ("badmpcount trap/end from %p", (void *)frame->tf_eip));
888 userret(lp, frame, sticks);
896 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
897 ("syscall: critical section count mismatch! %d/%d",
898 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
903 trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
906 struct vmspace *vm = NULL;
910 thread_t td = curthread;
911 struct lwp *lp = td->td_lwp;
913 va = trunc_page(eva);
914 if (va >= KERNBASE) {
916 * Don't allow user-mode faults in kernel address space.
917 * An exception: if the faulting address is the invalid
918 * instruction entry in the IDT, then the Intel Pentium
919 * F00F bug workaround was triggered, and we need to
920 * treat it is as an illegal instruction, and not a page
923 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
924 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) {
925 frame->tf_trapno = T_PRIVINFLT;
935 * This is a fault on non-kernel virtual memory.
936 * vm is initialized above to NULL. If curproc is NULL
937 * or curproc->p_vmspace is NULL the fault is fatal.
940 vm = lp->lwp_vmspace;
948 if (frame->tf_err & PGEX_W)
949 ftype = VM_PROT_WRITE;
951 ftype = VM_PROT_READ;
953 if (map != &kernel_map) {
955 * Keep swapout from messing with us during this
961 * Grow the stack if necessary
963 /* grow_stack returns false only if va falls into
964 * a growable stack region and the stack growth
965 * fails. It returns true if va was not within
966 * a growable stack region, or if the stack
969 if (!grow_stack(lp->lwp_proc, va)) {
975 /* Fault in the user page: */
976 rv = vm_fault(map, va, ftype,
977 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
983 * Don't have to worry about process locking or stacks in the kernel.
985 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
988 if (rv == KERN_SUCCESS)
992 if (td->td_gd->gd_intr_nesting_level == 0 &&
993 td->td_pcb->pcb_onfault) {
994 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
997 trap_fatal(frame, eva);
1001 /* kludge to pass faulting virtual address to sendsig */
1002 frame->tf_xflags = frame->tf_err;
1003 frame->tf_err = eva;
1005 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
1009 trap_fatal(struct trapframe *frame, vm_offset_t eva)
1011 int code, type, ss, esp;
1012 struct soft_segment_descriptor softseg;
1014 code = frame->tf_err;
1015 type = frame->tf_trapno;
1016 sdtossd(&gdt[mycpu->gd_cpuid * NGDT + IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
1018 if (type <= MAX_TRAP_MSG)
1019 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
1020 type, trap_msg[type],
1021 frame->tf_eflags & PSL_VM ? "vm86" :
1022 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
1024 /* three separate prints in case of a trap on an unmapped page */
1025 kprintf("mp_lock = %08x; ", mp_lock);
1026 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
1027 kprintf("lapic.id = %08x\n", lapic.id);
1029 if (type == T_PAGEFLT) {
1030 kprintf("fault virtual address = 0x%x\n", eva);
1031 kprintf("fault code = %s %s, %s\n",
1032 code & PGEX_U ? "user" : "supervisor",
1033 code & PGEX_W ? "write" : "read",
1034 code & PGEX_P ? "protection violation" : "page not present");
1036 kprintf("instruction pointer = 0x%x:0x%x\n",
1037 frame->tf_cs & 0xffff, frame->tf_eip);
1038 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
1039 ss = frame->tf_ss & 0xffff;
1040 esp = frame->tf_esp;
1042 ss = GSEL(GDATA_SEL, SEL_KPL);
1043 esp = (int)&frame->tf_esp;
1045 kprintf("stack pointer = 0x%x:0x%x\n", ss, esp);
1046 kprintf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
1047 kprintf("code segment = base 0x%x, limit 0x%x, type 0x%x\n",
1048 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
1049 kprintf(" = DPL %d, pres %d, def32 %d, gran %d\n",
1050 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
1052 kprintf("processor eflags = ");
1053 if (frame->tf_eflags & PSL_T)
1054 kprintf("trace trap, ");
1055 if (frame->tf_eflags & PSL_I)
1056 kprintf("interrupt enabled, ");
1057 if (frame->tf_eflags & PSL_NT)
1058 kprintf("nested task, ");
1059 if (frame->tf_eflags & PSL_RF)
1060 kprintf("resume, ");
1061 if (frame->tf_eflags & PSL_VM)
1063 kprintf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
1064 kprintf("current process = ");
1066 kprintf("%lu (%s)\n",
1067 (u_long)curproc->p_pid, curproc->p_comm ?
1068 curproc->p_comm : "");
1072 kprintf("current thread = pri %d ", curthread->td_pri);
1073 if (curthread->td_pri >= TDPRI_CRIT)
1079 * we probably SHOULD have stopped the other CPUs before now!
1080 * another CPU COULD have been touching cpl at this moment...
1082 kprintf(" <- SMP: XXX");
1091 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1094 kprintf("trap number = %d\n", type);
1095 if (type <= MAX_TRAP_MSG)
1096 panic("%s", trap_msg[type]);
1098 panic("unknown/reserved trap");
1102 * Double fault handler. Called when a fault occurs while writing
1103 * a frame for a trap/exception onto the stack. This usually occurs
1104 * when the stack overflows (such is the case with infinite recursion,
1107 * XXX Note that the current PTD gets replaced by IdlePTD when the
1108 * task switch occurs. This means that the stack that was active at
1109 * the time of the double fault is not available at <kstack> unless
1110 * the machine was idle when the double fault occurred. The downside
1111 * of this is that "trace <ebp>" in ddb won't work.
1114 dblfault_handler(void)
1116 struct mdglobaldata *gd = mdcpu;
1118 kprintf("\nFatal double fault:\n");
1119 kprintf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1120 kprintf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1121 kprintf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
1123 /* three separate prints in case of a trap on an unmapped page */
1124 kprintf("mp_lock = %08x; ", mp_lock);
1125 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
1126 kprintf("lapic.id = %08x\n", lapic.id);
1128 panic("double fault");
1132 * Compensate for 386 brain damage (missing URKR).
1133 * This is a little simpler than the pagefault handler in trap() because
1134 * it the page tables have already been faulted in and high addresses
1135 * are thrown out early for other reasons.
1138 trapwrite(unsigned addr)
1145 va = trunc_page((vm_offset_t)addr);
1147 * XXX - MAX is END. Changed > to >= for temp. fix.
1149 if (va >= VM_MAX_USER_ADDRESS)
1152 lp = curthread->td_lwp;
1153 vm = lp->lwp_vmspace;
1155 PHOLD(lp->lwp_proc);
1157 if (!grow_stack(lp->lwp_proc, va)) {
1158 PRELE(lp->lwp_proc);
1163 * fault the data page
1165 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY);
1167 PRELE(lp->lwp_proc);
1169 if (rv != KERN_SUCCESS)
1176 * syscall2 - MP aware system call request C handler
1178 * A system call is essentially treated as a trap except that the
1179 * MP lock is not held on entry or return. We are responsible for
1180 * obtaining the MP lock if necessary and for handling ASTs
1181 * (e.g. a task switch) prior to return.
1183 * In general, only simple access and manipulation of curproc and
1184 * the current stack is allowed without having to hold MP lock.
1186 * MPSAFE - note that large sections of this routine are run without
1191 syscall2(struct trapframe *frame)
1193 struct thread *td = curthread;
1194 struct proc *p = td->td_proc;
1195 struct lwp *lp = td->td_lwp;
1197 struct sysent *callp;
1198 register_t orig_tf_eflags;
1203 int crit_count = td->td_pri & ~TDPRI_MASK;
1206 int have_mplock = 0;
1209 union sysunion args;
1212 if (ISPL(frame->tf_cs) != SEL_UPL) {
1220 KASSERT(td->td_mpcount == 0, ("badmpcount syscall2 from %p", (void *)frame->tf_eip));
1221 if (syscall_mpsafe == 0)
1222 MAKEMPSAFE(have_mplock);
1224 userenter(td); /* lazy raise our priority */
1229 sticks = (int)td->td_sticks;
1230 orig_tf_eflags = frame->tf_eflags;
1233 * Virtual kernel intercept - if a VM context managed by a virtual
1234 * kernel issues a system call the virtual kernel handles it, not us.
1235 * Restore the virtual kernel context and return from its system
1236 * call. The current frame is copied out to the virtual kernel.
1238 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1239 error = vkernel_trap(lp, frame);
1240 frame->tf_eax = error;
1242 frame->tf_eflags |= PSL_C;
1243 error = EJUSTRETURN;
1248 * Get the system call parameters and account for time
1250 lp->lwp_md.md_regs = frame;
1251 params = (caddr_t)frame->tf_esp + sizeof(int);
1252 code = frame->tf_eax;
1254 if (p->p_sysent->sv_prepsyscall) {
1255 (*p->p_sysent->sv_prepsyscall)(
1256 frame, (int *)(&args.nosys.sysmsg + 1),
1260 * Need to check if this is a 32 bit or 64 bit syscall.
1261 * fuword is MP aware.
1263 if (code == SYS_syscall) {
1265 * Code is first argument, followed by actual args.
1267 code = fuword(params);
1268 params += sizeof(int);
1269 } else if (code == SYS___syscall) {
1271 * Like syscall, but code is a quad, so as to maintain
1272 * quad alignment for the rest of the arguments.
1274 code = fuword(params);
1275 params += sizeof(quad_t);
1279 code &= p->p_sysent->sv_mask;
1280 if (code >= p->p_sysent->sv_size)
1281 callp = &p->p_sysent->sv_table[0];
1283 callp = &p->p_sysent->sv_table[code];
1285 narg = callp->sy_narg & SYF_ARGMASK;
1288 * copyin is MP aware, but the tracing code is not
1290 if (narg && params) {
1291 error = copyin(params, (caddr_t)(&args.nosys.sysmsg + 1),
1292 narg * sizeof(register_t));
1295 if (KTRPOINT(td, KTR_SYSCALL)) {
1296 MAKEMPSAFE(have_mplock);
1298 ktrsyscall(p, code, narg,
1299 (void *)(&args.nosys.sysmsg + 1));
1307 if (KTRPOINT(td, KTR_SYSCALL)) {
1308 MAKEMPSAFE(have_mplock);
1309 ktrsyscall(p, code, narg, (void *)(&args.nosys.sysmsg + 1));
1314 * For traditional syscall code edx is left untouched when 32 bit
1315 * results are returned. Since edx is loaded from fds[1] when the
1316 * system call returns we pre-set it here.
1318 args.sysmsg_fds[0] = 0;
1319 args.sysmsg_fds[1] = frame->tf_edx;
1322 * The syscall might manipulate the trap frame. If it does it
1323 * will probably return EJUSTRETURN.
1325 args.sysmsg_frame = frame;
1327 STOPEVENT(p, S_SCE, narg); /* MP aware */
1331 * Try to run the syscall without the MP lock if the syscall
1332 * is MP safe. We have to obtain the MP lock no matter what if
1335 if ((callp->sy_narg & SYF_MPSAFE) == 0)
1336 MAKEMPSAFE(have_mplock);
1339 error = (*callp->sy_call)(&args);
1343 * MP SAFE (we may or may not have the MP lock at this point)
1348 * Reinitialize proc pointer `p' as it may be different
1349 * if this is a child returning from fork syscall.
1352 lp = curthread->td_lwp;
1353 frame->tf_eax = args.sysmsg_fds[0];
1354 frame->tf_edx = args.sysmsg_fds[1];
1355 frame->tf_eflags &= ~PSL_C;
1359 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1360 * int 0x80 is 2 bytes. We saved this in tf_err.
1362 frame->tf_eip -= frame->tf_err;
1367 panic("Unexpected EASYNC return value (for now)");
1370 if (p->p_sysent->sv_errsize) {
1371 if (error >= p->p_sysent->sv_errsize)
1372 error = -1; /* XXX */
1374 error = p->p_sysent->sv_errtbl[error];
1376 frame->tf_eax = error;
1377 frame->tf_eflags |= PSL_C;
1382 * Traced syscall. trapsignal() is not MP aware.
1384 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1385 MAKEMPSAFE(have_mplock);
1386 frame->tf_eflags &= ~PSL_T;
1387 trapsignal(lp, SIGTRAP, 0);
1391 * Handle reschedule and other end-of-syscall issues
1393 userret(lp, frame, sticks);
1396 if (KTRPOINT(td, KTR_SYSRET)) {
1397 MAKEMPSAFE(have_mplock);
1398 ktrsysret(p, code, error, args.sysmsg_result);
1403 * This works because errno is findable through the
1404 * register set. If we ever support an emulation where this
1405 * is not the case, this code will need to be revisited.
1407 STOPEVENT(p, S_SCX, code);
1412 * Release the MP lock if we had to get it
1414 KASSERT(td->td_mpcount == have_mplock,
1415 ("badmpcount syscall2/end from %p", (void *)frame->tf_eip));
1420 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
1421 ("syscall: critical section count mismatch! %d/%d",
1422 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
1427 fork_return(struct lwp *lp, struct trapframe *frame)
1429 frame->tf_eax = 0; /* Child returns zero */
1430 frame->tf_eflags &= ~PSL_C; /* success */
1433 generic_lwp_return(lp, frame);
1437 * Simplified back end of syscall(), used when returning from fork()
1438 * directly into user mode. MP lock is held on entry and should be
1439 * released on return. This code will return back into the fork
1440 * trampoline code which then runs doreti.
1443 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1445 struct proc *p = lp->lwp_proc;
1448 * Newly forked processes are given a kernel priority. We have to
1449 * adjust the priority to a normal user priority and fake entry
1450 * into the kernel (call userenter()) to install a passive release
1451 * function just in case userret() decides to stop the process. This
1452 * can occur when ^Z races a fork. If we do not install the passive
1453 * release function the current process designation will not be
1454 * released when the thread goes to sleep.
1456 lwkt_setpri_self(TDPRI_USER_NORM);
1457 userenter(lp->lwp_thread);
1458 userret(lp, frame, 0);
1460 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1461 ktrsysret(p, SYS_fork, 0, 0);
1463 p->p_flag |= P_PASSIVE_ACQ;
1465 p->p_flag &= ~P_PASSIVE_ACQ;
1467 KKASSERT(lp->lwp_thread->td_mpcount == 1);
1473 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1474 * fault (which is then passed back to the virtual kernel) if an attempt is
1475 * made to use the FP unit.
1477 * XXX this is a fairly big hack.
1480 set_vkernel_fp(struct trapframe *frame)
1482 struct thread *td = curthread;
1484 if (frame->tf_xflags & PGEX_FPFAULT) {
1485 td->td_pcb->pcb_flags |= FP_VIRTFP;
1486 if (mdcpu->gd_npxthread == td)
1489 td->td_pcb->pcb_flags &= ~FP_VIRTFP;