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/i386/i386/Attic/trap.c,v 1.54 2004/07/24 20:21:33 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/syscall.h>
63 #include <sys/sysctl.h>
64 #include <sys/sysent.h>
66 #include <sys/vmmeter.h>
67 #include <sys/malloc.h>
69 #include <sys/ktrace.h>
71 #include <sys/upcall.h>
72 #include <sys/sysproto.h>
73 #include <sys/sysunion.h>
76 #include <vm/vm_param.h>
79 #include <vm/vm_kern.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_page.h>
82 #include <vm/vm_extern.h>
84 #include <machine/cpu.h>
85 #include <machine/ipl.h>
86 #include <machine/md_var.h>
87 #include <machine/pcb.h>
89 #include <machine/smp.h>
91 #include <machine/tss.h>
92 #include <machine/globaldata.h>
94 #include <i386/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>
107 int (*pmath_emulate) (struct trapframe *);
109 extern void trap (struct trapframe frame);
110 extern int trapwrite (unsigned addr);
111 extern void syscall2 (struct trapframe frame);
112 extern void sendsys2 (struct trapframe frame);
114 static int trap_pfault (struct trapframe *, int, vm_offset_t);
115 static void trap_fatal (struct trapframe *, vm_offset_t);
116 void dblfault_handler (void);
118 extern inthand_t IDTVEC(syscall);
120 #define MAX_TRAP_MSG 28
121 static char *trap_msg[] = {
123 "privileged instruction fault", /* 1 T_PRIVINFLT */
125 "breakpoint instruction fault", /* 3 T_BPTFLT */
128 "arithmetic trap", /* 6 T_ARITHTRAP */
129 "system forced exception", /* 7 T_ASTFLT */
131 "general protection fault", /* 9 T_PROTFLT */
132 "trace trap", /* 10 T_TRCTRAP */
134 "page fault", /* 12 T_PAGEFLT */
136 "alignment fault", /* 14 T_ALIGNFLT */
140 "integer divide fault", /* 18 T_DIVIDE */
141 "non-maskable interrupt trap", /* 19 T_NMI */
142 "overflow trap", /* 20 T_OFLOW */
143 "FPU bounds check fault", /* 21 T_BOUND */
144 "FPU device not available", /* 22 T_DNA */
145 "double fault", /* 23 T_DOUBLEFLT */
146 "FPU operand fetch fault", /* 24 T_FPOPFLT */
147 "invalid TSS fault", /* 25 T_TSSFLT */
148 "segment not present fault", /* 26 T_SEGNPFLT */
149 "stack fault", /* 27 T_STKFLT */
150 "machine check trap", /* 28 T_MCHK */
153 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
154 extern int has_f00f_bug;
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 MALLOC_DEFINE(M_SYSMSG, "sysmsg", "sysmsg structure");
175 * Passive USER->KERNEL transition. This only occurs if we block in the
176 * kernel while still holding our userland priority. We have to fixup our
177 * priority in order to avoid potential deadlocks before we allow the system
178 * to switch us to another thread.
181 passive_release(struct thread *td)
183 struct proc *p = td->td_proc;
185 td->td_release = NULL;
186 lwkt_setpri_self(TDPRI_KERN_USER);
191 * userenter() passively intercepts the thread switch function to increase
192 * the thread priority from a user priority to a kernel priority, reducing
193 * syscall and trap overhead for the case where no switch occurs.
197 userenter(struct thread *curtd)
199 curtd->td_release = passive_release;
203 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
204 * must be completed before we can return to or try to return to userland.
206 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
207 * arithmatic on the delta calculation so the absolute tick values are
208 * truncated to an integer.
211 userret(struct proc *p, struct trapframe *frame, int sticks)
216 * Post any pending upcalls
218 if (p->p_flag & P_UPCALLPEND) {
219 p->p_flag &= ~P_UPCALLPEND;
224 * Post any pending signals
226 while ((sig = CURSIG(p)) != 0) {
231 * Charge system time if profiling. Note: times are in microseconds.
233 if (p->p_flag & P_PROFIL) {
234 addupc_task(p, frame->tf_eip,
235 (u_int)((int)p->p_thread->td_sticks - sticks));
239 * Post any pending signals XXX
241 while ((sig = CURSIG(p)) != 0)
246 * Cleanup from userenter and any passive release that might have occured.
247 * We must reclaim the current-process designation before we can return
248 * to usermode. We also handle both LWKT and USER reschedule requests.
251 userexit(struct proc *p)
253 struct thread *td = p->p_thread;
254 globaldata_t gd = td->td_gd;
258 * If a user reschedule is requested force a new process to be
259 * chosen by releasing the current process. Our process will only
260 * be chosen again if it has a considerably better priority.
262 if (user_resched_wanted())
268 * Handle a LWKT reschedule request first. Since our passive release
269 * is still in place we do not have to do anything special.
271 if (lwkt_resched_wanted())
275 * Acquire the current process designation if we do not own it.
276 * Note that acquire_curproc() does not reset the user reschedule
277 * bit on purpose, because we may need to accumulate over several
278 * threads waking up at the same time.
280 * NOTE: userland scheduler cruft: because processes are removed
281 * from the userland scheduler's queue we run through loops to try
282 * to figure out which is the best of [ existing, waking-up ]
285 if (p != gd->gd_uschedcp) {
288 /* We may have switched cpus on acquisition */
295 * Reduce our priority in preparation for a return to userland. If
296 * our passive release function was still in place, our priority was
297 * never raised and does not need to be reduced.
299 if (td->td_release == NULL)
300 lwkt_setpri_self(TDPRI_USER_NORM);
301 td->td_release = NULL;
304 * After reducing our priority there might be other kernel-level
305 * LWKTs that now have a greater priority. Run them as necessary.
306 * We don't have to worry about losing cpu to userland because
307 * we still control the current-process designation and we no longer
308 * have a passive release function installed.
310 if (lwkt_checkpri_self())
314 * If a userland reschedule is [still] pending we may not be the best
315 * selected process. Select a better one. If another LWKT resched
316 * is pending the trap will be re-entered.
318 if (user_resched_wanted()) {
320 if (p != gd->gd_uschedcp) {
321 lwkt_setpri_self(TDPRI_KERN_USER);
327 #ifdef DEVICE_POLLING
328 extern u_int32_t poll_in_trap;
329 extern int ether_poll (int count);
330 #endif /* DEVICE_POLLING */
333 * Exception, fault, and trap interface to the kernel.
334 * This common code is called from assembly language IDT gate entry
335 * routines that prepare a suitable stack frame, and restore this
336 * frame after the exception has been processed.
338 * This function is also called from doreti in an interlock to handle ASTs.
339 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
341 * NOTE! We have to retrieve the fault address prior to obtaining the
342 * MP lock because get_mplock() may switch out. YYY cr2 really ought
343 * to be retrieved by the assembly code, not here.
347 struct trapframe frame;
349 struct thread *td = curthread;
352 int i = 0, ucode = 0, type, code;
358 eva = (frame.tf_trapno == T_PAGEFLT ? rcr2() : 0);
360 trap_fatal(&frame, eva);
366 if (frame.tf_trapno == T_PAGEFLT) {
368 * For some Cyrix CPUs, %cr2 is clobbered by interrupts.
369 * This problem is worked around by using an interrupt
370 * gate for the pagefault handler. We are finally ready
371 * to read %cr2 and then must reenable interrupts.
373 * XXX this should be in the switch statement, but the
374 * NO_FOOF_HACK and VM86 goto and ifdefs obfuscate the
375 * flow of control too much for this to be obviously
385 * MP lock is held at this point
388 if (!(frame.tf_eflags & PSL_I)) {
390 * Buggy application or kernel code has disabled interrupts
391 * and then trapped. Enabling interrupts now is wrong, but
392 * it is better than running with interrupts disabled until
393 * they are accidentally enabled later.
395 type = frame.tf_trapno;
396 if (ISPL(frame.tf_cs)==SEL_UPL || (frame.tf_eflags & PSL_VM)) {
398 "pid %ld (%s): trap %d with interrupts disabled\n",
399 (long)curproc->p_pid, curproc->p_comm, type);
400 } else if (type != T_BPTFLT && type != T_TRCTRAP) {
402 * XXX not quite right, since this may be for a
403 * multiple fault in user mode.
405 printf("kernel trap %d with interrupts disabled\n",
412 #ifdef DEVICE_POLLING
414 ether_poll(poll_in_trap);
415 #endif /* DEVICE_POLLING */
417 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
420 type = frame.tf_trapno;
424 if (frame.tf_eflags & PSL_VM &&
425 (type == T_PROTFLT || type == T_STKFLT)) {
427 KKASSERT(td->td_mpcount > 0);
429 i = vm86_emulate((struct vm86frame *)&frame);
431 KKASSERT(td->td_mpcount > 0);
435 * returns to original process
437 vm86_trap((struct vm86frame *)&frame);
444 * these traps want either a process context, or
445 * assume a normal userspace trap.
449 trap_fatal(&frame, eva);
452 type = T_BPTFLT; /* kernel breakpoint */
455 goto kernel_trap; /* normal kernel trap handling */
458 if ((ISPL(frame.tf_cs) == SEL_UPL) || (frame.tf_eflags & PSL_VM)) {
463 sticks = (int)td->td_sticks;
464 p->p_md.md_regs = &frame;
467 case T_PRIVINFLT: /* privileged instruction fault */
472 case T_BPTFLT: /* bpt instruction fault */
473 case T_TRCTRAP: /* trace trap */
474 frame.tf_eflags &= ~PSL_T;
478 case T_ARITHTRAP: /* arithmetic trap */
483 case T_ASTFLT: /* Allow process switch */
484 mycpu->gd_cnt.v_soft++;
485 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
486 atomic_clear_int_nonlocked(&mycpu->gd_reqflags,
488 addupc_task(p, p->p_stats->p_prof.pr_addr,
489 p->p_stats->p_prof.pr_ticks);
494 * The following two traps can happen in
495 * vm86 mode, and, if so, we want to handle
498 case T_PROTFLT: /* general protection fault */
499 case T_STKFLT: /* stack fault */
500 if (frame.tf_eflags & PSL_VM) {
501 i = vm86_emulate((struct vm86frame *)&frame);
508 case T_SEGNPFLT: /* segment not present fault */
509 case T_TSSFLT: /* invalid TSS fault */
510 case T_DOUBLEFLT: /* double fault */
512 ucode = code + BUS_SEGM_FAULT ;
516 case T_PAGEFLT: /* page fault */
517 i = trap_pfault(&frame, TRUE, eva);
520 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
530 case T_DIVIDE: /* integer divide fault */
538 goto handle_powerfail;
539 #else /* !POWERFAIL_NMI */
540 /* machine/parity/power fail/"kitchen sink" faults */
541 if (isa_nmi(code) == 0) {
544 * NMI can be hooked up to a pushbutton
548 printf ("NMI ... going to debugger\n");
549 kdb_trap (type, 0, &frame);
553 } else if (panic_on_nmi)
554 panic("NMI indicates hardware failure");
556 #endif /* POWERFAIL_NMI */
557 #endif /* NISA > 0 */
559 case T_OFLOW: /* integer overflow fault */
564 case T_BOUND: /* bounds check fault */
572 * The kernel may have switched out the FP unit's
573 * state, causing the user process to take a fault
574 * when it tries to use the FP unit. Restore the
580 if (!pmath_emulate) {
582 ucode = FPE_FPU_NP_TRAP;
585 i = (*pmath_emulate)(&frame);
587 if (!(frame.tf_eflags & PSL_T))
589 frame.tf_eflags &= ~PSL_T;
592 /* else ucode = emulator_only_knows() XXX */
595 case T_FPOPFLT: /* FPU operand fetch fault */
600 case T_XMMFLT: /* SIMD floating-point exception */
610 case T_PAGEFLT: /* page fault */
611 (void) trap_pfault(&frame, FALSE, eva);
617 * The kernel may be using npx for copying or other
625 case T_PROTFLT: /* general protection fault */
626 case T_SEGNPFLT: /* segment not present fault */
628 * Invalid segment selectors and out of bounds
629 * %eip's and %esp's can be set up in user mode.
630 * This causes a fault in kernel mode when the
631 * kernel tries to return to user mode. We want
632 * to get this fault so that we can fix the
633 * problem here and not have to check all the
634 * selectors and pointers when the user changes
637 #define MAYBE_DORETI_FAULT(where, whereto) \
639 if (frame.tf_eip == (int)where) { \
640 frame.tf_eip = (int)whereto; \
645 * Since we don't save %gs across an interrupt
646 * frame this check must occur outside the intr
647 * nesting level check.
649 if (frame.tf_eip == (int)cpu_switch_load_gs) {
650 td->td_pcb->pcb_gs = 0;
654 if (mycpu->gd_intr_nesting_level == 0) {
656 * Invalid %fs's and %gs's can be created using
657 * procfs or PT_SETREGS or by invalidating the
658 * underlying LDT entry. This causes a fault
659 * in kernel mode when the kernel attempts to
660 * switch contexts. Lose the bad context
661 * (XXX) so that we can continue, and generate
664 MAYBE_DORETI_FAULT(doreti_iret,
666 MAYBE_DORETI_FAULT(doreti_popl_ds,
667 doreti_popl_ds_fault);
668 MAYBE_DORETI_FAULT(doreti_popl_es,
669 doreti_popl_es_fault);
670 MAYBE_DORETI_FAULT(doreti_popl_fs,
671 doreti_popl_fs_fault);
672 if (td->td_pcb->pcb_onfault) {
674 (register_t)td->td_pcb->pcb_onfault;
682 * PSL_NT can be set in user mode and isn't cleared
683 * automatically when the kernel is entered. This
684 * causes a TSS fault when the kernel attempts to
685 * `iret' because the TSS link is uninitialized. We
686 * want to get this fault so that we can fix the
687 * problem here and not every time the kernel is
690 if (frame.tf_eflags & PSL_NT) {
691 frame.tf_eflags &= ~PSL_NT;
696 case T_TRCTRAP: /* trace trap */
697 if (frame.tf_eip == (int)IDTVEC(syscall)) {
699 * We've just entered system mode via the
700 * syscall lcall. Continue single stepping
701 * silently until the syscall handler has
706 if (frame.tf_eip == (int)IDTVEC(syscall) + 1) {
708 * The syscall handler has now saved the
709 * flags. Stop single stepping it.
711 frame.tf_eflags &= ~PSL_T;
715 * Ignore debug register trace traps due to
716 * accesses in the user's address space, which
717 * can happen under several conditions such as
718 * if a user sets a watchpoint on a buffer and
719 * then passes that buffer to a system call.
720 * We still want to get TRCTRAPS for addresses
721 * in kernel space because that is useful when
722 * debugging the kernel.
724 if (user_dbreg_trap()) {
726 * Reset breakpoint bits because the
729 load_dr6(rdr6() & 0xfffffff0);
733 * Fall through (TRCTRAP kernel mode, kernel address)
737 * If DDB is enabled, let it handle the debugger trap.
738 * Otherwise, debugger traps "can't happen".
741 if (kdb_trap (type, 0, &frame))
750 # define TIMER_FREQ 1193182
754 static unsigned lastalert = 0;
756 if(time_second - lastalert > 10)
758 log(LOG_WARNING, "NMI: power fail\n");
759 sysbeep(TIMER_FREQ/880, hz);
760 lastalert = time_second;
765 #else /* !POWERFAIL_NMI */
766 /* machine/parity/power fail/"kitchen sink" faults */
767 if (isa_nmi(code) == 0) {
770 * NMI can be hooked up to a pushbutton
774 printf ("NMI ... going to debugger\n");
775 kdb_trap (type, 0, &frame);
779 } else if (panic_on_nmi == 0)
782 #endif /* POWERFAIL_NMI */
783 #endif /* NISA > 0 */
786 trap_fatal(&frame, eva);
790 /* Translate fault for emulators (e.g. Linux) */
791 if (*p->p_sysent->sv_transtrap)
792 i = (*p->p_sysent->sv_transtrap)(i, type);
794 trapsignal(p, i, ucode);
797 if (type <= MAX_TRAP_MSG) {
798 uprintf("fatal process exception: %s",
800 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
801 uprintf(", fault VA = 0x%lx", (u_long)eva);
808 if (ISPL(frame.tf_cs) == SEL_UPL)
809 KASSERT(td->td_mpcount == 1, ("badmpcount trap from %p", (void *)frame.tf_eip));
811 userret(p, &frame, sticks);
815 KKASSERT(td->td_mpcount > 0);
822 * This version doesn't allow a page fault to user space while
823 * in the kernel. The rest of the kernel needs to be made "safe"
824 * before this can be used. I think the only things remaining
825 * to be made safe are the iBCS2 code and the process tracing/
829 trap_pfault(frame, usermode, eva)
830 struct trapframe *frame;
835 struct vmspace *vm = NULL;
839 thread_t td = curthread;
840 struct proc *p = td->td_proc; /* may be NULL */
842 if (frame->tf_err & PGEX_W)
843 ftype = VM_PROT_WRITE;
845 ftype = VM_PROT_READ;
847 va = trunc_page(eva);
848 if (va < VM_MIN_KERNEL_ADDRESS) {
853 (!usermode && va < VM_MAXUSER_ADDRESS &&
854 (td->td_gd->gd_intr_nesting_level != 0 ||
855 td->td_pcb->pcb_onfault == NULL))) {
856 trap_fatal(frame, eva);
861 * This is a fault on non-kernel virtual memory.
862 * vm is initialized above to NULL. If curproc is NULL
863 * or curproc->p_vmspace is NULL the fault is fatal.
872 * Keep swapout from messing with us during this
878 * Grow the stack if necessary
880 /* grow_stack returns false only if va falls into
881 * a growable stack region and the stack growth
882 * fails. It returns true if va was not within
883 * a growable stack region, or if the stack
886 if (!grow_stack (p, va)) {
892 /* Fault in the user page: */
893 rv = vm_fault(map, va, ftype,
894 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
900 * Don't allow user-mode faults in kernel address space.
906 * Since we know that kernel virtual address addresses
907 * always have pte pages mapped, we just have to fault
910 rv = vm_fault(kernel_map, va, ftype, VM_FAULT_NORMAL);
913 if (rv == KERN_SUCCESS)
917 if (mtd->td_gd->gd_intr_nesting_level == 0 &&
918 td->td_pcb->pcb_onfault) {
919 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
922 trap_fatal(frame, eva);
926 /* kludge to pass faulting virtual address to sendsig */
929 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
934 trap_pfault(frame, usermode, eva)
935 struct trapframe *frame;
940 struct vmspace *vm = NULL;
944 thread_t td = curthread;
945 struct proc *p = td->td_proc;
947 va = trunc_page(eva);
948 if (va >= KERNBASE) {
950 * Don't allow user-mode faults in kernel address space.
951 * An exception: if the faulting address is the invalid
952 * instruction entry in the IDT, then the Intel Pentium
953 * F00F bug workaround was triggered, and we need to
954 * treat it is as an illegal instruction, and not a page
957 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
958 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) {
959 frame->tf_trapno = T_PRIVINFLT;
969 * This is a fault on non-kernel virtual memory.
970 * vm is initialized above to NULL. If curproc is NULL
971 * or curproc->p_vmspace is NULL the fault is fatal.
982 if (frame->tf_err & PGEX_W)
983 ftype = VM_PROT_WRITE;
985 ftype = VM_PROT_READ;
987 if (map != kernel_map) {
989 * Keep swapout from messing with us during this
995 * Grow the stack if necessary
997 /* grow_stack returns false only if va falls into
998 * a growable stack region and the stack growth
999 * fails. It returns true if va was not within
1000 * a growable stack region, or if the stack
1003 if (!grow_stack (p, va)) {
1009 /* Fault in the user page: */
1010 rv = vm_fault(map, va, ftype,
1011 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
1017 * Don't have to worry about process locking or stacks in the kernel.
1019 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
1022 if (rv == KERN_SUCCESS)
1026 if (td->td_gd->gd_intr_nesting_level == 0 &&
1027 td->td_pcb->pcb_onfault) {
1028 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
1031 trap_fatal(frame, eva);
1035 /* kludge to pass faulting virtual address to sendsig */
1036 frame->tf_err = eva;
1038 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
1042 trap_fatal(frame, eva)
1043 struct trapframe *frame;
1046 int code, type, ss, esp;
1047 struct soft_segment_descriptor softseg;
1049 code = frame->tf_err;
1050 type = frame->tf_trapno;
1051 sdtossd(&gdt[mycpu->gd_cpuid * NGDT + IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
1053 if (type <= MAX_TRAP_MSG)
1054 printf("\n\nFatal trap %d: %s while in %s mode\n",
1055 type, trap_msg[type],
1056 frame->tf_eflags & PSL_VM ? "vm86" :
1057 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
1059 /* three separate prints in case of a trap on an unmapped page */
1060 printf("mp_lock = %08x; ", mp_lock);
1061 printf("cpuid = %d; ", mycpu->gd_cpuid);
1062 printf("lapic.id = %08x\n", lapic.id);
1064 if (type == T_PAGEFLT) {
1065 printf("fault virtual address = 0x%x\n", eva);
1066 printf("fault code = %s %s, %s\n",
1067 code & PGEX_U ? "user" : "supervisor",
1068 code & PGEX_W ? "write" : "read",
1069 code & PGEX_P ? "protection violation" : "page not present");
1071 printf("instruction pointer = 0x%x:0x%x\n",
1072 frame->tf_cs & 0xffff, frame->tf_eip);
1073 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
1074 ss = frame->tf_ss & 0xffff;
1075 esp = frame->tf_esp;
1077 ss = GSEL(GDATA_SEL, SEL_KPL);
1078 esp = (int)&frame->tf_esp;
1080 printf("stack pointer = 0x%x:0x%x\n", ss, esp);
1081 printf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
1082 printf("code segment = base 0x%x, limit 0x%x, type 0x%x\n",
1083 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
1084 printf(" = DPL %d, pres %d, def32 %d, gran %d\n",
1085 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
1087 printf("processor eflags = ");
1088 if (frame->tf_eflags & PSL_T)
1089 printf("trace trap, ");
1090 if (frame->tf_eflags & PSL_I)
1091 printf("interrupt enabled, ");
1092 if (frame->tf_eflags & PSL_NT)
1093 printf("nested task, ");
1094 if (frame->tf_eflags & PSL_RF)
1096 if (frame->tf_eflags & PSL_VM)
1098 printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
1099 printf("current process = ");
1101 printf("%lu (%s)\n",
1102 (u_long)curproc->p_pid, curproc->p_comm ?
1103 curproc->p_comm : "");
1107 printf("current thread = pri %d ", curthread->td_pri);
1108 if (curthread->td_pri >= TDPRI_CRIT)
1111 printf("interrupt mask = ");
1112 if ((curthread->td_cpl & net_imask) == net_imask)
1114 if ((curthread->td_cpl & tty_imask) == tty_imask)
1116 if ((curthread->td_cpl & bio_imask) == bio_imask)
1118 if ((curthread->td_cpl & cam_imask) == cam_imask)
1120 if (curthread->td_cpl == 0)
1125 * we probably SHOULD have stopped the other CPUs before now!
1126 * another CPU COULD have been touching cpl at this moment...
1128 printf(" <- SMP: XXX");
1137 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1140 printf("trap number = %d\n", type);
1141 if (type <= MAX_TRAP_MSG)
1142 panic("%s", trap_msg[type]);
1144 panic("unknown/reserved trap");
1148 * Double fault handler. Called when a fault occurs while writing
1149 * a frame for a trap/exception onto the stack. This usually occurs
1150 * when the stack overflows (such is the case with infinite recursion,
1153 * XXX Note that the current PTD gets replaced by IdlePTD when the
1154 * task switch occurs. This means that the stack that was active at
1155 * the time of the double fault is not available at <kstack> unless
1156 * the machine was idle when the double fault occurred. The downside
1157 * of this is that "trace <ebp>" in ddb won't work.
1162 struct mdglobaldata *gd = mdcpu;
1164 printf("\nFatal double fault:\n");
1165 printf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1166 printf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1167 printf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
1169 /* three separate prints in case of a trap on an unmapped page */
1170 printf("mp_lock = %08x; ", mp_lock);
1171 printf("cpuid = %d; ", mycpu->gd_cpuid);
1172 printf("lapic.id = %08x\n", lapic.id);
1174 panic("double fault");
1178 * Compensate for 386 brain damage (missing URKR).
1179 * This is a little simpler than the pagefault handler in trap() because
1180 * it the page tables have already been faulted in and high addresses
1181 * are thrown out early for other reasons.
1191 va = trunc_page((vm_offset_t)addr);
1193 * XXX - MAX is END. Changed > to >= for temp. fix.
1195 if (va >= VM_MAXUSER_ADDRESS)
1203 if (!grow_stack (p, va)) {
1209 * fault the data page
1211 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY);
1215 if (rv != KERN_SUCCESS)
1222 * syscall2 - MP aware system call request C handler
1224 * A system call is essentially treated as a trap except that the
1225 * MP lock is not held on entry or return. We are responsible for
1226 * obtaining the MP lock if necessary and for handling ASTs
1227 * (e.g. a task switch) prior to return.
1229 * In general, only simple access and manipulation of curproc and
1230 * the current stack is allowed without having to hold MP lock.
1233 syscall2(struct trapframe frame)
1235 struct thread *td = curthread;
1236 struct proc *p = td->td_proc;
1238 struct sysent *callp;
1239 register_t orig_tf_eflags;
1244 union sysunion args;
1247 if (ISPL(frame.tf_cs) != SEL_UPL) {
1255 KASSERT(td->td_mpcount == 0, ("badmpcount syscall from %p", (void *)frame.tf_eip));
1258 userenter(td); /* lazy raise our priority */
1260 sticks = (int)td->td_sticks;
1262 p->p_md.md_regs = &frame;
1263 params = (caddr_t)frame.tf_esp + sizeof(int);
1264 code = frame.tf_eax;
1265 orig_tf_eflags = frame.tf_eflags;
1267 if (p->p_sysent->sv_prepsyscall) {
1269 * The prep code is not MP aware.
1271 (*p->p_sysent->sv_prepsyscall)(&frame, (int *)(&args.nosys.usrmsg + 1), &code, ¶ms);
1274 * Need to check if this is a 32 bit or 64 bit syscall.
1275 * fuword is MP aware.
1277 if (code == SYS_syscall) {
1279 * Code is first argument, followed by actual args.
1281 code = fuword(params);
1282 params += sizeof(int);
1283 } else if (code == SYS___syscall) {
1285 * Like syscall, but code is a quad, so as to maintain
1286 * quad alignment for the rest of the arguments.
1288 code = fuword(params);
1289 params += sizeof(quad_t);
1293 code &= p->p_sysent->sv_mask;
1294 if (code >= p->p_sysent->sv_size)
1295 callp = &p->p_sysent->sv_table[0];
1297 callp = &p->p_sysent->sv_table[code];
1299 narg = callp->sy_narg & SYF_ARGMASK;
1302 * copyin is MP aware, but the tracing code is not
1304 if (narg && params) {
1305 error = copyin(params, (caddr_t)(&args.nosys.usrmsg + 1),
1306 narg * sizeof(register_t));
1309 if (KTRPOINT(td, KTR_SYSCALL))
1310 ktrsyscall(p->p_tracep, code, narg,
1311 (void *)(&args.nosys.usrmsg + 1));
1319 * Try to run the syscall without the MP lock if the syscall
1320 * is MP safe. We have to obtain the MP lock no matter what if
1323 if ((callp->sy_narg & SYF_MPSAFE) == 0) {
1330 if (KTRPOINT(td, KTR_SYSCALL)) {
1331 ktrsyscall(p->p_tracep, code, narg, (void *)(&args.nosys.usrmsg + 1));
1336 * For traditional syscall code edx is left untouched when 32 bit
1337 * results are returned. Since edx is loaded from fds[1] when the
1338 * system call returns we pre-set it here.
1340 lwkt_initmsg(&args.lmsg, &td->td_msgport, 0,
1341 lwkt_cmd_op(code), lwkt_cmd_op_none);
1342 args.sysmsg_copyout = NULL;
1343 args.sysmsg_fds[0] = 0;
1344 args.sysmsg_fds[1] = frame.tf_edx;
1346 STOPEVENT(p, S_SCE, narg); /* MP aware */
1348 error = (*callp->sy_call)(&args);
1351 * MP SAFE (we may or may not have the MP lock at this point)
1356 * Reinitialize proc pointer `p' as it may be different
1357 * if this is a child returning from fork syscall.
1360 frame.tf_eax = args.sysmsg_fds[0];
1361 frame.tf_edx = args.sysmsg_fds[1];
1362 frame.tf_eflags &= ~PSL_C;
1366 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1367 * int 0x80 is 2 bytes. We saved this in tf_err.
1369 frame.tf_eip -= frame.tf_err;
1374 panic("Unexpected EASYNC return value (for now)");
1377 if (p->p_sysent->sv_errsize) {
1378 if (error >= p->p_sysent->sv_errsize)
1379 error = -1; /* XXX */
1381 error = p->p_sysent->sv_errtbl[error];
1383 frame.tf_eax = error;
1384 frame.tf_eflags |= PSL_C;
1389 * Traced syscall. trapsignal() is not MP aware.
1391 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1392 frame.tf_eflags &= ~PSL_T;
1393 trapsignal(p, SIGTRAP, 0);
1397 * Handle reschedule and other end-of-syscall issues
1399 userret(p, &frame, sticks);
1402 if (KTRPOINT(td, KTR_SYSRET)) {
1403 ktrsysret(p->p_tracep, code, error, args.sysmsg_result);
1408 * This works because errno is findable through the
1409 * register set. If we ever support an emulation where this
1410 * is not the case, this code will need to be revisited.
1412 STOPEVENT(p, S_SCX, code);
1417 * Release the MP lock if we had to get it
1419 KASSERT(td->td_mpcount == 1, ("badmpcount syscall from %p", (void *)frame.tf_eip));
1425 * sendsys2 - MP aware system message request C handler
1428 sendsys2(struct trapframe frame)
1430 struct globaldata *gd;
1431 struct thread *td = curthread;
1432 struct proc *p = td->td_proc;
1433 register_t orig_tf_eflags;
1434 struct sysent *callp;
1435 union sysunion *sysun;
1445 if (ISPL(frame.tf_cs) != SEL_UPL) {
1453 KASSERT(td->td_mpcount == 0, ("badmpcount syscall from %p", (void *)frame.tf_eip));
1457 * access non-atomic field from critical section. p_sticks is
1458 * updated by the clock interrupt. Also use this opportunity
1459 * to lazy-raise our LWKT priority.
1462 sticks = td->td_sticks;
1464 p->p_md.md_regs = &frame;
1465 orig_tf_eflags = frame.tf_eflags;
1469 * Handle the waitport/waitmsg/checkport/checkmsg case
1471 * YYY MOVE THIS TO INT 0x82! We don't really need to combine it
1474 if ((msgsize = frame.tf_edx) <= 0) {
1476 printf("waitmsg/checkmsg not yet supported: %08x\n",
1482 printf("waitport/checkport only the default port is supported at the moment\n");
1488 sysun = (void *)sysmsg_wait(p, NULL, 0);
1491 sysun = (void *)sysmsg_wait(p, NULL, 1);
1499 umsg = sysun->lmsg.opaque.ms_umsg;
1500 frame.tf_eax = (register_t)umsg;
1501 if (sysun->sysmsg_copyout)
1502 sysun->sysmsg_copyout(sysun);
1503 sysun->nosys.usrmsg.umsg.u.ms_fds[0] = sysun->lmsg.u.ms_fds[0];
1504 sysun->nosys.usrmsg.umsg.u.ms_fds[1] = sysun->lmsg.u.ms_fds[1];
1505 sysun->nosys.usrmsg.umsg.ms_error = sysun->lmsg.ms_error;
1506 error = sysun->lmsg.ms_error;
1507 result = sysun->lmsg.u.ms_fds[0]; /* for ktrace */
1508 if (error != 0 || code != SYS_execve) {
1510 &sysun->nosys.usrmsg.umsg.ms_copyout_start,
1511 &umsg->ms_copyout_start,
1514 crit_enter_quick(td);
1515 sysun->lmsg.opaque.ms_sysunnext = gd->gd_freesysun;
1516 gd->gd_freesysun = sysun;
1517 crit_exit_quick(td);
1528 * Extract the system call message. If msgsize is zero we are
1529 * blocking on a message and/or message port. If msgsize is -1
1530 * we are testing a message for completion or a message port for
1533 * The userland system call message size includes the size of the
1534 * userland lwkt_msg plus arguments. We load it into the userland
1535 * portion of our sysunion structure then we initialize the kerneland
1542 if (msgsize < sizeof(struct lwkt_msg) ||
1543 msgsize > sizeof(union sysunion) - sizeof(struct sysmsg)
1550 * Obtain a sysun from our per-cpu cache or allocate a new one. Use
1551 * the opaque field to store the original (user) message pointer.
1552 * A critical section is necessary to interlock against interrupts
1553 * returning system messages to the thread cache.
1556 crit_enter_quick(td);
1557 if ((sysun = gd->gd_freesysun) != NULL) {
1558 gd->gd_freesysun = sysun->lmsg.opaque.ms_sysunnext;
1559 crit_exit_quick(td);
1561 crit_exit_quick(td);
1562 sysun = malloc(sizeof(union sysunion), M_SYSMSG, M_WAITOK);
1566 * Copy the user request into the kernel copy of the user request.
1568 umsg = (void *)frame.tf_ecx;
1569 error = copyin(umsg, &sysun->nosys.usrmsg, msgsize);
1572 if ((sysun->nosys.usrmsg.umsg.ms_flags & MSGF_ASYNC) &&
1573 (error = suser(td)) != 0
1579 * Initialize the kernel message from the copied-in data and
1580 * pull in appropriate flags from the userland message.
1582 * ms_abort_port is usually initialized in sendmsg/domsg, but since
1583 * we are not calling those functions (yet), we have to do it manually.
1585 lwkt_initmsg(&sysun->lmsg, &td->td_msgport, 0,
1586 sysun->nosys.usrmsg.umsg.ms_cmd,
1588 sysun->lmsg.ms_abort_port = sysun->lmsg.ms_reply_port;
1589 sysun->sysmsg_copyout = NULL;
1590 sysun->lmsg.opaque.ms_umsg = umsg;
1591 sysun->lmsg.ms_flags |= sysun->nosys.usrmsg.umsg.ms_flags & MSGF_ASYNC;
1594 * Extract the system call number, lookup the system call, and
1595 * set the default return value.
1597 code = (u_int)sysun->lmsg.ms_cmd.cm_op;
1598 if (code >= p->p_sysent->sv_size) {
1603 callp = &p->p_sysent->sv_table[code];
1605 narg = (msgsize - sizeof(struct lwkt_msg)) / sizeof(register_t);
1608 if (KTRPOINT(td, KTR_SYSCALL)) {
1609 ktrsyscall(p->p_tracep, code, narg, (void *)(&sysun->nosys.usrmsg + 1));
1612 sysun->lmsg.u.ms_fds[0] = 0;
1613 sysun->lmsg.u.ms_fds[1] = 0;
1615 STOPEVENT(p, S_SCE, narg); /* MP aware */
1618 * Make the system call. An error code is always returned, results
1619 * are copied back via ms_result32 or ms_result64. YYY temporary
1620 * stage copy p_retval[] into ms_result32/64
1622 * NOTE! XXX if this is a child returning from a fork curproc
1623 * might be different. YYY huh? a child returning from a fork
1624 * should never 'return' from this call, it should go right to the
1625 * fork_trampoline function.
1627 error = (*callp->sy_call)(sysun);
1628 gd = td->td_gd; /* RELOAD, might have switched cpus */
1632 * If a synchronous return copy p_retval to ms_result64 and return
1633 * the sysmsg to the free pool.
1635 * YYY Don't writeback message if execve() YYY
1637 if (error == EASYNC) {
1639 * Since only the current process ever messes with msgq,
1640 * we can safely manipulate it in parallel with the async
1643 TAILQ_INSERT_TAIL(&p->p_sysmsgq, &sysun->sysmsg, msgq);
1645 sysun->nosys.usrmsg.umsg.u.ms_fds[0] = sysun->lmsg.u.ms_fds[0];
1646 sysun->nosys.usrmsg.umsg.u.ms_fds[1] = sysun->lmsg.u.ms_fds[1];
1647 result = sysun->nosys.usrmsg.umsg.u.ms_fds[0]; /* for ktrace */
1648 if (error != 0 || code != SYS_execve) {
1650 error2 = copyout(&sysun->nosys.usrmsg.umsg.ms_copyout_start,
1651 &umsg->ms_copyout_start,
1656 crit_enter_quick(td);
1657 sysun->lmsg.opaque.ms_sysunnext = gd->gd_freesysun;
1658 gd->gd_freesysun = sysun;
1659 crit_exit_quick(td);
1662 frame.tf_eax = error;
1666 * Traced syscall. trapsignal() is not MP aware.
1668 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1669 frame.tf_eflags &= ~PSL_T;
1670 trapsignal(p, SIGTRAP, 0);
1674 * Handle reschedule and other end-of-syscall issues
1676 userret(p, &frame, sticks);
1679 if (KTRPOINT(td, KTR_SYSRET)) {
1680 ktrsysret(p->p_tracep, code, error, result);
1685 * This works because errno is findable through the
1686 * register set. If we ever support an emulation where this
1687 * is not the case, this code will need to be revisited.
1689 STOPEVENT(p, S_SCX, code);
1694 * Release the MP lock if we had to get it
1696 KASSERT(td->td_mpcount == 1, ("badmpcount syscall from %p", (void *)frame.tf_eip));
1702 * Simplified back end of syscall(), used when returning from fork()
1703 * directly into user mode. MP lock is held on entry and should be
1704 * released on return. This code will return back into the fork
1705 * trampoline code which then runs doreti.
1708 fork_return(p, frame)
1710 struct trapframe frame;
1712 frame.tf_eax = 0; /* Child returns zero */
1713 frame.tf_eflags &= ~PSL_C; /* success */
1716 userret(p, &frame, 0);
1718 if (KTRPOINT(p->p_thread, KTR_SYSRET))
1719 ktrsysret(p->p_tracep, SYS_fork, 0, 0);
1721 p->p_flag |= P_PASSIVE_ACQ;
1723 p->p_flag &= ~P_PASSIVE_ACQ;
1725 KKASSERT(p->p_thread->td_mpcount == 1);