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 1.48 2004/03/30 19:14:04 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 * USER->KERNEL transition. Do not transition us out of userland from the
176 * point of view of the userland scheduler unless we actually have to
177 * switch. Switching typically occurs when a process blocks in the kernel.
179 * passive_release is called from within a critical section and the BGL will
180 * still be held. This function is NOT called for preemptions, only for
181 * switchouts. Note that other elements of the system (uio_yield()) assume
182 * that the user cruft will be released when lwkt_switch() is called.
185 passive_release(struct thread *td)
187 struct proc *p = td->td_proc;
189 td->td_release = NULL;
194 * userenter() passively intercepts the thread switch function to increase
195 * the thread priority from a user priority to a kernel priority, reducing
196 * syscall and trap overhead for the case where no switch occurs.
200 userenter(struct thread *curtd)
202 curtd->td_release = passive_release;
206 * Reacquire our current process designation. This will not return until
207 * we have it. Our LWKT priority will be adjusted for our return to
208 * userland. acquire_curproc() also handles cleaning up P_CP_RELEASED.
210 * This is always the last step before returning to user mode.
213 userexit(struct proc *p)
215 struct thread *td = p->p_thread;
217 td->td_release = NULL;
218 if (p->p_flag & P_CP_RELEASED)
226 userret(struct proc *p, struct trapframe *frame, u_quad_t oticks)
231 * Post any pending upcalls
233 if (p->p_flag & P_UPCALLPEND) {
234 p->p_flag &= ~P_UPCALLPEND;
239 * Post any pending signals
241 while ((sig = CURSIG(p)) != 0) {
246 * If a reschedule has been requested then we release the current
247 * process in order to shift the current process designation to
248 * another user process and/or to switch to a higher priority
249 * kernel thread at userexit() time.
251 if (any_resched_wanted()) {
252 p->p_thread->td_release = NULL;
257 * Charge system time if profiling. Note: times are in microseconds.
259 if (p->p_flag & P_PROFIL) {
260 addupc_task(p, frame->tf_eip,
261 (u_int)(curthread->td_sticks - oticks));
265 * Post any pending signals XXX
267 while ((sig = CURSIG(p)) != 0)
271 #ifdef DEVICE_POLLING
272 extern u_int32_t poll_in_trap;
273 extern int ether_poll (int count);
274 #endif /* DEVICE_POLLING */
277 * Exception, fault, and trap interface to the FreeBSD kernel.
278 * This common code is called from assembly language IDT gate entry
279 * routines that prepare a suitable stack frame, and restore this
280 * frame after the exception has been processed.
282 * This function is also called from doreti in an interlock to handle ASTs.
283 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
285 * NOTE! We have to retrieve the fault address prior to obtaining the
286 * MP lock because get_mplock() may switch out. YYY cr2 really ought
287 * to be retrieved by the assembly code, not here.
291 struct trapframe frame;
293 struct thread *td = curthread;
296 int i = 0, ucode = 0, type, code;
302 eva = (frame.tf_trapno == T_PAGEFLT ? rcr2() : 0);
304 trap_fatal(&frame, eva);
310 if (frame.tf_trapno == T_PAGEFLT) {
312 * For some Cyrix CPUs, %cr2 is clobbered by interrupts.
313 * This problem is worked around by using an interrupt
314 * gate for the pagefault handler. We are finally ready
315 * to read %cr2 and then must reenable interrupts.
317 * XXX this should be in the switch statement, but the
318 * NO_FOOF_HACK and VM86 goto and ifdefs obfuscate the
319 * flow of control too much for this to be obviously
329 * MP lock is held at this point
332 if (!(frame.tf_eflags & PSL_I)) {
334 * Buggy application or kernel code has disabled interrupts
335 * and then trapped. Enabling interrupts now is wrong, but
336 * it is better than running with interrupts disabled until
337 * they are accidentally enabled later.
339 type = frame.tf_trapno;
340 if (ISPL(frame.tf_cs)==SEL_UPL || (frame.tf_eflags & PSL_VM)) {
342 "pid %ld (%s): trap %d with interrupts disabled\n",
343 (long)curproc->p_pid, curproc->p_comm, type);
344 } else if (type != T_BPTFLT && type != T_TRCTRAP) {
346 * XXX not quite right, since this may be for a
347 * multiple fault in user mode.
349 printf("kernel trap %d with interrupts disabled\n",
356 #ifdef DEVICE_POLLING
358 ether_poll(poll_in_trap);
359 #endif /* DEVICE_POLLING */
361 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
364 type = frame.tf_trapno;
368 if (frame.tf_eflags & PSL_VM &&
369 (type == T_PROTFLT || type == T_STKFLT)) {
371 KKASSERT(curthread->td_mpcount > 0);
373 i = vm86_emulate((struct vm86frame *)&frame);
375 KKASSERT(curthread->td_mpcount > 0);
379 * returns to original process
381 vm86_trap((struct vm86frame *)&frame);
388 * these traps want either a process context, or
389 * assume a normal userspace trap.
393 trap_fatal(&frame, eva);
396 type = T_BPTFLT; /* kernel breakpoint */
399 goto kernel_trap; /* normal kernel trap handling */
402 if ((ISPL(frame.tf_cs) == SEL_UPL) || (frame.tf_eflags & PSL_VM)) {
407 sticks = curthread->td_sticks;
408 p->p_md.md_regs = &frame;
411 case T_PRIVINFLT: /* privileged instruction fault */
416 case T_BPTFLT: /* bpt instruction fault */
417 case T_TRCTRAP: /* trace trap */
418 frame.tf_eflags &= ~PSL_T;
422 case T_ARITHTRAP: /* arithmetic trap */
427 case T_ASTFLT: /* Allow process switch */
428 mycpu->gd_cnt.v_soft++;
429 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
430 atomic_clear_int_nonlocked(&mycpu->gd_reqflags,
432 addupc_task(p, p->p_stats->p_prof.pr_addr,
433 p->p_stats->p_prof.pr_ticks);
438 * The following two traps can happen in
439 * vm86 mode, and, if so, we want to handle
442 case T_PROTFLT: /* general protection fault */
443 case T_STKFLT: /* stack fault */
444 if (frame.tf_eflags & PSL_VM) {
445 i = vm86_emulate((struct vm86frame *)&frame);
452 case T_SEGNPFLT: /* segment not present fault */
453 case T_TSSFLT: /* invalid TSS fault */
454 case T_DOUBLEFLT: /* double fault */
456 ucode = code + BUS_SEGM_FAULT ;
460 case T_PAGEFLT: /* page fault */
461 i = trap_pfault(&frame, TRUE, eva);
464 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
474 case T_DIVIDE: /* integer divide fault */
482 goto handle_powerfail;
483 #else /* !POWERFAIL_NMI */
484 /* machine/parity/power fail/"kitchen sink" faults */
485 if (isa_nmi(code) == 0) {
488 * NMI can be hooked up to a pushbutton
492 printf ("NMI ... going to debugger\n");
493 kdb_trap (type, 0, &frame);
497 } else if (panic_on_nmi)
498 panic("NMI indicates hardware failure");
500 #endif /* POWERFAIL_NMI */
501 #endif /* NISA > 0 */
503 case T_OFLOW: /* integer overflow fault */
508 case T_BOUND: /* bounds check fault */
515 /* if a transparent fault (due to context switch "late") */
519 if (!pmath_emulate) {
521 ucode = FPE_FPU_NP_TRAP;
524 i = (*pmath_emulate)(&frame);
526 if (!(frame.tf_eflags & PSL_T))
528 frame.tf_eflags &= ~PSL_T;
531 /* else ucode = emulator_only_knows() XXX */
534 case T_FPOPFLT: /* FPU operand fetch fault */
539 case T_XMMFLT: /* SIMD floating-point exception */
549 case T_PAGEFLT: /* page fault */
550 (void) trap_pfault(&frame, FALSE, eva);
556 * The kernel is apparently using npx for copying.
557 * XXX this should be fatal unless the kernel has
558 * registered such use.
565 case T_PROTFLT: /* general protection fault */
566 case T_SEGNPFLT: /* segment not present fault */
568 * Invalid segment selectors and out of bounds
569 * %eip's and %esp's can be set up in user mode.
570 * This causes a fault in kernel mode when the
571 * kernel tries to return to user mode. We want
572 * to get this fault so that we can fix the
573 * problem here and not have to check all the
574 * selectors and pointers when the user changes
577 #define MAYBE_DORETI_FAULT(where, whereto) \
579 if (frame.tf_eip == (int)where) { \
580 frame.tf_eip = (int)whereto; \
585 * Since we don't save %gs across an interrupt
586 * frame this check must occur outside the intr
587 * nesting level check.
589 if (frame.tf_eip == (int)cpu_switch_load_gs) {
590 curthread->td_pcb->pcb_gs = 0;
594 if (mycpu->gd_intr_nesting_level == 0) {
596 * Invalid %fs's and %gs's can be created using
597 * procfs or PT_SETREGS or by invalidating the
598 * underlying LDT entry. This causes a fault
599 * in kernel mode when the kernel attempts to
600 * switch contexts. Lose the bad context
601 * (XXX) so that we can continue, and generate
604 MAYBE_DORETI_FAULT(doreti_iret,
606 MAYBE_DORETI_FAULT(doreti_popl_ds,
607 doreti_popl_ds_fault);
608 MAYBE_DORETI_FAULT(doreti_popl_es,
609 doreti_popl_es_fault);
610 MAYBE_DORETI_FAULT(doreti_popl_fs,
611 doreti_popl_fs_fault);
612 if (curthread->td_pcb->pcb_onfault) {
613 frame.tf_eip = (int)curthread->td_pcb->pcb_onfault;
621 * PSL_NT can be set in user mode and isn't cleared
622 * automatically when the kernel is entered. This
623 * causes a TSS fault when the kernel attempts to
624 * `iret' because the TSS link is uninitialized. We
625 * want to get this fault so that we can fix the
626 * problem here and not every time the kernel is
629 if (frame.tf_eflags & PSL_NT) {
630 frame.tf_eflags &= ~PSL_NT;
635 case T_TRCTRAP: /* trace trap */
636 if (frame.tf_eip == (int)IDTVEC(syscall)) {
638 * We've just entered system mode via the
639 * syscall lcall. Continue single stepping
640 * silently until the syscall handler has
645 if (frame.tf_eip == (int)IDTVEC(syscall) + 1) {
647 * The syscall handler has now saved the
648 * flags. Stop single stepping it.
650 frame.tf_eflags &= ~PSL_T;
654 * Ignore debug register trace traps due to
655 * accesses in the user's address space, which
656 * can happen under several conditions such as
657 * if a user sets a watchpoint on a buffer and
658 * then passes that buffer to a system call.
659 * We still want to get TRCTRAPS for addresses
660 * in kernel space because that is useful when
661 * debugging the kernel.
663 if (user_dbreg_trap()) {
665 * Reset breakpoint bits because the
668 load_dr6(rdr6() & 0xfffffff0);
672 * Fall through (TRCTRAP kernel mode, kernel address)
676 * If DDB is enabled, let it handle the debugger trap.
677 * Otherwise, debugger traps "can't happen".
680 if (kdb_trap (type, 0, &frame))
689 # define TIMER_FREQ 1193182
693 static unsigned lastalert = 0;
695 if(time_second - lastalert > 10)
697 log(LOG_WARNING, "NMI: power fail\n");
698 sysbeep(TIMER_FREQ/880, hz);
699 lastalert = time_second;
704 #else /* !POWERFAIL_NMI */
705 /* machine/parity/power fail/"kitchen sink" faults */
706 if (isa_nmi(code) == 0) {
709 * NMI can be hooked up to a pushbutton
713 printf ("NMI ... going to debugger\n");
714 kdb_trap (type, 0, &frame);
718 } else if (panic_on_nmi == 0)
721 #endif /* POWERFAIL_NMI */
722 #endif /* NISA > 0 */
725 trap_fatal(&frame, eva);
729 /* Translate fault for emulators (e.g. Linux) */
730 if (*p->p_sysent->sv_transtrap)
731 i = (*p->p_sysent->sv_transtrap)(i, type);
733 trapsignal(p, i, ucode);
736 if (type <= MAX_TRAP_MSG) {
737 uprintf("fatal process exception: %s",
739 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
740 uprintf(", fault VA = 0x%lx", (u_long)eva);
747 if (ISPL(frame.tf_cs) == SEL_UPL)
748 KASSERT(curthread->td_mpcount == 1, ("badmpcount trap from %p", (void *)frame.tf_eip));
750 userret(p, &frame, sticks);
754 KKASSERT(curthread->td_mpcount > 0);
761 * This version doesn't allow a page fault to user space while
762 * in the kernel. The rest of the kernel needs to be made "safe"
763 * before this can be used. I think the only things remaining
764 * to be made safe are the iBCS2 code and the process tracing/
768 trap_pfault(frame, usermode, eva)
769 struct trapframe *frame;
774 struct vmspace *vm = NULL;
778 struct proc *p = curproc;
780 if (frame->tf_err & PGEX_W)
781 ftype = VM_PROT_WRITE;
783 ftype = VM_PROT_READ;
785 va = trunc_page(eva);
786 if (va < VM_MIN_KERNEL_ADDRESS) {
791 (!usermode && va < VM_MAXUSER_ADDRESS &&
792 (mycpu->gd_intr_nesting_level != 0 ||
793 curthread->td_pcb->pcb_onfault == NULL))) {
794 trap_fatal(frame, eva);
799 * This is a fault on non-kernel virtual memory.
800 * vm is initialized above to NULL. If curproc is NULL
801 * or curproc->p_vmspace is NULL the fault is fatal.
810 * Keep swapout from messing with us during this
816 * Grow the stack if necessary
818 /* grow_stack returns false only if va falls into
819 * a growable stack region and the stack growth
820 * fails. It returns true if va was not within
821 * a growable stack region, or if the stack
824 if (!grow_stack (p, va)) {
830 /* Fault in the user page: */
831 rv = vm_fault(map, va, ftype,
832 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
838 * Don't allow user-mode faults in kernel address space.
844 * Since we know that kernel virtual address addresses
845 * always have pte pages mapped, we just have to fault
848 rv = vm_fault(kernel_map, va, ftype, VM_FAULT_NORMAL);
851 if (rv == KERN_SUCCESS)
855 if (mycpu->gd_intr_nesting_level == 0 && curthread->td_pcb->pcb_onfault) {
856 frame->tf_eip = (int)curthread->td_pcb->pcb_onfault;
859 trap_fatal(frame, eva);
863 /* kludge to pass faulting virtual address to sendsig */
866 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
871 trap_pfault(frame, usermode, eva)
872 struct trapframe *frame;
877 struct vmspace *vm = NULL;
881 struct proc *p = curproc;
883 va = trunc_page(eva);
884 if (va >= KERNBASE) {
886 * Don't allow user-mode faults in kernel address space.
887 * An exception: if the faulting address is the invalid
888 * instruction entry in the IDT, then the Intel Pentium
889 * F00F bug workaround was triggered, and we need to
890 * treat it is as an illegal instruction, and not a page
893 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
894 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) {
895 frame->tf_trapno = T_PRIVINFLT;
905 * This is a fault on non-kernel virtual memory.
906 * vm is initialized above to NULL. If curproc is NULL
907 * or curproc->p_vmspace is NULL the fault is fatal.
918 if (frame->tf_err & PGEX_W)
919 ftype = VM_PROT_WRITE;
921 ftype = VM_PROT_READ;
923 if (map != kernel_map) {
925 * Keep swapout from messing with us during this
931 * Grow the stack if necessary
933 /* grow_stack returns false only if va falls into
934 * a growable stack region and the stack growth
935 * fails. It returns true if va was not within
936 * a growable stack region, or if the stack
939 if (!grow_stack (p, va)) {
945 /* Fault in the user page: */
946 rv = vm_fault(map, va, ftype,
947 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
953 * Don't have to worry about process locking or stacks in the kernel.
955 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
958 if (rv == KERN_SUCCESS)
962 if (mycpu->gd_intr_nesting_level == 0 && curthread->td_pcb->pcb_onfault) {
963 frame->tf_eip = (int)curthread->td_pcb->pcb_onfault;
966 trap_fatal(frame, eva);
970 /* kludge to pass faulting virtual address to sendsig */
973 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
977 trap_fatal(frame, eva)
978 struct trapframe *frame;
981 int code, type, ss, esp;
982 struct soft_segment_descriptor softseg;
984 code = frame->tf_err;
985 type = frame->tf_trapno;
986 sdtossd(&gdt[mycpu->gd_cpuid * NGDT + IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
988 if (type <= MAX_TRAP_MSG)
989 printf("\n\nFatal trap %d: %s while in %s mode\n",
990 type, trap_msg[type],
991 frame->tf_eflags & PSL_VM ? "vm86" :
992 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
994 /* three separate prints in case of a trap on an unmapped page */
995 printf("mp_lock = %08x; ", mp_lock);
996 printf("cpuid = %d; ", mycpu->gd_cpuid);
997 printf("lapic.id = %08x\n", lapic.id);
999 if (type == T_PAGEFLT) {
1000 printf("fault virtual address = 0x%x\n", eva);
1001 printf("fault code = %s %s, %s\n",
1002 code & PGEX_U ? "user" : "supervisor",
1003 code & PGEX_W ? "write" : "read",
1004 code & PGEX_P ? "protection violation" : "page not present");
1006 printf("instruction pointer = 0x%x:0x%x\n",
1007 frame->tf_cs & 0xffff, frame->tf_eip);
1008 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
1009 ss = frame->tf_ss & 0xffff;
1010 esp = frame->tf_esp;
1012 ss = GSEL(GDATA_SEL, SEL_KPL);
1013 esp = (int)&frame->tf_esp;
1015 printf("stack pointer = 0x%x:0x%x\n", ss, esp);
1016 printf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
1017 printf("code segment = base 0x%x, limit 0x%x, type 0x%x\n",
1018 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
1019 printf(" = DPL %d, pres %d, def32 %d, gran %d\n",
1020 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
1022 printf("processor eflags = ");
1023 if (frame->tf_eflags & PSL_T)
1024 printf("trace trap, ");
1025 if (frame->tf_eflags & PSL_I)
1026 printf("interrupt enabled, ");
1027 if (frame->tf_eflags & PSL_NT)
1028 printf("nested task, ");
1029 if (frame->tf_eflags & PSL_RF)
1031 if (frame->tf_eflags & PSL_VM)
1033 printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
1034 printf("current process = ");
1036 printf("%lu (%s)\n",
1037 (u_long)curproc->p_pid, curproc->p_comm ?
1038 curproc->p_comm : "");
1042 printf("current thread = pri %d ", curthread->td_pri);
1043 if (curthread->td_pri >= TDPRI_CRIT)
1046 printf("interrupt mask = ");
1047 if ((curthread->td_cpl & net_imask) == net_imask)
1049 if ((curthread->td_cpl & tty_imask) == tty_imask)
1051 if ((curthread->td_cpl & bio_imask) == bio_imask)
1053 if ((curthread->td_cpl & cam_imask) == cam_imask)
1055 if (curthread->td_cpl == 0)
1060 * we probably SHOULD have stopped the other CPUs before now!
1061 * another CPU COULD have been touching cpl at this moment...
1063 printf(" <- SMP: XXX");
1072 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1075 printf("trap number = %d\n", type);
1076 if (type <= MAX_TRAP_MSG)
1077 panic("%s", trap_msg[type]);
1079 panic("unknown/reserved trap");
1083 * Double fault handler. Called when a fault occurs while writing
1084 * a frame for a trap/exception onto the stack. This usually occurs
1085 * when the stack overflows (such is the case with infinite recursion,
1088 * XXX Note that the current PTD gets replaced by IdlePTD when the
1089 * task switch occurs. This means that the stack that was active at
1090 * the time of the double fault is not available at <kstack> unless
1091 * the machine was idle when the double fault occurred. The downside
1092 * of this is that "trace <ebp>" in ddb won't work.
1097 struct mdglobaldata *gd = mdcpu;
1099 printf("\nFatal double fault:\n");
1100 printf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1101 printf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1102 printf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
1104 /* three separate prints in case of a trap on an unmapped page */
1105 printf("mp_lock = %08x; ", mp_lock);
1106 printf("cpuid = %d; ", mycpu->gd_cpuid);
1107 printf("lapic.id = %08x\n", lapic.id);
1109 panic("double fault");
1113 * Compensate for 386 brain damage (missing URKR).
1114 * This is a little simpler than the pagefault handler in trap() because
1115 * it the page tables have already been faulted in and high addresses
1116 * are thrown out early for other reasons.
1126 va = trunc_page((vm_offset_t)addr);
1128 * XXX - MAX is END. Changed > to >= for temp. fix.
1130 if (va >= VM_MAXUSER_ADDRESS)
1138 if (!grow_stack (p, va)) {
1144 * fault the data page
1146 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY);
1150 if (rv != KERN_SUCCESS)
1157 * syscall2 - MP aware system call request C handler
1159 * A system call is essentially treated as a trap except that the
1160 * MP lock is not held on entry or return. We are responsible for
1161 * obtaining the MP lock if necessary and for handling ASTs
1162 * (e.g. a task switch) prior to return.
1164 * In general, only simple access and manipulation of curproc and
1165 * the current stack is allowed without having to hold MP lock.
1168 syscall2(struct trapframe frame)
1170 struct thread *td = curthread;
1171 struct proc *p = td->td_proc;
1174 struct sysent *callp;
1175 register_t orig_tf_eflags;
1180 union sysunion args;
1183 if (ISPL(frame.tf_cs) != SEL_UPL) {
1191 KASSERT(curthread->td_mpcount == 0, ("badmpcount syscall from %p", (void *)frame.tf_eip));
1195 * access non-atomic field from critical section. p_sticks is
1196 * updated by the clock interrupt. Also use this opportunity
1197 * to lazy-raise our LWKT priority.
1200 crit_enter_quick(td);
1201 sticks = curthread->td_sticks;
1202 crit_exit_quick(td);
1204 p->p_md.md_regs = &frame;
1205 params = (caddr_t)frame.tf_esp + sizeof(int);
1206 code = frame.tf_eax;
1207 orig_tf_eflags = frame.tf_eflags;
1209 if (p->p_sysent->sv_prepsyscall) {
1211 * The prep code is not MP aware.
1213 (*p->p_sysent->sv_prepsyscall)(&frame, (int *)(&args.nosys.usrmsg + 1), &code, ¶ms);
1216 * Need to check if this is a 32 bit or 64 bit syscall.
1217 * fuword is MP aware.
1219 if (code == SYS_syscall) {
1221 * Code is first argument, followed by actual args.
1223 code = fuword(params);
1224 params += sizeof(int);
1225 } else if (code == SYS___syscall) {
1227 * Like syscall, but code is a quad, so as to maintain
1228 * quad alignment for the rest of the arguments.
1230 code = fuword(params);
1231 params += sizeof(quad_t);
1235 if (p->p_sysent->sv_mask)
1236 code &= p->p_sysent->sv_mask;
1238 if (code >= p->p_sysent->sv_size)
1239 callp = &p->p_sysent->sv_table[0];
1241 callp = &p->p_sysent->sv_table[code];
1243 narg = callp->sy_narg & SYF_ARGMASK;
1246 * copyin is MP aware, but the tracing code is not
1248 if (params && (i = narg * sizeof(register_t)) &&
1249 (error = copyin(params, (caddr_t)(&args.nosys.usrmsg + 1), (u_int)i))) {
1251 if (KTRPOINT(td, KTR_SYSCALL))
1252 ktrsyscall(p->p_tracep, code, narg, (void *)(&args.nosys.usrmsg + 1));
1259 * Try to run the syscall without the MP lock if the syscall
1260 * is MP safe. We have to obtain the MP lock no matter what if
1263 if ((callp->sy_narg & SYF_MPSAFE) == 0) {
1270 if (KTRPOINT(td, KTR_SYSCALL)) {
1271 ktrsyscall(p->p_tracep, code, narg, (void *)(&args.nosys.usrmsg + 1));
1276 * For traditional syscall code edx is left untouched when 32 bit
1277 * results are returned. Since edx is loaded from fds[1] when the
1278 * system call returns we pre-set it here.
1280 lwkt_initmsg_rp(&args.lmsg, &td->td_msgport, code);
1281 args.sysmsg_copyout = NULL;
1282 args.sysmsg_fds[0] = 0;
1283 args.sysmsg_fds[1] = frame.tf_edx;
1285 STOPEVENT(p, S_SCE, narg); /* MP aware */
1287 error = (*callp->sy_call)(&args);
1290 * MP SAFE (we may or may not have the MP lock at this point)
1295 * Reinitialize proc pointer `p' as it may be different
1296 * if this is a child returning from fork syscall.
1299 frame.tf_eax = args.sysmsg_fds[0];
1300 frame.tf_edx = args.sysmsg_fds[1];
1301 frame.tf_eflags &= ~PSL_C;
1305 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1306 * int 0x80 is 2 bytes. We saved this in tf_err.
1308 frame.tf_eip -= frame.tf_err;
1313 panic("Unexpected EASYNC return value (for now)");
1316 if (p->p_sysent->sv_errsize) {
1317 if (error >= p->p_sysent->sv_errsize)
1318 error = -1; /* XXX */
1320 error = p->p_sysent->sv_errtbl[error];
1322 frame.tf_eax = error;
1323 frame.tf_eflags |= PSL_C;
1328 * Traced syscall. trapsignal() is not MP aware.
1330 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1331 frame.tf_eflags &= ~PSL_T;
1332 trapsignal(p, SIGTRAP, 0);
1336 * Handle reschedule and other end-of-syscall issues
1338 userret(p, &frame, sticks);
1341 if (KTRPOINT(td, KTR_SYSRET)) {
1342 ktrsysret(p->p_tracep, code, error, args.sysmsg_result);
1347 * This works because errno is findable through the
1348 * register set. If we ever support an emulation where this
1349 * is not the case, this code will need to be revisited.
1351 STOPEVENT(p, S_SCX, code);
1356 * Release the MP lock if we had to get it
1358 KASSERT(curthread->td_mpcount == 1, ("badmpcount syscall from %p", (void *)frame.tf_eip));
1364 * sendsys2 - MP aware system message request C handler
1367 sendsys2(struct trapframe frame)
1369 struct globaldata *gd;
1370 struct thread *td = curthread;
1371 struct proc *p = td->td_proc;
1372 register_t orig_tf_eflags;
1373 struct sysent *callp;
1374 union sysunion *sysun;
1384 if (ISPL(frame.tf_cs) != SEL_UPL) {
1392 KASSERT(curthread->td_mpcount == 0, ("badmpcount syscall from %p", (void *)frame.tf_eip));
1396 * access non-atomic field from critical section. p_sticks is
1397 * updated by the clock interrupt. Also use this opportunity
1398 * to lazy-raise our LWKT priority.
1401 crit_enter_quick(td);
1402 sticks = curthread->td_sticks;
1403 crit_exit_quick(td);
1405 p->p_md.md_regs = &frame;
1406 orig_tf_eflags = frame.tf_eflags;
1410 * Handle the waitport/waitmsg/checkport/checkmsg case
1412 * YYY MOVE THIS TO INT 0x82! We don't really need to combine it
1415 if ((msgsize = frame.tf_edx) <= 0) {
1417 printf("waitmsg/checkmsg not yet supported: %08x\n",
1423 printf("waitport/checkport only the default port is supported at the moment\n");
1430 * Wait on port for message
1432 sysun = lwkt_getport(&td->td_msgport);
1437 * Test port for message
1439 sysun = lwkt_getport(&td->td_msgport);
1447 umsg = sysun->lmsg.opaque.ms_umsg;
1448 frame.tf_eax = (register_t)umsg;
1449 if (sysun->sysmsg_copyout)
1450 sysun->sysmsg_copyout(sysun);
1451 atomic_add_int_nonlocked(&td->td_msgport.mp_refs, -1);
1452 sysun->nosys.usrmsg.umsg.u.ms_fds[0] = sysun->lmsg.u.ms_fds[0];
1453 sysun->nosys.usrmsg.umsg.u.ms_fds[1] = sysun->lmsg.u.ms_fds[1];
1454 sysun->nosys.usrmsg.umsg.ms_error = sysun->lmsg.ms_error;
1455 error = sysun->lmsg.ms_error;
1456 result = sysun->lmsg.u.ms_fds[0]; /* for ktrace */
1457 if (error != 0 || code != SYS_execve) {
1459 &sysun->nosys.usrmsg.umsg.ms_copyout_start,
1460 &umsg->ms_copyout_start,
1463 crit_enter_quick(td);
1464 sysun->lmsg.opaque.ms_sysunnext = gd->gd_freesysun;
1465 gd->gd_freesysun = sysun;
1466 crit_exit_quick(td);
1477 * Extract the system call message. If msgsize is zero we are
1478 * blocking on a message and/or message port. If msgsize is -1
1479 * we are testing a message for completion or a message port for
1482 * The userland system call message size includes the size of the
1483 * userland lwkt_msg plus arguments. We load it into the userland
1484 * portion of our sysunion structure then we initialize the kerneland
1491 if (msgsize < sizeof(struct lwkt_msg) ||
1492 msgsize > sizeof(union sysunion) - sizeof(struct sysmsg)
1499 * Obtain a sysun from our per-cpu cache or allocate a new one. Use
1500 * the opaque field to store the original (user) message pointer.
1501 * A critical section is necessary to interlock against interrupts
1502 * returning system messages to the thread cache.
1505 crit_enter_quick(td);
1506 if ((sysun = gd->gd_freesysun) != NULL) {
1507 gd->gd_freesysun = sysun->lmsg.opaque.ms_sysunnext;
1508 crit_exit_quick(td);
1510 crit_exit_quick(td);
1511 sysun = malloc(sizeof(union sysunion), M_SYSMSG, M_WAITOK);
1513 atomic_add_int_nonlocked(&td->td_msgport.mp_refs, 1);
1516 * Copy the user request into the kernel copy of the user request.
1518 umsg = (void *)frame.tf_ecx;
1519 error = copyin(umsg, &sysun->nosys.usrmsg, msgsize);
1522 if ((sysun->nosys.usrmsg.umsg.ms_flags & MSGF_ASYNC) &&
1523 (error = suser(td)) != 0
1529 * Initialize the kernel message from the copied-in data and
1530 * pull in appropriate flags from the userland message.
1532 lwkt_initmsg_rp(&sysun->lmsg, &td->td_msgport,
1533 sysun->nosys.usrmsg.umsg.ms_cmd);
1534 sysun->sysmsg_copyout = NULL;
1535 sysun->lmsg.opaque.ms_umsg = umsg;
1536 sysun->lmsg.ms_flags |= sysun->nosys.usrmsg.umsg.ms_flags & MSGF_ASYNC;
1539 * Extract the system call number, lookup the system call, and
1540 * set the default return value.
1542 code = (u_int)sysun->lmsg.ms_cmd;
1543 if (code >= p->p_sysent->sv_size) {
1548 callp = &p->p_sysent->sv_table[code];
1550 narg = (msgsize - sizeof(struct lwkt_msg)) / sizeof(register_t);
1553 if (KTRPOINT(td, KTR_SYSCALL)) {
1554 ktrsyscall(p->p_tracep, code, narg, (void *)(&sysun->nosys.usrmsg + 1));
1557 sysun->lmsg.u.ms_fds[0] = 0;
1558 sysun->lmsg.u.ms_fds[1] = 0;
1560 STOPEVENT(p, S_SCE, narg); /* MP aware */
1563 * Make the system call. An error code is always returned, results
1564 * are copied back via ms_result32 or ms_result64. YYY temporary
1565 * stage copy p_retval[] into ms_result32/64
1567 * NOTE! XXX if this is a child returning from a fork curproc
1568 * might be different. YYY huh? a child returning from a fork
1569 * should never 'return' from this call, it should go right to the
1570 * fork_trampoline function.
1572 error = (*callp->sy_call)(sysun);
1573 gd = td->td_gd; /* RELOAD, might have switched cpus */
1577 * If a synchronous return copy p_retval to ms_result64 and return
1578 * the sysmsg to the free pool.
1580 * YYY Don't writeback message if execve() YYY
1582 if (error != EASYNC) {
1583 atomic_add_int_nonlocked(&td->td_msgport.mp_refs, -1);
1584 sysun->nosys.usrmsg.umsg.u.ms_fds[0] = sysun->lmsg.u.ms_fds[0];
1585 sysun->nosys.usrmsg.umsg.u.ms_fds[1] = sysun->lmsg.u.ms_fds[1];
1586 result = sysun->nosys.usrmsg.umsg.u.ms_fds[0]; /* for ktrace */
1587 if (error != 0 || code != SYS_execve) {
1589 error2 = copyout(&sysun->nosys.usrmsg.umsg.ms_copyout_start,
1590 &umsg->ms_copyout_start,
1595 crit_enter_quick(td);
1596 sysun->lmsg.opaque.ms_sysunnext = gd->gd_freesysun;
1597 gd->gd_freesysun = sysun;
1598 crit_exit_quick(td);
1601 frame.tf_eax = error;
1605 * Traced syscall. trapsignal() is not MP aware.
1607 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1608 frame.tf_eflags &= ~PSL_T;
1609 trapsignal(p, SIGTRAP, 0);
1613 * Handle reschedule and other end-of-syscall issues
1615 userret(p, &frame, sticks);
1618 if (KTRPOINT(td, KTR_SYSRET)) {
1619 ktrsysret(p->p_tracep, code, error, result);
1624 * This works because errno is findable through the
1625 * register set. If we ever support an emulation where this
1626 * is not the case, this code will need to be revisited.
1628 STOPEVENT(p, S_SCX, code);
1633 * Release the MP lock if we had to get it
1635 KASSERT(curthread->td_mpcount == 1, ("badmpcount syscall from %p", (void *)frame.tf_eip));
1641 * Simplified back end of syscall(), used when returning from fork()
1642 * directly into user mode. MP lock is held on entry and should be
1643 * released on return. This code will return back into the fork
1644 * trampoline code which then runs doreti.
1647 fork_return(p, frame)
1649 struct trapframe frame;
1651 frame.tf_eax = 0; /* Child returns zero */
1652 frame.tf_eflags &= ~PSL_C; /* success */
1655 userret(p, &frame, 0);
1657 if (KTRPOINT(p->p_thread, KTR_SYSRET))
1658 ktrsysret(p->p_tracep, SYS_fork, 0, 0);
1660 p->p_flag |= P_PASSIVE_ACQ;
1662 p->p_flag &= ~P_PASSIVE_ACQ;
1664 KKASSERT(curthread->td_mpcount == 1);