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 1.115 2008/09/09 04:06:17 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/kerneldump.h>
61 #include <sys/resourcevar.h>
62 #include <sys/signalvar.h>
63 #include <sys/signal2.h>
64 #include <sys/syscall.h>
65 #include <sys/sysctl.h>
66 #include <sys/sysent.h>
68 #include <sys/vmmeter.h>
69 #include <sys/malloc.h>
71 #include <sys/ktrace.h>
74 #include <sys/upcall.h>
75 #include <sys/vkernel.h>
76 #include <sys/sysproto.h>
77 #include <sys/sysunion.h>
80 #include <vm/vm_param.h>
83 #include <vm/vm_kern.h>
84 #include <vm/vm_map.h>
85 #include <vm/vm_page.h>
86 #include <vm/vm_extern.h>
88 #include <machine/cpu.h>
89 #include <machine/md_var.h>
90 #include <machine/pcb.h>
91 #include <machine/smp.h>
92 #include <machine/tss.h>
93 #include <machine/specialreg.h>
94 #include <machine/globaldata.h>
95 #include <machine/intr_machdep.h>
97 #include <machine_base/isa/isa_intr.h>
100 #include <sys/syslog.h>
101 #include <machine/clock.h>
104 #include <machine/vm86.h>
108 #include <sys/msgport2.h>
109 #include <sys/thread2.h>
110 #include <sys/mplock2.h>
114 #define MAKEMPSAFE(have_mplock) \
115 if (have_mplock == 0) { \
122 #define MAKEMPSAFE(have_mplock)
126 int (*pmath_emulate) (struct trapframe *);
128 extern void trap (struct trapframe *frame);
129 extern void syscall2 (struct trapframe *frame);
131 static int trap_pfault (struct trapframe *, int, vm_offset_t);
132 static void trap_fatal (struct trapframe *, vm_offset_t);
133 void dblfault_handler (void);
135 extern inthand_t IDTVEC(syscall);
137 #define MAX_TRAP_MSG 28
138 static char *trap_msg[] = {
140 "privileged instruction fault", /* 1 T_PRIVINFLT */
142 "breakpoint instruction fault", /* 3 T_BPTFLT */
145 "arithmetic trap", /* 6 T_ARITHTRAP */
146 "system forced exception", /* 7 T_ASTFLT */
148 "general protection fault", /* 9 T_PROTFLT */
149 "trace trap", /* 10 T_TRCTRAP */
151 "page fault", /* 12 T_PAGEFLT */
153 "alignment fault", /* 14 T_ALIGNFLT */
157 "integer divide fault", /* 18 T_DIVIDE */
158 "non-maskable interrupt trap", /* 19 T_NMI */
159 "overflow trap", /* 20 T_OFLOW */
160 "FPU bounds check fault", /* 21 T_BOUND */
161 "FPU device not available", /* 22 T_DNA */
162 "double fault", /* 23 T_DOUBLEFLT */
163 "FPU operand fetch fault", /* 24 T_FPOPFLT */
164 "invalid TSS fault", /* 25 T_TSSFLT */
165 "segment not present fault", /* 26 T_SEGNPFLT */
166 "stack fault", /* 27 T_STKFLT */
167 "machine check trap", /* 28 T_MCHK */
170 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
171 extern int has_f00f_bug;
175 static int ddb_on_nmi = 1;
176 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
177 &ddb_on_nmi, 0, "Go to DDB on NMI");
179 static int panic_on_nmi = 1;
180 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
181 &panic_on_nmi, 0, "Panic on NMI");
182 static int fast_release;
183 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
184 &fast_release, 0, "Passive Release was optimal");
185 static int slow_release;
186 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
187 &slow_release, 0, "Passive Release was nonoptimal");
189 MALLOC_DEFINE(M_SYSMSG, "sysmsg", "sysmsg structure");
190 extern int max_sysmsg;
193 * Passively intercepts the thread switch function to increase the thread
194 * priority from a user priority to a kernel priority, reducing
195 * syscall and trap overhead for the case where no switch occurs.
197 * Synchronizes td_ucred with p_ucred. This is used by system calls,
198 * signal handling, faults, AST traps, and anything else that enters the
199 * kernel from userland and provides the kernel with a stable read-only
200 * copy of the process ucred.
203 userenter(struct thread *curtd, struct proc *curp)
208 curtd->td_release = lwkt_passive_release;
210 if (curtd->td_ucred != curp->p_ucred) {
211 ncred = crhold(curp->p_ucred);
212 ocred = curtd->td_ucred;
213 curtd->td_ucred = ncred;
221 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
222 * must be completed before we can return to or try to return to userland.
224 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
225 * arithmatic on the delta calculation so the absolute tick values are
226 * truncated to an integer.
229 userret(struct lwp *lp, struct trapframe *frame, int sticks)
231 struct proc *p = lp->lwp_proc;
235 if (p->p_userret != NULL) {
242 * Charge system time if profiling. Note: times are in microseconds.
243 * This may do a copyout and block, so do it first even though it
244 * means some system time will be charged as user time.
246 if (p->p_flag & P_PROFIL) {
247 addupc_task(p, frame->tf_eip,
248 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
253 * If the jungle wants us dead, so be it.
255 if (lp->lwp_flag & LWP_WEXIT) {
256 lwkt_gettoken(&p->p_token);
258 lwkt_reltoken(&p->p_token); /* NOT REACHED */
262 * Block here if we are in a stopped state.
264 if (p->p_stat == SSTOP || dump_stop_usertds) {
272 * Post any pending upcalls. If running a virtual kernel be sure
273 * to restore the virtual kernel's vmspace before posting the upcall.
275 if (p->p_flag & P_UPCALLPEND) {
276 p->p_flag &= ~P_UPCALLPEND;
284 * Post any pending signals. If running a virtual kernel be sure
285 * to restore the virtual kernel's vmspace before posting the signal.
287 * WARNING! postsig() can exit and not return.
289 if ((sig = CURSIG_TRACE(lp)) != 0) {
297 * block here if we are swapped out, but still process signals
298 * (such as SIGKILL). proc0 (the swapin scheduler) is already
299 * aware of our situation, we do not have to wake it up.
301 if (p->p_flag & P_SWAPPEDOUT) {
303 p->p_flag |= P_SWAPWAIT;
305 if (p->p_flag & P_SWAPWAIT)
306 tsleep(p, PCATCH, "SWOUT", 0);
307 p->p_flag &= ~P_SWAPWAIT;
313 * Make sure postsig() handled request to restore old signal mask after
314 * running signal handler.
316 KKASSERT((lp->lwp_flag & LWP_OLDMASK) == 0);
320 * Cleanup from userenter and any passive release that might have occured.
321 * We must reclaim the current-process designation before we can return
322 * to usermode. We also handle both LWKT and USER reschedule requests.
325 userexit(struct lwp *lp)
327 struct thread *td = lp->lwp_thread;
328 /* globaldata_t gd = td->td_gd; */
331 * Handle stop requests at kernel priority. Any requests queued
332 * after this loop will generate another AST.
334 while (lp->lwp_proc->p_stat == SSTOP) {
341 * Reduce our priority in preparation for a return to userland. If
342 * our passive release function was still in place, our priority was
343 * never raised and does not need to be reduced.
345 lwkt_passive_recover(td);
348 * Become the current user scheduled process if we aren't already,
349 * and deal with reschedule requests and other factors.
351 lp->lwp_proc->p_usched->acquire_curproc(lp);
352 /* WARNING: we may have migrated cpu's */
353 /* gd = td->td_gd; */
356 #if !defined(KTR_KERNENTRY)
357 #define KTR_KERNENTRY KTR_ALL
359 KTR_INFO_MASTER(kernentry);
360 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, "pid=%d, tid=%d, trapno=%d, eva=%p",
361 sizeof(int) + sizeof(int) + sizeof(int) + sizeof(vm_offset_t));
362 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "pid=%d, tid=%d",
363 sizeof(int) + sizeof(int));
364 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "pid=%d, tid=%d, call=%d",
365 sizeof(int) + sizeof(int) + sizeof(int));
366 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "pid=%d, tid=%d, err=%d",
367 sizeof(int) + sizeof(int) + sizeof(int));
368 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "pid=%d, tid=%d",
369 sizeof(int) + sizeof(int));
372 * Exception, fault, and trap interface to the kernel.
373 * This common code is called from assembly language IDT gate entry
374 * routines that prepare a suitable stack frame, and restore this
375 * frame after the exception has been processed.
377 * This function is also called from doreti in an interlock to handle ASTs.
378 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
380 * NOTE! We have to retrieve the fault address prior to obtaining the
381 * MP lock because get_mplock() may switch out. YYY cr2 really ought
382 * to be retrieved by the assembly code, not here.
384 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
385 * if an attempt is made to switch from a fast interrupt or IPI. This is
386 * necessary to properly take fatal kernel traps on SMP machines if
387 * get_mplock() has to block.
391 trap(struct trapframe *frame)
393 struct globaldata *gd = mycpu;
394 struct thread *td = gd->gd_curthread;
395 struct lwp *lp = td->td_lwp;
398 int i = 0, ucode = 0, type, code;
403 int crit_count = td->td_critcount;
404 lwkt_tokref_t curstop = td->td_toks_stop;
411 * We need to allow T_DNA faults when the debugger is active since
412 * some dumping paths do large bcopy() which use the floating
413 * point registers for faster copying.
415 if (db_active && frame->tf_trapno != T_DNA) {
416 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
417 ++gd->gd_trap_nesting_level;
418 MAKEMPSAFE(have_mplock);
419 trap_fatal(frame, eva);
420 --gd->gd_trap_nesting_level;
426 ++gd->gd_trap_nesting_level;
427 if (frame->tf_trapno == T_PAGEFLT) {
429 * For some Cyrix CPUs, %cr2 is clobbered by interrupts.
430 * This problem is worked around by using an interrupt
431 * gate for the pagefault handler. We are finally ready
432 * to read %cr2 and then must reenable interrupts.
434 * XXX this should be in the switch statement, but the
435 * NO_FOOF_HACK and VM86 goto and ifdefs obfuscate the
436 * flow of control too much for this to be obviously
443 --gd->gd_trap_nesting_level;
445 if (!(frame->tf_eflags & PSL_I)) {
447 * Buggy application or kernel code has disabled interrupts
448 * and then trapped. Enabling interrupts now is wrong, but
449 * it is better than running with interrupts disabled until
450 * they are accidentally enabled later.
452 type = frame->tf_trapno;
453 if (ISPL(frame->tf_cs)==SEL_UPL || (frame->tf_eflags & PSL_VM)) {
454 MAKEMPSAFE(have_mplock);
456 "pid %ld (%s): trap %d with interrupts disabled\n",
457 (long)curproc->p_pid, curproc->p_comm, type);
458 } else if (type != T_BPTFLT && type != T_TRCTRAP) {
460 * XXX not quite right, since this may be for a
461 * multiple fault in user mode.
463 MAKEMPSAFE(have_mplock);
464 kprintf("kernel trap %d with interrupts disabled\n",
470 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
473 type = frame->tf_trapno;
474 code = frame->tf_err;
477 if (frame->tf_eflags & PSL_VM &&
478 (type == T_PROTFLT || type == T_STKFLT)) {
480 KKASSERT(get_mplock_count(curthread) > 0);
482 i = vm86_emulate((struct vm86frame *)frame);
484 KKASSERT(get_mplock_count(curthread) > 0);
488 * returns to original process
491 vm86_trap((struct vm86frame *)frame,
494 vm86_trap((struct vm86frame *)frame, 0);
496 KKASSERT(0); /* NOT REACHED */
502 * these traps want either a process context, or
503 * assume a normal userspace trap.
507 trap_fatal(frame, eva);
510 type = T_BPTFLT; /* kernel breakpoint */
513 goto kernel_trap; /* normal kernel trap handling */
516 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
519 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid,
520 frame->tf_trapno, eva);
524 sticks = (int)td->td_sticks;
525 lp->lwp_md.md_regs = frame;
528 case T_PRIVINFLT: /* privileged instruction fault */
533 case T_BPTFLT: /* bpt instruction fault */
534 case T_TRCTRAP: /* trace trap */
535 frame->tf_eflags &= ~PSL_T;
540 case T_ARITHTRAP: /* arithmetic trap */
545 case T_ASTFLT: /* Allow process switch */
546 mycpu->gd_cnt.v_soft++;
547 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
548 atomic_clear_int(&mycpu->gd_reqflags,
550 addupc_task(p, p->p_prof.pr_addr,
556 * The following two traps can happen in
557 * vm86 mode, and, if so, we want to handle
560 case T_PROTFLT: /* general protection fault */
561 case T_STKFLT: /* stack fault */
562 if (frame->tf_eflags & PSL_VM) {
563 i = vm86_emulate((struct vm86frame *)frame);
569 ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR;
571 case T_SEGNPFLT: /* segment not present fault */
575 case T_TSSFLT: /* invalid TSS fault */
576 case T_DOUBLEFLT: /* double fault */
581 ucode = code + BUS_SEGM_FAULT ; /* XXX: ???*/
587 case T_PAGEFLT: /* page fault */
588 i = trap_pfault(frame, TRUE, eva);
591 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
603 ucode = BUS_ADRERR; /* XXX */
606 case T_DIVIDE: /* integer divide fault */
613 MAKEMPSAFE(have_mplock);
615 goto handle_powerfail;
616 #else /* !POWERFAIL_NMI */
617 /* machine/parity/power fail/"kitchen sink" faults */
618 if (isa_nmi(code) == 0) {
621 * NMI can be hooked up to a pushbutton
625 kprintf ("NMI ... going to debugger\n");
626 kdb_trap (type, 0, frame);
630 } else if (panic_on_nmi)
631 panic("NMI indicates hardware failure");
633 #endif /* POWERFAIL_NMI */
634 #endif /* NISA > 0 */
636 case T_OFLOW: /* integer overflow fault */
641 case T_BOUND: /* bounds check fault */
648 * Virtual kernel intercept - pass the DNA exception
649 * to the virtual kernel if it asked to handle it.
650 * This occurs when the virtual kernel is holding
651 * onto the FP context for a different emulated
652 * process then the one currently running.
654 * We must still call npxdna() since we may have
655 * saved FP state that the virtual kernel needs
656 * to hand over to a different emulated process.
658 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
659 (td->td_pcb->pcb_flags & FP_VIRTFP)
667 * The kernel may have switched out the FP unit's
668 * state, causing the user process to take a fault
669 * when it tries to use the FP unit. Restore the
675 if (!pmath_emulate) {
677 ucode = FPE_FPU_NP_TRAP;
680 i = (*pmath_emulate)(frame);
682 if (!(frame->tf_eflags & PSL_T))
684 frame->tf_eflags &= ~PSL_T;
687 /* else ucode = emulator_only_knows() XXX */
690 case T_FPOPFLT: /* FPU operand fetch fault */
695 case T_XMMFLT: /* SIMD floating-point exception */
705 case T_PAGEFLT: /* page fault */
706 trap_pfault(frame, FALSE, eva);
712 * The kernel may be using npx for copying or other
720 case T_PROTFLT: /* general protection fault */
721 case T_SEGNPFLT: /* segment not present fault */
723 * Invalid segment selectors and out of bounds
724 * %eip's and %esp's can be set up in user mode.
725 * This causes a fault in kernel mode when the
726 * kernel tries to return to user mode. We want
727 * to get this fault so that we can fix the
728 * problem here and not have to check all the
729 * selectors and pointers when the user changes
732 #define MAYBE_DORETI_FAULT(where, whereto) \
734 if (frame->tf_eip == (int)where) { \
735 frame->tf_eip = (int)whereto; \
739 if (mycpu->gd_intr_nesting_level == 0) {
741 * Invalid %fs's and %gs's can be created using
742 * procfs or PT_SETREGS or by invalidating the
743 * underlying LDT entry. This causes a fault
744 * in kernel mode when the kernel attempts to
745 * switch contexts. Lose the bad context
746 * (XXX) so that we can continue, and generate
749 MAYBE_DORETI_FAULT(doreti_iret,
751 MAYBE_DORETI_FAULT(doreti_popl_ds,
752 doreti_popl_ds_fault);
753 MAYBE_DORETI_FAULT(doreti_popl_es,
754 doreti_popl_es_fault);
755 MAYBE_DORETI_FAULT(doreti_popl_fs,
756 doreti_popl_fs_fault);
757 MAYBE_DORETI_FAULT(doreti_popl_gs,
758 doreti_popl_gs_fault);
759 if (td->td_pcb->pcb_onfault) {
761 (register_t)td->td_pcb->pcb_onfault;
769 * PSL_NT can be set in user mode and isn't cleared
770 * automatically when the kernel is entered. This
771 * causes a TSS fault when the kernel attempts to
772 * `iret' because the TSS link is uninitialized. We
773 * want to get this fault so that we can fix the
774 * problem here and not every time the kernel is
777 if (frame->tf_eflags & PSL_NT) {
778 frame->tf_eflags &= ~PSL_NT;
783 case T_TRCTRAP: /* trace trap */
784 if (frame->tf_eip == (int)IDTVEC(syscall)) {
786 * We've just entered system mode via the
787 * syscall lcall. Continue single stepping
788 * silently until the syscall handler has
793 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
795 * The syscall handler has now saved the
796 * flags. Stop single stepping it.
798 frame->tf_eflags &= ~PSL_T;
802 * Ignore debug register trace traps due to
803 * accesses in the user's address space, which
804 * can happen under several conditions such as
805 * if a user sets a watchpoint on a buffer and
806 * then passes that buffer to a system call.
807 * We still want to get TRCTRAPS for addresses
808 * in kernel space because that is useful when
809 * debugging the kernel.
811 if (user_dbreg_trap()) {
813 * Reset breakpoint bits because the
816 load_dr6(rdr6() & 0xfffffff0);
820 * Fall through (TRCTRAP kernel mode, kernel address)
824 * If DDB is enabled, let it handle the debugger trap.
825 * Otherwise, debugger traps "can't happen".
829 MAKEMPSAFE(have_mplock);
830 if (kdb_trap (type, 0, frame))
837 MAKEMPSAFE(have_mplock);
840 # define TIMER_FREQ 1193182
844 static unsigned lastalert = 0;
846 if(time_second - lastalert > 10)
848 log(LOG_WARNING, "NMI: power fail\n");
849 sysbeep(TIMER_FREQ/880, hz);
850 lastalert = time_second;
855 #else /* !POWERFAIL_NMI */
856 /* machine/parity/power fail/"kitchen sink" faults */
857 if (isa_nmi(code) == 0) {
860 * NMI can be hooked up to a pushbutton
864 kprintf ("NMI ... going to debugger\n");
865 kdb_trap (type, 0, frame);
869 } else if (panic_on_nmi == 0)
872 #endif /* POWERFAIL_NMI */
873 #endif /* NISA > 0 */
876 MAKEMPSAFE(have_mplock);
877 trap_fatal(frame, eva);
882 * Virtual kernel intercept - if the fault is directly related to a
883 * VM context managed by a virtual kernel then let the virtual kernel
886 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
887 vkernel_trap(lp, frame);
892 * Translate fault for emulators (e.g. Linux)
894 if (*p->p_sysent->sv_transtrap)
895 i = (*p->p_sysent->sv_transtrap)(i, type);
897 MAKEMPSAFE(have_mplock);
898 trapsignal(lp, i, ucode);
901 if (type <= MAX_TRAP_MSG) {
902 uprintf("fatal process exception: %s",
904 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
905 uprintf(", fault VA = 0x%lx", (u_long)eva);
911 userret(lp, frame, sticks);
918 if (p != NULL && lp != NULL)
919 KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid);
921 KASSERT(crit_count == td->td_critcount,
922 ("trap: critical section count mismatch! %d/%d",
923 crit_count, td->td_pri));
924 KASSERT(curstop == td->td_toks_stop,
925 ("trap: extra tokens held after trap! %zd/%zd",
926 curstop - &td->td_toks_base,
927 td->td_toks_stop - &td->td_toks_base));
932 trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
935 struct vmspace *vm = NULL;
940 thread_t td = curthread;
941 struct lwp *lp = td->td_lwp;
943 va = trunc_page(eva);
944 if (va >= KERNBASE) {
946 * Don't allow user-mode faults in kernel address space.
947 * An exception: if the faulting address is the invalid
948 * instruction entry in the IDT, then the Intel Pentium
949 * F00F bug workaround was triggered, and we need to
950 * treat it is as an illegal instruction, and not a page
953 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
954 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) {
955 frame->tf_trapno = T_PRIVINFLT;
965 * This is a fault on non-kernel virtual memory.
966 * vm is initialized above to NULL. If curproc is NULL
967 * or curproc->p_vmspace is NULL the fault is fatal.
970 vm = lp->lwp_vmspace;
978 if (frame->tf_err & PGEX_W)
979 ftype = VM_PROT_WRITE;
981 ftype = VM_PROT_READ;
983 if (map != &kernel_map) {
985 * Keep swapout from messing with us during this
995 fault_flags |= VM_FAULT_BURST;
996 if (ftype & VM_PROT_WRITE)
997 fault_flags |= VM_FAULT_DIRTY;
999 fault_flags |= VM_FAULT_NORMAL;
1000 rv = vm_fault(map, va, ftype, fault_flags);
1001 PRELE(lp->lwp_proc);
1004 * Don't have to worry about process locking or stacks
1007 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
1010 if (rv == KERN_SUCCESS)
1014 if (td->td_gd->gd_intr_nesting_level == 0 &&
1015 td->td_pcb->pcb_onfault) {
1016 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
1019 trap_fatal(frame, eva);
1023 /* kludge to pass faulting virtual address to sendsig */
1024 frame->tf_xflags = frame->tf_err;
1025 frame->tf_err = eva;
1027 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
1031 trap_fatal(struct trapframe *frame, vm_offset_t eva)
1033 int code, type, ss, esp;
1034 struct soft_segment_descriptor softseg;
1036 code = frame->tf_err;
1037 type = frame->tf_trapno;
1038 sdtossd(&gdt[mycpu->gd_cpuid * NGDT + IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
1040 if (type <= MAX_TRAP_MSG)
1041 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
1042 type, trap_msg[type],
1043 frame->tf_eflags & PSL_VM ? "vm86" :
1044 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
1046 /* three separate prints in case of a trap on an unmapped page */
1047 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
1048 kprintf("lapic.id = %08x\n", lapic.id);
1050 if (type == T_PAGEFLT) {
1051 kprintf("fault virtual address = %p\n", (void *)eva);
1052 kprintf("fault code = %s %s, %s\n",
1053 code & PGEX_U ? "user" : "supervisor",
1054 code & PGEX_W ? "write" : "read",
1055 code & PGEX_P ? "protection violation" : "page not present");
1057 kprintf("instruction pointer = 0x%x:0x%x\n",
1058 frame->tf_cs & 0xffff, frame->tf_eip);
1059 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
1060 ss = frame->tf_ss & 0xffff;
1061 esp = frame->tf_esp;
1063 ss = GSEL(GDATA_SEL, SEL_KPL);
1064 esp = (int)&frame->tf_esp;
1066 kprintf("stack pointer = 0x%x:0x%x\n", ss, esp);
1067 kprintf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
1068 kprintf("code segment = base 0x%x, limit 0x%x, type 0x%x\n",
1069 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
1070 kprintf(" = DPL %d, pres %d, def32 %d, gran %d\n",
1071 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
1073 kprintf("processor eflags = ");
1074 if (frame->tf_eflags & PSL_T)
1075 kprintf("trace trap, ");
1076 if (frame->tf_eflags & PSL_I)
1077 kprintf("interrupt enabled, ");
1078 if (frame->tf_eflags & PSL_NT)
1079 kprintf("nested task, ");
1080 if (frame->tf_eflags & PSL_RF)
1081 kprintf("resume, ");
1082 if (frame->tf_eflags & PSL_VM)
1084 kprintf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
1085 kprintf("current process = ");
1087 kprintf("%lu (%s)\n",
1088 (u_long)curproc->p_pid, curproc->p_comm ?
1089 curproc->p_comm : "");
1093 kprintf("current thread = pri %d ", curthread->td_pri);
1094 if (curthread->td_critcount)
1100 * we probably SHOULD have stopped the other CPUs before now!
1101 * another CPU COULD have been touching cpl at this moment...
1103 kprintf(" <- SMP: XXX");
1112 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1115 kprintf("trap number = %d\n", type);
1116 if (type <= MAX_TRAP_MSG)
1117 panic("%s", trap_msg[type]);
1119 panic("unknown/reserved trap");
1123 * Double fault handler. Called when a fault occurs while writing
1124 * a frame for a trap/exception onto the stack. This usually occurs
1125 * when the stack overflows (such is the case with infinite recursion,
1128 * XXX Note that the current PTD gets replaced by IdlePTD when the
1129 * task switch occurs. This means that the stack that was active at
1130 * the time of the double fault is not available at <kstack> unless
1131 * the machine was idle when the double fault occurred. The downside
1132 * of this is that "trace <ebp>" in ddb won't work.
1136 in_kstack_guard(register_t rptr)
1138 thread_t td = curthread;
1140 if ((char *)rptr >= td->td_kstack &&
1141 (char *)rptr < td->td_kstack + PAGE_SIZE) {
1148 dblfault_handler(void)
1150 struct mdglobaldata *gd = mdcpu;
1152 if (in_kstack_guard(gd->gd_common_tss.tss_esp) ||
1153 in_kstack_guard(gd->gd_common_tss.tss_ebp)) {
1154 kprintf("DOUBLE FAULT - KERNEL STACK GUARD HIT!\n");
1156 kprintf("DOUBLE FAULT:\n");
1158 kprintf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1159 kprintf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1160 kprintf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
1162 /* three separate prints in case of a trap on an unmapped page */
1163 kprintf("cpuid = %d; ", gd->mi.gd_cpuid);
1164 kprintf("lapic.id = %08x\n", lapic.id);
1166 panic("double fault");
1170 * syscall2 - MP aware system call request C handler
1172 * A system call is essentially treated as a trap. The MP lock is not
1173 * held on entry or return. We are responsible for handling ASTs
1174 * (e.g. a task switch) prior to return.
1179 syscall2(struct trapframe *frame)
1181 struct thread *td = curthread;
1182 struct proc *p = td->td_proc;
1183 struct lwp *lp = td->td_lwp;
1185 struct sysent *callp;
1186 register_t orig_tf_eflags;
1191 int crit_count = td->td_critcount;
1194 int have_mplock = 0;
1197 union sysunion args;
1200 if (ISPL(frame->tf_cs) != SEL_UPL) {
1207 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid,
1210 userenter(td, p); /* lazy raise our priority */
1215 sticks = (int)td->td_sticks;
1216 orig_tf_eflags = frame->tf_eflags;
1219 * Virtual kernel intercept - if a VM context managed by a virtual
1220 * kernel issues a system call the virtual kernel handles it, not us.
1221 * Restore the virtual kernel context and return from its system
1222 * call. The current frame is copied out to the virtual kernel.
1224 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1225 vkernel_trap(lp, frame);
1226 error = EJUSTRETURN;
1232 * Get the system call parameters and account for time
1234 lp->lwp_md.md_regs = frame;
1235 params = (caddr_t)frame->tf_esp + sizeof(int);
1236 code = frame->tf_eax;
1238 if (p->p_sysent->sv_prepsyscall) {
1239 (*p->p_sysent->sv_prepsyscall)(
1240 frame, (int *)(&args.nosys.sysmsg + 1),
1244 * Need to check if this is a 32 bit or 64 bit syscall.
1245 * fuword is MP aware.
1247 if (code == SYS_syscall) {
1249 * Code is first argument, followed by actual args.
1251 code = fuword(params);
1252 params += sizeof(int);
1253 } else if (code == SYS___syscall) {
1255 * Like syscall, but code is a quad, so as to maintain
1256 * quad alignment for the rest of the arguments.
1258 code = fuword(params);
1259 params += sizeof(quad_t);
1263 code &= p->p_sysent->sv_mask;
1265 if (code >= p->p_sysent->sv_size)
1266 callp = &p->p_sysent->sv_table[0];
1268 callp = &p->p_sysent->sv_table[code];
1270 narg = callp->sy_narg & SYF_ARGMASK;
1273 if (p->p_sysent->sv_name[0] == 'L')
1274 kprintf("Linux syscall, code = %d\n", code);
1278 * copyin is MP aware, but the tracing code is not
1280 if (narg && params) {
1281 error = copyin(params, (caddr_t)(&args.nosys.sysmsg + 1),
1282 narg * sizeof(register_t));
1285 if (KTRPOINT(td, KTR_SYSCALL)) {
1286 MAKEMPSAFE(have_mplock);
1288 ktrsyscall(lp, code, narg,
1289 (void *)(&args.nosys.sysmsg + 1));
1297 if (KTRPOINT(td, KTR_SYSCALL)) {
1298 MAKEMPSAFE(have_mplock);
1299 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1304 * For traditional syscall code edx is left untouched when 32 bit
1305 * results are returned. Since edx is loaded from fds[1] when the
1306 * system call returns we pre-set it here.
1308 args.sysmsg_fds[0] = 0;
1309 args.sysmsg_fds[1] = frame->tf_edx;
1312 * The syscall might manipulate the trap frame. If it does it
1313 * will probably return EJUSTRETURN.
1315 args.sysmsg_frame = frame;
1317 STOPEVENT(p, S_SCE, narg); /* MP aware */
1320 * NOTE: All system calls run MPSAFE now. The system call itself
1321 * is responsible for getting the MP lock.
1323 error = (*callp->sy_call)(&args);
1327 * MP SAFE (we may or may not have the MP lock at this point)
1332 * Reinitialize proc pointer `p' as it may be different
1333 * if this is a child returning from fork syscall.
1336 lp = curthread->td_lwp;
1337 frame->tf_eax = args.sysmsg_fds[0];
1338 frame->tf_edx = args.sysmsg_fds[1];
1339 frame->tf_eflags &= ~PSL_C;
1343 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1344 * int 0x80 is 2 bytes. We saved this in tf_err.
1346 frame->tf_eip -= frame->tf_err;
1351 panic("Unexpected EASYNC return value (for now)");
1354 if (p->p_sysent->sv_errsize) {
1355 if (error >= p->p_sysent->sv_errsize)
1356 error = -1; /* XXX */
1358 error = p->p_sysent->sv_errtbl[error];
1360 frame->tf_eax = error;
1361 frame->tf_eflags |= PSL_C;
1366 * Traced syscall. trapsignal() is not MP aware.
1368 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1369 MAKEMPSAFE(have_mplock);
1370 frame->tf_eflags &= ~PSL_T;
1371 trapsignal(lp, SIGTRAP, TRAP_TRACE);
1375 * Handle reschedule and other end-of-syscall issues
1377 userret(lp, frame, sticks);
1380 if (KTRPOINT(td, KTR_SYSRET)) {
1381 MAKEMPSAFE(have_mplock);
1382 ktrsysret(lp, code, error, args.sysmsg_result);
1387 * This works because errno is findable through the
1388 * register set. If we ever support an emulation where this
1389 * is not the case, this code will need to be revisited.
1391 STOPEVENT(p, S_SCX, code);
1396 * Release the MP lock if we had to get it
1401 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error);
1403 KASSERT(crit_count == td->td_critcount,
1404 ("syscall: critical section count mismatch! %d/%d",
1405 crit_count, td->td_pri));
1406 KASSERT(&td->td_toks_base == td->td_toks_stop,
1407 ("syscall: extra tokens held after trap! %zd",
1408 td->td_toks_stop - &td->td_toks_base));
1413 * NOTE: MP lock not held at any point.
1416 fork_return(struct lwp *lp, struct trapframe *frame)
1418 frame->tf_eax = 0; /* Child returns zero */
1419 frame->tf_eflags &= ~PSL_C; /* success */
1422 generic_lwp_return(lp, frame);
1423 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1427 * Simplified back end of syscall(), used when returning from fork()
1428 * directly into user mode.
1430 * This code will return back into the fork trampoline code which then
1433 * NOTE: The mplock is not held at any point.
1436 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1438 struct proc *p = lp->lwp_proc;
1441 * Newly forked processes are given a kernel priority. We have to
1442 * adjust the priority to a normal user priority and fake entry
1443 * into the kernel (call userenter()) to install a passive release
1444 * function just in case userret() decides to stop the process. This
1445 * can occur when ^Z races a fork. If we do not install the passive
1446 * release function the current process designation will not be
1447 * released when the thread goes to sleep.
1449 lwkt_setpri_self(TDPRI_USER_NORM);
1450 userenter(lp->lwp_thread, p);
1451 userret(lp, frame, 0);
1453 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1454 ktrsysret(lp, SYS_fork, 0, 0);
1456 p->p_flag |= P_PASSIVE_ACQ;
1458 p->p_flag &= ~P_PASSIVE_ACQ;
1462 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1463 * fault (which is then passed back to the virtual kernel) if an attempt is
1464 * made to use the FP unit.
1466 * XXX this is a fairly big hack.
1469 set_vkernel_fp(struct trapframe *frame)
1471 struct thread *td = curthread;
1473 if (frame->tf_xflags & PGEX_FPFAULT) {
1474 td->td_pcb->pcb_flags |= FP_VIRTFP;
1475 if (mdcpu->gd_npxthread == td)
1478 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1483 * Called from vkernel_trap() to fixup the vkernel's syscall
1484 * frame for vmspace_ctl() return.
1487 cpu_vkernel_trap(struct trapframe *frame, int error)
1489 frame->tf_eax = error;
1491 frame->tf_eflags |= PSL_C;
1493 frame->tf_eflags &= ~PSL_C;