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>
98 #include <machine_base/apic/lapic.h>
101 #include <sys/syslog.h>
102 #include <machine/clock.h>
105 #include <machine/vm86.h>
109 #include <sys/msgport2.h>
110 #include <sys/thread2.h>
111 #include <sys/mplock2.h>
115 #define MAKEMPSAFE(have_mplock) \
116 if (have_mplock == 0) { \
123 #define MAKEMPSAFE(have_mplock)
127 int (*pmath_emulate) (struct trapframe *);
129 extern void trap (struct trapframe *frame);
130 extern void syscall2 (struct trapframe *frame);
132 static int trap_pfault (struct trapframe *, int, vm_offset_t);
133 static void trap_fatal (struct trapframe *, vm_offset_t);
134 void dblfault_handler (void);
136 extern inthand_t IDTVEC(syscall);
138 #define MAX_TRAP_MSG 28
139 static char *trap_msg[] = {
141 "privileged instruction fault", /* 1 T_PRIVINFLT */
143 "breakpoint instruction fault", /* 3 T_BPTFLT */
146 "arithmetic trap", /* 6 T_ARITHTRAP */
147 "system forced exception", /* 7 T_ASTFLT */
149 "general protection fault", /* 9 T_PROTFLT */
150 "trace trap", /* 10 T_TRCTRAP */
152 "page fault", /* 12 T_PAGEFLT */
154 "alignment fault", /* 14 T_ALIGNFLT */
158 "integer divide fault", /* 18 T_DIVIDE */
159 "non-maskable interrupt trap", /* 19 T_NMI */
160 "overflow trap", /* 20 T_OFLOW */
161 "FPU bounds check fault", /* 21 T_BOUND */
162 "FPU device not available", /* 22 T_DNA */
163 "double fault", /* 23 T_DOUBLEFLT */
164 "FPU operand fetch fault", /* 24 T_FPOPFLT */
165 "invalid TSS fault", /* 25 T_TSSFLT */
166 "segment not present fault", /* 26 T_SEGNPFLT */
167 "stack fault", /* 27 T_STKFLT */
168 "machine check trap", /* 28 T_MCHK */
171 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
172 extern int has_f00f_bug;
176 static int ddb_on_nmi = 1;
177 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
178 &ddb_on_nmi, 0, "Go to DDB on NMI");
180 static int panic_on_nmi = 1;
181 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
182 &panic_on_nmi, 0, "Panic on NMI");
183 static int fast_release;
184 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
185 &fast_release, 0, "Passive Release was optimal");
186 static int slow_release;
187 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
188 &slow_release, 0, "Passive Release was nonoptimal");
190 MALLOC_DEFINE(M_SYSMSG, "sysmsg", "sysmsg structure");
191 extern int max_sysmsg;
194 * Passively intercepts the thread switch function to increase the thread
195 * priority from a user priority to a kernel priority, reducing
196 * syscall and trap overhead for the case where no switch occurs.
198 * Synchronizes td_ucred with p_ucred. This is used by system calls,
199 * signal handling, faults, AST traps, and anything else that enters the
200 * kernel from userland and provides the kernel with a stable read-only
201 * copy of the process ucred.
204 userenter(struct thread *curtd, struct proc *curp)
209 curtd->td_release = lwkt_passive_release;
211 if (curtd->td_ucred != curp->p_ucred) {
212 ncred = crhold(curp->p_ucred);
213 ocred = curtd->td_ucred;
214 curtd->td_ucred = ncred;
222 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
223 * must be completed before we can return to or try to return to userland.
225 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
226 * arithmatic on the delta calculation so the absolute tick values are
227 * truncated to an integer.
230 userret(struct lwp *lp, struct trapframe *frame, int sticks)
232 struct proc *p = lp->lwp_proc;
236 if (p->p_userret != NULL) {
243 * Charge system time if profiling. Note: times are in microseconds.
244 * This may do a copyout and block, so do it first even though it
245 * means some system time will be charged as user time.
247 if (p->p_flag & P_PROFIL) {
248 addupc_task(p, frame->tf_eip,
249 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
254 * If the jungle wants us dead, so be it.
256 if (lp->lwp_flag & LWP_WEXIT) {
257 lwkt_gettoken(&p->p_token);
259 lwkt_reltoken(&p->p_token); /* NOT REACHED */
263 * Block here if we are in a stopped state.
265 if (p->p_stat == SSTOP || dump_stop_usertds) {
273 * Post any pending upcalls. If running a virtual kernel be sure
274 * to restore the virtual kernel's vmspace before posting the upcall.
276 if (p->p_flag & P_UPCALLPEND) {
277 p->p_flag &= ~P_UPCALLPEND;
285 * Post any pending signals. If running a virtual kernel be sure
286 * to restore the virtual kernel's vmspace before posting the signal.
288 * WARNING! postsig() can exit and not return.
290 if ((sig = CURSIG_TRACE(lp)) != 0) {
298 * block here if we are swapped out, but still process signals
299 * (such as SIGKILL). proc0 (the swapin scheduler) is already
300 * aware of our situation, we do not have to wake it up.
302 if (p->p_flag & P_SWAPPEDOUT) {
304 p->p_flag |= P_SWAPWAIT;
306 if (p->p_flag & P_SWAPWAIT)
307 tsleep(p, PCATCH, "SWOUT", 0);
308 p->p_flag &= ~P_SWAPWAIT;
314 * Make sure postsig() handled request to restore old signal mask after
315 * running signal handler.
317 KKASSERT((lp->lwp_flag & LWP_OLDMASK) == 0);
321 * Cleanup from userenter and any passive release that might have occured.
322 * We must reclaim the current-process designation before we can return
323 * to usermode. We also handle both LWKT and USER reschedule requests.
326 userexit(struct lwp *lp)
328 struct thread *td = lp->lwp_thread;
329 /* globaldata_t gd = td->td_gd; */
332 * Handle stop requests at kernel priority. Any requests queued
333 * after this loop will generate another AST.
335 while (lp->lwp_proc->p_stat == SSTOP) {
342 * Reduce our priority in preparation for a return to userland. If
343 * our passive release function was still in place, our priority was
344 * never raised and does not need to be reduced.
346 lwkt_passive_recover(td);
349 * Become the current user scheduled process if we aren't already,
350 * and deal with reschedule requests and other factors.
352 lp->lwp_proc->p_usched->acquire_curproc(lp);
353 /* WARNING: we may have migrated cpu's */
354 /* gd = td->td_gd; */
357 #if !defined(KTR_KERNENTRY)
358 #define KTR_KERNENTRY KTR_ALL
360 KTR_INFO_MASTER(kernentry);
361 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, "pid=%d, tid=%d, trapno=%d, eva=%p",
362 sizeof(int) + sizeof(int) + sizeof(int) + sizeof(vm_offset_t));
363 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "pid=%d, tid=%d",
364 sizeof(int) + sizeof(int));
365 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "pid=%d, tid=%d, call=%d",
366 sizeof(int) + sizeof(int) + sizeof(int));
367 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "pid=%d, tid=%d, err=%d",
368 sizeof(int) + sizeof(int) + sizeof(int));
369 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "pid=%d, tid=%d",
370 sizeof(int) + sizeof(int));
373 * Exception, fault, and trap interface to the kernel.
374 * This common code is called from assembly language IDT gate entry
375 * routines that prepare a suitable stack frame, and restore this
376 * frame after the exception has been processed.
378 * This function is also called from doreti in an interlock to handle ASTs.
379 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
381 * NOTE! We have to retrieve the fault address prior to obtaining the
382 * MP lock because get_mplock() may switch out. YYY cr2 really ought
383 * to be retrieved by the assembly code, not here.
385 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
386 * if an attempt is made to switch from a fast interrupt or IPI. This is
387 * necessary to properly take fatal kernel traps on SMP machines if
388 * get_mplock() has to block.
392 trap(struct trapframe *frame)
394 struct globaldata *gd = mycpu;
395 struct thread *td = gd->gd_curthread;
396 struct lwp *lp = td->td_lwp;
399 int i = 0, ucode = 0, type, code;
404 int crit_count = td->td_critcount;
405 lwkt_tokref_t curstop = td->td_toks_stop;
412 * We need to allow T_DNA faults when the debugger is active since
413 * some dumping paths do large bcopy() which use the floating
414 * point registers for faster copying.
416 if (db_active && frame->tf_trapno != T_DNA) {
417 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
418 ++gd->gd_trap_nesting_level;
419 MAKEMPSAFE(have_mplock);
420 trap_fatal(frame, eva);
421 --gd->gd_trap_nesting_level;
427 ++gd->gd_trap_nesting_level;
428 if (frame->tf_trapno == T_PAGEFLT) {
430 * For some Cyrix CPUs, %cr2 is clobbered by interrupts.
431 * This problem is worked around by using an interrupt
432 * gate for the pagefault handler. We are finally ready
433 * to read %cr2 and then must reenable interrupts.
435 * XXX this should be in the switch statement, but the
436 * NO_FOOF_HACK and VM86 goto and ifdefs obfuscate the
437 * flow of control too much for this to be obviously
444 --gd->gd_trap_nesting_level;
446 if (!(frame->tf_eflags & PSL_I)) {
448 * Buggy application or kernel code has disabled interrupts
449 * and then trapped. Enabling interrupts now is wrong, but
450 * it is better than running with interrupts disabled until
451 * they are accidentally enabled later.
453 type = frame->tf_trapno;
454 if (ISPL(frame->tf_cs)==SEL_UPL || (frame->tf_eflags & PSL_VM)) {
455 MAKEMPSAFE(have_mplock);
457 "pid %ld (%s): trap %d with interrupts disabled\n",
458 (long)curproc->p_pid, curproc->p_comm, type);
459 } else if (type != T_BPTFLT && type != T_TRCTRAP) {
461 * XXX not quite right, since this may be for a
462 * multiple fault in user mode.
464 MAKEMPSAFE(have_mplock);
465 kprintf("kernel trap %d with interrupts disabled\n",
471 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
474 type = frame->tf_trapno;
475 code = frame->tf_err;
478 if (frame->tf_eflags & PSL_VM &&
479 (type == T_PROTFLT || type == T_STKFLT)) {
481 KKASSERT(get_mplock_count(curthread) > 0);
483 i = vm86_emulate((struct vm86frame *)frame);
485 KKASSERT(get_mplock_count(curthread) > 0);
489 * returns to original process
492 vm86_trap((struct vm86frame *)frame,
495 vm86_trap((struct vm86frame *)frame, 0);
497 KKASSERT(0); /* NOT REACHED */
503 * these traps want either a process context, or
504 * assume a normal userspace trap.
508 trap_fatal(frame, eva);
511 type = T_BPTFLT; /* kernel breakpoint */
514 goto kernel_trap; /* normal kernel trap handling */
517 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
520 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid,
521 frame->tf_trapno, eva);
525 sticks = (int)td->td_sticks;
526 lp->lwp_md.md_regs = frame;
529 case T_PRIVINFLT: /* privileged instruction fault */
534 case T_BPTFLT: /* bpt instruction fault */
535 case T_TRCTRAP: /* trace trap */
536 frame->tf_eflags &= ~PSL_T;
541 case T_ARITHTRAP: /* arithmetic trap */
546 case T_ASTFLT: /* Allow process switch */
547 mycpu->gd_cnt.v_soft++;
548 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
549 atomic_clear_int(&mycpu->gd_reqflags,
551 addupc_task(p, p->p_prof.pr_addr,
557 * The following two traps can happen in
558 * vm86 mode, and, if so, we want to handle
561 case T_PROTFLT: /* general protection fault */
562 case T_STKFLT: /* stack fault */
563 if (frame->tf_eflags & PSL_VM) {
564 i = vm86_emulate((struct vm86frame *)frame);
570 ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR;
572 case T_SEGNPFLT: /* segment not present fault */
576 case T_TSSFLT: /* invalid TSS fault */
577 case T_DOUBLEFLT: /* double fault */
582 ucode = code + BUS_SEGM_FAULT ; /* XXX: ???*/
588 case T_PAGEFLT: /* page fault */
589 i = trap_pfault(frame, TRUE, eva);
592 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
604 ucode = BUS_ADRERR; /* XXX */
607 case T_DIVIDE: /* integer divide fault */
614 MAKEMPSAFE(have_mplock);
616 goto handle_powerfail;
617 #else /* !POWERFAIL_NMI */
618 /* machine/parity/power fail/"kitchen sink" faults */
619 if (isa_nmi(code) == 0) {
622 * NMI can be hooked up to a pushbutton
626 kprintf ("NMI ... going to debugger\n");
627 kdb_trap (type, 0, frame);
631 } else if (panic_on_nmi)
632 panic("NMI indicates hardware failure");
634 #endif /* POWERFAIL_NMI */
635 #endif /* NISA > 0 */
637 case T_OFLOW: /* integer overflow fault */
642 case T_BOUND: /* bounds check fault */
649 * Virtual kernel intercept - pass the DNA exception
650 * to the virtual kernel if it asked to handle it.
651 * This occurs when the virtual kernel is holding
652 * onto the FP context for a different emulated
653 * process then the one currently running.
655 * We must still call npxdna() since we may have
656 * saved FP state that the virtual kernel needs
657 * to hand over to a different emulated process.
659 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
660 (td->td_pcb->pcb_flags & FP_VIRTFP)
668 * The kernel may have switched out the FP unit's
669 * state, causing the user process to take a fault
670 * when it tries to use the FP unit. Restore the
676 if (!pmath_emulate) {
678 ucode = FPE_FPU_NP_TRAP;
681 i = (*pmath_emulate)(frame);
683 if (!(frame->tf_eflags & PSL_T))
685 frame->tf_eflags &= ~PSL_T;
688 /* else ucode = emulator_only_knows() XXX */
691 case T_FPOPFLT: /* FPU operand fetch fault */
696 case T_XMMFLT: /* SIMD floating-point exception */
706 case T_PAGEFLT: /* page fault */
707 trap_pfault(frame, FALSE, eva);
713 * The kernel may be using npx for copying or other
721 case T_PROTFLT: /* general protection fault */
722 case T_SEGNPFLT: /* segment not present fault */
724 * Invalid segment selectors and out of bounds
725 * %eip's and %esp's can be set up in user mode.
726 * This causes a fault in kernel mode when the
727 * kernel tries to return to user mode. We want
728 * to get this fault so that we can fix the
729 * problem here and not have to check all the
730 * selectors and pointers when the user changes
733 #define MAYBE_DORETI_FAULT(where, whereto) \
735 if (frame->tf_eip == (int)where) { \
736 frame->tf_eip = (int)whereto; \
740 if (mycpu->gd_intr_nesting_level == 0) {
742 * Invalid %fs's and %gs's can be created using
743 * procfs or PT_SETREGS or by invalidating the
744 * underlying LDT entry. This causes a fault
745 * in kernel mode when the kernel attempts to
746 * switch contexts. Lose the bad context
747 * (XXX) so that we can continue, and generate
750 MAYBE_DORETI_FAULT(doreti_iret,
752 MAYBE_DORETI_FAULT(doreti_popl_ds,
753 doreti_popl_ds_fault);
754 MAYBE_DORETI_FAULT(doreti_popl_es,
755 doreti_popl_es_fault);
756 MAYBE_DORETI_FAULT(doreti_popl_fs,
757 doreti_popl_fs_fault);
758 MAYBE_DORETI_FAULT(doreti_popl_gs,
759 doreti_popl_gs_fault);
760 if (td->td_pcb->pcb_onfault) {
762 (register_t)td->td_pcb->pcb_onfault;
770 * PSL_NT can be set in user mode and isn't cleared
771 * automatically when the kernel is entered. This
772 * causes a TSS fault when the kernel attempts to
773 * `iret' because the TSS link is uninitialized. We
774 * want to get this fault so that we can fix the
775 * problem here and not every time the kernel is
778 if (frame->tf_eflags & PSL_NT) {
779 frame->tf_eflags &= ~PSL_NT;
784 case T_TRCTRAP: /* trace trap */
785 if (frame->tf_eip == (int)IDTVEC(syscall)) {
787 * We've just entered system mode via the
788 * syscall lcall. Continue single stepping
789 * silently until the syscall handler has
794 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
796 * The syscall handler has now saved the
797 * flags. Stop single stepping it.
799 frame->tf_eflags &= ~PSL_T;
803 * Ignore debug register trace traps due to
804 * accesses in the user's address space, which
805 * can happen under several conditions such as
806 * if a user sets a watchpoint on a buffer and
807 * then passes that buffer to a system call.
808 * We still want to get TRCTRAPS for addresses
809 * in kernel space because that is useful when
810 * debugging the kernel.
812 if (user_dbreg_trap()) {
814 * Reset breakpoint bits because the
817 load_dr6(rdr6() & 0xfffffff0);
821 * Fall through (TRCTRAP kernel mode, kernel address)
825 * If DDB is enabled, let it handle the debugger trap.
826 * Otherwise, debugger traps "can't happen".
830 MAKEMPSAFE(have_mplock);
831 if (kdb_trap (type, 0, frame))
838 MAKEMPSAFE(have_mplock);
841 # define TIMER_FREQ 1193182
845 static unsigned lastalert = 0;
847 if(time_second - lastalert > 10)
849 log(LOG_WARNING, "NMI: power fail\n");
850 sysbeep(TIMER_FREQ/880, hz);
851 lastalert = time_second;
856 #else /* !POWERFAIL_NMI */
857 /* machine/parity/power fail/"kitchen sink" faults */
858 if (isa_nmi(code) == 0) {
861 * NMI can be hooked up to a pushbutton
865 kprintf ("NMI ... going to debugger\n");
866 kdb_trap (type, 0, frame);
870 } else if (panic_on_nmi == 0)
873 #endif /* POWERFAIL_NMI */
874 #endif /* NISA > 0 */
877 MAKEMPSAFE(have_mplock);
878 trap_fatal(frame, eva);
883 * Virtual kernel intercept - if the fault is directly related to a
884 * VM context managed by a virtual kernel then let the virtual kernel
887 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
888 vkernel_trap(lp, frame);
893 * Translate fault for emulators (e.g. Linux)
895 if (*p->p_sysent->sv_transtrap)
896 i = (*p->p_sysent->sv_transtrap)(i, type);
898 MAKEMPSAFE(have_mplock);
899 trapsignal(lp, i, ucode);
902 if (type <= MAX_TRAP_MSG) {
903 uprintf("fatal process exception: %s",
905 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
906 uprintf(", fault VA = 0x%lx", (u_long)eva);
912 userret(lp, frame, sticks);
919 if (p != NULL && lp != NULL)
920 KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid);
922 KASSERT(crit_count == td->td_critcount,
923 ("trap: critical section count mismatch! %d/%d",
924 crit_count, td->td_pri));
925 KASSERT(curstop == td->td_toks_stop,
926 ("trap: extra tokens held after trap! %zd/%zd",
927 curstop - &td->td_toks_base,
928 td->td_toks_stop - &td->td_toks_base));
933 trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
936 struct vmspace *vm = NULL;
941 thread_t td = curthread;
942 struct lwp *lp = td->td_lwp;
944 va = trunc_page(eva);
945 if (va >= KERNBASE) {
947 * Don't allow user-mode faults in kernel address space.
948 * An exception: if the faulting address is the invalid
949 * instruction entry in the IDT, then the Intel Pentium
950 * F00F bug workaround was triggered, and we need to
951 * treat it is as an illegal instruction, and not a page
954 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
955 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) {
956 frame->tf_trapno = T_PRIVINFLT;
966 * This is a fault on non-kernel virtual memory.
967 * vm is initialized above to NULL. If curproc is NULL
968 * or curproc->p_vmspace is NULL the fault is fatal.
971 vm = lp->lwp_vmspace;
979 if (frame->tf_err & PGEX_W)
980 ftype = VM_PROT_WRITE;
982 ftype = VM_PROT_READ;
984 if (map != &kernel_map) {
986 * Keep swapout from messing with us during this
996 fault_flags |= VM_FAULT_BURST;
997 if (ftype & VM_PROT_WRITE)
998 fault_flags |= VM_FAULT_DIRTY;
1000 fault_flags |= VM_FAULT_NORMAL;
1001 rv = vm_fault(map, va, ftype, fault_flags);
1002 PRELE(lp->lwp_proc);
1005 * Don't have to worry about process locking or stacks
1008 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
1011 if (rv == KERN_SUCCESS)
1015 if (td->td_gd->gd_intr_nesting_level == 0 &&
1016 td->td_pcb->pcb_onfault) {
1017 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
1020 trap_fatal(frame, eva);
1024 /* kludge to pass faulting virtual address to sendsig */
1025 frame->tf_xflags = frame->tf_err;
1026 frame->tf_err = eva;
1028 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
1032 trap_fatal(struct trapframe *frame, vm_offset_t eva)
1034 int code, type, ss, esp;
1035 struct soft_segment_descriptor softseg;
1037 code = frame->tf_err;
1038 type = frame->tf_trapno;
1039 sdtossd(&gdt[mycpu->gd_cpuid * NGDT + IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
1041 if (type <= MAX_TRAP_MSG)
1042 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
1043 type, trap_msg[type],
1044 frame->tf_eflags & PSL_VM ? "vm86" :
1045 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
1047 /* three separate prints in case of a trap on an unmapped page */
1048 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
1049 kprintf("lapic.id = %08x\n", lapic->id);
1051 if (type == T_PAGEFLT) {
1052 kprintf("fault virtual address = %p\n", (void *)eva);
1053 kprintf("fault code = %s %s, %s\n",
1054 code & PGEX_U ? "user" : "supervisor",
1055 code & PGEX_W ? "write" : "read",
1056 code & PGEX_P ? "protection violation" : "page not present");
1058 kprintf("instruction pointer = 0x%x:0x%x\n",
1059 frame->tf_cs & 0xffff, frame->tf_eip);
1060 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
1061 ss = frame->tf_ss & 0xffff;
1062 esp = frame->tf_esp;
1064 ss = GSEL(GDATA_SEL, SEL_KPL);
1065 esp = (int)&frame->tf_esp;
1067 kprintf("stack pointer = 0x%x:0x%x\n", ss, esp);
1068 kprintf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
1069 kprintf("code segment = base 0x%x, limit 0x%x, type 0x%x\n",
1070 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
1071 kprintf(" = DPL %d, pres %d, def32 %d, gran %d\n",
1072 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
1074 kprintf("processor eflags = ");
1075 if (frame->tf_eflags & PSL_T)
1076 kprintf("trace trap, ");
1077 if (frame->tf_eflags & PSL_I)
1078 kprintf("interrupt enabled, ");
1079 if (frame->tf_eflags & PSL_NT)
1080 kprintf("nested task, ");
1081 if (frame->tf_eflags & PSL_RF)
1082 kprintf("resume, ");
1083 if (frame->tf_eflags & PSL_VM)
1085 kprintf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
1086 kprintf("current process = ");
1088 kprintf("%lu (%s)\n",
1089 (u_long)curproc->p_pid, curproc->p_comm ?
1090 curproc->p_comm : "");
1094 kprintf("current thread = pri %d ", curthread->td_pri);
1095 if (curthread->td_critcount)
1101 * we probably SHOULD have stopped the other CPUs before now!
1102 * another CPU COULD have been touching cpl at this moment...
1104 kprintf(" <- SMP: XXX");
1113 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1116 kprintf("trap number = %d\n", type);
1117 if (type <= MAX_TRAP_MSG)
1118 panic("%s", trap_msg[type]);
1120 panic("unknown/reserved trap");
1124 * Double fault handler. Called when a fault occurs while writing
1125 * a frame for a trap/exception onto the stack. This usually occurs
1126 * when the stack overflows (such is the case with infinite recursion,
1129 * XXX Note that the current PTD gets replaced by IdlePTD when the
1130 * task switch occurs. This means that the stack that was active at
1131 * the time of the double fault is not available at <kstack> unless
1132 * the machine was idle when the double fault occurred. The downside
1133 * of this is that "trace <ebp>" in ddb won't work.
1137 in_kstack_guard(register_t rptr)
1139 thread_t td = curthread;
1141 if ((char *)rptr >= td->td_kstack &&
1142 (char *)rptr < td->td_kstack + PAGE_SIZE) {
1149 dblfault_handler(void)
1151 struct mdglobaldata *gd = mdcpu;
1153 if (in_kstack_guard(gd->gd_common_tss.tss_esp) ||
1154 in_kstack_guard(gd->gd_common_tss.tss_ebp)) {
1155 kprintf("DOUBLE FAULT - KERNEL STACK GUARD HIT!\n");
1157 kprintf("DOUBLE FAULT:\n");
1159 kprintf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1160 kprintf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1161 kprintf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
1163 /* three separate prints in case of a trap on an unmapped page */
1164 kprintf("cpuid = %d; ", gd->mi.gd_cpuid);
1165 kprintf("lapic.id = %08x\n", lapic->id);
1167 panic("double fault");
1171 * syscall2 - MP aware system call request C handler
1173 * A system call is essentially treated as a trap. The MP lock is not
1174 * held on entry or return. We are responsible for handling ASTs
1175 * (e.g. a task switch) prior to return.
1180 syscall2(struct trapframe *frame)
1182 struct thread *td = curthread;
1183 struct proc *p = td->td_proc;
1184 struct lwp *lp = td->td_lwp;
1186 struct sysent *callp;
1187 register_t orig_tf_eflags;
1192 int crit_count = td->td_critcount;
1195 int have_mplock = 0;
1198 union sysunion args;
1201 if (ISPL(frame->tf_cs) != SEL_UPL) {
1208 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid,
1211 userenter(td, p); /* lazy raise our priority */
1216 sticks = (int)td->td_sticks;
1217 orig_tf_eflags = frame->tf_eflags;
1220 * Virtual kernel intercept - if a VM context managed by a virtual
1221 * kernel issues a system call the virtual kernel handles it, not us.
1222 * Restore the virtual kernel context and return from its system
1223 * call. The current frame is copied out to the virtual kernel.
1225 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1226 vkernel_trap(lp, frame);
1227 error = EJUSTRETURN;
1233 * Get the system call parameters and account for time
1235 lp->lwp_md.md_regs = frame;
1236 params = (caddr_t)frame->tf_esp + sizeof(int);
1237 code = frame->tf_eax;
1239 if (p->p_sysent->sv_prepsyscall) {
1240 (*p->p_sysent->sv_prepsyscall)(
1241 frame, (int *)(&args.nosys.sysmsg + 1),
1245 * Need to check if this is a 32 bit or 64 bit syscall.
1246 * fuword is MP aware.
1248 if (code == SYS_syscall) {
1250 * Code is first argument, followed by actual args.
1252 code = fuword(params);
1253 params += sizeof(int);
1254 } else if (code == SYS___syscall) {
1256 * Like syscall, but code is a quad, so as to maintain
1257 * quad alignment for the rest of the arguments.
1259 code = fuword(params);
1260 params += sizeof(quad_t);
1264 code &= p->p_sysent->sv_mask;
1266 if (code >= p->p_sysent->sv_size)
1267 callp = &p->p_sysent->sv_table[0];
1269 callp = &p->p_sysent->sv_table[code];
1271 narg = callp->sy_narg & SYF_ARGMASK;
1274 if (p->p_sysent->sv_name[0] == 'L')
1275 kprintf("Linux syscall, code = %d\n", code);
1279 * copyin is MP aware, but the tracing code is not
1281 if (narg && params) {
1282 error = copyin(params, (caddr_t)(&args.nosys.sysmsg + 1),
1283 narg * sizeof(register_t));
1286 if (KTRPOINT(td, KTR_SYSCALL)) {
1287 MAKEMPSAFE(have_mplock);
1289 ktrsyscall(lp, code, narg,
1290 (void *)(&args.nosys.sysmsg + 1));
1298 if (KTRPOINT(td, KTR_SYSCALL)) {
1299 MAKEMPSAFE(have_mplock);
1300 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1305 * For traditional syscall code edx is left untouched when 32 bit
1306 * results are returned. Since edx is loaded from fds[1] when the
1307 * system call returns we pre-set it here.
1309 args.sysmsg_fds[0] = 0;
1310 args.sysmsg_fds[1] = frame->tf_edx;
1313 * The syscall might manipulate the trap frame. If it does it
1314 * will probably return EJUSTRETURN.
1316 args.sysmsg_frame = frame;
1318 STOPEVENT(p, S_SCE, narg); /* MP aware */
1321 * NOTE: All system calls run MPSAFE now. The system call itself
1322 * is responsible for getting the MP lock.
1324 error = (*callp->sy_call)(&args);
1328 * MP SAFE (we may or may not have the MP lock at this point)
1333 * Reinitialize proc pointer `p' as it may be different
1334 * if this is a child returning from fork syscall.
1337 lp = curthread->td_lwp;
1338 frame->tf_eax = args.sysmsg_fds[0];
1339 frame->tf_edx = args.sysmsg_fds[1];
1340 frame->tf_eflags &= ~PSL_C;
1344 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1345 * int 0x80 is 2 bytes. We saved this in tf_err.
1347 frame->tf_eip -= frame->tf_err;
1352 panic("Unexpected EASYNC return value (for now)");
1355 if (p->p_sysent->sv_errsize) {
1356 if (error >= p->p_sysent->sv_errsize)
1357 error = -1; /* XXX */
1359 error = p->p_sysent->sv_errtbl[error];
1361 frame->tf_eax = error;
1362 frame->tf_eflags |= PSL_C;
1367 * Traced syscall. trapsignal() is not MP aware.
1369 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1370 MAKEMPSAFE(have_mplock);
1371 frame->tf_eflags &= ~PSL_T;
1372 trapsignal(lp, SIGTRAP, TRAP_TRACE);
1376 * Handle reschedule and other end-of-syscall issues
1378 userret(lp, frame, sticks);
1381 if (KTRPOINT(td, KTR_SYSRET)) {
1382 MAKEMPSAFE(have_mplock);
1383 ktrsysret(lp, code, error, args.sysmsg_result);
1388 * This works because errno is findable through the
1389 * register set. If we ever support an emulation where this
1390 * is not the case, this code will need to be revisited.
1392 STOPEVENT(p, S_SCX, code);
1397 * Release the MP lock if we had to get it
1402 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error);
1404 KASSERT(crit_count == td->td_critcount,
1405 ("syscall: critical section count mismatch! %d/%d",
1406 crit_count, td->td_pri));
1407 KASSERT(&td->td_toks_base == td->td_toks_stop,
1408 ("syscall: extra tokens held after trap! %zd",
1409 td->td_toks_stop - &td->td_toks_base));
1414 * NOTE: MP lock not held at any point.
1417 fork_return(struct lwp *lp, struct trapframe *frame)
1419 frame->tf_eax = 0; /* Child returns zero */
1420 frame->tf_eflags &= ~PSL_C; /* success */
1423 generic_lwp_return(lp, frame);
1424 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1428 * Simplified back end of syscall(), used when returning from fork()
1429 * directly into user mode.
1431 * This code will return back into the fork trampoline code which then
1434 * NOTE: The mplock is not held at any point.
1437 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1439 struct proc *p = lp->lwp_proc;
1442 * Newly forked processes are given a kernel priority. We have to
1443 * adjust the priority to a normal user priority and fake entry
1444 * into the kernel (call userenter()) to install a passive release
1445 * function just in case userret() decides to stop the process. This
1446 * can occur when ^Z races a fork. If we do not install the passive
1447 * release function the current process designation will not be
1448 * released when the thread goes to sleep.
1450 lwkt_setpri_self(TDPRI_USER_NORM);
1451 userenter(lp->lwp_thread, p);
1452 userret(lp, frame, 0);
1454 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1455 ktrsysret(lp, SYS_fork, 0, 0);
1457 p->p_flag |= P_PASSIVE_ACQ;
1459 p->p_flag &= ~P_PASSIVE_ACQ;
1463 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1464 * fault (which is then passed back to the virtual kernel) if an attempt is
1465 * made to use the FP unit.
1467 * XXX this is a fairly big hack.
1470 set_vkernel_fp(struct trapframe *frame)
1472 struct thread *td = curthread;
1474 if (frame->tf_xflags & PGEX_FPFAULT) {
1475 td->td_pcb->pcb_flags |= FP_VIRTFP;
1476 if (mdcpu->gd_npxthread == td)
1479 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1484 * Called from vkernel_trap() to fixup the vkernel's syscall
1485 * frame for vmspace_ctl() return.
1488 cpu_vkernel_trap(struct trapframe *frame, int error)
1490 frame->tf_eax = error;
1492 frame->tf_eflags |= PSL_C;
1494 frame->tf_eflags &= ~PSL_C;