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/resourcevar.h>
61 #include <sys/signalvar.h>
62 #include <sys/signal2.h>
63 #include <sys/syscall.h>
64 #include <sys/sysctl.h>
65 #include <sys/sysent.h>
67 #include <sys/vmmeter.h>
68 #include <sys/malloc.h>
70 #include <sys/ktrace.h>
73 #include <sys/upcall.h>
74 #include <sys/vkernel.h>
75 #include <sys/sysproto.h>
76 #include <sys/sysunion.h>
79 #include <vm/vm_param.h>
82 #include <vm/vm_kern.h>
83 #include <vm/vm_map.h>
84 #include <vm/vm_page.h>
85 #include <vm/vm_extern.h>
87 #include <machine/cpu.h>
88 #include <machine/md_var.h>
89 #include <machine/pcb.h>
90 #include <machine/smp.h>
91 #include <machine/tss.h>
92 #include <machine/specialreg.h>
93 #include <machine/globaldata.h>
95 #include <machine_base/isa/intr_machdep.h>
98 #include <sys/syslog.h>
99 #include <machine/clock.h>
102 #include <machine/vm86.h>
105 #include <sys/msgport2.h>
106 #include <sys/thread2.h>
110 #define MAKEMPSAFE(have_mplock) \
111 if (have_mplock == 0) { \
118 #define MAKEMPSAFE(have_mplock)
122 int (*pmath_emulate) (struct trapframe *);
124 extern void trap (struct trapframe *frame);
125 extern int trapwrite (unsigned addr);
126 extern void syscall2 (struct trapframe *frame);
128 static int trap_pfault (struct trapframe *, int, vm_offset_t);
129 static void trap_fatal (struct trapframe *, vm_offset_t);
130 void dblfault_handler (void);
132 extern inthand_t IDTVEC(syscall);
134 #define MAX_TRAP_MSG 28
135 static char *trap_msg[] = {
137 "privileged instruction fault", /* 1 T_PRIVINFLT */
139 "breakpoint instruction fault", /* 3 T_BPTFLT */
142 "arithmetic trap", /* 6 T_ARITHTRAP */
143 "system forced exception", /* 7 T_ASTFLT */
145 "general protection fault", /* 9 T_PROTFLT */
146 "trace trap", /* 10 T_TRCTRAP */
148 "page fault", /* 12 T_PAGEFLT */
150 "alignment fault", /* 14 T_ALIGNFLT */
154 "integer divide fault", /* 18 T_DIVIDE */
155 "non-maskable interrupt trap", /* 19 T_NMI */
156 "overflow trap", /* 20 T_OFLOW */
157 "FPU bounds check fault", /* 21 T_BOUND */
158 "FPU device not available", /* 22 T_DNA */
159 "double fault", /* 23 T_DOUBLEFLT */
160 "FPU operand fetch fault", /* 24 T_FPOPFLT */
161 "invalid TSS fault", /* 25 T_TSSFLT */
162 "segment not present fault", /* 26 T_SEGNPFLT */
163 "stack fault", /* 27 T_STKFLT */
164 "machine check trap", /* 28 T_MCHK */
167 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
168 extern int has_f00f_bug;
172 static int ddb_on_nmi = 1;
173 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
174 &ddb_on_nmi, 0, "Go to DDB on NMI");
176 static int panic_on_nmi = 1;
177 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
178 &panic_on_nmi, 0, "Panic on NMI");
179 static int fast_release;
180 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
181 &fast_release, 0, "Passive Release was optimal");
182 static int slow_release;
183 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
184 &slow_release, 0, "Passive Release was nonoptimal");
186 static int syscall_mpsafe = 1;
187 SYSCTL_INT(_kern, OID_AUTO, syscall_mpsafe, CTLFLAG_RW,
188 &syscall_mpsafe, 0, "Allow MPSAFE marked syscalls to run without BGL");
189 TUNABLE_INT("kern.syscall_mpsafe", &syscall_mpsafe);
190 static int trap_mpsafe = 1;
191 SYSCTL_INT(_kern, OID_AUTO, trap_mpsafe, CTLFLAG_RW,
192 &trap_mpsafe, 0, "Allow traps to mostly run without the BGL");
193 TUNABLE_INT("kern.trap_mpsafe", &trap_mpsafe);
196 MALLOC_DEFINE(M_SYSMSG, "sysmsg", "sysmsg structure");
197 extern int max_sysmsg;
200 * userenter() passively intercepts the thread switch function to increase
201 * the thread priority from a user priority to a kernel priority, reducing
202 * syscall and trap overhead for the case where no switch occurs.
206 userenter(struct thread *curtd)
208 curtd->td_release = lwkt_passive_release;
212 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
213 * must be completed before we can return to or try to return to userland.
215 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
216 * arithmatic on the delta calculation so the absolute tick values are
217 * truncated to an integer.
220 userret(struct lwp *lp, struct trapframe *frame, int sticks)
222 struct proc *p = lp->lwp_proc;
226 * Charge system time if profiling. Note: times are in microseconds.
227 * This may do a copyout and block, so do it first even though it
228 * means some system time will be charged as user time.
230 if (p->p_flag & P_PROFIL) {
231 addupc_task(p, frame->tf_eip,
232 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
237 * If the jungle wants us dead, so be it.
239 if (lp->lwp_flag & LWP_WEXIT) {
242 rel_mplock(); /* NOT REACHED */
246 * Block here if we are in a stopped state.
248 if (p->p_stat == SSTOP) {
256 * Post any pending upcalls. If running a virtual kernel be sure
257 * to restore the virtual kernel's vmspace before posting the upcall.
259 if (p->p_flag & P_UPCALLPEND) {
260 p->p_flag &= ~P_UPCALLPEND;
268 * Post any pending signals. If running a virtual kernel be sure
269 * to restore the virtual kernel's vmspace before posting the signal.
271 if ((sig = CURSIG_TRACE(lp)) != 0) {
279 * block here if we are swapped out, but still process signals
280 * (such as SIGKILL). proc0 (the swapin scheduler) is already
281 * aware of our situation, we do not have to wake it up.
283 if (p->p_flag & P_SWAPPEDOUT) {
285 p->p_flag |= P_SWAPWAIT;
287 if (p->p_flag & P_SWAPWAIT)
288 tsleep(p, PCATCH, "SWOUT", 0);
289 p->p_flag &= ~P_SWAPWAIT;
295 * Make sure postsig() handled request to restore old signal mask after
296 * running signal handler.
298 KKASSERT((lp->lwp_flag & LWP_OLDMASK) == 0);
302 * Cleanup from userenter and any passive release that might have occured.
303 * We must reclaim the current-process designation before we can return
304 * to usermode. We also handle both LWKT and USER reschedule requests.
307 userexit(struct lwp *lp)
309 struct thread *td = lp->lwp_thread;
310 /* globaldata_t gd = td->td_gd; */
313 * Handle stop requests at kernel priority. Any requests queued
314 * after this loop will generate another AST.
316 while (lp->lwp_proc->p_stat == SSTOP) {
323 * Reduce our priority in preparation for a return to userland. If
324 * our passive release function was still in place, our priority was
325 * never raised and does not need to be reduced.
327 lwkt_passive_recover(td);
330 * Become the current user scheduled process if we aren't already,
331 * and deal with reschedule requests and other factors.
333 lp->lwp_proc->p_usched->acquire_curproc(lp);
334 /* WARNING: we may have migrated cpu's */
335 /* gd = td->td_gd; */
338 #if !defined(KTR_KERNENTRY)
339 #define KTR_KERNENTRY KTR_ALL
341 KTR_INFO_MASTER(kernentry);
342 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, "pid=%d, tid=%d, trapno=%d, eva=%p",
343 sizeof(int) + sizeof(int) + sizeof(int) + sizeof(vm_offset_t));
344 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "pid=%d, tid=%d",
345 sizeof(int) + sizeof(int));
346 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "pid=%d, tid=%d, call=%d",
347 sizeof(int) + sizeof(int) + sizeof(int));
348 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "pid=%d, tid=%d, err=%d",
349 sizeof(int) + sizeof(int) + sizeof(int));
350 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "pid=%d, tid=%d",
351 sizeof(int) + sizeof(int));
354 * Exception, fault, and trap interface to the kernel.
355 * This common code is called from assembly language IDT gate entry
356 * routines that prepare a suitable stack frame, and restore this
357 * frame after the exception has been processed.
359 * This function is also called from doreti in an interlock to handle ASTs.
360 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
362 * NOTE! We have to retrieve the fault address prior to obtaining the
363 * MP lock because get_mplock() may switch out. YYY cr2 really ought
364 * to be retrieved by the assembly code, not here.
366 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
367 * if an attempt is made to switch from a fast interrupt or IPI. This is
368 * necessary to properly take fatal kernel traps on SMP machines if
369 * get_mplock() has to block.
373 trap(struct trapframe *frame)
375 struct globaldata *gd = mycpu;
376 struct thread *td = gd->gd_curthread;
377 struct lwp *lp = td->td_lwp;
380 int i = 0, ucode = 0, type, code;
385 int crit_count = td->td_pri & ~TDPRI_MASK;
392 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
393 ++gd->gd_trap_nesting_level;
394 MAKEMPSAFE(have_mplock);
395 trap_fatal(frame, eva);
396 --gd->gd_trap_nesting_level;
402 ++gd->gd_trap_nesting_level;
403 if (frame->tf_trapno == T_PAGEFLT) {
405 * For some Cyrix CPUs, %cr2 is clobbered by interrupts.
406 * This problem is worked around by using an interrupt
407 * gate for the pagefault handler. We are finally ready
408 * to read %cr2 and then must reenable interrupts.
410 * XXX this should be in the switch statement, but the
411 * NO_FOOF_HACK and VM86 goto and ifdefs obfuscate the
412 * flow of control too much for this to be obviously
420 if (trap_mpsafe == 0)
421 MAKEMPSAFE(have_mplock);
424 --gd->gd_trap_nesting_level;
426 if (!(frame->tf_eflags & PSL_I)) {
428 * Buggy application or kernel code has disabled interrupts
429 * and then trapped. Enabling interrupts now is wrong, but
430 * it is better than running with interrupts disabled until
431 * they are accidentally enabled later.
433 type = frame->tf_trapno;
434 if (ISPL(frame->tf_cs)==SEL_UPL || (frame->tf_eflags & PSL_VM)) {
435 MAKEMPSAFE(have_mplock);
437 "pid %ld (%s): trap %d with interrupts disabled\n",
438 (long)curproc->p_pid, curproc->p_comm, type);
439 } else if (type != T_BPTFLT && type != T_TRCTRAP) {
441 * XXX not quite right, since this may be for a
442 * multiple fault in user mode.
444 MAKEMPSAFE(have_mplock);
445 kprintf("kernel trap %d with interrupts disabled\n",
451 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
454 type = frame->tf_trapno;
455 code = frame->tf_err;
458 if (frame->tf_eflags & PSL_VM &&
459 (type == T_PROTFLT || type == T_STKFLT)) {
461 KKASSERT(td->td_mpcount > 0);
463 i = vm86_emulate((struct vm86frame *)frame);
465 KKASSERT(td->td_mpcount > 0);
469 * returns to original process
472 vm86_trap((struct vm86frame *)frame,
475 vm86_trap((struct vm86frame *)frame, 0);
477 KKASSERT(0); /* NOT REACHED */
483 * these traps want either a process context, or
484 * assume a normal userspace trap.
488 trap_fatal(frame, eva);
491 type = T_BPTFLT; /* kernel breakpoint */
494 goto kernel_trap; /* normal kernel trap handling */
497 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
500 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid,
501 frame->tf_trapno, eva);
505 sticks = (int)td->td_sticks;
506 lp->lwp_md.md_regs = frame;
509 case T_PRIVINFLT: /* privileged instruction fault */
514 case T_BPTFLT: /* bpt instruction fault */
515 case T_TRCTRAP: /* trace trap */
516 frame->tf_eflags &= ~PSL_T;
520 case T_ARITHTRAP: /* arithmetic trap */
525 case T_ASTFLT: /* Allow process switch */
526 mycpu->gd_cnt.v_soft++;
527 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
528 atomic_clear_int_nonlocked(&mycpu->gd_reqflags,
530 addupc_task(p, p->p_prof.pr_addr,
536 * The following two traps can happen in
537 * vm86 mode, and, if so, we want to handle
540 case T_PROTFLT: /* general protection fault */
541 case T_STKFLT: /* stack fault */
542 if (frame->tf_eflags & PSL_VM) {
543 i = vm86_emulate((struct vm86frame *)frame);
550 case T_SEGNPFLT: /* segment not present fault */
551 case T_TSSFLT: /* invalid TSS fault */
552 case T_DOUBLEFLT: /* double fault */
554 ucode = code + BUS_SEGM_FAULT ;
558 case T_PAGEFLT: /* page fault */
559 MAKEMPSAFE(have_mplock);
560 i = trap_pfault(frame, TRUE, eva);
563 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
573 case T_DIVIDE: /* integer divide fault */
580 MAKEMPSAFE(have_mplock);
582 goto handle_powerfail;
583 #else /* !POWERFAIL_NMI */
584 /* machine/parity/power fail/"kitchen sink" faults */
585 if (isa_nmi(code) == 0) {
588 * NMI can be hooked up to a pushbutton
592 kprintf ("NMI ... going to debugger\n");
593 kdb_trap (type, 0, frame);
597 } else if (panic_on_nmi)
598 panic("NMI indicates hardware failure");
600 #endif /* POWERFAIL_NMI */
601 #endif /* NISA > 0 */
603 case T_OFLOW: /* integer overflow fault */
608 case T_BOUND: /* bounds check fault */
615 * Virtual kernel intercept - pass the DNA exception
616 * to the virtual kernel if it asked to handle it.
617 * This occurs when the virtual kernel is holding
618 * onto the FP context for a different emulated
619 * process then the one currently running.
621 * We must still call npxdna() since we may have
622 * saved FP state that the virtual kernel needs
623 * to hand over to a different emulated process.
625 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
626 (td->td_pcb->pcb_flags & FP_VIRTFP)
634 * The kernel may have switched out the FP unit's
635 * state, causing the user process to take a fault
636 * when it tries to use the FP unit. Restore the
642 if (!pmath_emulate) {
644 ucode = FPE_FPU_NP_TRAP;
647 i = (*pmath_emulate)(frame);
649 if (!(frame->tf_eflags & PSL_T))
651 frame->tf_eflags &= ~PSL_T;
654 /* else ucode = emulator_only_knows() XXX */
657 case T_FPOPFLT: /* FPU operand fetch fault */
662 case T_XMMFLT: /* SIMD floating-point exception */
672 case T_PAGEFLT: /* page fault */
673 MAKEMPSAFE(have_mplock);
674 trap_pfault(frame, FALSE, eva);
680 * The kernel may be using npx for copying or other
688 case T_PROTFLT: /* general protection fault */
689 case T_SEGNPFLT: /* segment not present fault */
691 * Invalid segment selectors and out of bounds
692 * %eip's and %esp's can be set up in user mode.
693 * This causes a fault in kernel mode when the
694 * kernel tries to return to user mode. We want
695 * to get this fault so that we can fix the
696 * problem here and not have to check all the
697 * selectors and pointers when the user changes
700 #define MAYBE_DORETI_FAULT(where, whereto) \
702 if (frame->tf_eip == (int)where) { \
703 frame->tf_eip = (int)whereto; \
707 if (mycpu->gd_intr_nesting_level == 0) {
709 * Invalid %fs's and %gs's can be created using
710 * procfs or PT_SETREGS or by invalidating the
711 * underlying LDT entry. This causes a fault
712 * in kernel mode when the kernel attempts to
713 * switch contexts. Lose the bad context
714 * (XXX) so that we can continue, and generate
717 MAYBE_DORETI_FAULT(doreti_iret,
719 MAYBE_DORETI_FAULT(doreti_popl_ds,
720 doreti_popl_ds_fault);
721 MAYBE_DORETI_FAULT(doreti_popl_es,
722 doreti_popl_es_fault);
723 MAYBE_DORETI_FAULT(doreti_popl_fs,
724 doreti_popl_fs_fault);
725 MAYBE_DORETI_FAULT(doreti_popl_gs,
726 doreti_popl_gs_fault);
727 if (td->td_pcb->pcb_onfault) {
729 (register_t)td->td_pcb->pcb_onfault;
737 * PSL_NT can be set in user mode and isn't cleared
738 * automatically when the kernel is entered. This
739 * causes a TSS fault when the kernel attempts to
740 * `iret' because the TSS link is uninitialized. We
741 * want to get this fault so that we can fix the
742 * problem here and not every time the kernel is
745 if (frame->tf_eflags & PSL_NT) {
746 frame->tf_eflags &= ~PSL_NT;
751 case T_TRCTRAP: /* trace trap */
752 if (frame->tf_eip == (int)IDTVEC(syscall)) {
754 * We've just entered system mode via the
755 * syscall lcall. Continue single stepping
756 * silently until the syscall handler has
761 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
763 * The syscall handler has now saved the
764 * flags. Stop single stepping it.
766 frame->tf_eflags &= ~PSL_T;
770 * Ignore debug register trace traps due to
771 * accesses in the user's address space, which
772 * can happen under several conditions such as
773 * if a user sets a watchpoint on a buffer and
774 * then passes that buffer to a system call.
775 * We still want to get TRCTRAPS for addresses
776 * in kernel space because that is useful when
777 * debugging the kernel.
779 if (user_dbreg_trap()) {
781 * Reset breakpoint bits because the
784 load_dr6(rdr6() & 0xfffffff0);
788 * Fall through (TRCTRAP kernel mode, kernel address)
792 * If DDB is enabled, let it handle the debugger trap.
793 * Otherwise, debugger traps "can't happen".
796 MAKEMPSAFE(have_mplock);
797 if (kdb_trap (type, 0, frame))
804 MAKEMPSAFE(have_mplock);
807 # define TIMER_FREQ 1193182
811 static unsigned lastalert = 0;
813 if(time_second - lastalert > 10)
815 log(LOG_WARNING, "NMI: power fail\n");
816 sysbeep(TIMER_FREQ/880, hz);
817 lastalert = time_second;
822 #else /* !POWERFAIL_NMI */
823 /* machine/parity/power fail/"kitchen sink" faults */
824 if (isa_nmi(code) == 0) {
827 * NMI can be hooked up to a pushbutton
831 kprintf ("NMI ... going to debugger\n");
832 kdb_trap (type, 0, frame);
836 } else if (panic_on_nmi == 0)
839 #endif /* POWERFAIL_NMI */
840 #endif /* NISA > 0 */
843 MAKEMPSAFE(have_mplock);
844 trap_fatal(frame, eva);
849 * Virtual kernel intercept - if the fault is directly related to a
850 * VM context managed by a virtual kernel then let the virtual kernel
853 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
854 vkernel_trap(lp, frame);
859 * Translate fault for emulators (e.g. Linux)
861 if (*p->p_sysent->sv_transtrap)
862 i = (*p->p_sysent->sv_transtrap)(i, type);
864 MAKEMPSAFE(have_mplock);
865 trapsignal(lp, i, ucode);
868 if (type <= MAX_TRAP_MSG) {
869 uprintf("fatal process exception: %s",
871 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
872 uprintf(", fault VA = 0x%lx", (u_long)eva);
879 if (ISPL(frame->tf_cs) == SEL_UPL)
880 KASSERT(td->td_mpcount == have_mplock, ("badmpcount trap/end from %p", (void *)frame->tf_eip));
882 userret(lp, frame, sticks);
889 if (p != NULL && lp != NULL)
890 KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid);
892 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
893 ("syscall: critical section count mismatch! %d/%d",
894 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
899 trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
902 struct vmspace *vm = NULL;
906 thread_t td = curthread;
907 struct lwp *lp = td->td_lwp;
909 va = trunc_page(eva);
910 if (va >= KERNBASE) {
912 * Don't allow user-mode faults in kernel address space.
913 * An exception: if the faulting address is the invalid
914 * instruction entry in the IDT, then the Intel Pentium
915 * F00F bug workaround was triggered, and we need to
916 * treat it is as an illegal instruction, and not a page
919 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
920 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) {
921 frame->tf_trapno = T_PRIVINFLT;
931 * This is a fault on non-kernel virtual memory.
932 * vm is initialized above to NULL. If curproc is NULL
933 * or curproc->p_vmspace is NULL the fault is fatal.
936 vm = lp->lwp_vmspace;
944 if (frame->tf_err & PGEX_W)
945 ftype = VM_PROT_WRITE;
947 ftype = VM_PROT_READ;
949 if (map != &kernel_map) {
951 * Keep swapout from messing with us during this
957 * Grow the stack if necessary
959 /* grow_stack returns false only if va falls into
960 * a growable stack region and the stack growth
961 * fails. It returns true if va was not within
962 * a growable stack region, or if the stack
965 if (!grow_stack(lp->lwp_proc, va)) {
971 /* Fault in the user page: */
972 rv = vm_fault(map, va, ftype,
973 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
979 * Don't have to worry about process locking or stacks
982 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
985 if (rv == KERN_SUCCESS)
989 if (td->td_gd->gd_intr_nesting_level == 0 &&
990 td->td_pcb->pcb_onfault) {
991 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
994 trap_fatal(frame, eva);
998 /* kludge to pass faulting virtual address to sendsig */
999 frame->tf_xflags = frame->tf_err;
1000 frame->tf_err = eva;
1002 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
1006 trap_fatal(struct trapframe *frame, vm_offset_t eva)
1008 int code, type, ss, esp;
1009 struct soft_segment_descriptor softseg;
1011 code = frame->tf_err;
1012 type = frame->tf_trapno;
1013 sdtossd(&gdt[mycpu->gd_cpuid * NGDT + IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
1015 if (type <= MAX_TRAP_MSG)
1016 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
1017 type, trap_msg[type],
1018 frame->tf_eflags & PSL_VM ? "vm86" :
1019 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
1021 /* three separate prints in case of a trap on an unmapped page */
1022 kprintf("mp_lock = %08x; ", mp_lock);
1023 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
1024 kprintf("lapic.id = %08x\n", lapic.id);
1026 if (type == T_PAGEFLT) {
1027 kprintf("fault virtual address = 0x%x\n", eva);
1028 kprintf("fault code = %s %s, %s\n",
1029 code & PGEX_U ? "user" : "supervisor",
1030 code & PGEX_W ? "write" : "read",
1031 code & PGEX_P ? "protection violation" : "page not present");
1033 kprintf("instruction pointer = 0x%x:0x%x\n",
1034 frame->tf_cs & 0xffff, frame->tf_eip);
1035 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
1036 ss = frame->tf_ss & 0xffff;
1037 esp = frame->tf_esp;
1039 ss = GSEL(GDATA_SEL, SEL_KPL);
1040 esp = (int)&frame->tf_esp;
1042 kprintf("stack pointer = 0x%x:0x%x\n", ss, esp);
1043 kprintf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
1044 kprintf("code segment = base 0x%x, limit 0x%x, type 0x%x\n",
1045 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
1046 kprintf(" = DPL %d, pres %d, def32 %d, gran %d\n",
1047 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
1049 kprintf("processor eflags = ");
1050 if (frame->tf_eflags & PSL_T)
1051 kprintf("trace trap, ");
1052 if (frame->tf_eflags & PSL_I)
1053 kprintf("interrupt enabled, ");
1054 if (frame->tf_eflags & PSL_NT)
1055 kprintf("nested task, ");
1056 if (frame->tf_eflags & PSL_RF)
1057 kprintf("resume, ");
1058 if (frame->tf_eflags & PSL_VM)
1060 kprintf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
1061 kprintf("current process = ");
1063 kprintf("%lu (%s)\n",
1064 (u_long)curproc->p_pid, curproc->p_comm ?
1065 curproc->p_comm : "");
1069 kprintf("current thread = pri %d ", curthread->td_pri);
1070 if (curthread->td_pri >= TDPRI_CRIT)
1076 * we probably SHOULD have stopped the other CPUs before now!
1077 * another CPU COULD have been touching cpl at this moment...
1079 kprintf(" <- SMP: XXX");
1088 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1091 kprintf("trap number = %d\n", type);
1092 if (type <= MAX_TRAP_MSG)
1093 panic("%s", trap_msg[type]);
1095 panic("unknown/reserved trap");
1099 * Double fault handler. Called when a fault occurs while writing
1100 * a frame for a trap/exception onto the stack. This usually occurs
1101 * when the stack overflows (such is the case with infinite recursion,
1104 * XXX Note that the current PTD gets replaced by IdlePTD when the
1105 * task switch occurs. This means that the stack that was active at
1106 * the time of the double fault is not available at <kstack> unless
1107 * the machine was idle when the double fault occurred. The downside
1108 * of this is that "trace <ebp>" in ddb won't work.
1111 dblfault_handler(void)
1113 struct mdglobaldata *gd = mdcpu;
1115 kprintf("\nFatal double fault:\n");
1116 kprintf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1117 kprintf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1118 kprintf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
1120 /* three separate prints in case of a trap on an unmapped page */
1121 kprintf("mp_lock = %08x; ", mp_lock);
1122 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
1123 kprintf("lapic.id = %08x\n", lapic.id);
1125 panic("double fault");
1129 * Compensate for 386 brain damage (missing URKR).
1130 * This is a little simpler than the pagefault handler in trap() because
1131 * it the page tables have already been faulted in and high addresses
1132 * are thrown out early for other reasons.
1135 trapwrite(unsigned addr)
1142 va = trunc_page((vm_offset_t)addr);
1144 * XXX - MAX is END. Changed > to >= for temp. fix.
1146 if (va >= VM_MAX_USER_ADDRESS)
1149 lp = curthread->td_lwp;
1150 vm = lp->lwp_vmspace;
1152 PHOLD(lp->lwp_proc);
1154 if (!grow_stack(lp->lwp_proc, va)) {
1155 PRELE(lp->lwp_proc);
1160 * fault the data page
1162 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY);
1164 PRELE(lp->lwp_proc);
1166 if (rv != KERN_SUCCESS)
1173 * syscall2 - MP aware system call request C handler
1175 * A system call is essentially treated as a trap except that the
1176 * MP lock is not held on entry or return. We are responsible for
1177 * obtaining the MP lock if necessary and for handling ASTs
1178 * (e.g. a task switch) prior to return.
1180 * In general, only simple access and manipulation of curproc and
1181 * the current stack is allowed without having to hold MP lock.
1183 * MPSAFE - note that large sections of this routine are run without
1188 syscall2(struct trapframe *frame)
1190 struct thread *td = curthread;
1191 struct proc *p = td->td_proc;
1192 struct lwp *lp = td->td_lwp;
1194 struct sysent *callp;
1195 register_t orig_tf_eflags;
1200 int crit_count = td->td_pri & ~TDPRI_MASK;
1203 int have_mplock = 0;
1206 union sysunion args;
1209 if (ISPL(frame->tf_cs) != SEL_UPL) {
1216 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid,
1220 KASSERT(td->td_mpcount == 0, ("badmpcount syscall2 from %p", (void *)frame->tf_eip));
1221 if (syscall_mpsafe == 0)
1222 MAKEMPSAFE(have_mplock);
1224 userenter(td); /* lazy raise our priority */
1229 sticks = (int)td->td_sticks;
1230 orig_tf_eflags = frame->tf_eflags;
1233 * Virtual kernel intercept - if a VM context managed by a virtual
1234 * kernel issues a system call the virtual kernel handles it, not us.
1235 * Restore the virtual kernel context and return from its system
1236 * call. The current frame is copied out to the virtual kernel.
1238 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1239 error = vkernel_trap(lp, frame);
1240 frame->tf_eax = error;
1242 frame->tf_eflags |= PSL_C;
1243 error = EJUSTRETURN;
1248 * Get the system call parameters and account for time
1250 lp->lwp_md.md_regs = frame;
1251 params = (caddr_t)frame->tf_esp + sizeof(int);
1252 code = frame->tf_eax;
1254 if (p->p_sysent->sv_prepsyscall) {
1255 (*p->p_sysent->sv_prepsyscall)(
1256 frame, (int *)(&args.nosys.sysmsg + 1),
1260 * Need to check if this is a 32 bit or 64 bit syscall.
1261 * fuword is MP aware.
1263 if (code == SYS_syscall) {
1265 * Code is first argument, followed by actual args.
1267 code = fuword(params);
1268 params += sizeof(int);
1269 } else if (code == SYS___syscall) {
1271 * Like syscall, but code is a quad, so as to maintain
1272 * quad alignment for the rest of the arguments.
1274 code = fuword(params);
1275 params += sizeof(quad_t);
1279 code &= p->p_sysent->sv_mask;
1280 if (code >= p->p_sysent->sv_size)
1281 callp = &p->p_sysent->sv_table[0];
1283 callp = &p->p_sysent->sv_table[code];
1285 narg = callp->sy_narg & SYF_ARGMASK;
1288 * copyin is MP aware, but the tracing code is not
1290 if (narg && params) {
1291 error = copyin(params, (caddr_t)(&args.nosys.sysmsg + 1),
1292 narg * sizeof(register_t));
1295 if (KTRPOINT(td, KTR_SYSCALL)) {
1296 MAKEMPSAFE(have_mplock);
1298 ktrsyscall(lp, code, narg,
1299 (void *)(&args.nosys.sysmsg + 1));
1307 if (KTRPOINT(td, KTR_SYSCALL)) {
1308 MAKEMPSAFE(have_mplock);
1309 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1314 * For traditional syscall code edx is left untouched when 32 bit
1315 * results are returned. Since edx is loaded from fds[1] when the
1316 * system call returns we pre-set it here.
1318 args.sysmsg_fds[0] = 0;
1319 args.sysmsg_fds[1] = frame->tf_edx;
1322 * The syscall might manipulate the trap frame. If it does it
1323 * will probably return EJUSTRETURN.
1325 args.sysmsg_frame = frame;
1327 STOPEVENT(p, S_SCE, narg); /* MP aware */
1331 * Try to run the syscall without the MP lock if the syscall
1332 * is MP safe. We have to obtain the MP lock no matter what if
1335 if ((callp->sy_narg & SYF_MPSAFE) == 0)
1336 MAKEMPSAFE(have_mplock);
1339 error = (*callp->sy_call)(&args);
1343 * MP SAFE (we may or may not have the MP lock at this point)
1348 * Reinitialize proc pointer `p' as it may be different
1349 * if this is a child returning from fork syscall.
1352 lp = curthread->td_lwp;
1353 frame->tf_eax = args.sysmsg_fds[0];
1354 frame->tf_edx = args.sysmsg_fds[1];
1355 frame->tf_eflags &= ~PSL_C;
1359 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1360 * int 0x80 is 2 bytes. We saved this in tf_err.
1362 frame->tf_eip -= frame->tf_err;
1367 panic("Unexpected EASYNC return value (for now)");
1370 if (p->p_sysent->sv_errsize) {
1371 if (error >= p->p_sysent->sv_errsize)
1372 error = -1; /* XXX */
1374 error = p->p_sysent->sv_errtbl[error];
1376 frame->tf_eax = error;
1377 frame->tf_eflags |= PSL_C;
1382 * Traced syscall. trapsignal() is not MP aware.
1384 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1385 MAKEMPSAFE(have_mplock);
1386 frame->tf_eflags &= ~PSL_T;
1387 trapsignal(lp, SIGTRAP, 0);
1391 * Handle reschedule and other end-of-syscall issues
1393 userret(lp, frame, sticks);
1396 if (KTRPOINT(td, KTR_SYSRET)) {
1397 MAKEMPSAFE(have_mplock);
1398 ktrsysret(lp, code, error, args.sysmsg_result);
1403 * This works because errno is findable through the
1404 * register set. If we ever support an emulation where this
1405 * is not the case, this code will need to be revisited.
1407 STOPEVENT(p, S_SCX, code);
1412 * Release the MP lock if we had to get it
1414 KASSERT(td->td_mpcount == have_mplock,
1415 ("badmpcount syscall2/end from %p", (void *)frame->tf_eip));
1419 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error);
1421 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
1422 ("syscall: critical section count mismatch! %d/%d",
1423 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
1428 fork_return(struct lwp *lp, struct trapframe *frame)
1430 frame->tf_eax = 0; /* Child returns zero */
1431 frame->tf_eflags &= ~PSL_C; /* success */
1434 generic_lwp_return(lp, frame);
1435 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1439 * Simplified back end of syscall(), used when returning from fork()
1440 * directly into user mode. MP lock is held on entry and should be
1441 * released on return. This code will return back into the fork
1442 * trampoline code which then runs doreti.
1445 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1447 struct proc *p = lp->lwp_proc;
1450 * Newly forked processes are given a kernel priority. We have to
1451 * adjust the priority to a normal user priority and fake entry
1452 * into the kernel (call userenter()) to install a passive release
1453 * function just in case userret() decides to stop the process. This
1454 * can occur when ^Z races a fork. If we do not install the passive
1455 * release function the current process designation will not be
1456 * released when the thread goes to sleep.
1458 lwkt_setpri_self(TDPRI_USER_NORM);
1459 userenter(lp->lwp_thread);
1460 userret(lp, frame, 0);
1462 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1463 ktrsysret(lp, SYS_fork, 0, 0);
1465 p->p_flag |= P_PASSIVE_ACQ;
1467 p->p_flag &= ~P_PASSIVE_ACQ;
1469 KKASSERT(lp->lwp_thread->td_mpcount == 1);
1475 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1476 * fault (which is then passed back to the virtual kernel) if an attempt is
1477 * made to use the FP unit.
1479 * XXX this is a fairly big hack.
1482 set_vkernel_fp(struct trapframe *frame)
1484 struct thread *td = curthread;
1486 if (frame->tf_xflags & PGEX_FPFAULT) {
1487 td->td_pcb->pcb_flags |= FP_VIRTFP;
1488 if (mdcpu->gd_npxthread == td)
1491 td->td_pcb->pcb_flags &= ~FP_VIRTFP;