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>
106 #include <sys/msgport2.h>
107 #include <sys/thread2.h>
108 #include <sys/mplock2.h>
112 #define MAKEMPSAFE(have_mplock) \
113 if (have_mplock == 0) { \
120 #define MAKEMPSAFE(have_mplock)
124 int (*pmath_emulate) (struct trapframe *);
126 extern void trap (struct trapframe *frame);
127 extern void syscall2 (struct trapframe *frame);
129 static int trap_pfault (struct trapframe *, int, vm_offset_t);
130 static void trap_fatal (struct trapframe *, vm_offset_t);
131 void dblfault_handler (void);
133 extern inthand_t IDTVEC(syscall);
135 #define MAX_TRAP_MSG 28
136 static char *trap_msg[] = {
138 "privileged instruction fault", /* 1 T_PRIVINFLT */
140 "breakpoint instruction fault", /* 3 T_BPTFLT */
143 "arithmetic trap", /* 6 T_ARITHTRAP */
144 "system forced exception", /* 7 T_ASTFLT */
146 "general protection fault", /* 9 T_PROTFLT */
147 "trace trap", /* 10 T_TRCTRAP */
149 "page fault", /* 12 T_PAGEFLT */
151 "alignment fault", /* 14 T_ALIGNFLT */
155 "integer divide fault", /* 18 T_DIVIDE */
156 "non-maskable interrupt trap", /* 19 T_NMI */
157 "overflow trap", /* 20 T_OFLOW */
158 "FPU bounds check fault", /* 21 T_BOUND */
159 "FPU device not available", /* 22 T_DNA */
160 "double fault", /* 23 T_DOUBLEFLT */
161 "FPU operand fetch fault", /* 24 T_FPOPFLT */
162 "invalid TSS fault", /* 25 T_TSSFLT */
163 "segment not present fault", /* 26 T_SEGNPFLT */
164 "stack fault", /* 27 T_STKFLT */
165 "machine check trap", /* 28 T_MCHK */
168 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
169 extern int has_f00f_bug;
173 static int ddb_on_nmi = 1;
174 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
175 &ddb_on_nmi, 0, "Go to DDB on NMI");
177 static int panic_on_nmi = 1;
178 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
179 &panic_on_nmi, 0, "Panic on NMI");
180 static int fast_release;
181 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
182 &fast_release, 0, "Passive Release was optimal");
183 static int slow_release;
184 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
185 &slow_release, 0, "Passive Release was nonoptimal");
187 MALLOC_DEFINE(M_SYSMSG, "sysmsg", "sysmsg structure");
188 extern int max_sysmsg;
191 * Passively intercepts the thread switch function to increase the thread
192 * priority from a user priority to a kernel priority, reducing
193 * syscall and trap overhead for the case where no switch occurs.
195 * Synchronizes td_ucred with p_ucred. This is used by system calls,
196 * signal handling, faults, AST traps, and anything else that enters the
197 * kernel from userland and provides the kernel with a stable read-only
198 * copy of the process ucred.
201 userenter(struct thread *curtd, struct proc *curp)
206 curtd->td_release = lwkt_passive_release;
208 if (curtd->td_ucred != curp->p_ucred) {
209 ncred = crhold(curp->p_ucred);
210 ocred = curtd->td_ucred;
211 curtd->td_ucred = ncred;
219 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
220 * must be completed before we can return to or try to return to userland.
222 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
223 * arithmatic on the delta calculation so the absolute tick values are
224 * truncated to an integer.
227 userret(struct lwp *lp, struct trapframe *frame, int sticks)
229 struct proc *p = lp->lwp_proc;
233 if (p->p_userret != NULL) {
240 * Charge system time if profiling. Note: times are in microseconds.
241 * This may do a copyout and block, so do it first even though it
242 * means some system time will be charged as user time.
244 if (p->p_flag & P_PROFIL) {
245 addupc_task(p, frame->tf_eip,
246 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
251 * If the jungle wants us dead, so be it.
253 if (lp->lwp_flag & LWP_WEXIT) {
256 rel_mplock(); /* NOT REACHED */
260 * Block here if we are in a stopped state.
262 if (p->p_stat == SSTOP) {
270 * Post any pending upcalls. If running a virtual kernel be sure
271 * to restore the virtual kernel's vmspace before posting the upcall.
273 if (p->p_flag & P_UPCALLPEND) {
274 p->p_flag &= ~P_UPCALLPEND;
282 * Post any pending signals. If running a virtual kernel be sure
283 * to restore the virtual kernel's vmspace before posting the signal.
285 if ((sig = CURSIG_TRACE(lp)) != 0) {
293 * block here if we are swapped out, but still process signals
294 * (such as SIGKILL). proc0 (the swapin scheduler) is already
295 * aware of our situation, we do not have to wake it up.
297 if (p->p_flag & P_SWAPPEDOUT) {
299 p->p_flag |= P_SWAPWAIT;
301 if (p->p_flag & P_SWAPWAIT)
302 tsleep(p, PCATCH, "SWOUT", 0);
303 p->p_flag &= ~P_SWAPWAIT;
309 * Make sure postsig() handled request to restore old signal mask after
310 * running signal handler.
312 KKASSERT((lp->lwp_flag & LWP_OLDMASK) == 0);
316 * Cleanup from userenter and any passive release that might have occured.
317 * We must reclaim the current-process designation before we can return
318 * to usermode. We also handle both LWKT and USER reschedule requests.
321 userexit(struct lwp *lp)
323 struct thread *td = lp->lwp_thread;
324 /* globaldata_t gd = td->td_gd; */
327 * Handle stop requests at kernel priority. Any requests queued
328 * after this loop will generate another AST.
330 while (lp->lwp_proc->p_stat == SSTOP) {
337 * Reduce our priority in preparation for a return to userland. If
338 * our passive release function was still in place, our priority was
339 * never raised and does not need to be reduced.
341 lwkt_passive_recover(td);
344 * Become the current user scheduled process if we aren't already,
345 * and deal with reschedule requests and other factors.
347 lp->lwp_proc->p_usched->acquire_curproc(lp);
348 /* WARNING: we may have migrated cpu's */
349 /* gd = td->td_gd; */
352 #if !defined(KTR_KERNENTRY)
353 #define KTR_KERNENTRY KTR_ALL
355 KTR_INFO_MASTER(kernentry);
356 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, "pid=%d, tid=%d, trapno=%d, eva=%p",
357 sizeof(int) + sizeof(int) + sizeof(int) + sizeof(vm_offset_t));
358 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "pid=%d, tid=%d",
359 sizeof(int) + sizeof(int));
360 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "pid=%d, tid=%d, call=%d",
361 sizeof(int) + sizeof(int) + sizeof(int));
362 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "pid=%d, tid=%d, err=%d",
363 sizeof(int) + sizeof(int) + sizeof(int));
364 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "pid=%d, tid=%d",
365 sizeof(int) + sizeof(int));
368 * Exception, fault, and trap interface to the kernel.
369 * This common code is called from assembly language IDT gate entry
370 * routines that prepare a suitable stack frame, and restore this
371 * frame after the exception has been processed.
373 * This function is also called from doreti in an interlock to handle ASTs.
374 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
376 * NOTE! We have to retrieve the fault address prior to obtaining the
377 * MP lock because get_mplock() may switch out. YYY cr2 really ought
378 * to be retrieved by the assembly code, not here.
380 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
381 * if an attempt is made to switch from a fast interrupt or IPI. This is
382 * necessary to properly take fatal kernel traps on SMP machines if
383 * get_mplock() has to block.
387 trap(struct trapframe *frame)
389 struct globaldata *gd = mycpu;
390 struct thread *td = gd->gd_curthread;
391 struct lwp *lp = td->td_lwp;
394 int i = 0, ucode = 0, type, code;
399 int crit_count = td->td_critcount;
400 lwkt_tokref_t curstop = td->td_toks_stop;
407 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
408 ++gd->gd_trap_nesting_level;
409 MAKEMPSAFE(have_mplock);
410 trap_fatal(frame, eva);
411 --gd->gd_trap_nesting_level;
417 ++gd->gd_trap_nesting_level;
418 if (frame->tf_trapno == T_PAGEFLT) {
420 * For some Cyrix CPUs, %cr2 is clobbered by interrupts.
421 * This problem is worked around by using an interrupt
422 * gate for the pagefault handler. We are finally ready
423 * to read %cr2 and then must reenable interrupts.
425 * XXX this should be in the switch statement, but the
426 * NO_FOOF_HACK and VM86 goto and ifdefs obfuscate the
427 * flow of control too much for this to be obviously
434 --gd->gd_trap_nesting_level;
436 if (!(frame->tf_eflags & PSL_I)) {
438 * Buggy application or kernel code has disabled interrupts
439 * and then trapped. Enabling interrupts now is wrong, but
440 * it is better than running with interrupts disabled until
441 * they are accidentally enabled later.
443 type = frame->tf_trapno;
444 if (ISPL(frame->tf_cs)==SEL_UPL || (frame->tf_eflags & PSL_VM)) {
445 MAKEMPSAFE(have_mplock);
447 "pid %ld (%s): trap %d with interrupts disabled\n",
448 (long)curproc->p_pid, curproc->p_comm, type);
449 } else if (type != T_BPTFLT && type != T_TRCTRAP) {
451 * XXX not quite right, since this may be for a
452 * multiple fault in user mode.
454 MAKEMPSAFE(have_mplock);
455 kprintf("kernel trap %d with interrupts disabled\n",
461 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
464 type = frame->tf_trapno;
465 code = frame->tf_err;
468 if (frame->tf_eflags & PSL_VM &&
469 (type == T_PROTFLT || type == T_STKFLT)) {
471 KKASSERT(td->td_mpcount > 0);
473 i = vm86_emulate((struct vm86frame *)frame);
475 KKASSERT(td->td_mpcount > 0);
479 * returns to original process
482 vm86_trap((struct vm86frame *)frame,
485 vm86_trap((struct vm86frame *)frame, 0);
487 KKASSERT(0); /* NOT REACHED */
493 * these traps want either a process context, or
494 * assume a normal userspace trap.
498 trap_fatal(frame, eva);
501 type = T_BPTFLT; /* kernel breakpoint */
504 goto kernel_trap; /* normal kernel trap handling */
507 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
510 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid,
511 frame->tf_trapno, eva);
515 sticks = (int)td->td_sticks;
516 lp->lwp_md.md_regs = frame;
519 case T_PRIVINFLT: /* privileged instruction fault */
524 case T_BPTFLT: /* bpt instruction fault */
525 case T_TRCTRAP: /* trace trap */
526 frame->tf_eflags &= ~PSL_T;
531 case T_ARITHTRAP: /* arithmetic trap */
536 case T_ASTFLT: /* Allow process switch */
537 mycpu->gd_cnt.v_soft++;
538 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
539 atomic_clear_int_nonlocked(&mycpu->gd_reqflags,
541 addupc_task(p, p->p_prof.pr_addr,
547 * The following two traps can happen in
548 * vm86 mode, and, if so, we want to handle
551 case T_PROTFLT: /* general protection fault */
552 case T_STKFLT: /* stack fault */
553 if (frame->tf_eflags & PSL_VM) {
554 i = vm86_emulate((struct vm86frame *)frame);
560 ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR;
562 case T_SEGNPFLT: /* segment not present fault */
566 case T_TSSFLT: /* invalid TSS fault */
567 case T_DOUBLEFLT: /* double fault */
572 ucode = code + BUS_SEGM_FAULT ; /* XXX: ???*/
578 case T_PAGEFLT: /* page fault */
579 MAKEMPSAFE(have_mplock);
580 i = trap_pfault(frame, TRUE, eva);
583 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
595 ucode = BUS_ADRERR; /* XXX */
598 case T_DIVIDE: /* integer divide fault */
605 MAKEMPSAFE(have_mplock);
607 goto handle_powerfail;
608 #else /* !POWERFAIL_NMI */
609 /* machine/parity/power fail/"kitchen sink" faults */
610 if (isa_nmi(code) == 0) {
613 * NMI can be hooked up to a pushbutton
617 kprintf ("NMI ... going to debugger\n");
618 kdb_trap (type, 0, frame);
622 } else if (panic_on_nmi)
623 panic("NMI indicates hardware failure");
625 #endif /* POWERFAIL_NMI */
626 #endif /* NISA > 0 */
628 case T_OFLOW: /* integer overflow fault */
633 case T_BOUND: /* bounds check fault */
640 * Virtual kernel intercept - pass the DNA exception
641 * to the virtual kernel if it asked to handle it.
642 * This occurs when the virtual kernel is holding
643 * onto the FP context for a different emulated
644 * process then the one currently running.
646 * We must still call npxdna() since we may have
647 * saved FP state that the virtual kernel needs
648 * to hand over to a different emulated process.
650 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
651 (td->td_pcb->pcb_flags & FP_VIRTFP)
659 * The kernel may have switched out the FP unit's
660 * state, causing the user process to take a fault
661 * when it tries to use the FP unit. Restore the
667 if (!pmath_emulate) {
669 ucode = FPE_FPU_NP_TRAP;
672 i = (*pmath_emulate)(frame);
674 if (!(frame->tf_eflags & PSL_T))
676 frame->tf_eflags &= ~PSL_T;
679 /* else ucode = emulator_only_knows() XXX */
682 case T_FPOPFLT: /* FPU operand fetch fault */
687 case T_XMMFLT: /* SIMD floating-point exception */
697 case T_PAGEFLT: /* page fault */
698 MAKEMPSAFE(have_mplock);
699 trap_pfault(frame, FALSE, eva);
705 * The kernel may be using npx for copying or other
713 case T_PROTFLT: /* general protection fault */
714 case T_SEGNPFLT: /* segment not present fault */
716 * Invalid segment selectors and out of bounds
717 * %eip's and %esp's can be set up in user mode.
718 * This causes a fault in kernel mode when the
719 * kernel tries to return to user mode. We want
720 * to get this fault so that we can fix the
721 * problem here and not have to check all the
722 * selectors and pointers when the user changes
725 #define MAYBE_DORETI_FAULT(where, whereto) \
727 if (frame->tf_eip == (int)where) { \
728 frame->tf_eip = (int)whereto; \
732 if (mycpu->gd_intr_nesting_level == 0) {
734 * Invalid %fs's and %gs's can be created using
735 * procfs or PT_SETREGS or by invalidating the
736 * underlying LDT entry. This causes a fault
737 * in kernel mode when the kernel attempts to
738 * switch contexts. Lose the bad context
739 * (XXX) so that we can continue, and generate
742 MAYBE_DORETI_FAULT(doreti_iret,
744 MAYBE_DORETI_FAULT(doreti_popl_ds,
745 doreti_popl_ds_fault);
746 MAYBE_DORETI_FAULT(doreti_popl_es,
747 doreti_popl_es_fault);
748 MAYBE_DORETI_FAULT(doreti_popl_fs,
749 doreti_popl_fs_fault);
750 MAYBE_DORETI_FAULT(doreti_popl_gs,
751 doreti_popl_gs_fault);
752 if (td->td_pcb->pcb_onfault) {
754 (register_t)td->td_pcb->pcb_onfault;
762 * PSL_NT can be set in user mode and isn't cleared
763 * automatically when the kernel is entered. This
764 * causes a TSS fault when the kernel attempts to
765 * `iret' because the TSS link is uninitialized. We
766 * want to get this fault so that we can fix the
767 * problem here and not every time the kernel is
770 if (frame->tf_eflags & PSL_NT) {
771 frame->tf_eflags &= ~PSL_NT;
776 case T_TRCTRAP: /* trace trap */
777 if (frame->tf_eip == (int)IDTVEC(syscall)) {
779 * We've just entered system mode via the
780 * syscall lcall. Continue single stepping
781 * silently until the syscall handler has
786 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
788 * The syscall handler has now saved the
789 * flags. Stop single stepping it.
791 frame->tf_eflags &= ~PSL_T;
795 * Ignore debug register trace traps due to
796 * accesses in the user's address space, which
797 * can happen under several conditions such as
798 * if a user sets a watchpoint on a buffer and
799 * then passes that buffer to a system call.
800 * We still want to get TRCTRAPS for addresses
801 * in kernel space because that is useful when
802 * debugging the kernel.
804 if (user_dbreg_trap()) {
806 * Reset breakpoint bits because the
809 load_dr6(rdr6() & 0xfffffff0);
813 * Fall through (TRCTRAP kernel mode, kernel address)
817 * If DDB is enabled, let it handle the debugger trap.
818 * Otherwise, debugger traps "can't happen".
822 MAKEMPSAFE(have_mplock);
823 if (kdb_trap (type, 0, frame))
830 MAKEMPSAFE(have_mplock);
833 # define TIMER_FREQ 1193182
837 static unsigned lastalert = 0;
839 if(time_second - lastalert > 10)
841 log(LOG_WARNING, "NMI: power fail\n");
842 sysbeep(TIMER_FREQ/880, hz);
843 lastalert = time_second;
848 #else /* !POWERFAIL_NMI */
849 /* machine/parity/power fail/"kitchen sink" faults */
850 if (isa_nmi(code) == 0) {
853 * NMI can be hooked up to a pushbutton
857 kprintf ("NMI ... going to debugger\n");
858 kdb_trap (type, 0, frame);
862 } else if (panic_on_nmi == 0)
865 #endif /* POWERFAIL_NMI */
866 #endif /* NISA > 0 */
869 MAKEMPSAFE(have_mplock);
870 trap_fatal(frame, eva);
875 * Virtual kernel intercept - if the fault is directly related to a
876 * VM context managed by a virtual kernel then let the virtual kernel
879 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
880 vkernel_trap(lp, frame);
885 * Translate fault for emulators (e.g. Linux)
887 if (*p->p_sysent->sv_transtrap)
888 i = (*p->p_sysent->sv_transtrap)(i, type);
890 MAKEMPSAFE(have_mplock);
891 trapsignal(lp, i, ucode);
894 if (type <= MAX_TRAP_MSG) {
895 uprintf("fatal process exception: %s",
897 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
898 uprintf(", fault VA = 0x%lx", (u_long)eva);
905 if (ISPL(frame->tf_cs) == SEL_UPL) {
906 KASSERT(td->td_mpcount == have_mplock,
907 ("badmpcount trap/end from %p", (void *)frame->tf_eip));
910 userret(lp, frame, sticks);
917 if (p != NULL && lp != NULL)
918 KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid);
920 KASSERT(crit_count == td->td_critcount,
921 ("trap: critical section count mismatch! %d/%d",
922 crit_count, td->td_pri));
923 KASSERT(curstop == td->td_toks_stop,
924 ("trap: extra tokens held after trap! %zd/%zd",
925 curstop - &td->td_toks_base,
926 td->td_toks_stop - &td->td_toks_base));
931 trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
934 struct vmspace *vm = NULL;
939 thread_t td = curthread;
940 struct lwp *lp = td->td_lwp;
942 va = trunc_page(eva);
943 if (va >= KERNBASE) {
945 * Don't allow user-mode faults in kernel address space.
946 * An exception: if the faulting address is the invalid
947 * instruction entry in the IDT, then the Intel Pentium
948 * F00F bug workaround was triggered, and we need to
949 * treat it is as an illegal instruction, and not a page
952 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
953 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) {
954 frame->tf_trapno = T_PRIVINFLT;
964 * This is a fault on non-kernel virtual memory.
965 * vm is initialized above to NULL. If curproc is NULL
966 * or curproc->p_vmspace is NULL the fault is fatal.
969 vm = lp->lwp_vmspace;
977 if (frame->tf_err & PGEX_W)
978 ftype = VM_PROT_WRITE;
980 ftype = VM_PROT_READ;
982 if (map != &kernel_map) {
984 * Keep swapout from messing with us during this
994 fault_flags |= VM_FAULT_BURST;
995 if (ftype & VM_PROT_WRITE)
996 fault_flags |= VM_FAULT_DIRTY;
998 fault_flags |= VM_FAULT_NORMAL;
999 rv = vm_fault(map, va, ftype, fault_flags);
1000 PRELE(lp->lwp_proc);
1003 * Don't have to worry about process locking or stacks
1006 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
1009 if (rv == KERN_SUCCESS)
1013 if (td->td_gd->gd_intr_nesting_level == 0 &&
1014 td->td_pcb->pcb_onfault) {
1015 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
1018 trap_fatal(frame, eva);
1022 /* kludge to pass faulting virtual address to sendsig */
1023 frame->tf_xflags = frame->tf_err;
1024 frame->tf_err = eva;
1026 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
1030 trap_fatal(struct trapframe *frame, vm_offset_t eva)
1032 int code, type, ss, esp;
1033 struct soft_segment_descriptor softseg;
1035 code = frame->tf_err;
1036 type = frame->tf_trapno;
1037 sdtossd(&gdt[mycpu->gd_cpuid * NGDT + IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
1039 if (type <= MAX_TRAP_MSG)
1040 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
1041 type, trap_msg[type],
1042 frame->tf_eflags & PSL_VM ? "vm86" :
1043 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
1045 /* three separate prints in case of a trap on an unmapped page */
1046 kprintf("mp_lock = %08x; ", mp_lock);
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.
1135 dblfault_handler(void)
1137 struct mdglobaldata *gd = mdcpu;
1139 kprintf("\nFatal double fault:\n");
1140 kprintf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1141 kprintf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1142 kprintf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
1144 /* three separate prints in case of a trap on an unmapped page */
1145 kprintf("mp_lock = %08x; ", mp_lock);
1146 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
1147 kprintf("lapic.id = %08x\n", lapic.id);
1149 panic("double fault");
1153 * syscall2 - MP aware system call request C handler
1155 * A system call is essentially treated as a trap. The MP lock is not
1156 * held on entry or return. We are responsible for handling ASTs
1157 * (e.g. a task switch) prior to return.
1162 syscall2(struct trapframe *frame)
1164 struct thread *td = curthread;
1165 struct proc *p = td->td_proc;
1166 struct lwp *lp = td->td_lwp;
1168 struct sysent *callp;
1169 register_t orig_tf_eflags;
1174 int crit_count = td->td_critcount;
1177 int have_mplock = 0;
1180 union sysunion args;
1183 if (ISPL(frame->tf_cs) != SEL_UPL) {
1190 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid,
1194 KASSERT(td->td_mpcount == 0,
1195 ("badmpcount syscall2 from %p", (void *)frame->tf_eip));
1197 userenter(td, p); /* lazy raise our priority */
1202 sticks = (int)td->td_sticks;
1203 orig_tf_eflags = frame->tf_eflags;
1206 * Virtual kernel intercept - if a VM context managed by a virtual
1207 * kernel issues a system call the virtual kernel handles it, not us.
1208 * Restore the virtual kernel context and return from its system
1209 * call. The current frame is copied out to the virtual kernel.
1211 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1212 vkernel_trap(lp, frame);
1213 error = EJUSTRETURN;
1219 * Get the system call parameters and account for time
1221 lp->lwp_md.md_regs = frame;
1222 params = (caddr_t)frame->tf_esp + sizeof(int);
1223 code = frame->tf_eax;
1225 if (p->p_sysent->sv_prepsyscall) {
1226 (*p->p_sysent->sv_prepsyscall)(
1227 frame, (int *)(&args.nosys.sysmsg + 1),
1231 * Need to check if this is a 32 bit or 64 bit syscall.
1232 * fuword is MP aware.
1234 if (code == SYS_syscall) {
1236 * Code is first argument, followed by actual args.
1238 code = fuword(params);
1239 params += sizeof(int);
1240 } else if (code == SYS___syscall) {
1242 * Like syscall, but code is a quad, so as to maintain
1243 * quad alignment for the rest of the arguments.
1245 code = fuword(params);
1246 params += sizeof(quad_t);
1250 code &= p->p_sysent->sv_mask;
1252 if (code >= p->p_sysent->sv_size)
1253 callp = &p->p_sysent->sv_table[0];
1255 callp = &p->p_sysent->sv_table[code];
1257 narg = callp->sy_narg & SYF_ARGMASK;
1260 if (p->p_sysent->sv_name[0] == 'L')
1261 kprintf("Linux syscall, code = %d\n", code);
1265 * copyin is MP aware, but the tracing code is not
1267 if (narg && params) {
1268 error = copyin(params, (caddr_t)(&args.nosys.sysmsg + 1),
1269 narg * sizeof(register_t));
1272 if (KTRPOINT(td, KTR_SYSCALL)) {
1273 MAKEMPSAFE(have_mplock);
1275 ktrsyscall(lp, code, narg,
1276 (void *)(&args.nosys.sysmsg + 1));
1284 if (KTRPOINT(td, KTR_SYSCALL)) {
1285 MAKEMPSAFE(have_mplock);
1286 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1291 * For traditional syscall code edx is left untouched when 32 bit
1292 * results are returned. Since edx is loaded from fds[1] when the
1293 * system call returns we pre-set it here.
1295 args.sysmsg_fds[0] = 0;
1296 args.sysmsg_fds[1] = frame->tf_edx;
1299 * The syscall might manipulate the trap frame. If it does it
1300 * will probably return EJUSTRETURN.
1302 args.sysmsg_frame = frame;
1304 STOPEVENT(p, S_SCE, narg); /* MP aware */
1307 * NOTE: All system calls run MPSAFE now. The system call itself
1308 * is responsible for getting the MP lock.
1310 error = (*callp->sy_call)(&args);
1314 * MP SAFE (we may or may not have the MP lock at this point)
1319 * Reinitialize proc pointer `p' as it may be different
1320 * if this is a child returning from fork syscall.
1323 lp = curthread->td_lwp;
1324 frame->tf_eax = args.sysmsg_fds[0];
1325 frame->tf_edx = args.sysmsg_fds[1];
1326 frame->tf_eflags &= ~PSL_C;
1330 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1331 * int 0x80 is 2 bytes. We saved this in tf_err.
1333 frame->tf_eip -= frame->tf_err;
1338 panic("Unexpected EASYNC return value (for now)");
1341 if (p->p_sysent->sv_errsize) {
1342 if (error >= p->p_sysent->sv_errsize)
1343 error = -1; /* XXX */
1345 error = p->p_sysent->sv_errtbl[error];
1347 frame->tf_eax = error;
1348 frame->tf_eflags |= PSL_C;
1353 * Traced syscall. trapsignal() is not MP aware.
1355 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1356 MAKEMPSAFE(have_mplock);
1357 frame->tf_eflags &= ~PSL_T;
1358 trapsignal(lp, SIGTRAP, TRAP_TRACE);
1362 * Handle reschedule and other end-of-syscall issues
1364 userret(lp, frame, sticks);
1367 if (KTRPOINT(td, KTR_SYSRET)) {
1368 MAKEMPSAFE(have_mplock);
1369 ktrsysret(lp, code, error, args.sysmsg_result);
1374 * This works because errno is findable through the
1375 * register set. If we ever support an emulation where this
1376 * is not the case, this code will need to be revisited.
1378 STOPEVENT(p, S_SCX, code);
1383 * Release the MP lock if we had to get it
1385 KASSERT(td->td_mpcount == have_mplock,
1386 ("badmpcount syscall2/end from %p callp %p",
1387 (void *)frame->tf_eip, callp));
1391 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error);
1393 KASSERT(crit_count == td->td_critcount,
1394 ("syscall: critical section count mismatch! %d/%d",
1395 crit_count, td->td_pri));
1396 KASSERT(&td->td_toks_base == td->td_toks_stop,
1397 ("syscall: extra tokens held after trap! %zd",
1398 td->td_toks_stop - &td->td_toks_base));
1403 * NOTE: MP lock not held at any point.
1406 fork_return(struct lwp *lp, struct trapframe *frame)
1408 frame->tf_eax = 0; /* Child returns zero */
1409 frame->tf_eflags &= ~PSL_C; /* success */
1412 generic_lwp_return(lp, frame);
1413 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1417 * Simplified back end of syscall(), used when returning from fork()
1418 * directly into user mode.
1420 * This code will return back into the fork trampoline code which then
1423 * NOTE: The mplock is not held at any point.
1426 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1428 struct proc *p = lp->lwp_proc;
1431 * Newly forked processes are given a kernel priority. We have to
1432 * adjust the priority to a normal user priority and fake entry
1433 * into the kernel (call userenter()) to install a passive release
1434 * function just in case userret() decides to stop the process. This
1435 * can occur when ^Z races a fork. If we do not install the passive
1436 * release function the current process designation will not be
1437 * released when the thread goes to sleep.
1439 lwkt_setpri_self(TDPRI_USER_NORM);
1440 userenter(lp->lwp_thread, p);
1441 userret(lp, frame, 0);
1443 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1444 ktrsysret(lp, SYS_fork, 0, 0);
1446 p->p_flag |= P_PASSIVE_ACQ;
1448 p->p_flag &= ~P_PASSIVE_ACQ;
1452 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1453 * fault (which is then passed back to the virtual kernel) if an attempt is
1454 * made to use the FP unit.
1456 * XXX this is a fairly big hack.
1459 set_vkernel_fp(struct trapframe *frame)
1461 struct thread *td = curthread;
1463 if (frame->tf_xflags & PGEX_FPFAULT) {
1464 td->td_pcb->pcb_flags |= FP_VIRTFP;
1465 if (mdcpu->gd_npxthread == td)
1468 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1473 * Called from vkernel_trap() to fixup the vkernel's syscall
1474 * frame for vmspace_ctl() return.
1477 cpu_vkernel_trap(struct trapframe *frame, int error)
1479 frame->tf_eax = error;
1481 frame->tf_eflags |= PSL_C;
1483 frame->tf_eflags &= ~PSL_C;