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 1.95 2007/01/14 21:07:12 dillon Exp $
43 * 386 Trap and System call handling
51 #include "opt_ktrace.h"
52 #include "opt_clock.h"
55 #include <sys/param.h>
56 #include <sys/systm.h>
58 #include <sys/pioctl.h>
59 #include <sys/kernel.h>
60 #include <sys/resourcevar.h>
61 #include <sys/signalvar.h>
62 #include <sys/syscall.h>
63 #include <sys/sysctl.h>
64 #include <sys/sysent.h>
66 #include <sys/vmmeter.h>
67 #include <sys/malloc.h>
69 #include <sys/ktrace.h>
71 #include <sys/upcall.h>
72 #include <sys/vkernel.h>
73 #include <sys/sysproto.h>
74 #include <sys/sysunion.h>
77 #include <vm/vm_param.h>
80 #include <vm/vm_kern.h>
81 #include <vm/vm_map.h>
82 #include <vm/vm_page.h>
83 #include <vm/vm_extern.h>
85 #include <machine/cpu.h>
86 #include <machine/md_var.h>
87 #include <machine/pcb.h>
88 #include <machine/smp.h>
89 #include <machine/tss.h>
90 #include <machine/specialreg.h>
91 #include <machine/globaldata.h>
93 #include <machine_base/isa/intr_machdep.h>
96 #include <sys/syslog.h>
97 #include <machine/clock.h>
100 #include <machine/vm86.h>
103 #include <sys/msgport2.h>
104 #include <sys/thread2.h>
108 #define MAKEMPSAFE(have_mplock) \
109 if (have_mplock == 0) { \
116 #define MAKEMPSAFE(have_mplock)
120 int (*pmath_emulate
) (struct trapframe
*);
122 extern void trap (struct trapframe frame
);
123 extern int trapwrite (unsigned addr
);
124 extern void syscall2 (struct trapframe frame
);
126 static int trap_pfault (struct trapframe
*, int, vm_offset_t
);
127 static void trap_fatal (struct trapframe
*, vm_offset_t
);
128 void dblfault_handler (void);
130 extern inthand_t
IDTVEC(syscall
);
132 #define MAX_TRAP_MSG 28
133 static char *trap_msg
[] = {
135 "privileged instruction fault", /* 1 T_PRIVINFLT */
137 "breakpoint instruction fault", /* 3 T_BPTFLT */
140 "arithmetic trap", /* 6 T_ARITHTRAP */
141 "system forced exception", /* 7 T_ASTFLT */
143 "general protection fault", /* 9 T_PROTFLT */
144 "trace trap", /* 10 T_TRCTRAP */
146 "page fault", /* 12 T_PAGEFLT */
148 "alignment fault", /* 14 T_ALIGNFLT */
152 "integer divide fault", /* 18 T_DIVIDE */
153 "non-maskable interrupt trap", /* 19 T_NMI */
154 "overflow trap", /* 20 T_OFLOW */
155 "FPU bounds check fault", /* 21 T_BOUND */
156 "FPU device not available", /* 22 T_DNA */
157 "double fault", /* 23 T_DOUBLEFLT */
158 "FPU operand fetch fault", /* 24 T_FPOPFLT */
159 "invalid TSS fault", /* 25 T_TSSFLT */
160 "segment not present fault", /* 26 T_SEGNPFLT */
161 "stack fault", /* 27 T_STKFLT */
162 "machine check trap", /* 28 T_MCHK */
165 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
166 extern int has_f00f_bug
;
170 static int ddb_on_nmi
= 1;
171 SYSCTL_INT(_machdep
, OID_AUTO
, ddb_on_nmi
, CTLFLAG_RW
,
172 &ddb_on_nmi
, 0, "Go to DDB on NMI");
174 static int panic_on_nmi
= 1;
175 SYSCTL_INT(_machdep
, OID_AUTO
, panic_on_nmi
, CTLFLAG_RW
,
176 &panic_on_nmi
, 0, "Panic on NMI");
177 static int fast_release
;
178 SYSCTL_INT(_machdep
, OID_AUTO
, fast_release
, CTLFLAG_RW
,
179 &fast_release
, 0, "Passive Release was optimal");
180 static int slow_release
;
181 SYSCTL_INT(_machdep
, OID_AUTO
, slow_release
, CTLFLAG_RW
,
182 &slow_release
, 0, "Passive Release was nonoptimal");
184 static int syscall_mpsafe
= 0;
185 SYSCTL_INT(_kern
, OID_AUTO
, syscall_mpsafe
, CTLFLAG_RW
,
186 &syscall_mpsafe
, 0, "Allow MPSAFE marked syscalls to run without BGL");
187 TUNABLE_INT("kern.syscall_mpsafe", &syscall_mpsafe
);
188 static int trap_mpsafe
= 0;
189 SYSCTL_INT(_kern
, OID_AUTO
, trap_mpsafe
, CTLFLAG_RW
,
190 &trap_mpsafe
, 0, "Allow traps to mostly run without the BGL");
191 TUNABLE_INT("kern.trap_mpsafe", &trap_mpsafe
);
194 MALLOC_DEFINE(M_SYSMSG
, "sysmsg", "sysmsg structure");
195 extern int max_sysmsg
;
198 * Passive USER->KERNEL transition. This only occurs if we block in the
199 * kernel while still holding our userland priority. We have to fixup our
200 * priority in order to avoid potential deadlocks before we allow the system
201 * to switch us to another thread.
204 passive_release(struct thread
*td
)
206 struct lwp
*lp
= td
->td_lwp
;
208 td
->td_release
= NULL
;
209 lwkt_setpri_self(TDPRI_KERN_USER
);
210 lp
->lwp_proc
->p_usched
->release_curproc(lp
);
214 * userenter() passively intercepts the thread switch function to increase
215 * the thread priority from a user priority to a kernel priority, reducing
216 * syscall and trap overhead for the case where no switch occurs.
220 userenter(struct thread
*curtd
)
222 curtd
->td_release
= passive_release
;
226 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
227 * must be completed before we can return to or try to return to userland.
229 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
230 * arithmatic on the delta calculation so the absolute tick values are
231 * truncated to an integer.
234 userret(struct lwp
*lp
, struct trapframe
*frame
, int sticks
)
236 struct proc
*p
= lp
->lwp_proc
;
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 * Block here if we are in a stopped state.
253 if (p
->p_flag
& P_STOPPED
) {
261 * Post any pending upcalls. If running a virtual kernel be sure
262 * to restore the virtual kernel's vmspace before posting the upcall.
264 if (p
->p_flag
& P_UPCALLPEND
) {
265 p
->p_flag
&= ~P_UPCALLPEND
;
273 * Post any pending signals. If running a virtual kernel be sure
274 * to restore the virtual kernel's vmspace before posting the signal.
276 if ((sig
= CURSIG(p
)) != 0) {
284 * block here if we are swapped out, but still process signals
285 * (such as SIGKILL). proc0 (the swapin scheduler) is already
286 * aware of our situation, we do not have to wake it up.
288 if (p
->p_flag
& P_SWAPPEDOUT
) {
290 p
->p_flag
|= P_SWAPWAIT
;
292 if (p
->p_flag
& P_SWAPWAIT
)
293 tsleep(p
, PCATCH
, "SWOUT", 0);
294 p
->p_flag
&= ~P_SWAPWAIT
;
301 * Cleanup from userenter and any passive release that might have occured.
302 * We must reclaim the current-process designation before we can return
303 * to usermode. We also handle both LWKT and USER reschedule requests.
306 userexit(struct lwp
*lp
)
308 struct thread
*td
= lp
->lwp_thread
;
309 globaldata_t gd
= td
->td_gd
;
313 * If a user reschedule is requested force a new process to be
314 * chosen by releasing the current process. Our process will only
315 * be chosen again if it has a considerably better priority.
317 if (user_resched_wanted())
318 lp
->lwp_proc
->p_usched
->release_curproc(lp
);
322 * Handle a LWKT reschedule request first. Since our passive release
323 * is still in place we do not have to do anything special.
325 if (lwkt_resched_wanted())
329 * Acquire the current process designation for this user scheduler
330 * on this cpu. This will also handle any user-reschedule requests.
332 lp
->lwp_proc
->p_usched
->acquire_curproc(lp
);
333 /* We may have switched cpus on acquisition */
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 if (td
->td_release
== NULL
)
342 lwkt_setpri_self(TDPRI_USER_NORM
);
343 td
->td_release
= NULL
;
346 * After reducing our priority there might be other kernel-level
347 * LWKTs that now have a greater priority. Run them as necessary.
348 * We don't have to worry about losing cpu to userland because
349 * we still control the current-process designation and we no longer
350 * have a passive release function installed.
352 if (lwkt_checkpri_self())
357 * Exception, fault, and trap interface to the kernel.
358 * This common code is called from assembly language IDT gate entry
359 * routines that prepare a suitable stack frame, and restore this
360 * frame after the exception has been processed.
362 * This function is also called from doreti in an interlock to handle ASTs.
363 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
365 * NOTE! We have to retrieve the fault address prior to obtaining the
366 * MP lock because get_mplock() may switch out. YYY cr2 really ought
367 * to be retrieved by the assembly code, not here.
369 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
370 * if an attempt is made to switch from a fast interrupt or IPI. This is
371 * necessary to properly take fatal kernel traps on SMP machines if
372 * get_mplock() has to block.
376 trap(struct trapframe frame
)
378 struct globaldata
*gd
= mycpu
;
379 struct thread
*td
= gd
->gd_curthread
;
380 struct lwp
*lp
= td
->td_lwp
;
383 int i
= 0, ucode
= 0, type
, code
;
388 int crit_count
= td
->td_pri
& ~TDPRI_MASK
;
395 eva
= (frame
.tf_trapno
== T_PAGEFLT
? rcr2() : 0);
396 ++gd
->gd_trap_nesting_level
;
397 MAKEMPSAFE(have_mplock
);
398 trap_fatal(&frame
, eva
);
399 --gd
->gd_trap_nesting_level
;
405 ++gd
->gd_trap_nesting_level
;
406 if (frame
.tf_trapno
== T_PAGEFLT
) {
408 * For some Cyrix CPUs, %cr2 is clobbered by interrupts.
409 * This problem is worked around by using an interrupt
410 * gate for the pagefault handler. We are finally ready
411 * to read %cr2 and then must reenable interrupts.
413 * XXX this should be in the switch statement, but the
414 * NO_FOOF_HACK and VM86 goto and ifdefs obfuscate the
415 * flow of control too much for this to be obviously
422 if (trap_mpsafe
== 0)
423 MAKEMPSAFE(have_mplock
);
426 --gd
->gd_trap_nesting_level
;
428 if (!(frame
.tf_eflags
& PSL_I
)) {
430 * Buggy application or kernel code has disabled interrupts
431 * and then trapped. Enabling interrupts now is wrong, but
432 * it is better than running with interrupts disabled until
433 * they are accidentally enabled later.
435 type
= frame
.tf_trapno
;
436 if (ISPL(frame
.tf_cs
)==SEL_UPL
|| (frame
.tf_eflags
& PSL_VM
)) {
437 MAKEMPSAFE(have_mplock
);
439 "pid %ld (%s): trap %d with interrupts disabled\n",
440 (long)curproc
->p_pid
, curproc
->p_comm
, type
);
441 } else if (type
!= T_BPTFLT
&& type
!= T_TRCTRAP
) {
443 * XXX not quite right, since this may be for a
444 * multiple fault in user mode.
446 MAKEMPSAFE(have_mplock
);
447 kprintf("kernel trap %d with interrupts disabled\n",
453 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
456 type
= frame
.tf_trapno
;
460 ASSERT_MP_LOCK_HELD(curthread
);
461 if (frame
.tf_eflags
& PSL_VM
&&
462 (type
== T_PROTFLT
|| type
== T_STKFLT
)) {
464 KKASSERT(td
->td_mpcount
> 0);
466 i
= vm86_emulate((struct vm86frame
*)&frame
);
468 KKASSERT(td
->td_mpcount
> 0);
472 * returns to original process
475 vm86_trap((struct vm86frame
*)&frame
,
478 vm86_trap((struct vm86frame
*)&frame
, 0);
480 KKASSERT(0); /* NOT REACHED */
486 * these traps want either a process context, or
487 * assume a normal userspace trap.
491 trap_fatal(&frame
, eva
);
494 type
= T_BPTFLT
; /* kernel breakpoint */
497 goto kernel_trap
; /* normal kernel trap handling */
500 if ((ISPL(frame
.tf_cs
) == SEL_UPL
) || (frame
.tf_eflags
& PSL_VM
)) {
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 (p
->p_vkernel
&& p
->p_vkernel
->vk_current
&&
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 (p
->p_vkernel
&& p
->p_vkernel
->vk_current
) {
854 vkernel_trap(p
, &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(p
, 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
);
890 KASSERT(crit_count
== (td
->td_pri
& ~TDPRI_MASK
),
891 ("syscall: critical section count mismatch! %d/%d",
892 crit_count
/ TDPRI_CRIT
, td
->td_pri
/ TDPRI_CRIT
));
897 trap_pfault(struct trapframe
*frame
, int usermode
, vm_offset_t eva
)
900 struct vmspace
*vm
= NULL
;
904 thread_t td
= curthread
;
905 struct proc
*p
= td
->td_proc
;
907 va
= trunc_page(eva
);
908 if (va
>= KERNBASE
) {
910 * Don't allow user-mode faults in kernel address space.
911 * An exception: if the faulting address is the invalid
912 * instruction entry in the IDT, then the Intel Pentium
913 * F00F bug workaround was triggered, and we need to
914 * treat it is as an illegal instruction, and not a page
917 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
918 if ((eva
== (unsigned int)&idt
[6]) && has_f00f_bug
) {
919 frame
->tf_trapno
= T_PRIVINFLT
;
929 * This is a fault on non-kernel virtual memory.
930 * vm is initialized above to NULL. If curproc is NULL
931 * or curproc->p_vmspace is NULL the fault is fatal.
942 if (frame
->tf_err
& PGEX_W
)
943 ftype
= VM_PROT_WRITE
;
945 ftype
= VM_PROT_READ
;
947 if (map
!= &kernel_map
) {
949 * Keep swapout from messing with us during this
955 * Grow the stack if necessary
957 /* grow_stack returns false only if va falls into
958 * a growable stack region and the stack growth
959 * fails. It returns true if va was not within
960 * a growable stack region, or if the stack
963 if (!grow_stack (p
, va
)) {
969 /* Fault in the user page: */
970 rv
= vm_fault(map
, va
, ftype
,
971 (ftype
& VM_PROT_WRITE
) ? VM_FAULT_DIRTY
977 * Don't have to worry about process locking or stacks in the kernel.
979 rv
= vm_fault(map
, va
, ftype
, VM_FAULT_NORMAL
);
982 if (rv
== KERN_SUCCESS
)
986 if (td
->td_gd
->gd_intr_nesting_level
== 0 &&
987 td
->td_pcb
->pcb_onfault
) {
988 frame
->tf_eip
= (register_t
)td
->td_pcb
->pcb_onfault
;
991 trap_fatal(frame
, eva
);
995 /* kludge to pass faulting virtual address to sendsig */
996 frame
->tf_xflags
= frame
->tf_err
;
999 return((rv
== KERN_PROTECTION_FAILURE
) ? SIGBUS
: SIGSEGV
);
1003 trap_fatal(struct trapframe
*frame
, vm_offset_t eva
)
1005 int code
, type
, ss
, esp
;
1006 struct soft_segment_descriptor softseg
;
1008 code
= frame
->tf_err
;
1009 type
= frame
->tf_trapno
;
1010 sdtossd(&gdt
[mycpu
->gd_cpuid
* NGDT
+ IDXSEL(frame
->tf_cs
& 0xffff)].sd
, &softseg
);
1012 if (type
<= MAX_TRAP_MSG
)
1013 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
1014 type
, trap_msg
[type
],
1015 frame
->tf_eflags
& PSL_VM
? "vm86" :
1016 ISPL(frame
->tf_cs
) == SEL_UPL
? "user" : "kernel");
1018 /* three separate prints in case of a trap on an unmapped page */
1019 kprintf("mp_lock = %08x; ", mp_lock
);
1020 kprintf("cpuid = %d; ", mycpu
->gd_cpuid
);
1021 kprintf("lapic.id = %08x\n", lapic
.id
);
1023 if (type
== T_PAGEFLT
) {
1024 kprintf("fault virtual address = 0x%x\n", eva
);
1025 kprintf("fault code = %s %s, %s\n",
1026 code
& PGEX_U
? "user" : "supervisor",
1027 code
& PGEX_W
? "write" : "read",
1028 code
& PGEX_P
? "protection violation" : "page not present");
1030 kprintf("instruction pointer = 0x%x:0x%x\n",
1031 frame
->tf_cs
& 0xffff, frame
->tf_eip
);
1032 if ((ISPL(frame
->tf_cs
) == SEL_UPL
) || (frame
->tf_eflags
& PSL_VM
)) {
1033 ss
= frame
->tf_ss
& 0xffff;
1034 esp
= frame
->tf_esp
;
1036 ss
= GSEL(GDATA_SEL
, SEL_KPL
);
1037 esp
= (int)&frame
->tf_esp
;
1039 kprintf("stack pointer = 0x%x:0x%x\n", ss
, esp
);
1040 kprintf("frame pointer = 0x%x:0x%x\n", ss
, frame
->tf_ebp
);
1041 kprintf("code segment = base 0x%x, limit 0x%x, type 0x%x\n",
1042 softseg
.ssd_base
, softseg
.ssd_limit
, softseg
.ssd_type
);
1043 kprintf(" = DPL %d, pres %d, def32 %d, gran %d\n",
1044 softseg
.ssd_dpl
, softseg
.ssd_p
, softseg
.ssd_def32
,
1046 kprintf("processor eflags = ");
1047 if (frame
->tf_eflags
& PSL_T
)
1048 kprintf("trace trap, ");
1049 if (frame
->tf_eflags
& PSL_I
)
1050 kprintf("interrupt enabled, ");
1051 if (frame
->tf_eflags
& PSL_NT
)
1052 kprintf("nested task, ");
1053 if (frame
->tf_eflags
& PSL_RF
)
1054 kprintf("resume, ");
1055 if (frame
->tf_eflags
& PSL_VM
)
1057 kprintf("IOPL = %d\n", (frame
->tf_eflags
& PSL_IOPL
) >> 12);
1058 kprintf("current process = ");
1060 kprintf("%lu (%s)\n",
1061 (u_long
)curproc
->p_pid
, curproc
->p_comm
?
1062 curproc
->p_comm
: "");
1066 kprintf("current thread = pri %d ", curthread
->td_pri
);
1067 if (curthread
->td_pri
>= TDPRI_CRIT
)
1073 * we probably SHOULD have stopped the other CPUs before now!
1074 * another CPU COULD have been touching cpl at this moment...
1076 kprintf(" <- SMP: XXX");
1085 if ((debugger_on_panic
|| db_active
) && kdb_trap(type
, code
, frame
))
1088 kprintf("trap number = %d\n", type
);
1089 if (type
<= MAX_TRAP_MSG
)
1090 panic("%s", trap_msg
[type
]);
1092 panic("unknown/reserved trap");
1096 * Double fault handler. Called when a fault occurs while writing
1097 * a frame for a trap/exception onto the stack. This usually occurs
1098 * when the stack overflows (such is the case with infinite recursion,
1101 * XXX Note that the current PTD gets replaced by IdlePTD when the
1102 * task switch occurs. This means that the stack that was active at
1103 * the time of the double fault is not available at <kstack> unless
1104 * the machine was idle when the double fault occurred. The downside
1105 * of this is that "trace <ebp>" in ddb won't work.
1108 dblfault_handler(void)
1110 struct mdglobaldata
*gd
= mdcpu
;
1112 kprintf("\nFatal double fault:\n");
1113 kprintf("eip = 0x%x\n", gd
->gd_common_tss
.tss_eip
);
1114 kprintf("esp = 0x%x\n", gd
->gd_common_tss
.tss_esp
);
1115 kprintf("ebp = 0x%x\n", gd
->gd_common_tss
.tss_ebp
);
1117 /* three separate prints in case of a trap on an unmapped page */
1118 kprintf("mp_lock = %08x; ", mp_lock
);
1119 kprintf("cpuid = %d; ", mycpu
->gd_cpuid
);
1120 kprintf("lapic.id = %08x\n", lapic
.id
);
1122 panic("double fault");
1126 * Compensate for 386 brain damage (missing URKR).
1127 * This is a little simpler than the pagefault handler in trap() because
1128 * it the page tables have already been faulted in and high addresses
1129 * are thrown out early for other reasons.
1132 trapwrite(unsigned addr
)
1139 va
= trunc_page((vm_offset_t
)addr
);
1141 * XXX - MAX is END. Changed > to >= for temp. fix.
1143 if (va
>= VM_MAX_USER_ADDRESS
)
1151 if (!grow_stack (p
, va
)) {
1157 * fault the data page
1159 rv
= vm_fault(&vm
->vm_map
, va
, VM_PROT_WRITE
, VM_FAULT_DIRTY
);
1163 if (rv
!= KERN_SUCCESS
)
1170 * syscall2 - MP aware system call request C handler
1172 * A system call is essentially treated as a trap except that the
1173 * MP lock is not held on entry or return. We are responsible for
1174 * obtaining the MP lock if necessary and for handling ASTs
1175 * (e.g. a task switch) prior to return.
1177 * In general, only simple access and manipulation of curproc and
1178 * the current stack is allowed without having to hold MP lock.
1180 * MPSAFE - note that large sections of this routine are run without
1185 syscall2(struct trapframe frame
)
1187 struct thread
*td
= curthread
;
1188 struct proc
*p
= td
->td_proc
;
1189 struct lwp
*lp
= td
->td_lwp
;
1191 struct sysent
*callp
;
1192 register_t orig_tf_eflags
;
1197 int crit_count
= td
->td_pri
& ~TDPRI_MASK
;
1200 int have_mplock
= 0;
1203 union sysunion args
;
1206 if (ISPL(frame
.tf_cs
) != SEL_UPL
) {
1214 KASSERT(td
->td_mpcount
== 0, ("badmpcount syscall2 from %p", (void *)frame
.tf_eip
));
1215 if (syscall_mpsafe
== 0)
1216 MAKEMPSAFE(have_mplock
);
1218 userenter(td
); /* lazy raise our priority */
1223 sticks
= (int)td
->td_sticks
;
1224 orig_tf_eflags
= frame
.tf_eflags
;
1227 * Virtual kernel intercept - if a VM context managed by a virtual
1228 * kernel issues a system call the virtual kernel handles it, not us.
1229 * Restore the virtual kernel context and return from its system
1230 * call. The current frame is copied out to the virtual kernel.
1232 if (p
->p_vkernel
&& p
->p_vkernel
->vk_current
) {
1233 error
= vkernel_trap(p
, &frame
);
1234 frame
.tf_eax
= error
;
1236 frame
.tf_eflags
|= PSL_C
;
1237 error
= EJUSTRETURN
;
1242 * Get the system call parameters and account for time
1244 lp
->lwp_md
.md_regs
= &frame
;
1245 params
= (caddr_t
)frame
.tf_esp
+ sizeof(int);
1246 code
= frame
.tf_eax
;
1248 if (p
->p_sysent
->sv_prepsyscall
) {
1249 (*p
->p_sysent
->sv_prepsyscall
)(
1250 &frame
, (int *)(&args
.nosys
.sysmsg
+ 1),
1254 * Need to check if this is a 32 bit or 64 bit syscall.
1255 * fuword is MP aware.
1257 if (code
== SYS_syscall
) {
1259 * Code is first argument, followed by actual args.
1261 code
= fuword(params
);
1262 params
+= sizeof(int);
1263 } else if (code
== SYS___syscall
) {
1265 * Like syscall, but code is a quad, so as to maintain
1266 * quad alignment for the rest of the arguments.
1268 code
= fuword(params
);
1269 params
+= sizeof(quad_t
);
1273 code
&= p
->p_sysent
->sv_mask
;
1274 if (code
>= p
->p_sysent
->sv_size
)
1275 callp
= &p
->p_sysent
->sv_table
[0];
1277 callp
= &p
->p_sysent
->sv_table
[code
];
1279 narg
= callp
->sy_narg
& SYF_ARGMASK
;
1282 * copyin is MP aware, but the tracing code is not
1284 if (narg
&& params
) {
1285 error
= copyin(params
, (caddr_t
)(&args
.nosys
.sysmsg
+ 1),
1286 narg
* sizeof(register_t
));
1289 if (KTRPOINT(td
, KTR_SYSCALL
)) {
1290 MAKEMPSAFE(have_mplock
);
1292 ktrsyscall(p
, code
, narg
,
1293 (void *)(&args
.nosys
.sysmsg
+ 1));
1301 if (KTRPOINT(td
, KTR_SYSCALL
)) {
1302 MAKEMPSAFE(have_mplock
);
1303 ktrsyscall(p
, code
, narg
, (void *)(&args
.nosys
.sysmsg
+ 1));
1308 * For traditional syscall code edx is left untouched when 32 bit
1309 * results are returned. Since edx is loaded from fds[1] when the
1310 * system call returns we pre-set it here.
1312 args
.sysmsg_fds
[0] = 0;
1313 args
.sysmsg_fds
[1] = frame
.tf_edx
;
1316 * The syscall might manipulate the trap frame. If it does it
1317 * will probably return EJUSTRETURN.
1319 args
.sysmsg_frame
= &frame
;
1321 STOPEVENT(p
, S_SCE
, narg
); /* MP aware */
1325 * Try to run the syscall without the MP lock if the syscall
1326 * is MP safe. We have to obtain the MP lock no matter what if
1329 if ((callp
->sy_narg
& SYF_MPSAFE
) == 0)
1330 MAKEMPSAFE(have_mplock
);
1333 error
= (*callp
->sy_call
)(&args
);
1337 * MP SAFE (we may or may not have the MP lock at this point)
1342 * Reinitialize proc pointer `p' as it may be different
1343 * if this is a child returning from fork syscall.
1346 lp
= curthread
->td_lwp
;
1347 frame
.tf_eax
= args
.sysmsg_fds
[0];
1348 frame
.tf_edx
= args
.sysmsg_fds
[1];
1349 frame
.tf_eflags
&= ~PSL_C
;
1353 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1354 * int 0x80 is 2 bytes. We saved this in tf_err.
1356 frame
.tf_eip
-= frame
.tf_err
;
1361 panic("Unexpected EASYNC return value (for now)");
1364 if (p
->p_sysent
->sv_errsize
) {
1365 if (error
>= p
->p_sysent
->sv_errsize
)
1366 error
= -1; /* XXX */
1368 error
= p
->p_sysent
->sv_errtbl
[error
];
1370 frame
.tf_eax
= error
;
1371 frame
.tf_eflags
|= PSL_C
;
1376 * Traced syscall. trapsignal() is not MP aware.
1378 if ((orig_tf_eflags
& PSL_T
) && !(orig_tf_eflags
& PSL_VM
)) {
1379 MAKEMPSAFE(have_mplock
);
1380 frame
.tf_eflags
&= ~PSL_T
;
1381 trapsignal(p
, SIGTRAP
, 0);
1385 * Handle reschedule and other end-of-syscall issues
1387 userret(lp
, &frame
, sticks
);
1390 if (KTRPOINT(td
, KTR_SYSRET
)) {
1391 MAKEMPSAFE(have_mplock
);
1392 ktrsysret(p
, code
, error
, args
.sysmsg_result
);
1397 * This works because errno is findable through the
1398 * register set. If we ever support an emulation where this
1399 * is not the case, this code will need to be revisited.
1401 STOPEVENT(p
, S_SCX
, code
);
1406 * Release the MP lock if we had to get it
1408 KASSERT(td
->td_mpcount
== have_mplock
,
1409 ("badmpcount syscall2/end from %p", (void *)frame
.tf_eip
));
1414 KASSERT(crit_count
== (td
->td_pri
& ~TDPRI_MASK
),
1415 ("syscall: critical section count mismatch! %d/%d",
1416 crit_count
/ TDPRI_CRIT
, td
->td_pri
/ TDPRI_CRIT
));
1421 * Simplified back end of syscall(), used when returning from fork()
1422 * directly into user mode. MP lock is held on entry and should be
1423 * released on return. This code will return back into the fork
1424 * trampoline code which then runs doreti.
1427 fork_return(struct lwp
*lp
, struct trapframe frame
)
1429 struct proc
*p
= lp
->lwp_proc
;
1431 frame
.tf_eax
= 0; /* Child returns zero */
1432 frame
.tf_eflags
&= ~PSL_C
; /* success */
1436 * Newly forked processes are given a kernel priority. We have to
1437 * adjust the priority to a normal user priority and fake entry
1438 * into the kernel (call userenter()) to install a passive release
1439 * function just in case userret() decides to stop the process. This
1440 * can occur when ^Z races a fork. If we do not install the passive
1441 * release function the current process designation will not be
1442 * released when the thread goes to sleep.
1444 lwkt_setpri_self(TDPRI_USER_NORM
);
1445 userenter(lp
->lwp_thread
);
1446 userret(lp
, &frame
, 0);
1448 if (KTRPOINT(lp
->lwp_thread
, KTR_SYSRET
))
1449 ktrsysret(p
, SYS_fork
, 0, 0);
1451 p
->p_flag
|= P_PASSIVE_ACQ
;
1453 p
->p_flag
&= ~P_PASSIVE_ACQ
;
1455 KKASSERT(lp
->lwp_thread
->td_mpcount
== 1);
1461 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1462 * fault (which is then passed back to the virtual kernel) if an attempt is
1463 * made to use the FP unit.
1465 * XXX this is a fairly big hack.
1468 set_vkernel_fp(struct trapframe
*frame
)
1470 struct thread
*td
= curthread
;
1472 if (frame
->tf_xflags
& PGEX_FPFAULT
) {
1473 td
->td_pcb
->pcb_flags
|= FP_VIRTFP
;
1474 if (mdcpu
->gd_npxthread
== td
)
1477 td
->td_pcb
->pcb_flags
&= ~FP_VIRTFP
;