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/vkernel/i386/trap.c,v 1.9 2007/01/11 11:15:17 dillon Exp $
43 * 386 Trap and System call handling
50 #include "opt_ktrace.h"
52 #include <sys/param.h>
53 #include <sys/systm.h>
55 #include <sys/pioctl.h>
56 #include <sys/kernel.h>
57 #include <sys/resourcevar.h>
58 #include <sys/signalvar.h>
59 #include <sys/syscall.h>
60 #include <sys/sysctl.h>
61 #include <sys/sysent.h>
63 #include <sys/vmmeter.h>
64 #include <sys/malloc.h>
66 #include <sys/ktrace.h>
68 #include <sys/upcall.h>
69 #include <sys/vkernel.h>
70 #include <sys/sysproto.h>
71 #include <sys/sysunion.h>
72 #include <sys/vmspace.h>
75 #include <vm/vm_param.h>
78 #include <vm/vm_kern.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_page.h>
81 #include <vm/vm_extern.h>
83 #include <machine/cpu.h>
84 #include <machine/md_var.h>
85 #include <machine/pcb.h>
86 #include <machine/smp.h>
87 #include <machine/tss.h>
88 #include <machine/globaldata.h>
90 #include <machine/vm86.h>
93 #include <sys/msgport2.h>
94 #include <sys/thread2.h>
98 #define MAKEMPSAFE(have_mplock) \
99 if (have_mplock == 0) { \
106 #define MAKEMPSAFE(have_mplock)
110 int (*pmath_emulate
) (struct trapframe
*);
112 extern int trapwrite (unsigned addr
);
114 static int trap_pfault (struct trapframe
*, int, vm_offset_t
);
115 static void trap_fatal (struct trapframe
*, int, vm_offset_t
);
116 void dblfault_handler (void);
119 extern inthand_t
IDTVEC(syscall
);
122 #define MAX_TRAP_MSG 28
123 static char *trap_msg
[] = {
125 "privileged instruction fault", /* 1 T_PRIVINFLT */
127 "breakpoint instruction fault", /* 3 T_BPTFLT */
130 "arithmetic trap", /* 6 T_ARITHTRAP */
131 "system forced exception", /* 7 T_ASTFLT */
133 "general protection fault", /* 9 T_PROTFLT */
134 "trace trap", /* 10 T_TRCTRAP */
136 "page fault", /* 12 T_PAGEFLT */
138 "alignment fault", /* 14 T_ALIGNFLT */
142 "integer divide fault", /* 18 T_DIVIDE */
143 "non-maskable interrupt trap", /* 19 T_NMI */
144 "overflow trap", /* 20 T_OFLOW */
145 "FPU bounds check fault", /* 21 T_BOUND */
146 "FPU device not available", /* 22 T_DNA */
147 "double fault", /* 23 T_DOUBLEFLT */
148 "FPU operand fetch fault", /* 24 T_FPOPFLT */
149 "invalid TSS fault", /* 25 T_TSSFLT */
150 "segment not present fault", /* 26 T_SEGNPFLT */
151 "stack fault", /* 27 T_STKFLT */
152 "machine check trap", /* 28 T_MCHK */
156 static int ddb_on_nmi
= 1;
157 SYSCTL_INT(_machdep
, OID_AUTO
, ddb_on_nmi
, CTLFLAG_RW
,
158 &ddb_on_nmi
, 0, "Go to DDB on NMI");
160 static int panic_on_nmi
= 1;
161 SYSCTL_INT(_machdep
, OID_AUTO
, panic_on_nmi
, CTLFLAG_RW
,
162 &panic_on_nmi
, 0, "Panic on NMI");
163 static int fast_release
;
164 SYSCTL_INT(_machdep
, OID_AUTO
, fast_release
, CTLFLAG_RW
,
165 &fast_release
, 0, "Passive Release was optimal");
166 static int slow_release
;
167 SYSCTL_INT(_machdep
, OID_AUTO
, slow_release
, CTLFLAG_RW
,
168 &slow_release
, 0, "Passive Release was nonoptimal");
170 static int syscall_mpsafe
= 0;
171 SYSCTL_INT(_kern
, OID_AUTO
, syscall_mpsafe
, CTLFLAG_RW
,
172 &syscall_mpsafe
, 0, "Allow MPSAFE marked syscalls to run without BGL");
173 TUNABLE_INT("kern.syscall_mpsafe", &syscall_mpsafe
);
174 static int trap_mpsafe
= 0;
175 SYSCTL_INT(_kern
, OID_AUTO
, trap_mpsafe
, CTLFLAG_RW
,
176 &trap_mpsafe
, 0, "Allow traps to mostly run without the BGL");
177 TUNABLE_INT("kern.trap_mpsafe", &trap_mpsafe
);
180 MALLOC_DEFINE(M_SYSMSG
, "sysmsg", "sysmsg structure");
181 extern int max_sysmsg
;
184 * Passive USER->KERNEL transition. This only occurs if we block in the
185 * kernel while still holding our userland priority. We have to fixup our
186 * priority in order to avoid potential deadlocks before we allow the system
187 * to switch us to another thread.
190 passive_release(struct thread
*td
)
192 struct lwp
*lp
= td
->td_lwp
;
194 td
->td_release
= NULL
;
195 lwkt_setpri_self(TDPRI_KERN_USER
);
196 lp
->lwp_proc
->p_usched
->release_curproc(lp
);
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
= 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 * Block here if we are in a stopped state.
239 if (p
->p_flag
& P_STOPPED
) {
247 * Post any pending upcalls
249 if (p
->p_flag
& P_UPCALLPEND
) {
251 p
->p_flag
&= ~P_UPCALLPEND
;
258 * Post any pending signals
260 if ((sig
= CURSIG(p
)) != 0) {
268 * block here if we are swapped out, but still process signals
269 * (such as SIGKILL). proc0 (the swapin scheduler) is already
270 * aware of our situation, we do not have to wake it up.
272 if (p
->p_flag
& P_SWAPPEDOUT
) {
274 p
->p_flag
|= P_SWAPWAIT
;
276 if (p
->p_flag
& P_SWAPWAIT
)
277 tsleep(p
, PCATCH
, "SWOUT", 0);
278 p
->p_flag
&= ~P_SWAPWAIT
;
285 * Cleanup from userenter and any passive release that might have occured.
286 * We must reclaim the current-process designation before we can return
287 * to usermode. We also handle both LWKT and USER reschedule requests.
290 userexit(struct lwp
*lp
)
292 struct thread
*td
= lp
->lwp_thread
;
293 globaldata_t gd
= td
->td_gd
;
297 * If a user reschedule is requested force a new process to be
298 * chosen by releasing the current process. Our process will only
299 * be chosen again if it has a considerably better priority.
301 if (user_resched_wanted())
302 lp
->lwp_proc
->p_usched
->release_curproc(lp
);
306 * Handle a LWKT reschedule request first. Since our passive release
307 * is still in place we do not have to do anything special.
309 if (lwkt_resched_wanted())
313 * Acquire the current process designation for this user scheduler
314 * on this cpu. This will also handle any user-reschedule requests.
316 lp
->lwp_proc
->p_usched
->acquire_curproc(lp
);
317 /* We may have switched cpus on acquisition */
321 * Reduce our priority in preparation for a return to userland. If
322 * our passive release function was still in place, our priority was
323 * never raised and does not need to be reduced.
325 if (td
->td_release
== NULL
)
326 lwkt_setpri_self(TDPRI_USER_NORM
);
327 td
->td_release
= NULL
;
330 * After reducing our priority there might be other kernel-level
331 * LWKTs that now have a greater priority. Run them as necessary.
332 * We don't have to worry about losing cpu to userland because
333 * we still control the current-process designation and we no longer
334 * have a passive release function installed.
336 if (lwkt_checkpri_self())
341 * Exception, fault, and trap interface to the kernel.
342 * This common code is called from assembly language IDT gate entry
343 * routines that prepare a suitable stack frame, and restore this
344 * frame after the exception has been processed.
346 * This function is also called from doreti in an interlock to handle ASTs.
347 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
349 * NOTE! We have to retrieve the fault address prior to obtaining the
350 * MP lock because get_mplock() may switch out. YYY cr2 really ought
351 * to be retrieved by the assembly code, not here.
353 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
354 * if an attempt is made to switch from a fast interrupt or IPI. This is
355 * necessary to properly take fatal kernel traps on SMP machines if
356 * get_mplock() has to block.
360 user_trap(struct trapframe
*frame
)
362 struct globaldata
*gd
= mycpu
;
363 struct thread
*td
= gd
->gd_curthread
;
364 struct lwp
*lp
= td
->td_lwp
;
367 int i
= 0, ucode
= 0, type
, code
;
372 int crit_count
= td
->td_pri
& ~TDPRI_MASK
;
379 * This is a bad kludge to avoid changing the various trapframe
380 * structures. Because we are enabled as a virtual kernel,
381 * the original tf_err field will be passed to us shifted 16
382 * over in the tf_trapno field for T_PAGEFLT.
384 if (frame
->tf_trapno
== T_PAGEFLT
)
389 kprintf("USER_TRAP AT %08x xflags %d trapno %d eva %08x\n",
390 frame
->tf_eip
, frame
->tf_xflags
, frame
->tf_trapno
, eva
);
394 * Everything coming from user mode runs through user_trap,
395 * including system calls.
397 if (frame
->tf_trapno
== T_SYSCALL80
) {
404 eva
= (frame
->tf_trapno
== T_PAGEFLT
? rcr2() : 0);
405 ++gd
->gd_trap_nesting_level
;
406 MAKEMPSAFE(have_mplock
);
407 trap_fatal(frame
, TRUE
, eva
);
408 --gd
->gd_trap_nesting_level
;
413 ++gd
->gd_trap_nesting_level
;
415 if (trap_mpsafe
== 0)
416 MAKEMPSAFE(have_mplock
);
419 --gd
->gd_trap_nesting_level
;
421 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
424 type
= frame
->tf_trapno
;
425 code
= frame
->tf_err
;
429 sticks
= (int)td
->td_sticks
;
430 lp
->lwp_md
.md_regs
= frame
;
433 case T_PRIVINFLT
: /* privileged instruction fault */
438 case T_BPTFLT
: /* bpt instruction fault */
439 case T_TRCTRAP
: /* trace trap */
440 frame
->tf_eflags
&= ~PSL_T
;
444 case T_ARITHTRAP
: /* arithmetic trap */
449 case T_ASTFLT
: /* Allow process switch */
450 mycpu
->gd_cnt
.v_soft
++;
451 if (mycpu
->gd_reqflags
& RQF_AST_OWEUPC
) {
452 atomic_clear_int_nonlocked(&mycpu
->gd_reqflags
,
454 addupc_task(p
, p
->p_prof
.pr_addr
,
460 * The following two traps can happen in
461 * vm86 mode, and, if so, we want to handle
464 case T_PROTFLT
: /* general protection fault */
465 case T_STKFLT
: /* stack fault */
467 if (frame
->tf_eflags
& PSL_VM
) {
468 i
= vm86_emulate((struct vm86frame
*)frame
);
476 case T_SEGNPFLT
: /* segment not present fault */
477 case T_TSSFLT
: /* invalid TSS fault */
478 case T_DOUBLEFLT
: /* double fault */
480 ucode
= code
+ BUS_SEGM_FAULT
;
484 case T_PAGEFLT
: /* page fault */
485 MAKEMPSAFE(have_mplock
);
486 i
= trap_pfault(frame
, TRUE
, eva
);
489 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
499 case T_DIVIDE
: /* integer divide fault */
506 MAKEMPSAFE(have_mplock
);
507 /* machine/parity/power fail/"kitchen sink" faults */
508 if (isa_nmi(code
) == 0) {
511 * NMI can be hooked up to a pushbutton
515 kprintf ("NMI ... going to debugger\n");
516 kdb_trap (type
, 0, frame
);
520 } else if (panic_on_nmi
)
521 panic("NMI indicates hardware failure");
523 #endif /* NISA > 0 */
525 case T_OFLOW
: /* integer overflow fault */
530 case T_BOUND
: /* bounds check fault */
537 * Virtual kernel intercept - pass the DNA exception
538 * to the (emulated) virtual kernel if it asked to handle
539 * it. This occurs when the virtual kernel is holding
540 * onto the FP context for a different emulated
541 * process then the one currently running.
543 * We must still call npxdna() since we may have
544 * saved FP state that the (emulated) virtual kernel
545 * needs to hand over to a different emulated process.
547 if (p
->p_vkernel
&& p
->p_vkernel
->vk_current
&&
548 (td
->td_pcb
->pcb_flags
& FP_VIRTFP
)
555 * The kernel may have switched out the FP unit's
556 * state, causing the user process to take a fault
557 * when it tries to use the FP unit. Restore the
563 if (!pmath_emulate
) {
565 ucode
= FPE_FPU_NP_TRAP
;
568 i
= (*pmath_emulate
)(frame
);
570 if (!(frame
->tf_eflags
& PSL_T
))
572 frame
->tf_eflags
&= ~PSL_T
;
575 /* else ucode = emulator_only_knows() XXX */
578 case T_FPOPFLT
: /* FPU operand fetch fault */
583 case T_XMMFLT
: /* SIMD floating-point exception */
590 * Virtual kernel intercept - if the fault is directly related to a
591 * VM context managed by a virtual kernel then let the virtual kernel
594 if (p
->p_vkernel
&& p
->p_vkernel
->vk_current
) {
595 vkernel_trap(p
, frame
);
600 * Translate fault for emulators (e.g. Linux)
602 if (*p
->p_sysent
->sv_transtrap
)
603 i
= (*p
->p_sysent
->sv_transtrap
)(i
, type
);
605 MAKEMPSAFE(have_mplock
);
606 trapsignal(p
, i
, ucode
);
609 if (type
<= MAX_TRAP_MSG
) {
610 uprintf("fatal process exception: %s",
612 if ((type
== T_PAGEFLT
) || (type
== T_PROTFLT
))
613 uprintf(", fault VA = 0x%lx", (u_long
)eva
);
620 KASSERT(td
->td_mpcount
== have_mplock
, ("badmpcount trap/end from %p", (void *)frame
->tf_eip
));
622 userret(lp
, frame
, sticks
);
630 KASSERT(crit_count
== (td
->td_pri
& ~TDPRI_MASK
),
631 ("syscall: critical section count mismatch! %d/%d",
632 crit_count
/ TDPRI_CRIT
, td
->td_pri
/ TDPRI_CRIT
));
637 kern_trap(struct trapframe
*frame
)
639 struct globaldata
*gd
= mycpu
;
640 struct thread
*td
= gd
->gd_curthread
;
642 int i
= 0, ucode
= 0, type
, code
;
647 int crit_count
= td
->td_pri
& ~TDPRI_MASK
;
653 if (frame
->tf_trapno
== T_PAGEFLT
)
660 ++gd
->gd_trap_nesting_level
;
661 MAKEMPSAFE(have_mplock
);
662 trap_fatal(frame
, FALSE
, eva
);
663 --gd
->gd_trap_nesting_level
;
668 ++gd
->gd_trap_nesting_level
;
671 if (trap_mpsafe
== 0)
672 MAKEMPSAFE(have_mplock
);
675 --gd
->gd_trap_nesting_level
;
677 type
= frame
->tf_trapno
;
678 code
= frame
->tf_err
;
686 case T_PAGEFLT
: /* page fault */
687 MAKEMPSAFE(have_mplock
);
688 trap_pfault(frame
, FALSE
, eva
);
694 * The kernel may be using npx for copying or other
697 panic("kernel NPX should not happen");
703 case T_PROTFLT
: /* general protection fault */
704 case T_SEGNPFLT
: /* segment not present fault */
706 * Invalid segment selectors and out of bounds
707 * %eip's and %esp's can be set up in user mode.
708 * This causes a fault in kernel mode when the
709 * kernel tries to return to user mode. We want
710 * to get this fault so that we can fix the
711 * problem here and not have to check all the
712 * selectors and pointers when the user changes
715 if (mycpu
->gd_intr_nesting_level
== 0) {
716 if (td
->td_pcb
->pcb_onfault
) {
718 (register_t
)td
->td_pcb
->pcb_onfault
;
726 * PSL_NT can be set in user mode and isn't cleared
727 * automatically when the kernel is entered. This
728 * causes a TSS fault when the kernel attempts to
729 * `iret' because the TSS link is uninitialized. We
730 * want to get this fault so that we can fix the
731 * problem here and not every time the kernel is
734 if (frame
->tf_eflags
& PSL_NT
) {
735 frame
->tf_eflags
&= ~PSL_NT
;
740 case T_TRCTRAP
: /* trace trap */
742 if (frame
->tf_eip
== (int)IDTVEC(syscall
)) {
744 * We've just entered system mode via the
745 * syscall lcall. Continue single stepping
746 * silently until the syscall handler has
751 if (frame
->tf_eip
== (int)IDTVEC(syscall
) + 1) {
753 * The syscall handler has now saved the
754 * flags. Stop single stepping it.
756 frame
->tf_eflags
&= ~PSL_T
;
762 * Ignore debug register trace traps due to
763 * accesses in the user's address space, which
764 * can happen under several conditions such as
765 * if a user sets a watchpoint on a buffer and
766 * then passes that buffer to a system call.
767 * We still want to get TRCTRAPS for addresses
768 * in kernel space because that is useful when
769 * debugging the kernel.
771 if (user_dbreg_trap()) {
773 * Reset breakpoint bits because the
776 load_dr6(rdr6() & 0xfffffff0);
781 * Fall through (TRCTRAP kernel mode, kernel address)
785 * If DDB is enabled, let it handle the debugger trap.
786 * Otherwise, debugger traps "can't happen".
789 MAKEMPSAFE(have_mplock
);
790 if (kdb_trap (type
, 0, frame
))
796 MAKEMPSAFE(have_mplock
);
797 trap_fatal(frame
, FALSE
, eva
);
802 * Translate fault for emulators (e.g. Linux)
804 if (*p
->p_sysent
->sv_transtrap
)
805 i
= (*p
->p_sysent
->sv_transtrap
)(i
, type
);
807 MAKEMPSAFE(have_mplock
);
808 trapsignal(p
, i
, ucode
);
811 if (type
<= MAX_TRAP_MSG
) {
812 uprintf("fatal process exception: %s",
814 if ((type
== T_PAGEFLT
) || (type
== T_PROTFLT
))
815 uprintf(", fault VA = 0x%lx", (u_long
)eva
);
827 KASSERT(crit_count
== (td
->td_pri
& ~TDPRI_MASK
),
828 ("syscall: critical section count mismatch! %d/%d",
829 crit_count
/ TDPRI_CRIT
, td
->td_pri
/ TDPRI_CRIT
));
834 trap_pfault(struct trapframe
*frame
, int usermode
, vm_offset_t eva
)
837 struct vmspace
*vm
= NULL
;
841 thread_t td
= curthread
;
842 struct proc
*p
= td
->td_proc
;
844 va
= trunc_page(eva
);
845 if (usermode
== FALSE
) {
847 * This is a fault on kernel virtual memory.
852 * This is a fault on non-kernel virtual memory.
853 * vm is initialized above to NULL. If curproc is NULL
854 * or curproc->p_vmspace is NULL the fault is fatal.
865 if (frame
->tf_xflags
& PGEX_W
)
866 ftype
= VM_PROT_READ
| VM_PROT_WRITE
;
868 ftype
= VM_PROT_READ
;
870 if (map
!= &kernel_map
) {
872 * Keep swapout from messing with us during this
878 * Grow the stack if necessary
880 /* grow_stack returns false only if va falls into
881 * a growable stack region and the stack growth
882 * fails. It returns true if va was not within
883 * a growable stack region, or if the stack
886 if (!grow_stack (p
, va
)) {
892 /* Fault in the user page: */
893 rv
= vm_fault(map
, va
, ftype
,
894 (ftype
& VM_PROT_WRITE
) ? VM_FAULT_DIRTY
900 * Don't have to worry about process locking or stacks in the kernel.
902 rv
= vm_fault(map
, va
, ftype
, VM_FAULT_NORMAL
);
905 if (rv
== KERN_SUCCESS
)
909 if (td
->td_gd
->gd_intr_nesting_level
== 0 &&
910 td
->td_pcb
->pcb_onfault
) {
911 frame
->tf_eip
= (register_t
)td
->td_pcb
->pcb_onfault
;
914 trap_fatal(frame
, usermode
, eva
);
917 return((rv
== KERN_PROTECTION_FAILURE
) ? SIGBUS
: SIGSEGV
);
921 trap_fatal(struct trapframe
*frame
, int usermode
, vm_offset_t eva
)
923 int code
, type
, ss
, esp
;
925 code
= frame
->tf_xflags
;
926 type
= frame
->tf_trapno
;
928 if (type
<= MAX_TRAP_MSG
)
929 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
930 type
, trap_msg
[type
],
931 /*frame->tf_eflags & PSL_VM ? "vm86" :*/
932 ISPL(frame
->tf_cs
) == SEL_UPL
? "user" : "kernel");
934 /* three separate prints in case of a trap on an unmapped page */
935 kprintf("mp_lock = %08x; ", mp_lock
);
936 kprintf("cpuid = %d; ", mycpu
->gd_cpuid
);
937 kprintf("lapic.id = %08x\n", lapic
.id
);
939 if (type
== T_PAGEFLT
) {
940 kprintf("fault virtual address = 0x%x\n", eva
);
941 kprintf("fault code = %s %s, %s\n",
942 usermode
? "user" : "supervisor",
943 code
& PGEX_W
? "write" : "read",
944 code
& PGEX_P
? "protection violation" : "page not present");
946 kprintf("instruction pointer = 0x%x:0x%x\n",
947 frame
->tf_cs
& 0xffff, frame
->tf_eip
);
948 if ((ISPL(frame
->tf_cs
) == SEL_UPL
) /*||(frame->tf_eflags&PSL_VM)*/) {
949 ss
= frame
->tf_ss
& 0xffff;
952 ss
= GSEL(GDATA_SEL
, SEL_KPL
);
953 esp
= (int)&frame
->tf_esp
;
955 kprintf("stack pointer = 0x%x:0x%x\n", ss
, esp
);
956 kprintf("frame pointer = 0x%x:0x%x\n", ss
, frame
->tf_ebp
);
957 kprintf("processor eflags = ");
958 if (frame
->tf_eflags
& PSL_T
)
959 kprintf("trace trap, ");
960 if (frame
->tf_eflags
& PSL_I
)
961 kprintf("interrupt enabled, ");
962 if (frame
->tf_eflags
& PSL_NT
)
963 kprintf("nested task, ");
964 if (frame
->tf_eflags
& PSL_RF
)
967 if (frame
->tf_eflags
& PSL_VM
)
970 kprintf("IOPL = %d\n", (frame
->tf_eflags
& PSL_IOPL
) >> 12);
971 kprintf("current process = ");
973 kprintf("%lu (%s)\n",
974 (u_long
)curproc
->p_pid
, curproc
->p_comm
?
975 curproc
->p_comm
: "");
979 kprintf("current thread = pri %d ", curthread
->td_pri
);
980 if (curthread
->td_pri
>= TDPRI_CRIT
)
986 * we probably SHOULD have stopped the other CPUs before now!
987 * another CPU COULD have been touching cpl at this moment...
989 kprintf(" <- SMP: XXX");
998 if ((debugger_on_panic
|| db_active
) && kdb_trap(type
, code
, frame
))
1001 kprintf("trap number = %d\n", type
);
1002 if (type
<= MAX_TRAP_MSG
)
1003 panic("%s", trap_msg
[type
]);
1005 panic("unknown/reserved trap");
1009 * Double fault handler. Called when a fault occurs while writing
1010 * a frame for a trap/exception onto the stack. This usually occurs
1011 * when the stack overflows (such is the case with infinite recursion,
1014 * XXX Note that the current PTD gets replaced by IdlePTD when the
1015 * task switch occurs. This means that the stack that was active at
1016 * the time of the double fault is not available at <kstack> unless
1017 * the machine was idle when the double fault occurred. The downside
1018 * of this is that "trace <ebp>" in ddb won't work.
1021 dblfault_handler(void)
1023 struct mdglobaldata
*gd
= mdcpu
;
1025 kprintf("\nFatal double fault:\n");
1026 kprintf("eip = 0x%x\n", gd
->gd_common_tss
.tss_eip
);
1027 kprintf("esp = 0x%x\n", gd
->gd_common_tss
.tss_esp
);
1028 kprintf("ebp = 0x%x\n", gd
->gd_common_tss
.tss_ebp
);
1030 /* three separate prints in case of a trap on an unmapped page */
1031 kprintf("mp_lock = %08x; ", mp_lock
);
1032 kprintf("cpuid = %d; ", mycpu
->gd_cpuid
);
1033 kprintf("lapic.id = %08x\n", lapic
.id
);
1035 panic("double fault");
1039 * Compensate for 386 brain damage (missing URKR).
1040 * This is a little simpler than the pagefault handler in trap() because
1041 * it the page tables have already been faulted in and high addresses
1042 * are thrown out early for other reasons.
1045 trapwrite(unsigned addr
)
1052 va
= trunc_page((vm_offset_t
)addr
);
1054 * XXX - MAX is END. Changed > to >= for temp. fix.
1056 if (va
>= VM_MAX_USER_ADDRESS
)
1064 if (!grow_stack (p
, va
)) {
1070 * fault the data page
1072 rv
= vm_fault(&vm
->vm_map
, va
, VM_PROT_WRITE
, VM_FAULT_DIRTY
);
1076 if (rv
!= KERN_SUCCESS
)
1083 * syscall2 - MP aware system call request C handler
1085 * A system call is essentially treated as a trap except that the
1086 * MP lock is not held on entry or return. We are responsible for
1087 * obtaining the MP lock if necessary and for handling ASTs
1088 * (e.g. a task switch) prior to return.
1090 * In general, only simple access and manipulation of curproc and
1091 * the current stack is allowed without having to hold MP lock.
1093 * MPSAFE - note that large sections of this routine are run without
1098 syscall2(struct trapframe
*frame
)
1100 struct thread
*td
= curthread
;
1101 struct proc
*p
= td
->td_proc
;
1102 struct lwp
*lp
= td
->td_lwp
;
1104 struct sysent
*callp
;
1105 register_t orig_tf_eflags
;
1110 int crit_count
= td
->td_pri
& ~TDPRI_MASK
;
1113 int have_mplock
= 0;
1116 union sysunion args
;
1119 if (ISPL(frame
->tf_cs
) != SEL_UPL
) {
1127 KASSERT(td
->td_mpcount
== 0, ("badmpcount syscall2 from %p", (void *)frame
->tf_eip
));
1128 if (syscall_mpsafe
== 0)
1129 MAKEMPSAFE(have_mplock
);
1131 userenter(td
); /* lazy raise our priority */
1136 sticks
= (int)td
->td_sticks
;
1137 orig_tf_eflags
= frame
->tf_eflags
;
1140 * Virtual kernel intercept - if a VM context managed by a virtual
1141 * kernel issues a system call the virtual kernel handles it, not us.
1142 * Restore the virtual kernel context and return from its system
1143 * call. The current frame is copied out to the virtual kernel.
1145 if (p
->p_vkernel
&& p
->p_vkernel
->vk_current
) {
1146 error
= vkernel_trap(p
, frame
);
1147 frame
->tf_eax
= error
;
1149 frame
->tf_eflags
|= PSL_C
;
1150 error
= EJUSTRETURN
;
1155 * Get the system call parameters and account for time
1157 lp
->lwp_md
.md_regs
= frame
;
1158 params
= (caddr_t
)frame
->tf_esp
+ sizeof(int);
1159 code
= frame
->tf_eax
;
1161 if (p
->p_sysent
->sv_prepsyscall
) {
1162 (*p
->p_sysent
->sv_prepsyscall
)(
1163 frame
, (int *)(&args
.nosys
.sysmsg
+ 1),
1167 * Need to check if this is a 32 bit or 64 bit syscall.
1168 * fuword is MP aware.
1170 if (code
== SYS_syscall
) {
1172 * Code is first argument, followed by actual args.
1174 code
= fuword(params
);
1175 params
+= sizeof(int);
1176 } else if (code
== SYS___syscall
) {
1178 * Like syscall, but code is a quad, so as to maintain
1179 * quad alignment for the rest of the arguments.
1181 code
= fuword(params
);
1182 params
+= sizeof(quad_t
);
1186 code
&= p
->p_sysent
->sv_mask
;
1187 if (code
>= p
->p_sysent
->sv_size
)
1188 callp
= &p
->p_sysent
->sv_table
[0];
1190 callp
= &p
->p_sysent
->sv_table
[code
];
1192 narg
= callp
->sy_narg
& SYF_ARGMASK
;
1195 * copyin is MP aware, but the tracing code is not
1197 if (narg
&& params
) {
1198 error
= copyin(params
, (caddr_t
)(&args
.nosys
.sysmsg
+ 1),
1199 narg
* sizeof(register_t
));
1202 if (KTRPOINT(td
, KTR_SYSCALL
)) {
1203 MAKEMPSAFE(have_mplock
);
1205 ktrsyscall(p
, code
, narg
,
1206 (void *)(&args
.nosys
.sysmsg
+ 1));
1214 if (KTRPOINT(td
, KTR_SYSCALL
)) {
1215 MAKEMPSAFE(have_mplock
);
1216 ktrsyscall(p
, code
, narg
, (void *)(&args
.nosys
.sysmsg
+ 1));
1221 * For traditional syscall code edx is left untouched when 32 bit
1222 * results are returned. Since edx is loaded from fds[1] when the
1223 * system call returns we pre-set it here.
1225 args
.sysmsg_fds
[0] = 0;
1226 args
.sysmsg_fds
[1] = frame
->tf_edx
;
1229 * The syscall might manipulate the trap frame. If it does it
1230 * will probably return EJUSTRETURN.
1232 args
.sysmsg_frame
= frame
;
1234 STOPEVENT(p
, S_SCE
, narg
); /* MP aware */
1238 * Try to run the syscall without the MP lock if the syscall
1239 * is MP safe. We have to obtain the MP lock no matter what if
1242 if ((callp
->sy_narg
& SYF_MPSAFE
) == 0)
1243 MAKEMPSAFE(have_mplock
);
1246 error
= (*callp
->sy_call
)(&args
);
1249 kprintf("system call %d returned %d\n", code
, error
);
1254 * MP SAFE (we may or may not have the MP lock at this point)
1259 * Reinitialize proc pointer `p' as it may be different
1260 * if this is a child returning from fork syscall.
1263 lp
= curthread
->td_lwp
;
1264 frame
->tf_eax
= args
.sysmsg_fds
[0];
1265 frame
->tf_edx
= args
.sysmsg_fds
[1];
1266 frame
->tf_eflags
&= ~PSL_C
;
1270 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1271 * int 0x80 is 2 bytes. We saved this in tf_err.
1273 frame
->tf_eip
-= frame
->tf_err
;
1278 panic("Unexpected EASYNC return value (for now)");
1281 if (p
->p_sysent
->sv_errsize
) {
1282 if (error
>= p
->p_sysent
->sv_errsize
)
1283 error
= -1; /* XXX */
1285 error
= p
->p_sysent
->sv_errtbl
[error
];
1287 frame
->tf_eax
= error
;
1288 frame
->tf_eflags
|= PSL_C
;
1293 * Traced syscall. trapsignal() is not MP aware.
1295 if ((orig_tf_eflags
& PSL_T
) /*&& !(orig_tf_eflags & PSL_VM)*/) {
1296 MAKEMPSAFE(have_mplock
);
1297 frame
->tf_eflags
&= ~PSL_T
;
1298 trapsignal(p
, SIGTRAP
, 0);
1302 * Handle reschedule and other end-of-syscall issues
1304 userret(lp
, frame
, sticks
);
1307 if (KTRPOINT(td
, KTR_SYSRET
)) {
1308 MAKEMPSAFE(have_mplock
);
1309 ktrsysret(p
, code
, error
, args
.sysmsg_result
);
1314 * This works because errno is findable through the
1315 * register set. If we ever support an emulation where this
1316 * is not the case, this code will need to be revisited.
1318 STOPEVENT(p
, S_SCX
, code
);
1323 * Release the MP lock if we had to get it
1325 KASSERT(td
->td_mpcount
== have_mplock
,
1326 ("badmpcount syscall2/end from %p", (void *)frame
->tf_eip
));
1331 KASSERT(crit_count
== (td
->td_pri
& ~TDPRI_MASK
),
1332 ("syscall: critical section count mismatch! %d/%d",
1333 crit_count
/ TDPRI_CRIT
, td
->td_pri
/ TDPRI_CRIT
));
1338 * Simplified back end of syscall(), used when returning from fork()
1339 * directly into user mode. MP lock is held on entry and should be
1340 * released on return. This code will return back into the fork
1341 * trampoline code which then runs doreti.
1344 fork_return(struct lwp
*lp
, struct trapframe frame
)
1346 struct proc
*p
= lp
->lwp_proc
;
1348 frame
.tf_eax
= 0; /* Child returns zero */
1349 frame
.tf_eflags
&= ~PSL_C
; /* success */
1353 * Newly forked processes are given a kernel priority. We have to
1354 * adjust the priority to a normal user priority and fake entry
1355 * into the kernel (call userenter()) to install a passive release
1356 * function just in case userret() decides to stop the process. This
1357 * can occur when ^Z races a fork. If we do not install the passive
1358 * release function the current process designation will not be
1359 * released when the thread goes to sleep.
1361 lwkt_setpri_self(TDPRI_USER_NORM
);
1362 userenter(lp
->lwp_thread
);
1363 userret(lp
, &frame
, 0);
1365 if (KTRPOINT(lp
->lwp_thread
, KTR_SYSRET
))
1366 ktrsysret(p
, SYS_fork
, 0, 0);
1368 p
->p_flag
|= P_PASSIVE_ACQ
;
1370 p
->p_flag
&= ~P_PASSIVE_ACQ
;
1372 KKASSERT(lp
->lwp_thread
->td_mpcount
== 1);
1378 * doreti has turned into this. The frame is directly on the stack. We
1379 * pull everything else we need (fpu and tls context) from the current
1382 * Note on fpu interactions: In a virtual kernel, the fpu context for
1383 * an emulated user mode process is not shared with the virtual kernel's
1384 * fpu context, so we only have to 'stack' fpu contexts within the virtual
1385 * kernel itself, and not even then since the signal() contexts that we care
1386 * about save and restore the FPU state (I think anyhow).
1388 * vmspace_ctl() returns an error only if it had problems instaling the
1389 * context we supplied or problems copying data to/from our VM space.
1392 go_user(struct trapframe frame
)
1397 * Interrupts may be disabled on entry, make sure all signals
1398 * can be received before beginning our loop.
1403 * Switch to the current simulated user process, then call
1404 * user_trap() when we break out of it (usually due to a signal).
1408 kprintf("GO USER VMSPC %p pid %-4d %s (blocked %08x)\n",
1409 &curproc
->p_vmspace
->vm_pmap
,
1410 curproc
->p_pid
, curproc
->p_comm
, sigblock(0));
1414 * Tell the real kernel whether it is ok to use the FP
1417 if (mdcpu
->gd_npxthread
== curthread
) {
1418 frame
.tf_xflags
&= ~FPEX_FAULT
;
1420 frame
.tf_xflags
|= FPEX_FAULT
;
1424 * Run emulated user process context
1426 r
= vmspace_ctl(&curproc
->p_vmspace
->vm_pmap
, VMSPACE_CTL_RUN
,
1427 &frame
, &curthread
->td_savevext
);
1429 panic("vmspace_ctl had problems with the context");
1431 if (frame
.tf_trapno
) {
1433 kprintf("User trapno %d eip %08x err %08x xflags %d\n",
1434 frame
.tf_trapno
, frame
.tf_eip
, frame
.tf_err
, frame
.tf_xflags
);
1437 } else if (mycpu
->gd_reqflags
& RQF_AST_MASK
) {
1439 kprintf("reqflags %08x %08x %d\n", mycpu
->gd_reqflags
, sigblock(0), curthread
->td_pri
);
1441 frame
.tf_trapno
= T_ASTFLT
;
1445 kprintf("Kernel AST %08x\n", sigblock(0));
1452 * If FPEX_FAULT 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 panic("set_vkernel_fp: vkernel-within-vkernel not yet supported");
1464 if (frame
->tf_xflags
& FPEX_FAULT
) {
1465 td
->td_pcb
->pcb_flags
|= FP_VIRTFP
;
1466 if (mdcpu
->gd_npxthread
== td
)
1469 td
->td_pcb
->pcb_flags
&= ~FP_VIRTFP
;