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 1.35 2008/09/09 04:06:19 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/signal2.h>
60 #include <sys/syscall.h>
61 #include <sys/sysctl.h>
62 #include <sys/sysent.h>
64 #include <sys/vmmeter.h>
65 #include <sys/malloc.h>
67 #include <sys/ktrace.h>
70 #include <sys/upcall.h>
71 #include <sys/vkernel.h>
72 #include <sys/sysproto.h>
73 #include <sys/sysunion.h>
74 #include <sys/vmspace.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/globaldata.h>
92 #include <machine/vm86.h>
96 #include <sys/msgport2.h>
97 #include <sys/thread2.h>
98 #include <sys/mplock2.h>
102 #define MAKEMPSAFE(have_mplock) \
103 if (have_mplock == 0) { \
110 #define MAKEMPSAFE(have_mplock)
114 int (*pmath_emulate
) (struct trapframe
*);
116 static int trap_pfault (struct trapframe
*, int, vm_offset_t
);
117 static void trap_fatal (struct trapframe
*, int, vm_offset_t
);
118 void dblfault_handler (void);
121 extern inthand_t
IDTVEC(syscall
);
124 #define MAX_TRAP_MSG 28
125 static char *trap_msg
[] = {
127 "privileged instruction fault", /* 1 T_PRIVINFLT */
129 "breakpoint instruction fault", /* 3 T_BPTFLT */
132 "arithmetic trap", /* 6 T_ARITHTRAP */
133 "system forced exception", /* 7 T_ASTFLT */
135 "general protection fault", /* 9 T_PROTFLT */
136 "trace trap", /* 10 T_TRCTRAP */
138 "page fault", /* 12 T_PAGEFLT */
140 "alignment fault", /* 14 T_ALIGNFLT */
144 "integer divide fault", /* 18 T_DIVIDE */
145 "non-maskable interrupt trap", /* 19 T_NMI */
146 "overflow trap", /* 20 T_OFLOW */
147 "FPU bounds check fault", /* 21 T_BOUND */
148 "FPU device not available", /* 22 T_DNA */
149 "double fault", /* 23 T_DOUBLEFLT */
150 "FPU operand fetch fault", /* 24 T_FPOPFLT */
151 "invalid TSS fault", /* 25 T_TSSFLT */
152 "segment not present fault", /* 26 T_SEGNPFLT */
153 "stack fault", /* 27 T_STKFLT */
154 "machine check trap", /* 28 T_MCHK */
158 static int ddb_on_nmi
= 1;
159 SYSCTL_INT(_machdep
, OID_AUTO
, ddb_on_nmi
, CTLFLAG_RW
,
160 &ddb_on_nmi
, 0, "Go to DDB on NMI");
162 static int panic_on_nmi
= 1;
163 SYSCTL_INT(_machdep
, OID_AUTO
, panic_on_nmi
, CTLFLAG_RW
,
164 &panic_on_nmi
, 0, "Panic on NMI");
165 static int fast_release
;
166 SYSCTL_INT(_machdep
, OID_AUTO
, fast_release
, CTLFLAG_RW
,
167 &fast_release
, 0, "Passive Release was optimal");
168 static int slow_release
;
169 SYSCTL_INT(_machdep
, OID_AUTO
, slow_release
, CTLFLAG_RW
,
170 &slow_release
, 0, "Passive Release was nonoptimal");
172 static int syscall_mpsafe
= 1;
173 SYSCTL_INT(_kern
, OID_AUTO
, syscall_mpsafe
, CTLFLAG_RW
,
174 &syscall_mpsafe
, 0, "Allow MPSAFE marked syscalls to run without BGL");
175 TUNABLE_INT("kern.syscall_mpsafe", &syscall_mpsafe
);
176 static int trap_mpsafe
= 1;
177 SYSCTL_INT(_kern
, OID_AUTO
, trap_mpsafe
, CTLFLAG_RW
,
178 &trap_mpsafe
, 0, "Allow traps to mostly run without the BGL");
179 TUNABLE_INT("kern.trap_mpsafe", &trap_mpsafe
);
182 MALLOC_DEFINE(M_SYSMSG
, "sysmsg", "sysmsg structure");
183 extern int max_sysmsg
;
186 * Passively intercepts the thread switch function to increase
187 * the thread priority from a user priority to a kernel priority, reducing
188 * syscall and trap overhead for the case where no switch occurs.
190 * Synchronizes td_ucred with p_ucred. This is used by system calls,
191 * signal handling, faults, AST traps, and anything else that enters the
192 * kernel from userland and provides the kernel with a stable read-only
193 * copy of the process ucred.
196 userenter(struct thread
*curtd
, struct proc
*curp
)
201 curtd
->td_release
= lwkt_passive_release
;
203 if (curtd
->td_ucred
!= curp
->p_ucred
) {
204 ncred
= crhold(curp
->p_ucred
);
205 ocred
= curtd
->td_ucred
;
206 curtd
->td_ucred
= ncred
;
213 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
214 * must be completed before we can return to or try to return to userland.
216 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
217 * arithmatic on the delta calculation so the absolute tick values are
218 * truncated to an integer.
221 userret(struct lwp
*lp
, struct trapframe
*frame
, int sticks
)
223 struct proc
*p
= lp
->lwp_proc
;
227 * Charge system time if profiling. Note: times are in microseconds.
228 * This may do a copyout and block, so do it first even though it
229 * means some system time will be charged as user time.
231 if (p
->p_flag
& P_PROFIL
) {
232 addupc_task(p
, frame
->tf_eip
,
233 (u_int
)((int)lp
->lwp_thread
->td_sticks
- sticks
));
238 * If the jungle wants us dead, so be it.
240 if (lp
->lwp_flag
& LWP_WEXIT
) {
243 rel_mplock(); /* NOT REACHED */
247 * Block here if we are in a stopped state.
249 if (p
->p_stat
== SSTOP
) {
257 * Post any pending upcalls
259 if (p
->p_flag
& P_UPCALLPEND
) {
261 p
->p_flag
&= ~P_UPCALLPEND
;
268 * Post any pending signals
270 if ((sig
= CURSIG_TRACE(lp
)) != 0) {
278 * block here if we are swapped out, but still process signals
279 * (such as SIGKILL). proc0 (the swapin scheduler) is already
280 * aware of our situation, we do not have to wake it up.
282 if (p
->p_flag
& P_SWAPPEDOUT
) {
284 p
->p_flag
|= P_SWAPWAIT
;
286 if (p
->p_flag
& P_SWAPWAIT
)
287 tsleep(p
, PCATCH
, "SWOUT", 0);
288 p
->p_flag
&= ~P_SWAPWAIT
;
294 * Make sure postsig() handled request to restore old signal mask after
295 * running signal handler.
297 KKASSERT((lp
->lwp_flag
& LWP_OLDMASK
) == 0);
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; */
312 * Handle stop requests at kernel priority. Any requests queued
313 * after this loop will generate another AST.
315 while (lp
->lwp_proc
->p_stat
== SSTOP
) {
322 * Reduce our priority in preparation for a return to userland. If
323 * our passive release function was still in place, our priority was
324 * never raised and does not need to be reduced.
326 lwkt_passive_recover(td
);
329 * Become the current user scheduled process if we aren't already,
330 * and deal with reschedule requests and other factors.
332 lp
->lwp_proc
->p_usched
->acquire_curproc(lp
);
333 /* WARNING: we may have migrated cpu's */
334 /* gd = td->td_gd; */
337 #if !defined(KTR_KERNENTRY)
338 #define KTR_KERNENTRY KTR_ALL
340 KTR_INFO_MASTER(kernentry
);
341 KTR_INFO(KTR_KERNENTRY
, kernentry
, trap
, 0, "pid=%d, tid=%d, trapno=%d, eva=%p",
342 sizeof(int) + sizeof(int) + sizeof(int) + sizeof(vm_offset_t
));
343 KTR_INFO(KTR_KERNENTRY
, kernentry
, trap_ret
, 0, "pid=%d, tid=%d",
344 sizeof(int) + sizeof(int));
345 KTR_INFO(KTR_KERNENTRY
, kernentry
, syscall
, 0, "pid=%d, tid=%d, call=%d",
346 sizeof(int) + sizeof(int) + sizeof(int));
347 KTR_INFO(KTR_KERNENTRY
, kernentry
, syscall_ret
, 0, "pid=%d, tid=%d, err=%d",
348 sizeof(int) + sizeof(int) + sizeof(int));
349 KTR_INFO(KTR_KERNENTRY
, kernentry
, fork_ret
, 0, "pid=%d, tid=%d",
350 sizeof(int) + sizeof(int));
353 * Exception, fault, and trap interface to the kernel.
354 * This common code is called from assembly language IDT gate entry
355 * routines that prepare a suitable stack frame, and restore this
356 * frame after the exception has been processed.
358 * This function is also called from doreti in an interlock to handle ASTs.
359 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
361 * NOTE! We have to retrieve the fault address prior to obtaining the
362 * MP lock because get_mplock() may switch out. YYY cr2 really ought
363 * to be retrieved by the assembly code, not here.
365 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
366 * if an attempt is made to switch from a fast interrupt or IPI. This is
367 * necessary to properly take fatal kernel traps on SMP machines if
368 * get_mplock() has to block.
372 user_trap(struct trapframe
*frame
)
374 struct globaldata
*gd
= mycpu
;
375 struct thread
*td
= gd
->gd_curthread
;
376 struct lwp
*lp
= td
->td_lwp
;
379 int i
= 0, ucode
= 0, type
, code
;
384 int crit_count
= td
->td_pri
& ~TDPRI_MASK
;
391 * This is a bad kludge to avoid changing the various trapframe
392 * structures. Because we are enabled as a virtual kernel,
393 * the original tf_err field will be passed to us shifted 16
394 * over in the tf_trapno field for T_PAGEFLT.
396 if (frame
->tf_trapno
== T_PAGEFLT
)
401 kprintf("USER_TRAP AT %08x xflags %d trapno %d eva %08x\n",
402 frame
->tf_eip
, frame
->tf_xflags
, frame
->tf_trapno
, eva
);
406 * Everything coming from user mode runs through user_trap,
407 * including system calls.
409 if (frame
->tf_trapno
== T_SYSCALL80
) {
414 KTR_LOG(kernentry_trap
, lp
->lwp_proc
->p_pid
, lp
->lwp_tid
,
415 frame
->tf_trapno
, eva
);
419 eva
= (frame
->tf_trapno
== T_PAGEFLT
? rcr2() : 0);
420 ++gd
->gd_trap_nesting_level
;
421 MAKEMPSAFE(have_mplock
);
422 trap_fatal(frame
, TRUE
, eva
);
423 --gd
->gd_trap_nesting_level
;
428 ++gd
->gd_trap_nesting_level
;
430 if (trap_mpsafe
== 0)
431 MAKEMPSAFE(have_mplock
);
434 --gd
->gd_trap_nesting_level
;
436 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
439 type
= frame
->tf_trapno
;
440 code
= frame
->tf_err
;
444 sticks
= (int)td
->td_sticks
;
445 lp
->lwp_md
.md_regs
= frame
;
448 case T_PRIVINFLT
: /* privileged instruction fault */
453 case T_BPTFLT
: /* bpt instruction fault */
454 case T_TRCTRAP
: /* trace trap */
455 frame
->tf_eflags
&= ~PSL_T
;
460 case T_ARITHTRAP
: /* arithmetic trap */
465 case T_ASTFLT
: /* Allow process switch */
466 mycpu
->gd_cnt
.v_soft
++;
467 if (mycpu
->gd_reqflags
& RQF_AST_OWEUPC
) {
468 atomic_clear_int_nonlocked(&mycpu
->gd_reqflags
,
470 addupc_task(p
, p
->p_prof
.pr_addr
,
476 * The following two traps can happen in
477 * vm86 mode, and, if so, we want to handle
480 case T_PROTFLT
: /* general protection fault */
481 case T_STKFLT
: /* stack fault */
483 if (frame
->tf_eflags
& PSL_VM
) {
484 i
= vm86_emulate((struct vm86frame
*)frame
);
491 ucode
= (type
== T_PROTFLT
) ? BUS_OBJERR
: BUS_ADRERR
;
493 case T_SEGNPFLT
: /* segment not present fault */
497 case T_TSSFLT
: /* invalid TSS fault */
498 case T_DOUBLEFLT
: /* double fault */
503 ucode
= code
+ BUS_SEGM_FAULT
; /* XXX: ???*/
509 case T_PAGEFLT
: /* page fault */
510 MAKEMPSAFE(have_mplock
);
511 i
= trap_pfault(frame
, TRUE
, eva
);
514 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
530 case T_DIVIDE
: /* integer divide fault */
537 MAKEMPSAFE(have_mplock
);
538 /* machine/parity/power fail/"kitchen sink" faults */
539 if (isa_nmi(code
) == 0) {
542 * NMI can be hooked up to a pushbutton
546 kprintf ("NMI ... going to debugger\n");
547 kdb_trap (type
, 0, frame
);
551 } else if (panic_on_nmi
)
552 panic("NMI indicates hardware failure");
554 #endif /* NISA > 0 */
556 case T_OFLOW
: /* integer overflow fault */
561 case T_BOUND
: /* bounds check fault */
568 * Virtual kernel intercept - pass the DNA exception
569 * to the (emulated) virtual kernel if it asked to handle
570 * it. This occurs when the virtual kernel is holding
571 * onto the FP context for a different emulated
572 * process then the one currently running.
574 * We must still call npxdna() since we may have
575 * saved FP state that the (emulated) virtual kernel
576 * needs to hand over to a different emulated process.
578 if (lp
->lwp_vkernel
&& lp
->lwp_vkernel
->ve
&&
579 (td
->td_pcb
->pcb_flags
& FP_VIRTFP
)
586 * The kernel may have switched out the FP unit's
587 * state, causing the user process to take a fault
588 * when it tries to use the FP unit. Restore the
594 if (!pmath_emulate
) {
596 ucode
= FPE_FPU_NP_TRAP
;
599 i
= (*pmath_emulate
)(frame
);
601 if (!(frame
->tf_eflags
& PSL_T
))
603 frame
->tf_eflags
&= ~PSL_T
;
606 /* else ucode = emulator_only_knows() XXX */
609 case T_FPOPFLT
: /* FPU operand fetch fault */
614 case T_XMMFLT
: /* SIMD floating-point exception */
621 * Virtual kernel intercept - if the fault is directly related to a
622 * VM context managed by a virtual kernel then let the virtual kernel
625 if (lp
->lwp_vkernel
&& lp
->lwp_vkernel
->ve
) {
626 vkernel_trap(lp
, frame
);
631 * Translate fault for emulators (e.g. Linux)
633 if (*p
->p_sysent
->sv_transtrap
)
634 i
= (*p
->p_sysent
->sv_transtrap
)(i
, type
);
636 MAKEMPSAFE(have_mplock
);
637 trapsignal(lp
, i
, ucode
);
640 if (type
<= MAX_TRAP_MSG
) {
641 uprintf("fatal process exception: %s",
643 if ((type
== T_PAGEFLT
) || (type
== T_PROTFLT
))
644 uprintf(", fault VA = 0x%lx", (u_long
)eva
);
651 KASSERT(td
->td_mpcount
== have_mplock
, ("badmpcount trap/end from %p", (void *)frame
->tf_eip
));
653 userret(lp
, frame
, sticks
);
660 KTR_LOG(kernentry_trap_ret
, lp
->lwp_proc
->p_pid
, lp
->lwp_tid
);
662 KASSERT(crit_count
== (td
->td_pri
& ~TDPRI_MASK
),
663 ("syscall: critical section count mismatch! %d/%d",
664 crit_count
/ TDPRI_CRIT
, td
->td_pri
/ TDPRI_CRIT
));
669 kern_trap(struct trapframe
*frame
)
671 struct globaldata
*gd
= mycpu
;
672 struct thread
*td
= gd
->gd_curthread
;
675 int i
= 0, ucode
= 0, type
, code
;
680 int crit_count
= td
->td_pri
& ~TDPRI_MASK
;
687 if (frame
->tf_trapno
== T_PAGEFLT
)
694 ++gd
->gd_trap_nesting_level
;
695 MAKEMPSAFE(have_mplock
);
696 trap_fatal(frame
, FALSE
, eva
);
697 --gd
->gd_trap_nesting_level
;
702 ++gd
->gd_trap_nesting_level
;
705 if (trap_mpsafe
== 0)
706 MAKEMPSAFE(have_mplock
);
709 --gd
->gd_trap_nesting_level
;
711 type
= frame
->tf_trapno
;
712 code
= frame
->tf_err
;
720 case T_PAGEFLT
: /* page fault */
721 MAKEMPSAFE(have_mplock
);
722 trap_pfault(frame
, FALSE
, eva
);
728 * The kernel may be using npx for copying or other
731 panic("kernel NPX should not happen");
737 case T_PROTFLT
: /* general protection fault */
738 case T_SEGNPFLT
: /* segment not present fault */
740 * Invalid segment selectors and out of bounds
741 * %eip's and %esp's can be set up in user mode.
742 * This causes a fault in kernel mode when the
743 * kernel tries to return to user mode. We want
744 * to get this fault so that we can fix the
745 * problem here and not have to check all the
746 * selectors and pointers when the user changes
749 if (mycpu
->gd_intr_nesting_level
== 0) {
750 if (td
->td_pcb
->pcb_onfault
) {
752 (register_t
)td
->td_pcb
->pcb_onfault
;
760 * PSL_NT can be set in user mode and isn't cleared
761 * automatically when the kernel is entered. This
762 * causes a TSS fault when the kernel attempts to
763 * `iret' because the TSS link is uninitialized. We
764 * want to get this fault so that we can fix the
765 * problem here and not every time the kernel is
768 if (frame
->tf_eflags
& PSL_NT
) {
769 frame
->tf_eflags
&= ~PSL_NT
;
774 case T_TRCTRAP
: /* trace trap */
776 if (frame
->tf_eip
== (int)IDTVEC(syscall
)) {
778 * We've just entered system mode via the
779 * syscall lcall. Continue single stepping
780 * silently until the syscall handler has
785 if (frame
->tf_eip
== (int)IDTVEC(syscall
) + 1) {
787 * The syscall handler has now saved the
788 * flags. Stop single stepping it.
790 frame
->tf_eflags
&= ~PSL_T
;
796 * Ignore debug register trace traps due to
797 * accesses in the user's address space, which
798 * can happen under several conditions such as
799 * if a user sets a watchpoint on a buffer and
800 * then passes that buffer to a system call.
801 * We still want to get TRCTRAPS for addresses
802 * in kernel space because that is useful when
803 * debugging the kernel.
805 if (user_dbreg_trap()) {
807 * Reset breakpoint bits because the
810 load_dr6(rdr6() & 0xfffffff0);
815 * Fall through (TRCTRAP kernel mode, kernel address)
819 * If DDB is enabled, let it handle the debugger trap.
820 * Otherwise, debugger traps "can't happen".
823 MAKEMPSAFE(have_mplock
);
824 if (kdb_trap (type
, 0, frame
))
829 MAKEMPSAFE(have_mplock
);
830 trap_fatal(frame
, FALSE
, eva
);
833 MAKEMPSAFE(have_mplock
);
834 trap_fatal(frame
, FALSE
, eva
);
838 * Ignore this trap generated from a spurious SIGTRAP.
840 * single stepping in / syscalls leads to spurious / SIGTRAP
843 * Haiku (c) 2007 Simon 'corecode' Schubert
849 * Translate fault for emulators (e.g. Linux)
851 if (*p
->p_sysent
->sv_transtrap
)
852 i
= (*p
->p_sysent
->sv_transtrap
)(i
, type
);
854 MAKEMPSAFE(have_mplock
);
855 trapsignal(lp
, i
, ucode
);
858 if (type
<= MAX_TRAP_MSG
) {
859 uprintf("fatal process exception: %s",
861 if ((type
== T_PAGEFLT
) || (type
== T_PROTFLT
))
862 uprintf(", fault VA = 0x%lx", (u_long
)eva
);
874 KASSERT(crit_count
== (td
->td_pri
& ~TDPRI_MASK
),
875 ("syscall: critical section count mismatch! %d/%d",
876 crit_count
/ TDPRI_CRIT
, td
->td_pri
/ TDPRI_CRIT
));
881 trap_pfault(struct trapframe
*frame
, int usermode
, vm_offset_t eva
)
884 struct vmspace
*vm
= NULL
;
889 thread_t td
= curthread
;
890 struct lwp
*lp
= td
->td_lwp
;
892 va
= trunc_page(eva
);
893 if (usermode
== FALSE
) {
895 * This is a fault on kernel virtual memory.
900 * This is a fault on non-kernel virtual memory.
901 * vm is initialized above to NULL. If curproc is NULL
902 * or curproc->p_vmspace is NULL the fault is fatal.
905 vm
= lp
->lwp_vmspace
;
913 if (frame
->tf_xflags
& PGEX_W
)
914 ftype
= VM_PROT_READ
| VM_PROT_WRITE
;
916 ftype
= VM_PROT_READ
;
918 if (map
!= &kernel_map
) {
920 * Keep swapout from messing with us during this
930 fault_flags
|= VM_FAULT_BURST
;
931 if (ftype
& VM_PROT_WRITE
)
932 fault_flags
|= VM_FAULT_DIRTY
;
934 fault_flags
|= VM_FAULT_NORMAL
;
935 rv
= vm_fault(map
, va
, ftype
, fault_flags
);
940 * Don't have to worry about process locking or stacks in the kernel.
942 rv
= vm_fault(map
, va
, ftype
, VM_FAULT_NORMAL
);
945 if (rv
== KERN_SUCCESS
)
949 if (td
->td_gd
->gd_intr_nesting_level
== 0 &&
950 td
->td_pcb
->pcb_onfault
) {
951 frame
->tf_eip
= (register_t
)td
->td_pcb
->pcb_onfault
;
954 trap_fatal(frame
, usermode
, eva
);
957 return((rv
== KERN_PROTECTION_FAILURE
) ? SIGBUS
: SIGSEGV
);
961 trap_fatal(struct trapframe
*frame
, int usermode
, vm_offset_t eva
)
963 int code
, type
, ss
, esp
;
965 code
= frame
->tf_xflags
;
966 type
= frame
->tf_trapno
;
968 if (type
<= MAX_TRAP_MSG
) {
969 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
970 type
, trap_msg
[type
],
971 (usermode
? "user" : "kernel"));
974 /* two separate prints in case of a trap on an unmapped page */
975 kprintf("mp_lock = %08x; ", mp_lock
);
976 kprintf("cpuid = %d\n", mycpu
->gd_cpuid
);
978 if (type
== T_PAGEFLT
) {
979 kprintf("fault virtual address = %p\n", (void *)eva
);
980 kprintf("fault code = %s %s, %s\n",
981 usermode
? "user" : "supervisor",
982 code
& PGEX_W
? "write" : "read",
983 code
& PGEX_P
? "protection violation" : "page not present");
985 kprintf("instruction pointer = 0x%x:0x%x\n",
986 frame
->tf_cs
& 0xffff, frame
->tf_eip
);
988 ss
= frame
->tf_ss
& 0xffff;
991 ss
= GSEL(GDATA_SEL
, SEL_KPL
);
992 esp
= (int)&frame
->tf_esp
;
994 kprintf("stack pointer = 0x%x:0x%x\n", ss
, esp
);
995 kprintf("frame pointer = 0x%x:0x%x\n", ss
, frame
->tf_ebp
);
996 kprintf("processor eflags = ");
997 if (frame
->tf_eflags
& PSL_T
)
998 kprintf("trace trap, ");
999 if (frame
->tf_eflags
& PSL_I
)
1000 kprintf("interrupt enabled, ");
1001 if (frame
->tf_eflags
& PSL_NT
)
1002 kprintf("nested task, ");
1003 if (frame
->tf_eflags
& PSL_RF
)
1004 kprintf("resume, ");
1006 if (frame
->tf_eflags
& PSL_VM
)
1009 kprintf("IOPL = %d\n", (frame
->tf_eflags
& PSL_IOPL
) >> 12);
1010 kprintf("current process = ");
1012 kprintf("%lu (%s)\n",
1013 (u_long
)curproc
->p_pid
, curproc
->p_comm
?
1014 curproc
->p_comm
: "");
1018 kprintf("current thread = pri %d ", curthread
->td_pri
);
1019 if (curthread
->td_pri
>= TDPRI_CRIT
)
1025 * we probably SHOULD have stopped the other CPUs before now!
1026 * another CPU COULD have been touching cpl at this moment...
1028 kprintf(" <- SMP: XXX");
1037 if ((debugger_on_panic
|| db_active
) && kdb_trap(type
, code
, frame
))
1040 kprintf("trap number = %d\n", type
);
1041 if (type
<= MAX_TRAP_MSG
)
1042 panic("%s", trap_msg
[type
]);
1044 panic("unknown/reserved trap");
1048 * Double fault handler. Called when a fault occurs while writing
1049 * a frame for a trap/exception onto the stack. This usually occurs
1050 * when the stack overflows (such is the case with infinite recursion,
1053 * XXX Note that the current PTD gets replaced by IdlePTD when the
1054 * task switch occurs. This means that the stack that was active at
1055 * the time of the double fault is not available at <kstack> unless
1056 * the machine was idle when the double fault occurred. The downside
1057 * of this is that "trace <ebp>" in ddb won't work.
1060 dblfault_handler(void)
1062 struct mdglobaldata
*gd
= mdcpu
;
1064 kprintf("\nFatal double fault:\n");
1065 kprintf("eip = 0x%x\n", gd
->gd_common_tss
.tss_eip
);
1066 kprintf("esp = 0x%x\n", gd
->gd_common_tss
.tss_esp
);
1067 kprintf("ebp = 0x%x\n", gd
->gd_common_tss
.tss_ebp
);
1069 /* two separate prints in case of a trap on an unmapped page */
1070 kprintf("mp_lock = %08x; ", mp_lock
);
1071 kprintf("cpuid = %d\n", mycpu
->gd_cpuid
);
1073 panic("double fault");
1077 * syscall2 - MP aware system call request C handler
1079 * A system call is essentially treated as a trap except that the
1080 * MP lock is not held on entry or return. We are responsible for
1081 * obtaining the MP lock if necessary and for handling ASTs
1082 * (e.g. a task switch) prior to return.
1087 syscall2(struct trapframe
*frame
)
1089 struct thread
*td
= curthread
;
1090 struct proc
*p
= td
->td_proc
;
1091 struct lwp
*lp
= td
->td_lwp
;
1093 struct sysent
*callp
;
1094 register_t orig_tf_eflags
;
1099 int crit_count
= td
->td_pri
& ~TDPRI_MASK
;
1102 int have_mplock
= 0;
1105 union sysunion args
;
1107 KTR_LOG(kernentry_syscall
, lp
->lwp_proc
->p_pid
, lp
->lwp_tid
,
1111 KASSERT(td
->td_mpcount
== 0, ("badmpcount syscall2 from %p", (void *)frame
->tf_eip
));
1112 if (syscall_mpsafe
== 0)
1113 MAKEMPSAFE(have_mplock
);
1115 userenter(td
, p
); /* lazy raise our priority */
1120 sticks
= (int)td
->td_sticks
;
1121 orig_tf_eflags
= frame
->tf_eflags
;
1124 * Virtual kernel intercept - if a VM context managed by a virtual
1125 * kernel issues a system call the virtual kernel handles it, not us.
1126 * Restore the virtual kernel context and return from its system
1127 * call. The current frame is copied out to the virtual kernel.
1129 if (lp
->lwp_vkernel
&& lp
->lwp_vkernel
->ve
) {
1130 vkernel_trap(lp
, frame
);
1131 error
= EJUSTRETURN
;
1136 * Get the system call parameters and account for time
1138 lp
->lwp_md
.md_regs
= frame
;
1139 params
= (caddr_t
)frame
->tf_esp
+ sizeof(int);
1140 code
= frame
->tf_eax
;
1142 if (p
->p_sysent
->sv_prepsyscall
) {
1143 (*p
->p_sysent
->sv_prepsyscall
)(
1144 frame
, (int *)(&args
.nosys
.sysmsg
+ 1),
1148 * Need to check if this is a 32 bit or 64 bit syscall.
1149 * fuword is MP aware.
1151 if (code
== SYS_syscall
) {
1153 * Code is first argument, followed by actual args.
1155 code
= fuword(params
);
1156 params
+= sizeof(int);
1157 } else if (code
== SYS___syscall
) {
1159 * Like syscall, but code is a quad, so as to maintain
1160 * quad alignment for the rest of the arguments.
1162 code
= fuword(params
);
1163 params
+= sizeof(quad_t
);
1167 code
&= p
->p_sysent
->sv_mask
;
1168 if (code
>= p
->p_sysent
->sv_size
)
1169 callp
= &p
->p_sysent
->sv_table
[0];
1171 callp
= &p
->p_sysent
->sv_table
[code
];
1173 narg
= callp
->sy_narg
& SYF_ARGMASK
;
1176 * copyin is MP aware, but the tracing code is not
1178 if (narg
&& params
) {
1179 error
= copyin(params
, (caddr_t
)(&args
.nosys
.sysmsg
+ 1),
1180 narg
* sizeof(register_t
));
1183 if (KTRPOINT(td
, KTR_SYSCALL
)) {
1184 MAKEMPSAFE(have_mplock
);
1186 ktrsyscall(lp
, code
, narg
,
1187 (void *)(&args
.nosys
.sysmsg
+ 1));
1195 if (KTRPOINT(td
, KTR_SYSCALL
)) {
1196 MAKEMPSAFE(have_mplock
);
1197 ktrsyscall(lp
, code
, narg
, (void *)(&args
.nosys
.sysmsg
+ 1));
1202 * For traditional syscall code edx is left untouched when 32 bit
1203 * results are returned. Since edx is loaded from fds[1] when the
1204 * system call returns we pre-set it here.
1206 args
.sysmsg_fds
[0] = 0;
1207 args
.sysmsg_fds
[1] = frame
->tf_edx
;
1210 * The syscall might manipulate the trap frame. If it does it
1211 * will probably return EJUSTRETURN.
1213 args
.sysmsg_frame
= frame
;
1215 STOPEVENT(p
, S_SCE
, narg
); /* MP aware */
1218 * NOTE: All system calls run MPSAFE now. The system call itself
1219 * is responsible for getting the MP lock.
1221 error
= (*callp
->sy_call
)(&args
);
1224 kprintf("system call %d returned %d\n", code
, error
);
1229 * MP SAFE (we may or may not have the MP lock at this point)
1234 * Reinitialize proc pointer `p' as it may be different
1235 * if this is a child returning from fork syscall.
1238 lp
= curthread
->td_lwp
;
1239 frame
->tf_eax
= args
.sysmsg_fds
[0];
1240 frame
->tf_edx
= args
.sysmsg_fds
[1];
1241 frame
->tf_eflags
&= ~PSL_C
;
1245 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1246 * int 0x80 is 2 bytes. We saved this in tf_err.
1248 frame
->tf_eip
-= frame
->tf_err
;
1253 panic("Unexpected EASYNC return value (for now)");
1256 if (p
->p_sysent
->sv_errsize
) {
1257 if (error
>= p
->p_sysent
->sv_errsize
)
1258 error
= -1; /* XXX */
1260 error
= p
->p_sysent
->sv_errtbl
[error
];
1262 frame
->tf_eax
= error
;
1263 frame
->tf_eflags
|= PSL_C
;
1268 * Traced syscall. trapsignal() is not MP aware.
1270 if ((orig_tf_eflags
& PSL_T
) /*&& !(orig_tf_eflags & PSL_VM)*/) {
1271 MAKEMPSAFE(have_mplock
);
1272 frame
->tf_eflags
&= ~PSL_T
;
1273 trapsignal(lp
, SIGTRAP
, TRAP_TRACE
);
1277 * Handle reschedule and other end-of-syscall issues
1279 userret(lp
, frame
, sticks
);
1282 if (KTRPOINT(td
, KTR_SYSRET
)) {
1283 MAKEMPSAFE(have_mplock
);
1284 ktrsysret(lp
, code
, error
, args
.sysmsg_result
);
1289 * This works because errno is findable through the
1290 * register set. If we ever support an emulation where this
1291 * is not the case, this code will need to be revisited.
1293 STOPEVENT(p
, S_SCX
, code
);
1298 * Release the MP lock if we had to get it
1300 KASSERT(td
->td_mpcount
== have_mplock
,
1301 ("badmpcount syscall2/end from %p", (void *)frame
->tf_eip
));
1305 KTR_LOG(kernentry_syscall_ret
, lp
->lwp_proc
->p_pid
, lp
->lwp_tid
, error
);
1307 KASSERT(crit_count
== (td
->td_pri
& ~TDPRI_MASK
),
1308 ("syscall: critical section count mismatch! %d/%d",
1309 crit_count
/ TDPRI_CRIT
, td
->td_pri
/ TDPRI_CRIT
));
1314 fork_return(struct lwp
*lp
, struct trapframe
*frame
)
1316 frame
->tf_eax
= 0; /* Child returns zero */
1317 frame
->tf_eflags
&= ~PSL_C
; /* success */
1320 generic_lwp_return(lp
, frame
);
1321 KTR_LOG(kernentry_fork_ret
, lp
->lwp_proc
->p_pid
, lp
->lwp_tid
);
1325 * Simplified back end of syscall(), used when returning from fork()
1326 * or lwp_create() directly into user mode. MP lock is held on entry and
1327 * should be released on return. This code will return back into the fork
1328 * trampoline code which then runs doreti.
1331 generic_lwp_return(struct lwp
*lp
, struct trapframe
*frame
)
1333 struct proc
*p
= lp
->lwp_proc
;
1336 * Newly forked processes are given a kernel priority. We have to
1337 * adjust the priority to a normal user priority and fake entry
1338 * into the kernel (call userenter()) to install a passive release
1339 * function just in case userret() decides to stop the process. This
1340 * can occur when ^Z races a fork. If we do not install the passive
1341 * release function the current process designation will not be
1342 * released when the thread goes to sleep.
1344 lwkt_setpri_self(TDPRI_USER_NORM
);
1345 userenter(lp
->lwp_thread
, p
);
1346 userret(lp
, frame
, 0);
1348 if (KTRPOINT(lp
->lwp_thread
, KTR_SYSRET
))
1349 ktrsysret(lp
, SYS_fork
, 0, 0);
1351 p
->p_flag
|= P_PASSIVE_ACQ
;
1353 p
->p_flag
&= ~P_PASSIVE_ACQ
;
1355 KKASSERT(lp
->lwp_thread
->td_mpcount
== 1);
1361 * doreti has turned into this. The frame is directly on the stack. We
1362 * pull everything else we need (fpu and tls context) from the current
1365 * Note on fpu interactions: In a virtual kernel, the fpu context for
1366 * an emulated user mode process is not shared with the virtual kernel's
1367 * fpu context, so we only have to 'stack' fpu contexts within the virtual
1368 * kernel itself, and not even then since the signal() contexts that we care
1369 * about save and restore the FPU state (I think anyhow).
1371 * vmspace_ctl() returns an error only if it had problems instaling the
1372 * context we supplied or problems copying data to/from our VM space.
1375 go_user(struct intrframe
*frame
)
1377 struct trapframe
*tf
= (void *)&frame
->if_gs
;
1381 * Interrupts may be disabled on entry, make sure all signals
1382 * can be received before beginning our loop.
1387 * Switch to the current simulated user process, then call
1388 * user_trap() when we break out of it (usually due to a signal).
1392 * Tell the real kernel whether it is ok to use the FP
1395 * The critical section is required to prevent an interrupt
1396 * from causing a preemptive task switch and changing
1400 if (mdcpu
->gd_npxthread
== curthread
) {
1401 tf
->tf_xflags
&= ~PGEX_FPFAULT
;
1403 tf
->tf_xflags
|= PGEX_FPFAULT
;
1407 * Run emulated user process context. This call interlocks
1408 * with new mailbox signals.
1410 * Set PGEX_U unconditionally, indicating a user frame (the
1411 * bit is normally set only by T_PAGEFLT).
1413 r
= vmspace_ctl(&curproc
->p_vmspace
->vm_pmap
, VMSPACE_CTL_RUN
,
1414 tf
, &curthread
->td_savevext
);
1416 frame
->if_xflags
|= PGEX_U
;
1418 kprintf("GO USER %d trap %d EVA %08x EIP %08x ESP %08x XFLAGS %02x/%02x\n",
1419 r
, tf
->tf_trapno
, tf
->tf_err
, tf
->tf_eip
, tf
->tf_esp
,
1420 tf
->tf_xflags
, frame
->if_xflags
);
1424 panic("vmspace_ctl failed error %d", errno
);
1426 if (tf
->tf_trapno
) {
1430 if (mycpu
->gd_reqflags
& RQF_AST_MASK
) {
1431 tf
->tf_trapno
= T_ASTFLT
;
1439 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1440 * fault (which is then passed back to the virtual kernel) if an attempt is
1441 * made to use the FP unit.
1443 * XXX this is a fairly big hack.
1446 set_vkernel_fp(struct trapframe
*frame
)
1448 struct thread
*td
= curthread
;
1450 if (frame
->tf_xflags
& PGEX_FPFAULT
) {
1451 td
->td_pcb
->pcb_flags
|= FP_VIRTFP
;
1452 if (mdcpu
->gd_npxthread
== td
)
1455 td
->td_pcb
->pcb_flags
&= ~FP_VIRTFP
;
1460 * Called from vkernel_trap() to fixup the vkernel's syscall
1461 * frame for vmspace_ctl() return.
1464 cpu_vkernel_trap(struct trapframe
*frame
, int error
)
1466 frame
->tf_eax
= error
;
1468 frame
->tf_eflags
|= PSL_C
;
1470 frame
->tf_eflags
&= ~PSL_C
;