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>
95 #include <sys/msgport2.h>
96 #include <sys/thread2.h>
100 #define MAKEMPSAFE(have_mplock) \
101 if (have_mplock == 0) { \
108 #define MAKEMPSAFE(have_mplock)
112 int (*pmath_emulate
) (struct trapframe
*);
114 extern int trapwrite (unsigned addr
);
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 * userenter() 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.
192 userenter(struct thread
*curtd
)
194 curtd
->td_release
= lwkt_passive_release
;
198 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
199 * must be completed before we can return to or try to return to userland.
201 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
202 * arithmatic on the delta calculation so the absolute tick values are
203 * truncated to an integer.
206 userret(struct lwp
*lp
, struct trapframe
*frame
, int sticks
)
208 struct proc
*p
= lp
->lwp_proc
;
212 * Charge system time if profiling. Note: times are in microseconds.
213 * This may do a copyout and block, so do it first even though it
214 * means some system time will be charged as user time.
216 if (p
->p_flag
& P_PROFIL
) {
217 addupc_task(p
, frame
->tf_eip
,
218 (u_int
)((int)lp
->lwp_thread
->td_sticks
- sticks
));
223 * If the jungle wants us dead, so be it.
225 if (lp
->lwp_flag
& LWP_WEXIT
) {
228 rel_mplock(); /* NOT REACHED */
232 * Block here if we are in a stopped state.
234 if (p
->p_stat
== SSTOP
) {
242 * Post any pending upcalls
244 if (p
->p_flag
& P_UPCALLPEND
) {
246 p
->p_flag
&= ~P_UPCALLPEND
;
253 * Post any pending signals
255 if ((sig
= CURSIG_TRACE(lp
)) != 0) {
263 * block here if we are swapped out, but still process signals
264 * (such as SIGKILL). proc0 (the swapin scheduler) is already
265 * aware of our situation, we do not have to wake it up.
267 if (p
->p_flag
& P_SWAPPEDOUT
) {
269 p
->p_flag
|= P_SWAPWAIT
;
271 if (p
->p_flag
& P_SWAPWAIT
)
272 tsleep(p
, PCATCH
, "SWOUT", 0);
273 p
->p_flag
&= ~P_SWAPWAIT
;
279 * Make sure postsig() handled request to restore old signal mask after
280 * running signal handler.
282 KKASSERT((lp
->lwp_flag
& LWP_OLDMASK
) == 0);
286 * Cleanup from userenter and any passive release that might have occured.
287 * We must reclaim the current-process designation before we can return
288 * to usermode. We also handle both LWKT and USER reschedule requests.
291 userexit(struct lwp
*lp
)
293 struct thread
*td
= lp
->lwp_thread
;
294 /* globaldata_t gd = td->td_gd; */
297 * Handle stop requests at kernel priority. Any requests queued
298 * after this loop will generate another AST.
300 while (lp
->lwp_proc
->p_stat
== SSTOP
) {
307 * Reduce our priority in preparation for a return to userland. If
308 * our passive release function was still in place, our priority was
309 * never raised and does not need to be reduced.
311 lwkt_passive_recover(td
);
314 * Become the current user scheduled process if we aren't already,
315 * and deal with reschedule requests and other factors.
317 lp
->lwp_proc
->p_usched
->acquire_curproc(lp
);
318 /* WARNING: we may have migrated cpu's */
319 /* gd = td->td_gd; */
322 #if !defined(KTR_KERNENTRY)
323 #define KTR_KERNENTRY KTR_ALL
325 KTR_INFO_MASTER(kernentry
);
326 KTR_INFO(KTR_KERNENTRY
, kernentry
, trap
, 0, "pid=%d, tid=%d, trapno=%d, eva=%p",
327 sizeof(int) + sizeof(int) + sizeof(int) + sizeof(vm_offset_t
));
328 KTR_INFO(KTR_KERNENTRY
, kernentry
, trap_ret
, 0, "pid=%d, tid=%d",
329 sizeof(int) + sizeof(int));
330 KTR_INFO(KTR_KERNENTRY
, kernentry
, syscall
, 0, "pid=%d, tid=%d, call=%d",
331 sizeof(int) + sizeof(int) + sizeof(int));
332 KTR_INFO(KTR_KERNENTRY
, kernentry
, syscall_ret
, 0, "pid=%d, tid=%d, err=%d",
333 sizeof(int) + sizeof(int) + sizeof(int));
334 KTR_INFO(KTR_KERNENTRY
, kernentry
, fork_ret
, 0, "pid=%d, tid=%d",
335 sizeof(int) + sizeof(int));
338 * Exception, fault, and trap interface to the kernel.
339 * This common code is called from assembly language IDT gate entry
340 * routines that prepare a suitable stack frame, and restore this
341 * frame after the exception has been processed.
343 * This function is also called from doreti in an interlock to handle ASTs.
344 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
346 * NOTE! We have to retrieve the fault address prior to obtaining the
347 * MP lock because get_mplock() may switch out. YYY cr2 really ought
348 * to be retrieved by the assembly code, not here.
350 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
351 * if an attempt is made to switch from a fast interrupt or IPI. This is
352 * necessary to properly take fatal kernel traps on SMP machines if
353 * get_mplock() has to block.
357 user_trap(struct trapframe
*frame
)
359 struct globaldata
*gd
= mycpu
;
360 struct thread
*td
= gd
->gd_curthread
;
361 struct lwp
*lp
= td
->td_lwp
;
364 int i
= 0, ucode
= 0, type
, code
;
369 int crit_count
= td
->td_pri
& ~TDPRI_MASK
;
376 * This is a bad kludge to avoid changing the various trapframe
377 * structures. Because we are enabled as a virtual kernel,
378 * the original tf_err field will be passed to us shifted 16
379 * over in the tf_trapno field for T_PAGEFLT.
381 if (frame
->tf_trapno
== T_PAGEFLT
)
386 kprintf("USER_TRAP AT %08x xflags %d trapno %d eva %08x\n",
387 frame
->tf_eip
, frame
->tf_xflags
, frame
->tf_trapno
, eva
);
391 * Everything coming from user mode runs through user_trap,
392 * including system calls.
394 if (frame
->tf_trapno
== T_SYSCALL80
) {
399 KTR_LOG(kernentry_trap
, lp
->lwp_proc
->p_pid
, lp
->lwp_tid
,
400 frame
->tf_trapno
, eva
);
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 (lp
->lwp_vkernel
&& lp
->lwp_vkernel
->ve
&&
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 (lp
->lwp_vkernel
&& lp
->lwp_vkernel
->ve
) {
595 vkernel_trap(lp
, 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(lp
, 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
);
629 KTR_LOG(kernentry_trap_ret
, lp
->lwp_proc
->p_pid
, lp
->lwp_tid
);
631 KASSERT(crit_count
== (td
->td_pri
& ~TDPRI_MASK
),
632 ("syscall: critical section count mismatch! %d/%d",
633 crit_count
/ TDPRI_CRIT
, td
->td_pri
/ TDPRI_CRIT
));
638 kern_trap(struct trapframe
*frame
)
640 struct globaldata
*gd
= mycpu
;
641 struct thread
*td
= gd
->gd_curthread
;
644 int i
= 0, ucode
= 0, type
, code
;
649 int crit_count
= td
->td_pri
& ~TDPRI_MASK
;
656 if (frame
->tf_trapno
== T_PAGEFLT
)
663 ++gd
->gd_trap_nesting_level
;
664 MAKEMPSAFE(have_mplock
);
665 trap_fatal(frame
, FALSE
, eva
);
666 --gd
->gd_trap_nesting_level
;
671 ++gd
->gd_trap_nesting_level
;
674 if (trap_mpsafe
== 0)
675 MAKEMPSAFE(have_mplock
);
678 --gd
->gd_trap_nesting_level
;
680 type
= frame
->tf_trapno
;
681 code
= frame
->tf_err
;
689 case T_PAGEFLT
: /* page fault */
690 MAKEMPSAFE(have_mplock
);
691 trap_pfault(frame
, FALSE
, eva
);
697 * The kernel may be using npx for copying or other
700 panic("kernel NPX should not happen");
706 case T_PROTFLT
: /* general protection fault */
707 case T_SEGNPFLT
: /* segment not present fault */
709 * Invalid segment selectors and out of bounds
710 * %eip's and %esp's can be set up in user mode.
711 * This causes a fault in kernel mode when the
712 * kernel tries to return to user mode. We want
713 * to get this fault so that we can fix the
714 * problem here and not have to check all the
715 * selectors and pointers when the user changes
718 if (mycpu
->gd_intr_nesting_level
== 0) {
719 if (td
->td_pcb
->pcb_onfault
) {
721 (register_t
)td
->td_pcb
->pcb_onfault
;
729 * PSL_NT can be set in user mode and isn't cleared
730 * automatically when the kernel is entered. This
731 * causes a TSS fault when the kernel attempts to
732 * `iret' because the TSS link is uninitialized. We
733 * want to get this fault so that we can fix the
734 * problem here and not every time the kernel is
737 if (frame
->tf_eflags
& PSL_NT
) {
738 frame
->tf_eflags
&= ~PSL_NT
;
743 case T_TRCTRAP
: /* trace trap */
745 if (frame
->tf_eip
== (int)IDTVEC(syscall
)) {
747 * We've just entered system mode via the
748 * syscall lcall. Continue single stepping
749 * silently until the syscall handler has
754 if (frame
->tf_eip
== (int)IDTVEC(syscall
) + 1) {
756 * The syscall handler has now saved the
757 * flags. Stop single stepping it.
759 frame
->tf_eflags
&= ~PSL_T
;
765 * Ignore debug register trace traps due to
766 * accesses in the user's address space, which
767 * can happen under several conditions such as
768 * if a user sets a watchpoint on a buffer and
769 * then passes that buffer to a system call.
770 * We still want to get TRCTRAPS for addresses
771 * in kernel space because that is useful when
772 * debugging the kernel.
774 if (user_dbreg_trap()) {
776 * Reset breakpoint bits because the
779 load_dr6(rdr6() & 0xfffffff0);
784 * Fall through (TRCTRAP kernel mode, kernel address)
788 * If DDB is enabled, let it handle the debugger trap.
789 * Otherwise, debugger traps "can't happen".
792 MAKEMPSAFE(have_mplock
);
793 if (kdb_trap (type
, 0, frame
))
798 MAKEMPSAFE(have_mplock
);
799 trap_fatal(frame
, FALSE
, eva
);
802 MAKEMPSAFE(have_mplock
);
803 trap_fatal(frame
, FALSE
, eva
);
807 * Ignore this trap generated from a spurious SIGTRAP.
809 * single stepping in / syscalls leads to spurious / SIGTRAP
812 * Haiku (c) 2007 Simon 'corecode' Schubert
818 * Translate fault for emulators (e.g. Linux)
820 if (*p
->p_sysent
->sv_transtrap
)
821 i
= (*p
->p_sysent
->sv_transtrap
)(i
, type
);
823 MAKEMPSAFE(have_mplock
);
824 trapsignal(lp
, i
, ucode
);
827 if (type
<= MAX_TRAP_MSG
) {
828 uprintf("fatal process exception: %s",
830 if ((type
== T_PAGEFLT
) || (type
== T_PROTFLT
))
831 uprintf(", fault VA = 0x%lx", (u_long
)eva
);
843 KASSERT(crit_count
== (td
->td_pri
& ~TDPRI_MASK
),
844 ("syscall: critical section count mismatch! %d/%d",
845 crit_count
/ TDPRI_CRIT
, td
->td_pri
/ TDPRI_CRIT
));
850 trap_pfault(struct trapframe
*frame
, int usermode
, vm_offset_t eva
)
853 struct vmspace
*vm
= NULL
;
857 thread_t td
= curthread
;
858 struct lwp
*lp
= td
->td_lwp
;
860 va
= trunc_page(eva
);
861 if (usermode
== FALSE
) {
863 * This is a fault on kernel virtual memory.
868 * This is a fault on non-kernel virtual memory.
869 * vm is initialized above to NULL. If curproc is NULL
870 * or curproc->p_vmspace is NULL the fault is fatal.
873 vm
= lp
->lwp_vmspace
;
881 if (frame
->tf_xflags
& PGEX_W
)
882 ftype
= VM_PROT_READ
| VM_PROT_WRITE
;
884 ftype
= VM_PROT_READ
;
886 if (map
!= &kernel_map
) {
888 * Keep swapout from messing with us during this
894 * Grow the stack if necessary
896 /* grow_stack returns false only if va falls into
897 * a growable stack region and the stack growth
898 * fails. It returns true if va was not within
899 * a growable stack region, or if the stack
902 if (!grow_stack (lp
->lwp_proc
, va
)) {
908 /* Fault in the user page: */
909 rv
= vm_fault(map
, va
, ftype
,
910 (ftype
& VM_PROT_WRITE
) ? VM_FAULT_DIRTY
916 * Don't have to worry about process locking or stacks in the kernel.
918 rv
= vm_fault(map
, va
, ftype
, VM_FAULT_NORMAL
);
921 if (rv
== KERN_SUCCESS
)
925 if (td
->td_gd
->gd_intr_nesting_level
== 0 &&
926 td
->td_pcb
->pcb_onfault
) {
927 frame
->tf_eip
= (register_t
)td
->td_pcb
->pcb_onfault
;
930 trap_fatal(frame
, usermode
, eva
);
933 return((rv
== KERN_PROTECTION_FAILURE
) ? SIGBUS
: SIGSEGV
);
937 trap_fatal(struct trapframe
*frame
, int usermode
, vm_offset_t eva
)
939 int code
, type
, ss
, esp
;
941 code
= frame
->tf_xflags
;
942 type
= frame
->tf_trapno
;
944 if (type
<= MAX_TRAP_MSG
) {
945 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
946 type
, trap_msg
[type
],
947 (usermode
? "user" : "kernel"));
950 /* two separate prints in case of a trap on an unmapped page */
951 kprintf("mp_lock = %08x; ", mp_lock
);
952 kprintf("cpuid = %d\n", mycpu
->gd_cpuid
);
954 if (type
== T_PAGEFLT
) {
955 kprintf("fault virtual address = %p\n", (void *)eva
);
956 kprintf("fault code = %s %s, %s\n",
957 usermode
? "user" : "supervisor",
958 code
& PGEX_W
? "write" : "read",
959 code
& PGEX_P
? "protection violation" : "page not present");
961 kprintf("instruction pointer = 0x%x:0x%x\n",
962 frame
->tf_cs
& 0xffff, frame
->tf_eip
);
964 ss
= frame
->tf_ss
& 0xffff;
967 ss
= GSEL(GDATA_SEL
, SEL_KPL
);
968 esp
= (int)&frame
->tf_esp
;
970 kprintf("stack pointer = 0x%x:0x%x\n", ss
, esp
);
971 kprintf("frame pointer = 0x%x:0x%x\n", ss
, frame
->tf_ebp
);
972 kprintf("processor eflags = ");
973 if (frame
->tf_eflags
& PSL_T
)
974 kprintf("trace trap, ");
975 if (frame
->tf_eflags
& PSL_I
)
976 kprintf("interrupt enabled, ");
977 if (frame
->tf_eflags
& PSL_NT
)
978 kprintf("nested task, ");
979 if (frame
->tf_eflags
& PSL_RF
)
982 if (frame
->tf_eflags
& PSL_VM
)
985 kprintf("IOPL = %d\n", (frame
->tf_eflags
& PSL_IOPL
) >> 12);
986 kprintf("current process = ");
988 kprintf("%lu (%s)\n",
989 (u_long
)curproc
->p_pid
, curproc
->p_comm
?
990 curproc
->p_comm
: "");
994 kprintf("current thread = pri %d ", curthread
->td_pri
);
995 if (curthread
->td_pri
>= TDPRI_CRIT
)
1001 * we probably SHOULD have stopped the other CPUs before now!
1002 * another CPU COULD have been touching cpl at this moment...
1004 kprintf(" <- SMP: XXX");
1013 if ((debugger_on_panic
|| db_active
) && kdb_trap(type
, code
, frame
))
1016 kprintf("trap number = %d\n", type
);
1017 if (type
<= MAX_TRAP_MSG
)
1018 panic("%s", trap_msg
[type
]);
1020 panic("unknown/reserved trap");
1024 * Double fault handler. Called when a fault occurs while writing
1025 * a frame for a trap/exception onto the stack. This usually occurs
1026 * when the stack overflows (such is the case with infinite recursion,
1029 * XXX Note that the current PTD gets replaced by IdlePTD when the
1030 * task switch occurs. This means that the stack that was active at
1031 * the time of the double fault is not available at <kstack> unless
1032 * the machine was idle when the double fault occurred. The downside
1033 * of this is that "trace <ebp>" in ddb won't work.
1036 dblfault_handler(void)
1038 struct mdglobaldata
*gd
= mdcpu
;
1040 kprintf("\nFatal double fault:\n");
1041 kprintf("eip = 0x%x\n", gd
->gd_common_tss
.tss_eip
);
1042 kprintf("esp = 0x%x\n", gd
->gd_common_tss
.tss_esp
);
1043 kprintf("ebp = 0x%x\n", gd
->gd_common_tss
.tss_ebp
);
1045 /* two separate prints in case of a trap on an unmapped page */
1046 kprintf("mp_lock = %08x; ", mp_lock
);
1047 kprintf("cpuid = %d\n", mycpu
->gd_cpuid
);
1049 panic("double fault");
1053 * Compensate for 386 brain damage (missing URKR).
1054 * This is a little simpler than the pagefault handler in trap() because
1055 * it the page tables have already been faulted in and high addresses
1056 * are thrown out early for other reasons.
1059 trapwrite(unsigned addr
)
1066 va
= trunc_page((vm_offset_t
)addr
);
1068 * XXX - MAX is END. Changed > to >= for temp. fix.
1070 if (va
>= VM_MAX_USER_ADDRESS
)
1073 lp
= curthread
->td_lwp
;
1074 vm
= lp
->lwp_vmspace
;
1076 PHOLD(lp
->lwp_proc
);
1078 if (!grow_stack (lp
->lwp_proc
, va
)) {
1079 PRELE(lp
->lwp_proc
);
1084 * fault the data page
1086 rv
= vm_fault(&vm
->vm_map
, va
, VM_PROT_WRITE
, VM_FAULT_DIRTY
);
1088 PRELE(lp
->lwp_proc
);
1090 if (rv
!= KERN_SUCCESS
)
1097 * syscall2 - MP aware system call request C handler
1099 * A system call is essentially treated as a trap except that the
1100 * MP lock is not held on entry or return. We are responsible for
1101 * obtaining the MP lock if necessary and for handling ASTs
1102 * (e.g. a task switch) prior to return.
1104 * In general, only simple access and manipulation of curproc and
1105 * the current stack is allowed without having to hold MP lock.
1107 * MPSAFE - note that large sections of this routine are run without
1112 syscall2(struct trapframe
*frame
)
1114 struct thread
*td
= curthread
;
1115 struct proc
*p
= td
->td_proc
;
1116 struct lwp
*lp
= td
->td_lwp
;
1118 struct sysent
*callp
;
1119 register_t orig_tf_eflags
;
1124 int crit_count
= td
->td_pri
& ~TDPRI_MASK
;
1127 int have_mplock
= 0;
1130 union sysunion args
;
1132 KTR_LOG(kernentry_syscall
, lp
->lwp_proc
->p_pid
, lp
->lwp_tid
,
1136 KASSERT(td
->td_mpcount
== 0, ("badmpcount syscall2 from %p", (void *)frame
->tf_eip
));
1137 if (syscall_mpsafe
== 0)
1138 MAKEMPSAFE(have_mplock
);
1140 userenter(td
); /* lazy raise our priority */
1145 sticks
= (int)td
->td_sticks
;
1146 orig_tf_eflags
= frame
->tf_eflags
;
1149 * Virtual kernel intercept - if a VM context managed by a virtual
1150 * kernel issues a system call the virtual kernel handles it, not us.
1151 * Restore the virtual kernel context and return from its system
1152 * call. The current frame is copied out to the virtual kernel.
1154 if (lp
->lwp_vkernel
&& lp
->lwp_vkernel
->ve
) {
1155 error
= vkernel_trap(lp
, frame
);
1156 frame
->tf_eax
= error
;
1158 frame
->tf_eflags
|= PSL_C
;
1159 error
= EJUSTRETURN
;
1164 * Get the system call parameters and account for time
1166 lp
->lwp_md
.md_regs
= frame
;
1167 params
= (caddr_t
)frame
->tf_esp
+ sizeof(int);
1168 code
= frame
->tf_eax
;
1170 if (p
->p_sysent
->sv_prepsyscall
) {
1171 (*p
->p_sysent
->sv_prepsyscall
)(
1172 frame
, (int *)(&args
.nosys
.sysmsg
+ 1),
1176 * Need to check if this is a 32 bit or 64 bit syscall.
1177 * fuword is MP aware.
1179 if (code
== SYS_syscall
) {
1181 * Code is first argument, followed by actual args.
1183 code
= fuword(params
);
1184 params
+= sizeof(int);
1185 } else if (code
== SYS___syscall
) {
1187 * Like syscall, but code is a quad, so as to maintain
1188 * quad alignment for the rest of the arguments.
1190 code
= fuword(params
);
1191 params
+= sizeof(quad_t
);
1195 code
&= p
->p_sysent
->sv_mask
;
1196 if (code
>= p
->p_sysent
->sv_size
)
1197 callp
= &p
->p_sysent
->sv_table
[0];
1199 callp
= &p
->p_sysent
->sv_table
[code
];
1201 narg
= callp
->sy_narg
& SYF_ARGMASK
;
1204 * copyin is MP aware, but the tracing code is not
1206 if (narg
&& params
) {
1207 error
= copyin(params
, (caddr_t
)(&args
.nosys
.sysmsg
+ 1),
1208 narg
* sizeof(register_t
));
1211 if (KTRPOINT(td
, KTR_SYSCALL
)) {
1212 MAKEMPSAFE(have_mplock
);
1214 ktrsyscall(lp
, code
, narg
,
1215 (void *)(&args
.nosys
.sysmsg
+ 1));
1223 if (KTRPOINT(td
, KTR_SYSCALL
)) {
1224 MAKEMPSAFE(have_mplock
);
1225 ktrsyscall(lp
, code
, narg
, (void *)(&args
.nosys
.sysmsg
+ 1));
1230 * For traditional syscall code edx is left untouched when 32 bit
1231 * results are returned. Since edx is loaded from fds[1] when the
1232 * system call returns we pre-set it here.
1234 args
.sysmsg_fds
[0] = 0;
1235 args
.sysmsg_fds
[1] = frame
->tf_edx
;
1238 * The syscall might manipulate the trap frame. If it does it
1239 * will probably return EJUSTRETURN.
1241 args
.sysmsg_frame
= frame
;
1243 STOPEVENT(p
, S_SCE
, narg
); /* MP aware */
1246 * NOTE: All system calls run MPSAFE now. The system call itself
1247 * is responsible for getting the MP lock.
1249 error
= (*callp
->sy_call
)(&args
);
1252 kprintf("system call %d returned %d\n", code
, error
);
1257 * MP SAFE (we may or may not have the MP lock at this point)
1262 * Reinitialize proc pointer `p' as it may be different
1263 * if this is a child returning from fork syscall.
1266 lp
= curthread
->td_lwp
;
1267 frame
->tf_eax
= args
.sysmsg_fds
[0];
1268 frame
->tf_edx
= args
.sysmsg_fds
[1];
1269 frame
->tf_eflags
&= ~PSL_C
;
1273 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1274 * int 0x80 is 2 bytes. We saved this in tf_err.
1276 frame
->tf_eip
-= frame
->tf_err
;
1281 panic("Unexpected EASYNC return value (for now)");
1284 if (p
->p_sysent
->sv_errsize
) {
1285 if (error
>= p
->p_sysent
->sv_errsize
)
1286 error
= -1; /* XXX */
1288 error
= p
->p_sysent
->sv_errtbl
[error
];
1290 frame
->tf_eax
= error
;
1291 frame
->tf_eflags
|= PSL_C
;
1296 * Traced syscall. trapsignal() is not MP aware.
1298 if ((orig_tf_eflags
& PSL_T
) /*&& !(orig_tf_eflags & PSL_VM)*/) {
1299 MAKEMPSAFE(have_mplock
);
1300 frame
->tf_eflags
&= ~PSL_T
;
1301 trapsignal(lp
, SIGTRAP
, 0);
1305 * Handle reschedule and other end-of-syscall issues
1307 userret(lp
, frame
, sticks
);
1310 if (KTRPOINT(td
, KTR_SYSRET
)) {
1311 MAKEMPSAFE(have_mplock
);
1312 ktrsysret(lp
, code
, error
, args
.sysmsg_result
);
1317 * This works because errno is findable through the
1318 * register set. If we ever support an emulation where this
1319 * is not the case, this code will need to be revisited.
1321 STOPEVENT(p
, S_SCX
, code
);
1326 * Release the MP lock if we had to get it
1328 KASSERT(td
->td_mpcount
== have_mplock
,
1329 ("badmpcount syscall2/end from %p", (void *)frame
->tf_eip
));
1333 KTR_LOG(kernentry_syscall_ret
, lp
->lwp_proc
->p_pid
, lp
->lwp_tid
, error
);
1335 KASSERT(crit_count
== (td
->td_pri
& ~TDPRI_MASK
),
1336 ("syscall: critical section count mismatch! %d/%d",
1337 crit_count
/ TDPRI_CRIT
, td
->td_pri
/ TDPRI_CRIT
));
1342 fork_return(struct lwp
*lp
, struct trapframe
*frame
)
1344 frame
->tf_eax
= 0; /* Child returns zero */
1345 frame
->tf_eflags
&= ~PSL_C
; /* success */
1348 generic_lwp_return(lp
, frame
);
1349 KTR_LOG(kernentry_fork_ret
, lp
->lwp_proc
->p_pid
, lp
->lwp_tid
);
1353 * Simplified back end of syscall(), used when returning from fork()
1354 * or lwp_create() directly into user mode. MP lock is held on entry and
1355 * should be released on return. This code will return back into the fork
1356 * trampoline code which then runs doreti.
1359 generic_lwp_return(struct lwp
*lp
, struct trapframe
*frame
)
1361 struct proc
*p
= lp
->lwp_proc
;
1364 * Newly forked processes are given a kernel priority. We have to
1365 * adjust the priority to a normal user priority and fake entry
1366 * into the kernel (call userenter()) to install a passive release
1367 * function just in case userret() decides to stop the process. This
1368 * can occur when ^Z races a fork. If we do not install the passive
1369 * release function the current process designation will not be
1370 * released when the thread goes to sleep.
1372 lwkt_setpri_self(TDPRI_USER_NORM
);
1373 userenter(lp
->lwp_thread
);
1374 userret(lp
, frame
, 0);
1376 if (KTRPOINT(lp
->lwp_thread
, KTR_SYSRET
))
1377 ktrsysret(lp
, SYS_fork
, 0, 0);
1379 p
->p_flag
|= P_PASSIVE_ACQ
;
1381 p
->p_flag
&= ~P_PASSIVE_ACQ
;
1383 KKASSERT(lp
->lwp_thread
->td_mpcount
== 1);
1389 * doreti has turned into this. The frame is directly on the stack. We
1390 * pull everything else we need (fpu and tls context) from the current
1393 * Note on fpu interactions: In a virtual kernel, the fpu context for
1394 * an emulated user mode process is not shared with the virtual kernel's
1395 * fpu context, so we only have to 'stack' fpu contexts within the virtual
1396 * kernel itself, and not even then since the signal() contexts that we care
1397 * about save and restore the FPU state (I think anyhow).
1399 * vmspace_ctl() returns an error only if it had problems instaling the
1400 * context we supplied or problems copying data to/from our VM space.
1403 go_user(struct intrframe
*frame
)
1405 struct trapframe
*tf
= (void *)&frame
->if_gs
;
1409 * Interrupts may be disabled on entry, make sure all signals
1410 * can be received before beginning our loop.
1415 * Switch to the current simulated user process, then call
1416 * user_trap() when we break out of it (usually due to a signal).
1420 * Tell the real kernel whether it is ok to use the FP
1423 * The critical section is required to prevent an interrupt
1424 * from causing a preemptive task switch and changing
1428 if (mdcpu
->gd_npxthread
== curthread
) {
1429 tf
->tf_xflags
&= ~PGEX_FPFAULT
;
1431 tf
->tf_xflags
|= PGEX_FPFAULT
;
1435 * Run emulated user process context. This call interlocks
1436 * with new mailbox signals.
1438 * Set PGEX_U unconditionally, indicating a user frame (the
1439 * bit is normally set only by T_PAGEFLT).
1441 r
= vmspace_ctl(&curproc
->p_vmspace
->vm_pmap
, VMSPACE_CTL_RUN
,
1442 tf
, &curthread
->td_savevext
);
1444 frame
->if_xflags
|= PGEX_U
;
1446 kprintf("GO USER %d trap %d EVA %08x EIP %08x ESP %08x XFLAGS %02x/%02x\n",
1447 r
, tf
->tf_trapno
, tf
->tf_err
, tf
->tf_eip
, tf
->tf_esp
,
1448 tf
->tf_xflags
, frame
->if_xflags
);
1452 panic("vmspace_ctl failed");
1454 if (tf
->tf_trapno
) {
1458 if (mycpu
->gd_reqflags
& RQF_AST_MASK
) {
1459 tf
->tf_trapno
= T_ASTFLT
;
1467 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1468 * fault (which is then passed back to the virtual kernel) if an attempt is
1469 * made to use the FP unit.
1471 * XXX this is a fairly big hack.
1474 set_vkernel_fp(struct trapframe
*frame
)
1476 struct thread
*td
= curthread
;
1478 if (frame
->tf_xflags
& PGEX_FPFAULT
) {
1479 td
->td_pcb
->pcb_flags
|= FP_VIRTFP
;
1480 if (mdcpu
->gd_npxthread
== td
)
1483 td
->td_pcb
->pcb_flags
&= ~FP_VIRTFP
;