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 * Passive USER->KERNEL transition. This only occurs if we block in the
187 * kernel while still holding our userland priority. We have to fixup our
188 * priority in order to avoid potential deadlocks before we allow the system
189 * to switch us to another thread.
192 passive_release(struct thread
*td
)
194 struct lwp
*lp
= td
->td_lwp
;
196 td
->td_release
= NULL
;
197 lwkt_setpri_self(TDPRI_KERN_USER
);
198 lp
->lwp_proc
->p_usched
->release_curproc(lp
);
202 * userenter() passively intercepts the thread switch function to increase
203 * the thread priority from a user priority to a kernel priority, reducing
204 * syscall and trap overhead for the case where no switch occurs.
208 userenter(struct thread
*curtd
)
210 curtd
->td_release
= passive_release
;
214 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
215 * must be completed before we can return to or try to return to userland.
217 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
218 * arithmatic on the delta calculation so the absolute tick values are
219 * truncated to an integer.
222 userret(struct lwp
*lp
, struct trapframe
*frame
, int sticks
)
224 struct proc
*p
= lp
->lwp_proc
;
228 * Charge system time if profiling. Note: times are in microseconds.
229 * This may do a copyout and block, so do it first even though it
230 * means some system time will be charged as user time.
232 if (p
->p_flag
& P_PROFIL
) {
233 addupc_task(p
, frame
->tf_eip
,
234 (u_int
)((int)lp
->lwp_thread
->td_sticks
- sticks
));
239 * If the jungle wants us dead, so be it.
241 if (lp
->lwp_flag
& LWP_WEXIT
) {
244 rel_mplock(); /* NOT REACHED */
248 * Block here if we are in a stopped state.
250 if (p
->p_stat
== SSTOP
) {
258 * Post any pending upcalls
260 if (p
->p_flag
& P_UPCALLPEND
) {
262 p
->p_flag
&= ~P_UPCALLPEND
;
269 * Post any pending signals
271 if ((sig
= CURSIG(lp
)) != 0) {
279 * block here if we are swapped out, but still process signals
280 * (such as SIGKILL). proc0 (the swapin scheduler) is already
281 * aware of our situation, we do not have to wake it up.
283 if (p
->p_flag
& P_SWAPPEDOUT
) {
285 p
->p_flag
|= P_SWAPWAIT
;
287 if (p
->p_flag
& P_SWAPWAIT
)
288 tsleep(p
, PCATCH
, "SWOUT", 0);
289 p
->p_flag
&= ~P_SWAPWAIT
;
295 * Make sure postsig() handled request to restore old signal mask after
296 * running signal handler.
298 KKASSERT((lp
->lwp_flag
& LWP_OLDMASK
) == 0);
302 * Cleanup from userenter and any passive release that might have occured.
303 * We must reclaim the current-process designation before we can return
304 * to usermode. We also handle both LWKT and USER reschedule requests.
307 userexit(struct lwp
*lp
)
309 struct thread
*td
= lp
->lwp_thread
;
310 globaldata_t gd
= td
->td_gd
;
313 * Handle stop requests at kernel priority. Any requests queued
314 * after this loop will generate another AST.
316 while (lp
->lwp_proc
->p_stat
== SSTOP
) {
323 * Reduce our priority in preparation for a return to userland. If
324 * our passive release function was still in place, our priority was
325 * never raised and does not need to be reduced.
327 if (td
->td_release
== NULL
)
328 lwkt_setpri_self(TDPRI_USER_NORM
);
329 td
->td_release
= NULL
;
332 * Become the current user scheduled process if we aren't already,
333 * and deal with reschedule requests and other factors.
335 lp
->lwp_proc
->p_usched
->acquire_curproc(lp
);
336 /* WARNING: we may have migrated cpu's */
337 /* gd = td->td_gd; */
340 #if !defined(KTR_KERNENTRY)
341 #define KTR_KERNENTRY KTR_ALL
343 KTR_INFO_MASTER(kernentry
);
344 KTR_INFO(KTR_KERNENTRY
, kernentry
, trap
, 0, "pid=%d, tid=%d, trapno=%d, eva=%p",
345 sizeof(int) + sizeof(int) + sizeof(int) + sizeof(vm_offset_t
));
346 KTR_INFO(KTR_KERNENTRY
, kernentry
, trap_ret
, 0, "pid=%d, tid=%d",
347 sizeof(int) + sizeof(int));
348 KTR_INFO(KTR_KERNENTRY
, kernentry
, syscall
, 0, "pid=%d, tid=%d, call=%d",
349 sizeof(int) + sizeof(int) + sizeof(int));
350 KTR_INFO(KTR_KERNENTRY
, kernentry
, syscall_ret
, 0, "pid=%d, tid=%d, err=%d",
351 sizeof(int) + sizeof(int) + sizeof(int));
352 KTR_INFO(KTR_KERNENTRY
, kernentry
, fork_ret
, 0, "pid=%d, tid=%d",
353 sizeof(int) + sizeof(int));
356 * Exception, fault, and trap interface to the kernel.
357 * This common code is called from assembly language IDT gate entry
358 * routines that prepare a suitable stack frame, and restore this
359 * frame after the exception has been processed.
361 * This function is also called from doreti in an interlock to handle ASTs.
362 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
364 * NOTE! We have to retrieve the fault address prior to obtaining the
365 * MP lock because get_mplock() may switch out. YYY cr2 really ought
366 * to be retrieved by the assembly code, not here.
368 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
369 * if an attempt is made to switch from a fast interrupt or IPI. This is
370 * necessary to properly take fatal kernel traps on SMP machines if
371 * get_mplock() has to block.
375 user_trap(struct trapframe
*frame
)
377 struct globaldata
*gd
= mycpu
;
378 struct thread
*td
= gd
->gd_curthread
;
379 struct lwp
*lp
= td
->td_lwp
;
382 int i
= 0, ucode
= 0, type
, code
;
387 int crit_count
= td
->td_pri
& ~TDPRI_MASK
;
394 * This is a bad kludge to avoid changing the various trapframe
395 * structures. Because we are enabled as a virtual kernel,
396 * the original tf_err field will be passed to us shifted 16
397 * over in the tf_trapno field for T_PAGEFLT.
399 if (frame
->tf_trapno
== T_PAGEFLT
)
404 kprintf("USER_TRAP AT %08x xflags %d trapno %d eva %08x\n",
405 frame
->tf_eip
, frame
->tf_xflags
, frame
->tf_trapno
, eva
);
409 * Everything coming from user mode runs through user_trap,
410 * including system calls.
412 if (frame
->tf_trapno
== T_SYSCALL80
) {
417 KTR_LOG(kernentry_trap
, lp
->lwp_proc
->p_pid
, lp
->lwp_tid
,
418 frame
->tf_trapno
, eva
);
422 eva
= (frame
->tf_trapno
== T_PAGEFLT
? rcr2() : 0);
423 ++gd
->gd_trap_nesting_level
;
424 MAKEMPSAFE(have_mplock
);
425 trap_fatal(frame
, TRUE
, eva
);
426 --gd
->gd_trap_nesting_level
;
431 ++gd
->gd_trap_nesting_level
;
433 if (trap_mpsafe
== 0)
434 MAKEMPSAFE(have_mplock
);
437 --gd
->gd_trap_nesting_level
;
439 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
442 type
= frame
->tf_trapno
;
443 code
= frame
->tf_err
;
447 sticks
= (int)td
->td_sticks
;
448 lp
->lwp_md
.md_regs
= frame
;
451 case T_PRIVINFLT
: /* privileged instruction fault */
456 case T_BPTFLT
: /* bpt instruction fault */
457 case T_TRCTRAP
: /* trace trap */
458 frame
->tf_eflags
&= ~PSL_T
;
462 case T_ARITHTRAP
: /* arithmetic trap */
467 case T_ASTFLT
: /* Allow process switch */
468 mycpu
->gd_cnt
.v_soft
++;
469 if (mycpu
->gd_reqflags
& RQF_AST_OWEUPC
) {
470 atomic_clear_int_nonlocked(&mycpu
->gd_reqflags
,
472 addupc_task(p
, p
->p_prof
.pr_addr
,
478 * The following two traps can happen in
479 * vm86 mode, and, if so, we want to handle
482 case T_PROTFLT
: /* general protection fault */
483 case T_STKFLT
: /* stack fault */
485 if (frame
->tf_eflags
& PSL_VM
) {
486 i
= vm86_emulate((struct vm86frame
*)frame
);
494 case T_SEGNPFLT
: /* segment not present fault */
495 case T_TSSFLT
: /* invalid TSS fault */
496 case T_DOUBLEFLT
: /* double fault */
498 ucode
= code
+ BUS_SEGM_FAULT
;
502 case T_PAGEFLT
: /* page fault */
503 MAKEMPSAFE(have_mplock
);
504 i
= trap_pfault(frame
, TRUE
, eva
);
507 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
517 case T_DIVIDE
: /* integer divide fault */
524 MAKEMPSAFE(have_mplock
);
525 /* machine/parity/power fail/"kitchen sink" faults */
526 if (isa_nmi(code
) == 0) {
529 * NMI can be hooked up to a pushbutton
533 kprintf ("NMI ... going to debugger\n");
534 kdb_trap (type
, 0, frame
);
538 } else if (panic_on_nmi
)
539 panic("NMI indicates hardware failure");
541 #endif /* NISA > 0 */
543 case T_OFLOW
: /* integer overflow fault */
548 case T_BOUND
: /* bounds check fault */
555 * Virtual kernel intercept - pass the DNA exception
556 * to the (emulated) virtual kernel if it asked to handle
557 * it. This occurs when the virtual kernel is holding
558 * onto the FP context for a different emulated
559 * process then the one currently running.
561 * We must still call npxdna() since we may have
562 * saved FP state that the (emulated) virtual kernel
563 * needs to hand over to a different emulated process.
565 if (lp
->lwp_vkernel
&& lp
->lwp_vkernel
->ve
&&
566 (td
->td_pcb
->pcb_flags
& FP_VIRTFP
)
573 * The kernel may have switched out the FP unit's
574 * state, causing the user process to take a fault
575 * when it tries to use the FP unit. Restore the
581 if (!pmath_emulate
) {
583 ucode
= FPE_FPU_NP_TRAP
;
586 i
= (*pmath_emulate
)(frame
);
588 if (!(frame
->tf_eflags
& PSL_T
))
590 frame
->tf_eflags
&= ~PSL_T
;
593 /* else ucode = emulator_only_knows() XXX */
596 case T_FPOPFLT
: /* FPU operand fetch fault */
601 case T_XMMFLT
: /* SIMD floating-point exception */
608 * Virtual kernel intercept - if the fault is directly related to a
609 * VM context managed by a virtual kernel then let the virtual kernel
612 if (lp
->lwp_vkernel
&& lp
->lwp_vkernel
->ve
) {
613 vkernel_trap(lp
, frame
);
618 * Translate fault for emulators (e.g. Linux)
620 if (*p
->p_sysent
->sv_transtrap
)
621 i
= (*p
->p_sysent
->sv_transtrap
)(i
, type
);
623 MAKEMPSAFE(have_mplock
);
624 trapsignal(lp
, i
, ucode
);
627 if (type
<= MAX_TRAP_MSG
) {
628 uprintf("fatal process exception: %s",
630 if ((type
== T_PAGEFLT
) || (type
== T_PROTFLT
))
631 uprintf(", fault VA = 0x%lx", (u_long
)eva
);
638 KASSERT(td
->td_mpcount
== have_mplock
, ("badmpcount trap/end from %p", (void *)frame
->tf_eip
));
640 userret(lp
, frame
, sticks
);
647 KTR_LOG(kernentry_trap_ret
, lp
->lwp_proc
->p_pid
, lp
->lwp_tid
);
649 KASSERT(crit_count
== (td
->td_pri
& ~TDPRI_MASK
),
650 ("syscall: critical section count mismatch! %d/%d",
651 crit_count
/ TDPRI_CRIT
, td
->td_pri
/ TDPRI_CRIT
));
656 kern_trap(struct trapframe
*frame
)
658 struct globaldata
*gd
= mycpu
;
659 struct thread
*td
= gd
->gd_curthread
;
662 int i
= 0, ucode
= 0, type
, code
;
667 int crit_count
= td
->td_pri
& ~TDPRI_MASK
;
674 if (frame
->tf_trapno
== T_PAGEFLT
)
681 ++gd
->gd_trap_nesting_level
;
682 MAKEMPSAFE(have_mplock
);
683 trap_fatal(frame
, FALSE
, eva
);
684 --gd
->gd_trap_nesting_level
;
689 ++gd
->gd_trap_nesting_level
;
692 if (trap_mpsafe
== 0)
693 MAKEMPSAFE(have_mplock
);
696 --gd
->gd_trap_nesting_level
;
698 type
= frame
->tf_trapno
;
699 code
= frame
->tf_err
;
707 case T_PAGEFLT
: /* page fault */
708 MAKEMPSAFE(have_mplock
);
709 trap_pfault(frame
, FALSE
, eva
);
715 * The kernel may be using npx for copying or other
718 panic("kernel NPX should not happen");
724 case T_PROTFLT
: /* general protection fault */
725 case T_SEGNPFLT
: /* segment not present fault */
727 * Invalid segment selectors and out of bounds
728 * %eip's and %esp's can be set up in user mode.
729 * This causes a fault in kernel mode when the
730 * kernel tries to return to user mode. We want
731 * to get this fault so that we can fix the
732 * problem here and not have to check all the
733 * selectors and pointers when the user changes
736 if (mycpu
->gd_intr_nesting_level
== 0) {
737 if (td
->td_pcb
->pcb_onfault
) {
739 (register_t
)td
->td_pcb
->pcb_onfault
;
747 * PSL_NT can be set in user mode and isn't cleared
748 * automatically when the kernel is entered. This
749 * causes a TSS fault when the kernel attempts to
750 * `iret' because the TSS link is uninitialized. We
751 * want to get this fault so that we can fix the
752 * problem here and not every time the kernel is
755 if (frame
->tf_eflags
& PSL_NT
) {
756 frame
->tf_eflags
&= ~PSL_NT
;
761 case T_TRCTRAP
: /* trace trap */
763 if (frame
->tf_eip
== (int)IDTVEC(syscall
)) {
765 * We've just entered system mode via the
766 * syscall lcall. Continue single stepping
767 * silently until the syscall handler has
772 if (frame
->tf_eip
== (int)IDTVEC(syscall
) + 1) {
774 * The syscall handler has now saved the
775 * flags. Stop single stepping it.
777 frame
->tf_eflags
&= ~PSL_T
;
783 * Ignore debug register trace traps due to
784 * accesses in the user's address space, which
785 * can happen under several conditions such as
786 * if a user sets a watchpoint on a buffer and
787 * then passes that buffer to a system call.
788 * We still want to get TRCTRAPS for addresses
789 * in kernel space because that is useful when
790 * debugging the kernel.
792 if (user_dbreg_trap()) {
794 * Reset breakpoint bits because the
797 load_dr6(rdr6() & 0xfffffff0);
802 * Fall through (TRCTRAP kernel mode, kernel address)
806 * If DDB is enabled, let it handle the debugger trap.
807 * Otherwise, debugger traps "can't happen".
810 MAKEMPSAFE(have_mplock
);
811 if (kdb_trap (type
, 0, frame
))
816 MAKEMPSAFE(have_mplock
);
817 trap_fatal(frame
, FALSE
, eva
);
820 MAKEMPSAFE(have_mplock
);
821 trap_fatal(frame
, FALSE
, eva
);
825 * Ignore this trap generated from a spurious SIGTRAP.
827 * single stepping in / syscalls leads to spurious / SIGTRAP
830 * Haiku (c) 2007 Simon 'corecode' Schubert
836 * Translate fault for emulators (e.g. Linux)
838 if (*p
->p_sysent
->sv_transtrap
)
839 i
= (*p
->p_sysent
->sv_transtrap
)(i
, type
);
841 MAKEMPSAFE(have_mplock
);
842 trapsignal(lp
, i
, ucode
);
845 if (type
<= MAX_TRAP_MSG
) {
846 uprintf("fatal process exception: %s",
848 if ((type
== T_PAGEFLT
) || (type
== T_PROTFLT
))
849 uprintf(", fault VA = 0x%lx", (u_long
)eva
);
861 KASSERT(crit_count
== (td
->td_pri
& ~TDPRI_MASK
),
862 ("syscall: critical section count mismatch! %d/%d",
863 crit_count
/ TDPRI_CRIT
, td
->td_pri
/ TDPRI_CRIT
));
868 trap_pfault(struct trapframe
*frame
, int usermode
, vm_offset_t eva
)
871 struct vmspace
*vm
= NULL
;
875 thread_t td
= curthread
;
876 struct lwp
*lp
= td
->td_lwp
;
878 va
= trunc_page(eva
);
879 if (usermode
== FALSE
) {
881 * This is a fault on kernel virtual memory.
886 * This is a fault on non-kernel virtual memory.
887 * vm is initialized above to NULL. If curproc is NULL
888 * or curproc->p_vmspace is NULL the fault is fatal.
891 vm
= lp
->lwp_vmspace
;
899 if (frame
->tf_xflags
& PGEX_W
)
900 ftype
= VM_PROT_READ
| VM_PROT_WRITE
;
902 ftype
= VM_PROT_READ
;
904 if (map
!= &kernel_map
) {
906 * Keep swapout from messing with us during this
912 * Grow the stack if necessary
914 /* grow_stack returns false only if va falls into
915 * a growable stack region and the stack growth
916 * fails. It returns true if va was not within
917 * a growable stack region, or if the stack
920 if (!grow_stack (lp
->lwp_proc
, va
)) {
926 /* Fault in the user page: */
927 rv
= vm_fault(map
, va
, ftype
,
928 (ftype
& VM_PROT_WRITE
) ? VM_FAULT_DIRTY
934 * Don't have to worry about process locking or stacks in the kernel.
936 rv
= vm_fault(map
, va
, ftype
, VM_FAULT_NORMAL
);
939 if (rv
== KERN_SUCCESS
)
943 if (td
->td_gd
->gd_intr_nesting_level
== 0 &&
944 td
->td_pcb
->pcb_onfault
) {
945 frame
->tf_eip
= (register_t
)td
->td_pcb
->pcb_onfault
;
948 trap_fatal(frame
, usermode
, eva
);
951 return((rv
== KERN_PROTECTION_FAILURE
) ? SIGBUS
: SIGSEGV
);
955 trap_fatal(struct trapframe
*frame
, int usermode
, vm_offset_t eva
)
957 int code
, type
, ss
, esp
;
959 code
= frame
->tf_xflags
;
960 type
= frame
->tf_trapno
;
962 if (type
<= MAX_TRAP_MSG
) {
963 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
964 type
, trap_msg
[type
],
965 (usermode
? "user" : "kernel"));
968 /* two separate prints in case of a trap on an unmapped page */
969 kprintf("mp_lock = %08x; ", mp_lock
);
970 kprintf("cpuid = %d\n", mycpu
->gd_cpuid
);
972 if (type
== T_PAGEFLT
) {
973 kprintf("fault virtual address = 0x%x\n", eva
);
974 kprintf("fault code = %s %s, %s\n",
975 usermode
? "user" : "supervisor",
976 code
& PGEX_W
? "write" : "read",
977 code
& PGEX_P
? "protection violation" : "page not present");
979 kprintf("instruction pointer = 0x%x:0x%x\n",
980 frame
->tf_cs
& 0xffff, frame
->tf_eip
);
982 ss
= frame
->tf_ss
& 0xffff;
985 ss
= GSEL(GDATA_SEL
, SEL_KPL
);
986 esp
= (int)&frame
->tf_esp
;
988 kprintf("stack pointer = 0x%x:0x%x\n", ss
, esp
);
989 kprintf("frame pointer = 0x%x:0x%x\n", ss
, frame
->tf_ebp
);
990 kprintf("processor eflags = ");
991 if (frame
->tf_eflags
& PSL_T
)
992 kprintf("trace trap, ");
993 if (frame
->tf_eflags
& PSL_I
)
994 kprintf("interrupt enabled, ");
995 if (frame
->tf_eflags
& PSL_NT
)
996 kprintf("nested task, ");
997 if (frame
->tf_eflags
& PSL_RF
)
1000 if (frame
->tf_eflags
& PSL_VM
)
1003 kprintf("IOPL = %d\n", (frame
->tf_eflags
& PSL_IOPL
) >> 12);
1004 kprintf("current process = ");
1006 kprintf("%lu (%s)\n",
1007 (u_long
)curproc
->p_pid
, curproc
->p_comm
?
1008 curproc
->p_comm
: "");
1012 kprintf("current thread = pri %d ", curthread
->td_pri
);
1013 if (curthread
->td_pri
>= TDPRI_CRIT
)
1019 * we probably SHOULD have stopped the other CPUs before now!
1020 * another CPU COULD have been touching cpl at this moment...
1022 kprintf(" <- SMP: XXX");
1031 if ((debugger_on_panic
|| db_active
) && kdb_trap(type
, code
, frame
))
1034 kprintf("trap number = %d\n", type
);
1035 if (type
<= MAX_TRAP_MSG
)
1036 panic("%s", trap_msg
[type
]);
1038 panic("unknown/reserved trap");
1042 * Double fault handler. Called when a fault occurs while writing
1043 * a frame for a trap/exception onto the stack. This usually occurs
1044 * when the stack overflows (such is the case with infinite recursion,
1047 * XXX Note that the current PTD gets replaced by IdlePTD when the
1048 * task switch occurs. This means that the stack that was active at
1049 * the time of the double fault is not available at <kstack> unless
1050 * the machine was idle when the double fault occurred. The downside
1051 * of this is that "trace <ebp>" in ddb won't work.
1054 dblfault_handler(void)
1056 struct mdglobaldata
*gd
= mdcpu
;
1058 kprintf("\nFatal double fault:\n");
1059 kprintf("eip = 0x%x\n", gd
->gd_common_tss
.tss_eip
);
1060 kprintf("esp = 0x%x\n", gd
->gd_common_tss
.tss_esp
);
1061 kprintf("ebp = 0x%x\n", gd
->gd_common_tss
.tss_ebp
);
1063 /* two separate prints in case of a trap on an unmapped page */
1064 kprintf("mp_lock = %08x; ", mp_lock
);
1065 kprintf("cpuid = %d\n", mycpu
->gd_cpuid
);
1067 panic("double fault");
1071 * Compensate for 386 brain damage (missing URKR).
1072 * This is a little simpler than the pagefault handler in trap() because
1073 * it the page tables have already been faulted in and high addresses
1074 * are thrown out early for other reasons.
1077 trapwrite(unsigned addr
)
1084 va
= trunc_page((vm_offset_t
)addr
);
1086 * XXX - MAX is END. Changed > to >= for temp. fix.
1088 if (va
>= VM_MAX_USER_ADDRESS
)
1091 lp
= curthread
->td_lwp
;
1092 vm
= lp
->lwp_vmspace
;
1094 PHOLD(lp
->lwp_proc
);
1096 if (!grow_stack (lp
->lwp_proc
, va
)) {
1097 PRELE(lp
->lwp_proc
);
1102 * fault the data page
1104 rv
= vm_fault(&vm
->vm_map
, va
, VM_PROT_WRITE
, VM_FAULT_DIRTY
);
1106 PRELE(lp
->lwp_proc
);
1108 if (rv
!= KERN_SUCCESS
)
1115 * syscall2 - MP aware system call request C handler
1117 * A system call is essentially treated as a trap except that the
1118 * MP lock is not held on entry or return. We are responsible for
1119 * obtaining the MP lock if necessary and for handling ASTs
1120 * (e.g. a task switch) prior to return.
1122 * In general, only simple access and manipulation of curproc and
1123 * the current stack is allowed without having to hold MP lock.
1125 * MPSAFE - note that large sections of this routine are run without
1130 syscall2(struct trapframe
*frame
)
1132 struct thread
*td
= curthread
;
1133 struct proc
*p
= td
->td_proc
;
1134 struct lwp
*lp
= td
->td_lwp
;
1136 struct sysent
*callp
;
1137 register_t orig_tf_eflags
;
1142 int crit_count
= td
->td_pri
& ~TDPRI_MASK
;
1145 int have_mplock
= 0;
1148 union sysunion args
;
1150 KTR_LOG(kernentry_syscall
, lp
->lwp_proc
->p_pid
, lp
->lwp_tid
,
1154 KASSERT(td
->td_mpcount
== 0, ("badmpcount syscall2 from %p", (void *)frame
->tf_eip
));
1155 if (syscall_mpsafe
== 0)
1156 MAKEMPSAFE(have_mplock
);
1158 userenter(td
); /* lazy raise our priority */
1163 sticks
= (int)td
->td_sticks
;
1164 orig_tf_eflags
= frame
->tf_eflags
;
1167 * Virtual kernel intercept - if a VM context managed by a virtual
1168 * kernel issues a system call the virtual kernel handles it, not us.
1169 * Restore the virtual kernel context and return from its system
1170 * call. The current frame is copied out to the virtual kernel.
1172 if (lp
->lwp_vkernel
&& lp
->lwp_vkernel
->ve
) {
1173 error
= vkernel_trap(lp
, frame
);
1174 frame
->tf_eax
= error
;
1176 frame
->tf_eflags
|= PSL_C
;
1177 error
= EJUSTRETURN
;
1182 * Get the system call parameters and account for time
1184 lp
->lwp_md
.md_regs
= frame
;
1185 params
= (caddr_t
)frame
->tf_esp
+ sizeof(int);
1186 code
= frame
->tf_eax
;
1188 if (p
->p_sysent
->sv_prepsyscall
) {
1189 (*p
->p_sysent
->sv_prepsyscall
)(
1190 frame
, (int *)(&args
.nosys
.sysmsg
+ 1),
1194 * Need to check if this is a 32 bit or 64 bit syscall.
1195 * fuword is MP aware.
1197 if (code
== SYS_syscall
) {
1199 * Code is first argument, followed by actual args.
1201 code
= fuword(params
);
1202 params
+= sizeof(int);
1203 } else if (code
== SYS___syscall
) {
1205 * Like syscall, but code is a quad, so as to maintain
1206 * quad alignment for the rest of the arguments.
1208 code
= fuword(params
);
1209 params
+= sizeof(quad_t
);
1213 code
&= p
->p_sysent
->sv_mask
;
1214 if (code
>= p
->p_sysent
->sv_size
)
1215 callp
= &p
->p_sysent
->sv_table
[0];
1217 callp
= &p
->p_sysent
->sv_table
[code
];
1219 narg
= callp
->sy_narg
& SYF_ARGMASK
;
1222 * copyin is MP aware, but the tracing code is not
1224 if (narg
&& params
) {
1225 error
= copyin(params
, (caddr_t
)(&args
.nosys
.sysmsg
+ 1),
1226 narg
* sizeof(register_t
));
1229 if (KTRPOINT(td
, KTR_SYSCALL
)) {
1230 MAKEMPSAFE(have_mplock
);
1232 ktrsyscall(lp
, code
, narg
,
1233 (void *)(&args
.nosys
.sysmsg
+ 1));
1241 if (KTRPOINT(td
, KTR_SYSCALL
)) {
1242 MAKEMPSAFE(have_mplock
);
1243 ktrsyscall(lp
, code
, narg
, (void *)(&args
.nosys
.sysmsg
+ 1));
1248 * For traditional syscall code edx is left untouched when 32 bit
1249 * results are returned. Since edx is loaded from fds[1] when the
1250 * system call returns we pre-set it here.
1252 args
.sysmsg_fds
[0] = 0;
1253 args
.sysmsg_fds
[1] = frame
->tf_edx
;
1256 * The syscall might manipulate the trap frame. If it does it
1257 * will probably return EJUSTRETURN.
1259 args
.sysmsg_frame
= frame
;
1261 STOPEVENT(p
, S_SCE
, narg
); /* MP aware */
1265 * Try to run the syscall without the MP lock if the syscall
1266 * is MP safe. We have to obtain the MP lock no matter what if
1269 if ((callp
->sy_narg
& SYF_MPSAFE
) == 0)
1270 MAKEMPSAFE(have_mplock
);
1273 error
= (*callp
->sy_call
)(&args
);
1276 kprintf("system call %d returned %d\n", code
, error
);
1281 * MP SAFE (we may or may not have the MP lock at this point)
1286 * Reinitialize proc pointer `p' as it may be different
1287 * if this is a child returning from fork syscall.
1290 lp
= curthread
->td_lwp
;
1291 frame
->tf_eax
= args
.sysmsg_fds
[0];
1292 frame
->tf_edx
= args
.sysmsg_fds
[1];
1293 frame
->tf_eflags
&= ~PSL_C
;
1297 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1298 * int 0x80 is 2 bytes. We saved this in tf_err.
1300 frame
->tf_eip
-= frame
->tf_err
;
1305 panic("Unexpected EASYNC return value (for now)");
1308 if (p
->p_sysent
->sv_errsize
) {
1309 if (error
>= p
->p_sysent
->sv_errsize
)
1310 error
= -1; /* XXX */
1312 error
= p
->p_sysent
->sv_errtbl
[error
];
1314 frame
->tf_eax
= error
;
1315 frame
->tf_eflags
|= PSL_C
;
1320 * Traced syscall. trapsignal() is not MP aware.
1322 if ((orig_tf_eflags
& PSL_T
) /*&& !(orig_tf_eflags & PSL_VM)*/) {
1323 MAKEMPSAFE(have_mplock
);
1324 frame
->tf_eflags
&= ~PSL_T
;
1325 trapsignal(lp
, SIGTRAP
, 0);
1329 * Handle reschedule and other end-of-syscall issues
1331 userret(lp
, frame
, sticks
);
1334 if (KTRPOINT(td
, KTR_SYSRET
)) {
1335 MAKEMPSAFE(have_mplock
);
1336 ktrsysret(lp
, code
, error
, args
.sysmsg_result
);
1341 * This works because errno is findable through the
1342 * register set. If we ever support an emulation where this
1343 * is not the case, this code will need to be revisited.
1345 STOPEVENT(p
, S_SCX
, code
);
1350 * Release the MP lock if we had to get it
1352 KASSERT(td
->td_mpcount
== have_mplock
,
1353 ("badmpcount syscall2/end from %p", (void *)frame
->tf_eip
));
1357 KTR_LOG(kernentry_syscall_ret
, lp
->lwp_proc
->p_pid
, lp
->lwp_tid
, error
);
1359 KASSERT(crit_count
== (td
->td_pri
& ~TDPRI_MASK
),
1360 ("syscall: critical section count mismatch! %d/%d",
1361 crit_count
/ TDPRI_CRIT
, td
->td_pri
/ TDPRI_CRIT
));
1366 fork_return(struct lwp
*lp
, struct trapframe
*frame
)
1368 frame
->tf_eax
= 0; /* Child returns zero */
1369 frame
->tf_eflags
&= ~PSL_C
; /* success */
1372 generic_lwp_return(lp
, frame
);
1373 KTR_LOG(kernentry_fork_ret
, lp
->lwp_proc
->p_pid
, lp
->lwp_tid
);
1377 * Simplified back end of syscall(), used when returning from fork()
1378 * or lwp_create() directly into user mode. MP lock is held on entry and
1379 * should be released on return. This code will return back into the fork
1380 * trampoline code which then runs doreti.
1383 generic_lwp_return(struct lwp
*lp
, struct trapframe
*frame
)
1385 struct proc
*p
= lp
->lwp_proc
;
1388 * Newly forked processes are given a kernel priority. We have to
1389 * adjust the priority to a normal user priority and fake entry
1390 * into the kernel (call userenter()) to install a passive release
1391 * function just in case userret() decides to stop the process. This
1392 * can occur when ^Z races a fork. If we do not install the passive
1393 * release function the current process designation will not be
1394 * released when the thread goes to sleep.
1396 lwkt_setpri_self(TDPRI_USER_NORM
);
1397 userenter(lp
->lwp_thread
);
1398 userret(lp
, frame
, 0);
1400 if (KTRPOINT(lp
->lwp_thread
, KTR_SYSRET
))
1401 ktrsysret(lp
, SYS_fork
, 0, 0);
1403 p
->p_flag
|= P_PASSIVE_ACQ
;
1405 p
->p_flag
&= ~P_PASSIVE_ACQ
;
1407 KKASSERT(lp
->lwp_thread
->td_mpcount
== 1);
1413 * doreti has turned into this. The frame is directly on the stack. We
1414 * pull everything else we need (fpu and tls context) from the current
1417 * Note on fpu interactions: In a virtual kernel, the fpu context for
1418 * an emulated user mode process is not shared with the virtual kernel's
1419 * fpu context, so we only have to 'stack' fpu contexts within the virtual
1420 * kernel itself, and not even then since the signal() contexts that we care
1421 * about save and restore the FPU state (I think anyhow).
1423 * vmspace_ctl() returns an error only if it had problems instaling the
1424 * context we supplied or problems copying data to/from our VM space.
1427 go_user(struct intrframe
*frame
)
1429 struct trapframe
*tf
= (void *)&frame
->if_gs
;
1433 * Interrupts may be disabled on entry, make sure all signals
1434 * can be received before beginning our loop.
1439 * Switch to the current simulated user process, then call
1440 * user_trap() when we break out of it (usually due to a signal).
1444 * Tell the real kernel whether it is ok to use the FP
1447 if (mdcpu
->gd_npxthread
== curthread
) {
1448 tf
->tf_xflags
&= ~PGEX_FPFAULT
;
1450 tf
->tf_xflags
|= PGEX_FPFAULT
;
1454 * Run emulated user process context. This call interlocks
1455 * with new mailbox signals.
1457 * Set PGEX_U unconditionally, indicating a user frame (the
1458 * bit is normally set only by T_PAGEFLT).
1460 r
= vmspace_ctl(&curproc
->p_vmspace
->vm_pmap
, VMSPACE_CTL_RUN
,
1461 tf
, &curthread
->td_savevext
);
1462 frame
->if_xflags
|= PGEX_U
;
1464 kprintf("GO USER %d trap %d EVA %08x EIP %08x ESP %08x XFLAGS %02x/%02x\n",
1465 r
, tf
->tf_trapno
, tf
->tf_err
, tf
->tf_eip
, tf
->tf_esp
,
1466 tf
->tf_xflags
, frame
->if_xflags
);
1470 panic("vmspace_ctl failed");
1472 if (tf
->tf_trapno
) {
1474 } else if (mycpu
->gd_reqflags
& RQF_AST_MASK
) {
1475 tf
->tf_trapno
= T_ASTFLT
;
1484 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1485 * fault (which is then passed back to the virtual kernel) if an attempt is
1486 * made to use the FP unit.
1488 * XXX this is a fairly big hack.
1491 set_vkernel_fp(struct trapframe
*frame
)
1493 struct thread
*td
= curthread
;
1495 if (frame
->tf_xflags
& PGEX_FPFAULT
) {
1496 td
->td_pcb
->pcb_flags
|= FP_VIRTFP
;
1497 if (mdcpu
->gd_npxthread
== td
)
1500 td
->td_pcb
->pcb_flags
&= ~FP_VIRTFP
;