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 extern int trapwrite (unsigned addr
);
118 static int trap_pfault (struct trapframe
*, int, vm_offset_t
);
119 static void trap_fatal (struct trapframe
*, int, vm_offset_t
);
120 void dblfault_handler (void);
123 extern inthand_t
IDTVEC(syscall
);
126 #define MAX_TRAP_MSG 28
127 static char *trap_msg
[] = {
129 "privileged instruction fault", /* 1 T_PRIVINFLT */
131 "breakpoint instruction fault", /* 3 T_BPTFLT */
134 "arithmetic trap", /* 6 T_ARITHTRAP */
135 "system forced exception", /* 7 T_ASTFLT */
137 "general protection fault", /* 9 T_PROTFLT */
138 "trace trap", /* 10 T_TRCTRAP */
140 "page fault", /* 12 T_PAGEFLT */
142 "alignment fault", /* 14 T_ALIGNFLT */
146 "integer divide fault", /* 18 T_DIVIDE */
147 "non-maskable interrupt trap", /* 19 T_NMI */
148 "overflow trap", /* 20 T_OFLOW */
149 "FPU bounds check fault", /* 21 T_BOUND */
150 "FPU device not available", /* 22 T_DNA */
151 "double fault", /* 23 T_DOUBLEFLT */
152 "FPU operand fetch fault", /* 24 T_FPOPFLT */
153 "invalid TSS fault", /* 25 T_TSSFLT */
154 "segment not present fault", /* 26 T_SEGNPFLT */
155 "stack fault", /* 27 T_STKFLT */
156 "machine check trap", /* 28 T_MCHK */
160 static int ddb_on_nmi
= 1;
161 SYSCTL_INT(_machdep
, OID_AUTO
, ddb_on_nmi
, CTLFLAG_RW
,
162 &ddb_on_nmi
, 0, "Go to DDB on NMI");
164 static int panic_on_nmi
= 1;
165 SYSCTL_INT(_machdep
, OID_AUTO
, panic_on_nmi
, CTLFLAG_RW
,
166 &panic_on_nmi
, 0, "Panic on NMI");
167 static int fast_release
;
168 SYSCTL_INT(_machdep
, OID_AUTO
, fast_release
, CTLFLAG_RW
,
169 &fast_release
, 0, "Passive Release was optimal");
170 static int slow_release
;
171 SYSCTL_INT(_machdep
, OID_AUTO
, slow_release
, CTLFLAG_RW
,
172 &slow_release
, 0, "Passive Release was nonoptimal");
174 static int syscall_mpsafe
= 1;
175 SYSCTL_INT(_kern
, OID_AUTO
, syscall_mpsafe
, CTLFLAG_RW
,
176 &syscall_mpsafe
, 0, "Allow MPSAFE marked syscalls to run without BGL");
177 TUNABLE_INT("kern.syscall_mpsafe", &syscall_mpsafe
);
178 static int trap_mpsafe
= 1;
179 SYSCTL_INT(_kern
, OID_AUTO
, trap_mpsafe
, CTLFLAG_RW
,
180 &trap_mpsafe
, 0, "Allow traps to mostly run without the BGL");
181 TUNABLE_INT("kern.trap_mpsafe", &trap_mpsafe
);
184 MALLOC_DEFINE(M_SYSMSG
, "sysmsg", "sysmsg structure");
185 extern int max_sysmsg
;
188 * Passively intercepts the thread switch function to increase
189 * the thread priority from a user priority to a kernel priority, reducing
190 * syscall and trap overhead for the case where no switch occurs.
192 * Synchronizes td_ucred with p_ucred. This is used by system calls,
193 * signal handling, faults, AST traps, and anything else that enters the
194 * kernel from userland and provides the kernel with a stable read-only
195 * copy of the process ucred.
198 userenter(struct thread
*curtd
, struct proc
*curp
)
203 curtd
->td_release
= lwkt_passive_release
;
205 if (curtd
->td_ucred
!= curp
->p_ucred
) {
206 ncred
= crhold(curp
->p_ucred
);
207 ocred
= curtd
->td_ucred
;
208 curtd
->td_ucred
= ncred
;
215 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
216 * must be completed before we can return to or try to return to userland.
218 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
219 * arithmatic on the delta calculation so the absolute tick values are
220 * truncated to an integer.
223 userret(struct lwp
*lp
, struct trapframe
*frame
, int sticks
)
225 struct proc
*p
= lp
->lwp_proc
;
229 * Charge system time if profiling. Note: times are in microseconds.
230 * This may do a copyout and block, so do it first even though it
231 * means some system time will be charged as user time.
233 if (p
->p_flag
& P_PROFIL
) {
234 addupc_task(p
, frame
->tf_eip
,
235 (u_int
)((int)lp
->lwp_thread
->td_sticks
- sticks
));
240 * If the jungle wants us dead, so be it.
242 if (lp
->lwp_flag
& LWP_WEXIT
) {
245 rel_mplock(); /* NOT REACHED */
249 * Block here if we are in a stopped state.
251 if (p
->p_stat
== SSTOP
) {
259 * Post any pending upcalls
261 if (p
->p_flag
& P_UPCALLPEND
) {
263 p
->p_flag
&= ~P_UPCALLPEND
;
270 * Post any pending signals
272 if ((sig
= CURSIG_TRACE(lp
)) != 0) {
280 * block here if we are swapped out, but still process signals
281 * (such as SIGKILL). proc0 (the swapin scheduler) is already
282 * aware of our situation, we do not have to wake it up.
284 if (p
->p_flag
& P_SWAPPEDOUT
) {
286 p
->p_flag
|= P_SWAPWAIT
;
288 if (p
->p_flag
& P_SWAPWAIT
)
289 tsleep(p
, PCATCH
, "SWOUT", 0);
290 p
->p_flag
&= ~P_SWAPWAIT
;
296 * Make sure postsig() handled request to restore old signal mask after
297 * running signal handler.
299 KKASSERT((lp
->lwp_flag
& LWP_OLDMASK
) == 0);
303 * Cleanup from userenter and any passive release that might have occured.
304 * We must reclaim the current-process designation before we can return
305 * to usermode. We also handle both LWKT and USER reschedule requests.
308 userexit(struct lwp
*lp
)
310 struct thread
*td
= lp
->lwp_thread
;
311 /* globaldata_t gd = td->td_gd; */
314 * Handle stop requests at kernel priority. Any requests queued
315 * after this loop will generate another AST.
317 while (lp
->lwp_proc
->p_stat
== SSTOP
) {
324 * Reduce our priority in preparation for a return to userland. If
325 * our passive release function was still in place, our priority was
326 * never raised and does not need to be reduced.
328 lwkt_passive_recover(td
);
331 * Become the current user scheduled process if we aren't already,
332 * and deal with reschedule requests and other factors.
334 lp
->lwp_proc
->p_usched
->acquire_curproc(lp
);
335 /* WARNING: we may have migrated cpu's */
336 /* gd = td->td_gd; */
339 #if !defined(KTR_KERNENTRY)
340 #define KTR_KERNENTRY KTR_ALL
342 KTR_INFO_MASTER(kernentry
);
343 KTR_INFO(KTR_KERNENTRY
, kernentry
, trap
, 0, "pid=%d, tid=%d, trapno=%d, eva=%p",
344 sizeof(int) + sizeof(int) + sizeof(int) + sizeof(vm_offset_t
));
345 KTR_INFO(KTR_KERNENTRY
, kernentry
, trap_ret
, 0, "pid=%d, tid=%d",
346 sizeof(int) + sizeof(int));
347 KTR_INFO(KTR_KERNENTRY
, kernentry
, syscall
, 0, "pid=%d, tid=%d, call=%d",
348 sizeof(int) + sizeof(int) + sizeof(int));
349 KTR_INFO(KTR_KERNENTRY
, kernentry
, syscall_ret
, 0, "pid=%d, tid=%d, err=%d",
350 sizeof(int) + sizeof(int) + sizeof(int));
351 KTR_INFO(KTR_KERNENTRY
, kernentry
, fork_ret
, 0, "pid=%d, tid=%d",
352 sizeof(int) + sizeof(int));
355 * Exception, fault, and trap interface to the kernel.
356 * This common code is called from assembly language IDT gate entry
357 * routines that prepare a suitable stack frame, and restore this
358 * frame after the exception has been processed.
360 * This function is also called from doreti in an interlock to handle ASTs.
361 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
363 * NOTE! We have to retrieve the fault address prior to obtaining the
364 * MP lock because get_mplock() may switch out. YYY cr2 really ought
365 * to be retrieved by the assembly code, not here.
367 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
368 * if an attempt is made to switch from a fast interrupt or IPI. This is
369 * necessary to properly take fatal kernel traps on SMP machines if
370 * get_mplock() has to block.
374 user_trap(struct trapframe
*frame
)
376 struct globaldata
*gd
= mycpu
;
377 struct thread
*td
= gd
->gd_curthread
;
378 struct lwp
*lp
= td
->td_lwp
;
381 int i
= 0, ucode
= 0, type
, code
;
386 int crit_count
= td
->td_pri
& ~TDPRI_MASK
;
393 * This is a bad kludge to avoid changing the various trapframe
394 * structures. Because we are enabled as a virtual kernel,
395 * the original tf_err field will be passed to us shifted 16
396 * over in the tf_trapno field for T_PAGEFLT.
398 if (frame
->tf_trapno
== T_PAGEFLT
)
403 kprintf("USER_TRAP AT %08x xflags %d trapno %d eva %08x\n",
404 frame
->tf_eip
, frame
->tf_xflags
, frame
->tf_trapno
, eva
);
408 * Everything coming from user mode runs through user_trap,
409 * including system calls.
411 if (frame
->tf_trapno
== T_SYSCALL80
) {
416 KTR_LOG(kernentry_trap
, lp
->lwp_proc
->p_pid
, lp
->lwp_tid
,
417 frame
->tf_trapno
, eva
);
421 eva
= (frame
->tf_trapno
== T_PAGEFLT
? rcr2() : 0);
422 ++gd
->gd_trap_nesting_level
;
423 MAKEMPSAFE(have_mplock
);
424 trap_fatal(frame
, TRUE
, eva
);
425 --gd
->gd_trap_nesting_level
;
430 ++gd
->gd_trap_nesting_level
;
432 if (trap_mpsafe
== 0)
433 MAKEMPSAFE(have_mplock
);
436 --gd
->gd_trap_nesting_level
;
438 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
441 type
= frame
->tf_trapno
;
442 code
= frame
->tf_err
;
446 sticks
= (int)td
->td_sticks
;
447 lp
->lwp_md
.md_regs
= frame
;
450 case T_PRIVINFLT
: /* privileged instruction fault */
455 case T_BPTFLT
: /* bpt instruction fault */
456 case T_TRCTRAP
: /* trace trap */
457 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
);
493 ucode
= (type
== T_PROTFLT
) ? BUS_OBJERR
: BUS_ADRERR
;
495 case T_SEGNPFLT
: /* segment not present fault */
499 case T_TSSFLT
: /* invalid TSS fault */
500 case T_DOUBLEFLT
: /* double fault */
505 ucode
= code
+ BUS_SEGM_FAULT
; /* XXX: ???*/
511 case T_PAGEFLT
: /* page fault */
512 MAKEMPSAFE(have_mplock
);
513 i
= trap_pfault(frame
, TRUE
, eva
);
516 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
532 case T_DIVIDE
: /* integer divide fault */
539 MAKEMPSAFE(have_mplock
);
540 /* machine/parity/power fail/"kitchen sink" faults */
541 if (isa_nmi(code
) == 0) {
544 * NMI can be hooked up to a pushbutton
548 kprintf ("NMI ... going to debugger\n");
549 kdb_trap (type
, 0, frame
);
553 } else if (panic_on_nmi
)
554 panic("NMI indicates hardware failure");
556 #endif /* NISA > 0 */
558 case T_OFLOW
: /* integer overflow fault */
563 case T_BOUND
: /* bounds check fault */
570 * Virtual kernel intercept - pass the DNA exception
571 * to the (emulated) virtual kernel if it asked to handle
572 * it. This occurs when the virtual kernel is holding
573 * onto the FP context for a different emulated
574 * process then the one currently running.
576 * We must still call npxdna() since we may have
577 * saved FP state that the (emulated) virtual kernel
578 * needs to hand over to a different emulated process.
580 if (lp
->lwp_vkernel
&& lp
->lwp_vkernel
->ve
&&
581 (td
->td_pcb
->pcb_flags
& FP_VIRTFP
)
588 * The kernel may have switched out the FP unit's
589 * state, causing the user process to take a fault
590 * when it tries to use the FP unit. Restore the
596 if (!pmath_emulate
) {
598 ucode
= FPE_FPU_NP_TRAP
;
601 i
= (*pmath_emulate
)(frame
);
603 if (!(frame
->tf_eflags
& PSL_T
))
605 frame
->tf_eflags
&= ~PSL_T
;
608 /* else ucode = emulator_only_knows() XXX */
611 case T_FPOPFLT
: /* FPU operand fetch fault */
616 case T_XMMFLT
: /* SIMD floating-point exception */
623 * Virtual kernel intercept - if the fault is directly related to a
624 * VM context managed by a virtual kernel then let the virtual kernel
627 if (lp
->lwp_vkernel
&& lp
->lwp_vkernel
->ve
) {
628 vkernel_trap(lp
, frame
);
633 * Translate fault for emulators (e.g. Linux)
635 if (*p
->p_sysent
->sv_transtrap
)
636 i
= (*p
->p_sysent
->sv_transtrap
)(i
, type
);
638 MAKEMPSAFE(have_mplock
);
639 trapsignal(lp
, i
, ucode
);
642 if (type
<= MAX_TRAP_MSG
) {
643 uprintf("fatal process exception: %s",
645 if ((type
== T_PAGEFLT
) || (type
== T_PROTFLT
))
646 uprintf(", fault VA = 0x%lx", (u_long
)eva
);
653 KASSERT(td
->td_mpcount
== have_mplock
, ("badmpcount trap/end from %p", (void *)frame
->tf_eip
));
655 userret(lp
, frame
, sticks
);
662 KTR_LOG(kernentry_trap_ret
, lp
->lwp_proc
->p_pid
, lp
->lwp_tid
);
664 KASSERT(crit_count
== (td
->td_pri
& ~TDPRI_MASK
),
665 ("syscall: critical section count mismatch! %d/%d",
666 crit_count
/ TDPRI_CRIT
, td
->td_pri
/ TDPRI_CRIT
));
671 kern_trap(struct trapframe
*frame
)
673 struct globaldata
*gd
= mycpu
;
674 struct thread
*td
= gd
->gd_curthread
;
677 int i
= 0, ucode
= 0, type
, code
;
682 int crit_count
= td
->td_pri
& ~TDPRI_MASK
;
689 if (frame
->tf_trapno
== T_PAGEFLT
)
696 ++gd
->gd_trap_nesting_level
;
697 MAKEMPSAFE(have_mplock
);
698 trap_fatal(frame
, FALSE
, eva
);
699 --gd
->gd_trap_nesting_level
;
704 ++gd
->gd_trap_nesting_level
;
707 if (trap_mpsafe
== 0)
708 MAKEMPSAFE(have_mplock
);
711 --gd
->gd_trap_nesting_level
;
713 type
= frame
->tf_trapno
;
714 code
= frame
->tf_err
;
722 case T_PAGEFLT
: /* page fault */
723 MAKEMPSAFE(have_mplock
);
724 trap_pfault(frame
, FALSE
, eva
);
730 * The kernel may be using npx for copying or other
733 panic("kernel NPX should not happen");
739 case T_PROTFLT
: /* general protection fault */
740 case T_SEGNPFLT
: /* segment not present fault */
742 * Invalid segment selectors and out of bounds
743 * %eip's and %esp's can be set up in user mode.
744 * This causes a fault in kernel mode when the
745 * kernel tries to return to user mode. We want
746 * to get this fault so that we can fix the
747 * problem here and not have to check all the
748 * selectors and pointers when the user changes
751 if (mycpu
->gd_intr_nesting_level
== 0) {
752 if (td
->td_pcb
->pcb_onfault
) {
754 (register_t
)td
->td_pcb
->pcb_onfault
;
762 * PSL_NT can be set in user mode and isn't cleared
763 * automatically when the kernel is entered. This
764 * causes a TSS fault when the kernel attempts to
765 * `iret' because the TSS link is uninitialized. We
766 * want to get this fault so that we can fix the
767 * problem here and not every time the kernel is
770 if (frame
->tf_eflags
& PSL_NT
) {
771 frame
->tf_eflags
&= ~PSL_NT
;
776 case T_TRCTRAP
: /* trace trap */
778 if (frame
->tf_eip
== (int)IDTVEC(syscall
)) {
780 * We've just entered system mode via the
781 * syscall lcall. Continue single stepping
782 * silently until the syscall handler has
787 if (frame
->tf_eip
== (int)IDTVEC(syscall
) + 1) {
789 * The syscall handler has now saved the
790 * flags. Stop single stepping it.
792 frame
->tf_eflags
&= ~PSL_T
;
798 * Ignore debug register trace traps due to
799 * accesses in the user's address space, which
800 * can happen under several conditions such as
801 * if a user sets a watchpoint on a buffer and
802 * then passes that buffer to a system call.
803 * We still want to get TRCTRAPS for addresses
804 * in kernel space because that is useful when
805 * debugging the kernel.
807 if (user_dbreg_trap()) {
809 * Reset breakpoint bits because the
812 load_dr6(rdr6() & 0xfffffff0);
817 * Fall through (TRCTRAP kernel mode, kernel address)
821 * If DDB is enabled, let it handle the debugger trap.
822 * Otherwise, debugger traps "can't happen".
825 MAKEMPSAFE(have_mplock
);
826 if (kdb_trap (type
, 0, frame
))
831 MAKEMPSAFE(have_mplock
);
832 trap_fatal(frame
, FALSE
, eva
);
835 MAKEMPSAFE(have_mplock
);
836 trap_fatal(frame
, FALSE
, eva
);
840 * Ignore this trap generated from a spurious SIGTRAP.
842 * single stepping in / syscalls leads to spurious / SIGTRAP
845 * Haiku (c) 2007 Simon 'corecode' Schubert
851 * Translate fault for emulators (e.g. Linux)
853 if (*p
->p_sysent
->sv_transtrap
)
854 i
= (*p
->p_sysent
->sv_transtrap
)(i
, type
);
856 MAKEMPSAFE(have_mplock
);
857 trapsignal(lp
, i
, ucode
);
860 if (type
<= MAX_TRAP_MSG
) {
861 uprintf("fatal process exception: %s",
863 if ((type
== T_PAGEFLT
) || (type
== T_PROTFLT
))
864 uprintf(", fault VA = 0x%lx", (u_long
)eva
);
876 KASSERT(crit_count
== (td
->td_pri
& ~TDPRI_MASK
),
877 ("syscall: critical section count mismatch! %d/%d",
878 crit_count
/ TDPRI_CRIT
, td
->td_pri
/ TDPRI_CRIT
));
883 trap_pfault(struct trapframe
*frame
, int usermode
, vm_offset_t eva
)
886 struct vmspace
*vm
= NULL
;
891 thread_t td
= curthread
;
892 struct lwp
*lp
= td
->td_lwp
;
894 va
= trunc_page(eva
);
895 if (usermode
== FALSE
) {
897 * This is a fault on kernel virtual memory.
902 * This is a fault on non-kernel virtual memory.
903 * vm is initialized above to NULL. If curproc is NULL
904 * or curproc->p_vmspace is NULL the fault is fatal.
907 vm
= lp
->lwp_vmspace
;
915 if (frame
->tf_xflags
& PGEX_W
)
916 ftype
= VM_PROT_READ
| VM_PROT_WRITE
;
918 ftype
= VM_PROT_READ
;
920 if (map
!= &kernel_map
) {
922 * Keep swapout from messing with us during this
928 * Grow the stack if necessary
930 /* grow_stack returns false only if va falls into
931 * a growable stack region and the stack growth
932 * fails. It returns true if va was not within
933 * a growable stack region, or if the stack
936 if (!grow_stack (lp
->lwp_proc
, va
)) {
947 fault_flags
|= VM_FAULT_BURST
;
948 if (ftype
& VM_PROT_WRITE
)
949 fault_flags
|= VM_FAULT_DIRTY
;
951 fault_flags
|= VM_FAULT_NORMAL
;
952 rv
= vm_fault(map
, va
, ftype
, fault_flags
);
957 * Don't have to worry about process locking or stacks in the kernel.
959 rv
= vm_fault(map
, va
, ftype
, VM_FAULT_NORMAL
);
962 if (rv
== KERN_SUCCESS
)
966 if (td
->td_gd
->gd_intr_nesting_level
== 0 &&
967 td
->td_pcb
->pcb_onfault
) {
968 frame
->tf_eip
= (register_t
)td
->td_pcb
->pcb_onfault
;
971 trap_fatal(frame
, usermode
, eva
);
974 return((rv
== KERN_PROTECTION_FAILURE
) ? SIGBUS
: SIGSEGV
);
978 trap_fatal(struct trapframe
*frame
, int usermode
, vm_offset_t eva
)
980 int code
, type
, ss
, esp
;
982 code
= frame
->tf_xflags
;
983 type
= frame
->tf_trapno
;
985 if (type
<= MAX_TRAP_MSG
) {
986 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
987 type
, trap_msg
[type
],
988 (usermode
? "user" : "kernel"));
991 /* two separate prints in case of a trap on an unmapped page */
992 kprintf("mp_lock = %08x; ", mp_lock
);
993 kprintf("cpuid = %d\n", mycpu
->gd_cpuid
);
995 if (type
== T_PAGEFLT
) {
996 kprintf("fault virtual address = %p\n", (void *)eva
);
997 kprintf("fault code = %s %s, %s\n",
998 usermode
? "user" : "supervisor",
999 code
& PGEX_W
? "write" : "read",
1000 code
& PGEX_P
? "protection violation" : "page not present");
1002 kprintf("instruction pointer = 0x%x:0x%x\n",
1003 frame
->tf_cs
& 0xffff, frame
->tf_eip
);
1005 ss
= frame
->tf_ss
& 0xffff;
1006 esp
= frame
->tf_esp
;
1008 ss
= GSEL(GDATA_SEL
, SEL_KPL
);
1009 esp
= (int)&frame
->tf_esp
;
1011 kprintf("stack pointer = 0x%x:0x%x\n", ss
, esp
);
1012 kprintf("frame pointer = 0x%x:0x%x\n", ss
, frame
->tf_ebp
);
1013 kprintf("processor eflags = ");
1014 if (frame
->tf_eflags
& PSL_T
)
1015 kprintf("trace trap, ");
1016 if (frame
->tf_eflags
& PSL_I
)
1017 kprintf("interrupt enabled, ");
1018 if (frame
->tf_eflags
& PSL_NT
)
1019 kprintf("nested task, ");
1020 if (frame
->tf_eflags
& PSL_RF
)
1021 kprintf("resume, ");
1023 if (frame
->tf_eflags
& PSL_VM
)
1026 kprintf("IOPL = %d\n", (frame
->tf_eflags
& PSL_IOPL
) >> 12);
1027 kprintf("current process = ");
1029 kprintf("%lu (%s)\n",
1030 (u_long
)curproc
->p_pid
, curproc
->p_comm
?
1031 curproc
->p_comm
: "");
1035 kprintf("current thread = pri %d ", curthread
->td_pri
);
1036 if (curthread
->td_pri
>= TDPRI_CRIT
)
1042 * we probably SHOULD have stopped the other CPUs before now!
1043 * another CPU COULD have been touching cpl at this moment...
1045 kprintf(" <- SMP: XXX");
1054 if ((debugger_on_panic
|| db_active
) && kdb_trap(type
, code
, frame
))
1057 kprintf("trap number = %d\n", type
);
1058 if (type
<= MAX_TRAP_MSG
)
1059 panic("%s", trap_msg
[type
]);
1061 panic("unknown/reserved trap");
1065 * Double fault handler. Called when a fault occurs while writing
1066 * a frame for a trap/exception onto the stack. This usually occurs
1067 * when the stack overflows (such is the case with infinite recursion,
1070 * XXX Note that the current PTD gets replaced by IdlePTD when the
1071 * task switch occurs. This means that the stack that was active at
1072 * the time of the double fault is not available at <kstack> unless
1073 * the machine was idle when the double fault occurred. The downside
1074 * of this is that "trace <ebp>" in ddb won't work.
1077 dblfault_handler(void)
1079 struct mdglobaldata
*gd
= mdcpu
;
1081 kprintf("\nFatal double fault:\n");
1082 kprintf("eip = 0x%x\n", gd
->gd_common_tss
.tss_eip
);
1083 kprintf("esp = 0x%x\n", gd
->gd_common_tss
.tss_esp
);
1084 kprintf("ebp = 0x%x\n", gd
->gd_common_tss
.tss_ebp
);
1086 /* two separate prints in case of a trap on an unmapped page */
1087 kprintf("mp_lock = %08x; ", mp_lock
);
1088 kprintf("cpuid = %d\n", mycpu
->gd_cpuid
);
1090 panic("double fault");
1094 * Compensate for 386 brain damage (missing URKR).
1095 * This is a little simpler than the pagefault handler in trap() because
1096 * it the page tables have already been faulted in and high addresses
1097 * are thrown out early for other reasons.
1100 trapwrite(unsigned addr
)
1107 va
= trunc_page((vm_offset_t
)addr
);
1109 * XXX - MAX is END. Changed > to >= for temp. fix.
1111 if (va
>= VM_MAX_USER_ADDRESS
)
1114 lp
= curthread
->td_lwp
;
1115 vm
= lp
->lwp_vmspace
;
1117 PHOLD(lp
->lwp_proc
);
1119 if (!grow_stack (lp
->lwp_proc
, va
)) {
1120 PRELE(lp
->lwp_proc
);
1125 * fault the data page
1127 rv
= vm_fault(&vm
->vm_map
, va
, VM_PROT_WRITE
, VM_FAULT_DIRTY
);
1129 PRELE(lp
->lwp_proc
);
1131 if (rv
!= KERN_SUCCESS
)
1138 * syscall2 - MP aware system call request C handler
1140 * A system call is essentially treated as a trap except that the
1141 * MP lock is not held on entry or return. We are responsible for
1142 * obtaining the MP lock if necessary and for handling ASTs
1143 * (e.g. a task switch) prior to return.
1148 syscall2(struct trapframe
*frame
)
1150 struct thread
*td
= curthread
;
1151 struct proc
*p
= td
->td_proc
;
1152 struct lwp
*lp
= td
->td_lwp
;
1154 struct sysent
*callp
;
1155 register_t orig_tf_eflags
;
1160 int crit_count
= td
->td_pri
& ~TDPRI_MASK
;
1163 int have_mplock
= 0;
1166 union sysunion args
;
1168 KTR_LOG(kernentry_syscall
, lp
->lwp_proc
->p_pid
, lp
->lwp_tid
,
1172 KASSERT(td
->td_mpcount
== 0, ("badmpcount syscall2 from %p", (void *)frame
->tf_eip
));
1173 if (syscall_mpsafe
== 0)
1174 MAKEMPSAFE(have_mplock
);
1176 userenter(td
, p
); /* lazy raise our priority */
1181 sticks
= (int)td
->td_sticks
;
1182 orig_tf_eflags
= frame
->tf_eflags
;
1185 * Virtual kernel intercept - if a VM context managed by a virtual
1186 * kernel issues a system call the virtual kernel handles it, not us.
1187 * Restore the virtual kernel context and return from its system
1188 * call. The current frame is copied out to the virtual kernel.
1190 if (lp
->lwp_vkernel
&& lp
->lwp_vkernel
->ve
) {
1191 error
= vkernel_trap(lp
, frame
);
1192 frame
->tf_eax
= error
;
1194 frame
->tf_eflags
|= PSL_C
;
1195 error
= EJUSTRETURN
;
1200 * Get the system call parameters and account for time
1202 lp
->lwp_md
.md_regs
= frame
;
1203 params
= (caddr_t
)frame
->tf_esp
+ sizeof(int);
1204 code
= frame
->tf_eax
;
1206 if (p
->p_sysent
->sv_prepsyscall
) {
1207 (*p
->p_sysent
->sv_prepsyscall
)(
1208 frame
, (int *)(&args
.nosys
.sysmsg
+ 1),
1212 * Need to check if this is a 32 bit or 64 bit syscall.
1213 * fuword is MP aware.
1215 if (code
== SYS_syscall
) {
1217 * Code is first argument, followed by actual args.
1219 code
= fuword(params
);
1220 params
+= sizeof(int);
1221 } else if (code
== SYS___syscall
) {
1223 * Like syscall, but code is a quad, so as to maintain
1224 * quad alignment for the rest of the arguments.
1226 code
= fuword(params
);
1227 params
+= sizeof(quad_t
);
1231 code
&= p
->p_sysent
->sv_mask
;
1232 if (code
>= p
->p_sysent
->sv_size
)
1233 callp
= &p
->p_sysent
->sv_table
[0];
1235 callp
= &p
->p_sysent
->sv_table
[code
];
1237 narg
= callp
->sy_narg
& SYF_ARGMASK
;
1240 * copyin is MP aware, but the tracing code is not
1242 if (narg
&& params
) {
1243 error
= copyin(params
, (caddr_t
)(&args
.nosys
.sysmsg
+ 1),
1244 narg
* sizeof(register_t
));
1247 if (KTRPOINT(td
, KTR_SYSCALL
)) {
1248 MAKEMPSAFE(have_mplock
);
1250 ktrsyscall(lp
, code
, narg
,
1251 (void *)(&args
.nosys
.sysmsg
+ 1));
1259 if (KTRPOINT(td
, KTR_SYSCALL
)) {
1260 MAKEMPSAFE(have_mplock
);
1261 ktrsyscall(lp
, code
, narg
, (void *)(&args
.nosys
.sysmsg
+ 1));
1266 * For traditional syscall code edx is left untouched when 32 bit
1267 * results are returned. Since edx is loaded from fds[1] when the
1268 * system call returns we pre-set it here.
1270 args
.sysmsg_fds
[0] = 0;
1271 args
.sysmsg_fds
[1] = frame
->tf_edx
;
1274 * The syscall might manipulate the trap frame. If it does it
1275 * will probably return EJUSTRETURN.
1277 args
.sysmsg_frame
= frame
;
1279 STOPEVENT(p
, S_SCE
, narg
); /* MP aware */
1282 * NOTE: All system calls run MPSAFE now. The system call itself
1283 * is responsible for getting the MP lock.
1285 error
= (*callp
->sy_call
)(&args
);
1288 kprintf("system call %d returned %d\n", code
, error
);
1293 * MP SAFE (we may or may not have the MP lock at this point)
1298 * Reinitialize proc pointer `p' as it may be different
1299 * if this is a child returning from fork syscall.
1302 lp
= curthread
->td_lwp
;
1303 frame
->tf_eax
= args
.sysmsg_fds
[0];
1304 frame
->tf_edx
= args
.sysmsg_fds
[1];
1305 frame
->tf_eflags
&= ~PSL_C
;
1309 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1310 * int 0x80 is 2 bytes. We saved this in tf_err.
1312 frame
->tf_eip
-= frame
->tf_err
;
1317 panic("Unexpected EASYNC return value (for now)");
1320 if (p
->p_sysent
->sv_errsize
) {
1321 if (error
>= p
->p_sysent
->sv_errsize
)
1322 error
= -1; /* XXX */
1324 error
= p
->p_sysent
->sv_errtbl
[error
];
1326 frame
->tf_eax
= error
;
1327 frame
->tf_eflags
|= PSL_C
;
1332 * Traced syscall. trapsignal() is not MP aware.
1334 if ((orig_tf_eflags
& PSL_T
) /*&& !(orig_tf_eflags & PSL_VM)*/) {
1335 MAKEMPSAFE(have_mplock
);
1336 frame
->tf_eflags
&= ~PSL_T
;
1337 trapsignal(lp
, SIGTRAP
, TRAP_TRACE
);
1341 * Handle reschedule and other end-of-syscall issues
1343 userret(lp
, frame
, sticks
);
1346 if (KTRPOINT(td
, KTR_SYSRET
)) {
1347 MAKEMPSAFE(have_mplock
);
1348 ktrsysret(lp
, code
, error
, args
.sysmsg_result
);
1353 * This works because errno is findable through the
1354 * register set. If we ever support an emulation where this
1355 * is not the case, this code will need to be revisited.
1357 STOPEVENT(p
, S_SCX
, code
);
1362 * Release the MP lock if we had to get it
1364 KASSERT(td
->td_mpcount
== have_mplock
,
1365 ("badmpcount syscall2/end from %p", (void *)frame
->tf_eip
));
1369 KTR_LOG(kernentry_syscall_ret
, lp
->lwp_proc
->p_pid
, lp
->lwp_tid
, error
);
1371 KASSERT(crit_count
== (td
->td_pri
& ~TDPRI_MASK
),
1372 ("syscall: critical section count mismatch! %d/%d",
1373 crit_count
/ TDPRI_CRIT
, td
->td_pri
/ TDPRI_CRIT
));
1378 fork_return(struct lwp
*lp
, struct trapframe
*frame
)
1380 frame
->tf_eax
= 0; /* Child returns zero */
1381 frame
->tf_eflags
&= ~PSL_C
; /* success */
1384 generic_lwp_return(lp
, frame
);
1385 KTR_LOG(kernentry_fork_ret
, lp
->lwp_proc
->p_pid
, lp
->lwp_tid
);
1389 * Simplified back end of syscall(), used when returning from fork()
1390 * or lwp_create() directly into user mode. MP lock is held on entry and
1391 * should be released on return. This code will return back into the fork
1392 * trampoline code which then runs doreti.
1395 generic_lwp_return(struct lwp
*lp
, struct trapframe
*frame
)
1397 struct proc
*p
= lp
->lwp_proc
;
1400 * Newly forked processes are given a kernel priority. We have to
1401 * adjust the priority to a normal user priority and fake entry
1402 * into the kernel (call userenter()) to install a passive release
1403 * function just in case userret() decides to stop the process. This
1404 * can occur when ^Z races a fork. If we do not install the passive
1405 * release function the current process designation will not be
1406 * released when the thread goes to sleep.
1408 lwkt_setpri_self(TDPRI_USER_NORM
);
1409 userenter(lp
->lwp_thread
, p
);
1410 userret(lp
, frame
, 0);
1412 if (KTRPOINT(lp
->lwp_thread
, KTR_SYSRET
))
1413 ktrsysret(lp
, SYS_fork
, 0, 0);
1415 p
->p_flag
|= P_PASSIVE_ACQ
;
1417 p
->p_flag
&= ~P_PASSIVE_ACQ
;
1419 KKASSERT(lp
->lwp_thread
->td_mpcount
== 1);
1425 * doreti has turned into this. The frame is directly on the stack. We
1426 * pull everything else we need (fpu and tls context) from the current
1429 * Note on fpu interactions: In a virtual kernel, the fpu context for
1430 * an emulated user mode process is not shared with the virtual kernel's
1431 * fpu context, so we only have to 'stack' fpu contexts within the virtual
1432 * kernel itself, and not even then since the signal() contexts that we care
1433 * about save and restore the FPU state (I think anyhow).
1435 * vmspace_ctl() returns an error only if it had problems instaling the
1436 * context we supplied or problems copying data to/from our VM space.
1439 go_user(struct intrframe
*frame
)
1441 struct trapframe
*tf
= (void *)&frame
->if_gs
;
1445 * Interrupts may be disabled on entry, make sure all signals
1446 * can be received before beginning our loop.
1451 * Switch to the current simulated user process, then call
1452 * user_trap() when we break out of it (usually due to a signal).
1456 * Tell the real kernel whether it is ok to use the FP
1459 * The critical section is required to prevent an interrupt
1460 * from causing a preemptive task switch and changing
1464 if (mdcpu
->gd_npxthread
== curthread
) {
1465 tf
->tf_xflags
&= ~PGEX_FPFAULT
;
1467 tf
->tf_xflags
|= PGEX_FPFAULT
;
1471 * Run emulated user process context. This call interlocks
1472 * with new mailbox signals.
1474 * Set PGEX_U unconditionally, indicating a user frame (the
1475 * bit is normally set only by T_PAGEFLT).
1477 r
= vmspace_ctl(&curproc
->p_vmspace
->vm_pmap
, VMSPACE_CTL_RUN
,
1478 tf
, &curthread
->td_savevext
);
1480 frame
->if_xflags
|= PGEX_U
;
1482 kprintf("GO USER %d trap %d EVA %08x EIP %08x ESP %08x XFLAGS %02x/%02x\n",
1483 r
, tf
->tf_trapno
, tf
->tf_err
, tf
->tf_eip
, tf
->tf_esp
,
1484 tf
->tf_xflags
, frame
->if_xflags
);
1488 panic("vmspace_ctl failed");
1490 if (tf
->tf_trapno
) {
1494 if (mycpu
->gd_reqflags
& RQF_AST_MASK
) {
1495 tf
->tf_trapno
= T_ASTFLT
;
1503 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1504 * fault (which is then passed back to the virtual kernel) if an attempt is
1505 * made to use the FP unit.
1507 * XXX this is a fairly big hack.
1510 set_vkernel_fp(struct trapframe
*frame
)
1512 struct thread
*td
= curthread
;
1514 if (frame
->tf_xflags
& PGEX_FPFAULT
) {
1515 td
->td_pcb
->pcb_flags
|= FP_VIRTFP
;
1516 if (mdcpu
->gd_npxthread
== td
)
1519 td
->td_pcb
->pcb_flags
&= ~FP_VIRTFP
;