2 * Copyright (c) 1990, 1993
3 * The Regents of the University of California. All rights reserved.
4 * Copyright (C) 1994, David Greenman
5 * Copyright (c) 2008 The DragonFly Project.
6 * Copyright (c) 2008 Jordan Gordeev.
8 * This code is derived from software contributed to Berkeley by
9 * the University of Utah, and William Jolitz.
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. All advertising materials mentioning features or use of this software
20 * must display the following acknowledgement:
21 * This product includes software developed by the University of
22 * California, Berkeley and its contributors.
23 * 4. Neither the name of the University nor the names of its contributors
24 * may be used to endorse or promote products derived from this software
25 * without specific prior written permission.
27 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39 * from: @(#)trap.c 7.4 (Berkeley) 5/13/91
40 * $FreeBSD: src/sys/i386/i386/trap.c,v 1.147.2.11 2003/02/27 19:09:59 luoqi Exp $
44 * x86_64 Trap and System call handling
50 #include "opt_ktrace.h"
52 #include <machine/frame.h>
53 #include <sys/param.h>
54 #include <sys/systm.h>
55 #include <sys/kernel.h>
56 #include <sys/kerneldump.h>
58 #include <sys/pioctl.h>
59 #include <sys/types.h>
60 #include <sys/signal2.h>
61 #include <sys/syscall.h>
62 #include <sys/sysctl.h>
63 #include <sys/sysent.h>
65 #include <sys/ktrace.h>
68 #include <sys/sysmsg.h>
69 #include <sys/sysproto.h>
70 #include <sys/sysunion.h>
74 #include <vm/vm_extern.h>
75 #include <vm/vm_kern.h>
76 #include <vm/vm_param.h>
77 #include <machine/cpu.h>
78 #include <machine/pcb.h>
79 #include <machine/smp.h>
80 #include <machine/thread.h>
81 #include <machine/clock.h>
82 #include <machine/vmparam.h>
83 #include <machine/md_var.h>
84 #include <machine_base/isa/isa_intr.h>
85 #include <machine_base/apic/lapic.h>
89 #include <sys/thread2.h>
90 #include <sys/spinlock2.h>
92 extern void trap(struct trapframe
*frame
);
94 static int trap_pfault(struct trapframe
*, int);
95 static void trap_fatal(struct trapframe
*, vm_offset_t
);
96 void dblfault_handler(struct trapframe
*frame
);
98 #define MAX_TRAP_MSG 30
99 static char *trap_msg
[] = {
101 "privileged instruction fault", /* 1 T_PRIVINFLT */
103 "breakpoint instruction fault", /* 3 T_BPTFLT */
106 "arithmetic trap", /* 6 T_ARITHTRAP */
107 "system forced exception", /* 7 T_ASTFLT */
109 "general protection fault", /* 9 T_PROTFLT */
110 "trace trap", /* 10 T_TRCTRAP */
112 "page fault", /* 12 T_PAGEFLT */
114 "alignment fault", /* 14 T_ALIGNFLT */
118 "integer divide fault", /* 18 T_DIVIDE */
119 "non-maskable interrupt trap", /* 19 T_NMI */
120 "overflow trap", /* 20 T_OFLOW */
121 "FPU bounds check fault", /* 21 T_BOUND */
122 "FPU device not available", /* 22 T_DNA */
123 "double fault", /* 23 T_DOUBLEFLT */
124 "FPU operand fetch fault", /* 24 T_FPOPFLT */
125 "invalid TSS fault", /* 25 T_TSSFLT */
126 "segment not present fault", /* 26 T_SEGNPFLT */
127 "stack fault", /* 27 T_STKFLT */
128 "machine check trap", /* 28 T_MCHK */
129 "SIMD floating-point exception", /* 29 T_XMMFLT */
130 "reserved (unknown) fault", /* 30 T_RESERVED */
134 static int ddb_on_nmi
= 1;
135 SYSCTL_INT(_machdep
, OID_AUTO
, ddb_on_nmi
, CTLFLAG_RW
,
136 &ddb_on_nmi
, 0, "Go to DDB on NMI");
137 static int ddb_on_seg_fault
= 0;
138 SYSCTL_INT(_machdep
, OID_AUTO
, ddb_on_seg_fault
, CTLFLAG_RW
,
139 &ddb_on_seg_fault
, 0, "Go to DDB on user seg-fault");
140 static int freeze_on_seg_fault
= 0;
141 SYSCTL_INT(_machdep
, OID_AUTO
, freeze_on_seg_fault
, CTLFLAG_RW
,
142 &freeze_on_seg_fault
, 0, "Go to DDB on user seg-fault");
144 static int panic_on_nmi
= 1;
145 SYSCTL_INT(_machdep
, OID_AUTO
, panic_on_nmi
, CTLFLAG_RW
,
146 &panic_on_nmi
, 0, "Panic on NMI");
147 static int fast_release
;
148 SYSCTL_INT(_machdep
, OID_AUTO
, fast_release
, CTLFLAG_RW
,
149 &fast_release
, 0, "Passive Release was optimal");
150 static int slow_release
;
151 SYSCTL_INT(_machdep
, OID_AUTO
, slow_release
, CTLFLAG_RW
,
152 &slow_release
, 0, "Passive Release was nonoptimal");
155 * System call debugging records the worst-case system call
156 * overhead (inclusive of blocking), but may be inaccurate.
158 /*#define SYSCALL_DEBUG*/
160 uint64_t SysCallsWorstCase
[SYS_MAXSYSCALL
];
164 * Passively intercepts the thread switch function to increase
165 * the thread priority from a user priority to a kernel priority, reducing
166 * syscall and trap overhead for the case where no switch occurs.
168 * Synchronizes td_ucred with p_ucred. This is used by system calls,
169 * signal handling, faults, AST traps, and anything else that enters the
170 * kernel from userland and provides the kernel with a stable read-only
171 * copy of the process ucred.
173 * To avoid races with another thread updating p_ucred we obtain p_spin.
174 * The other thread doing the update will obtain both p_token and p_spin.
175 * In the case where the cached cred pointer matches, we will already have
176 * the ref and we don't have to do one blessed thing.
179 userenter(struct thread
*curtd
, struct proc
*curp
)
184 curtd
->td_release
= lwkt_passive_release
;
186 if (curtd
->td_ucred
!= curp
->p_ucred
) {
187 spin_lock(&curp
->p_spin
);
188 ncred
= crhold(curp
->p_ucred
);
189 spin_unlock(&curp
->p_spin
);
190 ocred
= curtd
->td_ucred
;
191 curtd
->td_ucred
= ncred
;
198 * Debugging, remove top two user stack pages to catch kernel faults
200 if (freeze_on_seg_fault
> 1 && curtd
->td_lwp
) {
201 pmap_remove(vmspace_pmap(curtd
->td_lwp
->lwp_vmspace
),
202 0x00007FFFFFFFD000LU
,
203 0x0000800000000000LU
);
209 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
210 * must be completed before we can return to or try to return to userland.
212 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
213 * arithmatic on the delta calculation so the absolute tick values are
214 * truncated to an integer.
217 userret(struct lwp
*lp
, struct trapframe
*frame
, int sticks
)
219 struct proc
*p
= lp
->lwp_proc
;
223 * Charge system time if profiling. Note: times are in microseconds.
224 * This may do a copyout and block, so do it first even though it
225 * means some system time will be charged as user time.
227 if (p
->p_flags
& P_PROFIL
) {
228 addupc_task(p
, frame
->tf_rip
,
229 (u_int
)((int)lp
->lwp_thread
->td_sticks
- sticks
));
234 * Specific on-return-to-usermode checks (LWP_MP_WEXIT,
235 * LWP_MP_VNLRU, etc).
237 if (lp
->lwp_mpflags
& LWP_MP_URETMASK
)
241 * Block here if we are in a stopped state.
243 if (STOPLWP(p
, lp
)) {
244 lwkt_gettoken(&p
->p_token
);
246 lwkt_reltoken(&p
->p_token
);
249 while (dump_stop_usertds
) {
250 tsleep(&dump_stop_usertds
, 0, "dumpstp", 0);
254 * Post any pending upcalls. If running a virtual kernel be sure
255 * to restore the virtual kernel's vmspace before posting the upcall.
257 if (p
->p_flags
& (P_SIGVTALRM
| P_SIGPROF
)) {
258 lwkt_gettoken(&p
->p_token
);
259 if (p
->p_flags
& P_SIGVTALRM
) {
260 p
->p_flags
&= ~P_SIGVTALRM
;
261 ksignal(p
, SIGVTALRM
);
263 if (p
->p_flags
& P_SIGPROF
) {
264 p
->p_flags
&= ~P_SIGPROF
;
267 lwkt_reltoken(&p
->p_token
);
272 * Post any pending signals. If running a virtual kernel be sure
273 * to restore the virtual kernel's vmspace before posting the signal.
275 * WARNING! postsig() can exit and not return.
277 if ((sig
= CURSIG_TRACE(lp
)) != 0) {
278 lwkt_gettoken(&p
->p_token
);
280 lwkt_reltoken(&p
->p_token
);
285 * block here if we are swapped out, but still process signals
286 * (such as SIGKILL). proc0 (the swapin scheduler) is already
287 * aware of our situation, we do not have to wake it up.
289 if (p
->p_flags
& P_SWAPPEDOUT
) {
290 lwkt_gettoken(&p
->p_token
);
291 p
->p_flags
|= P_SWAPWAIT
;
293 if (p
->p_flags
& P_SWAPWAIT
)
294 tsleep(p
, PCATCH
, "SWOUT", 0);
295 p
->p_flags
&= ~P_SWAPWAIT
;
296 lwkt_reltoken(&p
->p_token
);
301 * In a multi-threaded program it is possible for a thread to change
302 * signal state during a system call which temporarily changes the
303 * signal mask. In this case postsig() might not be run and we
304 * have to restore the mask ourselves.
306 if (lp
->lwp_flags
& LWP_OLDMASK
) {
307 lp
->lwp_flags
&= ~LWP_OLDMASK
;
308 lp
->lwp_sigmask
= lp
->lwp_oldsigmask
;
314 * Cleanup from userenter and any passive release that might have occured.
315 * We must reclaim the current-process designation before we can return
316 * to usermode. We also handle both LWKT and USER reschedule requests.
319 userexit(struct lwp
*lp
)
321 struct thread
*td
= lp
->lwp_thread
;
322 /* globaldata_t gd = td->td_gd; */
325 * Handle stop requests at kernel priority. Any requests queued
326 * after this loop will generate another AST.
328 while (STOPLWP(lp
->lwp_proc
, lp
)) {
329 lwkt_gettoken(&lp
->lwp_proc
->p_token
);
331 lwkt_reltoken(&lp
->lwp_proc
->p_token
);
335 * Reduce our priority in preparation for a return to userland. If
336 * our passive release function was still in place, our priority was
337 * never raised and does not need to be reduced.
339 lwkt_passive_recover(td
);
341 /* WARNING: we may have migrated cpu's */
342 /* gd = td->td_gd; */
345 * Become the current user scheduled process if we aren't already,
346 * and deal with reschedule requests and other factors.
348 lp
->lwp_proc
->p_usched
->acquire_curproc(lp
);
351 #if !defined(KTR_KERNENTRY)
352 #define KTR_KERNENTRY KTR_ALL
354 KTR_INFO_MASTER(kernentry
);
355 KTR_INFO(KTR_KERNENTRY
, kernentry
, trap
, 0,
356 "TRAP(pid %d, tid %d, trapno %ld, eva %lu)",
357 pid_t pid
, lwpid_t tid
, register_t trapno
, vm_offset_t eva
);
358 KTR_INFO(KTR_KERNENTRY
, kernentry
, trap_ret
, 0, "TRAP_RET(pid %d, tid %d)",
359 pid_t pid
, lwpid_t tid
);
360 KTR_INFO(KTR_KERNENTRY
, kernentry
, syscall
, 0, "SYSC(pid %d, tid %d, nr %ld)",
361 pid_t pid
, lwpid_t tid
, register_t trapno
);
362 KTR_INFO(KTR_KERNENTRY
, kernentry
, syscall_ret
, 0, "SYSRET(pid %d, tid %d, err %d)",
363 pid_t pid
, lwpid_t tid
, int err
);
364 KTR_INFO(KTR_KERNENTRY
, kernentry
, fork_ret
, 0, "FORKRET(pid %d, tid %d)",
365 pid_t pid
, lwpid_t tid
);
368 * Exception, fault, and trap interface to the kernel.
369 * This common code is called from assembly language IDT gate entry
370 * routines that prepare a suitable stack frame, and restore this
371 * frame after the exception has been processed.
373 * This function is also called from doreti in an interlock to handle ASTs.
374 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
376 * NOTE! We have to retrieve the fault address prior to potentially
377 * blocking, including blocking on any token.
379 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
380 * if an attempt is made to switch from a fast interrupt or IPI.
383 trap(struct trapframe
*frame
)
385 struct globaldata
*gd
= mycpu
;
386 struct thread
*td
= gd
->gd_curthread
;
387 struct lwp
*lp
= td
->td_lwp
;
390 int i
= 0, ucode
= 0, type
, code
;
392 int crit_count
= td
->td_critcount
;
393 lwkt_tokref_t curstop
= td
->td_toks_stop
;
402 * We need to allow T_DNA faults when the debugger is active since
403 * some dumping paths do large bcopy() which use the floating
404 * point registers for faster copying.
406 if (db_active
&& frame
->tf_trapno
!= T_DNA
) {
407 eva
= (frame
->tf_trapno
== T_PAGEFLT
? frame
->tf_addr
: 0);
408 ++gd
->gd_trap_nesting_level
;
409 trap_fatal(frame
, eva
);
410 --gd
->gd_trap_nesting_level
;
417 if ((frame
->tf_rflags
& PSL_I
) == 0) {
419 * Buggy application or kernel code has disabled interrupts
420 * and then trapped. Enabling interrupts now is wrong, but
421 * it is better than running with interrupts disabled until
422 * they are accidentally enabled later.
424 type
= frame
->tf_trapno
;
425 if (ISPL(frame
->tf_cs
) == SEL_UPL
) {
426 /* JG curproc can be NULL */
428 "pid %ld (%s): trap %d with interrupts disabled\n",
429 (long)curproc
->p_pid
, curproc
->p_comm
, type
);
430 } else if (type
!= T_NMI
&& type
!= T_BPTFLT
&&
433 * XXX not quite right, since this may be for a
434 * multiple fault in user mode.
436 kprintf("kernel trap %d (%s @ 0x%016jx) with "
437 "interrupts disabled\n",
445 type
= frame
->tf_trapno
;
446 code
= frame
->tf_err
;
448 if (ISPL(frame
->tf_cs
) == SEL_UPL
) {
451 KTR_LOG(kernentry_trap
, p
->p_pid
, lp
->lwp_tid
,
452 frame
->tf_trapno
, eva
);
456 sticks
= (int)td
->td_sticks
;
457 KASSERT(lp
->lwp_md
.md_regs
== frame
,
458 ("Frame mismatch %p %p", lp
->lwp_md
.md_regs
, frame
));
461 case T_PRIVINFLT
: /* privileged instruction fault */
466 case T_BPTFLT
: /* bpt instruction fault */
467 case T_TRCTRAP
: /* trace trap */
468 frame
->tf_rflags
&= ~PSL_T
;
470 ucode
= (type
== T_TRCTRAP
? TRAP_TRACE
: TRAP_BRKPT
);
473 case T_ARITHTRAP
: /* arithmetic trap */
478 case T_ASTFLT
: /* Allow process switch */
479 mycpu
->gd_cnt
.v_soft
++;
480 if (mycpu
->gd_reqflags
& RQF_AST_OWEUPC
) {
481 atomic_clear_int(&mycpu
->gd_reqflags
,
483 addupc_task(p
, p
->p_prof
.pr_addr
,
488 case T_PROTFLT
: /* general protection fault */
492 case T_STKFLT
: /* stack fault */
493 case T_SEGNPFLT
: /* segment not present fault */
497 case T_TSSFLT
: /* invalid TSS fault */
498 case T_DOUBLEFLT
: /* double fault */
504 case T_PAGEFLT
: /* page fault */
505 i
= trap_pfault(frame
, TRUE
);
506 if (frame
->tf_rip
== 0) {
508 /* used for kernel debugging only */
509 while (freeze_on_seg_fault
)
510 tsleep(p
, 0, "freeze", hz
* 20);
513 if (i
== -1 || i
== 0)
525 case T_DIVIDE
: /* integer divide fault */
532 /* machine/parity/power fail/"kitchen sink" faults */
533 if (isa_nmi(code
) == 0) {
536 * NMI can be hooked up to a pushbutton
540 kprintf ("NMI ... going to debugger\n");
541 kdb_trap(type
, 0, frame
);
545 } else if (panic_on_nmi
)
546 panic("NMI indicates hardware failure");
548 #endif /* NISA > 0 */
550 case T_OFLOW
: /* integer overflow fault */
555 case T_BOUND
: /* bounds check fault */
562 * Virtual kernel intercept - pass the DNA exception
563 * to the virtual kernel if it asked to handle it.
564 * This occurs when the virtual kernel is holding
565 * onto the FP context for a different emulated
566 * process then the one currently running.
568 * We must still call npxdna() since we may have
569 * saved FP state that the virtual kernel needs
570 * to hand over to a different emulated process.
572 if (lp
->lwp_vkernel
&& lp
->lwp_vkernel
->ve
&&
573 (td
->td_pcb
->pcb_flags
& FP_VIRTFP
)
580 * The kernel may have switched out the FP unit's
581 * state, causing the user process to take a fault
582 * when it tries to use the FP unit. Restore the
588 ucode
= FPE_FPU_NP_TRAP
;
591 case T_FPOPFLT
: /* FPU operand fetch fault */
596 case T_XMMFLT
: /* SIMD floating-point exception */
605 case T_PAGEFLT
: /* page fault */
606 trap_pfault(frame
, FALSE
);
611 * The kernel is apparently using fpu for copying.
612 * XXX this should be fatal unless the kernel has
613 * registered such use.
619 case T_STKFLT
: /* stack fault */
620 case T_PROTFLT
: /* general protection fault */
621 case T_SEGNPFLT
: /* segment not present fault */
623 * Invalid segment selectors and out of bounds
624 * %rip's and %rsp's can be set up in user mode.
625 * This causes a fault in kernel mode when the
626 * kernel tries to return to user mode. We want
627 * to get this fault so that we can fix the
628 * problem here and not have to check all the
629 * selectors and pointers when the user changes
632 if (mycpu
->gd_intr_nesting_level
== 0) {
634 * NOTE: in 64-bit mode traps push rsp/ss
635 * even if no ring change occurs.
637 if (td
->td_pcb
->pcb_onfault
&&
638 td
->td_pcb
->pcb_onfault_sp
==
640 frame
->tf_rip
= (register_t
)
641 td
->td_pcb
->pcb_onfault
;
644 if (frame
->tf_rip
== (long)doreti_iret
) {
645 frame
->tf_rip
= (long)doreti_iret_fault
;
653 * PSL_NT can be set in user mode and isn't cleared
654 * automatically when the kernel is entered. This
655 * causes a TSS fault when the kernel attempts to
656 * `iret' because the TSS link is uninitialized. We
657 * want to get this fault so that we can fix the
658 * problem here and not every time the kernel is
661 if (frame
->tf_rflags
& PSL_NT
) {
662 frame
->tf_rflags
&= ~PSL_NT
;
667 case T_TRCTRAP
: /* trace trap */
669 if (frame
->tf_rip
== (int)IDTVEC(syscall
)) {
671 * We've just entered system mode via the
672 * syscall lcall. Continue single stepping
673 * silently until the syscall handler has
678 if (frame
->tf_rip
== (int)IDTVEC(syscall
) + 1) {
680 * The syscall handler has now saved the
681 * flags. Stop single stepping it.
683 frame
->tf_rflags
&= ~PSL_T
;
689 * Ignore debug register trace traps due to
690 * accesses in the user's address space, which
691 * can happen under several conditions such as
692 * if a user sets a watchpoint on a buffer and
693 * then passes that buffer to a system call.
694 * We still want to get TRCTRAPS for addresses
695 * in kernel space because that is useful when
696 * debugging the kernel.
699 if (user_dbreg_trap()) {
701 * Reset breakpoint bits because the
704 /* XXX check upper bits here */
705 load_dr6(rdr6() & 0xfffffff0);
710 * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
714 * If DDB is enabled, let it handle the debugger trap.
715 * Otherwise, debugger traps "can't happen".
719 if (kdb_trap(type
, 0, frame
))
726 /* machine/parity/power fail/"kitchen sink" faults */
727 if (isa_nmi(code
) == 0) {
730 * NMI can be hooked up to a pushbutton
734 kprintf ("NMI ... going to debugger\n");
735 kdb_trap(type
, 0, frame
);
739 } else if (panic_on_nmi
== 0)
742 #endif /* NISA > 0 */
744 trap_fatal(frame
, 0);
749 * Virtual kernel intercept - if the fault is directly related to a
750 * VM context managed by a virtual kernel then let the virtual kernel
753 if (lp
->lwp_vkernel
&& lp
->lwp_vkernel
->ve
) {
754 vkernel_trap(lp
, frame
);
758 /* Translate fault for emulators (e.g. Linux) */
759 if (*p
->p_sysent
->sv_transtrap
)
760 i
= (*p
->p_sysent
->sv_transtrap
)(i
, type
);
762 trapsignal(lp
, i
, ucode
);
765 if (type
<= MAX_TRAP_MSG
) {
766 uprintf("fatal process exception: %s",
768 if ((type
== T_PAGEFLT
) || (type
== T_PROTFLT
))
769 uprintf(", fault VA = 0x%lx", frame
->tf_addr
);
775 userret(lp
, frame
, sticks
);
778 if (p
!= NULL
&& lp
!= NULL
)
779 KTR_LOG(kernentry_trap_ret
, p
->p_pid
, lp
->lwp_tid
);
781 KASSERT(crit_count
== td
->td_critcount
,
782 ("trap: critical section count mismatch! %d/%d",
783 crit_count
, td
->td_pri
));
784 KASSERT(curstop
== td
->td_toks_stop
,
785 ("trap: extra tokens held after trap! %ld/%ld",
786 curstop
- &td
->td_toks_base
,
787 td
->td_toks_stop
- &td
->td_toks_base
));
792 trap_handle_userenter(struct thread
*td
)
794 userenter(td
, td
->td_proc
);
798 trap_handle_userexit(struct trapframe
*frame
, int sticks
)
800 struct lwp
*lp
= curthread
->td_lwp
;
803 userret(lp
, frame
, sticks
);
809 trap_pfault(struct trapframe
*frame
, int usermode
)
812 struct vmspace
*vm
= NULL
;
817 thread_t td
= curthread
;
818 struct lwp
*lp
= td
->td_lwp
;
821 va
= trunc_page(frame
->tf_addr
);
822 if (va
>= VM_MIN_KERNEL_ADDRESS
) {
824 * Don't allow user-mode faults in kernel address space.
835 * This is a fault on non-kernel virtual memory.
836 * vm is initialized above to NULL. If curproc is NULL
837 * or curproc->p_vmspace is NULL the fault is fatal.
840 vm
= lp
->lwp_vmspace
;
849 * Debugging, try to catch kernel faults on the user address
850 * space when not inside on onfault (e.g. copyin/copyout)
853 if (usermode
== 0 && (td
->td_pcb
== NULL
||
854 td
->td_pcb
->pcb_onfault
== NULL
)) {
856 if (freeze_on_seg_fault
) {
857 kprintf("trap_pfault: user address fault from kernel mode "
858 "%016lx\n", (long)frame
->tf_addr
);
859 while (freeze_on_seg_fault
)
860 tsleep(&freeze_on_seg_fault
, 0, "frzseg", hz
* 20);
868 * PGEX_I is defined only if the execute disable bit capability is
869 * supported and enabled.
871 if (frame
->tf_err
& PGEX_W
)
872 ftype
= VM_PROT_WRITE
;
874 else if ((frame
->tf_err
& PGEX_I
) && pg_nx
!= 0)
875 ftype
= VM_PROT_EXECUTE
;
878 ftype
= VM_PROT_READ
;
880 if (map
!= &kernel_map
) {
882 * Keep swapout from messing with us during this
892 fault_flags
|= VM_FAULT_BURST
| VM_FAULT_USERMODE
;
893 if (ftype
& VM_PROT_WRITE
)
894 fault_flags
|= VM_FAULT_DIRTY
;
896 fault_flags
|= VM_FAULT_NORMAL
;
897 rv
= vm_fault(map
, va
, ftype
, fault_flags
);
902 * Don't have to worry about process locking or stacks in the
905 fault_flags
= VM_FAULT_NORMAL
;
906 rv
= vm_fault(map
, va
, ftype
, VM_FAULT_NORMAL
);
908 if (rv
== KERN_SUCCESS
)
913 * NOTE: in 64-bit mode traps push rsp/ss
914 * even if no ring change occurs.
916 if (td
->td_pcb
->pcb_onfault
&&
917 td
->td_pcb
->pcb_onfault_sp
== frame
->tf_rsp
&&
918 td
->td_gd
->gd_intr_nesting_level
== 0) {
919 frame
->tf_rip
= (register_t
)td
->td_pcb
->pcb_onfault
;
922 trap_fatal(frame
, frame
->tf_addr
);
927 * NOTE: on x86_64 we have a tf_addr field in the trapframe, no
928 * kludge is needed to pass the fault address to signal handlers.
932 if (td
->td_lwp
->lwp_vkernel
== NULL
) {
933 while (freeze_on_seg_fault
) {
934 tsleep(p
, 0, "freeze", hz
* 20);
936 if (ddb_on_seg_fault
)
937 Debugger("ddb_on_seg_fault");
941 return((rv
== KERN_PROTECTION_FAILURE
) ? SIGBUS
: SIGSEGV
);
945 trap_fatal(struct trapframe
*frame
, vm_offset_t eva
)
950 struct soft_segment_descriptor softseg
;
953 code
= frame
->tf_err
;
954 type
= frame
->tf_trapno
;
955 sdtossd(&gdt
[IDXSEL(frame
->tf_cs
& 0xffff)], &softseg
);
957 if (type
<= MAX_TRAP_MSG
)
958 msg
= trap_msg
[type
];
961 kprintf("\n\nFatal trap %d: %s while in %s mode\n", type
, msg
,
962 ISPL(frame
->tf_cs
) == SEL_UPL
? "user" : "kernel");
963 /* three separate prints in case of a trap on an unmapped page */
964 kprintf("cpuid = %d; ", mycpu
->gd_cpuid
);
965 kprintf("lapic->id = %08x\n", lapic
->id
);
966 if (type
== T_PAGEFLT
) {
967 kprintf("fault virtual address = 0x%lx\n", eva
);
968 kprintf("fault code = %s %s %s, %s\n",
969 code
& PGEX_U
? "user" : "supervisor",
970 code
& PGEX_W
? "write" : "read",
971 code
& PGEX_I
? "instruction" : "data",
972 code
& PGEX_P
? "protection violation" : "page not present");
974 kprintf("instruction pointer = 0x%lx:0x%lx\n",
975 frame
->tf_cs
& 0xffff, frame
->tf_rip
);
976 if (ISPL(frame
->tf_cs
) == SEL_UPL
) {
977 ss
= frame
->tf_ss
& 0xffff;
981 * NOTE: in 64-bit mode traps push rsp/ss even if no ring
984 ss
= GSEL(GDATA_SEL
, SEL_KPL
);
987 kprintf("stack pointer = 0x%x:0x%lx\n", ss
, rsp
);
988 kprintf("frame pointer = 0x%x:0x%lx\n", ss
, frame
->tf_rbp
);
989 kprintf("code segment = base 0x%lx, limit 0x%lx, type 0x%x\n",
990 softseg
.ssd_base
, softseg
.ssd_limit
, softseg
.ssd_type
);
991 kprintf(" = DPL %d, pres %d, long %d, def32 %d, gran %d\n",
992 softseg
.ssd_dpl
, softseg
.ssd_p
, softseg
.ssd_long
, softseg
.ssd_def32
,
994 kprintf("processor eflags = ");
995 if (frame
->tf_rflags
& PSL_T
)
996 kprintf("trace trap, ");
997 if (frame
->tf_rflags
& PSL_I
)
998 kprintf("interrupt enabled, ");
999 if (frame
->tf_rflags
& PSL_NT
)
1000 kprintf("nested task, ");
1001 if (frame
->tf_rflags
& PSL_RF
)
1002 kprintf("resume, ");
1003 kprintf("IOPL = %ld\n", (frame
->tf_rflags
& PSL_IOPL
) >> 12);
1004 kprintf("current process = ");
1007 (u_long
)curproc
->p_pid
);
1011 kprintf("current thread = pri %d ", curthread
->td_pri
);
1012 if (curthread
->td_critcount
)
1017 if ((debugger_on_panic
|| db_active
) && kdb_trap(type
, code
, frame
))
1020 kprintf("trap number = %d\n", type
);
1021 if (type
<= MAX_TRAP_MSG
)
1022 panic("%s", trap_msg
[type
]);
1024 panic("unknown/reserved trap");
1028 * Double fault handler. Called when a fault occurs while writing
1029 * a frame for a trap/exception onto the stack. This usually occurs
1030 * when the stack overflows (such is the case with infinite recursion,
1035 in_kstack_guard(register_t rptr
)
1037 thread_t td
= curthread
;
1039 if ((char *)rptr
>= td
->td_kstack
&&
1040 (char *)rptr
< td
->td_kstack
+ PAGE_SIZE
) {
1047 dblfault_handler(struct trapframe
*frame
)
1049 thread_t td
= curthread
;
1051 if (in_kstack_guard(frame
->tf_rsp
) || in_kstack_guard(frame
->tf_rbp
)) {
1052 kprintf("DOUBLE FAULT - KERNEL STACK GUARD HIT!\n");
1053 if (in_kstack_guard(frame
->tf_rsp
))
1054 frame
->tf_rsp
= (register_t
)(td
->td_kstack
+ PAGE_SIZE
);
1055 if (in_kstack_guard(frame
->tf_rbp
))
1056 frame
->tf_rbp
= (register_t
)(td
->td_kstack
+ PAGE_SIZE
);
1058 kprintf("DOUBLE FAULT\n");
1060 kprintf("\nFatal double fault\n");
1061 kprintf("rip = 0x%lx\n", frame
->tf_rip
);
1062 kprintf("rsp = 0x%lx\n", frame
->tf_rsp
);
1063 kprintf("rbp = 0x%lx\n", frame
->tf_rbp
);
1064 /* three separate prints in case of a trap on an unmapped page */
1065 kprintf("cpuid = %d; ", mycpu
->gd_cpuid
);
1066 kprintf("lapic->id = %08x\n", lapic
->id
);
1067 panic("double fault");
1071 * syscall2 - MP aware system call request C handler
1073 * A system call is essentially treated as a trap except that the
1074 * MP lock is not held on entry or return. We are responsible for
1075 * obtaining the MP lock if necessary and for handling ASTs
1076 * (e.g. a task switch) prior to return.
1081 syscall2(struct trapframe
*frame
)
1083 struct thread
*td
= curthread
;
1084 struct proc
*p
= td
->td_proc
;
1085 struct lwp
*lp
= td
->td_lwp
;
1087 struct sysent
*callp
;
1088 register_t orig_tf_rflags
;
1093 int crit_count
= td
->td_critcount
;
1098 union sysunion args
;
1099 register_t
*argsdst
;
1101 mycpu
->gd_cnt
.v_syscall
++;
1104 if (ISPL(frame
->tf_cs
) != SEL_UPL
) {
1110 KTR_LOG(kernentry_syscall
, p
->p_pid
, lp
->lwp_tid
,
1113 userenter(td
, p
); /* lazy raise our priority */
1120 sticks
= (int)td
->td_sticks
;
1121 orig_tf_rflags
= frame
->tf_rflags
;
1124 * Virtual kernel intercept - if a VM context managed by a virtual
1125 * kernel issues a system call the virtual kernel handles it, not us.
1126 * Restore the virtual kernel context and return from its system
1127 * call. The current frame is copied out to the virtual kernel.
1129 if (lp
->lwp_vkernel
&& lp
->lwp_vkernel
->ve
) {
1130 vkernel_trap(lp
, frame
);
1131 error
= EJUSTRETURN
;
1137 * Get the system call parameters and account for time
1139 KASSERT(lp
->lwp_md
.md_regs
== frame
,
1140 ("Frame mismatch %p %p", lp
->lwp_md
.md_regs
, frame
));
1141 params
= (caddr_t
)frame
->tf_rsp
+ sizeof(register_t
);
1142 code
= frame
->tf_rax
;
1144 if (p
->p_sysent
->sv_prepsyscall
) {
1145 (*p
->p_sysent
->sv_prepsyscall
)(
1146 frame
, (int *)(&args
.nosys
.sysmsg
+ 1),
1149 if (code
== SYS_syscall
|| code
== SYS___syscall
) {
1150 code
= frame
->tf_rdi
;
1156 if (p
->p_sysent
->sv_mask
)
1157 code
&= p
->p_sysent
->sv_mask
;
1159 if (code
>= p
->p_sysent
->sv_size
)
1160 callp
= &p
->p_sysent
->sv_table
[0];
1162 callp
= &p
->p_sysent
->sv_table
[code
];
1164 narg
= callp
->sy_narg
& SYF_ARGMASK
;
1167 * On x86_64 we get up to six arguments in registers. The rest are
1168 * on the stack. The first six members of 'struct trapframe' happen
1169 * to be the registers used to pass arguments, in exactly the right
1172 argp
= &frame
->tf_rdi
;
1174 argsdst
= (register_t
*)(&args
.nosys
.sysmsg
+ 1);
1176 * JG can we overflow the space pointed to by 'argsdst'
1177 * either with 'bcopy' or with 'copyin'?
1179 bcopy(argp
, argsdst
, sizeof(register_t
) * regcnt
);
1181 * copyin is MP aware, but the tracing code is not
1183 if (narg
> regcnt
) {
1184 KASSERT(params
!= NULL
, ("copyin args with no params!"));
1185 error
= copyin(params
, &argsdst
[regcnt
],
1186 (narg
- regcnt
) * sizeof(register_t
));
1189 if (KTRPOINT(td
, KTR_SYSCALL
)) {
1190 ktrsyscall(lp
, code
, narg
,
1191 (void *)(&args
.nosys
.sysmsg
+ 1));
1199 if (KTRPOINT(td
, KTR_SYSCALL
)) {
1200 ktrsyscall(lp
, code
, narg
, (void *)(&args
.nosys
.sysmsg
+ 1));
1205 * Default return value is 0 (will be copied to %rax). Double-value
1206 * returns use %rax and %rdx. %rdx is left unchanged for system
1207 * calls which return only one result.
1209 args
.sysmsg_fds
[0] = 0;
1210 args
.sysmsg_fds
[1] = frame
->tf_rdx
;
1213 * The syscall might manipulate the trap frame. If it does it
1214 * will probably return EJUSTRETURN.
1216 args
.sysmsg_frame
= frame
;
1218 STOPEVENT(p
, S_SCE
, narg
); /* MP aware */
1221 * NOTE: All system calls run MPSAFE now. The system call itself
1222 * is responsible for getting the MP lock.
1224 #ifdef SYSCALL_DEBUG
1225 uint64_t tscval
= rdtsc();
1227 error
= (*callp
->sy_call
)(&args
);
1228 #ifdef SYSCALL_DEBUG
1229 tscval
= rdtsc() - tscval
;
1230 tscval
= tscval
* 1000000 / tsc_frequency
;
1231 if (SysCallsWorstCase
[code
] < tscval
)
1232 SysCallsWorstCase
[code
] = tscval
;
1237 * MP SAFE (we may or may not have the MP lock at this point)
1239 //kprintf("SYSMSG %d ", error);
1243 * Reinitialize proc pointer `p' as it may be different
1244 * if this is a child returning from fork syscall.
1247 lp
= curthread
->td_lwp
;
1248 frame
->tf_rax
= args
.sysmsg_fds
[0];
1249 frame
->tf_rdx
= args
.sysmsg_fds
[1];
1250 frame
->tf_rflags
&= ~PSL_C
;
1254 * Reconstruct pc, we know that 'syscall' is 2 bytes.
1255 * We have to do a full context restore so that %r10
1256 * (which was holding the value of %rcx) is restored for
1257 * the next iteration.
1259 if (frame
->tf_err
!= 0 && frame
->tf_err
!= 2)
1260 kprintf("lp %s:%d frame->tf_err is weird %ld\n",
1261 td
->td_comm
, lp
->lwp_proc
->p_pid
, frame
->tf_err
);
1262 frame
->tf_rip
-= frame
->tf_err
;
1263 frame
->tf_r10
= frame
->tf_rcx
;
1268 panic("Unexpected EASYNC return value (for now)");
1271 if (p
->p_sysent
->sv_errsize
) {
1272 if (error
>= p
->p_sysent
->sv_errsize
)
1273 error
= -1; /* XXX */
1275 error
= p
->p_sysent
->sv_errtbl
[error
];
1277 frame
->tf_rax
= error
;
1278 frame
->tf_rflags
|= PSL_C
;
1283 * Traced syscall. trapsignal() should now be MP aware
1285 if (orig_tf_rflags
& PSL_T
) {
1286 frame
->tf_rflags
&= ~PSL_T
;
1287 trapsignal(lp
, SIGTRAP
, TRAP_TRACE
);
1291 * Handle reschedule and other end-of-syscall issues
1293 userret(lp
, frame
, sticks
);
1296 if (KTRPOINT(td
, KTR_SYSRET
)) {
1297 ktrsysret(lp
, code
, error
, args
.sysmsg_result
);
1302 * This works because errno is findable through the
1303 * register set. If we ever support an emulation where this
1304 * is not the case, this code will need to be revisited.
1306 STOPEVENT(p
, S_SCX
, code
);
1309 KTR_LOG(kernentry_syscall_ret
, p
->p_pid
, lp
->lwp_tid
, error
);
1311 KASSERT(crit_count
== td
->td_critcount
,
1312 ("syscall: critical section count mismatch! %d/%d",
1313 crit_count
, td
->td_pri
));
1314 KASSERT(&td
->td_toks_base
== td
->td_toks_stop
,
1315 ("syscall: %ld extra tokens held after trap! syscall %p",
1316 td
->td_toks_stop
- &td
->td_toks_base
,
1322 fork_return(struct lwp
*lp
, struct trapframe
*frame
)
1324 frame
->tf_rax
= 0; /* Child returns zero */
1325 frame
->tf_rflags
&= ~PSL_C
; /* success */
1328 generic_lwp_return(lp
, frame
);
1329 KTR_LOG(kernentry_fork_ret
, lp
->lwp_proc
->p_pid
, lp
->lwp_tid
);
1333 * Simplified back end of syscall(), used when returning from fork()
1334 * directly into user mode.
1336 * This code will return back into the fork trampoline code which then
1340 generic_lwp_return(struct lwp
*lp
, struct trapframe
*frame
)
1342 struct proc
*p
= lp
->lwp_proc
;
1345 * Check for exit-race. If one lwp exits the process concurrent with
1346 * another lwp creating a new thread, the two operations may cross
1347 * each other resulting in the newly-created lwp not receiving a
1350 if (p
->p_flags
& P_WEXIT
) {
1351 lwpsignal(p
, lp
, SIGKILL
);
1355 * Newly forked processes are given a kernel priority. We have to
1356 * adjust the priority to a normal user priority and fake entry
1357 * into the kernel (call userenter()) to install a passive release
1358 * function just in case userret() decides to stop the process. This
1359 * can occur when ^Z races a fork. If we do not install the passive
1360 * release function the current process designation will not be
1361 * released when the thread goes to sleep.
1363 lwkt_setpri_self(TDPRI_USER_NORM
);
1364 userenter(lp
->lwp_thread
, p
);
1365 userret(lp
, frame
, 0);
1367 if (KTRPOINT(lp
->lwp_thread
, KTR_SYSRET
))
1368 ktrsysret(lp
, SYS_fork
, 0, 0);
1370 lp
->lwp_flags
|= LWP_PASSIVE_ACQ
;
1372 lp
->lwp_flags
&= ~LWP_PASSIVE_ACQ
;
1376 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1377 * fault (which is then passed back to the virtual kernel) if an attempt is
1378 * made to use the FP unit.
1380 * XXX this is a fairly big hack.
1383 set_vkernel_fp(struct trapframe
*frame
)
1385 struct thread
*td
= curthread
;
1387 if (frame
->tf_xflags
& PGEX_FPFAULT
) {
1388 td
->td_pcb
->pcb_flags
|= FP_VIRTFP
;
1389 if (mdcpu
->gd_npxthread
== td
)
1392 td
->td_pcb
->pcb_flags
&= ~FP_VIRTFP
;
1397 * Called from vkernel_trap() to fixup the vkernel's syscall
1398 * frame for vmspace_ctl() return.
1401 cpu_vkernel_trap(struct trapframe
*frame
, int error
)
1403 frame
->tf_rax
= error
;
1405 frame
->tf_rflags
|= PSL_C
;
1407 frame
->tf_rflags
&= ~PSL_C
;