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 $
42 * x86_64 Trap and System call handling
48 #include "opt_ktrace.h"
50 #include <sys/param.h>
51 #include <sys/systm.h>
53 #include <sys/pioctl.h>
54 #include <sys/kernel.h>
55 #include <sys/resourcevar.h>
56 #include <sys/signalvar.h>
57 #include <sys/signal2.h>
58 #include <sys/syscall.h>
59 #include <sys/sysctl.h>
60 #include <sys/sysent.h>
62 #include <sys/vmmeter.h>
63 #include <sys/malloc.h>
65 #include <sys/ktrace.h>
68 #include <sys/vkernel.h>
69 #include <sys/sysproto.h>
70 #include <sys/sysunion.h>
71 #include <sys/vmspace.h>
74 #include <vm/vm_param.h>
77 #include <vm/vm_kern.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_page.h>
80 #include <vm/vm_extern.h>
82 #include <machine/cpu.h>
83 #include <machine/md_var.h>
84 #include <machine/pcb.h>
85 #include <machine/smp.h>
86 #include <machine/tss.h>
87 #include <machine/globaldata.h>
91 #include <sys/msgport2.h>
92 #include <sys/thread2.h>
93 #include <sys/mplock2.h>
95 #define MAKEMPSAFE(have_mplock) \
96 if (have_mplock == 0) { \
101 int (*pmath_emulate
) (struct trapframe
*);
103 extern int trapwrite (unsigned addr
);
105 static int trap_pfault (struct trapframe
*, int, vm_offset_t
);
106 static void trap_fatal (struct trapframe
*, int, vm_offset_t
);
107 void dblfault_handler (void);
108 extern int vmm_enabled
;
111 extern inthand_t
IDTVEC(syscall
);
114 #define MAX_TRAP_MSG 30
115 static char *trap_msg
[] = {
117 "privileged instruction fault", /* 1 T_PRIVINFLT */
119 "breakpoint instruction fault", /* 3 T_BPTFLT */
122 "arithmetic trap", /* 6 T_ARITHTRAP */
123 "system forced exception", /* 7 T_ASTFLT */
125 "general protection fault", /* 9 T_PROTFLT */
126 "trace trap", /* 10 T_TRCTRAP */
128 "page fault", /* 12 T_PAGEFLT */
130 "alignment fault", /* 14 T_ALIGNFLT */
134 "integer divide fault", /* 18 T_DIVIDE */
135 "non-maskable interrupt trap", /* 19 T_NMI */
136 "overflow trap", /* 20 T_OFLOW */
137 "FPU bounds check fault", /* 21 T_BOUND */
138 "FPU device not available", /* 22 T_DNA */
139 "double fault", /* 23 T_DOUBLEFLT */
140 "FPU operand fetch fault", /* 24 T_FPOPFLT */
141 "invalid TSS fault", /* 25 T_TSSFLT */
142 "segment not present fault", /* 26 T_SEGNPFLT */
143 "stack fault", /* 27 T_STKFLT */
144 "machine check trap", /* 28 T_MCHK */
145 "SIMD floating-point exception", /* 29 T_XMMFLT */
146 "reserved (unknown) fault", /* 30 T_RESERVED */
150 static int ddb_on_nmi
= 1;
151 SYSCTL_INT(_machdep
, OID_AUTO
, ddb_on_nmi
, CTLFLAG_RW
,
152 &ddb_on_nmi
, 0, "Go to DDB on NMI");
154 static int panic_on_nmi
= 1;
155 SYSCTL_INT(_machdep
, OID_AUTO
, panic_on_nmi
, CTLFLAG_RW
,
156 &panic_on_nmi
, 0, "Panic on NMI");
157 static int fast_release
;
158 SYSCTL_INT(_machdep
, OID_AUTO
, fast_release
, CTLFLAG_RW
,
159 &fast_release
, 0, "Passive Release was optimal");
160 static int slow_release
;
161 SYSCTL_INT(_machdep
, OID_AUTO
, slow_release
, CTLFLAG_RW
,
162 &slow_release
, 0, "Passive Release was nonoptimal");
165 * Passively intercepts the thread switch function to increase
166 * the thread priority from a user priority to a kernel priority, reducing
167 * syscall and trap overhead for the case where no switch occurs.
169 * Synchronizes td_ucred with p_ucred. This is used by system calls,
170 * signal handling, faults, AST traps, and anything else that enters the
171 * kernel from userland and provides the kernel with a stable read-only
172 * copy of the process ucred.
175 userenter(struct thread
*curtd
, struct proc
*curp
)
180 curtd
->td_release
= lwkt_passive_release
;
182 if (curtd
->td_ucred
!= curp
->p_ucred
) {
183 ncred
= crhold(curp
->p_ucred
);
184 ocred
= curtd
->td_ucred
;
185 curtd
->td_ucred
= ncred
;
192 * Handle signals, profiling, and other AST's and/or tasks that
193 * must be completed before we can return to or try to return to userland.
195 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
196 * arithmatic on the delta calculation so the absolute tick values are
197 * truncated to an integer.
200 userret(struct lwp
*lp
, struct trapframe
*frame
, int sticks
)
202 struct proc
*p
= lp
->lwp_proc
;
206 * Charge system time if profiling. Note: times are in microseconds.
207 * This may do a copyout and block, so do it first even though it
208 * means some system time will be charged as user time.
210 if (p
->p_flags
& P_PROFIL
) {
211 addupc_task(p
, frame
->tf_rip
,
212 (u_int
)((int)lp
->lwp_thread
->td_sticks
- sticks
));
217 * Specific on-return-to-usermode checks (LWP_MP_WEXIT,
218 * LWP_MP_VNLRU, etc).
220 if (lp
->lwp_mpflags
& LWP_MP_URETMASK
)
224 * Block here if we are in a stopped state.
226 if (p
->p_stat
== SSTOP
|| p
->p_stat
== SCORE
) {
227 lwkt_gettoken(&p
->p_token
);
229 lwkt_reltoken(&p
->p_token
);
234 * Post any pending upcalls. If running a virtual kernel be sure
235 * to restore the virtual kernel's vmspace before posting the upcall.
237 if (p
->p_flags
& (P_SIGVTALRM
| P_SIGPROF
)) {
238 lwkt_gettoken(&p
->p_token
);
239 if (p
->p_flags
& P_SIGVTALRM
) {
240 p
->p_flags
&= ~P_SIGVTALRM
;
241 ksignal(p
, SIGVTALRM
);
243 if (p
->p_flags
& P_SIGPROF
) {
244 p
->p_flags
&= ~P_SIGPROF
;
247 lwkt_reltoken(&p
->p_token
);
252 * Post any pending signals
254 * WARNING! postsig() can exit and not return.
256 if ((sig
= CURSIG_TRACE(lp
)) != 0) {
257 lwkt_gettoken(&p
->p_token
);
259 lwkt_reltoken(&p
->p_token
);
264 * block here if we are swapped out, but still process signals
265 * (such as SIGKILL). proc0 (the swapin scheduler) is already
266 * aware of our situation, we do not have to wake it up.
268 if (p
->p_flags
& P_SWAPPEDOUT
) {
269 lwkt_gettoken(&p
->p_token
);
271 p
->p_flags
|= P_SWAPWAIT
;
273 if (p
->p_flags
& P_SWAPWAIT
)
274 tsleep(p
, PCATCH
, "SWOUT", 0);
275 p
->p_flags
&= ~P_SWAPWAIT
;
277 lwkt_reltoken(&p
->p_token
);
282 * In a multi-threaded program it is possible for a thread to change
283 * signal state during a system call which temporarily changes the
284 * signal mask. In this case postsig() might not be run and we
285 * have to restore the mask ourselves.
287 if (lp
->lwp_flags
& LWP_OLDMASK
) {
288 lp
->lwp_flags
&= ~LWP_OLDMASK
;
289 lp
->lwp_sigmask
= lp
->lwp_oldsigmask
;
295 * Cleanup from userenter and any passive release that might have occured.
296 * We must reclaim the current-process designation before we can return
297 * to usermode. We also handle both LWKT and USER reschedule requests.
300 userexit(struct lwp
*lp
)
302 struct thread
*td
= lp
->lwp_thread
;
303 /* globaldata_t gd = td->td_gd; */
306 * Handle stop requests at kernel priority. Any requests queued
307 * after this loop will generate another AST.
309 while (lp
->lwp_proc
->p_stat
== SSTOP
||
310 lp
->lwp_proc
->p_stat
== SCORE
) {
311 lwkt_gettoken(&lp
->lwp_proc
->p_token
);
313 lwkt_reltoken(&lp
->lwp_proc
->p_token
);
317 * Reduce our priority in preparation for a return to userland. If
318 * our passive release function was still in place, our priority was
319 * never raised and does not need to be reduced.
321 lwkt_passive_recover(td
);
324 * Become the current user scheduled process if we aren't already,
325 * and deal with reschedule requests and other factors.
327 lp
->lwp_proc
->p_usched
->acquire_curproc(lp
);
328 /* WARNING: we may have migrated cpu's */
329 /* gd = td->td_gd; */
332 #if !defined(KTR_KERNENTRY)
333 #define KTR_KERNENTRY KTR_ALL
335 KTR_INFO_MASTER(kernentry
);
336 KTR_INFO(KTR_KERNENTRY
, kernentry
, trap
, 0,
337 "TRAP(pid %hd, tid %hd, trapno %ld, eva %lu)",
338 pid_t pid
, lwpid_t tid
, register_t trapno
, vm_offset_t eva
);
339 KTR_INFO(KTR_KERNENTRY
, kernentry
, trap_ret
, 0, "TRAP_RET(pid %hd, tid %hd)",
340 pid_t pid
, lwpid_t tid
);
341 KTR_INFO(KTR_KERNENTRY
, kernentry
, syscall
, 0, "SYSC(pid %hd, tid %hd, nr %ld)",
342 pid_t pid
, lwpid_t tid
, register_t trapno
);
343 KTR_INFO(KTR_KERNENTRY
, kernentry
, syscall_ret
, 0, "SYSRET(pid %hd, tid %hd, err %d)",
344 pid_t pid
, lwpid_t tid
, int err
);
345 KTR_INFO(KTR_KERNENTRY
, kernentry
, fork_ret
, 0, "FORKRET(pid %hd, tid %hd)",
346 pid_t pid
, lwpid_t tid
);
349 * Exception, fault, and trap interface to the kernel.
350 * This common code is called from assembly language IDT gate entry
351 * routines that prepare a suitable stack frame, and restore this
352 * frame after the exception has been processed.
354 * This function is also called from doreti in an interlock to handle ASTs.
355 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
357 * NOTE! We have to retrieve the fault address prior to obtaining the
358 * MP lock because get_mplock() may switch out. YYY cr2 really ought
359 * to be retrieved by the assembly code, not here.
361 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
362 * if an attempt is made to switch from a fast interrupt or IPI. This is
363 * necessary to properly take fatal kernel traps on SMP machines if
364 * get_mplock() has to block.
368 user_trap(struct trapframe
*frame
)
370 struct globaldata
*gd
= mycpu
;
371 struct thread
*td
= gd
->gd_curthread
;
372 struct lwp
*lp
= td
->td_lwp
;
375 int i
= 0, ucode
= 0, type
, code
;
378 int crit_count
= td
->td_critcount
;
379 lwkt_tokref_t curstop
= td
->td_toks_stop
;
385 if (frame
->tf_trapno
== T_PAGEFLT
)
386 eva
= frame
->tf_addr
;
390 kprintf("USER_TRAP AT %08lx xflags %ld trapno %ld eva %08lx\n",
391 frame
->tf_rip
, frame
->tf_xflags
, frame
->tf_trapno
, eva
);
395 * Everything coming from user mode runs through user_trap,
396 * including system calls.
398 if (frame
->tf_trapno
== T_FAST_SYSCALL
) {
403 KTR_LOG(kernentry_trap
, lp
->lwp_proc
->p_pid
, lp
->lwp_tid
,
404 frame
->tf_trapno
, eva
);
408 eva
= (frame
->tf_trapno
== T_PAGEFLT
? rcr2() : 0);
409 ++gd
->gd_trap_nesting_level
;
410 MAKEMPSAFE(have_mplock
);
411 trap_fatal(frame
, TRUE
, eva
);
412 --gd
->gd_trap_nesting_level
;
417 type
= frame
->tf_trapno
;
418 code
= frame
->tf_err
;
422 sticks
= (int)td
->td_sticks
;
423 lp
->lwp_md
.md_regs
= frame
;
426 case T_PRIVINFLT
: /* privileged instruction fault */
431 case T_BPTFLT
: /* bpt instruction fault */
432 case T_TRCTRAP
: /* trace trap */
433 frame
->tf_rflags
&= ~PSL_T
;
435 ucode
= (type
== T_TRCTRAP
? TRAP_TRACE
: TRAP_BRKPT
);
438 case T_ARITHTRAP
: /* arithmetic trap */
443 case T_ASTFLT
: /* Allow process switch */
444 mycpu
->gd_cnt
.v_soft
++;
445 if (mycpu
->gd_reqflags
& RQF_AST_OWEUPC
) {
446 atomic_clear_int(&mycpu
->gd_reqflags
, RQF_AST_OWEUPC
);
447 addupc_task(p
, p
->p_prof
.pr_addr
, p
->p_prof
.pr_ticks
);
452 * The following two traps can happen in
453 * vm86 mode, and, if so, we want to handle
456 case T_PROTFLT
: /* general protection fault */
457 case T_STKFLT
: /* stack fault */
459 if (frame
->tf_eflags
& PSL_VM
) {
460 i
= vm86_emulate((struct vm86frame
*)frame
);
468 case T_SEGNPFLT
: /* segment not present fault */
469 case T_TSSFLT
: /* invalid TSS fault */
470 case T_DOUBLEFLT
: /* double fault */
473 ucode
= code
+ BUS_SEGM_FAULT
;
476 case T_PAGEFLT
: /* page fault */
477 i
= trap_pfault(frame
, TRUE
, eva
);
478 if (i
== -1 || i
== 0)
490 case T_DIVIDE
: /* integer divide fault */
497 MAKEMPSAFE(have_mplock
);
498 /* machine/parity/power fail/"kitchen sink" faults */
499 if (isa_nmi(code
) == 0) {
502 * NMI can be hooked up to a pushbutton
506 kprintf ("NMI ... going to debugger\n");
507 kdb_trap(type
, 0, frame
);
511 } else if (panic_on_nmi
)
512 panic("NMI indicates hardware failure");
514 #endif /* NISA > 0 */
516 case T_OFLOW
: /* integer overflow fault */
521 case T_BOUND
: /* bounds check fault */
528 * Virtual kernel intercept - pass the DNA exception
529 * to the (emulated) virtual kernel if it asked to handle
530 * it. This occurs when the virtual kernel is holding
531 * onto the FP context for a different emulated
532 * process then the one currently running.
534 * We must still call npxdna() since we may have
535 * saved FP state that the (emulated) virtual kernel
536 * needs to hand over to a different emulated process.
538 if (lp
->lwp_vkernel
&& lp
->lwp_vkernel
->ve
&&
539 (td
->td_pcb
->pcb_flags
& FP_VIRTFP
)
546 * The kernel may have switched out the FP unit's
547 * state, causing the user process to take a fault
548 * when it tries to use the FP unit. Restore the
553 if (!pmath_emulate
) {
555 ucode
= FPE_FPU_NP_TRAP
;
558 i
= (*pmath_emulate
)(frame
);
560 if (!(frame
->tf_rflags
& PSL_T
))
562 frame
->tf_rflags
&= ~PSL_T
;
565 /* else ucode = emulator_only_knows() XXX */
568 case T_FPOPFLT
: /* FPU operand fetch fault */
573 case T_XMMFLT
: /* SIMD floating-point exception */
580 * Virtual kernel intercept - if the fault is directly related to a
581 * VM context managed by a virtual kernel then let the virtual kernel
584 if (lp
->lwp_vkernel
&& lp
->lwp_vkernel
->ve
) {
585 vkernel_trap(lp
, frame
);
590 * Translate fault for emulators (e.g. Linux)
592 if (*p
->p_sysent
->sv_transtrap
)
593 i
= (*p
->p_sysent
->sv_transtrap
)(i
, type
);
595 MAKEMPSAFE(have_mplock
);
596 trapsignal(lp
, i
, ucode
);
599 if (type
<= MAX_TRAP_MSG
) {
600 uprintf("fatal process exception: %s",
602 if ((type
== T_PAGEFLT
) || (type
== T_PROTFLT
))
603 uprintf(", fault VA = 0x%lx", (u_long
)eva
);
609 userret(lp
, frame
, sticks
);
614 KTR_LOG(kernentry_trap_ret
, lp
->lwp_proc
->p_pid
, lp
->lwp_tid
);
616 KASSERT(crit_count
== td
->td_critcount
,
617 ("trap: critical section count mismatch! %d/%d",
618 crit_count
, td
->td_pri
));
619 KASSERT(curstop
== td
->td_toks_stop
,
620 ("trap: extra tokens held after trap! %ld/%ld",
621 curstop
- &td
->td_toks_base
,
622 td
->td_toks_stop
- &td
->td_toks_base
));
627 kern_trap(struct trapframe
*frame
)
629 struct globaldata
*gd
= mycpu
;
630 struct thread
*td
= gd
->gd_curthread
;
633 int i
= 0, ucode
= 0, type
, code
;
636 int crit_count
= td
->td_critcount
;
637 lwkt_tokref_t curstop
= td
->td_toks_stop
;
644 if (frame
->tf_trapno
== T_PAGEFLT
)
645 eva
= frame
->tf_addr
;
651 ++gd
->gd_trap_nesting_level
;
652 MAKEMPSAFE(have_mplock
);
653 trap_fatal(frame
, FALSE
, eva
);
654 --gd
->gd_trap_nesting_level
;
659 type
= frame
->tf_trapno
;
660 code
= frame
->tf_err
;
668 case T_PAGEFLT
: /* page fault */
669 trap_pfault(frame
, FALSE
, eva
);
674 * The kernel may be using npx for copying or other
677 panic("kernel NPX should not happen");
682 case T_PROTFLT
: /* general protection fault */
683 case T_SEGNPFLT
: /* segment not present fault */
685 * Invalid segment selectors and out of bounds
686 * %eip's and %esp's can be set up in user mode.
687 * This causes a fault in kernel mode when the
688 * kernel tries to return to user mode. We want
689 * to get this fault so that we can fix the
690 * problem here and not have to check all the
691 * selectors and pointers when the user changes
694 if (mycpu
->gd_intr_nesting_level
== 0) {
695 if (td
->td_pcb
->pcb_onfault
) {
697 (register_t
)td
->td_pcb
->pcb_onfault
;
705 * PSL_NT can be set in user mode and isn't cleared
706 * automatically when the kernel is entered. This
707 * causes a TSS fault when the kernel attempts to
708 * `iret' because the TSS link is uninitialized. We
709 * want to get this fault so that we can fix the
710 * problem here and not every time the kernel is
713 if (frame
->tf_rflags
& PSL_NT
) {
714 frame
->tf_rflags
&= ~PSL_NT
;
719 case T_TRCTRAP
: /* trace trap */
721 if (frame
->tf_eip
== (int)IDTVEC(syscall
)) {
723 * We've just entered system mode via the
724 * syscall lcall. Continue single stepping
725 * silently until the syscall handler has
730 if (frame
->tf_eip
== (int)IDTVEC(syscall
) + 1) {
732 * The syscall handler has now saved the
733 * flags. Stop single stepping it.
735 frame
->tf_eflags
&= ~PSL_T
;
741 * Ignore debug register trace traps due to
742 * accesses in the user's address space, which
743 * can happen under several conditions such as
744 * if a user sets a watchpoint on a buffer and
745 * then passes that buffer to a system call.
746 * We still want to get TRCTRAPS for addresses
747 * in kernel space because that is useful when
748 * debugging the kernel.
750 if (user_dbreg_trap()) {
752 * Reset breakpoint bits because the
755 load_dr6(rdr6() & 0xfffffff0);
760 * Fall through (TRCTRAP kernel mode, kernel address)
764 * If DDB is enabled, let it handle the debugger trap.
765 * Otherwise, debugger traps "can't happen".
768 MAKEMPSAFE(have_mplock
);
769 if (kdb_trap (type
, 0, frame
))
774 MAKEMPSAFE(have_mplock
);
775 trap_fatal(frame
, FALSE
, eva
);
778 MAKEMPSAFE(have_mplock
);
779 trap_fatal(frame
, FALSE
, eva
);
784 * Ignore this trap generated from a spurious SIGTRAP.
786 * single stepping in / syscalls leads to spurious / SIGTRAP
789 * Haiku (c) 2007 Simon 'corecode' Schubert
795 * Translate fault for emulators (e.g. Linux)
797 if (*p
->p_sysent
->sv_transtrap
)
798 i
= (*p
->p_sysent
->sv_transtrap
)(i
, type
);
800 MAKEMPSAFE(have_mplock
);
801 trapsignal(lp
, i
, ucode
);
804 if (type
<= MAX_TRAP_MSG
) {
805 uprintf("fatal process exception: %s",
807 if ((type
== T_PAGEFLT
) || (type
== T_PROTFLT
))
808 uprintf(", fault VA = 0x%lx", (u_long
)eva
);
818 KASSERT(crit_count
== td
->td_critcount
,
819 ("trap: critical section count mismatch! %d/%d",
820 crit_count
, td
->td_pri
));
821 KASSERT(curstop
== td
->td_toks_stop
,
822 ("trap: extra tokens held after trap! %ld/%ld",
823 curstop
- &td
->td_toks_base
,
824 td
->td_toks_stop
- &td
->td_toks_base
));
829 trap_pfault(struct trapframe
*frame
, int usermode
, vm_offset_t eva
)
832 struct vmspace
*vm
= NULL
;
836 thread_t td
= curthread
;
837 struct lwp
*lp
= td
->td_lwp
;
840 va
= trunc_page(eva
);
841 if (usermode
== FALSE
) {
843 * This is a fault on kernel virtual memory.
848 * This is a fault on non-kernel virtual memory.
849 * vm is initialized above to NULL. If curproc is NULL
850 * or curproc->p_vmspace is NULL the fault is fatal.
853 vm
= lp
->lwp_vmspace
;
861 if (frame
->tf_err
& PGEX_W
)
862 ftype
= VM_PROT_READ
| VM_PROT_WRITE
;
864 ftype
= VM_PROT_READ
;
866 if (map
!= &kernel_map
) {
868 * Keep swapout from messing with us during this
874 * Grow the stack if necessary
876 /* grow_stack returns false only if va falls into
877 * a growable stack region and the stack growth
878 * fails. It returns true if va was not within
879 * a growable stack region, or if the stack
882 if (!grow_stack (lp
->lwp_proc
, va
)) {
890 fault_flags
|= VM_FAULT_BURST
;
891 if (ftype
& VM_PROT_WRITE
)
892 fault_flags
|= VM_FAULT_DIRTY
;
894 fault_flags
|= VM_FAULT_NORMAL
;
895 rv
= vm_fault(map
, va
, ftype
, fault_flags
);
900 * Don't have to worry about process locking or stacks in the kernel.
902 rv
= vm_fault(map
, va
, ftype
, VM_FAULT_NORMAL
);
905 if (rv
== KERN_SUCCESS
)
909 if (td
->td_gd
->gd_intr_nesting_level
== 0 &&
910 td
->td_pcb
->pcb_onfault
) {
911 frame
->tf_rip
= (register_t
)td
->td_pcb
->pcb_onfault
;
914 trap_fatal(frame
, usermode
, eva
);
919 * NOTE: on x86_64 we have a tf_addr field in the trapframe, no
920 * kludge is needed to pass the fault address to signal handlers.
922 struct proc
*p
= td
->td_proc
;
923 kprintf("seg-fault accessing address %p rip=%p pid=%d p_comm=%s\n",
924 (void *)va
, (void *)frame
->tf_rip
, p
->p_pid
, p
->p_comm
);
925 /* Debugger("seg-fault"); */
927 return((rv
== KERN_PROTECTION_FAILURE
) ? SIGBUS
: SIGSEGV
);
931 trap_fatal(struct trapframe
*frame
, int usermode
, vm_offset_t eva
)
936 code
= frame
->tf_xflags
;
937 type
= frame
->tf_trapno
;
939 if (type
<= MAX_TRAP_MSG
) {
940 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
941 type
, trap_msg
[type
],
942 (usermode
? "user" : "kernel"));
944 /* two separate prints in case of a trap on an unmapped page */
945 kprintf("cpuid = %d\n", mycpu
->gd_cpuid
);
946 if (type
== T_PAGEFLT
) {
947 kprintf("fault virtual address = %p\n", (void *)eva
);
948 kprintf("fault code = %s %s, %s\n",
949 usermode
? "user" : "supervisor",
950 code
& PGEX_W
? "write" : "read",
951 code
& PGEX_P
? "protection violation" : "page not present");
953 kprintf("instruction pointer = 0x%lx:0x%lx\n",
954 frame
->tf_cs
& 0xffff, frame
->tf_rip
);
956 ss
= frame
->tf_ss
& 0xffff;
959 ss
= GSEL(GDATA_SEL
, SEL_KPL
);
960 rsp
= (long)&frame
->tf_rsp
;
962 kprintf("stack pointer = 0x%x:0x%lx\n", ss
, rsp
);
963 kprintf("frame pointer = 0x%x:0x%lx\n", ss
, frame
->tf_rbp
);
964 kprintf("processor eflags = ");
965 if (frame
->tf_rflags
& PSL_T
)
966 kprintf("trace trap, ");
967 if (frame
->tf_rflags
& PSL_I
)
968 kprintf("interrupt enabled, ");
969 if (frame
->tf_rflags
& PSL_NT
)
970 kprintf("nested task, ");
971 if (frame
->tf_rflags
& PSL_RF
)
974 if (frame
->tf_eflags
& PSL_VM
)
977 kprintf("IOPL = %jd\n", (intmax_t)((frame
->tf_rflags
& PSL_IOPL
) >> 12));
978 kprintf("current process = ");
980 kprintf("%lu (%s)\n",
981 (u_long
)curproc
->p_pid
, curproc
->p_comm
?
982 curproc
->p_comm
: "");
986 kprintf("current thread = pri %d ", curthread
->td_pri
);
987 if (curthread
->td_critcount
)
992 * we probably SHOULD have stopped the other CPUs before now!
993 * another CPU COULD have been touching cpl at this moment...
995 kprintf(" <- SMP: XXX");
1003 if ((debugger_on_panic
|| db_active
) && kdb_trap(type
, code
, frame
))
1006 kprintf("trap number = %d\n", type
);
1007 if (type
<= MAX_TRAP_MSG
)
1008 panic("%s", trap_msg
[type
]);
1010 panic("unknown/reserved trap");
1014 * Double fault handler. Called when a fault occurs while writing
1015 * a frame for a trap/exception onto the stack. This usually occurs
1016 * when the stack overflows (such is the case with infinite recursion,
1019 * XXX Note that the current PTD gets replaced by IdlePTD when the
1020 * task switch occurs. This means that the stack that was active at
1021 * the time of the double fault is not available at <kstack> unless
1022 * the machine was idle when the double fault occurred. The downside
1023 * of this is that "trace <ebp>" in ddb won't work.
1026 dblfault_handler(void)
1029 struct mdglobaldata
*gd
= mdcpu
;
1032 kprintf("\nFatal double fault:\n");
1034 kprintf("rip = 0x%lx\n", gd
->gd_common_tss
.tss_rip
);
1035 kprintf("rsp = 0x%lx\n", gd
->gd_common_tss
.tss_rsp
);
1036 kprintf("rbp = 0x%lx\n", gd
->gd_common_tss
.tss_rbp
);
1038 /* two separate prints in case of a trap on an unmapped page */
1039 kprintf("cpuid = %d\n", mycpu
->gd_cpuid
);
1040 panic("double fault");
1044 * Compensate for 386 brain damage (missing URKR).
1045 * This is a little simpler than the pagefault handler in trap() because
1046 * it the page tables have already been faulted in and high addresses
1047 * are thrown out early for other reasons.
1050 trapwrite(unsigned addr
)
1057 va
= trunc_page((vm_offset_t
)addr
);
1059 * XXX - MAX is END. Changed > to >= for temp. fix.
1061 if (va
>= VM_MAX_USER_ADDRESS
)
1064 lp
= curthread
->td_lwp
;
1065 vm
= lp
->lwp_vmspace
;
1067 PHOLD(lp
->lwp_proc
);
1069 if (!grow_stack (lp
->lwp_proc
, va
)) {
1070 PRELE(lp
->lwp_proc
);
1075 * fault the data page
1077 rv
= vm_fault(&vm
->vm_map
, va
, VM_PROT_WRITE
, VM_FAULT_DIRTY
);
1079 PRELE(lp
->lwp_proc
);
1081 if (rv
!= KERN_SUCCESS
)
1088 * syscall2 - MP aware system call request C handler
1090 * A system call is essentially treated as a trap except that the
1091 * MP lock is not held on entry or return. We are responsible for
1092 * obtaining the MP lock if necessary and for handling ASTs
1093 * (e.g. a task switch) prior to return.
1095 * In general, only simple access and manipulation of curproc and
1096 * the current stack is allowed without having to hold MP lock.
1098 * MPSAFE - note that large sections of this routine are run without
1102 syscall2(struct trapframe
*frame
)
1104 struct thread
*td
= curthread
;
1105 struct proc
*p
= td
->td_proc
;
1106 struct lwp
*lp
= td
->td_lwp
;
1108 struct sysent
*callp
;
1109 register_t orig_tf_rflags
;
1114 int crit_count
= td
->td_critcount
;
1115 lwkt_tokref_t curstop
= td
->td_toks_stop
;
1117 int have_mplock
= 0;
1121 union sysunion args
;
1122 register_t
*argsdst
;
1124 mycpu
->gd_cnt
.v_syscall
++;
1126 KTR_LOG(kernentry_syscall
, lp
->lwp_proc
->p_pid
, lp
->lwp_tid
,
1129 userenter(td
, p
); /* lazy raise our priority */
1136 sticks
= (int)td
->td_sticks
;
1137 orig_tf_rflags
= frame
->tf_rflags
;
1140 * Virtual kernel intercept - if a VM context managed by a virtual
1141 * kernel issues a system call the virtual kernel handles it, not us.
1142 * Restore the virtual kernel context and return from its system
1143 * call. The current frame is copied out to the virtual kernel.
1145 if (lp
->lwp_vkernel
&& lp
->lwp_vkernel
->ve
) {
1146 vkernel_trap(lp
, frame
);
1147 error
= EJUSTRETURN
;
1152 * Get the system call parameters and account for time
1154 lp
->lwp_md
.md_regs
= frame
;
1155 params
= (caddr_t
)frame
->tf_rsp
+ sizeof(register_t
);
1156 code
= frame
->tf_rax
;
1158 if (p
->p_sysent
->sv_prepsyscall
) {
1159 (*p
->p_sysent
->sv_prepsyscall
)(
1160 frame
, (int *)(&args
.nosys
.sysmsg
+ 1),
1163 if (code
== SYS_syscall
|| code
== SYS___syscall
) {
1164 code
= frame
->tf_rdi
;
1170 if (p
->p_sysent
->sv_mask
)
1171 code
&= p
->p_sysent
->sv_mask
;
1173 if (code
>= p
->p_sysent
->sv_size
)
1174 callp
= &p
->p_sysent
->sv_table
[0];
1176 callp
= &p
->p_sysent
->sv_table
[code
];
1178 narg
= callp
->sy_narg
& SYF_ARGMASK
;
1181 * On x86_64 we get up to six arguments in registers. The rest are
1182 * on the stack. The first six members of 'struct trapframe' happen
1183 * to be the registers used to pass arguments, in exactly the right
1186 argp
= &frame
->tf_rdi
;
1188 argsdst
= (register_t
*)(&args
.nosys
.sysmsg
+ 1);
1190 * JG can we overflow the space pointed to by 'argsdst'
1191 * either with 'bcopy' or with 'copyin'?
1193 bcopy(argp
, argsdst
, sizeof(register_t
) * regcnt
);
1195 * copyin is MP aware, but the tracing code is not
1197 if (narg
> regcnt
) {
1198 KASSERT(params
!= NULL
, ("copyin args with no params!"));
1199 error
= copyin(params
, &argsdst
[regcnt
],
1200 (narg
- regcnt
) * sizeof(register_t
));
1203 if (KTRPOINT(td
, KTR_SYSCALL
)) {
1204 MAKEMPSAFE(have_mplock
);
1206 ktrsyscall(lp
, code
, narg
,
1207 (void *)(&args
.nosys
.sysmsg
+ 1));
1215 if (KTRPOINT(td
, KTR_SYSCALL
)) {
1216 MAKEMPSAFE(have_mplock
);
1217 ktrsyscall(lp
, code
, narg
, (void *)(&args
.nosys
.sysmsg
+ 1));
1222 * Default return value is 0 (will be copied to %rax). Double-value
1223 * returns use %rax and %rdx. %rdx is left unchanged for system
1224 * calls which return only one result.
1226 args
.sysmsg_fds
[0] = 0;
1227 args
.sysmsg_fds
[1] = frame
->tf_rdx
;
1230 * The syscall might manipulate the trap frame. If it does it
1231 * will probably return EJUSTRETURN.
1233 args
.sysmsg_frame
= frame
;
1235 STOPEVENT(p
, S_SCE
, narg
); /* MP aware */
1238 * NOTE: All system calls run MPSAFE now. The system call itself
1239 * is responsible for getting the MP lock.
1241 error
= (*callp
->sy_call
)(&args
);
1244 kprintf("system call %d returned %d\n", code
, error
);
1249 * MP SAFE (we may or may not have the MP lock at this point)
1254 * Reinitialize proc pointer `p' as it may be different
1255 * if this is a child returning from fork syscall.
1258 lp
= curthread
->td_lwp
;
1259 frame
->tf_rax
= args
.sysmsg_fds
[0];
1260 frame
->tf_rdx
= args
.sysmsg_fds
[1];
1261 frame
->tf_rflags
&= ~PSL_C
;
1265 * Reconstruct pc, we know that 'syscall' is 2 bytes.
1266 * We have to do a full context restore so that %r10
1267 * (which was holding the value of %rcx) is restored for
1268 * the next iteration.
1270 frame
->tf_rip
-= frame
->tf_err
;
1271 frame
->tf_r10
= frame
->tf_rcx
;
1276 panic("Unexpected EASYNC return value (for now)");
1279 if (p
->p_sysent
->sv_errsize
) {
1280 if (error
>= p
->p_sysent
->sv_errsize
)
1281 error
= -1; /* XXX */
1283 error
= p
->p_sysent
->sv_errtbl
[error
];
1285 frame
->tf_rax
= error
;
1286 frame
->tf_rflags
|= PSL_C
;
1291 * Traced syscall. trapsignal() is not MP aware.
1293 if (orig_tf_rflags
& PSL_T
) {
1294 MAKEMPSAFE(have_mplock
);
1295 frame
->tf_rflags
&= ~PSL_T
;
1296 trapsignal(lp
, SIGTRAP
, 0);
1300 * Handle reschedule and other end-of-syscall issues
1302 userret(lp
, frame
, sticks
);
1305 if (KTRPOINT(td
, KTR_SYSRET
)) {
1306 MAKEMPSAFE(have_mplock
);
1307 ktrsysret(lp
, code
, error
, args
.sysmsg_result
);
1312 * This works because errno is findable through the
1313 * register set. If we ever support an emulation where this
1314 * is not the case, this code will need to be revisited.
1316 STOPEVENT(p
, S_SCX
, code
);
1320 * Release the MP lock if we had to get it
1324 KTR_LOG(kernentry_syscall_ret
, lp
->lwp_proc
->p_pid
, lp
->lwp_tid
, error
);
1326 KASSERT(&td
->td_toks_base
== td
->td_toks_stop
,
1327 ("syscall: critical section count mismatch! %d/%d",
1328 crit_count
, td
->td_pri
));
1329 KASSERT(curstop
== td
->td_toks_stop
,
1330 ("syscall: extra tokens held after trap! %ld",
1331 td
->td_toks_stop
- &td
->td_toks_base
));
1336 * NOTE: mplock not held at any point
1339 fork_return(struct lwp
*lp
, struct trapframe
*frame
)
1341 frame
->tf_rax
= 0; /* Child returns zero */
1342 frame
->tf_rflags
&= ~PSL_C
; /* success */
1345 generic_lwp_return(lp
, frame
);
1346 KTR_LOG(kernentry_fork_ret
, lp
->lwp_proc
->p_pid
, lp
->lwp_tid
);
1350 * Simplified back end of syscall(), used when returning from fork()
1351 * directly into user mode.
1353 * This code will return back into the fork trampoline code which then
1356 * NOTE: The mplock is not held at any point.
1359 generic_lwp_return(struct lwp
*lp
, struct trapframe
*frame
)
1361 struct proc
*p
= lp
->lwp_proc
;
1364 * Check for exit-race. If one lwp exits the process concurrent with
1365 * another lwp creating a new thread, the two operations may cross
1366 * each other resulting in the newly-created lwp not receiving a
1369 if (p
->p_flags
& P_WEXIT
) {
1370 lwpsignal(p
, lp
, SIGKILL
);
1374 * Newly forked processes are given a kernel priority. We have to
1375 * adjust the priority to a normal user priority and fake entry
1376 * into the kernel (call userenter()) to install a passive release
1377 * function just in case userret() decides to stop the process. This
1378 * can occur when ^Z races a fork. If we do not install the passive
1379 * release function the current process designation will not be
1380 * released when the thread goes to sleep.
1382 lwkt_setpri_self(TDPRI_USER_NORM
);
1383 userenter(lp
->lwp_thread
, p
);
1384 userret(lp
, frame
, 0);
1386 if (KTRPOINT(lp
->lwp_thread
, KTR_SYSRET
))
1387 ktrsysret(lp
, SYS_fork
, 0, 0);
1389 lp
->lwp_flags
|= LWP_PASSIVE_ACQ
;
1391 lp
->lwp_flags
&= ~LWP_PASSIVE_ACQ
;
1395 * doreti has turned into this. The frame is directly on the stack. We
1396 * pull everything else we need (fpu and tls context) from the current
1399 * Note on fpu interactions: In a virtual kernel, the fpu context for
1400 * an emulated user mode process is not shared with the virtual kernel's
1401 * fpu context, so we only have to 'stack' fpu contexts within the virtual
1402 * kernel itself, and not even then since the signal() contexts that we care
1403 * about save and restore the FPU state (I think anyhow).
1405 * vmspace_ctl() returns an error only if it had problems instaling the
1406 * context we supplied or problems copying data to/from our VM space.
1409 go_user(struct intrframe
*frame
)
1411 struct trapframe
*tf
= (void *)&frame
->if_rdi
;
1416 * Interrupts may be disabled on entry, make sure all signals
1417 * can be received before beginning our loop.
1422 * Switch to the current simulated user process, then call
1423 * user_trap() when we break out of it (usually due to a signal).
1428 * Always make the FPU state correct. This should generally
1429 * be faster because the cost of taking a #NM fault through
1430 * the vkernel to the real kernel is astronomical.
1432 tf
->tf_xflags
&= ~PGEX_FPFAULT
;
1433 if (mdcpu
->gd_npxthread
!= curthread
) {
1434 if (mdcpu
->gd_npxthread
)
1435 npxsave(mdcpu
->gd_npxthread
->td_savefpu
);
1440 * Tell the real kernel whether it is ok to use the FP
1441 * unit or not, allowing us to take a T_DNA exception
1442 * if the context tries to use the FP.
1444 if (mdcpu
->gd_npxthread
== curthread
) {
1445 tf
->tf_xflags
&= ~PGEX_FPFAULT
;
1447 tf
->tf_xflags
|= PGEX_FPFAULT
;
1452 * Run emulated user process context. This call interlocks
1453 * with new mailbox signals.
1455 * Set PGEX_U unconditionally, indicating a user frame (the
1456 * bit is normally set only by T_PAGEFLT).
1459 id
= (void *)vtophys(curproc
->p_vmspace
->vm_pmap
.pm_pml4
);
1461 id
= &curproc
->p_vmspace
->vm_pmap
;
1463 r
= vmspace_ctl(id
, VMSPACE_CTL_RUN
, tf
, &curthread
->td_savevext
);
1465 frame
->if_xflags
|= PGEX_U
;
1467 kprintf("GO USER %d trap %ld EVA %08lx RIP %08lx RSP %08lx XFLAGS %02lx/%02lx\n",
1468 r
, tf
->tf_trapno
, tf
->tf_addr
, tf
->tf_rip
, tf
->tf_rsp
,
1469 tf
->tf_xflags
, frame
->if_xflags
);
1473 panic("vmspace_ctl failed error %d", errno
);
1475 if (tf
->tf_trapno
) {
1479 if (mycpu
->gd_reqflags
& RQF_AST_MASK
) {
1480 tf
->tf_trapno
= T_ASTFLT
;
1488 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1489 * fault (which is then passed back to the virtual kernel) if an attempt is
1490 * made to use the FP unit.
1492 * XXX this is a fairly big hack.
1495 set_vkernel_fp(struct trapframe
*frame
)
1497 struct thread
*td
= curthread
;
1499 if (frame
->tf_xflags
& PGEX_FPFAULT
) {
1500 td
->td_pcb
->pcb_flags
|= FP_VIRTFP
;
1501 if (mdcpu
->gd_npxthread
== td
)
1504 td
->td_pcb
->pcb_flags
&= ~FP_VIRTFP
;
1509 * Called from vkernel_trap() to fixup the vkernel's syscall
1510 * frame for vmspace_ctl() return.
1513 cpu_vkernel_trap(struct trapframe
*frame
, int error
)
1515 frame
->tf_rax
= error
;
1517 frame
->tf_rflags
|= PSL_C
;
1519 frame
->tf_rflags
&= ~PSL_C
;