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
48 #include "opt_ktrace.h"
50 #include <machine/frame.h>
51 #include <sys/param.h>
52 #include <sys/systm.h>
53 #include <sys/kernel.h>
55 #include <sys/pioctl.h>
56 #include <sys/types.h>
57 #include <sys/signal2.h>
58 #include <sys/syscall.h>
59 #include <sys/sysctl.h>
60 #include <sys/sysent.h>
61 #include <sys/systm.h>
63 #include <sys/ktrace.h>
66 #include <sys/sysmsg.h>
67 #include <sys/sysproto.h>
68 #include <sys/sysunion.h>
72 #include <vm/vm_extern.h>
73 #include <vm/vm_kern.h>
74 #include <vm/vm_param.h>
75 #include <machine/cpu.h>
76 #include <machine/pcb.h>
77 #include <machine/smp.h>
78 #include <machine/thread.h>
79 #include <machine/vmparam.h>
80 #include <machine/md_var.h>
81 #include <machine_base/isa/intr_machdep.h>
85 #include <sys/thread2.h>
86 #include <sys/mplock2.h>
90 #define MAKEMPSAFE(have_mplock) \
91 if (have_mplock == 0) { \
98 #define MAKEMPSAFE(have_mplock)
102 extern void trap(struct trapframe
*frame
);
104 static int trap_pfault(struct trapframe
*, int);
105 static void trap_fatal(struct trapframe
*, vm_offset_t
);
106 void dblfault_handler(struct trapframe
*frame
);
108 #define MAX_TRAP_MSG 30
109 static char *trap_msg
[] = {
111 "privileged instruction fault", /* 1 T_PRIVINFLT */
113 "breakpoint instruction fault", /* 3 T_BPTFLT */
116 "arithmetic trap", /* 6 T_ARITHTRAP */
117 "system forced exception", /* 7 T_ASTFLT */
119 "general protection fault", /* 9 T_PROTFLT */
120 "trace trap", /* 10 T_TRCTRAP */
122 "page fault", /* 12 T_PAGEFLT */
124 "alignment fault", /* 14 T_ALIGNFLT */
128 "integer divide fault", /* 18 T_DIVIDE */
129 "non-maskable interrupt trap", /* 19 T_NMI */
130 "overflow trap", /* 20 T_OFLOW */
131 "FPU bounds check fault", /* 21 T_BOUND */
132 "FPU device not available", /* 22 T_DNA */
133 "double fault", /* 23 T_DOUBLEFLT */
134 "FPU operand fetch fault", /* 24 T_FPOPFLT */
135 "invalid TSS fault", /* 25 T_TSSFLT */
136 "segment not present fault", /* 26 T_SEGNPFLT */
137 "stack fault", /* 27 T_STKFLT */
138 "machine check trap", /* 28 T_MCHK */
139 "SIMD floating-point exception", /* 29 T_XMMFLT */
140 "reserved (unknown) fault", /* 30 T_RESERVED */
144 static int ddb_on_nmi
= 1;
145 SYSCTL_INT(_machdep
, OID_AUTO
, ddb_on_nmi
, CTLFLAG_RW
,
146 &ddb_on_nmi
, 0, "Go to DDB on NMI");
148 static int panic_on_nmi
= 1;
149 SYSCTL_INT(_machdep
, OID_AUTO
, panic_on_nmi
, CTLFLAG_RW
,
150 &panic_on_nmi
, 0, "Panic on NMI");
151 static int fast_release
;
152 SYSCTL_INT(_machdep
, OID_AUTO
, fast_release
, CTLFLAG_RW
,
153 &fast_release
, 0, "Passive Release was optimal");
154 static int slow_release
;
155 SYSCTL_INT(_machdep
, OID_AUTO
, slow_release
, CTLFLAG_RW
,
156 &slow_release
, 0, "Passive Release was nonoptimal");
159 * Passively intercepts the thread switch function to increase
160 * the thread priority from a user priority to a kernel priority, reducing
161 * syscall and trap overhead for the case where no switch occurs.
163 * Synchronizes td_ucred with p_ucred. This is used by system calls,
164 * signal handling, faults, AST traps, and anything else that enters the
165 * kernel from userland and provides the kernel with a stable read-only
166 * copy of the process ucred.
169 userenter(struct thread
*curtd
, struct proc
*curp
)
174 curtd
->td_release
= lwkt_passive_release
;
176 if (curtd
->td_ucred
!= curp
->p_ucred
) {
177 ncred
= crhold(curp
->p_ucred
);
178 ocred
= curtd
->td_ucred
;
179 curtd
->td_ucred
= ncred
;
186 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
187 * must be completed before we can return to or try to return to userland.
189 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
190 * arithmatic on the delta calculation so the absolute tick values are
191 * truncated to an integer.
194 userret(struct lwp
*lp
, struct trapframe
*frame
, int sticks
)
196 struct proc
*p
= lp
->lwp_proc
;
200 * Charge system time if profiling. Note: times are in microseconds.
201 * This may do a copyout and block, so do it first even though it
202 * means some system time will be charged as user time.
204 if (p
->p_flag
& P_PROFIL
) {
205 addupc_task(p
, frame
->tf_rip
,
206 (u_int
)((int)lp
->lwp_thread
->td_sticks
- sticks
));
211 * If the jungle wants us dead, so be it.
213 if (lp
->lwp_flag
& LWP_WEXIT
) {
216 rel_mplock(); /* NOT REACHED */
220 * Block here if we are in a stopped state.
222 if (p
->p_stat
== SSTOP
) {
230 * Post any pending upcalls. If running a virtual kernel be sure
231 * to restore the virtual kernel's vmspace before posting the upcall.
233 if (p
->p_flag
& P_UPCALLPEND
) {
234 p
->p_flag
&= ~P_UPCALLPEND
;
242 * Post any pending signals. If running a virtual kernel be sure
243 * to restore the virtual kernel's vmspace before posting the signal.
245 if ((sig
= CURSIG_TRACE(lp
)) != 0) {
253 * block here if we are swapped out, but still process signals
254 * (such as SIGKILL). proc0 (the swapin scheduler) is already
255 * aware of our situation, we do not have to wake it up.
257 if (p
->p_flag
& P_SWAPPEDOUT
) {
259 p
->p_flag
|= P_SWAPWAIT
;
261 if (p
->p_flag
& P_SWAPWAIT
)
262 tsleep(p
, PCATCH
, "SWOUT", 0);
263 p
->p_flag
&= ~P_SWAPWAIT
;
269 * Make sure postsig() handled request to restore old signal mask after
270 * running signal handler.
272 KKASSERT((lp
->lwp_flag
& LWP_OLDMASK
) == 0);
276 * Cleanup from userenter and any passive release that might have occured.
277 * We must reclaim the current-process designation before we can return
278 * to usermode. We also handle both LWKT and USER reschedule requests.
281 userexit(struct lwp
*lp
)
283 struct thread
*td
= lp
->lwp_thread
;
284 /* globaldata_t gd = td->td_gd;*/
287 * Handle stop requests at kernel priority. Any requests queued
288 * after this loop will generate another AST.
290 while (lp
->lwp_proc
->p_stat
== SSTOP
) {
297 * Reduce our priority in preparation for a return to userland. If
298 * our passive release function was still in place, our priority was
299 * never raised and does not need to be reduced.
301 lwkt_passive_recover(td
);
304 * Become the current user scheduled process if we aren't already,
305 * and deal with reschedule requests and other factors.
307 lp
->lwp_proc
->p_usched
->acquire_curproc(lp
);
308 /* WARNING: we may have migrated cpu's */
309 /* gd = td->td_gd; */
312 #if !defined(KTR_KERNENTRY)
313 #define KTR_KERNENTRY KTR_ALL
315 KTR_INFO_MASTER(kernentry
);
316 KTR_INFO(KTR_KERNENTRY
, kernentry
, trap
, 0, "STR",
317 sizeof(long) + sizeof(long) + sizeof(long) + sizeof(vm_offset_t
));
318 KTR_INFO(KTR_KERNENTRY
, kernentry
, trap_ret
, 0, "STR",
319 sizeof(long) + sizeof(long));
320 KTR_INFO(KTR_KERNENTRY
, kernentry
, syscall
, 0, "STR",
321 sizeof(long) + sizeof(long) + sizeof(long));
322 KTR_INFO(KTR_KERNENTRY
, kernentry
, syscall_ret
, 0, "STR",
323 sizeof(long) + sizeof(long) + sizeof(long));
324 KTR_INFO(KTR_KERNENTRY
, kernentry
, fork_ret
, 0, "STR",
325 sizeof(long) + sizeof(long));
328 * Exception, fault, and trap interface to the kernel.
329 * This common code is called from assembly language IDT gate entry
330 * routines that prepare a suitable stack frame, and restore this
331 * frame after the exception has been processed.
333 * This function is also called from doreti in an interlock to handle ASTs.
334 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
336 * NOTE! We have to retrieve the fault address prior to obtaining the
337 * MP lock because get_mplock() may switch out. YYY cr2 really ought
338 * to be retrieved by the assembly code, not here.
340 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
341 * if an attempt is made to switch from a fast interrupt or IPI. This is
342 * necessary to properly take fatal kernel traps on SMP machines if
343 * get_mplock() has to block.
347 trap(struct trapframe
*frame
)
349 struct globaldata
*gd
= mycpu
;
350 struct thread
*td
= gd
->gd_curthread
;
351 struct lwp
*lp
= td
->td_lwp
;
354 int i
= 0, ucode
= 0, type
, code
;
359 int crit_count
= td
->td_critcount
;
360 lwkt_tokref_t curstop
= td
->td_toks_stop
;
368 kprintf0("\"%s\" type=%ld\n",
369 trap_msg
[frame
->tf_trapno
], frame
->tf_trapno
);
370 kprintf0(" rip=%lx rsp=%lx\n", frame
->tf_rip
, frame
->tf_rsp
);
371 kprintf0(" err=%lx addr=%lx\n", frame
->tf_err
, frame
->tf_addr
);
372 kprintf0(" cs=%lx ss=%lx rflags=%lx\n", (unsigned long)frame
->tf_cs
, (unsigned long)frame
->tf_ss
, frame
->tf_rflags
);
377 ++gd
->gd_trap_nesting_level
;
378 MAKEMPSAFE(have_mplock
);
379 trap_fatal(frame
, frame
->tf_addr
);
380 --gd
->gd_trap_nesting_level
;
386 eva
= (frame
->tf_trapno
== T_PAGEFLT
? frame
->tf_addr
: 0);
387 ++gd
->gd_trap_nesting_level
;
388 MAKEMPSAFE(have_mplock
);
389 trap_fatal(frame
, eva
);
390 --gd
->gd_trap_nesting_level
;
397 if ((frame
->tf_rflags
& PSL_I
) == 0) {
399 * Buggy application or kernel code has disabled interrupts
400 * and then trapped. Enabling interrupts now is wrong, but
401 * it is better than running with interrupts disabled until
402 * they are accidentally enabled later.
404 type
= frame
->tf_trapno
;
405 if (ISPL(frame
->tf_cs
) == SEL_UPL
) {
406 MAKEMPSAFE(have_mplock
);
407 /* JG curproc can be NULL */
409 "pid %ld (%s): trap %d with interrupts disabled\n",
410 (long)curproc
->p_pid
, curproc
->p_comm
, type
);
411 } else if (type
!= T_NMI
&& type
!= T_BPTFLT
&&
414 * XXX not quite right, since this may be for a
415 * multiple fault in user mode.
417 MAKEMPSAFE(have_mplock
);
418 kprintf("kernel trap %d with interrupts disabled\n",
424 type
= frame
->tf_trapno
;
425 code
= frame
->tf_err
;
427 if (ISPL(frame
->tf_cs
) == SEL_UPL
) {
430 KTR_LOG(kernentry_trap
, p
->p_pid
, lp
->lwp_tid
,
431 frame
->tf_trapno
, eva
);
435 sticks
= (int)td
->td_sticks
;
436 lp
->lwp_md
.md_regs
= frame
;
439 case T_PRIVINFLT
: /* privileged instruction fault */
444 case T_BPTFLT
: /* bpt instruction fault */
445 case T_TRCTRAP
: /* trace trap */
446 frame
->tf_rflags
&= ~PSL_T
;
451 case T_ARITHTRAP
: /* arithmetic trap */
464 case T_ASTFLT
: /* Allow process switch */
465 mycpu
->gd_cnt
.v_soft
++;
466 if (mycpu
->gd_reqflags
& RQF_AST_OWEUPC
) {
467 atomic_clear_int_nonlocked(&mycpu
->gd_reqflags
,
469 addupc_task(p
, p
->p_prof
.pr_addr
,
474 case T_PROTFLT
: /* general protection fault */
478 case T_SEGNPFLT
: /* segment not present fault */
482 case T_TSSFLT
: /* invalid TSS fault */
483 case T_DOUBLEFLT
: /* double fault */
488 ucode
= code
+ BUS_SEGM_FAULT
; /* XXX: ???*/
494 case T_PAGEFLT
: /* page fault */
495 MAKEMPSAFE(have_mplock
);
496 i
= trap_pfault(frame
, TRUE
);
497 if (frame
->tf_rip
== 0)
498 kprintf("T_PAGEFLT: Warning %%rip == 0!\n");
513 case T_DIVIDE
: /* integer divide fault */
519 MAKEMPSAFE(have_mplock
);
520 /* machine/parity/power fail/"kitchen sink" faults */
521 if (isa_nmi(code
) == 0) {
524 * NMI can be hooked up to a pushbutton
528 kprintf ("NMI ... going to debugger\n");
529 kdb_trap(type
, 0, frame
);
533 } else if (panic_on_nmi
)
534 panic("NMI indicates hardware failure");
537 case T_OFLOW
: /* integer overflow fault */
542 case T_BOUND
: /* bounds check fault */
549 * Virtual kernel intercept - pass the DNA exception
550 * to the virtual kernel if it asked to handle it.
551 * This occurs when the virtual kernel is holding
552 * onto the FP context for a different emulated
553 * process then the one currently running.
555 * We must still call npxdna() since we may have
556 * saved FP state that the virtual kernel needs
557 * to hand over to a different emulated process.
559 if (lp
->lwp_vkernel
&& lp
->lwp_vkernel
->ve
&&
560 (td
->td_pcb
->pcb_flags
& FP_VIRTFP
)
567 * The kernel may have switched out the FP unit's
568 * state, causing the user process to take a fault
569 * when it tries to use the FP unit. Restore the
575 ucode
= FPE_FPU_NP_TRAP
;
578 case T_FPOPFLT
: /* FPU operand fetch fault */
583 case T_XMMFLT
: /* SIMD floating-point exception */
592 case T_PAGEFLT
: /* page fault */
593 MAKEMPSAFE(have_mplock
);
594 trap_pfault(frame
, FALSE
);
599 * The kernel is apparently using fpu for copying.
600 * XXX this should be fatal unless the kernel has
601 * registered such use.
607 case T_STKFLT
: /* stack fault */
610 case T_PROTFLT
: /* general protection fault */
611 case T_SEGNPFLT
: /* segment not present fault */
613 * Invalid segment selectors and out of bounds
614 * %rip's and %rsp's can be set up in user mode.
615 * This causes a fault in kernel mode when the
616 * kernel tries to return to user mode. We want
617 * to get this fault so that we can fix the
618 * problem here and not have to check all the
619 * selectors and pointers when the user changes
622 kprintf("trap.c line %d\n", __LINE__
);
623 if (mycpu
->gd_intr_nesting_level
== 0) {
624 if (td
->td_pcb
->pcb_onfault
) {
625 frame
->tf_rip
= (register_t
)
626 td
->td_pcb
->pcb_onfault
;
629 if (frame
->tf_rip
== (long)doreti_iret
) {
630 frame
->tf_rip
= (long)doreti_iret_fault
;
638 * PSL_NT can be set in user mode and isn't cleared
639 * automatically when the kernel is entered. This
640 * causes a TSS fault when the kernel attempts to
641 * `iret' because the TSS link is uninitialized. We
642 * want to get this fault so that we can fix the
643 * problem here and not every time the kernel is
646 if (frame
->tf_rflags
& PSL_NT
) {
647 frame
->tf_rflags
&= ~PSL_NT
;
652 case T_TRCTRAP
: /* trace trap */
654 if (frame
->tf_rip
== (int)IDTVEC(syscall
)) {
656 * We've just entered system mode via the
657 * syscall lcall. Continue single stepping
658 * silently until the syscall handler has
663 if (frame
->tf_rip
== (int)IDTVEC(syscall
) + 1) {
665 * The syscall handler has now saved the
666 * flags. Stop single stepping it.
668 frame
->tf_rflags
&= ~PSL_T
;
674 * Ignore debug register trace traps due to
675 * accesses in the user's address space, which
676 * can happen under several conditions such as
677 * if a user sets a watchpoint on a buffer and
678 * then passes that buffer to a system call.
679 * We still want to get TRCTRAPS for addresses
680 * in kernel space because that is useful when
681 * debugging the kernel.
684 if (user_dbreg_trap()) {
686 * Reset breakpoint bits because the
689 /* XXX check upper bits here */
690 load_dr6(rdr6() & 0xfffffff0);
695 * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
699 * If DDB is enabled, let it handle the debugger trap.
700 * Otherwise, debugger traps "can't happen".
704 MAKEMPSAFE(have_mplock
);
705 if (kdb_trap(type
, 0, frame
))
711 MAKEMPSAFE(have_mplock
);
712 /* machine/parity/power fail/"kitchen sink" faults */
714 if (isa_nmi(code
) == 0) {
717 * NMI can be hooked up to a pushbutton
721 kprintf ("NMI ... going to debugger\n");
722 kdb_trap(type
, 0, frame
);
726 } else if (panic_on_nmi
== 0)
729 #endif /* NISA > 0 */
731 MAKEMPSAFE(have_mplock
);
732 trap_fatal(frame
, 0);
737 * Virtual kernel intercept - if the fault is directly related to a
738 * VM context managed by a virtual kernel then let the virtual kernel
741 if (lp
->lwp_vkernel
&& lp
->lwp_vkernel
->ve
) {
742 vkernel_trap(lp
, frame
);
747 * Translate fault for emulators (e.g. Linux)
749 if (*p
->p_sysent
->sv_transtrap
)
750 i
= (*p
->p_sysent
->sv_transtrap
)(i
, type
);
752 MAKEMPSAFE(have_mplock
);
753 trapsignal(lp
, i
, ucode
);
756 if (type
<= MAX_TRAP_MSG
) {
757 uprintf("fatal process exception: %s",
759 if ((type
== T_PAGEFLT
) || (type
== T_PROTFLT
))
760 uprintf(", fault VA = 0x%lx", frame
->tf_addr
);
767 if (ISPL(frame
->tf_cs
) == SEL_UPL
) {
768 KASSERT(td
->td_mpcount
== have_mplock
,
769 ("badmpcount trap/end from %p", (void *)frame
->tf_rip
));
772 userret(lp
, frame
, sticks
);
779 if (p
!= NULL
&& lp
!= NULL
)
780 KTR_LOG(kernentry_trap_ret
, p
->p_pid
, lp
->lwp_tid
);
782 KASSERT(crit_count
== td
->td_critcount
,
783 ("trap: critical section count mismatch! %d/%d",
784 crit_count
, td
->td_pri
));
785 KASSERT(curstop
== td
->td_toks_stop
,
786 ("trap: extra tokens held after trap! %ld/%ld",
787 curstop
- &td
->td_toks_base
,
788 td
->td_toks_stop
- &td
->td_toks_base
));
793 trap_pfault(struct trapframe
*frame
, int usermode
)
796 struct vmspace
*vm
= NULL
;
801 thread_t td
= curthread
;
802 struct lwp
*lp
= td
->td_lwp
;
804 va
= trunc_page(frame
->tf_addr
);
805 if (va
>= VM_MIN_KERNEL_ADDRESS
) {
807 * Don't allow user-mode faults in kernel address space.
815 * This is a fault on non-kernel virtual memory.
816 * vm is initialized above to NULL. If curproc is NULL
817 * or curproc->p_vmspace is NULL the fault is fatal.
820 vm
= lp
->lwp_vmspace
;
829 * PGEX_I is defined only if the execute disable bit capability is
830 * supported and enabled.
832 if (frame
->tf_err
& PGEX_W
)
833 ftype
= VM_PROT_WRITE
;
835 else if ((frame
->tf_err
& PGEX_I
) && pg_nx
!= 0)
836 ftype
= VM_PROT_EXECUTE
;
839 ftype
= VM_PROT_READ
;
841 if (map
!= &kernel_map
) {
843 * Keep swapout from messing with us during this
853 fault_flags
|= VM_FAULT_BURST
;
854 if (ftype
& VM_PROT_WRITE
)
855 fault_flags
|= VM_FAULT_DIRTY
;
857 fault_flags
|= VM_FAULT_NORMAL
;
858 rv
= vm_fault(map
, va
, ftype
, fault_flags
);
863 * Don't have to worry about process locking or stacks
866 rv
= vm_fault(map
, va
, ftype
, VM_FAULT_NORMAL
);
869 if (rv
== KERN_SUCCESS
)
873 if (td
->td_gd
->gd_intr_nesting_level
== 0 &&
874 td
->td_pcb
->pcb_onfault
) {
875 frame
->tf_rip
= (register_t
)td
->td_pcb
->pcb_onfault
;
878 trap_fatal(frame
, frame
->tf_addr
);
883 * NOTE: on x86_64 we have a tf_addr field in the trapframe, no
884 * kludge is needed to pass the fault address to signal handlers.
886 struct proc
*p
= td
->td_proc
;
887 if (td
->td_lwp
->lwp_vkernel
== NULL
) {
888 kprintf("seg-fault accessing address %p rip=%p pid=%d p_comm=%s\n",
889 (void *)va
, (void *)frame
->tf_rip
, p
->p_pid
, p
->p_comm
);
891 /* Debugger("seg-fault"); */
893 return((rv
== KERN_PROTECTION_FAILURE
) ? SIGBUS
: SIGSEGV
);
897 trap_fatal(struct trapframe
*frame
, vm_offset_t eva
)
902 struct soft_segment_descriptor softseg
;
905 code
= frame
->tf_err
;
906 type
= frame
->tf_trapno
;
907 sdtossd(&gdt
[IDXSEL(frame
->tf_cs
& 0xffff)], &softseg
);
909 if (type
<= MAX_TRAP_MSG
)
910 msg
= trap_msg
[type
];
913 kprintf("\n\nFatal trap %d: %s while in %s mode\n", type
, msg
,
914 ISPL(frame
->tf_cs
) == SEL_UPL
? "user" : "kernel");
916 /* three separate prints in case of a trap on an unmapped page */
917 kprintf("mp_lock = %08x; ", mp_lock
);
918 kprintf("cpuid = %d; ", mycpu
->gd_cpuid
);
919 kprintf("lapic->id = %08x\n", lapic
->id
);
921 if (type
== T_PAGEFLT
) {
922 kprintf("fault virtual address = 0x%lx\n", eva
);
923 kprintf("fault code = %s %s %s, %s\n",
924 code
& PGEX_U
? "user" : "supervisor",
925 code
& PGEX_W
? "write" : "read",
926 code
& PGEX_I
? "instruction" : "data",
927 code
& PGEX_P
? "protection violation" : "page not present");
929 kprintf("instruction pointer = 0x%lx:0x%lx\n",
930 frame
->tf_cs
& 0xffff, frame
->tf_rip
);
931 if (ISPL(frame
->tf_cs
) == SEL_UPL
) {
932 ss
= frame
->tf_ss
& 0xffff;
935 ss
= GSEL(GDATA_SEL
, SEL_KPL
);
936 rsp
= (long)&frame
->tf_rsp
;
938 kprintf("stack pointer = 0x%x:0x%lx\n", ss
, rsp
);
939 kprintf("frame pointer = 0x%x:0x%lx\n", ss
, frame
->tf_rbp
);
940 kprintf("code segment = base 0x%lx, limit 0x%lx, type 0x%x\n",
941 softseg
.ssd_base
, softseg
.ssd_limit
, softseg
.ssd_type
);
942 kprintf(" = DPL %d, pres %d, long %d, def32 %d, gran %d\n",
943 softseg
.ssd_dpl
, softseg
.ssd_p
, softseg
.ssd_long
, softseg
.ssd_def32
,
945 kprintf("processor eflags = ");
946 if (frame
->tf_rflags
& PSL_T
)
947 kprintf("trace trap, ");
948 if (frame
->tf_rflags
& PSL_I
)
949 kprintf("interrupt enabled, ");
950 if (frame
->tf_rflags
& PSL_NT
)
951 kprintf("nested task, ");
952 if (frame
->tf_rflags
& PSL_RF
)
954 kprintf("IOPL = %ld\n", (frame
->tf_rflags
& PSL_IOPL
) >> 12);
955 kprintf("current process = ");
958 (u_long
)curproc
->p_pid
);
962 kprintf("current thread = pri %d ", curthread
->td_pri
);
963 if (curthread
->td_critcount
)
968 if ((debugger_on_panic
|| db_active
) && kdb_trap(type
, code
, frame
))
971 kprintf("trap number = %d\n", type
);
972 if (type
<= MAX_TRAP_MSG
)
973 panic("%s", trap_msg
[type
]);
975 panic("unknown/reserved trap");
979 * Double fault handler. Called when a fault occurs while writing
980 * a frame for a trap/exception onto the stack. This usually occurs
981 * when the stack overflows (such is the case with infinite recursion,
985 dblfault_handler(struct trapframe
*frame
)
987 kprintf0("DOUBLE FAULT\n");
988 kprintf("\nFatal double fault\n");
989 kprintf("rip = 0x%lx\n", frame
->tf_rip
);
990 kprintf("rsp = 0x%lx\n", frame
->tf_rsp
);
991 kprintf("rbp = 0x%lx\n", frame
->tf_rbp
);
993 /* three separate prints in case of a trap on an unmapped page */
994 kprintf("mp_lock = %08x; ", mp_lock
);
995 kprintf("cpuid = %d; ", mycpu
->gd_cpuid
);
996 kprintf("lapic->id = %08x\n", lapic
->id
);
998 panic("double fault");
1002 * syscall2 - MP aware system call request C handler
1004 * A system call is essentially treated as a trap except that the
1005 * MP lock is not held on entry or return. We are responsible for
1006 * obtaining the MP lock if necessary and for handling ASTs
1007 * (e.g. a task switch) prior to return.
1012 syscall2(struct trapframe
*frame
)
1014 struct thread
*td
= curthread
;
1015 struct proc
*p
= td
->td_proc
;
1016 struct lwp
*lp
= td
->td_lwp
;
1018 struct sysent
*callp
;
1019 register_t orig_tf_rflags
;
1024 int crit_count
= td
->td_critcount
;
1027 int have_mplock
= 0;
1032 union sysunion args
;
1033 register_t
*argsdst
;
1035 mycpu
->gd_cnt
.v_syscall
++;
1038 if (ISPL(frame
->tf_cs
) != SEL_UPL
) {
1045 KTR_LOG(kernentry_syscall
, p
->p_pid
, lp
->lwp_tid
,
1049 KASSERT(td
->td_mpcount
== 0,
1050 ("badmpcount syscall2 from %p", (void *)frame
->tf_rip
));
1052 userenter(td
, p
); /* lazy raise our priority */
1059 sticks
= (int)td
->td_sticks
;
1060 orig_tf_rflags
= frame
->tf_rflags
;
1063 * Virtual kernel intercept - if a VM context managed by a virtual
1064 * kernel issues a system call the virtual kernel handles it, not us.
1065 * Restore the virtual kernel context and return from its system
1066 * call. The current frame is copied out to the virtual kernel.
1068 if (lp
->lwp_vkernel
&& lp
->lwp_vkernel
->ve
) {
1069 vkernel_trap(lp
, frame
);
1070 error
= EJUSTRETURN
;
1075 * Get the system call parameters and account for time
1077 lp
->lwp_md
.md_regs
= frame
;
1078 params
= (caddr_t
)frame
->tf_rsp
+ sizeof(register_t
);
1079 code
= frame
->tf_rax
;
1081 if (p
->p_sysent
->sv_prepsyscall
) {
1082 (*p
->p_sysent
->sv_prepsyscall
)(
1083 frame
, (int *)(&args
.nosys
.sysmsg
+ 1),
1086 if (code
== SYS_syscall
|| code
== SYS___syscall
) {
1087 code
= frame
->tf_rdi
;
1093 if (p
->p_sysent
->sv_mask
)
1094 code
&= p
->p_sysent
->sv_mask
;
1096 if (code
>= p
->p_sysent
->sv_size
)
1097 callp
= &p
->p_sysent
->sv_table
[0];
1099 callp
= &p
->p_sysent
->sv_table
[code
];
1101 narg
= callp
->sy_narg
& SYF_ARGMASK
;
1104 * On x86_64 we get up to six arguments in registers. The rest are
1105 * on the stack. The first six members of 'struct trapframe' happen
1106 * to be the registers used to pass arguments, in exactly the right
1109 argp
= &frame
->tf_rdi
;
1111 argsdst
= (register_t
*)(&args
.nosys
.sysmsg
+ 1);
1113 * JG can we overflow the space pointed to by 'argsdst'
1114 * either with 'bcopy' or with 'copyin'?
1116 bcopy(argp
, argsdst
, sizeof(register_t
) * regcnt
);
1118 * copyin is MP aware, but the tracing code is not
1120 if (narg
> regcnt
) {
1121 KASSERT(params
!= NULL
, ("copyin args with no params!"));
1122 error
= copyin(params
, &argsdst
[regcnt
],
1123 (narg
- regcnt
) * sizeof(register_t
));
1126 if (KTRPOINT(td
, KTR_SYSCALL
)) {
1127 MAKEMPSAFE(have_mplock
);
1129 ktrsyscall(lp
, code
, narg
,
1130 (void *)(&args
.nosys
.sysmsg
+ 1));
1138 if (KTRPOINT(td
, KTR_SYSCALL
)) {
1139 MAKEMPSAFE(have_mplock
);
1140 ktrsyscall(lp
, code
, narg
, (void *)(&args
.nosys
.sysmsg
+ 1));
1145 * Default return value is 0 (will be copied to %rax). Double-value
1146 * returns use %rax and %rdx. %rdx is left unchanged for system
1147 * calls which return only one result.
1149 args
.sysmsg_fds
[0] = 0;
1150 args
.sysmsg_fds
[1] = frame
->tf_rdx
;
1153 * The syscall might manipulate the trap frame. If it does it
1154 * will probably return EJUSTRETURN.
1156 args
.sysmsg_frame
= frame
;
1158 STOPEVENT(p
, S_SCE
, narg
); /* MP aware */
1161 * NOTE: All system calls run MPSAFE now. The system call itself
1162 * is responsible for getting the MP lock.
1164 error
= (*callp
->sy_call
)(&args
);
1168 * MP SAFE (we may or may not have the MP lock at this point)
1170 //kprintf("SYSMSG %d ", error);
1174 * Reinitialize proc pointer `p' as it may be different
1175 * if this is a child returning from fork syscall.
1178 lp
= curthread
->td_lwp
;
1179 frame
->tf_rax
= args
.sysmsg_fds
[0];
1180 frame
->tf_rdx
= args
.sysmsg_fds
[1];
1181 frame
->tf_rflags
&= ~PSL_C
;
1185 * Reconstruct pc, we know that 'syscall' is 2 bytes.
1186 * We have to do a full context restore so that %r10
1187 * (which was holding the value of %rcx) is restored for
1188 * the next iteration.
1190 frame
->tf_rip
-= frame
->tf_err
;
1191 frame
->tf_r10
= frame
->tf_rcx
;
1196 panic("Unexpected EASYNC return value (for now)");
1199 if (p
->p_sysent
->sv_errsize
) {
1200 if (error
>= p
->p_sysent
->sv_errsize
)
1201 error
= -1; /* XXX */
1203 error
= p
->p_sysent
->sv_errtbl
[error
];
1205 frame
->tf_rax
= error
;
1206 frame
->tf_rflags
|= PSL_C
;
1211 * Traced syscall. trapsignal() is not MP aware.
1213 if (orig_tf_rflags
& PSL_T
) {
1214 MAKEMPSAFE(have_mplock
);
1215 frame
->tf_rflags
&= ~PSL_T
;
1216 trapsignal(lp
, SIGTRAP
, TRAP_TRACE
);
1220 * Handle reschedule and other end-of-syscall issues
1222 userret(lp
, frame
, sticks
);
1225 if (KTRPOINT(td
, KTR_SYSRET
)) {
1226 MAKEMPSAFE(have_mplock
);
1227 ktrsysret(lp
, code
, error
, args
.sysmsg_result
);
1232 * This works because errno is findable through the
1233 * register set. If we ever support an emulation where this
1234 * is not the case, this code will need to be revisited.
1236 STOPEVENT(p
, S_SCX
, code
);
1241 * Release the MP lock if we had to get it
1243 KASSERT(td
->td_mpcount
== have_mplock
,
1244 ("badmpcount syscall2/end from %p", (void *)frame
->tf_rip
));
1248 KTR_LOG(kernentry_syscall_ret
, p
->p_pid
, lp
->lwp_tid
, error
);
1250 KASSERT(crit_count
== td
->td_critcount
,
1251 ("syscall: critical section count mismatch! %d/%d",
1252 crit_count
, td
->td_pri
));
1253 KASSERT(&td
->td_toks_base
== td
->td_toks_stop
,
1254 ("syscall: extra tokens held after trap! %ld",
1255 td
->td_toks_stop
- &td
->td_toks_base
));
1260 * NOTE: mplock not held at any point
1263 fork_return(struct lwp
*lp
, struct trapframe
*frame
)
1265 frame
->tf_rax
= 0; /* Child returns zero */
1266 frame
->tf_rflags
&= ~PSL_C
; /* success */
1269 generic_lwp_return(lp
, frame
);
1270 KTR_LOG(kernentry_fork_ret
, lp
->lwp_proc
->p_pid
, lp
->lwp_tid
);
1274 * Simplified back end of syscall(), used when returning from fork()
1275 * directly into user mode.
1277 * This code will return back into the fork trampoline code which then
1280 * NOTE: The mplock is not held at any point.
1283 generic_lwp_return(struct lwp
*lp
, struct trapframe
*frame
)
1285 struct proc
*p
= lp
->lwp_proc
;
1288 * Newly forked processes are given a kernel priority. We have to
1289 * adjust the priority to a normal user priority and fake entry
1290 * into the kernel (call userenter()) to install a passive release
1291 * function just in case userret() decides to stop the process. This
1292 * can occur when ^Z races a fork. If we do not install the passive
1293 * release function the current process designation will not be
1294 * released when the thread goes to sleep.
1296 lwkt_setpri_self(TDPRI_USER_NORM
);
1297 userenter(lp
->lwp_thread
, p
);
1298 userret(lp
, frame
, 0);
1300 if (KTRPOINT(lp
->lwp_thread
, KTR_SYSRET
))
1301 ktrsysret(lp
, SYS_fork
, 0, 0);
1303 p
->p_flag
|= P_PASSIVE_ACQ
;
1305 p
->p_flag
&= ~P_PASSIVE_ACQ
;
1309 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1310 * fault (which is then passed back to the virtual kernel) if an attempt is
1311 * made to use the FP unit.
1313 * XXX this is a fairly big hack.
1316 set_vkernel_fp(struct trapframe
*frame
)
1318 struct thread
*td
= curthread
;
1320 if (frame
->tf_xflags
& PGEX_FPFAULT
) {
1321 td
->td_pcb
->pcb_flags
|= FP_VIRTFP
;
1322 if (mdcpu
->gd_npxthread
== td
)
1325 td
->td_pcb
->pcb_flags
&= ~FP_VIRTFP
;
1330 * Called from vkernel_trap() to fixup the vkernel's syscall
1331 * frame for vmspace_ctl() return.
1334 cpu_vkernel_trap(struct trapframe
*frame
, int error
)
1336 frame
->tf_rax
= error
;
1338 frame
->tf_rflags
|= PSL_C
;
1340 frame
->tf_rflags
&= ~PSL_C
;