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 $
41 * $DragonFly: src/sys/platform/pc64/amd64/trap.c,v 1.3 2008/09/09 04:06:18 dillon Exp $
45 * AMD64 Trap and System call handling
49 #include "opt_ktrace.h"
51 #include <machine/frame.h>
52 #include <sys/param.h>
53 #include <sys/systm.h>
54 #include <sys/kernel.h>
56 #include <sys/pioctl.h>
57 #include <sys/types.h>
58 #include <sys/signal2.h>
59 #include <sys/syscall.h>
60 #include <sys/sysctl.h>
61 #include <sys/sysent.h>
62 #include <sys/systm.h>
64 #include <sys/ktrace.h>
67 #include <sys/sysmsg.h>
68 #include <sys/sysproto.h>
69 #include <sys/sysunion.h>
73 #include <vm/vm_extern.h>
74 #include <vm/vm_kern.h>
75 #include <vm/vm_param.h>
76 #include <machine/cpu.h>
77 #include <machine/pcb.h>
78 #include <machine/thread.h>
79 #include <machine/vmparam.h>
80 #include <machine/md_var.h>
83 #include <sys/thread2.h>
87 #define MAKEMPSAFE(have_mplock) \
88 if (have_mplock == 0) { \
95 #define MAKEMPSAFE(have_mplock)
99 extern void trap(struct trapframe
*frame
);
101 static int trap_pfault(struct trapframe
*, int);
102 static void trap_fatal(struct trapframe
*, vm_offset_t
);
103 void dblfault_handler(struct trapframe
*frame
);
105 #define PCPU_GET(member) ((mycpu)->gd_##member)
106 #define PCPU_INC(member) ((mycpu)->gd_##member)++
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");
158 static int syscall_mpsafe
= 1;
159 SYSCTL_INT(_kern
, OID_AUTO
, syscall_mpsafe
, CTLFLAG_RW
,
160 &syscall_mpsafe
, 0, "Allow MPSAFE marked syscalls to run without BGL");
161 TUNABLE_INT("kern.syscall_mpsafe", &syscall_mpsafe
);
162 static int trap_mpsafe
= 1;
163 SYSCTL_INT(_kern
, OID_AUTO
, trap_mpsafe
, CTLFLAG_RW
,
164 &trap_mpsafe
, 0, "Allow traps to mostly run without the BGL");
165 TUNABLE_INT("kern.trap_mpsafe", &trap_mpsafe
);
169 * userenter() passively intercepts the thread switch function to increase
170 * the thread priority from a user priority to a kernel priority, reducing
171 * syscall and trap overhead for the case where no switch occurs.
175 userenter(struct thread
*curtd
)
177 curtd
->td_release
= lwkt_passive_release
;
181 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
182 * must be completed before we can return to or try to return to userland.
184 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
185 * arithmatic on the delta calculation so the absolute tick values are
186 * truncated to an integer.
189 userret(struct lwp
*lp
, struct trapframe
*frame
, int sticks
)
191 struct proc
*p
= lp
->lwp_proc
;
195 * Charge system time if profiling. Note: times are in microseconds.
196 * This may do a copyout and block, so do it first even though it
197 * means some system time will be charged as user time.
199 if (p
->p_flag
& P_PROFIL
) {
200 addupc_task(p
, frame
->tf_rip
,
201 (u_int
)((int)lp
->lwp_thread
->td_sticks
- sticks
));
206 * If the jungle wants us dead, so be it.
208 if (lp
->lwp_flag
& LWP_WEXIT
) {
211 rel_mplock(); /* NOT REACHED */
215 * Block here if we are in a stopped state.
217 if (p
->p_stat
== SSTOP
) {
225 * Post any pending upcalls. If running a virtual kernel be sure
226 * to restore the virtual kernel's vmspace before posting the upcall.
228 if (p
->p_flag
& P_UPCALLPEND
) {
229 p
->p_flag
&= ~P_UPCALLPEND
;
237 * Post any pending signals. If running a virtual kernel be sure
238 * to restore the virtual kernel's vmspace before posting the signal.
240 if ((sig
= CURSIG_TRACE(lp
)) != 0) {
248 * block here if we are swapped out, but still process signals
249 * (such as SIGKILL). proc0 (the swapin scheduler) is already
250 * aware of our situation, we do not have to wake it up.
252 if (p
->p_flag
& P_SWAPPEDOUT
) {
254 p
->p_flag
|= P_SWAPWAIT
;
256 if (p
->p_flag
& P_SWAPWAIT
)
257 tsleep(p
, PCATCH
, "SWOUT", 0);
258 p
->p_flag
&= ~P_SWAPWAIT
;
264 * Make sure postsig() handled request to restore old signal mask after
265 * running signal handler.
267 KKASSERT((lp
->lwp_flag
& LWP_OLDMASK
) == 0);
271 * Cleanup from userenter and any passive release that might have occured.
272 * We must reclaim the current-process designation before we can return
273 * to usermode. We also handle both LWKT and USER reschedule requests.
276 userexit(struct lwp
*lp
)
278 struct thread
*td
= lp
->lwp_thread
;
279 /* globaldata_t gd = td->td_gd;*/
282 * Handle stop requests at kernel priority. Any requests queued
283 * after this loop will generate another AST.
285 while (lp
->lwp_proc
->p_stat
== SSTOP
) {
292 * Reduce our priority in preparation for a return to userland. If
293 * our passive release function was still in place, our priority was
294 * never raised and does not need to be reduced.
296 lwkt_passive_recover(td
);
299 * Become the current user scheduled process if we aren't already,
300 * and deal with reschedule requests and other factors.
302 lp
->lwp_proc
->p_usched
->acquire_curproc(lp
);
303 /* WARNING: we may have migrated cpu's */
304 /* gd = td->td_gd; */
307 #if !defined(KTR_KERNENTRY)
308 #define KTR_KERNENTRY KTR_ALL
310 KTR_INFO_MASTER(kernentry
);
311 KTR_INFO(KTR_KERNENTRY
, kernentry
, trap
, 0, "STR",
312 sizeof(long) + sizeof(long) + sizeof(long) + sizeof(vm_offset_t
));
313 KTR_INFO(KTR_KERNENTRY
, kernentry
, trap_ret
, 0, "STR",
314 sizeof(long) + sizeof(long));
315 KTR_INFO(KTR_KERNENTRY
, kernentry
, syscall
, 0, "STR",
316 sizeof(long) + sizeof(long) + sizeof(long));
317 KTR_INFO(KTR_KERNENTRY
, kernentry
, syscall_ret
, 0, "STR",
318 sizeof(long) + sizeof(long) + sizeof(long));
319 KTR_INFO(KTR_KERNENTRY
, kernentry
, fork_ret
, 0, "STR",
320 sizeof(long) + sizeof(long));
323 * Exception, fault, and trap interface to the kernel.
324 * This common code is called from assembly language IDT gate entry
325 * routines that prepare a suitable stack frame, and restore this
326 * frame after the exception has been processed.
328 * This function is also called from doreti in an interlock to handle ASTs.
329 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
331 * NOTE! We have to retrieve the fault address prior to obtaining the
332 * MP lock because get_mplock() may switch out. YYY cr2 really ought
333 * to be retrieved by the assembly code, not here.
335 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
336 * if an attempt is made to switch from a fast interrupt or IPI. This is
337 * necessary to properly take fatal kernel traps on SMP machines if
338 * get_mplock() has to block.
342 trap(struct trapframe
*frame
)
344 struct globaldata
*gd
= mycpu
;
345 struct thread
*td
= gd
->gd_curthread
;
346 struct lwp
*lp
= td
->td_lwp
;
349 int i
= 0, ucode
= 0, type
, code
;
354 int crit_count
= td
->td_pri
& ~TDPRI_MASK
;
362 kprintf0("\"%s\" type=%ld\n",
363 trap_msg
[frame
->tf_trapno
], frame
->tf_trapno
);
364 kprintf0(" rip=%lx rsp=%lx\n", frame
->tf_rip
, frame
->tf_rsp
);
365 kprintf0(" err=%lx addr=%lx\n", frame
->tf_err
, frame
->tf_addr
);
366 kprintf0(" cs=%lx ss=%lx rflags=%lx\n", (unsigned long)frame
->tf_cs
, (unsigned long)frame
->tf_ss
, frame
->tf_rflags
);
371 ++gd
->gd_trap_nesting_level
;
372 MAKEMPSAFE(have_mplock
);
373 trap_fatal(frame
, frame
->tf_addr
);
374 --gd
->gd_trap_nesting_level
;
380 eva
= (frame
->tf_trapno
== T_PAGEFLT
? frame
->tf_addr
: 0);
381 ++gd
->gd_trap_nesting_level
;
382 MAKEMPSAFE(have_mplock
);
383 trap_fatal(frame
, eva
);
384 --gd
->gd_trap_nesting_level
;
392 if (trap_mpsafe
== 0) {
393 ++gd
->gd_trap_nesting_level
;
394 MAKEMPSAFE(have_mplock
);
395 --gd
->gd_trap_nesting_level
;
399 if ((frame
->tf_rflags
& PSL_I
) == 0) {
401 * Buggy application or kernel code has disabled interrupts
402 * and then trapped. Enabling interrupts now is wrong, but
403 * it is better than running with interrupts disabled until
404 * they are accidentally enabled later.
406 type
= frame
->tf_trapno
;
407 if (ISPL(frame
->tf_cs
) == SEL_UPL
) {
408 MAKEMPSAFE(have_mplock
);
409 /* JG curproc can be NULL */
411 "pid %ld (%s): trap %d with interrupts disabled\n",
412 (long)curproc
->p_pid
, curproc
->p_comm
, type
);
413 } else if (type
!= T_NMI
&& type
!= T_BPTFLT
&&
416 * XXX not quite right, since this may be for a
417 * multiple fault in user mode.
419 MAKEMPSAFE(have_mplock
);
420 kprintf("kernel trap %d with interrupts disabled\n",
426 type
= frame
->tf_trapno
;
427 code
= frame
->tf_err
;
429 if (ISPL(frame
->tf_cs
) == SEL_UPL
) {
432 KTR_LOG(kernentry_trap
, p
->p_pid
, lp
->lwp_tid
,
433 frame
->tf_trapno
, eva
);
437 sticks
= (int)td
->td_sticks
;
438 lp
->lwp_md
.md_regs
= frame
;
441 case T_PRIVINFLT
: /* privileged instruction fault */
446 case T_BPTFLT
: /* bpt instruction fault */
447 case T_TRCTRAP
: /* trace trap */
448 frame
->tf_rflags
&= ~PSL_T
;
452 case T_ARITHTRAP
: /* arithmetic trap */
465 case T_ASTFLT
: /* Allow process switch */
466 mycpu
->gd_cnt
.v_soft
++;
467 if (mycpu
->gd_reqflags
& RQF_AST_OWEUPC
) {
468 atomic_clear_int_nonlocked(&mycpu
->gd_reqflags
,
470 addupc_task(p
, p
->p_prof
.pr_addr
,
475 case T_PROTFLT
: /* general protection fault */
476 case T_SEGNPFLT
: /* segment not present fault */
477 case T_TSSFLT
: /* invalid TSS fault */
478 case T_DOUBLEFLT
: /* double fault */
480 ucode
= code
+ BUS_SEGM_FAULT
;
484 case T_PAGEFLT
: /* page fault */
485 MAKEMPSAFE(have_mplock
);
486 i
= trap_pfault(frame
, TRUE
);
487 if (frame
->tf_rip
== 0)
488 kprintf("T_PAGEFLT: Warning %%rip == 0!\n");
497 case T_DIVIDE
: /* integer divide fault */
503 MAKEMPSAFE(have_mplock
);
504 /* machine/parity/power fail/"kitchen sink" faults */
505 if (isa_nmi(code
) == 0) {
508 * NMI can be hooked up to a pushbutton
512 kprintf ("NMI ... going to debugger\n");
513 kdb_trap(type
, 0, frame
);
517 } else if (panic_on_nmi
)
518 panic("NMI indicates hardware failure");
521 case T_OFLOW
: /* integer overflow fault */
526 case T_BOUND
: /* bounds check fault */
533 * Virtual kernel intercept - pass the DNA exception
534 * to the virtual kernel if it asked to handle it.
535 * This occurs when the virtual kernel is holding
536 * onto the FP context for a different emulated
537 * process then the one currently running.
539 * We must still call npxdna() since we may have
540 * saved FP state that the virtual kernel needs
541 * to hand over to a different emulated process.
543 if (lp
->lwp_vkernel
&& lp
->lwp_vkernel
->ve
&&
544 (td
->td_pcb
->pcb_flags
& FP_VIRTFP
)
551 * The kernel may have switched out the FP unit's
552 * state, causing the user process to take a fault
553 * when it tries to use the FP unit. Restore the
559 ucode
= FPE_FPU_NP_TRAP
;
562 case T_FPOPFLT
: /* FPU operand fetch fault */
567 case T_XMMFLT
: /* SIMD floating-point exception */
576 case T_PAGEFLT
: /* page fault */
577 MAKEMPSAFE(have_mplock
);
578 trap_pfault(frame
, FALSE
);
583 * The kernel is apparently using fpu for copying.
584 * XXX this should be fatal unless the kernel has
585 * registered such use.
591 case T_STKFLT
: /* stack fault */
594 case T_PROTFLT
: /* general protection fault */
595 case T_SEGNPFLT
: /* segment not present fault */
597 * Invalid segment selectors and out of bounds
598 * %rip's and %rsp's can be set up in user mode.
599 * This causes a fault in kernel mode when the
600 * kernel tries to return to user mode. We want
601 * to get this fault so that we can fix the
602 * problem here and not have to check all the
603 * selectors and pointers when the user changes
606 kprintf0("trap.c line %d\n", __LINE__
);
607 if (mycpu
->gd_intr_nesting_level
== 0) {
608 if (td
->td_pcb
->pcb_onfault
) {
609 frame
->tf_rip
= (register_t
)
610 td
->td_pcb
->pcb_onfault
;
613 if (frame
->tf_rip
== (long)doreti_iret
) {
614 frame
->tf_rip
= (long)doreti_iret_fault
;
622 * PSL_NT can be set in user mode and isn't cleared
623 * automatically when the kernel is entered. This
624 * causes a TSS fault when the kernel attempts to
625 * `iret' because the TSS link is uninitialized. We
626 * want to get this fault so that we can fix the
627 * problem here and not every time the kernel is
630 if (frame
->tf_rflags
& PSL_NT
) {
631 frame
->tf_rflags
&= ~PSL_NT
;
636 case T_TRCTRAP
: /* trace trap */
638 if (frame
->tf_rip
== (int)IDTVEC(syscall
)) {
640 * We've just entered system mode via the
641 * syscall lcall. Continue single stepping
642 * silently until the syscall handler has
647 if (frame
->tf_rip
== (int)IDTVEC(syscall
) + 1) {
649 * The syscall handler has now saved the
650 * flags. Stop single stepping it.
652 frame
->tf_rflags
&= ~PSL_T
;
658 * Ignore debug register trace traps due to
659 * accesses in the user's address space, which
660 * can happen under several conditions such as
661 * if a user sets a watchpoint on a buffer and
662 * then passes that buffer to a system call.
663 * We still want to get TRCTRAPS for addresses
664 * in kernel space because that is useful when
665 * debugging the kernel.
668 if (user_dbreg_trap()) {
670 * Reset breakpoint bits because the
673 /* XXX check upper bits here */
674 load_dr6(rdr6() & 0xfffffff0);
679 * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
683 * If DDB is enabled, let it handle the debugger trap.
684 * Otherwise, debugger traps "can't happen".
687 MAKEMPSAFE(have_mplock
);
688 if (kdb_trap(type
, 0, frame
))
694 MAKEMPSAFE(have_mplock
);
695 /* machine/parity/power fail/"kitchen sink" faults */
697 if (isa_nmi(code
) == 0) {
700 * NMI can be hooked up to a pushbutton
704 kprintf ("NMI ... going to debugger\n");
705 kdb_trap(type
, 0, frame
);
709 } else if (panic_on_nmi
== 0)
712 #endif /* NISA > 0 */
714 MAKEMPSAFE(have_mplock
);
715 trap_fatal(frame
, 0);
720 * Virtual kernel intercept - if the fault is directly related to a
721 * VM context managed by a virtual kernel then let the virtual kernel
724 if (lp
->lwp_vkernel
&& lp
->lwp_vkernel
->ve
) {
725 vkernel_trap(lp
, frame
);
730 * Virtual kernel intercept - if the fault is directly related to a
731 * VM context managed by a virtual kernel then let the virtual kernel
734 if (lp
->lwp_vkernel
&& lp
->lwp_vkernel
->ve
) {
735 vkernel_trap(lp
, frame
);
740 * Translate fault for emulators (e.g. Linux)
742 if (*p
->p_sysent
->sv_transtrap
)
743 i
= (*p
->p_sysent
->sv_transtrap
)(i
, type
);
745 MAKEMPSAFE(have_mplock
);
746 trapsignal(lp
, i
, ucode
);
749 if (type
<= MAX_TRAP_MSG
) {
750 uprintf("fatal process exception: %s",
752 if ((type
== T_PAGEFLT
) || (type
== T_PROTFLT
))
753 uprintf(", fault VA = 0x%lx", frame
->tf_addr
);
760 if (ISPL(frame
->tf_cs
) == SEL_UPL
)
761 KASSERT(td
->td_mpcount
== have_mplock
, ("badmpcount trap/end from %p", (void *)frame
->tf_rip
));
763 userret(lp
, frame
, sticks
);
770 if (p
!= NULL
&& lp
!= NULL
)
771 KTR_LOG(kernentry_trap_ret
, p
->p_pid
, lp
->lwp_tid
);
773 KASSERT(crit_count
== (td
->td_pri
& ~TDPRI_MASK
),
774 ("syscall: critical section count mismatch! %d/%d",
775 crit_count
/ TDPRI_CRIT
, td
->td_pri
/ TDPRI_CRIT
));
780 trap_pfault(struct trapframe
*frame
, int usermode
)
783 struct vmspace
*vm
= NULL
;
787 thread_t td
= curthread
;
788 struct lwp
*lp
= td
->td_lwp
;
790 va
= trunc_page(frame
->tf_addr
);
791 if (va
>= VM_MIN_KERNEL_ADDRESS
) {
793 * Don't allow user-mode faults in kernel address space.
801 * This is a fault on non-kernel virtual memory.
802 * vm is initialized above to NULL. If curproc is NULL
803 * or curproc->p_vmspace is NULL the fault is fatal.
806 vm
= lp
->lwp_vmspace
;
815 * PGEX_I is defined only if the execute disable bit capability is
816 * supported and enabled.
818 if (frame
->tf_err
& PGEX_W
)
819 ftype
= VM_PROT_WRITE
;
821 else if ((frame
->tf_err
& PGEX_I
) && pg_nx
!= 0)
822 ftype
= VM_PROT_EXECUTE
;
825 ftype
= VM_PROT_READ
;
827 if (map
!= &kernel_map
) {
829 * Keep swapout from messing with us during this
835 * Grow the stack if necessary
837 /* grow_stack returns false only if va falls into
838 * a growable stack region and the stack growth
839 * fails. It returns true if va was not within
840 * a growable stack region, or if the stack
843 if (!grow_stack(lp
->lwp_proc
, va
)) {
849 /* Fault in the user page: */
850 rv
= vm_fault(map
, va
, ftype
,
851 (ftype
& VM_PROT_WRITE
) ? VM_FAULT_DIRTY
857 * Don't have to worry about process locking or stacks
860 rv
= vm_fault(map
, va
, ftype
, VM_FAULT_NORMAL
);
863 if (rv
== KERN_SUCCESS
)
867 if (td
->td_gd
->gd_intr_nesting_level
== 0 &&
868 td
->td_pcb
->pcb_onfault
) {
869 frame
->tf_rip
= (register_t
)td
->td_pcb
->pcb_onfault
;
872 trap_fatal(frame
, frame
->tf_addr
);
877 * NOTE: on amd64 we have a tf_addr field in the trapframe, no
878 * kludge is needed to pass the fault address to signal handlers.
880 kprintf("seg-fault accessing address %p ip=%p\n",
881 (void *)va
, (void *)frame
->tf_rip
);
882 /* Debugger("seg-fault"); */
884 return((rv
== KERN_PROTECTION_FAILURE
) ? SIGBUS
: SIGSEGV
);
888 trap_fatal(struct trapframe
*frame
, vm_offset_t eva
)
893 struct soft_segment_descriptor softseg
;
896 code
= frame
->tf_err
;
897 type
= frame
->tf_trapno
;
898 sdtossd(&gdt
[IDXSEL(frame
->tf_cs
& 0xffff)], &softseg
);
900 if (type
<= MAX_TRAP_MSG
)
901 msg
= trap_msg
[type
];
904 kprintf("\n\nFatal trap %d: %s while in %s mode\n", type
, msg
,
905 ISPL(frame
->tf_cs
) == SEL_UPL
? "user" : "kernel");
907 /* two separate prints in case of a trap on an unmapped page */
908 kprintf("cpuid = %d; ", PCPU_GET(cpuid
));
909 kprintf("apic id = %02x\n", PCPU_GET(apic_id
));
911 if (type
== T_PAGEFLT
) {
912 kprintf("fault virtual address = 0x%lx\n", eva
);
913 kprintf("fault code = %s %s %s, %s\n",
914 code
& PGEX_U
? "user" : "supervisor",
915 code
& PGEX_W
? "write" : "read",
916 code
& PGEX_I
? "instruction" : "data",
917 code
& PGEX_P
? "protection violation" : "page not present");
919 kprintf("instruction pointer = 0x%lx:0x%lx\n",
920 frame
->tf_cs
& 0xffff, frame
->tf_rip
);
921 if (ISPL(frame
->tf_cs
) == SEL_UPL
) {
922 ss
= frame
->tf_ss
& 0xffff;
925 ss
= GSEL(GDATA_SEL
, SEL_KPL
);
926 rsp
= (long)&frame
->tf_rsp
;
928 kprintf("stack pointer = 0x%x:0x%lx\n", ss
, rsp
);
929 kprintf("frame pointer = 0x%x:0x%lx\n", ss
, frame
->tf_rbp
);
930 kprintf("code segment = base 0x%lx, limit 0x%lx, type 0x%x\n",
931 softseg
.ssd_base
, softseg
.ssd_limit
, softseg
.ssd_type
);
932 kprintf(" = DPL %d, pres %d, long %d, def32 %d, gran %d\n",
933 softseg
.ssd_dpl
, softseg
.ssd_p
, softseg
.ssd_long
, softseg
.ssd_def32
,
935 kprintf("processor eflags = ");
936 if (frame
->tf_rflags
& PSL_T
)
937 kprintf("trace trap, ");
938 if (frame
->tf_rflags
& PSL_I
)
939 kprintf("interrupt enabled, ");
940 if (frame
->tf_rflags
& PSL_NT
)
941 kprintf("nested task, ");
942 if (frame
->tf_rflags
& PSL_RF
)
944 kprintf("IOPL = %ld\n", (frame
->tf_rflags
& PSL_IOPL
) >> 12);
945 kprintf("current process = ");
948 (u_long
)curproc
->p_pid
);
952 kprintf("current thread = pri %d ", curthread
->td_pri
);
953 if (curthread
->td_pri
>= TDPRI_CRIT
)
958 if ((debugger_on_panic
|| db_active
) && kdb_trap(type
, code
, frame
))
961 kprintf("trap number = %d\n", type
);
962 if (type
<= MAX_TRAP_MSG
)
963 panic("%s", trap_msg
[type
]);
965 panic("unknown/reserved trap");
969 * Double fault handler. Called when a fault occurs while writing
970 * a frame for a trap/exception onto the stack. This usually occurs
971 * when the stack overflows (such is the case with infinite recursion,
975 dblfault_handler(struct trapframe
*frame
)
977 kprintf0("DOUBLE FAULT\n");
978 kprintf("\nFatal double fault\n");
979 kprintf("rip = 0x%lx\n", frame
->tf_rip
);
980 kprintf("rsp = 0x%lx\n", frame
->tf_rsp
);
981 kprintf("rbp = 0x%lx\n", frame
->tf_rbp
);
983 /* two separate prints in case of a trap on an unmapped page */
984 kprintf("cpuid = %d; ", PCPU_GET(cpuid
));
985 kprintf("apic id = %02x\n", PCPU_GET(apic_id
));
987 panic("double fault");
991 * syscall2 - MP aware system call request C handler
993 * A system call is essentially treated as a trap except that the
994 * MP lock is not held on entry or return. We are responsible for
995 * obtaining the MP lock if necessary and for handling ASTs
996 * (e.g. a task switch) prior to return.
998 * In general, only simple access and manipulation of curproc and
999 * the current stack is allowed without having to hold MP lock.
1001 * MPSAFE - note that large sections of this routine are run without
1005 syscall2(struct trapframe
*frame
)
1007 struct thread
*td
= curthread
;
1008 struct proc
*p
= td
->td_proc
;
1009 struct lwp
*lp
= td
->td_lwp
;
1011 struct sysent
*callp
;
1012 register_t orig_tf_rflags
;
1017 int crit_count
= td
->td_pri
& ~TDPRI_MASK
;
1020 int have_mplock
= 0;
1025 union sysunion args
;
1026 register_t
*argsdst
;
1027 kprintf0("SYSCALL rip = %016llx\n", frame
->tf_rip
);
1029 PCPU_INC(cnt
.v_syscall
);
1031 kprintf0("\033[31mSYSCALL %ld\033[39m\n", frame
->tf_rax
);
1033 if (ISPL(frame
->tf_cs
) != SEL_UPL
) {
1040 KTR_LOG(kernentry_syscall
, p
->p_pid
, lp
->lwp_tid
,
1044 KASSERT(td
->td_mpcount
== 0, ("badmpcount syscall2 from %p", (void *)frame
->tf_eip
));
1045 if (syscall_mpsafe
== 0)
1046 MAKEMPSAFE(have_mplock
);
1048 userenter(td
); /* lazy raise our priority */
1055 sticks
= (int)td
->td_sticks
;
1056 orig_tf_rflags
= frame
->tf_rflags
;
1059 * Virtual kernel intercept - if a VM context managed by a virtual
1060 * kernel issues a system call the virtual kernel handles it, not us.
1061 * Restore the virtual kernel context and return from its system
1062 * call. The current frame is copied out to the virtual kernel.
1064 if (lp
->lwp_vkernel
&& lp
->lwp_vkernel
->ve
) {
1065 error
= vkernel_trap(lp
, frame
);
1066 frame
->tf_rax
= error
;
1068 frame
->tf_rflags
|= PSL_C
;
1069 error
= EJUSTRETURN
;
1074 * Get the system call parameters and account for time
1076 lp
->lwp_md
.md_regs
= frame
;
1077 params
= (caddr_t
)frame
->tf_rsp
+ sizeof(register_t
);
1078 code
= frame
->tf_rax
;
1080 if (p
->p_sysent
->sv_prepsyscall
) {
1081 (*p
->p_sysent
->sv_prepsyscall
)(
1082 frame
, (int *)(&args
.nosys
.sysmsg
+ 1),
1085 if (code
== SYS_syscall
|| code
== SYS___syscall
) {
1086 code
= frame
->tf_rdi
;
1092 if (p
->p_sysent
->sv_mask
)
1093 code
&= p
->p_sysent
->sv_mask
;
1095 if (code
>= p
->p_sysent
->sv_size
)
1096 callp
= &p
->p_sysent
->sv_table
[0];
1098 callp
= &p
->p_sysent
->sv_table
[code
];
1100 narg
= callp
->sy_narg
& SYF_ARGMASK
;
1103 * On amd64 we get up to six arguments in registers. The rest are
1104 * on the stack. The first six members of 'struct trapframe' happen
1105 * to be the registers used to pass arguments, in exactly the right
1108 argp
= &frame
->tf_rdi
;
1110 argsdst
= (register_t
*)(&args
.nosys
.sysmsg
+ 1);
1112 * JG can we overflow the space pointed to by 'argsdst'
1113 * either with 'bcopy' or with 'copyin'?
1115 bcopy(argp
, argsdst
, sizeof(register_t
) * regcnt
);
1117 * copyin is MP aware, but the tracing code is not
1119 if (narg
> regcnt
) {
1120 KASSERT(params
!= NULL
, ("copyin args with no params!"));
1121 error
= copyin(params
, &argsdst
[regcnt
],
1122 (narg
- regcnt
) * sizeof(register_t
));
1125 if (KTRPOINT(td
, KTR_SYSCALL
)) {
1126 MAKEMPSAFE(have_mplock
);
1128 ktrsyscall(lp
, code
, narg
,
1129 (void *)(&args
.nosys
.sysmsg
+ 1));
1137 if (KTRPOINT(td
, KTR_SYSCALL
)) {
1138 MAKEMPSAFE(have_mplock
);
1139 ktrsyscall(lp
, code
, narg
, (void *)(&args
.nosys
.sysmsg
+ 1));
1144 * Default return value is 0 (will be copied to %rax). Double-value
1145 * returns use %rax and %rdx. %rdx is left unchanged for system
1146 * calls which return only one result.
1148 args
.sysmsg_fds
[0] = 0;
1149 args
.sysmsg_fds
[1] = frame
->tf_rdx
;
1152 * The syscall might manipulate the trap frame. If it does it
1153 * will probably return EJUSTRETURN.
1155 args
.sysmsg_frame
= frame
;
1157 STOPEVENT(p
, S_SCE
, narg
); /* MP aware */
1161 * Try to run the syscall without the MP lock if the syscall
1162 * is MP safe. We have to obtain the MP lock no matter what if
1165 if ((callp
->sy_narg
& SYF_MPSAFE
) == 0)
1166 MAKEMPSAFE(have_mplock
);
1169 error
= (*callp
->sy_call
)(&args
);
1173 * MP SAFE (we may or may not have the MP lock at this point)
1175 //kprintf("SYSMSG %d ", error);
1179 * Reinitialize proc pointer `p' as it may be different
1180 * if this is a child returning from fork syscall.
1183 lp
= curthread
->td_lwp
;
1184 frame
->tf_rax
= args
.sysmsg_fds
[0];
1185 frame
->tf_rdx
= args
.sysmsg_fds
[1];
1186 kprintf0("RESULT %lld %lld\n", frame
->tf_rax
, frame
->tf_rdx
);
1187 frame
->tf_rflags
&= ~PSL_C
;
1191 * Reconstruct pc, we know that 'syscall' is 2 bytes.
1192 * We have to do a full context restore so that %r10
1193 * (which was holding the value of %rcx) is restored for
1194 * the next iteration.
1196 frame
->tf_rip
-= frame
->tf_err
;
1197 frame
->tf_r10
= frame
->tf_rcx
;
1198 td
->td_pcb
->pcb_flags
|= PCB_FULLCTX
;
1203 panic("Unexpected EASYNC return value (for now)");
1206 if (p
->p_sysent
->sv_errsize
) {
1207 if (error
>= p
->p_sysent
->sv_errsize
)
1208 error
= -1; /* XXX */
1210 error
= p
->p_sysent
->sv_errtbl
[error
];
1212 kprintf0("ERROR %d\n", error
);
1213 frame
->tf_rax
= error
;
1214 frame
->tf_rflags
|= PSL_C
;
1219 * Traced syscall. trapsignal() is not MP aware.
1221 if (orig_tf_rflags
& PSL_T
) {
1222 MAKEMPSAFE(have_mplock
);
1223 frame
->tf_rflags
&= ~PSL_T
;
1224 trapsignal(lp
, SIGTRAP
, 0);
1228 * Handle reschedule and other end-of-syscall issues
1230 userret(lp
, frame
, sticks
);
1233 if (KTRPOINT(td
, KTR_SYSRET
)) {
1234 MAKEMPSAFE(have_mplock
);
1235 ktrsysret(lp
, code
, error
, args
.sysmsg_result
);
1240 * This works because errno is findable through the
1241 * register set. If we ever support an emulation where this
1242 * is not the case, this code will need to be revisited.
1244 STOPEVENT(p
, S_SCX
, code
);
1249 * Release the MP lock if we had to get it
1251 KASSERT(td
->td_mpcount
== have_mplock
,
1252 ("badmpcount syscall2/end from %p", (void *)frame
->tf_eip
));
1256 KTR_LOG(kernentry_syscall_ret
, p
->p_pid
, lp
->lwp_tid
, error
);
1258 KASSERT(crit_count
== (td
->td_pri
& ~TDPRI_MASK
),
1259 ("syscall: critical section count mismatch! %d/%d",
1260 crit_count
/ TDPRI_CRIT
, td
->td_pri
/ TDPRI_CRIT
));
1265 fork_return(struct lwp
*lp
, struct trapframe
*frame
)
1267 kprintf0("fork return\n");
1268 frame
->tf_rax
= 0; /* Child returns zero */
1269 frame
->tf_rflags
&= ~PSL_C
; /* success */
1272 generic_lwp_return(lp
, frame
);
1273 KTR_LOG(kernentry_fork_ret
, lp
->lwp_proc
->p_pid
, lp
->lwp_tid
);
1277 * Simplified back end of syscall(), used when returning from fork()
1278 * directly into user mode. MP lock is held on entry and should be
1279 * released on return. This code will return back into the fork
1280 * trampoline code which then runs doreti.
1283 generic_lwp_return(struct lwp
*lp
, struct trapframe
*frame
)
1285 kprintf0("generic_lwp_return\n");
1286 struct proc
*p
= lp
->lwp_proc
;
1289 * Newly forked processes are given a kernel priority. We have to
1290 * adjust the priority to a normal user priority and fake entry
1291 * into the kernel (call userenter()) to install a passive release
1292 * function just in case userret() decides to stop the process. This
1293 * can occur when ^Z races a fork. If we do not install the passive
1294 * release function the current process designation will not be
1295 * released when the thread goes to sleep.
1297 lwkt_setpri_self(TDPRI_USER_NORM
);
1298 userenter(lp
->lwp_thread
);
1299 userret(lp
, frame
, 0);
1301 if (KTRPOINT(lp
->lwp_thread
, KTR_SYSRET
))
1302 ktrsysret(lp
, SYS_fork
, 0, 0);
1304 p
->p_flag
|= P_PASSIVE_ACQ
;
1306 p
->p_flag
&= ~P_PASSIVE_ACQ
;
1308 KKASSERT(lp
->lwp_thread
->td_mpcount
== 1);
1314 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1315 * fault (which is then passed back to the virtual kernel) if an attempt is
1316 * made to use the FP unit.
1318 * XXX this is a fairly big hack.
1321 set_vkernel_fp(struct trapframe
*frame
)