2 * Copyright (c) 1990, 1993
3 * The Regents of the University of California. All rights reserved.
4 * Copyright (C) 1994, David Greenman
5 * Copyright (c) 2008 The DragonFly Project.
6 * Copyright (c) 2008 Jordan Gordeev.
8 * This code is derived from software contributed to Berkeley by
9 * the University of Utah, and William Jolitz.
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. All advertising materials mentioning features or use of this software
20 * must display the following acknowledgement:
21 * This product includes software developed by the University of
22 * California, Berkeley and its contributors.
23 * 4. Neither the name of the University nor the names of its contributors
24 * may be used to endorse or promote products derived from this software
25 * without specific prior written permission.
27 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39 * from: @(#)trap.c 7.4 (Berkeley) 5/13/91
40 * $FreeBSD: src/sys/i386/i386/trap.c,v 1.147.2.11 2003/02/27 19:09:59 luoqi Exp $
44 * x86_64 Trap and System call handling
50 #include "opt_ktrace.h"
52 #include <machine/frame.h>
53 #include <sys/param.h>
54 #include <sys/systm.h>
55 #include <sys/kernel.h>
56 #include <sys/kerneldump.h>
58 #include <sys/pioctl.h>
59 #include <sys/types.h>
60 #include <sys/signal2.h>
61 #include <sys/syscall.h>
62 #include <sys/sysctl.h>
63 #include <sys/sysent.h>
64 #include <sys/systm.h>
66 #include <sys/ktrace.h>
69 #include <sys/sysmsg.h>
70 #include <sys/sysproto.h>
71 #include <sys/sysunion.h>
75 #include <vm/vm_extern.h>
76 #include <vm/vm_kern.h>
77 #include <vm/vm_param.h>
78 #include <machine/cpu.h>
79 #include <machine/pcb.h>
80 #include <machine/smp.h>
81 #include <machine/thread.h>
82 #include <machine/clock.h>
83 #include <machine/vmparam.h>
84 #include <machine/md_var.h>
85 #include <machine_base/isa/isa_intr.h>
86 #include <machine_base/apic/lapic.h>
90 #include <sys/thread2.h>
91 #include <sys/mplock2.h>
95 #define MAKEMPSAFE(have_mplock) \
96 if (have_mplock == 0) { \
103 #define MAKEMPSAFE(have_mplock)
107 extern void trap(struct trapframe *frame);
109 static int trap_pfault(struct trapframe *, int);
110 static void trap_fatal(struct trapframe *, vm_offset_t);
111 void dblfault_handler(struct trapframe *frame);
113 #define MAX_TRAP_MSG 30
114 static char *trap_msg[] = {
116 "privileged instruction fault", /* 1 T_PRIVINFLT */
118 "breakpoint instruction fault", /* 3 T_BPTFLT */
121 "arithmetic trap", /* 6 T_ARITHTRAP */
122 "system forced exception", /* 7 T_ASTFLT */
124 "general protection fault", /* 9 T_PROTFLT */
125 "trace trap", /* 10 T_TRCTRAP */
127 "page fault", /* 12 T_PAGEFLT */
129 "alignment fault", /* 14 T_ALIGNFLT */
133 "integer divide fault", /* 18 T_DIVIDE */
134 "non-maskable interrupt trap", /* 19 T_NMI */
135 "overflow trap", /* 20 T_OFLOW */
136 "FPU bounds check fault", /* 21 T_BOUND */
137 "FPU device not available", /* 22 T_DNA */
138 "double fault", /* 23 T_DOUBLEFLT */
139 "FPU operand fetch fault", /* 24 T_FPOPFLT */
140 "invalid TSS fault", /* 25 T_TSSFLT */
141 "segment not present fault", /* 26 T_SEGNPFLT */
142 "stack fault", /* 27 T_STKFLT */
143 "machine check trap", /* 28 T_MCHK */
144 "SIMD floating-point exception", /* 29 T_XMMFLT */
145 "reserved (unknown) fault", /* 30 T_RESERVED */
149 static int ddb_on_nmi = 1;
150 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
151 &ddb_on_nmi, 0, "Go to DDB on NMI");
152 static int ddb_on_seg_fault = 0;
153 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_seg_fault, CTLFLAG_RW,
154 &ddb_on_seg_fault, 0, "Go to DDB on user seg-fault");
155 static int freeze_on_seg_fault = 0;
156 SYSCTL_INT(_machdep, OID_AUTO, freeze_on_seg_fault, CTLFLAG_RW,
157 &freeze_on_seg_fault, 0, "Go to DDB on user seg-fault");
159 static int panic_on_nmi = 1;
160 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
161 &panic_on_nmi, 0, "Panic on NMI");
162 static int fast_release;
163 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
164 &fast_release, 0, "Passive Release was optimal");
165 static int slow_release;
166 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
167 &slow_release, 0, "Passive Release was nonoptimal");
170 * System call debugging records the worst-case system call
171 * overhead (inclusive of blocking), but may be inaccurate.
173 /*#define SYSCALL_DEBUG*/
175 uint64_t SysCallsWorstCase[SYS_MAXSYSCALL];
179 * Passively intercepts the thread switch function to increase
180 * the thread priority from a user priority to a kernel priority, reducing
181 * syscall and trap overhead for the case where no switch occurs.
183 * Synchronizes td_ucred with p_ucred. This is used by system calls,
184 * signal handling, faults, AST traps, and anything else that enters the
185 * kernel from userland and provides the kernel with a stable read-only
186 * copy of the process ucred.
189 userenter(struct thread *curtd, struct proc *curp)
194 curtd->td_release = lwkt_passive_release;
196 if (curtd->td_ucred != curp->p_ucred) {
197 ncred = crhold(curp->p_ucred);
198 ocred = curtd->td_ucred;
199 curtd->td_ucred = ncred;
206 * Debugging, remove top two user stack pages to catch kernel faults
208 if (freeze_on_seg_fault > 1 && curtd->td_lwp) {
209 pmap_remove(vmspace_pmap(curtd->td_lwp->lwp_vmspace),
210 0x00007FFFFFFFD000LU,
211 0x0000800000000000LU);
217 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
218 * must be completed before we can return to or try to return to userland.
220 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
221 * arithmatic on the delta calculation so the absolute tick values are
222 * truncated to an integer.
225 userret(struct lwp *lp, struct trapframe *frame, int sticks)
227 struct proc *p = lp->lwp_proc;
231 * Charge system time if profiling. Note: times are in microseconds.
232 * This may do a copyout and block, so do it first even though it
233 * means some system time will be charged as user time.
235 if (p->p_flags & P_PROFIL) {
236 addupc_task(p, frame->tf_rip,
237 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
242 * If the jungle wants us dead, so be it.
244 if (lp->lwp_mpflags & LWP_MP_WEXIT) {
245 lwkt_gettoken(&p->p_token);
247 lwkt_reltoken(&p->p_token); /* NOT REACHED */
251 * Block here if we are in a stopped state.
253 if (p->p_stat == SSTOP || dump_stop_usertds) {
254 lwkt_gettoken(&p->p_token);
256 lwkt_reltoken(&p->p_token);
261 * Post any pending upcalls. If running a virtual kernel be sure
262 * to restore the virtual kernel's vmspace before posting the upcall.
264 if (p->p_flags & (P_SIGVTALRM | P_SIGPROF | P_UPCALLPEND)) {
265 lwkt_gettoken(&p->p_token);
266 if (p->p_flags & P_SIGVTALRM) {
267 p->p_flags &= ~P_SIGVTALRM;
268 ksignal(p, SIGVTALRM);
270 if (p->p_flags & P_SIGPROF) {
271 p->p_flags &= ~P_SIGPROF;
274 if (p->p_flags & P_UPCALLPEND) {
275 p->p_flags &= ~P_UPCALLPEND;
278 lwkt_reltoken(&p->p_token);
283 * Post any pending signals. If running a virtual kernel be sure
284 * to restore the virtual kernel's vmspace before posting the signal.
286 * WARNING! postsig() can exit and not return.
288 if ((sig = CURSIG_TRACE(lp)) != 0) {
289 lwkt_gettoken(&p->p_token);
291 lwkt_reltoken(&p->p_token);
296 * block here if we are swapped out, but still process signals
297 * (such as SIGKILL). proc0 (the swapin scheduler) is already
298 * aware of our situation, we do not have to wake it up.
300 if (p->p_flags & P_SWAPPEDOUT) {
301 lwkt_gettoken(&p->p_token);
303 p->p_flags |= P_SWAPWAIT;
305 if (p->p_flags & P_SWAPWAIT)
306 tsleep(p, PCATCH, "SWOUT", 0);
307 p->p_flags &= ~P_SWAPWAIT;
309 lwkt_reltoken(&p->p_token);
314 * In a multi-threaded program it is possible for a thread to change
315 * signal state during a system call which temporarily changes the
316 * signal mask. In this case postsig() might not be run and we
317 * have to restore the mask ourselves.
319 if (lp->lwp_flags & LWP_OLDMASK) {
320 lp->lwp_flags &= ~LWP_OLDMASK;
321 lp->lwp_sigmask = lp->lwp_oldsigmask;
327 * Cleanup from userenter and any passive release that might have occured.
328 * We must reclaim the current-process designation before we can return
329 * to usermode. We also handle both LWKT and USER reschedule requests.
332 userexit(struct lwp *lp)
334 struct thread *td = lp->lwp_thread;
335 /* globaldata_t gd = td->td_gd; */
338 * Handle stop requests at kernel priority. Any requests queued
339 * after this loop will generate another AST.
341 while (lp->lwp_proc->p_stat == SSTOP) {
342 lwkt_gettoken(&lp->lwp_proc->p_token);
344 lwkt_reltoken(&lp->lwp_proc->p_token);
348 * Reduce our priority in preparation for a return to userland. If
349 * our passive release function was still in place, our priority was
350 * never raised and does not need to be reduced.
352 lwkt_passive_recover(td);
354 /* WARNING: we may have migrated cpu's */
355 /* gd = td->td_gd; */
358 * Become the current user scheduled process if we aren't already,
359 * and deal with reschedule requests and other factors.
361 lp->lwp_proc->p_usched->acquire_curproc(lp);
364 #if !defined(KTR_KERNENTRY)
365 #define KTR_KERNENTRY KTR_ALL
367 KTR_INFO_MASTER(kernentry);
368 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0,
369 "TRAP(pid %d, tid %d, trapno %ld, eva %lu)",
370 pid_t pid, lwpid_t tid, register_t trapno, vm_offset_t eva);
371 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "TRAP_RET(pid %d, tid %d)",
372 pid_t pid, lwpid_t tid);
373 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "SYSC(pid %d, tid %d, nr %ld)",
374 pid_t pid, lwpid_t tid, register_t trapno);
375 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "SYSRET(pid %d, tid %d, err %d)",
376 pid_t pid, lwpid_t tid, int err);
377 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "FORKRET(pid %d, tid %d)",
378 pid_t pid, lwpid_t tid);
381 * Exception, fault, and trap interface to the kernel.
382 * This common code is called from assembly language IDT gate entry
383 * routines that prepare a suitable stack frame, and restore this
384 * frame after the exception has been processed.
386 * This function is also called from doreti in an interlock to handle ASTs.
387 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
389 * NOTE! We have to retrieve the fault address prior to obtaining the
390 * MP lock because get_mplock() may switch out. YYY cr2 really ought
391 * to be retrieved by the assembly code, not here.
393 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
394 * if an attempt is made to switch from a fast interrupt or IPI. This is
395 * necessary to properly take fatal kernel traps on SMP machines if
396 * get_mplock() has to block.
400 trap(struct trapframe *frame)
402 struct globaldata *gd = mycpu;
403 struct thread *td = gd->gd_curthread;
404 struct lwp *lp = td->td_lwp;
407 int i = 0, ucode = 0, type, code;
412 int crit_count = td->td_critcount;
413 lwkt_tokref_t curstop = td->td_toks_stop;
422 * We need to allow T_DNA faults when the debugger is active since
423 * some dumping paths do large bcopy() which use the floating
424 * point registers for faster copying.
426 if (db_active && frame->tf_trapno != T_DNA) {
427 eva = (frame->tf_trapno == T_PAGEFLT ? frame->tf_addr : 0);
428 ++gd->gd_trap_nesting_level;
429 MAKEMPSAFE(have_mplock);
430 trap_fatal(frame, eva);
431 --gd->gd_trap_nesting_level;
438 if ((frame->tf_rflags & PSL_I) == 0) {
440 * Buggy application or kernel code has disabled interrupts
441 * and then trapped. Enabling interrupts now is wrong, but
442 * it is better than running with interrupts disabled until
443 * they are accidentally enabled later.
445 type = frame->tf_trapno;
446 if (ISPL(frame->tf_cs) == SEL_UPL) {
447 MAKEMPSAFE(have_mplock);
448 /* JG curproc can be NULL */
450 "pid %ld (%s): trap %d with interrupts disabled\n",
451 (long)curproc->p_pid, curproc->p_comm, type);
452 } else if (type != T_NMI && type != T_BPTFLT &&
455 * XXX not quite right, since this may be for a
456 * multiple fault in user mode.
458 MAKEMPSAFE(have_mplock);
459 kprintf("kernel trap %d with interrupts disabled\n",
465 type = frame->tf_trapno;
466 code = frame->tf_err;
468 if (ISPL(frame->tf_cs) == SEL_UPL) {
471 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid,
472 frame->tf_trapno, eva);
476 sticks = (int)td->td_sticks;
477 KASSERT(lp->lwp_md.md_regs == frame,
478 ("Frame mismatch %p %p", lp->lwp_md.md_regs, frame));
481 case T_PRIVINFLT: /* privileged instruction fault */
486 case T_BPTFLT: /* bpt instruction fault */
487 case T_TRCTRAP: /* trace trap */
488 frame->tf_rflags &= ~PSL_T;
490 ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
493 case T_ARITHTRAP: /* arithmetic trap */
498 case T_ASTFLT: /* Allow process switch */
499 mycpu->gd_cnt.v_soft++;
500 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
501 atomic_clear_int(&mycpu->gd_reqflags,
503 addupc_task(p, p->p_prof.pr_addr,
508 case T_PROTFLT: /* general protection fault */
512 case T_STKFLT: /* stack fault */
513 case T_SEGNPFLT: /* segment not present fault */
517 case T_TSSFLT: /* invalid TSS fault */
518 case T_DOUBLEFLT: /* double fault */
524 case T_PAGEFLT: /* page fault */
525 i = trap_pfault(frame, TRUE);
526 if (frame->tf_rip == 0) {
527 kprintf("T_PAGEFLT: Warning %%rip == 0!\n");
529 while (freeze_on_seg_fault)
530 tsleep(p, 0, "freeze", hz * 20);
533 if (i == -1 || i == 0)
545 case T_DIVIDE: /* integer divide fault */
552 MAKEMPSAFE(have_mplock);
553 /* machine/parity/power fail/"kitchen sink" faults */
554 if (isa_nmi(code) == 0) {
557 * NMI can be hooked up to a pushbutton
561 kprintf ("NMI ... going to debugger\n");
562 kdb_trap(type, 0, frame);
566 } else if (panic_on_nmi)
567 panic("NMI indicates hardware failure");
569 #endif /* NISA > 0 */
571 case T_OFLOW: /* integer overflow fault */
576 case T_BOUND: /* bounds check fault */
583 * Virtual kernel intercept - pass the DNA exception
584 * to the virtual kernel if it asked to handle it.
585 * This occurs when the virtual kernel is holding
586 * onto the FP context for a different emulated
587 * process then the one currently running.
589 * We must still call npxdna() since we may have
590 * saved FP state that the virtual kernel needs
591 * to hand over to a different emulated process.
593 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
594 (td->td_pcb->pcb_flags & FP_VIRTFP)
601 * The kernel may have switched out the FP unit's
602 * state, causing the user process to take a fault
603 * when it tries to use the FP unit. Restore the
609 ucode = FPE_FPU_NP_TRAP;
612 case T_FPOPFLT: /* FPU operand fetch fault */
617 case T_XMMFLT: /* SIMD floating-point exception */
626 case T_PAGEFLT: /* page fault */
627 trap_pfault(frame, FALSE);
632 * The kernel is apparently using fpu for copying.
633 * XXX this should be fatal unless the kernel has
634 * registered such use.
640 case T_STKFLT: /* stack fault */
643 case T_PROTFLT: /* general protection fault */
644 case T_SEGNPFLT: /* segment not present fault */
646 * Invalid segment selectors and out of bounds
647 * %rip's and %rsp's can be set up in user mode.
648 * This causes a fault in kernel mode when the
649 * kernel tries to return to user mode. We want
650 * to get this fault so that we can fix the
651 * problem here and not have to check all the
652 * selectors and pointers when the user changes
655 if (mycpu->gd_intr_nesting_level == 0) {
657 * NOTE: in 64-bit mode traps push rsp/ss
658 * even if no ring change occurs.
660 if (td->td_pcb->pcb_onfault &&
661 td->td_pcb->pcb_onfault_sp ==
663 frame->tf_rip = (register_t)
664 td->td_pcb->pcb_onfault;
667 if (frame->tf_rip == (long)doreti_iret) {
668 frame->tf_rip = (long)doreti_iret_fault;
676 * PSL_NT can be set in user mode and isn't cleared
677 * automatically when the kernel is entered. This
678 * causes a TSS fault when the kernel attempts to
679 * `iret' because the TSS link is uninitialized. We
680 * want to get this fault so that we can fix the
681 * problem here and not every time the kernel is
684 if (frame->tf_rflags & PSL_NT) {
685 frame->tf_rflags &= ~PSL_NT;
690 case T_TRCTRAP: /* trace trap */
692 if (frame->tf_rip == (int)IDTVEC(syscall)) {
694 * We've just entered system mode via the
695 * syscall lcall. Continue single stepping
696 * silently until the syscall handler has
701 if (frame->tf_rip == (int)IDTVEC(syscall) + 1) {
703 * The syscall handler has now saved the
704 * flags. Stop single stepping it.
706 frame->tf_rflags &= ~PSL_T;
712 * Ignore debug register trace traps due to
713 * accesses in the user's address space, which
714 * can happen under several conditions such as
715 * if a user sets a watchpoint on a buffer and
716 * then passes that buffer to a system call.
717 * We still want to get TRCTRAPS for addresses
718 * in kernel space because that is useful when
719 * debugging the kernel.
722 if (user_dbreg_trap()) {
724 * Reset breakpoint bits because the
727 /* XXX check upper bits here */
728 load_dr6(rdr6() & 0xfffffff0);
733 * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
737 * If DDB is enabled, let it handle the debugger trap.
738 * Otherwise, debugger traps "can't happen".
742 MAKEMPSAFE(have_mplock);
743 if (kdb_trap(type, 0, frame))
750 MAKEMPSAFE(have_mplock);
751 /* machine/parity/power fail/"kitchen sink" faults */
752 if (isa_nmi(code) == 0) {
755 * NMI can be hooked up to a pushbutton
759 kprintf ("NMI ... going to debugger\n");
760 kdb_trap(type, 0, frame);
764 } else if (panic_on_nmi == 0)
767 #endif /* NISA > 0 */
769 MAKEMPSAFE(have_mplock);
770 trap_fatal(frame, 0);
775 * Virtual kernel intercept - if the fault is directly related to a
776 * VM context managed by a virtual kernel then let the virtual kernel
779 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
780 vkernel_trap(lp, frame);
784 /* Translate fault for emulators (e.g. Linux) */
785 if (*p->p_sysent->sv_transtrap)
786 i = (*p->p_sysent->sv_transtrap)(i, type);
788 MAKEMPSAFE(have_mplock);
789 trapsignal(lp, i, ucode);
792 if (type <= MAX_TRAP_MSG) {
793 uprintf("fatal process exception: %s",
795 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
796 uprintf(", fault VA = 0x%lx", frame->tf_addr);
802 userret(lp, frame, sticks);
809 if (p != NULL && lp != NULL)
810 KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid);
812 KASSERT(crit_count == td->td_critcount,
813 ("trap: critical section count mismatch! %d/%d",
814 crit_count, td->td_pri));
815 KASSERT(curstop == td->td_toks_stop,
816 ("trap: extra tokens held after trap! %ld/%ld",
817 curstop - &td->td_toks_base,
818 td->td_toks_stop - &td->td_toks_base));
823 trap_pfault(struct trapframe *frame, int usermode)
826 struct vmspace *vm = NULL;
831 thread_t td = curthread;
832 struct lwp *lp = td->td_lwp;
835 va = trunc_page(frame->tf_addr);
836 if (va >= VM_MIN_KERNEL_ADDRESS) {
838 * Don't allow user-mode faults in kernel address space.
849 * This is a fault on non-kernel virtual memory.
850 * vm is initialized above to NULL. If curproc is NULL
851 * or curproc->p_vmspace is NULL the fault is fatal.
854 vm = lp->lwp_vmspace;
863 * Debugging, try to catch kernel faults on the user address space when not inside
864 * on onfault (e.g. copyin/copyout) routine.
866 if (usermode == 0 && (td->td_pcb == NULL ||
867 td->td_pcb->pcb_onfault == NULL)) {
869 if (freeze_on_seg_fault) {
870 kprintf("trap_pfault: user address fault from kernel mode "
871 "%016lx\n", (long)frame->tf_addr);
872 while (freeze_on_seg_fault)
873 tsleep(&freeze_on_seg_fault, 0, "frzseg", hz * 20);
881 * PGEX_I is defined only if the execute disable bit capability is
882 * supported and enabled.
884 if (frame->tf_err & PGEX_W)
885 ftype = VM_PROT_WRITE;
887 else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
888 ftype = VM_PROT_EXECUTE;
891 ftype = VM_PROT_READ;
893 if (map != &kernel_map) {
895 * Keep swapout from messing with us during this
905 fault_flags |= VM_FAULT_BURST;
906 if (ftype & VM_PROT_WRITE)
907 fault_flags |= VM_FAULT_DIRTY;
909 fault_flags |= VM_FAULT_NORMAL;
910 rv = vm_fault(map, va, ftype, fault_flags);
915 * Don't have to worry about process locking or stacks in the
918 fault_flags = VM_FAULT_NORMAL;
919 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
921 if (rv == KERN_SUCCESS)
926 * NOTE: in 64-bit mode traps push rsp/ss
927 * even if no ring change occurs.
929 if (td->td_pcb->pcb_onfault &&
930 td->td_pcb->pcb_onfault_sp == frame->tf_rsp &&
931 td->td_gd->gd_intr_nesting_level == 0) {
932 frame->tf_rip = (register_t)td->td_pcb->pcb_onfault;
935 trap_fatal(frame, frame->tf_addr);
940 * NOTE: on x86_64 we have a tf_addr field in the trapframe, no
941 * kludge is needed to pass the fault address to signal handlers.
944 if (td->td_lwp->lwp_vkernel == NULL) {
946 if (bootverbose || freeze_on_seg_fault || ddb_on_seg_fault) {
950 kprintf("seg-fault ft=%04x ff=%04x addr=%p rip=%p "
951 "pid=%d cpu=%d p_comm=%s\n",
953 (void *)frame->tf_addr,
954 (void *)frame->tf_rip,
955 p->p_pid, mycpu->gd_cpuid, p->p_comm);
958 while (freeze_on_seg_fault) {
959 tsleep(p, 0, "freeze", hz * 20);
961 if (ddb_on_seg_fault)
962 Debugger("ddb_on_seg_fault");
966 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
970 trap_fatal(struct trapframe *frame, vm_offset_t eva)
975 struct soft_segment_descriptor softseg;
978 code = frame->tf_err;
979 type = frame->tf_trapno;
980 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)], &softseg);
982 if (type <= MAX_TRAP_MSG)
983 msg = trap_msg[type];
986 kprintf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
987 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
989 /* three separate prints in case of a trap on an unmapped page */
990 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
991 kprintf("lapic->id = %08x\n", lapic->id);
993 if (type == T_PAGEFLT) {
994 kprintf("fault virtual address = 0x%lx\n", eva);
995 kprintf("fault code = %s %s %s, %s\n",
996 code & PGEX_U ? "user" : "supervisor",
997 code & PGEX_W ? "write" : "read",
998 code & PGEX_I ? "instruction" : "data",
999 code & PGEX_P ? "protection violation" : "page not present");
1001 kprintf("instruction pointer = 0x%lx:0x%lx\n",
1002 frame->tf_cs & 0xffff, frame->tf_rip);
1003 if (ISPL(frame->tf_cs) == SEL_UPL) {
1004 ss = frame->tf_ss & 0xffff;
1005 rsp = frame->tf_rsp;
1008 * NOTE: in 64-bit mode traps push rsp/ss even if no ring
1011 ss = GSEL(GDATA_SEL, SEL_KPL);
1012 rsp = frame->tf_rsp;
1014 kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp);
1015 kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp);
1016 kprintf("code segment = base 0x%lx, limit 0x%lx, type 0x%x\n",
1017 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
1018 kprintf(" = DPL %d, pres %d, long %d, def32 %d, gran %d\n",
1019 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32,
1021 kprintf("processor eflags = ");
1022 if (frame->tf_rflags & PSL_T)
1023 kprintf("trace trap, ");
1024 if (frame->tf_rflags & PSL_I)
1025 kprintf("interrupt enabled, ");
1026 if (frame->tf_rflags & PSL_NT)
1027 kprintf("nested task, ");
1028 if (frame->tf_rflags & PSL_RF)
1029 kprintf("resume, ");
1030 kprintf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12);
1031 kprintf("current process = ");
1034 (u_long)curproc->p_pid);
1038 kprintf("current thread = pri %d ", curthread->td_pri);
1039 if (curthread->td_critcount)
1044 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1047 kprintf("trap number = %d\n", type);
1048 if (type <= MAX_TRAP_MSG)
1049 panic("%s", trap_msg[type]);
1051 panic("unknown/reserved trap");
1055 * Double fault handler. Called when a fault occurs while writing
1056 * a frame for a trap/exception onto the stack. This usually occurs
1057 * when the stack overflows (such is the case with infinite recursion,
1062 in_kstack_guard(register_t rptr)
1064 thread_t td = curthread;
1066 if ((char *)rptr >= td->td_kstack &&
1067 (char *)rptr < td->td_kstack + PAGE_SIZE) {
1074 dblfault_handler(struct trapframe *frame)
1076 thread_t td = curthread;
1078 if (in_kstack_guard(frame->tf_rsp) || in_kstack_guard(frame->tf_rbp)) {
1079 kprintf("DOUBLE FAULT - KERNEL STACK GUARD HIT!\n");
1080 if (in_kstack_guard(frame->tf_rsp))
1081 frame->tf_rsp = (register_t)(td->td_kstack + PAGE_SIZE);
1082 if (in_kstack_guard(frame->tf_rbp))
1083 frame->tf_rbp = (register_t)(td->td_kstack + PAGE_SIZE);
1085 kprintf("DOUBLE FAULT\n");
1087 kprintf("\nFatal double fault\n");
1088 kprintf("rip = 0x%lx\n", frame->tf_rip);
1089 kprintf("rsp = 0x%lx\n", frame->tf_rsp);
1090 kprintf("rbp = 0x%lx\n", frame->tf_rbp);
1092 /* three separate prints in case of a trap on an unmapped page */
1093 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
1094 kprintf("lapic->id = %08x\n", lapic->id);
1096 panic("double fault");
1100 * syscall2 - MP aware system call request C handler
1102 * A system call is essentially treated as a trap except that the
1103 * MP lock is not held on entry or return. We are responsible for
1104 * obtaining the MP lock if necessary and for handling ASTs
1105 * (e.g. a task switch) prior to return.
1110 syscall2(struct trapframe *frame)
1112 struct thread *td = curthread;
1113 struct proc *p = td->td_proc;
1114 struct lwp *lp = td->td_lwp;
1116 struct sysent *callp;
1117 register_t orig_tf_rflags;
1122 int crit_count = td->td_critcount;
1125 int have_mplock = 0;
1130 union sysunion args;
1131 register_t *argsdst;
1133 mycpu->gd_cnt.v_syscall++;
1136 if (ISPL(frame->tf_cs) != SEL_UPL) {
1143 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid,
1146 userenter(td, p); /* lazy raise our priority */
1153 sticks = (int)td->td_sticks;
1154 orig_tf_rflags = frame->tf_rflags;
1157 * Virtual kernel intercept - if a VM context managed by a virtual
1158 * kernel issues a system call the virtual kernel handles it, not us.
1159 * Restore the virtual kernel context and return from its system
1160 * call. The current frame is copied out to the virtual kernel.
1162 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1163 vkernel_trap(lp, frame);
1164 error = EJUSTRETURN;
1169 * Get the system call parameters and account for time
1171 KASSERT(lp->lwp_md.md_regs == frame,
1172 ("Frame mismatch %p %p", lp->lwp_md.md_regs, frame));
1173 params = (caddr_t)frame->tf_rsp + sizeof(register_t);
1174 code = frame->tf_rax;
1176 if (p->p_sysent->sv_prepsyscall) {
1177 (*p->p_sysent->sv_prepsyscall)(
1178 frame, (int *)(&args.nosys.sysmsg + 1),
1181 if (code == SYS_syscall || code == SYS___syscall) {
1182 code = frame->tf_rdi;
1188 if (p->p_sysent->sv_mask)
1189 code &= p->p_sysent->sv_mask;
1191 if (code >= p->p_sysent->sv_size)
1192 callp = &p->p_sysent->sv_table[0];
1194 callp = &p->p_sysent->sv_table[code];
1196 narg = callp->sy_narg & SYF_ARGMASK;
1199 * On x86_64 we get up to six arguments in registers. The rest are
1200 * on the stack. The first six members of 'struct trapframe' happen
1201 * to be the registers used to pass arguments, in exactly the right
1204 argp = &frame->tf_rdi;
1206 argsdst = (register_t *)(&args.nosys.sysmsg + 1);
1208 * JG can we overflow the space pointed to by 'argsdst'
1209 * either with 'bcopy' or with 'copyin'?
1211 bcopy(argp, argsdst, sizeof(register_t) * regcnt);
1213 * copyin is MP aware, but the tracing code is not
1215 if (narg > regcnt) {
1216 KASSERT(params != NULL, ("copyin args with no params!"));
1217 error = copyin(params, &argsdst[regcnt],
1218 (narg - regcnt) * sizeof(register_t));
1221 if (KTRPOINT(td, KTR_SYSCALL)) {
1222 MAKEMPSAFE(have_mplock);
1224 ktrsyscall(lp, code, narg,
1225 (void *)(&args.nosys.sysmsg + 1));
1233 if (KTRPOINT(td, KTR_SYSCALL)) {
1234 MAKEMPSAFE(have_mplock);
1235 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1240 * Default return value is 0 (will be copied to %rax). Double-value
1241 * returns use %rax and %rdx. %rdx is left unchanged for system
1242 * calls which return only one result.
1244 args.sysmsg_fds[0] = 0;
1245 args.sysmsg_fds[1] = frame->tf_rdx;
1248 * The syscall might manipulate the trap frame. If it does it
1249 * will probably return EJUSTRETURN.
1251 args.sysmsg_frame = frame;
1253 STOPEVENT(p, S_SCE, narg); /* MP aware */
1256 * NOTE: All system calls run MPSAFE now. The system call itself
1257 * is responsible for getting the MP lock.
1259 #ifdef SYSCALL_DEBUG
1260 uint64_t tscval = rdtsc();
1262 error = (*callp->sy_call)(&args);
1263 #ifdef SYSCALL_DEBUG
1264 tscval = rdtsc() - tscval;
1265 tscval = tscval * 1000000 / tsc_frequency;
1266 if (SysCallsWorstCase[code] < tscval)
1267 SysCallsWorstCase[code] = tscval;
1272 * MP SAFE (we may or may not have the MP lock at this point)
1274 //kprintf("SYSMSG %d ", error);
1278 * Reinitialize proc pointer `p' as it may be different
1279 * if this is a child returning from fork syscall.
1282 lp = curthread->td_lwp;
1283 frame->tf_rax = args.sysmsg_fds[0];
1284 frame->tf_rdx = args.sysmsg_fds[1];
1285 frame->tf_rflags &= ~PSL_C;
1289 * Reconstruct pc, we know that 'syscall' is 2 bytes.
1290 * We have to do a full context restore so that %r10
1291 * (which was holding the value of %rcx) is restored for
1292 * the next iteration.
1294 if (frame->tf_err != 0 && frame->tf_err != 2)
1295 kprintf("lp %s:%d frame->tf_err is weird %ld\n",
1296 td->td_comm, lp->lwp_proc->p_pid, frame->tf_err);
1297 frame->tf_rip -= frame->tf_err;
1298 frame->tf_r10 = frame->tf_rcx;
1303 panic("Unexpected EASYNC return value (for now)");
1306 if (p->p_sysent->sv_errsize) {
1307 if (error >= p->p_sysent->sv_errsize)
1308 error = -1; /* XXX */
1310 error = p->p_sysent->sv_errtbl[error];
1312 frame->tf_rax = error;
1313 frame->tf_rflags |= PSL_C;
1318 * Traced syscall. trapsignal() is not MP aware.
1320 if (orig_tf_rflags & PSL_T) {
1321 MAKEMPSAFE(have_mplock);
1322 frame->tf_rflags &= ~PSL_T;
1323 trapsignal(lp, SIGTRAP, TRAP_TRACE);
1327 * Handle reschedule and other end-of-syscall issues
1329 userret(lp, frame, sticks);
1332 if (KTRPOINT(td, KTR_SYSRET)) {
1333 MAKEMPSAFE(have_mplock);
1334 ktrsysret(lp, code, error, args.sysmsg_result);
1339 * This works because errno is findable through the
1340 * register set. If we ever support an emulation where this
1341 * is not the case, this code will need to be revisited.
1343 STOPEVENT(p, S_SCX, code);
1348 * Release the MP lock if we had to get it
1353 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error);
1355 KASSERT(crit_count == td->td_critcount,
1356 ("syscall: critical section count mismatch! %d/%d",
1357 crit_count, td->td_pri));
1358 KASSERT(&td->td_toks_base == td->td_toks_stop,
1359 ("syscall: extra tokens held after trap! %ld",
1360 td->td_toks_stop - &td->td_toks_base));
1365 * NOTE: mplock not held at any point
1368 fork_return(struct lwp *lp, struct trapframe *frame)
1370 frame->tf_rax = 0; /* Child returns zero */
1371 frame->tf_rflags &= ~PSL_C; /* success */
1374 generic_lwp_return(lp, frame);
1375 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1379 * Simplified back end of syscall(), used when returning from fork()
1380 * directly into user mode.
1382 * This code will return back into the fork trampoline code which then
1385 * NOTE: The mplock is not held at any point.
1388 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1390 struct proc *p = lp->lwp_proc;
1393 * Newly forked processes are given a kernel priority. We have to
1394 * adjust the priority to a normal user priority and fake entry
1395 * into the kernel (call userenter()) to install a passive release
1396 * function just in case userret() decides to stop the process. This
1397 * can occur when ^Z races a fork. If we do not install the passive
1398 * release function the current process designation will not be
1399 * released when the thread goes to sleep.
1401 lwkt_setpri_self(TDPRI_USER_NORM);
1402 userenter(lp->lwp_thread, p);
1403 userret(lp, frame, 0);
1405 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1406 ktrsysret(lp, SYS_fork, 0, 0);
1408 lp->lwp_flags |= LWP_PASSIVE_ACQ;
1410 lp->lwp_flags &= ~LWP_PASSIVE_ACQ;
1414 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1415 * fault (which is then passed back to the virtual kernel) if an attempt is
1416 * made to use the FP unit.
1418 * XXX this is a fairly big hack.
1421 set_vkernel_fp(struct trapframe *frame)
1423 struct thread *td = curthread;
1425 if (frame->tf_xflags & PGEX_FPFAULT) {
1426 td->td_pcb->pcb_flags |= FP_VIRTFP;
1427 if (mdcpu->gd_npxthread == td)
1430 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1435 * Called from vkernel_trap() to fixup the vkernel's syscall
1436 * frame for vmspace_ctl() return.
1439 cpu_vkernel_trap(struct trapframe *frame, int error)
1441 frame->tf_rax = error;
1443 frame->tf_rflags |= PSL_C;
1445 frame->tf_rflags &= ~PSL_C;