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;
205 * Debugging, remove top two user stack pages to catch kernel faults
207 if (freeze_on_seg_fault > 1 && curtd->td_lwp) {
208 pmap_remove(vmspace_pmap(curtd->td_lwp->lwp_vmspace),
209 0x00007FFFFFFFD000LU,
210 0x0000800000000000LU);
215 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
216 * must be completed before we can return to or try to return to userland.
218 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
219 * arithmatic on the delta calculation so the absolute tick values are
220 * truncated to an integer.
223 userret(struct lwp *lp, struct trapframe *frame, int sticks)
225 struct proc *p = lp->lwp_proc;
229 * Charge system time if profiling. Note: times are in microseconds.
230 * This may do a copyout and block, so do it first even though it
231 * means some system time will be charged as user time.
233 if (p->p_flags & P_PROFIL) {
234 addupc_task(p, frame->tf_rip,
235 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
240 * If the jungle wants us dead, so be it.
242 if (lp->lwp_mpflags & LWP_MP_WEXIT) {
243 lwkt_gettoken(&p->p_token);
245 lwkt_reltoken(&p->p_token); /* NOT REACHED */
249 * Block here if we are in a stopped state.
251 if (p->p_stat == SSTOP || dump_stop_usertds) {
252 lwkt_gettoken(&p->p_token);
254 lwkt_reltoken(&p->p_token);
259 * Post any pending upcalls. If running a virtual kernel be sure
260 * to restore the virtual kernel's vmspace before posting the upcall.
262 if (p->p_flags & (P_SIGVTALRM | P_SIGPROF | P_UPCALLPEND)) {
263 lwkt_gettoken(&p->p_token);
264 if (p->p_flags & P_SIGVTALRM) {
265 p->p_flags &= ~P_SIGVTALRM;
266 ksignal(p, SIGVTALRM);
268 if (p->p_flags & P_SIGPROF) {
269 p->p_flags &= ~P_SIGPROF;
272 if (p->p_flags & P_UPCALLPEND) {
273 p->p_flags &= ~P_UPCALLPEND;
276 lwkt_reltoken(&p->p_token);
281 * Post any pending signals. If running a virtual kernel be sure
282 * to restore the virtual kernel's vmspace before posting the signal.
284 * WARNING! postsig() can exit and not return.
286 if ((sig = CURSIG_TRACE(lp)) != 0) {
287 lwkt_gettoken(&p->p_token);
289 lwkt_reltoken(&p->p_token);
294 * block here if we are swapped out, but still process signals
295 * (such as SIGKILL). proc0 (the swapin scheduler) is already
296 * aware of our situation, we do not have to wake it up.
298 if (p->p_flags & P_SWAPPEDOUT) {
299 lwkt_gettoken(&p->p_token);
301 p->p_flags |= P_SWAPWAIT;
303 if (p->p_flags & P_SWAPWAIT)
304 tsleep(p, PCATCH, "SWOUT", 0);
305 p->p_flags &= ~P_SWAPWAIT;
307 lwkt_reltoken(&p->p_token);
312 * Make sure postsig() handled request to restore old signal mask after
313 * running signal handler.
315 KKASSERT((lp->lwp_flags & LWP_OLDMASK) == 0);
319 * Cleanup from userenter and any passive release that might have occured.
320 * We must reclaim the current-process designation before we can return
321 * to usermode. We also handle both LWKT and USER reschedule requests.
324 userexit(struct lwp *lp)
326 struct thread *td = lp->lwp_thread;
327 /* globaldata_t gd = td->td_gd; */
330 * Handle stop requests at kernel priority. Any requests queued
331 * after this loop will generate another AST.
333 while (lp->lwp_proc->p_stat == SSTOP) {
334 lwkt_gettoken(&lp->lwp_proc->p_token);
336 lwkt_reltoken(&lp->lwp_proc->p_token);
340 * Reduce our priority in preparation for a return to userland. If
341 * our passive release function was still in place, our priority was
342 * never raised and does not need to be reduced.
344 lwkt_passive_recover(td);
347 * Become the current user scheduled process if we aren't already,
348 * and deal with reschedule requests and other factors.
350 lp->lwp_proc->p_usched->acquire_curproc(lp);
351 /* WARNING: we may have migrated cpu's */
352 /* gd = td->td_gd; */
355 #if !defined(KTR_KERNENTRY)
356 #define KTR_KERNENTRY KTR_ALL
358 KTR_INFO_MASTER(kernentry);
359 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0,
360 "TRAP(pid %d, tid %d, trapno %ld, eva %lu)",
361 pid_t pid, lwpid_t tid, register_t trapno, vm_offset_t eva);
362 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "TRAP_RET(pid %d, tid %d)",
363 pid_t pid, lwpid_t tid);
364 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "SYSC(pid %d, tid %d, nr %ld)",
365 pid_t pid, lwpid_t tid, register_t trapno);
366 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "SYSRET(pid %d, tid %d, err %d)",
367 pid_t pid, lwpid_t tid, int err);
368 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "FORKRET(pid %d, tid %d)",
369 pid_t pid, lwpid_t tid);
372 * Exception, fault, and trap interface to the kernel.
373 * This common code is called from assembly language IDT gate entry
374 * routines that prepare a suitable stack frame, and restore this
375 * frame after the exception has been processed.
377 * This function is also called from doreti in an interlock to handle ASTs.
378 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
380 * NOTE! We have to retrieve the fault address prior to obtaining the
381 * MP lock because get_mplock() may switch out. YYY cr2 really ought
382 * to be retrieved by the assembly code, not here.
384 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
385 * if an attempt is made to switch from a fast interrupt or IPI. This is
386 * necessary to properly take fatal kernel traps on SMP machines if
387 * get_mplock() has to block.
391 trap(struct trapframe *frame)
393 struct globaldata *gd = mycpu;
394 struct thread *td = gd->gd_curthread;
395 struct lwp *lp = td->td_lwp;
398 int i = 0, ucode = 0, type, code;
403 int crit_count = td->td_critcount;
404 lwkt_tokref_t curstop = td->td_toks_stop;
413 * We need to allow T_DNA faults when the debugger is active since
414 * some dumping paths do large bcopy() which use the floating
415 * point registers for faster copying.
417 if (db_active && frame->tf_trapno != T_DNA) {
418 eva = (frame->tf_trapno == T_PAGEFLT ? frame->tf_addr : 0);
419 ++gd->gd_trap_nesting_level;
420 MAKEMPSAFE(have_mplock);
421 trap_fatal(frame, eva);
422 --gd->gd_trap_nesting_level;
429 if ((frame->tf_rflags & PSL_I) == 0) {
431 * Buggy application or kernel code has disabled interrupts
432 * and then trapped. Enabling interrupts now is wrong, but
433 * it is better than running with interrupts disabled until
434 * they are accidentally enabled later.
436 type = frame->tf_trapno;
437 if (ISPL(frame->tf_cs) == SEL_UPL) {
438 MAKEMPSAFE(have_mplock);
439 /* JG curproc can be NULL */
441 "pid %ld (%s): trap %d with interrupts disabled\n",
442 (long)curproc->p_pid, curproc->p_comm, type);
443 } else if (type != T_NMI && type != T_BPTFLT &&
446 * XXX not quite right, since this may be for a
447 * multiple fault in user mode.
449 MAKEMPSAFE(have_mplock);
450 kprintf("kernel trap %d with interrupts disabled\n",
456 type = frame->tf_trapno;
457 code = frame->tf_err;
459 if (ISPL(frame->tf_cs) == SEL_UPL) {
462 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid,
463 frame->tf_trapno, eva);
467 sticks = (int)td->td_sticks;
468 KASSERT(lp->lwp_md.md_regs == frame,
469 ("Frame mismatch %p %p", lp->lwp_md.md_regs, frame));
472 case T_PRIVINFLT: /* privileged instruction fault */
477 case T_BPTFLT: /* bpt instruction fault */
478 case T_TRCTRAP: /* trace trap */
479 frame->tf_rflags &= ~PSL_T;
481 ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
484 case T_ARITHTRAP: /* arithmetic trap */
489 case T_ASTFLT: /* Allow process switch */
490 mycpu->gd_cnt.v_soft++;
491 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
492 atomic_clear_int(&mycpu->gd_reqflags,
494 addupc_task(p, p->p_prof.pr_addr,
499 case T_PROTFLT: /* general protection fault */
503 case T_STKFLT: /* stack fault */
504 case T_SEGNPFLT: /* segment not present fault */
508 case T_TSSFLT: /* invalid TSS fault */
509 case T_DOUBLEFLT: /* double fault */
515 case T_PAGEFLT: /* page fault */
516 i = trap_pfault(frame, TRUE);
517 if (frame->tf_rip == 0) {
518 kprintf("T_PAGEFLT: Warning %%rip == 0!\n");
519 while (freeze_on_seg_fault) {
520 tsleep(p, 0, "freeze", hz * 20);
523 if (i == -1 || i == 0)
535 case T_DIVIDE: /* integer divide fault */
542 MAKEMPSAFE(have_mplock);
543 /* machine/parity/power fail/"kitchen sink" faults */
544 if (isa_nmi(code) == 0) {
547 * NMI can be hooked up to a pushbutton
551 kprintf ("NMI ... going to debugger\n");
552 kdb_trap(type, 0, frame);
556 } else if (panic_on_nmi)
557 panic("NMI indicates hardware failure");
559 #endif /* NISA > 0 */
561 case T_OFLOW: /* integer overflow fault */
566 case T_BOUND: /* bounds check fault */
573 * Virtual kernel intercept - pass the DNA exception
574 * to the virtual kernel if it asked to handle it.
575 * This occurs when the virtual kernel is holding
576 * onto the FP context for a different emulated
577 * process then the one currently running.
579 * We must still call npxdna() since we may have
580 * saved FP state that the virtual kernel needs
581 * to hand over to a different emulated process.
583 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
584 (td->td_pcb->pcb_flags & FP_VIRTFP)
591 * The kernel may have switched out the FP unit's
592 * state, causing the user process to take a fault
593 * when it tries to use the FP unit. Restore the
599 ucode = FPE_FPU_NP_TRAP;
602 case T_FPOPFLT: /* FPU operand fetch fault */
607 case T_XMMFLT: /* SIMD floating-point exception */
616 case T_PAGEFLT: /* page fault */
617 trap_pfault(frame, FALSE);
622 * The kernel is apparently using fpu for copying.
623 * XXX this should be fatal unless the kernel has
624 * registered such use.
630 case T_STKFLT: /* stack fault */
633 case T_PROTFLT: /* general protection fault */
634 case T_SEGNPFLT: /* segment not present fault */
636 * Invalid segment selectors and out of bounds
637 * %rip's and %rsp's can be set up in user mode.
638 * This causes a fault in kernel mode when the
639 * kernel tries to return to user mode. We want
640 * to get this fault so that we can fix the
641 * problem here and not have to check all the
642 * selectors and pointers when the user changes
645 if (mycpu->gd_intr_nesting_level == 0) {
646 if (td->td_pcb->pcb_onfault) {
647 frame->tf_rip = (register_t)
648 td->td_pcb->pcb_onfault;
651 if (frame->tf_rip == (long)doreti_iret) {
652 frame->tf_rip = (long)doreti_iret_fault;
660 * PSL_NT can be set in user mode and isn't cleared
661 * automatically when the kernel is entered. This
662 * causes a TSS fault when the kernel attempts to
663 * `iret' because the TSS link is uninitialized. We
664 * want to get this fault so that we can fix the
665 * problem here and not every time the kernel is
668 if (frame->tf_rflags & PSL_NT) {
669 frame->tf_rflags &= ~PSL_NT;
674 case T_TRCTRAP: /* trace trap */
676 if (frame->tf_rip == (int)IDTVEC(syscall)) {
678 * We've just entered system mode via the
679 * syscall lcall. Continue single stepping
680 * silently until the syscall handler has
685 if (frame->tf_rip == (int)IDTVEC(syscall) + 1) {
687 * The syscall handler has now saved the
688 * flags. Stop single stepping it.
690 frame->tf_rflags &= ~PSL_T;
696 * Ignore debug register trace traps due to
697 * accesses in the user's address space, which
698 * can happen under several conditions such as
699 * if a user sets a watchpoint on a buffer and
700 * then passes that buffer to a system call.
701 * We still want to get TRCTRAPS for addresses
702 * in kernel space because that is useful when
703 * debugging the kernel.
706 if (user_dbreg_trap()) {
708 * Reset breakpoint bits because the
711 /* XXX check upper bits here */
712 load_dr6(rdr6() & 0xfffffff0);
717 * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
721 * If DDB is enabled, let it handle the debugger trap.
722 * Otherwise, debugger traps "can't happen".
726 MAKEMPSAFE(have_mplock);
727 if (kdb_trap(type, 0, frame))
734 MAKEMPSAFE(have_mplock);
735 /* machine/parity/power fail/"kitchen sink" faults */
736 if (isa_nmi(code) == 0) {
739 * NMI can be hooked up to a pushbutton
743 kprintf ("NMI ... going to debugger\n");
744 kdb_trap(type, 0, frame);
748 } else if (panic_on_nmi == 0)
751 #endif /* NISA > 0 */
753 MAKEMPSAFE(have_mplock);
754 trap_fatal(frame, 0);
759 * Virtual kernel intercept - if the fault is directly related to a
760 * VM context managed by a virtual kernel then let the virtual kernel
763 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
764 vkernel_trap(lp, frame);
768 /* Translate fault for emulators (e.g. Linux) */
769 if (*p->p_sysent->sv_transtrap)
770 i = (*p->p_sysent->sv_transtrap)(i, type);
772 MAKEMPSAFE(have_mplock);
773 trapsignal(lp, i, ucode);
776 if (type <= MAX_TRAP_MSG) {
777 uprintf("fatal process exception: %s",
779 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
780 uprintf(", fault VA = 0x%lx", frame->tf_addr);
786 userret(lp, frame, sticks);
793 if (p != NULL && lp != NULL)
794 KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid);
796 KASSERT(crit_count == td->td_critcount,
797 ("trap: critical section count mismatch! %d/%d",
798 crit_count, td->td_pri));
799 KASSERT(curstop == td->td_toks_stop,
800 ("trap: extra tokens held after trap! %ld/%ld",
801 curstop - &td->td_toks_base,
802 td->td_toks_stop - &td->td_toks_base));
807 trap_pfault(struct trapframe *frame, int usermode)
810 struct vmspace *vm = NULL;
815 thread_t td = curthread;
816 struct lwp *lp = td->td_lwp;
819 va = trunc_page(frame->tf_addr);
820 if (va >= VM_MIN_KERNEL_ADDRESS) {
822 * Don't allow user-mode faults in kernel address space.
833 * This is a fault on non-kernel virtual memory.
834 * vm is initialized above to NULL. If curproc is NULL
835 * or curproc->p_vmspace is NULL the fault is fatal.
838 vm = lp->lwp_vmspace;
847 * Debugging, try to catch kernel faults on the user address space when not inside
848 * on onfault (e.g. copyin/copyout) routine.
850 if (usermode == 0 && (td->td_pcb == NULL || td->td_pcb->pcb_onfault == NULL)) {
851 if (freeze_on_seg_fault) {
852 kprintf("trap_pfault: user address fault from kernel mode "
853 "%016lx\n", (long)frame->tf_addr);
854 while (freeze_on_seg_fault) {
855 tsleep(&freeze_on_seg_fault, 0, "frzseg", hz * 20);
863 * PGEX_I is defined only if the execute disable bit capability is
864 * supported and enabled.
866 if (frame->tf_err & PGEX_W)
867 ftype = VM_PROT_WRITE;
869 else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
870 ftype = VM_PROT_EXECUTE;
873 ftype = VM_PROT_READ;
875 if (map != &kernel_map) {
877 * Keep swapout from messing with us during this
887 fault_flags |= VM_FAULT_BURST;
888 if (ftype & VM_PROT_WRITE)
889 fault_flags |= VM_FAULT_DIRTY;
891 fault_flags |= VM_FAULT_NORMAL;
892 rv = vm_fault(map, va, ftype, fault_flags);
897 * Don't have to worry about process locking or stacks in the
900 fault_flags = VM_FAULT_NORMAL;
901 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
903 if (rv == KERN_SUCCESS)
907 if (td->td_gd->gd_intr_nesting_level == 0 &&
908 td->td_pcb->pcb_onfault) {
909 frame->tf_rip = (register_t)td->td_pcb->pcb_onfault;
912 trap_fatal(frame, frame->tf_addr);
917 * NOTE: on x86_64 we have a tf_addr field in the trapframe, no
918 * kludge is needed to pass the fault address to signal handlers.
921 if (td->td_lwp->lwp_vkernel == NULL) {
922 if (bootverbose || freeze_on_seg_fault || ddb_on_seg_fault) {
923 kprintf("seg-fault ft=%04x ff=%04x addr=%p rip=%p "
924 "pid=%d cpu=%d p_comm=%s\n",
926 (void *)frame->tf_addr,
927 (void *)frame->tf_rip,
928 p->p_pid, mycpu->gd_cpuid, p->p_comm);
931 while (freeze_on_seg_fault) {
932 tsleep(p, 0, "freeze", hz * 20);
934 if (ddb_on_seg_fault)
935 Debugger("ddb_on_seg_fault");
939 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
943 trap_fatal(struct trapframe *frame, vm_offset_t eva)
948 struct soft_segment_descriptor softseg;
951 code = frame->tf_err;
952 type = frame->tf_trapno;
953 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)], &softseg);
955 if (type <= MAX_TRAP_MSG)
956 msg = trap_msg[type];
959 kprintf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
960 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
962 /* three separate prints in case of a trap on an unmapped page */
963 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
964 kprintf("lapic->id = %08x\n", lapic->id);
966 if (type == T_PAGEFLT) {
967 kprintf("fault virtual address = 0x%lx\n", eva);
968 kprintf("fault code = %s %s %s, %s\n",
969 code & PGEX_U ? "user" : "supervisor",
970 code & PGEX_W ? "write" : "read",
971 code & PGEX_I ? "instruction" : "data",
972 code & PGEX_P ? "protection violation" : "page not present");
974 kprintf("instruction pointer = 0x%lx:0x%lx\n",
975 frame->tf_cs & 0xffff, frame->tf_rip);
976 if (ISPL(frame->tf_cs) == SEL_UPL) {
977 ss = frame->tf_ss & 0xffff;
980 ss = GSEL(GDATA_SEL, SEL_KPL);
981 rsp = (long)&frame->tf_rsp;
983 kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp);
984 kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp);
985 kprintf("code segment = base 0x%lx, limit 0x%lx, type 0x%x\n",
986 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
987 kprintf(" = DPL %d, pres %d, long %d, def32 %d, gran %d\n",
988 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32,
990 kprintf("processor eflags = ");
991 if (frame->tf_rflags & PSL_T)
992 kprintf("trace trap, ");
993 if (frame->tf_rflags & PSL_I)
994 kprintf("interrupt enabled, ");
995 if (frame->tf_rflags & PSL_NT)
996 kprintf("nested task, ");
997 if (frame->tf_rflags & PSL_RF)
999 kprintf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12);
1000 kprintf("current process = ");
1003 (u_long)curproc->p_pid);
1007 kprintf("current thread = pri %d ", curthread->td_pri);
1008 if (curthread->td_critcount)
1013 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1016 kprintf("trap number = %d\n", type);
1017 if (type <= MAX_TRAP_MSG)
1018 panic("%s", trap_msg[type]);
1020 panic("unknown/reserved trap");
1024 * Double fault handler. Called when a fault occurs while writing
1025 * a frame for a trap/exception onto the stack. This usually occurs
1026 * when the stack overflows (such is the case with infinite recursion,
1031 in_kstack_guard(register_t rptr)
1033 thread_t td = curthread;
1035 if ((char *)rptr >= td->td_kstack &&
1036 (char *)rptr < td->td_kstack + PAGE_SIZE) {
1043 dblfault_handler(struct trapframe *frame)
1045 thread_t td = curthread;
1047 if (in_kstack_guard(frame->tf_rsp) || in_kstack_guard(frame->tf_rbp)) {
1048 kprintf("DOUBLE FAULT - KERNEL STACK GUARD HIT!\n");
1049 if (in_kstack_guard(frame->tf_rsp))
1050 frame->tf_rsp = (register_t)(td->td_kstack + PAGE_SIZE);
1051 if (in_kstack_guard(frame->tf_rbp))
1052 frame->tf_rbp = (register_t)(td->td_kstack + PAGE_SIZE);
1054 kprintf("DOUBLE FAULT\n");
1056 kprintf("\nFatal double fault\n");
1057 kprintf("rip = 0x%lx\n", frame->tf_rip);
1058 kprintf("rsp = 0x%lx\n", frame->tf_rsp);
1059 kprintf("rbp = 0x%lx\n", frame->tf_rbp);
1061 /* three separate prints in case of a trap on an unmapped page */
1062 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
1063 kprintf("lapic->id = %08x\n", lapic->id);
1065 panic("double fault");
1069 * syscall2 - MP aware system call request C handler
1071 * A system call is essentially treated as a trap except that the
1072 * MP lock is not held on entry or return. We are responsible for
1073 * obtaining the MP lock if necessary and for handling ASTs
1074 * (e.g. a task switch) prior to return.
1079 syscall2(struct trapframe *frame)
1081 struct thread *td = curthread;
1082 struct proc *p = td->td_proc;
1083 struct lwp *lp = td->td_lwp;
1085 struct sysent *callp;
1086 register_t orig_tf_rflags;
1091 int crit_count = td->td_critcount;
1094 int have_mplock = 0;
1099 union sysunion args;
1100 register_t *argsdst;
1102 mycpu->gd_cnt.v_syscall++;
1105 if (ISPL(frame->tf_cs) != SEL_UPL) {
1112 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid,
1115 userenter(td, p); /* lazy raise our priority */
1122 sticks = (int)td->td_sticks;
1123 orig_tf_rflags = frame->tf_rflags;
1126 * Virtual kernel intercept - if a VM context managed by a virtual
1127 * kernel issues a system call the virtual kernel handles it, not us.
1128 * Restore the virtual kernel context and return from its system
1129 * call. The current frame is copied out to the virtual kernel.
1131 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1132 vkernel_trap(lp, frame);
1133 error = EJUSTRETURN;
1138 * Get the system call parameters and account for time
1140 KASSERT(lp->lwp_md.md_regs == frame,
1141 ("Frame mismatch %p %p", lp->lwp_md.md_regs, frame));
1142 params = (caddr_t)frame->tf_rsp + sizeof(register_t);
1143 code = frame->tf_rax;
1145 if (p->p_sysent->sv_prepsyscall) {
1146 (*p->p_sysent->sv_prepsyscall)(
1147 frame, (int *)(&args.nosys.sysmsg + 1),
1150 if (code == SYS_syscall || code == SYS___syscall) {
1151 code = frame->tf_rdi;
1157 if (p->p_sysent->sv_mask)
1158 code &= p->p_sysent->sv_mask;
1160 if (code >= p->p_sysent->sv_size)
1161 callp = &p->p_sysent->sv_table[0];
1163 callp = &p->p_sysent->sv_table[code];
1165 narg = callp->sy_narg & SYF_ARGMASK;
1168 * On x86_64 we get up to six arguments in registers. The rest are
1169 * on the stack. The first six members of 'struct trapframe' happen
1170 * to be the registers used to pass arguments, in exactly the right
1173 argp = &frame->tf_rdi;
1175 argsdst = (register_t *)(&args.nosys.sysmsg + 1);
1177 * JG can we overflow the space pointed to by 'argsdst'
1178 * either with 'bcopy' or with 'copyin'?
1180 bcopy(argp, argsdst, sizeof(register_t) * regcnt);
1182 * copyin is MP aware, but the tracing code is not
1184 if (narg > regcnt) {
1185 KASSERT(params != NULL, ("copyin args with no params!"));
1186 error = copyin(params, &argsdst[regcnt],
1187 (narg - regcnt) * sizeof(register_t));
1190 if (KTRPOINT(td, KTR_SYSCALL)) {
1191 MAKEMPSAFE(have_mplock);
1193 ktrsyscall(lp, code, narg,
1194 (void *)(&args.nosys.sysmsg + 1));
1202 if (KTRPOINT(td, KTR_SYSCALL)) {
1203 MAKEMPSAFE(have_mplock);
1204 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1209 * Default return value is 0 (will be copied to %rax). Double-value
1210 * returns use %rax and %rdx. %rdx is left unchanged for system
1211 * calls which return only one result.
1213 args.sysmsg_fds[0] = 0;
1214 args.sysmsg_fds[1] = frame->tf_rdx;
1217 * The syscall might manipulate the trap frame. If it does it
1218 * will probably return EJUSTRETURN.
1220 args.sysmsg_frame = frame;
1222 STOPEVENT(p, S_SCE, narg); /* MP aware */
1225 * NOTE: All system calls run MPSAFE now. The system call itself
1226 * is responsible for getting the MP lock.
1228 #ifdef SYSCALL_DEBUG
1229 uint64_t tscval = rdtsc();
1231 error = (*callp->sy_call)(&args);
1232 #ifdef SYSCALL_DEBUG
1233 tscval = rdtsc() - tscval;
1234 tscval = tscval * 1000000 / tsc_frequency;
1235 if (SysCallsWorstCase[code] < tscval)
1236 SysCallsWorstCase[code] = tscval;
1241 * MP SAFE (we may or may not have the MP lock at this point)
1243 //kprintf("SYSMSG %d ", error);
1247 * Reinitialize proc pointer `p' as it may be different
1248 * if this is a child returning from fork syscall.
1251 lp = curthread->td_lwp;
1252 frame->tf_rax = args.sysmsg_fds[0];
1253 frame->tf_rdx = args.sysmsg_fds[1];
1254 frame->tf_rflags &= ~PSL_C;
1258 * Reconstruct pc, we know that 'syscall' is 2 bytes.
1259 * We have to do a full context restore so that %r10
1260 * (which was holding the value of %rcx) is restored for
1261 * the next iteration.
1263 if (frame->tf_err != 0 && frame->tf_err != 2)
1264 kprintf("lp %s:%d frame->tf_err is weird %ld\n",
1265 td->td_comm, lp->lwp_proc->p_pid, frame->tf_err);
1266 frame->tf_rip -= frame->tf_err;
1267 frame->tf_r10 = frame->tf_rcx;
1272 panic("Unexpected EASYNC return value (for now)");
1275 if (p->p_sysent->sv_errsize) {
1276 if (error >= p->p_sysent->sv_errsize)
1277 error = -1; /* XXX */
1279 error = p->p_sysent->sv_errtbl[error];
1281 frame->tf_rax = error;
1282 frame->tf_rflags |= PSL_C;
1287 * Traced syscall. trapsignal() is not MP aware.
1289 if (orig_tf_rflags & PSL_T) {
1290 MAKEMPSAFE(have_mplock);
1291 frame->tf_rflags &= ~PSL_T;
1292 trapsignal(lp, SIGTRAP, TRAP_TRACE);
1296 * Handle reschedule and other end-of-syscall issues
1298 userret(lp, frame, sticks);
1301 if (KTRPOINT(td, KTR_SYSRET)) {
1302 MAKEMPSAFE(have_mplock);
1303 ktrsysret(lp, code, error, args.sysmsg_result);
1308 * This works because errno is findable through the
1309 * register set. If we ever support an emulation where this
1310 * is not the case, this code will need to be revisited.
1312 STOPEVENT(p, S_SCX, code);
1317 * Release the MP lock if we had to get it
1322 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error);
1324 KASSERT(crit_count == td->td_critcount,
1325 ("syscall: critical section count mismatch! %d/%d",
1326 crit_count, td->td_pri));
1327 KASSERT(&td->td_toks_base == td->td_toks_stop,
1328 ("syscall: extra tokens held after trap! %ld",
1329 td->td_toks_stop - &td->td_toks_base));
1334 * NOTE: mplock not held at any point
1337 fork_return(struct lwp *lp, struct trapframe *frame)
1339 frame->tf_rax = 0; /* Child returns zero */
1340 frame->tf_rflags &= ~PSL_C; /* success */
1343 generic_lwp_return(lp, frame);
1344 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1348 * Simplified back end of syscall(), used when returning from fork()
1349 * directly into user mode.
1351 * This code will return back into the fork trampoline code which then
1354 * NOTE: The mplock is not held at any point.
1357 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1359 struct proc *p = lp->lwp_proc;
1362 * Newly forked processes are given a kernel priority. We have to
1363 * adjust the priority to a normal user priority and fake entry
1364 * into the kernel (call userenter()) to install a passive release
1365 * function just in case userret() decides to stop the process. This
1366 * can occur when ^Z races a fork. If we do not install the passive
1367 * release function the current process designation will not be
1368 * released when the thread goes to sleep.
1370 lwkt_setpri_self(TDPRI_USER_NORM);
1371 userenter(lp->lwp_thread, p);
1372 userret(lp, frame, 0);
1374 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1375 ktrsysret(lp, SYS_fork, 0, 0);
1377 lp->lwp_flags |= LWP_PASSIVE_ACQ;
1379 lp->lwp_flags &= ~LWP_PASSIVE_ACQ;
1383 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1384 * fault (which is then passed back to the virtual kernel) if an attempt is
1385 * made to use the FP unit.
1387 * XXX this is a fairly big hack.
1390 set_vkernel_fp(struct trapframe *frame)
1392 struct thread *td = curthread;
1394 if (frame->tf_xflags & PGEX_FPFAULT) {
1395 td->td_pcb->pcb_flags |= FP_VIRTFP;
1396 if (mdcpu->gd_npxthread == td)
1399 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1404 * Called from vkernel_trap() to fixup the vkernel's syscall
1405 * frame for vmspace_ctl() return.
1408 cpu_vkernel_trap(struct trapframe *frame, int error)
1410 frame->tf_rax = error;
1412 frame->tf_rflags |= PSL_C;
1414 frame->tf_rflags &= ~PSL_C;