2 * Copyright (C) 1994, David Greenman
3 * Copyright (c) 1990, 1993
4 * The Regents of the University of California. All rights reserved.
6 * This code is derived from software contributed to Berkeley by
7 * the University of Utah, and William Jolitz.
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. All advertising materials mentioning features or use of this software
18 * must display the following acknowledgement:
19 * This product includes software developed by the University of
20 * California, Berkeley and its contributors.
21 * 4. Neither the name of the University nor the names of its contributors
22 * may be used to endorse or promote products derived from this software
23 * without specific prior written permission.
25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 * from: @(#)trap.c 7.4 (Berkeley) 5/13/91
38 * $FreeBSD: src/sys/i386/i386/trap.c,v 1.147.2.11 2003/02/27 19:09:59 luoqi Exp $
42 * x86_64 Trap and System call handling
48 #include "opt_ktrace.h"
50 #include <sys/param.h>
51 #include <sys/systm.h>
53 #include <sys/pioctl.h>
54 #include <sys/kernel.h>
55 #include <sys/resourcevar.h>
56 #include <sys/signalvar.h>
57 #include <sys/signal2.h>
58 #include <sys/syscall.h>
59 #include <sys/sysctl.h>
60 #include <sys/sysent.h>
62 #include <sys/vmmeter.h>
63 #include <sys/malloc.h>
65 #include <sys/ktrace.h>
68 #include <sys/upcall.h>
69 #include <sys/vkernel.h>
70 #include <sys/sysproto.h>
71 #include <sys/sysunion.h>
72 #include <sys/vmspace.h>
75 #include <vm/vm_param.h>
78 #include <vm/vm_kern.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_page.h>
81 #include <vm/vm_extern.h>
83 #include <machine/cpu.h>
84 #include <machine/md_var.h>
85 #include <machine/pcb.h>
86 #include <machine/smp.h>
87 #include <machine/tss.h>
88 #include <machine/globaldata.h>
92 #include <sys/msgport2.h>
93 #include <sys/thread2.h>
94 #include <sys/mplock2.h>
98 #define MAKEMPSAFE(have_mplock) \
99 if (have_mplock == 0) { \
106 #define MAKEMPSAFE(have_mplock)
110 int (*pmath_emulate) (struct trapframe *);
112 extern int trapwrite (unsigned addr);
114 static int trap_pfault (struct trapframe *, int, vm_offset_t);
115 static void trap_fatal (struct trapframe *, int, vm_offset_t);
116 void dblfault_handler (void);
119 extern inthand_t IDTVEC(syscall);
122 #define MAX_TRAP_MSG 30
123 static char *trap_msg[] = {
125 "privileged instruction fault", /* 1 T_PRIVINFLT */
127 "breakpoint instruction fault", /* 3 T_BPTFLT */
130 "arithmetic trap", /* 6 T_ARITHTRAP */
131 "system forced exception", /* 7 T_ASTFLT */
133 "general protection fault", /* 9 T_PROTFLT */
134 "trace trap", /* 10 T_TRCTRAP */
136 "page fault", /* 12 T_PAGEFLT */
138 "alignment fault", /* 14 T_ALIGNFLT */
142 "integer divide fault", /* 18 T_DIVIDE */
143 "non-maskable interrupt trap", /* 19 T_NMI */
144 "overflow trap", /* 20 T_OFLOW */
145 "FPU bounds check fault", /* 21 T_BOUND */
146 "FPU device not available", /* 22 T_DNA */
147 "double fault", /* 23 T_DOUBLEFLT */
148 "FPU operand fetch fault", /* 24 T_FPOPFLT */
149 "invalid TSS fault", /* 25 T_TSSFLT */
150 "segment not present fault", /* 26 T_SEGNPFLT */
151 "stack fault", /* 27 T_STKFLT */
152 "machine check trap", /* 28 T_MCHK */
153 "SIMD floating-point exception", /* 29 T_XMMFLT */
154 "reserved (unknown) fault", /* 30 T_RESERVED */
158 static int ddb_on_nmi = 1;
159 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
160 &ddb_on_nmi, 0, "Go to DDB on NMI");
162 static int panic_on_nmi = 1;
163 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
164 &panic_on_nmi, 0, "Panic on NMI");
165 static int fast_release;
166 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
167 &fast_release, 0, "Passive Release was optimal");
168 static int slow_release;
169 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
170 &slow_release, 0, "Passive Release was nonoptimal");
172 MALLOC_DEFINE(M_SYSMSG, "sysmsg", "sysmsg structure");
173 extern int max_sysmsg;
176 * Passively intercepts the thread switch function to increase
177 * the thread priority from a user priority to a kernel priority, reducing
178 * syscall and trap overhead for the case where no switch occurs.
180 * Synchronizes td_ucred with p_ucred. This is used by system calls,
181 * signal handling, faults, AST traps, and anything else that enters the
182 * kernel from userland and provides the kernel with a stable read-only
183 * copy of the process ucred.
186 userenter(struct thread *curtd, struct proc *curp)
191 curtd->td_release = lwkt_passive_release;
193 if (curtd->td_ucred != curp->p_ucred) {
194 ncred = crhold(curp->p_ucred);
195 ocred = curtd->td_ucred;
196 curtd->td_ucred = ncred;
203 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
204 * must be completed before we can return to or try to return to userland.
206 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
207 * arithmatic on the delta calculation so the absolute tick values are
208 * truncated to an integer.
211 userret(struct lwp *lp, struct trapframe *frame, int sticks)
213 struct proc *p = lp->lwp_proc;
217 * Charge system time if profiling. Note: times are in microseconds.
218 * This may do a copyout and block, so do it first even though it
219 * means some system time will be charged as user time.
221 if (p->p_flags & P_PROFIL) {
222 addupc_task(p, frame->tf_rip,
223 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
228 * If the jungle wants us dead, so be it.
230 if (lp->lwp_mpflags & LWP_MP_WEXIT) {
231 lwkt_gettoken(&p->p_token);
233 lwkt_reltoken(&p->p_token); /* NOT REACHED */
237 * Block here if we are in a stopped state.
239 if (p->p_stat == SSTOP) {
240 lwkt_gettoken(&p->p_token);
242 lwkt_reltoken(&p->p_token);
247 * Post any pending upcalls. If running a virtual kernel be sure
248 * to restore the virtual kernel's vmspace before posting the upcall.
250 if (p->p_flags & (P_SIGVTALRM | P_SIGPROF | P_UPCALLPEND)) {
251 lwkt_gettoken(&p->p_token);
252 if (p->p_flags & P_SIGVTALRM) {
253 p->p_flags &= ~P_SIGVTALRM;
254 ksignal(p, SIGVTALRM);
256 if (p->p_flags & P_SIGPROF) {
257 p->p_flags &= ~P_SIGPROF;
260 if (p->p_flags & P_UPCALLPEND) {
261 p->p_flags &= ~P_UPCALLPEND;
264 lwkt_reltoken(&p->p_token);
269 * Post any pending signals
271 * WARNING! postsig() can exit and not return.
273 if ((sig = CURSIG_TRACE(lp)) != 0) {
274 lwkt_gettoken(&p->p_token);
276 lwkt_reltoken(&p->p_token);
281 * block here if we are swapped out, but still process signals
282 * (such as SIGKILL). proc0 (the swapin scheduler) is already
283 * aware of our situation, we do not have to wake it up.
285 if (p->p_flags & P_SWAPPEDOUT) {
286 lwkt_gettoken(&p->p_token);
288 p->p_flags |= P_SWAPWAIT;
290 if (p->p_flags & P_SWAPWAIT)
291 tsleep(p, PCATCH, "SWOUT", 0);
292 p->p_flags &= ~P_SWAPWAIT;
294 lwkt_reltoken(&p->p_token);
299 * In a multi-threaded program it is possible for a thread to change
300 * signal state during a system call which temporarily changes the
301 * signal mask. In this case postsig() might not be run and we
302 * have to restore the mask ourselves.
304 if (lp->lwp_flags & LWP_OLDMASK) {
305 lp->lwp_flags &= ~LWP_OLDMASK;
306 lp->lwp_sigmask = lp->lwp_oldsigmask;
312 * Cleanup from userenter and any passive release that might have occured.
313 * We must reclaim the current-process designation before we can return
314 * to usermode. We also handle both LWKT and USER reschedule requests.
317 userexit(struct lwp *lp)
319 struct thread *td = lp->lwp_thread;
320 /* globaldata_t gd = td->td_gd; */
323 * Handle stop requests at kernel priority. Any requests queued
324 * after this loop will generate another AST.
326 while (lp->lwp_proc->p_stat == SSTOP) {
327 lwkt_gettoken(&lp->lwp_proc->p_token);
329 lwkt_reltoken(&lp->lwp_proc->p_token);
333 * Reduce our priority in preparation for a return to userland. If
334 * our passive release function was still in place, our priority was
335 * never raised and does not need to be reduced.
337 lwkt_passive_recover(td);
340 * Become the current user scheduled process if we aren't already,
341 * and deal with reschedule requests and other factors.
343 lp->lwp_proc->p_usched->acquire_curproc(lp);
344 /* WARNING: we may have migrated cpu's */
345 /* gd = td->td_gd; */
348 #if !defined(KTR_KERNENTRY)
349 #define KTR_KERNENTRY KTR_ALL
351 KTR_INFO_MASTER(kernentry);
352 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0,
353 "TRAP(pid %hd, tid %hd, trapno %ld, eva %lu)",
354 pid_t pid, lwpid_t tid, register_t trapno, vm_offset_t eva);
355 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "TRAP_RET(pid %hd, tid %hd)",
356 pid_t pid, lwpid_t tid);
357 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "SYSC(pid %hd, tid %hd, nr %ld)",
358 pid_t pid, lwpid_t tid, register_t trapno);
359 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "SYSRET(pid %hd, tid %hd, err %d)",
360 pid_t pid, lwpid_t tid, int err);
361 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "FORKRET(pid %hd, tid %hd)",
362 pid_t pid, lwpid_t tid);
365 * Exception, fault, and trap interface to the kernel.
366 * This common code is called from assembly language IDT gate entry
367 * routines that prepare a suitable stack frame, and restore this
368 * frame after the exception has been processed.
370 * This function is also called from doreti in an interlock to handle ASTs.
371 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
373 * NOTE! We have to retrieve the fault address prior to obtaining the
374 * MP lock because get_mplock() may switch out. YYY cr2 really ought
375 * to be retrieved by the assembly code, not here.
377 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
378 * if an attempt is made to switch from a fast interrupt or IPI. This is
379 * necessary to properly take fatal kernel traps on SMP machines if
380 * get_mplock() has to block.
384 user_trap(struct trapframe *frame)
386 struct globaldata *gd = mycpu;
387 struct thread *td = gd->gd_curthread;
388 struct lwp *lp = td->td_lwp;
391 int i = 0, ucode = 0, type, code;
396 int crit_count = td->td_critcount;
397 lwkt_tokref_t curstop = td->td_toks_stop;
403 if (frame->tf_trapno == T_PAGEFLT)
404 eva = frame->tf_addr;
408 kprintf("USER_TRAP AT %08lx xflags %ld trapno %ld eva %08lx\n",
409 frame->tf_rip, frame->tf_xflags, frame->tf_trapno, eva);
413 * Everything coming from user mode runs through user_trap,
414 * including system calls.
416 if (frame->tf_trapno == T_FAST_SYSCALL) {
421 KTR_LOG(kernentry_trap, lp->lwp_proc->p_pid, lp->lwp_tid,
422 frame->tf_trapno, eva);
426 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
427 ++gd->gd_trap_nesting_level;
428 MAKEMPSAFE(have_mplock);
429 trap_fatal(frame, TRUE, eva);
430 --gd->gd_trap_nesting_level;
435 type = frame->tf_trapno;
436 code = frame->tf_err;
440 sticks = (int)td->td_sticks;
441 lp->lwp_md.md_regs = frame;
444 case T_PRIVINFLT: /* privileged instruction fault */
449 case T_BPTFLT: /* bpt instruction fault */
450 case T_TRCTRAP: /* trace trap */
451 frame->tf_rflags &= ~PSL_T;
453 ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
456 case T_ARITHTRAP: /* arithmetic trap */
461 case T_ASTFLT: /* Allow process switch */
462 mycpu->gd_cnt.v_soft++;
463 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
464 atomic_clear_int(&mycpu->gd_reqflags, RQF_AST_OWEUPC);
465 addupc_task(p, p->p_prof.pr_addr, p->p_prof.pr_ticks);
470 * The following two traps can happen in
471 * vm86 mode, and, if so, we want to handle
474 case T_PROTFLT: /* general protection fault */
475 case T_STKFLT: /* stack fault */
477 if (frame->tf_eflags & PSL_VM) {
478 i = vm86_emulate((struct vm86frame *)frame);
486 case T_SEGNPFLT: /* segment not present fault */
487 case T_TSSFLT: /* invalid TSS fault */
488 case T_DOUBLEFLT: /* double fault */
491 ucode = code + BUS_SEGM_FAULT ;
494 case T_PAGEFLT: /* page fault */
495 MAKEMPSAFE(have_mplock);
496 i = trap_pfault(frame, TRUE, eva);
497 if (i == -1 || i == 0)
509 case T_DIVIDE: /* integer divide fault */
516 MAKEMPSAFE(have_mplock);
517 /* machine/parity/power fail/"kitchen sink" faults */
518 if (isa_nmi(code) == 0) {
521 * NMI can be hooked up to a pushbutton
525 kprintf ("NMI ... going to debugger\n");
526 kdb_trap(type, 0, frame);
530 } else if (panic_on_nmi)
531 panic("NMI indicates hardware failure");
533 #endif /* NISA > 0 */
535 case T_OFLOW: /* integer overflow fault */
540 case T_BOUND: /* bounds check fault */
547 * Virtual kernel intercept - pass the DNA exception
548 * to the (emulated) virtual kernel if it asked to handle
549 * it. This occurs when the virtual kernel is holding
550 * onto the FP context for a different emulated
551 * process then the one currently running.
553 * We must still call npxdna() since we may have
554 * saved FP state that the (emulated) virtual kernel
555 * needs to hand over to a different emulated process.
557 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
558 (td->td_pcb->pcb_flags & FP_VIRTFP)
565 * The kernel may have switched out the FP unit's
566 * state, causing the user process to take a fault
567 * when it tries to use the FP unit. Restore the
572 if (!pmath_emulate) {
574 ucode = FPE_FPU_NP_TRAP;
577 i = (*pmath_emulate)(frame);
579 if (!(frame->tf_rflags & PSL_T))
581 frame->tf_rflags &= ~PSL_T;
584 /* else ucode = emulator_only_knows() XXX */
587 case T_FPOPFLT: /* FPU operand fetch fault */
592 case T_XMMFLT: /* SIMD floating-point exception */
599 * Virtual kernel intercept - if the fault is directly related to a
600 * VM context managed by a virtual kernel then let the virtual kernel
603 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
604 vkernel_trap(lp, frame);
609 * Translate fault for emulators (e.g. Linux)
611 if (*p->p_sysent->sv_transtrap)
612 i = (*p->p_sysent->sv_transtrap)(i, type);
614 MAKEMPSAFE(have_mplock);
615 trapsignal(lp, i, ucode);
618 if (type <= MAX_TRAP_MSG) {
619 uprintf("fatal process exception: %s",
621 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
622 uprintf(", fault VA = 0x%lx", (u_long)eva);
628 userret(lp, frame, sticks);
635 KTR_LOG(kernentry_trap_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
637 KASSERT(crit_count == td->td_critcount,
638 ("trap: critical section count mismatch! %d/%d",
639 crit_count, td->td_pri));
640 KASSERT(curstop == td->td_toks_stop,
641 ("trap: extra tokens held after trap! %ld/%ld",
642 curstop - &td->td_toks_base,
643 td->td_toks_stop - &td->td_toks_base));
648 kern_trap(struct trapframe *frame)
650 struct globaldata *gd = mycpu;
651 struct thread *td = gd->gd_curthread;
654 int i = 0, ucode = 0, type, code;
659 int crit_count = td->td_critcount;
660 lwkt_tokref_t curstop = td->td_toks_stop;
667 if (frame->tf_trapno == T_PAGEFLT)
668 eva = frame->tf_addr;
674 ++gd->gd_trap_nesting_level;
675 MAKEMPSAFE(have_mplock);
676 trap_fatal(frame, FALSE, eva);
677 --gd->gd_trap_nesting_level;
682 type = frame->tf_trapno;
683 code = frame->tf_err;
691 case T_PAGEFLT: /* page fault */
692 MAKEMPSAFE(have_mplock);
693 trap_pfault(frame, FALSE, eva);
698 * The kernel may be using npx for copying or other
701 panic("kernel NPX should not happen");
706 case T_PROTFLT: /* general protection fault */
707 case T_SEGNPFLT: /* segment not present fault */
709 * Invalid segment selectors and out of bounds
710 * %eip's and %esp's can be set up in user mode.
711 * This causes a fault in kernel mode when the
712 * kernel tries to return to user mode. We want
713 * to get this fault so that we can fix the
714 * problem here and not have to check all the
715 * selectors and pointers when the user changes
718 if (mycpu->gd_intr_nesting_level == 0) {
719 if (td->td_pcb->pcb_onfault) {
721 (register_t)td->td_pcb->pcb_onfault;
729 * PSL_NT can be set in user mode and isn't cleared
730 * automatically when the kernel is entered. This
731 * causes a TSS fault when the kernel attempts to
732 * `iret' because the TSS link is uninitialized. We
733 * want to get this fault so that we can fix the
734 * problem here and not every time the kernel is
737 if (frame->tf_rflags & PSL_NT) {
738 frame->tf_rflags &= ~PSL_NT;
743 case T_TRCTRAP: /* trace trap */
745 if (frame->tf_eip == (int)IDTVEC(syscall)) {
747 * We've just entered system mode via the
748 * syscall lcall. Continue single stepping
749 * silently until the syscall handler has
754 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
756 * The syscall handler has now saved the
757 * flags. Stop single stepping it.
759 frame->tf_eflags &= ~PSL_T;
765 * Ignore debug register trace traps due to
766 * accesses in the user's address space, which
767 * can happen under several conditions such as
768 * if a user sets a watchpoint on a buffer and
769 * then passes that buffer to a system call.
770 * We still want to get TRCTRAPS for addresses
771 * in kernel space because that is useful when
772 * debugging the kernel.
774 if (user_dbreg_trap()) {
776 * Reset breakpoint bits because the
779 load_dr6(rdr6() & 0xfffffff0);
784 * Fall through (TRCTRAP kernel mode, kernel address)
788 * If DDB is enabled, let it handle the debugger trap.
789 * Otherwise, debugger traps "can't happen".
792 MAKEMPSAFE(have_mplock);
793 if (kdb_trap (type, 0, frame))
798 MAKEMPSAFE(have_mplock);
799 trap_fatal(frame, FALSE, eva);
802 MAKEMPSAFE(have_mplock);
803 trap_fatal(frame, FALSE, eva);
808 * Ignore this trap generated from a spurious SIGTRAP.
810 * single stepping in / syscalls leads to spurious / SIGTRAP
813 * Haiku (c) 2007 Simon 'corecode' Schubert
819 * Translate fault for emulators (e.g. Linux)
821 if (*p->p_sysent->sv_transtrap)
822 i = (*p->p_sysent->sv_transtrap)(i, type);
824 MAKEMPSAFE(have_mplock);
825 trapsignal(lp, i, ucode);
828 if (type <= MAX_TRAP_MSG) {
829 uprintf("fatal process exception: %s",
831 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
832 uprintf(", fault VA = 0x%lx", (u_long)eva);
844 KASSERT(crit_count == td->td_critcount,
845 ("trap: critical section count mismatch! %d/%d",
846 crit_count, td->td_pri));
847 KASSERT(curstop == td->td_toks_stop,
848 ("trap: extra tokens held after trap! %ld/%ld",
849 curstop - &td->td_toks_base,
850 td->td_toks_stop - &td->td_toks_base));
855 trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
858 struct vmspace *vm = NULL;
862 thread_t td = curthread;
863 struct lwp *lp = td->td_lwp;
866 va = trunc_page(eva);
867 if (usermode == FALSE) {
869 * This is a fault on kernel virtual memory.
874 * This is a fault on non-kernel virtual memory.
875 * vm is initialized above to NULL. If curproc is NULL
876 * or curproc->p_vmspace is NULL the fault is fatal.
879 vm = lp->lwp_vmspace;
887 if (frame->tf_err & PGEX_W)
888 ftype = VM_PROT_READ | VM_PROT_WRITE;
890 ftype = VM_PROT_READ;
892 if (map != &kernel_map) {
894 * Keep swapout from messing with us during this
900 * Grow the stack if necessary
902 /* grow_stack returns false only if va falls into
903 * a growable stack region and the stack growth
904 * fails. It returns true if va was not within
905 * a growable stack region, or if the stack
908 if (!grow_stack (lp->lwp_proc, va)) {
916 fault_flags |= VM_FAULT_BURST;
917 if (ftype & VM_PROT_WRITE)
918 fault_flags |= VM_FAULT_DIRTY;
920 fault_flags |= VM_FAULT_NORMAL;
921 rv = vm_fault(map, va, ftype, fault_flags);
926 * Don't have to worry about process locking or stacks in the kernel.
928 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
931 if (rv == KERN_SUCCESS)
935 if (td->td_gd->gd_intr_nesting_level == 0 &&
936 td->td_pcb->pcb_onfault) {
937 frame->tf_rip = (register_t)td->td_pcb->pcb_onfault;
940 trap_fatal(frame, usermode, eva);
945 * NOTE: on x86_64 we have a tf_addr field in the trapframe, no
946 * kludge is needed to pass the fault address to signal handlers.
948 struct proc *p = td->td_proc;
949 kprintf("seg-fault accessing address %p rip=%p pid=%d p_comm=%s\n",
950 (void *)va, (void *)frame->tf_rip, p->p_pid, p->p_comm);
951 /* Debugger("seg-fault"); */
953 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
957 trap_fatal(struct trapframe *frame, int usermode, vm_offset_t eva)
962 code = frame->tf_xflags;
963 type = frame->tf_trapno;
965 if (type <= MAX_TRAP_MSG) {
966 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
967 type, trap_msg[type],
968 (usermode ? "user" : "kernel"));
971 /* two separate prints in case of a trap on an unmapped page */
972 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
974 if (type == T_PAGEFLT) {
975 kprintf("fault virtual address = %p\n", (void *)eva);
976 kprintf("fault code = %s %s, %s\n",
977 usermode ? "user" : "supervisor",
978 code & PGEX_W ? "write" : "read",
979 code & PGEX_P ? "protection violation" : "page not present");
981 kprintf("instruction pointer = 0x%lx:0x%lx\n",
982 frame->tf_cs & 0xffff, frame->tf_rip);
984 ss = frame->tf_ss & 0xffff;
987 ss = GSEL(GDATA_SEL, SEL_KPL);
988 rsp = (long)&frame->tf_rsp;
990 kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp);
991 kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp);
992 kprintf("processor eflags = ");
993 if (frame->tf_rflags & PSL_T)
994 kprintf("trace trap, ");
995 if (frame->tf_rflags & PSL_I)
996 kprintf("interrupt enabled, ");
997 if (frame->tf_rflags & PSL_NT)
998 kprintf("nested task, ");
999 if (frame->tf_rflags & PSL_RF)
1000 kprintf("resume, ");
1002 if (frame->tf_eflags & PSL_VM)
1005 kprintf("IOPL = %jd\n", (intmax_t)((frame->tf_rflags & PSL_IOPL) >> 12));
1006 kprintf("current process = ");
1008 kprintf("%lu (%s)\n",
1009 (u_long)curproc->p_pid, curproc->p_comm ?
1010 curproc->p_comm : "");
1014 kprintf("current thread = pri %d ", curthread->td_pri);
1015 if (curthread->td_critcount)
1021 * we probably SHOULD have stopped the other CPUs before now!
1022 * another CPU COULD have been touching cpl at this moment...
1024 kprintf(" <- SMP: XXX");
1033 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1036 kprintf("trap number = %d\n", type);
1037 if (type <= MAX_TRAP_MSG)
1038 panic("%s", trap_msg[type]);
1040 panic("unknown/reserved trap");
1044 * Double fault handler. Called when a fault occurs while writing
1045 * a frame for a trap/exception onto the stack. This usually occurs
1046 * when the stack overflows (such is the case with infinite recursion,
1049 * XXX Note that the current PTD gets replaced by IdlePTD when the
1050 * task switch occurs. This means that the stack that was active at
1051 * the time of the double fault is not available at <kstack> unless
1052 * the machine was idle when the double fault occurred. The downside
1053 * of this is that "trace <ebp>" in ddb won't work.
1056 dblfault_handler(void)
1059 struct mdglobaldata *gd = mdcpu;
1062 kprintf("\nFatal double fault:\n");
1064 kprintf("rip = 0x%lx\n", gd->gd_common_tss.tss_rip);
1065 kprintf("rsp = 0x%lx\n", gd->gd_common_tss.tss_rsp);
1066 kprintf("rbp = 0x%lx\n", gd->gd_common_tss.tss_rbp);
1069 /* two separate prints in case of a trap on an unmapped page */
1070 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
1072 panic("double fault");
1076 * Compensate for 386 brain damage (missing URKR).
1077 * This is a little simpler than the pagefault handler in trap() because
1078 * it the page tables have already been faulted in and high addresses
1079 * are thrown out early for other reasons.
1082 trapwrite(unsigned addr)
1089 va = trunc_page((vm_offset_t)addr);
1091 * XXX - MAX is END. Changed > to >= for temp. fix.
1093 if (va >= VM_MAX_USER_ADDRESS)
1096 lp = curthread->td_lwp;
1097 vm = lp->lwp_vmspace;
1099 PHOLD(lp->lwp_proc);
1101 if (!grow_stack (lp->lwp_proc, va)) {
1102 PRELE(lp->lwp_proc);
1107 * fault the data page
1109 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY);
1111 PRELE(lp->lwp_proc);
1113 if (rv != KERN_SUCCESS)
1120 * syscall2 - MP aware system call request C handler
1122 * A system call is essentially treated as a trap except that the
1123 * MP lock is not held on entry or return. We are responsible for
1124 * obtaining the MP lock if necessary and for handling ASTs
1125 * (e.g. a task switch) prior to return.
1127 * In general, only simple access and manipulation of curproc and
1128 * the current stack is allowed without having to hold MP lock.
1130 * MPSAFE - note that large sections of this routine are run without
1134 syscall2(struct trapframe *frame)
1136 struct thread *td = curthread;
1137 struct proc *p = td->td_proc;
1138 struct lwp *lp = td->td_lwp;
1140 struct sysent *callp;
1141 register_t orig_tf_rflags;
1146 int crit_count = td->td_critcount;
1147 lwkt_tokref_t curstop = td->td_toks_stop;
1150 int have_mplock = 0;
1155 union sysunion args;
1156 register_t *argsdst;
1158 mycpu->gd_cnt.v_syscall++;
1160 KTR_LOG(kernentry_syscall, lp->lwp_proc->p_pid, lp->lwp_tid,
1163 userenter(td, p); /* lazy raise our priority */
1170 sticks = (int)td->td_sticks;
1171 orig_tf_rflags = frame->tf_rflags;
1174 * Virtual kernel intercept - if a VM context managed by a virtual
1175 * kernel issues a system call the virtual kernel handles it, not us.
1176 * Restore the virtual kernel context and return from its system
1177 * call. The current frame is copied out to the virtual kernel.
1179 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1180 vkernel_trap(lp, frame);
1181 error = EJUSTRETURN;
1186 * Get the system call parameters and account for time
1188 lp->lwp_md.md_regs = frame;
1189 params = (caddr_t)frame->tf_rsp + sizeof(register_t);
1190 code = frame->tf_rax;
1192 if (p->p_sysent->sv_prepsyscall) {
1193 (*p->p_sysent->sv_prepsyscall)(
1194 frame, (int *)(&args.nosys.sysmsg + 1),
1197 if (code == SYS_syscall || code == SYS___syscall) {
1198 code = frame->tf_rdi;
1204 if (p->p_sysent->sv_mask)
1205 code &= p->p_sysent->sv_mask;
1207 if (code >= p->p_sysent->sv_size)
1208 callp = &p->p_sysent->sv_table[0];
1210 callp = &p->p_sysent->sv_table[code];
1212 narg = callp->sy_narg & SYF_ARGMASK;
1215 * On x86_64 we get up to six arguments in registers. The rest are
1216 * on the stack. The first six members of 'struct trapframe' happen
1217 * to be the registers used to pass arguments, in exactly the right
1220 argp = &frame->tf_rdi;
1222 argsdst = (register_t *)(&args.nosys.sysmsg + 1);
1224 * JG can we overflow the space pointed to by 'argsdst'
1225 * either with 'bcopy' or with 'copyin'?
1227 bcopy(argp, argsdst, sizeof(register_t) * regcnt);
1229 * copyin is MP aware, but the tracing code is not
1231 if (narg > regcnt) {
1232 KASSERT(params != NULL, ("copyin args with no params!"));
1233 error = copyin(params, &argsdst[regcnt],
1234 (narg - regcnt) * sizeof(register_t));
1237 if (KTRPOINT(td, KTR_SYSCALL)) {
1238 MAKEMPSAFE(have_mplock);
1240 ktrsyscall(lp, code, narg,
1241 (void *)(&args.nosys.sysmsg + 1));
1249 if (KTRPOINT(td, KTR_SYSCALL)) {
1250 MAKEMPSAFE(have_mplock);
1251 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1256 * Default return value is 0 (will be copied to %rax). Double-value
1257 * returns use %rax and %rdx. %rdx is left unchanged for system
1258 * calls which return only one result.
1260 args.sysmsg_fds[0] = 0;
1261 args.sysmsg_fds[1] = frame->tf_rdx;
1264 * The syscall might manipulate the trap frame. If it does it
1265 * will probably return EJUSTRETURN.
1267 args.sysmsg_frame = frame;
1269 STOPEVENT(p, S_SCE, narg); /* MP aware */
1272 * NOTE: All system calls run MPSAFE now. The system call itself
1273 * is responsible for getting the MP lock.
1275 error = (*callp->sy_call)(&args);
1278 kprintf("system call %d returned %d\n", code, error);
1283 * MP SAFE (we may or may not have the MP lock at this point)
1288 * Reinitialize proc pointer `p' as it may be different
1289 * if this is a child returning from fork syscall.
1292 lp = curthread->td_lwp;
1293 frame->tf_rax = args.sysmsg_fds[0];
1294 frame->tf_rdx = args.sysmsg_fds[1];
1295 frame->tf_rflags &= ~PSL_C;
1299 * Reconstruct pc, we know that 'syscall' is 2 bytes.
1300 * We have to do a full context restore so that %r10
1301 * (which was holding the value of %rcx) is restored for
1302 * the next iteration.
1304 frame->tf_rip -= frame->tf_err;
1305 frame->tf_r10 = frame->tf_rcx;
1310 panic("Unexpected EASYNC return value (for now)");
1313 if (p->p_sysent->sv_errsize) {
1314 if (error >= p->p_sysent->sv_errsize)
1315 error = -1; /* XXX */
1317 error = p->p_sysent->sv_errtbl[error];
1319 frame->tf_rax = error;
1320 frame->tf_rflags |= PSL_C;
1325 * Traced syscall. trapsignal() is not MP aware.
1327 if (orig_tf_rflags & PSL_T) {
1328 MAKEMPSAFE(have_mplock);
1329 frame->tf_rflags &= ~PSL_T;
1330 trapsignal(lp, SIGTRAP, 0);
1334 * Handle reschedule and other end-of-syscall issues
1336 userret(lp, frame, sticks);
1339 if (KTRPOINT(td, KTR_SYSRET)) {
1340 MAKEMPSAFE(have_mplock);
1341 ktrsysret(lp, code, error, args.sysmsg_result);
1346 * This works because errno is findable through the
1347 * register set. If we ever support an emulation where this
1348 * is not the case, this code will need to be revisited.
1350 STOPEVENT(p, S_SCX, code);
1355 * Release the MP lock if we had to get it
1360 KTR_LOG(kernentry_syscall_ret, lp->lwp_proc->p_pid, lp->lwp_tid, error);
1362 KASSERT(&td->td_toks_base == td->td_toks_stop,
1363 ("syscall: critical section count mismatch! %d/%d",
1364 crit_count, td->td_pri));
1365 KASSERT(curstop == td->td_toks_stop,
1366 ("syscall: extra tokens held after trap! %ld",
1367 td->td_toks_stop - &td->td_toks_base));
1372 * NOTE: mplock not held at any point
1375 fork_return(struct lwp *lp, struct trapframe *frame)
1377 frame->tf_rax = 0; /* Child returns zero */
1378 frame->tf_rflags &= ~PSL_C; /* success */
1381 generic_lwp_return(lp, frame);
1382 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1386 * Simplified back end of syscall(), used when returning from fork()
1387 * directly into user mode.
1389 * This code will return back into the fork trampoline code which then
1392 * NOTE: The mplock is not held at any point.
1395 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1397 struct proc *p = lp->lwp_proc;
1400 * Newly forked processes are given a kernel priority. We have to
1401 * adjust the priority to a normal user priority and fake entry
1402 * into the kernel (call userenter()) to install a passive release
1403 * function just in case userret() decides to stop the process. This
1404 * can occur when ^Z races a fork. If we do not install the passive
1405 * release function the current process designation will not be
1406 * released when the thread goes to sleep.
1408 lwkt_setpri_self(TDPRI_USER_NORM);
1409 userenter(lp->lwp_thread, p);
1410 userret(lp, frame, 0);
1412 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1413 ktrsysret(lp, SYS_fork, 0, 0);
1415 lp->lwp_flags |= LWP_PASSIVE_ACQ;
1417 lp->lwp_flags &= ~LWP_PASSIVE_ACQ;
1421 * doreti has turned into this. The frame is directly on the stack. We
1422 * pull everything else we need (fpu and tls context) from the current
1425 * Note on fpu interactions: In a virtual kernel, the fpu context for
1426 * an emulated user mode process is not shared with the virtual kernel's
1427 * fpu context, so we only have to 'stack' fpu contexts within the virtual
1428 * kernel itself, and not even then since the signal() contexts that we care
1429 * about save and restore the FPU state (I think anyhow).
1431 * vmspace_ctl() returns an error only if it had problems instaling the
1432 * context we supplied or problems copying data to/from our VM space.
1435 go_user(struct intrframe *frame)
1437 struct trapframe *tf = (void *)&frame->if_rdi;
1441 * Interrupts may be disabled on entry, make sure all signals
1442 * can be received before beginning our loop.
1447 * Switch to the current simulated user process, then call
1448 * user_trap() when we break out of it (usually due to a signal).
1452 * Tell the real kernel whether it is ok to use the FP
1455 if (mdcpu->gd_npxthread == curthread) {
1456 tf->tf_xflags &= ~PGEX_FPFAULT;
1458 tf->tf_xflags |= PGEX_FPFAULT;
1462 * Run emulated user process context. This call interlocks
1463 * with new mailbox signals.
1465 * Set PGEX_U unconditionally, indicating a user frame (the
1466 * bit is normally set only by T_PAGEFLT).
1468 r = vmspace_ctl(&curproc->p_vmspace->vm_pmap, VMSPACE_CTL_RUN,
1469 tf, &curthread->td_savevext);
1470 frame->if_xflags |= PGEX_U;
1472 kprintf("GO USER %d trap %ld EVA %08lx RIP %08lx RSP %08lx XFLAGS %02lx/%02lx\n",
1473 r, tf->tf_trapno, tf->tf_addr, tf->tf_rip, tf->tf_rsp,
1474 tf->tf_xflags, frame->if_xflags);
1478 panic("vmspace_ctl failed error %d", errno);
1480 if (tf->tf_trapno) {
1484 if (mycpu->gd_reqflags & RQF_AST_MASK) {
1485 tf->tf_trapno = T_ASTFLT;
1493 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1494 * fault (which is then passed back to the virtual kernel) if an attempt is
1495 * made to use the FP unit.
1497 * XXX this is a fairly big hack.
1500 set_vkernel_fp(struct trapframe *frame)
1502 struct thread *td = curthread;
1504 if (frame->tf_xflags & PGEX_FPFAULT) {
1505 td->td_pcb->pcb_flags |= FP_VIRTFP;
1506 if (mdcpu->gd_npxthread == td)
1509 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1514 * Called from vkernel_trap() to fixup the vkernel's syscall
1515 * frame for vmspace_ctl() return.
1518 cpu_vkernel_trap(struct trapframe *frame, int error)
1520 frame->tf_rax = error;
1522 frame->tf_rflags |= PSL_C;
1524 frame->tf_rflags &= ~PSL_C;