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 * Make sure postsig() handled request to restore old signal mask after
300 * running signal handler.
302 KKASSERT((lp->lwp_flags & LWP_OLDMASK) == 0);
306 * Cleanup from userenter and any passive release that might have occured.
307 * We must reclaim the current-process designation before we can return
308 * to usermode. We also handle both LWKT and USER reschedule requests.
311 userexit(struct lwp *lp)
313 struct thread *td = lp->lwp_thread;
314 /* globaldata_t gd = td->td_gd; */
317 * Handle stop requests at kernel priority. Any requests queued
318 * after this loop will generate another AST.
320 while (lp->lwp_proc->p_stat == SSTOP) {
321 lwkt_gettoken(&lp->lwp_proc->p_token);
323 lwkt_reltoken(&lp->lwp_proc->p_token);
327 * Reduce our priority in preparation for a return to userland. If
328 * our passive release function was still in place, our priority was
329 * never raised and does not need to be reduced.
331 lwkt_passive_recover(td);
334 * Become the current user scheduled process if we aren't already,
335 * and deal with reschedule requests and other factors.
337 lp->lwp_proc->p_usched->acquire_curproc(lp);
338 /* WARNING: we may have migrated cpu's */
339 /* gd = td->td_gd; */
342 #if !defined(KTR_KERNENTRY)
343 #define KTR_KERNENTRY KTR_ALL
345 KTR_INFO_MASTER(kernentry);
346 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0,
347 "TRAP(pid %hd, tid %hd, trapno %ld, eva %lu)",
348 pid_t pid, lwpid_t tid, register_t trapno, vm_offset_t eva);
349 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "TRAP_RET(pid %hd, tid %hd)",
350 pid_t pid, lwpid_t tid);
351 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "SYSC(pid %hd, tid %hd, nr %ld)",
352 pid_t pid, lwpid_t tid, register_t trapno);
353 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "SYSRET(pid %hd, tid %hd, err %d)",
354 pid_t pid, lwpid_t tid, int err);
355 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "FORKRET(pid %hd, tid %hd)",
356 pid_t pid, lwpid_t tid);
359 * Exception, fault, and trap interface to the kernel.
360 * This common code is called from assembly language IDT gate entry
361 * routines that prepare a suitable stack frame, and restore this
362 * frame after the exception has been processed.
364 * This function is also called from doreti in an interlock to handle ASTs.
365 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
367 * NOTE! We have to retrieve the fault address prior to obtaining the
368 * MP lock because get_mplock() may switch out. YYY cr2 really ought
369 * to be retrieved by the assembly code, not here.
371 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
372 * if an attempt is made to switch from a fast interrupt or IPI. This is
373 * necessary to properly take fatal kernel traps on SMP machines if
374 * get_mplock() has to block.
378 user_trap(struct trapframe *frame)
380 struct globaldata *gd = mycpu;
381 struct thread *td = gd->gd_curthread;
382 struct lwp *lp = td->td_lwp;
385 int i = 0, ucode = 0, type, code;
390 int crit_count = td->td_critcount;
391 lwkt_tokref_t curstop = td->td_toks_stop;
397 if (frame->tf_trapno == T_PAGEFLT)
398 eva = frame->tf_addr;
402 kprintf("USER_TRAP AT %08lx xflags %ld trapno %ld eva %08lx\n",
403 frame->tf_rip, frame->tf_xflags, frame->tf_trapno, eva);
407 * Everything coming from user mode runs through user_trap,
408 * including system calls.
410 if (frame->tf_trapno == T_FAST_SYSCALL) {
415 KTR_LOG(kernentry_trap, lp->lwp_proc->p_pid, lp->lwp_tid,
416 frame->tf_trapno, eva);
420 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
421 ++gd->gd_trap_nesting_level;
422 MAKEMPSAFE(have_mplock);
423 trap_fatal(frame, TRUE, eva);
424 --gd->gd_trap_nesting_level;
429 type = frame->tf_trapno;
430 code = frame->tf_err;
434 sticks = (int)td->td_sticks;
435 lp->lwp_md.md_regs = frame;
438 case T_PRIVINFLT: /* privileged instruction fault */
443 case T_BPTFLT: /* bpt instruction fault */
444 case T_TRCTRAP: /* trace trap */
445 frame->tf_rflags &= ~PSL_T;
447 ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
450 case T_ARITHTRAP: /* arithmetic trap */
455 case T_ASTFLT: /* Allow process switch */
456 mycpu->gd_cnt.v_soft++;
457 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
458 atomic_clear_int(&mycpu->gd_reqflags, RQF_AST_OWEUPC);
459 addupc_task(p, p->p_prof.pr_addr, p->p_prof.pr_ticks);
464 * The following two traps can happen in
465 * vm86 mode, and, if so, we want to handle
468 case T_PROTFLT: /* general protection fault */
469 case T_STKFLT: /* stack fault */
471 if (frame->tf_eflags & PSL_VM) {
472 i = vm86_emulate((struct vm86frame *)frame);
480 case T_SEGNPFLT: /* segment not present fault */
481 case T_TSSFLT: /* invalid TSS fault */
482 case T_DOUBLEFLT: /* double fault */
485 ucode = code + BUS_SEGM_FAULT ;
488 case T_PAGEFLT: /* page fault */
489 MAKEMPSAFE(have_mplock);
490 i = trap_pfault(frame, TRUE, eva);
491 if (i == -1 || i == 0)
503 case T_DIVIDE: /* integer divide fault */
510 MAKEMPSAFE(have_mplock);
511 /* machine/parity/power fail/"kitchen sink" faults */
512 if (isa_nmi(code) == 0) {
515 * NMI can be hooked up to a pushbutton
519 kprintf ("NMI ... going to debugger\n");
520 kdb_trap(type, 0, frame);
524 } else if (panic_on_nmi)
525 panic("NMI indicates hardware failure");
527 #endif /* NISA > 0 */
529 case T_OFLOW: /* integer overflow fault */
534 case T_BOUND: /* bounds check fault */
541 * Virtual kernel intercept - pass the DNA exception
542 * to the (emulated) virtual kernel if it asked to handle
543 * it. This occurs when the virtual kernel is holding
544 * onto the FP context for a different emulated
545 * process then the one currently running.
547 * We must still call npxdna() since we may have
548 * saved FP state that the (emulated) virtual kernel
549 * needs to hand over to a different emulated process.
551 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
552 (td->td_pcb->pcb_flags & FP_VIRTFP)
559 * The kernel may have switched out the FP unit's
560 * state, causing the user process to take a fault
561 * when it tries to use the FP unit. Restore the
566 if (!pmath_emulate) {
568 ucode = FPE_FPU_NP_TRAP;
571 i = (*pmath_emulate)(frame);
573 if (!(frame->tf_rflags & PSL_T))
575 frame->tf_rflags &= ~PSL_T;
578 /* else ucode = emulator_only_knows() XXX */
581 case T_FPOPFLT: /* FPU operand fetch fault */
586 case T_XMMFLT: /* SIMD floating-point exception */
593 * Virtual kernel intercept - if the fault is directly related to a
594 * VM context managed by a virtual kernel then let the virtual kernel
597 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
598 vkernel_trap(lp, frame);
603 * Translate fault for emulators (e.g. Linux)
605 if (*p->p_sysent->sv_transtrap)
606 i = (*p->p_sysent->sv_transtrap)(i, type);
608 MAKEMPSAFE(have_mplock);
609 trapsignal(lp, i, ucode);
612 if (type <= MAX_TRAP_MSG) {
613 uprintf("fatal process exception: %s",
615 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
616 uprintf(", fault VA = 0x%lx", (u_long)eva);
622 userret(lp, frame, sticks);
629 KTR_LOG(kernentry_trap_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
631 KASSERT(crit_count == td->td_critcount,
632 ("trap: critical section count mismatch! %d/%d",
633 crit_count, td->td_pri));
634 KASSERT(curstop == td->td_toks_stop,
635 ("trap: extra tokens held after trap! %ld/%ld",
636 curstop - &td->td_toks_base,
637 td->td_toks_stop - &td->td_toks_base));
642 kern_trap(struct trapframe *frame)
644 struct globaldata *gd = mycpu;
645 struct thread *td = gd->gd_curthread;
648 int i = 0, ucode = 0, type, code;
653 int crit_count = td->td_critcount;
654 lwkt_tokref_t curstop = td->td_toks_stop;
661 if (frame->tf_trapno == T_PAGEFLT)
662 eva = frame->tf_addr;
668 ++gd->gd_trap_nesting_level;
669 MAKEMPSAFE(have_mplock);
670 trap_fatal(frame, FALSE, eva);
671 --gd->gd_trap_nesting_level;
676 type = frame->tf_trapno;
677 code = frame->tf_err;
685 case T_PAGEFLT: /* page fault */
686 MAKEMPSAFE(have_mplock);
687 trap_pfault(frame, FALSE, eva);
692 * The kernel may be using npx for copying or other
695 panic("kernel NPX should not happen");
700 case T_PROTFLT: /* general protection fault */
701 case T_SEGNPFLT: /* segment not present fault */
703 * Invalid segment selectors and out of bounds
704 * %eip's and %esp's can be set up in user mode.
705 * This causes a fault in kernel mode when the
706 * kernel tries to return to user mode. We want
707 * to get this fault so that we can fix the
708 * problem here and not have to check all the
709 * selectors and pointers when the user changes
712 if (mycpu->gd_intr_nesting_level == 0) {
713 if (td->td_pcb->pcb_onfault) {
715 (register_t)td->td_pcb->pcb_onfault;
723 * PSL_NT can be set in user mode and isn't cleared
724 * automatically when the kernel is entered. This
725 * causes a TSS fault when the kernel attempts to
726 * `iret' because the TSS link is uninitialized. We
727 * want to get this fault so that we can fix the
728 * problem here and not every time the kernel is
731 if (frame->tf_rflags & PSL_NT) {
732 frame->tf_rflags &= ~PSL_NT;
737 case T_TRCTRAP: /* trace trap */
739 if (frame->tf_eip == (int)IDTVEC(syscall)) {
741 * We've just entered system mode via the
742 * syscall lcall. Continue single stepping
743 * silently until the syscall handler has
748 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
750 * The syscall handler has now saved the
751 * flags. Stop single stepping it.
753 frame->tf_eflags &= ~PSL_T;
759 * Ignore debug register trace traps due to
760 * accesses in the user's address space, which
761 * can happen under several conditions such as
762 * if a user sets a watchpoint on a buffer and
763 * then passes that buffer to a system call.
764 * We still want to get TRCTRAPS for addresses
765 * in kernel space because that is useful when
766 * debugging the kernel.
768 if (user_dbreg_trap()) {
770 * Reset breakpoint bits because the
773 load_dr6(rdr6() & 0xfffffff0);
778 * Fall through (TRCTRAP kernel mode, kernel address)
782 * If DDB is enabled, let it handle the debugger trap.
783 * Otherwise, debugger traps "can't happen".
786 MAKEMPSAFE(have_mplock);
787 if (kdb_trap (type, 0, frame))
792 MAKEMPSAFE(have_mplock);
793 trap_fatal(frame, FALSE, eva);
796 MAKEMPSAFE(have_mplock);
797 trap_fatal(frame, FALSE, eva);
802 * Ignore this trap generated from a spurious SIGTRAP.
804 * single stepping in / syscalls leads to spurious / SIGTRAP
807 * Haiku (c) 2007 Simon 'corecode' Schubert
813 * Translate fault for emulators (e.g. Linux)
815 if (*p->p_sysent->sv_transtrap)
816 i = (*p->p_sysent->sv_transtrap)(i, type);
818 MAKEMPSAFE(have_mplock);
819 trapsignal(lp, i, ucode);
822 if (type <= MAX_TRAP_MSG) {
823 uprintf("fatal process exception: %s",
825 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
826 uprintf(", fault VA = 0x%lx", (u_long)eva);
838 KASSERT(crit_count == td->td_critcount,
839 ("trap: critical section count mismatch! %d/%d",
840 crit_count, td->td_pri));
841 KASSERT(curstop == td->td_toks_stop,
842 ("trap: extra tokens held after trap! %ld/%ld",
843 curstop - &td->td_toks_base,
844 td->td_toks_stop - &td->td_toks_base));
849 trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
852 struct vmspace *vm = NULL;
856 thread_t td = curthread;
857 struct lwp *lp = td->td_lwp;
860 va = trunc_page(eva);
861 if (usermode == FALSE) {
863 * This is a fault on kernel virtual memory.
868 * This is a fault on non-kernel virtual memory.
869 * vm is initialized above to NULL. If curproc is NULL
870 * or curproc->p_vmspace is NULL the fault is fatal.
873 vm = lp->lwp_vmspace;
881 if (frame->tf_err & PGEX_W)
882 ftype = VM_PROT_READ | VM_PROT_WRITE;
884 ftype = VM_PROT_READ;
886 if (map != &kernel_map) {
888 * Keep swapout from messing with us during this
894 * Grow the stack if necessary
896 /* grow_stack returns false only if va falls into
897 * a growable stack region and the stack growth
898 * fails. It returns true if va was not within
899 * a growable stack region, or if the stack
902 if (!grow_stack (lp->lwp_proc, va)) {
910 fault_flags |= VM_FAULT_BURST;
911 if (ftype & VM_PROT_WRITE)
912 fault_flags |= VM_FAULT_DIRTY;
914 fault_flags |= VM_FAULT_NORMAL;
915 rv = vm_fault(map, va, ftype, fault_flags);
920 * Don't have to worry about process locking or stacks in the kernel.
922 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
925 if (rv == KERN_SUCCESS)
929 if (td->td_gd->gd_intr_nesting_level == 0 &&
930 td->td_pcb->pcb_onfault) {
931 frame->tf_rip = (register_t)td->td_pcb->pcb_onfault;
934 trap_fatal(frame, usermode, eva);
939 * NOTE: on x86_64 we have a tf_addr field in the trapframe, no
940 * kludge is needed to pass the fault address to signal handlers.
942 struct proc *p = td->td_proc;
943 kprintf("seg-fault accessing address %p rip=%p pid=%d p_comm=%s\n",
944 (void *)va, (void *)frame->tf_rip, p->p_pid, p->p_comm);
945 /* Debugger("seg-fault"); */
947 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
951 trap_fatal(struct trapframe *frame, int usermode, vm_offset_t eva)
956 code = frame->tf_xflags;
957 type = frame->tf_trapno;
959 if (type <= MAX_TRAP_MSG) {
960 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
961 type, trap_msg[type],
962 (usermode ? "user" : "kernel"));
965 /* two separate prints in case of a trap on an unmapped page */
966 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
968 if (type == T_PAGEFLT) {
969 kprintf("fault virtual address = %p\n", (void *)eva);
970 kprintf("fault code = %s %s, %s\n",
971 usermode ? "user" : "supervisor",
972 code & PGEX_W ? "write" : "read",
973 code & PGEX_P ? "protection violation" : "page not present");
975 kprintf("instruction pointer = 0x%lx:0x%lx\n",
976 frame->tf_cs & 0xffff, frame->tf_rip);
978 ss = frame->tf_ss & 0xffff;
981 ss = GSEL(GDATA_SEL, SEL_KPL);
982 rsp = (long)&frame->tf_rsp;
984 kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp);
985 kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp);
986 kprintf("processor eflags = ");
987 if (frame->tf_rflags & PSL_T)
988 kprintf("trace trap, ");
989 if (frame->tf_rflags & PSL_I)
990 kprintf("interrupt enabled, ");
991 if (frame->tf_rflags & PSL_NT)
992 kprintf("nested task, ");
993 if (frame->tf_rflags & PSL_RF)
996 if (frame->tf_eflags & PSL_VM)
999 kprintf("IOPL = %jd\n", (intmax_t)((frame->tf_rflags & PSL_IOPL) >> 12));
1000 kprintf("current process = ");
1002 kprintf("%lu (%s)\n",
1003 (u_long)curproc->p_pid, curproc->p_comm ?
1004 curproc->p_comm : "");
1008 kprintf("current thread = pri %d ", curthread->td_pri);
1009 if (curthread->td_critcount)
1015 * we probably SHOULD have stopped the other CPUs before now!
1016 * another CPU COULD have been touching cpl at this moment...
1018 kprintf(" <- SMP: XXX");
1027 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1030 kprintf("trap number = %d\n", type);
1031 if (type <= MAX_TRAP_MSG)
1032 panic("%s", trap_msg[type]);
1034 panic("unknown/reserved trap");
1038 * Double fault handler. Called when a fault occurs while writing
1039 * a frame for a trap/exception onto the stack. This usually occurs
1040 * when the stack overflows (such is the case with infinite recursion,
1043 * XXX Note that the current PTD gets replaced by IdlePTD when the
1044 * task switch occurs. This means that the stack that was active at
1045 * the time of the double fault is not available at <kstack> unless
1046 * the machine was idle when the double fault occurred. The downside
1047 * of this is that "trace <ebp>" in ddb won't work.
1050 dblfault_handler(void)
1053 struct mdglobaldata *gd = mdcpu;
1056 kprintf("\nFatal double fault:\n");
1058 kprintf("rip = 0x%lx\n", gd->gd_common_tss.tss_rip);
1059 kprintf("rsp = 0x%lx\n", gd->gd_common_tss.tss_rsp);
1060 kprintf("rbp = 0x%lx\n", gd->gd_common_tss.tss_rbp);
1063 /* two separate prints in case of a trap on an unmapped page */
1064 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
1066 panic("double fault");
1070 * Compensate for 386 brain damage (missing URKR).
1071 * This is a little simpler than the pagefault handler in trap() because
1072 * it the page tables have already been faulted in and high addresses
1073 * are thrown out early for other reasons.
1076 trapwrite(unsigned addr)
1083 va = trunc_page((vm_offset_t)addr);
1085 * XXX - MAX is END. Changed > to >= for temp. fix.
1087 if (va >= VM_MAX_USER_ADDRESS)
1090 lp = curthread->td_lwp;
1091 vm = lp->lwp_vmspace;
1093 PHOLD(lp->lwp_proc);
1095 if (!grow_stack (lp->lwp_proc, va)) {
1096 PRELE(lp->lwp_proc);
1101 * fault the data page
1103 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY);
1105 PRELE(lp->lwp_proc);
1107 if (rv != KERN_SUCCESS)
1114 * syscall2 - MP aware system call request C handler
1116 * A system call is essentially treated as a trap except that the
1117 * MP lock is not held on entry or return. We are responsible for
1118 * obtaining the MP lock if necessary and for handling ASTs
1119 * (e.g. a task switch) prior to return.
1121 * In general, only simple access and manipulation of curproc and
1122 * the current stack is allowed without having to hold MP lock.
1124 * MPSAFE - note that large sections of this routine are run without
1128 syscall2(struct trapframe *frame)
1130 struct thread *td = curthread;
1131 struct proc *p = td->td_proc;
1132 struct lwp *lp = td->td_lwp;
1134 struct sysent *callp;
1135 register_t orig_tf_rflags;
1140 int crit_count = td->td_critcount;
1141 lwkt_tokref_t curstop = td->td_toks_stop;
1144 int have_mplock = 0;
1149 union sysunion args;
1150 register_t *argsdst;
1152 mycpu->gd_cnt.v_syscall++;
1154 KTR_LOG(kernentry_syscall, lp->lwp_proc->p_pid, lp->lwp_tid,
1157 userenter(td, p); /* lazy raise our priority */
1164 sticks = (int)td->td_sticks;
1165 orig_tf_rflags = frame->tf_rflags;
1168 * Virtual kernel intercept - if a VM context managed by a virtual
1169 * kernel issues a system call the virtual kernel handles it, not us.
1170 * Restore the virtual kernel context and return from its system
1171 * call. The current frame is copied out to the virtual kernel.
1173 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1174 vkernel_trap(lp, frame);
1175 error = EJUSTRETURN;
1180 * Get the system call parameters and account for time
1182 lp->lwp_md.md_regs = frame;
1183 params = (caddr_t)frame->tf_rsp + sizeof(register_t);
1184 code = frame->tf_rax;
1186 if (p->p_sysent->sv_prepsyscall) {
1187 (*p->p_sysent->sv_prepsyscall)(
1188 frame, (int *)(&args.nosys.sysmsg + 1),
1191 if (code == SYS_syscall || code == SYS___syscall) {
1192 code = frame->tf_rdi;
1198 if (p->p_sysent->sv_mask)
1199 code &= p->p_sysent->sv_mask;
1201 if (code >= p->p_sysent->sv_size)
1202 callp = &p->p_sysent->sv_table[0];
1204 callp = &p->p_sysent->sv_table[code];
1206 narg = callp->sy_narg & SYF_ARGMASK;
1209 * On x86_64 we get up to six arguments in registers. The rest are
1210 * on the stack. The first six members of 'struct trapframe' happen
1211 * to be the registers used to pass arguments, in exactly the right
1214 argp = &frame->tf_rdi;
1216 argsdst = (register_t *)(&args.nosys.sysmsg + 1);
1218 * JG can we overflow the space pointed to by 'argsdst'
1219 * either with 'bcopy' or with 'copyin'?
1221 bcopy(argp, argsdst, sizeof(register_t) * regcnt);
1223 * copyin is MP aware, but the tracing code is not
1225 if (narg > regcnt) {
1226 KASSERT(params != NULL, ("copyin args with no params!"));
1227 error = copyin(params, &argsdst[regcnt],
1228 (narg - regcnt) * sizeof(register_t));
1231 if (KTRPOINT(td, KTR_SYSCALL)) {
1232 MAKEMPSAFE(have_mplock);
1234 ktrsyscall(lp, code, narg,
1235 (void *)(&args.nosys.sysmsg + 1));
1243 if (KTRPOINT(td, KTR_SYSCALL)) {
1244 MAKEMPSAFE(have_mplock);
1245 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1250 * Default return value is 0 (will be copied to %rax). Double-value
1251 * returns use %rax and %rdx. %rdx is left unchanged for system
1252 * calls which return only one result.
1254 args.sysmsg_fds[0] = 0;
1255 args.sysmsg_fds[1] = frame->tf_rdx;
1258 * The syscall might manipulate the trap frame. If it does it
1259 * will probably return EJUSTRETURN.
1261 args.sysmsg_frame = frame;
1263 STOPEVENT(p, S_SCE, narg); /* MP aware */
1266 * NOTE: All system calls run MPSAFE now. The system call itself
1267 * is responsible for getting the MP lock.
1269 error = (*callp->sy_call)(&args);
1272 kprintf("system call %d returned %d\n", code, error);
1277 * MP SAFE (we may or may not have the MP lock at this point)
1282 * Reinitialize proc pointer `p' as it may be different
1283 * if this is a child returning from fork syscall.
1286 lp = curthread->td_lwp;
1287 frame->tf_rax = args.sysmsg_fds[0];
1288 frame->tf_rdx = args.sysmsg_fds[1];
1289 frame->tf_rflags &= ~PSL_C;
1293 * Reconstruct pc, we know that 'syscall' is 2 bytes.
1294 * We have to do a full context restore so that %r10
1295 * (which was holding the value of %rcx) is restored for
1296 * the next iteration.
1298 frame->tf_rip -= frame->tf_err;
1299 frame->tf_r10 = frame->tf_rcx;
1304 panic("Unexpected EASYNC return value (for now)");
1307 if (p->p_sysent->sv_errsize) {
1308 if (error >= p->p_sysent->sv_errsize)
1309 error = -1; /* XXX */
1311 error = p->p_sysent->sv_errtbl[error];
1313 frame->tf_rax = error;
1314 frame->tf_rflags |= PSL_C;
1319 * Traced syscall. trapsignal() is not MP aware.
1321 if (orig_tf_rflags & PSL_T) {
1322 MAKEMPSAFE(have_mplock);
1323 frame->tf_rflags &= ~PSL_T;
1324 trapsignal(lp, SIGTRAP, 0);
1328 * Handle reschedule and other end-of-syscall issues
1330 userret(lp, frame, sticks);
1333 if (KTRPOINT(td, KTR_SYSRET)) {
1334 MAKEMPSAFE(have_mplock);
1335 ktrsysret(lp, code, error, args.sysmsg_result);
1340 * This works because errno is findable through the
1341 * register set. If we ever support an emulation where this
1342 * is not the case, this code will need to be revisited.
1344 STOPEVENT(p, S_SCX, code);
1349 * Release the MP lock if we had to get it
1354 KTR_LOG(kernentry_syscall_ret, lp->lwp_proc->p_pid, lp->lwp_tid, error);
1356 KASSERT(&td->td_toks_base == td->td_toks_stop,
1357 ("syscall: critical section count mismatch! %d/%d",
1358 crit_count, td->td_pri));
1359 KASSERT(curstop == td->td_toks_stop,
1360 ("syscall: extra tokens held after trap! %ld",
1361 td->td_toks_stop - &td->td_toks_base));
1366 * NOTE: mplock not held at any point
1369 fork_return(struct lwp *lp, struct trapframe *frame)
1371 frame->tf_rax = 0; /* Child returns zero */
1372 frame->tf_rflags &= ~PSL_C; /* success */
1375 generic_lwp_return(lp, frame);
1376 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1380 * Simplified back end of syscall(), used when returning from fork()
1381 * directly into user mode.
1383 * This code will return back into the fork trampoline code which then
1386 * NOTE: The mplock is not held at any point.
1389 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1391 struct proc *p = lp->lwp_proc;
1394 * Newly forked processes are given a kernel priority. We have to
1395 * adjust the priority to a normal user priority and fake entry
1396 * into the kernel (call userenter()) to install a passive release
1397 * function just in case userret() decides to stop the process. This
1398 * can occur when ^Z races a fork. If we do not install the passive
1399 * release function the current process designation will not be
1400 * released when the thread goes to sleep.
1402 lwkt_setpri_self(TDPRI_USER_NORM);
1403 userenter(lp->lwp_thread, p);
1404 userret(lp, frame, 0);
1406 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1407 ktrsysret(lp, SYS_fork, 0, 0);
1409 lp->lwp_flags |= LWP_PASSIVE_ACQ;
1411 lp->lwp_flags &= ~LWP_PASSIVE_ACQ;
1415 * doreti has turned into this. The frame is directly on the stack. We
1416 * pull everything else we need (fpu and tls context) from the current
1419 * Note on fpu interactions: In a virtual kernel, the fpu context for
1420 * an emulated user mode process is not shared with the virtual kernel's
1421 * fpu context, so we only have to 'stack' fpu contexts within the virtual
1422 * kernel itself, and not even then since the signal() contexts that we care
1423 * about save and restore the FPU state (I think anyhow).
1425 * vmspace_ctl() returns an error only if it had problems instaling the
1426 * context we supplied or problems copying data to/from our VM space.
1429 go_user(struct intrframe *frame)
1431 struct trapframe *tf = (void *)&frame->if_rdi;
1435 * Interrupts may be disabled on entry, make sure all signals
1436 * can be received before beginning our loop.
1441 * Switch to the current simulated user process, then call
1442 * user_trap() when we break out of it (usually due to a signal).
1446 * Tell the real kernel whether it is ok to use the FP
1449 if (mdcpu->gd_npxthread == curthread) {
1450 tf->tf_xflags &= ~PGEX_FPFAULT;
1452 tf->tf_xflags |= PGEX_FPFAULT;
1456 * Run emulated user process context. This call interlocks
1457 * with new mailbox signals.
1459 * Set PGEX_U unconditionally, indicating a user frame (the
1460 * bit is normally set only by T_PAGEFLT).
1462 r = vmspace_ctl(&curproc->p_vmspace->vm_pmap, VMSPACE_CTL_RUN,
1463 tf, &curthread->td_savevext);
1464 frame->if_xflags |= PGEX_U;
1466 kprintf("GO USER %d trap %ld EVA %08lx RIP %08lx RSP %08lx XFLAGS %02lx/%02lx\n",
1467 r, tf->tf_trapno, tf->tf_addr, tf->tf_rip, tf->tf_rsp,
1468 tf->tf_xflags, frame->if_xflags);
1472 panic("vmspace_ctl failed error %d", errno);
1474 if (tf->tf_trapno) {
1478 if (mycpu->gd_reqflags & RQF_AST_MASK) {
1479 tf->tf_trapno = T_ASTFLT;
1487 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1488 * fault (which is then passed back to the virtual kernel) if an attempt is
1489 * made to use the FP unit.
1491 * XXX this is a fairly big hack.
1494 set_vkernel_fp(struct trapframe *frame)
1496 struct thread *td = curthread;
1498 if (frame->tf_xflags & PGEX_FPFAULT) {
1499 td->td_pcb->pcb_flags |= FP_VIRTFP;
1500 if (mdcpu->gd_npxthread == td)
1503 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1508 * Called from vkernel_trap() to fixup the vkernel's syscall
1509 * frame for vmspace_ctl() return.
1512 cpu_vkernel_trap(struct trapframe *frame, int error)
1514 frame->tf_rax = error;
1516 frame->tf_rflags |= PSL_C;
1518 frame->tf_rflags &= ~PSL_C;