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 * 386 Trap and System call handling
49 #include "opt_ktrace.h"
51 #include <sys/param.h>
52 #include <sys/systm.h>
54 #include <sys/pioctl.h>
55 #include <sys/kernel.h>
56 #include <sys/resourcevar.h>
57 #include <sys/signalvar.h>
58 #include <sys/signal2.h>
59 #include <sys/syscall.h>
60 #include <sys/sysctl.h>
61 #include <sys/sysent.h>
63 #include <sys/vmmeter.h>
64 #include <sys/malloc.h>
66 #include <sys/ktrace.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>
90 #include <machine/vm86.h>
94 #include <sys/msgport2.h>
95 #include <sys/thread2.h>
96 #include <sys/mplock2.h>
98 #define MAKEMPSAFE(have_mplock) \
99 if (have_mplock == 0) { \
104 int (*pmath_emulate) (struct trapframe *);
106 static int trap_pfault (struct trapframe *, int, vm_offset_t);
107 static void trap_fatal (struct trapframe *, int, vm_offset_t);
108 void dblfault_handler (void);
111 extern inthand_t IDTVEC(syscall);
114 #define MAX_TRAP_MSG 28
115 static char *trap_msg[] = {
117 "privileged instruction fault", /* 1 T_PRIVINFLT */
119 "breakpoint instruction fault", /* 3 T_BPTFLT */
122 "arithmetic trap", /* 6 T_ARITHTRAP */
123 "system forced exception", /* 7 T_ASTFLT */
125 "general protection fault", /* 9 T_PROTFLT */
126 "trace trap", /* 10 T_TRCTRAP */
128 "page fault", /* 12 T_PAGEFLT */
130 "alignment fault", /* 14 T_ALIGNFLT */
134 "integer divide fault", /* 18 T_DIVIDE */
135 "non-maskable interrupt trap", /* 19 T_NMI */
136 "overflow trap", /* 20 T_OFLOW */
137 "FPU bounds check fault", /* 21 T_BOUND */
138 "FPU device not available", /* 22 T_DNA */
139 "double fault", /* 23 T_DOUBLEFLT */
140 "FPU operand fetch fault", /* 24 T_FPOPFLT */
141 "invalid TSS fault", /* 25 T_TSSFLT */
142 "segment not present fault", /* 26 T_SEGNPFLT */
143 "stack fault", /* 27 T_STKFLT */
144 "machine check trap", /* 28 T_MCHK */
148 static int ddb_on_nmi = 1;
149 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
150 &ddb_on_nmi, 0, "Go to DDB on NMI");
152 static int panic_on_nmi = 1;
153 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
154 &panic_on_nmi, 0, "Panic on NMI");
155 static int fast_release;
156 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
157 &fast_release, 0, "Passive Release was optimal");
158 static int slow_release;
159 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
160 &slow_release, 0, "Passive Release was nonoptimal");
162 MALLOC_DEFINE(M_SYSMSG, "sysmsg", "sysmsg structure");
163 extern int max_sysmsg;
166 * Passively intercepts the thread switch function to increase the thread
167 * priority from a user priority to a kernel priority, reducing
168 * syscall and trap overhead for the case where no switch occurs.
170 * Synchronizes td_ucred with p_ucred. This is used by system calls,
171 * signal handling, faults, AST traps, and anything else that enters the
172 * kernel from userland and provides the kernel with a stable read-only
173 * copy of the process ucred.
176 userenter(struct thread *curtd, struct proc *curp)
181 curtd->td_release = lwkt_passive_release;
183 if (curtd->td_ucred != curp->p_ucred) {
184 ncred = crhold(curp->p_ucred);
185 ocred = curtd->td_ucred;
186 curtd->td_ucred = ncred;
194 * Handle signals, profiling, and other AST's and/or tasks that
195 * must be completed before we can return to or try to return to userland.
197 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
198 * arithmatic on the delta calculation so the absolute tick values are
199 * truncated to an integer.
202 userret(struct lwp *lp, struct trapframe *frame, int sticks)
204 struct proc *p = lp->lwp_proc;
208 * Charge system time if profiling. Note: times are in microseconds.
209 * This may do a copyout and block, so do it first even though it
210 * means some system time will be charged as user time.
212 if (p->p_flags & P_PROFIL) {
213 addupc_task(p, frame->tf_eip,
214 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
219 * Specific on-return-to-usermode checks (LWP_MP_WEXIT,
220 * LWP_MP_VNLRU, etc).
222 if (lp->lwp_mpflags & LWP_MP_URETMASK)
226 * Block here if we are in a stopped state.
228 if (p->p_stat == SSTOP) {
229 lwkt_gettoken(&p->p_token);
231 lwkt_reltoken(&p->p_token);
236 * Post any pending upcalls. If running a virtual kernel be sure
237 * to restore the virtual kernel's vmspace before posting the upcall.
239 if (p->p_flags & (P_SIGVTALRM | P_SIGPROF)) {
240 lwkt_gettoken(&p->p_token);
241 if (p->p_flags & P_SIGVTALRM) {
242 p->p_flags &= ~P_SIGVTALRM;
243 ksignal(p, SIGVTALRM);
245 if (p->p_flags & P_SIGPROF) {
246 p->p_flags &= ~P_SIGPROF;
249 lwkt_reltoken(&p->p_token);
254 * Post any pending signals
256 * WARNING! postsig() can exit and not return.
258 if ((sig = CURSIG_TRACE(lp)) != 0) {
259 lwkt_gettoken(&p->p_token);
261 lwkt_reltoken(&p->p_token);
266 * block here if we are swapped out, but still process signals
267 * (such as SIGKILL). proc0 (the swapin scheduler) is already
268 * aware of our situation, we do not have to wake it up.
270 if (p->p_flags & P_SWAPPEDOUT) {
271 lwkt_gettoken(&p->p_token);
273 p->p_flags |= P_SWAPWAIT;
275 if (p->p_flags & P_SWAPWAIT)
276 tsleep(p, PCATCH, "SWOUT", 0);
277 p->p_flags &= ~P_SWAPWAIT;
279 lwkt_reltoken(&p->p_token);
284 * In a multi-threaded program it is possible for a thread to change
285 * signal state during a system call which temporarily changes the
286 * signal mask. In this case postsig() might not be run and we
287 * have to restore the mask ourselves.
289 if (lp->lwp_flags & LWP_OLDMASK) {
290 lp->lwp_flags &= ~LWP_OLDMASK;
291 lp->lwp_sigmask = lp->lwp_oldsigmask;
297 * Cleanup from userenter and any passive release that might have occured.
298 * We must reclaim the current-process designation before we can return
299 * to usermode. We also handle both LWKT and USER reschedule requests.
302 userexit(struct lwp *lp)
304 struct thread *td = lp->lwp_thread;
305 /* globaldata_t gd = td->td_gd; */
308 * Handle stop requests at kernel priority. Any requests queued
309 * after this loop will generate another AST.
311 while (lp->lwp_proc->p_stat == SSTOP) {
312 lwkt_gettoken(&lp->lwp_proc->p_token);
314 lwkt_reltoken(&lp->lwp_proc->p_token);
318 * Become the current user scheduled process if we aren't already,
319 * and deal with reschedule requests and other factors.
321 lp->lwp_proc->p_usched->acquire_curproc(lp);
322 /* WARNING: we may have migrated cpu's */
323 /* gd = td->td_gd; */
326 * Reduce our priority in preparation for a return to userland. If
327 * our passive release function was still in place, our priority was
328 * never raised and does not need to be reduced.
330 lwkt_passive_recover(td);
333 #if !defined(KTR_KERNENTRY)
334 #define KTR_KERNENTRY KTR_ALL
336 KTR_INFO_MASTER(kernentry);
337 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0,
338 "TRAP(pid %d, tid %d, trapno %d, eva %lu)",
339 pid_t pid, lwpid_t tid, register_t trapno, vm_offset_t eva);
340 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "TRAP_RET(pid %d, tid %d)",
341 pid_t pid, lwpid_t tid);
342 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "SYSC(pid %d, tid %d, nr %d)",
343 pid_t pid, lwpid_t tid, register_t trapno);
344 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "SYSRET(pid %d, tid %d, err %d)",
345 pid_t pid, lwpid_t tid, int err);
346 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "FORKRET(pid %d, tid %d)",
347 pid_t pid, lwpid_t tid);
350 * Exception, fault, and trap interface to the kernel.
351 * This common code is called from assembly language IDT gate entry
352 * routines that prepare a suitable stack frame, and restore this
353 * frame after the exception has been processed.
355 * This function is also called from doreti in an interlock to handle ASTs.
356 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
358 * NOTE! We have to retrieve the fault address prior to obtaining the
359 * MP lock because get_mplock() may switch out. YYY cr2 really ought
360 * to be retrieved by the assembly code, not here.
362 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
363 * if an attempt is made to switch from a fast interrupt or IPI. This is
364 * necessary to properly take fatal kernel traps on SMP machines if
365 * get_mplock() has to block.
369 user_trap(struct trapframe *frame)
371 struct globaldata *gd = mycpu;
372 struct thread *td = gd->gd_curthread;
373 struct lwp *lp = td->td_lwp;
376 int i = 0, ucode = 0, type, code;
379 int crit_count = td->td_critcount;
380 lwkt_tokref_t curstop = td->td_toks_stop;
387 * This is a bad kludge to avoid changing the various trapframe
388 * structures. Because we are enabled as a virtual kernel,
389 * the original tf_err field will be passed to us shifted 16
390 * over in the tf_trapno field for T_PAGEFLT.
392 if (frame->tf_trapno == T_PAGEFLT)
397 kprintf("USER_TRAP AT %08x xflags %d trapno %d eva %08x\n",
398 frame->tf_eip, frame->tf_xflags, frame->tf_trapno, eva);
402 * Everything coming from user mode runs through user_trap,
403 * including system calls.
405 if (frame->tf_trapno == T_SYSCALL80) {
410 KTR_LOG(kernentry_trap, lp->lwp_proc->p_pid, lp->lwp_tid,
411 frame->tf_trapno, eva);
415 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
416 ++gd->gd_trap_nesting_level;
417 MAKEMPSAFE(have_mplock);
418 trap_fatal(frame, TRUE, eva);
419 --gd->gd_trap_nesting_level;
424 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
427 type = frame->tf_trapno;
428 code = frame->tf_err;
432 sticks = (int)td->td_sticks;
433 lp->lwp_md.md_regs = frame;
436 case T_PRIVINFLT: /* privileged instruction fault */
441 case T_BPTFLT: /* bpt instruction fault */
442 case T_TRCTRAP: /* trace trap */
443 frame->tf_eflags &= ~PSL_T;
445 ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
448 case T_ARITHTRAP: /* arithmetic trap */
453 case T_ASTFLT: /* Allow process switch */
454 mycpu->gd_cnt.v_soft++;
455 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
456 atomic_clear_int(&mycpu->gd_reqflags,
458 addupc_task(p, p->p_prof.pr_addr,
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);
479 ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR;
481 case T_SEGNPFLT: /* segment not present fault */
485 case T_TSSFLT: /* invalid TSS fault */
486 case T_DOUBLEFLT: /* double fault */
492 case T_PAGEFLT: /* page fault */
493 MAKEMPSAFE(have_mplock);
494 i = trap_pfault(frame, TRUE, eva);
497 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
512 case T_DIVIDE: /* integer divide fault */
519 MAKEMPSAFE(have_mplock);
520 /* machine/parity/power fail/"kitchen sink" faults */
521 if (isa_nmi(code) == 0) {
524 * NMI can be hooked up to a pushbutton
528 kprintf ("NMI ... going to debugger\n");
529 kdb_trap (type, 0, frame);
533 } else if (panic_on_nmi)
534 panic("NMI indicates hardware failure");
536 #endif /* NISA > 0 */
538 case T_OFLOW: /* integer overflow fault */
543 case T_BOUND: /* bounds check fault */
550 * Virtual kernel intercept - pass the DNA exception
551 * to the (emulated) virtual kernel if it asked to handle
552 * it. This occurs when the virtual kernel is holding
553 * onto the FP context for a different emulated
554 * process then the one currently running.
556 * We must still call npxdna() since we may have
557 * saved FP state that the (emulated) virtual kernel
558 * needs to hand over to a different emulated process.
560 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
561 (td->td_pcb->pcb_flags & FP_VIRTFP)
569 * The kernel may have switched out the FP unit's
570 * state, causing the user process to take a fault
571 * when it tries to use the FP unit. Restore the
577 if (!pmath_emulate) {
579 ucode = FPE_FPU_NP_TRAP;
582 i = (*pmath_emulate)(frame);
584 if (!(frame->tf_eflags & PSL_T))
586 frame->tf_eflags &= ~PSL_T;
589 /* else ucode = emulator_only_knows() XXX */
592 case T_FPOPFLT: /* FPU operand fetch fault */
597 case T_XMMFLT: /* SIMD floating-point exception */
604 * Virtual kernel intercept - if the fault is directly related to a
605 * VM context managed by a virtual kernel then let the virtual kernel
608 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
609 vkernel_trap(lp, frame);
614 * Translate fault for emulators (e.g. Linux)
616 if (*p->p_sysent->sv_transtrap)
617 i = (*p->p_sysent->sv_transtrap)(i, type);
619 MAKEMPSAFE(have_mplock);
620 trapsignal(lp, i, ucode);
623 if (type <= MAX_TRAP_MSG) {
624 uprintf("fatal process exception: %s",
626 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
627 uprintf(", fault VA = 0x%lx", (u_long)eva);
633 userret(lp, frame, sticks);
638 KTR_LOG(kernentry_trap_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
640 KASSERT(crit_count == td->td_critcount,
641 ("trap: critical section count mismatch! %d/%d",
642 crit_count, td->td_pri));
643 KASSERT(curstop == td->td_toks_stop,
644 ("trap: extra tokens held after trap! %zd/%zd",
645 curstop - &td->td_toks_base,
646 td->td_toks_stop - &td->td_toks_base));
651 kern_trap(struct trapframe *frame)
653 struct globaldata *gd = mycpu;
654 struct thread *td = gd->gd_curthread;
657 int i = 0, ucode = 0, type, code;
660 int crit_count = td->td_critcount;
661 lwkt_tokref_t curstop = td->td_toks_stop;
668 if (frame->tf_trapno == T_PAGEFLT)
675 ++gd->gd_trap_nesting_level;
676 MAKEMPSAFE(have_mplock);
677 trap_fatal(frame, FALSE, eva);
678 --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);
699 * The kernel may be using npx for copying or other
702 panic("kernel NPX should not happen");
708 case T_PROTFLT: /* general protection fault */
709 case T_SEGNPFLT: /* segment not present fault */
711 * Invalid segment selectors and out of bounds
712 * %eip's and %esp's can be set up in user mode.
713 * This causes a fault in kernel mode when the
714 * kernel tries to return to user mode. We want
715 * to get this fault so that we can fix the
716 * problem here and not have to check all the
717 * selectors and pointers when the user changes
720 if (mycpu->gd_intr_nesting_level == 0) {
721 if (td->td_pcb->pcb_onfault) {
723 (register_t)td->td_pcb->pcb_onfault;
731 * PSL_NT can be set in user mode and isn't cleared
732 * automatically when the kernel is entered. This
733 * causes a TSS fault when the kernel attempts to
734 * `iret' because the TSS link is uninitialized. We
735 * want to get this fault so that we can fix the
736 * problem here and not every time the kernel is
739 if (frame->tf_eflags & PSL_NT) {
740 frame->tf_eflags &= ~PSL_NT;
745 case T_TRCTRAP: /* trace trap */
747 if (frame->tf_eip == (int)IDTVEC(syscall)) {
749 * We've just entered system mode via the
750 * syscall lcall. Continue single stepping
751 * silently until the syscall handler has
756 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
758 * The syscall handler has now saved the
759 * flags. Stop single stepping it.
761 frame->tf_eflags &= ~PSL_T;
767 * Ignore debug register trace traps due to
768 * accesses in the user's address space, which
769 * can happen under several conditions such as
770 * if a user sets a watchpoint on a buffer and
771 * then passes that buffer to a system call.
772 * We still want to get TRCTRAPS for addresses
773 * in kernel space because that is useful when
774 * debugging the kernel.
776 if (user_dbreg_trap()) {
778 * Reset breakpoint bits because the
781 load_dr6(rdr6() & 0xfffffff0);
786 * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
790 * If DDB is enabled, let it handle the debugger trap.
791 * Otherwise, debugger traps "can't happen".
794 MAKEMPSAFE(have_mplock);
795 if (kdb_trap (type, 0, frame))
800 MAKEMPSAFE(have_mplock);
801 trap_fatal(frame, FALSE, eva);
804 MAKEMPSAFE(have_mplock);
805 trap_fatal(frame, FALSE, eva);
809 * Ignore this trap generated from a spurious SIGTRAP.
811 * single stepping in / syscalls leads to spurious / SIGTRAP
814 * Haiku (c) 2007 Simon 'corecode' Schubert
820 * Translate fault for emulators (e.g. Linux)
822 if (*p->p_sysent->sv_transtrap)
823 i = (*p->p_sysent->sv_transtrap)(i, type);
825 MAKEMPSAFE(have_mplock);
826 trapsignal(lp, i, ucode);
829 if (type <= MAX_TRAP_MSG) {
830 uprintf("fatal process exception: %s",
832 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
833 uprintf(", fault VA = 0x%lx", (u_long)eva);
843 KASSERT(crit_count == td->td_critcount,
844 ("trap: critical section count mismatch! %d/%d",
845 crit_count, td->td_pri));
846 KASSERT(curstop == td->td_toks_stop,
847 ("trap: extra tokens held after trap! %zd/%zd",
848 curstop - &td->td_toks_base,
849 td->td_toks_stop - &td->td_toks_base));
854 trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
857 struct vmspace *vm = NULL;
862 thread_t td = curthread;
863 struct lwp *lp = td->td_lwp;
865 va = trunc_page(eva);
866 if (usermode == FALSE) {
868 * This is a fault on kernel virtual memory.
873 * This is a fault on non-kernel virtual memory.
874 * vm is initialized above to NULL. If curproc is NULL
875 * or curproc->p_vmspace is NULL the fault is fatal.
878 vm = lp->lwp_vmspace;
886 if (frame->tf_xflags & PGEX_W)
887 ftype = VM_PROT_READ | VM_PROT_WRITE;
889 ftype = VM_PROT_READ;
891 if (map != &kernel_map) {
893 * Keep swapout from messing with us during this
903 fault_flags |= VM_FAULT_BURST;
904 if (ftype & VM_PROT_WRITE)
905 fault_flags |= VM_FAULT_DIRTY;
907 fault_flags |= VM_FAULT_NORMAL;
908 rv = vm_fault(map, va, ftype, fault_flags);
912 * Don't have to worry about process locking or stacks in the kernel.
914 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
916 if (rv == KERN_SUCCESS)
920 if (td->td_gd->gd_intr_nesting_level == 0 &&
921 td->td_pcb->pcb_onfault) {
922 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
925 trap_fatal(frame, usermode, eva);
928 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
932 trap_fatal(struct trapframe *frame, int usermode, vm_offset_t eva)
934 int code, type, ss, esp;
936 code = frame->tf_xflags;
937 type = frame->tf_trapno;
939 if (type <= MAX_TRAP_MSG) {
940 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
941 type, trap_msg[type],
942 (usermode ? "user" : "kernel"));
944 /* two separate prints in case of a trap on an unmapped page */
945 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
946 if (type == T_PAGEFLT) {
947 kprintf("fault virtual address = %p\n", (void *)eva);
948 kprintf("fault code = %s %s, %s\n",
949 usermode ? "user" : "supervisor",
950 code & PGEX_W ? "write" : "read",
951 code & PGEX_P ? "protection violation" : "page not present");
953 kprintf("instruction pointer = 0x%x:0x%x\n",
954 frame->tf_cs & 0xffff, frame->tf_eip);
956 ss = frame->tf_ss & 0xffff;
959 ss = GSEL(GDATA_SEL, SEL_KPL);
960 esp = (int)&frame->tf_esp;
962 kprintf("stack pointer = 0x%x:0x%x\n", ss, esp);
963 kprintf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
964 kprintf("processor eflags = ");
965 if (frame->tf_eflags & PSL_T)
966 kprintf("trace trap, ");
967 if (frame->tf_eflags & PSL_I)
968 kprintf("interrupt enabled, ");
969 if (frame->tf_eflags & PSL_NT)
970 kprintf("nested task, ");
971 if (frame->tf_eflags & PSL_RF)
974 if (frame->tf_eflags & PSL_VM)
977 kprintf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
978 kprintf("current process = ");
980 kprintf("%lu (%s)\n",
981 (u_long)curproc->p_pid, curproc->p_comm ?
982 curproc->p_comm : "");
986 kprintf("current thread = pri %d ", curthread->td_pri);
987 if (curthread->td_critcount)
992 * we probably SHOULD have stopped the other CPUs before now!
993 * another CPU COULD have been touching cpl at this moment...
995 kprintf(" <- SMP: XXX");
1003 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1006 kprintf("trap number = %d\n", type);
1007 if (type <= MAX_TRAP_MSG)
1008 panic("%s", trap_msg[type]);
1010 panic("unknown/reserved trap");
1014 * Double fault handler. Called when a fault occurs while writing
1015 * a frame for a trap/exception onto the stack. This usually occurs
1016 * when the stack overflows (such is the case with infinite recursion,
1019 * XXX Note that the current PTD gets replaced by IdlePTD when the
1020 * task switch occurs. This means that the stack that was active at
1021 * the time of the double fault is not available at <kstack> unless
1022 * the machine was idle when the double fault occurred. The downside
1023 * of this is that "trace <ebp>" in ddb won't work.
1026 dblfault_handler(void)
1028 struct mdglobaldata *gd = mdcpu;
1030 kprintf("\nFatal double fault:\n");
1031 kprintf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1032 kprintf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1033 kprintf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
1034 /* two separate prints in case of a trap on an unmapped page */
1035 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
1036 panic("double fault");
1040 * syscall2 - MP aware system call request C handler
1042 * A system call is essentially treated as a trap except that the
1043 * MP lock is not held on entry or return. We are responsible for
1044 * obtaining the MP lock if necessary and for handling ASTs
1045 * (e.g. a task switch) prior to return.
1050 syscall2(struct trapframe *frame)
1052 struct thread *td = curthread;
1053 struct proc *p = td->td_proc;
1054 struct lwp *lp = td->td_lwp;
1056 struct sysent *callp;
1057 register_t orig_tf_eflags;
1062 int crit_count = td->td_critcount;
1064 int have_mplock = 0;
1066 union sysunion args;
1068 KTR_LOG(kernentry_syscall, lp->lwp_proc->p_pid, lp->lwp_tid,
1071 userenter(td, p); /* lazy raise our priority */
1076 sticks = (int)td->td_sticks;
1077 orig_tf_eflags = frame->tf_eflags;
1080 * Virtual kernel intercept - if a VM context managed by a virtual
1081 * kernel issues a system call the virtual kernel handles it, not us.
1082 * Restore the virtual kernel context and return from its system
1083 * call. The current frame is copied out to the virtual kernel.
1085 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1086 vkernel_trap(lp, frame);
1087 error = EJUSTRETURN;
1092 * Get the system call parameters and account for time
1094 lp->lwp_md.md_regs = frame;
1095 params = (caddr_t)frame->tf_esp + sizeof(int);
1096 code = frame->tf_eax;
1098 if (p->p_sysent->sv_prepsyscall) {
1099 (*p->p_sysent->sv_prepsyscall)(
1100 frame, (int *)(&args.nosys.sysmsg + 1),
1104 * Need to check if this is a 32 bit or 64 bit syscall.
1105 * fuword is MP aware.
1107 if (code == SYS_syscall) {
1109 * Code is first argument, followed by actual args.
1111 code = fuword(params);
1112 params += sizeof(int);
1113 } else if (code == SYS___syscall) {
1115 * Like syscall, but code is a quad, so as to maintain
1116 * quad alignment for the rest of the arguments.
1118 code = fuword(params);
1119 params += sizeof(quad_t);
1123 code &= p->p_sysent->sv_mask;
1124 if (code >= p->p_sysent->sv_size)
1125 callp = &p->p_sysent->sv_table[0];
1127 callp = &p->p_sysent->sv_table[code];
1129 narg = callp->sy_narg & SYF_ARGMASK;
1132 * copyin is MP aware, but the tracing code is not
1134 if (narg && params) {
1135 error = copyin(params, (caddr_t)(&args.nosys.sysmsg + 1),
1136 narg * sizeof(register_t));
1139 if (KTRPOINT(td, KTR_SYSCALL)) {
1140 MAKEMPSAFE(have_mplock);
1142 ktrsyscall(lp, code, narg,
1143 (void *)(&args.nosys.sysmsg + 1));
1151 if (KTRPOINT(td, KTR_SYSCALL)) {
1152 MAKEMPSAFE(have_mplock);
1153 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1158 * For traditional syscall code edx is left untouched when 32 bit
1159 * results are returned. Since edx is loaded from fds[1] when the
1160 * system call returns we pre-set it here.
1162 args.sysmsg_fds[0] = 0;
1163 args.sysmsg_fds[1] = frame->tf_edx;
1166 * The syscall might manipulate the trap frame. If it does it
1167 * will probably return EJUSTRETURN.
1169 args.sysmsg_frame = frame;
1171 STOPEVENT(p, S_SCE, narg); /* MP aware */
1174 * NOTE: All system calls run MPSAFE now. The system call itself
1175 * is responsible for getting the MP lock.
1177 error = (*callp->sy_call)(&args);
1180 kprintf("system call %d returned %d\n", code, error);
1185 * MP SAFE (we may or may not have the MP lock at this point)
1190 * Reinitialize proc pointer `p' as it may be different
1191 * if this is a child returning from fork syscall.
1194 lp = curthread->td_lwp;
1195 frame->tf_eax = args.sysmsg_fds[0];
1196 frame->tf_edx = args.sysmsg_fds[1];
1197 frame->tf_eflags &= ~PSL_C;
1201 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1202 * int 0x80 is 2 bytes. We saved this in tf_err.
1204 frame->tf_eip -= frame->tf_err;
1209 panic("Unexpected EASYNC return value (for now)");
1212 if (p->p_sysent->sv_errsize) {
1213 if (error >= p->p_sysent->sv_errsize)
1214 error = -1; /* XXX */
1216 error = p->p_sysent->sv_errtbl[error];
1218 frame->tf_eax = error;
1219 frame->tf_eflags |= PSL_C;
1224 * Traced syscall. trapsignal() is not MP aware.
1226 if ((orig_tf_eflags & PSL_T) /*&& !(orig_tf_eflags & PSL_VM)*/) {
1227 MAKEMPSAFE(have_mplock);
1228 frame->tf_eflags &= ~PSL_T;
1229 trapsignal(lp, SIGTRAP, TRAP_TRACE);
1233 * Handle reschedule and other end-of-syscall issues
1235 userret(lp, frame, sticks);
1238 if (KTRPOINT(td, KTR_SYSRET)) {
1239 MAKEMPSAFE(have_mplock);
1240 ktrsysret(lp, code, error, args.sysmsg_result);
1245 * This works because errno is findable through the
1246 * register set. If we ever support an emulation where this
1247 * is not the case, this code will need to be revisited.
1249 STOPEVENT(p, S_SCX, code);
1253 * Release the MP lock if we had to get it
1257 KTR_LOG(kernentry_syscall_ret, lp->lwp_proc->p_pid, lp->lwp_tid, error);
1259 KASSERT(crit_count == td->td_critcount,
1260 ("syscall: critical section count mismatch! %d/%d",
1261 crit_count, td->td_pri));
1262 KASSERT(&td->td_toks_base == td->td_toks_stop,
1263 ("syscall: extra tokens held after trap! %zd",
1264 td->td_toks_stop - &td->td_toks_base));
1269 * NOTE: mplock not held at any point
1272 fork_return(struct lwp *lp, struct trapframe *frame)
1274 frame->tf_eax = 0; /* Child returns zero */
1275 frame->tf_eflags &= ~PSL_C; /* success */
1278 generic_lwp_return(lp, frame);
1279 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1283 * Simplified back end of syscall(), used when returning from fork()
1284 * directly into user mode.
1286 * This code will return back into the fork trampoline code which then
1289 * NOTE: The mplock is not held at any point.
1292 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1294 struct proc *p = lp->lwp_proc;
1297 * Newly forked processes are given a kernel priority. We have to
1298 * adjust the priority to a normal user priority and fake entry
1299 * into the kernel (call userenter()) to install a passive release
1300 * function just in case userret() decides to stop the process. This
1301 * can occur when ^Z races a fork. If we do not install the passive
1302 * release function the current process designation will not be
1303 * released when the thread goes to sleep.
1305 lwkt_setpri_self(TDPRI_USER_NORM);
1306 userenter(lp->lwp_thread, p);
1307 userret(lp, frame, 0);
1309 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1310 ktrsysret(lp, SYS_fork, 0, 0);
1312 lp->lwp_flags |= LWP_PASSIVE_ACQ;
1314 lp->lwp_flags &= ~LWP_PASSIVE_ACQ;
1318 * doreti has turned into this. The frame is directly on the stack. We
1319 * pull everything else we need (fpu and tls context) from the current
1322 * Note on fpu interactions: In a virtual kernel, the fpu context for
1323 * an emulated user mode process is not shared with the virtual kernel's
1324 * fpu context, so we only have to 'stack' fpu contexts within the virtual
1325 * kernel itself, and not even then since the signal() contexts that we care
1326 * about save and restore the FPU state (I think anyhow).
1328 * vmspace_ctl() returns an error only if it had problems instaling the
1329 * context we supplied or problems copying data to/from our VM space.
1332 go_user(struct intrframe *frame)
1334 struct trapframe *tf = (void *)&frame->if_gs;
1338 * Interrupts may be disabled on entry, make sure all signals
1339 * can be received before beginning our loop.
1344 * Switch to the current simulated user process, then call
1345 * user_trap() when we break out of it (usually due to a signal).
1349 * Tell the real kernel whether it is ok to use the FP
1352 * The critical section is required to prevent an interrupt
1353 * from causing a preemptive task switch and changing
1357 if (mdcpu->gd_npxthread == curthread) {
1358 tf->tf_xflags &= ~PGEX_FPFAULT;
1360 tf->tf_xflags |= PGEX_FPFAULT;
1364 * Run emulated user process context. This call interlocks
1365 * with new mailbox signals.
1367 * Set PGEX_U unconditionally, indicating a user frame (the
1368 * bit is normally set only by T_PAGEFLT).
1370 r = vmspace_ctl(&curproc->p_vmspace->vm_pmap, VMSPACE_CTL_RUN,
1371 tf, &curthread->td_savevext);
1373 frame->if_xflags |= PGEX_U;
1375 kprintf("GO USER %d trap %d EVA %08x EIP %08x ESP %08x XFLAGS %02x/%02x\n",
1376 r, tf->tf_trapno, tf->tf_err, tf->tf_eip, tf->tf_esp,
1377 tf->tf_xflags, frame->if_xflags);
1381 panic("vmspace_ctl failed error %d", errno);
1383 if (tf->tf_trapno) {
1387 if (mycpu->gd_reqflags & RQF_AST_MASK) {
1388 tf->tf_trapno = T_ASTFLT;
1396 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1397 * fault (which is then passed back to the virtual kernel) if an attempt is
1398 * made to use the FP unit.
1400 * XXX this is a fairly big hack.
1403 set_vkernel_fp(struct trapframe *frame)
1405 struct thread *td = curthread;
1407 if (frame->tf_xflags & PGEX_FPFAULT) {
1408 td->td_pcb->pcb_flags |= FP_VIRTFP;
1409 if (mdcpu->gd_npxthread == td)
1412 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1417 * Called from vkernel_trap() to fixup the vkernel's syscall
1418 * frame for vmspace_ctl() return.
1421 cpu_vkernel_trap(struct trapframe *frame, int error)
1423 frame->tf_eax = error;
1425 frame->tf_eflags |= PSL_C;
1427 frame->tf_eflags &= ~PSL_C;