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 $
39 * $DragonFly: src/sys/platform/vkernel/i386/trap.c,v 1.29 2008/01/10 22:30:28 nth Exp $
43 * 386 Trap and System call handling
50 #include "opt_ktrace.h"
52 #include <sys/param.h>
53 #include <sys/systm.h>
55 #include <sys/pioctl.h>
56 #include <sys/kernel.h>
57 #include <sys/resourcevar.h>
58 #include <sys/signalvar.h>
59 #include <sys/signal2.h>
60 #include <sys/syscall.h>
61 #include <sys/sysctl.h>
62 #include <sys/sysent.h>
64 #include <sys/vmmeter.h>
65 #include <sys/malloc.h>
67 #include <sys/ktrace.h>
69 #include <sys/upcall.h>
70 #include <sys/vkernel.h>
71 #include <sys/sysproto.h>
72 #include <sys/sysunion.h>
73 #include <sys/vmspace.h>
76 #include <vm/vm_param.h>
79 #include <vm/vm_kern.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_page.h>
82 #include <vm/vm_extern.h>
84 #include <machine/cpu.h>
85 #include <machine/md_var.h>
86 #include <machine/pcb.h>
87 #include <machine/smp.h>
88 #include <machine/tss.h>
89 #include <machine/globaldata.h>
91 #include <machine/vm86.h>
94 #include <sys/msgport2.h>
95 #include <sys/thread2.h>
99 #define MAKEMPSAFE(have_mplock) \
100 if (have_mplock == 0) { \
107 #define MAKEMPSAFE(have_mplock)
111 int (*pmath_emulate) (struct trapframe *);
113 extern int trapwrite (unsigned addr);
115 static int trap_pfault (struct trapframe *, int, vm_offset_t);
116 static void trap_fatal (struct trapframe *, int, vm_offset_t);
117 void dblfault_handler (void);
120 extern inthand_t IDTVEC(syscall);
123 #define MAX_TRAP_MSG 28
124 static char *trap_msg[] = {
126 "privileged instruction fault", /* 1 T_PRIVINFLT */
128 "breakpoint instruction fault", /* 3 T_BPTFLT */
131 "arithmetic trap", /* 6 T_ARITHTRAP */
132 "system forced exception", /* 7 T_ASTFLT */
134 "general protection fault", /* 9 T_PROTFLT */
135 "trace trap", /* 10 T_TRCTRAP */
137 "page fault", /* 12 T_PAGEFLT */
139 "alignment fault", /* 14 T_ALIGNFLT */
143 "integer divide fault", /* 18 T_DIVIDE */
144 "non-maskable interrupt trap", /* 19 T_NMI */
145 "overflow trap", /* 20 T_OFLOW */
146 "FPU bounds check fault", /* 21 T_BOUND */
147 "FPU device not available", /* 22 T_DNA */
148 "double fault", /* 23 T_DOUBLEFLT */
149 "FPU operand fetch fault", /* 24 T_FPOPFLT */
150 "invalid TSS fault", /* 25 T_TSSFLT */
151 "segment not present fault", /* 26 T_SEGNPFLT */
152 "stack fault", /* 27 T_STKFLT */
153 "machine check trap", /* 28 T_MCHK */
157 static int ddb_on_nmi = 1;
158 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
159 &ddb_on_nmi, 0, "Go to DDB on NMI");
161 static int panic_on_nmi = 1;
162 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
163 &panic_on_nmi, 0, "Panic on NMI");
164 static int fast_release;
165 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
166 &fast_release, 0, "Passive Release was optimal");
167 static int slow_release;
168 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
169 &slow_release, 0, "Passive Release was nonoptimal");
171 static int syscall_mpsafe = 0;
172 SYSCTL_INT(_kern, OID_AUTO, syscall_mpsafe, CTLFLAG_RW,
173 &syscall_mpsafe, 0, "Allow MPSAFE marked syscalls to run without BGL");
174 TUNABLE_INT("kern.syscall_mpsafe", &syscall_mpsafe);
175 static int trap_mpsafe = 0;
176 SYSCTL_INT(_kern, OID_AUTO, trap_mpsafe, CTLFLAG_RW,
177 &trap_mpsafe, 0, "Allow traps to mostly run without the BGL");
178 TUNABLE_INT("kern.trap_mpsafe", &trap_mpsafe);
181 MALLOC_DEFINE(M_SYSMSG, "sysmsg", "sysmsg structure");
182 extern int max_sysmsg;
185 * Passive USER->KERNEL transition. This only occurs if we block in the
186 * kernel while still holding our userland priority. We have to fixup our
187 * priority in order to avoid potential deadlocks before we allow the system
188 * to switch us to another thread.
191 passive_release(struct thread *td)
193 struct lwp *lp = td->td_lwp;
195 td->td_release = NULL;
196 lwkt_setpri_self(TDPRI_KERN_USER);
197 lp->lwp_proc->p_usched->release_curproc(lp);
201 * userenter() passively intercepts the thread switch function to increase
202 * the thread priority from a user priority to a kernel priority, reducing
203 * syscall and trap overhead for the case where no switch occurs.
207 userenter(struct thread *curtd)
209 curtd->td_release = passive_release;
213 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
214 * must be completed before we can return to or try to return to userland.
216 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
217 * arithmatic on the delta calculation so the absolute tick values are
218 * truncated to an integer.
221 userret(struct lwp *lp, struct trapframe *frame, int sticks)
223 struct proc *p = lp->lwp_proc;
227 * Charge system time if profiling. Note: times are in microseconds.
228 * This may do a copyout and block, so do it first even though it
229 * means some system time will be charged as user time.
231 if (p->p_flag & P_PROFIL) {
232 addupc_task(p, frame->tf_eip,
233 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
238 * If the jungle wants us dead, so be it.
240 if (lp->lwp_flag & LWP_WEXIT)
244 * Block here if we are in a stopped state.
246 if (p->p_stat == SSTOP) {
254 * Post any pending upcalls
256 if (p->p_flag & P_UPCALLPEND) {
258 p->p_flag &= ~P_UPCALLPEND;
265 * Post any pending signals
267 if ((sig = CURSIG(lp)) != 0) {
275 * block here if we are swapped out, but still process signals
276 * (such as SIGKILL). proc0 (the swapin scheduler) is already
277 * aware of our situation, we do not have to wake it up.
279 if (p->p_flag & P_SWAPPEDOUT) {
281 p->p_flag |= P_SWAPWAIT;
283 if (p->p_flag & P_SWAPWAIT)
284 tsleep(p, PCATCH, "SWOUT", 0);
285 p->p_flag &= ~P_SWAPWAIT;
291 * Make sure postsig() handled request to restore old signal mask after
292 * running signal handler.
294 KKASSERT((lp->lwp_flag & LWP_OLDMASK) == 0);
298 * Cleanup from userenter and any passive release that might have occured.
299 * We must reclaim the current-process designation before we can return
300 * to usermode. We also handle both LWKT and USER reschedule requests.
303 userexit(struct lwp *lp)
305 struct thread *td = lp->lwp_thread;
306 globaldata_t gd = td->td_gd;
310 * If a user reschedule is requested force a new process to be
311 * chosen by releasing the current process. Our process will only
312 * be chosen again if it has a considerably better priority.
314 if (user_resched_wanted())
315 lp->lwp_proc->p_usched->release_curproc(lp);
319 * Handle a LWKT reschedule request first. Since our passive release
320 * is still in place we do not have to do anything special.
322 if (lwkt_resched_wanted())
326 * Acquire the current process designation for this user scheduler
327 * on this cpu. This will also handle any user-reschedule requests.
329 lp->lwp_proc->p_usched->acquire_curproc(lp);
330 /* We may have switched cpus on acquisition */
334 * Reduce our priority in preparation for a return to userland. If
335 * our passive release function was still in place, our priority was
336 * never raised and does not need to be reduced.
338 if (td->td_release == NULL)
339 lwkt_setpri_self(TDPRI_USER_NORM);
340 td->td_release = NULL;
343 * After reducing our priority there might be other kernel-level
344 * LWKTs that now have a greater priority. Run them as necessary.
345 * We don't have to worry about losing cpu to userland because
346 * we still control the current-process designation and we no longer
347 * have a passive release function installed.
349 if (lwkt_checkpri_self())
354 * Exception, fault, and trap interface to the kernel.
355 * This common code is called from assembly language IDT gate entry
356 * routines that prepare a suitable stack frame, and restore this
357 * frame after the exception has been processed.
359 * This function is also called from doreti in an interlock to handle ASTs.
360 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
362 * NOTE! We have to retrieve the fault address prior to obtaining the
363 * MP lock because get_mplock() may switch out. YYY cr2 really ought
364 * to be retrieved by the assembly code, not here.
366 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
367 * if an attempt is made to switch from a fast interrupt or IPI. This is
368 * necessary to properly take fatal kernel traps on SMP machines if
369 * get_mplock() has to block.
373 user_trap(struct trapframe *frame)
375 struct globaldata *gd = mycpu;
376 struct thread *td = gd->gd_curthread;
377 struct lwp *lp = td->td_lwp;
380 int i = 0, ucode = 0, type, code;
385 int crit_count = td->td_pri & ~TDPRI_MASK;
392 * This is a bad kludge to avoid changing the various trapframe
393 * structures. Because we are enabled as a virtual kernel,
394 * the original tf_err field will be passed to us shifted 16
395 * over in the tf_trapno field for T_PAGEFLT.
397 if (frame->tf_trapno == T_PAGEFLT)
402 kprintf("USER_TRAP AT %08x xflags %d trapno %d eva %08x\n",
403 frame->tf_eip, 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_SYSCALL80) {
417 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
418 ++gd->gd_trap_nesting_level;
419 MAKEMPSAFE(have_mplock);
420 trap_fatal(frame, TRUE, eva);
421 --gd->gd_trap_nesting_level;
426 ++gd->gd_trap_nesting_level;
428 if (trap_mpsafe == 0)
429 MAKEMPSAFE(have_mplock);
432 --gd->gd_trap_nesting_level;
434 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
437 type = frame->tf_trapno;
438 code = frame->tf_err;
442 sticks = (int)td->td_sticks;
443 lp->lwp_md.md_regs = frame;
446 case T_PRIVINFLT: /* privileged instruction fault */
451 case T_BPTFLT: /* bpt instruction fault */
452 case T_TRCTRAP: /* trace trap */
453 frame->tf_eflags &= ~PSL_T;
457 case T_ARITHTRAP: /* arithmetic trap */
462 case T_ASTFLT: /* Allow process switch */
463 mycpu->gd_cnt.v_soft++;
464 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
465 atomic_clear_int_nonlocked(&mycpu->gd_reqflags,
467 addupc_task(p, p->p_prof.pr_addr,
473 * The following two traps can happen in
474 * vm86 mode, and, if so, we want to handle
477 case T_PROTFLT: /* general protection fault */
478 case T_STKFLT: /* stack fault */
480 if (frame->tf_eflags & PSL_VM) {
481 i = vm86_emulate((struct vm86frame *)frame);
489 case T_SEGNPFLT: /* segment not present fault */
490 case T_TSSFLT: /* invalid TSS fault */
491 case T_DOUBLEFLT: /* double fault */
493 ucode = code + BUS_SEGM_FAULT ;
497 case T_PAGEFLT: /* page fault */
498 MAKEMPSAFE(have_mplock);
499 i = trap_pfault(frame, TRUE, eva);
502 #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)
568 * The kernel may have switched out the FP unit's
569 * state, causing the user process to take a fault
570 * when it tries to use the FP unit. Restore the
576 if (!pmath_emulate) {
578 ucode = FPE_FPU_NP_TRAP;
581 i = (*pmath_emulate)(frame);
583 if (!(frame->tf_eflags & PSL_T))
585 frame->tf_eflags &= ~PSL_T;
588 /* else ucode = emulator_only_knows() XXX */
591 case T_FPOPFLT: /* FPU operand fetch fault */
596 case T_XMMFLT: /* SIMD floating-point exception */
603 * Virtual kernel intercept - if the fault is directly related to a
604 * VM context managed by a virtual kernel then let the virtual kernel
607 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
608 vkernel_trap(lp, frame);
613 * Translate fault for emulators (e.g. Linux)
615 if (*p->p_sysent->sv_transtrap)
616 i = (*p->p_sysent->sv_transtrap)(i, type);
618 MAKEMPSAFE(have_mplock);
619 trapsignal(lp, i, ucode);
622 if (type <= MAX_TRAP_MSG) {
623 uprintf("fatal process exception: %s",
625 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
626 uprintf(", fault VA = 0x%lx", (u_long)eva);
633 KASSERT(td->td_mpcount == have_mplock, ("badmpcount trap/end from %p", (void *)frame->tf_eip));
635 userret(lp, frame, sticks);
643 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
644 ("syscall: critical section count mismatch! %d/%d",
645 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
650 kern_trap(struct trapframe *frame)
652 struct globaldata *gd = mycpu;
653 struct thread *td = gd->gd_curthread;
656 int i = 0, ucode = 0, type, code;
661 int crit_count = td->td_pri & ~TDPRI_MASK;
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;
683 ++gd->gd_trap_nesting_level;
686 if (trap_mpsafe == 0)
687 MAKEMPSAFE(have_mplock);
690 --gd->gd_trap_nesting_level;
692 type = frame->tf_trapno;
693 code = frame->tf_err;
701 case T_PAGEFLT: /* page fault */
702 MAKEMPSAFE(have_mplock);
703 trap_pfault(frame, FALSE, eva);
709 * The kernel may be using npx for copying or other
712 panic("kernel NPX should not happen");
718 case T_PROTFLT: /* general protection fault */
719 case T_SEGNPFLT: /* segment not present fault */
721 * Invalid segment selectors and out of bounds
722 * %eip's and %esp's can be set up in user mode.
723 * This causes a fault in kernel mode when the
724 * kernel tries to return to user mode. We want
725 * to get this fault so that we can fix the
726 * problem here and not have to check all the
727 * selectors and pointers when the user changes
730 if (mycpu->gd_intr_nesting_level == 0) {
731 if (td->td_pcb->pcb_onfault) {
733 (register_t)td->td_pcb->pcb_onfault;
741 * PSL_NT can be set in user mode and isn't cleared
742 * automatically when the kernel is entered. This
743 * causes a TSS fault when the kernel attempts to
744 * `iret' because the TSS link is uninitialized. We
745 * want to get this fault so that we can fix the
746 * problem here and not every time the kernel is
749 if (frame->tf_eflags & PSL_NT) {
750 frame->tf_eflags &= ~PSL_NT;
755 case T_TRCTRAP: /* trace trap */
757 if (frame->tf_eip == (int)IDTVEC(syscall)) {
759 * We've just entered system mode via the
760 * syscall lcall. Continue single stepping
761 * silently until the syscall handler has
766 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
768 * The syscall handler has now saved the
769 * flags. Stop single stepping it.
771 frame->tf_eflags &= ~PSL_T;
777 * Ignore debug register trace traps due to
778 * accesses in the user's address space, which
779 * can happen under several conditions such as
780 * if a user sets a watchpoint on a buffer and
781 * then passes that buffer to a system call.
782 * We still want to get TRCTRAPS for addresses
783 * in kernel space because that is useful when
784 * debugging the kernel.
786 if (user_dbreg_trap()) {
788 * Reset breakpoint bits because the
791 load_dr6(rdr6() & 0xfffffff0);
796 * Fall through (TRCTRAP kernel mode, kernel address)
800 * If DDB is enabled, let it handle the debugger trap.
801 * Otherwise, debugger traps "can't happen".
804 MAKEMPSAFE(have_mplock);
805 if (kdb_trap (type, 0, frame))
810 MAKEMPSAFE(have_mplock);
811 trap_fatal(frame, FALSE, eva);
814 MAKEMPSAFE(have_mplock);
815 trap_fatal(frame, FALSE, eva);
819 * Ignore this trap generated from a spurious SIGTRAP.
821 * single stepping in / syscalls leads to spurious / SIGTRAP
824 * Haiku (c) 2007 Simon 'corecode' Schubert
830 * Translate fault for emulators (e.g. Linux)
832 if (*p->p_sysent->sv_transtrap)
833 i = (*p->p_sysent->sv_transtrap)(i, type);
835 MAKEMPSAFE(have_mplock);
836 trapsignal(lp, i, ucode);
839 if (type <= MAX_TRAP_MSG) {
840 uprintf("fatal process exception: %s",
842 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
843 uprintf(", fault VA = 0x%lx", (u_long)eva);
855 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
856 ("syscall: critical section count mismatch! %d/%d",
857 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
862 trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
865 struct vmspace *vm = NULL;
869 thread_t td = curthread;
870 struct lwp *lp = td->td_lwp;
872 va = trunc_page(eva);
873 if (usermode == FALSE) {
875 * This is a fault on kernel virtual memory.
880 * This is a fault on non-kernel virtual memory.
881 * vm is initialized above to NULL. If curproc is NULL
882 * or curproc->p_vmspace is NULL the fault is fatal.
885 vm = lp->lwp_vmspace;
893 if (frame->tf_xflags & PGEX_W)
894 ftype = VM_PROT_READ | VM_PROT_WRITE;
896 ftype = VM_PROT_READ;
898 if (map != &kernel_map) {
900 * Keep swapout from messing with us during this
906 * Grow the stack if necessary
908 /* grow_stack returns false only if va falls into
909 * a growable stack region and the stack growth
910 * fails. It returns true if va was not within
911 * a growable stack region, or if the stack
914 if (!grow_stack (lp->lwp_proc, va)) {
920 /* Fault in the user page: */
921 rv = vm_fault(map, va, ftype,
922 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
928 * Don't have to worry about process locking or stacks in the kernel.
930 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
933 if (rv == KERN_SUCCESS)
937 if (td->td_gd->gd_intr_nesting_level == 0 &&
938 td->td_pcb->pcb_onfault) {
939 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
942 trap_fatal(frame, usermode, eva);
945 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
949 trap_fatal(struct trapframe *frame, int usermode, vm_offset_t eva)
951 int code, type, ss, esp;
953 code = frame->tf_xflags;
954 type = frame->tf_trapno;
956 if (type <= MAX_TRAP_MSG) {
957 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
958 type, trap_msg[type],
959 (usermode ? "user" : "kernel"));
962 /* two separate prints in case of a trap on an unmapped page */
963 kprintf("mp_lock = %08x; ", mp_lock);
964 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
966 if (type == T_PAGEFLT) {
967 kprintf("fault virtual address = 0x%x\n", eva);
968 kprintf("fault code = %s %s, %s\n",
969 usermode ? "user" : "supervisor",
970 code & PGEX_W ? "write" : "read",
971 code & PGEX_P ? "protection violation" : "page not present");
973 kprintf("instruction pointer = 0x%x:0x%x\n",
974 frame->tf_cs & 0xffff, frame->tf_eip);
976 ss = frame->tf_ss & 0xffff;
979 ss = GSEL(GDATA_SEL, SEL_KPL);
980 esp = (int)&frame->tf_esp;
982 kprintf("stack pointer = 0x%x:0x%x\n", ss, esp);
983 kprintf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
984 kprintf("processor eflags = ");
985 if (frame->tf_eflags & PSL_T)
986 kprintf("trace trap, ");
987 if (frame->tf_eflags & PSL_I)
988 kprintf("interrupt enabled, ");
989 if (frame->tf_eflags & PSL_NT)
990 kprintf("nested task, ");
991 if (frame->tf_eflags & PSL_RF)
994 if (frame->tf_eflags & PSL_VM)
997 kprintf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
998 kprintf("current process = ");
1000 kprintf("%lu (%s)\n",
1001 (u_long)curproc->p_pid, curproc->p_comm ?
1002 curproc->p_comm : "");
1006 kprintf("current thread = pri %d ", curthread->td_pri);
1007 if (curthread->td_pri >= TDPRI_CRIT)
1013 * we probably SHOULD have stopped the other CPUs before now!
1014 * another CPU COULD have been touching cpl at this moment...
1016 kprintf(" <- SMP: XXX");
1025 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1028 kprintf("trap number = %d\n", type);
1029 if (type <= MAX_TRAP_MSG)
1030 panic("%s", trap_msg[type]);
1032 panic("unknown/reserved trap");
1036 * Double fault handler. Called when a fault occurs while writing
1037 * a frame for a trap/exception onto the stack. This usually occurs
1038 * when the stack overflows (such is the case with infinite recursion,
1041 * XXX Note that the current PTD gets replaced by IdlePTD when the
1042 * task switch occurs. This means that the stack that was active at
1043 * the time of the double fault is not available at <kstack> unless
1044 * the machine was idle when the double fault occurred. The downside
1045 * of this is that "trace <ebp>" in ddb won't work.
1048 dblfault_handler(void)
1050 struct mdglobaldata *gd = mdcpu;
1052 kprintf("\nFatal double fault:\n");
1053 kprintf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1054 kprintf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1055 kprintf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
1057 /* two separate prints in case of a trap on an unmapped page */
1058 kprintf("mp_lock = %08x; ", mp_lock);
1059 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
1061 panic("double fault");
1065 * Compensate for 386 brain damage (missing URKR).
1066 * This is a little simpler than the pagefault handler in trap() because
1067 * it the page tables have already been faulted in and high addresses
1068 * are thrown out early for other reasons.
1071 trapwrite(unsigned addr)
1078 va = trunc_page((vm_offset_t)addr);
1080 * XXX - MAX is END. Changed > to >= for temp. fix.
1082 if (va >= VM_MAX_USER_ADDRESS)
1085 lp = curthread->td_lwp;
1086 vm = lp->lwp_vmspace;
1088 PHOLD(lp->lwp_proc);
1090 if (!grow_stack (lp->lwp_proc, va)) {
1091 PRELE(lp->lwp_proc);
1096 * fault the data page
1098 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY);
1100 PRELE(lp->lwp_proc);
1102 if (rv != KERN_SUCCESS)
1109 * syscall2 - MP aware system call request C handler
1111 * A system call is essentially treated as a trap except that the
1112 * MP lock is not held on entry or return. We are responsible for
1113 * obtaining the MP lock if necessary and for handling ASTs
1114 * (e.g. a task switch) prior to return.
1116 * In general, only simple access and manipulation of curproc and
1117 * the current stack is allowed without having to hold MP lock.
1119 * MPSAFE - note that large sections of this routine are run without
1124 syscall2(struct trapframe *frame)
1126 struct thread *td = curthread;
1127 struct proc *p = td->td_proc;
1128 struct lwp *lp = td->td_lwp;
1130 struct sysent *callp;
1131 register_t orig_tf_eflags;
1136 int crit_count = td->td_pri & ~TDPRI_MASK;
1139 int have_mplock = 0;
1142 union sysunion args;
1145 KASSERT(td->td_mpcount == 0, ("badmpcount syscall2 from %p", (void *)frame->tf_eip));
1146 if (syscall_mpsafe == 0)
1147 MAKEMPSAFE(have_mplock);
1149 userenter(td); /* lazy raise our priority */
1154 sticks = (int)td->td_sticks;
1155 orig_tf_eflags = frame->tf_eflags;
1158 * Virtual kernel intercept - if a VM context managed by a virtual
1159 * kernel issues a system call the virtual kernel handles it, not us.
1160 * Restore the virtual kernel context and return from its system
1161 * call. The current frame is copied out to the virtual kernel.
1163 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1164 error = vkernel_trap(lp, frame);
1165 frame->tf_eax = error;
1167 frame->tf_eflags |= PSL_C;
1168 error = EJUSTRETURN;
1173 * Get the system call parameters and account for time
1175 lp->lwp_md.md_regs = frame;
1176 params = (caddr_t)frame->tf_esp + sizeof(int);
1177 code = frame->tf_eax;
1179 if (p->p_sysent->sv_prepsyscall) {
1180 (*p->p_sysent->sv_prepsyscall)(
1181 frame, (int *)(&args.nosys.sysmsg + 1),
1185 * Need to check if this is a 32 bit or 64 bit syscall.
1186 * fuword is MP aware.
1188 if (code == SYS_syscall) {
1190 * Code is first argument, followed by actual args.
1192 code = fuword(params);
1193 params += sizeof(int);
1194 } else if (code == SYS___syscall) {
1196 * Like syscall, but code is a quad, so as to maintain
1197 * quad alignment for the rest of the arguments.
1199 code = fuword(params);
1200 params += sizeof(quad_t);
1204 code &= p->p_sysent->sv_mask;
1205 if (code >= p->p_sysent->sv_size)
1206 callp = &p->p_sysent->sv_table[0];
1208 callp = &p->p_sysent->sv_table[code];
1210 narg = callp->sy_narg & SYF_ARGMASK;
1213 * copyin is MP aware, but the tracing code is not
1215 if (narg && params) {
1216 error = copyin(params, (caddr_t)(&args.nosys.sysmsg + 1),
1217 narg * sizeof(register_t));
1220 if (KTRPOINT(td, KTR_SYSCALL)) {
1221 MAKEMPSAFE(have_mplock);
1223 ktrsyscall(p, code, narg,
1224 (void *)(&args.nosys.sysmsg + 1));
1232 if (KTRPOINT(td, KTR_SYSCALL)) {
1233 MAKEMPSAFE(have_mplock);
1234 ktrsyscall(p, code, narg, (void *)(&args.nosys.sysmsg + 1));
1239 * For traditional syscall code edx is left untouched when 32 bit
1240 * results are returned. Since edx is loaded from fds[1] when the
1241 * system call returns we pre-set it here.
1243 args.sysmsg_fds[0] = 0;
1244 args.sysmsg_fds[1] = frame->tf_edx;
1247 * The syscall might manipulate the trap frame. If it does it
1248 * will probably return EJUSTRETURN.
1250 args.sysmsg_frame = frame;
1252 STOPEVENT(p, S_SCE, narg); /* MP aware */
1256 * Try to run the syscall without the MP lock if the syscall
1257 * is MP safe. We have to obtain the MP lock no matter what if
1260 if ((callp->sy_narg & SYF_MPSAFE) == 0)
1261 MAKEMPSAFE(have_mplock);
1264 error = (*callp->sy_call)(&args);
1267 kprintf("system call %d returned %d\n", code, error);
1272 * MP SAFE (we may or may not have the MP lock at this point)
1277 * Reinitialize proc pointer `p' as it may be different
1278 * if this is a child returning from fork syscall.
1281 lp = curthread->td_lwp;
1282 frame->tf_eax = args.sysmsg_fds[0];
1283 frame->tf_edx = args.sysmsg_fds[1];
1284 frame->tf_eflags &= ~PSL_C;
1288 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1289 * int 0x80 is 2 bytes. We saved this in tf_err.
1291 frame->tf_eip -= frame->tf_err;
1296 panic("Unexpected EASYNC return value (for now)");
1299 if (p->p_sysent->sv_errsize) {
1300 if (error >= p->p_sysent->sv_errsize)
1301 error = -1; /* XXX */
1303 error = p->p_sysent->sv_errtbl[error];
1305 frame->tf_eax = error;
1306 frame->tf_eflags |= PSL_C;
1311 * Traced syscall. trapsignal() is not MP aware.
1313 if ((orig_tf_eflags & PSL_T) /*&& !(orig_tf_eflags & PSL_VM)*/) {
1314 MAKEMPSAFE(have_mplock);
1315 frame->tf_eflags &= ~PSL_T;
1316 trapsignal(lp, SIGTRAP, 0);
1320 * Handle reschedule and other end-of-syscall issues
1322 userret(lp, frame, sticks);
1325 if (KTRPOINT(td, KTR_SYSRET)) {
1326 MAKEMPSAFE(have_mplock);
1327 ktrsysret(p, code, error, args.sysmsg_result);
1332 * This works because errno is findable through the
1333 * register set. If we ever support an emulation where this
1334 * is not the case, this code will need to be revisited.
1336 STOPEVENT(p, S_SCX, code);
1341 * Release the MP lock if we had to get it
1343 KASSERT(td->td_mpcount == have_mplock,
1344 ("badmpcount syscall2/end from %p", (void *)frame->tf_eip));
1349 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
1350 ("syscall: critical section count mismatch! %d/%d",
1351 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
1356 fork_return(struct lwp *lp, struct trapframe *frame)
1358 frame->tf_eax = 0; /* Child returns zero */
1359 frame->tf_eflags &= ~PSL_C; /* success */
1362 generic_lwp_return(lp, frame);
1366 * Simplified back end of syscall(), used when returning from fork()
1367 * or lwp_create() directly into user mode. MP lock is held on entry and
1368 * should be released on return. This code will return back into the fork
1369 * trampoline code which then runs doreti.
1372 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1374 struct proc *p = lp->lwp_proc;
1377 * Newly forked processes are given a kernel priority. We have to
1378 * adjust the priority to a normal user priority and fake entry
1379 * into the kernel (call userenter()) to install a passive release
1380 * function just in case userret() decides to stop the process. This
1381 * can occur when ^Z races a fork. If we do not install the passive
1382 * release function the current process designation will not be
1383 * released when the thread goes to sleep.
1385 lwkt_setpri_self(TDPRI_USER_NORM);
1386 userenter(lp->lwp_thread);
1387 userret(lp, frame, 0);
1389 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1390 ktrsysret(p, SYS_fork, 0, 0);
1392 p->p_flag |= P_PASSIVE_ACQ;
1394 p->p_flag &= ~P_PASSIVE_ACQ;
1396 KKASSERT(lp->lwp_thread->td_mpcount == 1);
1402 * doreti has turned into this. The frame is directly on the stack. We
1403 * pull everything else we need (fpu and tls context) from the current
1406 * Note on fpu interactions: In a virtual kernel, the fpu context for
1407 * an emulated user mode process is not shared with the virtual kernel's
1408 * fpu context, so we only have to 'stack' fpu contexts within the virtual
1409 * kernel itself, and not even then since the signal() contexts that we care
1410 * about save and restore the FPU state (I think anyhow).
1412 * vmspace_ctl() returns an error only if it had problems instaling the
1413 * context we supplied or problems copying data to/from our VM space.
1416 go_user(struct intrframe *frame)
1418 struct trapframe *tf = (void *)&frame->if_gs;
1422 * Interrupts may be disabled on entry, make sure all signals
1423 * can be received before beginning our loop.
1428 * Switch to the current simulated user process, then call
1429 * user_trap() when we break out of it (usually due to a signal).
1433 * Tell the real kernel whether it is ok to use the FP
1436 if (mdcpu->gd_npxthread == curthread) {
1437 tf->tf_xflags &= ~PGEX_FPFAULT;
1439 tf->tf_xflags |= PGEX_FPFAULT;
1443 * Run emulated user process context. This call interlocks
1444 * with new mailbox signals.
1446 * Set PGEX_U unconditionally, indicating a user frame (the
1447 * bit is normally set only by T_PAGEFLT).
1449 r = vmspace_ctl(&curproc->p_vmspace->vm_pmap, VMSPACE_CTL_RUN,
1450 tf, &curthread->td_savevext);
1451 frame->if_xflags |= PGEX_U;
1453 kprintf("GO USER %d trap %d EVA %08x EIP %08x ESP %08x XFLAGS %02x/%02x\n",
1454 r, tf->tf_trapno, tf->tf_err, tf->tf_eip, tf->tf_esp,
1455 tf->tf_xflags, frame->if_xflags);
1459 panic("vmspace_ctl failed");
1461 if (tf->tf_trapno) {
1463 } else if (mycpu->gd_reqflags & RQF_AST_MASK) {
1464 tf->tf_trapno = T_ASTFLT;
1473 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1474 * fault (which is then passed back to the virtual kernel) if an attempt is
1475 * made to use the FP unit.
1477 * XXX this is a fairly big hack.
1480 set_vkernel_fp(struct trapframe *frame)
1482 struct thread *td = curthread;
1484 if (frame->tf_xflags & PGEX_FPFAULT) {
1485 td->td_pcb->pcb_flags |= FP_VIRTFP;
1486 if (mdcpu->gd_npxthread == td)
1489 td->td_pcb->pcb_flags &= ~FP_VIRTFP;