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 1.32 2008/05/06 18:43:02 dillon 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 = 1;
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 = 1;
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) {
243 rel_mplock(); /* NOT REACHED */
247 * Block here if we are in a stopped state.
249 if (p->p_stat == SSTOP) {
257 * Post any pending upcalls
259 if (p->p_flag & P_UPCALLPEND) {
261 p->p_flag &= ~P_UPCALLPEND;
268 * Post any pending signals
270 if ((sig = CURSIG(lp)) != 0) {
278 * block here if we are swapped out, but still process signals
279 * (such as SIGKILL). proc0 (the swapin scheduler) is already
280 * aware of our situation, we do not have to wake it up.
282 if (p->p_flag & P_SWAPPEDOUT) {
284 p->p_flag |= P_SWAPWAIT;
286 if (p->p_flag & P_SWAPWAIT)
287 tsleep(p, PCATCH, "SWOUT", 0);
288 p->p_flag &= ~P_SWAPWAIT;
294 * Make sure postsig() handled request to restore old signal mask after
295 * running signal handler.
297 KKASSERT((lp->lwp_flag & LWP_OLDMASK) == 0);
301 * Cleanup from userenter and any passive release that might have occured.
302 * We must reclaim the current-process designation before we can return
303 * to usermode. We also handle both LWKT and USER reschedule requests.
306 userexit(struct lwp *lp)
308 struct thread *td = lp->lwp_thread;
309 globaldata_t gd = td->td_gd;
313 * If a user reschedule is requested force a new process to be
314 * chosen by releasing the current process. Our process will only
315 * be chosen again if it has a considerably better priority.
317 if (user_resched_wanted())
318 lp->lwp_proc->p_usched->release_curproc(lp);
322 * Handle a LWKT reschedule request first. Since our passive release
323 * is still in place we do not have to do anything special.
325 if (lwkt_resched_wanted())
329 * Acquire the current process designation for this user scheduler
330 * on this cpu. This will also handle any user-reschedule requests.
332 lp->lwp_proc->p_usched->acquire_curproc(lp);
333 /* We may have switched cpus on acquisition */
337 * Reduce our priority in preparation for a return to userland. If
338 * our passive release function was still in place, our priority was
339 * never raised and does not need to be reduced.
341 if (td->td_release == NULL)
342 lwkt_setpri_self(TDPRI_USER_NORM);
343 td->td_release = NULL;
346 * After reducing our priority there might be other kernel-level
347 * LWKTs that now have a greater priority. Run them as necessary.
348 * We don't have to worry about losing cpu to userland because
349 * we still control the current-process designation and we no longer
350 * have a passive release function installed.
352 if (lwkt_checkpri_self())
357 * Exception, fault, and trap interface to the kernel.
358 * This common code is called from assembly language IDT gate entry
359 * routines that prepare a suitable stack frame, and restore this
360 * frame after the exception has been processed.
362 * This function is also called from doreti in an interlock to handle ASTs.
363 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
365 * NOTE! We have to retrieve the fault address prior to obtaining the
366 * MP lock because get_mplock() may switch out. YYY cr2 really ought
367 * to be retrieved by the assembly code, not here.
369 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
370 * if an attempt is made to switch from a fast interrupt or IPI. This is
371 * necessary to properly take fatal kernel traps on SMP machines if
372 * get_mplock() has to block.
376 user_trap(struct trapframe *frame)
378 struct globaldata *gd = mycpu;
379 struct thread *td = gd->gd_curthread;
380 struct lwp *lp = td->td_lwp;
383 int i = 0, ucode = 0, type, code;
388 int crit_count = td->td_pri & ~TDPRI_MASK;
395 * This is a bad kludge to avoid changing the various trapframe
396 * structures. Because we are enabled as a virtual kernel,
397 * the original tf_err field will be passed to us shifted 16
398 * over in the tf_trapno field for T_PAGEFLT.
400 if (frame->tf_trapno == T_PAGEFLT)
405 kprintf("USER_TRAP AT %08x xflags %d trapno %d eva %08x\n",
406 frame->tf_eip, frame->tf_xflags, frame->tf_trapno, eva);
410 * Everything coming from user mode runs through user_trap,
411 * including system calls.
413 if (frame->tf_trapno == T_SYSCALL80) {
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 ++gd->gd_trap_nesting_level;
431 if (trap_mpsafe == 0)
432 MAKEMPSAFE(have_mplock);
435 --gd->gd_trap_nesting_level;
437 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
440 type = frame->tf_trapno;
441 code = frame->tf_err;
445 sticks = (int)td->td_sticks;
446 lp->lwp_md.md_regs = frame;
449 case T_PRIVINFLT: /* privileged instruction fault */
454 case T_BPTFLT: /* bpt instruction fault */
455 case T_TRCTRAP: /* trace trap */
456 frame->tf_eflags &= ~PSL_T;
460 case T_ARITHTRAP: /* arithmetic trap */
465 case T_ASTFLT: /* Allow process switch */
466 mycpu->gd_cnt.v_soft++;
467 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
468 atomic_clear_int_nonlocked(&mycpu->gd_reqflags,
470 addupc_task(p, p->p_prof.pr_addr,
476 * The following two traps can happen in
477 * vm86 mode, and, if so, we want to handle
480 case T_PROTFLT: /* general protection fault */
481 case T_STKFLT: /* stack fault */
483 if (frame->tf_eflags & PSL_VM) {
484 i = vm86_emulate((struct vm86frame *)frame);
492 case T_SEGNPFLT: /* segment not present fault */
493 case T_TSSFLT: /* invalid TSS fault */
494 case T_DOUBLEFLT: /* double fault */
496 ucode = code + BUS_SEGM_FAULT ;
500 case T_PAGEFLT: /* page fault */
501 MAKEMPSAFE(have_mplock);
502 i = trap_pfault(frame, TRUE, eva);
505 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
515 case T_DIVIDE: /* integer divide fault */
522 MAKEMPSAFE(have_mplock);
523 /* machine/parity/power fail/"kitchen sink" faults */
524 if (isa_nmi(code) == 0) {
527 * NMI can be hooked up to a pushbutton
531 kprintf ("NMI ... going to debugger\n");
532 kdb_trap (type, 0, frame);
536 } else if (panic_on_nmi)
537 panic("NMI indicates hardware failure");
539 #endif /* NISA > 0 */
541 case T_OFLOW: /* integer overflow fault */
546 case T_BOUND: /* bounds check fault */
553 * Virtual kernel intercept - pass the DNA exception
554 * to the (emulated) virtual kernel if it asked to handle
555 * it. This occurs when the virtual kernel is holding
556 * onto the FP context for a different emulated
557 * process then the one currently running.
559 * We must still call npxdna() since we may have
560 * saved FP state that the (emulated) virtual kernel
561 * needs to hand over to a different emulated process.
563 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
564 (td->td_pcb->pcb_flags & FP_VIRTFP)
571 * The kernel may have switched out the FP unit's
572 * state, causing the user process to take a fault
573 * when it tries to use the FP unit. Restore the
579 if (!pmath_emulate) {
581 ucode = FPE_FPU_NP_TRAP;
584 i = (*pmath_emulate)(frame);
586 if (!(frame->tf_eflags & PSL_T))
588 frame->tf_eflags &= ~PSL_T;
591 /* else ucode = emulator_only_knows() XXX */
594 case T_FPOPFLT: /* FPU operand fetch fault */
599 case T_XMMFLT: /* SIMD floating-point exception */
606 * Virtual kernel intercept - if the fault is directly related to a
607 * VM context managed by a virtual kernel then let the virtual kernel
610 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
611 vkernel_trap(lp, frame);
616 * Translate fault for emulators (e.g. Linux)
618 if (*p->p_sysent->sv_transtrap)
619 i = (*p->p_sysent->sv_transtrap)(i, type);
621 MAKEMPSAFE(have_mplock);
622 trapsignal(lp, i, ucode);
625 if (type <= MAX_TRAP_MSG) {
626 uprintf("fatal process exception: %s",
628 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
629 uprintf(", fault VA = 0x%lx", (u_long)eva);
636 KASSERT(td->td_mpcount == have_mplock, ("badmpcount trap/end from %p", (void *)frame->tf_eip));
638 userret(lp, frame, sticks);
646 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
647 ("syscall: critical section count mismatch! %d/%d",
648 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
653 kern_trap(struct trapframe *frame)
655 struct globaldata *gd = mycpu;
656 struct thread *td = gd->gd_curthread;
659 int i = 0, ucode = 0, type, code;
664 int crit_count = td->td_pri & ~TDPRI_MASK;
671 if (frame->tf_trapno == T_PAGEFLT)
678 ++gd->gd_trap_nesting_level;
679 MAKEMPSAFE(have_mplock);
680 trap_fatal(frame, FALSE, eva);
681 --gd->gd_trap_nesting_level;
686 ++gd->gd_trap_nesting_level;
689 if (trap_mpsafe == 0)
690 MAKEMPSAFE(have_mplock);
693 --gd->gd_trap_nesting_level;
695 type = frame->tf_trapno;
696 code = frame->tf_err;
704 case T_PAGEFLT: /* page fault */
705 MAKEMPSAFE(have_mplock);
706 trap_pfault(frame, FALSE, eva);
712 * The kernel may be using npx for copying or other
715 panic("kernel NPX should not happen");
721 case T_PROTFLT: /* general protection fault */
722 case T_SEGNPFLT: /* segment not present fault */
724 * Invalid segment selectors and out of bounds
725 * %eip's and %esp's can be set up in user mode.
726 * This causes a fault in kernel mode when the
727 * kernel tries to return to user mode. We want
728 * to get this fault so that we can fix the
729 * problem here and not have to check all the
730 * selectors and pointers when the user changes
733 if (mycpu->gd_intr_nesting_level == 0) {
734 if (td->td_pcb->pcb_onfault) {
736 (register_t)td->td_pcb->pcb_onfault;
744 * PSL_NT can be set in user mode and isn't cleared
745 * automatically when the kernel is entered. This
746 * causes a TSS fault when the kernel attempts to
747 * `iret' because the TSS link is uninitialized. We
748 * want to get this fault so that we can fix the
749 * problem here and not every time the kernel is
752 if (frame->tf_eflags & PSL_NT) {
753 frame->tf_eflags &= ~PSL_NT;
758 case T_TRCTRAP: /* trace trap */
760 if (frame->tf_eip == (int)IDTVEC(syscall)) {
762 * We've just entered system mode via the
763 * syscall lcall. Continue single stepping
764 * silently until the syscall handler has
769 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
771 * The syscall handler has now saved the
772 * flags. Stop single stepping it.
774 frame->tf_eflags &= ~PSL_T;
780 * Ignore debug register trace traps due to
781 * accesses in the user's address space, which
782 * can happen under several conditions such as
783 * if a user sets a watchpoint on a buffer and
784 * then passes that buffer to a system call.
785 * We still want to get TRCTRAPS for addresses
786 * in kernel space because that is useful when
787 * debugging the kernel.
789 if (user_dbreg_trap()) {
791 * Reset breakpoint bits because the
794 load_dr6(rdr6() & 0xfffffff0);
799 * Fall through (TRCTRAP kernel mode, kernel address)
803 * If DDB is enabled, let it handle the debugger trap.
804 * Otherwise, debugger traps "can't happen".
807 MAKEMPSAFE(have_mplock);
808 if (kdb_trap (type, 0, frame))
813 MAKEMPSAFE(have_mplock);
814 trap_fatal(frame, FALSE, eva);
817 MAKEMPSAFE(have_mplock);
818 trap_fatal(frame, FALSE, eva);
822 * Ignore this trap generated from a spurious SIGTRAP.
824 * single stepping in / syscalls leads to spurious / SIGTRAP
827 * Haiku (c) 2007 Simon 'corecode' Schubert
833 * Translate fault for emulators (e.g. Linux)
835 if (*p->p_sysent->sv_transtrap)
836 i = (*p->p_sysent->sv_transtrap)(i, type);
838 MAKEMPSAFE(have_mplock);
839 trapsignal(lp, i, ucode);
842 if (type <= MAX_TRAP_MSG) {
843 uprintf("fatal process exception: %s",
845 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
846 uprintf(", fault VA = 0x%lx", (u_long)eva);
858 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
859 ("syscall: critical section count mismatch! %d/%d",
860 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
865 trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
868 struct vmspace *vm = NULL;
872 thread_t td = curthread;
873 struct lwp *lp = td->td_lwp;
875 va = trunc_page(eva);
876 if (usermode == FALSE) {
878 * This is a fault on kernel virtual memory.
883 * This is a fault on non-kernel virtual memory.
884 * vm is initialized above to NULL. If curproc is NULL
885 * or curproc->p_vmspace is NULL the fault is fatal.
888 vm = lp->lwp_vmspace;
896 if (frame->tf_xflags & PGEX_W)
897 ftype = VM_PROT_READ | VM_PROT_WRITE;
899 ftype = VM_PROT_READ;
901 if (map != &kernel_map) {
903 * Keep swapout from messing with us during this
909 * Grow the stack if necessary
911 /* grow_stack returns false only if va falls into
912 * a growable stack region and the stack growth
913 * fails. It returns true if va was not within
914 * a growable stack region, or if the stack
917 if (!grow_stack (lp->lwp_proc, va)) {
923 /* Fault in the user page: */
924 rv = vm_fault(map, va, ftype,
925 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
931 * Don't have to worry about process locking or stacks in the kernel.
933 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
936 if (rv == KERN_SUCCESS)
940 if (td->td_gd->gd_intr_nesting_level == 0 &&
941 td->td_pcb->pcb_onfault) {
942 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
945 trap_fatal(frame, usermode, eva);
948 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
952 trap_fatal(struct trapframe *frame, int usermode, vm_offset_t eva)
954 int code, type, ss, esp;
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("mp_lock = %08x; ", mp_lock);
967 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
969 if (type == T_PAGEFLT) {
970 kprintf("fault virtual address = 0x%x\n", eva);
971 kprintf("fault code = %s %s, %s\n",
972 usermode ? "user" : "supervisor",
973 code & PGEX_W ? "write" : "read",
974 code & PGEX_P ? "protection violation" : "page not present");
976 kprintf("instruction pointer = 0x%x:0x%x\n",
977 frame->tf_cs & 0xffff, frame->tf_eip);
979 ss = frame->tf_ss & 0xffff;
982 ss = GSEL(GDATA_SEL, SEL_KPL);
983 esp = (int)&frame->tf_esp;
985 kprintf("stack pointer = 0x%x:0x%x\n", ss, esp);
986 kprintf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
987 kprintf("processor eflags = ");
988 if (frame->tf_eflags & PSL_T)
989 kprintf("trace trap, ");
990 if (frame->tf_eflags & PSL_I)
991 kprintf("interrupt enabled, ");
992 if (frame->tf_eflags & PSL_NT)
993 kprintf("nested task, ");
994 if (frame->tf_eflags & PSL_RF)
997 if (frame->tf_eflags & PSL_VM)
1000 kprintf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
1001 kprintf("current process = ");
1003 kprintf("%lu (%s)\n",
1004 (u_long)curproc->p_pid, curproc->p_comm ?
1005 curproc->p_comm : "");
1009 kprintf("current thread = pri %d ", curthread->td_pri);
1010 if (curthread->td_pri >= TDPRI_CRIT)
1016 * we probably SHOULD have stopped the other CPUs before now!
1017 * another CPU COULD have been touching cpl at this moment...
1019 kprintf(" <- SMP: XXX");
1028 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1031 kprintf("trap number = %d\n", type);
1032 if (type <= MAX_TRAP_MSG)
1033 panic("%s", trap_msg[type]);
1035 panic("unknown/reserved trap");
1039 * Double fault handler. Called when a fault occurs while writing
1040 * a frame for a trap/exception onto the stack. This usually occurs
1041 * when the stack overflows (such is the case with infinite recursion,
1044 * XXX Note that the current PTD gets replaced by IdlePTD when the
1045 * task switch occurs. This means that the stack that was active at
1046 * the time of the double fault is not available at <kstack> unless
1047 * the machine was idle when the double fault occurred. The downside
1048 * of this is that "trace <ebp>" in ddb won't work.
1051 dblfault_handler(void)
1053 struct mdglobaldata *gd = mdcpu;
1055 kprintf("\nFatal double fault:\n");
1056 kprintf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1057 kprintf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1058 kprintf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
1060 /* two separate prints in case of a trap on an unmapped page */
1061 kprintf("mp_lock = %08x; ", mp_lock);
1062 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
1064 panic("double fault");
1068 * Compensate for 386 brain damage (missing URKR).
1069 * This is a little simpler than the pagefault handler in trap() because
1070 * it the page tables have already been faulted in and high addresses
1071 * are thrown out early for other reasons.
1074 trapwrite(unsigned addr)
1081 va = trunc_page((vm_offset_t)addr);
1083 * XXX - MAX is END. Changed > to >= for temp. fix.
1085 if (va >= VM_MAX_USER_ADDRESS)
1088 lp = curthread->td_lwp;
1089 vm = lp->lwp_vmspace;
1091 PHOLD(lp->lwp_proc);
1093 if (!grow_stack (lp->lwp_proc, va)) {
1094 PRELE(lp->lwp_proc);
1099 * fault the data page
1101 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY);
1103 PRELE(lp->lwp_proc);
1105 if (rv != KERN_SUCCESS)
1112 * syscall2 - MP aware system call request C handler
1114 * A system call is essentially treated as a trap except that the
1115 * MP lock is not held on entry or return. We are responsible for
1116 * obtaining the MP lock if necessary and for handling ASTs
1117 * (e.g. a task switch) prior to return.
1119 * In general, only simple access and manipulation of curproc and
1120 * the current stack is allowed without having to hold MP lock.
1122 * MPSAFE - note that large sections of this routine are run without
1127 syscall2(struct trapframe *frame)
1129 struct thread *td = curthread;
1130 struct proc *p = td->td_proc;
1131 struct lwp *lp = td->td_lwp;
1133 struct sysent *callp;
1134 register_t orig_tf_eflags;
1139 int crit_count = td->td_pri & ~TDPRI_MASK;
1142 int have_mplock = 0;
1145 union sysunion args;
1148 KASSERT(td->td_mpcount == 0, ("badmpcount syscall2 from %p", (void *)frame->tf_eip));
1149 if (syscall_mpsafe == 0)
1150 MAKEMPSAFE(have_mplock);
1152 userenter(td); /* lazy raise our priority */
1157 sticks = (int)td->td_sticks;
1158 orig_tf_eflags = frame->tf_eflags;
1161 * Virtual kernel intercept - if a VM context managed by a virtual
1162 * kernel issues a system call the virtual kernel handles it, not us.
1163 * Restore the virtual kernel context and return from its system
1164 * call. The current frame is copied out to the virtual kernel.
1166 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1167 error = vkernel_trap(lp, frame);
1168 frame->tf_eax = error;
1170 frame->tf_eflags |= PSL_C;
1171 error = EJUSTRETURN;
1176 * Get the system call parameters and account for time
1178 lp->lwp_md.md_regs = frame;
1179 params = (caddr_t)frame->tf_esp + sizeof(int);
1180 code = frame->tf_eax;
1182 if (p->p_sysent->sv_prepsyscall) {
1183 (*p->p_sysent->sv_prepsyscall)(
1184 frame, (int *)(&args.nosys.sysmsg + 1),
1188 * Need to check if this is a 32 bit or 64 bit syscall.
1189 * fuword is MP aware.
1191 if (code == SYS_syscall) {
1193 * Code is first argument, followed by actual args.
1195 code = fuword(params);
1196 params += sizeof(int);
1197 } else if (code == SYS___syscall) {
1199 * Like syscall, but code is a quad, so as to maintain
1200 * quad alignment for the rest of the arguments.
1202 code = fuword(params);
1203 params += sizeof(quad_t);
1207 code &= p->p_sysent->sv_mask;
1208 if (code >= p->p_sysent->sv_size)
1209 callp = &p->p_sysent->sv_table[0];
1211 callp = &p->p_sysent->sv_table[code];
1213 narg = callp->sy_narg & SYF_ARGMASK;
1216 * copyin is MP aware, but the tracing code is not
1218 if (narg && params) {
1219 error = copyin(params, (caddr_t)(&args.nosys.sysmsg + 1),
1220 narg * sizeof(register_t));
1223 if (KTRPOINT(td, KTR_SYSCALL)) {
1224 MAKEMPSAFE(have_mplock);
1226 ktrsyscall(lp, code, narg,
1227 (void *)(&args.nosys.sysmsg + 1));
1235 if (KTRPOINT(td, KTR_SYSCALL)) {
1236 MAKEMPSAFE(have_mplock);
1237 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1242 * For traditional syscall code edx is left untouched when 32 bit
1243 * results are returned. Since edx is loaded from fds[1] when the
1244 * system call returns we pre-set it here.
1246 args.sysmsg_fds[0] = 0;
1247 args.sysmsg_fds[1] = frame->tf_edx;
1250 * The syscall might manipulate the trap frame. If it does it
1251 * will probably return EJUSTRETURN.
1253 args.sysmsg_frame = frame;
1255 STOPEVENT(p, S_SCE, narg); /* MP aware */
1259 * Try to run the syscall without the MP lock if the syscall
1260 * is MP safe. We have to obtain the MP lock no matter what if
1263 if ((callp->sy_narg & SYF_MPSAFE) == 0)
1264 MAKEMPSAFE(have_mplock);
1267 error = (*callp->sy_call)(&args);
1270 kprintf("system call %d returned %d\n", code, error);
1275 * MP SAFE (we may or may not have the MP lock at this point)
1280 * Reinitialize proc pointer `p' as it may be different
1281 * if this is a child returning from fork syscall.
1284 lp = curthread->td_lwp;
1285 frame->tf_eax = args.sysmsg_fds[0];
1286 frame->tf_edx = args.sysmsg_fds[1];
1287 frame->tf_eflags &= ~PSL_C;
1291 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1292 * int 0x80 is 2 bytes. We saved this in tf_err.
1294 frame->tf_eip -= frame->tf_err;
1299 panic("Unexpected EASYNC return value (for now)");
1302 if (p->p_sysent->sv_errsize) {
1303 if (error >= p->p_sysent->sv_errsize)
1304 error = -1; /* XXX */
1306 error = p->p_sysent->sv_errtbl[error];
1308 frame->tf_eax = error;
1309 frame->tf_eflags |= PSL_C;
1314 * Traced syscall. trapsignal() is not MP aware.
1316 if ((orig_tf_eflags & PSL_T) /*&& !(orig_tf_eflags & PSL_VM)*/) {
1317 MAKEMPSAFE(have_mplock);
1318 frame->tf_eflags &= ~PSL_T;
1319 trapsignal(lp, SIGTRAP, 0);
1323 * Handle reschedule and other end-of-syscall issues
1325 userret(lp, frame, sticks);
1328 if (KTRPOINT(td, KTR_SYSRET)) {
1329 MAKEMPSAFE(have_mplock);
1330 ktrsysret(lp, code, error, args.sysmsg_result);
1335 * This works because errno is findable through the
1336 * register set. If we ever support an emulation where this
1337 * is not the case, this code will need to be revisited.
1339 STOPEVENT(p, S_SCX, code);
1344 * Release the MP lock if we had to get it
1346 KASSERT(td->td_mpcount == have_mplock,
1347 ("badmpcount syscall2/end from %p", (void *)frame->tf_eip));
1352 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
1353 ("syscall: critical section count mismatch! %d/%d",
1354 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
1359 fork_return(struct lwp *lp, struct trapframe *frame)
1361 frame->tf_eax = 0; /* Child returns zero */
1362 frame->tf_eflags &= ~PSL_C; /* success */
1365 generic_lwp_return(lp, frame);
1369 * Simplified back end of syscall(), used when returning from fork()
1370 * or lwp_create() directly into user mode. MP lock is held on entry and
1371 * should be released on return. This code will return back into the fork
1372 * trampoline code which then runs doreti.
1375 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1377 struct proc *p = lp->lwp_proc;
1380 * Newly forked processes are given a kernel priority. We have to
1381 * adjust the priority to a normal user priority and fake entry
1382 * into the kernel (call userenter()) to install a passive release
1383 * function just in case userret() decides to stop the process. This
1384 * can occur when ^Z races a fork. If we do not install the passive
1385 * release function the current process designation will not be
1386 * released when the thread goes to sleep.
1388 lwkt_setpri_self(TDPRI_USER_NORM);
1389 userenter(lp->lwp_thread);
1390 userret(lp, frame, 0);
1392 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1393 ktrsysret(lp, SYS_fork, 0, 0);
1395 p->p_flag |= P_PASSIVE_ACQ;
1397 p->p_flag &= ~P_PASSIVE_ACQ;
1399 KKASSERT(lp->lwp_thread->td_mpcount == 1);
1405 * doreti has turned into this. The frame is directly on the stack. We
1406 * pull everything else we need (fpu and tls context) from the current
1409 * Note on fpu interactions: In a virtual kernel, the fpu context for
1410 * an emulated user mode process is not shared with the virtual kernel's
1411 * fpu context, so we only have to 'stack' fpu contexts within the virtual
1412 * kernel itself, and not even then since the signal() contexts that we care
1413 * about save and restore the FPU state (I think anyhow).
1415 * vmspace_ctl() returns an error only if it had problems instaling the
1416 * context we supplied or problems copying data to/from our VM space.
1419 go_user(struct intrframe *frame)
1421 struct trapframe *tf = (void *)&frame->if_gs;
1425 * Interrupts may be disabled on entry, make sure all signals
1426 * can be received before beginning our loop.
1431 * Switch to the current simulated user process, then call
1432 * user_trap() when we break out of it (usually due to a signal).
1436 * Tell the real kernel whether it is ok to use the FP
1439 if (mdcpu->gd_npxthread == curthread) {
1440 tf->tf_xflags &= ~PGEX_FPFAULT;
1442 tf->tf_xflags |= PGEX_FPFAULT;
1446 * Run emulated user process context. This call interlocks
1447 * with new mailbox signals.
1449 * Set PGEX_U unconditionally, indicating a user frame (the
1450 * bit is normally set only by T_PAGEFLT).
1452 r = vmspace_ctl(&curproc->p_vmspace->vm_pmap, VMSPACE_CTL_RUN,
1453 tf, &curthread->td_savevext);
1454 frame->if_xflags |= PGEX_U;
1456 kprintf("GO USER %d trap %d EVA %08x EIP %08x ESP %08x XFLAGS %02x/%02x\n",
1457 r, tf->tf_trapno, tf->tf_err, tf->tf_eip, tf->tf_esp,
1458 tf->tf_xflags, frame->if_xflags);
1462 panic("vmspace_ctl failed");
1464 if (tf->tf_trapno) {
1466 } else if (mycpu->gd_reqflags & RQF_AST_MASK) {
1467 tf->tf_trapno = T_ASTFLT;
1476 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1477 * fault (which is then passed back to the virtual kernel) if an attempt is
1478 * made to use the FP unit.
1480 * XXX this is a fairly big hack.
1483 set_vkernel_fp(struct trapframe *frame)
1485 struct thread *td = curthread;
1487 if (frame->tf_xflags & PGEX_FPFAULT) {
1488 td->td_pcb->pcb_flags |= FP_VIRTFP;
1489 if (mdcpu->gd_npxthread == td)
1492 td->td_pcb->pcb_flags &= ~FP_VIRTFP;