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/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>
95 #include <sys/msgport2.h>
96 #include <sys/thread2.h>
97 #include <sys/mplock2.h>
101 #define MAKEMPSAFE(have_mplock) \
102 if (have_mplock == 0) { \
109 #define MAKEMPSAFE(have_mplock)
113 int (*pmath_emulate) (struct trapframe *);
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 MALLOC_DEFINE(M_SYSMSG, "sysmsg", "sysmsg structure");
172 extern int max_sysmsg;
175 * Passively intercepts the thread switch function to increase the thread
176 * priority from a user priority to a kernel priority, reducing
177 * syscall and trap overhead for the case where no switch occurs.
179 * Synchronizes td_ucred with p_ucred. This is used by system calls,
180 * signal handling, faults, AST traps, and anything else that enters the
181 * kernel from userland and provides the kernel with a stable read-only
182 * copy of the process ucred.
185 userenter(struct thread *curtd, struct proc *curp)
190 curtd->td_release = lwkt_passive_release;
192 if (curtd->td_ucred != curp->p_ucred) {
193 ncred = crhold(curp->p_ucred);
194 ocred = curtd->td_ucred;
195 curtd->td_ucred = ncred;
203 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
204 * must be completed before we can return to or try to return to userland.
206 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
207 * arithmatic on the delta calculation so the absolute tick values are
208 * truncated to an integer.
211 userret(struct lwp *lp, struct trapframe *frame, int sticks)
213 struct proc *p = lp->lwp_proc;
217 * Charge system time if profiling. Note: times are in microseconds.
218 * This may do a copyout and block, so do it first even though it
219 * means some system time will be charged as user time.
221 if (p->p_flags & P_PROFIL) {
222 addupc_task(p, frame->tf_eip,
223 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
228 * If the jungle wants us dead, so be it.
230 if (lp->lwp_mpflags & LWP_MP_WEXIT) {
231 lwkt_gettoken(&p->p_token);
233 lwkt_reltoken(&p->p_token); /* NOT REACHED */
237 * Block here if we are in a stopped state.
239 if (p->p_stat == SSTOP) {
240 lwkt_gettoken(&p->p_token);
242 lwkt_reltoken(&p->p_token);
247 * Post any pending upcalls. If running a virtual kernel be sure
248 * to restore the virtual kernel's vmspace before posting the upcall.
250 if (p->p_flags & (P_SIGVTALRM | P_SIGPROF | P_UPCALLPEND)) {
251 lwkt_gettoken(&p->p_token);
252 if (p->p_flags & P_SIGVTALRM) {
253 p->p_flags &= ~P_SIGVTALRM;
254 ksignal(p, SIGVTALRM);
256 if (p->p_flags & P_SIGPROF) {
257 p->p_flags &= ~P_SIGPROF;
260 if (p->p_flags & P_UPCALLPEND) {
261 p->p_flags &= ~P_UPCALLPEND;
264 lwkt_reltoken(&p->p_token);
269 * Post any pending signals
271 * WARNING! postsig() can exit and not return.
273 if ((sig = CURSIG_TRACE(lp)) != 0) {
274 lwkt_gettoken(&p->p_token);
276 lwkt_reltoken(&p->p_token);
281 * block here if we are swapped out, but still process signals
282 * (such as SIGKILL). proc0 (the swapin scheduler) is already
283 * aware of our situation, we do not have to wake it up.
285 if (p->p_flags & P_SWAPPEDOUT) {
286 lwkt_gettoken(&p->p_token);
288 p->p_flags |= P_SWAPWAIT;
290 if (p->p_flags & P_SWAPWAIT)
291 tsleep(p, PCATCH, "SWOUT", 0);
292 p->p_flags &= ~P_SWAPWAIT;
294 lwkt_reltoken(&p->p_token);
299 * Make sure postsig() handled request to restore old signal mask after
300 * running signal handler.
302 KKASSERT((lp->lwp_flags & LWP_OLDMASK) == 0);
306 * Cleanup from userenter and any passive release that might have occured.
307 * We must reclaim the current-process designation before we can return
308 * to usermode. We also handle both LWKT and USER reschedule requests.
311 userexit(struct lwp *lp)
313 struct thread *td = lp->lwp_thread;
314 /* globaldata_t gd = td->td_gd; */
317 * Handle stop requests at kernel priority. Any requests queued
318 * after this loop will generate another AST.
320 while (lp->lwp_proc->p_stat == SSTOP) {
321 lwkt_gettoken(&lp->lwp_proc->p_token);
323 lwkt_reltoken(&lp->lwp_proc->p_token);
327 * Reduce our priority in preparation for a return to userland. If
328 * our passive release function was still in place, our priority was
329 * never raised and does not need to be reduced.
331 lwkt_passive_recover(td);
334 * Become the current user scheduled process if we aren't already,
335 * and deal with reschedule requests and other factors.
337 lp->lwp_proc->p_usched->acquire_curproc(lp);
338 /* WARNING: we may have migrated cpu's */
339 /* gd = td->td_gd; */
342 #if !defined(KTR_KERNENTRY)
343 #define KTR_KERNENTRY KTR_ALL
345 KTR_INFO_MASTER(kernentry);
346 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0,
347 "TRAP(pid %d, tid %d, trapno %d, eva %lu)",
348 pid_t pid, lwpid_t tid, register_t trapno, vm_offset_t eva);
349 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "TRAP_RET(pid %d, tid %d)",
350 pid_t pid, lwpid_t tid);
351 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "SYSC(pid %d, tid %d, nr %d)",
352 pid_t pid, lwpid_t tid, register_t trapno);
353 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "SYSRET(pid %d, tid %d, err %d)",
354 pid_t pid, lwpid_t tid, int err);
355 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "FORKRET(pid %d, tid %d)",
356 pid_t pid, lwpid_t tid);
359 * Exception, fault, and trap interface to the kernel.
360 * This common code is called from assembly language IDT gate entry
361 * routines that prepare a suitable stack frame, and restore this
362 * frame after the exception has been processed.
364 * This function is also called from doreti in an interlock to handle ASTs.
365 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
367 * NOTE! We have to retrieve the fault address prior to obtaining the
368 * MP lock because get_mplock() may switch out. YYY cr2 really ought
369 * to be retrieved by the assembly code, not here.
371 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
372 * if an attempt is made to switch from a fast interrupt or IPI. This is
373 * necessary to properly take fatal kernel traps on SMP machines if
374 * get_mplock() has to block.
378 user_trap(struct trapframe *frame)
380 struct globaldata *gd = mycpu;
381 struct thread *td = gd->gd_curthread;
382 struct lwp *lp = td->td_lwp;
385 int i = 0, ucode = 0, type, code;
390 int crit_count = td->td_critcount;
391 lwkt_tokref_t curstop = td->td_toks_stop;
398 * This is a bad kludge to avoid changing the various trapframe
399 * structures. Because we are enabled as a virtual kernel,
400 * the original tf_err field will be passed to us shifted 16
401 * over in the tf_trapno field for T_PAGEFLT.
403 if (frame->tf_trapno == T_PAGEFLT)
408 kprintf("USER_TRAP AT %08x xflags %d trapno %d eva %08x\n",
409 frame->tf_eip, frame->tf_xflags, frame->tf_trapno, eva);
413 * Everything coming from user mode runs through user_trap,
414 * including system calls.
416 if (frame->tf_trapno == T_SYSCALL80) {
421 KTR_LOG(kernentry_trap, lp->lwp_proc->p_pid, lp->lwp_tid,
422 frame->tf_trapno, eva);
426 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
427 ++gd->gd_trap_nesting_level;
428 MAKEMPSAFE(have_mplock);
429 trap_fatal(frame, TRUE, eva);
430 --gd->gd_trap_nesting_level;
435 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
438 type = frame->tf_trapno;
439 code = frame->tf_err;
443 sticks = (int)td->td_sticks;
444 lp->lwp_md.md_regs = frame;
447 case T_PRIVINFLT: /* privileged instruction fault */
452 case T_BPTFLT: /* bpt instruction fault */
453 case T_TRCTRAP: /* trace trap */
454 frame->tf_eflags &= ~PSL_T;
456 ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
459 case T_ARITHTRAP: /* arithmetic trap */
464 case T_ASTFLT: /* Allow process switch */
465 mycpu->gd_cnt.v_soft++;
466 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
467 atomic_clear_int(&mycpu->gd_reqflags,
469 addupc_task(p, p->p_prof.pr_addr,
475 * The following two traps can happen in
476 * vm86 mode, and, if so, we want to handle
479 case T_PROTFLT: /* general protection fault */
480 case T_STKFLT: /* stack fault */
482 if (frame->tf_eflags & PSL_VM) {
483 i = vm86_emulate((struct vm86frame *)frame);
490 ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR;
492 case T_SEGNPFLT: /* segment not present fault */
496 case T_TSSFLT: /* invalid TSS fault */
497 case T_DOUBLEFLT: /* double fault */
503 case T_PAGEFLT: /* page fault */
504 MAKEMPSAFE(have_mplock);
505 i = trap_pfault(frame, TRUE, eva);
508 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
523 case T_DIVIDE: /* integer divide fault */
530 MAKEMPSAFE(have_mplock);
531 /* machine/parity/power fail/"kitchen sink" faults */
532 if (isa_nmi(code) == 0) {
535 * NMI can be hooked up to a pushbutton
539 kprintf ("NMI ... going to debugger\n");
540 kdb_trap (type, 0, frame);
544 } else if (panic_on_nmi)
545 panic("NMI indicates hardware failure");
547 #endif /* NISA > 0 */
549 case T_OFLOW: /* integer overflow fault */
554 case T_BOUND: /* bounds check fault */
561 * Virtual kernel intercept - pass the DNA exception
562 * to the (emulated) virtual kernel if it asked to handle
563 * it. This occurs when the virtual kernel is holding
564 * onto the FP context for a different emulated
565 * process then the one currently running.
567 * We must still call npxdna() since we may have
568 * saved FP state that the (emulated) virtual kernel
569 * needs to hand over to a different emulated process.
571 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
572 (td->td_pcb->pcb_flags & FP_VIRTFP)
580 * The kernel may have switched out the FP unit's
581 * state, causing the user process to take a fault
582 * when it tries to use the FP unit. Restore the
588 if (!pmath_emulate) {
590 ucode = FPE_FPU_NP_TRAP;
593 i = (*pmath_emulate)(frame);
595 if (!(frame->tf_eflags & PSL_T))
597 frame->tf_eflags &= ~PSL_T;
600 /* else ucode = emulator_only_knows() XXX */
603 case T_FPOPFLT: /* FPU operand fetch fault */
608 case T_XMMFLT: /* SIMD floating-point exception */
615 * Virtual kernel intercept - if the fault is directly related to a
616 * VM context managed by a virtual kernel then let the virtual kernel
619 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
620 vkernel_trap(lp, frame);
625 * Translate fault for emulators (e.g. Linux)
627 if (*p->p_sysent->sv_transtrap)
628 i = (*p->p_sysent->sv_transtrap)(i, type);
630 MAKEMPSAFE(have_mplock);
631 trapsignal(lp, i, ucode);
634 if (type <= MAX_TRAP_MSG) {
635 uprintf("fatal process exception: %s",
637 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
638 uprintf(", fault VA = 0x%lx", (u_long)eva);
644 userret(lp, frame, sticks);
651 KTR_LOG(kernentry_trap_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
653 KASSERT(crit_count == td->td_critcount,
654 ("trap: critical section count mismatch! %d/%d",
655 crit_count, td->td_pri));
656 KASSERT(curstop == td->td_toks_stop,
657 ("trap: extra tokens held after trap! %zd/%zd",
658 curstop - &td->td_toks_base,
659 td->td_toks_stop - &td->td_toks_base));
664 kern_trap(struct trapframe *frame)
666 struct globaldata *gd = mycpu;
667 struct thread *td = gd->gd_curthread;
670 int i = 0, ucode = 0, type, code;
675 int crit_count = td->td_critcount;
676 lwkt_tokref_t curstop = td->td_toks_stop;
683 if (frame->tf_trapno == T_PAGEFLT)
690 ++gd->gd_trap_nesting_level;
691 MAKEMPSAFE(have_mplock);
692 trap_fatal(frame, FALSE, eva);
693 --gd->gd_trap_nesting_level;
697 type = frame->tf_trapno;
698 code = frame->tf_err;
706 case T_PAGEFLT: /* page fault */
707 MAKEMPSAFE(have_mplock);
708 trap_pfault(frame, FALSE, eva);
714 * The kernel may be using npx for copying or other
717 panic("kernel NPX should not happen");
723 case T_PROTFLT: /* general protection fault */
724 case T_SEGNPFLT: /* segment not present fault */
726 * Invalid segment selectors and out of bounds
727 * %eip's and %esp's can be set up in user mode.
728 * This causes a fault in kernel mode when the
729 * kernel tries to return to user mode. We want
730 * to get this fault so that we can fix the
731 * problem here and not have to check all the
732 * selectors and pointers when the user changes
735 if (mycpu->gd_intr_nesting_level == 0) {
736 if (td->td_pcb->pcb_onfault) {
738 (register_t)td->td_pcb->pcb_onfault;
746 * PSL_NT can be set in user mode and isn't cleared
747 * automatically when the kernel is entered. This
748 * causes a TSS fault when the kernel attempts to
749 * `iret' because the TSS link is uninitialized. We
750 * want to get this fault so that we can fix the
751 * problem here and not every time the kernel is
754 if (frame->tf_eflags & PSL_NT) {
755 frame->tf_eflags &= ~PSL_NT;
760 case T_TRCTRAP: /* trace trap */
762 if (frame->tf_eip == (int)IDTVEC(syscall)) {
764 * We've just entered system mode via the
765 * syscall lcall. Continue single stepping
766 * silently until the syscall handler has
771 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
773 * The syscall handler has now saved the
774 * flags. Stop single stepping it.
776 frame->tf_eflags &= ~PSL_T;
782 * Ignore debug register trace traps due to
783 * accesses in the user's address space, which
784 * can happen under several conditions such as
785 * if a user sets a watchpoint on a buffer and
786 * then passes that buffer to a system call.
787 * We still want to get TRCTRAPS for addresses
788 * in kernel space because that is useful when
789 * debugging the kernel.
791 if (user_dbreg_trap()) {
793 * Reset breakpoint bits because the
796 load_dr6(rdr6() & 0xfffffff0);
801 * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
805 * If DDB is enabled, let it handle the debugger trap.
806 * Otherwise, debugger traps "can't happen".
809 MAKEMPSAFE(have_mplock);
810 if (kdb_trap (type, 0, frame))
815 MAKEMPSAFE(have_mplock);
816 trap_fatal(frame, FALSE, eva);
819 MAKEMPSAFE(have_mplock);
820 trap_fatal(frame, FALSE, eva);
824 * Ignore this trap generated from a spurious SIGTRAP.
826 * single stepping in / syscalls leads to spurious / SIGTRAP
829 * Haiku (c) 2007 Simon 'corecode' Schubert
835 * Translate fault for emulators (e.g. Linux)
837 if (*p->p_sysent->sv_transtrap)
838 i = (*p->p_sysent->sv_transtrap)(i, type);
840 MAKEMPSAFE(have_mplock);
841 trapsignal(lp, i, ucode);
844 if (type <= MAX_TRAP_MSG) {
845 uprintf("fatal process exception: %s",
847 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
848 uprintf(", fault VA = 0x%lx", (u_long)eva);
860 KASSERT(crit_count == td->td_critcount,
861 ("trap: critical section count mismatch! %d/%d",
862 crit_count, td->td_pri));
863 KASSERT(curstop == td->td_toks_stop,
864 ("trap: extra tokens held after trap! %zd/%zd",
865 curstop - &td->td_toks_base,
866 td->td_toks_stop - &td->td_toks_base));
871 trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
874 struct vmspace *vm = NULL;
879 thread_t td = curthread;
880 struct lwp *lp = td->td_lwp;
882 va = trunc_page(eva);
883 if (usermode == FALSE) {
885 * This is a fault on kernel virtual memory.
890 * This is a fault on non-kernel virtual memory.
891 * vm is initialized above to NULL. If curproc is NULL
892 * or curproc->p_vmspace is NULL the fault is fatal.
895 vm = lp->lwp_vmspace;
903 if (frame->tf_xflags & PGEX_W)
904 ftype = VM_PROT_READ | VM_PROT_WRITE;
906 ftype = VM_PROT_READ;
908 if (map != &kernel_map) {
910 * Keep swapout from messing with us during this
920 fault_flags |= VM_FAULT_BURST;
921 if (ftype & VM_PROT_WRITE)
922 fault_flags |= VM_FAULT_DIRTY;
924 fault_flags |= VM_FAULT_NORMAL;
925 rv = vm_fault(map, va, ftype, fault_flags);
929 * Don't have to worry about process locking or stacks in the kernel.
931 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("cpuid = %d\n", mycpu->gd_cpuid);
965 if (type == T_PAGEFLT) {
966 kprintf("fault virtual address = %p\n", (void *)eva);
967 kprintf("fault code = %s %s, %s\n",
968 usermode ? "user" : "supervisor",
969 code & PGEX_W ? "write" : "read",
970 code & PGEX_P ? "protection violation" : "page not present");
972 kprintf("instruction pointer = 0x%x:0x%x\n",
973 frame->tf_cs & 0xffff, frame->tf_eip);
975 ss = frame->tf_ss & 0xffff;
978 ss = GSEL(GDATA_SEL, SEL_KPL);
979 esp = (int)&frame->tf_esp;
981 kprintf("stack pointer = 0x%x:0x%x\n", ss, esp);
982 kprintf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
983 kprintf("processor eflags = ");
984 if (frame->tf_eflags & PSL_T)
985 kprintf("trace trap, ");
986 if (frame->tf_eflags & PSL_I)
987 kprintf("interrupt enabled, ");
988 if (frame->tf_eflags & PSL_NT)
989 kprintf("nested task, ");
990 if (frame->tf_eflags & PSL_RF)
993 if (frame->tf_eflags & PSL_VM)
996 kprintf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
997 kprintf("current process = ");
999 kprintf("%lu (%s)\n",
1000 (u_long)curproc->p_pid, curproc->p_comm ?
1001 curproc->p_comm : "");
1005 kprintf("current thread = pri %d ", curthread->td_pri);
1006 if (curthread->td_critcount)
1012 * we probably SHOULD have stopped the other CPUs before now!
1013 * another CPU COULD have been touching cpl at this moment...
1015 kprintf(" <- SMP: XXX");
1024 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1027 kprintf("trap number = %d\n", type);
1028 if (type <= MAX_TRAP_MSG)
1029 panic("%s", trap_msg[type]);
1031 panic("unknown/reserved trap");
1035 * Double fault handler. Called when a fault occurs while writing
1036 * a frame for a trap/exception onto the stack. This usually occurs
1037 * when the stack overflows (such is the case with infinite recursion,
1040 * XXX Note that the current PTD gets replaced by IdlePTD when the
1041 * task switch occurs. This means that the stack that was active at
1042 * the time of the double fault is not available at <kstack> unless
1043 * the machine was idle when the double fault occurred. The downside
1044 * of this is that "trace <ebp>" in ddb won't work.
1047 dblfault_handler(void)
1049 struct mdglobaldata *gd = mdcpu;
1051 kprintf("\nFatal double fault:\n");
1052 kprintf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1053 kprintf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1054 kprintf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
1056 /* two separate prints in case of a trap on an unmapped page */
1057 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
1059 panic("double fault");
1063 * syscall2 - MP aware system call request C handler
1065 * A system call is essentially treated as a trap except that the
1066 * MP lock is not held on entry or return. We are responsible for
1067 * obtaining the MP lock if necessary and for handling ASTs
1068 * (e.g. a task switch) prior to return.
1073 syscall2(struct trapframe *frame)
1075 struct thread *td = curthread;
1076 struct proc *p = td->td_proc;
1077 struct lwp *lp = td->td_lwp;
1079 struct sysent *callp;
1080 register_t orig_tf_eflags;
1085 int crit_count = td->td_critcount;
1088 int have_mplock = 0;
1091 union sysunion args;
1093 KTR_LOG(kernentry_syscall, lp->lwp_proc->p_pid, lp->lwp_tid,
1096 userenter(td, p); /* lazy raise our priority */
1101 sticks = (int)td->td_sticks;
1102 orig_tf_eflags = frame->tf_eflags;
1105 * Virtual kernel intercept - if a VM context managed by a virtual
1106 * kernel issues a system call the virtual kernel handles it, not us.
1107 * Restore the virtual kernel context and return from its system
1108 * call. The current frame is copied out to the virtual kernel.
1110 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1111 vkernel_trap(lp, frame);
1112 error = EJUSTRETURN;
1117 * Get the system call parameters and account for time
1119 lp->lwp_md.md_regs = frame;
1120 params = (caddr_t)frame->tf_esp + sizeof(int);
1121 code = frame->tf_eax;
1123 if (p->p_sysent->sv_prepsyscall) {
1124 (*p->p_sysent->sv_prepsyscall)(
1125 frame, (int *)(&args.nosys.sysmsg + 1),
1129 * Need to check if this is a 32 bit or 64 bit syscall.
1130 * fuword is MP aware.
1132 if (code == SYS_syscall) {
1134 * Code is first argument, followed by actual args.
1136 code = fuword(params);
1137 params += sizeof(int);
1138 } else if (code == SYS___syscall) {
1140 * Like syscall, but code is a quad, so as to maintain
1141 * quad alignment for the rest of the arguments.
1143 code = fuword(params);
1144 params += sizeof(quad_t);
1148 code &= p->p_sysent->sv_mask;
1149 if (code >= p->p_sysent->sv_size)
1150 callp = &p->p_sysent->sv_table[0];
1152 callp = &p->p_sysent->sv_table[code];
1154 narg = callp->sy_narg & SYF_ARGMASK;
1157 * copyin is MP aware, but the tracing code is not
1159 if (narg && params) {
1160 error = copyin(params, (caddr_t)(&args.nosys.sysmsg + 1),
1161 narg * sizeof(register_t));
1164 if (KTRPOINT(td, KTR_SYSCALL)) {
1165 MAKEMPSAFE(have_mplock);
1167 ktrsyscall(lp, code, narg,
1168 (void *)(&args.nosys.sysmsg + 1));
1176 if (KTRPOINT(td, KTR_SYSCALL)) {
1177 MAKEMPSAFE(have_mplock);
1178 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1183 * For traditional syscall code edx is left untouched when 32 bit
1184 * results are returned. Since edx is loaded from fds[1] when the
1185 * system call returns we pre-set it here.
1187 args.sysmsg_fds[0] = 0;
1188 args.sysmsg_fds[1] = frame->tf_edx;
1191 * The syscall might manipulate the trap frame. If it does it
1192 * will probably return EJUSTRETURN.
1194 args.sysmsg_frame = frame;
1196 STOPEVENT(p, S_SCE, narg); /* MP aware */
1199 * NOTE: All system calls run MPSAFE now. The system call itself
1200 * is responsible for getting the MP lock.
1202 error = (*callp->sy_call)(&args);
1205 kprintf("system call %d returned %d\n", code, error);
1210 * MP SAFE (we may or may not have the MP lock at this point)
1215 * Reinitialize proc pointer `p' as it may be different
1216 * if this is a child returning from fork syscall.
1219 lp = curthread->td_lwp;
1220 frame->tf_eax = args.sysmsg_fds[0];
1221 frame->tf_edx = args.sysmsg_fds[1];
1222 frame->tf_eflags &= ~PSL_C;
1226 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1227 * int 0x80 is 2 bytes. We saved this in tf_err.
1229 frame->tf_eip -= frame->tf_err;
1234 panic("Unexpected EASYNC return value (for now)");
1237 if (p->p_sysent->sv_errsize) {
1238 if (error >= p->p_sysent->sv_errsize)
1239 error = -1; /* XXX */
1241 error = p->p_sysent->sv_errtbl[error];
1243 frame->tf_eax = error;
1244 frame->tf_eflags |= PSL_C;
1249 * Traced syscall. trapsignal() is not MP aware.
1251 if ((orig_tf_eflags & PSL_T) /*&& !(orig_tf_eflags & PSL_VM)*/) {
1252 MAKEMPSAFE(have_mplock);
1253 frame->tf_eflags &= ~PSL_T;
1254 trapsignal(lp, SIGTRAP, TRAP_TRACE);
1258 * Handle reschedule and other end-of-syscall issues
1260 userret(lp, frame, sticks);
1263 if (KTRPOINT(td, KTR_SYSRET)) {
1264 MAKEMPSAFE(have_mplock);
1265 ktrsysret(lp, code, error, args.sysmsg_result);
1270 * This works because errno is findable through the
1271 * register set. If we ever support an emulation where this
1272 * is not the case, this code will need to be revisited.
1274 STOPEVENT(p, S_SCX, code);
1279 * Release the MP lock if we had to get it
1284 KTR_LOG(kernentry_syscall_ret, lp->lwp_proc->p_pid, lp->lwp_tid, error);
1286 KASSERT(crit_count == td->td_critcount,
1287 ("syscall: critical section count mismatch! %d/%d",
1288 crit_count, td->td_pri));
1289 KASSERT(&td->td_toks_base == td->td_toks_stop,
1290 ("syscall: extra tokens held after trap! %zd",
1291 td->td_toks_stop - &td->td_toks_base));
1296 * NOTE: mplock not held at any point
1299 fork_return(struct lwp *lp, struct trapframe *frame)
1301 frame->tf_eax = 0; /* Child returns zero */
1302 frame->tf_eflags &= ~PSL_C; /* success */
1305 generic_lwp_return(lp, frame);
1306 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1310 * Simplified back end of syscall(), used when returning from fork()
1311 * directly into user mode.
1313 * This code will return back into the fork trampoline code which then
1316 * NOTE: The mplock is not held at any point.
1319 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1321 struct proc *p = lp->lwp_proc;
1324 * Newly forked processes are given a kernel priority. We have to
1325 * adjust the priority to a normal user priority and fake entry
1326 * into the kernel (call userenter()) to install a passive release
1327 * function just in case userret() decides to stop the process. This
1328 * can occur when ^Z races a fork. If we do not install the passive
1329 * release function the current process designation will not be
1330 * released when the thread goes to sleep.
1332 lwkt_setpri_self(TDPRI_USER_NORM);
1333 userenter(lp->lwp_thread, p);
1334 userret(lp, frame, 0);
1336 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1337 ktrsysret(lp, SYS_fork, 0, 0);
1339 lp->lwp_flags |= LWP_PASSIVE_ACQ;
1341 lp->lwp_flags &= ~LWP_PASSIVE_ACQ;
1345 * doreti has turned into this. The frame is directly on the stack. We
1346 * pull everything else we need (fpu and tls context) from the current
1349 * Note on fpu interactions: In a virtual kernel, the fpu context for
1350 * an emulated user mode process is not shared with the virtual kernel's
1351 * fpu context, so we only have to 'stack' fpu contexts within the virtual
1352 * kernel itself, and not even then since the signal() contexts that we care
1353 * about save and restore the FPU state (I think anyhow).
1355 * vmspace_ctl() returns an error only if it had problems instaling the
1356 * context we supplied or problems copying data to/from our VM space.
1359 go_user(struct intrframe *frame)
1361 struct trapframe *tf = (void *)&frame->if_gs;
1365 * Interrupts may be disabled on entry, make sure all signals
1366 * can be received before beginning our loop.
1371 * Switch to the current simulated user process, then call
1372 * user_trap() when we break out of it (usually due to a signal).
1376 * Tell the real kernel whether it is ok to use the FP
1379 * The critical section is required to prevent an interrupt
1380 * from causing a preemptive task switch and changing
1384 if (mdcpu->gd_npxthread == curthread) {
1385 tf->tf_xflags &= ~PGEX_FPFAULT;
1387 tf->tf_xflags |= PGEX_FPFAULT;
1391 * Run emulated user process context. This call interlocks
1392 * with new mailbox signals.
1394 * Set PGEX_U unconditionally, indicating a user frame (the
1395 * bit is normally set only by T_PAGEFLT).
1397 r = vmspace_ctl(&curproc->p_vmspace->vm_pmap, VMSPACE_CTL_RUN,
1398 tf, &curthread->td_savevext);
1400 frame->if_xflags |= PGEX_U;
1402 kprintf("GO USER %d trap %d EVA %08x EIP %08x ESP %08x XFLAGS %02x/%02x\n",
1403 r, tf->tf_trapno, tf->tf_err, tf->tf_eip, tf->tf_esp,
1404 tf->tf_xflags, frame->if_xflags);
1408 panic("vmspace_ctl failed error %d", errno);
1410 if (tf->tf_trapno) {
1414 if (mycpu->gd_reqflags & RQF_AST_MASK) {
1415 tf->tf_trapno = T_ASTFLT;
1423 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1424 * fault (which is then passed back to the virtual kernel) if an attempt is
1425 * made to use the FP unit.
1427 * XXX this is a fairly big hack.
1430 set_vkernel_fp(struct trapframe *frame)
1432 struct thread *td = curthread;
1434 if (frame->tf_xflags & PGEX_FPFAULT) {
1435 td->td_pcb->pcb_flags |= FP_VIRTFP;
1436 if (mdcpu->gd_npxthread == td)
1439 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1444 * Called from vkernel_trap() to fixup the vkernel's syscall
1445 * frame for vmspace_ctl() return.
1448 cpu_vkernel_trap(struct trapframe *frame, int error)
1450 frame->tf_eax = error;
1452 frame->tf_eflags |= PSL_C;
1454 frame->tf_eflags &= ~PSL_C;