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 1.35 2008/09/09 04:06:19 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>
70 #include <sys/upcall.h>
71 #include <sys/vkernel.h>
72 #include <sys/sysproto.h>
73 #include <sys/sysunion.h>
74 #include <sys/vmspace.h>
77 #include <vm/vm_param.h>
80 #include <vm/vm_kern.h>
81 #include <vm/vm_map.h>
82 #include <vm/vm_page.h>
83 #include <vm/vm_extern.h>
85 #include <machine/cpu.h>
86 #include <machine/md_var.h>
87 #include <machine/pcb.h>
88 #include <machine/smp.h>
89 #include <machine/tss.h>
90 #include <machine/globaldata.h>
92 #include <machine/vm86.h>
96 #include <sys/msgport2.h>
97 #include <sys/thread2.h>
98 #include <sys/mplock2.h>
102 #define MAKEMPSAFE(have_mplock) \
103 if (have_mplock == 0) { \
110 #define MAKEMPSAFE(have_mplock)
114 int (*pmath_emulate) (struct trapframe *);
116 static int trap_pfault (struct trapframe *, int, vm_offset_t);
117 static void trap_fatal (struct trapframe *, int, vm_offset_t);
118 void dblfault_handler (void);
121 extern inthand_t IDTVEC(syscall);
124 #define MAX_TRAP_MSG 28
125 static char *trap_msg[] = {
127 "privileged instruction fault", /* 1 T_PRIVINFLT */
129 "breakpoint instruction fault", /* 3 T_BPTFLT */
132 "arithmetic trap", /* 6 T_ARITHTRAP */
133 "system forced exception", /* 7 T_ASTFLT */
135 "general protection fault", /* 9 T_PROTFLT */
136 "trace trap", /* 10 T_TRCTRAP */
138 "page fault", /* 12 T_PAGEFLT */
140 "alignment fault", /* 14 T_ALIGNFLT */
144 "integer divide fault", /* 18 T_DIVIDE */
145 "non-maskable interrupt trap", /* 19 T_NMI */
146 "overflow trap", /* 20 T_OFLOW */
147 "FPU bounds check fault", /* 21 T_BOUND */
148 "FPU device not available", /* 22 T_DNA */
149 "double fault", /* 23 T_DOUBLEFLT */
150 "FPU operand fetch fault", /* 24 T_FPOPFLT */
151 "invalid TSS fault", /* 25 T_TSSFLT */
152 "segment not present fault", /* 26 T_SEGNPFLT */
153 "stack fault", /* 27 T_STKFLT */
154 "machine check trap", /* 28 T_MCHK */
158 static int ddb_on_nmi = 1;
159 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
160 &ddb_on_nmi, 0, "Go to DDB on NMI");
162 static int panic_on_nmi = 1;
163 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
164 &panic_on_nmi, 0, "Panic on NMI");
165 static int fast_release;
166 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
167 &fast_release, 0, "Passive Release was optimal");
168 static int slow_release;
169 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
170 &slow_release, 0, "Passive Release was nonoptimal");
172 MALLOC_DEFINE(M_SYSMSG, "sysmsg", "sysmsg structure");
173 extern int max_sysmsg;
176 * Passively intercepts the thread switch function to increase
177 * the thread priority from a user priority to a kernel priority, reducing
178 * syscall and trap overhead for the case where no switch occurs.
180 * Synchronizes td_ucred with p_ucred. This is used by system calls,
181 * signal handling, faults, AST traps, and anything else that enters the
182 * kernel from userland and provides the kernel with a stable read-only
183 * copy of the process ucred.
186 userenter(struct thread *curtd, struct proc *curp)
191 curtd->td_release = lwkt_passive_release;
193 if (curtd->td_ucred != curp->p_ucred) {
194 ncred = crhold(curp->p_ucred);
195 ocred = curtd->td_ucred;
196 curtd->td_ucred = ncred;
203 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
204 * must be completed before we can return to or try to return to userland.
206 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
207 * arithmatic on the delta calculation so the absolute tick values are
208 * truncated to an integer.
211 userret(struct lwp *lp, struct trapframe *frame, int sticks)
213 struct proc *p = lp->lwp_proc;
217 * Charge system time if profiling. Note: times are in microseconds.
218 * This may do a copyout and block, so do it first even though it
219 * means some system time will be charged as user time.
221 if (p->p_flag & 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_flag & LWP_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) {
247 * Post any pending upcalls
249 if (p->p_flag & P_UPCALLPEND) {
250 lwkt_gettoken(&p->p_token);
251 p->p_flag &= ~P_UPCALLPEND;
253 lwkt_reltoken(&p->p_token);
258 * Post any pending signals
260 * WARNING! postsig() can exit and not return.
262 if ((sig = CURSIG_TRACE(lp)) != 0) {
263 lwkt_gettoken(&p->p_token);
265 lwkt_reltoken(&p->p_token);
270 * block here if we are swapped out, but still process signals
271 * (such as SIGKILL). proc0 (the swapin scheduler) is already
272 * aware of our situation, we do not have to wake it up.
274 if (p->p_flag & P_SWAPPEDOUT) {
275 lwkt_gettoken(&p->p_token);
277 p->p_flag |= P_SWAPWAIT;
279 if (p->p_flag & P_SWAPWAIT)
280 tsleep(p, PCATCH, "SWOUT", 0);
281 p->p_flag &= ~P_SWAPWAIT;
283 lwkt_reltoken(&p->p_token);
288 * Make sure postsig() handled request to restore old signal mask after
289 * running signal handler.
291 KKASSERT((lp->lwp_flag & LWP_OLDMASK) == 0);
295 * Cleanup from userenter and any passive release that might have occured.
296 * We must reclaim the current-process designation before we can return
297 * to usermode. We also handle both LWKT and USER reschedule requests.
300 userexit(struct lwp *lp)
302 struct thread *td = lp->lwp_thread;
303 /* globaldata_t gd = td->td_gd; */
306 * Handle stop requests at kernel priority. Any requests queued
307 * after this loop will generate another AST.
309 while (lp->lwp_proc->p_stat == SSTOP) {
316 * Reduce our priority in preparation for a return to userland. If
317 * our passive release function was still in place, our priority was
318 * never raised and does not need to be reduced.
320 lwkt_passive_recover(td);
323 * Become the current user scheduled process if we aren't already,
324 * and deal with reschedule requests and other factors.
326 lp->lwp_proc->p_usched->acquire_curproc(lp);
327 /* WARNING: we may have migrated cpu's */
328 /* gd = td->td_gd; */
331 #if !defined(KTR_KERNENTRY)
332 #define KTR_KERNENTRY KTR_ALL
334 KTR_INFO_MASTER(kernentry);
335 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, "pid=%d, tid=%d, trapno=%d, eva=%p",
336 sizeof(int) + sizeof(int) + sizeof(int) + sizeof(vm_offset_t));
337 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "pid=%d, tid=%d",
338 sizeof(int) + sizeof(int));
339 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "pid=%d, tid=%d, call=%d",
340 sizeof(int) + sizeof(int) + sizeof(int));
341 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "pid=%d, tid=%d, err=%d",
342 sizeof(int) + sizeof(int) + sizeof(int));
343 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "pid=%d, tid=%d",
344 sizeof(int) + sizeof(int));
347 * Exception, fault, and trap interface to the kernel.
348 * This common code is called from assembly language IDT gate entry
349 * routines that prepare a suitable stack frame, and restore this
350 * frame after the exception has been processed.
352 * This function is also called from doreti in an interlock to handle ASTs.
353 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
355 * NOTE! We have to retrieve the fault address prior to obtaining the
356 * MP lock because get_mplock() may switch out. YYY cr2 really ought
357 * to be retrieved by the assembly code, not here.
359 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
360 * if an attempt is made to switch from a fast interrupt or IPI. This is
361 * necessary to properly take fatal kernel traps on SMP machines if
362 * get_mplock() has to block.
366 user_trap(struct trapframe *frame)
368 struct globaldata *gd = mycpu;
369 struct thread *td = gd->gd_curthread;
370 struct lwp *lp = td->td_lwp;
373 int i = 0, ucode = 0, type, code;
378 int crit_count = td->td_critcount;
379 lwkt_tokref_t curstop = td->td_toks_stop;
386 * This is a bad kludge to avoid changing the various trapframe
387 * structures. Because we are enabled as a virtual kernel,
388 * the original tf_err field will be passed to us shifted 16
389 * over in the tf_trapno field for T_PAGEFLT.
391 if (frame->tf_trapno == T_PAGEFLT)
396 kprintf("USER_TRAP AT %08x xflags %d trapno %d eva %08x\n",
397 frame->tf_eip, frame->tf_xflags, frame->tf_trapno, eva);
401 * Everything coming from user mode runs through user_trap,
402 * including system calls.
404 if (frame->tf_trapno == T_SYSCALL80) {
409 KTR_LOG(kernentry_trap, lp->lwp_proc->p_pid, lp->lwp_tid,
410 frame->tf_trapno, eva);
414 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
415 ++gd->gd_trap_nesting_level;
416 MAKEMPSAFE(have_mplock);
417 trap_fatal(frame, TRUE, eva);
418 --gd->gd_trap_nesting_level;
423 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
426 type = frame->tf_trapno;
427 code = frame->tf_err;
431 sticks = (int)td->td_sticks;
432 lp->lwp_md.md_regs = frame;
435 case T_PRIVINFLT: /* privileged instruction fault */
440 case T_BPTFLT: /* bpt instruction fault */
441 case T_TRCTRAP: /* trace trap */
442 frame->tf_eflags &= ~PSL_T;
447 case T_ARITHTRAP: /* arithmetic trap */
452 case T_ASTFLT: /* Allow process switch */
453 mycpu->gd_cnt.v_soft++;
454 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
455 atomic_clear_int(&mycpu->gd_reqflags,
457 addupc_task(p, p->p_prof.pr_addr,
463 * The following two traps can happen in
464 * vm86 mode, and, if so, we want to handle
467 case T_PROTFLT: /* general protection fault */
468 case T_STKFLT: /* stack fault */
470 if (frame->tf_eflags & PSL_VM) {
471 i = vm86_emulate((struct vm86frame *)frame);
478 ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR;
480 case T_SEGNPFLT: /* segment not present fault */
484 case T_TSSFLT: /* invalid TSS fault */
485 case T_DOUBLEFLT: /* double fault */
490 ucode = code + BUS_SEGM_FAULT ; /* XXX: ???*/
496 case T_PAGEFLT: /* page fault */
497 MAKEMPSAFE(have_mplock);
498 i = trap_pfault(frame, TRUE, eva);
501 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
517 case T_DIVIDE: /* integer divide fault */
524 MAKEMPSAFE(have_mplock);
525 /* machine/parity/power fail/"kitchen sink" faults */
526 if (isa_nmi(code) == 0) {
529 * NMI can be hooked up to a pushbutton
533 kprintf ("NMI ... going to debugger\n");
534 kdb_trap (type, 0, frame);
538 } else if (panic_on_nmi)
539 panic("NMI indicates hardware failure");
541 #endif /* NISA > 0 */
543 case T_OFLOW: /* integer overflow fault */
548 case T_BOUND: /* bounds check fault */
555 * Virtual kernel intercept - pass the DNA exception
556 * to the (emulated) virtual kernel if it asked to handle
557 * it. This occurs when the virtual kernel is holding
558 * onto the FP context for a different emulated
559 * process then the one currently running.
561 * We must still call npxdna() since we may have
562 * saved FP state that the (emulated) virtual kernel
563 * needs to hand over to a different emulated process.
565 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
566 (td->td_pcb->pcb_flags & FP_VIRTFP)
573 * The kernel may have switched out the FP unit's
574 * state, causing the user process to take a fault
575 * when it tries to use the FP unit. Restore the
581 if (!pmath_emulate) {
583 ucode = FPE_FPU_NP_TRAP;
586 i = (*pmath_emulate)(frame);
588 if (!(frame->tf_eflags & PSL_T))
590 frame->tf_eflags &= ~PSL_T;
593 /* else ucode = emulator_only_knows() XXX */
596 case T_FPOPFLT: /* FPU operand fetch fault */
601 case T_XMMFLT: /* SIMD floating-point exception */
608 * Virtual kernel intercept - if the fault is directly related to a
609 * VM context managed by a virtual kernel then let the virtual kernel
612 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
613 vkernel_trap(lp, frame);
618 * Translate fault for emulators (e.g. Linux)
620 if (*p->p_sysent->sv_transtrap)
621 i = (*p->p_sysent->sv_transtrap)(i, type);
623 MAKEMPSAFE(have_mplock);
624 trapsignal(lp, i, ucode);
627 if (type <= MAX_TRAP_MSG) {
628 uprintf("fatal process exception: %s",
630 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
631 uprintf(", fault VA = 0x%lx", (u_long)eva);
637 userret(lp, frame, sticks);
644 KTR_LOG(kernentry_trap_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
646 KASSERT(crit_count == td->td_critcount,
647 ("trap: critical section count mismatch! %d/%d",
648 crit_count, td->td_pri));
649 KASSERT(curstop == td->td_toks_stop,
650 ("trap: extra tokens held after trap! %zd/%zd",
651 curstop - &td->td_toks_base,
652 td->td_toks_stop - &td->td_toks_base));
657 kern_trap(struct trapframe *frame)
659 struct globaldata *gd = mycpu;
660 struct thread *td = gd->gd_curthread;
663 int i = 0, ucode = 0, type, code;
668 int crit_count = td->td_critcount;
669 lwkt_tokref_t curstop = td->td_toks_stop;
676 if (frame->tf_trapno == T_PAGEFLT)
683 ++gd->gd_trap_nesting_level;
684 MAKEMPSAFE(have_mplock);
685 trap_fatal(frame, FALSE, eva);
686 --gd->gd_trap_nesting_level;
690 type = frame->tf_trapno;
691 code = frame->tf_err;
699 case T_PAGEFLT: /* page fault */
700 MAKEMPSAFE(have_mplock);
701 trap_pfault(frame, FALSE, eva);
707 * The kernel may be using npx for copying or other
710 panic("kernel NPX should not happen");
716 case T_PROTFLT: /* general protection fault */
717 case T_SEGNPFLT: /* segment not present fault */
719 * Invalid segment selectors and out of bounds
720 * %eip's and %esp's can be set up in user mode.
721 * This causes a fault in kernel mode when the
722 * kernel tries to return to user mode. We want
723 * to get this fault so that we can fix the
724 * problem here and not have to check all the
725 * selectors and pointers when the user changes
728 if (mycpu->gd_intr_nesting_level == 0) {
729 if (td->td_pcb->pcb_onfault) {
731 (register_t)td->td_pcb->pcb_onfault;
739 * PSL_NT can be set in user mode and isn't cleared
740 * automatically when the kernel is entered. This
741 * causes a TSS fault when the kernel attempts to
742 * `iret' because the TSS link is uninitialized. We
743 * want to get this fault so that we can fix the
744 * problem here and not every time the kernel is
747 if (frame->tf_eflags & PSL_NT) {
748 frame->tf_eflags &= ~PSL_NT;
753 case T_TRCTRAP: /* trace trap */
755 if (frame->tf_eip == (int)IDTVEC(syscall)) {
757 * We've just entered system mode via the
758 * syscall lcall. Continue single stepping
759 * silently until the syscall handler has
764 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
766 * The syscall handler has now saved the
767 * flags. Stop single stepping it.
769 frame->tf_eflags &= ~PSL_T;
775 * Ignore debug register trace traps due to
776 * accesses in the user's address space, which
777 * can happen under several conditions such as
778 * if a user sets a watchpoint on a buffer and
779 * then passes that buffer to a system call.
780 * We still want to get TRCTRAPS for addresses
781 * in kernel space because that is useful when
782 * debugging the kernel.
784 if (user_dbreg_trap()) {
786 * Reset breakpoint bits because the
789 load_dr6(rdr6() & 0xfffffff0);
794 * Fall through (TRCTRAP kernel mode, kernel address)
798 * If DDB is enabled, let it handle the debugger trap.
799 * Otherwise, debugger traps "can't happen".
802 MAKEMPSAFE(have_mplock);
803 if (kdb_trap (type, 0, frame))
808 MAKEMPSAFE(have_mplock);
809 trap_fatal(frame, FALSE, eva);
812 MAKEMPSAFE(have_mplock);
813 trap_fatal(frame, FALSE, eva);
817 * Ignore this trap generated from a spurious SIGTRAP.
819 * single stepping in / syscalls leads to spurious / SIGTRAP
822 * Haiku (c) 2007 Simon 'corecode' Schubert
828 * Translate fault for emulators (e.g. Linux)
830 if (*p->p_sysent->sv_transtrap)
831 i = (*p->p_sysent->sv_transtrap)(i, type);
833 MAKEMPSAFE(have_mplock);
834 trapsignal(lp, i, ucode);
837 if (type <= MAX_TRAP_MSG) {
838 uprintf("fatal process exception: %s",
840 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
841 uprintf(", fault VA = 0x%lx", (u_long)eva);
853 KASSERT(crit_count == td->td_critcount,
854 ("trap: critical section count mismatch! %d/%d",
855 crit_count, td->td_pri));
856 KASSERT(curstop == td->td_toks_stop,
857 ("trap: extra tokens held after trap! %zd/%zd",
858 curstop - &td->td_toks_base,
859 td->td_toks_stop - &td->td_toks_base));
864 trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
867 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
913 fault_flags |= VM_FAULT_BURST;
914 if (ftype & VM_PROT_WRITE)
915 fault_flags |= VM_FAULT_DIRTY;
917 fault_flags |= VM_FAULT_NORMAL;
918 rv = vm_fault(map, va, ftype, fault_flags);
923 * Don't have to worry about process locking or stacks in the kernel.
925 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
928 if (rv == KERN_SUCCESS)
932 if (td->td_gd->gd_intr_nesting_level == 0 &&
933 td->td_pcb->pcb_onfault) {
934 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
937 trap_fatal(frame, usermode, eva);
940 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
944 trap_fatal(struct trapframe *frame, int usermode, vm_offset_t eva)
946 int code, type, ss, esp;
948 code = frame->tf_xflags;
949 type = frame->tf_trapno;
951 if (type <= MAX_TRAP_MSG) {
952 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
953 type, trap_msg[type],
954 (usermode ? "user" : "kernel"));
957 /* two separate prints in case of a trap on an unmapped page */
958 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
960 if (type == T_PAGEFLT) {
961 kprintf("fault virtual address = %p\n", (void *)eva);
962 kprintf("fault code = %s %s, %s\n",
963 usermode ? "user" : "supervisor",
964 code & PGEX_W ? "write" : "read",
965 code & PGEX_P ? "protection violation" : "page not present");
967 kprintf("instruction pointer = 0x%x:0x%x\n",
968 frame->tf_cs & 0xffff, frame->tf_eip);
970 ss = frame->tf_ss & 0xffff;
973 ss = GSEL(GDATA_SEL, SEL_KPL);
974 esp = (int)&frame->tf_esp;
976 kprintf("stack pointer = 0x%x:0x%x\n", ss, esp);
977 kprintf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
978 kprintf("processor eflags = ");
979 if (frame->tf_eflags & PSL_T)
980 kprintf("trace trap, ");
981 if (frame->tf_eflags & PSL_I)
982 kprintf("interrupt enabled, ");
983 if (frame->tf_eflags & PSL_NT)
984 kprintf("nested task, ");
985 if (frame->tf_eflags & PSL_RF)
988 if (frame->tf_eflags & PSL_VM)
991 kprintf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
992 kprintf("current process = ");
994 kprintf("%lu (%s)\n",
995 (u_long)curproc->p_pid, curproc->p_comm ?
996 curproc->p_comm : "");
1000 kprintf("current thread = pri %d ", curthread->td_pri);
1001 if (curthread->td_critcount)
1007 * we probably SHOULD have stopped the other CPUs before now!
1008 * another CPU COULD have been touching cpl at this moment...
1010 kprintf(" <- SMP: XXX");
1019 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1022 kprintf("trap number = %d\n", type);
1023 if (type <= MAX_TRAP_MSG)
1024 panic("%s", trap_msg[type]);
1026 panic("unknown/reserved trap");
1030 * Double fault handler. Called when a fault occurs while writing
1031 * a frame for a trap/exception onto the stack. This usually occurs
1032 * when the stack overflows (such is the case with infinite recursion,
1035 * XXX Note that the current PTD gets replaced by IdlePTD when the
1036 * task switch occurs. This means that the stack that was active at
1037 * the time of the double fault is not available at <kstack> unless
1038 * the machine was idle when the double fault occurred. The downside
1039 * of this is that "trace <ebp>" in ddb won't work.
1042 dblfault_handler(void)
1044 struct mdglobaldata *gd = mdcpu;
1046 kprintf("\nFatal double fault:\n");
1047 kprintf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1048 kprintf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1049 kprintf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
1051 /* two separate prints in case of a trap on an unmapped page */
1052 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
1054 panic("double fault");
1058 * syscall2 - MP aware system call request C handler
1060 * A system call is essentially treated as a trap except that the
1061 * MP lock is not held on entry or return. We are responsible for
1062 * obtaining the MP lock if necessary and for handling ASTs
1063 * (e.g. a task switch) prior to return.
1068 syscall2(struct trapframe *frame)
1070 struct thread *td = curthread;
1071 struct proc *p = td->td_proc;
1072 struct lwp *lp = td->td_lwp;
1074 struct sysent *callp;
1075 register_t orig_tf_eflags;
1080 int crit_count = td->td_critcount;
1083 int have_mplock = 0;
1086 union sysunion args;
1088 KTR_LOG(kernentry_syscall, lp->lwp_proc->p_pid, lp->lwp_tid,
1091 userenter(td, p); /* lazy raise our priority */
1096 sticks = (int)td->td_sticks;
1097 orig_tf_eflags = frame->tf_eflags;
1100 * Virtual kernel intercept - if a VM context managed by a virtual
1101 * kernel issues a system call the virtual kernel handles it, not us.
1102 * Restore the virtual kernel context and return from its system
1103 * call. The current frame is copied out to the virtual kernel.
1105 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1106 vkernel_trap(lp, frame);
1107 error = EJUSTRETURN;
1112 * Get the system call parameters and account for time
1114 lp->lwp_md.md_regs = frame;
1115 params = (caddr_t)frame->tf_esp + sizeof(int);
1116 code = frame->tf_eax;
1118 if (p->p_sysent->sv_prepsyscall) {
1119 (*p->p_sysent->sv_prepsyscall)(
1120 frame, (int *)(&args.nosys.sysmsg + 1),
1124 * Need to check if this is a 32 bit or 64 bit syscall.
1125 * fuword is MP aware.
1127 if (code == SYS_syscall) {
1129 * Code is first argument, followed by actual args.
1131 code = fuword(params);
1132 params += sizeof(int);
1133 } else if (code == SYS___syscall) {
1135 * Like syscall, but code is a quad, so as to maintain
1136 * quad alignment for the rest of the arguments.
1138 code = fuword(params);
1139 params += sizeof(quad_t);
1143 code &= p->p_sysent->sv_mask;
1144 if (code >= p->p_sysent->sv_size)
1145 callp = &p->p_sysent->sv_table[0];
1147 callp = &p->p_sysent->sv_table[code];
1149 narg = callp->sy_narg & SYF_ARGMASK;
1152 * copyin is MP aware, but the tracing code is not
1154 if (narg && params) {
1155 error = copyin(params, (caddr_t)(&args.nosys.sysmsg + 1),
1156 narg * sizeof(register_t));
1159 if (KTRPOINT(td, KTR_SYSCALL)) {
1160 MAKEMPSAFE(have_mplock);
1162 ktrsyscall(lp, code, narg,
1163 (void *)(&args.nosys.sysmsg + 1));
1171 if (KTRPOINT(td, KTR_SYSCALL)) {
1172 MAKEMPSAFE(have_mplock);
1173 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1178 * For traditional syscall code edx is left untouched when 32 bit
1179 * results are returned. Since edx is loaded from fds[1] when the
1180 * system call returns we pre-set it here.
1182 args.sysmsg_fds[0] = 0;
1183 args.sysmsg_fds[1] = frame->tf_edx;
1186 * The syscall might manipulate the trap frame. If it does it
1187 * will probably return EJUSTRETURN.
1189 args.sysmsg_frame = frame;
1191 STOPEVENT(p, S_SCE, narg); /* MP aware */
1194 * NOTE: All system calls run MPSAFE now. The system call itself
1195 * is responsible for getting the MP lock.
1197 error = (*callp->sy_call)(&args);
1200 kprintf("system call %d returned %d\n", code, error);
1205 * MP SAFE (we may or may not have the MP lock at this point)
1210 * Reinitialize proc pointer `p' as it may be different
1211 * if this is a child returning from fork syscall.
1214 lp = curthread->td_lwp;
1215 frame->tf_eax = args.sysmsg_fds[0];
1216 frame->tf_edx = args.sysmsg_fds[1];
1217 frame->tf_eflags &= ~PSL_C;
1221 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1222 * int 0x80 is 2 bytes. We saved this in tf_err.
1224 frame->tf_eip -= frame->tf_err;
1229 panic("Unexpected EASYNC return value (for now)");
1232 if (p->p_sysent->sv_errsize) {
1233 if (error >= p->p_sysent->sv_errsize)
1234 error = -1; /* XXX */
1236 error = p->p_sysent->sv_errtbl[error];
1238 frame->tf_eax = error;
1239 frame->tf_eflags |= PSL_C;
1244 * Traced syscall. trapsignal() is not MP aware.
1246 if ((orig_tf_eflags & PSL_T) /*&& !(orig_tf_eflags & PSL_VM)*/) {
1247 MAKEMPSAFE(have_mplock);
1248 frame->tf_eflags &= ~PSL_T;
1249 trapsignal(lp, SIGTRAP, TRAP_TRACE);
1253 * Handle reschedule and other end-of-syscall issues
1255 userret(lp, frame, sticks);
1258 if (KTRPOINT(td, KTR_SYSRET)) {
1259 MAKEMPSAFE(have_mplock);
1260 ktrsysret(lp, code, error, args.sysmsg_result);
1265 * This works because errno is findable through the
1266 * register set. If we ever support an emulation where this
1267 * is not the case, this code will need to be revisited.
1269 STOPEVENT(p, S_SCX, code);
1274 * Release the MP lock if we had to get it
1279 KTR_LOG(kernentry_syscall_ret, lp->lwp_proc->p_pid, lp->lwp_tid, error);
1281 KASSERT(crit_count == td->td_critcount,
1282 ("syscall: critical section count mismatch! %d/%d",
1283 crit_count, td->td_pri));
1284 KASSERT(&td->td_toks_base == td->td_toks_stop,
1285 ("syscall: extra tokens held after trap! %zd",
1286 td->td_toks_stop - &td->td_toks_base));
1291 * NOTE: mplock not held at any point
1294 fork_return(struct lwp *lp, struct trapframe *frame)
1296 frame->tf_eax = 0; /* Child returns zero */
1297 frame->tf_eflags &= ~PSL_C; /* success */
1300 generic_lwp_return(lp, frame);
1301 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1305 * Simplified back end of syscall(), used when returning from fork()
1306 * directly into user mode.
1308 * This code will return back into the fork trampoline code which then
1311 * NOTE: The mplock is not held at any point.
1314 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1316 struct proc *p = lp->lwp_proc;
1319 * Newly forked processes are given a kernel priority. We have to
1320 * adjust the priority to a normal user priority and fake entry
1321 * into the kernel (call userenter()) to install a passive release
1322 * function just in case userret() decides to stop the process. This
1323 * can occur when ^Z races a fork. If we do not install the passive
1324 * release function the current process designation will not be
1325 * released when the thread goes to sleep.
1327 lwkt_setpri_self(TDPRI_USER_NORM);
1328 userenter(lp->lwp_thread, p);
1329 userret(lp, frame, 0);
1331 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1332 ktrsysret(lp, SYS_fork, 0, 0);
1334 lp->lwp_flag |= LWP_PASSIVE_ACQ;
1336 lp->lwp_flag &= ~LWP_PASSIVE_ACQ;
1340 * doreti has turned into this. The frame is directly on the stack. We
1341 * pull everything else we need (fpu and tls context) from the current
1344 * Note on fpu interactions: In a virtual kernel, the fpu context for
1345 * an emulated user mode process is not shared with the virtual kernel's
1346 * fpu context, so we only have to 'stack' fpu contexts within the virtual
1347 * kernel itself, and not even then since the signal() contexts that we care
1348 * about save and restore the FPU state (I think anyhow).
1350 * vmspace_ctl() returns an error only if it had problems instaling the
1351 * context we supplied or problems copying data to/from our VM space.
1354 go_user(struct intrframe *frame)
1356 struct trapframe *tf = (void *)&frame->if_gs;
1360 * Interrupts may be disabled on entry, make sure all signals
1361 * can be received before beginning our loop.
1366 * Switch to the current simulated user process, then call
1367 * user_trap() when we break out of it (usually due to a signal).
1371 * Tell the real kernel whether it is ok to use the FP
1374 * The critical section is required to prevent an interrupt
1375 * from causing a preemptive task switch and changing
1379 if (mdcpu->gd_npxthread == curthread) {
1380 tf->tf_xflags &= ~PGEX_FPFAULT;
1382 tf->tf_xflags |= PGEX_FPFAULT;
1386 * Run emulated user process context. This call interlocks
1387 * with new mailbox signals.
1389 * Set PGEX_U unconditionally, indicating a user frame (the
1390 * bit is normally set only by T_PAGEFLT).
1392 r = vmspace_ctl(&curproc->p_vmspace->vm_pmap, VMSPACE_CTL_RUN,
1393 tf, &curthread->td_savevext);
1395 frame->if_xflags |= PGEX_U;
1397 kprintf("GO USER %d trap %d EVA %08x EIP %08x ESP %08x XFLAGS %02x/%02x\n",
1398 r, tf->tf_trapno, tf->tf_err, tf->tf_eip, tf->tf_esp,
1399 tf->tf_xflags, frame->if_xflags);
1403 panic("vmspace_ctl failed error %d", errno);
1405 if (tf->tf_trapno) {
1409 if (mycpu->gd_reqflags & RQF_AST_MASK) {
1410 tf->tf_trapno = T_ASTFLT;
1418 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1419 * fault (which is then passed back to the virtual kernel) if an attempt is
1420 * made to use the FP unit.
1422 * XXX this is a fairly big hack.
1425 set_vkernel_fp(struct trapframe *frame)
1427 struct thread *td = curthread;
1429 if (frame->tf_xflags & PGEX_FPFAULT) {
1430 td->td_pcb->pcb_flags |= FP_VIRTFP;
1431 if (mdcpu->gd_npxthread == td)
1434 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1439 * Called from vkernel_trap() to fixup the vkernel's syscall
1440 * frame for vmspace_ctl() return.
1443 cpu_vkernel_trap(struct trapframe *frame, int error)
1445 frame->tf_eax = error;
1447 frame->tf_eflags |= PSL_C;
1449 frame->tf_eflags &= ~PSL_C;