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) {
233 rel_mplock(); /* 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) {
251 p->p_flag &= ~P_UPCALLPEND;
258 * Post any pending signals
260 if ((sig = CURSIG_TRACE(lp)) != 0) {
268 * block here if we are swapped out, but still process signals
269 * (such as SIGKILL). proc0 (the swapin scheduler) is already
270 * aware of our situation, we do not have to wake it up.
272 if (p->p_flag & P_SWAPPEDOUT) {
274 p->p_flag |= P_SWAPWAIT;
276 if (p->p_flag & P_SWAPWAIT)
277 tsleep(p, PCATCH, "SWOUT", 0);
278 p->p_flag &= ~P_SWAPWAIT;
284 * Make sure postsig() handled request to restore old signal mask after
285 * running signal handler.
287 KKASSERT((lp->lwp_flag & LWP_OLDMASK) == 0);
291 * Cleanup from userenter and any passive release that might have occured.
292 * We must reclaim the current-process designation before we can return
293 * to usermode. We also handle both LWKT and USER reschedule requests.
296 userexit(struct lwp *lp)
298 struct thread *td = lp->lwp_thread;
299 /* globaldata_t gd = td->td_gd; */
302 * Handle stop requests at kernel priority. Any requests queued
303 * after this loop will generate another AST.
305 while (lp->lwp_proc->p_stat == SSTOP) {
312 * Reduce our priority in preparation for a return to userland. If
313 * our passive release function was still in place, our priority was
314 * never raised and does not need to be reduced.
316 lwkt_passive_recover(td);
319 * Become the current user scheduled process if we aren't already,
320 * and deal with reschedule requests and other factors.
322 lp->lwp_proc->p_usched->acquire_curproc(lp);
323 /* WARNING: we may have migrated cpu's */
324 /* gd = td->td_gd; */
327 #if !defined(KTR_KERNENTRY)
328 #define KTR_KERNENTRY KTR_ALL
330 KTR_INFO_MASTER(kernentry);
331 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, "pid=%d, tid=%d, trapno=%d, eva=%p",
332 sizeof(int) + sizeof(int) + sizeof(int) + sizeof(vm_offset_t));
333 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "pid=%d, tid=%d",
334 sizeof(int) + sizeof(int));
335 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "pid=%d, tid=%d, call=%d",
336 sizeof(int) + sizeof(int) + sizeof(int));
337 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "pid=%d, tid=%d, err=%d",
338 sizeof(int) + sizeof(int) + sizeof(int));
339 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "pid=%d, tid=%d",
340 sizeof(int) + sizeof(int));
343 * Exception, fault, and trap interface to the kernel.
344 * This common code is called from assembly language IDT gate entry
345 * routines that prepare a suitable stack frame, and restore this
346 * frame after the exception has been processed.
348 * This function is also called from doreti in an interlock to handle ASTs.
349 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
351 * NOTE! We have to retrieve the fault address prior to obtaining the
352 * MP lock because get_mplock() may switch out. YYY cr2 really ought
353 * to be retrieved by the assembly code, not here.
355 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
356 * if an attempt is made to switch from a fast interrupt or IPI. This is
357 * necessary to properly take fatal kernel traps on SMP machines if
358 * get_mplock() has to block.
362 user_trap(struct trapframe *frame)
364 struct globaldata *gd = mycpu;
365 struct thread *td = gd->gd_curthread;
366 struct lwp *lp = td->td_lwp;
369 int i = 0, ucode = 0, type, code;
374 int crit_count = td->td_critcount;
381 * This is a bad kludge to avoid changing the various trapframe
382 * structures. Because we are enabled as a virtual kernel,
383 * the original tf_err field will be passed to us shifted 16
384 * over in the tf_trapno field for T_PAGEFLT.
386 if (frame->tf_trapno == T_PAGEFLT)
391 kprintf("USER_TRAP AT %08x xflags %d trapno %d eva %08x\n",
392 frame->tf_eip, frame->tf_xflags, frame->tf_trapno, eva);
396 * Everything coming from user mode runs through user_trap,
397 * including system calls.
399 if (frame->tf_trapno == T_SYSCALL80) {
404 KTR_LOG(kernentry_trap, lp->lwp_proc->p_pid, lp->lwp_tid,
405 frame->tf_trapno, eva);
409 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
410 ++gd->gd_trap_nesting_level;
411 MAKEMPSAFE(have_mplock);
412 trap_fatal(frame, TRUE, eva);
413 --gd->gd_trap_nesting_level;
418 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
421 type = frame->tf_trapno;
422 code = frame->tf_err;
426 sticks = (int)td->td_sticks;
427 lp->lwp_md.md_regs = frame;
430 case T_PRIVINFLT: /* privileged instruction fault */
435 case T_BPTFLT: /* bpt instruction fault */
436 case T_TRCTRAP: /* trace trap */
437 frame->tf_eflags &= ~PSL_T;
442 case T_ARITHTRAP: /* arithmetic trap */
447 case T_ASTFLT: /* Allow process switch */
448 mycpu->gd_cnt.v_soft++;
449 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
450 atomic_clear_int_nonlocked(&mycpu->gd_reqflags,
452 addupc_task(p, p->p_prof.pr_addr,
458 * The following two traps can happen in
459 * vm86 mode, and, if so, we want to handle
462 case T_PROTFLT: /* general protection fault */
463 case T_STKFLT: /* stack fault */
465 if (frame->tf_eflags & PSL_VM) {
466 i = vm86_emulate((struct vm86frame *)frame);
473 ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR;
475 case T_SEGNPFLT: /* segment not present fault */
479 case T_TSSFLT: /* invalid TSS fault */
480 case T_DOUBLEFLT: /* double fault */
485 ucode = code + BUS_SEGM_FAULT ; /* XXX: ???*/
491 case T_PAGEFLT: /* page fault */
492 MAKEMPSAFE(have_mplock);
493 i = trap_pfault(frame, TRUE, eva);
496 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
512 case T_DIVIDE: /* integer divide fault */
519 MAKEMPSAFE(have_mplock);
520 /* machine/parity/power fail/"kitchen sink" faults */
521 if (isa_nmi(code) == 0) {
524 * NMI can be hooked up to a pushbutton
528 kprintf ("NMI ... going to debugger\n");
529 kdb_trap (type, 0, frame);
533 } else if (panic_on_nmi)
534 panic("NMI indicates hardware failure");
536 #endif /* NISA > 0 */
538 case T_OFLOW: /* integer overflow fault */
543 case T_BOUND: /* bounds check fault */
550 * Virtual kernel intercept - pass the DNA exception
551 * to the (emulated) virtual kernel if it asked to handle
552 * it. This occurs when the virtual kernel is holding
553 * onto the FP context for a different emulated
554 * process then the one currently running.
556 * We must still call npxdna() since we may have
557 * saved FP state that the (emulated) virtual kernel
558 * needs to hand over to a different emulated process.
560 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
561 (td->td_pcb->pcb_flags & FP_VIRTFP)
568 * The kernel may have switched out the FP unit's
569 * state, causing the user process to take a fault
570 * when it tries to use the FP unit. Restore the
576 if (!pmath_emulate) {
578 ucode = FPE_FPU_NP_TRAP;
581 i = (*pmath_emulate)(frame);
583 if (!(frame->tf_eflags & PSL_T))
585 frame->tf_eflags &= ~PSL_T;
588 /* else ucode = emulator_only_knows() XXX */
591 case T_FPOPFLT: /* FPU operand fetch fault */
596 case T_XMMFLT: /* SIMD floating-point exception */
603 * Virtual kernel intercept - if the fault is directly related to a
604 * VM context managed by a virtual kernel then let the virtual kernel
607 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
608 vkernel_trap(lp, frame);
613 * Translate fault for emulators (e.g. Linux)
615 if (*p->p_sysent->sv_transtrap)
616 i = (*p->p_sysent->sv_transtrap)(i, type);
618 MAKEMPSAFE(have_mplock);
619 trapsignal(lp, i, ucode);
622 if (type <= MAX_TRAP_MSG) {
623 uprintf("fatal process exception: %s",
625 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
626 uprintf(", fault VA = 0x%lx", (u_long)eva);
633 KASSERT(td->td_mpcount == have_mplock, ("badmpcount trap/end from %p", (void *)frame->tf_eip));
635 userret(lp, frame, sticks);
642 KTR_LOG(kernentry_trap_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
644 KASSERT(crit_count == td->td_critcount,
645 ("syscall: critical section count mismatch! %d/%d",
646 crit_count, td->td_pri));
651 kern_trap(struct trapframe *frame)
653 struct globaldata *gd = mycpu;
654 struct thread *td = gd->gd_curthread;
657 int i = 0, ucode = 0, type, code;
662 int crit_count = td->td_critcount;
669 if (frame->tf_trapno == T_PAGEFLT)
676 ++gd->gd_trap_nesting_level;
677 MAKEMPSAFE(have_mplock);
678 trap_fatal(frame, FALSE, eva);
679 --gd->gd_trap_nesting_level;
683 type = frame->tf_trapno;
684 code = frame->tf_err;
692 case T_PAGEFLT: /* page fault */
693 MAKEMPSAFE(have_mplock);
694 trap_pfault(frame, FALSE, eva);
700 * The kernel may be using npx for copying or other
703 panic("kernel NPX should not happen");
709 case T_PROTFLT: /* general protection fault */
710 case T_SEGNPFLT: /* segment not present fault */
712 * Invalid segment selectors and out of bounds
713 * %eip's and %esp's can be set up in user mode.
714 * This causes a fault in kernel mode when the
715 * kernel tries to return to user mode. We want
716 * to get this fault so that we can fix the
717 * problem here and not have to check all the
718 * selectors and pointers when the user changes
721 if (mycpu->gd_intr_nesting_level == 0) {
722 if (td->td_pcb->pcb_onfault) {
724 (register_t)td->td_pcb->pcb_onfault;
732 * PSL_NT can be set in user mode and isn't cleared
733 * automatically when the kernel is entered. This
734 * causes a TSS fault when the kernel attempts to
735 * `iret' because the TSS link is uninitialized. We
736 * want to get this fault so that we can fix the
737 * problem here and not every time the kernel is
740 if (frame->tf_eflags & PSL_NT) {
741 frame->tf_eflags &= ~PSL_NT;
746 case T_TRCTRAP: /* trace trap */
748 if (frame->tf_eip == (int)IDTVEC(syscall)) {
750 * We've just entered system mode via the
751 * syscall lcall. Continue single stepping
752 * silently until the syscall handler has
757 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
759 * The syscall handler has now saved the
760 * flags. Stop single stepping it.
762 frame->tf_eflags &= ~PSL_T;
768 * Ignore debug register trace traps due to
769 * accesses in the user's address space, which
770 * can happen under several conditions such as
771 * if a user sets a watchpoint on a buffer and
772 * then passes that buffer to a system call.
773 * We still want to get TRCTRAPS for addresses
774 * in kernel space because that is useful when
775 * debugging the kernel.
777 if (user_dbreg_trap()) {
779 * Reset breakpoint bits because the
782 load_dr6(rdr6() & 0xfffffff0);
787 * Fall through (TRCTRAP kernel mode, kernel address)
791 * If DDB is enabled, let it handle the debugger trap.
792 * Otherwise, debugger traps "can't happen".
795 MAKEMPSAFE(have_mplock);
796 if (kdb_trap (type, 0, frame))
801 MAKEMPSAFE(have_mplock);
802 trap_fatal(frame, FALSE, eva);
805 MAKEMPSAFE(have_mplock);
806 trap_fatal(frame, FALSE, eva);
810 * Ignore this trap generated from a spurious SIGTRAP.
812 * single stepping in / syscalls leads to spurious / SIGTRAP
815 * Haiku (c) 2007 Simon 'corecode' Schubert
821 * Translate fault for emulators (e.g. Linux)
823 if (*p->p_sysent->sv_transtrap)
824 i = (*p->p_sysent->sv_transtrap)(i, type);
826 MAKEMPSAFE(have_mplock);
827 trapsignal(lp, i, ucode);
830 if (type <= MAX_TRAP_MSG) {
831 uprintf("fatal process exception: %s",
833 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
834 uprintf(", fault VA = 0x%lx", (u_long)eva);
846 KASSERT(crit_count == td->td_critcount,
847 ("syscall: critical section count mismatch! %d/%d",
848 crit_count, td->td_pri));
853 trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
856 struct vmspace *vm = NULL;
861 thread_t td = curthread;
862 struct lwp *lp = td->td_lwp;
864 va = trunc_page(eva);
865 if (usermode == FALSE) {
867 * This is a fault on kernel virtual memory.
872 * This is a fault on non-kernel virtual memory.
873 * vm is initialized above to NULL. If curproc is NULL
874 * or curproc->p_vmspace is NULL the fault is fatal.
877 vm = lp->lwp_vmspace;
885 if (frame->tf_xflags & PGEX_W)
886 ftype = VM_PROT_READ | VM_PROT_WRITE;
888 ftype = VM_PROT_READ;
890 if (map != &kernel_map) {
892 * Keep swapout from messing with us during this
902 fault_flags |= VM_FAULT_BURST;
903 if (ftype & VM_PROT_WRITE)
904 fault_flags |= VM_FAULT_DIRTY;
906 fault_flags |= VM_FAULT_NORMAL;
907 rv = vm_fault(map, va, ftype, fault_flags);
912 * Don't have to worry about process locking or stacks in the kernel.
914 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
917 if (rv == KERN_SUCCESS)
921 if (td->td_gd->gd_intr_nesting_level == 0 &&
922 td->td_pcb->pcb_onfault) {
923 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
926 trap_fatal(frame, usermode, eva);
929 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
933 trap_fatal(struct trapframe *frame, int usermode, vm_offset_t eva)
935 int code, type, ss, esp;
937 code = frame->tf_xflags;
938 type = frame->tf_trapno;
940 if (type <= MAX_TRAP_MSG) {
941 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
942 type, trap_msg[type],
943 (usermode ? "user" : "kernel"));
946 /* two separate prints in case of a trap on an unmapped page */
947 kprintf("mp_lock = %08x; ", mp_lock);
948 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
950 if (type == T_PAGEFLT) {
951 kprintf("fault virtual address = %p\n", (void *)eva);
952 kprintf("fault code = %s %s, %s\n",
953 usermode ? "user" : "supervisor",
954 code & PGEX_W ? "write" : "read",
955 code & PGEX_P ? "protection violation" : "page not present");
957 kprintf("instruction pointer = 0x%x:0x%x\n",
958 frame->tf_cs & 0xffff, frame->tf_eip);
960 ss = frame->tf_ss & 0xffff;
963 ss = GSEL(GDATA_SEL, SEL_KPL);
964 esp = (int)&frame->tf_esp;
966 kprintf("stack pointer = 0x%x:0x%x\n", ss, esp);
967 kprintf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
968 kprintf("processor eflags = ");
969 if (frame->tf_eflags & PSL_T)
970 kprintf("trace trap, ");
971 if (frame->tf_eflags & PSL_I)
972 kprintf("interrupt enabled, ");
973 if (frame->tf_eflags & PSL_NT)
974 kprintf("nested task, ");
975 if (frame->tf_eflags & PSL_RF)
978 if (frame->tf_eflags & PSL_VM)
981 kprintf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
982 kprintf("current process = ");
984 kprintf("%lu (%s)\n",
985 (u_long)curproc->p_pid, curproc->p_comm ?
986 curproc->p_comm : "");
990 kprintf("current thread = pri %d ", curthread->td_pri);
991 if (curthread->td_critcount)
997 * we probably SHOULD have stopped the other CPUs before now!
998 * another CPU COULD have been touching cpl at this moment...
1000 kprintf(" <- SMP: XXX");
1009 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1012 kprintf("trap number = %d\n", type);
1013 if (type <= MAX_TRAP_MSG)
1014 panic("%s", trap_msg[type]);
1016 panic("unknown/reserved trap");
1020 * Double fault handler. Called when a fault occurs while writing
1021 * a frame for a trap/exception onto the stack. This usually occurs
1022 * when the stack overflows (such is the case with infinite recursion,
1025 * XXX Note that the current PTD gets replaced by IdlePTD when the
1026 * task switch occurs. This means that the stack that was active at
1027 * the time of the double fault is not available at <kstack> unless
1028 * the machine was idle when the double fault occurred. The downside
1029 * of this is that "trace <ebp>" in ddb won't work.
1032 dblfault_handler(void)
1034 struct mdglobaldata *gd = mdcpu;
1036 kprintf("\nFatal double fault:\n");
1037 kprintf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1038 kprintf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1039 kprintf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
1041 /* two separate prints in case of a trap on an unmapped page */
1042 kprintf("mp_lock = %08x; ", mp_lock);
1043 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
1045 panic("double fault");
1049 * syscall2 - MP aware system call request C handler
1051 * A system call is essentially treated as a trap except that the
1052 * MP lock is not held on entry or return. We are responsible for
1053 * obtaining the MP lock if necessary and for handling ASTs
1054 * (e.g. a task switch) prior to return.
1059 syscall2(struct trapframe *frame)
1061 struct thread *td = curthread;
1062 struct proc *p = td->td_proc;
1063 struct lwp *lp = td->td_lwp;
1065 struct sysent *callp;
1066 register_t orig_tf_eflags;
1071 int crit_count = td->td_critcount;
1074 int have_mplock = 0;
1077 union sysunion args;
1079 KTR_LOG(kernentry_syscall, lp->lwp_proc->p_pid, lp->lwp_tid,
1083 KASSERT(td->td_mpcount == 0,
1084 ("badmpcount syscall2 from %p", (void *)frame->tf_eip));
1086 userenter(td, p); /* lazy raise our priority */
1091 sticks = (int)td->td_sticks;
1092 orig_tf_eflags = frame->tf_eflags;
1095 * Virtual kernel intercept - if a VM context managed by a virtual
1096 * kernel issues a system call the virtual kernel handles it, not us.
1097 * Restore the virtual kernel context and return from its system
1098 * call. The current frame is copied out to the virtual kernel.
1100 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1101 vkernel_trap(lp, frame);
1102 error = EJUSTRETURN;
1107 * Get the system call parameters and account for time
1109 lp->lwp_md.md_regs = frame;
1110 params = (caddr_t)frame->tf_esp + sizeof(int);
1111 code = frame->tf_eax;
1113 if (p->p_sysent->sv_prepsyscall) {
1114 (*p->p_sysent->sv_prepsyscall)(
1115 frame, (int *)(&args.nosys.sysmsg + 1),
1119 * Need to check if this is a 32 bit or 64 bit syscall.
1120 * fuword is MP aware.
1122 if (code == SYS_syscall) {
1124 * Code is first argument, followed by actual args.
1126 code = fuword(params);
1127 params += sizeof(int);
1128 } else if (code == SYS___syscall) {
1130 * Like syscall, but code is a quad, so as to maintain
1131 * quad alignment for the rest of the arguments.
1133 code = fuword(params);
1134 params += sizeof(quad_t);
1138 code &= p->p_sysent->sv_mask;
1139 if (code >= p->p_sysent->sv_size)
1140 callp = &p->p_sysent->sv_table[0];
1142 callp = &p->p_sysent->sv_table[code];
1144 narg = callp->sy_narg & SYF_ARGMASK;
1147 * copyin is MP aware, but the tracing code is not
1149 if (narg && params) {
1150 error = copyin(params, (caddr_t)(&args.nosys.sysmsg + 1),
1151 narg * sizeof(register_t));
1154 if (KTRPOINT(td, KTR_SYSCALL)) {
1155 MAKEMPSAFE(have_mplock);
1157 ktrsyscall(lp, code, narg,
1158 (void *)(&args.nosys.sysmsg + 1));
1166 if (KTRPOINT(td, KTR_SYSCALL)) {
1167 MAKEMPSAFE(have_mplock);
1168 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1173 * For traditional syscall code edx is left untouched when 32 bit
1174 * results are returned. Since edx is loaded from fds[1] when the
1175 * system call returns we pre-set it here.
1177 args.sysmsg_fds[0] = 0;
1178 args.sysmsg_fds[1] = frame->tf_edx;
1181 * The syscall might manipulate the trap frame. If it does it
1182 * will probably return EJUSTRETURN.
1184 args.sysmsg_frame = frame;
1186 STOPEVENT(p, S_SCE, narg); /* MP aware */
1189 * NOTE: All system calls run MPSAFE now. The system call itself
1190 * is responsible for getting the MP lock.
1192 error = (*callp->sy_call)(&args);
1195 kprintf("system call %d returned %d\n", code, error);
1200 * MP SAFE (we may or may not have the MP lock at this point)
1205 * Reinitialize proc pointer `p' as it may be different
1206 * if this is a child returning from fork syscall.
1209 lp = curthread->td_lwp;
1210 frame->tf_eax = args.sysmsg_fds[0];
1211 frame->tf_edx = args.sysmsg_fds[1];
1212 frame->tf_eflags &= ~PSL_C;
1216 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1217 * int 0x80 is 2 bytes. We saved this in tf_err.
1219 frame->tf_eip -= frame->tf_err;
1224 panic("Unexpected EASYNC return value (for now)");
1227 if (p->p_sysent->sv_errsize) {
1228 if (error >= p->p_sysent->sv_errsize)
1229 error = -1; /* XXX */
1231 error = p->p_sysent->sv_errtbl[error];
1233 frame->tf_eax = error;
1234 frame->tf_eflags |= PSL_C;
1239 * Traced syscall. trapsignal() is not MP aware.
1241 if ((orig_tf_eflags & PSL_T) /*&& !(orig_tf_eflags & PSL_VM)*/) {
1242 MAKEMPSAFE(have_mplock);
1243 frame->tf_eflags &= ~PSL_T;
1244 trapsignal(lp, SIGTRAP, TRAP_TRACE);
1248 * Handle reschedule and other end-of-syscall issues
1250 userret(lp, frame, sticks);
1253 if (KTRPOINT(td, KTR_SYSRET)) {
1254 MAKEMPSAFE(have_mplock);
1255 ktrsysret(lp, code, error, args.sysmsg_result);
1260 * This works because errno is findable through the
1261 * register set. If we ever support an emulation where this
1262 * is not the case, this code will need to be revisited.
1264 STOPEVENT(p, S_SCX, code);
1269 * Release the MP lock if we had to get it
1271 KASSERT(td->td_mpcount == have_mplock,
1272 ("badmpcount syscall2/end from %p", (void *)frame->tf_eip));
1276 KTR_LOG(kernentry_syscall_ret, lp->lwp_proc->p_pid, lp->lwp_tid, error);
1278 KASSERT(crit_count == td->td_critcount,
1279 ("syscall: critical section count mismatch! %d/%d",
1280 crit_count, td->td_pri));
1285 * NOTE: mplock not held at any point
1288 fork_return(struct lwp *lp, struct trapframe *frame)
1290 frame->tf_eax = 0; /* Child returns zero */
1291 frame->tf_eflags &= ~PSL_C; /* success */
1294 generic_lwp_return(lp, frame);
1295 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1299 * Simplified back end of syscall(), used when returning from fork()
1300 * directly into user mode.
1302 * This code will return back into the fork trampoline code which then
1305 * NOTE: The mplock is not held at any point.
1308 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1310 struct proc *p = lp->lwp_proc;
1313 * Newly forked processes are given a kernel priority. We have to
1314 * adjust the priority to a normal user priority and fake entry
1315 * into the kernel (call userenter()) to install a passive release
1316 * function just in case userret() decides to stop the process. This
1317 * can occur when ^Z races a fork. If we do not install the passive
1318 * release function the current process designation will not be
1319 * released when the thread goes to sleep.
1321 lwkt_setpri_self(TDPRI_USER_NORM);
1322 userenter(lp->lwp_thread, p);
1323 userret(lp, frame, 0);
1325 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1326 ktrsysret(lp, SYS_fork, 0, 0);
1328 p->p_flag |= P_PASSIVE_ACQ;
1330 p->p_flag &= ~P_PASSIVE_ACQ;
1334 * doreti has turned into this. The frame is directly on the stack. We
1335 * pull everything else we need (fpu and tls context) from the current
1338 * Note on fpu interactions: In a virtual kernel, the fpu context for
1339 * an emulated user mode process is not shared with the virtual kernel's
1340 * fpu context, so we only have to 'stack' fpu contexts within the virtual
1341 * kernel itself, and not even then since the signal() contexts that we care
1342 * about save and restore the FPU state (I think anyhow).
1344 * vmspace_ctl() returns an error only if it had problems instaling the
1345 * context we supplied or problems copying data to/from our VM space.
1348 go_user(struct intrframe *frame)
1350 struct trapframe *tf = (void *)&frame->if_gs;
1354 * Interrupts may be disabled on entry, make sure all signals
1355 * can be received before beginning our loop.
1360 * Switch to the current simulated user process, then call
1361 * user_trap() when we break out of it (usually due to a signal).
1365 * Tell the real kernel whether it is ok to use the FP
1368 * The critical section is required to prevent an interrupt
1369 * from causing a preemptive task switch and changing
1373 if (mdcpu->gd_npxthread == curthread) {
1374 tf->tf_xflags &= ~PGEX_FPFAULT;
1376 tf->tf_xflags |= PGEX_FPFAULT;
1380 * Run emulated user process context. This call interlocks
1381 * with new mailbox signals.
1383 * Set PGEX_U unconditionally, indicating a user frame (the
1384 * bit is normally set only by T_PAGEFLT).
1386 r = vmspace_ctl(&curproc->p_vmspace->vm_pmap, VMSPACE_CTL_RUN,
1387 tf, &curthread->td_savevext);
1389 frame->if_xflags |= PGEX_U;
1391 kprintf("GO USER %d trap %d EVA %08x EIP %08x ESP %08x XFLAGS %02x/%02x\n",
1392 r, tf->tf_trapno, tf->tf_err, tf->tf_eip, tf->tf_esp,
1393 tf->tf_xflags, frame->if_xflags);
1397 panic("vmspace_ctl failed error %d", errno);
1399 if (tf->tf_trapno) {
1403 if (mycpu->gd_reqflags & RQF_AST_MASK) {
1404 tf->tf_trapno = T_ASTFLT;
1412 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1413 * fault (which is then passed back to the virtual kernel) if an attempt is
1414 * made to use the FP unit.
1416 * XXX this is a fairly big hack.
1419 set_vkernel_fp(struct trapframe *frame)
1421 struct thread *td = curthread;
1423 if (frame->tf_xflags & PGEX_FPFAULT) {
1424 td->td_pcb->pcb_flags |= FP_VIRTFP;
1425 if (mdcpu->gd_npxthread == td)
1428 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1433 * Called from vkernel_trap() to fixup the vkernel's syscall
1434 * frame for vmspace_ctl() return.
1437 cpu_vkernel_trap(struct trapframe *frame, int error)
1439 frame->tf_eax = error;
1441 frame->tf_eflags |= PSL_C;
1443 frame->tf_eflags &= ~PSL_C;