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/pc32/i386/trap.c,v 1.115 2008/09/09 04:06:17 dillon Exp $
43 * 386 Trap and System call handling
51 #include "opt_ktrace.h"
52 #include "opt_clock.h"
55 #include <sys/param.h>
56 #include <sys/systm.h>
58 #include <sys/pioctl.h>
59 #include <sys/kernel.h>
60 #include <sys/resourcevar.h>
61 #include <sys/signalvar.h>
62 #include <sys/signal2.h>
63 #include <sys/syscall.h>
64 #include <sys/sysctl.h>
65 #include <sys/sysent.h>
67 #include <sys/vmmeter.h>
68 #include <sys/malloc.h>
70 #include <sys/ktrace.h>
73 #include <sys/upcall.h>
74 #include <sys/vkernel.h>
75 #include <sys/sysproto.h>
76 #include <sys/sysunion.h>
79 #include <vm/vm_param.h>
82 #include <vm/vm_kern.h>
83 #include <vm/vm_map.h>
84 #include <vm/vm_page.h>
85 #include <vm/vm_extern.h>
87 #include <machine/cpu.h>
88 #include <machine/md_var.h>
89 #include <machine/pcb.h>
90 #include <machine/smp.h>
91 #include <machine/tss.h>
92 #include <machine/specialreg.h>
93 #include <machine/globaldata.h>
95 #include <machine_base/isa/intr_machdep.h>
98 #include <sys/syslog.h>
99 #include <machine/clock.h>
102 #include <machine/vm86.h>
105 #include <sys/msgport2.h>
106 #include <sys/thread2.h>
110 #define MAKEMPSAFE(have_mplock) \
111 if (have_mplock == 0) { \
118 #define MAKEMPSAFE(have_mplock)
122 int (*pmath_emulate) (struct trapframe *);
124 extern void trap (struct trapframe *frame);
125 extern int trapwrite (unsigned addr);
126 extern void syscall2 (struct trapframe *frame);
128 static int trap_pfault (struct trapframe *, int, vm_offset_t);
129 static void trap_fatal (struct trapframe *, vm_offset_t);
130 void dblfault_handler (void);
132 extern inthand_t IDTVEC(syscall);
134 #define MAX_TRAP_MSG 28
135 static char *trap_msg[] = {
137 "privileged instruction fault", /* 1 T_PRIVINFLT */
139 "breakpoint instruction fault", /* 3 T_BPTFLT */
142 "arithmetic trap", /* 6 T_ARITHTRAP */
143 "system forced exception", /* 7 T_ASTFLT */
145 "general protection fault", /* 9 T_PROTFLT */
146 "trace trap", /* 10 T_TRCTRAP */
148 "page fault", /* 12 T_PAGEFLT */
150 "alignment fault", /* 14 T_ALIGNFLT */
154 "integer divide fault", /* 18 T_DIVIDE */
155 "non-maskable interrupt trap", /* 19 T_NMI */
156 "overflow trap", /* 20 T_OFLOW */
157 "FPU bounds check fault", /* 21 T_BOUND */
158 "FPU device not available", /* 22 T_DNA */
159 "double fault", /* 23 T_DOUBLEFLT */
160 "FPU operand fetch fault", /* 24 T_FPOPFLT */
161 "invalid TSS fault", /* 25 T_TSSFLT */
162 "segment not present fault", /* 26 T_SEGNPFLT */
163 "stack fault", /* 27 T_STKFLT */
164 "machine check trap", /* 28 T_MCHK */
167 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
168 extern int has_f00f_bug;
172 static int ddb_on_nmi = 1;
173 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
174 &ddb_on_nmi, 0, "Go to DDB on NMI");
176 static int panic_on_nmi = 1;
177 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
178 &panic_on_nmi, 0, "Panic on NMI");
179 static int fast_release;
180 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
181 &fast_release, 0, "Passive Release was optimal");
182 static int slow_release;
183 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
184 &slow_release, 0, "Passive Release was nonoptimal");
186 static int syscall_mpsafe = 1;
187 SYSCTL_INT(_kern, OID_AUTO, syscall_mpsafe, CTLFLAG_RW,
188 &syscall_mpsafe, 0, "Allow MPSAFE marked syscalls to run without BGL");
189 TUNABLE_INT("kern.syscall_mpsafe", &syscall_mpsafe);
190 static int trap_mpsafe = 1;
191 SYSCTL_INT(_kern, OID_AUTO, trap_mpsafe, CTLFLAG_RW,
192 &trap_mpsafe, 0, "Allow traps to mostly run without the BGL");
193 TUNABLE_INT("kern.trap_mpsafe", &trap_mpsafe);
196 MALLOC_DEFINE(M_SYSMSG, "sysmsg", "sysmsg structure");
197 extern int max_sysmsg;
200 * Passively intercepts the thread switch function to increase the thread
201 * priority from a user priority to a kernel priority, reducing
202 * syscall and trap overhead for the case where no switch occurs.
204 * Synchronizes td_ucred with p_ucred. This is used by system calls,
205 * signal handling, faults, AST traps, and anything else that enters the
206 * kernel from userland and provides the kernel with a stable read-only
207 * copy of the process ucred.
210 userenter(struct thread *curtd, struct proc *curp)
215 curtd->td_release = lwkt_passive_release;
217 if (curtd->td_ucred != curp->p_ucred) {
218 ncred = crhold(curp->p_ucred);
219 ocred = curtd->td_ucred;
220 curtd->td_ucred = ncred;
228 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
229 * must be completed before we can return to or try to return to userland.
231 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
232 * arithmatic on the delta calculation so the absolute tick values are
233 * truncated to an integer.
236 userret(struct lwp *lp, struct trapframe *frame, int sticks)
238 struct proc *p = lp->lwp_proc;
242 * Charge system time if profiling. Note: times are in microseconds.
243 * This may do a copyout and block, so do it first even though it
244 * means some system time will be charged as user time.
246 if (p->p_flag & P_PROFIL) {
247 addupc_task(p, frame->tf_eip,
248 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
253 * If the jungle wants us dead, so be it.
255 if (lp->lwp_flag & LWP_WEXIT) {
258 rel_mplock(); /* NOT REACHED */
262 * Block here if we are in a stopped state.
264 if (p->p_stat == SSTOP) {
272 * Post any pending upcalls. If running a virtual kernel be sure
273 * to restore the virtual kernel's vmspace before posting the upcall.
275 if (p->p_flag & P_UPCALLPEND) {
276 p->p_flag &= ~P_UPCALLPEND;
284 * Post any pending signals. If running a virtual kernel be sure
285 * to restore the virtual kernel's vmspace before posting the signal.
287 if ((sig = CURSIG_TRACE(lp)) != 0) {
295 * block here if we are swapped out, but still process signals
296 * (such as SIGKILL). proc0 (the swapin scheduler) is already
297 * aware of our situation, we do not have to wake it up.
299 if (p->p_flag & P_SWAPPEDOUT) {
301 p->p_flag |= P_SWAPWAIT;
303 if (p->p_flag & P_SWAPWAIT)
304 tsleep(p, PCATCH, "SWOUT", 0);
305 p->p_flag &= ~P_SWAPWAIT;
311 * Make sure postsig() handled request to restore old signal mask after
312 * running signal handler.
314 KKASSERT((lp->lwp_flag & LWP_OLDMASK) == 0);
318 * Cleanup from userenter and any passive release that might have occured.
319 * We must reclaim the current-process designation before we can return
320 * to usermode. We also handle both LWKT and USER reschedule requests.
323 userexit(struct lwp *lp)
325 struct thread *td = lp->lwp_thread;
326 /* globaldata_t gd = td->td_gd; */
329 * Handle stop requests at kernel priority. Any requests queued
330 * after this loop will generate another AST.
332 while (lp->lwp_proc->p_stat == SSTOP) {
339 * Reduce our priority in preparation for a return to userland. If
340 * our passive release function was still in place, our priority was
341 * never raised and does not need to be reduced.
343 lwkt_passive_recover(td);
346 * Become the current user scheduled process if we aren't already,
347 * and deal with reschedule requests and other factors.
349 lp->lwp_proc->p_usched->acquire_curproc(lp);
350 /* WARNING: we may have migrated cpu's */
351 /* gd = td->td_gd; */
354 #if !defined(KTR_KERNENTRY)
355 #define KTR_KERNENTRY KTR_ALL
357 KTR_INFO_MASTER(kernentry);
358 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, "pid=%d, tid=%d, trapno=%d, eva=%p",
359 sizeof(int) + sizeof(int) + sizeof(int) + sizeof(vm_offset_t));
360 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "pid=%d, tid=%d",
361 sizeof(int) + sizeof(int));
362 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "pid=%d, tid=%d, call=%d",
363 sizeof(int) + sizeof(int) + sizeof(int));
364 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "pid=%d, tid=%d, err=%d",
365 sizeof(int) + sizeof(int) + sizeof(int));
366 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "pid=%d, tid=%d",
367 sizeof(int) + sizeof(int));
370 * Exception, fault, and trap interface to the kernel.
371 * This common code is called from assembly language IDT gate entry
372 * routines that prepare a suitable stack frame, and restore this
373 * frame after the exception has been processed.
375 * This function is also called from doreti in an interlock to handle ASTs.
376 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
378 * NOTE! We have to retrieve the fault address prior to obtaining the
379 * MP lock because get_mplock() may switch out. YYY cr2 really ought
380 * to be retrieved by the assembly code, not here.
382 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
383 * if an attempt is made to switch from a fast interrupt or IPI. This is
384 * necessary to properly take fatal kernel traps on SMP machines if
385 * get_mplock() has to block.
389 trap(struct trapframe *frame)
391 struct globaldata *gd = mycpu;
392 struct thread *td = gd->gd_curthread;
393 struct lwp *lp = td->td_lwp;
396 int i = 0, ucode = 0, type, code;
401 int crit_count = td->td_pri & ~TDPRI_MASK;
408 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
409 ++gd->gd_trap_nesting_level;
410 MAKEMPSAFE(have_mplock);
411 trap_fatal(frame, eva);
412 --gd->gd_trap_nesting_level;
418 ++gd->gd_trap_nesting_level;
419 if (frame->tf_trapno == T_PAGEFLT) {
421 * For some Cyrix CPUs, %cr2 is clobbered by interrupts.
422 * This problem is worked around by using an interrupt
423 * gate for the pagefault handler. We are finally ready
424 * to read %cr2 and then must reenable interrupts.
426 * XXX this should be in the switch statement, but the
427 * NO_FOOF_HACK and VM86 goto and ifdefs obfuscate the
428 * flow of control too much for this to be obviously
436 if (trap_mpsafe == 0)
437 MAKEMPSAFE(have_mplock);
440 --gd->gd_trap_nesting_level;
442 if (!(frame->tf_eflags & PSL_I)) {
444 * Buggy application or kernel code has disabled interrupts
445 * and then trapped. Enabling interrupts now is wrong, but
446 * it is better than running with interrupts disabled until
447 * they are accidentally enabled later.
449 type = frame->tf_trapno;
450 if (ISPL(frame->tf_cs)==SEL_UPL || (frame->tf_eflags & PSL_VM)) {
451 MAKEMPSAFE(have_mplock);
453 "pid %ld (%s): trap %d with interrupts disabled\n",
454 (long)curproc->p_pid, curproc->p_comm, type);
455 } else if (type != T_BPTFLT && type != T_TRCTRAP) {
457 * XXX not quite right, since this may be for a
458 * multiple fault in user mode.
460 MAKEMPSAFE(have_mplock);
461 kprintf("kernel trap %d with interrupts disabled\n",
467 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
470 type = frame->tf_trapno;
471 code = frame->tf_err;
474 if (frame->tf_eflags & PSL_VM &&
475 (type == T_PROTFLT || type == T_STKFLT)) {
477 KKASSERT(td->td_mpcount > 0);
479 i = vm86_emulate((struct vm86frame *)frame);
481 KKASSERT(td->td_mpcount > 0);
485 * returns to original process
488 vm86_trap((struct vm86frame *)frame,
491 vm86_trap((struct vm86frame *)frame, 0);
493 KKASSERT(0); /* NOT REACHED */
499 * these traps want either a process context, or
500 * assume a normal userspace trap.
504 trap_fatal(frame, eva);
507 type = T_BPTFLT; /* kernel breakpoint */
510 goto kernel_trap; /* normal kernel trap handling */
513 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
516 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid,
517 frame->tf_trapno, eva);
521 sticks = (int)td->td_sticks;
522 lp->lwp_md.md_regs = frame;
525 case T_PRIVINFLT: /* privileged instruction fault */
530 case T_BPTFLT: /* bpt instruction fault */
531 case T_TRCTRAP: /* trace trap */
532 frame->tf_eflags &= ~PSL_T;
536 case T_ARITHTRAP: /* arithmetic trap */
541 case T_ASTFLT: /* Allow process switch */
542 mycpu->gd_cnt.v_soft++;
543 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
544 atomic_clear_int_nonlocked(&mycpu->gd_reqflags,
546 addupc_task(p, p->p_prof.pr_addr,
552 * The following two traps can happen in
553 * vm86 mode, and, if so, we want to handle
556 case T_PROTFLT: /* general protection fault */
557 case T_STKFLT: /* stack fault */
558 if (frame->tf_eflags & PSL_VM) {
559 i = vm86_emulate((struct vm86frame *)frame);
566 case T_SEGNPFLT: /* segment not present fault */
567 case T_TSSFLT: /* invalid TSS fault */
568 case T_DOUBLEFLT: /* double fault */
570 ucode = code + BUS_SEGM_FAULT ;
574 case T_PAGEFLT: /* page fault */
575 MAKEMPSAFE(have_mplock);
576 i = trap_pfault(frame, TRUE, eva);
579 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
589 case T_DIVIDE: /* integer divide fault */
596 MAKEMPSAFE(have_mplock);
598 goto handle_powerfail;
599 #else /* !POWERFAIL_NMI */
600 /* machine/parity/power fail/"kitchen sink" faults */
601 if (isa_nmi(code) == 0) {
604 * NMI can be hooked up to a pushbutton
608 kprintf ("NMI ... going to debugger\n");
609 kdb_trap (type, 0, frame);
613 } else if (panic_on_nmi)
614 panic("NMI indicates hardware failure");
616 #endif /* POWERFAIL_NMI */
617 #endif /* NISA > 0 */
619 case T_OFLOW: /* integer overflow fault */
624 case T_BOUND: /* bounds check fault */
631 * Virtual kernel intercept - pass the DNA exception
632 * to the virtual kernel if it asked to handle it.
633 * This occurs when the virtual kernel is holding
634 * onto the FP context for a different emulated
635 * process then the one currently running.
637 * We must still call npxdna() since we may have
638 * saved FP state that the virtual kernel needs
639 * to hand over to a different emulated process.
641 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
642 (td->td_pcb->pcb_flags & FP_VIRTFP)
650 * The kernel may have switched out the FP unit's
651 * state, causing the user process to take a fault
652 * when it tries to use the FP unit. Restore the
658 if (!pmath_emulate) {
660 ucode = FPE_FPU_NP_TRAP;
663 i = (*pmath_emulate)(frame);
665 if (!(frame->tf_eflags & PSL_T))
667 frame->tf_eflags &= ~PSL_T;
670 /* else ucode = emulator_only_knows() XXX */
673 case T_FPOPFLT: /* FPU operand fetch fault */
678 case T_XMMFLT: /* SIMD floating-point exception */
688 case T_PAGEFLT: /* page fault */
689 MAKEMPSAFE(have_mplock);
690 trap_pfault(frame, FALSE, eva);
696 * The kernel may be using npx for copying or other
704 case T_PROTFLT: /* general protection fault */
705 case T_SEGNPFLT: /* segment not present fault */
707 * Invalid segment selectors and out of bounds
708 * %eip's and %esp's can be set up in user mode.
709 * This causes a fault in kernel mode when the
710 * kernel tries to return to user mode. We want
711 * to get this fault so that we can fix the
712 * problem here and not have to check all the
713 * selectors and pointers when the user changes
716 #define MAYBE_DORETI_FAULT(where, whereto) \
718 if (frame->tf_eip == (int)where) { \
719 frame->tf_eip = (int)whereto; \
723 if (mycpu->gd_intr_nesting_level == 0) {
725 * Invalid %fs's and %gs's can be created using
726 * procfs or PT_SETREGS or by invalidating the
727 * underlying LDT entry. This causes a fault
728 * in kernel mode when the kernel attempts to
729 * switch contexts. Lose the bad context
730 * (XXX) so that we can continue, and generate
733 MAYBE_DORETI_FAULT(doreti_iret,
735 MAYBE_DORETI_FAULT(doreti_popl_ds,
736 doreti_popl_ds_fault);
737 MAYBE_DORETI_FAULT(doreti_popl_es,
738 doreti_popl_es_fault);
739 MAYBE_DORETI_FAULT(doreti_popl_fs,
740 doreti_popl_fs_fault);
741 MAYBE_DORETI_FAULT(doreti_popl_gs,
742 doreti_popl_gs_fault);
743 if (td->td_pcb->pcb_onfault) {
745 (register_t)td->td_pcb->pcb_onfault;
753 * PSL_NT can be set in user mode and isn't cleared
754 * automatically when the kernel is entered. This
755 * causes a TSS fault when the kernel attempts to
756 * `iret' because the TSS link is uninitialized. We
757 * want to get this fault so that we can fix the
758 * problem here and not every time the kernel is
761 if (frame->tf_eflags & PSL_NT) {
762 frame->tf_eflags &= ~PSL_NT;
767 case T_TRCTRAP: /* trace trap */
768 if (frame->tf_eip == (int)IDTVEC(syscall)) {
770 * We've just entered system mode via the
771 * syscall lcall. Continue single stepping
772 * silently until the syscall handler has
777 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
779 * The syscall handler has now saved the
780 * flags. Stop single stepping it.
782 frame->tf_eflags &= ~PSL_T;
786 * Ignore debug register trace traps due to
787 * accesses in the user's address space, which
788 * can happen under several conditions such as
789 * if a user sets a watchpoint on a buffer and
790 * then passes that buffer to a system call.
791 * We still want to get TRCTRAPS for addresses
792 * in kernel space because that is useful when
793 * debugging the kernel.
795 if (user_dbreg_trap()) {
797 * Reset breakpoint bits because the
800 load_dr6(rdr6() & 0xfffffff0);
804 * Fall through (TRCTRAP kernel mode, kernel address)
808 * If DDB is enabled, let it handle the debugger trap.
809 * Otherwise, debugger traps "can't happen".
812 MAKEMPSAFE(have_mplock);
813 if (kdb_trap (type, 0, frame))
820 MAKEMPSAFE(have_mplock);
823 # define TIMER_FREQ 1193182
827 static unsigned lastalert = 0;
829 if(time_second - lastalert > 10)
831 log(LOG_WARNING, "NMI: power fail\n");
832 sysbeep(TIMER_FREQ/880, hz);
833 lastalert = time_second;
838 #else /* !POWERFAIL_NMI */
839 /* machine/parity/power fail/"kitchen sink" faults */
840 if (isa_nmi(code) == 0) {
843 * NMI can be hooked up to a pushbutton
847 kprintf ("NMI ... going to debugger\n");
848 kdb_trap (type, 0, frame);
852 } else if (panic_on_nmi == 0)
855 #endif /* POWERFAIL_NMI */
856 #endif /* NISA > 0 */
859 MAKEMPSAFE(have_mplock);
860 trap_fatal(frame, eva);
865 * Virtual kernel intercept - if the fault is directly related to a
866 * VM context managed by a virtual kernel then let the virtual kernel
869 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
870 vkernel_trap(lp, frame);
875 * Translate fault for emulators (e.g. Linux)
877 if (*p->p_sysent->sv_transtrap)
878 i = (*p->p_sysent->sv_transtrap)(i, type);
880 MAKEMPSAFE(have_mplock);
881 trapsignal(lp, i, ucode);
884 if (type <= MAX_TRAP_MSG) {
885 uprintf("fatal process exception: %s",
887 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
888 uprintf(", fault VA = 0x%lx", (u_long)eva);
895 if (ISPL(frame->tf_cs) == SEL_UPL)
896 KASSERT(td->td_mpcount == have_mplock, ("badmpcount trap/end from %p", (void *)frame->tf_eip));
898 userret(lp, frame, sticks);
905 if (p != NULL && lp != NULL)
906 KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid);
908 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
909 ("syscall: critical section count mismatch! %d/%d",
910 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
915 trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
918 struct vmspace *vm = NULL;
922 thread_t td = curthread;
923 struct lwp *lp = td->td_lwp;
925 va = trunc_page(eva);
926 if (va >= KERNBASE) {
928 * Don't allow user-mode faults in kernel address space.
929 * An exception: if the faulting address is the invalid
930 * instruction entry in the IDT, then the Intel Pentium
931 * F00F bug workaround was triggered, and we need to
932 * treat it is as an illegal instruction, and not a page
935 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
936 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) {
937 frame->tf_trapno = T_PRIVINFLT;
947 * This is a fault on non-kernel virtual memory.
948 * vm is initialized above to NULL. If curproc is NULL
949 * or curproc->p_vmspace is NULL the fault is fatal.
952 vm = lp->lwp_vmspace;
960 if (frame->tf_err & PGEX_W)
961 ftype = VM_PROT_WRITE;
963 ftype = VM_PROT_READ;
965 if (map != &kernel_map) {
967 * Keep swapout from messing with us during this
973 * Grow the stack if necessary
975 /* grow_stack returns false only if va falls into
976 * a growable stack region and the stack growth
977 * fails. It returns true if va was not within
978 * a growable stack region, or if the stack
981 if (!grow_stack(lp->lwp_proc, va)) {
987 /* Fault in the user page: */
988 rv = vm_fault(map, va, ftype,
989 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
995 * Don't have to worry about process locking or stacks
998 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
1001 if (rv == KERN_SUCCESS)
1005 if (td->td_gd->gd_intr_nesting_level == 0 &&
1006 td->td_pcb->pcb_onfault) {
1007 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
1010 trap_fatal(frame, eva);
1014 /* kludge to pass faulting virtual address to sendsig */
1015 frame->tf_xflags = frame->tf_err;
1016 frame->tf_err = eva;
1018 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
1022 trap_fatal(struct trapframe *frame, vm_offset_t eva)
1024 int code, type, ss, esp;
1025 struct soft_segment_descriptor softseg;
1027 code = frame->tf_err;
1028 type = frame->tf_trapno;
1029 sdtossd(&gdt[mycpu->gd_cpuid * NGDT + IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
1031 if (type <= MAX_TRAP_MSG)
1032 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
1033 type, trap_msg[type],
1034 frame->tf_eflags & PSL_VM ? "vm86" :
1035 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
1037 /* three separate prints in case of a trap on an unmapped page */
1038 kprintf("mp_lock = %08x; ", mp_lock);
1039 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
1040 kprintf("lapic.id = %08x\n", lapic.id);
1042 if (type == T_PAGEFLT) {
1043 kprintf("fault virtual address = %p\n", (void *)eva);
1044 kprintf("fault code = %s %s, %s\n",
1045 code & PGEX_U ? "user" : "supervisor",
1046 code & PGEX_W ? "write" : "read",
1047 code & PGEX_P ? "protection violation" : "page not present");
1049 kprintf("instruction pointer = 0x%x:0x%x\n",
1050 frame->tf_cs & 0xffff, frame->tf_eip);
1051 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
1052 ss = frame->tf_ss & 0xffff;
1053 esp = frame->tf_esp;
1055 ss = GSEL(GDATA_SEL, SEL_KPL);
1056 esp = (int)&frame->tf_esp;
1058 kprintf("stack pointer = 0x%x:0x%x\n", ss, esp);
1059 kprintf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
1060 kprintf("code segment = base 0x%x, limit 0x%x, type 0x%x\n",
1061 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
1062 kprintf(" = DPL %d, pres %d, def32 %d, gran %d\n",
1063 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
1065 kprintf("processor eflags = ");
1066 if (frame->tf_eflags & PSL_T)
1067 kprintf("trace trap, ");
1068 if (frame->tf_eflags & PSL_I)
1069 kprintf("interrupt enabled, ");
1070 if (frame->tf_eflags & PSL_NT)
1071 kprintf("nested task, ");
1072 if (frame->tf_eflags & PSL_RF)
1073 kprintf("resume, ");
1074 if (frame->tf_eflags & PSL_VM)
1076 kprintf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
1077 kprintf("current process = ");
1079 kprintf("%lu (%s)\n",
1080 (u_long)curproc->p_pid, curproc->p_comm ?
1081 curproc->p_comm : "");
1085 kprintf("current thread = pri %d ", curthread->td_pri);
1086 if (curthread->td_pri >= TDPRI_CRIT)
1092 * we probably SHOULD have stopped the other CPUs before now!
1093 * another CPU COULD have been touching cpl at this moment...
1095 kprintf(" <- SMP: XXX");
1104 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1107 kprintf("trap number = %d\n", type);
1108 if (type <= MAX_TRAP_MSG)
1109 panic("%s", trap_msg[type]);
1111 panic("unknown/reserved trap");
1115 * Double fault handler. Called when a fault occurs while writing
1116 * a frame for a trap/exception onto the stack. This usually occurs
1117 * when the stack overflows (such is the case with infinite recursion,
1120 * XXX Note that the current PTD gets replaced by IdlePTD when the
1121 * task switch occurs. This means that the stack that was active at
1122 * the time of the double fault is not available at <kstack> unless
1123 * the machine was idle when the double fault occurred. The downside
1124 * of this is that "trace <ebp>" in ddb won't work.
1127 dblfault_handler(void)
1129 struct mdglobaldata *gd = mdcpu;
1131 kprintf("\nFatal double fault:\n");
1132 kprintf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1133 kprintf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1134 kprintf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
1136 /* three separate prints in case of a trap on an unmapped page */
1137 kprintf("mp_lock = %08x; ", mp_lock);
1138 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
1139 kprintf("lapic.id = %08x\n", lapic.id);
1141 panic("double fault");
1145 * Compensate for 386 brain damage (missing URKR).
1146 * This is a little simpler than the pagefault handler in trap() because
1147 * it the page tables have already been faulted in and high addresses
1148 * are thrown out early for other reasons.
1151 trapwrite(unsigned addr)
1158 va = trunc_page((vm_offset_t)addr);
1160 * XXX - MAX is END. Changed > to >= for temp. fix.
1162 if (va >= VM_MAX_USER_ADDRESS)
1165 lp = curthread->td_lwp;
1166 vm = lp->lwp_vmspace;
1168 PHOLD(lp->lwp_proc);
1170 if (!grow_stack(lp->lwp_proc, va)) {
1171 PRELE(lp->lwp_proc);
1176 * fault the data page
1178 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY);
1180 PRELE(lp->lwp_proc);
1182 if (rv != KERN_SUCCESS)
1189 * syscall2 - MP aware system call request C handler
1191 * A system call is essentially treated as a trap. The MP lock is not
1192 * held on entry or return. We are responsible for handling ASTs
1193 * (e.g. a task switch) prior to return.
1198 syscall2(struct trapframe *frame)
1200 struct thread *td = curthread;
1201 struct proc *p = td->td_proc;
1202 struct lwp *lp = td->td_lwp;
1204 struct sysent *callp;
1205 register_t orig_tf_eflags;
1210 int crit_count = td->td_pri & ~TDPRI_MASK;
1213 int have_mplock = 0;
1216 union sysunion args;
1219 if (ISPL(frame->tf_cs) != SEL_UPL) {
1226 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid,
1230 KASSERT(td->td_mpcount == 0, ("badmpcount syscall2 from %p", (void *)frame->tf_eip));
1231 if (syscall_mpsafe == 0)
1232 MAKEMPSAFE(have_mplock);
1234 userenter(td, p); /* lazy raise our priority */
1239 sticks = (int)td->td_sticks;
1240 orig_tf_eflags = frame->tf_eflags;
1243 * Virtual kernel intercept - if a VM context managed by a virtual
1244 * kernel issues a system call the virtual kernel handles it, not us.
1245 * Restore the virtual kernel context and return from its system
1246 * call. The current frame is copied out to the virtual kernel.
1248 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1249 error = vkernel_trap(lp, frame);
1250 frame->tf_eax = error;
1252 frame->tf_eflags |= PSL_C;
1253 error = EJUSTRETURN;
1258 * Get the system call parameters and account for time
1260 lp->lwp_md.md_regs = frame;
1261 params = (caddr_t)frame->tf_esp + sizeof(int);
1262 code = frame->tf_eax;
1264 if (p->p_sysent->sv_prepsyscall) {
1265 (*p->p_sysent->sv_prepsyscall)(
1266 frame, (int *)(&args.nosys.sysmsg + 1),
1270 * Need to check if this is a 32 bit or 64 bit syscall.
1271 * fuword is MP aware.
1273 if (code == SYS_syscall) {
1275 * Code is first argument, followed by actual args.
1277 code = fuword(params);
1278 params += sizeof(int);
1279 } else if (code == SYS___syscall) {
1281 * Like syscall, but code is a quad, so as to maintain
1282 * quad alignment for the rest of the arguments.
1284 code = fuword(params);
1285 params += sizeof(quad_t);
1289 code &= p->p_sysent->sv_mask;
1290 if (code >= p->p_sysent->sv_size)
1291 callp = &p->p_sysent->sv_table[0];
1293 callp = &p->p_sysent->sv_table[code];
1295 narg = callp->sy_narg & SYF_ARGMASK;
1298 * copyin is MP aware, but the tracing code is not
1300 if (narg && params) {
1301 error = copyin(params, (caddr_t)(&args.nosys.sysmsg + 1),
1302 narg * sizeof(register_t));
1305 if (KTRPOINT(td, KTR_SYSCALL)) {
1306 MAKEMPSAFE(have_mplock);
1308 ktrsyscall(lp, code, narg,
1309 (void *)(&args.nosys.sysmsg + 1));
1317 if (KTRPOINT(td, KTR_SYSCALL)) {
1318 MAKEMPSAFE(have_mplock);
1319 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1324 * For traditional syscall code edx is left untouched when 32 bit
1325 * results are returned. Since edx is loaded from fds[1] when the
1326 * system call returns we pre-set it here.
1328 args.sysmsg_fds[0] = 0;
1329 args.sysmsg_fds[1] = frame->tf_edx;
1332 * The syscall might manipulate the trap frame. If it does it
1333 * will probably return EJUSTRETURN.
1335 args.sysmsg_frame = frame;
1337 STOPEVENT(p, S_SCE, narg); /* MP aware */
1340 * NOTE: All system calls run MPSAFE now. The system call itself
1341 * is responsible for getting the MP lock.
1343 error = (*callp->sy_call)(&args);
1347 * MP SAFE (we may or may not have the MP lock at this point)
1352 * Reinitialize proc pointer `p' as it may be different
1353 * if this is a child returning from fork syscall.
1356 lp = curthread->td_lwp;
1357 frame->tf_eax = args.sysmsg_fds[0];
1358 frame->tf_edx = args.sysmsg_fds[1];
1359 frame->tf_eflags &= ~PSL_C;
1363 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1364 * int 0x80 is 2 bytes. We saved this in tf_err.
1366 frame->tf_eip -= frame->tf_err;
1371 panic("Unexpected EASYNC return value (for now)");
1374 if (p->p_sysent->sv_errsize) {
1375 if (error >= p->p_sysent->sv_errsize)
1376 error = -1; /* XXX */
1378 error = p->p_sysent->sv_errtbl[error];
1380 frame->tf_eax = error;
1381 frame->tf_eflags |= PSL_C;
1386 * Traced syscall. trapsignal() is not MP aware.
1388 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1389 MAKEMPSAFE(have_mplock);
1390 frame->tf_eflags &= ~PSL_T;
1391 trapsignal(lp, SIGTRAP, 0);
1395 * Handle reschedule and other end-of-syscall issues
1397 userret(lp, frame, sticks);
1400 if (KTRPOINT(td, KTR_SYSRET)) {
1401 MAKEMPSAFE(have_mplock);
1402 ktrsysret(lp, code, error, args.sysmsg_result);
1407 * This works because errno is findable through the
1408 * register set. If we ever support an emulation where this
1409 * is not the case, this code will need to be revisited.
1411 STOPEVENT(p, S_SCX, code);
1416 * Release the MP lock if we had to get it
1418 KASSERT(td->td_mpcount == have_mplock,
1419 ("badmpcount syscall2/end from %p", (void *)frame->tf_eip));
1423 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error);
1425 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
1426 ("syscall: critical section count mismatch! %d/%d",
1427 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
1432 fork_return(struct lwp *lp, struct trapframe *frame)
1434 frame->tf_eax = 0; /* Child returns zero */
1435 frame->tf_eflags &= ~PSL_C; /* success */
1438 generic_lwp_return(lp, frame);
1439 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1443 * Simplified back end of syscall(), used when returning from fork()
1444 * directly into user mode. MP lock is held on entry and should be
1445 * released on return. This code will return back into the fork
1446 * trampoline code which then runs doreti.
1449 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1451 struct proc *p = lp->lwp_proc;
1454 * Newly forked processes are given a kernel priority. We have to
1455 * adjust the priority to a normal user priority and fake entry
1456 * into the kernel (call userenter()) to install a passive release
1457 * function just in case userret() decides to stop the process. This
1458 * can occur when ^Z races a fork. If we do not install the passive
1459 * release function the current process designation will not be
1460 * released when the thread goes to sleep.
1462 lwkt_setpri_self(TDPRI_USER_NORM);
1463 userenter(lp->lwp_thread, p);
1464 userret(lp, frame, 0);
1466 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1467 ktrsysret(lp, SYS_fork, 0, 0);
1469 p->p_flag |= P_PASSIVE_ACQ;
1471 p->p_flag &= ~P_PASSIVE_ACQ;
1473 KKASSERT(lp->lwp_thread->td_mpcount == 1);
1479 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1480 * fault (which is then passed back to the virtual kernel) if an attempt is
1481 * made to use the FP unit.
1483 * XXX this is a fairly big hack.
1486 set_vkernel_fp(struct trapframe *frame)
1488 struct thread *td = curthread;
1490 if (frame->tf_xflags & PGEX_FPFAULT) {
1491 td->td_pcb->pcb_flags |= FP_VIRTFP;
1492 if (mdcpu->gd_npxthread == td)
1495 td->td_pcb->pcb_flags &= ~FP_VIRTFP;