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>
106 #include <sys/msgport2.h>
107 #include <sys/thread2.h>
108 #include <sys/mplock2.h>
112 #define MAKEMPSAFE(have_mplock) \
113 if (have_mplock == 0) { \
120 #define MAKEMPSAFE(have_mplock)
124 int (*pmath_emulate) (struct trapframe *);
126 extern void trap (struct trapframe *frame);
127 extern void syscall2 (struct trapframe *frame);
129 static int trap_pfault (struct trapframe *, int, vm_offset_t);
130 static void trap_fatal (struct trapframe *, vm_offset_t);
131 void dblfault_handler (void);
133 extern inthand_t IDTVEC(syscall);
135 #define MAX_TRAP_MSG 28
136 static char *trap_msg[] = {
138 "privileged instruction fault", /* 1 T_PRIVINFLT */
140 "breakpoint instruction fault", /* 3 T_BPTFLT */
143 "arithmetic trap", /* 6 T_ARITHTRAP */
144 "system forced exception", /* 7 T_ASTFLT */
146 "general protection fault", /* 9 T_PROTFLT */
147 "trace trap", /* 10 T_TRCTRAP */
149 "page fault", /* 12 T_PAGEFLT */
151 "alignment fault", /* 14 T_ALIGNFLT */
155 "integer divide fault", /* 18 T_DIVIDE */
156 "non-maskable interrupt trap", /* 19 T_NMI */
157 "overflow trap", /* 20 T_OFLOW */
158 "FPU bounds check fault", /* 21 T_BOUND */
159 "FPU device not available", /* 22 T_DNA */
160 "double fault", /* 23 T_DOUBLEFLT */
161 "FPU operand fetch fault", /* 24 T_FPOPFLT */
162 "invalid TSS fault", /* 25 T_TSSFLT */
163 "segment not present fault", /* 26 T_SEGNPFLT */
164 "stack fault", /* 27 T_STKFLT */
165 "machine check trap", /* 28 T_MCHK */
168 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
169 extern int has_f00f_bug;
173 static int ddb_on_nmi = 1;
174 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
175 &ddb_on_nmi, 0, "Go to DDB on NMI");
177 static int panic_on_nmi = 1;
178 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
179 &panic_on_nmi, 0, "Panic on NMI");
180 static int fast_release;
181 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
182 &fast_release, 0, "Passive Release was optimal");
183 static int slow_release;
184 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
185 &slow_release, 0, "Passive Release was nonoptimal");
187 static int syscall_mpsafe = 1;
188 SYSCTL_INT(_kern, OID_AUTO, syscall_mpsafe, CTLFLAG_RW,
189 &syscall_mpsafe, 0, "Allow MPSAFE marked syscalls to run without BGL");
190 TUNABLE_INT("kern.syscall_mpsafe", &syscall_mpsafe);
191 static int trap_mpsafe = 1;
192 SYSCTL_INT(_kern, OID_AUTO, trap_mpsafe, CTLFLAG_RW,
193 &trap_mpsafe, 0, "Allow traps to mostly run without the BGL");
194 TUNABLE_INT("kern.trap_mpsafe", &trap_mpsafe);
197 MALLOC_DEFINE(M_SYSMSG, "sysmsg", "sysmsg structure");
198 extern int max_sysmsg;
201 * Passively intercepts the thread switch function to increase the thread
202 * priority from a user priority to a kernel priority, reducing
203 * syscall and trap overhead for the case where no switch occurs.
205 * Synchronizes td_ucred with p_ucred. This is used by system calls,
206 * signal handling, faults, AST traps, and anything else that enters the
207 * kernel from userland and provides the kernel with a stable read-only
208 * copy of the process ucred.
211 userenter(struct thread *curtd, struct proc *curp)
216 curtd->td_release = lwkt_passive_release;
218 if (curtd->td_ucred != curp->p_ucred) {
219 ncred = crhold(curp->p_ucred);
220 ocred = curtd->td_ucred;
221 curtd->td_ucred = ncred;
229 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
230 * must be completed before we can return to or try to return to userland.
232 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
233 * arithmatic on the delta calculation so the absolute tick values are
234 * truncated to an integer.
237 userret(struct lwp *lp, struct trapframe *frame, int sticks)
239 struct proc *p = lp->lwp_proc;
243 if (p->p_userret != NULL) {
250 * Charge system time if profiling. Note: times are in microseconds.
251 * This may do a copyout and block, so do it first even though it
252 * means some system time will be charged as user time.
254 if (p->p_flag & P_PROFIL) {
255 addupc_task(p, frame->tf_eip,
256 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
261 * If the jungle wants us dead, so be it.
263 if (lp->lwp_flag & LWP_WEXIT) {
266 rel_mplock(); /* NOT REACHED */
270 * Block here if we are in a stopped state.
272 if (p->p_stat == SSTOP) {
280 * Post any pending upcalls. If running a virtual kernel be sure
281 * to restore the virtual kernel's vmspace before posting the upcall.
283 if (p->p_flag & P_UPCALLPEND) {
284 p->p_flag &= ~P_UPCALLPEND;
292 * Post any pending signals. If running a virtual kernel be sure
293 * to restore the virtual kernel's vmspace before posting the signal.
295 if ((sig = CURSIG_TRACE(lp)) != 0) {
303 * block here if we are swapped out, but still process signals
304 * (such as SIGKILL). proc0 (the swapin scheduler) is already
305 * aware of our situation, we do not have to wake it up.
307 if (p->p_flag & P_SWAPPEDOUT) {
309 p->p_flag |= P_SWAPWAIT;
311 if (p->p_flag & P_SWAPWAIT)
312 tsleep(p, PCATCH, "SWOUT", 0);
313 p->p_flag &= ~P_SWAPWAIT;
319 * Make sure postsig() handled request to restore old signal mask after
320 * running signal handler.
322 KKASSERT((lp->lwp_flag & LWP_OLDMASK) == 0);
326 * Cleanup from userenter and any passive release that might have occured.
327 * We must reclaim the current-process designation before we can return
328 * to usermode. We also handle both LWKT and USER reschedule requests.
331 userexit(struct lwp *lp)
333 struct thread *td = lp->lwp_thread;
334 /* globaldata_t gd = td->td_gd; */
337 * Handle stop requests at kernel priority. Any requests queued
338 * after this loop will generate another AST.
340 while (lp->lwp_proc->p_stat == SSTOP) {
347 * Reduce our priority in preparation for a return to userland. If
348 * our passive release function was still in place, our priority was
349 * never raised and does not need to be reduced.
351 lwkt_passive_recover(td);
354 * Become the current user scheduled process if we aren't already,
355 * and deal with reschedule requests and other factors.
357 lp->lwp_proc->p_usched->acquire_curproc(lp);
358 /* WARNING: we may have migrated cpu's */
359 /* gd = td->td_gd; */
362 #if !defined(KTR_KERNENTRY)
363 #define KTR_KERNENTRY KTR_ALL
365 KTR_INFO_MASTER(kernentry);
366 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, "pid=%d, tid=%d, trapno=%d, eva=%p",
367 sizeof(int) + sizeof(int) + sizeof(int) + sizeof(vm_offset_t));
368 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "pid=%d, tid=%d",
369 sizeof(int) + sizeof(int));
370 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "pid=%d, tid=%d, call=%d",
371 sizeof(int) + sizeof(int) + sizeof(int));
372 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "pid=%d, tid=%d, err=%d",
373 sizeof(int) + sizeof(int) + sizeof(int));
374 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "pid=%d, tid=%d",
375 sizeof(int) + sizeof(int));
378 * Exception, fault, and trap interface to the kernel.
379 * This common code is called from assembly language IDT gate entry
380 * routines that prepare a suitable stack frame, and restore this
381 * frame after the exception has been processed.
383 * This function is also called from doreti in an interlock to handle ASTs.
384 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
386 * NOTE! We have to retrieve the fault address prior to obtaining the
387 * MP lock because get_mplock() may switch out. YYY cr2 really ought
388 * to be retrieved by the assembly code, not here.
390 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
391 * if an attempt is made to switch from a fast interrupt or IPI. This is
392 * necessary to properly take fatal kernel traps on SMP machines if
393 * get_mplock() has to block.
397 trap(struct trapframe *frame)
399 struct globaldata *gd = mycpu;
400 struct thread *td = gd->gd_curthread;
401 struct lwp *lp = td->td_lwp;
404 int i = 0, ucode = 0, type, code;
409 int crit_count = td->td_pri & ~TDPRI_MASK;
416 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
417 ++gd->gd_trap_nesting_level;
418 MAKEMPSAFE(have_mplock);
419 trap_fatal(frame, eva);
420 --gd->gd_trap_nesting_level;
426 ++gd->gd_trap_nesting_level;
427 if (frame->tf_trapno == T_PAGEFLT) {
429 * For some Cyrix CPUs, %cr2 is clobbered by interrupts.
430 * This problem is worked around by using an interrupt
431 * gate for the pagefault handler. We are finally ready
432 * to read %cr2 and then must reenable interrupts.
434 * XXX this should be in the switch statement, but the
435 * NO_FOOF_HACK and VM86 goto and ifdefs obfuscate the
436 * flow of control too much for this to be obviously
444 if (trap_mpsafe == 0)
445 MAKEMPSAFE(have_mplock);
448 --gd->gd_trap_nesting_level;
450 if (!(frame->tf_eflags & PSL_I)) {
452 * Buggy application or kernel code has disabled interrupts
453 * and then trapped. Enabling interrupts now is wrong, but
454 * it is better than running with interrupts disabled until
455 * they are accidentally enabled later.
457 type = frame->tf_trapno;
458 if (ISPL(frame->tf_cs)==SEL_UPL || (frame->tf_eflags & PSL_VM)) {
459 MAKEMPSAFE(have_mplock);
461 "pid %ld (%s): trap %d with interrupts disabled\n",
462 (long)curproc->p_pid, curproc->p_comm, type);
463 } else if (type != T_BPTFLT && type != T_TRCTRAP) {
465 * XXX not quite right, since this may be for a
466 * multiple fault in user mode.
468 MAKEMPSAFE(have_mplock);
469 kprintf("kernel trap %d with interrupts disabled\n",
475 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
478 type = frame->tf_trapno;
479 code = frame->tf_err;
482 if (frame->tf_eflags & PSL_VM &&
483 (type == T_PROTFLT || type == T_STKFLT)) {
485 KKASSERT(td->td_mpcount > 0);
487 i = vm86_emulate((struct vm86frame *)frame);
489 KKASSERT(td->td_mpcount > 0);
493 * returns to original process
496 vm86_trap((struct vm86frame *)frame,
499 vm86_trap((struct vm86frame *)frame, 0);
501 KKASSERT(0); /* NOT REACHED */
507 * these traps want either a process context, or
508 * assume a normal userspace trap.
512 trap_fatal(frame, eva);
515 type = T_BPTFLT; /* kernel breakpoint */
518 goto kernel_trap; /* normal kernel trap handling */
521 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
524 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid,
525 frame->tf_trapno, eva);
529 sticks = (int)td->td_sticks;
530 lp->lwp_md.md_regs = frame;
533 case T_PRIVINFLT: /* privileged instruction fault */
538 case T_BPTFLT: /* bpt instruction fault */
539 case T_TRCTRAP: /* trace trap */
540 frame->tf_eflags &= ~PSL_T;
545 case T_ARITHTRAP: /* arithmetic trap */
550 case T_ASTFLT: /* Allow process switch */
551 mycpu->gd_cnt.v_soft++;
552 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
553 atomic_clear_int_nonlocked(&mycpu->gd_reqflags,
555 addupc_task(p, p->p_prof.pr_addr,
561 * The following two traps can happen in
562 * vm86 mode, and, if so, we want to handle
565 case T_PROTFLT: /* general protection fault */
566 case T_STKFLT: /* stack fault */
567 if (frame->tf_eflags & PSL_VM) {
568 i = vm86_emulate((struct vm86frame *)frame);
574 ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR;
576 case T_SEGNPFLT: /* segment not present fault */
580 case T_TSSFLT: /* invalid TSS fault */
581 case T_DOUBLEFLT: /* double fault */
586 ucode = code + BUS_SEGM_FAULT ; /* XXX: ???*/
592 case T_PAGEFLT: /* page fault */
593 MAKEMPSAFE(have_mplock);
594 i = trap_pfault(frame, TRUE, eva);
597 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
609 ucode = BUS_ADRERR; /* XXX */
612 case T_DIVIDE: /* integer divide fault */
619 MAKEMPSAFE(have_mplock);
621 goto handle_powerfail;
622 #else /* !POWERFAIL_NMI */
623 /* machine/parity/power fail/"kitchen sink" faults */
624 if (isa_nmi(code) == 0) {
627 * NMI can be hooked up to a pushbutton
631 kprintf ("NMI ... going to debugger\n");
632 kdb_trap (type, 0, frame);
636 } else if (panic_on_nmi)
637 panic("NMI indicates hardware failure");
639 #endif /* POWERFAIL_NMI */
640 #endif /* NISA > 0 */
642 case T_OFLOW: /* integer overflow fault */
647 case T_BOUND: /* bounds check fault */
654 * Virtual kernel intercept - pass the DNA exception
655 * to the virtual kernel if it asked to handle it.
656 * This occurs when the virtual kernel is holding
657 * onto the FP context for a different emulated
658 * process then the one currently running.
660 * We must still call npxdna() since we may have
661 * saved FP state that the virtual kernel needs
662 * to hand over to a different emulated process.
664 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
665 (td->td_pcb->pcb_flags & FP_VIRTFP)
673 * The kernel may have switched out the FP unit's
674 * state, causing the user process to take a fault
675 * when it tries to use the FP unit. Restore the
681 if (!pmath_emulate) {
683 ucode = FPE_FPU_NP_TRAP;
686 i = (*pmath_emulate)(frame);
688 if (!(frame->tf_eflags & PSL_T))
690 frame->tf_eflags &= ~PSL_T;
693 /* else ucode = emulator_only_knows() XXX */
696 case T_FPOPFLT: /* FPU operand fetch fault */
701 case T_XMMFLT: /* SIMD floating-point exception */
711 case T_PAGEFLT: /* page fault */
712 MAKEMPSAFE(have_mplock);
713 trap_pfault(frame, FALSE, eva);
719 * The kernel may be using npx for copying or other
727 case T_PROTFLT: /* general protection fault */
728 case T_SEGNPFLT: /* segment not present fault */
730 * Invalid segment selectors and out of bounds
731 * %eip's and %esp's can be set up in user mode.
732 * This causes a fault in kernel mode when the
733 * kernel tries to return to user mode. We want
734 * to get this fault so that we can fix the
735 * problem here and not have to check all the
736 * selectors and pointers when the user changes
739 #define MAYBE_DORETI_FAULT(where, whereto) \
741 if (frame->tf_eip == (int)where) { \
742 frame->tf_eip = (int)whereto; \
746 if (mycpu->gd_intr_nesting_level == 0) {
748 * Invalid %fs's and %gs's can be created using
749 * procfs or PT_SETREGS or by invalidating the
750 * underlying LDT entry. This causes a fault
751 * in kernel mode when the kernel attempts to
752 * switch contexts. Lose the bad context
753 * (XXX) so that we can continue, and generate
756 MAYBE_DORETI_FAULT(doreti_iret,
758 MAYBE_DORETI_FAULT(doreti_popl_ds,
759 doreti_popl_ds_fault);
760 MAYBE_DORETI_FAULT(doreti_popl_es,
761 doreti_popl_es_fault);
762 MAYBE_DORETI_FAULT(doreti_popl_fs,
763 doreti_popl_fs_fault);
764 MAYBE_DORETI_FAULT(doreti_popl_gs,
765 doreti_popl_gs_fault);
766 if (td->td_pcb->pcb_onfault) {
768 (register_t)td->td_pcb->pcb_onfault;
776 * PSL_NT can be set in user mode and isn't cleared
777 * automatically when the kernel is entered. This
778 * causes a TSS fault when the kernel attempts to
779 * `iret' because the TSS link is uninitialized. We
780 * want to get this fault so that we can fix the
781 * problem here and not every time the kernel is
784 if (frame->tf_eflags & PSL_NT) {
785 frame->tf_eflags &= ~PSL_NT;
790 case T_TRCTRAP: /* trace trap */
791 if (frame->tf_eip == (int)IDTVEC(syscall)) {
793 * We've just entered system mode via the
794 * syscall lcall. Continue single stepping
795 * silently until the syscall handler has
800 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
802 * The syscall handler has now saved the
803 * flags. Stop single stepping it.
805 frame->tf_eflags &= ~PSL_T;
809 * Ignore debug register trace traps due to
810 * accesses in the user's address space, which
811 * can happen under several conditions such as
812 * if a user sets a watchpoint on a buffer and
813 * then passes that buffer to a system call.
814 * We still want to get TRCTRAPS for addresses
815 * in kernel space because that is useful when
816 * debugging the kernel.
818 if (user_dbreg_trap()) {
820 * Reset breakpoint bits because the
823 load_dr6(rdr6() & 0xfffffff0);
827 * Fall through (TRCTRAP kernel mode, kernel address)
831 * If DDB is enabled, let it handle the debugger trap.
832 * Otherwise, debugger traps "can't happen".
836 MAKEMPSAFE(have_mplock);
837 if (kdb_trap (type, 0, frame))
844 MAKEMPSAFE(have_mplock);
847 # define TIMER_FREQ 1193182
851 static unsigned lastalert = 0;
853 if(time_second - lastalert > 10)
855 log(LOG_WARNING, "NMI: power fail\n");
856 sysbeep(TIMER_FREQ/880, hz);
857 lastalert = time_second;
862 #else /* !POWERFAIL_NMI */
863 /* machine/parity/power fail/"kitchen sink" faults */
864 if (isa_nmi(code) == 0) {
867 * NMI can be hooked up to a pushbutton
871 kprintf ("NMI ... going to debugger\n");
872 kdb_trap (type, 0, frame);
876 } else if (panic_on_nmi == 0)
879 #endif /* POWERFAIL_NMI */
880 #endif /* NISA > 0 */
883 MAKEMPSAFE(have_mplock);
884 trap_fatal(frame, eva);
889 * Virtual kernel intercept - if the fault is directly related to a
890 * VM context managed by a virtual kernel then let the virtual kernel
893 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
894 vkernel_trap(lp, frame);
899 * Translate fault for emulators (e.g. Linux)
901 if (*p->p_sysent->sv_transtrap)
902 i = (*p->p_sysent->sv_transtrap)(i, type);
904 MAKEMPSAFE(have_mplock);
905 trapsignal(lp, i, ucode);
908 if (type <= MAX_TRAP_MSG) {
909 uprintf("fatal process exception: %s",
911 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
912 uprintf(", fault VA = 0x%lx", (u_long)eva);
919 if (ISPL(frame->tf_cs) == SEL_UPL)
920 KASSERT(td->td_mpcount == have_mplock, ("badmpcount trap/end from %p", (void *)frame->tf_eip));
922 userret(lp, frame, sticks);
929 if (p != NULL && lp != NULL)
930 KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid);
932 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
933 ("syscall: critical section count mismatch! %d/%d",
934 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
939 trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
942 struct vmspace *vm = NULL;
947 thread_t td = curthread;
948 struct lwp *lp = td->td_lwp;
950 va = trunc_page(eva);
951 if (va >= KERNBASE) {
953 * Don't allow user-mode faults in kernel address space.
954 * An exception: if the faulting address is the invalid
955 * instruction entry in the IDT, then the Intel Pentium
956 * F00F bug workaround was triggered, and we need to
957 * treat it is as an illegal instruction, and not a page
960 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
961 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) {
962 frame->tf_trapno = T_PRIVINFLT;
972 * This is a fault on non-kernel virtual memory.
973 * vm is initialized above to NULL. If curproc is NULL
974 * or curproc->p_vmspace is NULL the fault is fatal.
977 vm = lp->lwp_vmspace;
985 if (frame->tf_err & PGEX_W)
986 ftype = VM_PROT_WRITE;
988 ftype = VM_PROT_READ;
990 if (map != &kernel_map) {
992 * Keep swapout from messing with us during this
1002 fault_flags |= VM_FAULT_BURST;
1003 if (ftype & VM_PROT_WRITE)
1004 fault_flags |= VM_FAULT_DIRTY;
1006 fault_flags |= VM_FAULT_NORMAL;
1007 rv = vm_fault(map, va, ftype, fault_flags);
1008 PRELE(lp->lwp_proc);
1011 * Don't have to worry about process locking or stacks
1014 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
1017 if (rv == KERN_SUCCESS)
1021 if (td->td_gd->gd_intr_nesting_level == 0 &&
1022 td->td_pcb->pcb_onfault) {
1023 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
1026 trap_fatal(frame, eva);
1030 /* kludge to pass faulting virtual address to sendsig */
1031 frame->tf_xflags = frame->tf_err;
1032 frame->tf_err = eva;
1034 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
1038 trap_fatal(struct trapframe *frame, vm_offset_t eva)
1040 int code, type, ss, esp;
1041 struct soft_segment_descriptor softseg;
1043 code = frame->tf_err;
1044 type = frame->tf_trapno;
1045 sdtossd(&gdt[mycpu->gd_cpuid * NGDT + IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
1047 if (type <= MAX_TRAP_MSG)
1048 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
1049 type, trap_msg[type],
1050 frame->tf_eflags & PSL_VM ? "vm86" :
1051 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
1053 /* three separate prints in case of a trap on an unmapped page */
1054 kprintf("mp_lock = %08x; ", mp_lock);
1055 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
1056 kprintf("lapic.id = %08x\n", lapic.id);
1058 if (type == T_PAGEFLT) {
1059 kprintf("fault virtual address = %p\n", (void *)eva);
1060 kprintf("fault code = %s %s, %s\n",
1061 code & PGEX_U ? "user" : "supervisor",
1062 code & PGEX_W ? "write" : "read",
1063 code & PGEX_P ? "protection violation" : "page not present");
1065 kprintf("instruction pointer = 0x%x:0x%x\n",
1066 frame->tf_cs & 0xffff, frame->tf_eip);
1067 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
1068 ss = frame->tf_ss & 0xffff;
1069 esp = frame->tf_esp;
1071 ss = GSEL(GDATA_SEL, SEL_KPL);
1072 esp = (int)&frame->tf_esp;
1074 kprintf("stack pointer = 0x%x:0x%x\n", ss, esp);
1075 kprintf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
1076 kprintf("code segment = base 0x%x, limit 0x%x, type 0x%x\n",
1077 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
1078 kprintf(" = DPL %d, pres %d, def32 %d, gran %d\n",
1079 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
1081 kprintf("processor eflags = ");
1082 if (frame->tf_eflags & PSL_T)
1083 kprintf("trace trap, ");
1084 if (frame->tf_eflags & PSL_I)
1085 kprintf("interrupt enabled, ");
1086 if (frame->tf_eflags & PSL_NT)
1087 kprintf("nested task, ");
1088 if (frame->tf_eflags & PSL_RF)
1089 kprintf("resume, ");
1090 if (frame->tf_eflags & PSL_VM)
1092 kprintf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
1093 kprintf("current process = ");
1095 kprintf("%lu (%s)\n",
1096 (u_long)curproc->p_pid, curproc->p_comm ?
1097 curproc->p_comm : "");
1101 kprintf("current thread = pri %d ", curthread->td_pri);
1102 if (curthread->td_pri >= TDPRI_CRIT)
1108 * we probably SHOULD have stopped the other CPUs before now!
1109 * another CPU COULD have been touching cpl at this moment...
1111 kprintf(" <- SMP: XXX");
1120 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1123 kprintf("trap number = %d\n", type);
1124 if (type <= MAX_TRAP_MSG)
1125 panic("%s", trap_msg[type]);
1127 panic("unknown/reserved trap");
1131 * Double fault handler. Called when a fault occurs while writing
1132 * a frame for a trap/exception onto the stack. This usually occurs
1133 * when the stack overflows (such is the case with infinite recursion,
1136 * XXX Note that the current PTD gets replaced by IdlePTD when the
1137 * task switch occurs. This means that the stack that was active at
1138 * the time of the double fault is not available at <kstack> unless
1139 * the machine was idle when the double fault occurred. The downside
1140 * of this is that "trace <ebp>" in ddb won't work.
1143 dblfault_handler(void)
1145 struct mdglobaldata *gd = mdcpu;
1147 kprintf("\nFatal double fault:\n");
1148 kprintf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1149 kprintf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1150 kprintf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
1152 /* three separate prints in case of a trap on an unmapped page */
1153 kprintf("mp_lock = %08x; ", mp_lock);
1154 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
1155 kprintf("lapic.id = %08x\n", lapic.id);
1157 panic("double fault");
1161 * syscall2 - MP aware system call request C handler
1163 * A system call is essentially treated as a trap. The MP lock is not
1164 * held on entry or return. We are responsible for handling ASTs
1165 * (e.g. a task switch) prior to return.
1170 syscall2(struct trapframe *frame)
1172 struct thread *td = curthread;
1173 struct proc *p = td->td_proc;
1174 struct lwp *lp = td->td_lwp;
1176 struct sysent *callp;
1177 register_t orig_tf_eflags;
1182 int crit_count = td->td_pri & ~TDPRI_MASK;
1185 int have_mplock = 0;
1188 union sysunion args;
1191 if (ISPL(frame->tf_cs) != SEL_UPL) {
1198 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid,
1202 KASSERT(td->td_mpcount == 0, ("badmpcount syscall2 from %p", (void *)frame->tf_eip));
1203 if (syscall_mpsafe == 0)
1204 MAKEMPSAFE(have_mplock);
1206 userenter(td, p); /* lazy raise our priority */
1211 sticks = (int)td->td_sticks;
1212 orig_tf_eflags = frame->tf_eflags;
1215 * Virtual kernel intercept - if a VM context managed by a virtual
1216 * kernel issues a system call the virtual kernel handles it, not us.
1217 * Restore the virtual kernel context and return from its system
1218 * call. The current frame is copied out to the virtual kernel.
1220 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1221 vkernel_trap(lp, frame);
1222 error = EJUSTRETURN;
1228 * Get the system call parameters and account for time
1230 lp->lwp_md.md_regs = frame;
1231 params = (caddr_t)frame->tf_esp + sizeof(int);
1232 code = frame->tf_eax;
1234 if (p->p_sysent->sv_prepsyscall) {
1235 (*p->p_sysent->sv_prepsyscall)(
1236 frame, (int *)(&args.nosys.sysmsg + 1),
1240 * Need to check if this is a 32 bit or 64 bit syscall.
1241 * fuword is MP aware.
1243 if (code == SYS_syscall) {
1245 * Code is first argument, followed by actual args.
1247 code = fuword(params);
1248 params += sizeof(int);
1249 } else if (code == SYS___syscall) {
1251 * Like syscall, but code is a quad, so as to maintain
1252 * quad alignment for the rest of the arguments.
1254 code = fuword(params);
1255 params += sizeof(quad_t);
1259 code &= p->p_sysent->sv_mask;
1261 if (code >= p->p_sysent->sv_size)
1262 callp = &p->p_sysent->sv_table[0];
1264 callp = &p->p_sysent->sv_table[code];
1266 narg = callp->sy_narg & SYF_ARGMASK;
1269 if (p->p_sysent->sv_name[0] == 'L')
1270 kprintf("Linux syscall, code = %d\n", code);
1274 * copyin is MP aware, but the tracing code is not
1276 if (narg && params) {
1277 error = copyin(params, (caddr_t)(&args.nosys.sysmsg + 1),
1278 narg * sizeof(register_t));
1281 if (KTRPOINT(td, KTR_SYSCALL)) {
1282 MAKEMPSAFE(have_mplock);
1284 ktrsyscall(lp, code, narg,
1285 (void *)(&args.nosys.sysmsg + 1));
1293 if (KTRPOINT(td, KTR_SYSCALL)) {
1294 MAKEMPSAFE(have_mplock);
1295 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1300 * For traditional syscall code edx is left untouched when 32 bit
1301 * results are returned. Since edx is loaded from fds[1] when the
1302 * system call returns we pre-set it here.
1304 args.sysmsg_fds[0] = 0;
1305 args.sysmsg_fds[1] = frame->tf_edx;
1308 * The syscall might manipulate the trap frame. If it does it
1309 * will probably return EJUSTRETURN.
1311 args.sysmsg_frame = frame;
1313 STOPEVENT(p, S_SCE, narg); /* MP aware */
1316 * NOTE: All system calls run MPSAFE now. The system call itself
1317 * is responsible for getting the MP lock.
1319 error = (*callp->sy_call)(&args);
1323 * MP SAFE (we may or may not have the MP lock at this point)
1328 * Reinitialize proc pointer `p' as it may be different
1329 * if this is a child returning from fork syscall.
1332 lp = curthread->td_lwp;
1333 frame->tf_eax = args.sysmsg_fds[0];
1334 frame->tf_edx = args.sysmsg_fds[1];
1335 frame->tf_eflags &= ~PSL_C;
1339 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1340 * int 0x80 is 2 bytes. We saved this in tf_err.
1342 frame->tf_eip -= frame->tf_err;
1347 panic("Unexpected EASYNC return value (for now)");
1350 if (p->p_sysent->sv_errsize) {
1351 if (error >= p->p_sysent->sv_errsize)
1352 error = -1; /* XXX */
1354 error = p->p_sysent->sv_errtbl[error];
1356 frame->tf_eax = error;
1357 frame->tf_eflags |= PSL_C;
1362 * Traced syscall. trapsignal() is not MP aware.
1364 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1365 MAKEMPSAFE(have_mplock);
1366 frame->tf_eflags &= ~PSL_T;
1367 trapsignal(lp, SIGTRAP, TRAP_TRACE);
1371 * Handle reschedule and other end-of-syscall issues
1373 userret(lp, frame, sticks);
1376 if (KTRPOINT(td, KTR_SYSRET)) {
1377 MAKEMPSAFE(have_mplock);
1378 ktrsysret(lp, code, error, args.sysmsg_result);
1383 * This works because errno is findable through the
1384 * register set. If we ever support an emulation where this
1385 * is not the case, this code will need to be revisited.
1387 STOPEVENT(p, S_SCX, code);
1392 * Release the MP lock if we had to get it
1394 KASSERT(td->td_mpcount == have_mplock,
1395 ("badmpcount syscall2/end from %p callp %p",
1396 (void *)frame->tf_eip, callp));
1400 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error);
1402 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
1403 ("syscall: critical section count mismatch! %d/%d",
1404 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
1409 fork_return(struct lwp *lp, struct trapframe *frame)
1411 frame->tf_eax = 0; /* Child returns zero */
1412 frame->tf_eflags &= ~PSL_C; /* success */
1415 generic_lwp_return(lp, frame);
1416 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1420 * Simplified back end of syscall(), used when returning from fork()
1421 * directly into user mode. MP lock is held on entry and should be
1422 * released on return. This code will return back into the fork
1423 * trampoline code which then runs doreti.
1426 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1428 struct proc *p = lp->lwp_proc;
1431 * Newly forked processes are given a kernel priority. We have to
1432 * adjust the priority to a normal user priority and fake entry
1433 * into the kernel (call userenter()) to install a passive release
1434 * function just in case userret() decides to stop the process. This
1435 * can occur when ^Z races a fork. If we do not install the passive
1436 * release function the current process designation will not be
1437 * released when the thread goes to sleep.
1439 lwkt_setpri_self(TDPRI_USER_NORM);
1440 userenter(lp->lwp_thread, p);
1441 userret(lp, frame, 0);
1443 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1444 ktrsysret(lp, SYS_fork, 0, 0);
1446 p->p_flag |= P_PASSIVE_ACQ;
1448 p->p_flag &= ~P_PASSIVE_ACQ;
1450 KKASSERT(lp->lwp_thread->td_mpcount == 1);
1456 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1457 * fault (which is then passed back to the virtual kernel) if an attempt is
1458 * made to use the FP unit.
1460 * XXX this is a fairly big hack.
1463 set_vkernel_fp(struct trapframe *frame)
1465 struct thread *td = curthread;
1467 if (frame->tf_xflags & PGEX_FPFAULT) {
1468 td->td_pcb->pcb_flags |= FP_VIRTFP;
1469 if (mdcpu->gd_npxthread == td)
1472 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1477 * Called from vkernel_trap() to fixup the vkernel's syscall
1478 * frame for vmspace_ctl() return.
1481 cpu_vkernel_trap(struct trapframe *frame, int error)
1483 frame->tf_eax = error;
1485 frame->tf_eflags |= PSL_C;
1487 frame->tf_eflags &= ~PSL_C;