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 1.69 2005/11/22 00:49:18 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/syscall.h>
63 #include <sys/sysctl.h>
64 #include <sys/sysent.h>
66 #include <sys/vmmeter.h>
67 #include <sys/malloc.h>
69 #include <sys/ktrace.h>
71 #include <sys/upcall.h>
72 #include <sys/sysproto.h>
73 #include <sys/sysunion.h>
76 #include <vm/vm_param.h>
79 #include <vm/vm_kern.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_page.h>
82 #include <vm/vm_extern.h>
84 #include <machine/cpu.h>
85 #include <machine/ipl.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 <i386/isa/intr_machdep.h>
95 #include <sys/syslog.h>
96 #include <machine/clock.h>
99 #include <machine/vm86.h>
102 #include <sys/msgport2.h>
103 #include <sys/thread2.h>
105 int (*pmath_emulate) (struct trapframe *);
107 extern void trap (struct trapframe frame);
108 extern int trapwrite (unsigned addr);
109 extern void syscall2 (struct trapframe frame);
110 extern void sendsys2 (struct trapframe frame);
111 extern void waitsys2 (struct trapframe frame);
113 static int trap_pfault (struct trapframe *, int, vm_offset_t);
114 static void trap_fatal (struct trapframe *, vm_offset_t);
115 void dblfault_handler (void);
117 extern inthand_t IDTVEC(syscall);
119 #define MAX_TRAP_MSG 28
120 static char *trap_msg[] = {
122 "privileged instruction fault", /* 1 T_PRIVINFLT */
124 "breakpoint instruction fault", /* 3 T_BPTFLT */
127 "arithmetic trap", /* 6 T_ARITHTRAP */
128 "system forced exception", /* 7 T_ASTFLT */
130 "general protection fault", /* 9 T_PROTFLT */
131 "trace trap", /* 10 T_TRCTRAP */
133 "page fault", /* 12 T_PAGEFLT */
135 "alignment fault", /* 14 T_ALIGNFLT */
139 "integer divide fault", /* 18 T_DIVIDE */
140 "non-maskable interrupt trap", /* 19 T_NMI */
141 "overflow trap", /* 20 T_OFLOW */
142 "FPU bounds check fault", /* 21 T_BOUND */
143 "FPU device not available", /* 22 T_DNA */
144 "double fault", /* 23 T_DOUBLEFLT */
145 "FPU operand fetch fault", /* 24 T_FPOPFLT */
146 "invalid TSS fault", /* 25 T_TSSFLT */
147 "segment not present fault", /* 26 T_SEGNPFLT */
148 "stack fault", /* 27 T_STKFLT */
149 "machine check trap", /* 28 T_MCHK */
152 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
153 extern int has_f00f_bug;
157 static int ddb_on_nmi = 1;
158 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
159 &ddb_on_nmi, 0, "Go to DDB on NMI");
161 static int panic_on_nmi = 1;
162 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
163 &panic_on_nmi, 0, "Panic on NMI");
164 static int fast_release;
165 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
166 &fast_release, 0, "Passive Release was optimal");
167 static int slow_release;
168 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
169 &slow_release, 0, "Passive Release was nonoptimal");
171 static int syscall_mpsafe = 0;
172 SYSCTL_INT(_kern, OID_AUTO, syscall_mpsafe, CTLFLAG_RW,
173 &syscall_mpsafe, 0, "Allow MPSAFE marked syscalls to run without BGL");
174 TUNABLE_INT("kern.syscall_mpsafe", &syscall_mpsafe);
177 MALLOC_DEFINE(M_SYSMSG, "sysmsg", "sysmsg structure");
178 extern int max_sysmsg;
181 * Passive USER->KERNEL transition. This only occurs if we block in the
182 * kernel while still holding our userland priority. We have to fixup our
183 * priority in order to avoid potential deadlocks before we allow the system
184 * to switch us to another thread.
187 passive_release(struct thread *td)
189 struct lwp *lp = td->td_lwp;
191 td->td_release = NULL;
192 lwkt_setpri_self(TDPRI_KERN_USER);
193 lp->lwp_proc->p_usched->release_curproc(lp);
197 * userenter() passively intercepts the thread switch function to increase
198 * the thread priority from a user priority to a kernel priority, reducing
199 * syscall and trap overhead for the case where no switch occurs.
203 userenter(struct thread *curtd)
205 curtd->td_release = passive_release;
209 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
210 * must be completed before we can return to or try to return to userland.
212 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
213 * arithmatic on the delta calculation so the absolute tick values are
214 * truncated to an integer.
217 userret(struct lwp *lp, struct trapframe *frame, int sticks)
219 struct proc *p = lp->lwp_proc;
223 * Charge system time if profiling. Note: times are in microseconds.
224 * This may do a copyout and block, so do it first even though it
225 * means some system time will be charged as user time.
227 if (p->p_flag & P_PROFIL) {
228 addupc_task(p, frame->tf_eip,
229 (u_int)((int)p->p_thread->td_sticks - sticks));
234 * Block here if we are in a stopped state.
236 if (p->p_flag & P_STOPPED) {
244 * Post any pending upcalls
246 if (p->p_flag & P_UPCALLPEND) {
247 p->p_flag &= ~P_UPCALLPEND;
255 * Post any pending signals
257 if ((sig = CURSIG(p)) != 0) {
265 * block here if we are swapped out, but still process signals
266 * (such as SIGKILL). proc0 (the swapin scheduler) is already
267 * aware of our situation, we do not have to wake it up.
269 if (p->p_flag & P_SWAPPEDOUT) {
271 p->p_flag |= P_SWAPWAIT;
273 if (p->p_flag & P_SWAPWAIT)
274 tsleep(p, PCATCH, "SWOUT", 0);
275 p->p_flag &= ~P_SWAPWAIT;
282 * Cleanup from userenter and any passive release that might have occured.
283 * We must reclaim the current-process designation before we can return
284 * to usermode. We also handle both LWKT and USER reschedule requests.
287 userexit(struct lwp *lp)
289 struct thread *td = lp->lwp_thread;
290 globaldata_t gd = td->td_gd;
294 * If a user reschedule is requested force a new process to be
295 * chosen by releasing the current process. Our process will only
296 * be chosen again if it has a considerably better priority.
298 if (user_resched_wanted())
299 lp->lwp_proc->p_usched->release_curproc(lp);
304 * Handle a LWKT reschedule request first. Since our passive release
305 * is still in place we do not have to do anything special.
307 if (lwkt_resched_wanted())
311 * Acquire the current process designation if we do not own it.
312 * Note that acquire_curproc() does not reset the user reschedule
313 * bit on purpose, because we may need to accumulate over several
314 * threads waking up at the same time.
316 * NOTE: userland scheduler cruft: because processes are removed
317 * from the userland scheduler's queue we run through loops to try
318 * to figure out which is the best of [ existing, waking-up ]
321 if (lp != gd->gd_uschedcp) {
323 lp->lwp_proc->p_usched->acquire_curproc(lp);
324 /* We may have switched cpus on acquisition */
331 * Reduce our priority in preparation for a return to userland. If
332 * our passive release function was still in place, our priority was
333 * never raised and does not need to be reduced.
335 if (td->td_release == NULL)
336 lwkt_setpri_self(TDPRI_USER_NORM);
337 td->td_release = NULL;
340 * After reducing our priority there might be other kernel-level
341 * LWKTs that now have a greater priority. Run them as necessary.
342 * We don't have to worry about losing cpu to userland because
343 * we still control the current-process designation and we no longer
344 * have a passive release function installed.
346 if (lwkt_checkpri_self())
350 * If a userland reschedule is [still] pending we may not be the best
351 * selected process. Select a better one. If another LWKT resched
352 * is pending the trap will be re-entered.
354 if (user_resched_wanted()) {
355 lp->lwp_proc->p_usched->select_curproc(gd);
356 if (lp != gd->gd_uschedcp) {
357 lwkt_setpri_self(TDPRI_KERN_USER);
364 * Exception, fault, and trap interface to the kernel.
365 * This common code is called from assembly language IDT gate entry
366 * routines that prepare a suitable stack frame, and restore this
367 * frame after the exception has been processed.
369 * This function is also called from doreti in an interlock to handle ASTs.
370 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
372 * NOTE! We have to retrieve the fault address prior to obtaining the
373 * MP lock because get_mplock() may switch out. YYY cr2 really ought
374 * to be retrieved by the assembly code, not here.
376 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
377 * if an attempt is made to switch from a fast interrupt or IPI. This is
378 * necessary to properly take fatal kernel traps on SMP machines if
379 * get_mplock() has to block.
383 struct trapframe frame;
385 struct globaldata *gd = mycpu;
386 struct thread *td = gd->gd_curthread;
387 struct lwp *lp = td->td_lwp;
390 int i = 0, ucode = 0, type, code;
396 eva = (frame.tf_trapno == T_PAGEFLT ? rcr2() : 0);
397 ++gd->gd_trap_nesting_level;
399 trap_fatal(&frame, eva);
400 --gd->gd_trap_nesting_level;
406 ++gd->gd_trap_nesting_level;
407 if (frame.tf_trapno == T_PAGEFLT) {
409 * For some Cyrix CPUs, %cr2 is clobbered by interrupts.
410 * This problem is worked around by using an interrupt
411 * gate for the pagefault handler. We are finally ready
412 * to read %cr2 and then must reenable interrupts.
414 * XXX this should be in the switch statement, but the
415 * NO_FOOF_HACK and VM86 goto and ifdefs obfuscate the
416 * flow of control too much for this to be obviously
425 --gd->gd_trap_nesting_level;
427 * MP lock is held at this point
430 if (!(frame.tf_eflags & PSL_I)) {
432 * Buggy application or kernel code has disabled interrupts
433 * and then trapped. Enabling interrupts now is wrong, but
434 * it is better than running with interrupts disabled until
435 * they are accidentally enabled later.
437 type = frame.tf_trapno;
438 if (ISPL(frame.tf_cs)==SEL_UPL || (frame.tf_eflags & PSL_VM)) {
440 "pid %ld (%s): trap %d with interrupts disabled\n",
441 (long)curproc->p_pid, curproc->p_comm, type);
442 } else if (type != T_BPTFLT && type != T_TRCTRAP) {
444 * XXX not quite right, since this may be for a
445 * multiple fault in user mode.
447 printf("kernel trap %d with interrupts disabled\n",
453 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
456 type = frame.tf_trapno;
460 if (frame.tf_eflags & PSL_VM &&
461 (type == T_PROTFLT || type == T_STKFLT)) {
463 KKASSERT(td->td_mpcount > 0);
465 i = vm86_emulate((struct vm86frame *)&frame);
467 KKASSERT(td->td_mpcount > 0);
471 * returns to original process
473 vm86_trap((struct vm86frame *)&frame);
480 * these traps want either a process context, or
481 * assume a normal userspace trap.
485 trap_fatal(&frame, eva);
488 type = T_BPTFLT; /* kernel breakpoint */
491 goto kernel_trap; /* normal kernel trap handling */
494 if ((ISPL(frame.tf_cs) == SEL_UPL) || (frame.tf_eflags & PSL_VM)) {
499 sticks = (int)td->td_sticks;
500 lp->lwp_md.md_regs = &frame;
503 case T_PRIVINFLT: /* privileged instruction fault */
508 case T_BPTFLT: /* bpt instruction fault */
509 case T_TRCTRAP: /* trace trap */
510 frame.tf_eflags &= ~PSL_T;
514 case T_ARITHTRAP: /* arithmetic trap */
519 case T_ASTFLT: /* Allow process switch */
520 mycpu->gd_cnt.v_soft++;
521 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
522 atomic_clear_int_nonlocked(&mycpu->gd_reqflags,
524 addupc_task(p, p->p_prof.pr_addr,
530 * The following two traps can happen in
531 * vm86 mode, and, if so, we want to handle
534 case T_PROTFLT: /* general protection fault */
535 case T_STKFLT: /* stack fault */
536 if (frame.tf_eflags & PSL_VM) {
537 i = vm86_emulate((struct vm86frame *)&frame);
544 case T_SEGNPFLT: /* segment not present fault */
545 case T_TSSFLT: /* invalid TSS fault */
546 case T_DOUBLEFLT: /* double fault */
548 ucode = code + BUS_SEGM_FAULT ;
552 case T_PAGEFLT: /* page fault */
553 i = trap_pfault(&frame, TRUE, eva);
556 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
566 case T_DIVIDE: /* integer divide fault */
574 goto handle_powerfail;
575 #else /* !POWERFAIL_NMI */
576 /* machine/parity/power fail/"kitchen sink" faults */
577 if (isa_nmi(code) == 0) {
580 * NMI can be hooked up to a pushbutton
584 printf ("NMI ... going to debugger\n");
585 kdb_trap (type, 0, &frame);
589 } else if (panic_on_nmi)
590 panic("NMI indicates hardware failure");
592 #endif /* POWERFAIL_NMI */
593 #endif /* NISA > 0 */
595 case T_OFLOW: /* integer overflow fault */
600 case T_BOUND: /* bounds check fault */
608 * The kernel may have switched out the FP unit's
609 * state, causing the user process to take a fault
610 * when it tries to use the FP unit. Restore the
616 if (!pmath_emulate) {
618 ucode = FPE_FPU_NP_TRAP;
621 i = (*pmath_emulate)(&frame);
623 if (!(frame.tf_eflags & PSL_T))
625 frame.tf_eflags &= ~PSL_T;
628 /* else ucode = emulator_only_knows() XXX */
631 case T_FPOPFLT: /* FPU operand fetch fault */
636 case T_XMMFLT: /* SIMD floating-point exception */
646 case T_PAGEFLT: /* page fault */
647 (void) trap_pfault(&frame, FALSE, eva);
653 * The kernel may be using npx for copying or other
661 case T_PROTFLT: /* general protection fault */
662 case T_SEGNPFLT: /* segment not present fault */
664 * Invalid segment selectors and out of bounds
665 * %eip's and %esp's can be set up in user mode.
666 * This causes a fault in kernel mode when the
667 * kernel tries to return to user mode. We want
668 * to get this fault so that we can fix the
669 * problem here and not have to check all the
670 * selectors and pointers when the user changes
673 #define MAYBE_DORETI_FAULT(where, whereto) \
675 if (frame.tf_eip == (int)where) { \
676 frame.tf_eip = (int)whereto; \
681 * Since we don't save %gs across an interrupt
682 * frame this check must occur outside the intr
683 * nesting level check.
685 if (frame.tf_eip == (int)cpu_switch_load_gs) {
686 td->td_pcb->pcb_gs = 0;
690 if (mycpu->gd_intr_nesting_level == 0) {
692 * Invalid %fs's and %gs's can be created using
693 * procfs or PT_SETREGS or by invalidating the
694 * underlying LDT entry. This causes a fault
695 * in kernel mode when the kernel attempts to
696 * switch contexts. Lose the bad context
697 * (XXX) so that we can continue, and generate
700 MAYBE_DORETI_FAULT(doreti_iret,
702 MAYBE_DORETI_FAULT(doreti_popl_ds,
703 doreti_popl_ds_fault);
704 MAYBE_DORETI_FAULT(doreti_popl_es,
705 doreti_popl_es_fault);
706 MAYBE_DORETI_FAULT(doreti_popl_fs,
707 doreti_popl_fs_fault);
708 if (td->td_pcb->pcb_onfault) {
710 (register_t)td->td_pcb->pcb_onfault;
718 * PSL_NT can be set in user mode and isn't cleared
719 * automatically when the kernel is entered. This
720 * causes a TSS fault when the kernel attempts to
721 * `iret' because the TSS link is uninitialized. We
722 * want to get this fault so that we can fix the
723 * problem here and not every time the kernel is
726 if (frame.tf_eflags & PSL_NT) {
727 frame.tf_eflags &= ~PSL_NT;
732 case T_TRCTRAP: /* trace trap */
733 if (frame.tf_eip == (int)IDTVEC(syscall)) {
735 * We've just entered system mode via the
736 * syscall lcall. Continue single stepping
737 * silently until the syscall handler has
742 if (frame.tf_eip == (int)IDTVEC(syscall) + 1) {
744 * The syscall handler has now saved the
745 * flags. Stop single stepping it.
747 frame.tf_eflags &= ~PSL_T;
751 * Ignore debug register trace traps due to
752 * accesses in the user's address space, which
753 * can happen under several conditions such as
754 * if a user sets a watchpoint on a buffer and
755 * then passes that buffer to a system call.
756 * We still want to get TRCTRAPS for addresses
757 * in kernel space because that is useful when
758 * debugging the kernel.
760 if (user_dbreg_trap()) {
762 * Reset breakpoint bits because the
765 load_dr6(rdr6() & 0xfffffff0);
769 * Fall through (TRCTRAP kernel mode, kernel address)
773 * If DDB is enabled, let it handle the debugger trap.
774 * Otherwise, debugger traps "can't happen".
777 if (kdb_trap (type, 0, &frame))
786 # define TIMER_FREQ 1193182
790 static unsigned lastalert = 0;
792 if(time_second - lastalert > 10)
794 log(LOG_WARNING, "NMI: power fail\n");
795 sysbeep(TIMER_FREQ/880, hz);
796 lastalert = time_second;
801 #else /* !POWERFAIL_NMI */
802 /* machine/parity/power fail/"kitchen sink" faults */
803 if (isa_nmi(code) == 0) {
806 * NMI can be hooked up to a pushbutton
810 printf ("NMI ... going to debugger\n");
811 kdb_trap (type, 0, &frame);
815 } else if (panic_on_nmi == 0)
818 #endif /* POWERFAIL_NMI */
819 #endif /* NISA > 0 */
822 trap_fatal(&frame, eva);
826 /* Translate fault for emulators (e.g. Linux) */
827 if (*p->p_sysent->sv_transtrap)
828 i = (*p->p_sysent->sv_transtrap)(i, type);
830 trapsignal(p, i, ucode);
833 if (type <= MAX_TRAP_MSG) {
834 uprintf("fatal process exception: %s",
836 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
837 uprintf(", fault VA = 0x%lx", (u_long)eva);
844 if (ISPL(frame.tf_cs) == SEL_UPL)
845 KASSERT(td->td_mpcount == 1, ("badmpcount trap from %p", (void *)frame.tf_eip));
847 userret(lp, &frame, sticks);
851 KKASSERT(td->td_mpcount > 0);
858 * This version doesn't allow a page fault to user space while
859 * in the kernel. The rest of the kernel needs to be made "safe"
860 * before this can be used. I think the only things remaining
861 * to be made safe are the iBCS2 code and the process tracing/
865 trap_pfault(frame, usermode, eva)
866 struct trapframe *frame;
871 struct vmspace *vm = NULL;
875 thread_t td = curthread;
876 struct proc *p = td->td_proc; /* may be NULL */
878 if (frame->tf_err & PGEX_W)
879 ftype = VM_PROT_WRITE;
881 ftype = VM_PROT_READ;
883 va = trunc_page(eva);
884 if (va < VM_MIN_KERNEL_ADDRESS) {
889 (!usermode && va < VM_MAXUSER_ADDRESS &&
890 (td->td_gd->gd_intr_nesting_level != 0 ||
891 td->td_pcb->pcb_onfault == NULL))) {
892 trap_fatal(frame, eva);
897 * This is a fault on non-kernel virtual memory.
898 * vm is initialized above to NULL. If curproc is NULL
899 * or curproc->p_vmspace is NULL the fault is fatal.
908 * Keep swapout from messing with us during this
914 * Grow the stack if necessary
916 /* grow_stack returns false only if va falls into
917 * a growable stack region and the stack growth
918 * fails. It returns true if va was not within
919 * a growable stack region, or if the stack
922 if (!grow_stack (p, va)) {
928 /* Fault in the user page: */
929 rv = vm_fault(map, va, ftype,
930 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
936 * Don't allow user-mode faults in kernel address space.
942 * Since we know that kernel virtual address addresses
943 * always have pte pages mapped, we just have to fault
946 rv = vm_fault(kernel_map, va, ftype, VM_FAULT_NORMAL);
949 if (rv == KERN_SUCCESS)
953 if (mtd->td_gd->gd_intr_nesting_level == 0 &&
954 td->td_pcb->pcb_onfault) {
955 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
958 trap_fatal(frame, eva);
962 /* kludge to pass faulting virtual address to sendsig */
965 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
970 trap_pfault(frame, usermode, eva)
971 struct trapframe *frame;
976 struct vmspace *vm = NULL;
980 thread_t td = curthread;
981 struct proc *p = td->td_proc;
983 va = trunc_page(eva);
984 if (va >= KERNBASE) {
986 * Don't allow user-mode faults in kernel address space.
987 * An exception: if the faulting address is the invalid
988 * instruction entry in the IDT, then the Intel Pentium
989 * F00F bug workaround was triggered, and we need to
990 * treat it is as an illegal instruction, and not a page
993 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
994 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) {
995 frame->tf_trapno = T_PRIVINFLT;
1005 * This is a fault on non-kernel virtual memory.
1006 * vm is initialized above to NULL. If curproc is NULL
1007 * or curproc->p_vmspace is NULL the fault is fatal.
1018 if (frame->tf_err & PGEX_W)
1019 ftype = VM_PROT_WRITE;
1021 ftype = VM_PROT_READ;
1023 if (map != kernel_map) {
1025 * Keep swapout from messing with us during this
1031 * Grow the stack if necessary
1033 /* grow_stack returns false only if va falls into
1034 * a growable stack region and the stack growth
1035 * fails. It returns true if va was not within
1036 * a growable stack region, or if the stack
1039 if (!grow_stack (p, va)) {
1045 /* Fault in the user page: */
1046 rv = vm_fault(map, va, ftype,
1047 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
1053 * Don't have to worry about process locking or stacks in the kernel.
1055 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
1058 if (rv == KERN_SUCCESS)
1062 if (td->td_gd->gd_intr_nesting_level == 0 &&
1063 td->td_pcb->pcb_onfault) {
1064 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
1067 trap_fatal(frame, eva);
1071 /* kludge to pass faulting virtual address to sendsig */
1072 frame->tf_err = eva;
1074 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
1078 trap_fatal(frame, eva)
1079 struct trapframe *frame;
1082 int code, type, ss, esp;
1083 struct soft_segment_descriptor softseg;
1085 code = frame->tf_err;
1086 type = frame->tf_trapno;
1087 sdtossd(&gdt[mycpu->gd_cpuid * NGDT + IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
1089 if (type <= MAX_TRAP_MSG)
1090 printf("\n\nFatal trap %d: %s while in %s mode\n",
1091 type, trap_msg[type],
1092 frame->tf_eflags & PSL_VM ? "vm86" :
1093 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
1095 /* three separate prints in case of a trap on an unmapped page */
1096 printf("mp_lock = %08x; ", mp_lock);
1097 printf("cpuid = %d; ", mycpu->gd_cpuid);
1098 printf("lapic.id = %08x\n", lapic.id);
1100 if (type == T_PAGEFLT) {
1101 printf("fault virtual address = 0x%x\n", eva);
1102 printf("fault code = %s %s, %s\n",
1103 code & PGEX_U ? "user" : "supervisor",
1104 code & PGEX_W ? "write" : "read",
1105 code & PGEX_P ? "protection violation" : "page not present");
1107 printf("instruction pointer = 0x%x:0x%x\n",
1108 frame->tf_cs & 0xffff, frame->tf_eip);
1109 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
1110 ss = frame->tf_ss & 0xffff;
1111 esp = frame->tf_esp;
1113 ss = GSEL(GDATA_SEL, SEL_KPL);
1114 esp = (int)&frame->tf_esp;
1116 printf("stack pointer = 0x%x:0x%x\n", ss, esp);
1117 printf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
1118 printf("code segment = base 0x%x, limit 0x%x, type 0x%x\n",
1119 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
1120 printf(" = DPL %d, pres %d, def32 %d, gran %d\n",
1121 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
1123 printf("processor eflags = ");
1124 if (frame->tf_eflags & PSL_T)
1125 printf("trace trap, ");
1126 if (frame->tf_eflags & PSL_I)
1127 printf("interrupt enabled, ");
1128 if (frame->tf_eflags & PSL_NT)
1129 printf("nested task, ");
1130 if (frame->tf_eflags & PSL_RF)
1132 if (frame->tf_eflags & PSL_VM)
1134 printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
1135 printf("current process = ");
1137 printf("%lu (%s)\n",
1138 (u_long)curproc->p_pid, curproc->p_comm ?
1139 curproc->p_comm : "");
1143 printf("current thread = pri %d ", curthread->td_pri);
1144 if (curthread->td_pri >= TDPRI_CRIT)
1150 * we probably SHOULD have stopped the other CPUs before now!
1151 * another CPU COULD have been touching cpl at this moment...
1153 printf(" <- SMP: XXX");
1162 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1165 printf("trap number = %d\n", type);
1166 if (type <= MAX_TRAP_MSG)
1167 panic("%s", trap_msg[type]);
1169 panic("unknown/reserved trap");
1173 * Double fault handler. Called when a fault occurs while writing
1174 * a frame for a trap/exception onto the stack. This usually occurs
1175 * when the stack overflows (such is the case with infinite recursion,
1178 * XXX Note that the current PTD gets replaced by IdlePTD when the
1179 * task switch occurs. This means that the stack that was active at
1180 * the time of the double fault is not available at <kstack> unless
1181 * the machine was idle when the double fault occurred. The downside
1182 * of this is that "trace <ebp>" in ddb won't work.
1187 struct mdglobaldata *gd = mdcpu;
1189 printf("\nFatal double fault:\n");
1190 printf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1191 printf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1192 printf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
1194 /* three separate prints in case of a trap on an unmapped page */
1195 printf("mp_lock = %08x; ", mp_lock);
1196 printf("cpuid = %d; ", mycpu->gd_cpuid);
1197 printf("lapic.id = %08x\n", lapic.id);
1199 panic("double fault");
1203 * Compensate for 386 brain damage (missing URKR).
1204 * This is a little simpler than the pagefault handler in trap() because
1205 * it the page tables have already been faulted in and high addresses
1206 * are thrown out early for other reasons.
1216 va = trunc_page((vm_offset_t)addr);
1218 * XXX - MAX is END. Changed > to >= for temp. fix.
1220 if (va >= VM_MAXUSER_ADDRESS)
1228 if (!grow_stack (p, va)) {
1234 * fault the data page
1236 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY);
1240 if (rv != KERN_SUCCESS)
1247 * syscall2 - MP aware system call request C handler
1249 * A system call is essentially treated as a trap except that the
1250 * MP lock is not held on entry or return. We are responsible for
1251 * obtaining the MP lock if necessary and for handling ASTs
1252 * (e.g. a task switch) prior to return.
1254 * In general, only simple access and manipulation of curproc and
1255 * the current stack is allowed without having to hold MP lock.
1257 * MPSAFE - note that large sections of this routine are run without
1262 #define MAKEMPSAFE(have_mplock) \
1263 if (have_mplock == 0) { \
1270 #define MAKEMPSAFE(have_mplock)
1275 syscall2(struct trapframe frame)
1277 struct thread *td = curthread;
1278 struct proc *p = td->td_proc;
1279 struct lwp *lp = td->td_lwp;
1281 struct sysent *callp;
1282 register_t orig_tf_eflags;
1287 int have_mplock = 0;
1290 union sysunion args;
1293 if (ISPL(frame.tf_cs) != SEL_UPL) {
1301 KASSERT(td->td_mpcount == 0, ("badmpcount syscall from %p", (void *)frame.tf_eip));
1302 if (syscall_mpsafe == 0)
1303 MAKEMPSAFE(have_mplock);
1305 userenter(td); /* lazy raise our priority */
1307 sticks = (int)td->td_sticks;
1309 lp->lwp_md.md_regs = &frame;
1310 params = (caddr_t)frame.tf_esp + sizeof(int);
1311 code = frame.tf_eax;
1312 orig_tf_eflags = frame.tf_eflags;
1314 if (p->p_sysent->sv_prepsyscall) {
1315 (*p->p_sysent->sv_prepsyscall)(
1316 &frame, (int *)(&args.nosys.usrmsg + 1),
1320 * Need to check if this is a 32 bit or 64 bit syscall.
1321 * fuword is MP aware.
1323 if (code == SYS_syscall) {
1325 * Code is first argument, followed by actual args.
1327 code = fuword(params);
1328 params += sizeof(int);
1329 } else if (code == SYS___syscall) {
1331 * Like syscall, but code is a quad, so as to maintain
1332 * quad alignment for the rest of the arguments.
1334 code = fuword(params);
1335 params += sizeof(quad_t);
1339 code &= p->p_sysent->sv_mask;
1340 if (code >= p->p_sysent->sv_size)
1341 callp = &p->p_sysent->sv_table[0];
1343 callp = &p->p_sysent->sv_table[code];
1345 narg = callp->sy_narg & SYF_ARGMASK;
1348 * copyin is MP aware, but the tracing code is not
1350 if (narg && params) {
1351 error = copyin(params, (caddr_t)(&args.nosys.usrmsg + 1),
1352 narg * sizeof(register_t));
1355 if (KTRPOINT(td, KTR_SYSCALL)) {
1356 MAKEMPSAFE(have_mplock);
1358 ktrsyscall(p->p_tracep, code, narg,
1359 (void *)(&args.nosys.usrmsg + 1));
1367 if (KTRPOINT(td, KTR_SYSCALL)) {
1368 MAKEMPSAFE(have_mplock);
1369 ktrsyscall(p->p_tracep, code, narg, (void *)(&args.nosys.usrmsg + 1));
1374 * For traditional syscall code edx is left untouched when 32 bit
1375 * results are returned. Since edx is loaded from fds[1] when the
1376 * system call returns we pre-set it here.
1378 lwkt_initmsg(&args.lmsg, &td->td_msgport, 0,
1379 lwkt_cmd_op(code), lwkt_cmd_op_none);
1380 args.sysmsg_copyout = NULL;
1381 args.sysmsg_fds[0] = 0;
1382 args.sysmsg_fds[1] = frame.tf_edx;
1384 STOPEVENT(p, S_SCE, narg); /* MP aware */
1388 * Try to run the syscall without the MP lock if the syscall
1389 * is MP safe. We have to obtain the MP lock no matter what if
1392 if ((callp->sy_narg & SYF_MPSAFE) == 0)
1393 MAKEMPSAFE(have_mplock);
1396 error = (*callp->sy_call)(&args);
1399 * MP SAFE (we may or may not have the MP lock at this point)
1404 * Reinitialize proc pointer `p' as it may be different
1405 * if this is a child returning from fork syscall.
1408 lp = curthread->td_lwp;
1409 frame.tf_eax = args.sysmsg_fds[0];
1410 frame.tf_edx = args.sysmsg_fds[1];
1411 frame.tf_eflags &= ~PSL_C;
1415 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1416 * int 0x80 is 2 bytes. We saved this in tf_err.
1418 frame.tf_eip -= frame.tf_err;
1423 panic("Unexpected EASYNC return value (for now)");
1426 if (p->p_sysent->sv_errsize) {
1427 if (error >= p->p_sysent->sv_errsize)
1428 error = -1; /* XXX */
1430 error = p->p_sysent->sv_errtbl[error];
1432 frame.tf_eax = error;
1433 frame.tf_eflags |= PSL_C;
1438 * Traced syscall. trapsignal() is not MP aware.
1440 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1441 MAKEMPSAFE(have_mplock);
1442 frame.tf_eflags &= ~PSL_T;
1443 trapsignal(p, SIGTRAP, 0);
1447 * Handle reschedule and other end-of-syscall issues
1449 userret(lp, &frame, sticks);
1452 if (KTRPOINT(td, KTR_SYSRET)) {
1453 MAKEMPSAFE(have_mplock);
1454 ktrsysret(p->p_tracep, code, error, args.sysmsg_result);
1459 * This works because errno is findable through the
1460 * register set. If we ever support an emulation where this
1461 * is not the case, this code will need to be revisited.
1463 STOPEVENT(p, S_SCX, code);
1468 * Release the MP lock if we had to get it
1470 KASSERT(td->td_mpcount == have_mplock,
1471 ("badmpcount syscall from %p", (void *)frame.tf_eip));
1478 * free_sysun - Put an unused sysun on the free list.
1480 static __inline void
1481 free_sysun(struct thread *td, union sysunion *sysun)
1483 struct globaldata *gd = td->td_gd;
1485 crit_enter_quick(td);
1486 sysun->lmsg.opaque.ms_sysunnext = gd->gd_freesysun;
1487 gd->gd_freesysun = sysun;
1488 crit_exit_quick(td);
1492 * sendsys2 - MP aware system message request C handler
1495 sendsys2(struct trapframe frame)
1497 struct globaldata *gd;
1498 struct thread *td = curthread;
1499 struct proc *p = td->td_proc;
1500 struct lwp *lp = td->td_lwp;
1501 register_t orig_tf_eflags;
1502 struct sysent *callp;
1503 union sysunion *sysun = NULL;
1506 int have_mplock = 0;
1516 if (ISPL(frame.tf_cs) != SEL_UPL) {
1524 KASSERT(td->td_mpcount == 0,
1525 ("badmpcount syscall from %p", (void *)frame.tf_eip));
1526 if (syscall_mpsafe == 0)
1527 MAKEMPSAFE(have_mplock);
1530 * access non-atomic field from critical section. p_sticks is
1531 * updated by the clock interrupt. Also use this opportunity
1532 * to lazy-raise our LWKT priority.
1535 sticks = td->td_sticks;
1537 lp->lwp_md.md_regs = &frame;
1538 orig_tf_eflags = frame.tf_eflags;
1542 * Extract the system call message. If msgsize is zero we are
1543 * blocking on a message and/or message port. If msgsize is -1
1544 * we are testing a message for completion or a message port for
1547 * The userland system call message size includes the size of the
1548 * userland lwkt_msg plus arguments. We load it into the userland
1549 * portion of our sysunion structure then we initialize the kerneland
1556 if ((msgsize = frame.tf_edx) < sizeof(struct lwkt_msg) ||
1557 msgsize > sizeof(union sysunion) - sizeof(struct sysmsg)) {
1563 * Obtain a sysun from our per-cpu cache or allocate a new one. Use
1564 * the opaque field to store the original (user) message pointer.
1565 * A critical section is necessary to interlock against interrupts
1566 * returning system messages to the thread cache.
1569 crit_enter_quick(td);
1570 if ((sysun = gd->gd_freesysun) != NULL)
1571 gd->gd_freesysun = sysun->lmsg.opaque.ms_sysunnext;
1573 sysun = malloc(sizeof(union sysunion), M_SYSMSG, M_WAITOK);
1574 crit_exit_quick(td);
1577 * Copy the user request into the kernel copy of the user request.
1579 umsg = (void *)frame.tf_ecx;
1580 error = copyin(umsg, &sysun->nosys.usrmsg, msgsize);
1583 if ((sysun->nosys.usrmsg.umsg.ms_flags & MSGF_ASYNC)) {
1588 if (max_sysmsg > 0 && lp->lwp_nsysmsg >= max_sysmsg) {
1595 * Initialize the kernel message from the copied-in data and
1596 * pull in appropriate flags from the userland message.
1598 * ms_abort_port is usually initialized in sendmsg/domsg, but since
1599 * we are not calling those functions (yet), we have to do it manually.
1601 lwkt_initmsg(&sysun->lmsg, &td->td_msgport, 0,
1602 sysun->nosys.usrmsg.umsg.ms_cmd,
1604 sysun->lmsg.ms_abort_port = sysun->lmsg.ms_reply_port;
1605 sysun->sysmsg_copyout = NULL;
1606 sysun->lmsg.opaque.ms_umsg = umsg;
1607 sysun->lmsg.ms_flags |= sysun->nosys.usrmsg.umsg.ms_flags & MSGF_ASYNC;
1610 * Extract the system call number, lookup the system call, and
1611 * set the default return value.
1613 code = (u_int)sysun->lmsg.ms_cmd.cm_op;
1614 /* We don't handle the syscall() syscall yet */
1617 free_sysun(td, sysun);
1620 if (code >= p->p_sysent->sv_size) {
1622 free_sysun(td, sysun);
1626 callp = &p->p_sysent->sv_table[code];
1628 narg = (msgsize - sizeof(struct lwkt_msg)) / sizeof(register_t);
1631 if (KTRPOINT(td, KTR_SYSCALL)) {
1632 MAKEMPSAFE(have_mplock);
1633 ktrsyscall(p->p_tracep, code, narg, (void *)(&sysun->nosys.usrmsg + 1));
1636 sysun->lmsg.u.ms_fds[0] = 0;
1637 sysun->lmsg.u.ms_fds[1] = 0;
1639 STOPEVENT(p, S_SCE, narg); /* MP aware */
1642 * Make the system call. An error code is always returned, results
1643 * are copied back via ms_result32 or ms_result64. YYY temporary
1644 * stage copy p_retval[] into ms_result32/64
1646 * NOTE! XXX if this is a child returning from a fork curproc
1647 * might be different. YYY huh? a child returning from a fork
1648 * should never 'return' from this call, it should go right to the
1649 * fork_trampoline function.
1651 * Obtain the MP lock if necessary.
1654 if ((callp->sy_narg & SYF_MPSAFE) == 0)
1655 MAKEMPSAFE(have_mplock);
1657 error = (*callp->sy_call)(sysun);
1658 gd = td->td_gd; /* RELOAD, might have switched cpus */
1662 * If a synchronous return copy p_retval to ms_result64 and return
1663 * the sysmsg to the free pool.
1665 * YYY Don't writeback message if execve() YYY
1667 sysun->nosys.usrmsg.umsg.ms_error = error;
1668 sysun->nosys.usrmsg.umsg.u.ms_fds[0] = sysun->lmsg.u.ms_fds[0];
1669 sysun->nosys.usrmsg.umsg.u.ms_fds[1] = sysun->lmsg.u.ms_fds[1];
1670 result = sysun->nosys.usrmsg.umsg.u.ms_fds[0]; /* for ktrace */
1671 if (error != 0 || code != SYS_execve) {
1673 error2 = copyout(&sysun->nosys.usrmsg.umsg.ms_copyout_start,
1674 &umsg->ms_copyout_start,
1679 if (error == EASYNC) {
1681 * Since only the current process ever messes with msgq,
1682 * we can safely manipulate it in parallel with the async
1685 TAILQ_INSERT_TAIL(&lp->lwp_sysmsgq, &sysun->sysmsg, msgq);
1687 error = (int)&sysun->sysmsg;
1689 free_sysun(td, sysun);
1692 frame.tf_eax = (register_t)error;
1695 * Traced syscall. trapsignal() is not MP aware.
1697 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1698 MAKEMPSAFE(have_mplock);
1699 frame.tf_eflags &= ~PSL_T;
1700 trapsignal(p, SIGTRAP, 0);
1704 * Handle reschedule and other end-of-syscall issues
1706 userret(lp, &frame, sticks);
1709 if (KTRPOINT(td, KTR_SYSRET)) {
1710 MAKEMPSAFE(have_mplock);
1711 ktrsysret(p->p_tracep, code, error, result);
1716 * This works because errno is findable through the
1717 * register set. If we ever support an emulation where this
1718 * is not the case, this code will need to be revisited.
1720 STOPEVENT(p, S_SCX, code);
1725 * Release the MP lock if we had to get it
1727 KASSERT(td->td_mpcount == have_mplock,
1728 ("badmpcount syscall from %p", (void *)frame.tf_eip));
1735 * waitsys2 - MP aware system message wait C handler
1738 waitsys2(struct trapframe frame)
1740 struct globaldata *gd;
1741 struct thread *td = curthread;
1742 struct proc *p = td->td_proc;
1743 struct lwp *lp = td->td_lwp;
1744 union sysunion *sysun = NULL;
1746 register_t orig_tf_eflags;
1747 int error = 0, result, sticks;
1749 int have_mplock = 0;
1754 if (ISPL(frame.tf_cs) != SEL_UPL) {
1762 KASSERT(td->td_mpcount == 0, ("badmpcount syscall from %p",
1763 (void *)frame.tf_eip));
1764 if (syscall_mpsafe == 0)
1765 MAKEMPSAFE(have_mplock);
1769 * access non-atomic field from critical section. p_sticks is
1770 * updated by the clock interrupt. Also use this opportunity
1771 * to lazy-raise our LWKT priority.
1774 sticks = td->td_sticks;
1776 lp->lwp_md.md_regs = &frame;
1777 orig_tf_eflags = frame.tf_eflags;
1783 TAILQ_FOREACH(ptr, &lp->lwp_sysmsgq, msgq) {
1784 if ((void *)ptr == (void *)frame.tf_ecx) {
1785 sysun = (void *)sysmsg_wait(lp,
1786 (void *)frame.tf_ecx, 1);
1796 else if (frame.tf_eax) {
1797 printf("waitport/checkport only the default port is supported at the moment\n");
1802 switch(frame.tf_edx) {
1804 sysun = (void *)sysmsg_wait(lp, NULL, 0);
1807 sysun = (void *)sysmsg_wait(lp, NULL, 1);
1816 umsg = sysun->lmsg.opaque.ms_umsg;
1817 frame.tf_eax = (register_t)sysun;
1818 sysun->nosys.usrmsg.umsg.u.ms_fds[0] = sysun->lmsg.u.ms_fds[0];
1819 sysun->nosys.usrmsg.umsg.u.ms_fds[1] = sysun->lmsg.u.ms_fds[1];
1820 sysun->nosys.usrmsg.umsg.ms_error = sysun->lmsg.ms_error;
1821 error = sysun->lmsg.ms_error;
1822 result = sysun->lmsg.u.ms_fds[0]; /* for ktrace */
1823 error = copyout(&sysun->nosys.usrmsg.umsg.ms_copyout_start,
1824 &umsg->ms_copyout_start, ms_copyout_size);
1825 free_sysun(td, sysun);
1827 code = (u_int)sysun->lmsg.ms_cmd.cm_op;
1831 frame.tf_eax = error;
1833 * Traced syscall. trapsignal() is not MP aware.
1835 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1836 MAKEMPSAFE(have_mplock);
1837 frame.tf_eflags &= ~PSL_T;
1838 trapsignal(p, SIGTRAP, 0);
1842 * Handle reschedule and other end-of-syscall issues
1844 userret(lp, &frame, sticks);
1847 if (KTRPOINT(td, KTR_SYSRET)) {
1848 MAKEMPSAFE(have_mplock);
1849 ktrsysret(p->p_tracep, code, error, result);
1854 * This works because errno is findable through the
1855 * register set. If we ever support an emulation where this
1856 * is not the case, this code will need to be revisited.
1858 STOPEVENT(p, S_SCX, code);
1862 KASSERT(td->td_mpcount == 1, ("badmpcount syscall from %p",
1863 (void *)frame.tf_eip));
1870 * Simplified back end of syscall(), used when returning from fork()
1871 * directly into user mode. MP lock is held on entry and should be
1872 * released on return. This code will return back into the fork
1873 * trampoline code which then runs doreti.
1876 fork_return(p, frame)
1878 struct trapframe frame;
1882 KKASSERT(p->p_nthreads == 1);
1884 lp = LIST_FIRST(&p->p_lwps);
1886 frame.tf_eax = 0; /* Child returns zero */
1887 frame.tf_eflags &= ~PSL_C; /* success */
1891 * Newly forked processes are given a kernel priority. We have to
1892 * adjust the priority to a normal user priority and fake entry
1893 * into the kernel (call userenter()) to install a passive release
1894 * function just in case userret() decides to stop the process. This
1895 * can occur when ^Z races a fork. If we do not install the passive
1896 * release function the current process designation will not be
1897 * released when the thread goes to sleep.
1899 lwkt_setpri_self(TDPRI_USER_NORM);
1900 userenter(lp->lwp_thread);
1901 userret(lp, &frame, 0);
1903 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1904 ktrsysret(p->p_tracep, SYS_fork, 0, 0);
1906 p->p_flag |= P_PASSIVE_ACQ;
1908 p->p_flag &= ~P_PASSIVE_ACQ;
1910 KKASSERT(lp->lwp_thread->td_mpcount == 1);