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/i386/i386/Attic/trap.c,v 1.77 2006/06/05 20:59:19 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>
107 #define MAKEMPSAFE(have_mplock) \
108 if (have_mplock == 0) { \
115 #define MAKEMPSAFE(have_mplock)
119 int (*pmath_emulate) (struct trapframe *);
121 extern void trap (struct trapframe frame);
122 extern int trapwrite (unsigned addr);
123 extern void syscall2 (struct trapframe frame);
124 extern void sendsys2 (struct trapframe frame);
125 extern void waitsys2 (struct trapframe frame);
127 static int trap_pfault (struct trapframe *, int, vm_offset_t);
128 static void trap_fatal (struct trapframe *, vm_offset_t);
129 void dblfault_handler (void);
131 extern inthand_t IDTVEC(syscall);
133 #define MAX_TRAP_MSG 28
134 static char *trap_msg[] = {
136 "privileged instruction fault", /* 1 T_PRIVINFLT */
138 "breakpoint instruction fault", /* 3 T_BPTFLT */
141 "arithmetic trap", /* 6 T_ARITHTRAP */
142 "system forced exception", /* 7 T_ASTFLT */
144 "general protection fault", /* 9 T_PROTFLT */
145 "trace trap", /* 10 T_TRCTRAP */
147 "page fault", /* 12 T_PAGEFLT */
149 "alignment fault", /* 14 T_ALIGNFLT */
153 "integer divide fault", /* 18 T_DIVIDE */
154 "non-maskable interrupt trap", /* 19 T_NMI */
155 "overflow trap", /* 20 T_OFLOW */
156 "FPU bounds check fault", /* 21 T_BOUND */
157 "FPU device not available", /* 22 T_DNA */
158 "double fault", /* 23 T_DOUBLEFLT */
159 "FPU operand fetch fault", /* 24 T_FPOPFLT */
160 "invalid TSS fault", /* 25 T_TSSFLT */
161 "segment not present fault", /* 26 T_SEGNPFLT */
162 "stack fault", /* 27 T_STKFLT */
163 "machine check trap", /* 28 T_MCHK */
166 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
167 extern int has_f00f_bug;
171 static int ddb_on_nmi = 1;
172 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
173 &ddb_on_nmi, 0, "Go to DDB on NMI");
175 static int panic_on_nmi = 1;
176 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
177 &panic_on_nmi, 0, "Panic on NMI");
178 static int fast_release;
179 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
180 &fast_release, 0, "Passive Release was optimal");
181 static int slow_release;
182 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
183 &slow_release, 0, "Passive Release was nonoptimal");
185 static int syscall_mpsafe = 0;
186 SYSCTL_INT(_kern, OID_AUTO, syscall_mpsafe, CTLFLAG_RW,
187 &syscall_mpsafe, 0, "Allow MPSAFE marked syscalls to run without BGL");
188 TUNABLE_INT("kern.syscall_mpsafe", &syscall_mpsafe);
189 static int trap_mpsafe = 0;
190 SYSCTL_INT(_kern, OID_AUTO, trap_mpsafe, CTLFLAG_RW,
191 &trap_mpsafe, 0, "Allow traps to mostly run without the BGL");
192 TUNABLE_INT("kern.trap_mpsafe", &trap_mpsafe);
195 MALLOC_DEFINE(M_SYSMSG, "sysmsg", "sysmsg structure");
196 extern int max_sysmsg;
199 * Passive USER->KERNEL transition. This only occurs if we block in the
200 * kernel while still holding our userland priority. We have to fixup our
201 * priority in order to avoid potential deadlocks before we allow the system
202 * to switch us to another thread.
205 passive_release(struct thread *td)
207 struct lwp *lp = td->td_lwp;
209 td->td_release = NULL;
210 lwkt_setpri_self(TDPRI_KERN_USER);
211 lp->lwp_proc->p_usched->release_curproc(lp);
215 * userenter() passively intercepts the thread switch function to increase
216 * the thread priority from a user priority to a kernel priority, reducing
217 * syscall and trap overhead for the case where no switch occurs.
221 userenter(struct thread *curtd)
223 curtd->td_release = passive_release;
227 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
228 * must be completed before we can return to or try to return to userland.
230 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
231 * arithmatic on the delta calculation so the absolute tick values are
232 * truncated to an integer.
235 userret(struct lwp *lp, struct trapframe *frame, int sticks)
237 struct proc *p = lp->lwp_proc;
241 * Charge system time if profiling. Note: times are in microseconds.
242 * This may do a copyout and block, so do it first even though it
243 * means some system time will be charged as user time.
245 if (p->p_flag & P_PROFIL) {
246 addupc_task(p, frame->tf_eip,
247 (u_int)((int)p->p_thread->td_sticks - sticks));
252 * Block here if we are in a stopped state.
254 if (p->p_flag & P_STOPPED) {
262 * Post any pending upcalls
264 if (p->p_flag & P_UPCALLPEND) {
265 p->p_flag &= ~P_UPCALLPEND;
273 * Post any pending signals
275 if ((sig = CURSIG(p)) != 0) {
283 * block here if we are swapped out, but still process signals
284 * (such as SIGKILL). proc0 (the swapin scheduler) is already
285 * aware of our situation, we do not have to wake it up.
287 if (p->p_flag & P_SWAPPEDOUT) {
289 p->p_flag |= P_SWAPWAIT;
291 if (p->p_flag & P_SWAPWAIT)
292 tsleep(p, PCATCH, "SWOUT", 0);
293 p->p_flag &= ~P_SWAPWAIT;
300 * Cleanup from userenter and any passive release that might have occured.
301 * We must reclaim the current-process designation before we can return
302 * to usermode. We also handle both LWKT and USER reschedule requests.
305 userexit(struct lwp *lp)
307 struct thread *td = lp->lwp_thread;
308 globaldata_t gd = td->td_gd;
312 * If a user reschedule is requested force a new process to be
313 * chosen by releasing the current process. Our process will only
314 * be chosen again if it has a considerably better priority.
316 if (user_resched_wanted())
317 lp->lwp_proc->p_usched->release_curproc(lp);
321 * Handle a LWKT reschedule request first. Since our passive release
322 * is still in place we do not have to do anything special.
324 if (lwkt_resched_wanted())
328 * Acquire the current process designation for this user scheduler
329 * on this cpu. This will also handle any user-reschedule requests.
331 lp->lwp_proc->p_usched->acquire_curproc(lp);
332 /* We may have switched cpus on acquisition */
336 * Reduce our priority in preparation for a return to userland. If
337 * our passive release function was still in place, our priority was
338 * never raised and does not need to be reduced.
340 if (td->td_release == NULL)
341 lwkt_setpri_self(TDPRI_USER_NORM);
342 td->td_release = NULL;
345 * After reducing our priority there might be other kernel-level
346 * LWKTs that now have a greater priority. Run them as necessary.
347 * We don't have to worry about losing cpu to userland because
348 * we still control the current-process designation and we no longer
349 * have a passive release function installed.
351 if (lwkt_checkpri_self())
356 * Exception, fault, and trap interface to the kernel.
357 * This common code is called from assembly language IDT gate entry
358 * routines that prepare a suitable stack frame, and restore this
359 * frame after the exception has been processed.
361 * This function is also called from doreti in an interlock to handle ASTs.
362 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
364 * NOTE! We have to retrieve the fault address prior to obtaining the
365 * MP lock because get_mplock() may switch out. YYY cr2 really ought
366 * to be retrieved by the assembly code, not here.
368 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
369 * if an attempt is made to switch from a fast interrupt or IPI. This is
370 * necessary to properly take fatal kernel traps on SMP machines if
371 * get_mplock() has to block.
376 struct trapframe frame;
378 struct globaldata *gd = mycpu;
379 struct thread *td = gd->gd_curthread;
380 struct lwp *lp = td->td_lwp;
383 int i = 0, ucode = 0, type, code;
388 int crit_count = td->td_pri & ~TDPRI_MASK;
395 eva = (frame.tf_trapno == T_PAGEFLT ? rcr2() : 0);
396 ++gd->gd_trap_nesting_level;
397 MAKEMPSAFE(have_mplock);
398 trap_fatal(&frame, eva);
399 --gd->gd_trap_nesting_level;
405 ++gd->gd_trap_nesting_level;
406 if (frame.tf_trapno == T_PAGEFLT) {
408 * For some Cyrix CPUs, %cr2 is clobbered by interrupts.
409 * This problem is worked around by using an interrupt
410 * gate for the pagefault handler. We are finally ready
411 * to read %cr2 and then must reenable interrupts.
413 * XXX this should be in the switch statement, but the
414 * NO_FOOF_HACK and VM86 goto and ifdefs obfuscate the
415 * flow of control too much for this to be obviously
422 if (trap_mpsafe == 0)
423 MAKEMPSAFE(have_mplock);
426 --gd->gd_trap_nesting_level;
428 if (!(frame.tf_eflags & PSL_I)) {
430 * Buggy application or kernel code has disabled interrupts
431 * and then trapped. Enabling interrupts now is wrong, but
432 * it is better than running with interrupts disabled until
433 * they are accidentally enabled later.
435 type = frame.tf_trapno;
436 if (ISPL(frame.tf_cs)==SEL_UPL || (frame.tf_eflags & PSL_VM)) {
437 MAKEMPSAFE(have_mplock);
439 "pid %ld (%s): trap %d with interrupts disabled\n",
440 (long)curproc->p_pid, curproc->p_comm, type);
441 } else if (type != T_BPTFLT && type != T_TRCTRAP) {
443 * XXX not quite right, since this may be for a
444 * multiple fault in user mode.
446 MAKEMPSAFE(have_mplock);
447 printf("kernel trap %d with interrupts disabled\n",
453 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
456 type = frame.tf_trapno;
460 ASSERT_MP_LOCK_HELD(curthread);
461 if (frame.tf_eflags & PSL_VM &&
462 (type == T_PROTFLT || type == T_STKFLT)) {
464 KKASSERT(td->td_mpcount > 0);
466 i = vm86_emulate((struct vm86frame *)&frame);
468 KKASSERT(td->td_mpcount > 0);
472 * returns to original process
475 vm86_trap((struct vm86frame *)&frame,
478 vm86_trap((struct vm86frame *)&frame, 0);
480 KKASSERT(0); /* NOT REACHED */
486 * these traps want either a process context, or
487 * assume a normal userspace trap.
491 trap_fatal(&frame, eva);
494 type = T_BPTFLT; /* kernel breakpoint */
497 goto kernel_trap; /* normal kernel trap handling */
500 if ((ISPL(frame.tf_cs) == SEL_UPL) || (frame.tf_eflags & PSL_VM)) {
505 sticks = (int)td->td_sticks;
506 lp->lwp_md.md_regs = &frame;
509 case T_PRIVINFLT: /* privileged instruction fault */
514 case T_BPTFLT: /* bpt instruction fault */
515 case T_TRCTRAP: /* trace trap */
516 frame.tf_eflags &= ~PSL_T;
520 case T_ARITHTRAP: /* arithmetic trap */
525 case T_ASTFLT: /* Allow process switch */
526 mycpu->gd_cnt.v_soft++;
527 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
528 atomic_clear_int_nonlocked(&mycpu->gd_reqflags,
530 addupc_task(p, p->p_prof.pr_addr,
536 * The following two traps can happen in
537 * vm86 mode, and, if so, we want to handle
540 case T_PROTFLT: /* general protection fault */
541 case T_STKFLT: /* stack fault */
542 if (frame.tf_eflags & PSL_VM) {
543 i = vm86_emulate((struct vm86frame *)&frame);
550 case T_SEGNPFLT: /* segment not present fault */
551 case T_TSSFLT: /* invalid TSS fault */
552 case T_DOUBLEFLT: /* double fault */
554 ucode = code + BUS_SEGM_FAULT ;
558 case T_PAGEFLT: /* page fault */
559 MAKEMPSAFE(have_mplock);
560 i = trap_pfault(&frame, TRUE, eva);
563 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
573 case T_DIVIDE: /* integer divide fault */
580 MAKEMPSAFE(have_mplock);
582 goto handle_powerfail;
583 #else /* !POWERFAIL_NMI */
584 /* machine/parity/power fail/"kitchen sink" faults */
585 if (isa_nmi(code) == 0) {
588 * NMI can be hooked up to a pushbutton
592 printf ("NMI ... going to debugger\n");
593 kdb_trap (type, 0, &frame);
597 } else if (panic_on_nmi)
598 panic("NMI indicates hardware failure");
600 #endif /* POWERFAIL_NMI */
601 #endif /* NISA > 0 */
603 case T_OFLOW: /* integer overflow fault */
608 case T_BOUND: /* bounds check fault */
616 * The kernel may have switched out the FP unit's
617 * state, causing the user process to take a fault
618 * when it tries to use the FP unit. Restore the
624 if (!pmath_emulate) {
626 ucode = FPE_FPU_NP_TRAP;
629 i = (*pmath_emulate)(&frame);
631 if (!(frame.tf_eflags & PSL_T))
633 frame.tf_eflags &= ~PSL_T;
636 /* else ucode = emulator_only_knows() XXX */
639 case T_FPOPFLT: /* FPU operand fetch fault */
644 case T_XMMFLT: /* SIMD floating-point exception */
654 case T_PAGEFLT: /* page fault */
655 MAKEMPSAFE(have_mplock);
656 trap_pfault(&frame, FALSE, eva);
662 * The kernel may be using npx for copying or other
670 case T_PROTFLT: /* general protection fault */
671 case T_SEGNPFLT: /* segment not present fault */
673 * Invalid segment selectors and out of bounds
674 * %eip's and %esp's can be set up in user mode.
675 * This causes a fault in kernel mode when the
676 * kernel tries to return to user mode. We want
677 * to get this fault so that we can fix the
678 * problem here and not have to check all the
679 * selectors and pointers when the user changes
682 #define MAYBE_DORETI_FAULT(where, whereto) \
684 if (frame.tf_eip == (int)where) { \
685 frame.tf_eip = (int)whereto; \
690 * Since we don't save %gs across an interrupt
691 * frame this check must occur outside the intr
692 * nesting level check.
694 if (frame.tf_eip == (int)cpu_switch_load_gs) {
695 td->td_pcb->pcb_gs = 0;
696 MAKEMPSAFE(have_mplock);
700 if (mycpu->gd_intr_nesting_level == 0) {
702 * Invalid %fs's and %gs's can be created using
703 * procfs or PT_SETREGS or by invalidating the
704 * underlying LDT entry. This causes a fault
705 * in kernel mode when the kernel attempts to
706 * switch contexts. Lose the bad context
707 * (XXX) so that we can continue, and generate
710 MAYBE_DORETI_FAULT(doreti_iret,
712 MAYBE_DORETI_FAULT(doreti_popl_ds,
713 doreti_popl_ds_fault);
714 MAYBE_DORETI_FAULT(doreti_popl_es,
715 doreti_popl_es_fault);
716 MAYBE_DORETI_FAULT(doreti_popl_fs,
717 doreti_popl_fs_fault);
718 if (td->td_pcb->pcb_onfault) {
720 (register_t)td->td_pcb->pcb_onfault;
728 * PSL_NT can be set in user mode and isn't cleared
729 * automatically when the kernel is entered. This
730 * causes a TSS fault when the kernel attempts to
731 * `iret' because the TSS link is uninitialized. We
732 * want to get this fault so that we can fix the
733 * problem here and not every time the kernel is
736 if (frame.tf_eflags & PSL_NT) {
737 frame.tf_eflags &= ~PSL_NT;
742 case T_TRCTRAP: /* trace trap */
743 if (frame.tf_eip == (int)IDTVEC(syscall)) {
745 * We've just entered system mode via the
746 * syscall lcall. Continue single stepping
747 * silently until the syscall handler has
752 if (frame.tf_eip == (int)IDTVEC(syscall) + 1) {
754 * The syscall handler has now saved the
755 * flags. Stop single stepping it.
757 frame.tf_eflags &= ~PSL_T;
761 * Ignore debug register trace traps due to
762 * accesses in the user's address space, which
763 * can happen under several conditions such as
764 * if a user sets a watchpoint on a buffer and
765 * then passes that buffer to a system call.
766 * We still want to get TRCTRAPS for addresses
767 * in kernel space because that is useful when
768 * debugging the kernel.
770 if (user_dbreg_trap()) {
772 * Reset breakpoint bits because the
775 load_dr6(rdr6() & 0xfffffff0);
779 * Fall through (TRCTRAP kernel mode, kernel address)
783 * If DDB is enabled, let it handle the debugger trap.
784 * Otherwise, debugger traps "can't happen".
787 MAKEMPSAFE(have_mplock);
788 if (kdb_trap (type, 0, &frame))
795 MAKEMPSAFE(have_mplock);
798 # define TIMER_FREQ 1193182
802 static unsigned lastalert = 0;
804 if(time_second - lastalert > 10)
806 log(LOG_WARNING, "NMI: power fail\n");
807 sysbeep(TIMER_FREQ/880, hz);
808 lastalert = time_second;
813 #else /* !POWERFAIL_NMI */
814 /* machine/parity/power fail/"kitchen sink" faults */
815 if (isa_nmi(code) == 0) {
818 * NMI can be hooked up to a pushbutton
822 printf ("NMI ... going to debugger\n");
823 kdb_trap (type, 0, &frame);
827 } else if (panic_on_nmi == 0)
830 #endif /* POWERFAIL_NMI */
831 #endif /* NISA > 0 */
834 MAKEMPSAFE(have_mplock);
835 trap_fatal(&frame, eva);
839 /* Translate fault for emulators (e.g. Linux) */
840 if (*p->p_sysent->sv_transtrap)
841 i = (*p->p_sysent->sv_transtrap)(i, type);
843 MAKEMPSAFE(have_mplock);
844 trapsignal(p, i, ucode);
847 if (type <= MAX_TRAP_MSG) {
848 uprintf("fatal process exception: %s",
850 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
851 uprintf(", fault VA = 0x%lx", (u_long)eva);
858 if (ISPL(frame.tf_cs) == SEL_UPL)
859 KASSERT(td->td_mpcount == have_mplock, ("badmpcount trap/end from %p", (void *)frame.tf_eip));
861 userret(lp, &frame, sticks);
869 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
870 ("syscall: critical section count mismatch! %d/%d",
871 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
877 * This version doesn't allow a page fault to user space while
878 * in the kernel. The rest of the kernel needs to be made "safe"
879 * before this can be used. I think the only things remaining
880 * to be made safe is the process tracing/debugging code.
883 trap_pfault(frame, usermode, eva)
884 struct trapframe *frame;
889 struct vmspace *vm = NULL;
893 thread_t td = curthread;
894 struct proc *p = td->td_proc; /* may be NULL */
896 if (frame->tf_err & PGEX_W)
897 ftype = VM_PROT_WRITE;
899 ftype = VM_PROT_READ;
901 va = trunc_page(eva);
902 if (va < VM_MIN_KERNEL_ADDRESS) {
907 (!usermode && va < VM_MAXUSER_ADDRESS &&
908 (td->td_gd->gd_intr_nesting_level != 0 ||
909 td->td_pcb->pcb_onfault == NULL))) {
910 trap_fatal(frame, eva);
915 * This is a fault on non-kernel virtual memory.
916 * vm is initialized above to NULL. If curproc is NULL
917 * or curproc->p_vmspace is NULL the fault is fatal.
926 * Keep swapout from messing with us during this
932 * Grow the stack if necessary
934 /* grow_stack returns false only if va falls into
935 * a growable stack region and the stack growth
936 * fails. It returns true if va was not within
937 * a growable stack region, or if the stack
940 if (!grow_stack (p, va)) {
946 /* Fault in the user page: */
947 rv = vm_fault(map, va, ftype,
948 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
954 * Don't allow user-mode faults in kernel address space.
960 * Since we know that kernel virtual address addresses
961 * always have pte pages mapped, we just have to fault
964 rv = vm_fault(kernel_map, va, ftype, VM_FAULT_NORMAL);
967 if (rv == KERN_SUCCESS)
971 if (mtd->td_gd->gd_intr_nesting_level == 0 &&
972 td->td_pcb->pcb_onfault) {
973 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
976 trap_fatal(frame, eva);
980 /* kludge to pass faulting virtual address to sendsig */
983 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
988 trap_pfault(frame, usermode, eva)
989 struct trapframe *frame;
994 struct vmspace *vm = NULL;
998 thread_t td = curthread;
999 struct proc *p = td->td_proc;
1001 va = trunc_page(eva);
1002 if (va >= KERNBASE) {
1004 * Don't allow user-mode faults in kernel address space.
1005 * An exception: if the faulting address is the invalid
1006 * instruction entry in the IDT, then the Intel Pentium
1007 * F00F bug workaround was triggered, and we need to
1008 * treat it is as an illegal instruction, and not a page
1011 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
1012 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) {
1013 frame->tf_trapno = T_PRIVINFLT;
1023 * This is a fault on non-kernel virtual memory.
1024 * vm is initialized above to NULL. If curproc is NULL
1025 * or curproc->p_vmspace is NULL the fault is fatal.
1036 if (frame->tf_err & PGEX_W)
1037 ftype = VM_PROT_WRITE;
1039 ftype = VM_PROT_READ;
1041 if (map != kernel_map) {
1043 * Keep swapout from messing with us during this
1049 * Grow the stack if necessary
1051 /* grow_stack returns false only if va falls into
1052 * a growable stack region and the stack growth
1053 * fails. It returns true if va was not within
1054 * a growable stack region, or if the stack
1057 if (!grow_stack (p, va)) {
1063 /* Fault in the user page: */
1064 rv = vm_fault(map, va, ftype,
1065 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
1071 * Don't have to worry about process locking or stacks in the kernel.
1073 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
1076 if (rv == KERN_SUCCESS)
1080 if (td->td_gd->gd_intr_nesting_level == 0 &&
1081 td->td_pcb->pcb_onfault) {
1082 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
1085 trap_fatal(frame, eva);
1089 /* kludge to pass faulting virtual address to sendsig */
1090 frame->tf_err = eva;
1092 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
1096 trap_fatal(frame, eva)
1097 struct trapframe *frame;
1100 int code, type, ss, esp;
1101 struct soft_segment_descriptor softseg;
1103 code = frame->tf_err;
1104 type = frame->tf_trapno;
1105 sdtossd(&gdt[mycpu->gd_cpuid * NGDT + IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
1107 if (type <= MAX_TRAP_MSG)
1108 printf("\n\nFatal trap %d: %s while in %s mode\n",
1109 type, trap_msg[type],
1110 frame->tf_eflags & PSL_VM ? "vm86" :
1111 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
1113 /* three separate prints in case of a trap on an unmapped page */
1114 printf("mp_lock = %08x; ", mp_lock);
1115 printf("cpuid = %d; ", mycpu->gd_cpuid);
1116 printf("lapic.id = %08x\n", lapic.id);
1118 if (type == T_PAGEFLT) {
1119 printf("fault virtual address = 0x%x\n", eva);
1120 printf("fault code = %s %s, %s\n",
1121 code & PGEX_U ? "user" : "supervisor",
1122 code & PGEX_W ? "write" : "read",
1123 code & PGEX_P ? "protection violation" : "page not present");
1125 printf("instruction pointer = 0x%x:0x%x\n",
1126 frame->tf_cs & 0xffff, frame->tf_eip);
1127 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
1128 ss = frame->tf_ss & 0xffff;
1129 esp = frame->tf_esp;
1131 ss = GSEL(GDATA_SEL, SEL_KPL);
1132 esp = (int)&frame->tf_esp;
1134 printf("stack pointer = 0x%x:0x%x\n", ss, esp);
1135 printf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
1136 printf("code segment = base 0x%x, limit 0x%x, type 0x%x\n",
1137 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
1138 printf(" = DPL %d, pres %d, def32 %d, gran %d\n",
1139 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
1141 printf("processor eflags = ");
1142 if (frame->tf_eflags & PSL_T)
1143 printf("trace trap, ");
1144 if (frame->tf_eflags & PSL_I)
1145 printf("interrupt enabled, ");
1146 if (frame->tf_eflags & PSL_NT)
1147 printf("nested task, ");
1148 if (frame->tf_eflags & PSL_RF)
1150 if (frame->tf_eflags & PSL_VM)
1152 printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
1153 printf("current process = ");
1155 printf("%lu (%s)\n",
1156 (u_long)curproc->p_pid, curproc->p_comm ?
1157 curproc->p_comm : "");
1161 printf("current thread = pri %d ", curthread->td_pri);
1162 if (curthread->td_pri >= TDPRI_CRIT)
1168 * we probably SHOULD have stopped the other CPUs before now!
1169 * another CPU COULD have been touching cpl at this moment...
1171 printf(" <- SMP: XXX");
1180 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1183 printf("trap number = %d\n", type);
1184 if (type <= MAX_TRAP_MSG)
1185 panic("%s", trap_msg[type]);
1187 panic("unknown/reserved trap");
1191 * Double fault handler. Called when a fault occurs while writing
1192 * a frame for a trap/exception onto the stack. This usually occurs
1193 * when the stack overflows (such is the case with infinite recursion,
1196 * XXX Note that the current PTD gets replaced by IdlePTD when the
1197 * task switch occurs. This means that the stack that was active at
1198 * the time of the double fault is not available at <kstack> unless
1199 * the machine was idle when the double fault occurred. The downside
1200 * of this is that "trace <ebp>" in ddb won't work.
1205 struct mdglobaldata *gd = mdcpu;
1207 printf("\nFatal double fault:\n");
1208 printf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1209 printf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1210 printf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
1212 /* three separate prints in case of a trap on an unmapped page */
1213 printf("mp_lock = %08x; ", mp_lock);
1214 printf("cpuid = %d; ", mycpu->gd_cpuid);
1215 printf("lapic.id = %08x\n", lapic.id);
1217 panic("double fault");
1221 * Compensate for 386 brain damage (missing URKR).
1222 * This is a little simpler than the pagefault handler in trap() because
1223 * it the page tables have already been faulted in and high addresses
1224 * are thrown out early for other reasons.
1234 va = trunc_page((vm_offset_t)addr);
1236 * XXX - MAX is END. Changed > to >= for temp. fix.
1238 if (va >= VM_MAXUSER_ADDRESS)
1246 if (!grow_stack (p, va)) {
1252 * fault the data page
1254 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY);
1258 if (rv != KERN_SUCCESS)
1265 * syscall2 - MP aware system call request C handler
1267 * A system call is essentially treated as a trap except that the
1268 * MP lock is not held on entry or return. We are responsible for
1269 * obtaining the MP lock if necessary and for handling ASTs
1270 * (e.g. a task switch) prior to return.
1272 * In general, only simple access and manipulation of curproc and
1273 * the current stack is allowed without having to hold MP lock.
1275 * MPSAFE - note that large sections of this routine are run without
1280 syscall2(struct trapframe frame)
1282 struct thread *td = curthread;
1283 struct proc *p = td->td_proc;
1284 struct lwp *lp = td->td_lwp;
1286 struct sysent *callp;
1287 register_t orig_tf_eflags;
1292 int crit_count = td->td_pri & ~TDPRI_MASK;
1295 int have_mplock = 0;
1298 union sysunion args;
1301 if (ISPL(frame.tf_cs) != SEL_UPL) {
1309 KASSERT(td->td_mpcount == 0, ("badmpcount syscall2 from %p", (void *)frame.tf_eip));
1310 if (syscall_mpsafe == 0)
1311 MAKEMPSAFE(have_mplock);
1313 userenter(td); /* lazy raise our priority */
1315 sticks = (int)td->td_sticks;
1317 lp->lwp_md.md_regs = &frame;
1318 params = (caddr_t)frame.tf_esp + sizeof(int);
1319 code = frame.tf_eax;
1320 orig_tf_eflags = frame.tf_eflags;
1322 if (p->p_sysent->sv_prepsyscall) {
1323 (*p->p_sysent->sv_prepsyscall)(
1324 &frame, (int *)(&args.nosys.usrmsg + 1),
1328 * Need to check if this is a 32 bit or 64 bit syscall.
1329 * fuword is MP aware.
1331 if (code == SYS_syscall) {
1333 * Code is first argument, followed by actual args.
1335 code = fuword(params);
1336 params += sizeof(int);
1337 } else if (code == SYS___syscall) {
1339 * Like syscall, but code is a quad, so as to maintain
1340 * quad alignment for the rest of the arguments.
1342 code = fuword(params);
1343 params += sizeof(quad_t);
1347 code &= p->p_sysent->sv_mask;
1348 if (code >= p->p_sysent->sv_size)
1349 callp = &p->p_sysent->sv_table[0];
1351 callp = &p->p_sysent->sv_table[code];
1353 narg = callp->sy_narg & SYF_ARGMASK;
1356 * copyin is MP aware, but the tracing code is not
1358 if (narg && params) {
1359 error = copyin(params, (caddr_t)(&args.nosys.usrmsg + 1),
1360 narg * sizeof(register_t));
1363 if (KTRPOINT(td, KTR_SYSCALL)) {
1364 MAKEMPSAFE(have_mplock);
1366 ktrsyscall(p, code, narg,
1367 (void *)(&args.nosys.usrmsg + 1));
1375 if (KTRPOINT(td, KTR_SYSCALL)) {
1376 MAKEMPSAFE(have_mplock);
1377 ktrsyscall(p, code, narg, (void *)(&args.nosys.usrmsg + 1));
1382 * For traditional syscall code edx is left untouched when 32 bit
1383 * results are returned. Since edx is loaded from fds[1] when the
1384 * system call returns we pre-set it here.
1386 lwkt_initmsg(&args.lmsg, &td->td_msgport, 0,
1387 lwkt_cmd_op(code), lwkt_cmd_op_none);
1388 args.sysmsg_copyout = NULL;
1389 args.sysmsg_fds[0] = 0;
1390 args.sysmsg_fds[1] = frame.tf_edx;
1392 STOPEVENT(p, S_SCE, narg); /* MP aware */
1396 * Try to run the syscall without the MP lock if the syscall
1397 * is MP safe. We have to obtain the MP lock no matter what if
1400 if ((callp->sy_narg & SYF_MPSAFE) == 0)
1401 MAKEMPSAFE(have_mplock);
1404 error = (*callp->sy_call)(&args);
1407 * MP SAFE (we may or may not have the MP lock at this point)
1412 * Reinitialize proc pointer `p' as it may be different
1413 * if this is a child returning from fork syscall.
1416 lp = curthread->td_lwp;
1417 frame.tf_eax = args.sysmsg_fds[0];
1418 frame.tf_edx = args.sysmsg_fds[1];
1419 frame.tf_eflags &= ~PSL_C;
1423 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1424 * int 0x80 is 2 bytes. We saved this in tf_err.
1426 frame.tf_eip -= frame.tf_err;
1431 panic("Unexpected EASYNC return value (for now)");
1434 if (p->p_sysent->sv_errsize) {
1435 if (error >= p->p_sysent->sv_errsize)
1436 error = -1; /* XXX */
1438 error = p->p_sysent->sv_errtbl[error];
1440 frame.tf_eax = error;
1441 frame.tf_eflags |= PSL_C;
1446 * Traced syscall. trapsignal() is not MP aware.
1448 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1449 MAKEMPSAFE(have_mplock);
1450 frame.tf_eflags &= ~PSL_T;
1451 trapsignal(p, SIGTRAP, 0);
1455 * Handle reschedule and other end-of-syscall issues
1457 userret(lp, &frame, sticks);
1460 if (KTRPOINT(td, KTR_SYSRET)) {
1461 MAKEMPSAFE(have_mplock);
1462 ktrsysret(p, code, error, args.sysmsg_result);
1467 * This works because errno is findable through the
1468 * register set. If we ever support an emulation where this
1469 * is not the case, this code will need to be revisited.
1471 STOPEVENT(p, S_SCX, code);
1476 * Release the MP lock if we had to get it
1478 KASSERT(td->td_mpcount == have_mplock,
1479 ("badmpcount syscall2/end from %p", (void *)frame.tf_eip));
1484 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
1485 ("syscall: critical section count mismatch! %d/%d",
1486 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
1491 * free_sysun - Put an unused sysun on the free list.
1493 static __inline void
1494 free_sysun(struct thread *td, union sysunion *sysun)
1496 struct globaldata *gd = td->td_gd;
1498 crit_enter_quick(td);
1499 sysun->lmsg.opaque.ms_sysunnext = gd->gd_freesysun;
1500 gd->gd_freesysun = sysun;
1501 crit_exit_quick(td);
1505 * sendsys2 - MP aware system message request C handler
1508 sendsys2(struct trapframe frame)
1510 struct globaldata *gd;
1511 struct thread *td = curthread;
1512 struct proc *p = td->td_proc;
1513 struct lwp *lp = td->td_lwp;
1514 register_t orig_tf_eflags;
1515 struct sysent *callp;
1516 union sysunion *sysun = NULL;
1519 int have_mplock = 0;
1529 if (ISPL(frame.tf_cs) != SEL_UPL) {
1537 KASSERT(td->td_mpcount == 0,
1538 ("badmpcount sendsys2 from %p", (void *)frame.tf_eip));
1539 if (syscall_mpsafe == 0)
1540 MAKEMPSAFE(have_mplock);
1543 * access non-atomic field from critical section. p_sticks is
1544 * updated by the clock interrupt. Also use this opportunity
1545 * to lazy-raise our LWKT priority.
1548 sticks = td->td_sticks;
1550 lp->lwp_md.md_regs = &frame;
1551 orig_tf_eflags = frame.tf_eflags;
1555 * Extract the system call message. If msgsize is zero we are
1556 * blocking on a message and/or message port. If msgsize is -1
1557 * we are testing a message for completion or a message port for
1560 * The userland system call message size includes the size of the
1561 * userland lwkt_msg plus arguments. We load it into the userland
1562 * portion of our sysunion structure then we initialize the kerneland
1569 if ((msgsize = frame.tf_edx) < sizeof(struct lwkt_msg) ||
1570 msgsize > sizeof(union sysunion) - sizeof(struct sysmsg)) {
1576 * Obtain a sysun from our per-cpu cache or allocate a new one. Use
1577 * the opaque field to store the original (user) message pointer.
1578 * A critical section is necessary to interlock against interrupts
1579 * returning system messages to the thread cache.
1582 crit_enter_quick(td);
1583 if ((sysun = gd->gd_freesysun) != NULL)
1584 gd->gd_freesysun = sysun->lmsg.opaque.ms_sysunnext;
1586 sysun = malloc(sizeof(union sysunion), M_SYSMSG, M_WAITOK);
1587 crit_exit_quick(td);
1590 * Copy the user request into the kernel copy of the user request.
1592 umsg = (void *)frame.tf_ecx;
1593 error = copyin(umsg, &sysun->nosys.usrmsg, msgsize);
1596 if ((sysun->nosys.usrmsg.umsg.ms_flags & MSGF_ASYNC)) {
1601 if (max_sysmsg > 0 && lp->lwp_nsysmsg >= max_sysmsg) {
1608 * Initialize the kernel message from the copied-in data and
1609 * pull in appropriate flags from the userland message.
1611 * ms_abort_port is usually initialized in sendmsg/domsg, but since
1612 * we are not calling those functions (yet), we have to do it manually.
1614 lwkt_initmsg(&sysun->lmsg, &td->td_msgport, 0,
1615 sysun->nosys.usrmsg.umsg.ms_cmd,
1617 sysun->lmsg.ms_abort_port = sysun->lmsg.ms_reply_port;
1618 sysun->sysmsg_copyout = NULL;
1619 sysun->lmsg.opaque.ms_umsg = umsg;
1620 sysun->lmsg.ms_flags |= sysun->nosys.usrmsg.umsg.ms_flags & MSGF_ASYNC;
1623 * Extract the system call number, lookup the system call, and
1624 * set the default return value.
1626 code = (u_int)sysun->lmsg.ms_cmd.cm_op;
1627 /* We don't handle the syscall() syscall yet */
1630 free_sysun(td, sysun);
1633 if (code >= p->p_sysent->sv_size) {
1635 free_sysun(td, sysun);
1639 callp = &p->p_sysent->sv_table[code];
1641 narg = (msgsize - sizeof(struct lwkt_msg)) / sizeof(register_t);
1644 if (KTRPOINT(td, KTR_SYSCALL)) {
1645 MAKEMPSAFE(have_mplock);
1646 ktrsyscall(p, code, narg, (void *)(&sysun->nosys.usrmsg + 1));
1649 sysun->lmsg.u.ms_fds[0] = 0;
1650 sysun->lmsg.u.ms_fds[1] = 0;
1652 STOPEVENT(p, S_SCE, narg); /* MP aware */
1655 * Make the system call. An error code is always returned, results
1656 * are copied back via ms_result32 or ms_result64. YYY temporary
1657 * stage copy p_retval[] into ms_result32/64
1659 * NOTE! XXX if this is a child returning from a fork curproc
1660 * might be different. YYY huh? a child returning from a fork
1661 * should never 'return' from this call, it should go right to the
1662 * fork_trampoline function.
1664 * Obtain the MP lock if necessary.
1667 if ((callp->sy_narg & SYF_MPSAFE) == 0)
1668 MAKEMPSAFE(have_mplock);
1670 error = (*callp->sy_call)(sysun);
1671 gd = td->td_gd; /* RELOAD, might have switched cpus */
1675 * If a synchronous return copy p_retval to ms_result64 and return
1676 * the sysmsg to the free pool.
1678 * YYY Don't writeback message if execve() YYY
1680 sysun->nosys.usrmsg.umsg.ms_error = error;
1681 sysun->nosys.usrmsg.umsg.u.ms_fds[0] = sysun->lmsg.u.ms_fds[0];
1682 sysun->nosys.usrmsg.umsg.u.ms_fds[1] = sysun->lmsg.u.ms_fds[1];
1683 result = sysun->nosys.usrmsg.umsg.u.ms_fds[0]; /* for ktrace */
1684 if (error != 0 || code != SYS_execve) {
1686 error2 = copyout(&sysun->nosys.usrmsg.umsg.ms_copyout_start,
1687 &umsg->ms_copyout_start,
1692 if (error == EASYNC) {
1694 * Since only the current process ever messes with msgq,
1695 * we can safely manipulate it in parallel with the async
1698 TAILQ_INSERT_TAIL(&lp->lwp_sysmsgq, &sysun->sysmsg, msgq);
1700 error = (int)&sysun->sysmsg;
1702 free_sysun(td, sysun);
1705 frame.tf_eax = (register_t)error;
1708 * Traced syscall. trapsignal() is not MP aware.
1710 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1711 MAKEMPSAFE(have_mplock);
1712 frame.tf_eflags &= ~PSL_T;
1713 trapsignal(p, SIGTRAP, 0);
1717 * Handle reschedule and other end-of-syscall issues
1719 userret(lp, &frame, sticks);
1722 if (KTRPOINT(td, KTR_SYSRET)) {
1723 MAKEMPSAFE(have_mplock);
1724 ktrsysret(p, code, error, result);
1729 * This works because errno is findable through the
1730 * register set. If we ever support an emulation where this
1731 * is not the case, this code will need to be revisited.
1733 STOPEVENT(p, S_SCX, code);
1738 * Release the MP lock if we had to get it
1740 KASSERT(td->td_mpcount == have_mplock,
1741 ("badmpcount sendsys2/end from %p", (void *)frame.tf_eip));
1748 * waitsys2 - MP aware system message wait C handler
1751 waitsys2(struct trapframe frame)
1753 struct globaldata *gd;
1754 struct thread *td = curthread;
1755 struct proc *p = td->td_proc;
1756 struct lwp *lp = td->td_lwp;
1757 union sysunion *sysun = NULL;
1759 register_t orig_tf_eflags;
1760 int error = 0, result, sticks;
1762 int have_mplock = 0;
1767 if (ISPL(frame.tf_cs) != SEL_UPL) {
1775 KASSERT(td->td_mpcount == 0, ("badmpcount waitsys from %p",
1776 (void *)frame.tf_eip));
1777 if (syscall_mpsafe == 0)
1778 MAKEMPSAFE(have_mplock);
1782 * access non-atomic field from critical section. p_sticks is
1783 * updated by the clock interrupt. Also use this opportunity
1784 * to lazy-raise our LWKT priority.
1787 sticks = td->td_sticks;
1789 lp->lwp_md.md_regs = &frame;
1790 orig_tf_eflags = frame.tf_eflags;
1796 TAILQ_FOREACH(ptr, &lp->lwp_sysmsgq, msgq) {
1797 if ((void *)ptr == (void *)frame.tf_ecx) {
1798 sysun = (void *)sysmsg_wait(lp,
1799 (void *)frame.tf_ecx, 1);
1809 else if (frame.tf_eax) {
1810 printf("waitport/checkport only the default port is supported at the moment\n");
1815 switch(frame.tf_edx) {
1817 sysun = (void *)sysmsg_wait(lp, NULL, 0);
1820 sysun = (void *)sysmsg_wait(lp, NULL, 1);
1829 umsg = sysun->lmsg.opaque.ms_umsg;
1830 frame.tf_eax = (register_t)sysun;
1831 sysun->nosys.usrmsg.umsg.u.ms_fds[0] = sysun->lmsg.u.ms_fds[0];
1832 sysun->nosys.usrmsg.umsg.u.ms_fds[1] = sysun->lmsg.u.ms_fds[1];
1833 sysun->nosys.usrmsg.umsg.ms_error = sysun->lmsg.ms_error;
1834 error = sysun->lmsg.ms_error;
1835 result = sysun->lmsg.u.ms_fds[0]; /* for ktrace */
1836 error = copyout(&sysun->nosys.usrmsg.umsg.ms_copyout_start,
1837 &umsg->ms_copyout_start, ms_copyout_size);
1838 free_sysun(td, sysun);
1840 code = (u_int)sysun->lmsg.ms_cmd.cm_op;
1844 frame.tf_eax = error;
1846 * Traced syscall. trapsignal() is not MP aware.
1848 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1849 MAKEMPSAFE(have_mplock);
1850 frame.tf_eflags &= ~PSL_T;
1851 trapsignal(p, SIGTRAP, 0);
1855 * Handle reschedule and other end-of-syscall issues
1857 userret(lp, &frame, sticks);
1860 if (KTRPOINT(td, KTR_SYSRET)) {
1861 MAKEMPSAFE(have_mplock);
1862 ktrsysret(p, code, error, result);
1867 * This works because errno is findable through the
1868 * register set. If we ever support an emulation where this
1869 * is not the case, this code will need to be revisited.
1871 STOPEVENT(p, S_SCX, code);
1875 KASSERT(td->td_mpcount == have_mplock,
1876 ("badmpcount waitsys/end from %p", (void *)frame.tf_eip));
1883 * Simplified back end of syscall(), used when returning from fork()
1884 * directly into user mode. MP lock is held on entry and should be
1885 * released on return. This code will return back into the fork
1886 * trampoline code which then runs doreti.
1889 fork_return(p, frame)
1891 struct trapframe frame;
1895 KKASSERT(p->p_nthreads == 1);
1897 lp = LIST_FIRST(&p->p_lwps);
1899 frame.tf_eax = 0; /* Child returns zero */
1900 frame.tf_eflags &= ~PSL_C; /* success */
1904 * Newly forked processes are given a kernel priority. We have to
1905 * adjust the priority to a normal user priority and fake entry
1906 * into the kernel (call userenter()) to install a passive release
1907 * function just in case userret() decides to stop the process. This
1908 * can occur when ^Z races a fork. If we do not install the passive
1909 * release function the current process designation will not be
1910 * released when the thread goes to sleep.
1912 lwkt_setpri_self(TDPRI_USER_NORM);
1913 userenter(lp->lwp_thread);
1914 userret(lp, &frame, 0);
1916 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1917 ktrsysret(p, SYS_fork, 0, 0);
1919 p->p_flag |= P_PASSIVE_ACQ;
1921 p->p_flag &= ~P_PASSIVE_ACQ;
1923 KKASSERT(lp->lwp_thread->td_mpcount == 1);