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 1.71 2005/11/22 08:41:00 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);
322 * Handle a LWKT reschedule request first. Since our passive release
323 * is still in place we do not have to do anything special.
325 if (lwkt_resched_wanted())
329 * Acquire the current process designation if we do not own it.
330 * Note that acquire_curproc() does not reset the user reschedule
331 * bit on purpose, because we may need to accumulate over several
332 * threads waking up at the same time.
334 * NOTE: userland scheduler cruft: because processes are removed
335 * from the userland scheduler's queue we run through loops to try
336 * to figure out which is the best of [ existing, waking-up ]
339 if (lp != gd->gd_uschedcp) {
341 lp->lwp_proc->p_usched->acquire_curproc(lp);
342 /* We may have switched cpus on acquisition */
349 * Reduce our priority in preparation for a return to userland. If
350 * our passive release function was still in place, our priority was
351 * never raised and does not need to be reduced.
353 if (td->td_release == NULL)
354 lwkt_setpri_self(TDPRI_USER_NORM);
355 td->td_release = NULL;
358 * After reducing our priority there might be other kernel-level
359 * LWKTs that now have a greater priority. Run them as necessary.
360 * We don't have to worry about losing cpu to userland because
361 * we still control the current-process designation and we no longer
362 * have a passive release function installed.
364 if (lwkt_checkpri_self())
368 * If a userland reschedule is [still] pending we may not be the best
369 * selected process. Select a better one. If another LWKT resched
370 * is pending the trap will be re-entered.
372 if (user_resched_wanted()) {
373 lp->lwp_proc->p_usched->select_curproc(gd);
374 if (lp != gd->gd_uschedcp) {
375 lwkt_setpri_self(TDPRI_KERN_USER);
382 * Exception, fault, and trap interface to the kernel.
383 * This common code is called from assembly language IDT gate entry
384 * routines that prepare a suitable stack frame, and restore this
385 * frame after the exception has been processed.
387 * This function is also called from doreti in an interlock to handle ASTs.
388 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
390 * NOTE! We have to retrieve the fault address prior to obtaining the
391 * MP lock because get_mplock() may switch out. YYY cr2 really ought
392 * to be retrieved by the assembly code, not here.
394 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
395 * if an attempt is made to switch from a fast interrupt or IPI. This is
396 * necessary to properly take fatal kernel traps on SMP machines if
397 * get_mplock() has to block.
402 struct trapframe frame;
404 struct globaldata *gd = mycpu;
405 struct thread *td = gd->gd_curthread;
406 struct lwp *lp = td->td_lwp;
409 int i = 0, ucode = 0, type, code;
418 eva = (frame.tf_trapno == T_PAGEFLT ? rcr2() : 0);
419 ++gd->gd_trap_nesting_level;
420 MAKEMPSAFE(have_mplock);
421 trap_fatal(&frame, eva);
422 --gd->gd_trap_nesting_level;
428 ++gd->gd_trap_nesting_level;
429 if (frame.tf_trapno == T_PAGEFLT) {
431 * For some Cyrix CPUs, %cr2 is clobbered by interrupts.
432 * This problem is worked around by using an interrupt
433 * gate for the pagefault handler. We are finally ready
434 * to read %cr2 and then must reenable interrupts.
436 * XXX this should be in the switch statement, but the
437 * NO_FOOF_HACK and VM86 goto and ifdefs obfuscate the
438 * flow of control too much for this to be obviously
445 if (trap_mpsafe == 0)
446 MAKEMPSAFE(have_mplock);
449 --gd->gd_trap_nesting_level;
451 if (!(frame.tf_eflags & PSL_I)) {
453 * Buggy application or kernel code has disabled interrupts
454 * and then trapped. Enabling interrupts now is wrong, but
455 * it is better than running with interrupts disabled until
456 * they are accidentally enabled later.
458 type = frame.tf_trapno;
459 if (ISPL(frame.tf_cs)==SEL_UPL || (frame.tf_eflags & PSL_VM)) {
460 MAKEMPSAFE(have_mplock);
462 "pid %ld (%s): trap %d with interrupts disabled\n",
463 (long)curproc->p_pid, curproc->p_comm, type);
464 } else if (type != T_BPTFLT && type != T_TRCTRAP) {
466 * XXX not quite right, since this may be for a
467 * multiple fault in user mode.
469 MAKEMPSAFE(have_mplock);
470 printf("kernel trap %d with interrupts disabled\n",
476 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
479 type = frame.tf_trapno;
483 ASSERT_MP_LOCK_HELD(curthread);
484 if (frame.tf_eflags & PSL_VM &&
485 (type == T_PROTFLT || type == T_STKFLT)) {
487 KKASSERT(td->td_mpcount > 0);
489 i = vm86_emulate((struct vm86frame *)&frame);
491 KKASSERT(td->td_mpcount > 0);
495 * returns to original process
498 vm86_trap((struct vm86frame *)&frame,
501 vm86_trap((struct vm86frame *)&frame, 0);
503 KKASSERT(0); /* NOT REACHED */
509 * these traps want either a process context, or
510 * assume a normal userspace trap.
514 trap_fatal(&frame, eva);
517 type = T_BPTFLT; /* kernel breakpoint */
520 goto kernel_trap; /* normal kernel trap handling */
523 if ((ISPL(frame.tf_cs) == SEL_UPL) || (frame.tf_eflags & PSL_VM)) {
528 sticks = (int)td->td_sticks;
529 lp->lwp_md.md_regs = &frame;
532 case T_PRIVINFLT: /* privileged instruction fault */
537 case T_BPTFLT: /* bpt instruction fault */
538 case T_TRCTRAP: /* trace trap */
539 frame.tf_eflags &= ~PSL_T;
543 case T_ARITHTRAP: /* arithmetic trap */
548 case T_ASTFLT: /* Allow process switch */
549 mycpu->gd_cnt.v_soft++;
550 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
551 atomic_clear_int_nonlocked(&mycpu->gd_reqflags,
553 addupc_task(p, p->p_prof.pr_addr,
559 * The following two traps can happen in
560 * vm86 mode, and, if so, we want to handle
563 case T_PROTFLT: /* general protection fault */
564 case T_STKFLT: /* stack fault */
565 if (frame.tf_eflags & PSL_VM) {
566 i = vm86_emulate((struct vm86frame *)&frame);
573 case T_SEGNPFLT: /* segment not present fault */
574 case T_TSSFLT: /* invalid TSS fault */
575 case T_DOUBLEFLT: /* double fault */
577 ucode = code + BUS_SEGM_FAULT ;
581 case T_PAGEFLT: /* page fault */
582 MAKEMPSAFE(have_mplock);
583 i = trap_pfault(&frame, TRUE, eva);
586 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
596 case T_DIVIDE: /* integer divide fault */
603 MAKEMPSAFE(have_mplock);
605 goto handle_powerfail;
606 #else /* !POWERFAIL_NMI */
607 /* machine/parity/power fail/"kitchen sink" faults */
608 if (isa_nmi(code) == 0) {
611 * NMI can be hooked up to a pushbutton
615 printf ("NMI ... going to debugger\n");
616 kdb_trap (type, 0, &frame);
620 } else if (panic_on_nmi)
621 panic("NMI indicates hardware failure");
623 #endif /* POWERFAIL_NMI */
624 #endif /* NISA > 0 */
626 case T_OFLOW: /* integer overflow fault */
631 case T_BOUND: /* bounds check fault */
639 * The kernel may have switched out the FP unit's
640 * state, causing the user process to take a fault
641 * when it tries to use the FP unit. Restore the
647 if (!pmath_emulate) {
649 ucode = FPE_FPU_NP_TRAP;
652 i = (*pmath_emulate)(&frame);
654 if (!(frame.tf_eflags & PSL_T))
656 frame.tf_eflags &= ~PSL_T;
659 /* else ucode = emulator_only_knows() XXX */
662 case T_FPOPFLT: /* FPU operand fetch fault */
667 case T_XMMFLT: /* SIMD floating-point exception */
677 case T_PAGEFLT: /* page fault */
678 MAKEMPSAFE(have_mplock);
679 (void) trap_pfault(&frame, FALSE, eva);
685 * The kernel may be using npx for copying or other
693 case T_PROTFLT: /* general protection fault */
694 case T_SEGNPFLT: /* segment not present fault */
696 * Invalid segment selectors and out of bounds
697 * %eip's and %esp's can be set up in user mode.
698 * This causes a fault in kernel mode when the
699 * kernel tries to return to user mode. We want
700 * to get this fault so that we can fix the
701 * problem here and not have to check all the
702 * selectors and pointers when the user changes
705 #define MAYBE_DORETI_FAULT(where, whereto) \
707 if (frame.tf_eip == (int)where) { \
708 frame.tf_eip = (int)whereto; \
713 * Since we don't save %gs across an interrupt
714 * frame this check must occur outside the intr
715 * nesting level check.
717 if (frame.tf_eip == (int)cpu_switch_load_gs) {
718 td->td_pcb->pcb_gs = 0;
719 MAKEMPSAFE(have_mplock);
723 if (mycpu->gd_intr_nesting_level == 0) {
725 * Invalid %fs's and %gs's can be created using
726 * procfs or PT_SETREGS or by invalidating the
727 * underlying LDT entry. This causes a fault
728 * in kernel mode when the kernel attempts to
729 * switch contexts. Lose the bad context
730 * (XXX) so that we can continue, and generate
733 MAYBE_DORETI_FAULT(doreti_iret,
735 MAYBE_DORETI_FAULT(doreti_popl_ds,
736 doreti_popl_ds_fault);
737 MAYBE_DORETI_FAULT(doreti_popl_es,
738 doreti_popl_es_fault);
739 MAYBE_DORETI_FAULT(doreti_popl_fs,
740 doreti_popl_fs_fault);
741 if (td->td_pcb->pcb_onfault) {
743 (register_t)td->td_pcb->pcb_onfault;
751 * PSL_NT can be set in user mode and isn't cleared
752 * automatically when the kernel is entered. This
753 * causes a TSS fault when the kernel attempts to
754 * `iret' because the TSS link is uninitialized. We
755 * want to get this fault so that we can fix the
756 * problem here and not every time the kernel is
759 if (frame.tf_eflags & PSL_NT) {
760 frame.tf_eflags &= ~PSL_NT;
765 case T_TRCTRAP: /* trace trap */
766 if (frame.tf_eip == (int)IDTVEC(syscall)) {
768 * We've just entered system mode via the
769 * syscall lcall. Continue single stepping
770 * silently until the syscall handler has
775 if (frame.tf_eip == (int)IDTVEC(syscall) + 1) {
777 * The syscall handler has now saved the
778 * flags. Stop single stepping it.
780 frame.tf_eflags &= ~PSL_T;
784 * Ignore debug register trace traps due to
785 * accesses in the user's address space, which
786 * can happen under several conditions such as
787 * if a user sets a watchpoint on a buffer and
788 * then passes that buffer to a system call.
789 * We still want to get TRCTRAPS for addresses
790 * in kernel space because that is useful when
791 * debugging the kernel.
793 if (user_dbreg_trap()) {
795 * Reset breakpoint bits because the
798 load_dr6(rdr6() & 0xfffffff0);
802 * Fall through (TRCTRAP kernel mode, kernel address)
806 * If DDB is enabled, let it handle the debugger trap.
807 * Otherwise, debugger traps "can't happen".
810 MAKEMPSAFE(have_mplock);
811 if (kdb_trap (type, 0, &frame))
818 MAKEMPSAFE(have_mplock);
821 # define TIMER_FREQ 1193182
825 static unsigned lastalert = 0;
827 if(time_second - lastalert > 10)
829 log(LOG_WARNING, "NMI: power fail\n");
830 sysbeep(TIMER_FREQ/880, hz);
831 lastalert = time_second;
836 #else /* !POWERFAIL_NMI */
837 /* machine/parity/power fail/"kitchen sink" faults */
838 if (isa_nmi(code) == 0) {
841 * NMI can be hooked up to a pushbutton
845 printf ("NMI ... going to debugger\n");
846 kdb_trap (type, 0, &frame);
850 } else if (panic_on_nmi == 0)
853 #endif /* POWERFAIL_NMI */
854 #endif /* NISA > 0 */
857 MAKEMPSAFE(have_mplock);
858 trap_fatal(&frame, eva);
862 /* Translate fault for emulators (e.g. Linux) */
863 if (*p->p_sysent->sv_transtrap)
864 i = (*p->p_sysent->sv_transtrap)(i, type);
866 MAKEMPSAFE(have_mplock);
867 trapsignal(p, i, ucode);
870 if (type <= MAX_TRAP_MSG) {
871 uprintf("fatal process exception: %s",
873 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
874 uprintf(", fault VA = 0x%lx", (u_long)eva);
881 if (ISPL(frame.tf_cs) == SEL_UPL)
882 KASSERT(td->td_mpcount == have_mplock, ("badmpcount trap/end from %p", (void *)frame.tf_eip));
884 userret(lp, &frame, sticks);
895 * This version doesn't allow a page fault to user space while
896 * in the kernel. The rest of the kernel needs to be made "safe"
897 * before this can be used. I think the only things remaining
898 * to be made safe are the iBCS2 code and the process tracing/
902 trap_pfault(frame, usermode, eva)
903 struct trapframe *frame;
908 struct vmspace *vm = NULL;
912 thread_t td = curthread;
913 struct proc *p = td->td_proc; /* may be NULL */
915 if (frame->tf_err & PGEX_W)
916 ftype = VM_PROT_WRITE;
918 ftype = VM_PROT_READ;
920 va = trunc_page(eva);
921 if (va < VM_MIN_KERNEL_ADDRESS) {
926 (!usermode && va < VM_MAXUSER_ADDRESS &&
927 (td->td_gd->gd_intr_nesting_level != 0 ||
928 td->td_pcb->pcb_onfault == NULL))) {
929 trap_fatal(frame, eva);
934 * This is a fault on non-kernel virtual memory.
935 * vm is initialized above to NULL. If curproc is NULL
936 * or curproc->p_vmspace is NULL the fault is fatal.
945 * Keep swapout from messing with us during this
951 * Grow the stack if necessary
953 /* grow_stack returns false only if va falls into
954 * a growable stack region and the stack growth
955 * fails. It returns true if va was not within
956 * a growable stack region, or if the stack
959 if (!grow_stack (p, va)) {
965 /* Fault in the user page: */
966 rv = vm_fault(map, va, ftype,
967 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
973 * Don't allow user-mode faults in kernel address space.
979 * Since we know that kernel virtual address addresses
980 * always have pte pages mapped, we just have to fault
983 rv = vm_fault(kernel_map, va, ftype, VM_FAULT_NORMAL);
986 if (rv == KERN_SUCCESS)
990 if (mtd->td_gd->gd_intr_nesting_level == 0 &&
991 td->td_pcb->pcb_onfault) {
992 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
995 trap_fatal(frame, eva);
999 /* kludge to pass faulting virtual address to sendsig */
1000 frame->tf_err = eva;
1002 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
1007 trap_pfault(frame, usermode, eva)
1008 struct trapframe *frame;
1013 struct vmspace *vm = NULL;
1017 thread_t td = curthread;
1018 struct proc *p = td->td_proc;
1020 va = trunc_page(eva);
1021 if (va >= KERNBASE) {
1023 * Don't allow user-mode faults in kernel address space.
1024 * An exception: if the faulting address is the invalid
1025 * instruction entry in the IDT, then the Intel Pentium
1026 * F00F bug workaround was triggered, and we need to
1027 * treat it is as an illegal instruction, and not a page
1030 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
1031 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) {
1032 frame->tf_trapno = T_PRIVINFLT;
1042 * This is a fault on non-kernel virtual memory.
1043 * vm is initialized above to NULL. If curproc is NULL
1044 * or curproc->p_vmspace is NULL the fault is fatal.
1055 if (frame->tf_err & PGEX_W)
1056 ftype = VM_PROT_WRITE;
1058 ftype = VM_PROT_READ;
1060 if (map != kernel_map) {
1062 * Keep swapout from messing with us during this
1068 * Grow the stack if necessary
1070 /* grow_stack returns false only if va falls into
1071 * a growable stack region and the stack growth
1072 * fails. It returns true if va was not within
1073 * a growable stack region, or if the stack
1076 if (!grow_stack (p, va)) {
1082 /* Fault in the user page: */
1083 rv = vm_fault(map, va, ftype,
1084 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
1090 * Don't have to worry about process locking or stacks in the kernel.
1092 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
1095 if (rv == KERN_SUCCESS)
1099 if (td->td_gd->gd_intr_nesting_level == 0 &&
1100 td->td_pcb->pcb_onfault) {
1101 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
1104 trap_fatal(frame, eva);
1108 /* kludge to pass faulting virtual address to sendsig */
1109 frame->tf_err = eva;
1111 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
1115 trap_fatal(frame, eva)
1116 struct trapframe *frame;
1119 int code, type, ss, esp;
1120 struct soft_segment_descriptor softseg;
1122 code = frame->tf_err;
1123 type = frame->tf_trapno;
1124 sdtossd(&gdt[mycpu->gd_cpuid * NGDT + IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
1126 if (type <= MAX_TRAP_MSG)
1127 printf("\n\nFatal trap %d: %s while in %s mode\n",
1128 type, trap_msg[type],
1129 frame->tf_eflags & PSL_VM ? "vm86" :
1130 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
1132 /* three separate prints in case of a trap on an unmapped page */
1133 printf("mp_lock = %08x; ", mp_lock);
1134 printf("cpuid = %d; ", mycpu->gd_cpuid);
1135 printf("lapic.id = %08x\n", lapic.id);
1137 if (type == T_PAGEFLT) {
1138 printf("fault virtual address = 0x%x\n", eva);
1139 printf("fault code = %s %s, %s\n",
1140 code & PGEX_U ? "user" : "supervisor",
1141 code & PGEX_W ? "write" : "read",
1142 code & PGEX_P ? "protection violation" : "page not present");
1144 printf("instruction pointer = 0x%x:0x%x\n",
1145 frame->tf_cs & 0xffff, frame->tf_eip);
1146 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
1147 ss = frame->tf_ss & 0xffff;
1148 esp = frame->tf_esp;
1150 ss = GSEL(GDATA_SEL, SEL_KPL);
1151 esp = (int)&frame->tf_esp;
1153 printf("stack pointer = 0x%x:0x%x\n", ss, esp);
1154 printf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
1155 printf("code segment = base 0x%x, limit 0x%x, type 0x%x\n",
1156 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
1157 printf(" = DPL %d, pres %d, def32 %d, gran %d\n",
1158 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
1160 printf("processor eflags = ");
1161 if (frame->tf_eflags & PSL_T)
1162 printf("trace trap, ");
1163 if (frame->tf_eflags & PSL_I)
1164 printf("interrupt enabled, ");
1165 if (frame->tf_eflags & PSL_NT)
1166 printf("nested task, ");
1167 if (frame->tf_eflags & PSL_RF)
1169 if (frame->tf_eflags & PSL_VM)
1171 printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
1172 printf("current process = ");
1174 printf("%lu (%s)\n",
1175 (u_long)curproc->p_pid, curproc->p_comm ?
1176 curproc->p_comm : "");
1180 printf("current thread = pri %d ", curthread->td_pri);
1181 if (curthread->td_pri >= TDPRI_CRIT)
1187 * we probably SHOULD have stopped the other CPUs before now!
1188 * another CPU COULD have been touching cpl at this moment...
1190 printf(" <- SMP: XXX");
1199 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1202 printf("trap number = %d\n", type);
1203 if (type <= MAX_TRAP_MSG)
1204 panic("%s", trap_msg[type]);
1206 panic("unknown/reserved trap");
1210 * Double fault handler. Called when a fault occurs while writing
1211 * a frame for a trap/exception onto the stack. This usually occurs
1212 * when the stack overflows (such is the case with infinite recursion,
1215 * XXX Note that the current PTD gets replaced by IdlePTD when the
1216 * task switch occurs. This means that the stack that was active at
1217 * the time of the double fault is not available at <kstack> unless
1218 * the machine was idle when the double fault occurred. The downside
1219 * of this is that "trace <ebp>" in ddb won't work.
1224 struct mdglobaldata *gd = mdcpu;
1226 printf("\nFatal double fault:\n");
1227 printf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1228 printf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1229 printf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
1231 /* three separate prints in case of a trap on an unmapped page */
1232 printf("mp_lock = %08x; ", mp_lock);
1233 printf("cpuid = %d; ", mycpu->gd_cpuid);
1234 printf("lapic.id = %08x\n", lapic.id);
1236 panic("double fault");
1240 * Compensate for 386 brain damage (missing URKR).
1241 * This is a little simpler than the pagefault handler in trap() because
1242 * it the page tables have already been faulted in and high addresses
1243 * are thrown out early for other reasons.
1253 va = trunc_page((vm_offset_t)addr);
1255 * XXX - MAX is END. Changed > to >= for temp. fix.
1257 if (va >= VM_MAXUSER_ADDRESS)
1265 if (!grow_stack (p, va)) {
1271 * fault the data page
1273 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY);
1277 if (rv != KERN_SUCCESS)
1284 * syscall2 - MP aware system call request C handler
1286 * A system call is essentially treated as a trap except that the
1287 * MP lock is not held on entry or return. We are responsible for
1288 * obtaining the MP lock if necessary and for handling ASTs
1289 * (e.g. a task switch) prior to return.
1291 * In general, only simple access and manipulation of curproc and
1292 * the current stack is allowed without having to hold MP lock.
1294 * MPSAFE - note that large sections of this routine are run without
1299 syscall2(struct trapframe frame)
1301 struct thread *td = curthread;
1302 struct proc *p = td->td_proc;
1303 struct lwp *lp = td->td_lwp;
1305 struct sysent *callp;
1306 register_t orig_tf_eflags;
1311 int have_mplock = 0;
1314 union sysunion args;
1317 if (ISPL(frame.tf_cs) != SEL_UPL) {
1325 KASSERT(td->td_mpcount == 0, ("badmpcount syscall2 from %p", (void *)frame.tf_eip));
1326 if (syscall_mpsafe == 0)
1327 MAKEMPSAFE(have_mplock);
1329 userenter(td); /* lazy raise our priority */
1331 sticks = (int)td->td_sticks;
1333 lp->lwp_md.md_regs = &frame;
1334 params = (caddr_t)frame.tf_esp + sizeof(int);
1335 code = frame.tf_eax;
1336 orig_tf_eflags = frame.tf_eflags;
1338 if (p->p_sysent->sv_prepsyscall) {
1339 (*p->p_sysent->sv_prepsyscall)(
1340 &frame, (int *)(&args.nosys.usrmsg + 1),
1344 * Need to check if this is a 32 bit or 64 bit syscall.
1345 * fuword is MP aware.
1347 if (code == SYS_syscall) {
1349 * Code is first argument, followed by actual args.
1351 code = fuword(params);
1352 params += sizeof(int);
1353 } else if (code == SYS___syscall) {
1355 * Like syscall, but code is a quad, so as to maintain
1356 * quad alignment for the rest of the arguments.
1358 code = fuword(params);
1359 params += sizeof(quad_t);
1363 code &= p->p_sysent->sv_mask;
1364 if (code >= p->p_sysent->sv_size)
1365 callp = &p->p_sysent->sv_table[0];
1367 callp = &p->p_sysent->sv_table[code];
1369 narg = callp->sy_narg & SYF_ARGMASK;
1372 * copyin is MP aware, but the tracing code is not
1374 if (narg && params) {
1375 error = copyin(params, (caddr_t)(&args.nosys.usrmsg + 1),
1376 narg * sizeof(register_t));
1379 if (KTRPOINT(td, KTR_SYSCALL)) {
1380 MAKEMPSAFE(have_mplock);
1382 ktrsyscall(p->p_tracep, code, narg,
1383 (void *)(&args.nosys.usrmsg + 1));
1391 if (KTRPOINT(td, KTR_SYSCALL)) {
1392 MAKEMPSAFE(have_mplock);
1393 ktrsyscall(p->p_tracep, code, narg, (void *)(&args.nosys.usrmsg + 1));
1398 * For traditional syscall code edx is left untouched when 32 bit
1399 * results are returned. Since edx is loaded from fds[1] when the
1400 * system call returns we pre-set it here.
1402 lwkt_initmsg(&args.lmsg, &td->td_msgport, 0,
1403 lwkt_cmd_op(code), lwkt_cmd_op_none);
1404 args.sysmsg_copyout = NULL;
1405 args.sysmsg_fds[0] = 0;
1406 args.sysmsg_fds[1] = frame.tf_edx;
1408 STOPEVENT(p, S_SCE, narg); /* MP aware */
1412 * Try to run the syscall without the MP lock if the syscall
1413 * is MP safe. We have to obtain the MP lock no matter what if
1416 if ((callp->sy_narg & SYF_MPSAFE) == 0)
1417 MAKEMPSAFE(have_mplock);
1420 error = (*callp->sy_call)(&args);
1423 * MP SAFE (we may or may not have the MP lock at this point)
1428 * Reinitialize proc pointer `p' as it may be different
1429 * if this is a child returning from fork syscall.
1432 lp = curthread->td_lwp;
1433 frame.tf_eax = args.sysmsg_fds[0];
1434 frame.tf_edx = args.sysmsg_fds[1];
1435 frame.tf_eflags &= ~PSL_C;
1439 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1440 * int 0x80 is 2 bytes. We saved this in tf_err.
1442 frame.tf_eip -= frame.tf_err;
1447 panic("Unexpected EASYNC return value (for now)");
1450 if (p->p_sysent->sv_errsize) {
1451 if (error >= p->p_sysent->sv_errsize)
1452 error = -1; /* XXX */
1454 error = p->p_sysent->sv_errtbl[error];
1456 frame.tf_eax = error;
1457 frame.tf_eflags |= PSL_C;
1462 * Traced syscall. trapsignal() is not MP aware.
1464 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1465 MAKEMPSAFE(have_mplock);
1466 frame.tf_eflags &= ~PSL_T;
1467 trapsignal(p, SIGTRAP, 0);
1471 * Handle reschedule and other end-of-syscall issues
1473 userret(lp, &frame, sticks);
1476 if (KTRPOINT(td, KTR_SYSRET)) {
1477 MAKEMPSAFE(have_mplock);
1478 ktrsysret(p->p_tracep, code, error, args.sysmsg_result);
1483 * This works because errno is findable through the
1484 * register set. If we ever support an emulation where this
1485 * is not the case, this code will need to be revisited.
1487 STOPEVENT(p, S_SCX, code);
1492 * Release the MP lock if we had to get it
1494 KASSERT(td->td_mpcount == have_mplock,
1495 ("badmpcount syscall2/end from %p", (void *)frame.tf_eip));
1502 * free_sysun - Put an unused sysun on the free list.
1504 static __inline void
1505 free_sysun(struct thread *td, union sysunion *sysun)
1507 struct globaldata *gd = td->td_gd;
1509 crit_enter_quick(td);
1510 sysun->lmsg.opaque.ms_sysunnext = gd->gd_freesysun;
1511 gd->gd_freesysun = sysun;
1512 crit_exit_quick(td);
1516 * sendsys2 - MP aware system message request C handler
1519 sendsys2(struct trapframe frame)
1521 struct globaldata *gd;
1522 struct thread *td = curthread;
1523 struct proc *p = td->td_proc;
1524 struct lwp *lp = td->td_lwp;
1525 register_t orig_tf_eflags;
1526 struct sysent *callp;
1527 union sysunion *sysun = NULL;
1530 int have_mplock = 0;
1540 if (ISPL(frame.tf_cs) != SEL_UPL) {
1548 KASSERT(td->td_mpcount == 0,
1549 ("badmpcount sendsys2 from %p", (void *)frame.tf_eip));
1550 if (syscall_mpsafe == 0)
1551 MAKEMPSAFE(have_mplock);
1554 * access non-atomic field from critical section. p_sticks is
1555 * updated by the clock interrupt. Also use this opportunity
1556 * to lazy-raise our LWKT priority.
1559 sticks = td->td_sticks;
1561 lp->lwp_md.md_regs = &frame;
1562 orig_tf_eflags = frame.tf_eflags;
1566 * Extract the system call message. If msgsize is zero we are
1567 * blocking on a message and/or message port. If msgsize is -1
1568 * we are testing a message for completion or a message port for
1571 * The userland system call message size includes the size of the
1572 * userland lwkt_msg plus arguments. We load it into the userland
1573 * portion of our sysunion structure then we initialize the kerneland
1580 if ((msgsize = frame.tf_edx) < sizeof(struct lwkt_msg) ||
1581 msgsize > sizeof(union sysunion) - sizeof(struct sysmsg)) {
1587 * Obtain a sysun from our per-cpu cache or allocate a new one. Use
1588 * the opaque field to store the original (user) message pointer.
1589 * A critical section is necessary to interlock against interrupts
1590 * returning system messages to the thread cache.
1593 crit_enter_quick(td);
1594 if ((sysun = gd->gd_freesysun) != NULL)
1595 gd->gd_freesysun = sysun->lmsg.opaque.ms_sysunnext;
1597 sysun = malloc(sizeof(union sysunion), M_SYSMSG, M_WAITOK);
1598 crit_exit_quick(td);
1601 * Copy the user request into the kernel copy of the user request.
1603 umsg = (void *)frame.tf_ecx;
1604 error = copyin(umsg, &sysun->nosys.usrmsg, msgsize);
1607 if ((sysun->nosys.usrmsg.umsg.ms_flags & MSGF_ASYNC)) {
1612 if (max_sysmsg > 0 && lp->lwp_nsysmsg >= max_sysmsg) {
1619 * Initialize the kernel message from the copied-in data and
1620 * pull in appropriate flags from the userland message.
1622 * ms_abort_port is usually initialized in sendmsg/domsg, but since
1623 * we are not calling those functions (yet), we have to do it manually.
1625 lwkt_initmsg(&sysun->lmsg, &td->td_msgport, 0,
1626 sysun->nosys.usrmsg.umsg.ms_cmd,
1628 sysun->lmsg.ms_abort_port = sysun->lmsg.ms_reply_port;
1629 sysun->sysmsg_copyout = NULL;
1630 sysun->lmsg.opaque.ms_umsg = umsg;
1631 sysun->lmsg.ms_flags |= sysun->nosys.usrmsg.umsg.ms_flags & MSGF_ASYNC;
1634 * Extract the system call number, lookup the system call, and
1635 * set the default return value.
1637 code = (u_int)sysun->lmsg.ms_cmd.cm_op;
1638 /* We don't handle the syscall() syscall yet */
1641 free_sysun(td, sysun);
1644 if (code >= p->p_sysent->sv_size) {
1646 free_sysun(td, sysun);
1650 callp = &p->p_sysent->sv_table[code];
1652 narg = (msgsize - sizeof(struct lwkt_msg)) / sizeof(register_t);
1655 if (KTRPOINT(td, KTR_SYSCALL)) {
1656 MAKEMPSAFE(have_mplock);
1657 ktrsyscall(p->p_tracep, code, narg, (void *)(&sysun->nosys.usrmsg + 1));
1660 sysun->lmsg.u.ms_fds[0] = 0;
1661 sysun->lmsg.u.ms_fds[1] = 0;
1663 STOPEVENT(p, S_SCE, narg); /* MP aware */
1666 * Make the system call. An error code is always returned, results
1667 * are copied back via ms_result32 or ms_result64. YYY temporary
1668 * stage copy p_retval[] into ms_result32/64
1670 * NOTE! XXX if this is a child returning from a fork curproc
1671 * might be different. YYY huh? a child returning from a fork
1672 * should never 'return' from this call, it should go right to the
1673 * fork_trampoline function.
1675 * Obtain the MP lock if necessary.
1678 if ((callp->sy_narg & SYF_MPSAFE) == 0)
1679 MAKEMPSAFE(have_mplock);
1681 error = (*callp->sy_call)(sysun);
1682 gd = td->td_gd; /* RELOAD, might have switched cpus */
1686 * If a synchronous return copy p_retval to ms_result64 and return
1687 * the sysmsg to the free pool.
1689 * YYY Don't writeback message if execve() YYY
1691 sysun->nosys.usrmsg.umsg.ms_error = error;
1692 sysun->nosys.usrmsg.umsg.u.ms_fds[0] = sysun->lmsg.u.ms_fds[0];
1693 sysun->nosys.usrmsg.umsg.u.ms_fds[1] = sysun->lmsg.u.ms_fds[1];
1694 result = sysun->nosys.usrmsg.umsg.u.ms_fds[0]; /* for ktrace */
1695 if (error != 0 || code != SYS_execve) {
1697 error2 = copyout(&sysun->nosys.usrmsg.umsg.ms_copyout_start,
1698 &umsg->ms_copyout_start,
1703 if (error == EASYNC) {
1705 * Since only the current process ever messes with msgq,
1706 * we can safely manipulate it in parallel with the async
1709 TAILQ_INSERT_TAIL(&lp->lwp_sysmsgq, &sysun->sysmsg, msgq);
1711 error = (int)&sysun->sysmsg;
1713 free_sysun(td, sysun);
1716 frame.tf_eax = (register_t)error;
1719 * Traced syscall. trapsignal() is not MP aware.
1721 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1722 MAKEMPSAFE(have_mplock);
1723 frame.tf_eflags &= ~PSL_T;
1724 trapsignal(p, SIGTRAP, 0);
1728 * Handle reschedule and other end-of-syscall issues
1730 userret(lp, &frame, sticks);
1733 if (KTRPOINT(td, KTR_SYSRET)) {
1734 MAKEMPSAFE(have_mplock);
1735 ktrsysret(p->p_tracep, code, error, result);
1740 * This works because errno is findable through the
1741 * register set. If we ever support an emulation where this
1742 * is not the case, this code will need to be revisited.
1744 STOPEVENT(p, S_SCX, code);
1749 * Release the MP lock if we had to get it
1751 KASSERT(td->td_mpcount == have_mplock,
1752 ("badmpcount sendsys2/end from %p", (void *)frame.tf_eip));
1759 * waitsys2 - MP aware system message wait C handler
1762 waitsys2(struct trapframe frame)
1764 struct globaldata *gd;
1765 struct thread *td = curthread;
1766 struct proc *p = td->td_proc;
1767 struct lwp *lp = td->td_lwp;
1768 union sysunion *sysun = NULL;
1770 register_t orig_tf_eflags;
1771 int error = 0, result, sticks;
1773 int have_mplock = 0;
1778 if (ISPL(frame.tf_cs) != SEL_UPL) {
1786 KASSERT(td->td_mpcount == 0, ("badmpcount waitsys from %p",
1787 (void *)frame.tf_eip));
1788 if (syscall_mpsafe == 0)
1789 MAKEMPSAFE(have_mplock);
1793 * access non-atomic field from critical section. p_sticks is
1794 * updated by the clock interrupt. Also use this opportunity
1795 * to lazy-raise our LWKT priority.
1798 sticks = td->td_sticks;
1800 lp->lwp_md.md_regs = &frame;
1801 orig_tf_eflags = frame.tf_eflags;
1807 TAILQ_FOREACH(ptr, &lp->lwp_sysmsgq, msgq) {
1808 if ((void *)ptr == (void *)frame.tf_ecx) {
1809 sysun = (void *)sysmsg_wait(lp,
1810 (void *)frame.tf_ecx, 1);
1820 else if (frame.tf_eax) {
1821 printf("waitport/checkport only the default port is supported at the moment\n");
1826 switch(frame.tf_edx) {
1828 sysun = (void *)sysmsg_wait(lp, NULL, 0);
1831 sysun = (void *)sysmsg_wait(lp, NULL, 1);
1840 umsg = sysun->lmsg.opaque.ms_umsg;
1841 frame.tf_eax = (register_t)sysun;
1842 sysun->nosys.usrmsg.umsg.u.ms_fds[0] = sysun->lmsg.u.ms_fds[0];
1843 sysun->nosys.usrmsg.umsg.u.ms_fds[1] = sysun->lmsg.u.ms_fds[1];
1844 sysun->nosys.usrmsg.umsg.ms_error = sysun->lmsg.ms_error;
1845 error = sysun->lmsg.ms_error;
1846 result = sysun->lmsg.u.ms_fds[0]; /* for ktrace */
1847 error = copyout(&sysun->nosys.usrmsg.umsg.ms_copyout_start,
1848 &umsg->ms_copyout_start, ms_copyout_size);
1849 free_sysun(td, sysun);
1851 code = (u_int)sysun->lmsg.ms_cmd.cm_op;
1855 frame.tf_eax = error;
1857 * Traced syscall. trapsignal() is not MP aware.
1859 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1860 MAKEMPSAFE(have_mplock);
1861 frame.tf_eflags &= ~PSL_T;
1862 trapsignal(p, SIGTRAP, 0);
1866 * Handle reschedule and other end-of-syscall issues
1868 userret(lp, &frame, sticks);
1871 if (KTRPOINT(td, KTR_SYSRET)) {
1872 MAKEMPSAFE(have_mplock);
1873 ktrsysret(p->p_tracep, code, error, result);
1878 * This works because errno is findable through the
1879 * register set. If we ever support an emulation where this
1880 * is not the case, this code will need to be revisited.
1882 STOPEVENT(p, S_SCX, code);
1886 KASSERT(td->td_mpcount == 1, ("badmpcount waitsys/end from %p",
1887 (void *)frame.tf_eip));
1894 * Simplified back end of syscall(), used when returning from fork()
1895 * directly into user mode. MP lock is held on entry and should be
1896 * released on return. This code will return back into the fork
1897 * trampoline code which then runs doreti.
1900 fork_return(p, frame)
1902 struct trapframe frame;
1906 KKASSERT(p->p_nthreads == 1);
1908 lp = LIST_FIRST(&p->p_lwps);
1910 frame.tf_eax = 0; /* Child returns zero */
1911 frame.tf_eflags &= ~PSL_C; /* success */
1915 * Newly forked processes are given a kernel priority. We have to
1916 * adjust the priority to a normal user priority and fake entry
1917 * into the kernel (call userenter()) to install a passive release
1918 * function just in case userret() decides to stop the process. This
1919 * can occur when ^Z races a fork. If we do not install the passive
1920 * release function the current process designation will not be
1921 * released when the thread goes to sleep.
1923 lwkt_setpri_self(TDPRI_USER_NORM);
1924 userenter(lp->lwp_thread);
1925 userret(lp, &frame, 0);
1927 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1928 ktrsysret(p->p_tracep, SYS_fork, 0, 0);
1930 p->p_flag |= P_PASSIVE_ACQ;
1932 p->p_flag &= ~P_PASSIVE_ACQ;
1934 KKASSERT(lp->lwp_thread->td_mpcount == 1);