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 1.60 2005/07/19 19:25:43 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>
89 #include <machine/smp.h>
91 #include <machine/tss.h>
92 #include <machine/globaldata.h>
94 #include <i386/isa/intr_machdep.h>
97 #include <sys/syslog.h>
98 #include <machine/clock.h>
101 #include <machine/vm86.h>
104 #include <sys/msgport2.h>
105 #include <sys/thread2.h>
107 int (*pmath_emulate) (struct trapframe *);
109 extern void trap (struct trapframe frame);
110 extern int trapwrite (unsigned addr);
111 extern void syscall2 (struct trapframe frame);
112 extern void sendsys2 (struct trapframe frame);
113 extern void waitsys2 (struct trapframe frame);
115 static int trap_pfault (struct trapframe *, int, vm_offset_t);
116 static void trap_fatal (struct trapframe *, vm_offset_t);
117 void dblfault_handler (void);
119 extern inthand_t IDTVEC(syscall);
121 #define MAX_TRAP_MSG 28
122 static char *trap_msg[] = {
124 "privileged instruction fault", /* 1 T_PRIVINFLT */
126 "breakpoint instruction fault", /* 3 T_BPTFLT */
129 "arithmetic trap", /* 6 T_ARITHTRAP */
130 "system forced exception", /* 7 T_ASTFLT */
132 "general protection fault", /* 9 T_PROTFLT */
133 "trace trap", /* 10 T_TRCTRAP */
135 "page fault", /* 12 T_PAGEFLT */
137 "alignment fault", /* 14 T_ALIGNFLT */
141 "integer divide fault", /* 18 T_DIVIDE */
142 "non-maskable interrupt trap", /* 19 T_NMI */
143 "overflow trap", /* 20 T_OFLOW */
144 "FPU bounds check fault", /* 21 T_BOUND */
145 "FPU device not available", /* 22 T_DNA */
146 "double fault", /* 23 T_DOUBLEFLT */
147 "FPU operand fetch fault", /* 24 T_FPOPFLT */
148 "invalid TSS fault", /* 25 T_TSSFLT */
149 "segment not present fault", /* 26 T_SEGNPFLT */
150 "stack fault", /* 27 T_STKFLT */
151 "machine check trap", /* 28 T_MCHK */
154 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
155 extern int has_f00f_bug;
159 static int ddb_on_nmi = 1;
160 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
161 &ddb_on_nmi, 0, "Go to DDB on NMI");
163 static int panic_on_nmi = 1;
164 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
165 &panic_on_nmi, 0, "Panic on NMI");
166 static int fast_release;
167 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
168 &fast_release, 0, "Passive Release was optimal");
169 static int slow_release;
170 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
171 &slow_release, 0, "Passive Release was nonoptimal");
173 MALLOC_DEFINE(M_SYSMSG, "sysmsg", "sysmsg structure");
174 extern int max_sysmsg;
177 * Passive USER->KERNEL transition. This only occurs if we block in the
178 * kernel while still holding our userland priority. We have to fixup our
179 * priority in order to avoid potential deadlocks before we allow the system
180 * to switch us to another thread.
183 passive_release(struct thread *td)
185 struct proc *p = td->td_proc;
187 td->td_release = NULL;
188 lwkt_setpri_self(TDPRI_KERN_USER);
189 p->p_usched->release_curproc(p);
193 * userenter() passively intercepts the thread switch function to increase
194 * the thread priority from a user priority to a kernel priority, reducing
195 * syscall and trap overhead for the case where no switch occurs.
199 userenter(struct thread *curtd)
201 curtd->td_release = passive_release;
205 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
206 * must be completed before we can return to or try to return to userland.
208 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
209 * arithmatic on the delta calculation so the absolute tick values are
210 * truncated to an integer.
213 userret(struct proc *p, struct trapframe *frame, int sticks)
218 * Post any pending upcalls
220 if (p->p_flag & P_UPCALLPEND) {
221 p->p_flag &= ~P_UPCALLPEND;
226 * Post any pending signals
228 while ((sig = CURSIG(p)) != 0) {
233 * Charge system time if profiling. Note: times are in microseconds.
235 if (p->p_flag & P_PROFIL) {
236 addupc_task(p, frame->tf_eip,
237 (u_int)((int)p->p_thread->td_sticks - sticks));
241 * Post any pending signals XXX
243 while ((sig = CURSIG(p)) != 0)
248 * Cleanup from userenter and any passive release that might have occured.
249 * We must reclaim the current-process designation before we can return
250 * to usermode. We also handle both LWKT and USER reschedule requests.
253 userexit(struct proc *p)
255 struct thread *td = p->p_thread;
256 globaldata_t gd = td->td_gd;
260 * If a user reschedule is requested force a new process to be
261 * chosen by releasing the current process. Our process will only
262 * be chosen again if it has a considerably better priority.
264 if (user_resched_wanted())
265 p->p_usched->release_curproc(p);
270 * Handle a LWKT reschedule request first. Since our passive release
271 * is still in place we do not have to do anything special.
273 if (lwkt_resched_wanted())
277 * Acquire the current process designation if we do not own it.
278 * Note that acquire_curproc() does not reset the user reschedule
279 * bit on purpose, because we may need to accumulate over several
280 * threads waking up at the same time.
282 * NOTE: userland scheduler cruft: because processes are removed
283 * from the userland scheduler's queue we run through loops to try
284 * to figure out which is the best of [ existing, waking-up ]
287 if (p != gd->gd_uschedcp) {
289 p->p_usched->acquire_curproc(p);
290 /* We may have switched cpus on acquisition */
297 * Reduce our priority in preparation for a return to userland. If
298 * our passive release function was still in place, our priority was
299 * never raised and does not need to be reduced.
301 if (td->td_release == NULL)
302 lwkt_setpri_self(TDPRI_USER_NORM);
303 td->td_release = NULL;
306 * After reducing our priority there might be other kernel-level
307 * LWKTs that now have a greater priority. Run them as necessary.
308 * We don't have to worry about losing cpu to userland because
309 * we still control the current-process designation and we no longer
310 * have a passive release function installed.
312 if (lwkt_checkpri_self())
316 * If a userland reschedule is [still] pending we may not be the best
317 * selected process. Select a better one. If another LWKT resched
318 * is pending the trap will be re-entered.
320 if (user_resched_wanted()) {
321 p->p_usched->select_curproc(gd);
322 if (p != gd->gd_uschedcp) {
323 lwkt_setpri_self(TDPRI_KERN_USER);
330 * Exception, fault, and trap interface to the kernel.
331 * This common code is called from assembly language IDT gate entry
332 * routines that prepare a suitable stack frame, and restore this
333 * frame after the exception has been processed.
335 * This function is also called from doreti in an interlock to handle ASTs.
336 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
338 * NOTE! We have to retrieve the fault address prior to obtaining the
339 * MP lock because get_mplock() may switch out. YYY cr2 really ought
340 * to be retrieved by the assembly code, not here.
342 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
343 * if an attempt is made to switch from a fast interrupt or IPI. This is
344 * necessary to properly take fatal kernel traps on SMP machines if
345 * get_mplock() has to block.
349 struct trapframe frame;
351 struct globaldata *gd = mycpu;
352 struct thread *td = gd->gd_curthread;
355 int i = 0, ucode = 0, type, code;
361 eva = (frame.tf_trapno == T_PAGEFLT ? rcr2() : 0);
362 ++gd->gd_trap_nesting_level;
364 trap_fatal(&frame, eva);
365 --gd->gd_trap_nesting_level;
371 ++gd->gd_trap_nesting_level;
372 if (frame.tf_trapno == T_PAGEFLT) {
374 * For some Cyrix CPUs, %cr2 is clobbered by interrupts.
375 * This problem is worked around by using an interrupt
376 * gate for the pagefault handler. We are finally ready
377 * to read %cr2 and then must reenable interrupts.
379 * XXX this should be in the switch statement, but the
380 * NO_FOOF_HACK and VM86 goto and ifdefs obfuscate the
381 * flow of control too much for this to be obviously
390 --gd->gd_trap_nesting_level;
392 * MP lock is held at this point
395 if (!(frame.tf_eflags & PSL_I)) {
397 * Buggy application or kernel code has disabled interrupts
398 * and then trapped. Enabling interrupts now is wrong, but
399 * it is better than running with interrupts disabled until
400 * they are accidentally enabled later.
402 type = frame.tf_trapno;
403 if (ISPL(frame.tf_cs)==SEL_UPL || (frame.tf_eflags & PSL_VM)) {
405 "pid %ld (%s): trap %d with interrupts disabled\n",
406 (long)curproc->p_pid, curproc->p_comm, type);
407 } else if (type != T_BPTFLT && type != T_TRCTRAP) {
409 * XXX not quite right, since this may be for a
410 * multiple fault in user mode.
412 printf("kernel trap %d with interrupts disabled\n",
418 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
421 type = frame.tf_trapno;
425 if (frame.tf_eflags & PSL_VM &&
426 (type == T_PROTFLT || type == T_STKFLT)) {
428 KKASSERT(td->td_mpcount > 0);
430 i = vm86_emulate((struct vm86frame *)&frame);
432 KKASSERT(td->td_mpcount > 0);
436 * returns to original process
438 vm86_trap((struct vm86frame *)&frame);
445 * these traps want either a process context, or
446 * assume a normal userspace trap.
450 trap_fatal(&frame, eva);
453 type = T_BPTFLT; /* kernel breakpoint */
456 goto kernel_trap; /* normal kernel trap handling */
459 if ((ISPL(frame.tf_cs) == SEL_UPL) || (frame.tf_eflags & PSL_VM)) {
464 sticks = (int)td->td_sticks;
465 p->p_md.md_regs = &frame;
468 case T_PRIVINFLT: /* privileged instruction fault */
473 case T_BPTFLT: /* bpt instruction fault */
474 case T_TRCTRAP: /* trace trap */
475 frame.tf_eflags &= ~PSL_T;
479 case T_ARITHTRAP: /* arithmetic trap */
484 case T_ASTFLT: /* Allow process switch */
485 mycpu->gd_cnt.v_soft++;
486 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
487 atomic_clear_int_nonlocked(&mycpu->gd_reqflags,
489 addupc_task(p, p->p_stats->p_prof.pr_addr,
490 p->p_stats->p_prof.pr_ticks);
495 * The following two traps can happen in
496 * vm86 mode, and, if so, we want to handle
499 case T_PROTFLT: /* general protection fault */
500 case T_STKFLT: /* stack fault */
501 if (frame.tf_eflags & PSL_VM) {
502 i = vm86_emulate((struct vm86frame *)&frame);
509 case T_SEGNPFLT: /* segment not present fault */
510 case T_TSSFLT: /* invalid TSS fault */
511 case T_DOUBLEFLT: /* double fault */
513 ucode = code + BUS_SEGM_FAULT ;
517 case T_PAGEFLT: /* page fault */
518 i = trap_pfault(&frame, TRUE, eva);
521 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
531 case T_DIVIDE: /* integer divide fault */
539 goto handle_powerfail;
540 #else /* !POWERFAIL_NMI */
541 /* machine/parity/power fail/"kitchen sink" faults */
542 if (isa_nmi(code) == 0) {
545 * NMI can be hooked up to a pushbutton
549 printf ("NMI ... going to debugger\n");
550 kdb_trap (type, 0, &frame);
554 } else if (panic_on_nmi)
555 panic("NMI indicates hardware failure");
557 #endif /* POWERFAIL_NMI */
558 #endif /* NISA > 0 */
560 case T_OFLOW: /* integer overflow fault */
565 case T_BOUND: /* bounds check fault */
573 * The kernel may have switched out the FP unit's
574 * state, causing the user process to take a fault
575 * when it tries to use the FP unit. Restore the
581 if (!pmath_emulate) {
583 ucode = FPE_FPU_NP_TRAP;
586 i = (*pmath_emulate)(&frame);
588 if (!(frame.tf_eflags & PSL_T))
590 frame.tf_eflags &= ~PSL_T;
593 /* else ucode = emulator_only_knows() XXX */
596 case T_FPOPFLT: /* FPU operand fetch fault */
601 case T_XMMFLT: /* SIMD floating-point exception */
611 case T_PAGEFLT: /* page fault */
612 (void) trap_pfault(&frame, FALSE, eva);
618 * The kernel may be using npx for copying or other
626 case T_PROTFLT: /* general protection fault */
627 case T_SEGNPFLT: /* segment not present fault */
629 * Invalid segment selectors and out of bounds
630 * %eip's and %esp's can be set up in user mode.
631 * This causes a fault in kernel mode when the
632 * kernel tries to return to user mode. We want
633 * to get this fault so that we can fix the
634 * problem here and not have to check all the
635 * selectors and pointers when the user changes
638 #define MAYBE_DORETI_FAULT(where, whereto) \
640 if (frame.tf_eip == (int)where) { \
641 frame.tf_eip = (int)whereto; \
646 * Since we don't save %gs across an interrupt
647 * frame this check must occur outside the intr
648 * nesting level check.
650 if (frame.tf_eip == (int)cpu_switch_load_gs) {
651 td->td_pcb->pcb_gs = 0;
655 if (mycpu->gd_intr_nesting_level == 0) {
657 * Invalid %fs's and %gs's can be created using
658 * procfs or PT_SETREGS or by invalidating the
659 * underlying LDT entry. This causes a fault
660 * in kernel mode when the kernel attempts to
661 * switch contexts. Lose the bad context
662 * (XXX) so that we can continue, and generate
665 MAYBE_DORETI_FAULT(doreti_iret,
667 MAYBE_DORETI_FAULT(doreti_popl_ds,
668 doreti_popl_ds_fault);
669 MAYBE_DORETI_FAULT(doreti_popl_es,
670 doreti_popl_es_fault);
671 MAYBE_DORETI_FAULT(doreti_popl_fs,
672 doreti_popl_fs_fault);
673 if (td->td_pcb->pcb_onfault) {
675 (register_t)td->td_pcb->pcb_onfault;
683 * PSL_NT can be set in user mode and isn't cleared
684 * automatically when the kernel is entered. This
685 * causes a TSS fault when the kernel attempts to
686 * `iret' because the TSS link is uninitialized. We
687 * want to get this fault so that we can fix the
688 * problem here and not every time the kernel is
691 if (frame.tf_eflags & PSL_NT) {
692 frame.tf_eflags &= ~PSL_NT;
697 case T_TRCTRAP: /* trace trap */
698 if (frame.tf_eip == (int)IDTVEC(syscall)) {
700 * We've just entered system mode via the
701 * syscall lcall. Continue single stepping
702 * silently until the syscall handler has
707 if (frame.tf_eip == (int)IDTVEC(syscall) + 1) {
709 * The syscall handler has now saved the
710 * flags. Stop single stepping it.
712 frame.tf_eflags &= ~PSL_T;
716 * Ignore debug register trace traps due to
717 * accesses in the user's address space, which
718 * can happen under several conditions such as
719 * if a user sets a watchpoint on a buffer and
720 * then passes that buffer to a system call.
721 * We still want to get TRCTRAPS for addresses
722 * in kernel space because that is useful when
723 * debugging the kernel.
725 if (user_dbreg_trap()) {
727 * Reset breakpoint bits because the
730 load_dr6(rdr6() & 0xfffffff0);
734 * Fall through (TRCTRAP kernel mode, kernel address)
738 * If DDB is enabled, let it handle the debugger trap.
739 * Otherwise, debugger traps "can't happen".
742 if (kdb_trap (type, 0, &frame))
751 # define TIMER_FREQ 1193182
755 static unsigned lastalert = 0;
757 if(time_second - lastalert > 10)
759 log(LOG_WARNING, "NMI: power fail\n");
760 sysbeep(TIMER_FREQ/880, hz);
761 lastalert = time_second;
766 #else /* !POWERFAIL_NMI */
767 /* machine/parity/power fail/"kitchen sink" faults */
768 if (isa_nmi(code) == 0) {
771 * NMI can be hooked up to a pushbutton
775 printf ("NMI ... going to debugger\n");
776 kdb_trap (type, 0, &frame);
780 } else if (panic_on_nmi == 0)
783 #endif /* POWERFAIL_NMI */
784 #endif /* NISA > 0 */
787 trap_fatal(&frame, eva);
791 /* Translate fault for emulators (e.g. Linux) */
792 if (*p->p_sysent->sv_transtrap)
793 i = (*p->p_sysent->sv_transtrap)(i, type);
795 trapsignal(p, i, ucode);
798 if (type <= MAX_TRAP_MSG) {
799 uprintf("fatal process exception: %s",
801 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
802 uprintf(", fault VA = 0x%lx", (u_long)eva);
809 if (ISPL(frame.tf_cs) == SEL_UPL)
810 KASSERT(td->td_mpcount == 1, ("badmpcount trap from %p", (void *)frame.tf_eip));
812 userret(p, &frame, sticks);
816 KKASSERT(td->td_mpcount > 0);
823 * This version doesn't allow a page fault to user space while
824 * in the kernel. The rest of the kernel needs to be made "safe"
825 * before this can be used. I think the only things remaining
826 * to be made safe are the iBCS2 code and the process tracing/
830 trap_pfault(frame, usermode, eva)
831 struct trapframe *frame;
836 struct vmspace *vm = NULL;
840 thread_t td = curthread;
841 struct proc *p = td->td_proc; /* may be NULL */
843 if (frame->tf_err & PGEX_W)
844 ftype = VM_PROT_WRITE;
846 ftype = VM_PROT_READ;
848 va = trunc_page(eva);
849 if (va < VM_MIN_KERNEL_ADDRESS) {
854 (!usermode && va < VM_MAXUSER_ADDRESS &&
855 (td->td_gd->gd_intr_nesting_level != 0 ||
856 td->td_pcb->pcb_onfault == NULL))) {
857 trap_fatal(frame, eva);
862 * This is a fault on non-kernel virtual memory.
863 * vm is initialized above to NULL. If curproc is NULL
864 * or curproc->p_vmspace is NULL the fault is fatal.
873 * Keep swapout from messing with us during this
879 * Grow the stack if necessary
881 /* grow_stack returns false only if va falls into
882 * a growable stack region and the stack growth
883 * fails. It returns true if va was not within
884 * a growable stack region, or if the stack
887 if (!grow_stack (p, va)) {
893 /* Fault in the user page: */
894 rv = vm_fault(map, va, ftype,
895 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
901 * Don't allow user-mode faults in kernel address space.
907 * Since we know that kernel virtual address addresses
908 * always have pte pages mapped, we just have to fault
911 rv = vm_fault(kernel_map, va, ftype, VM_FAULT_NORMAL);
914 if (rv == KERN_SUCCESS)
918 if (mtd->td_gd->gd_intr_nesting_level == 0 &&
919 td->td_pcb->pcb_onfault) {
920 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
923 trap_fatal(frame, eva);
927 /* kludge to pass faulting virtual address to sendsig */
930 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
935 trap_pfault(frame, usermode, eva)
936 struct trapframe *frame;
941 struct vmspace *vm = NULL;
945 thread_t td = curthread;
946 struct proc *p = td->td_proc;
948 va = trunc_page(eva);
949 if (va >= KERNBASE) {
951 * Don't allow user-mode faults in kernel address space.
952 * An exception: if the faulting address is the invalid
953 * instruction entry in the IDT, then the Intel Pentium
954 * F00F bug workaround was triggered, and we need to
955 * treat it is as an illegal instruction, and not a page
958 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
959 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) {
960 frame->tf_trapno = T_PRIVINFLT;
970 * This is a fault on non-kernel virtual memory.
971 * vm is initialized above to NULL. If curproc is NULL
972 * or curproc->p_vmspace is NULL the fault is fatal.
983 if (frame->tf_err & PGEX_W)
984 ftype = VM_PROT_WRITE;
986 ftype = VM_PROT_READ;
988 if (map != kernel_map) {
990 * Keep swapout from messing with us during this
996 * Grow the stack if necessary
998 /* grow_stack returns false only if va falls into
999 * a growable stack region and the stack growth
1000 * fails. It returns true if va was not within
1001 * a growable stack region, or if the stack
1004 if (!grow_stack (p, va)) {
1010 /* Fault in the user page: */
1011 rv = vm_fault(map, va, ftype,
1012 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
1018 * Don't have to worry about process locking or stacks in the kernel.
1020 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
1023 if (rv == KERN_SUCCESS)
1027 if (td->td_gd->gd_intr_nesting_level == 0 &&
1028 td->td_pcb->pcb_onfault) {
1029 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
1032 trap_fatal(frame, eva);
1036 /* kludge to pass faulting virtual address to sendsig */
1037 frame->tf_err = eva;
1039 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
1043 trap_fatal(frame, eva)
1044 struct trapframe *frame;
1047 int code, type, ss, esp;
1048 struct soft_segment_descriptor softseg;
1050 code = frame->tf_err;
1051 type = frame->tf_trapno;
1052 sdtossd(&gdt[mycpu->gd_cpuid * NGDT + IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
1054 if (type <= MAX_TRAP_MSG)
1055 printf("\n\nFatal trap %d: %s while in %s mode\n",
1056 type, trap_msg[type],
1057 frame->tf_eflags & PSL_VM ? "vm86" :
1058 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
1060 /* three separate prints in case of a trap on an unmapped page */
1061 printf("mp_lock = %08x; ", mp_lock);
1062 printf("cpuid = %d; ", mycpu->gd_cpuid);
1063 printf("lapic.id = %08x\n", lapic.id);
1065 if (type == T_PAGEFLT) {
1066 printf("fault virtual address = 0x%x\n", eva);
1067 printf("fault code = %s %s, %s\n",
1068 code & PGEX_U ? "user" : "supervisor",
1069 code & PGEX_W ? "write" : "read",
1070 code & PGEX_P ? "protection violation" : "page not present");
1072 printf("instruction pointer = 0x%x:0x%x\n",
1073 frame->tf_cs & 0xffff, frame->tf_eip);
1074 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
1075 ss = frame->tf_ss & 0xffff;
1076 esp = frame->tf_esp;
1078 ss = GSEL(GDATA_SEL, SEL_KPL);
1079 esp = (int)&frame->tf_esp;
1081 printf("stack pointer = 0x%x:0x%x\n", ss, esp);
1082 printf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
1083 printf("code segment = base 0x%x, limit 0x%x, type 0x%x\n",
1084 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
1085 printf(" = DPL %d, pres %d, def32 %d, gran %d\n",
1086 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
1088 printf("processor eflags = ");
1089 if (frame->tf_eflags & PSL_T)
1090 printf("trace trap, ");
1091 if (frame->tf_eflags & PSL_I)
1092 printf("interrupt enabled, ");
1093 if (frame->tf_eflags & PSL_NT)
1094 printf("nested task, ");
1095 if (frame->tf_eflags & PSL_RF)
1097 if (frame->tf_eflags & PSL_VM)
1099 printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
1100 printf("current process = ");
1102 printf("%lu (%s)\n",
1103 (u_long)curproc->p_pid, curproc->p_comm ?
1104 curproc->p_comm : "");
1108 printf("current thread = pri %d ", curthread->td_pri);
1109 if (curthread->td_pri >= TDPRI_CRIT)
1115 * we probably SHOULD have stopped the other CPUs before now!
1116 * another CPU COULD have been touching cpl at this moment...
1118 printf(" <- SMP: XXX");
1127 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1130 printf("trap number = %d\n", type);
1131 if (type <= MAX_TRAP_MSG)
1132 panic("%s", trap_msg[type]);
1134 panic("unknown/reserved trap");
1138 * Double fault handler. Called when a fault occurs while writing
1139 * a frame for a trap/exception onto the stack. This usually occurs
1140 * when the stack overflows (such is the case with infinite recursion,
1143 * XXX Note that the current PTD gets replaced by IdlePTD when the
1144 * task switch occurs. This means that the stack that was active at
1145 * the time of the double fault is not available at <kstack> unless
1146 * the machine was idle when the double fault occurred. The downside
1147 * of this is that "trace <ebp>" in ddb won't work.
1152 struct mdglobaldata *gd = mdcpu;
1154 printf("\nFatal double fault:\n");
1155 printf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1156 printf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1157 printf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
1159 /* three separate prints in case of a trap on an unmapped page */
1160 printf("mp_lock = %08x; ", mp_lock);
1161 printf("cpuid = %d; ", mycpu->gd_cpuid);
1162 printf("lapic.id = %08x\n", lapic.id);
1164 panic("double fault");
1168 * Compensate for 386 brain damage (missing URKR).
1169 * This is a little simpler than the pagefault handler in trap() because
1170 * it the page tables have already been faulted in and high addresses
1171 * are thrown out early for other reasons.
1181 va = trunc_page((vm_offset_t)addr);
1183 * XXX - MAX is END. Changed > to >= for temp. fix.
1185 if (va >= VM_MAXUSER_ADDRESS)
1193 if (!grow_stack (p, va)) {
1199 * fault the data page
1201 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY);
1205 if (rv != KERN_SUCCESS)
1212 * syscall2 - MP aware system call request C handler
1214 * A system call is essentially treated as a trap except that the
1215 * MP lock is not held on entry or return. We are responsible for
1216 * obtaining the MP lock if necessary and for handling ASTs
1217 * (e.g. a task switch) prior to return.
1219 * In general, only simple access and manipulation of curproc and
1220 * the current stack is allowed without having to hold MP lock.
1223 syscall2(struct trapframe frame)
1225 struct thread *td = curthread;
1226 struct proc *p = td->td_proc;
1228 struct sysent *callp;
1229 register_t orig_tf_eflags;
1234 union sysunion args;
1237 if (ISPL(frame.tf_cs) != SEL_UPL) {
1245 KASSERT(td->td_mpcount == 0, ("badmpcount syscall from %p", (void *)frame.tf_eip));
1248 userenter(td); /* lazy raise our priority */
1250 sticks = (int)td->td_sticks;
1252 p->p_md.md_regs = &frame;
1253 params = (caddr_t)frame.tf_esp + sizeof(int);
1254 code = frame.tf_eax;
1255 orig_tf_eflags = frame.tf_eflags;
1257 if (p->p_sysent->sv_prepsyscall) {
1259 * The prep code is not MP aware.
1261 (*p->p_sysent->sv_prepsyscall)(&frame, (int *)(&args.nosys.usrmsg + 1), &code, ¶ms);
1264 * Need to check if this is a 32 bit or 64 bit syscall.
1265 * fuword is MP aware.
1267 if (code == SYS_syscall) {
1269 * Code is first argument, followed by actual args.
1271 code = fuword(params);
1272 params += sizeof(int);
1273 } else if (code == SYS___syscall) {
1275 * Like syscall, but code is a quad, so as to maintain
1276 * quad alignment for the rest of the arguments.
1278 code = fuword(params);
1279 params += sizeof(quad_t);
1283 code &= p->p_sysent->sv_mask;
1284 if (code >= p->p_sysent->sv_size)
1285 callp = &p->p_sysent->sv_table[0];
1287 callp = &p->p_sysent->sv_table[code];
1289 narg = callp->sy_narg & SYF_ARGMASK;
1292 * copyin is MP aware, but the tracing code is not
1294 if (narg && params) {
1295 error = copyin(params, (caddr_t)(&args.nosys.usrmsg + 1),
1296 narg * sizeof(register_t));
1299 if (KTRPOINT(td, KTR_SYSCALL))
1300 ktrsyscall(p->p_tracep, code, narg,
1301 (void *)(&args.nosys.usrmsg + 1));
1309 * Try to run the syscall without the MP lock if the syscall
1310 * is MP safe. We have to obtain the MP lock no matter what if
1313 if ((callp->sy_narg & SYF_MPSAFE) == 0) {
1320 if (KTRPOINT(td, KTR_SYSCALL)) {
1321 ktrsyscall(p->p_tracep, code, narg, (void *)(&args.nosys.usrmsg + 1));
1326 * For traditional syscall code edx is left untouched when 32 bit
1327 * results are returned. Since edx is loaded from fds[1] when the
1328 * system call returns we pre-set it here.
1330 lwkt_initmsg(&args.lmsg, &td->td_msgport, 0,
1331 lwkt_cmd_op(code), lwkt_cmd_op_none);
1332 args.sysmsg_copyout = NULL;
1333 args.sysmsg_fds[0] = 0;
1334 args.sysmsg_fds[1] = frame.tf_edx;
1336 STOPEVENT(p, S_SCE, narg); /* MP aware */
1338 error = (*callp->sy_call)(&args);
1341 * MP SAFE (we may or may not have the MP lock at this point)
1346 * Reinitialize proc pointer `p' as it may be different
1347 * if this is a child returning from fork syscall.
1350 frame.tf_eax = args.sysmsg_fds[0];
1351 frame.tf_edx = args.sysmsg_fds[1];
1352 frame.tf_eflags &= ~PSL_C;
1356 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1357 * int 0x80 is 2 bytes. We saved this in tf_err.
1359 frame.tf_eip -= frame.tf_err;
1364 panic("Unexpected EASYNC return value (for now)");
1367 if (p->p_sysent->sv_errsize) {
1368 if (error >= p->p_sysent->sv_errsize)
1369 error = -1; /* XXX */
1371 error = p->p_sysent->sv_errtbl[error];
1373 frame.tf_eax = error;
1374 frame.tf_eflags |= PSL_C;
1379 * Traced syscall. trapsignal() is not MP aware.
1381 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1382 frame.tf_eflags &= ~PSL_T;
1383 trapsignal(p, SIGTRAP, 0);
1387 * Handle reschedule and other end-of-syscall issues
1389 userret(p, &frame, sticks);
1392 if (KTRPOINT(td, KTR_SYSRET)) {
1393 ktrsysret(p->p_tracep, code, error, args.sysmsg_result);
1398 * This works because errno is findable through the
1399 * register set. If we ever support an emulation where this
1400 * is not the case, this code will need to be revisited.
1402 STOPEVENT(p, S_SCX, code);
1407 * Release the MP lock if we had to get it
1409 KASSERT(td->td_mpcount == 1, ("badmpcount syscall from %p", (void *)frame.tf_eip));
1415 * free_sysun - Put an unused sysun on the free list.
1417 static __inline void
1418 free_sysun(struct thread *td, union sysunion *sysun)
1420 struct globaldata *gd = td->td_gd;
1422 crit_enter_quick(td);
1423 sysun->lmsg.opaque.ms_sysunnext = gd->gd_freesysun;
1424 gd->gd_freesysun = sysun;
1425 crit_exit_quick(td);
1429 * sendsys2 - MP aware system message request C handler
1432 sendsys2(struct trapframe frame)
1434 struct globaldata *gd;
1435 struct thread *td = curthread;
1436 struct proc *p = td->td_proc;
1437 register_t orig_tf_eflags;
1438 struct sysent *callp;
1439 union sysunion *sysun = NULL;
1449 if (ISPL(frame.tf_cs) != SEL_UPL) {
1457 KASSERT(td->td_mpcount == 0, ("badmpcount syscall from %p", (void *)frame.tf_eip));
1461 * access non-atomic field from critical section. p_sticks is
1462 * updated by the clock interrupt. Also use this opportunity
1463 * to lazy-raise our LWKT priority.
1466 sticks = td->td_sticks;
1468 p->p_md.md_regs = &frame;
1469 orig_tf_eflags = frame.tf_eflags;
1473 * Extract the system call message. If msgsize is zero we are
1474 * blocking on a message and/or message port. If msgsize is -1
1475 * we are testing a message for completion or a message port for
1478 * The userland system call message size includes the size of the
1479 * userland lwkt_msg plus arguments. We load it into the userland
1480 * portion of our sysunion structure then we initialize the kerneland
1487 if ((msgsize = frame.tf_edx) < sizeof(struct lwkt_msg) ||
1488 msgsize > sizeof(union sysunion) - sizeof(struct sysmsg)) {
1494 * Obtain a sysun from our per-cpu cache or allocate a new one. Use
1495 * the opaque field to store the original (user) message pointer.
1496 * A critical section is necessary to interlock against interrupts
1497 * returning system messages to the thread cache.
1500 crit_enter_quick(td);
1501 if ((sysun = gd->gd_freesysun) != NULL)
1502 gd->gd_freesysun = sysun->lmsg.opaque.ms_sysunnext;
1504 sysun = malloc(sizeof(union sysunion), M_SYSMSG, M_WAITOK);
1505 crit_exit_quick(td);
1508 * Copy the user request into the kernel copy of the user request.
1510 umsg = (void *)frame.tf_ecx;
1511 error = copyin(umsg, &sysun->nosys.usrmsg, msgsize);
1514 if ((sysun->nosys.usrmsg.umsg.ms_flags & MSGF_ASYNC)) {
1519 if (max_sysmsg > 0 && p->p_num_sysmsg >= max_sysmsg) {
1526 * Initialize the kernel message from the copied-in data and
1527 * pull in appropriate flags from the userland message.
1529 * ms_abort_port is usually initialized in sendmsg/domsg, but since
1530 * we are not calling those functions (yet), we have to do it manually.
1532 lwkt_initmsg(&sysun->lmsg, &td->td_msgport, 0,
1533 sysun->nosys.usrmsg.umsg.ms_cmd,
1535 sysun->lmsg.ms_abort_port = sysun->lmsg.ms_reply_port;
1536 sysun->sysmsg_copyout = NULL;
1537 sysun->lmsg.opaque.ms_umsg = umsg;
1538 sysun->lmsg.ms_flags |= sysun->nosys.usrmsg.umsg.ms_flags & MSGF_ASYNC;
1541 * Extract the system call number, lookup the system call, and
1542 * set the default return value.
1544 code = (u_int)sysun->lmsg.ms_cmd.cm_op;
1545 /* We don't handle the syscall() syscall yet */
1548 free_sysun(td, sysun);
1551 if (code >= p->p_sysent->sv_size) {
1553 free_sysun(td, sysun);
1557 callp = &p->p_sysent->sv_table[code];
1559 narg = (msgsize - sizeof(struct lwkt_msg)) / sizeof(register_t);
1562 if (KTRPOINT(td, KTR_SYSCALL)) {
1563 ktrsyscall(p->p_tracep, code, narg, (void *)(&sysun->nosys.usrmsg + 1));
1566 sysun->lmsg.u.ms_fds[0] = 0;
1567 sysun->lmsg.u.ms_fds[1] = 0;
1569 STOPEVENT(p, S_SCE, narg); /* MP aware */
1572 * Make the system call. An error code is always returned, results
1573 * are copied back via ms_result32 or ms_result64. YYY temporary
1574 * stage copy p_retval[] into ms_result32/64
1576 * NOTE! XXX if this is a child returning from a fork curproc
1577 * might be different. YYY huh? a child returning from a fork
1578 * should never 'return' from this call, it should go right to the
1579 * fork_trampoline function.
1581 error = (*callp->sy_call)(sysun);
1582 gd = td->td_gd; /* RELOAD, might have switched cpus */
1586 * If a synchronous return copy p_retval to ms_result64 and return
1587 * the sysmsg to the free pool.
1589 * YYY Don't writeback message if execve() YYY
1591 sysun->nosys.usrmsg.umsg.ms_error = error;
1592 sysun->nosys.usrmsg.umsg.u.ms_fds[0] = sysun->lmsg.u.ms_fds[0];
1593 sysun->nosys.usrmsg.umsg.u.ms_fds[1] = sysun->lmsg.u.ms_fds[1];
1594 result = sysun->nosys.usrmsg.umsg.u.ms_fds[0]; /* for ktrace */
1595 if (error != 0 || code != SYS_execve) {
1597 error2 = copyout(&sysun->nosys.usrmsg.umsg.ms_copyout_start,
1598 &umsg->ms_copyout_start,
1603 if (error == EASYNC) {
1605 * Since only the current process ever messes with msgq,
1606 * we can safely manipulate it in parallel with the async
1609 TAILQ_INSERT_TAIL(&p->p_sysmsgq, &sysun->sysmsg, msgq);
1611 error = (int)&sysun->sysmsg;
1614 free_sysun(td, sysun);
1617 frame.tf_eax = (register_t)error;
1620 * Traced syscall. trapsignal() is not MP aware.
1622 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1623 frame.tf_eflags &= ~PSL_T;
1624 trapsignal(p, SIGTRAP, 0);
1628 * Handle reschedule and other end-of-syscall issues
1630 userret(p, &frame, sticks);
1633 if (KTRPOINT(td, KTR_SYSRET)) {
1634 ktrsysret(p->p_tracep, code, error, result);
1639 * This works because errno is findable through the
1640 * register set. If we ever support an emulation where this
1641 * is not the case, this code will need to be revisited.
1643 STOPEVENT(p, S_SCX, code);
1648 * Release the MP lock if we had to get it
1650 KASSERT(td->td_mpcount == 1, ("badmpcount syscall from %p", (void *)frame.tf_eip));
1656 * waitsys2 - MP aware system message wait C handler
1659 waitsys2(struct trapframe frame)
1661 struct globaldata *gd;
1662 struct thread *td = curthread;
1663 struct proc *p = td->td_proc;
1664 union sysunion *sysun = NULL;
1666 register_t orig_tf_eflags;
1667 int error = 0, result, sticks;
1671 if (ISPL(frame.tf_cs) != SEL_UPL) {
1679 KASSERT(td->td_mpcount == 0, ("badmpcount syscall from %p",
1680 (void *)frame.tf_eip));
1685 * access non-atomic field from critical section. p_sticks is
1686 * updated by the clock interrupt. Also use this opportunity
1687 * to lazy-raise our LWKT priority.
1690 sticks = td->td_sticks;
1692 p->p_md.md_regs = &frame;
1693 orig_tf_eflags = frame.tf_eflags;
1699 TAILQ_FOREACH(ptr, &p->p_sysmsgq, msgq) {
1700 if ((void *)ptr == (void *)frame.tf_ecx) {
1701 sysun = (void *)sysmsg_wait(p,
1702 (void *)frame.tf_ecx, 1);
1712 else if (frame.tf_eax) {
1713 printf("waitport/checkport only the default port is supported at the moment\n");
1718 switch(frame.tf_edx) {
1720 sysun = (void *)sysmsg_wait(p, NULL, 0);
1723 sysun = (void *)sysmsg_wait(p, NULL, 1);
1732 umsg = sysun->lmsg.opaque.ms_umsg;
1733 frame.tf_eax = (register_t)sysun;
1734 sysun->nosys.usrmsg.umsg.u.ms_fds[0] = sysun->lmsg.u.ms_fds[0];
1735 sysun->nosys.usrmsg.umsg.u.ms_fds[1] = sysun->lmsg.u.ms_fds[1];
1736 sysun->nosys.usrmsg.umsg.ms_error = sysun->lmsg.ms_error;
1737 error = sysun->lmsg.ms_error;
1738 result = sysun->lmsg.u.ms_fds[0]; /* for ktrace */
1739 error = copyout(&sysun->nosys.usrmsg.umsg.ms_copyout_start,
1740 &umsg->ms_copyout_start, ms_copyout_size);
1741 free_sysun(td, sysun);
1743 code = (u_int)sysun->lmsg.ms_cmd.cm_op;
1747 frame.tf_eax = error;
1749 * Traced syscall. trapsignal() is not MP aware.
1751 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1752 frame.tf_eflags &= ~PSL_T;
1753 trapsignal(p, SIGTRAP, 0);
1757 * Handle reschedule and other end-of-syscall issues
1759 userret(p, &frame, sticks);
1762 if (KTRPOINT(td, KTR_SYSRET)) {
1763 ktrsysret(p->p_tracep, code, error, result);
1768 * This works because errno is findable through the
1769 * register set. If we ever support an emulation where this
1770 * is not the case, this code will need to be revisited.
1772 STOPEVENT(p, S_SCX, code);
1776 KASSERT(td->td_mpcount == 1, ("badmpcount syscall from %p",
1777 (void *)frame.tf_eip));
1783 * Simplified back end of syscall(), used when returning from fork()
1784 * directly into user mode. MP lock is held on entry and should be
1785 * released on return. This code will return back into the fork
1786 * trampoline code which then runs doreti.
1789 fork_return(p, frame)
1791 struct trapframe frame;
1793 frame.tf_eax = 0; /* Child returns zero */
1794 frame.tf_eflags &= ~PSL_C; /* success */
1798 * Newly forked processes are given a kernel priority. We have to
1799 * adjust the priority to a normal user priority and fake entry
1800 * into the kernel (call userenter()) to install a passive release
1801 * function just in case userret() decides to stop the process. This
1802 * can occur when ^Z races a fork. If we do not install the passive
1803 * release function the current process designation will not be
1804 * released when the thread goes to sleep.
1806 lwkt_setpri_self(TDPRI_USER_NORM);
1807 userenter(p->p_thread);
1808 userret(p, &frame, 0);
1810 if (KTRPOINT(p->p_thread, KTR_SYSRET))
1811 ktrsysret(p->p_tracep, SYS_fork, 0, 0);
1813 p->p_flag |= P_PASSIVE_ACQ;
1815 p->p_flag &= ~P_PASSIVE_ACQ;
1817 KKASSERT(p->p_thread->td_mpcount == 1);