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 1.59 2005/06/25 20:03:34 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.
344 struct trapframe frame;
346 struct thread *td = curthread;
349 int i = 0, ucode = 0, type, code;
355 eva = (frame.tf_trapno == T_PAGEFLT ? rcr2() : 0);
357 trap_fatal(&frame, eva);
363 if (frame.tf_trapno == T_PAGEFLT) {
365 * For some Cyrix CPUs, %cr2 is clobbered by interrupts.
366 * This problem is worked around by using an interrupt
367 * gate for the pagefault handler. We are finally ready
368 * to read %cr2 and then must reenable interrupts.
370 * XXX this should be in the switch statement, but the
371 * NO_FOOF_HACK and VM86 goto and ifdefs obfuscate the
372 * flow of control too much for this to be obviously
382 * MP lock is held at this point
385 if (!(frame.tf_eflags & PSL_I)) {
387 * Buggy application or kernel code has disabled interrupts
388 * and then trapped. Enabling interrupts now is wrong, but
389 * it is better than running with interrupts disabled until
390 * they are accidentally enabled later.
392 type = frame.tf_trapno;
393 if (ISPL(frame.tf_cs)==SEL_UPL || (frame.tf_eflags & PSL_VM)) {
395 "pid %ld (%s): trap %d with interrupts disabled\n",
396 (long)curproc->p_pid, curproc->p_comm, type);
397 } else if (type != T_BPTFLT && type != T_TRCTRAP) {
399 * XXX not quite right, since this may be for a
400 * multiple fault in user mode.
402 printf("kernel trap %d with interrupts disabled\n",
408 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
411 type = frame.tf_trapno;
415 if (frame.tf_eflags & PSL_VM &&
416 (type == T_PROTFLT || type == T_STKFLT)) {
418 KKASSERT(td->td_mpcount > 0);
420 i = vm86_emulate((struct vm86frame *)&frame);
422 KKASSERT(td->td_mpcount > 0);
426 * returns to original process
428 vm86_trap((struct vm86frame *)&frame);
435 * these traps want either a process context, or
436 * assume a normal userspace trap.
440 trap_fatal(&frame, eva);
443 type = T_BPTFLT; /* kernel breakpoint */
446 goto kernel_trap; /* normal kernel trap handling */
449 if ((ISPL(frame.tf_cs) == SEL_UPL) || (frame.tf_eflags & PSL_VM)) {
454 sticks = (int)td->td_sticks;
455 p->p_md.md_regs = &frame;
458 case T_PRIVINFLT: /* privileged instruction fault */
463 case T_BPTFLT: /* bpt instruction fault */
464 case T_TRCTRAP: /* trace trap */
465 frame.tf_eflags &= ~PSL_T;
469 case T_ARITHTRAP: /* arithmetic trap */
474 case T_ASTFLT: /* Allow process switch */
475 mycpu->gd_cnt.v_soft++;
476 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
477 atomic_clear_int_nonlocked(&mycpu->gd_reqflags,
479 addupc_task(p, p->p_stats->p_prof.pr_addr,
480 p->p_stats->p_prof.pr_ticks);
485 * The following two traps can happen in
486 * vm86 mode, and, if so, we want to handle
489 case T_PROTFLT: /* general protection fault */
490 case T_STKFLT: /* stack fault */
491 if (frame.tf_eflags & PSL_VM) {
492 i = vm86_emulate((struct vm86frame *)&frame);
499 case T_SEGNPFLT: /* segment not present fault */
500 case T_TSSFLT: /* invalid TSS fault */
501 case T_DOUBLEFLT: /* double fault */
503 ucode = code + BUS_SEGM_FAULT ;
507 case T_PAGEFLT: /* page fault */
508 i = trap_pfault(&frame, TRUE, eva);
511 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
521 case T_DIVIDE: /* integer divide fault */
529 goto handle_powerfail;
530 #else /* !POWERFAIL_NMI */
531 /* machine/parity/power fail/"kitchen sink" faults */
532 if (isa_nmi(code) == 0) {
535 * NMI can be hooked up to a pushbutton
539 printf ("NMI ... going to debugger\n");
540 kdb_trap (type, 0, &frame);
544 } else if (panic_on_nmi)
545 panic("NMI indicates hardware failure");
547 #endif /* POWERFAIL_NMI */
548 #endif /* NISA > 0 */
550 case T_OFLOW: /* integer overflow fault */
555 case T_BOUND: /* bounds check fault */
563 * The kernel may have switched out the FP unit's
564 * state, causing the user process to take a fault
565 * when it tries to use the FP unit. Restore the
571 if (!pmath_emulate) {
573 ucode = FPE_FPU_NP_TRAP;
576 i = (*pmath_emulate)(&frame);
578 if (!(frame.tf_eflags & PSL_T))
580 frame.tf_eflags &= ~PSL_T;
583 /* else ucode = emulator_only_knows() XXX */
586 case T_FPOPFLT: /* FPU operand fetch fault */
591 case T_XMMFLT: /* SIMD floating-point exception */
601 case T_PAGEFLT: /* page fault */
602 (void) trap_pfault(&frame, FALSE, eva);
608 * The kernel may be using npx for copying or other
616 case T_PROTFLT: /* general protection fault */
617 case T_SEGNPFLT: /* segment not present fault */
619 * Invalid segment selectors and out of bounds
620 * %eip's and %esp's can be set up in user mode.
621 * This causes a fault in kernel mode when the
622 * kernel tries to return to user mode. We want
623 * to get this fault so that we can fix the
624 * problem here and not have to check all the
625 * selectors and pointers when the user changes
628 #define MAYBE_DORETI_FAULT(where, whereto) \
630 if (frame.tf_eip == (int)where) { \
631 frame.tf_eip = (int)whereto; \
636 * Since we don't save %gs across an interrupt
637 * frame this check must occur outside the intr
638 * nesting level check.
640 if (frame.tf_eip == (int)cpu_switch_load_gs) {
641 td->td_pcb->pcb_gs = 0;
645 if (mycpu->gd_intr_nesting_level == 0) {
647 * Invalid %fs's and %gs's can be created using
648 * procfs or PT_SETREGS or by invalidating the
649 * underlying LDT entry. This causes a fault
650 * in kernel mode when the kernel attempts to
651 * switch contexts. Lose the bad context
652 * (XXX) so that we can continue, and generate
655 MAYBE_DORETI_FAULT(doreti_iret,
657 MAYBE_DORETI_FAULT(doreti_popl_ds,
658 doreti_popl_ds_fault);
659 MAYBE_DORETI_FAULT(doreti_popl_es,
660 doreti_popl_es_fault);
661 MAYBE_DORETI_FAULT(doreti_popl_fs,
662 doreti_popl_fs_fault);
663 if (td->td_pcb->pcb_onfault) {
665 (register_t)td->td_pcb->pcb_onfault;
673 * PSL_NT can be set in user mode and isn't cleared
674 * automatically when the kernel is entered. This
675 * causes a TSS fault when the kernel attempts to
676 * `iret' because the TSS link is uninitialized. We
677 * want to get this fault so that we can fix the
678 * problem here and not every time the kernel is
681 if (frame.tf_eflags & PSL_NT) {
682 frame.tf_eflags &= ~PSL_NT;
687 case T_TRCTRAP: /* trace trap */
688 if (frame.tf_eip == (int)IDTVEC(syscall)) {
690 * We've just entered system mode via the
691 * syscall lcall. Continue single stepping
692 * silently until the syscall handler has
697 if (frame.tf_eip == (int)IDTVEC(syscall) + 1) {
699 * The syscall handler has now saved the
700 * flags. Stop single stepping it.
702 frame.tf_eflags &= ~PSL_T;
706 * Ignore debug register trace traps due to
707 * accesses in the user's address space, which
708 * can happen under several conditions such as
709 * if a user sets a watchpoint on a buffer and
710 * then passes that buffer to a system call.
711 * We still want to get TRCTRAPS for addresses
712 * in kernel space because that is useful when
713 * debugging the kernel.
715 if (user_dbreg_trap()) {
717 * Reset breakpoint bits because the
720 load_dr6(rdr6() & 0xfffffff0);
724 * Fall through (TRCTRAP kernel mode, kernel address)
728 * If DDB is enabled, let it handle the debugger trap.
729 * Otherwise, debugger traps "can't happen".
732 if (kdb_trap (type, 0, &frame))
741 # define TIMER_FREQ 1193182
745 static unsigned lastalert = 0;
747 if(time_second - lastalert > 10)
749 log(LOG_WARNING, "NMI: power fail\n");
750 sysbeep(TIMER_FREQ/880, hz);
751 lastalert = time_second;
756 #else /* !POWERFAIL_NMI */
757 /* machine/parity/power fail/"kitchen sink" faults */
758 if (isa_nmi(code) == 0) {
761 * NMI can be hooked up to a pushbutton
765 printf ("NMI ... going to debugger\n");
766 kdb_trap (type, 0, &frame);
770 } else if (panic_on_nmi == 0)
773 #endif /* POWERFAIL_NMI */
774 #endif /* NISA > 0 */
777 trap_fatal(&frame, eva);
781 /* Translate fault for emulators (e.g. Linux) */
782 if (*p->p_sysent->sv_transtrap)
783 i = (*p->p_sysent->sv_transtrap)(i, type);
785 trapsignal(p, i, ucode);
788 if (type <= MAX_TRAP_MSG) {
789 uprintf("fatal process exception: %s",
791 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
792 uprintf(", fault VA = 0x%lx", (u_long)eva);
799 if (ISPL(frame.tf_cs) == SEL_UPL)
800 KASSERT(td->td_mpcount == 1, ("badmpcount trap from %p", (void *)frame.tf_eip));
802 userret(p, &frame, sticks);
806 KKASSERT(td->td_mpcount > 0);
813 * This version doesn't allow a page fault to user space while
814 * in the kernel. The rest of the kernel needs to be made "safe"
815 * before this can be used. I think the only things remaining
816 * to be made safe are the iBCS2 code and the process tracing/
820 trap_pfault(frame, usermode, eva)
821 struct trapframe *frame;
826 struct vmspace *vm = NULL;
830 thread_t td = curthread;
831 struct proc *p = td->td_proc; /* may be NULL */
833 if (frame->tf_err & PGEX_W)
834 ftype = VM_PROT_WRITE;
836 ftype = VM_PROT_READ;
838 va = trunc_page(eva);
839 if (va < VM_MIN_KERNEL_ADDRESS) {
844 (!usermode && va < VM_MAXUSER_ADDRESS &&
845 (td->td_gd->gd_intr_nesting_level != 0 ||
846 td->td_pcb->pcb_onfault == NULL))) {
847 trap_fatal(frame, eva);
852 * This is a fault on non-kernel virtual memory.
853 * vm is initialized above to NULL. If curproc is NULL
854 * or curproc->p_vmspace is NULL the fault is fatal.
863 * Keep swapout from messing with us during this
869 * Grow the stack if necessary
871 /* grow_stack returns false only if va falls into
872 * a growable stack region and the stack growth
873 * fails. It returns true if va was not within
874 * a growable stack region, or if the stack
877 if (!grow_stack (p, va)) {
883 /* Fault in the user page: */
884 rv = vm_fault(map, va, ftype,
885 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
891 * Don't allow user-mode faults in kernel address space.
897 * Since we know that kernel virtual address addresses
898 * always have pte pages mapped, we just have to fault
901 rv = vm_fault(kernel_map, va, ftype, VM_FAULT_NORMAL);
904 if (rv == KERN_SUCCESS)
908 if (mtd->td_gd->gd_intr_nesting_level == 0 &&
909 td->td_pcb->pcb_onfault) {
910 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
913 trap_fatal(frame, eva);
917 /* kludge to pass faulting virtual address to sendsig */
920 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
925 trap_pfault(frame, usermode, eva)
926 struct trapframe *frame;
931 struct vmspace *vm = NULL;
935 thread_t td = curthread;
936 struct proc *p = td->td_proc;
938 va = trunc_page(eva);
939 if (va >= KERNBASE) {
941 * Don't allow user-mode faults in kernel address space.
942 * An exception: if the faulting address is the invalid
943 * instruction entry in the IDT, then the Intel Pentium
944 * F00F bug workaround was triggered, and we need to
945 * treat it is as an illegal instruction, and not a page
948 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
949 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug) {
950 frame->tf_trapno = T_PRIVINFLT;
960 * This is a fault on non-kernel virtual memory.
961 * vm is initialized above to NULL. If curproc is NULL
962 * or curproc->p_vmspace is NULL the fault is fatal.
973 if (frame->tf_err & PGEX_W)
974 ftype = VM_PROT_WRITE;
976 ftype = VM_PROT_READ;
978 if (map != kernel_map) {
980 * Keep swapout from messing with us during this
986 * Grow the stack if necessary
988 /* grow_stack returns false only if va falls into
989 * a growable stack region and the stack growth
990 * fails. It returns true if va was not within
991 * a growable stack region, or if the stack
994 if (!grow_stack (p, va)) {
1000 /* Fault in the user page: */
1001 rv = vm_fault(map, va, ftype,
1002 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
1008 * Don't have to worry about process locking or stacks in the kernel.
1010 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
1013 if (rv == KERN_SUCCESS)
1017 if (td->td_gd->gd_intr_nesting_level == 0 &&
1018 td->td_pcb->pcb_onfault) {
1019 frame->tf_eip = (register_t)td->td_pcb->pcb_onfault;
1022 trap_fatal(frame, eva);
1026 /* kludge to pass faulting virtual address to sendsig */
1027 frame->tf_err = eva;
1029 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
1033 trap_fatal(frame, eva)
1034 struct trapframe *frame;
1037 int code, type, ss, esp;
1038 struct soft_segment_descriptor softseg;
1040 code = frame->tf_err;
1041 type = frame->tf_trapno;
1042 sdtossd(&gdt[mycpu->gd_cpuid * NGDT + IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
1044 if (type <= MAX_TRAP_MSG)
1045 printf("\n\nFatal trap %d: %s while in %s mode\n",
1046 type, trap_msg[type],
1047 frame->tf_eflags & PSL_VM ? "vm86" :
1048 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
1050 /* three separate prints in case of a trap on an unmapped page */
1051 printf("mp_lock = %08x; ", mp_lock);
1052 printf("cpuid = %d; ", mycpu->gd_cpuid);
1053 printf("lapic.id = %08x\n", lapic.id);
1055 if (type == T_PAGEFLT) {
1056 printf("fault virtual address = 0x%x\n", eva);
1057 printf("fault code = %s %s, %s\n",
1058 code & PGEX_U ? "user" : "supervisor",
1059 code & PGEX_W ? "write" : "read",
1060 code & PGEX_P ? "protection violation" : "page not present");
1062 printf("instruction pointer = 0x%x:0x%x\n",
1063 frame->tf_cs & 0xffff, frame->tf_eip);
1064 if ((ISPL(frame->tf_cs) == SEL_UPL) || (frame->tf_eflags & PSL_VM)) {
1065 ss = frame->tf_ss & 0xffff;
1066 esp = frame->tf_esp;
1068 ss = GSEL(GDATA_SEL, SEL_KPL);
1069 esp = (int)&frame->tf_esp;
1071 printf("stack pointer = 0x%x:0x%x\n", ss, esp);
1072 printf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
1073 printf("code segment = base 0x%x, limit 0x%x, type 0x%x\n",
1074 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
1075 printf(" = DPL %d, pres %d, def32 %d, gran %d\n",
1076 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
1078 printf("processor eflags = ");
1079 if (frame->tf_eflags & PSL_T)
1080 printf("trace trap, ");
1081 if (frame->tf_eflags & PSL_I)
1082 printf("interrupt enabled, ");
1083 if (frame->tf_eflags & PSL_NT)
1084 printf("nested task, ");
1085 if (frame->tf_eflags & PSL_RF)
1087 if (frame->tf_eflags & PSL_VM)
1089 printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
1090 printf("current process = ");
1092 printf("%lu (%s)\n",
1093 (u_long)curproc->p_pid, curproc->p_comm ?
1094 curproc->p_comm : "");
1098 printf("current thread = pri %d ", curthread->td_pri);
1099 if (curthread->td_pri >= TDPRI_CRIT)
1105 * we probably SHOULD have stopped the other CPUs before now!
1106 * another CPU COULD have been touching cpl at this moment...
1108 printf(" <- SMP: XXX");
1117 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1120 printf("trap number = %d\n", type);
1121 if (type <= MAX_TRAP_MSG)
1122 panic("%s", trap_msg[type]);
1124 panic("unknown/reserved trap");
1128 * Double fault handler. Called when a fault occurs while writing
1129 * a frame for a trap/exception onto the stack. This usually occurs
1130 * when the stack overflows (such is the case with infinite recursion,
1133 * XXX Note that the current PTD gets replaced by IdlePTD when the
1134 * task switch occurs. This means that the stack that was active at
1135 * the time of the double fault is not available at <kstack> unless
1136 * the machine was idle when the double fault occurred. The downside
1137 * of this is that "trace <ebp>" in ddb won't work.
1142 struct mdglobaldata *gd = mdcpu;
1144 printf("\nFatal double fault:\n");
1145 printf("eip = 0x%x\n", gd->gd_common_tss.tss_eip);
1146 printf("esp = 0x%x\n", gd->gd_common_tss.tss_esp);
1147 printf("ebp = 0x%x\n", gd->gd_common_tss.tss_ebp);
1149 /* three separate prints in case of a trap on an unmapped page */
1150 printf("mp_lock = %08x; ", mp_lock);
1151 printf("cpuid = %d; ", mycpu->gd_cpuid);
1152 printf("lapic.id = %08x\n", lapic.id);
1154 panic("double fault");
1158 * Compensate for 386 brain damage (missing URKR).
1159 * This is a little simpler than the pagefault handler in trap() because
1160 * it the page tables have already been faulted in and high addresses
1161 * are thrown out early for other reasons.
1171 va = trunc_page((vm_offset_t)addr);
1173 * XXX - MAX is END. Changed > to >= for temp. fix.
1175 if (va >= VM_MAXUSER_ADDRESS)
1183 if (!grow_stack (p, va)) {
1189 * fault the data page
1191 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY);
1195 if (rv != KERN_SUCCESS)
1202 * syscall2 - MP aware system call request C handler
1204 * A system call is essentially treated as a trap except that the
1205 * MP lock is not held on entry or return. We are responsible for
1206 * obtaining the MP lock if necessary and for handling ASTs
1207 * (e.g. a task switch) prior to return.
1209 * In general, only simple access and manipulation of curproc and
1210 * the current stack is allowed without having to hold MP lock.
1213 syscall2(struct trapframe frame)
1215 struct thread *td = curthread;
1216 struct proc *p = td->td_proc;
1218 struct sysent *callp;
1219 register_t orig_tf_eflags;
1224 union sysunion args;
1227 if (ISPL(frame.tf_cs) != SEL_UPL) {
1235 KASSERT(td->td_mpcount == 0, ("badmpcount syscall from %p", (void *)frame.tf_eip));
1238 userenter(td); /* lazy raise our priority */
1240 sticks = (int)td->td_sticks;
1242 p->p_md.md_regs = &frame;
1243 params = (caddr_t)frame.tf_esp + sizeof(int);
1244 code = frame.tf_eax;
1245 orig_tf_eflags = frame.tf_eflags;
1247 if (p->p_sysent->sv_prepsyscall) {
1249 * The prep code is not MP aware.
1251 (*p->p_sysent->sv_prepsyscall)(&frame, (int *)(&args.nosys.usrmsg + 1), &code, ¶ms);
1254 * Need to check if this is a 32 bit or 64 bit syscall.
1255 * fuword is MP aware.
1257 if (code == SYS_syscall) {
1259 * Code is first argument, followed by actual args.
1261 code = fuword(params);
1262 params += sizeof(int);
1263 } else if (code == SYS___syscall) {
1265 * Like syscall, but code is a quad, so as to maintain
1266 * quad alignment for the rest of the arguments.
1268 code = fuword(params);
1269 params += sizeof(quad_t);
1273 code &= p->p_sysent->sv_mask;
1274 if (code >= p->p_sysent->sv_size)
1275 callp = &p->p_sysent->sv_table[0];
1277 callp = &p->p_sysent->sv_table[code];
1279 narg = callp->sy_narg & SYF_ARGMASK;
1282 * copyin is MP aware, but the tracing code is not
1284 if (narg && params) {
1285 error = copyin(params, (caddr_t)(&args.nosys.usrmsg + 1),
1286 narg * sizeof(register_t));
1289 if (KTRPOINT(td, KTR_SYSCALL))
1290 ktrsyscall(p->p_tracep, code, narg,
1291 (void *)(&args.nosys.usrmsg + 1));
1299 * Try to run the syscall without the MP lock if the syscall
1300 * is MP safe. We have to obtain the MP lock no matter what if
1303 if ((callp->sy_narg & SYF_MPSAFE) == 0) {
1310 if (KTRPOINT(td, KTR_SYSCALL)) {
1311 ktrsyscall(p->p_tracep, code, narg, (void *)(&args.nosys.usrmsg + 1));
1316 * For traditional syscall code edx is left untouched when 32 bit
1317 * results are returned. Since edx is loaded from fds[1] when the
1318 * system call returns we pre-set it here.
1320 lwkt_initmsg(&args.lmsg, &td->td_msgport, 0,
1321 lwkt_cmd_op(code), lwkt_cmd_op_none);
1322 args.sysmsg_copyout = NULL;
1323 args.sysmsg_fds[0] = 0;
1324 args.sysmsg_fds[1] = frame.tf_edx;
1326 STOPEVENT(p, S_SCE, narg); /* MP aware */
1328 error = (*callp->sy_call)(&args);
1331 * MP SAFE (we may or may not have the MP lock at this point)
1336 * Reinitialize proc pointer `p' as it may be different
1337 * if this is a child returning from fork syscall.
1340 frame.tf_eax = args.sysmsg_fds[0];
1341 frame.tf_edx = args.sysmsg_fds[1];
1342 frame.tf_eflags &= ~PSL_C;
1346 * Reconstruct pc, assuming lcall $X,y is 7 bytes,
1347 * int 0x80 is 2 bytes. We saved this in tf_err.
1349 frame.tf_eip -= frame.tf_err;
1354 panic("Unexpected EASYNC return value (for now)");
1357 if (p->p_sysent->sv_errsize) {
1358 if (error >= p->p_sysent->sv_errsize)
1359 error = -1; /* XXX */
1361 error = p->p_sysent->sv_errtbl[error];
1363 frame.tf_eax = error;
1364 frame.tf_eflags |= PSL_C;
1369 * Traced syscall. trapsignal() is not MP aware.
1371 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1372 frame.tf_eflags &= ~PSL_T;
1373 trapsignal(p, SIGTRAP, 0);
1377 * Handle reschedule and other end-of-syscall issues
1379 userret(p, &frame, sticks);
1382 if (KTRPOINT(td, KTR_SYSRET)) {
1383 ktrsysret(p->p_tracep, code, error, args.sysmsg_result);
1388 * This works because errno is findable through the
1389 * register set. If we ever support an emulation where this
1390 * is not the case, this code will need to be revisited.
1392 STOPEVENT(p, S_SCX, code);
1397 * Release the MP lock if we had to get it
1399 KASSERT(td->td_mpcount == 1, ("badmpcount syscall from %p", (void *)frame.tf_eip));
1405 * free_sysun - Put an unused sysun on the free list.
1407 static __inline void
1408 free_sysun(struct thread *td, union sysunion *sysun)
1410 struct globaldata *gd = td->td_gd;
1412 crit_enter_quick(td);
1413 sysun->lmsg.opaque.ms_sysunnext = gd->gd_freesysun;
1414 gd->gd_freesysun = sysun;
1415 crit_exit_quick(td);
1419 * sendsys2 - MP aware system message request C handler
1422 sendsys2(struct trapframe frame)
1424 struct globaldata *gd;
1425 struct thread *td = curthread;
1426 struct proc *p = td->td_proc;
1427 register_t orig_tf_eflags;
1428 struct sysent *callp;
1429 union sysunion *sysun = NULL;
1439 if (ISPL(frame.tf_cs) != SEL_UPL) {
1447 KASSERT(td->td_mpcount == 0, ("badmpcount syscall from %p", (void *)frame.tf_eip));
1451 * access non-atomic field from critical section. p_sticks is
1452 * updated by the clock interrupt. Also use this opportunity
1453 * to lazy-raise our LWKT priority.
1456 sticks = td->td_sticks;
1458 p->p_md.md_regs = &frame;
1459 orig_tf_eflags = frame.tf_eflags;
1463 * Extract the system call message. If msgsize is zero we are
1464 * blocking on a message and/or message port. If msgsize is -1
1465 * we are testing a message for completion or a message port for
1468 * The userland system call message size includes the size of the
1469 * userland lwkt_msg plus arguments. We load it into the userland
1470 * portion of our sysunion structure then we initialize the kerneland
1477 if ((msgsize = frame.tf_edx) < sizeof(struct lwkt_msg) ||
1478 msgsize > sizeof(union sysunion) - sizeof(struct sysmsg)) {
1484 * Obtain a sysun from our per-cpu cache or allocate a new one. Use
1485 * the opaque field to store the original (user) message pointer.
1486 * A critical section is necessary to interlock against interrupts
1487 * returning system messages to the thread cache.
1490 crit_enter_quick(td);
1491 if ((sysun = gd->gd_freesysun) != NULL)
1492 gd->gd_freesysun = sysun->lmsg.opaque.ms_sysunnext;
1494 sysun = malloc(sizeof(union sysunion), M_SYSMSG, M_WAITOK);
1495 crit_exit_quick(td);
1498 * Copy the user request into the kernel copy of the user request.
1500 umsg = (void *)frame.tf_ecx;
1501 error = copyin(umsg, &sysun->nosys.usrmsg, msgsize);
1504 if ((sysun->nosys.usrmsg.umsg.ms_flags & MSGF_ASYNC)) {
1509 if (max_sysmsg > 0 && p->p_num_sysmsg >= max_sysmsg) {
1516 * Initialize the kernel message from the copied-in data and
1517 * pull in appropriate flags from the userland message.
1519 * ms_abort_port is usually initialized in sendmsg/domsg, but since
1520 * we are not calling those functions (yet), we have to do it manually.
1522 lwkt_initmsg(&sysun->lmsg, &td->td_msgport, 0,
1523 sysun->nosys.usrmsg.umsg.ms_cmd,
1525 sysun->lmsg.ms_abort_port = sysun->lmsg.ms_reply_port;
1526 sysun->sysmsg_copyout = NULL;
1527 sysun->lmsg.opaque.ms_umsg = umsg;
1528 sysun->lmsg.ms_flags |= sysun->nosys.usrmsg.umsg.ms_flags & MSGF_ASYNC;
1531 * Extract the system call number, lookup the system call, and
1532 * set the default return value.
1534 code = (u_int)sysun->lmsg.ms_cmd.cm_op;
1535 /* We don't handle the syscall() syscall yet */
1538 free_sysun(td, sysun);
1541 if (code >= p->p_sysent->sv_size) {
1543 free_sysun(td, sysun);
1547 callp = &p->p_sysent->sv_table[code];
1549 narg = (msgsize - sizeof(struct lwkt_msg)) / sizeof(register_t);
1552 if (KTRPOINT(td, KTR_SYSCALL)) {
1553 ktrsyscall(p->p_tracep, code, narg, (void *)(&sysun->nosys.usrmsg + 1));
1556 sysun->lmsg.u.ms_fds[0] = 0;
1557 sysun->lmsg.u.ms_fds[1] = 0;
1559 STOPEVENT(p, S_SCE, narg); /* MP aware */
1562 * Make the system call. An error code is always returned, results
1563 * are copied back via ms_result32 or ms_result64. YYY temporary
1564 * stage copy p_retval[] into ms_result32/64
1566 * NOTE! XXX if this is a child returning from a fork curproc
1567 * might be different. YYY huh? a child returning from a fork
1568 * should never 'return' from this call, it should go right to the
1569 * fork_trampoline function.
1571 error = (*callp->sy_call)(sysun);
1572 gd = td->td_gd; /* RELOAD, might have switched cpus */
1576 * If a synchronous return copy p_retval to ms_result64 and return
1577 * the sysmsg to the free pool.
1579 * YYY Don't writeback message if execve() YYY
1581 sysun->nosys.usrmsg.umsg.ms_error = error;
1582 sysun->nosys.usrmsg.umsg.u.ms_fds[0] = sysun->lmsg.u.ms_fds[0];
1583 sysun->nosys.usrmsg.umsg.u.ms_fds[1] = sysun->lmsg.u.ms_fds[1];
1584 result = sysun->nosys.usrmsg.umsg.u.ms_fds[0]; /* for ktrace */
1585 if (error != 0 || code != SYS_execve) {
1587 error2 = copyout(&sysun->nosys.usrmsg.umsg.ms_copyout_start,
1588 &umsg->ms_copyout_start,
1593 if (error == EASYNC) {
1595 * Since only the current process ever messes with msgq,
1596 * we can safely manipulate it in parallel with the async
1599 TAILQ_INSERT_TAIL(&p->p_sysmsgq, &sysun->sysmsg, msgq);
1601 error = (int)&sysun->sysmsg;
1604 free_sysun(td, sysun);
1607 frame.tf_eax = (register_t)error;
1610 * Traced syscall. trapsignal() is not MP aware.
1612 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1613 frame.tf_eflags &= ~PSL_T;
1614 trapsignal(p, SIGTRAP, 0);
1618 * Handle reschedule and other end-of-syscall issues
1620 userret(p, &frame, sticks);
1623 if (KTRPOINT(td, KTR_SYSRET)) {
1624 ktrsysret(p->p_tracep, code, error, result);
1629 * This works because errno is findable through the
1630 * register set. If we ever support an emulation where this
1631 * is not the case, this code will need to be revisited.
1633 STOPEVENT(p, S_SCX, code);
1638 * Release the MP lock if we had to get it
1640 KASSERT(td->td_mpcount == 1, ("badmpcount syscall from %p", (void *)frame.tf_eip));
1646 * waitsys2 - MP aware system message wait C handler
1649 waitsys2(struct trapframe frame)
1651 struct globaldata *gd;
1652 struct thread *td = curthread;
1653 struct proc *p = td->td_proc;
1654 union sysunion *sysun = NULL;
1656 register_t orig_tf_eflags;
1657 int error = 0, result, sticks;
1661 if (ISPL(frame.tf_cs) != SEL_UPL) {
1669 KASSERT(td->td_mpcount == 0, ("badmpcount syscall from %p",
1670 (void *)frame.tf_eip));
1675 * access non-atomic field from critical section. p_sticks is
1676 * updated by the clock interrupt. Also use this opportunity
1677 * to lazy-raise our LWKT priority.
1680 sticks = td->td_sticks;
1682 p->p_md.md_regs = &frame;
1683 orig_tf_eflags = frame.tf_eflags;
1689 TAILQ_FOREACH(ptr, &p->p_sysmsgq, msgq) {
1690 if ((void *)ptr == (void *)frame.tf_ecx) {
1691 sysun = (void *)sysmsg_wait(p,
1692 (void *)frame.tf_ecx, 1);
1702 else if (frame.tf_eax) {
1703 printf("waitport/checkport only the default port is supported at the moment\n");
1708 switch(frame.tf_edx) {
1710 sysun = (void *)sysmsg_wait(p, NULL, 0);
1713 sysun = (void *)sysmsg_wait(p, NULL, 1);
1722 umsg = sysun->lmsg.opaque.ms_umsg;
1723 frame.tf_eax = (register_t)sysun;
1724 sysun->nosys.usrmsg.umsg.u.ms_fds[0] = sysun->lmsg.u.ms_fds[0];
1725 sysun->nosys.usrmsg.umsg.u.ms_fds[1] = sysun->lmsg.u.ms_fds[1];
1726 sysun->nosys.usrmsg.umsg.ms_error = sysun->lmsg.ms_error;
1727 error = sysun->lmsg.ms_error;
1728 result = sysun->lmsg.u.ms_fds[0]; /* for ktrace */
1729 error = copyout(&sysun->nosys.usrmsg.umsg.ms_copyout_start,
1730 &umsg->ms_copyout_start, ms_copyout_size);
1731 free_sysun(td, sysun);
1733 code = (u_int)sysun->lmsg.ms_cmd.cm_op;
1737 frame.tf_eax = error;
1739 * Traced syscall. trapsignal() is not MP aware.
1741 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1742 frame.tf_eflags &= ~PSL_T;
1743 trapsignal(p, SIGTRAP, 0);
1747 * Handle reschedule and other end-of-syscall issues
1749 userret(p, &frame, sticks);
1752 if (KTRPOINT(td, KTR_SYSRET)) {
1753 ktrsysret(p->p_tracep, code, error, result);
1758 * This works because errno is findable through the
1759 * register set. If we ever support an emulation where this
1760 * is not the case, this code will need to be revisited.
1762 STOPEVENT(p, S_SCX, code);
1766 KASSERT(td->td_mpcount == 1, ("badmpcount syscall from %p",
1767 (void *)frame.tf_eip));
1773 * Simplified back end of syscall(), used when returning from fork()
1774 * directly into user mode. MP lock is held on entry and should be
1775 * released on return. This code will return back into the fork
1776 * trampoline code which then runs doreti.
1779 fork_return(p, frame)
1781 struct trapframe frame;
1783 frame.tf_eax = 0; /* Child returns zero */
1784 frame.tf_eflags &= ~PSL_C; /* success */
1788 * Newly forked processes are given a kernel priority. We have to
1789 * adjust the priority to a normal user priority and fake entry
1790 * into the kernel (call userenter()) to install a passive release
1791 * function just in case userret() decides to stop the process. This
1792 * can occur when ^Z races a fork. If we do not install the passive
1793 * release function the current process designation will not be
1794 * released when the thread goes to sleep.
1796 lwkt_setpri_self(TDPRI_USER_NORM);
1797 userenter(p->p_thread);
1798 userret(p, &frame, 0);
1800 if (KTRPOINT(p->p_thread, KTR_SYSRET))
1801 ktrsysret(p->p_tracep, SYS_fork, 0, 0);
1803 p->p_flag |= P_PASSIVE_ACQ;
1805 p->p_flag &= ~P_PASSIVE_ACQ;
1807 KKASSERT(p->p_thread->td_mpcount == 1);