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/vkernel/i386/trap.c,v 1.35 2008/09/09 04:06:19 dillon Exp $
43 * AMD64 Trap and System call handling
49 #include "opt_ktrace.h"
51 #include <sys/param.h>
52 #include <sys/systm.h>
54 #include <sys/pioctl.h>
55 #include <sys/kernel.h>
56 #include <sys/resourcevar.h>
57 #include <sys/signalvar.h>
58 #include <sys/signal2.h>
59 #include <sys/syscall.h>
60 #include <sys/sysctl.h>
61 #include <sys/sysent.h>
63 #include <sys/vmmeter.h>
64 #include <sys/malloc.h>
66 #include <sys/ktrace.h>
69 #include <sys/upcall.h>
70 #include <sys/vkernel.h>
71 #include <sys/sysproto.h>
72 #include <sys/sysunion.h>
73 #include <sys/vmspace.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/md_var.h>
86 #include <machine/pcb.h>
87 #include <machine/smp.h>
88 #include <machine/tss.h>
89 #include <machine/globaldata.h>
93 #include <sys/msgport2.h>
94 #include <sys/thread2.h>
95 #include <sys/mplock2.h>
99 #define MAKEMPSAFE(have_mplock) \
100 if (have_mplock == 0) { \
107 #define MAKEMPSAFE(have_mplock)
111 int (*pmath_emulate) (struct trapframe *);
113 extern int trapwrite (unsigned addr);
115 static int trap_pfault (struct trapframe *, int, vm_offset_t);
116 static void trap_fatal (struct trapframe *, int, vm_offset_t);
117 void dblfault_handler (void);
120 extern inthand_t IDTVEC(syscall);
123 #define MAX_TRAP_MSG 30
124 static char *trap_msg[] = {
126 "privileged instruction fault", /* 1 T_PRIVINFLT */
128 "breakpoint instruction fault", /* 3 T_BPTFLT */
131 "arithmetic trap", /* 6 T_ARITHTRAP */
132 "system forced exception", /* 7 T_ASTFLT */
134 "general protection fault", /* 9 T_PROTFLT */
135 "trace trap", /* 10 T_TRCTRAP */
137 "page fault", /* 12 T_PAGEFLT */
139 "alignment fault", /* 14 T_ALIGNFLT */
143 "integer divide fault", /* 18 T_DIVIDE */
144 "non-maskable interrupt trap", /* 19 T_NMI */
145 "overflow trap", /* 20 T_OFLOW */
146 "FPU bounds check fault", /* 21 T_BOUND */
147 "FPU device not available", /* 22 T_DNA */
148 "double fault", /* 23 T_DOUBLEFLT */
149 "FPU operand fetch fault", /* 24 T_FPOPFLT */
150 "invalid TSS fault", /* 25 T_TSSFLT */
151 "segment not present fault", /* 26 T_SEGNPFLT */
152 "stack fault", /* 27 T_STKFLT */
153 "machine check trap", /* 28 T_MCHK */
154 "SIMD floating-point exception", /* 29 T_XMMFLT */
155 "reserved (unknown) fault", /* 30 T_RESERVED */
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 static int syscall_mpsafe = 1;
174 SYSCTL_INT(_kern, OID_AUTO, syscall_mpsafe, CTLFLAG_RW,
175 &syscall_mpsafe, 0, "Allow MPSAFE marked syscalls to run without BGL");
176 TUNABLE_INT("kern.syscall_mpsafe", &syscall_mpsafe);
177 static int trap_mpsafe = 1;
178 SYSCTL_INT(_kern, OID_AUTO, trap_mpsafe, CTLFLAG_RW,
179 &trap_mpsafe, 0, "Allow traps to mostly run without the BGL");
180 TUNABLE_INT("kern.trap_mpsafe", &trap_mpsafe);
183 MALLOC_DEFINE(M_SYSMSG, "sysmsg", "sysmsg structure");
184 extern int max_sysmsg;
187 * userenter() passively intercepts the thread switch function to increase
188 * the thread priority from a user priority to a kernel priority, reducing
189 * syscall and trap overhead for the case where no switch occurs.
193 userenter(struct thread *curtd)
195 curtd->td_release = lwkt_passive_release;
199 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
200 * must be completed before we can return to or try to return to userland.
202 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
203 * arithmatic on the delta calculation so the absolute tick values are
204 * truncated to an integer.
207 userret(struct lwp *lp, struct trapframe *frame, int sticks)
209 struct proc *p = lp->lwp_proc;
213 * Charge system time if profiling. Note: times are in microseconds.
214 * This may do a copyout and block, so do it first even though it
215 * means some system time will be charged as user time.
217 if (p->p_flag & P_PROFIL) {
218 addupc_task(p, frame->tf_rip,
219 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
224 * If the jungle wants us dead, so be it.
226 if (lp->lwp_flag & LWP_WEXIT) {
229 rel_mplock(); /* NOT REACHED */
233 * Block here if we are in a stopped state.
235 if (p->p_stat == SSTOP) {
243 * Post any pending upcalls
245 if (p->p_flag & P_UPCALLPEND) {
247 p->p_flag &= ~P_UPCALLPEND;
254 * Post any pending signals
256 if ((sig = CURSIG_TRACE(lp)) != 0) {
264 * block here if we are swapped out, but still process signals
265 * (such as SIGKILL). proc0 (the swapin scheduler) is already
266 * aware of our situation, we do not have to wake it up.
268 if (p->p_flag & P_SWAPPEDOUT) {
270 p->p_flag |= P_SWAPWAIT;
272 if (p->p_flag & P_SWAPWAIT)
273 tsleep(p, PCATCH, "SWOUT", 0);
274 p->p_flag &= ~P_SWAPWAIT;
280 * Make sure postsig() handled request to restore old signal mask after
281 * running signal handler.
283 KKASSERT((lp->lwp_flag & LWP_OLDMASK) == 0);
287 * Cleanup from userenter and any passive release that might have occured.
288 * We must reclaim the current-process designation before we can return
289 * to usermode. We also handle both LWKT and USER reschedule requests.
292 userexit(struct lwp *lp)
294 struct thread *td = lp->lwp_thread;
295 /* globaldata_t gd = td->td_gd; */
298 * Handle stop requests at kernel priority. Any requests queued
299 * after this loop will generate another AST.
301 while (lp->lwp_proc->p_stat == SSTOP) {
308 * Reduce our priority in preparation for a return to userland. If
309 * our passive release function was still in place, our priority was
310 * never raised and does not need to be reduced.
312 lwkt_passive_recover(td);
315 * Become the current user scheduled process if we aren't already,
316 * and deal with reschedule requests and other factors.
318 lp->lwp_proc->p_usched->acquire_curproc(lp);
319 /* WARNING: we may have migrated cpu's */
320 /* gd = td->td_gd; */
323 #if !defined(KTR_KERNENTRY)
324 #define KTR_KERNENTRY KTR_ALL
326 KTR_INFO_MASTER(kernentry);
327 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, "pid=%d, tid=%d, trapno=%d, eva=%p",
328 sizeof(int) + sizeof(int) + sizeof(int) + sizeof(vm_offset_t));
329 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "pid=%d, tid=%d",
330 sizeof(int) + sizeof(int));
331 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "pid=%d, tid=%d, call=%d",
332 sizeof(int) + sizeof(int) + sizeof(int));
333 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "pid=%d, tid=%d, err=%d",
334 sizeof(int) + sizeof(int) + sizeof(int));
335 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "pid=%d, tid=%d",
336 sizeof(int) + sizeof(int));
339 * Exception, fault, and trap interface to the kernel.
340 * This common code is called from assembly language IDT gate entry
341 * routines that prepare a suitable stack frame, and restore this
342 * frame after the exception has been processed.
344 * This function is also called from doreti in an interlock to handle ASTs.
345 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
347 * NOTE! We have to retrieve the fault address prior to obtaining the
348 * MP lock because get_mplock() may switch out. YYY cr2 really ought
349 * to be retrieved by the assembly code, not here.
351 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
352 * if an attempt is made to switch from a fast interrupt or IPI. This is
353 * necessary to properly take fatal kernel traps on SMP machines if
354 * get_mplock() has to block.
358 user_trap(struct trapframe *frame)
360 struct globaldata *gd = mycpu;
361 struct thread *td = gd->gd_curthread;
362 struct lwp *lp = td->td_lwp;
365 int i = 0, ucode = 0, type, code;
370 int crit_count = td->td_pri & ~TDPRI_MASK;
376 if (frame->tf_trapno == T_PAGEFLT)
377 eva = frame->tf_addr;
381 kprintf("USER_TRAP AT %08lx xflags %ld trapno %ld eva %08lx\n",
382 frame->tf_rip, frame->tf_xflags, frame->tf_trapno, eva);
386 * Everything coming from user mode runs through user_trap,
387 * including system calls.
389 if (frame->tf_trapno == T_FAST_SYSCALL) {
394 KTR_LOG(kernentry_trap, lp->lwp_proc->p_pid, lp->lwp_tid,
395 frame->tf_trapno, eva);
399 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
400 ++gd->gd_trap_nesting_level;
401 MAKEMPSAFE(have_mplock);
402 trap_fatal(frame, TRUE, eva);
403 --gd->gd_trap_nesting_level;
408 ++gd->gd_trap_nesting_level;
410 if (trap_mpsafe == 0)
411 MAKEMPSAFE(have_mplock);
414 --gd->gd_trap_nesting_level;
416 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
419 type = frame->tf_trapno;
420 code = frame->tf_err;
424 sticks = (int)td->td_sticks;
425 lp->lwp_md.md_regs = frame;
428 case T_PRIVINFLT: /* privileged instruction fault */
433 case T_BPTFLT: /* bpt instruction fault */
434 case T_TRCTRAP: /* trace trap */
435 frame->tf_rflags &= ~PSL_T;
439 case T_ARITHTRAP: /* arithmetic trap */
444 case T_ASTFLT: /* Allow process switch */
445 mycpu->gd_cnt.v_soft++;
446 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
447 atomic_clear_int_nonlocked(&mycpu->gd_reqflags,
449 addupc_task(p, p->p_prof.pr_addr,
455 * The following two traps can happen in
456 * vm86 mode, and, if so, we want to handle
459 case T_PROTFLT: /* general protection fault */
460 case T_STKFLT: /* stack fault */
462 if (frame->tf_eflags & PSL_VM) {
463 i = vm86_emulate((struct vm86frame *)frame);
471 case T_SEGNPFLT: /* segment not present fault */
472 case T_TSSFLT: /* invalid TSS fault */
473 case T_DOUBLEFLT: /* double fault */
475 ucode = code + BUS_SEGM_FAULT ;
479 case T_PAGEFLT: /* page fault */
480 MAKEMPSAFE(have_mplock);
481 i = trap_pfault(frame, TRUE, eva);
484 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
494 case T_DIVIDE: /* integer divide fault */
501 MAKEMPSAFE(have_mplock);
502 /* machine/parity/power fail/"kitchen sink" faults */
503 if (isa_nmi(code) == 0) {
506 * NMI can be hooked up to a pushbutton
510 kprintf ("NMI ... going to debugger\n");
511 kdb_trap (type, 0, frame);
515 } else if (panic_on_nmi)
516 panic("NMI indicates hardware failure");
518 #endif /* NISA > 0 */
520 case T_OFLOW: /* integer overflow fault */
525 case T_BOUND: /* bounds check fault */
532 * Virtual kernel intercept - pass the DNA exception
533 * to the (emulated) virtual kernel if it asked to handle
534 * it. This occurs when the virtual kernel is holding
535 * onto the FP context for a different emulated
536 * process then the one currently running.
538 * We must still call npxdna() since we may have
539 * saved FP state that the (emulated) virtual kernel
540 * needs to hand over to a different emulated process.
542 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
543 (td->td_pcb->pcb_flags & FP_VIRTFP)
549 * The kernel may have switched out the FP unit's
550 * state, causing the user process to take a fault
551 * when it tries to use the FP unit. Restore the
556 if (!pmath_emulate) {
558 ucode = FPE_FPU_NP_TRAP;
561 i = (*pmath_emulate)(frame);
563 if (!(frame->tf_rflags & PSL_T))
565 frame->tf_rflags &= ~PSL_T;
568 /* else ucode = emulator_only_knows() XXX */
571 case T_FPOPFLT: /* FPU operand fetch fault */
576 case T_XMMFLT: /* SIMD floating-point exception */
583 * Virtual kernel intercept - if the fault is directly related to a
584 * VM context managed by a virtual kernel then let the virtual kernel
587 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
588 vkernel_trap(lp, frame);
593 * Translate fault for emulators (e.g. Linux)
595 if (*p->p_sysent->sv_transtrap)
596 i = (*p->p_sysent->sv_transtrap)(i, type);
598 MAKEMPSAFE(have_mplock);
599 trapsignal(lp, i, ucode);
602 if (type <= MAX_TRAP_MSG) {
603 uprintf("fatal process exception: %s",
605 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
606 uprintf(", fault VA = 0x%lx", (u_long)eva);
613 KASSERT(td->td_mpcount == have_mplock, ("badmpcount trap/end from %p", (void *)frame->tf_rip));
615 userret(lp, frame, sticks);
622 KTR_LOG(kernentry_trap_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
624 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
625 ("syscall: critical section count mismatch! %d/%d",
626 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
631 kern_trap(struct trapframe *frame)
633 struct globaldata *gd = mycpu;
634 struct thread *td = gd->gd_curthread;
637 int i = 0, ucode = 0, type, code;
642 int crit_count = td->td_pri & ~TDPRI_MASK;
649 if (frame->tf_trapno == T_PAGEFLT)
650 eva = frame->tf_addr;
656 ++gd->gd_trap_nesting_level;
657 MAKEMPSAFE(have_mplock);
658 trap_fatal(frame, FALSE, eva);
659 --gd->gd_trap_nesting_level;
664 ++gd->gd_trap_nesting_level;
667 if (trap_mpsafe == 0)
668 MAKEMPSAFE(have_mplock);
671 --gd->gd_trap_nesting_level;
673 type = frame->tf_trapno;
674 code = frame->tf_err;
682 case T_PAGEFLT: /* page fault */
683 MAKEMPSAFE(have_mplock);
684 trap_pfault(frame, FALSE, eva);
689 * The kernel may be using npx for copying or other
692 panic("kernel NPX should not happen");
697 case T_PROTFLT: /* general protection fault */
698 case T_SEGNPFLT: /* segment not present fault */
700 * Invalid segment selectors and out of bounds
701 * %eip's and %esp's can be set up in user mode.
702 * This causes a fault in kernel mode when the
703 * kernel tries to return to user mode. We want
704 * to get this fault so that we can fix the
705 * problem here and not have to check all the
706 * selectors and pointers when the user changes
709 if (mycpu->gd_intr_nesting_level == 0) {
710 if (td->td_pcb->pcb_onfault) {
712 (register_t)td->td_pcb->pcb_onfault;
720 * PSL_NT can be set in user mode and isn't cleared
721 * automatically when the kernel is entered. This
722 * causes a TSS fault when the kernel attempts to
723 * `iret' because the TSS link is uninitialized. We
724 * want to get this fault so that we can fix the
725 * problem here and not every time the kernel is
728 if (frame->tf_rflags & PSL_NT) {
729 frame->tf_rflags &= ~PSL_NT;
734 case T_TRCTRAP: /* trace trap */
736 if (frame->tf_eip == (int)IDTVEC(syscall)) {
738 * We've just entered system mode via the
739 * syscall lcall. Continue single stepping
740 * silently until the syscall handler has
745 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
747 * The syscall handler has now saved the
748 * flags. Stop single stepping it.
750 frame->tf_eflags &= ~PSL_T;
756 * Ignore debug register trace traps due to
757 * accesses in the user's address space, which
758 * can happen under several conditions such as
759 * if a user sets a watchpoint on a buffer and
760 * then passes that buffer to a system call.
761 * We still want to get TRCTRAPS for addresses
762 * in kernel space because that is useful when
763 * debugging the kernel.
765 if (user_dbreg_trap()) {
767 * Reset breakpoint bits because the
770 load_dr6(rdr6() & 0xfffffff0);
775 * Fall through (TRCTRAP kernel mode, kernel address)
779 * If DDB is enabled, let it handle the debugger trap.
780 * Otherwise, debugger traps "can't happen".
783 MAKEMPSAFE(have_mplock);
784 if (kdb_trap (type, 0, frame))
789 MAKEMPSAFE(have_mplock);
790 trap_fatal(frame, FALSE, eva);
793 MAKEMPSAFE(have_mplock);
794 trap_fatal(frame, FALSE, eva);
799 * Ignore this trap generated from a spurious SIGTRAP.
801 * single stepping in / syscalls leads to spurious / SIGTRAP
804 * Haiku (c) 2007 Simon 'corecode' Schubert
810 * Translate fault for emulators (e.g. Linux)
812 if (*p->p_sysent->sv_transtrap)
813 i = (*p->p_sysent->sv_transtrap)(i, type);
815 MAKEMPSAFE(have_mplock);
816 trapsignal(lp, i, ucode);
819 if (type <= MAX_TRAP_MSG) {
820 uprintf("fatal process exception: %s",
822 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
823 uprintf(", fault VA = 0x%lx", (u_long)eva);
835 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
836 ("syscall: critical section count mismatch! %d/%d",
837 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
842 trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
845 struct vmspace *vm = NULL;
849 thread_t td = curthread;
850 struct lwp *lp = td->td_lwp;
852 va = trunc_page(eva);
853 if (usermode == FALSE) {
855 * This is a fault on kernel virtual memory.
860 * This is a fault on non-kernel virtual memory.
861 * vm is initialized above to NULL. If curproc is NULL
862 * or curproc->p_vmspace is NULL the fault is fatal.
865 vm = lp->lwp_vmspace;
873 if (frame->tf_err & PGEX_W)
874 ftype = VM_PROT_READ | VM_PROT_WRITE;
876 ftype = VM_PROT_READ;
878 if (map != &kernel_map) {
880 * Keep swapout from messing with us during this
886 * Grow the stack if necessary
888 /* grow_stack returns false only if va falls into
889 * a growable stack region and the stack growth
890 * fails. It returns true if va was not within
891 * a growable stack region, or if the stack
894 if (!grow_stack (lp->lwp_proc, va)) {
900 /* Fault in the user page: */
901 rv = vm_fault(map, va, ftype,
902 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
908 * Don't have to worry about process locking or stacks in the kernel.
910 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
913 if (rv == KERN_SUCCESS)
917 if (td->td_gd->gd_intr_nesting_level == 0 &&
918 td->td_pcb->pcb_onfault) {
919 frame->tf_rip = (register_t)td->td_pcb->pcb_onfault;
922 trap_fatal(frame, usermode, eva);
927 * NOTE: on amd64 we have a tf_addr field in the trapframe, no
928 * kludge is needed to pass the fault address to signal handlers.
930 struct proc *p = td->td_proc;
931 kprintf("seg-fault accessing address %p rip=%p pid=%d p_comm=%s\n",
932 (void *)va, (void *)frame->tf_rip, p->p_pid, p->p_comm);
933 /* Debugger("seg-fault"); */
935 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
939 trap_fatal(struct trapframe *frame, int usermode, vm_offset_t eva)
944 code = frame->tf_xflags;
945 type = frame->tf_trapno;
947 if (type <= MAX_TRAP_MSG) {
948 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
949 type, trap_msg[type],
950 (usermode ? "user" : "kernel"));
953 /* two separate prints in case of a trap on an unmapped page */
954 kprintf("mp_lock = %08x; ", mp_lock);
955 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
957 if (type == T_PAGEFLT) {
958 kprintf("fault virtual address = %p\n", (void *)eva);
959 kprintf("fault code = %s %s, %s\n",
960 usermode ? "user" : "supervisor",
961 code & PGEX_W ? "write" : "read",
962 code & PGEX_P ? "protection violation" : "page not present");
964 kprintf("instruction pointer = 0x%lx:0x%lx\n",
965 frame->tf_cs & 0xffff, frame->tf_rip);
967 ss = frame->tf_ss & 0xffff;
970 ss = GSEL(GDATA_SEL, SEL_KPL);
971 rsp = (long)&frame->tf_rsp;
973 kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp);
974 kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp);
975 kprintf("processor eflags = ");
976 if (frame->tf_rflags & PSL_T)
977 kprintf("trace trap, ");
978 if (frame->tf_rflags & PSL_I)
979 kprintf("interrupt enabled, ");
980 if (frame->tf_rflags & PSL_NT)
981 kprintf("nested task, ");
982 if (frame->tf_rflags & PSL_RF)
985 if (frame->tf_eflags & PSL_VM)
988 kprintf("IOPL = %jd\n", (intmax_t)((frame->tf_rflags & PSL_IOPL) >> 12));
989 kprintf("current process = ");
991 kprintf("%lu (%s)\n",
992 (u_long)curproc->p_pid, curproc->p_comm ?
993 curproc->p_comm : "");
997 kprintf("current thread = pri %d ", curthread->td_pri);
998 if (curthread->td_pri >= TDPRI_CRIT)
1004 * we probably SHOULD have stopped the other CPUs before now!
1005 * another CPU COULD have been touching cpl at this moment...
1007 kprintf(" <- SMP: XXX");
1016 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1019 kprintf("trap number = %d\n", type);
1020 if (type <= MAX_TRAP_MSG)
1021 panic("%s", trap_msg[type]);
1023 panic("unknown/reserved trap");
1027 * Double fault handler. Called when a fault occurs while writing
1028 * a frame for a trap/exception onto the stack. This usually occurs
1029 * when the stack overflows (such is the case with infinite recursion,
1032 * XXX Note that the current PTD gets replaced by IdlePTD when the
1033 * task switch occurs. This means that the stack that was active at
1034 * the time of the double fault is not available at <kstack> unless
1035 * the machine was idle when the double fault occurred. The downside
1036 * of this is that "trace <ebp>" in ddb won't work.
1039 dblfault_handler(void)
1042 struct mdglobaldata *gd = mdcpu;
1045 kprintf("\nFatal double fault:\n");
1047 kprintf("rip = 0x%lx\n", gd->gd_common_tss.tss_rip);
1048 kprintf("rsp = 0x%lx\n", gd->gd_common_tss.tss_rsp);
1049 kprintf("rbp = 0x%lx\n", gd->gd_common_tss.tss_rbp);
1052 /* two separate prints in case of a trap on an unmapped page */
1053 kprintf("mp_lock = %08x; ", mp_lock);
1054 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
1056 panic("double fault");
1060 * Compensate for 386 brain damage (missing URKR).
1061 * This is a little simpler than the pagefault handler in trap() because
1062 * it the page tables have already been faulted in and high addresses
1063 * are thrown out early for other reasons.
1066 trapwrite(unsigned addr)
1073 va = trunc_page((vm_offset_t)addr);
1075 * XXX - MAX is END. Changed > to >= for temp. fix.
1077 if (va >= VM_MAX_USER_ADDRESS)
1080 lp = curthread->td_lwp;
1081 vm = lp->lwp_vmspace;
1083 PHOLD(lp->lwp_proc);
1085 if (!grow_stack (lp->lwp_proc, va)) {
1086 PRELE(lp->lwp_proc);
1091 * fault the data page
1093 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY);
1095 PRELE(lp->lwp_proc);
1097 if (rv != KERN_SUCCESS)
1104 * syscall2 - MP aware system call request C handler
1106 * A system call is essentially treated as a trap except that the
1107 * MP lock is not held on entry or return. We are responsible for
1108 * obtaining the MP lock if necessary and for handling ASTs
1109 * (e.g. a task switch) prior to return.
1111 * In general, only simple access and manipulation of curproc and
1112 * the current stack is allowed without having to hold MP lock.
1114 * MPSAFE - note that large sections of this routine are run without
1118 syscall2(struct trapframe *frame)
1120 struct thread *td = curthread;
1121 struct proc *p = td->td_proc;
1122 struct lwp *lp = td->td_lwp;
1124 struct sysent *callp;
1125 register_t orig_tf_rflags;
1130 int crit_count = td->td_pri & ~TDPRI_MASK;
1133 int have_mplock = 0;
1138 union sysunion args;
1139 register_t *argsdst;
1141 mycpu->gd_cnt.v_syscall++;
1143 KTR_LOG(kernentry_syscall, lp->lwp_proc->p_pid, lp->lwp_tid,
1147 KASSERT(td->td_mpcount == 0, ("badmpcount syscall2 from %p", (void *)frame->tf_rip));
1148 if (syscall_mpsafe == 0)
1149 MAKEMPSAFE(have_mplock);
1151 userenter(td); /* lazy raise our priority */
1158 sticks = (int)td->td_sticks;
1159 orig_tf_rflags = frame->tf_rflags;
1162 * Virtual kernel intercept - if a VM context managed by a virtual
1163 * kernel issues a system call the virtual kernel handles it, not us.
1164 * Restore the virtual kernel context and return from its system
1165 * call. The current frame is copied out to the virtual kernel.
1167 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1168 vkernel_trap(lp, frame);
1169 error = EJUSTRETURN;
1174 * Get the system call parameters and account for time
1176 lp->lwp_md.md_regs = frame;
1177 params = (caddr_t)frame->tf_rsp + sizeof(register_t);
1178 code = frame->tf_rax;
1180 if (p->p_sysent->sv_prepsyscall) {
1181 (*p->p_sysent->sv_prepsyscall)(
1182 frame, (int *)(&args.nosys.sysmsg + 1),
1185 if (code == SYS_syscall || code == SYS___syscall) {
1186 code = frame->tf_rdi;
1192 if (p->p_sysent->sv_mask)
1193 code &= p->p_sysent->sv_mask;
1195 if (code >= p->p_sysent->sv_size)
1196 callp = &p->p_sysent->sv_table[0];
1198 callp = &p->p_sysent->sv_table[code];
1200 narg = callp->sy_narg & SYF_ARGMASK;
1203 * On amd64 we get up to six arguments in registers. The rest are
1204 * on the stack. The first six members of 'struct trapframe' happen
1205 * to be the registers used to pass arguments, in exactly the right
1208 argp = &frame->tf_rdi;
1210 argsdst = (register_t *)(&args.nosys.sysmsg + 1);
1212 * JG can we overflow the space pointed to by 'argsdst'
1213 * either with 'bcopy' or with 'copyin'?
1215 bcopy(argp, argsdst, sizeof(register_t) * regcnt);
1217 * copyin is MP aware, but the tracing code is not
1219 if (narg > regcnt) {
1220 KASSERT(params != NULL, ("copyin args with no params!"));
1221 error = copyin(params, &argsdst[regcnt],
1222 (narg - regcnt) * sizeof(register_t));
1225 if (KTRPOINT(td, KTR_SYSCALL)) {
1226 MAKEMPSAFE(have_mplock);
1228 ktrsyscall(lp, code, narg,
1229 (void *)(&args.nosys.sysmsg + 1));
1237 if (KTRPOINT(td, KTR_SYSCALL)) {
1238 MAKEMPSAFE(have_mplock);
1239 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1244 * Default return value is 0 (will be copied to %rax). Double-value
1245 * returns use %rax and %rdx. %rdx is left unchanged for system
1246 * calls which return only one result.
1248 args.sysmsg_fds[0] = 0;
1249 args.sysmsg_fds[1] = frame->tf_rdx;
1252 * The syscall might manipulate the trap frame. If it does it
1253 * will probably return EJUSTRETURN.
1255 args.sysmsg_frame = frame;
1257 STOPEVENT(p, S_SCE, narg); /* MP aware */
1260 * NOTE: All system calls run MPSAFE now. The system call itself
1261 * is responsible for getting the MP lock.
1263 error = (*callp->sy_call)(&args);
1266 kprintf("system call %d returned %d\n", code, error);
1271 * MP SAFE (we may or may not have the MP lock at this point)
1276 * Reinitialize proc pointer `p' as it may be different
1277 * if this is a child returning from fork syscall.
1280 lp = curthread->td_lwp;
1281 frame->tf_rax = args.sysmsg_fds[0];
1282 frame->tf_rdx = args.sysmsg_fds[1];
1283 frame->tf_rflags &= ~PSL_C;
1287 * Reconstruct pc, we know that 'syscall' is 2 bytes.
1288 * We have to do a full context restore so that %r10
1289 * (which was holding the value of %rcx) is restored for
1290 * the next iteration.
1292 frame->tf_rip -= frame->tf_err;
1293 frame->tf_r10 = frame->tf_rcx;
1298 panic("Unexpected EASYNC return value (for now)");
1301 if (p->p_sysent->sv_errsize) {
1302 if (error >= p->p_sysent->sv_errsize)
1303 error = -1; /* XXX */
1305 error = p->p_sysent->sv_errtbl[error];
1307 frame->tf_rax = error;
1308 frame->tf_rflags |= PSL_C;
1313 * Traced syscall. trapsignal() is not MP aware.
1315 if (orig_tf_rflags & PSL_T) {
1316 MAKEMPSAFE(have_mplock);
1317 frame->tf_rflags &= ~PSL_T;
1318 trapsignal(lp, SIGTRAP, 0);
1322 * Handle reschedule and other end-of-syscall issues
1324 userret(lp, frame, sticks);
1327 if (KTRPOINT(td, KTR_SYSRET)) {
1328 MAKEMPSAFE(have_mplock);
1329 ktrsysret(lp, code, error, args.sysmsg_result);
1334 * This works because errno is findable through the
1335 * register set. If we ever support an emulation where this
1336 * is not the case, this code will need to be revisited.
1338 STOPEVENT(p, S_SCX, code);
1343 * Release the MP lock if we had to get it
1345 KASSERT(td->td_mpcount == have_mplock,
1346 ("badmpcount syscall2/end from %p", (void *)frame->tf_rip));
1350 KTR_LOG(kernentry_syscall_ret, lp->lwp_proc->p_pid, lp->lwp_tid, error);
1352 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
1353 ("syscall: critical section count mismatch! %d/%d",
1354 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
1359 fork_return(struct lwp *lp, struct trapframe *frame)
1361 frame->tf_rax = 0; /* Child returns zero */
1362 frame->tf_rflags &= ~PSL_C; /* success */
1365 generic_lwp_return(lp, frame);
1366 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1370 * Simplified back end of syscall(), used when returning from fork()
1371 * or lwp_create() directly into user mode. MP lock is held on entry and
1372 * should be released on return. This code will return back into the fork
1373 * trampoline code which then runs doreti.
1376 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1378 struct proc *p = lp->lwp_proc;
1381 * Newly forked processes are given a kernel priority. We have to
1382 * adjust the priority to a normal user priority and fake entry
1383 * into the kernel (call userenter()) to install a passive release
1384 * function just in case userret() decides to stop the process. This
1385 * can occur when ^Z races a fork. If we do not install the passive
1386 * release function the current process designation will not be
1387 * released when the thread goes to sleep.
1389 lwkt_setpri_self(TDPRI_USER_NORM);
1390 userenter(lp->lwp_thread);
1391 userret(lp, frame, 0);
1393 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1394 ktrsysret(lp, SYS_fork, 0, 0);
1396 p->p_flag |= P_PASSIVE_ACQ;
1398 p->p_flag &= ~P_PASSIVE_ACQ;
1400 KKASSERT(lp->lwp_thread->td_mpcount == 1);
1406 * doreti has turned into this. The frame is directly on the stack. We
1407 * pull everything else we need (fpu and tls context) from the current
1410 * Note on fpu interactions: In a virtual kernel, the fpu context for
1411 * an emulated user mode process is not shared with the virtual kernel's
1412 * fpu context, so we only have to 'stack' fpu contexts within the virtual
1413 * kernel itself, and not even then since the signal() contexts that we care
1414 * about save and restore the FPU state (I think anyhow).
1416 * vmspace_ctl() returns an error only if it had problems instaling the
1417 * context we supplied or problems copying data to/from our VM space.
1420 go_user(struct intrframe *frame)
1422 struct trapframe *tf = (void *)&frame->if_rdi;
1426 * Interrupts may be disabled on entry, make sure all signals
1427 * can be received before beginning our loop.
1432 * Switch to the current simulated user process, then call
1433 * user_trap() when we break out of it (usually due to a signal).
1437 * Tell the real kernel whether it is ok to use the FP
1440 if (mdcpu->gd_npxthread == curthread) {
1441 tf->tf_xflags &= ~PGEX_FPFAULT;
1443 tf->tf_xflags |= PGEX_FPFAULT;
1447 * Run emulated user process context. This call interlocks
1448 * with new mailbox signals.
1450 * Set PGEX_U unconditionally, indicating a user frame (the
1451 * bit is normally set only by T_PAGEFLT).
1453 r = vmspace_ctl(&curproc->p_vmspace->vm_pmap, VMSPACE_CTL_RUN,
1454 tf, &curthread->td_savevext);
1455 frame->if_xflags |= PGEX_U;
1457 kprintf("GO USER %d trap %ld EVA %08lx RIP %08lx RSP %08lx XFLAGS %02lx/%02lx\n",
1458 r, tf->tf_trapno, tf->tf_addr, tf->tf_rip, tf->tf_rsp,
1459 tf->tf_xflags, frame->if_xflags);
1463 panic("vmspace_ctl failed error %d", errno);
1465 if (tf->tf_trapno) {
1469 if (mycpu->gd_reqflags & RQF_AST_MASK) {
1470 tf->tf_trapno = T_ASTFLT;
1478 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1479 * fault (which is then passed back to the virtual kernel) if an attempt is
1480 * made to use the FP unit.
1482 * XXX this is a fairly big hack.
1485 set_vkernel_fp(struct trapframe *frame)
1487 struct thread *td = curthread;
1489 if (frame->tf_xflags & PGEX_FPFAULT) {
1490 td->td_pcb->pcb_flags |= FP_VIRTFP;
1491 if (mdcpu->gd_npxthread == td)
1494 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1499 * Called from vkernel_trap() to fixup the vkernel's syscall
1500 * frame for vmspace_ctl() return.
1503 cpu_vkernel_trap(struct trapframe *frame, int error)
1505 frame->tf_rax = error;
1507 frame->tf_rflags |= PSL_C;
1509 frame->tf_rflags &= ~PSL_C;