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>
98 #define MAKEMPSAFE(have_mplock) \
99 if (have_mplock == 0) { \
106 #define MAKEMPSAFE(have_mplock)
110 int (*pmath_emulate) (struct trapframe *);
112 extern int trapwrite (unsigned addr);
114 static int trap_pfault (struct trapframe *, int, vm_offset_t);
115 static void trap_fatal (struct trapframe *, int, vm_offset_t);
116 void dblfault_handler (void);
119 extern inthand_t IDTVEC(syscall);
122 #define MAX_TRAP_MSG 30
123 static char *trap_msg[] = {
125 "privileged instruction fault", /* 1 T_PRIVINFLT */
127 "breakpoint instruction fault", /* 3 T_BPTFLT */
130 "arithmetic trap", /* 6 T_ARITHTRAP */
131 "system forced exception", /* 7 T_ASTFLT */
133 "general protection fault", /* 9 T_PROTFLT */
134 "trace trap", /* 10 T_TRCTRAP */
136 "page fault", /* 12 T_PAGEFLT */
138 "alignment fault", /* 14 T_ALIGNFLT */
142 "integer divide fault", /* 18 T_DIVIDE */
143 "non-maskable interrupt trap", /* 19 T_NMI */
144 "overflow trap", /* 20 T_OFLOW */
145 "FPU bounds check fault", /* 21 T_BOUND */
146 "FPU device not available", /* 22 T_DNA */
147 "double fault", /* 23 T_DOUBLEFLT */
148 "FPU operand fetch fault", /* 24 T_FPOPFLT */
149 "invalid TSS fault", /* 25 T_TSSFLT */
150 "segment not present fault", /* 26 T_SEGNPFLT */
151 "stack fault", /* 27 T_STKFLT */
152 "machine check trap", /* 28 T_MCHK */
153 "SIMD floating-point exception", /* 29 T_XMMFLT */
154 "reserved (unknown) fault", /* 30 T_RESERVED */
158 static int ddb_on_nmi = 1;
159 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
160 &ddb_on_nmi, 0, "Go to DDB on NMI");
162 static int panic_on_nmi = 1;
163 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
164 &panic_on_nmi, 0, "Panic on NMI");
165 static int fast_release;
166 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
167 &fast_release, 0, "Passive Release was optimal");
168 static int slow_release;
169 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
170 &slow_release, 0, "Passive Release was nonoptimal");
172 static int syscall_mpsafe = 1;
173 SYSCTL_INT(_kern, OID_AUTO, syscall_mpsafe, CTLFLAG_RW,
174 &syscall_mpsafe, 0, "Allow MPSAFE marked syscalls to run without BGL");
175 TUNABLE_INT("kern.syscall_mpsafe", &syscall_mpsafe);
176 static int trap_mpsafe = 1;
177 SYSCTL_INT(_kern, OID_AUTO, trap_mpsafe, CTLFLAG_RW,
178 &trap_mpsafe, 0, "Allow traps to mostly run without the BGL");
179 TUNABLE_INT("kern.trap_mpsafe", &trap_mpsafe);
182 MALLOC_DEFINE(M_SYSMSG, "sysmsg", "sysmsg structure");
183 extern int max_sysmsg;
186 * userenter() passively intercepts the thread switch function to increase
187 * the thread priority from a user priority to a kernel priority, reducing
188 * syscall and trap overhead for the case where no switch occurs.
192 userenter(struct thread *curtd)
194 curtd->td_release = lwkt_passive_release;
198 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
199 * must be completed before we can return to or try to return to userland.
201 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
202 * arithmatic on the delta calculation so the absolute tick values are
203 * truncated to an integer.
206 userret(struct lwp *lp, struct trapframe *frame, int sticks)
208 struct proc *p = lp->lwp_proc;
212 * Charge system time if profiling. Note: times are in microseconds.
213 * This may do a copyout and block, so do it first even though it
214 * means some system time will be charged as user time.
216 if (p->p_flag & P_PROFIL) {
217 addupc_task(p, frame->tf_rip,
218 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
223 * If the jungle wants us dead, so be it.
225 if (lp->lwp_flag & LWP_WEXIT) {
228 rel_mplock(); /* NOT REACHED */
232 * Block here if we are in a stopped state.
234 if (p->p_stat == SSTOP) {
242 * Post any pending upcalls
244 if (p->p_flag & P_UPCALLPEND) {
246 p->p_flag &= ~P_UPCALLPEND;
253 * Post any pending signals
255 if ((sig = CURSIG_TRACE(lp)) != 0) {
263 * block here if we are swapped out, but still process signals
264 * (such as SIGKILL). proc0 (the swapin scheduler) is already
265 * aware of our situation, we do not have to wake it up.
267 if (p->p_flag & P_SWAPPEDOUT) {
269 p->p_flag |= P_SWAPWAIT;
271 if (p->p_flag & P_SWAPWAIT)
272 tsleep(p, PCATCH, "SWOUT", 0);
273 p->p_flag &= ~P_SWAPWAIT;
279 * Make sure postsig() handled request to restore old signal mask after
280 * running signal handler.
282 KKASSERT((lp->lwp_flag & LWP_OLDMASK) == 0);
286 * Cleanup from userenter and any passive release that might have occured.
287 * We must reclaim the current-process designation before we can return
288 * to usermode. We also handle both LWKT and USER reschedule requests.
291 userexit(struct lwp *lp)
293 struct thread *td = lp->lwp_thread;
294 /* globaldata_t gd = td->td_gd; */
297 * Handle stop requests at kernel priority. Any requests queued
298 * after this loop will generate another AST.
300 while (lp->lwp_proc->p_stat == SSTOP) {
307 * Reduce our priority in preparation for a return to userland. If
308 * our passive release function was still in place, our priority was
309 * never raised and does not need to be reduced.
311 lwkt_passive_recover(td);
314 * Become the current user scheduled process if we aren't already,
315 * and deal with reschedule requests and other factors.
317 lp->lwp_proc->p_usched->acquire_curproc(lp);
318 /* WARNING: we may have migrated cpu's */
319 /* gd = td->td_gd; */
322 #if !defined(KTR_KERNENTRY)
323 #define KTR_KERNENTRY KTR_ALL
325 KTR_INFO_MASTER(kernentry);
326 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, "pid=%d, tid=%d, trapno=%d, eva=%p",
327 sizeof(int) + sizeof(int) + sizeof(int) + sizeof(vm_offset_t));
328 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "pid=%d, tid=%d",
329 sizeof(int) + sizeof(int));
330 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "pid=%d, tid=%d, call=%d",
331 sizeof(int) + sizeof(int) + sizeof(int));
332 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "pid=%d, tid=%d, err=%d",
333 sizeof(int) + sizeof(int) + sizeof(int));
334 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "pid=%d, tid=%d",
335 sizeof(int) + sizeof(int));
338 * Exception, fault, and trap interface to the kernel.
339 * This common code is called from assembly language IDT gate entry
340 * routines that prepare a suitable stack frame, and restore this
341 * frame after the exception has been processed.
343 * This function is also called from doreti in an interlock to handle ASTs.
344 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
346 * NOTE! We have to retrieve the fault address prior to obtaining the
347 * MP lock because get_mplock() may switch out. YYY cr2 really ought
348 * to be retrieved by the assembly code, not here.
350 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
351 * if an attempt is made to switch from a fast interrupt or IPI. This is
352 * necessary to properly take fatal kernel traps on SMP machines if
353 * get_mplock() has to block.
357 user_trap(struct trapframe *frame)
359 struct globaldata *gd = mycpu;
360 struct thread *td = gd->gd_curthread;
361 struct lwp *lp = td->td_lwp;
364 int i = 0, ucode = 0, type, code;
369 int crit_count = td->td_pri & ~TDPRI_MASK;
375 if (frame->tf_trapno == T_PAGEFLT)
376 eva = frame->tf_addr;
380 kprintf("USER_TRAP AT %08lx xflags %ld trapno %ld eva %08lx\n",
381 frame->tf_rip, frame->tf_xflags, frame->tf_trapno, eva);
385 * Everything coming from user mode runs through user_trap,
386 * including system calls.
388 if (frame->tf_trapno == T_SYSCALL80) {
393 KTR_LOG(kernentry_trap, lp->lwp_proc->p_pid, lp->lwp_tid,
394 frame->tf_trapno, eva);
398 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
399 ++gd->gd_trap_nesting_level;
400 MAKEMPSAFE(have_mplock);
401 trap_fatal(frame, TRUE, eva);
402 --gd->gd_trap_nesting_level;
407 ++gd->gd_trap_nesting_level;
409 if (trap_mpsafe == 0)
410 MAKEMPSAFE(have_mplock);
413 --gd->gd_trap_nesting_level;
415 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
418 type = frame->tf_trapno;
419 code = frame->tf_err;
423 sticks = (int)td->td_sticks;
424 lp->lwp_md.md_regs = frame;
427 case T_PRIVINFLT: /* privileged instruction fault */
432 case T_BPTFLT: /* bpt instruction fault */
433 case T_TRCTRAP: /* trace trap */
434 frame->tf_rflags &= ~PSL_T;
438 case T_ARITHTRAP: /* arithmetic trap */
443 case T_ASTFLT: /* Allow process switch */
444 mycpu->gd_cnt.v_soft++;
445 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
446 atomic_clear_int_nonlocked(&mycpu->gd_reqflags,
448 addupc_task(p, p->p_prof.pr_addr,
454 * The following two traps can happen in
455 * vm86 mode, and, if so, we want to handle
458 case T_PROTFLT: /* general protection fault */
459 case T_STKFLT: /* stack fault */
461 if (frame->tf_eflags & PSL_VM) {
462 i = vm86_emulate((struct vm86frame *)frame);
470 case T_SEGNPFLT: /* segment not present fault */
471 case T_TSSFLT: /* invalid TSS fault */
472 case T_DOUBLEFLT: /* double fault */
474 ucode = code + BUS_SEGM_FAULT ;
478 case T_PAGEFLT: /* page fault */
479 MAKEMPSAFE(have_mplock);
480 i = trap_pfault(frame, TRUE, eva);
483 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
493 case T_DIVIDE: /* integer divide fault */
500 MAKEMPSAFE(have_mplock);
501 /* machine/parity/power fail/"kitchen sink" faults */
502 if (isa_nmi(code) == 0) {
505 * NMI can be hooked up to a pushbutton
509 kprintf ("NMI ... going to debugger\n");
510 kdb_trap (type, 0, frame);
514 } else if (panic_on_nmi)
515 panic("NMI indicates hardware failure");
517 #endif /* NISA > 0 */
519 case T_OFLOW: /* integer overflow fault */
524 case T_BOUND: /* bounds check fault */
531 * Virtual kernel intercept - pass the DNA exception
532 * to the (emulated) virtual kernel if it asked to handle
533 * it. This occurs when the virtual kernel is holding
534 * onto the FP context for a different emulated
535 * process then the one currently running.
537 * We must still call npxdna() since we may have
538 * saved FP state that the (emulated) virtual kernel
539 * needs to hand over to a different emulated process.
541 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
542 (td->td_pcb->pcb_flags & FP_VIRTFP)
548 * The kernel may have switched out the FP unit's
549 * state, causing the user process to take a fault
550 * when it tries to use the FP unit. Restore the
555 if (!pmath_emulate) {
557 ucode = FPE_FPU_NP_TRAP;
560 i = (*pmath_emulate)(frame);
562 if (!(frame->tf_rflags & PSL_T))
564 frame->tf_rflags &= ~PSL_T;
567 /* else ucode = emulator_only_knows() XXX */
570 case T_FPOPFLT: /* FPU operand fetch fault */
575 case T_XMMFLT: /* SIMD floating-point exception */
582 * Virtual kernel intercept - if the fault is directly related to a
583 * VM context managed by a virtual kernel then let the virtual kernel
586 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
587 vkernel_trap(lp, frame);
592 * Translate fault for emulators (e.g. Linux)
594 if (*p->p_sysent->sv_transtrap)
595 i = (*p->p_sysent->sv_transtrap)(i, type);
597 MAKEMPSAFE(have_mplock);
598 trapsignal(lp, i, ucode);
601 if (type <= MAX_TRAP_MSG) {
602 uprintf("fatal process exception: %s",
604 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
605 uprintf(", fault VA = 0x%lx", (u_long)eva);
612 KASSERT(td->td_mpcount == have_mplock, ("badmpcount trap/end from %p", (void *)frame->tf_rip));
614 userret(lp, frame, sticks);
621 KTR_LOG(kernentry_trap_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
623 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
624 ("syscall: critical section count mismatch! %d/%d",
625 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
630 kern_trap(struct trapframe *frame)
632 struct globaldata *gd = mycpu;
633 struct thread *td = gd->gd_curthread;
636 int i = 0, ucode = 0, type, code;
641 int crit_count = td->td_pri & ~TDPRI_MASK;
648 if (frame->tf_trapno == T_PAGEFLT)
649 eva = frame->tf_addr;
655 ++gd->gd_trap_nesting_level;
656 MAKEMPSAFE(have_mplock);
657 trap_fatal(frame, FALSE, eva);
658 --gd->gd_trap_nesting_level;
663 ++gd->gd_trap_nesting_level;
666 if (trap_mpsafe == 0)
667 MAKEMPSAFE(have_mplock);
670 --gd->gd_trap_nesting_level;
672 type = frame->tf_trapno;
673 code = frame->tf_err;
681 case T_PAGEFLT: /* page fault */
682 MAKEMPSAFE(have_mplock);
683 trap_pfault(frame, FALSE, eva);
688 * The kernel may be using npx for copying or other
691 panic("kernel NPX should not happen");
696 case T_PROTFLT: /* general protection fault */
697 case T_SEGNPFLT: /* segment not present fault */
699 * Invalid segment selectors and out of bounds
700 * %eip's and %esp's can be set up in user mode.
701 * This causes a fault in kernel mode when the
702 * kernel tries to return to user mode. We want
703 * to get this fault so that we can fix the
704 * problem here and not have to check all the
705 * selectors and pointers when the user changes
708 if (mycpu->gd_intr_nesting_level == 0) {
709 if (td->td_pcb->pcb_onfault) {
711 (register_t)td->td_pcb->pcb_onfault;
719 * PSL_NT can be set in user mode and isn't cleared
720 * automatically when the kernel is entered. This
721 * causes a TSS fault when the kernel attempts to
722 * `iret' because the TSS link is uninitialized. We
723 * want to get this fault so that we can fix the
724 * problem here and not every time the kernel is
727 if (frame->tf_rflags & PSL_NT) {
728 frame->tf_rflags &= ~PSL_NT;
733 case T_TRCTRAP: /* trace trap */
735 if (frame->tf_eip == (int)IDTVEC(syscall)) {
737 * We've just entered system mode via the
738 * syscall lcall. Continue single stepping
739 * silently until the syscall handler has
744 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
746 * The syscall handler has now saved the
747 * flags. Stop single stepping it.
749 frame->tf_eflags &= ~PSL_T;
755 * Ignore debug register trace traps due to
756 * accesses in the user's address space, which
757 * can happen under several conditions such as
758 * if a user sets a watchpoint on a buffer and
759 * then passes that buffer to a system call.
760 * We still want to get TRCTRAPS for addresses
761 * in kernel space because that is useful when
762 * debugging the kernel.
764 if (user_dbreg_trap()) {
766 * Reset breakpoint bits because the
769 load_dr6(rdr6() & 0xfffffff0);
774 * Fall through (TRCTRAP kernel mode, kernel address)
778 * If DDB is enabled, let it handle the debugger trap.
779 * Otherwise, debugger traps "can't happen".
782 MAKEMPSAFE(have_mplock);
783 if (kdb_trap (type, 0, frame))
788 MAKEMPSAFE(have_mplock);
789 trap_fatal(frame, FALSE, eva);
792 MAKEMPSAFE(have_mplock);
793 trap_fatal(frame, FALSE, eva);
797 * Ignore this trap generated from a spurious SIGTRAP.
799 * single stepping in / syscalls leads to spurious / SIGTRAP
802 * Haiku (c) 2007 Simon 'corecode' Schubert
808 * Translate fault for emulators (e.g. Linux)
810 if (*p->p_sysent->sv_transtrap)
811 i = (*p->p_sysent->sv_transtrap)(i, type);
813 MAKEMPSAFE(have_mplock);
814 trapsignal(lp, i, ucode);
817 if (type <= MAX_TRAP_MSG) {
818 uprintf("fatal process exception: %s",
820 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
821 uprintf(", fault VA = 0x%lx", (u_long)eva);
833 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
834 ("syscall: critical section count mismatch! %d/%d",
835 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
840 trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
843 struct vmspace *vm = NULL;
847 thread_t td = curthread;
848 struct lwp *lp = td->td_lwp;
850 va = trunc_page(eva);
851 if (usermode == FALSE) {
853 * This is a fault on kernel virtual memory.
858 * This is a fault on non-kernel virtual memory.
859 * vm is initialized above to NULL. If curproc is NULL
860 * or curproc->p_vmspace is NULL the fault is fatal.
863 vm = lp->lwp_vmspace;
871 if (frame->tf_err & PGEX_W)
872 ftype = VM_PROT_READ | VM_PROT_WRITE;
874 ftype = VM_PROT_READ;
876 if (map != &kernel_map) {
878 * Keep swapout from messing with us during this
884 * Grow the stack if necessary
886 /* grow_stack returns false only if va falls into
887 * a growable stack region and the stack growth
888 * fails. It returns true if va was not within
889 * a growable stack region, or if the stack
892 if (!grow_stack (lp->lwp_proc, va)) {
898 /* Fault in the user page: */
899 rv = vm_fault(map, va, ftype,
900 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
906 * Don't have to worry about process locking or stacks in the kernel.
908 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
911 if (rv == KERN_SUCCESS)
915 if (td->td_gd->gd_intr_nesting_level == 0 &&
916 td->td_pcb->pcb_onfault) {
917 frame->tf_rip = (register_t)td->td_pcb->pcb_onfault;
920 trap_fatal(frame, usermode, eva);
925 * NOTE: on amd64 we have a tf_addr field in the trapframe, no
926 * kludge is needed to pass the fault address to signal handlers.
928 struct proc *p = td->td_proc;
929 kprintf("seg-fault accessing address %p rip=%p pid=%d p_comm=%s\n",
930 (void *)va, (void *)frame->tf_rip, p->p_pid, p->p_comm);
931 /* Debugger("seg-fault"); */
933 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
937 trap_fatal(struct trapframe *frame, int usermode, vm_offset_t eva)
942 code = frame->tf_xflags;
943 type = frame->tf_trapno;
945 if (type <= MAX_TRAP_MSG) {
946 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
947 type, trap_msg[type],
948 (usermode ? "user" : "kernel"));
951 /* two separate prints in case of a trap on an unmapped page */
952 kprintf("mp_lock = %08x; ", mp_lock);
953 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
955 if (type == T_PAGEFLT) {
956 kprintf("fault virtual address = 0x%x\n", eva);
957 kprintf("fault code = %s %s, %s\n",
958 usermode ? "user" : "supervisor",
959 code & PGEX_W ? "write" : "read",
960 code & PGEX_P ? "protection violation" : "page not present");
962 kprintf("instruction pointer = 0x%lx:0x%lx\n",
963 frame->tf_cs & 0xffff, frame->tf_rip);
965 ss = frame->tf_ss & 0xffff;
968 ss = GSEL(GDATA_SEL, SEL_KPL);
969 rsp = (long)&frame->tf_rsp;
971 kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp);
972 kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp);
973 kprintf("processor eflags = ");
974 if (frame->tf_rflags & PSL_T)
975 kprintf("trace trap, ");
976 if (frame->tf_rflags & PSL_I)
977 kprintf("interrupt enabled, ");
978 if (frame->tf_rflags & PSL_NT)
979 kprintf("nested task, ");
980 if (frame->tf_rflags & PSL_RF)
983 if (frame->tf_eflags & PSL_VM)
986 kprintf("IOPL = %d\n", (frame->tf_rflags & PSL_IOPL) >> 12);
987 kprintf("current process = ");
989 kprintf("%lu (%s)\n",
990 (u_long)curproc->p_pid, curproc->p_comm ?
991 curproc->p_comm : "");
995 kprintf("current thread = pri %d ", curthread->td_pri);
996 if (curthread->td_pri >= TDPRI_CRIT)
1002 * we probably SHOULD have stopped the other CPUs before now!
1003 * another CPU COULD have been touching cpl at this moment...
1005 kprintf(" <- SMP: XXX");
1014 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1017 kprintf("trap number = %d\n", type);
1018 if (type <= MAX_TRAP_MSG)
1019 panic("%s", trap_msg[type]);
1021 panic("unknown/reserved trap");
1025 * Double fault handler. Called when a fault occurs while writing
1026 * a frame for a trap/exception onto the stack. This usually occurs
1027 * when the stack overflows (such is the case with infinite recursion,
1030 * XXX Note that the current PTD gets replaced by IdlePTD when the
1031 * task switch occurs. This means that the stack that was active at
1032 * the time of the double fault is not available at <kstack> unless
1033 * the machine was idle when the double fault occurred. The downside
1034 * of this is that "trace <ebp>" in ddb won't work.
1037 dblfault_handler(void)
1039 struct mdglobaldata *gd = mdcpu;
1041 kprintf("\nFatal double fault:\n");
1043 kprintf("rip = 0x%lx\n", gd->gd_common_tss.tss_rip);
1044 kprintf("rsp = 0x%lx\n", gd->gd_common_tss.tss_rsp);
1045 kprintf("rbp = 0x%lx\n", gd->gd_common_tss.tss_rbp);
1048 /* two separate prints in case of a trap on an unmapped page */
1049 kprintf("mp_lock = %08x; ", mp_lock);
1050 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
1052 panic("double fault");
1056 * Compensate for 386 brain damage (missing URKR).
1057 * This is a little simpler than the pagefault handler in trap() because
1058 * it the page tables have already been faulted in and high addresses
1059 * are thrown out early for other reasons.
1062 trapwrite(unsigned addr)
1069 va = trunc_page((vm_offset_t)addr);
1071 * XXX - MAX is END. Changed > to >= for temp. fix.
1073 if (va >= VM_MAX_USER_ADDRESS)
1076 lp = curthread->td_lwp;
1077 vm = lp->lwp_vmspace;
1079 PHOLD(lp->lwp_proc);
1081 if (!grow_stack (lp->lwp_proc, va)) {
1082 PRELE(lp->lwp_proc);
1087 * fault the data page
1089 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY);
1091 PRELE(lp->lwp_proc);
1093 if (rv != KERN_SUCCESS)
1100 * syscall2 - MP aware system call request C handler
1102 * A system call is essentially treated as a trap except that the
1103 * MP lock is not held on entry or return. We are responsible for
1104 * obtaining the MP lock if necessary and for handling ASTs
1105 * (e.g. a task switch) prior to return.
1107 * In general, only simple access and manipulation of curproc and
1108 * the current stack is allowed without having to hold MP lock.
1110 * MPSAFE - note that large sections of this routine are run without
1114 syscall2(struct trapframe *frame)
1116 struct thread *td = curthread;
1117 struct proc *p = td->td_proc;
1118 struct lwp *lp = td->td_lwp;
1120 struct sysent *callp;
1121 register_t orig_tf_rflags;
1126 int crit_count = td->td_pri & ~TDPRI_MASK;
1129 int have_mplock = 0;
1134 union sysunion args;
1135 register_t *argsdst;
1137 mycpu->gd_cnt.v_syscall++;
1139 KTR_LOG(kernentry_syscall, lp->lwp_proc->p_pid, lp->lwp_tid,
1143 KASSERT(td->td_mpcount == 0, ("badmpcount syscall2 from %p", (void *)frame->tf_rip));
1144 if (syscall_mpsafe == 0)
1145 MAKEMPSAFE(have_mplock);
1147 userenter(td); /* lazy raise our priority */
1154 sticks = (int)td->td_sticks;
1155 orig_tf_rflags = frame->tf_rflags;
1158 * Virtual kernel intercept - if a VM context managed by a virtual
1159 * kernel issues a system call the virtual kernel handles it, not us.
1160 * Restore the virtual kernel context and return from its system
1161 * call. The current frame is copied out to the virtual kernel.
1163 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1164 error = vkernel_trap(lp, frame);
1165 frame->tf_rax = error;
1167 frame->tf_rflags |= PSL_C;
1168 error = EJUSTRETURN;
1173 * Get the system call parameters and account for time
1175 lp->lwp_md.md_regs = frame;
1176 params = (caddr_t)frame->tf_rsp + sizeof(register_t);
1177 code = frame->tf_rax;
1179 if (p->p_sysent->sv_prepsyscall) {
1180 (*p->p_sysent->sv_prepsyscall)(
1181 frame, (int *)(&args.nosys.sysmsg + 1),
1184 if (code == SYS_syscall || code == SYS___syscall) {
1185 code = frame->tf_rdi;
1191 if (p->p_sysent->sv_mask)
1192 code &= p->p_sysent->sv_mask;
1194 if (code >= p->p_sysent->sv_size)
1195 callp = &p->p_sysent->sv_table[0];
1197 callp = &p->p_sysent->sv_table[code];
1199 narg = callp->sy_narg & SYF_ARGMASK;
1202 * On amd64 we get up to six arguments in registers. The rest are
1203 * on the stack. The first six members of 'struct trapframe' happen
1204 * to be the registers used to pass arguments, in exactly the right
1207 argp = &frame->tf_rdi;
1209 argsdst = (register_t *)(&args.nosys.sysmsg + 1);
1211 * JG can we overflow the space pointed to by 'argsdst'
1212 * either with 'bcopy' or with 'copyin'?
1214 bcopy(argp, argsdst, sizeof(register_t) * regcnt);
1216 * copyin is MP aware, but the tracing code is not
1218 if (narg > regcnt) {
1219 KASSERT(params != NULL, ("copyin args with no params!"));
1220 error = copyin(params, &argsdst[regcnt],
1221 (narg - regcnt) * sizeof(register_t));
1224 if (KTRPOINT(td, KTR_SYSCALL)) {
1225 MAKEMPSAFE(have_mplock);
1227 ktrsyscall(lp, code, narg,
1228 (void *)(&args.nosys.sysmsg + 1));
1236 if (KTRPOINT(td, KTR_SYSCALL)) {
1237 MAKEMPSAFE(have_mplock);
1238 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1243 * Default return value is 0 (will be copied to %rax). Double-value
1244 * returns use %rax and %rdx. %rdx is left unchanged for system
1245 * calls which return only one result.
1247 args.sysmsg_fds[0] = 0;
1248 args.sysmsg_fds[1] = frame->tf_rdx;
1251 * The syscall might manipulate the trap frame. If it does it
1252 * will probably return EJUSTRETURN.
1254 args.sysmsg_frame = frame;
1256 STOPEVENT(p, S_SCE, narg); /* MP aware */
1260 * Try to run the syscall without the MP lock if the syscall
1261 * is MP safe. We have to obtain the MP lock no matter what if
1264 if ((callp->sy_narg & SYF_MPSAFE) == 0)
1265 MAKEMPSAFE(have_mplock);
1268 error = (*callp->sy_call)(&args);
1271 kprintf("system call %d returned %d\n", code, error);
1276 * MP SAFE (we may or may not have the MP lock at this point)
1281 * Reinitialize proc pointer `p' as it may be different
1282 * if this is a child returning from fork syscall.
1285 lp = curthread->td_lwp;
1286 frame->tf_rax = args.sysmsg_fds[0];
1287 frame->tf_rdx = args.sysmsg_fds[1];
1288 frame->tf_rflags &= ~PSL_C;
1292 * Reconstruct pc, we know that 'syscall' is 2 bytes.
1293 * We have to do a full context restore so that %r10
1294 * (which was holding the value of %rcx) is restored for
1295 * the next iteration.
1297 frame->tf_rip -= frame->tf_err;
1298 frame->tf_r10 = frame->tf_rcx;
1303 panic("Unexpected EASYNC return value (for now)");
1306 if (p->p_sysent->sv_errsize) {
1307 if (error >= p->p_sysent->sv_errsize)
1308 error = -1; /* XXX */
1310 error = p->p_sysent->sv_errtbl[error];
1312 frame->tf_rax = error;
1313 frame->tf_rflags |= PSL_C;
1318 * Traced syscall. trapsignal() is not MP aware.
1320 if (orig_tf_rflags & PSL_T) {
1321 MAKEMPSAFE(have_mplock);
1322 frame->tf_rflags &= ~PSL_T;
1323 trapsignal(lp, SIGTRAP, 0);
1327 * Handle reschedule and other end-of-syscall issues
1329 userret(lp, frame, sticks);
1332 if (KTRPOINT(td, KTR_SYSRET)) {
1333 MAKEMPSAFE(have_mplock);
1334 ktrsysret(lp, code, error, args.sysmsg_result);
1339 * This works because errno is findable through the
1340 * register set. If we ever support an emulation where this
1341 * is not the case, this code will need to be revisited.
1343 STOPEVENT(p, S_SCX, code);
1348 * Release the MP lock if we had to get it
1350 KASSERT(td->td_mpcount == have_mplock,
1351 ("badmpcount syscall2/end from %p", (void *)frame->tf_rip));
1355 KTR_LOG(kernentry_syscall_ret, lp->lwp_proc->p_pid, lp->lwp_tid, error);
1357 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
1358 ("syscall: critical section count mismatch! %d/%d",
1359 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
1364 fork_return(struct lwp *lp, struct trapframe *frame)
1366 frame->tf_rax = 0; /* Child returns zero */
1367 frame->tf_rflags &= ~PSL_C; /* success */
1370 generic_lwp_return(lp, frame);
1371 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1375 * Simplified back end of syscall(), used when returning from fork()
1376 * or lwp_create() directly into user mode. MP lock is held on entry and
1377 * should be released on return. This code will return back into the fork
1378 * trampoline code which then runs doreti.
1381 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1383 struct proc *p = lp->lwp_proc;
1386 * Newly forked processes are given a kernel priority. We have to
1387 * adjust the priority to a normal user priority and fake entry
1388 * into the kernel (call userenter()) to install a passive release
1389 * function just in case userret() decides to stop the process. This
1390 * can occur when ^Z races a fork. If we do not install the passive
1391 * release function the current process designation will not be
1392 * released when the thread goes to sleep.
1394 lwkt_setpri_self(TDPRI_USER_NORM);
1395 userenter(lp->lwp_thread);
1396 userret(lp, frame, 0);
1398 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1399 ktrsysret(lp, SYS_fork, 0, 0);
1401 p->p_flag |= P_PASSIVE_ACQ;
1403 p->p_flag &= ~P_PASSIVE_ACQ;
1405 KKASSERT(lp->lwp_thread->td_mpcount == 1);
1411 * doreti has turned into this. The frame is directly on the stack. We
1412 * pull everything else we need (fpu and tls context) from the current
1415 * Note on fpu interactions: In a virtual kernel, the fpu context for
1416 * an emulated user mode process is not shared with the virtual kernel's
1417 * fpu context, so we only have to 'stack' fpu contexts within the virtual
1418 * kernel itself, and not even then since the signal() contexts that we care
1419 * about save and restore the FPU state (I think anyhow).
1421 * vmspace_ctl() returns an error only if it had problems instaling the
1422 * context we supplied or problems copying data to/from our VM space.
1425 go_user(struct intrframe *frame)
1427 struct trapframe *tf = (void *)&frame->if_rdi;
1431 * Interrupts may be disabled on entry, make sure all signals
1432 * can be received before beginning our loop.
1437 * Switch to the current simulated user process, then call
1438 * user_trap() when we break out of it (usually due to a signal).
1442 * Tell the real kernel whether it is ok to use the FP
1445 if (mdcpu->gd_npxthread == curthread) {
1446 tf->tf_xflags &= ~PGEX_FPFAULT;
1448 tf->tf_xflags |= PGEX_FPFAULT;
1452 * Run emulated user process context. This call interlocks
1453 * with new mailbox signals.
1455 * Set PGEX_U unconditionally, indicating a user frame (the
1456 * bit is normally set only by T_PAGEFLT).
1458 r = vmspace_ctl(&curproc->p_vmspace->vm_pmap, VMSPACE_CTL_RUN,
1459 tf, &curthread->td_savevext);
1460 frame->if_xflags |= PGEX_U;
1462 kprintf("GO USER %d trap %ld EVA %08lx RIP %08lx RSP %08lx XFLAGS %02lx/%02lx\n",
1463 r, tf->tf_trapno, tf->tf_addr, tf->tf_rip, tf->tf_rsp,
1464 tf->tf_xflags, frame->if_xflags);
1468 panic("vmspace_ctl failed");
1470 if (tf->tf_trapno) {
1474 if (mycpu->gd_reqflags & RQF_AST_MASK) {
1475 tf->tf_trapno = T_ASTFLT;
1483 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1484 * fault (which is then passed back to the virtual kernel) if an attempt is
1485 * made to use the FP unit.
1487 * XXX this is a fairly big hack.
1490 set_vkernel_fp(struct trapframe *frame)
1492 struct thread *td = curthread;
1494 if (frame->tf_xflags & PGEX_FPFAULT) {
1495 td->td_pcb->pcb_flags |= FP_VIRTFP;
1496 if (mdcpu->gd_npxthread == td)
1499 td->td_pcb->pcb_flags &= ~FP_VIRTFP;