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 $
42 * x86_64 Trap and System call handling
48 #include "opt_ktrace.h"
50 #include <sys/param.h>
51 #include <sys/systm.h>
53 #include <sys/pioctl.h>
54 #include <sys/kernel.h>
55 #include <sys/resourcevar.h>
56 #include <sys/signalvar.h>
57 #include <sys/signal2.h>
58 #include <sys/syscall.h>
59 #include <sys/sysctl.h>
60 #include <sys/sysent.h>
62 #include <sys/vmmeter.h>
63 #include <sys/malloc.h>
65 #include <sys/ktrace.h>
68 #include <sys/upcall.h>
69 #include <sys/vkernel.h>
70 #include <sys/sysproto.h>
71 #include <sys/sysunion.h>
72 #include <sys/vmspace.h>
75 #include <vm/vm_param.h>
78 #include <vm/vm_kern.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_page.h>
81 #include <vm/vm_extern.h>
83 #include <machine/cpu.h>
84 #include <machine/md_var.h>
85 #include <machine/pcb.h>
86 #include <machine/smp.h>
87 #include <machine/tss.h>
88 #include <machine/globaldata.h>
92 #include <sys/msgport2.h>
93 #include <sys/thread2.h>
94 #include <sys/mplock2.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 MALLOC_DEFINE(M_SYSMSG, "sysmsg", "sysmsg structure");
173 extern int max_sysmsg;
176 * Passively intercepts the thread switch function to increase the thread
177 * priority from a user priority to a kernel priority, reducing
178 * syscall and trap overhead for the case where no switch occurs.
180 * Synchronizes td_ucred with p_ucred. This is used by system calls,
181 * signal handling, faults, AST traps, and anything else that enters the
182 * kernel from userland and provides the kernel with a stable read-only
183 * copy of the process ucred.
186 userenter(struct thread *curtd, struct proc *curp)
191 curtd->td_release = lwkt_passive_release;
193 if (curtd->td_ucred != curp->p_ucred) {
194 ncred = crhold(curp->p_ucred);
195 ocred = curtd->td_ucred;
196 curtd->td_ucred = ncred;
203 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
204 * must be completed before we can return to or try to return to userland.
206 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
207 * arithmatic on the delta calculation so the absolute tick values are
208 * truncated to an integer.
211 userret(struct lwp *lp, struct trapframe *frame, int sticks)
213 struct proc *p = lp->lwp_proc;
217 * Charge system time if profiling. Note: times are in microseconds.
218 * This may do a copyout and block, so do it first even though it
219 * means some system time will be charged as user time.
221 if (p->p_flag & P_PROFIL) {
222 addupc_task(p, frame->tf_rip,
223 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
228 * If the jungle wants us dead, so be it.
230 if (lp->lwp_flag & LWP_WEXIT) {
231 lwkt_gettoken(&p->p_token);
233 lwkt_reltoken(&p->p_token); /* NOT REACHED */
237 * Block here if we are in a stopped state.
239 if (p->p_stat == SSTOP) {
247 * Post any pending upcalls
249 if (p->p_flag & P_UPCALLPEND) {
251 p->p_flag &= ~P_UPCALLPEND;
258 * Post any pending signals
260 * WARNING! postsig() can exit and not return.
262 if ((sig = CURSIG_TRACE(lp)) != 0) {
270 * block here if we are swapped out, but still process signals
271 * (such as SIGKILL). proc0 (the swapin scheduler) is already
272 * aware of our situation, we do not have to wake it up.
274 if (p->p_flag & P_SWAPPEDOUT) {
276 p->p_flag |= P_SWAPWAIT;
278 if (p->p_flag & P_SWAPWAIT)
279 tsleep(p, PCATCH, "SWOUT", 0);
280 p->p_flag &= ~P_SWAPWAIT;
286 * Make sure postsig() handled request to restore old signal mask after
287 * running signal handler.
289 KKASSERT((lp->lwp_flag & LWP_OLDMASK) == 0);
293 * Cleanup from userenter and any passive release that might have occured.
294 * We must reclaim the current-process designation before we can return
295 * to usermode. We also handle both LWKT and USER reschedule requests.
298 userexit(struct lwp *lp)
300 struct thread *td = lp->lwp_thread;
301 /* globaldata_t gd = td->td_gd; */
304 * Handle stop requests at kernel priority. Any requests queued
305 * after this loop will generate another AST.
307 while (lp->lwp_proc->p_stat == SSTOP) {
314 * Reduce our priority in preparation for a return to userland. If
315 * our passive release function was still in place, our priority was
316 * never raised and does not need to be reduced.
318 lwkt_passive_recover(td);
321 * Become the current user scheduled process if we aren't already,
322 * and deal with reschedule requests and other factors.
324 lp->lwp_proc->p_usched->acquire_curproc(lp);
325 /* WARNING: we may have migrated cpu's */
326 /* gd = td->td_gd; */
329 #if !defined(KTR_KERNENTRY)
330 #define KTR_KERNENTRY KTR_ALL
332 KTR_INFO_MASTER(kernentry);
333 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, "pid=%d, tid=%d, trapno=%d, eva=%p",
334 sizeof(int) + sizeof(int) + sizeof(int) + sizeof(vm_offset_t));
335 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "pid=%d, tid=%d",
336 sizeof(int) + sizeof(int));
337 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "pid=%d, tid=%d, call=%d",
338 sizeof(int) + sizeof(int) + sizeof(int));
339 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "pid=%d, tid=%d, err=%d",
340 sizeof(int) + sizeof(int) + sizeof(int));
341 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "pid=%d, tid=%d",
342 sizeof(int) + sizeof(int));
345 * Exception, fault, and trap interface to the kernel.
346 * This common code is called from assembly language IDT gate entry
347 * routines that prepare a suitable stack frame, and restore this
348 * frame after the exception has been processed.
350 * This function is also called from doreti in an interlock to handle ASTs.
351 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
353 * NOTE! We have to retrieve the fault address prior to obtaining the
354 * MP lock because get_mplock() may switch out. YYY cr2 really ought
355 * to be retrieved by the assembly code, not here.
357 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
358 * if an attempt is made to switch from a fast interrupt or IPI. This is
359 * necessary to properly take fatal kernel traps on SMP machines if
360 * get_mplock() has to block.
364 user_trap(struct trapframe *frame)
366 struct globaldata *gd = mycpu;
367 struct thread *td = gd->gd_curthread;
368 struct lwp *lp = td->td_lwp;
371 int i = 0, ucode = 0, type, code;
376 int crit_count = td->td_critcount;
377 lwkt_tokref_t curstop = td->td_toks_stop;
383 if (frame->tf_trapno == T_PAGEFLT)
384 eva = frame->tf_addr;
388 kprintf("USER_TRAP AT %08lx xflags %ld trapno %ld eva %08lx\n",
389 frame->tf_rip, frame->tf_xflags, frame->tf_trapno, eva);
393 * Everything coming from user mode runs through user_trap,
394 * including system calls.
396 if (frame->tf_trapno == T_FAST_SYSCALL) {
401 KTR_LOG(kernentry_trap, lp->lwp_proc->p_pid, lp->lwp_tid,
402 frame->tf_trapno, eva);
406 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
407 ++gd->gd_trap_nesting_level;
408 MAKEMPSAFE(have_mplock);
409 trap_fatal(frame, TRUE, eva);
410 --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(&mycpu->gd_reqflags, RQF_AST_OWEUPC);
447 addupc_task(p, p->p_prof.pr_addr, p->p_prof.pr_ticks);
452 * The following two traps can happen in
453 * vm86 mode, and, if so, we want to handle
456 case T_PROTFLT: /* general protection fault */
457 case T_STKFLT: /* stack fault */
459 if (frame->tf_eflags & PSL_VM) {
460 i = vm86_emulate((struct vm86frame *)frame);
468 case T_SEGNPFLT: /* segment not present fault */
469 case T_TSSFLT: /* invalid TSS fault */
470 case T_DOUBLEFLT: /* double fault */
472 ucode = code + BUS_SEGM_FAULT ;
476 case T_PAGEFLT: /* page fault */
477 MAKEMPSAFE(have_mplock);
478 i = trap_pfault(frame, TRUE, eva);
481 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
491 case T_DIVIDE: /* integer divide fault */
498 MAKEMPSAFE(have_mplock);
499 /* machine/parity/power fail/"kitchen sink" faults */
500 if (isa_nmi(code) == 0) {
503 * NMI can be hooked up to a pushbutton
507 kprintf ("NMI ... going to debugger\n");
508 kdb_trap (type, 0, frame);
512 } else if (panic_on_nmi)
513 panic("NMI indicates hardware failure");
515 #endif /* NISA > 0 */
517 case T_OFLOW: /* integer overflow fault */
522 case T_BOUND: /* bounds check fault */
529 * Virtual kernel intercept - pass the DNA exception
530 * to the (emulated) virtual kernel if it asked to handle
531 * it. This occurs when the virtual kernel is holding
532 * onto the FP context for a different emulated
533 * process then the one currently running.
535 * We must still call npxdna() since we may have
536 * saved FP state that the (emulated) virtual kernel
537 * needs to hand over to a different emulated process.
539 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
540 (td->td_pcb->pcb_flags & FP_VIRTFP)
546 * The kernel may have switched out the FP unit's
547 * state, causing the user process to take a fault
548 * when it tries to use the FP unit. Restore the
553 if (!pmath_emulate) {
555 ucode = FPE_FPU_NP_TRAP;
558 i = (*pmath_emulate)(frame);
560 if (!(frame->tf_rflags & PSL_T))
562 frame->tf_rflags &= ~PSL_T;
565 /* else ucode = emulator_only_knows() XXX */
568 case T_FPOPFLT: /* FPU operand fetch fault */
573 case T_XMMFLT: /* SIMD floating-point exception */
580 * Virtual kernel intercept - if the fault is directly related to a
581 * VM context managed by a virtual kernel then let the virtual kernel
584 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
585 vkernel_trap(lp, frame);
590 * Translate fault for emulators (e.g. Linux)
592 if (*p->p_sysent->sv_transtrap)
593 i = (*p->p_sysent->sv_transtrap)(i, type);
595 MAKEMPSAFE(have_mplock);
596 trapsignal(lp, i, ucode);
599 if (type <= MAX_TRAP_MSG) {
600 uprintf("fatal process exception: %s",
602 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
603 uprintf(", fault VA = 0x%lx", (u_long)eva);
609 userret(lp, frame, sticks);
616 KTR_LOG(kernentry_trap_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
618 KASSERT(crit_count == td->td_critcount,
619 ("trap: critical section count mismatch! %d/%d",
620 crit_count, td->td_pri));
621 KASSERT(curstop == td->td_toks_stop,
622 ("trap: extra tokens held after trap! %ld/%ld",
623 curstop - &td->td_toks_base,
624 td->td_toks_stop - &td->td_toks_base));
629 kern_trap(struct trapframe *frame)
631 struct globaldata *gd = mycpu;
632 struct thread *td = gd->gd_curthread;
635 int i = 0, ucode = 0, type, code;
640 int crit_count = td->td_critcount;
641 lwkt_tokref_t curstop = td->td_toks_stop;
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 type = frame->tf_trapno;
664 code = frame->tf_err;
672 case T_PAGEFLT: /* page fault */
673 MAKEMPSAFE(have_mplock);
674 trap_pfault(frame, FALSE, eva);
679 * The kernel may be using npx for copying or other
682 panic("kernel NPX should not happen");
687 case T_PROTFLT: /* general protection fault */
688 case T_SEGNPFLT: /* segment not present fault */
690 * Invalid segment selectors and out of bounds
691 * %eip's and %esp's can be set up in user mode.
692 * This causes a fault in kernel mode when the
693 * kernel tries to return to user mode. We want
694 * to get this fault so that we can fix the
695 * problem here and not have to check all the
696 * selectors and pointers when the user changes
699 if (mycpu->gd_intr_nesting_level == 0) {
700 if (td->td_pcb->pcb_onfault) {
702 (register_t)td->td_pcb->pcb_onfault;
710 * PSL_NT can be set in user mode and isn't cleared
711 * automatically when the kernel is entered. This
712 * causes a TSS fault when the kernel attempts to
713 * `iret' because the TSS link is uninitialized. We
714 * want to get this fault so that we can fix the
715 * problem here and not every time the kernel is
718 if (frame->tf_rflags & PSL_NT) {
719 frame->tf_rflags &= ~PSL_NT;
724 case T_TRCTRAP: /* trace trap */
726 if (frame->tf_eip == (int)IDTVEC(syscall)) {
728 * We've just entered system mode via the
729 * syscall lcall. Continue single stepping
730 * silently until the syscall handler has
735 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
737 * The syscall handler has now saved the
738 * flags. Stop single stepping it.
740 frame->tf_eflags &= ~PSL_T;
746 * Ignore debug register trace traps due to
747 * accesses in the user's address space, which
748 * can happen under several conditions such as
749 * if a user sets a watchpoint on a buffer and
750 * then passes that buffer to a system call.
751 * We still want to get TRCTRAPS for addresses
752 * in kernel space because that is useful when
753 * debugging the kernel.
755 if (user_dbreg_trap()) {
757 * Reset breakpoint bits because the
760 load_dr6(rdr6() & 0xfffffff0);
765 * Fall through (TRCTRAP kernel mode, kernel address)
769 * If DDB is enabled, let it handle the debugger trap.
770 * Otherwise, debugger traps "can't happen".
773 MAKEMPSAFE(have_mplock);
774 if (kdb_trap (type, 0, frame))
779 MAKEMPSAFE(have_mplock);
780 trap_fatal(frame, FALSE, eva);
783 MAKEMPSAFE(have_mplock);
784 trap_fatal(frame, FALSE, eva);
789 * Ignore this trap generated from a spurious SIGTRAP.
791 * single stepping in / syscalls leads to spurious / SIGTRAP
794 * Haiku (c) 2007 Simon 'corecode' Schubert
800 * Translate fault for emulators (e.g. Linux)
802 if (*p->p_sysent->sv_transtrap)
803 i = (*p->p_sysent->sv_transtrap)(i, type);
805 MAKEMPSAFE(have_mplock);
806 trapsignal(lp, i, ucode);
809 if (type <= MAX_TRAP_MSG) {
810 uprintf("fatal process exception: %s",
812 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
813 uprintf(", fault VA = 0x%lx", (u_long)eva);
825 KASSERT(crit_count == td->td_critcount,
826 ("trap: critical section count mismatch! %d/%d",
827 crit_count, td->td_pri));
828 KASSERT(curstop == td->td_toks_stop,
829 ("trap: extra tokens held after trap! %ld/%ld",
830 curstop - &td->td_toks_base,
831 td->td_toks_stop - &td->td_toks_base));
836 trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
839 struct vmspace *vm = NULL;
843 thread_t td = curthread;
844 struct lwp *lp = td->td_lwp;
846 va = trunc_page(eva);
847 if (usermode == FALSE) {
849 * This is a fault on kernel virtual memory.
854 * This is a fault on non-kernel virtual memory.
855 * vm is initialized above to NULL. If curproc is NULL
856 * or curproc->p_vmspace is NULL the fault is fatal.
859 vm = lp->lwp_vmspace;
867 if (frame->tf_err & PGEX_W)
868 ftype = VM_PROT_READ | VM_PROT_WRITE;
870 ftype = VM_PROT_READ;
872 if (map != &kernel_map) {
874 * Keep swapout from messing with us during this
880 * Grow the stack if necessary
882 /* grow_stack returns false only if va falls into
883 * a growable stack region and the stack growth
884 * fails. It returns true if va was not within
885 * a growable stack region, or if the stack
888 if (!grow_stack (lp->lwp_proc, va)) {
894 /* Fault in the user page: */
895 rv = vm_fault(map, va, ftype,
896 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
902 * Don't have to worry about process locking or stacks in the kernel.
904 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
907 if (rv == KERN_SUCCESS)
911 if (td->td_gd->gd_intr_nesting_level == 0 &&
912 td->td_pcb->pcb_onfault) {
913 frame->tf_rip = (register_t)td->td_pcb->pcb_onfault;
916 trap_fatal(frame, usermode, eva);
921 * NOTE: on x86_64 we have a tf_addr field in the trapframe, no
922 * kludge is needed to pass the fault address to signal handlers.
924 struct proc *p = td->td_proc;
925 kprintf("seg-fault accessing address %p rip=%p pid=%d p_comm=%s\n",
926 (void *)va, (void *)frame->tf_rip, p->p_pid, p->p_comm);
927 /* Debugger("seg-fault"); */
929 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
933 trap_fatal(struct trapframe *frame, int usermode, vm_offset_t eva)
938 code = frame->tf_xflags;
939 type = frame->tf_trapno;
941 if (type <= MAX_TRAP_MSG) {
942 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
943 type, trap_msg[type],
944 (usermode ? "user" : "kernel"));
947 /* two separate prints in case of a trap on an unmapped page */
948 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
950 if (type == T_PAGEFLT) {
951 kprintf("fault virtual address = %p\n", (void *)eva);
952 kprintf("fault code = %s %s, %s\n",
953 usermode ? "user" : "supervisor",
954 code & PGEX_W ? "write" : "read",
955 code & PGEX_P ? "protection violation" : "page not present");
957 kprintf("instruction pointer = 0x%lx:0x%lx\n",
958 frame->tf_cs & 0xffff, frame->tf_rip);
960 ss = frame->tf_ss & 0xffff;
963 ss = GSEL(GDATA_SEL, SEL_KPL);
964 rsp = (long)&frame->tf_rsp;
966 kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp);
967 kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp);
968 kprintf("processor eflags = ");
969 if (frame->tf_rflags & PSL_T)
970 kprintf("trace trap, ");
971 if (frame->tf_rflags & PSL_I)
972 kprintf("interrupt enabled, ");
973 if (frame->tf_rflags & PSL_NT)
974 kprintf("nested task, ");
975 if (frame->tf_rflags & PSL_RF)
978 if (frame->tf_eflags & PSL_VM)
981 kprintf("IOPL = %jd\n", (intmax_t)((frame->tf_rflags & PSL_IOPL) >> 12));
982 kprintf("current process = ");
984 kprintf("%lu (%s)\n",
985 (u_long)curproc->p_pid, curproc->p_comm ?
986 curproc->p_comm : "");
990 kprintf("current thread = pri %d ", curthread->td_pri);
991 if (curthread->td_critcount)
997 * we probably SHOULD have stopped the other CPUs before now!
998 * another CPU COULD have been touching cpl at this moment...
1000 kprintf(" <- SMP: XXX");
1009 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1012 kprintf("trap number = %d\n", type);
1013 if (type <= MAX_TRAP_MSG)
1014 panic("%s", trap_msg[type]);
1016 panic("unknown/reserved trap");
1020 * Double fault handler. Called when a fault occurs while writing
1021 * a frame for a trap/exception onto the stack. This usually occurs
1022 * when the stack overflows (such is the case with infinite recursion,
1025 * XXX Note that the current PTD gets replaced by IdlePTD when the
1026 * task switch occurs. This means that the stack that was active at
1027 * the time of the double fault is not available at <kstack> unless
1028 * the machine was idle when the double fault occurred. The downside
1029 * of this is that "trace <ebp>" in ddb won't work.
1032 dblfault_handler(void)
1035 struct mdglobaldata *gd = mdcpu;
1038 kprintf("\nFatal double fault:\n");
1040 kprintf("rip = 0x%lx\n", gd->gd_common_tss.tss_rip);
1041 kprintf("rsp = 0x%lx\n", gd->gd_common_tss.tss_rsp);
1042 kprintf("rbp = 0x%lx\n", gd->gd_common_tss.tss_rbp);
1045 /* two separate prints in case of a trap on an unmapped page */
1046 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
1048 panic("double fault");
1052 * Compensate for 386 brain damage (missing URKR).
1053 * This is a little simpler than the pagefault handler in trap() because
1054 * it the page tables have already been faulted in and high addresses
1055 * are thrown out early for other reasons.
1058 trapwrite(unsigned addr)
1065 va = trunc_page((vm_offset_t)addr);
1067 * XXX - MAX is END. Changed > to >= for temp. fix.
1069 if (va >= VM_MAX_USER_ADDRESS)
1072 lp = curthread->td_lwp;
1073 vm = lp->lwp_vmspace;
1075 PHOLD(lp->lwp_proc);
1077 if (!grow_stack (lp->lwp_proc, va)) {
1078 PRELE(lp->lwp_proc);
1083 * fault the data page
1085 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY);
1087 PRELE(lp->lwp_proc);
1089 if (rv != KERN_SUCCESS)
1096 * syscall2 - MP aware system call request C handler
1098 * A system call is essentially treated as a trap except that the
1099 * MP lock is not held on entry or return. We are responsible for
1100 * obtaining the MP lock if necessary and for handling ASTs
1101 * (e.g. a task switch) prior to return.
1103 * In general, only simple access and manipulation of curproc and
1104 * the current stack is allowed without having to hold MP lock.
1106 * MPSAFE - note that large sections of this routine are run without
1110 syscall2(struct trapframe *frame)
1112 struct thread *td = curthread;
1113 struct proc *p = td->td_proc;
1114 struct lwp *lp = td->td_lwp;
1116 struct sysent *callp;
1117 register_t orig_tf_rflags;
1122 int crit_count = td->td_critcount;
1123 lwkt_tokref_t curstop = td->td_toks_stop;
1126 int have_mplock = 0;
1131 union sysunion args;
1132 register_t *argsdst;
1134 mycpu->gd_cnt.v_syscall++;
1136 KTR_LOG(kernentry_syscall, lp->lwp_proc->p_pid, lp->lwp_tid,
1139 userenter(td, p); /* lazy raise our priority */
1146 sticks = (int)td->td_sticks;
1147 orig_tf_rflags = frame->tf_rflags;
1150 * Virtual kernel intercept - if a VM context managed by a virtual
1151 * kernel issues a system call the virtual kernel handles it, not us.
1152 * Restore the virtual kernel context and return from its system
1153 * call. The current frame is copied out to the virtual kernel.
1155 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1156 vkernel_trap(lp, frame);
1157 error = EJUSTRETURN;
1162 * Get the system call parameters and account for time
1164 lp->lwp_md.md_regs = frame;
1165 params = (caddr_t)frame->tf_rsp + sizeof(register_t);
1166 code = frame->tf_rax;
1168 if (p->p_sysent->sv_prepsyscall) {
1169 (*p->p_sysent->sv_prepsyscall)(
1170 frame, (int *)(&args.nosys.sysmsg + 1),
1173 if (code == SYS_syscall || code == SYS___syscall) {
1174 code = frame->tf_rdi;
1180 if (p->p_sysent->sv_mask)
1181 code &= p->p_sysent->sv_mask;
1183 if (code >= p->p_sysent->sv_size)
1184 callp = &p->p_sysent->sv_table[0];
1186 callp = &p->p_sysent->sv_table[code];
1188 narg = callp->sy_narg & SYF_ARGMASK;
1191 * On x86_64 we get up to six arguments in registers. The rest are
1192 * on the stack. The first six members of 'struct trapframe' happen
1193 * to be the registers used to pass arguments, in exactly the right
1196 argp = &frame->tf_rdi;
1198 argsdst = (register_t *)(&args.nosys.sysmsg + 1);
1200 * JG can we overflow the space pointed to by 'argsdst'
1201 * either with 'bcopy' or with 'copyin'?
1203 bcopy(argp, argsdst, sizeof(register_t) * regcnt);
1205 * copyin is MP aware, but the tracing code is not
1207 if (narg > regcnt) {
1208 KASSERT(params != NULL, ("copyin args with no params!"));
1209 error = copyin(params, &argsdst[regcnt],
1210 (narg - regcnt) * sizeof(register_t));
1213 if (KTRPOINT(td, KTR_SYSCALL)) {
1214 MAKEMPSAFE(have_mplock);
1216 ktrsyscall(lp, code, narg,
1217 (void *)(&args.nosys.sysmsg + 1));
1225 if (KTRPOINT(td, KTR_SYSCALL)) {
1226 MAKEMPSAFE(have_mplock);
1227 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1232 * Default return value is 0 (will be copied to %rax). Double-value
1233 * returns use %rax and %rdx. %rdx is left unchanged for system
1234 * calls which return only one result.
1236 args.sysmsg_fds[0] = 0;
1237 args.sysmsg_fds[1] = frame->tf_rdx;
1240 * The syscall might manipulate the trap frame. If it does it
1241 * will probably return EJUSTRETURN.
1243 args.sysmsg_frame = frame;
1245 STOPEVENT(p, S_SCE, narg); /* MP aware */
1248 * NOTE: All system calls run MPSAFE now. The system call itself
1249 * is responsible for getting the MP lock.
1251 error = (*callp->sy_call)(&args);
1254 kprintf("system call %d returned %d\n", code, error);
1259 * MP SAFE (we may or may not have the MP lock at this point)
1264 * Reinitialize proc pointer `p' as it may be different
1265 * if this is a child returning from fork syscall.
1268 lp = curthread->td_lwp;
1269 frame->tf_rax = args.sysmsg_fds[0];
1270 frame->tf_rdx = args.sysmsg_fds[1];
1271 frame->tf_rflags &= ~PSL_C;
1275 * Reconstruct pc, we know that 'syscall' is 2 bytes.
1276 * We have to do a full context restore so that %r10
1277 * (which was holding the value of %rcx) is restored for
1278 * the next iteration.
1280 frame->tf_rip -= frame->tf_err;
1281 frame->tf_r10 = frame->tf_rcx;
1286 panic("Unexpected EASYNC return value (for now)");
1289 if (p->p_sysent->sv_errsize) {
1290 if (error >= p->p_sysent->sv_errsize)
1291 error = -1; /* XXX */
1293 error = p->p_sysent->sv_errtbl[error];
1295 frame->tf_rax = error;
1296 frame->tf_rflags |= PSL_C;
1301 * Traced syscall. trapsignal() is not MP aware.
1303 if (orig_tf_rflags & PSL_T) {
1304 MAKEMPSAFE(have_mplock);
1305 frame->tf_rflags &= ~PSL_T;
1306 trapsignal(lp, SIGTRAP, 0);
1310 * Handle reschedule and other end-of-syscall issues
1312 userret(lp, frame, sticks);
1315 if (KTRPOINT(td, KTR_SYSRET)) {
1316 MAKEMPSAFE(have_mplock);
1317 ktrsysret(lp, code, error, args.sysmsg_result);
1322 * This works because errno is findable through the
1323 * register set. If we ever support an emulation where this
1324 * is not the case, this code will need to be revisited.
1326 STOPEVENT(p, S_SCX, code);
1331 * Release the MP lock if we had to get it
1336 KTR_LOG(kernentry_syscall_ret, lp->lwp_proc->p_pid, lp->lwp_tid, error);
1338 KASSERT(&td->td_toks_base == td->td_toks_stop,
1339 ("syscall: critical section count mismatch! %d/%d",
1340 crit_count, td->td_pri));
1341 KASSERT(curstop == td->td_toks_stop,
1342 ("syscall: extra tokens held after trap! %ld",
1343 td->td_toks_stop - &td->td_toks_base));
1348 * NOTE: mplock not held at any point
1351 fork_return(struct lwp *lp, struct trapframe *frame)
1353 frame->tf_rax = 0; /* Child returns zero */
1354 frame->tf_rflags &= ~PSL_C; /* success */
1357 generic_lwp_return(lp, frame);
1358 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1362 * Simplified back end of syscall(), used when returning from fork()
1363 * directly into user mode.
1365 * This code will return back into the fork trampoline code which then
1368 * NOTE: The mplock is not held at any point.
1371 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1373 struct proc *p = lp->lwp_proc;
1376 * Newly forked processes are given a kernel priority. We have to
1377 * adjust the priority to a normal user priority and fake entry
1378 * into the kernel (call userenter()) to install a passive release
1379 * function just in case userret() decides to stop the process. This
1380 * can occur when ^Z races a fork. If we do not install the passive
1381 * release function the current process designation will not be
1382 * released when the thread goes to sleep.
1384 lwkt_setpri_self(TDPRI_USER_NORM);
1385 userenter(lp->lwp_thread, p);
1386 userret(lp, frame, 0);
1388 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1389 ktrsysret(lp, SYS_fork, 0, 0);
1391 p->p_flag |= P_PASSIVE_ACQ;
1393 p->p_flag &= ~P_PASSIVE_ACQ;
1397 * doreti has turned into this. The frame is directly on the stack. We
1398 * pull everything else we need (fpu and tls context) from the current
1401 * Note on fpu interactions: In a virtual kernel, the fpu context for
1402 * an emulated user mode process is not shared with the virtual kernel's
1403 * fpu context, so we only have to 'stack' fpu contexts within the virtual
1404 * kernel itself, and not even then since the signal() contexts that we care
1405 * about save and restore the FPU state (I think anyhow).
1407 * vmspace_ctl() returns an error only if it had problems instaling the
1408 * context we supplied or problems copying data to/from our VM space.
1411 go_user(struct intrframe *frame)
1413 struct trapframe *tf = (void *)&frame->if_rdi;
1417 * Interrupts may be disabled on entry, make sure all signals
1418 * can be received before beginning our loop.
1423 * Switch to the current simulated user process, then call
1424 * user_trap() when we break out of it (usually due to a signal).
1428 * Tell the real kernel whether it is ok to use the FP
1431 if (mdcpu->gd_npxthread == curthread) {
1432 tf->tf_xflags &= ~PGEX_FPFAULT;
1434 tf->tf_xflags |= PGEX_FPFAULT;
1438 * Run emulated user process context. This call interlocks
1439 * with new mailbox signals.
1441 * Set PGEX_U unconditionally, indicating a user frame (the
1442 * bit is normally set only by T_PAGEFLT).
1444 r = vmspace_ctl(&curproc->p_vmspace->vm_pmap, VMSPACE_CTL_RUN,
1445 tf, &curthread->td_savevext);
1446 frame->if_xflags |= PGEX_U;
1448 kprintf("GO USER %d trap %ld EVA %08lx RIP %08lx RSP %08lx XFLAGS %02lx/%02lx\n",
1449 r, tf->tf_trapno, tf->tf_addr, tf->tf_rip, tf->tf_rsp,
1450 tf->tf_xflags, frame->if_xflags);
1454 panic("vmspace_ctl failed error %d", errno);
1456 if (tf->tf_trapno) {
1460 if (mycpu->gd_reqflags & RQF_AST_MASK) {
1461 tf->tf_trapno = T_ASTFLT;
1469 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1470 * fault (which is then passed back to the virtual kernel) if an attempt is
1471 * made to use the FP unit.
1473 * XXX this is a fairly big hack.
1476 set_vkernel_fp(struct trapframe *frame)
1478 struct thread *td = curthread;
1480 if (frame->tf_xflags & PGEX_FPFAULT) {
1481 td->td_pcb->pcb_flags |= FP_VIRTFP;
1482 if (mdcpu->gd_npxthread == td)
1485 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1490 * Called from vkernel_trap() to fixup the vkernel's syscall
1491 * frame for vmspace_ctl() return.
1494 cpu_vkernel_trap(struct trapframe *frame, int error)
1496 frame->tf_rax = error;
1498 frame->tf_rflags |= PSL_C;
1500 frame->tf_rflags &= ~PSL_C;