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
177 * the thread 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_flags & P_PROFIL) {
222 addupc_task(p, frame->tf_rip,
223 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
228 * Specific on-return-to-usermode checks (LWP_MP_WEXIT,
229 * LWP_MP_VNLRU, etc).
231 if (lp->lwp_mpflags & LWP_MP_URETMASK)
235 * Block here if we are in a stopped state.
237 if (p->p_stat == SSTOP) {
238 lwkt_gettoken(&p->p_token);
240 lwkt_reltoken(&p->p_token);
245 * Post any pending upcalls. If running a virtual kernel be sure
246 * to restore the virtual kernel's vmspace before posting the upcall.
248 if (p->p_flags & (P_SIGVTALRM | P_SIGPROF | P_UPCALLPEND)) {
249 lwkt_gettoken(&p->p_token);
250 if (p->p_flags & P_SIGVTALRM) {
251 p->p_flags &= ~P_SIGVTALRM;
252 ksignal(p, SIGVTALRM);
254 if (p->p_flags & P_SIGPROF) {
255 p->p_flags &= ~P_SIGPROF;
258 if (p->p_flags & P_UPCALLPEND) {
259 p->p_flags &= ~P_UPCALLPEND;
262 lwkt_reltoken(&p->p_token);
267 * Post any pending signals
269 * WARNING! postsig() can exit and not return.
271 if ((sig = CURSIG_TRACE(lp)) != 0) {
272 lwkt_gettoken(&p->p_token);
274 lwkt_reltoken(&p->p_token);
279 * block here if we are swapped out, but still process signals
280 * (such as SIGKILL). proc0 (the swapin scheduler) is already
281 * aware of our situation, we do not have to wake it up.
283 if (p->p_flags & P_SWAPPEDOUT) {
284 lwkt_gettoken(&p->p_token);
286 p->p_flags |= P_SWAPWAIT;
288 if (p->p_flags & P_SWAPWAIT)
289 tsleep(p, PCATCH, "SWOUT", 0);
290 p->p_flags &= ~P_SWAPWAIT;
292 lwkt_reltoken(&p->p_token);
297 * In a multi-threaded program it is possible for a thread to change
298 * signal state during a system call which temporarily changes the
299 * signal mask. In this case postsig() might not be run and we
300 * have to restore the mask ourselves.
302 if (lp->lwp_flags & LWP_OLDMASK) {
303 lp->lwp_flags &= ~LWP_OLDMASK;
304 lp->lwp_sigmask = lp->lwp_oldsigmask;
310 * Cleanup from userenter and any passive release that might have occured.
311 * We must reclaim the current-process designation before we can return
312 * to usermode. We also handle both LWKT and USER reschedule requests.
315 userexit(struct lwp *lp)
317 struct thread *td = lp->lwp_thread;
318 /* globaldata_t gd = td->td_gd; */
321 * Handle stop requests at kernel priority. Any requests queued
322 * after this loop will generate another AST.
324 while (lp->lwp_proc->p_stat == SSTOP) {
325 lwkt_gettoken(&lp->lwp_proc->p_token);
327 lwkt_reltoken(&lp->lwp_proc->p_token);
331 * Reduce our priority in preparation for a return to userland. If
332 * our passive release function was still in place, our priority was
333 * never raised and does not need to be reduced.
335 lwkt_passive_recover(td);
338 * Become the current user scheduled process if we aren't already,
339 * and deal with reschedule requests and other factors.
341 lp->lwp_proc->p_usched->acquire_curproc(lp);
342 /* WARNING: we may have migrated cpu's */
343 /* gd = td->td_gd; */
346 #if !defined(KTR_KERNENTRY)
347 #define KTR_KERNENTRY KTR_ALL
349 KTR_INFO_MASTER(kernentry);
350 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0,
351 "TRAP(pid %hd, tid %hd, trapno %ld, eva %lu)",
352 pid_t pid, lwpid_t tid, register_t trapno, vm_offset_t eva);
353 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "TRAP_RET(pid %hd, tid %hd)",
354 pid_t pid, lwpid_t tid);
355 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "SYSC(pid %hd, tid %hd, nr %ld)",
356 pid_t pid, lwpid_t tid, register_t trapno);
357 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "SYSRET(pid %hd, tid %hd, err %d)",
358 pid_t pid, lwpid_t tid, int err);
359 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "FORKRET(pid %hd, tid %hd)",
360 pid_t pid, lwpid_t tid);
363 * Exception, fault, and trap interface to the kernel.
364 * This common code is called from assembly language IDT gate entry
365 * routines that prepare a suitable stack frame, and restore this
366 * frame after the exception has been processed.
368 * This function is also called from doreti in an interlock to handle ASTs.
369 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
371 * NOTE! We have to retrieve the fault address prior to obtaining the
372 * MP lock because get_mplock() may switch out. YYY cr2 really ought
373 * to be retrieved by the assembly code, not here.
375 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
376 * if an attempt is made to switch from a fast interrupt or IPI. This is
377 * necessary to properly take fatal kernel traps on SMP machines if
378 * get_mplock() has to block.
382 user_trap(struct trapframe *frame)
384 struct globaldata *gd = mycpu;
385 struct thread *td = gd->gd_curthread;
386 struct lwp *lp = td->td_lwp;
389 int i = 0, ucode = 0, type, code;
394 int crit_count = td->td_critcount;
395 lwkt_tokref_t curstop = td->td_toks_stop;
401 if (frame->tf_trapno == T_PAGEFLT)
402 eva = frame->tf_addr;
406 kprintf("USER_TRAP AT %08lx xflags %ld trapno %ld eva %08lx\n",
407 frame->tf_rip, frame->tf_xflags, frame->tf_trapno, eva);
411 * Everything coming from user mode runs through user_trap,
412 * including system calls.
414 if (frame->tf_trapno == T_FAST_SYSCALL) {
419 KTR_LOG(kernentry_trap, lp->lwp_proc->p_pid, lp->lwp_tid,
420 frame->tf_trapno, eva);
424 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
425 ++gd->gd_trap_nesting_level;
426 MAKEMPSAFE(have_mplock);
427 trap_fatal(frame, TRUE, eva);
428 --gd->gd_trap_nesting_level;
433 type = frame->tf_trapno;
434 code = frame->tf_err;
438 sticks = (int)td->td_sticks;
439 lp->lwp_md.md_regs = frame;
442 case T_PRIVINFLT: /* privileged instruction fault */
447 case T_BPTFLT: /* bpt instruction fault */
448 case T_TRCTRAP: /* trace trap */
449 frame->tf_rflags &= ~PSL_T;
451 ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
454 case T_ARITHTRAP: /* arithmetic trap */
459 case T_ASTFLT: /* Allow process switch */
460 mycpu->gd_cnt.v_soft++;
461 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
462 atomic_clear_int(&mycpu->gd_reqflags, RQF_AST_OWEUPC);
463 addupc_task(p, p->p_prof.pr_addr, p->p_prof.pr_ticks);
468 * The following two traps can happen in
469 * vm86 mode, and, if so, we want to handle
472 case T_PROTFLT: /* general protection fault */
473 case T_STKFLT: /* stack fault */
475 if (frame->tf_eflags & PSL_VM) {
476 i = vm86_emulate((struct vm86frame *)frame);
484 case T_SEGNPFLT: /* segment not present fault */
485 case T_TSSFLT: /* invalid TSS fault */
486 case T_DOUBLEFLT: /* double fault */
489 ucode = code + BUS_SEGM_FAULT ;
492 case T_PAGEFLT: /* page fault */
493 MAKEMPSAFE(have_mplock);
494 i = trap_pfault(frame, TRUE, eva);
495 if (i == -1 || i == 0)
507 case T_DIVIDE: /* integer divide fault */
514 MAKEMPSAFE(have_mplock);
515 /* machine/parity/power fail/"kitchen sink" faults */
516 if (isa_nmi(code) == 0) {
519 * NMI can be hooked up to a pushbutton
523 kprintf ("NMI ... going to debugger\n");
524 kdb_trap(type, 0, frame);
528 } else if (panic_on_nmi)
529 panic("NMI indicates hardware failure");
531 #endif /* NISA > 0 */
533 case T_OFLOW: /* integer overflow fault */
538 case T_BOUND: /* bounds check fault */
545 * Virtual kernel intercept - pass the DNA exception
546 * to the (emulated) virtual kernel if it asked to handle
547 * it. This occurs when the virtual kernel is holding
548 * onto the FP context for a different emulated
549 * process then the one currently running.
551 * We must still call npxdna() since we may have
552 * saved FP state that the (emulated) virtual kernel
553 * needs to hand over to a different emulated process.
555 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
556 (td->td_pcb->pcb_flags & FP_VIRTFP)
563 * The kernel may have switched out the FP unit's
564 * state, causing the user process to take a fault
565 * when it tries to use the FP unit. Restore the
570 if (!pmath_emulate) {
572 ucode = FPE_FPU_NP_TRAP;
575 i = (*pmath_emulate)(frame);
577 if (!(frame->tf_rflags & PSL_T))
579 frame->tf_rflags &= ~PSL_T;
582 /* else ucode = emulator_only_knows() XXX */
585 case T_FPOPFLT: /* FPU operand fetch fault */
590 case T_XMMFLT: /* SIMD floating-point exception */
597 * Virtual kernel intercept - if the fault is directly related to a
598 * VM context managed by a virtual kernel then let the virtual kernel
601 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
602 vkernel_trap(lp, frame);
607 * Translate fault for emulators (e.g. Linux)
609 if (*p->p_sysent->sv_transtrap)
610 i = (*p->p_sysent->sv_transtrap)(i, type);
612 MAKEMPSAFE(have_mplock);
613 trapsignal(lp, i, ucode);
616 if (type <= MAX_TRAP_MSG) {
617 uprintf("fatal process exception: %s",
619 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
620 uprintf(", fault VA = 0x%lx", (u_long)eva);
626 userret(lp, frame, sticks);
633 KTR_LOG(kernentry_trap_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
635 KASSERT(crit_count == td->td_critcount,
636 ("trap: critical section count mismatch! %d/%d",
637 crit_count, td->td_pri));
638 KASSERT(curstop == td->td_toks_stop,
639 ("trap: extra tokens held after trap! %ld/%ld",
640 curstop - &td->td_toks_base,
641 td->td_toks_stop - &td->td_toks_base));
646 kern_trap(struct trapframe *frame)
648 struct globaldata *gd = mycpu;
649 struct thread *td = gd->gd_curthread;
652 int i = 0, ucode = 0, type, code;
657 int crit_count = td->td_critcount;
658 lwkt_tokref_t curstop = td->td_toks_stop;
665 if (frame->tf_trapno == T_PAGEFLT)
666 eva = frame->tf_addr;
672 ++gd->gd_trap_nesting_level;
673 MAKEMPSAFE(have_mplock);
674 trap_fatal(frame, FALSE, eva);
675 --gd->gd_trap_nesting_level;
680 type = frame->tf_trapno;
681 code = frame->tf_err;
689 case T_PAGEFLT: /* page fault */
690 MAKEMPSAFE(have_mplock);
691 trap_pfault(frame, FALSE, eva);
696 * The kernel may be using npx for copying or other
699 panic("kernel NPX should not happen");
704 case T_PROTFLT: /* general protection fault */
705 case T_SEGNPFLT: /* segment not present fault */
707 * Invalid segment selectors and out of bounds
708 * %eip's and %esp's can be set up in user mode.
709 * This causes a fault in kernel mode when the
710 * kernel tries to return to user mode. We want
711 * to get this fault so that we can fix the
712 * problem here and not have to check all the
713 * selectors and pointers when the user changes
716 if (mycpu->gd_intr_nesting_level == 0) {
717 if (td->td_pcb->pcb_onfault) {
719 (register_t)td->td_pcb->pcb_onfault;
727 * PSL_NT can be set in user mode and isn't cleared
728 * automatically when the kernel is entered. This
729 * causes a TSS fault when the kernel attempts to
730 * `iret' because the TSS link is uninitialized. We
731 * want to get this fault so that we can fix the
732 * problem here and not every time the kernel is
735 if (frame->tf_rflags & PSL_NT) {
736 frame->tf_rflags &= ~PSL_NT;
741 case T_TRCTRAP: /* trace trap */
743 if (frame->tf_eip == (int)IDTVEC(syscall)) {
745 * We've just entered system mode via the
746 * syscall lcall. Continue single stepping
747 * silently until the syscall handler has
752 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
754 * The syscall handler has now saved the
755 * flags. Stop single stepping it.
757 frame->tf_eflags &= ~PSL_T;
763 * Ignore debug register trace traps due to
764 * accesses in the user's address space, which
765 * can happen under several conditions such as
766 * if a user sets a watchpoint on a buffer and
767 * then passes that buffer to a system call.
768 * We still want to get TRCTRAPS for addresses
769 * in kernel space because that is useful when
770 * debugging the kernel.
772 if (user_dbreg_trap()) {
774 * Reset breakpoint bits because the
777 load_dr6(rdr6() & 0xfffffff0);
782 * Fall through (TRCTRAP kernel mode, kernel address)
786 * If DDB is enabled, let it handle the debugger trap.
787 * Otherwise, debugger traps "can't happen".
790 MAKEMPSAFE(have_mplock);
791 if (kdb_trap (type, 0, frame))
796 MAKEMPSAFE(have_mplock);
797 trap_fatal(frame, FALSE, eva);
800 MAKEMPSAFE(have_mplock);
801 trap_fatal(frame, FALSE, eva);
806 * Ignore this trap generated from a spurious SIGTRAP.
808 * single stepping in / syscalls leads to spurious / SIGTRAP
811 * Haiku (c) 2007 Simon 'corecode' Schubert
817 * Translate fault for emulators (e.g. Linux)
819 if (*p->p_sysent->sv_transtrap)
820 i = (*p->p_sysent->sv_transtrap)(i, type);
822 MAKEMPSAFE(have_mplock);
823 trapsignal(lp, i, ucode);
826 if (type <= MAX_TRAP_MSG) {
827 uprintf("fatal process exception: %s",
829 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
830 uprintf(", fault VA = 0x%lx", (u_long)eva);
842 KASSERT(crit_count == td->td_critcount,
843 ("trap: critical section count mismatch! %d/%d",
844 crit_count, td->td_pri));
845 KASSERT(curstop == td->td_toks_stop,
846 ("trap: extra tokens held after trap! %ld/%ld",
847 curstop - &td->td_toks_base,
848 td->td_toks_stop - &td->td_toks_base));
853 trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
856 struct vmspace *vm = NULL;
860 thread_t td = curthread;
861 struct lwp *lp = td->td_lwp;
864 va = trunc_page(eva);
865 if (usermode == FALSE) {
867 * This is a fault on kernel virtual memory.
872 * This is a fault on non-kernel virtual memory.
873 * vm is initialized above to NULL. If curproc is NULL
874 * or curproc->p_vmspace is NULL the fault is fatal.
877 vm = lp->lwp_vmspace;
885 if (frame->tf_err & PGEX_W)
886 ftype = VM_PROT_READ | VM_PROT_WRITE;
888 ftype = VM_PROT_READ;
890 if (map != &kernel_map) {
892 * Keep swapout from messing with us during this
898 * Grow the stack if necessary
900 /* grow_stack returns false only if va falls into
901 * a growable stack region and the stack growth
902 * fails. It returns true if va was not within
903 * a growable stack region, or if the stack
906 if (!grow_stack (lp->lwp_proc, va)) {
914 fault_flags |= VM_FAULT_BURST;
915 if (ftype & VM_PROT_WRITE)
916 fault_flags |= VM_FAULT_DIRTY;
918 fault_flags |= VM_FAULT_NORMAL;
919 rv = vm_fault(map, va, ftype, fault_flags);
924 * Don't have to worry about process locking or stacks in the kernel.
926 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
929 if (rv == KERN_SUCCESS)
933 if (td->td_gd->gd_intr_nesting_level == 0 &&
934 td->td_pcb->pcb_onfault) {
935 frame->tf_rip = (register_t)td->td_pcb->pcb_onfault;
938 trap_fatal(frame, usermode, eva);
943 * NOTE: on x86_64 we have a tf_addr field in the trapframe, no
944 * kludge is needed to pass the fault address to signal handlers.
946 struct proc *p = td->td_proc;
947 kprintf("seg-fault accessing address %p rip=%p pid=%d p_comm=%s\n",
948 (void *)va, (void *)frame->tf_rip, p->p_pid, p->p_comm);
949 /* Debugger("seg-fault"); */
951 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
955 trap_fatal(struct trapframe *frame, int usermode, vm_offset_t eva)
960 code = frame->tf_xflags;
961 type = frame->tf_trapno;
963 if (type <= MAX_TRAP_MSG) {
964 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
965 type, trap_msg[type],
966 (usermode ? "user" : "kernel"));
969 /* two separate prints in case of a trap on an unmapped page */
970 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
972 if (type == T_PAGEFLT) {
973 kprintf("fault virtual address = %p\n", (void *)eva);
974 kprintf("fault code = %s %s, %s\n",
975 usermode ? "user" : "supervisor",
976 code & PGEX_W ? "write" : "read",
977 code & PGEX_P ? "protection violation" : "page not present");
979 kprintf("instruction pointer = 0x%lx:0x%lx\n",
980 frame->tf_cs & 0xffff, frame->tf_rip);
982 ss = frame->tf_ss & 0xffff;
985 ss = GSEL(GDATA_SEL, SEL_KPL);
986 rsp = (long)&frame->tf_rsp;
988 kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp);
989 kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp);
990 kprintf("processor eflags = ");
991 if (frame->tf_rflags & PSL_T)
992 kprintf("trace trap, ");
993 if (frame->tf_rflags & PSL_I)
994 kprintf("interrupt enabled, ");
995 if (frame->tf_rflags & PSL_NT)
996 kprintf("nested task, ");
997 if (frame->tf_rflags & PSL_RF)
1000 if (frame->tf_eflags & PSL_VM)
1003 kprintf("IOPL = %jd\n", (intmax_t)((frame->tf_rflags & PSL_IOPL) >> 12));
1004 kprintf("current process = ");
1006 kprintf("%lu (%s)\n",
1007 (u_long)curproc->p_pid, curproc->p_comm ?
1008 curproc->p_comm : "");
1012 kprintf("current thread = pri %d ", curthread->td_pri);
1013 if (curthread->td_critcount)
1019 * we probably SHOULD have stopped the other CPUs before now!
1020 * another CPU COULD have been touching cpl at this moment...
1022 kprintf(" <- SMP: XXX");
1031 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1034 kprintf("trap number = %d\n", type);
1035 if (type <= MAX_TRAP_MSG)
1036 panic("%s", trap_msg[type]);
1038 panic("unknown/reserved trap");
1042 * Double fault handler. Called when a fault occurs while writing
1043 * a frame for a trap/exception onto the stack. This usually occurs
1044 * when the stack overflows (such is the case with infinite recursion,
1047 * XXX Note that the current PTD gets replaced by IdlePTD when the
1048 * task switch occurs. This means that the stack that was active at
1049 * the time of the double fault is not available at <kstack> unless
1050 * the machine was idle when the double fault occurred. The downside
1051 * of this is that "trace <ebp>" in ddb won't work.
1054 dblfault_handler(void)
1057 struct mdglobaldata *gd = mdcpu;
1060 kprintf("\nFatal double fault:\n");
1062 kprintf("rip = 0x%lx\n", gd->gd_common_tss.tss_rip);
1063 kprintf("rsp = 0x%lx\n", gd->gd_common_tss.tss_rsp);
1064 kprintf("rbp = 0x%lx\n", gd->gd_common_tss.tss_rbp);
1067 /* two separate prints in case of a trap on an unmapped page */
1068 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
1070 panic("double fault");
1074 * Compensate for 386 brain damage (missing URKR).
1075 * This is a little simpler than the pagefault handler in trap() because
1076 * it the page tables have already been faulted in and high addresses
1077 * are thrown out early for other reasons.
1080 trapwrite(unsigned addr)
1087 va = trunc_page((vm_offset_t)addr);
1089 * XXX - MAX is END. Changed > to >= for temp. fix.
1091 if (va >= VM_MAX_USER_ADDRESS)
1094 lp = curthread->td_lwp;
1095 vm = lp->lwp_vmspace;
1097 PHOLD(lp->lwp_proc);
1099 if (!grow_stack (lp->lwp_proc, va)) {
1100 PRELE(lp->lwp_proc);
1105 * fault the data page
1107 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY);
1109 PRELE(lp->lwp_proc);
1111 if (rv != KERN_SUCCESS)
1118 * syscall2 - MP aware system call request C handler
1120 * A system call is essentially treated as a trap except that the
1121 * MP lock is not held on entry or return. We are responsible for
1122 * obtaining the MP lock if necessary and for handling ASTs
1123 * (e.g. a task switch) prior to return.
1125 * In general, only simple access and manipulation of curproc and
1126 * the current stack is allowed without having to hold MP lock.
1128 * MPSAFE - note that large sections of this routine are run without
1132 syscall2(struct trapframe *frame)
1134 struct thread *td = curthread;
1135 struct proc *p = td->td_proc;
1136 struct lwp *lp = td->td_lwp;
1138 struct sysent *callp;
1139 register_t orig_tf_rflags;
1144 int crit_count = td->td_critcount;
1145 lwkt_tokref_t curstop = td->td_toks_stop;
1148 int have_mplock = 0;
1153 union sysunion args;
1154 register_t *argsdst;
1156 mycpu->gd_cnt.v_syscall++;
1158 KTR_LOG(kernentry_syscall, lp->lwp_proc->p_pid, lp->lwp_tid,
1161 userenter(td, p); /* lazy raise our priority */
1168 sticks = (int)td->td_sticks;
1169 orig_tf_rflags = frame->tf_rflags;
1172 * Virtual kernel intercept - if a VM context managed by a virtual
1173 * kernel issues a system call the virtual kernel handles it, not us.
1174 * Restore the virtual kernel context and return from its system
1175 * call. The current frame is copied out to the virtual kernel.
1177 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1178 vkernel_trap(lp, frame);
1179 error = EJUSTRETURN;
1184 * Get the system call parameters and account for time
1186 lp->lwp_md.md_regs = frame;
1187 params = (caddr_t)frame->tf_rsp + sizeof(register_t);
1188 code = frame->tf_rax;
1190 if (p->p_sysent->sv_prepsyscall) {
1191 (*p->p_sysent->sv_prepsyscall)(
1192 frame, (int *)(&args.nosys.sysmsg + 1),
1195 if (code == SYS_syscall || code == SYS___syscall) {
1196 code = frame->tf_rdi;
1202 if (p->p_sysent->sv_mask)
1203 code &= p->p_sysent->sv_mask;
1205 if (code >= p->p_sysent->sv_size)
1206 callp = &p->p_sysent->sv_table[0];
1208 callp = &p->p_sysent->sv_table[code];
1210 narg = callp->sy_narg & SYF_ARGMASK;
1213 * On x86_64 we get up to six arguments in registers. The rest are
1214 * on the stack. The first six members of 'struct trapframe' happen
1215 * to be the registers used to pass arguments, in exactly the right
1218 argp = &frame->tf_rdi;
1220 argsdst = (register_t *)(&args.nosys.sysmsg + 1);
1222 * JG can we overflow the space pointed to by 'argsdst'
1223 * either with 'bcopy' or with 'copyin'?
1225 bcopy(argp, argsdst, sizeof(register_t) * regcnt);
1227 * copyin is MP aware, but the tracing code is not
1229 if (narg > regcnt) {
1230 KASSERT(params != NULL, ("copyin args with no params!"));
1231 error = copyin(params, &argsdst[regcnt],
1232 (narg - regcnt) * sizeof(register_t));
1235 if (KTRPOINT(td, KTR_SYSCALL)) {
1236 MAKEMPSAFE(have_mplock);
1238 ktrsyscall(lp, code, narg,
1239 (void *)(&args.nosys.sysmsg + 1));
1247 if (KTRPOINT(td, KTR_SYSCALL)) {
1248 MAKEMPSAFE(have_mplock);
1249 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1254 * Default return value is 0 (will be copied to %rax). Double-value
1255 * returns use %rax and %rdx. %rdx is left unchanged for system
1256 * calls which return only one result.
1258 args.sysmsg_fds[0] = 0;
1259 args.sysmsg_fds[1] = frame->tf_rdx;
1262 * The syscall might manipulate the trap frame. If it does it
1263 * will probably return EJUSTRETURN.
1265 args.sysmsg_frame = frame;
1267 STOPEVENT(p, S_SCE, narg); /* MP aware */
1270 * NOTE: All system calls run MPSAFE now. The system call itself
1271 * is responsible for getting the MP lock.
1273 error = (*callp->sy_call)(&args);
1276 kprintf("system call %d returned %d\n", code, error);
1281 * MP SAFE (we may or may not have the MP lock at this point)
1286 * Reinitialize proc pointer `p' as it may be different
1287 * if this is a child returning from fork syscall.
1290 lp = curthread->td_lwp;
1291 frame->tf_rax = args.sysmsg_fds[0];
1292 frame->tf_rdx = args.sysmsg_fds[1];
1293 frame->tf_rflags &= ~PSL_C;
1297 * Reconstruct pc, we know that 'syscall' is 2 bytes.
1298 * We have to do a full context restore so that %r10
1299 * (which was holding the value of %rcx) is restored for
1300 * the next iteration.
1302 frame->tf_rip -= frame->tf_err;
1303 frame->tf_r10 = frame->tf_rcx;
1308 panic("Unexpected EASYNC return value (for now)");
1311 if (p->p_sysent->sv_errsize) {
1312 if (error >= p->p_sysent->sv_errsize)
1313 error = -1; /* XXX */
1315 error = p->p_sysent->sv_errtbl[error];
1317 frame->tf_rax = error;
1318 frame->tf_rflags |= PSL_C;
1323 * Traced syscall. trapsignal() is not MP aware.
1325 if (orig_tf_rflags & PSL_T) {
1326 MAKEMPSAFE(have_mplock);
1327 frame->tf_rflags &= ~PSL_T;
1328 trapsignal(lp, SIGTRAP, 0);
1332 * Handle reschedule and other end-of-syscall issues
1334 userret(lp, frame, sticks);
1337 if (KTRPOINT(td, KTR_SYSRET)) {
1338 MAKEMPSAFE(have_mplock);
1339 ktrsysret(lp, code, error, args.sysmsg_result);
1344 * This works because errno is findable through the
1345 * register set. If we ever support an emulation where this
1346 * is not the case, this code will need to be revisited.
1348 STOPEVENT(p, S_SCX, code);
1353 * Release the MP lock if we had to get it
1358 KTR_LOG(kernentry_syscall_ret, lp->lwp_proc->p_pid, lp->lwp_tid, error);
1360 KASSERT(&td->td_toks_base == td->td_toks_stop,
1361 ("syscall: critical section count mismatch! %d/%d",
1362 crit_count, td->td_pri));
1363 KASSERT(curstop == td->td_toks_stop,
1364 ("syscall: extra tokens held after trap! %ld",
1365 td->td_toks_stop - &td->td_toks_base));
1370 * NOTE: mplock not held at any point
1373 fork_return(struct lwp *lp, struct trapframe *frame)
1375 frame->tf_rax = 0; /* Child returns zero */
1376 frame->tf_rflags &= ~PSL_C; /* success */
1379 generic_lwp_return(lp, frame);
1380 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1384 * Simplified back end of syscall(), used when returning from fork()
1385 * directly into user mode.
1387 * This code will return back into the fork trampoline code which then
1390 * NOTE: The mplock is not held at any point.
1393 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1395 struct proc *p = lp->lwp_proc;
1398 * Newly forked processes are given a kernel priority. We have to
1399 * adjust the priority to a normal user priority and fake entry
1400 * into the kernel (call userenter()) to install a passive release
1401 * function just in case userret() decides to stop the process. This
1402 * can occur when ^Z races a fork. If we do not install the passive
1403 * release function the current process designation will not be
1404 * released when the thread goes to sleep.
1406 lwkt_setpri_self(TDPRI_USER_NORM);
1407 userenter(lp->lwp_thread, p);
1408 userret(lp, frame, 0);
1410 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1411 ktrsysret(lp, SYS_fork, 0, 0);
1413 lp->lwp_flags |= LWP_PASSIVE_ACQ;
1415 lp->lwp_flags &= ~LWP_PASSIVE_ACQ;
1419 * doreti has turned into this. The frame is directly on the stack. We
1420 * pull everything else we need (fpu and tls context) from the current
1423 * Note on fpu interactions: In a virtual kernel, the fpu context for
1424 * an emulated user mode process is not shared with the virtual kernel's
1425 * fpu context, so we only have to 'stack' fpu contexts within the virtual
1426 * kernel itself, and not even then since the signal() contexts that we care
1427 * about save and restore the FPU state (I think anyhow).
1429 * vmspace_ctl() returns an error only if it had problems instaling the
1430 * context we supplied or problems copying data to/from our VM space.
1433 go_user(struct intrframe *frame)
1435 struct trapframe *tf = (void *)&frame->if_rdi;
1439 * Interrupts may be disabled on entry, make sure all signals
1440 * can be received before beginning our loop.
1445 * Switch to the current simulated user process, then call
1446 * user_trap() when we break out of it (usually due to a signal).
1450 * Tell the real kernel whether it is ok to use the FP
1453 if (mdcpu->gd_npxthread == curthread) {
1454 tf->tf_xflags &= ~PGEX_FPFAULT;
1456 tf->tf_xflags |= PGEX_FPFAULT;
1460 * Run emulated user process context. This call interlocks
1461 * with new mailbox signals.
1463 * Set PGEX_U unconditionally, indicating a user frame (the
1464 * bit is normally set only by T_PAGEFLT).
1466 r = vmspace_ctl(&curproc->p_vmspace->vm_pmap, VMSPACE_CTL_RUN,
1467 tf, &curthread->td_savevext);
1468 frame->if_xflags |= PGEX_U;
1470 kprintf("GO USER %d trap %ld EVA %08lx RIP %08lx RSP %08lx XFLAGS %02lx/%02lx\n",
1471 r, tf->tf_trapno, tf->tf_addr, tf->tf_rip, tf->tf_rsp,
1472 tf->tf_xflags, frame->if_xflags);
1476 panic("vmspace_ctl failed error %d", errno);
1478 if (tf->tf_trapno) {
1482 if (mycpu->gd_reqflags & RQF_AST_MASK) {
1483 tf->tf_trapno = T_ASTFLT;
1491 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1492 * fault (which is then passed back to the virtual kernel) if an attempt is
1493 * made to use the FP unit.
1495 * XXX this is a fairly big hack.
1498 set_vkernel_fp(struct trapframe *frame)
1500 struct thread *td = curthread;
1502 if (frame->tf_xflags & PGEX_FPFAULT) {
1503 td->td_pcb->pcb_flags |= FP_VIRTFP;
1504 if (mdcpu->gd_npxthread == td)
1507 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1512 * Called from vkernel_trap() to fixup the vkernel's syscall
1513 * frame for vmspace_ctl() return.
1516 cpu_vkernel_trap(struct trapframe *frame, int error)
1518 frame->tf_rax = error;
1520 frame->tf_rflags |= PSL_C;
1522 frame->tf_rflags &= ~PSL_C;