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>
96 #define MAKEMPSAFE(have_mplock) \
97 if (have_mplock == 0) { \
102 int (*pmath_emulate) (struct trapframe *);
104 extern int trapwrite (unsigned addr);
106 static int trap_pfault (struct trapframe *, int, vm_offset_t);
107 static void trap_fatal (struct trapframe *, int, vm_offset_t);
108 void dblfault_handler (void);
111 extern inthand_t IDTVEC(syscall);
114 #define MAX_TRAP_MSG 30
115 static char *trap_msg[] = {
117 "privileged instruction fault", /* 1 T_PRIVINFLT */
119 "breakpoint instruction fault", /* 3 T_BPTFLT */
122 "arithmetic trap", /* 6 T_ARITHTRAP */
123 "system forced exception", /* 7 T_ASTFLT */
125 "general protection fault", /* 9 T_PROTFLT */
126 "trace trap", /* 10 T_TRCTRAP */
128 "page fault", /* 12 T_PAGEFLT */
130 "alignment fault", /* 14 T_ALIGNFLT */
134 "integer divide fault", /* 18 T_DIVIDE */
135 "non-maskable interrupt trap", /* 19 T_NMI */
136 "overflow trap", /* 20 T_OFLOW */
137 "FPU bounds check fault", /* 21 T_BOUND */
138 "FPU device not available", /* 22 T_DNA */
139 "double fault", /* 23 T_DOUBLEFLT */
140 "FPU operand fetch fault", /* 24 T_FPOPFLT */
141 "invalid TSS fault", /* 25 T_TSSFLT */
142 "segment not present fault", /* 26 T_SEGNPFLT */
143 "stack fault", /* 27 T_STKFLT */
144 "machine check trap", /* 28 T_MCHK */
145 "SIMD floating-point exception", /* 29 T_XMMFLT */
146 "reserved (unknown) fault", /* 30 T_RESERVED */
150 static int ddb_on_nmi = 1;
151 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
152 &ddb_on_nmi, 0, "Go to DDB on NMI");
154 static int panic_on_nmi = 1;
155 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
156 &panic_on_nmi, 0, "Panic on NMI");
157 static int fast_release;
158 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
159 &fast_release, 0, "Passive Release was optimal");
160 static int slow_release;
161 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
162 &slow_release, 0, "Passive Release was nonoptimal");
164 MALLOC_DEFINE(M_SYSMSG, "sysmsg", "sysmsg structure");
165 extern int max_sysmsg;
168 * Passively intercepts the thread switch function to increase
169 * the thread priority from a user priority to a kernel priority, reducing
170 * syscall and trap overhead for the case where no switch occurs.
172 * Synchronizes td_ucred with p_ucred. This is used by system calls,
173 * signal handling, faults, AST traps, and anything else that enters the
174 * kernel from userland and provides the kernel with a stable read-only
175 * copy of the process ucred.
178 userenter(struct thread *curtd, struct proc *curp)
183 curtd->td_release = lwkt_passive_release;
185 if (curtd->td_ucred != curp->p_ucred) {
186 ncred = crhold(curp->p_ucred);
187 ocred = curtd->td_ucred;
188 curtd->td_ucred = ncred;
195 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
196 * must be completed before we can return to or try to return to userland.
198 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
199 * arithmatic on the delta calculation so the absolute tick values are
200 * truncated to an integer.
203 userret(struct lwp *lp, struct trapframe *frame, int sticks)
205 struct proc *p = lp->lwp_proc;
209 * Charge system time if profiling. Note: times are in microseconds.
210 * This may do a copyout and block, so do it first even though it
211 * means some system time will be charged as user time.
213 if (p->p_flags & P_PROFIL) {
214 addupc_task(p, frame->tf_rip,
215 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
220 * If the jungle wants us dead, so be it.
222 if (lp->lwp_mpflags & LWP_MP_WEXIT) {
223 lwkt_gettoken(&p->p_token);
225 lwkt_reltoken(&p->p_token); /* NOT REACHED */
229 * Block here if we are in a stopped state.
231 if (p->p_stat == SSTOP) {
232 lwkt_gettoken(&p->p_token);
234 lwkt_reltoken(&p->p_token);
239 * Post any pending upcalls. If running a virtual kernel be sure
240 * to restore the virtual kernel's vmspace before posting the upcall.
242 if (p->p_flags & (P_SIGVTALRM | P_SIGPROF | P_UPCALLPEND)) {
243 lwkt_gettoken(&p->p_token);
244 if (p->p_flags & P_SIGVTALRM) {
245 p->p_flags &= ~P_SIGVTALRM;
246 ksignal(p, SIGVTALRM);
248 if (p->p_flags & P_SIGPROF) {
249 p->p_flags &= ~P_SIGPROF;
252 if (p->p_flags & P_UPCALLPEND) {
253 p->p_flags &= ~P_UPCALLPEND;
256 lwkt_reltoken(&p->p_token);
261 * Post any pending signals
263 * WARNING! postsig() can exit and not return.
265 if ((sig = CURSIG_TRACE(lp)) != 0) {
266 lwkt_gettoken(&p->p_token);
268 lwkt_reltoken(&p->p_token);
273 * block here if we are swapped out, but still process signals
274 * (such as SIGKILL). proc0 (the swapin scheduler) is already
275 * aware of our situation, we do not have to wake it up.
277 if (p->p_flags & P_SWAPPEDOUT) {
278 lwkt_gettoken(&p->p_token);
280 p->p_flags |= P_SWAPWAIT;
282 if (p->p_flags & P_SWAPWAIT)
283 tsleep(p, PCATCH, "SWOUT", 0);
284 p->p_flags &= ~P_SWAPWAIT;
286 lwkt_reltoken(&p->p_token);
291 * In a multi-threaded program it is possible for a thread to change
292 * signal state during a system call which temporarily changes the
293 * signal mask. In this case postsig() might not be run and we
294 * have to restore the mask ourselves.
296 if (lp->lwp_flags & LWP_OLDMASK) {
297 lp->lwp_flags &= ~LWP_OLDMASK;
298 lp->lwp_sigmask = lp->lwp_oldsigmask;
304 * Cleanup from userenter and any passive release that might have occured.
305 * We must reclaim the current-process designation before we can return
306 * to usermode. We also handle both LWKT and USER reschedule requests.
309 userexit(struct lwp *lp)
311 struct thread *td = lp->lwp_thread;
312 /* globaldata_t gd = td->td_gd; */
315 * Handle stop requests at kernel priority. Any requests queued
316 * after this loop will generate another AST.
318 while (lp->lwp_proc->p_stat == SSTOP) {
319 lwkt_gettoken(&lp->lwp_proc->p_token);
321 lwkt_reltoken(&lp->lwp_proc->p_token);
325 * Reduce our priority in preparation for a return to userland. If
326 * our passive release function was still in place, our priority was
327 * never raised and does not need to be reduced.
329 lwkt_passive_recover(td);
332 * Become the current user scheduled process if we aren't already,
333 * and deal with reschedule requests and other factors.
335 lp->lwp_proc->p_usched->acquire_curproc(lp);
336 /* WARNING: we may have migrated cpu's */
337 /* gd = td->td_gd; */
340 #if !defined(KTR_KERNENTRY)
341 #define KTR_KERNENTRY KTR_ALL
343 KTR_INFO_MASTER(kernentry);
344 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0,
345 "TRAP(pid %hd, tid %hd, trapno %ld, eva %lu)",
346 pid_t pid, lwpid_t tid, register_t trapno, vm_offset_t eva);
347 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "TRAP_RET(pid %hd, tid %hd)",
348 pid_t pid, lwpid_t tid);
349 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "SYSC(pid %hd, tid %hd, nr %ld)",
350 pid_t pid, lwpid_t tid, register_t trapno);
351 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "SYSRET(pid %hd, tid %hd, err %d)",
352 pid_t pid, lwpid_t tid, int err);
353 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "FORKRET(pid %hd, tid %hd)",
354 pid_t pid, lwpid_t tid);
357 * Exception, fault, and trap interface to the kernel.
358 * This common code is called from assembly language IDT gate entry
359 * routines that prepare a suitable stack frame, and restore this
360 * frame after the exception has been processed.
362 * This function is also called from doreti in an interlock to handle ASTs.
363 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
365 * NOTE! We have to retrieve the fault address prior to obtaining the
366 * MP lock because get_mplock() may switch out. YYY cr2 really ought
367 * to be retrieved by the assembly code, not here.
369 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
370 * if an attempt is made to switch from a fast interrupt or IPI. This is
371 * necessary to properly take fatal kernel traps on SMP machines if
372 * get_mplock() has to block.
376 user_trap(struct trapframe *frame)
378 struct globaldata *gd = mycpu;
379 struct thread *td = gd->gd_curthread;
380 struct lwp *lp = td->td_lwp;
383 int i = 0, ucode = 0, type, code;
386 int crit_count = td->td_critcount;
387 lwkt_tokref_t curstop = td->td_toks_stop;
393 if (frame->tf_trapno == T_PAGEFLT)
394 eva = frame->tf_addr;
398 kprintf("USER_TRAP AT %08lx xflags %ld trapno %ld eva %08lx\n",
399 frame->tf_rip, frame->tf_xflags, frame->tf_trapno, eva);
403 * Everything coming from user mode runs through user_trap,
404 * including system calls.
406 if (frame->tf_trapno == T_FAST_SYSCALL) {
411 KTR_LOG(kernentry_trap, lp->lwp_proc->p_pid, lp->lwp_tid,
412 frame->tf_trapno, eva);
416 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
417 ++gd->gd_trap_nesting_level;
418 MAKEMPSAFE(have_mplock);
419 trap_fatal(frame, TRUE, eva);
420 --gd->gd_trap_nesting_level;
425 type = frame->tf_trapno;
426 code = frame->tf_err;
430 sticks = (int)td->td_sticks;
431 lp->lwp_md.md_regs = frame;
434 case T_PRIVINFLT: /* privileged instruction fault */
439 case T_BPTFLT: /* bpt instruction fault */
440 case T_TRCTRAP: /* trace trap */
441 frame->tf_rflags &= ~PSL_T;
443 ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
446 case T_ARITHTRAP: /* arithmetic trap */
451 case T_ASTFLT: /* Allow process switch */
452 mycpu->gd_cnt.v_soft++;
453 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
454 atomic_clear_int(&mycpu->gd_reqflags, RQF_AST_OWEUPC);
455 addupc_task(p, p->p_prof.pr_addr, p->p_prof.pr_ticks);
460 * The following two traps can happen in
461 * vm86 mode, and, if so, we want to handle
464 case T_PROTFLT: /* general protection fault */
465 case T_STKFLT: /* stack fault */
467 if (frame->tf_eflags & PSL_VM) {
468 i = vm86_emulate((struct vm86frame *)frame);
476 case T_SEGNPFLT: /* segment not present fault */
477 case T_TSSFLT: /* invalid TSS fault */
478 case T_DOUBLEFLT: /* double fault */
481 ucode = code + BUS_SEGM_FAULT ;
484 case T_PAGEFLT: /* page fault */
485 MAKEMPSAFE(have_mplock);
486 i = trap_pfault(frame, TRUE, eva);
487 if (i == -1 || i == 0)
499 case T_DIVIDE: /* integer divide fault */
506 MAKEMPSAFE(have_mplock);
507 /* machine/parity/power fail/"kitchen sink" faults */
508 if (isa_nmi(code) == 0) {
511 * NMI can be hooked up to a pushbutton
515 kprintf ("NMI ... going to debugger\n");
516 kdb_trap(type, 0, frame);
520 } else if (panic_on_nmi)
521 panic("NMI indicates hardware failure");
523 #endif /* NISA > 0 */
525 case T_OFLOW: /* integer overflow fault */
530 case T_BOUND: /* bounds check fault */
537 * Virtual kernel intercept - pass the DNA exception
538 * to the (emulated) virtual kernel if it asked to handle
539 * it. This occurs when the virtual kernel is holding
540 * onto the FP context for a different emulated
541 * process then the one currently running.
543 * We must still call npxdna() since we may have
544 * saved FP state that the (emulated) virtual kernel
545 * needs to hand over to a different emulated process.
547 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
548 (td->td_pcb->pcb_flags & FP_VIRTFP)
555 * The kernel may have switched out the FP unit's
556 * state, causing the user process to take a fault
557 * when it tries to use the FP unit. Restore the
562 if (!pmath_emulate) {
564 ucode = FPE_FPU_NP_TRAP;
567 i = (*pmath_emulate)(frame);
569 if (!(frame->tf_rflags & PSL_T))
571 frame->tf_rflags &= ~PSL_T;
574 /* else ucode = emulator_only_knows() XXX */
577 case T_FPOPFLT: /* FPU operand fetch fault */
582 case T_XMMFLT: /* SIMD floating-point exception */
589 * Virtual kernel intercept - if the fault is directly related to a
590 * VM context managed by a virtual kernel then let the virtual kernel
593 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
594 vkernel_trap(lp, frame);
599 * Translate fault for emulators (e.g. Linux)
601 if (*p->p_sysent->sv_transtrap)
602 i = (*p->p_sysent->sv_transtrap)(i, type);
604 MAKEMPSAFE(have_mplock);
605 trapsignal(lp, i, ucode);
608 if (type <= MAX_TRAP_MSG) {
609 uprintf("fatal process exception: %s",
611 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
612 uprintf(", fault VA = 0x%lx", (u_long)eva);
618 userret(lp, frame, sticks);
623 KTR_LOG(kernentry_trap_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
625 KASSERT(crit_count == td->td_critcount,
626 ("trap: critical section count mismatch! %d/%d",
627 crit_count, td->td_pri));
628 KASSERT(curstop == td->td_toks_stop,
629 ("trap: extra tokens held after trap! %ld/%ld",
630 curstop - &td->td_toks_base,
631 td->td_toks_stop - &td->td_toks_base));
636 kern_trap(struct trapframe *frame)
638 struct globaldata *gd = mycpu;
639 struct thread *td = gd->gd_curthread;
642 int i = 0, ucode = 0, type, code;
645 int crit_count = td->td_critcount;
646 lwkt_tokref_t curstop = td->td_toks_stop;
653 if (frame->tf_trapno == T_PAGEFLT)
654 eva = frame->tf_addr;
660 ++gd->gd_trap_nesting_level;
661 MAKEMPSAFE(have_mplock);
662 trap_fatal(frame, FALSE, eva);
663 --gd->gd_trap_nesting_level;
668 type = frame->tf_trapno;
669 code = frame->tf_err;
677 case T_PAGEFLT: /* page fault */
678 MAKEMPSAFE(have_mplock);
679 trap_pfault(frame, FALSE, eva);
684 * The kernel may be using npx for copying or other
687 panic("kernel NPX should not happen");
692 case T_PROTFLT: /* general protection fault */
693 case T_SEGNPFLT: /* segment not present fault */
695 * Invalid segment selectors and out of bounds
696 * %eip's and %esp's can be set up in user mode.
697 * This causes a fault in kernel mode when the
698 * kernel tries to return to user mode. We want
699 * to get this fault so that we can fix the
700 * problem here and not have to check all the
701 * selectors and pointers when the user changes
704 if (mycpu->gd_intr_nesting_level == 0) {
705 if (td->td_pcb->pcb_onfault) {
707 (register_t)td->td_pcb->pcb_onfault;
715 * PSL_NT can be set in user mode and isn't cleared
716 * automatically when the kernel is entered. This
717 * causes a TSS fault when the kernel attempts to
718 * `iret' because the TSS link is uninitialized. We
719 * want to get this fault so that we can fix the
720 * problem here and not every time the kernel is
723 if (frame->tf_rflags & PSL_NT) {
724 frame->tf_rflags &= ~PSL_NT;
729 case T_TRCTRAP: /* trace trap */
731 if (frame->tf_eip == (int)IDTVEC(syscall)) {
733 * We've just entered system mode via the
734 * syscall lcall. Continue single stepping
735 * silently until the syscall handler has
740 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
742 * The syscall handler has now saved the
743 * flags. Stop single stepping it.
745 frame->tf_eflags &= ~PSL_T;
751 * Ignore debug register trace traps due to
752 * accesses in the user's address space, which
753 * can happen under several conditions such as
754 * if a user sets a watchpoint on a buffer and
755 * then passes that buffer to a system call.
756 * We still want to get TRCTRAPS for addresses
757 * in kernel space because that is useful when
758 * debugging the kernel.
760 if (user_dbreg_trap()) {
762 * Reset breakpoint bits because the
765 load_dr6(rdr6() & 0xfffffff0);
770 * Fall through (TRCTRAP kernel mode, kernel address)
774 * If DDB is enabled, let it handle the debugger trap.
775 * Otherwise, debugger traps "can't happen".
778 MAKEMPSAFE(have_mplock);
779 if (kdb_trap (type, 0, frame))
784 MAKEMPSAFE(have_mplock);
785 trap_fatal(frame, FALSE, eva);
788 MAKEMPSAFE(have_mplock);
789 trap_fatal(frame, FALSE, eva);
794 * Ignore this trap generated from a spurious SIGTRAP.
796 * single stepping in / syscalls leads to spurious / SIGTRAP
799 * Haiku (c) 2007 Simon 'corecode' Schubert
805 * Translate fault for emulators (e.g. Linux)
807 if (*p->p_sysent->sv_transtrap)
808 i = (*p->p_sysent->sv_transtrap)(i, type);
810 MAKEMPSAFE(have_mplock);
811 trapsignal(lp, i, ucode);
814 if (type <= MAX_TRAP_MSG) {
815 uprintf("fatal process exception: %s",
817 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
818 uprintf(", fault VA = 0x%lx", (u_long)eva);
828 KASSERT(crit_count == td->td_critcount,
829 ("trap: critical section count mismatch! %d/%d",
830 crit_count, td->td_pri));
831 KASSERT(curstop == td->td_toks_stop,
832 ("trap: extra tokens held after trap! %ld/%ld",
833 curstop - &td->td_toks_base,
834 td->td_toks_stop - &td->td_toks_base));
839 trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
842 struct vmspace *vm = NULL;
846 thread_t td = curthread;
847 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)) {
900 fault_flags |= VM_FAULT_BURST;
901 if (ftype & VM_PROT_WRITE)
902 fault_flags |= VM_FAULT_DIRTY;
904 fault_flags |= VM_FAULT_NORMAL;
905 rv = vm_fault(map, va, ftype, fault_flags);
910 * Don't have to worry about process locking or stacks in the kernel.
912 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
915 if (rv == KERN_SUCCESS)
919 if (td->td_gd->gd_intr_nesting_level == 0 &&
920 td->td_pcb->pcb_onfault) {
921 frame->tf_rip = (register_t)td->td_pcb->pcb_onfault;
924 trap_fatal(frame, usermode, eva);
929 * NOTE: on x86_64 we have a tf_addr field in the trapframe, no
930 * kludge is needed to pass the fault address to signal handlers.
932 struct proc *p = td->td_proc;
933 kprintf("seg-fault accessing address %p rip=%p pid=%d p_comm=%s\n",
934 (void *)va, (void *)frame->tf_rip, p->p_pid, p->p_comm);
935 /* Debugger("seg-fault"); */
937 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
941 trap_fatal(struct trapframe *frame, int usermode, vm_offset_t eva)
946 code = frame->tf_xflags;
947 type = frame->tf_trapno;
949 if (type <= MAX_TRAP_MSG) {
950 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
951 type, trap_msg[type],
952 (usermode ? "user" : "kernel"));
954 /* two separate prints in case of a trap on an unmapped page */
955 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
956 if (type == T_PAGEFLT) {
957 kprintf("fault virtual address = %p\n", (void *)eva);
958 kprintf("fault code = %s %s, %s\n",
959 usermode ? "user" : "supervisor",
960 code & PGEX_W ? "write" : "read",
961 code & PGEX_P ? "protection violation" : "page not present");
963 kprintf("instruction pointer = 0x%lx:0x%lx\n",
964 frame->tf_cs & 0xffff, frame->tf_rip);
966 ss = frame->tf_ss & 0xffff;
969 ss = GSEL(GDATA_SEL, SEL_KPL);
970 rsp = (long)&frame->tf_rsp;
972 kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp);
973 kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp);
974 kprintf("processor eflags = ");
975 if (frame->tf_rflags & PSL_T)
976 kprintf("trace trap, ");
977 if (frame->tf_rflags & PSL_I)
978 kprintf("interrupt enabled, ");
979 if (frame->tf_rflags & PSL_NT)
980 kprintf("nested task, ");
981 if (frame->tf_rflags & PSL_RF)
984 if (frame->tf_eflags & PSL_VM)
987 kprintf("IOPL = %jd\n", (intmax_t)((frame->tf_rflags & PSL_IOPL) >> 12));
988 kprintf("current process = ");
990 kprintf("%lu (%s)\n",
991 (u_long)curproc->p_pid, curproc->p_comm ?
992 curproc->p_comm : "");
996 kprintf("current thread = pri %d ", curthread->td_pri);
997 if (curthread->td_critcount)
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");
1013 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1016 kprintf("trap number = %d\n", type);
1017 if (type <= MAX_TRAP_MSG)
1018 panic("%s", trap_msg[type]);
1020 panic("unknown/reserved trap");
1024 * Double fault handler. Called when a fault occurs while writing
1025 * a frame for a trap/exception onto the stack. This usually occurs
1026 * when the stack overflows (such is the case with infinite recursion,
1029 * XXX Note that the current PTD gets replaced by IdlePTD when the
1030 * task switch occurs. This means that the stack that was active at
1031 * the time of the double fault is not available at <kstack> unless
1032 * the machine was idle when the double fault occurred. The downside
1033 * of this is that "trace <ebp>" in ddb won't work.
1036 dblfault_handler(void)
1039 struct mdglobaldata *gd = mdcpu;
1042 kprintf("\nFatal double fault:\n");
1044 kprintf("rip = 0x%lx\n", gd->gd_common_tss.tss_rip);
1045 kprintf("rsp = 0x%lx\n", gd->gd_common_tss.tss_rsp);
1046 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("cpuid = %d\n", mycpu->gd_cpuid);
1050 panic("double fault");
1054 * Compensate for 386 brain damage (missing URKR).
1055 * This is a little simpler than the pagefault handler in trap() because
1056 * it the page tables have already been faulted in and high addresses
1057 * are thrown out early for other reasons.
1060 trapwrite(unsigned addr)
1067 va = trunc_page((vm_offset_t)addr);
1069 * XXX - MAX is END. Changed > to >= for temp. fix.
1071 if (va >= VM_MAX_USER_ADDRESS)
1074 lp = curthread->td_lwp;
1075 vm = lp->lwp_vmspace;
1077 PHOLD(lp->lwp_proc);
1079 if (!grow_stack (lp->lwp_proc, va)) {
1080 PRELE(lp->lwp_proc);
1085 * fault the data page
1087 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY);
1089 PRELE(lp->lwp_proc);
1091 if (rv != KERN_SUCCESS)
1098 * syscall2 - MP aware system call request C handler
1100 * A system call is essentially treated as a trap except that the
1101 * MP lock is not held on entry or return. We are responsible for
1102 * obtaining the MP lock if necessary and for handling ASTs
1103 * (e.g. a task switch) prior to return.
1105 * In general, only simple access and manipulation of curproc and
1106 * the current stack is allowed without having to hold MP lock.
1108 * MPSAFE - note that large sections of this routine are run without
1112 syscall2(struct trapframe *frame)
1114 struct thread *td = curthread;
1115 struct proc *p = td->td_proc;
1116 struct lwp *lp = td->td_lwp;
1118 struct sysent *callp;
1119 register_t orig_tf_rflags;
1124 int crit_count = td->td_critcount;
1125 lwkt_tokref_t curstop = td->td_toks_stop;
1127 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);
1330 * Release the MP lock if we had to get it
1334 KTR_LOG(kernentry_syscall_ret, lp->lwp_proc->p_pid, lp->lwp_tid, error);
1336 KASSERT(&td->td_toks_base == td->td_toks_stop,
1337 ("syscall: critical section count mismatch! %d/%d",
1338 crit_count, td->td_pri));
1339 KASSERT(curstop == td->td_toks_stop,
1340 ("syscall: extra tokens held after trap! %ld",
1341 td->td_toks_stop - &td->td_toks_base));
1346 * NOTE: mplock not held at any point
1349 fork_return(struct lwp *lp, struct trapframe *frame)
1351 frame->tf_rax = 0; /* Child returns zero */
1352 frame->tf_rflags &= ~PSL_C; /* success */
1355 generic_lwp_return(lp, frame);
1356 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1360 * Simplified back end of syscall(), used when returning from fork()
1361 * directly into user mode.
1363 * This code will return back into the fork trampoline code which then
1366 * NOTE: The mplock is not held at any point.
1369 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1371 struct proc *p = lp->lwp_proc;
1374 * Newly forked processes are given a kernel priority. We have to
1375 * adjust the priority to a normal user priority and fake entry
1376 * into the kernel (call userenter()) to install a passive release
1377 * function just in case userret() decides to stop the process. This
1378 * can occur when ^Z races a fork. If we do not install the passive
1379 * release function the current process designation will not be
1380 * released when the thread goes to sleep.
1382 lwkt_setpri_self(TDPRI_USER_NORM);
1383 userenter(lp->lwp_thread, p);
1384 userret(lp, frame, 0);
1386 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1387 ktrsysret(lp, SYS_fork, 0, 0);
1389 lp->lwp_flags |= LWP_PASSIVE_ACQ;
1391 lp->lwp_flags &= ~LWP_PASSIVE_ACQ;
1395 * doreti has turned into this. The frame is directly on the stack. We
1396 * pull everything else we need (fpu and tls context) from the current
1399 * Note on fpu interactions: In a virtual kernel, the fpu context for
1400 * an emulated user mode process is not shared with the virtual kernel's
1401 * fpu context, so we only have to 'stack' fpu contexts within the virtual
1402 * kernel itself, and not even then since the signal() contexts that we care
1403 * about save and restore the FPU state (I think anyhow).
1405 * vmspace_ctl() returns an error only if it had problems instaling the
1406 * context we supplied or problems copying data to/from our VM space.
1409 go_user(struct intrframe *frame)
1411 struct trapframe *tf = (void *)&frame->if_rdi;
1415 * Interrupts may be disabled on entry, make sure all signals
1416 * can be received before beginning our loop.
1421 * Switch to the current simulated user process, then call
1422 * user_trap() when we break out of it (usually due to a signal).
1426 * Tell the real kernel whether it is ok to use the FP
1429 if (mdcpu->gd_npxthread == curthread) {
1430 tf->tf_xflags &= ~PGEX_FPFAULT;
1432 tf->tf_xflags |= PGEX_FPFAULT;
1436 * Run emulated user process context. This call interlocks
1437 * with new mailbox signals.
1439 * Set PGEX_U unconditionally, indicating a user frame (the
1440 * bit is normally set only by T_PAGEFLT).
1442 r = vmspace_ctl(&curproc->p_vmspace->vm_pmap, VMSPACE_CTL_RUN,
1443 tf, &curthread->td_savevext);
1444 frame->if_xflags |= PGEX_U;
1446 kprintf("GO USER %d trap %ld EVA %08lx RIP %08lx RSP %08lx XFLAGS %02lx/%02lx\n",
1447 r, tf->tf_trapno, tf->tf_addr, tf->tf_rip, tf->tf_rsp,
1448 tf->tf_xflags, frame->if_xflags);
1452 panic("vmspace_ctl failed error %d", errno);
1454 if (tf->tf_trapno) {
1458 if (mycpu->gd_reqflags & RQF_AST_MASK) {
1459 tf->tf_trapno = T_ASTFLT;
1467 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1468 * fault (which is then passed back to the virtual kernel) if an attempt is
1469 * made to use the FP unit.
1471 * XXX this is a fairly big hack.
1474 set_vkernel_fp(struct trapframe *frame)
1476 struct thread *td = curthread;
1478 if (frame->tf_xflags & PGEX_FPFAULT) {
1479 td->td_pcb->pcb_flags |= FP_VIRTFP;
1480 if (mdcpu->gd_npxthread == td)
1483 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1488 * Called from vkernel_trap() to fixup the vkernel's syscall
1489 * frame for vmspace_ctl() return.
1492 cpu_vkernel_trap(struct trapframe *frame, int error)
1494 frame->tf_rax = error;
1496 frame->tf_rflags |= PSL_C;
1498 frame->tf_rflags &= ~PSL_C;