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) {
233 rel_mplock(); /* 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_nonlocked(&mycpu->gd_reqflags,
448 addupc_task(p, p->p_prof.pr_addr,
454 * The following two traps can happen in
455 * vm86 mode, and, if so, we want to handle
458 case T_PROTFLT: /* general protection fault */
459 case T_STKFLT: /* stack fault */
461 if (frame->tf_eflags & PSL_VM) {
462 i = vm86_emulate((struct vm86frame *)frame);
470 case T_SEGNPFLT: /* segment not present fault */
471 case T_TSSFLT: /* invalid TSS fault */
472 case T_DOUBLEFLT: /* double fault */
474 ucode = code + BUS_SEGM_FAULT ;
478 case T_PAGEFLT: /* page fault */
479 MAKEMPSAFE(have_mplock);
480 i = trap_pfault(frame, TRUE, eva);
483 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
493 case T_DIVIDE: /* integer divide fault */
500 MAKEMPSAFE(have_mplock);
501 /* machine/parity/power fail/"kitchen sink" faults */
502 if (isa_nmi(code) == 0) {
505 * NMI can be hooked up to a pushbutton
509 kprintf ("NMI ... going to debugger\n");
510 kdb_trap (type, 0, frame);
514 } else if (panic_on_nmi)
515 panic("NMI indicates hardware failure");
517 #endif /* NISA > 0 */
519 case T_OFLOW: /* integer overflow fault */
524 case T_BOUND: /* bounds check fault */
531 * Virtual kernel intercept - pass the DNA exception
532 * to the (emulated) virtual kernel if it asked to handle
533 * it. This occurs when the virtual kernel is holding
534 * onto the FP context for a different emulated
535 * process then the one currently running.
537 * We must still call npxdna() since we may have
538 * saved FP state that the (emulated) virtual kernel
539 * needs to hand over to a different emulated process.
541 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
542 (td->td_pcb->pcb_flags & FP_VIRTFP)
548 * The kernel may have switched out the FP unit's
549 * state, causing the user process to take a fault
550 * when it tries to use the FP unit. Restore the
555 if (!pmath_emulate) {
557 ucode = FPE_FPU_NP_TRAP;
560 i = (*pmath_emulate)(frame);
562 if (!(frame->tf_rflags & PSL_T))
564 frame->tf_rflags &= ~PSL_T;
567 /* else ucode = emulator_only_knows() XXX */
570 case T_FPOPFLT: /* FPU operand fetch fault */
575 case T_XMMFLT: /* SIMD floating-point exception */
582 * Virtual kernel intercept - if the fault is directly related to a
583 * VM context managed by a virtual kernel then let the virtual kernel
586 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
587 vkernel_trap(lp, frame);
592 * Translate fault for emulators (e.g. Linux)
594 if (*p->p_sysent->sv_transtrap)
595 i = (*p->p_sysent->sv_transtrap)(i, type);
597 MAKEMPSAFE(have_mplock);
598 trapsignal(lp, i, ucode);
601 if (type <= MAX_TRAP_MSG) {
602 uprintf("fatal process exception: %s",
604 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
605 uprintf(", fault VA = 0x%lx", (u_long)eva);
611 userret(lp, frame, sticks);
618 KTR_LOG(kernentry_trap_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
620 KASSERT(crit_count == td->td_critcount,
621 ("trap: critical section count mismatch! %d/%d",
622 crit_count, td->td_pri));
623 KASSERT(curstop == td->td_toks_stop,
624 ("trap: extra tokens held after trap! %ld/%ld",
625 curstop - &td->td_toks_base,
626 td->td_toks_stop - &td->td_toks_base));
631 kern_trap(struct trapframe *frame)
633 struct globaldata *gd = mycpu;
634 struct thread *td = gd->gd_curthread;
637 int i = 0, ucode = 0, type, code;
642 int crit_count = td->td_critcount;
643 lwkt_tokref_t curstop = td->td_toks_stop;
650 if (frame->tf_trapno == T_PAGEFLT)
651 eva = frame->tf_addr;
657 ++gd->gd_trap_nesting_level;
658 MAKEMPSAFE(have_mplock);
659 trap_fatal(frame, FALSE, eva);
660 --gd->gd_trap_nesting_level;
665 type = frame->tf_trapno;
666 code = frame->tf_err;
674 case T_PAGEFLT: /* page fault */
675 MAKEMPSAFE(have_mplock);
676 trap_pfault(frame, FALSE, eva);
681 * The kernel may be using npx for copying or other
684 panic("kernel NPX should not happen");
689 case T_PROTFLT: /* general protection fault */
690 case T_SEGNPFLT: /* segment not present fault */
692 * Invalid segment selectors and out of bounds
693 * %eip's and %esp's can be set up in user mode.
694 * This causes a fault in kernel mode when the
695 * kernel tries to return to user mode. We want
696 * to get this fault so that we can fix the
697 * problem here and not have to check all the
698 * selectors and pointers when the user changes
701 if (mycpu->gd_intr_nesting_level == 0) {
702 if (td->td_pcb->pcb_onfault) {
704 (register_t)td->td_pcb->pcb_onfault;
712 * PSL_NT can be set in user mode and isn't cleared
713 * automatically when the kernel is entered. This
714 * causes a TSS fault when the kernel attempts to
715 * `iret' because the TSS link is uninitialized. We
716 * want to get this fault so that we can fix the
717 * problem here and not every time the kernel is
720 if (frame->tf_rflags & PSL_NT) {
721 frame->tf_rflags &= ~PSL_NT;
726 case T_TRCTRAP: /* trace trap */
728 if (frame->tf_eip == (int)IDTVEC(syscall)) {
730 * We've just entered system mode via the
731 * syscall lcall. Continue single stepping
732 * silently until the syscall handler has
737 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
739 * The syscall handler has now saved the
740 * flags. Stop single stepping it.
742 frame->tf_eflags &= ~PSL_T;
748 * Ignore debug register trace traps due to
749 * accesses in the user's address space, which
750 * can happen under several conditions such as
751 * if a user sets a watchpoint on a buffer and
752 * then passes that buffer to a system call.
753 * We still want to get TRCTRAPS for addresses
754 * in kernel space because that is useful when
755 * debugging the kernel.
757 if (user_dbreg_trap()) {
759 * Reset breakpoint bits because the
762 load_dr6(rdr6() & 0xfffffff0);
767 * Fall through (TRCTRAP kernel mode, kernel address)
771 * If DDB is enabled, let it handle the debugger trap.
772 * Otherwise, debugger traps "can't happen".
775 MAKEMPSAFE(have_mplock);
776 if (kdb_trap (type, 0, frame))
781 MAKEMPSAFE(have_mplock);
782 trap_fatal(frame, FALSE, eva);
785 MAKEMPSAFE(have_mplock);
786 trap_fatal(frame, FALSE, eva);
791 * Ignore this trap generated from a spurious SIGTRAP.
793 * single stepping in / syscalls leads to spurious / SIGTRAP
796 * Haiku (c) 2007 Simon 'corecode' Schubert
802 * Translate fault for emulators (e.g. Linux)
804 if (*p->p_sysent->sv_transtrap)
805 i = (*p->p_sysent->sv_transtrap)(i, type);
807 MAKEMPSAFE(have_mplock);
808 trapsignal(lp, i, ucode);
811 if (type <= MAX_TRAP_MSG) {
812 uprintf("fatal process exception: %s",
814 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
815 uprintf(", fault VA = 0x%lx", (u_long)eva);
827 KASSERT(crit_count == td->td_critcount,
828 ("trap: critical section count mismatch! %d/%d",
829 crit_count, td->td_pri));
830 KASSERT(curstop == td->td_toks_stop,
831 ("trap: extra tokens held after trap! %ld/%ld",
832 curstop - &td->td_toks_base,
833 td->td_toks_stop - &td->td_toks_base));
838 trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
841 struct vmspace *vm = NULL;
845 thread_t td = curthread;
846 struct lwp *lp = td->td_lwp;
848 va = trunc_page(eva);
849 if (usermode == FALSE) {
851 * This is a fault on kernel virtual memory.
856 * This is a fault on non-kernel virtual memory.
857 * vm is initialized above to NULL. If curproc is NULL
858 * or curproc->p_vmspace is NULL the fault is fatal.
861 vm = lp->lwp_vmspace;
869 if (frame->tf_err & PGEX_W)
870 ftype = VM_PROT_READ | VM_PROT_WRITE;
872 ftype = VM_PROT_READ;
874 if (map != &kernel_map) {
876 * Keep swapout from messing with us during this
882 * Grow the stack if necessary
884 /* grow_stack returns false only if va falls into
885 * a growable stack region and the stack growth
886 * fails. It returns true if va was not within
887 * a growable stack region, or if the stack
890 if (!grow_stack (lp->lwp_proc, va)) {
896 /* Fault in the user page: */
897 rv = vm_fault(map, va, ftype,
898 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
904 * Don't have to worry about process locking or stacks in the kernel.
906 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
909 if (rv == KERN_SUCCESS)
913 if (td->td_gd->gd_intr_nesting_level == 0 &&
914 td->td_pcb->pcb_onfault) {
915 frame->tf_rip = (register_t)td->td_pcb->pcb_onfault;
918 trap_fatal(frame, usermode, eva);
923 * NOTE: on x86_64 we have a tf_addr field in the trapframe, no
924 * kludge is needed to pass the fault address to signal handlers.
926 struct proc *p = td->td_proc;
927 kprintf("seg-fault accessing address %p rip=%p pid=%d p_comm=%s\n",
928 (void *)va, (void *)frame->tf_rip, p->p_pid, p->p_comm);
929 /* Debugger("seg-fault"); */
931 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
935 trap_fatal(struct trapframe *frame, int usermode, vm_offset_t eva)
940 code = frame->tf_xflags;
941 type = frame->tf_trapno;
943 if (type <= MAX_TRAP_MSG) {
944 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
945 type, trap_msg[type],
946 (usermode ? "user" : "kernel"));
949 /* two separate prints in case of a trap on an unmapped page */
950 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
952 if (type == T_PAGEFLT) {
953 kprintf("fault virtual address = %p\n", (void *)eva);
954 kprintf("fault code = %s %s, %s\n",
955 usermode ? "user" : "supervisor",
956 code & PGEX_W ? "write" : "read",
957 code & PGEX_P ? "protection violation" : "page not present");
959 kprintf("instruction pointer = 0x%lx:0x%lx\n",
960 frame->tf_cs & 0xffff, frame->tf_rip);
962 ss = frame->tf_ss & 0xffff;
965 ss = GSEL(GDATA_SEL, SEL_KPL);
966 rsp = (long)&frame->tf_rsp;
968 kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp);
969 kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp);
970 kprintf("processor eflags = ");
971 if (frame->tf_rflags & PSL_T)
972 kprintf("trace trap, ");
973 if (frame->tf_rflags & PSL_I)
974 kprintf("interrupt enabled, ");
975 if (frame->tf_rflags & PSL_NT)
976 kprintf("nested task, ");
977 if (frame->tf_rflags & PSL_RF)
980 if (frame->tf_eflags & PSL_VM)
983 kprintf("IOPL = %jd\n", (intmax_t)((frame->tf_rflags & PSL_IOPL) >> 12));
984 kprintf("current process = ");
986 kprintf("%lu (%s)\n",
987 (u_long)curproc->p_pid, curproc->p_comm ?
988 curproc->p_comm : "");
992 kprintf("current thread = pri %d ", curthread->td_pri);
993 if (curthread->td_critcount)
999 * we probably SHOULD have stopped the other CPUs before now!
1000 * another CPU COULD have been touching cpl at this moment...
1002 kprintf(" <- SMP: XXX");
1011 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1014 kprintf("trap number = %d\n", type);
1015 if (type <= MAX_TRAP_MSG)
1016 panic("%s", trap_msg[type]);
1018 panic("unknown/reserved trap");
1022 * Double fault handler. Called when a fault occurs while writing
1023 * a frame for a trap/exception onto the stack. This usually occurs
1024 * when the stack overflows (such is the case with infinite recursion,
1027 * XXX Note that the current PTD gets replaced by IdlePTD when the
1028 * task switch occurs. This means that the stack that was active at
1029 * the time of the double fault is not available at <kstack> unless
1030 * the machine was idle when the double fault occurred. The downside
1031 * of this is that "trace <ebp>" in ddb won't work.
1034 dblfault_handler(void)
1037 struct mdglobaldata *gd = mdcpu;
1040 kprintf("\nFatal double fault:\n");
1042 kprintf("rip = 0x%lx\n", gd->gd_common_tss.tss_rip);
1043 kprintf("rsp = 0x%lx\n", gd->gd_common_tss.tss_rsp);
1044 kprintf("rbp = 0x%lx\n", gd->gd_common_tss.tss_rbp);
1047 /* two separate prints in case of a trap on an unmapped page */
1048 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;
1128 int have_mplock = 0;
1133 union sysunion args;
1134 register_t *argsdst;
1136 mycpu->gd_cnt.v_syscall++;
1138 KTR_LOG(kernentry_syscall, lp->lwp_proc->p_pid, lp->lwp_tid,
1141 userenter(td, p); /* lazy raise our priority */
1148 sticks = (int)td->td_sticks;
1149 orig_tf_rflags = frame->tf_rflags;
1152 * Virtual kernel intercept - if a VM context managed by a virtual
1153 * kernel issues a system call the virtual kernel handles it, not us.
1154 * Restore the virtual kernel context and return from its system
1155 * call. The current frame is copied out to the virtual kernel.
1157 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1158 vkernel_trap(lp, frame);
1159 error = EJUSTRETURN;
1164 * Get the system call parameters and account for time
1166 lp->lwp_md.md_regs = frame;
1167 params = (caddr_t)frame->tf_rsp + sizeof(register_t);
1168 code = frame->tf_rax;
1170 if (p->p_sysent->sv_prepsyscall) {
1171 (*p->p_sysent->sv_prepsyscall)(
1172 frame, (int *)(&args.nosys.sysmsg + 1),
1175 if (code == SYS_syscall || code == SYS___syscall) {
1176 code = frame->tf_rdi;
1182 if (p->p_sysent->sv_mask)
1183 code &= p->p_sysent->sv_mask;
1185 if (code >= p->p_sysent->sv_size)
1186 callp = &p->p_sysent->sv_table[0];
1188 callp = &p->p_sysent->sv_table[code];
1190 narg = callp->sy_narg & SYF_ARGMASK;
1193 * On x86_64 we get up to six arguments in registers. The rest are
1194 * on the stack. The first six members of 'struct trapframe' happen
1195 * to be the registers used to pass arguments, in exactly the right
1198 argp = &frame->tf_rdi;
1200 argsdst = (register_t *)(&args.nosys.sysmsg + 1);
1202 * JG can we overflow the space pointed to by 'argsdst'
1203 * either with 'bcopy' or with 'copyin'?
1205 bcopy(argp, argsdst, sizeof(register_t) * regcnt);
1207 * copyin is MP aware, but the tracing code is not
1209 if (narg > regcnt) {
1210 KASSERT(params != NULL, ("copyin args with no params!"));
1211 error = copyin(params, &argsdst[regcnt],
1212 (narg - regcnt) * sizeof(register_t));
1215 if (KTRPOINT(td, KTR_SYSCALL)) {
1216 MAKEMPSAFE(have_mplock);
1218 ktrsyscall(lp, code, narg,
1219 (void *)(&args.nosys.sysmsg + 1));
1227 if (KTRPOINT(td, KTR_SYSCALL)) {
1228 MAKEMPSAFE(have_mplock);
1229 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1234 * Default return value is 0 (will be copied to %rax). Double-value
1235 * returns use %rax and %rdx. %rdx is left unchanged for system
1236 * calls which return only one result.
1238 args.sysmsg_fds[0] = 0;
1239 args.sysmsg_fds[1] = frame->tf_rdx;
1242 * The syscall might manipulate the trap frame. If it does it
1243 * will probably return EJUSTRETURN.
1245 args.sysmsg_frame = frame;
1247 STOPEVENT(p, S_SCE, narg); /* MP aware */
1250 * NOTE: All system calls run MPSAFE now. The system call itself
1251 * is responsible for getting the MP lock.
1253 error = (*callp->sy_call)(&args);
1256 kprintf("system call %d returned %d\n", code, error);
1261 * MP SAFE (we may or may not have the MP lock at this point)
1266 * Reinitialize proc pointer `p' as it may be different
1267 * if this is a child returning from fork syscall.
1270 lp = curthread->td_lwp;
1271 frame->tf_rax = args.sysmsg_fds[0];
1272 frame->tf_rdx = args.sysmsg_fds[1];
1273 frame->tf_rflags &= ~PSL_C;
1277 * Reconstruct pc, we know that 'syscall' is 2 bytes.
1278 * We have to do a full context restore so that %r10
1279 * (which was holding the value of %rcx) is restored for
1280 * the next iteration.
1282 frame->tf_rip -= frame->tf_err;
1283 frame->tf_r10 = frame->tf_rcx;
1288 panic("Unexpected EASYNC return value (for now)");
1291 if (p->p_sysent->sv_errsize) {
1292 if (error >= p->p_sysent->sv_errsize)
1293 error = -1; /* XXX */
1295 error = p->p_sysent->sv_errtbl[error];
1297 frame->tf_rax = error;
1298 frame->tf_rflags |= PSL_C;
1303 * Traced syscall. trapsignal() is not MP aware.
1305 if (orig_tf_rflags & PSL_T) {
1306 MAKEMPSAFE(have_mplock);
1307 frame->tf_rflags &= ~PSL_T;
1308 trapsignal(lp, SIGTRAP, 0);
1312 * Handle reschedule and other end-of-syscall issues
1314 userret(lp, frame, sticks);
1317 if (KTRPOINT(td, KTR_SYSRET)) {
1318 MAKEMPSAFE(have_mplock);
1319 ktrsysret(lp, code, error, args.sysmsg_result);
1324 * This works because errno is findable through the
1325 * register set. If we ever support an emulation where this
1326 * is not the case, this code will need to be revisited.
1328 STOPEVENT(p, S_SCX, code);
1333 * Release the MP lock if we had to get it
1338 KTR_LOG(kernentry_syscall_ret, lp->lwp_proc->p_pid, lp->lwp_tid, error);
1340 KASSERT(&td->td_toks_base == td->td_toks_stop,
1341 ("syscall: critical section count mismatch! %d/%d",
1342 crit_count, td->td_pri));
1343 KASSERT(curstop == td->td_toks_stop,
1344 ("syscall: extra tokens held after trap! %ld",
1345 td->td_toks_stop - &td->td_toks_base));
1350 * NOTE: mplock not held at any point
1353 fork_return(struct lwp *lp, struct trapframe *frame)
1355 frame->tf_rax = 0; /* Child returns zero */
1356 frame->tf_rflags &= ~PSL_C; /* success */
1359 generic_lwp_return(lp, frame);
1360 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1364 * Simplified back end of syscall(), used when returning from fork()
1365 * directly into user mode.
1367 * This code will return back into the fork trampoline code which then
1370 * NOTE: The mplock is not held at any point.
1373 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1375 struct proc *p = lp->lwp_proc;
1378 * Newly forked processes are given a kernel priority. We have to
1379 * adjust the priority to a normal user priority and fake entry
1380 * into the kernel (call userenter()) to install a passive release
1381 * function just in case userret() decides to stop the process. This
1382 * can occur when ^Z races a fork. If we do not install the passive
1383 * release function the current process designation will not be
1384 * released when the thread goes to sleep.
1386 lwkt_setpri_self(TDPRI_USER_NORM);
1387 userenter(lp->lwp_thread, p);
1388 userret(lp, frame, 0);
1390 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1391 ktrsysret(lp, SYS_fork, 0, 0);
1393 p->p_flag |= P_PASSIVE_ACQ;
1395 p->p_flag &= ~P_PASSIVE_ACQ;
1399 * doreti has turned into this. The frame is directly on the stack. We
1400 * pull everything else we need (fpu and tls context) from the current
1403 * Note on fpu interactions: In a virtual kernel, the fpu context for
1404 * an emulated user mode process is not shared with the virtual kernel's
1405 * fpu context, so we only have to 'stack' fpu contexts within the virtual
1406 * kernel itself, and not even then since the signal() contexts that we care
1407 * about save and restore the FPU state (I think anyhow).
1409 * vmspace_ctl() returns an error only if it had problems instaling the
1410 * context we supplied or problems copying data to/from our VM space.
1413 go_user(struct intrframe *frame)
1415 struct trapframe *tf = (void *)&frame->if_rdi;
1419 * Interrupts may be disabled on entry, make sure all signals
1420 * can be received before beginning our loop.
1425 * Switch to the current simulated user process, then call
1426 * user_trap() when we break out of it (usually due to a signal).
1430 * Tell the real kernel whether it is ok to use the FP
1433 if (mdcpu->gd_npxthread == curthread) {
1434 tf->tf_xflags &= ~PGEX_FPFAULT;
1436 tf->tf_xflags |= PGEX_FPFAULT;
1440 * Run emulated user process context. This call interlocks
1441 * with new mailbox signals.
1443 * Set PGEX_U unconditionally, indicating a user frame (the
1444 * bit is normally set only by T_PAGEFLT).
1446 r = vmspace_ctl(&curproc->p_vmspace->vm_pmap, VMSPACE_CTL_RUN,
1447 tf, &curthread->td_savevext);
1448 frame->if_xflags |= PGEX_U;
1450 kprintf("GO USER %d trap %ld EVA %08lx RIP %08lx RSP %08lx XFLAGS %02lx/%02lx\n",
1451 r, tf->tf_trapno, tf->tf_addr, tf->tf_rip, tf->tf_rsp,
1452 tf->tf_xflags, frame->if_xflags);
1456 panic("vmspace_ctl failed error %d", errno);
1458 if (tf->tf_trapno) {
1462 if (mycpu->gd_reqflags & RQF_AST_MASK) {
1463 tf->tf_trapno = T_ASTFLT;
1471 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1472 * fault (which is then passed back to the virtual kernel) if an attempt is
1473 * made to use the FP unit.
1475 * XXX this is a fairly big hack.
1478 set_vkernel_fp(struct trapframe *frame)
1480 struct thread *td = curthread;
1482 if (frame->tf_xflags & PGEX_FPFAULT) {
1483 td->td_pcb->pcb_flags |= FP_VIRTFP;
1484 if (mdcpu->gd_npxthread == td)
1487 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1492 * Called from vkernel_trap() to fixup the vkernel's syscall
1493 * frame for vmspace_ctl() return.
1496 cpu_vkernel_trap(struct trapframe *frame, int error)
1498 frame->tf_rax = error;
1500 frame->tf_rflags |= PSL_C;
1502 frame->tf_rflags &= ~PSL_C;