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 static int syscall_mpsafe = 1;
173 SYSCTL_INT(_kern, OID_AUTO, syscall_mpsafe, CTLFLAG_RW,
174 &syscall_mpsafe, 0, "Allow MPSAFE marked syscalls to run without BGL");
175 TUNABLE_INT("kern.syscall_mpsafe", &syscall_mpsafe);
176 static int trap_mpsafe = 1;
177 SYSCTL_INT(_kern, OID_AUTO, trap_mpsafe, CTLFLAG_RW,
178 &trap_mpsafe, 0, "Allow traps to mostly run without the BGL");
179 TUNABLE_INT("kern.trap_mpsafe", &trap_mpsafe);
182 MALLOC_DEFINE(M_SYSMSG, "sysmsg", "sysmsg structure");
183 extern int max_sysmsg;
186 * Passively intercepts the thread switch function to increase the thread
187 * priority from a user priority to a kernel priority, reducing
188 * syscall and trap overhead for the case where no switch occurs.
190 * Synchronizes td_ucred with p_ucred. This is used by system calls,
191 * signal handling, faults, AST traps, and anything else that enters the
192 * kernel from userland and provides the kernel with a stable read-only
193 * copy of the process ucred.
196 userenter(struct thread *curtd, struct proc *curp)
201 curtd->td_release = lwkt_passive_release;
203 if (curtd->td_ucred != curp->p_ucred) {
204 ncred = crhold(curp->p_ucred);
205 ocred = curtd->td_ucred;
206 curtd->td_ucred = ncred;
213 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
214 * must be completed before we can return to or try to return to userland.
216 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
217 * arithmatic on the delta calculation so the absolute tick values are
218 * truncated to an integer.
221 userret(struct lwp *lp, struct trapframe *frame, int sticks)
223 struct proc *p = lp->lwp_proc;
227 * Charge system time if profiling. Note: times are in microseconds.
228 * This may do a copyout and block, so do it first even though it
229 * means some system time will be charged as user time.
231 if (p->p_flag & P_PROFIL) {
232 addupc_task(p, frame->tf_rip,
233 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
238 * If the jungle wants us dead, so be it.
240 if (lp->lwp_flag & LWP_WEXIT) {
243 rel_mplock(); /* NOT REACHED */
247 * Block here if we are in a stopped state.
249 if (p->p_stat == SSTOP) {
257 * Post any pending upcalls
259 if (p->p_flag & P_UPCALLPEND) {
261 p->p_flag &= ~P_UPCALLPEND;
268 * Post any pending signals
270 if ((sig = CURSIG_TRACE(lp)) != 0) {
278 * block here if we are swapped out, but still process signals
279 * (such as SIGKILL). proc0 (the swapin scheduler) is already
280 * aware of our situation, we do not have to wake it up.
282 if (p->p_flag & P_SWAPPEDOUT) {
284 p->p_flag |= P_SWAPWAIT;
286 if (p->p_flag & P_SWAPWAIT)
287 tsleep(p, PCATCH, "SWOUT", 0);
288 p->p_flag &= ~P_SWAPWAIT;
294 * Make sure postsig() handled request to restore old signal mask after
295 * running signal handler.
297 KKASSERT((lp->lwp_flag & LWP_OLDMASK) == 0);
301 * Cleanup from userenter and any passive release that might have occured.
302 * We must reclaim the current-process designation before we can return
303 * to usermode. We also handle both LWKT and USER reschedule requests.
306 userexit(struct lwp *lp)
308 struct thread *td = lp->lwp_thread;
309 /* globaldata_t gd = td->td_gd; */
312 * Handle stop requests at kernel priority. Any requests queued
313 * after this loop will generate another AST.
315 while (lp->lwp_proc->p_stat == SSTOP) {
322 * Reduce our priority in preparation for a return to userland. If
323 * our passive release function was still in place, our priority was
324 * never raised and does not need to be reduced.
326 lwkt_passive_recover(td);
329 * Become the current user scheduled process if we aren't already,
330 * and deal with reschedule requests and other factors.
332 lp->lwp_proc->p_usched->acquire_curproc(lp);
333 /* WARNING: we may have migrated cpu's */
334 /* gd = td->td_gd; */
337 #if !defined(KTR_KERNENTRY)
338 #define KTR_KERNENTRY KTR_ALL
340 KTR_INFO_MASTER(kernentry);
341 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, "pid=%d, tid=%d, trapno=%d, eva=%p",
342 sizeof(int) + sizeof(int) + sizeof(int) + sizeof(vm_offset_t));
343 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "pid=%d, tid=%d",
344 sizeof(int) + sizeof(int));
345 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "pid=%d, tid=%d, call=%d",
346 sizeof(int) + sizeof(int) + sizeof(int));
347 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "pid=%d, tid=%d, err=%d",
348 sizeof(int) + sizeof(int) + sizeof(int));
349 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "pid=%d, tid=%d",
350 sizeof(int) + sizeof(int));
353 * Exception, fault, and trap interface to the kernel.
354 * This common code is called from assembly language IDT gate entry
355 * routines that prepare a suitable stack frame, and restore this
356 * frame after the exception has been processed.
358 * This function is also called from doreti in an interlock to handle ASTs.
359 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
361 * NOTE! We have to retrieve the fault address prior to obtaining the
362 * MP lock because get_mplock() may switch out. YYY cr2 really ought
363 * to be retrieved by the assembly code, not here.
365 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
366 * if an attempt is made to switch from a fast interrupt or IPI. This is
367 * necessary to properly take fatal kernel traps on SMP machines if
368 * get_mplock() has to block.
372 user_trap(struct trapframe *frame)
374 struct globaldata *gd = mycpu;
375 struct thread *td = gd->gd_curthread;
376 struct lwp *lp = td->td_lwp;
379 int i = 0, ucode = 0, type, code;
384 int crit_count = td->td_pri & ~TDPRI_MASK;
390 if (frame->tf_trapno == T_PAGEFLT)
391 eva = frame->tf_addr;
395 kprintf("USER_TRAP AT %08lx xflags %ld trapno %ld eva %08lx\n",
396 frame->tf_rip, frame->tf_xflags, frame->tf_trapno, eva);
400 * Everything coming from user mode runs through user_trap,
401 * including system calls.
403 if (frame->tf_trapno == T_FAST_SYSCALL) {
408 KTR_LOG(kernentry_trap, lp->lwp_proc->p_pid, lp->lwp_tid,
409 frame->tf_trapno, eva);
413 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
414 ++gd->gd_trap_nesting_level;
415 MAKEMPSAFE(have_mplock);
416 trap_fatal(frame, TRUE, eva);
417 --gd->gd_trap_nesting_level;
422 ++gd->gd_trap_nesting_level;
424 if (trap_mpsafe == 0)
425 MAKEMPSAFE(have_mplock);
428 --gd->gd_trap_nesting_level;
430 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
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;
453 case T_ARITHTRAP: /* arithmetic trap */
458 case T_ASTFLT: /* Allow process switch */
459 mycpu->gd_cnt.v_soft++;
460 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
461 atomic_clear_int_nonlocked(&mycpu->gd_reqflags,
463 addupc_task(p, p->p_prof.pr_addr,
469 * The following two traps can happen in
470 * vm86 mode, and, if so, we want to handle
473 case T_PROTFLT: /* general protection fault */
474 case T_STKFLT: /* stack fault */
476 if (frame->tf_eflags & PSL_VM) {
477 i = vm86_emulate((struct vm86frame *)frame);
485 case T_SEGNPFLT: /* segment not present fault */
486 case T_TSSFLT: /* invalid TSS fault */
487 case T_DOUBLEFLT: /* double fault */
489 ucode = code + BUS_SEGM_FAULT ;
493 case T_PAGEFLT: /* page fault */
494 MAKEMPSAFE(have_mplock);
495 i = trap_pfault(frame, TRUE, eva);
498 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
508 case T_DIVIDE: /* integer divide fault */
515 MAKEMPSAFE(have_mplock);
516 /* machine/parity/power fail/"kitchen sink" faults */
517 if (isa_nmi(code) == 0) {
520 * NMI can be hooked up to a pushbutton
524 kprintf ("NMI ... going to debugger\n");
525 kdb_trap (type, 0, frame);
529 } else if (panic_on_nmi)
530 panic("NMI indicates hardware failure");
532 #endif /* NISA > 0 */
534 case T_OFLOW: /* integer overflow fault */
539 case T_BOUND: /* bounds check fault */
546 * Virtual kernel intercept - pass the DNA exception
547 * to the (emulated) virtual kernel if it asked to handle
548 * it. This occurs when the virtual kernel is holding
549 * onto the FP context for a different emulated
550 * process then the one currently running.
552 * We must still call npxdna() since we may have
553 * saved FP state that the (emulated) virtual kernel
554 * needs to hand over to a different emulated process.
556 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
557 (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);
627 KASSERT(td->td_mpcount == have_mplock, ("badmpcount trap/end from %p", (void *)frame->tf_rip));
629 userret(lp, frame, sticks);
636 KTR_LOG(kernentry_trap_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
638 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
639 ("syscall: critical section count mismatch! %d/%d",
640 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
645 kern_trap(struct trapframe *frame)
647 struct globaldata *gd = mycpu;
648 struct thread *td = gd->gd_curthread;
651 int i = 0, ucode = 0, type, code;
656 int crit_count = td->td_pri & ~TDPRI_MASK;
663 if (frame->tf_trapno == T_PAGEFLT)
664 eva = frame->tf_addr;
670 ++gd->gd_trap_nesting_level;
671 MAKEMPSAFE(have_mplock);
672 trap_fatal(frame, FALSE, eva);
673 --gd->gd_trap_nesting_level;
678 ++gd->gd_trap_nesting_level;
681 if (trap_mpsafe == 0)
682 MAKEMPSAFE(have_mplock);
685 --gd->gd_trap_nesting_level;
687 type = frame->tf_trapno;
688 code = frame->tf_err;
696 case T_PAGEFLT: /* page fault */
697 MAKEMPSAFE(have_mplock);
698 trap_pfault(frame, FALSE, eva);
703 * The kernel may be using npx for copying or other
706 panic("kernel NPX should not happen");
711 case T_PROTFLT: /* general protection fault */
712 case T_SEGNPFLT: /* segment not present fault */
714 * Invalid segment selectors and out of bounds
715 * %eip's and %esp's can be set up in user mode.
716 * This causes a fault in kernel mode when the
717 * kernel tries to return to user mode. We want
718 * to get this fault so that we can fix the
719 * problem here and not have to check all the
720 * selectors and pointers when the user changes
723 if (mycpu->gd_intr_nesting_level == 0) {
724 if (td->td_pcb->pcb_onfault) {
726 (register_t)td->td_pcb->pcb_onfault;
734 * PSL_NT can be set in user mode and isn't cleared
735 * automatically when the kernel is entered. This
736 * causes a TSS fault when the kernel attempts to
737 * `iret' because the TSS link is uninitialized. We
738 * want to get this fault so that we can fix the
739 * problem here and not every time the kernel is
742 if (frame->tf_rflags & PSL_NT) {
743 frame->tf_rflags &= ~PSL_NT;
748 case T_TRCTRAP: /* trace trap */
750 if (frame->tf_eip == (int)IDTVEC(syscall)) {
752 * We've just entered system mode via the
753 * syscall lcall. Continue single stepping
754 * silently until the syscall handler has
759 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
761 * The syscall handler has now saved the
762 * flags. Stop single stepping it.
764 frame->tf_eflags &= ~PSL_T;
770 * Ignore debug register trace traps due to
771 * accesses in the user's address space, which
772 * can happen under several conditions such as
773 * if a user sets a watchpoint on a buffer and
774 * then passes that buffer to a system call.
775 * We still want to get TRCTRAPS for addresses
776 * in kernel space because that is useful when
777 * debugging the kernel.
779 if (user_dbreg_trap()) {
781 * Reset breakpoint bits because the
784 load_dr6(rdr6() & 0xfffffff0);
789 * Fall through (TRCTRAP kernel mode, kernel address)
793 * If DDB is enabled, let it handle the debugger trap.
794 * Otherwise, debugger traps "can't happen".
797 MAKEMPSAFE(have_mplock);
798 if (kdb_trap (type, 0, frame))
803 MAKEMPSAFE(have_mplock);
804 trap_fatal(frame, FALSE, eva);
807 MAKEMPSAFE(have_mplock);
808 trap_fatal(frame, FALSE, eva);
813 * Ignore this trap generated from a spurious SIGTRAP.
815 * single stepping in / syscalls leads to spurious / SIGTRAP
818 * Haiku (c) 2007 Simon 'corecode' Schubert
824 * Translate fault for emulators (e.g. Linux)
826 if (*p->p_sysent->sv_transtrap)
827 i = (*p->p_sysent->sv_transtrap)(i, type);
829 MAKEMPSAFE(have_mplock);
830 trapsignal(lp, i, ucode);
833 if (type <= MAX_TRAP_MSG) {
834 uprintf("fatal process exception: %s",
836 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
837 uprintf(", fault VA = 0x%lx", (u_long)eva);
849 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
850 ("syscall: critical section count mismatch! %d/%d",
851 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
856 trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
859 struct vmspace *vm = NULL;
863 thread_t td = curthread;
864 struct lwp *lp = td->td_lwp;
866 va = trunc_page(eva);
867 if (usermode == FALSE) {
869 * This is a fault on kernel virtual memory.
874 * This is a fault on non-kernel virtual memory.
875 * vm is initialized above to NULL. If curproc is NULL
876 * or curproc->p_vmspace is NULL the fault is fatal.
879 vm = lp->lwp_vmspace;
887 if (frame->tf_err & PGEX_W)
888 ftype = VM_PROT_READ | VM_PROT_WRITE;
890 ftype = VM_PROT_READ;
892 if (map != &kernel_map) {
894 * Keep swapout from messing with us during this
900 * Grow the stack if necessary
902 /* grow_stack returns false only if va falls into
903 * a growable stack region and the stack growth
904 * fails. It returns true if va was not within
905 * a growable stack region, or if the stack
908 if (!grow_stack (lp->lwp_proc, va)) {
914 /* Fault in the user page: */
915 rv = vm_fault(map, va, ftype,
916 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
922 * Don't have to worry about process locking or stacks in the kernel.
924 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
927 if (rv == KERN_SUCCESS)
931 if (td->td_gd->gd_intr_nesting_level == 0 &&
932 td->td_pcb->pcb_onfault) {
933 frame->tf_rip = (register_t)td->td_pcb->pcb_onfault;
936 trap_fatal(frame, usermode, eva);
941 * NOTE: on x86_64 we have a tf_addr field in the trapframe, no
942 * kludge is needed to pass the fault address to signal handlers.
944 struct proc *p = td->td_proc;
945 kprintf("seg-fault accessing address %p rip=%p pid=%d p_comm=%s\n",
946 (void *)va, (void *)frame->tf_rip, p->p_pid, p->p_comm);
947 /* Debugger("seg-fault"); */
949 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
953 trap_fatal(struct trapframe *frame, int usermode, vm_offset_t eva)
958 code = frame->tf_xflags;
959 type = frame->tf_trapno;
961 if (type <= MAX_TRAP_MSG) {
962 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
963 type, trap_msg[type],
964 (usermode ? "user" : "kernel"));
967 /* two separate prints in case of a trap on an unmapped page */
968 kprintf("mp_lock = %08x; ", mp_lock);
969 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
971 if (type == T_PAGEFLT) {
972 kprintf("fault virtual address = %p\n", (void *)eva);
973 kprintf("fault code = %s %s, %s\n",
974 usermode ? "user" : "supervisor",
975 code & PGEX_W ? "write" : "read",
976 code & PGEX_P ? "protection violation" : "page not present");
978 kprintf("instruction pointer = 0x%lx:0x%lx\n",
979 frame->tf_cs & 0xffff, frame->tf_rip);
981 ss = frame->tf_ss & 0xffff;
984 ss = GSEL(GDATA_SEL, SEL_KPL);
985 rsp = (long)&frame->tf_rsp;
987 kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp);
988 kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp);
989 kprintf("processor eflags = ");
990 if (frame->tf_rflags & PSL_T)
991 kprintf("trace trap, ");
992 if (frame->tf_rflags & PSL_I)
993 kprintf("interrupt enabled, ");
994 if (frame->tf_rflags & PSL_NT)
995 kprintf("nested task, ");
996 if (frame->tf_rflags & PSL_RF)
999 if (frame->tf_eflags & PSL_VM)
1002 kprintf("IOPL = %jd\n", (intmax_t)((frame->tf_rflags & PSL_IOPL) >> 12));
1003 kprintf("current process = ");
1005 kprintf("%lu (%s)\n",
1006 (u_long)curproc->p_pid, curproc->p_comm ?
1007 curproc->p_comm : "");
1011 kprintf("current thread = pri %d ", curthread->td_pri);
1012 if (curthread->td_pri >= TDPRI_CRIT)
1018 * we probably SHOULD have stopped the other CPUs before now!
1019 * another CPU COULD have been touching cpl at this moment...
1021 kprintf(" <- SMP: XXX");
1030 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1033 kprintf("trap number = %d\n", type);
1034 if (type <= MAX_TRAP_MSG)
1035 panic("%s", trap_msg[type]);
1037 panic("unknown/reserved trap");
1041 * Double fault handler. Called when a fault occurs while writing
1042 * a frame for a trap/exception onto the stack. This usually occurs
1043 * when the stack overflows (such is the case with infinite recursion,
1046 * XXX Note that the current PTD gets replaced by IdlePTD when the
1047 * task switch occurs. This means that the stack that was active at
1048 * the time of the double fault is not available at <kstack> unless
1049 * the machine was idle when the double fault occurred. The downside
1050 * of this is that "trace <ebp>" in ddb won't work.
1053 dblfault_handler(void)
1056 struct mdglobaldata *gd = mdcpu;
1059 kprintf("\nFatal double fault:\n");
1061 kprintf("rip = 0x%lx\n", gd->gd_common_tss.tss_rip);
1062 kprintf("rsp = 0x%lx\n", gd->gd_common_tss.tss_rsp);
1063 kprintf("rbp = 0x%lx\n", gd->gd_common_tss.tss_rbp);
1066 /* two separate prints in case of a trap on an unmapped page */
1067 kprintf("mp_lock = %08x; ", mp_lock);
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_pri & ~TDPRI_MASK;
1147 int have_mplock = 0;
1152 union sysunion args;
1153 register_t *argsdst;
1155 mycpu->gd_cnt.v_syscall++;
1157 KTR_LOG(kernentry_syscall, lp->lwp_proc->p_pid, lp->lwp_tid,
1161 KASSERT(td->td_mpcount == 0, ("badmpcount syscall2 from %p", (void *)frame->tf_rip));
1162 if (syscall_mpsafe == 0)
1163 MAKEMPSAFE(have_mplock);
1165 userenter(td, p); /* lazy raise our priority */
1172 sticks = (int)td->td_sticks;
1173 orig_tf_rflags = frame->tf_rflags;
1176 * Virtual kernel intercept - if a VM context managed by a virtual
1177 * kernel issues a system call the virtual kernel handles it, not us.
1178 * Restore the virtual kernel context and return from its system
1179 * call. The current frame is copied out to the virtual kernel.
1181 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1182 vkernel_trap(lp, frame);
1183 error = EJUSTRETURN;
1188 * Get the system call parameters and account for time
1190 lp->lwp_md.md_regs = frame;
1191 params = (caddr_t)frame->tf_rsp + sizeof(register_t);
1192 code = frame->tf_rax;
1194 if (p->p_sysent->sv_prepsyscall) {
1195 (*p->p_sysent->sv_prepsyscall)(
1196 frame, (int *)(&args.nosys.sysmsg + 1),
1199 if (code == SYS_syscall || code == SYS___syscall) {
1200 code = frame->tf_rdi;
1206 if (p->p_sysent->sv_mask)
1207 code &= p->p_sysent->sv_mask;
1209 if (code >= p->p_sysent->sv_size)
1210 callp = &p->p_sysent->sv_table[0];
1212 callp = &p->p_sysent->sv_table[code];
1214 narg = callp->sy_narg & SYF_ARGMASK;
1217 * On x86_64 we get up to six arguments in registers. The rest are
1218 * on the stack. The first six members of 'struct trapframe' happen
1219 * to be the registers used to pass arguments, in exactly the right
1222 argp = &frame->tf_rdi;
1224 argsdst = (register_t *)(&args.nosys.sysmsg + 1);
1226 * JG can we overflow the space pointed to by 'argsdst'
1227 * either with 'bcopy' or with 'copyin'?
1229 bcopy(argp, argsdst, sizeof(register_t) * regcnt);
1231 * copyin is MP aware, but the tracing code is not
1233 if (narg > regcnt) {
1234 KASSERT(params != NULL, ("copyin args with no params!"));
1235 error = copyin(params, &argsdst[regcnt],
1236 (narg - regcnt) * sizeof(register_t));
1239 if (KTRPOINT(td, KTR_SYSCALL)) {
1240 MAKEMPSAFE(have_mplock);
1242 ktrsyscall(lp, code, narg,
1243 (void *)(&args.nosys.sysmsg + 1));
1251 if (KTRPOINT(td, KTR_SYSCALL)) {
1252 MAKEMPSAFE(have_mplock);
1253 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1258 * Default return value is 0 (will be copied to %rax). Double-value
1259 * returns use %rax and %rdx. %rdx is left unchanged for system
1260 * calls which return only one result.
1262 args.sysmsg_fds[0] = 0;
1263 args.sysmsg_fds[1] = frame->tf_rdx;
1266 * The syscall might manipulate the trap frame. If it does it
1267 * will probably return EJUSTRETURN.
1269 args.sysmsg_frame = frame;
1271 STOPEVENT(p, S_SCE, narg); /* MP aware */
1274 * NOTE: All system calls run MPSAFE now. The system call itself
1275 * is responsible for getting the MP lock.
1277 error = (*callp->sy_call)(&args);
1280 kprintf("system call %d returned %d\n", code, error);
1285 * MP SAFE (we may or may not have the MP lock at this point)
1290 * Reinitialize proc pointer `p' as it may be different
1291 * if this is a child returning from fork syscall.
1294 lp = curthread->td_lwp;
1295 frame->tf_rax = args.sysmsg_fds[0];
1296 frame->tf_rdx = args.sysmsg_fds[1];
1297 frame->tf_rflags &= ~PSL_C;
1301 * Reconstruct pc, we know that 'syscall' is 2 bytes.
1302 * We have to do a full context restore so that %r10
1303 * (which was holding the value of %rcx) is restored for
1304 * the next iteration.
1306 frame->tf_rip -= frame->tf_err;
1307 frame->tf_r10 = frame->tf_rcx;
1312 panic("Unexpected EASYNC return value (for now)");
1315 if (p->p_sysent->sv_errsize) {
1316 if (error >= p->p_sysent->sv_errsize)
1317 error = -1; /* XXX */
1319 error = p->p_sysent->sv_errtbl[error];
1321 frame->tf_rax = error;
1322 frame->tf_rflags |= PSL_C;
1327 * Traced syscall. trapsignal() is not MP aware.
1329 if (orig_tf_rflags & PSL_T) {
1330 MAKEMPSAFE(have_mplock);
1331 frame->tf_rflags &= ~PSL_T;
1332 trapsignal(lp, SIGTRAP, 0);
1336 * Handle reschedule and other end-of-syscall issues
1338 userret(lp, frame, sticks);
1341 if (KTRPOINT(td, KTR_SYSRET)) {
1342 MAKEMPSAFE(have_mplock);
1343 ktrsysret(lp, code, error, args.sysmsg_result);
1348 * This works because errno is findable through the
1349 * register set. If we ever support an emulation where this
1350 * is not the case, this code will need to be revisited.
1352 STOPEVENT(p, S_SCX, code);
1357 * Release the MP lock if we had to get it
1359 KASSERT(td->td_mpcount == have_mplock,
1360 ("badmpcount syscall2/end from %p", (void *)frame->tf_rip));
1364 KTR_LOG(kernentry_syscall_ret, lp->lwp_proc->p_pid, lp->lwp_tid, error);
1366 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
1367 ("syscall: critical section count mismatch! %d/%d",
1368 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
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 * or lwp_create() directly into user mode. MP lock is held on entry and
1386 * should be released on return. This code will return back into the fork
1387 * trampoline code which then runs doreti.
1390 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1392 struct proc *p = lp->lwp_proc;
1395 * Newly forked processes are given a kernel priority. We have to
1396 * adjust the priority to a normal user priority and fake entry
1397 * into the kernel (call userenter()) to install a passive release
1398 * function just in case userret() decides to stop the process. This
1399 * can occur when ^Z races a fork. If we do not install the passive
1400 * release function the current process designation will not be
1401 * released when the thread goes to sleep.
1403 lwkt_setpri_self(TDPRI_USER_NORM);
1404 userenter(lp->lwp_thread, p);
1405 userret(lp, frame, 0);
1407 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1408 ktrsysret(lp, SYS_fork, 0, 0);
1410 p->p_flag |= P_PASSIVE_ACQ;
1412 p->p_flag &= ~P_PASSIVE_ACQ;
1414 KKASSERT(lp->lwp_thread->td_mpcount == 1);
1420 * doreti has turned into this. The frame is directly on the stack. We
1421 * pull everything else we need (fpu and tls context) from the current
1424 * Note on fpu interactions: In a virtual kernel, the fpu context for
1425 * an emulated user mode process is not shared with the virtual kernel's
1426 * fpu context, so we only have to 'stack' fpu contexts within the virtual
1427 * kernel itself, and not even then since the signal() contexts that we care
1428 * about save and restore the FPU state (I think anyhow).
1430 * vmspace_ctl() returns an error only if it had problems instaling the
1431 * context we supplied or problems copying data to/from our VM space.
1434 go_user(struct intrframe *frame)
1436 struct trapframe *tf = (void *)&frame->if_rdi;
1440 * Interrupts may be disabled on entry, make sure all signals
1441 * can be received before beginning our loop.
1446 * Switch to the current simulated user process, then call
1447 * user_trap() when we break out of it (usually due to a signal).
1451 * Tell the real kernel whether it is ok to use the FP
1454 if (mdcpu->gd_npxthread == curthread) {
1455 tf->tf_xflags &= ~PGEX_FPFAULT;
1457 tf->tf_xflags |= PGEX_FPFAULT;
1461 * Run emulated user process context. This call interlocks
1462 * with new mailbox signals.
1464 * Set PGEX_U unconditionally, indicating a user frame (the
1465 * bit is normally set only by T_PAGEFLT).
1467 r = vmspace_ctl(&curproc->p_vmspace->vm_pmap, VMSPACE_CTL_RUN,
1468 tf, &curthread->td_savevext);
1469 frame->if_xflags |= PGEX_U;
1471 kprintf("GO USER %d trap %ld EVA %08lx RIP %08lx RSP %08lx XFLAGS %02lx/%02lx\n",
1472 r, tf->tf_trapno, tf->tf_addr, tf->tf_rip, tf->tf_rsp,
1473 tf->tf_xflags, frame->if_xflags);
1477 panic("vmspace_ctl failed error %d", errno);
1479 if (tf->tf_trapno) {
1483 if (mycpu->gd_reqflags & RQF_AST_MASK) {
1484 tf->tf_trapno = T_ASTFLT;
1492 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1493 * fault (which is then passed back to the virtual kernel) if an attempt is
1494 * made to use the FP unit.
1496 * XXX this is a fairly big hack.
1499 set_vkernel_fp(struct trapframe *frame)
1501 struct thread *td = curthread;
1503 if (frame->tf_xflags & PGEX_FPFAULT) {
1504 td->td_pcb->pcb_flags |= FP_VIRTFP;
1505 if (mdcpu->gd_npxthread == td)
1508 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1513 * Called from vkernel_trap() to fixup the vkernel's syscall
1514 * frame for vmspace_ctl() return.
1517 cpu_vkernel_trap(struct trapframe *frame, int error)
1519 frame->tf_rax = error;
1521 frame->tf_rflags |= PSL_C;
1523 frame->tf_rflags &= ~PSL_C;