2 * Copyright (c) 1990, 1993
3 * The Regents of the University of California. All rights reserved.
4 * Copyright (C) 1994, David Greenman
5 * Copyright (c) 2008 The DragonFly Project.
6 * Copyright (c) 2008 Jordan Gordeev.
8 * This code is derived from software contributed to Berkeley by
9 * the University of Utah, and William Jolitz.
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. All advertising materials mentioning features or use of this software
20 * must display the following acknowledgement:
21 * This product includes software developed by the University of
22 * California, Berkeley and its contributors.
23 * 4. Neither the name of the University nor the names of its contributors
24 * may be used to endorse or promote products derived from this software
25 * without specific prior written permission.
27 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39 * from: @(#)trap.c 7.4 (Berkeley) 5/13/91
40 * $FreeBSD: src/sys/i386/i386/trap.c,v 1.147.2.11 2003/02/27 19:09:59 luoqi Exp $
44 * x86_64 Trap and System call handling
48 #include "opt_ktrace.h"
50 #include <machine/frame.h>
51 #include <sys/param.h>
52 #include <sys/systm.h>
53 #include <sys/kernel.h>
55 #include <sys/pioctl.h>
56 #include <sys/types.h>
57 #include <sys/signal2.h>
58 #include <sys/syscall.h>
59 #include <sys/sysctl.h>
60 #include <sys/sysent.h>
61 #include <sys/systm.h>
63 #include <sys/ktrace.h>
66 #include <sys/sysmsg.h>
67 #include <sys/sysproto.h>
68 #include <sys/sysunion.h>
72 #include <vm/vm_extern.h>
73 #include <vm/vm_kern.h>
74 #include <vm/vm_param.h>
75 #include <machine/cpu.h>
76 #include <machine/pcb.h>
77 #include <machine/smp.h>
78 #include <machine/thread.h>
79 #include <machine/vmparam.h>
80 #include <machine/md_var.h>
81 #include <machine_base/isa/intr_machdep.h>
84 #include <sys/thread2.h>
88 #define MAKEMPSAFE(have_mplock) \
89 if (have_mplock == 0) { \
96 #define MAKEMPSAFE(have_mplock)
100 extern void trap(struct trapframe *frame);
102 static int trap_pfault(struct trapframe *, int);
103 static void trap_fatal(struct trapframe *, vm_offset_t);
104 void dblfault_handler(struct trapframe *frame);
106 #define MAX_TRAP_MSG 30
107 static char *trap_msg[] = {
109 "privileged instruction fault", /* 1 T_PRIVINFLT */
111 "breakpoint instruction fault", /* 3 T_BPTFLT */
114 "arithmetic trap", /* 6 T_ARITHTRAP */
115 "system forced exception", /* 7 T_ASTFLT */
117 "general protection fault", /* 9 T_PROTFLT */
118 "trace trap", /* 10 T_TRCTRAP */
120 "page fault", /* 12 T_PAGEFLT */
122 "alignment fault", /* 14 T_ALIGNFLT */
126 "integer divide fault", /* 18 T_DIVIDE */
127 "non-maskable interrupt trap", /* 19 T_NMI */
128 "overflow trap", /* 20 T_OFLOW */
129 "FPU bounds check fault", /* 21 T_BOUND */
130 "FPU device not available", /* 22 T_DNA */
131 "double fault", /* 23 T_DOUBLEFLT */
132 "FPU operand fetch fault", /* 24 T_FPOPFLT */
133 "invalid TSS fault", /* 25 T_TSSFLT */
134 "segment not present fault", /* 26 T_SEGNPFLT */
135 "stack fault", /* 27 T_STKFLT */
136 "machine check trap", /* 28 T_MCHK */
137 "SIMD floating-point exception", /* 29 T_XMMFLT */
138 "reserved (unknown) fault", /* 30 T_RESERVED */
142 static int ddb_on_nmi = 1;
143 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
144 &ddb_on_nmi, 0, "Go to DDB on NMI");
146 static int panic_on_nmi = 1;
147 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
148 &panic_on_nmi, 0, "Panic on NMI");
149 static int fast_release;
150 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
151 &fast_release, 0, "Passive Release was optimal");
152 static int slow_release;
153 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
154 &slow_release, 0, "Passive Release was nonoptimal");
156 static int syscall_mpsafe = 1;
157 SYSCTL_INT(_kern, OID_AUTO, syscall_mpsafe, CTLFLAG_RW,
158 &syscall_mpsafe, 0, "Allow MPSAFE marked syscalls to run without BGL");
159 TUNABLE_INT("kern.syscall_mpsafe", &syscall_mpsafe);
160 static int trap_mpsafe = 1;
161 SYSCTL_INT(_kern, OID_AUTO, trap_mpsafe, CTLFLAG_RW,
162 &trap_mpsafe, 0, "Allow traps to mostly run without the BGL");
163 TUNABLE_INT("kern.trap_mpsafe", &trap_mpsafe);
167 * Passively intercepts the thread switch function to increase
168 * the thread priority from a user priority to a kernel priority, reducing
169 * syscall and trap overhead for the case where no switch occurs.
171 * Synchronizes lwp_ucred with p_ucred. This is used by system calls,
172 * signal handling, faults, AST traps, and anything else that enters the
173 * kernel from userland and provides the kernel with a stable read-only
174 * copy of the process ucred.
177 userenter(struct thread *curtd, struct lwp *curlp)
179 struct proc *p = curlp->lwp_proc;
183 curtd->td_release = lwkt_passive_release;
185 if (curlp->lwp_ucred != p->p_ucred) {
186 ncred = crhold(p->p_ucred);
187 ocred = curlp->lwp_ucred;
188 curlp->lwp_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_flag & 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_flag & LWP_WEXIT) {
225 rel_mplock(); /* NOT REACHED */
229 * Block here if we are in a stopped state.
231 if (p->p_stat == SSTOP) {
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_flag & P_UPCALLPEND) {
243 p->p_flag &= ~P_UPCALLPEND;
251 * Post any pending signals. If running a virtual kernel be sure
252 * to restore the virtual kernel's vmspace before posting the signal.
254 if ((sig = CURSIG_TRACE(lp)) != 0) {
262 * block here if we are swapped out, but still process signals
263 * (such as SIGKILL). proc0 (the swapin scheduler) is already
264 * aware of our situation, we do not have to wake it up.
266 if (p->p_flag & P_SWAPPEDOUT) {
268 p->p_flag |= P_SWAPWAIT;
270 if (p->p_flag & P_SWAPWAIT)
271 tsleep(p, PCATCH, "SWOUT", 0);
272 p->p_flag &= ~P_SWAPWAIT;
278 * Make sure postsig() handled request to restore old signal mask after
279 * running signal handler.
281 KKASSERT((lp->lwp_flag & LWP_OLDMASK) == 0);
285 * Cleanup from userenter and any passive release that might have occured.
286 * We must reclaim the current-process designation before we can return
287 * to usermode. We also handle both LWKT and USER reschedule requests.
290 userexit(struct lwp *lp)
292 struct thread *td = lp->lwp_thread;
293 /* globaldata_t gd = td->td_gd;*/
296 * Handle stop requests at kernel priority. Any requests queued
297 * after this loop will generate another AST.
299 while (lp->lwp_proc->p_stat == SSTOP) {
306 * Reduce our priority in preparation for a return to userland. If
307 * our passive release function was still in place, our priority was
308 * never raised and does not need to be reduced.
310 lwkt_passive_recover(td);
313 * Become the current user scheduled process if we aren't already,
314 * and deal with reschedule requests and other factors.
316 lp->lwp_proc->p_usched->acquire_curproc(lp);
317 /* WARNING: we may have migrated cpu's */
318 /* gd = td->td_gd; */
321 #if !defined(KTR_KERNENTRY)
322 #define KTR_KERNENTRY KTR_ALL
324 KTR_INFO_MASTER(kernentry);
325 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, "STR",
326 sizeof(long) + sizeof(long) + sizeof(long) + sizeof(vm_offset_t));
327 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "STR",
328 sizeof(long) + sizeof(long));
329 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "STR",
330 sizeof(long) + sizeof(long) + sizeof(long));
331 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "STR",
332 sizeof(long) + sizeof(long) + sizeof(long));
333 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "STR",
334 sizeof(long) + sizeof(long));
337 * Exception, fault, and trap interface to the kernel.
338 * This common code is called from assembly language IDT gate entry
339 * routines that prepare a suitable stack frame, and restore this
340 * frame after the exception has been processed.
342 * This function is also called from doreti in an interlock to handle ASTs.
343 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
345 * NOTE! We have to retrieve the fault address prior to obtaining the
346 * MP lock because get_mplock() may switch out. YYY cr2 really ought
347 * to be retrieved by the assembly code, not here.
349 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
350 * if an attempt is made to switch from a fast interrupt or IPI. This is
351 * necessary to properly take fatal kernel traps on SMP machines if
352 * get_mplock() has to block.
356 trap(struct trapframe *frame)
358 struct globaldata *gd = mycpu;
359 struct thread *td = gd->gd_curthread;
360 struct lwp *lp = td->td_lwp;
363 int i = 0, ucode = 0, type, code;
368 int crit_count = td->td_pri & ~TDPRI_MASK;
376 kprintf0("\"%s\" type=%ld\n",
377 trap_msg[frame->tf_trapno], frame->tf_trapno);
378 kprintf0(" rip=%lx rsp=%lx\n", frame->tf_rip, frame->tf_rsp);
379 kprintf0(" err=%lx addr=%lx\n", frame->tf_err, frame->tf_addr);
380 kprintf0(" cs=%lx ss=%lx rflags=%lx\n", (unsigned long)frame->tf_cs, (unsigned long)frame->tf_ss, frame->tf_rflags);
385 ++gd->gd_trap_nesting_level;
386 MAKEMPSAFE(have_mplock);
387 trap_fatal(frame, frame->tf_addr);
388 --gd->gd_trap_nesting_level;
394 eva = (frame->tf_trapno == T_PAGEFLT ? frame->tf_addr : 0);
395 ++gd->gd_trap_nesting_level;
396 MAKEMPSAFE(have_mplock);
397 trap_fatal(frame, eva);
398 --gd->gd_trap_nesting_level;
406 if (trap_mpsafe == 0) {
407 ++gd->gd_trap_nesting_level;
408 MAKEMPSAFE(have_mplock);
409 --gd->gd_trap_nesting_level;
413 if ((frame->tf_rflags & PSL_I) == 0) {
415 * Buggy application or kernel code has disabled interrupts
416 * and then trapped. Enabling interrupts now is wrong, but
417 * it is better than running with interrupts disabled until
418 * they are accidentally enabled later.
420 type = frame->tf_trapno;
421 if (ISPL(frame->tf_cs) == SEL_UPL) {
422 MAKEMPSAFE(have_mplock);
423 /* JG curproc can be NULL */
425 "pid %ld (%s): trap %d with interrupts disabled\n",
426 (long)curproc->p_pid, curproc->p_comm, type);
427 } else if (type != T_NMI && type != T_BPTFLT &&
430 * XXX not quite right, since this may be for a
431 * multiple fault in user mode.
433 MAKEMPSAFE(have_mplock);
434 kprintf("kernel trap %d with interrupts disabled\n",
440 type = frame->tf_trapno;
441 code = frame->tf_err;
443 if (ISPL(frame->tf_cs) == SEL_UPL) {
446 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid,
447 frame->tf_trapno, eva);
451 sticks = (int)td->td_sticks;
452 lp->lwp_md.md_regs = frame;
455 case T_PRIVINFLT: /* privileged instruction fault */
460 case T_BPTFLT: /* bpt instruction fault */
461 case T_TRCTRAP: /* trace trap */
462 frame->tf_rflags &= ~PSL_T;
466 case T_ARITHTRAP: /* arithmetic trap */
479 case T_ASTFLT: /* Allow process switch */
480 mycpu->gd_cnt.v_soft++;
481 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
482 atomic_clear_int_nonlocked(&mycpu->gd_reqflags,
484 addupc_task(p, p->p_prof.pr_addr,
489 case T_PROTFLT: /* general protection fault */
490 case T_SEGNPFLT: /* segment not present fault */
491 case T_TSSFLT: /* invalid TSS fault */
492 case T_DOUBLEFLT: /* double fault */
494 ucode = code + BUS_SEGM_FAULT ;
498 case T_PAGEFLT: /* page fault */
499 MAKEMPSAFE(have_mplock);
500 i = trap_pfault(frame, TRUE);
501 if (frame->tf_rip == 0)
502 kprintf("T_PAGEFLT: Warning %%rip == 0!\n");
511 case T_DIVIDE: /* integer divide fault */
517 MAKEMPSAFE(have_mplock);
518 /* machine/parity/power fail/"kitchen sink" faults */
519 if (isa_nmi(code) == 0) {
522 * NMI can be hooked up to a pushbutton
526 kprintf ("NMI ... going to debugger\n");
527 kdb_trap(type, 0, frame);
531 } else if (panic_on_nmi)
532 panic("NMI indicates hardware failure");
535 case T_OFLOW: /* integer overflow fault */
540 case T_BOUND: /* bounds check fault */
547 * Virtual kernel intercept - pass the DNA exception
548 * to the virtual kernel if it asked to handle it.
549 * This occurs when the virtual kernel is holding
550 * onto the FP context for a different emulated
551 * process then the one currently running.
553 * We must still call npxdna() since we may have
554 * saved FP state that the virtual kernel needs
555 * to hand over to a different emulated process.
557 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
558 (td->td_pcb->pcb_flags & FP_VIRTFP)
565 * The kernel may have switched out the FP unit's
566 * state, causing the user process to take a fault
567 * when it tries to use the FP unit. Restore the
573 ucode = FPE_FPU_NP_TRAP;
576 case T_FPOPFLT: /* FPU operand fetch fault */
581 case T_XMMFLT: /* SIMD floating-point exception */
590 case T_PAGEFLT: /* page fault */
591 MAKEMPSAFE(have_mplock);
592 trap_pfault(frame, FALSE);
597 * The kernel is apparently using fpu for copying.
598 * XXX this should be fatal unless the kernel has
599 * registered such use.
605 case T_STKFLT: /* stack fault */
608 case T_PROTFLT: /* general protection fault */
609 case T_SEGNPFLT: /* segment not present fault */
611 * Invalid segment selectors and out of bounds
612 * %rip's and %rsp's can be set up in user mode.
613 * This causes a fault in kernel mode when the
614 * kernel tries to return to user mode. We want
615 * to get this fault so that we can fix the
616 * problem here and not have to check all the
617 * selectors and pointers when the user changes
620 kprintf("trap.c line %d\n", __LINE__);
621 if (mycpu->gd_intr_nesting_level == 0) {
622 if (td->td_pcb->pcb_onfault) {
623 frame->tf_rip = (register_t)
624 td->td_pcb->pcb_onfault;
627 if (frame->tf_rip == (long)doreti_iret) {
628 frame->tf_rip = (long)doreti_iret_fault;
636 * PSL_NT can be set in user mode and isn't cleared
637 * automatically when the kernel is entered. This
638 * causes a TSS fault when the kernel attempts to
639 * `iret' because the TSS link is uninitialized. We
640 * want to get this fault so that we can fix the
641 * problem here and not every time the kernel is
644 if (frame->tf_rflags & PSL_NT) {
645 frame->tf_rflags &= ~PSL_NT;
650 case T_TRCTRAP: /* trace trap */
652 if (frame->tf_rip == (int)IDTVEC(syscall)) {
654 * We've just entered system mode via the
655 * syscall lcall. Continue single stepping
656 * silently until the syscall handler has
661 if (frame->tf_rip == (int)IDTVEC(syscall) + 1) {
663 * The syscall handler has now saved the
664 * flags. Stop single stepping it.
666 frame->tf_rflags &= ~PSL_T;
672 * Ignore debug register trace traps due to
673 * accesses in the user's address space, which
674 * can happen under several conditions such as
675 * if a user sets a watchpoint on a buffer and
676 * then passes that buffer to a system call.
677 * We still want to get TRCTRAPS for addresses
678 * in kernel space because that is useful when
679 * debugging the kernel.
682 if (user_dbreg_trap()) {
684 * Reset breakpoint bits because the
687 /* XXX check upper bits here */
688 load_dr6(rdr6() & 0xfffffff0);
693 * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
697 * If DDB is enabled, let it handle the debugger trap.
698 * Otherwise, debugger traps "can't happen".
701 MAKEMPSAFE(have_mplock);
702 if (kdb_trap(type, 0, frame))
708 MAKEMPSAFE(have_mplock);
709 /* machine/parity/power fail/"kitchen sink" faults */
711 if (isa_nmi(code) == 0) {
714 * NMI can be hooked up to a pushbutton
718 kprintf ("NMI ... going to debugger\n");
719 kdb_trap(type, 0, frame);
723 } else if (panic_on_nmi == 0)
726 #endif /* NISA > 0 */
728 MAKEMPSAFE(have_mplock);
729 trap_fatal(frame, 0);
734 * Virtual kernel intercept - if the fault is directly related to a
735 * VM context managed by a virtual kernel then let the virtual kernel
738 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
739 vkernel_trap(lp, frame);
744 * Virtual kernel intercept - if the fault is directly related to a
745 * VM context managed by a virtual kernel then let the virtual kernel
748 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
749 vkernel_trap(lp, frame);
754 * Translate fault for emulators (e.g. Linux)
756 if (*p->p_sysent->sv_transtrap)
757 i = (*p->p_sysent->sv_transtrap)(i, type);
759 MAKEMPSAFE(have_mplock);
760 trapsignal(lp, i, ucode);
763 if (type <= MAX_TRAP_MSG) {
764 uprintf("fatal process exception: %s",
766 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
767 uprintf(", fault VA = 0x%lx", frame->tf_addr);
774 if (ISPL(frame->tf_cs) == SEL_UPL)
775 KASSERT(td->td_mpcount == have_mplock, ("badmpcount trap/end from %p", (void *)frame->tf_rip));
777 userret(lp, frame, sticks);
784 if (p != NULL && lp != NULL)
785 KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid);
787 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
788 ("syscall: critical section count mismatch! %d/%d",
789 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
794 trap_pfault(struct trapframe *frame, int usermode)
797 struct vmspace *vm = NULL;
801 thread_t td = curthread;
802 struct lwp *lp = td->td_lwp;
804 va = trunc_page(frame->tf_addr);
805 if (va >= VM_MIN_KERNEL_ADDRESS) {
807 * Don't allow user-mode faults in kernel address space.
815 * This is a fault on non-kernel virtual memory.
816 * vm is initialized above to NULL. If curproc is NULL
817 * or curproc->p_vmspace is NULL the fault is fatal.
820 vm = lp->lwp_vmspace;
829 * PGEX_I is defined only if the execute disable bit capability is
830 * supported and enabled.
832 if (frame->tf_err & PGEX_W)
833 ftype = VM_PROT_WRITE;
835 else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
836 ftype = VM_PROT_EXECUTE;
839 ftype = VM_PROT_READ;
841 if (map != &kernel_map) {
843 * Keep swapout from messing with us during this
849 * Grow the stack if necessary
851 /* grow_stack returns false only if va falls into
852 * a growable stack region and the stack growth
853 * fails. It returns true if va was not within
854 * a growable stack region, or if the stack
857 if (!grow_stack(lp->lwp_proc, va)) {
863 /* Fault in the user page: */
864 rv = vm_fault(map, va, ftype,
865 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
871 * Don't have to worry about process locking or stacks
874 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
877 if (rv == KERN_SUCCESS)
881 if (td->td_gd->gd_intr_nesting_level == 0 &&
882 td->td_pcb->pcb_onfault) {
883 frame->tf_rip = (register_t)td->td_pcb->pcb_onfault;
886 trap_fatal(frame, frame->tf_addr);
891 * NOTE: on x86_64 we have a tf_addr field in the trapframe, no
892 * kludge is needed to pass the fault address to signal handlers.
894 struct proc *p = td->td_proc;
895 kprintf("seg-fault accessing address %p rip=%p pid=%d p_comm=%s\n",
896 (void *)va, (void *)frame->tf_rip, p->p_pid, p->p_comm);
897 /* Debugger("seg-fault"); */
899 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
903 trap_fatal(struct trapframe *frame, vm_offset_t eva)
908 struct soft_segment_descriptor softseg;
911 code = frame->tf_err;
912 type = frame->tf_trapno;
913 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)], &softseg);
915 if (type <= MAX_TRAP_MSG)
916 msg = trap_msg[type];
919 kprintf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
920 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
922 /* three separate prints in case of a trap on an unmapped page */
923 kprintf("mp_lock = %08x; ", mp_lock);
924 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
925 kprintf("lapic->id = %08x\n", lapic->id);
927 if (type == T_PAGEFLT) {
928 kprintf("fault virtual address = 0x%lx\n", eva);
929 kprintf("fault code = %s %s %s, %s\n",
930 code & PGEX_U ? "user" : "supervisor",
931 code & PGEX_W ? "write" : "read",
932 code & PGEX_I ? "instruction" : "data",
933 code & PGEX_P ? "protection violation" : "page not present");
935 kprintf("instruction pointer = 0x%lx:0x%lx\n",
936 frame->tf_cs & 0xffff, frame->tf_rip);
937 if (ISPL(frame->tf_cs) == SEL_UPL) {
938 ss = frame->tf_ss & 0xffff;
941 ss = GSEL(GDATA_SEL, SEL_KPL);
942 rsp = (long)&frame->tf_rsp;
944 kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp);
945 kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp);
946 kprintf("code segment = base 0x%lx, limit 0x%lx, type 0x%x\n",
947 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
948 kprintf(" = DPL %d, pres %d, long %d, def32 %d, gran %d\n",
949 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32,
951 kprintf("processor eflags = ");
952 if (frame->tf_rflags & PSL_T)
953 kprintf("trace trap, ");
954 if (frame->tf_rflags & PSL_I)
955 kprintf("interrupt enabled, ");
956 if (frame->tf_rflags & PSL_NT)
957 kprintf("nested task, ");
958 if (frame->tf_rflags & PSL_RF)
960 kprintf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12);
961 kprintf("current process = ");
964 (u_long)curproc->p_pid);
968 kprintf("current thread = pri %d ", curthread->td_pri);
969 if (curthread->td_pri >= TDPRI_CRIT)
974 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
977 kprintf("trap number = %d\n", type);
978 if (type <= MAX_TRAP_MSG)
979 panic("%s", trap_msg[type]);
981 panic("unknown/reserved trap");
985 * Double fault handler. Called when a fault occurs while writing
986 * a frame for a trap/exception onto the stack. This usually occurs
987 * when the stack overflows (such is the case with infinite recursion,
991 dblfault_handler(struct trapframe *frame)
993 kprintf0("DOUBLE FAULT\n");
994 kprintf("\nFatal double fault\n");
995 kprintf("rip = 0x%lx\n", frame->tf_rip);
996 kprintf("rsp = 0x%lx\n", frame->tf_rsp);
997 kprintf("rbp = 0x%lx\n", frame->tf_rbp);
999 /* three separate prints in case of a trap on an unmapped page */
1000 kprintf("mp_lock = %08x; ", mp_lock);
1001 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
1002 kprintf("lapic->id = %08x\n", lapic->id);
1004 panic("double fault");
1008 * syscall2 - MP aware system call request C handler
1010 * A system call is essentially treated as a trap except that the
1011 * MP lock is not held on entry or return. We are responsible for
1012 * obtaining the MP lock if necessary and for handling ASTs
1013 * (e.g. a task switch) prior to return.
1018 syscall2(struct trapframe *frame)
1020 struct thread *td = curthread;
1021 struct proc *p = td->td_proc;
1022 struct lwp *lp = td->td_lwp;
1024 struct sysent *callp;
1025 register_t orig_tf_rflags;
1030 int crit_count = td->td_pri & ~TDPRI_MASK;
1033 int have_mplock = 0;
1038 union sysunion args;
1039 register_t *argsdst;
1041 mycpu->gd_cnt.v_syscall++;
1044 if (ISPL(frame->tf_cs) != SEL_UPL) {
1051 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid,
1055 KASSERT(td->td_mpcount == 0, ("badmpcount syscall2 from %p", (void *)frame->tf_rip));
1056 if (syscall_mpsafe == 0)
1057 MAKEMPSAFE(have_mplock);
1059 userenter(td, lp); /* lazy raise our priority */
1066 sticks = (int)td->td_sticks;
1067 orig_tf_rflags = frame->tf_rflags;
1070 * Virtual kernel intercept - if a VM context managed by a virtual
1071 * kernel issues a system call the virtual kernel handles it, not us.
1072 * Restore the virtual kernel context and return from its system
1073 * call. The current frame is copied out to the virtual kernel.
1075 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1076 error = vkernel_trap(lp, frame);
1077 frame->tf_rax = error;
1079 frame->tf_rflags |= PSL_C;
1080 error = EJUSTRETURN;
1085 * Get the system call parameters and account for time
1087 lp->lwp_md.md_regs = frame;
1088 params = (caddr_t)frame->tf_rsp + sizeof(register_t);
1089 code = frame->tf_rax;
1091 if (p->p_sysent->sv_prepsyscall) {
1092 (*p->p_sysent->sv_prepsyscall)(
1093 frame, (int *)(&args.nosys.sysmsg + 1),
1096 if (code == SYS_syscall || code == SYS___syscall) {
1097 code = frame->tf_rdi;
1103 if (p->p_sysent->sv_mask)
1104 code &= p->p_sysent->sv_mask;
1106 if (code >= p->p_sysent->sv_size)
1107 callp = &p->p_sysent->sv_table[0];
1109 callp = &p->p_sysent->sv_table[code];
1111 narg = callp->sy_narg & SYF_ARGMASK;
1114 * On x86_64 we get up to six arguments in registers. The rest are
1115 * on the stack. The first six members of 'struct trapframe' happen
1116 * to be the registers used to pass arguments, in exactly the right
1119 argp = &frame->tf_rdi;
1121 argsdst = (register_t *)(&args.nosys.sysmsg + 1);
1123 * JG can we overflow the space pointed to by 'argsdst'
1124 * either with 'bcopy' or with 'copyin'?
1126 bcopy(argp, argsdst, sizeof(register_t) * regcnt);
1128 * copyin is MP aware, but the tracing code is not
1130 if (narg > regcnt) {
1131 KASSERT(params != NULL, ("copyin args with no params!"));
1132 error = copyin(params, &argsdst[regcnt],
1133 (narg - regcnt) * sizeof(register_t));
1136 if (KTRPOINT(td, KTR_SYSCALL)) {
1137 MAKEMPSAFE(have_mplock);
1139 ktrsyscall(lp, code, narg,
1140 (void *)(&args.nosys.sysmsg + 1));
1148 if (KTRPOINT(td, KTR_SYSCALL)) {
1149 MAKEMPSAFE(have_mplock);
1150 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1155 * Default return value is 0 (will be copied to %rax). Double-value
1156 * returns use %rax and %rdx. %rdx is left unchanged for system
1157 * calls which return only one result.
1159 args.sysmsg_fds[0] = 0;
1160 args.sysmsg_fds[1] = frame->tf_rdx;
1163 * The syscall might manipulate the trap frame. If it does it
1164 * will probably return EJUSTRETURN.
1166 args.sysmsg_frame = frame;
1168 STOPEVENT(p, S_SCE, narg); /* MP aware */
1171 * NOTE: All system calls run MPSAFE now. The system call itself
1172 * is responsible for getting the MP lock.
1174 error = (*callp->sy_call)(&args);
1178 * MP SAFE (we may or may not have the MP lock at this point)
1180 //kprintf("SYSMSG %d ", error);
1184 * Reinitialize proc pointer `p' as it may be different
1185 * if this is a child returning from fork syscall.
1188 lp = curthread->td_lwp;
1189 frame->tf_rax = args.sysmsg_fds[0];
1190 frame->tf_rdx = args.sysmsg_fds[1];
1191 frame->tf_rflags &= ~PSL_C;
1195 * Reconstruct pc, we know that 'syscall' is 2 bytes.
1196 * We have to do a full context restore so that %r10
1197 * (which was holding the value of %rcx) is restored for
1198 * the next iteration.
1200 frame->tf_rip -= frame->tf_err;
1201 frame->tf_r10 = frame->tf_rcx;
1206 panic("Unexpected EASYNC return value (for now)");
1209 if (p->p_sysent->sv_errsize) {
1210 if (error >= p->p_sysent->sv_errsize)
1211 error = -1; /* XXX */
1213 error = p->p_sysent->sv_errtbl[error];
1215 frame->tf_rax = error;
1216 frame->tf_rflags |= PSL_C;
1221 * Traced syscall. trapsignal() is not MP aware.
1223 if (orig_tf_rflags & PSL_T) {
1224 MAKEMPSAFE(have_mplock);
1225 frame->tf_rflags &= ~PSL_T;
1226 trapsignal(lp, SIGTRAP, 0);
1230 * Handle reschedule and other end-of-syscall issues
1232 userret(lp, frame, sticks);
1235 if (KTRPOINT(td, KTR_SYSRET)) {
1236 MAKEMPSAFE(have_mplock);
1237 ktrsysret(lp, code, error, args.sysmsg_result);
1242 * This works because errno is findable through the
1243 * register set. If we ever support an emulation where this
1244 * is not the case, this code will need to be revisited.
1246 STOPEVENT(p, S_SCX, code);
1251 * Release the MP lock if we had to get it
1253 KASSERT(td->td_mpcount == have_mplock,
1254 ("badmpcount syscall2/end from %p", (void *)frame->tf_rip));
1258 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error);
1260 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
1261 ("syscall: critical section count mismatch! %d/%d",
1262 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
1267 fork_return(struct lwp *lp, struct trapframe *frame)
1269 frame->tf_rax = 0; /* Child returns zero */
1270 frame->tf_rflags &= ~PSL_C; /* success */
1273 generic_lwp_return(lp, frame);
1274 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1278 * Simplified back end of syscall(), used when returning from fork()
1279 * directly into user mode. MP lock is held on entry and should be
1280 * released on return. This code will return back into the fork
1281 * trampoline code which then runs doreti.
1284 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1286 struct proc *p = lp->lwp_proc;
1289 * Newly forked processes are given a kernel priority. We have to
1290 * adjust the priority to a normal user priority and fake entry
1291 * into the kernel (call userenter()) to install a passive release
1292 * function just in case userret() decides to stop the process. This
1293 * can occur when ^Z races a fork. If we do not install the passive
1294 * release function the current process designation will not be
1295 * released when the thread goes to sleep.
1297 lwkt_setpri_self(TDPRI_USER_NORM);
1298 userenter(lp->lwp_thread, lp);
1299 userret(lp, frame, 0);
1301 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1302 ktrsysret(lp, SYS_fork, 0, 0);
1304 p->p_flag |= P_PASSIVE_ACQ;
1306 p->p_flag &= ~P_PASSIVE_ACQ;
1308 KKASSERT(lp->lwp_thread->td_mpcount == 1);
1314 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1315 * fault (which is then passed back to the virtual kernel) if an attempt is
1316 * made to use the FP unit.
1318 * XXX this is a fairly big hack.
1321 set_vkernel_fp(struct trapframe *frame)