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>
54 #include <sys/kerneldump.h>
56 #include <sys/pioctl.h>
57 #include <sys/types.h>
58 #include <sys/signal2.h>
59 #include <sys/syscall.h>
60 #include <sys/sysctl.h>
61 #include <sys/sysent.h>
62 #include <sys/systm.h>
64 #include <sys/ktrace.h>
67 #include <sys/sysmsg.h>
68 #include <sys/sysproto.h>
69 #include <sys/sysunion.h>
73 #include <vm/vm_extern.h>
74 #include <vm/vm_kern.h>
75 #include <vm/vm_param.h>
76 #include <machine/cpu.h>
77 #include <machine/pcb.h>
78 #include <machine/smp.h>
79 #include <machine/thread.h>
80 #include <machine/vmparam.h>
81 #include <machine/md_var.h>
82 #include <machine_base/isa/intr_machdep.h>
86 #include <sys/thread2.h>
87 #include <sys/mplock2.h>
91 #define MAKEMPSAFE(have_mplock) \
92 if (have_mplock == 0) { \
99 #define MAKEMPSAFE(have_mplock)
103 extern void trap(struct trapframe *frame);
105 static int trap_pfault(struct trapframe *, int);
106 static void trap_fatal(struct trapframe *, vm_offset_t);
107 void dblfault_handler(struct trapframe *frame);
109 #define MAX_TRAP_MSG 30
110 static char *trap_msg[] = {
112 "privileged instruction fault", /* 1 T_PRIVINFLT */
114 "breakpoint instruction fault", /* 3 T_BPTFLT */
117 "arithmetic trap", /* 6 T_ARITHTRAP */
118 "system forced exception", /* 7 T_ASTFLT */
120 "general protection fault", /* 9 T_PROTFLT */
121 "trace trap", /* 10 T_TRCTRAP */
123 "page fault", /* 12 T_PAGEFLT */
125 "alignment fault", /* 14 T_ALIGNFLT */
129 "integer divide fault", /* 18 T_DIVIDE */
130 "non-maskable interrupt trap", /* 19 T_NMI */
131 "overflow trap", /* 20 T_OFLOW */
132 "FPU bounds check fault", /* 21 T_BOUND */
133 "FPU device not available", /* 22 T_DNA */
134 "double fault", /* 23 T_DOUBLEFLT */
135 "FPU operand fetch fault", /* 24 T_FPOPFLT */
136 "invalid TSS fault", /* 25 T_TSSFLT */
137 "segment not present fault", /* 26 T_SEGNPFLT */
138 "stack fault", /* 27 T_STKFLT */
139 "machine check trap", /* 28 T_MCHK */
140 "SIMD floating-point exception", /* 29 T_XMMFLT */
141 "reserved (unknown) fault", /* 30 T_RESERVED */
145 static int ddb_on_nmi = 1;
146 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
147 &ddb_on_nmi, 0, "Go to DDB on NMI");
148 static int ddb_on_seg_fault = 0;
149 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_seg_fault, CTLFLAG_RW,
150 &ddb_on_seg_fault, 0, "Go to DDB on user seg-fault");
152 static int panic_on_nmi = 1;
153 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
154 &panic_on_nmi, 0, "Panic on NMI");
155 static int fast_release;
156 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
157 &fast_release, 0, "Passive Release was optimal");
158 static int slow_release;
159 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
160 &slow_release, 0, "Passive Release was nonoptimal");
163 * Passively intercepts the thread switch function to increase
164 * the thread priority from a user priority to a kernel priority, reducing
165 * syscall and trap overhead for the case where no switch occurs.
167 * Synchronizes td_ucred with p_ucred. This is used by system calls,
168 * signal handling, faults, AST traps, and anything else that enters the
169 * kernel from userland and provides the kernel with a stable read-only
170 * copy of the process ucred.
173 userenter(struct thread *curtd, struct proc *curp)
178 curtd->td_release = lwkt_passive_release;
180 if (curtd->td_ucred != curp->p_ucred) {
181 ncred = crhold(curp->p_ucred);
182 ocred = curtd->td_ucred;
183 curtd->td_ucred = ncred;
190 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
191 * must be completed before we can return to or try to return to userland.
193 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
194 * arithmatic on the delta calculation so the absolute tick values are
195 * truncated to an integer.
198 userret(struct lwp *lp, struct trapframe *frame, int sticks)
200 struct proc *p = lp->lwp_proc;
204 * Charge system time if profiling. Note: times are in microseconds.
205 * This may do a copyout and block, so do it first even though it
206 * means some system time will be charged as user time.
208 if (p->p_flag & P_PROFIL) {
209 addupc_task(p, frame->tf_rip,
210 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
215 * If the jungle wants us dead, so be it.
217 if (lp->lwp_flag & LWP_WEXIT) {
220 rel_mplock(); /* NOT REACHED */
224 * Block here if we are in a stopped state.
226 if (p->p_stat == SSTOP || dump_stop_usertds) {
234 * Post any pending upcalls. If running a virtual kernel be sure
235 * to restore the virtual kernel's vmspace before posting the upcall.
237 if (p->p_flag & P_UPCALLPEND) {
238 p->p_flag &= ~P_UPCALLPEND;
246 * Post any pending signals. If running a virtual kernel be sure
247 * to restore the virtual kernel's vmspace before posting the signal.
249 * WARNING! postsig() can exit and not return.
251 if ((sig = CURSIG_TRACE(lp)) != 0) {
259 * block here if we are swapped out, but still process signals
260 * (such as SIGKILL). proc0 (the swapin scheduler) is already
261 * aware of our situation, we do not have to wake it up.
263 if (p->p_flag & P_SWAPPEDOUT) {
265 p->p_flag |= P_SWAPWAIT;
267 if (p->p_flag & P_SWAPWAIT)
268 tsleep(p, PCATCH, "SWOUT", 0);
269 p->p_flag &= ~P_SWAPWAIT;
275 * Make sure postsig() handled request to restore old signal mask after
276 * running signal handler.
278 KKASSERT((lp->lwp_flag & LWP_OLDMASK) == 0);
282 * Cleanup from userenter and any passive release that might have occured.
283 * We must reclaim the current-process designation before we can return
284 * to usermode. We also handle both LWKT and USER reschedule requests.
287 userexit(struct lwp *lp)
289 struct thread *td = lp->lwp_thread;
290 /* globaldata_t gd = td->td_gd;*/
293 * Handle stop requests at kernel priority. Any requests queued
294 * after this loop will generate another AST.
296 while (lp->lwp_proc->p_stat == SSTOP) {
303 * Reduce our priority in preparation for a return to userland. If
304 * our passive release function was still in place, our priority was
305 * never raised and does not need to be reduced.
307 lwkt_passive_recover(td);
310 * Become the current user scheduled process if we aren't already,
311 * and deal with reschedule requests and other factors.
313 lp->lwp_proc->p_usched->acquire_curproc(lp);
314 /* WARNING: we may have migrated cpu's */
315 /* gd = td->td_gd; */
318 #if !defined(KTR_KERNENTRY)
319 #define KTR_KERNENTRY KTR_ALL
321 KTR_INFO_MASTER(kernentry);
322 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, "STR",
323 sizeof(long) + sizeof(long) + sizeof(long) + sizeof(vm_offset_t));
324 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "STR",
325 sizeof(long) + sizeof(long));
326 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "STR",
327 sizeof(long) + sizeof(long) + sizeof(long));
328 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "STR",
329 sizeof(long) + sizeof(long) + sizeof(long));
330 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "STR",
331 sizeof(long) + sizeof(long));
334 * Exception, fault, and trap interface to the kernel.
335 * This common code is called from assembly language IDT gate entry
336 * routines that prepare a suitable stack frame, and restore this
337 * frame after the exception has been processed.
339 * This function is also called from doreti in an interlock to handle ASTs.
340 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
342 * NOTE! We have to retrieve the fault address prior to obtaining the
343 * MP lock because get_mplock() may switch out. YYY cr2 really ought
344 * to be retrieved by the assembly code, not here.
346 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
347 * if an attempt is made to switch from a fast interrupt or IPI. This is
348 * necessary to properly take fatal kernel traps on SMP machines if
349 * get_mplock() has to block.
353 trap(struct trapframe *frame)
355 struct globaldata *gd = mycpu;
356 struct thread *td = gd->gd_curthread;
357 struct lwp *lp = td->td_lwp;
360 int i = 0, ucode = 0, type, code;
365 int crit_count = td->td_critcount;
366 lwkt_tokref_t curstop = td->td_toks_stop;
374 kprintf0("\"%s\" type=%ld\n",
375 trap_msg[frame->tf_trapno], frame->tf_trapno);
376 kprintf0(" rip=%lx rsp=%lx\n", frame->tf_rip, frame->tf_rsp);
377 kprintf0(" err=%lx addr=%lx\n", frame->tf_err, frame->tf_addr);
378 kprintf0(" cs=%lx ss=%lx rflags=%lx\n", (unsigned long)frame->tf_cs, (unsigned long)frame->tf_ss, frame->tf_rflags);
382 if (db_active && frame->tf_trapno != T_DNA) {
383 eva = (frame->tf_trapno == T_PAGEFLT ? frame->tf_addr : 0);
384 ++gd->gd_trap_nesting_level;
385 MAKEMPSAFE(have_mplock);
386 trap_fatal(frame, eva);
387 --gd->gd_trap_nesting_level;
394 if ((frame->tf_rflags & PSL_I) == 0) {
396 * Buggy application or kernel code has disabled interrupts
397 * and then trapped. Enabling interrupts now is wrong, but
398 * it is better than running with interrupts disabled until
399 * they are accidentally enabled later.
401 type = frame->tf_trapno;
402 if (ISPL(frame->tf_cs) == SEL_UPL) {
403 MAKEMPSAFE(have_mplock);
404 /* JG curproc can be NULL */
406 "pid %ld (%s): trap %d with interrupts disabled\n",
407 (long)curproc->p_pid, curproc->p_comm, type);
408 } else if (type != T_NMI && type != T_BPTFLT &&
411 * XXX not quite right, since this may be for a
412 * multiple fault in user mode.
414 MAKEMPSAFE(have_mplock);
415 kprintf("kernel trap %d with interrupts disabled\n",
421 type = frame->tf_trapno;
422 code = frame->tf_err;
424 if (ISPL(frame->tf_cs) == SEL_UPL) {
427 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid,
428 frame->tf_trapno, eva);
432 sticks = (int)td->td_sticks;
433 lp->lwp_md.md_regs = frame;
436 case T_PRIVINFLT: /* privileged instruction fault */
441 case T_BPTFLT: /* bpt instruction fault */
442 case T_TRCTRAP: /* trace trap */
443 frame->tf_rflags &= ~PSL_T;
448 case T_ARITHTRAP: /* arithmetic trap */
461 case T_ASTFLT: /* Allow process switch */
462 mycpu->gd_cnt.v_soft++;
463 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
464 atomic_clear_int_nonlocked(&mycpu->gd_reqflags,
466 addupc_task(p, p->p_prof.pr_addr,
471 case T_PROTFLT: /* general protection fault */
475 case T_SEGNPFLT: /* segment not present fault */
479 case T_TSSFLT: /* invalid TSS fault */
480 case T_DOUBLEFLT: /* double fault */
485 ucode = code + BUS_SEGM_FAULT ; /* XXX: ???*/
491 case T_PAGEFLT: /* page fault */
492 MAKEMPSAFE(have_mplock);
493 i = trap_pfault(frame, TRUE);
494 if (frame->tf_rip == 0)
495 kprintf("T_PAGEFLT: Warning %%rip == 0!\n");
510 case T_DIVIDE: /* integer divide fault */
516 MAKEMPSAFE(have_mplock);
517 /* machine/parity/power fail/"kitchen sink" faults */
518 if (isa_nmi(code) == 0) {
521 * NMI can be hooked up to a pushbutton
525 kprintf ("NMI ... going to debugger\n");
526 kdb_trap(type, 0, frame);
530 } else if (panic_on_nmi)
531 panic("NMI indicates hardware failure");
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 virtual kernel if it asked to handle it.
548 * 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 virtual kernel needs
554 * 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)
564 * The kernel may have switched out the FP unit's
565 * state, causing the user process to take a fault
566 * when it tries to use the FP unit. Restore the
572 ucode = FPE_FPU_NP_TRAP;
575 case T_FPOPFLT: /* FPU operand fetch fault */
580 case T_XMMFLT: /* SIMD floating-point exception */
589 case T_PAGEFLT: /* page fault */
590 MAKEMPSAFE(have_mplock);
591 trap_pfault(frame, FALSE);
596 * The kernel is apparently using fpu for copying.
597 * XXX this should be fatal unless the kernel has
598 * registered such use.
604 case T_STKFLT: /* stack fault */
607 case T_PROTFLT: /* general protection fault */
608 case T_SEGNPFLT: /* segment not present fault */
610 * Invalid segment selectors and out of bounds
611 * %rip's and %rsp's can be set up in user mode.
612 * This causes a fault in kernel mode when the
613 * kernel tries to return to user mode. We want
614 * to get this fault so that we can fix the
615 * problem here and not have to check all the
616 * selectors and pointers when the user changes
619 kprintf("trap.c line %d\n", __LINE__);
620 if (mycpu->gd_intr_nesting_level == 0) {
621 if (td->td_pcb->pcb_onfault) {
622 frame->tf_rip = (register_t)
623 td->td_pcb->pcb_onfault;
626 if (frame->tf_rip == (long)doreti_iret) {
627 frame->tf_rip = (long)doreti_iret_fault;
635 * PSL_NT can be set in user mode and isn't cleared
636 * automatically when the kernel is entered. This
637 * causes a TSS fault when the kernel attempts to
638 * `iret' because the TSS link is uninitialized. We
639 * want to get this fault so that we can fix the
640 * problem here and not every time the kernel is
643 if (frame->tf_rflags & PSL_NT) {
644 frame->tf_rflags &= ~PSL_NT;
649 case T_TRCTRAP: /* trace trap */
651 if (frame->tf_rip == (int)IDTVEC(syscall)) {
653 * We've just entered system mode via the
654 * syscall lcall. Continue single stepping
655 * silently until the syscall handler has
660 if (frame->tf_rip == (int)IDTVEC(syscall) + 1) {
662 * The syscall handler has now saved the
663 * flags. Stop single stepping it.
665 frame->tf_rflags &= ~PSL_T;
671 * Ignore debug register trace traps due to
672 * accesses in the user's address space, which
673 * can happen under several conditions such as
674 * if a user sets a watchpoint on a buffer and
675 * then passes that buffer to a system call.
676 * We still want to get TRCTRAPS for addresses
677 * in kernel space because that is useful when
678 * debugging the kernel.
681 if (user_dbreg_trap()) {
683 * Reset breakpoint bits because the
686 /* XXX check upper bits here */
687 load_dr6(rdr6() & 0xfffffff0);
692 * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
696 * If DDB is enabled, let it handle the debugger trap.
697 * 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 * Translate fault for emulators (e.g. Linux)
746 if (*p->p_sysent->sv_transtrap)
747 i = (*p->p_sysent->sv_transtrap)(i, type);
749 MAKEMPSAFE(have_mplock);
750 trapsignal(lp, i, ucode);
753 if (type <= MAX_TRAP_MSG) {
754 uprintf("fatal process exception: %s",
756 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
757 uprintf(", fault VA = 0x%lx", frame->tf_addr);
764 if (ISPL(frame->tf_cs) == SEL_UPL) {
765 KASSERT(td->td_mpcount == have_mplock,
766 ("badmpcount trap/end from %p", (void *)frame->tf_rip));
769 userret(lp, frame, sticks);
776 if (p != NULL && lp != NULL)
777 KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid);
779 KASSERT(crit_count == td->td_critcount,
780 ("trap: critical section count mismatch! %d/%d",
781 crit_count, td->td_pri));
782 KASSERT(curstop == td->td_toks_stop,
783 ("trap: extra tokens held after trap! %ld/%ld",
784 curstop - &td->td_toks_base,
785 td->td_toks_stop - &td->td_toks_base));
790 trap_pfault(struct trapframe *frame, int usermode)
793 struct vmspace *vm = NULL;
798 thread_t td = curthread;
799 struct lwp *lp = td->td_lwp;
802 va = trunc_page(frame->tf_addr);
803 if (va >= VM_MIN_KERNEL_ADDRESS) {
805 * Don't allow user-mode faults in kernel address space.
816 * This is a fault on non-kernel virtual memory.
817 * vm is initialized above to NULL. If curproc is NULL
818 * or curproc->p_vmspace is NULL the fault is fatal.
821 vm = lp->lwp_vmspace;
833 * PGEX_I is defined only if the execute disable bit capability is
834 * supported and enabled.
836 if (frame->tf_err & PGEX_W)
837 ftype = VM_PROT_WRITE;
839 else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
840 ftype = VM_PROT_EXECUTE;
843 ftype = VM_PROT_READ;
845 if (map != &kernel_map) {
847 * Keep swapout from messing with us during this
857 fault_flags |= VM_FAULT_BURST;
858 if (ftype & VM_PROT_WRITE)
859 fault_flags |= VM_FAULT_DIRTY;
861 fault_flags |= VM_FAULT_NORMAL;
862 rv = vm_fault(map, va, ftype, fault_flags);
867 * Don't have to worry about process locking or stacks
870 fault_flags = VM_FAULT_NORMAL;
871 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
874 if (rv == KERN_SUCCESS)
878 if (td->td_gd->gd_intr_nesting_level == 0 &&
879 td->td_pcb->pcb_onfault) {
880 frame->tf_rip = (register_t)td->td_pcb->pcb_onfault;
883 trap_fatal(frame, frame->tf_addr);
888 * NOTE: on x86_64 we have a tf_addr field in the trapframe, no
889 * kludge is needed to pass the fault address to signal handlers.
892 if (td->td_lwp->lwp_vkernel == NULL) {
893 kprintf("seg-fault ft=%04x ff=%04x addr=%p rip=%p "
894 "pid=%d p_comm=%s\n",
896 (void *)frame->tf_addr,
897 (void *)frame->tf_rip,
898 p->p_pid, p->p_comm);
899 if (ddb_on_seg_fault)
900 Debugger("ddb_on_seg_fault");
902 /* Debugger("seg-fault"); */
904 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
908 trap_fatal(struct trapframe *frame, vm_offset_t eva)
913 struct soft_segment_descriptor softseg;
916 code = frame->tf_err;
917 type = frame->tf_trapno;
918 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)], &softseg);
920 if (type <= MAX_TRAP_MSG)
921 msg = trap_msg[type];
924 kprintf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
925 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
927 /* three separate prints in case of a trap on an unmapped page */
928 kprintf("mp_lock = %08x; ", mp_lock);
929 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
930 kprintf("lapic->id = %08x\n", lapic->id);
932 if (type == T_PAGEFLT) {
933 kprintf("fault virtual address = 0x%lx\n", eva);
934 kprintf("fault code = %s %s %s, %s\n",
935 code & PGEX_U ? "user" : "supervisor",
936 code & PGEX_W ? "write" : "read",
937 code & PGEX_I ? "instruction" : "data",
938 code & PGEX_P ? "protection violation" : "page not present");
940 kprintf("instruction pointer = 0x%lx:0x%lx\n",
941 frame->tf_cs & 0xffff, frame->tf_rip);
942 if (ISPL(frame->tf_cs) == SEL_UPL) {
943 ss = frame->tf_ss & 0xffff;
946 ss = GSEL(GDATA_SEL, SEL_KPL);
947 rsp = (long)&frame->tf_rsp;
949 kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp);
950 kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp);
951 kprintf("code segment = base 0x%lx, limit 0x%lx, type 0x%x\n",
952 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
953 kprintf(" = DPL %d, pres %d, long %d, def32 %d, gran %d\n",
954 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32,
956 kprintf("processor eflags = ");
957 if (frame->tf_rflags & PSL_T)
958 kprintf("trace trap, ");
959 if (frame->tf_rflags & PSL_I)
960 kprintf("interrupt enabled, ");
961 if (frame->tf_rflags & PSL_NT)
962 kprintf("nested task, ");
963 if (frame->tf_rflags & PSL_RF)
965 kprintf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12);
966 kprintf("current process = ");
969 (u_long)curproc->p_pid);
973 kprintf("current thread = pri %d ", curthread->td_pri);
974 if (curthread->td_critcount)
979 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
982 kprintf("trap number = %d\n", type);
983 if (type <= MAX_TRAP_MSG)
984 panic("%s", trap_msg[type]);
986 panic("unknown/reserved trap");
990 * Double fault handler. Called when a fault occurs while writing
991 * a frame for a trap/exception onto the stack. This usually occurs
992 * when the stack overflows (such is the case with infinite recursion,
996 dblfault_handler(struct trapframe *frame)
998 kprintf0("DOUBLE FAULT\n");
999 kprintf("\nFatal double fault\n");
1000 kprintf("rip = 0x%lx\n", frame->tf_rip);
1001 kprintf("rsp = 0x%lx\n", frame->tf_rsp);
1002 kprintf("rbp = 0x%lx\n", frame->tf_rbp);
1004 /* three separate prints in case of a trap on an unmapped page */
1005 kprintf("mp_lock = %08x; ", mp_lock);
1006 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
1007 kprintf("lapic->id = %08x\n", lapic->id);
1009 panic("double fault");
1013 * syscall2 - MP aware system call request C handler
1015 * A system call is essentially treated as a trap except that the
1016 * MP lock is not held on entry or return. We are responsible for
1017 * obtaining the MP lock if necessary and for handling ASTs
1018 * (e.g. a task switch) prior to return.
1023 syscall2(struct trapframe *frame)
1025 struct thread *td = curthread;
1026 struct proc *p = td->td_proc;
1027 struct lwp *lp = td->td_lwp;
1029 struct sysent *callp;
1030 register_t orig_tf_rflags;
1035 int crit_count = td->td_critcount;
1038 int have_mplock = 0;
1043 union sysunion args;
1044 register_t *argsdst;
1046 mycpu->gd_cnt.v_syscall++;
1049 if (ISPL(frame->tf_cs) != SEL_UPL) {
1056 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid,
1060 KASSERT(td->td_mpcount == 0,
1061 ("badmpcount syscall2 from %p", (void *)frame->tf_rip));
1063 userenter(td, p); /* lazy raise our priority */
1070 sticks = (int)td->td_sticks;
1071 orig_tf_rflags = frame->tf_rflags;
1074 * Virtual kernel intercept - if a VM context managed by a virtual
1075 * kernel issues a system call the virtual kernel handles it, not us.
1076 * Restore the virtual kernel context and return from its system
1077 * call. The current frame is copied out to the virtual kernel.
1079 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1080 vkernel_trap(lp, frame);
1081 error = EJUSTRETURN;
1086 * Get the system call parameters and account for time
1088 lp->lwp_md.md_regs = frame;
1089 params = (caddr_t)frame->tf_rsp + sizeof(register_t);
1090 code = frame->tf_rax;
1092 if (p->p_sysent->sv_prepsyscall) {
1093 (*p->p_sysent->sv_prepsyscall)(
1094 frame, (int *)(&args.nosys.sysmsg + 1),
1097 if (code == SYS_syscall || code == SYS___syscall) {
1098 code = frame->tf_rdi;
1104 if (p->p_sysent->sv_mask)
1105 code &= p->p_sysent->sv_mask;
1107 if (code >= p->p_sysent->sv_size)
1108 callp = &p->p_sysent->sv_table[0];
1110 callp = &p->p_sysent->sv_table[code];
1112 narg = callp->sy_narg & SYF_ARGMASK;
1115 * On x86_64 we get up to six arguments in registers. The rest are
1116 * on the stack. The first six members of 'struct trapframe' happen
1117 * to be the registers used to pass arguments, in exactly the right
1120 argp = &frame->tf_rdi;
1122 argsdst = (register_t *)(&args.nosys.sysmsg + 1);
1124 * JG can we overflow the space pointed to by 'argsdst'
1125 * either with 'bcopy' or with 'copyin'?
1127 bcopy(argp, argsdst, sizeof(register_t) * regcnt);
1129 * copyin is MP aware, but the tracing code is not
1131 if (narg > regcnt) {
1132 KASSERT(params != NULL, ("copyin args with no params!"));
1133 error = copyin(params, &argsdst[regcnt],
1134 (narg - regcnt) * sizeof(register_t));
1137 if (KTRPOINT(td, KTR_SYSCALL)) {
1138 MAKEMPSAFE(have_mplock);
1140 ktrsyscall(lp, code, narg,
1141 (void *)(&args.nosys.sysmsg + 1));
1149 if (KTRPOINT(td, KTR_SYSCALL)) {
1150 MAKEMPSAFE(have_mplock);
1151 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1156 * Default return value is 0 (will be copied to %rax). Double-value
1157 * returns use %rax and %rdx. %rdx is left unchanged for system
1158 * calls which return only one result.
1160 args.sysmsg_fds[0] = 0;
1161 args.sysmsg_fds[1] = frame->tf_rdx;
1164 * The syscall might manipulate the trap frame. If it does it
1165 * will probably return EJUSTRETURN.
1167 args.sysmsg_frame = frame;
1169 STOPEVENT(p, S_SCE, narg); /* MP aware */
1172 * NOTE: All system calls run MPSAFE now. The system call itself
1173 * is responsible for getting the MP lock.
1175 error = (*callp->sy_call)(&args);
1179 * MP SAFE (we may or may not have the MP lock at this point)
1181 //kprintf("SYSMSG %d ", error);
1185 * Reinitialize proc pointer `p' as it may be different
1186 * if this is a child returning from fork syscall.
1189 lp = curthread->td_lwp;
1190 frame->tf_rax = args.sysmsg_fds[0];
1191 frame->tf_rdx = args.sysmsg_fds[1];
1192 frame->tf_rflags &= ~PSL_C;
1196 * Reconstruct pc, we know that 'syscall' is 2 bytes.
1197 * We have to do a full context restore so that %r10
1198 * (which was holding the value of %rcx) is restored for
1199 * the next iteration.
1201 frame->tf_rip -= frame->tf_err;
1202 frame->tf_r10 = frame->tf_rcx;
1207 panic("Unexpected EASYNC return value (for now)");
1210 if (p->p_sysent->sv_errsize) {
1211 if (error >= p->p_sysent->sv_errsize)
1212 error = -1; /* XXX */
1214 error = p->p_sysent->sv_errtbl[error];
1216 frame->tf_rax = error;
1217 frame->tf_rflags |= PSL_C;
1222 * Traced syscall. trapsignal() is not MP aware.
1224 if (orig_tf_rflags & PSL_T) {
1225 MAKEMPSAFE(have_mplock);
1226 frame->tf_rflags &= ~PSL_T;
1227 trapsignal(lp, SIGTRAP, TRAP_TRACE);
1231 * Handle reschedule and other end-of-syscall issues
1233 userret(lp, frame, sticks);
1236 if (KTRPOINT(td, KTR_SYSRET)) {
1237 MAKEMPSAFE(have_mplock);
1238 ktrsysret(lp, code, error, args.sysmsg_result);
1243 * This works because errno is findable through the
1244 * register set. If we ever support an emulation where this
1245 * is not the case, this code will need to be revisited.
1247 STOPEVENT(p, S_SCX, code);
1252 * Release the MP lock if we had to get it
1254 KASSERT(td->td_mpcount == have_mplock,
1255 ("badmpcount syscall2/end from %p", (void *)frame->tf_rip));
1259 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error);
1261 KASSERT(crit_count == td->td_critcount,
1262 ("syscall: critical section count mismatch! %d/%d",
1263 crit_count, td->td_pri));
1264 KASSERT(&td->td_toks_base == td->td_toks_stop,
1265 ("syscall: extra tokens held after trap! %ld",
1266 td->td_toks_stop - &td->td_toks_base));
1271 * NOTE: mplock not held at any point
1274 fork_return(struct lwp *lp, struct trapframe *frame)
1276 frame->tf_rax = 0; /* Child returns zero */
1277 frame->tf_rflags &= ~PSL_C; /* success */
1280 generic_lwp_return(lp, frame);
1281 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1285 * Simplified back end of syscall(), used when returning from fork()
1286 * directly into user mode.
1288 * This code will return back into the fork trampoline code which then
1291 * NOTE: The mplock is not held at any point.
1294 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1296 struct proc *p = lp->lwp_proc;
1299 * Newly forked processes are given a kernel priority. We have to
1300 * adjust the priority to a normal user priority and fake entry
1301 * into the kernel (call userenter()) to install a passive release
1302 * function just in case userret() decides to stop the process. This
1303 * can occur when ^Z races a fork. If we do not install the passive
1304 * release function the current process designation will not be
1305 * released when the thread goes to sleep.
1307 lwkt_setpri_self(TDPRI_USER_NORM);
1308 userenter(lp->lwp_thread, p);
1309 userret(lp, frame, 0);
1311 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1312 ktrsysret(lp, SYS_fork, 0, 0);
1314 p->p_flag |= P_PASSIVE_ACQ;
1316 p->p_flag &= ~P_PASSIVE_ACQ;
1320 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1321 * fault (which is then passed back to the virtual kernel) if an attempt is
1322 * made to use the FP unit.
1324 * XXX this is a fairly big hack.
1327 set_vkernel_fp(struct trapframe *frame)
1329 struct thread *td = curthread;
1331 if (frame->tf_xflags & PGEX_FPFAULT) {
1332 td->td_pcb->pcb_flags |= FP_VIRTFP;
1333 if (mdcpu->gd_npxthread == td)
1336 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1341 * Called from vkernel_trap() to fixup the vkernel's syscall
1342 * frame for vmspace_ctl() return.
1345 cpu_vkernel_trap(struct trapframe *frame, int error)
1347 frame->tf_rax = error;
1349 frame->tf_rflags |= PSL_C;
1351 frame->tf_rflags &= ~PSL_C;