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
50 #include "opt_ktrace.h"
52 #include <machine/frame.h>
53 #include <sys/param.h>
54 #include <sys/systm.h>
55 #include <sys/kernel.h>
56 #include <sys/kerneldump.h>
58 #include <sys/pioctl.h>
59 #include <sys/types.h>
60 #include <sys/signal2.h>
61 #include <sys/syscall.h>
62 #include <sys/sysctl.h>
63 #include <sys/sysent.h>
64 #include <sys/systm.h>
66 #include <sys/ktrace.h>
69 #include <sys/sysmsg.h>
70 #include <sys/sysproto.h>
71 #include <sys/sysunion.h>
75 #include <vm/vm_extern.h>
76 #include <vm/vm_kern.h>
77 #include <vm/vm_param.h>
78 #include <machine/cpu.h>
79 #include <machine/pcb.h>
80 #include <machine/smp.h>
81 #include <machine/thread.h>
82 #include <machine/clock.h>
83 #include <machine/vmparam.h>
84 #include <machine/md_var.h>
85 #include <machine_base/isa/isa_intr.h>
86 #include <machine_base/apic/lapic.h>
90 #include <sys/thread2.h>
91 #include <sys/mplock2.h>
95 #define MAKEMPSAFE(have_mplock) \
96 if (have_mplock == 0) { \
103 #define MAKEMPSAFE(have_mplock)
107 extern void trap(struct trapframe *frame);
109 static int trap_pfault(struct trapframe *, int);
110 static void trap_fatal(struct trapframe *, vm_offset_t);
111 void dblfault_handler(struct trapframe *frame);
113 #define MAX_TRAP_MSG 30
114 static char *trap_msg[] = {
116 "privileged instruction fault", /* 1 T_PRIVINFLT */
118 "breakpoint instruction fault", /* 3 T_BPTFLT */
121 "arithmetic trap", /* 6 T_ARITHTRAP */
122 "system forced exception", /* 7 T_ASTFLT */
124 "general protection fault", /* 9 T_PROTFLT */
125 "trace trap", /* 10 T_TRCTRAP */
127 "page fault", /* 12 T_PAGEFLT */
129 "alignment fault", /* 14 T_ALIGNFLT */
133 "integer divide fault", /* 18 T_DIVIDE */
134 "non-maskable interrupt trap", /* 19 T_NMI */
135 "overflow trap", /* 20 T_OFLOW */
136 "FPU bounds check fault", /* 21 T_BOUND */
137 "FPU device not available", /* 22 T_DNA */
138 "double fault", /* 23 T_DOUBLEFLT */
139 "FPU operand fetch fault", /* 24 T_FPOPFLT */
140 "invalid TSS fault", /* 25 T_TSSFLT */
141 "segment not present fault", /* 26 T_SEGNPFLT */
142 "stack fault", /* 27 T_STKFLT */
143 "machine check trap", /* 28 T_MCHK */
144 "SIMD floating-point exception", /* 29 T_XMMFLT */
145 "reserved (unknown) fault", /* 30 T_RESERVED */
149 static int ddb_on_nmi = 1;
150 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
151 &ddb_on_nmi, 0, "Go to DDB on NMI");
152 static int ddb_on_seg_fault = 0;
153 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_seg_fault, CTLFLAG_RW,
154 &ddb_on_seg_fault, 0, "Go to DDB on user seg-fault");
155 static int freeze_on_seg_fault = 0;
156 SYSCTL_INT(_machdep, OID_AUTO, freeze_on_seg_fault, CTLFLAG_RW,
157 &freeze_on_seg_fault, 0, "Go to DDB on user seg-fault");
159 static int panic_on_nmi = 1;
160 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
161 &panic_on_nmi, 0, "Panic on NMI");
162 static int fast_release;
163 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
164 &fast_release, 0, "Passive Release was optimal");
165 static int slow_release;
166 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
167 &slow_release, 0, "Passive Release was nonoptimal");
170 * System call debugging records the worst-case system call
171 * overhead (inclusive of blocking), but may be inaccurate.
173 /*#define SYSCALL_DEBUG*/
175 uint64_t SysCallsWorstCase[SYS_MAXSYSCALL];
179 * Passively intercepts the thread switch function to increase
180 * the thread priority from a user priority to a kernel priority, reducing
181 * syscall and trap overhead for the case where no switch occurs.
183 * Synchronizes td_ucred with p_ucred. This is used by system calls,
184 * signal handling, faults, AST traps, and anything else that enters the
185 * kernel from userland and provides the kernel with a stable read-only
186 * copy of the process ucred.
189 userenter(struct thread *curtd, struct proc *curp)
194 curtd->td_release = lwkt_passive_release;
196 if (curtd->td_ucred != curp->p_ucred) {
197 ncred = crhold(curp->p_ucred);
198 ocred = curtd->td_ucred;
199 curtd->td_ucred = ncred;
206 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
207 * must be completed before we can return to or try to return to userland.
209 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
210 * arithmatic on the delta calculation so the absolute tick values are
211 * truncated to an integer.
214 userret(struct lwp *lp, struct trapframe *frame, int sticks)
216 struct proc *p = lp->lwp_proc;
220 * Charge system time if profiling. Note: times are in microseconds.
221 * This may do a copyout and block, so do it first even though it
222 * means some system time will be charged as user time.
224 if (p->p_flag & P_PROFIL) {
225 addupc_task(p, frame->tf_rip,
226 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
231 * If the jungle wants us dead, so be it.
233 if (lp->lwp_flag & LWP_WEXIT) {
234 lwkt_gettoken(&p->p_token);
236 lwkt_reltoken(&p->p_token); /* NOT REACHED */
240 * Block here if we are in a stopped state.
242 if (p->p_stat == SSTOP || dump_stop_usertds) {
250 * Post any pending upcalls. If running a virtual kernel be sure
251 * to restore the virtual kernel's vmspace before posting the upcall.
253 if (p->p_flag & P_UPCALLPEND) {
254 lwkt_gettoken(&p->p_token);
255 p->p_flag &= ~P_UPCALLPEND;
257 lwkt_reltoken(&p->p_token);
262 * Post any pending signals. If running a virtual kernel be sure
263 * to restore the virtual kernel's vmspace before posting the signal.
265 * WARNING! postsig() can exit and not return.
267 if ((sig = CURSIG_TRACE(lp)) != 0) {
268 lwkt_gettoken(&p->p_token);
270 lwkt_reltoken(&p->p_token);
275 * block here if we are swapped out, but still process signals
276 * (such as SIGKILL). proc0 (the swapin scheduler) is already
277 * aware of our situation, we do not have to wake it up.
279 if (p->p_flag & P_SWAPPEDOUT) {
281 p->p_flag |= P_SWAPWAIT;
283 if (p->p_flag & P_SWAPWAIT)
284 tsleep(p, PCATCH, "SWOUT", 0);
285 p->p_flag &= ~P_SWAPWAIT;
291 * Make sure postsig() handled request to restore old signal mask after
292 * running signal handler.
294 KKASSERT((lp->lwp_flag & LWP_OLDMASK) == 0);
298 * Cleanup from userenter and any passive release that might have occured.
299 * We must reclaim the current-process designation before we can return
300 * to usermode. We also handle both LWKT and USER reschedule requests.
303 userexit(struct lwp *lp)
305 struct thread *td = lp->lwp_thread;
306 /* globaldata_t gd = td->td_gd;*/
309 * Handle stop requests at kernel priority. Any requests queued
310 * after this loop will generate another AST.
312 while (lp->lwp_proc->p_stat == SSTOP) {
319 * Reduce our priority in preparation for a return to userland. If
320 * our passive release function was still in place, our priority was
321 * never raised and does not need to be reduced.
323 lwkt_passive_recover(td);
326 * Become the current user scheduled process if we aren't already,
327 * and deal with reschedule requests and other factors.
329 lp->lwp_proc->p_usched->acquire_curproc(lp);
330 /* WARNING: we may have migrated cpu's */
331 /* gd = td->td_gd; */
334 #if !defined(KTR_KERNENTRY)
335 #define KTR_KERNENTRY KTR_ALL
337 KTR_INFO_MASTER(kernentry);
338 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, "STR",
339 sizeof(long) + sizeof(long) + sizeof(long) + sizeof(vm_offset_t));
340 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "STR",
341 sizeof(long) + sizeof(long));
342 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "STR",
343 sizeof(long) + sizeof(long) + sizeof(long));
344 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "STR",
345 sizeof(long) + sizeof(long) + sizeof(long));
346 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "STR",
347 sizeof(long) + sizeof(long));
350 * Exception, fault, and trap interface to the kernel.
351 * This common code is called from assembly language IDT gate entry
352 * routines that prepare a suitable stack frame, and restore this
353 * frame after the exception has been processed.
355 * This function is also called from doreti in an interlock to handle ASTs.
356 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
358 * NOTE! We have to retrieve the fault address prior to obtaining the
359 * MP lock because get_mplock() may switch out. YYY cr2 really ought
360 * to be retrieved by the assembly code, not here.
362 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
363 * if an attempt is made to switch from a fast interrupt or IPI. This is
364 * necessary to properly take fatal kernel traps on SMP machines if
365 * get_mplock() has to block.
369 trap(struct trapframe *frame)
371 struct globaldata *gd = mycpu;
372 struct thread *td = gd->gd_curthread;
373 struct lwp *lp = td->td_lwp;
376 int i = 0, ucode = 0, type, code;
381 int crit_count = td->td_critcount;
382 lwkt_tokref_t curstop = td->td_toks_stop;
390 * We need to allow T_DNA faults when the debugger is active since
391 * some dumping paths do large bcopy() which use the floating
392 * point registers for faster copying.
394 if (db_active && frame->tf_trapno != T_DNA) {
395 eva = (frame->tf_trapno == T_PAGEFLT ? frame->tf_addr : 0);
396 ++gd->gd_trap_nesting_level;
397 MAKEMPSAFE(have_mplock);
398 trap_fatal(frame, eva);
399 --gd->gd_trap_nesting_level;
406 if ((frame->tf_rflags & PSL_I) == 0) {
408 * Buggy application or kernel code has disabled interrupts
409 * and then trapped. Enabling interrupts now is wrong, but
410 * it is better than running with interrupts disabled until
411 * they are accidentally enabled later.
413 type = frame->tf_trapno;
414 if (ISPL(frame->tf_cs) == SEL_UPL) {
415 MAKEMPSAFE(have_mplock);
416 /* JG curproc can be NULL */
418 "pid %ld (%s): trap %d with interrupts disabled\n",
419 (long)curproc->p_pid, curproc->p_comm, type);
420 } else if (type != T_NMI && type != T_BPTFLT &&
423 * XXX not quite right, since this may be for a
424 * multiple fault in user mode.
426 MAKEMPSAFE(have_mplock);
427 kprintf("kernel trap %d with interrupts disabled\n",
433 type = frame->tf_trapno;
434 code = frame->tf_err;
436 if (ISPL(frame->tf_cs) == SEL_UPL) {
439 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid,
440 frame->tf_trapno, eva);
444 sticks = (int)td->td_sticks;
445 KASSERT(lp->lwp_md.md_regs == frame,
446 ("Frame mismatch %p %p", lp->lwp_md.md_regs, frame));
449 case T_PRIVINFLT: /* privileged instruction fault */
454 case T_BPTFLT: /* bpt instruction fault */
455 case T_TRCTRAP: /* trace trap */
456 frame->tf_rflags &= ~PSL_T;
461 case T_ARITHTRAP: /* arithmetic trap */
474 case T_ASTFLT: /* Allow process switch */
475 mycpu->gd_cnt.v_soft++;
476 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
477 atomic_clear_int(&mycpu->gd_reqflags,
479 addupc_task(p, p->p_prof.pr_addr,
484 case T_PROTFLT: /* general protection fault */
488 case T_SEGNPFLT: /* segment not present fault */
492 case T_TSSFLT: /* invalid TSS fault */
493 case T_DOUBLEFLT: /* double fault */
498 ucode = code + BUS_SEGM_FAULT ; /* XXX: ???*/
504 case T_PAGEFLT: /* page fault */
505 i = trap_pfault(frame, TRUE);
506 if (frame->tf_rip == 0) {
507 kprintf("T_PAGEFLT: Warning %%rip == 0!\n");
508 while (freeze_on_seg_fault) {
509 tsleep(p, 0, "freeze", hz * 20);
526 case T_DIVIDE: /* integer divide fault */
533 MAKEMPSAFE(have_mplock);
534 /* machine/parity/power fail/"kitchen sink" faults */
535 if (isa_nmi(code) == 0) {
538 * NMI can be hooked up to a pushbutton
542 kprintf ("NMI ... going to debugger\n");
543 kdb_trap(type, 0, frame);
547 } else if (panic_on_nmi)
548 panic("NMI indicates hardware failure");
550 #endif /* NISA > 0 */
552 case T_OFLOW: /* integer overflow fault */
557 case T_BOUND: /* bounds check fault */
564 * Virtual kernel intercept - pass the DNA exception
565 * to the virtual kernel if it asked to handle it.
566 * This occurs when the virtual kernel is holding
567 * onto the FP context for a different emulated
568 * process then the one currently running.
570 * We must still call npxdna() since we may have
571 * saved FP state that the virtual kernel needs
572 * to hand over to a different emulated process.
574 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
575 (td->td_pcb->pcb_flags & FP_VIRTFP)
582 * The kernel may have switched out the FP unit's
583 * state, causing the user process to take a fault
584 * when it tries to use the FP unit. Restore the
590 ucode = FPE_FPU_NP_TRAP;
593 case T_FPOPFLT: /* FPU operand fetch fault */
598 case T_XMMFLT: /* SIMD floating-point exception */
607 case T_PAGEFLT: /* page fault */
608 trap_pfault(frame, FALSE);
613 * The kernel is apparently using fpu for copying.
614 * XXX this should be fatal unless the kernel has
615 * registered such use.
621 case T_STKFLT: /* stack fault */
624 case T_PROTFLT: /* general protection fault */
625 case T_SEGNPFLT: /* segment not present fault */
627 * Invalid segment selectors and out of bounds
628 * %rip's and %rsp's can be set up in user mode.
629 * This causes a fault in kernel mode when the
630 * kernel tries to return to user mode. We want
631 * to get this fault so that we can fix the
632 * problem here and not have to check all the
633 * selectors and pointers when the user changes
636 if (mycpu->gd_intr_nesting_level == 0) {
637 if (td->td_pcb->pcb_onfault) {
638 frame->tf_rip = (register_t)
639 td->td_pcb->pcb_onfault;
642 if (frame->tf_rip == (long)doreti_iret) {
643 frame->tf_rip = (long)doreti_iret_fault;
651 * PSL_NT can be set in user mode and isn't cleared
652 * automatically when the kernel is entered. This
653 * causes a TSS fault when the kernel attempts to
654 * `iret' because the TSS link is uninitialized. We
655 * want to get this fault so that we can fix the
656 * problem here and not every time the kernel is
659 if (frame->tf_rflags & PSL_NT) {
660 frame->tf_rflags &= ~PSL_NT;
665 case T_TRCTRAP: /* trace trap */
667 if (frame->tf_rip == (int)IDTVEC(syscall)) {
669 * We've just entered system mode via the
670 * syscall lcall. Continue single stepping
671 * silently until the syscall handler has
676 if (frame->tf_rip == (int)IDTVEC(syscall) + 1) {
678 * The syscall handler has now saved the
679 * flags. Stop single stepping it.
681 frame->tf_rflags &= ~PSL_T;
687 * Ignore debug register trace traps due to
688 * accesses in the user's address space, which
689 * can happen under several conditions such as
690 * if a user sets a watchpoint on a buffer and
691 * then passes that buffer to a system call.
692 * We still want to get TRCTRAPS for addresses
693 * in kernel space because that is useful when
694 * debugging the kernel.
697 if (user_dbreg_trap()) {
699 * Reset breakpoint bits because the
702 /* XXX check upper bits here */
703 load_dr6(rdr6() & 0xfffffff0);
708 * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
712 * If DDB is enabled, let it handle the debugger trap.
713 * Otherwise, debugger traps "can't happen".
717 MAKEMPSAFE(have_mplock);
718 if (kdb_trap(type, 0, frame))
725 MAKEMPSAFE(have_mplock);
726 /* machine/parity/power fail/"kitchen sink" faults */
727 if (isa_nmi(code) == 0) {
730 * NMI can be hooked up to a pushbutton
734 kprintf ("NMI ... going to debugger\n");
735 kdb_trap(type, 0, frame);
739 } else if (panic_on_nmi == 0)
742 #endif /* NISA > 0 */
744 MAKEMPSAFE(have_mplock);
745 trap_fatal(frame, 0);
750 * Virtual kernel intercept - if the fault is directly related to a
751 * VM context managed by a virtual kernel then let the virtual kernel
754 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
755 vkernel_trap(lp, frame);
760 * Translate fault for emulators (e.g. Linux)
762 if (*p->p_sysent->sv_transtrap)
763 i = (*p->p_sysent->sv_transtrap)(i, type);
765 MAKEMPSAFE(have_mplock);
766 trapsignal(lp, i, ucode);
769 if (type <= MAX_TRAP_MSG) {
770 uprintf("fatal process exception: %s",
772 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
773 uprintf(", fault VA = 0x%lx", frame->tf_addr);
779 userret(lp, frame, sticks);
786 if (p != NULL && lp != NULL)
787 KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid);
789 KASSERT(crit_count == td->td_critcount,
790 ("trap: critical section count mismatch! %d/%d",
791 crit_count, td->td_pri));
792 KASSERT(curstop == td->td_toks_stop,
793 ("trap: extra tokens held after trap! %ld/%ld",
794 curstop - &td->td_toks_base,
795 td->td_toks_stop - &td->td_toks_base));
800 trap_pfault(struct trapframe *frame, int usermode)
803 struct vmspace *vm = NULL;
808 thread_t td = curthread;
809 struct lwp *lp = td->td_lwp;
812 va = trunc_page(frame->tf_addr);
813 if (va >= VM_MIN_KERNEL_ADDRESS) {
815 * Don't allow user-mode faults in kernel address space.
826 * This is a fault on non-kernel virtual memory.
827 * vm is initialized above to NULL. If curproc is NULL
828 * or curproc->p_vmspace is NULL the fault is fatal.
831 vm = lp->lwp_vmspace;
843 * PGEX_I is defined only if the execute disable bit capability is
844 * supported and enabled.
846 if (frame->tf_err & PGEX_W)
847 ftype = VM_PROT_WRITE;
849 else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
850 ftype = VM_PROT_EXECUTE;
853 ftype = VM_PROT_READ;
855 if (map != &kernel_map) {
857 * Keep swapout from messing with us during this
867 fault_flags |= VM_FAULT_BURST;
868 if (ftype & VM_PROT_WRITE)
869 fault_flags |= VM_FAULT_DIRTY;
871 fault_flags |= VM_FAULT_NORMAL;
872 rv = vm_fault(map, va, ftype, fault_flags);
877 * Don't have to worry about process locking or stacks
880 fault_flags = VM_FAULT_NORMAL;
881 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
884 if (rv == KERN_SUCCESS)
888 if (td->td_gd->gd_intr_nesting_level == 0 &&
889 td->td_pcb->pcb_onfault) {
890 frame->tf_rip = (register_t)td->td_pcb->pcb_onfault;
893 trap_fatal(frame, frame->tf_addr);
898 * NOTE: on x86_64 we have a tf_addr field in the trapframe, no
899 * kludge is needed to pass the fault address to signal handlers.
902 if (td->td_lwp->lwp_vkernel == NULL) {
903 if (bootverbose || freeze_on_seg_fault || ddb_on_seg_fault) {
904 kprintf("seg-fault ft=%04x ff=%04x addr=%p rip=%p "
905 "pid=%d p_comm=%s\n",
907 (void *)frame->tf_addr,
908 (void *)frame->tf_rip,
909 p->p_pid, p->p_comm);
912 while (freeze_on_seg_fault) {
913 tsleep(p, 0, "freeze", hz * 20);
915 if (ddb_on_seg_fault)
916 Debugger("ddb_on_seg_fault");
920 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
924 trap_fatal(struct trapframe *frame, vm_offset_t eva)
929 struct soft_segment_descriptor softseg;
932 code = frame->tf_err;
933 type = frame->tf_trapno;
934 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)], &softseg);
936 if (type <= MAX_TRAP_MSG)
937 msg = trap_msg[type];
940 kprintf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
941 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
943 /* three separate prints in case of a trap on an unmapped page */
944 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
945 kprintf("lapic->id = %08x\n", lapic->id);
947 if (type == T_PAGEFLT) {
948 kprintf("fault virtual address = 0x%lx\n", eva);
949 kprintf("fault code = %s %s %s, %s\n",
950 code & PGEX_U ? "user" : "supervisor",
951 code & PGEX_W ? "write" : "read",
952 code & PGEX_I ? "instruction" : "data",
953 code & PGEX_P ? "protection violation" : "page not present");
955 kprintf("instruction pointer = 0x%lx:0x%lx\n",
956 frame->tf_cs & 0xffff, frame->tf_rip);
957 if (ISPL(frame->tf_cs) == SEL_UPL) {
958 ss = frame->tf_ss & 0xffff;
961 ss = GSEL(GDATA_SEL, SEL_KPL);
962 rsp = (long)&frame->tf_rsp;
964 kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp);
965 kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp);
966 kprintf("code segment = base 0x%lx, limit 0x%lx, type 0x%x\n",
967 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
968 kprintf(" = DPL %d, pres %d, long %d, def32 %d, gran %d\n",
969 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32,
971 kprintf("processor eflags = ");
972 if (frame->tf_rflags & PSL_T)
973 kprintf("trace trap, ");
974 if (frame->tf_rflags & PSL_I)
975 kprintf("interrupt enabled, ");
976 if (frame->tf_rflags & PSL_NT)
977 kprintf("nested task, ");
978 if (frame->tf_rflags & PSL_RF)
980 kprintf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12);
981 kprintf("current process = ");
984 (u_long)curproc->p_pid);
988 kprintf("current thread = pri %d ", curthread->td_pri);
989 if (curthread->td_critcount)
994 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
997 kprintf("trap number = %d\n", type);
998 if (type <= MAX_TRAP_MSG)
999 panic("%s", trap_msg[type]);
1001 panic("unknown/reserved trap");
1005 * Double fault handler. Called when a fault occurs while writing
1006 * a frame for a trap/exception onto the stack. This usually occurs
1007 * when the stack overflows (such is the case with infinite recursion,
1012 in_kstack_guard(register_t rptr)
1014 thread_t td = curthread;
1016 if ((char *)rptr >= td->td_kstack &&
1017 (char *)rptr < td->td_kstack + PAGE_SIZE) {
1024 dblfault_handler(struct trapframe *frame)
1026 thread_t td = curthread;
1028 if (in_kstack_guard(frame->tf_rsp) || in_kstack_guard(frame->tf_rbp)) {
1029 kprintf("DOUBLE FAULT - KERNEL STACK GUARD HIT!\n");
1030 if (in_kstack_guard(frame->tf_rsp))
1031 frame->tf_rsp = (register_t)(td->td_kstack + PAGE_SIZE);
1032 if (in_kstack_guard(frame->tf_rbp))
1033 frame->tf_rbp = (register_t)(td->td_kstack + PAGE_SIZE);
1035 kprintf("DOUBLE FAULT\n");
1037 kprintf("\nFatal double fault\n");
1038 kprintf("rip = 0x%lx\n", frame->tf_rip);
1039 kprintf("rsp = 0x%lx\n", frame->tf_rsp);
1040 kprintf("rbp = 0x%lx\n", frame->tf_rbp);
1042 /* three separate prints in case of a trap on an unmapped page */
1043 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
1044 kprintf("lapic->id = %08x\n", lapic->id);
1046 panic("double fault");
1050 * syscall2 - MP aware system call request C handler
1052 * A system call is essentially treated as a trap except that the
1053 * MP lock is not held on entry or return. We are responsible for
1054 * obtaining the MP lock if necessary and for handling ASTs
1055 * (e.g. a task switch) prior to return.
1060 syscall2(struct trapframe *frame)
1062 struct thread *td = curthread;
1063 struct proc *p = td->td_proc;
1064 struct lwp *lp = td->td_lwp;
1066 struct sysent *callp;
1067 register_t orig_tf_rflags;
1072 int crit_count = td->td_critcount;
1075 int have_mplock = 0;
1080 union sysunion args;
1081 register_t *argsdst;
1083 mycpu->gd_cnt.v_syscall++;
1086 if (ISPL(frame->tf_cs) != SEL_UPL) {
1093 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid,
1096 userenter(td, p); /* lazy raise our priority */
1103 sticks = (int)td->td_sticks;
1104 orig_tf_rflags = frame->tf_rflags;
1107 * Virtual kernel intercept - if a VM context managed by a virtual
1108 * kernel issues a system call the virtual kernel handles it, not us.
1109 * Restore the virtual kernel context and return from its system
1110 * call. The current frame is copied out to the virtual kernel.
1112 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1113 vkernel_trap(lp, frame);
1114 error = EJUSTRETURN;
1119 * Get the system call parameters and account for time
1121 KASSERT(lp->lwp_md.md_regs == frame,
1122 ("Frame mismatch %p %p", lp->lwp_md.md_regs, frame));
1123 params = (caddr_t)frame->tf_rsp + sizeof(register_t);
1124 code = frame->tf_rax;
1126 if (p->p_sysent->sv_prepsyscall) {
1127 (*p->p_sysent->sv_prepsyscall)(
1128 frame, (int *)(&args.nosys.sysmsg + 1),
1131 if (code == SYS_syscall || code == SYS___syscall) {
1132 code = frame->tf_rdi;
1138 if (p->p_sysent->sv_mask)
1139 code &= p->p_sysent->sv_mask;
1141 if (code >= p->p_sysent->sv_size)
1142 callp = &p->p_sysent->sv_table[0];
1144 callp = &p->p_sysent->sv_table[code];
1146 narg = callp->sy_narg & SYF_ARGMASK;
1149 * On x86_64 we get up to six arguments in registers. The rest are
1150 * on the stack. The first six members of 'struct trapframe' happen
1151 * to be the registers used to pass arguments, in exactly the right
1154 argp = &frame->tf_rdi;
1156 argsdst = (register_t *)(&args.nosys.sysmsg + 1);
1158 * JG can we overflow the space pointed to by 'argsdst'
1159 * either with 'bcopy' or with 'copyin'?
1161 bcopy(argp, argsdst, sizeof(register_t) * regcnt);
1163 * copyin is MP aware, but the tracing code is not
1165 if (narg > regcnt) {
1166 KASSERT(params != NULL, ("copyin args with no params!"));
1167 error = copyin(params, &argsdst[regcnt],
1168 (narg - regcnt) * sizeof(register_t));
1171 if (KTRPOINT(td, KTR_SYSCALL)) {
1172 MAKEMPSAFE(have_mplock);
1174 ktrsyscall(lp, code, narg,
1175 (void *)(&args.nosys.sysmsg + 1));
1183 if (KTRPOINT(td, KTR_SYSCALL)) {
1184 MAKEMPSAFE(have_mplock);
1185 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1190 * Default return value is 0 (will be copied to %rax). Double-value
1191 * returns use %rax and %rdx. %rdx is left unchanged for system
1192 * calls which return only one result.
1194 args.sysmsg_fds[0] = 0;
1195 args.sysmsg_fds[1] = frame->tf_rdx;
1198 * The syscall might manipulate the trap frame. If it does it
1199 * will probably return EJUSTRETURN.
1201 args.sysmsg_frame = frame;
1203 STOPEVENT(p, S_SCE, narg); /* MP aware */
1206 * NOTE: All system calls run MPSAFE now. The system call itself
1207 * is responsible for getting the MP lock.
1209 #ifdef SYSCALL_DEBUG
1210 uint64_t tscval = rdtsc();
1212 error = (*callp->sy_call)(&args);
1213 #ifdef SYSCALL_DEBUG
1214 tscval = rdtsc() - tscval;
1215 tscval = tscval * 1000000 / tsc_frequency;
1216 if (SysCallsWorstCase[code] < tscval)
1217 SysCallsWorstCase[code] = tscval;
1222 * MP SAFE (we may or may not have the MP lock at this point)
1224 //kprintf("SYSMSG %d ", error);
1228 * Reinitialize proc pointer `p' as it may be different
1229 * if this is a child returning from fork syscall.
1232 lp = curthread->td_lwp;
1233 frame->tf_rax = args.sysmsg_fds[0];
1234 frame->tf_rdx = args.sysmsg_fds[1];
1235 frame->tf_rflags &= ~PSL_C;
1239 * Reconstruct pc, we know that 'syscall' is 2 bytes.
1240 * We have to do a full context restore so that %r10
1241 * (which was holding the value of %rcx) is restored for
1242 * the next iteration.
1244 frame->tf_rip -= frame->tf_err;
1245 frame->tf_r10 = frame->tf_rcx;
1250 panic("Unexpected EASYNC return value (for now)");
1253 if (p->p_sysent->sv_errsize) {
1254 if (error >= p->p_sysent->sv_errsize)
1255 error = -1; /* XXX */
1257 error = p->p_sysent->sv_errtbl[error];
1259 frame->tf_rax = error;
1260 frame->tf_rflags |= PSL_C;
1265 * Traced syscall. trapsignal() is not MP aware.
1267 if (orig_tf_rflags & PSL_T) {
1268 MAKEMPSAFE(have_mplock);
1269 frame->tf_rflags &= ~PSL_T;
1270 trapsignal(lp, SIGTRAP, TRAP_TRACE);
1274 * Handle reschedule and other end-of-syscall issues
1276 userret(lp, frame, sticks);
1279 if (KTRPOINT(td, KTR_SYSRET)) {
1280 MAKEMPSAFE(have_mplock);
1281 ktrsysret(lp, code, error, args.sysmsg_result);
1286 * This works because errno is findable through the
1287 * register set. If we ever support an emulation where this
1288 * is not the case, this code will need to be revisited.
1290 STOPEVENT(p, S_SCX, code);
1295 * Release the MP lock if we had to get it
1300 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error);
1302 KASSERT(crit_count == td->td_critcount,
1303 ("syscall: critical section count mismatch! %d/%d",
1304 crit_count, td->td_pri));
1305 KASSERT(&td->td_toks_base == td->td_toks_stop,
1306 ("syscall: extra tokens held after trap! %ld",
1307 td->td_toks_stop - &td->td_toks_base));
1312 * NOTE: mplock not held at any point
1315 fork_return(struct lwp *lp, struct trapframe *frame)
1317 frame->tf_rax = 0; /* Child returns zero */
1318 frame->tf_rflags &= ~PSL_C; /* success */
1321 generic_lwp_return(lp, frame);
1322 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1326 * Simplified back end of syscall(), used when returning from fork()
1327 * directly into user mode.
1329 * This code will return back into the fork trampoline code which then
1332 * NOTE: The mplock is not held at any point.
1335 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1337 struct proc *p = lp->lwp_proc;
1340 * Newly forked processes are given a kernel priority. We have to
1341 * adjust the priority to a normal user priority and fake entry
1342 * into the kernel (call userenter()) to install a passive release
1343 * function just in case userret() decides to stop the process. This
1344 * can occur when ^Z races a fork. If we do not install the passive
1345 * release function the current process designation will not be
1346 * released when the thread goes to sleep.
1348 lwkt_setpri_self(TDPRI_USER_NORM);
1349 userenter(lp->lwp_thread, p);
1350 userret(lp, frame, 0);
1352 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1353 ktrsysret(lp, SYS_fork, 0, 0);
1355 p->p_flag |= P_PASSIVE_ACQ;
1357 p->p_flag &= ~P_PASSIVE_ACQ;
1361 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1362 * fault (which is then passed back to the virtual kernel) if an attempt is
1363 * made to use the FP unit.
1365 * XXX this is a fairly big hack.
1368 set_vkernel_fp(struct trapframe *frame)
1370 struct thread *td = curthread;
1372 if (frame->tf_xflags & PGEX_FPFAULT) {
1373 td->td_pcb->pcb_flags |= FP_VIRTFP;
1374 if (mdcpu->gd_npxthread == td)
1377 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1382 * Called from vkernel_trap() to fixup the vkernel's syscall
1383 * frame for vmspace_ctl() return.
1386 cpu_vkernel_trap(struct trapframe *frame, int error)
1388 frame->tf_rax = error;
1390 frame->tf_rflags |= PSL_C;
1392 frame->tf_rflags &= ~PSL_C;