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/vmparam.h>
83 #include <machine/md_var.h>
84 #include <machine_base/isa/isa_intr.h>
85 #include <machine_base/apic/lapic.h>
89 #include <sys/thread2.h>
90 #include <sys/mplock2.h>
94 #define MAKEMPSAFE(have_mplock) \
95 if (have_mplock == 0) { \
102 #define MAKEMPSAFE(have_mplock)
106 extern void trap(struct trapframe *frame);
108 static int trap_pfault(struct trapframe *, int);
109 static void trap_fatal(struct trapframe *, vm_offset_t);
110 void dblfault_handler(struct trapframe *frame);
112 #define MAX_TRAP_MSG 30
113 static char *trap_msg[] = {
115 "privileged instruction fault", /* 1 T_PRIVINFLT */
117 "breakpoint instruction fault", /* 3 T_BPTFLT */
120 "arithmetic trap", /* 6 T_ARITHTRAP */
121 "system forced exception", /* 7 T_ASTFLT */
123 "general protection fault", /* 9 T_PROTFLT */
124 "trace trap", /* 10 T_TRCTRAP */
126 "page fault", /* 12 T_PAGEFLT */
128 "alignment fault", /* 14 T_ALIGNFLT */
132 "integer divide fault", /* 18 T_DIVIDE */
133 "non-maskable interrupt trap", /* 19 T_NMI */
134 "overflow trap", /* 20 T_OFLOW */
135 "FPU bounds check fault", /* 21 T_BOUND */
136 "FPU device not available", /* 22 T_DNA */
137 "double fault", /* 23 T_DOUBLEFLT */
138 "FPU operand fetch fault", /* 24 T_FPOPFLT */
139 "invalid TSS fault", /* 25 T_TSSFLT */
140 "segment not present fault", /* 26 T_SEGNPFLT */
141 "stack fault", /* 27 T_STKFLT */
142 "machine check trap", /* 28 T_MCHK */
143 "SIMD floating-point exception", /* 29 T_XMMFLT */
144 "reserved (unknown) fault", /* 30 T_RESERVED */
148 static int ddb_on_nmi = 1;
149 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
150 &ddb_on_nmi, 0, "Go to DDB on NMI");
151 static int ddb_on_seg_fault = 0;
152 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_seg_fault, CTLFLAG_RW,
153 &ddb_on_seg_fault, 0, "Go to DDB on user seg-fault");
155 static int panic_on_nmi = 1;
156 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
157 &panic_on_nmi, 0, "Panic on NMI");
158 static int fast_release;
159 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
160 &fast_release, 0, "Passive Release was optimal");
161 static int slow_release;
162 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
163 &slow_release, 0, "Passive Release was nonoptimal");
166 * Passively intercepts the thread switch function to increase
167 * the thread priority from a user priority to a kernel priority, reducing
168 * syscall and trap overhead for the case where no switch occurs.
170 * Synchronizes td_ucred with p_ucred. This is used by system calls,
171 * signal handling, faults, AST traps, and anything else that enters the
172 * kernel from userland and provides the kernel with a stable read-only
173 * copy of the process ucred.
176 userenter(struct thread *curtd, struct proc *curp)
181 curtd->td_release = lwkt_passive_release;
183 if (curtd->td_ucred != curp->p_ucred) {
184 ncred = crhold(curp->p_ucred);
185 ocred = curtd->td_ucred;
186 curtd->td_ucred = ncred;
193 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
194 * must be completed before we can return to or try to return to userland.
196 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
197 * arithmatic on the delta calculation so the absolute tick values are
198 * truncated to an integer.
201 userret(struct lwp *lp, struct trapframe *frame, int sticks)
203 struct proc *p = lp->lwp_proc;
207 * Charge system time if profiling. Note: times are in microseconds.
208 * This may do a copyout and block, so do it first even though it
209 * means some system time will be charged as user time.
211 if (p->p_flag & P_PROFIL) {
212 addupc_task(p, frame->tf_rip,
213 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
218 * If the jungle wants us dead, so be it.
220 if (lp->lwp_flag & LWP_WEXIT) {
221 lwkt_gettoken(&p->p_token);
223 lwkt_reltoken(&p->p_token); /* NOT REACHED */
227 * Block here if we are in a stopped state.
229 if (p->p_stat == SSTOP || dump_stop_usertds) {
237 * Post any pending upcalls. If running a virtual kernel be sure
238 * to restore the virtual kernel's vmspace before posting the upcall.
240 if (p->p_flag & P_UPCALLPEND) {
241 p->p_flag &= ~P_UPCALLPEND;
249 * Post any pending signals. If running a virtual kernel be sure
250 * to restore the virtual kernel's vmspace before posting the signal.
252 * WARNING! postsig() can exit and not return.
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_critcount;
369 lwkt_tokref_t curstop = td->td_toks_stop;
377 * We need to allow T_DNA faults when the debugger is active since
378 * some dumping paths do large bcopy() which use the floating
379 * point registers for faster copying.
381 if (db_active && frame->tf_trapno != T_DNA) {
382 eva = (frame->tf_trapno == T_PAGEFLT ? frame->tf_addr : 0);
383 ++gd->gd_trap_nesting_level;
384 MAKEMPSAFE(have_mplock);
385 trap_fatal(frame, eva);
386 --gd->gd_trap_nesting_level;
393 if ((frame->tf_rflags & PSL_I) == 0) {
395 * Buggy application or kernel code has disabled interrupts
396 * and then trapped. Enabling interrupts now is wrong, but
397 * it is better than running with interrupts disabled until
398 * they are accidentally enabled later.
400 type = frame->tf_trapno;
401 if (ISPL(frame->tf_cs) == SEL_UPL) {
402 MAKEMPSAFE(have_mplock);
403 /* JG curproc can be NULL */
405 "pid %ld (%s): trap %d with interrupts disabled\n",
406 (long)curproc->p_pid, curproc->p_comm, type);
407 } else if (type != T_NMI && type != T_BPTFLT &&
410 * XXX not quite right, since this may be for a
411 * multiple fault in user mode.
413 MAKEMPSAFE(have_mplock);
414 kprintf("kernel trap %d with interrupts disabled\n",
420 type = frame->tf_trapno;
421 code = frame->tf_err;
423 if (ISPL(frame->tf_cs) == SEL_UPL) {
426 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid,
427 frame->tf_trapno, eva);
431 sticks = (int)td->td_sticks;
432 KASSERT(lp->lwp_md.md_regs == frame,
433 ("Frame mismatch %p %p", 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(&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 i = trap_pfault(frame, TRUE);
493 if (frame->tf_rip == 0)
494 kprintf("T_PAGEFLT: Warning %%rip == 0!\n");
509 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");
533 #endif /* NISA > 0 */
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 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 if (mycpu->gd_intr_nesting_level == 0) {
620 if (td->td_pcb->pcb_onfault) {
621 frame->tf_rip = (register_t)
622 td->td_pcb->pcb_onfault;
625 if (frame->tf_rip == (long)doreti_iret) {
626 frame->tf_rip = (long)doreti_iret_fault;
634 * PSL_NT can be set in user mode and isn't cleared
635 * automatically when the kernel is entered. This
636 * causes a TSS fault when the kernel attempts to
637 * `iret' because the TSS link is uninitialized. We
638 * want to get this fault so that we can fix the
639 * problem here and not every time the kernel is
642 if (frame->tf_rflags & PSL_NT) {
643 frame->tf_rflags &= ~PSL_NT;
648 case T_TRCTRAP: /* trace trap */
650 if (frame->tf_rip == (int)IDTVEC(syscall)) {
652 * We've just entered system mode via the
653 * syscall lcall. Continue single stepping
654 * silently until the syscall handler has
659 if (frame->tf_rip == (int)IDTVEC(syscall) + 1) {
661 * The syscall handler has now saved the
662 * flags. Stop single stepping it.
664 frame->tf_rflags &= ~PSL_T;
670 * Ignore debug register trace traps due to
671 * accesses in the user's address space, which
672 * can happen under several conditions such as
673 * if a user sets a watchpoint on a buffer and
674 * then passes that buffer to a system call.
675 * We still want to get TRCTRAPS for addresses
676 * in kernel space because that is useful when
677 * debugging the kernel.
680 if (user_dbreg_trap()) {
682 * Reset breakpoint bits because the
685 /* XXX check upper bits here */
686 load_dr6(rdr6() & 0xfffffff0);
691 * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
695 * If DDB is enabled, let it handle the debugger trap.
696 * Otherwise, debugger traps "can't happen".
700 MAKEMPSAFE(have_mplock);
701 if (kdb_trap(type, 0, frame))
708 MAKEMPSAFE(have_mplock);
709 /* machine/parity/power fail/"kitchen sink" faults */
710 if (isa_nmi(code) == 0) {
713 * NMI can be hooked up to a pushbutton
717 kprintf ("NMI ... going to debugger\n");
718 kdb_trap(type, 0, frame);
722 } else if (panic_on_nmi == 0)
725 #endif /* NISA > 0 */
727 MAKEMPSAFE(have_mplock);
728 trap_fatal(frame, 0);
733 * Virtual kernel intercept - if the fault is directly related to a
734 * VM context managed by a virtual kernel then let the virtual kernel
737 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
738 vkernel_trap(lp, frame);
743 * Translate fault for emulators (e.g. Linux)
745 if (*p->p_sysent->sv_transtrap)
746 i = (*p->p_sysent->sv_transtrap)(i, type);
748 MAKEMPSAFE(have_mplock);
749 trapsignal(lp, i, ucode);
752 if (type <= MAX_TRAP_MSG) {
753 uprintf("fatal process exception: %s",
755 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
756 uprintf(", fault VA = 0x%lx", frame->tf_addr);
762 userret(lp, frame, sticks);
769 if (p != NULL && lp != NULL)
770 KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid);
772 KASSERT(crit_count == td->td_critcount,
773 ("trap: critical section count mismatch! %d/%d",
774 crit_count, td->td_pri));
775 KASSERT(curstop == td->td_toks_stop,
776 ("trap: extra tokens held after trap! %ld/%ld",
777 curstop - &td->td_toks_base,
778 td->td_toks_stop - &td->td_toks_base));
783 trap_pfault(struct trapframe *frame, int usermode)
786 struct vmspace *vm = NULL;
791 thread_t td = curthread;
792 struct lwp *lp = td->td_lwp;
795 va = trunc_page(frame->tf_addr);
796 if (va >= VM_MIN_KERNEL_ADDRESS) {
798 * Don't allow user-mode faults in kernel address space.
809 * This is a fault on non-kernel virtual memory.
810 * vm is initialized above to NULL. If curproc is NULL
811 * or curproc->p_vmspace is NULL the fault is fatal.
814 vm = lp->lwp_vmspace;
826 * PGEX_I is defined only if the execute disable bit capability is
827 * supported and enabled.
829 if (frame->tf_err & PGEX_W)
830 ftype = VM_PROT_WRITE;
832 else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
833 ftype = VM_PROT_EXECUTE;
836 ftype = VM_PROT_READ;
838 if (map != &kernel_map) {
840 * Keep swapout from messing with us during this
850 fault_flags |= VM_FAULT_BURST;
851 if (ftype & VM_PROT_WRITE)
852 fault_flags |= VM_FAULT_DIRTY;
854 fault_flags |= VM_FAULT_NORMAL;
855 rv = vm_fault(map, va, ftype, fault_flags);
860 * Don't have to worry about process locking or stacks
863 fault_flags = VM_FAULT_NORMAL;
864 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
867 if (rv == KERN_SUCCESS)
871 if (td->td_gd->gd_intr_nesting_level == 0 &&
872 td->td_pcb->pcb_onfault) {
873 frame->tf_rip = (register_t)td->td_pcb->pcb_onfault;
876 trap_fatal(frame, frame->tf_addr);
881 * NOTE: on x86_64 we have a tf_addr field in the trapframe, no
882 * kludge is needed to pass the fault address to signal handlers.
885 if (td->td_lwp->lwp_vkernel == NULL) {
887 kprintf("seg-fault ft=%04x ff=%04x addr=%p rip=%p "
888 "pid=%d p_comm=%s\n",
890 (void *)frame->tf_addr,
891 (void *)frame->tf_rip,
892 p->p_pid, p->p_comm);
894 if (ddb_on_seg_fault)
895 Debugger("ddb_on_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("cpuid = %d; ", mycpu->gd_cpuid);
924 kprintf("lapic->id = %08x\n", lapic->id);
926 if (type == T_PAGEFLT) {
927 kprintf("fault virtual address = 0x%lx\n", eva);
928 kprintf("fault code = %s %s %s, %s\n",
929 code & PGEX_U ? "user" : "supervisor",
930 code & PGEX_W ? "write" : "read",
931 code & PGEX_I ? "instruction" : "data",
932 code & PGEX_P ? "protection violation" : "page not present");
934 kprintf("instruction pointer = 0x%lx:0x%lx\n",
935 frame->tf_cs & 0xffff, frame->tf_rip);
936 if (ISPL(frame->tf_cs) == SEL_UPL) {
937 ss = frame->tf_ss & 0xffff;
940 ss = GSEL(GDATA_SEL, SEL_KPL);
941 rsp = (long)&frame->tf_rsp;
943 kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp);
944 kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp);
945 kprintf("code segment = base 0x%lx, limit 0x%lx, type 0x%x\n",
946 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
947 kprintf(" = DPL %d, pres %d, long %d, def32 %d, gran %d\n",
948 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32,
950 kprintf("processor eflags = ");
951 if (frame->tf_rflags & PSL_T)
952 kprintf("trace trap, ");
953 if (frame->tf_rflags & PSL_I)
954 kprintf("interrupt enabled, ");
955 if (frame->tf_rflags & PSL_NT)
956 kprintf("nested task, ");
957 if (frame->tf_rflags & PSL_RF)
959 kprintf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12);
960 kprintf("current process = ");
963 (u_long)curproc->p_pid);
967 kprintf("current thread = pri %d ", curthread->td_pri);
968 if (curthread->td_critcount)
973 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
976 kprintf("trap number = %d\n", type);
977 if (type <= MAX_TRAP_MSG)
978 panic("%s", trap_msg[type]);
980 panic("unknown/reserved trap");
984 * Double fault handler. Called when a fault occurs while writing
985 * a frame for a trap/exception onto the stack. This usually occurs
986 * when the stack overflows (such is the case with infinite recursion,
991 in_kstack_guard(register_t rptr)
993 thread_t td = curthread;
995 if ((char *)rptr >= td->td_kstack &&
996 (char *)rptr < td->td_kstack + PAGE_SIZE) {
1003 dblfault_handler(struct trapframe *frame)
1005 thread_t td = curthread;
1007 if (in_kstack_guard(frame->tf_rsp) || in_kstack_guard(frame->tf_rbp)) {
1008 kprintf("DOUBLE FAULT - KERNEL STACK GUARD HIT!\n");
1009 if (in_kstack_guard(frame->tf_rsp))
1010 frame->tf_rsp = (register_t)(td->td_kstack + PAGE_SIZE);
1011 if (in_kstack_guard(frame->tf_rbp))
1012 frame->tf_rbp = (register_t)(td->td_kstack + PAGE_SIZE);
1014 kprintf("DOUBLE FAULT\n");
1016 kprintf("\nFatal double fault\n");
1017 kprintf("rip = 0x%lx\n", frame->tf_rip);
1018 kprintf("rsp = 0x%lx\n", frame->tf_rsp);
1019 kprintf("rbp = 0x%lx\n", frame->tf_rbp);
1021 /* three separate prints in case of a trap on an unmapped page */
1022 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
1023 kprintf("lapic->id = %08x\n", lapic->id);
1025 panic("double fault");
1029 * syscall2 - MP aware system call request C handler
1031 * A system call is essentially treated as a trap except that the
1032 * MP lock is not held on entry or return. We are responsible for
1033 * obtaining the MP lock if necessary and for handling ASTs
1034 * (e.g. a task switch) prior to return.
1039 syscall2(struct trapframe *frame)
1041 struct thread *td = curthread;
1042 struct proc *p = td->td_proc;
1043 struct lwp *lp = td->td_lwp;
1045 struct sysent *callp;
1046 register_t orig_tf_rflags;
1051 int crit_count = td->td_critcount;
1054 int have_mplock = 0;
1059 union sysunion args;
1060 register_t *argsdst;
1062 mycpu->gd_cnt.v_syscall++;
1065 if (ISPL(frame->tf_cs) != SEL_UPL) {
1072 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid,
1075 userenter(td, p); /* lazy raise our priority */
1082 sticks = (int)td->td_sticks;
1083 orig_tf_rflags = frame->tf_rflags;
1086 * Virtual kernel intercept - if a VM context managed by a virtual
1087 * kernel issues a system call the virtual kernel handles it, not us.
1088 * Restore the virtual kernel context and return from its system
1089 * call. The current frame is copied out to the virtual kernel.
1091 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1092 vkernel_trap(lp, frame);
1093 error = EJUSTRETURN;
1098 * Get the system call parameters and account for time
1100 KASSERT(lp->lwp_md.md_regs == frame,
1101 ("Frame mismatch %p %p", lp->lwp_md.md_regs, frame));
1102 params = (caddr_t)frame->tf_rsp + sizeof(register_t);
1103 code = frame->tf_rax;
1105 if (p->p_sysent->sv_prepsyscall) {
1106 (*p->p_sysent->sv_prepsyscall)(
1107 frame, (int *)(&args.nosys.sysmsg + 1),
1110 if (code == SYS_syscall || code == SYS___syscall) {
1111 code = frame->tf_rdi;
1117 if (p->p_sysent->sv_mask)
1118 code &= p->p_sysent->sv_mask;
1120 if (code >= p->p_sysent->sv_size)
1121 callp = &p->p_sysent->sv_table[0];
1123 callp = &p->p_sysent->sv_table[code];
1125 narg = callp->sy_narg & SYF_ARGMASK;
1128 * On x86_64 we get up to six arguments in registers. The rest are
1129 * on the stack. The first six members of 'struct trapframe' happen
1130 * to be the registers used to pass arguments, in exactly the right
1133 argp = &frame->tf_rdi;
1135 argsdst = (register_t *)(&args.nosys.sysmsg + 1);
1137 * JG can we overflow the space pointed to by 'argsdst'
1138 * either with 'bcopy' or with 'copyin'?
1140 bcopy(argp, argsdst, sizeof(register_t) * regcnt);
1142 * copyin is MP aware, but the tracing code is not
1144 if (narg > regcnt) {
1145 KASSERT(params != NULL, ("copyin args with no params!"));
1146 error = copyin(params, &argsdst[regcnt],
1147 (narg - regcnt) * sizeof(register_t));
1150 if (KTRPOINT(td, KTR_SYSCALL)) {
1151 MAKEMPSAFE(have_mplock);
1153 ktrsyscall(lp, code, narg,
1154 (void *)(&args.nosys.sysmsg + 1));
1162 if (KTRPOINT(td, KTR_SYSCALL)) {
1163 MAKEMPSAFE(have_mplock);
1164 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1169 * Default return value is 0 (will be copied to %rax). Double-value
1170 * returns use %rax and %rdx. %rdx is left unchanged for system
1171 * calls which return only one result.
1173 args.sysmsg_fds[0] = 0;
1174 args.sysmsg_fds[1] = frame->tf_rdx;
1177 * The syscall might manipulate the trap frame. If it does it
1178 * will probably return EJUSTRETURN.
1180 args.sysmsg_frame = frame;
1182 STOPEVENT(p, S_SCE, narg); /* MP aware */
1185 * NOTE: All system calls run MPSAFE now. The system call itself
1186 * is responsible for getting the MP lock.
1188 error = (*callp->sy_call)(&args);
1192 * MP SAFE (we may or may not have the MP lock at this point)
1194 //kprintf("SYSMSG %d ", error);
1198 * Reinitialize proc pointer `p' as it may be different
1199 * if this is a child returning from fork syscall.
1202 lp = curthread->td_lwp;
1203 frame->tf_rax = args.sysmsg_fds[0];
1204 frame->tf_rdx = args.sysmsg_fds[1];
1205 frame->tf_rflags &= ~PSL_C;
1209 * Reconstruct pc, we know that 'syscall' is 2 bytes.
1210 * We have to do a full context restore so that %r10
1211 * (which was holding the value of %rcx) is restored for
1212 * the next iteration.
1214 frame->tf_rip -= frame->tf_err;
1215 frame->tf_r10 = frame->tf_rcx;
1220 panic("Unexpected EASYNC return value (for now)");
1223 if (p->p_sysent->sv_errsize) {
1224 if (error >= p->p_sysent->sv_errsize)
1225 error = -1; /* XXX */
1227 error = p->p_sysent->sv_errtbl[error];
1229 frame->tf_rax = error;
1230 frame->tf_rflags |= PSL_C;
1235 * Traced syscall. trapsignal() is not MP aware.
1237 if (orig_tf_rflags & PSL_T) {
1238 MAKEMPSAFE(have_mplock);
1239 frame->tf_rflags &= ~PSL_T;
1240 trapsignal(lp, SIGTRAP, TRAP_TRACE);
1244 * Handle reschedule and other end-of-syscall issues
1246 userret(lp, frame, sticks);
1249 if (KTRPOINT(td, KTR_SYSRET)) {
1250 MAKEMPSAFE(have_mplock);
1251 ktrsysret(lp, code, error, args.sysmsg_result);
1256 * This works because errno is findable through the
1257 * register set. If we ever support an emulation where this
1258 * is not the case, this code will need to be revisited.
1260 STOPEVENT(p, S_SCX, code);
1265 * Release the MP lock if we had to get it
1270 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error);
1272 KASSERT(crit_count == td->td_critcount,
1273 ("syscall: critical section count mismatch! %d/%d",
1274 crit_count, td->td_pri));
1275 KASSERT(&td->td_toks_base == td->td_toks_stop,
1276 ("syscall: extra tokens held after trap! %ld",
1277 td->td_toks_stop - &td->td_toks_base));
1282 * NOTE: mplock not held at any point
1285 fork_return(struct lwp *lp, struct trapframe *frame)
1287 frame->tf_rax = 0; /* Child returns zero */
1288 frame->tf_rflags &= ~PSL_C; /* success */
1291 generic_lwp_return(lp, frame);
1292 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1296 * Simplified back end of syscall(), used when returning from fork()
1297 * directly into user mode.
1299 * This code will return back into the fork trampoline code which then
1302 * NOTE: The mplock is not held at any point.
1305 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1307 struct proc *p = lp->lwp_proc;
1310 * Newly forked processes are given a kernel priority. We have to
1311 * adjust the priority to a normal user priority and fake entry
1312 * into the kernel (call userenter()) to install a passive release
1313 * function just in case userret() decides to stop the process. This
1314 * can occur when ^Z races a fork. If we do not install the passive
1315 * release function the current process designation will not be
1316 * released when the thread goes to sleep.
1318 lwkt_setpri_self(TDPRI_USER_NORM);
1319 userenter(lp->lwp_thread, p);
1320 userret(lp, frame, 0);
1322 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1323 ktrsysret(lp, SYS_fork, 0, 0);
1325 p->p_flag |= P_PASSIVE_ACQ;
1327 p->p_flag &= ~P_PASSIVE_ACQ;
1331 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1332 * fault (which is then passed back to the virtual kernel) if an attempt is
1333 * made to use the FP unit.
1335 * XXX this is a fairly big hack.
1338 set_vkernel_fp(struct trapframe *frame)
1340 struct thread *td = curthread;
1342 if (frame->tf_xflags & PGEX_FPFAULT) {
1343 td->td_pcb->pcb_flags |= FP_VIRTFP;
1344 if (mdcpu->gd_npxthread == td)
1347 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1352 * Called from vkernel_trap() to fixup the vkernel's syscall
1353 * frame for vmspace_ctl() return.
1356 cpu_vkernel_trap(struct trapframe *frame, int error)
1358 frame->tf_rax = error;
1360 frame->tf_rflags |= PSL_C;
1362 frame->tf_rflags &= ~PSL_C;