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>
88 #include <sys/thread2.h>
89 #include <sys/mplock2.h>
93 #define MAKEMPSAFE(have_mplock) \
94 if (have_mplock == 0) { \
101 #define MAKEMPSAFE(have_mplock)
105 extern void trap(struct trapframe *frame);
107 static int trap_pfault(struct trapframe *, int);
108 static void trap_fatal(struct trapframe *, vm_offset_t);
109 void dblfault_handler(struct trapframe *frame);
111 #define MAX_TRAP_MSG 30
112 static char *trap_msg[] = {
114 "privileged instruction fault", /* 1 T_PRIVINFLT */
116 "breakpoint instruction fault", /* 3 T_BPTFLT */
119 "arithmetic trap", /* 6 T_ARITHTRAP */
120 "system forced exception", /* 7 T_ASTFLT */
122 "general protection fault", /* 9 T_PROTFLT */
123 "trace trap", /* 10 T_TRCTRAP */
125 "page fault", /* 12 T_PAGEFLT */
127 "alignment fault", /* 14 T_ALIGNFLT */
131 "integer divide fault", /* 18 T_DIVIDE */
132 "non-maskable interrupt trap", /* 19 T_NMI */
133 "overflow trap", /* 20 T_OFLOW */
134 "FPU bounds check fault", /* 21 T_BOUND */
135 "FPU device not available", /* 22 T_DNA */
136 "double fault", /* 23 T_DOUBLEFLT */
137 "FPU operand fetch fault", /* 24 T_FPOPFLT */
138 "invalid TSS fault", /* 25 T_TSSFLT */
139 "segment not present fault", /* 26 T_SEGNPFLT */
140 "stack fault", /* 27 T_STKFLT */
141 "machine check trap", /* 28 T_MCHK */
142 "SIMD floating-point exception", /* 29 T_XMMFLT */
143 "reserved (unknown) fault", /* 30 T_RESERVED */
147 static int ddb_on_nmi = 1;
148 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
149 &ddb_on_nmi, 0, "Go to DDB on NMI");
150 static int ddb_on_seg_fault = 0;
151 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_seg_fault, CTLFLAG_RW,
152 &ddb_on_seg_fault, 0, "Go to DDB on user seg-fault");
154 static int panic_on_nmi = 1;
155 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
156 &panic_on_nmi, 0, "Panic on NMI");
157 static int fast_release;
158 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
159 &fast_release, 0, "Passive Release was optimal");
160 static int slow_release;
161 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
162 &slow_release, 0, "Passive Release was nonoptimal");
165 * Passively intercepts the thread switch function to increase
166 * the thread priority from a user priority to a kernel priority, reducing
167 * syscall and trap overhead for the case where no switch occurs.
169 * Synchronizes td_ucred with p_ucred. This is used by system calls,
170 * signal handling, faults, AST traps, and anything else that enters the
171 * kernel from userland and provides the kernel with a stable read-only
172 * copy of the process ucred.
175 userenter(struct thread *curtd, struct proc *curp)
180 curtd->td_release = lwkt_passive_release;
182 if (curtd->td_ucred != curp->p_ucred) {
183 ncred = crhold(curp->p_ucred);
184 ocred = curtd->td_ucred;
185 curtd->td_ucred = ncred;
192 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
193 * must be completed before we can return to or try to return to userland.
195 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
196 * arithmatic on the delta calculation so the absolute tick values are
197 * truncated to an integer.
200 userret(struct lwp *lp, struct trapframe *frame, int sticks)
202 struct proc *p = lp->lwp_proc;
206 * Charge system time if profiling. Note: times are in microseconds.
207 * This may do a copyout and block, so do it first even though it
208 * means some system time will be charged as user time.
210 if (p->p_flag & P_PROFIL) {
211 addupc_task(p, frame->tf_rip,
212 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
217 * If the jungle wants us dead, so be it.
219 if (lp->lwp_flag & LWP_WEXIT) {
220 lwkt_gettoken(&p->p_token);
222 lwkt_reltoken(&p->p_token); /* NOT REACHED */
226 * Block here if we are in a stopped state.
228 if (p->p_stat == SSTOP || dump_stop_usertds) {
236 * Post any pending upcalls. If running a virtual kernel be sure
237 * to restore the virtual kernel's vmspace before posting the upcall.
239 if (p->p_flag & P_UPCALLPEND) {
240 p->p_flag &= ~P_UPCALLPEND;
248 * Post any pending signals. If running a virtual kernel be sure
249 * to restore the virtual kernel's vmspace before posting the signal.
251 * WARNING! postsig() can exit and not return.
253 if ((sig = CURSIG_TRACE(lp)) != 0) {
261 * block here if we are swapped out, but still process signals
262 * (such as SIGKILL). proc0 (the swapin scheduler) is already
263 * aware of our situation, we do not have to wake it up.
265 if (p->p_flag & P_SWAPPEDOUT) {
267 p->p_flag |= P_SWAPWAIT;
269 if (p->p_flag & P_SWAPWAIT)
270 tsleep(p, PCATCH, "SWOUT", 0);
271 p->p_flag &= ~P_SWAPWAIT;
277 * Make sure postsig() handled request to restore old signal mask after
278 * running signal handler.
280 KKASSERT((lp->lwp_flag & LWP_OLDMASK) == 0);
284 * Cleanup from userenter and any passive release that might have occured.
285 * We must reclaim the current-process designation before we can return
286 * to usermode. We also handle both LWKT and USER reschedule requests.
289 userexit(struct lwp *lp)
291 struct thread *td = lp->lwp_thread;
292 /* globaldata_t gd = td->td_gd;*/
295 * Handle stop requests at kernel priority. Any requests queued
296 * after this loop will generate another AST.
298 while (lp->lwp_proc->p_stat == SSTOP) {
305 * Reduce our priority in preparation for a return to userland. If
306 * our passive release function was still in place, our priority was
307 * never raised and does not need to be reduced.
309 lwkt_passive_recover(td);
312 * Become the current user scheduled process if we aren't already,
313 * and deal with reschedule requests and other factors.
315 lp->lwp_proc->p_usched->acquire_curproc(lp);
316 /* WARNING: we may have migrated cpu's */
317 /* gd = td->td_gd; */
320 #if !defined(KTR_KERNENTRY)
321 #define KTR_KERNENTRY KTR_ALL
323 KTR_INFO_MASTER(kernentry);
324 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, "STR",
325 sizeof(long) + sizeof(long) + sizeof(long) + sizeof(vm_offset_t));
326 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "STR",
327 sizeof(long) + sizeof(long));
328 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "STR",
329 sizeof(long) + sizeof(long) + sizeof(long));
330 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "STR",
331 sizeof(long) + sizeof(long) + sizeof(long));
332 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "STR",
333 sizeof(long) + sizeof(long));
336 * Exception, fault, and trap interface to the kernel.
337 * This common code is called from assembly language IDT gate entry
338 * routines that prepare a suitable stack frame, and restore this
339 * frame after the exception has been processed.
341 * This function is also called from doreti in an interlock to handle ASTs.
342 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
344 * NOTE! We have to retrieve the fault address prior to obtaining the
345 * MP lock because get_mplock() may switch out. YYY cr2 really ought
346 * to be retrieved by the assembly code, not here.
348 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
349 * if an attempt is made to switch from a fast interrupt or IPI. This is
350 * necessary to properly take fatal kernel traps on SMP machines if
351 * get_mplock() has to block.
355 trap(struct trapframe *frame)
357 struct globaldata *gd = mycpu;
358 struct thread *td = gd->gd_curthread;
359 struct lwp *lp = td->td_lwp;
362 int i = 0, ucode = 0, type, code;
367 int crit_count = td->td_critcount;
368 lwkt_tokref_t curstop = td->td_toks_stop;
376 * We need to allow T_DNA faults when the debugger is active since
377 * some dumping paths do large bcopy() which use the floating
378 * point registers for faster copying.
380 if (db_active && frame->tf_trapno != T_DNA) {
381 eva = (frame->tf_trapno == T_PAGEFLT ? frame->tf_addr : 0);
382 ++gd->gd_trap_nesting_level;
383 MAKEMPSAFE(have_mplock);
384 trap_fatal(frame, eva);
385 --gd->gd_trap_nesting_level;
392 if ((frame->tf_rflags & PSL_I) == 0) {
394 * Buggy application or kernel code has disabled interrupts
395 * and then trapped. Enabling interrupts now is wrong, but
396 * it is better than running with interrupts disabled until
397 * they are accidentally enabled later.
399 type = frame->tf_trapno;
400 if (ISPL(frame->tf_cs) == SEL_UPL) {
401 MAKEMPSAFE(have_mplock);
402 /* JG curproc can be NULL */
404 "pid %ld (%s): trap %d with interrupts disabled\n",
405 (long)curproc->p_pid, curproc->p_comm, type);
406 } else if (type != T_NMI && type != T_BPTFLT &&
409 * XXX not quite right, since this may be for a
410 * multiple fault in user mode.
412 MAKEMPSAFE(have_mplock);
413 kprintf("kernel trap %d with interrupts disabled\n",
419 type = frame->tf_trapno;
420 code = frame->tf_err;
422 if (ISPL(frame->tf_cs) == SEL_UPL) {
425 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid,
426 frame->tf_trapno, eva);
430 sticks = (int)td->td_sticks;
431 KASSERT(lp->lwp_md.md_regs == frame,
432 ("Frame mismatch %p %p", lp->lwp_md.md_regs, frame));
435 case T_PRIVINFLT: /* privileged instruction fault */
440 case T_BPTFLT: /* bpt instruction fault */
441 case T_TRCTRAP: /* trace trap */
442 frame->tf_rflags &= ~PSL_T;
447 case T_ARITHTRAP: /* arithmetic trap */
460 case T_ASTFLT: /* Allow process switch */
461 mycpu->gd_cnt.v_soft++;
462 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
463 atomic_clear_int(&mycpu->gd_reqflags,
465 addupc_task(p, p->p_prof.pr_addr,
470 case T_PROTFLT: /* general protection fault */
474 case T_SEGNPFLT: /* segment not present fault */
478 case T_TSSFLT: /* invalid TSS fault */
479 case T_DOUBLEFLT: /* double fault */
484 ucode = code + BUS_SEGM_FAULT ; /* XXX: ???*/
490 case T_PAGEFLT: /* page fault */
491 i = trap_pfault(frame, TRUE);
492 if (frame->tf_rip == 0)
493 kprintf("T_PAGEFLT: Warning %%rip == 0!\n");
508 case T_DIVIDE: /* integer divide fault */
515 MAKEMPSAFE(have_mplock);
516 /* machine/parity/power fail/"kitchen sink" faults */
517 if (isa_nmi(code) == 0) {
520 * NMI can be hooked up to a pushbutton
524 kprintf ("NMI ... going to debugger\n");
525 kdb_trap(type, 0, frame);
529 } else if (panic_on_nmi)
530 panic("NMI indicates hardware failure");
532 #endif /* NISA > 0 */
534 case T_OFLOW: /* integer overflow fault */
539 case T_BOUND: /* bounds check fault */
546 * Virtual kernel intercept - pass the DNA exception
547 * to the 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 trap_pfault(frame, FALSE);
595 * The kernel is apparently using fpu for copying.
596 * XXX this should be fatal unless the kernel has
597 * registered such use.
603 case T_STKFLT: /* stack fault */
606 case T_PROTFLT: /* general protection fault */
607 case T_SEGNPFLT: /* segment not present fault */
609 * Invalid segment selectors and out of bounds
610 * %rip's and %rsp's can be set up in user mode.
611 * This causes a fault in kernel mode when the
612 * kernel tries to return to user mode. We want
613 * to get this fault so that we can fix the
614 * problem here and not have to check all the
615 * selectors and pointers when the user changes
618 if (mycpu->gd_intr_nesting_level == 0) {
619 if (td->td_pcb->pcb_onfault) {
620 frame->tf_rip = (register_t)
621 td->td_pcb->pcb_onfault;
624 if (frame->tf_rip == (long)doreti_iret) {
625 frame->tf_rip = (long)doreti_iret_fault;
633 * PSL_NT can be set in user mode and isn't cleared
634 * automatically when the kernel is entered. This
635 * causes a TSS fault when the kernel attempts to
636 * `iret' because the TSS link is uninitialized. We
637 * want to get this fault so that we can fix the
638 * problem here and not every time the kernel is
641 if (frame->tf_rflags & PSL_NT) {
642 frame->tf_rflags &= ~PSL_NT;
647 case T_TRCTRAP: /* trace trap */
649 if (frame->tf_rip == (int)IDTVEC(syscall)) {
651 * We've just entered system mode via the
652 * syscall lcall. Continue single stepping
653 * silently until the syscall handler has
658 if (frame->tf_rip == (int)IDTVEC(syscall) + 1) {
660 * The syscall handler has now saved the
661 * flags. Stop single stepping it.
663 frame->tf_rflags &= ~PSL_T;
669 * Ignore debug register trace traps due to
670 * accesses in the user's address space, which
671 * can happen under several conditions such as
672 * if a user sets a watchpoint on a buffer and
673 * then passes that buffer to a system call.
674 * We still want to get TRCTRAPS for addresses
675 * in kernel space because that is useful when
676 * debugging the kernel.
679 if (user_dbreg_trap()) {
681 * Reset breakpoint bits because the
684 /* XXX check upper bits here */
685 load_dr6(rdr6() & 0xfffffff0);
690 * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
694 * If DDB is enabled, let it handle the debugger trap.
695 * Otherwise, debugger traps "can't happen".
699 MAKEMPSAFE(have_mplock);
700 if (kdb_trap(type, 0, frame))
707 MAKEMPSAFE(have_mplock);
708 /* machine/parity/power fail/"kitchen sink" faults */
709 if (isa_nmi(code) == 0) {
712 * NMI can be hooked up to a pushbutton
716 kprintf ("NMI ... going to debugger\n");
717 kdb_trap(type, 0, frame);
721 } else if (panic_on_nmi == 0)
724 #endif /* NISA > 0 */
726 MAKEMPSAFE(have_mplock);
727 trap_fatal(frame, 0);
732 * Virtual kernel intercept - if the fault is directly related to a
733 * VM context managed by a virtual kernel then let the virtual kernel
736 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
737 vkernel_trap(lp, frame);
742 * Translate fault for emulators (e.g. Linux)
744 if (*p->p_sysent->sv_transtrap)
745 i = (*p->p_sysent->sv_transtrap)(i, type);
747 MAKEMPSAFE(have_mplock);
748 trapsignal(lp, i, ucode);
751 if (type <= MAX_TRAP_MSG) {
752 uprintf("fatal process exception: %s",
754 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
755 uprintf(", fault VA = 0x%lx", frame->tf_addr);
761 userret(lp, frame, sticks);
768 if (p != NULL && lp != NULL)
769 KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid);
771 KASSERT(crit_count == td->td_critcount,
772 ("trap: critical section count mismatch! %d/%d",
773 crit_count, td->td_pri));
774 KASSERT(curstop == td->td_toks_stop,
775 ("trap: extra tokens held after trap! %ld/%ld",
776 curstop - &td->td_toks_base,
777 td->td_toks_stop - &td->td_toks_base));
782 trap_pfault(struct trapframe *frame, int usermode)
785 struct vmspace *vm = NULL;
790 thread_t td = curthread;
791 struct lwp *lp = td->td_lwp;
794 va = trunc_page(frame->tf_addr);
795 if (va >= VM_MIN_KERNEL_ADDRESS) {
797 * Don't allow user-mode faults in kernel address space.
808 * This is a fault on non-kernel virtual memory.
809 * vm is initialized above to NULL. If curproc is NULL
810 * or curproc->p_vmspace is NULL the fault is fatal.
813 vm = lp->lwp_vmspace;
825 * PGEX_I is defined only if the execute disable bit capability is
826 * supported and enabled.
828 if (frame->tf_err & PGEX_W)
829 ftype = VM_PROT_WRITE;
831 else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
832 ftype = VM_PROT_EXECUTE;
835 ftype = VM_PROT_READ;
837 if (map != &kernel_map) {
839 * Keep swapout from messing with us during this
849 fault_flags |= VM_FAULT_BURST;
850 if (ftype & VM_PROT_WRITE)
851 fault_flags |= VM_FAULT_DIRTY;
853 fault_flags |= VM_FAULT_NORMAL;
854 rv = vm_fault(map, va, ftype, fault_flags);
859 * Don't have to worry about process locking or stacks
862 fault_flags = VM_FAULT_NORMAL;
863 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
866 if (rv == KERN_SUCCESS)
870 if (td->td_gd->gd_intr_nesting_level == 0 &&
871 td->td_pcb->pcb_onfault) {
872 frame->tf_rip = (register_t)td->td_pcb->pcb_onfault;
875 trap_fatal(frame, frame->tf_addr);
880 * NOTE: on x86_64 we have a tf_addr field in the trapframe, no
881 * kludge is needed to pass the fault address to signal handlers.
884 if (td->td_lwp->lwp_vkernel == NULL) {
886 kprintf("seg-fault ft=%04x ff=%04x addr=%p rip=%p "
887 "pid=%d p_comm=%s\n",
889 (void *)frame->tf_addr,
890 (void *)frame->tf_rip,
891 p->p_pid, p->p_comm);
893 if (ddb_on_seg_fault)
894 Debugger("ddb_on_seg_fault");
898 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
902 trap_fatal(struct trapframe *frame, vm_offset_t eva)
907 struct soft_segment_descriptor softseg;
910 code = frame->tf_err;
911 type = frame->tf_trapno;
912 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)], &softseg);
914 if (type <= MAX_TRAP_MSG)
915 msg = trap_msg[type];
918 kprintf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
919 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
921 /* three separate prints in case of a trap on an unmapped page */
922 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
923 kprintf("lapic->id = %08x\n", lapic->id);
925 if (type == T_PAGEFLT) {
926 kprintf("fault virtual address = 0x%lx\n", eva);
927 kprintf("fault code = %s %s %s, %s\n",
928 code & PGEX_U ? "user" : "supervisor",
929 code & PGEX_W ? "write" : "read",
930 code & PGEX_I ? "instruction" : "data",
931 code & PGEX_P ? "protection violation" : "page not present");
933 kprintf("instruction pointer = 0x%lx:0x%lx\n",
934 frame->tf_cs & 0xffff, frame->tf_rip);
935 if (ISPL(frame->tf_cs) == SEL_UPL) {
936 ss = frame->tf_ss & 0xffff;
939 ss = GSEL(GDATA_SEL, SEL_KPL);
940 rsp = (long)&frame->tf_rsp;
942 kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp);
943 kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp);
944 kprintf("code segment = base 0x%lx, limit 0x%lx, type 0x%x\n",
945 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
946 kprintf(" = DPL %d, pres %d, long %d, def32 %d, gran %d\n",
947 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32,
949 kprintf("processor eflags = ");
950 if (frame->tf_rflags & PSL_T)
951 kprintf("trace trap, ");
952 if (frame->tf_rflags & PSL_I)
953 kprintf("interrupt enabled, ");
954 if (frame->tf_rflags & PSL_NT)
955 kprintf("nested task, ");
956 if (frame->tf_rflags & PSL_RF)
958 kprintf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12);
959 kprintf("current process = ");
962 (u_long)curproc->p_pid);
966 kprintf("current thread = pri %d ", curthread->td_pri);
967 if (curthread->td_critcount)
972 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
975 kprintf("trap number = %d\n", type);
976 if (type <= MAX_TRAP_MSG)
977 panic("%s", trap_msg[type]);
979 panic("unknown/reserved trap");
983 * Double fault handler. Called when a fault occurs while writing
984 * a frame for a trap/exception onto the stack. This usually occurs
985 * when the stack overflows (such is the case with infinite recursion,
990 in_kstack_guard(register_t rptr)
992 thread_t td = curthread;
994 if ((char *)rptr >= td->td_kstack &&
995 (char *)rptr < td->td_kstack + PAGE_SIZE) {
1002 dblfault_handler(struct trapframe *frame)
1004 thread_t td = curthread;
1006 if (in_kstack_guard(frame->tf_rsp) || in_kstack_guard(frame->tf_rbp)) {
1007 kprintf("DOUBLE FAULT - KERNEL STACK GUARD HIT!\n");
1008 if (in_kstack_guard(frame->tf_rsp))
1009 frame->tf_rsp = (register_t)(td->td_kstack + PAGE_SIZE);
1010 if (in_kstack_guard(frame->tf_rbp))
1011 frame->tf_rbp = (register_t)(td->td_kstack + PAGE_SIZE);
1013 kprintf("DOUBLE FAULT\n");
1015 kprintf("\nFatal double fault\n");
1016 kprintf("rip = 0x%lx\n", frame->tf_rip);
1017 kprintf("rsp = 0x%lx\n", frame->tf_rsp);
1018 kprintf("rbp = 0x%lx\n", frame->tf_rbp);
1020 /* three separate prints in case of a trap on an unmapped page */
1021 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
1022 kprintf("lapic->id = %08x\n", lapic->id);
1024 panic("double fault");
1028 * syscall2 - MP aware system call request C handler
1030 * A system call is essentially treated as a trap except that the
1031 * MP lock is not held on entry or return. We are responsible for
1032 * obtaining the MP lock if necessary and for handling ASTs
1033 * (e.g. a task switch) prior to return.
1038 syscall2(struct trapframe *frame)
1040 struct thread *td = curthread;
1041 struct proc *p = td->td_proc;
1042 struct lwp *lp = td->td_lwp;
1044 struct sysent *callp;
1045 register_t orig_tf_rflags;
1050 int crit_count = td->td_critcount;
1053 int have_mplock = 0;
1058 union sysunion args;
1059 register_t *argsdst;
1061 mycpu->gd_cnt.v_syscall++;
1064 if (ISPL(frame->tf_cs) != SEL_UPL) {
1071 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid,
1074 userenter(td, p); /* lazy raise our priority */
1081 sticks = (int)td->td_sticks;
1082 orig_tf_rflags = frame->tf_rflags;
1085 * Virtual kernel intercept - if a VM context managed by a virtual
1086 * kernel issues a system call the virtual kernel handles it, not us.
1087 * Restore the virtual kernel context and return from its system
1088 * call. The current frame is copied out to the virtual kernel.
1090 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1091 vkernel_trap(lp, frame);
1092 error = EJUSTRETURN;
1097 * Get the system call parameters and account for time
1099 KASSERT(lp->lwp_md.md_regs == frame,
1100 ("Frame mismatch %p %p", lp->lwp_md.md_regs, frame));
1101 params = (caddr_t)frame->tf_rsp + sizeof(register_t);
1102 code = frame->tf_rax;
1104 if (p->p_sysent->sv_prepsyscall) {
1105 (*p->p_sysent->sv_prepsyscall)(
1106 frame, (int *)(&args.nosys.sysmsg + 1),
1109 if (code == SYS_syscall || code == SYS___syscall) {
1110 code = frame->tf_rdi;
1116 if (p->p_sysent->sv_mask)
1117 code &= p->p_sysent->sv_mask;
1119 if (code >= p->p_sysent->sv_size)
1120 callp = &p->p_sysent->sv_table[0];
1122 callp = &p->p_sysent->sv_table[code];
1124 narg = callp->sy_narg & SYF_ARGMASK;
1127 * On x86_64 we get up to six arguments in registers. The rest are
1128 * on the stack. The first six members of 'struct trapframe' happen
1129 * to be the registers used to pass arguments, in exactly the right
1132 argp = &frame->tf_rdi;
1134 argsdst = (register_t *)(&args.nosys.sysmsg + 1);
1136 * JG can we overflow the space pointed to by 'argsdst'
1137 * either with 'bcopy' or with 'copyin'?
1139 bcopy(argp, argsdst, sizeof(register_t) * regcnt);
1141 * copyin is MP aware, but the tracing code is not
1143 if (narg > regcnt) {
1144 KASSERT(params != NULL, ("copyin args with no params!"));
1145 error = copyin(params, &argsdst[regcnt],
1146 (narg - regcnt) * sizeof(register_t));
1149 if (KTRPOINT(td, KTR_SYSCALL)) {
1150 MAKEMPSAFE(have_mplock);
1152 ktrsyscall(lp, code, narg,
1153 (void *)(&args.nosys.sysmsg + 1));
1161 if (KTRPOINT(td, KTR_SYSCALL)) {
1162 MAKEMPSAFE(have_mplock);
1163 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1168 * Default return value is 0 (will be copied to %rax). Double-value
1169 * returns use %rax and %rdx. %rdx is left unchanged for system
1170 * calls which return only one result.
1172 args.sysmsg_fds[0] = 0;
1173 args.sysmsg_fds[1] = frame->tf_rdx;
1176 * The syscall might manipulate the trap frame. If it does it
1177 * will probably return EJUSTRETURN.
1179 args.sysmsg_frame = frame;
1181 STOPEVENT(p, S_SCE, narg); /* MP aware */
1184 * NOTE: All system calls run MPSAFE now. The system call itself
1185 * is responsible for getting the MP lock.
1187 error = (*callp->sy_call)(&args);
1191 * MP SAFE (we may or may not have the MP lock at this point)
1193 //kprintf("SYSMSG %d ", error);
1197 * Reinitialize proc pointer `p' as it may be different
1198 * if this is a child returning from fork syscall.
1201 lp = curthread->td_lwp;
1202 frame->tf_rax = args.sysmsg_fds[0];
1203 frame->tf_rdx = args.sysmsg_fds[1];
1204 frame->tf_rflags &= ~PSL_C;
1208 * Reconstruct pc, we know that 'syscall' is 2 bytes.
1209 * We have to do a full context restore so that %r10
1210 * (which was holding the value of %rcx) is restored for
1211 * the next iteration.
1213 frame->tf_rip -= frame->tf_err;
1214 frame->tf_r10 = frame->tf_rcx;
1219 panic("Unexpected EASYNC return value (for now)");
1222 if (p->p_sysent->sv_errsize) {
1223 if (error >= p->p_sysent->sv_errsize)
1224 error = -1; /* XXX */
1226 error = p->p_sysent->sv_errtbl[error];
1228 frame->tf_rax = error;
1229 frame->tf_rflags |= PSL_C;
1234 * Traced syscall. trapsignal() is not MP aware.
1236 if (orig_tf_rflags & PSL_T) {
1237 MAKEMPSAFE(have_mplock);
1238 frame->tf_rflags &= ~PSL_T;
1239 trapsignal(lp, SIGTRAP, TRAP_TRACE);
1243 * Handle reschedule and other end-of-syscall issues
1245 userret(lp, frame, sticks);
1248 if (KTRPOINT(td, KTR_SYSRET)) {
1249 MAKEMPSAFE(have_mplock);
1250 ktrsysret(lp, code, error, args.sysmsg_result);
1255 * This works because errno is findable through the
1256 * register set. If we ever support an emulation where this
1257 * is not the case, this code will need to be revisited.
1259 STOPEVENT(p, S_SCX, code);
1264 * Release the MP lock if we had to get it
1269 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error);
1271 KASSERT(crit_count == td->td_critcount,
1272 ("syscall: critical section count mismatch! %d/%d",
1273 crit_count, td->td_pri));
1274 KASSERT(&td->td_toks_base == td->td_toks_stop,
1275 ("syscall: extra tokens held after trap! %ld",
1276 td->td_toks_stop - &td->td_toks_base));
1281 * NOTE: mplock not held at any point
1284 fork_return(struct lwp *lp, struct trapframe *frame)
1286 frame->tf_rax = 0; /* Child returns zero */
1287 frame->tf_rflags &= ~PSL_C; /* success */
1290 generic_lwp_return(lp, frame);
1291 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1295 * Simplified back end of syscall(), used when returning from fork()
1296 * directly into user mode.
1298 * This code will return back into the fork trampoline code which then
1301 * NOTE: The mplock is not held at any point.
1304 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1306 struct proc *p = lp->lwp_proc;
1309 * Newly forked processes are given a kernel priority. We have to
1310 * adjust the priority to a normal user priority and fake entry
1311 * into the kernel (call userenter()) to install a passive release
1312 * function just in case userret() decides to stop the process. This
1313 * can occur when ^Z races a fork. If we do not install the passive
1314 * release function the current process designation will not be
1315 * released when the thread goes to sleep.
1317 lwkt_setpri_self(TDPRI_USER_NORM);
1318 userenter(lp->lwp_thread, p);
1319 userret(lp, frame, 0);
1321 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1322 ktrsysret(lp, SYS_fork, 0, 0);
1324 p->p_flag |= P_PASSIVE_ACQ;
1326 p->p_flag &= ~P_PASSIVE_ACQ;
1330 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1331 * fault (which is then passed back to the virtual kernel) if an attempt is
1332 * made to use the FP unit.
1334 * XXX this is a fairly big hack.
1337 set_vkernel_fp(struct trapframe *frame)
1339 struct thread *td = curthread;
1341 if (frame->tf_xflags & PGEX_FPFAULT) {
1342 td->td_pcb->pcb_flags |= FP_VIRTFP;
1343 if (mdcpu->gd_npxthread == td)
1346 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1351 * Called from vkernel_trap() to fixup the vkernel's syscall
1352 * frame for vmspace_ctl() return.
1355 cpu_vkernel_trap(struct trapframe *frame, int error)
1357 frame->tf_rax = error;
1359 frame->tf_rflags |= PSL_C;
1361 frame->tf_rflags &= ~PSL_C;