2 * Copyright (c) 1990, 1993
3 * The Regents of the University of California. All rights reserved.
4 * Copyright (C) 1994, David Greenman
5 * Copyright (c) 2008 The DragonFly Project.
6 * Copyright (c) 2008 Jordan Gordeev.
8 * This code is derived from software contributed to Berkeley by
9 * the University of Utah, and William Jolitz.
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. All advertising materials mentioning features or use of this software
20 * must display the following acknowledgement:
21 * This product includes software developed by the University of
22 * California, Berkeley and its contributors.
23 * 4. Neither the name of the University nor the names of its contributors
24 * may be used to endorse or promote products derived from this software
25 * without specific prior written permission.
27 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39 * from: @(#)trap.c 7.4 (Berkeley) 5/13/91
40 * $FreeBSD: src/sys/i386/i386/trap.c,v 1.147.2.11 2003/02/27 19:09:59 luoqi Exp $
44 * x86_64 Trap and System call handling
48 #include "opt_ktrace.h"
50 #include <machine/frame.h>
51 #include <sys/param.h>
52 #include <sys/systm.h>
53 #include <sys/kernel.h>
55 #include <sys/pioctl.h>
56 #include <sys/types.h>
57 #include <sys/signal2.h>
58 #include <sys/syscall.h>
59 #include <sys/sysctl.h>
60 #include <sys/sysent.h>
61 #include <sys/systm.h>
63 #include <sys/ktrace.h>
66 #include <sys/sysmsg.h>
67 #include <sys/sysproto.h>
68 #include <sys/sysunion.h>
72 #include <vm/vm_extern.h>
73 #include <vm/vm_kern.h>
74 #include <vm/vm_param.h>
75 #include <machine/cpu.h>
76 #include <machine/pcb.h>
77 #include <machine/smp.h>
78 #include <machine/thread.h>
79 #include <machine/vmparam.h>
80 #include <machine/md_var.h>
81 #include <machine_base/isa/intr_machdep.h>
84 #include <sys/thread2.h>
88 #define MAKEMPSAFE(have_mplock) \
89 if (have_mplock == 0) { \
96 #define MAKEMPSAFE(have_mplock)
100 extern void trap(struct trapframe *frame);
102 static int trap_pfault(struct trapframe *, int);
103 static void trap_fatal(struct trapframe *, vm_offset_t);
104 void dblfault_handler(struct trapframe *frame);
106 #define MAX_TRAP_MSG 30
107 static char *trap_msg[] = {
109 "privileged instruction fault", /* 1 T_PRIVINFLT */
111 "breakpoint instruction fault", /* 3 T_BPTFLT */
114 "arithmetic trap", /* 6 T_ARITHTRAP */
115 "system forced exception", /* 7 T_ASTFLT */
117 "general protection fault", /* 9 T_PROTFLT */
118 "trace trap", /* 10 T_TRCTRAP */
120 "page fault", /* 12 T_PAGEFLT */
122 "alignment fault", /* 14 T_ALIGNFLT */
126 "integer divide fault", /* 18 T_DIVIDE */
127 "non-maskable interrupt trap", /* 19 T_NMI */
128 "overflow trap", /* 20 T_OFLOW */
129 "FPU bounds check fault", /* 21 T_BOUND */
130 "FPU device not available", /* 22 T_DNA */
131 "double fault", /* 23 T_DOUBLEFLT */
132 "FPU operand fetch fault", /* 24 T_FPOPFLT */
133 "invalid TSS fault", /* 25 T_TSSFLT */
134 "segment not present fault", /* 26 T_SEGNPFLT */
135 "stack fault", /* 27 T_STKFLT */
136 "machine check trap", /* 28 T_MCHK */
137 "SIMD floating-point exception", /* 29 T_XMMFLT */
138 "reserved (unknown) fault", /* 30 T_RESERVED */
142 static int ddb_on_nmi = 1;
143 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
144 &ddb_on_nmi, 0, "Go to DDB on NMI");
146 static int panic_on_nmi = 1;
147 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
148 &panic_on_nmi, 0, "Panic on NMI");
149 static int fast_release;
150 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
151 &fast_release, 0, "Passive Release was optimal");
152 static int slow_release;
153 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
154 &slow_release, 0, "Passive Release was nonoptimal");
156 static int syscall_mpsafe = 1;
157 SYSCTL_INT(_kern, OID_AUTO, syscall_mpsafe, CTLFLAG_RW,
158 &syscall_mpsafe, 0, "Allow MPSAFE marked syscalls to run without BGL");
159 TUNABLE_INT("kern.syscall_mpsafe", &syscall_mpsafe);
160 static int trap_mpsafe = 1;
161 SYSCTL_INT(_kern, OID_AUTO, trap_mpsafe, CTLFLAG_RW,
162 &trap_mpsafe, 0, "Allow traps to mostly run without the BGL");
163 TUNABLE_INT("kern.trap_mpsafe", &trap_mpsafe);
167 * userenter() passively intercepts the thread switch function to increase
168 * the thread priority from a user priority to a kernel priority, reducing
169 * syscall and trap overhead for the case where no switch occurs.
173 userenter(struct thread *curtd)
175 curtd->td_release = lwkt_passive_release;
179 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
180 * must be completed before we can return to or try to return to userland.
182 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
183 * arithmatic on the delta calculation so the absolute tick values are
184 * truncated to an integer.
187 userret(struct lwp *lp, struct trapframe *frame, int sticks)
189 struct proc *p = lp->lwp_proc;
193 * Charge system time if profiling. Note: times are in microseconds.
194 * This may do a copyout and block, so do it first even though it
195 * means some system time will be charged as user time.
197 if (p->p_flag & P_PROFIL) {
198 addupc_task(p, frame->tf_rip,
199 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
204 * If the jungle wants us dead, so be it.
206 if (lp->lwp_flag & LWP_WEXIT) {
209 rel_mplock(); /* NOT REACHED */
213 * Block here if we are in a stopped state.
215 if (p->p_stat == SSTOP) {
223 * Post any pending upcalls. If running a virtual kernel be sure
224 * to restore the virtual kernel's vmspace before posting the upcall.
226 if (p->p_flag & P_UPCALLPEND) {
227 p->p_flag &= ~P_UPCALLPEND;
235 * Post any pending signals. If running a virtual kernel be sure
236 * to restore the virtual kernel's vmspace before posting the signal.
238 if ((sig = CURSIG_TRACE(lp)) != 0) {
246 * block here if we are swapped out, but still process signals
247 * (such as SIGKILL). proc0 (the swapin scheduler) is already
248 * aware of our situation, we do not have to wake it up.
250 if (p->p_flag & P_SWAPPEDOUT) {
252 p->p_flag |= P_SWAPWAIT;
254 if (p->p_flag & P_SWAPWAIT)
255 tsleep(p, PCATCH, "SWOUT", 0);
256 p->p_flag &= ~P_SWAPWAIT;
262 * Make sure postsig() handled request to restore old signal mask after
263 * running signal handler.
265 KKASSERT((lp->lwp_flag & LWP_OLDMASK) == 0);
269 * Cleanup from userenter and any passive release that might have occured.
270 * We must reclaim the current-process designation before we can return
271 * to usermode. We also handle both LWKT and USER reschedule requests.
274 userexit(struct lwp *lp)
276 struct thread *td = lp->lwp_thread;
277 /* globaldata_t gd = td->td_gd;*/
280 * Handle stop requests at kernel priority. Any requests queued
281 * after this loop will generate another AST.
283 while (lp->lwp_proc->p_stat == SSTOP) {
290 * Reduce our priority in preparation for a return to userland. If
291 * our passive release function was still in place, our priority was
292 * never raised and does not need to be reduced.
294 lwkt_passive_recover(td);
297 * Become the current user scheduled process if we aren't already,
298 * and deal with reschedule requests and other factors.
300 lp->lwp_proc->p_usched->acquire_curproc(lp);
301 /* WARNING: we may have migrated cpu's */
302 /* gd = td->td_gd; */
305 #if !defined(KTR_KERNENTRY)
306 #define KTR_KERNENTRY KTR_ALL
308 KTR_INFO_MASTER(kernentry);
309 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, "STR",
310 sizeof(long) + sizeof(long) + sizeof(long) + sizeof(vm_offset_t));
311 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "STR",
312 sizeof(long) + sizeof(long));
313 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "STR",
314 sizeof(long) + sizeof(long) + sizeof(long));
315 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "STR",
316 sizeof(long) + sizeof(long) + sizeof(long));
317 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "STR",
318 sizeof(long) + sizeof(long));
321 * Exception, fault, and trap interface to the kernel.
322 * This common code is called from assembly language IDT gate entry
323 * routines that prepare a suitable stack frame, and restore this
324 * frame after the exception has been processed.
326 * This function is also called from doreti in an interlock to handle ASTs.
327 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
329 * NOTE! We have to retrieve the fault address prior to obtaining the
330 * MP lock because get_mplock() may switch out. YYY cr2 really ought
331 * to be retrieved by the assembly code, not here.
333 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
334 * if an attempt is made to switch from a fast interrupt or IPI. This is
335 * necessary to properly take fatal kernel traps on SMP machines if
336 * get_mplock() has to block.
340 trap(struct trapframe *frame)
342 struct globaldata *gd = mycpu;
343 struct thread *td = gd->gd_curthread;
344 struct lwp *lp = td->td_lwp;
347 int i = 0, ucode = 0, type, code;
352 int crit_count = td->td_pri & ~TDPRI_MASK;
360 kprintf0("\"%s\" type=%ld\n",
361 trap_msg[frame->tf_trapno], frame->tf_trapno);
362 kprintf0(" rip=%lx rsp=%lx\n", frame->tf_rip, frame->tf_rsp);
363 kprintf0(" err=%lx addr=%lx\n", frame->tf_err, frame->tf_addr);
364 kprintf0(" cs=%lx ss=%lx rflags=%lx\n", (unsigned long)frame->tf_cs, (unsigned long)frame->tf_ss, frame->tf_rflags);
369 ++gd->gd_trap_nesting_level;
370 MAKEMPSAFE(have_mplock);
371 trap_fatal(frame, frame->tf_addr);
372 --gd->gd_trap_nesting_level;
378 eva = (frame->tf_trapno == T_PAGEFLT ? frame->tf_addr : 0);
379 ++gd->gd_trap_nesting_level;
380 MAKEMPSAFE(have_mplock);
381 trap_fatal(frame, eva);
382 --gd->gd_trap_nesting_level;
390 if (trap_mpsafe == 0) {
391 ++gd->gd_trap_nesting_level;
392 MAKEMPSAFE(have_mplock);
393 --gd->gd_trap_nesting_level;
397 if ((frame->tf_rflags & PSL_I) == 0) {
399 * Buggy application or kernel code has disabled interrupts
400 * and then trapped. Enabling interrupts now is wrong, but
401 * it is better than running with interrupts disabled until
402 * they are accidentally enabled later.
404 type = frame->tf_trapno;
405 if (ISPL(frame->tf_cs) == SEL_UPL) {
406 MAKEMPSAFE(have_mplock);
407 /* JG curproc can be NULL */
409 "pid %ld (%s): trap %d with interrupts disabled\n",
410 (long)curproc->p_pid, curproc->p_comm, type);
411 } else if (type != T_NMI && type != T_BPTFLT &&
414 * XXX not quite right, since this may be for a
415 * multiple fault in user mode.
417 MAKEMPSAFE(have_mplock);
418 kprintf("kernel trap %d with interrupts disabled\n",
424 type = frame->tf_trapno;
425 code = frame->tf_err;
427 if (ISPL(frame->tf_cs) == SEL_UPL) {
430 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid,
431 frame->tf_trapno, eva);
435 sticks = (int)td->td_sticks;
436 lp->lwp_md.md_regs = frame;
439 case T_PRIVINFLT: /* privileged instruction fault */
444 case T_BPTFLT: /* bpt instruction fault */
445 case T_TRCTRAP: /* trace trap */
446 frame->tf_rflags &= ~PSL_T;
450 case T_ARITHTRAP: /* arithmetic trap */
463 case T_ASTFLT: /* Allow process switch */
464 mycpu->gd_cnt.v_soft++;
465 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
466 atomic_clear_int_nonlocked(&mycpu->gd_reqflags,
468 addupc_task(p, p->p_prof.pr_addr,
473 case T_PROTFLT: /* general protection fault */
474 case T_SEGNPFLT: /* segment not present fault */
475 case T_TSSFLT: /* invalid TSS fault */
476 case T_DOUBLEFLT: /* double fault */
478 ucode = code + BUS_SEGM_FAULT ;
482 case T_PAGEFLT: /* page fault */
483 MAKEMPSAFE(have_mplock);
484 i = trap_pfault(frame, TRUE);
485 if (frame->tf_rip == 0)
486 kprintf("T_PAGEFLT: Warning %%rip == 0!\n");
495 case T_DIVIDE: /* integer divide fault */
501 MAKEMPSAFE(have_mplock);
502 /* machine/parity/power fail/"kitchen sink" faults */
503 if (isa_nmi(code) == 0) {
506 * NMI can be hooked up to a pushbutton
510 kprintf ("NMI ... going to debugger\n");
511 kdb_trap(type, 0, frame);
515 } else if (panic_on_nmi)
516 panic("NMI indicates hardware failure");
519 case T_OFLOW: /* integer overflow fault */
524 case T_BOUND: /* bounds check fault */
531 * Virtual kernel intercept - pass the DNA exception
532 * to the virtual kernel if it asked to handle it.
533 * This occurs when the virtual kernel is holding
534 * onto the FP context for a different emulated
535 * process then the one currently running.
537 * We must still call npxdna() since we may have
538 * saved FP state that the virtual kernel needs
539 * to hand over to a different emulated process.
541 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
542 (td->td_pcb->pcb_flags & FP_VIRTFP)
549 * The kernel may have switched out the FP unit's
550 * state, causing the user process to take a fault
551 * when it tries to use the FP unit. Restore the
557 ucode = FPE_FPU_NP_TRAP;
560 case T_FPOPFLT: /* FPU operand fetch fault */
565 case T_XMMFLT: /* SIMD floating-point exception */
574 case T_PAGEFLT: /* page fault */
575 MAKEMPSAFE(have_mplock);
576 trap_pfault(frame, FALSE);
581 * The kernel is apparently using fpu for copying.
582 * XXX this should be fatal unless the kernel has
583 * registered such use.
589 case T_STKFLT: /* stack fault */
592 case T_PROTFLT: /* general protection fault */
593 case T_SEGNPFLT: /* segment not present fault */
595 * Invalid segment selectors and out of bounds
596 * %rip's and %rsp's can be set up in user mode.
597 * This causes a fault in kernel mode when the
598 * kernel tries to return to user mode. We want
599 * to get this fault so that we can fix the
600 * problem here and not have to check all the
601 * selectors and pointers when the user changes
604 kprintf("trap.c line %d\n", __LINE__);
605 if (mycpu->gd_intr_nesting_level == 0) {
606 if (td->td_pcb->pcb_onfault) {
607 frame->tf_rip = (register_t)
608 td->td_pcb->pcb_onfault;
611 if (frame->tf_rip == (long)doreti_iret) {
612 frame->tf_rip = (long)doreti_iret_fault;
620 * PSL_NT can be set in user mode and isn't cleared
621 * automatically when the kernel is entered. This
622 * causes a TSS fault when the kernel attempts to
623 * `iret' because the TSS link is uninitialized. We
624 * want to get this fault so that we can fix the
625 * problem here and not every time the kernel is
628 if (frame->tf_rflags & PSL_NT) {
629 frame->tf_rflags &= ~PSL_NT;
634 case T_TRCTRAP: /* trace trap */
636 if (frame->tf_rip == (int)IDTVEC(syscall)) {
638 * We've just entered system mode via the
639 * syscall lcall. Continue single stepping
640 * silently until the syscall handler has
645 if (frame->tf_rip == (int)IDTVEC(syscall) + 1) {
647 * The syscall handler has now saved the
648 * flags. Stop single stepping it.
650 frame->tf_rflags &= ~PSL_T;
656 * Ignore debug register trace traps due to
657 * accesses in the user's address space, which
658 * can happen under several conditions such as
659 * if a user sets a watchpoint on a buffer and
660 * then passes that buffer to a system call.
661 * We still want to get TRCTRAPS for addresses
662 * in kernel space because that is useful when
663 * debugging the kernel.
666 if (user_dbreg_trap()) {
668 * Reset breakpoint bits because the
671 /* XXX check upper bits here */
672 load_dr6(rdr6() & 0xfffffff0);
677 * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
681 * If DDB is enabled, let it handle the debugger trap.
682 * Otherwise, debugger traps "can't happen".
685 MAKEMPSAFE(have_mplock);
686 if (kdb_trap(type, 0, frame))
692 MAKEMPSAFE(have_mplock);
693 /* machine/parity/power fail/"kitchen sink" faults */
695 if (isa_nmi(code) == 0) {
698 * NMI can be hooked up to a pushbutton
702 kprintf ("NMI ... going to debugger\n");
703 kdb_trap(type, 0, frame);
707 } else if (panic_on_nmi == 0)
710 #endif /* NISA > 0 */
712 MAKEMPSAFE(have_mplock);
713 trap_fatal(frame, 0);
718 * Virtual kernel intercept - if the fault is directly related to a
719 * VM context managed by a virtual kernel then let the virtual kernel
722 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
723 vkernel_trap(lp, frame);
728 * Virtual kernel intercept - if the fault is directly related to a
729 * VM context managed by a virtual kernel then let the virtual kernel
732 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
733 vkernel_trap(lp, frame);
738 * Translate fault for emulators (e.g. Linux)
740 if (*p->p_sysent->sv_transtrap)
741 i = (*p->p_sysent->sv_transtrap)(i, type);
743 MAKEMPSAFE(have_mplock);
744 trapsignal(lp, i, ucode);
747 if (type <= MAX_TRAP_MSG) {
748 uprintf("fatal process exception: %s",
750 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
751 uprintf(", fault VA = 0x%lx", frame->tf_addr);
758 if (ISPL(frame->tf_cs) == SEL_UPL)
759 KASSERT(td->td_mpcount == have_mplock, ("badmpcount trap/end from %p", (void *)frame->tf_rip));
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_pri & ~TDPRI_MASK),
772 ("syscall: critical section count mismatch! %d/%d",
773 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
778 trap_pfault(struct trapframe *frame, int usermode)
781 struct vmspace *vm = NULL;
785 thread_t td = curthread;
786 struct lwp *lp = td->td_lwp;
788 va = trunc_page(frame->tf_addr);
789 if (va >= VM_MIN_KERNEL_ADDRESS) {
791 * Don't allow user-mode faults in kernel address space.
799 * This is a fault on non-kernel virtual memory.
800 * vm is initialized above to NULL. If curproc is NULL
801 * or curproc->p_vmspace is NULL the fault is fatal.
804 vm = lp->lwp_vmspace;
813 * PGEX_I is defined only if the execute disable bit capability is
814 * supported and enabled.
816 if (frame->tf_err & PGEX_W)
817 ftype = VM_PROT_WRITE;
819 else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
820 ftype = VM_PROT_EXECUTE;
823 ftype = VM_PROT_READ;
825 if (map != &kernel_map) {
827 * Keep swapout from messing with us during this
833 * Grow the stack if necessary
835 /* grow_stack returns false only if va falls into
836 * a growable stack region and the stack growth
837 * fails. It returns true if va was not within
838 * a growable stack region, or if the stack
841 if (!grow_stack(lp->lwp_proc, va)) {
847 /* Fault in the user page: */
848 rv = vm_fault(map, va, ftype,
849 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
855 * Don't have to worry about process locking or stacks
858 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
861 if (rv == KERN_SUCCESS)
865 if (td->td_gd->gd_intr_nesting_level == 0 &&
866 td->td_pcb->pcb_onfault) {
867 frame->tf_rip = (register_t)td->td_pcb->pcb_onfault;
870 trap_fatal(frame, frame->tf_addr);
875 * NOTE: on x86_64 we have a tf_addr field in the trapframe, no
876 * kludge is needed to pass the fault address to signal handlers.
878 struct proc *p = td->td_proc;
879 kprintf("seg-fault accessing address %p rip=%p pid=%d p_comm=%s\n",
880 (void *)va, (void *)frame->tf_rip, p->p_pid, p->p_comm);
881 /* Debugger("seg-fault"); */
883 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
887 trap_fatal(struct trapframe *frame, vm_offset_t eva)
892 struct soft_segment_descriptor softseg;
895 code = frame->tf_err;
896 type = frame->tf_trapno;
897 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)], &softseg);
899 if (type <= MAX_TRAP_MSG)
900 msg = trap_msg[type];
903 kprintf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
904 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
906 /* three separate prints in case of a trap on an unmapped page */
907 kprintf("mp_lock = %08x; ", mp_lock);
908 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
909 kprintf("lapic->id = %08x\n", lapic->id);
911 if (type == T_PAGEFLT) {
912 kprintf("fault virtual address = 0x%lx\n", eva);
913 kprintf("fault code = %s %s %s, %s\n",
914 code & PGEX_U ? "user" : "supervisor",
915 code & PGEX_W ? "write" : "read",
916 code & PGEX_I ? "instruction" : "data",
917 code & PGEX_P ? "protection violation" : "page not present");
919 kprintf("instruction pointer = 0x%lx:0x%lx\n",
920 frame->tf_cs & 0xffff, frame->tf_rip);
921 if (ISPL(frame->tf_cs) == SEL_UPL) {
922 ss = frame->tf_ss & 0xffff;
925 ss = GSEL(GDATA_SEL, SEL_KPL);
926 rsp = (long)&frame->tf_rsp;
928 kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp);
929 kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp);
930 kprintf("code segment = base 0x%lx, limit 0x%lx, type 0x%x\n",
931 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
932 kprintf(" = DPL %d, pres %d, long %d, def32 %d, gran %d\n",
933 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32,
935 kprintf("processor eflags = ");
936 if (frame->tf_rflags & PSL_T)
937 kprintf("trace trap, ");
938 if (frame->tf_rflags & PSL_I)
939 kprintf("interrupt enabled, ");
940 if (frame->tf_rflags & PSL_NT)
941 kprintf("nested task, ");
942 if (frame->tf_rflags & PSL_RF)
944 kprintf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12);
945 kprintf("current process = ");
948 (u_long)curproc->p_pid);
952 kprintf("current thread = pri %d ", curthread->td_pri);
953 if (curthread->td_pri >= TDPRI_CRIT)
958 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
961 kprintf("trap number = %d\n", type);
962 if (type <= MAX_TRAP_MSG)
963 panic("%s", trap_msg[type]);
965 panic("unknown/reserved trap");
969 * Double fault handler. Called when a fault occurs while writing
970 * a frame for a trap/exception onto the stack. This usually occurs
971 * when the stack overflows (such is the case with infinite recursion,
975 dblfault_handler(struct trapframe *frame)
977 kprintf0("DOUBLE FAULT\n");
978 kprintf("\nFatal double fault\n");
979 kprintf("rip = 0x%lx\n", frame->tf_rip);
980 kprintf("rsp = 0x%lx\n", frame->tf_rsp);
981 kprintf("rbp = 0x%lx\n", frame->tf_rbp);
983 /* three separate prints in case of a trap on an unmapped page */
984 kprintf("mp_lock = %08x; ", mp_lock);
985 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
986 kprintf("lapic->id = %08x\n", lapic->id);
988 panic("double fault");
992 * syscall2 - MP aware system call request C handler
994 * A system call is essentially treated as a trap except that the
995 * MP lock is not held on entry or return. We are responsible for
996 * obtaining the MP lock if necessary and for handling ASTs
997 * (e.g. a task switch) prior to return.
999 * In general, only simple access and manipulation of curproc and
1000 * the current stack is allowed without having to hold MP lock.
1002 * MPSAFE - note that large sections of this routine are run without
1006 syscall2(struct trapframe *frame)
1008 struct thread *td = curthread;
1009 struct proc *p = td->td_proc;
1010 struct lwp *lp = td->td_lwp;
1012 struct sysent *callp;
1013 register_t orig_tf_rflags;
1018 int crit_count = td->td_pri & ~TDPRI_MASK;
1021 int have_mplock = 0;
1026 union sysunion args;
1027 register_t *argsdst;
1029 mycpu->gd_cnt.v_syscall++;
1032 if (ISPL(frame->tf_cs) != SEL_UPL) {
1039 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid,
1043 KASSERT(td->td_mpcount == 0, ("badmpcount syscall2 from %p", (void *)frame->tf_rip));
1044 if (syscall_mpsafe == 0)
1045 MAKEMPSAFE(have_mplock);
1047 userenter(td); /* lazy raise our priority */
1054 sticks = (int)td->td_sticks;
1055 orig_tf_rflags = frame->tf_rflags;
1058 * Virtual kernel intercept - if a VM context managed by a virtual
1059 * kernel issues a system call the virtual kernel handles it, not us.
1060 * Restore the virtual kernel context and return from its system
1061 * call. The current frame is copied out to the virtual kernel.
1063 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1064 error = vkernel_trap(lp, frame);
1065 frame->tf_rax = error;
1067 frame->tf_rflags |= PSL_C;
1068 error = EJUSTRETURN;
1073 * Get the system call parameters and account for time
1075 lp->lwp_md.md_regs = frame;
1076 params = (caddr_t)frame->tf_rsp + sizeof(register_t);
1077 code = frame->tf_rax;
1079 if (p->p_sysent->sv_prepsyscall) {
1080 (*p->p_sysent->sv_prepsyscall)(
1081 frame, (int *)(&args.nosys.sysmsg + 1),
1084 if (code == SYS_syscall || code == SYS___syscall) {
1085 code = frame->tf_rdi;
1091 if (p->p_sysent->sv_mask)
1092 code &= p->p_sysent->sv_mask;
1094 if (code >= p->p_sysent->sv_size)
1095 callp = &p->p_sysent->sv_table[0];
1097 callp = &p->p_sysent->sv_table[code];
1099 narg = callp->sy_narg & SYF_ARGMASK;
1102 * On x86_64 we get up to six arguments in registers. The rest are
1103 * on the stack. The first six members of 'struct trapframe' happen
1104 * to be the registers used to pass arguments, in exactly the right
1107 argp = &frame->tf_rdi;
1109 argsdst = (register_t *)(&args.nosys.sysmsg + 1);
1111 * JG can we overflow the space pointed to by 'argsdst'
1112 * either with 'bcopy' or with 'copyin'?
1114 bcopy(argp, argsdst, sizeof(register_t) * regcnt);
1116 * copyin is MP aware, but the tracing code is not
1118 if (narg > regcnt) {
1119 KASSERT(params != NULL, ("copyin args with no params!"));
1120 error = copyin(params, &argsdst[regcnt],
1121 (narg - regcnt) * sizeof(register_t));
1124 if (KTRPOINT(td, KTR_SYSCALL)) {
1125 MAKEMPSAFE(have_mplock);
1127 ktrsyscall(lp, code, narg,
1128 (void *)(&args.nosys.sysmsg + 1));
1136 if (KTRPOINT(td, KTR_SYSCALL)) {
1137 MAKEMPSAFE(have_mplock);
1138 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1143 * Default return value is 0 (will be copied to %rax). Double-value
1144 * returns use %rax and %rdx. %rdx is left unchanged for system
1145 * calls which return only one result.
1147 args.sysmsg_fds[0] = 0;
1148 args.sysmsg_fds[1] = frame->tf_rdx;
1151 * The syscall might manipulate the trap frame. If it does it
1152 * will probably return EJUSTRETURN.
1154 args.sysmsg_frame = frame;
1156 STOPEVENT(p, S_SCE, narg); /* MP aware */
1160 * Try to run the syscall without the MP lock if the syscall
1161 * is MP safe. We have to obtain the MP lock no matter what if
1164 if ((callp->sy_narg & SYF_MPSAFE) == 0)
1165 MAKEMPSAFE(have_mplock);
1168 error = (*callp->sy_call)(&args);
1172 * MP SAFE (we may or may not have the MP lock at this point)
1174 //kprintf("SYSMSG %d ", error);
1178 * Reinitialize proc pointer `p' as it may be different
1179 * if this is a child returning from fork syscall.
1182 lp = curthread->td_lwp;
1183 frame->tf_rax = args.sysmsg_fds[0];
1184 frame->tf_rdx = args.sysmsg_fds[1];
1185 frame->tf_rflags &= ~PSL_C;
1189 * Reconstruct pc, we know that 'syscall' is 2 bytes.
1190 * We have to do a full context restore so that %r10
1191 * (which was holding the value of %rcx) is restored for
1192 * the next iteration.
1194 frame->tf_rip -= frame->tf_err;
1195 frame->tf_r10 = frame->tf_rcx;
1200 panic("Unexpected EASYNC return value (for now)");
1203 if (p->p_sysent->sv_errsize) {
1204 if (error >= p->p_sysent->sv_errsize)
1205 error = -1; /* XXX */
1207 error = p->p_sysent->sv_errtbl[error];
1209 frame->tf_rax = error;
1210 frame->tf_rflags |= PSL_C;
1215 * Traced syscall. trapsignal() is not MP aware.
1217 if (orig_tf_rflags & PSL_T) {
1218 MAKEMPSAFE(have_mplock);
1219 frame->tf_rflags &= ~PSL_T;
1220 trapsignal(lp, SIGTRAP, 0);
1224 * Handle reschedule and other end-of-syscall issues
1226 userret(lp, frame, sticks);
1229 if (KTRPOINT(td, KTR_SYSRET)) {
1230 MAKEMPSAFE(have_mplock);
1231 ktrsysret(lp, code, error, args.sysmsg_result);
1236 * This works because errno is findable through the
1237 * register set. If we ever support an emulation where this
1238 * is not the case, this code will need to be revisited.
1240 STOPEVENT(p, S_SCX, code);
1245 * Release the MP lock if we had to get it
1247 KASSERT(td->td_mpcount == have_mplock,
1248 ("badmpcount syscall2/end from %p", (void *)frame->tf_rip));
1252 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error);
1254 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
1255 ("syscall: critical section count mismatch! %d/%d",
1256 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
1261 fork_return(struct lwp *lp, struct trapframe *frame)
1263 frame->tf_rax = 0; /* Child returns zero */
1264 frame->tf_rflags &= ~PSL_C; /* success */
1267 generic_lwp_return(lp, frame);
1268 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1272 * Simplified back end of syscall(), used when returning from fork()
1273 * directly into user mode. MP lock is held on entry and should be
1274 * released on return. This code will return back into the fork
1275 * trampoline code which then runs doreti.
1278 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1280 struct proc *p = lp->lwp_proc;
1283 * Newly forked processes are given a kernel priority. We have to
1284 * adjust the priority to a normal user priority and fake entry
1285 * into the kernel (call userenter()) to install a passive release
1286 * function just in case userret() decides to stop the process. This
1287 * can occur when ^Z races a fork. If we do not install the passive
1288 * release function the current process designation will not be
1289 * released when the thread goes to sleep.
1291 lwkt_setpri_self(TDPRI_USER_NORM);
1292 userenter(lp->lwp_thread);
1293 userret(lp, frame, 0);
1295 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1296 ktrsysret(lp, SYS_fork, 0, 0);
1298 p->p_flag |= P_PASSIVE_ACQ;
1300 p->p_flag &= ~P_PASSIVE_ACQ;
1302 KKASSERT(lp->lwp_thread->td_mpcount == 1);
1308 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1309 * fault (which is then passed back to the virtual kernel) if an attempt is
1310 * made to use the FP unit.
1312 * XXX this is a fairly big hack.
1315 set_vkernel_fp(struct trapframe *frame)