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>
85 #include <sys/thread2.h>
86 #include <sys/mplock2.h>
90 #define MAKEMPSAFE(have_mplock) \
91 if (have_mplock == 0) { \
98 #define MAKEMPSAFE(have_mplock)
102 extern void trap(struct trapframe *frame);
104 static int trap_pfault(struct trapframe *, int);
105 static void trap_fatal(struct trapframe *, vm_offset_t);
106 void dblfault_handler(struct trapframe *frame);
108 #define MAX_TRAP_MSG 30
109 static char *trap_msg[] = {
111 "privileged instruction fault", /* 1 T_PRIVINFLT */
113 "breakpoint instruction fault", /* 3 T_BPTFLT */
116 "arithmetic trap", /* 6 T_ARITHTRAP */
117 "system forced exception", /* 7 T_ASTFLT */
119 "general protection fault", /* 9 T_PROTFLT */
120 "trace trap", /* 10 T_TRCTRAP */
122 "page fault", /* 12 T_PAGEFLT */
124 "alignment fault", /* 14 T_ALIGNFLT */
128 "integer divide fault", /* 18 T_DIVIDE */
129 "non-maskable interrupt trap", /* 19 T_NMI */
130 "overflow trap", /* 20 T_OFLOW */
131 "FPU bounds check fault", /* 21 T_BOUND */
132 "FPU device not available", /* 22 T_DNA */
133 "double fault", /* 23 T_DOUBLEFLT */
134 "FPU operand fetch fault", /* 24 T_FPOPFLT */
135 "invalid TSS fault", /* 25 T_TSSFLT */
136 "segment not present fault", /* 26 T_SEGNPFLT */
137 "stack fault", /* 27 T_STKFLT */
138 "machine check trap", /* 28 T_MCHK */
139 "SIMD floating-point exception", /* 29 T_XMMFLT */
140 "reserved (unknown) fault", /* 30 T_RESERVED */
144 static int ddb_on_nmi = 1;
145 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
146 &ddb_on_nmi, 0, "Go to DDB on NMI");
148 static int panic_on_nmi = 1;
149 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
150 &panic_on_nmi, 0, "Panic on NMI");
151 static int fast_release;
152 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
153 &fast_release, 0, "Passive Release was optimal");
154 static int slow_release;
155 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
156 &slow_release, 0, "Passive Release was nonoptimal");
158 static int syscall_mpsafe = 1;
159 SYSCTL_INT(_kern, OID_AUTO, syscall_mpsafe, CTLFLAG_RW,
160 &syscall_mpsafe, 0, "Allow MPSAFE marked syscalls to run without BGL");
161 TUNABLE_INT("kern.syscall_mpsafe", &syscall_mpsafe);
162 static int trap_mpsafe = 1;
163 SYSCTL_INT(_kern, OID_AUTO, trap_mpsafe, CTLFLAG_RW,
164 &trap_mpsafe, 0, "Allow traps to mostly run without the BGL");
165 TUNABLE_INT("kern.trap_mpsafe", &trap_mpsafe);
169 * Passively intercepts the thread switch function to increase
170 * the thread priority from a user priority to a kernel priority, reducing
171 * syscall and trap overhead for the case where no switch occurs.
173 * Synchronizes td_ucred with p_ucred. This is used by system calls,
174 * signal handling, faults, AST traps, and anything else that enters the
175 * kernel from userland and provides the kernel with a stable read-only
176 * copy of the process ucred.
179 userenter(struct thread *curtd, struct proc *curp)
184 curtd->td_release = lwkt_passive_release;
186 if (curtd->td_ucred != curp->p_ucred) {
187 ncred = crhold(curp->p_ucred);
188 ocred = curtd->td_ucred;
189 curtd->td_ucred = ncred;
196 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
197 * must be completed before we can return to or try to return to userland.
199 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
200 * arithmatic on the delta calculation so the absolute tick values are
201 * truncated to an integer.
204 userret(struct lwp *lp, struct trapframe *frame, int sticks)
206 struct proc *p = lp->lwp_proc;
210 * Charge system time if profiling. Note: times are in microseconds.
211 * This may do a copyout and block, so do it first even though it
212 * means some system time will be charged as user time.
214 if (p->p_flag & P_PROFIL) {
215 addupc_task(p, frame->tf_rip,
216 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
221 * If the jungle wants us dead, so be it.
223 if (lp->lwp_flag & LWP_WEXIT) {
226 rel_mplock(); /* NOT REACHED */
230 * Block here if we are in a stopped state.
232 if (p->p_stat == SSTOP) {
240 * Post any pending upcalls. If running a virtual kernel be sure
241 * to restore the virtual kernel's vmspace before posting the upcall.
243 if (p->p_flag & P_UPCALLPEND) {
244 p->p_flag &= ~P_UPCALLPEND;
252 * Post any pending signals. If running a virtual kernel be sure
253 * to restore the virtual kernel's vmspace before posting the signal.
255 if ((sig = CURSIG_TRACE(lp)) != 0) {
263 * block here if we are swapped out, but still process signals
264 * (such as SIGKILL). proc0 (the swapin scheduler) is already
265 * aware of our situation, we do not have to wake it up.
267 if (p->p_flag & P_SWAPPEDOUT) {
269 p->p_flag |= P_SWAPWAIT;
271 if (p->p_flag & P_SWAPWAIT)
272 tsleep(p, PCATCH, "SWOUT", 0);
273 p->p_flag &= ~P_SWAPWAIT;
279 * Make sure postsig() handled request to restore old signal mask after
280 * running signal handler.
282 KKASSERT((lp->lwp_flag & LWP_OLDMASK) == 0);
286 * Cleanup from userenter and any passive release that might have occured.
287 * We must reclaim the current-process designation before we can return
288 * to usermode. We also handle both LWKT and USER reschedule requests.
291 userexit(struct lwp *lp)
293 struct thread *td = lp->lwp_thread;
294 /* globaldata_t gd = td->td_gd;*/
297 * Handle stop requests at kernel priority. Any requests queued
298 * after this loop will generate another AST.
300 while (lp->lwp_proc->p_stat == SSTOP) {
307 * Reduce our priority in preparation for a return to userland. If
308 * our passive release function was still in place, our priority was
309 * never raised and does not need to be reduced.
311 lwkt_passive_recover(td);
314 * Become the current user scheduled process if we aren't already,
315 * and deal with reschedule requests and other factors.
317 lp->lwp_proc->p_usched->acquire_curproc(lp);
318 /* WARNING: we may have migrated cpu's */
319 /* gd = td->td_gd; */
322 #if !defined(KTR_KERNENTRY)
323 #define KTR_KERNENTRY KTR_ALL
325 KTR_INFO_MASTER(kernentry);
326 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, "STR",
327 sizeof(long) + sizeof(long) + sizeof(long) + sizeof(vm_offset_t));
328 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "STR",
329 sizeof(long) + sizeof(long));
330 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "STR",
331 sizeof(long) + sizeof(long) + sizeof(long));
332 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "STR",
333 sizeof(long) + sizeof(long) + sizeof(long));
334 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "STR",
335 sizeof(long) + sizeof(long));
338 * Exception, fault, and trap interface to the kernel.
339 * This common code is called from assembly language IDT gate entry
340 * routines that prepare a suitable stack frame, and restore this
341 * frame after the exception has been processed.
343 * This function is also called from doreti in an interlock to handle ASTs.
344 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
346 * NOTE! We have to retrieve the fault address prior to obtaining the
347 * MP lock because get_mplock() may switch out. YYY cr2 really ought
348 * to be retrieved by the assembly code, not here.
350 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
351 * if an attempt is made to switch from a fast interrupt or IPI. This is
352 * necessary to properly take fatal kernel traps on SMP machines if
353 * get_mplock() has to block.
357 trap(struct trapframe *frame)
359 struct globaldata *gd = mycpu;
360 struct thread *td = gd->gd_curthread;
361 struct lwp *lp = td->td_lwp;
364 int i = 0, ucode = 0, type, code;
369 int crit_count = td->td_pri & ~TDPRI_MASK;
377 kprintf0("\"%s\" type=%ld\n",
378 trap_msg[frame->tf_trapno], frame->tf_trapno);
379 kprintf0(" rip=%lx rsp=%lx\n", frame->tf_rip, frame->tf_rsp);
380 kprintf0(" err=%lx addr=%lx\n", frame->tf_err, frame->tf_addr);
381 kprintf0(" cs=%lx ss=%lx rflags=%lx\n", (unsigned long)frame->tf_cs, (unsigned long)frame->tf_ss, frame->tf_rflags);
386 ++gd->gd_trap_nesting_level;
387 MAKEMPSAFE(have_mplock);
388 trap_fatal(frame, frame->tf_addr);
389 --gd->gd_trap_nesting_level;
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;
407 if (trap_mpsafe == 0) {
408 ++gd->gd_trap_nesting_level;
409 MAKEMPSAFE(have_mplock);
410 --gd->gd_trap_nesting_level;
414 if ((frame->tf_rflags & PSL_I) == 0) {
416 * Buggy application or kernel code has disabled interrupts
417 * and then trapped. Enabling interrupts now is wrong, but
418 * it is better than running with interrupts disabled until
419 * they are accidentally enabled later.
421 type = frame->tf_trapno;
422 if (ISPL(frame->tf_cs) == SEL_UPL) {
423 MAKEMPSAFE(have_mplock);
424 /* JG curproc can be NULL */
426 "pid %ld (%s): trap %d with interrupts disabled\n",
427 (long)curproc->p_pid, curproc->p_comm, type);
428 } else if (type != T_NMI && type != T_BPTFLT &&
431 * XXX not quite right, since this may be for a
432 * multiple fault in user mode.
434 MAKEMPSAFE(have_mplock);
435 kprintf("kernel trap %d with interrupts disabled\n",
441 type = frame->tf_trapno;
442 code = frame->tf_err;
444 if (ISPL(frame->tf_cs) == SEL_UPL) {
447 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid,
448 frame->tf_trapno, eva);
452 sticks = (int)td->td_sticks;
453 lp->lwp_md.md_regs = frame;
456 case T_PRIVINFLT: /* privileged instruction fault */
461 case T_BPTFLT: /* bpt instruction fault */
462 case T_TRCTRAP: /* trace trap */
463 frame->tf_rflags &= ~PSL_T;
468 case T_ARITHTRAP: /* arithmetic trap */
481 case T_ASTFLT: /* Allow process switch */
482 mycpu->gd_cnt.v_soft++;
483 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
484 atomic_clear_int_nonlocked(&mycpu->gd_reqflags,
486 addupc_task(p, p->p_prof.pr_addr,
491 case T_PROTFLT: /* general protection fault */
495 case T_SEGNPFLT: /* segment not present fault */
499 case T_TSSFLT: /* invalid TSS fault */
500 case T_DOUBLEFLT: /* double fault */
505 ucode = code + BUS_SEGM_FAULT ; /* XXX: ???*/
511 case T_PAGEFLT: /* page fault */
512 MAKEMPSAFE(have_mplock);
513 i = trap_pfault(frame, TRUE);
514 if (frame->tf_rip == 0)
515 kprintf("T_PAGEFLT: Warning %%rip == 0!\n");
530 case T_DIVIDE: /* integer divide fault */
536 MAKEMPSAFE(have_mplock);
537 /* machine/parity/power fail/"kitchen sink" faults */
538 if (isa_nmi(code) == 0) {
541 * NMI can be hooked up to a pushbutton
545 kprintf ("NMI ... going to debugger\n");
546 kdb_trap(type, 0, frame);
550 } else if (panic_on_nmi)
551 panic("NMI indicates hardware failure");
554 case T_OFLOW: /* integer overflow fault */
559 case T_BOUND: /* bounds check fault */
566 * Virtual kernel intercept - pass the DNA exception
567 * to the virtual kernel if it asked to handle it.
568 * This occurs when the virtual kernel is holding
569 * onto the FP context for a different emulated
570 * process then the one currently running.
572 * We must still call npxdna() since we may have
573 * saved FP state that the virtual kernel needs
574 * to hand over to a different emulated process.
576 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
577 (td->td_pcb->pcb_flags & FP_VIRTFP)
584 * The kernel may have switched out the FP unit's
585 * state, causing the user process to take a fault
586 * when it tries to use the FP unit. Restore the
592 ucode = FPE_FPU_NP_TRAP;
595 case T_FPOPFLT: /* FPU operand fetch fault */
600 case T_XMMFLT: /* SIMD floating-point exception */
609 case T_PAGEFLT: /* page fault */
610 MAKEMPSAFE(have_mplock);
611 trap_pfault(frame, FALSE);
616 * The kernel is apparently using fpu for copying.
617 * XXX this should be fatal unless the kernel has
618 * registered such use.
624 case T_STKFLT: /* stack fault */
627 case T_PROTFLT: /* general protection fault */
628 case T_SEGNPFLT: /* segment not present fault */
630 * Invalid segment selectors and out of bounds
631 * %rip's and %rsp's can be set up in user mode.
632 * This causes a fault in kernel mode when the
633 * kernel tries to return to user mode. We want
634 * to get this fault so that we can fix the
635 * problem here and not have to check all the
636 * selectors and pointers when the user changes
639 kprintf("trap.c line %d\n", __LINE__);
640 if (mycpu->gd_intr_nesting_level == 0) {
641 if (td->td_pcb->pcb_onfault) {
642 frame->tf_rip = (register_t)
643 td->td_pcb->pcb_onfault;
646 if (frame->tf_rip == (long)doreti_iret) {
647 frame->tf_rip = (long)doreti_iret_fault;
655 * PSL_NT can be set in user mode and isn't cleared
656 * automatically when the kernel is entered. This
657 * causes a TSS fault when the kernel attempts to
658 * `iret' because the TSS link is uninitialized. We
659 * want to get this fault so that we can fix the
660 * problem here and not every time the kernel is
663 if (frame->tf_rflags & PSL_NT) {
664 frame->tf_rflags &= ~PSL_NT;
669 case T_TRCTRAP: /* trace trap */
671 if (frame->tf_rip == (int)IDTVEC(syscall)) {
673 * We've just entered system mode via the
674 * syscall lcall. Continue single stepping
675 * silently until the syscall handler has
680 if (frame->tf_rip == (int)IDTVEC(syscall) + 1) {
682 * The syscall handler has now saved the
683 * flags. Stop single stepping it.
685 frame->tf_rflags &= ~PSL_T;
691 * Ignore debug register trace traps due to
692 * accesses in the user's address space, which
693 * can happen under several conditions such as
694 * if a user sets a watchpoint on a buffer and
695 * then passes that buffer to a system call.
696 * We still want to get TRCTRAPS for addresses
697 * in kernel space because that is useful when
698 * debugging the kernel.
701 if (user_dbreg_trap()) {
703 * Reset breakpoint bits because the
706 /* XXX check upper bits here */
707 load_dr6(rdr6() & 0xfffffff0);
712 * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
716 * If DDB is enabled, let it handle the debugger trap.
717 * Otherwise, debugger traps "can't happen".
721 MAKEMPSAFE(have_mplock);
722 if (kdb_trap(type, 0, frame))
728 MAKEMPSAFE(have_mplock);
729 /* machine/parity/power fail/"kitchen sink" faults */
731 if (isa_nmi(code) == 0) {
734 * NMI can be hooked up to a pushbutton
738 kprintf ("NMI ... going to debugger\n");
739 kdb_trap(type, 0, frame);
743 } else if (panic_on_nmi == 0)
746 #endif /* NISA > 0 */
748 MAKEMPSAFE(have_mplock);
749 trap_fatal(frame, 0);
754 * Virtual kernel intercept - if the fault is directly related to a
755 * VM context managed by a virtual kernel then let the virtual kernel
758 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
759 vkernel_trap(lp, frame);
764 * Translate fault for emulators (e.g. Linux)
766 if (*p->p_sysent->sv_transtrap)
767 i = (*p->p_sysent->sv_transtrap)(i, type);
769 MAKEMPSAFE(have_mplock);
770 trapsignal(lp, i, ucode);
773 if (type <= MAX_TRAP_MSG) {
774 uprintf("fatal process exception: %s",
776 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
777 uprintf(", fault VA = 0x%lx", frame->tf_addr);
784 if (ISPL(frame->tf_cs) == SEL_UPL)
785 KASSERT(td->td_mpcount == have_mplock, ("badmpcount trap/end from %p", (void *)frame->tf_rip));
787 userret(lp, frame, sticks);
794 if (p != NULL && lp != NULL)
795 KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid);
797 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
798 ("syscall: critical section count mismatch! %d/%d",
799 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
804 trap_pfault(struct trapframe *frame, int usermode)
807 struct vmspace *vm = NULL;
812 thread_t td = curthread;
813 struct lwp *lp = td->td_lwp;
815 va = trunc_page(frame->tf_addr);
816 if (va >= VM_MIN_KERNEL_ADDRESS) {
818 * 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;
840 * PGEX_I is defined only if the execute disable bit capability is
841 * supported and enabled.
843 if (frame->tf_err & PGEX_W)
844 ftype = VM_PROT_WRITE;
846 else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
847 ftype = VM_PROT_EXECUTE;
850 ftype = VM_PROT_READ;
852 if (map != &kernel_map) {
854 * Keep swapout from messing with us during this
860 * Grow the stack if necessary
862 /* grow_stack returns false only if va falls into
863 * a growable stack region and the stack growth
864 * fails. It returns true if va was not within
865 * a growable stack region, or if the stack
868 if (!grow_stack(lp->lwp_proc, va)) {
879 fault_flags |= VM_FAULT_BURST;
880 if (ftype & VM_PROT_WRITE)
881 fault_flags |= VM_FAULT_DIRTY;
883 fault_flags |= VM_FAULT_NORMAL;
884 rv = vm_fault(map, va, ftype, fault_flags);
889 * Don't have to worry about process locking or stacks
892 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
895 if (rv == KERN_SUCCESS)
899 if (td->td_gd->gd_intr_nesting_level == 0 &&
900 td->td_pcb->pcb_onfault) {
901 frame->tf_rip = (register_t)td->td_pcb->pcb_onfault;
904 trap_fatal(frame, frame->tf_addr);
909 * NOTE: on x86_64 we have a tf_addr field in the trapframe, no
910 * kludge is needed to pass the fault address to signal handlers.
912 struct proc *p = td->td_proc;
913 kprintf("seg-fault accessing address %p rip=%p pid=%d p_comm=%s\n",
914 (void *)va, (void *)frame->tf_rip, p->p_pid, p->p_comm);
915 /* Debugger("seg-fault"); */
917 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
921 trap_fatal(struct trapframe *frame, vm_offset_t eva)
926 struct soft_segment_descriptor softseg;
929 code = frame->tf_err;
930 type = frame->tf_trapno;
931 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)], &softseg);
933 if (type <= MAX_TRAP_MSG)
934 msg = trap_msg[type];
937 kprintf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
938 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
940 /* three separate prints in case of a trap on an unmapped page */
941 kprintf("mp_lock = %08x; ", mp_lock);
942 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
943 kprintf("lapic->id = %08x\n", lapic->id);
945 if (type == T_PAGEFLT) {
946 kprintf("fault virtual address = 0x%lx\n", eva);
947 kprintf("fault code = %s %s %s, %s\n",
948 code & PGEX_U ? "user" : "supervisor",
949 code & PGEX_W ? "write" : "read",
950 code & PGEX_I ? "instruction" : "data",
951 code & PGEX_P ? "protection violation" : "page not present");
953 kprintf("instruction pointer = 0x%lx:0x%lx\n",
954 frame->tf_cs & 0xffff, frame->tf_rip);
955 if (ISPL(frame->tf_cs) == SEL_UPL) {
956 ss = frame->tf_ss & 0xffff;
959 ss = GSEL(GDATA_SEL, SEL_KPL);
960 rsp = (long)&frame->tf_rsp;
962 kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp);
963 kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp);
964 kprintf("code segment = base 0x%lx, limit 0x%lx, type 0x%x\n",
965 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
966 kprintf(" = DPL %d, pres %d, long %d, def32 %d, gran %d\n",
967 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32,
969 kprintf("processor eflags = ");
970 if (frame->tf_rflags & PSL_T)
971 kprintf("trace trap, ");
972 if (frame->tf_rflags & PSL_I)
973 kprintf("interrupt enabled, ");
974 if (frame->tf_rflags & PSL_NT)
975 kprintf("nested task, ");
976 if (frame->tf_rflags & PSL_RF)
978 kprintf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12);
979 kprintf("current process = ");
982 (u_long)curproc->p_pid);
986 kprintf("current thread = pri %d ", curthread->td_pri);
987 if (curthread->td_pri >= TDPRI_CRIT)
992 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
995 kprintf("trap number = %d\n", type);
996 if (type <= MAX_TRAP_MSG)
997 panic("%s", trap_msg[type]);
999 panic("unknown/reserved trap");
1003 * Double fault handler. Called when a fault occurs while writing
1004 * a frame for a trap/exception onto the stack. This usually occurs
1005 * when the stack overflows (such is the case with infinite recursion,
1009 dblfault_handler(struct trapframe *frame)
1011 kprintf0("DOUBLE FAULT\n");
1012 kprintf("\nFatal double fault\n");
1013 kprintf("rip = 0x%lx\n", frame->tf_rip);
1014 kprintf("rsp = 0x%lx\n", frame->tf_rsp);
1015 kprintf("rbp = 0x%lx\n", frame->tf_rbp);
1017 /* three separate prints in case of a trap on an unmapped page */
1018 kprintf("mp_lock = %08x; ", mp_lock);
1019 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
1020 kprintf("lapic->id = %08x\n", lapic->id);
1022 panic("double fault");
1026 * syscall2 - MP aware system call request C handler
1028 * A system call is essentially treated as a trap except that the
1029 * MP lock is not held on entry or return. We are responsible for
1030 * obtaining the MP lock if necessary and for handling ASTs
1031 * (e.g. a task switch) prior to return.
1036 syscall2(struct trapframe *frame)
1038 struct thread *td = curthread;
1039 struct proc *p = td->td_proc;
1040 struct lwp *lp = td->td_lwp;
1042 struct sysent *callp;
1043 register_t orig_tf_rflags;
1048 int crit_count = td->td_pri & ~TDPRI_MASK;
1051 int have_mplock = 0;
1056 union sysunion args;
1057 register_t *argsdst;
1059 mycpu->gd_cnt.v_syscall++;
1062 if (ISPL(frame->tf_cs) != SEL_UPL) {
1069 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid,
1073 KASSERT(td->td_mpcount == 0, ("badmpcount syscall2 from %p", (void *)frame->tf_rip));
1074 if (syscall_mpsafe == 0)
1075 MAKEMPSAFE(have_mplock);
1077 userenter(td, p); /* lazy raise our priority */
1084 sticks = (int)td->td_sticks;
1085 orig_tf_rflags = frame->tf_rflags;
1088 * Virtual kernel intercept - if a VM context managed by a virtual
1089 * kernel issues a system call the virtual kernel handles it, not us.
1090 * Restore the virtual kernel context and return from its system
1091 * call. The current frame is copied out to the virtual kernel.
1093 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1094 vkernel_trap(lp, frame);
1095 error = EJUSTRETURN;
1100 * Get the system call parameters and account for time
1102 lp->lwp_md.md_regs = frame;
1103 params = (caddr_t)frame->tf_rsp + sizeof(register_t);
1104 code = frame->tf_rax;
1106 if (p->p_sysent->sv_prepsyscall) {
1107 (*p->p_sysent->sv_prepsyscall)(
1108 frame, (int *)(&args.nosys.sysmsg + 1),
1111 if (code == SYS_syscall || code == SYS___syscall) {
1112 code = frame->tf_rdi;
1118 if (p->p_sysent->sv_mask)
1119 code &= p->p_sysent->sv_mask;
1121 if (code >= p->p_sysent->sv_size)
1122 callp = &p->p_sysent->sv_table[0];
1124 callp = &p->p_sysent->sv_table[code];
1126 narg = callp->sy_narg & SYF_ARGMASK;
1129 * On x86_64 we get up to six arguments in registers. The rest are
1130 * on the stack. The first six members of 'struct trapframe' happen
1131 * to be the registers used to pass arguments, in exactly the right
1134 argp = &frame->tf_rdi;
1136 argsdst = (register_t *)(&args.nosys.sysmsg + 1);
1138 * JG can we overflow the space pointed to by 'argsdst'
1139 * either with 'bcopy' or with 'copyin'?
1141 bcopy(argp, argsdst, sizeof(register_t) * regcnt);
1143 * copyin is MP aware, but the tracing code is not
1145 if (narg > regcnt) {
1146 KASSERT(params != NULL, ("copyin args with no params!"));
1147 error = copyin(params, &argsdst[regcnt],
1148 (narg - regcnt) * sizeof(register_t));
1151 if (KTRPOINT(td, KTR_SYSCALL)) {
1152 MAKEMPSAFE(have_mplock);
1154 ktrsyscall(lp, code, narg,
1155 (void *)(&args.nosys.sysmsg + 1));
1163 if (KTRPOINT(td, KTR_SYSCALL)) {
1164 MAKEMPSAFE(have_mplock);
1165 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1170 * Default return value is 0 (will be copied to %rax). Double-value
1171 * returns use %rax and %rdx. %rdx is left unchanged for system
1172 * calls which return only one result.
1174 args.sysmsg_fds[0] = 0;
1175 args.sysmsg_fds[1] = frame->tf_rdx;
1178 * The syscall might manipulate the trap frame. If it does it
1179 * will probably return EJUSTRETURN.
1181 args.sysmsg_frame = frame;
1183 STOPEVENT(p, S_SCE, narg); /* MP aware */
1186 * NOTE: All system calls run MPSAFE now. The system call itself
1187 * is responsible for getting the MP lock.
1189 error = (*callp->sy_call)(&args);
1193 * MP SAFE (we may or may not have the MP lock at this point)
1195 //kprintf("SYSMSG %d ", error);
1199 * Reinitialize proc pointer `p' as it may be different
1200 * if this is a child returning from fork syscall.
1203 lp = curthread->td_lwp;
1204 frame->tf_rax = args.sysmsg_fds[0];
1205 frame->tf_rdx = args.sysmsg_fds[1];
1206 frame->tf_rflags &= ~PSL_C;
1210 * Reconstruct pc, we know that 'syscall' is 2 bytes.
1211 * We have to do a full context restore so that %r10
1212 * (which was holding the value of %rcx) is restored for
1213 * the next iteration.
1215 frame->tf_rip -= frame->tf_err;
1216 frame->tf_r10 = frame->tf_rcx;
1221 panic("Unexpected EASYNC return value (for now)");
1224 if (p->p_sysent->sv_errsize) {
1225 if (error >= p->p_sysent->sv_errsize)
1226 error = -1; /* XXX */
1228 error = p->p_sysent->sv_errtbl[error];
1230 frame->tf_rax = error;
1231 frame->tf_rflags |= PSL_C;
1236 * Traced syscall. trapsignal() is not MP aware.
1238 if (orig_tf_rflags & PSL_T) {
1239 MAKEMPSAFE(have_mplock);
1240 frame->tf_rflags &= ~PSL_T;
1241 trapsignal(lp, SIGTRAP, TRAP_TRACE);
1245 * Handle reschedule and other end-of-syscall issues
1247 userret(lp, frame, sticks);
1250 if (KTRPOINT(td, KTR_SYSRET)) {
1251 MAKEMPSAFE(have_mplock);
1252 ktrsysret(lp, code, error, args.sysmsg_result);
1257 * This works because errno is findable through the
1258 * register set. If we ever support an emulation where this
1259 * is not the case, this code will need to be revisited.
1261 STOPEVENT(p, S_SCX, code);
1266 * Release the MP lock if we had to get it
1268 KASSERT(td->td_mpcount == have_mplock,
1269 ("badmpcount syscall2/end from %p", (void *)frame->tf_rip));
1273 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error);
1275 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
1276 ("syscall: critical section count mismatch! %d/%d",
1277 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
1282 fork_return(struct lwp *lp, struct trapframe *frame)
1284 frame->tf_rax = 0; /* Child returns zero */
1285 frame->tf_rflags &= ~PSL_C; /* success */
1288 generic_lwp_return(lp, frame);
1289 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1293 * Simplified back end of syscall(), used when returning from fork()
1294 * directly into user mode. MP lock is held on entry and should be
1295 * released on return. This code will return back into the fork
1296 * trampoline code which then runs doreti.
1299 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1301 struct proc *p = lp->lwp_proc;
1304 * Newly forked processes are given a kernel priority. We have to
1305 * adjust the priority to a normal user priority and fake entry
1306 * into the kernel (call userenter()) to install a passive release
1307 * function just in case userret() decides to stop the process. This
1308 * can occur when ^Z races a fork. If we do not install the passive
1309 * release function the current process designation will not be
1310 * released when the thread goes to sleep.
1312 lwkt_setpri_self(TDPRI_USER_NORM);
1313 userenter(lp->lwp_thread, p);
1314 userret(lp, frame, 0);
1316 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1317 ktrsysret(lp, SYS_fork, 0, 0);
1319 p->p_flag |= P_PASSIVE_ACQ;
1321 p->p_flag &= ~P_PASSIVE_ACQ;
1323 KKASSERT(lp->lwp_thread->td_mpcount == 1);
1329 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1330 * fault (which is then passed back to the virtual kernel) if an attempt is
1331 * made to use the FP unit.
1333 * XXX this is a fairly big hack.
1336 set_vkernel_fp(struct trapframe *frame)
1342 * Called from vkernel_trap() to fixup the vkernel's syscall
1343 * frame for vmspace_ctl() return.
1346 cpu_vkernel_trap(struct trapframe *frame, int error)
1348 frame->tf_rax = error;
1350 frame->tf_rflags |= PSL_C;
1352 frame->tf_rflags &= ~PSL_C;