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");
159 * Passively intercepts the thread switch function to increase
160 * the thread priority from a user priority to a kernel priority, reducing
161 * syscall and trap overhead for the case where no switch occurs.
163 * Synchronizes td_ucred with p_ucred. This is used by system calls,
164 * signal handling, faults, AST traps, and anything else that enters the
165 * kernel from userland and provides the kernel with a stable read-only
166 * copy of the process ucred.
169 userenter(struct thread *curtd, struct proc *curp)
174 curtd->td_release = lwkt_passive_release;
176 if (curtd->td_ucred != curp->p_ucred) {
177 ncred = crhold(curp->p_ucred);
178 ocred = curtd->td_ucred;
179 curtd->td_ucred = ncred;
186 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
187 * must be completed before we can return to or try to return to userland.
189 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
190 * arithmatic on the delta calculation so the absolute tick values are
191 * truncated to an integer.
194 userret(struct lwp *lp, struct trapframe *frame, int sticks)
196 struct proc *p = lp->lwp_proc;
200 * Charge system time if profiling. Note: times are in microseconds.
201 * This may do a copyout and block, so do it first even though it
202 * means some system time will be charged as user time.
204 if (p->p_flag & P_PROFIL) {
205 addupc_task(p, frame->tf_rip,
206 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
211 * If the jungle wants us dead, so be it.
213 if (lp->lwp_flag & LWP_WEXIT) {
216 rel_mplock(); /* NOT REACHED */
220 * Block here if we are in a stopped state.
222 if (p->p_stat == SSTOP) {
230 * Post any pending upcalls. If running a virtual kernel be sure
231 * to restore the virtual kernel's vmspace before posting the upcall.
233 if (p->p_flag & P_UPCALLPEND) {
234 p->p_flag &= ~P_UPCALLPEND;
242 * Post any pending signals. If running a virtual kernel be sure
243 * to restore the virtual kernel's vmspace before posting the signal.
245 if ((sig = CURSIG_TRACE(lp)) != 0) {
253 * block here if we are swapped out, but still process signals
254 * (such as SIGKILL). proc0 (the swapin scheduler) is already
255 * aware of our situation, we do not have to wake it up.
257 if (p->p_flag & P_SWAPPEDOUT) {
259 p->p_flag |= P_SWAPWAIT;
261 if (p->p_flag & P_SWAPWAIT)
262 tsleep(p, PCATCH, "SWOUT", 0);
263 p->p_flag &= ~P_SWAPWAIT;
269 * Make sure postsig() handled request to restore old signal mask after
270 * running signal handler.
272 KKASSERT((lp->lwp_flag & LWP_OLDMASK) == 0);
276 * Cleanup from userenter and any passive release that might have occured.
277 * We must reclaim the current-process designation before we can return
278 * to usermode. We also handle both LWKT and USER reschedule requests.
281 userexit(struct lwp *lp)
283 struct thread *td = lp->lwp_thread;
284 /* globaldata_t gd = td->td_gd;*/
287 * Handle stop requests at kernel priority. Any requests queued
288 * after this loop will generate another AST.
290 while (lp->lwp_proc->p_stat == SSTOP) {
297 * Reduce our priority in preparation for a return to userland. If
298 * our passive release function was still in place, our priority was
299 * never raised and does not need to be reduced.
301 lwkt_passive_recover(td);
304 * Become the current user scheduled process if we aren't already,
305 * and deal with reschedule requests and other factors.
307 lp->lwp_proc->p_usched->acquire_curproc(lp);
308 /* WARNING: we may have migrated cpu's */
309 /* gd = td->td_gd; */
312 #if !defined(KTR_KERNENTRY)
313 #define KTR_KERNENTRY KTR_ALL
315 KTR_INFO_MASTER(kernentry);
316 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, "STR",
317 sizeof(long) + sizeof(long) + sizeof(long) + sizeof(vm_offset_t));
318 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "STR",
319 sizeof(long) + sizeof(long));
320 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "STR",
321 sizeof(long) + sizeof(long) + sizeof(long));
322 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "STR",
323 sizeof(long) + sizeof(long) + sizeof(long));
324 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "STR",
325 sizeof(long) + sizeof(long));
328 * Exception, fault, and trap interface to the kernel.
329 * This common code is called from assembly language IDT gate entry
330 * routines that prepare a suitable stack frame, and restore this
331 * frame after the exception has been processed.
333 * This function is also called from doreti in an interlock to handle ASTs.
334 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
336 * NOTE! We have to retrieve the fault address prior to obtaining the
337 * MP lock because get_mplock() may switch out. YYY cr2 really ought
338 * to be retrieved by the assembly code, not here.
340 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
341 * if an attempt is made to switch from a fast interrupt or IPI. This is
342 * necessary to properly take fatal kernel traps on SMP machines if
343 * get_mplock() has to block.
347 trap(struct trapframe *frame)
349 struct globaldata *gd = mycpu;
350 struct thread *td = gd->gd_curthread;
351 struct lwp *lp = td->td_lwp;
354 int i = 0, ucode = 0, type, code;
359 int crit_count = td->td_critcount;
367 kprintf0("\"%s\" type=%ld\n",
368 trap_msg[frame->tf_trapno], frame->tf_trapno);
369 kprintf0(" rip=%lx rsp=%lx\n", frame->tf_rip, frame->tf_rsp);
370 kprintf0(" err=%lx addr=%lx\n", frame->tf_err, frame->tf_addr);
371 kprintf0(" cs=%lx ss=%lx rflags=%lx\n", (unsigned long)frame->tf_cs, (unsigned long)frame->tf_ss, frame->tf_rflags);
376 ++gd->gd_trap_nesting_level;
377 MAKEMPSAFE(have_mplock);
378 trap_fatal(frame, frame->tf_addr);
379 --gd->gd_trap_nesting_level;
385 eva = (frame->tf_trapno == T_PAGEFLT ? frame->tf_addr : 0);
386 ++gd->gd_trap_nesting_level;
387 MAKEMPSAFE(have_mplock);
388 trap_fatal(frame, eva);
389 --gd->gd_trap_nesting_level;
396 if ((frame->tf_rflags & PSL_I) == 0) {
398 * Buggy application or kernel code has disabled interrupts
399 * and then trapped. Enabling interrupts now is wrong, but
400 * it is better than running with interrupts disabled until
401 * they are accidentally enabled later.
403 type = frame->tf_trapno;
404 if (ISPL(frame->tf_cs) == SEL_UPL) {
405 MAKEMPSAFE(have_mplock);
406 /* JG curproc can be NULL */
408 "pid %ld (%s): trap %d with interrupts disabled\n",
409 (long)curproc->p_pid, curproc->p_comm, type);
410 } else if (type != T_NMI && type != T_BPTFLT &&
413 * XXX not quite right, since this may be for a
414 * multiple fault in user mode.
416 MAKEMPSAFE(have_mplock);
417 kprintf("kernel trap %d with interrupts disabled\n",
423 type = frame->tf_trapno;
424 code = frame->tf_err;
426 if (ISPL(frame->tf_cs) == SEL_UPL) {
429 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid,
430 frame->tf_trapno, eva);
434 sticks = (int)td->td_sticks;
435 lp->lwp_md.md_regs = frame;
438 case T_PRIVINFLT: /* privileged instruction fault */
443 case T_BPTFLT: /* bpt instruction fault */
444 case T_TRCTRAP: /* trace trap */
445 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 */
477 case T_SEGNPFLT: /* segment not present fault */
481 case T_TSSFLT: /* invalid TSS fault */
482 case T_DOUBLEFLT: /* double fault */
487 ucode = code + BUS_SEGM_FAULT ; /* XXX: ???*/
493 case T_PAGEFLT: /* page fault */
494 MAKEMPSAFE(have_mplock);
495 i = trap_pfault(frame, TRUE);
496 if (frame->tf_rip == 0)
497 kprintf("T_PAGEFLT: Warning %%rip == 0!\n");
512 case T_DIVIDE: /* integer divide fault */
518 MAKEMPSAFE(have_mplock);
519 /* machine/parity/power fail/"kitchen sink" faults */
520 if (isa_nmi(code) == 0) {
523 * NMI can be hooked up to a pushbutton
527 kprintf ("NMI ... going to debugger\n");
528 kdb_trap(type, 0, frame);
532 } else if (panic_on_nmi)
533 panic("NMI indicates hardware failure");
536 case T_OFLOW: /* integer overflow fault */
541 case T_BOUND: /* bounds check fault */
548 * Virtual kernel intercept - pass the DNA exception
549 * to the virtual kernel if it asked to handle it.
550 * This occurs when the virtual kernel is holding
551 * onto the FP context for a different emulated
552 * process then the one currently running.
554 * We must still call npxdna() since we may have
555 * saved FP state that the virtual kernel needs
556 * to hand over to a different emulated process.
558 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
559 (td->td_pcb->pcb_flags & FP_VIRTFP)
566 * The kernel may have switched out the FP unit's
567 * state, causing the user process to take a fault
568 * when it tries to use the FP unit. Restore the
574 ucode = FPE_FPU_NP_TRAP;
577 case T_FPOPFLT: /* FPU operand fetch fault */
582 case T_XMMFLT: /* SIMD floating-point exception */
591 case T_PAGEFLT: /* page fault */
592 MAKEMPSAFE(have_mplock);
593 trap_pfault(frame, FALSE);
598 * The kernel is apparently using fpu for copying.
599 * XXX this should be fatal unless the kernel has
600 * registered such use.
606 case T_STKFLT: /* stack fault */
609 case T_PROTFLT: /* general protection fault */
610 case T_SEGNPFLT: /* segment not present fault */
612 * Invalid segment selectors and out of bounds
613 * %rip's and %rsp's can be set up in user mode.
614 * This causes a fault in kernel mode when the
615 * kernel tries to return to user mode. We want
616 * to get this fault so that we can fix the
617 * problem here and not have to check all the
618 * selectors and pointers when the user changes
621 kprintf("trap.c line %d\n", __LINE__);
622 if (mycpu->gd_intr_nesting_level == 0) {
623 if (td->td_pcb->pcb_onfault) {
624 frame->tf_rip = (register_t)
625 td->td_pcb->pcb_onfault;
628 if (frame->tf_rip == (long)doreti_iret) {
629 frame->tf_rip = (long)doreti_iret_fault;
637 * PSL_NT can be set in user mode and isn't cleared
638 * automatically when the kernel is entered. This
639 * causes a TSS fault when the kernel attempts to
640 * `iret' because the TSS link is uninitialized. We
641 * want to get this fault so that we can fix the
642 * problem here and not every time the kernel is
645 if (frame->tf_rflags & PSL_NT) {
646 frame->tf_rflags &= ~PSL_NT;
651 case T_TRCTRAP: /* trace trap */
653 if (frame->tf_rip == (int)IDTVEC(syscall)) {
655 * We've just entered system mode via the
656 * syscall lcall. Continue single stepping
657 * silently until the syscall handler has
662 if (frame->tf_rip == (int)IDTVEC(syscall) + 1) {
664 * The syscall handler has now saved the
665 * flags. Stop single stepping it.
667 frame->tf_rflags &= ~PSL_T;
673 * Ignore debug register trace traps due to
674 * accesses in the user's address space, which
675 * can happen under several conditions such as
676 * if a user sets a watchpoint on a buffer and
677 * then passes that buffer to a system call.
678 * We still want to get TRCTRAPS for addresses
679 * in kernel space because that is useful when
680 * debugging the kernel.
683 if (user_dbreg_trap()) {
685 * Reset breakpoint bits because the
688 /* XXX check upper bits here */
689 load_dr6(rdr6() & 0xfffffff0);
694 * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
698 * If DDB is enabled, let it handle the debugger trap.
699 * Otherwise, debugger traps "can't happen".
703 MAKEMPSAFE(have_mplock);
704 if (kdb_trap(type, 0, frame))
710 MAKEMPSAFE(have_mplock);
711 /* machine/parity/power fail/"kitchen sink" faults */
713 if (isa_nmi(code) == 0) {
716 * NMI can be hooked up to a pushbutton
720 kprintf ("NMI ... going to debugger\n");
721 kdb_trap(type, 0, frame);
725 } else if (panic_on_nmi == 0)
728 #endif /* NISA > 0 */
730 MAKEMPSAFE(have_mplock);
731 trap_fatal(frame, 0);
736 * Virtual kernel intercept - if the fault is directly related to a
737 * VM context managed by a virtual kernel then let the virtual kernel
740 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
741 vkernel_trap(lp, frame);
746 * Translate fault for emulators (e.g. Linux)
748 if (*p->p_sysent->sv_transtrap)
749 i = (*p->p_sysent->sv_transtrap)(i, type);
751 MAKEMPSAFE(have_mplock);
752 trapsignal(lp, i, ucode);
755 if (type <= MAX_TRAP_MSG) {
756 uprintf("fatal process exception: %s",
758 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
759 uprintf(", fault VA = 0x%lx", frame->tf_addr);
766 if (ISPL(frame->tf_cs) == SEL_UPL)
767 KASSERT(td->td_mpcount == have_mplock, ("badmpcount trap/end from %p", (void *)frame->tf_rip));
769 userret(lp, frame, sticks);
776 if (p != NULL && lp != NULL)
777 KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid);
779 KASSERT(crit_count == td->td_critcount,
780 ("syscall: critical section count mismatch! %d/%d",
781 crit_count, td->td_pri));
786 trap_pfault(struct trapframe *frame, int usermode)
789 struct vmspace *vm = NULL;
794 thread_t td = curthread;
795 struct lwp *lp = td->td_lwp;
797 va = trunc_page(frame->tf_addr);
798 if (va >= VM_MIN_KERNEL_ADDRESS) {
800 * 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;
822 * PGEX_I is defined only if the execute disable bit capability is
823 * supported and enabled.
825 if (frame->tf_err & PGEX_W)
826 ftype = VM_PROT_WRITE;
828 else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
829 ftype = VM_PROT_EXECUTE;
832 ftype = VM_PROT_READ;
834 if (map != &kernel_map) {
836 * Keep swapout from messing with us during this
846 fault_flags |= VM_FAULT_BURST;
847 if (ftype & VM_PROT_WRITE)
848 fault_flags |= VM_FAULT_DIRTY;
850 fault_flags |= VM_FAULT_NORMAL;
851 rv = vm_fault(map, va, ftype, fault_flags);
856 * Don't have to worry about process locking or stacks
859 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
862 if (rv == KERN_SUCCESS)
866 if (td->td_gd->gd_intr_nesting_level == 0 &&
867 td->td_pcb->pcb_onfault) {
868 frame->tf_rip = (register_t)td->td_pcb->pcb_onfault;
871 trap_fatal(frame, frame->tf_addr);
876 * NOTE: on x86_64 we have a tf_addr field in the trapframe, no
877 * kludge is needed to pass the fault address to signal handlers.
879 struct proc *p = td->td_proc;
880 if (td->td_lwp->lwp_vkernel == NULL) {
881 kprintf("seg-fault accessing address %p rip=%p pid=%d p_comm=%s\n",
882 (void *)va, (void *)frame->tf_rip, p->p_pid, p->p_comm);
884 /* Debugger("seg-fault"); */
886 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
890 trap_fatal(struct trapframe *frame, vm_offset_t eva)
895 struct soft_segment_descriptor softseg;
898 code = frame->tf_err;
899 type = frame->tf_trapno;
900 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)], &softseg);
902 if (type <= MAX_TRAP_MSG)
903 msg = trap_msg[type];
906 kprintf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
907 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
909 /* three separate prints in case of a trap on an unmapped page */
910 kprintf("mp_lock = %08x; ", mp_lock);
911 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
912 kprintf("lapic->id = %08x\n", lapic->id);
914 if (type == T_PAGEFLT) {
915 kprintf("fault virtual address = 0x%lx\n", eva);
916 kprintf("fault code = %s %s %s, %s\n",
917 code & PGEX_U ? "user" : "supervisor",
918 code & PGEX_W ? "write" : "read",
919 code & PGEX_I ? "instruction" : "data",
920 code & PGEX_P ? "protection violation" : "page not present");
922 kprintf("instruction pointer = 0x%lx:0x%lx\n",
923 frame->tf_cs & 0xffff, frame->tf_rip);
924 if (ISPL(frame->tf_cs) == SEL_UPL) {
925 ss = frame->tf_ss & 0xffff;
928 ss = GSEL(GDATA_SEL, SEL_KPL);
929 rsp = (long)&frame->tf_rsp;
931 kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp);
932 kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp);
933 kprintf("code segment = base 0x%lx, limit 0x%lx, type 0x%x\n",
934 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
935 kprintf(" = DPL %d, pres %d, long %d, def32 %d, gran %d\n",
936 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32,
938 kprintf("processor eflags = ");
939 if (frame->tf_rflags & PSL_T)
940 kprintf("trace trap, ");
941 if (frame->tf_rflags & PSL_I)
942 kprintf("interrupt enabled, ");
943 if (frame->tf_rflags & PSL_NT)
944 kprintf("nested task, ");
945 if (frame->tf_rflags & PSL_RF)
947 kprintf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12);
948 kprintf("current process = ");
951 (u_long)curproc->p_pid);
955 kprintf("current thread = pri %d ", curthread->td_pri);
956 if (curthread->td_critcount)
961 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
964 kprintf("trap number = %d\n", type);
965 if (type <= MAX_TRAP_MSG)
966 panic("%s", trap_msg[type]);
968 panic("unknown/reserved trap");
972 * Double fault handler. Called when a fault occurs while writing
973 * a frame for a trap/exception onto the stack. This usually occurs
974 * when the stack overflows (such is the case with infinite recursion,
978 dblfault_handler(struct trapframe *frame)
980 kprintf0("DOUBLE FAULT\n");
981 kprintf("\nFatal double fault\n");
982 kprintf("rip = 0x%lx\n", frame->tf_rip);
983 kprintf("rsp = 0x%lx\n", frame->tf_rsp);
984 kprintf("rbp = 0x%lx\n", frame->tf_rbp);
986 /* three separate prints in case of a trap on an unmapped page */
987 kprintf("mp_lock = %08x; ", mp_lock);
988 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
989 kprintf("lapic->id = %08x\n", lapic->id);
991 panic("double fault");
995 * syscall2 - MP aware system call request C handler
997 * A system call is essentially treated as a trap except that the
998 * MP lock is not held on entry or return. We are responsible for
999 * obtaining the MP lock if necessary and for handling ASTs
1000 * (e.g. a task switch) prior to return.
1005 syscall2(struct trapframe *frame)
1007 struct thread *td = curthread;
1008 struct proc *p = td->td_proc;
1009 struct lwp *lp = td->td_lwp;
1011 struct sysent *callp;
1012 register_t orig_tf_rflags;
1017 int crit_count = td->td_critcount;
1020 int have_mplock = 0;
1025 union sysunion args;
1026 register_t *argsdst;
1028 mycpu->gd_cnt.v_syscall++;
1031 if (ISPL(frame->tf_cs) != SEL_UPL) {
1038 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid,
1042 KASSERT(td->td_mpcount == 0,
1043 ("badmpcount syscall2 from %p", (void *)frame->tf_rip));
1045 userenter(td, p); /* lazy raise our priority */
1052 sticks = (int)td->td_sticks;
1053 orig_tf_rflags = frame->tf_rflags;
1056 * Virtual kernel intercept - if a VM context managed by a virtual
1057 * kernel issues a system call the virtual kernel handles it, not us.
1058 * Restore the virtual kernel context and return from its system
1059 * call. The current frame is copied out to the virtual kernel.
1061 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1062 vkernel_trap(lp, frame);
1063 error = EJUSTRETURN;
1068 * Get the system call parameters and account for time
1070 lp->lwp_md.md_regs = frame;
1071 params = (caddr_t)frame->tf_rsp + sizeof(register_t);
1072 code = frame->tf_rax;
1074 if (p->p_sysent->sv_prepsyscall) {
1075 (*p->p_sysent->sv_prepsyscall)(
1076 frame, (int *)(&args.nosys.sysmsg + 1),
1079 if (code == SYS_syscall || code == SYS___syscall) {
1080 code = frame->tf_rdi;
1086 if (p->p_sysent->sv_mask)
1087 code &= p->p_sysent->sv_mask;
1089 if (code >= p->p_sysent->sv_size)
1090 callp = &p->p_sysent->sv_table[0];
1092 callp = &p->p_sysent->sv_table[code];
1094 narg = callp->sy_narg & SYF_ARGMASK;
1097 * On x86_64 we get up to six arguments in registers. The rest are
1098 * on the stack. The first six members of 'struct trapframe' happen
1099 * to be the registers used to pass arguments, in exactly the right
1102 argp = &frame->tf_rdi;
1104 argsdst = (register_t *)(&args.nosys.sysmsg + 1);
1106 * JG can we overflow the space pointed to by 'argsdst'
1107 * either with 'bcopy' or with 'copyin'?
1109 bcopy(argp, argsdst, sizeof(register_t) * regcnt);
1111 * copyin is MP aware, but the tracing code is not
1113 if (narg > regcnt) {
1114 KASSERT(params != NULL, ("copyin args with no params!"));
1115 error = copyin(params, &argsdst[regcnt],
1116 (narg - regcnt) * sizeof(register_t));
1119 if (KTRPOINT(td, KTR_SYSCALL)) {
1120 MAKEMPSAFE(have_mplock);
1122 ktrsyscall(lp, code, narg,
1123 (void *)(&args.nosys.sysmsg + 1));
1131 if (KTRPOINT(td, KTR_SYSCALL)) {
1132 MAKEMPSAFE(have_mplock);
1133 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1138 * Default return value is 0 (will be copied to %rax). Double-value
1139 * returns use %rax and %rdx. %rdx is left unchanged for system
1140 * calls which return only one result.
1142 args.sysmsg_fds[0] = 0;
1143 args.sysmsg_fds[1] = frame->tf_rdx;
1146 * The syscall might manipulate the trap frame. If it does it
1147 * will probably return EJUSTRETURN.
1149 args.sysmsg_frame = frame;
1151 STOPEVENT(p, S_SCE, narg); /* MP aware */
1154 * NOTE: All system calls run MPSAFE now. The system call itself
1155 * is responsible for getting the MP lock.
1157 error = (*callp->sy_call)(&args);
1161 * MP SAFE (we may or may not have the MP lock at this point)
1163 //kprintf("SYSMSG %d ", error);
1167 * Reinitialize proc pointer `p' as it may be different
1168 * if this is a child returning from fork syscall.
1171 lp = curthread->td_lwp;
1172 frame->tf_rax = args.sysmsg_fds[0];
1173 frame->tf_rdx = args.sysmsg_fds[1];
1174 frame->tf_rflags &= ~PSL_C;
1178 * Reconstruct pc, we know that 'syscall' is 2 bytes.
1179 * We have to do a full context restore so that %r10
1180 * (which was holding the value of %rcx) is restored for
1181 * the next iteration.
1183 frame->tf_rip -= frame->tf_err;
1184 frame->tf_r10 = frame->tf_rcx;
1189 panic("Unexpected EASYNC return value (for now)");
1192 if (p->p_sysent->sv_errsize) {
1193 if (error >= p->p_sysent->sv_errsize)
1194 error = -1; /* XXX */
1196 error = p->p_sysent->sv_errtbl[error];
1198 frame->tf_rax = error;
1199 frame->tf_rflags |= PSL_C;
1204 * Traced syscall. trapsignal() is not MP aware.
1206 if (orig_tf_rflags & PSL_T) {
1207 MAKEMPSAFE(have_mplock);
1208 frame->tf_rflags &= ~PSL_T;
1209 trapsignal(lp, SIGTRAP, TRAP_TRACE);
1213 * Handle reschedule and other end-of-syscall issues
1215 userret(lp, frame, sticks);
1218 if (KTRPOINT(td, KTR_SYSRET)) {
1219 MAKEMPSAFE(have_mplock);
1220 ktrsysret(lp, code, error, args.sysmsg_result);
1225 * This works because errno is findable through the
1226 * register set. If we ever support an emulation where this
1227 * is not the case, this code will need to be revisited.
1229 STOPEVENT(p, S_SCX, code);
1234 * Release the MP lock if we had to get it
1236 KASSERT(td->td_mpcount == have_mplock,
1237 ("badmpcount syscall2/end from %p", (void *)frame->tf_rip));
1241 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error);
1243 KASSERT(crit_count == td->td_critcount,
1244 ("syscall: critical section count mismatch! %d/%d",
1245 crit_count, td->td_pri));
1250 * NOTE: mplock not held at any point
1253 fork_return(struct lwp *lp, struct trapframe *frame)
1255 frame->tf_rax = 0; /* Child returns zero */
1256 frame->tf_rflags &= ~PSL_C; /* success */
1259 generic_lwp_return(lp, frame);
1260 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1264 * Simplified back end of syscall(), used when returning from fork()
1265 * directly into user mode.
1267 * This code will return back into the fork trampoline code which then
1270 * NOTE: The mplock is not held at any point.
1273 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1275 struct proc *p = lp->lwp_proc;
1278 * Newly forked processes are given a kernel priority. We have to
1279 * adjust the priority to a normal user priority and fake entry
1280 * into the kernel (call userenter()) to install a passive release
1281 * function just in case userret() decides to stop the process. This
1282 * can occur when ^Z races a fork. If we do not install the passive
1283 * release function the current process designation will not be
1284 * released when the thread goes to sleep.
1286 lwkt_setpri_self(TDPRI_USER_NORM);
1287 userenter(lp->lwp_thread, p);
1288 userret(lp, frame, 0);
1290 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1291 ktrsysret(lp, SYS_fork, 0, 0);
1293 p->p_flag |= P_PASSIVE_ACQ;
1295 p->p_flag &= ~P_PASSIVE_ACQ;
1299 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1300 * fault (which is then passed back to the virtual kernel) if an attempt is
1301 * made to use the FP unit.
1303 * XXX this is a fairly big hack.
1306 set_vkernel_fp(struct trapframe *frame)
1308 struct thread *td = curthread;
1310 if (frame->tf_xflags & PGEX_FPFAULT) {
1311 td->td_pcb->pcb_flags |= FP_VIRTFP;
1312 if (mdcpu->gd_npxthread == td)
1315 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1320 * Called from vkernel_trap() to fixup the vkernel's syscall
1321 * frame for vmspace_ctl() return.
1324 cpu_vkernel_trap(struct trapframe *frame, int error)
1326 frame->tf_rax = error;
1328 frame->tf_rflags |= PSL_C;
1330 frame->tf_rflags &= ~PSL_C;