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>
54 #include <sys/kerneldump.h>
56 #include <sys/pioctl.h>
57 #include <sys/types.h>
58 #include <sys/signal2.h>
59 #include <sys/syscall.h>
60 #include <sys/sysctl.h>
61 #include <sys/sysent.h>
62 #include <sys/systm.h>
64 #include <sys/ktrace.h>
67 #include <sys/sysmsg.h>
68 #include <sys/sysproto.h>
69 #include <sys/sysunion.h>
73 #include <vm/vm_extern.h>
74 #include <vm/vm_kern.h>
75 #include <vm/vm_param.h>
76 #include <machine/cpu.h>
77 #include <machine/pcb.h>
78 #include <machine/smp.h>
79 #include <machine/thread.h>
80 #include <machine/vmparam.h>
81 #include <machine/md_var.h>
82 #include <machine_base/isa/isa_intr.h>
86 #include <sys/thread2.h>
87 #include <sys/mplock2.h>
91 #define MAKEMPSAFE(have_mplock) \
92 if (have_mplock == 0) { \
99 #define MAKEMPSAFE(have_mplock)
103 extern void trap(struct trapframe *frame);
105 static int trap_pfault(struct trapframe *, int);
106 static void trap_fatal(struct trapframe *, vm_offset_t);
107 void dblfault_handler(struct trapframe *frame);
109 #define MAX_TRAP_MSG 30
110 static char *trap_msg[] = {
112 "privileged instruction fault", /* 1 T_PRIVINFLT */
114 "breakpoint instruction fault", /* 3 T_BPTFLT */
117 "arithmetic trap", /* 6 T_ARITHTRAP */
118 "system forced exception", /* 7 T_ASTFLT */
120 "general protection fault", /* 9 T_PROTFLT */
121 "trace trap", /* 10 T_TRCTRAP */
123 "page fault", /* 12 T_PAGEFLT */
125 "alignment fault", /* 14 T_ALIGNFLT */
129 "integer divide fault", /* 18 T_DIVIDE */
130 "non-maskable interrupt trap", /* 19 T_NMI */
131 "overflow trap", /* 20 T_OFLOW */
132 "FPU bounds check fault", /* 21 T_BOUND */
133 "FPU device not available", /* 22 T_DNA */
134 "double fault", /* 23 T_DOUBLEFLT */
135 "FPU operand fetch fault", /* 24 T_FPOPFLT */
136 "invalid TSS fault", /* 25 T_TSSFLT */
137 "segment not present fault", /* 26 T_SEGNPFLT */
138 "stack fault", /* 27 T_STKFLT */
139 "machine check trap", /* 28 T_MCHK */
140 "SIMD floating-point exception", /* 29 T_XMMFLT */
141 "reserved (unknown) fault", /* 30 T_RESERVED */
145 static int ddb_on_nmi = 1;
146 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
147 &ddb_on_nmi, 0, "Go to DDB on NMI");
148 static int ddb_on_seg_fault = 0;
149 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_seg_fault, CTLFLAG_RW,
150 &ddb_on_seg_fault, 0, "Go to DDB on user seg-fault");
152 static int panic_on_nmi = 1;
153 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
154 &panic_on_nmi, 0, "Panic on NMI");
155 static int fast_release;
156 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
157 &fast_release, 0, "Passive Release was optimal");
158 static int slow_release;
159 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
160 &slow_release, 0, "Passive Release was nonoptimal");
163 * Passively intercepts the thread switch function to increase
164 * the thread priority from a user priority to a kernel priority, reducing
165 * syscall and trap overhead for the case where no switch occurs.
167 * Synchronizes td_ucred with p_ucred. This is used by system calls,
168 * signal handling, faults, AST traps, and anything else that enters the
169 * kernel from userland and provides the kernel with a stable read-only
170 * copy of the process ucred.
173 userenter(struct thread *curtd, struct proc *curp)
178 curtd->td_release = lwkt_passive_release;
180 if (curtd->td_ucred != curp->p_ucred) {
181 ncred = crhold(curp->p_ucred);
182 ocred = curtd->td_ucred;
183 curtd->td_ucred = ncred;
190 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
191 * must be completed before we can return to or try to return to userland.
193 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
194 * arithmatic on the delta calculation so the absolute tick values are
195 * truncated to an integer.
198 userret(struct lwp *lp, struct trapframe *frame, int sticks)
200 struct proc *p = lp->lwp_proc;
204 * Charge system time if profiling. Note: times are in microseconds.
205 * This may do a copyout and block, so do it first even though it
206 * means some system time will be charged as user time.
208 if (p->p_flag & P_PROFIL) {
209 addupc_task(p, frame->tf_rip,
210 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
215 * If the jungle wants us dead, so be it.
217 if (lp->lwp_flag & LWP_WEXIT) {
218 lwkt_gettoken(&p->p_token);
220 lwkt_reltoken(&p->p_token); /* NOT REACHED */
224 * Block here if we are in a stopped state.
226 if (p->p_stat == SSTOP || dump_stop_usertds) {
234 * Post any pending upcalls. If running a virtual kernel be sure
235 * to restore the virtual kernel's vmspace before posting the upcall.
237 if (p->p_flag & P_UPCALLPEND) {
238 p->p_flag &= ~P_UPCALLPEND;
246 * Post any pending signals. If running a virtual kernel be sure
247 * to restore the virtual kernel's vmspace before posting the signal.
249 * WARNING! postsig() can exit and not return.
251 if ((sig = CURSIG_TRACE(lp)) != 0) {
259 * block here if we are swapped out, but still process signals
260 * (such as SIGKILL). proc0 (the swapin scheduler) is already
261 * aware of our situation, we do not have to wake it up.
263 if (p->p_flag & P_SWAPPEDOUT) {
265 p->p_flag |= P_SWAPWAIT;
267 if (p->p_flag & P_SWAPWAIT)
268 tsleep(p, PCATCH, "SWOUT", 0);
269 p->p_flag &= ~P_SWAPWAIT;
275 * Make sure postsig() handled request to restore old signal mask after
276 * running signal handler.
278 KKASSERT((lp->lwp_flag & LWP_OLDMASK) == 0);
282 * Cleanup from userenter and any passive release that might have occured.
283 * We must reclaim the current-process designation before we can return
284 * to usermode. We also handle both LWKT and USER reschedule requests.
287 userexit(struct lwp *lp)
289 struct thread *td = lp->lwp_thread;
290 /* globaldata_t gd = td->td_gd;*/
293 * Handle stop requests at kernel priority. Any requests queued
294 * after this loop will generate another AST.
296 while (lp->lwp_proc->p_stat == SSTOP) {
303 * Reduce our priority in preparation for a return to userland. If
304 * our passive release function was still in place, our priority was
305 * never raised and does not need to be reduced.
307 lwkt_passive_recover(td);
310 * Become the current user scheduled process if we aren't already,
311 * and deal with reschedule requests and other factors.
313 lp->lwp_proc->p_usched->acquire_curproc(lp);
314 /* WARNING: we may have migrated cpu's */
315 /* gd = td->td_gd; */
318 #if !defined(KTR_KERNENTRY)
319 #define KTR_KERNENTRY KTR_ALL
321 KTR_INFO_MASTER(kernentry);
322 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, "STR",
323 sizeof(long) + sizeof(long) + sizeof(long) + sizeof(vm_offset_t));
324 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "STR",
325 sizeof(long) + sizeof(long));
326 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "STR",
327 sizeof(long) + sizeof(long) + sizeof(long));
328 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "STR",
329 sizeof(long) + sizeof(long) + sizeof(long));
330 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "STR",
331 sizeof(long) + sizeof(long));
334 * Exception, fault, and trap interface to the kernel.
335 * This common code is called from assembly language IDT gate entry
336 * routines that prepare a suitable stack frame, and restore this
337 * frame after the exception has been processed.
339 * This function is also called from doreti in an interlock to handle ASTs.
340 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
342 * NOTE! We have to retrieve the fault address prior to obtaining the
343 * MP lock because get_mplock() may switch out. YYY cr2 really ought
344 * to be retrieved by the assembly code, not here.
346 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
347 * if an attempt is made to switch from a fast interrupt or IPI. This is
348 * necessary to properly take fatal kernel traps on SMP machines if
349 * get_mplock() has to block.
353 trap(struct trapframe *frame)
355 struct globaldata *gd = mycpu;
356 struct thread *td = gd->gd_curthread;
357 struct lwp *lp = td->td_lwp;
360 int i = 0, ucode = 0, type, code;
365 int crit_count = td->td_critcount;
366 lwkt_tokref_t curstop = td->td_toks_stop;
374 * We need to allow T_DNA faults when the debugger is active since
375 * some dumping paths do large bcopy() which use the floating
376 * point registers for faster copying.
378 if (db_active && frame->tf_trapno != T_DNA) {
379 eva = (frame->tf_trapno == T_PAGEFLT ? frame->tf_addr : 0);
380 ++gd->gd_trap_nesting_level;
381 MAKEMPSAFE(have_mplock);
382 trap_fatal(frame, eva);
383 --gd->gd_trap_nesting_level;
390 if ((frame->tf_rflags & PSL_I) == 0) {
392 * Buggy application or kernel code has disabled interrupts
393 * and then trapped. Enabling interrupts now is wrong, but
394 * it is better than running with interrupts disabled until
395 * they are accidentally enabled later.
397 type = frame->tf_trapno;
398 if (ISPL(frame->tf_cs) == SEL_UPL) {
399 MAKEMPSAFE(have_mplock);
400 /* JG curproc can be NULL */
402 "pid %ld (%s): trap %d with interrupts disabled\n",
403 (long)curproc->p_pid, curproc->p_comm, type);
404 } else if (type != T_NMI && type != T_BPTFLT &&
407 * XXX not quite right, since this may be for a
408 * multiple fault in user mode.
410 MAKEMPSAFE(have_mplock);
411 kprintf("kernel trap %d with interrupts disabled\n",
417 type = frame->tf_trapno;
418 code = frame->tf_err;
420 if (ISPL(frame->tf_cs) == SEL_UPL) {
423 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid,
424 frame->tf_trapno, eva);
428 sticks = (int)td->td_sticks;
429 KASSERT(lp->lwp_md.md_regs == frame,
430 ("Frame mismatch %p %p", lp->lwp_md.md_regs, frame));
433 case T_PRIVINFLT: /* privileged instruction fault */
438 case T_BPTFLT: /* bpt instruction fault */
439 case T_TRCTRAP: /* trace trap */
440 frame->tf_rflags &= ~PSL_T;
445 case T_ARITHTRAP: /* arithmetic trap */
458 case T_ASTFLT: /* Allow process switch */
459 mycpu->gd_cnt.v_soft++;
460 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
461 atomic_clear_int(&mycpu->gd_reqflags,
463 addupc_task(p, p->p_prof.pr_addr,
468 case T_PROTFLT: /* general protection fault */
472 case T_SEGNPFLT: /* segment not present fault */
476 case T_TSSFLT: /* invalid TSS fault */
477 case T_DOUBLEFLT: /* double fault */
482 ucode = code + BUS_SEGM_FAULT ; /* XXX: ???*/
488 case T_PAGEFLT: /* page fault */
489 i = trap_pfault(frame, TRUE);
490 if (frame->tf_rip == 0)
491 kprintf("T_PAGEFLT: Warning %%rip == 0!\n");
506 case T_DIVIDE: /* integer divide fault */
512 MAKEMPSAFE(have_mplock);
513 /* machine/parity/power fail/"kitchen sink" faults */
514 if (isa_nmi(code) == 0) {
517 * NMI can be hooked up to a pushbutton
521 kprintf ("NMI ... going to debugger\n");
522 kdb_trap(type, 0, frame);
526 } else if (panic_on_nmi)
527 panic("NMI indicates hardware failure");
530 case T_OFLOW: /* integer overflow fault */
535 case T_BOUND: /* bounds check fault */
542 * Virtual kernel intercept - pass the DNA exception
543 * to the virtual kernel if it asked to handle it.
544 * This occurs when the virtual kernel is holding
545 * onto the FP context for a different emulated
546 * process then the one currently running.
548 * We must still call npxdna() since we may have
549 * saved FP state that the virtual kernel needs
550 * to hand over to a different emulated process.
552 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
553 (td->td_pcb->pcb_flags & FP_VIRTFP)
560 * The kernel may have switched out the FP unit's
561 * state, causing the user process to take a fault
562 * when it tries to use the FP unit. Restore the
568 ucode = FPE_FPU_NP_TRAP;
571 case T_FPOPFLT: /* FPU operand fetch fault */
576 case T_XMMFLT: /* SIMD floating-point exception */
585 case T_PAGEFLT: /* page fault */
586 trap_pfault(frame, FALSE);
591 * The kernel is apparently using fpu for copying.
592 * XXX this should be fatal unless the kernel has
593 * registered such use.
599 case T_STKFLT: /* stack fault */
602 case T_PROTFLT: /* general protection fault */
603 case T_SEGNPFLT: /* segment not present fault */
605 * Invalid segment selectors and out of bounds
606 * %rip's and %rsp's can be set up in user mode.
607 * This causes a fault in kernel mode when the
608 * kernel tries to return to user mode. We want
609 * to get this fault so that we can fix the
610 * problem here and not have to check all the
611 * selectors and pointers when the user changes
614 if (mycpu->gd_intr_nesting_level == 0) {
615 if (td->td_pcb->pcb_onfault) {
616 frame->tf_rip = (register_t)
617 td->td_pcb->pcb_onfault;
620 if (frame->tf_rip == (long)doreti_iret) {
621 frame->tf_rip = (long)doreti_iret_fault;
629 * PSL_NT can be set in user mode and isn't cleared
630 * automatically when the kernel is entered. This
631 * causes a TSS fault when the kernel attempts to
632 * `iret' because the TSS link is uninitialized. We
633 * want to get this fault so that we can fix the
634 * problem here and not every time the kernel is
637 if (frame->tf_rflags & PSL_NT) {
638 frame->tf_rflags &= ~PSL_NT;
643 case T_TRCTRAP: /* trace trap */
645 if (frame->tf_rip == (int)IDTVEC(syscall)) {
647 * We've just entered system mode via the
648 * syscall lcall. Continue single stepping
649 * silently until the syscall handler has
654 if (frame->tf_rip == (int)IDTVEC(syscall) + 1) {
656 * The syscall handler has now saved the
657 * flags. Stop single stepping it.
659 frame->tf_rflags &= ~PSL_T;
665 * Ignore debug register trace traps due to
666 * accesses in the user's address space, which
667 * can happen under several conditions such as
668 * if a user sets a watchpoint on a buffer and
669 * then passes that buffer to a system call.
670 * We still want to get TRCTRAPS for addresses
671 * in kernel space because that is useful when
672 * debugging the kernel.
675 if (user_dbreg_trap()) {
677 * Reset breakpoint bits because the
680 /* XXX check upper bits here */
681 load_dr6(rdr6() & 0xfffffff0);
686 * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
690 * If DDB is enabled, let it handle the debugger trap.
691 * Otherwise, debugger traps "can't happen".
695 MAKEMPSAFE(have_mplock);
696 if (kdb_trap(type, 0, frame))
702 MAKEMPSAFE(have_mplock);
703 /* machine/parity/power fail/"kitchen sink" faults */
705 if (isa_nmi(code) == 0) {
708 * NMI can be hooked up to a pushbutton
712 kprintf ("NMI ... going to debugger\n");
713 kdb_trap(type, 0, frame);
717 } else if (panic_on_nmi == 0)
720 #endif /* NISA > 0 */
722 MAKEMPSAFE(have_mplock);
723 trap_fatal(frame, 0);
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);
757 userret(lp, frame, sticks);
764 if (p != NULL && lp != NULL)
765 KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid);
767 KASSERT(crit_count == td->td_critcount,
768 ("trap: critical section count mismatch! %d/%d",
769 crit_count, td->td_pri));
770 KASSERT(curstop == td->td_toks_stop,
771 ("trap: extra tokens held after trap! %ld/%ld",
772 curstop - &td->td_toks_base,
773 td->td_toks_stop - &td->td_toks_base));
778 trap_pfault(struct trapframe *frame, int usermode)
781 struct vmspace *vm = NULL;
786 thread_t td = curthread;
787 struct lwp *lp = td->td_lwp;
790 va = trunc_page(frame->tf_addr);
791 if (va >= VM_MIN_KERNEL_ADDRESS) {
793 * Don't allow user-mode faults in kernel address space.
804 * This is a fault on non-kernel virtual memory.
805 * vm is initialized above to NULL. If curproc is NULL
806 * or curproc->p_vmspace is NULL the fault is fatal.
809 vm = lp->lwp_vmspace;
821 * PGEX_I is defined only if the execute disable bit capability is
822 * supported and enabled.
824 if (frame->tf_err & PGEX_W)
825 ftype = VM_PROT_WRITE;
827 else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
828 ftype = VM_PROT_EXECUTE;
831 ftype = VM_PROT_READ;
833 if (map != &kernel_map) {
835 * Keep swapout from messing with us during this
845 fault_flags |= VM_FAULT_BURST;
846 if (ftype & VM_PROT_WRITE)
847 fault_flags |= VM_FAULT_DIRTY;
849 fault_flags |= VM_FAULT_NORMAL;
850 rv = vm_fault(map, va, ftype, fault_flags);
855 * Don't have to worry about process locking or stacks
858 fault_flags = VM_FAULT_NORMAL;
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.
880 if (td->td_lwp->lwp_vkernel == NULL) {
882 kprintf("seg-fault ft=%04x ff=%04x addr=%p rip=%p "
883 "pid=%d p_comm=%s\n",
885 (void *)frame->tf_addr,
886 (void *)frame->tf_rip,
887 p->p_pid, p->p_comm);
889 if (ddb_on_seg_fault)
890 Debugger("ddb_on_seg_fault");
894 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
898 trap_fatal(struct trapframe *frame, vm_offset_t eva)
903 struct soft_segment_descriptor softseg;
906 code = frame->tf_err;
907 type = frame->tf_trapno;
908 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)], &softseg);
910 if (type <= MAX_TRAP_MSG)
911 msg = trap_msg[type];
914 kprintf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
915 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
917 /* three separate prints in case of a trap on an unmapped page */
918 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
919 kprintf("lapic->id = %08x\n", lapic->id);
921 if (type == T_PAGEFLT) {
922 kprintf("fault virtual address = 0x%lx\n", eva);
923 kprintf("fault code = %s %s %s, %s\n",
924 code & PGEX_U ? "user" : "supervisor",
925 code & PGEX_W ? "write" : "read",
926 code & PGEX_I ? "instruction" : "data",
927 code & PGEX_P ? "protection violation" : "page not present");
929 kprintf("instruction pointer = 0x%lx:0x%lx\n",
930 frame->tf_cs & 0xffff, frame->tf_rip);
931 if (ISPL(frame->tf_cs) == SEL_UPL) {
932 ss = frame->tf_ss & 0xffff;
935 ss = GSEL(GDATA_SEL, SEL_KPL);
936 rsp = (long)&frame->tf_rsp;
938 kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp);
939 kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp);
940 kprintf("code segment = base 0x%lx, limit 0x%lx, type 0x%x\n",
941 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
942 kprintf(" = DPL %d, pres %d, long %d, def32 %d, gran %d\n",
943 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32,
945 kprintf("processor eflags = ");
946 if (frame->tf_rflags & PSL_T)
947 kprintf("trace trap, ");
948 if (frame->tf_rflags & PSL_I)
949 kprintf("interrupt enabled, ");
950 if (frame->tf_rflags & PSL_NT)
951 kprintf("nested task, ");
952 if (frame->tf_rflags & PSL_RF)
954 kprintf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12);
955 kprintf("current process = ");
958 (u_long)curproc->p_pid);
962 kprintf("current thread = pri %d ", curthread->td_pri);
963 if (curthread->td_critcount)
968 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
971 kprintf("trap number = %d\n", type);
972 if (type <= MAX_TRAP_MSG)
973 panic("%s", trap_msg[type]);
975 panic("unknown/reserved trap");
979 * Double fault handler. Called when a fault occurs while writing
980 * a frame for a trap/exception onto the stack. This usually occurs
981 * when the stack overflows (such is the case with infinite recursion,
986 in_kstack_guard(register_t rptr)
988 thread_t td = curthread;
990 if ((char *)rptr >= td->td_kstack &&
991 (char *)rptr < td->td_kstack + PAGE_SIZE) {
998 dblfault_handler(struct trapframe *frame)
1000 thread_t td = curthread;
1002 if (in_kstack_guard(frame->tf_rsp) || in_kstack_guard(frame->tf_rbp)) {
1003 kprintf("DOUBLE FAULT - KERNEL STACK GUARD HIT!\n");
1004 if (in_kstack_guard(frame->tf_rsp))
1005 frame->tf_rsp = (register_t)(td->td_kstack + PAGE_SIZE);
1006 if (in_kstack_guard(frame->tf_rbp))
1007 frame->tf_rbp = (register_t)(td->td_kstack + PAGE_SIZE);
1009 kprintf("DOUBLE FAULT\n");
1011 kprintf("\nFatal double fault\n");
1012 kprintf("rip = 0x%lx\n", frame->tf_rip);
1013 kprintf("rsp = 0x%lx\n", frame->tf_rsp);
1014 kprintf("rbp = 0x%lx\n", frame->tf_rbp);
1016 /* three separate prints in case of a trap on an unmapped page */
1017 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
1018 kprintf("lapic->id = %08x\n", lapic->id);
1020 panic("double fault");
1024 * syscall2 - MP aware system call request C handler
1026 * A system call is essentially treated as a trap except that the
1027 * MP lock is not held on entry or return. We are responsible for
1028 * obtaining the MP lock if necessary and for handling ASTs
1029 * (e.g. a task switch) prior to return.
1034 syscall2(struct trapframe *frame)
1036 struct thread *td = curthread;
1037 struct proc *p = td->td_proc;
1038 struct lwp *lp = td->td_lwp;
1040 struct sysent *callp;
1041 register_t orig_tf_rflags;
1046 int crit_count = td->td_critcount;
1049 int have_mplock = 0;
1054 union sysunion args;
1055 register_t *argsdst;
1057 mycpu->gd_cnt.v_syscall++;
1060 if (ISPL(frame->tf_cs) != SEL_UPL) {
1067 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid,
1070 userenter(td, p); /* lazy raise our priority */
1077 sticks = (int)td->td_sticks;
1078 orig_tf_rflags = frame->tf_rflags;
1081 * Virtual kernel intercept - if a VM context managed by a virtual
1082 * kernel issues a system call the virtual kernel handles it, not us.
1083 * Restore the virtual kernel context and return from its system
1084 * call. The current frame is copied out to the virtual kernel.
1086 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1087 vkernel_trap(lp, frame);
1088 error = EJUSTRETURN;
1093 * Get the system call parameters and account for time
1095 KASSERT(lp->lwp_md.md_regs == frame,
1096 ("Frame mismatch %p %p", lp->lwp_md.md_regs, frame));
1097 params = (caddr_t)frame->tf_rsp + sizeof(register_t);
1098 code = frame->tf_rax;
1100 if (p->p_sysent->sv_prepsyscall) {
1101 (*p->p_sysent->sv_prepsyscall)(
1102 frame, (int *)(&args.nosys.sysmsg + 1),
1105 if (code == SYS_syscall || code == SYS___syscall) {
1106 code = frame->tf_rdi;
1112 if (p->p_sysent->sv_mask)
1113 code &= p->p_sysent->sv_mask;
1115 if (code >= p->p_sysent->sv_size)
1116 callp = &p->p_sysent->sv_table[0];
1118 callp = &p->p_sysent->sv_table[code];
1120 narg = callp->sy_narg & SYF_ARGMASK;
1123 * On x86_64 we get up to six arguments in registers. The rest are
1124 * on the stack. The first six members of 'struct trapframe' happen
1125 * to be the registers used to pass arguments, in exactly the right
1128 argp = &frame->tf_rdi;
1130 argsdst = (register_t *)(&args.nosys.sysmsg + 1);
1132 * JG can we overflow the space pointed to by 'argsdst'
1133 * either with 'bcopy' or with 'copyin'?
1135 bcopy(argp, argsdst, sizeof(register_t) * regcnt);
1137 * copyin is MP aware, but the tracing code is not
1139 if (narg > regcnt) {
1140 KASSERT(params != NULL, ("copyin args with no params!"));
1141 error = copyin(params, &argsdst[regcnt],
1142 (narg - regcnt) * sizeof(register_t));
1145 if (KTRPOINT(td, KTR_SYSCALL)) {
1146 MAKEMPSAFE(have_mplock);
1148 ktrsyscall(lp, code, narg,
1149 (void *)(&args.nosys.sysmsg + 1));
1157 if (KTRPOINT(td, KTR_SYSCALL)) {
1158 MAKEMPSAFE(have_mplock);
1159 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1164 * Default return value is 0 (will be copied to %rax). Double-value
1165 * returns use %rax and %rdx. %rdx is left unchanged for system
1166 * calls which return only one result.
1168 args.sysmsg_fds[0] = 0;
1169 args.sysmsg_fds[1] = frame->tf_rdx;
1172 * The syscall might manipulate the trap frame. If it does it
1173 * will probably return EJUSTRETURN.
1175 args.sysmsg_frame = frame;
1177 STOPEVENT(p, S_SCE, narg); /* MP aware */
1180 * NOTE: All system calls run MPSAFE now. The system call itself
1181 * is responsible for getting the MP lock.
1183 error = (*callp->sy_call)(&args);
1187 * MP SAFE (we may or may not have the MP lock at this point)
1189 //kprintf("SYSMSG %d ", error);
1193 * Reinitialize proc pointer `p' as it may be different
1194 * if this is a child returning from fork syscall.
1197 lp = curthread->td_lwp;
1198 frame->tf_rax = args.sysmsg_fds[0];
1199 frame->tf_rdx = args.sysmsg_fds[1];
1200 frame->tf_rflags &= ~PSL_C;
1204 * Reconstruct pc, we know that 'syscall' is 2 bytes.
1205 * We have to do a full context restore so that %r10
1206 * (which was holding the value of %rcx) is restored for
1207 * the next iteration.
1209 frame->tf_rip -= frame->tf_err;
1210 frame->tf_r10 = frame->tf_rcx;
1215 panic("Unexpected EASYNC return value (for now)");
1218 if (p->p_sysent->sv_errsize) {
1219 if (error >= p->p_sysent->sv_errsize)
1220 error = -1; /* XXX */
1222 error = p->p_sysent->sv_errtbl[error];
1224 frame->tf_rax = error;
1225 frame->tf_rflags |= PSL_C;
1230 * Traced syscall. trapsignal() is not MP aware.
1232 if (orig_tf_rflags & PSL_T) {
1233 MAKEMPSAFE(have_mplock);
1234 frame->tf_rflags &= ~PSL_T;
1235 trapsignal(lp, SIGTRAP, TRAP_TRACE);
1239 * Handle reschedule and other end-of-syscall issues
1241 userret(lp, frame, sticks);
1244 if (KTRPOINT(td, KTR_SYSRET)) {
1245 MAKEMPSAFE(have_mplock);
1246 ktrsysret(lp, code, error, args.sysmsg_result);
1251 * This works because errno is findable through the
1252 * register set. If we ever support an emulation where this
1253 * is not the case, this code will need to be revisited.
1255 STOPEVENT(p, S_SCX, code);
1260 * Release the MP lock if we had to get it
1265 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error);
1267 KASSERT(crit_count == td->td_critcount,
1268 ("syscall: critical section count mismatch! %d/%d",
1269 crit_count, td->td_pri));
1270 KASSERT(&td->td_toks_base == td->td_toks_stop,
1271 ("syscall: extra tokens held after trap! %ld",
1272 td->td_toks_stop - &td->td_toks_base));
1277 * NOTE: mplock not held at any point
1280 fork_return(struct lwp *lp, struct trapframe *frame)
1282 frame->tf_rax = 0; /* Child returns zero */
1283 frame->tf_rflags &= ~PSL_C; /* success */
1286 generic_lwp_return(lp, frame);
1287 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1291 * Simplified back end of syscall(), used when returning from fork()
1292 * directly into user mode.
1294 * This code will return back into the fork trampoline code which then
1297 * NOTE: The mplock is not held at any point.
1300 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1302 struct proc *p = lp->lwp_proc;
1305 * Newly forked processes are given a kernel priority. We have to
1306 * adjust the priority to a normal user priority and fake entry
1307 * into the kernel (call userenter()) to install a passive release
1308 * function just in case userret() decides to stop the process. This
1309 * can occur when ^Z races a fork. If we do not install the passive
1310 * release function the current process designation will not be
1311 * released when the thread goes to sleep.
1313 lwkt_setpri_self(TDPRI_USER_NORM);
1314 userenter(lp->lwp_thread, p);
1315 userret(lp, frame, 0);
1317 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1318 ktrsysret(lp, SYS_fork, 0, 0);
1320 p->p_flag |= P_PASSIVE_ACQ;
1322 p->p_flag &= ~P_PASSIVE_ACQ;
1326 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1327 * fault (which is then passed back to the virtual kernel) if an attempt is
1328 * made to use the FP unit.
1330 * XXX this is a fairly big hack.
1333 set_vkernel_fp(struct trapframe *frame)
1335 struct thread *td = curthread;
1337 if (frame->tf_xflags & PGEX_FPFAULT) {
1338 td->td_pcb->pcb_flags |= FP_VIRTFP;
1339 if (mdcpu->gd_npxthread == td)
1342 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1347 * Called from vkernel_trap() to fixup the vkernel's syscall
1348 * frame for vmspace_ctl() return.
1351 cpu_vkernel_trap(struct trapframe *frame, int error)
1353 frame->tf_rax = error;
1355 frame->tf_rflags |= PSL_C;
1357 frame->tf_rflags &= ~PSL_C;