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) 2008-2018 The DragonFly Project.
6 * Copyright (c) 2008 Jordan Gordeev.
8 * This code is derived from software contributed to Berkeley by
9 * the University of Utah, and William Jolitz.
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. All advertising materials mentioning features or use of this software
20 * must display the following acknowledgement:
21 * This product includes software developed by the University of
22 * California, Berkeley and its contributors.
23 * 4. Neither the name of the University nor the names of its contributors
24 * may be used to endorse or promote products derived from this software
25 * without specific prior written permission.
27 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39 * from: @(#)trap.c 7.4 (Berkeley) 5/13/91
40 * $FreeBSD: src/sys/i386/i386/trap.c,v 1.147.2.11 2003/02/27 19:09:59 luoqi Exp $
44 * x86_64 Trap and System call handling
50 #include "opt_ktrace.h"
52 #include <machine/frame.h>
53 #include <sys/param.h>
54 #include <sys/systm.h>
55 #include <sys/kernel.h>
56 #include <sys/kerneldump.h>
58 #include <sys/pioctl.h>
59 #include <sys/types.h>
60 #include <sys/signal2.h>
61 #include <sys/syscall.h>
62 #include <sys/sysctl.h>
63 #include <sys/sysent.h>
65 #include <sys/ktrace.h>
68 #include <sys/sysmsg.h>
69 #include <sys/sysproto.h>
70 #include <sys/sysunion.h>
74 #include <vm/vm_extern.h>
75 #include <vm/vm_kern.h>
76 #include <vm/vm_param.h>
77 #include <machine/cpu.h>
78 #include <machine/pcb.h>
79 #include <machine/smp.h>
80 #include <machine/thread.h>
81 #include <machine/clock.h>
82 #include <machine/vmparam.h>
83 #include <machine/md_var.h>
84 #include <machine_base/isa/isa_intr.h>
85 #include <machine_base/apic/lapic.h>
89 #include <sys/thread2.h>
90 #include <sys/spinlock2.h>
93 * These %rip's are used to detect a historical CPU artifact on syscall or
94 * int $3 entry, if not shortcutted in exception.S via
95 * DIRECT_DISALLOW_SS_CPUBUG.
97 extern void Xbpt(void);
98 extern void Xfast_syscall(void);
99 #define IDTVEC(vec) X##vec
101 extern void trap(struct trapframe *frame);
103 static int trap_pfault(struct trapframe *, int);
104 static void trap_fatal(struct trapframe *, vm_offset_t);
105 void dblfault_handler(struct trapframe *frame);
107 #define MAX_TRAP_MSG 30
108 static char *trap_msg[] = {
110 "privileged instruction fault", /* 1 T_PRIVINFLT */
112 "breakpoint instruction fault", /* 3 T_BPTFLT */
115 "arithmetic trap", /* 6 T_ARITHTRAP */
116 "system forced exception", /* 7 T_ASTFLT */
118 "general protection fault", /* 9 T_PROTFLT */
119 "trace trap", /* 10 T_TRCTRAP */
121 "page fault", /* 12 T_PAGEFLT */
123 "alignment fault", /* 14 T_ALIGNFLT */
127 "integer divide fault", /* 18 T_DIVIDE */
128 "non-maskable interrupt trap", /* 19 T_NMI */
129 "overflow trap", /* 20 T_OFLOW */
130 "FPU bounds check fault", /* 21 T_BOUND */
131 "FPU device not available", /* 22 T_DNA */
132 "double fault", /* 23 T_DOUBLEFLT */
133 "FPU operand fetch fault", /* 24 T_FPOPFLT */
134 "invalid TSS fault", /* 25 T_TSSFLT */
135 "segment not present fault", /* 26 T_SEGNPFLT */
136 "stack fault", /* 27 T_STKFLT */
137 "machine check trap", /* 28 T_MCHK */
138 "SIMD floating-point exception", /* 29 T_XMMFLT */
139 "reserved (unknown) fault", /* 30 T_RESERVED */
143 static int ddb_on_nmi = 1;
144 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
145 &ddb_on_nmi, 0, "Go to DDB on NMI");
146 static int ddb_on_seg_fault = 0;
147 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_seg_fault, CTLFLAG_RW,
148 &ddb_on_seg_fault, 0, "Go to DDB on user seg-fault");
149 static int freeze_on_seg_fault = 0;
150 SYSCTL_INT(_machdep, OID_AUTO, freeze_on_seg_fault, CTLFLAG_RW,
151 &freeze_on_seg_fault, 0, "Go to DDB on user seg-fault");
153 static int panic_on_nmi = 1;
154 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
155 &panic_on_nmi, 0, "Panic on NMI");
156 static int fast_release;
157 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
158 &fast_release, 0, "Passive Release was optimal");
159 static int slow_release;
160 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
161 &slow_release, 0, "Passive Release was nonoptimal");
164 * System call debugging records the worst-case system call
165 * overhead (inclusive of blocking), but may be inaccurate.
167 /*#define SYSCALL_DEBUG*/
169 uint64_t SysCallsWorstCase[SYS_MAXSYSCALL];
173 * Passively intercepts the thread switch function to increase
174 * the thread priority from a user priority to a kernel priority, reducing
175 * syscall and trap overhead for the case where no switch occurs.
177 * Synchronizes td_ucred with p_ucred. This is used by system calls,
178 * signal handling, faults, AST traps, and anything else that enters the
179 * kernel from userland and provides the kernel with a stable read-only
180 * copy of the process ucred.
182 * To avoid races with another thread updating p_ucred we obtain p_spin.
183 * The other thread doing the update will obtain both p_token and p_spin.
184 * In the case where the cached cred pointer matches, we will already have
185 * the ref and we don't have to do one blessed thing.
188 userenter(struct thread *curtd, struct proc *curp)
193 curtd->td_release = lwkt_passive_release;
195 if (curtd->td_ucred != curp->p_ucred) {
196 spin_lock(&curp->p_spin);
197 ncred = crhold(curp->p_ucred);
198 spin_unlock(&curp->p_spin);
199 ocred = curtd->td_ucred;
200 curtd->td_ucred = ncred;
207 * Debugging, remove top two user stack pages to catch kernel faults
209 if (freeze_on_seg_fault > 1 && curtd->td_lwp) {
210 pmap_remove(vmspace_pmap(curtd->td_lwp->lwp_vmspace),
211 0x00007FFFFFFFD000LU,
212 0x0000800000000000LU);
218 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
219 * must be completed before we can return to or try to return to userland.
221 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
222 * arithmatic on the delta calculation so the absolute tick values are
223 * truncated to an integer.
226 userret(struct lwp *lp, struct trapframe *frame, int sticks)
228 struct proc *p = lp->lwp_proc;
233 * Charge system time if profiling. Note: times are in microseconds.
234 * This may do a copyout and block, so do it first even though it
235 * means some system time will be charged as user time.
237 if (p->p_flags & P_PROFIL) {
238 addupc_task(p, frame->tf_rip,
239 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
244 * Specific on-return-to-usermode checks (LWP_MP_WEXIT,
245 * LWP_MP_VNLRU, etc).
247 if (lp->lwp_mpflags & LWP_MP_URETMASK)
251 * Block here if we are in a stopped state.
253 if (STOPLWP(p, lp)) {
254 lwkt_gettoken(&p->p_token);
256 lwkt_reltoken(&p->p_token);
259 while (dump_stop_usertds) {
260 tsleep(&dump_stop_usertds, 0, "dumpstp", 0);
264 * Post any pending upcalls. If running a virtual kernel be sure
265 * to restore the virtual kernel's vmspace before posting the upcall.
267 if (p->p_flags & (P_SIGVTALRM | P_SIGPROF)) {
268 lwkt_gettoken(&p->p_token);
269 if (p->p_flags & P_SIGVTALRM) {
270 p->p_flags &= ~P_SIGVTALRM;
271 ksignal(p, SIGVTALRM);
273 if (p->p_flags & P_SIGPROF) {
274 p->p_flags &= ~P_SIGPROF;
277 lwkt_reltoken(&p->p_token);
282 * Post any pending signals. If running a virtual kernel be sure
283 * to restore the virtual kernel's vmspace before posting the signal.
285 * WARNING! postsig() can exit and not return.
287 if ((sig = CURSIG_LCK_TRACE(lp, &ptok)) != 0) {
293 * block here if we are swapped out, but still process signals
294 * (such as SIGKILL). proc0 (the swapin scheduler) is already
295 * aware of our situation, we do not have to wake it up.
297 if (p->p_flags & P_SWAPPEDOUT) {
298 lwkt_gettoken(&p->p_token);
299 p->p_flags |= P_SWAPWAIT;
301 if (p->p_flags & P_SWAPWAIT)
302 tsleep(p, PCATCH, "SWOUT", 0);
303 p->p_flags &= ~P_SWAPWAIT;
304 lwkt_reltoken(&p->p_token);
309 * In a multi-threaded program it is possible for a thread to change
310 * signal state during a system call which temporarily changes the
311 * signal mask. In this case postsig() might not be run and we
312 * have to restore the mask ourselves.
314 if (lp->lwp_flags & LWP_OLDMASK) {
315 lp->lwp_flags &= ~LWP_OLDMASK;
316 lp->lwp_sigmask = lp->lwp_oldsigmask;
322 * Cleanup from userenter and any passive release that might have occured.
323 * We must reclaim the current-process designation before we can return
324 * to usermode. We also handle both LWKT and USER reschedule requests.
327 userexit(struct lwp *lp)
329 struct thread *td = lp->lwp_thread;
330 /* globaldata_t gd = td->td_gd; */
333 * Handle stop requests at kernel priority. Any requests queued
334 * after this loop will generate another AST.
336 while (STOPLWP(lp->lwp_proc, lp)) {
337 lwkt_gettoken(&lp->lwp_proc->p_token);
339 lwkt_reltoken(&lp->lwp_proc->p_token);
343 * Reduce our priority in preparation for a return to userland. If
344 * our passive release function was still in place, our priority was
345 * never raised and does not need to be reduced.
347 lwkt_passive_recover(td);
349 /* WARNING: we may have migrated cpu's */
350 /* gd = td->td_gd; */
353 * Become the current user scheduled process if we aren't already,
354 * and deal with reschedule requests and other factors.
356 lp->lwp_proc->p_usched->acquire_curproc(lp);
360 * A page fault on a userspace address is classified as SMAP-induced
362 * - SMAP is supported
363 * - kernel mode accessed present data page
364 * - rflags.AC was cleared
367 trap_is_smap(struct trapframe *frame)
369 if ((cpu_stdext_feature & CPUID_STDEXT_SMAP) != 0 &&
370 (frame->tf_err & (PGEX_P | PGEX_U | PGEX_I | PGEX_RSV)) == PGEX_P &&
371 (frame->tf_rflags & PSL_AC) == 0) {
378 #if !defined(KTR_KERNENTRY)
379 #define KTR_KERNENTRY KTR_ALL
381 KTR_INFO_MASTER(kernentry);
382 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0,
383 "TRAP(pid %d, tid %d, trapno %ld, eva %lu)",
384 pid_t pid, lwpid_t tid, register_t trapno, vm_offset_t eva);
385 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "TRAP_RET(pid %d, tid %d)",
386 pid_t pid, lwpid_t tid);
387 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "SYSC(pid %d, tid %d, nr %ld)",
388 pid_t pid, lwpid_t tid, register_t trapno);
389 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "SYSRET(pid %d, tid %d, err %d)",
390 pid_t pid, lwpid_t tid, int err);
391 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "FORKRET(pid %d, tid %d)",
392 pid_t pid, lwpid_t tid);
395 * Exception, fault, and trap interface to the kernel.
396 * This common code is called from assembly language IDT gate entry
397 * routines that prepare a suitable stack frame, and restore this
398 * frame after the exception has been processed.
400 * This function is also called from doreti in an interlock to handle ASTs.
401 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
403 * NOTE! We have to retrieve the fault address prior to potentially
404 * blocking, including blocking on any token.
406 * NOTE! NMI and kernel DBG traps remain on their respective pcpu IST
407 * stacks if taken from a kernel RPL. trap() cannot block in this
408 * situation. DDB entry or a direct report-and-return is ok.
410 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
411 * if an attempt is made to switch from a fast interrupt or IPI.
414 trap(struct trapframe *frame)
416 static struct krate sscpubugrate = { 1 };
417 struct globaldata *gd = mycpu;
418 struct thread *td = gd->gd_curthread;
419 struct lwp *lp = td->td_lwp;
422 int i = 0, ucode = 0, type, code;
424 int crit_count = td->td_critcount;
425 lwkt_tokref_t curstop = td->td_toks_stop;
434 * We need to allow T_DNA faults when the debugger is active since
435 * some dumping paths do large bcopy() which use the floating
436 * point registers for faster copying.
438 if (db_active && frame->tf_trapno != T_DNA) {
439 eva = (frame->tf_trapno == T_PAGEFLT ? frame->tf_addr : 0);
440 ++gd->gd_trap_nesting_level;
441 trap_fatal(frame, eva);
442 --gd->gd_trap_nesting_level;
449 if ((frame->tf_rflags & PSL_I) == 0) {
451 * Buggy application or kernel code has disabled interrupts
452 * and then trapped. Enabling interrupts now is wrong, but
453 * it is better than running with interrupts disabled until
454 * they are accidentally enabled later.
457 type = frame->tf_trapno;
458 if (ISPL(frame->tf_cs) == SEL_UPL) {
459 /* JG curproc can be NULL */
461 "pid %ld (%s): trap %d with interrupts disabled\n",
462 (long)curproc->p_pid, curproc->p_comm, type);
463 } else if ((type == T_STKFLT || type == T_PROTFLT ||
464 type == T_SEGNPFLT) &&
465 frame->tf_rip == (long)doreti_iret) {
467 * iretq fault from kernel mode during return to
470 * This situation is expected, don't complain.
472 } else if (type != T_NMI && type != T_BPTFLT &&
475 * XXX not quite right, since this may be for a
476 * multiple fault in user mode.
478 kprintf("kernel trap %d (%s @ 0x%016jx) with "
479 "interrupts disabled\n",
487 type = frame->tf_trapno;
488 code = frame->tf_err;
490 if (ISPL(frame->tf_cs) == SEL_UPL) {
493 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid,
494 frame->tf_trapno, eva);
498 sticks = (int)td->td_sticks;
499 KASSERT(lp->lwp_md.md_regs == frame,
500 ("Frame mismatch %p %p", lp->lwp_md.md_regs, frame));
503 case T_PRIVINFLT: /* privileged instruction fault */
508 case T_BPTFLT: /* bpt instruction fault */
509 case T_TRCTRAP: /* trace trap */
510 frame->tf_rflags &= ~PSL_T;
512 ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
515 case T_ARITHTRAP: /* arithmetic trap */
520 case T_ASTFLT: /* Allow process switch */
521 mycpu->gd_cnt.v_soft++;
522 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
523 atomic_clear_int(&mycpu->gd_reqflags,
525 addupc_task(p, p->p_prof.pr_addr,
530 case T_PROTFLT: /* general protection fault */
534 case T_STKFLT: /* stack fault */
535 case T_SEGNPFLT: /* segment not present fault */
539 case T_TSSFLT: /* invalid TSS fault */
540 case T_DOUBLEFLT: /* double fault */
546 case T_PAGEFLT: /* page fault */
547 i = trap_pfault(frame, TRUE);
549 if (frame->tf_rip == 0) {
550 /* used for kernel debugging only */
551 while (freeze_on_seg_fault)
552 tsleep(p, 0, "freeze", hz * 20);
555 if (i == -1 || i == 0)
565 case T_DIVIDE: /* integer divide fault */
572 /* machine/parity/power fail/"kitchen sink" faults */
573 if (isa_nmi(code) == 0) {
576 * NMI can be hooked up to a pushbutton
580 kprintf ("NMI ... going to debugger\n");
581 kdb_trap(type, 0, frame);
585 } else if (panic_on_nmi)
586 panic("NMI indicates hardware failure");
588 #endif /* NISA > 0 */
590 case T_OFLOW: /* integer overflow fault */
595 case T_BOUND: /* bounds check fault */
602 * Virtual kernel intercept - pass the DNA exception
603 * to the virtual kernel if it asked to handle it.
604 * This occurs when the virtual kernel is holding
605 * onto the FP context for a different emulated
606 * process then the one currently running.
608 * We must still call npxdna() since we may have
609 * saved FP state that the virtual kernel needs
610 * to hand over to a different emulated process.
612 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
613 (td->td_pcb->pcb_flags & FP_VIRTFP)
620 * The kernel may have switched out the FP unit's
621 * state, causing the user process to take a fault
622 * when it tries to use the FP unit. Restore the
630 ucode = FPE_FPU_NP_TRAP;
633 case T_FPOPFLT: /* FPU operand fetch fault */
638 case T_XMMFLT: /* SIMD floating-point exception */
647 case T_PAGEFLT: /* page fault */
648 trap_pfault(frame, FALSE);
653 * The kernel is apparently using fpu for copying.
654 * XXX this should be fatal unless the kernel has
655 * registered such use.
663 case T_STKFLT: /* stack fault */
664 case T_PROTFLT: /* general protection fault */
665 case T_SEGNPFLT: /* segment not present fault */
667 * Invalid segment selectors and out of bounds
668 * %rip's and %rsp's can be set up in user mode.
669 * This causes a fault in kernel mode when the
670 * kernel tries to return to user mode. We want
671 * to get this fault so that we can fix the
672 * problem here and not have to check all the
673 * selectors and pointers when the user changes
676 if (mycpu->gd_intr_nesting_level == 0) {
678 * NOTE: in 64-bit mode traps push rsp/ss
679 * even if no ring change occurs.
681 if (td->td_pcb->pcb_onfault &&
682 td->td_pcb->pcb_onfault_sp ==
684 frame->tf_rip = (register_t)
685 td->td_pcb->pcb_onfault;
690 * If the iretq in doreti faults during
691 * return to user, it will be special-cased
692 * in IDTVEC(prot) to get here. We want
693 * to 'return' to doreti_iret_fault in
694 * ipl.s in approximately the same state we
695 * were in at the iretq.
697 if (frame->tf_rip == (long)doreti_iret) {
698 frame->tf_rip = (long)doreti_iret_fault;
706 * PSL_NT can be set in user mode and isn't cleared
707 * automatically when the kernel is entered. This
708 * causes a TSS fault when the kernel attempts to
709 * `iret' because the TSS link is uninitialized. We
710 * want to get this fault so that we can fix the
711 * problem here and not every time the kernel is
714 if (frame->tf_rflags & PSL_NT) {
715 frame->tf_rflags &= ~PSL_NT;
717 /* do we need this? */
718 if (frame->tf_rip == (long)doreti_iret)
719 frame->tf_rip = (long)doreti_iret_fault;
725 case T_TRCTRAP: /* trace trap */
727 * Detect historical CPU artifact on syscall or int $3
728 * entry (if not shortcutted in exception.s via
729 * DIRECT_DISALLOW_SS_CPUBUG).
732 if (frame->tf_rip == (register_t)IDTVEC(fast_syscall)) {
733 krateprintf(&sscpubugrate,
734 "Caught #DB at syscall cpu artifact\n");
737 if (frame->tf_rip == (register_t)IDTVEC(bpt)) {
738 krateprintf(&sscpubugrate,
739 "Caught #DB at int $N cpu artifact\n");
744 * Ignore debug register trace traps due to
745 * accesses in the user's address space, which
746 * can happen under several conditions such as
747 * if a user sets a watchpoint on a buffer and
748 * then passes that buffer to a system call.
749 * We still want to get TRCTRAPS for addresses
750 * in kernel space because that is useful when
751 * debugging the kernel.
753 if (user_dbreg_trap()) {
755 * Reset breakpoint bits because the
758 load_dr6(rdr6() & ~0xf);
762 * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
766 * If DDB is enabled, let it handle the debugger trap.
767 * Otherwise, debugger traps "can't happen".
771 if (kdb_trap(type, 0, frame))
778 /* machine/parity/power fail/"kitchen sink" faults */
779 if (isa_nmi(code) == 0) {
782 * NMI can be hooked up to a pushbutton
786 kprintf ("NMI ... going to debugger\n");
787 kdb_trap(type, 0, frame);
791 } else if (panic_on_nmi == 0)
794 #endif /* NISA > 0 */
796 trap_fatal(frame, 0);
801 * Fault from user mode, virtual kernel interecept.
803 * If the fault is directly related to a VM context managed by a
804 * virtual kernel then let the virtual kernel handle it.
806 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
807 vkernel_trap(lp, frame);
811 /* Translate fault for emulators (e.g. Linux) */
812 if (*p->p_sysent->sv_transtrap)
813 i = (*p->p_sysent->sv_transtrap)(i, type);
816 trapsignal(lp, i, ucode);
819 if (type <= MAX_TRAP_MSG) {
820 uprintf("fatal process exception: %s",
822 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
823 uprintf(", fault VA = 0x%lx", frame->tf_addr);
829 userret(lp, frame, sticks);
832 if (p != NULL && lp != NULL)
833 KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid);
835 KASSERT(crit_count == td->td_critcount,
836 ("trap: critical section count mismatch! %d/%d",
837 crit_count, td->td_pri));
838 KASSERT(curstop == td->td_toks_stop,
839 ("trap: extra tokens held after trap! %ld/%ld (%s)",
840 curstop - &td->td_toks_base,
841 td->td_toks_stop - &td->td_toks_base,
842 td->td_toks_stop[-1].tr_tok->t_desc));
847 trap_handle_userenter(struct thread *td)
849 userenter(td, td->td_proc);
853 trap_handle_userexit(struct trapframe *frame, int sticks)
855 struct lwp *lp = curthread->td_lwp;
858 userret(lp, frame, sticks);
864 trap_pfault(struct trapframe *frame, int usermode)
867 struct vmspace *vm = NULL;
872 thread_t td = curthread;
873 struct lwp *lp = td->td_lwp;
876 va = trunc_page(frame->tf_addr);
877 if (va >= VM_MIN_KERNEL_ADDRESS) {
879 * Don't allow user-mode faults in kernel address space.
890 * This is a fault on non-kernel virtual memory.
891 * vm is initialized above to NULL. If curproc is NULL
892 * or curproc->p_vmspace is NULL the fault is fatal.
895 vm = lp->lwp_vmspace;
906 * Debugging, catch kernel faults on the user address
907 * space when not inside on onfault (e.g. copyin/
910 if (td->td_pcb == NULL ||
911 td->td_pcb->pcb_onfault == NULL) {
912 if (freeze_on_seg_fault) {
913 kprintf("trap_pfault: user address "
914 "fault from kernel mode "
916 (long)frame->tf_addr);
917 while (freeze_on_seg_fault) {
918 tsleep(&freeze_on_seg_fault,
925 if (td->td_gd->gd_intr_nesting_level ||
926 trap_is_smap(frame) ||
927 td->td_pcb == NULL ||
928 td->td_pcb->pcb_onfault == NULL) {
929 kprintf("Fatal user address access "
930 "from kernel mode from %s at %016jx\n",
931 td->td_comm, frame->tf_rip);
932 trap_fatal(frame, frame->tf_addr);
941 * PGEX_I is defined only if the execute disable bit capability is
942 * supported and enabled.
944 if (frame->tf_err & PGEX_W)
945 ftype = VM_PROT_WRITE;
946 else if (frame->tf_err & PGEX_I)
947 ftype = VM_PROT_EXECUTE;
949 ftype = VM_PROT_READ;
951 lwkt_tokref_t stop = td->td_toks_stop;
953 if (map != &kernel_map) {
955 * Keep swapout from messing with us during this
965 fault_flags |= VM_FAULT_BURST | VM_FAULT_USERMODE;
966 if (ftype & VM_PROT_WRITE)
967 fault_flags |= VM_FAULT_DIRTY;
969 fault_flags |= VM_FAULT_NORMAL;
970 rv = vm_fault(map, va, ftype, fault_flags);
971 if (td->td_toks_stop != stop) {
972 stop = td->td_toks_stop - 1;
973 kprintf("A-HELD TOKENS DURING PFAULT td=%p(%s) map=%p va=%p ftype=%d fault_flags=%d\n", td, td->td_comm, map, (void *)va, ftype, fault_flags);
974 panic("held tokens");
980 * Don't have to worry about process locking or stacks in the
983 fault_flags = VM_FAULT_NORMAL;
984 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
985 if (td->td_toks_stop != stop) {
986 stop = td->td_toks_stop - 1;
987 kprintf("B-HELD TOKENS DURING PFAULT td=%p(%s) map=%p va=%p ftype=%d fault_flags=%d\n", td, td->td_comm, map, (void *)va, ftype, VM_FAULT_NORMAL);
988 panic("held tokens");
991 if (rv == KERN_SUCCESS)
996 * NOTE: in 64-bit mode traps push rsp/ss
997 * even if no ring change occurs.
999 if (td->td_pcb->pcb_onfault &&
1000 td->td_pcb->pcb_onfault_sp == frame->tf_rsp &&
1001 td->td_gd->gd_intr_nesting_level == 0) {
1002 frame->tf_rip = (register_t)td->td_pcb->pcb_onfault;
1005 trap_fatal(frame, frame->tf_addr);
1010 * NOTE: on x86_64 we have a tf_addr field in the trapframe, no
1011 * kludge is needed to pass the fault address to signal handlers.
1015 if (td->td_lwp->lwp_vkernel == NULL) {
1016 while (freeze_on_seg_fault) {
1017 tsleep(p, 0, "freeze", hz * 20);
1019 if (ddb_on_seg_fault)
1020 Debugger("ddb_on_seg_fault");
1024 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
1028 trap_fatal(struct trapframe *frame, vm_offset_t eva)
1033 struct soft_segment_descriptor softseg;
1036 code = frame->tf_err;
1037 type = frame->tf_trapno;
1038 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)], &softseg);
1040 if (type <= MAX_TRAP_MSG)
1041 msg = trap_msg[type];
1044 kprintf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
1045 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
1046 /* three separate prints in case of a trap on an unmapped page */
1047 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
1049 kprintf("lapic id = %u\n", LAPIC_READID);
1050 if (type == T_PAGEFLT) {
1051 kprintf("fault virtual address = 0x%lx\n", eva);
1052 kprintf("fault code = %s %s %s, %s\n",
1053 code & PGEX_U ? "user" : "supervisor",
1054 code & PGEX_W ? "write" : "read",
1055 code & PGEX_I ? "instruction" : "data",
1056 code & PGEX_P ? "protection violation" : "page not present");
1058 kprintf("instruction pointer = 0x%lx:0x%lx\n",
1059 frame->tf_cs & 0xffff, frame->tf_rip);
1060 if (ISPL(frame->tf_cs) == SEL_UPL) {
1061 ss = frame->tf_ss & 0xffff;
1062 rsp = frame->tf_rsp;
1065 * NOTE: in 64-bit mode traps push rsp/ss even if no ring
1068 ss = GSEL(GDATA_SEL, SEL_KPL);
1069 rsp = frame->tf_rsp;
1071 kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp);
1072 kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp);
1073 kprintf("code segment = base 0x%lx, limit 0x%lx, type 0x%x\n",
1074 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
1075 kprintf(" = DPL %d, pres %d, long %d, def32 %d, gran %d\n",
1076 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32,
1078 kprintf("processor eflags = ");
1079 if (frame->tf_rflags & PSL_T)
1080 kprintf("trace trap, ");
1081 if (frame->tf_rflags & PSL_I)
1082 kprintf("interrupt enabled, ");
1083 if (frame->tf_rflags & PSL_NT)
1084 kprintf("nested task, ");
1085 if (frame->tf_rflags & PSL_RF)
1086 kprintf("resume, ");
1087 kprintf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12);
1088 kprintf("current process = ");
1091 (u_long)curproc->p_pid);
1095 kprintf("current thread = pri %d ", curthread->td_pri);
1096 if (curthread->td_critcount)
1101 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1104 kprintf("trap number = %d\n", type);
1105 if (type <= MAX_TRAP_MSG)
1106 panic("%s", trap_msg[type]);
1108 panic("unknown/reserved trap");
1112 * Double fault handler. Called when a fault occurs while writing
1113 * a frame for a trap/exception onto the stack. This usually occurs
1114 * when the stack overflows (such is the case with infinite recursion,
1119 in_kstack_guard(register_t rptr)
1121 thread_t td = curthread;
1123 if ((char *)rptr >= td->td_kstack &&
1124 (char *)rptr < td->td_kstack + PAGE_SIZE) {
1131 dblfault_handler(struct trapframe *frame)
1133 thread_t td = curthread;
1135 if (in_kstack_guard(frame->tf_rsp) || in_kstack_guard(frame->tf_rbp)) {
1136 kprintf("DOUBLE FAULT - KERNEL STACK GUARD HIT!\n");
1137 if (in_kstack_guard(frame->tf_rsp))
1138 frame->tf_rsp = (register_t)(td->td_kstack + PAGE_SIZE);
1139 if (in_kstack_guard(frame->tf_rbp))
1140 frame->tf_rbp = (register_t)(td->td_kstack + PAGE_SIZE);
1142 kprintf("DOUBLE FAULT\n");
1144 kprintf("\nFatal double fault\n");
1145 kprintf("rip = 0x%lx\n", frame->tf_rip);
1146 kprintf("rsp = 0x%lx\n", frame->tf_rsp);
1147 kprintf("rbp = 0x%lx\n", frame->tf_rbp);
1148 /* three separate prints in case of a trap on an unmapped page */
1149 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
1151 kprintf("lapic id = %u\n", LAPIC_READID);
1152 panic("double fault");
1156 * syscall2 - MP aware system call request C handler
1158 * A system call is essentially treated as a trap except that the
1159 * MP lock is not held on entry or return. We are responsible for
1160 * obtaining the MP lock if necessary and for handling ASTs
1161 * (e.g. a task switch) prior to return.
1166 syscall2(struct trapframe *frame)
1168 struct thread *td = curthread;
1169 struct proc *p = td->td_proc;
1170 struct lwp *lp = td->td_lwp;
1171 struct sysent *callp;
1172 register_t orig_tf_rflags;
1177 int crit_count = td->td_critcount;
1181 int regcnt, optimized_regcnt;
1182 union sysunion args;
1183 register_t *argsdst;
1185 mycpu->gd_cnt.v_syscall++;
1188 if (ISPL(frame->tf_cs) != SEL_UPL) {
1194 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid,
1197 userenter(td, p); /* lazy raise our priority */
1200 optimized_regcnt = 6;
1205 sticks = (int)td->td_sticks;
1206 orig_tf_rflags = frame->tf_rflags;
1209 * Virtual kernel intercept - if a VM context managed by a virtual
1210 * kernel issues a system call the virtual kernel handles it, not us.
1211 * Restore the virtual kernel context and return from its system
1212 * call. The current frame is copied out to the virtual kernel.
1214 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1215 vkernel_trap(lp, frame);
1216 error = EJUSTRETURN;
1223 * Get the system call parameters and account for time
1225 KASSERT(lp->lwp_md.md_regs == frame,
1226 ("Frame mismatch %p %p", lp->lwp_md.md_regs, frame));
1227 code = (u_int)frame->tf_rax;
1229 if (code == SYS_syscall || code == SYS___syscall) {
1230 code = frame->tf_rdi;
1232 argp = &frame->tf_rdi + 1;
1234 argp = &frame->tf_rdi;
1237 if (code >= p->p_sysent->sv_size)
1238 callp = &p->p_sysent->sv_table[0];
1240 callp = &p->p_sysent->sv_table[code];
1242 narg = callp->sy_narg & SYF_ARGMASK;
1245 * On x86_64 we get up to six arguments in registers. The rest are
1246 * on the stack. The first six members of 'struct trapframe' happen
1247 * to be the registers used to pass arguments, in exactly the right
1250 argsdst = (register_t *)(&args.nosys.sysmsg + 1);
1253 * Its easier to copy up to the highest number of syscall arguments
1254 * passed in registers, which is 6, than to conditionalize it.
1256 bcopy(argp, argsdst, sizeof(register_t) * optimized_regcnt);
1259 * Any arguments beyond available argument-passing registers must
1260 * be copyin()'d from the user stack.
1262 if (narg > regcnt) {
1265 params = (caddr_t)frame->tf_rsp + sizeof(register_t);
1266 error = copyin(params, &argsdst[regcnt],
1267 (narg - regcnt) * sizeof(register_t));
1270 if (KTRPOINT(td, KTR_SYSCALL)) {
1271 ktrsyscall(lp, code, narg,
1272 (void *)(&args.nosys.sysmsg + 1));
1280 if (KTRPOINT(td, KTR_SYSCALL)) {
1281 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1286 * Default return value is 0 (will be copied to %rax). Double-value
1287 * returns use %rax and %rdx. %rdx is left unchanged for system
1288 * calls which return only one result.
1290 args.sysmsg_fds[0] = 0;
1291 args.sysmsg_fds[1] = frame->tf_rdx;
1294 * The syscall might manipulate the trap frame. If it does it
1295 * will probably return EJUSTRETURN.
1297 args.sysmsg_frame = frame;
1299 STOPEVENT(p, S_SCE, narg); /* MP aware */
1302 * NOTE: All system calls run MPSAFE now. The system call itself
1303 * is responsible for getting the MP lock.
1305 #ifdef SYSCALL_DEBUG
1306 tsc_uclock_t tscval = rdtsc();
1308 error = (*callp->sy_call)(&args);
1309 #ifdef SYSCALL_DEBUG
1310 tscval = rdtsc() - tscval;
1311 tscval = tscval * 1000000 / tsc_frequency;
1312 if (SysCallsWorstCase[code] < tscval)
1313 SysCallsWorstCase[code] = tscval;
1318 * MP SAFE (we may or may not have the MP lock at this point)
1320 //kprintf("SYSMSG %d ", error);
1324 * Reinitialize proc pointer `p' as it may be different
1325 * if this is a child returning from fork syscall.
1328 lp = curthread->td_lwp;
1329 frame->tf_rax = args.sysmsg_fds[0];
1330 frame->tf_rdx = args.sysmsg_fds[1];
1331 frame->tf_rflags &= ~PSL_C;
1335 * Reconstruct pc, we know that 'syscall' is 2 bytes.
1336 * We have to do a full context restore so that %r10
1337 * (which was holding the value of %rcx) is restored for
1338 * the next iteration.
1340 if (frame->tf_err != 0 && frame->tf_err != 2)
1341 kprintf("lp %s:%d frame->tf_err is weird %ld\n",
1342 td->td_comm, lp->lwp_proc->p_pid, frame->tf_err);
1343 frame->tf_rip -= frame->tf_err;
1344 frame->tf_r10 = frame->tf_rcx;
1349 panic("Unexpected EASYNC return value (for now)");
1352 if (p->p_sysent->sv_errsize) {
1353 if (error >= p->p_sysent->sv_errsize)
1354 error = -1; /* XXX */
1356 error = p->p_sysent->sv_errtbl[error];
1358 frame->tf_rax = error;
1359 frame->tf_rflags |= PSL_C;
1364 * Traced syscall. trapsignal() should now be MP aware
1366 if (orig_tf_rflags & PSL_T) {
1367 frame->tf_rflags &= ~PSL_T;
1368 trapsignal(lp, SIGTRAP, TRAP_TRACE);
1372 * Handle reschedule and other end-of-syscall issues
1374 userret(lp, frame, sticks);
1377 if (KTRPOINT(td, KTR_SYSRET)) {
1378 ktrsysret(lp, code, error, args.sysmsg_result);
1383 * This works because errno is findable through the
1384 * register set. If we ever support an emulation where this
1385 * is not the case, this code will need to be revisited.
1387 STOPEVENT(p, S_SCX, code);
1390 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error);
1392 KASSERT(crit_count == td->td_critcount,
1393 ("syscall: critical section count mismatch! %d/%d",
1394 crit_count, td->td_pri));
1395 KASSERT(&td->td_toks_base == td->td_toks_stop,
1396 ("syscall: %ld extra tokens held after trap! syscall %p",
1397 td->td_toks_stop - &td->td_toks_base,
1403 fork_return(struct lwp *lp, struct trapframe *frame)
1405 frame->tf_rax = 0; /* Child returns zero */
1406 frame->tf_rflags &= ~PSL_C; /* success */
1409 generic_lwp_return(lp, frame);
1410 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1414 * Simplified back end of syscall(), used when returning from fork()
1415 * directly into user mode.
1417 * This code will return back into the fork trampoline code which then
1421 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1423 struct proc *p = lp->lwp_proc;
1426 * Check for exit-race. If one lwp exits the process concurrent with
1427 * another lwp creating a new thread, the two operations may cross
1428 * each other resulting in the newly-created lwp not receiving a
1431 if (p->p_flags & P_WEXIT) {
1432 lwpsignal(p, lp, SIGKILL);
1436 * Newly forked processes are given a kernel priority. We have to
1437 * adjust the priority to a normal user priority and fake entry
1438 * into the kernel (call userenter()) to install a passive release
1439 * function just in case userret() decides to stop the process. This
1440 * can occur when ^Z races a fork. If we do not install the passive
1441 * release function the current process designation will not be
1442 * released when the thread goes to sleep.
1444 lwkt_setpri_self(TDPRI_USER_NORM);
1445 userenter(lp->lwp_thread, p);
1446 userret(lp, frame, 0);
1448 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1449 ktrsysret(lp, SYS_fork, 0, 0);
1451 lp->lwp_flags |= LWP_PASSIVE_ACQ;
1453 lp->lwp_flags &= ~LWP_PASSIVE_ACQ;
1457 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1458 * fault (which is then passed back to the virtual kernel) if an attempt is
1459 * made to use the FP unit.
1461 * XXX this is a fairly big hack.
1464 set_vkernel_fp(struct trapframe *frame)
1466 struct thread *td = curthread;
1468 if (frame->tf_xflags & PGEX_FPFAULT) {
1469 td->td_pcb->pcb_flags |= FP_VIRTFP;
1470 if (mdcpu->gd_npxthread == td)
1473 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1478 * Called from vkernel_trap() to fixup the vkernel's syscall
1479 * frame for vmspace_ctl() return.
1482 cpu_vkernel_trap(struct trapframe *frame, int error)
1484 frame->tf_rax = error;
1486 frame->tf_rflags |= PSL_C;
1488 frame->tf_rflags &= ~PSL_C;