2 * Copyright (C) 1994, David Greenman
3 * Copyright (c) 1990, 1993
4 * The Regents of the University of California. All rights reserved.
6 * This code is derived from software contributed to Berkeley by
7 * the University of Utah, and William Jolitz.
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. All advertising materials mentioning features or use of this software
18 * must display the following acknowledgement:
19 * This product includes software developed by the University of
20 * California, Berkeley and its contributors.
21 * 4. Neither the name of the University nor the names of its contributors
22 * may be used to endorse or promote products derived from this software
23 * without specific prior written permission.
25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 * from: @(#)trap.c 7.4 (Berkeley) 5/13/91
38 * $FreeBSD: src/sys/i386/i386/trap.c,v 1.147.2.11 2003/02/27 19:09:59 luoqi Exp $
42 * x86_64 Trap and System call handling
48 #include "opt_ktrace.h"
50 #include <sys/param.h>
51 #include <sys/systm.h>
53 #include <sys/pioctl.h>
54 #include <sys/kernel.h>
55 #include <sys/resourcevar.h>
56 #include <sys/signalvar.h>
57 #include <sys/signal2.h>
58 #include <sys/syscall.h>
59 #include <sys/sysctl.h>
60 #include <sys/sysent.h>
62 #include <sys/vmmeter.h>
63 #include <sys/malloc.h>
65 #include <sys/ktrace.h>
68 #include <sys/vkernel.h>
69 #include <sys/sysproto.h>
70 #include <sys/sysunion.h>
71 #include <sys/vmspace.h>
74 #include <vm/vm_param.h>
77 #include <vm/vm_kern.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_page.h>
80 #include <vm/vm_extern.h>
82 #include <machine/cpu.h>
83 #include <machine/md_var.h>
84 #include <machine/pcb.h>
85 #include <machine/smp.h>
86 #include <machine/tss.h>
87 #include <machine/globaldata.h>
91 #include <sys/msgport2.h>
92 #include <sys/thread2.h>
93 #include <sys/mplock2.h>
95 #define MAKEMPSAFE(have_mplock) \
96 if (have_mplock == 0) { \
101 int (*pmath_emulate) (struct trapframe *);
103 extern int trapwrite (unsigned addr);
105 static int trap_pfault (struct trapframe *, int, vm_offset_t);
106 static void trap_fatal (struct trapframe *, int, vm_offset_t);
107 void dblfault_handler (void);
108 extern int vmm_enabled;
111 extern inthand_t IDTVEC(syscall);
114 #define MAX_TRAP_MSG 30
115 static char *trap_msg[] = {
117 "privileged instruction fault", /* 1 T_PRIVINFLT */
119 "breakpoint instruction fault", /* 3 T_BPTFLT */
122 "arithmetic trap", /* 6 T_ARITHTRAP */
123 "system forced exception", /* 7 T_ASTFLT */
125 "general protection fault", /* 9 T_PROTFLT */
126 "trace trap", /* 10 T_TRCTRAP */
128 "page fault", /* 12 T_PAGEFLT */
130 "alignment fault", /* 14 T_ALIGNFLT */
134 "integer divide fault", /* 18 T_DIVIDE */
135 "non-maskable interrupt trap", /* 19 T_NMI */
136 "overflow trap", /* 20 T_OFLOW */
137 "FPU bounds check fault", /* 21 T_BOUND */
138 "FPU device not available", /* 22 T_DNA */
139 "double fault", /* 23 T_DOUBLEFLT */
140 "FPU operand fetch fault", /* 24 T_FPOPFLT */
141 "invalid TSS fault", /* 25 T_TSSFLT */
142 "segment not present fault", /* 26 T_SEGNPFLT */
143 "stack fault", /* 27 T_STKFLT */
144 "machine check trap", /* 28 T_MCHK */
145 "SIMD floating-point exception", /* 29 T_XMMFLT */
146 "reserved (unknown) fault", /* 30 T_RESERVED */
150 static int ddb_on_nmi = 1;
151 SYSCTL_INT(_machdep, OID_AUTO, ddb_on_nmi, CTLFLAG_RW,
152 &ddb_on_nmi, 0, "Go to DDB on NMI");
154 static int panic_on_nmi = 1;
155 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
156 &panic_on_nmi, 0, "Panic on NMI");
157 static int fast_release;
158 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
159 &fast_release, 0, "Passive Release was optimal");
160 static int slow_release;
161 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
162 &slow_release, 0, "Passive Release was nonoptimal");
165 * Passively intercepts the thread switch function to increase
166 * the thread priority from a user priority to a kernel priority, reducing
167 * syscall and trap overhead for the case where no switch occurs.
169 * Synchronizes td_ucred with p_ucred. This is used by system calls,
170 * signal handling, faults, AST traps, and anything else that enters the
171 * kernel from userland and provides the kernel with a stable read-only
172 * copy of the process ucred.
175 userenter(struct thread *curtd, struct proc *curp)
180 curtd->td_release = lwkt_passive_release;
182 if (curtd->td_ucred != curp->p_ucred) {
183 ncred = crhold(curp->p_ucred);
184 ocred = curtd->td_ucred;
185 curtd->td_ucred = ncred;
192 * Handle signals, profiling, and other AST's and/or tasks that
193 * must be completed before we can return to or try to return to userland.
195 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
196 * arithmatic on the delta calculation so the absolute tick values are
197 * truncated to an integer.
200 userret(struct lwp *lp, struct trapframe *frame, int sticks)
202 struct proc *p = lp->lwp_proc;
206 * Charge system time if profiling. Note: times are in microseconds.
207 * This may do a copyout and block, so do it first even though it
208 * means some system time will be charged as user time.
210 if (p->p_flags & P_PROFIL) {
211 addupc_task(p, frame->tf_rip,
212 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
217 * Specific on-return-to-usermode checks (LWP_MP_WEXIT,
218 * LWP_MP_VNLRU, etc).
220 if (lp->lwp_mpflags & LWP_MP_URETMASK)
224 * Block here if we are in a stopped state.
226 if (p->p_stat == SSTOP) {
227 lwkt_gettoken(&p->p_token);
229 lwkt_reltoken(&p->p_token);
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_flags & (P_SIGVTALRM | P_SIGPROF)) {
238 lwkt_gettoken(&p->p_token);
239 if (p->p_flags & P_SIGVTALRM) {
240 p->p_flags &= ~P_SIGVTALRM;
241 ksignal(p, SIGVTALRM);
243 if (p->p_flags & P_SIGPROF) {
244 p->p_flags &= ~P_SIGPROF;
247 lwkt_reltoken(&p->p_token);
252 * Post any pending signals
254 * WARNING! postsig() can exit and not return.
256 if ((sig = CURSIG_TRACE(lp)) != 0) {
257 lwkt_gettoken(&p->p_token);
259 lwkt_reltoken(&p->p_token);
264 * block here if we are swapped out, but still process signals
265 * (such as SIGKILL). proc0 (the swapin scheduler) is already
266 * aware of our situation, we do not have to wake it up.
268 if (p->p_flags & P_SWAPPEDOUT) {
269 lwkt_gettoken(&p->p_token);
271 p->p_flags |= P_SWAPWAIT;
273 if (p->p_flags & P_SWAPWAIT)
274 tsleep(p, PCATCH, "SWOUT", 0);
275 p->p_flags &= ~P_SWAPWAIT;
277 lwkt_reltoken(&p->p_token);
282 * In a multi-threaded program it is possible for a thread to change
283 * signal state during a system call which temporarily changes the
284 * signal mask. In this case postsig() might not be run and we
285 * have to restore the mask ourselves.
287 if (lp->lwp_flags & LWP_OLDMASK) {
288 lp->lwp_flags &= ~LWP_OLDMASK;
289 lp->lwp_sigmask = lp->lwp_oldsigmask;
295 * Cleanup from userenter and any passive release that might have occured.
296 * We must reclaim the current-process designation before we can return
297 * to usermode. We also handle both LWKT and USER reschedule requests.
300 userexit(struct lwp *lp)
302 struct thread *td = lp->lwp_thread;
303 /* globaldata_t gd = td->td_gd; */
306 * Handle stop requests at kernel priority. Any requests queued
307 * after this loop will generate another AST.
309 while (lp->lwp_proc->p_stat == SSTOP) {
310 lwkt_gettoken(&lp->lwp_proc->p_token);
312 lwkt_reltoken(&lp->lwp_proc->p_token);
316 * Reduce our priority in preparation for a return to userland. If
317 * our passive release function was still in place, our priority was
318 * never raised and does not need to be reduced.
320 lwkt_passive_recover(td);
323 * Become the current user scheduled process if we aren't already,
324 * and deal with reschedule requests and other factors.
326 lp->lwp_proc->p_usched->acquire_curproc(lp);
327 /* WARNING: we may have migrated cpu's */
328 /* gd = td->td_gd; */
331 #if !defined(KTR_KERNENTRY)
332 #define KTR_KERNENTRY KTR_ALL
334 KTR_INFO_MASTER(kernentry);
335 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0,
336 "TRAP(pid %hd, tid %hd, trapno %ld, eva %lu)",
337 pid_t pid, lwpid_t tid, register_t trapno, vm_offset_t eva);
338 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "TRAP_RET(pid %hd, tid %hd)",
339 pid_t pid, lwpid_t tid);
340 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "SYSC(pid %hd, tid %hd, nr %ld)",
341 pid_t pid, lwpid_t tid, register_t trapno);
342 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "SYSRET(pid %hd, tid %hd, err %d)",
343 pid_t pid, lwpid_t tid, int err);
344 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "FORKRET(pid %hd, tid %hd)",
345 pid_t pid, lwpid_t tid);
348 * Exception, fault, and trap interface to the kernel.
349 * This common code is called from assembly language IDT gate entry
350 * routines that prepare a suitable stack frame, and restore this
351 * frame after the exception has been processed.
353 * This function is also called from doreti in an interlock to handle ASTs.
354 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
356 * NOTE! We have to retrieve the fault address prior to obtaining the
357 * MP lock because get_mplock() may switch out. YYY cr2 really ought
358 * to be retrieved by the assembly code, not here.
360 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
361 * if an attempt is made to switch from a fast interrupt or IPI. This is
362 * necessary to properly take fatal kernel traps on SMP machines if
363 * get_mplock() has to block.
367 user_trap(struct trapframe *frame)
369 struct globaldata *gd = mycpu;
370 struct thread *td = gd->gd_curthread;
371 struct lwp *lp = td->td_lwp;
374 int i = 0, ucode = 0, type, code;
377 int crit_count = td->td_critcount;
378 lwkt_tokref_t curstop = td->td_toks_stop;
384 if (frame->tf_trapno == T_PAGEFLT)
385 eva = frame->tf_addr;
389 kprintf("USER_TRAP AT %08lx xflags %ld trapno %ld eva %08lx\n",
390 frame->tf_rip, frame->tf_xflags, frame->tf_trapno, eva);
394 * Everything coming from user mode runs through user_trap,
395 * including system calls.
397 if (frame->tf_trapno == T_FAST_SYSCALL) {
402 KTR_LOG(kernentry_trap, lp->lwp_proc->p_pid, lp->lwp_tid,
403 frame->tf_trapno, eva);
407 eva = (frame->tf_trapno == T_PAGEFLT ? rcr2() : 0);
408 ++gd->gd_trap_nesting_level;
409 MAKEMPSAFE(have_mplock);
410 trap_fatal(frame, TRUE, eva);
411 --gd->gd_trap_nesting_level;
416 type = frame->tf_trapno;
417 code = frame->tf_err;
421 sticks = (int)td->td_sticks;
422 lp->lwp_md.md_regs = frame;
425 case T_PRIVINFLT: /* privileged instruction fault */
430 case T_BPTFLT: /* bpt instruction fault */
431 case T_TRCTRAP: /* trace trap */
432 frame->tf_rflags &= ~PSL_T;
434 ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
437 case T_ARITHTRAP: /* arithmetic trap */
442 case T_ASTFLT: /* Allow process switch */
443 mycpu->gd_cnt.v_soft++;
444 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
445 atomic_clear_int(&mycpu->gd_reqflags, RQF_AST_OWEUPC);
446 addupc_task(p, p->p_prof.pr_addr, p->p_prof.pr_ticks);
451 * The following two traps can happen in
452 * vm86 mode, and, if so, we want to handle
455 case T_PROTFLT: /* general protection fault */
456 case T_STKFLT: /* stack fault */
458 if (frame->tf_eflags & PSL_VM) {
459 i = vm86_emulate((struct vm86frame *)frame);
467 case T_SEGNPFLT: /* segment not present fault */
468 case T_TSSFLT: /* invalid TSS fault */
469 case T_DOUBLEFLT: /* double fault */
472 ucode = code + BUS_SEGM_FAULT ;
475 case T_PAGEFLT: /* page fault */
476 i = trap_pfault(frame, TRUE, eva);
477 if (i == -1 || i == 0)
489 case T_DIVIDE: /* integer divide fault */
496 MAKEMPSAFE(have_mplock);
497 /* machine/parity/power fail/"kitchen sink" faults */
498 if (isa_nmi(code) == 0) {
501 * NMI can be hooked up to a pushbutton
505 kprintf ("NMI ... going to debugger\n");
506 kdb_trap(type, 0, frame);
510 } else if (panic_on_nmi)
511 panic("NMI indicates hardware failure");
513 #endif /* NISA > 0 */
515 case T_OFLOW: /* integer overflow fault */
520 case T_BOUND: /* bounds check fault */
527 * Virtual kernel intercept - pass the DNA exception
528 * to the (emulated) virtual kernel if it asked to handle
529 * it. This occurs when the virtual kernel is holding
530 * onto the FP context for a different emulated
531 * process then the one currently running.
533 * We must still call npxdna() since we may have
534 * saved FP state that the (emulated) virtual kernel
535 * needs to hand over to a different emulated process.
537 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
538 (td->td_pcb->pcb_flags & FP_VIRTFP)
545 * The kernel may have switched out the FP unit's
546 * state, causing the user process to take a fault
547 * when it tries to use the FP unit. Restore the
552 if (!pmath_emulate) {
554 ucode = FPE_FPU_NP_TRAP;
557 i = (*pmath_emulate)(frame);
559 if (!(frame->tf_rflags & PSL_T))
561 frame->tf_rflags &= ~PSL_T;
564 /* else ucode = emulator_only_knows() XXX */
567 case T_FPOPFLT: /* FPU operand fetch fault */
572 case T_XMMFLT: /* SIMD floating-point exception */
579 * Virtual kernel intercept - if the fault is directly related to a
580 * VM context managed by a virtual kernel then let the virtual kernel
583 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
584 vkernel_trap(lp, frame);
589 * Translate fault for emulators (e.g. Linux)
591 if (*p->p_sysent->sv_transtrap)
592 i = (*p->p_sysent->sv_transtrap)(i, type);
594 MAKEMPSAFE(have_mplock);
595 trapsignal(lp, i, ucode);
598 if (type <= MAX_TRAP_MSG) {
599 uprintf("fatal process exception: %s",
601 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
602 uprintf(", fault VA = 0x%lx", (u_long)eva);
608 userret(lp, frame, sticks);
613 KTR_LOG(kernentry_trap_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
615 KASSERT(crit_count == td->td_critcount,
616 ("trap: critical section count mismatch! %d/%d",
617 crit_count, td->td_pri));
618 KASSERT(curstop == td->td_toks_stop,
619 ("trap: extra tokens held after trap! %ld/%ld",
620 curstop - &td->td_toks_base,
621 td->td_toks_stop - &td->td_toks_base));
626 kern_trap(struct trapframe *frame)
628 struct globaldata *gd = mycpu;
629 struct thread *td = gd->gd_curthread;
632 int i = 0, ucode = 0, type, code;
635 int crit_count = td->td_critcount;
636 lwkt_tokref_t curstop = td->td_toks_stop;
643 if (frame->tf_trapno == T_PAGEFLT)
644 eva = frame->tf_addr;
650 ++gd->gd_trap_nesting_level;
651 MAKEMPSAFE(have_mplock);
652 trap_fatal(frame, FALSE, eva);
653 --gd->gd_trap_nesting_level;
658 type = frame->tf_trapno;
659 code = frame->tf_err;
667 case T_PAGEFLT: /* page fault */
668 trap_pfault(frame, FALSE, eva);
673 * The kernel may be using npx for copying or other
676 panic("kernel NPX should not happen");
681 case T_PROTFLT: /* general protection fault */
682 case T_SEGNPFLT: /* segment not present fault */
684 * Invalid segment selectors and out of bounds
685 * %eip's and %esp's can be set up in user mode.
686 * This causes a fault in kernel mode when the
687 * kernel tries to return to user mode. We want
688 * to get this fault so that we can fix the
689 * problem here and not have to check all the
690 * selectors and pointers when the user changes
693 if (mycpu->gd_intr_nesting_level == 0) {
694 if (td->td_pcb->pcb_onfault) {
696 (register_t)td->td_pcb->pcb_onfault;
704 * PSL_NT can be set in user mode and isn't cleared
705 * automatically when the kernel is entered. This
706 * causes a TSS fault when the kernel attempts to
707 * `iret' because the TSS link is uninitialized. We
708 * want to get this fault so that we can fix the
709 * problem here and not every time the kernel is
712 if (frame->tf_rflags & PSL_NT) {
713 frame->tf_rflags &= ~PSL_NT;
718 case T_TRCTRAP: /* trace trap */
720 if (frame->tf_eip == (int)IDTVEC(syscall)) {
722 * We've just entered system mode via the
723 * syscall lcall. Continue single stepping
724 * silently until the syscall handler has
729 if (frame->tf_eip == (int)IDTVEC(syscall) + 1) {
731 * The syscall handler has now saved the
732 * flags. Stop single stepping it.
734 frame->tf_eflags &= ~PSL_T;
740 * Ignore debug register trace traps due to
741 * accesses in the user's address space, which
742 * can happen under several conditions such as
743 * if a user sets a watchpoint on a buffer and
744 * then passes that buffer to a system call.
745 * We still want to get TRCTRAPS for addresses
746 * in kernel space because that is useful when
747 * debugging the kernel.
749 if (user_dbreg_trap()) {
751 * Reset breakpoint bits because the
754 load_dr6(rdr6() & 0xfffffff0);
759 * Fall through (TRCTRAP kernel mode, kernel address)
763 * If DDB is enabled, let it handle the debugger trap.
764 * Otherwise, debugger traps "can't happen".
767 MAKEMPSAFE(have_mplock);
768 if (kdb_trap (type, 0, frame))
773 MAKEMPSAFE(have_mplock);
774 trap_fatal(frame, FALSE, eva);
777 MAKEMPSAFE(have_mplock);
778 trap_fatal(frame, FALSE, eva);
783 * Ignore this trap generated from a spurious SIGTRAP.
785 * single stepping in / syscalls leads to spurious / SIGTRAP
788 * Haiku (c) 2007 Simon 'corecode' Schubert
794 * Translate fault for emulators (e.g. Linux)
796 if (*p->p_sysent->sv_transtrap)
797 i = (*p->p_sysent->sv_transtrap)(i, type);
799 MAKEMPSAFE(have_mplock);
800 trapsignal(lp, i, ucode);
803 if (type <= MAX_TRAP_MSG) {
804 uprintf("fatal process exception: %s",
806 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
807 uprintf(", fault VA = 0x%lx", (u_long)eva);
817 KASSERT(crit_count == td->td_critcount,
818 ("trap: critical section count mismatch! %d/%d",
819 crit_count, td->td_pri));
820 KASSERT(curstop == td->td_toks_stop,
821 ("trap: extra tokens held after trap! %ld/%ld",
822 curstop - &td->td_toks_base,
823 td->td_toks_stop - &td->td_toks_base));
828 trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
831 struct vmspace *vm = NULL;
835 thread_t td = curthread;
836 struct lwp *lp = td->td_lwp;
839 va = trunc_page(eva);
840 if (usermode == FALSE) {
842 * This is a fault on kernel virtual memory.
847 * This is a fault on non-kernel virtual memory.
848 * vm is initialized above to NULL. If curproc is NULL
849 * or curproc->p_vmspace is NULL the fault is fatal.
852 vm = lp->lwp_vmspace;
860 if (frame->tf_err & PGEX_W)
861 ftype = VM_PROT_READ | VM_PROT_WRITE;
863 ftype = VM_PROT_READ;
865 if (map != &kernel_map) {
867 * Keep swapout from messing with us during this
873 * Grow the stack if necessary
875 /* grow_stack returns false only if va falls into
876 * a growable stack region and the stack growth
877 * fails. It returns true if va was not within
878 * a growable stack region, or if the stack
881 if (!grow_stack (lp->lwp_proc, va)) {
889 fault_flags |= VM_FAULT_BURST;
890 if (ftype & VM_PROT_WRITE)
891 fault_flags |= VM_FAULT_DIRTY;
893 fault_flags |= VM_FAULT_NORMAL;
894 rv = vm_fault(map, va, ftype, fault_flags);
899 * Don't have to worry about process locking or stacks in the kernel.
901 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
904 if (rv == KERN_SUCCESS)
908 if (td->td_gd->gd_intr_nesting_level == 0 &&
909 td->td_pcb->pcb_onfault) {
910 frame->tf_rip = (register_t)td->td_pcb->pcb_onfault;
913 trap_fatal(frame, usermode, eva);
918 * NOTE: on x86_64 we have a tf_addr field in the trapframe, no
919 * kludge is needed to pass the fault address to signal handlers.
921 struct proc *p = td->td_proc;
922 kprintf("seg-fault accessing address %p rip=%p pid=%d p_comm=%s\n",
923 (void *)va, (void *)frame->tf_rip, p->p_pid, p->p_comm);
924 /* Debugger("seg-fault"); */
926 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
930 trap_fatal(struct trapframe *frame, int usermode, vm_offset_t eva)
935 code = frame->tf_xflags;
936 type = frame->tf_trapno;
938 if (type <= MAX_TRAP_MSG) {
939 kprintf("\n\nFatal trap %d: %s while in %s mode\n",
940 type, trap_msg[type],
941 (usermode ? "user" : "kernel"));
943 /* two separate prints in case of a trap on an unmapped page */
944 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
945 if (type == T_PAGEFLT) {
946 kprintf("fault virtual address = %p\n", (void *)eva);
947 kprintf("fault code = %s %s, %s\n",
948 usermode ? "user" : "supervisor",
949 code & PGEX_W ? "write" : "read",
950 code & PGEX_P ? "protection violation" : "page not present");
952 kprintf("instruction pointer = 0x%lx:0x%lx\n",
953 frame->tf_cs & 0xffff, frame->tf_rip);
955 ss = frame->tf_ss & 0xffff;
958 ss = GSEL(GDATA_SEL, SEL_KPL);
959 rsp = (long)&frame->tf_rsp;
961 kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp);
962 kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp);
963 kprintf("processor eflags = ");
964 if (frame->tf_rflags & PSL_T)
965 kprintf("trace trap, ");
966 if (frame->tf_rflags & PSL_I)
967 kprintf("interrupt enabled, ");
968 if (frame->tf_rflags & PSL_NT)
969 kprintf("nested task, ");
970 if (frame->tf_rflags & PSL_RF)
973 if (frame->tf_eflags & PSL_VM)
976 kprintf("IOPL = %jd\n", (intmax_t)((frame->tf_rflags & PSL_IOPL) >> 12));
977 kprintf("current process = ");
979 kprintf("%lu (%s)\n",
980 (u_long)curproc->p_pid, curproc->p_comm ?
981 curproc->p_comm : "");
985 kprintf("current thread = pri %d ", curthread->td_pri);
986 if (curthread->td_critcount)
991 * we probably SHOULD have stopped the other CPUs before now!
992 * another CPU COULD have been touching cpl at this moment...
994 kprintf(" <- SMP: XXX");
1002 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
1005 kprintf("trap number = %d\n", type);
1006 if (type <= MAX_TRAP_MSG)
1007 panic("%s", trap_msg[type]);
1009 panic("unknown/reserved trap");
1013 * Double fault handler. Called when a fault occurs while writing
1014 * a frame for a trap/exception onto the stack. This usually occurs
1015 * when the stack overflows (such is the case with infinite recursion,
1018 * XXX Note that the current PTD gets replaced by IdlePTD when the
1019 * task switch occurs. This means that the stack that was active at
1020 * the time of the double fault is not available at <kstack> unless
1021 * the machine was idle when the double fault occurred. The downside
1022 * of this is that "trace <ebp>" in ddb won't work.
1025 dblfault_handler(void)
1028 struct mdglobaldata *gd = mdcpu;
1031 kprintf("\nFatal double fault:\n");
1033 kprintf("rip = 0x%lx\n", gd->gd_common_tss.tss_rip);
1034 kprintf("rsp = 0x%lx\n", gd->gd_common_tss.tss_rsp);
1035 kprintf("rbp = 0x%lx\n", gd->gd_common_tss.tss_rbp);
1037 /* two separate prints in case of a trap on an unmapped page */
1038 kprintf("cpuid = %d\n", mycpu->gd_cpuid);
1039 panic("double fault");
1043 * Compensate for 386 brain damage (missing URKR).
1044 * This is a little simpler than the pagefault handler in trap() because
1045 * it the page tables have already been faulted in and high addresses
1046 * are thrown out early for other reasons.
1049 trapwrite(unsigned addr)
1056 va = trunc_page((vm_offset_t)addr);
1058 * XXX - MAX is END. Changed > to >= for temp. fix.
1060 if (va >= VM_MAX_USER_ADDRESS)
1063 lp = curthread->td_lwp;
1064 vm = lp->lwp_vmspace;
1066 PHOLD(lp->lwp_proc);
1068 if (!grow_stack (lp->lwp_proc, va)) {
1069 PRELE(lp->lwp_proc);
1074 * fault the data page
1076 rv = vm_fault(&vm->vm_map, va, VM_PROT_WRITE, VM_FAULT_DIRTY);
1078 PRELE(lp->lwp_proc);
1080 if (rv != KERN_SUCCESS)
1087 * syscall2 - MP aware system call request C handler
1089 * A system call is essentially treated as a trap except that the
1090 * MP lock is not held on entry or return. We are responsible for
1091 * obtaining the MP lock if necessary and for handling ASTs
1092 * (e.g. a task switch) prior to return.
1094 * In general, only simple access and manipulation of curproc and
1095 * the current stack is allowed without having to hold MP lock.
1097 * MPSAFE - note that large sections of this routine are run without
1101 syscall2(struct trapframe *frame)
1103 struct thread *td = curthread;
1104 struct proc *p = td->td_proc;
1105 struct lwp *lp = td->td_lwp;
1107 struct sysent *callp;
1108 register_t orig_tf_rflags;
1113 int crit_count = td->td_critcount;
1114 lwkt_tokref_t curstop = td->td_toks_stop;
1116 int have_mplock = 0;
1120 union sysunion args;
1121 register_t *argsdst;
1123 mycpu->gd_cnt.v_syscall++;
1125 KTR_LOG(kernentry_syscall, lp->lwp_proc->p_pid, lp->lwp_tid,
1128 userenter(td, p); /* lazy raise our priority */
1135 sticks = (int)td->td_sticks;
1136 orig_tf_rflags = frame->tf_rflags;
1139 * Virtual kernel intercept - if a VM context managed by a virtual
1140 * kernel issues a system call the virtual kernel handles it, not us.
1141 * Restore the virtual kernel context and return from its system
1142 * call. The current frame is copied out to the virtual kernel.
1144 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1145 vkernel_trap(lp, frame);
1146 error = EJUSTRETURN;
1151 * Get the system call parameters and account for time
1153 lp->lwp_md.md_regs = frame;
1154 params = (caddr_t)frame->tf_rsp + sizeof(register_t);
1155 code = frame->tf_rax;
1157 if (p->p_sysent->sv_prepsyscall) {
1158 (*p->p_sysent->sv_prepsyscall)(
1159 frame, (int *)(&args.nosys.sysmsg + 1),
1162 if (code == SYS_syscall || code == SYS___syscall) {
1163 code = frame->tf_rdi;
1169 if (p->p_sysent->sv_mask)
1170 code &= p->p_sysent->sv_mask;
1172 if (code >= p->p_sysent->sv_size)
1173 callp = &p->p_sysent->sv_table[0];
1175 callp = &p->p_sysent->sv_table[code];
1177 narg = callp->sy_narg & SYF_ARGMASK;
1180 * On x86_64 we get up to six arguments in registers. The rest are
1181 * on the stack. The first six members of 'struct trapframe' happen
1182 * to be the registers used to pass arguments, in exactly the right
1185 argp = &frame->tf_rdi;
1187 argsdst = (register_t *)(&args.nosys.sysmsg + 1);
1189 * JG can we overflow the space pointed to by 'argsdst'
1190 * either with 'bcopy' or with 'copyin'?
1192 bcopy(argp, argsdst, sizeof(register_t) * regcnt);
1194 * copyin is MP aware, but the tracing code is not
1196 if (narg > regcnt) {
1197 KASSERT(params != NULL, ("copyin args with no params!"));
1198 error = copyin(params, &argsdst[regcnt],
1199 (narg - regcnt) * sizeof(register_t));
1202 if (KTRPOINT(td, KTR_SYSCALL)) {
1203 MAKEMPSAFE(have_mplock);
1205 ktrsyscall(lp, code, narg,
1206 (void *)(&args.nosys.sysmsg + 1));
1214 if (KTRPOINT(td, KTR_SYSCALL)) {
1215 MAKEMPSAFE(have_mplock);
1216 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1221 * Default return value is 0 (will be copied to %rax). Double-value
1222 * returns use %rax and %rdx. %rdx is left unchanged for system
1223 * calls which return only one result.
1225 args.sysmsg_fds[0] = 0;
1226 args.sysmsg_fds[1] = frame->tf_rdx;
1229 * The syscall might manipulate the trap frame. If it does it
1230 * will probably return EJUSTRETURN.
1232 args.sysmsg_frame = frame;
1234 STOPEVENT(p, S_SCE, narg); /* MP aware */
1237 * NOTE: All system calls run MPSAFE now. The system call itself
1238 * is responsible for getting the MP lock.
1240 error = (*callp->sy_call)(&args);
1243 kprintf("system call %d returned %d\n", code, error);
1248 * MP SAFE (we may or may not have the MP lock at this point)
1253 * Reinitialize proc pointer `p' as it may be different
1254 * if this is a child returning from fork syscall.
1257 lp = curthread->td_lwp;
1258 frame->tf_rax = args.sysmsg_fds[0];
1259 frame->tf_rdx = args.sysmsg_fds[1];
1260 frame->tf_rflags &= ~PSL_C;
1264 * Reconstruct pc, we know that 'syscall' is 2 bytes.
1265 * We have to do a full context restore so that %r10
1266 * (which was holding the value of %rcx) is restored for
1267 * the next iteration.
1269 frame->tf_rip -= frame->tf_err;
1270 frame->tf_r10 = frame->tf_rcx;
1275 panic("Unexpected EASYNC return value (for now)");
1278 if (p->p_sysent->sv_errsize) {
1279 if (error >= p->p_sysent->sv_errsize)
1280 error = -1; /* XXX */
1282 error = p->p_sysent->sv_errtbl[error];
1284 frame->tf_rax = error;
1285 frame->tf_rflags |= PSL_C;
1290 * Traced syscall. trapsignal() is not MP aware.
1292 if (orig_tf_rflags & PSL_T) {
1293 MAKEMPSAFE(have_mplock);
1294 frame->tf_rflags &= ~PSL_T;
1295 trapsignal(lp, SIGTRAP, 0);
1299 * Handle reschedule and other end-of-syscall issues
1301 userret(lp, frame, sticks);
1304 if (KTRPOINT(td, KTR_SYSRET)) {
1305 MAKEMPSAFE(have_mplock);
1306 ktrsysret(lp, code, error, args.sysmsg_result);
1311 * This works because errno is findable through the
1312 * register set. If we ever support an emulation where this
1313 * is not the case, this code will need to be revisited.
1315 STOPEVENT(p, S_SCX, code);
1319 * Release the MP lock if we had to get it
1323 KTR_LOG(kernentry_syscall_ret, lp->lwp_proc->p_pid, lp->lwp_tid, error);
1325 KASSERT(&td->td_toks_base == td->td_toks_stop,
1326 ("syscall: critical section count mismatch! %d/%d",
1327 crit_count, td->td_pri));
1328 KASSERT(curstop == td->td_toks_stop,
1329 ("syscall: extra tokens held after trap! %ld",
1330 td->td_toks_stop - &td->td_toks_base));
1335 * NOTE: mplock not held at any point
1338 fork_return(struct lwp *lp, struct trapframe *frame)
1340 frame->tf_rax = 0; /* Child returns zero */
1341 frame->tf_rflags &= ~PSL_C; /* success */
1344 generic_lwp_return(lp, frame);
1345 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1349 * Simplified back end of syscall(), used when returning from fork()
1350 * directly into user mode.
1352 * This code will return back into the fork trampoline code which then
1355 * NOTE: The mplock is not held at any point.
1358 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1360 struct proc *p = lp->lwp_proc;
1363 * Newly forked processes are given a kernel priority. We have to
1364 * adjust the priority to a normal user priority and fake entry
1365 * into the kernel (call userenter()) to install a passive release
1366 * function just in case userret() decides to stop the process. This
1367 * can occur when ^Z races a fork. If we do not install the passive
1368 * release function the current process designation will not be
1369 * released when the thread goes to sleep.
1371 lwkt_setpri_self(TDPRI_USER_NORM);
1372 userenter(lp->lwp_thread, p);
1373 userret(lp, frame, 0);
1375 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1376 ktrsysret(lp, SYS_fork, 0, 0);
1378 lp->lwp_flags |= LWP_PASSIVE_ACQ;
1380 lp->lwp_flags &= ~LWP_PASSIVE_ACQ;
1384 * doreti has turned into this. The frame is directly on the stack. We
1385 * pull everything else we need (fpu and tls context) from the current
1388 * Note on fpu interactions: In a virtual kernel, the fpu context for
1389 * an emulated user mode process is not shared with the virtual kernel's
1390 * fpu context, so we only have to 'stack' fpu contexts within the virtual
1391 * kernel itself, and not even then since the signal() contexts that we care
1392 * about save and restore the FPU state (I think anyhow).
1394 * vmspace_ctl() returns an error only if it had problems instaling the
1395 * context we supplied or problems copying data to/from our VM space.
1398 go_user(struct intrframe *frame)
1400 struct trapframe *tf = (void *)&frame->if_rdi;
1405 * Interrupts may be disabled on entry, make sure all signals
1406 * can be received before beginning our loop.
1411 * Switch to the current simulated user process, then call
1412 * user_trap() when we break out of it (usually due to a signal).
1416 * Tell the real kernel whether it is ok to use the FP
1419 if (mdcpu->gd_npxthread == curthread) {
1420 tf->tf_xflags &= ~PGEX_FPFAULT;
1422 tf->tf_xflags |= PGEX_FPFAULT;
1426 * Run emulated user process context. This call interlocks
1427 * with new mailbox signals.
1429 * Set PGEX_U unconditionally, indicating a user frame (the
1430 * bit is normally set only by T_PAGEFLT).
1433 id = (void *)vtophys(curproc->p_vmspace->vm_pmap.pm_pml4);
1435 id = &curproc->p_vmspace->vm_pmap;
1437 r = vmspace_ctl(id, VMSPACE_CTL_RUN, tf, &curthread->td_savevext);
1439 frame->if_xflags |= PGEX_U;
1441 kprintf("GO USER %d trap %ld EVA %08lx RIP %08lx RSP %08lx XFLAGS %02lx/%02lx\n",
1442 r, tf->tf_trapno, tf->tf_addr, tf->tf_rip, tf->tf_rsp,
1443 tf->tf_xflags, frame->if_xflags);
1447 panic("vmspace_ctl failed error %d", errno);
1449 if (tf->tf_trapno) {
1453 if (mycpu->gd_reqflags & RQF_AST_MASK) {
1454 tf->tf_trapno = T_ASTFLT;
1462 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1463 * fault (which is then passed back to the virtual kernel) if an attempt is
1464 * made to use the FP unit.
1466 * XXX this is a fairly big hack.
1469 set_vkernel_fp(struct trapframe *frame)
1471 struct thread *td = curthread;
1473 if (frame->tf_xflags & PGEX_FPFAULT) {
1474 td->td_pcb->pcb_flags |= FP_VIRTFP;
1475 if (mdcpu->gd_npxthread == td)
1478 td->td_pcb->pcb_flags &= ~FP_VIRTFP;
1483 * Called from vkernel_trap() to fixup the vkernel's syscall
1484 * frame for vmspace_ctl() return.
1487 cpu_vkernel_trap(struct trapframe *frame, int error)
1489 frame->tf_rax = error;
1491 frame->tf_rflags |= PSL_C;
1493 frame->tf_rflags &= ~PSL_C;