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 $
41 * $DragonFly: src/sys/platform/pc64/amd64/trap.c,v 1.3 2008/09/09 04:06:18 dillon Exp $
45 * AMD64 Trap and System call handling
49 #include "opt_ktrace.h"
51 #include <machine/frame.h>
52 #include <sys/param.h>
53 #include <sys/systm.h>
54 #include <sys/kernel.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/intr_machdep.h>
85 #include <sys/thread2.h>
89 #define MAKEMPSAFE(have_mplock) \
90 if (have_mplock == 0) { \
97 #define MAKEMPSAFE(have_mplock)
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");
147 static int panic_on_nmi = 1;
148 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
149 &panic_on_nmi, 0, "Panic on NMI");
150 static int fast_release;
151 SYSCTL_INT(_machdep, OID_AUTO, fast_release, CTLFLAG_RW,
152 &fast_release, 0, "Passive Release was optimal");
153 static int slow_release;
154 SYSCTL_INT(_machdep, OID_AUTO, slow_release, CTLFLAG_RW,
155 &slow_release, 0, "Passive Release was nonoptimal");
157 static int syscall_mpsafe = 1;
158 SYSCTL_INT(_kern, OID_AUTO, syscall_mpsafe, CTLFLAG_RW,
159 &syscall_mpsafe, 0, "Allow MPSAFE marked syscalls to run without BGL");
160 TUNABLE_INT("kern.syscall_mpsafe", &syscall_mpsafe);
161 static int trap_mpsafe = 1;
162 SYSCTL_INT(_kern, OID_AUTO, trap_mpsafe, CTLFLAG_RW,
163 &trap_mpsafe, 0, "Allow traps to mostly run without the BGL");
164 TUNABLE_INT("kern.trap_mpsafe", &trap_mpsafe);
168 * userenter() passively intercepts the thread switch function to increase
169 * the thread priority from a user priority to a kernel priority, reducing
170 * syscall and trap overhead for the case where no switch occurs.
174 userenter(struct thread *curtd)
176 curtd->td_release = lwkt_passive_release;
180 * Handle signals, upcalls, profiling, and other AST's and/or tasks that
181 * must be completed before we can return to or try to return to userland.
183 * Note that td_sticks is a 64 bit quantity, but there's no point doing 64
184 * arithmatic on the delta calculation so the absolute tick values are
185 * truncated to an integer.
188 userret(struct lwp *lp, struct trapframe *frame, int sticks)
190 struct proc *p = lp->lwp_proc;
194 * Charge system time if profiling. Note: times are in microseconds.
195 * This may do a copyout and block, so do it first even though it
196 * means some system time will be charged as user time.
198 if (p->p_flag & P_PROFIL) {
199 addupc_task(p, frame->tf_rip,
200 (u_int)((int)lp->lwp_thread->td_sticks - sticks));
205 * If the jungle wants us dead, so be it.
207 if (lp->lwp_flag & LWP_WEXIT) {
210 rel_mplock(); /* NOT REACHED */
214 * Block here if we are in a stopped state.
216 if (p->p_stat == SSTOP) {
224 * Post any pending upcalls. If running a virtual kernel be sure
225 * to restore the virtual kernel's vmspace before posting the upcall.
227 if (p->p_flag & P_UPCALLPEND) {
228 p->p_flag &= ~P_UPCALLPEND;
236 * Post any pending signals. If running a virtual kernel be sure
237 * to restore the virtual kernel's vmspace before posting the signal.
239 if ((sig = CURSIG_TRACE(lp)) != 0) {
247 * block here if we are swapped out, but still process signals
248 * (such as SIGKILL). proc0 (the swapin scheduler) is already
249 * aware of our situation, we do not have to wake it up.
251 if (p->p_flag & P_SWAPPEDOUT) {
253 p->p_flag |= P_SWAPWAIT;
255 if (p->p_flag & P_SWAPWAIT)
256 tsleep(p, PCATCH, "SWOUT", 0);
257 p->p_flag &= ~P_SWAPWAIT;
263 * Make sure postsig() handled request to restore old signal mask after
264 * running signal handler.
266 KKASSERT((lp->lwp_flag & LWP_OLDMASK) == 0);
270 * Cleanup from userenter and any passive release that might have occured.
271 * We must reclaim the current-process designation before we can return
272 * to usermode. We also handle both LWKT and USER reschedule requests.
275 userexit(struct lwp *lp)
277 struct thread *td = lp->lwp_thread;
278 /* globaldata_t gd = td->td_gd;*/
281 * Handle stop requests at kernel priority. Any requests queued
282 * after this loop will generate another AST.
284 while (lp->lwp_proc->p_stat == SSTOP) {
291 * Reduce our priority in preparation for a return to userland. If
292 * our passive release function was still in place, our priority was
293 * never raised and does not need to be reduced.
295 lwkt_passive_recover(td);
298 * Become the current user scheduled process if we aren't already,
299 * and deal with reschedule requests and other factors.
301 lp->lwp_proc->p_usched->acquire_curproc(lp);
302 /* WARNING: we may have migrated cpu's */
303 /* gd = td->td_gd; */
306 #if !defined(KTR_KERNENTRY)
307 #define KTR_KERNENTRY KTR_ALL
309 KTR_INFO_MASTER(kernentry);
310 KTR_INFO(KTR_KERNENTRY, kernentry, trap, 0, "STR",
311 sizeof(long) + sizeof(long) + sizeof(long) + sizeof(vm_offset_t));
312 KTR_INFO(KTR_KERNENTRY, kernentry, trap_ret, 0, "STR",
313 sizeof(long) + sizeof(long));
314 KTR_INFO(KTR_KERNENTRY, kernentry, syscall, 0, "STR",
315 sizeof(long) + sizeof(long) + sizeof(long));
316 KTR_INFO(KTR_KERNENTRY, kernentry, syscall_ret, 0, "STR",
317 sizeof(long) + sizeof(long) + sizeof(long));
318 KTR_INFO(KTR_KERNENTRY, kernentry, fork_ret, 0, "STR",
319 sizeof(long) + sizeof(long));
322 * Exception, fault, and trap interface to the kernel.
323 * This common code is called from assembly language IDT gate entry
324 * routines that prepare a suitable stack frame, and restore this
325 * frame after the exception has been processed.
327 * This function is also called from doreti in an interlock to handle ASTs.
328 * For example: hardwareint->INTROUTINE->(set ast)->doreti->trap
330 * NOTE! We have to retrieve the fault address prior to obtaining the
331 * MP lock because get_mplock() may switch out. YYY cr2 really ought
332 * to be retrieved by the assembly code, not here.
334 * XXX gd_trap_nesting_level currently prevents lwkt_switch() from panicing
335 * if an attempt is made to switch from a fast interrupt or IPI. This is
336 * necessary to properly take fatal kernel traps on SMP machines if
337 * get_mplock() has to block.
341 trap(struct trapframe *frame)
343 struct globaldata *gd = mycpu;
344 struct thread *td = gd->gd_curthread;
345 struct lwp *lp = td->td_lwp;
348 int i = 0, ucode = 0, type, code;
353 int crit_count = td->td_pri & ~TDPRI_MASK;
361 kprintf0("\"%s\" type=%ld\n",
362 trap_msg[frame->tf_trapno], frame->tf_trapno);
363 kprintf0(" rip=%lx rsp=%lx\n", frame->tf_rip, frame->tf_rsp);
364 kprintf0(" err=%lx addr=%lx\n", frame->tf_err, frame->tf_addr);
365 kprintf0(" cs=%lx ss=%lx rflags=%lx\n", (unsigned long)frame->tf_cs, (unsigned long)frame->tf_ss, frame->tf_rflags);
370 ++gd->gd_trap_nesting_level;
371 MAKEMPSAFE(have_mplock);
372 trap_fatal(frame, frame->tf_addr);
373 --gd->gd_trap_nesting_level;
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;
391 if (trap_mpsafe == 0) {
392 ++gd->gd_trap_nesting_level;
393 MAKEMPSAFE(have_mplock);
394 --gd->gd_trap_nesting_level;
398 if ((frame->tf_rflags & PSL_I) == 0) {
400 * Buggy application or kernel code has disabled interrupts
401 * and then trapped. Enabling interrupts now is wrong, but
402 * it is better than running with interrupts disabled until
403 * they are accidentally enabled later.
405 type = frame->tf_trapno;
406 if (ISPL(frame->tf_cs) == SEL_UPL) {
407 MAKEMPSAFE(have_mplock);
408 /* JG curproc can be NULL */
410 "pid %ld (%s): trap %d with interrupts disabled\n",
411 (long)curproc->p_pid, curproc->p_comm, type);
412 } else if (type != T_NMI && type != T_BPTFLT &&
415 * XXX not quite right, since this may be for a
416 * multiple fault in user mode.
418 MAKEMPSAFE(have_mplock);
419 kprintf("kernel trap %d with interrupts disabled\n",
425 type = frame->tf_trapno;
426 code = frame->tf_err;
428 if (ISPL(frame->tf_cs) == SEL_UPL) {
431 KTR_LOG(kernentry_trap, p->p_pid, lp->lwp_tid,
432 frame->tf_trapno, eva);
436 sticks = (int)td->td_sticks;
437 lp->lwp_md.md_regs = frame;
440 case T_PRIVINFLT: /* privileged instruction fault */
445 case T_BPTFLT: /* bpt instruction fault */
446 case T_TRCTRAP: /* trace trap */
447 frame->tf_rflags &= ~PSL_T;
451 case T_ARITHTRAP: /* arithmetic trap */
464 case T_ASTFLT: /* Allow process switch */
465 mycpu->gd_cnt.v_soft++;
466 if (mycpu->gd_reqflags & RQF_AST_OWEUPC) {
467 atomic_clear_int_nonlocked(&mycpu->gd_reqflags,
469 addupc_task(p, p->p_prof.pr_addr,
474 case T_PROTFLT: /* general protection fault */
475 case T_SEGNPFLT: /* segment not present fault */
476 case T_TSSFLT: /* invalid TSS fault */
477 case T_DOUBLEFLT: /* double fault */
479 ucode = code + BUS_SEGM_FAULT ;
483 case T_PAGEFLT: /* page fault */
484 MAKEMPSAFE(have_mplock);
485 i = trap_pfault(frame, TRUE);
486 if (frame->tf_rip == 0)
487 kprintf("T_PAGEFLT: Warning %%rip == 0!\n");
496 case T_DIVIDE: /* integer divide fault */
502 MAKEMPSAFE(have_mplock);
503 /* machine/parity/power fail/"kitchen sink" faults */
504 if (isa_nmi(code) == 0) {
507 * NMI can be hooked up to a pushbutton
511 kprintf ("NMI ... going to debugger\n");
512 kdb_trap(type, 0, frame);
516 } else if (panic_on_nmi)
517 panic("NMI indicates hardware failure");
520 case T_OFLOW: /* integer overflow fault */
525 case T_BOUND: /* bounds check fault */
532 * Virtual kernel intercept - pass the DNA exception
533 * to the virtual kernel if it asked to handle it.
534 * This occurs when the virtual kernel is holding
535 * onto the FP context for a different emulated
536 * process then the one currently running.
538 * We must still call npxdna() since we may have
539 * saved FP state that the virtual kernel needs
540 * to hand over to a different emulated process.
542 if (lp->lwp_vkernel && lp->lwp_vkernel->ve &&
543 (td->td_pcb->pcb_flags & FP_VIRTFP)
550 * The kernel may have switched out the FP unit's
551 * state, causing the user process to take a fault
552 * when it tries to use the FP unit. Restore the
558 ucode = FPE_FPU_NP_TRAP;
561 case T_FPOPFLT: /* FPU operand fetch fault */
566 case T_XMMFLT: /* SIMD floating-point exception */
575 case T_PAGEFLT: /* page fault */
576 MAKEMPSAFE(have_mplock);
577 trap_pfault(frame, FALSE);
582 * The kernel is apparently using fpu for copying.
583 * XXX this should be fatal unless the kernel has
584 * registered such use.
590 case T_STKFLT: /* stack fault */
593 case T_PROTFLT: /* general protection fault */
594 case T_SEGNPFLT: /* segment not present fault */
596 * Invalid segment selectors and out of bounds
597 * %rip's and %rsp's can be set up in user mode.
598 * This causes a fault in kernel mode when the
599 * kernel tries to return to user mode. We want
600 * to get this fault so that we can fix the
601 * problem here and not have to check all the
602 * selectors and pointers when the user changes
605 kprintf("trap.c line %d\n", __LINE__);
606 if (mycpu->gd_intr_nesting_level == 0) {
607 if (td->td_pcb->pcb_onfault) {
608 frame->tf_rip = (register_t)
609 td->td_pcb->pcb_onfault;
612 if (frame->tf_rip == (long)doreti_iret) {
613 frame->tf_rip = (long)doreti_iret_fault;
621 * PSL_NT can be set in user mode and isn't cleared
622 * automatically when the kernel is entered. This
623 * causes a TSS fault when the kernel attempts to
624 * `iret' because the TSS link is uninitialized. We
625 * want to get this fault so that we can fix the
626 * problem here and not every time the kernel is
629 if (frame->tf_rflags & PSL_NT) {
630 frame->tf_rflags &= ~PSL_NT;
635 case T_TRCTRAP: /* trace trap */
637 if (frame->tf_rip == (int)IDTVEC(syscall)) {
639 * We've just entered system mode via the
640 * syscall lcall. Continue single stepping
641 * silently until the syscall handler has
646 if (frame->tf_rip == (int)IDTVEC(syscall) + 1) {
648 * The syscall handler has now saved the
649 * flags. Stop single stepping it.
651 frame->tf_rflags &= ~PSL_T;
657 * Ignore debug register trace traps due to
658 * accesses in the user's address space, which
659 * can happen under several conditions such as
660 * if a user sets a watchpoint on a buffer and
661 * then passes that buffer to a system call.
662 * We still want to get TRCTRAPS for addresses
663 * in kernel space because that is useful when
664 * debugging the kernel.
667 if (user_dbreg_trap()) {
669 * Reset breakpoint bits because the
672 /* XXX check upper bits here */
673 load_dr6(rdr6() & 0xfffffff0);
678 * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
682 * If DDB is enabled, let it handle the debugger trap.
683 * Otherwise, debugger traps "can't happen".
686 MAKEMPSAFE(have_mplock);
687 if (kdb_trap(type, 0, frame))
693 MAKEMPSAFE(have_mplock);
694 /* machine/parity/power fail/"kitchen sink" faults */
696 if (isa_nmi(code) == 0) {
699 * NMI can be hooked up to a pushbutton
703 kprintf ("NMI ... going to debugger\n");
704 kdb_trap(type, 0, frame);
708 } else if (panic_on_nmi == 0)
711 #endif /* NISA > 0 */
713 MAKEMPSAFE(have_mplock);
714 trap_fatal(frame, 0);
719 * Virtual kernel intercept - if the fault is directly related to a
720 * VM context managed by a virtual kernel then let the virtual kernel
723 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
724 vkernel_trap(lp, frame);
729 * Virtual kernel intercept - if the fault is directly related to a
730 * VM context managed by a virtual kernel then let the virtual kernel
733 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
734 vkernel_trap(lp, frame);
739 * Translate fault for emulators (e.g. Linux)
741 if (*p->p_sysent->sv_transtrap)
742 i = (*p->p_sysent->sv_transtrap)(i, type);
744 MAKEMPSAFE(have_mplock);
745 trapsignal(lp, i, ucode);
748 if (type <= MAX_TRAP_MSG) {
749 uprintf("fatal process exception: %s",
751 if ((type == T_PAGEFLT) || (type == T_PROTFLT))
752 uprintf(", fault VA = 0x%lx", frame->tf_addr);
759 if (ISPL(frame->tf_cs) == SEL_UPL)
760 KASSERT(td->td_mpcount == have_mplock, ("badmpcount trap/end from %p", (void *)frame->tf_rip));
762 userret(lp, frame, sticks);
769 if (p != NULL && lp != NULL)
770 KTR_LOG(kernentry_trap_ret, p->p_pid, lp->lwp_tid);
772 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
773 ("syscall: critical section count mismatch! %d/%d",
774 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
779 trap_pfault(struct trapframe *frame, int usermode)
782 struct vmspace *vm = NULL;
786 thread_t td = curthread;
787 struct lwp *lp = td->td_lwp;
789 va = trunc_page(frame->tf_addr);
790 if (va >= VM_MIN_KERNEL_ADDRESS) {
792 * Don't allow user-mode faults in kernel address space.
800 * This is a fault on non-kernel virtual memory.
801 * vm is initialized above to NULL. If curproc is NULL
802 * or curproc->p_vmspace is NULL the fault is fatal.
805 vm = lp->lwp_vmspace;
814 * PGEX_I is defined only if the execute disable bit capability is
815 * supported and enabled.
817 if (frame->tf_err & PGEX_W)
818 ftype = VM_PROT_WRITE;
820 else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
821 ftype = VM_PROT_EXECUTE;
824 ftype = VM_PROT_READ;
826 if (map != &kernel_map) {
828 * Keep swapout from messing with us during this
834 * Grow the stack if necessary
836 /* grow_stack returns false only if va falls into
837 * a growable stack region and the stack growth
838 * fails. It returns true if va was not within
839 * a growable stack region, or if the stack
842 if (!grow_stack(lp->lwp_proc, va)) {
848 /* Fault in the user page: */
849 rv = vm_fault(map, va, ftype,
850 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
856 * Don't have to worry about process locking or stacks
859 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
862 if (rv == KERN_SUCCESS)
866 if (td->td_gd->gd_intr_nesting_level == 0 &&
867 td->td_pcb->pcb_onfault) {
868 frame->tf_rip = (register_t)td->td_pcb->pcb_onfault;
871 trap_fatal(frame, frame->tf_addr);
876 * NOTE: on amd64 we have a tf_addr field in the trapframe, no
877 * kludge is needed to pass the fault address to signal handlers.
879 struct proc *p = td->td_proc;
880 kprintf("seg-fault accessing address %p rip=%p pid=%d p_comm=%s\n",
881 (void *)va, (void *)frame->tf_rip, p->p_pid, p->p_comm);
882 /* Debugger("seg-fault"); */
884 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
888 trap_fatal(struct trapframe *frame, vm_offset_t eva)
893 struct soft_segment_descriptor softseg;
896 code = frame->tf_err;
897 type = frame->tf_trapno;
898 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)], &softseg);
900 if (type <= MAX_TRAP_MSG)
901 msg = trap_msg[type];
904 kprintf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
905 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
907 /* three separate prints in case of a trap on an unmapped page */
908 kprintf("mp_lock = %08x; ", mp_lock);
909 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
910 kprintf("lapic->id = %08x\n", lapic->id);
912 if (type == T_PAGEFLT) {
913 kprintf("fault virtual address = 0x%lx\n", eva);
914 kprintf("fault code = %s %s %s, %s\n",
915 code & PGEX_U ? "user" : "supervisor",
916 code & PGEX_W ? "write" : "read",
917 code & PGEX_I ? "instruction" : "data",
918 code & PGEX_P ? "protection violation" : "page not present");
920 kprintf("instruction pointer = 0x%lx:0x%lx\n",
921 frame->tf_cs & 0xffff, frame->tf_rip);
922 if (ISPL(frame->tf_cs) == SEL_UPL) {
923 ss = frame->tf_ss & 0xffff;
926 ss = GSEL(GDATA_SEL, SEL_KPL);
927 rsp = (long)&frame->tf_rsp;
929 kprintf("stack pointer = 0x%x:0x%lx\n", ss, rsp);
930 kprintf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp);
931 kprintf("code segment = base 0x%lx, limit 0x%lx, type 0x%x\n",
932 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
933 kprintf(" = DPL %d, pres %d, long %d, def32 %d, gran %d\n",
934 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32,
936 kprintf("processor eflags = ");
937 if (frame->tf_rflags & PSL_T)
938 kprintf("trace trap, ");
939 if (frame->tf_rflags & PSL_I)
940 kprintf("interrupt enabled, ");
941 if (frame->tf_rflags & PSL_NT)
942 kprintf("nested task, ");
943 if (frame->tf_rflags & PSL_RF)
945 kprintf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12);
946 kprintf("current process = ");
949 (u_long)curproc->p_pid);
953 kprintf("current thread = pri %d ", curthread->td_pri);
954 if (curthread->td_pri >= TDPRI_CRIT)
959 if ((debugger_on_panic || db_active) && kdb_trap(type, code, frame))
962 kprintf("trap number = %d\n", type);
963 if (type <= MAX_TRAP_MSG)
964 panic("%s", trap_msg[type]);
966 panic("unknown/reserved trap");
970 * Double fault handler. Called when a fault occurs while writing
971 * a frame for a trap/exception onto the stack. This usually occurs
972 * when the stack overflows (such is the case with infinite recursion,
976 dblfault_handler(struct trapframe *frame)
978 kprintf0("DOUBLE FAULT\n");
979 kprintf("\nFatal double fault\n");
980 kprintf("rip = 0x%lx\n", frame->tf_rip);
981 kprintf("rsp = 0x%lx\n", frame->tf_rsp);
982 kprintf("rbp = 0x%lx\n", frame->tf_rbp);
984 /* three separate prints in case of a trap on an unmapped page */
985 kprintf("mp_lock = %08x; ", mp_lock);
986 kprintf("cpuid = %d; ", mycpu->gd_cpuid);
987 kprintf("lapic->id = %08x\n", lapic->id);
989 panic("double fault");
993 * syscall2 - MP aware system call request C handler
995 * A system call is essentially treated as a trap except that the
996 * MP lock is not held on entry or return. We are responsible for
997 * obtaining the MP lock if necessary and for handling ASTs
998 * (e.g. a task switch) prior to return.
1000 * In general, only simple access and manipulation of curproc and
1001 * the current stack is allowed without having to hold MP lock.
1003 * MPSAFE - note that large sections of this routine are run without
1007 syscall2(struct trapframe *frame)
1009 struct thread *td = curthread;
1010 struct proc *p = td->td_proc;
1011 struct lwp *lp = td->td_lwp;
1013 struct sysent *callp;
1014 register_t orig_tf_rflags;
1019 int crit_count = td->td_pri & ~TDPRI_MASK;
1022 int have_mplock = 0;
1027 union sysunion args;
1028 register_t *argsdst;
1030 mycpu->gd_cnt.v_syscall++;
1033 if (ISPL(frame->tf_cs) != SEL_UPL) {
1040 KTR_LOG(kernentry_syscall, p->p_pid, lp->lwp_tid,
1044 KASSERT(td->td_mpcount == 0, ("badmpcount syscall2 from %p", (void *)frame->tf_rip));
1045 if (syscall_mpsafe == 0)
1046 MAKEMPSAFE(have_mplock);
1048 userenter(td); /* lazy raise our priority */
1055 sticks = (int)td->td_sticks;
1056 orig_tf_rflags = frame->tf_rflags;
1059 * Virtual kernel intercept - if a VM context managed by a virtual
1060 * kernel issues a system call the virtual kernel handles it, not us.
1061 * Restore the virtual kernel context and return from its system
1062 * call. The current frame is copied out to the virtual kernel.
1064 if (lp->lwp_vkernel && lp->lwp_vkernel->ve) {
1065 error = vkernel_trap(lp, frame);
1066 frame->tf_rax = error;
1068 frame->tf_rflags |= PSL_C;
1069 error = EJUSTRETURN;
1074 * Get the system call parameters and account for time
1076 lp->lwp_md.md_regs = frame;
1077 params = (caddr_t)frame->tf_rsp + sizeof(register_t);
1078 code = frame->tf_rax;
1080 if (p->p_sysent->sv_prepsyscall) {
1081 (*p->p_sysent->sv_prepsyscall)(
1082 frame, (int *)(&args.nosys.sysmsg + 1),
1085 if (code == SYS_syscall || code == SYS___syscall) {
1086 code = frame->tf_rdi;
1092 if (p->p_sysent->sv_mask)
1093 code &= p->p_sysent->sv_mask;
1095 if (code >= p->p_sysent->sv_size)
1096 callp = &p->p_sysent->sv_table[0];
1098 callp = &p->p_sysent->sv_table[code];
1100 narg = callp->sy_narg & SYF_ARGMASK;
1103 * On amd64 we get up to six arguments in registers. The rest are
1104 * on the stack. The first six members of 'struct trapframe' happen
1105 * to be the registers used to pass arguments, in exactly the right
1108 argp = &frame->tf_rdi;
1110 argsdst = (register_t *)(&args.nosys.sysmsg + 1);
1112 * JG can we overflow the space pointed to by 'argsdst'
1113 * either with 'bcopy' or with 'copyin'?
1115 bcopy(argp, argsdst, sizeof(register_t) * regcnt);
1117 * copyin is MP aware, but the tracing code is not
1119 if (narg > regcnt) {
1120 KASSERT(params != NULL, ("copyin args with no params!"));
1121 error = copyin(params, &argsdst[regcnt],
1122 (narg - regcnt) * sizeof(register_t));
1125 if (KTRPOINT(td, KTR_SYSCALL)) {
1126 MAKEMPSAFE(have_mplock);
1128 ktrsyscall(lp, code, narg,
1129 (void *)(&args.nosys.sysmsg + 1));
1137 if (KTRPOINT(td, KTR_SYSCALL)) {
1138 MAKEMPSAFE(have_mplock);
1139 ktrsyscall(lp, code, narg, (void *)(&args.nosys.sysmsg + 1));
1144 * Default return value is 0 (will be copied to %rax). Double-value
1145 * returns use %rax and %rdx. %rdx is left unchanged for system
1146 * calls which return only one result.
1148 args.sysmsg_fds[0] = 0;
1149 args.sysmsg_fds[1] = frame->tf_rdx;
1152 * The syscall might manipulate the trap frame. If it does it
1153 * will probably return EJUSTRETURN.
1155 args.sysmsg_frame = frame;
1157 STOPEVENT(p, S_SCE, narg); /* MP aware */
1161 * Try to run the syscall without the MP lock if the syscall
1162 * is MP safe. We have to obtain the MP lock no matter what if
1165 if ((callp->sy_narg & SYF_MPSAFE) == 0)
1166 MAKEMPSAFE(have_mplock);
1169 error = (*callp->sy_call)(&args);
1173 * MP SAFE (we may or may not have the MP lock at this point)
1175 //kprintf("SYSMSG %d ", error);
1179 * Reinitialize proc pointer `p' as it may be different
1180 * if this is a child returning from fork syscall.
1183 lp = curthread->td_lwp;
1184 frame->tf_rax = args.sysmsg_fds[0];
1185 frame->tf_rdx = args.sysmsg_fds[1];
1186 frame->tf_rflags &= ~PSL_C;
1190 * Reconstruct pc, we know that 'syscall' is 2 bytes.
1191 * We have to do a full context restore so that %r10
1192 * (which was holding the value of %rcx) is restored for
1193 * the next iteration.
1195 frame->tf_rip -= frame->tf_err;
1196 frame->tf_r10 = frame->tf_rcx;
1201 panic("Unexpected EASYNC return value (for now)");
1204 if (p->p_sysent->sv_errsize) {
1205 if (error >= p->p_sysent->sv_errsize)
1206 error = -1; /* XXX */
1208 error = p->p_sysent->sv_errtbl[error];
1210 frame->tf_rax = error;
1211 frame->tf_rflags |= PSL_C;
1216 * Traced syscall. trapsignal() is not MP aware.
1218 if (orig_tf_rflags & PSL_T) {
1219 MAKEMPSAFE(have_mplock);
1220 frame->tf_rflags &= ~PSL_T;
1221 trapsignal(lp, SIGTRAP, 0);
1225 * Handle reschedule and other end-of-syscall issues
1227 userret(lp, frame, sticks);
1230 if (KTRPOINT(td, KTR_SYSRET)) {
1231 MAKEMPSAFE(have_mplock);
1232 ktrsysret(lp, code, error, args.sysmsg_result);
1237 * This works because errno is findable through the
1238 * register set. If we ever support an emulation where this
1239 * is not the case, this code will need to be revisited.
1241 STOPEVENT(p, S_SCX, code);
1246 * Release the MP lock if we had to get it
1248 KASSERT(td->td_mpcount == have_mplock,
1249 ("badmpcount syscall2/end from %p", (void *)frame->tf_rip));
1253 KTR_LOG(kernentry_syscall_ret, p->p_pid, lp->lwp_tid, error);
1255 KASSERT(crit_count == (td->td_pri & ~TDPRI_MASK),
1256 ("syscall: critical section count mismatch! %d/%d",
1257 crit_count / TDPRI_CRIT, td->td_pri / TDPRI_CRIT));
1262 fork_return(struct lwp *lp, struct trapframe *frame)
1264 frame->tf_rax = 0; /* Child returns zero */
1265 frame->tf_rflags &= ~PSL_C; /* success */
1268 generic_lwp_return(lp, frame);
1269 KTR_LOG(kernentry_fork_ret, lp->lwp_proc->p_pid, lp->lwp_tid);
1273 * Simplified back end of syscall(), used when returning from fork()
1274 * directly into user mode. MP lock is held on entry and should be
1275 * released on return. This code will return back into the fork
1276 * trampoline code which then runs doreti.
1279 generic_lwp_return(struct lwp *lp, struct trapframe *frame)
1281 struct proc *p = lp->lwp_proc;
1284 * Newly forked processes are given a kernel priority. We have to
1285 * adjust the priority to a normal user priority and fake entry
1286 * into the kernel (call userenter()) to install a passive release
1287 * function just in case userret() decides to stop the process. This
1288 * can occur when ^Z races a fork. If we do not install the passive
1289 * release function the current process designation will not be
1290 * released when the thread goes to sleep.
1292 lwkt_setpri_self(TDPRI_USER_NORM);
1293 userenter(lp->lwp_thread);
1294 userret(lp, frame, 0);
1296 if (KTRPOINT(lp->lwp_thread, KTR_SYSRET))
1297 ktrsysret(lp, SYS_fork, 0, 0);
1299 p->p_flag |= P_PASSIVE_ACQ;
1301 p->p_flag &= ~P_PASSIVE_ACQ;
1303 KKASSERT(lp->lwp_thread->td_mpcount == 1);
1309 * If PGEX_FPFAULT is set then set FP_VIRTFP in the PCB to force a T_DNA
1310 * fault (which is then passed back to the virtual kernel) if an attempt is
1311 * made to use the FP unit.
1313 * XXX this is a fairly big hack.
1316 set_vkernel_fp(struct trapframe *frame)