2 * SPDX-License-Identifier: BSD-4-Clause
4 * Copyright (C) 1994, David Greenman
5 * Copyright (c) 1990, 1993
6 * The Regents of the University of California. All rights reserved.
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
42 #include <sys/cdefs.h>
43 __FBSDID("$FreeBSD$");
46 * 386 Trap and System call handling
49 #include "opt_clock.h"
50 #include "opt_compat.h"
52 #include "opt_hwpmc_hooks.h"
55 #include "opt_stack.h"
58 #include <sys/param.h>
60 #include <sys/systm.h>
62 #include <sys/pioctl.h>
63 #include <sys/ptrace.h>
65 #include <sys/kernel.h>
68 #include <sys/mutex.h>
69 #include <sys/resourcevar.h>
70 #include <sys/signalvar.h>
71 #include <sys/syscall.h>
72 #include <sys/sysctl.h>
73 #include <sys/sysent.h>
75 #include <sys/vmmeter.h>
77 #include <sys/pmckern.h>
78 PMC_SOFT_DEFINE( , , page_fault, all);
79 PMC_SOFT_DEFINE( , , page_fault, read);
80 PMC_SOFT_DEFINE( , , page_fault, write);
82 #include <security/audit/audit.h>
85 #include <vm/vm_param.h>
87 #include <vm/vm_kern.h>
88 #include <vm/vm_map.h>
89 #include <vm/vm_page.h>
90 #include <vm/vm_extern.h>
92 #include <machine/cpu.h>
93 #include <machine/intr_machdep.h>
95 #include <machine/md_var.h>
96 #include <machine/pcb.h>
98 #include <machine/smp.h>
100 #include <machine/stack.h>
101 #include <machine/trap.h>
102 #include <machine/tss.h>
103 #include <machine/vm86.h>
106 #include <sys/syslog.h>
107 #include <machine/clock.h>
111 #include <sys/dtrace_bsd.h>
114 void trap(struct trapframe *frame);
115 void syscall(struct trapframe *frame);
117 static int trap_pfault(struct trapframe *, int, vm_offset_t);
118 static void trap_fatal(struct trapframe *, vm_offset_t);
119 void dblfault_handler(void);
121 extern inthand_t IDTVEC(bpt), IDTVEC(dbg), IDTVEC(int0x80_syscall);
123 #define MAX_TRAP_MSG 32
130 static const struct trap_data trap_data[] = {
131 [T_PRIVINFLT] = { .ei = true, .msg = "privileged instruction fault" },
132 [T_BPTFLT] = { .ei = false, .msg = "breakpoint instruction fault" },
133 [T_ARITHTRAP] = { .ei = true, .msg = "arithmetic trap" },
134 [T_PROTFLT] = { .ei = true, .msg = "general protection fault" },
135 [T_TRCTRAP] = { .ei = false, .msg = "debug exception" },
136 [T_PAGEFLT] = { .ei = true, .msg = "page fault" },
137 [T_ALIGNFLT] = { .ei = true, .msg = "alignment fault" },
138 [T_DIVIDE] = { .ei = true, .msg = "integer divide fault" },
139 [T_NMI] = { .ei = false, .msg = "non-maskable interrupt trap" },
140 [T_OFLOW] = { .ei = true, .msg = "overflow trap" },
141 [T_BOUND] = { .ei = true, .msg = "FPU bounds check fault" },
142 [T_DNA] = { .ei = true, .msg = "FPU device not available" },
143 [T_DOUBLEFLT] = { .ei = false, .msg = "double fault" },
144 [T_FPOPFLT] = { .ei = true, .msg = "FPU operand fetch fault" },
145 [T_TSSFLT] = { .ei = true, .msg = "invalid TSS fault" },
146 [T_SEGNPFLT] = { .ei = true, .msg = "segment not present fault" },
147 [T_STKFLT] = { .ei = true, .msg = "stack fault" },
148 [T_MCHK] = { .ei = true, .msg = "machine check trap" },
149 [T_XMMFLT] = { .ei = true, .msg = "SIMD floating-point exception" },
150 [T_DTRACE_RET] ={ .ei = true, .msg = "DTrace pid return trap" },
154 trap_enable_intr(int trapno)
158 if (trapno < nitems(trap_data) && trap_data[trapno].msg != NULL)
159 return (trap_data[trapno].ei);
167 static const char unkn[] = "UNKNOWN";
170 if (trapno < nitems(trap_data))
171 res = trap_data[trapno].msg;
177 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
178 int has_f00f_bug = 0; /* Initialized so that it can be patched. */
181 static int prot_fault_translation = 0;
182 SYSCTL_INT(_machdep, OID_AUTO, prot_fault_translation, CTLFLAG_RW,
183 &prot_fault_translation, 0, "Select signal to deliver on protection fault");
184 static int uprintf_signal;
185 SYSCTL_INT(_machdep, OID_AUTO, uprintf_signal, CTLFLAG_RW,
187 "Print debugging information on trap signal to ctty");
190 * Exception, fault, and trap interface to the FreeBSD kernel.
191 * This common code is called from assembly language IDT gate entry
192 * routines that prepare a suitable stack frame, and restore this
193 * frame after the exception has been processed.
197 trap(struct trapframe *frame)
204 register_t addr, dr6;
207 static int lastalert = 0;
218 type = frame->tf_trapno;
220 KASSERT((read_eflags() & PSL_I) == 0,
221 ("trap: interrupts enabled, type %d frame %p", type, frame));
224 /* Handler for NMI IPIs used for stopping CPUs. */
225 if (type == T_NMI && ipi_nmi_handler() == 0)
236 if (type == T_RESERVED) {
237 trap_fatal(frame, 0);
244 * CPU PMCs interrupt using an NMI so we check for that first.
245 * If the HWPMC module is active, 'pmc_hook' will point to
246 * the function to be called. A non-zero return value from the
247 * hook means that the NMI was consumed by it and that we can
248 * return immediately.
250 if (pmc_intr != NULL &&
251 (*pmc_intr)(frame) != 0)
256 if (stack_nmi_handler(frame) != 0)
261 if (type == T_MCHK) {
268 * A trap can occur while DTrace executes a probe. Before
269 * executing the probe, DTrace blocks re-scheduling and sets
270 * a flag in its per-cpu flags to indicate that it doesn't
271 * want to fault. On returning from the probe, the no-fault
272 * flag is cleared and finally re-scheduling is enabled.
274 if ((type == T_PROTFLT || type == T_PAGEFLT) &&
275 dtrace_trap_func != NULL && (*dtrace_trap_func)(frame, type))
280 * We must not allow context switches until %cr2 is read.
281 * Also, for some Cyrix CPUs, %cr2 is clobbered by interrupts.
282 * All faults use interrupt gates, so %cr2 can be safely read
283 * now, before optional enable of the interrupts below.
285 if (type == T_PAGEFLT)
289 * Buggy application or kernel code has disabled interrupts
290 * and then trapped. Enabling interrupts now is wrong, but it
291 * is better than running with interrupts disabled until they
292 * are accidentally enabled later.
294 if ((frame->tf_eflags & PSL_I) == 0 && TRAPF_USERMODE(frame) &&
295 (curpcb->pcb_flags & PCB_VM86CALL) == 0)
296 uprintf("pid %ld (%s): trap %d with interrupts disabled\n",
297 (long)curproc->p_pid, curthread->td_name, type);
300 * Conditionally reenable interrupts. If we hold a spin lock,
301 * then we must not reenable interrupts. This might be a
302 * spurious page fault.
304 if (trap_enable_intr(type) && td->td_md.md_spinlock_count == 0 &&
305 frame->tf_eip != (int)cpu_switch_load_gs)
308 if (TRAPF_USERMODE(frame) && (curpcb->pcb_flags & PCB_VM86CALL) == 0) {
312 td->td_frame = frame;
313 addr = frame->tf_eip;
314 if (td->td_cowgen != p->p_cowgen)
315 thread_cow_update(td);
318 case T_PRIVINFLT: /* privileged instruction fault */
323 case T_BPTFLT: /* bpt instruction fault */
326 if (dtrace_pid_probe_ptr != NULL &&
327 dtrace_pid_probe_ptr(frame) == 0)
334 case T_TRCTRAP: /* debug exception */
340 if ((dr6 & DBREG_DR6_BS) != 0) {
341 PROC_LOCK(td->td_proc);
342 if ((td->td_dbgflags & TDB_STEP) != 0) {
343 td->td_frame->tf_eflags &= ~PSL_T;
344 td->td_dbgflags &= ~TDB_STEP;
346 PROC_UNLOCK(td->td_proc);
350 case T_ARITHTRAP: /* arithmetic trap */
351 ucode = npxtrap_x87();
358 * The following two traps can happen in vm86 mode,
359 * and, if so, we want to handle them specially.
361 case T_PROTFLT: /* general protection fault */
362 case T_STKFLT: /* stack fault */
363 if (frame->tf_eflags & PSL_VM) {
364 signo = vm86_emulate((struct vm86frame *)frame);
365 if (signo == SIGTRAP) {
366 load_dr6(rdr6() | 0x4000);
367 goto user_trctrap_out;
374 ucode = (type == T_PROTFLT) ? BUS_OBJERR : BUS_ADRERR;
376 case T_SEGNPFLT: /* segment not present fault */
380 case T_TSSFLT: /* invalid TSS fault */
388 case T_DOUBLEFLT: /* double fault */
394 case T_PAGEFLT: /* page fault */
395 signo = trap_pfault(frame, TRUE, eva);
396 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
399 * The f00f hack workaround has triggered, so
400 * treat the fault as an illegal instruction
401 * (T_PRIVINFLT) instead of a page fault.
403 type = frame->tf_trapno = T_PRIVINFLT;
405 /* Proceed as in that case. */
416 if (signo == SIGSEGV)
418 else if (prot_fault_translation == 0) {
420 * Autodetect. This check also covers
421 * the images without the ABI-tag ELF
424 if (SV_CURPROC_ABI() == SV_ABI_FREEBSD &&
425 p->p_osrel >= P_OSREL_SIGSEGV) {
432 } else if (prot_fault_translation == 1) {
434 * Always compat mode.
440 * Always SIGSEGV mode.
448 case T_DIVIDE: /* integer divide fault */
457 # define TIMER_FREQ 1193182
459 if (time_second - lastalert > 10) {
460 log(LOG_WARNING, "NMI: power fail\n");
462 lastalert = time_second;
465 #else /* !POWERFAIL_NMI */
466 nmi_handle_intr(type, frame);
468 #endif /* POWERFAIL_NMI */
471 case T_OFLOW: /* integer overflow fault */
476 case T_BOUND: /* bounds check fault */
482 KASSERT(PCB_USER_FPU(td->td_pcb),
483 ("kernel FPU ctx has leaked"));
484 /* transparent fault (due to context switch "late") */
487 uprintf("pid %d killed due to lack of floating point\n",
493 case T_FPOPFLT: /* FPU operand fetch fault */
498 case T_XMMFLT: /* SIMD floating-point exception */
499 ucode = npxtrap_sse();
507 if (dtrace_return_probe_ptr != NULL)
508 dtrace_return_probe_ptr(frame);
515 KASSERT(cold || td->td_ucred != NULL,
516 ("kernel trap doesn't have ucred"));
518 case T_PAGEFLT: /* page fault */
519 (void) trap_pfault(frame, FALSE, eva);
523 if (PCB_USER_FPU(td->td_pcb))
524 panic("Unregistered use of FPU in kernel");
529 case T_ARITHTRAP: /* arithmetic trap */
530 case T_XMMFLT: /* SIMD floating-point exception */
531 case T_FPOPFLT: /* FPU operand fetch fault */
533 * XXXKIB for now disable any FPU traps in kernel
534 * handler registration seems to be overkill
536 trap_fatal(frame, 0);
540 * The following two traps can happen in
541 * vm86 mode, and, if so, we want to handle
544 case T_PROTFLT: /* general protection fault */
545 case T_STKFLT: /* stack fault */
546 if (frame->tf_eflags & PSL_VM) {
547 signo = vm86_emulate((struct vm86frame *)frame);
548 if (signo == SIGTRAP) {
550 load_dr6(rdr6() | 0x4000);
555 * returns to original process
557 vm86_trap((struct vm86frame *)frame);
561 case T_SEGNPFLT: /* segment not present fault */
562 if (curpcb->pcb_flags & PCB_VM86CALL)
566 * Invalid %fs's and %gs's can be created using
567 * procfs or PT_SETREGS or by invalidating the
568 * underlying LDT entry. This causes a fault
569 * in kernel mode when the kernel attempts to
570 * switch contexts. Lose the bad context
571 * (XXX) so that we can continue, and generate
574 if (frame->tf_eip == (int)cpu_switch_load_gs) {
578 kern_psignal(p, SIGBUS);
584 if (td->td_intr_nesting_level != 0)
588 * Invalid segment selectors and out of bounds
589 * %eip's and %esp's can be set up in user mode.
590 * This causes a fault in kernel mode when the
591 * kernel tries to return to user mode. We want
592 * to get this fault so that we can fix the
593 * problem here and not have to check all the
594 * selectors and pointers when the user changes
597 * N.B. Comparing to long mode, 32-bit mode
598 * does not push %esp on the trap frame,
599 * because iretl faulted while in ring 0. As
600 * the consequence, there is no need to fixup
601 * the stack pointer for doreti_iret_fault,
602 * the fixup and the complimentary trap() call
603 * are executed on the main thread stack, not
604 * on the trampoline stack.
606 if (frame->tf_eip == (int)doreti_iret + setidt_disp) {
607 frame->tf_eip = (int)doreti_iret_fault +
611 if (type == T_STKFLT)
614 if (frame->tf_eip == (int)doreti_popl_ds +
616 frame->tf_eip = (int)doreti_popl_ds_fault +
620 if (frame->tf_eip == (int)doreti_popl_es +
622 frame->tf_eip = (int)doreti_popl_es_fault +
626 if (frame->tf_eip == (int)doreti_popl_fs +
628 frame->tf_eip = (int)doreti_popl_fs_fault +
632 if (curpcb->pcb_onfault != NULL) {
633 frame->tf_eip = (int)curpcb->pcb_onfault;
640 * PSL_NT can be set in user mode and isn't cleared
641 * automatically when the kernel is entered. This
642 * causes a TSS fault when the kernel attempts to
643 * `iret' because the TSS link is uninitialized. We
644 * want to get this fault so that we can fix the
645 * problem here and not every time the kernel is
648 if (frame->tf_eflags & PSL_NT) {
649 frame->tf_eflags &= ~PSL_NT;
654 case T_TRCTRAP: /* debug exception */
656 /* Clear any pending debug events. */
661 * Ignore debug register exceptions due to
662 * accesses in the user's address space, which
663 * can happen under several conditions such as
664 * if a user sets a watchpoint on a buffer and
665 * then passes that buffer to a system call.
666 * We still want to get TRCTRAPS for addresses
667 * in kernel space because that is useful when
668 * debugging the kernel.
670 if (user_dbreg_trap(dr6) &&
671 !(curpcb->pcb_flags & PCB_VM86CALL))
675 * Malicious user code can configure a debug
676 * register watchpoint to trap on data access
677 * to the top of stack and then execute 'pop
678 * %ss; int 3'. Due to exception deferral for
679 * 'pop %ss', the CPU will not interrupt 'int
680 * 3' to raise the DB# exception for the debug
681 * register but will postpone the DB# until
682 * execution of the first instruction of the
683 * BP# handler (in kernel mode). Normally the
684 * previous check would ignore DB# exceptions
685 * for watchpoints on user addresses raised in
686 * kernel mode. However, some CPU errata
687 * include cases where DB# exceptions do not
688 * properly set bits in %dr6, e.g. Haswell
689 * HSD23 and Skylake-X SKZ24.
691 * A deferred DB# can also be raised on the
692 * first instructions of system call entry
693 * points or single-step traps via similar use
694 * of 'pop %ss' or 'mov xxx, %ss'.
697 (uintptr_t)IDTVEC(int0x80_syscall) + setidt_disp ||
698 frame->tf_eip == (uintptr_t)IDTVEC(bpt) +
700 frame->tf_eip == (uintptr_t)IDTVEC(dbg) +
704 * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
708 * If KDB is enabled, let it handle the debugger trap.
709 * Otherwise, debugger traps "can't happen".
712 if (kdb_trap(type, dr6, frame))
720 if (time_second - lastalert > 10) {
721 log(LOG_WARNING, "NMI: power fail\n");
723 lastalert = time_second;
726 #else /* !POWERFAIL_NMI */
727 nmi_handle_intr(type, frame);
729 #endif /* POWERFAIL_NMI */
733 trap_fatal(frame, eva);
737 /* Translate fault for emulators (e.g. Linux) */
738 if (*p->p_sysent->sv_transtrap != NULL)
739 signo = (*p->p_sysent->sv_transtrap)(signo, type);
741 ksiginfo_init_trap(&ksi);
742 ksi.ksi_signo = signo;
743 ksi.ksi_code = ucode;
744 ksi.ksi_addr = (void *)addr;
745 ksi.ksi_trapno = type;
746 if (uprintf_signal) {
747 uprintf("pid %d comm %s: signal %d err %x code %d type %d "
748 "addr 0x%x ss 0x%04x esp 0x%08x cs 0x%04x eip 0x%08x "
749 "<%02x %02x %02x %02x %02x %02x %02x %02x>\n",
750 p->p_pid, p->p_comm, signo, frame->tf_err, ucode, type,
751 addr, frame->tf_ss, frame->tf_esp, frame->tf_cs,
753 fubyte((void *)(frame->tf_eip + 0)),
754 fubyte((void *)(frame->tf_eip + 1)),
755 fubyte((void *)(frame->tf_eip + 2)),
756 fubyte((void *)(frame->tf_eip + 3)),
757 fubyte((void *)(frame->tf_eip + 4)),
758 fubyte((void *)(frame->tf_eip + 5)),
759 fubyte((void *)(frame->tf_eip + 6)),
760 fubyte((void *)(frame->tf_eip + 7)));
762 KASSERT((read_eflags() & PSL_I) != 0, ("interrupts disabled"));
763 trapsignal(td, &ksi);
767 KASSERT(PCB_USER_FPU(td->td_pcb),
768 ("Return from trap with kernel FPU ctx leaked"));
772 trap_pfault(struct trapframe *frame, int usermode, vm_offset_t eva)
784 if (__predict_false((td->td_pflags & TDP_NOFAULTING) != 0)) {
786 * Due to both processor errata and lazy TLB invalidation when
787 * access restrictions are removed from virtual pages, memory
788 * accesses that are allowed by the physical mapping layer may
789 * nonetheless cause one spurious page fault per virtual page.
790 * When the thread is executing a "no faulting" section that
791 * is bracketed by vm_fault_{disable,enable}_pagefaults(),
792 * every page fault is treated as a spurious page fault,
793 * unless it accesses the same virtual address as the most
794 * recent page fault within the same "no faulting" section.
796 if (td->td_md.md_spurflt_addr != eva ||
797 (td->td_pflags & TDP_RESETSPUR) != 0) {
799 * Do nothing to the TLB. A stale TLB entry is
800 * flushed automatically by a page fault.
802 td->td_md.md_spurflt_addr = eva;
803 td->td_pflags &= ~TDP_RESETSPUR;
808 * If we get a page fault while in a critical section, then
809 * it is most likely a fatal kernel page fault. The kernel
810 * is already going to panic trying to get a sleep lock to
811 * do the VM lookup, so just consider it a fatal trap so the
812 * kernel can print out a useful trap message and even get
815 * If we get a page fault while holding a non-sleepable
816 * lock, then it is most likely a fatal kernel page fault.
817 * If WITNESS is enabled, then it's going to whine about
818 * bogus LORs with various VM locks, so just skip to the
819 * fatal trap handling directly.
821 if (td->td_critnest != 0 ||
822 WITNESS_CHECK(WARN_SLEEPOK | WARN_GIANTOK, NULL,
823 "Kernel page fault") != 0) {
824 trap_fatal(frame, eva);
828 va = trunc_page(eva);
829 if (va >= PMAP_TRM_MIN_ADDRESS) {
831 * Don't allow user-mode faults in kernel address space.
832 * An exception: if the faulting address is the invalid
833 * instruction entry in the IDT, then the Intel Pentium
834 * F00F bug workaround was triggered, and we need to
835 * treat it is as an illegal instruction, and not a page
838 #if defined(I586_CPU) && !defined(NO_F00F_HACK)
839 if ((eva == (unsigned int)&idt[6]) && has_f00f_bug)
844 trap_fatal(frame, eva);
847 map = usermode ? &p->p_vmspace->vm_map : kernel_map;
850 * Kernel cannot access a user-space address directly
851 * because user pages are not mapped. Also, page
852 * faults must not be caused during the interrupts.
854 if (!usermode && td->td_intr_nesting_level != 0) {
855 trap_fatal(frame, eva);
861 * If the trap was caused by errant bits in the PTE then panic.
863 if (frame->tf_err & PGEX_RSV) {
864 trap_fatal(frame, eva);
869 * PGEX_I is defined only if the execute disable bit capability is
870 * supported and enabled.
872 if (frame->tf_err & PGEX_W)
873 ftype = VM_PROT_WRITE;
874 #if defined(PAE) || defined(PAE_TABLES)
875 else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
876 ftype = VM_PROT_EXECUTE;
879 ftype = VM_PROT_READ;
881 /* Fault in the page. */
882 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
883 if (rv == KERN_SUCCESS) {
885 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
886 PMC_SOFT_CALL_TF( , , page_fault, all, frame);
887 if (ftype == VM_PROT_READ)
888 PMC_SOFT_CALL_TF( , , page_fault, read,
891 PMC_SOFT_CALL_TF( , , page_fault, write,
898 if (td->td_intr_nesting_level == 0 &&
899 curpcb->pcb_onfault != NULL) {
900 frame->tf_eip = (int)curpcb->pcb_onfault;
903 trap_fatal(frame, eva);
906 return ((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
910 trap_fatal(frame, eva)
911 struct trapframe *frame;
916 struct soft_segment_descriptor softseg;
921 code = frame->tf_err;
922 type = frame->tf_trapno;
923 sdtossd(&gdt[IDXSEL(frame->tf_cs & 0xffff)].sd, &softseg);
925 printf("\n\nFatal trap %d: %s while in %s mode\n", type, trap_msg(type),
926 frame->tf_eflags & PSL_VM ? "vm86" :
927 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
929 /* two separate prints in case of a trap on an unmapped page */
930 printf("cpuid = %d; ", PCPU_GET(cpuid));
931 printf("apic id = %02x\n", PCPU_GET(apic_id));
933 if (type == T_PAGEFLT) {
934 printf("fault virtual address = 0x%x\n", eva);
935 printf("fault code = %s %s%s, %s\n",
936 code & PGEX_U ? "user" : "supervisor",
937 code & PGEX_W ? "write" : "read",
938 #if defined(PAE) || defined(PAE_TABLES)
940 (code & PGEX_I ? " instruction" : " data") :
943 code & PGEX_RSV ? "reserved bits in PTE" :
944 code & PGEX_P ? "protection violation" : "page not present");
946 printf("error code = %#x\n", code);
948 printf("instruction pointer = 0x%x:0x%x\n",
949 frame->tf_cs & 0xffff, frame->tf_eip);
950 if (TF_HAS_STACKREGS(frame)) {
951 ss = frame->tf_ss & 0xffff;
954 ss = GSEL(GDATA_SEL, SEL_KPL);
955 esp = (int)&frame->tf_esp;
957 printf("stack pointer = 0x%x:0x%x\n", ss, esp);
958 printf("frame pointer = 0x%x:0x%x\n", ss, frame->tf_ebp);
959 printf("code segment = base 0x%x, limit 0x%x, type 0x%x\n",
960 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
961 printf(" = DPL %d, pres %d, def32 %d, gran %d\n",
962 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_def32,
964 printf("processor eflags = ");
965 if (frame->tf_eflags & PSL_T)
966 printf("trace trap, ");
967 if (frame->tf_eflags & PSL_I)
968 printf("interrupt enabled, ");
969 if (frame->tf_eflags & PSL_NT)
970 printf("nested task, ");
971 if (frame->tf_eflags & PSL_RF)
973 if (frame->tf_eflags & PSL_VM)
975 printf("IOPL = %d\n", (frame->tf_eflags & PSL_IOPL) >> 12);
976 printf("current process = %d (%s)\n",
977 curproc->p_pid, curthread->td_name);
980 if (debugger_on_trap) {
981 kdb_why = KDB_WHY_TRAP;
982 frame->tf_err = eva; /* smuggle fault address to ddb */
983 handled = kdb_trap(type, 0, frame);
984 frame->tf_err = code; /* restore error code */
985 kdb_why = KDB_WHY_UNSET;
990 printf("trap number = %d\n", type);
991 if (trap_msg(type) != NULL)
992 panic("%s", trap_msg(type));
994 panic("unknown/reserved trap");
998 * Double fault handler. Called when a fault occurs while writing
999 * a frame for a trap/exception onto the stack. This usually occurs
1000 * when the stack overflows (such is the case with infinite recursion,
1003 * XXX Note that the current PTD gets replaced by IdlePTD when the
1004 * task switch occurs. This means that the stack that was active at
1005 * the time of the double fault is not available at <kstack> unless
1006 * the machine was idle when the double fault occurred. The downside
1007 * of this is that "trace <ebp>" in ddb won't work.
1010 dblfault_handler(void)
1012 #ifdef KDTRACE_HOOKS
1013 if (dtrace_doubletrap_func != NULL)
1014 (*dtrace_doubletrap_func)();
1016 printf("\nFatal double fault:\n");
1017 printf("eip = 0x%x\n", PCPU_GET(common_tssp)->tss_eip);
1018 printf("esp = 0x%x\n", PCPU_GET(common_tssp)->tss_esp);
1019 printf("ebp = 0x%x\n", PCPU_GET(common_tssp)->tss_ebp);
1021 /* two separate prints in case of a trap on an unmapped page */
1022 printf("cpuid = %d; ", PCPU_GET(cpuid));
1023 printf("apic id = %02x\n", PCPU_GET(apic_id));
1025 panic("double fault");
1029 cpu_fetch_syscall_args(struct thread *td)
1032 struct trapframe *frame;
1033 struct syscall_args *sa;
1043 frame = td->td_frame;
1047 if (__predict_false(frame->tf_cs == 7 && frame->tf_eip == 2)) {
1049 * In lcall $7,$0 after int $0x80. Convert the user
1050 * frame to what it would be for a direct int 0x80 instead
1051 * of lcall $7,$0, by popping the lcall return address.
1053 error = fueword32((void *)frame->tf_esp, &eip);
1056 cs = fuword16((void *)(frame->tf_esp + sizeof(u_int32_t)));
1061 * Unwind in-kernel frame after all stack frame pieces
1062 * were successfully read.
1064 frame->tf_eip = eip;
1066 frame->tf_esp += 2 * sizeof(u_int32_t);
1067 frame->tf_err = 7; /* size of lcall $7,$0 */
1071 sa->code = frame->tf_eax;
1072 params = (caddr_t)frame->tf_esp + sizeof(uint32_t);
1075 * Need to check if this is a 32 bit or 64 bit syscall.
1077 if (sa->code == SYS_syscall) {
1079 * Code is first argument, followed by actual args.
1081 error = fueword(params, &tmp);
1085 params += sizeof(uint32_t);
1086 } else if (sa->code == SYS___syscall) {
1088 * Like syscall, but code is a quad, so as to maintain
1089 * quad alignment for the rest of the arguments.
1091 error = fueword(params, &tmp);
1095 params += sizeof(quad_t);
1098 if (p->p_sysent->sv_mask)
1099 sa->code &= p->p_sysent->sv_mask;
1100 if (sa->code >= p->p_sysent->sv_size)
1101 sa->callp = &p->p_sysent->sv_table[0];
1103 sa->callp = &p->p_sysent->sv_table[sa->code];
1104 sa->narg = sa->callp->sy_narg;
1106 if (params != NULL && sa->narg != 0)
1107 error = copyin(params, (caddr_t)sa->args,
1108 (u_int)(sa->narg * sizeof(uint32_t)));
1113 td->td_retval[0] = 0;
1114 td->td_retval[1] = frame->tf_edx;
1120 #include "../../kern/subr_syscall.c"
1123 * syscall - system call request C handler. A system call is
1124 * essentially treated as a trap by reusing the frame layout.
1127 syscall(struct trapframe *frame)
1130 register_t orig_tf_eflags;
1135 if (!(TRAPF_USERMODE(frame) &&
1136 (curpcb->pcb_flags & PCB_VM86CALL) == 0)) {
1141 orig_tf_eflags = frame->tf_eflags;
1144 td->td_frame = frame;
1146 error = syscallenter(td);
1151 if ((orig_tf_eflags & PSL_T) && !(orig_tf_eflags & PSL_VM)) {
1152 frame->tf_eflags &= ~PSL_T;
1153 ksiginfo_init_trap(&ksi);
1154 ksi.ksi_signo = SIGTRAP;
1155 ksi.ksi_code = TRAP_TRACE;
1156 ksi.ksi_addr = (void *)frame->tf_eip;
1157 trapsignal(td, &ksi);
1160 KASSERT(PCB_USER_FPU(td->td_pcb),
1161 ("System call %s returning with kernel FPU ctx leaked",
1162 syscallname(td->td_proc, td->td_sa.code)));
1163 KASSERT(td->td_pcb->pcb_save == get_pcb_user_save_td(td),
1164 ("System call %s returning with mangled pcb_save",
1165 syscallname(td->td_proc, td->td_sa.code)));
1167 syscallret(td, error);