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
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
44 * AMD64 Trap and System call handling
47 #include "opt_clock.h"
49 #include "opt_hwpmc_hooks.h"
52 #include "opt_kdtrace.h"
53 #include "opt_ktrace.h"
55 #include <sys/param.h>
57 #include <sys/systm.h>
59 #include <sys/pioctl.h>
60 #include <sys/ptrace.h>
62 #include <sys/kernel.h>
65 #include <sys/mutex.h>
66 #include <sys/resourcevar.h>
67 #include <sys/signalvar.h>
68 #include <sys/syscall.h>
69 #include <sys/sysctl.h>
70 #include <sys/sysent.h>
72 #include <sys/vmmeter.h>
74 #include <sys/ktrace.h>
77 #include <sys/pmckern.h>
81 #include <vm/vm_param.h>
83 #include <vm/vm_kern.h>
84 #include <vm/vm_map.h>
85 #include <vm/vm_page.h>
86 #include <vm/vm_extern.h>
88 #include <machine/cpu.h>
89 #include <machine/intr_machdep.h>
90 #include <machine/mca.h>
91 #include <machine/md_var.h>
92 #include <machine/pcb.h>
94 #include <machine/smp.h>
96 #include <machine/tss.h>
99 #include <sys/dtrace_bsd.h>
102 * This is a hook which is initialised by the dtrace module
103 * to handle traps which might occur during DTrace probe
106 dtrace_trap_func_t dtrace_trap_func;
108 dtrace_doubletrap_func_t dtrace_doubletrap_func;
111 * This is a hook which is initialised by the systrace module
112 * when it is loaded. This keeps the DTrace syscall provider
113 * implementation opaque.
115 systrace_probe_func_t systrace_probe_func;
118 extern void trap(struct trapframe *frame);
119 extern void syscall(struct trapframe *frame);
120 void dblfault_handler(struct trapframe *frame);
122 static int trap_pfault(struct trapframe *, int);
123 static void trap_fatal(struct trapframe *, vm_offset_t);
125 #define MAX_TRAP_MSG 30
126 static char *trap_msg[] = {
128 "privileged instruction fault", /* 1 T_PRIVINFLT */
130 "breakpoint instruction fault", /* 3 T_BPTFLT */
133 "arithmetic trap", /* 6 T_ARITHTRAP */
136 "general protection fault", /* 9 T_PROTFLT */
137 "trace trap", /* 10 T_TRCTRAP */
139 "page fault", /* 12 T_PAGEFLT */
141 "alignment fault", /* 14 T_ALIGNFLT */
145 "integer divide fault", /* 18 T_DIVIDE */
146 "non-maskable interrupt trap", /* 19 T_NMI */
147 "overflow trap", /* 20 T_OFLOW */
148 "FPU bounds check fault", /* 21 T_BOUND */
149 "FPU device not available", /* 22 T_DNA */
150 "double fault", /* 23 T_DOUBLEFLT */
151 "FPU operand fetch fault", /* 24 T_FPOPFLT */
152 "invalid TSS fault", /* 25 T_TSSFLT */
153 "segment not present fault", /* 26 T_SEGNPFLT */
154 "stack fault", /* 27 T_STKFLT */
155 "machine check trap", /* 28 T_MCHK */
156 "SIMD floating-point exception", /* 29 T_XMMFLT */
157 "reserved (unknown) fault", /* 30 T_RESERVED */
161 static int kdb_on_nmi = 1;
162 SYSCTL_INT(_machdep, OID_AUTO, kdb_on_nmi, CTLFLAG_RW,
163 &kdb_on_nmi, 0, "Go to KDB on NMI");
165 static int panic_on_nmi = 1;
166 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
167 &panic_on_nmi, 0, "Panic on NMI");
168 static int prot_fault_translation = 0;
169 SYSCTL_INT(_machdep, OID_AUTO, prot_fault_translation, CTLFLAG_RW,
170 &prot_fault_translation, 0, "Select signal to deliver on protection fault");
173 * Exception, fault, and trap interface to the FreeBSD kernel.
174 * This common code is called from assembly language IDT gate entry
175 * routines that prepare a suitable stack frame, and restore this
176 * frame after the exception has been processed.
180 trap(struct trapframe *frame)
182 struct thread *td = curthread;
183 struct proc *p = td->td_proc;
184 int i = 0, ucode = 0, code;
189 PCPU_INC(cnt.v_trap);
190 type = frame->tf_trapno;
193 /* Handler for NMI IPIs used for stopping CPUs. */
195 if (ipi_nmi_handler() == 0)
207 if (type == T_RESERVED) {
208 trap_fatal(frame, 0);
214 * CPU PMCs interrupt using an NMI. If the PMC module is
215 * active, pass the 'rip' value to the PMC module's interrupt
216 * handler. A return value of '1' from the handler means that
217 * the NMI was handled by it and we can return immediately.
219 if (type == T_NMI && pmc_intr &&
220 (*pmc_intr)(PCPU_GET(cpuid), frame))
224 if (type == T_MCHK) {
226 trap_fatal(frame, 0);
232 * A trap can occur while DTrace executes a probe. Before
233 * executing the probe, DTrace blocks re-scheduling and sets
234 * a flag in it's per-cpu flags to indicate that it doesn't
235 * want to fault. On returning from the the probe, the no-fault
236 * flag is cleared and finally re-scheduling is enabled.
238 * If the DTrace kernel module has registered a trap handler,
239 * call it and if it returns non-zero, assume that it has
240 * handled the trap and modified the trap frame so that this
241 * function can return normally.
243 if (dtrace_trap_func != NULL)
244 if ((*dtrace_trap_func)(frame, type))
248 if ((frame->tf_rflags & PSL_I) == 0) {
250 * Buggy application or kernel code has disabled
251 * interrupts and then trapped. Enabling interrupts
252 * now is wrong, but it is better than running with
253 * interrupts disabled until they are accidentally
256 if (ISPL(frame->tf_cs) == SEL_UPL)
258 "pid %ld (%s): trap %d with interrupts disabled\n",
259 (long)curproc->p_pid, curthread->td_name, type);
260 else if (type != T_NMI && type != T_BPTFLT &&
263 * XXX not quite right, since this may be for a
264 * multiple fault in user mode.
266 printf("kernel trap %d with interrupts disabled\n",
270 * We shouldn't enable interrupts while holding a
273 if (td->td_md.md_spinlock_count == 0)
278 code = frame->tf_err;
279 if (type == T_PAGEFLT) {
281 * If we get a page fault while in a critical section, then
282 * it is most likely a fatal kernel page fault. The kernel
283 * is already going to panic trying to get a sleep lock to
284 * do the VM lookup, so just consider it a fatal trap so the
285 * kernel can print out a useful trap message and even get
288 * If we get a page fault while holding a non-sleepable
289 * lock, then it is most likely a fatal kernel page fault.
290 * If WITNESS is enabled, then it's going to whine about
291 * bogus LORs with various VM locks, so just skip to the
292 * fatal trap handling directly.
294 if (td->td_critnest != 0 ||
295 WITNESS_CHECK(WARN_SLEEPOK | WARN_GIANTOK, NULL,
296 "Kernel page fault") != 0)
297 trap_fatal(frame, frame->tf_addr);
300 if (ISPL(frame->tf_cs) == SEL_UPL) {
304 td->td_frame = frame;
305 addr = frame->tf_rip;
306 if (td->td_ucred != p->p_ucred)
307 cred_update_thread(td);
310 case T_PRIVINFLT: /* privileged instruction fault */
315 case T_BPTFLT: /* bpt instruction fault */
316 case T_TRCTRAP: /* trace trap */
318 frame->tf_rflags &= ~PSL_T;
320 ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
323 case T_ARITHTRAP: /* arithmetic trap */
330 case T_PROTFLT: /* general protection fault */
334 case T_STKFLT: /* stack fault */
335 case T_SEGNPFLT: /* segment not present fault */
339 case T_TSSFLT: /* invalid TSS fault */
343 case T_DOUBLEFLT: /* double fault */
349 case T_PAGEFLT: /* page fault */
350 addr = frame->tf_addr;
351 i = trap_pfault(frame, TRUE);
360 if (prot_fault_translation == 0) {
363 * This check also covers the images
364 * without the ABI-tag ELF note.
366 if (SV_CURPROC_ABI() == SV_ABI_FREEBSD
367 && p->p_osrel >= P_OSREL_SIGSEGV) {
372 ucode = BUS_PAGE_FAULT;
374 } else if (prot_fault_translation == 1) {
376 * Always compat mode.
379 ucode = BUS_PAGE_FAULT;
382 * Always SIGSEGV mode.
390 case T_DIVIDE: /* integer divide fault */
397 /* machine/parity/power fail/"kitchen sink" faults */
398 if (isa_nmi(code) == 0) {
401 * NMI can be hooked up to a pushbutton
405 printf ("NMI ... going to debugger\n");
406 kdb_trap(type, 0, frame);
410 } else if (panic_on_nmi)
411 panic("NMI indicates hardware failure");
415 case T_OFLOW: /* integer overflow fault */
420 case T_BOUND: /* bounds check fault */
426 /* transparent fault (due to context switch "late") */
427 KASSERT(PCB_USER_FPU(td->td_pcb),
428 ("kernel FPU ctx has leaked"));
432 case T_FPOPFLT: /* FPU operand fetch fault */
437 case T_XMMFLT: /* SIMD floating-point exception */
445 KASSERT(cold || td->td_ucred != NULL,
446 ("kernel trap doesn't have ucred"));
448 case T_PAGEFLT: /* page fault */
449 (void) trap_pfault(frame, FALSE);
453 KASSERT(!PCB_USER_FPU(td->td_pcb),
454 ("Unregistered use of FPU in kernel"));
458 case T_ARITHTRAP: /* arithmetic trap */
459 case T_XMMFLT: /* SIMD floating-point exception */
460 case T_FPOPFLT: /* FPU operand fetch fault */
462 * XXXKIB for now disable any FPU traps in kernel
463 * handler registration seems to be overkill
465 trap_fatal(frame, 0);
468 case T_STKFLT: /* stack fault */
471 case T_PROTFLT: /* general protection fault */
472 case T_SEGNPFLT: /* segment not present fault */
473 if (td->td_intr_nesting_level != 0)
477 * Invalid segment selectors and out of bounds
478 * %rip's and %rsp's can be set up in user mode.
479 * This causes a fault in kernel mode when the
480 * kernel tries to return to user mode. We want
481 * to get this fault so that we can fix the
482 * problem here and not have to check all the
483 * selectors and pointers when the user changes
486 if (frame->tf_rip == (long)doreti_iret) {
487 frame->tf_rip = (long)doreti_iret_fault;
490 if (frame->tf_rip == (long)ld_ds) {
491 frame->tf_rip = (long)ds_load_fault;
494 if (frame->tf_rip == (long)ld_es) {
495 frame->tf_rip = (long)es_load_fault;
498 if (frame->tf_rip == (long)ld_fs) {
499 frame->tf_rip = (long)fs_load_fault;
502 if (frame->tf_rip == (long)ld_gs) {
503 frame->tf_rip = (long)gs_load_fault;
506 if (frame->tf_rip == (long)ld_gsbase) {
507 frame->tf_rip = (long)gsbase_load_fault;
510 if (frame->tf_rip == (long)ld_fsbase) {
511 frame->tf_rip = (long)fsbase_load_fault;
514 if (PCPU_GET(curpcb)->pcb_onfault != NULL) {
516 (long)PCPU_GET(curpcb)->pcb_onfault;
523 * PSL_NT can be set in user mode and isn't cleared
524 * automatically when the kernel is entered. This
525 * causes a TSS fault when the kernel attempts to
526 * `iret' because the TSS link is uninitialized. We
527 * want to get this fault so that we can fix the
528 * problem here and not every time the kernel is
531 if (frame->tf_rflags & PSL_NT) {
532 frame->tf_rflags &= ~PSL_NT;
537 case T_TRCTRAP: /* trace trap */
539 * Ignore debug register trace traps due to
540 * accesses in the user's address space, which
541 * can happen under several conditions such as
542 * if a user sets a watchpoint on a buffer and
543 * then passes that buffer to a system call.
544 * We still want to get TRCTRAPS for addresses
545 * in kernel space because that is useful when
546 * debugging the kernel.
548 if (user_dbreg_trap()) {
550 * Reset breakpoint bits because the
553 /* XXX check upper bits here */
554 load_dr6(rdr6() & 0xfffffff0);
558 * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
562 * If KDB is enabled, let it handle the debugger trap.
563 * Otherwise, debugger traps "can't happen".
566 if (kdb_trap(type, 0, frame))
573 /* machine/parity/power fail/"kitchen sink" faults */
574 if (isa_nmi(code) == 0) {
577 * NMI can be hooked up to a pushbutton
581 printf ("NMI ... going to debugger\n");
582 kdb_trap(type, 0, frame);
586 } else if (panic_on_nmi == 0)
592 trap_fatal(frame, 0);
596 /* Translate fault for emulators (e.g. Linux) */
597 if (*p->p_sysent->sv_transtrap)
598 i = (*p->p_sysent->sv_transtrap)(i, type);
600 ksiginfo_init_trap(&ksi);
602 ksi.ksi_code = ucode;
603 ksi.ksi_trapno = type;
604 ksi.ksi_addr = (void *)addr;
605 trapsignal(td, &ksi);
609 mtx_assert(&Giant, MA_NOTOWNED);
610 KASSERT(PCB_USER_FPU(td->td_pcb),
611 ("Return from trap with kernel FPU ctx leaked"));
618 trap_pfault(frame, usermode)
619 struct trapframe *frame;
623 struct vmspace *vm = NULL;
627 struct thread *td = curthread;
628 struct proc *p = td->td_proc;
629 vm_offset_t eva = frame->tf_addr;
631 va = trunc_page(eva);
632 if (va >= VM_MIN_KERNEL_ADDRESS) {
634 * Don't allow user-mode faults in kernel address space.
642 * This is a fault on non-kernel virtual memory.
643 * vm is initialized above to NULL. If curproc is NULL
644 * or curproc->p_vmspace is NULL the fault is fatal.
656 * PGEX_I is defined only if the execute disable bit capability is
657 * supported and enabled.
659 if (frame->tf_err & PGEX_W)
660 ftype = VM_PROT_WRITE;
661 else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
662 ftype = VM_PROT_EXECUTE;
664 ftype = VM_PROT_READ;
666 if (map != kernel_map) {
668 * Keep swapout from messing with us during this
675 /* Fault in the user page: */
676 rv = vm_fault(map, va, ftype,
677 (ftype & VM_PROT_WRITE) ? VM_FAULT_DIRTY
685 * Don't have to worry about process locking or stacks in the
688 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
690 if (rv == KERN_SUCCESS)
694 if (td->td_intr_nesting_level == 0 &&
695 PCPU_GET(curpcb)->pcb_onfault != NULL) {
696 frame->tf_rip = (long)PCPU_GET(curpcb)->pcb_onfault;
699 trap_fatal(frame, eva);
703 return((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
707 trap_fatal(frame, eva)
708 struct trapframe *frame;
714 struct soft_segment_descriptor softseg;
717 code = frame->tf_err;
718 type = frame->tf_trapno;
719 sdtossd(&gdt[NGDT * PCPU_GET(cpuid) + IDXSEL(frame->tf_cs & 0xffff)],
722 if (type <= MAX_TRAP_MSG)
723 msg = trap_msg[type];
726 printf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
727 ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
729 /* two separate prints in case of a trap on an unmapped page */
730 printf("cpuid = %d; ", PCPU_GET(cpuid));
731 printf("apic id = %02x\n", PCPU_GET(apic_id));
733 if (type == T_PAGEFLT) {
734 printf("fault virtual address = 0x%lx\n", eva);
735 printf("fault code = %s %s %s, %s\n",
736 code & PGEX_U ? "user" : "supervisor",
737 code & PGEX_W ? "write" : "read",
738 code & PGEX_I ? "instruction" : "data",
739 code & PGEX_P ? "protection violation" : "page not present");
741 printf("instruction pointer = 0x%lx:0x%lx\n",
742 frame->tf_cs & 0xffff, frame->tf_rip);
743 if (ISPL(frame->tf_cs) == SEL_UPL) {
744 ss = frame->tf_ss & 0xffff;
747 ss = GSEL(GDATA_SEL, SEL_KPL);
748 esp = (long)&frame->tf_rsp;
750 printf("stack pointer = 0x%x:0x%lx\n", ss, esp);
751 printf("frame pointer = 0x%x:0x%lx\n", ss, frame->tf_rbp);
752 printf("code segment = base 0x%lx, limit 0x%lx, type 0x%x\n",
753 softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
754 printf(" = DPL %d, pres %d, long %d, def32 %d, gran %d\n",
755 softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32,
757 printf("processor eflags = ");
758 if (frame->tf_rflags & PSL_T)
759 printf("trace trap, ");
760 if (frame->tf_rflags & PSL_I)
761 printf("interrupt enabled, ");
762 if (frame->tf_rflags & PSL_NT)
763 printf("nested task, ");
764 if (frame->tf_rflags & PSL_RF)
766 printf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12);
767 printf("current process = ");
770 (u_long)curproc->p_pid, curthread->td_name ?
771 curthread->td_name : "");
777 if (debugger_on_panic || kdb_active)
778 if (kdb_trap(type, 0, frame))
781 printf("trap number = %d\n", type);
782 if (type <= MAX_TRAP_MSG)
783 panic("%s", trap_msg[type]);
785 panic("unknown/reserved trap");
789 * Double fault handler. Called when a fault occurs while writing
790 * a frame for a trap/exception onto the stack. This usually occurs
791 * when the stack overflows (such is the case with infinite recursion,
795 dblfault_handler(struct trapframe *frame)
798 if (dtrace_doubletrap_func != NULL)
799 (*dtrace_doubletrap_func)();
801 printf("\nFatal double fault\n");
802 printf("rip = 0x%lx\n", frame->tf_rip);
803 printf("rsp = 0x%lx\n", frame->tf_rsp);
804 printf("rbp = 0x%lx\n", frame->tf_rbp);
806 /* two separate prints in case of a trap on an unmapped page */
807 printf("cpuid = %d; ", PCPU_GET(cpuid));
808 printf("apic id = %02x\n", PCPU_GET(apic_id));
810 panic("double fault");
814 cpu_fetch_syscall_args(struct thread *td, struct syscall_args *sa)
817 struct trapframe *frame;
820 int reg, regcnt, error;
823 frame = td->td_frame;
827 params = (caddr_t)frame->tf_rsp + sizeof(register_t);
828 sa->code = frame->tf_rax;
830 if (sa->code == SYS_syscall || sa->code == SYS___syscall) {
831 sa->code = frame->tf_rdi;
835 if (p->p_sysent->sv_mask)
836 sa->code &= p->p_sysent->sv_mask;
838 if (sa->code >= p->p_sysent->sv_size)
839 sa->callp = &p->p_sysent->sv_table[0];
841 sa->callp = &p->p_sysent->sv_table[sa->code];
843 sa->narg = sa->callp->sy_narg;
844 KASSERT(sa->narg <= sizeof(sa->args) / sizeof(sa->args[0]),
845 ("Too many syscall arguments!"));
847 argp = &frame->tf_rdi;
849 bcopy(argp, sa->args, sizeof(sa->args[0]) * regcnt);
850 if (sa->narg > regcnt) {
851 KASSERT(params != NULL, ("copyin args with no params!"));
852 error = copyin(params, &sa->args[regcnt],
853 (sa->narg - regcnt) * sizeof(sa->args[0]));
857 td->td_retval[0] = 0;
858 td->td_retval[1] = frame->tf_rdx;
865 * syscall - system call request C handler
867 * A system call is essentially treated as a trap.
870 syscall(struct trapframe *frame)
873 struct syscall_args sa;
874 register_t orig_tf_rflags;
879 if (ISPL(frame->tf_cs) != SEL_UPL) {
884 orig_tf_rflags = frame->tf_rflags;
886 td->td_frame = frame;
888 error = syscallenter(td, &sa);
893 if (orig_tf_rflags & PSL_T) {
894 frame->tf_rflags &= ~PSL_T;
895 ksiginfo_init_trap(&ksi);
896 ksi.ksi_signo = SIGTRAP;
897 ksi.ksi_code = TRAP_TRACE;
898 ksi.ksi_addr = (void *)frame->tf_rip;
899 trapsignal(td, &ksi);
902 KASSERT(PCB_USER_FPU(td->td_pcb),
903 ("System call %s returing with kernel FPU ctx leaked",
904 syscallname(td->td_proc, sa.code)));
905 KASSERT(td->td_pcb->pcb_save == &td->td_pcb->pcb_user_save,
906 ("System call %s returning with mangled pcb_save",
907 syscallname(td->td_proc, sa.code)));
909 syscallret(td, error, &sa);
912 * If the user-supplied value of %rip is not a canonical
913 * address, then some CPUs will trigger a ring 0 #GP during
914 * the sysret instruction. However, the fault handler would
915 * execute with the user's %gs and %rsp in ring 0 which would
916 * not be safe. Instead, preemptively kill the thread with a
919 if (td->td_frame->tf_rip >= VM_MAXUSER_ADDRESS) {
920 ksiginfo_init_trap(&ksi);
921 ksi.ksi_signo = SIGBUS;
922 ksi.ksi_code = BUS_OBJERR;
923 ksi.ksi_trapno = T_PROTFLT;
924 ksi.ksi_addr = (void *)td->td_frame->tf_rip;
925 trapsignal(td, &ksi);