2 * Copyright (c) 2000-2010 Apple Inc. All rights reserved.
4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
6 * This file contains Original Code and/or Modifications of Original Code
7 * as defined in and that are subject to the Apple Public Source License
8 * Version 2.0 (the 'License'). You may not use this file except in
9 * compliance with the License. The rights granted to you under the License
10 * may not be used to create, or enable the creation or redistribution of,
11 * unlawful or unlicensed copies of an Apple operating system, or to
12 * circumvent, violate, or enable the circumvention or violation of, any
13 * terms of an Apple operating system software license agreement.
15 * Please obtain a copy of the License at
16 * http://www.opensource.apple.com/apsl/ and read it before using this file.
18 * The Original Code and all software distributed under the License are
19 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23 * Please see the License for the specific language governing rights and
24 * limitations under the License.
26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
32 * Mach Operating System
33 * Copyright (c) 1991,1990,1989,1988 Carnegie Mellon University
34 * All Rights Reserved.
36 * Permission to use, copy, modify and distribute this software and its
37 * documentation is hereby granted, provided that both the copyright
38 * notice and this permission notice appear in all copies of the
39 * software, derivative works or modified versions, and any portions
40 * thereof, and that both notices appear in supporting documentation.
42 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
43 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND FOR
44 * ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
46 * Carnegie Mellon requests users of this software to return to
48 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
49 * School of Computer Science
50 * Carnegie Mellon University
51 * Pittsburgh PA 15213-3890
53 * any improvements or extensions that they make and grant Carnegie Mellon
54 * the rights to redistribute these changes.
60 * Hardware trap/fault handler.
64 #include <mach_ldebug.h>
67 #include <i386/eflags.h>
68 #include <i386/trap.h>
69 #include <i386/pmap.h>
71 #include <i386/misc_protos.h> /* panic_io_port_read() */
72 #include <i386/lapic.h>
74 #include <mach/exception.h>
75 #include <mach/kern_return.h>
76 #include <mach/vm_param.h>
77 #include <mach/i386/thread_status.h>
79 #include <vm/vm_kern.h>
80 #include <vm/vm_fault.h>
82 #include <kern/kern_types.h>
83 #include <kern/processor.h>
84 #include <kern/thread.h>
85 #include <kern/task.h>
86 #include <kern/sched.h>
87 #include <kern/sched_prim.h>
88 #include <kern/exception.h>
90 #include <kern/misc_protos.h>
91 #include <kern/debug.h>
93 #include <sys/kdebug.h>
97 #include <i386/postcode.h>
98 #include <i386/mp_desc.h>
99 #include <i386/proc_reg.h>
101 #include <i386/machine_check.h>
103 #include <mach/i386/syscall_sw.h>
105 #include <libkern/OSDebug.h>
107 #include <machine/pal_routines.h>
109 extern void throttle_lowpri_io(int);
110 extern void kprint_state(x86_saved_state64_t
*saved_state
);
113 * Forward declarations
115 static void user_page_fault_continue(kern_return_t kret
);
117 static void panic_trap(x86_saved_state32_t
*saved_state
);
118 static void set_recovery_ip(x86_saved_state32_t
*saved_state
, vm_offset_t ip
);
119 extern void panic_64(x86_saved_state_t
*, int, const char *, boolean_t
);
121 static void panic_trap(x86_saved_state64_t
*saved_state
);
122 static void set_recovery_ip(x86_saved_state64_t
*saved_state
, vm_offset_t ip
);
125 volatile perfCallback perfTrapHook
= NULL
; /* Pointer to CHUD trap hook routine */
128 /* See <rdar://problem/4613924> */
129 perfCallback tempDTraceTrapHook
= NULL
; /* Pointer to DTrace fbt trap hook routine */
131 extern boolean_t
dtrace_tally_fault(user_addr_t
);
134 extern boolean_t pmap_smep_enabled
;
137 thread_syscall_return(
140 thread_t thr_act
= current_thread();
144 pal_register_cache_state(thr_act
, DIRTY
);
146 if (thread_is_64bit(thr_act
)) {
147 x86_saved_state64_t
*regs
;
149 regs
= USER_REGS64(thr_act
);
151 code
= (int) (regs
->rax
& SYSCALL_NUMBER_MASK
);
152 is_mach
= (regs
->rax
& SYSCALL_CLASS_MASK
)
153 == (SYSCALL_CLASS_MACH
<< SYSCALL_CLASS_SHIFT
);
154 if (kdebug_enable
&& is_mach
) {
156 KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE
,
157 MACHDBG_CODE(DBG_MACH_EXCP_SC
,code
)|DBG_FUNC_END
,
163 DEBUG_KPRINT_SYSCALL_MACH(
164 "thread_syscall_return: 64-bit mach ret=%u\n",
167 DEBUG_KPRINT_SYSCALL_UNIX(
168 "thread_syscall_return: 64-bit unix ret=%u\n",
172 x86_saved_state32_t
*regs
;
174 regs
= USER_REGS32(thr_act
);
176 code
= ((int) regs
->eax
);
177 is_mach
= (code
< 0);
178 if (kdebug_enable
&& is_mach
) {
180 KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE
,
181 MACHDBG_CODE(DBG_MACH_EXCP_SC
,-code
)|DBG_FUNC_END
,
187 DEBUG_KPRINT_SYSCALL_MACH(
188 "thread_syscall_return: 32-bit mach ret=%u\n",
191 DEBUG_KPRINT_SYSCALL_UNIX(
192 "thread_syscall_return: 32-bit unix ret=%u\n",
196 throttle_lowpri_io(TRUE
);
198 thread_exception_return();
204 user_page_fault_continue(
207 thread_t thread
= current_thread();
210 if (thread_is_64bit(thread
)) {
211 x86_saved_state64_t
*uregs
;
213 uregs
= USER_REGS64(thread
);
215 vaddr
= (user_addr_t
)uregs
->cr2
;
217 x86_saved_state32_t
*uregs
;
219 uregs
= USER_REGS32(thread
);
224 if (__probable((kr
== KERN_SUCCESS
) || (kr
== KERN_ABORTED
))) {
225 thread_exception_return();
230 pal_dbg_page_fault( thread
, vaddr
, kr
);
232 i386_exception(EXC_BAD_ACCESS
, kr
, vaddr
);
237 * Fault recovery in copyin/copyout routines.
240 uintptr_t fault_addr
;
241 uintptr_t recover_addr
;
244 extern struct recovery recover_table
[];
245 extern struct recovery recover_table_end
[];
247 const char * trap_type
[] = {TRAP_NAMES
};
248 unsigned TRAP_TYPES
= sizeof(trap_type
)/sizeof(trap_type
[0]);
250 extern void PE_incoming_interrupt(int interrupt
);
252 #if defined(__x86_64__) && DEBUG
254 kprint_state(x86_saved_state64_t
*saved_state
)
256 kprintf("current_cpu_datap() 0x%lx\n", (uintptr_t)current_cpu_datap());
257 kprintf("Current GS base MSR 0x%llx\n", rdmsr64(MSR_IA32_GS_BASE
));
258 kprintf("Kernel GS base MSR 0x%llx\n", rdmsr64(MSR_IA32_KERNEL_GS_BASE
));
259 kprintf("state at 0x%lx:\n", (uintptr_t) saved_state
);
261 kprintf(" rdi 0x%llx\n", saved_state
->rdi
);
262 kprintf(" rsi 0x%llx\n", saved_state
->rsi
);
263 kprintf(" rdx 0x%llx\n", saved_state
->rdx
);
264 kprintf(" r10 0x%llx\n", saved_state
->r10
);
265 kprintf(" r8 0x%llx\n", saved_state
->r8
);
266 kprintf(" r9 0x%llx\n", saved_state
->r9
);
267 kprintf(" v_arg6 0x%llx\n", saved_state
->v_arg6
);
268 kprintf(" v_arg7 0x%llx\n", saved_state
->v_arg7
);
269 kprintf(" v_arg8 0x%llx\n", saved_state
->v_arg8
);
271 kprintf(" cr2 0x%llx\n", saved_state
->cr2
);
272 kprintf("real cr2 0x%lx\n", get_cr2());
273 kprintf(" r15 0x%llx\n", saved_state
->r15
);
274 kprintf(" r14 0x%llx\n", saved_state
->r14
);
275 kprintf(" r13 0x%llx\n", saved_state
->r13
);
276 kprintf(" r12 0x%llx\n", saved_state
->r12
);
277 kprintf(" r11 0x%llx\n", saved_state
->r11
);
278 kprintf(" rbp 0x%llx\n", saved_state
->rbp
);
279 kprintf(" rbx 0x%llx\n", saved_state
->rbx
);
280 kprintf(" rcx 0x%llx\n", saved_state
->rcx
);
281 kprintf(" rax 0x%llx\n", saved_state
->rax
);
283 kprintf(" gs 0x%x\n", saved_state
->gs
);
284 kprintf(" fs 0x%x\n", saved_state
->fs
);
286 kprintf(" isf.trapno 0x%x\n", saved_state
->isf
.trapno
);
287 kprintf(" isf._pad 0x%x\n", saved_state
->isf
._pad
);
288 kprintf(" isf.trapfn 0x%llx\n", saved_state
->isf
.trapfn
);
289 kprintf(" isf.err 0x%llx\n", saved_state
->isf
.err
);
290 kprintf(" isf.rip 0x%llx\n", saved_state
->isf
.rip
);
291 kprintf(" isf.cs 0x%llx\n", saved_state
->isf
.cs
);
292 kprintf(" isf.rflags 0x%llx\n", saved_state
->isf
.rflags
);
293 kprintf(" isf.rsp 0x%llx\n", saved_state
->isf
.rsp
);
294 kprintf(" isf.ss 0x%llx\n", saved_state
->isf
.ss
);
300 * Non-zero indicates latency assert is enabled and capped at valued
301 * absolute time units.
304 uint64_t interrupt_latency_cap
= 0;
305 boolean_t ilat_assert
= FALSE
;
308 interrupt_latency_tracker_setup(void) {
309 uint32_t ilat_cap_us
;
310 if (PE_parse_boot_argn("interrupt_latency_cap_us", &ilat_cap_us
, sizeof(ilat_cap_us
))) {
311 interrupt_latency_cap
= ilat_cap_us
* NSEC_PER_USEC
;
312 nanoseconds_to_absolutetime(interrupt_latency_cap
, &interrupt_latency_cap
);
314 interrupt_latency_cap
= LockTimeOut
;
316 PE_parse_boot_argn("-interrupt_latency_assert_enable", &ilat_assert
, sizeof(ilat_assert
));
319 void interrupt_reset_latency_stats(void) {
321 for (i
= 0; i
< real_ncpus
; i
++) {
322 cpu_data_ptr
[i
]->cpu_max_observed_int_latency
=
323 cpu_data_ptr
[i
]->cpu_max_observed_int_latency_vector
= 0;
327 void interrupt_populate_latency_stats(char *buf
, unsigned bufsize
) {
328 uint32_t i
, tcpu
= ~0;
329 uint64_t cur_max
= 0;
331 for (i
= 0; i
< real_ncpus
; i
++) {
332 if (cur_max
< cpu_data_ptr
[i
]->cpu_max_observed_int_latency
) {
333 cur_max
= cpu_data_ptr
[i
]->cpu_max_observed_int_latency
;
338 if (tcpu
< real_ncpus
)
339 snprintf(buf
, bufsize
, "0x%x 0x%x 0x%llx", tcpu
, cpu_data_ptr
[tcpu
]->cpu_max_observed_int_latency_vector
, cpu_data_ptr
[tcpu
]->cpu_max_observed_int_latency
);
344 * - local APIC interrupts (IPIs, timers, etc) are handled by the kernel,
345 * - device interrupts go to the platform expert.
348 interrupt(x86_saved_state_t
*state
)
353 boolean_t user_mode
= FALSE
;
355 int cnum
= cpu_number();
358 if (is_saved_state64(state
) == TRUE
) {
359 x86_saved_state64_t
*state64
;
361 state64
= saved_state64(state
);
362 rip
= state64
->isf
.rip
;
363 rsp
= state64
->isf
.rsp
;
364 interrupt_num
= state64
->isf
.trapno
;
366 if(state64
->isf
.cs
& 0x03)
370 x86_saved_state32_t
*state32
;
372 state32
= saved_state32(state
);
373 if (state32
->cs
& 0x03)
377 interrupt_num
= state32
->trapno
;
380 if (interrupt_num
== (LAPIC_DEFAULT_INTERRUPT_BASE
+ LAPIC_INTERPROCESSOR_INTERRUPT
))
382 else if (interrupt_num
== (LAPIC_DEFAULT_INTERRUPT_BASE
+ LAPIC_TIMER_INTERRUPT
))
387 KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE
,
388 MACHDBG_CODE(DBG_MACH_EXCP_INTR
, 0) | DBG_FUNC_START
,
390 (user_mode
? rip
: VM_KERNEL_UNSLIDE(rip
)),
391 user_mode
, itype
, 0);
393 SCHED_STATS_INTERRUPT(current_processor());
395 ipl
= get_preemption_level();
398 * Handle local APIC interrupts
399 * else call platform expert for devices.
401 if (!lapic_interrupt(interrupt_num
, state
))
402 PE_incoming_interrupt(interrupt_num
);
404 if (__improbable(get_preemption_level() != ipl
)) {
405 panic("Preemption level altered by interrupt vector 0x%x: initial 0x%x, final: 0x%x\n", interrupt_num
, ipl
, get_preemption_level());
409 KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE
,
410 MACHDBG_CODE(DBG_MACH_EXCP_INTR
, 0) | DBG_FUNC_END
,
411 interrupt_num
, 0, 0, 0, 0);
413 if (cpu_data_ptr
[cnum
]->cpu_nested_istack
) {
414 cpu_data_ptr
[cnum
]->cpu_nested_istack_events
++;
417 uint64_t int_latency
= mach_absolute_time() - cpu_data_ptr
[cnum
]->cpu_int_event_time
;
418 if (ilat_assert
&& (int_latency
> interrupt_latency_cap
) && !machine_timeout_suspended()) {
419 panic("Interrupt vector 0x%x exceeded interrupt latency threshold, 0x%llx absolute time delta, prior signals: 0x%x, current signals: 0x%x", interrupt_num
, int_latency
, cpu_data_ptr
[cnum
]->cpu_prior_signals
, cpu_data_ptr
[cnum
]->cpu_signals
);
421 if (int_latency
> cpu_data_ptr
[cnum
]->cpu_max_observed_int_latency
) {
422 cpu_data_ptr
[cnum
]->cpu_max_observed_int_latency
= int_latency
;
423 cpu_data_ptr
[cnum
]->cpu_max_observed_int_latency_vector
= interrupt_num
;
428 * Having serviced the interrupt first, look at the interrupted stack depth.
431 uint64_t depth
= cpu_data_ptr
[cnum
]->cpu_kernel_stack
432 + sizeof(struct x86_kernel_state
)
433 + sizeof(struct i386_exception_link
*)
435 if (depth
> kernel_stack_depth_max
) {
436 kernel_stack_depth_max
= (vm_offset_t
)depth
;
437 KERNEL_DEBUG_CONSTANT(
438 MACHDBG_CODE(DBG_MACH_SCHED
, MACH_STACK_DEPTH
),
439 (long) depth
, (long) VM_KERNEL_UNSLIDE(rip
), 0, 0, 0);
447 long dr7
= 0x400; /* magic dr7 reset value; 32 bit on i386, 64 bit on x86_64 */
448 __asm__
volatile("mov %0,%%dr7" : : "r" (dr7
));
451 unsigned kdp_has_active_watchpoints
= 0;
452 #define NO_WATCHPOINTS (!kdp_has_active_watchpoints)
454 #define NO_WATCHPOINTS 1
457 * Trap from kernel mode. Only page-fault errors are recoverable,
458 * and then only in special circumstances. All other errors are
459 * fatal. Return value indicates if trap was handled.
464 x86_saved_state_t
*state
,
468 x86_saved_state32_t
*saved_state
;
470 x86_saved_state64_t
*saved_state
;
475 vm_map_t map
= 0; /* protected by T_PAGE_FAULT */
476 kern_return_t result
= KERN_FAILURE
;
483 #if NCOPY_WINDOWS > 0
484 int fault_in_copy_window
= -1;
488 thread
= current_thread();
491 if (__improbable(is_saved_state64(state
))) {
492 panic_64(state
, 0, "Kernel trap with 64-bit state", FALSE
);
495 saved_state
= saved_state32(state
);
497 /* Record cpu where state was captured (trampolines don't set this) */
498 saved_state
->cpu
= cpu_number();
500 vaddr
= (user_addr_t
)saved_state
->cr2
;
501 type
= saved_state
->trapno
;
502 code
= saved_state
->err
& 0xffff;
503 intr
= (saved_state
->efl
& EFL_IF
) != 0; /* state of ints at trap */
504 kern_ip
= (vm_offset_t
)saved_state
->eip
;
506 if (__improbable(is_saved_state32(state
)))
507 panic("kernel_trap(%p) with 32-bit state", state
);
508 saved_state
= saved_state64(state
);
510 /* Record cpu where state was captured */
511 saved_state
->isf
.cpu
= cpu_number();
513 vaddr
= (user_addr_t
)saved_state
->cr2
;
514 type
= saved_state
->isf
.trapno
;
515 code
= (int)(saved_state
->isf
.err
& 0xffff);
516 intr
= (saved_state
->isf
.rflags
& EFL_IF
) != 0; /* state of ints at trap */
517 kern_ip
= (vm_offset_t
)saved_state
->isf
.rip
;
520 myast
= ast_pending();
522 perfASTCallback astfn
= perfASTHook
;
523 if (__improbable(astfn
!= NULL
)) {
524 if (*myast
& AST_CHUD_ALL
)
525 astfn(AST_CHUD_ALL
, myast
);
527 *myast
&= ~AST_CHUD_ALL
;
532 perfCallback fn
= perfTrapHook
;
533 if (__improbable(fn
!= NULL
)) {
534 if (fn(type
, NULL
, 0, 0) == KERN_SUCCESS
) {
536 * If it succeeds, we are done...
543 if (__improbable(tempDTraceTrapHook
!= NULL
)) {
544 if (tempDTraceTrapHook(type
, state
, lo_spp
, 0) == KERN_SUCCESS
) {
546 * If it succeeds, we are done...
551 #endif /* CONFIG_DTRACE */
554 * we come here with interrupts off as we don't want to recurse
555 * on preemption below. but we do want to re-enable interrupts
556 * as soon we possibly can to hold latency down
558 if (__improbable(T_PREEMPT
== type
)) {
559 ast_taken(AST_PREEMPTION
, FALSE
);
561 KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE
,
562 (MACHDBG_CODE(DBG_MACH_EXCP_KTRAP_x86
, type
)) | DBG_FUNC_NONE
,
563 0, 0, 0, VM_KERNEL_UNSLIDE(kern_ip
), 0);
567 if (T_PAGE_FAULT
== type
) {
569 * assume we're faulting in the kernel map
573 if (__probable(thread
!= THREAD_NULL
&& thread
->map
!= kernel_map
)) {
574 #if NCOPY_WINDOWS > 0
575 vm_offset_t copy_window_base
;
579 kvaddr
= (vm_offset_t
)vaddr
;
581 * must determine if fault occurred in
582 * the copy window while pre-emption is
583 * disabled for this processor so that
584 * we only need to look at the window
585 * associated with this processor
587 copy_window_base
= current_cpu_datap()->cpu_copywindow_base
;
589 if (kvaddr
>= copy_window_base
&& kvaddr
< (copy_window_base
+ (NBPDE
* NCOPY_WINDOWS
)) ) {
591 window_index
= (int)((kvaddr
- copy_window_base
) / NBPDE
);
593 if (thread
->machine
.copy_window
[window_index
].user_base
!= (user_addr_t
)-1) {
595 kvaddr
-= (copy_window_base
+ (NBPDE
* window_index
));
596 vaddr
= thread
->machine
.copy_window
[window_index
].user_base
+ kvaddr
;
599 fault_in_copy_window
= window_index
;
604 if (__probable(vaddr
< VM_MAX_USER_PAGE_ADDRESS
)) {
605 /* fault occurred in userspace */
609 /* Intercept a potential Supervisor Mode Execute
610 * Protection fault. These criteria identify
611 * both NX faults and SMEP faults, but both
612 * are fatal. We avoid checking PTEs (racy).
613 * (The VM could just redrive a SMEP fault, hence
616 if (__improbable((code
== (T_PF_PROT
| T_PF_EXECUTE
)) && (pmap_smep_enabled
) && (saved_state
->isf
.rip
== vaddr
))) {
621 * If we're not sharing cr3 with the user
622 * and we faulted in copyio,
623 * then switch cr3 here and dismiss the fault.
626 (thread
->machine
.specFlags
&CopyIOActive
) &&
627 map
->pmap
->pm_cr3
!= get_cr3_base()) {
628 pmap_assert(current_cpu_datap()->cpu_pmap_pcid_enabled
== FALSE
);
629 set_cr3_raw(map
->pmap
->pm_cr3
);
636 user_addr_t kd_vaddr
= is_user
? vaddr
: VM_KERNEL_UNSLIDE(vaddr
);
637 KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE
,
638 (MACHDBG_CODE(DBG_MACH_EXCP_KTRAP_x86
, type
)) | DBG_FUNC_NONE
,
639 (unsigned)(kd_vaddr
>> 32), (unsigned)kd_vaddr
, is_user
,
640 VM_KERNEL_UNSLIDE(kern_ip
), 0);
643 (void) ml_set_interrupts_enabled(intr
);
655 case T_FLOATING_POINT_ERROR
:
659 case T_SSE_FLOAT_ERROR
:
664 if ((saved_state
->efl
& EFL_TF
) == 0 && NO_WATCHPOINTS
)
666 if ((saved_state
->isf
.rflags
& EFL_TF
) == 0 && NO_WATCHPOINTS
)
669 /* We've somehow encountered a debug
670 * register match that does not belong
671 * to the kernel debugger.
672 * This isn't supposed to happen.
685 if (thread
!= THREAD_NULL
&& thread
->options
& TH_OPT_DTRACE
) { /* Executing under dtrace_probe? */
686 if (dtrace_tally_fault(vaddr
)) { /* Should a fault under dtrace be ignored? */
688 * DTrace has "anticipated" the possibility of this fault, and has
689 * established the suitable recovery state. Drop down now into the
690 * recovery handling code in "case T_GENERAL_PROTECTION:".
695 #endif /* CONFIG_DTRACE */
699 if (code
& T_PF_WRITE
)
700 prot
|= VM_PROT_WRITE
;
702 if (code
& T_PF_EXECUTE
)
703 prot
|= VM_PROT_EXECUTE
;
706 result
= vm_fault(map
,
707 vm_map_trunc_page(vaddr
),
710 THREAD_UNINT
, NULL
, 0);
712 if (result
== KERN_SUCCESS
) {
713 #if NCOPY_WINDOWS > 0
714 if (fault_in_copy_window
!= -1) {
715 ml_set_interrupts_enabled(FALSE
);
716 copy_window_fault(thread
, map
,
717 fault_in_copy_window
);
718 (void) ml_set_interrupts_enabled(intr
);
720 #endif /* NCOPY_WINDOWS > 0 */
728 #endif /* CONFIG_DTRACE */
730 case T_GENERAL_PROTECTION
:
732 * If there is a failure recovery address
733 * for this fault, go there.
735 for (rp
= recover_table
; rp
< recover_table_end
; rp
++) {
736 if (kern_ip
== rp
->fault_addr
) {
737 set_recovery_ip(saved_state
, rp
->recover_addr
);
743 * Check thread recovery address also.
745 if (thread
!= THREAD_NULL
&& thread
->recover
) {
746 set_recovery_ip(saved_state
, thread
->recover
);
751 * Unanticipated page-fault errors in kernel
758 * Exception 15 is reserved but some chips may generate it
759 * spuriously. Seen at startup on AMD Athlon-64.
762 kprintf("kernel_trap() ignoring spurious trap 15\n");
766 /* Ensure that the i386_kernel_state at the base of the
767 * current thread's stack (if any) is synchronized with the
768 * context at the moment of the trap, to facilitate
769 * access through the debugger.
771 sync_iss_to_iks(state
);
773 if (current_debugger
!= KDB_CUR_DB
) {
774 if (kdp_i386_trap(type
, saved_state
, result
, (vm_offset_t
)vaddr
))
780 panic_trap(saved_state
);
789 set_recovery_ip(x86_saved_state32_t
*saved_state
, vm_offset_t ip
)
791 saved_state
->eip
= ip
;
795 set_recovery_ip(x86_saved_state64_t
*saved_state
, vm_offset_t ip
)
797 saved_state
->isf
.rip
= ip
;
804 panic_trap(x86_saved_state32_t
*regs
)
806 const char *trapname
= "Unknown";
807 pal_cr_t cr0
, cr2
, cr3
, cr4
;
809 pal_get_control_registers( &cr0
, &cr2
, &cr3
, &cr4
);
812 * Issue an I/O port read if one has been requested - this is an
813 * event logic analyzers can use as a trigger point.
815 panic_io_port_read();
817 kprintf("panic trap number 0x%x, eip 0x%x\n", regs
->trapno
, regs
->eip
);
818 kprintf("cr0 0x%08x cr2 0x%08x cr3 0x%08x cr4 0x%08x\n",
821 if (regs
->trapno
< TRAP_TYPES
)
822 trapname
= trap_type
[regs
->trapno
];
824 panic("Kernel trap at 0x%08x, type %d=%s, registers:\n"
825 "CR0: 0x%08x, CR2: 0x%08x, CR3: 0x%08x, CR4: 0x%08x\n"
826 "EAX: 0x%08x, EBX: 0x%08x, ECX: 0x%08x, EDX: 0x%08x\n"
827 "CR2: 0x%08x, EBP: 0x%08x, ESI: 0x%08x, EDI: 0x%08x\n"
828 "EFL: 0x%08x, EIP: 0x%08x, CS: 0x%08x, DS: 0x%08x\n"
829 "Error code: 0x%08x%s\n",
830 regs
->eip
, regs
->trapno
, trapname
, cr0
, cr2
, cr3
, cr4
,
831 regs
->eax
,regs
->ebx
,regs
->ecx
,regs
->edx
,
832 regs
->cr2
,regs
->ebp
,regs
->esi
,regs
->edi
,
833 regs
->efl
,regs
->eip
,regs
->cs
& 0xFFFF, regs
->ds
& 0xFFFF, regs
->err
,
834 virtualized
? " VMM" : "");
836 * This next statement is not executed,
837 * but it's needed to stop the compiler using tail call optimization
838 * for the panic call - which confuses the subsequent backtrace.
846 panic_trap(x86_saved_state64_t
*regs
)
848 const char *trapname
= "Unknown";
849 pal_cr_t cr0
, cr2
, cr3
, cr4
;
850 boolean_t potential_smep_fault
= FALSE
, potential_kernel_NX_fault
= FALSE
;
852 pal_get_control_registers( &cr0
, &cr2
, &cr3
, &cr4
);
853 assert(ml_get_interrupts_enabled() == FALSE
);
854 current_cpu_datap()->cpu_fatal_trap_state
= regs
;
856 * Issue an I/O port read if one has been requested - this is an
857 * event logic analyzers can use as a trigger point.
859 panic_io_port_read();
861 kprintf("panic trap number 0x%x, rip 0x%016llx\n",
862 regs
->isf
.trapno
, regs
->isf
.rip
);
863 kprintf("cr0 0x%016llx cr2 0x%016llx cr3 0x%016llx cr4 0x%016llx\n",
866 if (regs
->isf
.trapno
< TRAP_TYPES
)
867 trapname
= trap_type
[regs
->isf
.trapno
];
869 if ((regs
->isf
.trapno
== T_PAGE_FAULT
) && (regs
->isf
.err
== (T_PF_PROT
| T_PF_EXECUTE
)) && (regs
->isf
.rip
== regs
->cr2
)) {
870 if (pmap_smep_enabled
&& (regs
->isf
.rip
< VM_MAX_USER_PAGE_ADDRESS
)) {
871 potential_smep_fault
= TRUE
;
872 } else if (regs
->isf
.rip
>= VM_MIN_KERNEL_AND_KEXT_ADDRESS
) {
873 potential_kernel_NX_fault
= TRUE
;
878 panic("Kernel trap at 0x%016llx, type %d=%s, registers:\n"
879 "CR0: 0x%016llx, CR2: 0x%016llx, CR3: 0x%016llx, CR4: 0x%016llx\n"
880 "RAX: 0x%016llx, RBX: 0x%016llx, RCX: 0x%016llx, RDX: 0x%016llx\n"
881 "RSP: 0x%016llx, RBP: 0x%016llx, RSI: 0x%016llx, RDI: 0x%016llx\n"
882 "R8: 0x%016llx, R9: 0x%016llx, R10: 0x%016llx, R11: 0x%016llx\n"
883 "R12: 0x%016llx, R13: 0x%016llx, R14: 0x%016llx, R15: 0x%016llx\n"
884 "RFL: 0x%016llx, RIP: 0x%016llx, CS: 0x%016llx, SS: 0x%016llx\n"
885 "Fault CR2: 0x%016llx, Error code: 0x%016llx, Fault CPU: 0x%x%s%s%s\n",
886 regs
->isf
.rip
, regs
->isf
.trapno
, trapname
,
888 regs
->rax
, regs
->rbx
, regs
->rcx
, regs
->rdx
,
889 regs
->isf
.rsp
, regs
->rbp
, regs
->rsi
, regs
->rdi
,
890 regs
->r8
, regs
->r9
, regs
->r10
, regs
->r11
,
891 regs
->r12
, regs
->r13
, regs
->r14
, regs
->r15
,
892 regs
->isf
.rflags
, regs
->isf
.rip
, regs
->isf
.cs
& 0xFFFF,
893 regs
->isf
.ss
& 0xFFFF,regs
->cr2
, regs
->isf
.err
, regs
->isf
.cpu
,
894 virtualized
? " VMM" : "",
895 potential_kernel_NX_fault
? " Kernel NX fault" : "",
896 potential_smep_fault
? " SMEP/User NX fault" : "");
898 * This next statement is not executed,
899 * but it's needed to stop the compiler using tail call optimization
900 * for the panic call - which confuses the subsequent backtrace.
907 extern kern_return_t
dtrace_user_probe(x86_saved_state_t
*);
911 * Trap from user mode.
915 x86_saved_state_t
*saved_state
)
919 mach_exception_code_t code
;
920 mach_exception_subcode_t subcode
;
924 thread_t thread
= current_thread();
928 unsigned long dr6
= 0; /* 32 bit for i386, 64 bit for x86_64 */
930 assert((is_saved_state32(saved_state
) && !thread_is_64bit(thread
)) ||
931 (is_saved_state64(saved_state
) && thread_is_64bit(thread
)));
933 if (is_saved_state64(saved_state
)) {
934 x86_saved_state64_t
*regs
;
936 regs
= saved_state64(saved_state
);
938 /* Record cpu where state was captured */
939 regs
->isf
.cpu
= cpu_number();
941 type
= regs
->isf
.trapno
;
942 err
= (int)regs
->isf
.err
& 0xffff;
943 vaddr
= (user_addr_t
)regs
->cr2
;
944 rip
= (user_addr_t
)regs
->isf
.rip
;
946 x86_saved_state32_t
*regs
;
948 regs
= saved_state32(saved_state
);
950 /* Record cpu where state was captured */
951 regs
->cpu
= cpu_number();
954 err
= regs
->err
& 0xffff;
955 vaddr
= (user_addr_t
)regs
->cr2
;
956 rip
= (user_addr_t
)regs
->eip
;
959 if ((type
== T_DEBUG
) && thread
->machine
.ids
) {
960 unsigned long clear
= 0;
961 /* Stash and clear this processor's DR6 value, in the event
962 * this was a debug register match
964 __asm__
volatile ("mov %%db6, %0" : "=r" (dr6
));
965 __asm__
volatile ("mov %0, %%db6" : : "r" (clear
));
970 KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE
,
971 (MACHDBG_CODE(DBG_MACH_EXCP_UTRAP_x86
, type
)) | DBG_FUNC_NONE
,
972 (unsigned)(vaddr
>>32), (unsigned)vaddr
,
973 (unsigned)(rip
>>32), (unsigned)rip
, 0);
980 kprintf("user_trap(0x%08x) type=%d vaddr=0x%016llx\n",
981 saved_state
, type
, vaddr
);
984 perfASTCallback astfn
= perfASTHook
;
985 if (__improbable(astfn
!= NULL
)) {
986 myast
= ast_pending();
987 if (*myast
& AST_CHUD_ALL
) {
988 astfn(AST_CHUD_ALL
, myast
);
992 /* Is there a hook? */
993 perfCallback fn
= perfTrapHook
;
994 if (__improbable(fn
!= NULL
)) {
995 if (fn(type
, saved_state
, 0, 0) == KERN_SUCCESS
)
996 return; /* If it succeeds, we are done... */
1000 * DTrace does not consume all user traps, only INT_3's for now.
1001 * Avoid needlessly calling tempDTraceTrapHook here, and let the
1002 * INT_3 case handle them.
1004 DEBUG_KPRINT_SYSCALL_MASK(1,
1005 "user_trap: type=0x%x(%s) err=0x%x cr2=%p rip=%p\n",
1006 type
, trap_type
[type
], err
, (void *)(long) vaddr
, (void *)(long) rip
);
1010 case T_DIVIDE_ERROR
:
1011 exc
= EXC_ARITHMETIC
;
1012 code
= EXC_I386_DIV
;
1019 * Update the PCB with this processor's DR6 value
1020 * in the event this was a debug register match.
1022 pcb
= THREAD_TO_PCB(thread
);
1025 * We can get and set the status register
1026 * in 32-bit mode even on a 64-bit thread
1027 * because the high order bits are not
1030 if (thread_is_64bit(thread
)) {
1031 x86_debug_state64_t
*ids
= pcb
->ids
;
1033 } else { /* 32 bit thread */
1034 x86_debug_state32_t
*ids
= pcb
->ids
;
1035 ids
->dr6
= (uint32_t) dr6
;
1038 exc
= EXC_BREAKPOINT
;
1039 code
= EXC_I386_SGL
;
1044 if (dtrace_user_probe(saved_state
) == KERN_SUCCESS
)
1045 return; /* If it succeeds, we are done... */
1047 exc
= EXC_BREAKPOINT
;
1048 code
= EXC_I386_BPT
;
1052 exc
= EXC_ARITHMETIC
;
1053 code
= EXC_I386_INTO
;
1056 case T_OUT_OF_BOUNDS
:
1058 code
= EXC_I386_BOUND
;
1061 case T_INVALID_OPCODE
:
1062 exc
= EXC_BAD_INSTRUCTION
;
1063 code
= EXC_I386_INVOP
;
1071 fpextovrflt(); /* Propagates exception directly, doesn't return */
1074 case T_INVALID_TSS
: /* invalid TSS == iret with NT flag set */
1075 exc
= EXC_BAD_INSTRUCTION
;
1076 code
= EXC_I386_INVTSSFLT
;
1080 case T_SEGMENT_NOT_PRESENT
:
1081 exc
= EXC_BAD_INSTRUCTION
;
1082 code
= EXC_I386_SEGNPFLT
;
1087 exc
= EXC_BAD_INSTRUCTION
;
1088 code
= EXC_I386_STKFLT
;
1092 case T_GENERAL_PROTECTION
:
1094 * There's a wide range of circumstances which generate this
1095 * class of exception. From user-space, many involve bad
1096 * addresses (such as a non-canonical 64-bit address).
1097 * So we map this to EXC_BAD_ACCESS (and thereby SIGSEGV).
1098 * The trouble is cr2 doesn't contain the faulting address;
1099 * we'd need to decode the faulting instruction to really
1100 * determine this. We'll leave that to debuggers.
1101 * However, attempted execution of privileged instructions
1102 * (e.g. cli) also generate GP faults and so we map these to
1103 * to EXC_BAD_ACCESS (and thence SIGSEGV) also - rather than
1104 * EXC_BAD_INSTRUCTION which is more accurate. We just can't
1107 exc
= EXC_BAD_ACCESS
;
1108 code
= EXC_I386_GPFLT
;
1113 prot
= VM_PROT_READ
;
1115 if (err
& T_PF_WRITE
)
1116 prot
|= VM_PROT_WRITE
;
1118 if (__improbable(err
& T_PF_EXECUTE
))
1119 prot
|= VM_PROT_EXECUTE
;
1121 kret
= vm_fault(thread
->map
, vm_map_trunc_page(vaddr
),
1123 THREAD_ABORTSAFE
, NULL
, 0);
1125 user_page_fault_continue(kret
);
1130 case T_SSE_FLOAT_ERROR
:
1131 fpSSEexterrflt(); /* Propagates exception directly, doesn't return */
1135 case T_FLOATING_POINT_ERROR
:
1136 fpexterrflt(); /* Propagates exception directly, doesn't return */
1141 if (dtrace_user_probe(saved_state
) == KERN_SUCCESS
)
1142 return; /* If it succeeds, we are done... */
1145 * If we get an INT 0x7f when we do not expect to,
1146 * treat it as an illegal instruction
1148 exc
= EXC_BAD_INSTRUCTION
;
1149 code
= EXC_I386_INVOP
;
1153 panic("Unexpected user trap, type %d", type
);
1156 /* Note: Codepaths that directly return from user_trap() have pending
1157 * ASTs processed in locore
1159 i386_exception(exc
, code
, subcode
);
1165 * Handle AST traps for i386.
1168 extern void log_thread_action (thread_t
, char *);
1171 i386_astintr(int preemption
)
1173 ast_t mask
= AST_ALL
;
1177 mask
= AST_PREEMPTION
;
1187 * Handle exceptions for i386.
1189 * If we are an AT bus machine, we must turn off the AST for a
1190 * delayed floating-point exception.
1192 * If we are providing floating-point emulation, we may have
1193 * to retrieve the real register values from the floating point
1199 mach_exception_code_t code
,
1200 mach_exception_subcode_t subcode
)
1202 mach_exception_data_type_t codes
[EXCEPTION_CODE_MAX
];
1204 DEBUG_KPRINT_SYSCALL_MACH("i386_exception: exc=%d code=0x%llx subcode=0x%llx\n",
1205 exc
, code
, subcode
);
1206 codes
[0] = code
; /* new exception interface */
1208 exception_triage(exc
, codes
, 2);
1213 /* Synchronize a thread's i386_kernel_state (if any) with the given
1214 * i386_saved_state_t obtained from the trap/IPI handler; called in
1215 * kernel_trap() prior to entering the debugger, and when receiving
1220 sync_iss_to_iks(x86_saved_state_t
*saved_state
)
1222 struct x86_kernel_state
*iks
;
1224 boolean_t record_active_regs
= FALSE
;
1226 /* The PAL may have a special way to sync registers */
1227 if( saved_state
->flavor
== THREAD_STATE_NONE
)
1228 pal_get_kern_regs( saved_state
);
1230 if ((kstack
= current_thread()->kernel_stack
) != 0) {
1232 x86_saved_state32_t
*regs
= saved_state32(saved_state
);
1234 x86_saved_state64_t
*regs
= saved_state64(saved_state
);
1237 iks
= STACK_IKS(kstack
);
1239 /* Did we take the trap/interrupt in kernel mode? */
1241 if (regs
== USER_REGS32(current_thread()))
1242 record_active_regs
= TRUE
;
1244 iks
->k_ebx
= regs
->ebx
;
1245 iks
->k_esp
= (int)regs
;
1246 iks
->k_ebp
= regs
->ebp
;
1247 iks
->k_edi
= regs
->edi
;
1248 iks
->k_esi
= regs
->esi
;
1249 iks
->k_eip
= regs
->eip
;
1252 if (regs
== USER_REGS64(current_thread()))
1253 record_active_regs
= TRUE
;
1255 iks
->k_rbx
= regs
->rbx
;
1256 iks
->k_rsp
= regs
->isf
.rsp
;
1257 iks
->k_rbp
= regs
->rbp
;
1258 iks
->k_r12
= regs
->r12
;
1259 iks
->k_r13
= regs
->r13
;
1260 iks
->k_r14
= regs
->r14
;
1261 iks
->k_r15
= regs
->r15
;
1262 iks
->k_rip
= regs
->isf
.rip
;
1267 if (record_active_regs
== TRUE
) {
1269 /* Show the trap handler path */
1270 __asm__
volatile("movl %%ebx, %0" : "=m" (iks
->k_ebx
));
1271 __asm__
volatile("movl %%esp, %0" : "=m" (iks
->k_esp
));
1272 __asm__
volatile("movl %%ebp, %0" : "=m" (iks
->k_ebp
));
1273 __asm__
volatile("movl %%edi, %0" : "=m" (iks
->k_edi
));
1274 __asm__
volatile("movl %%esi, %0" : "=m" (iks
->k_esi
));
1275 /* "Current" instruction pointer */
1276 __asm__
volatile("movl $1f, %0\n1:" : "=m" (iks
->k_eip
));
1278 /* Show the trap handler path */
1279 __asm__
volatile("movq %%rbx, %0" : "=m" (iks
->k_rbx
));
1280 __asm__
volatile("movq %%rsp, %0" : "=m" (iks
->k_rsp
));
1281 __asm__
volatile("movq %%rbp, %0" : "=m" (iks
->k_rbp
));
1282 __asm__
volatile("movq %%r12, %0" : "=m" (iks
->k_r12
));
1283 __asm__
volatile("movq %%r13, %0" : "=m" (iks
->k_r13
));
1284 __asm__
volatile("movq %%r14, %0" : "=m" (iks
->k_r14
));
1285 __asm__
volatile("movq %%r15, %0" : "=m" (iks
->k_r15
));
1286 /* "Current" instruction pointer */
1287 __asm__
volatile("leaq 1f(%%rip), %%rax; mov %%rax, %0\n1:"
1296 * This is used by the NMI interrupt handler (from mp.c) to
1297 * uncondtionally sync the trap handler context to the IKS
1298 * irrespective of whether the NMI was fielded in kernel
1302 sync_iss_to_iks_unconditionally(__unused x86_saved_state_t
*saved_state
) {
1303 struct x86_kernel_state
*iks
;
1306 if ((kstack
= current_thread()->kernel_stack
) != 0) {
1307 iks
= STACK_IKS(kstack
);
1309 /* Display the trap handler path */
1310 __asm__
volatile("movl %%ebx, %0" : "=m" (iks
->k_ebx
));
1311 __asm__
volatile("movl %%esp, %0" : "=m" (iks
->k_esp
));
1312 __asm__
volatile("movl %%ebp, %0" : "=m" (iks
->k_ebp
));
1313 __asm__
volatile("movl %%edi, %0" : "=m" (iks
->k_edi
));
1314 __asm__
volatile("movl %%esi, %0" : "=m" (iks
->k_esi
));
1315 /* "Current" instruction pointer */
1316 __asm__
volatile("movl $1f, %0\n1:" : "=m" (iks
->k_eip
));
1318 /* Display the trap handler path */
1319 __asm__
volatile("movq %%rbx, %0" : "=m" (iks
->k_rbx
));
1320 __asm__
volatile("movq %%rsp, %0" : "=m" (iks
->k_rsp
));
1321 __asm__
volatile("movq %%rbp, %0" : "=m" (iks
->k_rbp
));
1322 __asm__
volatile("movq %%r12, %0" : "=m" (iks
->k_r12
));
1323 __asm__
volatile("movq %%r13, %0" : "=m" (iks
->k_r13
));
1324 __asm__
volatile("movq %%r14, %0" : "=m" (iks
->k_r14
));
1325 __asm__
volatile("movq %%r15, %0" : "=m" (iks
->k_r15
));
1326 /* "Current" instruction pointer */
1327 __asm__
volatile("leaq 1f(%%rip), %%rax; mov %%rax, %0\n1:" : "=m" (iks
->k_rip
)::"rax");