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
);
226 pal_dbg_page_fault( thread
, vaddr
, kr
);
228 i386_exception(EXC_BAD_ACCESS
, kr
, vaddr
);
233 * Fault recovery in copyin/copyout routines.
236 uintptr_t fault_addr
;
237 uintptr_t recover_addr
;
240 extern struct recovery recover_table
[];
241 extern struct recovery recover_table_end
[];
243 const char * trap_type
[] = {TRAP_NAMES
};
244 unsigned TRAP_TYPES
= sizeof(trap_type
)/sizeof(trap_type
[0]);
246 extern void PE_incoming_interrupt(int interrupt
);
248 #if defined(__x86_64__) && DEBUG
250 kprint_state(x86_saved_state64_t
*saved_state
)
252 kprintf("current_cpu_datap() 0x%lx\n", (uintptr_t)current_cpu_datap());
253 kprintf("Current GS base MSR 0x%llx\n", rdmsr64(MSR_IA32_GS_BASE
));
254 kprintf("Kernel GS base MSR 0x%llx\n", rdmsr64(MSR_IA32_KERNEL_GS_BASE
));
255 kprintf("state at 0x%lx:\n", (uintptr_t) saved_state
);
257 kprintf(" rdi 0x%llx\n", saved_state
->rdi
);
258 kprintf(" rsi 0x%llx\n", saved_state
->rsi
);
259 kprintf(" rdx 0x%llx\n", saved_state
->rdx
);
260 kprintf(" r10 0x%llx\n", saved_state
->r10
);
261 kprintf(" r8 0x%llx\n", saved_state
->r8
);
262 kprintf(" r9 0x%llx\n", saved_state
->r9
);
263 kprintf(" v_arg6 0x%llx\n", saved_state
->v_arg6
);
264 kprintf(" v_arg7 0x%llx\n", saved_state
->v_arg7
);
265 kprintf(" v_arg8 0x%llx\n", saved_state
->v_arg8
);
267 kprintf(" cr2 0x%llx\n", saved_state
->cr2
);
268 kprintf("real cr2 0x%lx\n", get_cr2());
269 kprintf(" r15 0x%llx\n", saved_state
->r15
);
270 kprintf(" r14 0x%llx\n", saved_state
->r14
);
271 kprintf(" r13 0x%llx\n", saved_state
->r13
);
272 kprintf(" r12 0x%llx\n", saved_state
->r12
);
273 kprintf(" r11 0x%llx\n", saved_state
->r11
);
274 kprintf(" rbp 0x%llx\n", saved_state
->rbp
);
275 kprintf(" rbx 0x%llx\n", saved_state
->rbx
);
276 kprintf(" rcx 0x%llx\n", saved_state
->rcx
);
277 kprintf(" rax 0x%llx\n", saved_state
->rax
);
279 kprintf(" gs 0x%x\n", saved_state
->gs
);
280 kprintf(" fs 0x%x\n", saved_state
->fs
);
282 kprintf(" isf.trapno 0x%x\n", saved_state
->isf
.trapno
);
283 kprintf(" isf._pad 0x%x\n", saved_state
->isf
._pad
);
284 kprintf(" isf.trapfn 0x%llx\n", saved_state
->isf
.trapfn
);
285 kprintf(" isf.err 0x%llx\n", saved_state
->isf
.err
);
286 kprintf(" isf.rip 0x%llx\n", saved_state
->isf
.rip
);
287 kprintf(" isf.cs 0x%llx\n", saved_state
->isf
.cs
);
288 kprintf(" isf.rflags 0x%llx\n", saved_state
->isf
.rflags
);
289 kprintf(" isf.rsp 0x%llx\n", saved_state
->isf
.rsp
);
290 kprintf(" isf.ss 0x%llx\n", saved_state
->isf
.ss
);
296 * Non-zero indicates latency assert is enabled and capped at valued
297 * absolute time units.
300 uint64_t interrupt_latency_cap
= 0;
301 boolean_t ilat_assert
= FALSE
;
304 interrupt_latency_tracker_setup(void) {
305 uint32_t ilat_cap_us
;
306 if (PE_parse_boot_argn("interrupt_latency_cap_us", &ilat_cap_us
, sizeof(ilat_cap_us
))) {
307 interrupt_latency_cap
= ilat_cap_us
* NSEC_PER_USEC
;
308 nanoseconds_to_absolutetime(interrupt_latency_cap
, &interrupt_latency_cap
);
310 interrupt_latency_cap
= LockTimeOut
;
312 PE_parse_boot_argn("-interrupt_latency_assert_enable", &ilat_assert
, sizeof(ilat_assert
));
315 void interrupt_reset_latency_stats(void) {
317 for (i
= 0; i
< real_ncpus
; i
++) {
318 cpu_data_ptr
[i
]->cpu_max_observed_int_latency
=
319 cpu_data_ptr
[i
]->cpu_max_observed_int_latency_vector
= 0;
323 void interrupt_populate_latency_stats(char *buf
, unsigned bufsize
) {
324 uint32_t i
, tcpu
= ~0;
325 uint64_t cur_max
= 0;
327 for (i
= 0; i
< real_ncpus
; i
++) {
328 if (cur_max
< cpu_data_ptr
[i
]->cpu_max_observed_int_latency
) {
329 cur_max
= cpu_data_ptr
[i
]->cpu_max_observed_int_latency
;
334 if (tcpu
< real_ncpus
)
335 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
);
340 * - local APIC interrupts (IPIs, timers, etc) are handled by the kernel,
341 * - device interrupts go to the platform expert.
344 interrupt(x86_saved_state_t
*state
)
349 boolean_t user_mode
= FALSE
;
351 int cnum
= cpu_number();
354 if (is_saved_state64(state
) == TRUE
) {
355 x86_saved_state64_t
*state64
;
357 state64
= saved_state64(state
);
358 rip
= state64
->isf
.rip
;
359 rsp
= state64
->isf
.rsp
;
360 interrupt_num
= state64
->isf
.trapno
;
362 if(state64
->isf
.cs
& 0x03)
366 x86_saved_state32_t
*state32
;
368 state32
= saved_state32(state
);
369 if (state32
->cs
& 0x03)
373 interrupt_num
= state32
->trapno
;
376 if (interrupt_num
== (LAPIC_DEFAULT_INTERRUPT_BASE
+ LAPIC_INTERPROCESSOR_INTERRUPT
))
378 else if (interrupt_num
== (LAPIC_DEFAULT_INTERRUPT_BASE
+ LAPIC_TIMER_INTERRUPT
))
383 KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE
,
384 MACHDBG_CODE(DBG_MACH_EXCP_INTR
, 0) | DBG_FUNC_START
,
386 (user_mode
? rip
: VM_KERNEL_UNSLIDE(rip
)),
387 user_mode
, itype
, 0);
389 SCHED_STATS_INTERRUPT(current_processor());
391 ipl
= get_preemption_level();
394 * Handle local APIC interrupts
395 * else call platform expert for devices.
397 if (!lapic_interrupt(interrupt_num
, state
))
398 PE_incoming_interrupt(interrupt_num
);
400 if (__improbable(get_preemption_level() != ipl
)) {
401 panic("Preemption level altered by interrupt vector 0x%x: initial 0x%x, final: 0x%x\n", interrupt_num
, ipl
, get_preemption_level());
405 KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE
,
406 MACHDBG_CODE(DBG_MACH_EXCP_INTR
, 0) | DBG_FUNC_END
,
407 interrupt_num
, 0, 0, 0, 0);
409 if (cpu_data_ptr
[cnum
]->cpu_nested_istack
) {
410 cpu_data_ptr
[cnum
]->cpu_nested_istack_events
++;
413 uint64_t int_latency
= mach_absolute_time() - cpu_data_ptr
[cnum
]->cpu_int_event_time
;
414 if (ilat_assert
&& (int_latency
> interrupt_latency_cap
) && !machine_timeout_suspended()) {
415 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
);
417 if (int_latency
> cpu_data_ptr
[cnum
]->cpu_max_observed_int_latency
) {
418 cpu_data_ptr
[cnum
]->cpu_max_observed_int_latency
= int_latency
;
419 cpu_data_ptr
[cnum
]->cpu_max_observed_int_latency_vector
= interrupt_num
;
424 * Having serviced the interrupt first, look at the interrupted stack depth.
427 uint64_t depth
= cpu_data_ptr
[cnum
]->cpu_kernel_stack
428 + sizeof(struct x86_kernel_state
)
429 + sizeof(struct i386_exception_link
*)
431 if (depth
> kernel_stack_depth_max
) {
432 kernel_stack_depth_max
= (vm_offset_t
)depth
;
433 KERNEL_DEBUG_CONSTANT(
434 MACHDBG_CODE(DBG_MACH_SCHED
, MACH_STACK_DEPTH
),
435 (long) depth
, (long) VM_KERNEL_UNSLIDE(rip
), 0, 0, 0);
443 long dr7
= 0x400; /* magic dr7 reset value; 32 bit on i386, 64 bit on x86_64 */
444 __asm__
volatile("mov %0,%%dr7" : : "r" (dr7
));
447 unsigned kdp_has_active_watchpoints
= 0;
448 #define NO_WATCHPOINTS (!kdp_has_active_watchpoints)
450 #define NO_WATCHPOINTS 1
453 * Trap from kernel mode. Only page-fault errors are recoverable,
454 * and then only in special circumstances. All other errors are
455 * fatal. Return value indicates if trap was handled.
460 x86_saved_state_t
*state
,
464 x86_saved_state32_t
*saved_state
;
466 x86_saved_state64_t
*saved_state
;
471 vm_map_t map
= 0; /* protected by T_PAGE_FAULT */
472 kern_return_t result
= KERN_FAILURE
;
479 #if NCOPY_WINDOWS > 0
480 int fault_in_copy_window
= -1;
484 thread
= current_thread();
487 if (__improbable(is_saved_state64(state
))) {
488 panic_64(state
, 0, "Kernel trap with 64-bit state", FALSE
);
491 saved_state
= saved_state32(state
);
493 /* Record cpu where state was captured (trampolines don't set this) */
494 saved_state
->cpu
= cpu_number();
496 vaddr
= (user_addr_t
)saved_state
->cr2
;
497 type
= saved_state
->trapno
;
498 code
= saved_state
->err
& 0xffff;
499 intr
= (saved_state
->efl
& EFL_IF
) != 0; /* state of ints at trap */
500 kern_ip
= (vm_offset_t
)saved_state
->eip
;
502 if (__improbable(is_saved_state32(state
)))
503 panic("kernel_trap(%p) with 32-bit state", state
);
504 saved_state
= saved_state64(state
);
506 /* Record cpu where state was captured */
507 saved_state
->isf
.cpu
= cpu_number();
509 vaddr
= (user_addr_t
)saved_state
->cr2
;
510 type
= saved_state
->isf
.trapno
;
511 code
= (int)(saved_state
->isf
.err
& 0xffff);
512 intr
= (saved_state
->isf
.rflags
& EFL_IF
) != 0; /* state of ints at trap */
513 kern_ip
= (vm_offset_t
)saved_state
->isf
.rip
;
516 myast
= ast_pending();
518 perfASTCallback astfn
= perfASTHook
;
519 if (__improbable(astfn
!= NULL
)) {
520 if (*myast
& AST_CHUD_ALL
)
521 astfn(AST_CHUD_ALL
, myast
);
523 *myast
&= ~AST_CHUD_ALL
;
528 perfCallback fn
= perfTrapHook
;
529 if (__improbable(fn
!= NULL
)) {
530 if (fn(type
, NULL
, 0, 0) == KERN_SUCCESS
) {
532 * If it succeeds, we are done...
539 if (__improbable(tempDTraceTrapHook
!= NULL
)) {
540 if (tempDTraceTrapHook(type
, state
, lo_spp
, 0) == KERN_SUCCESS
) {
542 * If it succeeds, we are done...
547 #endif /* CONFIG_DTRACE */
550 * we come here with interrupts off as we don't want to recurse
551 * on preemption below. but we do want to re-enable interrupts
552 * as soon we possibly can to hold latency down
554 if (__improbable(T_PREEMPT
== type
)) {
555 ast_taken(AST_PREEMPTION
, FALSE
);
557 KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE
,
558 (MACHDBG_CODE(DBG_MACH_EXCP_KTRAP_x86
, type
)) | DBG_FUNC_NONE
,
559 0, 0, 0, VM_KERNEL_UNSLIDE(kern_ip
), 0);
563 if (T_PAGE_FAULT
== type
) {
565 * assume we're faulting in the kernel map
569 if (__probable(thread
!= THREAD_NULL
&& thread
->map
!= kernel_map
)) {
570 #if NCOPY_WINDOWS > 0
571 vm_offset_t copy_window_base
;
575 kvaddr
= (vm_offset_t
)vaddr
;
577 * must determine if fault occurred in
578 * the copy window while pre-emption is
579 * disabled for this processor so that
580 * we only need to look at the window
581 * associated with this processor
583 copy_window_base
= current_cpu_datap()->cpu_copywindow_base
;
585 if (kvaddr
>= copy_window_base
&& kvaddr
< (copy_window_base
+ (NBPDE
* NCOPY_WINDOWS
)) ) {
587 window_index
= (int)((kvaddr
- copy_window_base
) / NBPDE
);
589 if (thread
->machine
.copy_window
[window_index
].user_base
!= (user_addr_t
)-1) {
591 kvaddr
-= (copy_window_base
+ (NBPDE
* window_index
));
592 vaddr
= thread
->machine
.copy_window
[window_index
].user_base
+ kvaddr
;
595 fault_in_copy_window
= window_index
;
600 if (__probable(vaddr
< VM_MAX_USER_PAGE_ADDRESS
)) {
601 /* fault occurred in userspace */
605 /* Intercept a potential Supervisor Mode Execute
606 * Protection fault. These criteria identify
607 * both NX faults and SMEP faults, but both
608 * are fatal. We avoid checking PTEs (racy).
609 * (The VM could just redrive a SMEP fault, hence
612 if (__improbable((code
== (T_PF_PROT
| T_PF_EXECUTE
)) && (pmap_smep_enabled
) && (saved_state
->isf
.rip
== vaddr
))) {
617 * If we're not sharing cr3 with the user
618 * and we faulted in copyio,
619 * then switch cr3 here and dismiss the fault.
622 (thread
->machine
.specFlags
&CopyIOActive
) &&
623 map
->pmap
->pm_cr3
!= get_cr3_base()) {
624 pmap_assert(current_cpu_datap()->cpu_pmap_pcid_enabled
== FALSE
);
625 set_cr3_raw(map
->pmap
->pm_cr3
);
632 user_addr_t kd_vaddr
= is_user
? vaddr
: VM_KERNEL_UNSLIDE(vaddr
);
633 KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE
,
634 (MACHDBG_CODE(DBG_MACH_EXCP_KTRAP_x86
, type
)) | DBG_FUNC_NONE
,
635 (unsigned)(kd_vaddr
>> 32), (unsigned)kd_vaddr
, is_user
,
636 VM_KERNEL_UNSLIDE(kern_ip
), 0);
639 (void) ml_set_interrupts_enabled(intr
);
651 case T_FLOATING_POINT_ERROR
:
655 case T_SSE_FLOAT_ERROR
:
660 if ((saved_state
->efl
& EFL_TF
) == 0 && NO_WATCHPOINTS
)
662 if ((saved_state
->isf
.rflags
& EFL_TF
) == 0 && NO_WATCHPOINTS
)
665 /* We've somehow encountered a debug
666 * register match that does not belong
667 * to the kernel debugger.
668 * This isn't supposed to happen.
681 if (thread
!= THREAD_NULL
&& thread
->options
& TH_OPT_DTRACE
) { /* Executing under dtrace_probe? */
682 if (dtrace_tally_fault(vaddr
)) { /* Should a fault under dtrace be ignored? */
684 * DTrace has "anticipated" the possibility of this fault, and has
685 * established the suitable recovery state. Drop down now into the
686 * recovery handling code in "case T_GENERAL_PROTECTION:".
691 #endif /* CONFIG_DTRACE */
695 if (code
& T_PF_WRITE
)
696 prot
|= VM_PROT_WRITE
;
698 if (code
& T_PF_EXECUTE
)
699 prot
|= VM_PROT_EXECUTE
;
702 result
= vm_fault(map
,
703 vm_map_trunc_page(vaddr
),
706 THREAD_UNINT
, NULL
, 0);
708 if (result
== KERN_SUCCESS
) {
709 #if NCOPY_WINDOWS > 0
710 if (fault_in_copy_window
!= -1) {
711 ml_set_interrupts_enabled(FALSE
);
712 copy_window_fault(thread
, map
,
713 fault_in_copy_window
);
714 (void) ml_set_interrupts_enabled(intr
);
716 #endif /* NCOPY_WINDOWS > 0 */
724 #endif /* CONFIG_DTRACE */
726 case T_GENERAL_PROTECTION
:
728 * If there is a failure recovery address
729 * for this fault, go there.
731 for (rp
= recover_table
; rp
< recover_table_end
; rp
++) {
732 if (kern_ip
== rp
->fault_addr
) {
733 set_recovery_ip(saved_state
, rp
->recover_addr
);
739 * Check thread recovery address also.
741 if (thread
!= THREAD_NULL
&& thread
->recover
) {
742 set_recovery_ip(saved_state
, thread
->recover
);
747 * Unanticipated page-fault errors in kernel
754 * Exception 15 is reserved but some chips may generate it
755 * spuriously. Seen at startup on AMD Athlon-64.
758 kprintf("kernel_trap() ignoring spurious trap 15\n");
762 /* Ensure that the i386_kernel_state at the base of the
763 * current thread's stack (if any) is synchronized with the
764 * context at the moment of the trap, to facilitate
765 * access through the debugger.
767 sync_iss_to_iks(state
);
769 if (current_debugger
!= KDB_CUR_DB
) {
770 if (kdp_i386_trap(type
, saved_state
, result
, (vm_offset_t
)vaddr
))
776 panic_trap(saved_state
);
785 set_recovery_ip(x86_saved_state32_t
*saved_state
, vm_offset_t ip
)
787 saved_state
->eip
= ip
;
791 set_recovery_ip(x86_saved_state64_t
*saved_state
, vm_offset_t ip
)
793 saved_state
->isf
.rip
= ip
;
800 panic_trap(x86_saved_state32_t
*regs
)
802 const char *trapname
= "Unknown";
803 pal_cr_t cr0
, cr2
, cr3
, cr4
;
805 pal_get_control_registers( &cr0
, &cr2
, &cr3
, &cr4
);
808 * Issue an I/O port read if one has been requested - this is an
809 * event logic analyzers can use as a trigger point.
811 panic_io_port_read();
813 kprintf("panic trap number 0x%x, eip 0x%x\n", regs
->trapno
, regs
->eip
);
814 kprintf("cr0 0x%08x cr2 0x%08x cr3 0x%08x cr4 0x%08x\n",
817 if (regs
->trapno
< TRAP_TYPES
)
818 trapname
= trap_type
[regs
->trapno
];
820 panic("Kernel trap at 0x%08x, type %d=%s, registers:\n"
821 "CR0: 0x%08x, CR2: 0x%08x, CR3: 0x%08x, CR4: 0x%08x\n"
822 "EAX: 0x%08x, EBX: 0x%08x, ECX: 0x%08x, EDX: 0x%08x\n"
823 "CR2: 0x%08x, EBP: 0x%08x, ESI: 0x%08x, EDI: 0x%08x\n"
824 "EFL: 0x%08x, EIP: 0x%08x, CS: 0x%08x, DS: 0x%08x\n"
825 "Error code: 0x%08x%s\n",
826 regs
->eip
, regs
->trapno
, trapname
, cr0
, cr2
, cr3
, cr4
,
827 regs
->eax
,regs
->ebx
,regs
->ecx
,regs
->edx
,
828 regs
->cr2
,regs
->ebp
,regs
->esi
,regs
->edi
,
829 regs
->efl
,regs
->eip
,regs
->cs
& 0xFFFF, regs
->ds
& 0xFFFF, regs
->err
,
830 virtualized
? " VMM" : "");
832 * This next statement is not executed,
833 * but it's needed to stop the compiler using tail call optimization
834 * for the panic call - which confuses the subsequent backtrace.
842 panic_trap(x86_saved_state64_t
*regs
)
844 const char *trapname
= "Unknown";
845 pal_cr_t cr0
, cr2
, cr3
, cr4
;
846 boolean_t potential_smep_fault
= FALSE
, potential_kernel_NX_fault
= FALSE
;
848 pal_get_control_registers( &cr0
, &cr2
, &cr3
, &cr4
);
849 assert(ml_get_interrupts_enabled() == FALSE
);
850 current_cpu_datap()->cpu_fatal_trap_state
= regs
;
852 * Issue an I/O port read if one has been requested - this is an
853 * event logic analyzers can use as a trigger point.
855 panic_io_port_read();
857 kprintf("panic trap number 0x%x, rip 0x%016llx\n",
858 regs
->isf
.trapno
, regs
->isf
.rip
);
859 kprintf("cr0 0x%016llx cr2 0x%016llx cr3 0x%016llx cr4 0x%016llx\n",
862 if (regs
->isf
.trapno
< TRAP_TYPES
)
863 trapname
= trap_type
[regs
->isf
.trapno
];
865 if ((regs
->isf
.trapno
== T_PAGE_FAULT
) && (regs
->isf
.err
== (T_PF_PROT
| T_PF_EXECUTE
)) && (regs
->isf
.rip
== regs
->cr2
)) {
866 if (pmap_smep_enabled
&& (regs
->isf
.rip
< VM_MAX_USER_PAGE_ADDRESS
)) {
867 potential_smep_fault
= TRUE
;
868 } else if (regs
->isf
.rip
>= VM_MIN_KERNEL_AND_KEXT_ADDRESS
) {
869 potential_kernel_NX_fault
= TRUE
;
874 panic("Kernel trap at 0x%016llx, type %d=%s, registers:\n"
875 "CR0: 0x%016llx, CR2: 0x%016llx, CR3: 0x%016llx, CR4: 0x%016llx\n"
876 "RAX: 0x%016llx, RBX: 0x%016llx, RCX: 0x%016llx, RDX: 0x%016llx\n"
877 "RSP: 0x%016llx, RBP: 0x%016llx, RSI: 0x%016llx, RDI: 0x%016llx\n"
878 "R8: 0x%016llx, R9: 0x%016llx, R10: 0x%016llx, R11: 0x%016llx\n"
879 "R12: 0x%016llx, R13: 0x%016llx, R14: 0x%016llx, R15: 0x%016llx\n"
880 "RFL: 0x%016llx, RIP: 0x%016llx, CS: 0x%016llx, SS: 0x%016llx\n"
881 "Fault CR2: 0x%016llx, Error code: 0x%016llx, Fault CPU: 0x%x%s%s%s\n",
882 regs
->isf
.rip
, regs
->isf
.trapno
, trapname
,
884 regs
->rax
, regs
->rbx
, regs
->rcx
, regs
->rdx
,
885 regs
->isf
.rsp
, regs
->rbp
, regs
->rsi
, regs
->rdi
,
886 regs
->r8
, regs
->r9
, regs
->r10
, regs
->r11
,
887 regs
->r12
, regs
->r13
, regs
->r14
, regs
->r15
,
888 regs
->isf
.rflags
, regs
->isf
.rip
, regs
->isf
.cs
& 0xFFFF,
889 regs
->isf
.ss
& 0xFFFF,regs
->cr2
, regs
->isf
.err
, regs
->isf
.cpu
,
890 virtualized
? " VMM" : "",
891 potential_kernel_NX_fault
? " Kernel NX fault" : "",
892 potential_smep_fault
? " SMEP/User NX fault" : "");
894 * This next statement is not executed,
895 * but it's needed to stop the compiler using tail call optimization
896 * for the panic call - which confuses the subsequent backtrace.
903 extern kern_return_t
dtrace_user_probe(x86_saved_state_t
*);
907 * Trap from user mode.
911 x86_saved_state_t
*saved_state
)
915 mach_exception_code_t code
;
916 mach_exception_subcode_t subcode
;
920 thread_t thread
= current_thread();
924 unsigned long dr6
= 0; /* 32 bit for i386, 64 bit for x86_64 */
926 assert((is_saved_state32(saved_state
) && !thread_is_64bit(thread
)) ||
927 (is_saved_state64(saved_state
) && thread_is_64bit(thread
)));
929 if (is_saved_state64(saved_state
)) {
930 x86_saved_state64_t
*regs
;
932 regs
= saved_state64(saved_state
);
934 /* Record cpu where state was captured */
935 regs
->isf
.cpu
= cpu_number();
937 type
= regs
->isf
.trapno
;
938 err
= (int)regs
->isf
.err
& 0xffff;
939 vaddr
= (user_addr_t
)regs
->cr2
;
940 rip
= (user_addr_t
)regs
->isf
.rip
;
942 x86_saved_state32_t
*regs
;
944 regs
= saved_state32(saved_state
);
946 /* Record cpu where state was captured */
947 regs
->cpu
= cpu_number();
950 err
= regs
->err
& 0xffff;
951 vaddr
= (user_addr_t
)regs
->cr2
;
952 rip
= (user_addr_t
)regs
->eip
;
955 if ((type
== T_DEBUG
) && thread
->machine
.ids
) {
956 unsigned long clear
= 0;
957 /* Stash and clear this processor's DR6 value, in the event
958 * this was a debug register match
960 __asm__
volatile ("mov %%db6, %0" : "=r" (dr6
));
961 __asm__
volatile ("mov %0, %%db6" : : "r" (clear
));
966 KERNEL_DEBUG_CONSTANT_IST(KDEBUG_TRACE
,
967 (MACHDBG_CODE(DBG_MACH_EXCP_UTRAP_x86
, type
)) | DBG_FUNC_NONE
,
968 (unsigned)(vaddr
>>32), (unsigned)vaddr
,
969 (unsigned)(rip
>>32), (unsigned)rip
, 0);
976 kprintf("user_trap(0x%08x) type=%d vaddr=0x%016llx\n",
977 saved_state
, type
, vaddr
);
980 perfASTCallback astfn
= perfASTHook
;
981 if (__improbable(astfn
!= NULL
)) {
982 myast
= ast_pending();
983 if (*myast
& AST_CHUD_ALL
) {
984 astfn(AST_CHUD_ALL
, myast
);
988 /* Is there a hook? */
989 perfCallback fn
= perfTrapHook
;
990 if (__improbable(fn
!= NULL
)) {
991 if (fn(type
, saved_state
, 0, 0) == KERN_SUCCESS
)
992 return; /* If it succeeds, we are done... */
996 * DTrace does not consume all user traps, only INT_3's for now.
997 * Avoid needlessly calling tempDTraceTrapHook here, and let the
998 * INT_3 case handle them.
1000 DEBUG_KPRINT_SYSCALL_MASK(1,
1001 "user_trap: type=0x%x(%s) err=0x%x cr2=%p rip=%p\n",
1002 type
, trap_type
[type
], err
, (void *)(long) vaddr
, (void *)(long) rip
);
1006 case T_DIVIDE_ERROR
:
1007 exc
= EXC_ARITHMETIC
;
1008 code
= EXC_I386_DIV
;
1015 * Update the PCB with this processor's DR6 value
1016 * in the event this was a debug register match.
1018 pcb
= THREAD_TO_PCB(thread
);
1021 * We can get and set the status register
1022 * in 32-bit mode even on a 64-bit thread
1023 * because the high order bits are not
1026 if (thread_is_64bit(thread
)) {
1027 x86_debug_state64_t
*ids
= pcb
->ids
;
1029 } else { /* 32 bit thread */
1030 x86_debug_state32_t
*ids
= pcb
->ids
;
1031 ids
->dr6
= (uint32_t) dr6
;
1034 exc
= EXC_BREAKPOINT
;
1035 code
= EXC_I386_SGL
;
1040 if (dtrace_user_probe(saved_state
) == KERN_SUCCESS
)
1041 return; /* If it succeeds, we are done... */
1043 exc
= EXC_BREAKPOINT
;
1044 code
= EXC_I386_BPT
;
1048 exc
= EXC_ARITHMETIC
;
1049 code
= EXC_I386_INTO
;
1052 case T_OUT_OF_BOUNDS
:
1054 code
= EXC_I386_BOUND
;
1057 case T_INVALID_OPCODE
:
1058 exc
= EXC_BAD_INSTRUCTION
;
1059 code
= EXC_I386_INVOP
;
1067 fpextovrflt(); /* Propagates exception directly, doesn't return */
1070 case T_INVALID_TSS
: /* invalid TSS == iret with NT flag set */
1071 exc
= EXC_BAD_INSTRUCTION
;
1072 code
= EXC_I386_INVTSSFLT
;
1076 case T_SEGMENT_NOT_PRESENT
:
1077 exc
= EXC_BAD_INSTRUCTION
;
1078 code
= EXC_I386_SEGNPFLT
;
1083 exc
= EXC_BAD_INSTRUCTION
;
1084 code
= EXC_I386_STKFLT
;
1088 case T_GENERAL_PROTECTION
:
1090 * There's a wide range of circumstances which generate this
1091 * class of exception. From user-space, many involve bad
1092 * addresses (such as a non-canonical 64-bit address).
1093 * So we map this to EXC_BAD_ACCESS (and thereby SIGSEGV).
1094 * The trouble is cr2 doesn't contain the faulting address;
1095 * we'd need to decode the faulting instruction to really
1096 * determine this. We'll leave that to debuggers.
1097 * However, attempted execution of privileged instructions
1098 * (e.g. cli) also generate GP faults and so we map these to
1099 * to EXC_BAD_ACCESS (and thence SIGSEGV) also - rather than
1100 * EXC_BAD_INSTRUCTION which is more accurate. We just can't
1103 exc
= EXC_BAD_ACCESS
;
1104 code
= EXC_I386_GPFLT
;
1110 prot
= VM_PROT_READ
;
1112 if (err
& T_PF_WRITE
)
1113 prot
|= VM_PROT_WRITE
;
1115 if (__improbable(err
& T_PF_EXECUTE
))
1116 prot
|= VM_PROT_EXECUTE
;
1118 kret
= vm_fault(thread
->map
, vm_map_trunc_page(vaddr
),
1120 THREAD_ABORTSAFE
, NULL
, 0);
1122 if (__probable((kret
== KERN_SUCCESS
) || (kret
== KERN_ABORTED
))) {
1123 thread_exception_return();
1127 user_page_fault_continue(kret
);
1131 case T_SSE_FLOAT_ERROR
:
1132 fpSSEexterrflt(); /* Propagates exception directly, doesn't return */
1136 case T_FLOATING_POINT_ERROR
:
1137 fpexterrflt(); /* Propagates exception directly, doesn't return */
1142 if (dtrace_user_probe(saved_state
) == KERN_SUCCESS
)
1143 return; /* If it succeeds, we are done... */
1146 * If we get an INT 0x7f when we do not expect to,
1147 * treat it as an illegal instruction
1149 exc
= EXC_BAD_INSTRUCTION
;
1150 code
= EXC_I386_INVOP
;
1154 panic("Unexpected user trap, type %d", type
);
1157 /* Note: Codepaths that directly return from user_trap() have pending
1158 * ASTs processed in locore
1160 i386_exception(exc
, code
, subcode
);
1166 * Handle AST traps for i386.
1169 extern void log_thread_action (thread_t
, char *);
1172 i386_astintr(int preemption
)
1174 ast_t mask
= AST_ALL
;
1178 mask
= AST_PREEMPTION
;
1188 * Handle exceptions for i386.
1190 * If we are an AT bus machine, we must turn off the AST for a
1191 * delayed floating-point exception.
1193 * If we are providing floating-point emulation, we may have
1194 * to retrieve the real register values from the floating point
1200 mach_exception_code_t code
,
1201 mach_exception_subcode_t subcode
)
1203 mach_exception_data_type_t codes
[EXCEPTION_CODE_MAX
];
1205 DEBUG_KPRINT_SYSCALL_MACH("i386_exception: exc=%d code=0x%llx subcode=0x%llx\n",
1206 exc
, code
, subcode
);
1207 codes
[0] = code
; /* new exception interface */
1209 exception_triage(exc
, codes
, 2);
1214 /* Synchronize a thread's i386_kernel_state (if any) with the given
1215 * i386_saved_state_t obtained from the trap/IPI handler; called in
1216 * kernel_trap() prior to entering the debugger, and when receiving
1221 sync_iss_to_iks(x86_saved_state_t
*saved_state
)
1223 struct x86_kernel_state
*iks
;
1225 boolean_t record_active_regs
= FALSE
;
1227 /* The PAL may have a special way to sync registers */
1228 if( saved_state
->flavor
== THREAD_STATE_NONE
)
1229 pal_get_kern_regs( saved_state
);
1231 if ((kstack
= current_thread()->kernel_stack
) != 0) {
1233 x86_saved_state32_t
*regs
= saved_state32(saved_state
);
1235 x86_saved_state64_t
*regs
= saved_state64(saved_state
);
1238 iks
= STACK_IKS(kstack
);
1240 /* Did we take the trap/interrupt in kernel mode? */
1242 if (regs
== USER_REGS32(current_thread()))
1243 record_active_regs
= TRUE
;
1245 iks
->k_ebx
= regs
->ebx
;
1246 iks
->k_esp
= (int)regs
;
1247 iks
->k_ebp
= regs
->ebp
;
1248 iks
->k_edi
= regs
->edi
;
1249 iks
->k_esi
= regs
->esi
;
1250 iks
->k_eip
= regs
->eip
;
1253 if (regs
== USER_REGS64(current_thread()))
1254 record_active_regs
= TRUE
;
1256 iks
->k_rbx
= regs
->rbx
;
1257 iks
->k_rsp
= regs
->isf
.rsp
;
1258 iks
->k_rbp
= regs
->rbp
;
1259 iks
->k_r12
= regs
->r12
;
1260 iks
->k_r13
= regs
->r13
;
1261 iks
->k_r14
= regs
->r14
;
1262 iks
->k_r15
= regs
->r15
;
1263 iks
->k_rip
= regs
->isf
.rip
;
1268 if (record_active_regs
== TRUE
) {
1270 /* Show the trap handler path */
1271 __asm__
volatile("movl %%ebx, %0" : "=m" (iks
->k_ebx
));
1272 __asm__
volatile("movl %%esp, %0" : "=m" (iks
->k_esp
));
1273 __asm__
volatile("movl %%ebp, %0" : "=m" (iks
->k_ebp
));
1274 __asm__
volatile("movl %%edi, %0" : "=m" (iks
->k_edi
));
1275 __asm__
volatile("movl %%esi, %0" : "=m" (iks
->k_esi
));
1276 /* "Current" instruction pointer */
1277 __asm__
volatile("movl $1f, %0\n1:" : "=m" (iks
->k_eip
));
1279 /* Show the trap handler path */
1280 __asm__
volatile("movq %%rbx, %0" : "=m" (iks
->k_rbx
));
1281 __asm__
volatile("movq %%rsp, %0" : "=m" (iks
->k_rsp
));
1282 __asm__
volatile("movq %%rbp, %0" : "=m" (iks
->k_rbp
));
1283 __asm__
volatile("movq %%r12, %0" : "=m" (iks
->k_r12
));
1284 __asm__
volatile("movq %%r13, %0" : "=m" (iks
->k_r13
));
1285 __asm__
volatile("movq %%r14, %0" : "=m" (iks
->k_r14
));
1286 __asm__
volatile("movq %%r15, %0" : "=m" (iks
->k_r15
));
1287 /* "Current" instruction pointer */
1288 __asm__
volatile("leaq 1f(%%rip), %%rax; mov %%rax, %0\n1:"
1297 * This is used by the NMI interrupt handler (from mp.c) to
1298 * uncondtionally sync the trap handler context to the IKS
1299 * irrespective of whether the NMI was fielded in kernel
1303 sync_iss_to_iks_unconditionally(__unused x86_saved_state_t
*saved_state
) {
1304 struct x86_kernel_state
*iks
;
1307 if ((kstack
= current_thread()->kernel_stack
) != 0) {
1308 iks
= STACK_IKS(kstack
);
1310 /* Display the trap handler path */
1311 __asm__
volatile("movl %%ebx, %0" : "=m" (iks
->k_ebx
));
1312 __asm__
volatile("movl %%esp, %0" : "=m" (iks
->k_esp
));
1313 __asm__
volatile("movl %%ebp, %0" : "=m" (iks
->k_ebp
));
1314 __asm__
volatile("movl %%edi, %0" : "=m" (iks
->k_edi
));
1315 __asm__
volatile("movl %%esi, %0" : "=m" (iks
->k_esi
));
1316 /* "Current" instruction pointer */
1317 __asm__
volatile("movl $1f, %0\n1:" : "=m" (iks
->k_eip
));
1319 /* Display the trap handler path */
1320 __asm__
volatile("movq %%rbx, %0" : "=m" (iks
->k_rbx
));
1321 __asm__
volatile("movq %%rsp, %0" : "=m" (iks
->k_rsp
));
1322 __asm__
volatile("movq %%rbp, %0" : "=m" (iks
->k_rbp
));
1323 __asm__
volatile("movq %%r12, %0" : "=m" (iks
->k_r12
));
1324 __asm__
volatile("movq %%r13, %0" : "=m" (iks
->k_r13
));
1325 __asm__
volatile("movq %%r14, %0" : "=m" (iks
->k_r14
));
1326 __asm__
volatile("movq %%r15, %0" : "=m" (iks
->k_r15
));
1327 /* "Current" instruction pointer */
1328 __asm__
volatile("leaq 1f(%%rip), %%rax; mov %%rax, %0\n1:" : "=m" (iks
->k_rip
)::"rax");